]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
391e43da | 2 | * kernel/sched/core.c |
1da177e4 | 3 | * |
d1ccc66d | 4 | * Core kernel scheduler code and related syscalls |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
1da177e4 | 7 | */ |
004172bd | 8 | #include <linux/sched.h> |
1da177e4 | 9 | #include <linux/cpuset.h> |
0ff92245 | 10 | #include <linux/delayacct.h> |
f1c6f1a7 | 11 | #include <linux/init_task.h> |
91d1aa43 | 12 | #include <linux/context_tracking.h> |
f9411ebe | 13 | #include <linux/rcupdate_wait.h> |
004172bd IM |
14 | |
15 | #include <linux/blkdev.h> | |
16 | #include <linux/kprobes.h> | |
17 | #include <linux/mmu_context.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/nmi.h> | |
6075620b | 20 | #include <linux/prefetch.h> |
004172bd IM |
21 | #include <linux/profile.h> |
22 | #include <linux/security.h> | |
23 | #include <linux/syscalls.h> | |
1da177e4 | 24 | |
96f951ed | 25 | #include <asm/switch_to.h> |
5517d86b | 26 | #include <asm/tlb.h> |
fd4a61e0 MB |
27 | #ifdef CONFIG_PARAVIRT |
28 | #include <asm/paravirt.h> | |
29 | #endif | |
1da177e4 | 30 | |
029632fb | 31 | #include "sched.h" |
ea138446 | 32 | #include "../workqueue_internal.h" |
29d5e047 | 33 | #include "../smpboot.h" |
6e0534f2 | 34 | |
a8d154b0 | 35 | #define CREATE_TRACE_POINTS |
ad8d75ff | 36 | #include <trace/events/sched.h> |
a8d154b0 | 37 | |
029632fb | 38 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 39 | |
bf5c91ba IM |
40 | /* |
41 | * Debugging: various feature bits | |
42 | */ | |
f00b45c1 | 43 | |
f00b45c1 PZ |
44 | #define SCHED_FEAT(name, enabled) \ |
45 | (1UL << __SCHED_FEAT_##name) * enabled | | |
46 | ||
bf5c91ba | 47 | const_debug unsigned int sysctl_sched_features = |
391e43da | 48 | #include "features.h" |
f00b45c1 PZ |
49 | 0; |
50 | ||
51 | #undef SCHED_FEAT | |
52 | ||
b82d9fdd PZ |
53 | /* |
54 | * Number of tasks to iterate in a single balance run. | |
55 | * Limited because this is done with IRQs disabled. | |
56 | */ | |
57 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
58 | ||
e9e9250b PZ |
59 | /* |
60 | * period over which we average the RT time consumption, measured | |
61 | * in ms. | |
62 | * | |
63 | * default: 1s | |
64 | */ | |
65 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
66 | ||
fa85ae24 | 67 | /* |
d1ccc66d | 68 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
69 | * default: 1s |
70 | */ | |
9f0c1e56 | 71 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 72 | |
029632fb | 73 | __read_mostly int scheduler_running; |
6892b75e | 74 | |
9f0c1e56 PZ |
75 | /* |
76 | * part of the period that we allow rt tasks to run in us. | |
77 | * default: 0.95s | |
78 | */ | |
79 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 80 | |
d1ccc66d | 81 | /* CPUs with isolated domains */ |
3fa0818b RR |
82 | cpumask_var_t cpu_isolated_map; |
83 | ||
1da177e4 | 84 | /* |
cc2a73b5 | 85 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 86 | */ |
a9957449 | 87 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
88 | __acquires(rq->lock) |
89 | { | |
70b97a7f | 90 | struct rq *rq; |
1da177e4 LT |
91 | |
92 | local_irq_disable(); | |
93 | rq = this_rq(); | |
05fa785c | 94 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
95 | |
96 | return rq; | |
97 | } | |
98 | ||
3e71a462 PZ |
99 | /* |
100 | * __task_rq_lock - lock the rq @p resides on. | |
101 | */ | |
eb580751 | 102 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
103 | __acquires(rq->lock) |
104 | { | |
105 | struct rq *rq; | |
106 | ||
107 | lockdep_assert_held(&p->pi_lock); | |
108 | ||
109 | for (;;) { | |
110 | rq = task_rq(p); | |
111 | raw_spin_lock(&rq->lock); | |
112 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 113 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
114 | return rq; |
