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" |
29d5e047 | 86 | #include "../smpboot.h" |
6e0534f2 | 87 | |
a8d154b0 | 88 | #define CREATE_TRACE_POINTS |
ad8d75ff | 89 | #include <trace/events/sched.h> |
a8d154b0 | 90 | |
029632fb | 91 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 92 | { |
58088ad0 PT |
93 | unsigned long delta; |
94 | ktime_t soft, hard, now; | |
d0b27fa7 | 95 | |
58088ad0 PT |
96 | for (;;) { |
97 | if (hrtimer_active(period_timer)) | |
98 | break; | |
99 | ||
100 | now = hrtimer_cb_get_time(period_timer); | |
101 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 102 | |
58088ad0 PT |
103 | soft = hrtimer_get_softexpires(period_timer); |
104 | hard = hrtimer_get_expires(period_timer); | |
105 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
106 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
107 | HRTIMER_MODE_ABS_PINNED, 0); | |
108 | } | |
109 | } | |
110 | ||
029632fb PZ |
111 | DEFINE_MUTEX(sched_domains_mutex); |
112 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 113 | |
fe44d621 | 114 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 115 | |
029632fb | 116 | void update_rq_clock(struct rq *rq) |
3e51f33f | 117 | { |
fe44d621 | 118 | s64 delta; |
305e6835 | 119 | |
61eadef6 | 120 | if (rq->skip_clock_update > 0) |
f26f9aff | 121 | return; |
aa483808 | 122 | |
fe44d621 PZ |
123 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
124 | rq->clock += delta; | |
125 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
126 | } |
127 | ||
bf5c91ba IM |
128 | /* |
129 | * Debugging: various feature bits | |
130 | */ | |
f00b45c1 | 131 | |
f00b45c1 PZ |
132 | #define SCHED_FEAT(name, enabled) \ |
133 | (1UL << __SCHED_FEAT_##name) * enabled | | |
134 | ||
bf5c91ba | 135 | const_debug unsigned int sysctl_sched_features = |
391e43da | 136 | #include "features.h" |
f00b45c1 PZ |
137 | 0; |
138 | ||
139 | #undef SCHED_FEAT | |
140 | ||
141 | #ifdef CONFIG_SCHED_DEBUG | |
142 | #define SCHED_FEAT(name, enabled) \ | |
143 | #name , | |
144 | ||
1292531f | 145 | static const char * const sched_feat_names[] = { |
391e43da | 146 | #include "features.h" |
f00b45c1 PZ |
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 | ||
45bf76df IM |
695 | static void set_load_weight(struct task_struct *p) |
696 | { | |
f05998d4 NR |
697 | int prio = p->static_prio - MAX_RT_PRIO; |
698 | struct load_weight *load = &p->se.load; | |
699 | ||
dd41f596 IM |
700 | /* |
701 | * SCHED_IDLE tasks get minimal weight: | |
702 | */ | |
703 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 704 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 705 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
706 | return; |
707 | } | |
71f8bd46 | 708 | |
c8b28116 | 709 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 710 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
711 | } |
712 | ||
371fd7e7 | 713 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 714 | { |
a64692a3 | 715 | update_rq_clock(rq); |
dd41f596 | 716 | sched_info_queued(p); |
371fd7e7 | 717 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
718 | } |
719 | ||
371fd7e7 | 720 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 721 | { |
a64692a3 | 722 | update_rq_clock(rq); |
46ac22ba | 723 | sched_info_dequeued(p); |
371fd7e7 | 724 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
725 | } |
726 | ||
029632fb | 727 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
728 | { |
729 | if (task_contributes_to_load(p)) | |
730 | rq->nr_uninterruptible--; | |
731 | ||
371fd7e7 | 732 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
733 | } |
734 | ||
029632fb | 735 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
736 | { |
737 | if (task_contributes_to_load(p)) | |
738 | rq->nr_uninterruptible++; | |
739 | ||
371fd7e7 | 740 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
741 | } |
742 | ||
b52bfee4 VP |
743 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
744 | ||
305e6835 VP |
745 | /* |
746 | * There are no locks covering percpu hardirq/softirq time. | |
747 | * They are only modified in account_system_vtime, on corresponding CPU | |
748 | * with interrupts disabled. So, writes are safe. | |
749 | * They are read and saved off onto struct rq in update_rq_clock(). | |
750 | * This may result in other CPU reading this CPU's irq time and can | |
751 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
752 | * or new value with a side effect of accounting a slice of irq time to wrong |
753 | * task when irq is in progress while we read rq->clock. That is a worthy | |
754 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 755 | */ |
b52bfee4 VP |
756 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
757 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
758 | ||
759 | static DEFINE_PER_CPU(u64, irq_start_time); | |
760 | static int sched_clock_irqtime; | |
761 | ||
762 | void enable_sched_clock_irqtime(void) | |
763 | { | |
764 | sched_clock_irqtime = 1; | |
765 | } | |
766 | ||
767 | void disable_sched_clock_irqtime(void) | |
768 | { | |
769 | sched_clock_irqtime = 0; | |
770 | } | |
771 | ||
8e92c201 PZ |
772 | #ifndef CONFIG_64BIT |
773 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
774 | ||
775 | static inline void irq_time_write_begin(void) | |
776 | { | |
777 | __this_cpu_inc(irq_time_seq.sequence); | |
778 | smp_wmb(); | |
779 | } | |
780 | ||
781 | static inline void irq_time_write_end(void) | |
782 | { | |
783 | smp_wmb(); | |
784 | __this_cpu_inc(irq_time_seq.sequence); | |
785 | } | |
786 | ||
787 | static inline u64 irq_time_read(int cpu) | |
788 | { | |
789 | u64 irq_time; | |
790 | unsigned seq; | |
791 | ||
792 | do { | |
793 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
794 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
795 | per_cpu(cpu_hardirq_time, cpu); | |
796 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
797 | ||
798 | return irq_time; | |
799 | } | |
800 | #else /* CONFIG_64BIT */ | |
801 | static inline void irq_time_write_begin(void) | |
802 | { | |
803 | } | |
804 | ||
805 | static inline void irq_time_write_end(void) | |
806 | { | |
807 | } | |
808 | ||
809 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 810 | { |
305e6835 VP |
811 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
812 | } | |
8e92c201 | 813 | #endif /* CONFIG_64BIT */ |
305e6835 | 814 | |
fe44d621 PZ |
815 | /* |
816 | * Called before incrementing preempt_count on {soft,}irq_enter | |
817 | * and before decrementing preempt_count on {soft,}irq_exit. | |
818 | */ | |
b52bfee4 VP |
819 | void account_system_vtime(struct task_struct *curr) |
820 | { | |
821 | unsigned long flags; | |
fe44d621 | 822 | s64 delta; |
b52bfee4 | 823 | int cpu; |
b52bfee4 VP |
824 | |
825 | if (!sched_clock_irqtime) | |
826 | return; | |
827 | ||
828 | local_irq_save(flags); | |
829 | ||
b52bfee4 | 830 | cpu = smp_processor_id(); |
fe44d621 PZ |
831 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
832 | __this_cpu_add(irq_start_time, delta); | |
833 | ||
8e92c201 | 834 | irq_time_write_begin(); |
b52bfee4 VP |
835 | /* |
836 | * We do not account for softirq time from ksoftirqd here. | |
837 | * We want to continue accounting softirq time to ksoftirqd thread | |
838 | * in that case, so as not to confuse scheduler with a special task | |
839 | * that do not consume any time, but still wants to run. | |
840 | */ | |
841 | if (hardirq_count()) | |
fe44d621 | 842 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 843 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 844 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 845 | |
8e92c201 | 846 | irq_time_write_end(); |
b52bfee4 VP |
847 | local_irq_restore(flags); |
848 | } | |
b7dadc38 | 849 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 850 | |
e6e6685a GC |
851 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
852 | ||
853 | #ifdef CONFIG_PARAVIRT | |
854 | static inline u64 steal_ticks(u64 steal) | |
aa483808 | 855 | { |
e6e6685a GC |
856 | if (unlikely(steal > NSEC_PER_SEC)) |
857 | return div_u64(steal, TICK_NSEC); | |
fe44d621 | 858 | |
e6e6685a GC |
859 | return __iter_div_u64_rem(steal, TICK_NSEC, &steal); |
860 | } | |
861 | #endif | |
862 | ||
fe44d621 | 863 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 864 | { |
095c0aa8 GC |
865 | /* |
866 | * In theory, the compile should just see 0 here, and optimize out the call | |
867 | * to sched_rt_avg_update. But I don't trust it... | |
868 | */ | |
869 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
870 | s64 steal = 0, irq_delta = 0; | |
871 | #endif | |
872 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 873 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
874 | |
875 | /* | |
876 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
877 | * this case when a previous update_rq_clock() happened inside a | |
878 | * {soft,}irq region. | |
879 | * | |
880 | * When this happens, we stop ->clock_task and only update the | |
881 | * prev_irq_time stamp to account for the part that fit, so that a next | |
882 | * update will consume the rest. This ensures ->clock_task is | |
883 | * monotonic. | |
884 | * | |
885 | * It does however cause some slight miss-attribution of {soft,}irq | |
886 | * time, a more accurate solution would be to update the irq_time using | |
887 | * the current rq->clock timestamp, except that would require using | |
888 | * atomic ops. | |
889 | */ | |
890 | if (irq_delta > delta) | |
891 | irq_delta = delta; | |
892 | ||
893 | rq->prev_irq_time += irq_delta; | |
894 | delta -= irq_delta; | |
095c0aa8 GC |
895 | #endif |
896 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 897 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
898 | u64 st; |
899 | ||
900 | steal = paravirt_steal_clock(cpu_of(rq)); | |
901 | steal -= rq->prev_steal_time_rq; | |
902 | ||
903 | if (unlikely(steal > delta)) | |
904 | steal = delta; | |
905 | ||
906 | st = steal_ticks(steal); | |
907 | steal = st * TICK_NSEC; | |
908 | ||
909 | rq->prev_steal_time_rq += steal; | |
910 | ||
911 | delta -= steal; | |
912 | } | |
913 | #endif | |
914 | ||
fe44d621 PZ |
915 | rq->clock_task += delta; |
916 | ||
095c0aa8 GC |
917 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
918 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
919 | sched_rt_avg_update(rq, irq_delta + steal); | |
920 | #endif | |
aa483808 VP |
921 | } |
922 | ||
095c0aa8 | 923 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
abb74cef VP |
924 | static int irqtime_account_hi_update(void) |
925 | { | |
3292beb3 | 926 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
927 | unsigned long flags; |
928 | u64 latest_ns; | |
929 | int ret = 0; | |
930 | ||
931 | local_irq_save(flags); | |
932 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
612ef28a | 933 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) |
abb74cef VP |
934 | ret = 1; |
935 | local_irq_restore(flags); | |
936 | return ret; | |
937 | } | |
938 | ||
939 | static int irqtime_account_si_update(void) | |
940 | { | |
3292beb3 | 941 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef VP |
942 | unsigned long flags; |
943 | u64 latest_ns; | |
944 | int ret = 0; | |
945 | ||
946 | local_irq_save(flags); | |
947 | latest_ns = this_cpu_read(cpu_softirq_time); | |
612ef28a | 948 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) |
abb74cef VP |
949 | ret = 1; |
950 | local_irq_restore(flags); | |
951 | return ret; | |
952 | } | |
953 | ||
fe44d621 | 954 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 955 | |
abb74cef VP |
956 | #define sched_clock_irqtime (0) |
957 | ||
095c0aa8 | 958 | #endif |
b52bfee4 | 959 | |
34f971f6 PZ |
960 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
961 | { | |
962 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
963 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
964 | ||
965 | if (stop) { | |
966 | /* | |
967 | * Make it appear like a SCHED_FIFO task, its something | |
968 | * userspace knows about and won't get confused about. | |
969 | * | |
970 | * Also, it will make PI more or less work without too | |
971 | * much confusion -- but then, stop work should not | |
972 | * rely on PI working anyway. | |
973 | */ | |
974 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
975 | ||
976 | stop->sched_class = &stop_sched_class; | |
977 | } | |
978 | ||
979 | cpu_rq(cpu)->stop = stop; | |
980 | ||
981 | if (old_stop) { | |
982 | /* | |
983 | * Reset it back to a normal scheduling class so that | |
984 | * it can die in pieces. | |
985 | */ | |
986 | old_stop->sched_class = &rt_sched_class; | |
987 | } | |
988 | } | |
989 | ||
14531189 | 990 | /* |
dd41f596 | 991 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 992 | */ |
14531189 IM |
993 | static inline int __normal_prio(struct task_struct *p) |
994 | { | |
dd41f596 | 995 | return p->static_prio; |
14531189 IM |
996 | } |
997 | ||
b29739f9 IM |
998 | /* |
999 | * Calculate the expected normal priority: i.e. priority | |
1000 | * without taking RT-inheritance into account. Might be | |
1001 | * boosted by interactivity modifiers. Changes upon fork, | |
1002 | * setprio syscalls, and whenever the interactivity | |
1003 | * estimator recalculates. | |
1004 | */ | |
36c8b586 | 1005 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1006 | { |
1007 | int prio; | |
1008 | ||
e05606d3 | 1009 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1010 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1011 | else | |
1012 | prio = __normal_prio(p); | |
1013 | return prio; | |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * Calculate the current priority, i.e. the priority | |
1018 | * taken into account by the scheduler. This value might | |
1019 | * be boosted by RT tasks, or might be boosted by | |
1020 | * interactivity modifiers. Will be RT if the task got | |
1021 | * RT-boosted. If not then it returns p->normal_prio. | |
1022 | */ | |
36c8b586 | 1023 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1024 | { |
1025 | p->normal_prio = normal_prio(p); | |
1026 | /* | |
1027 | * If we are RT tasks or we were boosted to RT priority, | |
1028 | * keep the priority unchanged. Otherwise, update priority | |
1029 | * to the normal priority: | |
1030 | */ | |
1031 | if (!rt_prio(p->prio)) | |
1032 | return p->normal_prio; | |
1033 | return p->prio; | |
1034 | } | |
1035 | ||
1da177e4 LT |
1036 | /** |
1037 | * task_curr - is this task currently executing on a CPU? | |
1038 | * @p: the task in question. | |
1039 | */ | |
36c8b586 | 1040 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1041 | { |
1042 | return cpu_curr(task_cpu(p)) == p; | |
1043 | } | |
1044 | ||
cb469845 SR |
1045 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1046 | const struct sched_class *prev_class, | |
da7a735e | 1047 | int oldprio) |
cb469845 SR |
1048 | { |
1049 | if (prev_class != p->sched_class) { | |
1050 | if (prev_class->switched_from) | |
da7a735e PZ |
1051 | prev_class->switched_from(rq, p); |
1052 | p->sched_class->switched_to(rq, p); | |
1053 | } else if (oldprio != p->prio) | |
1054 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
1055 | } |
1056 | ||
029632fb | 1057 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1058 | { |
1059 | const struct sched_class *class; | |
1060 | ||
1061 | if (p->sched_class == rq->curr->sched_class) { | |
1062 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1063 | } else { | |
1064 | for_each_class(class) { | |
1065 | if (class == rq->curr->sched_class) | |
1066 | break; | |
1067 | if (class == p->sched_class) { | |
1068 | resched_task(rq->curr); | |
1069 | break; | |
1070 | } | |
1071 | } | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * A queue event has occurred, and we're going to schedule. In | |
1076 | * this case, we can save a useless back to back clock update. | |
1077 | */ | |
fd2f4419 | 1078 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1079 | rq->skip_clock_update = 1; |
1080 | } | |
1081 | ||
1da177e4 | 1082 | #ifdef CONFIG_SMP |
dd41f596 | 1083 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1084 | { |
e2912009 PZ |
1085 | #ifdef CONFIG_SCHED_DEBUG |
1086 | /* | |
1087 | * We should never call set_task_cpu() on a blocked task, | |
1088 | * ttwu() will sort out the placement. | |
1089 | */ | |
077614ee PZ |
1090 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
1091 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
1092 | |
1093 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1094 | /* |
1095 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1096 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1097 | * | |
1098 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1099 | * see task_group(). |
6c6c54e1 PZ |
1100 | * |
1101 | * Furthermore, all task_rq users should acquire both locks, see | |
1102 | * task_rq_lock(). | |
1103 | */ | |
0122ec5b PZ |
1104 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1105 | lockdep_is_held(&task_rq(p)->lock))); | |
1106 | #endif | |
e2912009 PZ |
1107 | #endif |
1108 | ||
de1d7286 | 1109 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1110 | |
0c69774e PZ |
1111 | if (task_cpu(p) != new_cpu) { |
1112 | p->se.nr_migrations++; | |
a8b0ca17 | 1113 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1114 | } |
dd41f596 IM |
1115 | |
1116 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1117 | } |
1118 | ||
969c7921 | 1119 | struct migration_arg { |
36c8b586 | 1120 | struct task_struct *task; |
1da177e4 | 1121 | int dest_cpu; |
70b97a7f | 1122 | }; |
1da177e4 | 1123 | |
969c7921 TH |
1124 | static int migration_cpu_stop(void *data); |
1125 | ||
1da177e4 LT |
1126 | /* |
1127 | * wait_task_inactive - wait for a thread to unschedule. | |
1128 | * | |
85ba2d86 RM |
1129 | * If @match_state is nonzero, it's the @p->state value just checked and |
1130 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1131 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1132 | * we return a positive number (its total switch count). If a second call | |
1133 | * a short while later returns the same number, the caller can be sure that | |
1134 | * @p has remained unscheduled the whole time. | |
1135 | * | |
1da177e4 LT |
1136 | * The caller must ensure that the task *will* unschedule sometime soon, |
1137 | * else this function might spin for a *long* time. This function can't | |
1138 | * be called with interrupts off, or it may introduce deadlock with | |
1139 | * smp_call_function() if an IPI is sent by the same process we are | |
1140 | * waiting to become inactive. | |
1141 | */ | |
85ba2d86 | 1142 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1143 | { |
1144 | unsigned long flags; | |
dd41f596 | 1145 | int running, on_rq; |
85ba2d86 | 1146 | unsigned long ncsw; |
70b97a7f | 1147 | struct rq *rq; |
1da177e4 | 1148 | |
3a5c359a AK |
1149 | for (;;) { |
1150 | /* | |
1151 | * We do the initial early heuristics without holding | |
1152 | * any task-queue locks at all. We'll only try to get | |
1153 | * the runqueue lock when things look like they will | |
1154 | * work out! | |
1155 | */ | |
1156 | rq = task_rq(p); | |
fa490cfd | 1157 | |
3a5c359a AK |
1158 | /* |
1159 | * If the task is actively running on another CPU | |
1160 | * still, just relax and busy-wait without holding | |
1161 | * any locks. | |
1162 | * | |
1163 | * NOTE! Since we don't hold any locks, it's not | |
1164 | * even sure that "rq" stays as the right runqueue! | |
1165 | * But we don't care, since "task_running()" will | |
1166 | * return false if the runqueue has changed and p | |
1167 | * is actually now running somewhere else! | |
1168 | */ | |
85ba2d86 RM |
1169 | while (task_running(rq, p)) { |
1170 | if (match_state && unlikely(p->state != match_state)) | |
1171 | return 0; | |
3a5c359a | 1172 | cpu_relax(); |
85ba2d86 | 1173 | } |
fa490cfd | 1174 | |
3a5c359a AK |
1175 | /* |
1176 | * Ok, time to look more closely! We need the rq | |
1177 | * lock now, to be *sure*. If we're wrong, we'll | |
1178 | * just go back and repeat. | |
1179 | */ | |
1180 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1181 | trace_sched_wait_task(p); |
3a5c359a | 1182 | running = task_running(rq, p); |
fd2f4419 | 1183 | on_rq = p->on_rq; |
85ba2d86 | 1184 | ncsw = 0; |
f31e11d8 | 1185 | if (!match_state || p->state == match_state) |
93dcf55f | 1186 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1187 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1188 | |
85ba2d86 RM |
1189 | /* |
1190 | * If it changed from the expected state, bail out now. | |
1191 | */ | |
1192 | if (unlikely(!ncsw)) | |
1193 | break; | |
1194 | ||
3a5c359a AK |
1195 | /* |
1196 | * Was it really running after all now that we | |
1197 | * checked with the proper locks actually held? | |
1198 | * | |
1199 | * Oops. Go back and try again.. | |
1200 | */ | |
1201 | if (unlikely(running)) { | |
1202 | cpu_relax(); | |
1203 | continue; | |
1204 | } | |
fa490cfd | 1205 | |
3a5c359a AK |
1206 | /* |
1207 | * It's not enough that it's not actively running, | |
1208 | * it must be off the runqueue _entirely_, and not | |
1209 | * preempted! | |
1210 | * | |
80dd99b3 | 1211 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1212 | * running right now), it's preempted, and we should |
1213 | * yield - it could be a while. | |
1214 | */ | |
1215 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1216 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1217 | ||
1218 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1219 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1220 | continue; |
1221 | } | |
fa490cfd | 1222 | |
3a5c359a AK |
1223 | /* |
1224 | * Ahh, all good. It wasn't running, and it wasn't | |
1225 | * runnable, which means that it will never become | |
1226 | * running in the future either. We're all done! | |
1227 | */ | |
1228 | break; | |
1229 | } | |
85ba2d86 RM |
1230 | |
1231 | return ncsw; | |
1da177e4 LT |
1232 | } |
1233 | ||
1234 | /*** | |
1235 | * kick_process - kick a running thread to enter/exit the kernel | |
1236 | * @p: the to-be-kicked thread | |
1237 | * | |
1238 | * Cause a process which is running on another CPU to enter | |
1239 | * kernel-mode, without any delay. (to get signals handled.) | |
1240 | * | |
25985edc | 1241 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1242 | * because all it wants to ensure is that the remote task enters |
1243 | * the kernel. If the IPI races and the task has been migrated | |
1244 | * to another CPU then no harm is done and the purpose has been | |
1245 | * achieved as well. | |
1246 | */ | |
36c8b586 | 1247 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1248 | { |
1249 | int cpu; | |
1250 | ||
1251 | preempt_disable(); | |
1252 | cpu = task_cpu(p); | |
1253 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1254 | smp_send_reschedule(cpu); | |
1255 | preempt_enable(); | |
1256 | } | |
b43e3521 | 1257 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1258 | #endif /* CONFIG_SMP */ |
1da177e4 | 1259 | |
970b13ba | 1260 | #ifdef CONFIG_SMP |
30da688e | 1261 | /* |
013fdb80 | 1262 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1263 | */ |
5da9a0fb PZ |
1264 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1265 | { | |
5da9a0fb | 1266 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); |
2baab4e9 PZ |
1267 | enum { cpuset, possible, fail } state = cpuset; |
1268 | int dest_cpu; | |
5da9a0fb PZ |
1269 | |
1270 | /* Look for allowed, online CPU in same node. */ | |
e3831edd | 1271 | for_each_cpu(dest_cpu, nodemask) { |
2baab4e9 PZ |
1272 | if (!cpu_online(dest_cpu)) |
1273 | continue; | |
1274 | if (!cpu_active(dest_cpu)) | |
1275 | continue; | |
fa17b507 | 1276 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
5da9a0fb | 1277 | return dest_cpu; |
2baab4e9 | 1278 | } |
5da9a0fb | 1279 | |
2baab4e9 PZ |
1280 | for (;;) { |
1281 | /* Any allowed, online CPU? */ | |
e3831edd | 1282 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1283 | if (!cpu_online(dest_cpu)) |
1284 | continue; | |
1285 | if (!cpu_active(dest_cpu)) | |
1286 | continue; | |
1287 | goto out; | |
1288 | } | |
5da9a0fb | 1289 | |
2baab4e9 PZ |
1290 | switch (state) { |
1291 | case cpuset: | |
1292 | /* No more Mr. Nice Guy. */ | |
1293 | cpuset_cpus_allowed_fallback(p); | |
1294 | state = possible; | |
1295 | break; | |
1296 | ||
1297 | case possible: | |
1298 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1299 | state = fail; | |
1300 | break; | |
1301 | ||
1302 | case fail: | |
1303 | BUG(); | |
1304 | break; | |
1305 | } | |
1306 | } | |
1307 | ||
1308 | out: | |
1309 | if (state != cpuset) { | |
1310 | /* | |
1311 | * Don't tell them about moving exiting tasks or | |
1312 | * kernel threads (both mm NULL), since they never | |
1313 | * leave kernel. | |
1314 | */ | |
1315 | if (p->mm && printk_ratelimit()) { | |
1316 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1317 | task_pid_nr(p), p->comm, cpu); | |
1318 | } | |
5da9a0fb PZ |
1319 | } |
1320 | ||
1321 | return dest_cpu; | |
1322 | } | |
1323 | ||
e2912009 | 1324 | /* |
013fdb80 | 1325 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1326 | */ |
970b13ba | 1327 | static inline |
7608dec2 | 1328 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 1329 | { |
7608dec2 | 1330 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
1331 | |
1332 | /* | |
1333 | * In order not to call set_task_cpu() on a blocking task we need | |
1334 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1335 | * cpu. | |
1336 | * | |
1337 | * Since this is common to all placement strategies, this lives here. | |
1338 | * | |
1339 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1340 | * not worry about this generic constraint ] | |
1341 | */ | |
fa17b507 | 1342 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1343 | !cpu_online(cpu))) |
5da9a0fb | 1344 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1345 | |
1346 | return cpu; | |
970b13ba | 1347 | } |
09a40af5 MG |
1348 | |
1349 | static void update_avg(u64 *avg, u64 sample) | |
1350 | { | |
1351 | s64 diff = sample - *avg; | |
1352 | *avg += diff >> 3; | |
1353 | } | |
970b13ba PZ |
1354 | #endif |
1355 | ||
d7c01d27 | 1356 | static void |
b84cb5df | 1357 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1358 | { |
d7c01d27 | 1359 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1360 | struct rq *rq = this_rq(); |
1361 | ||
d7c01d27 PZ |
1362 | #ifdef CONFIG_SMP |
1363 | int this_cpu = smp_processor_id(); | |
1364 | ||
1365 | if (cpu == this_cpu) { | |
1366 | schedstat_inc(rq, ttwu_local); | |
1367 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1368 | } else { | |
1369 | struct sched_domain *sd; | |
1370 | ||
1371 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1372 | rcu_read_lock(); |
d7c01d27 PZ |
1373 | for_each_domain(this_cpu, sd) { |
1374 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1375 | schedstat_inc(sd, ttwu_wake_remote); | |
1376 | break; | |
1377 | } | |
1378 | } | |
057f3fad | 1379 | rcu_read_unlock(); |
d7c01d27 | 1380 | } |
f339b9dc PZ |
1381 | |
1382 | if (wake_flags & WF_MIGRATED) | |
1383 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1384 | ||
d7c01d27 PZ |
1385 | #endif /* CONFIG_SMP */ |
1386 | ||
1387 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1388 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1389 | |
1390 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1391 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1392 | |
d7c01d27 PZ |
1393 | #endif /* CONFIG_SCHEDSTATS */ |
1394 | } | |
1395 | ||
1396 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1397 | { | |
9ed3811a | 1398 | activate_task(rq, p, en_flags); |
fd2f4419 | 1399 | p->on_rq = 1; |
c2f7115e PZ |
1400 | |
1401 | /* if a worker is waking up, notify workqueue */ | |
1402 | if (p->flags & PF_WQ_WORKER) | |
1403 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1404 | } |
1405 | ||
23f41eeb PZ |
1406 | /* |
1407 | * Mark the task runnable and perform wakeup-preemption. | |
1408 | */ | |
89363381 | 1409 | static void |
23f41eeb | 1410 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1411 | { |
89363381 | 1412 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1413 | check_preempt_curr(rq, p, wake_flags); |
1414 | ||
1415 | p->state = TASK_RUNNING; | |
1416 | #ifdef CONFIG_SMP | |
1417 | if (p->sched_class->task_woken) | |
1418 | p->sched_class->task_woken(rq, p); | |
1419 | ||
e69c6341 | 1420 | if (rq->idle_stamp) { |
9ed3811a TH |
1421 | u64 delta = rq->clock - rq->idle_stamp; |
1422 | u64 max = 2*sysctl_sched_migration_cost; | |
1423 | ||
1424 | if (delta > max) | |
1425 | rq->avg_idle = max; | |
1426 | else | |
1427 | update_avg(&rq->avg_idle, delta); | |
1428 | rq->idle_stamp = 0; | |
1429 | } | |
1430 | #endif | |
1431 | } | |
1432 | ||
c05fbafb PZ |
1433 | static void |
1434 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1435 | { | |
1436 | #ifdef CONFIG_SMP | |
1437 | if (p->sched_contributes_to_load) | |
1438 | rq->nr_uninterruptible--; | |
1439 | #endif | |
1440 | ||
1441 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1442 | ttwu_do_wakeup(rq, p, wake_flags); | |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1447 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1448 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1449 | * the task is still ->on_rq. | |
1450 | */ | |
1451 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1452 | { | |
1453 | struct rq *rq; | |
1454 | int ret = 0; | |
1455 | ||
1456 | rq = __task_rq_lock(p); | |
1457 | if (p->on_rq) { | |
1458 | ttwu_do_wakeup(rq, p, wake_flags); | |
1459 | ret = 1; | |
1460 | } | |
1461 | __task_rq_unlock(rq); | |
1462 | ||
1463 | return ret; | |
1464 | } | |
1465 | ||
317f3941 | 1466 | #ifdef CONFIG_SMP |
fa14ff4a | 1467 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1468 | { |
1469 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1470 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1471 | struct task_struct *p; | |
317f3941 PZ |
1472 | |
1473 | raw_spin_lock(&rq->lock); | |
1474 | ||
fa14ff4a PZ |
1475 | while (llist) { |
1476 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1477 | llist = llist_next(llist); | |
317f3941 PZ |
1478 | ttwu_do_activate(rq, p, 0); |
1479 | } | |
1480 | ||
1481 | raw_spin_unlock(&rq->lock); | |
1482 | } | |
1483 | ||
1484 | void scheduler_ipi(void) | |
1485 | { | |
ca38062e | 1486 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1487 | return; |
1488 | ||
1489 | /* | |
1490 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1491 | * traditionally all their work was done from the interrupt return | |
1492 | * path. Now that we actually do some work, we need to make sure | |
1493 | * we do call them. | |
1494 | * | |
1495 | * Some archs already do call them, luckily irq_enter/exit nest | |
1496 | * properly. | |
1497 | * | |
1498 | * Arguably we should visit all archs and update all handlers, | |
1499 | * however a fair share of IPIs are still resched only so this would | |
1500 | * somewhat pessimize the simple resched case. | |
1501 | */ | |
1502 | irq_enter(); | |
fa14ff4a | 1503 | sched_ttwu_pending(); |
ca38062e SS |
1504 | |
1505 | /* | |
1506 | * Check if someone kicked us for doing the nohz idle load balance. | |
1507 | */ | |
6eb57e0d SS |
1508 | if (unlikely(got_nohz_idle_kick() && !need_resched())) { |
1509 | this_rq()->idle_balance = 1; | |
ca38062e | 1510 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1511 | } |
c5d753a5 | 1512 | irq_exit(); |
317f3941 PZ |
1513 | } |
1514 | ||
1515 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1516 | { | |
fa14ff4a | 1517 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1518 | smp_send_reschedule(cpu); |
1519 | } | |
d6aa8f85 PZ |
1520 | |
1521 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1522 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
1523 | { | |
1524 | struct rq *rq; | |
1525 | int ret = 0; | |
1526 | ||
1527 | rq = __task_rq_lock(p); | |
1528 | if (p->on_cpu) { | |
1529 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
1530 | ttwu_do_wakeup(rq, p, wake_flags); | |
1531 | ret = 1; | |
1532 | } | |
1533 | __task_rq_unlock(rq); | |
1534 | ||
1535 | return ret; | |
1536 | ||
1537 | } | |
1538 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
518cd623 | 1539 | |
39be3501 | 1540 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1541 | { |
1542 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1543 | } | |
d6aa8f85 | 1544 | #endif /* CONFIG_SMP */ |
317f3941 | 1545 | |
c05fbafb PZ |
1546 | static void ttwu_queue(struct task_struct *p, int cpu) |
1547 | { | |
1548 | struct rq *rq = cpu_rq(cpu); | |
1549 | ||
17d9f311 | 1550 | #if defined(CONFIG_SMP) |
39be3501 | 1551 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1552 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1553 | ttwu_queue_remote(p, cpu); |
1554 | return; | |
1555 | } | |
1556 | #endif | |
1557 | ||
c05fbafb PZ |
1558 | raw_spin_lock(&rq->lock); |
1559 | ttwu_do_activate(rq, p, 0); | |
1560 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1561 | } |
1562 | ||
1563 | /** | |
1da177e4 | 1564 | * try_to_wake_up - wake up a thread |
9ed3811a | 1565 | * @p: the thread to be awakened |
1da177e4 | 1566 | * @state: the mask of task states that can be woken |
9ed3811a | 1567 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1568 | * |
1569 | * Put it on the run-queue if it's not already there. The "current" | |
1570 | * thread is always on the run-queue (except when the actual | |
1571 | * re-schedule is in progress), and as such you're allowed to do | |
1572 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1573 | * runnable without the overhead of this. | |
1574 | * | |
9ed3811a TH |
1575 | * Returns %true if @p was woken up, %false if it was already running |
1576 | * or @state didn't match @p's state. | |
1da177e4 | 1577 | */ |
e4a52bcb PZ |
1578 | static int |
1579 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1580 | { |
1da177e4 | 1581 | unsigned long flags; |
c05fbafb | 1582 | int cpu, success = 0; |
2398f2c6 | 1583 | |
04e2f174 | 1584 | smp_wmb(); |
013fdb80 | 1585 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1586 | if (!(p->state & state)) |
1da177e4 LT |
1587 | goto out; |
1588 | ||
c05fbafb | 1589 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1590 | cpu = task_cpu(p); |
1da177e4 | 1591 | |
c05fbafb PZ |
1592 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1593 | goto stat; | |
1da177e4 | 1594 | |
1da177e4 | 1595 | #ifdef CONFIG_SMP |
e9c84311 | 1596 | /* |
c05fbafb PZ |
1597 | * If the owning (remote) cpu is still in the middle of schedule() with |
1598 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1599 | */ |
e4a52bcb PZ |
1600 | while (p->on_cpu) { |
1601 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
1602 | /* | |
d6aa8f85 PZ |
1603 | * In case the architecture enables interrupts in |
1604 | * context_switch(), we cannot busy wait, since that | |
1605 | * would lead to deadlocks when an interrupt hits and | |
1606 | * tries to wake up @prev. So bail and do a complete | |
1607 | * remote wakeup. | |
e4a52bcb | 1608 | */ |
d6aa8f85 | 1609 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 1610 | goto stat; |
d6aa8f85 | 1611 | #else |
e4a52bcb | 1612 | cpu_relax(); |
d6aa8f85 | 1613 | #endif |
371fd7e7 | 1614 | } |
0970d299 | 1615 | /* |
e4a52bcb | 1616 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1617 | */ |
e4a52bcb | 1618 | smp_rmb(); |
1da177e4 | 1619 | |
a8e4f2ea | 1620 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1621 | p->state = TASK_WAKING; |
e7693a36 | 1622 | |
e4a52bcb | 1623 | if (p->sched_class->task_waking) |
74f8e4b2 | 1624 | p->sched_class->task_waking(p); |
efbbd05a | 1625 | |
7608dec2 | 1626 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1627 | if (task_cpu(p) != cpu) { |
1628 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1629 | set_task_cpu(p, cpu); |
f339b9dc | 1630 | } |
1da177e4 | 1631 | #endif /* CONFIG_SMP */ |
1da177e4 | 1632 | |
c05fbafb PZ |
1633 | ttwu_queue(p, cpu); |
1634 | stat: | |
b84cb5df | 1635 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1636 | out: |
013fdb80 | 1637 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1638 | |
1639 | return success; | |
1640 | } | |
1641 | ||
21aa9af0 TH |
1642 | /** |
1643 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1644 | * @p: the thread to be awakened | |
1645 | * | |
2acca55e | 1646 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1647 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1648 | * the current task. |
