]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 LT |
3 | * |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
8f4d37ec | 67 | #include <linux/hrtimer.h> |
30914a58 | 68 | #include <linux/tick.h> |
f00b45c1 PZ |
69 | #include <linux/debugfs.h> |
70 | #include <linux/ctype.h> | |
6cd8a4bb | 71 | #include <linux/ftrace.h> |
5a0e3ad6 | 72 | #include <linux/slab.h> |
f1c6f1a7 | 73 | #include <linux/init_task.h> |
40401530 | 74 | #include <linux/binfmts.h> |
91d1aa43 | 75 | #include <linux/context_tracking.h> |
52f5684c | 76 | #include <linux/compiler.h> |
1da177e4 | 77 | |
96f951ed | 78 | #include <asm/switch_to.h> |
5517d86b | 79 | #include <asm/tlb.h> |
838225b4 | 80 | #include <asm/irq_regs.h> |
db7e527d | 81 | #include <asm/mutex.h> |
e6e6685a GC |
82 | #ifdef CONFIG_PARAVIRT |
83 | #include <asm/paravirt.h> | |
84 | #endif | |
1da177e4 | 85 | |
029632fb | 86 | #include "sched.h" |
ea138446 | 87 | #include "../workqueue_internal.h" |
29d5e047 | 88 | #include "../smpboot.h" |
6e0534f2 | 89 | |
a8d154b0 | 90 | #define CREATE_TRACE_POINTS |
ad8d75ff | 91 | #include <trace/events/sched.h> |
a8d154b0 | 92 | |
029632fb | 93 | void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period) |
d0b27fa7 | 94 | { |
58088ad0 PT |
95 | unsigned long delta; |
96 | ktime_t soft, hard, now; | |
d0b27fa7 | 97 | |
58088ad0 PT |
98 | for (;;) { |
99 | if (hrtimer_active(period_timer)) | |
100 | break; | |
101 | ||
102 | now = hrtimer_cb_get_time(period_timer); | |
103 | hrtimer_forward(period_timer, now, period); | |
d0b27fa7 | 104 | |
58088ad0 PT |
105 | soft = hrtimer_get_softexpires(period_timer); |
106 | hard = hrtimer_get_expires(period_timer); | |
107 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
108 | __hrtimer_start_range_ns(period_timer, soft, delta, | |
109 | HRTIMER_MODE_ABS_PINNED, 0); | |
110 | } | |
111 | } | |
112 | ||
029632fb PZ |
113 | DEFINE_MUTEX(sched_domains_mutex); |
114 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); | |
dc61b1d6 | 115 | |
fe44d621 | 116 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 117 | |
029632fb | 118 | void update_rq_clock(struct rq *rq) |
3e51f33f | 119 | { |
fe44d621 | 120 | s64 delta; |
305e6835 | 121 | |
61eadef6 | 122 | if (rq->skip_clock_update > 0) |
f26f9aff | 123 | return; |
aa483808 | 124 | |
fe44d621 PZ |
125 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
126 | rq->clock += delta; | |
127 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
128 | } |
129 | ||
bf5c91ba IM |
130 | /* |
131 | * Debugging: various feature bits | |
132 | */ | |
f00b45c1 | 133 | |
f00b45c1 PZ |
134 | #define SCHED_FEAT(name, enabled) \ |
135 | (1UL << __SCHED_FEAT_##name) * enabled | | |
136 | ||
bf5c91ba | 137 | const_debug unsigned int sysctl_sched_features = |
391e43da | 138 | #include "features.h" |
f00b45c1 PZ |
139 | 0; |
140 | ||
141 | #undef SCHED_FEAT | |
142 | ||
143 | #ifdef CONFIG_SCHED_DEBUG | |
144 | #define SCHED_FEAT(name, enabled) \ | |
145 | #name , | |
146 | ||
1292531f | 147 | static const char * const sched_feat_names[] = { |
391e43da | 148 | #include "features.h" |
f00b45c1 PZ |
149 | }; |
150 | ||
151 | #undef SCHED_FEAT | |
152 | ||
34f3a814 | 153 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 154 | { |
f00b45c1 PZ |
155 | int i; |
156 | ||
f8b6d1cc | 157 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
34f3a814 LZ |
158 | if (!(sysctl_sched_features & (1UL << i))) |
159 | seq_puts(m, "NO_"); | |
160 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 161 | } |
34f3a814 | 162 | seq_puts(m, "\n"); |
f00b45c1 | 163 | |
34f3a814 | 164 | return 0; |
f00b45c1 PZ |
165 | } |
166 | ||
f8b6d1cc PZ |
167 | #ifdef HAVE_JUMP_LABEL |
168 | ||
c5905afb IM |
169 | #define jump_label_key__true STATIC_KEY_INIT_TRUE |
170 | #define jump_label_key__false STATIC_KEY_INIT_FALSE | |
f8b6d1cc PZ |
171 | |
172 | #define SCHED_FEAT(name, enabled) \ | |
173 | jump_label_key__##enabled , | |
174 | ||
c5905afb | 175 | struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { |
f8b6d1cc PZ |
176 | #include "features.h" |
177 | }; | |
178 | ||
179 | #undef SCHED_FEAT | |
180 | ||
181 | static void sched_feat_disable(int i) | |
182 | { | |
c5905afb IM |
183 | if (static_key_enabled(&sched_feat_keys[i])) |
184 | static_key_slow_dec(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
185 | } |
186 | ||
187 | static void sched_feat_enable(int i) | |
188 | { | |
c5905afb IM |
189 | if (!static_key_enabled(&sched_feat_keys[i])) |
190 | static_key_slow_inc(&sched_feat_keys[i]); | |
f8b6d1cc PZ |
191 | } |
192 | #else | |
193 | static void sched_feat_disable(int i) { }; | |
194 | static void sched_feat_enable(int i) { }; | |
195 | #endif /* HAVE_JUMP_LABEL */ | |
196 | ||
1a687c2e | 197 | static int sched_feat_set(char *cmp) |
f00b45c1 | 198 | { |
f00b45c1 | 199 | int i; |
1a687c2e | 200 | int neg = 0; |
f00b45c1 | 201 | |
524429c3 | 202 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
203 | neg = 1; |
204 | cmp += 3; | |
205 | } | |
206 | ||
f8b6d1cc | 207 | for (i = 0; i < __SCHED_FEAT_NR; i++) { |
7740191c | 208 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f8b6d1cc | 209 | if (neg) { |
f00b45c1 | 210 | sysctl_sched_features &= ~(1UL << i); |
f8b6d1cc PZ |
211 | sched_feat_disable(i); |
212 | } else { | |
f00b45c1 | 213 | sysctl_sched_features |= (1UL << i); |
f8b6d1cc PZ |
214 | sched_feat_enable(i); |
215 | } | |
f00b45c1 PZ |
216 | break; |
217 | } | |
218 | } | |
219 | ||
1a687c2e MG |
220 | return i; |
221 | } | |
222 | ||
223 | static ssize_t | |
224 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
225 | size_t cnt, loff_t *ppos) | |
226 | { | |
227 | char buf[64]; | |
228 | char *cmp; | |
229 | int i; | |
230 | ||
231 | if (cnt > 63) | |
232 | cnt = 63; | |
233 | ||
234 | if (copy_from_user(&buf, ubuf, cnt)) | |
235 | return -EFAULT; | |
236 | ||
237 | buf[cnt] = 0; | |
238 | cmp = strstrip(buf); | |
239 | ||
240 | i = sched_feat_set(cmp); | |
f8b6d1cc | 241 | if (i == __SCHED_FEAT_NR) |
f00b45c1 PZ |
242 | return -EINVAL; |
243 | ||
42994724 | 244 | *ppos += cnt; |
f00b45c1 PZ |
245 | |
246 | return cnt; | |
247 | } | |
248 | ||
34f3a814 LZ |
249 | static int sched_feat_open(struct inode *inode, struct file *filp) |
250 | { | |
251 | return single_open(filp, sched_feat_show, NULL); | |
252 | } | |
253 | ||
828c0950 | 254 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
255 | .open = sched_feat_open, |
256 | .write = sched_feat_write, | |
257 | .read = seq_read, | |
258 | .llseek = seq_lseek, | |
259 | .release = single_release, | |
f00b45c1 PZ |
260 | }; |
261 | ||
262 | static __init int sched_init_debug(void) | |
263 | { | |
f00b45c1 PZ |
264 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
265 | &sched_feat_fops); | |
266 | ||
267 | return 0; | |
268 | } | |
269 | late_initcall(sched_init_debug); | |
f8b6d1cc | 270 | #endif /* CONFIG_SCHED_DEBUG */ |
bf5c91ba | 271 | |
b82d9fdd PZ |
272 | /* |
273 | * Number of tasks to iterate in a single balance run. | |
274 | * Limited because this is done with IRQs disabled. | |
275 | */ | |
276 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
277 | ||
e9e9250b PZ |
278 | /* |
279 | * period over which we average the RT time consumption, measured | |
280 | * in ms. | |
281 | * | |
282 | * default: 1s | |
283 | */ | |
284 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
285 | ||
fa85ae24 | 286 | /* |
9f0c1e56 | 287 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
288 | * default: 1s |
289 | */ | |
9f0c1e56 | 290 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 291 | |
029632fb | 292 | __read_mostly int scheduler_running; |
6892b75e | 293 | |
9f0c1e56 PZ |
294 | /* |
295 | * part of the period that we allow rt tasks to run in us. | |
296 | * default: 0.95s | |
297 | */ | |
298 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 299 | |
0970d299 | 300 | /* |
0122ec5b | 301 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 302 | */ |
70b97a7f | 303 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
304 | __acquires(rq->lock) |
305 | { | |
0970d299 PZ |
306 | struct rq *rq; |
307 | ||
0122ec5b PZ |
308 | lockdep_assert_held(&p->pi_lock); |
309 | ||
3a5c359a | 310 | for (;;) { |
0970d299 | 311 | rq = task_rq(p); |
05fa785c | 312 | raw_spin_lock(&rq->lock); |
65cc8e48 | 313 | if (likely(rq == task_rq(p))) |
3a5c359a | 314 | return rq; |
05fa785c | 315 | raw_spin_unlock(&rq->lock); |
b29739f9 | 316 | } |
b29739f9 IM |
317 | } |
318 | ||
1da177e4 | 319 | /* |
0122ec5b | 320 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 321 | */ |
70b97a7f | 322 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 323 | __acquires(p->pi_lock) |
1da177e4 LT |
324 | __acquires(rq->lock) |
325 | { | |
70b97a7f | 326 | struct rq *rq; |
1da177e4 | 327 | |
3a5c359a | 328 | for (;;) { |
0122ec5b | 329 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 330 | rq = task_rq(p); |
05fa785c | 331 | raw_spin_lock(&rq->lock); |
65cc8e48 | 332 | if (likely(rq == task_rq(p))) |
3a5c359a | 333 | return rq; |
0122ec5b PZ |
334 | raw_spin_unlock(&rq->lock); |
335 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 336 | } |
1da177e4 LT |
337 | } |
338 | ||
a9957449 | 339 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
340 | __releases(rq->lock) |
341 | { | |
05fa785c | 342 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
343 | } |
344 | ||
0122ec5b PZ |
345 | static inline void |
346 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 347 | __releases(rq->lock) |
0122ec5b | 348 | __releases(p->pi_lock) |
1da177e4 | 349 | { |
0122ec5b PZ |
350 | raw_spin_unlock(&rq->lock); |
351 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
352 | } |
353 | ||
1da177e4 | 354 | /* |
cc2a73b5 | 355 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 356 | */ |
a9957449 | 357 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
358 | __acquires(rq->lock) |
359 | { | |
70b97a7f | 360 | struct rq *rq; |
1da177e4 LT |
361 | |
362 | local_irq_disable(); | |
363 | rq = this_rq(); | |
05fa785c | 364 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
365 | |
366 | return rq; | |
367 | } | |
368 | ||
8f4d37ec PZ |
369 | #ifdef CONFIG_SCHED_HRTICK |
370 | /* | |
371 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 372 | */ |
8f4d37ec | 373 | |
8f4d37ec PZ |
374 | static void hrtick_clear(struct rq *rq) |
375 | { | |
376 | if (hrtimer_active(&rq->hrtick_timer)) | |
377 | hrtimer_cancel(&rq->hrtick_timer); | |
378 | } | |
379 | ||
8f4d37ec PZ |
380 | /* |
381 | * High-resolution timer tick. | |
382 | * Runs from hardirq context with interrupts disabled. | |
383 | */ | |
384 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
385 | { | |
386 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
387 | ||
388 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
389 | ||
05fa785c | 390 | raw_spin_lock(&rq->lock); |
3e51f33f | 391 | update_rq_clock(rq); |
8f4d37ec | 392 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 393 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
394 | |
395 | return HRTIMER_NORESTART; | |
396 | } | |
397 | ||
95e904c7 | 398 | #ifdef CONFIG_SMP |
971ee28c PZ |
399 | |
400 | static int __hrtick_restart(struct rq *rq) | |
401 | { | |
402 | struct hrtimer *timer = &rq->hrtick_timer; | |
403 | ktime_t time = hrtimer_get_softexpires(timer); | |
404 | ||
405 | return __hrtimer_start_range_ns(timer, time, 0, HRTIMER_MODE_ABS_PINNED, 0); | |
406 | } | |
407 | ||
31656519 PZ |
408 | /* |
409 | * called from hardirq (IPI) context | |
410 | */ | |
411 | static void __hrtick_start(void *arg) | |
b328ca18 | 412 | { |
31656519 | 413 | struct rq *rq = arg; |
b328ca18 | 414 | |
05fa785c | 415 | raw_spin_lock(&rq->lock); |
971ee28c | 416 | __hrtick_restart(rq); |
31656519 | 417 | rq->hrtick_csd_pending = 0; |
05fa785c | 418 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
419 | } |
420 | ||
31656519 PZ |
421 | /* |
422 | * Called to set the hrtick timer state. | |
423 | * | |
424 | * called with rq->lock held and irqs disabled | |
425 | */ | |
029632fb | 426 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 427 | { |
31656519 PZ |
428 | struct hrtimer *timer = &rq->hrtick_timer; |
429 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 430 | |
cc584b21 | 431 | hrtimer_set_expires(timer, time); |
31656519 PZ |
432 | |
433 | if (rq == this_rq()) { | |
971ee28c | 434 | __hrtick_restart(rq); |
31656519 | 435 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 436 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
437 | rq->hrtick_csd_pending = 1; |
438 | } | |
b328ca18 PZ |
439 | } |
440 | ||
441 | static int | |
442 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
443 | { | |
444 | int cpu = (int)(long)hcpu; | |
445 | ||
446 | switch (action) { | |
447 | case CPU_UP_CANCELED: | |
448 | case CPU_UP_CANCELED_FROZEN: | |
449 | case CPU_DOWN_PREPARE: | |
450 | case CPU_DOWN_PREPARE_FROZEN: | |
451 | case CPU_DEAD: | |
452 | case CPU_DEAD_FROZEN: | |
31656519 | 453 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
454 | return NOTIFY_OK; |
455 | } | |
456 | ||
457 | return NOTIFY_DONE; | |
458 | } | |
459 | ||
fa748203 | 460 | static __init void init_hrtick(void) |
b328ca18 PZ |
461 | { |
462 | hotcpu_notifier(hotplug_hrtick, 0); | |
463 | } | |
31656519 PZ |
464 | #else |
465 | /* | |
466 | * Called to set the hrtick timer state. | |
467 | * | |
468 | * called with rq->lock held and irqs disabled | |
469 | */ | |
029632fb | 470 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 471 | { |
7f1e2ca9 | 472 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 473 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 474 | } |
b328ca18 | 475 | |
006c75f1 | 476 | static inline void init_hrtick(void) |
8f4d37ec | 477 | { |
8f4d37ec | 478 | } |
31656519 | 479 | #endif /* CONFIG_SMP */ |
8f4d37ec | 480 | |
31656519 | 481 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 482 | { |
31656519 PZ |
483 | #ifdef CONFIG_SMP |
484 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 485 | |
31656519 PZ |
486 | rq->hrtick_csd.flags = 0; |
487 | rq->hrtick_csd.func = __hrtick_start; | |
488 | rq->hrtick_csd.info = rq; | |
489 | #endif | |
8f4d37ec | 490 | |
31656519 PZ |
491 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
492 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 493 | } |
006c75f1 | 494 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
495 | static inline void hrtick_clear(struct rq *rq) |
496 | { | |
497 | } | |
498 | ||
8f4d37ec PZ |
499 | static inline void init_rq_hrtick(struct rq *rq) |
500 | { | |
501 | } | |
502 | ||
b328ca18 PZ |
503 | static inline void init_hrtick(void) |
504 | { | |
505 | } | |
006c75f1 | 506 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 507 | |
fd99f91a PZ |
508 | /* |
509 | * cmpxchg based fetch_or, macro so it works for different integer types | |
510 | */ | |
511 | #define fetch_or(ptr, val) \ | |
512 | ({ typeof(*(ptr)) __old, __val = *(ptr); \ | |
513 | for (;;) { \ | |
514 | __old = cmpxchg((ptr), __val, __val | (val)); \ | |
515 | if (__old == __val) \ | |
516 | break; \ | |
517 | __val = __old; \ | |
518 | } \ | |
519 | __old; \ | |
520 | }) | |
521 | ||
522 | #ifdef TIF_POLLING_NRFLAG | |
523 | /* | |
524 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
525 | * this avoids any races wrt polling state changes and thereby avoids | |
526 | * spurious IPIs. | |
527 | */ | |
528 | static bool set_nr_and_not_polling(struct task_struct *p) | |
529 | { | |
530 | struct thread_info *ti = task_thread_info(p); | |
531 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
532 | } | |
533 | #else | |
534 | static bool set_nr_and_not_polling(struct task_struct *p) | |
535 | { | |
536 | set_tsk_need_resched(p); | |
537 | return true; | |
538 | } | |
539 | #endif | |
540 | ||
c24d20db IM |
541 | /* |
542 | * resched_task - mark a task 'to be rescheduled now'. | |
543 | * | |
544 | * On UP this means the setting of the need_resched flag, on SMP it | |
545 | * might also involve a cross-CPU call to trigger the scheduler on | |
546 | * the target CPU. | |
547 | */ | |
029632fb | 548 | void resched_task(struct task_struct *p) |
c24d20db IM |
549 | { |
550 | int cpu; | |
551 | ||
b021fe3e | 552 | lockdep_assert_held(&task_rq(p)->lock); |
c24d20db | 553 | |
5ed0cec0 | 554 | if (test_tsk_need_resched(p)) |
c24d20db IM |
555 | return; |
556 | ||
c24d20db | 557 | cpu = task_cpu(p); |
fd99f91a | 558 | |
f27dde8d | 559 | if (cpu == smp_processor_id()) { |
fd99f91a | 560 | set_tsk_need_resched(p); |
f27dde8d | 561 | set_preempt_need_resched(); |
c24d20db | 562 | return; |
f27dde8d | 563 | } |
c24d20db | 564 | |
fd99f91a | 565 | if (set_nr_and_not_polling(p)) |
c24d20db IM |
566 | smp_send_reschedule(cpu); |
567 | } | |
568 | ||
029632fb | 569 | void resched_cpu(int cpu) |
c24d20db IM |
570 | { |
571 | struct rq *rq = cpu_rq(cpu); | |
572 | unsigned long flags; | |
573 | ||
05fa785c | 574 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
575 | return; |
576 | resched_task(cpu_curr(cpu)); | |
05fa785c | 577 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 578 | } |
06d8308c | 579 | |
b021fe3e | 580 | #ifdef CONFIG_SMP |
3451d024 | 581 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
582 | /* |
583 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
584 | * from an idle cpu. This is good for power-savings. | |
585 | * | |
586 | * We don't do similar optimization for completely idle system, as | |
587 | * selecting an idle cpu will add more delays to the timers than intended | |
588 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
589 | */ | |
6201b4d6 | 590 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
591 | { |
592 | int cpu = smp_processor_id(); | |
593 | int i; | |
594 | struct sched_domain *sd; | |
595 | ||
6201b4d6 VK |
596 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
597 | return cpu; | |
598 | ||
057f3fad | 599 | rcu_read_lock(); |
83cd4fe2 | 600 | for_each_domain(cpu, sd) { |
057f3fad PZ |
601 | for_each_cpu(i, sched_domain_span(sd)) { |
602 | if (!idle_cpu(i)) { | |
603 | cpu = i; | |
604 | goto unlock; | |
605 | } | |
606 | } | |
83cd4fe2 | 607 | } |
057f3fad PZ |
608 | unlock: |
609 | rcu_read_unlock(); | |
83cd4fe2 VP |
610 | return cpu; |
611 | } | |
06d8308c TG |
612 | /* |
613 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
614 | * idle CPU then this timer might expire before the next timer event | |
615 | * which is scheduled to wake up that CPU. In case of a completely | |
616 | * idle system the next event might even be infinite time into the | |
617 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
618 | * leaves the inner idle loop so the newly added timer is taken into | |
619 | * account when the CPU goes back to idle and evaluates the timer | |
620 | * wheel for the next timer event. | |
621 | */ | |
1c20091e | 622 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
623 | { |
624 | struct rq *rq = cpu_rq(cpu); | |
625 | ||
626 | if (cpu == smp_processor_id()) | |
627 | return; | |
628 | ||
629 | /* | |
630 | * This is safe, as this function is called with the timer | |
631 | * wheel base lock of (cpu) held. When the CPU is on the way | |
632 | * to idle and has not yet set rq->curr to idle then it will | |
633 | * be serialized on the timer wheel base lock and take the new | |
634 | * timer into account automatically. | |
635 | */ | |
636 | if (rq->curr != rq->idle) | |
637 | return; | |
45bf76df | 638 | |
45bf76df | 639 | /* |
06d8308c TG |
640 | * We can set TIF_RESCHED on the idle task of the other CPU |
641 | * lockless. The worst case is that the other CPU runs the | |
642 | * idle task through an additional NOOP schedule() | |
45bf76df | 643 | */ |
5ed0cec0 | 644 | set_tsk_need_resched(rq->idle); |
45bf76df | 645 | |
06d8308c TG |
646 | /* NEED_RESCHED must be visible before we test polling */ |
647 | smp_mb(); | |
648 | if (!tsk_is_polling(rq->idle)) | |
649 | smp_send_reschedule(cpu); | |
45bf76df IM |
650 | } |
651 | ||
c5bfece2 | 652 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 653 | { |
c5bfece2 | 654 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
655 | if (cpu != smp_processor_id() || |
656 | tick_nohz_tick_stopped()) | |
657 | smp_send_reschedule(cpu); | |
658 | return true; | |
659 | } | |
660 | ||
661 | return false; | |
662 | } | |
663 | ||
664 | void wake_up_nohz_cpu(int cpu) | |
665 | { | |
c5bfece2 | 666 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
667 | wake_up_idle_cpu(cpu); |
668 | } | |
669 | ||
ca38062e | 670 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 671 | { |
1c792db7 | 672 | int cpu = smp_processor_id(); |
873b4c65 VG |
673 | |
674 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
675 | return false; | |
676 | ||
677 | if (idle_cpu(cpu) && !need_resched()) | |
678 | return true; | |
679 | ||
680 | /* | |
681 | * We can't run Idle Load Balance on this CPU for this time so we | |
682 | * cancel it and clear NOHZ_BALANCE_KICK | |
683 | */ | |
684 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
685 | return false; | |
45bf76df IM |
686 | } |
687 | ||
3451d024 | 688 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 689 | |
ca38062e | 690 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 691 | { |
ca38062e | 692 | return false; |
2069dd75 PZ |
693 | } |
694 | ||
3451d024 | 695 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 696 | |
ce831b38 FW |
697 | #ifdef CONFIG_NO_HZ_FULL |
698 | bool sched_can_stop_tick(void) | |
699 | { | |
700 | struct rq *rq; | |
701 | ||
702 | rq = this_rq(); | |
703 | ||
704 | /* Make sure rq->nr_running update is visible after the IPI */ | |
705 | smp_rmb(); | |
706 | ||
707 | /* More than one running task need preemption */ | |
708 | if (rq->nr_running > 1) | |
709 | return false; | |
710 | ||
711 | return true; | |
712 | } | |
713 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 714 | |
029632fb | 715 | void sched_avg_update(struct rq *rq) |
18d95a28 | 716 | { |
e9e9250b PZ |
717 | s64 period = sched_avg_period(); |
718 | ||
78becc27 | 719 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
720 | /* |
721 | * Inline assembly required to prevent the compiler | |
722 | * optimising this loop into a divmod call. | |
723 | * See __iter_div_u64_rem() for another example of this. | |
724 | */ | |
725 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
726 | rq->age_stamp += period; |
727 | rq->rt_avg /= 2; | |
728 | } | |
18d95a28 PZ |
729 | } |
730 | ||
6d6bc0ad | 731 | #endif /* CONFIG_SMP */ |
18d95a28 | 732 | |
a790de99 PT |
733 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
734 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 735 | /* |
8277434e PT |
736 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
737 | * node and @up when leaving it for the final time. | |
738 | * | |
739 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 740 | */ |
029632fb | 741 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 742 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
743 | { |
744 | struct task_group *parent, *child; | |
eb755805 | 745 | int ret; |
c09595f6 | 746 | |
8277434e PT |
747 | parent = from; |
748 | ||
c09595f6 | 749 | down: |
eb755805 PZ |
750 | ret = (*down)(parent, data); |
751 | if (ret) | |
8277434e | 752 | goto out; |
c09595f6 PZ |
753 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
754 | parent = child; | |
755 | goto down; | |
756 | ||
757 | up: | |
758 | continue; | |
759 | } | |
eb755805 | 760 | ret = (*up)(parent, data); |
8277434e PT |
761 | if (ret || parent == from) |
762 | goto out; | |
c09595f6 PZ |
763 | |
764 | child = parent; | |
765 | parent = parent->parent; | |
766 | if (parent) | |
767 | goto up; | |
8277434e | 768 | out: |
eb755805 | 769 | return ret; |
c09595f6 PZ |
770 | } |
771 | ||
029632fb | 772 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 773 | { |
e2b245f8 | 774 | return 0; |
eb755805 | 775 | } |
18d95a28 PZ |
776 | #endif |
777 | ||
45bf76df IM |
778 | static void set_load_weight(struct task_struct *p) |
779 | { | |
f05998d4 NR |
780 | int prio = p->static_prio - MAX_RT_PRIO; |
781 | struct load_weight *load = &p->se.load; | |
782 | ||
dd41f596 IM |
783 | /* |
784 | * SCHED_IDLE tasks get minimal weight: | |
785 | */ | |
786 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 787 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 788 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
789 | return; |
790 | } | |
71f8bd46 | 791 | |
c8b28116 | 792 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 793 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
794 | } |
795 | ||
371fd7e7 | 796 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 797 | { |
a64692a3 | 798 | update_rq_clock(rq); |
43148951 | 799 | sched_info_queued(rq, p); |
371fd7e7 | 800 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
801 | } |
802 | ||
371fd7e7 | 803 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 804 | { |
a64692a3 | 805 | update_rq_clock(rq); |
43148951 | 806 | sched_info_dequeued(rq, p); |
371fd7e7 | 807 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
808 | } |
809 | ||
029632fb | 810 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
811 | { |
812 | if (task_contributes_to_load(p)) | |
813 | rq->nr_uninterruptible--; | |
814 | ||
371fd7e7 | 815 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
816 | } |
817 | ||
029632fb | 818 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
819 | { |
820 | if (task_contributes_to_load(p)) | |
821 | rq->nr_uninterruptible++; | |
822 | ||
371fd7e7 | 823 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
824 | } |
825 | ||
fe44d621 | 826 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 827 | { |
095c0aa8 GC |
828 | /* |
829 | * In theory, the compile should just see 0 here, and optimize out the call | |
830 | * to sched_rt_avg_update. But I don't trust it... | |
831 | */ | |
832 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
833 | s64 steal = 0, irq_delta = 0; | |
834 | #endif | |
835 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 836 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
837 | |
838 | /* | |
839 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
840 | * this case when a previous update_rq_clock() happened inside a | |
841 | * {soft,}irq region. | |
842 | * | |
843 | * When this happens, we stop ->clock_task and only update the | |
844 | * prev_irq_time stamp to account for the part that fit, so that a next | |
845 | * update will consume the rest. This ensures ->clock_task is | |
846 | * monotonic. | |
847 | * | |
848 | * It does however cause some slight miss-attribution of {soft,}irq | |
849 | * time, a more accurate solution would be to update the irq_time using | |
850 | * the current rq->clock timestamp, except that would require using | |
851 | * atomic ops. | |
852 | */ | |
853 | if (irq_delta > delta) | |
854 | irq_delta = delta; | |
855 | ||
856 | rq->prev_irq_time += irq_delta; | |
857 | delta -= irq_delta; | |
095c0aa8 GC |
858 | #endif |
859 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 860 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
861 | steal = paravirt_steal_clock(cpu_of(rq)); |
862 | steal -= rq->prev_steal_time_rq; | |
863 | ||
864 | if (unlikely(steal > delta)) | |
865 | steal = delta; | |
866 | ||
095c0aa8 | 867 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
868 | delta -= steal; |
869 | } | |
870 | #endif | |
871 | ||
fe44d621 PZ |
872 | rq->clock_task += delta; |
873 | ||
095c0aa8 GC |
874 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
875 | if ((irq_delta + steal) && sched_feat(NONTASK_POWER)) | |
876 | sched_rt_avg_update(rq, irq_delta + steal); | |
877 | #endif | |
aa483808 VP |
878 | } |
879 | ||
34f971f6 PZ |
880 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
881 | { | |
882 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
883 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
884 | ||
885 | if (stop) { | |
886 | /* | |
887 | * Make it appear like a SCHED_FIFO task, its something | |
888 | * userspace knows about and won't get confused about. | |
889 | * | |
890 | * Also, it will make PI more or less work without too | |
891 | * much confusion -- but then, stop work should not | |
892 | * rely on PI working anyway. | |
893 | */ | |
894 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
895 | ||
896 | stop->sched_class = &stop_sched_class; | |
897 | } | |
898 | ||
899 | cpu_rq(cpu)->stop = stop; | |
900 | ||
901 | if (old_stop) { | |
902 | /* | |
903 | * Reset it back to a normal scheduling class so that | |
904 | * it can die in pieces. | |
905 | */ | |
906 | old_stop->sched_class = &rt_sched_class; | |
907 | } | |
908 | } | |
909 | ||
14531189 | 910 | /* |
dd41f596 | 911 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 912 | */ |
14531189 IM |
913 | static inline int __normal_prio(struct task_struct *p) |
914 | { | |
dd41f596 | 915 | return p->static_prio; |
14531189 IM |
916 | } |
917 | ||
b29739f9 IM |
918 | /* |
919 | * Calculate the expected normal priority: i.e. priority | |
920 | * without taking RT-inheritance into account. Might be | |
921 | * boosted by interactivity modifiers. Changes upon fork, | |
922 | * setprio syscalls, and whenever the interactivity | |
923 | * estimator recalculates. | |
924 | */ | |
36c8b586 | 925 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
926 | { |
927 | int prio; | |
928 | ||
aab03e05 DF |
929 | if (task_has_dl_policy(p)) |
930 | prio = MAX_DL_PRIO-1; | |
931 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
932 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
933 | else | |
934 | prio = __normal_prio(p); | |
935 | return prio; | |
936 | } | |
937 | ||
938 | /* | |
939 | * Calculate the current priority, i.e. the priority | |
940 | * taken into account by the scheduler. This value might | |
941 | * be boosted by RT tasks, or might be boosted by | |
942 | * interactivity modifiers. Will be RT if the task got | |
943 | * RT-boosted. If not then it returns p->normal_prio. | |
944 | */ | |
36c8b586 | 945 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
946 | { |
947 | p->normal_prio = normal_prio(p); | |
948 | /* | |
949 | * If we are RT tasks or we were boosted to RT priority, | |
950 | * keep the priority unchanged. Otherwise, update priority | |
951 | * to the normal priority: | |
952 | */ | |
953 | if (!rt_prio(p->prio)) | |
954 | return p->normal_prio; | |
955 | return p->prio; | |
956 | } | |
957 | ||
1da177e4 LT |
958 | /** |
959 | * task_curr - is this task currently executing on a CPU? | |
960 | * @p: the task in question. | |
e69f6186 YB |
961 | * |
962 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 963 | */ |
36c8b586 | 964 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
965 | { |
966 | return cpu_curr(task_cpu(p)) == p; | |
967 | } | |
968 | ||
cb469845 SR |
969 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
970 | const struct sched_class *prev_class, | |
da7a735e | 971 | int oldprio) |
cb469845 SR |
972 | { |
973 | if (prev_class != p->sched_class) { | |
974 | if (prev_class->switched_from) | |
da7a735e PZ |
975 | prev_class->switched_from(rq, p); |
976 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 977 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 978 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
979 | } |
980 | ||
029632fb | 981 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
982 | { |
983 | const struct sched_class *class; | |
984 | ||
985 | if (p->sched_class == rq->curr->sched_class) { | |
986 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
987 | } else { | |
988 | for_each_class(class) { | |
989 | if (class == rq->curr->sched_class) | |
990 | break; | |
991 | if (class == p->sched_class) { | |
992 | resched_task(rq->curr); | |
993 | break; | |
994 | } | |
995 | } | |
996 | } | |
997 | ||
998 | /* | |
999 | * A queue event has occurred, and we're going to schedule. In | |
1000 | * this case, we can save a useless back to back clock update. | |
1001 | */ | |
fd2f4419 | 1002 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
1003 | rq->skip_clock_update = 1; |
1004 | } | |
1005 | ||
1da177e4 | 1006 | #ifdef CONFIG_SMP |
dd41f596 | 1007 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1008 | { |
e2912009 PZ |
1009 | #ifdef CONFIG_SCHED_DEBUG |
1010 | /* | |
1011 | * We should never call set_task_cpu() on a blocked task, | |
1012 | * ttwu() will sort out the placement. | |
1013 | */ | |
077614ee | 1014 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 1015 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
1016 | |
1017 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1018 | /* |
1019 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1020 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1021 | * | |
1022 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1023 | * see task_group(). |
6c6c54e1 PZ |
1024 | * |
1025 | * Furthermore, all task_rq users should acquire both locks, see | |
1026 | * task_rq_lock(). | |
1027 | */ | |
0122ec5b PZ |
1028 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1029 | lockdep_is_held(&task_rq(p)->lock))); | |
1030 | #endif | |
e2912009 PZ |
1031 | #endif |
1032 | ||
de1d7286 | 1033 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1034 | |
0c69774e | 1035 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1036 | if (p->sched_class->migrate_task_rq) |
1037 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1038 | p->se.nr_migrations++; |
a8b0ca17 | 1039 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1040 | } |
dd41f596 IM |
1041 | |
1042 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1043 | } |
1044 | ||
ac66f547 PZ |
1045 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1046 | { | |
1047 | if (p->on_rq) { | |
1048 | struct rq *src_rq, *dst_rq; | |
1049 | ||
1050 | src_rq = task_rq(p); | |
1051 | dst_rq = cpu_rq(cpu); | |
1052 | ||
1053 | deactivate_task(src_rq, p, 0); | |
1054 | set_task_cpu(p, cpu); | |
1055 | activate_task(dst_rq, p, 0); | |
1056 | check_preempt_curr(dst_rq, p, 0); | |
1057 | } else { | |
1058 | /* | |
1059 | * Task isn't running anymore; make it appear like we migrated | |
1060 | * it before it went to sleep. This means on wakeup we make the | |
1061 | * previous cpu our targer instead of where it really is. | |
1062 | */ | |
1063 | p->wake_cpu = cpu; | |
1064 | } | |
1065 | } | |
1066 | ||
1067 | struct migration_swap_arg { | |
1068 | struct task_struct *src_task, *dst_task; | |
1069 | int src_cpu, dst_cpu; | |
1070 | }; | |
1071 | ||
1072 | static int migrate_swap_stop(void *data) | |
1073 | { | |
1074 | struct migration_swap_arg *arg = data; | |
1075 | struct rq *src_rq, *dst_rq; | |
1076 | int ret = -EAGAIN; | |
1077 | ||
1078 | src_rq = cpu_rq(arg->src_cpu); | |
1079 | dst_rq = cpu_rq(arg->dst_cpu); | |
1080 | ||
74602315 PZ |
1081 | double_raw_lock(&arg->src_task->pi_lock, |
1082 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1083 | double_rq_lock(src_rq, dst_rq); |
1084 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1085 | goto unlock; | |
1086 | ||
1087 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1088 | goto unlock; | |
1089 | ||
1090 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1091 | goto unlock; | |
1092 | ||
1093 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1094 | goto unlock; | |
1095 | ||
1096 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1097 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1098 | ||
1099 | ret = 0; | |
1100 | ||
1101 | unlock: | |
1102 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1103 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1104 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1105 | |
1106 | return ret; | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Cross migrate two tasks | |
1111 | */ | |
1112 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1113 | { | |
1114 | struct migration_swap_arg arg; | |
1115 | int ret = -EINVAL; | |
1116 | ||
ac66f547 PZ |
1117 | arg = (struct migration_swap_arg){ |
1118 | .src_task = cur, | |