115 | } | |
116 | raw_spin_unlock(&rq->lock); | |
117 | ||
118 | while (unlikely(task_on_rq_migrating(p))) | |
119 | cpu_relax(); | |
120 | } | |
121 | } | |
122 | ||
123 | /* | |
124 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
125 | */ | |
eb580751 | 126 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
127 | __acquires(p->pi_lock) |
128 | __acquires(rq->lock) | |
129 | { | |
130 | struct rq *rq; | |
131 | ||
132 | for (;;) { | |
eb580751 | 133 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
134 | rq = task_rq(p); |
135 | raw_spin_lock(&rq->lock); | |
136 | /* | |
137 | * move_queued_task() task_rq_lock() | |
138 | * | |
139 | * ACQUIRE (rq->lock) | |
140 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
141 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
142 | * [S] ->cpu = new_cpu [L] task_rq() | |
143 | * [L] ->on_rq | |
144 | * RELEASE (rq->lock) | |
145 | * | |
146 | * If we observe the old cpu in task_rq_lock, the acquire of | |
147 | * the old rq->lock will fully serialize against the stores. | |
148 | * | |
d1ccc66d | 149 | * If we observe the new CPU in task_rq_lock, the acquire will |
3e71a462 PZ |
150 | * pair with the WMB to ensure we must then also see migrating. |
151 | */ | |
152 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 153 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
154 | return rq; |
155 | } | |
156 | raw_spin_unlock(&rq->lock); | |
eb580751 | 157 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
158 | |
159 | while (unlikely(task_on_rq_migrating(p))) | |
160 | cpu_relax(); | |
161 | } | |
162 | } | |
163 | ||
535b9552 IM |
164 | /* |
165 | * RQ-clock updating methods: | |
166 | */ | |
167 | ||
168 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
169 | { | |
170 | /* | |
171 | * In theory, the compile should just see 0 here, and optimize out the call | |
172 | * to sched_rt_avg_update. But I don't trust it... | |
173 | */ | |
174 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
175 | s64 steal = 0, irq_delta = 0; | |
176 | #endif | |
177 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
178 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
179 | ||
180 | /* | |
181 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
182 | * this case when a previous update_rq_clock() happened inside a | |
183 | * {soft,}irq region. | |
184 | * | |
185 | * When this happens, we stop ->clock_task and only update the | |
186 | * prev_irq_time stamp to account for the part that fit, so that a next | |
187 | * update will consume the rest. This ensures ->clock_task is | |
188 | * monotonic. | |
189 | * | |
190 | * It does however cause some slight miss-attribution of {soft,}irq | |
191 | * time, a more accurate solution would be to update the irq_time using | |
192 | * the current rq->clock timestamp, except that would require using | |
193 | * atomic ops. | |
194 | */ | |
195 | if (irq_delta > delta) | |
196 | irq_delta = delta; | |
197 | ||
198 | rq->prev_irq_time += irq_delta; | |
199 | delta -= irq_delta; | |
200 | #endif | |
201 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
202 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
203 | steal = paravirt_steal_clock(cpu_of(rq)); | |
204 | steal -= rq->prev_steal_time_rq; | |
205 | ||
206 | if (unlikely(steal > delta)) | |
207 | steal = delta; | |
208 | ||
209 | rq->prev_steal_time_rq += steal; | |
210 | delta -= steal; | |
211 | } | |
212 | #endif | |
213 | ||
214 | rq->clock_task += delta; | |
215 | ||
216 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) | |
217 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) | |
218 | sched_rt_avg_update(rq, irq_delta + steal); | |
219 | #endif | |
220 | } | |
221 | ||
222 | void update_rq_clock(struct rq *rq) | |
223 | { | |
224 | s64 delta; | |
