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 | 566 | smp_send_reschedule(cpu); |
dfc68f29 AL |
567 | else |
568 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
569 | } |
570 | ||
029632fb | 571 | void resched_cpu(int cpu) |
c24d20db IM |
572 | { |
573 | struct rq *rq = cpu_rq(cpu); | |
574 | unsigned long flags; | |
575 | ||
05fa785c | 576 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
577 | return; |
578 | resched_task(cpu_curr(cpu)); | |
05fa785c | 579 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 580 | } |
06d8308c | 581 | |
b021fe3e | 582 | #ifdef CONFIG_SMP |
3451d024 | 583 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 VP |
584 | /* |
585 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
586 | * from an idle cpu. This is good for power-savings. | |
587 | * | |
588 | * We don't do similar optimization for completely idle system, as | |
589 | * selecting an idle cpu will add more delays to the timers than intended | |
590 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
591 | */ | |
6201b4d6 | 592 | int get_nohz_timer_target(int pinned) |
83cd4fe2 VP |
593 | { |
594 | int cpu = smp_processor_id(); | |
595 | int i; | |
596 | struct sched_domain *sd; | |
597 | ||
6201b4d6 VK |
598 | if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu)) |
599 | return cpu; | |
600 | ||
057f3fad | 601 | rcu_read_lock(); |
83cd4fe2 | 602 | for_each_domain(cpu, sd) { |
057f3fad PZ |
603 | for_each_cpu(i, sched_domain_span(sd)) { |
604 | if (!idle_cpu(i)) { | |
605 | cpu = i; | |
606 | goto unlock; | |
607 | } | |
608 | } | |
83cd4fe2 | 609 | } |
057f3fad PZ |
610 | unlock: |
611 | rcu_read_unlock(); | |
83cd4fe2 VP |
612 | return cpu; |
613 | } | |
06d8308c TG |
614 | /* |
615 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
616 | * idle CPU then this timer might expire before the next timer event | |
617 | * which is scheduled to wake up that CPU. In case of a completely | |
618 | * idle system the next event might even be infinite time into the | |
619 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
620 | * leaves the inner idle loop so the newly added timer is taken into | |
621 | * account when the CPU goes back to idle and evaluates the timer | |
622 | * wheel for the next timer event. | |
623 | */ | |
1c20091e | 624 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
625 | { |
626 | struct rq *rq = cpu_rq(cpu); | |
627 | ||
628 | if (cpu == smp_processor_id()) | |
629 | return; | |
630 | ||
67b9ca70 | 631 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 632 | smp_send_reschedule(cpu); |
dfc68f29 AL |
633 | else |
634 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
635 | } |
636 | ||
c5bfece2 | 637 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 638 | { |
c5bfece2 | 639 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
640 | if (cpu != smp_processor_id() || |
641 | tick_nohz_tick_stopped()) | |
642 | smp_send_reschedule(cpu); | |
643 | return true; | |
644 | } | |
645 | ||
646 | return false; | |
647 | } | |
648 | ||
649 | void wake_up_nohz_cpu(int cpu) | |
650 | { | |
c5bfece2 | 651 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
652 | wake_up_idle_cpu(cpu); |
653 | } | |
654 | ||
ca38062e | 655 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 656 | { |
1c792db7 | 657 | int cpu = smp_processor_id(); |
873b4c65 VG |
658 | |
659 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
660 | return false; | |
661 | ||
662 | if (idle_cpu(cpu) && !need_resched()) | |
663 | return true; | |
664 | ||
665 | /* | |
666 | * We can't run Idle Load Balance on this CPU for this time so we | |
667 | * cancel it and clear NOHZ_BALANCE_KICK | |
668 | */ | |
669 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
670 | return false; | |
45bf76df IM |
671 | } |
672 | ||
3451d024 | 673 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 674 | |
ca38062e | 675 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 676 | { |
ca38062e | 677 | return false; |
2069dd75 PZ |
678 | } |
679 | ||
3451d024 | 680 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 681 | |
ce831b38 FW |
682 | #ifdef CONFIG_NO_HZ_FULL |
683 | bool sched_can_stop_tick(void) | |
684 | { | |
685 | struct rq *rq; | |
686 | ||
687 | rq = this_rq(); | |
688 | ||
689 | /* Make sure rq->nr_running update is visible after the IPI */ | |
690 | smp_rmb(); | |
691 | ||
692 | /* More than one running task need preemption */ | |
693 | if (rq->nr_running > 1) | |
694 | return false; | |
695 | ||
696 | return true; | |
697 | } | |
698 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 699 | |
029632fb | 700 | void sched_avg_update(struct rq *rq) |
18d95a28 | 701 | { |
e9e9250b PZ |
702 | s64 period = sched_avg_period(); |
703 | ||
78becc27 | 704 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
705 | /* |
706 | * Inline assembly required to prevent the compiler | |
707 | * optimising this loop into a divmod call. | |
708 | * See __iter_div_u64_rem() for another example of this. | |
709 | */ | |
710 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
711 | rq->age_stamp += period; |
712 | rq->rt_avg /= 2; | |
713 | } | |
18d95a28 PZ |
714 | } |
715 | ||
6d6bc0ad | 716 | #endif /* CONFIG_SMP */ |
18d95a28 | 717 | |
a790de99 PT |
718 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
719 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 720 | /* |
8277434e PT |
721 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
722 | * node and @up when leaving it for the final time. | |
723 | * | |
724 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 725 | */ |
029632fb | 726 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 727 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
728 | { |
729 | struct task_group *parent, *child; | |
eb755805 | 730 | int ret; |
c09595f6 | 731 | |
8277434e PT |
732 | parent = from; |
733 | ||
c09595f6 | 734 | down: |
eb755805 PZ |
735 | ret = (*down)(parent, data); |
736 | if (ret) | |
8277434e | 737 | goto out; |
c09595f6 PZ |
738 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
739 | parent = child; | |
740 | goto down; | |
741 | ||
742 | up: | |
743 | continue; | |
744 | } | |
eb755805 | 745 | ret = (*up)(parent, data); |
8277434e PT |
746 | if (ret || parent == from) |
747 | goto out; | |
c09595f6 PZ |
748 | |
749 | child = parent; | |
750 | parent = parent->parent; | |
751 | if (parent) | |
752 | goto up; | |
8277434e | 753 | out: |
eb755805 | 754 | return ret; |
c09595f6 PZ |
755 | } |
756 | ||
029632fb | 757 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 758 | { |
e2b245f8 | 759 | return 0; |
eb755805 | 760 | } |
18d95a28 PZ |
761 | #endif |
762 | ||
45bf76df IM |
763 | static void set_load_weight(struct task_struct *p) |
764 | { | |
f05998d4 NR |
765 | int prio = p->static_prio - MAX_RT_PRIO; |
766 | struct load_weight *load = &p->se.load; | |
767 | ||
dd41f596 IM |
768 | /* |
769 | * SCHED_IDLE tasks get minimal weight: | |
770 | */ | |
771 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 772 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 773 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
774 | return; |
775 | } | |
71f8bd46 | 776 | |
c8b28116 | 777 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 778 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
779 | } |
780 | ||
371fd7e7 | 781 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 782 | { |
a64692a3 | 783 | update_rq_clock(rq); |
43148951 | 784 | sched_info_queued(rq, p); |
371fd7e7 | 785 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
786 | } |
787 | ||
371fd7e7 | 788 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 789 | { |
a64692a3 | 790 | update_rq_clock(rq); |
43148951 | 791 | sched_info_dequeued(rq, p); |
371fd7e7 | 792 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
793 | } |
794 | ||
029632fb | 795 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
796 | { |
797 | if (task_contributes_to_load(p)) | |
798 | rq->nr_uninterruptible--; | |
799 | ||
371fd7e7 | 800 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
801 | } |
802 | ||
029632fb | 803 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
804 | { |
805 | if (task_contributes_to_load(p)) | |
806 | rq->nr_uninterruptible++; | |
807 | ||
371fd7e7 | 808 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
809 | } |
810 | ||
fe44d621 | 811 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 812 | { |
095c0aa8 GC |
813 | /* |
814 | * In theory, the compile should just see 0 here, and optimize out the call | |
815 | * to sched_rt_avg_update. But I don't trust it... | |
816 | */ | |
817 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
818 | s64 steal = 0, irq_delta = 0; | |
819 | #endif | |
820 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8e92c201 | 821 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
822 | |
823 | /* | |
824 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
825 | * this case when a previous update_rq_clock() happened inside a | |
826 | * {soft,}irq region. | |
827 | * | |
828 | * When this happens, we stop ->clock_task and only update the | |
829 | * prev_irq_time stamp to account for the part that fit, so that a next | |
830 | * update will consume the rest. This ensures ->clock_task is | |
831 | * monotonic. | |
832 | * | |
833 | * It does however cause some slight miss-attribution of {soft,}irq | |
834 | * time, a more accurate solution would be to update the irq_time using | |
835 | * the current rq->clock timestamp, except that would require using | |
836 | * atomic ops. | |
837 | */ | |
838 | if (irq_delta > delta) | |
839 | irq_delta = delta; | |
840 | ||
841 | rq->prev_irq_time += irq_delta; | |
842 | delta -= irq_delta; | |
095c0aa8 GC |
843 | #endif |
844 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
c5905afb | 845 | if (static_key_false((¶virt_steal_rq_enabled))) { |
095c0aa8 GC |
846 | steal = paravirt_steal_clock(cpu_of(rq)); |
847 | steal -= rq->prev_steal_time_rq; | |
848 | ||
849 | if (unlikely(steal > delta)) | |
850 | steal = delta; | |
851 | ||
095c0aa8 | 852 | rq->prev_steal_time_rq += steal; |
095c0aa8 GC |
853 | delta -= steal; |
854 | } | |
855 | #endif | |
856 | ||
fe44d621 PZ |
857 | rq->clock_task += delta; |
858 | ||
095c0aa8 | 859 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
5d4dfddd | 860 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
095c0aa8 GC |
861 | sched_rt_avg_update(rq, irq_delta + steal); |
862 | #endif | |
aa483808 VP |
863 | } |
864 | ||
34f971f6 PZ |
865 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
866 | { | |
867 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
868 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
869 | ||
870 | if (stop) { | |
871 | /* | |
872 | * Make it appear like a SCHED_FIFO task, its something | |
873 | * userspace knows about and won't get confused about. | |
874 | * | |
875 | * Also, it will make PI more or less work without too | |
876 | * much confusion -- but then, stop work should not | |
877 | * rely on PI working anyway. | |
878 | */ | |
879 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
880 | ||
881 | stop->sched_class = &stop_sched_class; | |
882 | } | |
883 | ||
884 | cpu_rq(cpu)->stop = stop; | |
885 | ||
886 | if (old_stop) { | |
887 | /* | |
888 | * Reset it back to a normal scheduling class so that | |
889 | * it can die in pieces. | |
890 | */ | |
891 | old_stop->sched_class = &rt_sched_class; | |
892 | } | |
893 | } | |
894 | ||
14531189 | 895 | /* |
dd41f596 | 896 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 897 | */ |
14531189 IM |
898 | static inline int __normal_prio(struct task_struct *p) |
899 | { | |
dd41f596 | 900 | return p->static_prio; |
14531189 IM |
901 | } |
902 | ||
b29739f9 IM |
903 | /* |
904 | * Calculate the expected normal priority: i.e. priority | |
905 | * without taking RT-inheritance into account. Might be | |
906 | * boosted by interactivity modifiers. Changes upon fork, | |
907 | * setprio syscalls, and whenever the interactivity | |
908 | * estimator recalculates. | |
909 | */ | |
36c8b586 | 910 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
911 | { |
912 | int prio; | |
913 | ||
aab03e05 DF |
914 | if (task_has_dl_policy(p)) |
915 | prio = MAX_DL_PRIO-1; | |
916 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
917 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
918 | else | |
919 | prio = __normal_prio(p); | |
920 | return prio; | |
921 | } | |
922 | ||
923 | /* | |
924 | * Calculate the current priority, i.e. the priority | |
925 | * taken into account by the scheduler. This value might | |
926 | * be boosted by RT tasks, or might be boosted by | |
927 | * interactivity modifiers. Will be RT if the task got | |
928 | * RT-boosted. If not then it returns p->normal_prio. | |
929 | */ | |
36c8b586 | 930 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
931 | { |
932 | p->normal_prio = normal_prio(p); | |
933 | /* | |
934 | * If we are RT tasks or we were boosted to RT priority, | |
935 | * keep the priority unchanged. Otherwise, update priority | |
936 | * to the normal priority: | |
937 | */ | |
938 | if (!rt_prio(p->prio)) | |
939 | return p->normal_prio; | |
940 | return p->prio; | |
941 | } | |
942 | ||
1da177e4 LT |
943 | /** |
944 | * task_curr - is this task currently executing on a CPU? | |
945 | * @p: the task in question. | |
e69f6186 YB |
946 | * |
947 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 948 | */ |
36c8b586 | 949 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
950 | { |
951 | return cpu_curr(task_cpu(p)) == p; | |
952 | } | |
953 | ||
cb469845 SR |
954 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
955 | const struct sched_class *prev_class, | |
da7a735e | 956 | int oldprio) |
cb469845 SR |
957 | { |
958 | if (prev_class != p->sched_class) { | |
959 | if (prev_class->switched_from) | |
da7a735e PZ |
960 | prev_class->switched_from(rq, p); |
961 | p->sched_class->switched_to(rq, p); | |
2d3d891d | 962 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 963 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
964 | } |
965 | ||
029632fb | 966 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
967 | { |
968 | const struct sched_class *class; | |
969 | ||
970 | if (p->sched_class == rq->curr->sched_class) { | |
971 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
972 | } else { | |
973 | for_each_class(class) { | |
974 | if (class == rq->curr->sched_class) | |
975 | break; | |
976 | if (class == p->sched_class) { | |
977 | resched_task(rq->curr); | |
978 | break; | |
979 | } | |
980 | } | |
981 | } | |
982 | ||
983 | /* | |
984 | * A queue event has occurred, and we're going to schedule. In | |
985 | * this case, we can save a useless back to back clock update. | |
986 | */ | |
fd2f4419 | 987 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
988 | rq->skip_clock_update = 1; |
989 | } | |
990 | ||
1da177e4 | 991 | #ifdef CONFIG_SMP |
dd41f596 | 992 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 993 | { |
e2912009 PZ |
994 | #ifdef CONFIG_SCHED_DEBUG |
995 | /* | |
996 | * We should never call set_task_cpu() on a blocked task, | |
997 | * ttwu() will sort out the placement. | |
998 | */ | |
077614ee | 999 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
01028747 | 1000 | !(task_preempt_count(p) & PREEMPT_ACTIVE)); |
0122ec5b PZ |
1001 | |
1002 | #ifdef CONFIG_LOCKDEP | |
6c6c54e1 PZ |
1003 | /* |
1004 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1005 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1006 | * | |
1007 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1008 | * see task_group(). |
6c6c54e1 PZ |
1009 | * |
1010 | * Furthermore, all task_rq users should acquire both locks, see | |
1011 | * task_rq_lock(). | |
1012 | */ | |
0122ec5b PZ |
1013 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1014 | lockdep_is_held(&task_rq(p)->lock))); | |
1015 | #endif | |
e2912009 PZ |
1016 | #endif |
1017 | ||
de1d7286 | 1018 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1019 | |
0c69774e | 1020 | if (task_cpu(p) != new_cpu) { |
0a74bef8 PT |
1021 | if (p->sched_class->migrate_task_rq) |
1022 | p->sched_class->migrate_task_rq(p, new_cpu); | |
0c69774e | 1023 | p->se.nr_migrations++; |
a8b0ca17 | 1024 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0); |
0c69774e | 1025 | } |
dd41f596 IM |
1026 | |
1027 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1028 | } |
1029 | ||
ac66f547 PZ |
1030 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1031 | { | |
1032 | if (p->on_rq) { | |
1033 | struct rq *src_rq, *dst_rq; | |
1034 | ||
1035 | src_rq = task_rq(p); | |
1036 | dst_rq = cpu_rq(cpu); | |
1037 | ||
1038 | deactivate_task(src_rq, p, 0); | |
1039 | set_task_cpu(p, cpu); | |
1040 | activate_task(dst_rq, p, 0); | |
1041 | check_preempt_curr(dst_rq, p, 0); | |
1042 | } else { | |
1043 | /* | |
1044 | * Task isn't running anymore; make it appear like we migrated | |
1045 | * it before it went to sleep. This means on wakeup we make the | |
1046 | * previous cpu our targer instead of where it really is. | |
1047 | */ | |
1048 | p->wake_cpu = cpu; | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | struct migration_swap_arg { | |
1053 | struct task_struct *src_task, *dst_task; | |
1054 | int src_cpu, dst_cpu; | |
1055 | }; | |
1056 | ||
1057 | static int migrate_swap_stop(void *data) | |
1058 | { | |
1059 | struct migration_swap_arg *arg = data; | |
1060 | struct rq *src_rq, *dst_rq; | |
1061 | int ret = -EAGAIN; | |
1062 | ||
1063 | src_rq = cpu_rq(arg->src_cpu); | |
1064 | dst_rq = cpu_rq(arg->dst_cpu); | |
1065 | ||
74602315 PZ |
1066 | double_raw_lock(&arg->src_task->pi_lock, |
1067 | &arg->dst_task->pi_lock); | |
ac66f547 PZ |
1068 | double_rq_lock(src_rq, dst_rq); |
1069 | if (task_cpu(arg->dst_task) != arg->dst_cpu) | |
1070 | goto unlock; | |
1071 | ||
1072 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1073 | goto unlock; | |
1074 | ||
1075 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) | |
1076 | goto unlock; | |
1077 | ||
1078 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) | |
1079 | goto unlock; | |
1080 | ||
1081 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1082 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1083 | ||
1084 | ret = 0; | |
1085 | ||
1086 | unlock: | |
1087 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1088 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1089 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1090 | |
1091 | return ret; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * Cross migrate two tasks | |
1096 | */ | |
1097 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1098 | { | |
1099 | struct migration_swap_arg arg; | |
1100 | int ret = -EINVAL; | |
1101 | ||
ac66f547 PZ |
1102 | arg = (struct migration_swap_arg){ |
1103 | .src_task = cur, | |
1104 | .src_cpu = task_cpu(cur), | |
1105 | .dst_task = p, | |
1106 | .dst_cpu = task_cpu(p), | |
1107 | }; | |
1108 | ||
1109 | if (arg.src_cpu == arg.dst_cpu) | |
1110 | goto out; | |
1111 | ||
6acce3ef PZ |
1112 | /* |
1113 | * These three tests are all lockless; this is OK since all of them | |
1114 | * will be re-checked with proper locks held further down the line. | |
1115 | */ | |
ac66f547 PZ |
1116 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1117 | goto out; | |
1118 | ||
1119 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) | |
1120 | goto out; | |
1121 | ||
1122 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) | |
1123 | goto out; | |
1124 | ||
286549dc | 1125 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1126 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1127 | ||
1128 | out: | |
ac66f547 PZ |
1129 | return ret; |
1130 | } | |
1131 | ||
969c7921 | 1132 | struct migration_arg { |
36c8b586 | 1133 | struct task_struct *task; |
1da177e4 | 1134 | int dest_cpu; |
70b97a7f | 1135 | }; |
1da177e4 | 1136 | |
969c7921 TH |
1137 | static int migration_cpu_stop(void *data); |
1138 | ||
1da177e4 LT |
1139 | /* |
1140 | * wait_task_inactive - wait for a thread to unschedule. | |
1141 | * | |
85ba2d86 RM |
1142 | * If @match_state is nonzero, it's the @p->state value just checked and |
1143 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1144 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1145 | * we return a positive number (its total switch count). If a second call | |
1146 | * a short while later returns the same number, the caller can be sure that | |
1147 | * @p has remained unscheduled the whole time. | |
1148 | * | |
1da177e4 LT |
1149 | * The caller must ensure that the task *will* unschedule sometime soon, |
1150 | * else this function might spin for a *long* time. This function can't | |
1151 | * be called with interrupts off, or it may introduce deadlock with | |
1152 | * smp_call_function() if an IPI is sent by the same process we are | |
1153 | * waiting to become inactive. | |
1154 | */ | |
85ba2d86 | 1155 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1156 | { |
1157 | unsigned long flags; | |
dd41f596 | 1158 | int running, on_rq; |
85ba2d86 | 1159 | unsigned long ncsw; |
70b97a7f | 1160 | struct rq *rq; |
1da177e4 | 1161 | |
3a5c359a AK |
1162 | for (;;) { |
1163 | /* | |
1164 | * We do the initial early heuristics without holding | |
1165 | * any task-queue locks at all. We'll only try to get | |
1166 | * the runqueue lock when things look like they will | |
1167 | * work out! | |
1168 | */ | |
1169 | rq = task_rq(p); | |
fa490cfd | 1170 | |
3a5c359a AK |
1171 | /* |
1172 | * If the task is actively running on another CPU | |
1173 | * still, just relax and busy-wait without holding | |
1174 | * any locks. | |
1175 | * | |
1176 | * NOTE! Since we don't hold any locks, it's not | |
1177 | * even sure that "rq" stays as the right runqueue! | |
1178 | * But we don't care, since "task_running()" will | |
1179 | * return false if the runqueue has changed and p | |
1180 | * is actually now running somewhere else! | |
1181 | */ | |
85ba2d86 RM |
1182 | while (task_running(rq, p)) { |
1183 | if (match_state && unlikely(p->state != match_state)) | |
1184 | return 0; | |
3a5c359a | 1185 | cpu_relax(); |
85ba2d86 | 1186 | } |
fa490cfd | 1187 | |
3a5c359a AK |
1188 | /* |
1189 | * Ok, time to look more closely! We need the rq | |
1190 | * lock now, to be *sure*. If we're wrong, we'll | |
1191 | * just go back and repeat. | |
1192 | */ | |
1193 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 1194 | trace_sched_wait_task(p); |
3a5c359a | 1195 | running = task_running(rq, p); |
fd2f4419 | 1196 | on_rq = p->on_rq; |
85ba2d86 | 1197 | ncsw = 0; |
f31e11d8 | 1198 | if (!match_state || p->state == match_state) |
93dcf55f | 1199 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 1200 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 1201 | |
85ba2d86 RM |
1202 | /* |
1203 | * If it changed from the expected state, bail out now. | |
1204 | */ | |
1205 | if (unlikely(!ncsw)) | |
1206 | break; | |
1207 | ||
3a5c359a AK |
1208 | /* |
1209 | * Was it really running after all now that we | |
1210 | * checked with the proper locks actually held? | |
1211 | * | |
1212 | * Oops. Go back and try again.. | |
1213 | */ | |
1214 | if (unlikely(running)) { | |
1215 | cpu_relax(); | |
1216 | continue; | |
1217 | } | |
fa490cfd | 1218 | |
3a5c359a AK |
1219 | /* |
1220 | * It's not enough that it's not actively running, | |
1221 | * it must be off the runqueue _entirely_, and not | |
1222 | * preempted! | |
1223 | * | |
80dd99b3 | 1224 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1225 | * running right now), it's preempted, and we should |
1226 | * yield - it could be a while. | |
1227 | */ | |
1228 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
1229 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
1230 | ||
1231 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1232 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1233 | continue; |
1234 | } | |
fa490cfd | 1235 | |
3a5c359a AK |
1236 | /* |
1237 | * Ahh, all good. It wasn't running, and it wasn't | |
1238 | * runnable, which means that it will never become | |
1239 | * running in the future either. We're all done! | |
1240 | */ | |
1241 | break; | |
1242 | } | |
85ba2d86 RM |
1243 | |
1244 | return ncsw; | |
1da177e4 LT |
1245 | } |
1246 | ||
1247 | /*** | |
1248 | * kick_process - kick a running thread to enter/exit the kernel | |
1249 | * @p: the to-be-kicked thread | |
1250 | * | |
1251 | * Cause a process which is running on another CPU to enter | |
1252 | * kernel-mode, without any delay. (to get signals handled.) | |
1253 | * | |
25985edc | 1254 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1255 | * because all it wants to ensure is that the remote task enters |
1256 | * the kernel. If the IPI races and the task has been migrated | |
1257 | * to another CPU then no harm is done and the purpose has been | |
1258 | * achieved as well. | |
1259 | */ | |
36c8b586 | 1260 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1261 | { |
1262 | int cpu; | |
1263 | ||
1264 | preempt_disable(); | |
1265 | cpu = task_cpu(p); | |
1266 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1267 | smp_send_reschedule(cpu); | |
1268 | preempt_enable(); | |
1269 | } | |
b43e3521 | 1270 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 1271 | #endif /* CONFIG_SMP */ |
1da177e4 | 1272 | |
970b13ba | 1273 | #ifdef CONFIG_SMP |
30da688e | 1274 | /* |
013fdb80 | 1275 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 1276 | */ |
5da9a0fb PZ |
1277 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1278 | { | |
aa00d89c TC |
1279 | int nid = cpu_to_node(cpu); |
1280 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1281 | enum { cpuset, possible, fail } state = cpuset; |
1282 | int dest_cpu; | |
5da9a0fb | 1283 | |
aa00d89c TC |
1284 | /* |
1285 | * If the node that the cpu is on has been offlined, cpu_to_node() | |
1286 | * will return -1. There is no cpu on the node, and we should | |
1287 | * select the cpu on the other node. | |
1288 | */ | |
1289 | if (nid != -1) { | |
1290 | nodemask = cpumask_of_node(nid); | |
1291 | ||
1292 | /* Look for allowed, online CPU in same node. */ | |
1293 | for_each_cpu(dest_cpu, nodemask) { | |
1294 | if (!cpu_online(dest_cpu)) | |
1295 | continue; | |
1296 | if (!cpu_active(dest_cpu)) | |
1297 | continue; | |
1298 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) | |
1299 | return dest_cpu; | |
1300 | } | |
2baab4e9 | 1301 | } |
5da9a0fb | 1302 | |
2baab4e9 PZ |
1303 | for (;;) { |
1304 | /* Any allowed, online CPU? */ | |
e3831edd | 1305 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
2baab4e9 PZ |
1306 | if (!cpu_online(dest_cpu)) |
1307 | continue; | |
1308 | if (!cpu_active(dest_cpu)) | |
1309 | continue; | |
1310 | goto out; | |
1311 | } | |
5da9a0fb | 1312 | |
2baab4e9 PZ |
1313 | switch (state) { |
1314 | case cpuset: | |
1315 | /* No more Mr. Nice Guy. */ | |
1316 | cpuset_cpus_allowed_fallback(p); | |
1317 | state = possible; | |
1318 | break; | |
1319 | ||
1320 | case possible: | |
1321 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1322 | state = fail; | |
1323 | break; | |
1324 | ||
1325 | case fail: | |
1326 | BUG(); | |
1327 | break; | |
1328 | } | |
1329 | } | |
1330 | ||
1331 | out: | |
1332 | if (state != cpuset) { | |
1333 | /* | |
1334 | * Don't tell them about moving exiting tasks or | |
1335 | * kernel threads (both mm NULL), since they never | |
1336 | * leave kernel. | |
1337 | */ | |
1338 | if (p->mm && printk_ratelimit()) { | |
1339 | printk_sched("process %d (%s) no longer affine to cpu%d\n", | |
1340 | task_pid_nr(p), p->comm, cpu); | |
1341 | } | |
5da9a0fb PZ |
1342 | } |
1343 | ||
1344 | return dest_cpu; | |
1345 | } | |
1346 | ||
e2912009 | 1347 | /* |
013fdb80 | 1348 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1349 | */ |
970b13ba | 1350 | static inline |
ac66f547 | 1351 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1352 | { |
ac66f547 | 1353 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e2912009 PZ |
1354 | |
1355 | /* | |
1356 | * In order not to call set_task_cpu() on a blocking task we need | |
1357 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
1358 | * cpu. | |
1359 | * | |
1360 | * Since this is common to all placement strategies, this lives here. | |
1361 | * | |
1362 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1363 | * not worry about this generic constraint ] | |
1364 | */ | |
fa17b507 | 1365 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
70f11205 | 1366 | !cpu_online(cpu))) |
5da9a0fb | 1367 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1368 | |
1369 | return cpu; | |
970b13ba | 1370 | } |
09a40af5 MG |
1371 | |
1372 | static void update_avg(u64 *avg, u64 sample) | |
1373 | { | |
1374 | s64 diff = sample - *avg; | |
1375 | *avg += diff >> 3; | |
1376 | } | |
970b13ba PZ |
1377 | #endif |
1378 | ||
d7c01d27 | 1379 | static void |
b84cb5df | 1380 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1381 | { |
d7c01d27 | 1382 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
1383 | struct rq *rq = this_rq(); |
1384 | ||
d7c01d27 PZ |
1385 | #ifdef CONFIG_SMP |
1386 | int this_cpu = smp_processor_id(); | |
1387 | ||
1388 | if (cpu == this_cpu) { | |
1389 | schedstat_inc(rq, ttwu_local); | |
1390 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
1391 | } else { | |
1392 | struct sched_domain *sd; | |
1393 | ||
1394 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 1395 | rcu_read_lock(); |
d7c01d27 PZ |
1396 | for_each_domain(this_cpu, sd) { |
1397 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
1398 | schedstat_inc(sd, ttwu_wake_remote); | |
1399 | break; | |
1400 | } | |
1401 | } | |
057f3fad | 1402 | rcu_read_unlock(); |
d7c01d27 | 1403 | } |
f339b9dc PZ |
1404 | |
1405 | if (wake_flags & WF_MIGRATED) | |
1406 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
1407 | ||
d7c01d27 PZ |
1408 | #endif /* CONFIG_SMP */ |
1409 | ||
1410 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 1411 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
1412 | |
1413 | if (wake_flags & WF_SYNC) | |
9ed3811a | 1414 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 | 1415 | |
d7c01d27 PZ |
1416 | #endif /* CONFIG_SCHEDSTATS */ |
1417 | } | |
1418 | ||
1419 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
1420 | { | |
9ed3811a | 1421 | activate_task(rq, p, en_flags); |
fd2f4419 | 1422 | p->on_rq = 1; |
c2f7115e PZ |
1423 | |
1424 | /* if a worker is waking up, notify workqueue */ | |
1425 | if (p->flags & PF_WQ_WORKER) | |
1426 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1427 | } |
1428 | ||
23f41eeb PZ |
1429 | /* |
1430 | * Mark the task runnable and perform wakeup-preemption. | |
1431 | */ | |
89363381 | 1432 | static void |
23f41eeb | 1433 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 1434 | { |
9ed3811a | 1435 | check_preempt_curr(rq, p, wake_flags); |
a8d7ad52 | 1436 | trace_sched_wakeup(p, true); |
9ed3811a TH |
1437 | |
1438 | p->state = TASK_RUNNING; | |
1439 | #ifdef CONFIG_SMP | |
1440 | if (p->sched_class->task_woken) | |
1441 | p->sched_class->task_woken(rq, p); | |
1442 | ||
e69c6341 | 1443 | if (rq->idle_stamp) { |
78becc27 | 1444 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1445 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1446 | |
abfafa54 JL |
1447 | update_avg(&rq->avg_idle, delta); |
1448 | ||
1449 | if (rq->avg_idle > max) | |
9ed3811a | 1450 | rq->avg_idle = max; |
abfafa54 | 1451 | |
9ed3811a TH |
1452 | rq->idle_stamp = 0; |
1453 | } | |
1454 | #endif | |
1455 | } | |
1456 | ||
c05fbafb PZ |
1457 | static void |
1458 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
1459 | { | |
1460 | #ifdef CONFIG_SMP | |
1461 | if (p->sched_contributes_to_load) | |
1462 | rq->nr_uninterruptible--; | |
1463 | #endif | |
1464 | ||
1465 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
1466 | ttwu_do_wakeup(rq, p, wake_flags); | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1471 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1472 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1473 | * the task is still ->on_rq. | |
1474 | */ | |
1475 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1476 | { | |
1477 | struct rq *rq; | |
1478 | int ret = 0; | |
1479 | ||
1480 | rq = __task_rq_lock(p); | |
1481 | if (p->on_rq) { | |
1ad4ec0d FW |
1482 | /* check_preempt_curr() may use rq clock */ |
1483 | update_rq_clock(rq); | |
c05fbafb PZ |
1484 | ttwu_do_wakeup(rq, p, wake_flags); |
1485 | ret = 1; | |
1486 | } | |
1487 | __task_rq_unlock(rq); | |
1488 | ||
1489 | return ret; | |
1490 | } | |
1491 | ||
317f3941 | 1492 | #ifdef CONFIG_SMP |
fa14ff4a | 1493 | static void sched_ttwu_pending(void) |
317f3941 PZ |
1494 | { |
1495 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1496 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1497 | struct task_struct *p; | |
317f3941 PZ |
1498 | |
1499 | raw_spin_lock(&rq->lock); | |
1500 | ||
fa14ff4a PZ |
1501 | while (llist) { |
1502 | p = llist_entry(llist, struct task_struct, wake_entry); | |
1503 | llist = llist_next(llist); | |
317f3941 PZ |
1504 | ttwu_do_activate(rq, p, 0); |
1505 | } | |
1506 | ||
1507 | raw_spin_unlock(&rq->lock); | |
1508 | } | |
1509 | ||
1510 | void scheduler_ipi(void) | |
1511 | { | |
f27dde8d PZ |
1512 | /* |
1513 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1514 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1515 | * this IPI. | |
1516 | */ | |
8cb75e0c | 1517 | preempt_fold_need_resched(); |
f27dde8d | 1518 | |
873b4c65 VG |
1519 | if (llist_empty(&this_rq()->wake_list) |
1520 | && !tick_nohz_full_cpu(smp_processor_id()) | |
1521 | && !got_nohz_idle_kick()) | |
c5d753a5 PZ |
1522 | return; |
1523 | ||
1524 | /* | |
1525 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1526 | * traditionally all their work was done from the interrupt return | |
1527 | * path. Now that we actually do some work, we need to make sure | |
1528 | * we do call them. | |
1529 | * | |
1530 | * Some archs already do call them, luckily irq_enter/exit nest | |
1531 | * properly. | |
1532 | * | |
1533 | * Arguably we should visit all archs and update all handlers, | |
1534 | * however a fair share of IPIs are still resched only so this would | |
