]>
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 | 3035 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3036 | int nice_rlim = nice_to_rlimit(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 | */ | |
a9467fa3 | 3060 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3061 | nice = task_nice(current) + increment; |
1da177e4 | 3062 | |
a9467fa3 | 3063 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3064 | if (increment < 0 && !can_nice(current, nice)) |
3065 | return -EPERM; | |
3066 | ||
1da177e4 LT |
3067 | retval = security_task_setnice(current, nice); |
3068 | if (retval) | |
3069 | return retval; | |
3070 | ||
3071 | set_user_nice(current, nice); | |
3072 | return 0; | |
3073 | } | |
3074 | ||
3075 | #endif | |
3076 | ||
3077 | /** | |
3078 | * task_prio - return the priority value of a given task. | |
3079 | * @p: the task in question. | |
3080 | * | |
e69f6186 | 3081 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3082 | * RT tasks are offset by -200. Normal tasks are centered |
3083 | * around 0, value goes from -16 to +15. | |
3084 | */ | |
36c8b586 | 3085 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3086 | { |
3087 | return p->prio - MAX_RT_PRIO; | |
3088 | } | |
3089 | ||
1da177e4 LT |
3090 | /** |
3091 | * idle_cpu - is a given cpu idle currently? | |
3092 | * @cpu: the processor in question. | |
e69f6186 YB |
3093 | * |
3094 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3095 | */ |
3096 | int idle_cpu(int cpu) | |
3097 | { | |
908a3283 TG |
3098 | struct rq *rq = cpu_rq(cpu); |
3099 | ||
3100 | if (rq->curr != rq->idle) | |
3101 | return 0; | |
3102 | ||
3103 | if (rq->nr_running) | |
3104 | return 0; | |
3105 | ||
3106 | #ifdef CONFIG_SMP | |
3107 | if (!llist_empty(&rq->wake_list)) | |
3108 | return 0; | |
3109 | #endif | |
3110 | ||
3111 | return 1; | |
1da177e4 LT |
3112 | } |
3113 | ||
1da177e4 LT |
3114 | /** |
3115 | * idle_task - return the idle task for a given cpu. | |
3116 | * @cpu: the processor in question. | |
e69f6186 YB |
3117 | * |
3118 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3119 | */ |
36c8b586 | 3120 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3121 | { |
3122 | return cpu_rq(cpu)->idle; | |
3123 | } | |
3124 | ||
3125 | /** | |
3126 | * find_process_by_pid - find a process with a matching PID value. | |
3127 | * @pid: the pid in question. | |
e69f6186 YB |
3128 | * |
3129 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3130 | */ |
a9957449 | 3131 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3132 | { |
228ebcbe | 3133 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3134 | } |
3135 | ||
aab03e05 DF |
3136 | /* |
3137 | * This function initializes the sched_dl_entity of a newly becoming | |
3138 | * SCHED_DEADLINE task. | |
3139 | * | |
3140 | * Only the static values are considered here, the actual runtime and the | |
3141 | * absolute deadline will be properly calculated when the task is enqueued | |
3142 | * for the first time with its new policy. | |
3143 | */ | |
3144 | static void | |
3145 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3146 | { | |
3147 | struct sched_dl_entity *dl_se = &p->dl; | |
3148 | ||
3149 | init_dl_task_timer(dl_se); | |
3150 | dl_se->dl_runtime = attr->sched_runtime; | |
3151 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3152 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3153 | dl_se->flags = attr->sched_flags; |
332ac17e | 3154 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3155 | dl_se->dl_throttled = 0; |
3156 | dl_se->dl_new = 1; | |
5bfd126e | 3157 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3158 | } |
3159 | ||
c365c292 TG |
3160 | static void __setscheduler_params(struct task_struct *p, |
3161 | const struct sched_attr *attr) | |
1da177e4 | 3162 | { |
d50dde5a DF |
3163 | int policy = attr->sched_policy; |
3164 | ||
39fd8fd2 PZ |
3165 | if (policy == -1) /* setparam */ |
3166 | policy = p->policy; | |
3167 | ||
1da177e4 | 3168 | p->policy = policy; |
d50dde5a | 3169 | |
aab03e05 DF |
3170 | if (dl_policy(policy)) |
3171 | __setparam_dl(p, attr); | |
39fd8fd2 | 3172 | else if (fair_policy(policy)) |
d50dde5a DF |
3173 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3174 | ||
39fd8fd2 PZ |
3175 | /* |
3176 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3177 | * !rt_policy. Always setting this ensures that things like | |
3178 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3179 | */ | |
3180 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3181 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3182 | set_load_weight(p); |
3183 | } | |
39fd8fd2 | 3184 | |
c365c292 TG |
3185 | /* Actually do priority change: must hold pi & rq lock. */ |
3186 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3187 | const struct sched_attr *attr) | |
3188 | { | |
3189 | __setscheduler_params(p, attr); | |
d50dde5a | 3190 | |
383afd09 SR |
3191 | /* |
3192 | * If we get here, there was no pi waiters boosting the | |
3193 | * task. It is safe to use the normal prio. | |
3194 | */ | |
3195 | p->prio = normal_prio(p); | |
3196 | ||
aab03e05 DF |
3197 | if (dl_prio(p->prio)) |
3198 | p->sched_class = &dl_sched_class; | |
3199 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3200 | p->sched_class = &rt_sched_class; |
3201 | else | |
3202 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3203 | } |
aab03e05 DF |
3204 | |
3205 | static void | |
3206 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3207 | { | |
3208 | struct sched_dl_entity *dl_se = &p->dl; | |
3209 | ||
3210 | attr->sched_priority = p->rt_priority; | |
3211 | attr->sched_runtime = dl_se->dl_runtime; | |
3212 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3213 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3214 | attr->sched_flags = dl_se->flags; |
3215 | } | |
3216 | ||
3217 | /* | |
3218 | * This function validates the new parameters of a -deadline task. | |
3219 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3220 | * than the runtime, as well as the period of being zero or |
332ac17e | 3221 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3222 | * user parameters are above the internal resolution of 1us (we |
3223 | * check sched_runtime only since it is always the smaller one) and | |
3224 | * below 2^63 ns (we have to check both sched_deadline and | |
3225 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3226 | */ |
3227 | static bool | |
3228 | __checkparam_dl(const struct sched_attr *attr) | |
3229 | { | |
b0827819 JL |
3230 | /* deadline != 0 */ |
3231 | if (attr->sched_deadline == 0) | |
3232 | return false; | |
3233 | ||
3234 | /* | |
3235 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3236 | * that big. | |
3237 | */ | |
3238 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3239 | return false; | |
3240 | ||
3241 | /* | |
3242 | * Since we use the MSB for wrap-around and sign issues, make | |
3243 | * sure it's not set (mind that period can be equal to zero). | |
3244 | */ | |
3245 | if (attr->sched_deadline & (1ULL << 63) || | |
3246 | attr->sched_period & (1ULL << 63)) | |
3247 | return false; | |
3248 | ||
3249 | /* runtime <= deadline <= period (if period != 0) */ | |
3250 | if ((attr->sched_period != 0 && | |
3251 | attr->sched_period < attr->sched_deadline) || | |
3252 | attr->sched_deadline < attr->sched_runtime) | |
3253 | return false; | |
3254 | ||
3255 | return true; | |
aab03e05 DF |
3256 | } |
3257 | ||
c69e8d9c DH |
3258 | /* |
3259 | * check the target process has a UID that matches the current process's | |
3260 | */ | |
3261 | static bool check_same_owner(struct task_struct *p) | |
3262 | { | |
3263 | const struct cred *cred = current_cred(), *pcred; | |
3264 | bool match; | |
3265 | ||
3266 | rcu_read_lock(); | |
3267 | pcred = __task_cred(p); | |
9c806aa0 EB |
3268 | match = (uid_eq(cred->euid, pcred->euid) || |
3269 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3270 | rcu_read_unlock(); |
3271 | return match; | |
3272 | } | |
3273 | ||
d50dde5a DF |
3274 | static int __sched_setscheduler(struct task_struct *p, |
3275 | const struct sched_attr *attr, | |
3276 | bool user) | |
1da177e4 | 3277 | { |
383afd09 SR |
3278 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3279 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
83b699ed | 3280 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3281 | int policy = attr->sched_policy; |
1da177e4 | 3282 | unsigned long flags; |
83ab0aa0 | 3283 | const struct sched_class *prev_class; |
70b97a7f | 3284 | struct rq *rq; |
ca94c442 | 3285 | int reset_on_fork; |
1da177e4 | 3286 | |
66e5393a SR |
3287 | /* may grab non-irq protected spin_locks */ |
3288 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3289 | recheck: |
3290 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3291 | if (policy < 0) { |
3292 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3293 | policy = oldpolicy = p->policy; |
ca94c442 | 3294 | } else { |
7479f3c9 | 3295 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3296 | |
aab03e05 DF |
3297 | if (policy != SCHED_DEADLINE && |
3298 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3299 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3300 | policy != SCHED_IDLE) | |
3301 | return -EINVAL; | |
3302 | } | |
3303 | ||
7479f3c9 PZ |
3304 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3305 | return -EINVAL; | |
3306 | ||
1da177e4 LT |
3307 | /* |
3308 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3309 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3310 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3311 | */ |
0bb040a4 | 3312 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3313 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3314 | return -EINVAL; |
aab03e05 DF |
3315 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3316 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3317 | return -EINVAL; |
3318 | ||
37e4ab3f OC |
3319 | /* |
3320 | * Allow unprivileged RT tasks to decrease priority: | |
3321 | */ | |
961ccddd | 3322 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3323 | if (fair_policy(policy)) { |
d0ea0268 | 3324 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3325 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3326 | return -EPERM; |
3327 | } | |
3328 | ||
e05606d3 | 3329 | if (rt_policy(policy)) { |
a44702e8 ON |
3330 | unsigned long rlim_rtprio = |
3331 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3332 | |
3333 | /* can't set/change the rt policy */ | |
3334 | if (policy != p->policy && !rlim_rtprio) | |
3335 | return -EPERM; | |
3336 | ||
3337 | /* can't increase priority */ | |
d50dde5a DF |
3338 | if (attr->sched_priority > p->rt_priority && |
3339 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3340 | return -EPERM; |
3341 | } | |
c02aa73b | 3342 | |
d44753b8 JL |
3343 | /* |
3344 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3345 | * (safest behavior); in the future we would like to allow | |
3346 | * unprivileged DL tasks to increase their relative deadline | |
3347 | * or reduce their runtime (both ways reducing utilization) | |
3348 | */ | |
3349 | if (dl_policy(policy)) | |
3350 | return -EPERM; | |
3351 | ||
dd41f596 | 3352 | /* |
c02aa73b DH |
3353 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3354 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3355 | */ |
c02aa73b | 3356 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3357 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3358 | return -EPERM; |
3359 | } | |
5fe1d75f | 3360 | |
37e4ab3f | 3361 | /* can't change other user's priorities */ |
c69e8d9c | 3362 | if (!check_same_owner(p)) |
37e4ab3f | 3363 | return -EPERM; |
ca94c442 LP |
3364 | |
3365 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3366 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3367 | return -EPERM; | |
37e4ab3f | 3368 | } |
1da177e4 | 3369 | |
725aad24 | 3370 | if (user) { |
b0ae1981 | 3371 | retval = security_task_setscheduler(p); |
725aad24 JF |
3372 | if (retval) |
3373 | return retval; | |
3374 | } | |
3375 | ||
b29739f9 IM |
3376 | /* |
3377 | * make sure no PI-waiters arrive (or leave) while we are | |
3378 | * changing the priority of the task: | |
0122ec5b | 3379 | * |
25985edc | 3380 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3381 | * runqueue lock must be held. |
3382 | */ | |
0122ec5b | 3383 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3384 | |
34f971f6 PZ |
3385 | /* |
3386 | * Changing the policy of the stop threads its a very bad idea | |
3387 | */ | |
3388 | if (p == rq->stop) { | |
0122ec5b | 3389 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3390 | return -EINVAL; |
3391 | } | |
3392 | ||
a51e9198 | 3393 | /* |
d6b1e911 TG |
3394 | * If not changing anything there's no need to proceed further, |
3395 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3396 | */ |
d50dde5a | 3397 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3398 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3399 | goto change; |
3400 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3401 | goto change; | |
aab03e05 DF |
3402 | if (dl_policy(policy)) |
3403 | goto change; | |
d50dde5a | 3404 | |
d6b1e911 | 3405 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3406 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3407 | return 0; |
3408 | } | |
d50dde5a | 3409 | change: |
a51e9198 | 3410 | |
dc61b1d6 | 3411 | if (user) { |
332ac17e | 3412 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3413 | /* |
3414 | * Do not allow realtime tasks into groups that have no runtime | |
3415 | * assigned. | |
3416 | */ | |
3417 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3418 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3419 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3420 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3421 | return -EPERM; |
3422 | } | |
dc61b1d6 | 3423 | #endif |
332ac17e DF |
3424 | #ifdef CONFIG_SMP |
3425 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3426 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3427 | |
3428 | /* | |
3429 | * Don't allow tasks with an affinity mask smaller than | |
3430 | * the entire root_domain to become SCHED_DEADLINE. We | |
3431 | * will also fail if there's no bandwidth available. | |
3432 | */ | |
e4099a5e PZ |
3433 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3434 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3435 | task_rq_unlock(rq, p, &flags); |
3436 | return -EPERM; | |
3437 | } | |
3438 | } | |
3439 | #endif | |
3440 | } | |
dc61b1d6 | 3441 | |
1da177e4 LT |
3442 | /* recheck policy now with rq lock held */ |
3443 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3444 | policy = oldpolicy = -1; | |
0122ec5b | 3445 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3446 | goto recheck; |
3447 | } | |
332ac17e DF |
3448 | |
3449 | /* | |
3450 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3451 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3452 | * is available. | |
3453 | */ | |
e4099a5e | 3454 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3455 | task_rq_unlock(rq, p, &flags); |
3456 | return -EBUSY; | |
3457 | } | |
3458 | ||
c365c292 TG |
3459 | p->sched_reset_on_fork = reset_on_fork; |
3460 | oldprio = p->prio; | |
3461 | ||
3462 | /* | |
3463 | * Special case for priority boosted tasks. | |
3464 | * | |
3465 | * If the new priority is lower or equal (user space view) | |
3466 | * than the current (boosted) priority, we just store the new | |
3467 | * normal parameters and do not touch the scheduler class and | |
3468 | * the runqueue. This will be done when the task deboost | |
3469 | * itself. | |
3470 | */ | |
3471 | if (rt_mutex_check_prio(p, newprio)) { | |
3472 | __setscheduler_params(p, attr); | |
3473 | task_rq_unlock(rq, p, &flags); | |
3474 | return 0; | |
3475 | } | |
3476 | ||
fd2f4419 | 3477 | on_rq = p->on_rq; |
051a1d1a | 3478 | running = task_current(rq, p); |
0e1f3483 | 3479 | if (on_rq) |
4ca9b72b | 3480 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3481 | if (running) |
3482 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3483 | |
83ab0aa0 | 3484 | prev_class = p->sched_class; |
d50dde5a | 3485 | __setscheduler(rq, p, attr); |
f6b53205 | 3486 | |
0e1f3483 HS |
3487 | if (running) |
3488 | p->sched_class->set_curr_task(rq); | |
81a44c54 TG |
3489 | if (on_rq) { |
3490 | /* | |
3491 | * We enqueue to tail when the priority of a task is | |
3492 | * increased (user space view). | |
3493 | */ | |
3494 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3495 | } | |
cb469845 | 3496 | |
da7a735e | 3497 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3498 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3499 | |
95e02ca9 TG |
3500 | rt_mutex_adjust_pi(p); |
3501 | ||
1da177e4 LT |
3502 | return 0; |
3503 | } | |
961ccddd | 3504 | |
7479f3c9 PZ |
3505 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3506 | const struct sched_param *param, bool check) | |
3507 | { | |
3508 | struct sched_attr attr = { | |
3509 | .sched_policy = policy, | |
3510 | .sched_priority = param->sched_priority, | |
3511 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3512 | }; | |
3513 | ||
3514 | /* | |
3515 | * Fixup the legacy SCHED_RESET_ON_FORK hack | |
3516 | */ | |
3517 | if (policy & SCHED_RESET_ON_FORK) { | |
3518 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
3519 | policy &= ~SCHED_RESET_ON_FORK; | |
3520 | attr.sched_policy = policy; | |
3521 | } | |
3522 | ||
3523 | return __sched_setscheduler(p, &attr, check); | |
3524 | } | |
961ccddd RR |
3525 | /** |
3526 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3527 | * @p: the task in question. | |
3528 | * @policy: new policy. | |
3529 | * @param: structure containing the new RT priority. | |
3530 | * | |
e69f6186 YB |
3531 | * Return: 0 on success. An error code otherwise. |
3532 | * | |
961ccddd RR |
3533 | * NOTE that the task may be already dead. |
3534 | */ | |
3535 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3536 | const struct sched_param *param) |
961ccddd | 3537 | { |
7479f3c9 | 3538 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3539 | } |
1da177e4 LT |
3540 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3541 | ||
d50dde5a DF |
3542 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3543 | { | |
3544 | return __sched_setscheduler(p, attr, true); | |
3545 | } | |
3546 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3547 | ||
961ccddd RR |
3548 | /** |
3549 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3550 | * @p: the task in question. | |
3551 | * @policy: new policy. | |
3552 | * @param: structure containing the new RT priority. | |
3553 | * | |
3554 | * Just like sched_setscheduler, only don't bother checking if the | |
3555 | * current context has permission. For example, this is needed in | |
3556 | * stop_machine(): we create temporary high priority worker threads, | |
3557 | * but our caller might not have that capability. | |
e69f6186 YB |
3558 | * |
3559 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3560 | */ |
3561 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3562 | const struct sched_param *param) |
961ccddd | 3563 | { |
7479f3c9 | 3564 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3565 | } |
3566 | ||
95cdf3b7 IM |
3567 | static int |
3568 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3569 | { |
1da177e4 LT |
3570 | struct sched_param lparam; |
3571 | struct task_struct *p; | |
36c8b586 | 3572 | int retval; |
1da177e4 LT |
3573 | |
3574 | if (!param || pid < 0) | |
3575 | return -EINVAL; | |
3576 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3577 | return -EFAULT; | |
5fe1d75f ON |
3578 | |
3579 | rcu_read_lock(); | |
3580 | retval = -ESRCH; | |
1da177e4 | 3581 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3582 | if (p != NULL) |
3583 | retval = sched_setscheduler(p, policy, &lparam); | |
3584 | rcu_read_unlock(); | |
36c8b586 | 3585 | |
1da177e4 LT |
3586 | return retval; |
3587 | } | |
3588 | ||
d50dde5a DF |
3589 | /* |
3590 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3591 | */ | |
3592 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3593 | struct sched_attr *attr) | |
3594 | { | |
3595 | u32 size; | |
3596 | int ret; | |
3597 | ||
3598 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3599 | return -EFAULT; | |
3600 | ||
3601 | /* | |
3602 | * zero the full structure, so that a short copy will be nice. | |
3603 | */ | |
3604 | memset(attr, 0, sizeof(*attr)); | |
3605 | ||
3606 | ret = get_user(size, &uattr->size); | |
3607 | if (ret) | |
3608 | return ret; | |
3609 | ||
3610 | if (size > PAGE_SIZE) /* silly large */ | |
3611 | goto err_size; | |
3612 | ||
3613 | if (!size) /* abi compat */ | |
3614 | size = SCHED_ATTR_SIZE_VER0; | |
3615 | ||
3616 | if (size < SCHED_ATTR_SIZE_VER0) | |
3617 | goto err_size; | |
3618 | ||
3619 | /* | |
3620 | * If we're handed a bigger struct than we know of, | |
3621 | * ensure all the unknown bits are 0 - i.e. new | |
3622 | * user-space does not rely on any kernel feature | |
3623 | * extensions we dont know about yet. | |
3624 | */ | |
3625 | if (size > sizeof(*attr)) { | |
3626 | unsigned char __user *addr; | |
3627 | unsigned char __user *end; | |
3628 | unsigned char val; | |
3629 | ||
3630 | addr = (void __user *)uattr + sizeof(*attr); | |
3631 | end = (void __user *)uattr + size; | |
3632 | ||
3633 | for (; addr < end; addr++) { | |
3634 | ret = get_user(val, addr); | |
3635 | if (ret) | |
3636 | return ret; | |
3637 | if (val) | |
3638 | goto err_size; | |
3639 | } | |
3640 | size = sizeof(*attr); | |
3641 | } | |
3642 | ||
3643 | ret = copy_from_user(attr, uattr, size); | |
3644 | if (ret) | |
3645 | return -EFAULT; | |
3646 | ||
3647 | /* | |
3648 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3649 | * to be strict and return an error on out-of-bounds values? | |
3650 | */ | |
75e45d51 | 3651 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3652 | |
e78c7bca | 3653 | return 0; |
d50dde5a DF |
3654 | |
3655 | err_size: | |
3656 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3657 | return -E2BIG; |
d50dde5a DF |
3658 | } |
3659 | ||
1da177e4 LT |
3660 | /** |
3661 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3662 | * @pid: the pid in question. | |
3663 | * @policy: new policy. | |
3664 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3665 | * |
3666 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3667 | */ |
5add95d4 HC |
3668 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3669 | struct sched_param __user *, param) | |