225 | ||
226 | lockdep_assert_held(&rq->lock); | |
227 | ||
228 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
229 | return; | |
230 | ||
231 | #ifdef CONFIG_SCHED_DEBUG | |
232 | rq->clock_update_flags |= RQCF_UPDATED; | |
233 | #endif | |
234 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; | |
235 | if (delta < 0) | |
236 | return; | |
237 | rq->clock += delta; | |
238 | update_rq_clock_task(rq, delta); | |
239 | } | |
240 | ||
241 | ||
8f4d37ec PZ |
242 | #ifdef CONFIG_SCHED_HRTICK |
243 | /* | |
244 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 245 | */ |
8f4d37ec | 246 | |
8f4d37ec PZ |
247 | static void hrtick_clear(struct rq *rq) |
248 | { | |
249 | if (hrtimer_active(&rq->hrtick_timer)) | |
250 | hrtimer_cancel(&rq->hrtick_timer); | |
251 | } | |
252 | ||
8f4d37ec PZ |
253 | /* |
254 | * High-resolution timer tick. | |
255 | * Runs from hardirq context with interrupts disabled. | |
256 | */ | |
257 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
258 | { | |
259 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
260 | ||
261 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
262 | ||
05fa785c | 263 | raw_spin_lock(&rq->lock); |
3e51f33f | 264 | update_rq_clock(rq); |
8f4d37ec | 265 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 266 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
267 | |
268 | return HRTIMER_NORESTART; | |
269 | } | |
270 | ||
95e904c7 | 271 | #ifdef CONFIG_SMP |
971ee28c | 272 | |
4961b6e1 | 273 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
274 | { |
275 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 276 | |
4961b6e1 | 277 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
278 | } |
279 | ||
31656519 PZ |
280 | /* |
281 | * called from hardirq (IPI) context | |
282 | */ | |
283 | static void __hrtick_start(void *arg) | |
b328ca18 | 284 | { |
31656519 | 285 | struct rq *rq = arg; |
b328ca18 | 286 | |
05fa785c | 287 | raw_spin_lock(&rq->lock); |
971ee28c | 288 | __hrtick_restart(rq); |
31656519 | 289 | rq->hrtick_csd_pending = 0; |
05fa785c | 290 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
291 | } |
292 | ||
31656519 PZ |
293 | /* |
294 | * Called to set the hrtick timer state. | |
295 | * | |
296 | * called with rq->lock held and irqs disabled | |
297 | */ | |
029632fb | 298 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 299 | { |
31656519 | 300 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 301 | ktime_t time; |
302 | s64 delta; | |
303 | ||
304 | /* | |
305 | * Don't schedule slices shorter than 10000ns, that just | |
306 | * doesn't make sense and can cause timer DoS. | |
307 | */ | |
308 | delta = max_t(s64, delay, 10000LL); | |
309 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 310 | |
cc584b21 | 311 | hrtimer_set_expires(timer, time); |
31656519 PZ |
312 | |
313 | if (rq == this_rq()) { | |
971ee28c | 314 | __hrtick_restart(rq); |
31656519 | 315 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 316 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
317 | rq->hrtick_csd_pending = 1; |
318 | } | |
b328ca18 PZ |
319 | } |
320 | ||
31656519 PZ |
321 | #else |
322 | /* | |
323 | * Called to set the hrtick timer state. | |
324 | * | |
325 | * called with rq->lock held and irqs disabled | |
326 | */ | |
029632fb | 327 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 328 | { |
86893335 WL |
329 | /* |
330 | * Don't schedule slices shorter than 10000ns, that just | |
331 | * doesn't make sense. Rely on vruntime for fairness. | |
332 | */ | |
333 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
334 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
335 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 336 | } |
31656519 | 337 | #endif /* CONFIG_SMP */ |
8f4d37ec | 338 | |