1119 | .src_cpu = task_cpu(cur), | |
1120 | .dst_task = p, | |
1121 | .dst_cpu = task_cpu(p), | |
1122 | }; | |
1123 | ||
1124 | if (arg.src_cpu == arg.dst_cpu) | |
1125 | goto out; | |
1126 | ||
6acce3ef PZ |
1127 | /* |
1128 | * These three tests are all lockless; this is OK since all of them | |
1129 | * will be re-checked with proper locks held further down the line. | |
1130 | */ | |
ac66f547 PZ |
1131 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1132 | goto out; | |
1133 | ||
1134 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1135 | goto out; | |
1136 | ||
1137 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1138 | goto out; | |
1139 | ||
286549dc | 1140 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1141 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1142 | ||
1143 | out: | |
ac66f547 PZ |
1144 | return ret; |
1145 | } | |
1146 | ||
969c7921 | 1147 | struct migration_arg { |
36c8b586 | 1148 | struct task_struct *task; |
1da177e4 | 1149 | int dest_cpu; |
70b97a7f | 1150 | }; |
1da177e4 | 1151 | |
969c7921 TH |
1152 | static int migration_cpu_stop(void *data); |
1153 | ||
1da177e4 LT |
1154 | /* |
1155 | * wait_task_inactive - wait for a thread to unschedule. | |
1156 | * | |
85ba2d86 RM |
1157 | * If @match_state is nonzero, it's the @p->state value just checked and |
1158 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1159 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1160 | * we return a positive number (its total switch count). If a second call | |
1161 | * a short while later returns the same number, the caller can be sure that | |
1162 | * @p has remained unscheduled the whole time. | |
1163 | * | |
1da177e4 LT |
1164 | * The caller must ensure that the task *will* unschedule sometime soon, |
1165 | * else this function might spin for a *long* time. This function can't | |
1166 | * be called with interrupts off, or it may introduce deadlock with | |
1167 | * smp_call_function() if an IPI is sent by the same process we are | |
1168 | * waiting to become inactive. | |
1169 | */ | |
85ba2d86 | 1170 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1171 | { |
1172 | unsigned long flags; | |
dd41f596 | 1173 | int running, on_rq; |
85ba2d86 | 1174 | unsigned long ncsw; |
70b97a7f | 1175 | struct rq *rq; |
1da177e4 | 1176 | |
3a5c359a AK |
1177 | for (;;) { |
1178 | /* | |
1179 | * We do the initial early heuristics without holding | |
1180 | * any task-queue locks at all. We'll only try to get | |
1181 | * the runqueue lock when things look like they will | |
1182 | * work out! | |
1183 | */ | |
1184 | rq = task_rq(p); | |
fa490cfd | 1185 | |
3a5c359a AK |
1186 | /* |
1187 | * If the task is actively running on another CPU | |
1188 | * still, just relax and busy-wait without holding | |
1189 | * any locks. | |
1190 | * | |
1191 | * NOTE! Since we don't hold any locks, it's not | |
1192 | * even sure that "rq" stays as the right runqueue! | |
1193 | * But we don't care, since "task_running()" will | |
1194 | * return false if the runqueue has changed and p | |
1195 | * is actually now running somewhere else! | |
1196 | */ | |
85ba2d86 RM |
1197 | while (task_running(rq, p)) { |
1198 | if (match_state && unlikely(p->state != match_state)) | |
1199 | return 0; | |
3a5c359a | 1200 | cpu_relax(); |
85ba2d86 | 1201 | } |
fa490cfd | 1202 | |
3a5c359a AK |
1203 | /* |
1204 | * Ok, time to look more closely! We need the rq | |
1205 | * lock now, to be *sure*. If we're wrong, we'll | |
1206 | * just go back and repeat. | |
1207 | */ | |
1208 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1209 | trace_sched_wait_task(p); |
3a5c359a | 1210 | running = task_running(rq, p); |
fd2f4419 | 1211 | on_rq = p->on_rq; |
85ba2d86 | 1212 | ncsw = 0; |
f31e11d8 | 1213 | if (!match_state || p->state == match_state) |
93dcf55f | 1214 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1215 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1216 | |
85ba2d86 RM |
1217 | /* |
1218 | * If it changed from the expected state, bail out now. | |
1219 | */ | |
1220 | if (unlikely(!ncsw)) | |
1221 | break; | |
1222 | ||
3a5c359a AK |
1223 | /* |
1224 | * Was it really running after all now that we | |
1225 | * checked with the proper locks actually held? | |
1226 | * | |
1227 | * Oops. Go back and try again.. | |
1228 | */ | |
1229 | if (unlikely(running)) { | |
1230 | cpu_relax(); | |
1231 | continue; | |
1232 | } | |
fa490cfd | 1233 | |
3a5c359a AK |
1234 | /* |
1235 | * It's not enough that it's not actively running, | |
1236 | * it must be off the runqueue _entirely_, and not | |
1237 | * preempted! | |
1238 | * | |
80dd99b3 | 1239 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1240 | * running right now), it's preempted, and we should |
1241 | * yield - it could be a while. | |
1242 | */ | |
1243 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1244 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1245 | ||
1246 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1247 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1248 | continue; |
1249 | } | |
fa490cfd | 1250 | |
3a5c359a AK |
1251 | /* |
1252 | * Ahh, all good. It wasn't running, and it wasn't | |
1253 | * runnable, which means that it will never become | |
1254 | * running in the future either. We're all done! | |
1255 | */ | |
1256 | break; | |
1257 | } | |
85ba2d86 RM |
1258 | |
1259 | return ncsw; | |
1da177e4 LT |
1260 | } |
1261 | ||
1262 | /*** | |
1263 | * kick_process - kick a running thread to enter/exit the kernel | |
1264 | * @p: the to-be-kicked thread | |
1265 | * | |
1266 | * Cause a process which is running on another CPU to enter | |
1267 | * kernel-mode, without any delay. (to get signals handled.) | |
1268 | * | |
25985edc | 1269 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1270 | * because all it wants to ensure is that the remote task enters |
1271 | * the kernel. If the IPI races and the task has been migrated | |
1272 | * to another CPU then no harm is done and the purpose has been | |
1273 | * achieved as well. | |
1274 | */ | |
36c8b586 | 1275 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1276 | { |
1277 | int cpu; | |
1278 | ||
1279 | preempt_disable(); | |
1280 | cpu = task_cpu(p); | |
1281 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1282 | smp_send_reschedule(cpu); | |
1283 | preempt_enable(); | |
1284 | } | |
b43e3521 | 1285 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1286 | #endif /* CONFIG_SMP */ |
1da177e4 | 1287 | |
970b13ba | 1288 | #ifdef CONFIG_SMP |
30da688e | 1289 | /* |
013fdb80 | 1290 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1291 | */ |
5da9a0fb PZ |
1292 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1293 | { | |
aa00d89c TC |
1294 | int nid = cpu_to_node(cpu); |
1295 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1296 | enum { cpuset, possible, fail } state = cpuset; |
1297 | int dest_cpu; | |
5da9a0fb | 1298 | |
aa00d89c TC |
1299 | /* |
1300 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1301 | * will return -1. There is no cpu on the node, and we should | |
1302 | * select the cpu on the other node. | |
1303 | */ | |
1304 | if (nid != -1) { | |
1305 | nodemask = cpumask_of_node(nid); | |
1306 | ||
1307 | /* Look for allowed, online CPU in same node. */ | |
1308 | for_each_cpu(dest_cpu, nodemask) { | |
1309 | if (!cpu_online(dest_cpu)) | |
1310 | continue; | |
1311 | if (!cpu_active(dest_cpu)) | |
1312 | continue; | |
1313 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1314 | return dest_cpu; | |
1315 | } | |
2baab4e9 | 1316 | } |
5da9a0fb | 1317 | |
2baab4e9 PZ |
1318 | for (;;) { |
1319 | /* Any allowed, online CPU? */ | |
e3831edd | 1320 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1321 | if (!cpu_online(dest_cpu)) |
1322 | continue; | |
1323 | if (!cpu_active(dest_cpu)) | |
1324 | continue; | |
1325 | goto out; | |
1326 | } | |
5da9a0fb | 1327 | |
2baab4e9 PZ |
1328 | switch (state) { |
1329 | case cpuset: | |
1330 | /* No more Mr. Nice Guy. */ | |
1331 | cpuset_cpus_allowed_fallback(p); | |
1332 | state = possible; | |
1333 | break; | |
1334 | ||
1335 | case possible: | |
1336 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1337 | state = fail; | |
1338 | break; | |
1339 | ||
1340 | case fail: | |
1341 | BUG(); | |
1342 | break; | |
1343 | } | |
1344 | } | |
1345 | ||
1346 | out: | |
1347 | if (state != cpuset) { | |
1348 | /* | |
1349 | * Don't tell them about moving exiting tasks or | |
1350 | * kernel threads (both mm NULL), since they never | |
1351 | * leave kernel. | |
1352 | */ | |
1353 | if (p->mm && printk_ratelimit()) { | |
1354 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1355 | task_pid_nr(p), p->comm, cpu); | |
1356 | } | |
5da9a0fb PZ |
1357 | } |
1358 | ||
1359 | return dest_cpu; | |
1360 | } | |
1361 | ||
e2912009 | 1362 | /* |
013fdb80 | 1363 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1364 | */ |
970b13ba | 1365 | static inline |
ac66f547 | 1366 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1367 | { |
ac66f547 | 1368 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1369 | |
1370 | /* | |
1371 | * In order not to call set_task_cpu() on a blocking task we need | |
1372 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1373 | * cpu. | |
1374 | * | |
1375 | * Since this is common to all placement strategies, this lives here. | |
1376 | * | |
1377 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1378 | * not worry about this generic constraint ] | |
1379 | */ | |
fa17b507 | 1380 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1381 | !cpu_online(cpu))) |
5da9a0fb | 1382 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1383 | |
1384 | return cpu; | |
970b13ba | 1385 | } |
09a40af5 MG |
1386 | |
1387 | static void update_avg(u64 *avg, u64 sample) | |
1388 | { | |
1389 | s64 diff = sample - *avg; | |
1390 | *avg += diff >> 3; | |
1391 | } | |
970b13ba PZ |
1392 | #endif |
1393 | ||
d7c01d27 | 1394 | static void |
b84cb5df | 1395 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1396 | { |
d7c01d27 | 1397 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1398 | struct rq *rq = this_rq(); |
1399 | ||
d7c01d27 PZ |
1400 | #ifdef CONFIG_SMP |
1401 | int this_cpu = smp_processor_id(); | |
1402 | ||
1403 | if (cpu == this_cpu) { | |
1404 | schedstat_inc(rq, ttwu_local); | |
1405 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1406 | } else { | |
1407 | struct sched_domain *sd; | |
1408 | ||
1409 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1410 | rcu_read_lock(); |
d7c01d27 PZ |
1411 | for_each_domain(this_cpu, sd) { |
1412 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1413 | schedstat_inc(sd, ttwu_wake_remote); | |
1414 | break; | |
1415 | } | |
1416 | } | |
057f3fad | 1417 | rcu_read_unlock(); |
d7c01d27 | 1418 | } |
f339b9dc PZ |
1419 | |
1420 | if (wake_flags & WF_MIGRATED) | |
1421 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1422 | ||
d7c01d27 PZ |
1423 | #endif /* CONFIG_SMP */ |
1424 | ||
1425 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1426 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1427 | |
1428 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1429 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1430 | |
d7c01d27 PZ |
1431 | #endif /* CONFIG_SCHEDSTATS */ |
1432 | } | |
1433 | ||
1434 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1435 | { | |
9ed3811a | 1436 | activate_task(rq, p, en_flags); |
fd2f4419 | 1437 | p->on_rq = 1; |
c2f7115e PZ |
1438 | |
1439 | /* if a worker is waking up, notify workqueue */ | |
1440 | if (p->flags & PF_WQ_WORKER) | |
1441 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1442 | } |
1443 | ||
23f41eeb PZ |
1444 | /* |
1445 | * Mark the task runnable and perform wakeup-preemption. | |
1446 | */ | |
89363381 | 1447 | static void |
23f41eeb | 1448 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1449 | { |
9ed3811a | 1450 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1451 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1452 | |
1453 | p->state = TASK_RUNNING; | |
1454 | #ifdef CONFIG_SMP | |
1455 | if (p->sched_class->task_woken) | |
1456 | p->sched_class->task_woken(rq, p); | |
1457 | ||
e69c6341 | 1458 | if (rq->idle_stamp) { |
78becc27 | 1459 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1460 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1461 | |
abfafa54 JL |
1462 | update_avg(&rq->avg_idle, delta); |
1463 | ||
1464 | if (rq->avg_idle > max) | |
9ed3811a | 1465 | rq->avg_idle = max; |
abfafa54 | 1466 | |
9ed3811a TH |
1467 | rq->idle_stamp = 0; |
1468 | } | |
1469 | #endif | |
1470 | } | |
1471 | ||
c05fbafb PZ |
1472 | static void |
1473 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1474 | { | |
1475 | #ifdef CONFIG_SMP | |
1476 | if (p->sched_contributes_to_load) | |
1477 | rq->nr_uninterruptible--; | |
1478 | #endif | |
1479 | ||
1480 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1481 | ttwu_do_wakeup(rq, p, wake_flags); | |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1486 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1487 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1488 | * the task is still ->on_rq. | |
1489 | */ | |
1490 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1491 | { | |
1492 | struct rq *rq; | |
1493 | int ret = 0; | |
1494 | ||
1495 | rq = __task_rq_lock(p); | |
1496 | if (p->on_rq) { | |
1ad4ec0d FW |
1497 | /* check_preempt_curr() may use rq clock */ |
1498 | update_rq_clock(rq); | |
c05fbafb PZ |
1499 | ttwu_do_wakeup(rq, p, wake_flags); |
1500 | ret = 1; | |
1501 | } | |
1502 | __task_rq_unlock(rq); | |
1503 | ||
1504 | return ret; | |
1505 | } | |
1506 | ||
317f3941 | 1507 | #ifdef CONFIG_SMP |
fa14ff4a | 1508 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1509 | { |
1510 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1511 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1512 | struct task_struct *p; | |
317f3941 PZ |
1513 | |
1514 | raw_spin_lock(&rq->lock); | |
1515 | ||
fa14ff4a PZ |
1516 | while (llist) { |
1517 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1518 | llist = llist_next(llist); | |
317f3941 PZ |
1519 | ttwu_do_activate(rq, p, 0); |
1520 | } | |
1521 | ||
1522 | raw_spin_unlock(&rq->lock); | |
1523 | } | |
1524 | ||
1525 | void scheduler_ipi(void) | |
1526 | { | |
f27dde8d PZ |
1527 | /* |
1528 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1529 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1530 | * this IPI. | |
1531 | */ | |
8cb75e0c | 1532 | preempt_fold_need_resched(); |
f27dde8d | 1533 | |
873b4c65 VG |
1534 | if (llist_empty(&this_rq()->wake_list) |
1535 | && !tick_nohz_full_cpu(smp_processor_id()) | |
1536 | && !got_nohz_idle_kick()) | |
c5d753a5 PZ |
1537 | return; |
1538 | ||
1539 | /* | |
1540 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1541 | * traditionally all their work was done from the interrupt return | |
1542 | * path. Now that we actually do some work, we need to make sure | |
1543 | * we do call them. | |
1544 | * | |
1545 | * Some archs already do call them, luckily irq_enter/exit nest | |
1546 | * properly. | |
1547 | * | |
1548 | * Arguably we should visit all archs and update all handlers, | |
1549 | * however a fair share of IPIs are still resched only so this would | |
1550 | * somewhat pessimize the simple resched case. | |
1551 | */ | |
1552 | irq_enter(); | |
ff442c51 | 1553 | tick_nohz_full_check(); |
fa14ff4a | 1554 | sched_ttwu_pending(); |
ca38062e SS |
1555 | |
1556 | /* | |
1557 | * Check if someone kicked us for doing the nohz idle load balance. | |
1558 | */ | |
873b4c65 | 1559 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1560 | this_rq()->idle_balance = 1; |
ca38062e | 1561 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1562 | } |
c5d753a5 | 1563 | irq_exit(); |
317f3941 PZ |
1564 | } |
1565 | ||
1566 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1567 | { | |
fa14ff4a | 1568 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1569 | smp_send_reschedule(cpu); |
1570 | } | |
d6aa8f85 | 1571 | |
39be3501 | 1572 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1573 | { |
1574 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1575 | } | |
d6aa8f85 | 1576 | #endif /* CONFIG_SMP */ |
317f3941 | 1577 | |
c05fbafb PZ |
1578 | static void ttwu_queue(struct task_struct *p, int cpu) |
1579 | { | |
1580 | struct rq *rq = cpu_rq(cpu); | |
1581 | ||
17d9f311 | 1582 | #if defined(CONFIG_SMP) |
39be3501 | 1583 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1584 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1585 | ttwu_queue_remote(p, cpu); |
1586 | return; | |
1587 | } | |
1588 | #endif | |
1589 | ||
c05fbafb PZ |
1590 | raw_spin_lock(&rq->lock); |
1591 | ttwu_do_activate(rq, p, 0); | |
1592 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1593 | } |
1594 | ||
1595 | /** | |
1da177e4 | 1596 | * try_to_wake_up - wake up a thread |
9ed3811a | 1597 | * @p: the thread to be awakened |
1da177e4 | 1598 | * @state: the mask of task states that can be woken |
9ed3811a | 1599 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1600 | * |
1601 | * Put it on the run-queue if it's not already there. The "current" | |
1602 | * thread is always on the run-queue (except when the actual | |
1603 | * re-schedule is in progress), and as such you're allowed to do | |
1604 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1605 | * runnable without the overhead of this. | |
1606 | * | |
e69f6186 | 1607 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1608 | * or @state didn't match @p's state. |
1da177e4 | 1609 | */ |
e4a52bcb PZ |
1610 | static int |
1611 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1612 | { |
1da177e4 | 1613 | unsigned long flags; |
c05fbafb | 1614 | int cpu, success = 0; |
2398f2c6 | 1615 | |
e0acd0a6 ON |
1616 | /* |
1617 | * If we are going to wake up a thread waiting for CONDITION we | |
1618 | * need to ensure that CONDITION=1 done by the caller can not be | |
1619 | * reordered with p->state check below. This pairs with mb() in | |
1620 | * set_current_state() the waiting thread does. | |
1621 | */ | |
1622 | smp_mb__before_spinlock(); | |
013fdb80 | 1623 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1624 | if (!(p->state & state)) |
1da177e4 LT |
1625 | goto out; |
1626 | ||
c05fbafb | 1627 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1628 | cpu = task_cpu(p); |
1da177e4 | 1629 | |
c05fbafb PZ |
1630 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1631 | goto stat; | |
1da177e4 | 1632 | |
1da177e4 | 1633 | #ifdef CONFIG_SMP |
e9c84311 | 1634 | /* |
c05fbafb PZ |
1635 | * If the owning (remote) cpu is still in the middle of schedule() with |
1636 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1637 | */ |
f3e94786 | 1638 | while (p->on_cpu) |
e4a52bcb | 1639 | cpu_relax(); |
0970d299 | 1640 | /* |
e4a52bcb | 1641 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1642 | */ |
e4a52bcb | 1643 | smp_rmb(); |
1da177e4 | 1644 | |
a8e4f2ea | 1645 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1646 | p->state = TASK_WAKING; |
e7693a36 | 1647 | |
e4a52bcb | 1648 | if (p->sched_class->task_waking) |
74f8e4b2 | 1649 | p->sched_class->task_waking(p); |
efbbd05a | 1650 | |
ac66f547 | 1651 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1652 | if (task_cpu(p) != cpu) { |
1653 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1654 | set_task_cpu(p, cpu); |
f339b9dc | 1655 | } |
1da177e4 | 1656 | #endif /* CONFIG_SMP */ |
1da177e4 | 1657 | |
c05fbafb PZ |
1658 | ttwu_queue(p, cpu); |
1659 | stat: | |
b84cb5df | 1660 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1661 | out: |
013fdb80 | 1662 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1663 | |
1664 | return success; | |
1665 | } | |
1666 | ||
21aa9af0 TH |
1667 | /** |
1668 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1669 | * @p: the thread to be awakened | |
1670 | * | |
2acca55e | 1671 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1672 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1673 | * the current task. |
21aa9af0 TH |
1674 | */ |
1675 | static void try_to_wake_up_local(struct task_struct *p) | |
1676 | { | |
1677 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1678 | |
383efcd0 TH |
1679 | if (WARN_ON_ONCE(rq != this_rq()) || |
1680 | WARN_ON_ONCE(p == current)) | |
1681 | return; | |
1682 | ||
21aa9af0 TH |
1683 | lockdep_assert_held(&rq->lock); |
1684 | ||
2acca55e PZ |
1685 | if (!raw_spin_trylock(&p->pi_lock)) { |
1686 | raw_spin_unlock(&rq->lock); | |
1687 | raw_spin_lock(&p->pi_lock); | |
1688 | raw_spin_lock(&rq->lock); | |
1689 | } | |
1690 | ||
21aa9af0 | 1691 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1692 | goto out; |
21aa9af0 | 1693 | |
fd2f4419 | 1694 | if (!p->on_rq) |
d7c01d27 PZ |
1695 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1696 | ||
23f41eeb | 1697 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1698 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1699 | out: |
1700 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1701 | } |
1702 | ||
50fa610a DH |
1703 | /** |
1704 | * wake_up_process - Wake up a specific process | |
1705 | * @p: The process to be woken up. | |
1706 | * | |
1707 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1708 | * processes. |
1709 | * | |
1710 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1711 | * |
1712 | * It may be assumed that this function implies a write memory barrier before | |
1713 | * changing the task state if and only if any tasks are woken up. | |
1714 | */ | |
7ad5b3a5 | 1715 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1716 | { |
9067ac85 ON |
1717 | WARN_ON(task_is_stopped_or_traced(p)); |
1718 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1719 | } |
1da177e4 LT |
1720 | EXPORT_SYMBOL(wake_up_process); |
1721 | ||
7ad5b3a5 | 1722 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1723 | { |
1724 | return try_to_wake_up(p, state, 0); | |
1725 | } | |
1726 | ||
1da177e4 LT |
1727 | /* |
1728 | * Perform scheduler related setup for a newly forked process p. | |
1729 | * p is forked by current. | |
dd41f596 IM |
1730 | * |
1731 | * __sched_fork() is basic setup used by init_idle() too: | |
1732 | */ | |
5e1576ed | 1733 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1734 | { |
fd2f4419 PZ |
1735 | p->on_rq = 0; |
1736 | ||
1737 | p->se.on_rq = 0; | |
dd41f596 IM |
1738 | p->se.exec_start = 0; |
1739 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1740 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1741 | p->se.nr_migrations = 0; |
da7a735e | 1742 | p->se.vruntime = 0; |
fd2f4419 | 1743 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1744 | |
1745 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1746 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1747 | #endif |
476d139c | 1748 | |
aab03e05 DF |
1749 | RB_CLEAR_NODE(&p->dl.rb_node); |
1750 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1751 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1752 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1753 | p->dl.dl_period = 0; |
aab03e05 DF |
1754 | p->dl.flags = 0; |
1755 | ||
fa717060 | 1756 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1757 | |
e107be36 AK |
1758 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1759 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1760 | #endif | |
cbee9f88 PZ |
1761 | |
1762 | #ifdef CONFIG_NUMA_BALANCING | |
1763 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1764 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1765 | p->mm->numa_scan_seq = 0; |
1766 | } | |
1767 | ||
5e1576ed RR |
1768 | if (clone_flags & CLONE_VM) |
1769 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1770 | else | |
1771 | p->numa_preferred_nid = -1; | |
1772 | ||
cbee9f88 PZ |
1773 | p->node_stamp = 0ULL; |
1774 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1775 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1776 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1777 | p->numa_faults_memory = NULL; |
1778 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1779 | p->last_task_numa_placement = 0; |
1780 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1781 | |
1782 | INIT_LIST_HEAD(&p->numa_entry); | |
1783 | p->numa_group = NULL; | |
cbee9f88 | 1784 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1785 | } |
1786 | ||
1a687c2e | 1787 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1788 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1789 | void set_numabalancing_state(bool enabled) |
1790 | { | |
1791 | if (enabled) | |
1792 | sched_feat_set("NUMA"); | |
1793 | else | |
1794 | sched_feat_set("NO_NUMA"); | |
1795 | } | |
3105b86a MG |
1796 | #else |
1797 | __read_mostly bool numabalancing_enabled; | |
1798 | ||
1799 | void set_numabalancing_state(bool enabled) | |
1800 | { | |
1801 | numabalancing_enabled = enabled; | |
dd41f596 | 1802 | } |
3105b86a | 1803 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1804 | |
1805 | #ifdef CONFIG_PROC_SYSCTL | |
1806 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1807 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1808 | { | |
1809 | struct ctl_table t; | |
1810 | int err; | |
1811 | int state = numabalancing_enabled; | |
1812 | ||
1813 | if (write && !capable(CAP_SYS_ADMIN)) | |
1814 | return -EPERM; | |
1815 | ||
1816 | t = *table; | |
1817 | t.data = &state; | |
1818 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1819 | if (err < 0) | |
1820 | return err; | |
1821 | if (write) | |
1822 | set_numabalancing_state(state); | |
1823 | return err; | |
1824 | } | |
1825 | #endif | |
1826 | #endif | |
dd41f596 IM |
1827 | |
1828 | /* | |
1829 | * fork()/clone()-time setup: | |
1830 | */ | |
aab03e05 | 1831 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1832 | { |
0122ec5b | 1833 | unsigned long flags; |
dd41f596 IM |
1834 | int cpu = get_cpu(); |
1835 | ||
5e1576ed | 1836 | __sched_fork(clone_flags, p); |
06b83b5f | 1837 | /* |
0017d735 | 1838 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1839 | * nobody will actually run it, and a signal or other external |
1840 | * event cannot wake it up and insert it on the runqueue either. | |
1841 | */ | |
0017d735 | 1842 | p->state = TASK_RUNNING; |
dd41f596 | 1843 | |
c350a04e MG |
1844 | /* |
1845 | * Make sure we do not leak PI boosting priority to the child. | |
1846 | */ | |
1847 | p->prio = current->normal_prio; | |
1848 | ||
b9dc29e7 MG |
1849 | /* |
1850 | * Revert to default priority/policy on fork if requested. | |
1851 | */ | |
1852 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1853 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1854 | p->policy = SCHED_NORMAL; |
6c697bdf | 1855 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1856 | p->rt_priority = 0; |
1857 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1858 | p->static_prio = NICE_TO_PRIO(0); | |
1859 | ||
1860 | p->prio = p->normal_prio = __normal_prio(p); | |
1861 | set_load_weight(p); | |
6c697bdf | 1862 | |
b9dc29e7 MG |
1863 | /* |
1864 | * We don't need the reset flag anymore after the fork. It has | |
1865 | * fulfilled its duty: | |
1866 | */ | |
1867 | p->sched_reset_on_fork = 0; | |
1868 | } | |
ca94c442 | 1869 | |
aab03e05 DF |
1870 | if (dl_prio(p->prio)) { |
1871 | put_cpu(); | |
1872 | return -EAGAIN; | |
1873 | } else if (rt_prio(p->prio)) { | |
1874 | p->sched_class = &rt_sched_class; | |
1875 | } else { | |
2ddbf952 | 1876 | p->sched_class = &fair_sched_class; |
aab03e05 | 1877 | } |
b29739f9 | 1878 | |
cd29fe6f PZ |
1879 | if (p->sched_class->task_fork) |
1880 | p->sched_class->task_fork(p); | |
1881 | ||
86951599 PZ |
1882 | /* |
1883 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1884 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1885 | * is ran before sched_fork(). | |
1886 | * | |
1887 | * Silence PROVE_RCU. | |
1888 | */ | |
0122ec5b | 1889 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1890 | set_task_cpu(p, cpu); |
0122ec5b | 1891 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1892 | |
52f17b6c | 1893 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1894 | if (likely(sched_info_on())) |
52f17b6c | 1895 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1896 | #endif |
3ca7a440 PZ |
1897 | #if defined(CONFIG_SMP) |
1898 | p->on_cpu = 0; | |
4866cde0 | 1899 | #endif |
01028747 | 1900 | init_task_preempt_count(p); |
806c09a7 | 1901 | #ifdef CONFIG_SMP |
917b627d | 1902 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1903 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1904 | #endif |
917b627d | 1905 | |
476d139c | 1906 | put_cpu(); |
aab03e05 | 1907 | return 0; |
1da177e4 LT |
1908 | } |
1909 | ||
332ac17e DF |
1910 | unsigned long to_ratio(u64 period, u64 runtime) |
1911 | { | |
1912 | if (runtime == RUNTIME_INF) | |
1913 | return 1ULL << 20; | |
1914 | ||
1915 | /* | |
1916 | * Doing this here saves a lot of checks in all | |
1917 | * the calling paths, and returning zero seems | |
1918 | * safe for them anyway. | |
1919 | */ | |
1920 | if (period == 0) | |
1921 | return 0; | |
1922 | ||
1923 | return div64_u64(runtime << 20, period); | |
1924 | } | |
1925 | ||
1926 | #ifdef CONFIG_SMP | |
1927 | inline struct dl_bw *dl_bw_of(int i) | |
1928 | { | |
1929 | return &cpu_rq(i)->rd->dl_bw; | |
1930 | } | |
1931 | ||
de212f18 | 1932 | static inline int dl_bw_cpus(int i) |
332ac17e | 1933 | { |
de212f18 PZ |
1934 | struct root_domain *rd = cpu_rq(i)->rd; |
1935 | int cpus = 0; | |
1936 | ||
1937 | for_each_cpu_and(i, rd->span, cpu_active_mask) | |
1938 | cpus++; | |
1939 | ||
1940 | return cpus; | |
332ac17e DF |
1941 | } |
1942 | #else | |
1943 | inline struct dl_bw *dl_bw_of(int i) | |
1944 | { | |
1945 | return &cpu_rq(i)->dl.dl_bw; | |
1946 | } | |
1947 | ||
de212f18 | 1948 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
1949 | { |
1950 | return 1; | |
1951 | } | |
1952 | #endif | |
1953 | ||
1954 | static inline | |
1955 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
1956 | { | |
1957 | dl_b->total_bw -= tsk_bw; | |
1958 | } | |
1959 | ||
1960 | static inline | |
1961 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
1962 | { | |
1963 | dl_b->total_bw += tsk_bw; | |
1964 | } | |
1965 | ||
1966 | static inline | |
1967 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
1968 | { | |
1969 | return dl_b->bw != -1 && | |
1970 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * We must be sure that accepting a new task (or allowing changing the | |
1975 | * parameters of an existing one) is consistent with the bandwidth | |
1976 | * constraints. If yes, this function also accordingly updates the currently | |
1977 | * allocated bandwidth to reflect the new situation. | |
1978 | * | |
1979 | * This function is called while holding p's rq->lock. | |
1980 | */ | |
1981 | static int dl_overflow(struct task_struct *p, int policy, | |
1982 | const struct sched_attr *attr) | |
1983 | { | |
1984 | ||
1985 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 1986 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
1987 | u64 runtime = attr->sched_runtime; |
1988 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 1989 | int cpus, err = -1; |
332ac17e DF |
1990 | |
1991 | if (new_bw == p->dl.dl_bw) | |
1992 | return 0; | |
1993 | ||
1994 | /* | |
1995 | * Either if a task, enters, leave, or stays -deadline but changes | |
1996 | * its parameters, we may need to update accordingly the total | |
1997 | * allocated bandwidth of the container. | |
1998 | */ | |
1999 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2000 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2001 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2002 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2003 | __dl_add(dl_b, new_bw); | |
2004 | err = 0; | |
2005 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2006 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2007 | __dl_clear(dl_b, p->dl.dl_bw); | |
2008 | __dl_add(dl_b, new_bw); | |
2009 | err = 0; | |
2010 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2011 | __dl_clear(dl_b, p->dl.dl_bw); | |
2012 | err = 0; | |
2013 | } | |
2014 | raw_spin_unlock(&dl_b->lock); | |
2015 | ||
2016 | return err; | |
2017 | } | |
2018 | ||
2019 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2020 | ||
1da177e4 LT |
2021 | /* |
2022 | * wake_up_new_task - wake up a newly created task for the first time. | |
2023 | * | |
2024 | * This function will do some initial scheduler statistics housekeeping | |
2025 | * that must be done for every newly created context, then puts the task | |
2026 | * on the runqueue and wakes it. | |
2027 | */ | |
3e51e3ed | 2028 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2029 | { |
2030 | unsigned long flags; | |
dd41f596 | 2031 | struct rq *rq; |
fabf318e | 2032 | |
ab2515c4 | 2033 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2034 | #ifdef CONFIG_SMP |
2035 | /* | |
2036 | * Fork balancing, do it here and not earlier because: | |
2037 | * - cpus_allowed can change in the fork path | |
2038 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2039 | */ |
ac66f547 | 2040 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2041 | #endif |
2042 | ||
a75cdaa9 AS |
2043 | /* Initialize new task's runnable average */ |
2044 | init_task_runnable_average(p); | |
ab2515c4 | 2045 | rq = __task_rq_lock(p); |
cd29fe6f | 2046 | activate_task(rq, p, 0); |
fd2f4419 | 2047 | p->on_rq = 1; |
89363381 | 2048 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2049 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2050 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2051 | if (p->sched_class->task_woken) |
2052 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2053 | #endif |
0122ec5b | 2054 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2055 | } |
2056 | ||
e107be36 AK |
2057 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2058 | ||
2059 | /** | |
80dd99b3 | 2060 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2061 | * @notifier: notifier struct to register |
e107be36 AK |
2062 | */ |
2063 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2064 | { | |
2065 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2066 | } | |
2067 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2068 | ||
2069 | /** | |
2070 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2071 | * @notifier: notifier struct to unregister |
e107be36 AK |
2072 | * |
2073 | * This is safe to call from within a preemption notifier. | |
2074 | */ | |
2075 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2076 | { | |
2077 | hlist_del(¬ifier->link); | |
2078 | } | |
2079 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2080 | ||
2081 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2082 | { | |
2083 | struct preempt_notifier *notifier; | |
e107be36 | 2084 | |
b67bfe0d | 2085 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2086 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2087 | } | |
2088 | ||
2089 | static void | |
2090 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2091 | struct task_struct *next) | |
2092 | { | |
2093 | struct preempt_notifier *notifier; | |
e107be36 | 2094 | |
b67bfe0d | 2095 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2096 | notifier->ops->sched_out(notifier, next); |
2097 | } | |
2098 | ||
6d6bc0ad | 2099 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2100 | |
2101 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2102 | { | |
2103 | } | |
2104 | ||
2105 | static void | |
2106 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2107 | struct task_struct *next) | |
2108 | { | |
2109 | } | |
2110 | ||
6d6bc0ad | 2111 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2112 | |
4866cde0 NP |
2113 | /** |
2114 | * prepare_task_switch - prepare to switch tasks | |
2115 | * @rq: the runqueue preparing to switch | |
421cee29 | 2116 | * @prev: the current task that is being switched out |
4866cde0 NP |
2117 | * @next: the task we are going to switch to. |
