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> |
1da177e4 | 75 | |
96f951ed | 76 | #include <asm/switch_to.h> |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
db7e527d | 79 | #include <asm/mutex.h> |
e6e6685a GC |
80 | #ifdef CONFIG_PARAVIRT |
81 | #include <asm/paravirt.h> | |
82 | #endif | |
1da177e4 | 83 | |
029632fb | 84 | #include "sched.h" |
391e43da | 85 | #include "../workqueue_sched.h" |
6e0534f2 | 86 | |
a8d154b0 | 87 | #define CREATE_TRACE_POINTS |
ad8d75ff | 88 | #include <trace/events/sched.h> |
a8d154b0 | 89 | |
029632fb | 90 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 91 | { |
58088ad0 PT |
92 | unsigned long delta; |
93 | ktime_t soft, hard, now; | |
d0b27fa7 | 94 | |
58088ad0 PT |
95 | for (;;) { |
96 | if (hrtimer_active(period_timer)) | |
97 | break; | |
98 | ||
99 | now = hrtimer_cb_get_time(period_timer); | |
100 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 101 | |
58088ad0 PT |
102 | soft = hrtimer_get_softexpires(period_timer); |
103 | hard = hrtimer_get_expires(period_timer); | |
104 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
105 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
106 | HRTIMER_MODE_ABS_PINNED, 0); | |
107 | } | |
108 | } | |
109 | ||
029632fb PZ |
110 | DEFINE_MUTEX(sched_domains_mutex); |
111 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 112 | |
fe44d621 | 113 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 114 | |
029632fb | 115 | void update_rq_clock(struct rq *rq) |
3e51f33f | 116 | { |
fe44d621 | 117 | s64 delta; |
305e6835 | 118 | |
61eadef6 | 119 | if (rq->skip_clock_update > 0) |
f26f9aff | 120 | return; |
aa483808 | 121 | |
fe44d621 PZ |
122 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
123 | rq->clock += delta; | |
124 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
125 | } |
126 | ||
bf5c91ba IM |
127 | /* |
128 | * Debugging: various feature bits | |
129 | */ | |
f00b45c1 | 130 | |
f00b45c1 PZ |
131 | #define SCHED_FEAT(name, enabled) \ |
132 | (1UL << __SCHED_FEAT_##name) * enabled | | |
133 | ||
bf5c91ba | 134 | const_debug unsigned int sysctl_sched_features = |
391e43da | 135 | #include "features.h" |
f00b45c1 PZ |
136 | 0; |
137 | ||
138 | #undef SCHED_FEAT | |
139 | ||
140 | #ifdef CONFIG_SCHED_DEBUG | |
141 | #define SCHED_FEAT(name, enabled) \ | |
142 | #name , | |
143 | ||
983ed7a6 | 144 | static __read_mostly char *sched_feat_names[] = { |
391e43da | 145 | #include "features.h" |
f00b45c1 PZ |
146 | NULL |
147 | }; | |
148 | ||
149 | #undef SCHED_FEAT | |
150 | ||
34f3a814 | 151 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 152 | { |
f00b45c1 PZ |
153 | int i; |
154 | ||
f8b6d1cc | 155 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
156 | if (!(sysctl_sched_features & (1UL << i))) |
157 | seq_puts(m, "NO_"); | |
158 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 159 | } |
34f3a814 | 160 | seq_puts(m, "\n"); |
f00b45c1 | 161 | |
34f3a814 | 162 | return 0; |
f00b45c1 PZ |
163 | } |
164 | ||
f8b6d1cc PZ |
165 | #ifdef HAVE_JUMP_LABEL |
166 | ||
c5905afb IM |
167 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
168 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
169 | |
170 | #define SCHED_FEAT(name, enabled) \ | |
171 | jump_label_key__##enabled , | |
172 | ||
c5905afb | 173 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
174 | #include "features.h" |
175 | }; | |
176 | ||
177 | #undef SCHED_FEAT | |
178 | ||
179 | static void sched_feat_disable(int i) | |
180 | { | |
c5905afb IM |
181 | if (static_key_enabled(&sched_feat_keys[i])) |
182 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
183 | } |
184 | ||
185 | static void sched_feat_enable(int i) | |
186 | { | |
c5905afb IM |
187 | if (!static_key_enabled(&sched_feat_keys[i])) |
188 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
189 | } |
190 | #else | |
191 | static void sched_feat_disable(int i) { }; | |
192 | static void sched_feat_enable(int i) { }; | |
193 | #endif /* HAVE_JUMP_LABEL */ | |
194 | ||
f00b45c1 PZ |
195 | static ssize_t |
196 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
197 | size_t cnt, loff_t *ppos) | |
198 | { | |
199 | char buf[64]; | |
7740191c | 200 | char *cmp; |
f00b45c1 PZ |
201 | int neg = 0; |
202 | int i; | |
203 | ||
204 | if (cnt > 63) | |
205 | cnt = 63; | |
206 | ||
207 | if (copy_from_user(&buf, ubuf, cnt)) | |
208 | return -EFAULT; | |
209 | ||
210 | buf[cnt] = 0; | |
7740191c | 211 | cmp = strstrip(buf); |
f00b45c1 | 212 | |
524429c3 | 213 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
214 | neg = 1; |
215 | cmp += 3; | |
216 | } | |
217 | ||
f8b6d1cc | 218 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 219 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 220 | if (neg) { |
f00b45c1 | 221 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
222 | sched_feat_disable(i); |
223 | } else { | |
f00b45c1 | 224 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
225 | sched_feat_enable(i); |
226 | } | |
f00b45c1 PZ |
227 | break; |
228 | } | |
229 | } | |
230 | ||
f8b6d1cc | 231 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
232 | return -EINVAL; |
233 | ||
42994724 | 234 | *ppos += cnt; |
f00b45c1 PZ |
235 | |
236 | return cnt; | |
237 | } | |
238 | ||
34f3a814 LZ |
239 | static int sched_feat_open(struct inode *inode, struct file *filp) |
240 | { | |
241 | return single_open(filp, sched_feat_show, NULL); | |
242 | } | |
243 | ||
828c0950 | 244 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
245 | .open = sched_feat_open, |
246 | .write = sched_feat_write, | |
247 | .read = seq_read, | |
248 | .llseek = seq_lseek, | |
249 | .release = single_release, | |
f00b45c1 PZ |
250 | }; |
251 | ||
252 | static __init int sched_init_debug(void) | |
253 | { | |
f00b45c1 PZ |
254 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
255 | &sched_feat_fops); | |
256 | ||
257 | return 0; | |
258 | } | |
259 | late_initcall(sched_init_debug); | |
f8b6d1cc | 260 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 261 | |
b82d9fdd PZ |
262 | /* |
263 | * Number of tasks to iterate in a single balance run. | |
264 | * Limited because this is done with IRQs disabled. | |
265 | */ | |
266 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
267 | ||
e9e9250b PZ |
268 | /* |
269 | * period over which we average the RT time consumption, measured | |
270 | * in ms. | |
271 | * | |
272 | * default: 1s | |
273 | */ | |
274 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
275 | ||
fa85ae24 | 276 | /* |
9f0c1e56 | 277 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
278 | * default: 1s |
279 | */ | |
9f0c1e56 | 280 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 281 | |
029632fb | 282 | __read_mostly int scheduler_running; |
6892b75e | 283 | |
9f0c1e56 PZ |
284 | /* |
285 | * part of the period that we allow rt tasks to run in us. | |
286 | * default: 0.95s | |
287 | */ | |
288 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 289 | |
fa85ae24 | 290 | |
1da177e4 | 291 | |
0970d299 | 292 | /* |
0122ec5b | 293 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 294 | */ |
70b97a7f | 295 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
296 | __acquires(rq->lock) |
297 | { | |
0970d299 PZ |
298 | struct rq *rq; |
299 | ||
0122ec5b PZ |
300 | lockdep_assert_held(&p->pi_lock); |
301 | ||
3a5c359a | 302 | for (;;) { |
0970d299 | 303 | rq = task_rq(p); |
05fa785c | 304 | raw_spin_lock(&rq->lock); |
65cc8e48 | 305 | if (likely(rq == task_rq(p))) |
3a5c359a | 306 | return rq; |
05fa785c | 307 | raw_spin_unlock(&rq->lock); |
b29739f9 | 308 | } |
b29739f9 IM |
309 | } |
310 | ||
1da177e4 | 311 | /* |
0122ec5b | 312 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 313 | */ |
70b97a7f | 314 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 315 | __acquires(p->pi_lock) |
1da177e4 LT |
316 | __acquires(rq->lock) |
317 | { | |
70b97a7f | 318 | struct rq *rq; |
1da177e4 | 319 | |
3a5c359a | 320 | for (;;) { |
0122ec5b | 321 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 322 | rq = task_rq(p); |
05fa785c | 323 | raw_spin_lock(&rq->lock); |
65cc8e48 | 324 | if (likely(rq == task_rq(p))) |
3a5c359a | 325 | return rq; |
0122ec5b PZ |
326 | raw_spin_unlock(&rq->lock); |
327 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 328 | } |
1da177e4 LT |
329 | } |
330 | ||
a9957449 | 331 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
332 | __releases(rq->lock) |
333 | { | |
05fa785c | 334 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
335 | } |
336 | ||
0122ec5b PZ |
337 | static inline void |
338 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 339 | __releases(rq->lock) |
0122ec5b | 340 | __releases(p->pi_lock) |
1da177e4 | 341 | { |
0122ec5b PZ |
342 | raw_spin_unlock(&rq->lock); |
343 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
344 | } |
345 | ||
1da177e4 | 346 | /* |
cc2a73b5 | 347 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 348 | */ |
a9957449 | 349 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
350 | __acquires(rq->lock) |
351 | { | |
70b97a7f | 352 | struct rq *rq; |
1da177e4 LT |
353 | |
354 | local_irq_disable(); | |
355 | rq = this_rq(); | |
05fa785c | 356 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
357 | |
358 | return rq; | |
359 | } | |
360 | ||
8f4d37ec PZ |
361 | #ifdef CONFIG_SCHED_HRTICK |
362 | /* | |
363 | * Use HR-timers to deliver accurate preemption points. | |
364 | * | |
365 | * Its all a bit involved since we cannot program an hrt while holding the | |
366 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
367 | * reschedule event. | |
368 | * | |
369 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
370 | * rq->lock. | |
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 |
31656519 PZ |
398 | /* |
399 | * called from hardirq (IPI) context | |
400 | */ | |
401 | static void __hrtick_start(void *arg) | |
b328ca18 | 402 | { |
31656519 | 403 | struct rq *rq = arg; |
b328ca18 | 404 | |
05fa785c | 405 | raw_spin_lock(&rq->lock); |
31656519 PZ |
406 | hrtimer_restart(&rq->hrtick_timer); |
407 | rq->hrtick_csd_pending = 0; | |
05fa785c | 408 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
409 | } |
410 | ||
31656519 PZ |
411 | /* |
412 | * Called to set the hrtick timer state. | |
413 | * | |
414 | * called with rq->lock held and irqs disabled | |
415 | */ | |
029632fb | 416 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 417 | { |
31656519 PZ |
418 | struct hrtimer *timer = &rq->hrtick_timer; |
419 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 420 | |
cc584b21 | 421 | hrtimer_set_expires(timer, time); |
31656519 PZ |
422 | |
423 | if (rq == this_rq()) { | |
424 | hrtimer_restart(timer); | |
425 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 426 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
427 | rq->hrtick_csd_pending = 1; |
428 | } | |
b328ca18 PZ |
429 | } |
430 | ||
431 | static int | |
432 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
433 | { | |
434 | int cpu = (int)(long)hcpu; | |
435 | ||
436 | switch (action) { | |
437 | case CPU_UP_CANCELED: | |
438 | case CPU_UP_CANCELED_FROZEN: | |
439 | case CPU_DOWN_PREPARE: | |
440 | case CPU_DOWN_PREPARE_FROZEN: | |
441 | case CPU_DEAD: | |
442 | case CPU_DEAD_FROZEN: | |
31656519 | 443 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
444 | return NOTIFY_OK; |
445 | } | |
446 | ||
447 | return NOTIFY_DONE; | |
448 | } | |
449 | ||
fa748203 | 450 | static __init void init_hrtick(void) |
b328ca18 PZ |
451 | { |
452 | hotcpu_notifier(hotplug_hrtick, 0); | |
453 | } | |
31656519 PZ |
454 | #else |
455 | /* | |
456 | * Called to set the hrtick timer state. | |
457 | * | |
458 | * called with rq->lock held and irqs disabled | |
459 | */ | |
029632fb | 460 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 461 | { |
7f1e2ca9 | 462 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 463 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 464 | } |
b328ca18 | 465 | |
006c75f1 | 466 | static inline void init_hrtick(void) |
8f4d37ec | 467 | { |
8f4d37ec | 468 | } |
31656519 | 469 | #endif /* CONFIG_SMP */ |
8f4d37ec | 470 | |
31656519 | 471 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 472 | { |
31656519 PZ |
473 | #ifdef CONFIG_SMP |
474 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 475 | |
31656519 PZ |
476 | rq->hrtick_csd.flags = 0; |
477 | rq->hrtick_csd.func = __hrtick_start; | |
478 | rq->hrtick_csd.info = rq; | |
479 | #endif | |
8f4d37ec | 480 | |
31656519 PZ |
481 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
482 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 483 | } |
006c75f1 | 484 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
485 | static inline void hrtick_clear(struct rq *rq) |
486 | { | |
487 | } | |
488 | ||
8f4d37ec PZ |
489 | static inline void init_rq_hrtick(struct rq *rq) |
490 | { | |
491 | } | |
492 | ||
b328ca18 PZ |
493 | static inline void init_hrtick(void) |
494 | { | |
495 | } | |
006c75f1 | 496 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 497 | |
c24d20db IM |
498 | /* |
499 | * resched_task - mark a task 'to be rescheduled now'. | |
500 | * | |
501 | * On UP this means the setting of the need_resched flag, on SMP it | |
502 | * might also involve a cross-CPU call to trigger the scheduler on | |
503 | * the target CPU. | |
504 | */ | |
505 | #ifdef CONFIG_SMP | |
506 | ||
507 | #ifndef tsk_is_polling | |
508 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
509 | #endif | |
510 | ||
029632fb | 511 | void resched_task(struct task_struct *p) |
c24d20db IM |
512 | { |
513 | int cpu; | |
514 | ||
05fa785c | 515 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 516 | |
5ed0cec0 | 517 | if (test_tsk_need_resched(p)) |
c24d20db IM |
518 | return; |
519 | ||
5ed0cec0 | 520 | set_tsk_need_resched(p); |
c24d20db IM |
521 | |
522 | cpu = task_cpu(p); | |
523 | if (cpu == smp_processor_id()) | |
524 | return; | |
525 | ||
526 | /* NEED_RESCHED must be visible before we test polling */ | |
527 | smp_mb(); | |
528 | if (!tsk_is_polling(p)) | |
529 | smp_send_reschedule(cpu); | |
530 | } | |
531 | ||
029632fb | 532 | void resched_cpu(int cpu) |
c24d20db IM |
533 | { |
534 | struct rq *rq = cpu_rq(cpu); | |
535 | unsigned long flags; | |
536 | ||
05fa785c | 537 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
538 | return; |
539 | resched_task(cpu_curr(cpu)); | |
05fa785c | 540 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 541 | } |
06d8308c TG |
542 | |
543 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
544 | /* |
545 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
546 | * from an idle cpu. This is good for power-savings. | |
547 | * | |
548 | * We don't do similar optimization for completely idle system, as | |
549 | * selecting an idle cpu will add more delays to the timers than intended | |
550 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
551 | */ | |
552 | int get_nohz_timer_target(void) | |
553 | { | |
554 | int cpu = smp_processor_id(); | |
555 | int i; | |
556 | struct sched_domain *sd; | |
557 | ||
057f3fad | 558 | rcu_read_lock(); |
83cd4fe2 | 559 | for_each_domain(cpu, sd) { |
057f3fad PZ |
560 | for_each_cpu(i, sched_domain_span(sd)) { |
561 | if (!idle_cpu(i)) { | |
562 | cpu = i; | |
563 | goto unlock; | |
564 | } | |
565 | } | |
83cd4fe2 | 566 | } |
057f3fad PZ |
567 | unlock: |
568 | rcu_read_unlock(); | |
83cd4fe2 VP |
569 | return cpu; |
570 | } | |
06d8308c TG |
571 | /* |
572 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
573 | * idle CPU then this timer might expire before the next timer event | |
574 | * which is scheduled to wake up that CPU. In case of a completely | |
575 | * idle system the next event might even be infinite time into the | |
576 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
577 | * leaves the inner idle loop so the newly added timer is taken into | |
578 | * account when the CPU goes back to idle and evaluates the timer | |
579 | * wheel for the next timer event. | |
580 | */ | |
581 | void wake_up_idle_cpu(int cpu) | |
582 | { | |
583 | struct rq *rq = cpu_rq(cpu); | |
584 | ||
585 | if (cpu == smp_processor_id()) | |
586 | return; | |
587 | ||
588 | /* | |
589 | * This is safe, as this function is called with the timer | |
590 | * wheel base lock of (cpu) held. When the CPU is on the way | |
591 | * to idle and has not yet set rq->curr to idle then it will | |
592 | * be serialized on the timer wheel base lock and take the new | |
593 | * timer into account automatically. | |
594 | */ | |
595 | if (rq->curr != rq->idle) | |
596 | return; | |
45bf76df | 597 | |
45bf76df | 598 | /* |
06d8308c TG |
599 | * We can set TIF_RESCHED on the idle task of the other CPU |
600 | * lockless. The worst case is that the other CPU runs the | |
601 | * idle task through an additional NOOP schedule() | |
45bf76df | 602 | */ |
5ed0cec0 | 603 | set_tsk_need_resched(rq->idle); |
45bf76df | 604 | |
06d8308c TG |
605 | /* NEED_RESCHED must be visible before we test polling */ |
606 | smp_mb(); | |
607 | if (!tsk_is_polling(rq->idle)) | |
608 | smp_send_reschedule(cpu); | |
45bf76df IM |
609 | } |
610 | ||
ca38062e | 611 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 612 | { |
1c792db7 SS |
613 | int cpu = smp_processor_id(); |
614 | return idle_cpu(cpu) && test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
45bf76df IM |
615 | } |
616 | ||
ca38062e | 617 | #else /* CONFIG_NO_HZ */ |
45bf76df | 618 | |
ca38062e | 619 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 620 | { |
ca38062e | 621 | return false; |
2069dd75 PZ |
622 | } |
623 | ||
6d6bc0ad | 624 | #endif /* CONFIG_NO_HZ */ |
d842de87 | 625 | |
029632fb | 626 | void sched_avg_update(struct rq *rq) |
18d95a28 | 627 | { |
e9e9250b PZ |
628 | s64 period = sched_avg_period(); |
629 | ||
630 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
631 | /* |
632 | * Inline assembly required to prevent the compiler | |
633 | * optimising this loop into a divmod call. | |
634 | * See __iter_div_u64_rem() for another example of this. | |
635 | */ | |
636 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
637 | rq->age_stamp += period; |
638 | rq->rt_avg /= 2; | |
639 | } | |
18d95a28 PZ |
640 | } |
641 | ||
6d6bc0ad | 642 | #else /* !CONFIG_SMP */ |
029632fb | 643 | void resched_task(struct task_struct *p) |
18d95a28 | 644 | { |
05fa785c | 645 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 646 | set_tsk_need_resched(p); |
18d95a28 | 647 | } |
6d6bc0ad | 648 | #endif /* CONFIG_SMP */ |
18d95a28 | 649 | |
a790de99 PT |
650 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
651 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 652 | /* |
8277434e PT |
653 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
654 | * node and @up when leaving it for the final time. | |
655 | * | |
656 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 657 | */ |
029632fb | 658 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 659 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
660 | { |
661 | struct task_group *parent, *child; | |
eb755805 | 662 | int ret; |
c09595f6 | 663 | |
8277434e PT |
664 | parent = from; |
665 | ||
c09595f6 | 666 | down: |
eb755805 PZ |
667 | ret = (*down)(parent, data); |
668 | if (ret) | |
8277434e | 669 | goto out; |
c09595f6 PZ |
670 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
671 | parent = child; | |
672 | goto down; | |
673 | ||
674 | up: | |
675 | continue; | |
676 | } | |
eb755805 | 677 | ret = (*up)(parent, data); |
8277434e PT |
678 | if (ret || parent == from) |
679 | goto out; | |
c09595f6 PZ |
680 | |
681 | child = parent; | |
682 | parent = parent->parent; | |
683 | if (parent) | |
684 | goto up; | |
8277434e | 685 | out: |
eb755805 | 686 | return ret; |
c09595f6 PZ |
687 | } |
688 | ||
029632fb | 689 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 690 | { |
e2b245f8 | 691 | return 0; |
eb755805 | 692 | } |
18d95a28 PZ |
693 | #endif |
694 | ||
029632fb | 695 | void update_cpu_load(struct rq *this_rq); |
9c217245 | 696 | |
45bf76df IM |
697 | static void set_load_weight(struct task_struct *p) |
698 | { | |
f05998d4 NR |
699 | int prio = p->static_prio - MAX_RT_PRIO; |
700 | struct load_weight *load = &p->se.load; | |
701 | ||
dd41f596 IM |
702 | /* |
703 | * SCHED_IDLE tasks get minimal weight: | |
704 | */ | |
705 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 706 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 707 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
708 | return; |
709 | } | |
71f8bd46 | 710 | |
c8b28116 | 711 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 712 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
713 | } |
714 | ||
371fd7e7 | 715 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 716 | { |
a64692a3 | 717 | update_rq_clock(rq); |
dd41f596 | 718 | sched_info_queued(p); |
371fd7e7 | 719 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
720 | } |
721 | ||
371fd7e7 | 722 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 723 | { |
a64692a3 | 724 | update_rq_clock(rq); |
46ac22ba | 725 | sched_info_dequeued(p); |
371fd7e7 | 726 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
727 | } |
728 | ||
029632fb | 729 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
730 | { |
731 | if (task_contributes_to_load(p)) | |
732 | rq->nr_uninterruptible--; | |
733 | ||
371fd7e7 | 734 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
735 | } |
736 | ||
029632fb | 737 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
738 | { |
739 | if (task_contributes_to_load(p)) | |
740 | rq->nr_uninterruptible++; | |
741 | ||
371fd7e7 | 742 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
743 | } |
744 | ||
b52bfee4 VP |
745 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
746 | ||
305e6835 VP |
747 | /* |
748 | * There are no locks covering percpu hardirq/softirq time. | |
749 | * They are only modified in account_system_vtime, on corresponding CPU | |
750 | * with interrupts disabled. So, writes are safe. | |
751 | * They are read and saved off onto struct rq in update_rq_clock(). | |
752 | * This may result in other CPU reading this CPU's irq time and can | |
753 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
754 | * or new value with a side effect of accounting a slice of irq time to wrong |
755 | * task when irq is in progress while we read rq->clock. That is a worthy | |
756 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 757 | */ |
b52bfee4 VP |
758 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
759 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
760 | ||
761 | static DEFINE_PER_CPU(u64, irq_start_time); | |
762 | static int sched_clock_irqtime; | |
763 | ||
764 | void enable_sched_clock_irqtime(void) | |
765 | { | |
766 | sched_clock_irqtime = 1; | |
767 | } | |
768 | ||
769 | void disable_sched_clock_irqtime(void) | |
770 | { | |
771 | sched_clock_irqtime = 0; | |
772 | } | |
773 | ||
8e92c201 PZ |
774 | #ifndef CONFIG_64BIT |
775 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
776 | ||
777 | static inline void irq_time_write_begin(void) | |
778 | { | |
779 | __this_cpu_inc(irq_time_seq.sequence); | |
780 | smp_wmb(); | |
781 | } | |
782 | ||
783 | static inline void irq_time_write_end(void) | |
784 | { | |
785 | smp_wmb(); | |
786 | __this_cpu_inc(irq_time_seq.sequence); | |
787 | } | |
788 | ||
789 | static inline u64 irq_time_read(int cpu) | |
790 | { | |
791 | u64 irq_time; | |
792 | unsigned seq; | |
793 | ||
794 | do { | |
795 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
796 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
797 | per_cpu(cpu_hardirq_time, cpu); | |
798 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
799 | ||
800 | return irq_time; | |
801 | } | |
802 | #else /* CONFIG_64BIT */ | |
803 | static inline void irq_time_write_begin(void) | |
804 | { | |
805 | } | |
806 | ||
807 | static inline void irq_time_write_end(void) | |
808 | { | |
809 | } | |
810 | ||
811 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 812 | { |
305e6835 VP |
813 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
814 | } | |
8e92c201 | 815 | #endif /* CONFIG_64BIT */ |
305e6835 | 816 | |
fe44d621 PZ |
817 | /* |
818 | * Called before incrementing preempt_count on {soft,}irq_enter | |
819 | * and before decrementing preempt_count on {soft,}irq_exit. | |
820 | */ | |
b52bfee4 VP |
821 | void account_system_vtime(struct task_struct *curr) |
822 | { | |
823 | unsigned long flags; | |
fe44d621 | 824 | s64 delta; |
b52bfee4 | 825 | int cpu; |
b52bfee4 VP |
826 | |
827 | if (!sched_clock_irqtime) | |
828 | return; | |
829 | ||
830 | local_irq_save(flags); | |
831 | ||
b52bfee4 | 832 | cpu = smp_processor_id(); |
fe44d621 PZ |
833 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
834 | __this_cpu_add(irq_start_time, delta); | |
835 | ||
8e92c201 | 836 | irq_time_write_begin(); |
b52bfee4 VP |
837 | /* |
838 | * We do not account for softirq time from ksoftirqd here. | |
839 | * We want to continue accounting softirq time to ksoftirqd thread | |
840 | * in that case, so as not to confuse scheduler with a special task | |
841 | * that do not consume any time, but still wants to run. | |
842 | */ | |
843 | if (hardirq_count()) | |
fe44d621 | 844 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 845 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 846 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 847 | |
8e92c201 | 848 | irq_time_write_end(); |
b52bfee4 VP |
849 | local_irq_restore(flags); |
850 | } | |
b7dadc38 | 851 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 852 | |
e6e6685a GC |
853 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
854 | ||
855 | #ifdef CONFIG_PARAVIRT | |
856 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 857 | { |
e6e6685a GC |
858 | if (unlikely(steal > NSEC_PER_SEC)) |
859 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 860 | |
e6e6685a GC |
861 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
862 | } | |
863 | #endif | |
864 | ||
fe44d621 | 865 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 866 | { |
095c0aa8 GC |
867 | /* |
868 | * In theory, the compile should just see 0 here, and optimize out the call | |
869 | * to sched_rt_avg_update. But I don't trust it... | |
870 | */ | |
871 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
872 | s64 steal = 0, irq_delta = 0; | |
873 | #endif | |
874 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 875 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
876 | |
877 | /* | |
878 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
879 | * this case when a previous update_rq_clock() happened inside a | |
880 | * {soft,}irq region. | |
881 | * | |
882 | * When this happens, we stop ->clock_task and only update the | |
883 | * prev_irq_time stamp to account for the part that fit, so that a next | |
884 | * update will consume the rest. This ensures ->clock_task is | |
885 | * monotonic. | |
886 | * | |
887 | * It does however cause some slight miss-attribution of {soft,}irq | |
888 | * time, a more accurate solution would be to update the irq_time using | |
889 | * the current rq->clock timestamp, except that would require using | |
890 | * atomic ops. | |
891 | */ | |
892 | if (irq_delta > delta) | |
893 | irq_delta = delta; | |
894 | ||
895 | rq->prev_irq_time += irq_delta; | |
896 | delta -= irq_delta; | |
095c0aa8 GC |
897 | #endif |
898 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 899 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
900 | u64 st; |
901 | ||
902 | steal = paravirt_steal_clock(cpu_of(rq)); | |
903 | steal -= rq->prev_steal_time_rq; | |
904 | ||
905 | if (unlikely(steal > delta)) | |
906 | steal = delta; | |
907 | ||
908 | st = steal_ticks(steal); | |
909 | steal = st * TICK_NSEC; | |
910 | ||
911 | rq->prev_steal_time_rq += steal; | |
912 | ||
913 | delta -= steal; | |
914 | } | |
915 | #endif | |
916 | ||
fe44d621 PZ |
917 | rq->clock_task += delta; |
918 | ||
095c0aa8 GC |
919 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
920 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
921 | sched_rt_avg_update(rq, irq_delta + steal); | |
922 | #endif | |
aa483808 VP |
923 | } |
924 | ||
095c0aa8 | 925 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
926 | static int irqtime_account_hi_update(void) |
927 | { | |
3292beb3 | 928 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
929 | unsigned long flags; |
930 | u64 latest_ns; | |
931 | int ret = 0; | |
932 | ||
933 | local_irq_save(flags); | |
934 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
612ef28a | 935 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) |
abb74cef VP |
936 | ret = 1; |
937 | local_irq_restore(flags); | |
938 | return ret; | |
939 | } | |
940 | ||
941 | static int irqtime_account_si_update(void) | |
942 | { | |
3292beb3 | 943 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
944 | unsigned long flags; |
945 | u64 latest_ns; | |
946 | int ret = 0; | |
947 | ||
948 | local_irq_save(flags); | |
949 | latest_ns = this_cpu_read(cpu_softirq_time); | |
612ef28a | 950 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) |
abb74cef VP |
951 | ret = 1; |
952 | local_irq_restore(flags); | |
953 | return ret; | |
954 | } | |
955 | ||
fe44d621 | 956 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 957 | |
abb74cef VP |
958 | #define sched_clock_irqtime (0) |
959 | ||
095c0aa8 | 960 | #endif |
b52bfee4 | 961 | |
34f971f6 PZ |
962 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
963 | { | |
964 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
965 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
966 | ||
967 | if (stop) { | |
968 | /* | |
969 | * Make it appear like a SCHED_FIFO task, its something | |
970 | * userspace knows about and won't get confused about. | |
971 | * | |
972 | * Also, it will make PI more or less work without too | |
973 | * much confusion -- but then, stop work should not | |
974 | * rely on PI working anyway. | |
975 | */ | |
976 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
977 | ||
978 | stop->sched_class = &stop_sched_class; | |
979 | } | |
980 | ||
981 | cpu_rq(cpu)->stop = stop; | |
982 | ||
983 | if (old_stop) { | |
984 | /* | |
985 | * Reset it back to a normal scheduling class so that | |
986 | * it can die in pieces. | |
987 | */ | |
988 | old_stop->sched_class = &rt_sched_class; | |
989 | } | |
990 | } | |
991 | ||
14531189 | 992 | /* |
dd41f596 | 993 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 994 | */ |
14531189 IM |
995 | static inline int __normal_prio(struct task_struct *p) |
996 | { | |
dd41f596 | 997 | return p->static_prio; |
14531189 IM |
998 | } |
999 | ||
b29739f9 IM |
1000 | /* |
1001 | * Calculate the expected normal priority: i.e. priority | |
1002 | * without taking RT-inheritance into account. Might be | |
1003 | * boosted by interactivity modifiers. Changes upon fork, | |
1004 | * setprio syscalls, and whenever the interactivity | |
1005 | * estimator recalculates. | |
1006 | */ | |
36c8b586 | 1007 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1008 | { |
1009 | int prio; | |
1010 | ||
e05606d3 | 1011 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1012 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1013 | else | |
1014 | prio = __normal_prio(p); | |
1015 | return prio; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Calculate the current priority, i.e. the priority | |
1020 | * taken into account by the scheduler. This value might | |
1021 | * be boosted by RT tasks, or might be boosted by | |
1022 | * interactivity modifiers. Will be RT if the task got | |
1023 | * RT-boosted. If not then it returns p->normal_prio. | |
1024 | */ | |
36c8b586 | 1025 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1026 | { |
1027 | p->normal_prio = normal_prio(p); | |
1028 | /* | |
1029 | * If we are RT tasks or we were boosted to RT priority, | |
1030 | * keep the priority unchanged. Otherwise, update priority | |
1031 | * to the normal priority: | |
1032 | */ | |
1033 | if (!rt_prio(p->prio)) | |
1034 | return p->normal_prio; | |
1035 | return p->prio; | |
1036 | } | |
1037 | ||
1da177e4 LT |
1038 | /** |
1039 | * task_curr - is this task currently executing on a CPU? | |
1040 | * @p: the task in question. | |
1041 | */ | |
36c8b586 | 1042 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1043 | { |
1044 | return cpu_curr(task_cpu(p)) == p; | |
1045 | } | |
1046 | ||
cb469845 SR |
1047 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1048 | const struct sched_class *prev_class, | |
da7a735e | 1049 | int oldprio) |
cb469845 SR |
1050 | { |
1051 | if (prev_class != p->sched_class) { | |
1052 | if (prev_class->switched_from) | |
da7a735e PZ |
1053 | prev_class->switched_from(rq, p); |
1054 | p->sched_class->switched_to(rq, p); | |
1055 | } else if (oldprio != p->prio) | |
1056 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
1057 | } |
1058 | ||
029632fb | 1059 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1060 | { |
1061 | const struct sched_class *class; | |
1062 | ||
1063 | if (p->sched_class == rq->curr->sched_class) { | |
1064 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1065 | } else { | |
1066 | for_each_class(class) { | |
1067 | if (class == rq->curr->sched_class) | |
1068 | break; | |
1069 | if (class == p->sched_class) { | |
1070 | resched_task(rq->curr); | |
1071 | break; | |
1072 | } | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * A queue event has occurred, and we're going to schedule. In | |
1078 | * this case, we can save a useless back to back clock update. | |
1079 | */ | |
fd2f4419 | 1080 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1081 | rq->skip_clock_update = 1; |
1082 | } | |
1083 | ||
1da177e4 | 1084 | #ifdef CONFIG_SMP |
dd41f596 | 1085 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1086 | { |
e2912009 PZ |
1087 | #ifdef CONFIG_SCHED_DEBUG |
1088 | /* | |
1089 | * We should never call set_task_cpu() on a blocked task, | |
1090 | * ttwu() will sort out the placement. | |
1091 | */ | |
077614ee PZ |
1092 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
1093 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
1094 | |
1095 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1096 | /* |
1097 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1098 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1099 | * | |
1100 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
1101 | * see set_task_rq(). | |
1102 | * | |
1103 | * Furthermore, all task_rq users should acquire both locks, see | |
1104 | * task_rq_lock(). | |
1105 | */ | |
0122ec5b PZ |
1106 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1107 | lockdep_is_held(&task_rq(p)->lock))); | |
1108 | #endif | |
e2912009 PZ |
1109 | #endif |
1110 | ||
de1d7286 | 1111 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1112 | |
0c69774e PZ |
1113 | if (task_cpu(p) != new_cpu) { |
1114 | p->se.nr_migrations++; | |
a8b0ca17 | 1115 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1116 | } |
dd41f596 IM |
1117 | |
1118 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1119 | } |
1120 | ||
969c7921 | 1121 | struct migration_arg { |
36c8b586 | 1122 | struct task_struct *task; |
1da177e4 | 1123 | int dest_cpu; |
70b97a7f | 1124 | }; |
1da177e4 | 1125 | |
969c7921 TH |
1126 | static int migration_cpu_stop(void *data); |
1127 | ||
1da177e4 LT |
1128 | /* |
1129 | * wait_task_inactive - wait for a thread to unschedule. | |
1130 | * | |
85ba2d86 RM |
1131 | * If @match_state is nonzero, it's the @p->state value just checked and |
1132 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1133 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1134 | * we return a positive number (its total switch count). If a second call | |
1135 | * a short while later returns the same number, the caller can be sure that | |
1136 | * @p has remained unscheduled the whole time. | |
1137 | * | |
1da177e4 LT |
1138 | * The caller must ensure that the task *will* unschedule sometime soon, |
1139 | * else this function might spin for a *long* time. This function can't | |
1140 | * be called with interrupts off, or it may introduce deadlock with | |
1141 | * smp_call_function() if an IPI is sent by the same process we are | |
1142 | * waiting to become inactive. | |
1143 | */ | |
85ba2d86 | 1144 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1145 | { |
1146 | unsigned long flags; | |
dd41f596 | 1147 | int running, on_rq; |
85ba2d86 | 1148 | unsigned long ncsw; |
70b97a7f | 1149 | struct rq *rq; |
1da177e4 | 1150 | |
3a5c359a AK |
1151 | for (;;) { |
1152 | /* | |
1153 | * We do the initial early heuristics without holding | |
1154 | * any task-queue locks at all. We'll only try to get | |
1155 | * the runqueue lock when things look like they will | |
1156 | * work out! | |
1157 | */ | |
1158 | rq = task_rq(p); | |
fa490cfd | 1159 | |
3a5c359a AK |
1160 | /* |
1161 | * If the task is actively running on another CPU | |
1162 | * still, just relax and busy-wait without holding | |
1163 | * any locks. | |
1164 | * | |
1165 | * NOTE! Since we don't hold any locks, it's not | |
1166 | * even sure that "rq" stays as the right runqueue! | |