1535 | * somewhat pessimize the simple resched case. | |
1536 | */ | |
1537 | irq_enter(); | |
ff442c51 | 1538 | tick_nohz_full_check(); |
fa14ff4a | 1539 | sched_ttwu_pending(); |
ca38062e SS |
1540 | |
1541 | /* | |
1542 | * Check if someone kicked us for doing the nohz idle load balance. | |
1543 | */ | |
873b4c65 | 1544 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1545 | this_rq()->idle_balance = 1; |
ca38062e | 1546 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1547 | } |
c5d753a5 | 1548 | irq_exit(); |
317f3941 PZ |
1549 | } |
1550 | ||
1551 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
1552 | { | |
fa14ff4a | 1553 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) |
317f3941 PZ |
1554 | smp_send_reschedule(cpu); |
1555 | } | |
d6aa8f85 | 1556 | |
39be3501 | 1557 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1558 | { |
1559 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1560 | } | |
d6aa8f85 | 1561 | #endif /* CONFIG_SMP */ |
317f3941 | 1562 | |
c05fbafb PZ |
1563 | static void ttwu_queue(struct task_struct *p, int cpu) |
1564 | { | |
1565 | struct rq *rq = cpu_rq(cpu); | |
1566 | ||
17d9f311 | 1567 | #if defined(CONFIG_SMP) |
39be3501 | 1568 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
f01114cb | 1569 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
1570 | ttwu_queue_remote(p, cpu); |
1571 | return; | |
1572 | } | |
1573 | #endif | |
1574 | ||
c05fbafb PZ |
1575 | raw_spin_lock(&rq->lock); |
1576 | ttwu_do_activate(rq, p, 0); | |
1577 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
1578 | } |
1579 | ||
1580 | /** | |
1da177e4 | 1581 | * try_to_wake_up - wake up a thread |
9ed3811a | 1582 | * @p: the thread to be awakened |
1da177e4 | 1583 | * @state: the mask of task states that can be woken |
9ed3811a | 1584 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
1585 | * |
1586 | * Put it on the run-queue if it's not already there. The "current" | |
1587 | * thread is always on the run-queue (except when the actual | |
1588 | * re-schedule is in progress), and as such you're allowed to do | |
1589 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1590 | * runnable without the overhead of this. | |
1591 | * | |
e69f6186 | 1592 | * Return: %true if @p was woken up, %false if it was already running. |
9ed3811a | 1593 | * or @state didn't match @p's state. |
1da177e4 | 1594 | */ |
e4a52bcb PZ |
1595 | static int |
1596 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1597 | { |
1da177e4 | 1598 | unsigned long flags; |
c05fbafb | 1599 | int cpu, success = 0; |
2398f2c6 | 1600 | |
e0acd0a6 ON |
1601 | /* |
1602 | * If we are going to wake up a thread waiting for CONDITION we | |
1603 | * need to ensure that CONDITION=1 done by the caller can not be | |
1604 | * reordered with p->state check below. This pairs with mb() in | |
1605 | * set_current_state() the waiting thread does. | |
1606 | */ | |
1607 | smp_mb__before_spinlock(); | |
013fdb80 | 1608 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1609 | if (!(p->state & state)) |
1da177e4 LT |
1610 | goto out; |
1611 | ||
c05fbafb | 1612 | success = 1; /* we're going to change ->state */ |
1da177e4 | 1613 | cpu = task_cpu(p); |
1da177e4 | 1614 | |
c05fbafb PZ |
1615 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1616 | goto stat; | |
1da177e4 | 1617 | |
1da177e4 | 1618 | #ifdef CONFIG_SMP |
e9c84311 | 1619 | /* |
c05fbafb PZ |
1620 | * If the owning (remote) cpu is still in the middle of schedule() with |
1621 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 1622 | */ |
f3e94786 | 1623 | while (p->on_cpu) |
e4a52bcb | 1624 | cpu_relax(); |
0970d299 | 1625 | /* |
e4a52bcb | 1626 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 1627 | */ |
e4a52bcb | 1628 | smp_rmb(); |
1da177e4 | 1629 | |
a8e4f2ea | 1630 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 1631 | p->state = TASK_WAKING; |
e7693a36 | 1632 | |
e4a52bcb | 1633 | if (p->sched_class->task_waking) |
74f8e4b2 | 1634 | p->sched_class->task_waking(p); |
efbbd05a | 1635 | |
ac66f547 | 1636 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
1637 | if (task_cpu(p) != cpu) { |
1638 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 1639 | set_task_cpu(p, cpu); |
f339b9dc | 1640 | } |
1da177e4 | 1641 | #endif /* CONFIG_SMP */ |
1da177e4 | 1642 | |
c05fbafb PZ |
1643 | ttwu_queue(p, cpu); |
1644 | stat: | |
b84cb5df | 1645 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 1646 | out: |
013fdb80 | 1647 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
1648 | |
1649 | return success; | |
1650 | } | |
1651 | ||
21aa9af0 TH |
1652 | /** |
1653 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
1654 | * @p: the thread to be awakened | |
1655 | * | |
2acca55e | 1656 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 1657 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 1658 | * the current task. |
21aa9af0 TH |
1659 | */ |
1660 | static void try_to_wake_up_local(struct task_struct *p) | |
1661 | { | |
1662 | struct rq *rq = task_rq(p); | |
21aa9af0 | 1663 | |
383efcd0 TH |
1664 | if (WARN_ON_ONCE(rq != this_rq()) || |
1665 | WARN_ON_ONCE(p == current)) | |
1666 | return; | |
1667 | ||
21aa9af0 TH |
1668 | lockdep_assert_held(&rq->lock); |
1669 | ||
2acca55e PZ |
1670 | if (!raw_spin_trylock(&p->pi_lock)) { |
1671 | raw_spin_unlock(&rq->lock); | |
1672 | raw_spin_lock(&p->pi_lock); | |
1673 | raw_spin_lock(&rq->lock); | |
1674 | } | |
1675 | ||
21aa9af0 | 1676 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 1677 | goto out; |
21aa9af0 | 1678 | |
fd2f4419 | 1679 | if (!p->on_rq) |
d7c01d27 PZ |
1680 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
1681 | ||
23f41eeb | 1682 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 1683 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
1684 | out: |
1685 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
1686 | } |
1687 | ||
50fa610a DH |
1688 | /** |
1689 | * wake_up_process - Wake up a specific process | |
1690 | * @p: The process to be woken up. | |
1691 | * | |
1692 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
1693 | * processes. |
1694 | * | |
1695 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
1696 | * |
1697 | * It may be assumed that this function implies a write memory barrier before | |
1698 | * changing the task state if and only if any tasks are woken up. | |
1699 | */ | |
7ad5b3a5 | 1700 | int wake_up_process(struct task_struct *p) |
1da177e4 | 1701 | { |
9067ac85 ON |
1702 | WARN_ON(task_is_stopped_or_traced(p)); |
1703 | return try_to_wake_up(p, TASK_NORMAL, 0); | |
1da177e4 | 1704 | } |
1da177e4 LT |
1705 | EXPORT_SYMBOL(wake_up_process); |
1706 | ||
7ad5b3a5 | 1707 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1708 | { |
1709 | return try_to_wake_up(p, state, 0); | |
1710 | } | |
1711 | ||
1da177e4 LT |
1712 | /* |
1713 | * Perform scheduler related setup for a newly forked process p. | |
1714 | * p is forked by current. | |
dd41f596 IM |
1715 | * |
1716 | * __sched_fork() is basic setup used by init_idle() too: | |
1717 | */ | |
5e1576ed | 1718 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1719 | { |
fd2f4419 PZ |
1720 | p->on_rq = 0; |
1721 | ||
1722 | p->se.on_rq = 0; | |
dd41f596 IM |
1723 | p->se.exec_start = 0; |
1724 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1725 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 1726 | p->se.nr_migrations = 0; |
da7a735e | 1727 | p->se.vruntime = 0; |
fd2f4419 | 1728 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
1729 | |
1730 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 1731 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 1732 | #endif |
476d139c | 1733 | |
aab03e05 DF |
1734 | RB_CLEAR_NODE(&p->dl.rb_node); |
1735 | hrtimer_init(&p->dl.dl_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1736 | p->dl.dl_runtime = p->dl.runtime = 0; | |
1737 | p->dl.dl_deadline = p->dl.deadline = 0; | |
755378a4 | 1738 | p->dl.dl_period = 0; |
aab03e05 DF |
1739 | p->dl.flags = 0; |
1740 | ||
fa717060 | 1741 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 1742 | |
e107be36 AK |
1743 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1744 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1745 | #endif | |
cbee9f88 PZ |
1746 | |
1747 | #ifdef CONFIG_NUMA_BALANCING | |
1748 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 1749 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
1750 | p->mm->numa_scan_seq = 0; |
1751 | } | |
1752 | ||
5e1576ed RR |
1753 | if (clone_flags & CLONE_VM) |
1754 | p->numa_preferred_nid = current->numa_preferred_nid; | |
1755 | else | |
1756 | p->numa_preferred_nid = -1; | |
1757 | ||
cbee9f88 PZ |
1758 | p->node_stamp = 0ULL; |
1759 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 1760 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 1761 | p->numa_work.next = &p->numa_work; |
ff1df896 RR |
1762 | p->numa_faults_memory = NULL; |
1763 | p->numa_faults_buffer_memory = NULL; | |
7e2703e6 RR |
1764 | p->last_task_numa_placement = 0; |
1765 | p->last_sum_exec_runtime = 0; | |
8c8a743c PZ |
1766 | |
1767 | INIT_LIST_HEAD(&p->numa_entry); | |
1768 | p->numa_group = NULL; | |
cbee9f88 | 1769 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
1770 | } |
1771 | ||
1a687c2e | 1772 | #ifdef CONFIG_NUMA_BALANCING |
3105b86a | 1773 | #ifdef CONFIG_SCHED_DEBUG |
1a687c2e MG |
1774 | void set_numabalancing_state(bool enabled) |
1775 | { | |
1776 | if (enabled) | |
1777 | sched_feat_set("NUMA"); | |
1778 | else | |
1779 | sched_feat_set("NO_NUMA"); | |
1780 | } | |
3105b86a MG |
1781 | #else |
1782 | __read_mostly bool numabalancing_enabled; | |
1783 | ||
1784 | void set_numabalancing_state(bool enabled) | |
1785 | { | |
1786 | numabalancing_enabled = enabled; | |
dd41f596 | 1787 | } |
3105b86a | 1788 | #endif /* CONFIG_SCHED_DEBUG */ |
54a43d54 AK |
1789 | |
1790 | #ifdef CONFIG_PROC_SYSCTL | |
1791 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
1792 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1793 | { | |
1794 | struct ctl_table t; | |
1795 | int err; | |
1796 | int state = numabalancing_enabled; | |
1797 | ||
1798 | if (write && !capable(CAP_SYS_ADMIN)) | |
1799 | return -EPERM; | |
1800 | ||
1801 | t = *table; | |
1802 | t.data = &state; | |
1803 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
1804 | if (err < 0) | |
1805 | return err; | |
1806 | if (write) | |
1807 | set_numabalancing_state(state); | |
1808 | return err; | |
1809 | } | |
1810 | #endif | |
1811 | #endif | |
dd41f596 IM |
1812 | |
1813 | /* | |
1814 | * fork()/clone()-time setup: | |
1815 | */ | |
aab03e05 | 1816 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 1817 | { |
0122ec5b | 1818 | unsigned long flags; |
dd41f596 IM |
1819 | int cpu = get_cpu(); |
1820 | ||
5e1576ed | 1821 | __sched_fork(clone_flags, p); |
06b83b5f | 1822 | /* |
0017d735 | 1823 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
1824 | * nobody will actually run it, and a signal or other external |
1825 | * event cannot wake it up and insert it on the runqueue either. | |
1826 | */ | |
0017d735 | 1827 | p->state = TASK_RUNNING; |
dd41f596 | 1828 | |
c350a04e MG |
1829 | /* |
1830 | * Make sure we do not leak PI boosting priority to the child. | |
1831 | */ | |
1832 | p->prio = current->normal_prio; | |
1833 | ||
b9dc29e7 MG |
1834 | /* |
1835 | * Revert to default priority/policy on fork if requested. | |
1836 | */ | |
1837 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 1838 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 1839 | p->policy = SCHED_NORMAL; |
6c697bdf | 1840 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
1841 | p->rt_priority = 0; |
1842 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
1843 | p->static_prio = NICE_TO_PRIO(0); | |
1844 | ||
1845 | p->prio = p->normal_prio = __normal_prio(p); | |
1846 | set_load_weight(p); | |
6c697bdf | 1847 | |
b9dc29e7 MG |
1848 | /* |
1849 | * We don't need the reset flag anymore after the fork. It has | |
1850 | * fulfilled its duty: | |
1851 | */ | |
1852 | p->sched_reset_on_fork = 0; | |
1853 | } | |
ca94c442 | 1854 | |
aab03e05 DF |
1855 | if (dl_prio(p->prio)) { |
1856 | put_cpu(); | |
1857 | return -EAGAIN; | |
1858 | } else if (rt_prio(p->prio)) { | |
1859 | p->sched_class = &rt_sched_class; | |
1860 | } else { | |
2ddbf952 | 1861 | p->sched_class = &fair_sched_class; |
aab03e05 | 1862 | } |
b29739f9 | 1863 | |
cd29fe6f PZ |
1864 | if (p->sched_class->task_fork) |
1865 | p->sched_class->task_fork(p); | |
1866 | ||
86951599 PZ |
1867 | /* |
1868 | * The child is not yet in the pid-hash so no cgroup attach races, | |
1869 | * and the cgroup is pinned to this child due to cgroup_fork() | |
1870 | * is ran before sched_fork(). | |
1871 | * | |
1872 | * Silence PROVE_RCU. | |
1873 | */ | |
0122ec5b | 1874 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 1875 | set_task_cpu(p, cpu); |
0122ec5b | 1876 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 1877 | |
52f17b6c | 1878 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1879 | if (likely(sched_info_on())) |
52f17b6c | 1880 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1881 | #endif |
3ca7a440 PZ |
1882 | #if defined(CONFIG_SMP) |
1883 | p->on_cpu = 0; | |
4866cde0 | 1884 | #endif |
01028747 | 1885 | init_task_preempt_count(p); |
806c09a7 | 1886 | #ifdef CONFIG_SMP |
917b627d | 1887 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 1888 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 1889 | #endif |
917b627d | 1890 | |
476d139c | 1891 | put_cpu(); |
aab03e05 | 1892 | return 0; |
1da177e4 LT |
1893 | } |
1894 | ||
332ac17e DF |
1895 | unsigned long to_ratio(u64 period, u64 runtime) |
1896 | { | |
1897 | if (runtime == RUNTIME_INF) | |
1898 | return 1ULL << 20; | |
1899 | ||
1900 | /* | |
1901 | * Doing this here saves a lot of checks in all | |
1902 | * the calling paths, and returning zero seems | |
1903 | * safe for them anyway. | |
1904 | */ | |
1905 | if (period == 0) | |
1906 | return 0; | |
1907 | ||
1908 | return div64_u64(runtime << 20, period); | |
1909 | } | |
1910 | ||
1911 | #ifdef CONFIG_SMP | |
1912 | inline struct dl_bw *dl_bw_of(int i) | |
1913 | { | |
1914 | return &cpu_rq(i)->rd->dl_bw; | |
1915 | } | |
1916 | ||
de212f18 | 1917 | static inline int dl_bw_cpus(int i) |
332ac17e | 1918 | { |
de212f18 PZ |
1919 | struct root_domain *rd = cpu_rq(i)->rd; |
1920 | int cpus = 0; | |
1921 | ||
1922 | for_each_cpu_and(i, rd->span, cpu_active_mask) | |
1923 | cpus++; | |
1924 | ||
1925 | return cpus; | |
332ac17e DF |
1926 | } |
1927 | #else | |
1928 | inline struct dl_bw *dl_bw_of(int i) | |
1929 | { | |
1930 | return &cpu_rq(i)->dl.dl_bw; | |
1931 | } | |
1932 | ||
de212f18 | 1933 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
1934 | { |
1935 | return 1; | |
1936 | } | |
1937 | #endif | |
1938 | ||
1939 | static inline | |
1940 | void __dl_clear(struct dl_bw *dl_b, u64 tsk_bw) | |
1941 | { | |
1942 | dl_b->total_bw -= tsk_bw; | |
1943 | } | |
1944 | ||
1945 | static inline | |
1946 | void __dl_add(struct dl_bw *dl_b, u64 tsk_bw) | |
1947 | { | |
1948 | dl_b->total_bw += tsk_bw; | |
1949 | } | |
1950 | ||
1951 | static inline | |
1952 | bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) | |
1953 | { | |
1954 | return dl_b->bw != -1 && | |
1955 | dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; | |
1956 | } | |
1957 | ||
1958 | /* | |
1959 | * We must be sure that accepting a new task (or allowing changing the | |
1960 | * parameters of an existing one) is consistent with the bandwidth | |
1961 | * constraints. If yes, this function also accordingly updates the currently | |
1962 | * allocated bandwidth to reflect the new situation. | |
1963 | * | |
1964 | * This function is called while holding p's rq->lock. | |
1965 | */ | |
1966 | static int dl_overflow(struct task_struct *p, int policy, | |
1967 | const struct sched_attr *attr) | |
1968 | { | |
1969 | ||
1970 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 1971 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
1972 | u64 runtime = attr->sched_runtime; |
1973 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 1974 | int cpus, err = -1; |
332ac17e DF |
1975 | |
1976 | if (new_bw == p->dl.dl_bw) | |
1977 | return 0; | |
1978 | ||
1979 | /* | |
1980 | * Either if a task, enters, leave, or stays -deadline but changes | |
1981 | * its parameters, we may need to update accordingly the total | |
1982 | * allocated bandwidth of the container. | |
1983 | */ | |
1984 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 1985 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
1986 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
1987 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
1988 | __dl_add(dl_b, new_bw); | |
1989 | err = 0; | |
1990 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
1991 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
1992 | __dl_clear(dl_b, p->dl.dl_bw); | |
1993 | __dl_add(dl_b, new_bw); | |
1994 | err = 0; | |
1995 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
1996 | __dl_clear(dl_b, p->dl.dl_bw); | |
1997 | err = 0; | |
1998 | } | |
1999 | raw_spin_unlock(&dl_b->lock); | |
2000 | ||
2001 | return err; | |
2002 | } | |
2003 | ||
2004 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2005 | ||
1da177e4 LT |
2006 | /* |
2007 | * wake_up_new_task - wake up a newly created task for the first time. | |
2008 | * | |
2009 | * This function will do some initial scheduler statistics housekeeping | |
2010 | * that must be done for every newly created context, then puts the task | |
2011 | * on the runqueue and wakes it. | |
2012 | */ | |
3e51e3ed | 2013 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2014 | { |
2015 | unsigned long flags; | |
dd41f596 | 2016 | struct rq *rq; |
fabf318e | 2017 | |
ab2515c4 | 2018 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2019 | #ifdef CONFIG_SMP |
2020 | /* | |
2021 | * Fork balancing, do it here and not earlier because: | |
2022 | * - cpus_allowed can change in the fork path | |
2023 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2024 | */ |
ac66f547 | 2025 | set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2026 | #endif |
2027 | ||
a75cdaa9 AS |
2028 | /* Initialize new task's runnable average */ |
2029 | init_task_runnable_average(p); | |
ab2515c4 | 2030 | rq = __task_rq_lock(p); |
cd29fe6f | 2031 | activate_task(rq, p, 0); |
fd2f4419 | 2032 | p->on_rq = 1; |
89363381 | 2033 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2034 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2035 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2036 | if (p->sched_class->task_woken) |
2037 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2038 | #endif |
0122ec5b | 2039 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2040 | } |
2041 | ||
e107be36 AK |
2042 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2043 | ||
2044 | /** | |
80dd99b3 | 2045 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2046 | * @notifier: notifier struct to register |
e107be36 AK |
2047 | */ |
2048 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2049 | { | |
2050 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2051 | } | |
2052 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2053 | ||
2054 | /** | |
2055 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2056 | * @notifier: notifier struct to unregister |
e107be36 AK |
2057 | * |
2058 | * This is safe to call from within a preemption notifier. | |
2059 | */ | |
2060 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2061 | { | |
2062 | hlist_del(¬ifier->link); | |
2063 | } | |
2064 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2065 | ||
2066 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2067 | { | |
2068 | struct preempt_notifier *notifier; | |
e107be36 | 2069 | |
b67bfe0d | 2070 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2071 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2072 | } | |
2073 | ||
2074 | static void | |
2075 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2076 | struct task_struct *next) | |
2077 | { | |
2078 | struct preempt_notifier *notifier; | |
e107be36 | 2079 | |
b67bfe0d | 2080 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2081 | notifier->ops->sched_out(notifier, next); |
2082 | } | |
2083 | ||
6d6bc0ad | 2084 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2085 | |
2086 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2087 | { | |
2088 | } | |
2089 | ||
2090 | static void | |
2091 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2092 | struct task_struct *next) | |
2093 | { | |
2094 | } | |
2095 | ||
6d6bc0ad | 2096 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2097 | |
4866cde0 NP |
2098 | /** |
2099 | * prepare_task_switch - prepare to switch tasks | |
2100 | * @rq: the runqueue preparing to switch | |
421cee29 | 2101 | * @prev: the current task that is being switched out |
4866cde0 NP |
2102 | * @next: the task we are going to switch to. |
2103 | * | |
2104 | * This is called with the rq lock held and interrupts off. It must | |
2105 | * be paired with a subsequent finish_task_switch after the context | |
2106 | * switch. | |
2107 | * | |
2108 | * prepare_task_switch sets up locking and calls architecture specific | |
2109 | * hooks. | |
2110 | */ | |
e107be36 AK |
2111 | static inline void |
2112 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2113 | struct task_struct *next) | |
4866cde0 | 2114 | { |
895dd92c | 2115 | trace_sched_switch(prev, next); |
43148951 | 2116 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2117 | perf_event_task_sched_out(prev, next); |
e107be36 | 2118 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2119 | prepare_lock_switch(rq, next); |
2120 | prepare_arch_switch(next); | |
2121 | } | |
2122 | ||
1da177e4 LT |
2123 | /** |
2124 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2125 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2126 | * @prev: the thread we just switched away from. |
2127 | * | |
4866cde0 NP |
2128 | * finish_task_switch must be called after the context switch, paired |
2129 | * with a prepare_task_switch call before the context switch. | |
2130 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2131 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2132 | * |
2133 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2134 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2135 | * with the lock held can cause deadlocks; see schedule() for |
2136 | * details.) | |
2137 | */ | |
a9957449 | 2138 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2139 | __releases(rq->lock) |
2140 | { | |
1da177e4 | 2141 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2142 | long prev_state; |
1da177e4 LT |
2143 | |
2144 | rq->prev_mm = NULL; | |
2145 | ||
2146 | /* | |
2147 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2148 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2149 | * schedule one last time. The schedule call will never return, and |
2150 | * the scheduled task must drop that reference. | |
c394cc9f | 2151 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2152 | * still held, otherwise prev could be scheduled on another cpu, die |
2153 | * there before we look at prev->state, and then the reference would | |
2154 | * be dropped twice. | |
2155 | * Manfred Spraul <manfred@colorfullife.com> | |
2156 | */ | |
55a101f8 | 2157 | prev_state = prev->state; |
bf9fae9f | 2158 | vtime_task_switch(prev); |
4866cde0 | 2159 | finish_arch_switch(prev); |
a8d757ef | 2160 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2161 | finish_lock_switch(rq, prev); |
01f23e16 | 2162 | finish_arch_post_lock_switch(); |
e8fa1362 | 2163 | |
e107be36 | 2164 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2165 | if (mm) |
2166 | mmdrop(mm); | |
c394cc9f | 2167 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2168 | if (prev->sched_class->task_dead) |
2169 | prev->sched_class->task_dead(prev); | |
2170 | ||
c6fd91f0 | 2171 | /* |
2172 | * Remove function-return probe instances associated with this | |
2173 | * task and put them back on the free list. | |
9761eea8 | 2174 | */ |
c6fd91f0 | 2175 | kprobe_flush_task(prev); |
1da177e4 | 2176 | put_task_struct(prev); |
c6fd91f0 | 2177 | } |
99e5ada9 FW |
2178 | |
2179 | tick_nohz_task_switch(current); | |
1da177e4 LT |
2180 | } |
2181 | ||
3f029d3c GH |
2182 | #ifdef CONFIG_SMP |
2183 | ||
3f029d3c GH |
2184 | /* rq->lock is NOT held, but preemption is disabled */ |
2185 | static inline void post_schedule(struct rq *rq) | |
2186 | { | |
2187 | if (rq->post_schedule) { | |
2188 | unsigned long flags; | |
2189 | ||
05fa785c | 2190 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2191 | if (rq->curr->sched_class->post_schedule) |
2192 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2193 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2194 | |
2195 | rq->post_schedule = 0; | |
2196 | } | |
2197 | } | |
2198 | ||
2199 | #else | |
da19ab51 | 2200 | |
3f029d3c GH |
2201 | static inline void post_schedule(struct rq *rq) |
2202 | { | |
1da177e4 LT |
2203 | } |
2204 | ||
3f029d3c GH |
2205 | #endif |
2206 | ||
1da177e4 LT |
2207 | /** |
2208 | * schedule_tail - first thing a freshly forked thread must call. | |
2209 | * @prev: the thread we just switched away from. | |
2210 | */ | |
722a9f92 | 2211 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2212 | __releases(rq->lock) |
2213 | { | |
70b97a7f IM |
2214 | struct rq *rq = this_rq(); |
2215 | ||
4866cde0 | 2216 | finish_task_switch(rq, prev); |
da19ab51 | 2217 | |
3f029d3c GH |
2218 | /* |
2219 | * FIXME: do we need to worry about rq being invalidated by the | |
2220 | * task_switch? | |
2221 | */ | |
2222 | post_schedule(rq); | |
70b97a7f | 2223 | |
4866cde0 NP |
2224 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2225 | /* In this case, finish_task_switch does not reenable preemption */ | |
2226 | preempt_enable(); | |
2227 | #endif | |
1da177e4 | 2228 | if (current->set_child_tid) |
b488893a | 2229 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2230 | } |
2231 | ||
2232 | /* | |
2233 | * context_switch - switch to the new MM and the new | |
2234 | * thread's register state. | |
2235 | */ | |
dd41f596 | 2236 | static inline void |
70b97a7f | 2237 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2238 | struct task_struct *next) |
1da177e4 | 2239 | { |
dd41f596 | 2240 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2241 | |
e107be36 | 2242 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2243 | |
dd41f596 IM |
2244 | mm = next->mm; |
2245 | oldmm = prev->active_mm; | |
9226d125 ZA |
2246 | /* |
2247 | * For paravirt, this is coupled with an exit in switch_to to | |
2248 | * combine the page table reload and the switch backend into | |
2249 | * one hypercall. | |
2250 | */ | |
224101ed | 2251 | arch_start_context_switch(prev); |
9226d125 | 2252 | |
31915ab4 | 2253 | if (!mm) { |
1da177e4 LT |
2254 | next->active_mm = oldmm; |
2255 | atomic_inc(&oldmm->mm_count); | |
2256 | enter_lazy_tlb(oldmm, next); | |
2257 | } else | |
2258 | switch_mm(oldmm, mm, next); | |
2259 | ||
31915ab4 | 2260 | if (!prev->mm) { |
1da177e4 | 2261 | prev->active_mm = NULL; |
1da177e4 LT |
2262 | rq->prev_mm = oldmm; |
2263 | } | |
3a5f5e48 IM |
2264 | /* |
2265 | * Since the runqueue lock will be released by the next | |
2266 | * task (which is an invalid locking op but in the case | |
2267 | * of the scheduler it's an obvious special-case), so we | |
2268 | * do an early lockdep release here: | |
2269 | */ | |
2270 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2271 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2272 | #endif |
1da177e4 | 2273 | |
91d1aa43 | 2274 | context_tracking_task_switch(prev, next); |
1da177e4 LT |
2275 | /* Here we just switch the register state and the stack. */ |
2276 | switch_to(prev, next, prev); | |
2277 | ||
dd41f596 IM |
2278 | barrier(); |
2279 | /* | |
2280 | * this_rq must be evaluated again because prev may have moved | |
2281 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2282 | * frame will be invalid. | |
2283 | */ | |
2284 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2285 | } |
2286 | ||
2287 | /* | |
1c3e8264 | 2288 | * nr_running and nr_context_switches: |
1da177e4 LT |
2289 | * |
2290 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2291 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2292 | */ |
2293 | unsigned long nr_running(void) | |
2294 | { | |
2295 | unsigned long i, sum = 0; | |
2296 | ||
2297 | for_each_online_cpu(i) | |
2298 | sum += cpu_rq(i)->nr_running; | |
2299 | ||
2300 | return sum; | |
f711f609 | 2301 | } |
1da177e4 | 2302 | |
1da177e4 | 2303 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2304 | { |
cc94abfc SR |
2305 | int i; |
2306 | unsigned long long sum = 0; | |
46cb4b7c | 2307 | |
0a945022 | 2308 | for_each_possible_cpu(i) |
1da177e4 | 2309 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2310 | |
1da177e4 LT |
2311 | return sum; |
2312 | } | |
483b4ee6 | 2313 | |
1da177e4 LT |
2314 | unsigned long nr_iowait(void) |
2315 | { | |
2316 | unsigned long i, sum = 0; | |
483b4ee6 | 2317 | |
0a945022 | 2318 | for_each_possible_cpu(i) |
1da177e4 | 2319 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2320 | |
1da177e4 LT |
2321 | return sum; |
2322 | } | |
483b4ee6 | 2323 | |
8c215bd3 | 2324 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2325 | { |
8c215bd3 | 2326 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2327 | return atomic_read(&this->nr_iowait); |
2328 | } | |
46cb4b7c | 2329 | |
dd41f596 | 2330 | #ifdef CONFIG_SMP |
8a0be9ef | 2331 | |
46cb4b7c | 2332 | /* |
38022906 PZ |
2333 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2334 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2335 | */ |
38022906 | 2336 | void sched_exec(void) |
46cb4b7c | 2337 | { |
38022906 | 2338 | struct task_struct *p = current; |
1da177e4 | 2339 | unsigned long flags; |
0017d735 | 2340 | int dest_cpu; |
46cb4b7c | 2341 | |
8f42ced9 | 2342 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2343 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2344 | if (dest_cpu == smp_processor_id()) |
2345 | goto unlock; | |
38022906 | 2346 | |
8f42ced9 | 2347 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2348 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2349 | |
8f42ced9 PZ |
2350 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2351 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2352 | return; |
2353 | } | |
0017d735 | 2354 | unlock: |
8f42ced9 | 2355 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2356 | } |
dd41f596 | 2357 | |
1da177e4 LT |
2358 | #endif |
2359 | ||
1da177e4 | 2360 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2361 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2362 | |
2363 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2364 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 LT |
2365 | |
2366 | /* | |
c5f8d995 | 2367 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 2368 | * @p in case that task is currently running. |
c5f8d995 HS |
2369 | * |
2370 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 2371 | */ |
c5f8d995 HS |
2372 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
2373 | { | |
2374 | u64 ns = 0; | |
2375 | ||
2376 | if (task_current(rq, p)) { | |
2377 | update_rq_clock(rq); | |
78becc27 | 2378 | ns = rq_clock_task(rq) - p->se.exec_start; |
c5f8d995 HS |
2379 | if ((s64)ns < 0) |
2380 | ns = 0; | |
2381 | } | |
2382 | ||
2383 | return ns; | |
2384 | } | |
2385 | ||
bb34d92f | 2386 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 2387 | { |
1da177e4 | 2388 | unsigned long flags; |
41b86e9c | 2389 | struct rq *rq; |
bb34d92f | 2390 | u64 ns = 0; |
48f24c4d | 2391 | |
41b86e9c | 2392 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 2393 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 2394 | task_rq_unlock(rq, p, &flags); |
1508487e | 2395 | |
c5f8d995 HS |
2396 | return ns; |
2397 | } | |
f06febc9 | 2398 | |
c5f8d995 HS |
2399 | /* |
2400 | * Return accounted runtime for the task. | |
2401 | * In case the task is currently running, return the runtime plus current's | |
2402 | * pending runtime that have not been accounted yet. | |
2403 | */ | |
2404 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2405 | { | |
2406 | unsigned long flags; | |
2407 | struct rq *rq; | |
2408 | u64 ns = 0; | |
2409 | ||
911b2898 PZ |
2410 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
2411 | /* | |
2412 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
2413 | * So we have a optimization chance when the task's delta_exec is 0. | |
2414 | * Reading ->on_cpu is racy, but this is ok. | |
2415 | * | |
2416 | * If we race with it leaving cpu, we'll take a lock. So we're correct. | |
2417 | * If we race with it entering cpu, unaccounted time is 0. This is | |
2418 | * indistinguishable from the read occurring a few cycles earlier. | |
2419 | */ | |
2420 | if (!p->on_cpu) | |
2421 | return p->se.sum_exec_runtime; | |
2422 | #endif | |
2423 | ||
c5f8d995 HS |
2424 | rq = task_rq_lock(p, &flags); |
2425 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 2426 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
2427 | |
2428 | return ns; | |
2429 | } | |
48f24c4d | 2430 | |
7835b98b CL |
2431 | /* |
2432 | * This function gets called by the timer code, with HZ frequency. | |
2433 | * We call it with interrupts disabled. | |
7835b98b CL |
2434 | */ |
2435 | void scheduler_tick(void) | |
2436 | { | |
7835b98b CL |
2437 | int cpu = smp_processor_id(); |
2438 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 2439 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
2440 | |
2441 | sched_clock_tick(); | |
dd41f596 | 2442 | |
05fa785c | 2443 | raw_spin_lock(&rq->lock); |
3e51f33f | 2444 | update_rq_clock(rq); |
fa85ae24 | 2445 | curr->sched_class->task_tick(rq, curr, 0); |
83dfd523 | 2446 | update_cpu_load_active(rq); |
05fa785c | 2447 | raw_spin_unlock(&rq->lock); |
7835b98b | 2448 | |
e9d2b064 | 2449 | perf_event_task_tick(); |
e220d2dc | 2450 | |
e418e1c2 | 2451 | #ifdef CONFIG_SMP |
6eb57e0d | 2452 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 2453 | trigger_load_balance(rq); |
e418e1c2 | 2454 | #endif |
265f22a9 | 2455 | rq_last_tick_reset(rq); |