1da177e4 | 3670 | { |
c21761f1 JB |
3671 | /* negative values for policy are not valid */ |
3672 | if (policy < 0) | |
3673 | return -EINVAL; | |
3674 | ||
1da177e4 LT |
3675 | return do_sched_setscheduler(pid, policy, param); |
3676 | } | |
3677 | ||
3678 | /** | |
3679 | * sys_sched_setparam - set/change the RT priority of a thread | |
3680 | * @pid: the pid in question. | |
3681 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3682 | * |
3683 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3684 | */ |
5add95d4 | 3685 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3686 | { |
3687 | return do_sched_setscheduler(pid, -1, param); | |
3688 | } | |
3689 | ||
d50dde5a DF |
3690 | /** |
3691 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3692 | * @pid: the pid in question. | |
5778fccf | 3693 | * @uattr: structure containing the extended parameters. |
db66d756 | 3694 | * @flags: for future extension. |
d50dde5a | 3695 | */ |
6d35ab48 PZ |
3696 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3697 | unsigned int, flags) | |
d50dde5a DF |
3698 | { |
3699 | struct sched_attr attr; | |
3700 | struct task_struct *p; | |
3701 | int retval; | |
3702 | ||
6d35ab48 | 3703 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3704 | return -EINVAL; |
3705 | ||
143cf23d MK |
3706 | retval = sched_copy_attr(uattr, &attr); |
3707 | if (retval) | |
3708 | return retval; | |
d50dde5a | 3709 | |
dbdb2275 PZ |
3710 | if (attr.sched_policy < 0) |
3711 | return -EINVAL; | |
d50dde5a DF |
3712 | |
3713 | rcu_read_lock(); | |
3714 | retval = -ESRCH; | |
3715 | p = find_process_by_pid(pid); | |
3716 | if (p != NULL) | |
3717 | retval = sched_setattr(p, &attr); | |
3718 | rcu_read_unlock(); | |
3719 | ||
3720 | return retval; | |
3721 | } | |
3722 | ||
1da177e4 LT |
3723 | /** |
3724 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3725 | * @pid: the pid in question. | |
e69f6186 YB |
3726 | * |
3727 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3728 | * code. | |
1da177e4 | 3729 | */ |
5add95d4 | 3730 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3731 | { |
36c8b586 | 3732 | struct task_struct *p; |
3a5c359a | 3733 | int retval; |
1da177e4 LT |
3734 | |
3735 | if (pid < 0) | |
3a5c359a | 3736 | return -EINVAL; |
1da177e4 LT |
3737 | |
3738 | retval = -ESRCH; | |
5fe85be0 | 3739 | rcu_read_lock(); |
1da177e4 LT |
3740 | p = find_process_by_pid(pid); |
3741 | if (p) { | |
3742 | retval = security_task_getscheduler(p); | |
3743 | if (!retval) | |
ca94c442 LP |
3744 | retval = p->policy |
3745 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3746 | } |
5fe85be0 | 3747 | rcu_read_unlock(); |
1da177e4 LT |
3748 | return retval; |
3749 | } | |
3750 | ||
3751 | /** | |
ca94c442 | 3752 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3753 | * @pid: the pid in question. |
3754 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3755 | * |
3756 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3757 | * code. | |
1da177e4 | 3758 | */ |
5add95d4 | 3759 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3760 | { |
ce5f7f82 | 3761 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3762 | struct task_struct *p; |
3a5c359a | 3763 | int retval; |
1da177e4 LT |
3764 | |
3765 | if (!param || pid < 0) | |
3a5c359a | 3766 | return -EINVAL; |
1da177e4 | 3767 | |
5fe85be0 | 3768 | rcu_read_lock(); |
1da177e4 LT |
3769 | p = find_process_by_pid(pid); |
3770 | retval = -ESRCH; | |
3771 | if (!p) | |
3772 | goto out_unlock; | |
3773 | ||
3774 | retval = security_task_getscheduler(p); | |
3775 | if (retval) | |
3776 | goto out_unlock; | |
3777 | ||
ce5f7f82 PZ |
3778 | if (task_has_rt_policy(p)) |
3779 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3780 | rcu_read_unlock(); |
1da177e4 LT |
3781 | |
3782 | /* | |
3783 | * This one might sleep, we cannot do it with a spinlock held ... | |
3784 | */ | |
3785 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3786 | ||
1da177e4 LT |
3787 | return retval; |
3788 | ||
3789 | out_unlock: | |
5fe85be0 | 3790 | rcu_read_unlock(); |
1da177e4 LT |
3791 | return retval; |
3792 | } | |
3793 | ||
d50dde5a DF |
3794 | static int sched_read_attr(struct sched_attr __user *uattr, |
3795 | struct sched_attr *attr, | |
3796 | unsigned int usize) | |
3797 | { | |
3798 | int ret; | |
3799 | ||
3800 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3801 | return -EFAULT; | |
3802 | ||
3803 | /* | |
3804 | * If we're handed a smaller struct than we know of, | |
3805 | * ensure all the unknown bits are 0 - i.e. old | |
3806 | * user-space does not get uncomplete information. | |
3807 | */ | |
3808 | if (usize < sizeof(*attr)) { | |
3809 | unsigned char *addr; | |
3810 | unsigned char *end; | |
3811 | ||
3812 | addr = (void *)attr + usize; | |
3813 | end = (void *)attr + sizeof(*attr); | |
3814 | ||
3815 | for (; addr < end; addr++) { | |
3816 | if (*addr) | |
22400674 | 3817 | return -EFBIG; |
d50dde5a DF |
3818 | } |
3819 | ||
3820 | attr->size = usize; | |
3821 | } | |
3822 | ||
4efbc454 | 3823 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3824 | if (ret) |
3825 | return -EFAULT; | |
3826 | ||
22400674 | 3827 | return 0; |
d50dde5a DF |
3828 | } |
3829 | ||
3830 | /** | |
aab03e05 | 3831 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3832 | * @pid: the pid in question. |
5778fccf | 3833 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3834 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3835 | * @flags: for future extension. |
d50dde5a | 3836 | */ |
6d35ab48 PZ |
3837 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3838 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3839 | { |
3840 | struct sched_attr attr = { | |
3841 | .size = sizeof(struct sched_attr), | |
3842 | }; | |
3843 | struct task_struct *p; | |
3844 | int retval; | |
3845 | ||
3846 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3847 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3848 | return -EINVAL; |
3849 | ||
3850 | rcu_read_lock(); | |
3851 | p = find_process_by_pid(pid); | |
3852 | retval = -ESRCH; | |
3853 | if (!p) | |
3854 | goto out_unlock; | |
3855 | ||
3856 | retval = security_task_getscheduler(p); | |
3857 | if (retval) | |
3858 | goto out_unlock; | |
3859 | ||
3860 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3861 | if (p->sched_reset_on_fork) |
3862 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3863 | if (task_has_dl_policy(p)) |
3864 | __getparam_dl(p, &attr); | |
3865 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3866 | attr.sched_priority = p->rt_priority; |
3867 | else | |
d0ea0268 | 3868 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3869 | |
3870 | rcu_read_unlock(); | |
3871 | ||
3872 | retval = sched_read_attr(uattr, &attr, size); | |
3873 | return retval; | |
3874 | ||
3875 | out_unlock: | |
3876 | rcu_read_unlock(); | |
3877 | return retval; | |
3878 | } | |
3879 | ||
96f874e2 | 3880 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3881 | { |
5a16f3d3 | 3882 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3883 | struct task_struct *p; |
3884 | int retval; | |
1da177e4 | 3885 | |
23f5d142 | 3886 | rcu_read_lock(); |
1da177e4 LT |
3887 | |
3888 | p = find_process_by_pid(pid); | |
3889 | if (!p) { | |
23f5d142 | 3890 | rcu_read_unlock(); |
1da177e4 LT |
3891 | return -ESRCH; |
3892 | } | |
3893 | ||
23f5d142 | 3894 | /* Prevent p going away */ |
1da177e4 | 3895 | get_task_struct(p); |
23f5d142 | 3896 | rcu_read_unlock(); |
1da177e4 | 3897 | |
14a40ffc TH |
3898 | if (p->flags & PF_NO_SETAFFINITY) { |
3899 | retval = -EINVAL; | |
3900 | goto out_put_task; | |
3901 | } | |
5a16f3d3 RR |
3902 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3903 | retval = -ENOMEM; | |
3904 | goto out_put_task; | |
3905 | } | |
3906 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3907 | retval = -ENOMEM; | |
3908 | goto out_free_cpus_allowed; | |
3909 | } | |
1da177e4 | 3910 | retval = -EPERM; |
4c44aaaf EB |
3911 | if (!check_same_owner(p)) { |
3912 | rcu_read_lock(); | |
3913 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3914 | rcu_read_unlock(); | |
3915 | goto out_unlock; | |
3916 | } | |
3917 | rcu_read_unlock(); | |
3918 | } | |
1da177e4 | 3919 | |
b0ae1981 | 3920 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3921 | if (retval) |
3922 | goto out_unlock; | |
3923 | ||
e4099a5e PZ |
3924 | |
3925 | cpuset_cpus_allowed(p, cpus_allowed); | |
3926 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3927 | ||
332ac17e DF |
3928 | /* |
3929 | * Since bandwidth control happens on root_domain basis, | |
3930 | * if admission test is enabled, we only admit -deadline | |
3931 | * tasks allowed to run on all the CPUs in the task's | |
3932 | * root_domain. | |
3933 | */ | |
3934 | #ifdef CONFIG_SMP | |
3935 | if (task_has_dl_policy(p)) { | |
3936 | const struct cpumask *span = task_rq(p)->rd->span; | |
3937 | ||
e4099a5e | 3938 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
3939 | retval = -EBUSY; |
3940 | goto out_unlock; | |
3941 | } | |
3942 | } | |
3943 | #endif | |
49246274 | 3944 | again: |
5a16f3d3 | 3945 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3946 | |
8707d8b8 | 3947 | if (!retval) { |
5a16f3d3 RR |
3948 | cpuset_cpus_allowed(p, cpus_allowed); |
3949 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3950 | /* |
3951 | * We must have raced with a concurrent cpuset | |
3952 | * update. Just reset the cpus_allowed to the | |
3953 | * cpuset's cpus_allowed | |
3954 | */ | |
5a16f3d3 | 3955 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3956 | goto again; |
3957 | } | |
3958 | } | |
1da177e4 | 3959 | out_unlock: |
5a16f3d3 RR |
3960 | free_cpumask_var(new_mask); |
3961 | out_free_cpus_allowed: | |
3962 | free_cpumask_var(cpus_allowed); | |
3963 | out_put_task: | |
1da177e4 | 3964 | put_task_struct(p); |
1da177e4 LT |
3965 | return retval; |
3966 | } | |
3967 | ||
3968 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3969 | struct cpumask *new_mask) |
1da177e4 | 3970 | { |
96f874e2 RR |
3971 | if (len < cpumask_size()) |
3972 | cpumask_clear(new_mask); | |
3973 | else if (len > cpumask_size()) | |
3974 | len = cpumask_size(); | |
3975 | ||
1da177e4 LT |
3976 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3977 | } | |
3978 | ||
3979 | /** | |
3980 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3981 | * @pid: pid of the process | |
3982 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3983 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
3984 | * |
3985 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3986 | */ |
5add95d4 HC |
3987 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
3988 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 3989 | { |
5a16f3d3 | 3990 | cpumask_var_t new_mask; |
1da177e4 LT |
3991 | int retval; |
3992 | ||
5a16f3d3 RR |
3993 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
3994 | return -ENOMEM; | |
1da177e4 | 3995 | |
5a16f3d3 RR |
3996 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
3997 | if (retval == 0) | |
3998 | retval = sched_setaffinity(pid, new_mask); | |
3999 | free_cpumask_var(new_mask); | |
4000 | return retval; | |
1da177e4 LT |
4001 | } |
4002 | ||
96f874e2 | 4003 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4004 | { |
36c8b586 | 4005 | struct task_struct *p; |
31605683 | 4006 | unsigned long flags; |
1da177e4 | 4007 | int retval; |
1da177e4 | 4008 | |
23f5d142 | 4009 | rcu_read_lock(); |
1da177e4 LT |
4010 | |
4011 | retval = -ESRCH; | |
4012 | p = find_process_by_pid(pid); | |
4013 | if (!p) | |
4014 | goto out_unlock; | |
4015 | ||
e7834f8f DQ |
4016 | retval = security_task_getscheduler(p); |
4017 | if (retval) | |
4018 | goto out_unlock; | |
4019 | ||
013fdb80 | 4020 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4021 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4022 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4023 | |
4024 | out_unlock: | |
23f5d142 | 4025 | rcu_read_unlock(); |
1da177e4 | 4026 | |
9531b62f | 4027 | return retval; |
1da177e4 LT |
4028 | } |
4029 | ||
4030 | /** | |
4031 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4032 | * @pid: pid of the process | |
4033 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4034 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4035 | * |
4036 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4037 | */ |
5add95d4 HC |
4038 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4039 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4040 | { |
4041 | int ret; | |
f17c8607 | 4042 | cpumask_var_t mask; |
1da177e4 | 4043 | |
84fba5ec | 4044 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4045 | return -EINVAL; |
4046 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4047 | return -EINVAL; |
4048 | ||
f17c8607 RR |
4049 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4050 | return -ENOMEM; | |
1da177e4 | 4051 | |
f17c8607 RR |
4052 | ret = sched_getaffinity(pid, mask); |
4053 | if (ret == 0) { | |
8bc037fb | 4054 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4055 | |
4056 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4057 | ret = -EFAULT; |
4058 | else | |
cd3d8031 | 4059 | ret = retlen; |
f17c8607 RR |
4060 | } |
4061 | free_cpumask_var(mask); | |
1da177e4 | 4062 | |
f17c8607 | 4063 | return ret; |
1da177e4 LT |
4064 | } |
4065 | ||
4066 | /** | |
4067 | * sys_sched_yield - yield the current processor to other threads. | |
4068 | * | |
dd41f596 IM |
4069 | * This function yields the current CPU to other tasks. If there are no |
4070 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4071 | * |
4072 | * Return: 0. | |
1da177e4 | 4073 | */ |
5add95d4 | 4074 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4075 | { |
70b97a7f | 4076 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4077 | |
2d72376b | 4078 | schedstat_inc(rq, yld_count); |
4530d7ab | 4079 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4080 | |
4081 | /* | |
4082 | * Since we are going to call schedule() anyway, there's | |
4083 | * no need to preempt or enable interrupts: | |
4084 | */ | |
4085 | __release(rq->lock); | |
8a25d5de | 4086 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4087 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4088 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4089 | |
4090 | schedule(); | |
4091 | ||
4092 | return 0; | |
4093 | } | |
4094 | ||
e7b38404 | 4095 | static void __cond_resched(void) |
1da177e4 | 4096 | { |
bdb43806 | 4097 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4098 | __schedule(); |
bdb43806 | 4099 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4100 | } |
4101 | ||
02b67cc3 | 4102 | int __sched _cond_resched(void) |
1da177e4 | 4103 | { |
d86ee480 | 4104 | if (should_resched()) { |
1da177e4 LT |
4105 | __cond_resched(); |
4106 | return 1; | |
4107 | } | |
4108 | return 0; | |
4109 | } | |
02b67cc3 | 4110 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4111 | |
4112 | /* | |
613afbf8 | 4113 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4114 | * call schedule, and on return reacquire the lock. |
4115 | * | |
41a2d6cf | 4116 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4117 | * operations here to prevent schedule() from being called twice (once via |
4118 | * spin_unlock(), once by hand). | |
4119 | */ | |
613afbf8 | 4120 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4121 | { |
d86ee480 | 4122 | int resched = should_resched(); |
6df3cecb JK |
4123 | int ret = 0; |
4124 | ||
f607c668 PZ |
4125 | lockdep_assert_held(lock); |
4126 | ||
95c354fe | 4127 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4128 | spin_unlock(lock); |
d86ee480 | 4129 | if (resched) |
95c354fe NP |
4130 | __cond_resched(); |
4131 | else | |
4132 | cpu_relax(); | |
6df3cecb | 4133 | ret = 1; |
1da177e4 | 4134 | spin_lock(lock); |
1da177e4 | 4135 | } |
6df3cecb | 4136 | return ret; |
1da177e4 | 4137 | } |
613afbf8 | 4138 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4139 | |
613afbf8 | 4140 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4141 | { |
4142 | BUG_ON(!in_softirq()); | |
4143 | ||
d86ee480 | 4144 | if (should_resched()) { |
98d82567 | 4145 | local_bh_enable(); |
1da177e4 LT |
4146 | __cond_resched(); |
4147 | local_bh_disable(); | |
4148 | return 1; | |
4149 | } | |
4150 | return 0; | |
4151 | } | |
613afbf8 | 4152 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4153 | |
1da177e4 LT |
4154 | /** |
4155 | * yield - yield the current processor to other threads. | |
4156 | * | |
8e3fabfd PZ |
4157 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4158 | * | |
4159 | * The scheduler is at all times free to pick the calling task as the most | |
4160 | * eligible task to run, if removing the yield() call from your code breaks | |
4161 | * it, its already broken. | |
4162 | * | |
4163 | * Typical broken usage is: | |
4164 | * | |
4165 | * while (!event) | |
4166 | * yield(); | |
4167 | * | |
4168 | * where one assumes that yield() will let 'the other' process run that will | |
4169 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4170 | * happen. Never use yield() as a progress guarantee!! | |
4171 | * | |
4172 | * If you want to use yield() to wait for something, use wait_event(). | |
4173 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4174 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4175 | */ |
4176 | void __sched yield(void) | |
4177 | { | |
4178 | set_current_state(TASK_RUNNING); | |
4179 | sys_sched_yield(); | |
4180 | } | |
1da177e4 LT |
4181 | EXPORT_SYMBOL(yield); |
4182 | ||
d95f4122 MG |
4183 | /** |
4184 | * yield_to - yield the current processor to another thread in | |
4185 | * your thread group, or accelerate that thread toward the | |
4186 | * processor it's on. | |
16addf95 RD |
4187 | * @p: target task |
4188 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4189 | * |
4190 | * It's the caller's job to ensure that the target task struct | |
4191 | * can't go away on us before we can do any checks. | |
4192 | * | |
e69f6186 | 4193 | * Return: |
7b270f60 PZ |
4194 | * true (>0) if we indeed boosted the target task. |
4195 | * false (0) if we failed to boost the target. | |
4196 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4197 | */ |
fa93384f | 4198 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4199 | { |
4200 | struct task_struct *curr = current; | |
4201 | struct rq *rq, *p_rq; | |
4202 | unsigned long flags; | |
c3c18640 | 4203 | int yielded = 0; |
d95f4122 MG |
4204 | |
4205 | local_irq_save(flags); | |
4206 | rq = this_rq(); | |
4207 | ||
4208 | again: | |
4209 | p_rq = task_rq(p); | |
7b270f60 PZ |
4210 | /* |
4211 | * If we're the only runnable task on the rq and target rq also | |
4212 | * has only one task, there's absolutely no point in yielding. | |
4213 | */ | |
4214 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4215 | yielded = -ESRCH; | |
4216 | goto out_irq; | |
4217 | } | |
4218 | ||
d95f4122 | 4219 | double_rq_lock(rq, p_rq); |
39e24d8f | 4220 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4221 | double_rq_unlock(rq, p_rq); |
4222 | goto again; | |
4223 | } | |
4224 | ||
4225 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4226 | goto out_unlock; |
d95f4122 MG |
4227 | |
4228 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4229 | goto out_unlock; |
d95f4122 MG |
4230 | |
4231 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4232 | goto out_unlock; |
d95f4122 MG |
4233 | |
4234 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4235 | if (yielded) { |
d95f4122 | 4236 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4237 | /* |
4238 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4239 | * fairness. | |
4240 | */ | |
4241 | if (preempt && rq != p_rq) | |
4242 | resched_task(p_rq->curr); | |
4243 | } | |
d95f4122 | 4244 | |
7b270f60 | 4245 | out_unlock: |
d95f4122 | 4246 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4247 | out_irq: |
d95f4122 MG |
4248 | local_irq_restore(flags); |
4249 | ||
7b270f60 | 4250 | if (yielded > 0) |
d95f4122 MG |
4251 | schedule(); |
4252 | ||
4253 | return yielded; | |
4254 | } | |
4255 | EXPORT_SYMBOL_GPL(yield_to); | |
4256 | ||
1da177e4 | 4257 | /* |
41a2d6cf | 4258 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4259 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4260 | */ |
4261 | void __sched io_schedule(void) | |
4262 | { | |
54d35f29 | 4263 | struct rq *rq = raw_rq(); |
1da177e4 | 4264 | |
0ff92245 | 4265 | delayacct_blkio_start(); |
1da177e4 | 4266 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4267 | blk_flush_plug(current); |
8f0dfc34 | 4268 | current->in_iowait = 1; |
1da177e4 | 4269 | schedule(); |
8f0dfc34 | 4270 | current->in_iowait = 0; |
1da177e4 | 4271 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4272 | delayacct_blkio_end(); |
1da177e4 | 4273 | } |
1da177e4 LT |
4274 | EXPORT_SYMBOL(io_schedule); |
4275 | ||
4276 | long __sched io_schedule_timeout(long timeout) | |
4277 | { | |
54d35f29 | 4278 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4279 | long ret; |
4280 | ||
0ff92245 | 4281 | delayacct_blkio_start(); |
1da177e4 | 4282 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4283 | blk_flush_plug(current); |
8f0dfc34 | 4284 | current->in_iowait = 1; |
1da177e4 | 4285 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4286 | current->in_iowait = 0; |
1da177e4 | 4287 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4288 | delayacct_blkio_end(); |
1da177e4 LT |
4289 | return ret; |
4290 | } | |
4291 | ||
4292 | /** | |
4293 | * sys_sched_get_priority_max - return maximum RT priority. | |
4294 | * @policy: scheduling class. | |
4295 | * | |
e69f6186 YB |
4296 | * Return: On success, this syscall returns the maximum |
4297 | * rt_priority that can be used by a given scheduling class. | |
4298 | * On failure, a negative error code is returned. | |
1da177e4 | 4299 | */ |
5add95d4 | 4300 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4301 | { |
4302 | int ret = -EINVAL; | |
4303 | ||
4304 | switch (policy) { | |
4305 | case SCHED_FIFO: | |
4306 | case SCHED_RR: | |
4307 | ret = MAX_USER_RT_PRIO-1; | |
4308 | break; | |
aab03e05 | 4309 | case SCHED_DEADLINE: |
1da177e4 | 4310 | case SCHED_NORMAL: |
b0a9499c | 4311 | case SCHED_BATCH: |
dd41f596 | 4312 | case SCHED_IDLE: |
1da177e4 LT |
4313 | ret = 0; |
4314 | break; | |
4315 | } | |
4316 | return ret; | |
4317 | } | |
4318 | ||
4319 | /** | |
4320 | * sys_sched_get_priority_min - return minimum RT priority. | |
4321 | * @policy: scheduling class. | |
4322 | * | |
e69f6186 YB |
4323 | * Return: On success, this syscall returns the minimum |