1167 | * But we don't care, since "task_running()" will | |
1168 | * return false if the runqueue has changed and p | |
1169 | * is actually now running somewhere else! | |
1170 | */ | |
85ba2d86 RM |
1171 | while (task_running(rq, p)) { |
1172 | if (match_state && unlikely(p->state != match_state)) | |
1173 | return 0; | |
3a5c359a | 1174 | cpu_relax(); |
85ba2d86 | 1175 | } |
fa490cfd | 1176 | |
3a5c359a AK |
1177 | /* |
1178 | * Ok, time to look more closely! We need the rq | |
1179 | * lock now, to be *sure*. If we're wrong, we'll | |
1180 | * just go back and repeat. | |
1181 | */ | |
1182 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1183 | trace_sched_wait_task(p); |
3a5c359a | 1184 | running = task_running(rq, p); |
fd2f4419 | 1185 | on_rq = p->on_rq; |
85ba2d86 | 1186 | ncsw = 0; |
f31e11d8 | 1187 | if (!match_state || p->state == match_state) |
93dcf55f | 1188 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1189 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1190 | |
85ba2d86 RM |
1191 | /* |
1192 | * If it changed from the expected state, bail out now. | |
1193 | */ | |
1194 | if (unlikely(!ncsw)) | |
1195 | break; | |
1196 | ||
3a5c359a AK |
1197 | /* |
1198 | * Was it really running after all now that we | |
1199 | * checked with the proper locks actually held? | |
1200 | * | |
1201 | * Oops. Go back and try again.. | |
1202 | */ | |
1203 | if (unlikely(running)) { | |
1204 | cpu_relax(); | |
1205 | continue; | |
1206 | } | |
fa490cfd | 1207 | |
3a5c359a AK |
1208 | /* |
1209 | * It's not enough that it's not actively running, | |
1210 | * it must be off the runqueue _entirely_, and not | |
1211 | * preempted! | |
1212 | * | |
80dd99b3 | 1213 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1214 | * running right now), it's preempted, and we should |
1215 | * yield - it could be a while. | |
1216 | */ | |
1217 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1218 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1219 | ||
1220 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1221 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1222 | continue; |
1223 | } | |
fa490cfd | 1224 | |
3a5c359a AK |
1225 | /* |
1226 | * Ahh, all good. It wasn't running, and it wasn't | |
1227 | * runnable, which means that it will never become | |
1228 | * running in the future either. We're all done! | |
1229 | */ | |
1230 | break; | |
1231 | } | |
85ba2d86 RM |
1232 | |
1233 | return ncsw; | |
1da177e4 LT |
1234 | } |
1235 | ||
1236 | /*** | |
1237 | * kick_process - kick a running thread to enter/exit the kernel | |
1238 | * @p: the to-be-kicked thread | |
1239 | * | |
1240 | * Cause a process which is running on another CPU to enter | |
1241 | * kernel-mode, without any delay. (to get signals handled.) | |
1242 | * | |
25985edc | 1243 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1244 | * because all it wants to ensure is that the remote task enters |
1245 | * the kernel. If the IPI races and the task has been migrated | |
1246 | * to another CPU then no harm is done and the purpose has been | |
1247 | * achieved as well. | |
1248 | */ | |
36c8b586 | 1249 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1250 | { |
1251 | int cpu; | |
1252 | ||
1253 | preempt_disable(); | |
1254 | cpu = task_cpu(p); | |
1255 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1256 | smp_send_reschedule(cpu); | |
1257 | preempt_enable(); | |
1258 | } | |
b43e3521 | 1259 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1260 | #endif /* CONFIG_SMP */ |
1da177e4 | 1261 | |
970b13ba | 1262 | #ifdef CONFIG_SMP |
30da688e | 1263 | /* |
013fdb80 | 1264 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1265 | */ |
5da9a0fb PZ |
1266 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1267 | { | |
5da9a0fb | 1268 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); |
2baab4e9 PZ |
1269 | enum { cpuset, possible, fail } state = cpuset; |
1270 | int dest_cpu; | |
5da9a0fb PZ |
1271 | |
1272 | /* Look for allowed, online CPU in same node. */ | |
e3831edd | 1273 | for_each_cpu(dest_cpu, nodemask) { |
2baab4e9 PZ |
1274 | if (!cpu_online(dest_cpu)) |
1275 | continue; | |
1276 | if (!cpu_active(dest_cpu)) | |
1277 | continue; | |
fa17b507 | 1278 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
5da9a0fb | 1279 | return dest_cpu; |
2baab4e9 | 1280 | } |
5da9a0fb | 1281 | |
2baab4e9 PZ |
1282 | for (;;) { |
1283 | /* Any allowed, online CPU? */ | |
e3831edd | 1284 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1285 | if (!cpu_online(dest_cpu)) |
1286 | continue; | |
1287 | if (!cpu_active(dest_cpu)) | |
1288 | continue; | |
1289 | goto out; | |
1290 | } | |
5da9a0fb | 1291 | |
2baab4e9 PZ |
1292 | switch (state) { |
1293 | case cpuset: | |
1294 | /* No more Mr. Nice Guy. */ | |
1295 | cpuset_cpus_allowed_fallback(p); | |
1296 | state = possible; | |
1297 | break; | |
1298 | ||
1299 | case possible: | |
1300 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1301 | state = fail; | |
1302 | break; | |
1303 | ||
1304 | case fail: | |
1305 | BUG(); | |
1306 | break; | |
1307 | } | |
1308 | } | |
1309 | ||
1310 | out: | |
1311 | if (state != cpuset) { | |
1312 | /* | |
1313 | * Don't tell them about moving exiting tasks or | |
1314 | * kernel threads (both mm NULL), since they never | |
1315 | * leave kernel. | |
1316 | */ | |
1317 | if (p->mm && printk_ratelimit()) { | |
1318 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1319 | task_pid_nr(p), p->comm, cpu); | |
1320 | } | |
5da9a0fb PZ |
1321 | } |
1322 | ||
1323 | return dest_cpu; | |
1324 | } | |
1325 | ||
e2912009 | 1326 | /* |
013fdb80 | 1327 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1328 | */ |
970b13ba | 1329 | static inline |
7608dec2 | 1330 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1331 | { |
7608dec2 | 1332 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1333 | |
1334 | /* | |
1335 | * In order not to call set_task_cpu() on a blocking task we need | |
1336 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1337 | * cpu. | |
1338 | * | |
1339 | * Since this is common to all placement strategies, this lives here. | |
1340 | * | |
1341 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1342 | * not worry about this generic constraint ] | |
1343 | */ | |
fa17b507 | 1344 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1345 | !cpu_online(cpu))) |
5da9a0fb | 1346 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1347 | |
1348 | return cpu; | |
970b13ba | 1349 | } |
09a40af5 MG |
1350 | |
1351 | static void update_avg(u64 *avg, u64 sample) | |
1352 | { | |
1353 | s64 diff = sample - *avg; | |
1354 | *avg += diff >> 3; | |
1355 | } | |
970b13ba PZ |
1356 | #endif |
1357 | ||
d7c01d27 | 1358 | static void |
b84cb5df | 1359 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1360 | { |
d7c01d27 | 1361 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1362 | struct rq *rq = this_rq(); |
1363 | ||
d7c01d27 PZ |
1364 | #ifdef CONFIG_SMP |
1365 | int this_cpu = smp_processor_id(); | |
1366 | ||
1367 | if (cpu == this_cpu) { | |
1368 | schedstat_inc(rq, ttwu_local); | |
1369 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1370 | } else { | |
1371 | struct sched_domain *sd; | |
1372 | ||
1373 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1374 | rcu_read_lock(); |
d7c01d27 PZ |
1375 | for_each_domain(this_cpu, sd) { |
1376 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1377 | schedstat_inc(sd, ttwu_wake_remote); | |
1378 | break; | |
1379 | } | |
1380 | } | |
057f3fad | 1381 | rcu_read_unlock(); |
d7c01d27 | 1382 | } |
f339b9dc PZ |
1383 | |
1384 | if (wake_flags & WF_MIGRATED) | |
1385 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1386 | ||
d7c01d27 PZ |
1387 | #endif /* CONFIG_SMP */ |
1388 | ||
1389 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1390 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1391 | |
1392 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1393 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1394 | |
d7c01d27 PZ |
1395 | #endif /* CONFIG_SCHEDSTATS */ |
1396 | } | |
1397 | ||
1398 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1399 | { | |
9ed3811a | 1400 | activate_task(rq, p, en_flags); |
fd2f4419 | 1401 | p->on_rq = 1; |
c2f7115e PZ |
1402 | |
1403 | /* if a worker is waking up, notify workqueue */ | |
1404 | if (p->flags & PF_WQ_WORKER) | |
1405 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1406 | } |
1407 | ||
23f41eeb PZ |
1408 | /* |
1409 | * Mark the task runnable and perform wakeup-preemption. | |
1410 | */ | |
89363381 | 1411 | static void |
23f41eeb | 1412 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1413 | { |
89363381 | 1414 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1415 | check_preempt_curr(rq, p, wake_flags); |
1416 | ||
1417 | p->state = TASK_RUNNING; | |
1418 | #ifdef CONFIG_SMP | |
1419 | if (p->sched_class->task_woken) | |
1420 | p->sched_class->task_woken(rq, p); | |
1421 | ||
e69c6341 | 1422 | if (rq->idle_stamp) { |
9ed3811a TH |
1423 | u64 delta = rq->clock - rq->idle_stamp; |
1424 | u64 max = 2*sysctl_sched_migration_cost; | |
1425 | ||
1426 | if (delta > max) | |
1427 | rq->avg_idle = max; | |
1428 | else | |
1429 | update_avg(&rq->avg_idle, delta); | |
1430 | rq->idle_stamp = 0; | |
1431 | } | |
1432 | #endif | |
1433 | } | |
1434 | ||
c05fbafb PZ |
1435 | static void |
1436 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1437 | { | |
1438 | #ifdef CONFIG_SMP | |
1439 | if (p->sched_contributes_to_load) | |
1440 | rq->nr_uninterruptible--; | |
1441 | #endif | |
1442 | ||
1443 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1444 | ttwu_do_wakeup(rq, p, wake_flags); | |
1445 | } | |
1446 | ||
1447 | /* | |
1448 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1449 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1450 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1451 | * the task is still ->on_rq. | |
1452 | */ | |
1453 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1454 | { | |
1455 | struct rq *rq; | |
1456 | int ret = 0; | |
1457 | ||
1458 | rq = __task_rq_lock(p); | |
1459 | if (p->on_rq) { | |
1460 | ttwu_do_wakeup(rq, p, wake_flags); | |
1461 | ret = 1; | |
1462 | } | |
1463 | __task_rq_unlock(rq); | |
1464 | ||
1465 | return ret; | |
1466 | } | |
1467 | ||
317f3941 | 1468 | #ifdef CONFIG_SMP |
fa14ff4a | 1469 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1470 | { |
1471 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1472 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1473 | struct task_struct *p; | |
317f3941 PZ |
1474 | |
1475 | raw_spin_lock(&rq->lock); | |
1476 | ||
fa14ff4a PZ |
1477 | while (llist) { |
1478 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1479 | llist = llist_next(llist); | |
317f3941 PZ |
1480 | ttwu_do_activate(rq, p, 0); |
1481 | } | |
1482 | ||
1483 | raw_spin_unlock(&rq->lock); | |
1484 | } | |
1485 | ||
1486 | void scheduler_ipi(void) | |
1487 | { | |
ca38062e | 1488 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1489 | return; |
1490 | ||
1491 | /* | |
1492 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1493 | * traditionally all their work was done from the interrupt return | |
1494 | * path. Now that we actually do some work, we need to make sure | |
1495 | * we do call them. | |
1496 | * | |
1497 | * Some archs already do call them, luckily irq_enter/exit nest | |
1498 | * properly. | |
1499 | * | |
1500 | * Arguably we should visit all archs and update all handlers, | |
1501 | * however a fair share of IPIs are still resched only so this would | |
1502 | * somewhat pessimize the simple resched case. | |
1503 | */ | |
1504 | irq_enter(); | |
fa14ff4a | 1505 | sched_ttwu_pending(); |
ca38062e SS |
1506 | |
1507 | /* | |
1508 | * Check if someone kicked us for doing the nohz idle load balance. | |
1509 | */ | |
6eb57e0d SS |
1510 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1511 | this_rq()->idle_balance = 1; | |
ca38062e | 1512 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1513 | } |
c5d753a5 | 1514 | irq_exit(); |
317f3941 PZ |
1515 | } |
1516 | ||
1517 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1518 | { | |
fa14ff4a | 1519 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1520 | smp_send_reschedule(cpu); |
1521 | } | |
d6aa8f85 PZ |
1522 | |
1523 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1524 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
1525 | { | |
1526 | struct rq *rq; | |
1527 | int ret = 0; | |
1528 | ||
1529 | rq = __task_rq_lock(p); | |
1530 | if (p->on_cpu) { | |
1531 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
1532 | ttwu_do_wakeup(rq, p, wake_flags); | |
1533 | ret = 1; | |
1534 | } | |
1535 | __task_rq_unlock(rq); | |
1536 | ||
1537 | return ret; | |
1538 | ||
1539 | } | |
1540 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
518cd623 | 1541 | |
39be3501 | 1542 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1543 | { |
1544 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1545 | } | |
d6aa8f85 | 1546 | #endif /* CONFIG_SMP */ |
317f3941 | 1547 | |
c05fbafb PZ |
1548 | static void ttwu_queue(struct task_struct *p, int cpu) |
1549 | { | |
1550 | struct rq *rq = cpu_rq(cpu); | |
1551 | ||
17d9f311 | 1552 | #if defined(CONFIG_SMP) |
39be3501 | 1553 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1554 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1555 | ttwu_queue_remote(p, cpu); |
1556 | return; | |
1557 | } | |
1558 | #endif | |
1559 | ||
c05fbafb PZ |
1560 | raw_spin_lock(&rq->lock); |
1561 | ttwu_do_activate(rq, p, 0); | |
1562 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1563 | } |
1564 | ||
1565 | /** | |
1da177e4 | 1566 | * try_to_wake_up - wake up a thread |
9ed3811a | 1567 | * @p: the thread to be awakened |
1da177e4 | 1568 | * @state: the mask of task states that can be woken |
9ed3811a | 1569 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1570 | * |
1571 | * Put it on the run-queue if it's not already there. The "current" | |
1572 | * thread is always on the run-queue (except when the actual | |
1573 | * re-schedule is in progress), and as such you're allowed to do | |
1574 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1575 | * runnable without the overhead of this. | |
1576 | * | |
9ed3811a TH |
1577 | * Returns %true if @p was woken up, %false if it was already running |
1578 | * or @state didn't match @p's state. | |
1da177e4 | 1579 | */ |
e4a52bcb PZ |
1580 | static int |
1581 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1582 | { |
1da177e4 | 1583 | unsigned long flags; |
c05fbafb | 1584 | int cpu, success = 0; |
2398f2c6 | 1585 | |
04e2f174 | 1586 | smp_wmb(); |
013fdb80 | 1587 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1588 | if (!(p->state & state)) |
1da177e4 LT |
1589 | goto out; |
1590 | ||
c05fbafb | 1591 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1592 | cpu = task_cpu(p); |
1da177e4 | 1593 | |
c05fbafb PZ |
1594 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1595 | goto stat; | |
1da177e4 | 1596 | |
1da177e4 | 1597 | #ifdef CONFIG_SMP |
e9c84311 | 1598 | /* |
c05fbafb PZ |
1599 | * If the owning (remote) cpu is still in the middle of schedule() with |
1600 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1601 | */ |
e4a52bcb PZ |
1602 | while (p->on_cpu) { |
1603 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1604 | /* | |
d6aa8f85 PZ |
1605 | * In case the architecture enables interrupts in |
1606 | * context_switch(), we cannot busy wait, since that | |
1607 | * would lead to deadlocks when an interrupt hits and | |
1608 | * tries to wake up @prev. So bail and do a complete | |
1609 | * remote wakeup. | |
e4a52bcb | 1610 | */ |
d6aa8f85 | 1611 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 1612 | goto stat; |
d6aa8f85 | 1613 | #else |
e4a52bcb | 1614 | cpu_relax(); |
d6aa8f85 | 1615 | #endif |
371fd7e7 | 1616 | } |
0970d299 | 1617 | /* |
e4a52bcb | 1618 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1619 | */ |
e4a52bcb | 1620 | smp_rmb(); |
1da177e4 | 1621 | |
a8e4f2ea | 1622 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1623 | p->state = TASK_WAKING; |
e7693a36 | 1624 | |
e4a52bcb | 1625 | if (p->sched_class->task_waking) |
74f8e4b2 | 1626 | p->sched_class->task_waking(p); |
efbbd05a | 1627 | |
7608dec2 | 1628 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1629 | if (task_cpu(p) != cpu) { |
1630 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1631 | set_task_cpu(p, cpu); |
f339b9dc | 1632 | } |
1da177e4 | 1633 | #endif /* CONFIG_SMP */ |
1da177e4 | 1634 | |
c05fbafb PZ |
1635 | ttwu_queue(p, cpu); |
1636 | stat: | |
b84cb5df | 1637 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1638 | out: |
013fdb80 | 1639 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1640 | |
1641 | return success; | |
1642 | } | |
1643 | ||
21aa9af0 TH |
1644 | /** |
1645 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1646 | * @p: the thread to be awakened | |
1647 | * | |
2acca55e | 1648 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1649 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1650 | * the current task. |
21aa9af0 TH |
1651 | */ |
1652 | static void try_to_wake_up_local(struct task_struct *p) | |
1653 | { | |
1654 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1655 | |
1656 | BUG_ON(rq != this_rq()); | |
1657 | BUG_ON(p == current); | |
1658 | lockdep_assert_held(&rq->lock); | |
1659 | ||
2acca55e PZ |
1660 | if (!raw_spin_trylock(&p->pi_lock)) { |
1661 | raw_spin_unlock(&rq->lock); | |
1662 | raw_spin_lock(&p->pi_lock); | |
1663 | raw_spin_lock(&rq->lock); | |
1664 | } | |
1665 | ||
21aa9af0 | 1666 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1667 | goto out; |
21aa9af0 | 1668 | |
fd2f4419 | 1669 | if (!p->on_rq) |
d7c01d27 PZ |
1670 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1671 | ||
23f41eeb | 1672 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1673 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1674 | out: |
1675 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1676 | } |
1677 | ||
50fa610a DH |
1678 | /** |
1679 | * wake_up_process - Wake up a specific process | |
1680 | * @p: The process to be woken up. | |
1681 | * | |
1682 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1683 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1684 | * running. | |
1685 | * | |
1686 | * It may be assumed that this function implies a write memory barrier before | |
1687 | * changing the task state if and only if any tasks are woken up. | |
1688 | */ | |
7ad5b3a5 | 1689 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1690 | { |
d9514f6c | 1691 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1692 | } |
1da177e4 LT |
1693 | EXPORT_SYMBOL(wake_up_process); |
1694 | ||
7ad5b3a5 | 1695 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1696 | { |
1697 | return try_to_wake_up(p, state, 0); | |
1698 | } | |
1699 | ||
1da177e4 LT |
1700 | /* |
1701 | * Perform scheduler related setup for a newly forked process p. | |
1702 | * p is forked by current. | |
dd41f596 IM |
1703 | * |
1704 | * __sched_fork() is basic setup used by init_idle() too: | |
1705 | */ | |
1706 | static void __sched_fork(struct task_struct *p) | |
1707 | { | |
fd2f4419 PZ |
1708 | p->on_rq = 0; |
1709 | ||
1710 | p->se.on_rq = 0; | |
dd41f596 IM |
1711 | p->se.exec_start = 0; |
1712 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1713 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1714 | p->se.nr_migrations = 0; |
da7a735e | 1715 | p->se.vruntime = 0; |
fd2f4419 | 1716 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1717 | |
1718 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1719 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1720 | #endif |
476d139c | 1721 | |
fa717060 | 1722 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1723 | |
e107be36 AK |
1724 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1725 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1726 | #endif | |
dd41f596 IM |
1727 | } |
1728 | ||
1729 | /* | |
1730 | * fork()/clone()-time setup: | |
1731 | */ | |
3e51e3ed | 1732 | void sched_fork(struct task_struct *p) |
dd41f596 | 1733 | { |
0122ec5b | 1734 | unsigned long flags; |
dd41f596 IM |
1735 | int cpu = get_cpu(); |
1736 | ||
1737 | __sched_fork(p); | |
06b83b5f | 1738 | /* |
0017d735 | 1739 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1740 | * nobody will actually run it, and a signal or other external |
1741 | * event cannot wake it up and insert it on the runqueue either. | |
1742 | */ | |
0017d735 | 1743 | p->state = TASK_RUNNING; |
dd41f596 | 1744 | |
c350a04e MG |
1745 | /* |
1746 | * Make sure we do not leak PI boosting priority to the child. | |
1747 | */ | |
1748 | p->prio = current->normal_prio; | |
1749 | ||
b9dc29e7 MG |
1750 | /* |
1751 | * Revert to default priority/policy on fork if requested. | |
1752 | */ | |
1753 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1754 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1755 | p->policy = SCHED_NORMAL; |
6c697bdf | 1756 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1757 | p->rt_priority = 0; |
1758 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1759 | p->static_prio = NICE_TO_PRIO(0); | |
1760 | ||
1761 | p->prio = p->normal_prio = __normal_prio(p); | |
1762 | set_load_weight(p); | |
6c697bdf | 1763 | |
b9dc29e7 MG |
1764 | /* |
1765 | * We don't need the reset flag anymore after the fork. It has | |
1766 | * fulfilled its duty: | |
1767 | */ | |
1768 | p->sched_reset_on_fork = 0; | |
1769 | } | |
ca94c442 | 1770 | |
2ddbf952 HS |
1771 | if (!rt_prio(p->prio)) |
1772 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1773 | |
cd29fe6f PZ |
1774 | if (p->sched_class->task_fork) |
1775 | p->sched_class->task_fork(p); | |
1776 | ||
86951599 PZ |
1777 | /* |
1778 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1779 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1780 | * is ran before sched_fork(). | |
1781 | * | |
1782 | * Silence PROVE_RCU. | |
1783 | */ | |
0122ec5b | 1784 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1785 | set_task_cpu(p, cpu); |
0122ec5b | 1786 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1787 | |
52f17b6c | 1788 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1789 | if (likely(sched_info_on())) |
52f17b6c | 1790 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1791 | #endif |
3ca7a440 PZ |
1792 | #if defined(CONFIG_SMP) |
1793 | p->on_cpu = 0; | |
4866cde0 | 1794 | #endif |
bdd4e85d | 1795 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1796 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1797 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1798 | #endif |
806c09a7 | 1799 | #ifdef CONFIG_SMP |
917b627d | 1800 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1801 | #endif |
917b627d | 1802 | |
476d139c | 1803 | put_cpu(); |
1da177e4 LT |
1804 | } |
1805 | ||
1806 | /* | |
1807 | * wake_up_new_task - wake up a newly created task for the first time. | |
1808 | * | |
1809 | * This function will do some initial scheduler statistics housekeeping | |
1810 | * that must be done for every newly created context, then puts the task | |
1811 | * on the runqueue and wakes it. | |
1812 | */ | |
3e51e3ed | 1813 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1814 | { |
1815 | unsigned long flags; | |
dd41f596 | 1816 | struct rq *rq; |
fabf318e | 1817 | |
ab2515c4 | 1818 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1819 | #ifdef CONFIG_SMP |
1820 | /* | |
1821 | * Fork balancing, do it here and not earlier because: | |
1822 | * - cpus_allowed can change in the fork path | |
1823 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1824 | */ |
ab2515c4 | 1825 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1826 | #endif |
1827 | ||
ab2515c4 | 1828 | rq = __task_rq_lock(p); |
cd29fe6f | 1829 | activate_task(rq, p, 0); |
fd2f4419 | 1830 | p->on_rq = 1; |
89363381 | 1831 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1832 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1833 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1834 | if (p->sched_class->task_woken) |
1835 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1836 | #endif |
0122ec5b | 1837 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1838 | } |
1839 | ||
e107be36 AK |
1840 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1841 | ||
1842 | /** | |
80dd99b3 | 1843 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1844 | * @notifier: notifier struct to register |
e107be36 AK |
1845 | */ |
1846 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1847 | { | |
1848 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1849 | } | |
1850 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1851 | ||
1852 | /** | |
1853 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1854 | * @notifier: notifier struct to unregister |
e107be36 AK |
1855 | * |
1856 | * This is safe to call from within a preemption notifier. | |
1857 | */ | |
1858 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1859 | { | |
1860 | hlist_del(¬ifier->link); | |
1861 | } | |
1862 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1863 | ||
1864 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1865 | { | |
1866 | struct preempt_notifier *notifier; | |
1867 | struct hlist_node *node; | |
1868 | ||
1869 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1870 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1871 | } | |
1872 | ||
1873 | static void | |
1874 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1875 | struct task_struct *next) | |
1876 | { | |
1877 | struct preempt_notifier *notifier; | |
1878 | struct hlist_node *node; | |
1879 | ||
1880 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1881 | notifier->ops->sched_out(notifier, next); | |
1882 | } | |
1883 | ||
6d6bc0ad | 1884 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1885 | |
1886 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1887 | { | |
1888 | } | |
1889 | ||
1890 | static void | |
1891 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1892 | struct task_struct *next) | |
1893 | { | |
1894 | } | |
1895 | ||
6d6bc0ad | 1896 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1897 | |
4866cde0 NP |
1898 | /** |
1899 | * prepare_task_switch - prepare to switch tasks | |
1900 | * @rq: the runqueue preparing to switch | |
421cee29 | 1901 | * @prev: the current task that is being switched out |
4866cde0 NP |
1902 | * @next: the task we are going to switch to. |
1903 | * | |
1904 | * This is called with the rq lock held and interrupts off. It must | |
1905 | * be paired with a subsequent finish_task_switch after the context | |
1906 | * switch. | |
1907 | * | |
1908 | * prepare_task_switch sets up locking and calls architecture specific | |
1909 | * hooks. | |
1910 | */ | |
e107be36 AK |
1911 | static inline void |
1912 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1913 | struct task_struct *next) | |
4866cde0 | 1914 | { |
fe4b04fa PZ |
1915 | sched_info_switch(prev, next); |
1916 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1917 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1918 | prepare_lock_switch(rq, next); |
1919 | prepare_arch_switch(next); | |
fe4b04fa | 1920 | trace_sched_switch(prev, next); |
4866cde0 NP |
1921 | } |
1922 | ||
1da177e4 LT |
1923 | /** |
1924 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1925 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1926 | * @prev: the thread we just switched away from. |
1927 | * | |
4866cde0 NP |
1928 | * finish_task_switch must be called after the context switch, paired |
1929 | * with a prepare_task_switch call before the context switch. | |
1930 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1931 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1932 | * |
1933 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1934 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1935 | * with the lock held can cause deadlocks; see schedule() for |
1936 | * details.) | |
1937 | */ | |
a9957449 | 1938 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1939 | __releases(rq->lock) |
1940 | { | |
1da177e4 | 1941 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1942 | long prev_state; |
1da177e4 LT |
1943 | |
1944 | rq->prev_mm = NULL; | |
1945 | ||
1946 | /* | |
1947 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1948 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1949 | * schedule one last time. The schedule call will never return, and |
1950 | * the scheduled task must drop that reference. | |
c394cc9f | 1951 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1952 | * still held, otherwise prev could be scheduled on another cpu, die |
1953 | * there before we look at prev->state, and then the reference would | |
1954 | * be dropped twice. | |
1955 | * Manfred Spraul <manfred@colorfullife.com> | |
1956 | */ | |
55a101f8 | 1957 | prev_state = prev->state; |
4866cde0 | 1958 | finish_arch_switch(prev); |
8381f65d JI |
1959 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1960 | local_irq_disable(); | |
1961 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
a8d757ef | 1962 | perf_event_task_sched_in(prev, current); |
8381f65d JI |
1963 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1964 | local_irq_enable(); | |
1965 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 1966 | finish_lock_switch(rq, prev); |
01f23e16 | 1967 | finish_arch_post_lock_switch(); |
e8fa1362 | 1968 | |
e107be36 | 1969 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1970 | if (mm) |
1971 | mmdrop(mm); | |
c394cc9f | 1972 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1973 | /* |
1974 | * Remove function-return probe instances associated with this | |
1975 | * task and put them back on the free list. | |
9761eea8 | 1976 | */ |
c6fd91f0 | 1977 | kprobe_flush_task(prev); |
1da177e4 | 1978 | put_task_struct(prev); |
c6fd91f0 | 1979 | } |
1da177e4 LT |
1980 | } |
1981 | ||
3f029d3c GH |
1982 | #ifdef CONFIG_SMP |
1983 | ||
1984 | /* assumes rq->lock is held */ | |
1985 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1986 | { | |
1987 | if (prev->sched_class->pre_schedule) | |
1988 | prev->sched_class->pre_schedule(rq, prev); | |
1989 | } | |
1990 | ||
1991 | /* rq->lock is NOT held, but preemption is disabled */ | |
1992 | static inline void post_schedule(struct rq *rq) | |
1993 | { | |
1994 | if (rq->post_schedule) { | |
1995 | unsigned long flags; | |
1996 | ||
05fa785c | 1997 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1998 | if (rq->curr->sched_class->post_schedule) |
1999 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2000 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2001 | |
2002 | rq->post_schedule = 0; | |
2003 | } | |
2004 | } | |
2005 | ||
2006 | #else | |
da19ab51 | 2007 | |
3f029d3c GH |
2008 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2009 | { | |
2010 | } | |
2011 | ||
2012 | static inline void post_schedule(struct rq *rq) | |
2013 | { | |
1da177e4 LT |
2014 | } |
2015 | ||
3f029d3c GH |
2016 | #endif |
2017 | ||
1da177e4 LT |
2018 | /** |
2019 | * schedule_tail - first thing a freshly forked thread must call. | |
2020 | * @prev: the thread we just switched away from. | |
2021 | */ | |
36c8b586 | 2022 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2023 | __releases(rq->lock) |
2024 | { | |
70b97a7f IM |
2025 | struct rq *rq = this_rq(); |
2026 | ||
4866cde0 | 2027 | finish_task_switch(rq, prev); |
da19ab51 | 2028 | |
3f029d3c GH |
2029 | /* |
2030 | * FIXME: do we need to worry about rq being invalidated by the | |
2031 | * task_switch? | |
2032 | */ | |
2033 | post_schedule(rq); | |
70b97a7f | 2034 | |
4866cde0 NP |
2035 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2036 | /* In this case, finish_task_switch does not reenable preemption */ | |
2037 | preempt_enable(); | |
2038 | #endif | |
1da177e4 | 2039 | if (current->set_child_tid) |
b488893a | 2040 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2041 | } |
2042 | ||
2043 | /* | |
2044 | * context_switch - switch to the new MM and the new | |
2045 | * thread's register state. | |
2046 | */ | |
dd41f596 | 2047 | static inline void |
70b97a7f | 2048 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2049 | struct task_struct *next) |
1da177e4 | 2050 | { |
dd41f596 | 2051 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2052 | |
e107be36 | 2053 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2054 | |
dd41f596 IM |
2055 | mm = next->mm; |
2056 | oldmm = prev->active_mm; | |
9226d125 ZA |
2057 | /* |
2058 | * For paravirt, this is coupled with an exit in switch_to to | |
2059 | * combine the page table reload and the switch backend into | |
2060 | * one hypercall. | |
2061 | */ | |
224101ed | 2062 | arch_start_context_switch(prev); |
9226d125 | 2063 | |
31915ab4 | 2064 | if (!mm) { |
1da177e4 LT |
2065 | next->active_mm = oldmm; |
2066 | atomic_inc(&oldmm->mm_count); | |
2067 | enter_lazy_tlb(oldmm, next); | |
2068 | } else | |
2069 | switch_mm(oldmm, mm, next); | |
2070 | ||
31915ab4 | 2071 | if (!prev->mm) { |
1da177e4 | 2072 | prev->active_mm = NULL; |
1da177e4 LT |
2073 | rq->prev_mm = oldmm; |
2074 | } | |
3a5f5e48 IM |
2075 | /* |
2076 | * Since the runqueue lock will be released by the next | |
2077 | * task (which is an invalid locking op but in the case | |
2078 | * of the scheduler it's an obvious special-case), so we | |
2079 | * do an early lockdep release here: | |
2080 | */ | |
2081 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2082 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2083 | #endif |
1da177e4 LT |
2084 | |
2085 | /* Here we just switch the register state and the stack. */ | |
2086 | switch_to(prev, next, prev); | |
2087 | ||
dd41f596 IM |
2088 | barrier(); |
2089 | /* | |
2090 | * this_rq must be evaluated again because prev may have moved | |
2091 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2092 | * frame will be invalid. | |
2093 | */ | |
2094 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2095 | } |
2096 | ||
2097 | /* | |
2098 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2099 | * | |
2100 | * externally visible scheduler statistics: current number of runnable | |
2101 | * threads, current number of uninterruptible-sleeping threads, total | |
2102 | * number of context switches performed since bootup. | |
2103 | */ | |
2104 | unsigned long nr_running(void) | |
2105 | { | |
2106 | unsigned long i, sum = 0; | |
2107 | ||
2108 | for_each_online_cpu(i) | |
2109 | sum += cpu_rq(i)->nr_running; | |
2110 | ||
2111 | return sum; | |
f711f609 | 2112 | } |
1da177e4 LT |
2113 | |
2114 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2115 | { |
1da177e4 | 2116 | unsigned long i, sum = 0; |
f711f609 | 2117 | |
0a945022 | 2118 | for_each_possible_cpu(i) |
1da177e4 | 2119 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2120 | |
2121 | /* | |
1da177e4 LT |
2122 | * Since we read the counters lockless, it might be slightly |
2123 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2124 | */ |
1da177e4 LT |
2125 | if (unlikely((long)sum < 0)) |
2126 | sum = 0; | |
f711f609 | 2127 | |
1da177e4 | 2128 | return sum; |
f711f609 | 2129 | } |
f711f609 | 2130 | |
1da177e4 | 2131 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2132 | { |
cc94abfc SR |
2133 | int i; |
2134 | unsigned long long sum = 0; | |
46cb4b7c | 2135 | |
0a945022 | 2136 | for_each_possible_cpu(i) |
1da177e4 | 2137 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2138 | |
1da177e4 LT |
2139 | return sum; |
2140 | } | |
483b4ee6 | 2141 | |
1da177e4 LT |
2142 | unsigned long nr_iowait(void) |
2143 | { | |
2144 | unsigned long i, sum = 0; | |
483b4ee6 | 2145 | |
0a945022 | 2146 | for_each_possible_cpu(i) |
1da177e4 | 2147 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2148 | |
1da177e4 LT |
2149 | return sum; |
2150 | } | |
483b4ee6 | 2151 | |
8c215bd3 | 2152 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2153 | { |
8c215bd3 | 2154 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2155 | return atomic_read(&this->nr_iowait); |
2156 | } | |
46cb4b7c | 2157 | |
69d25870 AV |
2158 | unsigned long this_cpu_load(void) |
2159 | { | |
2160 | struct rq *this = this_rq(); | |
2161 | return this->cpu_load[0]; | |
2162 | } | |
e790fb0b | 2163 | |
46cb4b7c | 2164 | |
dce48a84 TG |
2165 | /* Variables and functions for calc_load */ |
2166 | static atomic_long_t calc_load_tasks; | |
2167 | static unsigned long calc_load_update; | |
2168 | unsigned long avenrun[3]; | |
2169 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2170 | |
74f5187a PZ |
2171 | static long calc_load_fold_active(struct rq *this_rq) |
2172 | { | |
2173 | long nr_active, delta = 0; | |
2174 | ||
2175 | nr_active = this_rq->nr_running; | |
2176 | nr_active += (long) this_rq->nr_uninterruptible; | |
2177 | ||
2178 | if (nr_active != this_rq->calc_load_active) { | |
2179 | delta = nr_active - this_rq->calc_load_active; | |
2180 | this_rq->calc_load_active = nr_active; | |
2181 | } | |
2182 | ||
2183 | return delta; | |
2184 | } | |
2185 | ||
0f004f5a PZ |