1da177e4 LT |
2456 | } |
2457 | ||
265f22a9 FW |
2458 | #ifdef CONFIG_NO_HZ_FULL |
2459 | /** | |
2460 | * scheduler_tick_max_deferment | |
2461 | * | |
2462 | * Keep at least one tick per second when a single | |
2463 | * active task is running because the scheduler doesn't | |
2464 | * yet completely support full dynticks environment. | |
2465 | * | |
2466 | * This makes sure that uptime, CFS vruntime, load | |
2467 | * balancing, etc... continue to move forward, even | |
2468 | * with a very low granularity. | |
e69f6186 YB |
2469 | * |
2470 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
2471 | */ |
2472 | u64 scheduler_tick_max_deferment(void) | |
2473 | { | |
2474 | struct rq *rq = this_rq(); | |
2475 | unsigned long next, now = ACCESS_ONCE(jiffies); | |
2476 | ||
2477 | next = rq->last_sched_tick + HZ; | |
2478 | ||
2479 | if (time_before_eq(next, now)) | |
2480 | return 0; | |
2481 | ||
8fe8ff09 | 2482 | return jiffies_to_nsecs(next - now); |
1da177e4 | 2483 | } |
265f22a9 | 2484 | #endif |
1da177e4 | 2485 | |
132380a0 | 2486 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
2487 | { |
2488 | if (in_lock_functions(addr)) { | |
2489 | addr = CALLER_ADDR2; | |
2490 | if (in_lock_functions(addr)) | |
2491 | addr = CALLER_ADDR3; | |
2492 | } | |
2493 | return addr; | |
2494 | } | |
1da177e4 | 2495 | |
7e49fcce SR |
2496 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
2497 | defined(CONFIG_PREEMPT_TRACER)) | |
2498 | ||
bdb43806 | 2499 | void __kprobes preempt_count_add(int val) |
1da177e4 | 2500 | { |
6cd8a4bb | 2501 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2502 | /* |
2503 | * Underflow? | |
2504 | */ | |
9a11b49a IM |
2505 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
2506 | return; | |
6cd8a4bb | 2507 | #endif |
bdb43806 | 2508 | __preempt_count_add(val); |
6cd8a4bb | 2509 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2510 | /* |
2511 | * Spinlock count overflowing soon? | |
2512 | */ | |
33859f7f MOS |
2513 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
2514 | PREEMPT_MASK - 10); | |
6cd8a4bb | 2515 | #endif |
8f47b187 TG |
2516 | if (preempt_count() == val) { |
2517 | unsigned long ip = get_parent_ip(CALLER_ADDR1); | |
2518 | #ifdef CONFIG_DEBUG_PREEMPT | |
2519 | current->preempt_disable_ip = ip; | |
2520 | #endif | |
2521 | trace_preempt_off(CALLER_ADDR0, ip); | |
2522 | } | |
1da177e4 | 2523 | } |
bdb43806 | 2524 | EXPORT_SYMBOL(preempt_count_add); |
1da177e4 | 2525 | |
bdb43806 | 2526 | void __kprobes preempt_count_sub(int val) |
1da177e4 | 2527 | { |
6cd8a4bb | 2528 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
2529 | /* |
2530 | * Underflow? | |
2531 | */ | |
01e3eb82 | 2532 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 2533 | return; |
1da177e4 LT |
2534 | /* |
2535 | * Is the spinlock portion underflowing? | |
2536 | */ | |
9a11b49a IM |
2537 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
2538 | !(preempt_count() & PREEMPT_MASK))) | |
2539 | return; | |
6cd8a4bb | 2540 | #endif |
9a11b49a | 2541 | |
6cd8a4bb SR |
2542 | if (preempt_count() == val) |
2543 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
bdb43806 | 2544 | __preempt_count_sub(val); |
1da177e4 | 2545 | } |
bdb43806 | 2546 | EXPORT_SYMBOL(preempt_count_sub); |
1da177e4 LT |
2547 | |
2548 | #endif | |
2549 | ||
2550 | /* | |
dd41f596 | 2551 | * Print scheduling while atomic bug: |
1da177e4 | 2552 | */ |
dd41f596 | 2553 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 2554 | { |
664dfa65 DJ |
2555 | if (oops_in_progress) |
2556 | return; | |
2557 | ||
3df0fc5b PZ |
2558 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
2559 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 2560 | |
dd41f596 | 2561 | debug_show_held_locks(prev); |
e21f5b15 | 2562 | print_modules(); |
dd41f596 IM |
2563 | if (irqs_disabled()) |
2564 | print_irqtrace_events(prev); | |
8f47b187 TG |
2565 | #ifdef CONFIG_DEBUG_PREEMPT |
2566 | if (in_atomic_preempt_off()) { | |
2567 | pr_err("Preemption disabled at:"); | |
2568 | print_ip_sym(current->preempt_disable_ip); | |
2569 | pr_cont("\n"); | |
2570 | } | |
2571 | #endif | |
6135fc1e | 2572 | dump_stack(); |
373d4d09 | 2573 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 2574 | } |
1da177e4 | 2575 | |
dd41f596 IM |
2576 | /* |
2577 | * Various schedule()-time debugging checks and statistics: | |
2578 | */ | |
2579 | static inline void schedule_debug(struct task_struct *prev) | |
2580 | { | |
1da177e4 | 2581 | /* |
41a2d6cf | 2582 | * Test if we are atomic. Since do_exit() needs to call into |
192301e7 ON |
2583 | * schedule() atomically, we ignore that path. Otherwise whine |
2584 | * if we are scheduling when we should not. | |
1da177e4 | 2585 | */ |
192301e7 | 2586 | if (unlikely(in_atomic_preempt_off() && prev->state != TASK_DEAD)) |
dd41f596 | 2587 | __schedule_bug(prev); |
b3fbab05 | 2588 | rcu_sleep_check(); |
dd41f596 | 2589 | |
1da177e4 LT |
2590 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
2591 | ||
2d72376b | 2592 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
2593 | } |
2594 | ||
2595 | /* | |
2596 | * Pick up the highest-prio task: | |
2597 | */ | |
2598 | static inline struct task_struct * | |
606dba2e | 2599 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 2600 | { |
37e117c0 | 2601 | const struct sched_class *class = &fair_sched_class; |
dd41f596 | 2602 | struct task_struct *p; |
1da177e4 LT |
2603 | |
2604 | /* | |
dd41f596 IM |
2605 | * Optimization: we know that if all tasks are in |
2606 | * the fair class we can call that function directly: | |
1da177e4 | 2607 | */ |
37e117c0 | 2608 | if (likely(prev->sched_class == class && |
38033c37 | 2609 | rq->nr_running == rq->cfs.h_nr_running)) { |
606dba2e | 2610 | p = fair_sched_class.pick_next_task(rq, prev); |
6ccdc84b PZ |
2611 | if (unlikely(p == RETRY_TASK)) |
2612 | goto again; | |
2613 | ||
2614 | /* assumes fair_sched_class->next == idle_sched_class */ | |
2615 | if (unlikely(!p)) | |
2616 | p = idle_sched_class.pick_next_task(rq, prev); | |
2617 | ||
2618 | return p; | |
1da177e4 LT |
2619 | } |
2620 | ||
37e117c0 | 2621 | again: |
34f971f6 | 2622 | for_each_class(class) { |
606dba2e | 2623 | p = class->pick_next_task(rq, prev); |
37e117c0 PZ |
2624 | if (p) { |
2625 | if (unlikely(p == RETRY_TASK)) | |
2626 | goto again; | |
dd41f596 | 2627 | return p; |
37e117c0 | 2628 | } |
dd41f596 | 2629 | } |
34f971f6 PZ |
2630 | |
2631 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 2632 | } |
1da177e4 | 2633 | |
dd41f596 | 2634 | /* |
c259e01a | 2635 | * __schedule() is the main scheduler function. |
edde96ea PE |
2636 | * |
2637 | * The main means of driving the scheduler and thus entering this function are: | |
2638 | * | |
2639 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
2640 | * | |
2641 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
2642 | * paths. For example, see arch/x86/entry_64.S. | |
2643 | * | |
2644 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
2645 | * interrupt handler scheduler_tick(). | |
2646 | * | |
2647 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
2648 | * task to the run-queue and that's it. | |
2649 | * | |
2650 | * Now, if the new task added to the run-queue preempts the current | |
2651 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
2652 | * called on the nearest possible occasion: | |
2653 | * | |
2654 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
2655 | * | |
2656 | * - in syscall or exception context, at the next outmost | |
2657 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
2658 | * spin_unlock()!) | |
2659 | * | |
2660 | * - in IRQ context, return from interrupt-handler to | |
2661 | * preemptible context | |
2662 | * | |
2663 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
2664 | * then at the next: | |
2665 | * | |
2666 | * - cond_resched() call | |
2667 | * - explicit schedule() call | |
2668 | * - return from syscall or exception to user-space | |
2669 | * - return from interrupt-handler to user-space | |
dd41f596 | 2670 | */ |
c259e01a | 2671 | static void __sched __schedule(void) |
dd41f596 IM |
2672 | { |
2673 | struct task_struct *prev, *next; | |
67ca7bde | 2674 | unsigned long *switch_count; |
dd41f596 | 2675 | struct rq *rq; |
31656519 | 2676 | int cpu; |
dd41f596 | 2677 | |
ff743345 PZ |
2678 | need_resched: |
2679 | preempt_disable(); | |
dd41f596 IM |
2680 | cpu = smp_processor_id(); |
2681 | rq = cpu_rq(cpu); | |
25502a6c | 2682 | rcu_note_context_switch(cpu); |
dd41f596 | 2683 | prev = rq->curr; |
dd41f596 | 2684 | |
dd41f596 | 2685 | schedule_debug(prev); |
1da177e4 | 2686 | |
31656519 | 2687 | if (sched_feat(HRTICK)) |
f333fdc9 | 2688 | hrtick_clear(rq); |
8f4d37ec | 2689 | |
e0acd0a6 ON |
2690 | /* |
2691 | * Make sure that signal_pending_state()->signal_pending() below | |
2692 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
2693 | * done by the caller to avoid the race with signal_wake_up(). | |
2694 | */ | |
2695 | smp_mb__before_spinlock(); | |
05fa785c | 2696 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 2697 | |
246d86b5 | 2698 | switch_count = &prev->nivcsw; |
1da177e4 | 2699 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 2700 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 2701 | prev->state = TASK_RUNNING; |
21aa9af0 | 2702 | } else { |
2acca55e PZ |
2703 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
2704 | prev->on_rq = 0; | |
2705 | ||
21aa9af0 | 2706 | /* |
2acca55e PZ |
2707 | * If a worker went to sleep, notify and ask workqueue |
2708 | * whether it wants to wake up a task to maintain | |
2709 | * concurrency. | |
21aa9af0 TH |
2710 | */ |
2711 | if (prev->flags & PF_WQ_WORKER) { | |
2712 | struct task_struct *to_wakeup; | |
2713 | ||
2714 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
2715 | if (to_wakeup) | |
2716 | try_to_wake_up_local(to_wakeup); | |
2717 | } | |
21aa9af0 | 2718 | } |
dd41f596 | 2719 | switch_count = &prev->nvcsw; |
1da177e4 LT |
2720 | } |
2721 | ||
606dba2e PZ |
2722 | if (prev->on_rq || rq->skip_clock_update < 0) |
2723 | update_rq_clock(rq); | |
2724 | ||
2725 | next = pick_next_task(rq, prev); | |
f26f9aff | 2726 | clear_tsk_need_resched(prev); |
f27dde8d | 2727 | clear_preempt_need_resched(); |
f26f9aff | 2728 | rq->skip_clock_update = 0; |
1da177e4 | 2729 | |
1da177e4 | 2730 | if (likely(prev != next)) { |
1da177e4 LT |
2731 | rq->nr_switches++; |
2732 | rq->curr = next; | |
2733 | ++*switch_count; | |
2734 | ||
dd41f596 | 2735 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 2736 | /* |
246d86b5 ON |
2737 | * The context switch have flipped the stack from under us |
2738 | * and restored the local variables which were saved when | |
2739 | * this task called schedule() in the past. prev == current | |
2740 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
2741 | */ |
2742 | cpu = smp_processor_id(); | |
2743 | rq = cpu_rq(cpu); | |
1da177e4 | 2744 | } else |
05fa785c | 2745 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 2746 | |
3f029d3c | 2747 | post_schedule(rq); |
1da177e4 | 2748 | |
ba74c144 | 2749 | sched_preempt_enable_no_resched(); |
ff743345 | 2750 | if (need_resched()) |
1da177e4 LT |
2751 | goto need_resched; |
2752 | } | |
c259e01a | 2753 | |
9c40cef2 TG |
2754 | static inline void sched_submit_work(struct task_struct *tsk) |
2755 | { | |
3c7d5184 | 2756 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
2757 | return; |
2758 | /* | |
2759 | * If we are going to sleep and we have plugged IO queued, | |
2760 | * make sure to submit it to avoid deadlocks. | |
2761 | */ | |
2762 | if (blk_needs_flush_plug(tsk)) | |
2763 | blk_schedule_flush_plug(tsk); | |
2764 | } | |
2765 | ||
722a9f92 | 2766 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 2767 | { |
9c40cef2 TG |
2768 | struct task_struct *tsk = current; |
2769 | ||
2770 | sched_submit_work(tsk); | |
c259e01a TG |
2771 | __schedule(); |
2772 | } | |
1da177e4 LT |
2773 | EXPORT_SYMBOL(schedule); |
2774 | ||
91d1aa43 | 2775 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 2776 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
2777 | { |
2778 | /* | |
2779 | * If we come here after a random call to set_need_resched(), | |
2780 | * or we have been woken up remotely but the IPI has not yet arrived, | |
2781 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
2782 | * we find a better solution. | |
2783 | */ | |
91d1aa43 | 2784 | user_exit(); |
20ab65e3 | 2785 | schedule(); |
91d1aa43 | 2786 | user_enter(); |
20ab65e3 FW |
2787 | } |
2788 | #endif | |
2789 | ||
c5491ea7 TG |
2790 | /** |
2791 | * schedule_preempt_disabled - called with preemption disabled | |
2792 | * | |
2793 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
2794 | */ | |
2795 | void __sched schedule_preempt_disabled(void) | |
2796 | { | |
ba74c144 | 2797 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
2798 | schedule(); |
2799 | preempt_disable(); | |
2800 | } | |
2801 | ||
1da177e4 LT |
2802 | #ifdef CONFIG_PREEMPT |
2803 | /* | |
2ed6e34f | 2804 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 2805 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
2806 | * occur there and call schedule directly. |
2807 | */ | |
722a9f92 | 2808 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 2809 | { |
1da177e4 LT |
2810 | /* |
2811 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 2812 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 2813 | */ |
fbb00b56 | 2814 | if (likely(!preemptible())) |
1da177e4 LT |
2815 | return; |
2816 | ||
3a5c359a | 2817 | do { |
bdb43806 | 2818 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 2819 | __schedule(); |
bdb43806 | 2820 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2821 | |
3a5c359a AK |
2822 | /* |
2823 | * Check again in case we missed a preemption opportunity | |
2824 | * between schedule and now. | |
2825 | */ | |
2826 | barrier(); | |
5ed0cec0 | 2827 | } while (need_resched()); |
1da177e4 | 2828 | } |
1da177e4 | 2829 | EXPORT_SYMBOL(preempt_schedule); |
32e475d7 | 2830 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
2831 | |
2832 | /* | |
2ed6e34f | 2833 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
2834 | * off of irq context. |
2835 | * Note, that this is called and return with irqs disabled. This will | |
2836 | * protect us against recursive calling from irq. | |
2837 | */ | |
722a9f92 | 2838 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 2839 | { |
b22366cd | 2840 | enum ctx_state prev_state; |
6478d880 | 2841 | |
2ed6e34f | 2842 | /* Catch callers which need to be fixed */ |
f27dde8d | 2843 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 2844 | |
b22366cd FW |
2845 | prev_state = exception_enter(); |
2846 | ||
3a5c359a | 2847 | do { |
bdb43806 | 2848 | __preempt_count_add(PREEMPT_ACTIVE); |
3a5c359a | 2849 | local_irq_enable(); |
c259e01a | 2850 | __schedule(); |
3a5c359a | 2851 | local_irq_disable(); |
bdb43806 | 2852 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 | 2853 | |
3a5c359a AK |
2854 | /* |
2855 | * Check again in case we missed a preemption opportunity | |
2856 | * between schedule and now. | |
2857 | */ | |
2858 | barrier(); | |
5ed0cec0 | 2859 | } while (need_resched()); |
b22366cd FW |
2860 | |
2861 | exception_exit(prev_state); | |
1da177e4 LT |
2862 | } |
2863 | ||
63859d4f | 2864 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 2865 | void *key) |
1da177e4 | 2866 | { |
63859d4f | 2867 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 2868 | } |
1da177e4 LT |
2869 | EXPORT_SYMBOL(default_wake_function); |
2870 | ||
b29739f9 IM |
2871 | #ifdef CONFIG_RT_MUTEXES |
2872 | ||
2873 | /* | |
2874 | * rt_mutex_setprio - set the current priority of a task | |
2875 | * @p: task | |
2876 | * @prio: prio value (kernel-internal form) | |
2877 | * | |
2878 | * This function changes the 'effective' priority of a task. It does | |
2879 | * not touch ->normal_prio like __setscheduler(). | |
2880 | * | |
c365c292 TG |
2881 | * Used by the rt_mutex code to implement priority inheritance |
2882 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 2883 | */ |
36c8b586 | 2884 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 2885 | { |
2d3d891d | 2886 | int oldprio, on_rq, running, enqueue_flag = 0; |
70b97a7f | 2887 | struct rq *rq; |
83ab0aa0 | 2888 | const struct sched_class *prev_class; |
b29739f9 | 2889 | |
aab03e05 | 2890 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 2891 | |
0122ec5b | 2892 | rq = __task_rq_lock(p); |
b29739f9 | 2893 | |
1c4dd99b TG |
2894 | /* |
2895 | * Idle task boosting is a nono in general. There is one | |
2896 | * exception, when PREEMPT_RT and NOHZ is active: | |
2897 | * | |
2898 | * The idle task calls get_next_timer_interrupt() and holds | |
2899 | * the timer wheel base->lock on the CPU and another CPU wants | |
2900 | * to access the timer (probably to cancel it). We can safely | |
2901 | * ignore the boosting request, as the idle CPU runs this code | |
2902 | * with interrupts disabled and will complete the lock | |
2903 | * protected section without being interrupted. So there is no | |
2904 | * real need to boost. | |
2905 | */ | |
2906 | if (unlikely(p == rq->idle)) { | |
2907 | WARN_ON(p != rq->curr); | |
2908 | WARN_ON(p->pi_blocked_on); | |
2909 | goto out_unlock; | |
2910 | } | |
2911 | ||
a8027073 | 2912 | trace_sched_pi_setprio(p, prio); |
2d3d891d | 2913 | p->pi_top_task = rt_mutex_get_top_task(p); |
d5f9f942 | 2914 | oldprio = p->prio; |
83ab0aa0 | 2915 | prev_class = p->sched_class; |
fd2f4419 | 2916 | on_rq = p->on_rq; |
051a1d1a | 2917 | running = task_current(rq, p); |
0e1f3483 | 2918 | if (on_rq) |
69be72c1 | 2919 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
2920 | if (running) |
2921 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 | 2922 | |
2d3d891d DF |
2923 | /* |
2924 | * Boosting condition are: | |
2925 | * 1. -rt task is running and holds mutex A | |
2926 | * --> -dl task blocks on mutex A | |
2927 | * | |
2928 | * 2. -dl task is running and holds mutex A | |
2929 | * --> -dl task blocks on mutex A and could preempt the | |
2930 | * running task | |
2931 | */ | |
2932 | if (dl_prio(prio)) { | |
2933 | if (!dl_prio(p->normal_prio) || (p->pi_top_task && | |
2934 | dl_entity_preempt(&p->pi_top_task->dl, &p->dl))) { | |
2935 | p->dl.dl_boosted = 1; | |
2936 | p->dl.dl_throttled = 0; | |
2937 | enqueue_flag = ENQUEUE_REPLENISH; | |
2938 | } else | |
2939 | p->dl.dl_boosted = 0; | |
aab03e05 | 2940 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
2941 | } else if (rt_prio(prio)) { |
2942 | if (dl_prio(oldprio)) | |
2943 | p->dl.dl_boosted = 0; | |
2944 | if (oldprio < prio) | |
2945 | enqueue_flag = ENQUEUE_HEAD; | |
dd41f596 | 2946 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
2947 | } else { |
2948 | if (dl_prio(oldprio)) | |
2949 | p->dl.dl_boosted = 0; | |
dd41f596 | 2950 | p->sched_class = &fair_sched_class; |
2d3d891d | 2951 | } |
dd41f596 | 2952 | |
b29739f9 IM |
2953 | p->prio = prio; |
2954 | ||
0e1f3483 HS |
2955 | if (running) |
2956 | p->sched_class->set_curr_task(rq); | |
da7a735e | 2957 | if (on_rq) |
2d3d891d | 2958 | enqueue_task(rq, p, enqueue_flag); |
cb469845 | 2959 | |
da7a735e | 2960 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 2961 | out_unlock: |
0122ec5b | 2962 | __task_rq_unlock(rq); |
b29739f9 | 2963 | } |
b29739f9 | 2964 | #endif |
d50dde5a | 2965 | |
36c8b586 | 2966 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 2967 | { |
dd41f596 | 2968 | int old_prio, delta, on_rq; |
1da177e4 | 2969 | unsigned long flags; |
70b97a7f | 2970 | struct rq *rq; |
1da177e4 | 2971 | |
75e45d51 | 2972 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
2973 | return; |
2974 | /* | |
2975 | * We have to be careful, if called from sys_setpriority(), | |
2976 | * the task might be in the middle of scheduling on another CPU. | |
2977 | */ | |
2978 | rq = task_rq_lock(p, &flags); | |
2979 | /* | |
2980 | * The RT priorities are set via sched_setscheduler(), but we still | |
2981 | * allow the 'normal' nice value to be set - but as expected | |
2982 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 2983 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 2984 | */ |
aab03e05 | 2985 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
2986 | p->static_prio = NICE_TO_PRIO(nice); |
2987 | goto out_unlock; | |
2988 | } | |
fd2f4419 | 2989 | on_rq = p->on_rq; |
c09595f6 | 2990 | if (on_rq) |
69be72c1 | 2991 | dequeue_task(rq, p, 0); |
1da177e4 | 2992 | |
1da177e4 | 2993 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 2994 | set_load_weight(p); |
b29739f9 IM |
2995 | old_prio = p->prio; |
2996 | p->prio = effective_prio(p); | |
2997 | delta = p->prio - old_prio; | |
1da177e4 | 2998 | |
dd41f596 | 2999 | if (on_rq) { |
371fd7e7 | 3000 | enqueue_task(rq, p, 0); |
1da177e4 | 3001 | /* |
d5f9f942 AM |
3002 | * If the task increased its priority or is running and |
3003 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3004 | */ |
d5f9f942 | 3005 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3006 | resched_task(rq->curr); |
3007 | } | |
3008 | out_unlock: | |
0122ec5b | 3009 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 3010 | } |
1da177e4 LT |
3011 | EXPORT_SYMBOL(set_user_nice); |
3012 | ||
e43379f1 MM |
3013 | /* |
3014 | * can_nice - check if a task can reduce its nice value | |
3015 | * @p: task | |
3016 | * @nice: nice value | |
3017 | */ | |
36c8b586 | 3018 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3019 | { |
024f4747 | 3020 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
7aa2c016 | 3021 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3022 | |
78d7d407 | 3023 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3024 | capable(CAP_SYS_NICE)); |
3025 | } | |
3026 | ||
1da177e4 LT |
3027 | #ifdef __ARCH_WANT_SYS_NICE |
3028 | ||
3029 | /* | |
3030 | * sys_nice - change the priority of the current process. | |
3031 | * @increment: priority increment | |
3032 | * | |
3033 | * sys_setpriority is a more generic, but much slower function that | |
3034 | * does similar things. | |
3035 | */ | |
5add95d4 | 3036 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3037 | { |
48f24c4d | 3038 | long nice, retval; |
1da177e4 LT |
3039 | |
3040 | /* | |
3041 | * Setpriority might change our priority at the same moment. | |
3042 | * We don't have to worry. Conceptually one call occurs first | |
3043 | * and we have a single winner. | |
3044 | */ | |
a9467fa3 | 3045 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3046 | nice = task_nice(current) + increment; |
1da177e4 | 3047 | |
a9467fa3 | 3048 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3049 | if (increment < 0 && !can_nice(current, nice)) |
3050 | return -EPERM; | |
3051 | ||
1da177e4 LT |
3052 | retval = security_task_setnice(current, nice); |
3053 | if (retval) | |
3054 | return retval; | |
3055 | ||
3056 | set_user_nice(current, nice); | |
3057 | return 0; | |
3058 | } | |
3059 | ||
3060 | #endif | |
3061 | ||
3062 | /** | |
3063 | * task_prio - return the priority value of a given task. | |
3064 | * @p: the task in question. | |
3065 | * | |
e69f6186 | 3066 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3067 | * RT tasks are offset by -200. Normal tasks are centered |
3068 | * around 0, value goes from -16 to +15. | |
3069 | */ | |
36c8b586 | 3070 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3071 | { |
3072 | return p->prio - MAX_RT_PRIO; | |
3073 | } | |
3074 | ||
1da177e4 LT |
3075 | /** |
3076 | * idle_cpu - is a given cpu idle currently? | |
3077 | * @cpu: the processor in question. | |
e69f6186 YB |
3078 | * |
3079 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3080 | */ |
3081 | int idle_cpu(int cpu) | |
3082 | { | |
908a3283 TG |
3083 | struct rq *rq = cpu_rq(cpu); |
3084 | ||
3085 | if (rq->curr != rq->idle) | |
3086 | return 0; | |
3087 | ||
3088 | if (rq->nr_running) | |
3089 | return 0; | |
3090 | ||
3091 | #ifdef CONFIG_SMP | |
3092 | if (!llist_empty(&rq->wake_list)) | |
3093 | return 0; | |
3094 | #endif | |
3095 | ||
3096 | return 1; | |
1da177e4 LT |
3097 | } |
3098 | ||
1da177e4 LT |
3099 | /** |
3100 | * idle_task - return the idle task for a given cpu. | |
3101 | * @cpu: the processor in question. | |
e69f6186 YB |
3102 | * |
3103 | * Return: The idle task for the cpu @cpu. | |
1da177e4 | 3104 | */ |
36c8b586 | 3105 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3106 | { |
3107 | return cpu_rq(cpu)->idle; | |
3108 | } | |
3109 | ||
3110 | /** | |
3111 | * find_process_by_pid - find a process with a matching PID value. | |
3112 | * @pid: the pid in question. | |
e69f6186 YB |
3113 | * |
3114 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3115 | */ |
a9957449 | 3116 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3117 | { |
228ebcbe | 3118 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3119 | } |
3120 | ||
aab03e05 DF |
3121 | /* |
3122 | * This function initializes the sched_dl_entity of a newly becoming | |
3123 | * SCHED_DEADLINE task. | |
3124 | * | |
3125 | * Only the static values are considered here, the actual runtime and the | |
3126 | * absolute deadline will be properly calculated when the task is enqueued | |
3127 | * for the first time with its new policy. | |
3128 | */ | |
3129 | static void | |
3130 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3131 | { | |
3132 | struct sched_dl_entity *dl_se = &p->dl; | |
3133 | ||
3134 | init_dl_task_timer(dl_se); | |
3135 | dl_se->dl_runtime = attr->sched_runtime; | |
3136 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3137 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3138 | dl_se->flags = attr->sched_flags; |
332ac17e | 3139 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
aab03e05 DF |
3140 | dl_se->dl_throttled = 0; |
3141 | dl_se->dl_new = 1; | |
5bfd126e | 3142 | dl_se->dl_yielded = 0; |
aab03e05 DF |
3143 | } |
3144 | ||
c365c292 TG |
3145 | static void __setscheduler_params(struct task_struct *p, |
3146 | const struct sched_attr *attr) | |
1da177e4 | 3147 | { |
d50dde5a DF |
3148 | int policy = attr->sched_policy; |
3149 | ||
39fd8fd2 PZ |
3150 | if (policy == -1) /* setparam */ |
3151 | policy = p->policy; | |
3152 | ||
1da177e4 | 3153 | p->policy = policy; |
d50dde5a | 3154 | |
aab03e05 DF |
3155 | if (dl_policy(policy)) |
3156 | __setparam_dl(p, attr); | |
39fd8fd2 | 3157 | else if (fair_policy(policy)) |
d50dde5a DF |
3158 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3159 | ||
39fd8fd2 PZ |
3160 | /* |
3161 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3162 | * !rt_policy. Always setting this ensures that things like | |
3163 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3164 | */ | |
3165 | p->rt_priority = attr->sched_priority; | |
383afd09 | 3166 | p->normal_prio = normal_prio(p); |
c365c292 TG |
3167 | set_load_weight(p); |
3168 | } | |
39fd8fd2 | 3169 | |
c365c292 TG |
3170 | /* Actually do priority change: must hold pi & rq lock. */ |
3171 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
3172 | const struct sched_attr *attr) | |
3173 | { | |
3174 | __setscheduler_params(p, attr); | |
d50dde5a | 3175 | |
383afd09 SR |
3176 | /* |
3177 | * If we get here, there was no pi waiters boosting the | |
3178 | * task. It is safe to use the normal prio. | |
3179 | */ | |
3180 | p->prio = normal_prio(p); | |
3181 | ||
aab03e05 DF |
3182 | if (dl_prio(p->prio)) |
3183 | p->sched_class = &dl_sched_class; | |
3184 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
3185 | p->sched_class = &rt_sched_class; |
3186 | else | |
3187 | p->sched_class = &fair_sched_class; | |
1da177e4 | 3188 | } |
aab03e05 DF |
3189 | |
3190 | static void | |
3191 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
3192 | { | |
3193 | struct sched_dl_entity *dl_se = &p->dl; | |
3194 | ||
3195 | attr->sched_priority = p->rt_priority; | |
3196 | attr->sched_runtime = dl_se->dl_runtime; | |
3197 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 3198 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
3199 | attr->sched_flags = dl_se->flags; |
3200 | } | |
3201 | ||
3202 | /* | |
3203 | * This function validates the new parameters of a -deadline task. | |
3204 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 3205 | * than the runtime, as well as the period of being zero or |
332ac17e | 3206 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
3207 | * user parameters are above the internal resolution of 1us (we |
3208 | * check sched_runtime only since it is always the smaller one) and | |
3209 | * below 2^63 ns (we have to check both sched_deadline and | |
3210 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
3211 | */ |
3212 | static bool | |
3213 | __checkparam_dl(const struct sched_attr *attr) | |
3214 | { | |
b0827819 JL |
3215 | /* deadline != 0 */ |
3216 | if (attr->sched_deadline == 0) | |
3217 | return false; | |
3218 | ||
3219 | /* | |
3220 | * Since we truncate DL_SCALE bits, make sure we're at least | |
3221 | * that big. | |
3222 | */ | |
3223 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
3224 | return false; | |
3225 | ||
3226 | /* | |
3227 | * Since we use the MSB for wrap-around and sign issues, make | |
3228 | * sure it's not set (mind that period can be equal to zero). | |
3229 | */ | |
3230 | if (attr->sched_deadline & (1ULL << 63) || | |
3231 | attr->sched_period & (1ULL << 63)) | |
3232 | return false; | |
3233 | ||
3234 | /* runtime <= deadline <= period (if period != 0) */ | |
3235 | if ((attr->sched_period != 0 && | |
3236 | attr->sched_period < attr->sched_deadline) || | |
3237 | attr->sched_deadline < attr->sched_runtime) | |
3238 | return false; | |
3239 | ||
3240 | return true; | |
aab03e05 DF |
3241 | } |
3242 | ||
c69e8d9c DH |
3243 | /* |
3244 | * check the target process has a UID that matches the current process's | |
3245 | */ | |
3246 | static bool check_same_owner(struct task_struct *p) | |
3247 | { | |
3248 | const struct cred *cred = current_cred(), *pcred; | |
3249 | bool match; | |
3250 | ||
3251 | rcu_read_lock(); | |
3252 | pcred = __task_cred(p); | |
9c806aa0 EB |
3253 | match = (uid_eq(cred->euid, pcred->euid) || |
3254 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
3255 | rcu_read_unlock(); |
3256 | return match; | |
3257 | } | |
3258 | ||
d50dde5a DF |
3259 | static int __sched_setscheduler(struct task_struct *p, |
3260 | const struct sched_attr *attr, | |
3261 | bool user) | |
1da177e4 | 3262 | { |
383afd09 SR |
3263 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
3264 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
83b699ed | 3265 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
d50dde5a | 3266 | int policy = attr->sched_policy; |
1da177e4 | 3267 | unsigned long flags; |
83ab0aa0 | 3268 | const struct sched_class *prev_class; |
70b97a7f | 3269 | struct rq *rq; |
ca94c442 | 3270 | int reset_on_fork; |
1da177e4 | 3271 | |
66e5393a SR |
3272 | /* may grab non-irq protected spin_locks */ |
3273 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
3274 | recheck: |
3275 | /* double check policy once rq lock held */ | |
ca94c442 LP |
3276 | if (policy < 0) { |
3277 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 3278 | policy = oldpolicy = p->policy; |
ca94c442 | 3279 | } else { |
7479f3c9 | 3280 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 3281 | |
aab03e05 DF |
3282 | if (policy != SCHED_DEADLINE && |
3283 | policy != SCHED_FIFO && policy != SCHED_RR && | |
ca94c442 LP |
3284 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
3285 | policy != SCHED_IDLE) | |
3286 | return -EINVAL; | |
3287 | } | |
3288 | ||
7479f3c9 PZ |
3289 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
3290 | return -EINVAL; | |
3291 | ||
1da177e4 LT |
3292 | /* |
3293 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
3294 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
3295 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 3296 | */ |
0bb040a4 | 3297 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 3298 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 3299 | return -EINVAL; |
aab03e05 DF |
3300 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
3301 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
3302 | return -EINVAL; |
3303 | ||
37e4ab3f OC |
3304 | /* |
3305 | * Allow unprivileged RT tasks to decrease priority: | |
3306 | */ | |
961ccddd | 3307 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 3308 | if (fair_policy(policy)) { |
d0ea0268 | 3309 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 3310 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
3311 | return -EPERM; |
3312 | } | |
3313 | ||
e05606d3 | 3314 | if (rt_policy(policy)) { |
a44702e8 ON |
3315 | unsigned long rlim_rtprio = |
3316 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
3317 | |
3318 | /* can't set/change the rt policy */ | |
3319 | if (policy != p->policy && !rlim_rtprio) | |
3320 | return -EPERM; | |
3321 | ||
3322 | /* can't increase priority */ | |
d50dde5a DF |
3323 | if (attr->sched_priority > p->rt_priority && |
3324 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
3325 | return -EPERM; |
3326 | } | |
c02aa73b | 3327 | |
d44753b8 JL |
3328 | /* |
3329 | * Can't set/change SCHED_DEADLINE policy at all for now | |
3330 | * (safest behavior); in the future we would like to allow | |
3331 | * unprivileged DL tasks to increase their relative deadline | |
3332 | * or reduce their runtime (both ways reducing utilization) | |
3333 | */ | |
3334 | if (dl_policy(policy)) | |
3335 | return -EPERM; | |
3336 | ||
dd41f596 | 3337 | /* |
c02aa73b DH |
3338 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
3339 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 3340 | */ |
c02aa73b | 3341 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
d0ea0268 | 3342 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
3343 | return -EPERM; |
3344 | } | |
5fe1d75f | 3345 | |
37e4ab3f | 3346 | /* can't change other user's priorities */ |
c69e8d9c | 3347 | if (!check_same_owner(p)) |
37e4ab3f | 3348 | return -EPERM; |
ca94c442 LP |
3349 | |
3350 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
3351 | if (p->sched_reset_on_fork && !reset_on_fork) | |
3352 | return -EPERM; | |
37e4ab3f | 3353 | } |
1da177e4 | 3354 | |
725aad24 | 3355 | if (user) { |
