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