2118 | * | |
2119 | * This is called with the rq lock held and interrupts off. It must | |
2120 | * be paired with a subsequent finish_task_switch after the context | |
2121 | * switch. | |
2122 | * | |
2123 | * prepare_task_switch sets up locking and calls architecture specific | |
2124 | * hooks. | |
2125 | */ | |
e107be36 AK |
2126 | static inline void |
2127 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2128 | struct task_struct *next) | |
4866cde0 | 2129 | { |
895dd92c | 2130 | trace_sched_switch(prev, next); |
43148951 | 2131 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2132 | perf_event_task_sched_out(prev, next); |
e107be36 | 2133 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2134 | prepare_lock_switch(rq, next); |
2135 | prepare_arch_switch(next); | |
2136 | } | |
2137 | ||
1da177e4 LT |
2138 | /** |
2139 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2140 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2141 | * @prev: the thread we just switched away from. |
2142 | * | |
4866cde0 NP |
2143 | * finish_task_switch must be called after the context switch, paired |
2144 | * with a prepare_task_switch call before the context switch. | |
2145 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2146 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2147 | * |
2148 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2149 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2150 | * with the lock held can cause deadlocks; see schedule() for |
2151 | * details.) | |
2152 | */ | |
a9957449 | 2153 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2154 | __releases(rq->lock) |
2155 | { | |
1da177e4 | 2156 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2157 | long prev_state; |
1da177e4 LT |
2158 | |
2159 | rq->prev_mm = NULL; | |
2160 | ||
2161 | /* | |
2162 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2163 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2164 | * schedule one last time. The schedule call will never return, and |
2165 | * the scheduled task must drop that reference. | |
c394cc9f | 2166 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2167 | * still held, otherwise prev could be scheduled on another cpu, die |
2168 | * there before we look at prev->state, and then the reference would | |
2169 | * be dropped twice. | |
2170 | * Manfred Spraul <manfred@colorfullife.com> | |
2171 | */ | |
55a101f8 | 2172 | prev_state = prev->state; |
bf9fae9f | 2173 | vtime_task_switch(prev); |
4866cde0 | 2174 | finish_arch_switch(prev); |
a8d757ef | 2175 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2176 | finish_lock_switch(rq, prev); |
01f23e16 | 2177 | finish_arch_post_lock_switch(); |
e8fa1362 | 2178 | |
e107be36 | 2179 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2180 | if (mm) |
2181 | mmdrop(mm); | |
c394cc9f | 2182 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2183 | if (prev->sched_class->task_dead) |
2184 | prev->sched_class->task_dead(prev); | |
2185 | ||
c6fd91f0 | 2186 | /* |
2187 | * Remove function-return probe instances associated with this | |
2188 | * task and put them back on the free list. | |
9761eea8 | 2189 | */ |
c6fd91f0 | 2190 | kprobe_flush_task(prev); |
1da177e4 | 2191 | put_task_struct(prev); |
c6fd91f0 | 2192 | } |
99e5ada9 FW |
2193 | |
2194 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2195 | } |
2196 | ||
3f029d3c GH |
2197 | #ifdef CONFIG_SMP |
2198 | ||
3f029d3c GH |
2199 | /* rq->lock is NOT held, but preemption is disabled */ |
2200 | static inline void post_schedule(struct rq *rq) | |
2201 | { | |
2202 | if (rq->post_schedule) { | |
2203 | unsigned long flags; | |
2204 | ||
05fa785c | 2205 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2206 | if (rq->curr->sched_class->post_schedule) |
2207 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2208 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2209 | |
2210 | rq->post_schedule = 0; | |
2211 | } | |
2212 | } | |
2213 | ||
2214 | #else | |
da19ab51 | 2215 | |
3f029d3c GH |
2216 | static inline void post_schedule(struct rq *rq) |
2217 | { | |
1da177e4 LT |
2218 | } |
2219 | ||
3f029d3c GH |
2220 | #endif |
2221 | ||
1da177e4 LT |
2222 | /** |
2223 | * schedule_tail - first thing a freshly forked thread must call. | |
2224 | * @prev: the thread we just switched away from. | |
2225 | */ | |
722a9f92 | 2226 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2227 | __releases(rq->lock) |
2228 | { | |
70b97a7f IM |
2229 | struct rq *rq = this_rq(); |
2230 | ||
4866cde0 | 2231 | finish_task_switch(rq, prev); |
da19ab51 | 2232 | |
3f029d3c GH |
2233 | /* |
2234 | * FIXME: do we need to worry about rq being invalidated by the | |
2235 | * task_switch? | |
2236 | */ | |
2237 | post_schedule(rq); | |
70b97a7f | 2238 | |
4866cde0 NP |
2239 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2240 | /* In this case, finish_task_switch does not reenable preemption */ | |
2241 | preempt_enable(); | |
2242 | #endif | |
1da177e4 | 2243 | if (current->set_child_tid) |
b488893a | 2244 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2245 | } |
2246 | ||
2247 | /* | |
2248 | * context_switch - switch to the new MM and the new | |
2249 | * thread's register state. | |
2250 | */ | |
dd41f596 | 2251 | static inline void |
70b97a7f | 2252 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2253 | struct task_struct *next) |
1da177e4 | 2254 | { |
dd41f596 | 2255 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2256 | |
e107be36 | 2257 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2258 | |
dd41f596 IM |
2259 | mm = next->mm; |
2260 | oldmm = prev->active_mm; | |
9226d125 ZA |
2261 | /* |
2262 | * For paravirt, this is coupled with an exit in switch_to to | |
2263 | * combine the page table reload and the switch backend into | |
2264 | * one hypercall. | |
2265 | */ | |
224101ed | 2266 | arch_start_context_switch(prev); |
9226d125 | 2267 | |
31915ab4 | 2268 | if (!mm) { |
1da177e4 LT |
2269 | next->active_mm = oldmm; |
2270 | atomic_inc(&oldmm->mm_count); | |
2271 | enter_lazy_tlb(oldmm, next); | |
2272 | } else | |
2273 | switch_mm(oldmm, mm, next); | |
2274 | ||
31915ab4 | 2275 | if (!prev->mm) { |
1da177e4 | 2276 | prev->active_mm = NULL; |
1da177e4 LT |
2277 | rq->prev_mm = oldmm; |
2278 | } | |
3a5f5e48 IM |
2279 | /* |
2280 | * Since the runqueue lock will be released by the next | |
2281 | * task (which is an invalid locking op but in the case | |
2282 | * of the scheduler it's an obvious special-case), so we | |
2283 | * do an early lockdep release here: | |
2284 | */ | |
2285 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2286 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2287 | #endif |
1da177e4 | 2288 | |
91d1aa43 | 2289 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2290 | /* Here we just switch the register state and the stack. */ |
2291 | switch_to(prev, next, prev); | |
2292 | ||
dd41f596 IM |
2293 | barrier(); |
2294 | /* | |
2295 | * this_rq must be evaluated again because prev may have moved | |
2296 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2297 | * frame will be invalid. | |
2298 | */ | |
2299 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2300 | } |
2301 | ||
2302 | /* | |
1c3e8264 | 2303 | * nr_running and nr_context_switches: |
1da177e4 LT |
2304 | * |
2305 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2306 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2307 | */ |
2308 | unsigned long nr_running(void) | |
2309 | { | |
2310 | unsigned long i, sum = 0; | |
2311 | ||
2312 | for_each_online_cpu(i) | |
2313 | sum += cpu_rq(i)->nr_running; | |
2314 | ||
2315 | return sum; | |
f711f609 | 2316 | } |
1da177e4 | 2317 | |
1da177e4 | 2318 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2319 | { |
cc94abfc SR |
2320 | int i; |
2321 | unsigned long long sum = 0; | |
46cb4b7c | 2322 | |
0a945022 | 2323 | for_each_possible_cpu(i) |
1da177e4 | 2324 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2325 | |
1da177e4 LT |
2326 | return sum; |
2327 | } | |
483b4ee6 | 2328 | |
1da177e4 LT |
2329 | unsigned long nr_iowait(void) |
2330 | { | |
2331 | unsigned long i, sum = 0; | |
483b4ee6 | 2332 | |
0a945022 | 2333 | for_each_possible_cpu(i) |
1da177e4 | 2334 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2335 | |
1da177e4 LT |
2336 | return sum; |
2337 | } | |
483b4ee6 | 2338 | |
8c215bd3 | 2339 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2340 | { |
8c215bd3 | 2341 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2342 | return atomic_read(&this->nr_iowait); |
2343 | } | |
46cb4b7c | 2344 | |
dd41f596 | 2345 | #ifdef CONFIG_SMP |
8a0be9ef | 2346 | |
46cb4b7c | 2347 | /* |
38022906 PZ |
2348 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2349 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2350 | */ |
38022906 | 2351 | void sched_exec(void) |
46cb4b7c | 2352 | { |
38022906 | 2353 | struct task_struct *p = current; |
1da177e4 | 2354 | unsigned long flags; |
0017d735 | 2355 | int dest_cpu; |
46cb4b7c | 2356 | |
8f42ced9 | 2357 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2358 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2359 | if (dest_cpu == smp_processor_id()) |
2360 | goto unlock; | |
38022906 | 2361 | |
8f42ced9 | 2362 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2363 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2364 | |
8f42ced9 PZ |
2365 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2366 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2367 | return; |
2368 | } | |
0017d735 | 2369 | unlock: |
8f42ced9 | 2370 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2371 | } |
dd41f596 | 2372 | |
1da177e4 LT |
2373 | #endif |
2374 | ||
1da177e4 | 2375 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2376 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2377 | |
2378 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2379 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2380 | |
2381 | /* | |
c5f8d995 | 2382 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2383 | * @p in case that task is currently running. |
c5f8d995 HS |
2384 | * |
2385 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2386 | */ |
c5f8d995 HS |
2387 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2388 | { | |
2389 | u64 ns = 0; | |
2390 | ||
2391 | if (task_current(rq, p)) { | |
2392 | update_rq_clock(rq); | |
78becc27 | 2393 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2394 | if ((s64)ns < 0) |
2395 | ns = 0; | |
2396 | } | |
2397 | ||
2398 | return ns; | |
2399 | } | |
2400 | ||
bb34d92f | 2401 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2402 | { |
1da177e4 | 2403 | unsigned long flags; |
41b86e9c | 2404 | struct rq *rq; |
bb34d92f | 2405 | u64 ns = 0; |
48f24c4d | 2406 | |
41b86e9c | 2407 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2408 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2409 | task_rq_unlock(rq, p, &flags); |
1508487e | 2410 | |
c5f8d995 HS |
2411 | return ns; |
2412 | } | |
f06febc9 | 2413 | |
c5f8d995 HS |
2414 | /* |
2415 | * Return accounted runtime for the task. | |
2416 | * In case the task is currently running, return the runtime plus current's | |
2417 | * pending runtime that have not been accounted yet. | |
2418 | */ | |
2419 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2420 | { | |
2421 | unsigned long flags; | |
2422 | struct rq *rq; | |
2423 | u64 ns = 0; | |
2424 | ||
911b2898 PZ |
2425 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2426 | /* | |
2427 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2428 | * So we have a optimization chance when the task's delta_exec is 0. | |
2429 | * Reading ->on_cpu is racy, but this is ok. | |
2430 | * | |
2431 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2432 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2433 | * indistinguishable from the read occurring a few cycles earlier. | |
2434 | */ | |
2435 | if (!p->on_cpu) | |
2436 | return p->se.sum_exec_runtime; | |
2437 | #endif | |
2438 | ||
c5f8d995 HS |
2439 | rq = task_rq_lock(p, &flags); |
2440 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2441 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2442 | |
2443 | return ns; | |
2444 | } | |
48f24c4d | 2445 | |
7835b98b CL |
2446 | /* |
2447 | * This function gets called by the timer code, with HZ frequency. | |
2448 | * We call it with interrupts disabled. | |
7835b98b CL |
2449 | */ |
2450 | void scheduler_tick(void) | |
2451 | { | |
7835b98b CL |
2452 | int cpu = smp_processor_id(); |
2453 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2454 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2455 | |
2456 | sched_clock_tick(); | |
dd41f596 | 2457 | |
05fa785c | 2458 | raw_spin_lock(&rq->lock); |
3e51f33f | 2459 | update_rq_clock(rq); |
fa85ae24 | 2460 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2461 | update_cpu_load_active(rq); |
05fa785c | 2462 | raw_spin_unlock(&rq->lock); |
7835b98b | 2463 | |
e9d2b064 | 2464 | perf_event_task_tick(); |
e220d2dc | 2465 | |
e418e1c2 | 2466 | #ifdef CONFIG_SMP |
6eb57e0d | 2467 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2468 | trigger_load_balance(rq); |
e418e1c2 | 2469 | #endif |
265f22a9 | 2470 | rq_last_tick_reset(rq); |
1da177e4 LT |
2471 | } |
2472 | ||
265f22a9 FW |
2473 | #ifdef CONFIG_NO_HZ_FULL |
2474 | /** | |
2475 | * scheduler_tick_max_deferment | |
2476 | * | |
2477 | * Keep at least one tick per second when a single | |
2478 | * active task is running because the scheduler doesn't | |
2479 | * yet completely support full dynticks environment. | |
2480 | * | |
2481 | * This makes sure that uptime, CFS vruntime, load | |
2482 | * balancing, etc... continue to move forward, even | |
2483 | * with a very low granularity. | |
e69f6186 YB |
2484 | * |
2485 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2486 | */ |
2487 | u64 scheduler_tick_max_deferment(void) | |
2488 | { | |
2489 | struct rq *rq = this_rq(); | |
2490 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2491 | ||
2492 | next = rq->last_sched_tick + HZ; | |
2493 | ||
2494 | if (time_before_eq(next, now)) | |
2495 | return 0; | |
2496 | ||
8fe8ff09 | 2497 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2498 | } |
265f22a9 | 2499 | #endif |
1da177e4 | 2500 | |
132380a0 | 2501 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2502 | { |
2503 | if (in_lock_functions(addr)) { | |
2504 | addr = CALLER_ADDR2; | |
2505 | if (in_lock_functions(addr)) | |
2506 | addr = CALLER_ADDR3; | |
2507 | } | |
2508 | return addr; | |
2509 | } | |
1da177e4 | 2510 | |
7e49fcce SR |
2511 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2512 | defined(CONFIG_PREEMPT_TRACER)) | |
2513 | ||
bdb43806 | 2514 | void __kprobes preempt_count_add(int val) |
1da177e4 | 2515 | { |
6cd8a4bb | 2516 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2517 | /* |
2518 | * Underflow? | |
2519 | */ | |
9a11b49a IM |
2520 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2521 | return; | |
6cd8a4bb | 2522 | #endif |
bdb43806 | 2523 | __preempt_count_add(val); |
6cd8a4bb | 2524 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2525 | /* |
2526 | * Spinlock count overflowing soon? | |
2527 | */ | |
33859f7f MOS |
2528 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2529 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2530 | #endif |
8f47b187 TG |
2531 | if (preempt_count() == val) { |
2532 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2533 | #ifdef CONFIG_DEBUG_PREEMPT | |
2534 | current->preempt_disable_ip = ip; | |
2535 | #endif | |
2536 | trace_preempt_off(CALLER_ADDR0, ip); | |
2537 | } | |
1da177e4 | 2538 | } |
bdb43806 | 2539 | EXPORT_SYMBOL(preempt_count_add); |
1da177e4 | 2540 | |
bdb43806 | 2541 | void __kprobes preempt_count_sub(int val) |
1da177e4 | 2542 | { |
6cd8a4bb | 2543 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2544 | /* |
2545 | * Underflow? | |
2546 | */ | |
01e3eb82 | 2547 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2548 | return; |
1da177e4 LT |
2549 | /* |
2550 | * Is the spinlock portion underflowing? | |
2551 | */ | |
9a11b49a IM |
2552 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2553 | !(preempt_count() & PREEMPT_MASK))) | |
2554 | return; | |
6cd8a4bb | 2555 | #endif |
9a11b49a | 2556 | |
6cd8a4bb SR |
2557 | if (preempt_count() == val) |
2558 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2559 | __preempt_count_sub(val); |
1da177e4 | 2560 | } |
bdb43806 | 2561 | EXPORT_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2562 | |
2563 | #endif | |
2564 | ||
2565 | /* | |
dd41f596 | 2566 | * Print scheduling while atomic bug: |
1da177e4 | 2567 | */ |
dd41f596 | 2568 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2569 | { |
664dfa65 DJ |
2570 | if (oops_in_progress) |
2571 | return; | |
2572 | ||
3df0fc5b PZ |
2573 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2574 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2575 | |
dd41f596 | 2576 | debug_show_held_locks(prev); |
e21f5b15 | 2577 | print_modules(); |
dd41f596 IM |
2578 | if (irqs_disabled()) |
2579 | print_irqtrace_events(prev); | |
8f47b187 TG |
2580 | #ifdef CONFIG_DEBUG_PREEMPT |
2581 | if (in_atomic_preempt_off()) { | |
2582 | pr_err("Preemption disabled at:"); | |
2583 | print_ip_sym(current->preempt_disable_ip); | |
2584 | pr_cont("\n"); | |
2585 | } | |
2586 | #endif | |
6135fc1e | 2587 | dump_stack(); |
373d4d09 | 2588 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2589 | } |
1da177e4 | 2590 | |
dd41f596 IM |
2591 | /* |
2592 | * Various schedule()-time debugging checks and statistics: | |
2593 | */ | |
2594 | static inline void schedule_debug(struct task_struct *prev) | |
2595 | { | |
1da177e4 | 2596 | /* |
41a2d6cf | 2597 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2598 | * schedule() atomically, we ignore that path. Otherwise whine |
2599 | * if we are scheduling when we should not. | |
1da177e4 | 2600 | */ |
192301e7 | 2601 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2602 | __schedule_bug(prev); |
b3fbab05 | 2603 | rcu_sleep_check(); |
dd41f596 | 2604 | |
1da177e4 LT |
2605 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2606 | ||
2d72376b | 2607 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2608 | } |
2609 | ||
2610 | /* | |
2611 | * Pick up the highest-prio task: | |
2612 | */ | |
2613 | static inline struct task_struct * | |
606dba2e | 2614 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2615 | { |
37e117c0 | 2616 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2617 | struct task_struct *p; |
1da177e4 LT |
2618 | |
2619 | /* | |
dd41f596 IM |
2620 | * Optimization: we know that if all tasks are in |
2621 | * the fair class we can call that function directly: | |
1da177e4 | 2622 | */ |
37e117c0 | 2623 | if (likely(prev->sched_class == class && |
38033c37 | 2624 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2625 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2626 | if (unlikely(p == RETRY_TASK)) |
2627 | goto again; | |
2628 | ||
2629 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2630 | if (unlikely(!p)) | |
2631 | p = idle_sched_class.pick_next_task(rq, prev); | |
2632 | ||
2633 | return p; | |
1da177e4 LT |
2634 | } |
2635 | ||
37e117c0 | 2636 | again: |
34f971f6 | 2637 | for_each_class(class) { |
606dba2e | 2638 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2639 | if (p) { |
2640 | if (unlikely(p == RETRY_TASK)) | |
2641 | goto again; | |
dd41f596 | 2642 | return p; |
37e117c0 | 2643 | } |
dd41f596 | 2644 | } |
34f971f6 PZ |
2645 | |
2646 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2647 | } |
1da177e4 | 2648 | |
dd41f596 | 2649 | /* |
c259e01a | 2650 | * __schedule() is the main scheduler function. |
edde96ea PE |
2651 | * |
2652 | * The main means of driving the scheduler and thus entering this function are: | |
2653 | * | |
2654 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2655 | * | |
2656 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2657 | * paths. For example, see arch/x86/entry_64.S. | |
2658 | * | |
2659 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2660 | * interrupt handler scheduler_tick(). | |
2661 | * | |
2662 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2663 | * task to the run-queue and that's it. | |
2664 | * | |
2665 | * Now, if the new task added to the run-queue preempts the current | |
2666 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2667 | * called on the nearest possible occasion: | |
2668 | * | |
2669 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2670 | * | |
2671 | * - in syscall or exception context, at the next outmost | |
2672 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2673 | * spin_unlock()!) | |
2674 | * | |
2675 | * - in IRQ context, return from interrupt-handler to | |
2676 | * preemptible context | |
2677 | * | |
2678 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2679 | * then at the next: | |
2680 | * | |
2681 | * - cond_resched() call | |
2682 | * - explicit schedule() call | |
2683 | * - return from syscall or exception to user-space | |
2684 | * - return from interrupt-handler to user-space | |
dd41f596 | 2685 | */ |
c259e01a | 2686 | static void __sched __schedule(void) |
dd41f596 IM |
2687 | { |
2688 | struct task_struct *prev, *next; | |
67ca7bde | 2689 | unsigned long *switch_count; |
dd41f596 | 2690 | struct rq *rq; |
31656519 | 2691 | int cpu; |
dd41f596 | 2692 | |
ff743345 PZ |
2693 | need_resched: |
2694 | preempt_disable(); | |
dd41f596 IM |
2695 | cpu = smp_processor_id(); |
2696 | rq = cpu_rq(cpu); | |
25502a6c | 2697 | rcu_note_context_switch(cpu); |
dd41f596 | 2698 | prev = rq->curr; |
dd41f596 | 2699 | |
dd41f596 | 2700 | schedule_debug(prev); |
1da177e4 | 2701 | |
31656519 | 2702 | if (sched_feat(HRTICK)) |
f333fdc9 | 2703 | hrtick_clear(rq); |
8f4d37ec | 2704 | |
e0acd0a6 ON |
2705 | /* |
2706 | * Make sure that signal_pending_state()->signal_pending() below | |
2707 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2708 | * done by the caller to avoid the race with signal_wake_up(). | |
2709 | */ | |
2710 | smp_mb__before_spinlock(); | |
05fa785c | 2711 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2712 | |
246d86b5 | 2713 | switch_count = &prev->nivcsw; |
1da177e4 | 2714 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2715 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2716 | prev->state = TASK_RUNNING; |
21aa9af0 | 2717 | } else { |
2acca55e PZ |
2718 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2719 | prev->on_rq = 0; | |
2720 | ||
21aa9af0 | 2721 | /* |
2acca55e PZ |
2722 | * If a worker went to sleep, notify and ask workqueue |
2723 | * whether it wants to wake up a task to maintain | |
2724 | * concurrency. | |
21aa9af0 TH |
2725 | */ |
2726 | if (prev->flags & PF_WQ_WORKER) { | |
2727 | struct task_struct *to_wakeup; | |
2728 | ||
2729 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2730 | if (to_wakeup) | |
2731 | try_to_wake_up_local(to_wakeup); | |
2732 | } | |
21aa9af0 | 2733 | } |
dd41f596 | 2734 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2735 | } |
2736 | ||
606dba2e PZ |
2737 | if (prev->on_rq || rq->skip_clock_update < 0) |
2738 | update_rq_clock(rq); | |
2739 | ||
2740 | next = pick_next_task(rq, prev); | |
f26f9aff | 2741 | clear_tsk_need_resched(prev); |
f27dde8d | 2742 | clear_preempt_need_resched(); |
f26f9aff | 2743 | rq->skip_clock_update = 0; |
1da177e4 | 2744 | |
1da177e4 | 2745 | if (likely(prev != next)) { |
1da177e4 LT |
2746 | rq->nr_switches++; |
2747 | rq->curr = next; | |
2748 | ++*switch_count; | |
2749 | ||
dd41f596 | 2750 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2751 | /* |
246d86b5 ON |
2752 | * The context switch have flipped the stack from under us |
2753 | * and restored the local variables which were saved when | |
2754 | * this task called schedule() in the past. prev == current | |
2755 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2756 | */ |
2757 | cpu = smp_processor_id(); | |
2758 | rq = cpu_rq(cpu); | |
1da177e4 | 2759 | } else |
05fa785c | 2760 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2761 | |
3f029d3c | 2762 | post_schedule(rq); |
1da177e4 | 2763 | |
ba74c144 | 2764 | sched_preempt_enable_no_resched(); |
ff743345 | 2765 | if (need_resched()) |
1da177e4 LT |
2766 | goto need_resched; |
2767 | } | |
c259e01a | 2768 | |
9c40cef2 TG |
2769 | static inline void sched_submit_work(struct task_struct *tsk) |
2770 | { | |
3c7d5184 | 2771 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2772 | return; |
2773 | /* | |
2774 | * If we are going to sleep and we have plugged IO queued, | |
2775 | * make sure to submit it to avoid deadlocks. | |
2776 | */ | |
2777 | if (blk_needs_flush_plug(tsk)) | |
2778 | blk_schedule_flush_plug(tsk); | |
2779 | } | |
2780 | ||
722a9f92 | 2781 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2782 | { |
9c40cef2 TG |
2783 | struct task_struct *tsk = current; |
2784 | ||
2785 | sched_submit_work(tsk); | |
c259e01a TG |
2786 | __schedule(); |
2787 | } | |
1da177e4 LT |
2788 | EXPORT_SYMBOL(schedule); |
2789 | ||
91d1aa43 | 2790 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2791 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2792 | { |
2793 | /* | |
2794 | * If we come here after a random call to set_need_resched(), | |
2795 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2796 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2797 | * we find a better solution. | |
2798 | */ | |
91d1aa43 | 2799 | user_exit(); |
20ab65e3 | 2800 | schedule(); |
91d1aa43 | 2801 | user_enter(); |
20ab65e3 FW |
2802 | } |
2803 | #endif | |
2804 | ||
c5491ea7 TG |
2805 | /** |
2806 | * schedule_preempt_disabled - called with preemption disabled | |
2807 | * | |
2808 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2809 | */ | |
2810 | void __sched schedule_preempt_disabled(void) | |
2811 | { | |
ba74c144 | 2812 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2813 | schedule(); |
2814 | preempt_disable(); | |
2815 | } | |
2816 | ||
1da177e4 LT |
2817 | #ifdef CONFIG_PREEMPT |
2818 | /* | |
2ed6e34f | 2819 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2820 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2821 | * occur there and call schedule directly. |
2822 | */ | |
722a9f92 | 2823 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2824 | { |
1da177e4 LT |
2825 | /* |
2826 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2827 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2828 | */ |
fbb00b56 | 2829 | if (likely(!preemptible())) |
1da177e4 LT |
2830 | return; |
2831 | ||
3a5c359a | 2832 | do { |
bdb43806 | 2833 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2834 | __schedule(); |
bdb43806 | 2835 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2836 | |
3a5c359a AK |
2837 | /* |
2838 | * Check again in case we missed a preemption opportunity | |
2839 | * between schedule and now. | |
2840 | */ | |
2841 | barrier(); | |
5ed0cec0 | 2842 | } while (need_resched()); |
1da177e4 | 2843 | } |
1da177e4 | 2844 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2845 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2846 | |
2847 | /* | |
2ed6e34f | 2848 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2849 | * off of irq context. |
2850 | * Note, that this is called and return with irqs disabled. This will | |
2851 | * protect us against recursive calling from irq. | |
2852 | */ | |
722a9f92 | 2853 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2854 | { |
b22366cd | 2855 | enum ctx_state prev_state; |
6478d880 | 2856 | |
2ed6e34f | 2857 | /* Catch callers which need to be fixed */ |
f27dde8d | 2858 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2859 | |
b22366cd FW |
2860 | prev_state = exception_enter(); |
2861 | ||
3a5c359a | 2862 | do { |
bdb43806 | 2863 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2864 | local_irq_enable(); |
c259e01a | 2865 | __schedule(); |
3a5c359a | 2866 | local_irq_disable(); |
bdb43806 | 2867 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2868 | |
3a5c359a AK |
2869 | /* |
2870 | * Check again in case we missed a preemption opportunity | |
2871 | * between schedule and now. | |
2872 | */ | |
2873 | barrier(); | |
5ed0cec0 | 2874 | } while (need_resched()); |
b22366cd FW |
2875 | |
2876 | exception_exit(prev_state); | |
1da177e4 LT |
2877 | } |
2878 | ||
63859d4f | 2879 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2880 | void *key) |
1da177e4 | 2881 | { |
63859d4f | 2882 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2883 | } |
1da177e4 LT |
2884 | EXPORT_SYMBOL(default_wake_function); |
2885 | ||
b29739f9 IM |
2886 | #ifdef CONFIG_RT_MUTEXES |
2887 | ||
2888 | /* | |
2889 | * rt_mutex_setprio - set the current priority of a task | |
2890 | * @p: task | |
2891 | * @prio: prio value (kernel-internal form) | |
2892 | * | |
2893 | * This function changes the 'effective' priority of a task. It does | |
2894 | * not touch ->normal_prio like __setscheduler(). | |
2895 | * | |
c365c292 TG |
2896 | * Used by the rt_mutex code to implement priority inheritance |
2897 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2898 | */ |
36c8b586 | 2899 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2900 | { |
2d3d891d | 2901 | int oldprio, on_rq, running, enqueue_flag = 0; |
70b97a7f | 2902 | struct rq *rq; |
83ab0aa0 | 2903 | const struct sched_class *prev_class; |
b29739f9 | 2904 | |
aab03e05 | 2905 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2906 | |
0122ec5b | 2907 | rq = __task_rq_lock(p); |
b29739f9 | 2908 | |
1c4dd99b TG |
2909 | /* |
2910 | * Idle task boosting is a nono in general. There is one | |
2911 | * exception, when PREEMPT_RT and NOHZ is active: | |
2912 | * | |
2913 | * The idle task calls get_next_timer_interrupt() and holds | |
2914 | * the timer wheel base->lock on the CPU and another CPU wants | |
2915 | * to access the timer (probably to cancel it). We can safely | |
2916 | * ignore the boosting request, as the idle CPU runs this code | |
2917 | * with interrupts disabled and will complete the lock | |
2918 | * protected section without being interrupted. So there is no | |
2919 | * real need to boost. | |
2920 | */ | |
2921 | if (unlikely(p == rq->idle)) { | |
2922 | WARN_ON(p != rq->curr); | |
2923 | WARN_ON(p->pi_blocked_on); | |
2924 | goto out_unlock; | |
2925 | } | |
2926 | ||
a8027073 | 2927 | trace_sched_pi_setprio(p, prio); |
2d3d891d | 2928 | p->pi_top_task = rt_mutex_get_top_task(p); |
d5f9f942 | 2929 | oldprio = p->prio; |
83ab0aa0 | 2930 | prev_class = p->sched_class; |
fd2f4419 | 2931 | on_rq = p->on_rq; |
051a1d1a | 2932 | running = task_current(rq, p); |
0e1f3483 | 2933 | if (on_rq) |
69be72c1 | 2934 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2935 | if (running) |
2936 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2937 | |
2d3d891d DF |
2938 | /* |
2939 | * Boosting condition are: | |
2940 | * 1. -rt task is running and holds mutex A | |
2941 | * --> -dl task blocks on mutex A | |
2942 | * | |
2943 | * 2. -dl task is running and holds mutex A | |
2944 | * --> -dl task blocks on mutex A and could preempt the | |
2945 | * running task | |
2946 | */ | |
2947 | if (dl_prio(prio)) { | |
2948 | if (!dl_prio(p->normal_prio) || (p->pi_top_task && | |
2949 | dl_entity_preempt(&p->pi_top_task->dl, &p->dl))) { | |
2950 | p->dl.dl_boosted = 1; | |
2951 | p->dl.dl_throttled = 0; | |
2952 | enqueue_flag = ENQUEUE_REPLENISH; | |
2953 | } else | |
2954 | p->dl.dl_boosted = 0; | |
aab03e05 | 2955 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
2956 | } else if (rt_prio(prio)) { |
2957 | if (dl_prio(oldprio)) | |
2958 | p->dl.dl_boosted = 0; | |
2959 | if (oldprio < prio) | |
2960 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 2961 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
2962 | } else { |
2963 | if (dl_prio(oldprio)) | |
2964 | p->dl.dl_boosted = 0; | |
dd41f596 | 2965 | p->sched_class = &fair_sched_class; |
2d3d891d | 2966 | } |
dd41f596 | 2967 | |
b29739f9 IM |
2968 | p->prio = prio; |
2969 | ||
0e1f3483 HS |
2970 | if (running) |
2971 | p->sched_class->set_curr_task(rq); | |
da7a735e | 2972 | if (on_rq) |
2d3d891d | 2973 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 2974 | |
da7a735e | 2975 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 2976 | out_unlock: |
0122ec5b | 2977 | __task_rq_unlock(rq); |
b29739f9 | 2978 | } |
b29739f9 | 2979 | #endif |
d50dde5a | 2980 | |
36c8b586 | 2981 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 2982 | { |
dd41f596 | 2983 | int old_prio, delta, on_rq; |
1da177e4 | 2984 | unsigned long flags; |
70b97a7f | 2985 | struct rq *rq; |
1da177e4 | 2986 | |
75e45d51 | 2987 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
2988 | return; |
2989 | /* | |
2990 | * We have to be careful, if called from sys_setpriority(), | |
2991 | * the task might be in the middle of scheduling on another CPU. | |
2992 | */ | |
2993 | rq = task_rq_lock(p, &flags); | |
2994 | /* | |
2995 | * The RT priorities are set via sched_setscheduler(), but we still | |
2996 | * allow the 'normal' nice value to be set - but as expected | |
2997 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 2998 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 2999 | */ |
aab03e05 | 3000 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3001 | p->static_prio = NICE_TO_PRIO(nice); |
3002 | goto out_unlock; | |
3003 | } | |
fd2f4419 | 3004 | on_rq = p->on_rq; |
c09595f6 | 3005 | if (on_rq) |
69be72c1 | 3006 | dequeue_task(rq, p, 0); |
1da177e4 | 3007 | |
1da177e4 | 3008 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3009 | set_load_weight(p); |
b29739f9 IM |
3010 | old_prio = p->prio; |
3011 | p->prio = effective_prio(p); | |
3012 | delta = p->prio - old_prio; | |
1da177e4 | 3013 | |
dd41f596 | 3014 | if (on_rq) { |
371fd7e7 | 3015 | enqueue_task(rq, p, 0); |
1da177e4 | 3016 | /* |
d5f9f942 AM |
3017 | * If the task increased its priority or is running and |
3018 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3019 | */ |
d5f9f942 | 3020 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3021 | resched_task(rq->curr); |
3022 | } | |
3023 | out_unlock: | |
0122ec5b | 3024 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3025 | } |
1da177e4 LT |
3026 | EXPORT_SYMBOL(set_user_nice); |
3027 | ||
e43379f1 MM |
3028 | /* |
3029 | * can_nice - check if a task can reduce its nice value | |
3030 | * @p: task | |
3031 | * @nice: nice value | |
3032 | */ | |
36c8b586 | 3033 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3034 | { |
024f4747 MM |
3035 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3036 | int nice_rlim = 20 - nice; | |
48f24c4d | 3037 | |
78d7d407 | 3038 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3039 | capable(CAP_SYS_NICE)); |
3040 | } | |
3041 | ||
1da177e4 LT |
3042 | #ifdef __ARCH_WANT_SYS_NICE |
3043 | ||
3044 | /* | |
3045 | * sys_nice - change the priority of the current process. | |
3046 | * @increment: priority increment | |
3047 | * | |
3048 | * sys_setpriority is a more generic, but much slower function that | |
3049 | * does similar things. | |
3050 | */ | |
5add95d4 | 3051 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3052 | { |
48f24c4d | 3053 | long nice, retval; |
1da177e4 LT |
3054 | |
3055 | /* | |
3056 | * Setpriority might change our priority at the same moment. | |
3057 | * We don't have to worry. Conceptually one call occurs first | |
3058 | * and we have a single winner. | |
3059 | */ | |
e43379f1 MM |
3060 | if (increment < -40) |
3061 | increment = -40; | |
1da177e4 LT |
3062 | if (increment > 40) |
3063 | increment = 40; | |
3064 | ||
d0ea0268 | 3065 | nice = task_nice(current) + increment; |
75e45d51 DY |
3066 | if (nice < MIN_NICE) |
3067 | nice = MIN_NICE; | |
3068 | if (nice > MAX_NICE) | |
3069 | nice = MAX_NICE; | |
1da177e4 | 3070 | |
e43379f1 MM |
3071 | if (increment < 0 && !can_nice(current, nice)) |
3072 | return -EPERM; | |
3073 | ||
1da177e4 LT |