b0ae1981 | 3356 | retval = security_task_setscheduler(p); |
725aad24 JF |
3357 | if (retval) |
3358 | return retval; | |
3359 | } | |
3360 | ||
b29739f9 IM |
3361 | /* |
3362 | * make sure no PI-waiters arrive (or leave) while we are | |
3363 | * changing the priority of the task: | |
0122ec5b | 3364 | * |
25985edc | 3365 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
3366 | * runqueue lock must be held. |
3367 | */ | |
0122ec5b | 3368 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 3369 | |
34f971f6 PZ |
3370 | /* |
3371 | * Changing the policy of the stop threads its a very bad idea | |
3372 | */ | |
3373 | if (p == rq->stop) { | |
0122ec5b | 3374 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
3375 | return -EINVAL; |
3376 | } | |
3377 | ||
a51e9198 | 3378 | /* |
d6b1e911 TG |
3379 | * If not changing anything there's no need to proceed further, |
3380 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 3381 | */ |
d50dde5a | 3382 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 3383 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
3384 | goto change; |
3385 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
3386 | goto change; | |
aab03e05 DF |
3387 | if (dl_policy(policy)) |
3388 | goto change; | |
d50dde5a | 3389 | |
d6b1e911 | 3390 | p->sched_reset_on_fork = reset_on_fork; |
45afb173 | 3391 | task_rq_unlock(rq, p, &flags); |
a51e9198 DF |
3392 | return 0; |
3393 | } | |
d50dde5a | 3394 | change: |
a51e9198 | 3395 | |
dc61b1d6 | 3396 | if (user) { |
332ac17e | 3397 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
3398 | /* |
3399 | * Do not allow realtime tasks into groups that have no runtime | |
3400 | * assigned. | |
3401 | */ | |
3402 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
3403 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
3404 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 3405 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
3406 | return -EPERM; |
3407 | } | |
dc61b1d6 | 3408 | #endif |
332ac17e DF |
3409 | #ifdef CONFIG_SMP |
3410 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
3411 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
3412 | |
3413 | /* | |
3414 | * Don't allow tasks with an affinity mask smaller than | |
3415 | * the entire root_domain to become SCHED_DEADLINE. We | |
3416 | * will also fail if there's no bandwidth available. | |
3417 | */ | |
e4099a5e PZ |
3418 | if (!cpumask_subset(span, &p->cpus_allowed) || |
3419 | rq->rd->dl_bw.bw == 0) { | |
332ac17e DF |
3420 | task_rq_unlock(rq, p, &flags); |
3421 | return -EPERM; | |
3422 | } | |
3423 | } | |
3424 | #endif | |
3425 | } | |
dc61b1d6 | 3426 | |
1da177e4 LT |
3427 | /* recheck policy now with rq lock held */ |
3428 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
3429 | policy = oldpolicy = -1; | |
0122ec5b | 3430 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
3431 | goto recheck; |
3432 | } | |
332ac17e DF |
3433 | |
3434 | /* | |
3435 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
3436 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
3437 | * is available. | |
3438 | */ | |
e4099a5e | 3439 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
332ac17e DF |
3440 | task_rq_unlock(rq, p, &flags); |
3441 | return -EBUSY; | |
3442 | } | |
3443 | ||
c365c292 TG |
3444 | p->sched_reset_on_fork = reset_on_fork; |
3445 | oldprio = p->prio; | |
3446 | ||
3447 | /* | |
3448 | * Special case for priority boosted tasks. | |
3449 | * | |
3450 | * If the new priority is lower or equal (user space view) | |
3451 | * than the current (boosted) priority, we just store the new | |
3452 | * normal parameters and do not touch the scheduler class and | |
3453 | * the runqueue. This will be done when the task deboost | |
3454 | * itself. | |
3455 | */ | |
3456 | if (rt_mutex_check_prio(p, newprio)) { | |
3457 | __setscheduler_params(p, attr); | |
3458 | task_rq_unlock(rq, p, &flags); | |
3459 | return 0; | |
3460 | } | |
3461 | ||
fd2f4419 | 3462 | on_rq = p->on_rq; |
051a1d1a | 3463 | running = task_current(rq, p); |
0e1f3483 | 3464 | if (on_rq) |
4ca9b72b | 3465 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
3466 | if (running) |
3467 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 3468 | |
83ab0aa0 | 3469 | prev_class = p->sched_class; |
d50dde5a | 3470 | __setscheduler(rq, p, attr); |
f6b53205 | 3471 | |
0e1f3483 HS |
3472 | if (running) |
3473 | p->sched_class->set_curr_task(rq); | |
81a44c54 TG |
3474 | if (on_rq) { |
3475 | /* | |
3476 | * We enqueue to tail when the priority of a task is | |
3477 | * increased (user space view). | |
3478 | */ | |
3479 | enqueue_task(rq, p, oldprio <= p->prio ? ENQUEUE_HEAD : 0); | |
3480 | } | |
cb469845 | 3481 | |
da7a735e | 3482 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 3483 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 3484 | |
95e02ca9 TG |
3485 | rt_mutex_adjust_pi(p); |
3486 | ||
1da177e4 LT |
3487 | return 0; |
3488 | } | |
961ccddd | 3489 | |
7479f3c9 PZ |
3490 | static int _sched_setscheduler(struct task_struct *p, int policy, |
3491 | const struct sched_param *param, bool check) | |
3492 | { | |
3493 | struct sched_attr attr = { | |
3494 | .sched_policy = policy, | |
3495 | .sched_priority = param->sched_priority, | |
3496 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
3497 | }; | |
3498 | ||
3499 | /* | |
3500 | * Fixup the legacy SCHED_RESET_ON_FORK hack | |
3501 | */ | |
3502 | if (policy & SCHED_RESET_ON_FORK) { | |
3503 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
3504 | policy &= ~SCHED_RESET_ON_FORK; | |
3505 | attr.sched_policy = policy; | |
3506 | } | |
3507 | ||
3508 | return __sched_setscheduler(p, &attr, check); | |
3509 | } | |
961ccddd RR |
3510 | /** |
3511 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
3512 | * @p: the task in question. | |
3513 | * @policy: new policy. | |
3514 | * @param: structure containing the new RT priority. | |
3515 | * | |
e69f6186 YB |
3516 | * Return: 0 on success. An error code otherwise. |
3517 | * | |
961ccddd RR |
3518 | * NOTE that the task may be already dead. |
3519 | */ | |
3520 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 3521 | const struct sched_param *param) |
961ccddd | 3522 | { |
7479f3c9 | 3523 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 3524 | } |
1da177e4 LT |
3525 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
3526 | ||
d50dde5a DF |
3527 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
3528 | { | |
3529 | return __sched_setscheduler(p, attr, true); | |
3530 | } | |
3531 | EXPORT_SYMBOL_GPL(sched_setattr); | |
3532 | ||
961ccddd RR |
3533 | /** |
3534 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
3535 | * @p: the task in question. | |
3536 | * @policy: new policy. | |
3537 | * @param: structure containing the new RT priority. | |
3538 | * | |
3539 | * Just like sched_setscheduler, only don't bother checking if the | |
3540 | * current context has permission. For example, this is needed in | |
3541 | * stop_machine(): we create temporary high priority worker threads, | |
3542 | * but our caller might not have that capability. | |
e69f6186 YB |
3543 | * |
3544 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
3545 | */ |
3546 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 3547 | const struct sched_param *param) |
961ccddd | 3548 | { |
7479f3c9 | 3549 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
3550 | } |
3551 | ||
95cdf3b7 IM |
3552 | static int |
3553 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 3554 | { |
1da177e4 LT |
3555 | struct sched_param lparam; |
3556 | struct task_struct *p; | |
36c8b586 | 3557 | int retval; |
1da177e4 LT |
3558 | |
3559 | if (!param || pid < 0) | |
3560 | return -EINVAL; | |
3561 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
3562 | return -EFAULT; | |
5fe1d75f ON |
3563 | |
3564 | rcu_read_lock(); | |
3565 | retval = -ESRCH; | |
1da177e4 | 3566 | p = find_process_by_pid(pid); |
5fe1d75f ON |
3567 | if (p != NULL) |
3568 | retval = sched_setscheduler(p, policy, &lparam); | |
3569 | rcu_read_unlock(); | |
36c8b586 | 3570 | |
1da177e4 LT |
3571 | return retval; |
3572 | } | |
3573 | ||
d50dde5a DF |
3574 | /* |
3575 | * Mimics kernel/events/core.c perf_copy_attr(). | |
3576 | */ | |
3577 | static int sched_copy_attr(struct sched_attr __user *uattr, | |
3578 | struct sched_attr *attr) | |
3579 | { | |
3580 | u32 size; | |
3581 | int ret; | |
3582 | ||
3583 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
3584 | return -EFAULT; | |
3585 | ||
3586 | /* | |
3587 | * zero the full structure, so that a short copy will be nice. | |
3588 | */ | |
3589 | memset(attr, 0, sizeof(*attr)); | |
3590 | ||
3591 | ret = get_user(size, &uattr->size); | |
3592 | if (ret) | |
3593 | return ret; | |
3594 | ||
3595 | if (size > PAGE_SIZE) /* silly large */ | |
3596 | goto err_size; | |
3597 | ||
3598 | if (!size) /* abi compat */ | |
3599 | size = SCHED_ATTR_SIZE_VER0; | |
3600 | ||
3601 | if (size < SCHED_ATTR_SIZE_VER0) | |
3602 | goto err_size; | |
3603 | ||
3604 | /* | |
3605 | * If we're handed a bigger struct than we know of, | |
3606 | * ensure all the unknown bits are 0 - i.e. new | |
3607 | * user-space does not rely on any kernel feature | |
3608 | * extensions we dont know about yet. | |
3609 | */ | |
3610 | if (size > sizeof(*attr)) { | |
3611 | unsigned char __user *addr; | |
3612 | unsigned char __user *end; | |
3613 | unsigned char val; | |
3614 | ||
3615 | addr = (void __user *)uattr + sizeof(*attr); | |
3616 | end = (void __user *)uattr + size; | |
3617 | ||
3618 | for (; addr < end; addr++) { | |
3619 | ret = get_user(val, addr); | |
3620 | if (ret) | |
3621 | return ret; | |
3622 | if (val) | |
3623 | goto err_size; | |
3624 | } | |
3625 | size = sizeof(*attr); | |
3626 | } | |
3627 | ||
3628 | ret = copy_from_user(attr, uattr, size); | |
3629 | if (ret) | |
3630 | return -EFAULT; | |
3631 | ||
3632 | /* | |
3633 | * XXX: do we want to be lenient like existing syscalls; or do we want | |
3634 | * to be strict and return an error on out-of-bounds values? | |
3635 | */ | |
75e45d51 | 3636 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 3637 | |
e78c7bca | 3638 | return 0; |
d50dde5a DF |
3639 | |
3640 | err_size: | |
3641 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 3642 | return -E2BIG; |
d50dde5a DF |
3643 | } |
3644 | ||
1da177e4 LT |
3645 | /** |
3646 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
3647 | * @pid: the pid in question. | |
3648 | * @policy: new policy. | |
3649 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3650 | * |
3651 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3652 | */ |
5add95d4 HC |
3653 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
3654 | struct sched_param __user *, param) | |
1da177e4 | 3655 | { |
c21761f1 JB |
3656 | /* negative values for policy are not valid */ |
3657 | if (policy < 0) | |
3658 | return -EINVAL; | |
3659 | ||
1da177e4 LT |
3660 | return do_sched_setscheduler(pid, policy, param); |
3661 | } | |
3662 | ||
3663 | /** | |
3664 | * sys_sched_setparam - set/change the RT priority of a thread | |
3665 | * @pid: the pid in question. | |
3666 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
3667 | * |
3668 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3669 | */ |
5add95d4 | 3670 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
3671 | { |
3672 | return do_sched_setscheduler(pid, -1, param); | |
3673 | } | |
3674 | ||
d50dde5a DF |
3675 | /** |
3676 | * sys_sched_setattr - same as above, but with extended sched_attr | |
3677 | * @pid: the pid in question. | |
5778fccf | 3678 | * @uattr: structure containing the extended parameters. |
db66d756 | 3679 | * @flags: for future extension. |
d50dde5a | 3680 | */ |
6d35ab48 PZ |
3681 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
3682 | unsigned int, flags) | |
d50dde5a DF |
3683 | { |
3684 | struct sched_attr attr; | |
3685 | struct task_struct *p; | |
3686 | int retval; | |
3687 | ||
6d35ab48 | 3688 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
3689 | return -EINVAL; |
3690 | ||
143cf23d MK |
3691 | retval = sched_copy_attr(uattr, &attr); |
3692 | if (retval) | |
3693 | return retval; | |
d50dde5a | 3694 | |
dbdb2275 PZ |
3695 | if (attr.sched_policy < 0) |
3696 | return -EINVAL; | |
d50dde5a DF |
3697 | |
3698 | rcu_read_lock(); | |
3699 | retval = -ESRCH; | |
3700 | p = find_process_by_pid(pid); | |
3701 | if (p != NULL) | |
3702 | retval = sched_setattr(p, &attr); | |
3703 | rcu_read_unlock(); | |
3704 | ||
3705 | return retval; | |
3706 | } | |
3707 | ||
1da177e4 LT |
3708 | /** |
3709 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
3710 | * @pid: the pid in question. | |
e69f6186 YB |
3711 | * |
3712 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
3713 | * code. | |
1da177e4 | 3714 | */ |
5add95d4 | 3715 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 3716 | { |
36c8b586 | 3717 | struct task_struct *p; |
3a5c359a | 3718 | int retval; |
1da177e4 LT |
3719 | |
3720 | if (pid < 0) | |
3a5c359a | 3721 | return -EINVAL; |
1da177e4 LT |
3722 | |
3723 | retval = -ESRCH; | |
5fe85be0 | 3724 | rcu_read_lock(); |
1da177e4 LT |
3725 | p = find_process_by_pid(pid); |
3726 | if (p) { | |
3727 | retval = security_task_getscheduler(p); | |
3728 | if (!retval) | |
ca94c442 LP |
3729 | retval = p->policy |
3730 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 3731 | } |
5fe85be0 | 3732 | rcu_read_unlock(); |
1da177e4 LT |
3733 | return retval; |
3734 | } | |
3735 | ||
3736 | /** | |
ca94c442 | 3737 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
3738 | * @pid: the pid in question. |
3739 | * @param: structure containing the RT priority. | |
e69f6186 YB |
3740 | * |
3741 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
3742 | * code. | |
1da177e4 | 3743 | */ |
5add95d4 | 3744 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 3745 | { |
ce5f7f82 | 3746 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 3747 | struct task_struct *p; |
3a5c359a | 3748 | int retval; |
1da177e4 LT |
3749 | |
3750 | if (!param || pid < 0) | |
3a5c359a | 3751 | return -EINVAL; |
1da177e4 | 3752 | |
5fe85be0 | 3753 | rcu_read_lock(); |
1da177e4 LT |
3754 | p = find_process_by_pid(pid); |
3755 | retval = -ESRCH; | |
3756 | if (!p) | |
3757 | goto out_unlock; | |
3758 | ||
3759 | retval = security_task_getscheduler(p); | |
3760 | if (retval) | |
3761 | goto out_unlock; | |
3762 | ||
ce5f7f82 PZ |
3763 | if (task_has_rt_policy(p)) |
3764 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 3765 | rcu_read_unlock(); |
1da177e4 LT |
3766 | |
3767 | /* | |
3768 | * This one might sleep, we cannot do it with a spinlock held ... | |
3769 | */ | |
3770 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
3771 | ||
1da177e4 LT |
3772 | return retval; |
3773 | ||
3774 | out_unlock: | |
5fe85be0 | 3775 | rcu_read_unlock(); |
1da177e4 LT |
3776 | return retval; |
3777 | } | |
3778 | ||
d50dde5a DF |
3779 | static int sched_read_attr(struct sched_attr __user *uattr, |
3780 | struct sched_attr *attr, | |
3781 | unsigned int usize) | |
3782 | { | |
3783 | int ret; | |
3784 | ||
3785 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
3786 | return -EFAULT; | |
3787 | ||
3788 | /* | |
3789 | * If we're handed a smaller struct than we know of, | |
3790 | * ensure all the unknown bits are 0 - i.e. old | |
3791 | * user-space does not get uncomplete information. | |
3792 | */ | |
3793 | if (usize < sizeof(*attr)) { | |
3794 | unsigned char *addr; | |
3795 | unsigned char *end; | |
3796 | ||
3797 | addr = (void *)attr + usize; | |
3798 | end = (void *)attr + sizeof(*attr); | |
3799 | ||
3800 | for (; addr < end; addr++) { | |
3801 | if (*addr) | |
22400674 | 3802 | return -EFBIG; |
d50dde5a DF |
3803 | } |
3804 | ||
3805 | attr->size = usize; | |
3806 | } | |
3807 | ||
4efbc454 | 3808 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
3809 | if (ret) |
3810 | return -EFAULT; | |
3811 | ||
22400674 | 3812 | return 0; |
d50dde5a DF |
3813 | } |
3814 | ||
3815 | /** | |
aab03e05 | 3816 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 3817 | * @pid: the pid in question. |
5778fccf | 3818 | * @uattr: structure containing the extended parameters. |
d50dde5a | 3819 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 3820 | * @flags: for future extension. |
d50dde5a | 3821 | */ |
6d35ab48 PZ |
3822 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
3823 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
3824 | { |
3825 | struct sched_attr attr = { | |
3826 | .size = sizeof(struct sched_attr), | |
3827 | }; | |
3828 | struct task_struct *p; | |
3829 | int retval; | |
3830 | ||
3831 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 3832 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
3833 | return -EINVAL; |
3834 | ||
3835 | rcu_read_lock(); | |
3836 | p = find_process_by_pid(pid); | |
3837 | retval = -ESRCH; | |
3838 | if (!p) | |
3839 | goto out_unlock; | |
3840 | ||
3841 | retval = security_task_getscheduler(p); | |
3842 | if (retval) | |
3843 | goto out_unlock; | |
3844 | ||
3845 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
3846 | if (p->sched_reset_on_fork) |
3847 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
3848 | if (task_has_dl_policy(p)) |
3849 | __getparam_dl(p, &attr); | |
3850 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
3851 | attr.sched_priority = p->rt_priority; |
3852 | else | |
d0ea0268 | 3853 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
3854 | |
3855 | rcu_read_unlock(); | |
3856 | ||
3857 | retval = sched_read_attr(uattr, &attr, size); | |
3858 | return retval; | |
3859 | ||
3860 | out_unlock: | |
3861 | rcu_read_unlock(); | |
3862 | return retval; | |
3863 | } | |
3864 | ||
96f874e2 | 3865 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 3866 | { |
5a16f3d3 | 3867 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
3868 | struct task_struct *p; |
3869 | int retval; | |
1da177e4 | 3870 | |
23f5d142 | 3871 | rcu_read_lock(); |
1da177e4 LT |
3872 | |
3873 | p = find_process_by_pid(pid); | |
3874 | if (!p) { | |
23f5d142 | 3875 | rcu_read_unlock(); |
1da177e4 LT |
3876 | return -ESRCH; |
3877 | } | |
3878 | ||
23f5d142 | 3879 | /* Prevent p going away */ |
1da177e4 | 3880 | get_task_struct(p); |
23f5d142 | 3881 | rcu_read_unlock(); |
1da177e4 | 3882 | |
14a40ffc TH |
3883 | if (p->flags & PF_NO_SETAFFINITY) { |
3884 | retval = -EINVAL; | |
3885 | goto out_put_task; | |
3886 | } | |
5a16f3d3 RR |
3887 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
3888 | retval = -ENOMEM; | |
3889 | goto out_put_task; | |
3890 | } | |
3891 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
3892 | retval = -ENOMEM; | |
3893 | goto out_free_cpus_allowed; | |
3894 | } | |
1da177e4 | 3895 | retval = -EPERM; |
4c44aaaf EB |
3896 | if (!check_same_owner(p)) { |
3897 | rcu_read_lock(); | |
3898 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
3899 | rcu_read_unlock(); | |
3900 | goto out_unlock; | |
3901 | } | |
3902 | rcu_read_unlock(); | |
3903 | } | |
1da177e4 | 3904 | |
b0ae1981 | 3905 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
3906 | if (retval) |
3907 | goto out_unlock; | |
3908 | ||
e4099a5e PZ |
3909 | |
3910 | cpuset_cpus_allowed(p, cpus_allowed); | |
3911 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
3912 | ||
332ac17e DF |
3913 | /* |
3914 | * Since bandwidth control happens on root_domain basis, | |
3915 | * if admission test is enabled, we only admit -deadline | |
3916 | * tasks allowed to run on all the CPUs in the task's | |
3917 | * root_domain. | |
3918 | */ | |
3919 | #ifdef CONFIG_SMP | |
3920 | if (task_has_dl_policy(p)) { | |
3921 | const struct cpumask *span = task_rq(p)->rd->span; | |
3922 | ||
e4099a5e | 3923 | if (dl_bandwidth_enabled() && !cpumask_subset(span, new_mask)) { |
332ac17e DF |
3924 | retval = -EBUSY; |
3925 | goto out_unlock; | |
3926 | } | |
3927 | } | |
3928 | #endif | |
49246274 | 3929 | again: |
5a16f3d3 | 3930 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 3931 | |
8707d8b8 | 3932 | if (!retval) { |
5a16f3d3 RR |
3933 | cpuset_cpus_allowed(p, cpus_allowed); |
3934 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
3935 | /* |
3936 | * We must have raced with a concurrent cpuset | |
3937 | * update. Just reset the cpus_allowed to the | |
3938 | * cpuset's cpus_allowed | |
3939 | */ | |
5a16f3d3 | 3940 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
3941 | goto again; |
3942 | } | |
3943 | } | |
1da177e4 | 3944 | out_unlock: |
5a16f3d3 RR |
3945 | free_cpumask_var(new_mask); |
3946 | out_free_cpus_allowed: | |
3947 | free_cpumask_var(cpus_allowed); | |
3948 | out_put_task: | |
1da177e4 | 3949 | put_task_struct(p); |
1da177e4 LT |
3950 | return retval; |
3951 | } | |
3952 | ||
3953 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 3954 | struct cpumask *new_mask) |
1da177e4 | 3955 | { |
96f874e2 RR |
3956 | if (len < cpumask_size()) |
3957 | cpumask_clear(new_mask); | |
3958 | else if (len > cpumask_size()) | |
3959 | len = cpumask_size(); | |
3960 | ||
1da177e4 LT |
3961 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
3962 | } | |
3963 | ||
3964 | /** | |
3965 | * sys_sched_setaffinity - set the cpu affinity of a process | |
3966 | * @pid: pid of the process | |
3967 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
3968 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
e69f6186 YB |
3969 | * |
3970 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 3971 | */ |
5add95d4 HC |
3972 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
3973 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 3974 | { |
5a16f3d3 | 3975 | cpumask_var_t new_mask; |
1da177e4 LT |
3976 | int retval; |
3977 | ||
5a16f3d3 RR |
3978 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
3979 | return -ENOMEM; | |
1da177e4 | 3980 | |
5a16f3d3 RR |
3981 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
3982 | if (retval == 0) | |
3983 | retval = sched_setaffinity(pid, new_mask); | |
3984 | free_cpumask_var(new_mask); | |
3985 | return retval; | |
1da177e4 LT |
3986 | } |
3987 | ||
96f874e2 | 3988 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 3989 | { |
36c8b586 | 3990 | struct task_struct *p; |
31605683 | 3991 | unsigned long flags; |
1da177e4 | 3992 | int retval; |
1da177e4 | 3993 | |
23f5d142 | 3994 | rcu_read_lock(); |
1da177e4 LT |
3995 | |
3996 | retval = -ESRCH; | |
3997 | p = find_process_by_pid(pid); | |
3998 | if (!p) | |
3999 | goto out_unlock; | |
4000 | ||
e7834f8f DQ |
4001 | retval = security_task_getscheduler(p); |
4002 | if (retval) | |
4003 | goto out_unlock; | |
4004 | ||
013fdb80 | 4005 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4006 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4007 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4008 | |
4009 | out_unlock: | |
23f5d142 | 4010 | rcu_read_unlock(); |
1da177e4 | 4011 | |
9531b62f | 4012 | return retval; |
1da177e4 LT |
4013 | } |
4014 | ||
4015 | /** | |
4016 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4017 | * @pid: pid of the process | |
4018 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4019 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
e69f6186 YB |
4020 | * |
4021 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4022 | */ |
5add95d4 HC |
4023 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4024 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4025 | { |
4026 | int ret; | |
f17c8607 | 4027 | cpumask_var_t mask; |
1da177e4 | 4028 | |
84fba5ec | 4029 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4030 | return -EINVAL; |
4031 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4032 | return -EINVAL; |
4033 | ||
f17c8607 RR |
4034 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4035 | return -ENOMEM; | |
1da177e4 | 4036 | |
f17c8607 RR |
4037 | ret = sched_getaffinity(pid, mask); |
4038 | if (ret == 0) { | |
8bc037fb | 4039 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4040 | |
4041 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4042 | ret = -EFAULT; |
4043 | else | |
cd3d8031 | 4044 | ret = retlen; |
f17c8607 RR |
4045 | } |
4046 | free_cpumask_var(mask); | |
1da177e4 | 4047 | |
f17c8607 | 4048 | return ret; |
1da177e4 LT |
4049 | } |
4050 | ||
4051 | /** | |
4052 | * sys_sched_yield - yield the current processor to other threads. | |
4053 | * | |
dd41f596 IM |
4054 | * This function yields the current CPU to other tasks. If there are no |
4055 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4056 | * |
4057 | * Return: 0. | |
1da177e4 | 4058 | */ |
5add95d4 | 4059 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4060 | { |
70b97a7f | 4061 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4062 | |
2d72376b | 4063 | schedstat_inc(rq, yld_count); |
4530d7ab | 4064 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4065 | |
4066 | /* | |
4067 | * Since we are going to call schedule() anyway, there's | |
4068 | * no need to preempt or enable interrupts: | |
4069 | */ | |
4070 | __release(rq->lock); | |
8a25d5de | 4071 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4072 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4073 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4074 | |
4075 | schedule(); | |
4076 | ||
4077 | return 0; | |
4078 | } | |
4079 | ||
e7b38404 | 4080 | static void __cond_resched(void) |
1da177e4 | 4081 | { |
bdb43806 | 4082 | __preempt_count_add(PREEMPT_ACTIVE); |
c259e01a | 4083 | __schedule(); |
bdb43806 | 4084 | __preempt_count_sub(PREEMPT_ACTIVE); |
1da177e4 LT |
4085 | } |
4086 | ||
02b67cc3 | 4087 | int __sched _cond_resched(void) |
1da177e4 | 4088 | { |
d86ee480 | 4089 | if (should_resched()) { |
1da177e4 LT |
4090 | __cond_resched(); |
4091 | return 1; | |
4092 | } | |
4093 | return 0; | |
4094 | } | |
02b67cc3 | 4095 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4096 | |
4097 | /* | |
613afbf8 | 4098 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4099 | * call schedule, and on return reacquire the lock. |
4100 | * | |
41a2d6cf | 4101 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4102 | * operations here to prevent schedule() from being called twice (once via |
4103 | * spin_unlock(), once by hand). | |
4104 | */ | |
613afbf8 | 4105 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4106 | { |
d86ee480 | 4107 | int resched = should_resched(); |
6df3cecb JK |
4108 | int ret = 0; |
4109 | ||
f607c668 PZ |
4110 | lockdep_assert_held(lock); |
4111 | ||
95c354fe | 4112 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4113 | spin_unlock(lock); |
d86ee480 | 4114 | if (resched) |
95c354fe NP |
4115 | __cond_resched(); |
4116 | else | |
4117 | cpu_relax(); | |
6df3cecb | 4118 | ret = 1; |
1da177e4 | 4119 | spin_lock(lock); |
1da177e4 | 4120 | } |
6df3cecb | 4121 | return ret; |
1da177e4 | 4122 | } |
613afbf8 | 4123 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4124 | |
613afbf8 | 4125 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4126 | { |
4127 | BUG_ON(!in_softirq()); | |
4128 | ||
d86ee480 | 4129 | if (should_resched()) { |
98d82567 | 4130 | local_bh_enable(); |
1da177e4 LT |
4131 | __cond_resched(); |
4132 | local_bh_disable(); | |
4133 | return 1; | |
4134 | } | |
4135 | return 0; | |
4136 | } | |
613afbf8 | 4137 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4138 | |
1da177e4 LT |
4139 | /** |
4140 | * yield - yield the current processor to other threads. | |
4141 | * | |
8e3fabfd PZ |
4142 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4143 | * | |
4144 | * The scheduler is at all times free to pick the calling task as the most | |
4145 | * eligible task to run, if removing the yield() call from your code breaks | |
4146 | * it, its already broken. | |
4147 | * | |
4148 | * Typical broken usage is: | |
4149 | * | |
4150 | * while (!event) | |
4151 | * yield(); | |
4152 | * | |
4153 | * where one assumes that yield() will let 'the other' process run that will | |
4154 | * make event true. If the current task is a SCHED_FIFO task that will never | |
4155 | * happen. Never use yield() as a progress guarantee!! | |
4156 | * | |
4157 | * If you want to use yield() to wait for something, use wait_event(). | |
4158 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
4159 | * If you still want to use yield(), do not! | |
1da177e4 LT |
4160 | */ |
4161 | void __sched yield(void) | |
4162 | { | |
4163 | set_current_state(TASK_RUNNING); | |
4164 | sys_sched_yield(); | |
4165 | } | |
1da177e4 LT |
4166 | EXPORT_SYMBOL(yield); |
4167 | ||
d95f4122 MG |
4168 | /** |
4169 | * yield_to - yield the current processor to another thread in | |
4170 | * your thread group, or accelerate that thread toward the | |
4171 | * processor it's on. | |
16addf95 RD |
4172 | * @p: target task |
4173 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
4174 | * |
4175 | * It's the caller's job to ensure that the target task struct | |
4176 | * can't go away on us before we can do any checks. | |
4177 | * | |
e69f6186 | 4178 | * Return: |
7b270f60 PZ |
4179 | * true (>0) if we indeed boosted the target task. |
4180 | * false (0) if we failed to boost the target. | |
4181 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 4182 | */ |
fa93384f | 4183 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
4184 | { |
4185 | struct task_struct *curr = current; | |
4186 | struct rq *rq, *p_rq; | |
4187 | unsigned long flags; | |
c3c18640 | 4188 | int yielded = 0; |
d95f4122 MG |
4189 | |
4190 | local_irq_save(flags); | |
4191 | rq = this_rq(); | |
4192 | ||
4193 | again: | |
4194 | p_rq = task_rq(p); | |
7b270f60 PZ |
4195 | /* |
4196 | * If we're the only runnable task on the rq and target rq also | |
4197 | * has only one task, there's absolutely no point in yielding. | |
4198 | */ | |
4199 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
4200 | yielded = -ESRCH; | |
4201 | goto out_irq; | |
4202 | } | |
4203 | ||
d95f4122 | 4204 | double_rq_lock(rq, p_rq); |
39e24d8f | 4205 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
4206 | double_rq_unlock(rq, p_rq); |
4207 | goto again; | |
4208 | } | |
4209 | ||
4210 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 4211 | goto out_unlock; |
d95f4122 MG |
4212 | |
4213 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 4214 | goto out_unlock; |
d95f4122 MG |
4215 | |
4216 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 4217 | goto out_unlock; |
d95f4122 MG |
4218 | |
4219 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 4220 | if (yielded) { |
d95f4122 | 4221 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
4222 | /* |
4223 | * Make p's CPU reschedule; pick_next_entity takes care of | |
4224 | * fairness. | |
4225 | */ | |
4226 | if (preempt && rq != p_rq) | |
4227 | resched_task(p_rq->curr); | |
4228 | } | |
d95f4122 | 4229 | |
7b270f60 | 4230 | out_unlock: |
d95f4122 | 4231 | double_rq_unlock(rq, p_rq); |
7b270f60 | 4232 | out_irq: |
d95f4122 MG |
4233 | local_irq_restore(flags); |
4234 | ||
7b270f60 | 4235 | if (yielded > 0) |
d95f4122 MG |
4236 | schedule(); |
4237 | ||
4238 | return yielded; | |
4239 | } | |
4240 | EXPORT_SYMBOL_GPL(yield_to); | |
4241 | ||
1da177e4 | 4242 | /* |
41a2d6cf | 4243 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4244 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4245 | */ |
4246 | void __sched io_schedule(void) | |
4247 | { | |
54d35f29 | 4248 | struct rq *rq = raw_rq(); |
1da177e4 | 4249 | |
0ff92245 | 4250 | delayacct_blkio_start(); |
1da177e4 | 4251 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4252 | blk_flush_plug(current); |
8f0dfc34 | 4253 | current->in_iowait = 1; |
1da177e4 | 4254 | schedule(); |
8f0dfc34 | 4255 | current->in_iowait = 0; |
1da177e4 | 4256 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4257 | delayacct_blkio_end(); |
1da177e4 | 4258 | } |
1da177e4 LT |
4259 | EXPORT_SYMBOL(io_schedule); |
4260 | ||
4261 | long __sched io_schedule_timeout(long timeout) | |
4262 | { | |
54d35f29 | 4263 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4264 | long ret; |
4265 | ||
0ff92245 | 4266 | delayacct_blkio_start(); |
1da177e4 | 4267 | atomic_inc(&rq->nr_iowait); |
73c10101 | 4268 | blk_flush_plug(current); |
8f0dfc34 | 4269 | current->in_iowait = 1; |
1da177e4 | 4270 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4271 | current->in_iowait = 0; |
1da177e4 | 4272 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4273 | delayacct_blkio_end(); |
1da177e4 LT |
4274 | return ret; |
4275 | } | |
4276 | ||
4277 | /** | |
4278 | * sys_sched_get_priority_max - return maximum RT priority. | |
4279 | * @policy: scheduling class. | |
4280 | * | |
e69f6186 YB |
4281 | * Return: On success, this syscall returns the maximum |
4282 | * rt_priority that can be used by a given scheduling class. | |
4283 | * On failure, a negative error code is returned. | |
1da177e4 | 4284 | */ |
5add95d4 | 4285 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4286 | { |
4287 | int ret = -EINVAL; | |
4288 | ||
4289 | switch (policy) { | |
4290 | case SCHED_FIFO: | |
4291 | case SCHED_RR: | |
4292 | ret = MAX_USER_RT_PRIO-1; | |
4293 | break; | |
aab03e05 | 4294 | case SCHED_DEADLINE: |
1da177e4 | 4295 | case SCHED_NORMAL: |
b0a9499c | 4296 | case SCHED_BATCH: |
dd41f596 | 4297 | case SCHED_IDLE: |
1da177e4 LT |
4298 | ret = 0; |
4299 | break; | |
4300 | } | |
4301 | return ret; | |
4302 | } | |
4303 | ||
4304 | /** | |
4305 | * sys_sched_get_priority_min - return minimum RT priority. | |
4306 | * @policy: scheduling class. | |
4307 | * | |
e69f6186 YB |
4308 | * Return: On success, this syscall returns the minimum |
4309 | * rt_priority that can be used by a given scheduling class. | |
4310 | * On failure, a negative error code is returned. | |
1da177e4 | 4311 | */ |
5add95d4 | 4312 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
4313 | { |
4314 | int ret = -EINVAL; | |
4315 | ||
4316 | switch (policy) { | |
4317 | case SCHED_FIFO: | |
4318 | case SCHED_RR: | |
4319 | ret = 1; | |
4320 | break; | |
aab03e05 | 4321 | case SCHED_DEADLINE: |
1da177e4 | 4322 | case SCHED_NORMAL: |
b0a9499c | 4323 | case SCHED_BATCH: |
dd41f596 | 4324 | case SCHED_IDLE: |
1da177e4 LT |
4325 | ret = 0; |
4326 | } | |
4327 | return ret; | |
4328 | } | |
4329 | ||
4330 | /** | |
4331 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4332 | * @pid: pid of the process. | |