4324 | * rt_priority that can be used by a given scheduling class. | |
4325 | * On failure, a negative error code is returned. | |
1da177e4 | 4326 | */ |
5add95d4 | 4327 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4328 | { |
4329 | int ret = -EINVAL; | |
4330 | ||
4331 | switch (policy) { | |
4332 | case SCHED_FIFO: | |
4333 | case SCHED_RR: | |
4334 | ret = 1; | |
4335 | break; | |
aab03e05 | 4336 | case SCHED_DEADLINE: |
1da177e4 | 4337 | case SCHED_NORMAL: |
b0a9499c | 4338 | case SCHED_BATCH: |
dd41f596 | 4339 | case SCHED_IDLE: |
1da177e4 LT |
4340 | ret = 0; |
4341 | } | |
4342 | return ret; | |
4343 | } | |
4344 | ||
4345 | /** | |
4346 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4347 | * @pid: pid of the process. | |
4348 | * @interval: userspace pointer to the timeslice value. | |
4349 | * | |
4350 | * this syscall writes the default timeslice value of a given process | |
4351 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4352 | * |
4353 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4354 | * an error code. | |
1da177e4 | 4355 | */ |
17da2bd9 | 4356 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4357 | struct timespec __user *, interval) |
1da177e4 | 4358 | { |
36c8b586 | 4359 | struct task_struct *p; |
a4ec24b4 | 4360 | unsigned int time_slice; |
dba091b9 TG |
4361 | unsigned long flags; |
4362 | struct rq *rq; | |
3a5c359a | 4363 | int retval; |
1da177e4 | 4364 | struct timespec t; |
1da177e4 LT |
4365 | |
4366 | if (pid < 0) | |
3a5c359a | 4367 | return -EINVAL; |
1da177e4 LT |
4368 | |
4369 | retval = -ESRCH; | |
1a551ae7 | 4370 | rcu_read_lock(); |
1da177e4 LT |
4371 | p = find_process_by_pid(pid); |
4372 | if (!p) | |
4373 | goto out_unlock; | |
4374 | ||
4375 | retval = security_task_getscheduler(p); | |
4376 | if (retval) | |
4377 | goto out_unlock; | |
4378 | ||
dba091b9 | 4379 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4380 | time_slice = 0; |
4381 | if (p->sched_class->get_rr_interval) | |
4382 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4383 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4384 | |
1a551ae7 | 4385 | rcu_read_unlock(); |
a4ec24b4 | 4386 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4387 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4388 | return retval; |
3a5c359a | 4389 | |
1da177e4 | 4390 | out_unlock: |
1a551ae7 | 4391 | rcu_read_unlock(); |
1da177e4 LT |
4392 | return retval; |
4393 | } | |
4394 | ||
7c731e0a | 4395 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4396 | |
82a1fcb9 | 4397 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4398 | { |
1da177e4 | 4399 | unsigned long free = 0; |
4e79752c | 4400 | int ppid; |
36c8b586 | 4401 | unsigned state; |
1da177e4 | 4402 | |
1da177e4 | 4403 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4404 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4405 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4406 | #if BITS_PER_LONG == 32 |
1da177e4 | 4407 | if (state == TASK_RUNNING) |
3df0fc5b | 4408 | printk(KERN_CONT " running "); |
1da177e4 | 4409 | else |
3df0fc5b | 4410 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4411 | #else |
4412 | if (state == TASK_RUNNING) | |
3df0fc5b | 4413 | printk(KERN_CONT " running task "); |
1da177e4 | 4414 | else |
3df0fc5b | 4415 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4416 | #endif |
4417 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4418 | free = stack_not_used(p); |
1da177e4 | 4419 | #endif |
4e79752c PM |
4420 | rcu_read_lock(); |
4421 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4422 | rcu_read_unlock(); | |
3df0fc5b | 4423 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4424 | task_pid_nr(p), ppid, |
aa47b7e0 | 4425 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4426 | |
3d1cb205 | 4427 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4428 | show_stack(p, NULL); |
1da177e4 LT |
4429 | } |
4430 | ||
e59e2ae2 | 4431 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4432 | { |
36c8b586 | 4433 | struct task_struct *g, *p; |
1da177e4 | 4434 | |
4bd77321 | 4435 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4436 | printk(KERN_INFO |
4437 | " task PC stack pid father\n"); | |
1da177e4 | 4438 | #else |
3df0fc5b PZ |
4439 | printk(KERN_INFO |
4440 | " task PC stack pid father\n"); | |
1da177e4 | 4441 | #endif |
510f5acc | 4442 | rcu_read_lock(); |
1da177e4 LT |
4443 | do_each_thread(g, p) { |
4444 | /* | |
4445 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4446 | * console might take a lot of time: |
1da177e4 LT |
4447 | */ |
4448 | touch_nmi_watchdog(); | |
39bc89fd | 4449 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4450 | sched_show_task(p); |
1da177e4 LT |
4451 | } while_each_thread(g, p); |
4452 | ||
04c9167f JF |
4453 | touch_all_softlockup_watchdogs(); |
4454 | ||
dd41f596 IM |
4455 | #ifdef CONFIG_SCHED_DEBUG |
4456 | sysrq_sched_debug_show(); | |
4457 | #endif | |
510f5acc | 4458 | rcu_read_unlock(); |
e59e2ae2 IM |
4459 | /* |
4460 | * Only show locks if all tasks are dumped: | |
4461 | */ | |
93335a21 | 4462 | if (!state_filter) |
e59e2ae2 | 4463 | debug_show_all_locks(); |
1da177e4 LT |
4464 | } |
4465 | ||
0db0628d | 4466 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4467 | { |
dd41f596 | 4468 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4469 | } |
4470 | ||
f340c0d1 IM |
4471 | /** |
4472 | * init_idle - set up an idle thread for a given CPU | |
4473 | * @idle: task in question | |
4474 | * @cpu: cpu the idle task belongs to | |
4475 | * | |
4476 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4477 | * flag, to make booting more robust. | |
4478 | */ | |
0db0628d | 4479 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4480 | { |
70b97a7f | 4481 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4482 | unsigned long flags; |
4483 | ||
05fa785c | 4484 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4485 | |
5e1576ed | 4486 | __sched_fork(0, idle); |
06b83b5f | 4487 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4488 | idle->se.exec_start = sched_clock(); |
4489 | ||
1e1b6c51 | 4490 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4491 | /* |
4492 | * We're having a chicken and egg problem, even though we are | |
4493 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4494 | * lockdep check in task_group() will fail. | |
4495 | * | |
4496 | * Similar case to sched_fork(). / Alternatively we could | |
4497 | * use task_rq_lock() here and obtain the other rq->lock. | |
4498 | * | |
4499 | * Silence PROVE_RCU | |
4500 | */ | |
4501 | rcu_read_lock(); | |
dd41f596 | 4502 | __set_task_cpu(idle, cpu); |
6506cf6c | 4503 | rcu_read_unlock(); |
1da177e4 | 4504 | |
1da177e4 | 4505 | rq->curr = rq->idle = idle; |
77177856 | 4506 | idle->on_rq = 1; |
3ca7a440 PZ |
4507 | #if defined(CONFIG_SMP) |
4508 | idle->on_cpu = 1; | |
4866cde0 | 4509 | #endif |
05fa785c | 4510 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4511 | |
4512 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4513 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4514 | |
dd41f596 IM |
4515 | /* |
4516 | * The idle tasks have their own, simple scheduling class: | |
4517 | */ | |
4518 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4519 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4520 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4521 | #if defined(CONFIG_SMP) |
4522 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4523 | #endif | |
19978ca6 IM |
4524 | } |
4525 | ||
1da177e4 | 4526 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4527 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4528 | { | |
4529 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4530 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4531 | |
4532 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4533 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4534 | } |
4535 | ||
1da177e4 LT |
4536 | /* |
4537 | * This is how migration works: | |
4538 | * | |
969c7921 TH |
4539 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4540 | * stop_one_cpu(). | |
4541 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4542 | * off the CPU) | |
4543 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4544 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4545 | * it and puts it into the right queue. |
969c7921 TH |
4546 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4547 | * is done. | |
1da177e4 LT |
4548 | */ |
4549 | ||
4550 | /* | |
4551 | * Change a given task's CPU affinity. Migrate the thread to a | |
4552 | * proper CPU and schedule it away if the CPU it's executing on | |
4553 | * is removed from the allowed bitmask. | |
4554 | * | |
4555 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4556 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4557 | * call is not atomic; no spinlocks may be held. |
4558 | */ | |
96f874e2 | 4559 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4560 | { |
4561 | unsigned long flags; | |
70b97a7f | 4562 | struct rq *rq; |
969c7921 | 4563 | unsigned int dest_cpu; |
48f24c4d | 4564 | int ret = 0; |
1da177e4 LT |
4565 | |
4566 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4567 | |
db44fc01 YZ |
4568 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4569 | goto out; | |
4570 | ||
6ad4c188 | 4571 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4572 | ret = -EINVAL; |
4573 | goto out; | |
4574 | } | |
4575 | ||
1e1b6c51 | 4576 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4577 | |
1da177e4 | 4578 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4579 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4580 | goto out; |
4581 | ||
969c7921 | 4582 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4583 | if (p->on_rq) { |
969c7921 | 4584 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4585 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4586 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4587 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4588 | tlb_migrate_finish(p->mm); |
4589 | return 0; | |
4590 | } | |
4591 | out: | |
0122ec5b | 4592 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4593 | |
1da177e4 LT |
4594 | return ret; |
4595 | } | |
cd8ba7cd | 4596 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4597 | |
4598 | /* | |
41a2d6cf | 4599 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4600 | * this because either it can't run here any more (set_cpus_allowed() |
4601 | * away from this CPU, or CPU going down), or because we're | |
4602 | * attempting to rebalance this task on exec (sched_exec). | |
4603 | * | |
4604 | * So we race with normal scheduler movements, but that's OK, as long | |
4605 | * as the task is no longer on this CPU. | |
efc30814 KK |
4606 | * |
4607 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4608 | */ |
efc30814 | 4609 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4610 | { |
70b97a7f | 4611 | struct rq *rq_dest, *rq_src; |
e2912009 | 4612 | int ret = 0; |
1da177e4 | 4613 | |
e761b772 | 4614 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4615 | return ret; |
1da177e4 LT |
4616 | |
4617 | rq_src = cpu_rq(src_cpu); | |
4618 | rq_dest = cpu_rq(dest_cpu); | |
4619 | ||
0122ec5b | 4620 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4621 | double_rq_lock(rq_src, rq_dest); |
4622 | /* Already moved. */ | |
4623 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4624 | goto done; |
1da177e4 | 4625 | /* Affinity changed (again). */ |
fa17b507 | 4626 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4627 | goto fail; |
1da177e4 | 4628 | |
e2912009 PZ |
4629 | /* |
4630 | * If we're not on a rq, the next wake-up will ensure we're | |
4631 | * placed properly. | |
4632 | */ | |
fd2f4419 | 4633 | if (p->on_rq) { |
4ca9b72b | 4634 | dequeue_task(rq_src, p, 0); |
e2912009 | 4635 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4636 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4637 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4638 | } |
b1e38734 | 4639 | done: |
efc30814 | 4640 | ret = 1; |
b1e38734 | 4641 | fail: |
1da177e4 | 4642 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4643 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4644 | return ret; |
1da177e4 LT |
4645 | } |
4646 | ||
e6628d5b MG |
4647 | #ifdef CONFIG_NUMA_BALANCING |
4648 | /* Migrate current task p to target_cpu */ | |
4649 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4650 | { | |
4651 | struct migration_arg arg = { p, target_cpu }; | |
4652 | int curr_cpu = task_cpu(p); | |
4653 | ||
4654 | if (curr_cpu == target_cpu) | |
4655 | return 0; | |
4656 | ||
4657 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4658 | return -EINVAL; | |
4659 | ||
4660 | /* TODO: This is not properly updating schedstats */ | |
4661 | ||
286549dc | 4662 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4663 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4664 | } | |
0ec8aa00 PZ |
4665 | |
4666 | /* | |
4667 | * Requeue a task on a given node and accurately track the number of NUMA | |
4668 | * tasks on the runqueues | |
4669 | */ | |
4670 | void sched_setnuma(struct task_struct *p, int nid) | |
4671 | { | |
4672 | struct rq *rq; | |
4673 | unsigned long flags; | |
4674 | bool on_rq, running; | |
4675 | ||
4676 | rq = task_rq_lock(p, &flags); | |
4677 | on_rq = p->on_rq; | |
4678 | running = task_current(rq, p); | |
4679 | ||
4680 | if (on_rq) | |
4681 | dequeue_task(rq, p, 0); | |
4682 | if (running) | |
4683 | p->sched_class->put_prev_task(rq, p); | |
4684 | ||
4685 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4686 | |
4687 | if (running) | |
4688 | p->sched_class->set_curr_task(rq); | |
4689 | if (on_rq) | |
4690 | enqueue_task(rq, p, 0); | |
4691 | task_rq_unlock(rq, p, &flags); | |
4692 | } | |
e6628d5b MG |
4693 | #endif |
4694 | ||
1da177e4 | 4695 | /* |
969c7921 TH |
4696 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4697 | * and performs thread migration by bumping thread off CPU then | |
4698 | * 'pushing' onto another runqueue. | |
1da177e4 | 4699 | */ |
969c7921 | 4700 | static int migration_cpu_stop(void *data) |
1da177e4 | 4701 | { |
969c7921 | 4702 | struct migration_arg *arg = data; |
f7b4cddc | 4703 | |
969c7921 TH |
4704 | /* |
4705 | * The original target cpu might have gone down and we might | |
4706 | * be on another cpu but it doesn't matter. | |
4707 | */ | |
f7b4cddc | 4708 | local_irq_disable(); |
969c7921 | 4709 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4710 | local_irq_enable(); |
1da177e4 | 4711 | return 0; |
f7b4cddc ON |
4712 | } |
4713 | ||
1da177e4 | 4714 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4715 | |
054b9108 | 4716 | /* |
48c5ccae PZ |
4717 | * Ensures that the idle task is using init_mm right before its cpu goes |
4718 | * offline. | |
054b9108 | 4719 | */ |
48c5ccae | 4720 | void idle_task_exit(void) |
1da177e4 | 4721 | { |
48c5ccae | 4722 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4723 | |
48c5ccae | 4724 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4725 | |
a53efe5f | 4726 | if (mm != &init_mm) { |
48c5ccae | 4727 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4728 | finish_arch_post_lock_switch(); |
4729 | } | |
48c5ccae | 4730 | mmdrop(mm); |
1da177e4 LT |
4731 | } |
4732 | ||
4733 | /* | |
5d180232 PZ |
4734 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4735 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4736 | * nr_active count is stable. | |
4737 | * | |
4738 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4739 | */ |
5d180232 | 4740 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4741 | { |
5d180232 PZ |
4742 | long delta = calc_load_fold_active(rq); |
4743 | if (delta) | |
4744 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4745 | } |
4746 | ||
3f1d2a31 PZ |
4747 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4748 | { | |
4749 | } | |
4750 | ||
4751 | static const struct sched_class fake_sched_class = { | |
4752 | .put_prev_task = put_prev_task_fake, | |
4753 | }; | |
4754 | ||
4755 | static struct task_struct fake_task = { | |
4756 | /* | |
4757 | * Avoid pull_{rt,dl}_task() | |
4758 | */ | |
4759 | .prio = MAX_PRIO + 1, | |
4760 | .sched_class = &fake_sched_class, | |
4761 | }; | |
4762 | ||
48f24c4d | 4763 | /* |
48c5ccae PZ |
4764 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4765 | * try_to_wake_up()->select_task_rq(). | |
4766 | * | |
4767 | * Called with rq->lock held even though we'er in stop_machine() and | |
4768 | * there's no concurrency possible, we hold the required locks anyway | |
4769 | * because of lock validation efforts. | |
1da177e4 | 4770 | */ |
48c5ccae | 4771 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4772 | { |
70b97a7f | 4773 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4774 | struct task_struct *next, *stop = rq->stop; |
4775 | int dest_cpu; | |
1da177e4 LT |
4776 | |
4777 | /* | |
48c5ccae PZ |
4778 | * Fudge the rq selection such that the below task selection loop |
4779 | * doesn't get stuck on the currently eligible stop task. | |
4780 | * | |
4781 | * We're currently inside stop_machine() and the rq is either stuck | |
4782 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4783 | * either way we should never end up calling schedule() until we're | |
4784 | * done here. | |
1da177e4 | 4785 | */ |
48c5ccae | 4786 | rq->stop = NULL; |
48f24c4d | 4787 | |
77bd3970 FW |
4788 | /* |
4789 | * put_prev_task() and pick_next_task() sched | |
4790 | * class method both need to have an up-to-date | |
4791 | * value of rq->clock[_task] | |
4792 | */ | |
4793 | update_rq_clock(rq); | |
4794 | ||
dd41f596 | 4795 | for ( ; ; ) { |
48c5ccae PZ |
4796 | /* |
4797 | * There's this thread running, bail when that's the only | |
4798 | * remaining thread. | |
4799 | */ | |
4800 | if (rq->nr_running == 1) | |
dd41f596 | 4801 | break; |
48c5ccae | 4802 | |
3f1d2a31 | 4803 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4804 | BUG_ON(!next); |
79c53799 | 4805 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4806 | |
48c5ccae PZ |
4807 | /* Find suitable destination for @next, with force if needed. */ |
4808 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4809 | raw_spin_unlock(&rq->lock); | |
4810 | ||
4811 | __migrate_task(next, dead_cpu, dest_cpu); | |
4812 | ||
4813 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4814 | } |
dce48a84 | 4815 | |
48c5ccae | 4816 | rq->stop = stop; |
dce48a84 | 4817 | } |
48c5ccae | 4818 | |
1da177e4 LT |
4819 | #endif /* CONFIG_HOTPLUG_CPU */ |
4820 | ||
e692ab53 NP |
4821 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4822 | ||
4823 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4824 | { |
4825 | .procname = "sched_domain", | |
c57baf1e | 4826 | .mode = 0555, |
e0361851 | 4827 | }, |
56992309 | 4828 | {} |
e692ab53 NP |
4829 | }; |
4830 | ||
4831 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4832 | { |
4833 | .procname = "kernel", | |
c57baf1e | 4834 | .mode = 0555, |
e0361851 AD |
4835 | .child = sd_ctl_dir, |
4836 | }, | |
56992309 | 4837 | {} |
e692ab53 NP |
4838 | }; |
4839 | ||
4840 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4841 | { | |
4842 | struct ctl_table *entry = | |
5cf9f062 | 4843 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4844 | |
e692ab53 NP |
4845 | return entry; |
4846 | } | |
4847 | ||
6382bc90 MM |
4848 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4849 | { | |
cd790076 | 4850 | struct ctl_table *entry; |
6382bc90 | 4851 | |
cd790076 MM |
4852 | /* |
4853 | * In the intermediate directories, both the child directory and | |
4854 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4855 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4856 | * static strings and all have proc handlers. |
4857 | */ | |
4858 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4859 | if (entry->child) |
4860 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4861 | if (entry->proc_handler == NULL) |
4862 | kfree(entry->procname); | |
4863 | } | |
6382bc90 MM |
4864 | |
4865 | kfree(*tablep); | |
4866 | *tablep = NULL; | |
4867 | } | |
4868 | ||
201c373e | 4869 | static int min_load_idx = 0; |
fd9b86d3 | 4870 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4871 | |
e692ab53 | 4872 | static void |
e0361851 | 4873 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4874 | const char *procname, void *data, int maxlen, |
201c373e NK |
4875 | umode_t mode, proc_handler *proc_handler, |
4876 | bool load_idx) | |
e692ab53 | 4877 | { |
e692ab53 NP |
4878 | entry->procname = procname; |
4879 | entry->data = data; | |
4880 | entry->maxlen = maxlen; | |
4881 | entry->mode = mode; | |
4882 | entry->proc_handler = proc_handler; | |
201c373e NK |
4883 | |
4884 | if (load_idx) { | |
4885 | entry->extra1 = &min_load_idx; | |
4886 | entry->extra2 = &max_load_idx; | |
4887 | } | |
e692ab53 NP |
4888 | } |
4889 | ||
4890 | static struct ctl_table * | |
4891 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4892 | { | |
37e6bae8 | 4893 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4894 | |
ad1cdc1d MM |
4895 | if (table == NULL) |
4896 | return NULL; | |
4897 | ||
e0361851 | 4898 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4899 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4900 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4901 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4902 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4903 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4904 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4905 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4906 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4907 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4908 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4909 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4910 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4911 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4912 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4913 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4914 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4915 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4916 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4917 | &sd->cache_nice_tries, |
201c373e | 4918 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4919 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4920 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4921 | set_table_entry(&table[11], "max_newidle_lb_cost", |
4922 | &sd->max_newidle_lb_cost, | |
4923 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