3074 | retval = security_task_setnice(current, nice); |
3075 | if (retval) | |
3076 | return retval; | |
3077 | ||
3078 | set_user_nice(current, nice); | |
3079 | return 0; | |
3080 | } | |
3081 | ||
3082 | #endif | |
3083 | ||
3084 | /** | |
3085 | * task_prio - return the priority value of a given task. | |
3086 | * @p: the task in question. | |
3087 | * | |
e69f6186 | 3088 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3089 | * RT tasks are offset by -200. Normal tasks are centered |
3090 | * around 0, value goes from -16 to +15. | |
3091 | */ | |
36c8b586 | 3092 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3093 | { |
3094 | return p->prio - MAX_RT_PRIO; | |
3095 | } | |
3096 | ||
1da177e4 LT |
3097 | /** |
3098 | * idle_cpu - is a given cpu idle currently? | |
3099 | * @cpu: the processor in question. | |
e69f6186 YB |
3100 | * |
3101 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3102 | */ |
3103 | int idle_cpu(int cpu) | |
3104 | { | |
908a3283 TG |
3105 | struct rq *rq = cpu_rq(cpu); |
3106 | ||
3107 | if (rq->curr != rq->idle) | |
3108 | return 0; | |
3109 | ||
3110 | if (rq->nr_running) | |
3111 | return 0; | |
3112 | ||
3113 | #ifdef CONFIG_SMP | |
3114 | if (!llist_empty(&rq->wake_list)) | |
3115 | return 0; | |
3116 | #endif | |
3117 | ||
3118 | return 1; | |
1da177e4 LT |
3119 | } |
3120 | ||
1da177e4 LT |
3121 | /** |
3122 | * idle_task - return the idle task for a given cpu. | |
3123 | * @cpu: the processor in question. | |
e69f6186 YB |
3124 | * |
3125 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3126 | */ |
36c8b586 | 3127 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3128 | { |
3129 | return cpu_rq(cpu)->idle; | |
3130 | } | |
3131 | ||
3132 | /** | |
3133 | * find_process_by_pid - find a process with a matching PID value. | |
3134 | * @pid: the pid in question. | |
e69f6186 YB |
3135 | * |
3136 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3137 | */ |
a9957449 | 3138 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3139 | { |
228ebcbe | 3140 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3141 | } |
3142 | ||
aab03e05 DF |
3143 | /* |
3144 | * This function initializes the sched_dl_entity of a newly becoming | |
3145 | * SCHED_DEADLINE task. | |
3146 | * | |
3147 | * Only the static values are considered here, the actual runtime and the | |
3148 | * absolute deadline will be properly calculated when the task is enqueued | |
3149 | * for the first time with its new policy. | |
3150 | */ | |
3151 | static void | |
3152 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3153 | { | |
3154 | struct sched_dl_entity *dl_se = &p->dl; | |
3155 | ||
3156 | init_dl_task_timer(dl_se); | |
3157 | dl_se->dl_runtime = attr->sched_runtime; | |
3158 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3159 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3160 | dl_se->flags = attr->sched_flags; |
332ac17e | 3161 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3162 | dl_se->dl_throttled = 0; |
3163 | dl_se->dl_new = 1; | |
5bfd126e | 3164 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3165 | } |
3166 | ||
c365c292 TG |
3167 | static void __setscheduler_params(struct task_struct *p, |
3168 | const struct sched_attr *attr) | |
1da177e4 | 3169 | { |
d50dde5a DF |
3170 | int policy = attr->sched_policy; |
3171 | ||
39fd8fd2 PZ |
3172 | if (policy == -1) /* setparam */ |
3173 | policy = p->policy; | |
3174 | ||
1da177e4 | 3175 | p->policy = policy; |
d50dde5a | 3176 | |
aab03e05 DF |
3177 | if (dl_policy(policy)) |
3178 | __setparam_dl(p, attr); | |
39fd8fd2 | 3179 | else if (fair_policy(policy)) |
d50dde5a DF |
3180 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3181 | ||
39fd8fd2 PZ |
3182 | /* |
3183 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3184 | * !rt_policy. Always setting this ensures that things like | |
3185 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3186 | */ | |
3187 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3188 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3189 | set_load_weight(p); |
3190 | } | |
39fd8fd2 | 3191 | |
c365c292 TG |
3192 | /* Actually do priority change: must hold pi & rq lock. */ |
3193 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3194 | const struct sched_attr *attr) | |
3195 | { | |
3196 | __setscheduler_params(p, attr); | |
d50dde5a | 3197 | |
383afd09 SR |
3198 | /* |
3199 | * If we get here, there was no pi waiters boosting the | |
3200 | * task. It is safe to use the normal prio. | |
3201 | */ | |
3202 | p->prio = normal_prio(p); | |
3203 | ||
aab03e05 DF |
3204 | if (dl_prio(p->prio)) |
3205 | p->sched_class = &dl_sched_class; | |
3206 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3207 | p->sched_class = &rt_sched_class; |
3208 | else | |
3209 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3210 | } |
aab03e05 DF |
3211 | |
3212 | static void | |
3213 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3214 | { | |
3215 | struct sched_dl_entity *dl_se = &p->dl; | |
3216 | ||
3217 | attr->sched_priority = p->rt_priority; | |
3218 | attr->sched_runtime = dl_se->dl_runtime; | |
3219 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3220 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3221 | attr->sched_flags = dl_se->flags; |
3222 | } | |
3223 | ||
3224 | /* | |
3225 | * This function validates the new parameters of a -deadline task. | |
3226 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3227 | * than the runtime, as well as the period of being zero or |
332ac17e | 3228 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3229 | * user parameters are above the internal resolution of 1us (we |
3230 | * check sched_runtime only since it is always the smaller one) and | |
3231 | * below 2^63 ns (we have to check both sched_deadline and | |
3232 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3233 | */ |
3234 | static bool | |
3235 | __checkparam_dl(const struct sched_attr *attr) | |
3236 | { | |
b0827819 JL |
3237 | /* deadline != 0 */ |
3238 | if (attr->sched_deadline == 0) | |
3239 | return false; | |
3240 | ||
3241 | /* | |
3242 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3243 | * that big. | |
3244 | */ | |
3245 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3246 | return false; | |
3247 | ||
3248 | /* | |
3249 | * Since we use the MSB for wrap-around and sign issues, make | |
3250 | * sure it's not set (mind that period can be equal to zero). | |
3251 | */ | |
3252 | if (attr->sched_deadline & (1ULL << 63) || | |
3253 | attr->sched_period & (1ULL << 63)) | |
3254 | return false; | |
3255 | ||
3256 | /* runtime <= deadline <= period (if period != 0) */ | |
3257 | if ((attr->sched_period != 0 && | |
3258 | attr->sched_period < attr->sched_deadline) || | |
3259 | attr->sched_deadline < attr->sched_runtime) | |
3260 | return false; | |
3261 | ||
3262 | return true; | |
aab03e05 DF |
3263 | } |
3264 | ||
c69e8d9c DH |
3265 | /* |
3266 | * check the target process has a UID that matches the current process's | |
3267 | */ | |
3268 | static bool check_same_owner(struct task_struct *p) | |
3269 | { | |
3270 | const struct cred *cred = current_cred(), *pcred; | |
3271 | bool match; | |
3272 | ||
3273 | rcu_read_lock(); | |
3274 | pcred = __task_cred(p); | |
9c806aa0 EB |
3275 | match = (uid_eq(cred->euid, pcred->euid) || |
3276 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3277 | rcu_read_unlock(); |
3278 | return match; | |
3279 | } | |
3280 | ||
d50dde5a DF |
3281 | static int __sched_setscheduler(struct task_struct *p, |
3282 | const struct sched_attr *attr, | |
3283 | bool user) | |
1da177e4 | 3284 | { |
383afd09 SR |
3285 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3286 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
83b699ed | 3287 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3288 | int policy = attr->sched_policy; |
1da177e4 | 3289 | unsigned long flags; |
83ab0aa0 | 3290 | const struct sched_class *prev_class; |
70b97a7f | 3291 | struct rq *rq; |
ca94c442 | 3292 | int reset_on_fork; |
1da177e4 | 3293 | |
66e5393a SR |
3294 | /* may grab non-irq protected spin_locks */ |
3295 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3296 | recheck: |
3297 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3298 | if (policy < 0) { |
3299 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3300 | policy = oldpolicy = p->policy; |
ca94c442 | 3301 | } else { |
7479f3c9 | 3302 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3303 | |
aab03e05 DF |
3304 | if (policy != SCHED_DEADLINE && |
3305 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3306 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3307 | policy != SCHED_IDLE) | |
3308 | return -EINVAL; | |
3309 | } | |
3310 | ||
7479f3c9 PZ |
3311 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3312 | return -EINVAL; | |
3313 | ||
1da177e4 LT |
3314 | /* |
3315 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3316 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3317 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3318 | */ |
0bb040a4 | 3319 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3320 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3321 | return -EINVAL; |
aab03e05 DF |
3322 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3323 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3324 | return -EINVAL; |
3325 | ||
37e4ab3f OC |
3326 | /* |
3327 | * Allow unprivileged RT tasks to decrease priority: | |
3328 | */ | |
961ccddd | 3329 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3330 | if (fair_policy(policy)) { |
d0ea0268 | 3331 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3332 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3333 | return -EPERM; |
3334 | } | |
3335 | ||
e05606d3 | 3336 | if (rt_policy(policy)) { |
a44702e8 ON |
3337 | unsigned long rlim_rtprio = |
3338 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3339 | |
3340 | /* can't set/change the rt policy */ | |
3341 | if (policy != p->policy && !rlim_rtprio) | |
3342 | return -EPERM; | |
3343 | ||
3344 | /* can't increase priority */ | |
d50dde5a DF |
3345 | if (attr->sched_priority > p->rt_priority && |
3346 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3347 | return -EPERM; |
3348 | } | |
c02aa73b | 3349 | |
d44753b8 JL |
3350 | /* |
3351 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3352 | * (safest behavior); in the future we would like to allow | |
3353 | * unprivileged DL tasks to increase their relative deadline | |
3354 | * or reduce their runtime (both ways reducing utilization) | |
3355 | */ | |
3356 | if (dl_policy(policy)) | |
3357 | return -EPERM; | |
3358 | ||
dd41f596 | 3359 | /* |
c02aa73b DH |
3360 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3361 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3362 | */ |
c02aa73b | 3363 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3364 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3365 | return -EPERM; |
3366 | } | |
5fe1d75f | 3367 | |
37e4ab3f | 3368 | /* can't change other user's priorities */ |
c69e8d9c | 3369 | if (!check_same_owner(p)) |
37e4ab3f | 3370 | return -EPERM; |
ca94c442 LP |
3371 | |
3372 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3373 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3374 | return -EPERM; | |
37e4ab3f | 3375 | } |
1da177e4 | 3376 | |
725aad24 | 3377 | if (user) { |
b0ae1981 | 3378 | retval = security_task_setscheduler(p); |
725aad24 JF |
3379 | if (retval) |
3380 | return retval; | |
3381 | } | |
3382 | ||
b29739f9 IM |
3383 | /* |
3384 | * make sure no PI-waiters arrive (or leave) while we are | |
3385 | * changing the priority of the task: | |
0122ec5b | 3386 | * |
25985edc | 3387 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3388 | * runqueue lock must be held. |
3389 | */ | |
0122ec5b | 3390 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3391 | |
34f971f6 PZ |
3392 | /* |
3393 | * Changing the policy of the stop threads its a very bad idea | |
3394 | */ | |
3395 | if (p == rq->stop) { | |
0122ec5b | 3396 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3397 | return -EINVAL; |
3398 | } | |
3399 | ||
a51e9198 | 3400 | /* |
d6b1e911 TG |
3401 | * If not changing anything there's no need to proceed further, |
3402 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3403 | */ |
d50dde5a | 3404 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3405 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3406 | goto change; |
3407 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3408 | goto change; | |
aab03e05 DF |
3409 | if (dl_policy(policy)) |
3410 | goto change; | |
d50dde5a | 3411 | |
d6b1e911 | 3412 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3413 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3414 | return 0; |
3415 | } | |
d50dde5a | 3416 | change: |
a51e9198 | 3417 | |
dc61b1d6 | 3418 | if (user) { |
332ac17e | 3419 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3420 | /* |
3421 | * Do not allow realtime tasks into groups that have no runtime | |
3422 | * assigned. | |
3423 | */ | |
3424 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3425 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3426 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3427 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3428 | return -EPERM; |
3429 | } | |
dc61b1d6 | 3430 | #endif |
332ac17e DF |
3431 | #ifdef CONFIG_SMP |
3432 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3433 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3434 | |
3435 | /* | |
3436 | * Don't allow tasks with an affinity mask smaller than | |
3437 | * the entire root_domain to become SCHED_DEADLINE. We | |
3438 | * will also fail if there's no bandwidth available. | |
3439 | */ | |
e4099a5e PZ |
3440 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3441 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3442 | task_rq_unlock(rq, p, &flags); |
3443 | return -EPERM; | |
3444 | } | |
3445 | } | |
3446 | #endif | |
3447 | } | |
dc61b1d6 | 3448 | |
1da177e4 LT |
3449 | /* recheck policy now with rq lock held */ |
3450 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3451 | policy = oldpolicy = -1; | |
0122ec5b | 3452 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3453 | goto recheck; |
3454 | } | |
332ac17e DF |
3455 | |
3456 | /* | |
3457 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3458 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3459 | * is available. | |
3460 | */ | |
e4099a5e | 3461 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3462 | task_rq_unlock(rq, p, &flags); |
3463 | return -EBUSY; | |
3464 | } | |
3465 | ||
c365c292 TG |
3466 | p->sched_reset_on_fork = reset_on_fork; |
3467 | oldprio = p->prio; | |
3468 | ||
3469 | /* | |
3470 | * Special case for priority boosted tasks. | |
3471 | * | |
3472 | * If the new priority is lower or equal (user space view) | |
3473 | * than the current (boosted) priority, we just store the new | |
3474 | * normal parameters and do not touch the scheduler class and | |
3475 | * the runqueue. This will be done when the task deboost | |
3476 | * itself. | |
3477 | */ | |
3478 | if (rt_mutex_check_prio(p, newprio)) { | |
3479 | __setscheduler_params(p, attr); | |
3480 | task_rq_unlock(rq, p, &flags); | |
3481 | return 0; | |
3482 | } | |
3483 | ||
fd2f4419 | 3484 | on_rq = p->on_rq; |
051a1d1a | 3485 | running = task_current(rq, p); |
0e1f3483 | 3486 | if (on_rq) |
4ca9b72b | 3487 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3488 | if (running) |
3489 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3490 | |
83ab0aa0 | 3491 | prev_class = p->sched_class; |
d50dde5a | 3492 | __setscheduler(rq, p, attr); |
f6b53205 | 3493 | |
0e1f3483 HS |
3494 | if (running) |
3495 | p->sched_class->set_curr_task(rq); | |
81a44c54 TG |
3496 | if (on_rq) { |
3497 | /* | |
3498 | * We enqueue to tail when the priority of a task is | |
3499 | * increased (user space view). | |
3500 | */ | |
3501 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3502 | } | |
cb469845 | 3503 | |
da7a735e | 3504 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3505 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3506 | |
95e02ca9 TG |
3507 | rt_mutex_adjust_pi(p); |
3508 | ||
1da177e4 LT |
3509 | return 0; |
3510 | } | |
961ccddd | 3511 | |
7479f3c9 PZ |
3512 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3513 | const struct sched_param *param, bool check) | |
3514 | { | |
3515 | struct sched_attr attr = { | |
3516 | .sched_policy = policy, | |
3517 | .sched_priority = param->sched_priority, | |
3518 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3519 | }; | |
3520 | ||
3521 | /* | |
3522 | * Fixup the legacy SCHED_RESET_ON_FORK hack | |
3523 | */ | |
3524 | if (policy & SCHED_RESET_ON_FORK) { | |
3525 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
3526 | policy &= ~SCHED_RESET_ON_FORK; | |
3527 | attr.sched_policy = policy; | |
3528 | } | |
3529 | ||
3530 | return __sched_setscheduler(p, &attr, check); | |
3531 | } | |
961ccddd RR |
3532 | /** |
3533 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3534 | * @p: the task in question. | |
3535 | * @policy: new policy. | |
3536 | * @param: structure containing the new RT priority. | |
3537 | * | |
e69f6186 YB |
3538 | * Return: 0 on success. An error code otherwise. |
3539 | * | |
961ccddd RR |
3540 | * NOTE that the task may be already dead. |
3541 | */ | |
3542 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3543 | const struct sched_param *param) |
961ccddd | 3544 | { |
7479f3c9 | 3545 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3546 | } |
1da177e4 LT |
3547 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3548 | ||
d50dde5a DF |
3549 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3550 | { | |
3551 | return __sched_setscheduler(p, attr, true); | |
3552 | } | |
3553 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3554 | ||
961ccddd RR |
3555 | /** |
3556 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3557 | * @p: the task in question. | |
3558 | * @policy: new policy. | |
3559 | * @param: structure containing the new RT priority. | |
3560 | * | |
3561 | * Just like sched_setscheduler, only don't bother checking if the | |
3562 | * current context has permission. For example, this is needed in | |
3563 | * stop_machine(): we create temporary high priority worker threads, | |
3564 | * but our caller might not have that capability. | |
e69f6186 YB |
3565 | * |
3566 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3567 | */ |
3568 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3569 | const struct sched_param *param) |
961ccddd | 3570 | { |
7479f3c9 | 3571 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3572 | } |
3573 | ||
95cdf3b7 IM |
3574 | static int |
3575 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3576 | { |
1da177e4 LT |
3577 | struct sched_param lparam; |
3578 | struct task_struct *p; | |
36c8b586 | 3579 | int retval; |
1da177e4 LT |
3580 | |
3581 | if (!param || pid < 0) | |
3582 | return -EINVAL; | |
3583 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3584 | return -EFAULT; | |
5fe1d75f ON |
3585 | |
3586 | rcu_read_lock(); | |
3587 | retval = -ESRCH; | |
1da177e4 | 3588 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3589 | if (p != NULL) |
3590 | retval = sched_setscheduler(p, policy, &lparam); | |
3591 | rcu_read_unlock(); | |
36c8b586 | 3592 | |
1da177e4 LT |
3593 | return retval; |
3594 | } | |
3595 | ||
d50dde5a DF |
3596 | /* |
3597 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3598 | */ | |
3599 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3600 | struct sched_attr *attr) | |
3601 | { | |
3602 | u32 size; | |
3603 | int ret; | |
3604 | ||
3605 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3606 | return -EFAULT; | |
3607 | ||
3608 | /* | |
3609 | * zero the full structure, so that a short copy will be nice. | |
3610 | */ | |
3611 | memset(attr, 0, sizeof(*attr)); | |
3612 | ||
3613 | ret = get_user(size, &uattr->size); | |
3614 | if (ret) | |
3615 | return ret; | |
3616 | ||
3617 | if (size > PAGE_SIZE) /* silly large */ | |
3618 | goto err_size; | |
3619 | ||
3620 | if (!size) /* abi compat */ | |
3621 | size = SCHED_ATTR_SIZE_VER0; | |
3622 | ||
3623 | if (size < SCHED_ATTR_SIZE_VER0) | |
3624 | goto err_size; | |
3625 | ||
3626 | /* | |
3627 | * If we're handed a bigger struct than we know of, | |
3628 | * ensure all the unknown bits are 0 - i.e. new | |
3629 | * user-space does not rely on any kernel feature | |
3630 | * extensions we dont know about yet. | |
3631 | */ | |
3632 | if (size > sizeof(*attr)) { | |
3633 | unsigned char __user *addr; | |
3634 | unsigned char __user *end; | |
3635 | unsigned char val; | |
3636 | ||
3637 | addr = (void __user *)uattr + sizeof(*attr); | |
3638 | end = (void __user *)uattr + size; | |
3639 | ||
3640 | for (; addr < end; addr++) { | |
3641 | ret = get_user(val, addr); | |
3642 | if (ret) | |
3643 | return ret; | |
3644 | if (val) | |
3645 | goto err_size; | |
3646 | } | |
3647 | size = sizeof(*attr); | |
3648 | } | |
3649 | ||
3650 | ret = copy_from_user(attr, uattr, size); | |
3651 | if (ret) | |
3652 | return -EFAULT; | |
3653 | ||
3654 | /* | |
3655 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3656 | * to be strict and return an error on out-of-bounds values? | |
3657 | */ | |
75e45d51 | 3658 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3659 | |
e78c7bca | 3660 | return 0; |
d50dde5a DF |
3661 | |
3662 | err_size: | |
3663 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3664 | return -E2BIG; |
d50dde5a DF |
3665 | } |
3666 | ||
1da177e4 LT |
3667 | /** |
3668 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3669 | * @pid: the pid in question. | |
3670 | * @policy: new policy. | |
3671 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3672 | * |
3673 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3674 | */ |
5add95d4 HC |
3675 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3676 | struct sched_param __user *, param) | |
1da177e4 | 3677 | { |
c21761f1 JB |
3678 | /* negative values for policy are not valid */ |
3679 | if (policy < 0) | |
3680 | return -EINVAL; | |
3681 | ||
1da177e4 LT |
3682 | return do_sched_setscheduler(pid, policy, param); |
3683 | } | |
3684 | ||
3685 | /** | |
3686 | * sys_sched_setparam - set/change the RT priority of a thread | |
3687 | * @pid: the pid in question. | |
3688 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3689 | * |
3690 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3691 | */ |
5add95d4 | 3692 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3693 | { |
3694 | return do_sched_setscheduler(pid, -1, param); | |
3695 | } | |
3696 | ||
d50dde5a DF |
3697 | /** |
3698 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3699 | * @pid: the pid in question. | |
5778fccf | 3700 | * @uattr: structure containing the extended parameters. |
db66d756 | 3701 | * @flags: for future extension. |
d50dde5a | 3702 | */ |
6d35ab48 PZ |
3703 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3704 | unsigned int, flags) | |
d50dde5a DF |
3705 | { |
3706 | struct sched_attr attr; | |
3707 | struct task_struct *p; | |
3708 | int retval; | |
3709 | ||
6d35ab48 | 3710 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3711 | return -EINVAL; |
3712 | ||
143cf23d MK |
3713 | retval = sched_copy_attr(uattr, &attr); |
3714 | if (retval) | |
3715 | return retval; | |
d50dde5a | 3716 | |
dbdb2275 PZ |
3717 | if (attr.sched_policy < 0) |
3718 | return -EINVAL; | |
d50dde5a DF |
3719 | |
3720 | rcu_read_lock(); | |
3721 | retval = -ESRCH; | |
3722 | p = find_process_by_pid(pid); | |
3723 | if (p != NULL) | |
3724 | retval = sched_setattr(p, &attr); | |
3725 | rcu_read_unlock(); | |
3726 | ||
3727 | return retval; | |
3728 | } | |
3729 | ||
1da177e4 LT |
3730 | /** |
3731 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3732 | * @pid: the pid in question. | |
e69f6186 YB |
3733 | * |
3734 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3735 | * code. | |
1da177e4 | 3736 | */ |
5add95d4 | 3737 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3738 | { |
36c8b586 | 3739 | struct task_struct *p; |
3a5c359a | 3740 | int retval; |
1da177e4 LT |
3741 | |
3742 | if (pid < 0) | |
3a5c359a | 3743 | return -EINVAL; |
1da177e4 LT |
3744 | |
3745 | retval = -ESRCH; | |
5fe85be0 | 3746 | rcu_read_lock(); |
1da177e4 LT |
3747 | p = find_process_by_pid(pid); |
3748 | if (p) { | |
3749 | retval = security_task_getscheduler(p); | |
3750 | if (!retval) | |
ca94c442 LP |
3751 | retval = p->policy |
3752 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3753 | } |
5fe85be0 | 3754 | rcu_read_unlock(); |
1da177e4 LT |
3755 | return retval; |
3756 | } | |
3757 | ||
3758 | /** | |
ca94c442 | 3759 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3760 | * @pid: the pid in question. |
3761 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3762 | * |
3763 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3764 | * code. | |
1da177e4 | 3765 | */ |
5add95d4 | 3766 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3767 | { |
ce5f7f82 | 3768 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3769 | struct task_struct *p; |
3a5c359a | 3770 | int retval; |
1da177e4 LT |
3771 | |
3772 | if (!param || pid < 0) | |
3a5c359a | 3773 | return -EINVAL; |
1da177e4 | 3774 | |
5fe85be0 | 3775 | rcu_read_lock(); |
1da177e4 LT |
3776 | p = find_process_by_pid(pid); |
3777 | retval = -ESRCH; | |
3778 | if (!p) | |
3779 | goto out_unlock; | |
3780 | ||
3781 | retval = security_task_getscheduler(p); | |
3782 | if (retval) | |
3783 | goto out_unlock; | |
3784 | ||
ce5f7f82 PZ |
3785 | if (task_has_rt_policy(p)) |
3786 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3787 | rcu_read_unlock(); |
1da177e4 LT |
3788 | |
3789 | /* | |
3790 | * This one might sleep, we cannot do it with a spinlock held ... | |
3791 | */ | |
3792 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3793 | ||
1da177e4 LT |
3794 | return retval; |
3795 | ||
3796 | out_unlock: | |
5fe85be0 | 3797 | rcu_read_unlock(); |
1da177e4 LT |
3798 | return retval; |
3799 | } | |
3800 | ||
d50dde5a DF |
3801 | static int sched_read_attr(struct sched_attr __user *uattr, |
3802 | struct sched_attr *attr, | |
3803 | unsigned int usize) | |
3804 | { | |
3805 | int ret; | |
3806 | ||
3807 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3808 | return -EFAULT; | |
3809 | ||
3810 | /* | |
3811 | * If we're handed a smaller struct than we know of, | |
3812 | * ensure all the unknown bits are 0 - i.e. old | |
3813 | * user-space does not get uncomplete information. | |
3814 | */ | |
3815 | if (usize < sizeof(*attr)) { | |
3816 | unsigned char *addr; | |
3817 | unsigned char *end; | |
3818 | ||
3819 | addr = (void *)attr + usize; | |
3820 | end = (void *)attr + sizeof(*attr); | |
3821 | ||
3822 | for (; addr < end; addr++) { | |
3823 | if (*addr) | |
3824 | goto err_size; | |
3825 | } | |
3826 | ||
3827 | attr->size = usize; | |
3828 | } | |
3829 | ||
4efbc454 | 3830 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3831 | if (ret) |
3832 | return -EFAULT; | |
3833 | ||
3834 | out: | |
3835 | return ret; | |
3836 | ||
3837 | err_size: | |
3838 | ret = -E2BIG; | |
3839 | goto out; | |
3840 | } | |
3841 | ||
3842 | /** | |
aab03e05 | 3843 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3844 | * @pid: the pid in question. |
5778fccf | 3845 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3846 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3847 | * @flags: for future extension. |
d50dde5a | 3848 | */ |
6d35ab48 PZ |
3849 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3850 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3851 | { |
3852 | struct sched_attr attr = { | |
3853 | .size = sizeof(struct sched_attr), | |
3854 | }; | |
3855 | struct task_struct *p; | |
3856 | int retval; | |
3857 | ||
3858 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3859 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3860 | return -EINVAL; |
3861 | ||
3862 | rcu_read_lock(); | |
3863 | p = find_process_by_pid(pid); | |
3864 | retval = -ESRCH; | |
3865 | if (!p) | |
3866 | goto out_unlock; | |
3867 | ||
3868 | retval = security_task_getscheduler(p); | |
3869 | if (retval) | |
3870 | goto out_unlock; | |
3871 | ||
3872 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3873 | if (p->sched_reset_on_fork) |
3874 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3875 | if (task_has_dl_policy(p)) |
3876 | __getparam_dl(p, &attr); | |
3877 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3878 | attr.sched_priority = p->rt_priority; |
3879 | else | |
d0ea0268 | 3880 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3881 | |
3882 | rcu_read_unlock(); | |
3883 | ||
3884 | retval = sched_read_attr(uattr, &attr, size); | |
3885 | return retval; | |
3886 | ||
3887 | out_unlock: | |
3888 | rcu_read_unlock(); | |
3889 | return retval; | |
3890 | } | |
3891 | ||
96f874e2 | 3892 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3893 | { |
5a16f3d3 | 3894 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3895 | struct task_struct *p; |
3896 | int retval; | |
1da177e4 | 3897 | |
23f5d142 | 3898 | rcu_read_lock(); |
1da177e4 LT |
3899 | |
3900 | p = find_process_by_pid(pid); | |
3901 | if (!p) { | |
23f5d142 | 3902 | rcu_read_unlock(); |
1da177e4 LT |
3903 | return -ESRCH; |
3904 | } | |
3905 | ||
23f5d142 | 3906 | /* Prevent p going away */ |
1da177e4 | 3907 | get_task_struct(p); |
23f5d142 | 3908 | rcu_read_unlock(); |
1da177e4 | 3909 | |
14a40ffc TH |
3910 | if (p->flags & PF_NO_SETAFFINITY) { |
3911 | retval = -EINVAL; | |
3912 | goto out_put_task; | |
3913 | } | |
5a16f3d3 RR |
3914 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3915 | retval = -ENOMEM; | |
3916 | goto out_put_task; | |
3917 | } | |
3918 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3919 | retval = -ENOMEM; | |
3920 | goto out_free_cpus_allowed; | |
3921 | } | |
1da177e4 | 3922 | retval = -EPERM; |
4c44aaaf EB |
3923 | if (!check_same_owner(p)) { |
3924 | rcu_read_lock(); | |
3925 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3926 | rcu_read_unlock(); | |
3927 | goto out_unlock; | |
3928 | } | |
3929 | rcu_read_unlock(); | |
3930 | } | |
1da177e4 | 3931 | |
b0ae1981 | 3932 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3933 | if (retval) |
3934 | goto out_unlock; | |
3935 | ||
e4099a5e PZ |
3936 | |
3937 | cpuset_cpus_allowed(p, cpus_allowed); | |
3938 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3939 | ||
332ac17e DF |
3940 | /* |
3941 | * Since bandwidth control happens on root_domain basis, | |
3942 | * if admission test is enabled, we only admit -deadline | |
3943 | * tasks allowed to run on all the CPUs in the task's | |
3944 | * root_domain. | |
3945 | */ | |
3946 | #ifdef CONFIG_SMP | |
3947 | if (task_has_dl_policy(p)) { | |
3948 | const struct cpumask *span = task_rq(p)->rd->span; | |
3949 | ||
e4099a5e | 3950 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
3951 | retval = -EBUSY; |
3952 | goto out_unlock; | |
3953 | } | |
3954 | } | |
3955 | #endif | |
49246274 | 3956 | again: |
5a16f3d3 | 3957 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3958 | |
8707d8b8 | 3959 | if (!retval) { |
5a16f3d3 RR |
3960 | cpuset_cpus_allowed(p, cpus_allowed); |
3961 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3962 | /* |
3963 | * We must have raced with a concurrent cpuset | |
3964 | * update. Just reset the cpus_allowed to the | |
3965 | * cpuset's cpus_allowed | |
3966 | */ | |
5a16f3d3 | 3967 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3968 | goto again; |
3969 | } | |
3970 | } | |
1da177e4 | 3971 | out_unlock: |
5a16f3d3 RR |
3972 | free_cpumask_var(new_mask); |
3973 | out_free_cpus_allowed: | |
3974 | free_cpumask_var(cpus_allowed); | |
3975 | out_put_task: | |
1da177e4 | 3976 | put_task_struct(p); |
1da177e4 LT |
3977 | return retval; |
3978 | } | |
3979 | ||
3980 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3981 | struct cpumask *new_mask) |
1da177e4 | 3982 | { |
96f874e2 RR |
3983 | if (len < cpumask_size()) |
3984 | cpumask_clear(new_mask); | |
3985 | else if (len > cpumask_size()) | |
3986 | len = cpumask_size(); | |
3987 | ||
1da177e4 LT |
3988 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3989 | } | |
3990 | ||
3991 | /** | |
3992 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3993 | * @pid: pid of the process | |
3994 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3995 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
3996 | * |
3997 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3998 | */ |
5add95d4 HC |
3999 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4000 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4001 | { |
5a16f3d3 | 4002 | cpumask_var_t new_mask; |
1da177e4 LT |
4003 | int retval; |
4004 | ||
5a16f3d3 RR |
4005 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4006 | return -ENOMEM; | |
1da177e4 | 4007 | |
5a16f3d3 RR |
4008 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4009 | if (retval == 0) | |
4010 | retval = sched_setaffinity(pid, new_mask); | |
4011 | free_cpumask_var(new_mask); | |
4012 | return retval; | |
1da177e4 LT |
4013 | } |
4014 | ||
96f874e2 | 4015 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4016 | { |
36c8b586 | 4017 | struct task_struct *p; |
31605683 | 4018 | unsigned long flags; |
1da177e4 | 4019 | int retval; |
1da177e4 | 4020 | |
23f5d142 | 4021 | rcu_read_lock(); |
1da177e4 LT |
4022 | |
4023 | retval = -ESRCH; | |
4024 | p = find_process_by_pid(pid); | |
4025 | if (!p) | |
4026 | goto out_unlock; | |
4027 | ||
e7834f8f DQ |
4028 | retval = security_task_getscheduler(p); |
4029 | if (retval) | |
4030 | goto out_unlock; | |
4031 | ||
013fdb80 | 4032 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4033 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4034 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4035 | |
4036 | out_unlock: | |
23f5d142 | 4037 | rcu_read_unlock(); |
1da177e4 | 4038 | |
9531b62f | 4039 | return retval; |
1da177e4 LT |
4040 | } |
4041 | ||
4042 | /** | |
4043 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4044 | * @pid: pid of the process | |
4045 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4046 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4047 | * |
4048 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4049 | */ |
5add95d4 HC |
4050 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4051 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4052 | { |
4053 | int ret; | |
f17c8607 | 4054 | cpumask_var_t mask; |
1da177e4 | 4055 | |
84fba5ec | 4056 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4057 | return -EINVAL; |
4058 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4059 | return -EINVAL; |
4060 | ||
f17c8607 RR |
4061 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4062 | return -ENOMEM; | |
1da177e4 | 4063 | |
f17c8607 RR |
4064 | ret = sched_getaffinity(pid, mask); |
4065 | if (ret == 0) { | |
8bc037fb | 4066 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4067 | |
4068 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4069 | ret = -EFAULT; |
4070 | else | |
cd3d8031 | 4071 | ret = retlen; |
f17c8607 RR |
4072 | } |
4073 | free_cpumask_var(mask); | |
1da177e4 | 4074 | |
f17c8607 | 4075 | return ret; |
1da177e4 LT |
4076 | } |
4077 | ||
4078 | /** | |
4079 | * sys_sched_yield - yield the current processor to other threads. | |
4080 | * | |
dd41f596 IM |
4081 | * This function yields the current CPU to other tasks. If there are no |
4082 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4083 | * |
4084 | * Return: 0. | |
1da177e4 | 4085 | */ |
5add95d4 | 4086 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4087 | { |
70b97a7f | 4088 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4089 | |
2d72376b | 4090 | schedstat_inc(rq, yld_count); |
4530d7ab | 4091 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4092 | |
4093 | /* | |