31656519 | 339 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 340 | { |
31656519 PZ |
341 | #ifdef CONFIG_SMP |
342 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 343 | |
31656519 PZ |
344 | rq->hrtick_csd.flags = 0; |
345 | rq->hrtick_csd.func = __hrtick_start; | |
346 | rq->hrtick_csd.info = rq; | |
347 | #endif | |
8f4d37ec | 348 | |
31656519 PZ |
349 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
350 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 351 | } |
006c75f1 | 352 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
353 | static inline void hrtick_clear(struct rq *rq) |
354 | { | |
355 | } | |
356 | ||
8f4d37ec PZ |
357 | static inline void init_rq_hrtick(struct rq *rq) |
358 | { | |
359 | } | |
006c75f1 | 360 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 361 | |
5529578a FW |
362 | /* |
363 | * cmpxchg based fetch_or, macro so it works for different integer types | |
364 | */ | |
365 | #define fetch_or(ptr, mask) \ | |
366 | ({ \ | |
367 | typeof(ptr) _ptr = (ptr); \ | |
368 | typeof(mask) _mask = (mask); \ | |
369 | typeof(*_ptr) _old, _val = *_ptr; \ | |
370 | \ | |
371 | for (;;) { \ | |
372 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
373 | if (_old == _val) \ | |
374 | break; \ | |
375 | _val = _old; \ | |
376 | } \ | |
377 | _old; \ | |
378 | }) | |
379 | ||
e3baac47 | 380 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
381 | /* |
382 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
383 | * this avoids any races wrt polling state changes and thereby avoids | |
384 | * spurious IPIs. | |
385 | */ | |
386 | static bool set_nr_and_not_polling(struct task_struct *p) | |
387 | { | |
388 | struct thread_info *ti = task_thread_info(p); | |
389 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
390 | } | |
e3baac47 PZ |
391 | |
392 | /* | |
393 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
394 | * | |
395 | * If this returns true, then the idle task promises to call | |
396 | * sched_ttwu_pending() and reschedule soon. | |
397 | */ | |
398 | static bool set_nr_if_polling(struct task_struct *p) | |
399 | { | |
400 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 401 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
402 | |
403 | for (;;) { | |
404 | if (!(val & _TIF_POLLING_NRFLAG)) | |
405 | return false; | |
406 | if (val & _TIF_NEED_RESCHED) | |
407 | return true; | |
408 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
409 | if (old == val) | |
410 | break; | |
411 | val = old; | |
412 | } | |
413 | return true; | |
414 | } | |
415 | ||
fd99f91a PZ |
416 | #else |
417 | static bool set_nr_and_not_polling(struct task_struct *p) | |
418 | { | |
419 | set_tsk_need_resched(p); | |
420 | return true; | |
421 | } | |
e3baac47 PZ |
422 | |
423 | #ifdef CONFIG_SMP | |
424 | static bool set_nr_if_polling(struct task_struct *p) | |
425 | { | |
426 | return false; | |
427 | } | |
428 | #endif | |
fd99f91a PZ |
429 | #endif |
430 | ||
76751049 PZ |
431 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
432 | { | |
433 | struct wake_q_node *node = &task->wake_q; | |
434 | ||
435 | /* | |
436 | * Atomically grab the task, if ->wake_q is !nil already it means | |
437 | * its already queued (either by us or someone else) and will get the | |
438 | * wakeup due to that. | |
439 | * | |
440 | * This cmpxchg() implies a full barrier, which pairs with the write | |
58fe9c46 | 441 | * barrier implied by the wakeup in wake_up_q(). |
76751049 PZ |
442 | */ |
443 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
444 | return; | |
445 | ||
446 | get_task_struct(task); | |
447 | ||
448 | /* | |
449 | * The head is context local, there can be no concurrency. | |
450 | */ | |
451 | *head->lastp = node; | |
452 | head->lastp = &node->next; | |
453 | } | |
454 | ||