4924 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 4925 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 4926 | /* &table[13] is terminator */ |
e692ab53 NP |
4927 | |
4928 | return table; | |
4929 | } | |
4930 | ||
be7002e6 | 4931 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4932 | { |
4933 | struct ctl_table *entry, *table; | |
4934 | struct sched_domain *sd; | |
4935 | int domain_num = 0, i; | |
4936 | char buf[32]; | |
4937 | ||
4938 | for_each_domain(cpu, sd) | |
4939 | domain_num++; | |
4940 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4941 | if (table == NULL) |
4942 | return NULL; | |
e692ab53 NP |
4943 | |
4944 | i = 0; | |
4945 | for_each_domain(cpu, sd) { | |
4946 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4947 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4948 | entry->mode = 0555; |
e692ab53 NP |
4949 | entry->child = sd_alloc_ctl_domain_table(sd); |
4950 | entry++; | |
4951 | i++; | |
4952 | } | |
4953 | return table; | |
4954 | } | |
4955 | ||
4956 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4957 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4958 | { |
6ad4c188 | 4959 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4960 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4961 | char buf[32]; | |
4962 | ||
7378547f MM |
4963 | WARN_ON(sd_ctl_dir[0].child); |
4964 | sd_ctl_dir[0].child = entry; | |
4965 | ||
ad1cdc1d MM |
4966 | if (entry == NULL) |
4967 | return; | |
4968 | ||
6ad4c188 | 4969 | for_each_possible_cpu(i) { |
e692ab53 | 4970 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4971 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4972 | entry->mode = 0555; |
e692ab53 | 4973 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4974 | entry++; |
e692ab53 | 4975 | } |
7378547f MM |
4976 | |
4977 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4978 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4979 | } | |
6382bc90 | 4980 | |
7378547f | 4981 | /* may be called multiple times per register */ |
6382bc90 MM |
4982 | static void unregister_sched_domain_sysctl(void) |
4983 | { | |
7378547f MM |
4984 | if (sd_sysctl_header) |
4985 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4986 | sd_sysctl_header = NULL; |
7378547f MM |
4987 | if (sd_ctl_dir[0].child) |
4988 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 4989 | } |
e692ab53 | 4990 | #else |
6382bc90 MM |
4991 | static void register_sched_domain_sysctl(void) |
4992 | { | |
4993 | } | |
4994 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
4995 | { |
4996 | } | |
4997 | #endif | |
4998 | ||
1f11eb6a GH |
4999 | static void set_rq_online(struct rq *rq) |
5000 | { | |
5001 | if (!rq->online) { | |
5002 | const struct sched_class *class; | |
5003 | ||
c6c4927b | 5004 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5005 | rq->online = 1; |
5006 | ||
5007 | for_each_class(class) { | |
5008 | if (class->rq_online) | |
5009 | class->rq_online(rq); | |
5010 | } | |
5011 | } | |
5012 | } | |
5013 | ||
5014 | static void set_rq_offline(struct rq *rq) | |
5015 | { | |
5016 | if (rq->online) { | |
5017 | const struct sched_class *class; | |
5018 | ||
5019 | for_each_class(class) { | |
5020 | if (class->rq_offline) | |
5021 | class->rq_offline(rq); | |
5022 | } | |
5023 | ||
c6c4927b | 5024 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5025 | rq->online = 0; |
5026 | } | |
5027 | } | |
5028 | ||
1da177e4 LT |
5029 | /* |
5030 | * migration_call - callback that gets triggered when a CPU is added. | |
5031 | * Here we can start up the necessary migration thread for the new CPU. | |
5032 | */ | |
0db0628d | 5033 | static int |
48f24c4d | 5034 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5035 | { |
48f24c4d | 5036 | int cpu = (long)hcpu; |
1da177e4 | 5037 | unsigned long flags; |
969c7921 | 5038 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5039 | |
48c5ccae | 5040 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5041 | |
1da177e4 | 5042 | case CPU_UP_PREPARE: |
a468d389 | 5043 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5044 | break; |
48f24c4d | 5045 | |
1da177e4 | 5046 | case CPU_ONLINE: |
1f94ef59 | 5047 | /* Update our root-domain */ |
05fa785c | 5048 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5049 | if (rq->rd) { |
c6c4927b | 5050 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5051 | |
5052 | set_rq_online(rq); | |
1f94ef59 | 5053 | } |
05fa785c | 5054 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5055 | break; |
48f24c4d | 5056 | |
1da177e4 | 5057 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5058 | case CPU_DYING: |
317f3941 | 5059 | sched_ttwu_pending(); |
57d885fe | 5060 | /* Update our root-domain */ |
05fa785c | 5061 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5062 | if (rq->rd) { |
c6c4927b | 5063 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5064 | set_rq_offline(rq); |
57d885fe | 5065 | } |
48c5ccae PZ |
5066 | migrate_tasks(cpu); |
5067 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5068 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5069 | break; |
48c5ccae | 5070 | |
5d180232 | 5071 | case CPU_DEAD: |
f319da0c | 5072 | calc_load_migrate(rq); |
57d885fe | 5073 | break; |
1da177e4 LT |
5074 | #endif |
5075 | } | |
49c022e6 PZ |
5076 | |
5077 | update_max_interval(); | |
5078 | ||
1da177e4 LT |
5079 | return NOTIFY_OK; |
5080 | } | |
5081 | ||
f38b0820 PM |
5082 | /* |
5083 | * Register at high priority so that task migration (migrate_all_tasks) | |
5084 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5085 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5086 | */ |
0db0628d | 5087 | static struct notifier_block migration_notifier = { |
1da177e4 | 5088 | .notifier_call = migration_call, |
50a323b7 | 5089 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5090 | }; |
5091 | ||
a803f026 CM |
5092 | static void __cpuinit set_cpu_rq_start_time(void) |
5093 | { | |
5094 | int cpu = smp_processor_id(); | |
5095 | struct rq *rq = cpu_rq(cpu); | |
5096 | rq->age_stamp = sched_clock_cpu(cpu); | |
5097 | } | |
5098 | ||
0db0628d | 5099 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5100 | unsigned long action, void *hcpu) |
5101 | { | |
5102 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5103 | case CPU_STARTING: |
5104 | set_cpu_rq_start_time(); | |
5105 | return NOTIFY_OK; | |
3a101d05 TH |
5106 | case CPU_DOWN_FAILED: |
5107 | set_cpu_active((long)hcpu, true); | |
5108 | return NOTIFY_OK; | |
5109 | default: | |
5110 | return NOTIFY_DONE; | |
5111 | } | |
5112 | } | |
5113 | ||
0db0628d | 5114 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5115 | unsigned long action, void *hcpu) |
5116 | { | |
de212f18 PZ |
5117 | unsigned long flags; |
5118 | long cpu = (long)hcpu; | |
5119 | ||
3a101d05 TH |
5120 | switch (action & ~CPU_TASKS_FROZEN) { |
5121 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5122 | set_cpu_active(cpu, false); |
5123 | ||
5124 | /* explicitly allow suspend */ | |
5125 | if (!(action & CPU_TASKS_FROZEN)) { | |
5126 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5127 | bool overflow; | |
5128 | int cpus; | |
5129 | ||
5130 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5131 | cpus = dl_bw_cpus(cpu); | |
5132 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5133 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5134 | ||
5135 | if (overflow) | |
5136 | return notifier_from_errno(-EBUSY); | |
5137 | } | |
3a101d05 | 5138 | return NOTIFY_OK; |
3a101d05 | 5139 | } |
de212f18 PZ |
5140 | |
5141 | return NOTIFY_DONE; | |
3a101d05 TH |
5142 | } |
5143 | ||
7babe8db | 5144 | static int __init migration_init(void) |
1da177e4 LT |
5145 | { |
5146 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5147 | int err; |
48f24c4d | 5148 | |
3a101d05 | 5149 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5150 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5151 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5152 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5153 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5154 | |
3a101d05 TH |
5155 | /* Register cpu active notifiers */ |
5156 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5157 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5158 | ||
a004cd42 | 5159 | return 0; |
1da177e4 | 5160 | } |
7babe8db | 5161 | early_initcall(migration_init); |
1da177e4 LT |
5162 | #endif |
5163 | ||
5164 | #ifdef CONFIG_SMP | |
476f3534 | 5165 | |
4cb98839 PZ |
5166 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5167 | ||
3e9830dc | 5168 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5169 | |
d039ac60 | 5170 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5171 | |
d039ac60 | 5172 | static int __init sched_debug_setup(char *str) |
f6630114 | 5173 | { |
d039ac60 | 5174 | sched_debug_enabled = 1; |
f6630114 MT |
5175 | |
5176 | return 0; | |
5177 | } | |
d039ac60 PZ |
5178 | early_param("sched_debug", sched_debug_setup); |
5179 | ||
5180 | static inline bool sched_debug(void) | |
5181 | { | |
5182 | return sched_debug_enabled; | |
5183 | } | |
f6630114 | 5184 | |
7c16ec58 | 5185 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5186 | struct cpumask *groupmask) |
1da177e4 | 5187 | { |
4dcf6aff | 5188 | struct sched_group *group = sd->groups; |
434d53b0 | 5189 | char str[256]; |
1da177e4 | 5190 | |
968ea6d8 | 5191 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5192 | cpumask_clear(groupmask); |
4dcf6aff IM |
5193 | |
5194 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5195 | ||
5196 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5197 | printk("does not load-balance\n"); |
4dcf6aff | 5198 | if (sd->parent) |
3df0fc5b PZ |
5199 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5200 | " has parent"); | |
4dcf6aff | 5201 | return -1; |
41c7ce9a NP |
5202 | } |
5203 | ||
3df0fc5b | 5204 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5205 | |
758b2cdc | 5206 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5207 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5208 | "CPU%d\n", cpu); | |
4dcf6aff | 5209 | } |
758b2cdc | 5210 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5211 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5212 | " CPU%d\n", cpu); | |
4dcf6aff | 5213 | } |
1da177e4 | 5214 | |
4dcf6aff | 5215 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5216 | do { |
4dcf6aff | 5217 | if (!group) { |
3df0fc5b PZ |
5218 | printk("\n"); |
5219 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5220 | break; |
5221 | } | |
5222 | ||
c3decf0d | 5223 | /* |
63b2ca30 NP |
5224 | * Even though we initialize ->capacity to something semi-sane, |
5225 | * we leave capacity_orig unset. This allows us to detect if | |
c3decf0d PZ |
5226 | * domain iteration is still funny without causing /0 traps. |
5227 | */ | |
63b2ca30 | 5228 | if (!group->sgc->capacity_orig) { |
3df0fc5b | 5229 | printk(KERN_CONT "\n"); |
63b2ca30 | 5230 | printk(KERN_ERR "ERROR: domain->cpu_capacity not set\n"); |
4dcf6aff IM |
5231 | break; |
5232 | } | |
1da177e4 | 5233 | |
758b2cdc | 5234 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5235 | printk(KERN_CONT "\n"); |
5236 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5237 | break; |
5238 | } | |
1da177e4 | 5239 | |
cb83b629 PZ |
5240 | if (!(sd->flags & SD_OVERLAP) && |
5241 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5242 | printk(KERN_CONT "\n"); |
5243 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5244 | break; |
5245 | } | |
1da177e4 | 5246 | |
758b2cdc | 5247 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5248 | |
968ea6d8 | 5249 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5250 | |
3df0fc5b | 5251 | printk(KERN_CONT " %s", str); |
63b2ca30 NP |
5252 | if (group->sgc->capacity != SCHED_POWER_SCALE) { |
5253 | printk(KERN_CONT " (cpu_capacity = %d)", | |
5254 | group->sgc->capacity); | |
381512cf | 5255 | } |
1da177e4 | 5256 | |
4dcf6aff IM |
5257 | group = group->next; |
5258 | } while (group != sd->groups); | |
3df0fc5b | 5259 | printk(KERN_CONT "\n"); |
1da177e4 | 5260 | |
758b2cdc | 5261 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5262 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5263 | |
758b2cdc RR |
5264 | if (sd->parent && |
5265 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5266 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5267 | "of domain->span\n"); | |
4dcf6aff IM |
5268 | return 0; |
5269 | } | |
1da177e4 | 5270 | |
4dcf6aff IM |
5271 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5272 | { | |
5273 | int level = 0; | |
1da177e4 | 5274 | |
d039ac60 | 5275 | if (!sched_debug_enabled) |
f6630114 MT |
5276 | return; |
5277 | ||
4dcf6aff IM |
5278 | if (!sd) { |
5279 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5280 | return; | |
5281 | } | |
1da177e4 | 5282 | |
4dcf6aff IM |
5283 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5284 | ||
5285 | for (;;) { | |
4cb98839 | 5286 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5287 | break; |
1da177e4 LT |
5288 | level++; |
5289 | sd = sd->parent; | |
33859f7f | 5290 | if (!sd) |
4dcf6aff IM |
5291 | break; |
5292 | } | |
1da177e4 | 5293 | } |
6d6bc0ad | 5294 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5295 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5296 | static inline bool sched_debug(void) |
5297 | { | |
5298 | return false; | |
5299 | } | |
6d6bc0ad | 5300 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5301 | |
1a20ff27 | 5302 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5303 | { |
758b2cdc | 5304 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5305 | return 1; |
5306 | ||
5307 | /* Following flags need at least 2 groups */ | |
5308 | if (sd->flags & (SD_LOAD_BALANCE | | |
5309 | SD_BALANCE_NEWIDLE | | |
5310 | SD_BALANCE_FORK | | |
89c4710e SS |
5311 | SD_BALANCE_EXEC | |
5312 | SD_SHARE_CPUPOWER | | |
d77b3ed5 VG |
5313 | SD_SHARE_PKG_RESOURCES | |
5314 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5315 | if (sd->groups != sd->groups->next) |
5316 | return 0; | |
5317 | } | |
5318 | ||
5319 | /* Following flags don't use groups */ | |
c88d5910 | 5320 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5321 | return 0; |
5322 | ||
5323 | return 1; | |
5324 | } | |
5325 | ||
48f24c4d IM |
5326 | static int |
5327 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5328 | { |
5329 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5330 | ||
5331 | if (sd_degenerate(parent)) | |
5332 | return 1; | |
5333 | ||
758b2cdc | 5334 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5335 | return 0; |
5336 | ||
245af2c7 SS |
5337 | /* Flags needing groups don't count if only 1 group in parent */ |
5338 | if (parent->groups == parent->groups->next) { | |
5339 | pflags &= ~(SD_LOAD_BALANCE | | |
5340 | SD_BALANCE_NEWIDLE | | |
5341 | SD_BALANCE_FORK | | |
89c4710e SS |
5342 | SD_BALANCE_EXEC | |
5343 | SD_SHARE_CPUPOWER | | |
10866e62 | 5344 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5345 | SD_PREFER_SIBLING | |
5346 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5347 | if (nr_node_ids == 1) |
5348 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5349 | } |
5350 | if (~cflags & pflags) | |
5351 | return 0; | |
5352 | ||
5353 | return 1; | |
5354 | } | |
5355 | ||
dce840a0 | 5356 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5357 | { |
dce840a0 | 5358 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5359 | |
68e74568 | 5360 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5361 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5362 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5363 | free_cpumask_var(rd->rto_mask); |
5364 | free_cpumask_var(rd->online); | |
5365 | free_cpumask_var(rd->span); | |
5366 | kfree(rd); | |
5367 | } | |
5368 | ||
57d885fe GH |
5369 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5370 | { | |
a0490fa3 | 5371 | struct root_domain *old_rd = NULL; |
57d885fe | 5372 | unsigned long flags; |
57d885fe | 5373 | |
05fa785c | 5374 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5375 | |
5376 | if (rq->rd) { | |
a0490fa3 | 5377 | old_rd = rq->rd; |
57d885fe | 5378 | |
c6c4927b | 5379 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5380 | set_rq_offline(rq); |
57d885fe | 5381 | |
c6c4927b | 5382 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5383 | |
a0490fa3 | 5384 | /* |
0515973f | 5385 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5386 | * set old_rd to NULL to skip the freeing later |
5387 | * in this function: | |
5388 | */ | |
5389 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5390 | old_rd = NULL; | |
57d885fe GH |
5391 | } |
5392 | ||
5393 | atomic_inc(&rd->refcount); | |
5394 | rq->rd = rd; | |
5395 | ||
c6c4927b | 5396 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5397 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5398 | set_rq_online(rq); |
57d885fe | 5399 | |
05fa785c | 5400 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5401 | |
5402 | if (old_rd) | |
dce840a0 | 5403 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5404 | } |
5405 | ||
68c38fc3 | 5406 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5407 | { |
5408 | memset(rd, 0, sizeof(*rd)); | |
5409 | ||
68c38fc3 | 5410 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5411 | goto out; |
68c38fc3 | 5412 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5413 | goto free_span; |
1baca4ce | 5414 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5415 | goto free_online; |
1baca4ce JL |
5416 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5417 | goto free_dlo_mask; | |
6e0534f2 | 5418 | |
332ac17e | 5419 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5420 | if (cpudl_init(&rd->cpudl) != 0) |
5421 | goto free_dlo_mask; | |
332ac17e | 5422 | |
68c38fc3 | 5423 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5424 | goto free_rto_mask; |
c6c4927b | 5425 | return 0; |
6e0534f2 | 5426 | |
68e74568 RR |
5427 | free_rto_mask: |
5428 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5429 | free_dlo_mask: |
5430 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5431 | free_online: |
5432 | free_cpumask_var(rd->online); | |
5433 | free_span: | |
5434 | free_cpumask_var(rd->span); | |
0c910d28 | 5435 | out: |
c6c4927b | 5436 | return -ENOMEM; |
57d885fe GH |
5437 | } |
5438 | ||
029632fb PZ |
5439 | /* |
5440 | * By default the system creates a single root-domain with all cpus as | |
5441 | * members (mimicking the global state we have today). | |
5442 | */ | |
5443 | struct root_domain def_root_domain; | |
5444 | ||
57d885fe GH |
5445 | static void init_defrootdomain(void) |
5446 | { | |
68c38fc3 | 5447 | init_rootdomain(&def_root_domain); |
c6c4927b | 5448 | |
57d885fe GH |
5449 | atomic_set(&def_root_domain.refcount, 1); |
5450 | } | |
5451 | ||
dc938520 | 5452 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5453 | { |
5454 | struct root_domain *rd; | |
5455 | ||
5456 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5457 | if (!rd) | |
5458 | return NULL; | |
5459 | ||
68c38fc3 | 5460 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5461 | kfree(rd); |
5462 | return NULL; | |
5463 | } | |
57d885fe GH |
5464 | |
5465 | return rd; | |
5466 | } | |
5467 | ||
63b2ca30 | 5468 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5469 | { |
5470 | struct sched_group *tmp, *first; | |
5471 | ||
5472 | if (!sg) | |
5473 | return; | |
5474 | ||
5475 | first = sg; | |
5476 | do { | |
5477 | tmp = sg->next; | |
5478 | ||
63b2ca30 NP |
5479 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5480 | kfree(sg->sgc); | |
e3589f6c PZ |
5481 | |
5482 | kfree(sg); | |
5483 | sg = tmp; | |
5484 | } while (sg != first); | |
5485 | } | |
5486 | ||
dce840a0 PZ |
5487 | static void free_sched_domain(struct rcu_head *rcu) |
5488 | { | |
5489 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5490 | |
5491 | /* | |
5492 | * If its an overlapping domain it has private groups, iterate and | |
5493 | * nuke them all. | |
5494 | */ | |
5495 | if (sd->flags & SD_OVERLAP) { | |
5496 | free_sched_groups(sd->groups, 1); | |
5497 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5498 | kfree(sd->groups->sgc); |
dce840a0 | 5499 | kfree(sd->groups); |
9c3f75cb | 5500 | } |
dce840a0 PZ |
5501 | kfree(sd); |
5502 | } | |
5503 | ||
5504 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5505 | { | |
5506 | call_rcu(&sd->rcu, free_sched_domain); | |
5507 | } | |
5508 | ||
5509 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5510 | { | |
5511 | for (; sd; sd = sd->parent) | |
5512 | destroy_sched_domain(sd, cpu); | |
5513 | } | |
5514 | ||
518cd623 PZ |
5515 | /* |
5516 | * Keep a special pointer to the highest sched_domain that has | |
5517 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5518 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5519 | * | |
5520 | * Also keep a unique ID per domain (we use the first cpu number in | |
5521 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5522 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5523 | */ |
5524 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5525 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5526 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5527 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5528 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5529 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5530 | |
5531 | static void update_top_cache_domain(int cpu) | |
5532 | { | |
5533 | struct sched_domain *sd; | |
5d4cf996 | 5534 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5535 | int id = cpu; |
7d9ffa89 | 5536 | int size = 1; |
518cd623 PZ |
5537 | |
5538 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5539 | if (sd) { |
518cd623 | 5540 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5541 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5542 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5543 | } |
5d4cf996 | 5544 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5545 | |
5546 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5547 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5548 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5549 | |
5550 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5551 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5552 | |
5553 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5554 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5555 | } |
5556 | ||
1da177e4 | 5557 | /* |
0eab9146 | 5558 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5559 | * hold the hotplug lock. |
5560 | */ | |
0eab9146 IM |