4094 | * Since we are going to call schedule() anyway, there's | |
4095 | * no need to preempt or enable interrupts: | |
4096 | */ | |
4097 | __release(rq->lock); | |
8a25d5de | 4098 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4099 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4100 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4101 | |
4102 | schedule(); | |
4103 | ||
4104 | return 0; | |
4105 | } | |
4106 | ||
e7b38404 | 4107 | static void __cond_resched(void) |
1da177e4 | 4108 | { |
bdb43806 | 4109 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4110 | __schedule(); |
bdb43806 | 4111 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4112 | } |
4113 | ||
02b67cc3 | 4114 | int __sched _cond_resched(void) |
1da177e4 | 4115 | { |
d86ee480 | 4116 | if (should_resched()) { |
1da177e4 LT |
4117 | __cond_resched(); |
4118 | return 1; | |
4119 | } | |
4120 | return 0; | |
4121 | } | |
02b67cc3 | 4122 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4123 | |
4124 | /* | |
613afbf8 | 4125 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4126 | * call schedule, and on return reacquire the lock. |
4127 | * | |
41a2d6cf | 4128 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4129 | * operations here to prevent schedule() from being called twice (once via |
4130 | * spin_unlock(), once by hand). | |
4131 | */ | |
613afbf8 | 4132 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4133 | { |
d86ee480 | 4134 | int resched = should_resched(); |
6df3cecb JK |
4135 | int ret = 0; |
4136 | ||
f607c668 PZ |
4137 | lockdep_assert_held(lock); |
4138 | ||
95c354fe | 4139 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4140 | spin_unlock(lock); |
d86ee480 | 4141 | if (resched) |
95c354fe NP |
4142 | __cond_resched(); |
4143 | else | |
4144 | cpu_relax(); | |
6df3cecb | 4145 | ret = 1; |
1da177e4 | 4146 | spin_lock(lock); |
1da177e4 | 4147 | } |
6df3cecb | 4148 | return ret; |
1da177e4 | 4149 | } |
613afbf8 | 4150 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4151 | |
613afbf8 | 4152 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4153 | { |
4154 | BUG_ON(!in_softirq()); | |
4155 | ||
d86ee480 | 4156 | if (should_resched()) { |
98d82567 | 4157 | local_bh_enable(); |
1da177e4 LT |
4158 | __cond_resched(); |
4159 | local_bh_disable(); | |
4160 | return 1; | |
4161 | } | |
4162 | return 0; | |
4163 | } | |
613afbf8 | 4164 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4165 | |
1da177e4 LT |
4166 | /** |
4167 | * yield - yield the current processor to other threads. | |
4168 | * | |
8e3fabfd PZ |
4169 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4170 | * | |
4171 | * The scheduler is at all times free to pick the calling task as the most | |
4172 | * eligible task to run, if removing the yield() call from your code breaks | |
4173 | * it, its already broken. | |
4174 | * | |
4175 | * Typical broken usage is: | |
4176 | * | |
4177 | * while (!event) | |
4178 | * yield(); | |
4179 | * | |
4180 | * where one assumes that yield() will let 'the other' process run that will | |
4181 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4182 | * happen. Never use yield() as a progress guarantee!! | |
4183 | * | |
4184 | * If you want to use yield() to wait for something, use wait_event(). | |
4185 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4186 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4187 | */ |
4188 | void __sched yield(void) | |
4189 | { | |
4190 | set_current_state(TASK_RUNNING); | |
4191 | sys_sched_yield(); | |
4192 | } | |
1da177e4 LT |
4193 | EXPORT_SYMBOL(yield); |
4194 | ||
d95f4122 MG |
4195 | /** |
4196 | * yield_to - yield the current processor to another thread in | |
4197 | * your thread group, or accelerate that thread toward the | |
4198 | * processor it's on. | |
16addf95 RD |
4199 | * @p: target task |
4200 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4201 | * |
4202 | * It's the caller's job to ensure that the target task struct | |
4203 | * can't go away on us before we can do any checks. | |
4204 | * | |
e69f6186 | 4205 | * Return: |
7b270f60 PZ |
4206 | * true (>0) if we indeed boosted the target task. |
4207 | * false (0) if we failed to boost the target. | |
4208 | * -ESRCH if there's no task to yield to. | |
d95f4122 MG |
4209 | */ |
4210 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
4211 | { | |
4212 | struct task_struct *curr = current; | |
4213 | struct rq *rq, *p_rq; | |
4214 | unsigned long flags; | |
c3c18640 | 4215 | int yielded = 0; |
d95f4122 MG |
4216 | |
4217 | local_irq_save(flags); | |
4218 | rq = this_rq(); | |
4219 | ||
4220 | again: | |
4221 | p_rq = task_rq(p); | |
7b270f60 PZ |
4222 | /* |
4223 | * If we're the only runnable task on the rq and target rq also | |
4224 | * has only one task, there's absolutely no point in yielding. | |
4225 | */ | |
4226 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4227 | yielded = -ESRCH; | |
4228 | goto out_irq; | |
4229 | } | |
4230 | ||
d95f4122 | 4231 | double_rq_lock(rq, p_rq); |
39e24d8f | 4232 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4233 | double_rq_unlock(rq, p_rq); |
4234 | goto again; | |
4235 | } | |
4236 | ||
4237 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4238 | goto out_unlock; |
d95f4122 MG |
4239 | |
4240 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4241 | goto out_unlock; |
d95f4122 MG |
4242 | |
4243 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4244 | goto out_unlock; |
d95f4122 MG |
4245 | |
4246 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4247 | if (yielded) { |
d95f4122 | 4248 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4249 | /* |
4250 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4251 | * fairness. | |
4252 | */ | |
4253 | if (preempt && rq != p_rq) | |
4254 | resched_task(p_rq->curr); | |
4255 | } | |
d95f4122 | 4256 | |
7b270f60 | 4257 | out_unlock: |
d95f4122 | 4258 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4259 | out_irq: |
d95f4122 MG |
4260 | local_irq_restore(flags); |
4261 | ||
7b270f60 | 4262 | if (yielded > 0) |
d95f4122 MG |
4263 | schedule(); |
4264 | ||
4265 | return yielded; | |
4266 | } | |
4267 | EXPORT_SYMBOL_GPL(yield_to); | |
4268 | ||
1da177e4 | 4269 | /* |
41a2d6cf | 4270 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4271 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4272 | */ |
4273 | void __sched io_schedule(void) | |
4274 | { | |
54d35f29 | 4275 | struct rq *rq = raw_rq(); |
1da177e4 | 4276 | |
0ff92245 | 4277 | delayacct_blkio_start(); |
1da177e4 | 4278 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4279 | blk_flush_plug(current); |
8f0dfc34 | 4280 | current->in_iowait = 1; |
1da177e4 | 4281 | schedule(); |
8f0dfc34 | 4282 | current->in_iowait = 0; |
1da177e4 | 4283 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4284 | delayacct_blkio_end(); |
1da177e4 | 4285 | } |
1da177e4 LT |
4286 | EXPORT_SYMBOL(io_schedule); |
4287 | ||
4288 | long __sched io_schedule_timeout(long timeout) | |
4289 | { | |
54d35f29 | 4290 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4291 | long ret; |
4292 | ||
0ff92245 | 4293 | delayacct_blkio_start(); |
1da177e4 | 4294 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4295 | blk_flush_plug(current); |
8f0dfc34 | 4296 | current->in_iowait = 1; |
1da177e4 | 4297 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4298 | current->in_iowait = 0; |
1da177e4 | 4299 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4300 | delayacct_blkio_end(); |
1da177e4 LT |
4301 | return ret; |
4302 | } | |
4303 | ||
4304 | /** | |
4305 | * sys_sched_get_priority_max - return maximum RT priority. | |
4306 | * @policy: scheduling class. | |
4307 | * | |
e69f6186 YB |
4308 | * Return: On success, this syscall returns the maximum |
4309 | * rt_priority that can be used by a given scheduling class. | |
4310 | * On failure, a negative error code is returned. | |
1da177e4 | 4311 | */ |
5add95d4 | 4312 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4313 | { |
4314 | int ret = -EINVAL; | |
4315 | ||
4316 | switch (policy) { | |
4317 | case SCHED_FIFO: | |
4318 | case SCHED_RR: | |
4319 | ret = MAX_USER_RT_PRIO-1; | |
4320 | break; | |
aab03e05 | 4321 | case SCHED_DEADLINE: |
1da177e4 | 4322 | case SCHED_NORMAL: |
b0a9499c | 4323 | case SCHED_BATCH: |
dd41f596 | 4324 | case SCHED_IDLE: |
1da177e4 LT |
4325 | ret = 0; |
4326 | break; | |
4327 | } | |
4328 | return ret; | |
4329 | } | |
4330 | ||
4331 | /** | |
4332 | * sys_sched_get_priority_min - return minimum RT priority. | |
4333 | * @policy: scheduling class. | |
4334 | * | |
e69f6186 YB |
4335 | * Return: On success, this syscall returns the minimum |
4336 | * rt_priority that can be used by a given scheduling class. | |
4337 | * On failure, a negative error code is returned. | |
1da177e4 | 4338 | */ |
5add95d4 | 4339 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4340 | { |
4341 | int ret = -EINVAL; | |
4342 | ||
4343 | switch (policy) { | |
4344 | case SCHED_FIFO: | |
4345 | case SCHED_RR: | |
4346 | ret = 1; | |
4347 | break; | |
aab03e05 | 4348 | case SCHED_DEADLINE: |
1da177e4 | 4349 | case SCHED_NORMAL: |
b0a9499c | 4350 | case SCHED_BATCH: |
dd41f596 | 4351 | case SCHED_IDLE: |
1da177e4 LT |
4352 | ret = 0; |
4353 | } | |
4354 | return ret; | |
4355 | } | |
4356 | ||
4357 | /** | |
4358 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4359 | * @pid: pid of the process. | |
4360 | * @interval: userspace pointer to the timeslice value. | |
4361 | * | |
4362 | * this syscall writes the default timeslice value of a given process | |
4363 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4364 | * |
4365 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4366 | * an error code. | |
1da177e4 | 4367 | */ |
17da2bd9 | 4368 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4369 | struct timespec __user *, interval) |
1da177e4 | 4370 | { |
36c8b586 | 4371 | struct task_struct *p; |
a4ec24b4 | 4372 | unsigned int time_slice; |
dba091b9 TG |
4373 | unsigned long flags; |
4374 | struct rq *rq; | |
3a5c359a | 4375 | int retval; |
1da177e4 | 4376 | struct timespec t; |
1da177e4 LT |
4377 | |
4378 | if (pid < 0) | |
3a5c359a | 4379 | return -EINVAL; |
1da177e4 LT |
4380 | |
4381 | retval = -ESRCH; | |
1a551ae7 | 4382 | rcu_read_lock(); |
1da177e4 LT |
4383 | p = find_process_by_pid(pid); |
4384 | if (!p) | |
4385 | goto out_unlock; | |
4386 | ||
4387 | retval = security_task_getscheduler(p); | |
4388 | if (retval) | |
4389 | goto out_unlock; | |
4390 | ||
dba091b9 | 4391 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4392 | time_slice = 0; |
4393 | if (p->sched_class->get_rr_interval) | |
4394 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4395 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4396 | |
1a551ae7 | 4397 | rcu_read_unlock(); |
a4ec24b4 | 4398 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4399 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4400 | return retval; |
3a5c359a | 4401 | |
1da177e4 | 4402 | out_unlock: |
1a551ae7 | 4403 | rcu_read_unlock(); |
1da177e4 LT |
4404 | return retval; |
4405 | } | |
4406 | ||
7c731e0a | 4407 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4408 | |
82a1fcb9 | 4409 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4410 | { |
1da177e4 | 4411 | unsigned long free = 0; |
4e79752c | 4412 | int ppid; |
36c8b586 | 4413 | unsigned state; |
1da177e4 | 4414 | |
1da177e4 | 4415 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4416 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4417 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4418 | #if BITS_PER_LONG == 32 |
1da177e4 | 4419 | if (state == TASK_RUNNING) |
3df0fc5b | 4420 | printk(KERN_CONT " running "); |
1da177e4 | 4421 | else |
3df0fc5b | 4422 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4423 | #else |
4424 | if (state == TASK_RUNNING) | |
3df0fc5b | 4425 | printk(KERN_CONT " running task "); |
1da177e4 | 4426 | else |
3df0fc5b | 4427 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4428 | #endif |
4429 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4430 | free = stack_not_used(p); |
1da177e4 | 4431 | #endif |
4e79752c PM |
4432 | rcu_read_lock(); |
4433 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4434 | rcu_read_unlock(); | |
3df0fc5b | 4435 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4436 | task_pid_nr(p), ppid, |
aa47b7e0 | 4437 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4438 | |
3d1cb205 | 4439 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4440 | show_stack(p, NULL); |
1da177e4 LT |
4441 | } |
4442 | ||
e59e2ae2 | 4443 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4444 | { |
36c8b586 | 4445 | struct task_struct *g, *p; |
1da177e4 | 4446 | |
4bd77321 | 4447 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4448 | printk(KERN_INFO |
4449 | " task PC stack pid father\n"); | |
1da177e4 | 4450 | #else |
3df0fc5b PZ |
4451 | printk(KERN_INFO |
4452 | " task PC stack pid father\n"); | |
1da177e4 | 4453 | #endif |
510f5acc | 4454 | rcu_read_lock(); |
1da177e4 LT |
4455 | do_each_thread(g, p) { |
4456 | /* | |
4457 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4458 | * console might take a lot of time: |
1da177e4 LT |
4459 | */ |
4460 | touch_nmi_watchdog(); | |
39bc89fd | 4461 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4462 | sched_show_task(p); |
1da177e4 LT |
4463 | } while_each_thread(g, p); |
4464 | ||
04c9167f JF |
4465 | touch_all_softlockup_watchdogs(); |
4466 | ||
dd41f596 IM |
4467 | #ifdef CONFIG_SCHED_DEBUG |
4468 | sysrq_sched_debug_show(); | |
4469 | #endif | |
510f5acc | 4470 | rcu_read_unlock(); |
e59e2ae2 IM |
4471 | /* |
4472 | * Only show locks if all tasks are dumped: | |
4473 | */ | |
93335a21 | 4474 | if (!state_filter) |
e59e2ae2 | 4475 | debug_show_all_locks(); |
1da177e4 LT |
4476 | } |
4477 | ||
0db0628d | 4478 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4479 | { |
dd41f596 | 4480 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4481 | } |
4482 | ||
f340c0d1 IM |
4483 | /** |
4484 | * init_idle - set up an idle thread for a given CPU | |
4485 | * @idle: task in question | |
4486 | * @cpu: cpu the idle task belongs to | |
4487 | * | |
4488 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4489 | * flag, to make booting more robust. | |
4490 | */ | |
0db0628d | 4491 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4492 | { |
70b97a7f | 4493 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4494 | unsigned long flags; |
4495 | ||
05fa785c | 4496 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4497 | |
5e1576ed | 4498 | __sched_fork(0, idle); |
06b83b5f | 4499 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4500 | idle->se.exec_start = sched_clock(); |
4501 | ||
1e1b6c51 | 4502 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4503 | /* |
4504 | * We're having a chicken and egg problem, even though we are | |
4505 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4506 | * lockdep check in task_group() will fail. | |
4507 | * | |
4508 | * Similar case to sched_fork(). / Alternatively we could | |
4509 | * use task_rq_lock() here and obtain the other rq->lock. | |
4510 | * | |
4511 | * Silence PROVE_RCU | |
4512 | */ | |
4513 | rcu_read_lock(); | |
dd41f596 | 4514 | __set_task_cpu(idle, cpu); |
6506cf6c | 4515 | rcu_read_unlock(); |
1da177e4 | 4516 | |
1da177e4 | 4517 | rq->curr = rq->idle = idle; |
77177856 | 4518 | idle->on_rq = 1; |
3ca7a440 PZ |
4519 | #if defined(CONFIG_SMP) |
4520 | idle->on_cpu = 1; | |
4866cde0 | 4521 | #endif |
05fa785c | 4522 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4523 | |
4524 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4525 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4526 | |
dd41f596 IM |
4527 | /* |
4528 | * The idle tasks have their own, simple scheduling class: | |
4529 | */ | |
4530 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4531 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4532 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4533 | #if defined(CONFIG_SMP) |
4534 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4535 | #endif | |
19978ca6 IM |
4536 | } |
4537 | ||
1da177e4 | 4538 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4539 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4540 | { | |
4541 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4542 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4543 | |
4544 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4545 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4546 | } |
4547 | ||
1da177e4 LT |
4548 | /* |
4549 | * This is how migration works: | |
4550 | * | |
969c7921 TH |
4551 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4552 | * stop_one_cpu(). | |
4553 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4554 | * off the CPU) | |
4555 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4556 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4557 | * it and puts it into the right queue. |
969c7921 TH |
4558 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4559 | * is done. | |
1da177e4 LT |
4560 | */ |
4561 | ||
4562 | /* | |
4563 | * Change a given task's CPU affinity. Migrate the thread to a | |
4564 | * proper CPU and schedule it away if the CPU it's executing on | |
4565 | * is removed from the allowed bitmask. | |
4566 | * | |
4567 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4568 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4569 | * call is not atomic; no spinlocks may be held. |
4570 | */ | |
96f874e2 | 4571 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4572 | { |
4573 | unsigned long flags; | |
70b97a7f | 4574 | struct rq *rq; |
969c7921 | 4575 | unsigned int dest_cpu; |
48f24c4d | 4576 | int ret = 0; |
1da177e4 LT |
4577 | |
4578 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4579 | |
db44fc01 YZ |
4580 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4581 | goto out; | |
4582 | ||
6ad4c188 | 4583 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4584 | ret = -EINVAL; |
4585 | goto out; | |
4586 | } | |
4587 | ||
1e1b6c51 | 4588 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4589 | |
1da177e4 | 4590 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4591 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4592 | goto out; |
4593 | ||
969c7921 | 4594 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4595 | if (p->on_rq) { |
969c7921 | 4596 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4597 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4598 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4599 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4600 | tlb_migrate_finish(p->mm); |
4601 | return 0; | |
4602 | } | |
4603 | out: | |
0122ec5b | 4604 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4605 | |
1da177e4 LT |
4606 | return ret; |
4607 | } | |
cd8ba7cd | 4608 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4609 | |
4610 | /* | |
41a2d6cf | 4611 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4612 | * this because either it can't run here any more (set_cpus_allowed() |
4613 | * away from this CPU, or CPU going down), or because we're | |
4614 | * attempting to rebalance this task on exec (sched_exec). | |
4615 | * | |
4616 | * So we race with normal scheduler movements, but that's OK, as long | |
4617 | * as the task is no longer on this CPU. | |
efc30814 KK |
4618 | * |
4619 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4620 | */ |
efc30814 | 4621 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4622 | { |
70b97a7f | 4623 | struct rq *rq_dest, *rq_src; |
e2912009 | 4624 | int ret = 0; |
1da177e4 | 4625 | |
e761b772 | 4626 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4627 | return ret; |
1da177e4 LT |
4628 | |
4629 | rq_src = cpu_rq(src_cpu); | |
4630 | rq_dest = cpu_rq(dest_cpu); | |
4631 | ||
0122ec5b | 4632 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4633 | double_rq_lock(rq_src, rq_dest); |
4634 | /* Already moved. */ | |
4635 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4636 | goto done; |
1da177e4 | 4637 | /* Affinity changed (again). */ |
fa17b507 | 4638 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4639 | goto fail; |
1da177e4 | 4640 | |
e2912009 PZ |
4641 | /* |
4642 | * If we're not on a rq, the next wake-up will ensure we're | |
4643 | * placed properly. | |
4644 | */ | |
fd2f4419 | 4645 | if (p->on_rq) { |
4ca9b72b | 4646 | dequeue_task(rq_src, p, 0); |
e2912009 | 4647 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4648 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4649 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4650 | } |
b1e38734 | 4651 | done: |
efc30814 | 4652 | ret = 1; |
b1e38734 | 4653 | fail: |
1da177e4 | 4654 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4655 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4656 | return ret; |
1da177e4 LT |
4657 | } |
4658 | ||
e6628d5b MG |
4659 | #ifdef CONFIG_NUMA_BALANCING |
4660 | /* Migrate current task p to target_cpu */ | |
4661 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4662 | { | |
4663 | struct migration_arg arg = { p, target_cpu }; | |
4664 | int curr_cpu = task_cpu(p); | |
4665 | ||
4666 | if (curr_cpu == target_cpu) | |
4667 | return 0; | |
4668 | ||
4669 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4670 | return -EINVAL; | |
4671 | ||
4672 | /* TODO: This is not properly updating schedstats */ | |
4673 | ||
286549dc | 4674 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4675 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4676 | } | |
0ec8aa00 PZ |
4677 | |
4678 | /* | |
4679 | * Requeue a task on a given node and accurately track the number of NUMA | |
4680 | * tasks on the runqueues | |
4681 | */ | |
4682 | void sched_setnuma(struct task_struct *p, int nid) | |
4683 | { | |
4684 | struct rq *rq; | |
4685 | unsigned long flags; | |
4686 | bool on_rq, running; | |
4687 | ||
4688 | rq = task_rq_lock(p, &flags); | |
4689 | on_rq = p->on_rq; | |
4690 | running = task_current(rq, p); | |
4691 | ||
4692 | if (on_rq) | |
4693 | dequeue_task(rq, p, 0); | |
4694 | if (running) | |
4695 | p->sched_class->put_prev_task(rq, p); | |
4696 | ||
4697 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4698 | |
4699 | if (running) | |
4700 | p->sched_class->set_curr_task(rq); | |
4701 | if (on_rq) | |
4702 | enqueue_task(rq, p, 0); | |
4703 | task_rq_unlock(rq, p, &flags); | |
4704 | } | |
e6628d5b MG |
4705 | #endif |
4706 | ||
1da177e4 | 4707 | /* |
969c7921 TH |
4708 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4709 | * and performs thread migration by bumping thread off CPU then | |
4710 | * 'pushing' onto another runqueue. | |
1da177e4 | 4711 | */ |
969c7921 | 4712 | static int migration_cpu_stop(void *data) |
1da177e4 | 4713 | { |
969c7921 | 4714 | struct migration_arg *arg = data; |
f7b4cddc | 4715 | |
969c7921 TH |
4716 | /* |
4717 | * The original target cpu might have gone down and we might | |
4718 | * be on another cpu but it doesn't matter. | |
4719 | */ | |
f7b4cddc | 4720 | local_irq_disable(); |
969c7921 | 4721 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4722 | local_irq_enable(); |
1da177e4 | 4723 | return 0; |
f7b4cddc ON |
4724 | } |
4725 | ||
1da177e4 | 4726 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4727 | |
054b9108 | 4728 | /* |
48c5ccae PZ |
4729 | * Ensures that the idle task is using init_mm right before its cpu goes |
4730 | * offline. | |
054b9108 | 4731 | */ |
48c5ccae | 4732 | void idle_task_exit(void) |
1da177e4 | 4733 | { |
48c5ccae | 4734 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4735 | |
48c5ccae | 4736 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4737 | |
a53efe5f | 4738 | if (mm != &init_mm) { |
48c5ccae | 4739 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4740 | finish_arch_post_lock_switch(); |
4741 | } | |
48c5ccae | 4742 | mmdrop(mm); |
1da177e4 LT |
4743 | } |
4744 | ||
4745 | /* | |
5d180232 PZ |
4746 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4747 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4748 | * nr_active count is stable. | |
4749 | * | |
4750 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4751 | */ |
5d180232 | 4752 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4753 | { |
5d180232 PZ |
4754 | long delta = calc_load_fold_active(rq); |
4755 | if (delta) | |
4756 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4757 | } |
4758 | ||
3f1d2a31 PZ |
4759 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4760 | { | |
4761 | } | |
4762 | ||
4763 | static const struct sched_class fake_sched_class = { | |
4764 | .put_prev_task = put_prev_task_fake, | |
4765 | }; | |
4766 | ||
4767 | static struct task_struct fake_task = { | |
4768 | /* | |
4769 | * Avoid pull_{rt,dl}_task() | |
4770 | */ | |
4771 | .prio = MAX_PRIO + 1, | |
4772 | .sched_class = &fake_sched_class, | |
4773 | }; | |
4774 | ||
48f24c4d | 4775 | /* |
48c5ccae PZ |
4776 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4777 | * try_to_wake_up()->select_task_rq(). | |
4778 | * | |
4779 | * Called with rq->lock held even though we'er in stop_machine() and | |
4780 | * there's no concurrency possible, we hold the required locks anyway | |
4781 | * because of lock validation efforts. | |
1da177e4 | 4782 | */ |
48c5ccae | 4783 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4784 | { |
70b97a7f | 4785 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4786 | struct task_struct *next, *stop = rq->stop; |
4787 | int dest_cpu; | |
1da177e4 LT |
4788 | |
4789 | /* | |
48c5ccae PZ |
4790 | * Fudge the rq selection such that the below task selection loop |
4791 | * doesn't get stuck on the currently eligible stop task. | |
4792 | * | |
4793 | * We're currently inside stop_machine() and the rq is either stuck | |
4794 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4795 | * either way we should never end up calling schedule() until we're | |
4796 | * done here. | |
1da177e4 | 4797 | */ |
48c5ccae | 4798 | rq->stop = NULL; |
48f24c4d | 4799 | |
77bd3970 FW |
4800 | /* |
4801 | * put_prev_task() and pick_next_task() sched | |
4802 | * class method both need to have an up-to-date | |
4803 | * value of rq->clock[_task] | |
4804 | */ | |
4805 | update_rq_clock(rq); | |
4806 | ||
dd41f596 | 4807 | for ( ; ; ) { |
48c5ccae PZ |
4808 | /* |
4809 | * There's this thread running, bail when that's the only | |
4810 | * remaining thread. | |
4811 | */ | |
4812 | if (rq->nr_running == 1) | |
dd41f596 | 4813 | break; |
48c5ccae | 4814 | |
3f1d2a31 | 4815 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4816 | BUG_ON(!next); |
79c53799 | 4817 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4818 | |
48c5ccae PZ |
4819 | /* Find suitable destination for @next, with force if needed. */ |
4820 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4821 | raw_spin_unlock(&rq->lock); | |
4822 | ||
4823 | __migrate_task(next, dead_cpu, dest_cpu); | |
4824 | ||
4825 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4826 | } |
dce48a84 | 4827 | |
48c5ccae | 4828 | rq->stop = stop; |
dce48a84 | 4829 | } |
48c5ccae | 4830 | |
1da177e4 LT |
4831 | #endif /* CONFIG_HOTPLUG_CPU */ |
4832 | ||
e692ab53 NP |
4833 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4834 | ||
4835 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4836 | { |
4837 | .procname = "sched_domain", | |
c57baf1e | 4838 | .mode = 0555, |
e0361851 | 4839 | }, |
56992309 | 4840 | {} |
e692ab53 NP |
4841 | }; |
4842 | ||
4843 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4844 | { |
4845 | .procname = "kernel", | |
c57baf1e | 4846 | .mode = 0555, |
e0361851 AD |
4847 | .child = sd_ctl_dir, |
4848 | }, | |
56992309 | 4849 | {} |
e692ab53 NP |
4850 | }; |
4851 | ||
4852 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4853 | { | |
4854 | struct ctl_table *entry = | |
5cf9f062 | 4855 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4856 | |
e692ab53 NP |
4857 | return entry; |
4858 | } | |
4859 | ||
6382bc90 MM |
4860 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4861 | { | |
cd790076 | 4862 | struct ctl_table *entry; |
6382bc90 | 4863 | |
cd790076 MM |
4864 | /* |
4865 | * In the intermediate directories, both the child directory and | |
4866 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4867 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4868 | * static strings and all have proc handlers. |
4869 | */ | |
4870 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4871 | if (entry->child) |
4872 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4873 | if (entry->proc_handler == NULL) |
4874 | kfree(entry->procname); | |
4875 | } | |
6382bc90 MM |
4876 | |
4877 | kfree(*tablep); | |
4878 | *tablep = NULL; | |
4879 | } | |
4880 | ||
201c373e | 4881 | static int min_load_idx = 0; |
fd9b86d3 | 4882 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4883 | |
e692ab53 | 4884 | static void |
e0361851 | 4885 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4886 | const char *procname, void *data, int maxlen, |
201c373e NK |
4887 | umode_t mode, proc_handler *proc_handler, |
4888 | bool load_idx) | |
e692ab53 | 4889 | { |
e692ab53 NP |
4890 | entry->procname = procname; |
4891 | entry->data = data; | |
4892 | entry->maxlen = maxlen; | |
4893 | entry->mode = mode; | |
4894 | entry->proc_handler = proc_handler; | |
201c373e NK |
4895 | |
4896 | if (load_idx) { | |
4897 | entry->extra1 = &min_load_idx; | |
4898 | entry->extra2 = &max_load_idx; | |
4899 | } | |
e692ab53 NP |
4900 | } |
4901 | ||
4902 | static struct ctl_table * | |
4903 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4904 | { | |
37e6bae8 | 4905 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4906 | |
ad1cdc1d MM |
4907 | if (table == NULL) |
4908 | return NULL; | |
4909 | ||
e0361851 | 4910 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4911 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4912 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4913 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4914 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4915 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4916 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4917 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4918 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4919 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4920 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4921 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4922 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4923 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4924 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4925 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4926 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4927 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4928 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4929 | &sd->cache_nice_tries, |
201c373e | 4930 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4931 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4932 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4933 | set_table_entry(&table[11], "max_newidle_lb_cost", |
4934 | &sd->max_newidle_lb_cost, | |
4935 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
4936 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 4937 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 4938 | /* &table[13] is terminator */ |
e692ab53 NP |
4939 | |
4940 | return table; | |
4941 | } | |
4942 | ||
be7002e6 | 4943 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4944 | { |
4945 | struct ctl_table *entry, *table; | |
4946 | struct sched_domain *sd; | |
4947 | int domain_num = 0, i; | |
4948 | char buf[32]; | |
4949 | ||
4950 | for_each_domain(cpu, sd) | |
4951 | domain_num++; | |
4952 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4953 | if (table == NULL) |
4954 | return NULL; | |
e692ab53 NP |
4955 | |
4956 | i = 0; | |
4957 | for_each_domain(cpu, sd) { | |
4958 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4959 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4960 | entry->mode = 0555; |
e692ab53 NP |
4961 | entry->child = sd_alloc_ctl_domain_table(sd); |
4962 | entry++; | |
4963 | i++; | |
4964 | } | |
4965 | return table; | |
4966 | } | |
4967 | ||
4968 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4969 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4970 | { |
6ad4c188 | 4971 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4972 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4973 | char buf[32]; | |
4974 | ||
7378547f MM |
4975 | WARN_ON(sd_ctl_dir[0].child); |
4976 | sd_ctl_dir[0].child = entry; | |
4977 | ||
ad1cdc1d MM |
4978 | if (entry == NULL) |
4979 | return; | |
4980 | ||
6ad4c188 | 4981 | for_each_possible_cpu(i) { |
e692ab53 | 4982 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4983 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4984 | entry->mode = 0555; |
e692ab53 | 4985 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4986 | entry++; |
e692ab53 | 4987 | } |
7378547f MM |
4988 | |
4989 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4990 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4991 | } | |
6382bc90 | 4992 | |
7378547f | 4993 | /* may be called multiple times per register */ |
6382bc90 MM |
4994 | static void unregister_sched_domain_sysctl(void) |
4995 | { | |
7378547f MM |
4996 | if (sd_sysctl_header) |
4997 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4998 | sd_sysctl_header = NULL; |
7378547f MM |
4999 | if (sd_ctl_dir[0].child) |
5000 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5001 | } |
e692ab53 | 5002 | #else |
6382bc90 MM |
5003 | static void register_sched_domain_sysctl(void) |
5004 | { | |
5005 | } | |
5006 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5007 | { |
5008 | } | |
5009 | #endif | |
5010 | ||
1f11eb6a GH |
5011 | static void set_rq_online(struct rq *rq) |
5012 | { | |
5013 | if (!rq->online) { | |
5014 | const struct sched_class *class; | |
5015 | ||
c6c4927b | 5016 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5017 | rq->online = 1; |
5018 | ||
5019 | for_each_class(class) { | |
5020 | if (class->rq_online) | |
5021 | class->rq_online(rq); | |
5022 | } | |
5023 | } | |
5024 | } | |
5025 | ||
5026 | static void set_rq_offline(struct rq *rq) | |
5027 | { | |
5028 | if (rq->online) { | |
5029 | const struct sched_class *class; | |
5030 | ||
5031 | for_each_class(class) { | |
5032 | if (class->rq_offline) | |
5033 | class->rq_offline(rq); | |
5034 | } | |
5035 | ||
c6c4927b | 5036 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5037 | rq->online = 0; |
5038 | } | |
5039 | } | |
5040 | ||
1da177e4 LT |
5041 | /* |
5042 | * migration_call - callback that gets triggered when a CPU is added. | |
5043 | * Here we can start up the necessary migration thread for the new CPU. | |
5044 | */ | |
0db0628d | 5045 | static int |
48f24c4d | 5046 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5047 | { |
48f24c4d | 5048 | int cpu = (long)hcpu; |
1da177e4 | 5049 | unsigned long flags; |
969c7921 | 5050 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5051 | |
48c5ccae | 5052 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5053 | |
1da177e4 | 5054 | case CPU_UP_PREPARE: |
a468d389 | 5055 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5056 | break; |
48f24c4d | 5057 | |
1da177e4 | 5058 | case CPU_ONLINE: |
1f94ef59 | 5059 | /* Update our root-domain */ |
05fa785c | 5060 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5061 | if (rq->rd) { |
c6c4927b | 5062 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5063 | |
5064 | set_rq_online(rq); | |
1f94ef59 | 5065 | } |
05fa785c | 5066 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5067 | break; |
48f24c4d | 5068 | |
1da177e4 | 5069 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5070 | case CPU_DYING: |
317f3941 | 5071 | sched_ttwu_pending(); |
57d885fe | 5072 | /* Update our root-domain */ |