2186 | static unsigned long |
2187 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2188 | { | |
2189 | load *= exp; | |
2190 | load += active * (FIXED_1 - exp); | |
2191 | load += 1UL << (FSHIFT - 1); | |
2192 | return load >> FSHIFT; | |
2193 | } | |
2194 | ||
74f5187a PZ |
2195 | #ifdef CONFIG_NO_HZ |
2196 | /* | |
2197 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2198 | * | |
2199 | * When making the ILB scale, we should try to pull this in as well. | |
2200 | */ | |
2201 | static atomic_long_t calc_load_tasks_idle; | |
2202 | ||
029632fb | 2203 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2204 | { |
2205 | long delta; | |
2206 | ||
2207 | delta = calc_load_fold_active(this_rq); | |
2208 | if (delta) | |
2209 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2210 | } | |
2211 | ||
2212 | static long calc_load_fold_idle(void) | |
2213 | { | |
2214 | long delta = 0; | |
2215 | ||
2216 | /* | |
2217 | * Its got a race, we don't care... | |
2218 | */ | |
2219 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2220 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2221 | ||
2222 | return delta; | |
2223 | } | |
0f004f5a PZ |
2224 | |
2225 | /** | |
2226 | * fixed_power_int - compute: x^n, in O(log n) time | |
2227 | * | |
2228 | * @x: base of the power | |
2229 | * @frac_bits: fractional bits of @x | |
2230 | * @n: power to raise @x to. | |
2231 | * | |
2232 | * By exploiting the relation between the definition of the natural power | |
2233 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2234 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2235 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2236 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2237 | * of course trivially computable in O(log_2 n), the length of our binary | |
2238 | * vector. | |
2239 | */ | |
2240 | static unsigned long | |
2241 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2242 | { | |
2243 | unsigned long result = 1UL << frac_bits; | |
2244 | ||
2245 | if (n) for (;;) { | |
2246 | if (n & 1) { | |
2247 | result *= x; | |
2248 | result += 1UL << (frac_bits - 1); | |
2249 | result >>= frac_bits; | |
2250 | } | |
2251 | n >>= 1; | |
2252 | if (!n) | |
2253 | break; | |
2254 | x *= x; | |
2255 | x += 1UL << (frac_bits - 1); | |
2256 | x >>= frac_bits; | |
2257 | } | |
2258 | ||
2259 | return result; | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * a1 = a0 * e + a * (1 - e) | |
2264 | * | |
2265 | * a2 = a1 * e + a * (1 - e) | |
2266 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2267 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2268 | * | |
2269 | * a3 = a2 * e + a * (1 - e) | |
2270 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2271 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2272 | * | |
2273 | * ... | |
2274 | * | |
2275 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2276 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2277 | * = a0 * e^n + a * (1 - e^n) | |
2278 | * | |
2279 | * [1] application of the geometric series: | |
2280 | * | |
2281 | * n 1 - x^(n+1) | |
2282 | * S_n := \Sum x^i = ------------- | |
2283 | * i=0 1 - x | |
2284 | */ | |
2285 | static unsigned long | |
2286 | calc_load_n(unsigned long load, unsigned long exp, | |
2287 | unsigned long active, unsigned int n) | |
2288 | { | |
2289 | ||
2290 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2291 | } | |
2292 | ||
2293 | /* | |
2294 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2295 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2296 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2297 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2298 | * | |
2299 | * Once we've updated the global active value, we need to apply the exponential | |
2300 | * weights adjusted to the number of cycles missed. | |
2301 | */ | |
c308b56b | 2302 | static void calc_global_nohz(void) |
0f004f5a PZ |
2303 | { |
2304 | long delta, active, n; | |
2305 | ||
0f004f5a PZ |
2306 | /* |
2307 | * If we crossed a calc_load_update boundary, make sure to fold | |
2308 | * any pending idle changes, the respective CPUs might have | |
2309 | * missed the tick driven calc_load_account_active() update | |
2310 | * due to NO_HZ. | |
2311 | */ | |
2312 | delta = calc_load_fold_idle(); | |
2313 | if (delta) | |
2314 | atomic_long_add(delta, &calc_load_tasks); | |
2315 | ||
2316 | /* | |
c308b56b | 2317 | * It could be the one fold was all it took, we done! |
0f004f5a | 2318 | */ |
c308b56b PZ |
2319 | if (time_before(jiffies, calc_load_update + 10)) |
2320 | return; | |
0f004f5a | 2321 | |
c308b56b PZ |
2322 | /* |
2323 | * Catch-up, fold however many we are behind still | |
2324 | */ | |
2325 | delta = jiffies - calc_load_update - 10; | |
2326 | n = 1 + (delta / LOAD_FREQ); | |
0f004f5a | 2327 | |
c308b56b PZ |
2328 | active = atomic_long_read(&calc_load_tasks); |
2329 | active = active > 0 ? active * FIXED_1 : 0; | |
0f004f5a | 2330 | |
c308b56b PZ |
2331 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); |
2332 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2333 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
0f004f5a | 2334 | |
c308b56b | 2335 | calc_load_update += n * LOAD_FREQ; |
0f004f5a | 2336 | } |
74f5187a | 2337 | #else |
029632fb | 2338 | void calc_load_account_idle(struct rq *this_rq) |
74f5187a PZ |
2339 | { |
2340 | } | |
2341 | ||
2342 | static inline long calc_load_fold_idle(void) | |
2343 | { | |
2344 | return 0; | |
2345 | } | |
0f004f5a | 2346 | |
c308b56b | 2347 | static void calc_global_nohz(void) |
0f004f5a PZ |
2348 | { |
2349 | } | |
74f5187a PZ |
2350 | #endif |
2351 | ||
2d02494f TG |
2352 | /** |
2353 | * get_avenrun - get the load average array | |
2354 | * @loads: pointer to dest load array | |
2355 | * @offset: offset to add | |
2356 | * @shift: shift count to shift the result left | |
2357 | * | |
2358 | * These values are estimates at best, so no need for locking. | |
2359 | */ | |
2360 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2361 | { | |
2362 | loads[0] = (avenrun[0] + offset) << shift; | |
2363 | loads[1] = (avenrun[1] + offset) << shift; | |
2364 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2365 | } |
46cb4b7c | 2366 | |
46cb4b7c | 2367 | /* |
dce48a84 TG |
2368 | * calc_load - update the avenrun load estimates 10 ticks after the |
2369 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2370 | */ |
0f004f5a | 2371 | void calc_global_load(unsigned long ticks) |
7835b98b | 2372 | { |
dce48a84 | 2373 | long active; |
1da177e4 | 2374 | |
0f004f5a | 2375 | if (time_before(jiffies, calc_load_update + 10)) |
dce48a84 | 2376 | return; |
1da177e4 | 2377 | |
dce48a84 TG |
2378 | active = atomic_long_read(&calc_load_tasks); |
2379 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2380 | |
dce48a84 TG |
2381 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2382 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2383 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2384 | |
dce48a84 | 2385 | calc_load_update += LOAD_FREQ; |
c308b56b PZ |
2386 | |
2387 | /* | |
2388 | * Account one period with whatever state we found before | |
2389 | * folding in the nohz state and ageing the entire idle period. | |
2390 | * | |
2391 | * This avoids loosing a sample when we go idle between | |
2392 | * calc_load_account_active() (10 ticks ago) and now and thus | |
2393 | * under-accounting. | |
2394 | */ | |
2395 | calc_global_nohz(); | |
dce48a84 | 2396 | } |
1da177e4 | 2397 | |
dce48a84 | 2398 | /* |
74f5187a PZ |
2399 | * Called from update_cpu_load() to periodically update this CPU's |
2400 | * active count. | |
dce48a84 TG |
2401 | */ |
2402 | static void calc_load_account_active(struct rq *this_rq) | |
2403 | { | |
74f5187a | 2404 | long delta; |
08c183f3 | 2405 | |
74f5187a PZ |
2406 | if (time_before(jiffies, this_rq->calc_load_update)) |
2407 | return; | |
783609c6 | 2408 | |
74f5187a PZ |
2409 | delta = calc_load_fold_active(this_rq); |
2410 | delta += calc_load_fold_idle(); | |
2411 | if (delta) | |
dce48a84 | 2412 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2413 | |
2414 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2415 | } |
2416 | ||
fdf3e95d VP |
2417 | /* |
2418 | * The exact cpuload at various idx values, calculated at every tick would be | |
2419 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2420 | * | |
2421 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2422 | * on nth tick when cpu may be busy, then we have: | |
2423 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2424 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2425 | * | |
2426 | * decay_load_missed() below does efficient calculation of | |
2427 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2428 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2429 | * | |
2430 | * The calculation is approximated on a 128 point scale. | |
2431 | * degrade_zero_ticks is the number of ticks after which load at any | |
2432 | * particular idx is approximated to be zero. | |
2433 | * degrade_factor is a precomputed table, a row for each load idx. | |
2434 | * Each column corresponds to degradation factor for a power of two ticks, | |
2435 | * based on 128 point scale. | |
2436 | * Example: | |
2437 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2438 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2439 | * | |
2440 | * With this power of 2 load factors, we can degrade the load n times | |
2441 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2442 | * n mult/shifts needed by the exact degradation. | |
2443 | */ | |
2444 | #define DEGRADE_SHIFT 7 | |
2445 | static const unsigned char | |
2446 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2447 | static const unsigned char | |
2448 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2449 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2450 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2451 | {96, 72, 40, 12, 1, 0, 0}, | |
2452 | {112, 98, 75, 43, 15, 1, 0}, | |
2453 | {120, 112, 98, 76, 45, 16, 2} }; | |
2454 | ||
2455 | /* | |
2456 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2457 | * would be when CPU is idle and so we just decay the old load without | |
2458 | * adding any new load. | |
2459 | */ | |
2460 | static unsigned long | |
2461 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2462 | { | |
2463 | int j = 0; | |
2464 | ||
2465 | if (!missed_updates) | |
2466 | return load; | |
2467 | ||
2468 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2469 | return 0; | |
2470 | ||
2471 | if (idx == 1) | |
2472 | return load >> missed_updates; | |
2473 | ||
2474 | while (missed_updates) { | |
2475 | if (missed_updates % 2) | |
2476 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2477 | ||
2478 | missed_updates >>= 1; | |
2479 | j++; | |
2480 | } | |
2481 | return load; | |
2482 | } | |
2483 | ||
46cb4b7c | 2484 | /* |
dd41f596 | 2485 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2486 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2487 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2488 | */ |
029632fb | 2489 | void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 2490 | { |
495eca49 | 2491 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
2492 | unsigned long curr_jiffies = jiffies; |
2493 | unsigned long pending_updates; | |
dd41f596 | 2494 | int i, scale; |
46cb4b7c | 2495 | |
dd41f596 | 2496 | this_rq->nr_load_updates++; |
46cb4b7c | 2497 | |
fdf3e95d VP |
2498 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
2499 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2500 | return; | |
2501 | ||
2502 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2503 | this_rq->last_load_update_tick = curr_jiffies; | |
2504 | ||
dd41f596 | 2505 | /* Update our load: */ |
fdf3e95d VP |
2506 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2507 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2508 | unsigned long old_load, new_load; |
7d1e6a9b | 2509 | |
dd41f596 | 2510 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2511 | |
dd41f596 | 2512 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2513 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2514 | new_load = this_load; |
a25707f3 IM |
2515 | /* |
2516 | * Round up the averaging division if load is increasing. This | |
2517 | * prevents us from getting stuck on 9 if the load is 10, for | |
2518 | * example. | |
2519 | */ | |
2520 | if (new_load > old_load) | |
fdf3e95d VP |
2521 | new_load += scale - 1; |
2522 | ||
2523 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2524 | } |
da2b71ed SS |
2525 | |
2526 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2527 | } |
2528 | ||
2529 | static void update_cpu_load_active(struct rq *this_rq) | |
2530 | { | |
2531 | update_cpu_load(this_rq); | |
46cb4b7c | 2532 | |
74f5187a | 2533 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2534 | } |
2535 | ||
dd41f596 | 2536 | #ifdef CONFIG_SMP |
8a0be9ef | 2537 | |
46cb4b7c | 2538 | /* |
38022906 PZ |
2539 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2540 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2541 | */ |
38022906 | 2542 | void sched_exec(void) |
46cb4b7c | 2543 | { |
38022906 | 2544 | struct task_struct *p = current; |
1da177e4 | 2545 | unsigned long flags; |
0017d735 | 2546 | int dest_cpu; |
46cb4b7c | 2547 | |
8f42ced9 | 2548 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2549 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2550 | if (dest_cpu == smp_processor_id()) |
2551 | goto unlock; | |
38022906 | 2552 | |
8f42ced9 | 2553 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2554 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2555 | |
8f42ced9 PZ |
2556 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2557 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2558 | return; |
2559 | } | |
0017d735 | 2560 | unlock: |
8f42ced9 | 2561 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2562 | } |
dd41f596 | 2563 | |
1da177e4 LT |
2564 | #endif |
2565 | ||
1da177e4 | 2566 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2567 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2568 | |
2569 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2570 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2571 | |
2572 | /* | |
c5f8d995 | 2573 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2574 | * @p in case that task is currently running. |
c5f8d995 HS |
2575 | * |
2576 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2577 | */ |
c5f8d995 HS |
2578 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2579 | { | |
2580 | u64 ns = 0; | |
2581 | ||
2582 | if (task_current(rq, p)) { | |
2583 | update_rq_clock(rq); | |
305e6835 | 2584 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2585 | if ((s64)ns < 0) |
2586 | ns = 0; | |
2587 | } | |
2588 | ||
2589 | return ns; | |
2590 | } | |
2591 | ||
bb34d92f | 2592 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2593 | { |
1da177e4 | 2594 | unsigned long flags; |
41b86e9c | 2595 | struct rq *rq; |
bb34d92f | 2596 | u64 ns = 0; |
48f24c4d | 2597 | |
41b86e9c | 2598 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2599 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2600 | task_rq_unlock(rq, p, &flags); |
1508487e | 2601 | |
c5f8d995 HS |
2602 | return ns; |
2603 | } | |
f06febc9 | 2604 | |
c5f8d995 HS |
2605 | /* |
2606 | * Return accounted runtime for the task. | |
2607 | * In case the task is currently running, return the runtime plus current's | |
2608 | * pending runtime that have not been accounted yet. | |
2609 | */ | |
2610 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2611 | { | |
2612 | unsigned long flags; | |
2613 | struct rq *rq; | |
2614 | u64 ns = 0; | |
2615 | ||
2616 | rq = task_rq_lock(p, &flags); | |
2617 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2618 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2619 | |
2620 | return ns; | |
2621 | } | |
48f24c4d | 2622 | |
54c707e9 GC |
2623 | #ifdef CONFIG_CGROUP_CPUACCT |
2624 | struct cgroup_subsys cpuacct_subsys; | |
2625 | struct cpuacct root_cpuacct; | |
2626 | #endif | |
2627 | ||
be726ffd GC |
2628 | static inline void task_group_account_field(struct task_struct *p, int index, |
2629 | u64 tmp) | |
54c707e9 GC |
2630 | { |
2631 | #ifdef CONFIG_CGROUP_CPUACCT | |
2632 | struct kernel_cpustat *kcpustat; | |
2633 | struct cpuacct *ca; | |
2634 | #endif | |
2635 | /* | |
2636 | * Since all updates are sure to touch the root cgroup, we | |
2637 | * get ourselves ahead and touch it first. If the root cgroup | |
2638 | * is the only cgroup, then nothing else should be necessary. | |
2639 | * | |
2640 | */ | |
2641 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | |
2642 | ||
2643 | #ifdef CONFIG_CGROUP_CPUACCT | |
2644 | if (unlikely(!cpuacct_subsys.active)) | |
2645 | return; | |
2646 | ||
2647 | rcu_read_lock(); | |
2648 | ca = task_ca(p); | |
2649 | while (ca && (ca != &root_cpuacct)) { | |
2650 | kcpustat = this_cpu_ptr(ca->cpustat); | |
2651 | kcpustat->cpustat[index] += tmp; | |
2652 | ca = parent_ca(ca); | |
2653 | } | |
2654 | rcu_read_unlock(); | |
2655 | #endif | |
2656 | } | |
2657 | ||
2658 | ||
1da177e4 LT |
2659 | /* |
2660 | * Account user cpu time to a process. | |
2661 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 2662 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 2663 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 2664 | */ |
457533a7 MS |
2665 | void account_user_time(struct task_struct *p, cputime_t cputime, |
2666 | cputime_t cputime_scaled) | |
1da177e4 | 2667 | { |
3292beb3 | 2668 | int index; |
1da177e4 | 2669 | |
457533a7 | 2670 | /* Add user time to process. */ |
64861634 MS |
2671 | p->utime += cputime; |
2672 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2673 | account_group_user_time(p, cputime); |
1da177e4 | 2674 | |
3292beb3 | 2675 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
ef12fefa | 2676 | |
1da177e4 | 2677 | /* Add user time to cpustat. */ |
612ef28a | 2678 | task_group_account_field(p, index, (__force u64) cputime); |
ef12fefa | 2679 | |
49b5cf34 JL |
2680 | /* Account for user time used */ |
2681 | acct_update_integrals(p); | |
1da177e4 LT |
2682 | } |
2683 | ||
94886b84 LV |
2684 | /* |
2685 | * Account guest cpu time to a process. | |
2686 | * @p: the process that the cpu time gets accounted to | |
2687 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 2688 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 2689 | */ |
457533a7 MS |
2690 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
2691 | cputime_t cputime_scaled) | |
94886b84 | 2692 | { |
3292beb3 | 2693 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
94886b84 | 2694 | |
457533a7 | 2695 | /* Add guest time to process. */ |
64861634 MS |
2696 | p->utime += cputime; |
2697 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2698 | account_group_user_time(p, cputime); |
64861634 | 2699 | p->gtime += cputime; |
94886b84 | 2700 | |
457533a7 | 2701 | /* Add guest time to cpustat. */ |
ce0e7b28 | 2702 | if (TASK_NICE(p) > 0) { |
612ef28a MS |
2703 | cpustat[CPUTIME_NICE] += (__force u64) cputime; |
2704 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
ce0e7b28 | 2705 | } else { |
612ef28a MS |
2706 | cpustat[CPUTIME_USER] += (__force u64) cputime; |
2707 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
ce0e7b28 | 2708 | } |
94886b84 LV |
2709 | } |
2710 | ||
70a89a66 VP |
2711 | /* |
2712 | * Account system cpu time to a process and desired cpustat field | |
2713 | * @p: the process that the cpu time gets accounted to | |
2714 | * @cputime: the cpu time spent in kernel space since the last update | |
2715 | * @cputime_scaled: cputime scaled by cpu frequency | |
2716 | * @target_cputime64: pointer to cpustat field that has to be updated | |
2717 | */ | |
2718 | static inline | |
2719 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3292beb3 | 2720 | cputime_t cputime_scaled, int index) |
70a89a66 | 2721 | { |
70a89a66 | 2722 | /* Add system time to process. */ |
64861634 MS |
2723 | p->stime += cputime; |
2724 | p->stimescaled += cputime_scaled; | |
70a89a66 VP |
2725 | account_group_system_time(p, cputime); |
2726 | ||
2727 | /* Add system time to cpustat. */ | |
612ef28a | 2728 | task_group_account_field(p, index, (__force u64) cputime); |
70a89a66 VP |
2729 | |
2730 | /* Account for system time used */ | |
2731 | acct_update_integrals(p); | |
2732 | } | |
2733 | ||
1da177e4 LT |
2734 | /* |
2735 | * Account system cpu time to a process. | |
2736 | * @p: the process that the cpu time gets accounted to | |
2737 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
2738 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 2739 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
2740 | */ |
2741 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 2742 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 | 2743 | { |
3292beb3 | 2744 | int index; |
1da177e4 | 2745 | |
983ed7a6 | 2746 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 2747 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
2748 | return; |
2749 | } | |
94886b84 | 2750 | |
1da177e4 | 2751 | if (hardirq_count() - hardirq_offset) |
3292beb3 | 2752 | index = CPUTIME_IRQ; |
75e1056f | 2753 | else if (in_serving_softirq()) |
3292beb3 | 2754 | index = CPUTIME_SOFTIRQ; |
1da177e4 | 2755 | else |
3292beb3 | 2756 | index = CPUTIME_SYSTEM; |
ef12fefa | 2757 | |
3292beb3 | 2758 | __account_system_time(p, cputime, cputime_scaled, index); |
1da177e4 LT |
2759 | } |
2760 | ||
c66f08be | 2761 | /* |
1da177e4 | 2762 | * Account for involuntary wait time. |
544b4a1f | 2763 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 2764 | */ |
79741dd3 | 2765 | void account_steal_time(cputime_t cputime) |
c66f08be | 2766 | { |
3292beb3 | 2767 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
79741dd3 | 2768 | |
612ef28a | 2769 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; |
c66f08be MN |
2770 | } |
2771 | ||
1da177e4 | 2772 | /* |
79741dd3 MS |
2773 | * Account for idle time. |
2774 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 2775 | */ |
79741dd3 | 2776 | void account_idle_time(cputime_t cputime) |
1da177e4 | 2777 | { |
3292beb3 | 2778 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
70b97a7f | 2779 | struct rq *rq = this_rq(); |
1da177e4 | 2780 | |
79741dd3 | 2781 | if (atomic_read(&rq->nr_iowait) > 0) |
612ef28a | 2782 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; |
79741dd3 | 2783 | else |
612ef28a | 2784 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; |
1da177e4 LT |
2785 | } |
2786 | ||
e6e6685a GC |
2787 | static __always_inline bool steal_account_process_tick(void) |
2788 | { | |
2789 | #ifdef CONFIG_PARAVIRT | |
c5905afb | 2790 | if (static_key_false(¶virt_steal_enabled)) { |
e6e6685a GC |
2791 | u64 steal, st = 0; |
2792 | ||
2793 | steal = paravirt_steal_clock(smp_processor_id()); | |
2794 | steal -= this_rq()->prev_steal_time; | |
2795 | ||
2796 | st = steal_ticks(steal); | |
2797 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
2798 | ||
2799 | account_steal_time(st); | |
2800 | return st; | |
2801 | } | |
2802 | #endif | |
2803 | return false; | |
2804 | } | |
2805 | ||
79741dd3 MS |
2806 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
2807 | ||
abb74cef VP |
2808 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
2809 | /* | |
2810 | * Account a tick to a process and cpustat | |
2811 | * @p: the process that the cpu time gets accounted to | |
2812 | * @user_tick: is the tick from userspace | |
2813 | * @rq: the pointer to rq | |
2814 | * | |
2815 | * Tick demultiplexing follows the order | |
2816 | * - pending hardirq update | |
2817 | * - pending softirq update | |
2818 | * - user_time | |
2819 | * - idle_time | |
2820 | * - system time | |
2821 | * - check for guest_time | |
2822 | * - else account as system_time | |
2823 | * | |
2824 | * Check for hardirq is done both for system and user time as there is | |
2825 | * no timer going off while we are on hardirq and hence we may never get an | |
2826 | * opportunity to update it solely in system time. | |
2827 | * p->stime and friends are only updated on system time and not on irq | |
2828 | * softirq as those do not count in task exec_runtime any more. | |
2829 | */ | |
2830 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2831 | struct rq *rq) | |
2832 | { | |
2833 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3292beb3 | 2834 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef | 2835 | |
e6e6685a GC |
2836 | if (steal_account_process_tick()) |
2837 | return; | |
2838 | ||
abb74cef | 2839 | if (irqtime_account_hi_update()) { |
612ef28a | 2840 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; |
abb74cef | 2841 | } else if (irqtime_account_si_update()) { |
612ef28a | 2842 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; |
414bee9b VP |
2843 | } else if (this_cpu_ksoftirqd() == p) { |
2844 | /* | |
2845 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
2846 | * So, we have to handle it separately here. | |
2847 | * Also, p->stime needs to be updated for ksoftirqd. | |
2848 | */ | |
2849 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2850 | CPUTIME_SOFTIRQ); |
abb74cef VP |
2851 | } else if (user_tick) { |
2852 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2853 | } else if (p == rq->idle) { | |
2854 | account_idle_time(cputime_one_jiffy); | |
2855 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
2856 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
2857 | } else { | |
2858 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 2859 | CPUTIME_SYSTEM); |
abb74cef VP |
2860 | } |
2861 | } | |
2862 | ||
2863 | static void irqtime_account_idle_ticks(int ticks) | |
2864 | { | |
2865 | int i; | |
2866 | struct rq *rq = this_rq(); | |
2867 | ||
2868 | for (i = 0; i < ticks; i++) | |
2869 | irqtime_account_process_tick(current, 0, rq); | |
2870 | } | |
544b4a1f | 2871 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
2872 | static void irqtime_account_idle_ticks(int ticks) {} |
2873 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2874 | struct rq *rq) {} | |
544b4a1f | 2875 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
2876 | |
2877 | /* | |
2878 | * Account a single tick of cpu time. | |
2879 | * @p: the process that the cpu time gets accounted to | |
2880 | * @user_tick: indicates if the tick is a user or a system tick | |
2881 | */ | |
2882 | void account_process_tick(struct task_struct *p, int user_tick) | |
2883 | { | |
a42548a1 | 2884 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
2885 | struct rq *rq = this_rq(); |
2886 | ||
abb74cef VP |
2887 | if (sched_clock_irqtime) { |
2888 | irqtime_account_process_tick(p, user_tick, rq); | |
2889 | return; | |
2890 | } | |
2891 | ||
e6e6685a GC |
2892 | if (steal_account_process_tick()) |
2893 | return; | |
2894 | ||
79741dd3 | 2895 | if (user_tick) |
a42548a1 | 2896 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 2897 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 2898 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
2899 | one_jiffy_scaled); |
2900 | else | |
a42548a1 | 2901 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
2902 | } |
2903 | ||
2904 | /* | |
2905 | * Account multiple ticks of steal time. | |
2906 | * @p: the process from which the cpu time has been stolen | |
2907 | * @ticks: number of stolen ticks | |
2908 | */ | |
2909 | void account_steal_ticks(unsigned long ticks) | |
2910 | { | |
2911 | account_steal_time(jiffies_to_cputime(ticks)); | |
2912 | } | |
2913 | ||
2914 | /* | |
2915 | * Account multiple ticks of idle time. | |
2916 | * @ticks: number of stolen ticks | |
2917 | */ | |
2918 | void account_idle_ticks(unsigned long ticks) | |
2919 | { | |
abb74cef VP |
2920 | |
2921 | if (sched_clock_irqtime) { | |
2922 | irqtime_account_idle_ticks(ticks); | |
2923 | return; | |
2924 | } | |
2925 | ||
79741dd3 | 2926 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
2927 | } |
2928 | ||
79741dd3 MS |
2929 | #endif |
2930 | ||
49048622 BS |
2931 | /* |
2932 | * Use precise platform statistics if available: | |
2933 | */ | |
2934 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 2935 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2936 | { |
d99ca3b9 HS |
2937 | *ut = p->utime; |
2938 | *st = p->stime; | |
49048622 BS |
2939 | } |
2940 | ||
0cf55e1e | 2941 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2942 | { |
0cf55e1e HS |
2943 | struct task_cputime cputime; |
2944 | ||
2945 | thread_group_cputime(p, &cputime); | |
2946 | ||
2947 | *ut = cputime.utime; | |
2948 | *st = cputime.stime; | |
49048622 BS |
2949 | } |
2950 | #else | |
761b1d26 HS |
2951 | |
2952 | #ifndef nsecs_to_cputime | |
b7b20df9 | 2953 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
2954 | #endif |
2955 | ||
d180c5bc | 2956 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 2957 | { |
64861634 | 2958 | cputime_t rtime, utime = p->utime, total = utime + p->stime; |
49048622 BS |
2959 | |
2960 | /* | |
2961 | * Use CFS's precise accounting: | |
2962 | */ | |
d180c5bc | 2963 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
2964 | |
2965 | if (total) { | |
64861634 | 2966 | u64 temp = (__force u64) rtime; |
d180c5bc | 2967 | |
64861634 MS |
2968 | temp *= (__force u64) utime; |
2969 | do_div(temp, (__force u32) total); | |
2970 | utime = (__force cputime_t) temp; | |
d180c5bc HS |
2971 | } else |
2972 | utime = rtime; | |
49048622 | 2973 | |
d180c5bc HS |
2974 | /* |
2975 | * Compare with previous values, to keep monotonicity: | |
2976 | */ | |
761b1d26 | 2977 | p->prev_utime = max(p->prev_utime, utime); |
64861634 | 2978 | p->prev_stime = max(p->prev_stime, rtime - p->prev_utime); |
49048622 | 2979 | |
d99ca3b9 HS |
2980 | *ut = p->prev_utime; |
2981 | *st = p->prev_stime; | |
49048622 BS |
2982 | } |
2983 | ||
0cf55e1e HS |
2984 | /* |
2985 | * Must be called with siglock held. | |
2986 | */ | |
2987 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 2988 | { |
0cf55e1e HS |
2989 | struct signal_struct *sig = p->signal; |
2990 | struct task_cputime cputime; | |
2991 | cputime_t rtime, utime, total; | |
49048622 | 2992 | |
0cf55e1e | 2993 | thread_group_cputime(p, &cputime); |
49048622 | 2994 | |
64861634 | 2995 | total = cputime.utime + cputime.stime; |
0cf55e1e | 2996 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); |
49048622 | 2997 | |
0cf55e1e | 2998 | if (total) { |
64861634 | 2999 | u64 temp = (__force u64) rtime; |
49048622 | 3000 | |
64861634 MS |
3001 | temp *= (__force u64) cputime.utime; |
3002 | do_div(temp, (__force u32) total); | |
3003 | utime = (__force cputime_t) temp; | |
0cf55e1e HS |
3004 | } else |
3005 | utime = rtime; | |
3006 | ||
3007 | sig->prev_utime = max(sig->prev_utime, utime); | |
64861634 | 3008 | sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime); |
0cf55e1e HS |
3009 | |
3010 | *ut = sig->prev_utime; | |
3011 | *st = sig->prev_stime; | |
49048622 | 3012 | } |
49048622 | 3013 | #endif |
49048622 | 3014 | |
7835b98b CL |
3015 | /* |
3016 | * This function gets called by the timer code, with HZ frequency. | |
3017 | * We call it with interrupts disabled. | |
7835b98b CL |
3018 | */ |
3019 | void scheduler_tick(void) | |
3020 | { | |
7835b98b CL |
3021 | int cpu = smp_processor_id(); |
3022 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3023 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3024 | |
3025 | sched_clock_tick(); | |
dd41f596 | 3026 | |
05fa785c | 3027 | raw_spin_lock(&rq->lock); |
3e51f33f | 3028 | update_rq_clock(rq); |
fdf3e95d | 3029 | update_cpu_load_active(rq); |
fa85ae24 | 3030 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3031 | raw_spin_unlock(&rq->lock); |
7835b98b | 3032 | |
e9d2b064 | 3033 | perf_event_task_tick(); |
e220d2dc | 3034 | |
e418e1c2 | 3035 | #ifdef CONFIG_SMP |
6eb57e0d | 3036 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 3037 | trigger_load_balance(rq, cpu); |
e418e1c2 | 3038 | #endif |
1da177e4 LT |
3039 | } |
3040 | ||
132380a0 | 3041 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3042 | { |
3043 | if (in_lock_functions(addr)) { | |
3044 | addr = CALLER_ADDR2; | |
3045 | if (in_lock_functions(addr)) | |
3046 | addr = CALLER_ADDR3; | |
3047 | } | |
3048 | return addr; | |
3049 | } | |
1da177e4 | 3050 | |
7e49fcce SR |
3051 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3052 | defined(CONFIG_PREEMPT_TRACER)) | |
3053 | ||
43627582 | 3054 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3055 | { |
6cd8a4bb | 3056 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3057 | /* |
3058 | * Underflow? | |
3059 | */ | |
9a11b49a IM |
3060 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3061 | return; | |
6cd8a4bb | 3062 | #endif |
1da177e4 | 3063 | preempt_count() += val; |
6cd8a4bb | 3064 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3065 | /* |
3066 | * Spinlock count overflowing soon? | |
3067 | */ | |
33859f7f MOS |
3068 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3069 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3070 | #endif |
3071 | if (preempt_count() == val) | |
3072 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3073 | } |
3074 | EXPORT_SYMBOL(add_preempt_count); | |
3075 | ||
43627582 | 3076 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3077 | { |
6cd8a4bb | 3078 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3079 | /* |
3080 | * Underflow? | |
3081 | */ | |
01e3eb82 | 3082 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3083 | return; |
1da177e4 LT |
3084 | /* |
3085 | * Is the spinlock portion underflowing? | |
3086 | */ | |
9a11b49a IM |
3087 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3088 | !(preempt_count() & PREEMPT_MASK))) | |
3089 | return; | |
6cd8a4bb | 3090 | #endif |
9a11b49a | 3091 | |
6cd8a4bb SR |
3092 | if (preempt_count() == val) |
3093 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3094 | preempt_count() -= val; |
3095 | } | |
3096 | EXPORT_SYMBOL(sub_preempt_count); | |
3097 | ||
3098 | #endif | |
3099 | ||
3100 | /* | |
dd41f596 | 3101 | * Print scheduling while atomic bug: |
1da177e4 | 3102 | */ |
dd41f596 | 3103 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3104 | { |
664dfa65 DJ |
3105 | if (oops_in_progress) |
3106 | return; | |
3107 | ||
3df0fc5b PZ |
3108 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3109 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3110 | |
dd41f596 | 3111 | debug_show_held_locks(prev); |
e21f5b15 | 3112 | print_modules(); |
dd41f596 IM |
3113 | if (irqs_disabled()) |
3114 | print_irqtrace_events(prev); | |
6135fc1e | 3115 | dump_stack(); |
dd41f596 | 3116 | } |
1da177e4 | 3117 | |
dd41f596 IM |
3118 | /* |
3119 | * Various schedule()-time debugging checks and statistics: | |
3120 | */ | |
3121 | static inline void schedule_debug(struct task_struct *prev) | |
3122 | { | |
1da177e4 | 3123 | /* |
41a2d6cf | 3124 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3125 | * schedule() atomically, we ignore that path for now. |
3126 | * Otherwise, whine if we are scheduling when we should not be. | |
3127 | */ | |
3f33a7ce | 3128 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 3129 | __schedule_bug(prev); |
b3fbab05 | 3130 | rcu_sleep_check(); |
dd41f596 | 3131 | |
1da177e4 LT |
3132 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3133 | ||
2d72376b | 3134 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3135 | } |
3136 | ||
6cecd084 | 3137 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3138 | { |
61eadef6 | 3139 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 3140 | update_rq_clock(rq); |
6cecd084 | 3141 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3142 | } |
3143 | ||
dd41f596 IM |
3144 | /* |
3145 | * Pick up the highest-prio task: | |
3146 | */ | |
3147 | static inline struct task_struct * | |
b67802ea | 3148 | pick_next_task(struct rq *rq) |
dd41f596 | 3149 | { |
5522d5d5 | 3150 | const struct sched_class *class; |
dd41f596 | 3151 | struct task_struct *p; |
1da177e4 LT |
3152 | |
3153 | /* | |
dd41f596 IM |
3154 | * Optimization: we know that if all tasks are in |
3155 | * the fair class we can call that function directly: | |
1da177e4 | 3156 | */ |
953bfcd1 | 3157 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 3158 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3159 | if (likely(p)) |
3160 | return p; | |
1da177e4 LT |
3161 | } |
3162 | ||
34f971f6 | 3163 | for_each_class(class) { |
fb8d4724 | 3164 | p = class->pick_next_task(rq); |
dd41f596 IM |
3165 | if (p) |
3166 | return p; | |
dd41f596 | 3167 | } |
34f971f6 PZ |
3168 | |
3169 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3170 | } |
1da177e4 | 3171 | |
dd41f596 | 3172 | /* |
c259e01a | 3173 | * __schedule() is the main scheduler function. |
dd41f596 | 3174 | */ |
c259e01a | 3175 | static void __sched __schedule(void) |
dd41f596 IM |
3176 | { |
3177 | struct task_struct *prev, *next; | |
67ca7bde | 3178 | unsigned long *switch_count; |
dd41f596 | 3179 | struct rq *rq; |
31656519 | 3180 | int cpu; |
dd41f596 | 3181 | |
ff743345 PZ |
3182 | need_resched: |
3183 | preempt_disable(); | |
dd41f596 IM |