455 | void wake_up_q(struct wake_q_head *head) | |
456 | { | |
457 | struct wake_q_node *node = head->first; | |
458 | ||
459 | while (node != WAKE_Q_TAIL) { | |
460 | struct task_struct *task; | |
461 | ||
462 | task = container_of(node, struct task_struct, wake_q); | |
463 | BUG_ON(!task); | |
d1ccc66d | 464 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
465 | node = node->next; |
466 | task->wake_q.next = NULL; | |
467 | ||
468 | /* | |
469 | * wake_up_process() implies a wmb() to pair with the queueing | |
470 | * in wake_q_add() so as not to miss wakeups. | |
471 | */ | |
472 | wake_up_process(task); | |
473 | put_task_struct(task); | |
474 | } | |
475 | } | |
476 | ||
c24d20db | 477 | /* |
8875125e | 478 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
479 | * |
480 | * On UP this means the setting of the need_resched flag, on SMP it | |
481 | * might also involve a cross-CPU call to trigger the scheduler on | |
482 | * the target CPU. | |
483 | */ | |
8875125e | 484 | void resched_curr(struct rq *rq) |
c24d20db | 485 | { |
8875125e | 486 | struct task_struct *curr = rq->curr; |
c24d20db IM |
487 | int cpu; |
488 | ||
8875125e | 489 | lockdep_assert_held(&rq->lock); |
c24d20db | 490 | |
8875125e | 491 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
492 | return; |
493 | ||
8875125e | 494 | cpu = cpu_of(rq); |
fd99f91a | 495 | |
f27dde8d | 496 | if (cpu == smp_processor_id()) { |
8875125e | 497 | set_tsk_need_resched(curr); |
f27dde8d | 498 | set_preempt_need_resched(); |
c24d20db | 499 | return; |
f27dde8d | 500 | } |
c24d20db | 501 | |
8875125e | 502 | if (set_nr_and_not_polling(curr)) |
c24d20db | 503 | smp_send_reschedule(cpu); |
dfc68f29 AL |
504 | else |
505 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
506 | } |
507 | ||
029632fb | 508 | void resched_cpu(int cpu) |
c24d20db IM |
509 | { |
510 | struct rq *rq = cpu_rq(cpu); | |
511 | unsigned long flags; | |
512 | ||
05fa785c | 513 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db | 514 | return; |
8875125e | 515 | resched_curr(rq); |
05fa785c | 516 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 517 | } |
06d8308c | 518 | |
b021fe3e | 519 | #ifdef CONFIG_SMP |
3451d024 | 520 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 521 | /* |
d1ccc66d IM |
522 | * In the semi idle case, use the nearest busy CPU for migrating timers |
523 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
524 | * |
525 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
526 | * selecting an idle CPU will add more delays to the timers than intended |
527 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 528 | */ |
bc7a34b8 | 529 | int get_nohz_timer_target(void) |
83cd4fe2 | 530 | { |
bc7a34b8 | 531 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
532 | struct sched_domain *sd; |
533 | ||
9642d18e | 534 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
6201b4d6 VK |
535 | return cpu; |
536 | ||
057f3fad | 537 | rcu_read_lock(); |
83cd4fe2 | 538 | for_each_domain(cpu, sd) { |
057f3fad | 539 | for_each_cpu(i, sched_domain_span(sd)) { |
44496922 WL |
540 | if (cpu == i) |
541 | continue; | |
542 | ||
543 | if (!idle_cpu(i) && is_housekeeping_cpu(i)) { | |
057f3fad PZ |
544 | cpu = i; |
545 | goto unlock; | |
546 | } | |
547 | } | |
83cd4fe2 | 548 | } |
9642d18e VH |
549 | |
550 | if (!is_housekeeping_cpu(cpu)) | |
551 | cpu = housekeeping_any_cpu(); | |
057f3fad PZ |
552 | unlock: |
553 | rcu_read_unlock(); | |
83cd4fe2 VP |
554 | return cpu; |
555 | } | |
d1ccc66d | 556 | |
06d8308c TG |
557 | /* |
558 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
559 | * idle CPU then this timer might expire before the next timer event | |
560 | * which is scheduled to wake up that CPU. In case of a completely | |