4333 | * @interval: userspace pointer to the timeslice value. | |
4334 | * | |
4335 | * this syscall writes the default timeslice value of a given process | |
4336 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
4337 | * |
4338 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
4339 | * an error code. | |
1da177e4 | 4340 | */ |
17da2bd9 | 4341 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 4342 | struct timespec __user *, interval) |
1da177e4 | 4343 | { |
36c8b586 | 4344 | struct task_struct *p; |
a4ec24b4 | 4345 | unsigned int time_slice; |
dba091b9 TG |
4346 | unsigned long flags; |
4347 | struct rq *rq; | |
3a5c359a | 4348 | int retval; |
1da177e4 | 4349 | struct timespec t; |
1da177e4 LT |
4350 | |
4351 | if (pid < 0) | |
3a5c359a | 4352 | return -EINVAL; |
1da177e4 LT |
4353 | |
4354 | retval = -ESRCH; | |
1a551ae7 | 4355 | rcu_read_lock(); |
1da177e4 LT |
4356 | p = find_process_by_pid(pid); |
4357 | if (!p) | |
4358 | goto out_unlock; | |
4359 | ||
4360 | retval = security_task_getscheduler(p); | |
4361 | if (retval) | |
4362 | goto out_unlock; | |
4363 | ||
dba091b9 | 4364 | rq = task_rq_lock(p, &flags); |
a57beec5 PZ |
4365 | time_slice = 0; |
4366 | if (p->sched_class->get_rr_interval) | |
4367 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 4368 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 4369 | |
1a551ae7 | 4370 | rcu_read_unlock(); |
a4ec24b4 | 4371 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4372 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4373 | return retval; |
3a5c359a | 4374 | |
1da177e4 | 4375 | out_unlock: |
1a551ae7 | 4376 | rcu_read_unlock(); |
1da177e4 LT |
4377 | return retval; |
4378 | } | |
4379 | ||
7c731e0a | 4380 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 4381 | |
82a1fcb9 | 4382 | void sched_show_task(struct task_struct *p) |
1da177e4 | 4383 | { |
1da177e4 | 4384 | unsigned long free = 0; |
4e79752c | 4385 | int ppid; |
36c8b586 | 4386 | unsigned state; |
1da177e4 | 4387 | |
1da177e4 | 4388 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 4389 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 4390 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4391 | #if BITS_PER_LONG == 32 |
1da177e4 | 4392 | if (state == TASK_RUNNING) |
3df0fc5b | 4393 | printk(KERN_CONT " running "); |
1da177e4 | 4394 | else |
3df0fc5b | 4395 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4396 | #else |
4397 | if (state == TASK_RUNNING) | |
3df0fc5b | 4398 | printk(KERN_CONT " running task "); |
1da177e4 | 4399 | else |
3df0fc5b | 4400 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4401 | #endif |
4402 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 4403 | free = stack_not_used(p); |
1da177e4 | 4404 | #endif |
4e79752c PM |
4405 | rcu_read_lock(); |
4406 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4407 | rcu_read_unlock(); | |
3df0fc5b | 4408 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 4409 | task_pid_nr(p), ppid, |
aa47b7e0 | 4410 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 4411 | |
3d1cb205 | 4412 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 4413 | show_stack(p, NULL); |
1da177e4 LT |
4414 | } |
4415 | ||
e59e2ae2 | 4416 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4417 | { |
36c8b586 | 4418 | struct task_struct *g, *p; |
1da177e4 | 4419 | |
4bd77321 | 4420 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
4421 | printk(KERN_INFO |
4422 | " task PC stack pid father\n"); | |
1da177e4 | 4423 | #else |
3df0fc5b PZ |
4424 | printk(KERN_INFO |
4425 | " task PC stack pid father\n"); | |
1da177e4 | 4426 | #endif |
510f5acc | 4427 | rcu_read_lock(); |
1da177e4 LT |
4428 | do_each_thread(g, p) { |
4429 | /* | |
4430 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 4431 | * console might take a lot of time: |
1da177e4 LT |
4432 | */ |
4433 | touch_nmi_watchdog(); | |
39bc89fd | 4434 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 4435 | sched_show_task(p); |
1da177e4 LT |
4436 | } while_each_thread(g, p); |
4437 | ||
04c9167f JF |
4438 | touch_all_softlockup_watchdogs(); |
4439 | ||
dd41f596 IM |
4440 | #ifdef CONFIG_SCHED_DEBUG |
4441 | sysrq_sched_debug_show(); | |
4442 | #endif | |
510f5acc | 4443 | rcu_read_unlock(); |
e59e2ae2 IM |
4444 | /* |
4445 | * Only show locks if all tasks are dumped: | |
4446 | */ | |
93335a21 | 4447 | if (!state_filter) |
e59e2ae2 | 4448 | debug_show_all_locks(); |
1da177e4 LT |
4449 | } |
4450 | ||
0db0628d | 4451 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 4452 | { |
dd41f596 | 4453 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4454 | } |
4455 | ||
f340c0d1 IM |
4456 | /** |
4457 | * init_idle - set up an idle thread for a given CPU | |
4458 | * @idle: task in question | |
4459 | * @cpu: cpu the idle task belongs to | |
4460 | * | |
4461 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4462 | * flag, to make booting more robust. | |
4463 | */ | |
0db0628d | 4464 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4465 | { |
70b97a7f | 4466 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4467 | unsigned long flags; |
4468 | ||
05fa785c | 4469 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 4470 | |
5e1576ed | 4471 | __sched_fork(0, idle); |
06b83b5f | 4472 | idle->state = TASK_RUNNING; |
dd41f596 IM |
4473 | idle->se.exec_start = sched_clock(); |
4474 | ||
1e1b6c51 | 4475 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
4476 | /* |
4477 | * We're having a chicken and egg problem, even though we are | |
4478 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
4479 | * lockdep check in task_group() will fail. | |
4480 | * | |
4481 | * Similar case to sched_fork(). / Alternatively we could | |
4482 | * use task_rq_lock() here and obtain the other rq->lock. | |
4483 | * | |
4484 | * Silence PROVE_RCU | |
4485 | */ | |
4486 | rcu_read_lock(); | |
dd41f596 | 4487 | __set_task_cpu(idle, cpu); |
6506cf6c | 4488 | rcu_read_unlock(); |
1da177e4 | 4489 | |
1da177e4 | 4490 | rq->curr = rq->idle = idle; |
77177856 | 4491 | idle->on_rq = 1; |
3ca7a440 PZ |
4492 | #if defined(CONFIG_SMP) |
4493 | idle->on_cpu = 1; | |
4866cde0 | 4494 | #endif |
05fa785c | 4495 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
4496 | |
4497 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 4498 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 4499 | |
dd41f596 IM |
4500 | /* |
4501 | * The idle tasks have their own, simple scheduling class: | |
4502 | */ | |
4503 | idle->sched_class = &idle_sched_class; | |
868baf07 | 4504 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 4505 | vtime_init_idle(idle, cpu); |
f1c6f1a7 CE |
4506 | #if defined(CONFIG_SMP) |
4507 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); | |
4508 | #endif | |
19978ca6 IM |
4509 | } |
4510 | ||
1da177e4 | 4511 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
4512 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
4513 | { | |
4514 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
4515 | p->sched_class->set_cpus_allowed(p, new_mask); | |
4939602a PZ |
4516 | |
4517 | cpumask_copy(&p->cpus_allowed, new_mask); | |
29baa747 | 4518 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1e1b6c51 KM |
4519 | } |
4520 | ||
1da177e4 LT |
4521 | /* |
4522 | * This is how migration works: | |
4523 | * | |
969c7921 TH |
4524 | * 1) we invoke migration_cpu_stop() on the target CPU using |
4525 | * stop_one_cpu(). | |
4526 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
4527 | * off the CPU) | |
4528 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
4529 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 4530 | * it and puts it into the right queue. |
969c7921 TH |
4531 | * 5) stopper completes and stop_one_cpu() returns and the migration |
4532 | * is done. | |
1da177e4 LT |
4533 | */ |
4534 | ||
4535 | /* | |
4536 | * Change a given task's CPU affinity. Migrate the thread to a | |
4537 | * proper CPU and schedule it away if the CPU it's executing on | |
4538 | * is removed from the allowed bitmask. | |
4539 | * | |
4540 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 4541 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
4542 | * call is not atomic; no spinlocks may be held. |
4543 | */ | |
96f874e2 | 4544 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
4545 | { |
4546 | unsigned long flags; | |
70b97a7f | 4547 | struct rq *rq; |
969c7921 | 4548 | unsigned int dest_cpu; |
48f24c4d | 4549 | int ret = 0; |
1da177e4 LT |
4550 | |
4551 | rq = task_rq_lock(p, &flags); | |
e2912009 | 4552 | |
db44fc01 YZ |
4553 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
4554 | goto out; | |
4555 | ||
6ad4c188 | 4556 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
4557 | ret = -EINVAL; |
4558 | goto out; | |
4559 | } | |
4560 | ||
1e1b6c51 | 4561 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 4562 | |
1da177e4 | 4563 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 4564 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
4565 | goto out; |
4566 | ||
969c7921 | 4567 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 4568 | if (p->on_rq) { |
969c7921 | 4569 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 4570 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 4571 | task_rq_unlock(rq, p, &flags); |
969c7921 | 4572 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
4573 | tlb_migrate_finish(p->mm); |
4574 | return 0; | |
4575 | } | |
4576 | out: | |
0122ec5b | 4577 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 4578 | |
1da177e4 LT |
4579 | return ret; |
4580 | } | |
cd8ba7cd | 4581 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
4582 | |
4583 | /* | |
41a2d6cf | 4584 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
4585 | * this because either it can't run here any more (set_cpus_allowed() |
4586 | * away from this CPU, or CPU going down), or because we're | |
4587 | * attempting to rebalance this task on exec (sched_exec). | |
4588 | * | |
4589 | * So we race with normal scheduler movements, but that's OK, as long | |
4590 | * as the task is no longer on this CPU. | |
efc30814 KK |
4591 | * |
4592 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4593 | */ |
efc30814 | 4594 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4595 | { |
70b97a7f | 4596 | struct rq *rq_dest, *rq_src; |
e2912009 | 4597 | int ret = 0; |
1da177e4 | 4598 | |
e761b772 | 4599 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 4600 | return ret; |
1da177e4 LT |
4601 | |
4602 | rq_src = cpu_rq(src_cpu); | |
4603 | rq_dest = cpu_rq(dest_cpu); | |
4604 | ||
0122ec5b | 4605 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
4606 | double_rq_lock(rq_src, rq_dest); |
4607 | /* Already moved. */ | |
4608 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 4609 | goto done; |
1da177e4 | 4610 | /* Affinity changed (again). */ |
fa17b507 | 4611 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
b1e38734 | 4612 | goto fail; |
1da177e4 | 4613 | |
e2912009 PZ |
4614 | /* |
4615 | * If we're not on a rq, the next wake-up will ensure we're | |
4616 | * placed properly. | |
4617 | */ | |
fd2f4419 | 4618 | if (p->on_rq) { |
4ca9b72b | 4619 | dequeue_task(rq_src, p, 0); |
e2912009 | 4620 | set_task_cpu(p, dest_cpu); |
4ca9b72b | 4621 | enqueue_task(rq_dest, p, 0); |
15afe09b | 4622 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 4623 | } |
b1e38734 | 4624 | done: |
efc30814 | 4625 | ret = 1; |
b1e38734 | 4626 | fail: |
1da177e4 | 4627 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 4628 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 4629 | return ret; |
1da177e4 LT |
4630 | } |
4631 | ||
e6628d5b MG |
4632 | #ifdef CONFIG_NUMA_BALANCING |
4633 | /* Migrate current task p to target_cpu */ | |
4634 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
4635 | { | |
4636 | struct migration_arg arg = { p, target_cpu }; | |
4637 | int curr_cpu = task_cpu(p); | |
4638 | ||
4639 | if (curr_cpu == target_cpu) | |
4640 | return 0; | |
4641 | ||
4642 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) | |
4643 | return -EINVAL; | |
4644 | ||
4645 | /* TODO: This is not properly updating schedstats */ | |
4646 | ||
286549dc | 4647 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
4648 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
4649 | } | |
0ec8aa00 PZ |
4650 | |
4651 | /* | |
4652 | * Requeue a task on a given node and accurately track the number of NUMA | |
4653 | * tasks on the runqueues | |
4654 | */ | |
4655 | void sched_setnuma(struct task_struct *p, int nid) | |
4656 | { | |
4657 | struct rq *rq; | |
4658 | unsigned long flags; | |
4659 | bool on_rq, running; | |
4660 | ||
4661 | rq = task_rq_lock(p, &flags); | |
4662 | on_rq = p->on_rq; | |
4663 | running = task_current(rq, p); | |
4664 | ||
4665 | if (on_rq) | |
4666 | dequeue_task(rq, p, 0); | |
4667 | if (running) | |
4668 | p->sched_class->put_prev_task(rq, p); | |
4669 | ||
4670 | p->numa_preferred_nid = nid; | |
0ec8aa00 PZ |
4671 | |
4672 | if (running) | |
4673 | p->sched_class->set_curr_task(rq); | |
4674 | if (on_rq) | |
4675 | enqueue_task(rq, p, 0); | |
4676 | task_rq_unlock(rq, p, &flags); | |
4677 | } | |
e6628d5b MG |
4678 | #endif |
4679 | ||
1da177e4 | 4680 | /* |
969c7921 TH |
4681 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
4682 | * and performs thread migration by bumping thread off CPU then | |
4683 | * 'pushing' onto another runqueue. | |
1da177e4 | 4684 | */ |
969c7921 | 4685 | static int migration_cpu_stop(void *data) |
1da177e4 | 4686 | { |
969c7921 | 4687 | struct migration_arg *arg = data; |
f7b4cddc | 4688 | |
969c7921 TH |
4689 | /* |
4690 | * The original target cpu might have gone down and we might | |
4691 | * be on another cpu but it doesn't matter. | |
4692 | */ | |
f7b4cddc | 4693 | local_irq_disable(); |
969c7921 | 4694 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 4695 | local_irq_enable(); |
1da177e4 | 4696 | return 0; |
f7b4cddc ON |
4697 | } |
4698 | ||
1da177e4 | 4699 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 4700 | |
054b9108 | 4701 | /* |
48c5ccae PZ |
4702 | * Ensures that the idle task is using init_mm right before its cpu goes |
4703 | * offline. | |
054b9108 | 4704 | */ |
48c5ccae | 4705 | void idle_task_exit(void) |
1da177e4 | 4706 | { |
48c5ccae | 4707 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 4708 | |
48c5ccae | 4709 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 4710 | |
a53efe5f | 4711 | if (mm != &init_mm) { |
48c5ccae | 4712 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
4713 | finish_arch_post_lock_switch(); |
4714 | } | |
48c5ccae | 4715 | mmdrop(mm); |
1da177e4 LT |
4716 | } |
4717 | ||
4718 | /* | |
5d180232 PZ |
4719 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
4720 | * we might have. Assumes we're called after migrate_tasks() so that the | |
4721 | * nr_active count is stable. | |
4722 | * | |
4723 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 4724 | */ |
5d180232 | 4725 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 4726 | { |
5d180232 PZ |
4727 | long delta = calc_load_fold_active(rq); |
4728 | if (delta) | |
4729 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
4730 | } |
4731 | ||
3f1d2a31 PZ |
4732 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
4733 | { | |
4734 | } | |
4735 | ||
4736 | static const struct sched_class fake_sched_class = { | |
4737 | .put_prev_task = put_prev_task_fake, | |
4738 | }; | |
4739 | ||
4740 | static struct task_struct fake_task = { | |
4741 | /* | |
4742 | * Avoid pull_{rt,dl}_task() | |
4743 | */ | |
4744 | .prio = MAX_PRIO + 1, | |
4745 | .sched_class = &fake_sched_class, | |
4746 | }; | |
4747 | ||
48f24c4d | 4748 | /* |
48c5ccae PZ |
4749 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
4750 | * try_to_wake_up()->select_task_rq(). | |
4751 | * | |
4752 | * Called with rq->lock held even though we'er in stop_machine() and | |
4753 | * there's no concurrency possible, we hold the required locks anyway | |
4754 | * because of lock validation efforts. | |
1da177e4 | 4755 | */ |
48c5ccae | 4756 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 4757 | { |
70b97a7f | 4758 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
4759 | struct task_struct *next, *stop = rq->stop; |
4760 | int dest_cpu; | |
1da177e4 LT |
4761 | |
4762 | /* | |
48c5ccae PZ |
4763 | * Fudge the rq selection such that the below task selection loop |
4764 | * doesn't get stuck on the currently eligible stop task. | |
4765 | * | |
4766 | * We're currently inside stop_machine() and the rq is either stuck | |
4767 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
4768 | * either way we should never end up calling schedule() until we're | |
4769 | * done here. | |
1da177e4 | 4770 | */ |
48c5ccae | 4771 | rq->stop = NULL; |
48f24c4d | 4772 | |
77bd3970 FW |
4773 | /* |
4774 | * put_prev_task() and pick_next_task() sched | |
4775 | * class method both need to have an up-to-date | |
4776 | * value of rq->clock[_task] | |
4777 | */ | |
4778 | update_rq_clock(rq); | |
4779 | ||
dd41f596 | 4780 | for ( ; ; ) { |
48c5ccae PZ |
4781 | /* |
4782 | * There's this thread running, bail when that's the only | |
4783 | * remaining thread. | |
4784 | */ | |
4785 | if (rq->nr_running == 1) | |
dd41f596 | 4786 | break; |
48c5ccae | 4787 | |
3f1d2a31 | 4788 | next = pick_next_task(rq, &fake_task); |
48c5ccae | 4789 | BUG_ON(!next); |
79c53799 | 4790 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 4791 | |
48c5ccae PZ |
4792 | /* Find suitable destination for @next, with force if needed. */ |
4793 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
4794 | raw_spin_unlock(&rq->lock); | |
4795 | ||
4796 | __migrate_task(next, dead_cpu, dest_cpu); | |
4797 | ||
4798 | raw_spin_lock(&rq->lock); | |
1da177e4 | 4799 | } |
dce48a84 | 4800 | |
48c5ccae | 4801 | rq->stop = stop; |
dce48a84 | 4802 | } |
48c5ccae | 4803 | |
1da177e4 LT |
4804 | #endif /* CONFIG_HOTPLUG_CPU */ |
4805 | ||
e692ab53 NP |
4806 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
4807 | ||
4808 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
4809 | { |
4810 | .procname = "sched_domain", | |
c57baf1e | 4811 | .mode = 0555, |
e0361851 | 4812 | }, |
56992309 | 4813 | {} |
e692ab53 NP |
4814 | }; |
4815 | ||
4816 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
4817 | { |
4818 | .procname = "kernel", | |
c57baf1e | 4819 | .mode = 0555, |
e0361851 AD |
4820 | .child = sd_ctl_dir, |
4821 | }, | |
56992309 | 4822 | {} |
e692ab53 NP |
4823 | }; |
4824 | ||
4825 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
4826 | { | |
4827 | struct ctl_table *entry = | |
5cf9f062 | 4828 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 4829 | |
e692ab53 NP |
4830 | return entry; |
4831 | } | |
4832 | ||
6382bc90 MM |
4833 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
4834 | { | |
cd790076 | 4835 | struct ctl_table *entry; |
6382bc90 | 4836 | |
cd790076 MM |
4837 | /* |
4838 | * In the intermediate directories, both the child directory and | |
4839 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 4840 | * will always be set. In the lowest directory the names are |
cd790076 MM |
4841 | * static strings and all have proc handlers. |
4842 | */ | |
4843 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
4844 | if (entry->child) |
4845 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
4846 | if (entry->proc_handler == NULL) |
4847 | kfree(entry->procname); | |
4848 | } | |
6382bc90 MM |
4849 | |
4850 | kfree(*tablep); | |
4851 | *tablep = NULL; | |
4852 | } | |
4853 | ||
201c373e | 4854 | static int min_load_idx = 0; |
fd9b86d3 | 4855 | static int max_load_idx = CPU_LOAD_IDX_MAX-1; |
201c373e | 4856 | |
e692ab53 | 4857 | static void |
e0361851 | 4858 | set_table_entry(struct ctl_table *entry, |
e692ab53 | 4859 | const char *procname, void *data, int maxlen, |
201c373e NK |
4860 | umode_t mode, proc_handler *proc_handler, |
4861 | bool load_idx) | |
e692ab53 | 4862 | { |
e692ab53 NP |
4863 | entry->procname = procname; |
4864 | entry->data = data; | |
4865 | entry->maxlen = maxlen; | |
4866 | entry->mode = mode; | |
4867 | entry->proc_handler = proc_handler; | |
201c373e NK |
4868 | |
4869 | if (load_idx) { | |
4870 | entry->extra1 = &min_load_idx; | |
4871 | entry->extra2 = &max_load_idx; | |
4872 | } | |
e692ab53 NP |
4873 | } |
4874 | ||
4875 | static struct ctl_table * | |
4876 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
4877 | { | |
37e6bae8 | 4878 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
e692ab53 | 4879 | |
ad1cdc1d MM |
4880 | if (table == NULL) |
4881 | return NULL; | |
4882 | ||
e0361851 | 4883 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
201c373e | 4884 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4885 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
201c373e | 4886 | sizeof(long), 0644, proc_doulongvec_minmax, false); |
e0361851 | 4887 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
201c373e | 4888 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4889 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
201c373e | 4890 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4891 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
201c373e | 4892 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4893 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
201c373e | 4894 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4895 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
201c373e | 4896 | sizeof(int), 0644, proc_dointvec_minmax, true); |
e0361851 | 4897 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
201c373e | 4898 | sizeof(int), 0644, proc_dointvec_minmax, false); |
e0361851 | 4899 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
201c373e | 4900 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4901 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 | 4902 | &sd->cache_nice_tries, |
201c373e | 4903 | sizeof(int), 0644, proc_dointvec_minmax, false); |
ace8b3d6 | 4904 | set_table_entry(&table[10], "flags", &sd->flags, |
201c373e | 4905 | sizeof(int), 0644, proc_dointvec_minmax, false); |
37e6bae8 AS |
4906 | set_table_entry(&table[11], "max_newidle_lb_cost", |
4907 | &sd->max_newidle_lb_cost, | |
4908 | sizeof(long), 0644, proc_doulongvec_minmax, false); | |
4909 | set_table_entry(&table[12], "name", sd->name, | |
201c373e | 4910 | CORENAME_MAX_SIZE, 0444, proc_dostring, false); |
37e6bae8 | 4911 | /* &table[13] is terminator */ |
e692ab53 NP |
4912 | |
4913 | return table; | |
4914 | } | |
4915 | ||
be7002e6 | 4916 | static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
4917 | { |
4918 | struct ctl_table *entry, *table; | |
4919 | struct sched_domain *sd; | |
4920 | int domain_num = 0, i; | |
4921 | char buf[32]; | |
4922 | ||
4923 | for_each_domain(cpu, sd) | |
4924 | domain_num++; | |
4925 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
4926 | if (table == NULL) |
4927 | return NULL; | |
e692ab53 NP |
4928 | |
4929 | i = 0; | |
4930 | for_each_domain(cpu, sd) { | |
4931 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 4932 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4933 | entry->mode = 0555; |
e692ab53 NP |
4934 | entry->child = sd_alloc_ctl_domain_table(sd); |
4935 | entry++; | |
4936 | i++; | |
4937 | } | |
4938 | return table; | |
4939 | } | |
4940 | ||
4941 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 4942 | static void register_sched_domain_sysctl(void) |
e692ab53 | 4943 | { |
6ad4c188 | 4944 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
4945 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
4946 | char buf[32]; | |
4947 | ||
7378547f MM |
4948 | WARN_ON(sd_ctl_dir[0].child); |
4949 | sd_ctl_dir[0].child = entry; | |
4950 | ||
ad1cdc1d MM |
4951 | if (entry == NULL) |
4952 | return; | |
4953 | ||
6ad4c188 | 4954 | for_each_possible_cpu(i) { |
e692ab53 | 4955 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 4956 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 4957 | entry->mode = 0555; |
e692ab53 | 4958 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 4959 | entry++; |
e692ab53 | 4960 | } |
7378547f MM |
4961 | |
4962 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
4963 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
4964 | } | |
6382bc90 | 4965 | |
7378547f | 4966 | /* may be called multiple times per register */ |
6382bc90 MM |
4967 | static void unregister_sched_domain_sysctl(void) |
4968 | { | |
7378547f MM |
4969 | if (sd_sysctl_header) |
4970 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 4971 | sd_sysctl_header = NULL; |
7378547f MM |
4972 | if (sd_ctl_dir[0].child) |
4973 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 4974 | } |
e692ab53 | 4975 | #else |
6382bc90 MM |
4976 | static void register_sched_domain_sysctl(void) |
4977 | { | |
4978 | } | |
4979 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
4980 | { |
4981 | } | |
4982 | #endif | |
4983 | ||
1f11eb6a GH |
4984 | static void set_rq_online(struct rq *rq) |
4985 | { | |
4986 | if (!rq->online) { | |
4987 | const struct sched_class *class; | |
4988 | ||
c6c4927b | 4989 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
4990 | rq->online = 1; |
4991 | ||
4992 | for_each_class(class) { | |
4993 | if (class->rq_online) | |
4994 | class->rq_online(rq); | |
4995 | } | |
4996 | } | |
4997 | } | |
4998 | ||
4999 | static void set_rq_offline(struct rq *rq) | |
5000 | { | |
5001 | if (rq->online) { | |
5002 | const struct sched_class *class; | |
5003 | ||
5004 | for_each_class(class) { | |
5005 | if (class->rq_offline) | |
5006 | class->rq_offline(rq); | |
5007 | } | |
5008 | ||
c6c4927b | 5009 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5010 | rq->online = 0; |
5011 | } | |
5012 | } | |
5013 | ||
1da177e4 LT |
5014 | /* |
5015 | * migration_call - callback that gets triggered when a CPU is added. | |
5016 | * Here we can start up the necessary migration thread for the new CPU. | |
5017 | */ | |
0db0628d | 5018 | static int |
48f24c4d | 5019 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
1da177e4 | 5020 | { |
48f24c4d | 5021 | int cpu = (long)hcpu; |
1da177e4 | 5022 | unsigned long flags; |
969c7921 | 5023 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5024 | |
48c5ccae | 5025 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5026 | |
1da177e4 | 5027 | case CPU_UP_PREPARE: |
a468d389 | 5028 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5029 | break; |
48f24c4d | 5030 | |
1da177e4 | 5031 | case CPU_ONLINE: |
1f94ef59 | 5032 | /* Update our root-domain */ |
05fa785c | 5033 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5034 | if (rq->rd) { |
c6c4927b | 5035 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5036 | |
5037 | set_rq_online(rq); | |
1f94ef59 | 5038 | } |
05fa785c | 5039 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5040 | break; |
48f24c4d | 5041 | |
1da177e4 | 5042 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5043 | case CPU_DYING: |
317f3941 | 5044 | sched_ttwu_pending(); |
57d885fe | 5045 | /* Update our root-domain */ |
05fa785c | 5046 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5047 | if (rq->rd) { |
c6c4927b | 5048 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5049 | set_rq_offline(rq); |
57d885fe | 5050 | } |
48c5ccae PZ |
5051 | migrate_tasks(cpu); |
5052 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5053 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
5d180232 | 5054 | break; |
48c5ccae | 5055 | |
5d180232 | 5056 | case CPU_DEAD: |
f319da0c | 5057 | calc_load_migrate(rq); |
57d885fe | 5058 | break; |
1da177e4 LT |
5059 | #endif |
5060 | } | |
49c022e6 PZ |
5061 | |
5062 | update_max_interval(); | |
5063 | ||
1da177e4 LT |
5064 | return NOTIFY_OK; |
5065 | } | |
5066 | ||
f38b0820 PM |
5067 | /* |
5068 | * Register at high priority so that task migration (migrate_all_tasks) | |
5069 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5070 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5071 | */ |
0db0628d | 5072 | static struct notifier_block migration_notifier = { |
1da177e4 | 5073 | .notifier_call = migration_call, |
50a323b7 | 5074 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5075 | }; |
5076 | ||
a803f026 CM |
5077 | static void __cpuinit set_cpu_rq_start_time(void) |
5078 | { | |
5079 | int cpu = smp_processor_id(); | |
5080 | struct rq *rq = cpu_rq(cpu); | |
5081 | rq->age_stamp = sched_clock_cpu(cpu); | |
5082 | } | |
5083 | ||
0db0628d | 5084 | static int sched_cpu_active(struct notifier_block *nfb, |
3a101d05 TH |
5085 | unsigned long action, void *hcpu) |
5086 | { | |
5087 | switch (action & ~CPU_TASKS_FROZEN) { | |
a803f026 CM |
5088 | case CPU_STARTING: |
5089 | set_cpu_rq_start_time(); | |
5090 | return NOTIFY_OK; | |
3a101d05 TH |
5091 | case CPU_DOWN_FAILED: |
5092 | set_cpu_active((long)hcpu, true); | |
5093 | return NOTIFY_OK; | |
5094 | default: | |
5095 | return NOTIFY_DONE; | |
5096 | } | |
5097 | } | |
5098 | ||
0db0628d | 5099 | static int sched_cpu_inactive(struct notifier_block *nfb, |
3a101d05 TH |
5100 | unsigned long action, void *hcpu) |
5101 | { | |
de212f18 PZ |
5102 | unsigned long flags; |
5103 | long cpu = (long)hcpu; | |
5104 | ||
3a101d05 TH |
5105 | switch (action & ~CPU_TASKS_FROZEN) { |
5106 | case CPU_DOWN_PREPARE: | |
de212f18 PZ |
5107 | set_cpu_active(cpu, false); |
5108 | ||
5109 | /* explicitly allow suspend */ | |
5110 | if (!(action & CPU_TASKS_FROZEN)) { | |
5111 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
5112 | bool overflow; | |
5113 | int cpus; | |
5114 | ||
5115 | raw_spin_lock_irqsave(&dl_b->lock, flags); | |
5116 | cpus = dl_bw_cpus(cpu); | |
5117 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5118 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
5119 | ||
5120 | if (overflow) | |
5121 | return notifier_from_errno(-EBUSY); | |
5122 | } | |
3a101d05 | 5123 | return NOTIFY_OK; |
3a101d05 | 5124 | } |
de212f18 PZ |
5125 | |
5126 | return NOTIFY_DONE; | |
3a101d05 TH |
5127 | } |
5128 | ||
7babe8db | 5129 | static int __init migration_init(void) |
1da177e4 LT |
5130 | { |
5131 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5132 | int err; |
48f24c4d | 5133 | |
3a101d05 | 5134 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5135 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5136 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5137 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5138 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5139 | |
3a101d05 TH |
5140 | /* Register cpu active notifiers */ |
5141 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5142 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5143 | ||
a004cd42 | 5144 | return 0; |
1da177e4 | 5145 | } |
7babe8db | 5146 | early_initcall(migration_init); |
1da177e4 LT |
5147 | #endif |
5148 | ||
5149 | #ifdef CONFIG_SMP | |
476f3534 | 5150 | |
4cb98839 PZ |
5151 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
5152 | ||
3e9830dc | 5153 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5154 | |
d039ac60 | 5155 | static __read_mostly int sched_debug_enabled; |
f6630114 | 5156 | |
d039ac60 | 5157 | static int __init sched_debug_setup(char *str) |
f6630114 | 5158 | { |
d039ac60 | 5159 | sched_debug_enabled = 1; |
f6630114 MT |
5160 | |
5161 | return 0; | |
5162 | } | |
d039ac60 PZ |
5163 | early_param("sched_debug", sched_debug_setup); |
5164 | ||
5165 | static inline bool sched_debug(void) | |
5166 | { | |
5167 | return sched_debug_enabled; | |
5168 | } | |
f6630114 | 5169 | |
7c16ec58 | 5170 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5171 | struct cpumask *groupmask) |
1da177e4 | 5172 | { |
4dcf6aff | 5173 | struct sched_group *group = sd->groups; |
434d53b0 | 5174 | char str[256]; |
1da177e4 | 5175 | |
968ea6d8 | 5176 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5177 | cpumask_clear(groupmask); |
4dcf6aff IM |
5178 | |
5179 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5180 | ||
5181 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5182 | printk("does not load-balance\n"); |
4dcf6aff | 5183 | if (sd->parent) |
3df0fc5b PZ |
5184 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5185 | " has parent"); | |
4dcf6aff | 5186 | return -1; |
41c7ce9a NP |
5187 | } |
5188 | ||
3df0fc5b | 5189 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5190 | |
758b2cdc | 5191 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5192 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5193 | "CPU%d\n", cpu); | |
4dcf6aff | 5194 | } |
758b2cdc | 5195 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5196 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5197 | " CPU%d\n", cpu); | |
4dcf6aff | 5198 | } |
1da177e4 | 5199 | |
4dcf6aff | 5200 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5201 | do { |
4dcf6aff | 5202 | if (!group) { |
3df0fc5b PZ |
5203 | printk("\n"); |
5204 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5205 | break; |
5206 | } | |
5207 | ||
c3decf0d | 5208 | /* |
63b2ca30 NP |
5209 | * Even though we initialize ->capacity to something semi-sane, |
5210 | * we leave capacity_orig unset. This allows us to detect if | |
c3decf0d PZ |
5211 | * domain iteration is still funny without causing /0 traps. |
5212 | */ | |
63b2ca30 | 5213 | if (!group->sgc->capacity_orig) { |
3df0fc5b | 5214 | printk(KERN_CONT "\n"); |
63b2ca30 | 5215 | printk(KERN_ERR "ERROR: domain->cpu_capacity not set\n"); |
4dcf6aff IM |
5216 | break; |
5217 | } | |
1da177e4 | 5218 | |
758b2cdc | 5219 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5220 | printk(KERN_CONT "\n"); |
5221 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5222 | break; |
5223 | } | |
1da177e4 | 5224 | |
cb83b629 PZ |
5225 | if (!(sd->flags & SD_OVERLAP) && |
5226 | cpumask_intersects(groupmask, sched_group_cpus(group))) { | |
3df0fc5b PZ |
5227 | printk(KERN_CONT "\n"); |
5228 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5229 | break; |
5230 | } | |
1da177e4 | 5231 | |
758b2cdc | 5232 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5233 | |
968ea6d8 | 5234 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5235 | |
3df0fc5b | 5236 | printk(KERN_CONT " %s", str); |
ca8ce3d0 | 5237 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
63b2ca30 NP |