3184 | cpu = smp_processor_id(); |
3185 | rq = cpu_rq(cpu); | |
25502a6c | 3186 | rcu_note_context_switch(cpu); |
dd41f596 | 3187 | prev = rq->curr; |
dd41f596 | 3188 | |
dd41f596 | 3189 | schedule_debug(prev); |
1da177e4 | 3190 | |
31656519 | 3191 | if (sched_feat(HRTICK)) |
f333fdc9 | 3192 | hrtick_clear(rq); |
8f4d37ec | 3193 | |
05fa785c | 3194 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3195 | |
246d86b5 | 3196 | switch_count = &prev->nivcsw; |
1da177e4 | 3197 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3198 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3199 | prev->state = TASK_RUNNING; |
21aa9af0 | 3200 | } else { |
2acca55e PZ |
3201 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3202 | prev->on_rq = 0; | |
3203 | ||
21aa9af0 | 3204 | /* |
2acca55e PZ |
3205 | * If a worker went to sleep, notify and ask workqueue |
3206 | * whether it wants to wake up a task to maintain | |
3207 | * concurrency. | |
21aa9af0 TH |
3208 | */ |
3209 | if (prev->flags & PF_WQ_WORKER) { | |
3210 | struct task_struct *to_wakeup; | |
3211 | ||
3212 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3213 | if (to_wakeup) | |
3214 | try_to_wake_up_local(to_wakeup); | |
3215 | } | |
21aa9af0 | 3216 | } |
dd41f596 | 3217 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3218 | } |
3219 | ||
3f029d3c | 3220 | pre_schedule(rq, prev); |
f65eda4f | 3221 | |
dd41f596 | 3222 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3223 | idle_balance(cpu, rq); |
1da177e4 | 3224 | |
df1c99d4 | 3225 | put_prev_task(rq, prev); |
b67802ea | 3226 | next = pick_next_task(rq); |
f26f9aff MG |
3227 | clear_tsk_need_resched(prev); |
3228 | rq->skip_clock_update = 0; | |
1da177e4 | 3229 | |
1da177e4 | 3230 | if (likely(prev != next)) { |
1da177e4 LT |
3231 | rq->nr_switches++; |
3232 | rq->curr = next; | |
3233 | ++*switch_count; | |
3234 | ||
dd41f596 | 3235 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3236 | /* |
246d86b5 ON |
3237 | * The context switch have flipped the stack from under us |
3238 | * and restored the local variables which were saved when | |
3239 | * this task called schedule() in the past. prev == current | |
3240 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3241 | */ |
3242 | cpu = smp_processor_id(); | |
3243 | rq = cpu_rq(cpu); | |
1da177e4 | 3244 | } else |
05fa785c | 3245 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3246 | |
3f029d3c | 3247 | post_schedule(rq); |
1da177e4 | 3248 | |
ba74c144 | 3249 | sched_preempt_enable_no_resched(); |
ff743345 | 3250 | if (need_resched()) |
1da177e4 LT |
3251 | goto need_resched; |
3252 | } | |
c259e01a | 3253 | |
9c40cef2 TG |
3254 | static inline void sched_submit_work(struct task_struct *tsk) |
3255 | { | |
3c7d5184 | 3256 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3257 | return; |
3258 | /* | |
3259 | * If we are going to sleep and we have plugged IO queued, | |
3260 | * make sure to submit it to avoid deadlocks. | |
3261 | */ | |
3262 | if (blk_needs_flush_plug(tsk)) | |
3263 | blk_schedule_flush_plug(tsk); | |
3264 | } | |
3265 | ||
6ebbe7a0 | 3266 | asmlinkage void __sched schedule(void) |
c259e01a | 3267 | { |
9c40cef2 TG |
3268 | struct task_struct *tsk = current; |
3269 | ||
3270 | sched_submit_work(tsk); | |
c259e01a TG |
3271 | __schedule(); |
3272 | } | |
1da177e4 LT |
3273 | EXPORT_SYMBOL(schedule); |
3274 | ||
c5491ea7 TG |
3275 | /** |
3276 | * schedule_preempt_disabled - called with preemption disabled | |
3277 | * | |
3278 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3279 | */ | |
3280 | void __sched schedule_preempt_disabled(void) | |
3281 | { | |
ba74c144 | 3282 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3283 | schedule(); |
3284 | preempt_disable(); | |
3285 | } | |
3286 | ||
c08f7829 | 3287 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3288 | |
c6eb3dda PZ |
3289 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3290 | { | |
c6eb3dda | 3291 | if (lock->owner != owner) |
307bf980 | 3292 | return false; |
0d66bf6d PZ |
3293 | |
3294 | /* | |
c6eb3dda PZ |
3295 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3296 | * lock->owner still matches owner, if that fails, owner might | |
3297 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3298 | * ensures the memory stays valid. | |
0d66bf6d | 3299 | */ |
c6eb3dda | 3300 | barrier(); |
0d66bf6d | 3301 | |
307bf980 | 3302 | return owner->on_cpu; |
c6eb3dda | 3303 | } |
0d66bf6d | 3304 | |
c6eb3dda PZ |
3305 | /* |
3306 | * Look out! "owner" is an entirely speculative pointer | |
3307 | * access and not reliable. | |
3308 | */ | |
3309 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3310 | { | |
3311 | if (!sched_feat(OWNER_SPIN)) | |
3312 | return 0; | |
0d66bf6d | 3313 | |
307bf980 | 3314 | rcu_read_lock(); |
c6eb3dda PZ |
3315 | while (owner_running(lock, owner)) { |
3316 | if (need_resched()) | |
307bf980 | 3317 | break; |
0d66bf6d | 3318 | |
335d7afb | 3319 | arch_mutex_cpu_relax(); |
0d66bf6d | 3320 | } |
307bf980 | 3321 | rcu_read_unlock(); |
4b402210 | 3322 | |
c6eb3dda | 3323 | /* |
307bf980 TG |
3324 | * We break out the loop above on need_resched() and when the |
3325 | * owner changed, which is a sign for heavy contention. Return | |
3326 | * success only when lock->owner is NULL. | |
c6eb3dda | 3327 | */ |
307bf980 | 3328 | return lock->owner == NULL; |
0d66bf6d PZ |
3329 | } |
3330 | #endif | |
3331 | ||
1da177e4 LT |
3332 | #ifdef CONFIG_PREEMPT |
3333 | /* | |
2ed6e34f | 3334 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3335 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3336 | * occur there and call schedule directly. |
3337 | */ | |
d1f74e20 | 3338 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3339 | { |
3340 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3341 | |
1da177e4 LT |
3342 | /* |
3343 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3344 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3345 | */ |
beed33a8 | 3346 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3347 | return; |
3348 | ||
3a5c359a | 3349 | do { |
d1f74e20 | 3350 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3351 | __schedule(); |
d1f74e20 | 3352 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3353 | |
3a5c359a AK |
3354 | /* |
3355 | * Check again in case we missed a preemption opportunity | |
3356 | * between schedule and now. | |
3357 | */ | |
3358 | barrier(); | |
5ed0cec0 | 3359 | } while (need_resched()); |
1da177e4 | 3360 | } |
1da177e4 LT |
3361 | EXPORT_SYMBOL(preempt_schedule); |
3362 | ||
3363 | /* | |
2ed6e34f | 3364 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3365 | * off of irq context. |
3366 | * Note, that this is called and return with irqs disabled. This will | |
3367 | * protect us against recursive calling from irq. | |
3368 | */ | |
3369 | asmlinkage void __sched preempt_schedule_irq(void) | |
3370 | { | |
3371 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3372 | |
2ed6e34f | 3373 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3374 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3375 | ||
3a5c359a AK |
3376 | do { |
3377 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3378 | local_irq_enable(); |
c259e01a | 3379 | __schedule(); |
3a5c359a | 3380 | local_irq_disable(); |
3a5c359a | 3381 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3382 | |
3a5c359a AK |
3383 | /* |
3384 | * Check again in case we missed a preemption opportunity | |
3385 | * between schedule and now. | |
3386 | */ | |
3387 | barrier(); | |
5ed0cec0 | 3388 | } while (need_resched()); |
1da177e4 LT |
3389 | } |
3390 | ||
3391 | #endif /* CONFIG_PREEMPT */ | |
3392 | ||
63859d4f | 3393 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3394 | void *key) |
1da177e4 | 3395 | { |
63859d4f | 3396 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3397 | } |
1da177e4 LT |
3398 | EXPORT_SYMBOL(default_wake_function); |
3399 | ||
3400 | /* | |
41a2d6cf IM |
3401 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3402 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3403 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3404 | * | |
3405 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3406 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3407 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3408 | */ | |
78ddb08f | 3409 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3410 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3411 | { |
2e45874c | 3412 | wait_queue_t *curr, *next; |
1da177e4 | 3413 | |
2e45874c | 3414 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3415 | unsigned flags = curr->flags; |
3416 | ||
63859d4f | 3417 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3418 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3419 | break; |
3420 | } | |
3421 | } | |
3422 | ||
3423 | /** | |
3424 | * __wake_up - wake up threads blocked on a waitqueue. | |
3425 | * @q: the waitqueue | |
3426 | * @mode: which threads | |
3427 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3428 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3429 | * |
3430 | * It may be assumed that this function implies a write memory barrier before | |
3431 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3432 | */ |
7ad5b3a5 | 3433 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3434 | int nr_exclusive, void *key) |
1da177e4 LT |
3435 | { |
3436 | unsigned long flags; | |
3437 | ||
3438 | spin_lock_irqsave(&q->lock, flags); | |
3439 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3440 | spin_unlock_irqrestore(&q->lock, flags); | |
3441 | } | |
1da177e4 LT |
3442 | EXPORT_SYMBOL(__wake_up); |
3443 | ||
3444 | /* | |
3445 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3446 | */ | |
63b20011 | 3447 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 3448 | { |
63b20011 | 3449 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 3450 | } |
22c43c81 | 3451 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3452 | |
4ede816a DL |
3453 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3454 | { | |
3455 | __wake_up_common(q, mode, 1, 0, key); | |
3456 | } | |
bf294b41 | 3457 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3458 | |
1da177e4 | 3459 | /** |
4ede816a | 3460 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3461 | * @q: the waitqueue |
3462 | * @mode: which threads | |
3463 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3464 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3465 | * |
3466 | * The sync wakeup differs that the waker knows that it will schedule | |
3467 | * away soon, so while the target thread will be woken up, it will not | |
3468 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3469 | * with each other. This can prevent needless bouncing between CPUs. | |
3470 | * | |
3471 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3472 | * |
3473 | * It may be assumed that this function implies a write memory barrier before | |
3474 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3475 | */ |
4ede816a DL |
3476 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3477 | int nr_exclusive, void *key) | |
1da177e4 LT |
3478 | { |
3479 | unsigned long flags; | |
7d478721 | 3480 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3481 | |
3482 | if (unlikely(!q)) | |
3483 | return; | |
3484 | ||
3485 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3486 | wake_flags = 0; |
1da177e4 LT |
3487 | |
3488 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3489 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3490 | spin_unlock_irqrestore(&q->lock, flags); |
3491 | } | |
4ede816a DL |
3492 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3493 | ||
3494 | /* | |
3495 | * __wake_up_sync - see __wake_up_sync_key() | |
3496 | */ | |
3497 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3498 | { | |
3499 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3500 | } | |
1da177e4 LT |
3501 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3502 | ||
65eb3dc6 KD |
3503 | /** |
3504 | * complete: - signals a single thread waiting on this completion | |
3505 | * @x: holds the state of this particular completion | |
3506 | * | |
3507 | * This will wake up a single thread waiting on this completion. Threads will be | |
3508 | * awakened in the same order in which they were queued. | |
3509 | * | |
3510 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3511 | * |
3512 | * It may be assumed that this function implies a write memory barrier before | |
3513 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3514 | */ |
b15136e9 | 3515 | void complete(struct completion *x) |
1da177e4 LT |
3516 | { |
3517 | unsigned long flags; | |
3518 | ||
3519 | spin_lock_irqsave(&x->wait.lock, flags); | |
3520 | x->done++; | |
d9514f6c | 3521 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3522 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3523 | } | |
3524 | EXPORT_SYMBOL(complete); | |
3525 | ||
65eb3dc6 KD |
3526 | /** |
3527 | * complete_all: - signals all threads waiting on this completion | |
3528 | * @x: holds the state of this particular completion | |
3529 | * | |
3530 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3531 | * |
3532 | * It may be assumed that this function implies a write memory barrier before | |
3533 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3534 | */ |
b15136e9 | 3535 | void complete_all(struct completion *x) |
1da177e4 LT |
3536 | { |
3537 | unsigned long flags; | |
3538 | ||
3539 | spin_lock_irqsave(&x->wait.lock, flags); | |
3540 | x->done += UINT_MAX/2; | |
d9514f6c | 3541 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3542 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3543 | } | |
3544 | EXPORT_SYMBOL(complete_all); | |
3545 | ||
8cbbe86d AK |
3546 | static inline long __sched |
3547 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3548 | { |
1da177e4 LT |
3549 | if (!x->done) { |
3550 | DECLARE_WAITQUEUE(wait, current); | |
3551 | ||
a93d2f17 | 3552 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3553 | do { |
94d3d824 | 3554 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3555 | timeout = -ERESTARTSYS; |
3556 | break; | |
8cbbe86d AK |
3557 | } |
3558 | __set_current_state(state); | |
1da177e4 LT |
3559 | spin_unlock_irq(&x->wait.lock); |
3560 | timeout = schedule_timeout(timeout); | |
3561 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3562 | } while (!x->done && timeout); |
1da177e4 | 3563 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3564 | if (!x->done) |
3565 | return timeout; | |
1da177e4 LT |
3566 | } |
3567 | x->done--; | |
ea71a546 | 3568 | return timeout ?: 1; |
1da177e4 | 3569 | } |
1da177e4 | 3570 | |
8cbbe86d AK |
3571 | static long __sched |
3572 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3573 | { |
1da177e4 LT |
3574 | might_sleep(); |
3575 | ||
3576 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3577 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3578 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3579 | return timeout; |
3580 | } | |
1da177e4 | 3581 | |
65eb3dc6 KD |
3582 | /** |
3583 | * wait_for_completion: - waits for completion of a task | |
3584 | * @x: holds the state of this particular completion | |
3585 | * | |
3586 | * This waits to be signaled for completion of a specific task. It is NOT | |
3587 | * interruptible and there is no timeout. | |
3588 | * | |
3589 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3590 | * and interrupt capability. Also see complete(). | |
3591 | */ | |
b15136e9 | 3592 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3593 | { |
3594 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3595 | } |
8cbbe86d | 3596 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3597 | |
65eb3dc6 KD |
3598 | /** |
3599 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3600 | * @x: holds the state of this particular completion | |
3601 | * @timeout: timeout value in jiffies | |
3602 | * | |
3603 | * This waits for either a completion of a specific task to be signaled or for a | |
3604 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3605 | * interruptible. | |
c6dc7f05 BF |
3606 | * |
3607 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3608 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3609 | */ |
b15136e9 | 3610 | unsigned long __sched |
8cbbe86d | 3611 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3612 | { |
8cbbe86d | 3613 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3614 | } |
8cbbe86d | 3615 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3616 | |
65eb3dc6 KD |
3617 | /** |
3618 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3619 | * @x: holds the state of this particular completion | |
3620 | * | |
3621 | * This waits for completion of a specific task to be signaled. It is | |
3622 | * interruptible. | |
c6dc7f05 BF |
3623 | * |
3624 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3625 | */ |
8cbbe86d | 3626 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3627 | { |
51e97990 AK |
3628 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3629 | if (t == -ERESTARTSYS) | |
3630 | return t; | |
3631 | return 0; | |
0fec171c | 3632 | } |
8cbbe86d | 3633 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3634 | |
65eb3dc6 KD |
3635 | /** |
3636 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3637 | * @x: holds the state of this particular completion | |
3638 | * @timeout: timeout value in jiffies | |
3639 | * | |
3640 | * This waits for either a completion of a specific task to be signaled or for a | |
3641 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3642 | * |
3643 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3644 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3645 | */ |
6bf41237 | 3646 | long __sched |
8cbbe86d AK |
3647 | wait_for_completion_interruptible_timeout(struct completion *x, |
3648 | unsigned long timeout) | |
0fec171c | 3649 | { |
8cbbe86d | 3650 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3651 | } |
8cbbe86d | 3652 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3653 | |
65eb3dc6 KD |
3654 | /** |
3655 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3656 | * @x: holds the state of this particular completion | |
3657 | * | |
3658 | * This waits to be signaled for completion of a specific task. It can be | |
3659 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3660 | * |
3661 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3662 | */ |
009e577e MW |
3663 | int __sched wait_for_completion_killable(struct completion *x) |
3664 | { | |
3665 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3666 | if (t == -ERESTARTSYS) | |
3667 | return t; | |
3668 | return 0; | |
3669 | } | |
3670 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3671 | ||
0aa12fb4 SW |
3672 | /** |
3673 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3674 | * @x: holds the state of this particular completion | |
3675 | * @timeout: timeout value in jiffies | |
3676 | * | |
3677 | * This waits for either a completion of a specific task to be | |
3678 | * signaled or for a specified timeout to expire. It can be | |
3679 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3680 | * |
3681 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3682 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3683 | */ |
6bf41237 | 3684 | long __sched |
0aa12fb4 SW |
3685 | wait_for_completion_killable_timeout(struct completion *x, |
3686 | unsigned long timeout) | |
3687 | { | |
3688 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3689 | } | |
3690 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3691 | ||
be4de352 DC |
3692 | /** |
3693 | * try_wait_for_completion - try to decrement a completion without blocking | |
3694 | * @x: completion structure | |
3695 | * | |
3696 | * Returns: 0 if a decrement cannot be done without blocking | |
3697 | * 1 if a decrement succeeded. | |
3698 | * | |
3699 | * If a completion is being used as a counting completion, | |
3700 | * attempt to decrement the counter without blocking. This | |
3701 | * enables us to avoid waiting if the resource the completion | |
3702 | * is protecting is not available. | |
3703 | */ | |
3704 | bool try_wait_for_completion(struct completion *x) | |
3705 | { | |
7539a3b3 | 3706 | unsigned long flags; |
be4de352 DC |
3707 | int ret = 1; |
3708 | ||
7539a3b3 | 3709 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3710 | if (!x->done) |
3711 | ret = 0; | |
3712 | else | |
3713 | x->done--; | |
7539a3b3 | 3714 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3715 | return ret; |
3716 | } | |
3717 | EXPORT_SYMBOL(try_wait_for_completion); | |
3718 | ||
3719 | /** | |
3720 | * completion_done - Test to see if a completion has any waiters | |
3721 | * @x: completion structure | |
3722 | * | |
3723 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3724 | * 1 if there are no waiters. | |
3725 | * | |
3726 | */ | |
3727 | bool completion_done(struct completion *x) | |
3728 | { | |
7539a3b3 | 3729 | unsigned long flags; |
be4de352 DC |
3730 | int ret = 1; |
3731 | ||
7539a3b3 | 3732 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3733 | if (!x->done) |
3734 | ret = 0; | |
7539a3b3 | 3735 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3736 | return ret; |
3737 | } | |
3738 | EXPORT_SYMBOL(completion_done); | |
3739 | ||
8cbbe86d AK |
3740 | static long __sched |
3741 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3742 | { |
0fec171c IM |
3743 | unsigned long flags; |
3744 | wait_queue_t wait; | |
3745 | ||
3746 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3747 | |
8cbbe86d | 3748 | __set_current_state(state); |
1da177e4 | 3749 | |
8cbbe86d AK |
3750 | spin_lock_irqsave(&q->lock, flags); |
3751 | __add_wait_queue(q, &wait); | |
3752 | spin_unlock(&q->lock); | |
3753 | timeout = schedule_timeout(timeout); | |
3754 | spin_lock_irq(&q->lock); | |
3755 | __remove_wait_queue(q, &wait); | |
3756 | spin_unlock_irqrestore(&q->lock, flags); | |
3757 | ||
3758 | return timeout; | |
3759 | } | |
3760 | ||
3761 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3762 | { | |
3763 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3764 | } |
1da177e4 LT |
3765 | EXPORT_SYMBOL(interruptible_sleep_on); |
3766 | ||
0fec171c | 3767 | long __sched |
95cdf3b7 | 3768 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3769 | { |
8cbbe86d | 3770 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3771 | } |
1da177e4 LT |
3772 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3773 | ||
0fec171c | 3774 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3775 | { |
8cbbe86d | 3776 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3777 | } |
1da177e4 LT |
3778 | EXPORT_SYMBOL(sleep_on); |
3779 | ||
0fec171c | 3780 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3781 | { |
8cbbe86d | 3782 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3783 | } |
1da177e4 LT |
3784 | EXPORT_SYMBOL(sleep_on_timeout); |
3785 | ||
b29739f9 IM |
3786 | #ifdef CONFIG_RT_MUTEXES |
3787 | ||
3788 | /* | |
3789 | * rt_mutex_setprio - set the current priority of a task | |
3790 | * @p: task | |
3791 | * @prio: prio value (kernel-internal form) | |
3792 | * | |
3793 | * This function changes the 'effective' priority of a task. It does | |
3794 | * not touch ->normal_prio like __setscheduler(). | |
3795 | * | |
3796 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3797 | */ | |
36c8b586 | 3798 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3799 | { |
83b699ed | 3800 | int oldprio, on_rq, running; |
70b97a7f | 3801 | struct rq *rq; |
83ab0aa0 | 3802 | const struct sched_class *prev_class; |
b29739f9 IM |
3803 | |
3804 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3805 | ||
0122ec5b | 3806 | rq = __task_rq_lock(p); |
b29739f9 | 3807 | |
1c4dd99b TG |
3808 | /* |
3809 | * Idle task boosting is a nono in general. There is one | |
3810 | * exception, when PREEMPT_RT and NOHZ is active: | |
3811 | * | |
3812 | * The idle task calls get_next_timer_interrupt() and holds | |
3813 | * the timer wheel base->lock on the CPU and another CPU wants | |
3814 | * to access the timer (probably to cancel it). We can safely | |
3815 | * ignore the boosting request, as the idle CPU runs this code | |
3816 | * with interrupts disabled and will complete the lock | |
3817 | * protected section without being interrupted. So there is no | |
3818 | * real need to boost. | |
3819 | */ | |
3820 | if (unlikely(p == rq->idle)) { | |
3821 | WARN_ON(p != rq->curr); | |
3822 | WARN_ON(p->pi_blocked_on); | |
3823 | goto out_unlock; | |
3824 | } | |
3825 | ||
a8027073 | 3826 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3827 | oldprio = p->prio; |
83ab0aa0 | 3828 | prev_class = p->sched_class; |
fd2f4419 | 3829 | on_rq = p->on_rq; |
051a1d1a | 3830 | running = task_current(rq, p); |
0e1f3483 | 3831 | if (on_rq) |
69be72c1 | 3832 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3833 | if (running) |
3834 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
3835 | |
3836 | if (rt_prio(prio)) | |
3837 | p->sched_class = &rt_sched_class; | |
3838 | else | |
3839 | p->sched_class = &fair_sched_class; | |
3840 | ||
b29739f9 IM |
3841 | p->prio = prio; |
3842 | ||
0e1f3483 HS |
3843 | if (running) |
3844 | p->sched_class->set_curr_task(rq); | |
da7a735e | 3845 | if (on_rq) |
371fd7e7 | 3846 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 3847 | |
da7a735e | 3848 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3849 | out_unlock: |
0122ec5b | 3850 | __task_rq_unlock(rq); |
b29739f9 | 3851 | } |
b29739f9 | 3852 | #endif |
36c8b586 | 3853 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3854 | { |
dd41f596 | 3855 | int old_prio, delta, on_rq; |
1da177e4 | 3856 | unsigned long flags; |
70b97a7f | 3857 | struct rq *rq; |
1da177e4 LT |
3858 | |
3859 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3860 | return; | |
3861 | /* | |
3862 | * We have to be careful, if called from sys_setpriority(), | |
3863 | * the task might be in the middle of scheduling on another CPU. | |
3864 | */ | |
3865 | rq = task_rq_lock(p, &flags); | |
3866 | /* | |
3867 | * The RT priorities are set via sched_setscheduler(), but we still | |
3868 | * allow the 'normal' nice value to be set - but as expected | |
3869 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3870 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3871 | */ |
e05606d3 | 3872 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3873 | p->static_prio = NICE_TO_PRIO(nice); |
3874 | goto out_unlock; | |
3875 | } | |
fd2f4419 | 3876 | on_rq = p->on_rq; |
c09595f6 | 3877 | if (on_rq) |
69be72c1 | 3878 | dequeue_task(rq, p, 0); |
1da177e4 | 3879 | |
1da177e4 | 3880 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3881 | set_load_weight(p); |
b29739f9 IM |
3882 | old_prio = p->prio; |
3883 | p->prio = effective_prio(p); | |
3884 | delta = p->prio - old_prio; | |
1da177e4 | 3885 | |
dd41f596 | 3886 | if (on_rq) { |
371fd7e7 | 3887 | enqueue_task(rq, p, 0); |
1da177e4 | 3888 | /* |
d5f9f942 AM |
3889 | * If the task increased its priority or is running and |
3890 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3891 | */ |
d5f9f942 | 3892 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3893 | resched_task(rq->curr); |
3894 | } | |
3895 | out_unlock: | |
0122ec5b | 3896 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3897 | } |
1da177e4 LT |
3898 | EXPORT_SYMBOL(set_user_nice); |
3899 | ||
e43379f1 MM |
3900 | /* |
3901 | * can_nice - check if a task can reduce its nice value | |
3902 | * @p: task | |
3903 | * @nice: nice value | |
3904 | */ | |
36c8b586 | 3905 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3906 | { |
024f4747 MM |
3907 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3908 | int nice_rlim = 20 - nice; | |
48f24c4d | 3909 | |
78d7d407 | 3910 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3911 | capable(CAP_SYS_NICE)); |
3912 | } | |
3913 | ||
1da177e4 LT |
3914 | #ifdef __ARCH_WANT_SYS_NICE |
3915 | ||
3916 | /* | |
3917 | * sys_nice - change the priority of the current process. | |
3918 | * @increment: priority increment | |
3919 | * | |
3920 | * sys_setpriority is a more generic, but much slower function that | |
3921 | * does similar things. | |
3922 | */ | |
5add95d4 | 3923 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3924 | { |
48f24c4d | 3925 | long nice, retval; |
1da177e4 LT |
3926 | |
3927 | /* | |
3928 | * Setpriority might change our priority at the same moment. | |
3929 | * We don't have to worry. Conceptually one call occurs first | |
3930 | * and we have a single winner. | |
3931 | */ | |
e43379f1 MM |
3932 | if (increment < -40) |
3933 | increment = -40; | |
1da177e4 LT |
3934 | if (increment > 40) |
3935 | increment = 40; | |
3936 | ||
2b8f836f | 3937 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
3938 | if (nice < -20) |
3939 | nice = -20; | |
3940 | if (nice > 19) | |
3941 | nice = 19; | |
3942 | ||
e43379f1 MM |
3943 | if (increment < 0 && !can_nice(current, nice)) |
3944 | return -EPERM; | |
3945 | ||
1da177e4 LT |
3946 | retval = security_task_setnice(current, nice); |
3947 | if (retval) | |
3948 | return retval; | |
3949 | ||
3950 | set_user_nice(current, nice); | |
3951 | return 0; | |
3952 | } | |
3953 | ||
3954 | #endif | |
3955 | ||
3956 | /** | |
3957 | * task_prio - return the priority value of a given task. | |
3958 | * @p: the task in question. | |
3959 | * | |
3960 | * This is the priority value as seen by users in /proc. | |
3961 | * RT tasks are offset by -200. Normal tasks are centered | |
3962 | * around 0, value goes from -16 to +15. | |
3963 | */ | |
36c8b586 | 3964 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3965 | { |
3966 | return p->prio - MAX_RT_PRIO; | |
3967 | } | |
3968 | ||
3969 | /** | |
3970 | * task_nice - return the nice value of a given task. | |
3971 | * @p: the task in question. | |
3972 | */ | |
36c8b586 | 3973 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3974 | { |
3975 | return TASK_NICE(p); | |
3976 | } | |
150d8bed | 3977 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
3978 | |
3979 | /** | |
3980 | * idle_cpu - is a given cpu idle currently? | |
3981 | * @cpu: the processor in question. | |
3982 | */ | |
3983 | int idle_cpu(int cpu) | |
3984 | { | |
908a3283 TG |
3985 | struct rq *rq = cpu_rq(cpu); |
3986 | ||
3987 | if (rq->curr != rq->idle) | |
3988 | return 0; | |
3989 | ||
3990 | if (rq->nr_running) | |
3991 | return 0; | |
3992 | ||
3993 | #ifdef CONFIG_SMP | |
3994 | if (!llist_empty(&rq->wake_list)) | |
3995 | return 0; | |
3996 | #endif | |
3997 | ||
3998 | return 1; | |
1da177e4 LT |
3999 | } |
4000 | ||
1da177e4 LT |
4001 | /** |
4002 | * idle_task - return the idle task for a given cpu. | |
4003 | * @cpu: the processor in question. | |
4004 | */ | |
36c8b586 | 4005 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4006 | { |
4007 | return cpu_rq(cpu)->idle; | |
4008 | } | |
4009 | ||
4010 | /** | |
4011 | * find_process_by_pid - find a process with a matching PID value. | |
4012 | * @pid: the pid in question. | |
4013 | */ | |
a9957449 | 4014 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4015 | { |
228ebcbe | 4016 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4017 | } |
4018 | ||
4019 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4020 | static void |
4021 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4022 | { |
1da177e4 LT |
4023 | p->policy = policy; |
4024 | p->rt_priority = prio; | |
b29739f9 IM |
4025 | p->normal_prio = normal_prio(p); |
4026 | /* we are holding p->pi_lock already */ | |
4027 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4028 | if (rt_prio(p->prio)) |
4029 | p->sched_class = &rt_sched_class; | |
4030 | else | |
4031 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4032 | set_load_weight(p); |
1da177e4 LT |
4033 | } |
4034 | ||
c69e8d9c DH |
4035 | /* |
4036 | * check the target process has a UID that matches the current process's | |
4037 | */ | |
4038 | static bool check_same_owner(struct task_struct *p) | |
4039 | { | |
4040 | const struct cred *cred = current_cred(), *pcred; | |
4041 | bool match; | |
4042 | ||
4043 | rcu_read_lock(); | |
4044 | pcred = __task_cred(p); | |
b0e77598 SH |
4045 | if (cred->user->user_ns == pcred->user->user_ns) |
4046 | match = (cred->euid == pcred->euid || | |
4047 | cred->euid == pcred->uid); | |
4048 | else | |
4049 | match = false; | |
c69e8d9c DH |
4050 | rcu_read_unlock(); |
4051 | return match; | |
4052 | } | |
4053 | ||
961ccddd | 4054 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4055 | const struct sched_param *param, bool user) |
1da177e4 | 4056 | { |
83b699ed | 4057 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4058 | unsigned long flags; |
83ab0aa0 | 4059 | const struct sched_class *prev_class; |
70b97a7f | 4060 | struct rq *rq; |
ca94c442 | 4061 | int reset_on_fork; |
1da177e4 | 4062 | |
66e5393a SR |
4063 | /* may grab non-irq protected spin_locks */ |
4064 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4065 | recheck: |
4066 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4067 | if (policy < 0) { |
4068 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4069 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4070 | } else { |
4071 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4072 | policy &= ~SCHED_RESET_ON_FORK; | |
4073 | ||
4074 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4075 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4076 | policy != SCHED_IDLE) | |
4077 | return -EINVAL; | |
4078 | } | |
4079 | ||
1da177e4 LT |
4080 | /* |
4081 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4082 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4083 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4084 | */ |
4085 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4086 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4087 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4088 | return -EINVAL; |
e05606d3 | 4089 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4090 | return -EINVAL; |
4091 | ||
37e4ab3f OC |
4092 | /* |
4093 | * Allow unprivileged RT tasks to decrease priority: | |
4094 | */ | |
961ccddd | 4095 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4096 | if (rt_policy(policy)) { |
a44702e8 ON |
4097 | unsigned long rlim_rtprio = |
4098 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4099 | |
4100 | /* can't set/change the rt policy */ | |
4101 | if (policy != p->policy && !rlim_rtprio) | |
4102 | return -EPERM; | |
4103 | ||
4104 | /* can't increase priority */ | |
4105 | if (param->sched_priority > p->rt_priority && | |
4106 | param->sched_priority > rlim_rtprio) | |
4107 | return -EPERM; | |
4108 | } | |
c02aa73b | 4109 | |
dd41f596 | 4110 | /* |
c02aa73b DH |
4111 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4112 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4113 | */ |
c02aa73b DH |
4114 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4115 | if (!can_nice(p, TASK_NICE(p))) | |
4116 | return -EPERM; | |
4117 | } | |
5fe1d75f | 4118 | |
37e4ab3f | 4119 | /* can't change other user's priorities */ |
c69e8d9c | 4120 | if (!check_same_owner(p)) |
37e4ab3f | 4121 | return -EPERM; |
ca94c442 LP |
4122 | |
4123 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4124 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4125 | return -EPERM; | |
37e4ab3f | 4126 | } |
1da177e4 | 4127 | |
725aad24 | 4128 | if (user) { |
b0ae1981 | 4129 | retval = security_task_setscheduler(p); |
725aad24 JF |
4130 | if (retval) |
4131 | return retval; | |
4132 | } | |
4133 | ||
b29739f9 IM |
4134 | /* |
4135 | * make sure no PI-waiters arrive (or leave) while we are | |
4136 | * changing the priority of the task: | |
0122ec5b | 4137 | * |
25985edc | 4138 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4139 | * runqueue lock must be held. |
4140 | */ | |
0122ec5b | 4141 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 4142 | |
34f971f6 PZ |
4143 | /* |
4144 | * Changing the policy of the stop threads its a very bad idea | |
4145 | */ | |
4146 | if (p == rq->stop) { | |
0122ec5b | 4147 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
4148 | return -EINVAL; |
4149 | } | |
4150 | ||
a51e9198 DF |
4151 | /* |
4152 | * If not changing anything there's no need to proceed further: | |
4153 | */ | |
4154 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
4155 | param->sched_priority == p->rt_priority))) { | |
4156 | ||
4157 | __task_rq_unlock(rq); | |
4158 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4159 | return 0; | |
4160 | } | |
4161 | ||
dc61b1d6 PZ |
4162 | #ifdef CONFIG_RT_GROUP_SCHED |
4163 | if (user) { | |
4164 | /* | |
4165 | * Do not allow realtime tasks into groups that have no runtime | |
4166 | * assigned. | |
4167 | */ | |
4168 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4169 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4170 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4171 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4172 | return -EPERM; |
4173 | } | |
4174 | } | |
4175 | #endif | |
4176 | ||
1da177e4 LT |