21aa9af0 TH |
1649 | */ |
1650 | static void try_to_wake_up_local(struct task_struct *p) | |
1651 | { | |
1652 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
1653 | |
1654 | BUG_ON(rq != this_rq()); | |
1655 | BUG_ON(p == current); | |
1656 | lockdep_assert_held(&rq->lock); | |
1657 | ||
2acca55e PZ |
1658 | if (!raw_spin_trylock(&p->pi_lock)) { |
1659 | raw_spin_unlock(&rq->lock); | |
1660 | raw_spin_lock(&p->pi_lock); | |
1661 | raw_spin_lock(&rq->lock); | |
1662 | } | |
1663 | ||
21aa9af0 | 1664 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1665 | goto out; |
21aa9af0 | 1666 | |
fd2f4419 | 1667 | if (!p->on_rq) |
d7c01d27 PZ |
1668 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1669 | ||
23f41eeb | 1670 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1671 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1672 | out: |
1673 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1674 | } |
1675 | ||
50fa610a DH |
1676 | /** |
1677 | * wake_up_process - Wake up a specific process | |
1678 | * @p: The process to be woken up. | |
1679 | * | |
1680 | * Attempt to wake up the nominated process and move it to the set of runnable | |
1681 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
1682 | * running. | |
1683 | * | |
1684 | * It may be assumed that this function implies a write memory barrier before | |
1685 | * changing the task state if and only if any tasks are woken up. | |
1686 | */ | |
7ad5b3a5 | 1687 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1688 | { |
d9514f6c | 1689 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1690 | } |
1da177e4 LT |
1691 | EXPORT_SYMBOL(wake_up_process); |
1692 | ||
7ad5b3a5 | 1693 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1694 | { |
1695 | return try_to_wake_up(p, state, 0); | |
1696 | } | |
1697 | ||
1da177e4 LT |
1698 | /* |
1699 | * Perform scheduler related setup for a newly forked process p. | |
1700 | * p is forked by current. | |
dd41f596 IM |
1701 | * |
1702 | * __sched_fork() is basic setup used by init_idle() too: | |
1703 | */ | |
1704 | static void __sched_fork(struct task_struct *p) | |
1705 | { | |
fd2f4419 PZ |
1706 | p->on_rq = 0; |
1707 | ||
1708 | p->se.on_rq = 0; | |
dd41f596 IM |
1709 | p->se.exec_start = 0; |
1710 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1711 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1712 | p->se.nr_migrations = 0; |
da7a735e | 1713 | p->se.vruntime = 0; |
fd2f4419 | 1714 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1715 | |
1716 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1717 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1718 | #endif |
476d139c | 1719 | |
fa717060 | 1720 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1721 | |
e107be36 AK |
1722 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1723 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1724 | #endif | |
dd41f596 IM |
1725 | } |
1726 | ||
1727 | /* | |
1728 | * fork()/clone()-time setup: | |
1729 | */ | |
3e51e3ed | 1730 | void sched_fork(struct task_struct *p) |
dd41f596 | 1731 | { |
0122ec5b | 1732 | unsigned long flags; |
dd41f596 IM |
1733 | int cpu = get_cpu(); |
1734 | ||
1735 | __sched_fork(p); | |
06b83b5f | 1736 | /* |
0017d735 | 1737 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1738 | * nobody will actually run it, and a signal or other external |
1739 | * event cannot wake it up and insert it on the runqueue either. | |
1740 | */ | |
0017d735 | 1741 | p->state = TASK_RUNNING; |
dd41f596 | 1742 | |
c350a04e MG |
1743 | /* |
1744 | * Make sure we do not leak PI boosting priority to the child. | |
1745 | */ | |
1746 | p->prio = current->normal_prio; | |
1747 | ||
b9dc29e7 MG |
1748 | /* |
1749 | * Revert to default priority/policy on fork if requested. | |
1750 | */ | |
1751 | if (unlikely(p->sched_reset_on_fork)) { | |
c350a04e | 1752 | if (task_has_rt_policy(p)) { |
b9dc29e7 | 1753 | p->policy = SCHED_NORMAL; |
6c697bdf | 1754 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1755 | p->rt_priority = 0; |
1756 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1757 | p->static_prio = NICE_TO_PRIO(0); | |
1758 | ||
1759 | p->prio = p->normal_prio = __normal_prio(p); | |
1760 | set_load_weight(p); | |
6c697bdf | 1761 | |
b9dc29e7 MG |
1762 | /* |
1763 | * We don't need the reset flag anymore after the fork. It has | |
1764 | * fulfilled its duty: | |
1765 | */ | |
1766 | p->sched_reset_on_fork = 0; | |
1767 | } | |
ca94c442 | 1768 | |
2ddbf952 HS |
1769 | if (!rt_prio(p->prio)) |
1770 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1771 | |
cd29fe6f PZ |
1772 | if (p->sched_class->task_fork) |
1773 | p->sched_class->task_fork(p); | |
1774 | ||
86951599 PZ |
1775 | /* |
1776 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1777 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1778 | * is ran before sched_fork(). | |
1779 | * | |
1780 | * Silence PROVE_RCU. | |
1781 | */ | |
0122ec5b | 1782 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1783 | set_task_cpu(p, cpu); |
0122ec5b | 1784 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1785 | |
52f17b6c | 1786 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1787 | if (likely(sched_info_on())) |
52f17b6c | 1788 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1789 | #endif |
3ca7a440 PZ |
1790 | #if defined(CONFIG_SMP) |
1791 | p->on_cpu = 0; | |
4866cde0 | 1792 | #endif |
bdd4e85d | 1793 | #ifdef CONFIG_PREEMPT_COUNT |
4866cde0 | 1794 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1795 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1796 | #endif |
806c09a7 | 1797 | #ifdef CONFIG_SMP |
917b627d | 1798 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 1799 | #endif |
917b627d | 1800 | |
476d139c | 1801 | put_cpu(); |
1da177e4 LT |
1802 | } |
1803 | ||
1804 | /* | |
1805 | * wake_up_new_task - wake up a newly created task for the first time. | |
1806 | * | |
1807 | * This function will do some initial scheduler statistics housekeeping | |
1808 | * that must be done for every newly created context, then puts the task | |
1809 | * on the runqueue and wakes it. | |
1810 | */ | |
3e51e3ed | 1811 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
1812 | { |
1813 | unsigned long flags; | |
dd41f596 | 1814 | struct rq *rq; |
fabf318e | 1815 | |
ab2515c4 | 1816 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
1817 | #ifdef CONFIG_SMP |
1818 | /* | |
1819 | * Fork balancing, do it here and not earlier because: | |
1820 | * - cpus_allowed can change in the fork path | |
1821 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 1822 | */ |
ab2515c4 | 1823 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
1824 | #endif |
1825 | ||
ab2515c4 | 1826 | rq = __task_rq_lock(p); |
cd29fe6f | 1827 | activate_task(rq, p, 0); |
fd2f4419 | 1828 | p->on_rq = 1; |
89363381 | 1829 | trace_sched_wakeup_new(p, true); |
a7558e01 | 1830 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 1831 | #ifdef CONFIG_SMP |
efbbd05a PZ |
1832 | if (p->sched_class->task_woken) |
1833 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 1834 | #endif |
0122ec5b | 1835 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
1836 | } |
1837 | ||
e107be36 AK |
1838 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1839 | ||
1840 | /** | |
80dd99b3 | 1841 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 1842 | * @notifier: notifier struct to register |
e107be36 AK |
1843 | */ |
1844 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1845 | { | |
1846 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1847 | } | |
1848 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1849 | ||
1850 | /** | |
1851 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1852 | * @notifier: notifier struct to unregister |
e107be36 AK |
1853 | * |
1854 | * This is safe to call from within a preemption notifier. | |
1855 | */ | |
1856 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1857 | { | |
1858 | hlist_del(¬ifier->link); | |
1859 | } | |
1860 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1861 | ||
1862 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1863 | { | |
1864 | struct preempt_notifier *notifier; | |
1865 | struct hlist_node *node; | |
1866 | ||
1867 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1868 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1869 | } | |
1870 | ||
1871 | static void | |
1872 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1873 | struct task_struct *next) | |
1874 | { | |
1875 | struct preempt_notifier *notifier; | |
1876 | struct hlist_node *node; | |
1877 | ||
1878 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1879 | notifier->ops->sched_out(notifier, next); | |
1880 | } | |
1881 | ||
6d6bc0ad | 1882 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
1883 | |
1884 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1885 | { | |
1886 | } | |
1887 | ||
1888 | static void | |
1889 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1890 | struct task_struct *next) | |
1891 | { | |
1892 | } | |
1893 | ||
6d6bc0ad | 1894 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 1895 | |
4866cde0 NP |
1896 | /** |
1897 | * prepare_task_switch - prepare to switch tasks | |
1898 | * @rq: the runqueue preparing to switch | |
421cee29 | 1899 | * @prev: the current task that is being switched out |
4866cde0 NP |
1900 | * @next: the task we are going to switch to. |
1901 | * | |
1902 | * This is called with the rq lock held and interrupts off. It must | |
1903 | * be paired with a subsequent finish_task_switch after the context | |
1904 | * switch. | |
1905 | * | |
1906 | * prepare_task_switch sets up locking and calls architecture specific | |
1907 | * hooks. | |
1908 | */ | |
e107be36 AK |
1909 | static inline void |
1910 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1911 | struct task_struct *next) | |
4866cde0 | 1912 | { |
895dd92c | 1913 | trace_sched_switch(prev, next); |
fe4b04fa PZ |
1914 | sched_info_switch(prev, next); |
1915 | perf_event_task_sched_out(prev, next); | |
e107be36 | 1916 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1917 | prepare_lock_switch(rq, next); |
1918 | prepare_arch_switch(next); | |
1919 | } | |
1920 | ||
1da177e4 LT |
1921 | /** |
1922 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1923 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1924 | * @prev: the thread we just switched away from. |
1925 | * | |
4866cde0 NP |
1926 | * finish_task_switch must be called after the context switch, paired |
1927 | * with a prepare_task_switch call before the context switch. | |
1928 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1929 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1930 | * |
1931 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1932 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1933 | * with the lock held can cause deadlocks; see schedule() for |
1934 | * details.) | |
1935 | */ | |
a9957449 | 1936 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1937 | __releases(rq->lock) |
1938 | { | |
1da177e4 | 1939 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1940 | long prev_state; |
1da177e4 LT |
1941 | |
1942 | rq->prev_mm = NULL; | |
1943 | ||
1944 | /* | |
1945 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1946 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1947 | * schedule one last time. The schedule call will never return, and |
1948 | * the scheduled task must drop that reference. | |
c394cc9f | 1949 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1950 | * still held, otherwise prev could be scheduled on another cpu, die |
1951 | * there before we look at prev->state, and then the reference would | |
1952 | * be dropped twice. | |
1953 | * Manfred Spraul <manfred@colorfullife.com> | |
1954 | */ | |
55a101f8 | 1955 | prev_state = prev->state; |
4866cde0 | 1956 | finish_arch_switch(prev); |
8381f65d JI |
1957 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1958 | local_irq_disable(); | |
1959 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
a8d757ef | 1960 | perf_event_task_sched_in(prev, current); |
8381f65d JI |
1961 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
1962 | local_irq_enable(); | |
1963 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 1964 | finish_lock_switch(rq, prev); |
01f23e16 | 1965 | finish_arch_post_lock_switch(); |
e8fa1362 | 1966 | |
e107be36 | 1967 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1968 | if (mm) |
1969 | mmdrop(mm); | |
c394cc9f | 1970 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1971 | /* |
1972 | * Remove function-return probe instances associated with this | |
1973 | * task and put them back on the free list. | |
9761eea8 | 1974 | */ |
c6fd91f0 | 1975 | kprobe_flush_task(prev); |
1da177e4 | 1976 | put_task_struct(prev); |
c6fd91f0 | 1977 | } |
1da177e4 LT |
1978 | } |
1979 | ||
3f029d3c GH |
1980 | #ifdef CONFIG_SMP |
1981 | ||
1982 | /* assumes rq->lock is held */ | |
1983 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
1984 | { | |
1985 | if (prev->sched_class->pre_schedule) | |
1986 | prev->sched_class->pre_schedule(rq, prev); | |
1987 | } | |
1988 | ||
1989 | /* rq->lock is NOT held, but preemption is disabled */ | |
1990 | static inline void post_schedule(struct rq *rq) | |
1991 | { | |
1992 | if (rq->post_schedule) { | |
1993 | unsigned long flags; | |
1994 | ||
05fa785c | 1995 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
1996 | if (rq->curr->sched_class->post_schedule) |
1997 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 1998 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
1999 | |
2000 | rq->post_schedule = 0; | |
2001 | } | |
2002 | } | |
2003 | ||
2004 | #else | |
da19ab51 | 2005 | |
3f029d3c GH |
2006 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2007 | { | |
2008 | } | |
2009 | ||
2010 | static inline void post_schedule(struct rq *rq) | |
2011 | { | |
1da177e4 LT |
2012 | } |
2013 | ||
3f029d3c GH |
2014 | #endif |
2015 | ||
1da177e4 LT |
2016 | /** |
2017 | * schedule_tail - first thing a freshly forked thread must call. | |
2018 | * @prev: the thread we just switched away from. | |
2019 | */ | |
36c8b586 | 2020 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2021 | __releases(rq->lock) |
2022 | { | |
70b97a7f IM |
2023 | struct rq *rq = this_rq(); |
2024 | ||
4866cde0 | 2025 | finish_task_switch(rq, prev); |
da19ab51 | 2026 | |
3f029d3c GH |
2027 | /* |
2028 | * FIXME: do we need to worry about rq being invalidated by the | |
2029 | * task_switch? | |
2030 | */ | |
2031 | post_schedule(rq); | |
70b97a7f | 2032 | |
4866cde0 NP |
2033 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2034 | /* In this case, finish_task_switch does not reenable preemption */ | |
2035 | preempt_enable(); | |
2036 | #endif | |
1da177e4 | 2037 | if (current->set_child_tid) |
b488893a | 2038 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2039 | } |
2040 | ||
2041 | /* | |
2042 | * context_switch - switch to the new MM and the new | |
2043 | * thread's register state. | |
2044 | */ | |
dd41f596 | 2045 | static inline void |
70b97a7f | 2046 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2047 | struct task_struct *next) |
1da177e4 | 2048 | { |
dd41f596 | 2049 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2050 | |
e107be36 | 2051 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2052 | |
dd41f596 IM |
2053 | mm = next->mm; |
2054 | oldmm = prev->active_mm; | |
9226d125 ZA |
2055 | /* |
2056 | * For paravirt, this is coupled with an exit in switch_to to | |
2057 | * combine the page table reload and the switch backend into | |
2058 | * one hypercall. | |
2059 | */ | |
224101ed | 2060 | arch_start_context_switch(prev); |
9226d125 | 2061 | |
31915ab4 | 2062 | if (!mm) { |
1da177e4 LT |
2063 | next->active_mm = oldmm; |
2064 | atomic_inc(&oldmm->mm_count); | |
2065 | enter_lazy_tlb(oldmm, next); | |
2066 | } else | |
2067 | switch_mm(oldmm, mm, next); | |
2068 | ||
31915ab4 | 2069 | if (!prev->mm) { |
1da177e4 | 2070 | prev->active_mm = NULL; |
1da177e4 LT |
2071 | rq->prev_mm = oldmm; |
2072 | } | |
3a5f5e48 IM |
2073 | /* |
2074 | * Since the runqueue lock will be released by the next | |
2075 | * task (which is an invalid locking op but in the case | |
2076 | * of the scheduler it's an obvious special-case), so we | |
2077 | * do an early lockdep release here: | |
2078 | */ | |
2079 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2080 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2081 | #endif |
1da177e4 LT |
2082 | |
2083 | /* Here we just switch the register state and the stack. */ | |
2084 | switch_to(prev, next, prev); | |
2085 | ||
dd41f596 IM |
2086 | barrier(); |
2087 | /* | |
2088 | * this_rq must be evaluated again because prev may have moved | |
2089 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2090 | * frame will be invalid. | |
2091 | */ | |
2092 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2093 | } |
2094 | ||
2095 | /* | |
2096 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2097 | * | |
2098 | * externally visible scheduler statistics: current number of runnable | |
2099 | * threads, current number of uninterruptible-sleeping threads, total | |
2100 | * number of context switches performed since bootup. | |
2101 | */ | |
2102 | unsigned long nr_running(void) | |
2103 | { | |
2104 | unsigned long i, sum = 0; | |
2105 | ||
2106 | for_each_online_cpu(i) | |
2107 | sum += cpu_rq(i)->nr_running; | |
2108 | ||
2109 | return sum; | |
f711f609 | 2110 | } |
1da177e4 LT |
2111 | |
2112 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2113 | { |
1da177e4 | 2114 | unsigned long i, sum = 0; |
f711f609 | 2115 | |
0a945022 | 2116 | for_each_possible_cpu(i) |
1da177e4 | 2117 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2118 | |
2119 | /* | |
1da177e4 LT |
2120 | * Since we read the counters lockless, it might be slightly |
2121 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2122 | */ |
1da177e4 LT |
2123 | if (unlikely((long)sum < 0)) |
2124 | sum = 0; | |
f711f609 | 2125 | |
1da177e4 | 2126 | return sum; |
f711f609 | 2127 | } |
f711f609 | 2128 | |
1da177e4 | 2129 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2130 | { |
cc94abfc SR |
2131 | int i; |
2132 | unsigned long long sum = 0; | |
46cb4b7c | 2133 | |
0a945022 | 2134 | for_each_possible_cpu(i) |
1da177e4 | 2135 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2136 | |
1da177e4 LT |
2137 | return sum; |
2138 | } | |
483b4ee6 | 2139 | |
1da177e4 LT |
2140 | unsigned long nr_iowait(void) |
2141 | { | |
2142 | unsigned long i, sum = 0; | |
483b4ee6 | 2143 | |
0a945022 | 2144 | for_each_possible_cpu(i) |
1da177e4 | 2145 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2146 | |
1da177e4 LT |
2147 | return sum; |
2148 | } | |
483b4ee6 | 2149 | |
8c215bd3 | 2150 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2151 | { |
8c215bd3 | 2152 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2153 | return atomic_read(&this->nr_iowait); |
2154 | } | |
46cb4b7c | 2155 | |
69d25870 AV |
2156 | unsigned long this_cpu_load(void) |
2157 | { | |
2158 | struct rq *this = this_rq(); | |
2159 | return this->cpu_load[0]; | |
2160 | } | |
e790fb0b | 2161 | |
46cb4b7c | 2162 | |
5167e8d5 PZ |
2163 | /* |
2164 | * Global load-average calculations | |
2165 | * | |
2166 | * We take a distributed and async approach to calculating the global load-avg | |
2167 | * in order to minimize overhead. | |
2168 | * | |
2169 | * The global load average is an exponentially decaying average of nr_running + | |
2170 | * nr_uninterruptible. | |
2171 | * | |
2172 | * Once every LOAD_FREQ: | |
2173 | * | |
2174 | * nr_active = 0; | |
2175 | * for_each_possible_cpu(cpu) | |
2176 | * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible; | |
2177 | * | |
2178 | * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n) | |
2179 | * | |
2180 | * Due to a number of reasons the above turns in the mess below: | |
2181 | * | |
2182 | * - for_each_possible_cpu() is prohibitively expensive on machines with | |
2183 | * serious number of cpus, therefore we need to take a distributed approach | |
2184 | * to calculating nr_active. | |
2185 | * | |
2186 | * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0 | |
2187 | * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) } | |
2188 | * | |
2189 | * So assuming nr_active := 0 when we start out -- true per definition, we | |
2190 | * can simply take per-cpu deltas and fold those into a global accumulate | |
2191 | * to obtain the same result. See calc_load_fold_active(). | |
2192 | * | |
2193 | * Furthermore, in order to avoid synchronizing all per-cpu delta folding | |
2194 | * across the machine, we assume 10 ticks is sufficient time for every | |
2195 | * cpu to have completed this task. | |
2196 | * | |
2197 | * This places an upper-bound on the IRQ-off latency of the machine. Then | |
2198 | * again, being late doesn't loose the delta, just wrecks the sample. | |
2199 | * | |
2200 | * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because | |
2201 | * this would add another cross-cpu cacheline miss and atomic operation | |
2202 | * to the wakeup path. Instead we increment on whatever cpu the task ran | |
2203 | * when it went into uninterruptible state and decrement on whatever cpu | |
2204 | * did the wakeup. This means that only the sum of nr_uninterruptible over | |
2205 | * all cpus yields the correct result. | |
2206 | * | |
2207 | * This covers the NO_HZ=n code, for extra head-aches, see the comment below. | |
2208 | */ | |
2209 | ||
dce48a84 TG |
2210 | /* Variables and functions for calc_load */ |
2211 | static atomic_long_t calc_load_tasks; | |
2212 | static unsigned long calc_load_update; | |
2213 | unsigned long avenrun[3]; | |
5167e8d5 PZ |
2214 | EXPORT_SYMBOL(avenrun); /* should be removed */ |
2215 | ||
2216 | /** | |
2217 | * get_avenrun - get the load average array | |
2218 | * @loads: pointer to dest load array | |
2219 | * @offset: offset to add | |
2220 | * @shift: shift count to shift the result left | |
2221 | * | |
2222 | * These values are estimates at best, so no need for locking. | |
2223 | */ | |
2224 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2225 | { | |
2226 | loads[0] = (avenrun[0] + offset) << shift; | |
2227 | loads[1] = (avenrun[1] + offset) << shift; | |
2228 | loads[2] = (avenrun[2] + offset) << shift; | |
2229 | } | |
46cb4b7c | 2230 | |
74f5187a PZ |
2231 | static long calc_load_fold_active(struct rq *this_rq) |
2232 | { | |
2233 | long nr_active, delta = 0; | |
2234 | ||
2235 | nr_active = this_rq->nr_running; | |
2236 | nr_active += (long) this_rq->nr_uninterruptible; | |
2237 | ||
2238 | if (nr_active != this_rq->calc_load_active) { | |
2239 | delta = nr_active - this_rq->calc_load_active; | |
2240 | this_rq->calc_load_active = nr_active; | |
2241 | } | |
2242 | ||
2243 | return delta; | |
2244 | } | |
2245 | ||
5167e8d5 PZ |
2246 | /* |
2247 | * a1 = a0 * e + a * (1 - e) | |
2248 | */ | |
0f004f5a PZ |
2249 | static unsigned long |
2250 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
2251 | { | |
2252 | load *= exp; | |
2253 | load += active * (FIXED_1 - exp); | |
2254 | load += 1UL << (FSHIFT - 1); | |
2255 | return load >> FSHIFT; | |
2256 | } | |
2257 | ||
74f5187a PZ |
2258 | #ifdef CONFIG_NO_HZ |
2259 | /* | |
5167e8d5 PZ |
2260 | * Handle NO_HZ for the global load-average. |
2261 | * | |
2262 | * Since the above described distributed algorithm to compute the global | |
2263 | * load-average relies on per-cpu sampling from the tick, it is affected by | |
2264 | * NO_HZ. | |
2265 | * | |
2266 | * The basic idea is to fold the nr_active delta into a global idle-delta upon | |
2267 | * entering NO_HZ state such that we can include this as an 'extra' cpu delta | |
2268 | * when we read the global state. | |
2269 | * | |
2270 | * Obviously reality has to ruin such a delightfully simple scheme: | |
2271 | * | |
2272 | * - When we go NO_HZ idle during the window, we can negate our sample | |
2273 | * contribution, causing under-accounting. | |
2274 | * | |
2275 | * We avoid this by keeping two idle-delta counters and flipping them | |
2276 | * when the window starts, thus separating old and new NO_HZ load. | |
2277 | * | |
2278 | * The only trick is the slight shift in index flip for read vs write. | |
2279 | * | |
2280 | * 0s 5s 10s 15s | |
2281 | * +10 +10 +10 +10 | |
2282 | * |-|-----------|-|-----------|-|-----------|-| | |
2283 | * r:0 0 1 1 0 0 1 1 0 | |
2284 | * w:0 1 1 0 0 1 1 0 0 | |
2285 | * | |
2286 | * This ensures we'll fold the old idle contribution in this window while | |
2287 | * accumlating the new one. | |
2288 | * | |
2289 | * - When we wake up from NO_HZ idle during the window, we push up our | |
2290 | * contribution, since we effectively move our sample point to a known | |
2291 | * busy state. | |
2292 | * | |
2293 | * This is solved by pushing the window forward, and thus skipping the | |
2294 | * sample, for this cpu (effectively using the idle-delta for this cpu which | |
2295 | * was in effect at the time the window opened). This also solves the issue | |
2296 | * of having to deal with a cpu having been in NOHZ idle for multiple | |
2297 | * LOAD_FREQ intervals. | |
74f5187a PZ |
2298 | * |
2299 | * When making the ILB scale, we should try to pull this in as well. | |
2300 | */ | |
5167e8d5 PZ |
2301 | static atomic_long_t calc_load_idle[2]; |
2302 | static int calc_load_idx; | |
74f5187a | 2303 | |
5167e8d5 | 2304 | static inline int calc_load_write_idx(void) |
74f5187a | 2305 | { |
5167e8d5 PZ |
2306 | int idx = calc_load_idx; |
2307 | ||
2308 | /* | |
2309 | * See calc_global_nohz(), if we observe the new index, we also | |
2310 | * need to observe the new update time. | |
2311 | */ | |
2312 | smp_rmb(); | |
2313 | ||
2314 | /* | |
2315 | * If the folding window started, make sure we start writing in the | |
2316 | * next idle-delta. | |
2317 | */ | |
2318 | if (!time_before(jiffies, calc_load_update)) | |
2319 | idx++; | |
2320 | ||
2321 | return idx & 1; | |
2322 | } | |
2323 | ||
2324 | static inline int calc_load_read_idx(void) | |
2325 | { | |
2326 | return calc_load_idx & 1; | |
2327 | } | |
2328 | ||
2329 | void calc_load_enter_idle(void) | |
2330 | { | |
2331 | struct rq *this_rq = this_rq(); | |
74f5187a PZ |
2332 | long delta; |
2333 | ||
5167e8d5 PZ |
2334 | /* |
2335 | * We're going into NOHZ mode, if there's any pending delta, fold it | |
2336 | * into the pending idle delta. | |
2337 | */ | |
74f5187a | 2338 | delta = calc_load_fold_active(this_rq); |
5167e8d5 PZ |
2339 | if (delta) { |
2340 | int idx = calc_load_write_idx(); | |
2341 | atomic_long_add(delta, &calc_load_idle[idx]); | |
2342 | } | |
74f5187a PZ |
2343 | } |
2344 | ||
5167e8d5 | 2345 | void calc_load_exit_idle(void) |
74f5187a | 2346 | { |
5167e8d5 PZ |
2347 | struct rq *this_rq = this_rq(); |
2348 | ||
2349 | /* | |
2350 | * If we're still before the sample window, we're done. | |
2351 | */ | |
2352 | if (time_before(jiffies, this_rq->calc_load_update)) | |
2353 | return; | |
74f5187a PZ |
2354 | |
2355 | /* | |
5167e8d5 PZ |
2356 | * We woke inside or after the sample window, this means we're already |
2357 | * accounted through the nohz accounting, so skip the entire deal and | |
2358 | * sync up for the next window. | |
74f5187a | 2359 | */ |
5167e8d5 PZ |
2360 | this_rq->calc_load_update = calc_load_update; |
2361 | if (time_before(jiffies, this_rq->calc_load_update + 10)) | |
2362 | this_rq->calc_load_update += LOAD_FREQ; | |
2363 | } | |
2364 | ||
2365 | static long calc_load_fold_idle(void) | |
2366 | { | |
2367 | int idx = calc_load_read_idx(); | |
2368 | long delta = 0; | |
2369 | ||
2370 | if (atomic_long_read(&calc_load_idle[idx])) | |
2371 | delta = atomic_long_xchg(&calc_load_idle[idx], 0); | |
74f5187a PZ |
2372 | |
2373 | return delta; | |
2374 | } | |
0f004f5a PZ |
2375 | |
2376 | /** | |
2377 | * fixed_power_int - compute: x^n, in O(log n) time | |
2378 | * | |
2379 | * @x: base of the power | |
2380 | * @frac_bits: fractional bits of @x | |
2381 | * @n: power to raise @x to. | |
2382 | * | |
2383 | * By exploiting the relation between the definition of the natural power | |
2384 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
2385 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
2386 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
2387 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
2388 | * of course trivially computable in O(log_2 n), the length of our binary | |
2389 | * vector. | |
2390 | */ | |
2391 | static unsigned long | |
2392 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
2393 | { | |
2394 | unsigned long result = 1UL << frac_bits; | |
2395 | ||
2396 | if (n) for (;;) { | |
2397 | if (n & 1) { | |
2398 | result *= x; | |
2399 | result += 1UL << (frac_bits - 1); | |
2400 | result >>= frac_bits; | |
2401 | } | |
2402 | n >>= 1; | |
2403 | if (!n) | |
2404 | break; | |
2405 | x *= x; | |
2406 | x += 1UL << (frac_bits - 1); | |
2407 | x >>= frac_bits; | |
2408 | } | |
2409 | ||
2410 | return result; | |
2411 | } | |
2412 | ||
2413 | /* | |
2414 | * a1 = a0 * e + a * (1 - e) | |
2415 | * | |
2416 | * a2 = a1 * e + a * (1 - e) | |
2417 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
2418 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
2419 | * | |
2420 | * a3 = a2 * e + a * (1 - e) | |
2421 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
2422 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
2423 | * | |
2424 | * ... | |
2425 | * | |
2426 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
2427 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
2428 | * = a0 * e^n + a * (1 - e^n) | |
2429 | * | |
2430 | * [1] application of the geometric series: | |
2431 | * | |
2432 | * n 1 - x^(n+1) | |
2433 | * S_n := \Sum x^i = ------------- | |
2434 | * i=0 1 - x | |
2435 | */ | |
2436 | static unsigned long | |
2437 | calc_load_n(unsigned long load, unsigned long exp, | |
2438 | unsigned long active, unsigned int n) | |
2439 | { | |
2440 | ||
2441 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
2442 | } | |
2443 | ||
2444 | /* | |
2445 | * NO_HZ can leave us missing all per-cpu ticks calling | |
2446 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
2447 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
2448 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
2449 | * | |
2450 | * Once we've updated the global active value, we need to apply the exponential | |
2451 | * weights adjusted to the number of cycles missed. | |
2452 | */ | |
c308b56b | 2453 | static void calc_global_nohz(void) |
0f004f5a PZ |
2454 | { |
2455 | long delta, active, n; | |
2456 | ||
5167e8d5 PZ |
2457 | if (!time_before(jiffies, calc_load_update + 10)) { |
2458 | /* | |
2459 | * Catch-up, fold however many we are behind still | |
2460 | */ | |
2461 | delta = jiffies - calc_load_update - 10; | |
2462 | n = 1 + (delta / LOAD_FREQ); | |
0f004f5a | 2463 | |
5167e8d5 PZ |
2464 | active = atomic_long_read(&calc_load_tasks); |
2465 | active = active > 0 ? active * FIXED_1 : 0; | |
0f004f5a | 2466 | |
5167e8d5 PZ |
2467 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); |
2468 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
2469 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
0f004f5a | 2470 | |
5167e8d5 PZ |
2471 | calc_load_update += n * LOAD_FREQ; |
2472 | } | |
74f5187a | 2473 | |
5167e8d5 PZ |
2474 | /* |
2475 | * Flip the idle index... | |
2476 | * | |
2477 | * Make sure we first write the new time then flip the index, so that | |
2478 | * calc_load_write_idx() will see the new time when it reads the new | |
2479 | * index, this avoids a double flip messing things up. | |
2480 | */ | |
2481 | smp_wmb(); | |
2482 | calc_load_idx++; | |
74f5187a | 2483 | } |
5167e8d5 | 2484 | #else /* !CONFIG_NO_HZ */ |
0f004f5a | 2485 | |
5167e8d5 PZ |
2486 | static inline long calc_load_fold_idle(void) { return 0; } |
2487 | static inline void calc_global_nohz(void) { } | |
74f5187a | 2488 | |
5167e8d5 | 2489 | #endif /* CONFIG_NO_HZ */ |
46cb4b7c | 2490 | |
46cb4b7c | 2491 | /* |
dce48a84 TG |
2492 | * calc_load - update the avenrun load estimates 10 ticks after the |
2493 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2494 | */ |
0f004f5a | 2495 | void calc_global_load(unsigned long ticks) |
7835b98b | 2496 | { |
5167e8d5 | 2497 | long active, delta; |
1da177e4 | 2498 | |
0f004f5a | 2499 | if (time_before(jiffies, calc_load_update + 10)) |
dce48a84 | 2500 | return; |
1da177e4 | 2501 | |
5167e8d5 PZ |
2502 | /* |
2503 | * Fold the 'old' idle-delta to include all NO_HZ cpus. | |
2504 | */ | |
2505 | delta = calc_load_fold_idle(); | |
2506 | if (delta) | |
2507 | atomic_long_add(delta, &calc_load_tasks); | |
2508 | ||
dce48a84 TG |
2509 | active = atomic_long_read(&calc_load_tasks); |
2510 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2511 | |
dce48a84 TG |
2512 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2513 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2514 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2515 | |
dce48a84 | 2516 | calc_load_update += LOAD_FREQ; |
c308b56b PZ |
2517 | |
2518 | /* | |
5167e8d5 | 2519 | * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk. |
c308b56b PZ |
2520 | */ |
2521 | calc_global_nohz(); | |
dce48a84 | 2522 | } |
1da177e4 | 2523 | |
dce48a84 | 2524 | /* |
74f5187a PZ |
2525 | * Called from update_cpu_load() to periodically update this CPU's |
2526 | * active count. | |
dce48a84 TG |
2527 | */ |
2528 | static void calc_load_account_active(struct rq *this_rq) | |
2529 | { | |
74f5187a | 2530 | long delta; |
08c183f3 | 2531 | |
74f5187a PZ |
2532 | if (time_before(jiffies, this_rq->calc_load_update)) |
2533 | return; | |
783609c6 | 2534 | |
74f5187a | 2535 | delta = calc_load_fold_active(this_rq); |
74f5187a | 2536 | if (delta) |
dce48a84 | 2537 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
2538 | |
2539 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
2540 | } |
2541 | ||
5167e8d5 PZ |
2542 | /* |
2543 | * End of global load-average stuff | |
2544 | */ | |
2545 | ||
fdf3e95d VP |
2546 | /* |
2547 | * The exact cpuload at various idx values, calculated at every tick would be | |
2548 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
2549 | * | |
2550 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
2551 | * on nth tick when cpu may be busy, then we have: | |
2552 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2553 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
2554 | * | |
2555 | * decay_load_missed() below does efficient calculation of | |
2556 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
2557 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
2558 | * | |
2559 | * The calculation is approximated on a 128 point scale. | |
2560 | * degrade_zero_ticks is the number of ticks after which load at any | |
2561 | * particular idx is approximated to be zero. | |
2562 | * degrade_factor is a precomputed table, a row for each load idx. | |
2563 | * Each column corresponds to degradation factor for a power of two ticks, | |
2564 | * based on 128 point scale. | |
2565 | * Example: | |
2566 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
2567 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
2568 | * | |
2569 | * With this power of 2 load factors, we can degrade the load n times | |
2570 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
2571 | * n mult/shifts needed by the exact degradation. | |
2572 | */ | |
2573 | #define DEGRADE_SHIFT 7 | |
2574 | static const unsigned char | |
2575 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
2576 | static const unsigned char | |
2577 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
2578 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
2579 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
2580 | {96, 72, 40, 12, 1, 0, 0}, | |
2581 | {112, 98, 75, 43, 15, 1, 0}, | |
2582 | {120, 112, 98, 76, 45, 16, 2} }; | |
2583 | ||
2584 | /* | |
2585 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
2586 | * would be when CPU is idle and so we just decay the old load without | |
2587 | * adding any new load. | |
2588 | */ | |
2589 | static unsigned long | |
2590 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
2591 | { | |
2592 | int j = 0; | |
2593 | ||
2594 | if (!missed_updates) | |
2595 | return load; | |
2596 | ||
2597 | if (missed_updates >= degrade_zero_ticks[idx]) | |