5238 | printk(KERN_CONT " (cpu_capacity = %d)", |
5239 | group->sgc->capacity); | |
381512cf | 5240 | } |
1da177e4 | 5241 | |
4dcf6aff IM |
5242 | group = group->next; |
5243 | } while (group != sd->groups); | |
3df0fc5b | 5244 | printk(KERN_CONT "\n"); |
1da177e4 | 5245 | |
758b2cdc | 5246 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5247 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5248 | |
758b2cdc RR |
5249 | if (sd->parent && |
5250 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5251 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5252 | "of domain->span\n"); | |
4dcf6aff IM |
5253 | return 0; |
5254 | } | |
1da177e4 | 5255 | |
4dcf6aff IM |
5256 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5257 | { | |
5258 | int level = 0; | |
1da177e4 | 5259 | |
d039ac60 | 5260 | if (!sched_debug_enabled) |
f6630114 MT |
5261 | return; |
5262 | ||
4dcf6aff IM |
5263 | if (!sd) { |
5264 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5265 | return; | |
5266 | } | |
1da177e4 | 5267 | |
4dcf6aff IM |
5268 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5269 | ||
5270 | for (;;) { | |
4cb98839 | 5271 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 5272 | break; |
1da177e4 LT |
5273 | level++; |
5274 | sd = sd->parent; | |
33859f7f | 5275 | if (!sd) |
4dcf6aff IM |
5276 | break; |
5277 | } | |
1da177e4 | 5278 | } |
6d6bc0ad | 5279 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5280 | # define sched_domain_debug(sd, cpu) do { } while (0) |
d039ac60 PZ |
5281 | static inline bool sched_debug(void) |
5282 | { | |
5283 | return false; | |
5284 | } | |
6d6bc0ad | 5285 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5286 | |
1a20ff27 | 5287 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5288 | { |
758b2cdc | 5289 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5290 | return 1; |
5291 | ||
5292 | /* Following flags need at least 2 groups */ | |
5293 | if (sd->flags & (SD_LOAD_BALANCE | | |
5294 | SD_BALANCE_NEWIDLE | | |
5295 | SD_BALANCE_FORK | | |
89c4710e | 5296 | SD_BALANCE_EXEC | |
5d4dfddd | 5297 | SD_SHARE_CPUCAPACITY | |
d77b3ed5 VG |
5298 | SD_SHARE_PKG_RESOURCES | |
5299 | SD_SHARE_POWERDOMAIN)) { | |
245af2c7 SS |
5300 | if (sd->groups != sd->groups->next) |
5301 | return 0; | |
5302 | } | |
5303 | ||
5304 | /* Following flags don't use groups */ | |
c88d5910 | 5305 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5306 | return 0; |
5307 | ||
5308 | return 1; | |
5309 | } | |
5310 | ||
48f24c4d IM |
5311 | static int |
5312 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5313 | { |
5314 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5315 | ||
5316 | if (sd_degenerate(parent)) | |
5317 | return 1; | |
5318 | ||
758b2cdc | 5319 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5320 | return 0; |
5321 | ||
245af2c7 SS |
5322 | /* Flags needing groups don't count if only 1 group in parent */ |
5323 | if (parent->groups == parent->groups->next) { | |
5324 | pflags &= ~(SD_LOAD_BALANCE | | |
5325 | SD_BALANCE_NEWIDLE | | |
5326 | SD_BALANCE_FORK | | |
89c4710e | 5327 | SD_BALANCE_EXEC | |
5d4dfddd | 5328 | SD_SHARE_CPUCAPACITY | |
10866e62 | 5329 | SD_SHARE_PKG_RESOURCES | |
d77b3ed5 VG |
5330 | SD_PREFER_SIBLING | |
5331 | SD_SHARE_POWERDOMAIN); | |
5436499e KC |
5332 | if (nr_node_ids == 1) |
5333 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
5334 | } |
5335 | if (~cflags & pflags) | |
5336 | return 0; | |
5337 | ||
5338 | return 1; | |
5339 | } | |
5340 | ||
dce840a0 | 5341 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 5342 | { |
dce840a0 | 5343 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 5344 | |
68e74568 | 5345 | cpupri_cleanup(&rd->cpupri); |
6bfd6d72 | 5346 | cpudl_cleanup(&rd->cpudl); |
1baca4ce | 5347 | free_cpumask_var(rd->dlo_mask); |
c6c4927b RR |
5348 | free_cpumask_var(rd->rto_mask); |
5349 | free_cpumask_var(rd->online); | |
5350 | free_cpumask_var(rd->span); | |
5351 | kfree(rd); | |
5352 | } | |
5353 | ||
57d885fe GH |
5354 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
5355 | { | |
a0490fa3 | 5356 | struct root_domain *old_rd = NULL; |
57d885fe | 5357 | unsigned long flags; |
57d885fe | 5358 | |
05fa785c | 5359 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
5360 | |
5361 | if (rq->rd) { | |
a0490fa3 | 5362 | old_rd = rq->rd; |
57d885fe | 5363 | |
c6c4927b | 5364 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 5365 | set_rq_offline(rq); |
57d885fe | 5366 | |
c6c4927b | 5367 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 5368 | |
a0490fa3 | 5369 | /* |
0515973f | 5370 | * If we dont want to free the old_rd yet then |
a0490fa3 IM |
5371 | * set old_rd to NULL to skip the freeing later |
5372 | * in this function: | |
5373 | */ | |
5374 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
5375 | old_rd = NULL; | |
57d885fe GH |
5376 | } |
5377 | ||
5378 | atomic_inc(&rd->refcount); | |
5379 | rq->rd = rd; | |
5380 | ||
c6c4927b | 5381 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 5382 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 5383 | set_rq_online(rq); |
57d885fe | 5384 | |
05fa785c | 5385 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
5386 | |
5387 | if (old_rd) | |
dce840a0 | 5388 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
5389 | } |
5390 | ||
68c38fc3 | 5391 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
5392 | { |
5393 | memset(rd, 0, sizeof(*rd)); | |
5394 | ||
68c38fc3 | 5395 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 5396 | goto out; |
68c38fc3 | 5397 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 5398 | goto free_span; |
1baca4ce | 5399 | if (!alloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
c6c4927b | 5400 | goto free_online; |
1baca4ce JL |
5401 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
5402 | goto free_dlo_mask; | |
6e0534f2 | 5403 | |
332ac17e | 5404 | init_dl_bw(&rd->dl_bw); |
6bfd6d72 JL |
5405 | if (cpudl_init(&rd->cpudl) != 0) |
5406 | goto free_dlo_mask; | |
332ac17e | 5407 | |
68c38fc3 | 5408 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 5409 | goto free_rto_mask; |
c6c4927b | 5410 | return 0; |
6e0534f2 | 5411 | |
68e74568 RR |
5412 | free_rto_mask: |
5413 | free_cpumask_var(rd->rto_mask); | |
1baca4ce JL |
5414 | free_dlo_mask: |
5415 | free_cpumask_var(rd->dlo_mask); | |
c6c4927b RR |
5416 | free_online: |
5417 | free_cpumask_var(rd->online); | |
5418 | free_span: | |
5419 | free_cpumask_var(rd->span); | |
0c910d28 | 5420 | out: |
c6c4927b | 5421 | return -ENOMEM; |
57d885fe GH |
5422 | } |
5423 | ||
029632fb PZ |
5424 | /* |
5425 | * By default the system creates a single root-domain with all cpus as | |
5426 | * members (mimicking the global state we have today). | |
5427 | */ | |
5428 | struct root_domain def_root_domain; | |
5429 | ||
57d885fe GH |
5430 | static void init_defrootdomain(void) |
5431 | { | |
68c38fc3 | 5432 | init_rootdomain(&def_root_domain); |
c6c4927b | 5433 | |
57d885fe GH |
5434 | atomic_set(&def_root_domain.refcount, 1); |
5435 | } | |
5436 | ||
dc938520 | 5437 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
5438 | { |
5439 | struct root_domain *rd; | |
5440 | ||
5441 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
5442 | if (!rd) | |
5443 | return NULL; | |
5444 | ||
68c38fc3 | 5445 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
5446 | kfree(rd); |
5447 | return NULL; | |
5448 | } | |
57d885fe GH |
5449 | |
5450 | return rd; | |
5451 | } | |
5452 | ||
63b2ca30 | 5453 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
e3589f6c PZ |
5454 | { |
5455 | struct sched_group *tmp, *first; | |
5456 | ||
5457 | if (!sg) | |
5458 | return; | |
5459 | ||
5460 | first = sg; | |
5461 | do { | |
5462 | tmp = sg->next; | |
5463 | ||
63b2ca30 NP |
5464 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
5465 | kfree(sg->sgc); | |
e3589f6c PZ |
5466 | |
5467 | kfree(sg); | |
5468 | sg = tmp; | |
5469 | } while (sg != first); | |
5470 | } | |
5471 | ||
dce840a0 PZ |
5472 | static void free_sched_domain(struct rcu_head *rcu) |
5473 | { | |
5474 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
e3589f6c PZ |
5475 | |
5476 | /* | |
5477 | * If its an overlapping domain it has private groups, iterate and | |
5478 | * nuke them all. | |
5479 | */ | |
5480 | if (sd->flags & SD_OVERLAP) { | |
5481 | free_sched_groups(sd->groups, 1); | |
5482 | } else if (atomic_dec_and_test(&sd->groups->ref)) { | |
63b2ca30 | 5483 | kfree(sd->groups->sgc); |
dce840a0 | 5484 | kfree(sd->groups); |
9c3f75cb | 5485 | } |
dce840a0 PZ |
5486 | kfree(sd); |
5487 | } | |
5488 | ||
5489 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
5490 | { | |
5491 | call_rcu(&sd->rcu, free_sched_domain); | |
5492 | } | |
5493 | ||
5494 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
5495 | { | |
5496 | for (; sd; sd = sd->parent) | |
5497 | destroy_sched_domain(sd, cpu); | |
5498 | } | |
5499 | ||
518cd623 PZ |
5500 | /* |
5501 | * Keep a special pointer to the highest sched_domain that has | |
5502 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this | |
5503 | * allows us to avoid some pointer chasing select_idle_sibling(). | |
5504 | * | |
5505 | * Also keep a unique ID per domain (we use the first cpu number in | |
5506 | * the cpumask of the domain), this allows us to quickly tell if | |
39be3501 | 5507 | * two cpus are in the same cache domain, see cpus_share_cache(). |
518cd623 PZ |
5508 | */ |
5509 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); | |
7d9ffa89 | 5510 | DEFINE_PER_CPU(int, sd_llc_size); |
518cd623 | 5511 | DEFINE_PER_CPU(int, sd_llc_id); |
fb13c7ee | 5512 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
37dc6b50 PM |
5513 | DEFINE_PER_CPU(struct sched_domain *, sd_busy); |
5514 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); | |
518cd623 PZ |
5515 | |
5516 | static void update_top_cache_domain(int cpu) | |
5517 | { | |
5518 | struct sched_domain *sd; | |
5d4cf996 | 5519 | struct sched_domain *busy_sd = NULL; |
518cd623 | 5520 | int id = cpu; |
7d9ffa89 | 5521 | int size = 1; |
518cd623 PZ |
5522 | |
5523 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); | |
7d9ffa89 | 5524 | if (sd) { |
518cd623 | 5525 | id = cpumask_first(sched_domain_span(sd)); |
7d9ffa89 | 5526 | size = cpumask_weight(sched_domain_span(sd)); |
5d4cf996 | 5527 | busy_sd = sd->parent; /* sd_busy */ |
7d9ffa89 | 5528 | } |
5d4cf996 | 5529 | rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd); |
518cd623 PZ |
5530 | |
5531 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); | |
7d9ffa89 | 5532 | per_cpu(sd_llc_size, cpu) = size; |
518cd623 | 5533 | per_cpu(sd_llc_id, cpu) = id; |
fb13c7ee MG |
5534 | |
5535 | sd = lowest_flag_domain(cpu, SD_NUMA); | |
5536 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); | |
37dc6b50 PM |
5537 | |
5538 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); | |
5539 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); | |
518cd623 PZ |
5540 | } |
5541 | ||
1da177e4 | 5542 | /* |
0eab9146 | 5543 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
5544 | * hold the hotplug lock. |
5545 | */ | |
0eab9146 IM |
5546 | static void |
5547 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 5548 | { |
70b97a7f | 5549 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5550 | struct sched_domain *tmp; |
5551 | ||
5552 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 5553 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
5554 | struct sched_domain *parent = tmp->parent; |
5555 | if (!parent) | |
5556 | break; | |
f29c9b1c | 5557 | |
1a848870 | 5558 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5559 | tmp->parent = parent->parent; |
1a848870 SS |
5560 | if (parent->parent) |
5561 | parent->parent->child = tmp; | |
10866e62 PZ |
5562 | /* |
5563 | * Transfer SD_PREFER_SIBLING down in case of a | |
5564 | * degenerate parent; the spans match for this | |
5565 | * so the property transfers. | |
5566 | */ | |
5567 | if (parent->flags & SD_PREFER_SIBLING) | |
5568 | tmp->flags |= SD_PREFER_SIBLING; | |
dce840a0 | 5569 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
5570 | } else |
5571 | tmp = tmp->parent; | |
245af2c7 SS |
5572 | } |
5573 | ||
1a848870 | 5574 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 5575 | tmp = sd; |
245af2c7 | 5576 | sd = sd->parent; |
dce840a0 | 5577 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
5578 | if (sd) |
5579 | sd->child = NULL; | |
5580 | } | |
1da177e4 | 5581 | |
4cb98839 | 5582 | sched_domain_debug(sd, cpu); |
1da177e4 | 5583 | |
57d885fe | 5584 | rq_attach_root(rq, rd); |
dce840a0 | 5585 | tmp = rq->sd; |
674311d5 | 5586 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 5587 | destroy_sched_domains(tmp, cpu); |
518cd623 PZ |
5588 | |
5589 | update_top_cache_domain(cpu); | |
1da177e4 LT |
5590 | } |
5591 | ||
5592 | /* cpus with isolated domains */ | |
dcc30a35 | 5593 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
5594 | |
5595 | /* Setup the mask of cpus configured for isolated domains */ | |
5596 | static int __init isolated_cpu_setup(char *str) | |
5597 | { | |
bdddd296 | 5598 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 5599 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
5600 | return 1; |
5601 | } | |
5602 | ||
8927f494 | 5603 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 5604 | |
49a02c51 | 5605 | struct s_data { |
21d42ccf | 5606 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
5607 | struct root_domain *rd; |
5608 | }; | |
5609 | ||
2109b99e | 5610 | enum s_alloc { |
2109b99e | 5611 | sa_rootdomain, |
21d42ccf | 5612 | sa_sd, |
dce840a0 | 5613 | sa_sd_storage, |
2109b99e AH |
5614 | sa_none, |
5615 | }; | |
5616 | ||
c1174876 PZ |
5617 | /* |
5618 | * Build an iteration mask that can exclude certain CPUs from the upwards | |
5619 | * domain traversal. | |
5620 | * | |
5621 | * Asymmetric node setups can result in situations where the domain tree is of | |
5622 | * unequal depth, make sure to skip domains that already cover the entire | |
5623 | * range. | |
5624 | * | |
5625 | * In that case build_sched_domains() will have terminated the iteration early | |
5626 | * and our sibling sd spans will be empty. Domains should always include the | |
5627 | * cpu they're built on, so check that. | |
5628 | * | |
5629 | */ | |
5630 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) | |
5631 | { | |
5632 | const struct cpumask *span = sched_domain_span(sd); | |
5633 | struct sd_data *sdd = sd->private; | |
5634 | struct sched_domain *sibling; | |
5635 | int i; | |
5636 | ||
5637 | for_each_cpu(i, span) { | |
5638 | sibling = *per_cpu_ptr(sdd->sd, i); | |
5639 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) | |
5640 | continue; | |
5641 | ||
5642 | cpumask_set_cpu(i, sched_group_mask(sg)); | |
5643 | } | |
5644 | } | |
5645 | ||
5646 | /* | |
5647 | * Return the canonical balance cpu for this group, this is the first cpu | |
5648 | * of this group that's also in the iteration mask. | |
5649 | */ | |
5650 | int group_balance_cpu(struct sched_group *sg) | |
5651 | { | |
5652 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); | |
5653 | } | |
5654 | ||
e3589f6c PZ |
5655 | static int |
5656 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) | |
5657 | { | |
5658 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; | |
5659 | const struct cpumask *span = sched_domain_span(sd); | |
5660 | struct cpumask *covered = sched_domains_tmpmask; | |
5661 | struct sd_data *sdd = sd->private; | |
5662 | struct sched_domain *child; | |
5663 | int i; | |
5664 | ||
5665 | cpumask_clear(covered); | |
5666 | ||
5667 | for_each_cpu(i, span) { | |
5668 | struct cpumask *sg_span; | |
5669 | ||
5670 | if (cpumask_test_cpu(i, covered)) | |
5671 | continue; | |
5672 | ||
c1174876 PZ |
5673 | child = *per_cpu_ptr(sdd->sd, i); |
5674 | ||
5675 | /* See the comment near build_group_mask(). */ | |
5676 | if (!cpumask_test_cpu(i, sched_domain_span(child))) | |
5677 | continue; | |
5678 | ||
e3589f6c | 5679 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
4d78a223 | 5680 | GFP_KERNEL, cpu_to_node(cpu)); |
e3589f6c PZ |
5681 | |
5682 | if (!sg) | |
5683 | goto fail; | |
5684 | ||
5685 | sg_span = sched_group_cpus(sg); | |
e3589f6c PZ |
5686 | if (child->child) { |
5687 | child = child->child; | |
5688 | cpumask_copy(sg_span, sched_domain_span(child)); | |
5689 | } else | |
5690 | cpumask_set_cpu(i, sg_span); | |
5691 | ||
5692 | cpumask_or(covered, covered, sg_span); | |
5693 | ||
63b2ca30 NP |
5694 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
5695 | if (atomic_inc_return(&sg->sgc->ref) == 1) | |
c1174876 PZ |
5696 | build_group_mask(sd, sg); |
5697 | ||
c3decf0d | 5698 | /* |
63b2ca30 | 5699 | * Initialize sgc->capacity such that even if we mess up the |
c3decf0d PZ |
5700 | * domains and no possible iteration will get us here, we won't |
5701 | * die on a /0 trap. | |
5702 | */ | |
ca8ce3d0 | 5703 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
63b2ca30 | 5704 | sg->sgc->capacity_orig = sg->sgc->capacity; |
e3589f6c | 5705 | |
c1174876 PZ |
5706 | /* |
5707 | * Make sure the first group of this domain contains the | |
5708 | * canonical balance cpu. Otherwise the sched_domain iteration | |
5709 | * breaks. See update_sg_lb_stats(). | |
5710 | */ | |
74a5ce20 | 5711 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
c1174876 | 5712 | group_balance_cpu(sg) == cpu) |
e3589f6c PZ |
5713 | groups = sg; |
5714 | ||
5715 | if (!first) | |
5716 | first = sg; | |
5717 | if (last) | |
5718 | last->next = sg; | |
5719 | last = sg; | |
5720 | last->next = first; | |
5721 | } | |
5722 | sd->groups = groups; | |
5723 | ||
5724 | return 0; | |
5725 | ||
5726 | fail: | |
5727 | free_sched_groups(first, 0); | |
5728 | ||
5729 | return -ENOMEM; | |
5730 | } | |
5731 | ||
dce840a0 | 5732 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 5733 | { |
dce840a0 PZ |
5734 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
5735 | struct sched_domain *child = sd->child; | |
1da177e4 | 5736 | |
dce840a0 PZ |
5737 | if (child) |
5738 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 5739 | |
9c3f75cb | 5740 | if (sg) { |
dce840a0 | 5741 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
63b2ca30 NP |
5742 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
5743 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ | |
9c3f75cb | 5744 | } |
dce840a0 PZ |
5745 | |
5746 | return cpu; | |
1e9f28fa | 5747 | } |
1e9f28fa | 5748 | |
01a08546 | 5749 | /* |
dce840a0 PZ |
5750 | * build_sched_groups will build a circular linked list of the groups |
5751 | * covered by the given span, and will set each group's ->cpumask correctly, | |
ced549fa | 5752 | * and ->cpu_capacity to 0. |
e3589f6c PZ |
5753 | * |
5754 | * Assumes the sched_domain tree is fully constructed | |
01a08546 | 5755 | */ |
e3589f6c PZ |
5756 | static int |
5757 | build_sched_groups(struct sched_domain *sd, int cpu) | |
1da177e4 | 5758 | { |
dce840a0 PZ |
5759 | struct sched_group *first = NULL, *last = NULL; |
5760 | struct sd_data *sdd = sd->private; | |
5761 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 5762 | struct cpumask *covered; |
dce840a0 | 5763 | int i; |
9c1cfda2 | 5764 | |
e3589f6c PZ |
5765 | get_group(cpu, sdd, &sd->groups); |
5766 | atomic_inc(&sd->groups->ref); | |
5767 | ||
0936629f | 5768 | if (cpu != cpumask_first(span)) |
e3589f6c PZ |
5769 | return 0; |
5770 | ||
f96225fd PZ |
5771 | lockdep_assert_held(&sched_domains_mutex); |
5772 | covered = sched_domains_tmpmask; | |
5773 | ||
dce840a0 | 5774 | cpumask_clear(covered); |
6711cab4 | 5775 | |
dce840a0 PZ |
5776 | for_each_cpu(i, span) { |
5777 | struct sched_group *sg; | |
cd08e923 | 5778 | int group, j; |
6711cab4 | 5779 | |
dce840a0 PZ |
5780 | if (cpumask_test_cpu(i, covered)) |
5781 | continue; | |
6711cab4 | 5782 | |
cd08e923 | 5783 | group = get_group(i, sdd, &sg); |
c1174876 | 5784 | cpumask_setall(sched_group_mask(sg)); |
0601a88d | 5785 | |
dce840a0 PZ |
5786 | for_each_cpu(j, span) { |
5787 | if (get_group(j, sdd, NULL) != group) | |
5788 | continue; | |
0601a88d | 5789 | |
dce840a0 PZ |
5790 | cpumask_set_cpu(j, covered); |
5791 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
5792 | } | |
0601a88d | 5793 | |
dce840a0 PZ |
5794 | if (!first) |
5795 | first = sg; | |
5796 | if (last) | |
5797 | last->next = sg; | |
5798 | last = sg; | |
5799 | } | |
5800 | last->next = first; | |
e3589f6c PZ |
5801 | |
5802 | return 0; | |
0601a88d | 5803 | } |
51888ca2 | 5804 | |
89c4710e | 5805 | /* |
63b2ca30 | 5806 | * Initialize sched groups cpu_capacity. |
89c4710e | 5807 | * |
63b2ca30 | 5808 | * cpu_capacity indicates the capacity of sched group, which is used while |
89c4710e | 5809 | * distributing the load between different sched groups in a sched domain. |
63b2ca30 NP |
5810 | * Typically cpu_capacity for all the groups in a sched domain will be same |
5811 | * unless there are asymmetries in the topology. If there are asymmetries, | |
5812 | * group having more cpu_capacity will pickup more load compared to the | |
5813 | * group having less cpu_capacity. | |
89c4710e | 5814 | */ |
63b2ca30 | 5815 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
89c4710e | 5816 | { |
e3589f6c | 5817 | struct sched_group *sg = sd->groups; |
89c4710e | 5818 | |
94c95ba6 | 5819 | WARN_ON(!sg); |
e3589f6c PZ |
5820 | |
5821 | do { | |
5822 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); | |
5823 | sg = sg->next; | |
5824 | } while (sg != sd->groups); | |
89c4710e | 5825 | |
c1174876 | 5826 | if (cpu != group_balance_cpu(sg)) |
e3589f6c | 5827 | return; |
aae6d3dd | 5828 | |
63b2ca30 NP |
5829 | update_group_capacity(sd, cpu); |
5830 | atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight); | |
89c4710e SS |
5831 | } |
5832 | ||
7c16ec58 MT |
5833 | /* |
5834 | * Initializers for schedule domains | |
5835 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
5836 | */ | |
5837 | ||
1d3504fc | 5838 | static int default_relax_domain_level = -1; |
60495e77 | 5839 | int sched_domain_level_max; |
1d3504fc HS |
5840 | |
5841 | static int __init setup_relax_domain_level(char *str) | |
5842 | { | |
a841f8ce DS |
5843 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
5844 | pr_warn("Unable to set relax_domain_level\n"); | |
30e0e178 | 5845 | |
1d3504fc HS |
5846 | return 1; |
5847 | } | |
5848 | __setup("relax_domain_level=", setup_relax_domain_level); | |
5849 | ||
5850 | static void set_domain_attribute(struct sched_domain *sd, | |
5851 | struct sched_domain_attr *attr) | |
5852 | { | |
5853 | int request; | |
5854 | ||
5855 | if (!attr || attr->relax_domain_level < 0) { | |
5856 | if (default_relax_domain_level < 0) | |
5857 | return; | |
5858 | else | |
5859 | request = default_relax_domain_level; | |
5860 | } else | |
5861 | request = attr->relax_domain_level; | |
5862 | if (request < sd->level) { | |
5863 | /* turn off idle balance on this domain */ | |
c88d5910 | 5864 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5865 | } else { |
5866 | /* turn on idle balance on this domain */ | |
c88d5910 | 5867 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
5868 | } |
5869 | } | |
5870 | ||
54ab4ff4 PZ |
5871 | static void __sdt_free(const struct cpumask *cpu_map); |
5872 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
5873 | ||
2109b99e AH |
5874 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
5875 | const struct cpumask *cpu_map) | |
5876 | { | |
5877 | switch (what) { | |
2109b99e | 5878 | case sa_rootdomain: |
822ff793 PZ |
5879 | if (!atomic_read(&d->rd->refcount)) |
5880 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
5881 | case sa_sd: |
5882 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 5883 | case sa_sd_storage: |
54ab4ff4 | 5884 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
5885 | case sa_none: |
5886 | break; | |
5887 | } | |
5888 | } | |
3404c8d9 | 5889 | |
2109b99e AH |
5890 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
5891 | const struct cpumask *cpu_map) | |
5892 | { | |
dce840a0 PZ |
5893 | memset(d, 0, sizeof(*d)); |
5894 | ||
54ab4ff4 PZ |
5895 | if (__sdt_alloc(cpu_map)) |
5896 | return sa_sd_storage; | |
dce840a0 PZ |
5897 | d->sd = alloc_percpu(struct sched_domain *); |
5898 | if (!d->sd) | |
5899 | return sa_sd_storage; | |
2109b99e | 5900 | d->rd = alloc_rootdomain(); |
dce840a0 | 5901 | if (!d->rd) |
21d42ccf | 5902 | return sa_sd; |
2109b99e AH |
5903 | return sa_rootdomain; |
5904 | } | |
57d885fe | 5905 | |
dce840a0 PZ |
5906 | /* |
5907 | * NULL the sd_data elements we've used to build the sched_domain and | |
5908 | * sched_group structure so that the subsequent __free_domain_allocs() | |
5909 | * will not free the data we're using. | |
5910 | */ | |
5911 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
5912 | { | |
5913 | struct sd_data *sdd = sd->private; | |
dce840a0 PZ |
5914 | |
5915 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
5916 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
5917 | ||
e3589f6c | 5918 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
dce840a0 | 5919 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
e3589f6c | 5920 | |
63b2ca30 NP |
5921 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
5922 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; | |
dce840a0 PZ |
5923 | } |
5924 | ||
cb83b629 | 5925 | #ifdef CONFIG_NUMA |
cb83b629 | 5926 | static int sched_domains_numa_levels; |
cb83b629 PZ |
5927 | static int *sched_domains_numa_distance; |
5928 | static struct cpumask ***sched_domains_numa_masks; | |
5929 | static int sched_domains_curr_level; | |
143e1e28 | 5930 | #endif |
cb83b629 | 5931 | |
143e1e28 VG |
5932 | /* |
5933 | * SD_flags allowed in topology descriptions. | |
5934 | * | |
5d4dfddd | 5935 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
143e1e28 VG |
5936 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
5937 | * SD_NUMA - describes NUMA topologies | |
d77b3ed5 | 5938 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
143e1e28 VG |
5939 | * |
5940 | * Odd one out: | |
5941 | * SD_ASYM_PACKING - describes SMT quirks | |
5942 | */ | |
5943 | #define TOPOLOGY_SD_FLAGS \ | |
5d4dfddd | 5944 | (SD_SHARE_CPUCAPACITY | \ |
143e1e28 VG |
5945 | SD_SHARE_PKG_RESOURCES | \ |
5946 | SD_NUMA | \ | |
d77b3ed5 VG |
5947 | SD_ASYM_PACKING | \ |
5948 | SD_SHARE_POWERDOMAIN) | |
cb83b629 PZ |
5949 | |
5950 | static struct sched_domain * | |
143e1e28 | 5951 | sd_init(struct sched_domain_topology_level *tl, int cpu) |
cb83b629 PZ |
5952 | { |
5953 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); | |
143e1e28 VG |
5954 | int sd_weight, sd_flags = 0; |
5955 | ||
5956 | #ifdef CONFIG_NUMA | |
5957 | /* | |
5958 | * Ugly hack to pass state to sd_numa_mask()... | |
5959 | */ | |
5960 | sched_domains_curr_level = tl->numa_level; | |
5961 | #endif | |
5962 | ||
5963 | sd_weight = cpumask_weight(tl->mask(cpu)); | |
5964 | ||
5965 | if (tl->sd_flags) | |
5966 | sd_flags = (*tl->sd_flags)(); | |
5967 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, | |
5968 | "wrong sd_flags in topology description\n")) | |
5969 | sd_flags &= ~TOPOLOGY_SD_FLAGS; | |
cb83b629 PZ |
5970 | |
5971 | *sd = (struct sched_domain){ | |
5972 | .min_interval = sd_weight, | |
5973 | .max_interval = 2*sd_weight, | |
5974 | .busy_factor = 32, | |
870a0bb5 | 5975 | .imbalance_pct = 125, |
143e1e28 VG |
5976 | |
5977 | .cache_nice_tries = 0, | |
5978 | .busy_idx = 0, | |
5979 | .idle_idx = 0, | |
cb83b629 PZ |
5980 | .newidle_idx = 0, |
5981 | .wake_idx = 0, | |
5982 | .forkexec_idx = 0, | |
5983 | ||
5984 | .flags = 1*SD_LOAD_BALANCE | |
5985 | | 1*SD_BALANCE_NEWIDLE | |
143e1e28 VG |
5986 | | 1*SD_BALANCE_EXEC |
5987 | | 1*SD_BALANCE_FORK | |
cb83b629 | 5988 | | 0*SD_BALANCE_WAKE |
143e1e28 | 5989 | | 1*SD_WAKE_AFFINE |
5d4dfddd | 5990 | | 0*SD_SHARE_CPUCAPACITY |
cb83b629 | 5991 | | 0*SD_SHARE_PKG_RESOURCES |
143e1e28 | 5992 | | 0*SD_SERIALIZE |
cb83b629 | 5993 | | 0*SD_PREFER_SIBLING |
143e1e28 VG |
5994 | | 0*SD_NUMA |
5995 | | sd_flags | |
cb83b629 | 5996 | , |
143e1e28 | 5997 | |
cb83b629 PZ |
5998 | .last_balance = jiffies, |
5999 | .balance_interval = sd_weight, | |
143e1e28 | 6000 | .smt_gain = 0, |
2b4cfe64 JL |
6001 | .max_newidle_lb_cost = 0, |
6002 | .next_decay_max_lb_cost = jiffies, | |
143e1e28 VG |
6003 | #ifdef CONFIG_SCHED_DEBUG |
6004 | .name = tl->name, | |
6005 | #endif | |
cb83b629 | 6006 | }; |
cb83b629 PZ |
6007 | |
6008 | /* | |
143e1e28 | 6009 | * Convert topological properties into behaviour. |
cb83b629 | 6010 | */ |
143e1e28 | 6011 | |
5d4dfddd | 6012 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
143e1e28 VG |
6013 | sd->imbalance_pct = 110; |
6014 | sd->smt_gain = 1178; /* ~15% */ | |
143e1e28 VG |
6015 | |
6016 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { | |
6017 | sd->imbalance_pct = 117; | |
6018 | sd->cache_nice_tries = 1; | |
6019 | sd->busy_idx = 2; | |
6020 | ||
6021 | #ifdef CONFIG_NUMA | |
6022 | } else if (sd->flags & SD_NUMA) { | |
6023 | sd->cache_nice_tries = 2; | |
6024 | sd->busy_idx = 3; | |
6025 | sd->idle_idx = 2; | |
6026 | ||
6027 | sd->flags |= SD_SERIALIZE; | |
6028 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { | |
6029 | sd->flags &= ~(SD_BALANCE_EXEC | | |
6030 | SD_BALANCE_FORK | | |
6031 | SD_WAKE_AFFINE); | |
6032 | } | |
6033 | ||
6034 | #endif | |
6035 | } else { | |
6036 | sd->flags |= SD_PREFER_SIBLING; | |
6037 | sd->cache_nice_tries = 1; | |
6038 | sd->busy_idx = 2; | |
6039 | sd->idle_idx = 1; | |
6040 | } | |
6041 | ||
6042 | sd->private = &tl->data; | |
cb83b629 PZ |
6043 | |
6044 | return sd; | |
6045 | } | |
6046 | ||
143e1e28 VG |
6047 | /* |
6048 | * Topology list, bottom-up. | |
6049 | */ | |
6050 | static struct sched_domain_topology_level default_topology[] = { | |
6051 | #ifdef CONFIG_SCHED_SMT | |
6052 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, | |
6053 | #endif | |
6054 | #ifdef CONFIG_SCHED_MC | |
6055 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, | |
143e1e28 VG |
6056 | #endif |
6057 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, | |
6058 | { NULL, }, | |
6059 | }; | |
6060 | ||
6061 | struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
6062 | ||
6063 | #define for_each_sd_topology(tl) \ | |
6064 | for (tl = sched_domain_topology; tl->mask; tl++) | |
6065 | ||
6066 | void set_sched_topology(struct sched_domain_topology_level *tl) | |
6067 | { | |
6068 | sched_domain_topology = tl; | |
6069 | } | |
6070 | ||
6071 | #ifdef CONFIG_NUMA | |
6072 | ||
cb83b629 PZ |
6073 | static const struct cpumask *sd_numa_mask(int cpu) |
6074 | { | |
6075 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; | |
6076 | } | |
6077 | ||
d039ac60 PZ |
6078 | static void sched_numa_warn(const char *str) |
6079 | { | |
6080 | static int done = false; | |
6081 | int i,j; | |
6082 | ||
6083 | if (done) | |
6084 | return; | |
6085 | ||
6086 | done = true; | |
6087 | ||
6088 | printk(KERN_WARNING "ERROR: %s\n\n", str); | |
6089 | ||
6090 | for (i = 0; i < nr_node_ids; i++) { | |
6091 | printk(KERN_WARNING " "); | |
6092 | for (j = 0; j < nr_node_ids; j++) | |
6093 | printk(KERN_CONT "%02d ", node_distance(i,j)); | |
6094 | printk(KERN_CONT "\n"); | |
6095 | } | |
6096 | printk(KERN_WARNING "\n"); | |
6097 | } | |
6098 | ||
6099 | static bool find_numa_distance(int distance) | |
6100 | { | |
6101 | int i; | |
6102 | ||
6103 | if (distance == node_distance(0, 0)) | |
6104 | return true; | |
6105 | ||
6106 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6107 | if (sched_domains_numa_distance[i] == distance) | |
6108 | return true; | |
6109 | } | |
6110 | ||
6111 | return false; | |
6112 | } | |
6113 | ||
cb83b629 PZ |
6114 | static void sched_init_numa(void) |
6115 | { | |
6116 | int next_distance, curr_distance = node_distance(0, 0); | |
6117 | struct sched_domain_topology_level *tl; | |
6118 | int level = 0; | |
6119 | int i, j, k; | |
6120 | ||
cb83b629 PZ |
6121 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
6122 | if (!sched_domains_numa_distance) | |
6123 | return; | |
6124 | ||
6125 | /* | |
6126 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the | |
6127 | * unique distances in the node_distance() table. | |
6128 | * | |
6129 | * Assumes node_distance(0,j) includes all distances in | |
6130 | * node_distance(i,j) in order to avoid cubic time. | |
cb83b629 PZ |
6131 | */ |
6132 | next_distance = curr_distance; | |
6133 | for (i = 0; i < nr_node_ids; i++) { | |
6134 | for (j = 0; j < nr_node_ids; j++) { | |
d039ac60 PZ |
6135 | for (k = 0; k < nr_node_ids; k++) { |
6136 | int distance = node_distance(i, k); | |
6137 | ||
6138 | if (distance > curr_distance && | |
6139 | (distance < next_distance || | |
6140 | next_distance == curr_distance)) | |
6141 | next_distance = distance; | |
6142 | ||
6143 | /* | |
6144 | * While not a strong assumption it would be nice to know | |
6145 | * about cases where if node A is connected to B, B is not | |
6146 | * equally connected to A. | |
6147 | */ | |
6148 | if (sched_debug() && node_distance(k, i) != distance) | |
6149 | sched_numa_warn("Node-distance not symmetric"); | |
6150 | ||
6151 | if (sched_debug() && i && !find_numa_distance(distance)) | |
6152 | sched_numa_warn("Node-0 not representative"); | |
6153 | } | |
6154 | if (next_distance != curr_distance) { | |
6155 | sched_domains_numa_distance[level++] = next_distance; | |
6156 | sched_domains_numa_levels = level; | |
6157 | curr_distance = next_distance; | |
6158 | } else break; | |
cb83b629 | 6159 | } |
d039ac60 PZ |
6160 | |
6161 | /* | |
6162 | * In case of sched_debug() we verify the above assumption. | |
6163 | */ | |
6164 | if (!sched_debug()) | |
6165 | break; | |
cb83b629 PZ |
6166 | } |
6167 | /* | |
6168 | * 'level' contains the number of unique distances, excluding the | |
6169 | * identity distance node_distance(i,i). | |
6170 | * | |
28b4a521 | 6171 | * The sched_domains_numa_distance[] array includes the actual distance |
cb83b629 PZ |
6172 | * numbers. |
6173 | */ | |
6174 | ||
5f7865f3 TC |
6175 | /* |
6176 | * Here, we should temporarily reset sched_domains_numa_levels to 0. | |
6177 | * If it fails to allocate memory for array sched_domains_numa_masks[][], | |
6178 | * the array will contain less then 'level' members. This could be | |
6179 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] | |