4177 | /* recheck policy now with rq lock held */ |
4178 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4179 | policy = oldpolicy = -1; | |
0122ec5b | 4180 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4181 | goto recheck; |
4182 | } | |
fd2f4419 | 4183 | on_rq = p->on_rq; |
051a1d1a | 4184 | running = task_current(rq, p); |
0e1f3483 | 4185 | if (on_rq) |
4ca9b72b | 4186 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4187 | if (running) |
4188 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4189 | |
ca94c442 LP |
4190 | p->sched_reset_on_fork = reset_on_fork; |
4191 | ||
1da177e4 | 4192 | oldprio = p->prio; |
83ab0aa0 | 4193 | prev_class = p->sched_class; |
dd41f596 | 4194 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4195 | |
0e1f3483 HS |
4196 | if (running) |
4197 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4198 | if (on_rq) |
4ca9b72b | 4199 | enqueue_task(rq, p, 0); |
cb469845 | 4200 | |
da7a735e | 4201 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4202 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4203 | |
95e02ca9 TG |
4204 | rt_mutex_adjust_pi(p); |
4205 | ||
1da177e4 LT |
4206 | return 0; |
4207 | } | |
961ccddd RR |
4208 | |
4209 | /** | |
4210 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4211 | * @p: the task in question. | |
4212 | * @policy: new policy. | |
4213 | * @param: structure containing the new RT priority. | |
4214 | * | |
4215 | * NOTE that the task may be already dead. | |
4216 | */ | |
4217 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4218 | const struct sched_param *param) |
961ccddd RR |
4219 | { |
4220 | return __sched_setscheduler(p, policy, param, true); | |
4221 | } | |
1da177e4 LT |
4222 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4223 | ||
961ccddd RR |
4224 | /** |
4225 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4226 | * @p: the task in question. | |
4227 | * @policy: new policy. | |
4228 | * @param: structure containing the new RT priority. | |
4229 | * | |
4230 | * Just like sched_setscheduler, only don't bother checking if the | |
4231 | * current context has permission. For example, this is needed in | |
4232 | * stop_machine(): we create temporary high priority worker threads, | |
4233 | * but our caller might not have that capability. | |
4234 | */ | |
4235 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4236 | const struct sched_param *param) |
961ccddd RR |
4237 | { |
4238 | return __sched_setscheduler(p, policy, param, false); | |
4239 | } | |
4240 | ||
95cdf3b7 IM |
4241 | static int |
4242 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4243 | { |
1da177e4 LT |
4244 | struct sched_param lparam; |
4245 | struct task_struct *p; | |
36c8b586 | 4246 | int retval; |
1da177e4 LT |
4247 | |
4248 | if (!param || pid < 0) | |
4249 | return -EINVAL; | |
4250 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4251 | return -EFAULT; | |
5fe1d75f ON |
4252 | |
4253 | rcu_read_lock(); | |
4254 | retval = -ESRCH; | |
1da177e4 | 4255 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4256 | if (p != NULL) |
4257 | retval = sched_setscheduler(p, policy, &lparam); | |
4258 | rcu_read_unlock(); | |
36c8b586 | 4259 | |
1da177e4 LT |
4260 | return retval; |
4261 | } | |
4262 | ||
4263 | /** | |
4264 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4265 | * @pid: the pid in question. | |
4266 | * @policy: new policy. | |
4267 | * @param: structure containing the new RT priority. | |
4268 | */ | |
5add95d4 HC |
4269 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4270 | struct sched_param __user *, param) | |
1da177e4 | 4271 | { |
c21761f1 JB |
4272 | /* negative values for policy are not valid */ |
4273 | if (policy < 0) | |
4274 | return -EINVAL; | |
4275 | ||
1da177e4 LT |
4276 | return do_sched_setscheduler(pid, policy, param); |
4277 | } | |
4278 | ||
4279 | /** | |
4280 | * sys_sched_setparam - set/change the RT priority of a thread | |
4281 | * @pid: the pid in question. | |
4282 | * @param: structure containing the new RT priority. | |
4283 | */ | |
5add95d4 | 4284 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4285 | { |
4286 | return do_sched_setscheduler(pid, -1, param); | |
4287 | } | |
4288 | ||
4289 | /** | |
4290 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4291 | * @pid: the pid in question. | |
4292 | */ | |
5add95d4 | 4293 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4294 | { |
36c8b586 | 4295 | struct task_struct *p; |
3a5c359a | 4296 | int retval; |
1da177e4 LT |
4297 | |
4298 | if (pid < 0) | |
3a5c359a | 4299 | return -EINVAL; |
1da177e4 LT |
4300 | |
4301 | retval = -ESRCH; | |
5fe85be0 | 4302 | rcu_read_lock(); |
1da177e4 LT |
4303 | p = find_process_by_pid(pid); |
4304 | if (p) { | |
4305 | retval = security_task_getscheduler(p); | |
4306 | if (!retval) | |
ca94c442 LP |
4307 | retval = p->policy |
4308 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4309 | } |
5fe85be0 | 4310 | rcu_read_unlock(); |
1da177e4 LT |
4311 | return retval; |
4312 | } | |
4313 | ||
4314 | /** | |
ca94c442 | 4315 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4316 | * @pid: the pid in question. |
4317 | * @param: structure containing the RT priority. | |
4318 | */ | |
5add95d4 | 4319 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4320 | { |
4321 | struct sched_param lp; | |
36c8b586 | 4322 | struct task_struct *p; |
3a5c359a | 4323 | int retval; |
1da177e4 LT |
4324 | |
4325 | if (!param || pid < 0) | |
3a5c359a | 4326 | return -EINVAL; |
1da177e4 | 4327 | |
5fe85be0 | 4328 | rcu_read_lock(); |
1da177e4 LT |
4329 | p = find_process_by_pid(pid); |
4330 | retval = -ESRCH; | |
4331 | if (!p) | |
4332 | goto out_unlock; | |
4333 | ||
4334 | retval = security_task_getscheduler(p); | |
4335 | if (retval) | |
4336 | goto out_unlock; | |
4337 | ||
4338 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4339 | rcu_read_unlock(); |
1da177e4 LT |
4340 | |
4341 | /* | |
4342 | * This one might sleep, we cannot do it with a spinlock held ... | |
4343 | */ | |
4344 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4345 | ||
1da177e4 LT |
4346 | return retval; |
4347 | ||
4348 | out_unlock: | |
5fe85be0 | 4349 | rcu_read_unlock(); |
1da177e4 LT |
4350 | return retval; |
4351 | } | |
4352 | ||
96f874e2 | 4353 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4354 | { |
5a16f3d3 | 4355 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4356 | struct task_struct *p; |
4357 | int retval; | |
1da177e4 | 4358 | |
95402b38 | 4359 | get_online_cpus(); |
23f5d142 | 4360 | rcu_read_lock(); |
1da177e4 LT |
4361 | |
4362 | p = find_process_by_pid(pid); | |
4363 | if (!p) { | |
23f5d142 | 4364 | rcu_read_unlock(); |
95402b38 | 4365 | put_online_cpus(); |
1da177e4 LT |
4366 | return -ESRCH; |
4367 | } | |
4368 | ||
23f5d142 | 4369 | /* Prevent p going away */ |
1da177e4 | 4370 | get_task_struct(p); |
23f5d142 | 4371 | rcu_read_unlock(); |
1da177e4 | 4372 | |
5a16f3d3 RR |
4373 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4374 | retval = -ENOMEM; | |
4375 | goto out_put_task; | |
4376 | } | |
4377 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4378 | retval = -ENOMEM; | |
4379 | goto out_free_cpus_allowed; | |
4380 | } | |
1da177e4 | 4381 | retval = -EPERM; |
f1c84dae | 4382 | if (!check_same_owner(p) && !ns_capable(task_user_ns(p), CAP_SYS_NICE)) |
1da177e4 LT |
4383 | goto out_unlock; |
4384 | ||
b0ae1981 | 4385 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4386 | if (retval) |
4387 | goto out_unlock; | |
4388 | ||
5a16f3d3 RR |
4389 | cpuset_cpus_allowed(p, cpus_allowed); |
4390 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4391 | again: |
5a16f3d3 | 4392 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4393 | |
8707d8b8 | 4394 | if (!retval) { |
5a16f3d3 RR |
4395 | cpuset_cpus_allowed(p, cpus_allowed); |
4396 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4397 | /* |
4398 | * We must have raced with a concurrent cpuset | |
4399 | * update. Just reset the cpus_allowed to the | |
4400 | * cpuset's cpus_allowed | |
4401 | */ | |
5a16f3d3 | 4402 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4403 | goto again; |
4404 | } | |
4405 | } | |
1da177e4 | 4406 | out_unlock: |
5a16f3d3 RR |
4407 | free_cpumask_var(new_mask); |
4408 | out_free_cpus_allowed: | |
4409 | free_cpumask_var(cpus_allowed); | |
4410 | out_put_task: | |
1da177e4 | 4411 | put_task_struct(p); |
95402b38 | 4412 | put_online_cpus(); |
1da177e4 LT |
4413 | return retval; |
4414 | } | |
4415 | ||
4416 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4417 | struct cpumask *new_mask) |
1da177e4 | 4418 | { |
96f874e2 RR |
4419 | if (len < cpumask_size()) |
4420 | cpumask_clear(new_mask); | |
4421 | else if (len > cpumask_size()) | |
4422 | len = cpumask_size(); | |
4423 | ||
1da177e4 LT |
4424 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4425 | } | |
4426 | ||
4427 | /** | |
4428 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4429 | * @pid: pid of the process | |
4430 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4431 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4432 | */ | |
5add95d4 HC |
4433 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4434 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4435 | { |
5a16f3d3 | 4436 | cpumask_var_t new_mask; |
1da177e4 LT |
4437 | int retval; |
4438 | ||
5a16f3d3 RR |
4439 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4440 | return -ENOMEM; | |
1da177e4 | 4441 | |
5a16f3d3 RR |
4442 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4443 | if (retval == 0) | |
4444 | retval = sched_setaffinity(pid, new_mask); | |
4445 | free_cpumask_var(new_mask); | |
4446 | return retval; | |
1da177e4 LT |
4447 | } |
4448 | ||
96f874e2 | 4449 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4450 | { |
36c8b586 | 4451 | struct task_struct *p; |
31605683 | 4452 | unsigned long flags; |
1da177e4 | 4453 | int retval; |
1da177e4 | 4454 | |
95402b38 | 4455 | get_online_cpus(); |
23f5d142 | 4456 | rcu_read_lock(); |
1da177e4 LT |
4457 | |
4458 | retval = -ESRCH; | |
4459 | p = find_process_by_pid(pid); | |
4460 | if (!p) | |
4461 | goto out_unlock; | |
4462 | ||
e7834f8f DQ |
4463 | retval = security_task_getscheduler(p); |
4464 | if (retval) | |
4465 | goto out_unlock; | |
4466 | ||
013fdb80 | 4467 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4468 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4469 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4470 | |
4471 | out_unlock: | |
23f5d142 | 4472 | rcu_read_unlock(); |
95402b38 | 4473 | put_online_cpus(); |
1da177e4 | 4474 | |
9531b62f | 4475 | return retval; |
1da177e4 LT |
4476 | } |
4477 | ||
4478 | /** | |
4479 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4480 | * @pid: pid of the process | |
4481 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4482 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4483 | */ | |
5add95d4 HC |
4484 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4485 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4486 | { |
4487 | int ret; | |
f17c8607 | 4488 | cpumask_var_t mask; |
1da177e4 | 4489 | |
84fba5ec | 4490 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4491 | return -EINVAL; |
4492 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4493 | return -EINVAL; |
4494 | ||
f17c8607 RR |
4495 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4496 | return -ENOMEM; | |
1da177e4 | 4497 | |
f17c8607 RR |
4498 | ret = sched_getaffinity(pid, mask); |
4499 | if (ret == 0) { | |
8bc037fb | 4500 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4501 | |
4502 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4503 | ret = -EFAULT; |
4504 | else | |
cd3d8031 | 4505 | ret = retlen; |
f17c8607 RR |
4506 | } |
4507 | free_cpumask_var(mask); | |
1da177e4 | 4508 | |
f17c8607 | 4509 | return ret; |
1da177e4 LT |
4510 | } |
4511 | ||
4512 | /** | |
4513 | * sys_sched_yield - yield the current processor to other threads. | |
4514 | * | |
dd41f596 IM |
4515 | * This function yields the current CPU to other tasks. If there are no |
4516 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4517 | */ |
5add95d4 | 4518 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4519 | { |
70b97a7f | 4520 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4521 | |
2d72376b | 4522 | schedstat_inc(rq, yld_count); |
4530d7ab | 4523 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4524 | |
4525 | /* | |
4526 | * Since we are going to call schedule() anyway, there's | |
4527 | * no need to preempt or enable interrupts: | |
4528 | */ | |
4529 | __release(rq->lock); | |
8a25d5de | 4530 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4531 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4532 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4533 | |
4534 | schedule(); | |
4535 | ||
4536 | return 0; | |
4537 | } | |
4538 | ||
d86ee480 PZ |
4539 | static inline int should_resched(void) |
4540 | { | |
4541 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4542 | } | |
4543 | ||
e7b38404 | 4544 | static void __cond_resched(void) |
1da177e4 | 4545 | { |
e7aaaa69 | 4546 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4547 | __schedule(); |
e7aaaa69 | 4548 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4549 | } |
4550 | ||
02b67cc3 | 4551 | int __sched _cond_resched(void) |
1da177e4 | 4552 | { |
d86ee480 | 4553 | if (should_resched()) { |
1da177e4 LT |
4554 | __cond_resched(); |
4555 | return 1; | |
4556 | } | |
4557 | return 0; | |
4558 | } | |
02b67cc3 | 4559 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4560 | |
4561 | /* | |
613afbf8 | 4562 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4563 | * call schedule, and on return reacquire the lock. |
4564 | * | |
41a2d6cf | 4565 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4566 | * operations here to prevent schedule() from being called twice (once via |
4567 | * spin_unlock(), once by hand). | |
4568 | */ | |
613afbf8 | 4569 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4570 | { |
d86ee480 | 4571 | int resched = should_resched(); |
6df3cecb JK |
4572 | int ret = 0; |
4573 | ||
f607c668 PZ |
4574 | lockdep_assert_held(lock); |
4575 | ||
95c354fe | 4576 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4577 | spin_unlock(lock); |
d86ee480 | 4578 | if (resched) |
95c354fe NP |
4579 | __cond_resched(); |
4580 | else | |
4581 | cpu_relax(); | |
6df3cecb | 4582 | ret = 1; |
1da177e4 | 4583 | spin_lock(lock); |
1da177e4 | 4584 | } |
6df3cecb | 4585 | return ret; |
1da177e4 | 4586 | } |
613afbf8 | 4587 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4588 | |
613afbf8 | 4589 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4590 | { |
4591 | BUG_ON(!in_softirq()); | |
4592 | ||
d86ee480 | 4593 | if (should_resched()) { |
98d82567 | 4594 | local_bh_enable(); |
1da177e4 LT |
4595 | __cond_resched(); |
4596 | local_bh_disable(); | |
4597 | return 1; | |
4598 | } | |
4599 | return 0; | |
4600 | } | |
613afbf8 | 4601 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4602 | |
1da177e4 LT |
4603 | /** |
4604 | * yield - yield the current processor to other threads. | |
4605 | * | |
8e3fabfd PZ |
4606 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4607 | * | |
4608 | * The scheduler is at all times free to pick the calling task as the most | |
4609 | * eligible task to run, if removing the yield() call from your code breaks | |
4610 | * it, its already broken. | |
4611 | * | |
4612 | * Typical broken usage is: | |
4613 | * | |
4614 | * while (!event) | |
4615 | * yield(); | |
4616 | * | |
4617 | * where one assumes that yield() will let 'the other' process run that will | |
4618 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4619 | * happen. Never use yield() as a progress guarantee!! | |
4620 | * | |
4621 | * If you want to use yield() to wait for something, use wait_event(). | |
4622 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4623 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4624 | */ |
4625 | void __sched yield(void) | |
4626 | { | |
4627 | set_current_state(TASK_RUNNING); | |
4628 | sys_sched_yield(); | |
4629 | } | |
1da177e4 LT |
4630 | EXPORT_SYMBOL(yield); |
4631 | ||
d95f4122 MG |
4632 | /** |
4633 | * yield_to - yield the current processor to another thread in | |
4634 | * your thread group, or accelerate that thread toward the | |
4635 | * processor it's on. | |
16addf95 RD |
4636 | * @p: target task |
4637 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4638 | * |
4639 | * It's the caller's job to ensure that the target task struct | |
4640 | * can't go away on us before we can do any checks. | |
4641 | * | |
4642 | * Returns true if we indeed boosted the target task. | |
4643 | */ | |
4644 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4645 | { | |
4646 | struct task_struct *curr = current; | |
4647 | struct rq *rq, *p_rq; | |
4648 | unsigned long flags; | |
4649 | bool yielded = 0; | |
4650 | ||
4651 | local_irq_save(flags); | |
4652 | rq = this_rq(); | |
4653 | ||
4654 | again: | |
4655 | p_rq = task_rq(p); | |
4656 | double_rq_lock(rq, p_rq); | |
4657 | while (task_rq(p) != p_rq) { | |
4658 | double_rq_unlock(rq, p_rq); | |
4659 | goto again; | |
4660 | } | |
4661 | ||
4662 | if (!curr->sched_class->yield_to_task) | |
4663 | goto out; | |
4664 | ||
4665 | if (curr->sched_class != p->sched_class) | |
4666 | goto out; | |
4667 | ||
4668 | if (task_running(p_rq, p) || p->state) | |
4669 | goto out; | |
4670 | ||
4671 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4672 | if (yielded) { |
d95f4122 | 4673 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4674 | /* |
4675 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4676 | * fairness. | |
4677 | */ | |
4678 | if (preempt && rq != p_rq) | |
4679 | resched_task(p_rq->curr); | |
916671c0 MG |
4680 | } else { |
4681 | /* | |
4682 | * We might have set it in task_yield_fair(), but are | |
4683 | * not going to schedule(), so don't want to skip | |
4684 | * the next update. | |
4685 | */ | |
4686 | rq->skip_clock_update = 0; | |
6d1cafd8 | 4687 | } |
d95f4122 MG |
4688 | |
4689 | out: | |
4690 | double_rq_unlock(rq, p_rq); | |
4691 | local_irq_restore(flags); | |
4692 | ||
4693 | if (yielded) | |
4694 | schedule(); | |
4695 | ||
4696 | return yielded; | |
4697 | } | |
4698 | EXPORT_SYMBOL_GPL(yield_to); | |
4699 | ||
1da177e4 | 4700 | /* |
41a2d6cf | 4701 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4702 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4703 | */ |
4704 | void __sched io_schedule(void) | |
4705 | { | |
54d35f29 | 4706 | struct rq *rq = raw_rq(); |
1da177e4 | 4707 | |
0ff92245 | 4708 | delayacct_blkio_start(); |
1da177e4 | 4709 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4710 | blk_flush_plug(current); |
8f0dfc34 | 4711 | current->in_iowait = 1; |
1da177e4 | 4712 | schedule(); |
8f0dfc34 | 4713 | current->in_iowait = 0; |
1da177e4 | 4714 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4715 | delayacct_blkio_end(); |
1da177e4 | 4716 | } |
1da177e4 LT |
4717 | EXPORT_SYMBOL(io_schedule); |
4718 | ||
4719 | long __sched io_schedule_timeout(long timeout) | |
4720 | { | |
54d35f29 | 4721 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4722 | long ret; |
4723 | ||
0ff92245 | 4724 | delayacct_blkio_start(); |
1da177e4 | 4725 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4726 | blk_flush_plug(current); |
8f0dfc34 | 4727 | current->in_iowait = 1; |
1da177e4 | 4728 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4729 | current->in_iowait = 0; |
1da177e4 | 4730 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4731 | delayacct_blkio_end(); |
1da177e4 LT |
4732 | return ret; |
4733 | } | |
4734 | ||
4735 | /** | |
4736 | * sys_sched_get_priority_max - return maximum RT priority. | |
4737 | * @policy: scheduling class. | |
4738 | * | |
4739 | * this syscall returns the maximum rt_priority that can be used | |
4740 | * by a given scheduling class. | |
4741 | */ | |
5add95d4 | 4742 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4743 | { |
4744 | int ret = -EINVAL; | |
4745 | ||
4746 | switch (policy) { | |
4747 | case SCHED_FIFO: | |
4748 | case SCHED_RR: | |
4749 | ret = MAX_USER_RT_PRIO-1; | |
4750 | break; | |
4751 | case SCHED_NORMAL: | |
b0a9499c | 4752 | case SCHED_BATCH: |
dd41f596 | 4753 | case SCHED_IDLE: |
1da177e4 LT |
4754 | ret = 0; |
4755 | break; | |
4756 | } | |
4757 | return ret; | |
4758 | } | |
4759 | ||
4760 | /** | |
4761 | * sys_sched_get_priority_min - return minimum RT priority. | |
4762 | * @policy: scheduling class. | |
4763 | * | |
4764 | * this syscall returns the minimum rt_priority that can be used | |
4765 | * by a given scheduling class. | |
4766 | */ | |
5add95d4 | 4767 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4768 | { |
4769 | int ret = -EINVAL; | |
4770 | ||
4771 | switch (policy) { | |
4772 | case SCHED_FIFO: | |
4773 | case SCHED_RR: | |
4774 | ret = 1; | |
4775 | break; | |
4776 | case SCHED_NORMAL: | |
b0a9499c | 4777 | case SCHED_BATCH: |
dd41f596 | 4778 | case SCHED_IDLE: |
1da177e4 LT |
4779 | ret = 0; |
4780 | } | |
4781 | return ret; | |
4782 | } | |
4783 | ||
4784 | /** | |
4785 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4786 | * @pid: pid of the process. | |
4787 | * @interval: userspace pointer to the timeslice value. | |
4788 | * | |
4789 | * this syscall writes the default timeslice value of a given process | |
4790 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4791 | */ | |
17da2bd9 | 4792 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4793 | struct timespec __user *, interval) |
1da177e4 | 4794 | { |
36c8b586 | 4795 | struct task_struct *p; |
a4ec24b4 | 4796 | unsigned int time_slice; |
dba091b9 TG |
4797 | unsigned long flags; |
4798 | struct rq *rq; | |
3a5c359a | 4799 | int retval; |
1da177e4 | 4800 | struct timespec t; |
1da177e4 LT |
4801 | |
4802 | if (pid < 0) | |
3a5c359a | 4803 | return -EINVAL; |
1da177e4 LT |
4804 | |
4805 | retval = -ESRCH; | |
1a551ae7 | 4806 | rcu_read_lock(); |
1da177e4 LT |
4807 | p = find_process_by_pid(pid); |
4808 | if (!p) | |
4809 | goto out_unlock; | |
4810 | ||
4811 | retval = security_task_getscheduler(p); | |
4812 | if (retval) | |
4813 | goto out_unlock; | |
4814 | ||
dba091b9 TG |
4815 | rq = task_rq_lock(p, &flags); |
4816 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4817 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4818 | |
1a551ae7 | 4819 | rcu_read_unlock(); |
a4ec24b4 | 4820 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4821 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4822 | return retval; |
3a5c359a | 4823 | |
1da177e4 | 4824 | out_unlock: |
1a551ae7 | 4825 | rcu_read_unlock(); |
1da177e4 LT |
4826 | return retval; |
4827 | } | |
4828 | ||
7c731e0a | 4829 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4830 | |
82a1fcb9 | 4831 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4832 | { |
1da177e4 | 4833 | unsigned long free = 0; |
36c8b586 | 4834 | unsigned state; |
1da177e4 | 4835 | |
1da177e4 | 4836 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4837 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4838 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4839 | #if BITS_PER_LONG == 32 |
1da177e4 | 4840 | if (state == TASK_RUNNING) |
3df0fc5b | 4841 | printk(KERN_CONT " running "); |
1da177e4 | 4842 | else |
3df0fc5b | 4843 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4844 | #else |
4845 | if (state == TASK_RUNNING) | |
3df0fc5b | 4846 | printk(KERN_CONT " running task "); |
1da177e4 | 4847 | else |
3df0fc5b | 4848 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4849 | #endif |
4850 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4851 | free = stack_not_used(p); |
1da177e4 | 4852 | #endif |
3df0fc5b | 4853 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
07cde260 | 4854 | task_pid_nr(p), task_pid_nr(rcu_dereference(p->real_parent)), |
aa47b7e0 | 4855 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4856 | |
5fb5e6de | 4857 | show_stack(p, NULL); |
1da177e4 LT |
4858 | } |
4859 | ||
e59e2ae2 | 4860 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4861 | { |
36c8b586 | 4862 | struct task_struct *g, *p; |
1da177e4 | 4863 | |
4bd77321 | 4864 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4865 | printk(KERN_INFO |
4866 | " task PC stack pid father\n"); | |
1da177e4 | 4867 | #else |
3df0fc5b PZ |
4868 | printk(KERN_INFO |
4869 | " task PC stack pid father\n"); | |
1da177e4 | 4870 | #endif |
510f5acc | 4871 | rcu_read_lock(); |
1da177e4 LT |
4872 | do_each_thread(g, p) { |
4873 | /* | |
4874 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4875 | * console might take a lot of time: |
1da177e4 LT |
4876 | */ |
4877 | touch_nmi_watchdog(); | |
39bc89fd | 4878 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4879 | sched_show_task(p); |
1da177e4 LT |
4880 | } while_each_thread(g, p); |
4881 | ||
04c9167f JF |
4882 | touch_all_softlockup_watchdogs(); |
4883 | ||
dd41f596 IM |
4884 | #ifdef CONFIG_SCHED_DEBUG |
4885 | sysrq_sched_debug_show(); | |
4886 | #endif | |
510f5acc | 4887 | rcu_read_unlock(); |
e59e2ae2 IM |
4888 | /* |
4889 | * Only show locks if all tasks are dumped: | |
4890 | */ | |
93335a21 | 4891 | if (!state_filter) |
e59e2ae2 | 4892 | debug_show_all_locks(); |
1da177e4 LT |
4893 | } |
4894 | ||
1df21055 IM |
4895 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4896 | { | |
dd41f596 | 4897 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4898 | } |
4899 | ||
f340c0d1 IM |
4900 | /** |
4901 | * init_idle - set up an idle thread for a given CPU | |
4902 | * @idle: task in question | |
4903 | * @cpu: cpu the idle task belongs to | |
4904 | * | |
4905 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4906 | * flag, to make booting more robust. | |
4907 | */ | |
5c1e1767 | 4908 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4909 | { |
70b97a7f | 4910 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4911 | unsigned long flags; |
4912 | ||
05fa785c | 4913 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4914 | |
dd41f596 | 4915 | __sched_fork(idle); |
06b83b5f | 4916 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4917 | idle->se.exec_start = sched_clock(); |
4918 | ||
1e1b6c51 | 4919 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4920 | /* |
4921 | * We're having a chicken and egg problem, even though we are | |
4922 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4923 | * lockdep check in task_group() will fail. | |
4924 | * | |
4925 | * Similar case to sched_fork(). / Alternatively we could | |
4926 | * use task_rq_lock() here and obtain the other rq->lock. | |
4927 | * | |
4928 | * Silence PROVE_RCU | |
4929 | */ | |
4930 | rcu_read_lock(); | |
dd41f596 | 4931 | __set_task_cpu(idle, cpu); |
6506cf6c | 4932 | rcu_read_unlock(); |
1da177e4 | 4933 | |
1da177e4 | 4934 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
4935 | #if defined(CONFIG_SMP) |
4936 | idle->on_cpu = 1; | |
4866cde0 | 4937 | #endif |
05fa785c | 4938 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4939 | |
4940 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 4941 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 4942 | |
dd41f596 IM |
4943 | /* |
4944 | * The idle tasks have their own, simple scheduling class: | |
4945 | */ | |
4946 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4947 | ftrace_graph_init_idle_task(idle, cpu); |
f1c6f1a7 CE |
4948 | #if defined(CONFIG_SMP) |
4949 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4950 | #endif | |
19978ca6 IM |
4951 | } |
4952 | ||
1da177e4 | 4953 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4954 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4955 | { | |
4956 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4957 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4958 | |
4959 | cpumask_copy(&p->cpus_allowed, new_mask); | |
4960 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
1e1b6c51 KM |
4961 | } |
4962 | ||
1da177e4 LT |
4963 | /* |
4964 | * This is how migration works: | |
4965 | * | |
969c7921 TH |
4966 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4967 | * stop_one_cpu(). | |
4968 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4969 | * off the CPU) | |
4970 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4971 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4972 | * it and puts it into the right queue. |
969c7921 TH |
4973 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4974 | * is done. | |
1da177e4 LT |
4975 | */ |
4976 | ||
4977 | /* | |
4978 | * Change a given task's CPU affinity. Migrate the thread to a | |
4979 | * proper CPU and schedule it away if the CPU it's executing on | |
4980 | * is removed from the allowed bitmask. | |
4981 | * | |
4982 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4983 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4984 | * call is not atomic; no spinlocks may be held. |
4985 | */ | |
96f874e2 | 4986 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4987 | { |
4988 | unsigned long flags; | |
70b97a7f | 4989 | struct rq *rq; |
969c7921 | 4990 | unsigned int dest_cpu; |
48f24c4d | 4991 | int ret = 0; |
1da177e4 LT |
4992 | |
4993 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4994 | |
db44fc01 YZ |
4995 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4996 | goto out; | |
4997 | ||
6ad4c188 | 4998 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4999 | ret = -EINVAL; |
5000 | goto out; | |
5001 | } | |
5002 | ||
db44fc01 | 5003 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
5004 | ret = -EINVAL; |
5005 | goto out; | |
5006 | } | |
5007 | ||
1e1b6c51 | 5008 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 5009 | |
1da177e4 | 5010 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5011 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5012 | goto out; |
5013 | ||
969c7921 | 5014 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 5015 | if (p->on_rq) { |
969c7921 | 5016 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 5017 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 5018 | task_rq_unlock(rq, p, &flags); |
969c7921 | 5019 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5020 | tlb_migrate_finish(p->mm); |
5021 | return 0; | |
5022 | } | |
5023 | out: | |
0122ec5b | 5024 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 5025 | |
1da177e4 LT |
5026 | return ret; |
5027 | } | |
cd8ba7cd | 5028 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5029 | |
5030 | /* | |
41a2d6cf | 5031 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5032 | * this because either it can't run here any more (set_cpus_allowed() |
5033 | * away from this CPU, or CPU going down), or because we're | |
5034 | * attempting to rebalance this task on exec (sched_exec). | |
5035 | * | |
5036 | * So we race with normal scheduler movements, but that's OK, as long | |
5037 | * as the task is no longer on this CPU. | |
efc30814 KK |
5038 | * |
5039 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5040 | */ |
efc30814 | 5041 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5042 | { |
70b97a7f | 5043 | struct rq *rq_dest, *rq_src; |
e2912009 | 5044 | int ret = 0; |
1da177e4 | 5045 | |
e761b772 | 5046 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5047 | return ret; |
1da177e4 LT |
5048 | |
5049 | rq_src = cpu_rq(src_cpu); | |
5050 | rq_dest = cpu_rq(dest_cpu); | |
5051 | ||
0122ec5b | 5052 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
5053 | double_rq_lock(rq_src, rq_dest); |
5054 | /* Already moved. */ | |
5055 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5056 | goto done; |
1da177e4 | 5057 | /* Affinity changed (again). */ |
fa17b507 | 5058 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 5059 | goto fail; |
1da177e4 | 5060 | |
e2912009 PZ |
5061 | /* |
5062 | * If we're not on a rq, the next wake-up will ensure we're | |
5063 | * placed properly. | |
5064 | */ | |
fd2f4419 | 5065 | if (p->on_rq) { |
4ca9b72b | 5066 | dequeue_task(rq_src, p, 0); |
e2912009 | 5067 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 5068 | enqueue_task(rq_dest, p, 0); |
15afe09b | 5069 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5070 | } |
b1e38734 | 5071 | done: |
efc30814 | 5072 | ret = 1; |
b1e38734 | 5073 | fail: |
1da177e4 | 5074 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 5075 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 5076 | return ret; |
1da177e4 LT |
5077 | } |
5078 | ||
5079 | /* | |
969c7921 TH |
5080 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5081 | * and performs thread migration by bumping thread off CPU then | |
5082 | * 'pushing' onto another runqueue. | |
1da177e4 | 5083 | */ |
969c7921 | 5084 | static int migration_cpu_stop(void *data) |
1da177e4 | 5085 | { |
969c7921 | 5086 | struct migration_arg *arg = data; |
f7b4cddc | 5087 | |
969c7921 TH |
5088 | /* |
5089 | * The original target cpu might have gone down and we might | |
5090 | * be on another cpu but it doesn't matter. | |
5091 | */ | |
f7b4cddc | 5092 | local_irq_disable(); |
969c7921 | 5093 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5094 | local_irq_enable(); |
1da177e4 | 5095 | return 0; |
f7b4cddc ON |
5096 | } |
5097 | ||
1da177e4 | 5098 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5099 | |
054b9108 | 5100 | /* |
48c5ccae PZ |
5101 | * Ensures that the idle task is using init_mm right before its cpu goes |
5102 | * offline. | |
054b9108 | 5103 | */ |
48c5ccae | 5104 | void idle_task_exit(void) |
1da177e4 | 5105 | { |
48c5ccae | 5106 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5107 | |
48c5ccae | 5108 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5109 | |
48c5ccae PZ |
5110 | if (mm != &init_mm) |
5111 | switch_mm(mm, &init_mm, current); | |
5112 | mmdrop(mm); | |
1da177e4 LT |
5113 | } |
5114 | ||
5115 | /* | |
5116 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5117 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5118 | * for performance reasons the counter is not stricly tracking tasks to | |
5119 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5120 | * to keep the global sum constant after CPU-down: | |
5121 | */ | |
70b97a7f | 5122 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5123 | { |
6ad4c188 | 5124 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5125 | |
1da177e4 LT |
5126 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5127 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5128 | } |
5129 | ||
dd41f596 | 5130 | /* |
48c5ccae | 5131 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5132 | */ |
48c5ccae | 5133 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5134 | { |
48c5ccae PZ |
5135 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5136 | rq->calc_load_active = 0; | |
1da177e4 LT |
5137 | } |
5138 | ||
48f24c4d | 5139 | /* |
48c5ccae PZ |
5140 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5141 | * try_to_wake_up()->select_task_rq(). | |
5142 | * | |
5143 | * Called with rq->lock held even though we'er in stop_machine() and | |
5144 | * there's no concurrency possible, we hold the required locks anyway | |
5145 | * because of lock validation efforts. | |
1da177e4 | 5146 | */ |
48c5ccae | 5147 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5148 | { |
70b97a7f | 5149 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5150 | struct task_struct *next, *stop = rq->stop; |
5151 | int dest_cpu; | |
1da177e4 LT |
5152 | |
5153 | /* | |
48c5ccae PZ |
5154 | * Fudge the rq selection such that the below task selection loop |
5155 | * doesn't get stuck on the currently eligible stop task. | |
5156 | * | |
5157 | * We're currently inside stop_machine() and the rq is either stuck | |
5158 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5159 | * either way we should never end up calling schedule() until we're | |
5160 | * done here. | |
1da177e4 | 5161 | */ |
48c5ccae | 5162 | rq->stop = NULL; |
48f24c4d | 5163 | |
8cb120d3 PT |
5164 | /* Ensure any throttled groups are reachable by pick_next_task */ |
5165 | unthrottle_offline_cfs_rqs(rq); | |
5166 | ||
dd41f596 | 5167 | for ( ; ; ) { |
48c5ccae PZ |
5168 | /* |
5169 | * There's this thread running, bail when that's the only | |
5170 | * remaining thread. | |
5171 | */ | |
5172 | if (rq->nr_running == 1) | |
dd41f596 | 5173 | break; |
48c5ccae | 5174 | |
b67802ea | 5175 | next = pick_next_task(rq); |
48c5ccae | 5176 | BUG_ON(!next); |
79c53799 | 5177 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5178 | |
48c5ccae PZ |
5179 | /* Find suitable destination for @next, with force if needed. */ |
5180 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5181 | raw_spin_unlock(&rq->lock); | |
5182 | ||
5183 | __migrate_task(next, dead_cpu, dest_cpu); | |
5184 | ||
5185 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5186 | } |
dce48a84 | 5187 | |
48c5ccae | 5188 | rq->stop = stop; |
dce48a84 | 5189 | } |
48c5ccae | 5190 | |
1da177e4 LT |
5191 | #endif /* CONFIG_HOTPLUG_CPU */ |
5192 | ||
e692ab53 NP |
5193 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5194 | ||
5195 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5196 | { |