05fa785c | 5073 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5074 | if (rq->rd) { |
c6c4927b | 5075 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5076 | set_rq_offline(rq); |
57d885fe | 5077 | } |
48c5ccae PZ |
5078 | migrate_tasks(cpu); |
5079 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5080 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5081 | break; |
48c5ccae | 5082 | |
5d180232 | 5083 | case CPU_DEAD: |
f319da0c | 5084 | calc_load_migrate(rq); |
57d885fe | 5085 | break; |
1da177e4 LT |
5086 | #endif |
5087 | } | |
49c022e6 PZ |
5088 | |
5089 | update_max_interval(); | |
5090 | ||
1da177e4 LT |
5091 | return NOTIFY_OK; |
5092 | } | |
5093 | ||
f38b0820 PM |
5094 | /* |
5095 | * Register at high priority so that task migration (migrate_all_tasks) | |
5096 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5097 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5098 | */ |
0db0628d | 5099 | static struct notifier_block migration_notifier = { |
1da177e4 | 5100 | .notifier_call = migration_call, |
50a323b7 | 5101 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5102 | }; |
5103 | ||
0db0628d | 5104 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5105 | unsigned long action, void *hcpu) |
5106 | { | |
5107 | switch (action & ~CPU_TASKS_FROZEN) { | |
3a101d05 TH |
5108 | case CPU_DOWN_FAILED: |
5109 | set_cpu_active((long)hcpu, true); | |
5110 | return NOTIFY_OK; | |
5111 | default: | |
5112 | return NOTIFY_DONE; | |
5113 | } | |
5114 | } | |
5115 | ||
0db0628d | 5116 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5117 | unsigned long action, void *hcpu) |
5118 | { | |
de212f18 PZ |
5119 | unsigned long flags; |
5120 | long cpu = (long)hcpu; | |
5121 | ||
3a101d05 TH |
5122 | switch (action & ~CPU_TASKS_FROZEN) { |
5123 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5124 | set_cpu_active(cpu, false); |
5125 | ||
5126 | /* explicitly allow suspend */ | |
5127 | if (!(action & CPU_TASKS_FROZEN)) { | |
5128 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5129 | bool overflow; | |
5130 | int cpus; | |
5131 | ||
5132 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5133 | cpus = dl_bw_cpus(cpu); | |
5134 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5135 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5136 | ||
5137 | if (overflow) | |
5138 | return notifier_from_errno(-EBUSY); | |
5139 | } | |
3a101d05 | 5140 | return NOTIFY_OK; |
3a101d05 | 5141 | } |
de212f18 PZ |
5142 | |
5143 | return NOTIFY_DONE; | |
3a101d05 TH |
5144 | } |
5145 | ||
7babe8db | 5146 | static int __init migration_init(void) |
1da177e4 LT |
5147 | { |
5148 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5149 | int err; |
48f24c4d | 5150 | |
3a101d05 | 5151 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5152 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5153 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5154 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5155 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5156 | |
3a101d05 TH |
5157 | /* Register cpu active notifiers */ |
5158 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5159 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5160 | ||
a004cd42 | 5161 | return 0; |
1da177e4 | 5162 | } |
7babe8db | 5163 | early_initcall(migration_init); |
1da177e4 LT |
5164 | #endif |
5165 | ||
5166 | #ifdef CONFIG_SMP | |
476f3534 | 5167 | |
4cb98839 PZ |
5168 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5169 | ||
3e9830dc | 5170 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5171 | |
d039ac60 | 5172 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5173 | |
d039ac60 | 5174 | static int __init sched_debug_setup(char *str) |
f6630114 | 5175 | { |
d039ac60 | 5176 | sched_debug_enabled = 1; |
f6630114 MT |
5177 | |
5178 | return 0; | |
5179 | } | |
d039ac60 PZ |
5180 | early_param("sched_debug", sched_debug_setup); |
5181 | ||
5182 | static inline bool sched_debug(void) | |
5183 | { | |
5184 | return sched_debug_enabled; | |
5185 | } | |
f6630114 | 5186 | |
7c16ec58 | 5187 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5188 | struct cpumask *groupmask) |
1da177e4 | 5189 | { |
4dcf6aff | 5190 | struct sched_group *group = sd->groups; |
434d53b0 | 5191 | char str[256]; |
1da177e4 | 5192 | |
968ea6d8 | 5193 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5194 | cpumask_clear(groupmask); |
4dcf6aff IM |
5195 | |
5196 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5197 | ||
5198 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5199 | printk("does not load-balance\n"); |
4dcf6aff | 5200 | if (sd->parent) |
3df0fc5b PZ |
5201 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5202 | " has parent"); | |
4dcf6aff | 5203 | return -1; |
41c7ce9a NP |
5204 | } |
5205 | ||
3df0fc5b | 5206 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5207 | |
758b2cdc | 5208 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5209 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5210 | "CPU%d\n", cpu); | |
4dcf6aff | 5211 | } |
758b2cdc | 5212 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5213 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5214 | " CPU%d\n", cpu); | |
4dcf6aff | 5215 | } |
1da177e4 | 5216 | |
4dcf6aff | 5217 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5218 | do { |
4dcf6aff | 5219 | if (!group) { |
3df0fc5b PZ |
5220 | printk("\n"); |
5221 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5222 | break; |
5223 | } | |
5224 | ||
c3decf0d PZ |
5225 | /* |
5226 | * Even though we initialize ->power to something semi-sane, | |
5227 | * we leave power_orig unset. This allows us to detect if | |
5228 | * domain iteration is still funny without causing /0 traps. | |
5229 | */ | |
5230 | if (!group->sgp->power_orig) { | |
3df0fc5b PZ |
5231 | printk(KERN_CONT "\n"); |
5232 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5233 | "set\n"); | |
4dcf6aff IM |
5234 | break; |
5235 | } | |
1da177e4 | 5236 | |
758b2cdc | 5237 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5238 | printk(KERN_CONT "\n"); |
5239 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5240 | break; |
5241 | } | |
1da177e4 | 5242 | |
cb83b629 PZ |
5243 | if (!(sd->flags & SD_OVERLAP) && |
5244 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5245 | printk(KERN_CONT "\n"); |
5246 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5247 | break; |
5248 | } | |
1da177e4 | 5249 | |
758b2cdc | 5250 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5251 | |
968ea6d8 | 5252 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5253 | |
3df0fc5b | 5254 | printk(KERN_CONT " %s", str); |
9c3f75cb | 5255 | if (group->sgp->power != SCHED_POWER_SCALE) { |
3df0fc5b | 5256 | printk(KERN_CONT " (cpu_power = %d)", |
9c3f75cb | 5257 | group->sgp->power); |
381512cf | 5258 | } |
1da177e4 | 5259 | |
4dcf6aff IM |
5260 | group = group->next; |
5261 | } while (group != sd->groups); | |
3df0fc5b | 5262 | printk(KERN_CONT "\n"); |
1da177e4 | 5263 | |
758b2cdc | 5264 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5265 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5266 | |
758b2cdc RR |
5267 | if (sd->parent && |
5268 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5269 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5270 | "of domain->span\n"); | |
4dcf6aff IM |
5271 | return 0; |
5272 | } | |
1da177e4 | 5273 | |
4dcf6aff IM |
5274 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5275 | { | |
5276 | int level = 0; | |
1da177e4 | 5277 | |
d039ac60 | 5278 | if (!sched_debug_enabled) |
f6630114 MT |
5279 | return; |
5280 | ||
4dcf6aff IM |
5281 | if (!sd) { |
5282 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5283 | return; | |
5284 | } | |
1da177e4 | 5285 | |
4dcf6aff IM |
5286 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5287 | ||
5288 | for (;;) { | |
4cb98839 | 5289 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5290 | break; |
1da177e4 LT |
5291 | level++; |
5292 | sd = sd->parent; | |
33859f7f | 5293 | if (!sd) |
4dcf6aff IM |
5294 | break; |
5295 | } | |
1da177e4 | 5296 | } |
6d6bc0ad | 5297 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5298 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5299 | static inline bool sched_debug(void) |
5300 | { | |
5301 | return false; | |
5302 | } | |
6d6bc0ad | 5303 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5304 | |
1a20ff27 | 5305 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5306 | { |
758b2cdc | 5307 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5308 | return 1; |
5309 | ||
5310 | /* Following flags need at least 2 groups */ | |
5311 | if (sd->flags & (SD_LOAD_BALANCE | | |
5312 | SD_BALANCE_NEWIDLE | | |
5313 | SD_BALANCE_FORK | | |
89c4710e SS |
5314 | SD_BALANCE_EXEC | |
5315 | SD_SHARE_CPUPOWER | | |
d77b3ed5 VG |
5316 | SD_SHARE_PKG_RESOURCES | |
5317 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5318 | if (sd->groups != sd->groups->next) |
5319 | return 0; | |
5320 | } | |
5321 | ||
5322 | /* Following flags don't use groups */ | |
c88d5910 | 5323 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5324 | return 0; |
5325 | ||
5326 | return 1; | |
5327 | } | |
5328 | ||
48f24c4d IM |
5329 | static int |
5330 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5331 | { |
5332 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5333 | ||
5334 | if (sd_degenerate(parent)) | |
5335 | return 1; | |
5336 | ||
758b2cdc | 5337 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5338 | return 0; |
5339 | ||
245af2c7 SS |
5340 | /* Flags needing groups don't count if only 1 group in parent */ |
5341 | if (parent->groups == parent->groups->next) { | |
5342 | pflags &= ~(SD_LOAD_BALANCE | | |
5343 | SD_BALANCE_NEWIDLE | | |
5344 | SD_BALANCE_FORK | | |
89c4710e SS |
5345 | SD_BALANCE_EXEC | |
5346 | SD_SHARE_CPUPOWER | | |
10866e62 | 5347 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5348 | SD_PREFER_SIBLING | |
5349 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5350 | if (nr_node_ids == 1) |
5351 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5352 | } |
5353 | if (~cflags & pflags) | |
5354 | return 0; | |
5355 | ||
5356 | return 1; | |
5357 | } | |
5358 | ||
dce840a0 | 5359 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5360 | { |
dce840a0 | 5361 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5362 | |
68e74568 | 5363 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5364 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5365 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5366 | free_cpumask_var(rd->rto_mask); |
5367 | free_cpumask_var(rd->online); | |
5368 | free_cpumask_var(rd->span); | |
5369 | kfree(rd); | |
5370 | } | |
5371 | ||
57d885fe GH |
5372 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5373 | { | |
a0490fa3 | 5374 | struct root_domain *old_rd = NULL; |
57d885fe | 5375 | unsigned long flags; |
57d885fe | 5376 | |
05fa785c | 5377 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5378 | |
5379 | if (rq->rd) { | |
a0490fa3 | 5380 | old_rd = rq->rd; |
57d885fe | 5381 | |
c6c4927b | 5382 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5383 | set_rq_offline(rq); |
57d885fe | 5384 | |
c6c4927b | 5385 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5386 | |
a0490fa3 | 5387 | /* |
0515973f | 5388 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5389 | * set old_rd to NULL to skip the freeing later |
5390 | * in this function: | |
5391 | */ | |
5392 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5393 | old_rd = NULL; | |
57d885fe GH |
5394 | } |
5395 | ||
5396 | atomic_inc(&rd->refcount); | |
5397 | rq->rd = rd; | |
5398 | ||
c6c4927b | 5399 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5400 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5401 | set_rq_online(rq); |
57d885fe | 5402 | |
05fa785c | 5403 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5404 | |
5405 | if (old_rd) | |
dce840a0 | 5406 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5407 | } |
5408 | ||
68c38fc3 | 5409 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5410 | { |
5411 | memset(rd, 0, sizeof(*rd)); | |
5412 | ||
68c38fc3 | 5413 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5414 | goto out; |
68c38fc3 | 5415 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5416 | goto free_span; |
1baca4ce | 5417 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5418 | goto free_online; |
1baca4ce JL |
5419 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5420 | goto free_dlo_mask; | |
6e0534f2 | 5421 | |
332ac17e | 5422 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5423 | if (cpudl_init(&rd->cpudl) != 0) |
5424 | goto free_dlo_mask; | |
332ac17e | 5425 | |
68c38fc3 | 5426 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5427 | goto free_rto_mask; |
c6c4927b | 5428 | return 0; |
6e0534f2 | 5429 | |
68e74568 RR |
5430 | free_rto_mask: |
5431 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5432 | free_dlo_mask: |
5433 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5434 | free_online: |
5435 | free_cpumask_var(rd->online); | |
5436 | free_span: | |
5437 | free_cpumask_var(rd->span); | |
0c910d28 | 5438 | out: |
c6c4927b | 5439 | return -ENOMEM; |
57d885fe GH |
5440 | } |
5441 | ||
029632fb PZ |
5442 | /* |
5443 | * By default the system creates a single root-domain with all cpus as | |
5444 | * members (mimicking the global state we have today). | |
5445 | */ | |
5446 | struct root_domain def_root_domain; | |
5447 | ||
57d885fe GH |
5448 | static void init_defrootdomain(void) |
5449 | { | |
68c38fc3 | 5450 | init_rootdomain(&def_root_domain); |
c6c4927b | 5451 | |
57d885fe GH |
5452 | atomic_set(&def_root_domain.refcount, 1); |
5453 | } | |
5454 | ||
dc938520 | 5455 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5456 | { |
5457 | struct root_domain *rd; | |
5458 | ||
5459 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5460 | if (!rd) | |
5461 | return NULL; | |
5462 | ||
68c38fc3 | 5463 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5464 | kfree(rd); |
5465 | return NULL; | |
5466 | } | |
57d885fe GH |
5467 | |
5468 | return rd; | |
5469 | } | |
5470 | ||
e3589f6c PZ |
5471 | static void free_sched_groups(struct sched_group *sg, int free_sgp) |
5472 | { | |
5473 | struct sched_group *tmp, *first; | |
5474 | ||
5475 | if (!sg) | |
5476 | return; | |
5477 | ||
5478 | first = sg; | |
5479 | do { | |
5480 | tmp = sg->next; | |
5481 | ||
5482 | if (free_sgp && atomic_dec_and_test(&sg->sgp->ref)) | |
5483 | kfree(sg->sgp); | |
5484 | ||
5485 | kfree(sg); | |
5486 | sg = tmp; | |
5487 | } while (sg != first); | |
5488 | } | |
5489 | ||
dce840a0 PZ |
5490 | static void free_sched_domain(struct rcu_head *rcu) |
5491 | { | |
5492 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5493 | |
5494 | /* | |
5495 | * If its an overlapping domain it has private groups, iterate and | |
5496 | * nuke them all. | |
5497 | */ | |
5498 | if (sd->flags & SD_OVERLAP) { | |
5499 | free_sched_groups(sd->groups, 1); | |
5500 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
9c3f75cb | 5501 | kfree(sd->groups->sgp); |
dce840a0 | 5502 | kfree(sd->groups); |
9c3f75cb | 5503 | } |
dce840a0 PZ |
5504 | kfree(sd); |
5505 | } | |
5506 | ||
5507 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5508 | { | |
5509 | call_rcu(&sd->rcu, free_sched_domain); | |
5510 | } | |
5511 | ||
5512 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5513 | { | |
5514 | for (; sd; sd = sd->parent) | |
5515 | destroy_sched_domain(sd, cpu); | |
5516 | } | |
5517 | ||
518cd623 PZ |
5518 | /* |
5519 | * Keep a special pointer to the highest sched_domain that has | |
5520 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5521 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5522 | * | |
5523 | * Also keep a unique ID per domain (we use the first cpu number in | |
5524 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5525 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5526 | */ |
5527 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5528 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5529 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5530 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5531 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5532 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5533 | |
5534 | static void update_top_cache_domain(int cpu) | |
5535 | { | |
5536 | struct sched_domain *sd; | |
5d4cf996 | 5537 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5538 | int id = cpu; |
7d9ffa89 | 5539 | int size = 1; |
518cd623 PZ |
5540 | |
5541 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5542 | if (sd) { |
518cd623 | 5543 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5544 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5545 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5546 | } |
5d4cf996 | 5547 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5548 | |
5549 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5550 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5551 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5552 | |
5553 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5554 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5555 | |
5556 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5557 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5558 | } |
5559 | ||
1da177e4 | 5560 | /* |
0eab9146 | 5561 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5562 | * hold the hotplug lock. |
5563 | */ | |
0eab9146 IM |
5564 | static void |
5565 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5566 | { |
70b97a7f | 5567 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5568 | struct sched_domain *tmp; |
5569 | ||
5570 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5571 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5572 | struct sched_domain *parent = tmp->parent; |
5573 | if (!parent) | |
5574 | break; | |
f29c9b1c | 5575 | |
1a848870 | 5576 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5577 | tmp->parent = parent->parent; |
1a848870 SS |
5578 | if (parent->parent) |
5579 | parent->parent->child = tmp; | |
10866e62 PZ |
5580 | /* |
5581 | * Transfer SD_PREFER_SIBLING down in case of a | |
5582 | * degenerate parent; the spans match for this | |
5583 | * so the property transfers. | |
5584 | */ | |
5585 | if (parent->flags & SD_PREFER_SIBLING) | |
5586 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5587 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5588 | } else |
5589 | tmp = tmp->parent; | |
245af2c7 SS |
5590 | } |
5591 | ||
1a848870 | 5592 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5593 | tmp = sd; |
245af2c7 | 5594 | sd = sd->parent; |
dce840a0 | 5595 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5596 | if (sd) |
5597 | sd->child = NULL; | |
5598 | } | |
1da177e4 | 5599 | |
4cb98839 | 5600 | sched_domain_debug(sd, cpu); |
1da177e4 | 5601 | |
57d885fe | 5602 | rq_attach_root(rq, rd); |
dce840a0 | 5603 | tmp = rq->sd; |
674311d5 | 5604 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5605 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5606 | |
5607 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5608 | } |
5609 | ||
5610 | /* cpus with isolated domains */ | |
dcc30a35 | 5611 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5612 | |
5613 | /* Setup the mask of cpus configured for isolated domains */ | |
5614 | static int __init isolated_cpu_setup(char *str) | |
5615 | { | |
bdddd296 | 5616 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5617 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5618 | return 1; |
5619 | } | |
5620 | ||
8927f494 | 5621 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5622 | |
49a02c51 | 5623 | struct s_data { |
21d42ccf | 5624 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5625 | struct root_domain *rd; |
5626 | }; | |
5627 | ||
2109b99e | 5628 | enum s_alloc { |
2109b99e | 5629 | sa_rootdomain, |
21d42ccf | 5630 | sa_sd, |
dce840a0 | 5631 | sa_sd_storage, |
2109b99e AH |
5632 | sa_none, |
5633 | }; | |
5634 | ||
c1174876 PZ |
5635 | /* |
5636 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5637 | * domain traversal. | |
5638 | * | |
5639 | * Asymmetric node setups can result in situations where the domain tree is of | |
5640 | * unequal depth, make sure to skip domains that already cover the entire | |
5641 | * range. | |
5642 | * | |
5643 | * In that case build_sched_domains() will have terminated the iteration early | |
5644 | * and our sibling sd spans will be empty. Domains should always include the | |
5645 | * cpu they're built on, so check that. | |
5646 | * | |
5647 | */ | |
5648 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5649 | { | |
5650 | const struct cpumask *span = sched_domain_span(sd); | |
5651 | struct sd_data *sdd = sd->private; | |
5652 | struct sched_domain *sibling; | |
5653 | int i; | |
5654 | ||
5655 | for_each_cpu(i, span) { | |
5656 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5657 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5658 | continue; | |
5659 | ||
5660 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5661 | } | |
5662 | } | |
5663 | ||
5664 | /* | |
5665 | * Return the canonical balance cpu for this group, this is the first cpu | |
5666 | * of this group that's also in the iteration mask. | |
5667 | */ | |
5668 | int group_balance_cpu(struct sched_group *sg) | |
5669 | { | |
5670 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5671 | } | |
5672 | ||
e3589f6c PZ |
5673 | static int |
5674 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5675 | { | |
5676 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5677 | const struct cpumask *span = sched_domain_span(sd); | |
5678 | struct cpumask *covered = sched_domains_tmpmask; | |
5679 | struct sd_data *sdd = sd->private; | |
5680 | struct sched_domain *child; | |
5681 | int i; | |
5682 | ||
5683 | cpumask_clear(covered); | |
5684 | ||
5685 | for_each_cpu(i, span) { | |
5686 | struct cpumask *sg_span; | |
5687 | ||
5688 | if (cpumask_test_cpu(i, covered)) | |
5689 | continue; | |
5690 | ||
c1174876 PZ |
5691 | child = *per_cpu_ptr(sdd->sd, i); |
5692 | ||
5693 | /* See the comment near build_group_mask(). */ | |
5694 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5695 | continue; | |
5696 | ||
e3589f6c | 5697 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5698 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5699 | |
5700 | if (!sg) | |
5701 | goto fail; | |
5702 | ||
5703 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5704 | if (child->child) { |
5705 | child = child->child; | |
5706 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5707 | } else | |
5708 | cpumask_set_cpu(i, sg_span); | |
5709 | ||
5710 | cpumask_or(covered, covered, sg_span); | |
5711 | ||
74a5ce20 | 5712 | sg->sgp = *per_cpu_ptr(sdd->sgp, i); |
c1174876 PZ |
5713 | if (atomic_inc_return(&sg->sgp->ref) == 1) |
5714 | build_group_mask(sd, sg); | |
5715 | ||
c3decf0d PZ |
5716 | /* |
5717 | * Initialize sgp->power such that even if we mess up the | |
5718 | * domains and no possible iteration will get us here, we won't | |
5719 | * die on a /0 trap. | |
5720 | */ | |
5721 | sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span); | |
8e8339a3 | 5722 | sg->sgp->power_orig = sg->sgp->power; |
e3589f6c | 5723 | |
c1174876 PZ |
5724 | /* |
5725 | * Make sure the first group of this domain contains the | |
5726 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5727 | * breaks. See update_sg_lb_stats(). | |
5728 | */ | |
74a5ce20 | 5729 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5730 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5731 | groups = sg; |
5732 | ||
5733 | if (!first) | |
5734 | first = sg; | |
5735 | if (last) | |
5736 | last->next = sg; | |
5737 | last = sg; | |
5738 | last->next = first; | |
5739 | } | |
5740 | sd->groups = groups; | |
5741 | ||
5742 | return 0; | |
5743 | ||
5744 | fail: | |
5745 | free_sched_groups(first, 0); | |
5746 | ||
5747 | return -ENOMEM; | |
5748 | } | |
5749 | ||
dce840a0 | 5750 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5751 | { |
dce840a0 PZ |
5752 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5753 | struct sched_domain *child = sd->child; | |
1da177e4 | 5754 | |
dce840a0 PZ |
5755 | if (child) |
5756 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5757 | |
9c3f75cb | 5758 | if (sg) { |
dce840a0 | 5759 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
9c3f75cb | 5760 | (*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu); |
e3589f6c | 5761 | atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */ |
9c3f75cb | 5762 | } |
dce840a0 PZ |
5763 | |
5764 | return cpu; | |
1e9f28fa | 5765 | } |
1e9f28fa | 5766 | |
01a08546 | 5767 | /* |
dce840a0 PZ |
5768 | * build_sched_groups will build a circular linked list of the groups |
5769 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5770 | * and ->cpu_power to 0. | |
e3589f6c PZ |
5771 | * |
5772 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5773 | */ |
e3589f6c PZ |
5774 | static int |
5775 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5776 | { |
dce840a0 PZ |
5777 | struct sched_group *first = NULL, *last = NULL; |
5778 | struct sd_data *sdd = sd->private; | |
5779 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5780 | struct cpumask *covered; |
dce840a0 | 5781 | int i; |
9c1cfda2 | 5782 | |
e3589f6c PZ |
5783 | get_group(cpu, sdd, &sd->groups); |
5784 | atomic_inc(&sd->groups->ref); | |
5785 | ||
0936629f | 5786 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5787 | return 0; |
5788 | ||
f96225fd PZ |
5789 | lockdep_assert_held(&sched_domains_mutex); |
5790 | covered = sched_domains_tmpmask; | |
5791 | ||
dce840a0 | 5792 | cpumask_clear(covered); |
6711cab4 | 5793 | |
dce840a0 PZ |
5794 | for_each_cpu(i, span) { |
5795 | struct sched_group *sg; | |
cd08e923 | 5796 | int group, j; |
6711cab4 | 5797 | |
dce840a0 PZ |
5798 | if (cpumask_test_cpu(i, covered)) |
5799 | continue; | |
6711cab4 | 5800 | |
cd08e923 | 5801 | group = get_group(i, sdd, &sg); |
dce840a0 | 5802 | cpumask_clear(sched_group_cpus(sg)); |
9c3f75cb | 5803 | sg->sgp->power = 0; |
c1174876 | 5804 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5805 | |
dce840a0 PZ |
5806 | for_each_cpu(j, span) { |
5807 | if (get_group(j, sdd, NULL) != group) | |
5808 | continue; | |
0601a88d | 5809 | |
dce840a0 PZ |
5810 | cpumask_set_cpu(j, covered); |
5811 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5812 | } | |
0601a88d | 5813 | |
dce840a0 PZ |
5814 | if (!first) |
5815 | first = sg; | |
5816 | if (last) | |
5817 | last->next = sg; | |
5818 | last = sg; | |
5819 | } | |
5820 | last->next = first; | |
e3589f6c PZ |
5821 | |
5822 | return 0; | |
0601a88d | 5823 | } |
51888ca2 | 5824 | |
89c4710e SS |
5825 | /* |
5826 | * Initialize sched groups cpu_power. | |
5827 | * | |
5828 | * cpu_power indicates the capacity of sched group, which is used while | |
5829 | * distributing the load between different sched groups in a sched domain. | |
5830 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5831 | * there are asymmetries in the topology. If there are asymmetries, group | |
5832 | * having more cpu_power will pickup more load compared to the group having | |
5833 | * less cpu_power. | |
89c4710e SS |
5834 | */ |
5835 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5836 | { | |
e3589f6c | 5837 | struct sched_group *sg = sd->groups; |
89c4710e | 5838 | |
94c95ba6 | 5839 | WARN_ON(!sg); |
e3589f6c PZ |
5840 | |
5841 | do { | |
5842 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5843 | sg = sg->next; | |
5844 | } while (sg != sd->groups); | |
89c4710e | 5845 | |
c1174876 | 5846 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5847 | return; |
aae6d3dd | 5848 | |
d274cb30 | 5849 | update_group_power(sd, cpu); |
69e1e811 | 5850 | atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight); |
89c4710e SS |
5851 | } |
5852 | ||
7c16ec58 MT |
5853 | /* |
5854 | * Initializers for schedule domains | |
5855 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5856 | */ | |
5857 | ||
1d3504fc | 5858 | static int default_relax_domain_level = -1; |
60495e77 | 5859 | int sched_domain_level_max; |
1d3504fc HS |
5860 | |
5861 | static int __init setup_relax_domain_level(char *str) | |
5862 | { | |
a841f8ce DS |
5863 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5864 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5865 | |
1d3504fc HS |
5866 | return 1; |
5867 | } | |
5868 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5869 | ||
5870 | static void set_domain_attribute(struct sched_domain *sd, | |
5871 | struct sched_domain_attr *attr) | |
5872 | { | |
5873 | int request; | |
5874 | ||
5875 | if (!attr || attr->relax_domain_level < 0) { | |
5876 | if (default_relax_domain_level < 0) | |
5877 | return; | |
5878 | else | |
5879 | request = default_relax_domain_level; | |
5880 | } else | |
5881 | request = attr->relax_domain_level; | |
5882 | if (request < sd->level) { | |
5883 | /* turn off idle balance on this domain */ | |
c88d5910 | 5884 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5885 | } else { |
5886 | /* turn on idle balance on this domain */ | |
c88d5910 | 5887 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5888 | } |
5889 | } | |
5890 | ||
54ab4ff4 PZ |
5891 | static void __sdt_free(const struct cpumask *cpu_map); |
5892 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5893 | ||
2109b99e AH |
5894 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5895 | const struct cpumask *cpu_map) | |
5896 | { | |
5897 | switch (what) { | |
2109b99e | 5898 | case sa_rootdomain: |
822ff793 PZ |
5899 | if (!atomic_read(&d->rd->refcount)) |
5900 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5901 | case sa_sd: |
5902 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5903 | case sa_sd_storage: |
54ab4ff4 | 5904 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5905 | case sa_none: |
5906 | break; | |
5907 | } | |
5908 | } | |
3404c8d9 | 5909 | |
2109b99e AH |
5910 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5911 | const struct cpumask *cpu_map) | |
5912 | { | |
dce840a0 PZ |
5913 | memset(d, 0, sizeof(*d)); |
5914 | ||
54ab4ff4 PZ |
5915 | if (__sdt_alloc(cpu_map)) |
5916 | return sa_sd_storage; | |
dce840a0 PZ |
5917 | d->sd = alloc_percpu(struct sched_domain *); |
5918 | if (!d->sd) | |
5919 | return sa_sd_storage; | |
2109b99e | 5920 | d->rd = alloc_rootdomain(); |
dce840a0 | 5921 | if (!d->rd) |
21d42ccf | 5922 | return sa_sd; |
2109b99e AH |
5923 | return sa_rootdomain; |
5924 | } | |
57d885fe | 5925 | |
dce840a0 PZ |
5926 | /* |
5927 | * NULL the sd_data elements we've used to build the sched_domain and | |
5928 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5929 | * will not free the data we're using. | |
5930 | */ | |
5931 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5932 | { | |
5933 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5934 | |
5935 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5936 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5937 | ||
e3589f6c | 5938 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5939 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c PZ |
5940 | |
5941 | if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref)) | |
9c3f75cb | 5942 | *per_cpu_ptr(sdd->sgp, cpu) = NULL; |
dce840a0 PZ |
5943 | } |
5944 | ||
cb83b629 | 5945 | #ifdef CONFIG_NUMA |
cb83b629 | 5946 | static int sched_domains_numa_levels; |
cb83b629 PZ |
5947 | static int *sched_domains_numa_distance; |
5948 | static struct cpumask ***sched_domains_numa_masks; | |
5949 | static int sched_domains_curr_level; | |
143e1e28 | 5950 | #endif |
cb83b629 | 5951 | |
143e1e28 VG |
5952 | /* |
5953 | * SD_flags allowed in topology descriptions. | |
5954 | * | |
5955 | * SD_SHARE_CPUPOWER - describes SMT topologies | |
5956 | * SD_SHARE_PKG_RESOURCES - describes shared caches | |
5957 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 5958 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
5959 | * |
5960 | * Odd one out: | |
5961 | * SD_ASYM_PACKING - describes SMT quirks | |
5962 | */ | |
5963 | #define TOPOLOGY_SD_FLAGS \ | |
5964 | (SD_SHARE_CPUPOWER | \ | |
5965 | SD_SHARE_PKG_RESOURCES | \ | |
5966 | SD_NUMA | \ | |
d77b3ed5 VG |
5967 | SD_ASYM_PACKING | \ |
5968 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
5969 | |
5970 | static struct sched_domain * | |
143e1e28 | 5971 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
5972 | { |
5973 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
5974 | int sd_weight, sd_flags = 0; |
5975 | ||
5976 | #ifdef CONFIG_NUMA | |
5977 | /* | |
5978 | * Ugly hack to pass state to sd_numa_mask()... | |
5979 | */ | |
5980 | sched_domains_curr_level = tl->numa_level; | |
5981 | #endif | |
5982 | ||
5983 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
5984 | ||
5985 | if (tl->sd_flags) | |
5986 | sd_flags = (*tl->sd_flags)(); | |
5987 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
5988 | "wrong sd_flags in topology description\n")) | |
5989 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
5990 | |
5991 | *sd = (struct sched_domain){ | |
5992 | .min_interval = sd_weight, | |
5993 | .max_interval = 2*sd_weight, | |
5994 | .busy_factor = 32, | |
870a0bb5 | 5995 | .imbalance_pct = 125, |
143e1e28 VG |
5996 | |
5997 | .cache_nice_tries = 0, | |
5998 | .busy_idx = 0, | |
5999 | .idle_idx = 0, | |
cb83b629 PZ |
6000 | .newidle_idx = 0, |
6001 | .wake_idx = 0, | |
6002 | .forkexec_idx = 0, | |
6003 | ||
6004 | .flags = 1*SD_LOAD_BALANCE | |
6005 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6006 | | 1*SD_BALANCE_EXEC |
6007 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6008 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6009 | | 1*SD_WAKE_AFFINE |
cb83b629 | 6010 | | 0*SD_SHARE_CPUPOWER |
cb83b629 | 6011 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6012 | | 0*SD_SERIALIZE |
cb83b629 | 6013 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6014 | | 0*SD_NUMA |
6015 | | sd_flags | |
cb83b629 | 6016 | , |
143e1e28 | 6017 | |
cb83b629 PZ |
6018 | .last_balance = jiffies, |
6019 | .balance_interval = sd_weight, | |
143e1e28 | 6020 | .smt_gain = 0, |
2b4cfe64 JL |
6021 | .max_newidle_lb_cost = 0, |
6022 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6023 | #ifdef CONFIG_SCHED_DEBUG |
6024 | .name = tl->name, | |
6025 | #endif | |
cb83b629 | 6026 | }; |
cb83b629 PZ |
6027 | |
6028 | /* | |
143e1e28 | 6029 | * Convert topological properties into behaviour. |
cb83b629 | 6030 | */ |
143e1e28 VG |
6031 | |
6032 | if (sd->flags & SD_SHARE_CPUPOWER) { | |
6033 | sd->imbalance_pct = 110; | |
6034 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6035 | |
6036 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6037 | sd->imbalance_pct = 117; | |
6038 | sd->cache_nice_tries = 1; | |
6039 | sd->busy_idx = 2; | |
6040 | ||
6041 | #ifdef CONFIG_NUMA | |
6042 | } else if (sd->flags & SD_NUMA) { | |
6043 | sd->cache_nice_tries = 2; | |
6044 | sd->busy_idx = 3; | |
6045 | sd->idle_idx = 2; | |
6046 | ||
6047 | sd->flags |= SD_SERIALIZE; | |
6048 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6049 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6050 | SD_BALANCE_FORK | | |
6051 | SD_WAKE_AFFINE); | |
6052 | } | |
6053 | ||
6054 | #endif | |
6055 | } else { | |
6056 | sd->flags |= SD_PREFER_SIBLING; | |
6057 | sd->cache_nice_tries = 1; | |
6058 | sd->busy_idx = 2; | |
6059 | sd->idle_idx = 1; | |
6060 | } | |
6061 | ||
6062 | sd->private = &tl->data; | |
cb83b629 PZ |
6063 | |
6064 | return sd; | |
6065 | } | |
6066 | ||
143e1e28 VG |
6067 | /* |
6068 | * Topology list, bottom-up. | |
6069 | */ | |
6070 | static struct sched_domain_topology_level default_topology[] = { | |
6071 | #ifdef CONFIG_SCHED_SMT | |
6072 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6073 | #endif | |
6074 | #ifdef CONFIG_SCHED_MC | |