6180 | * in other functions. | |
6181 | * | |
6182 | * We reset it to 'level' at the end of this function. | |
6183 | */ | |
6184 | sched_domains_numa_levels = 0; | |
6185 | ||
cb83b629 PZ |
6186 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
6187 | if (!sched_domains_numa_masks) | |
6188 | return; | |
6189 | ||
6190 | /* | |
6191 | * Now for each level, construct a mask per node which contains all | |
6192 | * cpus of nodes that are that many hops away from us. | |
6193 | */ | |
6194 | for (i = 0; i < level; i++) { | |
6195 | sched_domains_numa_masks[i] = | |
6196 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); | |
6197 | if (!sched_domains_numa_masks[i]) | |
6198 | return; | |
6199 | ||
6200 | for (j = 0; j < nr_node_ids; j++) { | |
2ea45800 | 6201 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
cb83b629 PZ |
6202 | if (!mask) |
6203 | return; | |
6204 | ||
6205 | sched_domains_numa_masks[i][j] = mask; | |
6206 | ||
6207 | for (k = 0; k < nr_node_ids; k++) { | |
dd7d8634 | 6208 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
cb83b629 PZ |
6209 | continue; |
6210 | ||
6211 | cpumask_or(mask, mask, cpumask_of_node(k)); | |
6212 | } | |
6213 | } | |
6214 | } | |
6215 | ||
143e1e28 VG |
6216 | /* Compute default topology size */ |
6217 | for (i = 0; sched_domain_topology[i].mask; i++); | |
6218 | ||
c515db8c | 6219 | tl = kzalloc((i + level + 1) * |
cb83b629 PZ |
6220 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
6221 | if (!tl) | |
6222 | return; | |
6223 | ||
6224 | /* | |
6225 | * Copy the default topology bits.. | |
6226 | */ | |
143e1e28 VG |
6227 | for (i = 0; sched_domain_topology[i].mask; i++) |
6228 | tl[i] = sched_domain_topology[i]; | |
cb83b629 PZ |
6229 | |
6230 | /* | |
6231 | * .. and append 'j' levels of NUMA goodness. | |
6232 | */ | |
6233 | for (j = 0; j < level; i++, j++) { | |
6234 | tl[i] = (struct sched_domain_topology_level){ | |
cb83b629 | 6235 | .mask = sd_numa_mask, |
143e1e28 | 6236 | .sd_flags = cpu_numa_flags, |
cb83b629 PZ |
6237 | .flags = SDTL_OVERLAP, |
6238 | .numa_level = j, | |
143e1e28 | 6239 | SD_INIT_NAME(NUMA) |
cb83b629 PZ |
6240 | }; |
6241 | } | |
6242 | ||
6243 | sched_domain_topology = tl; | |
5f7865f3 TC |
6244 | |
6245 | sched_domains_numa_levels = level; | |
cb83b629 | 6246 | } |
301a5cba TC |
6247 | |
6248 | static void sched_domains_numa_masks_set(int cpu) | |
6249 | { | |
6250 | int i, j; | |
6251 | int node = cpu_to_node(cpu); | |
6252 | ||
6253 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6254 | for (j = 0; j < nr_node_ids; j++) { | |
6255 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) | |
6256 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6257 | } | |
6258 | } | |
6259 | } | |
6260 | ||
6261 | static void sched_domains_numa_masks_clear(int cpu) | |
6262 | { | |
6263 | int i, j; | |
6264 | for (i = 0; i < sched_domains_numa_levels; i++) { | |
6265 | for (j = 0; j < nr_node_ids; j++) | |
6266 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); | |
6267 | } | |
6268 | } | |
6269 | ||
6270 | /* | |
6271 | * Update sched_domains_numa_masks[level][node] array when new cpus | |
6272 | * are onlined. | |
6273 | */ | |
6274 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6275 | unsigned long action, | |
6276 | void *hcpu) | |
6277 | { | |
6278 | int cpu = (long)hcpu; | |
6279 | ||
6280 | switch (action & ~CPU_TASKS_FROZEN) { | |
6281 | case CPU_ONLINE: | |
6282 | sched_domains_numa_masks_set(cpu); | |
6283 | break; | |
6284 | ||
6285 | case CPU_DEAD: | |
6286 | sched_domains_numa_masks_clear(cpu); | |
6287 | break; | |
6288 | ||
6289 | default: | |
6290 | return NOTIFY_DONE; | |
6291 | } | |
6292 | ||
6293 | return NOTIFY_OK; | |
cb83b629 PZ |
6294 | } |
6295 | #else | |
6296 | static inline void sched_init_numa(void) | |
6297 | { | |
6298 | } | |
301a5cba TC |
6299 | |
6300 | static int sched_domains_numa_masks_update(struct notifier_block *nfb, | |
6301 | unsigned long action, | |
6302 | void *hcpu) | |
6303 | { | |
6304 | return 0; | |
6305 | } | |
cb83b629 PZ |
6306 | #endif /* CONFIG_NUMA */ |
6307 | ||
54ab4ff4 PZ |
6308 | static int __sdt_alloc(const struct cpumask *cpu_map) |
6309 | { | |
6310 | struct sched_domain_topology_level *tl; | |
6311 | int j; | |
6312 | ||
27723a68 | 6313 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6314 | struct sd_data *sdd = &tl->data; |
6315 | ||
6316 | sdd->sd = alloc_percpu(struct sched_domain *); | |
6317 | if (!sdd->sd) | |
6318 | return -ENOMEM; | |
6319 | ||
6320 | sdd->sg = alloc_percpu(struct sched_group *); | |
6321 | if (!sdd->sg) | |
6322 | return -ENOMEM; | |
6323 | ||
63b2ca30 NP |
6324 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
6325 | if (!sdd->sgc) | |
9c3f75cb PZ |
6326 | return -ENOMEM; |
6327 | ||
54ab4ff4 PZ |
6328 | for_each_cpu(j, cpu_map) { |
6329 | struct sched_domain *sd; | |
6330 | struct sched_group *sg; | |
63b2ca30 | 6331 | struct sched_group_capacity *sgc; |
54ab4ff4 PZ |
6332 | |
6333 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
6334 | GFP_KERNEL, cpu_to_node(j)); | |
6335 | if (!sd) | |
6336 | return -ENOMEM; | |
6337 | ||
6338 | *per_cpu_ptr(sdd->sd, j) = sd; | |
6339 | ||
6340 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6341 | GFP_KERNEL, cpu_to_node(j)); | |
6342 | if (!sg) | |
6343 | return -ENOMEM; | |
6344 | ||
30b4e9eb IM |
6345 | sg->next = sg; |
6346 | ||
54ab4ff4 | 6347 | *per_cpu_ptr(sdd->sg, j) = sg; |
9c3f75cb | 6348 | |
63b2ca30 | 6349 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
9c3f75cb | 6350 | GFP_KERNEL, cpu_to_node(j)); |
63b2ca30 | 6351 | if (!sgc) |
9c3f75cb PZ |
6352 | return -ENOMEM; |
6353 | ||
63b2ca30 | 6354 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
54ab4ff4 PZ |
6355 | } |
6356 | } | |
6357 | ||
6358 | return 0; | |
6359 | } | |
6360 | ||
6361 | static void __sdt_free(const struct cpumask *cpu_map) | |
6362 | { | |
6363 | struct sched_domain_topology_level *tl; | |
6364 | int j; | |
6365 | ||
27723a68 | 6366 | for_each_sd_topology(tl) { |
54ab4ff4 PZ |
6367 | struct sd_data *sdd = &tl->data; |
6368 | ||
6369 | for_each_cpu(j, cpu_map) { | |
fb2cf2c6 | 6370 | struct sched_domain *sd; |
6371 | ||
6372 | if (sdd->sd) { | |
6373 | sd = *per_cpu_ptr(sdd->sd, j); | |
6374 | if (sd && (sd->flags & SD_OVERLAP)) | |
6375 | free_sched_groups(sd->groups, 0); | |
6376 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
6377 | } | |
6378 | ||
6379 | if (sdd->sg) | |
6380 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
63b2ca30 NP |
6381 | if (sdd->sgc) |
6382 | kfree(*per_cpu_ptr(sdd->sgc, j)); | |
54ab4ff4 PZ |
6383 | } |
6384 | free_percpu(sdd->sd); | |
fb2cf2c6 | 6385 | sdd->sd = NULL; |
54ab4ff4 | 6386 | free_percpu(sdd->sg); |
fb2cf2c6 | 6387 | sdd->sg = NULL; |
63b2ca30 NP |
6388 | free_percpu(sdd->sgc); |
6389 | sdd->sgc = NULL; | |
54ab4ff4 PZ |
6390 | } |
6391 | } | |
6392 | ||
2c402dc3 | 6393 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
4a850cbe VK |
6394 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
6395 | struct sched_domain *child, int cpu) | |
2c402dc3 | 6396 | { |
143e1e28 | 6397 | struct sched_domain *sd = sd_init(tl, cpu); |
2c402dc3 | 6398 | if (!sd) |
d069b916 | 6399 | return child; |
2c402dc3 | 6400 | |
2c402dc3 | 6401 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
60495e77 PZ |
6402 | if (child) { |
6403 | sd->level = child->level + 1; | |
6404 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 6405 | child->parent = sd; |
c75e0128 | 6406 | sd->child = child; |
60495e77 | 6407 | } |
a841f8ce | 6408 | set_domain_attribute(sd, attr); |
2c402dc3 PZ |
6409 | |
6410 | return sd; | |
6411 | } | |
6412 | ||
2109b99e AH |
6413 | /* |
6414 | * Build sched domains for a given set of cpus and attach the sched domains | |
6415 | * to the individual cpus | |
6416 | */ | |
dce840a0 PZ |
6417 | static int build_sched_domains(const struct cpumask *cpu_map, |
6418 | struct sched_domain_attr *attr) | |
2109b99e | 6419 | { |
1c632169 | 6420 | enum s_alloc alloc_state; |
dce840a0 | 6421 | struct sched_domain *sd; |
2109b99e | 6422 | struct s_data d; |
822ff793 | 6423 | int i, ret = -ENOMEM; |
9c1cfda2 | 6424 | |
2109b99e AH |
6425 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6426 | if (alloc_state != sa_rootdomain) | |
6427 | goto error; | |
9c1cfda2 | 6428 | |
dce840a0 | 6429 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 6430 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
6431 | struct sched_domain_topology_level *tl; |
6432 | ||
3bd65a80 | 6433 | sd = NULL; |
27723a68 | 6434 | for_each_sd_topology(tl) { |
4a850cbe | 6435 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
22da9569 VK |
6436 | if (tl == sched_domain_topology) |
6437 | *per_cpu_ptr(d.sd, i) = sd; | |
e3589f6c PZ |
6438 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
6439 | sd->flags |= SD_OVERLAP; | |
d110235d PZ |
6440 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
6441 | break; | |
e3589f6c | 6442 | } |
dce840a0 PZ |
6443 | } |
6444 | ||
6445 | /* Build the groups for the domains */ | |
6446 | for_each_cpu(i, cpu_map) { | |
6447 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
6448 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
e3589f6c PZ |
6449 | if (sd->flags & SD_OVERLAP) { |
6450 | if (build_overlap_sched_groups(sd, i)) | |
6451 | goto error; | |
6452 | } else { | |
6453 | if (build_sched_groups(sd, i)) | |
6454 | goto error; | |
6455 | } | |
1cf51902 | 6456 | } |
a06dadbe | 6457 | } |
9c1cfda2 | 6458 | |
ced549fa | 6459 | /* Calculate CPU capacity for physical packages and nodes */ |
a9c9a9b6 PZ |
6460 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
6461 | if (!cpumask_test_cpu(i, cpu_map)) | |
6462 | continue; | |
9c1cfda2 | 6463 | |
dce840a0 PZ |
6464 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
6465 | claim_allocations(i, sd); | |
63b2ca30 | 6466 | init_sched_groups_capacity(i, sd); |
dce840a0 | 6467 | } |
f712c0c7 | 6468 | } |
9c1cfda2 | 6469 | |
1da177e4 | 6470 | /* Attach the domains */ |
dce840a0 | 6471 | rcu_read_lock(); |
abcd083a | 6472 | for_each_cpu(i, cpu_map) { |
21d42ccf | 6473 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 6474 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6475 | } |
dce840a0 | 6476 | rcu_read_unlock(); |
51888ca2 | 6477 | |
822ff793 | 6478 | ret = 0; |
51888ca2 | 6479 | error: |
2109b99e | 6480 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 6481 | return ret; |
1da177e4 | 6482 | } |
029190c5 | 6483 | |
acc3f5d7 | 6484 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6485 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6486 | static struct sched_domain_attr *dattr_cur; |
6487 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6488 | |
6489 | /* | |
6490 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6491 | * cpumask) fails, then fallback to a single sched domain, |
6492 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 6493 | */ |
4212823f | 6494 | static cpumask_var_t fallback_doms; |
029190c5 | 6495 | |
ee79d1bd HC |
6496 | /* |
6497 | * arch_update_cpu_topology lets virtualized architectures update the | |
6498 | * cpu core maps. It is supposed to return 1 if the topology changed | |
6499 | * or 0 if it stayed the same. | |
6500 | */ | |
52f5684c | 6501 | int __weak arch_update_cpu_topology(void) |
22e52b07 | 6502 | { |
ee79d1bd | 6503 | return 0; |
22e52b07 HC |
6504 | } |
6505 | ||
acc3f5d7 RR |
6506 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
6507 | { | |
6508 | int i; | |
6509 | cpumask_var_t *doms; | |
6510 | ||
6511 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
6512 | if (!doms) | |
6513 | return NULL; | |
6514 | for (i = 0; i < ndoms; i++) { | |
6515 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
6516 | free_sched_domains(doms, i); | |
6517 | return NULL; | |
6518 | } | |
6519 | } | |
6520 | return doms; | |
6521 | } | |
6522 | ||
6523 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
6524 | { | |
6525 | unsigned int i; | |
6526 | for (i = 0; i < ndoms; i++) | |
6527 | free_cpumask_var(doms[i]); | |
6528 | kfree(doms); | |
6529 | } | |
6530 | ||
1a20ff27 | 6531 | /* |
41a2d6cf | 6532 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6533 | * For now this just excludes isolated cpus, but could be used to |
6534 | * exclude other special cases in the future. | |
1a20ff27 | 6535 | */ |
c4a8849a | 6536 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 6537 | { |
7378547f MM |
6538 | int err; |
6539 | ||
22e52b07 | 6540 | arch_update_cpu_topology(); |
029190c5 | 6541 | ndoms_cur = 1; |
acc3f5d7 | 6542 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 6543 | if (!doms_cur) |
acc3f5d7 RR |
6544 | doms_cur = &fallback_doms; |
6545 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
dce840a0 | 6546 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 6547 | register_sched_domain_sysctl(); |
7378547f MM |
6548 | |
6549 | return err; | |
1a20ff27 DG |
6550 | } |
6551 | ||
1a20ff27 DG |
6552 | /* |
6553 | * Detach sched domains from a group of cpus specified in cpu_map | |
6554 | * These cpus will now be attached to the NULL domain | |
6555 | */ | |
96f874e2 | 6556 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
6557 | { |
6558 | int i; | |
6559 | ||
dce840a0 | 6560 | rcu_read_lock(); |
abcd083a | 6561 | for_each_cpu(i, cpu_map) |
57d885fe | 6562 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 6563 | rcu_read_unlock(); |
1a20ff27 DG |
6564 | } |
6565 | ||
1d3504fc HS |
6566 | /* handle null as "default" */ |
6567 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
6568 | struct sched_domain_attr *new, int idx_new) | |
6569 | { | |
6570 | struct sched_domain_attr tmp; | |
6571 | ||
6572 | /* fast path */ | |
6573 | if (!new && !cur) | |
6574 | return 1; | |
6575 | ||
6576 | tmp = SD_ATTR_INIT; | |
6577 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
6578 | new ? (new + idx_new) : &tmp, | |
6579 | sizeof(struct sched_domain_attr)); | |
6580 | } | |
6581 | ||
029190c5 PJ |
6582 | /* |
6583 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6584 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6585 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6586 | * It destroys each deleted domain and builds each new domain. | |
6587 | * | |
acc3f5d7 | 6588 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
6589 | * The masks don't intersect (don't overlap.) We should setup one |
6590 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6591 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6592 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6593 | * it as it is. | |
6594 | * | |
acc3f5d7 RR |
6595 | * The passed in 'doms_new' should be allocated using |
6596 | * alloc_sched_domains. This routine takes ownership of it and will | |
6597 | * free_sched_domains it when done with it. If the caller failed the | |
6598 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
6599 | * and partition_sched_domains() will fallback to the single partition | |
6600 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 6601 | * |
96f874e2 | 6602 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
6603 | * ndoms_new == 0 is a special case for destroying existing domains, |
6604 | * and it will not create the default domain. | |
dfb512ec | 6605 | * |
029190c5 PJ |
6606 | * Call with hotplug lock held |
6607 | */ | |
acc3f5d7 | 6608 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 6609 | struct sched_domain_attr *dattr_new) |
029190c5 | 6610 | { |
dfb512ec | 6611 | int i, j, n; |
d65bd5ec | 6612 | int new_topology; |
029190c5 | 6613 | |
712555ee | 6614 | mutex_lock(&sched_domains_mutex); |
a1835615 | 6615 | |
7378547f MM |
6616 | /* always unregister in case we don't destroy any domains */ |
6617 | unregister_sched_domain_sysctl(); | |
6618 | ||
d65bd5ec HC |
6619 | /* Let architecture update cpu core mappings. */ |
6620 | new_topology = arch_update_cpu_topology(); | |
6621 | ||
dfb512ec | 6622 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
6623 | |
6624 | /* Destroy deleted domains */ | |
6625 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 6626 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6627 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 6628 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
6629 | goto match1; |
6630 | } | |
6631 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 6632 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
6633 | match1: |
6634 | ; | |
6635 | } | |
6636 | ||
c8d2d47a | 6637 | n = ndoms_cur; |
e761b772 | 6638 | if (doms_new == NULL) { |
c8d2d47a | 6639 | n = 0; |
acc3f5d7 | 6640 | doms_new = &fallback_doms; |
6ad4c188 | 6641 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 6642 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
6643 | } |
6644 | ||
029190c5 PJ |
6645 | /* Build new domains */ |
6646 | for (i = 0; i < ndoms_new; i++) { | |
c8d2d47a | 6647 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 6648 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 6649 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
6650 | goto match2; |
6651 | } | |
6652 | /* no match - add a new doms_new */ | |
dce840a0 | 6653 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
6654 | match2: |
6655 | ; | |
6656 | } | |
6657 | ||
6658 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
6659 | if (doms_cur != &fallback_doms) |
6660 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 6661 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 6662 | doms_cur = doms_new; |
1d3504fc | 6663 | dattr_cur = dattr_new; |
029190c5 | 6664 | ndoms_cur = ndoms_new; |
7378547f MM |
6665 | |
6666 | register_sched_domain_sysctl(); | |
a1835615 | 6667 | |
712555ee | 6668 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
6669 | } |
6670 | ||
d35be8ba SB |
6671 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
6672 | ||
1da177e4 | 6673 | /* |
3a101d05 TH |
6674 | * Update cpusets according to cpu_active mask. If cpusets are |
6675 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6676 | * around partition_sched_domains(). | |
d35be8ba SB |
6677 | * |
6678 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6679 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6680 | */ |
0b2e918a TH |
6681 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
6682 | void *hcpu) | |
e761b772 | 6683 | { |
d35be8ba SB |
6684 | switch (action) { |
6685 | case CPU_ONLINE_FROZEN: | |
6686 | case CPU_DOWN_FAILED_FROZEN: | |
6687 | ||
6688 | /* | |
6689 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6690 | * resume sequence. As long as this is not the last online | |
6691 | * operation in the resume sequence, just build a single sched | |
6692 | * domain, ignoring cpusets. | |
6693 | */ | |
6694 | num_cpus_frozen--; | |
6695 | if (likely(num_cpus_frozen)) { | |
6696 | partition_sched_domains(1, NULL, NULL); | |
6697 | break; | |
6698 | } | |
6699 | ||
6700 | /* | |
6701 | * This is the last CPU online operation. So fall through and | |
6702 | * restore the original sched domains by considering the | |
6703 | * cpuset configurations. | |
6704 | */ | |
6705 | ||
e761b772 | 6706 | case CPU_ONLINE: |
6ad4c188 | 6707 | case CPU_DOWN_FAILED: |
7ddf96b0 | 6708 | cpuset_update_active_cpus(true); |
d35be8ba | 6709 | break; |
3a101d05 TH |
6710 | default: |
6711 | return NOTIFY_DONE; | |
6712 | } | |
d35be8ba | 6713 | return NOTIFY_OK; |
3a101d05 | 6714 | } |
e761b772 | 6715 | |
0b2e918a TH |
6716 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
6717 | void *hcpu) | |
3a101d05 | 6718 | { |
d35be8ba | 6719 | switch (action) { |
3a101d05 | 6720 | case CPU_DOWN_PREPARE: |
7ddf96b0 | 6721 | cpuset_update_active_cpus(false); |
d35be8ba SB |
6722 | break; |
6723 | case CPU_DOWN_PREPARE_FROZEN: | |
6724 | num_cpus_frozen++; | |
6725 | partition_sched_domains(1, NULL, NULL); | |
6726 | break; | |
e761b772 MK |
6727 | default: |
6728 | return NOTIFY_DONE; | |
6729 | } | |
d35be8ba | 6730 | return NOTIFY_OK; |
e761b772 | 6731 | } |
e761b772 | 6732 | |
1da177e4 LT |
6733 | void __init sched_init_smp(void) |
6734 | { | |
dcc30a35 RR |
6735 | cpumask_var_t non_isolated_cpus; |
6736 | ||
6737 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 6738 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 6739 | |
cb83b629 PZ |
6740 | sched_init_numa(); |
6741 | ||
6acce3ef PZ |
6742 | /* |
6743 | * There's no userspace yet to cause hotplug operations; hence all the | |
6744 | * cpu masks are stable and all blatant races in the below code cannot | |
6745 | * happen. | |
6746 | */ | |
712555ee | 6747 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 6748 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
6749 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
6750 | if (cpumask_empty(non_isolated_cpus)) | |
6751 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 6752 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6753 | |
301a5cba | 6754 | hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE); |
3a101d05 TH |
6755 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
6756 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 | 6757 | |
b328ca18 | 6758 | init_hrtick(); |
5c1e1767 NP |
6759 | |
6760 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 6761 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 6762 | BUG(); |
19978ca6 | 6763 | sched_init_granularity(); |
dcc30a35 | 6764 | free_cpumask_var(non_isolated_cpus); |
4212823f | 6765 | |
0e3900e6 | 6766 | init_sched_rt_class(); |
1baca4ce | 6767 | init_sched_dl_class(); |
1da177e4 LT |
6768 | } |
6769 | #else | |
6770 | void __init sched_init_smp(void) | |
6771 | { | |
19978ca6 | 6772 | sched_init_granularity(); |
1da177e4 LT |
6773 | } |
6774 | #endif /* CONFIG_SMP */ | |
6775 | ||
cd1bb94b AB |
6776 | const_debug unsigned int sysctl_timer_migration = 1; |
6777 | ||
1da177e4 LT |
6778 | int in_sched_functions(unsigned long addr) |
6779 | { | |
1da177e4 LT |
6780 | return in_lock_functions(addr) || |
6781 | (addr >= (unsigned long)__sched_text_start | |
6782 | && addr < (unsigned long)__sched_text_end); | |
6783 | } | |
6784 | ||
029632fb | 6785 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6786 | /* |
6787 | * Default task group. | |
6788 | * Every task in system belongs to this group at bootup. | |
6789 | */ | |
029632fb | 6790 | struct task_group root_task_group; |
35cf4e50 | 6791 | LIST_HEAD(task_groups); |
052f1dc7 | 6792 | #endif |
6f505b16 | 6793 | |
e6252c3e | 6794 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
6f505b16 | 6795 | |
1da177e4 LT |
6796 | void __init sched_init(void) |
6797 | { | |
dd41f596 | 6798 | int i, j; |
434d53b0 MT |
6799 | unsigned long alloc_size = 0, ptr; |
6800 | ||
6801 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6802 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
6803 | #endif | |
6804 | #ifdef CONFIG_RT_GROUP_SCHED | |
6805 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 6806 | #endif |
df7c8e84 | 6807 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 6808 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 6809 | #endif |
434d53b0 | 6810 | if (alloc_size) { |
36b7b6d4 | 6811 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
6812 | |
6813 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6814 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6815 | ptr += nr_cpu_ids * sizeof(void **); |
6816 | ||
07e06b01 | 6817 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6818 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6819 | |
6d6bc0ad | 6820 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6821 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6822 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6823 | ptr += nr_cpu_ids * sizeof(void **); |
6824 | ||
07e06b01 | 6825 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6826 | ptr += nr_cpu_ids * sizeof(void **); |
6827 | ||
6d6bc0ad | 6828 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
6829 | #ifdef CONFIG_CPUMASK_OFFSTACK |
6830 | for_each_possible_cpu(i) { | |
e6252c3e | 6831 | per_cpu(load_balance_mask, i) = (void *)ptr; |
df7c8e84 RR |
6832 | ptr += cpumask_size(); |
6833 | } | |
6834 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 6835 | } |
dd41f596 | 6836 | |
332ac17e DF |
6837 | init_rt_bandwidth(&def_rt_bandwidth, |
6838 | global_rt_period(), global_rt_runtime()); | |
6839 | init_dl_bandwidth(&def_dl_bandwidth, | |
1724813d | 6840 | global_rt_period(), global_rt_runtime()); |
332ac17e | 6841 | |
57d885fe GH |
6842 | #ifdef CONFIG_SMP |
6843 | init_defrootdomain(); | |
6844 | #endif | |
6845 | ||
d0b27fa7 | 6846 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6847 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6848 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6849 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6850 | |
7c941438 | 6851 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
6852 | list_add(&root_task_group.list, &task_groups); |
6853 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6854 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6855 | autogroup_init(&init_task); |
54c707e9 | 6856 | |
7c941438 | 6857 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6858 | |
0a945022 | 6859 | for_each_possible_cpu(i) { |
70b97a7f | 6860 | struct rq *rq; |
1da177e4 LT |
6861 | |
6862 | rq = cpu_rq(i); | |
05fa785c | 6863 | raw_spin_lock_init(&rq->lock); |
7897986b | 6864 | rq->nr_running = 0; |
dce48a84 TG |
6865 | rq->calc_load_active = 0; |
6866 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6867 | init_cfs_rq(&rq->cfs); |
6f505b16 | 6868 | init_rt_rq(&rq->rt, rq); |
aab03e05 | 6869 | init_dl_rq(&rq->dl, rq); |
dd41f596 | 6870 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6871 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6872 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 6873 | /* |
07e06b01 | 6874 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
6875 | * |
6876 | * In case of task-groups formed thr' the cgroup filesystem, it | |
6877 | * gets 100% of the cpu resources in the system. This overall | |
6878 | * system cpu resource is divided among the tasks of | |
07e06b01 | 6879 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6880 | * based on each entity's (task or task-group's) weight |
6881 | * (se->load.weight). | |
6882 | * | |
07e06b01 | 6883 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
6884 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
6885 | * then A0's share of the cpu resource is: | |
6886 | * | |
0d905bca | 6887 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6888 | * |
07e06b01 YZ |
6889 | * We achieve this by letting root_task_group's tasks sit |
6890 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6891 | */ |
ab84d31e | 6892 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6893 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6894 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6895 | ||
6896 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6897 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6898 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6899 | #endif |
1da177e4 | 6900 | |
dd41f596 IM |
6901 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6902 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
6903 | |
6904 | rq->last_load_update_tick = jiffies; | |
6905 | ||
1da177e4 | 6906 | #ifdef CONFIG_SMP |
41c7ce9a | 6907 | rq->sd = NULL; |
57d885fe | 6908 | rq->rd = NULL; |
ca8ce3d0 | 6909 | rq->cpu_capacity = SCHED_CAPACITY_SCALE; |
3f029d3c | 6910 | rq->post_schedule = 0; |
1da177e4 | 6911 | rq->active_balance = 0; |
dd41f596 | 6912 | rq->next_balance = jiffies; |
1da177e4 | 6913 | rq->push_cpu = 0; |
0a2966b4 | 6914 | rq->cpu = i; |
1f11eb6a | 6915 | rq->online = 0; |
eae0c9df MG |
6916 | rq->idle_stamp = 0; |
6917 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6918 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6919 | |
6920 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6921 | ||
dc938520 | 6922 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6923 | #ifdef CONFIG_NO_HZ_COMMON |
1c792db7 | 6924 | rq->nohz_flags = 0; |
83cd4fe2 | 6925 | #endif |
265f22a9 FW |
6926 | #ifdef CONFIG_NO_HZ_FULL |
6927 | rq->last_sched_tick = 0; | |
6928 | #endif | |
1da177e4 | 6929 | #endif |
8f4d37ec | 6930 | init_rq_hrtick(rq); |
1da177e4 | 6931 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6932 | } |
6933 | ||
2dd73a4f | 6934 | set_load_weight(&init_task); |
b50f60ce | 6935 | |
e107be36 AK |
6936 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6937 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6938 | #endif | |
6939 | ||
1da177e4 LT |
6940 | /* |
6941 | * The boot idle thread does lazy MMU switching as well: | |
6942 | */ | |
6943 | atomic_inc(&init_mm.mm_count); | |
6944 | enter_lazy_tlb(&init_mm, current); | |
6945 | ||
6946 | /* | |
6947 | * Make us the idle thread. Technically, schedule() should not be | |
6948 | * called from this thread, however somewhere below it might be, | |
6949 | * but because we are the idle thread, we just pick up running again | |
6950 | * when this runqueue becomes "idle". | |
6951 | */ | |
6952 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6953 | |
6954 | calc_load_update = jiffies + LOAD_FREQ; | |
6955 | ||
dd41f596 IM |
6956 | /* |
6957 | * During early bootup we pretend to be a normal task: | |
6958 | */ | |
6959 | current->sched_class = &fair_sched_class; | |
6892b75e | 6960 | |
bf4d83f6 | 6961 | #ifdef CONFIG_SMP |
4cb98839 | 6962 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6963 | /* May be allocated at isolcpus cmdline parse time */ |
6964 | if (cpu_isolated_map == NULL) | |
6965 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6966 | idle_thread_set_boot_cpu(); |
a803f026 | 6967 | set_cpu_rq_start_time(); |
029632fb PZ |
6968 | #endif |
6969 | init_sched_fair_class(); | |
6a7b3dc3 | 6970 | |
6892b75e | 6971 | scheduler_running = 1; |
1da177e4 LT |
6972 | } |
6973 | ||
d902db1e | 6974 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6975 | static inline int preempt_count_equals(int preempt_offset) |
6976 | { | |
234da7bc | 6977 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 6978 | |
4ba8216c | 6979 | return (nested == preempt_offset); |
e4aafea2 FW |
6980 | } |
6981 | ||
d894837f | 6982 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6983 | { |
1da177e4 LT |
6984 | static unsigned long prev_jiffy; /* ratelimiting */ |
6985 | ||
b3fbab05 | 6986 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
db273be2 TG |
6987 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
6988 | !is_idle_task(current)) || | |
e4aafea2 | 6989 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
6990 | return; |
6991 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6992 | return; | |
6993 | prev_jiffy = jiffies; | |
6994 | ||
3df0fc5b PZ |
6995 | printk(KERN_ERR |
6996 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6997 | file, line); | |
6998 | printk(KERN_ERR | |
6999 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7000 | in_atomic(), irqs_disabled(), | |
7001 | current->pid, current->comm); | |
aef745fc IM |
7002 | |
7003 | debug_show_held_locks(current); | |
7004 | if (irqs_disabled()) | |
7005 | print_irqtrace_events(current); | |
8f47b187 TG |
7006 | #ifdef CONFIG_DEBUG_PREEMPT |
7007 | if (!preempt_count_equals(preempt_offset)) { | |
7008 | pr_err("Preemption disabled at:"); | |
7009 | print_ip_sym(current->preempt_disable_ip); | |
7010 | pr_cont("\n"); | |
7011 | } | |
7012 | #endif | |
aef745fc | 7013 | dump_stack(); |
1da177e4 LT |
7014 | } |
7015 | EXPORT_SYMBOL(__might_sleep); | |
7016 | #endif | |
7017 | ||
7018 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7019 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7020 | { | |
da7a735e | 7021 | const struct sched_class *prev_class = p->sched_class; |
d50dde5a DF |
7022 | struct sched_attr attr = { |
7023 | .sched_policy = SCHED_NORMAL, | |
7024 | }; | |
da7a735e | 7025 | int old_prio = p->prio; |
3a5e4dc1 | 7026 | int on_rq; |
3e51f33f | 7027 | |
fd2f4419 | 7028 | on_rq = p->on_rq; |
3a5e4dc1 | 7029 | if (on_rq) |
4ca9b72b | 7030 | dequeue_task(rq, p, 0); |
d50dde5a | 7031 | __setscheduler(rq, p, &attr); |
3a5e4dc1 | 7032 | if (on_rq) { |
4ca9b72b | 7033 | enqueue_task(rq, p, 0); |
3a5e4dc1 AK |
7034 | resched_task(rq->curr); |
7035 | } | |
da7a735e PZ |
7036 | |
7037 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
7038 | } |
7039 | ||
1da177e4 LT |
7040 | void normalize_rt_tasks(void) |
7041 | { | |
a0f98a1c | 7042 | struct task_struct *g, *p; |
1da177e4 | 7043 | unsigned long flags; |
70b97a7f | 7044 | struct rq *rq; |
1da177e4 | 7045 | |
4cf5d77a | 7046 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7047 | do_each_thread(g, p) { |
178be793 IM |
7048 | /* |
7049 | * Only normalize user tasks: | |
7050 | */ | |
7051 | if (!p->mm) | |
7052 | continue; | |
7053 | ||
6cfb0d5d | 7054 | p->se.exec_start = 0; |
6cfb0d5d | 7055 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7056 | p->se.statistics.wait_start = 0; |
7057 | p->se.statistics.sleep_start = 0; | |
7058 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7059 | #endif |
dd41f596 | 7060 | |
aab03e05 | 7061 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7062 | /* |
7063 | * Renice negative nice level userspace | |
7064 | * tasks back to 0: | |
7065 | */ | |
d0ea0268 | 7066 | if (task_nice(p) < 0 && p->mm) |
dd41f596 | 7067 | set_user_nice(p, 0); |
1da177e4 | 7068 | continue; |
dd41f596 | 7069 | } |
1da177e4 | 7070 | |
1d615482 | 7071 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7072 | rq = __task_rq_lock(p); |
1da177e4 | 7073 | |
178be793 | 7074 | normalize_task(rq, p); |
3a5e4dc1 | 7075 | |
b29739f9 | 7076 | __task_rq_unlock(rq); |
1d615482 | 7077 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7078 | } while_each_thread(g, p); |
7079 | ||
4cf5d77a | 7080 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7081 | } |
7082 | ||
7083 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7084 | |
67fc4e0c | 7085 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7086 | /* |
67fc4e0c | 7087 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7088 | * |
7089 | * They can only be called when the whole system has been | |