2598 | return 0; | |
2599 | ||
2600 | if (idx == 1) | |
2601 | return load >> missed_updates; | |
2602 | ||
2603 | while (missed_updates) { | |
2604 | if (missed_updates % 2) | |
2605 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
2606 | ||
2607 | missed_updates >>= 1; | |
2608 | j++; | |
2609 | } | |
2610 | return load; | |
2611 | } | |
2612 | ||
46cb4b7c | 2613 | /* |
dd41f596 | 2614 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
2615 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
2616 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 2617 | */ |
556061b0 PZ |
2618 | static void __update_cpu_load(struct rq *this_rq, unsigned long this_load, |
2619 | unsigned long pending_updates) | |
46cb4b7c | 2620 | { |
dd41f596 | 2621 | int i, scale; |
46cb4b7c | 2622 | |
dd41f596 | 2623 | this_rq->nr_load_updates++; |
46cb4b7c | 2624 | |
dd41f596 | 2625 | /* Update our load: */ |
fdf3e95d VP |
2626 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
2627 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 2628 | unsigned long old_load, new_load; |
7d1e6a9b | 2629 | |
dd41f596 | 2630 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 2631 | |
dd41f596 | 2632 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 2633 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 2634 | new_load = this_load; |
a25707f3 IM |
2635 | /* |
2636 | * Round up the averaging division if load is increasing. This | |
2637 | * prevents us from getting stuck on 9 if the load is 10, for | |
2638 | * example. | |
2639 | */ | |
2640 | if (new_load > old_load) | |
fdf3e95d VP |
2641 | new_load += scale - 1; |
2642 | ||
2643 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 2644 | } |
da2b71ed SS |
2645 | |
2646 | sched_avg_update(this_rq); | |
fdf3e95d VP |
2647 | } |
2648 | ||
5aaa0b7a PZ |
2649 | #ifdef CONFIG_NO_HZ |
2650 | /* | |
2651 | * There is no sane way to deal with nohz on smp when using jiffies because the | |
2652 | * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading | |
2653 | * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}. | |
2654 | * | |
2655 | * Therefore we cannot use the delta approach from the regular tick since that | |
2656 | * would seriously skew the load calculation. However we'll make do for those | |
2657 | * updates happening while idle (nohz_idle_balance) or coming out of idle | |
2658 | * (tick_nohz_idle_exit). | |
2659 | * | |
2660 | * This means we might still be one tick off for nohz periods. | |
2661 | */ | |
2662 | ||
556061b0 PZ |
2663 | /* |
2664 | * Called from nohz_idle_balance() to update the load ratings before doing the | |
2665 | * idle balance. | |
2666 | */ | |
2667 | void update_idle_cpu_load(struct rq *this_rq) | |
2668 | { | |
5aaa0b7a | 2669 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); |
556061b0 PZ |
2670 | unsigned long load = this_rq->load.weight; |
2671 | unsigned long pending_updates; | |
2672 | ||
2673 | /* | |
5aaa0b7a | 2674 | * bail if there's load or we're actually up-to-date. |
556061b0 PZ |
2675 | */ |
2676 | if (load || curr_jiffies == this_rq->last_load_update_tick) | |
2677 | return; | |
2678 | ||
2679 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2680 | this_rq->last_load_update_tick = curr_jiffies; | |
2681 | ||
2682 | __update_cpu_load(this_rq, load, pending_updates); | |
2683 | } | |
2684 | ||
5aaa0b7a PZ |
2685 | /* |
2686 | * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed. | |
2687 | */ | |
2688 | void update_cpu_load_nohz(void) | |
2689 | { | |
2690 | struct rq *this_rq = this_rq(); | |
2691 | unsigned long curr_jiffies = ACCESS_ONCE(jiffies); | |
2692 | unsigned long pending_updates; | |
2693 | ||
2694 | if (curr_jiffies == this_rq->last_load_update_tick) | |
2695 | return; | |
2696 | ||
2697 | raw_spin_lock(&this_rq->lock); | |
2698 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
2699 | if (pending_updates) { | |
2700 | this_rq->last_load_update_tick = curr_jiffies; | |
2701 | /* | |
2702 | * We were idle, this means load 0, the current load might be | |
2703 | * !0 due to remote wakeups and the sort. | |
2704 | */ | |
2705 | __update_cpu_load(this_rq, 0, pending_updates); | |
2706 | } | |
2707 | raw_spin_unlock(&this_rq->lock); | |
2708 | } | |
2709 | #endif /* CONFIG_NO_HZ */ | |
2710 | ||
556061b0 PZ |
2711 | /* |
2712 | * Called from scheduler_tick() | |
2713 | */ | |
fdf3e95d VP |
2714 | static void update_cpu_load_active(struct rq *this_rq) |
2715 | { | |
556061b0 | 2716 | /* |
5aaa0b7a | 2717 | * See the mess around update_idle_cpu_load() / update_cpu_load_nohz(). |
556061b0 PZ |
2718 | */ |
2719 | this_rq->last_load_update_tick = jiffies; | |
2720 | __update_cpu_load(this_rq, this_rq->load.weight, 1); | |
46cb4b7c | 2721 | |
74f5187a | 2722 | calc_load_account_active(this_rq); |
46cb4b7c SS |
2723 | } |
2724 | ||
dd41f596 | 2725 | #ifdef CONFIG_SMP |
8a0be9ef | 2726 | |
46cb4b7c | 2727 | /* |
38022906 PZ |
2728 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2729 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2730 | */ |
38022906 | 2731 | void sched_exec(void) |
46cb4b7c | 2732 | { |
38022906 | 2733 | struct task_struct *p = current; |
1da177e4 | 2734 | unsigned long flags; |
0017d735 | 2735 | int dest_cpu; |
46cb4b7c | 2736 | |
8f42ced9 | 2737 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 2738 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2739 | if (dest_cpu == smp_processor_id()) |
2740 | goto unlock; | |
38022906 | 2741 | |
8f42ced9 | 2742 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2743 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2744 | |
8f42ced9 PZ |
2745 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2746 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2747 | return; |
2748 | } | |
0017d735 | 2749 | unlock: |
8f42ced9 | 2750 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2751 | } |
dd41f596 | 2752 | |
1da177e4 LT |
2753 | #endif |
2754 | ||
1da177e4 | 2755 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2756 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2757 | |
2758 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2759 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2760 | |
2761 | /* | |
c5f8d995 | 2762 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2763 | * @p in case that task is currently running. |
c5f8d995 HS |
2764 | * |
2765 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2766 | */ |
c5f8d995 HS |
2767 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2768 | { | |
2769 | u64 ns = 0; | |
2770 | ||
2771 | if (task_current(rq, p)) { | |
2772 | update_rq_clock(rq); | |
305e6835 | 2773 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
2774 | if ((s64)ns < 0) |
2775 | ns = 0; | |
2776 | } | |
2777 | ||
2778 | return ns; | |
2779 | } | |
2780 | ||
bb34d92f | 2781 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2782 | { |
1da177e4 | 2783 | unsigned long flags; |
41b86e9c | 2784 | struct rq *rq; |
bb34d92f | 2785 | u64 ns = 0; |
48f24c4d | 2786 | |
41b86e9c | 2787 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2788 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2789 | task_rq_unlock(rq, p, &flags); |
1508487e | 2790 | |
c5f8d995 HS |
2791 | return ns; |
2792 | } | |
f06febc9 | 2793 | |
c5f8d995 HS |
2794 | /* |
2795 | * Return accounted runtime for the task. | |
2796 | * In case the task is currently running, return the runtime plus current's | |
2797 | * pending runtime that have not been accounted yet. | |
2798 | */ | |
2799 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2800 | { | |
2801 | unsigned long flags; | |
2802 | struct rq *rq; | |
2803 | u64 ns = 0; | |
2804 | ||
2805 | rq = task_rq_lock(p, &flags); | |
2806 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2807 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2808 | |
2809 | return ns; | |
2810 | } | |
48f24c4d | 2811 | |
54c707e9 GC |
2812 | #ifdef CONFIG_CGROUP_CPUACCT |
2813 | struct cgroup_subsys cpuacct_subsys; | |
2814 | struct cpuacct root_cpuacct; | |
2815 | #endif | |
2816 | ||
be726ffd GC |
2817 | static inline void task_group_account_field(struct task_struct *p, int index, |
2818 | u64 tmp) | |
54c707e9 GC |
2819 | { |
2820 | #ifdef CONFIG_CGROUP_CPUACCT | |
2821 | struct kernel_cpustat *kcpustat; | |
2822 | struct cpuacct *ca; | |
2823 | #endif | |
2824 | /* | |
2825 | * Since all updates are sure to touch the root cgroup, we | |
2826 | * get ourselves ahead and touch it first. If the root cgroup | |
2827 | * is the only cgroup, then nothing else should be necessary. | |
2828 | * | |
2829 | */ | |
2830 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | |
2831 | ||
2832 | #ifdef CONFIG_CGROUP_CPUACCT | |
2833 | if (unlikely(!cpuacct_subsys.active)) | |
2834 | return; | |
2835 | ||
2836 | rcu_read_lock(); | |
2837 | ca = task_ca(p); | |
2838 | while (ca && (ca != &root_cpuacct)) { | |
2839 | kcpustat = this_cpu_ptr(ca->cpustat); | |
2840 | kcpustat->cpustat[index] += tmp; | |
2841 | ca = parent_ca(ca); | |
2842 | } | |
2843 | rcu_read_unlock(); | |
2844 | #endif | |
2845 | } | |
2846 | ||
2847 | ||
1da177e4 LT |
2848 | /* |
2849 | * Account user cpu time to a process. | |
2850 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 2851 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 2852 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 2853 | */ |
457533a7 MS |
2854 | void account_user_time(struct task_struct *p, cputime_t cputime, |
2855 | cputime_t cputime_scaled) | |
1da177e4 | 2856 | { |
3292beb3 | 2857 | int index; |
1da177e4 | 2858 | |
457533a7 | 2859 | /* Add user time to process. */ |
64861634 MS |
2860 | p->utime += cputime; |
2861 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2862 | account_group_user_time(p, cputime); |
1da177e4 | 2863 | |
3292beb3 | 2864 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
ef12fefa | 2865 | |
1da177e4 | 2866 | /* Add user time to cpustat. */ |
612ef28a | 2867 | task_group_account_field(p, index, (__force u64) cputime); |
ef12fefa | 2868 | |
49b5cf34 JL |
2869 | /* Account for user time used */ |
2870 | acct_update_integrals(p); | |
1da177e4 LT |
2871 | } |
2872 | ||
94886b84 LV |
2873 | /* |
2874 | * Account guest cpu time to a process. | |
2875 | * @p: the process that the cpu time gets accounted to | |
2876 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 2877 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 2878 | */ |
457533a7 MS |
2879 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
2880 | cputime_t cputime_scaled) | |
94886b84 | 2881 | { |
3292beb3 | 2882 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
94886b84 | 2883 | |
457533a7 | 2884 | /* Add guest time to process. */ |
64861634 MS |
2885 | p->utime += cputime; |
2886 | p->utimescaled += cputime_scaled; | |
f06febc9 | 2887 | account_group_user_time(p, cputime); |
64861634 | 2888 | p->gtime += cputime; |
94886b84 | 2889 | |
457533a7 | 2890 | /* Add guest time to cpustat. */ |
ce0e7b28 | 2891 | if (TASK_NICE(p) > 0) { |
612ef28a MS |
2892 | cpustat[CPUTIME_NICE] += (__force u64) cputime; |
2893 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
ce0e7b28 | 2894 | } else { |
612ef28a MS |
2895 | cpustat[CPUTIME_USER] += (__force u64) cputime; |
2896 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
ce0e7b28 | 2897 | } |
94886b84 LV |
2898 | } |
2899 | ||
70a89a66 VP |
2900 | /* |
2901 | * Account system cpu time to a process and desired cpustat field | |
2902 | * @p: the process that the cpu time gets accounted to | |
2903 | * @cputime: the cpu time spent in kernel space since the last update | |
2904 | * @cputime_scaled: cputime scaled by cpu frequency | |
2905 | * @target_cputime64: pointer to cpustat field that has to be updated | |
2906 | */ | |
2907 | static inline | |
2908 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3292beb3 | 2909 | cputime_t cputime_scaled, int index) |
70a89a66 | 2910 | { |
70a89a66 | 2911 | /* Add system time to process. */ |
64861634 MS |
2912 | p->stime += cputime; |
2913 | p->stimescaled += cputime_scaled; | |
70a89a66 VP |
2914 | account_group_system_time(p, cputime); |
2915 | ||
2916 | /* Add system time to cpustat. */ | |
612ef28a | 2917 | task_group_account_field(p, index, (__force u64) cputime); |
70a89a66 VP |
2918 | |
2919 | /* Account for system time used */ | |
2920 | acct_update_integrals(p); | |
2921 | } | |
2922 | ||
1da177e4 LT |
2923 | /* |
2924 | * Account system cpu time to a process. | |
2925 | * @p: the process that the cpu time gets accounted to | |
2926 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
2927 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 2928 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
2929 | */ |
2930 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 2931 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 | 2932 | { |
3292beb3 | 2933 | int index; |
1da177e4 | 2934 | |
983ed7a6 | 2935 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 2936 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
2937 | return; |
2938 | } | |
94886b84 | 2939 | |
1da177e4 | 2940 | if (hardirq_count() - hardirq_offset) |
3292beb3 | 2941 | index = CPUTIME_IRQ; |
75e1056f | 2942 | else if (in_serving_softirq()) |
3292beb3 | 2943 | index = CPUTIME_SOFTIRQ; |
1da177e4 | 2944 | else |
3292beb3 | 2945 | index = CPUTIME_SYSTEM; |
ef12fefa | 2946 | |
3292beb3 | 2947 | __account_system_time(p, cputime, cputime_scaled, index); |
1da177e4 LT |
2948 | } |
2949 | ||
c66f08be | 2950 | /* |
1da177e4 | 2951 | * Account for involuntary wait time. |
544b4a1f | 2952 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 2953 | */ |
79741dd3 | 2954 | void account_steal_time(cputime_t cputime) |
c66f08be | 2955 | { |
3292beb3 | 2956 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
79741dd3 | 2957 | |
612ef28a | 2958 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; |
c66f08be MN |
2959 | } |
2960 | ||
1da177e4 | 2961 | /* |
79741dd3 MS |
2962 | * Account for idle time. |
2963 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 2964 | */ |
79741dd3 | 2965 | void account_idle_time(cputime_t cputime) |
1da177e4 | 2966 | { |
3292beb3 | 2967 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
70b97a7f | 2968 | struct rq *rq = this_rq(); |
1da177e4 | 2969 | |
79741dd3 | 2970 | if (atomic_read(&rq->nr_iowait) > 0) |
612ef28a | 2971 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; |
79741dd3 | 2972 | else |
612ef28a | 2973 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; |
1da177e4 LT |
2974 | } |
2975 | ||
e6e6685a GC |
2976 | static __always_inline bool steal_account_process_tick(void) |
2977 | { | |
2978 | #ifdef CONFIG_PARAVIRT | |
c5905afb | 2979 | if (static_key_false(¶virt_steal_enabled)) { |
e6e6685a GC |
2980 | u64 steal, st = 0; |
2981 | ||
2982 | steal = paravirt_steal_clock(smp_processor_id()); | |
2983 | steal -= this_rq()->prev_steal_time; | |
2984 | ||
2985 | st = steal_ticks(steal); | |
2986 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
2987 | ||
2988 | account_steal_time(st); | |
2989 | return st; | |
2990 | } | |
2991 | #endif | |
2992 | return false; | |
2993 | } | |
2994 | ||
79741dd3 MS |
2995 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
2996 | ||
abb74cef VP |
2997 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
2998 | /* | |
2999 | * Account a tick to a process and cpustat | |
3000 | * @p: the process that the cpu time gets accounted to | |
3001 | * @user_tick: is the tick from userspace | |
3002 | * @rq: the pointer to rq | |
3003 | * | |
3004 | * Tick demultiplexing follows the order | |
3005 | * - pending hardirq update | |
3006 | * - pending softirq update | |
3007 | * - user_time | |
3008 | * - idle_time | |
3009 | * - system time | |
3010 | * - check for guest_time | |
3011 | * - else account as system_time | |
3012 | * | |
3013 | * Check for hardirq is done both for system and user time as there is | |
3014 | * no timer going off while we are on hardirq and hence we may never get an | |
3015 | * opportunity to update it solely in system time. | |
3016 | * p->stime and friends are only updated on system time and not on irq | |
3017 | * softirq as those do not count in task exec_runtime any more. | |
3018 | */ | |
3019 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3020 | struct rq *rq) | |
3021 | { | |
3022 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3292beb3 | 3023 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
abb74cef | 3024 | |
e6e6685a GC |
3025 | if (steal_account_process_tick()) |
3026 | return; | |
3027 | ||
abb74cef | 3028 | if (irqtime_account_hi_update()) { |
612ef28a | 3029 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; |
abb74cef | 3030 | } else if (irqtime_account_si_update()) { |
612ef28a | 3031 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; |
414bee9b VP |
3032 | } else if (this_cpu_ksoftirqd() == p) { |
3033 | /* | |
3034 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3035 | * So, we have to handle it separately here. | |
3036 | * Also, p->stime needs to be updated for ksoftirqd. | |
3037 | */ | |
3038 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 3039 | CPUTIME_SOFTIRQ); |
abb74cef VP |
3040 | } else if (user_tick) { |
3041 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3042 | } else if (p == rq->idle) { | |
3043 | account_idle_time(cputime_one_jiffy); | |
3044 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3045 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3046 | } else { | |
3047 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3292beb3 | 3048 | CPUTIME_SYSTEM); |
abb74cef VP |
3049 | } |
3050 | } | |
3051 | ||
3052 | static void irqtime_account_idle_ticks(int ticks) | |
3053 | { | |
3054 | int i; | |
3055 | struct rq *rq = this_rq(); | |
3056 | ||
3057 | for (i = 0; i < ticks; i++) | |
3058 | irqtime_account_process_tick(current, 0, rq); | |
3059 | } | |
544b4a1f | 3060 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
3061 | static void irqtime_account_idle_ticks(int ticks) {} |
3062 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3063 | struct rq *rq) {} | |
544b4a1f | 3064 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
3065 | |
3066 | /* | |
3067 | * Account a single tick of cpu time. | |
3068 | * @p: the process that the cpu time gets accounted to | |
3069 | * @user_tick: indicates if the tick is a user or a system tick | |
3070 | */ | |
3071 | void account_process_tick(struct task_struct *p, int user_tick) | |
3072 | { | |
a42548a1 | 3073 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3074 | struct rq *rq = this_rq(); |
3075 | ||
abb74cef VP |
3076 | if (sched_clock_irqtime) { |
3077 | irqtime_account_process_tick(p, user_tick, rq); | |
3078 | return; | |
3079 | } | |
3080 | ||
e6e6685a GC |
3081 | if (steal_account_process_tick()) |
3082 | return; | |
3083 | ||
79741dd3 | 3084 | if (user_tick) |
a42548a1 | 3085 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3086 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3087 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3088 | one_jiffy_scaled); |
3089 | else | |
a42548a1 | 3090 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3091 | } |
3092 | ||
3093 | /* | |
3094 | * Account multiple ticks of steal time. | |
3095 | * @p: the process from which the cpu time has been stolen | |
3096 | * @ticks: number of stolen ticks | |
3097 | */ | |
3098 | void account_steal_ticks(unsigned long ticks) | |
3099 | { | |
3100 | account_steal_time(jiffies_to_cputime(ticks)); | |
3101 | } | |
3102 | ||
3103 | /* | |
3104 | * Account multiple ticks of idle time. | |
3105 | * @ticks: number of stolen ticks | |
3106 | */ | |
3107 | void account_idle_ticks(unsigned long ticks) | |
3108 | { | |
abb74cef VP |
3109 | |
3110 | if (sched_clock_irqtime) { | |
3111 | irqtime_account_idle_ticks(ticks); | |
3112 | return; | |
3113 | } | |
3114 | ||
79741dd3 | 3115 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3116 | } |
3117 | ||
79741dd3 MS |
3118 | #endif |
3119 | ||
49048622 BS |
3120 | /* |
3121 | * Use precise platform statistics if available: | |
3122 | */ | |
3123 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3124 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3125 | { |
d99ca3b9 HS |
3126 | *ut = p->utime; |
3127 | *st = p->stime; | |
49048622 BS |
3128 | } |
3129 | ||
0cf55e1e | 3130 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3131 | { |
0cf55e1e HS |
3132 | struct task_cputime cputime; |
3133 | ||
3134 | thread_group_cputime(p, &cputime); | |
3135 | ||
3136 | *ut = cputime.utime; | |
3137 | *st = cputime.stime; | |
49048622 BS |
3138 | } |
3139 | #else | |
761b1d26 HS |
3140 | |
3141 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3142 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3143 | #endif |
3144 | ||
bea6832c SG |
3145 | static cputime_t scale_utime(cputime_t utime, cputime_t rtime, cputime_t total) |
3146 | { | |
3147 | u64 temp = (__force u64) rtime; | |
3148 | ||
3149 | temp *= (__force u64) utime; | |
3150 | ||
3151 | if (sizeof(cputime_t) == 4) | |
3152 | temp = div_u64(temp, (__force u32) total); | |
3153 | else | |
3154 | temp = div64_u64(temp, (__force u64) total); | |
3155 | ||
3156 | return (__force cputime_t) temp; | |
3157 | } | |
3158 | ||
d180c5bc | 3159 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3160 | { |
64861634 | 3161 | cputime_t rtime, utime = p->utime, total = utime + p->stime; |
49048622 BS |
3162 | |
3163 | /* | |
3164 | * Use CFS's precise accounting: | |
3165 | */ | |
d180c5bc | 3166 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 | 3167 | |
bea6832c SG |
3168 | if (total) |
3169 | utime = scale_utime(utime, rtime, total); | |
3170 | else | |
d180c5bc | 3171 | utime = rtime; |
49048622 | 3172 | |
d180c5bc HS |
3173 | /* |
3174 | * Compare with previous values, to keep monotonicity: | |
3175 | */ | |
761b1d26 | 3176 | p->prev_utime = max(p->prev_utime, utime); |
64861634 | 3177 | p->prev_stime = max(p->prev_stime, rtime - p->prev_utime); |
49048622 | 3178 | |
d99ca3b9 HS |
3179 | *ut = p->prev_utime; |
3180 | *st = p->prev_stime; | |
49048622 BS |
3181 | } |
3182 | ||
0cf55e1e HS |
3183 | /* |
3184 | * Must be called with siglock held. | |
3185 | */ | |
3186 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3187 | { |
0cf55e1e HS |
3188 | struct signal_struct *sig = p->signal; |
3189 | struct task_cputime cputime; | |
3190 | cputime_t rtime, utime, total; | |
49048622 | 3191 | |
0cf55e1e | 3192 | thread_group_cputime(p, &cputime); |
49048622 | 3193 | |
64861634 | 3194 | total = cputime.utime + cputime.stime; |
0cf55e1e | 3195 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); |
49048622 | 3196 | |
bea6832c SG |
3197 | if (total) |
3198 | utime = scale_utime(cputime.utime, rtime, total); | |
3199 | else | |
0cf55e1e HS |
3200 | utime = rtime; |
3201 | ||
3202 | sig->prev_utime = max(sig->prev_utime, utime); | |
64861634 | 3203 | sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime); |
0cf55e1e HS |
3204 | |
3205 | *ut = sig->prev_utime; | |
3206 | *st = sig->prev_stime; | |
49048622 | 3207 | } |
49048622 | 3208 | #endif |
49048622 | 3209 | |
7835b98b CL |
3210 | /* |
3211 | * This function gets called by the timer code, with HZ frequency. | |
3212 | * We call it with interrupts disabled. | |
7835b98b CL |
3213 | */ |
3214 | void scheduler_tick(void) | |
3215 | { | |
7835b98b CL |
3216 | int cpu = smp_processor_id(); |
3217 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3218 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3219 | |
3220 | sched_clock_tick(); | |
dd41f596 | 3221 | |
05fa785c | 3222 | raw_spin_lock(&rq->lock); |
3e51f33f | 3223 | update_rq_clock(rq); |
fdf3e95d | 3224 | update_cpu_load_active(rq); |
fa85ae24 | 3225 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3226 | raw_spin_unlock(&rq->lock); |
7835b98b | 3227 | |
e9d2b064 | 3228 | perf_event_task_tick(); |
e220d2dc | 3229 | |
e418e1c2 | 3230 | #ifdef CONFIG_SMP |
6eb57e0d | 3231 | rq->idle_balance = idle_cpu(cpu); |
dd41f596 | 3232 | trigger_load_balance(rq, cpu); |
e418e1c2 | 3233 | #endif |
1da177e4 LT |
3234 | } |
3235 | ||
132380a0 | 3236 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3237 | { |
3238 | if (in_lock_functions(addr)) { | |
3239 | addr = CALLER_ADDR2; | |
3240 | if (in_lock_functions(addr)) | |
3241 | addr = CALLER_ADDR3; | |
3242 | } | |
3243 | return addr; | |
3244 | } | |
1da177e4 | 3245 | |
7e49fcce SR |
3246 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3247 | defined(CONFIG_PREEMPT_TRACER)) | |
3248 | ||
43627582 | 3249 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3250 | { |
6cd8a4bb | 3251 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3252 | /* |
3253 | * Underflow? | |
3254 | */ | |
9a11b49a IM |
3255 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3256 | return; | |
6cd8a4bb | 3257 | #endif |
1da177e4 | 3258 | preempt_count() += val; |
6cd8a4bb | 3259 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3260 | /* |
3261 | * Spinlock count overflowing soon? | |
3262 | */ | |
33859f7f MOS |
3263 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3264 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3265 | #endif |
3266 | if (preempt_count() == val) | |
3267 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3268 | } |
3269 | EXPORT_SYMBOL(add_preempt_count); | |
3270 | ||
43627582 | 3271 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3272 | { |
6cd8a4bb | 3273 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3274 | /* |
3275 | * Underflow? | |
3276 | */ | |
01e3eb82 | 3277 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3278 | return; |
1da177e4 LT |
3279 | /* |
3280 | * Is the spinlock portion underflowing? | |
3281 | */ | |
9a11b49a IM |
3282 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3283 | !(preempt_count() & PREEMPT_MASK))) | |
3284 | return; | |
6cd8a4bb | 3285 | #endif |
9a11b49a | 3286 | |
6cd8a4bb SR |
3287 | if (preempt_count() == val) |
3288 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3289 | preempt_count() -= val; |
3290 | } | |
3291 | EXPORT_SYMBOL(sub_preempt_count); | |
3292 | ||
3293 | #endif | |
3294 | ||
3295 | /* | |
dd41f596 | 3296 | * Print scheduling while atomic bug: |
1da177e4 | 3297 | */ |
dd41f596 | 3298 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3299 | { |
664dfa65 DJ |
3300 | if (oops_in_progress) |
3301 | return; | |
3302 | ||
3df0fc5b PZ |
3303 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3304 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3305 | |
dd41f596 | 3306 | debug_show_held_locks(prev); |
e21f5b15 | 3307 | print_modules(); |
dd41f596 IM |
3308 | if (irqs_disabled()) |
3309 | print_irqtrace_events(prev); | |
6135fc1e | 3310 | dump_stack(); |
1c2927f1 | 3311 | add_taint(TAINT_WARN); |
dd41f596 | 3312 | } |
1da177e4 | 3313 | |
dd41f596 IM |
3314 | /* |
3315 | * Various schedule()-time debugging checks and statistics: | |
3316 | */ | |
3317 | static inline void schedule_debug(struct task_struct *prev) | |
3318 | { | |
1da177e4 | 3319 | /* |
41a2d6cf | 3320 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3321 | * schedule() atomically, we ignore that path for now. |
3322 | * Otherwise, whine if we are scheduling when we should not be. | |
3323 | */ | |
3f33a7ce | 3324 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 | 3325 | __schedule_bug(prev); |
b3fbab05 | 3326 | rcu_sleep_check(); |
dd41f596 | 3327 | |
1da177e4 LT |
3328 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3329 | ||
2d72376b | 3330 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
3331 | } |
3332 | ||
6cecd084 | 3333 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3334 | { |
61eadef6 | 3335 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 3336 | update_rq_clock(rq); |
6cecd084 | 3337 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3338 | } |
3339 | ||
dd41f596 IM |
3340 | /* |
3341 | * Pick up the highest-prio task: | |
3342 | */ | |
3343 | static inline struct task_struct * | |
b67802ea | 3344 | pick_next_task(struct rq *rq) |
dd41f596 | 3345 | { |
5522d5d5 | 3346 | const struct sched_class *class; |
dd41f596 | 3347 | struct task_struct *p; |
1da177e4 LT |
3348 | |
3349 | /* | |
dd41f596 IM |
3350 | * Optimization: we know that if all tasks are in |
3351 | * the fair class we can call that function directly: | |
1da177e4 | 3352 | */ |
953bfcd1 | 3353 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
fb8d4724 | 3354 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3355 | if (likely(p)) |
3356 | return p; | |
1da177e4 LT |
3357 | } |
3358 | ||
34f971f6 | 3359 | for_each_class(class) { |
fb8d4724 | 3360 | p = class->pick_next_task(rq); |
dd41f596 IM |
3361 | if (p) |
3362 | return p; | |
dd41f596 | 3363 | } |
34f971f6 PZ |
3364 | |
3365 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3366 | } |
1da177e4 | 3367 | |
dd41f596 | 3368 | /* |
c259e01a | 3369 | * __schedule() is the main scheduler function. |
dd41f596 | 3370 | */ |
c259e01a | 3371 | static void __sched __schedule(void) |
dd41f596 IM |
3372 | { |
3373 | struct task_struct *prev, *next; | |
67ca7bde | 3374 | unsigned long *switch_count; |
dd41f596 | 3375 | struct rq *rq; |
31656519 | 3376 | int cpu; |
dd41f596 | 3377 | |
ff743345 PZ |
3378 | need_resched: |
3379 | preempt_disable(); | |
dd41f596 IM |
3380 | cpu = smp_processor_id(); |
3381 | rq = cpu_rq(cpu); | |
25502a6c | 3382 | rcu_note_context_switch(cpu); |
dd41f596 | 3383 | prev = rq->curr; |
dd41f596 | 3384 | |
dd41f596 | 3385 | schedule_debug(prev); |
1da177e4 | 3386 | |
31656519 | 3387 | if (sched_feat(HRTICK)) |
f333fdc9 | 3388 | hrtick_clear(rq); |
8f4d37ec | 3389 | |
05fa785c | 3390 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3391 | |
246d86b5 | 3392 | switch_count = &prev->nivcsw; |
1da177e4 | 3393 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3394 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3395 | prev->state = TASK_RUNNING; |
21aa9af0 | 3396 | } else { |
2acca55e PZ |
3397 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3398 | prev->on_rq = 0; | |
3399 | ||
21aa9af0 | 3400 | /* |
2acca55e PZ |
3401 | * If a worker went to sleep, notify and ask workqueue |
3402 | * whether it wants to wake up a task to maintain | |
3403 | * concurrency. | |
21aa9af0 TH |
3404 | */ |
3405 | if (prev->flags & PF_WQ_WORKER) { | |
3406 | struct task_struct *to_wakeup; | |
3407 | ||
3408 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3409 | if (to_wakeup) | |
3410 | try_to_wake_up_local(to_wakeup); | |
3411 | } | |
21aa9af0 | 3412 | } |
dd41f596 | 3413 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3414 | } |
3415 | ||
3f029d3c | 3416 | pre_schedule(rq, prev); |
f65eda4f | 3417 | |
dd41f596 | 3418 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3419 | idle_balance(cpu, rq); |
1da177e4 | 3420 | |
df1c99d4 | 3421 | put_prev_task(rq, prev); |
b67802ea | 3422 | next = pick_next_task(rq); |
f26f9aff MG |
3423 | clear_tsk_need_resched(prev); |
3424 | rq->skip_clock_update = 0; | |
1da177e4 | 3425 | |
1da177e4 | 3426 | if (likely(prev != next)) { |
1da177e4 LT |
3427 | rq->nr_switches++; |
3428 | rq->curr = next; | |
3429 | ++*switch_count; | |
3430 | ||
dd41f596 | 3431 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3432 | /* |
246d86b5 ON |
3433 | * The context switch have flipped the stack from under us |
3434 | * and restored the local variables which were saved when | |
3435 | * this task called schedule() in the past. prev == current | |
3436 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3437 | */ |
3438 | cpu = smp_processor_id(); | |
3439 | rq = cpu_rq(cpu); | |
1da177e4 | 3440 | } else |
05fa785c | 3441 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3442 | |
3f029d3c | 3443 | post_schedule(rq); |
1da177e4 | 3444 | |
ba74c144 | 3445 | sched_preempt_enable_no_resched(); |
ff743345 | 3446 | if (need_resched()) |
1da177e4 LT |
3447 | goto need_resched; |
3448 | } | |
c259e01a | 3449 | |
9c40cef2 TG |
3450 | static inline void sched_submit_work(struct task_struct *tsk) |
3451 | { | |
3c7d5184 | 3452 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3453 | return; |
3454 | /* | |
3455 | * If we are going to sleep and we have plugged IO queued, | |
3456 | * make sure to submit it to avoid deadlocks. | |
3457 | */ | |
3458 | if (blk_needs_flush_plug(tsk)) | |
3459 | blk_schedule_flush_plug(tsk); | |
3460 | } | |
3461 | ||
6ebbe7a0 | 3462 | asmlinkage void __sched schedule(void) |
c259e01a | 3463 | { |
9c40cef2 TG |
3464 | struct task_struct *tsk = current; |
3465 | ||
3466 | sched_submit_work(tsk); | |
c259e01a TG |
3467 | __schedule(); |
3468 | } | |
1da177e4 LT |
3469 | EXPORT_SYMBOL(schedule); |
3470 | ||
c5491ea7 TG |
3471 | /** |
3472 | * schedule_preempt_disabled - called with preemption disabled | |
3473 | * | |
3474 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3475 | */ | |
3476 | void __sched schedule_preempt_disabled(void) | |
3477 | { | |
ba74c144 | 3478 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3479 | schedule(); |
3480 | preempt_disable(); | |
3481 | } | |
3482 | ||
c08f7829 | 3483 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 3484 | |
c6eb3dda PZ |
3485 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
3486 | { | |
c6eb3dda | 3487 | if (lock->owner != owner) |
307bf980 | 3488 | return false; |
0d66bf6d PZ |
3489 | |
3490 | /* | |
c6eb3dda PZ |
3491 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
3492 | * lock->owner still matches owner, if that fails, owner might | |
3493 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
3494 | * ensures the memory stays valid. | |
0d66bf6d | 3495 | */ |
c6eb3dda | 3496 | barrier(); |
0d66bf6d | 3497 | |
307bf980 | 3498 | return owner->on_cpu; |
c6eb3dda | 3499 | } |
0d66bf6d | 3500 | |
c6eb3dda PZ |
3501 | /* |
3502 | * Look out! "owner" is an entirely speculative pointer | |
3503 | * access and not reliable. | |
3504 | */ | |
3505 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
3506 | { | |
3507 | if (!sched_feat(OWNER_SPIN)) | |
3508 | return 0; | |
0d66bf6d | 3509 | |
307bf980 | 3510 | rcu_read_lock(); |
c6eb3dda PZ |
3511 | while (owner_running(lock, owner)) { |
3512 | if (need_resched()) | |
307bf980 | 3513 | break; |
0d66bf6d | 3514 | |
335d7afb | 3515 | arch_mutex_cpu_relax(); |
0d66bf6d | 3516 | } |
307bf980 | 3517 | rcu_read_unlock(); |
4b402210 | 3518 | |
c6eb3dda | 3519 | /* |
307bf980 TG |
3520 | * We break out the loop above on need_resched() and when the |
3521 | * owner changed, which is a sign for heavy contention. Return | |
3522 | * success only when lock->owner is NULL. | |
c6eb3dda | 3523 | */ |
307bf980 | 3524 | return lock->owner == NULL; |
0d66bf6d PZ |
3525 | } |
3526 | #endif | |
3527 | ||
1da177e4 LT |
3528 | #ifdef CONFIG_PREEMPT |
3529 | /* | |
2ed6e34f | 3530 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3531 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3532 | * occur there and call schedule directly. |
3533 | */ | |
d1f74e20 | 3534 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3535 | { |
3536 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3537 | |
1da177e4 LT |
3538 | /* |
3539 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3540 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3541 | */ |
beed33a8 | 3542 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3543 | return; |
3544 | ||
3a5c359a | 3545 | do { |
d1f74e20 | 3546 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
c259e01a | 3547 | __schedule(); |
d1f74e20 | 3548 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3549 | |
3a5c359a AK |
3550 | /* |
3551 | * Check again in case we missed a preemption opportunity | |
3552 | * between schedule and now. | |
3553 | */ | |
3554 | barrier(); | |
5ed0cec0 | 3555 | } while (need_resched()); |
1da177e4 | 3556 | } |
1da177e4 LT |
3557 | EXPORT_SYMBOL(preempt_schedule); |
3558 | ||
3559 | /* | |
2ed6e34f | 3560 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3561 | * off of irq context. |
3562 | * Note, that this is called and return with irqs disabled. This will | |
3563 | * protect us against recursive calling from irq. | |
3564 | */ | |
3565 | asmlinkage void __sched preempt_schedule_irq(void) | |
3566 | { | |
3567 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3568 | |
2ed6e34f | 3569 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3570 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3571 | ||
3a5c359a AK |
3572 | do { |
3573 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3574 | local_irq_enable(); |