6075 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6076 | #endif |
6077 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6078 | { NULL, }, | |
6079 | }; | |
6080 | ||
6081 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6082 | ||
6083 | #define for_each_sd_topology(tl) \ | |
6084 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6085 | ||
6086 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6087 | { | |
6088 | sched_domain_topology = tl; | |
6089 | } | |
6090 | ||
6091 | #ifdef CONFIG_NUMA | |
6092 | ||
cb83b629 PZ |
6093 | static const struct cpumask *sd_numa_mask(int cpu) |
6094 | { | |
6095 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6096 | } | |
6097 | ||
d039ac60 PZ |
6098 | static void sched_numa_warn(const char *str) |
6099 | { | |
6100 | static int done = false; | |
6101 | int i,j; | |
6102 | ||
6103 | if (done) | |
6104 | return; | |
6105 | ||
6106 | done = true; | |
6107 | ||
6108 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6109 | ||
6110 | for (i = 0; i < nr_node_ids; i++) { | |
6111 | printk(KERN_WARNING " "); | |
6112 | for (j = 0; j < nr_node_ids; j++) | |
6113 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6114 | printk(KERN_CONT "\n"); | |
6115 | } | |
6116 | printk(KERN_WARNING "\n"); | |
6117 | } | |
6118 | ||
6119 | static bool find_numa_distance(int distance) | |
6120 | { | |
6121 | int i; | |
6122 | ||
6123 | if (distance == node_distance(0, 0)) | |
6124 | return true; | |
6125 | ||
6126 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6127 | if (sched_domains_numa_distance[i] == distance) | |
6128 | return true; | |
6129 | } | |
6130 | ||
6131 | return false; | |
6132 | } | |
6133 | ||
cb83b629 PZ |
6134 | static void sched_init_numa(void) |
6135 | { | |
6136 | int next_distance, curr_distance = node_distance(0, 0); | |
6137 | struct sched_domain_topology_level *tl; | |
6138 | int level = 0; | |
6139 | int i, j, k; | |
6140 | ||
cb83b629 PZ |
6141 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6142 | if (!sched_domains_numa_distance) | |
6143 | return; | |
6144 | ||
6145 | /* | |
6146 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6147 | * unique distances in the node_distance() table. | |
6148 | * | |
6149 | * Assumes node_distance(0,j) includes all distances in | |
6150 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6151 | */ |
6152 | next_distance = curr_distance; | |
6153 | for (i = 0; i < nr_node_ids; i++) { | |
6154 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6155 | for (k = 0; k < nr_node_ids; k++) { |
6156 | int distance = node_distance(i, k); | |
6157 | ||
6158 | if (distance > curr_distance && | |
6159 | (distance < next_distance || | |
6160 | next_distance == curr_distance)) | |
6161 | next_distance = distance; | |
6162 | ||
6163 | /* | |
6164 | * While not a strong assumption it would be nice to know | |
6165 | * about cases where if node A is connected to B, B is not | |
6166 | * equally connected to A. | |
6167 | */ | |
6168 | if (sched_debug() && node_distance(k, i) != distance) | |
6169 | sched_numa_warn("Node-distance not symmetric"); | |
6170 | ||
6171 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6172 | sched_numa_warn("Node-0 not representative"); | |
6173 | } | |
6174 | if (next_distance != curr_distance) { | |
6175 | sched_domains_numa_distance[level++] = next_distance; | |
6176 | sched_domains_numa_levels = level; | |
6177 | curr_distance = next_distance; | |
6178 | } else break; | |
cb83b629 | 6179 | } |
d039ac60 PZ |
6180 | |
6181 | /* | |
6182 | * In case of sched_debug() we verify the above assumption. | |
6183 | */ | |
6184 | if (!sched_debug()) | |
6185 | break; | |
cb83b629 PZ |
6186 | } |
6187 | /* | |
6188 | * 'level' contains the number of unique distances, excluding the | |
6189 | * identity distance node_distance(i,i). | |
6190 | * | |
28b4a521 | 6191 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6192 | * numbers. |
6193 | */ | |
6194 | ||
5f7865f3 TC |
6195 | /* |
6196 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6197 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6198 | * the array will contain less then 'level' members. This could be | |
6199 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6200 | * in other functions. | |
6201 | * | |
6202 | * We reset it to 'level' at the end of this function. | |
6203 | */ | |
6204 | sched_domains_numa_levels = 0; | |
6205 | ||
cb83b629 PZ |
6206 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6207 | if (!sched_domains_numa_masks) | |
6208 | return; | |
6209 | ||
6210 | /* | |
6211 | * Now for each level, construct a mask per node which contains all | |
6212 | * cpus of nodes that are that many hops away from us. | |
6213 | */ | |
6214 | for (i = 0; i < level; i++) { | |
6215 | sched_domains_numa_masks[i] = | |
6216 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6217 | if (!sched_domains_numa_masks[i]) | |
6218 | return; | |
6219 | ||
6220 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6221 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6222 | if (!mask) |
6223 | return; | |
6224 | ||
6225 | sched_domains_numa_masks[i][j] = mask; | |
6226 | ||
6227 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6228 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6229 | continue; |
6230 | ||
6231 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6232 | } | |
6233 | } | |
6234 | } | |
6235 | ||
143e1e28 VG |
6236 | /* Compute default topology size */ |
6237 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6238 | ||
6239 | tl = kzalloc((i + level) * | |
cb83b629 PZ |
6240 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6241 | if (!tl) | |
6242 | return; | |
6243 | ||
6244 | /* | |
6245 | * Copy the default topology bits.. | |
6246 | */ | |
143e1e28 VG |
6247 | for (i = 0; sched_domain_topology[i].mask; i++) |
6248 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6249 | |
6250 | /* | |
6251 | * .. and append 'j' levels of NUMA goodness. | |
6252 | */ | |
6253 | for (j = 0; j < level; i++, j++) { | |
6254 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6255 | .mask = sd_numa_mask, |
143e1e28 | 6256 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6257 | .flags = SDTL_OVERLAP, |
6258 | .numa_level = j, | |
143e1e28 | 6259 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6260 | }; |
6261 | } | |
6262 | ||
6263 | sched_domain_topology = tl; | |
5f7865f3 TC |
6264 | |
6265 | sched_domains_numa_levels = level; | |
cb83b629 | 6266 | } |
301a5cba TC |
6267 | |
6268 | static void sched_domains_numa_masks_set(int cpu) | |
6269 | { | |
6270 | int i, j; | |
6271 | int node = cpu_to_node(cpu); | |
6272 | ||
6273 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6274 | for (j = 0; j < nr_node_ids; j++) { | |
6275 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6276 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6277 | } | |
6278 | } | |
6279 | } | |
6280 | ||
6281 | static void sched_domains_numa_masks_clear(int cpu) | |
6282 | { | |
6283 | int i, j; | |
6284 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6285 | for (j = 0; j < nr_node_ids; j++) | |
6286 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6287 | } | |
6288 | } | |
6289 | ||
6290 | /* | |
6291 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6292 | * are onlined. | |
6293 | */ | |
6294 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6295 | unsigned long action, | |
6296 | void *hcpu) | |
6297 | { | |
6298 | int cpu = (long)hcpu; | |
6299 | ||
6300 | switch (action & ~CPU_TASKS_FROZEN) { | |
6301 | case CPU_ONLINE: | |
6302 | sched_domains_numa_masks_set(cpu); | |
6303 | break; | |
6304 | ||
6305 | case CPU_DEAD: | |
6306 | sched_domains_numa_masks_clear(cpu); | |
6307 | break; | |
6308 | ||
6309 | default: | |
6310 | return NOTIFY_DONE; | |
6311 | } | |
6312 | ||
6313 | return NOTIFY_OK; | |
cb83b629 PZ |
6314 | } |
6315 | #else | |
6316 | static inline void sched_init_numa(void) | |
6317 | { | |
6318 | } | |
301a5cba TC |
6319 | |
6320 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6321 | unsigned long action, | |
6322 | void *hcpu) | |
6323 | { | |
6324 | return 0; | |
6325 | } | |
cb83b629 PZ |
6326 | #endif /* CONFIG_NUMA */ |
6327 | ||
54ab4ff4 PZ |
6328 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6329 | { | |
6330 | struct sched_domain_topology_level *tl; | |
6331 | int j; | |
6332 | ||
27723a68 | 6333 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6334 | struct sd_data *sdd = &tl->data; |
6335 | ||
6336 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6337 | if (!sdd->sd) | |
6338 | return -ENOMEM; | |
6339 | ||
6340 | sdd->sg = alloc_percpu(struct sched_group *); | |
6341 | if (!sdd->sg) | |
6342 | return -ENOMEM; | |
6343 | ||
9c3f75cb PZ |
6344 | sdd->sgp = alloc_percpu(struct sched_group_power *); |
6345 | if (!sdd->sgp) | |
6346 | return -ENOMEM; | |
6347 | ||
54ab4ff4 PZ |
6348 | for_each_cpu(j, cpu_map) { |
6349 | struct sched_domain *sd; | |
6350 | struct sched_group *sg; | |
9c3f75cb | 6351 | struct sched_group_power *sgp; |
54ab4ff4 PZ |
6352 | |
6353 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6354 | GFP_KERNEL, cpu_to_node(j)); | |
6355 | if (!sd) | |
6356 | return -ENOMEM; | |
6357 | ||
6358 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6359 | ||
6360 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6361 | GFP_KERNEL, cpu_to_node(j)); | |
6362 | if (!sg) | |
6363 | return -ENOMEM; | |
6364 | ||
30b4e9eb IM |
6365 | sg->next = sg; |
6366 | ||
54ab4ff4 | 6367 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6368 | |
c1174876 | 6369 | sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(), |
9c3f75cb PZ |
6370 | GFP_KERNEL, cpu_to_node(j)); |
6371 | if (!sgp) | |
6372 | return -ENOMEM; | |
6373 | ||
6374 | *per_cpu_ptr(sdd->sgp, j) = sgp; | |
54ab4ff4 PZ |
6375 | } |
6376 | } | |
6377 | ||
6378 | return 0; | |
6379 | } | |
6380 | ||
6381 | static void __sdt_free(const struct cpumask *cpu_map) | |
6382 | { | |
6383 | struct sched_domain_topology_level *tl; | |
6384 | int j; | |
6385 | ||
27723a68 | 6386 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6387 | struct sd_data *sdd = &tl->data; |
6388 | ||
6389 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6390 | struct sched_domain *sd; |
6391 | ||
6392 | if (sdd->sd) { | |
6393 | sd = *per_cpu_ptr(sdd->sd, j); | |
6394 | if (sd && (sd->flags & SD_OVERLAP)) | |
6395 | free_sched_groups(sd->groups, 0); | |
6396 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6397 | } | |
6398 | ||
6399 | if (sdd->sg) | |
6400 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
6401 | if (sdd->sgp) | |
6402 | kfree(*per_cpu_ptr(sdd->sgp, j)); | |
54ab4ff4 PZ |
6403 | } |
6404 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6405 | sdd->sd = NULL; |
54ab4ff4 | 6406 | free_percpu(sdd->sg); |
fb2cf2c6 | 6407 | sdd->sg = NULL; |
9c3f75cb | 6408 | free_percpu(sdd->sgp); |
fb2cf2c6 | 6409 | sdd->sgp = NULL; |
54ab4ff4 PZ |
6410 | } |
6411 | } | |
6412 | ||
2c402dc3 | 6413 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6414 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6415 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6416 | { |
143e1e28 | 6417 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6418 | if (!sd) |
d069b916 | 6419 | return child; |
2c402dc3 | 6420 | |
2c402dc3 | 6421 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6422 | if (child) { |
6423 | sd->level = child->level + 1; | |
6424 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6425 | child->parent = sd; |
c75e0128 | 6426 | sd->child = child; |
60495e77 | 6427 | } |
a841f8ce | 6428 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6429 | |
6430 | return sd; | |
6431 | } | |
6432 | ||
2109b99e AH |
6433 | /* |
6434 | * Build sched domains for a given set of cpus and attach the sched domains | |
6435 | * to the individual cpus | |
6436 | */ | |
dce840a0 PZ |
6437 | static int build_sched_domains(const struct cpumask *cpu_map, |
6438 | struct sched_domain_attr *attr) | |
2109b99e | 6439 | { |
1c632169 | 6440 | enum s_alloc alloc_state; |
dce840a0 | 6441 | struct sched_domain *sd; |
2109b99e | 6442 | struct s_data d; |
822ff793 | 6443 | int i, ret = -ENOMEM; |
9c1cfda2 | 6444 | |
2109b99e AH |
6445 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6446 | if (alloc_state != sa_rootdomain) | |
6447 | goto error; | |
9c1cfda2 | 6448 | |
dce840a0 | 6449 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6450 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6451 | struct sched_domain_topology_level *tl; |
6452 | ||
3bd65a80 | 6453 | sd = NULL; |
27723a68 | 6454 | for_each_sd_topology(tl) { |
4a850cbe | 6455 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6456 | if (tl == sched_domain_topology) |
6457 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6458 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6459 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6460 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6461 | break; | |
e3589f6c | 6462 | } |
dce840a0 PZ |
6463 | } |
6464 | ||
6465 | /* Build the groups for the domains */ | |
6466 | for_each_cpu(i, cpu_map) { | |
6467 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6468 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6469 | if (sd->flags & SD_OVERLAP) { |
6470 | if (build_overlap_sched_groups(sd, i)) | |
6471 | goto error; | |
6472 | } else { | |
6473 | if (build_sched_groups(sd, i)) | |
6474 | goto error; | |
6475 | } | |
1cf51902 | 6476 | } |
a06dadbe | 6477 | } |
9c1cfda2 | 6478 | |
1da177e4 | 6479 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
6480 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6481 | if (!cpumask_test_cpu(i, cpu_map)) | |
6482 | continue; | |
9c1cfda2 | 6483 | |
dce840a0 PZ |
6484 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6485 | claim_allocations(i, sd); | |
cd4ea6ae | 6486 | init_sched_groups_power(i, sd); |
dce840a0 | 6487 | } |
f712c0c7 | 6488 | } |
9c1cfda2 | 6489 | |
1da177e4 | 6490 | /* Attach the domains */ |
dce840a0 | 6491 | rcu_read_lock(); |
abcd083a | 6492 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6493 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6494 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6495 | } |
dce840a0 | 6496 | rcu_read_unlock(); |
51888ca2 | 6497 | |
822ff793 | 6498 | ret = 0; |
51888ca2 | 6499 | error: |
2109b99e | 6500 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6501 | return ret; |
1da177e4 | 6502 | } |
029190c5 | 6503 | |
acc3f5d7 | 6504 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6505 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6506 | static struct sched_domain_attr *dattr_cur; |
6507 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6508 | |
6509 | /* | |
6510 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6511 | * cpumask) fails, then fallback to a single sched domain, |
6512 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6513 | */ |
4212823f | 6514 | static cpumask_var_t fallback_doms; |
029190c5 | 6515 | |
ee79d1bd HC |
6516 | /* |
6517 | * arch_update_cpu_topology lets virtualized architectures update the | |
6518 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6519 | * or 0 if it stayed the same. | |
6520 | */ | |
52f5684c | 6521 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6522 | { |
ee79d1bd | 6523 | return 0; |
22e52b07 HC |
6524 | } |
6525 | ||
acc3f5d7 RR |
6526 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6527 | { | |
6528 | int i; | |
6529 | cpumask_var_t *doms; | |
6530 | ||
6531 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6532 | if (!doms) | |
6533 | return NULL; | |
6534 | for (i = 0; i < ndoms; i++) { | |
6535 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6536 | free_sched_domains(doms, i); | |
6537 | return NULL; | |
6538 | } | |
6539 | } | |
6540 | return doms; | |
6541 | } | |
6542 | ||
6543 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6544 | { | |
6545 | unsigned int i; | |
6546 | for (i = 0; i < ndoms; i++) | |
6547 | free_cpumask_var(doms[i]); | |
6548 | kfree(doms); | |
6549 | } | |
6550 | ||
1a20ff27 | 6551 | /* |
41a2d6cf | 6552 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6553 | * For now this just excludes isolated cpus, but could be used to |
6554 | * exclude other special cases in the future. | |
1a20ff27 | 6555 | */ |
c4a8849a | 6556 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6557 | { |
7378547f MM |
6558 | int err; |
6559 | ||
22e52b07 | 6560 | arch_update_cpu_topology(); |
029190c5 | 6561 | ndoms_cur = 1; |
acc3f5d7 | 6562 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6563 | if (!doms_cur) |
acc3f5d7 RR |
6564 | doms_cur = &fallback_doms; |
6565 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6566 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6567 | register_sched_domain_sysctl(); |
7378547f MM |
6568 | |
6569 | return err; | |
1a20ff27 DG |
6570 | } |
6571 | ||
1a20ff27 DG |
6572 | /* |
6573 | * Detach sched domains from a group of cpus specified in cpu_map | |
6574 | * These cpus will now be attached to the NULL domain | |
6575 | */ | |
96f874e2 | 6576 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6577 | { |
6578 | int i; | |
6579 | ||
dce840a0 | 6580 | rcu_read_lock(); |
abcd083a | 6581 | for_each_cpu(i, cpu_map) |
57d885fe | 6582 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6583 | rcu_read_unlock(); |
1a20ff27 DG |
6584 | } |
6585 | ||
1d3504fc HS |
6586 | /* handle null as "default" */ |
6587 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6588 | struct sched_domain_attr *new, int idx_new) | |
6589 | { | |
6590 | struct sched_domain_attr tmp; | |
6591 | ||
6592 | /* fast path */ | |
6593 | if (!new && !cur) | |
6594 | return 1; | |
6595 | ||
6596 | tmp = SD_ATTR_INIT; | |
6597 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6598 | new ? (new + idx_new) : &tmp, | |
6599 | sizeof(struct sched_domain_attr)); | |
6600 | } | |
6601 | ||
029190c5 PJ |
6602 | /* |
6603 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6604 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6605 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6606 | * It destroys each deleted domain and builds each new domain. | |
6607 | * | |
acc3f5d7 | 6608 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6609 | * The masks don't intersect (don't overlap.) We should setup one |
6610 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6611 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6612 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6613 | * it as it is. | |
6614 | * | |
acc3f5d7 RR |
6615 | * The passed in 'doms_new' should be allocated using |
6616 | * alloc_sched_domains. This routine takes ownership of it and will | |
6617 | * free_sched_domains it when done with it. If the caller failed the | |
6618 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6619 | * and partition_sched_domains() will fallback to the single partition | |
6620 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6621 | * |
96f874e2 | 6622 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6623 | * ndoms_new == 0 is a special case for destroying existing domains, |
6624 | * and it will not create the default domain. | |
dfb512ec | 6625 | * |
029190c5 PJ |
6626 | * Call with hotplug lock held |
6627 | */ | |
acc3f5d7 | 6628 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6629 | struct sched_domain_attr *dattr_new) |
029190c5 | 6630 | { |
dfb512ec | 6631 | int i, j, n; |
d65bd5ec | 6632 | int new_topology; |
029190c5 | 6633 | |
712555ee | 6634 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6635 | |
7378547f MM |
6636 | /* always unregister in case we don't destroy any domains */ |
6637 | unregister_sched_domain_sysctl(); | |
6638 | ||
d65bd5ec HC |
6639 | /* Let architecture update cpu core mappings. */ |
6640 | new_topology = arch_update_cpu_topology(); | |
6641 | ||
dfb512ec | 6642 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6643 | |
6644 | /* Destroy deleted domains */ | |
6645 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6646 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6647 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6648 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6649 | goto match1; |
6650 | } | |
6651 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6652 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6653 | match1: |
6654 | ; | |
6655 | } | |
6656 | ||
c8d2d47a | 6657 | n = ndoms_cur; |
e761b772 | 6658 | if (doms_new == NULL) { |
c8d2d47a | 6659 | n = 0; |
acc3f5d7 | 6660 | doms_new = &fallback_doms; |
6ad4c188 | 6661 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6662 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6663 | } |
6664 | ||
029190c5 PJ |
6665 | /* Build new domains */ |
6666 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6667 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6668 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6669 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6670 | goto match2; |
6671 | } | |
6672 | /* no match - add a new doms_new */ | |
dce840a0 | 6673 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6674 | match2: |
6675 | ; | |
6676 | } | |
6677 | ||
6678 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6679 | if (doms_cur != &fallback_doms) |
6680 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6681 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6682 | doms_cur = doms_new; |
1d3504fc | 6683 | dattr_cur = dattr_new; |
029190c5 | 6684 | ndoms_cur = ndoms_new; |
7378547f MM |
6685 | |
6686 | register_sched_domain_sysctl(); | |
a1835615 | 6687 | |
712555ee | 6688 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6689 | } |
6690 | ||
d35be8ba SB |
6691 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6692 | ||
1da177e4 | 6693 | /* |
3a101d05 TH |
6694 | * Update cpusets according to cpu_active mask. If cpusets are |
6695 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6696 | * around partition_sched_domains(). | |
d35be8ba SB |
6697 | * |
6698 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6699 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6700 | */ |
0b2e918a TH |
6701 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6702 | void *hcpu) | |
e761b772 | 6703 | { |
d35be8ba SB |
6704 | switch (action) { |
6705 | case CPU_ONLINE_FROZEN: | |
6706 | case CPU_DOWN_FAILED_FROZEN: | |
6707 | ||
6708 | /* | |
6709 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6710 | * resume sequence. As long as this is not the last online | |
6711 | * operation in the resume sequence, just build a single sched | |
6712 | * domain, ignoring cpusets. | |
6713 | */ | |
6714 | num_cpus_frozen--; | |
6715 | if (likely(num_cpus_frozen)) { | |
6716 | partition_sched_domains(1, NULL, NULL); | |
6717 | break; | |
6718 | } | |
6719 | ||
6720 | /* | |
6721 | * This is the last CPU online operation. So fall through and | |
6722 | * restore the original sched domains by considering the | |
6723 | * cpuset configurations. | |
6724 | */ | |
6725 | ||
e761b772 | 6726 | case CPU_ONLINE: |
6ad4c188 | 6727 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6728 | cpuset_update_active_cpus(true); |
d35be8ba | 6729 | break; |
3a101d05 TH |
6730 | default: |
6731 | return NOTIFY_DONE; | |
6732 | } | |
d35be8ba | 6733 | return NOTIFY_OK; |
3a101d05 | 6734 | } |
e761b772 | 6735 | |
0b2e918a TH |
6736 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6737 | void *hcpu) | |
3a101d05 | 6738 | { |
d35be8ba | 6739 | switch (action) { |
3a101d05 | 6740 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6741 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6742 | break; |
6743 | case CPU_DOWN_PREPARE_FROZEN: | |
6744 | num_cpus_frozen++; | |
6745 | partition_sched_domains(1, NULL, NULL); | |
6746 | break; | |
e761b772 MK |
6747 | default: |
6748 | return NOTIFY_DONE; | |
6749 | } | |
d35be8ba | 6750 | return NOTIFY_OK; |
e761b772 | 6751 | } |
e761b772 | 6752 | |
1da177e4 LT |
6753 | void __init sched_init_smp(void) |
6754 | { | |
dcc30a35 RR |
6755 | cpumask_var_t non_isolated_cpus; |
6756 | ||
6757 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6758 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6759 | |
cb83b629 PZ |
6760 | sched_init_numa(); |
6761 | ||
6acce3ef PZ |
6762 | /* |
6763 | * There's no userspace yet to cause hotplug operations; hence all the | |
6764 | * cpu masks are stable and all blatant races in the below code cannot | |
6765 | * happen. | |
6766 | */ | |
712555ee | 6767 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6768 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6769 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6770 | if (cpumask_empty(non_isolated_cpus)) | |
6771 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6772 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6773 | |
301a5cba | 6774 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6775 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6776 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6777 | |
b328ca18 | 6778 | init_hrtick(); |
5c1e1767 NP |
6779 | |
6780 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6781 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6782 | BUG(); |
19978ca6 | 6783 | sched_init_granularity(); |
dcc30a35 | 6784 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6785 | |
0e3900e6 | 6786 | init_sched_rt_class(); |
1baca4ce | 6787 | init_sched_dl_class(); |
1da177e4 LT |
6788 | } |
6789 | #else | |
6790 | void __init sched_init_smp(void) | |
6791 | { | |
19978ca6 | 6792 | sched_init_granularity(); |
1da177e4 LT |
6793 | } |
6794 | #endif /* CONFIG_SMP */ | |
6795 | ||
cd1bb94b AB |
6796 | const_debug unsigned int sysctl_timer_migration = 1; |
6797 | ||
1da177e4 LT |
6798 | int in_sched_functions(unsigned long addr) |
6799 | { | |
1da177e4 LT |
6800 | return in_lock_functions(addr) || |
6801 | (addr >= (unsigned long)__sched_text_start | |
6802 | && addr < (unsigned long)__sched_text_end); | |
6803 | } | |
6804 | ||
029632fb | 6805 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6806 | /* |
6807 | * Default task group. | |
6808 | * Every task in system belongs to this group at bootup. | |
6809 | */ | |
029632fb | 6810 | struct task_group root_task_group; |
35cf4e50 | 6811 | LIST_HEAD(task_groups); |
052f1dc7 | 6812 | #endif |
6f505b16 | 6813 | |
e6252c3e | 6814 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6815 | |
1da177e4 LT |
6816 | void __init sched_init(void) |
6817 | { | |
dd41f596 | 6818 | int i, j; |
434d53b0 MT |
6819 | unsigned long alloc_size = 0, ptr; |
6820 | ||
6821 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6822 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6823 | #endif | |
6824 | #ifdef CONFIG_RT_GROUP_SCHED | |
6825 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6826 | #endif |
df7c8e84 | 6827 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6828 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6829 | #endif |
434d53b0 | 6830 | if (alloc_size) { |
36b7b6d4 | 6831 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6832 | |
6833 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6834 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6835 | ptr += nr_cpu_ids * sizeof(void **); |
6836 | ||
07e06b01 | 6837 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6838 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6839 | |
6d6bc0ad | 6840 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6841 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6842 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6843 | ptr += nr_cpu_ids * sizeof(void **); |
6844 | ||
07e06b01 | 6845 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6846 | ptr += nr_cpu_ids * sizeof(void **); |
6847 | ||
6d6bc0ad | 6848 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6849 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6850 | for_each_possible_cpu(i) { | |
e6252c3e | 6851 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6852 | ptr += cpumask_size(); |
6853 | } | |
6854 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6855 | } |
dd41f596 | 6856 | |
332ac17e DF |
6857 | init_rt_bandwidth(&def_rt_bandwidth, |
6858 | global_rt_period(), global_rt_runtime()); | |
6859 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6860 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6861 | |
57d885fe GH |
6862 | #ifdef CONFIG_SMP |
6863 | init_defrootdomain(); | |
6864 | #endif | |
6865 | ||
d0b27fa7 | 6866 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6867 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6868 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6869 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6870 | |
7c941438 | 6871 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6872 | list_add(&root_task_group.list, &task_groups); |
6873 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6874 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6875 | autogroup_init(&init_task); |
54c707e9 | 6876 | |
7c941438 | 6877 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6878 | |
0a945022 | 6879 | for_each_possible_cpu(i) { |
70b97a7f | 6880 | struct rq *rq; |
1da177e4 LT |
6881 | |
6882 | rq = cpu_rq(i); | |
05fa785c | 6883 | raw_spin_lock_init(&rq->lock); |
7897986b | 6884 | rq->nr_running = 0; |
dce48a84 TG |
6885 | rq->calc_load_active = 0; |
6886 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6887 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6888 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6889 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6890 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6891 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6892 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6893 | /* |
07e06b01 | 6894 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6895 | * |
6896 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6897 | * gets 100% of the cpu resources in the system. This overall | |
6898 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6899 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6900 | * based on each entity's (task or task-group's) weight |
6901 | * (se->load.weight). | |
6902 | * | |
07e06b01 | 6903 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6904 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6905 | * then A0's share of the cpu resource is: | |
6906 | * | |
0d905bca | 6907 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6908 | * |
07e06b01 YZ |
6909 | * We achieve this by letting root_task_group's tasks sit |
6910 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6911 | */ |
ab84d31e | 6912 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6913 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6914 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6915 | ||
6916 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6917 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6918 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6919 | #endif |
1da177e4 | 6920 | |
dd41f596 IM |
6921 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6922 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6923 | |
6924 | rq->last_load_update_tick = jiffies; | |
6925 | ||
1da177e4 | 6926 | #ifdef CONFIG_SMP |
41c7ce9a | 6927 | rq->sd = NULL; |
57d885fe | 6928 | rq->rd = NULL; |
1399fa78 | 6929 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 6930 | rq->post_schedule = 0; |
1da177e4 | 6931 | rq->active_balance = 0; |
dd41f596 | 6932 | rq->next_balance = jiffies; |
1da177e4 | 6933 | rq->push_cpu = 0; |
0a2966b4 | 6934 | rq->cpu = i; |
1f11eb6a | 6935 | rq->online = 0; |
eae0c9df MG |
6936 | rq->idle_stamp = 0; |
6937 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6938 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6939 | |
6940 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6941 | ||
dc938520 | 6942 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6943 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6944 | rq->nohz_flags = 0; |
83cd4fe2 | 6945 | #endif |
265f22a9 FW |
6946 | #ifdef CONFIG_NO_HZ_FULL |
6947 | rq->last_sched_tick = 0; | |
6948 | #endif | |
1da177e4 | 6949 | #endif |
8f4d37ec | 6950 | init_rq_hrtick(rq); |
1da177e4 | 6951 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6952 | } |
6953 | ||
2dd73a4f | 6954 | set_load_weight(&init_task); |
b50f60ce | 6955 | |
e107be36 AK |
6956 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6957 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6958 | #endif | |
6959 | ||
1da177e4 LT |
6960 | /* |
6961 | * The boot idle thread does lazy MMU switching as well: | |
6962 | */ | |
6963 | atomic_inc(&init_mm.mm_count); | |
6964 | enter_lazy_tlb(&init_mm, current); | |
6965 | ||
6966 | /* | |
6967 | * Make us the idle thread. Technically, schedule() should not be | |
6968 | * called from this thread, however somewhere below it might be, | |
6969 | * but because we are the idle thread, we just pick up running again | |
6970 | * when this runqueue becomes "idle". | |
6971 | */ | |
6972 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6973 | |
6974 | calc_load_update = jiffies + LOAD_FREQ; | |
6975 | ||
dd41f596 IM |
6976 | /* |
6977 | * During early bootup we pretend to be a normal task: | |
6978 | */ | |
6979 | current->sched_class = &fair_sched_class; | |
6892b75e | 6980 | |
bf4d83f6 | 6981 | #ifdef CONFIG_SMP |
4cb98839 | 6982 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6983 | /* May be allocated at isolcpus cmdline parse time */ |
6984 | if (cpu_isolated_map == NULL) | |
6985 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6986 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6987 | #endif |
6988 | init_sched_fair_class(); | |
6a7b3dc3 | 6989 | |
6892b75e | 6990 | scheduler_running = 1; |
1da177e4 LT |
6991 | } |
6992 | ||
d902db1e | 6993 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6994 | static inline int preempt_count_equals(int preempt_offset) |
6995 | { | |
234da7bc | 6996 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6997 | |
4ba8216c | 6998 | return (nested == preempt_offset); |
e4aafea2 FW |
6999 | } |
7000 | ||
d894837f | 7001 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7002 | { |
1da177e4 LT |
7003 | static unsigned long prev_jiffy; /* ratelimiting */ |
7004 | ||
b3fbab05 | 7005 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7006 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7007 | !is_idle_task(current)) || | |
e4aafea2 | 7008 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7009 | return; |
7010 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7011 | return; | |
7012 | prev_jiffy = jiffies; | |
7013 | ||
3df0fc5b PZ |
7014 | printk(KERN_ERR |
7015 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7016 | file, line); | |
7017 | printk(KERN_ERR | |
7018 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7019 | in_atomic(), irqs_disabled(), | |
7020 | current->pid, current->comm); | |
aef745fc IM |
7021 | |
7022 | debug_show_held_locks(current); | |
7023 | if (irqs_disabled()) | |
7024 | print_irqtrace_events(current); | |
8f47b187 TG |
7025 | #ifdef CONFIG_DEBUG_PREEMPT |
7026 | if (!preempt_count_equals(preempt_offset)) { | |
7027 | pr_err("Preemption disabled at:"); | |
7028 | print_ip_sym(current->preempt_disable_ip); | |
7029 | pr_cont("\n"); | |
7030 | } | |
7031 | #endif | |
aef745fc | 7032 | dump_stack(); |
1da177e4 LT |
7033 | } |
7034 | EXPORT_SYMBOL(__might_sleep); | |
7035 | #endif | |
7036 | ||
7037 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7038 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7039 | { | |
da7a735e | 7040 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7041 | struct sched_attr attr = { |
7042 | .sched_policy = SCHED_NORMAL, | |
7043 | }; | |
da7a735e | 7044 | int old_prio = p->prio; |
3a5e4dc1 | 7045 | int on_rq; |
3e51f33f | 7046 | |
fd2f4419 | 7047 | on_rq = p->on_rq; |
3a5e4dc1 | 7048 | if (on_rq) |
4ca9b72b | 7049 | dequeue_task(rq, p, 0); |