5197 | .procname = "sched_domain", | |
c57baf1e | 5198 | .mode = 0555, |
e0361851 | 5199 | }, |
56992309 | 5200 | {} |
e692ab53 NP |
5201 | }; |
5202 | ||
5203 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5204 | { |
5205 | .procname = "kernel", | |
c57baf1e | 5206 | .mode = 0555, |
e0361851 AD |
5207 | .child = sd_ctl_dir, |
5208 | }, | |
56992309 | 5209 | {} |
e692ab53 NP |
5210 | }; |
5211 | ||
5212 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5213 | { | |
5214 | struct ctl_table *entry = | |
5cf9f062 | 5215 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5216 | |
e692ab53 NP |
5217 | return entry; |
5218 | } | |
5219 | ||
6382bc90 MM |
5220 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5221 | { | |
cd790076 | 5222 | struct ctl_table *entry; |
6382bc90 | 5223 | |
cd790076 MM |
5224 | /* |
5225 | * In the intermediate directories, both the child directory and | |
5226 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5227 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5228 | * static strings and all have proc handlers. |
5229 | */ | |
5230 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5231 | if (entry->child) |
5232 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5233 | if (entry->proc_handler == NULL) |
5234 | kfree(entry->procname); | |
5235 | } | |
6382bc90 MM |
5236 | |
5237 | kfree(*tablep); | |
5238 | *tablep = NULL; | |
5239 | } | |
5240 | ||
e692ab53 | 5241 | static void |
e0361851 | 5242 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5243 | const char *procname, void *data, int maxlen, |
36fcb589 | 5244 | umode_t mode, proc_handler *proc_handler) |
e692ab53 | 5245 | { |
e692ab53 NP |
5246 | entry->procname = procname; |
5247 | entry->data = data; | |
5248 | entry->maxlen = maxlen; | |
5249 | entry->mode = mode; | |
5250 | entry->proc_handler = proc_handler; | |
5251 | } | |
5252 | ||
5253 | static struct ctl_table * | |
5254 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5255 | { | |
a5d8c348 | 5256 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5257 | |
ad1cdc1d MM |
5258 | if (table == NULL) |
5259 | return NULL; | |
5260 | ||
e0361851 | 5261 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5262 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5263 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5264 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5265 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5266 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5267 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5268 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5269 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5270 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5271 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5272 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5273 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5274 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5275 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5276 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5277 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5278 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5279 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5280 | &sd->cache_nice_tries, |
5281 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5282 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5283 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5284 | set_table_entry(&table[11], "name", sd->name, |
5285 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5286 | /* &table[12] is terminator */ | |
e692ab53 NP |
5287 | |
5288 | return table; | |
5289 | } | |
5290 | ||
9a4e7159 | 5291 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5292 | { |
5293 | struct ctl_table *entry, *table; | |
5294 | struct sched_domain *sd; | |
5295 | int domain_num = 0, i; | |
5296 | char buf[32]; | |
5297 | ||
5298 | for_each_domain(cpu, sd) | |
5299 | domain_num++; | |
5300 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5301 | if (table == NULL) |
5302 | return NULL; | |
e692ab53 NP |
5303 | |
5304 | i = 0; | |
5305 | for_each_domain(cpu, sd) { | |
5306 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5307 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5308 | entry->mode = 0555; |
e692ab53 NP |
5309 | entry->child = sd_alloc_ctl_domain_table(sd); |
5310 | entry++; | |
5311 | i++; | |
5312 | } | |
5313 | return table; | |
5314 | } | |
5315 | ||
5316 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5317 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5318 | { |
6ad4c188 | 5319 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5320 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5321 | char buf[32]; | |
5322 | ||
7378547f MM |
5323 | WARN_ON(sd_ctl_dir[0].child); |
5324 | sd_ctl_dir[0].child = entry; | |
5325 | ||
ad1cdc1d MM |
5326 | if (entry == NULL) |
5327 | return; | |
5328 | ||
6ad4c188 | 5329 | for_each_possible_cpu(i) { |
e692ab53 | 5330 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5331 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5332 | entry->mode = 0555; |
e692ab53 | 5333 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5334 | entry++; |
e692ab53 | 5335 | } |
7378547f MM |
5336 | |
5337 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5338 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5339 | } | |
6382bc90 | 5340 | |
7378547f | 5341 | /* may be called multiple times per register */ |
6382bc90 MM |
5342 | static void unregister_sched_domain_sysctl(void) |
5343 | { | |
7378547f MM |
5344 | if (sd_sysctl_header) |
5345 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5346 | sd_sysctl_header = NULL; |
7378547f MM |
5347 | if (sd_ctl_dir[0].child) |
5348 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5349 | } |
e692ab53 | 5350 | #else |
6382bc90 MM |
5351 | static void register_sched_domain_sysctl(void) |
5352 | { | |
5353 | } | |
5354 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5355 | { |
5356 | } | |
5357 | #endif | |
5358 | ||
1f11eb6a GH |
5359 | static void set_rq_online(struct rq *rq) |
5360 | { | |
5361 | if (!rq->online) { | |
5362 | const struct sched_class *class; | |
5363 | ||
c6c4927b | 5364 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5365 | rq->online = 1; |
5366 | ||
5367 | for_each_class(class) { | |
5368 | if (class->rq_online) | |
5369 | class->rq_online(rq); | |
5370 | } | |
5371 | } | |
5372 | } | |
5373 | ||
5374 | static void set_rq_offline(struct rq *rq) | |
5375 | { | |
5376 | if (rq->online) { | |
5377 | const struct sched_class *class; | |
5378 | ||
5379 | for_each_class(class) { | |
5380 | if (class->rq_offline) | |
5381 | class->rq_offline(rq); | |
5382 | } | |
5383 | ||
c6c4927b | 5384 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5385 | rq->online = 0; |
5386 | } | |
5387 | } | |
5388 | ||
1da177e4 LT |
5389 | /* |
5390 | * migration_call - callback that gets triggered when a CPU is added. | |
5391 | * Here we can start up the necessary migration thread for the new CPU. | |
5392 | */ | |
48f24c4d IM |
5393 | static int __cpuinit |
5394 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5395 | { |
48f24c4d | 5396 | int cpu = (long)hcpu; |
1da177e4 | 5397 | unsigned long flags; |
969c7921 | 5398 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5399 | |
48c5ccae | 5400 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5401 | |
1da177e4 | 5402 | case CPU_UP_PREPARE: |
a468d389 | 5403 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5404 | break; |
48f24c4d | 5405 | |
1da177e4 | 5406 | case CPU_ONLINE: |
1f94ef59 | 5407 | /* Update our root-domain */ |
05fa785c | 5408 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5409 | if (rq->rd) { |
c6c4927b | 5410 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5411 | |
5412 | set_rq_online(rq); | |
1f94ef59 | 5413 | } |
05fa785c | 5414 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5415 | break; |
48f24c4d | 5416 | |
1da177e4 | 5417 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5418 | case CPU_DYING: |
317f3941 | 5419 | sched_ttwu_pending(); |
57d885fe | 5420 | /* Update our root-domain */ |
05fa785c | 5421 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5422 | if (rq->rd) { |
c6c4927b | 5423 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5424 | set_rq_offline(rq); |
57d885fe | 5425 | } |
48c5ccae PZ |
5426 | migrate_tasks(cpu); |
5427 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5428 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5429 | |
5430 | migrate_nr_uninterruptible(rq); | |
5431 | calc_global_load_remove(rq); | |
57d885fe | 5432 | break; |
1da177e4 LT |
5433 | #endif |
5434 | } | |
49c022e6 PZ |
5435 | |
5436 | update_max_interval(); | |
5437 | ||
1da177e4 LT |
5438 | return NOTIFY_OK; |
5439 | } | |
5440 | ||
f38b0820 PM |
5441 | /* |
5442 | * Register at high priority so that task migration (migrate_all_tasks) | |
5443 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5444 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5445 | */ |
26c2143b | 5446 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5447 | .notifier_call = migration_call, |
50a323b7 | 5448 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5449 | }; |
5450 | ||
3a101d05 TH |
5451 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5452 | unsigned long action, void *hcpu) | |
5453 | { | |
5454 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5455 | case CPU_STARTING: |
3a101d05 TH |
5456 | case CPU_DOWN_FAILED: |
5457 | set_cpu_active((long)hcpu, true); | |
5458 | return NOTIFY_OK; | |
5459 | default: | |
5460 | return NOTIFY_DONE; | |
5461 | } | |
5462 | } | |
5463 | ||
5464 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5465 | unsigned long action, void *hcpu) | |
5466 | { | |
5467 | switch (action & ~CPU_TASKS_FROZEN) { | |
5468 | case CPU_DOWN_PREPARE: | |
5469 | set_cpu_active((long)hcpu, false); | |
5470 | return NOTIFY_OK; | |
5471 | default: | |
5472 | return NOTIFY_DONE; | |
5473 | } | |
5474 | } | |
5475 | ||
7babe8db | 5476 | static int __init migration_init(void) |
1da177e4 LT |
5477 | { |
5478 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5479 | int err; |
48f24c4d | 5480 | |
3a101d05 | 5481 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5482 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5483 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5484 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5485 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5486 | |
3a101d05 TH |
5487 | /* Register cpu active notifiers */ |
5488 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5489 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5490 | ||
a004cd42 | 5491 | return 0; |
1da177e4 | 5492 | } |
7babe8db | 5493 | early_initcall(migration_init); |
1da177e4 LT |
5494 | #endif |
5495 | ||
5496 | #ifdef CONFIG_SMP | |
476f3534 | 5497 | |
4cb98839 PZ |
5498 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5499 | ||
3e9830dc | 5500 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5501 | |
f6630114 MT |
5502 | static __read_mostly int sched_domain_debug_enabled; |
5503 | ||
5504 | static int __init sched_domain_debug_setup(char *str) | |
5505 | { | |
5506 | sched_domain_debug_enabled = 1; | |
5507 | ||
5508 | return 0; | |
5509 | } | |
5510 | early_param("sched_debug", sched_domain_debug_setup); | |
5511 | ||
7c16ec58 | 5512 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5513 | struct cpumask *groupmask) |
1da177e4 | 5514 | { |
4dcf6aff | 5515 | struct sched_group *group = sd->groups; |
434d53b0 | 5516 | char str[256]; |
1da177e4 | 5517 | |
968ea6d8 | 5518 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5519 | cpumask_clear(groupmask); |
4dcf6aff IM |
5520 | |
5521 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5522 | ||
5523 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5524 | printk("does not load-balance\n"); |
4dcf6aff | 5525 | if (sd->parent) |
3df0fc5b PZ |
5526 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5527 | " has parent"); | |
4dcf6aff | 5528 | return -1; |
41c7ce9a NP |
5529 | } |
5530 | ||
3df0fc5b | 5531 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5532 | |
758b2cdc | 5533 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5534 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5535 | "CPU%d\n", cpu); | |
4dcf6aff | 5536 | } |
758b2cdc | 5537 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5538 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5539 | " CPU%d\n", cpu); | |
4dcf6aff | 5540 | } |
1da177e4 | 5541 | |
4dcf6aff | 5542 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5543 | do { |
4dcf6aff | 5544 | if (!group) { |
3df0fc5b PZ |
5545 | printk("\n"); |
5546 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5547 | break; |
5548 | } | |
5549 | ||
9c3f75cb | 5550 | if (!group->sgp->power) { |
3df0fc5b PZ |
5551 | printk(KERN_CONT "\n"); |
5552 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5553 | "set\n"); | |
4dcf6aff IM |
5554 | break; |
5555 | } | |
1da177e4 | 5556 | |
758b2cdc | 5557 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5558 | printk(KERN_CONT "\n"); |
5559 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5560 | break; |
5561 | } | |
1da177e4 | 5562 | |
cb83b629 PZ |
5563 | if (!(sd->flags & SD_OVERLAP) && |
5564 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5565 | printk(KERN_CONT "\n"); |
5566 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5567 | break; |
5568 | } | |
1da177e4 | 5569 | |
758b2cdc | 5570 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5571 | |
968ea6d8 | 5572 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5573 | |
3df0fc5b | 5574 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5575 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5576 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5577 | group->sgp->power); |
381512cf | 5578 | } |
1da177e4 | 5579 | |
4dcf6aff IM |
5580 | group = group->next; |
5581 | } while (group != sd->groups); | |
3df0fc5b | 5582 | printk(KERN_CONT "\n"); |
1da177e4 | 5583 | |
758b2cdc | 5584 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5585 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5586 | |
758b2cdc RR |
5587 | if (sd->parent && |
5588 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5589 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5590 | "of domain->span\n"); | |
4dcf6aff IM |
5591 | return 0; |
5592 | } | |
1da177e4 | 5593 | |
4dcf6aff IM |
5594 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5595 | { | |
5596 | int level = 0; | |
1da177e4 | 5597 | |
f6630114 MT |
5598 | if (!sched_domain_debug_enabled) |
5599 | return; | |
5600 | ||
4dcf6aff IM |
5601 | if (!sd) { |
5602 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5603 | return; | |
5604 | } | |
1da177e4 | 5605 | |
4dcf6aff IM |
5606 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5607 | ||
5608 | for (;;) { | |
4cb98839 | 5609 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5610 | break; |
1da177e4 LT |
5611 | level++; |
5612 | sd = sd->parent; | |
33859f7f | 5613 | if (!sd) |
4dcf6aff IM |
5614 | break; |
5615 | } | |
1da177e4 | 5616 | } |
6d6bc0ad | 5617 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5618 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5619 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5620 | |
1a20ff27 | 5621 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5622 | { |
758b2cdc | 5623 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5624 | return 1; |
5625 | ||
5626 | /* Following flags need at least 2 groups */ | |
5627 | if (sd->flags & (SD_LOAD_BALANCE | | |
5628 | SD_BALANCE_NEWIDLE | | |
5629 | SD_BALANCE_FORK | | |
89c4710e SS |
5630 | SD_BALANCE_EXEC | |
5631 | SD_SHARE_CPUPOWER | | |
5632 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5633 | if (sd->groups != sd->groups->next) |
5634 | return 0; | |
5635 | } | |
5636 | ||
5637 | /* Following flags don't use groups */ | |
c88d5910 | 5638 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5639 | return 0; |
5640 | ||
5641 | return 1; | |
5642 | } | |
5643 | ||
48f24c4d IM |
5644 | static int |
5645 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5646 | { |
5647 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5648 | ||
5649 | if (sd_degenerate(parent)) | |
5650 | return 1; | |
5651 | ||
758b2cdc | 5652 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5653 | return 0; |
5654 | ||
245af2c7 SS |
5655 | /* Flags needing groups don't count if only 1 group in parent */ |
5656 | if (parent->groups == parent->groups->next) { | |
5657 | pflags &= ~(SD_LOAD_BALANCE | | |
5658 | SD_BALANCE_NEWIDLE | | |
5659 | SD_BALANCE_FORK | | |
89c4710e SS |
5660 | SD_BALANCE_EXEC | |
5661 | SD_SHARE_CPUPOWER | | |
5662 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5663 | if (nr_node_ids == 1) |
5664 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5665 | } |
5666 | if (~cflags & pflags) | |
5667 | return 0; | |
5668 | ||
5669 | return 1; | |
5670 | } | |
5671 | ||
dce840a0 | 5672 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5673 | { |
dce840a0 | 5674 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5675 | |
68e74568 | 5676 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5677 | free_cpumask_var(rd->rto_mask); |
5678 | free_cpumask_var(rd->online); | |
5679 | free_cpumask_var(rd->span); | |
5680 | kfree(rd); | |
5681 | } | |
5682 | ||
57d885fe GH |
5683 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5684 | { | |
a0490fa3 | 5685 | struct root_domain *old_rd = NULL; |
57d885fe | 5686 | unsigned long flags; |
57d885fe | 5687 | |
05fa785c | 5688 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5689 | |
5690 | if (rq->rd) { | |
a0490fa3 | 5691 | old_rd = rq->rd; |
57d885fe | 5692 | |
c6c4927b | 5693 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5694 | set_rq_offline(rq); |
57d885fe | 5695 | |
c6c4927b | 5696 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5697 | |
a0490fa3 IM |
5698 | /* |
5699 | * If we dont want to free the old_rt yet then | |
5700 | * set old_rd to NULL to skip the freeing later | |
5701 | * in this function: | |
5702 | */ | |
5703 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5704 | old_rd = NULL; | |
57d885fe GH |
5705 | } |
5706 | ||
5707 | atomic_inc(&rd->refcount); | |
5708 | rq->rd = rd; | |
5709 | ||
c6c4927b | 5710 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5711 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5712 | set_rq_online(rq); |
57d885fe | 5713 | |
05fa785c | 5714 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5715 | |
5716 | if (old_rd) | |
dce840a0 | 5717 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5718 | } |
5719 | ||
68c38fc3 | 5720 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5721 | { |
5722 | memset(rd, 0, sizeof(*rd)); | |
5723 | ||
68c38fc3 | 5724 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5725 | goto out; |
68c38fc3 | 5726 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5727 | goto free_span; |
68c38fc3 | 5728 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5729 | goto free_online; |
6e0534f2 | 5730 | |
68c38fc3 | 5731 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5732 | goto free_rto_mask; |
c6c4927b | 5733 | return 0; |
6e0534f2 | 5734 | |
68e74568 RR |
5735 | free_rto_mask: |
5736 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5737 | free_online: |
5738 | free_cpumask_var(rd->online); | |
5739 | free_span: | |
5740 | free_cpumask_var(rd->span); | |
0c910d28 | 5741 | out: |
c6c4927b | 5742 | return -ENOMEM; |
57d885fe GH |
5743 | } |
5744 | ||
029632fb PZ |
5745 | /* |
5746 | * By default the system creates a single root-domain with all cpus as | |
5747 | * members (mimicking the global state we have today). | |
5748 | */ | |
5749 | struct root_domain def_root_domain; | |
5750 | ||
57d885fe GH |
5751 | static void init_defrootdomain(void) |
5752 | { | |
68c38fc3 | 5753 | init_rootdomain(&def_root_domain); |
c6c4927b | 5754 | |
57d885fe GH |
5755 | atomic_set(&def_root_domain.refcount, 1); |
5756 | } | |
5757 | ||
dc938520 | 5758 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5759 | { |
5760 | struct root_domain *rd; | |
5761 | ||
5762 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5763 | if (!rd) | |
5764 | return NULL; | |
5765 | ||
68c38fc3 | 5766 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5767 | kfree(rd); |
5768 | return NULL; | |
5769 | } | |
57d885fe GH |
5770 | |
5771 | return rd; | |
5772 | } | |
5773 | ||
e3589f6c PZ |
5774 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5775 | { | |
5776 | struct sched_group *tmp, *first; | |
5777 | ||
5778 | if (!sg) | |
5779 | return; | |
5780 | ||
5781 | first = sg; | |
5782 | do { | |
5783 | tmp = sg->next; | |
5784 | ||
5785 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5786 | kfree(sg->sgp); | |
5787 | ||
5788 | kfree(sg); | |
5789 | sg = tmp; | |
5790 | } while (sg != first); | |
5791 | } | |
5792 | ||
dce840a0 PZ |
5793 | static void free_sched_domain(struct rcu_head *rcu) |
5794 | { | |
5795 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5796 | |
5797 | /* | |
5798 | * If its an overlapping domain it has private groups, iterate and | |
5799 | * nuke them all. | |
5800 | */ | |
5801 | if (sd->flags & SD_OVERLAP) { | |
5802 | free_sched_groups(sd->groups, 1); | |
5803 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5804 | kfree(sd->groups->sgp); |
dce840a0 | 5805 | kfree(sd->groups); |
9c3f75cb | 5806 | } |
dce840a0 PZ |
5807 | kfree(sd); |
5808 | } | |
5809 | ||
5810 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5811 | { | |
5812 | call_rcu(&sd->rcu, free_sched_domain); | |
5813 | } | |
5814 | ||
5815 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5816 | { | |
5817 | for (; sd; sd = sd->parent) | |
5818 | destroy_sched_domain(sd, cpu); | |
5819 | } | |
5820 | ||
518cd623 PZ |
5821 | /* |
5822 | * Keep a special pointer to the highest sched_domain that has | |
5823 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5824 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5825 | * | |
5826 | * Also keep a unique ID per domain (we use the first cpu number in | |
5827 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5828 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5829 | */ |
5830 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
5831 | DEFINE_PER_CPU(int, sd_llc_id); | |
5832 | ||
5833 | static void update_top_cache_domain(int cpu) | |
5834 | { | |
5835 | struct sched_domain *sd; | |
5836 | int id = cpu; | |
5837 | ||
5838 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
5839 | if (sd) | |
5840 | id = cpumask_first(sched_domain_span(sd)); | |
5841 | ||
5842 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
5843 | per_cpu(sd_llc_id, cpu) = id; | |
5844 | } | |
5845 | ||
1da177e4 | 5846 | /* |
0eab9146 | 5847 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5848 | * hold the hotplug lock. |
5849 | */ | |
0eab9146 IM |
5850 | static void |
5851 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5852 | { |
70b97a7f | 5853 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5854 | struct sched_domain *tmp; |
5855 | ||
5856 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5857 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5858 | struct sched_domain *parent = tmp->parent; |
5859 | if (!parent) | |
5860 | break; | |
f29c9b1c | 5861 | |
1a848870 | 5862 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5863 | tmp->parent = parent->parent; |
1a848870 SS |
5864 | if (parent->parent) |
5865 | parent->parent->child = tmp; | |
dce840a0 | 5866 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5867 | } else |
5868 | tmp = tmp->parent; | |
245af2c7 SS |
5869 | } |
5870 | ||
1a848870 | 5871 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5872 | tmp = sd; |
245af2c7 | 5873 | sd = sd->parent; |
dce840a0 | 5874 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5875 | if (sd) |
5876 | sd->child = NULL; | |
5877 | } | |
1da177e4 | 5878 | |
4cb98839 | 5879 | sched_domain_debug(sd, cpu); |
1da177e4 | 5880 | |
57d885fe | 5881 | rq_attach_root(rq, rd); |
dce840a0 | 5882 | tmp = rq->sd; |
674311d5 | 5883 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5884 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5885 | |
5886 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5887 | } |
5888 | ||
5889 | /* cpus with isolated domains */ | |
dcc30a35 | 5890 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5891 | |
5892 | /* Setup the mask of cpus configured for isolated domains */ | |
5893 | static int __init isolated_cpu_setup(char *str) | |
5894 | { | |
bdddd296 | 5895 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5896 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5897 | return 1; |
5898 | } | |
5899 | ||
8927f494 | 5900 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5901 | |
d3081f52 PZ |
5902 | static const struct cpumask *cpu_cpu_mask(int cpu) |
5903 | { | |
5904 | return cpumask_of_node(cpu_to_node(cpu)); | |
5905 | } | |
5906 | ||
5c45bf27 | 5907 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5908 | |
dce840a0 PZ |
5909 | struct sd_data { |
5910 | struct sched_domain **__percpu sd; | |
5911 | struct sched_group **__percpu sg; | |
9c3f75cb | 5912 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
5913 | }; |
5914 | ||
49a02c51 | 5915 | struct s_data { |
21d42ccf | 5916 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5917 | struct root_domain *rd; |
5918 | }; | |
5919 | ||
2109b99e | 5920 | enum s_alloc { |
2109b99e | 5921 | sa_rootdomain, |
21d42ccf | 5922 | sa_sd, |
dce840a0 | 5923 | sa_sd_storage, |
2109b99e AH |
5924 | sa_none, |
5925 | }; | |
5926 | ||
54ab4ff4 PZ |
5927 | struct sched_domain_topology_level; |
5928 | ||
5929 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
5930 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
5931 | ||
e3589f6c PZ |
5932 | #define SDTL_OVERLAP 0x01 |
5933 | ||
eb7a74e6 | 5934 | struct sched_domain_topology_level { |
2c402dc3 PZ |
5935 | sched_domain_init_f init; |
5936 | sched_domain_mask_f mask; | |
e3589f6c | 5937 | int flags; |
cb83b629 | 5938 | int numa_level; |
54ab4ff4 | 5939 | struct sd_data data; |
eb7a74e6 PZ |
5940 | }; |
5941 | ||
e3589f6c PZ |
5942 | static int |
5943 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5944 | { | |
5945 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5946 | const struct cpumask *span = sched_domain_span(sd); | |
5947 | struct cpumask *covered = sched_domains_tmpmask; | |
5948 | struct sd_data *sdd = sd->private; | |
5949 | struct sched_domain *child; | |
5950 | int i; | |
5951 | ||
5952 | cpumask_clear(covered); | |
5953 | ||
5954 | for_each_cpu(i, span) { | |
5955 | struct cpumask *sg_span; | |
5956 | ||
5957 | if (cpumask_test_cpu(i, covered)) | |
5958 | continue; | |
5959 | ||
5960 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
4d78a223 | 5961 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5962 | |
5963 | if (!sg) | |
5964 | goto fail; | |
5965 | ||
5966 | sg_span = sched_group_cpus(sg); | |
5967 | ||
5968 | child = *per_cpu_ptr(sdd->sd, i); | |
5969 | if (child->child) { | |
5970 | child = child->child; | |
5971 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5972 | } else | |
5973 | cpumask_set_cpu(i, sg_span); | |
5974 | ||
5975 | cpumask_or(covered, covered, sg_span); | |
5976 | ||
5977 | sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span)); | |
5978 | atomic_inc(&sg->sgp->ref); | |
5979 | ||
5980 | if (cpumask_test_cpu(cpu, sg_span)) | |
5981 | groups = sg; | |
5982 | ||
5983 | if (!first) | |
5984 | first = sg; | |
5985 | if (last) | |
5986 | last->next = sg; | |
5987 | last = sg; | |
5988 | last->next = first; | |
5989 | } | |
5990 | sd->groups = groups; | |
5991 | ||
5992 | return 0; | |
5993 | ||
5994 | fail: | |
5995 | free_sched_groups(first, 0); | |
5996 | ||
5997 | return -ENOMEM; | |
5998 | } | |
5999 | ||
dce840a0 | 6000 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6001 | { |
dce840a0 PZ |
6002 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6003 | struct sched_domain *child = sd->child; | |
1da177e4 | 6004 | |
dce840a0 PZ |
6005 | if (child) |
6006 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6007 | |
9c3f75cb | 6008 | if (sg) { |
dce840a0 | 6009 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 6010 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 6011 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 6012 | } |
dce840a0 PZ |
6013 | |
6014 | return cpu; | |
1e9f28fa | 6015 | } |
1e9f28fa | 6016 | |
01a08546 | 6017 | /* |
dce840a0 PZ |
6018 | * build_sched_groups will build a circular linked list of the groups |
6019 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6020 | * and ->cpu_power to 0. | |
e3589f6c PZ |
6021 | * |
6022 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6023 | */ |
e3589f6c PZ |
6024 | static int |
6025 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6026 | { |
dce840a0 PZ |
6027 | struct sched_group *first = NULL, *last = NULL; |
6028 | struct sd_data *sdd = sd->private; | |
6029 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6030 | struct cpumask *covered; |
dce840a0 | 6031 | int i; |
9c1cfda2 | 6032 | |
e3589f6c PZ |
6033 | get_group(cpu, sdd, &sd->groups); |
6034 | atomic_inc(&sd->groups->ref); | |
6035 | ||
6036 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
6037 | return 0; | |
6038 | ||
f96225fd PZ |
6039 | lockdep_assert_held(&sched_domains_mutex); |
6040 | covered = sched_domains_tmpmask; | |
6041 | ||
dce840a0 | 6042 | cpumask_clear(covered); |
6711cab4 | 6043 | |
dce840a0 PZ |
6044 | for_each_cpu(i, span) { |
6045 | struct sched_group *sg; | |
6046 | int group = get_group(i, sdd, &sg); | |
6047 | int j; | |
6711cab4 | 6048 | |
dce840a0 PZ |
6049 | if (cpumask_test_cpu(i, covered)) |
6050 | continue; | |
6711cab4 | 6051 | |
dce840a0 | 6052 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 6053 | sg->sgp->power = 0; |
0601a88d | 6054 | |
dce840a0 PZ |
6055 | for_each_cpu(j, span) { |
6056 | if (get_group(j, sdd, NULL) != group) | |
6057 | continue; | |
0601a88d | 6058 | |
dce840a0 PZ |
6059 | cpumask_set_cpu(j, covered); |
6060 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6061 | } | |
0601a88d | 6062 | |
dce840a0 PZ |
6063 | if (!first) |
6064 | first = sg; | |
6065 | if (last) | |
6066 | last->next = sg; | |
6067 | last = sg; | |
6068 | } | |
6069 | last->next = first; | |
e3589f6c PZ |
6070 | |
6071 | return 0; | |
0601a88d | 6072 | } |
51888ca2 | 6073 | |
89c4710e SS |
6074 | /* |
6075 | * Initialize sched groups cpu_power. | |
6076 | * | |
6077 | * cpu_power indicates the capacity of sched group, which is used while | |
6078 | * distributing the load between different sched groups in a sched domain. | |
6079 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6080 | * there are asymmetries in the topology. If there are asymmetries, group | |
6081 | * having more cpu_power will pickup more load compared to the group having | |
6082 | * less cpu_power. | |
89c4710e SS |
6083 | */ |
6084 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6085 | { | |
e3589f6c | 6086 | struct sched_group *sg = sd->groups; |
89c4710e | 6087 | |
e3589f6c PZ |
6088 | WARN_ON(!sd || !sg); |
6089 | ||
6090 | do { | |
6091 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6092 | sg = sg->next; | |
6093 | } while (sg != sd->groups); | |
89c4710e | 6094 | |
e3589f6c PZ |
6095 | if (cpu != group_first_cpu(sg)) |
6096 | return; | |
aae6d3dd | 6097 | |
d274cb30 | 6098 | update_group_power(sd, cpu); |
69e1e811 | 6099 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
6100 | } |
6101 | ||
029632fb PZ |
6102 | int __weak arch_sd_sibling_asym_packing(void) |
6103 | { | |
6104 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
6105 | } |
6106 | ||
7c16ec58 MT |
6107 | /* |
6108 | * Initializers for schedule domains | |
6109 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6110 | */ | |
6111 | ||
a5d8c348 IM |
6112 | #ifdef CONFIG_SCHED_DEBUG |
6113 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6114 | #else | |
6115 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6116 | #endif | |
6117 | ||
54ab4ff4 PZ |
6118 | #define SD_INIT_FUNC(type) \ |
6119 | static noinline struct sched_domain * \ | |
6120 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
6121 | { \ | |
6122 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
6123 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
6124 | SD_INIT_NAME(sd, type); \ |
6125 | sd->private = &tl->data; \ | |
6126 | return sd; \ | |
7c16ec58 MT |
6127 | } |
6128 | ||
6129 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
6130 | #ifdef CONFIG_SCHED_SMT |
6131 | SD_INIT_FUNC(SIBLING) | |
6132 | #endif | |
6133 | #ifdef CONFIG_SCHED_MC | |
6134 | SD_INIT_FUNC(MC) | |
6135 | #endif | |
01a08546 HC |
6136 | #ifdef CONFIG_SCHED_BOOK |
6137 | SD_INIT_FUNC(BOOK) | |
6138 | #endif | |
7c16ec58 | 6139 | |
1d3504fc | 6140 | static int default_relax_domain_level = -1; |
60495e77 | 6141 | int sched_domain_level_max; |
1d3504fc HS |
6142 | |
6143 | static int __init setup_relax_domain_level(char *str) | |
6144 | { | |
30e0e178 LZ |
6145 | unsigned long val; |
6146 | ||
6147 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 6148 | if (val < sched_domain_level_max) |
30e0e178 LZ |
6149 | default_relax_domain_level = val; |
6150 | ||
1d3504fc HS |
6151 | return 1; |
6152 | } | |
6153 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6154 | ||
6155 | static void set_domain_attribute(struct sched_domain *sd, | |
6156 | struct sched_domain_attr *attr) | |
6157 | { | |
6158 | int request; | |
6159 | ||
6160 | if (!attr || attr->relax_domain_level < 0) { | |
6161 | if (default_relax_domain_level < 0) | |
6162 | return; | |
6163 | else | |
6164 | request = default_relax_domain_level; | |
6165 | } else | |
6166 | request = attr->relax_domain_level; | |
6167 | if (request < sd->level) { | |
6168 | /* turn off idle balance on this domain */ | |
c88d5910 | 6169 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6170 | } else { |
6171 | /* turn on idle balance on this domain */ | |
c88d5910 | 6172 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6173 | } |
6174 | } | |
6175 | ||
54ab4ff4 PZ |
6176 | static void __sdt_free(const struct cpumask *cpu_map); |
6177 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6178 | ||
2109b99e AH |
6179 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6180 | const struct cpumask *cpu_map) | |
6181 | { | |
6182 | switch (what) { | |
2109b99e | 6183 | case sa_rootdomain: |
822ff793 PZ |
6184 | if (!atomic_read(&d->rd->refcount)) |
6185 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6186 | case sa_sd: |
6187 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6188 | case sa_sd_storage: |
54ab4ff4 | 6189 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6190 | case sa_none: |
6191 | break; | |
6192 | } | |
6193 | } | |
3404c8d9 | 6194 | |
2109b99e AH |
6195 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6196 | const struct cpumask *cpu_map) | |
6197 | { | |
dce840a0 PZ |
6198 | memset(d, 0, sizeof(*d)); |
6199 | ||
54ab4ff4 PZ |
6200 | if (__sdt_alloc(cpu_map)) |
6201 | return sa_sd_storage; | |
dce840a0 PZ |
6202 | d->sd = alloc_percpu(struct sched_domain *); |
6203 | if (!d->sd) | |
6204 | return sa_sd_storage; | |
2109b99e | 6205 | d->rd = alloc_rootdomain(); |
dce840a0 | 6206 | if (!d->rd) |
21d42ccf | 6207 | return sa_sd; |
2109b99e AH |
6208 | return sa_rootdomain; |
6209 | } | |
57d885fe | 6210 | |
dce840a0 PZ |
6211 | /* |
6212 | * NULL the sd_data elements we've used to build the sched_domain and | |
6213 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6214 | * will not free the data we're using. | |
6215 | */ | |
6216 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6217 | { | |
6218 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6219 | |
6220 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6221 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6222 | ||
e3589f6c | 6223 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6224 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6225 | |
6226 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6227 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6228 | } |
6229 | ||
2c402dc3 PZ |
6230 | #ifdef CONFIG_SCHED_SMT |
6231 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6232 | { |
2c402dc3 | 6233 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6234 | } |
2c402dc3 | 6235 | #endif |
7f4588f3 | 6236 | |
d069b916 PZ |
6237 | /* |
6238 | * Topology list, bottom-up. | |
6239 | */ | |
2c402dc3 | 6240 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6241 | #ifdef CONFIG_SCHED_SMT |
6242 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6243 | #endif |
1e9f28fa | 6244 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6245 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6246 | #endif |
d069b916 PZ |
6247 | #ifdef CONFIG_SCHED_BOOK |
6248 | { sd_init_BOOK, cpu_book_mask, }, | |
6249 | #endif | |
6250 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