c259e01a | 3575 | __schedule(); |
3a5c359a | 3576 | local_irq_disable(); |
3a5c359a | 3577 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3578 | |
3a5c359a AK |
3579 | /* |
3580 | * Check again in case we missed a preemption opportunity | |
3581 | * between schedule and now. | |
3582 | */ | |
3583 | barrier(); | |
5ed0cec0 | 3584 | } while (need_resched()); |
1da177e4 LT |
3585 | } |
3586 | ||
3587 | #endif /* CONFIG_PREEMPT */ | |
3588 | ||
63859d4f | 3589 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3590 | void *key) |
1da177e4 | 3591 | { |
63859d4f | 3592 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3593 | } |
1da177e4 LT |
3594 | EXPORT_SYMBOL(default_wake_function); |
3595 | ||
3596 | /* | |
41a2d6cf IM |
3597 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3598 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3599 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3600 | * | |
3601 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3602 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3603 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3604 | */ | |
78ddb08f | 3605 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3606 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3607 | { |
2e45874c | 3608 | wait_queue_t *curr, *next; |
1da177e4 | 3609 | |
2e45874c | 3610 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3611 | unsigned flags = curr->flags; |
3612 | ||
63859d4f | 3613 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3614 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3615 | break; |
3616 | } | |
3617 | } | |
3618 | ||
3619 | /** | |
3620 | * __wake_up - wake up threads blocked on a waitqueue. | |
3621 | * @q: the waitqueue | |
3622 | * @mode: which threads | |
3623 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3624 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3625 | * |
3626 | * It may be assumed that this function implies a write memory barrier before | |
3627 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3628 | */ |
7ad5b3a5 | 3629 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3630 | int nr_exclusive, void *key) |
1da177e4 LT |
3631 | { |
3632 | unsigned long flags; | |
3633 | ||
3634 | spin_lock_irqsave(&q->lock, flags); | |
3635 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3636 | spin_unlock_irqrestore(&q->lock, flags); | |
3637 | } | |
1da177e4 LT |
3638 | EXPORT_SYMBOL(__wake_up); |
3639 | ||
3640 | /* | |
3641 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3642 | */ | |
63b20011 | 3643 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) |
1da177e4 | 3644 | { |
63b20011 | 3645 | __wake_up_common(q, mode, nr, 0, NULL); |
1da177e4 | 3646 | } |
22c43c81 | 3647 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3648 | |
4ede816a DL |
3649 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3650 | { | |
3651 | __wake_up_common(q, mode, 1, 0, key); | |
3652 | } | |
bf294b41 | 3653 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 3654 | |
1da177e4 | 3655 | /** |
4ede816a | 3656 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3657 | * @q: the waitqueue |
3658 | * @mode: which threads | |
3659 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3660 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3661 | * |
3662 | * The sync wakeup differs that the waker knows that it will schedule | |
3663 | * away soon, so while the target thread will be woken up, it will not | |
3664 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3665 | * with each other. This can prevent needless bouncing between CPUs. | |
3666 | * | |
3667 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3668 | * |
3669 | * It may be assumed that this function implies a write memory barrier before | |
3670 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3671 | */ |
4ede816a DL |
3672 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3673 | int nr_exclusive, void *key) | |
1da177e4 LT |
3674 | { |
3675 | unsigned long flags; | |
7d478721 | 3676 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3677 | |
3678 | if (unlikely(!q)) | |
3679 | return; | |
3680 | ||
3681 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3682 | wake_flags = 0; |
1da177e4 LT |
3683 | |
3684 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3685 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3686 | spin_unlock_irqrestore(&q->lock, flags); |
3687 | } | |
4ede816a DL |
3688 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3689 | ||
3690 | /* | |
3691 | * __wake_up_sync - see __wake_up_sync_key() | |
3692 | */ | |
3693 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3694 | { | |
3695 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3696 | } | |
1da177e4 LT |
3697 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3698 | ||
65eb3dc6 KD |
3699 | /** |
3700 | * complete: - signals a single thread waiting on this completion | |
3701 | * @x: holds the state of this particular completion | |
3702 | * | |
3703 | * This will wake up a single thread waiting on this completion. Threads will be | |
3704 | * awakened in the same order in which they were queued. | |
3705 | * | |
3706 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3707 | * |
3708 | * It may be assumed that this function implies a write memory barrier before | |
3709 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3710 | */ |
b15136e9 | 3711 | void complete(struct completion *x) |
1da177e4 LT |
3712 | { |
3713 | unsigned long flags; | |
3714 | ||
3715 | spin_lock_irqsave(&x->wait.lock, flags); | |
3716 | x->done++; | |
d9514f6c | 3717 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3718 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3719 | } | |
3720 | EXPORT_SYMBOL(complete); | |
3721 | ||
65eb3dc6 KD |
3722 | /** |
3723 | * complete_all: - signals all threads waiting on this completion | |
3724 | * @x: holds the state of this particular completion | |
3725 | * | |
3726 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3727 | * |
3728 | * It may be assumed that this function implies a write memory barrier before | |
3729 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3730 | */ |
b15136e9 | 3731 | void complete_all(struct completion *x) |
1da177e4 LT |
3732 | { |
3733 | unsigned long flags; | |
3734 | ||
3735 | spin_lock_irqsave(&x->wait.lock, flags); | |
3736 | x->done += UINT_MAX/2; | |
d9514f6c | 3737 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3738 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3739 | } | |
3740 | EXPORT_SYMBOL(complete_all); | |
3741 | ||
8cbbe86d AK |
3742 | static inline long __sched |
3743 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3744 | { |
1da177e4 LT |
3745 | if (!x->done) { |
3746 | DECLARE_WAITQUEUE(wait, current); | |
3747 | ||
a93d2f17 | 3748 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3749 | do { |
94d3d824 | 3750 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3751 | timeout = -ERESTARTSYS; |
3752 | break; | |
8cbbe86d AK |
3753 | } |
3754 | __set_current_state(state); | |
1da177e4 LT |
3755 | spin_unlock_irq(&x->wait.lock); |
3756 | timeout = schedule_timeout(timeout); | |
3757 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3758 | } while (!x->done && timeout); |
1da177e4 | 3759 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3760 | if (!x->done) |
3761 | return timeout; | |
1da177e4 LT |
3762 | } |
3763 | x->done--; | |
ea71a546 | 3764 | return timeout ?: 1; |
1da177e4 | 3765 | } |
1da177e4 | 3766 | |
8cbbe86d AK |
3767 | static long __sched |
3768 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3769 | { |
1da177e4 LT |
3770 | might_sleep(); |
3771 | ||
3772 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3773 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3774 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3775 | return timeout; |
3776 | } | |
1da177e4 | 3777 | |
65eb3dc6 KD |
3778 | /** |
3779 | * wait_for_completion: - waits for completion of a task | |
3780 | * @x: holds the state of this particular completion | |
3781 | * | |
3782 | * This waits to be signaled for completion of a specific task. It is NOT | |
3783 | * interruptible and there is no timeout. | |
3784 | * | |
3785 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3786 | * and interrupt capability. Also see complete(). | |
3787 | */ | |
b15136e9 | 3788 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3789 | { |
3790 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3791 | } |
8cbbe86d | 3792 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3793 | |
65eb3dc6 KD |
3794 | /** |
3795 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
3796 | * @x: holds the state of this particular completion | |
3797 | * @timeout: timeout value in jiffies | |
3798 | * | |
3799 | * This waits for either a completion of a specific task to be signaled or for a | |
3800 | * specified timeout to expire. The timeout is in jiffies. It is not | |
3801 | * interruptible. | |
c6dc7f05 BF |
3802 | * |
3803 | * The return value is 0 if timed out, and positive (at least 1, or number of | |
3804 | * jiffies left till timeout) if completed. | |
65eb3dc6 | 3805 | */ |
b15136e9 | 3806 | unsigned long __sched |
8cbbe86d | 3807 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3808 | { |
8cbbe86d | 3809 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3810 | } |
8cbbe86d | 3811 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3812 | |
65eb3dc6 KD |
3813 | /** |
3814 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
3815 | * @x: holds the state of this particular completion | |
3816 | * | |
3817 | * This waits for completion of a specific task to be signaled. It is | |
3818 | * interruptible. | |
c6dc7f05 BF |
3819 | * |
3820 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3821 | */ |
8cbbe86d | 3822 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3823 | { |
51e97990 AK |
3824 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3825 | if (t == -ERESTARTSYS) | |
3826 | return t; | |
3827 | return 0; | |
0fec171c | 3828 | } |
8cbbe86d | 3829 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3830 | |
65eb3dc6 KD |
3831 | /** |
3832 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
3833 | * @x: holds the state of this particular completion | |
3834 | * @timeout: timeout value in jiffies | |
3835 | * | |
3836 | * This waits for either a completion of a specific task to be signaled or for a | |
3837 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
c6dc7f05 BF |
3838 | * |
3839 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3840 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
65eb3dc6 | 3841 | */ |
6bf41237 | 3842 | long __sched |
8cbbe86d AK |
3843 | wait_for_completion_interruptible_timeout(struct completion *x, |
3844 | unsigned long timeout) | |
0fec171c | 3845 | { |
8cbbe86d | 3846 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3847 | } |
8cbbe86d | 3848 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3849 | |
65eb3dc6 KD |
3850 | /** |
3851 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
3852 | * @x: holds the state of this particular completion | |
3853 | * | |
3854 | * This waits to be signaled for completion of a specific task. It can be | |
3855 | * interrupted by a kill signal. | |
c6dc7f05 BF |
3856 | * |
3857 | * The return value is -ERESTARTSYS if interrupted, 0 if completed. | |
65eb3dc6 | 3858 | */ |
009e577e MW |
3859 | int __sched wait_for_completion_killable(struct completion *x) |
3860 | { | |
3861 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
3862 | if (t == -ERESTARTSYS) | |
3863 | return t; | |
3864 | return 0; | |
3865 | } | |
3866 | EXPORT_SYMBOL(wait_for_completion_killable); | |
3867 | ||
0aa12fb4 SW |
3868 | /** |
3869 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
3870 | * @x: holds the state of this particular completion | |
3871 | * @timeout: timeout value in jiffies | |
3872 | * | |
3873 | * This waits for either a completion of a specific task to be | |
3874 | * signaled or for a specified timeout to expire. It can be | |
3875 | * interrupted by a kill signal. The timeout is in jiffies. | |
c6dc7f05 BF |
3876 | * |
3877 | * The return value is -ERESTARTSYS if interrupted, 0 if timed out, | |
3878 | * positive (at least 1, or number of jiffies left till timeout) if completed. | |
0aa12fb4 | 3879 | */ |
6bf41237 | 3880 | long __sched |
0aa12fb4 SW |
3881 | wait_for_completion_killable_timeout(struct completion *x, |
3882 | unsigned long timeout) | |
3883 | { | |
3884 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
3885 | } | |
3886 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
3887 | ||
be4de352 DC |
3888 | /** |
3889 | * try_wait_for_completion - try to decrement a completion without blocking | |
3890 | * @x: completion structure | |
3891 | * | |
3892 | * Returns: 0 if a decrement cannot be done without blocking | |
3893 | * 1 if a decrement succeeded. | |
3894 | * | |
3895 | * If a completion is being used as a counting completion, | |
3896 | * attempt to decrement the counter without blocking. This | |
3897 | * enables us to avoid waiting if the resource the completion | |
3898 | * is protecting is not available. | |
3899 | */ | |
3900 | bool try_wait_for_completion(struct completion *x) | |
3901 | { | |
7539a3b3 | 3902 | unsigned long flags; |
be4de352 DC |
3903 | int ret = 1; |
3904 | ||
7539a3b3 | 3905 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3906 | if (!x->done) |
3907 | ret = 0; | |
3908 | else | |
3909 | x->done--; | |
7539a3b3 | 3910 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3911 | return ret; |
3912 | } | |
3913 | EXPORT_SYMBOL(try_wait_for_completion); | |
3914 | ||
3915 | /** | |
3916 | * completion_done - Test to see if a completion has any waiters | |
3917 | * @x: completion structure | |
3918 | * | |
3919 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
3920 | * 1 if there are no waiters. | |
3921 | * | |
3922 | */ | |
3923 | bool completion_done(struct completion *x) | |
3924 | { | |
7539a3b3 | 3925 | unsigned long flags; |
be4de352 DC |
3926 | int ret = 1; |
3927 | ||
7539a3b3 | 3928 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
3929 | if (!x->done) |
3930 | ret = 0; | |
7539a3b3 | 3931 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
3932 | return ret; |
3933 | } | |
3934 | EXPORT_SYMBOL(completion_done); | |
3935 | ||
8cbbe86d AK |
3936 | static long __sched |
3937 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3938 | { |
0fec171c IM |
3939 | unsigned long flags; |
3940 | wait_queue_t wait; | |
3941 | ||
3942 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3943 | |
8cbbe86d | 3944 | __set_current_state(state); |
1da177e4 | 3945 | |
8cbbe86d AK |
3946 | spin_lock_irqsave(&q->lock, flags); |
3947 | __add_wait_queue(q, &wait); | |
3948 | spin_unlock(&q->lock); | |
3949 | timeout = schedule_timeout(timeout); | |
3950 | spin_lock_irq(&q->lock); | |
3951 | __remove_wait_queue(q, &wait); | |
3952 | spin_unlock_irqrestore(&q->lock, flags); | |
3953 | ||
3954 | return timeout; | |
3955 | } | |
3956 | ||
3957 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3958 | { | |
3959 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3960 | } |
1da177e4 LT |
3961 | EXPORT_SYMBOL(interruptible_sleep_on); |
3962 | ||
0fec171c | 3963 | long __sched |
95cdf3b7 | 3964 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3965 | { |
8cbbe86d | 3966 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3967 | } |
1da177e4 LT |
3968 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3969 | ||
0fec171c | 3970 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3971 | { |
8cbbe86d | 3972 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3973 | } |
1da177e4 LT |
3974 | EXPORT_SYMBOL(sleep_on); |
3975 | ||
0fec171c | 3976 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3977 | { |
8cbbe86d | 3978 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3979 | } |
1da177e4 LT |
3980 | EXPORT_SYMBOL(sleep_on_timeout); |
3981 | ||
b29739f9 IM |
3982 | #ifdef CONFIG_RT_MUTEXES |
3983 | ||
3984 | /* | |
3985 | * rt_mutex_setprio - set the current priority of a task | |
3986 | * @p: task | |
3987 | * @prio: prio value (kernel-internal form) | |
3988 | * | |
3989 | * This function changes the 'effective' priority of a task. It does | |
3990 | * not touch ->normal_prio like __setscheduler(). | |
3991 | * | |
3992 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3993 | */ | |
36c8b586 | 3994 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3995 | { |
83b699ed | 3996 | int oldprio, on_rq, running; |
70b97a7f | 3997 | struct rq *rq; |
83ab0aa0 | 3998 | const struct sched_class *prev_class; |
b29739f9 IM |
3999 | |
4000 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4001 | ||
0122ec5b | 4002 | rq = __task_rq_lock(p); |
b29739f9 | 4003 | |
1c4dd99b TG |
4004 | /* |
4005 | * Idle task boosting is a nono in general. There is one | |
4006 | * exception, when PREEMPT_RT and NOHZ is active: | |
4007 | * | |
4008 | * The idle task calls get_next_timer_interrupt() and holds | |
4009 | * the timer wheel base->lock on the CPU and another CPU wants | |
4010 | * to access the timer (probably to cancel it). We can safely | |
4011 | * ignore the boosting request, as the idle CPU runs this code | |
4012 | * with interrupts disabled and will complete the lock | |
4013 | * protected section without being interrupted. So there is no | |
4014 | * real need to boost. | |
4015 | */ | |
4016 | if (unlikely(p == rq->idle)) { | |
4017 | WARN_ON(p != rq->curr); | |
4018 | WARN_ON(p->pi_blocked_on); | |
4019 | goto out_unlock; | |
4020 | } | |
4021 | ||
a8027073 | 4022 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4023 | oldprio = p->prio; |
83ab0aa0 | 4024 | prev_class = p->sched_class; |
fd2f4419 | 4025 | on_rq = p->on_rq; |
051a1d1a | 4026 | running = task_current(rq, p); |
0e1f3483 | 4027 | if (on_rq) |
69be72c1 | 4028 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4029 | if (running) |
4030 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4031 | |
4032 | if (rt_prio(prio)) | |
4033 | p->sched_class = &rt_sched_class; | |
4034 | else | |
4035 | p->sched_class = &fair_sched_class; | |
4036 | ||
b29739f9 IM |
4037 | p->prio = prio; |
4038 | ||
0e1f3483 HS |
4039 | if (running) |
4040 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4041 | if (on_rq) |
371fd7e7 | 4042 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4043 | |
da7a735e | 4044 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 4045 | out_unlock: |
0122ec5b | 4046 | __task_rq_unlock(rq); |
b29739f9 | 4047 | } |
b29739f9 | 4048 | #endif |
36c8b586 | 4049 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4050 | { |
dd41f596 | 4051 | int old_prio, delta, on_rq; |
1da177e4 | 4052 | unsigned long flags; |
70b97a7f | 4053 | struct rq *rq; |
1da177e4 LT |
4054 | |
4055 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4056 | return; | |
4057 | /* | |
4058 | * We have to be careful, if called from sys_setpriority(), | |
4059 | * the task might be in the middle of scheduling on another CPU. | |
4060 | */ | |
4061 | rq = task_rq_lock(p, &flags); | |
4062 | /* | |
4063 | * The RT priorities are set via sched_setscheduler(), but we still | |
4064 | * allow the 'normal' nice value to be set - but as expected | |
4065 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4066 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4067 | */ |
e05606d3 | 4068 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4069 | p->static_prio = NICE_TO_PRIO(nice); |
4070 | goto out_unlock; | |
4071 | } | |
fd2f4419 | 4072 | on_rq = p->on_rq; |
c09595f6 | 4073 | if (on_rq) |
69be72c1 | 4074 | dequeue_task(rq, p, 0); |
1da177e4 | 4075 | |
1da177e4 | 4076 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4077 | set_load_weight(p); |
b29739f9 IM |
4078 | old_prio = p->prio; |
4079 | p->prio = effective_prio(p); | |
4080 | delta = p->prio - old_prio; | |
1da177e4 | 4081 | |
dd41f596 | 4082 | if (on_rq) { |
371fd7e7 | 4083 | enqueue_task(rq, p, 0); |
1da177e4 | 4084 | /* |
d5f9f942 AM |
4085 | * If the task increased its priority or is running and |
4086 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4087 | */ |
d5f9f942 | 4088 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4089 | resched_task(rq->curr); |
4090 | } | |
4091 | out_unlock: | |
0122ec5b | 4092 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 4093 | } |
1da177e4 LT |
4094 | EXPORT_SYMBOL(set_user_nice); |
4095 | ||
e43379f1 MM |
4096 | /* |
4097 | * can_nice - check if a task can reduce its nice value | |
4098 | * @p: task | |
4099 | * @nice: nice value | |
4100 | */ | |
36c8b586 | 4101 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4102 | { |
024f4747 MM |
4103 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4104 | int nice_rlim = 20 - nice; | |
48f24c4d | 4105 | |
78d7d407 | 4106 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4107 | capable(CAP_SYS_NICE)); |
4108 | } | |
4109 | ||
1da177e4 LT |
4110 | #ifdef __ARCH_WANT_SYS_NICE |
4111 | ||
4112 | /* | |
4113 | * sys_nice - change the priority of the current process. | |
4114 | * @increment: priority increment | |
4115 | * | |
4116 | * sys_setpriority is a more generic, but much slower function that | |
4117 | * does similar things. | |
4118 | */ | |
5add95d4 | 4119 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4120 | { |
48f24c4d | 4121 | long nice, retval; |
1da177e4 LT |
4122 | |
4123 | /* | |
4124 | * Setpriority might change our priority at the same moment. | |
4125 | * We don't have to worry. Conceptually one call occurs first | |
4126 | * and we have a single winner. | |
4127 | */ | |
e43379f1 MM |
4128 | if (increment < -40) |
4129 | increment = -40; | |
1da177e4 LT |
4130 | if (increment > 40) |
4131 | increment = 40; | |
4132 | ||
2b8f836f | 4133 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4134 | if (nice < -20) |
4135 | nice = -20; | |
4136 | if (nice > 19) | |
4137 | nice = 19; | |
4138 | ||
e43379f1 MM |
4139 | if (increment < 0 && !can_nice(current, nice)) |
4140 | return -EPERM; | |
4141 | ||
1da177e4 LT |
4142 | retval = security_task_setnice(current, nice); |
4143 | if (retval) | |
4144 | return retval; | |
4145 | ||
4146 | set_user_nice(current, nice); | |
4147 | return 0; | |
4148 | } | |
4149 | ||
4150 | #endif | |
4151 | ||
4152 | /** | |
4153 | * task_prio - return the priority value of a given task. | |
4154 | * @p: the task in question. | |
4155 | * | |
4156 | * This is the priority value as seen by users in /proc. | |
4157 | * RT tasks are offset by -200. Normal tasks are centered | |
4158 | * around 0, value goes from -16 to +15. | |
4159 | */ | |
36c8b586 | 4160 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4161 | { |
4162 | return p->prio - MAX_RT_PRIO; | |
4163 | } | |
4164 | ||
4165 | /** | |
4166 | * task_nice - return the nice value of a given task. | |
4167 | * @p: the task in question. | |
4168 | */ | |
36c8b586 | 4169 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4170 | { |
4171 | return TASK_NICE(p); | |
4172 | } | |
150d8bed | 4173 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4174 | |
4175 | /** | |
4176 | * idle_cpu - is a given cpu idle currently? | |
4177 | * @cpu: the processor in question. | |
4178 | */ | |
4179 | int idle_cpu(int cpu) | |
4180 | { | |
908a3283 TG |
4181 | struct rq *rq = cpu_rq(cpu); |
4182 | ||
4183 | if (rq->curr != rq->idle) | |
4184 | return 0; | |
4185 | ||
4186 | if (rq->nr_running) | |
4187 | return 0; | |
4188 | ||
4189 | #ifdef CONFIG_SMP | |
4190 | if (!llist_empty(&rq->wake_list)) | |
4191 | return 0; | |
4192 | #endif | |
4193 | ||
4194 | return 1; | |
1da177e4 LT |
4195 | } |
4196 | ||
1da177e4 LT |
4197 | /** |
4198 | * idle_task - return the idle task for a given cpu. | |
4199 | * @cpu: the processor in question. | |
4200 | */ | |
36c8b586 | 4201 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4202 | { |
4203 | return cpu_rq(cpu)->idle; | |
4204 | } | |
4205 | ||
4206 | /** | |
4207 | * find_process_by_pid - find a process with a matching PID value. | |
4208 | * @pid: the pid in question. | |
4209 | */ | |
a9957449 | 4210 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4211 | { |
228ebcbe | 4212 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4213 | } |
4214 | ||
4215 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4216 | static void |
4217 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4218 | { |
1da177e4 LT |
4219 | p->policy = policy; |
4220 | p->rt_priority = prio; | |
b29739f9 IM |
4221 | p->normal_prio = normal_prio(p); |
4222 | /* we are holding p->pi_lock already */ | |
4223 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4224 | if (rt_prio(p->prio)) |
4225 | p->sched_class = &rt_sched_class; | |
4226 | else | |
4227 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4228 | set_load_weight(p); |
1da177e4 LT |
4229 | } |
4230 | ||
c69e8d9c DH |
4231 | /* |
4232 | * check the target process has a UID that matches the current process's | |
4233 | */ | |
4234 | static bool check_same_owner(struct task_struct *p) | |
4235 | { | |
4236 | const struct cred *cred = current_cred(), *pcred; | |
4237 | bool match; | |
4238 | ||
4239 | rcu_read_lock(); | |
4240 | pcred = __task_cred(p); | |
9c806aa0 EB |
4241 | match = (uid_eq(cred->euid, pcred->euid) || |
4242 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4243 | rcu_read_unlock(); |
4244 | return match; | |
4245 | } | |
4246 | ||
961ccddd | 4247 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4248 | const struct sched_param *param, bool user) |
1da177e4 | 4249 | { |
83b699ed | 4250 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4251 | unsigned long flags; |
83ab0aa0 | 4252 | const struct sched_class *prev_class; |
70b97a7f | 4253 | struct rq *rq; |
ca94c442 | 4254 | int reset_on_fork; |
1da177e4 | 4255 | |
66e5393a SR |
4256 | /* may grab non-irq protected spin_locks */ |
4257 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4258 | recheck: |
4259 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4260 | if (policy < 0) { |
4261 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4262 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4263 | } else { |
4264 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4265 | policy &= ~SCHED_RESET_ON_FORK; | |
4266 | ||
4267 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4268 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4269 | policy != SCHED_IDLE) | |
4270 | return -EINVAL; | |
4271 | } | |
4272 | ||
1da177e4 LT |
4273 | /* |
4274 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4275 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4276 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4277 | */ |
4278 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4279 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4280 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4281 | return -EINVAL; |
e05606d3 | 4282 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4283 | return -EINVAL; |
4284 | ||
37e4ab3f OC |
4285 | /* |
4286 | * Allow unprivileged RT tasks to decrease priority: | |
4287 | */ | |
961ccddd | 4288 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4289 | if (rt_policy(policy)) { |
a44702e8 ON |
4290 | unsigned long rlim_rtprio = |
4291 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4292 | |
4293 | /* can't set/change the rt policy */ | |
4294 | if (policy != p->policy && !rlim_rtprio) | |
4295 | return -EPERM; | |
4296 | ||
4297 | /* can't increase priority */ | |
4298 | if (param->sched_priority > p->rt_priority && | |
4299 | param->sched_priority > rlim_rtprio) | |
4300 | return -EPERM; | |
4301 | } | |
c02aa73b | 4302 | |
dd41f596 | 4303 | /* |
c02aa73b DH |
4304 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4305 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4306 | */ |
c02aa73b DH |
4307 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
4308 | if (!can_nice(p, TASK_NICE(p))) | |
4309 | return -EPERM; | |
4310 | } | |
5fe1d75f | 4311 | |
37e4ab3f | 4312 | /* can't change other user's priorities */ |
c69e8d9c | 4313 | if (!check_same_owner(p)) |
37e4ab3f | 4314 | return -EPERM; |
ca94c442 LP |
4315 | |
4316 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4317 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4318 | return -EPERM; | |
37e4ab3f | 4319 | } |
1da177e4 | 4320 | |
725aad24 | 4321 | if (user) { |
b0ae1981 | 4322 | retval = security_task_setscheduler(p); |
725aad24 JF |
4323 | if (retval) |
4324 | return retval; | |
4325 | } | |
4326 | ||
b29739f9 IM |
4327 | /* |
4328 | * make sure no PI-waiters arrive (or leave) while we are | |
4329 | * changing the priority of the task: | |
0122ec5b | 4330 | * |
25985edc | 4331 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4332 | * runqueue lock must be held. |
4333 | */ | |
0122ec5b | 4334 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 4335 | |
34f971f6 PZ |
4336 | /* |
4337 | * Changing the policy of the stop threads its a very bad idea | |
4338 | */ | |
4339 | if (p == rq->stop) { | |
0122ec5b | 4340 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
4341 | return -EINVAL; |
4342 | } | |
4343 | ||
a51e9198 DF |
4344 | /* |
4345 | * If not changing anything there's no need to proceed further: | |
4346 | */ | |
4347 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
4348 | param->sched_priority == p->rt_priority))) { | |
45afb173 | 4349 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
4350 | return 0; |
4351 | } | |
4352 | ||
dc61b1d6 PZ |
4353 | #ifdef CONFIG_RT_GROUP_SCHED |
4354 | if (user) { | |
4355 | /* | |
4356 | * Do not allow realtime tasks into groups that have no runtime | |
4357 | * assigned. | |
4358 | */ | |
4359 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4360 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4361 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 4362 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
4363 | return -EPERM; |
4364 | } | |
4365 | } | |
4366 | #endif | |
4367 | ||
1da177e4 LT |
4368 | /* recheck policy now with rq lock held */ |
4369 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4370 | policy = oldpolicy = -1; | |
0122ec5b | 4371 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
4372 | goto recheck; |
4373 | } | |
fd2f4419 | 4374 | on_rq = p->on_rq; |
051a1d1a | 4375 | running = task_current(rq, p); |
0e1f3483 | 4376 | if (on_rq) |
4ca9b72b | 4377 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4378 | if (running) |
4379 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4380 | |
ca94c442 LP |
4381 | p->sched_reset_on_fork = reset_on_fork; |
4382 | ||
1da177e4 | 4383 | oldprio = p->prio; |
83ab0aa0 | 4384 | prev_class = p->sched_class; |
dd41f596 | 4385 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4386 | |
0e1f3483 HS |
4387 | if (running) |
4388 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4389 | if (on_rq) |
4ca9b72b | 4390 | enqueue_task(rq, p, 0); |
cb469845 | 4391 | |
da7a735e | 4392 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4393 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 4394 | |
95e02ca9 TG |
4395 | rt_mutex_adjust_pi(p); |
4396 | ||
1da177e4 LT |
4397 | return 0; |
4398 | } | |
961ccddd RR |
4399 | |
4400 | /** | |
4401 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4402 | * @p: the task in question. | |
4403 | * @policy: new policy. | |
4404 | * @param: structure containing the new RT priority. | |
4405 | * | |
4406 | * NOTE that the task may be already dead. | |
4407 | */ | |
4408 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4409 | const struct sched_param *param) |
961ccddd RR |
4410 | { |
4411 | return __sched_setscheduler(p, policy, param, true); | |
4412 | } | |
1da177e4 LT |
4413 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4414 | ||
961ccddd RR |
4415 | /** |
4416 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4417 | * @p: the task in question. | |
4418 | * @policy: new policy. | |
4419 | * @param: structure containing the new RT priority. | |
4420 | * | |
4421 | * Just like sched_setscheduler, only don't bother checking if the | |
4422 | * current context has permission. For example, this is needed in | |
4423 | * stop_machine(): we create temporary high priority worker threads, | |
4424 | * but our caller might not have that capability. | |
4425 | */ | |
4426 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4427 | const struct sched_param *param) |
961ccddd RR |
4428 | { |
4429 | return __sched_setscheduler(p, policy, param, false); | |
4430 | } | |
4431 | ||
95cdf3b7 IM |
4432 | static int |
4433 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4434 | { |
1da177e4 LT |
4435 | struct sched_param lparam; |
4436 | struct task_struct *p; | |
36c8b586 | 4437 | int retval; |
1da177e4 LT |
4438 | |
4439 | if (!param || pid < 0) | |
4440 | return -EINVAL; | |
4441 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4442 | return -EFAULT; | |
5fe1d75f ON |
4443 | |
4444 | rcu_read_lock(); | |
4445 | retval = -ESRCH; | |
1da177e4 | 4446 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4447 | if (p != NULL) |
4448 | retval = sched_setscheduler(p, policy, &lparam); | |
4449 | rcu_read_unlock(); | |
36c8b586 | 4450 | |
1da177e4 LT |
4451 | return retval; |
4452 | } | |
4453 | ||
4454 | /** | |
4455 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4456 | * @pid: the pid in question. | |
4457 | * @policy: new policy. | |
4458 | * @param: structure containing the new RT priority. | |
4459 | */ | |
5add95d4 HC |
4460 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4461 | struct sched_param __user *, param) | |
1da177e4 | 4462 | { |
c21761f1 JB |
4463 | /* negative values for policy are not valid */ |
4464 | if (policy < 0) | |
4465 | return -EINVAL; | |
4466 | ||
1da177e4 LT |
4467 | return do_sched_setscheduler(pid, policy, param); |
4468 | } | |
4469 | ||
4470 | /** | |
4471 | * sys_sched_setparam - set/change the RT priority of a thread | |
4472 | * @pid: the pid in question. | |
4473 | * @param: structure containing the new RT priority. | |
4474 | */ | |
5add95d4 | 4475 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4476 | { |
4477 | return do_sched_setscheduler(pid, -1, param); | |
4478 | } | |
4479 | ||
4480 | /** | |
4481 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4482 | * @pid: the pid in question. | |
4483 | */ | |
5add95d4 | 4484 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4485 | { |
36c8b586 | 4486 | struct task_struct *p; |
3a5c359a | 4487 | int retval; |
1da177e4 LT |
4488 | |
4489 | if (pid < 0) | |
3a5c359a | 4490 | return -EINVAL; |
1da177e4 LT |
4491 | |
4492 | retval = -ESRCH; | |
5fe85be0 | 4493 | rcu_read_lock(); |
1da177e4 LT |
4494 | p = find_process_by_pid(pid); |
4495 | if (p) { | |
4496 | retval = security_task_getscheduler(p); | |
4497 | if (!retval) | |
ca94c442 LP |
4498 | retval = p->policy |
4499 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4500 | } |
5fe85be0 | 4501 | rcu_read_unlock(); |
1da177e4 LT |
4502 | return retval; |
4503 | } | |
4504 | ||
4505 | /** | |
ca94c442 | 4506 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4507 | * @pid: the pid in question. |
4508 | * @param: structure containing the RT priority. | |
4509 | */ | |
5add95d4 | 4510 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4511 | { |
4512 | struct sched_param lp; | |
36c8b586 | 4513 | struct task_struct *p; |
3a5c359a | 4514 | int retval; |
1da177e4 LT |
4515 | |
4516 | if (!param || pid < 0) | |
3a5c359a | 4517 | return -EINVAL; |
1da177e4 | 4518 | |
5fe85be0 | 4519 | rcu_read_lock(); |
1da177e4 LT |
4520 | p = find_process_by_pid(pid); |
4521 | retval = -ESRCH; | |
4522 | if (!p) | |
4523 | goto out_unlock; | |
4524 | ||
4525 | retval = security_task_getscheduler(p); | |
4526 | if (retval) | |
4527 | goto out_unlock; | |
4528 | ||
4529 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4530 | rcu_read_unlock(); |
1da177e4 LT |
4531 | |
4532 | /* | |
4533 | * This one might sleep, we cannot do it with a spinlock held ... | |
4534 | */ | |
4535 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4536 | ||
1da177e4 LT |
4537 | return retval; |
4538 | ||
4539 | out_unlock: | |
5fe85be0 | 4540 | rcu_read_unlock(); |
1da177e4 LT |
4541 | return retval; |
4542 | } | |
4543 | ||
96f874e2 | 4544 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4545 | { |
5a16f3d3 | 4546 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4547 | struct task_struct *p; |
4548 | int retval; | |