d50dde5a | 7050 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 7051 | if (on_rq) { |
4ca9b72b | 7052 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7053 | resched_task(rq->curr); |
7054 | } | |
da7a735e PZ |
7055 | |
7056 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7057 | } |
7058 | ||
1da177e4 LT |
7059 | void normalize_rt_tasks(void) |
7060 | { | |
a0f98a1c | 7061 | struct task_struct *g, *p; |
1da177e4 | 7062 | unsigned long flags; |
70b97a7f | 7063 | struct rq *rq; |
1da177e4 | 7064 | |
4cf5d77a | 7065 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7066 | do_each_thread(g, p) { |
178be793 IM |
7067 | /* |
7068 | * Only normalize user tasks: | |
7069 | */ | |
7070 | if (!p->mm) | |
7071 | continue; | |
7072 | ||
6cfb0d5d | 7073 | p->se.exec_start = 0; |
6cfb0d5d | 7074 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7075 | p->se.statistics.wait_start = 0; |
7076 | p->se.statistics.sleep_start = 0; | |
7077 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7078 | #endif |
dd41f596 | 7079 | |
aab03e05 | 7080 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7081 | /* |
7082 | * Renice negative nice level userspace | |
7083 | * tasks back to 0: | |
7084 | */ | |
d0ea0268 | 7085 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7086 | set_user_nice(p, 0); |
1da177e4 | 7087 | continue; |
dd41f596 | 7088 | } |
1da177e4 | 7089 | |
1d615482 | 7090 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7091 | rq = __task_rq_lock(p); |
1da177e4 | 7092 | |
178be793 | 7093 | normalize_task(rq, p); |
3a5e4dc1 | 7094 | |
b29739f9 | 7095 | __task_rq_unlock(rq); |
1d615482 | 7096 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7097 | } while_each_thread(g, p); |
7098 | ||
4cf5d77a | 7099 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7100 | } |
7101 | ||
7102 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7103 | |
67fc4e0c | 7104 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7105 | /* |
67fc4e0c | 7106 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7107 | * |
7108 | * They can only be called when the whole system has been | |
7109 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7110 | * activity can take place. Using them for anything else would | |
7111 | * be a serious bug, and as a result, they aren't even visible | |
7112 | * under any other configuration. | |
7113 | */ | |
7114 | ||
7115 | /** | |
7116 | * curr_task - return the current task for a given cpu. | |
7117 | * @cpu: the processor in question. | |
7118 | * | |
7119 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7120 | * |
7121 | * Return: The current task for @cpu. | |
1df5c10a | 7122 | */ |
36c8b586 | 7123 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7124 | { |
7125 | return cpu_curr(cpu); | |
7126 | } | |
7127 | ||
67fc4e0c JW |
7128 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7129 | ||
7130 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7131 | /** |
7132 | * set_curr_task - set the current task for a given cpu. | |
7133 | * @cpu: the processor in question. | |
7134 | * @p: the task pointer to set. | |
7135 | * | |
7136 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7137 | * are serviced on a separate stack. It allows the architecture to switch the |
7138 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7139 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7140 | * and caller must save the original value of the current task (see | |
7141 | * curr_task() above) and restore that value before reenabling interrupts and | |
7142 | * re-starting the system. | |
7143 | * | |
7144 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7145 | */ | |
36c8b586 | 7146 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7147 | { |
7148 | cpu_curr(cpu) = p; | |
7149 | } | |
7150 | ||
7151 | #endif | |
29f59db3 | 7152 | |
7c941438 | 7153 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7154 | /* task_group_lock serializes the addition/removal of task groups */ |
7155 | static DEFINE_SPINLOCK(task_group_lock); | |
7156 | ||
bccbe08a PZ |
7157 | static void free_sched_group(struct task_group *tg) |
7158 | { | |
7159 | free_fair_sched_group(tg); | |
7160 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7161 | autogroup_free(tg); |
bccbe08a PZ |
7162 | kfree(tg); |
7163 | } | |
7164 | ||
7165 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7166 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7167 | { |
7168 | struct task_group *tg; | |
bccbe08a PZ |
7169 | |
7170 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7171 | if (!tg) | |
7172 | return ERR_PTR(-ENOMEM); | |
7173 | ||
ec7dc8ac | 7174 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7175 | goto err; |
7176 | ||
ec7dc8ac | 7177 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7178 | goto err; |
7179 | ||
ace783b9 LZ |
7180 | return tg; |
7181 | ||
7182 | err: | |
7183 | free_sched_group(tg); | |
7184 | return ERR_PTR(-ENOMEM); | |
7185 | } | |
7186 | ||
7187 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7188 | { | |
7189 | unsigned long flags; | |
7190 | ||
8ed36996 | 7191 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7192 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7193 | |
7194 | WARN_ON(!parent); /* root should already exist */ | |
7195 | ||
7196 | tg->parent = parent; | |
f473aa5e | 7197 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7198 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7199 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7200 | } |
7201 | ||
9b5b7751 | 7202 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7203 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7204 | { |
29f59db3 | 7205 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7206 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7207 | } |
7208 | ||
9b5b7751 | 7209 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7210 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7211 | { |
7212 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7213 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7214 | } | |
7215 | ||
7216 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7217 | { |
8ed36996 | 7218 | unsigned long flags; |
9b5b7751 | 7219 | int i; |
29f59db3 | 7220 | |
3d4b47b4 PZ |
7221 | /* end participation in shares distribution */ |
7222 | for_each_possible_cpu(i) | |
bccbe08a | 7223 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7224 | |
7225 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7226 | list_del_rcu(&tg->list); |
f473aa5e | 7227 | list_del_rcu(&tg->siblings); |
8ed36996 | 7228 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7229 | } |
7230 | ||
9b5b7751 | 7231 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7232 | * The caller of this function should have put the task in its new group |
7233 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7234 | * reflect its new group. | |
9b5b7751 SV |
7235 | */ |
7236 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7237 | { |
8323f26c | 7238 | struct task_group *tg; |
29f59db3 SV |
7239 | int on_rq, running; |
7240 | unsigned long flags; | |
7241 | struct rq *rq; | |
7242 | ||
7243 | rq = task_rq_lock(tsk, &flags); | |
7244 | ||
051a1d1a | 7245 | running = task_current(rq, tsk); |
fd2f4419 | 7246 | on_rq = tsk->on_rq; |
29f59db3 | 7247 | |
0e1f3483 | 7248 | if (on_rq) |
29f59db3 | 7249 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7250 | if (unlikely(running)) |
7251 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7252 | |
073219e9 | 7253 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7254 | lockdep_is_held(&tsk->sighand->siglock)), |
7255 | struct task_group, css); | |
7256 | tg = autogroup_task_group(tsk, tg); | |
7257 | tsk->sched_task_group = tg; | |
7258 | ||
810b3817 | 7259 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7260 | if (tsk->sched_class->task_move_group) |
7261 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7262 | else | |
810b3817 | 7263 | #endif |
b2b5ce02 | 7264 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7265 | |
0e1f3483 HS |
7266 | if (unlikely(running)) |
7267 | tsk->sched_class->set_curr_task(rq); | |
7268 | if (on_rq) | |
371fd7e7 | 7269 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7270 | |
0122ec5b | 7271 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7272 | } |
7c941438 | 7273 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7274 | |
a790de99 PT |
7275 | #ifdef CONFIG_RT_GROUP_SCHED |
7276 | /* | |
7277 | * Ensure that the real time constraints are schedulable. | |
7278 | */ | |
7279 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7280 | |
9a7e0b18 PZ |
7281 | /* Must be called with tasklist_lock held */ |
7282 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7283 | { |
9a7e0b18 | 7284 | struct task_struct *g, *p; |
b40b2e8e | 7285 | |
9a7e0b18 | 7286 | do_each_thread(g, p) { |
029632fb | 7287 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7288 | return 1; |
7289 | } while_each_thread(g, p); | |
b40b2e8e | 7290 | |
9a7e0b18 PZ |
7291 | return 0; |
7292 | } | |
b40b2e8e | 7293 | |
9a7e0b18 PZ |
7294 | struct rt_schedulable_data { |
7295 | struct task_group *tg; | |
7296 | u64 rt_period; | |
7297 | u64 rt_runtime; | |
7298 | }; | |
b40b2e8e | 7299 | |
a790de99 | 7300 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7301 | { |
7302 | struct rt_schedulable_data *d = data; | |
7303 | struct task_group *child; | |
7304 | unsigned long total, sum = 0; | |
7305 | u64 period, runtime; | |
b40b2e8e | 7306 | |
9a7e0b18 PZ |
7307 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7308 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7309 | |
9a7e0b18 PZ |
7310 | if (tg == d->tg) { |
7311 | period = d->rt_period; | |
7312 | runtime = d->rt_runtime; | |
b40b2e8e | 7313 | } |
b40b2e8e | 7314 | |
4653f803 PZ |
7315 | /* |
7316 | * Cannot have more runtime than the period. | |
7317 | */ | |
7318 | if (runtime > period && runtime != RUNTIME_INF) | |
7319 | return -EINVAL; | |
6f505b16 | 7320 | |
4653f803 PZ |
7321 | /* |
7322 | * Ensure we don't starve existing RT tasks. | |
7323 | */ | |
9a7e0b18 PZ |
7324 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7325 | return -EBUSY; | |
6f505b16 | 7326 | |
9a7e0b18 | 7327 | total = to_ratio(period, runtime); |
6f505b16 | 7328 | |
4653f803 PZ |
7329 | /* |
7330 | * Nobody can have more than the global setting allows. | |
7331 | */ | |
7332 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7333 | return -EINVAL; | |
6f505b16 | 7334 | |
4653f803 PZ |
7335 | /* |
7336 | * The sum of our children's runtime should not exceed our own. | |
7337 | */ | |
9a7e0b18 PZ |
7338 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7339 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7340 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7341 | |
9a7e0b18 PZ |
7342 | if (child == d->tg) { |
7343 | period = d->rt_period; | |
7344 | runtime = d->rt_runtime; | |
7345 | } | |
6f505b16 | 7346 | |
9a7e0b18 | 7347 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7348 | } |
6f505b16 | 7349 | |
9a7e0b18 PZ |
7350 | if (sum > total) |
7351 | return -EINVAL; | |
7352 | ||
7353 | return 0; | |
6f505b16 PZ |
7354 | } |
7355 | ||
9a7e0b18 | 7356 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7357 | { |
8277434e PT |
7358 | int ret; |
7359 | ||
9a7e0b18 PZ |
7360 | struct rt_schedulable_data data = { |
7361 | .tg = tg, | |
7362 | .rt_period = period, | |
7363 | .rt_runtime = runtime, | |
7364 | }; | |
7365 | ||
8277434e PT |
7366 | rcu_read_lock(); |
7367 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7368 | rcu_read_unlock(); | |
7369 | ||
7370 | return ret; | |
521f1a24 DG |
7371 | } |
7372 | ||
ab84d31e | 7373 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7374 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7375 | { |
ac086bc2 | 7376 | int i, err = 0; |
9f0c1e56 | 7377 | |
9f0c1e56 | 7378 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7379 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7380 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7381 | if (err) | |
9f0c1e56 | 7382 | goto unlock; |
ac086bc2 | 7383 | |
0986b11b | 7384 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7385 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7386 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7387 | |
7388 | for_each_possible_cpu(i) { | |
7389 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7390 | ||
0986b11b | 7391 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7392 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7393 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7394 | } |
0986b11b | 7395 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7396 | unlock: |
521f1a24 | 7397 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7398 | mutex_unlock(&rt_constraints_mutex); |
7399 | ||
7400 | return err; | |
6f505b16 PZ |
7401 | } |
7402 | ||
25cc7da7 | 7403 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7404 | { |
7405 | u64 rt_runtime, rt_period; | |
7406 | ||
7407 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7408 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7409 | if (rt_runtime_us < 0) | |
7410 | rt_runtime = RUNTIME_INF; | |
7411 | ||
ab84d31e | 7412 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7413 | } |
7414 | ||
25cc7da7 | 7415 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7416 | { |
7417 | u64 rt_runtime_us; | |
7418 | ||
d0b27fa7 | 7419 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7420 | return -1; |
7421 | ||
d0b27fa7 | 7422 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7423 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7424 | return rt_runtime_us; | |
7425 | } | |
d0b27fa7 | 7426 | |
25cc7da7 | 7427 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7428 | { |
7429 | u64 rt_runtime, rt_period; | |
7430 | ||
7431 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7432 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7433 | ||
619b0488 R |
7434 | if (rt_period == 0) |
7435 | return -EINVAL; | |
7436 | ||
ab84d31e | 7437 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7438 | } |
7439 | ||
25cc7da7 | 7440 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7441 | { |
7442 | u64 rt_period_us; | |
7443 | ||
7444 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7445 | do_div(rt_period_us, NSEC_PER_USEC); | |
7446 | return rt_period_us; | |
7447 | } | |
332ac17e | 7448 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7449 | |
332ac17e | 7450 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7451 | static int sched_rt_global_constraints(void) |
7452 | { | |
7453 | int ret = 0; | |
7454 | ||
7455 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7456 | read_lock(&tasklist_lock); |
4653f803 | 7457 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7458 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7459 | mutex_unlock(&rt_constraints_mutex); |
7460 | ||
7461 | return ret; | |
7462 | } | |
54e99124 | 7463 | |
25cc7da7 | 7464 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7465 | { |
7466 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7467 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7468 | return 0; | |
7469 | ||
7470 | return 1; | |
7471 | } | |
7472 | ||
6d6bc0ad | 7473 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7474 | static int sched_rt_global_constraints(void) |
7475 | { | |
ac086bc2 | 7476 | unsigned long flags; |
332ac17e | 7477 | int i, ret = 0; |
ec5d4989 | 7478 | |
0986b11b | 7479 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7480 | for_each_possible_cpu(i) { |
7481 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7482 | ||
0986b11b | 7483 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7484 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7485 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7486 | } |
0986b11b | 7487 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7488 | |
332ac17e | 7489 | return ret; |
d0b27fa7 | 7490 | } |
6d6bc0ad | 7491 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7492 | |
332ac17e DF |
7493 | static int sched_dl_global_constraints(void) |
7494 | { | |
1724813d PZ |
7495 | u64 runtime = global_rt_runtime(); |
7496 | u64 period = global_rt_period(); | |
332ac17e | 7497 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7498 | int cpu, ret = 0; |
49516342 | 7499 | unsigned long flags; |
332ac17e DF |
7500 | |
7501 | /* | |
7502 | * Here we want to check the bandwidth not being set to some | |
7503 | * value smaller than the currently allocated bandwidth in | |
7504 | * any of the root_domains. | |
7505 | * | |
7506 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7507 | * cycling on root_domains... Discussion on different/better | |
7508 | * solutions is welcome! | |
7509 | */ | |
1724813d PZ |
7510 | for_each_possible_cpu(cpu) { |
7511 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e | 7512 | |
49516342 | 7513 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7514 | if (new_bw < dl_b->total_bw) |
7515 | ret = -EBUSY; | |
49516342 | 7516 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d PZ |
7517 | |
7518 | if (ret) | |
7519 | break; | |
332ac17e DF |
7520 | } |
7521 | ||
1724813d | 7522 | return ret; |
332ac17e DF |
7523 | } |
7524 | ||
1724813d | 7525 | static void sched_dl_do_global(void) |
ce0dbbbb | 7526 | { |
1724813d PZ |
7527 | u64 new_bw = -1; |
7528 | int cpu; | |
49516342 | 7529 | unsigned long flags; |
ce0dbbbb | 7530 | |
1724813d PZ |
7531 | def_dl_bandwidth.dl_period = global_rt_period(); |
7532 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7533 | ||
7534 | if (global_rt_runtime() != RUNTIME_INF) | |
7535 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7536 | ||
7537 | /* | |
7538 | * FIXME: As above... | |
7539 | */ | |
7540 | for_each_possible_cpu(cpu) { | |
7541 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7542 | ||
49516342 | 7543 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7544 | dl_b->bw = new_bw; |
49516342 | 7545 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
ce0dbbbb | 7546 | } |
1724813d PZ |
7547 | } |
7548 | ||
7549 | static int sched_rt_global_validate(void) | |
7550 | { | |
7551 | if (sysctl_sched_rt_period <= 0) | |
7552 | return -EINVAL; | |
7553 | ||
e9e7cb38 JL |
7554 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7555 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7556 | return -EINVAL; |
7557 | ||
7558 | return 0; | |
7559 | } | |
7560 | ||
7561 | static void sched_rt_do_global(void) | |
7562 | { | |
7563 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7564 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7565 | } |
7566 | ||
d0b27fa7 | 7567 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7568 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7569 | loff_t *ppos) |
7570 | { | |
d0b27fa7 PZ |
7571 | int old_period, old_runtime; |
7572 | static DEFINE_MUTEX(mutex); | |
1724813d | 7573 | int ret; |
d0b27fa7 PZ |
7574 | |
7575 | mutex_lock(&mutex); | |
7576 | old_period = sysctl_sched_rt_period; | |
7577 | old_runtime = sysctl_sched_rt_runtime; | |
7578 | ||
8d65af78 | 7579 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7580 | |
7581 | if (!ret && write) { | |
1724813d PZ |
7582 | ret = sched_rt_global_validate(); |
7583 | if (ret) | |
7584 | goto undo; | |
7585 | ||
d0b27fa7 | 7586 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7587 | if (ret) |
7588 | goto undo; | |
7589 | ||
7590 | ret = sched_dl_global_constraints(); | |
7591 | if (ret) | |
7592 | goto undo; | |
7593 | ||
7594 | sched_rt_do_global(); | |
7595 | sched_dl_do_global(); | |
7596 | } | |
7597 | if (0) { | |
7598 | undo: | |
7599 | sysctl_sched_rt_period = old_period; | |
7600 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7601 | } |
7602 | mutex_unlock(&mutex); | |
7603 | ||
7604 | return ret; | |
7605 | } | |
68318b8e | 7606 | |
1724813d | 7607 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7608 | void __user *buffer, size_t *lenp, |
7609 | loff_t *ppos) | |
7610 | { | |
7611 | int ret; | |
332ac17e | 7612 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7613 | |
7614 | mutex_lock(&mutex); | |
332ac17e | 7615 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7616 | /* make sure that internally we keep jiffies */ |
7617 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7618 | if (!ret && write) { |
1724813d PZ |
7619 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7620 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7621 | } |
7622 | mutex_unlock(&mutex); | |
332ac17e DF |
7623 | return ret; |
7624 | } | |
7625 | ||
052f1dc7 | 7626 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7627 | |
a7c6d554 | 7628 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7629 | { |
a7c6d554 | 7630 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7631 | } |
7632 | ||
eb95419b TH |
7633 | static struct cgroup_subsys_state * |
7634 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7635 | { |
eb95419b TH |
7636 | struct task_group *parent = css_tg(parent_css); |
7637 | struct task_group *tg; | |
68318b8e | 7638 | |
eb95419b | 7639 | if (!parent) { |
68318b8e | 7640 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7641 | return &root_task_group.css; |
68318b8e SV |
7642 | } |
7643 | ||
ec7dc8ac | 7644 | tg = sched_create_group(parent); |
68318b8e SV |
7645 | if (IS_ERR(tg)) |
7646 | return ERR_PTR(-ENOMEM); | |
7647 | ||
68318b8e SV |
7648 | return &tg->css; |
7649 | } | |
7650 | ||
eb95419b | 7651 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7652 | { |
eb95419b TH |
7653 | struct task_group *tg = css_tg(css); |
7654 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7655 | |
63876986 TH |
7656 | if (parent) |
7657 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7658 | return 0; |
7659 | } | |
7660 | ||
eb95419b | 7661 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7662 | { |
eb95419b | 7663 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7664 | |
7665 | sched_destroy_group(tg); | |
7666 | } | |
7667 | ||
eb95419b | 7668 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7669 | { |
eb95419b | 7670 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7671 | |
7672 | sched_offline_group(tg); | |
7673 | } | |
7674 | ||
eb95419b | 7675 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7676 | struct cgroup_taskset *tset) |
68318b8e | 7677 | { |
bb9d97b6 TH |
7678 | struct task_struct *task; |
7679 | ||
924f0d9a | 7680 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7681 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7682 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7683 | return -EINVAL; |
b68aa230 | 7684 | #else |
bb9d97b6 TH |
7685 | /* We don't support RT-tasks being in separate groups */ |
7686 | if (task->sched_class != &fair_sched_class) | |
7687 | return -EINVAL; | |
b68aa230 | 7688 | #endif |
bb9d97b6 | 7689 | } |
be367d09 BB |
7690 | return 0; |
7691 | } | |
68318b8e | 7692 | |
eb95419b | 7693 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7694 | struct cgroup_taskset *tset) |
68318b8e | 7695 | { |
bb9d97b6 TH |
7696 | struct task_struct *task; |
7697 | ||
924f0d9a | 7698 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7699 | sched_move_task(task); |
68318b8e SV |
7700 | } |
7701 | ||
eb95419b TH |
7702 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7703 | struct cgroup_subsys_state *old_css, | |
7704 | struct task_struct *task) | |
068c5cc5 PZ |
7705 | { |
7706 | /* | |
7707 | * cgroup_exit() is called in the copy_process() failure path. | |
7708 | * Ignore this case since the task hasn't ran yet, this avoids | |
7709 | * trying to poke a half freed task state from generic code. | |
7710 | */ | |
7711 | if (!(task->flags & PF_EXITING)) | |
7712 | return; | |
7713 | ||
7714 | sched_move_task(task); | |
7715 | } | |
7716 | ||
052f1dc7 | 7717 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7718 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7719 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7720 | { |
182446d0 | 7721 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7722 | } |
7723 | ||
182446d0 TH |
7724 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7725 | struct cftype *cft) | |
68318b8e | 7726 | { |
182446d0 | 7727 | struct task_group *tg = css_tg(css); |
68318b8e | 7728 | |
c8b28116 | 7729 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7730 | } |
ab84d31e PT |
7731 | |
7732 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7733 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7734 | ||
ab84d31e PT |
7735 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7736 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7737 | ||
a790de99 PT |
7738 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7739 | ||
ab84d31e PT |
7740 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7741 | { | |
56f570e5 | 7742 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7743 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7744 | |
7745 | if (tg == &root_task_group) | |
7746 | return -EINVAL; | |
7747 | ||
7748 | /* | |
7749 | * Ensure we have at some amount of bandwidth every period. This is | |
7750 | * to prevent reaching a state of large arrears when throttled via | |
7751 | * entity_tick() resulting in prolonged exit starvation. | |
7752 | */ | |
7753 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7754 | return -EINVAL; | |
7755 | ||
7756 | /* | |
7757 | * Likewise, bound things on the otherside by preventing insane quota | |
7758 | * periods. This also allows us to normalize in computing quota | |
7759 | * feasibility. | |
7760 | */ | |
7761 | if (period > max_cfs_quota_period) | |
7762 | return -EINVAL; | |
7763 | ||
a790de99 PT |
7764 | mutex_lock(&cfs_constraints_mutex); |
7765 | ret = __cfs_schedulable(tg, period, quota); | |
7766 | if (ret) | |
7767 | goto out_unlock; | |
7768 | ||
58088ad0 | 7769 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7770 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7771 | /* |
7772 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7773 | * before making related changes, and on->off must occur afterwards | |
7774 | */ | |
7775 | if (runtime_enabled && !runtime_was_enabled) | |
7776 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7777 | raw_spin_lock_irq(&cfs_b->lock); |
7778 | cfs_b->period = ns_to_ktime(period); | |
7779 | cfs_b->quota = quota; | |
58088ad0 | 7780 | |
a9cf55b2 | 7781 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7782 | /* restart the period timer (if active) to handle new period expiry */ |
7783 | if (runtime_enabled && cfs_b->timer_active) { | |
7784 | /* force a reprogram */ | |
7785 | cfs_b->timer_active = 0; | |
7786 | __start_cfs_bandwidth(cfs_b); | |
7787 | } | |
ab84d31e PT |
7788 | raw_spin_unlock_irq(&cfs_b->lock); |
7789 | ||
7790 | for_each_possible_cpu(i) { | |
7791 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7792 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7793 | |
7794 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7795 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7796 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7797 | |
029632fb | 7798 | if (cfs_rq->throttled) |
671fd9da | 7799 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7800 | raw_spin_unlock_irq(&rq->lock); |
7801 | } | |
1ee14e6c BS |
7802 | if (runtime_was_enabled && !runtime_enabled) |
7803 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7804 | out_unlock: |
7805 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7806 | |
a790de99 | 7807 | return ret; |
ab84d31e PT |
7808 | } |
7809 | ||
7810 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7811 | { | |
7812 | u64 quota, period; | |
7813 | ||
029632fb | 7814 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7815 | if (cfs_quota_us < 0) |
7816 | quota = RUNTIME_INF; | |
7817 | else | |
7818 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7819 | ||
7820 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7821 | } | |
7822 | ||
7823 | long tg_get_cfs_quota(struct task_group *tg) | |
7824 | { | |
7825 | u64 quota_us; | |
7826 | ||
029632fb | 7827 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7828 | return -1; |
7829 | ||
029632fb | 7830 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7831 | do_div(quota_us, NSEC_PER_USEC); |
7832 | ||
7833 | return quota_us; | |
7834 | } | |
7835 | ||
7836 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7837 | { | |
7838 | u64 quota, period; | |
7839 | ||
7840 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7841 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7842 | |
ab84d31e PT |
7843 | return tg_set_cfs_bandwidth(tg, period, quota); |
7844 | } | |
7845 | ||
7846 | long tg_get_cfs_period(struct task_group *tg) | |
7847 | { | |
7848 | u64 cfs_period_us; | |
7849 | ||
029632fb | 7850 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7851 | do_div(cfs_period_us, NSEC_PER_USEC); |
7852 | ||
7853 | return cfs_period_us; | |
7854 | } | |
7855 | ||
182446d0 TH |
7856 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7857 | struct cftype *cft) | |
ab84d31e | 7858 | { |
182446d0 | 7859 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7860 | } |
7861 | ||
182446d0 TH |
7862 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7863 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7864 | { |
182446d0 | 7865 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7866 | } |
7867 | ||
182446d0 TH |
7868 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7869 | struct cftype *cft) | |
ab84d31e | 7870 | { |
182446d0 | 7871 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7872 | } |
7873 | ||
182446d0 TH |
7874 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7875 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7876 | { |
182446d0 | 7877 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7878 | } |
7879 | ||
a790de99 PT |
7880 | struct cfs_schedulable_data { |
7881 | struct task_group *tg; | |
7882 | u64 period, quota; | |
7883 | }; | |
7884 | ||
7885 | /* | |
7886 | * normalize group quota/period to be quota/max_period | |
7887 | * note: units are usecs | |
7888 | */ | |
7889 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7890 | struct cfs_schedulable_data *d) | |
7891 | { | |
7892 | u64 quota, period; | |
7893 | ||
7894 | if (tg == d->tg) { | |
7895 | period = d->period; | |
7896 | quota = d->quota; | |
7897 | } else { | |
7898 | period = tg_get_cfs_period(tg); | |
7899 | quota = tg_get_cfs_quota(tg); | |
7900 | } | |
7901 | ||
7902 | /* note: these should typically be equivalent */ | |
7903 | if (quota == RUNTIME_INF || quota == -1) | |
7904 | return RUNTIME_INF; | |
7905 | ||
7906 | return to_ratio(period, quota); | |
7907 | } | |
7908 | ||
7909 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7910 | { | |
7911 | struct cfs_schedulable_data *d = data; | |
029632fb | 7912 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7913 | s64 quota = 0, parent_quota = -1; |
7914 | ||
7915 | if (!tg->parent) { | |
7916 | quota = RUNTIME_INF; | |
7917 | } else { | |
029632fb | 7918 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7919 | |
7920 | quota = normalize_cfs_quota(tg, d); | |
7921 | parent_quota = parent_b->hierarchal_quota; | |
7922 | ||
7923 | /* | |
7924 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7925 | * limit is set | |
7926 | */ | |
7927 | if (quota == RUNTIME_INF) | |
7928 | quota = parent_quota; | |
7929 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7930 | return -EINVAL; | |
7931 | } | |
7932 | cfs_b->hierarchal_quota = quota; | |
7933 | ||
7934 | return 0; | |
7935 | } | |
7936 | ||
7937 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7938 | { | |
8277434e | 7939 | int ret; |
a790de99 PT |
7940 | struct cfs_schedulable_data data = { |
7941 | .tg = tg, | |
7942 | .period = period, | |
7943 | .quota = quota, | |
7944 | }; | |
7945 | ||
7946 | if (quota != RUNTIME_INF) { | |
7947 | do_div(data.period, NSEC_PER_USEC); | |
7948 | do_div(data.quota, NSEC_PER_USEC); | |
7949 | } | |
7950 | ||
8277434e PT |
7951 | rcu_read_lock(); |
7952 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7953 | rcu_read_unlock(); | |
7954 | ||
7955 | return ret; | |
a790de99 | 7956 | } |
e8da1b18 | 7957 | |
2da8ca82 | 7958 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 7959 | { |
2da8ca82 | 7960 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7961 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7962 | |
44ffc75b TH |
7963 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7964 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7965 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
7966 | |
7967 | return 0; | |
7968 | } | |
ab84d31e | 7969 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7970 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7971 | |
052f1dc7 | 7972 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7973 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7974 | struct cftype *cft, s64 val) | |
6f505b16 | 7975 | { |
182446d0 | 7976 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7977 | } |
7978 | ||
182446d0 TH |
7979 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7980 | struct cftype *cft) | |
6f505b16 | 7981 | { |
182446d0 | 7982 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7983 | } |
d0b27fa7 | 7984 | |
182446d0 TH |
7985 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7986 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7987 | { |
182446d0 | 7988 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7989 | } |
7990 | ||
182446d0 TH |
7991 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7992 | struct cftype *cft) | |
d0b27fa7 | 7993 | { |
182446d0 | 7994 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7995 | } |
6d6bc0ad | 7996 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7997 | |
fe5c7cc2 | 7998 | static struct cftype cpu_files[] = { |
052f1dc7 | 7999 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8000 | { |
8001 | .name = "shares", | |
f4c753b7 PM |
8002 | .read_u64 = cpu_shares_read_u64, |
8003 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8004 | }, |
052f1dc7 | 8005 | #endif |
ab84d31e PT |
8006 | #ifdef CONFIG_CFS_BANDWIDTH |
8007 | { | |
8008 | .name = "cfs_quota_us", | |
8009 | .read_s64 = cpu_cfs_quota_read_s64, | |
8010 | .write_s64 = cpu_cfs_quota_write_s64, | |
8011 | }, | |
8012 | { | |
8013 | .name = "cfs_period_us", | |
8014 | .read_u64 = cpu_cfs_period_read_u64, | |
8015 | .write_u64 = cpu_cfs_period_write_u64, | |
8016 | }, | |
e8da1b18 NR |
8017 | { |
8018 | .name = "stat", | |
2da8ca82 | 8019 | .seq_show = cpu_stats_show, |
e8da1b18 | 8020 | }, |
ab84d31e | 8021 | #endif |
052f1dc7 | 8022 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8023 | { |
9f0c1e56 | 8024 | .name = "rt_runtime_us", |
06ecb27c PM |
8025 | .read_s64 = cpu_rt_runtime_read, |
8026 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8027 | }, |
d0b27fa7 PZ |
8028 | { |
8029 | .name = "rt_period_us", | |
f4c753b7 PM |
8030 | .read_u64 = cpu_rt_period_read_uint, |
8031 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8032 | }, |
052f1dc7 | 8033 | #endif |
4baf6e33 | 8034 | { } /* terminate */ |
68318b8e SV |
8035 | }; |
8036 | ||
073219e9 | 8037 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8038 | .css_alloc = cpu_cgroup_css_alloc, |
8039 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8040 | .css_online = cpu_cgroup_css_online, |
8041 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8042 | .can_attach = cpu_cgroup_can_attach, |
8043 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8044 | .exit = cpu_cgroup_exit, |
4baf6e33 | 8045 | .base_cftypes = cpu_files, |
68318b8e SV |
8046 | .early_init = 1, |
8047 | }; | |
8048 | ||
052f1dc7 | 8049 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8050 | |
b637a328 PM |
8051 | void dump_cpu_task(int cpu) |
8052 | { | |
8053 | pr_info("Task dump for CPU %d:\n", cpu); | |
8054 | sched_show_task(cpu_curr(cpu)); | |
8055 | } |