6251 | { NULL, }, |
6252 | }; | |
6253 | ||
6254 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6255 | ||
cb83b629 PZ |
6256 | #ifdef CONFIG_NUMA |
6257 | ||
6258 | static int sched_domains_numa_levels; | |
6259 | static int sched_domains_numa_scale; | |
6260 | static int *sched_domains_numa_distance; | |
6261 | static struct cpumask ***sched_domains_numa_masks; | |
6262 | static int sched_domains_curr_level; | |
6263 | ||
6264 | static inline unsigned long numa_scale(unsigned long x, int level) | |
6265 | { | |
6266 | return x * sched_domains_numa_distance[level] / sched_domains_numa_scale; | |
6267 | } | |
6268 | ||
6269 | static inline int sd_local_flags(int level) | |
6270 | { | |
6271 | if (sched_domains_numa_distance[level] > REMOTE_DISTANCE) | |
6272 | return 0; | |
6273 | ||
6274 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
6275 | } | |
6276 | ||
6277 | static struct sched_domain * | |
6278 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
6279 | { | |
6280 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
6281 | int level = tl->numa_level; | |
6282 | int sd_weight = cpumask_weight( | |
6283 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
6284 | ||
6285 | *sd = (struct sched_domain){ | |
6286 | .min_interval = sd_weight, | |
6287 | .max_interval = 2*sd_weight, | |
6288 | .busy_factor = 32, | |
6289 | .imbalance_pct = 100 + numa_scale(25, level), | |
6290 | .cache_nice_tries = 2, | |
6291 | .busy_idx = 3, | |
6292 | .idle_idx = 2, | |
6293 | .newidle_idx = 0, | |
6294 | .wake_idx = 0, | |
6295 | .forkexec_idx = 0, | |
6296 | ||
6297 | .flags = 1*SD_LOAD_BALANCE | |
6298 | | 1*SD_BALANCE_NEWIDLE | |
6299 | | 0*SD_BALANCE_EXEC | |
6300 | | 0*SD_BALANCE_FORK | |
6301 | | 0*SD_BALANCE_WAKE | |
6302 | | 0*SD_WAKE_AFFINE | |
6303 | | 0*SD_PREFER_LOCAL | |
6304 | | 0*SD_SHARE_CPUPOWER | |
6305 | | 0*SD_POWERSAVINGS_BALANCE | |
6306 | | 0*SD_SHARE_PKG_RESOURCES | |
6307 | | 1*SD_SERIALIZE | |
6308 | | 0*SD_PREFER_SIBLING | |
6309 | | sd_local_flags(level) | |
6310 | , | |
6311 | .last_balance = jiffies, | |
6312 | .balance_interval = sd_weight, | |
6313 | }; | |
6314 | SD_INIT_NAME(sd, NUMA); | |
6315 | sd->private = &tl->data; | |
6316 | ||
6317 | /* | |
6318 | * Ugly hack to pass state to sd_numa_mask()... | |
6319 | */ | |
6320 | sched_domains_curr_level = tl->numa_level; | |
6321 | ||
6322 | return sd; | |
6323 | } | |
6324 | ||
6325 | static const struct cpumask *sd_numa_mask(int cpu) | |
6326 | { | |
6327 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6328 | } | |
6329 | ||
6330 | static void sched_init_numa(void) | |
6331 | { | |
6332 | int next_distance, curr_distance = node_distance(0, 0); | |
6333 | struct sched_domain_topology_level *tl; | |
6334 | int level = 0; | |
6335 | int i, j, k; | |
6336 | ||
6337 | sched_domains_numa_scale = curr_distance; | |
6338 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); | |
6339 | if (!sched_domains_numa_distance) | |
6340 | return; | |
6341 | ||
6342 | /* | |
6343 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6344 | * unique distances in the node_distance() table. | |
6345 | * | |
6346 | * Assumes node_distance(0,j) includes all distances in | |
6347 | * node_distance(i,j) in order to avoid cubic time. | |
6348 | * | |
6349 | * XXX: could be optimized to O(n log n) by using sort() | |
6350 | */ | |
6351 | next_distance = curr_distance; | |
6352 | for (i = 0; i < nr_node_ids; i++) { | |
6353 | for (j = 0; j < nr_node_ids; j++) { | |
6354 | int distance = node_distance(0, j); | |
6355 | if (distance > curr_distance && | |
6356 | (distance < next_distance || | |
6357 | next_distance == curr_distance)) | |
6358 | next_distance = distance; | |
6359 | } | |
6360 | if (next_distance != curr_distance) { | |
6361 | sched_domains_numa_distance[level++] = next_distance; | |
6362 | sched_domains_numa_levels = level; | |
6363 | curr_distance = next_distance; | |
6364 | } else break; | |
6365 | } | |
6366 | /* | |
6367 | * 'level' contains the number of unique distances, excluding the | |
6368 | * identity distance node_distance(i,i). | |
6369 | * | |
6370 | * The sched_domains_nume_distance[] array includes the actual distance | |
6371 | * numbers. | |
6372 | */ | |
6373 | ||
6374 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); | |
6375 | if (!sched_domains_numa_masks) | |
6376 | return; | |
6377 | ||
6378 | /* | |
6379 | * Now for each level, construct a mask per node which contains all | |
6380 | * cpus of nodes that are that many hops away from us. | |
6381 | */ | |
6382 | for (i = 0; i < level; i++) { | |
6383 | sched_domains_numa_masks[i] = | |
6384 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6385 | if (!sched_domains_numa_masks[i]) | |
6386 | return; | |
6387 | ||
6388 | for (j = 0; j < nr_node_ids; j++) { | |
6389 | struct cpumask *mask = kzalloc_node(cpumask_size(), GFP_KERNEL, j); | |
6390 | if (!mask) | |
6391 | return; | |
6392 | ||
6393 | sched_domains_numa_masks[i][j] = mask; | |
6394 | ||
6395 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6396 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6397 | continue; |
6398 | ||
6399 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6400 | } | |
6401 | } | |
6402 | } | |
6403 | ||
6404 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6405 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6406 | if (!tl) | |
6407 | return; | |
6408 | ||
6409 | /* | |
6410 | * Copy the default topology bits.. | |
6411 | */ | |
6412 | for (i = 0; default_topology[i].init; i++) | |
6413 | tl[i] = default_topology[i]; | |
6414 | ||
6415 | /* | |
6416 | * .. and append 'j' levels of NUMA goodness. | |
6417 | */ | |
6418 | for (j = 0; j < level; i++, j++) { | |
6419 | tl[i] = (struct sched_domain_topology_level){ | |
6420 | .init = sd_numa_init, | |
6421 | .mask = sd_numa_mask, | |
6422 | .flags = SDTL_OVERLAP, | |
6423 | .numa_level = j, | |
6424 | }; | |
6425 | } | |
6426 | ||
6427 | sched_domain_topology = tl; | |
6428 | } | |
6429 | #else | |
6430 | static inline void sched_init_numa(void) | |
6431 | { | |
6432 | } | |
6433 | #endif /* CONFIG_NUMA */ | |
6434 | ||
54ab4ff4 PZ |
6435 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6436 | { | |
6437 | struct sched_domain_topology_level *tl; | |
6438 | int j; | |
6439 | ||
6440 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6441 | struct sd_data *sdd = &tl->data; | |
6442 | ||
6443 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6444 | if (!sdd->sd) | |
6445 | return -ENOMEM; | |
6446 | ||
6447 | sdd->sg = alloc_percpu(struct sched_group *); | |
6448 | if (!sdd->sg) | |
6449 | return -ENOMEM; | |
6450 | ||
9c3f75cb PZ |
6451 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6452 | if (!sdd->sgp) | |
6453 | return -ENOMEM; | |
6454 | ||
54ab4ff4 PZ |
6455 | for_each_cpu(j, cpu_map) { |
6456 | struct sched_domain *sd; | |
6457 | struct sched_group *sg; | |
9c3f75cb | 6458 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6459 | |
6460 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6461 | GFP_KERNEL, cpu_to_node(j)); | |
6462 | if (!sd) | |
6463 | return -ENOMEM; | |
6464 | ||
6465 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6466 | ||
6467 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6468 | GFP_KERNEL, cpu_to_node(j)); | |
6469 | if (!sg) | |
6470 | return -ENOMEM; | |
6471 | ||
6472 | *per_cpu_ptr(sdd->sg, j) = sg; | |
9c3f75cb PZ |
6473 | |
6474 | sgp = kzalloc_node(sizeof(struct sched_group_power), | |
6475 | GFP_KERNEL, cpu_to_node(j)); | |
6476 | if (!sgp) | |
6477 | return -ENOMEM; | |
6478 | ||
6479 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6480 | } |
6481 | } | |
6482 | ||
6483 | return 0; | |
6484 | } | |
6485 | ||
6486 | static void __sdt_free(const struct cpumask *cpu_map) | |
6487 | { | |
6488 | struct sched_domain_topology_level *tl; | |
6489 | int j; | |
6490 | ||
6491 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6492 | struct sd_data *sdd = &tl->data; | |
6493 | ||
6494 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6495 | struct sched_domain *sd; |
6496 | ||
6497 | if (sdd->sd) { | |
6498 | sd = *per_cpu_ptr(sdd->sd, j); | |
6499 | if (sd && (sd->flags & SD_OVERLAP)) | |
6500 | free_sched_groups(sd->groups, 0); | |
6501 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6502 | } | |
6503 | ||
6504 | if (sdd->sg) | |
6505 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6506 | if (sdd->sgp) | |
6507 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6508 | } |
6509 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6510 | sdd->sd = NULL; |
54ab4ff4 | 6511 | free_percpu(sdd->sg); |
fb2cf2c6 | 6512 | sdd->sg = NULL; |
9c3f75cb | 6513 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6514 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6515 | } |
6516 | } | |
6517 | ||
2c402dc3 PZ |
6518 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6519 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6520 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6521 | int cpu) |
6522 | { | |
54ab4ff4 | 6523 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6524 | if (!sd) |
d069b916 | 6525 | return child; |
2c402dc3 PZ |
6526 | |
6527 | set_domain_attribute(sd, attr); | |
6528 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
6529 | if (child) { |
6530 | sd->level = child->level + 1; | |
6531 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6532 | child->parent = sd; |
60495e77 | 6533 | } |
d069b916 | 6534 | sd->child = child; |
2c402dc3 PZ |
6535 | |
6536 | return sd; | |
6537 | } | |
6538 | ||
2109b99e AH |
6539 | /* |
6540 | * Build sched domains for a given set of cpus and attach the sched domains | |
6541 | * to the individual cpus | |
6542 | */ | |
dce840a0 PZ |
6543 | static int build_sched_domains(const struct cpumask *cpu_map, |
6544 | struct sched_domain_attr *attr) | |
2109b99e AH |
6545 | { |
6546 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6547 | struct sched_domain *sd; |
2109b99e | 6548 | struct s_data d; |
822ff793 | 6549 | int i, ret = -ENOMEM; |
9c1cfda2 | 6550 | |
2109b99e AH |
6551 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6552 | if (alloc_state != sa_rootdomain) | |
6553 | goto error; | |
9c1cfda2 | 6554 | |
dce840a0 | 6555 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6556 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6557 | struct sched_domain_topology_level *tl; |
6558 | ||
3bd65a80 | 6559 | sd = NULL; |
e3589f6c | 6560 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6561 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6562 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6563 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6564 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6565 | break; | |
e3589f6c | 6566 | } |
d274cb30 | 6567 | |
d069b916 PZ |
6568 | while (sd->child) |
6569 | sd = sd->child; | |
6570 | ||
21d42ccf | 6571 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6572 | } |
6573 | ||
6574 | /* Build the groups for the domains */ | |
6575 | for_each_cpu(i, cpu_map) { | |
6576 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6577 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6578 | if (sd->flags & SD_OVERLAP) { |
6579 | if (build_overlap_sched_groups(sd, i)) | |
6580 | goto error; | |
6581 | } else { | |
6582 | if (build_sched_groups(sd, i)) | |
6583 | goto error; | |
6584 | } | |
1cf51902 | 6585 | } |
a06dadbe | 6586 | } |
9c1cfda2 | 6587 | |
1da177e4 | 6588 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6589 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6590 | if (!cpumask_test_cpu(i, cpu_map)) | |
6591 | continue; | |
9c1cfda2 | 6592 | |
dce840a0 PZ |
6593 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6594 | claim_allocations(i, sd); | |
cd4ea6ae | 6595 | init_sched_groups_power(i, sd); |
dce840a0 | 6596 | } |
f712c0c7 | 6597 | } |
9c1cfda2 | 6598 | |
1da177e4 | 6599 | /* Attach the domains */ |
dce840a0 | 6600 | rcu_read_lock(); |
abcd083a | 6601 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6602 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6603 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6604 | } |
dce840a0 | 6605 | rcu_read_unlock(); |
51888ca2 | 6606 | |
822ff793 | 6607 | ret = 0; |
51888ca2 | 6608 | error: |
2109b99e | 6609 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6610 | return ret; |
1da177e4 | 6611 | } |
029190c5 | 6612 | |
acc3f5d7 | 6613 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6614 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6615 | static struct sched_domain_attr *dattr_cur; |
6616 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6617 | |
6618 | /* | |
6619 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6620 | * cpumask) fails, then fallback to a single sched domain, |
6621 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6622 | */ |
4212823f | 6623 | static cpumask_var_t fallback_doms; |
029190c5 | 6624 | |
ee79d1bd HC |
6625 | /* |
6626 | * arch_update_cpu_topology lets virtualized architectures update the | |
6627 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6628 | * or 0 if it stayed the same. | |
6629 | */ | |
6630 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6631 | { |
ee79d1bd | 6632 | return 0; |
22e52b07 HC |
6633 | } |
6634 | ||
acc3f5d7 RR |
6635 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6636 | { | |
6637 | int i; | |
6638 | cpumask_var_t *doms; | |
6639 | ||
6640 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6641 | if (!doms) | |
6642 | return NULL; | |
6643 | for (i = 0; i < ndoms; i++) { | |
6644 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6645 | free_sched_domains(doms, i); | |
6646 | return NULL; | |
6647 | } | |
6648 | } | |
6649 | return doms; | |
6650 | } | |
6651 | ||
6652 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6653 | { | |
6654 | unsigned int i; | |
6655 | for (i = 0; i < ndoms; i++) | |
6656 | free_cpumask_var(doms[i]); | |
6657 | kfree(doms); | |
6658 | } | |
6659 | ||
1a20ff27 | 6660 | /* |
41a2d6cf | 6661 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6662 | * For now this just excludes isolated cpus, but could be used to |
6663 | * exclude other special cases in the future. | |
1a20ff27 | 6664 | */ |
c4a8849a | 6665 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6666 | { |
7378547f MM |
6667 | int err; |
6668 | ||
22e52b07 | 6669 | arch_update_cpu_topology(); |
029190c5 | 6670 | ndoms_cur = 1; |
acc3f5d7 | 6671 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6672 | if (!doms_cur) |
acc3f5d7 RR |
6673 | doms_cur = &fallback_doms; |
6674 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 6675 | dattr_cur = NULL; |
dce840a0 | 6676 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6677 | register_sched_domain_sysctl(); |
7378547f MM |
6678 | |
6679 | return err; | |
1a20ff27 DG |
6680 | } |
6681 | ||
1a20ff27 DG |
6682 | /* |
6683 | * Detach sched domains from a group of cpus specified in cpu_map | |
6684 | * These cpus will now be attached to the NULL domain | |
6685 | */ | |
96f874e2 | 6686 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6687 | { |
6688 | int i; | |
6689 | ||
dce840a0 | 6690 | rcu_read_lock(); |
abcd083a | 6691 | for_each_cpu(i, cpu_map) |
57d885fe | 6692 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6693 | rcu_read_unlock(); |
1a20ff27 DG |
6694 | } |
6695 | ||
1d3504fc HS |
6696 | /* handle null as "default" */ |
6697 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6698 | struct sched_domain_attr *new, int idx_new) | |
6699 | { | |
6700 | struct sched_domain_attr tmp; | |
6701 | ||
6702 | /* fast path */ | |
6703 | if (!new && !cur) | |
6704 | return 1; | |
6705 | ||
6706 | tmp = SD_ATTR_INIT; | |
6707 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6708 | new ? (new + idx_new) : &tmp, | |
6709 | sizeof(struct sched_domain_attr)); | |
6710 | } | |
6711 | ||
029190c5 PJ |
6712 | /* |
6713 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6714 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6715 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6716 | * It destroys each deleted domain and builds each new domain. | |
6717 | * | |
acc3f5d7 | 6718 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6719 | * The masks don't intersect (don't overlap.) We should setup one |
6720 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6721 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6722 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6723 | * it as it is. | |
6724 | * | |
acc3f5d7 RR |
6725 | * The passed in 'doms_new' should be allocated using |
6726 | * alloc_sched_domains. This routine takes ownership of it and will | |
6727 | * free_sched_domains it when done with it. If the caller failed the | |
6728 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6729 | * and partition_sched_domains() will fallback to the single partition | |
6730 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6731 | * |
96f874e2 | 6732 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6733 | * ndoms_new == 0 is a special case for destroying existing domains, |
6734 | * and it will not create the default domain. | |
dfb512ec | 6735 | * |
029190c5 PJ |
6736 | * Call with hotplug lock held |
6737 | */ | |
acc3f5d7 | 6738 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6739 | struct sched_domain_attr *dattr_new) |
029190c5 | 6740 | { |
dfb512ec | 6741 | int i, j, n; |
d65bd5ec | 6742 | int new_topology; |
029190c5 | 6743 | |
712555ee | 6744 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6745 | |
7378547f MM |
6746 | /* always unregister in case we don't destroy any domains */ |
6747 | unregister_sched_domain_sysctl(); | |
6748 | ||
d65bd5ec HC |
6749 | /* Let architecture update cpu core mappings. */ |
6750 | new_topology = arch_update_cpu_topology(); | |
6751 | ||
dfb512ec | 6752 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6753 | |
6754 | /* Destroy deleted domains */ | |
6755 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6756 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6757 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6758 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6759 | goto match1; |
6760 | } | |
6761 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6762 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6763 | match1: |
6764 | ; | |
6765 | } | |
6766 | ||
e761b772 MK |
6767 | if (doms_new == NULL) { |
6768 | ndoms_cur = 0; | |
acc3f5d7 | 6769 | doms_new = &fallback_doms; |
6ad4c188 | 6770 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6771 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6772 | } |
6773 | ||
029190c5 PJ |
6774 | /* Build new domains */ |
6775 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 6776 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 6777 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6778 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6779 | goto match2; |
6780 | } | |
6781 | /* no match - add a new doms_new */ | |
dce840a0 | 6782 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6783 | match2: |
6784 | ; | |
6785 | } | |
6786 | ||
6787 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6788 | if (doms_cur != &fallback_doms) |
6789 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6790 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6791 | doms_cur = doms_new; |
1d3504fc | 6792 | dattr_cur = dattr_new; |
029190c5 | 6793 | ndoms_cur = ndoms_new; |
7378547f MM |
6794 | |
6795 | register_sched_domain_sysctl(); | |
a1835615 | 6796 | |
712555ee | 6797 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6798 | } |
6799 | ||
5c45bf27 | 6800 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 6801 | static void reinit_sched_domains(void) |
5c45bf27 | 6802 | { |
95402b38 | 6803 | get_online_cpus(); |
dfb512ec MK |
6804 | |
6805 | /* Destroy domains first to force the rebuild */ | |
6806 | partition_sched_domains(0, NULL, NULL); | |
6807 | ||
e761b772 | 6808 | rebuild_sched_domains(); |
95402b38 | 6809 | put_online_cpus(); |
5c45bf27 SS |
6810 | } |
6811 | ||
6812 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6813 | { | |
afb8a9b7 | 6814 | unsigned int level = 0; |
5c45bf27 | 6815 | |
afb8a9b7 GS |
6816 | if (sscanf(buf, "%u", &level) != 1) |
6817 | return -EINVAL; | |
6818 | ||
6819 | /* | |
6820 | * level is always be positive so don't check for | |
6821 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
6822 | * What happens on 0 or 1 byte write, | |
6823 | * need to check for count as well? | |
6824 | */ | |
6825 | ||
6826 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
6827 | return -EINVAL; |
6828 | ||
6829 | if (smt) | |
afb8a9b7 | 6830 | sched_smt_power_savings = level; |
5c45bf27 | 6831 | else |
afb8a9b7 | 6832 | sched_mc_power_savings = level; |
5c45bf27 | 6833 | |
c4a8849a | 6834 | reinit_sched_domains(); |
5c45bf27 | 6835 | |
c70f22d2 | 6836 | return count; |
5c45bf27 SS |
6837 | } |
6838 | ||
5c45bf27 | 6839 | #ifdef CONFIG_SCHED_MC |
8a25a2fd KS |
6840 | static ssize_t sched_mc_power_savings_show(struct device *dev, |
6841 | struct device_attribute *attr, | |
6842 | char *buf) | |
5c45bf27 | 6843 | { |
8a25a2fd | 6844 | return sprintf(buf, "%u\n", sched_mc_power_savings); |
5c45bf27 | 6845 | } |
8a25a2fd KS |
6846 | static ssize_t sched_mc_power_savings_store(struct device *dev, |
6847 | struct device_attribute *attr, | |
48f24c4d | 6848 | const char *buf, size_t count) |
5c45bf27 SS |
6849 | { |
6850 | return sched_power_savings_store(buf, count, 0); | |
6851 | } | |
8a25a2fd KS |
6852 | static DEVICE_ATTR(sched_mc_power_savings, 0644, |
6853 | sched_mc_power_savings_show, | |
6854 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6855 | #endif |
6856 | ||
6857 | #ifdef CONFIG_SCHED_SMT | |
8a25a2fd KS |
6858 | static ssize_t sched_smt_power_savings_show(struct device *dev, |
6859 | struct device_attribute *attr, | |
6860 | char *buf) | |
5c45bf27 | 6861 | { |
8a25a2fd | 6862 | return sprintf(buf, "%u\n", sched_smt_power_savings); |
5c45bf27 | 6863 | } |
8a25a2fd KS |
6864 | static ssize_t sched_smt_power_savings_store(struct device *dev, |
6865 | struct device_attribute *attr, | |
48f24c4d | 6866 | const char *buf, size_t count) |
5c45bf27 SS |
6867 | { |
6868 | return sched_power_savings_store(buf, count, 1); | |
6869 | } | |
8a25a2fd | 6870 | static DEVICE_ATTR(sched_smt_power_savings, 0644, |
f718cd4a | 6871 | sched_smt_power_savings_show, |
6707de00 AB |
6872 | sched_smt_power_savings_store); |
6873 | #endif | |
6874 | ||
8a25a2fd | 6875 | int __init sched_create_sysfs_power_savings_entries(struct device *dev) |
6707de00 AB |
6876 | { |
6877 | int err = 0; | |
6878 | ||
6879 | #ifdef CONFIG_SCHED_SMT | |
6880 | if (smt_capable()) | |
8a25a2fd | 6881 | err = device_create_file(dev, &dev_attr_sched_smt_power_savings); |
6707de00 AB |
6882 | #endif |
6883 | #ifdef CONFIG_SCHED_MC | |
6884 | if (!err && mc_capable()) | |
8a25a2fd | 6885 | err = device_create_file(dev, &dev_attr_sched_mc_power_savings); |
6707de00 AB |
6886 | #endif |
6887 | return err; | |
6888 | } | |
6d6bc0ad | 6889 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 6890 | |
1da177e4 | 6891 | /* |
3a101d05 TH |
6892 | * Update cpusets according to cpu_active mask. If cpusets are |
6893 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6894 | * around partition_sched_domains(). | |
1da177e4 | 6895 | */ |
0b2e918a TH |
6896 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6897 | void *hcpu) | |
e761b772 | 6898 | { |
3a101d05 | 6899 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 6900 | case CPU_ONLINE: |
6ad4c188 | 6901 | case CPU_DOWN_FAILED: |
3a101d05 | 6902 | cpuset_update_active_cpus(); |
e761b772 | 6903 | return NOTIFY_OK; |
3a101d05 TH |
6904 | default: |
6905 | return NOTIFY_DONE; | |
6906 | } | |
6907 | } | |
e761b772 | 6908 | |
0b2e918a TH |
6909 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6910 | void *hcpu) | |
3a101d05 TH |
6911 | { |
6912 | switch (action & ~CPU_TASKS_FROZEN) { | |
6913 | case CPU_DOWN_PREPARE: | |
6914 | cpuset_update_active_cpus(); | |
6915 | return NOTIFY_OK; | |
e761b772 MK |
6916 | default: |
6917 | return NOTIFY_DONE; | |
6918 | } | |
6919 | } | |
e761b772 | 6920 | |
1da177e4 LT |
6921 | void __init sched_init_smp(void) |
6922 | { | |
dcc30a35 RR |
6923 | cpumask_var_t non_isolated_cpus; |
6924 | ||
6925 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6926 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6927 | |
cb83b629 PZ |
6928 | sched_init_numa(); |
6929 | ||
95402b38 | 6930 | get_online_cpus(); |
712555ee | 6931 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6932 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6933 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6934 | if (cpumask_empty(non_isolated_cpus)) | |
6935 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6936 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 6937 | put_online_cpus(); |
e761b772 | 6938 | |
3a101d05 TH |
6939 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6940 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
6941 | |
6942 | /* RT runtime code needs to handle some hotplug events */ | |
6943 | hotcpu_notifier(update_runtime, 0); | |
6944 | ||
b328ca18 | 6945 | init_hrtick(); |
5c1e1767 NP |
6946 | |
6947 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6948 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6949 | BUG(); |
19978ca6 | 6950 | sched_init_granularity(); |
dcc30a35 | 6951 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6952 | |
0e3900e6 | 6953 | init_sched_rt_class(); |
1da177e4 LT |
6954 | } |
6955 | #else | |
6956 | void __init sched_init_smp(void) | |
6957 | { | |
19978ca6 | 6958 | sched_init_granularity(); |
1da177e4 LT |
6959 | } |
6960 | #endif /* CONFIG_SMP */ | |
6961 | ||
cd1bb94b AB |
6962 | const_debug unsigned int sysctl_timer_migration = 1; |
6963 | ||
1da177e4 LT |
6964 | int in_sched_functions(unsigned long addr) |
6965 | { | |
1da177e4 LT |
6966 | return in_lock_functions(addr) || |
6967 | (addr >= (unsigned long)__sched_text_start | |
6968 | && addr < (unsigned long)__sched_text_end); | |
6969 | } | |
6970 | ||
029632fb PZ |
6971 | #ifdef CONFIG_CGROUP_SCHED |
6972 | struct task_group root_task_group; | |
052f1dc7 | 6973 | #endif |
6f505b16 | 6974 | |
029632fb | 6975 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 6976 | |
1da177e4 LT |
6977 | void __init sched_init(void) |
6978 | { | |
dd41f596 | 6979 | int i, j; |
434d53b0 MT |
6980 | unsigned long alloc_size = 0, ptr; |
6981 | ||
6982 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6983 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6984 | #endif | |
6985 | #ifdef CONFIG_RT_GROUP_SCHED | |
6986 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6987 | #endif |
df7c8e84 | 6988 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6989 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6990 | #endif |
434d53b0 | 6991 | if (alloc_size) { |
36b7b6d4 | 6992 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6993 | |
6994 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6995 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6996 | ptr += nr_cpu_ids * sizeof(void **); |
6997 | ||
07e06b01 | 6998 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6999 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7000 | |
6d6bc0ad | 7001 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7002 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7003 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7004 | ptr += nr_cpu_ids * sizeof(void **); |
7005 | ||
07e06b01 | 7006 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7007 | ptr += nr_cpu_ids * sizeof(void **); |
7008 | ||
6d6bc0ad | 7009 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7010 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7011 | for_each_possible_cpu(i) { | |
7012 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7013 | ptr += cpumask_size(); | |
7014 | } | |
7015 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7016 | } |
dd41f596 | 7017 | |
57d885fe GH |
7018 | #ifdef CONFIG_SMP |
7019 | init_defrootdomain(); | |
7020 | #endif | |
7021 | ||
d0b27fa7 PZ |
7022 | init_rt_bandwidth(&def_rt_bandwidth, |
7023 | global_rt_period(), global_rt_runtime()); | |
7024 | ||
7025 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 7026 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7027 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7028 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7029 | |
7c941438 | 7030 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7031 | list_add(&root_task_group.list, &task_groups); |
7032 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7033 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7034 | autogroup_init(&init_task); |
54c707e9 | 7035 | |
7c941438 | 7036 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7037 | |
54c707e9 GC |
7038 | #ifdef CONFIG_CGROUP_CPUACCT |
7039 | root_cpuacct.cpustat = &kernel_cpustat; | |
7040 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
7041 | /* Too early, not expected to fail */ | |
7042 | BUG_ON(!root_cpuacct.cpuusage); | |
7043 | #endif | |
0a945022 | 7044 | for_each_possible_cpu(i) { |
70b97a7f | 7045 | struct rq *rq; |
1da177e4 LT |
7046 | |
7047 | rq = cpu_rq(i); | |
05fa785c | 7048 | raw_spin_lock_init(&rq->lock); |
7897986b | 7049 | rq->nr_running = 0; |
dce48a84 TG |
7050 | rq->calc_load_active = 0; |
7051 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7052 | init_cfs_rq(&rq->cfs); |
6f505b16 | 7053 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7054 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7055 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7056 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7057 | /* |
07e06b01 | 7058 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7059 | * |
7060 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7061 | * gets 100% of the cpu resources in the system. This overall | |
7062 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7063 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7064 | * based on each entity's (task or task-group's) weight |
7065 | * (se->load.weight). | |
7066 | * | |
07e06b01 | 7067 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7068 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7069 | * then A0's share of the cpu resource is: | |
7070 | * | |
0d905bca | 7071 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7072 | * |
07e06b01 YZ |
7073 | * We achieve this by letting root_task_group's tasks sit |
7074 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7075 | */ |
ab84d31e | 7076 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7077 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7078 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7079 | ||
7080 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7081 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7082 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7083 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7084 | #endif |
1da177e4 | 7085 | |
dd41f596 IM |
7086 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7087 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7088 | |
7089 | rq->last_load_update_tick = jiffies; | |
7090 | ||
1da177e4 | 7091 | #ifdef CONFIG_SMP |
41c7ce9a | 7092 | rq->sd = NULL; |
57d885fe | 7093 | rq->rd = NULL; |
1399fa78 | 7094 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 7095 | rq->post_schedule = 0; |
1da177e4 | 7096 | rq->active_balance = 0; |
dd41f596 | 7097 | rq->next_balance = jiffies; |
1da177e4 | 7098 | rq->push_cpu = 0; |
0a2966b4 | 7099 | rq->cpu = i; |
1f11eb6a | 7100 | rq->online = 0; |
eae0c9df MG |
7101 | rq->idle_stamp = 0; |
7102 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
367456c7 PZ |
7103 | |
7104 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7105 | ||
dc938520 | 7106 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 | 7107 | #ifdef CONFIG_NO_HZ |
1c792db7 | 7108 | rq->nohz_flags = 0; |
83cd4fe2 | 7109 | #endif |
1da177e4 | 7110 | #endif |
8f4d37ec | 7111 | init_rq_hrtick(rq); |
1da177e4 | 7112 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7113 | } |
7114 | ||
2dd73a4f | 7115 | set_load_weight(&init_task); |
b50f60ce | 7116 | |
e107be36 AK |
7117 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7118 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7119 | #endif | |
7120 | ||
b50f60ce | 7121 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 7122 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
7123 | #endif |
7124 | ||
1da177e4 LT |
7125 | /* |
7126 | * The boot idle thread does lazy MMU switching as well: | |
7127 | */ | |
7128 | atomic_inc(&init_mm.mm_count); | |
7129 | enter_lazy_tlb(&init_mm, current); | |
7130 | ||
7131 | /* | |
7132 | * Make us the idle thread. Technically, schedule() should not be | |
7133 | * called from this thread, however somewhere below it might be, | |
7134 | * but because we are the idle thread, we just pick up running again | |
7135 | * when this runqueue becomes "idle". | |
7136 | */ | |
7137 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7138 | |
7139 | calc_load_update = jiffies + LOAD_FREQ; | |
7140 | ||
dd41f596 IM |
7141 | /* |
7142 | * During early bootup we pretend to be a normal task: | |
7143 | */ | |
7144 | current->sched_class = &fair_sched_class; | |
6892b75e | 7145 | |
bf4d83f6 | 7146 | #ifdef CONFIG_SMP |
4cb98839 | 7147 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7148 | /* May be allocated at isolcpus cmdline parse time */ |
7149 | if (cpu_isolated_map == NULL) | |
7150 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
029632fb PZ |
7151 | #endif |
7152 | init_sched_fair_class(); | |
6a7b3dc3 | 7153 | |
6892b75e | 7154 | scheduler_running = 1; |
1da177e4 LT |
7155 | } |
7156 | ||
d902db1e | 7157 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7158 | static inline int preempt_count_equals(int preempt_offset) |
7159 | { | |
234da7bc | 7160 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7161 | |
4ba8216c | 7162 | return (nested == preempt_offset); |
e4aafea2 FW |
7163 | } |
7164 | ||
d894837f | 7165 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7166 | { |
1da177e4 LT |
7167 | static unsigned long prev_jiffy; /* ratelimiting */ |
7168 | ||
b3fbab05 | 7169 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
7170 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7171 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7172 | return; |
7173 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7174 | return; | |
7175 | prev_jiffy = jiffies; | |
7176 | ||
3df0fc5b PZ |
7177 | printk(KERN_ERR |
7178 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7179 | file, line); | |
7180 | printk(KERN_ERR | |
7181 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7182 | in_atomic(), irqs_disabled(), | |
7183 | current->pid, current->comm); | |
aef745fc IM |
7184 | |
7185 | debug_show_held_locks(current); | |
7186 | if (irqs_disabled()) | |
7187 | print_irqtrace_events(current); | |
7188 | dump_stack(); | |
1da177e4 LT |
7189 | } |
7190 | EXPORT_SYMBOL(__might_sleep); | |
7191 | #endif | |
7192 | ||
7193 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7194 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7195 | { | |
da7a735e PZ |
7196 | const struct sched_class *prev_class = p->sched_class; |
7197 | int old_prio = p->prio; | |
3a5e4dc1 | 7198 | int on_rq; |
3e51f33f | 7199 | |
fd2f4419 | 7200 | on_rq = p->on_rq; |
3a5e4dc1 | 7201 | if (on_rq) |
4ca9b72b | 7202 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
7203 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7204 | if (on_rq) { | |
4ca9b72b | 7205 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7206 | resched_task(rq->curr); |
7207 | } | |
da7a735e PZ |
7208 | |
7209 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7210 | } |
7211 | ||
1da177e4 LT |
7212 | void normalize_rt_tasks(void) |
7213 | { | |
a0f98a1c | 7214 | struct task_struct *g, *p; |
1da177e4 | 7215 | unsigned long flags; |
70b97a7f | 7216 | struct rq *rq; |
1da177e4 | 7217 | |
4cf5d77a | 7218 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7219 | do_each_thread(g, p) { |
178be793 IM |
7220 | /* |
7221 | * Only normalize user tasks: | |
7222 | */ | |
7223 | if (!p->mm) | |
7224 | continue; | |
7225 | ||
6cfb0d5d | 7226 | p->se.exec_start = 0; |
6cfb0d5d | 7227 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7228 | p->se.statistics.wait_start = 0; |
7229 | p->se.statistics.sleep_start = 0; | |
7230 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7231 | #endif |
dd41f596 IM |
7232 | |
7233 | if (!rt_task(p)) { | |
7234 | /* | |
7235 | * Renice negative nice level userspace | |
7236 | * tasks back to 0: | |
7237 | */ | |
7238 | if (TASK_NICE(p) < 0 && p->mm) | |
7239 | set_user_nice(p, 0); | |
1da177e4 | 7240 | continue; |
dd41f596 | 7241 | } |
1da177e4 | 7242 | |
1d615482 | 7243 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7244 | rq = __task_rq_lock(p); |
1da177e4 | 7245 | |
178be793 | 7246 | normalize_task(rq, p); |
3a5e4dc1 | 7247 | |