1da177e4 | 4549 | |
95402b38 | 4550 | get_online_cpus(); |
23f5d142 | 4551 | rcu_read_lock(); |
1da177e4 LT |
4552 | |
4553 | p = find_process_by_pid(pid); | |
4554 | if (!p) { | |
23f5d142 | 4555 | rcu_read_unlock(); |
95402b38 | 4556 | put_online_cpus(); |
1da177e4 LT |
4557 | return -ESRCH; |
4558 | } | |
4559 | ||
23f5d142 | 4560 | /* Prevent p going away */ |
1da177e4 | 4561 | get_task_struct(p); |
23f5d142 | 4562 | rcu_read_unlock(); |
1da177e4 | 4563 | |
5a16f3d3 RR |
4564 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4565 | retval = -ENOMEM; | |
4566 | goto out_put_task; | |
4567 | } | |
4568 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4569 | retval = -ENOMEM; | |
4570 | goto out_free_cpus_allowed; | |
4571 | } | |
1da177e4 | 4572 | retval = -EPERM; |
f1c84dae | 4573 | if (!check_same_owner(p) && !ns_capable(task_user_ns(p), CAP_SYS_NICE)) |
1da177e4 LT |
4574 | goto out_unlock; |
4575 | ||
b0ae1981 | 4576 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4577 | if (retval) |
4578 | goto out_unlock; | |
4579 | ||
5a16f3d3 RR |
4580 | cpuset_cpus_allowed(p, cpus_allowed); |
4581 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4582 | again: |
5a16f3d3 | 4583 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4584 | |
8707d8b8 | 4585 | if (!retval) { |
5a16f3d3 RR |
4586 | cpuset_cpus_allowed(p, cpus_allowed); |
4587 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4588 | /* |
4589 | * We must have raced with a concurrent cpuset | |
4590 | * update. Just reset the cpus_allowed to the | |
4591 | * cpuset's cpus_allowed | |
4592 | */ | |
5a16f3d3 | 4593 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4594 | goto again; |
4595 | } | |
4596 | } | |
1da177e4 | 4597 | out_unlock: |
5a16f3d3 RR |
4598 | free_cpumask_var(new_mask); |
4599 | out_free_cpus_allowed: | |
4600 | free_cpumask_var(cpus_allowed); | |
4601 | out_put_task: | |
1da177e4 | 4602 | put_task_struct(p); |
95402b38 | 4603 | put_online_cpus(); |
1da177e4 LT |
4604 | return retval; |
4605 | } | |
4606 | ||
4607 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4608 | struct cpumask *new_mask) |
1da177e4 | 4609 | { |
96f874e2 RR |
4610 | if (len < cpumask_size()) |
4611 | cpumask_clear(new_mask); | |
4612 | else if (len > cpumask_size()) | |
4613 | len = cpumask_size(); | |
4614 | ||
1da177e4 LT |
4615 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4616 | } | |
4617 | ||
4618 | /** | |
4619 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4620 | * @pid: pid of the process | |
4621 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4622 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4623 | */ | |
5add95d4 HC |
4624 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4625 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4626 | { |
5a16f3d3 | 4627 | cpumask_var_t new_mask; |
1da177e4 LT |
4628 | int retval; |
4629 | ||
5a16f3d3 RR |
4630 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4631 | return -ENOMEM; | |
1da177e4 | 4632 | |
5a16f3d3 RR |
4633 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4634 | if (retval == 0) | |
4635 | retval = sched_setaffinity(pid, new_mask); | |
4636 | free_cpumask_var(new_mask); | |
4637 | return retval; | |
1da177e4 LT |
4638 | } |
4639 | ||
96f874e2 | 4640 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4641 | { |
36c8b586 | 4642 | struct task_struct *p; |
31605683 | 4643 | unsigned long flags; |
1da177e4 | 4644 | int retval; |
1da177e4 | 4645 | |
95402b38 | 4646 | get_online_cpus(); |
23f5d142 | 4647 | rcu_read_lock(); |
1da177e4 LT |
4648 | |
4649 | retval = -ESRCH; | |
4650 | p = find_process_by_pid(pid); | |
4651 | if (!p) | |
4652 | goto out_unlock; | |
4653 | ||
e7834f8f DQ |
4654 | retval = security_task_getscheduler(p); |
4655 | if (retval) | |
4656 | goto out_unlock; | |
4657 | ||
013fdb80 | 4658 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 4659 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 4660 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4661 | |
4662 | out_unlock: | |
23f5d142 | 4663 | rcu_read_unlock(); |
95402b38 | 4664 | put_online_cpus(); |
1da177e4 | 4665 | |
9531b62f | 4666 | return retval; |
1da177e4 LT |
4667 | } |
4668 | ||
4669 | /** | |
4670 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4671 | * @pid: pid of the process | |
4672 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4673 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4674 | */ | |
5add95d4 HC |
4675 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4676 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4677 | { |
4678 | int ret; | |
f17c8607 | 4679 | cpumask_var_t mask; |
1da177e4 | 4680 | |
84fba5ec | 4681 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4682 | return -EINVAL; |
4683 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4684 | return -EINVAL; |
4685 | ||
f17c8607 RR |
4686 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4687 | return -ENOMEM; | |
1da177e4 | 4688 | |
f17c8607 RR |
4689 | ret = sched_getaffinity(pid, mask); |
4690 | if (ret == 0) { | |
8bc037fb | 4691 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4692 | |
4693 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4694 | ret = -EFAULT; |
4695 | else | |
cd3d8031 | 4696 | ret = retlen; |
f17c8607 RR |
4697 | } |
4698 | free_cpumask_var(mask); | |
1da177e4 | 4699 | |
f17c8607 | 4700 | return ret; |
1da177e4 LT |
4701 | } |
4702 | ||
4703 | /** | |
4704 | * sys_sched_yield - yield the current processor to other threads. | |
4705 | * | |
dd41f596 IM |
4706 | * This function yields the current CPU to other tasks. If there are no |
4707 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4708 | */ |
5add95d4 | 4709 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4710 | { |
70b97a7f | 4711 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4712 | |
2d72376b | 4713 | schedstat_inc(rq, yld_count); |
4530d7ab | 4714 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4715 | |
4716 | /* | |
4717 | * Since we are going to call schedule() anyway, there's | |
4718 | * no need to preempt or enable interrupts: | |
4719 | */ | |
4720 | __release(rq->lock); | |
8a25d5de | 4721 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4722 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4723 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4724 | |
4725 | schedule(); | |
4726 | ||
4727 | return 0; | |
4728 | } | |
4729 | ||
d86ee480 PZ |
4730 | static inline int should_resched(void) |
4731 | { | |
4732 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4733 | } | |
4734 | ||
e7b38404 | 4735 | static void __cond_resched(void) |
1da177e4 | 4736 | { |
e7aaaa69 | 4737 | add_preempt_count(PREEMPT_ACTIVE); |
c259e01a | 4738 | __schedule(); |
e7aaaa69 | 4739 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 LT |
4740 | } |
4741 | ||
02b67cc3 | 4742 | int __sched _cond_resched(void) |
1da177e4 | 4743 | { |
d86ee480 | 4744 | if (should_resched()) { |
1da177e4 LT |
4745 | __cond_resched(); |
4746 | return 1; | |
4747 | } | |
4748 | return 0; | |
4749 | } | |
02b67cc3 | 4750 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4751 | |
4752 | /* | |
613afbf8 | 4753 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4754 | * call schedule, and on return reacquire the lock. |
4755 | * | |
41a2d6cf | 4756 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4757 | * operations here to prevent schedule() from being called twice (once via |
4758 | * spin_unlock(), once by hand). | |
4759 | */ | |
613afbf8 | 4760 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4761 | { |
d86ee480 | 4762 | int resched = should_resched(); |
6df3cecb JK |
4763 | int ret = 0; |
4764 | ||
f607c668 PZ |
4765 | lockdep_assert_held(lock); |
4766 | ||
95c354fe | 4767 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4768 | spin_unlock(lock); |
d86ee480 | 4769 | if (resched) |
95c354fe NP |
4770 | __cond_resched(); |
4771 | else | |
4772 | cpu_relax(); | |
6df3cecb | 4773 | ret = 1; |
1da177e4 | 4774 | spin_lock(lock); |
1da177e4 | 4775 | } |
6df3cecb | 4776 | return ret; |
1da177e4 | 4777 | } |
613afbf8 | 4778 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4779 | |
613afbf8 | 4780 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4781 | { |
4782 | BUG_ON(!in_softirq()); | |
4783 | ||
d86ee480 | 4784 | if (should_resched()) { |
98d82567 | 4785 | local_bh_enable(); |
1da177e4 LT |
4786 | __cond_resched(); |
4787 | local_bh_disable(); | |
4788 | return 1; | |
4789 | } | |
4790 | return 0; | |
4791 | } | |
613afbf8 | 4792 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4793 | |
1da177e4 LT |
4794 | /** |
4795 | * yield - yield the current processor to other threads. | |
4796 | * | |
8e3fabfd PZ |
4797 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4798 | * | |
4799 | * The scheduler is at all times free to pick the calling task as the most | |
4800 | * eligible task to run, if removing the yield() call from your code breaks | |
4801 | * it, its already broken. | |
4802 | * | |
4803 | * Typical broken usage is: | |
4804 | * | |
4805 | * while (!event) | |
4806 | * yield(); | |
4807 | * | |
4808 | * where one assumes that yield() will let 'the other' process run that will | |
4809 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4810 | * happen. Never use yield() as a progress guarantee!! | |
4811 | * | |
4812 | * If you want to use yield() to wait for something, use wait_event(). | |
4813 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4814 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4815 | */ |
4816 | void __sched yield(void) | |
4817 | { | |
4818 | set_current_state(TASK_RUNNING); | |
4819 | sys_sched_yield(); | |
4820 | } | |
1da177e4 LT |
4821 | EXPORT_SYMBOL(yield); |
4822 | ||
d95f4122 MG |
4823 | /** |
4824 | * yield_to - yield the current processor to another thread in | |
4825 | * your thread group, or accelerate that thread toward the | |
4826 | * processor it's on. | |
16addf95 RD |
4827 | * @p: target task |
4828 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4829 | * |
4830 | * It's the caller's job to ensure that the target task struct | |
4831 | * can't go away on us before we can do any checks. | |
4832 | * | |
4833 | * Returns true if we indeed boosted the target task. | |
4834 | */ | |
4835 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4836 | { | |
4837 | struct task_struct *curr = current; | |
4838 | struct rq *rq, *p_rq; | |
4839 | unsigned long flags; | |
4840 | bool yielded = 0; | |
4841 | ||
4842 | local_irq_save(flags); | |
4843 | rq = this_rq(); | |
4844 | ||
4845 | again: | |
4846 | p_rq = task_rq(p); | |
4847 | double_rq_lock(rq, p_rq); | |
4848 | while (task_rq(p) != p_rq) { | |
4849 | double_rq_unlock(rq, p_rq); | |
4850 | goto again; | |
4851 | } | |
4852 | ||
4853 | if (!curr->sched_class->yield_to_task) | |
4854 | goto out; | |
4855 | ||
4856 | if (curr->sched_class != p->sched_class) | |
4857 | goto out; | |
4858 | ||
4859 | if (task_running(p_rq, p) || p->state) | |
4860 | goto out; | |
4861 | ||
4862 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4863 | if (yielded) { |
d95f4122 | 4864 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4865 | /* |
4866 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4867 | * fairness. | |
4868 | */ | |
4869 | if (preempt && rq != p_rq) | |
4870 | resched_task(p_rq->curr); | |
916671c0 MG |
4871 | } else { |
4872 | /* | |
4873 | * We might have set it in task_yield_fair(), but are | |
4874 | * not going to schedule(), so don't want to skip | |
4875 | * the next update. | |
4876 | */ | |
4877 | rq->skip_clock_update = 0; | |
6d1cafd8 | 4878 | } |
d95f4122 MG |
4879 | |
4880 | out: | |
4881 | double_rq_unlock(rq, p_rq); | |
4882 | local_irq_restore(flags); | |
4883 | ||
4884 | if (yielded) | |
4885 | schedule(); | |
4886 | ||
4887 | return yielded; | |
4888 | } | |
4889 | EXPORT_SYMBOL_GPL(yield_to); | |
4890 | ||
1da177e4 | 4891 | /* |
41a2d6cf | 4892 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4893 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4894 | */ |
4895 | void __sched io_schedule(void) | |
4896 | { | |
54d35f29 | 4897 | struct rq *rq = raw_rq(); |
1da177e4 | 4898 | |
0ff92245 | 4899 | delayacct_blkio_start(); |
1da177e4 | 4900 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4901 | blk_flush_plug(current); |
8f0dfc34 | 4902 | current->in_iowait = 1; |
1da177e4 | 4903 | schedule(); |
8f0dfc34 | 4904 | current->in_iowait = 0; |
1da177e4 | 4905 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4906 | delayacct_blkio_end(); |
1da177e4 | 4907 | } |
1da177e4 LT |
4908 | EXPORT_SYMBOL(io_schedule); |
4909 | ||
4910 | long __sched io_schedule_timeout(long timeout) | |
4911 | { | |
54d35f29 | 4912 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4913 | long ret; |
4914 | ||
0ff92245 | 4915 | delayacct_blkio_start(); |
1da177e4 | 4916 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4917 | blk_flush_plug(current); |
8f0dfc34 | 4918 | current->in_iowait = 1; |
1da177e4 | 4919 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4920 | current->in_iowait = 0; |
1da177e4 | 4921 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4922 | delayacct_blkio_end(); |
1da177e4 LT |
4923 | return ret; |
4924 | } | |
4925 | ||
4926 | /** | |
4927 | * sys_sched_get_priority_max - return maximum RT priority. | |
4928 | * @policy: scheduling class. | |
4929 | * | |
4930 | * this syscall returns the maximum rt_priority that can be used | |
4931 | * by a given scheduling class. | |
4932 | */ | |
5add95d4 | 4933 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4934 | { |
4935 | int ret = -EINVAL; | |
4936 | ||
4937 | switch (policy) { | |
4938 | case SCHED_FIFO: | |
4939 | case SCHED_RR: | |
4940 | ret = MAX_USER_RT_PRIO-1; | |
4941 | break; | |
4942 | case SCHED_NORMAL: | |
b0a9499c | 4943 | case SCHED_BATCH: |
dd41f596 | 4944 | case SCHED_IDLE: |
1da177e4 LT |
4945 | ret = 0; |
4946 | break; | |
4947 | } | |
4948 | return ret; | |
4949 | } | |
4950 | ||
4951 | /** | |
4952 | * sys_sched_get_priority_min - return minimum RT priority. | |
4953 | * @policy: scheduling class. | |
4954 | * | |
4955 | * this syscall returns the minimum rt_priority that can be used | |
4956 | * by a given scheduling class. | |
4957 | */ | |
5add95d4 | 4958 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4959 | { |
4960 | int ret = -EINVAL; | |
4961 | ||
4962 | switch (policy) { | |
4963 | case SCHED_FIFO: | |
4964 | case SCHED_RR: | |
4965 | ret = 1; | |
4966 | break; | |
4967 | case SCHED_NORMAL: | |
b0a9499c | 4968 | case SCHED_BATCH: |
dd41f596 | 4969 | case SCHED_IDLE: |
1da177e4 LT |
4970 | ret = 0; |
4971 | } | |
4972 | return ret; | |
4973 | } | |
4974 | ||
4975 | /** | |
4976 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4977 | * @pid: pid of the process. | |
4978 | * @interval: userspace pointer to the timeslice value. | |
4979 | * | |
4980 | * this syscall writes the default timeslice value of a given process | |
4981 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4982 | */ | |
17da2bd9 | 4983 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4984 | struct timespec __user *, interval) |
1da177e4 | 4985 | { |
36c8b586 | 4986 | struct task_struct *p; |
a4ec24b4 | 4987 | unsigned int time_slice; |
dba091b9 TG |
4988 | unsigned long flags; |
4989 | struct rq *rq; | |
3a5c359a | 4990 | int retval; |
1da177e4 | 4991 | struct timespec t; |
1da177e4 LT |
4992 | |
4993 | if (pid < 0) | |
3a5c359a | 4994 | return -EINVAL; |
1da177e4 LT |
4995 | |
4996 | retval = -ESRCH; | |
1a551ae7 | 4997 | rcu_read_lock(); |
1da177e4 LT |
4998 | p = find_process_by_pid(pid); |
4999 | if (!p) | |
5000 | goto out_unlock; | |
5001 | ||
5002 | retval = security_task_getscheduler(p); | |
5003 | if (retval) | |
5004 | goto out_unlock; | |
5005 | ||
dba091b9 TG |
5006 | rq = task_rq_lock(p, &flags); |
5007 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 5008 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 5009 | |
1a551ae7 | 5010 | rcu_read_unlock(); |
a4ec24b4 | 5011 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5012 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5013 | return retval; |
3a5c359a | 5014 | |
1da177e4 | 5015 | out_unlock: |
1a551ae7 | 5016 | rcu_read_unlock(); |
1da177e4 LT |
5017 | return retval; |
5018 | } | |
5019 | ||
7c731e0a | 5020 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5021 | |
82a1fcb9 | 5022 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5023 | { |
1da177e4 | 5024 | unsigned long free = 0; |
36c8b586 | 5025 | unsigned state; |
1da177e4 | 5026 | |
1da177e4 | 5027 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5028 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5029 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5030 | #if BITS_PER_LONG == 32 |
1da177e4 | 5031 | if (state == TASK_RUNNING) |
3df0fc5b | 5032 | printk(KERN_CONT " running "); |
1da177e4 | 5033 | else |
3df0fc5b | 5034 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5035 | #else |
5036 | if (state == TASK_RUNNING) | |
3df0fc5b | 5037 | printk(KERN_CONT " running task "); |
1da177e4 | 5038 | else |
3df0fc5b | 5039 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5040 | #endif |
5041 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5042 | free = stack_not_used(p); |
1da177e4 | 5043 | #endif |
3df0fc5b | 5044 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
07cde260 | 5045 | task_pid_nr(p), task_pid_nr(rcu_dereference(p->real_parent)), |
aa47b7e0 | 5046 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5047 | |
5fb5e6de | 5048 | show_stack(p, NULL); |
1da177e4 LT |
5049 | } |
5050 | ||
e59e2ae2 | 5051 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5052 | { |
36c8b586 | 5053 | struct task_struct *g, *p; |
1da177e4 | 5054 | |
4bd77321 | 5055 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5056 | printk(KERN_INFO |
5057 | " task PC stack pid father\n"); | |
1da177e4 | 5058 | #else |
3df0fc5b PZ |
5059 | printk(KERN_INFO |
5060 | " task PC stack pid father\n"); | |
1da177e4 | 5061 | #endif |
510f5acc | 5062 | rcu_read_lock(); |
1da177e4 LT |
5063 | do_each_thread(g, p) { |
5064 | /* | |
5065 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5066 | * console might take a lot of time: |
1da177e4 LT |
5067 | */ |
5068 | touch_nmi_watchdog(); | |
39bc89fd | 5069 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5070 | sched_show_task(p); |
1da177e4 LT |
5071 | } while_each_thread(g, p); |
5072 | ||
04c9167f JF |
5073 | touch_all_softlockup_watchdogs(); |
5074 | ||
dd41f596 IM |
5075 | #ifdef CONFIG_SCHED_DEBUG |
5076 | sysrq_sched_debug_show(); | |
5077 | #endif | |
510f5acc | 5078 | rcu_read_unlock(); |
e59e2ae2 IM |
5079 | /* |
5080 | * Only show locks if all tasks are dumped: | |
5081 | */ | |
93335a21 | 5082 | if (!state_filter) |
e59e2ae2 | 5083 | debug_show_all_locks(); |
1da177e4 LT |
5084 | } |
5085 | ||
1df21055 IM |
5086 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5087 | { | |
dd41f596 | 5088 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5089 | } |
5090 | ||
f340c0d1 IM |
5091 | /** |
5092 | * init_idle - set up an idle thread for a given CPU | |
5093 | * @idle: task in question | |
5094 | * @cpu: cpu the idle task belongs to | |
5095 | * | |
5096 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5097 | * flag, to make booting more robust. | |
5098 | */ | |
5c1e1767 | 5099 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5100 | { |
70b97a7f | 5101 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5102 | unsigned long flags; |
5103 | ||
05fa785c | 5104 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5105 | |
dd41f596 | 5106 | __sched_fork(idle); |
06b83b5f | 5107 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5108 | idle->se.exec_start = sched_clock(); |
5109 | ||
1e1b6c51 | 5110 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
5111 | /* |
5112 | * We're having a chicken and egg problem, even though we are | |
5113 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5114 | * lockdep check in task_group() will fail. | |
5115 | * | |
5116 | * Similar case to sched_fork(). / Alternatively we could | |
5117 | * use task_rq_lock() here and obtain the other rq->lock. | |
5118 | * | |
5119 | * Silence PROVE_RCU | |
5120 | */ | |
5121 | rcu_read_lock(); | |
dd41f596 | 5122 | __set_task_cpu(idle, cpu); |
6506cf6c | 5123 | rcu_read_unlock(); |
1da177e4 | 5124 | |
1da177e4 | 5125 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
5126 | #if defined(CONFIG_SMP) |
5127 | idle->on_cpu = 1; | |
4866cde0 | 5128 | #endif |
05fa785c | 5129 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5130 | |
5131 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5132 | task_thread_info(idle)->preempt_count = 0; |
55cd5340 | 5133 | |
dd41f596 IM |
5134 | /* |
5135 | * The idle tasks have their own, simple scheduling class: | |
5136 | */ | |
5137 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5138 | ftrace_graph_init_idle_task(idle, cpu); |
f1c6f1a7 CE |
5139 | #if defined(CONFIG_SMP) |
5140 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
5141 | #endif | |
19978ca6 IM |
5142 | } |
5143 | ||
1da177e4 | 5144 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
5145 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
5146 | { | |
5147 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
5148 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
5149 | |
5150 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 5151 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
5152 | } |
5153 | ||
1da177e4 LT |
5154 | /* |
5155 | * This is how migration works: | |
5156 | * | |
969c7921 TH |
5157 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5158 | * stop_one_cpu(). | |
5159 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5160 | * off the CPU) | |
5161 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5162 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5163 | * it and puts it into the right queue. |
969c7921 TH |
5164 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5165 | * is done. | |
1da177e4 LT |
5166 | */ |
5167 | ||
5168 | /* | |
5169 | * Change a given task's CPU affinity. Migrate the thread to a | |
5170 | * proper CPU and schedule it away if the CPU it's executing on | |
5171 | * is removed from the allowed bitmask. | |
5172 | * | |
5173 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5174 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5175 | * call is not atomic; no spinlocks may be held. |
5176 | */ | |
96f874e2 | 5177 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5178 | { |
5179 | unsigned long flags; | |
70b97a7f | 5180 | struct rq *rq; |
969c7921 | 5181 | unsigned int dest_cpu; |
48f24c4d | 5182 | int ret = 0; |
1da177e4 LT |
5183 | |
5184 | rq = task_rq_lock(p, &flags); | |
e2912009 | 5185 | |
db44fc01 YZ |
5186 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
5187 | goto out; | |
5188 | ||
6ad4c188 | 5189 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5190 | ret = -EINVAL; |
5191 | goto out; | |
5192 | } | |
5193 | ||
db44fc01 | 5194 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
5195 | ret = -EINVAL; |
5196 | goto out; | |
5197 | } | |
5198 | ||
1e1b6c51 | 5199 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 5200 | |
1da177e4 | 5201 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5202 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5203 | goto out; |
5204 | ||
969c7921 | 5205 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 5206 | if (p->on_rq) { |
969c7921 | 5207 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 5208 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 5209 | task_rq_unlock(rq, p, &flags); |
969c7921 | 5210 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5211 | tlb_migrate_finish(p->mm); |
5212 | return 0; | |
5213 | } | |
5214 | out: | |
0122ec5b | 5215 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 5216 | |
1da177e4 LT |
5217 | return ret; |
5218 | } | |
cd8ba7cd | 5219 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5220 | |
5221 | /* | |
41a2d6cf | 5222 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5223 | * this because either it can't run here any more (set_cpus_allowed() |
5224 | * away from this CPU, or CPU going down), or because we're | |
5225 | * attempting to rebalance this task on exec (sched_exec). | |
5226 | * | |
5227 | * So we race with normal scheduler movements, but that's OK, as long | |
5228 | * as the task is no longer on this CPU. | |
efc30814 KK |
5229 | * |
5230 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5231 | */ |
efc30814 | 5232 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5233 | { |
70b97a7f | 5234 | struct rq *rq_dest, *rq_src; |
e2912009 | 5235 | int ret = 0; |
1da177e4 | 5236 | |
e761b772 | 5237 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5238 | return ret; |
1da177e4 LT |
5239 | |
5240 | rq_src = cpu_rq(src_cpu); | |
5241 | rq_dest = cpu_rq(dest_cpu); | |
5242 | ||
0122ec5b | 5243 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
5244 | double_rq_lock(rq_src, rq_dest); |
5245 | /* Already moved. */ | |
5246 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5247 | goto done; |
1da177e4 | 5248 | /* Affinity changed (again). */ |
fa17b507 | 5249 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 5250 | goto fail; |
1da177e4 | 5251 | |
e2912009 PZ |
5252 | /* |
5253 | * If we're not on a rq, the next wake-up will ensure we're | |
5254 | * placed properly. | |
5255 | */ | |
fd2f4419 | 5256 | if (p->on_rq) { |
4ca9b72b | 5257 | dequeue_task(rq_src, p, 0); |
e2912009 | 5258 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 5259 | enqueue_task(rq_dest, p, 0); |
15afe09b | 5260 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5261 | } |
b1e38734 | 5262 | done: |
efc30814 | 5263 | ret = 1; |
b1e38734 | 5264 | fail: |
1da177e4 | 5265 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 5266 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 5267 | return ret; |
1da177e4 LT |
5268 | } |
5269 | ||
5270 | /* | |
969c7921 TH |
5271 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5272 | * and performs thread migration by bumping thread off CPU then | |
5273 | * 'pushing' onto another runqueue. | |
1da177e4 | 5274 | */ |
969c7921 | 5275 | static int migration_cpu_stop(void *data) |
1da177e4 | 5276 | { |
969c7921 | 5277 | struct migration_arg *arg = data; |
f7b4cddc | 5278 | |
969c7921 TH |
5279 | /* |
5280 | * The original target cpu might have gone down and we might | |
5281 | * be on another cpu but it doesn't matter. | |
5282 | */ | |
f7b4cddc | 5283 | local_irq_disable(); |
969c7921 | 5284 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5285 | local_irq_enable(); |
1da177e4 | 5286 | return 0; |
f7b4cddc ON |
5287 | } |
5288 | ||
1da177e4 | 5289 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5290 | |
054b9108 | 5291 | /* |
48c5ccae PZ |
5292 | * Ensures that the idle task is using init_mm right before its cpu goes |
5293 | * offline. | |
054b9108 | 5294 | */ |
48c5ccae | 5295 | void idle_task_exit(void) |
1da177e4 | 5296 | { |
48c5ccae | 5297 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5298 | |
48c5ccae | 5299 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5300 | |
48c5ccae PZ |
5301 | if (mm != &init_mm) |
5302 | switch_mm(mm, &init_mm, current); | |
5303 | mmdrop(mm); | |
1da177e4 LT |
5304 | } |
5305 | ||
5306 | /* | |
5307 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5308 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5309 | * for performance reasons the counter is not stricly tracking tasks to | |
5310 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5311 | * to keep the global sum constant after CPU-down: | |
5312 | */ | |
70b97a7f | 5313 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5314 | { |
6ad4c188 | 5315 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5316 | |
1da177e4 LT |
5317 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5318 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5319 | } |
5320 | ||
dd41f596 | 5321 | /* |
48c5ccae | 5322 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5323 | */ |
48c5ccae | 5324 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5325 | { |
48c5ccae PZ |
5326 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5327 | rq->calc_load_active = 0; | |
1da177e4 LT |
5328 | } |
5329 | ||
48f24c4d | 5330 | /* |
48c5ccae PZ |
5331 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5332 | * try_to_wake_up()->select_task_rq(). | |
5333 | * | |
5334 | * Called with rq->lock held even though we'er in stop_machine() and | |
5335 | * there's no concurrency possible, we hold the required locks anyway | |
5336 | * because of lock validation efforts. | |
1da177e4 | 5337 | */ |
48c5ccae | 5338 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5339 | { |
70b97a7f | 5340 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5341 | struct task_struct *next, *stop = rq->stop; |
5342 | int dest_cpu; | |
1da177e4 LT |
5343 | |
5344 | /* | |
48c5ccae PZ |
5345 | * Fudge the rq selection such that the below task selection loop |
5346 | * doesn't get stuck on the currently eligible stop task. | |
5347 | * | |
5348 | * We're currently inside stop_machine() and the rq is either stuck | |
5349 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5350 | * either way we should never end up calling schedule() until we're | |
5351 | * done here. | |
1da177e4 | 5352 | */ |
48c5ccae | 5353 | rq->stop = NULL; |
48f24c4d | 5354 | |
8cb120d3 PT |
5355 | /* Ensure any throttled groups are reachable by pick_next_task */ |
5356 | unthrottle_offline_cfs_rqs(rq); | |
5357 | ||
dd41f596 | 5358 | for ( ; ; ) { |
48c5ccae PZ |
5359 | /* |
5360 | * There's this thread running, bail when that's the only | |
5361 | * remaining thread. | |
5362 | */ | |
5363 | if (rq->nr_running == 1) | |
dd41f596 | 5364 | break; |
48c5ccae | 5365 | |
b67802ea | 5366 | next = pick_next_task(rq); |
48c5ccae | 5367 | BUG_ON(!next); |
79c53799 | 5368 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5369 | |
48c5ccae PZ |
5370 | /* Find suitable destination for @next, with force if needed. */ |
5371 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5372 | raw_spin_unlock(&rq->lock); | |
5373 | ||
5374 | __migrate_task(next, dead_cpu, dest_cpu); | |
5375 | ||
5376 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5377 | } |
dce48a84 | 5378 | |
48c5ccae | 5379 | rq->stop = stop; |
dce48a84 | 5380 | } |
48c5ccae | 5381 | |
1da177e4 LT |
5382 | #endif /* CONFIG_HOTPLUG_CPU */ |
5383 | ||
e692ab53 NP |
5384 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5385 | ||
5386 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5387 | { |
5388 | .procname = "sched_domain", | |
c57baf1e | 5389 | .mode = 0555, |
e0361851 | 5390 | }, |
56992309 | 5391 | {} |
e692ab53 NP |
5392 | }; |
5393 | ||
5394 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5395 | { |
5396 | .procname = "kernel", | |
c57baf1e | 5397 | .mode = 0555, |
e0361851 AD |
5398 | .child = sd_ctl_dir, |
5399 | }, | |
56992309 | 5400 | {} |
e692ab53 NP |
5401 | }; |
5402 | ||
5403 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5404 | { | |
5405 | struct ctl_table *entry = | |
5cf9f062 | 5406 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5407 | |
e692ab53 NP |
5408 | return entry; |
5409 | } | |
5410 | ||
6382bc90 MM |
5411 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5412 | { | |
cd790076 | 5413 | struct ctl_table *entry; |
6382bc90 | 5414 | |
cd790076 MM |
5415 | /* |
5416 | * In the intermediate directories, both the child directory and | |
5417 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5418 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5419 | * static strings and all have proc handlers. |
5420 | */ | |
5421 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5422 | if (entry->child) |
5423 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5424 | if (entry->proc_handler == NULL) |
5425 | kfree(entry->procname); | |
5426 | } | |
6382bc90 MM |
5427 | |
5428 | kfree(*tablep); | |
5429 | *tablep = NULL; | |
5430 | } | |
5431 | ||
e692ab53 | 5432 | static void |
e0361851 | 5433 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 5434 | const char *procname, void *data, int maxlen, |
36fcb589 | 5435 | umode_t mode, proc_handler *proc_handler) |
e692ab53 | 5436 | { |
e692ab53 NP |
5437 | entry->procname = procname; |
5438 | entry->data = data; | |
5439 | entry->maxlen = maxlen; | |
5440 | entry->mode = mode; | |
5441 | entry->proc_handler = proc_handler; | |
5442 | } | |
5443 | ||
5444 | static struct ctl_table * | |
5445 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5446 | { | |
a5d8c348 | 5447 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5448 | |
ad1cdc1d MM |
5449 | if (table == NULL) |
5450 | return NULL; | |
5451 | ||
e0361851 | 5452 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5453 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5454 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5455 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5456 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5457 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5458 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5459 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5460 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5461 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5462 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5463 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5464 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5465 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5466 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5467 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5468 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5469 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5470 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5471 | &sd->cache_nice_tries, |
5472 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5473 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5474 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5475 | set_table_entry(&table[11], "name", sd->name, |
5476 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5477 | /* &table[12] is terminator */ | |
e692ab53 NP |
5478 | |
5479 | return table; | |
5480 | } | |
5481 | ||
9a4e7159 | 5482 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5483 | { |
5484 | struct ctl_table *entry, *table; | |
5485 | struct sched_domain *sd; | |
5486 | int domain_num = 0, i; | |
5487 | char buf[32]; | |
5488 | ||
5489 | for_each_domain(cpu, sd) | |
5490 | domain_num++; | |
5491 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5492 | if (table == NULL) |
5493 | return NULL; | |
e692ab53 NP |
5494 | |
5495 | i = 0; | |
5496 | for_each_domain(cpu, sd) { | |
5497 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5498 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5499 | entry->mode = 0555; |
e692ab53 NP |
5500 | entry->child = sd_alloc_ctl_domain_table(sd); |
5501 | entry++; | |
5502 | i++; | |
5503 | } | |
5504 | return table; | |
5505 | } | |
5506 | ||
5507 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5508 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5509 | { |
6ad4c188 | 5510 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5511 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5512 | char buf[32]; | |
5513 | ||
7378547f MM |
5514 | WARN_ON(sd_ctl_dir[0].child); |
5515 | sd_ctl_dir[0].child = entry; | |
5516 | ||
ad1cdc1d MM |
5517 | if (entry == NULL) |
5518 | return; | |
5519 | ||
6ad4c188 | 5520 | for_each_possible_cpu(i) { |
e692ab53 | 5521 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5522 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5523 | entry->mode = 0555; |
e692ab53 | 5524 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5525 | entry++; |
e692ab53 | 5526 | } |
7378547f MM |
5527 | |
5528 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5529 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5530 | } | |
6382bc90 | 5531 | |
7378547f | 5532 | /* may be called multiple times per register */ |
6382bc90 MM |
5533 | static void unregister_sched_domain_sysctl(void) |
5534 | { | |
7378547f MM |
5535 | if (sd_sysctl_header) |
5536 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5537 | sd_sysctl_header = NULL; |
7378547f MM |
5538 | if (sd_ctl_dir[0].child) |