7090 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7091 | * activity can take place. Using them for anything else would | |
7092 | * be a serious bug, and as a result, they aren't even visible | |
7093 | * under any other configuration. | |
7094 | */ | |
7095 | ||
7096 | /** | |
7097 | * curr_task - return the current task for a given cpu. | |
7098 | * @cpu: the processor in question. | |
7099 | * | |
7100 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7101 | * |
7102 | * Return: The current task for @cpu. | |
1df5c10a | 7103 | */ |
36c8b586 | 7104 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7105 | { |
7106 | return cpu_curr(cpu); | |
7107 | } | |
7108 | ||
67fc4e0c JW |
7109 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7110 | ||
7111 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7112 | /** |
7113 | * set_curr_task - set the current task for a given cpu. | |
7114 | * @cpu: the processor in question. | |
7115 | * @p: the task pointer to set. | |
7116 | * | |
7117 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7118 | * are serviced on a separate stack. It allows the architecture to switch the |
7119 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7120 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7121 | * and caller must save the original value of the current task (see | |
7122 | * curr_task() above) and restore that value before reenabling interrupts and | |
7123 | * re-starting the system. | |
7124 | * | |
7125 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7126 | */ | |
36c8b586 | 7127 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7128 | { |
7129 | cpu_curr(cpu) = p; | |
7130 | } | |
7131 | ||
7132 | #endif | |
29f59db3 | 7133 | |
7c941438 | 7134 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7135 | /* task_group_lock serializes the addition/removal of task groups */ |
7136 | static DEFINE_SPINLOCK(task_group_lock); | |
7137 | ||
bccbe08a PZ |
7138 | static void free_sched_group(struct task_group *tg) |
7139 | { | |
7140 | free_fair_sched_group(tg); | |
7141 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7142 | autogroup_free(tg); |
bccbe08a PZ |
7143 | kfree(tg); |
7144 | } | |
7145 | ||
7146 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7147 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7148 | { |
7149 | struct task_group *tg; | |
bccbe08a PZ |
7150 | |
7151 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7152 | if (!tg) | |
7153 | return ERR_PTR(-ENOMEM); | |
7154 | ||
ec7dc8ac | 7155 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7156 | goto err; |
7157 | ||
ec7dc8ac | 7158 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7159 | goto err; |
7160 | ||
ace783b9 LZ |
7161 | return tg; |
7162 | ||
7163 | err: | |
7164 | free_sched_group(tg); | |
7165 | return ERR_PTR(-ENOMEM); | |
7166 | } | |
7167 | ||
7168 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7169 | { | |
7170 | unsigned long flags; | |
7171 | ||
8ed36996 | 7172 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7173 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
7174 | |
7175 | WARN_ON(!parent); /* root should already exist */ | |
7176 | ||
7177 | tg->parent = parent; | |
f473aa5e | 7178 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7179 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7180 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7181 | } |
7182 | ||
9b5b7751 | 7183 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 7184 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7185 | { |
29f59db3 | 7186 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 7187 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7188 | } |
7189 | ||
9b5b7751 | 7190 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7191 | void sched_destroy_group(struct task_group *tg) |
ace783b9 LZ |
7192 | { |
7193 | /* wait for possible concurrent references to cfs_rqs complete */ | |
7194 | call_rcu(&tg->rcu, free_sched_group_rcu); | |
7195 | } | |
7196 | ||
7197 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7198 | { |
8ed36996 | 7199 | unsigned long flags; |
9b5b7751 | 7200 | int i; |
29f59db3 | 7201 | |
3d4b47b4 PZ |
7202 | /* end participation in shares distribution */ |
7203 | for_each_possible_cpu(i) | |
bccbe08a | 7204 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
7205 | |
7206 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7207 | list_del_rcu(&tg->list); |
f473aa5e | 7208 | list_del_rcu(&tg->siblings); |
8ed36996 | 7209 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7210 | } |
7211 | ||
9b5b7751 | 7212 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7213 | * The caller of this function should have put the task in its new group |
7214 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7215 | * reflect its new group. | |
9b5b7751 SV |
7216 | */ |
7217 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 | 7218 | { |
8323f26c | 7219 | struct task_group *tg; |
29f59db3 SV |
7220 | int on_rq, running; |
7221 | unsigned long flags; | |
7222 | struct rq *rq; | |
7223 | ||
7224 | rq = task_rq_lock(tsk, &flags); | |
7225 | ||
051a1d1a | 7226 | running = task_current(rq, tsk); |
fd2f4419 | 7227 | on_rq = tsk->on_rq; |
29f59db3 | 7228 | |
0e1f3483 | 7229 | if (on_rq) |
29f59db3 | 7230 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
7231 | if (unlikely(running)) |
7232 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 7233 | |
073219e9 | 7234 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, |
8323f26c PZ |
7235 | lockdep_is_held(&tsk->sighand->siglock)), |
7236 | struct task_group, css); | |
7237 | tg = autogroup_task_group(tsk, tg); | |
7238 | tsk->sched_task_group = tg; | |
7239 | ||
810b3817 | 7240 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
7241 | if (tsk->sched_class->task_move_group) |
7242 | tsk->sched_class->task_move_group(tsk, on_rq); | |
7243 | else | |
810b3817 | 7244 | #endif |
b2b5ce02 | 7245 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 7246 | |
0e1f3483 HS |
7247 | if (unlikely(running)) |
7248 | tsk->sched_class->set_curr_task(rq); | |
7249 | if (on_rq) | |
371fd7e7 | 7250 | enqueue_task(rq, tsk, 0); |
29f59db3 | 7251 | |
0122ec5b | 7252 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 7253 | } |
7c941438 | 7254 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 7255 | |
a790de99 PT |
7256 | #ifdef CONFIG_RT_GROUP_SCHED |
7257 | /* | |
7258 | * Ensure that the real time constraints are schedulable. | |
7259 | */ | |
7260 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 7261 | |
9a7e0b18 PZ |
7262 | /* Must be called with tasklist_lock held */ |
7263 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 7264 | { |
9a7e0b18 | 7265 | struct task_struct *g, *p; |
b40b2e8e | 7266 | |
9a7e0b18 | 7267 | do_each_thread(g, p) { |
029632fb | 7268 | if (rt_task(p) && task_rq(p)->rt.tg == tg) |
9a7e0b18 PZ |
7269 | return 1; |
7270 | } while_each_thread(g, p); | |
b40b2e8e | 7271 | |
9a7e0b18 PZ |
7272 | return 0; |
7273 | } | |
b40b2e8e | 7274 | |
9a7e0b18 PZ |
7275 | struct rt_schedulable_data { |
7276 | struct task_group *tg; | |
7277 | u64 rt_period; | |
7278 | u64 rt_runtime; | |
7279 | }; | |
b40b2e8e | 7280 | |
a790de99 | 7281 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
7282 | { |
7283 | struct rt_schedulable_data *d = data; | |
7284 | struct task_group *child; | |
7285 | unsigned long total, sum = 0; | |
7286 | u64 period, runtime; | |
b40b2e8e | 7287 | |
9a7e0b18 PZ |
7288 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
7289 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 7290 | |
9a7e0b18 PZ |
7291 | if (tg == d->tg) { |
7292 | period = d->rt_period; | |
7293 | runtime = d->rt_runtime; | |
b40b2e8e | 7294 | } |
b40b2e8e | 7295 | |
4653f803 PZ |
7296 | /* |
7297 | * Cannot have more runtime than the period. | |
7298 | */ | |
7299 | if (runtime > period && runtime != RUNTIME_INF) | |
7300 | return -EINVAL; | |
6f505b16 | 7301 | |
4653f803 PZ |
7302 | /* |
7303 | * Ensure we don't starve existing RT tasks. | |
7304 | */ | |
9a7e0b18 PZ |
7305 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
7306 | return -EBUSY; | |
6f505b16 | 7307 | |
9a7e0b18 | 7308 | total = to_ratio(period, runtime); |
6f505b16 | 7309 | |
4653f803 PZ |
7310 | /* |
7311 | * Nobody can have more than the global setting allows. | |
7312 | */ | |
7313 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
7314 | return -EINVAL; | |
6f505b16 | 7315 | |
4653f803 PZ |
7316 | /* |
7317 | * The sum of our children's runtime should not exceed our own. | |
7318 | */ | |
9a7e0b18 PZ |
7319 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
7320 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
7321 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 7322 | |
9a7e0b18 PZ |
7323 | if (child == d->tg) { |
7324 | period = d->rt_period; | |
7325 | runtime = d->rt_runtime; | |
7326 | } | |
6f505b16 | 7327 | |
9a7e0b18 | 7328 | sum += to_ratio(period, runtime); |
9f0c1e56 | 7329 | } |
6f505b16 | 7330 | |
9a7e0b18 PZ |
7331 | if (sum > total) |
7332 | return -EINVAL; | |
7333 | ||
7334 | return 0; | |
6f505b16 PZ |
7335 | } |
7336 | ||
9a7e0b18 | 7337 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 7338 | { |
8277434e PT |
7339 | int ret; |
7340 | ||
9a7e0b18 PZ |
7341 | struct rt_schedulable_data data = { |
7342 | .tg = tg, | |
7343 | .rt_period = period, | |
7344 | .rt_runtime = runtime, | |
7345 | }; | |
7346 | ||
8277434e PT |
7347 | rcu_read_lock(); |
7348 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
7349 | rcu_read_unlock(); | |
7350 | ||
7351 | return ret; | |
521f1a24 DG |
7352 | } |
7353 | ||
ab84d31e | 7354 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 7355 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 7356 | { |
ac086bc2 | 7357 | int i, err = 0; |
9f0c1e56 | 7358 | |
9f0c1e56 | 7359 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 7360 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
7361 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
7362 | if (err) | |
9f0c1e56 | 7363 | goto unlock; |
ac086bc2 | 7364 | |
0986b11b | 7365 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
7366 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
7367 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
7368 | |
7369 | for_each_possible_cpu(i) { | |
7370 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
7371 | ||
0986b11b | 7372 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7373 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 7374 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7375 | } |
0986b11b | 7376 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 7377 | unlock: |
521f1a24 | 7378 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
7379 | mutex_unlock(&rt_constraints_mutex); |
7380 | ||
7381 | return err; | |
6f505b16 PZ |
7382 | } |
7383 | ||
25cc7da7 | 7384 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
7385 | { |
7386 | u64 rt_runtime, rt_period; | |
7387 | ||
7388 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7389 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
7390 | if (rt_runtime_us < 0) | |
7391 | rt_runtime = RUNTIME_INF; | |
7392 | ||
ab84d31e | 7393 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7394 | } |
7395 | ||
25cc7da7 | 7396 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
7397 | { |
7398 | u64 rt_runtime_us; | |
7399 | ||
d0b27fa7 | 7400 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
7401 | return -1; |
7402 | ||
d0b27fa7 | 7403 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
7404 | do_div(rt_runtime_us, NSEC_PER_USEC); |
7405 | return rt_runtime_us; | |
7406 | } | |
d0b27fa7 | 7407 | |
25cc7da7 | 7408 | static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) |
d0b27fa7 PZ |
7409 | { |
7410 | u64 rt_runtime, rt_period; | |
7411 | ||
7412 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
7413 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
7414 | ||
619b0488 R |
7415 | if (rt_period == 0) |
7416 | return -EINVAL; | |
7417 | ||
ab84d31e | 7418 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
7419 | } |
7420 | ||
25cc7da7 | 7421 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
7422 | { |
7423 | u64 rt_period_us; | |
7424 | ||
7425 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
7426 | do_div(rt_period_us, NSEC_PER_USEC); | |
7427 | return rt_period_us; | |
7428 | } | |
332ac17e | 7429 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7430 | |
332ac17e | 7431 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
7432 | static int sched_rt_global_constraints(void) |
7433 | { | |
7434 | int ret = 0; | |
7435 | ||
7436 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 7437 | read_lock(&tasklist_lock); |
4653f803 | 7438 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 7439 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
7440 | mutex_unlock(&rt_constraints_mutex); |
7441 | ||
7442 | return ret; | |
7443 | } | |
54e99124 | 7444 | |
25cc7da7 | 7445 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
7446 | { |
7447 | /* Don't accept realtime tasks when there is no way for them to run */ | |
7448 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
7449 | return 0; | |
7450 | ||
7451 | return 1; | |
7452 | } | |
7453 | ||
6d6bc0ad | 7454 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
7455 | static int sched_rt_global_constraints(void) |
7456 | { | |
ac086bc2 | 7457 | unsigned long flags; |
332ac17e | 7458 | int i, ret = 0; |
ec5d4989 | 7459 | |
0986b11b | 7460 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
7461 | for_each_possible_cpu(i) { |
7462 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
7463 | ||
0986b11b | 7464 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7465 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 7466 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 7467 | } |
0986b11b | 7468 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 7469 | |
332ac17e | 7470 | return ret; |
d0b27fa7 | 7471 | } |
6d6bc0ad | 7472 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7473 | |
332ac17e DF |
7474 | static int sched_dl_global_constraints(void) |
7475 | { | |
1724813d PZ |
7476 | u64 runtime = global_rt_runtime(); |
7477 | u64 period = global_rt_period(); | |
332ac17e | 7478 | u64 new_bw = to_ratio(period, runtime); |
1724813d | 7479 | int cpu, ret = 0; |
49516342 | 7480 | unsigned long flags; |
332ac17e DF |
7481 | |
7482 | /* | |
7483 | * Here we want to check the bandwidth not being set to some | |
7484 | * value smaller than the currently allocated bandwidth in | |
7485 | * any of the root_domains. | |
7486 | * | |
7487 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
7488 | * cycling on root_domains... Discussion on different/better | |
7489 | * solutions is welcome! | |
7490 | */ | |
1724813d PZ |
7491 | for_each_possible_cpu(cpu) { |
7492 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
332ac17e | 7493 | |
49516342 | 7494 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
7495 | if (new_bw < dl_b->total_bw) |
7496 | ret = -EBUSY; | |
49516342 | 7497 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d PZ |
7498 | |
7499 | if (ret) | |
7500 | break; | |
332ac17e DF |
7501 | } |
7502 | ||
1724813d | 7503 | return ret; |
332ac17e DF |
7504 | } |
7505 | ||
1724813d | 7506 | static void sched_dl_do_global(void) |
ce0dbbbb | 7507 | { |
1724813d PZ |
7508 | u64 new_bw = -1; |
7509 | int cpu; | |
49516342 | 7510 | unsigned long flags; |
ce0dbbbb | 7511 | |
1724813d PZ |
7512 | def_dl_bandwidth.dl_period = global_rt_period(); |
7513 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
7514 | ||
7515 | if (global_rt_runtime() != RUNTIME_INF) | |
7516 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
7517 | ||
7518 | /* | |
7519 | * FIXME: As above... | |
7520 | */ | |
7521 | for_each_possible_cpu(cpu) { | |
7522 | struct dl_bw *dl_b = dl_bw_of(cpu); | |
7523 | ||
49516342 | 7524 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 7525 | dl_b->bw = new_bw; |
49516342 | 7526 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
ce0dbbbb | 7527 | } |
1724813d PZ |
7528 | } |
7529 | ||
7530 | static int sched_rt_global_validate(void) | |
7531 | { | |
7532 | if (sysctl_sched_rt_period <= 0) | |
7533 | return -EINVAL; | |
7534 | ||
e9e7cb38 JL |
7535 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
7536 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
7537 | return -EINVAL; |
7538 | ||
7539 | return 0; | |
7540 | } | |
7541 | ||
7542 | static void sched_rt_do_global(void) | |
7543 | { | |
7544 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
7545 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
7546 | } |
7547 | ||
d0b27fa7 | 7548 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 7549 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
7550 | loff_t *ppos) |
7551 | { | |
d0b27fa7 PZ |
7552 | int old_period, old_runtime; |
7553 | static DEFINE_MUTEX(mutex); | |
1724813d | 7554 | int ret; |
d0b27fa7 PZ |
7555 | |
7556 | mutex_lock(&mutex); | |
7557 | old_period = sysctl_sched_rt_period; | |
7558 | old_runtime = sysctl_sched_rt_runtime; | |
7559 | ||
8d65af78 | 7560 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
7561 | |
7562 | if (!ret && write) { | |
1724813d PZ |
7563 | ret = sched_rt_global_validate(); |
7564 | if (ret) | |
7565 | goto undo; | |
7566 | ||
d0b27fa7 | 7567 | ret = sched_rt_global_constraints(); |
1724813d PZ |
7568 | if (ret) |
7569 | goto undo; | |
7570 | ||
7571 | ret = sched_dl_global_constraints(); | |
7572 | if (ret) | |
7573 | goto undo; | |
7574 | ||
7575 | sched_rt_do_global(); | |
7576 | sched_dl_do_global(); | |
7577 | } | |
7578 | if (0) { | |
7579 | undo: | |
7580 | sysctl_sched_rt_period = old_period; | |
7581 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
7582 | } |
7583 | mutex_unlock(&mutex); | |
7584 | ||
7585 | return ret; | |
7586 | } | |
68318b8e | 7587 | |
1724813d | 7588 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
7589 | void __user *buffer, size_t *lenp, |
7590 | loff_t *ppos) | |
7591 | { | |
7592 | int ret; | |
332ac17e | 7593 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
7594 | |
7595 | mutex_lock(&mutex); | |
332ac17e | 7596 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
1724813d PZ |
7597 | /* make sure that internally we keep jiffies */ |
7598 | /* also, writing zero resets timeslice to default */ | |
332ac17e | 7599 | if (!ret && write) { |
1724813d PZ |
7600 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
7601 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); | |
332ac17e DF |
7602 | } |
7603 | mutex_unlock(&mutex); | |
332ac17e DF |
7604 | return ret; |
7605 | } | |
7606 | ||
052f1dc7 | 7607 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 7608 | |
a7c6d554 | 7609 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7610 | { |
a7c6d554 | 7611 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7612 | } |
7613 | ||
eb95419b TH |
7614 | static struct cgroup_subsys_state * |
7615 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7616 | { |
eb95419b TH |
7617 | struct task_group *parent = css_tg(parent_css); |
7618 | struct task_group *tg; | |
68318b8e | 7619 | |
eb95419b | 7620 | if (!parent) { |
68318b8e | 7621 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7622 | return &root_task_group.css; |
68318b8e SV |
7623 | } |
7624 | ||
ec7dc8ac | 7625 | tg = sched_create_group(parent); |
68318b8e SV |
7626 | if (IS_ERR(tg)) |
7627 | return ERR_PTR(-ENOMEM); | |
7628 | ||
68318b8e SV |
7629 | return &tg->css; |
7630 | } | |
7631 | ||
eb95419b | 7632 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
ace783b9 | 7633 | { |
eb95419b TH |
7634 | struct task_group *tg = css_tg(css); |
7635 | struct task_group *parent = css_tg(css_parent(css)); | |
ace783b9 | 7636 | |
63876986 TH |
7637 | if (parent) |
7638 | sched_online_group(tg, parent); | |
ace783b9 LZ |
7639 | return 0; |
7640 | } | |
7641 | ||
eb95419b | 7642 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7643 | { |
eb95419b | 7644 | struct task_group *tg = css_tg(css); |
68318b8e SV |
7645 | |
7646 | sched_destroy_group(tg); | |
7647 | } | |
7648 | ||
eb95419b | 7649 | static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css) |
ace783b9 | 7650 | { |
eb95419b | 7651 | struct task_group *tg = css_tg(css); |
ace783b9 LZ |
7652 | |
7653 | sched_offline_group(tg); | |
7654 | } | |
7655 | ||
eb95419b | 7656 | static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7657 | struct cgroup_taskset *tset) |
68318b8e | 7658 | { |
bb9d97b6 TH |
7659 | struct task_struct *task; |
7660 | ||
924f0d9a | 7661 | cgroup_taskset_for_each(task, tset) { |
b68aa230 | 7662 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7663 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7664 | return -EINVAL; |
b68aa230 | 7665 | #else |
bb9d97b6 TH |
7666 | /* We don't support RT-tasks being in separate groups */ |
7667 | if (task->sched_class != &fair_sched_class) | |
7668 | return -EINVAL; | |
b68aa230 | 7669 | #endif |
bb9d97b6 | 7670 | } |
be367d09 BB |
7671 | return 0; |
7672 | } | |
68318b8e | 7673 | |
eb95419b | 7674 | static void cpu_cgroup_attach(struct cgroup_subsys_state *css, |
bb9d97b6 | 7675 | struct cgroup_taskset *tset) |
68318b8e | 7676 | { |
bb9d97b6 TH |
7677 | struct task_struct *task; |
7678 | ||
924f0d9a | 7679 | cgroup_taskset_for_each(task, tset) |
bb9d97b6 | 7680 | sched_move_task(task); |
68318b8e SV |
7681 | } |
7682 | ||
eb95419b TH |
7683 | static void cpu_cgroup_exit(struct cgroup_subsys_state *css, |
7684 | struct cgroup_subsys_state *old_css, | |
7685 | struct task_struct *task) | |
068c5cc5 PZ |
7686 | { |
7687 | /* | |
7688 | * cgroup_exit() is called in the copy_process() failure path. | |
7689 | * Ignore this case since the task hasn't ran yet, this avoids | |
7690 | * trying to poke a half freed task state from generic code. | |
7691 | */ | |
7692 | if (!(task->flags & PF_EXITING)) | |
7693 | return; | |
7694 | ||
7695 | sched_move_task(task); | |
7696 | } | |
7697 | ||
052f1dc7 | 7698 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7699 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7700 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7701 | { |
182446d0 | 7702 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7703 | } |
7704 | ||
182446d0 TH |
7705 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7706 | struct cftype *cft) | |
68318b8e | 7707 | { |
182446d0 | 7708 | struct task_group *tg = css_tg(css); |
68318b8e | 7709 | |
c8b28116 | 7710 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7711 | } |
ab84d31e PT |
7712 | |
7713 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7714 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7715 | ||
ab84d31e PT |
7716 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
7717 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
7718 | ||
a790de99 PT |
7719 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7720 | ||
ab84d31e PT |
7721 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7722 | { | |
56f570e5 | 7723 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7724 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7725 | |
7726 | if (tg == &root_task_group) | |
7727 | return -EINVAL; | |
7728 | ||
7729 | /* | |
7730 | * Ensure we have at some amount of bandwidth every period. This is | |
7731 | * to prevent reaching a state of large arrears when throttled via | |
7732 | * entity_tick() resulting in prolonged exit starvation. | |
7733 | */ | |
7734 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7735 | return -EINVAL; | |
7736 | ||
7737 | /* | |
7738 | * Likewise, bound things on the otherside by preventing insane quota | |
7739 | * periods. This also allows us to normalize in computing quota | |
7740 | * feasibility. | |
7741 | */ | |
7742 | if (period > max_cfs_quota_period) | |
7743 | return -EINVAL; | |
7744 | ||
a790de99 PT |
7745 | mutex_lock(&cfs_constraints_mutex); |
7746 | ret = __cfs_schedulable(tg, period, quota); | |
7747 | if (ret) | |
7748 | goto out_unlock; | |
7749 | ||
58088ad0 | 7750 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7751 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7752 | /* |
7753 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7754 | * before making related changes, and on->off must occur afterwards | |
7755 | */ | |
7756 | if (runtime_enabled && !runtime_was_enabled) | |
7757 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7758 | raw_spin_lock_irq(&cfs_b->lock); |
7759 | cfs_b->period = ns_to_ktime(period); | |
7760 | cfs_b->quota = quota; | |
58088ad0 | 7761 | |
a9cf55b2 | 7762 | __refill_cfs_bandwidth_runtime(cfs_b); |
58088ad0 PT |
7763 | /* restart the period timer (if active) to handle new period expiry */ |
7764 | if (runtime_enabled && cfs_b->timer_active) { | |
7765 | /* force a reprogram */ | |
7766 | cfs_b->timer_active = 0; | |
7767 | __start_cfs_bandwidth(cfs_b); | |
7768 | } | |
ab84d31e PT |
7769 | raw_spin_unlock_irq(&cfs_b->lock); |
7770 | ||
7771 | for_each_possible_cpu(i) { | |
7772 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; | |
029632fb | 7773 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7774 | |
7775 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7776 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7777 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7778 | |
029632fb | 7779 | if (cfs_rq->throttled) |
671fd9da | 7780 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7781 | raw_spin_unlock_irq(&rq->lock); |
7782 | } | |
1ee14e6c BS |
7783 | if (runtime_was_enabled && !runtime_enabled) |
7784 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7785 | out_unlock: |
7786 | mutex_unlock(&cfs_constraints_mutex); | |
ab84d31e | 7787 | |
a790de99 | 7788 | return ret; |
ab84d31e PT |
7789 | } |
7790 | ||
7791 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7792 | { | |
7793 | u64 quota, period; | |
7794 | ||
029632fb | 7795 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7796 | if (cfs_quota_us < 0) |
7797 | quota = RUNTIME_INF; | |
7798 | else | |
7799 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7800 | ||
7801 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7802 | } | |
7803 | ||
7804 | long tg_get_cfs_quota(struct task_group *tg) | |
7805 | { | |
7806 | u64 quota_us; | |
7807 | ||
029632fb | 7808 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7809 | return -1; |
7810 | ||
029632fb | 7811 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7812 | do_div(quota_us, NSEC_PER_USEC); |
7813 | ||
7814 | return quota_us; | |
7815 | } | |
7816 | ||
7817 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7818 | { | |
7819 | u64 quota, period; | |
7820 | ||
7821 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7822 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7823 | |
ab84d31e PT |
7824 | return tg_set_cfs_bandwidth(tg, period, quota); |
7825 | } | |
7826 | ||
7827 | long tg_get_cfs_period(struct task_group *tg) | |
7828 | { | |
7829 | u64 cfs_period_us; | |
7830 | ||
029632fb | 7831 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7832 | do_div(cfs_period_us, NSEC_PER_USEC); |
7833 | ||
7834 | return cfs_period_us; | |
7835 | } | |
7836 | ||
182446d0 TH |
7837 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7838 | struct cftype *cft) | |
ab84d31e | 7839 | { |
182446d0 | 7840 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7841 | } |
7842 | ||
182446d0 TH |
7843 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7844 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7845 | { |
182446d0 | 7846 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7847 | } |
7848 | ||
182446d0 TH |
7849 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7850 | struct cftype *cft) | |
ab84d31e | 7851 | { |
182446d0 | 7852 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7853 | } |
7854 | ||
182446d0 TH |
7855 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7856 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7857 | { |
182446d0 | 7858 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7859 | } |
7860 | ||
a790de99 PT |
7861 | struct cfs_schedulable_data { |
7862 | struct task_group *tg; | |
7863 | u64 period, quota; | |
7864 | }; | |
7865 | ||
7866 | /* | |
7867 | * normalize group quota/period to be quota/max_period | |
7868 | * note: units are usecs | |
7869 | */ | |
7870 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7871 | struct cfs_schedulable_data *d) | |
7872 | { | |
7873 | u64 quota, period; | |
7874 | ||
7875 | if (tg == d->tg) { | |
7876 | period = d->period; | |
7877 | quota = d->quota; | |
7878 | } else { | |
7879 | period = tg_get_cfs_period(tg); | |
7880 | quota = tg_get_cfs_quota(tg); | |
7881 | } | |
7882 | ||
7883 | /* note: these should typically be equivalent */ | |
7884 | if (quota == RUNTIME_INF || quota == -1) | |
7885 | return RUNTIME_INF; | |
7886 | ||
7887 | return to_ratio(period, quota); | |
7888 | } | |
7889 | ||
7890 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7891 | { | |
7892 | struct cfs_schedulable_data *d = data; | |
029632fb | 7893 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7894 | s64 quota = 0, parent_quota = -1; |
7895 | ||
7896 | if (!tg->parent) { | |
7897 | quota = RUNTIME_INF; | |
7898 | } else { | |
029632fb | 7899 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7900 | |
7901 | quota = normalize_cfs_quota(tg, d); | |
7902 | parent_quota = parent_b->hierarchal_quota; | |
7903 | ||
7904 | /* | |
7905 | * ensure max(child_quota) <= parent_quota, inherit when no | |
7906 | * limit is set | |
7907 | */ | |
7908 | if (quota == RUNTIME_INF) | |
7909 | quota = parent_quota; | |
7910 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7911 | return -EINVAL; | |
7912 | } | |
7913 | cfs_b->hierarchal_quota = quota; | |
7914 | ||
7915 | return 0; | |
7916 | } | |
7917 | ||
7918 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7919 | { | |
8277434e | 7920 | int ret; |
a790de99 PT |
7921 | struct cfs_schedulable_data data = { |
7922 | .tg = tg, | |
7923 | .period = period, | |
7924 | .quota = quota, | |
7925 | }; | |
7926 | ||
7927 | if (quota != RUNTIME_INF) { | |
7928 | do_div(data.period, NSEC_PER_USEC); | |
7929 | do_div(data.quota, NSEC_PER_USEC); | |
7930 | } | |
7931 | ||
8277434e PT |
7932 | rcu_read_lock(); |
7933 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7934 | rcu_read_unlock(); | |
7935 | ||
7936 | return ret; | |
a790de99 | 7937 | } |
e8da1b18 | 7938 | |
2da8ca82 | 7939 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 7940 | { |
2da8ca82 | 7941 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7942 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7943 | |
44ffc75b TH |
7944 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7945 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7946 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
7947 | |
7948 | return 0; | |
7949 | } | |
ab84d31e | 7950 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7951 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7952 | |
052f1dc7 | 7953 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7954 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7955 | struct cftype *cft, s64 val) | |
6f505b16 | 7956 | { |
182446d0 | 7957 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7958 | } |
7959 | ||
182446d0 TH |
7960 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7961 | struct cftype *cft) | |
6f505b16 | 7962 | { |
182446d0 | 7963 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7964 | } |
d0b27fa7 | 7965 | |
182446d0 TH |
7966 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7967 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7968 | { |
182446d0 | 7969 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7970 | } |
7971 | ||
182446d0 TH |
7972 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7973 | struct cftype *cft) | |
d0b27fa7 | 7974 | { |
182446d0 | 7975 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7976 | } |
6d6bc0ad | 7977 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7978 | |
fe5c7cc2 | 7979 | static struct cftype cpu_files[] = { |
052f1dc7 | 7980 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7981 | { |
7982 | .name = "shares", | |
f4c753b7 PM |
7983 | .read_u64 = cpu_shares_read_u64, |
7984 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7985 | }, |
052f1dc7 | 7986 | #endif |
ab84d31e PT |
7987 | #ifdef CONFIG_CFS_BANDWIDTH |
7988 | { | |
7989 | .name = "cfs_quota_us", | |
7990 | .read_s64 = cpu_cfs_quota_read_s64, | |
7991 | .write_s64 = cpu_cfs_quota_write_s64, | |
7992 | }, | |
7993 | { | |
7994 | .name = "cfs_period_us", | |
7995 | .read_u64 = cpu_cfs_period_read_u64, | |
7996 | .write_u64 = cpu_cfs_period_write_u64, | |
7997 | }, | |
e8da1b18 NR |
7998 | { |
7999 | .name = "stat", | |
2da8ca82 | 8000 | .seq_show = cpu_stats_show, |
e8da1b18 | 8001 | }, |
ab84d31e | 8002 | #endif |
052f1dc7 | 8003 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8004 | { |
9f0c1e56 | 8005 | .name = "rt_runtime_us", |
06ecb27c PM |
8006 | .read_s64 = cpu_rt_runtime_read, |
8007 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8008 | }, |
d0b27fa7 PZ |
8009 | { |
8010 | .name = "rt_period_us", | |
f4c753b7 PM |
8011 | .read_u64 = cpu_rt_period_read_uint, |
8012 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8013 | }, |
052f1dc7 | 8014 | #endif |
4baf6e33 | 8015 | { } /* terminate */ |
68318b8e SV |
8016 | }; |
8017 | ||
073219e9 | 8018 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 TH |
8019 | .css_alloc = cpu_cgroup_css_alloc, |
8020 | .css_free = cpu_cgroup_css_free, | |
ace783b9 LZ |
8021 | .css_online = cpu_cgroup_css_online, |
8022 | .css_offline = cpu_cgroup_css_offline, | |
bb9d97b6 TH |
8023 | .can_attach = cpu_cgroup_can_attach, |
8024 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8025 | .exit = cpu_cgroup_exit, |
4baf6e33 | 8026 | .base_cftypes = cpu_files, |
68318b8e SV |
8027 | .early_init = 1, |
8028 | }; | |
8029 | ||
052f1dc7 | 8030 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8031 | |
b637a328 PM |
8032 | void dump_cpu_task(int cpu) |
8033 | { | |
8034 | pr_info("Task dump for CPU %d:\n", cpu); | |
8035 | sched_show_task(cpu_curr(cpu)); | |
8036 | } |