b29739f9 | 7248 | __task_rq_unlock(rq); |
1d615482 | 7249 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7250 | } while_each_thread(g, p); |
7251 | ||
4cf5d77a | 7252 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7253 | } |
7254 | ||
7255 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7256 | |
67fc4e0c | 7257 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7258 | /* |
67fc4e0c | 7259 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7260 | * |
7261 | * They can only be called when the whole system has been | |
7262 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7263 | * activity can take place. Using them for anything else would | |
7264 | * be a serious bug, and as a result, they aren't even visible | |
7265 | * under any other configuration. | |
7266 | */ | |
7267 | ||
7268 | /** | |
7269 | * curr_task - return the current task for a given cpu. | |
7270 | * @cpu: the processor in question. | |
7271 | * | |
7272 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7273 | */ | |
36c8b586 | 7274 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7275 | { |
7276 | return cpu_curr(cpu); | |
7277 | } | |
7278 | ||
67fc4e0c JW |
7279 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7280 | ||
7281 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7282 | /** |
7283 | * set_curr_task - set the current task for a given cpu. | |
7284 | * @cpu: the processor in question. | |
7285 | * @p: the task pointer to set. | |
7286 | * | |
7287 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7288 | * are serviced on a separate stack. It allows the architecture to switch the |
7289 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7290 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7291 | * and caller must save the original value of the current task (see | |
7292 | * curr_task() above) and restore that value before reenabling interrupts and | |
7293 | * re-starting the system. | |
7294 | * | |
7295 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7296 | */ | |
36c8b586 | 7297 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7298 | { |
7299 | cpu_curr(cpu) = p; | |
7300 | } | |
7301 | ||
7302 | #endif | |
29f59db3 | 7303 | |
7c941438 | 7304 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7305 | /* task_group_lock serializes the addition/removal of task groups */ |
7306 | static DEFINE_SPINLOCK(task_group_lock); | |
7307 | ||
bccbe08a PZ |
7308 | static void free_sched_group(struct task_group *tg) |
7309 | { | |
7310 | free_fair_sched_group(tg); | |
7311 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7312 | autogroup_free(tg); |
bccbe08a PZ |
7313 | kfree(tg); |
7314 | } | |
7315 | ||
7316 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7317 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7318 | { |
7319 | struct task_group *tg; | |
7320 | unsigned long flags; | |
bccbe08a PZ |
7321 | |
7322 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7323 | if (!tg) | |
7324 | return ERR_PTR(-ENOMEM); | |
7325 | ||
ec7dc8ac | 7326 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7327 | goto err; |
7328 | ||
ec7dc8ac | 7329 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7330 | goto err; |
7331 | ||
8ed36996 | 7332 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7333 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7334 | |
7335 | WARN_ON(!parent); /* root should already exist */ | |
7336 | ||
7337 | tg->parent = parent; | |
f473aa5e | 7338 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7339 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7340 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7341 | |
9b5b7751 | 7342 | return tg; |
29f59db3 SV |
7343 | |
7344 | err: | |
6f505b16 | 7345 | free_sched_group(tg); |
29f59db3 SV |
7346 | return ERR_PTR(-ENOMEM); |
7347 | } | |
7348 | ||
9b5b7751 | 7349 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7350 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7351 | { |
29f59db3 | 7352 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7353 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7354 | } |
7355 | ||
9b5b7751 | 7356 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7357 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7358 | { |
8ed36996 | 7359 | unsigned long flags; |
9b5b7751 | 7360 | int i; |
29f59db3 | 7361 | |
3d4b47b4 PZ |
7362 | /* end participation in shares distribution */ |
7363 | for_each_possible_cpu(i) | |
bccbe08a | 7364 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7365 | |
7366 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7367 | list_del_rcu(&tg->list); |
f473aa5e | 7368 | list_del_rcu(&tg->siblings); |
8ed36996 | 7369 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7370 | |
9b5b7751 | 7371 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7372 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7373 | } |
7374 | ||
9b5b7751 | 7375 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7376 | * The caller of this function should have put the task in its new group |
7377 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7378 | * reflect its new group. | |
9b5b7751 SV |
7379 | */ |
7380 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7381 | { |
7382 | int on_rq, running; | |
7383 | unsigned long flags; | |
7384 | struct rq *rq; | |
7385 | ||
7386 | rq = task_rq_lock(tsk, &flags); | |
7387 | ||
051a1d1a | 7388 | running = task_current(rq, tsk); |
fd2f4419 | 7389 | on_rq = tsk->on_rq; |
29f59db3 | 7390 | |
0e1f3483 | 7391 | if (on_rq) |
29f59db3 | 7392 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7393 | if (unlikely(running)) |
7394 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7395 | |
810b3817 | 7396 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7397 | if (tsk->sched_class->task_move_group) |
7398 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7399 | else | |
810b3817 | 7400 | #endif |
b2b5ce02 | 7401 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7402 | |
0e1f3483 HS |
7403 | if (unlikely(running)) |
7404 | tsk->sched_class->set_curr_task(rq); | |
7405 | if (on_rq) | |
371fd7e7 | 7406 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7407 | |
0122ec5b | 7408 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7409 | } |
7c941438 | 7410 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7411 | |
a790de99 | 7412 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7413 | static unsigned long to_ratio(u64 period, u64 runtime) |
7414 | { | |
7415 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7416 | return 1ULL << 20; |
9f0c1e56 | 7417 | |
9a7e0b18 | 7418 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7419 | } |
a790de99 PT |
7420 | #endif |
7421 | ||
7422 | #ifdef CONFIG_RT_GROUP_SCHED | |
7423 | /* | |
7424 | * Ensure that the real time constraints are schedulable. | |
7425 | */ | |
7426 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7427 | |
9a7e0b18 PZ |
7428 | /* Must be called with tasklist_lock held */ |
7429 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7430 | { |
9a7e0b18 | 7431 | struct task_struct *g, *p; |
b40b2e8e | 7432 | |
9a7e0b18 | 7433 | do_each_thread(g, p) { |
029632fb | 7434 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7435 | return 1; |
7436 | } while_each_thread(g, p); | |
b40b2e8e | 7437 | |
9a7e0b18 PZ |
7438 | return 0; |
7439 | } | |
b40b2e8e | 7440 | |
9a7e0b18 PZ |
7441 | struct rt_schedulable_data { |
7442 | struct task_group *tg; | |
7443 | u64 rt_period; | |
7444 | u64 rt_runtime; | |
7445 | }; | |
b40b2e8e | 7446 | |
a790de99 | 7447 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7448 | { |
7449 | struct rt_schedulable_data *d = data; | |
7450 | struct task_group *child; | |
7451 | unsigned long total, sum = 0; | |
7452 | u64 period, runtime; | |
b40b2e8e | 7453 | |
9a7e0b18 PZ |
7454 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7455 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7456 | |
9a7e0b18 PZ |
7457 | if (tg == d->tg) { |
7458 | period = d->rt_period; | |
7459 | runtime = d->rt_runtime; | |
b40b2e8e | 7460 | } |
b40b2e8e | 7461 | |
4653f803 PZ |
7462 | /* |
7463 | * Cannot have more runtime than the period. | |
7464 | */ | |
7465 | if (runtime > period && runtime != RUNTIME_INF) | |
7466 | return -EINVAL; | |
6f505b16 | 7467 | |
4653f803 PZ |
7468 | /* |
7469 | * Ensure we don't starve existing RT tasks. | |
7470 | */ | |
9a7e0b18 PZ |
7471 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7472 | return -EBUSY; | |
6f505b16 | 7473 | |
9a7e0b18 | 7474 | total = to_ratio(period, runtime); |
6f505b16 | 7475 | |
4653f803 PZ |
7476 | /* |
7477 | * Nobody can have more than the global setting allows. | |
7478 | */ | |
7479 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7480 | return -EINVAL; | |
6f505b16 | 7481 | |
4653f803 PZ |
7482 | /* |
7483 | * The sum of our children's runtime should not exceed our own. | |
7484 | */ | |
9a7e0b18 PZ |
7485 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7486 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7487 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7488 | |
9a7e0b18 PZ |
7489 | if (child == d->tg) { |
7490 | period = d->rt_period; | |
7491 | runtime = d->rt_runtime; | |
7492 | } | |
6f505b16 | 7493 | |
9a7e0b18 | 7494 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7495 | } |
6f505b16 | 7496 | |
9a7e0b18 PZ |
7497 | if (sum > total) |
7498 | return -EINVAL; | |
7499 | ||
7500 | return 0; | |
6f505b16 PZ |
7501 | } |
7502 | ||
9a7e0b18 | 7503 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7504 | { |
8277434e PT |
7505 | int ret; |
7506 | ||
9a7e0b18 PZ |
7507 | struct rt_schedulable_data data = { |
7508 | .tg = tg, | |
7509 | .rt_period = period, | |
7510 | .rt_runtime = runtime, | |
7511 | }; | |
7512 | ||
8277434e PT |
7513 | rcu_read_lock(); |
7514 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7515 | rcu_read_unlock(); | |
7516 | ||
7517 | return ret; | |
521f1a24 DG |
7518 | } |
7519 | ||
ab84d31e | 7520 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7521 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7522 | { |
ac086bc2 | 7523 | int i, err = 0; |
9f0c1e56 | 7524 | |
9f0c1e56 | 7525 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7526 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7527 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7528 | if (err) | |
9f0c1e56 | 7529 | goto unlock; |
ac086bc2 | 7530 | |
0986b11b | 7531 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7532 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7533 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7534 | |
7535 | for_each_possible_cpu(i) { | |
7536 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7537 | ||
0986b11b | 7538 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7539 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7540 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7541 | } |
0986b11b | 7542 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7543 | unlock: |
521f1a24 | 7544 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7545 | mutex_unlock(&rt_constraints_mutex); |
7546 | ||
7547 | return err; | |
6f505b16 PZ |
7548 | } |
7549 | ||
d0b27fa7 PZ |
7550 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7551 | { | |
7552 | u64 rt_runtime, rt_period; | |
7553 | ||
7554 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7555 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7556 | if (rt_runtime_us < 0) | |
7557 | rt_runtime = RUNTIME_INF; | |
7558 | ||
ab84d31e | 7559 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7560 | } |
7561 | ||
9f0c1e56 PZ |
7562 | long sched_group_rt_runtime(struct task_group *tg) |
7563 | { | |
7564 | u64 rt_runtime_us; | |
7565 | ||
d0b27fa7 | 7566 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7567 | return -1; |
7568 | ||
d0b27fa7 | 7569 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7570 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7571 | return rt_runtime_us; | |
7572 | } | |
d0b27fa7 PZ |
7573 | |
7574 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7575 | { | |
7576 | u64 rt_runtime, rt_period; | |
7577 | ||
7578 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7579 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7580 | ||
619b0488 R |
7581 | if (rt_period == 0) |
7582 | return -EINVAL; | |
7583 | ||
ab84d31e | 7584 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7585 | } |
7586 | ||
7587 | long sched_group_rt_period(struct task_group *tg) | |
7588 | { | |
7589 | u64 rt_period_us; | |
7590 | ||
7591 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7592 | do_div(rt_period_us, NSEC_PER_USEC); | |
7593 | return rt_period_us; | |
7594 | } | |
7595 | ||
7596 | static int sched_rt_global_constraints(void) | |
7597 | { | |
4653f803 | 7598 | u64 runtime, period; |
d0b27fa7 PZ |
7599 | int ret = 0; |
7600 | ||
ec5d4989 HS |
7601 | if (sysctl_sched_rt_period <= 0) |
7602 | return -EINVAL; | |
7603 | ||
4653f803 PZ |
7604 | runtime = global_rt_runtime(); |
7605 | period = global_rt_period(); | |
7606 | ||
7607 | /* | |
7608 | * Sanity check on the sysctl variables. | |
7609 | */ | |
7610 | if (runtime > period && runtime != RUNTIME_INF) | |
7611 | return -EINVAL; | |
10b612f4 | 7612 | |
d0b27fa7 | 7613 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7614 | read_lock(&tasklist_lock); |
4653f803 | 7615 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7616 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7617 | mutex_unlock(&rt_constraints_mutex); |
7618 | ||
7619 | return ret; | |
7620 | } | |
54e99124 DG |
7621 | |
7622 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7623 | { | |
7624 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7625 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7626 | return 0; | |
7627 | ||
7628 | return 1; | |
7629 | } | |
7630 | ||
6d6bc0ad | 7631 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7632 | static int sched_rt_global_constraints(void) |
7633 | { | |
ac086bc2 PZ |
7634 | unsigned long flags; |
7635 | int i; | |
7636 | ||
ec5d4989 HS |
7637 | if (sysctl_sched_rt_period <= 0) |
7638 | return -EINVAL; | |
7639 | ||
60aa605d PZ |
7640 | /* |
7641 | * There's always some RT tasks in the root group | |
7642 | * -- migration, kstopmachine etc.. | |
7643 | */ | |
7644 | if (sysctl_sched_rt_runtime == 0) | |
7645 | return -EBUSY; | |
7646 | ||
0986b11b | 7647 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7648 | for_each_possible_cpu(i) { |
7649 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7650 | ||
0986b11b | 7651 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7652 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7653 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7654 | } |
0986b11b | 7655 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7656 | |
d0b27fa7 PZ |
7657 | return 0; |
7658 | } | |
6d6bc0ad | 7659 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7660 | |
7661 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 7662 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7663 | loff_t *ppos) |
7664 | { | |
7665 | int ret; | |
7666 | int old_period, old_runtime; | |
7667 | static DEFINE_MUTEX(mutex); | |
7668 | ||
7669 | mutex_lock(&mutex); | |
7670 | old_period = sysctl_sched_rt_period; | |
7671 | old_runtime = sysctl_sched_rt_runtime; | |
7672 | ||
8d65af78 | 7673 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7674 | |
7675 | if (!ret && write) { | |
7676 | ret = sched_rt_global_constraints(); | |
7677 | if (ret) { | |
7678 | sysctl_sched_rt_period = old_period; | |
7679 | sysctl_sched_rt_runtime = old_runtime; | |
7680 | } else { | |
7681 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7682 | def_rt_bandwidth.rt_period = | |
7683 | ns_to_ktime(global_rt_period()); | |
7684 | } | |
7685 | } | |
7686 | mutex_unlock(&mutex); | |
7687 | ||
7688 | return ret; | |
7689 | } | |
68318b8e | 7690 | |
052f1dc7 | 7691 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7692 | |
7693 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7694 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7695 | { |
2b01dfe3 PM |
7696 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7697 | struct task_group, css); | |
68318b8e SV |
7698 | } |
7699 | ||
761b3ef5 | 7700 | static struct cgroup_subsys_state *cpu_cgroup_create(struct cgroup *cgrp) |
68318b8e | 7701 | { |
ec7dc8ac | 7702 | struct task_group *tg, *parent; |
68318b8e | 7703 | |
2b01dfe3 | 7704 | if (!cgrp->parent) { |
68318b8e | 7705 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7706 | return &root_task_group.css; |
68318b8e SV |
7707 | } |
7708 | ||
ec7dc8ac DG |
7709 | parent = cgroup_tg(cgrp->parent); |
7710 | tg = sched_create_group(parent); | |
68318b8e SV |
7711 | if (IS_ERR(tg)) |
7712 | return ERR_PTR(-ENOMEM); | |
7713 | ||
68318b8e SV |
7714 | return &tg->css; |
7715 | } | |
7716 | ||
761b3ef5 | 7717 | static void cpu_cgroup_destroy(struct cgroup *cgrp) |
68318b8e | 7718 | { |
2b01dfe3 | 7719 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7720 | |
7721 | sched_destroy_group(tg); | |
7722 | } | |
7723 | ||
761b3ef5 | 7724 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
bb9d97b6 | 7725 | struct cgroup_taskset *tset) |
68318b8e | 7726 | { |
bb9d97b6 TH |
7727 | struct task_struct *task; |
7728 | ||
7729 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 7730 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
7731 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
7732 | return -EINVAL; | |
b68aa230 | 7733 | #else |
bb9d97b6 TH |
7734 | /* We don't support RT-tasks being in separate groups */ |
7735 | if (task->sched_class != &fair_sched_class) | |
7736 | return -EINVAL; | |
b68aa230 | 7737 | #endif |
bb9d97b6 | 7738 | } |
be367d09 BB |
7739 | return 0; |
7740 | } | |
68318b8e | 7741 | |
761b3ef5 | 7742 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
bb9d97b6 | 7743 | struct cgroup_taskset *tset) |
68318b8e | 7744 | { |
bb9d97b6 TH |
7745 | struct task_struct *task; |
7746 | ||
7747 | cgroup_taskset_for_each(task, cgrp, tset) | |
7748 | sched_move_task(task); | |
68318b8e SV |
7749 | } |
7750 | ||
068c5cc5 | 7751 | static void |
761b3ef5 LZ |
7752 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
7753 | struct task_struct *task) | |
068c5cc5 PZ |
7754 | { |
7755 | /* | |
7756 | * cgroup_exit() is called in the copy_process() failure path. | |
7757 | * Ignore this case since the task hasn't ran yet, this avoids | |
7758 | * trying to poke a half freed task state from generic code. | |
7759 | */ | |
7760 | if (!(task->flags & PF_EXITING)) | |
7761 | return; | |
7762 | ||
7763 | sched_move_task(task); | |
7764 | } | |
7765 | ||
052f1dc7 | 7766 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 7767 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 7768 | u64 shareval) |
68318b8e | 7769 | { |
c8b28116 | 7770 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
7771 | } |
7772 | ||
f4c753b7 | 7773 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7774 | { |
2b01dfe3 | 7775 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 7776 | |
c8b28116 | 7777 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7778 | } |
ab84d31e PT |
7779 | |
7780 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7781 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7782 | ||
ab84d31e PT |
7783 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7784 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7785 | ||
a790de99 PT |
7786 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7787 | ||
ab84d31e PT |
7788 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7789 | { | |
56f570e5 | 7790 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7791 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7792 | |
7793 | if (tg == &root_task_group) | |
7794 | return -EINVAL; | |
7795 | ||
7796 | /* | |
7797 | * Ensure we have at some amount of bandwidth every period. This is | |
7798 | * to prevent reaching a state of large arrears when throttled via | |
7799 | * entity_tick() resulting in prolonged exit starvation. | |
7800 | */ | |
7801 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7802 | return -EINVAL; | |
7803 | ||
7804 | /* | |
7805 | * Likewise, bound things on the otherside by preventing insane quota | |
7806 | * periods. This also allows us to normalize in computing quota | |
7807 | * feasibility. | |
7808 | */ | |
7809 | if (period > max_cfs_quota_period) | |
7810 | return -EINVAL; | |
7811 | ||
a790de99 PT |
7812 | mutex_lock(&cfs_constraints_mutex); |
7813 | ret = __cfs_schedulable(tg, period, quota); | |
7814 | if (ret) | |
7815 | goto out_unlock; | |
7816 | ||
58088ad0 | 7817 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
7818 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
7819 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
7820 | raw_spin_lock_irq(&cfs_b->lock); |
7821 | cfs_b->period = ns_to_ktime(period); | |
7822 | cfs_b->quota = quota; | |
58088ad0 | 7823 | |
a9cf55b2 | 7824 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7825 | /* restart the period timer (if active) to handle new period expiry */ |
7826 | if (runtime_enabled && cfs_b->timer_active) { | |
7827 | /* force a reprogram */ | |
7828 | cfs_b->timer_active = 0; | |
7829 | __start_cfs_bandwidth(cfs_b); | |
7830 | } | |
ab84d31e PT |
7831 | raw_spin_unlock_irq(&cfs_b->lock); |
7832 | ||
7833 | for_each_possible_cpu(i) { | |
7834 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7835 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7836 | |
7837 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7838 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7839 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7840 | |
029632fb | 7841 | if (cfs_rq->throttled) |
671fd9da | 7842 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7843 | raw_spin_unlock_irq(&rq->lock); |
7844 | } | |
a790de99 PT |
7845 | out_unlock: |
7846 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7847 | |
a790de99 | 7848 | return ret; |
ab84d31e PT |
7849 | } |
7850 | ||
7851 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7852 | { | |
7853 | u64 quota, period; | |
7854 | ||
029632fb | 7855 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7856 | if (cfs_quota_us < 0) |
7857 | quota = RUNTIME_INF; | |
7858 | else | |
7859 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7860 | ||
7861 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7862 | } | |
7863 | ||
7864 | long tg_get_cfs_quota(struct task_group *tg) | |
7865 | { | |
7866 | u64 quota_us; | |
7867 | ||
029632fb | 7868 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7869 | return -1; |
7870 | ||
029632fb | 7871 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7872 | do_div(quota_us, NSEC_PER_USEC); |
7873 | ||
7874 | return quota_us; | |
7875 | } | |
7876 | ||
7877 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7878 | { | |
7879 | u64 quota, period; | |
7880 | ||
7881 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7882 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7883 | |
ab84d31e PT |
7884 | return tg_set_cfs_bandwidth(tg, period, quota); |
7885 | } | |
7886 | ||
7887 | long tg_get_cfs_period(struct task_group *tg) | |
7888 | { | |
7889 | u64 cfs_period_us; | |
7890 | ||
029632fb | 7891 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7892 | do_div(cfs_period_us, NSEC_PER_USEC); |
7893 | ||
7894 | return cfs_period_us; | |
7895 | } | |
7896 | ||
7897 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
7898 | { | |
7899 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
7900 | } | |
7901 | ||
7902 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
7903 | s64 cfs_quota_us) | |
7904 | { | |
7905 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
7906 | } | |
7907 | ||
7908 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
7909 | { | |
7910 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
7911 | } | |
7912 | ||
7913 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
7914 | u64 cfs_period_us) | |
7915 | { | |
7916 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
7917 | } | |
7918 | ||
a790de99 PT |
7919 | struct cfs_schedulable_data { |
7920 | struct task_group *tg; | |
7921 | u64 period, quota; | |
7922 | }; | |
7923 | ||
7924 | /* | |
7925 | * normalize group quota/period to be quota/max_period | |
7926 | * note: units are usecs | |
7927 | */ | |
7928 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7929 | struct cfs_schedulable_data *d) | |
7930 | { | |
7931 | u64 quota, period; | |
7932 | ||
7933 | if (tg == d->tg) { | |
7934 | period = d->period; | |
7935 | quota = d->quota; | |
7936 | } else { | |
7937 | period = tg_get_cfs_period(tg); | |
7938 | quota = tg_get_cfs_quota(tg); | |
7939 | } | |
7940 | ||
7941 | /* note: these should typically be equivalent */ | |
7942 | if (quota == RUNTIME_INF || quota == -1) | |
7943 | return RUNTIME_INF; | |
7944 | ||
7945 | return to_ratio(period, quota); | |
7946 | } | |
7947 | ||
7948 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7949 | { | |
7950 | struct cfs_schedulable_data *d = data; | |
029632fb | 7951 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7952 | s64 quota = 0, parent_quota = -1; |
7953 | ||
7954 | if (!tg->parent) { | |
7955 | quota = RUNTIME_INF; | |
7956 | } else { | |
029632fb | 7957 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7958 | |
7959 | quota = normalize_cfs_quota(tg, d); | |
7960 | parent_quota = parent_b->hierarchal_quota; | |
7961 | ||
7962 | /* | |
7963 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7964 | * limit is set | |
7965 | */ | |
7966 | if (quota == RUNTIME_INF) | |
7967 | quota = parent_quota; | |
7968 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7969 | return -EINVAL; | |
7970 | } | |
7971 | cfs_b->hierarchal_quota = quota; | |
7972 | ||
7973 | return 0; | |
7974 | } | |
7975 | ||
7976 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7977 | { | |
8277434e | 7978 | int ret; |
a790de99 PT |
7979 | struct cfs_schedulable_data data = { |
7980 | .tg = tg, | |
7981 | .period = period, | |
7982 | .quota = quota, | |
7983 | }; | |
7984 | ||
7985 | if (quota != RUNTIME_INF) { | |
7986 | do_div(data.period, NSEC_PER_USEC); | |
7987 | do_div(data.quota, NSEC_PER_USEC); | |
7988 | } | |
7989 | ||
8277434e PT |
7990 | rcu_read_lock(); |
7991 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7992 | rcu_read_unlock(); | |
7993 | ||
7994 | return ret; | |
a790de99 | 7995 | } |
e8da1b18 NR |
7996 | |
7997 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
7998 | struct cgroup_map_cb *cb) | |
7999 | { | |
8000 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 8001 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
8002 | |
8003 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
8004 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
8005 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
8006 | ||
8007 | return 0; | |
8008 | } | |
ab84d31e | 8009 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8010 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8011 | |
052f1dc7 | 8012 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8013 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8014 | s64 val) |
6f505b16 | 8015 | { |
06ecb27c | 8016 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8017 | } |
8018 | ||
06ecb27c | 8019 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8020 | { |
06ecb27c | 8021 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8022 | } |
d0b27fa7 PZ |
8023 | |
8024 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8025 | u64 rt_period_us) | |
8026 | { | |
8027 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8028 | } | |
8029 | ||
8030 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8031 | { | |
8032 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8033 | } | |
6d6bc0ad | 8034 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8035 | |
fe5c7cc2 | 8036 | static struct cftype cpu_files[] = { |
052f1dc7 | 8037 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8038 | { |
8039 | .name = "shares", | |
f4c753b7 PM |
8040 | .read_u64 = cpu_shares_read_u64, |
8041 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8042 | }, |
052f1dc7 | 8043 | #endif |
ab84d31e PT |
8044 | #ifdef CONFIG_CFS_BANDWIDTH |
8045 | { | |
8046 | .name = "cfs_quota_us", | |
8047 | .read_s64 = cpu_cfs_quota_read_s64, | |
8048 | .write_s64 = cpu_cfs_quota_write_s64, | |
8049 | }, | |
8050 | { | |
8051 | .name = "cfs_period_us", | |
8052 | .read_u64 = cpu_cfs_period_read_u64, | |
8053 | .write_u64 = cpu_cfs_period_write_u64, | |
8054 | }, | |
e8da1b18 NR |
8055 | { |
8056 | .name = "stat", | |
8057 | .read_map = cpu_stats_show, | |
8058 | }, | |
ab84d31e | 8059 | #endif |
052f1dc7 | 8060 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8061 | { |
9f0c1e56 | 8062 | .name = "rt_runtime_us", |
06ecb27c PM |
8063 | .read_s64 = cpu_rt_runtime_read, |
8064 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8065 | }, |
d0b27fa7 PZ |
8066 | { |
8067 | .name = "rt_period_us", | |
f4c753b7 PM |
8068 | .read_u64 = cpu_rt_period_read_uint, |
8069 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8070 | }, |
052f1dc7 | 8071 | #endif |
68318b8e SV |
8072 | }; |
8073 | ||
8074 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8075 | { | |
fe5c7cc2 | 8076 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8077 | } |
8078 | ||
8079 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8080 | .name = "cpu", |
8081 | .create = cpu_cgroup_create, | |
8082 | .destroy = cpu_cgroup_destroy, | |
bb9d97b6 TH |
8083 | .can_attach = cpu_cgroup_can_attach, |
8084 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8085 | .exit = cpu_cgroup_exit, |
38605cae IM |
8086 | .populate = cpu_cgroup_populate, |
8087 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8088 | .early_init = 1, |
8089 | }; | |
8090 | ||
052f1dc7 | 8091 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8092 | |
8093 | #ifdef CONFIG_CGROUP_CPUACCT | |
8094 | ||
8095 | /* | |
8096 | * CPU accounting code for task groups. | |
8097 | * | |
8098 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8099 | * (balbir@in.ibm.com). | |
8100 | */ | |
8101 | ||
d842de87 | 8102 | /* create a new cpu accounting group */ |
761b3ef5 | 8103 | static struct cgroup_subsys_state *cpuacct_create(struct cgroup *cgrp) |
d842de87 | 8104 | { |
54c707e9 | 8105 | struct cpuacct *ca; |
d842de87 | 8106 | |
54c707e9 GC |
8107 | if (!cgrp->parent) |
8108 | return &root_cpuacct.css; | |
8109 | ||
8110 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 8111 | if (!ca) |
ef12fefa | 8112 | goto out; |
d842de87 SV |
8113 | |
8114 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8115 | if (!ca->cpuusage) |
8116 | goto out_free_ca; | |
8117 | ||
54c707e9 GC |
8118 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
8119 | if (!ca->cpustat) | |
8120 | goto out_free_cpuusage; | |
934352f2 | 8121 | |
d842de87 | 8122 | return &ca->css; |
ef12fefa | 8123 | |
54c707e9 | 8124 | out_free_cpuusage: |
ef12fefa BR |
8125 | free_percpu(ca->cpuusage); |
8126 | out_free_ca: | |
8127 | kfree(ca); | |
8128 | out: | |
8129 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8130 | } |
8131 | ||
8132 | /* destroy an existing cpu accounting group */ | |
761b3ef5 | 8133 | static void cpuacct_destroy(struct cgroup *cgrp) |
d842de87 | 8134 | { |
32cd756a | 8135 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 8136 | |
54c707e9 | 8137 | free_percpu(ca->cpustat); |
d842de87 SV |
8138 | free_percpu(ca->cpuusage); |
8139 | kfree(ca); | |
8140 | } | |
8141 | ||
720f5498 KC |
8142 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8143 | { | |
b36128c8 | 8144 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8145 | u64 data; |
8146 | ||
8147 | #ifndef CONFIG_64BIT | |
8148 | /* | |
8149 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8150 | */ | |
05fa785c | 8151 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8152 | data = *cpuusage; |
05fa785c | 8153 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8154 | #else |
8155 | data = *cpuusage; | |
8156 | #endif | |
8157 | ||
8158 | return data; | |
8159 | } | |
8160 | ||
8161 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8162 | { | |
b36128c8 | 8163 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8164 | |
8165 | #ifndef CONFIG_64BIT | |
8166 | /* | |
8167 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8168 | */ | |
05fa785c | 8169 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8170 | *cpuusage = val; |
05fa785c | 8171 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8172 | #else |
8173 | *cpuusage = val; | |
8174 | #endif | |
8175 | } | |
8176 | ||
d842de87 | 8177 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8178 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8179 | { |
32cd756a | 8180 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8181 | u64 totalcpuusage = 0; |
8182 | int i; | |
8183 | ||
720f5498 KC |
8184 | for_each_present_cpu(i) |
8185 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8186 | |
8187 | return totalcpuusage; | |
8188 | } | |
8189 | ||
0297b803 DG |
8190 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8191 | u64 reset) | |
8192 | { | |
8193 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8194 | int err = 0; | |
8195 | int i; | |
8196 | ||
8197 | if (reset) { | |
8198 | err = -EINVAL; | |
8199 | goto out; | |
8200 | } | |
8201 | ||
720f5498 KC |
8202 | for_each_present_cpu(i) |
8203 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8204 | |
0297b803 DG |
8205 | out: |
8206 | return err; | |
8207 | } | |
8208 | ||
e9515c3c KC |
8209 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8210 | struct seq_file *m) | |
8211 | { | |
8212 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8213 | u64 percpu; | |
8214 | int i; | |
8215 | ||
8216 | for_each_present_cpu(i) { | |
8217 | percpu = cpuacct_cpuusage_read(ca, i); | |
8218 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8219 | } | |
8220 | seq_printf(m, "\n"); | |
8221 | return 0; | |
8222 | } | |
8223 | ||
ef12fefa BR |
8224 | static const char *cpuacct_stat_desc[] = { |
8225 | [CPUACCT_STAT_USER] = "user", | |
8226 | [CPUACCT_STAT_SYSTEM] = "system", | |
8227 | }; | |
8228 | ||
8229 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8230 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8231 | { |
8232 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8233 | int cpu; |
8234 | s64 val = 0; | |
ef12fefa | 8235 | |
54c707e9 GC |
8236 | for_each_online_cpu(cpu) { |
8237 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8238 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8239 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8240 | } |
54c707e9 GC |
8241 | val = cputime64_to_clock_t(val); |
8242 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8243 | |
54c707e9 GC |
8244 | val = 0; |
8245 | for_each_online_cpu(cpu) { | |
8246 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8247 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8248 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8249 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8250 | } |
54c707e9 GC |
8251 | |
8252 | val = cputime64_to_clock_t(val); | |
8253 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8254 | ||
ef12fefa BR |
8255 | return 0; |
8256 | } | |
8257 | ||
d842de87 SV |
8258 | static struct cftype files[] = { |
8259 | { | |
8260 | .name = "usage", | |
f4c753b7 PM |
8261 | .read_u64 = cpuusage_read, |
8262 | .write_u64 = cpuusage_write, | |
d842de87 | 8263 | }, |
e9515c3c KC |
8264 | { |
8265 | .name = "usage_percpu", | |
8266 | .read_seq_string = cpuacct_percpu_seq_read, | |
8267 | }, | |
ef12fefa BR |
8268 | { |
8269 | .name = "stat", | |
8270 | .read_map = cpuacct_stats_show, | |
8271 | }, | |
d842de87 SV |
8272 | }; |
8273 | ||
32cd756a | 8274 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8275 | { |
32cd756a | 8276 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8277 | } |
8278 | ||
8279 | /* | |
8280 | * charge this task's execution time to its accounting group. | |
8281 | * | |
8282 | * called with rq->lock held. | |
8283 | */ | |
029632fb | 8284 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8285 | { |
8286 | struct cpuacct *ca; | |
934352f2 | 8287 | int cpu; |
d842de87 | 8288 | |
c40c6f85 | 8289 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8290 | return; |
8291 | ||
934352f2 | 8292 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8293 | |
8294 | rcu_read_lock(); | |
8295 | ||
d842de87 | 8296 | ca = task_ca(tsk); |
d842de87 | 8297 | |
44252e42 | 8298 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8299 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8300 | *cpuusage += cputime; |
8301 | } | |
a18b83b7 BR |
8302 | |
8303 | rcu_read_unlock(); | |
d842de87 SV |
8304 | } |
8305 | ||
8306 | struct cgroup_subsys cpuacct_subsys = { | |
8307 | .name = "cpuacct", | |
8308 | .create = cpuacct_create, | |
8309 | .destroy = cpuacct_destroy, | |
8310 | .populate = cpuacct_populate, | |
8311 | .subsys_id = cpuacct_subsys_id, | |
8312 | }; | |
8313 | #endif /* CONFIG_CGROUP_CPUACCT */ |