5539 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5540 | } |
e692ab53 | 5541 | #else |
6382bc90 MM |
5542 | static void register_sched_domain_sysctl(void) |
5543 | { | |
5544 | } | |
5545 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5546 | { |
5547 | } | |
5548 | #endif | |
5549 | ||
1f11eb6a GH |
5550 | static void set_rq_online(struct rq *rq) |
5551 | { | |
5552 | if (!rq->online) { | |
5553 | const struct sched_class *class; | |
5554 | ||
c6c4927b | 5555 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5556 | rq->online = 1; |
5557 | ||
5558 | for_each_class(class) { | |
5559 | if (class->rq_online) | |
5560 | class->rq_online(rq); | |
5561 | } | |
5562 | } | |
5563 | } | |
5564 | ||
5565 | static void set_rq_offline(struct rq *rq) | |
5566 | { | |
5567 | if (rq->online) { | |
5568 | const struct sched_class *class; | |
5569 | ||
5570 | for_each_class(class) { | |
5571 | if (class->rq_offline) | |
5572 | class->rq_offline(rq); | |
5573 | } | |
5574 | ||
c6c4927b | 5575 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5576 | rq->online = 0; |
5577 | } | |
5578 | } | |
5579 | ||
1da177e4 LT |
5580 | /* |
5581 | * migration_call - callback that gets triggered when a CPU is added. | |
5582 | * Here we can start up the necessary migration thread for the new CPU. | |
5583 | */ | |
48f24c4d IM |
5584 | static int __cpuinit |
5585 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5586 | { |
48f24c4d | 5587 | int cpu = (long)hcpu; |
1da177e4 | 5588 | unsigned long flags; |
969c7921 | 5589 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5590 | |
48c5ccae | 5591 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5592 | |
1da177e4 | 5593 | case CPU_UP_PREPARE: |
a468d389 | 5594 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5595 | break; |
48f24c4d | 5596 | |
1da177e4 | 5597 | case CPU_ONLINE: |
1f94ef59 | 5598 | /* Update our root-domain */ |
05fa785c | 5599 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5600 | if (rq->rd) { |
c6c4927b | 5601 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5602 | |
5603 | set_rq_online(rq); | |
1f94ef59 | 5604 | } |
05fa785c | 5605 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5606 | break; |
48f24c4d | 5607 | |
1da177e4 | 5608 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5609 | case CPU_DYING: |
317f3941 | 5610 | sched_ttwu_pending(); |
57d885fe | 5611 | /* Update our root-domain */ |
05fa785c | 5612 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5613 | if (rq->rd) { |
c6c4927b | 5614 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5615 | set_rq_offline(rq); |
57d885fe | 5616 | } |
48c5ccae PZ |
5617 | migrate_tasks(cpu); |
5618 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5619 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5620 | |
5621 | migrate_nr_uninterruptible(rq); | |
5622 | calc_global_load_remove(rq); | |
57d885fe | 5623 | break; |
1da177e4 LT |
5624 | #endif |
5625 | } | |
49c022e6 PZ |
5626 | |
5627 | update_max_interval(); | |
5628 | ||
1da177e4 LT |
5629 | return NOTIFY_OK; |
5630 | } | |
5631 | ||
f38b0820 PM |
5632 | /* |
5633 | * Register at high priority so that task migration (migrate_all_tasks) | |
5634 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5635 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5636 | */ |
26c2143b | 5637 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5638 | .notifier_call = migration_call, |
50a323b7 | 5639 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5640 | }; |
5641 | ||
3a101d05 TH |
5642 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5643 | unsigned long action, void *hcpu) | |
5644 | { | |
5645 | switch (action & ~CPU_TASKS_FROZEN) { | |
5fbd036b | 5646 | case CPU_STARTING: |
3a101d05 TH |
5647 | case CPU_DOWN_FAILED: |
5648 | set_cpu_active((long)hcpu, true); | |
5649 | return NOTIFY_OK; | |
5650 | default: | |
5651 | return NOTIFY_DONE; | |
5652 | } | |
5653 | } | |
5654 | ||
5655 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5656 | unsigned long action, void *hcpu) | |
5657 | { | |
5658 | switch (action & ~CPU_TASKS_FROZEN) { | |
5659 | case CPU_DOWN_PREPARE: | |
5660 | set_cpu_active((long)hcpu, false); | |
5661 | return NOTIFY_OK; | |
5662 | default: | |
5663 | return NOTIFY_DONE; | |
5664 | } | |
5665 | } | |
5666 | ||
7babe8db | 5667 | static int __init migration_init(void) |
1da177e4 LT |
5668 | { |
5669 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5670 | int err; |
48f24c4d | 5671 | |
3a101d05 | 5672 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5673 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5674 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5675 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5676 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5677 | |
3a101d05 TH |
5678 | /* Register cpu active notifiers */ |
5679 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5680 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5681 | ||
a004cd42 | 5682 | return 0; |
1da177e4 | 5683 | } |
7babe8db | 5684 | early_initcall(migration_init); |
1da177e4 LT |
5685 | #endif |
5686 | ||
5687 | #ifdef CONFIG_SMP | |
476f3534 | 5688 | |
4cb98839 PZ |
5689 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5690 | ||
3e9830dc | 5691 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5692 | |
d039ac60 | 5693 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5694 | |
d039ac60 | 5695 | static int __init sched_debug_setup(char *str) |
f6630114 | 5696 | { |
d039ac60 | 5697 | sched_debug_enabled = 1; |
f6630114 MT |
5698 | |
5699 | return 0; | |
5700 | } | |
d039ac60 PZ |
5701 | early_param("sched_debug", sched_debug_setup); |
5702 | ||
5703 | static inline bool sched_debug(void) | |
5704 | { | |
5705 | return sched_debug_enabled; | |
5706 | } | |
f6630114 | 5707 | |
7c16ec58 | 5708 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5709 | struct cpumask *groupmask) |
1da177e4 | 5710 | { |
4dcf6aff | 5711 | struct sched_group *group = sd->groups; |
434d53b0 | 5712 | char str[256]; |
1da177e4 | 5713 | |
968ea6d8 | 5714 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5715 | cpumask_clear(groupmask); |
4dcf6aff IM |
5716 | |
5717 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5718 | ||
5719 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5720 | printk("does not load-balance\n"); |
4dcf6aff | 5721 | if (sd->parent) |
3df0fc5b PZ |
5722 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5723 | " has parent"); | |
4dcf6aff | 5724 | return -1; |
41c7ce9a NP |
5725 | } |
5726 | ||
3df0fc5b | 5727 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5728 | |
758b2cdc | 5729 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5730 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5731 | "CPU%d\n", cpu); | |
4dcf6aff | 5732 | } |
758b2cdc | 5733 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5734 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5735 | " CPU%d\n", cpu); | |
4dcf6aff | 5736 | } |
1da177e4 | 5737 | |
4dcf6aff | 5738 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5739 | do { |
4dcf6aff | 5740 | if (!group) { |
3df0fc5b PZ |
5741 | printk("\n"); |
5742 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5743 | break; |
5744 | } | |
5745 | ||
c3decf0d PZ |
5746 | /* |
5747 | * Even though we initialize ->power to something semi-sane, | |
5748 | * we leave power_orig unset. This allows us to detect if | |
5749 | * domain iteration is still funny without causing /0 traps. | |
5750 | */ | |
5751 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5752 | printk(KERN_CONT "\n"); |
5753 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5754 | "set\n"); | |
4dcf6aff IM |
5755 | break; |
5756 | } | |
1da177e4 | 5757 | |
758b2cdc | 5758 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5759 | printk(KERN_CONT "\n"); |
5760 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5761 | break; |
5762 | } | |
1da177e4 | 5763 | |
cb83b629 PZ |
5764 | if (!(sd->flags & SD_OVERLAP) && |
5765 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5766 | printk(KERN_CONT "\n"); |
5767 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5768 | break; |
5769 | } | |
1da177e4 | 5770 | |
758b2cdc | 5771 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5772 | |
968ea6d8 | 5773 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5774 | |
3df0fc5b | 5775 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5776 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5777 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5778 | group->sgp->power); |
381512cf | 5779 | } |
1da177e4 | 5780 | |
4dcf6aff IM |
5781 | group = group->next; |
5782 | } while (group != sd->groups); | |
3df0fc5b | 5783 | printk(KERN_CONT "\n"); |
1da177e4 | 5784 | |
758b2cdc | 5785 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5786 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5787 | |
758b2cdc RR |
5788 | if (sd->parent && |
5789 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5790 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5791 | "of domain->span\n"); | |
4dcf6aff IM |
5792 | return 0; |
5793 | } | |
1da177e4 | 5794 | |
4dcf6aff IM |
5795 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5796 | { | |
5797 | int level = 0; | |
1da177e4 | 5798 | |
d039ac60 | 5799 | if (!sched_debug_enabled) |
f6630114 MT |
5800 | return; |
5801 | ||
4dcf6aff IM |
5802 | if (!sd) { |
5803 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5804 | return; | |
5805 | } | |
1da177e4 | 5806 | |
4dcf6aff IM |
5807 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5808 | ||
5809 | for (;;) { | |
4cb98839 | 5810 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5811 | break; |
1da177e4 LT |
5812 | level++; |
5813 | sd = sd->parent; | |
33859f7f | 5814 | if (!sd) |
4dcf6aff IM |
5815 | break; |
5816 | } | |
1da177e4 | 5817 | } |
6d6bc0ad | 5818 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5819 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5820 | static inline bool sched_debug(void) |
5821 | { | |
5822 | return false; | |
5823 | } | |
6d6bc0ad | 5824 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5825 | |
1a20ff27 | 5826 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5827 | { |
758b2cdc | 5828 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5829 | return 1; |
5830 | ||
5831 | /* Following flags need at least 2 groups */ | |
5832 | if (sd->flags & (SD_LOAD_BALANCE | | |
5833 | SD_BALANCE_NEWIDLE | | |
5834 | SD_BALANCE_FORK | | |
89c4710e SS |
5835 | SD_BALANCE_EXEC | |
5836 | SD_SHARE_CPUPOWER | | |
5837 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5838 | if (sd->groups != sd->groups->next) |
5839 | return 0; | |
5840 | } | |
5841 | ||
5842 | /* Following flags don't use groups */ | |
c88d5910 | 5843 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5844 | return 0; |
5845 | ||
5846 | return 1; | |
5847 | } | |
5848 | ||
48f24c4d IM |
5849 | static int |
5850 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5851 | { |
5852 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5853 | ||
5854 | if (sd_degenerate(parent)) | |
5855 | return 1; | |
5856 | ||
758b2cdc | 5857 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5858 | return 0; |
5859 | ||
245af2c7 SS |
5860 | /* Flags needing groups don't count if only 1 group in parent */ |
5861 | if (parent->groups == parent->groups->next) { | |
5862 | pflags &= ~(SD_LOAD_BALANCE | | |
5863 | SD_BALANCE_NEWIDLE | | |
5864 | SD_BALANCE_FORK | | |
89c4710e SS |
5865 | SD_BALANCE_EXEC | |
5866 | SD_SHARE_CPUPOWER | | |
5867 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5868 | if (nr_node_ids == 1) |
5869 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5870 | } |
5871 | if (~cflags & pflags) | |
5872 | return 0; | |
5873 | ||
5874 | return 1; | |
5875 | } | |
5876 | ||
dce840a0 | 5877 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5878 | { |
dce840a0 | 5879 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5880 | |
68e74568 | 5881 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
5882 | free_cpumask_var(rd->rto_mask); |
5883 | free_cpumask_var(rd->online); | |
5884 | free_cpumask_var(rd->span); | |
5885 | kfree(rd); | |
5886 | } | |
5887 | ||
57d885fe GH |
5888 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5889 | { | |
a0490fa3 | 5890 | struct root_domain *old_rd = NULL; |
57d885fe | 5891 | unsigned long flags; |
57d885fe | 5892 | |
05fa785c | 5893 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5894 | |
5895 | if (rq->rd) { | |
a0490fa3 | 5896 | old_rd = rq->rd; |
57d885fe | 5897 | |
c6c4927b | 5898 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5899 | set_rq_offline(rq); |
57d885fe | 5900 | |
c6c4927b | 5901 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5902 | |
a0490fa3 IM |
5903 | /* |
5904 | * If we dont want to free the old_rt yet then | |
5905 | * set old_rd to NULL to skip the freeing later | |
5906 | * in this function: | |
5907 | */ | |
5908 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5909 | old_rd = NULL; | |
57d885fe GH |
5910 | } |
5911 | ||
5912 | atomic_inc(&rd->refcount); | |
5913 | rq->rd = rd; | |
5914 | ||
c6c4927b | 5915 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5916 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5917 | set_rq_online(rq); |
57d885fe | 5918 | |
05fa785c | 5919 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5920 | |
5921 | if (old_rd) | |
dce840a0 | 5922 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5923 | } |
5924 | ||
68c38fc3 | 5925 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5926 | { |
5927 | memset(rd, 0, sizeof(*rd)); | |
5928 | ||
68c38fc3 | 5929 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5930 | goto out; |
68c38fc3 | 5931 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5932 | goto free_span; |
68c38fc3 | 5933 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 5934 | goto free_online; |
6e0534f2 | 5935 | |
68c38fc3 | 5936 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5937 | goto free_rto_mask; |
c6c4927b | 5938 | return 0; |
6e0534f2 | 5939 | |
68e74568 RR |
5940 | free_rto_mask: |
5941 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
5942 | free_online: |
5943 | free_cpumask_var(rd->online); | |
5944 | free_span: | |
5945 | free_cpumask_var(rd->span); | |
0c910d28 | 5946 | out: |
c6c4927b | 5947 | return -ENOMEM; |
57d885fe GH |
5948 | } |
5949 | ||
029632fb PZ |
5950 | /* |
5951 | * By default the system creates a single root-domain with all cpus as | |
5952 | * members (mimicking the global state we have today). | |
5953 | */ | |
5954 | struct root_domain def_root_domain; | |
5955 | ||
57d885fe GH |
5956 | static void init_defrootdomain(void) |
5957 | { | |
68c38fc3 | 5958 | init_rootdomain(&def_root_domain); |
c6c4927b | 5959 | |
57d885fe GH |
5960 | atomic_set(&def_root_domain.refcount, 1); |
5961 | } | |
5962 | ||
dc938520 | 5963 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5964 | { |
5965 | struct root_domain *rd; | |
5966 | ||
5967 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5968 | if (!rd) | |
5969 | return NULL; | |
5970 | ||
68c38fc3 | 5971 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5972 | kfree(rd); |
5973 | return NULL; | |
5974 | } | |
57d885fe GH |
5975 | |
5976 | return rd; | |
5977 | } | |
5978 | ||
e3589f6c PZ |
5979 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5980 | { | |
5981 | struct sched_group *tmp, *first; | |
5982 | ||
5983 | if (!sg) | |
5984 | return; | |
5985 | ||
5986 | first = sg; | |
5987 | do { | |
5988 | tmp = sg->next; | |
5989 | ||
5990 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5991 | kfree(sg->sgp); | |
5992 | ||
5993 | kfree(sg); | |
5994 | sg = tmp; | |
5995 | } while (sg != first); | |
5996 | } | |
5997 | ||
dce840a0 PZ |
5998 | static void free_sched_domain(struct rcu_head *rcu) |
5999 | { | |
6000 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
6001 | |
6002 | /* | |
6003 | * If its an overlapping domain it has private groups, iterate and | |
6004 | * nuke them all. | |
6005 | */ | |
6006 | if (sd->flags & SD_OVERLAP) { | |
6007 | free_sched_groups(sd->groups, 1); | |
6008 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 6009 | kfree(sd->groups->sgp); |
dce840a0 | 6010 | kfree(sd->groups); |
9c3f75cb | 6011 | } |
dce840a0 PZ |
6012 | kfree(sd); |
6013 | } | |
6014 | ||
6015 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
6016 | { | |
6017 | call_rcu(&sd->rcu, free_sched_domain); | |
6018 | } | |
6019 | ||
6020 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
6021 | { | |
6022 | for (; sd; sd = sd->parent) | |
6023 | destroy_sched_domain(sd, cpu); | |
6024 | } | |
6025 | ||
518cd623 PZ |
6026 | /* |
6027 | * Keep a special pointer to the highest sched_domain that has | |
6028 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
6029 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
6030 | * | |
970e1789 MG |
6031 | * Iterate domains and sched_groups downward, assigning CPUs to be |
6032 | * select_idle_sibling() hw buddy. Cross-wiring hw makes bouncing | |
6033 | * due to random perturbation self canceling, ie sw buddies pull | |
6034 | * their counterpart to their CPU's hw counterpart. | |
6035 | * | |
518cd623 PZ |
6036 | * Also keep a unique ID per domain (we use the first cpu number in |
6037 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 6038 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
6039 | */ |
6040 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
6041 | DEFINE_PER_CPU(int, sd_llc_id); | |
6042 | ||
6043 | static void update_top_cache_domain(int cpu) | |
6044 | { | |
6045 | struct sched_domain *sd; | |
6046 | int id = cpu; | |
6047 | ||
6048 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
970e1789 MG |
6049 | if (sd) { |
6050 | struct sched_domain *tmp = sd; | |
6051 | struct sched_group *sg, *prev; | |
6052 | bool right; | |
6053 | ||
6054 | /* | |
6055 | * Traverse to first CPU in group, and count hops | |
6056 | * to cpu from there, switching direction on each | |
6057 | * hop, never ever pointing the last CPU rightward. | |
6058 | */ | |
6059 | do { | |
6060 | id = cpumask_first(sched_domain_span(tmp)); | |
6061 | prev = sg = tmp->groups; | |
6062 | right = 1; | |
6063 | ||
6064 | while (cpumask_first(sched_group_cpus(sg)) != id) | |
6065 | sg = sg->next; | |
6066 | ||
6067 | while (!cpumask_test_cpu(cpu, sched_group_cpus(sg))) { | |
6068 | prev = sg; | |
6069 | sg = sg->next; | |
6070 | right = !right; | |
6071 | } | |
6072 | ||
6073 | /* A CPU went down, never point back to domain start. */ | |
6074 | if (right && cpumask_first(sched_group_cpus(sg->next)) == id) | |
6075 | right = false; | |
6076 | ||
6077 | sg = right ? sg->next : prev; | |
6078 | tmp->idle_buddy = cpumask_first(sched_group_cpus(sg)); | |
6079 | } while ((tmp = tmp->child)); | |
6080 | ||
518cd623 | 6081 | id = cpumask_first(sched_domain_span(sd)); |
970e1789 | 6082 | } |
518cd623 PZ |
6083 | |
6084 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
6085 | per_cpu(sd_llc_id, cpu) = id; | |
6086 | } | |
6087 | ||
1da177e4 | 6088 | /* |
0eab9146 | 6089 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6090 | * hold the hotplug lock. |
6091 | */ | |
0eab9146 IM |
6092 | static void |
6093 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6094 | { |
70b97a7f | 6095 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6096 | struct sched_domain *tmp; |
6097 | ||
6098 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6099 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6100 | struct sched_domain *parent = tmp->parent; |
6101 | if (!parent) | |
6102 | break; | |
f29c9b1c | 6103 | |
1a848870 | 6104 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6105 | tmp->parent = parent->parent; |
1a848870 SS |
6106 | if (parent->parent) |
6107 | parent->parent->child = tmp; | |
dce840a0 | 6108 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
6109 | } else |
6110 | tmp = tmp->parent; | |
245af2c7 SS |
6111 | } |
6112 | ||
1a848870 | 6113 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6114 | tmp = sd; |
245af2c7 | 6115 | sd = sd->parent; |
dce840a0 | 6116 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
6117 | if (sd) |
6118 | sd->child = NULL; | |
6119 | } | |
1da177e4 | 6120 | |
4cb98839 | 6121 | sched_domain_debug(sd, cpu); |
1da177e4 | 6122 | |
57d885fe | 6123 | rq_attach_root(rq, rd); |
dce840a0 | 6124 | tmp = rq->sd; |
674311d5 | 6125 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 6126 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
6127 | |
6128 | update_top_cache_domain(cpu); | |
1da177e4 LT |
6129 | } |
6130 | ||
6131 | /* cpus with isolated domains */ | |
dcc30a35 | 6132 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6133 | |
6134 | /* Setup the mask of cpus configured for isolated domains */ | |
6135 | static int __init isolated_cpu_setup(char *str) | |
6136 | { | |
bdddd296 | 6137 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6138 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6139 | return 1; |
6140 | } | |
6141 | ||
8927f494 | 6142 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6143 | |
d3081f52 PZ |
6144 | static const struct cpumask *cpu_cpu_mask(int cpu) |
6145 | { | |
6146 | return cpumask_of_node(cpu_to_node(cpu)); | |
6147 | } | |
6148 | ||
dce840a0 PZ |
6149 | struct sd_data { |
6150 | struct sched_domain **__percpu sd; | |
6151 | struct sched_group **__percpu sg; | |
9c3f75cb | 6152 | struct sched_group_power **__percpu sgp; |
dce840a0 PZ |
6153 | }; |
6154 | ||
49a02c51 | 6155 | struct s_data { |
21d42ccf | 6156 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
6157 | struct root_domain *rd; |
6158 | }; | |
6159 | ||
2109b99e | 6160 | enum s_alloc { |
2109b99e | 6161 | sa_rootdomain, |
21d42ccf | 6162 | sa_sd, |
dce840a0 | 6163 | sa_sd_storage, |
2109b99e AH |
6164 | sa_none, |
6165 | }; | |
6166 | ||
54ab4ff4 PZ |
6167 | struct sched_domain_topology_level; |
6168 | ||
6169 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
6170 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
6171 | ||
e3589f6c PZ |
6172 | #define SDTL_OVERLAP 0x01 |
6173 | ||
eb7a74e6 | 6174 | struct sched_domain_topology_level { |
2c402dc3 PZ |
6175 | sched_domain_init_f init; |
6176 | sched_domain_mask_f mask; | |
e3589f6c | 6177 | int flags; |
cb83b629 | 6178 | int numa_level; |
54ab4ff4 | 6179 | struct sd_data data; |
eb7a74e6 PZ |
6180 | }; |
6181 | ||
c1174876 PZ |
6182 | /* |
6183 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
6184 | * domain traversal. | |
6185 | * | |
6186 | * Asymmetric node setups can result in situations where the domain tree is of | |
6187 | * unequal depth, make sure to skip domains that already cover the entire | |
6188 | * range. | |
6189 | * | |
6190 | * In that case build_sched_domains() will have terminated the iteration early | |
6191 | * and our sibling sd spans will be empty. Domains should always include the | |
6192 | * cpu they're built on, so check that. | |
6193 | * | |
6194 | */ | |
6195 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
6196 | { | |
6197 | const struct cpumask *span = sched_domain_span(sd); | |
6198 | struct sd_data *sdd = sd->private; | |
6199 | struct sched_domain *sibling; | |
6200 | int i; | |
6201 | ||
6202 | for_each_cpu(i, span) { | |
6203 | sibling = *per_cpu_ptr(sdd->sd, i); | |
6204 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
6205 | continue; | |
6206 | ||
6207 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
6208 | } | |
6209 | } | |
6210 | ||
6211 | /* | |
6212 | * Return the canonical balance cpu for this group, this is the first cpu | |
6213 | * of this group that's also in the iteration mask. | |
6214 | */ | |
6215 | int group_balance_cpu(struct sched_group *sg) | |
6216 | { | |
6217 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
6218 | } | |
6219 | ||
e3589f6c PZ |
6220 | static int |
6221 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
6222 | { | |
6223 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
6224 | const struct cpumask *span = sched_domain_span(sd); | |
6225 | struct cpumask *covered = sched_domains_tmpmask; | |
6226 | struct sd_data *sdd = sd->private; | |
6227 | struct sched_domain *child; | |
6228 | int i; | |
6229 | ||
6230 | cpumask_clear(covered); | |
6231 | ||
6232 | for_each_cpu(i, span) { | |
6233 | struct cpumask *sg_span; | |
6234 | ||
6235 | if (cpumask_test_cpu(i, covered)) | |
6236 | continue; | |
6237 | ||
c1174876 PZ |
6238 | child = *per_cpu_ptr(sdd->sd, i); |
6239 | ||
6240 | /* See the comment near build_group_mask(). */ | |
6241 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
6242 | continue; | |
6243 | ||
e3589f6c | 6244 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 6245 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
6246 | |
6247 | if (!sg) | |
6248 | goto fail; | |
6249 | ||
6250 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
6251 | if (child->child) { |
6252 | child = child->child; | |
6253 | cpumask_copy(sg_span, sched_domain_span(child)); | |
6254 | } else | |
6255 | cpumask_set_cpu(i, sg_span); | |
6256 | ||
6257 | cpumask_or(covered, covered, sg_span); | |
6258 | ||
74a5ce20 | 6259 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
6260 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
6261 | build_group_mask(sd, sg); | |
6262 | ||
c3decf0d PZ |
6263 | /* |
6264 | * Initialize sgp->power such that even if we mess up the | |
6265 | * domains and no possible iteration will get us here, we won't | |
6266 | * die on a /0 trap. | |
6267 | */ | |
6268 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
e3589f6c | 6269 | |
c1174876 PZ |
6270 | /* |
6271 | * Make sure the first group of this domain contains the | |
6272 | * canonical balance cpu. Otherwise the sched_domain iteration | |
6273 | * breaks. See update_sg_lb_stats(). | |
6274 | */ | |
74a5ce20 | 6275 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 6276 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
6277 | groups = sg; |
6278 | ||
6279 | if (!first) | |
6280 | first = sg; | |
6281 | if (last) | |
6282 | last->next = sg; | |
6283 | last = sg; | |
6284 | last->next = first; | |
6285 | } | |
6286 | sd->groups = groups; | |
6287 | ||
6288 | return 0; | |
6289 | ||
6290 | fail: | |
6291 | free_sched_groups(first, 0); | |
6292 | ||
6293 | return -ENOMEM; | |
6294 | } | |
6295 | ||
dce840a0 | 6296 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6297 | { |
dce840a0 PZ |
6298 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6299 | struct sched_domain *child = sd->child; | |
1da177e4 | 6300 | |
dce840a0 PZ |
6301 | if (child) |
6302 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6303 | |
9c3f75cb | 6304 | if (sg) { |
dce840a0 | 6305 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 6306 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 6307 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 6308 | } |
dce840a0 PZ |
6309 | |
6310 | return cpu; | |
1e9f28fa | 6311 | } |
1e9f28fa | 6312 | |
01a08546 | 6313 | /* |
dce840a0 PZ |
6314 | * build_sched_groups will build a circular linked list of the groups |
6315 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6316 | * and ->cpu_power to 0. | |
e3589f6c PZ |
6317 | * |
6318 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 6319 | */ |
e3589f6c PZ |
6320 | static int |
6321 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 6322 | { |
dce840a0 PZ |
6323 | struct sched_group *first = NULL, *last = NULL; |
6324 | struct sd_data *sdd = sd->private; | |
6325 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6326 | struct cpumask *covered; |
dce840a0 | 6327 | int i; |
9c1cfda2 | 6328 | |
e3589f6c PZ |
6329 | get_group(cpu, sdd, &sd->groups); |
6330 | atomic_inc(&sd->groups->ref); | |
6331 | ||
6332 | if (cpu != cpumask_first(sched_domain_span(sd))) | |
6333 | return 0; | |
6334 | ||
f96225fd PZ |
6335 | lockdep_assert_held(&sched_domains_mutex); |
6336 | covered = sched_domains_tmpmask; | |
6337 | ||
dce840a0 | 6338 | cpumask_clear(covered); |
6711cab4 | 6339 | |
dce840a0 PZ |
6340 | for_each_cpu(i, span) { |
6341 | struct sched_group *sg; | |
6342 | int group = get_group(i, sdd, &sg); | |
6343 | int j; | |
6711cab4 | 6344 | |
dce840a0 PZ |
6345 | if (cpumask_test_cpu(i, covered)) |
6346 | continue; | |
6711cab4 | 6347 | |
dce840a0 | 6348 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 6349 | sg->sgp->power = 0; |
c1174876 | 6350 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 6351 | |
dce840a0 PZ |
6352 | for_each_cpu(j, span) { |
6353 | if (get_group(j, sdd, NULL) != group) | |
6354 | continue; | |
0601a88d | 6355 | |
dce840a0 PZ |
6356 | cpumask_set_cpu(j, covered); |
6357 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
6358 | } | |
0601a88d | 6359 | |
dce840a0 PZ |
6360 | if (!first) |
6361 | first = sg; | |
6362 | if (last) | |
6363 | last->next = sg; | |
6364 | last = sg; | |
6365 | } | |
6366 | last->next = first; | |
e3589f6c PZ |
6367 | |
6368 | return 0; | |
0601a88d | 6369 | } |
51888ca2 | 6370 | |
89c4710e SS |
6371 | /* |
6372 | * Initialize sched groups cpu_power. | |
6373 | * | |
6374 | * cpu_power indicates the capacity of sched group, which is used while | |
6375 | * distributing the load between different sched groups in a sched domain. | |
6376 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6377 | * there are asymmetries in the topology. If there are asymmetries, group | |
6378 | * having more cpu_power will pickup more load compared to the group having | |
6379 | * less cpu_power. | |
89c4710e SS |
6380 | */ |
6381 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6382 | { | |
e3589f6c | 6383 | struct sched_group *sg = sd->groups; |
89c4710e | 6384 | |
e3589f6c PZ |
6385 | WARN_ON(!sd || !sg); |
6386 | ||
6387 | do { | |
6388 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
6389 | sg = sg->next; | |
6390 | } while (sg != sd->groups); | |
89c4710e | 6391 | |
c1174876 | 6392 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 6393 | return; |
aae6d3dd | 6394 | |
d274cb30 | 6395 | update_group_power(sd, cpu); |
69e1e811 | 6396 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
6397 | } |
6398 | ||
029632fb PZ |
6399 | int __weak arch_sd_sibling_asym_packing(void) |
6400 | { | |
6401 | return 0*SD_ASYM_PACKING; | |
89c4710e SS |
6402 | } |
6403 | ||
7c16ec58 MT |
6404 | /* |
6405 | * Initializers for schedule domains | |
6406 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6407 | */ | |
6408 | ||
a5d8c348 IM |
6409 | #ifdef CONFIG_SCHED_DEBUG |
6410 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6411 | #else | |
6412 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6413 | #endif | |
6414 | ||
54ab4ff4 PZ |
6415 | #define SD_INIT_FUNC(type) \ |
6416 | static noinline struct sched_domain * \ | |
6417 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
6418 | { \ | |
6419 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
6420 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
6421 | SD_INIT_NAME(sd, type); \ |
6422 | sd->private = &tl->data; \ | |
6423 | return sd; \ | |
7c16ec58 MT |
6424 | } |
6425 | ||
6426 | SD_INIT_FUNC(CPU) | |
7c16ec58 MT |
6427 | #ifdef CONFIG_SCHED_SMT |
6428 | SD_INIT_FUNC(SIBLING) | |
6429 | #endif | |
6430 | #ifdef CONFIG_SCHED_MC | |
6431 | SD_INIT_FUNC(MC) | |
6432 | #endif | |
01a08546 HC |
6433 | #ifdef CONFIG_SCHED_BOOK |
6434 | SD_INIT_FUNC(BOOK) | |
6435 | #endif | |
7c16ec58 | 6436 | |
1d3504fc | 6437 | static int default_relax_domain_level = -1; |
60495e77 | 6438 | int sched_domain_level_max; |
1d3504fc HS |
6439 | |
6440 | static int __init setup_relax_domain_level(char *str) | |
6441 | { | |
a841f8ce DS |
6442 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
6443 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 6444 | |
1d3504fc HS |
6445 | return 1; |
6446 | } | |
6447 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6448 | ||
6449 | static void set_domain_attribute(struct sched_domain *sd, | |
6450 | struct sched_domain_attr *attr) | |
6451 | { | |
6452 | int request; | |
6453 | ||
6454 | if (!attr || attr->relax_domain_level < 0) { | |
6455 | if (default_relax_domain_level < 0) | |
6456 | return; | |
6457 | else | |
6458 | request = default_relax_domain_level; | |
6459 | } else | |
6460 | request = attr->relax_domain_level; | |
6461 | if (request < sd->level) { | |
6462 | /* turn off idle balance on this domain */ | |
c88d5910 | 6463 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6464 | } else { |
6465 | /* turn on idle balance on this domain */ | |
c88d5910 | 6466 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6467 | } |
6468 | } | |
6469 | ||
54ab4ff4 PZ |
6470 | static void __sdt_free(const struct cpumask *cpu_map); |
6471 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
6472 | ||
2109b99e AH |
6473 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6474 | const struct cpumask *cpu_map) | |
6475 | { | |
6476 | switch (what) { | |
2109b99e | 6477 | case sa_rootdomain: |
822ff793 PZ |
6478 | if (!atomic_read(&d->rd->refcount)) |
6479 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
6480 | case sa_sd: |
6481 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 6482 | case sa_sd_storage: |
54ab4ff4 | 6483 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
6484 | case sa_none: |
6485 | break; | |
6486 | } | |
6487 | } | |
3404c8d9 | 6488 | |
2109b99e AH |
6489 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6490 | const struct cpumask *cpu_map) | |
6491 | { | |
dce840a0 PZ |
6492 | memset(d, 0, sizeof(*d)); |
6493 | ||
54ab4ff4 PZ |
6494 | if (__sdt_alloc(cpu_map)) |
6495 | return sa_sd_storage; | |
dce840a0 PZ |
6496 | d->sd = alloc_percpu(struct sched_domain *); |
6497 | if (!d->sd) | |
6498 | return sa_sd_storage; | |
2109b99e | 6499 | d->rd = alloc_rootdomain(); |
dce840a0 | 6500 | if (!d->rd) |
21d42ccf | 6501 | return sa_sd; |
2109b99e AH |
6502 | return sa_rootdomain; |
6503 | } | |
57d885fe | 6504 | |
dce840a0 PZ |
6505 | /* |
6506 | * NULL the sd_data elements we've used to build the sched_domain and | |
6507 | * sched_group structure so that the subsequent __free_domain_allocs() | |
6508 | * will not free the data we're using. | |
6509 | */ | |
6510 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
6511 | { | |
6512 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
6513 | |
6514 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
6515 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
6516 | ||
e3589f6c | 6517 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 6518 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
6519 | |
6520 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 6521 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
6522 | } |
6523 | ||
2c402dc3 PZ |
6524 | #ifdef CONFIG_SCHED_SMT |
6525 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 6526 | { |
2c402dc3 | 6527 | return topology_thread_cpumask(cpu); |
3bd65a80 | 6528 | } |
2c402dc3 | 6529 | #endif |
7f4588f3 | 6530 | |
d069b916 PZ |
6531 | /* |
6532 | * Topology list, bottom-up. | |
6533 | */ | |
2c402dc3 | 6534 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
6535 | #ifdef CONFIG_SCHED_SMT |
6536 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 6537 | #endif |
1e9f28fa | 6538 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 6539 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 6540 | #endif |
d069b916 PZ |
6541 | #ifdef CONFIG_SCHED_BOOK |
6542 | { sd_init_BOOK, cpu_book_mask, }, | |
6543 | #endif | |
6544 | { sd_init_CPU, cpu_cpu_mask, }, | |
eb7a74e6 PZ |
6545 | { NULL, }, |
6546 | }; | |
6547 | ||
6548 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6549 | ||
cb83b629 PZ |
6550 | #ifdef CONFIG_NUMA |
6551 | ||
6552 | static int sched_domains_numa_levels; | |
cb83b629 PZ |
6553 | static int *sched_domains_numa_distance; |
6554 | static struct cpumask ***sched_domains_numa_masks; | |
6555 | static int sched_domains_curr_level; | |
6556 | ||
cb83b629 PZ |
6557 | static inline int sd_local_flags(int level) |
6558 | { | |
10717dcd | 6559 | if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE) |
cb83b629 PZ |
6560 | return 0; |
6561 | ||
6562 | return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE; | |
6563 | } | |
6564 | ||
6565 | static struct sched_domain * | |
6566 | sd_numa_init(struct sched_domain_topology_level *tl, int cpu) | |
6567 | { | |
6568 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
6569 | int level = tl->numa_level; | |
6570 | int sd_weight = cpumask_weight( | |