5561 | static void |
5562 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5563 | { |
70b97a7f | 5564 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5565 | struct sched_domain *tmp; |
5566 | ||
5567 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5568 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5569 | struct sched_domain *parent = tmp->parent; |
5570 | if (!parent) | |
5571 | break; | |
f29c9b1c | 5572 | |
1a848870 | 5573 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5574 | tmp->parent = parent->parent; |
1a848870 SS |
5575 | if (parent->parent) |
5576 | parent->parent->child = tmp; | |
10866e62 PZ |
5577 | /* |
5578 | * Transfer SD_PREFER_SIBLING down in case of a | |
5579 | * degenerate parent; the spans match for this | |
5580 | * so the property transfers. | |
5581 | */ | |
5582 | if (parent->flags & SD_PREFER_SIBLING) | |
5583 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5584 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5585 | } else |
5586 | tmp = tmp->parent; | |
245af2c7 SS |
5587 | } |
5588 | ||
1a848870 | 5589 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5590 | tmp = sd; |
245af2c7 | 5591 | sd = sd->parent; |
dce840a0 | 5592 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5593 | if (sd) |
5594 | sd->child = NULL; | |
5595 | } | |
1da177e4 | 5596 | |
4cb98839 | 5597 | sched_domain_debug(sd, cpu); |
1da177e4 | 5598 | |
57d885fe | 5599 | rq_attach_root(rq, rd); |
dce840a0 | 5600 | tmp = rq->sd; |
674311d5 | 5601 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5602 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5603 | |
5604 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5605 | } |
5606 | ||
5607 | /* cpus with isolated domains */ | |
dcc30a35 | 5608 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5609 | |
5610 | /* Setup the mask of cpus configured for isolated domains */ | |
5611 | static int __init isolated_cpu_setup(char *str) | |
5612 | { | |
bdddd296 | 5613 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5614 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5615 | return 1; |
5616 | } | |
5617 | ||
8927f494 | 5618 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5619 | |
49a02c51 | 5620 | struct s_data { |
21d42ccf | 5621 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5622 | struct root_domain *rd; |
5623 | }; | |
5624 | ||
2109b99e | 5625 | enum s_alloc { |
2109b99e | 5626 | sa_rootdomain, |
21d42ccf | 5627 | sa_sd, |
dce840a0 | 5628 | sa_sd_storage, |
2109b99e AH |
5629 | sa_none, |
5630 | }; | |
5631 | ||
c1174876 PZ |
5632 | /* |
5633 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5634 | * domain traversal. | |
5635 | * | |
5636 | * Asymmetric node setups can result in situations where the domain tree is of | |
5637 | * unequal depth, make sure to skip domains that already cover the entire | |
5638 | * range. | |
5639 | * | |
5640 | * In that case build_sched_domains() will have terminated the iteration early | |
5641 | * and our sibling sd spans will be empty. Domains should always include the | |
5642 | * cpu they're built on, so check that. | |
5643 | * | |
5644 | */ | |
5645 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5646 | { | |
5647 | const struct cpumask *span = sched_domain_span(sd); | |
5648 | struct sd_data *sdd = sd->private; | |
5649 | struct sched_domain *sibling; | |
5650 | int i; | |
5651 | ||
5652 | for_each_cpu(i, span) { | |
5653 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5654 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5655 | continue; | |
5656 | ||
5657 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5658 | } | |
5659 | } | |
5660 | ||
5661 | /* | |
5662 | * Return the canonical balance cpu for this group, this is the first cpu | |
5663 | * of this group that's also in the iteration mask. | |
5664 | */ | |
5665 | int group_balance_cpu(struct sched_group *sg) | |
5666 | { | |
5667 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5668 | } | |
5669 | ||
e3589f6c PZ |
5670 | static int |
5671 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5672 | { | |
5673 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5674 | const struct cpumask *span = sched_domain_span(sd); | |
5675 | struct cpumask *covered = sched_domains_tmpmask; | |
5676 | struct sd_data *sdd = sd->private; | |
5677 | struct sched_domain *child; | |
5678 | int i; | |
5679 | ||
5680 | cpumask_clear(covered); | |
5681 | ||
5682 | for_each_cpu(i, span) { | |
5683 | struct cpumask *sg_span; | |
5684 | ||
5685 | if (cpumask_test_cpu(i, covered)) | |
5686 | continue; | |
5687 | ||
c1174876 PZ |
5688 | child = *per_cpu_ptr(sdd->sd, i); |
5689 | ||
5690 | /* See the comment near build_group_mask(). */ | |
5691 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5692 | continue; | |
5693 | ||
e3589f6c | 5694 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5695 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5696 | |
5697 | if (!sg) | |
5698 | goto fail; | |
5699 | ||
5700 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5701 | if (child->child) { |
5702 | child = child->child; | |
5703 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5704 | } else | |
5705 | cpumask_set_cpu(i, sg_span); | |
5706 | ||
5707 | cpumask_or(covered, covered, sg_span); | |
5708 | ||
63b2ca30 NP |
5709 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5710 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5711 | build_group_mask(sd, sg); |
5712 | ||
c3decf0d | 5713 | /* |
63b2ca30 | 5714 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5715 | * domains and no possible iteration will get us here, we won't |
5716 | * die on a /0 trap. | |
5717 | */ | |
63b2ca30 NP |
5718 | sg->sgc->capacity = SCHED_POWER_SCALE * cpumask_weight(sg_span); |
5719 | sg->sgc->capacity_orig = sg->sgc->capacity; | |
e3589f6c | 5720 | |
c1174876 PZ |
5721 | /* |
5722 | * Make sure the first group of this domain contains the | |
5723 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5724 | * breaks. See update_sg_lb_stats(). | |
5725 | */ | |
74a5ce20 | 5726 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5727 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5728 | groups = sg; |
5729 | ||
5730 | if (!first) | |
5731 | first = sg; | |
5732 | if (last) | |
5733 | last->next = sg; | |
5734 | last = sg; | |
5735 | last->next = first; | |
5736 | } | |
5737 | sd->groups = groups; | |
5738 | ||
5739 | return 0; | |
5740 | ||
5741 | fail: | |
5742 | free_sched_groups(first, 0); | |
5743 | ||
5744 | return -ENOMEM; | |
5745 | } | |
5746 | ||
dce840a0 | 5747 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5748 | { |
dce840a0 PZ |
5749 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5750 | struct sched_domain *child = sd->child; | |
1da177e4 | 5751 | |
dce840a0 PZ |
5752 | if (child) |
5753 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5754 | |
9c3f75cb | 5755 | if (sg) { |
dce840a0 | 5756 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5757 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5758 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5759 | } |
dce840a0 PZ |
5760 | |
5761 | return cpu; | |
1e9f28fa | 5762 | } |
1e9f28fa | 5763 | |
01a08546 | 5764 | /* |
dce840a0 PZ |
5765 | * build_sched_groups will build a circular linked list of the groups |
5766 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 5767 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
5768 | * |
5769 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5770 | */ |
e3589f6c PZ |
5771 | static int |
5772 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5773 | { |
dce840a0 PZ |
5774 | struct sched_group *first = NULL, *last = NULL; |
5775 | struct sd_data *sdd = sd->private; | |
5776 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5777 | struct cpumask *covered; |
dce840a0 | 5778 | int i; |
9c1cfda2 | 5779 | |
e3589f6c PZ |
5780 | get_group(cpu, sdd, &sd->groups); |
5781 | atomic_inc(&sd->groups->ref); | |
5782 | ||
0936629f | 5783 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5784 | return 0; |
5785 | ||
f96225fd PZ |
5786 | lockdep_assert_held(&sched_domains_mutex); |
5787 | covered = sched_domains_tmpmask; | |
5788 | ||
dce840a0 | 5789 | cpumask_clear(covered); |
6711cab4 | 5790 | |
dce840a0 PZ |
5791 | for_each_cpu(i, span) { |
5792 | struct sched_group *sg; | |
cd08e923 | 5793 | int group, j; |
6711cab4 | 5794 | |
dce840a0 PZ |
5795 | if (cpumask_test_cpu(i, covered)) |
5796 | continue; | |
6711cab4 | 5797 | |
cd08e923 | 5798 | group = get_group(i, sdd, &sg); |
c1174876 | 5799 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5800 | |
dce840a0 PZ |
5801 | for_each_cpu(j, span) { |
5802 | if (get_group(j, sdd, NULL) != group) | |
5803 | continue; | |
0601a88d | 5804 | |
dce840a0 PZ |
5805 | cpumask_set_cpu(j, covered); |
5806 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5807 | } | |
0601a88d | 5808 | |
dce840a0 PZ |
5809 | if (!first) |
5810 | first = sg; | |
5811 | if (last) | |
5812 | last->next = sg; | |
5813 | last = sg; | |
5814 | } | |
5815 | last->next = first; | |
e3589f6c PZ |
5816 | |
5817 | return 0; | |
0601a88d | 5818 | } |
51888ca2 | 5819 | |
89c4710e | 5820 | /* |
63b2ca30 | 5821 | * Initialize sched groups cpu_capacity. |
89c4710e | 5822 | * |
63b2ca30 | 5823 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 5824 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
5825 | * Typically cpu_capacity for all the groups in a sched domain will be same |
5826 | * unless there are asymmetries in the topology. If there are asymmetries, | |
5827 | * group having more cpu_capacity will pickup more load compared to the | |
5828 | * group having less cpu_capacity. | |
89c4710e | 5829 | */ |
63b2ca30 | 5830 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 5831 | { |
e3589f6c | 5832 | struct sched_group *sg = sd->groups; |
89c4710e | 5833 | |
94c95ba6 | 5834 | WARN_ON(!sg); |
e3589f6c PZ |
5835 | |
5836 | do { | |
5837 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5838 | sg = sg->next; | |
5839 | } while (sg != sd->groups); | |
89c4710e | 5840 | |
c1174876 | 5841 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5842 | return; |
aae6d3dd | 5843 | |
63b2ca30 NP |
5844 | update_group_capacity(sd, cpu); |
5845 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
5846 | } |
5847 | ||
7c16ec58 MT |
5848 | /* |
5849 | * Initializers for schedule domains | |
5850 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5851 | */ | |
5852 | ||
1d3504fc | 5853 | static int default_relax_domain_level = -1; |
60495e77 | 5854 | int sched_domain_level_max; |
1d3504fc HS |
5855 | |
5856 | static int __init setup_relax_domain_level(char *str) | |
5857 | { | |
a841f8ce DS |
5858 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5859 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5860 | |
1d3504fc HS |
5861 | return 1; |
5862 | } | |
5863 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5864 | ||
5865 | static void set_domain_attribute(struct sched_domain *sd, | |
5866 | struct sched_domain_attr *attr) | |
5867 | { | |
5868 | int request; | |
5869 | ||
5870 | if (!attr || attr->relax_domain_level < 0) { | |
5871 | if (default_relax_domain_level < 0) | |
5872 | return; | |
5873 | else | |
5874 | request = default_relax_domain_level; | |
5875 | } else | |
5876 | request = attr->relax_domain_level; | |
5877 | if (request < sd->level) { | |
5878 | /* turn off idle balance on this domain */ | |
c88d5910 | 5879 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5880 | } else { |
5881 | /* turn on idle balance on this domain */ | |
c88d5910 | 5882 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5883 | } |
5884 | } | |
5885 | ||
54ab4ff4 PZ |
5886 | static void __sdt_free(const struct cpumask *cpu_map); |
5887 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5888 | ||
2109b99e AH |
5889 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5890 | const struct cpumask *cpu_map) | |
5891 | { | |
5892 | switch (what) { | |
2109b99e | 5893 | case sa_rootdomain: |
822ff793 PZ |
5894 | if (!atomic_read(&d->rd->refcount)) |
5895 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5896 | case sa_sd: |
5897 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5898 | case sa_sd_storage: |
54ab4ff4 | 5899 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5900 | case sa_none: |
5901 | break; | |
5902 | } | |
5903 | } | |
3404c8d9 | 5904 | |
2109b99e AH |
5905 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5906 | const struct cpumask *cpu_map) | |
5907 | { | |
dce840a0 PZ |
5908 | memset(d, 0, sizeof(*d)); |
5909 | ||
54ab4ff4 PZ |
5910 | if (__sdt_alloc(cpu_map)) |
5911 | return sa_sd_storage; | |
dce840a0 PZ |
5912 | d->sd = alloc_percpu(struct sched_domain *); |
5913 | if (!d->sd) | |
5914 | return sa_sd_storage; | |
2109b99e | 5915 | d->rd = alloc_rootdomain(); |
dce840a0 | 5916 | if (!d->rd) |
21d42ccf | 5917 | return sa_sd; |
2109b99e AH |
5918 | return sa_rootdomain; |
5919 | } | |
57d885fe | 5920 | |
dce840a0 PZ |
5921 | /* |
5922 | * NULL the sd_data elements we've used to build the sched_domain and | |
5923 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5924 | * will not free the data we're using. | |
5925 | */ | |
5926 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5927 | { | |
5928 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5929 | |
5930 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5931 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5932 | ||
e3589f6c | 5933 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5934 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 5935 | |
63b2ca30 NP |
5936 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
5937 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
5938 | } |
5939 | ||
cb83b629 | 5940 | #ifdef CONFIG_NUMA |
cb83b629 | 5941 | static int sched_domains_numa_levels; |
cb83b629 PZ |
5942 | static int *sched_domains_numa_distance; |
5943 | static struct cpumask ***sched_domains_numa_masks; | |
5944 | static int sched_domains_curr_level; | |
143e1e28 | 5945 | #endif |
cb83b629 | 5946 | |
143e1e28 VG |
5947 | /* |
5948 | * SD_flags allowed in topology descriptions. | |
5949 | * | |
5950 | * SD_SHARE_CPUPOWER - describes SMT topologies | |
5951 | * SD_SHARE_PKG_RESOURCES - describes shared caches | |
5952 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 5953 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
5954 | * |
5955 | * Odd one out: | |
5956 | * SD_ASYM_PACKING - describes SMT quirks | |
5957 | */ | |
5958 | #define TOPOLOGY_SD_FLAGS \ | |
5959 | (SD_SHARE_CPUPOWER | \ | |
5960 | SD_SHARE_PKG_RESOURCES | \ | |
5961 | SD_NUMA | \ | |
d77b3ed5 VG |
5962 | SD_ASYM_PACKING | \ |
5963 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
5964 | |
5965 | static struct sched_domain * | |
143e1e28 | 5966 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
5967 | { |
5968 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
5969 | int sd_weight, sd_flags = 0; |
5970 | ||
5971 | #ifdef CONFIG_NUMA | |
5972 | /* | |
5973 | * Ugly hack to pass state to sd_numa_mask()... | |
5974 | */ | |
5975 | sched_domains_curr_level = tl->numa_level; | |
5976 | #endif | |
5977 | ||
5978 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
5979 | ||
5980 | if (tl->sd_flags) | |
5981 | sd_flags = (*tl->sd_flags)(); | |
5982 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
5983 | "wrong sd_flags in topology description\n")) | |
5984 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
5985 | |
5986 | *sd = (struct sched_domain){ | |
5987 | .min_interval = sd_weight, | |
5988 | .max_interval = 2*sd_weight, | |
5989 | .busy_factor = 32, | |
870a0bb5 | 5990 | .imbalance_pct = 125, |
143e1e28 VG |
5991 | |
5992 | .cache_nice_tries = 0, | |
5993 | .busy_idx = 0, | |
5994 | .idle_idx = 0, | |
cb83b629 PZ |
5995 | .newidle_idx = 0, |
5996 | .wake_idx = 0, | |
5997 | .forkexec_idx = 0, | |
5998 | ||
5999 | .flags = 1*SD_LOAD_BALANCE | |
6000 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
6001 | | 1*SD_BALANCE_EXEC |
6002 | | 1*SD_BALANCE_FORK | |
cb83b629 | 6003 | | 0*SD_BALANCE_WAKE |
143e1e28 | 6004 | | 1*SD_WAKE_AFFINE |
cb83b629 | 6005 | | 0*SD_SHARE_CPUPOWER |
cb83b629 | 6006 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 6007 | | 0*SD_SERIALIZE |
cb83b629 | 6008 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
6009 | | 0*SD_NUMA |
6010 | | sd_flags | |
cb83b629 | 6011 | , |
143e1e28 | 6012 | |
cb83b629 PZ |
6013 | .last_balance = jiffies, |
6014 | .balance_interval = sd_weight, | |
143e1e28 | 6015 | .smt_gain = 0, |
2b4cfe64 JL |
6016 | .max_newidle_lb_cost = 0, |
6017 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6018 | #ifdef CONFIG_SCHED_DEBUG |
6019 | .name = tl->name, | |
6020 | #endif | |
cb83b629 | 6021 | }; |
cb83b629 PZ |
6022 | |
6023 | /* | |
143e1e28 | 6024 | * Convert topological properties into behaviour. |
cb83b629 | 6025 | */ |
143e1e28 VG |
6026 | |
6027 | if (sd->flags & SD_SHARE_CPUPOWER) { | |
6028 | sd->imbalance_pct = 110; | |
6029 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6030 | |
6031 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6032 | sd->imbalance_pct = 117; | |
6033 | sd->cache_nice_tries = 1; | |
6034 | sd->busy_idx = 2; | |
6035 | ||
6036 | #ifdef CONFIG_NUMA | |
6037 | } else if (sd->flags & SD_NUMA) { | |
6038 | sd->cache_nice_tries = 2; | |
6039 | sd->busy_idx = 3; | |
6040 | sd->idle_idx = 2; | |
6041 | ||
6042 | sd->flags |= SD_SERIALIZE; | |
6043 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6044 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6045 | SD_BALANCE_FORK | | |
6046 | SD_WAKE_AFFINE); | |
6047 | } | |
6048 | ||
6049 | #endif | |
6050 | } else { | |
6051 | sd->flags |= SD_PREFER_SIBLING; | |
6052 | sd->cache_nice_tries = 1; | |
6053 | sd->busy_idx = 2; | |
6054 | sd->idle_idx = 1; | |
6055 | } | |
6056 | ||
6057 | sd->private = &tl->data; | |
cb83b629 PZ |
6058 | |
6059 | return sd; | |
6060 | } | |
6061 | ||
143e1e28 VG |
6062 | /* |
6063 | * Topology list, bottom-up. | |
6064 | */ | |
6065 | static struct sched_domain_topology_level default_topology[] = { | |
6066 | #ifdef CONFIG_SCHED_SMT | |
6067 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6068 | #endif | |
6069 | #ifdef CONFIG_SCHED_MC | |
6070 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6071 | #endif |
6072 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6073 | { NULL, }, | |
6074 | }; | |
6075 | ||
6076 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6077 | ||
6078 | #define for_each_sd_topology(tl) \ | |
6079 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6080 | ||
6081 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6082 | { | |
6083 | sched_domain_topology = tl; | |
6084 | } | |
6085 | ||
6086 | #ifdef CONFIG_NUMA | |
6087 | ||
cb83b629 PZ |
6088 | static const struct cpumask *sd_numa_mask(int cpu) |
6089 | { | |
6090 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6091 | } | |
6092 | ||
d039ac60 PZ |
6093 | static void sched_numa_warn(const char *str) |
6094 | { | |
6095 | static int done = false; | |
6096 | int i,j; | |
6097 | ||
6098 | if (done) | |
6099 | return; | |
6100 | ||
6101 | done = true; | |
6102 | ||
6103 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6104 | ||
6105 | for (i = 0; i < nr_node_ids; i++) { | |
6106 | printk(KERN_WARNING " "); | |
6107 | for (j = 0; j < nr_node_ids; j++) | |
6108 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6109 | printk(KERN_CONT "\n"); | |
6110 | } | |
6111 | printk(KERN_WARNING "\n"); | |
6112 | } | |
6113 | ||
6114 | static bool find_numa_distance(int distance) | |
6115 | { | |
6116 | int i; | |
6117 | ||
6118 | if (distance == node_distance(0, 0)) | |
6119 | return true; | |
6120 | ||
6121 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6122 | if (sched_domains_numa_distance[i] == distance) | |
6123 | return true; | |
6124 | } | |
6125 | ||
6126 | return false; | |
6127 | } | |
6128 | ||
cb83b629 PZ |
6129 | static void sched_init_numa(void) |
6130 | { | |
6131 | int next_distance, curr_distance = node_distance(0, 0); | |
6132 | struct sched_domain_topology_level *tl; | |
6133 | int level = 0; | |
6134 | int i, j, k; | |
6135 | ||
cb83b629 PZ |
6136 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6137 | if (!sched_domains_numa_distance) | |
6138 | return; | |
6139 | ||
6140 | /* | |
6141 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6142 | * unique distances in the node_distance() table. | |
6143 | * | |
6144 | * Assumes node_distance(0,j) includes all distances in | |
6145 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6146 | */ |
6147 | next_distance = curr_distance; | |
6148 | for (i = 0; i < nr_node_ids; i++) { | |
6149 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6150 | for (k = 0; k < nr_node_ids; k++) { |
6151 | int distance = node_distance(i, k); | |
6152 | ||
6153 | if (distance > curr_distance && | |
6154 | (distance < next_distance || | |
6155 | next_distance == curr_distance)) | |
6156 | next_distance = distance; | |
6157 | ||
6158 | /* | |
6159 | * While not a strong assumption it would be nice to know | |
6160 | * about cases where if node A is connected to B, B is not | |
6161 | * equally connected to A. | |
6162 | */ | |
6163 | if (sched_debug() && node_distance(k, i) != distance) | |
6164 | sched_numa_warn("Node-distance not symmetric"); | |
6165 | ||
6166 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6167 | sched_numa_warn("Node-0 not representative"); | |
6168 | } | |
6169 | if (next_distance != curr_distance) { | |
6170 | sched_domains_numa_distance[level++] = next_distance; | |
6171 | sched_domains_numa_levels = level; | |
6172 | curr_distance = next_distance; | |
6173 | } else break; | |
cb83b629 | 6174 | } |
d039ac60 PZ |
6175 | |
6176 | /* | |
6177 | * In case of sched_debug() we verify the above assumption. | |
6178 | */ | |
6179 | if (!sched_debug()) | |
6180 | break; | |
cb83b629 PZ |
6181 | } |
6182 | /* | |
6183 | * 'level' contains the number of unique distances, excluding the | |
6184 | * identity distance node_distance(i,i). | |
6185 | * | |
28b4a521 | 6186 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6187 | * numbers. |
6188 | */ | |
6189 | ||
5f7865f3 TC |
6190 | /* |
6191 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6192 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6193 | * the array will contain less then 'level' members. This could be | |
6194 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6195 | * in other functions. | |
6196 | * | |
6197 | * We reset it to 'level' at the end of this function. | |