6571 | sched_domains_numa_masks[level][cpu_to_node(cpu)]); | |
6572 | ||
6573 | *sd = (struct sched_domain){ | |
6574 | .min_interval = sd_weight, | |
6575 | .max_interval = 2*sd_weight, | |
6576 | .busy_factor = 32, | |
870a0bb5 | 6577 | .imbalance_pct = 125, |
cb83b629 PZ |
6578 | .cache_nice_tries = 2, |
6579 | .busy_idx = 3, | |
6580 | .idle_idx = 2, | |
6581 | .newidle_idx = 0, | |
6582 | .wake_idx = 0, | |
6583 | .forkexec_idx = 0, | |
6584 | ||
6585 | .flags = 1*SD_LOAD_BALANCE | |
6586 | | 1*SD_BALANCE_NEWIDLE | |
6587 | | 0*SD_BALANCE_EXEC | |
6588 | | 0*SD_BALANCE_FORK | |
6589 | | 0*SD_BALANCE_WAKE | |
6590 | | 0*SD_WAKE_AFFINE | |
6591 | | 0*SD_PREFER_LOCAL | |
6592 | | 0*SD_SHARE_CPUPOWER | |
cb83b629 PZ |
6593 | | 0*SD_SHARE_PKG_RESOURCES |
6594 | | 1*SD_SERIALIZE | |
6595 | | 0*SD_PREFER_SIBLING | |
6596 | | sd_local_flags(level) | |
6597 | , | |
6598 | .last_balance = jiffies, | |
6599 | .balance_interval = sd_weight, | |
6600 | }; | |
6601 | SD_INIT_NAME(sd, NUMA); | |
6602 | sd->private = &tl->data; | |
6603 | ||
6604 | /* | |
6605 | * Ugly hack to pass state to sd_numa_mask()... | |
6606 | */ | |
6607 | sched_domains_curr_level = tl->numa_level; | |
6608 | ||
6609 | return sd; | |
6610 | } | |
6611 | ||
6612 | static const struct cpumask *sd_numa_mask(int cpu) | |
6613 | { | |
6614 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6615 | } | |
6616 | ||
d039ac60 PZ |
6617 | static void sched_numa_warn(const char *str) |
6618 | { | |
6619 | static int done = false; | |
6620 | int i,j; | |
6621 | ||
6622 | if (done) | |
6623 | return; | |
6624 | ||
6625 | done = true; | |
6626 | ||
6627 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6628 | ||
6629 | for (i = 0; i < nr_node_ids; i++) { | |
6630 | printk(KERN_WARNING " "); | |
6631 | for (j = 0; j < nr_node_ids; j++) | |
6632 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6633 | printk(KERN_CONT "\n"); | |
6634 | } | |
6635 | printk(KERN_WARNING "\n"); | |
6636 | } | |
6637 | ||
6638 | static bool find_numa_distance(int distance) | |
6639 | { | |
6640 | int i; | |
6641 | ||
6642 | if (distance == node_distance(0, 0)) | |
6643 | return true; | |
6644 | ||
6645 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6646 | if (sched_domains_numa_distance[i] == distance) | |
6647 | return true; | |
6648 | } | |
6649 | ||
6650 | return false; | |
6651 | } | |
6652 | ||
cb83b629 PZ |
6653 | static void sched_init_numa(void) |
6654 | { | |
6655 | int next_distance, curr_distance = node_distance(0, 0); | |
6656 | struct sched_domain_topology_level *tl; | |
6657 | int level = 0; | |
6658 | int i, j, k; | |
6659 | ||
cb83b629 PZ |
6660 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6661 | if (!sched_domains_numa_distance) | |
6662 | return; | |
6663 | ||
6664 | /* | |
6665 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6666 | * unique distances in the node_distance() table. | |
6667 | * | |
6668 | * Assumes node_distance(0,j) includes all distances in | |
6669 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6670 | */ |
6671 | next_distance = curr_distance; | |
6672 | for (i = 0; i < nr_node_ids; i++) { | |
6673 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6674 | for (k = 0; k < nr_node_ids; k++) { |
6675 | int distance = node_distance(i, k); | |
6676 | ||
6677 | if (distance > curr_distance && | |
6678 | (distance < next_distance || | |
6679 | next_distance == curr_distance)) | |
6680 | next_distance = distance; | |
6681 | ||
6682 | /* | |
6683 | * While not a strong assumption it would be nice to know | |
6684 | * about cases where if node A is connected to B, B is not | |
6685 | * equally connected to A. | |
6686 | */ | |
6687 | if (sched_debug() && node_distance(k, i) != distance) | |
6688 | sched_numa_warn("Node-distance not symmetric"); | |
6689 | ||
6690 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6691 | sched_numa_warn("Node-0 not representative"); | |
6692 | } | |
6693 | if (next_distance != curr_distance) { | |
6694 | sched_domains_numa_distance[level++] = next_distance; | |
6695 | sched_domains_numa_levels = level; | |
6696 | curr_distance = next_distance; | |
6697 | } else break; | |
cb83b629 | 6698 | } |
d039ac60 PZ |
6699 | |
6700 | /* | |
6701 | * In case of sched_debug() we verify the above assumption. | |
6702 | */ | |
6703 | if (!sched_debug()) | |
6704 | break; | |
cb83b629 PZ |
6705 | } |
6706 | /* | |
6707 | * 'level' contains the number of unique distances, excluding the | |
6708 | * identity distance node_distance(i,i). | |
6709 | * | |
6710 | * The sched_domains_nume_distance[] array includes the actual distance | |
6711 | * numbers. | |
6712 | */ | |
6713 | ||
6714 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); | |
6715 | if (!sched_domains_numa_masks) | |
6716 | return; | |
6717 | ||
6718 | /* | |
6719 | * Now for each level, construct a mask per node which contains all | |
6720 | * cpus of nodes that are that many hops away from us. | |
6721 | */ | |
6722 | for (i = 0; i < level; i++) { | |
6723 | sched_domains_numa_masks[i] = | |
6724 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6725 | if (!sched_domains_numa_masks[i]) | |
6726 | return; | |
6727 | ||
6728 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6729 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6730 | if (!mask) |
6731 | return; | |
6732 | ||
6733 | sched_domains_numa_masks[i][j] = mask; | |
6734 | ||
6735 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6736 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6737 | continue; |
6738 | ||
6739 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6740 | } | |
6741 | } | |
6742 | } | |
6743 | ||
6744 | tl = kzalloc((ARRAY_SIZE(default_topology) + level) * | |
6745 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); | |
6746 | if (!tl) | |
6747 | return; | |
6748 | ||
6749 | /* | |
6750 | * Copy the default topology bits.. | |
6751 | */ | |
6752 | for (i = 0; default_topology[i].init; i++) | |
6753 | tl[i] = default_topology[i]; | |
6754 | ||
6755 | /* | |
6756 | * .. and append 'j' levels of NUMA goodness. | |
6757 | */ | |
6758 | for (j = 0; j < level; i++, j++) { | |
6759 | tl[i] = (struct sched_domain_topology_level){ | |
6760 | .init = sd_numa_init, | |
6761 | .mask = sd_numa_mask, | |
6762 | .flags = SDTL_OVERLAP, | |
6763 | .numa_level = j, | |
6764 | }; | |
6765 | } | |
6766 | ||
6767 | sched_domain_topology = tl; | |
6768 | } | |
6769 | #else | |
6770 | static inline void sched_init_numa(void) | |
6771 | { | |
6772 | } | |
6773 | #endif /* CONFIG_NUMA */ | |
6774 | ||
54ab4ff4 PZ |
6775 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6776 | { | |
6777 | struct sched_domain_topology_level *tl; | |
6778 | int j; | |
6779 | ||
6780 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6781 | struct sd_data *sdd = &tl->data; | |
6782 | ||
6783 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6784 | if (!sdd->sd) | |
6785 | return -ENOMEM; | |
6786 | ||
6787 | sdd->sg = alloc_percpu(struct sched_group *); | |
6788 | if (!sdd->sg) | |
6789 | return -ENOMEM; | |
6790 | ||
9c3f75cb PZ |
6791 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6792 | if (!sdd->sgp) | |
6793 | return -ENOMEM; | |
6794 | ||
54ab4ff4 PZ |
6795 | for_each_cpu(j, cpu_map) { |
6796 | struct sched_domain *sd; | |
6797 | struct sched_group *sg; | |
9c3f75cb | 6798 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6799 | |
6800 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6801 | GFP_KERNEL, cpu_to_node(j)); | |
6802 | if (!sd) | |
6803 | return -ENOMEM; | |
6804 | ||
6805 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6806 | ||
6807 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6808 | GFP_KERNEL, cpu_to_node(j)); | |
6809 | if (!sg) | |
6810 | return -ENOMEM; | |
6811 | ||
30b4e9eb IM |
6812 | sg->next = sg; |
6813 | ||
54ab4ff4 | 6814 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6815 | |
c1174876 | 6816 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6817 | GFP_KERNEL, cpu_to_node(j)); |
6818 | if (!sgp) | |
6819 | return -ENOMEM; | |
6820 | ||
6821 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6822 | } |
6823 | } | |
6824 | ||
6825 | return 0; | |
6826 | } | |
6827 | ||
6828 | static void __sdt_free(const struct cpumask *cpu_map) | |
6829 | { | |
6830 | struct sched_domain_topology_level *tl; | |
6831 | int j; | |
6832 | ||
6833 | for (tl = sched_domain_topology; tl->init; tl++) { | |
6834 | struct sd_data *sdd = &tl->data; | |
6835 | ||
6836 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6837 | struct sched_domain *sd; |
6838 | ||
6839 | if (sdd->sd) { | |
6840 | sd = *per_cpu_ptr(sdd->sd, j); | |
6841 | if (sd && (sd->flags & SD_OVERLAP)) | |
6842 | free_sched_groups(sd->groups, 0); | |
6843 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6844 | } | |
6845 | ||
6846 | if (sdd->sg) | |
6847 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6848 | if (sdd->sgp) | |
6849 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6850 | } |
6851 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6852 | sdd->sd = NULL; |
54ab4ff4 | 6853 | free_percpu(sdd->sg); |
fb2cf2c6 | 6854 | sdd->sg = NULL; |
9c3f75cb | 6855 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6856 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6857 | } |
6858 | } | |
6859 | ||
2c402dc3 PZ |
6860 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
6861 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 6862 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
6863 | int cpu) |
6864 | { | |
54ab4ff4 | 6865 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 6866 | if (!sd) |
d069b916 | 6867 | return child; |
2c402dc3 | 6868 | |
2c402dc3 | 6869 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6870 | if (child) { |
6871 | sd->level = child->level + 1; | |
6872 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6873 | child->parent = sd; |
60495e77 | 6874 | } |
d069b916 | 6875 | sd->child = child; |
a841f8ce | 6876 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6877 | |
6878 | return sd; | |
6879 | } | |
6880 | ||
2109b99e AH |
6881 | /* |
6882 | * Build sched domains for a given set of cpus and attach the sched domains | |
6883 | * to the individual cpus | |
6884 | */ | |
dce840a0 PZ |
6885 | static int build_sched_domains(const struct cpumask *cpu_map, |
6886 | struct sched_domain_attr *attr) | |
2109b99e AH |
6887 | { |
6888 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 6889 | struct sched_domain *sd; |
2109b99e | 6890 | struct s_data d; |
822ff793 | 6891 | int i, ret = -ENOMEM; |
9c1cfda2 | 6892 | |
2109b99e AH |
6893 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6894 | if (alloc_state != sa_rootdomain) | |
6895 | goto error; | |
9c1cfda2 | 6896 | |
dce840a0 | 6897 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6898 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6899 | struct sched_domain_topology_level *tl; |
6900 | ||
3bd65a80 | 6901 | sd = NULL; |
e3589f6c | 6902 | for (tl = sched_domain_topology; tl->init; tl++) { |
2c402dc3 | 6903 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); |
e3589f6c PZ |
6904 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6905 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6906 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6907 | break; | |
e3589f6c | 6908 | } |
d274cb30 | 6909 | |
d069b916 PZ |
6910 | while (sd->child) |
6911 | sd = sd->child; | |
6912 | ||
21d42ccf | 6913 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
6914 | } |
6915 | ||
6916 | /* Build the groups for the domains */ | |
6917 | for_each_cpu(i, cpu_map) { | |
6918 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6919 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6920 | if (sd->flags & SD_OVERLAP) { |
6921 | if (build_overlap_sched_groups(sd, i)) | |
6922 | goto error; | |
6923 | } else { | |
6924 | if (build_sched_groups(sd, i)) | |
6925 | goto error; | |
6926 | } | |
1cf51902 | 6927 | } |
a06dadbe | 6928 | } |
9c1cfda2 | 6929 | |
1da177e4 | 6930 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6931 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6932 | if (!cpumask_test_cpu(i, cpu_map)) | |
6933 | continue; | |
9c1cfda2 | 6934 | |
dce840a0 PZ |
6935 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6936 | claim_allocations(i, sd); | |
cd4ea6ae | 6937 | init_sched_groups_power(i, sd); |
dce840a0 | 6938 | } |
f712c0c7 | 6939 | } |
9c1cfda2 | 6940 | |
1da177e4 | 6941 | /* Attach the domains */ |
dce840a0 | 6942 | rcu_read_lock(); |
abcd083a | 6943 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6944 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6945 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6946 | } |
dce840a0 | 6947 | rcu_read_unlock(); |
51888ca2 | 6948 | |
822ff793 | 6949 | ret = 0; |
51888ca2 | 6950 | error: |
2109b99e | 6951 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6952 | return ret; |
1da177e4 | 6953 | } |
029190c5 | 6954 | |
acc3f5d7 | 6955 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6956 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6957 | static struct sched_domain_attr *dattr_cur; |
6958 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6959 | |
6960 | /* | |
6961 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6962 | * cpumask) fails, then fallback to a single sched domain, |
6963 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6964 | */ |
4212823f | 6965 | static cpumask_var_t fallback_doms; |
029190c5 | 6966 | |
ee79d1bd HC |
6967 | /* |
6968 | * arch_update_cpu_topology lets virtualized architectures update the | |
6969 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6970 | * or 0 if it stayed the same. | |
6971 | */ | |
6972 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 6973 | { |
ee79d1bd | 6974 | return 0; |
22e52b07 HC |
6975 | } |
6976 | ||
acc3f5d7 RR |
6977 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6978 | { | |
6979 | int i; | |
6980 | cpumask_var_t *doms; | |
6981 | ||
6982 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6983 | if (!doms) | |
6984 | return NULL; | |
6985 | for (i = 0; i < ndoms; i++) { | |
6986 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6987 | free_sched_domains(doms, i); | |
6988 | return NULL; | |
6989 | } | |
6990 | } | |
6991 | return doms; | |
6992 | } | |
6993 | ||
6994 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6995 | { | |
6996 | unsigned int i; | |
6997 | for (i = 0; i < ndoms; i++) | |
6998 | free_cpumask_var(doms[i]); | |
6999 | kfree(doms); | |
7000 | } | |
7001 | ||
1a20ff27 | 7002 | /* |
41a2d6cf | 7003 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7004 | * For now this just excludes isolated cpus, but could be used to |
7005 | * exclude other special cases in the future. | |
1a20ff27 | 7006 | */ |
c4a8849a | 7007 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7008 | { |
7378547f MM |
7009 | int err; |
7010 | ||
22e52b07 | 7011 | arch_update_cpu_topology(); |
029190c5 | 7012 | ndoms_cur = 1; |
acc3f5d7 | 7013 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7014 | if (!doms_cur) |
acc3f5d7 RR |
7015 | doms_cur = &fallback_doms; |
7016 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 7017 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7018 | register_sched_domain_sysctl(); |
7378547f MM |
7019 | |
7020 | return err; | |
1a20ff27 DG |
7021 | } |
7022 | ||
1a20ff27 DG |
7023 | /* |
7024 | * Detach sched domains from a group of cpus specified in cpu_map | |
7025 | * These cpus will now be attached to the NULL domain | |
7026 | */ | |
96f874e2 | 7027 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7028 | { |
7029 | int i; | |
7030 | ||
dce840a0 | 7031 | rcu_read_lock(); |
abcd083a | 7032 | for_each_cpu(i, cpu_map) |
57d885fe | 7033 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7034 | rcu_read_unlock(); |
1a20ff27 DG |
7035 | } |
7036 | ||
1d3504fc HS |
7037 | /* handle null as "default" */ |
7038 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7039 | struct sched_domain_attr *new, int idx_new) | |
7040 | { | |
7041 | struct sched_domain_attr tmp; | |
7042 | ||
7043 | /* fast path */ | |
7044 | if (!new && !cur) | |
7045 | return 1; | |
7046 | ||
7047 | tmp = SD_ATTR_INIT; | |
7048 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7049 | new ? (new + idx_new) : &tmp, | |
7050 | sizeof(struct sched_domain_attr)); | |
7051 | } | |
7052 | ||
029190c5 PJ |
7053 | /* |
7054 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7055 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7056 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7057 | * It destroys each deleted domain and builds each new domain. | |
7058 | * | |
acc3f5d7 | 7059 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7060 | * The masks don't intersect (don't overlap.) We should setup one |
7061 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7062 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7063 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7064 | * it as it is. | |
7065 | * | |
acc3f5d7 RR |
7066 | * The passed in 'doms_new' should be allocated using |
7067 | * alloc_sched_domains. This routine takes ownership of it and will | |
7068 | * free_sched_domains it when done with it. If the caller failed the | |
7069 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7070 | * and partition_sched_domains() will fallback to the single partition | |
7071 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7072 | * |
96f874e2 | 7073 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7074 | * ndoms_new == 0 is a special case for destroying existing domains, |
7075 | * and it will not create the default domain. | |
dfb512ec | 7076 | * |
029190c5 PJ |
7077 | * Call with hotplug lock held |
7078 | */ | |
acc3f5d7 | 7079 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7080 | struct sched_domain_attr *dattr_new) |
029190c5 | 7081 | { |
dfb512ec | 7082 | int i, j, n; |
d65bd5ec | 7083 | int new_topology; |
029190c5 | 7084 | |
712555ee | 7085 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7086 | |
7378547f MM |
7087 | /* always unregister in case we don't destroy any domains */ |
7088 | unregister_sched_domain_sysctl(); | |
7089 | ||
d65bd5ec HC |
7090 | /* Let architecture update cpu core mappings. */ |
7091 | new_topology = arch_update_cpu_topology(); | |
7092 | ||
dfb512ec | 7093 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7094 | |
7095 | /* Destroy deleted domains */ | |
7096 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7097 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7098 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7099 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7100 | goto match1; |
7101 | } | |
7102 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7103 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7104 | match1: |
7105 | ; | |
7106 | } | |
7107 | ||
e761b772 MK |
7108 | if (doms_new == NULL) { |
7109 | ndoms_cur = 0; | |
acc3f5d7 | 7110 | doms_new = &fallback_doms; |
6ad4c188 | 7111 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7112 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7113 | } |
7114 | ||
029190c5 PJ |
7115 | /* Build new domains */ |
7116 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7117 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7118 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7119 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7120 | goto match2; |
7121 | } | |
7122 | /* no match - add a new doms_new */ | |
dce840a0 | 7123 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7124 | match2: |
7125 | ; | |
7126 | } | |
7127 | ||
7128 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7129 | if (doms_cur != &fallback_doms) |
7130 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7131 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7132 | doms_cur = doms_new; |
1d3504fc | 7133 | dattr_cur = dattr_new; |
029190c5 | 7134 | ndoms_cur = ndoms_new; |
7378547f MM |
7135 | |
7136 | register_sched_domain_sysctl(); | |
a1835615 | 7137 | |
712555ee | 7138 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7139 | } |
7140 | ||
d35be8ba SB |
7141 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
7142 | ||
1da177e4 | 7143 | /* |
3a101d05 TH |
7144 | * Update cpusets according to cpu_active mask. If cpusets are |
7145 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7146 | * around partition_sched_domains(). | |
d35be8ba SB |
7147 | * |
7148 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7149 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7150 | */ |
0b2e918a TH |
7151 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7152 | void *hcpu) | |
e761b772 | 7153 | { |
d35be8ba SB |
7154 | switch (action) { |
7155 | case CPU_ONLINE_FROZEN: | |
7156 | case CPU_DOWN_FAILED_FROZEN: | |
7157 | ||
7158 | /* | |
7159 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7160 | * resume sequence. As long as this is not the last online | |
7161 | * operation in the resume sequence, just build a single sched | |
7162 | * domain, ignoring cpusets. | |
7163 | */ | |
7164 | num_cpus_frozen--; | |
7165 | if (likely(num_cpus_frozen)) { | |
7166 | partition_sched_domains(1, NULL, NULL); | |
7167 | break; | |
7168 | } | |
7169 | ||
7170 | /* | |
7171 | * This is the last CPU online operation. So fall through and | |
7172 | * restore the original sched domains by considering the | |
7173 | * cpuset configurations. | |
7174 | */ | |
7175 | ||
e761b772 | 7176 | case CPU_ONLINE: |
6ad4c188 | 7177 | case CPU_DOWN_FAILED: |
7ddf96b0 | 7178 | cpuset_update_active_cpus(true); |
d35be8ba | 7179 | break; |
3a101d05 TH |
7180 | default: |
7181 | return NOTIFY_DONE; | |
7182 | } | |
d35be8ba | 7183 | return NOTIFY_OK; |
3a101d05 | 7184 | } |
e761b772 | 7185 | |
0b2e918a TH |
7186 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7187 | void *hcpu) | |
3a101d05 | 7188 | { |
d35be8ba | 7189 | switch (action) { |
3a101d05 | 7190 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 7191 | cpuset_update_active_cpus(false); |
d35be8ba SB |
7192 | break; |
7193 | case CPU_DOWN_PREPARE_FROZEN: | |
7194 | num_cpus_frozen++; | |
7195 | partition_sched_domains(1, NULL, NULL); | |
7196 | break; | |
e761b772 MK |
7197 | default: |
7198 | return NOTIFY_DONE; | |
7199 | } | |
d35be8ba | 7200 | return NOTIFY_OK; |
e761b772 | 7201 | } |
e761b772 | 7202 | |
1da177e4 LT |
7203 | void __init sched_init_smp(void) |
7204 | { | |
dcc30a35 RR |
7205 | cpumask_var_t non_isolated_cpus; |
7206 | ||
7207 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7208 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7209 | |
cb83b629 PZ |
7210 | sched_init_numa(); |
7211 | ||
95402b38 | 7212 | get_online_cpus(); |
712555ee | 7213 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7214 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7215 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7216 | if (cpumask_empty(non_isolated_cpus)) | |
7217 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7218 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7219 | put_online_cpus(); |
e761b772 | 7220 | |
3a101d05 TH |
7221 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7222 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7223 | |
7224 | /* RT runtime code needs to handle some hotplug events */ | |
7225 | hotcpu_notifier(update_runtime, 0); | |
7226 | ||
b328ca18 | 7227 | init_hrtick(); |
5c1e1767 NP |
7228 | |
7229 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7230 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7231 | BUG(); |
19978ca6 | 7232 | sched_init_granularity(); |
dcc30a35 | 7233 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7234 | |
0e3900e6 | 7235 | init_sched_rt_class(); |
1da177e4 LT |
7236 | } |
7237 | #else | |
7238 | void __init sched_init_smp(void) | |
7239 | { | |
19978ca6 | 7240 | sched_init_granularity(); |
1da177e4 LT |
7241 | } |
7242 | #endif /* CONFIG_SMP */ | |
7243 | ||
cd1bb94b AB |
7244 | const_debug unsigned int sysctl_timer_migration = 1; |
7245 | ||
1da177e4 LT |
7246 | int in_sched_functions(unsigned long addr) |
7247 | { | |
1da177e4 LT |
7248 | return in_lock_functions(addr) || |
7249 | (addr >= (unsigned long)__sched_text_start | |
7250 | && addr < (unsigned long)__sched_text_end); | |
7251 | } | |
7252 | ||
029632fb PZ |
7253 | #ifdef CONFIG_CGROUP_SCHED |
7254 | struct task_group root_task_group; | |
35cf4e50 | 7255 | LIST_HEAD(task_groups); |
052f1dc7 | 7256 | #endif |
6f505b16 | 7257 | |
029632fb | 7258 | DECLARE_PER_CPU(cpumask_var_t, load_balance_tmpmask); |
6f505b16 | 7259 | |
1da177e4 LT |
7260 | void __init sched_init(void) |
7261 | { | |
dd41f596 | 7262 | int i, j; |
434d53b0 MT |
7263 | unsigned long alloc_size = 0, ptr; |
7264 | ||
7265 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7266 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7267 | #endif | |
7268 | #ifdef CONFIG_RT_GROUP_SCHED | |
7269 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7270 | #endif |
df7c8e84 | 7271 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7272 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7273 | #endif |
434d53b0 | 7274 | if (alloc_size) { |
36b7b6d4 | 7275 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7276 | |
7277 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7278 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7279 | ptr += nr_cpu_ids * sizeof(void **); |
7280 | ||
07e06b01 | 7281 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7282 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7283 | |
6d6bc0ad | 7284 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7285 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7286 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7287 | ptr += nr_cpu_ids * sizeof(void **); |
7288 | ||
07e06b01 | 7289 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7290 | ptr += nr_cpu_ids * sizeof(void **); |
7291 | ||
6d6bc0ad | 7292 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7293 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7294 | for_each_possible_cpu(i) { | |
7295 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7296 | ptr += cpumask_size(); | |
7297 | } | |
7298 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7299 | } |
dd41f596 | 7300 | |
57d885fe GH |
7301 | #ifdef CONFIG_SMP |
7302 | init_defrootdomain(); | |
7303 | #endif | |
7304 | ||
d0b27fa7 PZ |
7305 | init_rt_bandwidth(&def_rt_bandwidth, |
7306 | global_rt_period(), global_rt_runtime()); | |
7307 | ||
7308 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 7309 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7310 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7311 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7312 | |
7c941438 | 7313 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7314 | list_add(&root_task_group.list, &task_groups); |
7315 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7316 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7317 | autogroup_init(&init_task); |
54c707e9 | 7318 | |
7c941438 | 7319 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7320 | |
54c707e9 GC |
7321 | #ifdef CONFIG_CGROUP_CPUACCT |
7322 | root_cpuacct.cpustat = &kernel_cpustat; | |
7323 | root_cpuacct.cpuusage = alloc_percpu(u64); | |
7324 | /* Too early, not expected to fail */ | |
7325 | BUG_ON(!root_cpuacct.cpuusage); | |
7326 | #endif | |
0a945022 | 7327 | for_each_possible_cpu(i) { |
70b97a7f | 7328 | struct rq *rq; |
1da177e4 LT |
7329 | |
7330 | rq = cpu_rq(i); | |
05fa785c | 7331 | raw_spin_lock_init(&rq->lock); |
7897986b | 7332 | rq->nr_running = 0; |
dce48a84 TG |
7333 | rq->calc_load_active = 0; |
7334 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7335 | init_cfs_rq(&rq->cfs); |
6f505b16 | 7336 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7337 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 7338 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 7339 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7340 | /* |
07e06b01 | 7341 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7342 | * |
7343 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7344 | * gets 100% of the cpu resources in the system. This overall | |
7345 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7346 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7347 | * based on each entity's (task or task-group's) weight |
7348 | * (se->load.weight). | |
7349 | * | |
07e06b01 | 7350 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7351 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7352 | * then A0's share of the cpu resource is: | |
7353 | * | |
0d905bca | 7354 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7355 | * |
07e06b01 YZ |
7356 | * We achieve this by letting root_task_group's tasks sit |
7357 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7358 | */ |
ab84d31e | 7359 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 7360 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7361 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7362 | ||
7363 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7364 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7365 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7366 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7367 | #endif |
1da177e4 | 7368 | |
dd41f596 IM |
7369 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7370 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7371 | |
7372 | rq->last_load_update_tick = jiffies; | |
7373 | ||
1da177e4 | 7374 | #ifdef CONFIG_SMP |
41c7ce9a | 7375 | rq->sd = NULL; |
57d885fe | 7376 | rq->rd = NULL; |
1399fa78 | 7377 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 7378 | rq->post_schedule = 0; |
1da177e4 | 7379 | rq->active_balance = 0; |
dd41f596 | 7380 | rq->next_balance = jiffies; |
1da177e4 | 7381 | rq->push_cpu = 0; |
0a2966b4 | 7382 | rq->cpu = i; |
1f11eb6a | 7383 | rq->online = 0; |
eae0c9df MG |
7384 | rq->idle_stamp = 0; |
7385 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
367456c7 PZ |
7386 | |
7387 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7388 | ||
dc938520 | 7389 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 | 7390 | #ifdef CONFIG_NO_HZ |
1c792db7 | 7391 | rq->nohz_flags = 0; |
83cd4fe2 | 7392 | #endif |
1da177e4 | 7393 | #endif |
8f4d37ec | 7394 | init_rq_hrtick(rq); |
1da177e4 | 7395 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7396 | } |
7397 | ||
2dd73a4f | 7398 | set_load_weight(&init_task); |
b50f60ce | 7399 | |
e107be36 AK |
7400 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7401 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7402 | #endif | |
7403 | ||
b50f60ce | 7404 | #ifdef CONFIG_RT_MUTEXES |
732375c6 | 7405 | plist_head_init(&init_task.pi_waiters); |
b50f60ce HC |
7406 | #endif |
7407 | ||
1da177e4 LT |
7408 | /* |
7409 | * The boot idle thread does lazy MMU switching as well: | |
7410 | */ | |
7411 | atomic_inc(&init_mm.mm_count); | |
7412 | enter_lazy_tlb(&init_mm, current); | |
7413 | ||
7414 | /* | |
7415 | * Make us the idle thread. Technically, schedule() should not be | |
7416 | * called from this thread, however somewhere below it might be, | |
7417 | * but because we are the idle thread, we just pick up running again | |
7418 | * when this runqueue becomes "idle". | |
7419 | */ | |
7420 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7421 | |
7422 | calc_load_update = jiffies + LOAD_FREQ; | |
7423 | ||
dd41f596 IM |
7424 | /* |
7425 | * During early bootup we pretend to be a normal task: | |
7426 | */ | |
7427 | current->sched_class = &fair_sched_class; | |
6892b75e | 7428 | |
bf4d83f6 | 7429 | #ifdef CONFIG_SMP |
4cb98839 | 7430 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
7431 | /* May be allocated at isolcpus cmdline parse time */ |
7432 | if (cpu_isolated_map == NULL) | |
7433 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 7434 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7435 | #endif |
7436 | init_sched_fair_class(); | |
6a7b3dc3 | 7437 | |
6892b75e | 7438 | scheduler_running = 1; |
1da177e4 LT |
7439 | } |
7440 | ||
d902db1e | 7441 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7442 | static inline int preempt_count_equals(int preempt_offset) |
7443 | { | |
234da7bc | 7444 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 7445 | |
4ba8216c | 7446 | return (nested == preempt_offset); |
e4aafea2 FW |
7447 | } |
7448 | ||
d894837f | 7449 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7450 | { |
1da177e4 LT |
7451 | static unsigned long prev_jiffy; /* ratelimiting */ |
7452 | ||
b3fbab05 | 7453 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
e4aafea2 FW |
7454 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7455 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7456 | return; |
7457 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7458 | return; | |
7459 | prev_jiffy = jiffies; | |
7460 | ||
3df0fc5b PZ |
7461 | printk(KERN_ERR |
7462 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7463 | file, line); | |
7464 | printk(KERN_ERR | |
7465 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7466 | in_atomic(), irqs_disabled(), | |
7467 | current->pid, current->comm); | |
aef745fc IM |
7468 | |
7469 | debug_show_held_locks(current); | |
7470 | if (irqs_disabled()) | |
7471 | print_irqtrace_events(current); | |
7472 | dump_stack(); | |
1da177e4 LT |
7473 | } |
7474 | EXPORT_SYMBOL(__might_sleep); | |
7475 | #endif | |
7476 | ||
7477 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7478 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7479 | { | |
da7a735e PZ |
7480 | const struct sched_class *prev_class = p->sched_class; |
7481 | int old_prio = p->prio; | |
3a5e4dc1 | 7482 | int on_rq; |
3e51f33f | 7483 | |
fd2f4419 | 7484 | on_rq = p->on_rq; |
3a5e4dc1 | 7485 | if (on_rq) |
4ca9b72b | 7486 | dequeue_task(rq, p, 0); |
3a5e4dc1 AK |
7487 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
7488 | if (on_rq) { | |
4ca9b72b | 7489 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7490 | resched_task(rq->curr); |
7491 | } | |
da7a735e PZ |
7492 | |
7493 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7494 | } |
7495 | ||
1da177e4 LT |
7496 | void normalize_rt_tasks(void) |
7497 | { | |
a0f98a1c | 7498 | struct task_struct *g, *p; |
1da177e4 | 7499 | unsigned long flags; |
70b97a7f | 7500 | struct rq *rq; |
1da177e4 | 7501 | |
4cf5d77a | 7502 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7503 | do_each_thread(g, p) { |
178be793 IM |
7504 | /* |
7505 | * Only normalize user tasks: | |
7506 | */ | |
7507 | if (!p->mm) | |
7508 | continue; | |
7509 | ||
6cfb0d5d | 7510 | p->se.exec_start = 0; |
6cfb0d5d | 7511 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7512 | p->se.statistics.wait_start = 0; |
7513 | p->se.statistics.sleep_start = 0; | |
7514 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7515 | #endif |
dd41f596 IM |
7516 | |
7517 | if (!rt_task(p)) { | |
7518 | /* | |
7519 | * Renice negative nice level userspace | |
7520 | * tasks back to 0: | |
7521 | */ | |
7522 | if (TASK_NICE(p) < 0 && p->mm) | |
7523 | set_user_nice(p, 0); | |
1da177e4 | 7524 | continue; |
dd41f596 | 7525 | } |
1da177e4 | 7526 | |
1d615482 | 7527 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7528 | rq = __task_rq_lock(p); |
1da177e4 | 7529 | |
178be793 | 7530 | normalize_task(rq, p); |
3a5e4dc1 | 7531 | |
b29739f9 | 7532 | __task_rq_unlock(rq); |
1d615482 | 7533 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7534 | } while_each_thread(g, p); |
7535 | ||
4cf5d77a | 7536 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7537 | } |
7538 | ||
7539 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7540 | |
67fc4e0c | 7541 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7542 | /* |
67fc4e0c | 7543 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7544 | * |
7545 | * They can only be called when the whole system has been | |
7546 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7547 | * activity can take place. Using them for anything else would | |
7548 | * be a serious bug, and as a result, they aren't even visible | |
7549 | * under any other configuration. | |
7550 | */ | |
7551 | ||
7552 | /** | |
7553 | * curr_task - return the current task for a given cpu. | |
7554 | * @cpu: the processor in question. | |
7555 | * | |
7556 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7557 | */ | |
36c8b586 | 7558 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7559 | { |
7560 | return cpu_curr(cpu); | |
7561 | } | |
7562 | ||
67fc4e0c JW |
7563 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7564 | ||
7565 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7566 | /** |
7567 | * set_curr_task - set the current task for a given cpu. | |
7568 | * @cpu: the processor in question. | |
7569 | * @p: the task pointer to set. | |
7570 | * | |
7571 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7572 | * are serviced on a separate stack. It allows the architecture to switch the |
7573 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7574 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7575 | * and caller must save the original value of the current task (see | |
7576 | * curr_task() above) and restore that value before reenabling interrupts and | |
7577 | * re-starting the system. | |
7578 | * | |
7579 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7580 | */ | |
36c8b586 | 7581 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7582 | { |
7583 | cpu_curr(cpu) = p; | |
7584 | } | |
7585 | ||
7586 | #endif | |
29f59db3 | 7587 | |
7c941438 | 7588 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7589 | /* task_group_lock serializes the addition/removal of task groups */ |
7590 | static DEFINE_SPINLOCK(task_group_lock); | |
7591 | ||
bccbe08a PZ |
7592 | static void free_sched_group(struct task_group *tg) |
7593 | { | |
7594 | free_fair_sched_group(tg); | |
7595 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7596 | autogroup_free(tg); |
bccbe08a PZ |
7597 | kfree(tg); |
7598 | } | |
7599 | ||
7600 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7601 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7602 | { |
7603 | struct task_group *tg; | |
7604 | unsigned long flags; | |
bccbe08a PZ |
7605 | |
7606 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7607 | if (!tg) | |
7608 | return ERR_PTR(-ENOMEM); | |
7609 | ||
ec7dc8ac | 7610 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7611 | goto err; |
7612 | ||
ec7dc8ac | 7613 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7614 | goto err; |
7615 | ||
8ed36996 | 7616 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7617 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7618 | |
7619 | WARN_ON(!parent); /* root should already exist */ | |
7620 | ||
7621 | tg->parent = parent; | |
f473aa5e | 7622 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7623 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7624 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 7625 | |
9b5b7751 | 7626 | return tg; |
29f59db3 SV |
7627 | |
7628 | err: | |
6f505b16 | 7629 | free_sched_group(tg); |
29f59db3 SV |
7630 | return ERR_PTR(-ENOMEM); |
7631 | } | |
7632 | ||
9b5b7751 | 7633 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7634 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7635 | { |
29f59db3 | 7636 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7637 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7638 | } |
7639 | ||
9b5b7751 | 7640 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7641 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7642 | { |
8ed36996 | 7643 | unsigned long flags; |
9b5b7751 | 7644 | int i; |
29f59db3 | 7645 | |
3d4b47b4 PZ |
7646 | /* end participation in shares distribution */ |
7647 | for_each_possible_cpu(i) | |
bccbe08a | 7648 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7649 | |
7650 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7651 | list_del_rcu(&tg->list); |
f473aa5e | 7652 | list_del_rcu(&tg->siblings); |
8ed36996 | 7653 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 7654 | |
9b5b7751 | 7655 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 7656 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
7657 | } |
7658 | ||
9b5b7751 | 7659 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7660 | * The caller of this function should have put the task in its new group |
7661 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7662 | * reflect its new group. | |
9b5b7751 SV |
7663 | */ |
7664 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7665 | { |
8323f26c | 7666 | struct task_group *tg; |
29f59db3 SV |
7667 | int on_rq, running; |
7668 | unsigned long flags; | |
7669 | struct rq *rq; | |
7670 | ||
7671 | rq = task_rq_lock(tsk, &flags); | |
7672 | ||
051a1d1a | 7673 | running = task_current(rq, tsk); |
fd2f4419 | 7674 | on_rq = tsk->on_rq; |
29f59db3 | 7675 | |
0e1f3483 | 7676 | if (on_rq) |
29f59db3 | 7677 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7678 | if (unlikely(running)) |
7679 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7680 | |
8323f26c PZ |
7681 | tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id, |
7682 | lockdep_is_held(&tsk->sighand->siglock)), | |
7683 | struct task_group, css); | |
7684 | tg = autogroup_task_group(tsk, tg); | |
7685 | tsk->sched_task_group = tg; | |
7686 | ||
810b3817 | 7687 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7688 | if (tsk->sched_class->task_move_group) |
7689 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7690 | else | |
810b3817 | 7691 | #endif |
b2b5ce02 | 7692 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7693 | |
0e1f3483 HS |
7694 | if (unlikely(running)) |
7695 | tsk->sched_class->set_curr_task(rq); | |
7696 | if (on_rq) | |
371fd7e7 | 7697 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7698 | |
0122ec5b | 7699 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7700 | } |
7c941438 | 7701 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7702 | |
a790de99 | 7703 | #if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH) |
9f0c1e56 PZ |
7704 | static unsigned long to_ratio(u64 period, u64 runtime) |
7705 | { | |
7706 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 7707 | return 1ULL << 20; |
9f0c1e56 | 7708 | |
9a7e0b18 | 7709 | return div64_u64(runtime << 20, period); |
9f0c1e56 | 7710 | } |
a790de99 PT |
7711 | #endif |
7712 | ||
7713 | #ifdef CONFIG_RT_GROUP_SCHED | |
7714 | /* | |
7715 | * Ensure that the real time constraints are schedulable. | |
7716 | */ | |
7717 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7718 | |
9a7e0b18 PZ |
7719 | /* Must be called with tasklist_lock held */ |
7720 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7721 | { |
9a7e0b18 | 7722 | struct task_struct *g, *p; |
b40b2e8e | 7723 | |
9a7e0b18 | 7724 | do_each_thread(g, p) { |
029632fb | 7725 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7726 | return 1; |
7727 | } while_each_thread(g, p); | |
b40b2e8e | 7728 | |
9a7e0b18 PZ |
7729 | return 0; |
7730 | } | |
b40b2e8e | 7731 | |
9a7e0b18 PZ |
7732 | struct rt_schedulable_data { |
7733 | struct task_group *tg; | |
7734 | u64 rt_period; | |
7735 | u64 rt_runtime; | |
7736 | }; | |
b40b2e8e | 7737 | |
a790de99 | 7738 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7739 | { |
7740 | struct rt_schedulable_data *d = data; | |
7741 | struct task_group *child; | |
7742 | unsigned long total, sum = 0; | |
7743 | u64 period, runtime; | |
b40b2e8e | 7744 | |
9a7e0b18 PZ |
7745 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7746 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7747 | |
9a7e0b18 PZ |
7748 | if (tg == d->tg) { |
7749 | period = d->rt_period; | |
7750 | runtime = d->rt_runtime; | |
b40b2e8e | 7751 | } |
b40b2e8e | 7752 | |
4653f803 PZ |
7753 | /* |
7754 | * Cannot have more runtime than the period. | |
7755 | */ | |
7756 | if (runtime > period && runtime != RUNTIME_INF) | |
7757 | return -EINVAL; | |
6f505b16 | 7758 | |
4653f803 PZ |
7759 | /* |
7760 | * Ensure we don't starve existing RT tasks. | |
7761 | */ | |
9a7e0b18 PZ |
7762 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7763 | return -EBUSY; | |
6f505b16 | 7764 | |
9a7e0b18 | 7765 | total = to_ratio(period, runtime); |
6f505b16 | 7766 | |
4653f803 PZ |
7767 | /* |
7768 | * Nobody can have more than the global setting allows. | |
7769 | */ | |
7770 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7771 | return -EINVAL; | |
6f505b16 | 7772 | |
4653f803 PZ |
7773 | /* |
7774 | * The sum of our children's runtime should not exceed our own. | |
7775 | */ | |
9a7e0b18 PZ |
7776 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7777 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7778 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7779 | |
9a7e0b18 PZ |
7780 | if (child == d->tg) { |
7781 | period = d->rt_period; | |
7782 | runtime = d->rt_runtime; | |
7783 | } | |
6f505b16 | 7784 | |
9a7e0b18 | 7785 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7786 | } |
6f505b16 | 7787 | |
9a7e0b18 PZ |
7788 | if (sum > total) |
7789 | return -EINVAL; | |
7790 | ||
7791 | return 0; | |
6f505b16 PZ |
7792 | } |
7793 | ||
9a7e0b18 | 7794 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7795 | { |
8277434e PT |
7796 | int ret; |
7797 | ||
9a7e0b18 PZ |
7798 | struct rt_schedulable_data data = { |
7799 | .tg = tg, | |
7800 | .rt_period = period, | |
7801 | .rt_runtime = runtime, | |
7802 | }; | |
7803 | ||
8277434e PT |
7804 | rcu_read_lock(); |
7805 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7806 | rcu_read_unlock(); | |
7807 | ||
7808 | return ret; | |
521f1a24 DG |
7809 | } |
7810 | ||
ab84d31e | 7811 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7812 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7813 | { |
ac086bc2 | 7814 | int i, err = 0; |
9f0c1e56 | 7815 | |
9f0c1e56 | 7816 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7817 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7818 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7819 | if (err) | |
9f0c1e56 | 7820 | goto unlock; |
ac086bc2 | 7821 | |
0986b11b | 7822 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7823 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7824 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7825 | |
7826 | for_each_possible_cpu(i) { | |
7827 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7828 | ||
0986b11b | 7829 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7830 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7831 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7832 | } |
0986b11b | 7833 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7834 | unlock: |
521f1a24 | 7835 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7836 | mutex_unlock(&rt_constraints_mutex); |
7837 | ||
7838 | return err; | |
6f505b16 PZ |
7839 | } |
7840 | ||
d0b27fa7 PZ |
7841 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
7842 | { | |
7843 | u64 rt_runtime, rt_period; | |
7844 | ||
7845 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7846 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7847 | if (rt_runtime_us < 0) | |
7848 | rt_runtime = RUNTIME_INF; | |
7849 | ||
ab84d31e | 7850 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7851 | } |
7852 | ||
9f0c1e56 PZ |
7853 | long sched_group_rt_runtime(struct task_group *tg) |
7854 | { | |
7855 | u64 rt_runtime_us; | |
7856 | ||
d0b27fa7 | 7857 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7858 | return -1; |
7859 | ||
d0b27fa7 | 7860 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7861 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7862 | return rt_runtime_us; | |
7863 | } | |
d0b27fa7 PZ |
7864 | |
7865 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
7866 | { | |
7867 | u64 rt_runtime, rt_period; | |
7868 | ||
7869 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7870 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7871 | ||
619b0488 R |
7872 | if (rt_period == 0) |
7873 | return -EINVAL; | |
7874 | ||
ab84d31e | 7875 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7876 | } |
7877 | ||
7878 | long sched_group_rt_period(struct task_group *tg) | |
7879 | { | |
7880 | u64 rt_period_us; | |
7881 | ||
7882 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7883 | do_div(rt_period_us, NSEC_PER_USEC); | |
7884 | return rt_period_us; | |
7885 | } | |
7886 | ||
7887 | static int sched_rt_global_constraints(void) | |
7888 | { | |
4653f803 | 7889 | u64 runtime, period; |
d0b27fa7 PZ |
7890 | int ret = 0; |
7891 | ||
ec5d4989 HS |
7892 | if (sysctl_sched_rt_period <= 0) |
7893 | return -EINVAL; | |
7894 | ||
4653f803 PZ |
7895 | runtime = global_rt_runtime(); |
7896 | period = global_rt_period(); | |
7897 | ||
7898 | /* | |
7899 | * Sanity check on the sysctl variables. | |
7900 | */ | |
7901 | if (runtime > period && runtime != RUNTIME_INF) | |
7902 | return -EINVAL; | |
10b612f4 | 7903 | |
d0b27fa7 | 7904 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 7905 | read_lock(&tasklist_lock); |
4653f803 | 7906 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7907 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7908 | mutex_unlock(&rt_constraints_mutex); |
7909 | ||
7910 | return ret; | |
7911 | } | |
54e99124 DG |
7912 | |
7913 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
7914 | { | |
7915 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7916 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7917 | return 0; | |
7918 | ||
7919 | return 1; | |
7920 | } | |
7921 | ||
6d6bc0ad | 7922 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7923 | static int sched_rt_global_constraints(void) |
7924 | { | |
ac086bc2 PZ |
7925 | unsigned long flags; |
7926 | int i; | |
7927 | ||
ec5d4989 HS |
7928 | if (sysctl_sched_rt_period <= 0) |
7929 | return -EINVAL; | |
7930 | ||
60aa605d PZ |
7931 | /* |
7932 | * There's always some RT tasks in the root group | |
7933 | * -- migration, kstopmachine etc.. | |
7934 | */ | |
7935 | if (sysctl_sched_rt_runtime == 0) | |
7936 | return -EBUSY; | |
7937 | ||
0986b11b | 7938 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7939 | for_each_possible_cpu(i) { |
7940 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7941 | ||
0986b11b | 7942 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7943 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7944 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7945 | } |
0986b11b | 7946 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7947 | |
d0b27fa7 PZ |
7948 | return 0; |
7949 | } | |
6d6bc0ad | 7950 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7951 | |
7952 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 7953 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7954 | loff_t *ppos) |
7955 | { | |
7956 | int ret; | |
7957 | int old_period, old_runtime; | |
7958 | static DEFINE_MUTEX(mutex); | |
7959 | ||
7960 | mutex_lock(&mutex); | |
7961 | old_period = sysctl_sched_rt_period; | |
7962 | old_runtime = sysctl_sched_rt_runtime; | |
7963 | ||
8d65af78 | 7964 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7965 | |
7966 | if (!ret && write) { | |
7967 | ret = sched_rt_global_constraints(); | |
7968 | if (ret) { | |
7969 | sysctl_sched_rt_period = old_period; | |
7970 | sysctl_sched_rt_runtime = old_runtime; | |
7971 | } else { | |
7972 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7973 | def_rt_bandwidth.rt_period = | |
7974 | ns_to_ktime(global_rt_period()); | |
7975 | } | |
7976 | } | |
7977 | mutex_unlock(&mutex); | |
7978 | ||
7979 | return ret; | |
7980 | } | |
68318b8e | 7981 | |
052f1dc7 | 7982 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
7983 | |
7984 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7985 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7986 | { |
2b01dfe3 PM |
7987 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7988 | struct task_group, css); | |
68318b8e SV |
7989 | } |
7990 | ||
761b3ef5 | 7991 | static struct cgroup_subsys_state *cpu_cgroup_create(struct cgroup *cgrp) |
68318b8e | 7992 | { |
ec7dc8ac | 7993 | struct task_group *tg, *parent; |
68318b8e | 7994 | |
2b01dfe3 | 7995 | if (!cgrp->parent) { |
68318b8e | 7996 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7997 | return &root_task_group.css; |
68318b8e SV |
7998 | } |
7999 | ||
ec7dc8ac DG |
8000 | parent = cgroup_tg(cgrp->parent); |
8001 | tg = sched_create_group(parent); | |
68318b8e SV |
8002 | if (IS_ERR(tg)) |
8003 | return ERR_PTR(-ENOMEM); | |
8004 | ||
68318b8e SV |
8005 | return &tg->css; |
8006 | } | |
8007 | ||
761b3ef5 | 8008 | static void cpu_cgroup_destroy(struct cgroup *cgrp) |
68318b8e | 8009 | { |
2b01dfe3 | 8010 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8011 | |
8012 | sched_destroy_group(tg); | |
8013 | } | |
8014 | ||
761b3ef5 | 8015 | static int cpu_cgroup_can_attach(struct cgroup *cgrp, |
bb9d97b6 | 8016 | struct cgroup_taskset *tset) |
68318b8e | 8017 | { |
bb9d97b6 TH |
8018 | struct task_struct *task; |
8019 | ||
8020 | cgroup_taskset_for_each(task, cgrp, tset) { | |
b68aa230 | 8021 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 TH |
8022 | if (!sched_rt_can_attach(cgroup_tg(cgrp), task)) |
8023 | return -EINVAL; | |
b68aa230 | 8024 | #else |
bb9d97b6 TH |
8025 | /* We don't support RT-tasks being in separate groups */ |
8026 | if (task->sched_class != &fair_sched_class) | |
8027 | return -EINVAL; | |
b68aa230 | 8028 | #endif |
bb9d97b6 | 8029 | } |
be367d09 BB |
8030 | return 0; |
8031 | } | |
68318b8e | 8032 | |
761b3ef5 | 8033 | static void cpu_cgroup_attach(struct cgroup *cgrp, |
bb9d97b6 | 8034 | struct cgroup_taskset *tset) |
68318b8e | 8035 | { |
bb9d97b6 TH |
8036 | struct task_struct *task; |
8037 | ||
8038 | cgroup_taskset_for_each(task, cgrp, tset) | |
8039 | sched_move_task(task); | |
68318b8e SV |
8040 | } |
8041 | ||
068c5cc5 | 8042 | static void |
761b3ef5 LZ |
8043 | cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp, |
8044 | struct task_struct *task) | |
068c5cc5 PZ |
8045 | { |
8046 | /* | |
8047 | * cgroup_exit() is called in the copy_process() failure path. | |
8048 | * Ignore this case since the task hasn't ran yet, this avoids | |
8049 | * trying to poke a half freed task state from generic code. | |
8050 | */ | |
8051 | if (!(task->flags & PF_EXITING)) | |
8052 | return; | |
8053 | ||
8054 | sched_move_task(task); | |
8055 | } | |
8056 | ||
052f1dc7 | 8057 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8058 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8059 | u64 shareval) |
68318b8e | 8060 | { |
c8b28116 | 8061 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
8062 | } |
8063 | ||
f4c753b7 | 8064 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8065 | { |
2b01dfe3 | 8066 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 8067 | |
c8b28116 | 8068 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8069 | } |
ab84d31e PT |
8070 | |
8071 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8072 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8073 | ||
ab84d31e PT |
8074 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
8075 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
8076 | ||
a790de99 PT |
8077 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8078 | ||
ab84d31e PT |
8079 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8080 | { | |
56f570e5 | 8081 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8082 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8083 | |
8084 | if (tg == &root_task_group) | |
8085 | return -EINVAL; | |
8086 | ||
8087 | /* | |
8088 | * Ensure we have at some amount of bandwidth every period. This is | |
8089 | * to prevent reaching a state of large arrears when throttled via | |
8090 | * entity_tick() resulting in prolonged exit starvation. | |
8091 | */ | |
8092 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8093 | return -EINVAL; | |
8094 | ||
8095 | /* | |
8096 | * Likewise, bound things on the otherside by preventing insane quota | |
8097 | * periods. This also allows us to normalize in computing quota | |
8098 | * feasibility. | |
8099 | */ | |
8100 | if (period > max_cfs_quota_period) | |
8101 | return -EINVAL; | |
8102 | ||
a790de99 PT |
8103 | mutex_lock(&cfs_constraints_mutex); |
8104 | ret = __cfs_schedulable(tg, period, quota); | |
8105 | if (ret) | |
8106 | goto out_unlock; | |
8107 | ||
58088ad0 | 8108 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 PT |
8109 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
8110 | account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled); | |
ab84d31e PT |
8111 | raw_spin_lock_irq(&cfs_b->lock); |
8112 | cfs_b->period = ns_to_ktime(period); | |
8113 | cfs_b->quota = quota; | |
58088ad0 | 8114 | |
a9cf55b2 | 8115 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
8116 | /* restart the period timer (if active) to handle new period expiry */ |
8117 | if (runtime_enabled && cfs_b->timer_active) { | |
8118 | /* force a reprogram */ | |
8119 | cfs_b->timer_active = 0; | |
8120 | __start_cfs_bandwidth(cfs_b); | |
8121 | } | |
ab84d31e PT |
8122 | raw_spin_unlock_irq(&cfs_b->lock); |
8123 | ||
8124 | for_each_possible_cpu(i) { | |
8125 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 8126 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
8127 | |
8128 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 8129 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8130 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8131 | |
029632fb | 8132 | if (cfs_rq->throttled) |
671fd9da | 8133 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
8134 | raw_spin_unlock_irq(&rq->lock); |
8135 | } | |
a790de99 PT |
8136 | out_unlock: |
8137 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 8138 | |
a790de99 | 8139 | return ret; |
ab84d31e PT |
8140 | } |
8141 | ||
8142 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
8143 | { | |
8144 | u64 quota, period; | |
8145 | ||
029632fb | 8146 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8147 | if (cfs_quota_us < 0) |
8148 | quota = RUNTIME_INF; | |
8149 | else | |
8150 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
8151 | ||
8152 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8153 | } | |
8154 | ||
8155 | long tg_get_cfs_quota(struct task_group *tg) | |
8156 | { | |
8157 | u64 quota_us; | |
8158 | ||
029632fb | 8159 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8160 | return -1; |
8161 | ||
029632fb | 8162 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8163 | do_div(quota_us, NSEC_PER_USEC); |
8164 | ||
8165 | return quota_us; | |
8166 | } | |
8167 | ||
8168 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
8169 | { | |
8170 | u64 quota, period; | |
8171 | ||
8172 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 8173 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8174 | |
ab84d31e PT |
8175 | return tg_set_cfs_bandwidth(tg, period, quota); |
8176 | } | |
8177 | ||
8178 | long tg_get_cfs_period(struct task_group *tg) | |
8179 | { | |
8180 | u64 cfs_period_us; | |
8181 | ||
029632fb | 8182 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8183 | do_div(cfs_period_us, NSEC_PER_USEC); |
8184 | ||
8185 | return cfs_period_us; | |
8186 | } | |
8187 | ||
8188 | static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft) | |
8189 | { | |
8190 | return tg_get_cfs_quota(cgroup_tg(cgrp)); | |
8191 | } | |
8192 | ||
8193 | static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype, | |
8194 | s64 cfs_quota_us) | |
8195 | { | |
8196 | return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us); | |
8197 | } | |
8198 | ||
8199 | static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft) | |
8200 | { | |
8201 | return tg_get_cfs_period(cgroup_tg(cgrp)); | |
8202 | } | |
8203 | ||
8204 | static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype, | |
8205 | u64 cfs_period_us) | |
8206 | { | |
8207 | return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us); | |
8208 | } | |
8209 | ||
a790de99 PT |
8210 | struct cfs_schedulable_data { |
8211 | struct task_group *tg; | |
8212 | u64 period, quota; | |
8213 | }; | |
8214 | ||
8215 | /* | |
8216 | * normalize group quota/period to be quota/max_period | |
8217 | * note: units are usecs | |
8218 | */ | |
8219 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8220 | struct cfs_schedulable_data *d) | |
8221 | { | |
8222 | u64 quota, period; | |
8223 | ||
8224 | if (tg == d->tg) { | |
8225 | period = d->period; | |
8226 | quota = d->quota; | |
8227 | } else { | |
8228 | period = tg_get_cfs_period(tg); | |
8229 | quota = tg_get_cfs_quota(tg); | |
8230 | } | |
8231 | ||
8232 | /* note: these should typically be equivalent */ | |
8233 | if (quota == RUNTIME_INF || quota == -1) | |
8234 | return RUNTIME_INF; | |
8235 | ||
8236 | return to_ratio(period, quota); | |
8237 | } | |
8238 | ||
8239 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8240 | { | |
8241 | struct cfs_schedulable_data *d = data; | |
029632fb | 8242 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8243 | s64 quota = 0, parent_quota = -1; |
8244 | ||
8245 | if (!tg->parent) { | |
8246 | quota = RUNTIME_INF; | |
8247 | } else { | |
029632fb | 8248 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8249 | |
8250 | quota = normalize_cfs_quota(tg, d); | |
8251 | parent_quota = parent_b->hierarchal_quota; | |
8252 | ||
8253 | /* | |
8254 | * ensure max(child_quota) <= parent_quota, inherit when no | |
8255 | * limit is set | |
8256 | */ | |
8257 | if (quota == RUNTIME_INF) | |
8258 | quota = parent_quota; | |
8259 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8260 | return -EINVAL; | |
8261 | } | |
8262 | cfs_b->hierarchal_quota = quota; | |
8263 | ||
8264 | return 0; | |
8265 | } | |
8266 | ||
8267 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8268 | { | |
8277434e | 8269 | int ret; |
a790de99 PT |
8270 | struct cfs_schedulable_data data = { |
8271 | .tg = tg, | |
8272 | .period = period, | |
8273 | .quota = quota, | |
8274 | }; | |
8275 | ||
8276 | if (quota != RUNTIME_INF) { | |
8277 | do_div(data.period, NSEC_PER_USEC); | |
8278 | do_div(data.quota, NSEC_PER_USEC); | |
8279 | } | |
8280 | ||
8277434e PT |
8281 | rcu_read_lock(); |
8282 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8283 | rcu_read_unlock(); | |
8284 | ||
8285 | return ret; | |
a790de99 | 8286 | } |
e8da1b18 NR |
8287 | |
8288 | static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8289 | struct cgroup_map_cb *cb) | |
8290 | { | |
8291 | struct task_group *tg = cgroup_tg(cgrp); | |
029632fb | 8292 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 NR |
8293 | |
8294 | cb->fill(cb, "nr_periods", cfs_b->nr_periods); | |
8295 | cb->fill(cb, "nr_throttled", cfs_b->nr_throttled); | |
8296 | cb->fill(cb, "throttled_time", cfs_b->throttled_time); | |
8297 | ||
8298 | return 0; | |
8299 | } | |
ab84d31e | 8300 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8301 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8302 | |
052f1dc7 | 8303 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8304 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8305 | s64 val) |
6f505b16 | 8306 | { |
06ecb27c | 8307 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8308 | } |
8309 | ||
06ecb27c | 8310 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8311 | { |
06ecb27c | 8312 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8313 | } |
d0b27fa7 PZ |
8314 | |
8315 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8316 | u64 rt_period_us) | |
8317 | { | |
8318 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8319 | } | |
8320 | ||
8321 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8322 | { | |
8323 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8324 | } | |
6d6bc0ad | 8325 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8326 | |
fe5c7cc2 | 8327 | static struct cftype cpu_files[] = { |
052f1dc7 | 8328 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8329 | { |
8330 | .name = "shares", | |
f4c753b7 PM |
8331 | .read_u64 = cpu_shares_read_u64, |
8332 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8333 | }, |
052f1dc7 | 8334 | #endif |
ab84d31e PT |
8335 | #ifdef CONFIG_CFS_BANDWIDTH |
8336 | { | |
8337 | .name = "cfs_quota_us", | |
8338 | .read_s64 = cpu_cfs_quota_read_s64, | |
8339 | .write_s64 = cpu_cfs_quota_write_s64, | |
8340 | }, | |
8341 | { | |
8342 | .name = "cfs_period_us", | |
8343 | .read_u64 = cpu_cfs_period_read_u64, | |
8344 | .write_u64 = cpu_cfs_period_write_u64, | |
8345 | }, | |
e8da1b18 NR |
8346 | { |
8347 | .name = "stat", | |
8348 | .read_map = cpu_stats_show, | |
8349 | }, | |
ab84d31e | 8350 | #endif |
052f1dc7 | 8351 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8352 | { |
9f0c1e56 | 8353 | .name = "rt_runtime_us", |
06ecb27c PM |
8354 | .read_s64 = cpu_rt_runtime_read, |
8355 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8356 | }, |
d0b27fa7 PZ |
8357 | { |
8358 | .name = "rt_period_us", | |
f4c753b7 PM |
8359 | .read_u64 = cpu_rt_period_read_uint, |
8360 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8361 | }, |
052f1dc7 | 8362 | #endif |
4baf6e33 | 8363 | { } /* terminate */ |
68318b8e SV |
8364 | }; |
8365 | ||
68318b8e | 8366 | struct cgroup_subsys cpu_cgroup_subsys = { |
38605cae IM |
8367 | .name = "cpu", |
8368 | .create = cpu_cgroup_create, | |
8369 | .destroy = cpu_cgroup_destroy, | |
bb9d97b6 TH |
8370 | .can_attach = cpu_cgroup_can_attach, |
8371 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8372 | .exit = cpu_cgroup_exit, |
38605cae | 8373 | .subsys_id = cpu_cgroup_subsys_id, |
4baf6e33 | 8374 | .base_cftypes = cpu_files, |
68318b8e SV |
8375 | .early_init = 1, |
8376 | }; | |
8377 | ||
052f1dc7 | 8378 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8379 | |
8380 | #ifdef CONFIG_CGROUP_CPUACCT | |
8381 | ||
8382 | /* | |
8383 | * CPU accounting code for task groups. | |
8384 | * | |
8385 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8386 | * (balbir@in.ibm.com). | |
8387 | */ | |
8388 | ||
d842de87 | 8389 | /* create a new cpu accounting group */ |
761b3ef5 | 8390 | static struct cgroup_subsys_state *cpuacct_create(struct cgroup *cgrp) |
d842de87 | 8391 | { |
54c707e9 | 8392 | struct cpuacct *ca; |
d842de87 | 8393 | |
54c707e9 GC |
8394 | if (!cgrp->parent) |
8395 | return &root_cpuacct.css; | |
8396 | ||
8397 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
d842de87 | 8398 | if (!ca) |
ef12fefa | 8399 | goto out; |
d842de87 SV |
8400 | |
8401 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8402 | if (!ca->cpuusage) |
8403 | goto out_free_ca; | |
8404 | ||
54c707e9 GC |
8405 | ca->cpustat = alloc_percpu(struct kernel_cpustat); |
8406 | if (!ca->cpustat) | |
8407 | goto out_free_cpuusage; | |
934352f2 | 8408 | |
d842de87 | 8409 | return &ca->css; |
ef12fefa | 8410 | |
54c707e9 | 8411 | out_free_cpuusage: |
ef12fefa BR |
8412 | free_percpu(ca->cpuusage); |
8413 | out_free_ca: | |
8414 | kfree(ca); | |
8415 | out: | |
8416 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8417 | } |
8418 | ||
8419 | /* destroy an existing cpu accounting group */ | |
761b3ef5 | 8420 | static void cpuacct_destroy(struct cgroup *cgrp) |
d842de87 | 8421 | { |
32cd756a | 8422 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 | 8423 | |
54c707e9 | 8424 | free_percpu(ca->cpustat); |
d842de87 SV |
8425 | free_percpu(ca->cpuusage); |
8426 | kfree(ca); | |
8427 | } | |
8428 | ||
720f5498 KC |
8429 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8430 | { | |
b36128c8 | 8431 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8432 | u64 data; |
8433 | ||
8434 | #ifndef CONFIG_64BIT | |
8435 | /* | |
8436 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8437 | */ | |
05fa785c | 8438 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8439 | data = *cpuusage; |
05fa785c | 8440 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8441 | #else |
8442 | data = *cpuusage; | |
8443 | #endif | |
8444 | ||
8445 | return data; | |
8446 | } | |
8447 | ||
8448 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8449 | { | |
b36128c8 | 8450 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8451 | |
8452 | #ifndef CONFIG_64BIT | |
8453 | /* | |
8454 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8455 | */ | |
05fa785c | 8456 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8457 | *cpuusage = val; |
05fa785c | 8458 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8459 | #else |
8460 | *cpuusage = val; | |
8461 | #endif | |
8462 | } | |
8463 | ||
d842de87 | 8464 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8465 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8466 | { |
32cd756a | 8467 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8468 | u64 totalcpuusage = 0; |
8469 | int i; | |
8470 | ||
720f5498 KC |
8471 | for_each_present_cpu(i) |
8472 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8473 | |
8474 | return totalcpuusage; | |
8475 | } | |
8476 | ||
0297b803 DG |
8477 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8478 | u64 reset) | |
8479 | { | |
8480 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8481 | int err = 0; | |
8482 | int i; | |
8483 | ||
8484 | if (reset) { | |
8485 | err = -EINVAL; | |
8486 | goto out; | |
8487 | } | |
8488 | ||
720f5498 KC |
8489 | for_each_present_cpu(i) |
8490 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8491 | |
0297b803 DG |
8492 | out: |
8493 | return err; | |
8494 | } | |
8495 | ||
e9515c3c KC |
8496 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8497 | struct seq_file *m) | |
8498 | { | |
8499 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8500 | u64 percpu; | |
8501 | int i; | |
8502 | ||
8503 | for_each_present_cpu(i) { | |
8504 | percpu = cpuacct_cpuusage_read(ca, i); | |
8505 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8506 | } | |
8507 | seq_printf(m, "\n"); | |
8508 | return 0; | |
8509 | } | |
8510 | ||
ef12fefa BR |
8511 | static const char *cpuacct_stat_desc[] = { |
8512 | [CPUACCT_STAT_USER] = "user", | |
8513 | [CPUACCT_STAT_SYSTEM] = "system", | |
8514 | }; | |
8515 | ||
8516 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
54c707e9 | 8517 | struct cgroup_map_cb *cb) |
ef12fefa BR |
8518 | { |
8519 | struct cpuacct *ca = cgroup_ca(cgrp); | |
54c707e9 GC |
8520 | int cpu; |
8521 | s64 val = 0; | |
ef12fefa | 8522 | |
54c707e9 GC |
8523 | for_each_online_cpu(cpu) { |
8524 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8525 | val += kcpustat->cpustat[CPUTIME_USER]; | |
8526 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
ef12fefa | 8527 | } |
54c707e9 GC |
8528 | val = cputime64_to_clock_t(val); |
8529 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
ef12fefa | 8530 | |
54c707e9 GC |
8531 | val = 0; |
8532 | for_each_online_cpu(cpu) { | |
8533 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
8534 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
8535 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
8536 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
ef12fefa | 8537 | } |
54c707e9 GC |
8538 | |
8539 | val = cputime64_to_clock_t(val); | |
8540 | cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
8541 | ||
ef12fefa BR |
8542 | return 0; |
8543 | } | |
8544 | ||
d842de87 SV |
8545 | static struct cftype files[] = { |
8546 | { | |
8547 | .name = "usage", | |
f4c753b7 PM |
8548 | .read_u64 = cpuusage_read, |
8549 | .write_u64 = cpuusage_write, | |
d842de87 | 8550 | }, |
e9515c3c KC |
8551 | { |
8552 | .name = "usage_percpu", | |
8553 | .read_seq_string = cpuacct_percpu_seq_read, | |
8554 | }, | |
ef12fefa BR |
8555 | { |
8556 | .name = "stat", | |
8557 | .read_map = cpuacct_stats_show, | |
8558 | }, | |
4baf6e33 | 8559 | { } /* terminate */ |
d842de87 SV |
8560 | }; |
8561 | ||
d842de87 SV |
8562 | /* |
8563 | * charge this task's execution time to its accounting group. | |
8564 | * | |
8565 | * called with rq->lock held. | |
8566 | */ | |
029632fb | 8567 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) |
d842de87 SV |
8568 | { |
8569 | struct cpuacct *ca; | |
934352f2 | 8570 | int cpu; |
d842de87 | 8571 | |
c40c6f85 | 8572 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8573 | return; |
8574 | ||
934352f2 | 8575 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8576 | |
8577 | rcu_read_lock(); | |
8578 | ||
d842de87 | 8579 | ca = task_ca(tsk); |
d842de87 | 8580 | |
44252e42 | 8581 | for (; ca; ca = parent_ca(ca)) { |
b36128c8 | 8582 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8583 | *cpuusage += cputime; |
8584 | } | |
a18b83b7 BR |
8585 | |
8586 | rcu_read_unlock(); | |
d842de87 SV |
8587 | } |
8588 | ||
8589 | struct cgroup_subsys cpuacct_subsys = { | |
8590 | .name = "cpuacct", | |
8591 | .create = cpuacct_create, | |
8592 | .destroy = cpuacct_destroy, | |
d842de87 | 8593 | .subsys_id = cpuacct_subsys_id, |
4baf6e33 | 8594 | .base_cftypes = files, |
d842de87 SV |
8595 | }; |
8596 | #endif /* CONFIG_CGROUP_CPUACCT */ |