6198 | */ | |
6199 | sched_domains_numa_levels = 0; | |
6200 | ||
cb83b629 PZ |
6201 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6202 | if (!sched_domains_numa_masks) | |
6203 | return; | |
6204 | ||
6205 | /* | |
6206 | * Now for each level, construct a mask per node which contains all | |
6207 | * cpus of nodes that are that many hops away from us. | |
6208 | */ | |
6209 | for (i = 0; i < level; i++) { | |
6210 | sched_domains_numa_masks[i] = | |
6211 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6212 | if (!sched_domains_numa_masks[i]) | |
6213 | return; | |
6214 | ||
6215 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6216 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6217 | if (!mask) |
6218 | return; | |
6219 | ||
6220 | sched_domains_numa_masks[i][j] = mask; | |
6221 | ||
6222 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6223 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6224 | continue; |
6225 | ||
6226 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6227 | } | |
6228 | } | |
6229 | } | |
6230 | ||
143e1e28 VG |
6231 | /* Compute default topology size */ |
6232 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6233 | ||
c515db8c | 6234 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6235 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6236 | if (!tl) | |
6237 | return; | |
6238 | ||
6239 | /* | |
6240 | * Copy the default topology bits.. | |
6241 | */ | |
143e1e28 VG |
6242 | for (i = 0; sched_domain_topology[i].mask; i++) |
6243 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6244 | |
6245 | /* | |
6246 | * .. and append 'j' levels of NUMA goodness. | |
6247 | */ | |
6248 | for (j = 0; j < level; i++, j++) { | |
6249 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6250 | .mask = sd_numa_mask, |
143e1e28 | 6251 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6252 | .flags = SDTL_OVERLAP, |
6253 | .numa_level = j, | |
143e1e28 | 6254 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6255 | }; |
6256 | } | |
6257 | ||
6258 | sched_domain_topology = tl; | |
5f7865f3 TC |
6259 | |
6260 | sched_domains_numa_levels = level; | |
cb83b629 | 6261 | } |
301a5cba TC |
6262 | |
6263 | static void sched_domains_numa_masks_set(int cpu) | |
6264 | { | |
6265 | int i, j; | |
6266 | int node = cpu_to_node(cpu); | |
6267 | ||
6268 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6269 | for (j = 0; j < nr_node_ids; j++) { | |
6270 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6271 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6272 | } | |
6273 | } | |
6274 | } | |
6275 | ||
6276 | static void sched_domains_numa_masks_clear(int cpu) | |
6277 | { | |
6278 | int i, j; | |
6279 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6280 | for (j = 0; j < nr_node_ids; j++) | |
6281 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6282 | } | |
6283 | } | |
6284 | ||
6285 | /* | |
6286 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6287 | * are onlined. | |
6288 | */ | |
6289 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6290 | unsigned long action, | |
6291 | void *hcpu) | |
6292 | { | |
6293 | int cpu = (long)hcpu; | |
6294 | ||
6295 | switch (action & ~CPU_TASKS_FROZEN) { | |
6296 | case CPU_ONLINE: | |
6297 | sched_domains_numa_masks_set(cpu); | |
6298 | break; | |
6299 | ||
6300 | case CPU_DEAD: | |
6301 | sched_domains_numa_masks_clear(cpu); | |
6302 | break; | |
6303 | ||
6304 | default: | |
6305 | return NOTIFY_DONE; | |
6306 | } | |
6307 | ||
6308 | return NOTIFY_OK; | |
cb83b629 PZ |
6309 | } |
6310 | #else | |
6311 | static inline void sched_init_numa(void) | |
6312 | { | |
6313 | } | |
301a5cba TC |
6314 | |
6315 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6316 | unsigned long action, | |
6317 | void *hcpu) | |
6318 | { | |
6319 | return 0; | |
6320 | } | |
cb83b629 PZ |
6321 | #endif /* CONFIG_NUMA */ |
6322 | ||
54ab4ff4 PZ |
6323 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6324 | { | |
6325 | struct sched_domain_topology_level *tl; | |
6326 | int j; | |
6327 | ||
27723a68 | 6328 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6329 | struct sd_data *sdd = &tl->data; |
6330 | ||
6331 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6332 | if (!sdd->sd) | |
6333 | return -ENOMEM; | |
6334 | ||
6335 | sdd->sg = alloc_percpu(struct sched_group *); | |
6336 | if (!sdd->sg) | |
6337 | return -ENOMEM; | |
6338 | ||
63b2ca30 NP |
6339 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6340 | if (!sdd->sgc) | |
9c3f75cb PZ |
6341 | return -ENOMEM; |
6342 | ||
54ab4ff4 PZ |
6343 | for_each_cpu(j, cpu_map) { |
6344 | struct sched_domain *sd; | |
6345 | struct sched_group *sg; | |
63b2ca30 | 6346 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6347 | |
6348 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6349 | GFP_KERNEL, cpu_to_node(j)); | |
6350 | if (!sd) | |
6351 | return -ENOMEM; | |
6352 | ||
6353 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6354 | ||
6355 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6356 | GFP_KERNEL, cpu_to_node(j)); | |
6357 | if (!sg) | |
6358 | return -ENOMEM; | |
6359 | ||
30b4e9eb IM |
6360 | sg->next = sg; |
6361 | ||
54ab4ff4 | 6362 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6363 | |
63b2ca30 | 6364 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6365 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6366 | if (!sgc) |
9c3f75cb PZ |
6367 | return -ENOMEM; |
6368 | ||
63b2ca30 | 6369 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6370 | } |
6371 | } | |
6372 | ||
6373 | return 0; | |
6374 | } | |
6375 | ||
6376 | static void __sdt_free(const struct cpumask *cpu_map) | |
6377 | { | |
6378 | struct sched_domain_topology_level *tl; | |
6379 | int j; | |
6380 | ||
27723a68 | 6381 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6382 | struct sd_data *sdd = &tl->data; |
6383 | ||
6384 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6385 | struct sched_domain *sd; |
6386 | ||
6387 | if (sdd->sd) { | |
6388 | sd = *per_cpu_ptr(sdd->sd, j); | |
6389 | if (sd && (sd->flags & SD_OVERLAP)) | |
6390 | free_sched_groups(sd->groups, 0); | |
6391 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6392 | } | |
6393 | ||
6394 | if (sdd->sg) | |
6395 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6396 | if (sdd->sgc) |
6397 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6398 | } |
6399 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6400 | sdd->sd = NULL; |
54ab4ff4 | 6401 | free_percpu(sdd->sg); |
fb2cf2c6 | 6402 | sdd->sg = NULL; |
63b2ca30 NP |
6403 | free_percpu(sdd->sgc); |
6404 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6405 | } |
6406 | } | |
6407 | ||
2c402dc3 | 6408 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6409 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6410 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6411 | { |
143e1e28 | 6412 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6413 | if (!sd) |
d069b916 | 6414 | return child; |
2c402dc3 | 6415 | |
2c402dc3 | 6416 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6417 | if (child) { |
6418 | sd->level = child->level + 1; | |
6419 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6420 | child->parent = sd; |
c75e0128 | 6421 | sd->child = child; |
60495e77 | 6422 | } |
a841f8ce | 6423 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6424 | |
6425 | return sd; | |
6426 | } | |
6427 | ||
2109b99e AH |
6428 | /* |
6429 | * Build sched domains for a given set of cpus and attach the sched domains | |
6430 | * to the individual cpus | |
6431 | */ | |
dce840a0 PZ |
6432 | static int build_sched_domains(const struct cpumask *cpu_map, |
6433 | struct sched_domain_attr *attr) | |
2109b99e | 6434 | { |
1c632169 | 6435 | enum s_alloc alloc_state; |
dce840a0 | 6436 | struct sched_domain *sd; |
2109b99e | 6437 | struct s_data d; |
822ff793 | 6438 | int i, ret = -ENOMEM; |
9c1cfda2 | 6439 | |
2109b99e AH |
6440 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6441 | if (alloc_state != sa_rootdomain) | |
6442 | goto error; | |
9c1cfda2 | 6443 | |
dce840a0 | 6444 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6445 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6446 | struct sched_domain_topology_level *tl; |
6447 | ||
3bd65a80 | 6448 | sd = NULL; |
27723a68 | 6449 | for_each_sd_topology(tl) { |
4a850cbe | 6450 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6451 | if (tl == sched_domain_topology) |
6452 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6453 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6454 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6455 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6456 | break; | |
e3589f6c | 6457 | } |
dce840a0 PZ |
6458 | } |
6459 | ||
6460 | /* Build the groups for the domains */ | |
6461 | for_each_cpu(i, cpu_map) { | |
6462 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6463 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6464 | if (sd->flags & SD_OVERLAP) { |
6465 | if (build_overlap_sched_groups(sd, i)) | |
6466 | goto error; | |
6467 | } else { | |
6468 | if (build_sched_groups(sd, i)) | |
6469 | goto error; | |
6470 | } | |
1cf51902 | 6471 | } |
a06dadbe | 6472 | } |
9c1cfda2 | 6473 | |
ced549fa | 6474 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6475 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6476 | if (!cpumask_test_cpu(i, cpu_map)) | |
6477 | continue; | |
9c1cfda2 | 6478 | |
dce840a0 PZ |
6479 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6480 | claim_allocations(i, sd); | |
63b2ca30 | 6481 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6482 | } |
f712c0c7 | 6483 | } |
9c1cfda2 | 6484 | |
1da177e4 | 6485 | /* Attach the domains */ |
dce840a0 | 6486 | rcu_read_lock(); |
abcd083a | 6487 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6488 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6489 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6490 | } |
dce840a0 | 6491 | rcu_read_unlock(); |
51888ca2 | 6492 | |
822ff793 | 6493 | ret = 0; |
51888ca2 | 6494 | error: |
2109b99e | 6495 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6496 | return ret; |
1da177e4 | 6497 | } |
029190c5 | 6498 | |
acc3f5d7 | 6499 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6500 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6501 | static struct sched_domain_attr *dattr_cur; |
6502 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6503 | |
6504 | /* | |
6505 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6506 | * cpumask) fails, then fallback to a single sched domain, |
6507 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6508 | */ |
4212823f | 6509 | static cpumask_var_t fallback_doms; |
029190c5 | 6510 | |
ee79d1bd HC |
6511 | /* |
6512 | * arch_update_cpu_topology lets virtualized architectures update the | |
6513 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6514 | * or 0 if it stayed the same. | |
6515 | */ | |
52f5684c | 6516 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6517 | { |
ee79d1bd | 6518 | return 0; |
22e52b07 HC |
6519 | } |
6520 | ||
acc3f5d7 RR |
6521 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6522 | { | |
6523 | int i; | |
6524 | cpumask_var_t *doms; | |
6525 | ||
6526 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6527 | if (!doms) | |
6528 | return NULL; | |
6529 | for (i = 0; i < ndoms; i++) { | |
6530 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6531 | free_sched_domains(doms, i); | |
6532 | return NULL; | |
6533 | } | |
6534 | } | |
6535 | return doms; | |
6536 | } | |
6537 | ||
6538 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6539 | { | |
6540 | unsigned int i; | |
6541 | for (i = 0; i < ndoms; i++) | |
6542 | free_cpumask_var(doms[i]); | |
6543 | kfree(doms); | |
6544 | } | |
6545 | ||
1a20ff27 | 6546 | /* |
41a2d6cf | 6547 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6548 | * For now this just excludes isolated cpus, but could be used to |
6549 | * exclude other special cases in the future. | |
1a20ff27 | 6550 | */ |
c4a8849a | 6551 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6552 | { |
7378547f MM |
6553 | int err; |
6554 | ||
22e52b07 | 6555 | arch_update_cpu_topology(); |
029190c5 | 6556 | ndoms_cur = 1; |
acc3f5d7 | 6557 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6558 | if (!doms_cur) |
acc3f5d7 RR |
6559 | doms_cur = &fallback_doms; |
6560 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6561 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6562 | register_sched_domain_sysctl(); |
7378547f MM |
6563 | |
6564 | return err; | |
1a20ff27 DG |
6565 | } |
6566 | ||
1a20ff27 DG |
6567 | /* |
6568 | * Detach sched domains from a group of cpus specified in cpu_map | |
6569 | * These cpus will now be attached to the NULL domain | |
6570 | */ | |
96f874e2 | 6571 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6572 | { |
6573 | int i; | |
6574 | ||
dce840a0 | 6575 | rcu_read_lock(); |
abcd083a | 6576 | for_each_cpu(i, cpu_map) |
57d885fe | 6577 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6578 | rcu_read_unlock(); |
1a20ff27 DG |
6579 | } |
6580 | ||
1d3504fc HS |
6581 | /* handle null as "default" */ |
6582 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6583 | struct sched_domain_attr *new, int idx_new) | |
6584 | { | |
6585 | struct sched_domain_attr tmp; | |
6586 | ||
6587 | /* fast path */ | |
6588 | if (!new && !cur) | |
6589 | return 1; | |
6590 | ||
6591 | tmp = SD_ATTR_INIT; | |
6592 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6593 | new ? (new + idx_new) : &tmp, | |
6594 | sizeof(struct sched_domain_attr)); | |
6595 | } | |
6596 | ||
029190c5 PJ |
6597 | /* |
6598 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6599 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6600 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6601 | * It destroys each deleted domain and builds each new domain. | |
6602 | * | |
acc3f5d7 | 6603 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6604 | * The masks don't intersect (don't overlap.) We should setup one |
6605 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6606 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6607 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6608 | * it as it is. | |
6609 | * | |
acc3f5d7 RR |
6610 | * The passed in 'doms_new' should be allocated using |
6611 | * alloc_sched_domains. This routine takes ownership of it and will | |
6612 | * free_sched_domains it when done with it. If the caller failed the | |
6613 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6614 | * and partition_sched_domains() will fallback to the single partition | |
6615 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6616 | * |
96f874e2 | 6617 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6618 | * ndoms_new == 0 is a special case for destroying existing domains, |
6619 | * and it will not create the default domain. | |
dfb512ec | 6620 | * |
029190c5 PJ |
6621 | * Call with hotplug lock held |
6622 | */ | |
acc3f5d7 | 6623 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6624 | struct sched_domain_attr *dattr_new) |
029190c5 | 6625 | { |
dfb512ec | 6626 | int i, j, n; |
d65bd5ec | 6627 | int new_topology; |
029190c5 | 6628 | |
712555ee | 6629 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6630 | |
7378547f MM |
6631 | /* always unregister in case we don't destroy any domains */ |
6632 | unregister_sched_domain_sysctl(); | |
6633 | ||
d65bd5ec HC |
6634 | /* Let architecture update cpu core mappings. */ |
6635 | new_topology = arch_update_cpu_topology(); | |
6636 | ||
dfb512ec | 6637 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6638 | |
6639 | /* Destroy deleted domains */ | |
6640 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6641 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6642 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6643 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6644 | goto match1; |
6645 | } | |
6646 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6647 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6648 | match1: |
6649 | ; | |
6650 | } | |
6651 | ||
c8d2d47a | 6652 | n = ndoms_cur; |
e761b772 | 6653 | if (doms_new == NULL) { |
c8d2d47a | 6654 | n = 0; |
acc3f5d7 | 6655 | doms_new = &fallback_doms; |
6ad4c188 | 6656 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6657 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6658 | } |
6659 | ||
029190c5 PJ |
6660 | /* Build new domains */ |
6661 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6662 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6663 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6664 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6665 | goto match2; |
6666 | } | |
6667 | /* no match - add a new doms_new */ | |
dce840a0 | 6668 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6669 | match2: |
6670 | ; | |
6671 | } | |
6672 | ||
6673 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6674 | if (doms_cur != &fallback_doms) |
6675 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6676 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6677 | doms_cur = doms_new; |
1d3504fc | 6678 | dattr_cur = dattr_new; |
029190c5 | 6679 | ndoms_cur = ndoms_new; |
7378547f MM |
6680 | |
6681 | register_sched_domain_sysctl(); | |
a1835615 | 6682 | |
712555ee | 6683 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6684 | } |
6685 | ||
d35be8ba SB |
6686 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6687 | ||
1da177e4 | 6688 | /* |
3a101d05 TH |
6689 | * Update cpusets according to cpu_active mask. If cpusets are |
6690 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6691 | * around partition_sched_domains(). | |
d35be8ba SB |
6692 | * |
6693 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6694 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6695 | */ |
0b2e918a TH |
6696 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6697 | void *hcpu) | |
e761b772 | 6698 | { |
d35be8ba SB |
6699 | switch (action) { |
6700 | case CPU_ONLINE_FROZEN: | |
6701 | case CPU_DOWN_FAILED_FROZEN: | |
6702 | ||
6703 | /* | |
6704 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6705 | * resume sequence. As long as this is not the last online | |
6706 | * operation in the resume sequence, just build a single sched | |
6707 | * domain, ignoring cpusets. | |
6708 | */ | |
6709 | num_cpus_frozen--; | |
6710 | if (likely(num_cpus_frozen)) { | |
6711 | partition_sched_domains(1, NULL, NULL); | |
6712 | break; | |
6713 | } | |
6714 | ||
6715 | /* | |
6716 | * This is the last CPU online operation. So fall through and | |
6717 | * restore the original sched domains by considering the | |
6718 | * cpuset configurations. | |
6719 | */ | |
6720 | ||
e761b772 | 6721 | case CPU_ONLINE: |
6ad4c188 | 6722 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6723 | cpuset_update_active_cpus(true); |
d35be8ba | 6724 | break; |
3a101d05 TH |
6725 | default: |
6726 | return NOTIFY_DONE; | |
6727 | } | |
d35be8ba | 6728 | return NOTIFY_OK; |
3a101d05 | 6729 | } |
e761b772 | 6730 | |
0b2e918a TH |
6731 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6732 | void *hcpu) | |
3a101d05 | 6733 | { |
d35be8ba | 6734 | switch (action) { |
3a101d05 | 6735 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6736 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6737 | break; |
6738 | case CPU_DOWN_PREPARE_FROZEN: | |
6739 | num_cpus_frozen++; | |
6740 | partition_sched_domains(1, NULL, NULL); | |
6741 | break; | |
e761b772 MK |
6742 | default: |
6743 | return NOTIFY_DONE; | |
6744 | } | |
d35be8ba | 6745 | return NOTIFY_OK; |
e761b772 | 6746 | } |
e761b772 | 6747 | |
1da177e4 LT |
6748 | void __init sched_init_smp(void) |
6749 | { | |
dcc30a35 RR |
6750 | cpumask_var_t non_isolated_cpus; |
6751 | ||
6752 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6753 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6754 | |
cb83b629 PZ |
6755 | sched_init_numa(); |
6756 | ||
6acce3ef PZ |
6757 | /* |
6758 | * There's no userspace yet to cause hotplug operations; hence all the | |
6759 | * cpu masks are stable and all blatant races in the below code cannot | |
6760 | * happen. | |
6761 | */ | |
712555ee | 6762 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6763 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6764 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6765 | if (cpumask_empty(non_isolated_cpus)) | |
6766 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6767 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6768 | |
301a5cba | 6769 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6770 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6771 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6772 | |
b328ca18 | 6773 | init_hrtick(); |
5c1e1767 NP |
6774 | |
6775 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6776 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6777 | BUG(); |
19978ca6 | 6778 | sched_init_granularity(); |
dcc30a35 | 6779 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6780 | |
0e3900e6 | 6781 | init_sched_rt_class(); |
1baca4ce | 6782 | init_sched_dl_class(); |
1da177e4 LT |
6783 | } |
6784 | #else | |
6785 | void __init sched_init_smp(void) | |
6786 | { | |
19978ca6 | 6787 | sched_init_granularity(); |
1da177e4 LT |
6788 | } |
6789 | #endif /* CONFIG_SMP */ | |
6790 | ||
cd1bb94b AB |
6791 | const_debug unsigned int sysctl_timer_migration = 1; |
6792 | ||
1da177e4 LT |
6793 | int in_sched_functions(unsigned long addr) |
6794 | { | |
1da177e4 LT |
6795 | return in_lock_functions(addr) || |
6796 | (addr >= (unsigned long)__sched_text_start | |
6797 | && addr < (unsigned long)__sched_text_end); | |
6798 | } | |
6799 | ||
029632fb | 6800 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6801 | /* |
6802 | * Default task group. | |
6803 | * Every task in system belongs to this group at bootup. | |
6804 | */ | |
029632fb | 6805 | struct task_group root_task_group; |
35cf4e50 | 6806 | LIST_HEAD(task_groups); |
052f1dc7 | 6807 | #endif |
6f505b16 | 6808 | |
e6252c3e | 6809 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6810 | |
1da177e4 LT |
6811 | void __init sched_init(void) |
6812 | { | |
dd41f596 | 6813 | int i, j; |
434d53b0 MT |
6814 | unsigned long alloc_size = 0, ptr; |
6815 | ||
6816 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6817 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6818 | #endif | |
6819 | #ifdef CONFIG_RT_GROUP_SCHED | |
6820 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6821 | #endif |
df7c8e84 | 6822 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6823 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6824 | #endif |
434d53b0 | 6825 | if (alloc_size) { |
36b7b6d4 | 6826 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6827 | |
6828 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6829 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6830 | ptr += nr_cpu_ids * sizeof(void **); |
6831 | ||
07e06b01 | 6832 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6833 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6834 | |
6d6bc0ad | 6835 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6836 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6837 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6838 | ptr += nr_cpu_ids * sizeof(void **); |
6839 | ||
07e06b01 | 6840 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6841 | ptr += nr_cpu_ids * sizeof(void **); |
6842 | ||
6d6bc0ad | 6843 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6844 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6845 | for_each_possible_cpu(i) { | |
e6252c3e | 6846 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6847 | ptr += cpumask_size(); |
6848 | } | |
6849 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6850 | } |
dd41f596 | 6851 | |
332ac17e DF |
6852 | init_rt_bandwidth(&def_rt_bandwidth, |
6853 | global_rt_period(), global_rt_runtime()); | |
6854 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6855 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6856 | |
57d885fe GH |
6857 | #ifdef CONFIG_SMP |
6858 | init_defrootdomain(); | |
6859 | #endif | |
6860 | ||
d0b27fa7 | 6861 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6862 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6863 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6864 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6865 | |
7c941438 | 6866 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6867 | list_add(&root_task_group.list, &task_groups); |
6868 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6869 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6870 | autogroup_init(&init_task); |
54c707e9 | 6871 | |
7c941438 | 6872 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6873 | |
0a945022 | 6874 | for_each_possible_cpu(i) { |
70b97a7f | 6875 | struct rq *rq; |
1da177e4 LT |
6876 | |
6877 | rq = cpu_rq(i); | |
05fa785c | 6878 | raw_spin_lock_init(&rq->lock); |
7897986b | 6879 | rq->nr_running = 0; |
dce48a84 TG |
6880 | rq->calc_load_active = 0; |
6881 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6882 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6883 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6884 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6885 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6886 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6887 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6888 | /* |
07e06b01 | 6889 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6890 | * |
6891 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6892 | * gets 100% of the cpu resources in the system. This overall | |
6893 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6894 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6895 | * based on each entity's (task or task-group's) weight |
6896 | * (se->load.weight). | |
6897 | * | |
07e06b01 | 6898 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6899 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6900 | * then A0's share of the cpu resource is: | |
6901 | * | |
0d905bca | 6902 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6903 | * |
07e06b01 YZ |
6904 | * We achieve this by letting root_task_group's tasks sit |
6905 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6906 | */ |
ab84d31e | 6907 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6908 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6909 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6910 | ||
6911 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6912 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6913 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6914 | #endif |
1da177e4 | 6915 | |
dd41f596 IM |
6916 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6917 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6918 | |
6919 | rq->last_load_update_tick = jiffies; | |
6920 | ||
1da177e4 | 6921 | #ifdef CONFIG_SMP |
41c7ce9a | 6922 | rq->sd = NULL; |
57d885fe | 6923 | rq->rd = NULL; |
ced549fa | 6924 | rq->cpu_capacity = SCHED_POWER_SCALE; |
3f029d3c | 6925 | rq->post_schedule = 0; |
1da177e4 | 6926 | rq->active_balance = 0; |
dd41f596 | 6927 | rq->next_balance = jiffies; |
1da177e4 | 6928 | rq->push_cpu = 0; |
0a2966b4 | 6929 | rq->cpu = i; |
1f11eb6a | 6930 | rq->online = 0; |
eae0c9df MG |
6931 | rq->idle_stamp = 0; |
6932 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6933 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6934 | |
6935 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6936 | ||
dc938520 | 6937 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6938 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6939 | rq->nohz_flags = 0; |
83cd4fe2 | 6940 | #endif |
265f22a9 FW |
6941 | #ifdef CONFIG_NO_HZ_FULL |
6942 | rq->last_sched_tick = 0; | |
6943 | #endif | |
1da177e4 | 6944 | #endif |
8f4d37ec | 6945 | init_rq_hrtick(rq); |
1da177e4 | 6946 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6947 | } |
6948 | ||
2dd73a4f | 6949 | set_load_weight(&init_task); |
b50f60ce | 6950 | |
e107be36 AK |
6951 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6952 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6953 | #endif | |
6954 | ||
1da177e4 LT |
6955 | /* |
6956 | * The boot idle thread does lazy MMU switching as well: | |
6957 | */ | |
6958 | atomic_inc(&init_mm.mm_count); | |
6959 | enter_lazy_tlb(&init_mm, current); | |
6960 | ||
6961 | /* | |
6962 | * Make us the idle thread. Technically, schedule() should not be | |
6963 | * called from this thread, however somewhere below it might be, | |
6964 | * but because we are the idle thread, we just pick up running again | |
6965 | * when this runqueue becomes "idle". | |
6966 | */ | |
6967 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6968 | |
6969 | calc_load_update = jiffies + LOAD_FREQ; | |
6970 | ||
dd41f596 IM |
6971 | /* |
6972 | * During early bootup we pretend to be a normal task: | |
6973 | */ | |
6974 | current->sched_class = &fair_sched_class; | |
6892b75e | 6975 | |
bf4d83f6 | 6976 | #ifdef CONFIG_SMP |
4cb98839 | 6977 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6978 | /* May be allocated at isolcpus cmdline parse time */ |
6979 | if (cpu_isolated_map == NULL) | |
6980 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6981 | idle_thread_set_boot_cpu(); |
a803f026 | 6982 | set_cpu_rq_start_time(); |
029632fb PZ |
6983 | #endif |
6984 | init_sched_fair_class(); | |
6a7b3dc3 | 6985 | |
6892b75e | 6986 | scheduler_running = 1; |
1da177e4 LT |
6987 | } |
6988 | ||
d902db1e | 6989 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6990 | static inline int preempt_count_equals(int preempt_offset) |
6991 | { | |
234da7bc | 6992 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6993 | |
4ba8216c | 6994 | return (nested == preempt_offset); |
e4aafea2 FW |
6995 | } |
6996 | ||
d894837f | 6997 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6998 | { |
1da177e4 LT |
6999 | static unsigned long prev_jiffy; /* ratelimiting */ |
7000 | ||
b3fbab05 | 7001 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
7002 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
7003 | !is_idle_task(current)) || | |
e4aafea2 | 7004 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
7005 | return; |
7006 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7007 | return; | |
7008 | prev_jiffy = jiffies; | |
7009 | ||
3df0fc5b PZ |
7010 | printk(KERN_ERR |
7011 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7012 | file, line); | |
7013 | printk(KERN_ERR | |
7014 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7015 | in_atomic(), irqs_disabled(), | |
7016 | current->pid, current->comm); | |
aef745fc IM |
7017 | |
7018 | debug_show_held_locks(current); | |
7019 | if (irqs_disabled()) | |
7020 | print_irqtrace_events(current); | |
8f47b187 TG |
7021 | #ifdef CONFIG_DEBUG_PREEMPT |
7022 | if (!preempt_count_equals(preempt_offset)) { | |
7023 | pr_err("Preemption disabled at:"); | |
7024 | print_ip_sym(current->preempt_disable_ip); | |
7025 | pr_cont("\n"); | |
7026 | } | |
7027 | #endif | |
aef745fc | 7028 | dump_stack(); |
1da177e4 LT |
7029 | } |
7030 | EXPORT_SYMBOL(__might_sleep); | |
7031 | #endif | |
7032 | ||
7033 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7034 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7035 | { | |
da7a735e | 7036 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7037 | struct sched_attr attr = { |
7038 | .sched_policy = SCHED_NORMAL, | |
7039 | }; | |
da7a735e | 7040 | int old_prio = p->prio; |
3a5e4dc1 | 7041 | int on_rq; |
3e51f33f | 7042 | |
fd2f4419 | 7043 | on_rq = p->on_rq; |
3a5e4dc1 | 7044 | if (on_rq) |
4ca9b72b | 7045 | dequeue_task(rq, p, 0); |
d50dde5a | 7046 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 7047 | if (on_rq) { |
4ca9b72b | 7048 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7049 | resched_task(rq->curr); |
7050 | } | |
da7a735e PZ |
7051 | |
7052 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7053 | } |
7054 | ||
1da177e4 LT |
7055 | void normalize_rt_tasks(void) |
7056 | { | |
a0f98a1c | 7057 | struct task_struct *g, *p; |
1da177e4 | 7058 | unsigned long flags; |
70b97a7f | 7059 | struct rq *rq; |
1da177e4 | 7060 | |
4cf5d77a | 7061 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7062 | do_each_thread(g, p) { |
178be793 IM |
7063 | /* |
7064 | * Only normalize user tasks: | |
7065 | */ | |
7066 | if (!p->mm) | |
7067 | continue; | |
7068 | ||
6cfb0d5d | 7069 | p->se.exec_start = 0; |
6cfb0d5d | 7070 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7071 | p->se.statistics.wait_start = 0; |
7072 | p->se.statistics.sleep_start = 0; | |
7073 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7074 | #endif |
dd41f596 | 7075 | |
aab03e05 | 7076 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7077 | /* |
7078 | * Renice negative nice level userspace | |
7079 | * tasks back to 0: | |
7080 | */ | |
d0ea0268 | 7081 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7082 | set_user_nice(p, 0); |
1da177e4 | 7083 | continue; |
dd41f596 | 7084 | } |
1da177e4 | 7085 | |
1d615482 | 7086 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7087 | rq = __task_rq_lock(p); |
1da177e4 | 7088 | |
178be793 | 7089 | normalize_task(rq, p); |
3a5e4dc1 | 7090 | |
b29739f9 | 7091 | __task_rq_unlock(rq); |
1d615482 | 7092 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7093 | } while_each_thread(g, p); |
7094 | ||
4cf5d77a | 7095 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7096 | } |
7097 | ||
7098 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7099 | |
67fc4e0c | 7100 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7101 | /* |
67fc4e0c | 7102 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7103 | * |
7104 | * They can only be called when the whole system has been | |
7105 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7106 | * activity can take place. Using them for anything else would | |
7107 | * be a serious bug, and as a result, they aren't even visible | |
7108 | * under any other configuration. | |
7109 | */ | |
7110 | ||
7111 | /** | |
7112 | * curr_task - return the current task for a given cpu. | |
7113 | * @cpu: the processor in question. | |
7114 | * | |
7115 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7116 | * |
7117 | * Return: The current task for @cpu. | |
1df5c10a | 7118 | */ |
36c8b586 | 7119 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7120 | { |
7121 | return cpu_curr(cpu); | |
7122 | } | |
7123 | ||
67fc4e0c JW |
7124 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7125 | ||
7126 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7127 | /** |
7128 | * set_curr_task - set the current task for a given cpu. | |
7129 | * @cpu: the processor in question. | |
7130 | * @p: the task pointer to set. | |
7131 | * | |
7132 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7133 | * are serviced on a separate stack. It allows the architecture to switch the |
7134 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7135 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7136 | * and caller must save the original value of the current task (see | |
7137 | * curr_task() above) and restore that value before reenabling interrupts and | |
7138 | * re-starting the system. | |
7139 | * | |
7140 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7141 | */ | |
36c8b586 | 7142 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7143 | { |
7144 | cpu_curr(cpu) = p; | |
7145 | } | |
7146 | ||
7147 | #endif | |
29f59db3 | 7148 | |
7c941438 | 7149 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7150 | /* task_group_lock serializes the addition/removal of task groups */ |
7151 | static DEFINE_SPINLOCK(task_group_lock); | |
7152 | ||
bccbe08a PZ |
7153 | static void free_sched_group(struct task_group *tg) |
7154 | { | |
7155 | free_fair_sched_group(tg); | |
7156 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7157 | autogroup_free(tg); |
bccbe08a PZ |
7158 | kfree(tg); |
7159 | } | |
7160 | ||
7161 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7162 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7163 | { |
7164 | struct task_group *tg; | |
bccbe08a PZ |
7165 | |
7166 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7167 | if (!tg) | |
7168 | return ERR_PTR(-ENOMEM); | |
7169 | ||
ec7dc8ac | 7170 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7171 | goto err; |
7172 | ||
ec7dc8ac | 7173 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7174 | goto err; |
7175 | ||
ace783b9 LZ |
7176 | return tg; |
7177 | ||
7178 | err: | |
7179 | free_sched_group(tg); | |
7180 | return ERR_PTR(-ENOMEM); | |
7181 | } | |
7182 | ||
7183 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7184 | { | |
7185 | unsigned long flags; | |
7186 | ||
8ed36996 | 7187 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7188 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7189 | |
7190 | WARN_ON(!parent); /* root should already exist */ | |
7191 | ||
7192 | tg->parent = parent; | |
f473aa5e | 7193 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7194 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7195 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7196 | } |
7197 | ||
9b5b7751 | 7198 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7199 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7200 | { |
29f59db3 | 7201 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7202 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7203 | } |
7204 | ||
9b5b7751 | 7205 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7206 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7207 | { |
7208 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7209 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7210 | } | |
7211 | ||
7212 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7213 | { |
8ed36996 | 7214 | unsigned long flags; |
9b5b7751 | 7215 | int i; |
29f59db3 | 7216 | |
3d4b47b4 PZ |
7217 | /* end participation in shares distribution */ |
7218 | for_each_possible_cpu(i) | |
bccbe08a | 7219 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7220 | |
7221 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7222 | list_del_rcu(&tg->list); |
f473aa5e | 7223 | list_del_rcu(&tg->siblings); |
8ed36996 | 7224 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7225 | } |
7226 | ||
9b5b7751 | 7227 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7228 | * The caller of this function should have put the task in its new group |
7229 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7230 | * reflect its new group. | |
9b5b7751 SV |
7231 | */ |
7232 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7233 | { |
8323f26c | 7234 | struct task_group *tg; |
29f59db3 SV |
7235 | int on_rq, running; |
7236 | unsigned long flags; | |
7237 | struct rq *rq; | |
7238 | ||
7239 | rq = task_rq_lock(tsk, &flags); | |
7240 | ||
051a1d1a | 7241 | running = task_current(rq, tsk); |
fd2f4419 | 7242 | on_rq = tsk->on_rq; |
29f59db3 | 7243 | |
0e1f3483 | 7244 | if (on_rq) |
29f59db3 | 7245 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7246 | if (unlikely(running)) |
7247 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7248 | |
073219e9 | 7249 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7250 | lockdep_is_held(&tsk->sighand->siglock)), |
7251 | struct task_group, css); | |
7252 | tg = autogroup_task_group(tsk, tg); | |
7253 | tsk->sched_task_group = tg; | |
7254 | ||
810b3817 | 7255 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7256 | if (tsk->sched_class->task_move_group) |
7257 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7258 | else | |
810b3817 | 7259 | #endif |
b2b5ce02 | 7260 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7261 | |
0e1f3483 HS |
7262 | if (unlikely(running)) |
7263 | tsk->sched_class->set_curr_task(rq); | |
7264 | if (on_rq) | |
371fd7e7 | 7265 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7266 | |
0122ec5b | 7267 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7268 | } |
7c941438 | 7269 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7270 | |
a790de99 PT |
7271 | #ifdef CONFIG_RT_GROUP_SCHED |
7272 | /* | |
7273 | * Ensure that the real time constraints are schedulable. | |
7274 | */ | |
7275 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7276 | |
9a7e0b18 PZ |
7277 | /* Must be called with tasklist_lock held */ |
7278 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7279 | { |
9a7e0b18 | 7280 | struct task_struct *g, *p; |
b40b2e8e | 7281 | |
9a7e0b18 | 7282 | do_each_thread(g, p) { |
029632fb | 7283 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7284 | return 1; |
7285 | } while_each_thread(g, p); | |
b40b2e8e | 7286 | |
9a7e0b18 PZ |
7287 | return 0; |
7288 | } | |
b40b2e8e | 7289 | |
9a7e0b18 PZ |
7290 | struct rt_schedulable_data { |
7291 | struct task_group *tg; | |
7292 | u64 rt_period; | |
7293 | u64 rt_runtime; | |
7294 | }; | |
b40b2e8e | 7295 | |
a790de99 | 7296 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7297 | { |
7298 | struct rt_schedulable_data *d = data; | |
7299 | struct task_group *child; | |
7300 | unsigned long total, sum = 0; | |
7301 | u64 period, runtime; | |
b40b2e8e | 7302 | |
9a7e0b18 PZ |
7303 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7304 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7305 | |
9a7e0b18 PZ |
7306 | if (tg == d->tg) { |
7307 | period = d->rt_period; | |
7308 | runtime = d->rt_runtime; | |
b40b2e8e | 7309 | } |
b40b2e8e | 7310 | |
4653f803 PZ |
7311 | /* |
7312 | * Cannot have more runtime than the period. | |
7313 | */ | |
7314 | if (runtime > period && runtime != RUNTIME_INF) | |
7315 | return -EINVAL; | |
6f505b16 | 7316 | |
4653f803 PZ |
7317 | /* |
7318 | * Ensure we don't starve existing RT tasks. | |
7319 | */ | |
9a7e0b18 PZ |
7320 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7321 | return -EBUSY; | |
6f505b16 | 7322 | |
9a7e0b18 | 7323 | total = to_ratio(period, runtime); |
6f505b16 | 7324 | |
4653f803 PZ |
7325 | /* |
7326 | * Nobody can have more than the global setting allows. | |
7327 | */ | |
7328 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7329 | return -EINVAL; | |
6f505b16 | 7330 | |
4653f803 PZ |
7331 | /* |
7332 | * The sum of our children's runtime should not exceed our own. | |
7333 | */ | |
9a7e0b18 PZ |
7334 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7335 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7336 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7337 | |
9a7e0b18 PZ |
7338 | if (child == d->tg) { |
7339 | period = d->rt_period; | |
7340 | runtime = d->rt_runtime; | |
7341 | } | |
6f505b16 | 7342 | |
9a7e0b18 | 7343 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7344 | } |
6f505b16 | 7345 | |
9a7e0b18 PZ |
7346 | if (sum > total) |
7347 | return -EINVAL; | |
7348 | ||
7349 | return 0; | |
6f505b16 PZ |
7350 | } |
7351 | ||
9a7e0b18 | 7352 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7353 | { |
8277434e PT |
7354 | int ret; |
7355 | ||
9a7e0b18 PZ |
7356 | struct rt_schedulable_data data = { |
7357 | .tg = tg, | |
7358 | .rt_period = period, | |
7359 | .rt_runtime = runtime, | |
7360 | }; | |
7361 | ||
8277434e PT |
7362 | rcu_read_lock(); |
7363 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7364 | rcu_read_unlock(); | |
7365 | ||
7366 | return ret; | |
521f1a24 DG |
7367 | } |
7368 | ||
ab84d31e | 7369 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7370 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7371 | { |
ac086bc2 | 7372 | int i, err = 0; |
9f0c1e56 | 7373 | |
9f0c1e56 | 7374 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7375 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7376 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7377 | if (err) | |
9f0c1e56 | 7378 | goto unlock; |
ac086bc2 | 7379 | |
0986b11b | 7380 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7381 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7382 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7383 | |
7384 | for_each_possible_cpu(i) { | |
7385 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7386 | ||
0986b11b | 7387 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7388 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7389 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7390 | } |
0986b11b | 7391 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7392 | unlock: |
521f1a24 | 7393 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7394 | mutex_unlock(&rt_constraints_mutex); |
7395 | ||
7396 | return err; | |
6f505b16 PZ |
7397 | } |
7398 | ||
25cc7da7 | 7399 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7400 | { |
7401 | u64 rt_runtime, rt_period; | |
7402 | ||
7403 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7404 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7405 | if (rt_runtime_us < 0) | |
7406 | rt_runtime = RUNTIME_INF; | |
7407 | ||
ab84d31e | 7408 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7409 | } |
7410 | ||
25cc7da7 | 7411 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7412 | { |
7413 | u64 rt_runtime_us; | |
7414 | ||
d0b27fa7 | 7415 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7416 | return -1; |
7417 | ||
d0b27fa7 | 7418 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7419 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7420 | return rt_runtime_us; | |
7421 | } | |
d0b27fa7 | 7422 | |
25cc7da7 | 7423 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7424 | { |
7425 | u64 rt_runtime, rt_period; | |
7426 | ||
7427 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7428 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7429 | ||
619b0488 R |
7430 | if (rt_period == 0) |
7431 | return -EINVAL; | |
7432 | ||
ab84d31e | 7433 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7434 | } |
7435 | ||
25cc7da7 | 7436 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7437 | { |
7438 | u64 rt_period_us; | |
7439 | ||
7440 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7441 | do_div(rt_period_us, NSEC_PER_USEC); | |
7442 | return rt_period_us; | |
7443 | } | |
332ac17e | 7444 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7445 | |
332ac17e | 7446 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7447 | static int sched_rt_global_constraints(void) |
7448 | { | |
7449 | int ret = 0; | |
7450 | ||
7451 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7452 | read_lock(&tasklist_lock); |
4653f803 | 7453 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7454 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7455 | mutex_unlock(&rt_constraints_mutex); |
7456 | ||
7457 | return ret; | |
7458 | } | |
54e99124 | 7459 | |
25cc7da7 | 7460 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7461 | { |
7462 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7463 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7464 | return 0; | |
7465 | ||
7466 | return 1; | |
7467 | } | |
7468 | ||
6d6bc0ad | 7469 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7470 | static int sched_rt_global_constraints(void) |
7471 | { | |
ac086bc2 | 7472 | unsigned long flags; |
332ac17e | 7473 | int i, ret = 0; |
ec5d4989 | 7474 | |
0986b11b | 7475 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7476 | for_each_possible_cpu(i) { |
7477 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7478 | ||
0986b11b | 7479 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7480 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7481 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7482 | } |
0986b11b | 7483 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7484 | |
332ac17e | 7485 | return ret; |
d0b27fa7 | 7486 | } |
6d6bc0ad | 7487 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7488 | |
332ac17e DF |
7489 | static int sched_dl_global_constraints(void) |
7490 | { | |
1724813d PZ |
7491 | u64 runtime = global_rt_runtime(); |
7492 | u64 period = global_rt_period(); | |
332ac17e | 7493 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7494 | int cpu, ret = 0; |
49516342 | 7495 | unsigned long flags; |
332ac17e DF |
7496 | |
7497 | /* | |
7498 | * Here we want to check the bandwidth not being set to some | |
7499 | * value smaller than the currently allocated bandwidth in | |
7500 | * any of the root_domains. | |
7501 | * | |
7502 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7503 | * cycling on root_domains... Discussion on different/better | |
7504 | * solutions is welcome! | |
7505 | */ | |
1724813d PZ |
7506 | for_each_possible_cpu(cpu) { |
7507 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e | 7508 | |
49516342 | 7509 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7510 | if (new_bw < dl_b->total_bw) |
7511 | ret = -EBUSY; | |
49516342 | 7512 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d PZ |
7513 | |
7514 | if (ret) | |
7515 | break; | |
332ac17e DF |
7516 | } |
7517 | ||
1724813d | 7518 | return ret; |
332ac17e DF |
7519 | } |
7520 | ||
1724813d | 7521 | static void sched_dl_do_global(void) |
ce0dbbbb | 7522 | { |
1724813d PZ |
7523 | u64 new_bw = -1; |
7524 | int cpu; | |
49516342 | 7525 | unsigned long flags; |
ce0dbbbb | 7526 | |
1724813d PZ |
7527 | def_dl_bandwidth.dl_period = global_rt_period(); |
7528 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7529 | ||
7530 | if (global_rt_runtime() != RUNTIME_INF) | |
7531 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7532 | ||
7533 | /* | |
7534 | * FIXME: As above... | |
7535 | */ | |
7536 | for_each_possible_cpu(cpu) { | |
7537 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7538 | ||
49516342 | 7539 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7540 | dl_b->bw = new_bw; |
49516342 | 7541 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
ce0dbbbb | 7542 | } |
1724813d PZ |
7543 | } |
7544 | ||
7545 | static int sched_rt_global_validate(void) | |
7546 | { | |
7547 | if (sysctl_sched_rt_period <= 0) | |
7548 | return -EINVAL; | |
7549 | ||
e9e7cb38 JL |
7550 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7551 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7552 | return -EINVAL; |
7553 | ||
7554 | return 0; | |
7555 | } | |
7556 | ||
7557 | static void sched_rt_do_global(void) | |
7558 | { | |
7559 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7560 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7561 | } |
7562 | ||
d0b27fa7 | 7563 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7564 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7565 | loff_t *ppos) |
7566 | { | |
d0b27fa7 PZ |
7567 | int old_period, old_runtime; |
7568 | static DEFINE_MUTEX(mutex); | |
1724813d | 7569 | int ret; |
d0b27fa7 PZ |
7570 | |
7571 | mutex_lock(&mutex); | |
7572 | old_period = sysctl_sched_rt_period; | |
7573 | old_runtime = sysctl_sched_rt_runtime; | |
7574 | ||
8d65af78 | 7575 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7576 | |
7577 | if (!ret && write) { | |
1724813d PZ |
7578 | ret = sched_rt_global_validate(); |
7579 | if (ret) | |
7580 | goto undo; | |
7581 | ||
d0b27fa7 | 7582 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7583 | if (ret) |
7584 | goto undo; | |
7585 | ||
7586 | ret = sched_dl_global_constraints(); | |
7587 | if (ret) | |
7588 | goto undo; | |
7589 | ||
7590 | sched_rt_do_global(); | |
7591 | sched_dl_do_global(); | |
7592 | } | |
7593 | if (0) { | |
7594 | undo: | |
7595 | sysctl_sched_rt_period = old_period; | |
7596 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7597 | } |
7598 | mutex_unlock(&mutex); | |
7599 | ||
7600 | return ret; | |
7601 | } | |
68318b8e | 7602 | |
1724813d | 7603 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7604 | void __user *buffer, size_t *lenp, |
7605 | loff_t *ppos) | |
7606 | { | |
7607 | int ret; | |
332ac17e | 7608 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7609 | |
7610 | mutex_lock(&mutex); | |
332ac17e | 7611 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7612 | /* make sure that internally we keep jiffies */ |
7613 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7614 | if (!ret && write) { |
1724813d PZ |
7615 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7616 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7617 | } |
7618 | mutex_unlock(&mutex); | |
332ac17e DF |
7619 | return ret; |
7620 | } | |
7621 | ||
052f1dc7 | 7622 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7623 | |
a7c6d554 | 7624 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7625 | { |
a7c6d554 | 7626 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7627 | } |
7628 | ||
eb95419b TH |
7629 | static struct cgroup_subsys_state * |
7630 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7631 | { |
eb95419b TH |
7632 | struct task_group *parent = css_tg(parent_css); |
7633 | struct task_group *tg; | |
68318b8e | 7634 | |
eb95419b | 7635 | if (!parent) { |
68318b8e | 7636 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7637 | return &root_task_group.css; |
68318b8e SV |
7638 | } |
7639 | ||
ec7dc8ac | 7640 | tg = sched_create_group(parent); |
68318b8e SV |
7641 | if (IS_ERR(tg)) |
7642 | return ERR_PTR(-ENOMEM); | |
7643 | ||
68318b8e SV |
7644 | return &tg->css; |
7645 | } | |
7646 | ||
eb95419b | 7647 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7648 | { |
eb95419b TH |
7649 | struct task_group *tg = css_tg(css); |
7650 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7651 | |
63876986 TH |
7652 | if (parent) |
7653 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7654 | return 0; |
7655 | } | |
7656 | ||
eb95419b | 7657 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7658 | { |
eb95419b | 7659 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7660 | |
7661 | sched_destroy_group(tg); | |
7662 | } | |
7663 | ||
eb95419b | 7664 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7665 | { |
eb95419b | 7666 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7667 | |
7668 | sched_offline_group(tg); | |
7669 | } | |
7670 | ||
eb95419b | 7671 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7672 | struct cgroup_taskset *tset) |
68318b8e | 7673 | { |
bb9d97b6 TH |
7674 | struct task_struct *task; |
7675 | ||
924f0d9a | 7676 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7677 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7678 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7679 | return -EINVAL; |
b68aa230 | 7680 | #else |
bb9d97b6 TH |
7681 | /* We don't support RT-tasks being in separate groups */ |
7682 | if (task->sched_class != &fair_sched_class) | |
7683 | return -EINVAL; | |
b68aa230 | 7684 | #endif |
bb9d97b6 | 7685 | } |
be367d09 BB |
7686 | return 0; |
7687 | } | |
68318b8e | 7688 | |
eb95419b | 7689 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7690 | struct cgroup_taskset *tset) |
68318b8e | 7691 | { |
bb9d97b6 TH |
7692 | struct task_struct *task; |
7693 | ||
924f0d9a | 7694 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7695 | sched_move_task(task); |
68318b8e SV |
7696 | } |
7697 | ||
eb95419b TH |
7698 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7699 | struct cgroup_subsys_state *old_css, | |
7700 | struct task_struct *task) | |
068c5cc5 PZ |
7701 | { |
7702 | /* | |
7703 | * cgroup_exit() is called in the copy_process() failure path. | |
7704 | * Ignore this case since the task hasn't ran yet, this avoids | |
7705 | * trying to poke a half freed task state from generic code. | |
7706 | */ | |
7707 | if (!(task->flags & PF_EXITING)) | |
7708 | return; | |
7709 | ||
7710 | sched_move_task(task); | |
7711 | } | |
7712 | ||
052f1dc7 | 7713 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7714 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7715 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7716 | { |
182446d0 | 7717 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7718 | } |
7719 | ||
182446d0 TH |
7720 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7721 | struct cftype *cft) | |
68318b8e | 7722 | { |
182446d0 | 7723 | struct task_group *tg = css_tg(css); |
68318b8e | 7724 | |
c8b28116 | 7725 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7726 | } |
ab84d31e PT |
7727 | |
7728 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7729 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7730 | ||
ab84d31e PT |
7731 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7732 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7733 | ||
a790de99 PT |
7734 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7735 | ||
ab84d31e PT |
7736 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7737 | { | |
56f570e5 | 7738 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7739 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7740 | |
7741 | if (tg == &root_task_group) | |
7742 | return -EINVAL; | |
7743 | ||
7744 | /* | |
7745 | * Ensure we have at some amount of bandwidth every period. This is | |
7746 | * to prevent reaching a state of large arrears when throttled via | |
7747 | * entity_tick() resulting in prolonged exit starvation. | |
7748 | */ | |
7749 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7750 | return -EINVAL; | |
7751 | ||
7752 | /* | |
7753 | * Likewise, bound things on the otherside by preventing insane quota | |
7754 | * periods. This also allows us to normalize in computing quota | |
7755 | * feasibility. | |
7756 | */ | |
7757 | if (period > max_cfs_quota_period) | |
7758 | return -EINVAL; | |
7759 | ||
a790de99 PT |
7760 | mutex_lock(&cfs_constraints_mutex); |
7761 | ret = __cfs_schedulable(tg, period, quota); | |
7762 | if (ret) | |
7763 | goto out_unlock; | |
7764 | ||
58088ad0 | 7765 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7766 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7767 | /* |
7768 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7769 | * before making related changes, and on->off must occur afterwards | |
7770 | */ | |
7771 | if (runtime_enabled && !runtime_was_enabled) | |
7772 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7773 | raw_spin_lock_irq(&cfs_b->lock); |
7774 | cfs_b->period = ns_to_ktime(period); | |
7775 | cfs_b->quota = quota; | |
58088ad0 | 7776 | |
a9cf55b2 | 7777 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7778 | /* restart the period timer (if active) to handle new period expiry */ |
7779 | if (runtime_enabled && cfs_b->timer_active) { | |
7780 | /* force a reprogram */ | |
7781 | cfs_b->timer_active = 0; | |
7782 | __start_cfs_bandwidth(cfs_b); | |
7783 | } | |
ab84d31e PT |
7784 | raw_spin_unlock_irq(&cfs_b->lock); |
7785 | ||
7786 | for_each_possible_cpu(i) { | |
7787 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7788 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7789 | |
7790 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7791 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7792 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7793 | |
029632fb | 7794 | if (cfs_rq->throttled) |
671fd9da | 7795 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7796 | raw_spin_unlock_irq(&rq->lock); |
7797 | } | |
1ee14e6c BS |
7798 | if (runtime_was_enabled && !runtime_enabled) |
7799 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7800 | out_unlock: |
7801 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7802 | |
a790de99 | 7803 | return ret; |
ab84d31e PT |
7804 | } |
7805 | ||
7806 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7807 | { | |
7808 | u64 quota, period; | |
7809 | ||
029632fb | 7810 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7811 | if (cfs_quota_us < 0) |
7812 | quota = RUNTIME_INF; | |
7813 | else | |
7814 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7815 | ||
7816 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7817 | } | |
7818 | ||
7819 | long tg_get_cfs_quota(struct task_group *tg) | |
7820 | { | |
7821 | u64 quota_us; | |
7822 | ||
029632fb | 7823 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7824 | return -1; |
7825 | ||
029632fb | 7826 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7827 | do_div(quota_us, NSEC_PER_USEC); |
7828 | ||
7829 | return quota_us; | |
7830 | } | |
7831 | ||
7832 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7833 | { | |
7834 | u64 quota, period; | |
7835 | ||
7836 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7837 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7838 | |
ab84d31e PT |
7839 | return tg_set_cfs_bandwidth(tg, period, quota); |
7840 | } | |
7841 | ||
7842 | long tg_get_cfs_period(struct task_group *tg) | |
7843 | { | |
7844 | u64 cfs_period_us; | |
7845 | ||
029632fb | 7846 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7847 | do_div(cfs_period_us, NSEC_PER_USEC); |
7848 | ||
7849 | return cfs_period_us; | |
7850 | } | |
7851 | ||
182446d0 TH |
7852 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7853 | struct cftype *cft) | |
ab84d31e | 7854 | { |
182446d0 | 7855 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7856 | } |
7857 | ||
182446d0 TH |
7858 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7859 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7860 | { |
182446d0 | 7861 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7862 | } |
7863 | ||
182446d0 TH |
7864 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7865 | struct cftype *cft) | |
ab84d31e | 7866 | { |
182446d0 | 7867 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7868 | } |
7869 | ||
182446d0 TH |
7870 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7871 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7872 | { |
182446d0 | 7873 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7874 | } |
7875 | ||
a790de99 PT |
7876 | struct cfs_schedulable_data { |
7877 | struct task_group *tg; | |
7878 | u64 period, quota; | |
7879 | }; | |
7880 | ||
7881 | /* | |
7882 | * normalize group quota/period to be quota/max_period | |
7883 | * note: units are usecs | |
7884 | */ | |
7885 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7886 | struct cfs_schedulable_data *d) | |
7887 | { | |
7888 | u64 quota, period; | |
7889 | ||
7890 | if (tg == d->tg) { | |
7891 | period = d->period; | |
7892 | quota = d->quota; | |
7893 | } else { | |
7894 | period = tg_get_cfs_period(tg); | |
7895 | quota = tg_get_cfs_quota(tg); | |
7896 | } | |
7897 | ||
7898 | /* note: these should typically be equivalent */ | |
7899 | if (quota == RUNTIME_INF || quota == -1) | |
7900 | return RUNTIME_INF; | |
7901 | ||
7902 | return to_ratio(period, quota); | |
7903 | } | |
7904 | ||
7905 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7906 | { | |
7907 | struct cfs_schedulable_data *d = data; | |
029632fb | 7908 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7909 | s64 quota = 0, parent_quota = -1; |
7910 | ||
7911 | if (!tg->parent) { | |
7912 | quota = RUNTIME_INF; | |
7913 | } else { | |
029632fb | 7914 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7915 | |
7916 | quota = normalize_cfs_quota(tg, d); | |
7917 | parent_quota = parent_b->hierarchal_quota; | |
7918 | ||
7919 | /* | |
7920 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7921 | * limit is set | |
7922 | */ | |
7923 | if (quota == RUNTIME_INF) | |
7924 | quota = parent_quota; | |
7925 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7926 | return -EINVAL; | |
7927 | } | |
7928 | cfs_b->hierarchal_quota = quota; | |
7929 | ||
7930 | return 0; | |
7931 | } | |
7932 | ||
7933 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7934 | { | |
8277434e | 7935 | int ret; |
a790de99 PT |
7936 | struct cfs_schedulable_data data = { |
7937 | .tg = tg, | |
7938 | .period = period, | |
7939 | .quota = quota, | |
7940 | }; | |
7941 | ||
7942 | if (quota != RUNTIME_INF) { | |
7943 | do_div(data.period, NSEC_PER_USEC); | |
7944 | do_div(data.quota, NSEC_PER_USEC); | |
7945 | } | |
7946 | ||
8277434e PT |
7947 | rcu_read_lock(); |
7948 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7949 | rcu_read_unlock(); | |
7950 | ||
7951 | return ret; | |
a790de99 | 7952 | } |
e8da1b18 | 7953 | |
2da8ca82 | 7954 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 7955 | { |
2da8ca82 | 7956 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7957 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7958 | |
44ffc75b TH |
7959 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7960 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7961 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
7962 | |
7963 | return 0; | |
7964 | } | |
ab84d31e | 7965 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7966 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7967 | |
052f1dc7 | 7968 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7969 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7970 | struct cftype *cft, s64 val) | |
6f505b16 | 7971 | { |
182446d0 | 7972 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7973 | } |
7974 | ||
182446d0 TH |
7975 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7976 | struct cftype *cft) | |
6f505b16 | 7977 | { |
182446d0 | 7978 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7979 | } |
d0b27fa7 | 7980 | |
182446d0 TH |
7981 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7982 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7983 | { |
182446d0 | 7984 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7985 | } |
7986 | ||
182446d0 TH |
7987 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7988 | struct cftype *cft) | |
d0b27fa7 | 7989 | { |
182446d0 | 7990 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7991 | } |
6d6bc0ad | 7992 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7993 | |
fe5c7cc2 | 7994 | static struct cftype cpu_files[] = { |
052f1dc7 | 7995 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7996 | { |
7997 | .name = "shares", | |
f4c753b7 PM |
7998 | .read_u64 = cpu_shares_read_u64, |
7999 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8000 | }, |
052f1dc7 | 8001 | #endif |
ab84d31e PT |
8002 | #ifdef CONFIG_CFS_BANDWIDTH |
8003 | { | |
8004 | .name = "cfs_quota_us", | |
8005 | .read_s64 = cpu_cfs_quota_read_s64, | |
8006 | .write_s64 = cpu_cfs_quota_write_s64, | |
8007 | }, | |
8008 | { | |
8009 | .name = "cfs_period_us", | |
8010 | .read_u64 = cpu_cfs_period_read_u64, | |
8011 | .write_u64 = cpu_cfs_period_write_u64, | |
8012 | }, | |
e8da1b18 NR |
8013 | { |
8014 | .name = "stat", | |
2da8ca82 | 8015 | .seq_show = cpu_stats_show, |
e8da1b18 | 8016 | }, |
ab84d31e | 8017 | #endif |
052f1dc7 | 8018 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8019 | { |
9f0c1e56 | 8020 | .name = "rt_runtime_us", |
06ecb27c PM |
8021 | .read_s64 = cpu_rt_runtime_read, |
8022 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8023 | }, |
d0b27fa7 PZ |
8024 | { |
8025 | .name = "rt_period_us", | |
f4c753b7 PM |
8026 | .read_u64 = cpu_rt_period_read_uint, |
8027 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8028 | }, |
052f1dc7 | 8029 | #endif |
4baf6e33 | 8030 | { } /* terminate */ |
68318b8e SV |
8031 | }; |
8032 | ||
073219e9 | 8033 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8034 | .css_alloc = cpu_cgroup_css_alloc, |
8035 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8036 | .css_online = cpu_cgroup_css_online, |
8037 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8038 | .can_attach = cpu_cgroup_can_attach, |
8039 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8040 | .exit = cpu_cgroup_exit, |
4baf6e33 | 8041 | .base_cftypes = cpu_files, |
68318b8e SV |
8042 | .early_init = 1, |
8043 | }; | |
8044 | ||
052f1dc7 | 8045 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8046 | |
b637a328 PM |
8047 | void dump_cpu_task(int cpu) |
8048 | { | |
8049 | pr_info("Task dump for CPU %d:\n", cpu); | |
8050 | sched_show_task(cpu_curr(cpu)); | |
8051 | } |