561 | * idle system the next event might even be infinite time into the | |
562 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
563 | * leaves the inner idle loop so the newly added timer is taken into | |
564 | * account when the CPU goes back to idle and evaluates the timer | |
565 | * wheel for the next timer event. | |
566 | */ | |
1c20091e | 567 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
568 | { |
569 | struct rq *rq = cpu_rq(cpu); | |
570 | ||
571 | if (cpu == smp_processor_id()) | |
572 | return; | |
573 | ||
67b9ca70 | 574 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 575 | smp_send_reschedule(cpu); |
dfc68f29 AL |
576 | else |
577 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
578 | } |
579 | ||
c5bfece2 | 580 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 581 | { |
53c5fa16 FW |
582 | /* |
583 | * We just need the target to call irq_exit() and re-evaluate | |
584 | * the next tick. The nohz full kick at least implies that. | |
585 | * If needed we can still optimize that later with an | |
586 | * empty IRQ. | |
587 | */ | |
379d9ecb PM |
588 | if (cpu_is_offline(cpu)) |
589 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 590 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
591 | if (cpu != smp_processor_id() || |
592 | tick_nohz_tick_stopped()) | |
53c5fa16 | 593 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
594 | return true; |
595 | } | |
596 | ||
597 | return false; | |
598 | } | |
599 | ||
379d9ecb PM |
600 | /* |
601 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
602 | * caller's responsibility to deal with the lost wakeup, for example, | |
603 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
604 | */ | |
1c20091e FW |
605 | void wake_up_nohz_cpu(int cpu) |
606 | { | |
c5bfece2 | 607 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
608 | wake_up_idle_cpu(cpu); |
609 | } | |
610 | ||
ca38062e | 611 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 612 | { |
1c792db7 | 613 | int cpu = smp_processor_id(); |
873b4c65 VG |
614 | |
615 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) | |
616 | return false; | |
617 | ||
618 | if (idle_cpu(cpu) && !need_resched()) | |
619 | return true; | |
620 | ||
621 | /* | |
622 | * We can't run Idle Load Balance on this CPU for this time so we | |
623 | * cancel it and clear NOHZ_BALANCE_KICK | |
624 | */ | |
625 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); | |
626 | return false; | |
45bf76df IM |
627 | } |
628 | ||
3451d024 | 629 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 630 | |
ca38062e | 631 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 632 | { |
ca38062e | 633 | return false; |
2069dd75 PZ |
634 | } |
635 | ||
3451d024 | 636 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 637 | |
ce831b38 | 638 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 639 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 640 | { |
76d92ac3 FW |
641 | int fifo_nr_running; |
642 | ||
643 | /* Deadline tasks, even if single, need the tick */ | |
644 | if (rq->dl.dl_nr_running) | |
645 | return false; | |
646 | ||
1e78cdbd | 647 | /* |
2548d546 PZ |
648 | * If there are more than one RR tasks, we need the tick to effect the |
649 | * actual RR behaviour. | |
1e78cdbd | 650 | */ |
76d92ac3 FW |
651 | if (rq->rt.rr_nr_running) { |
652 | if (rq->rt.rr_nr_running == 1) | |
653 | return true; | |
654 | else | |
655 | return false; | |
1e78cdbd RR |
656 | } |
657 | ||
2548d546 PZ |
658 | /* |
659 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
660 | * forced preemption between FIFO tasks. | |
661 | */ | |
662 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
663 | if (fifo_nr_running) | |
664 | return true; | |
665 | ||
666 | /* | |
667 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
668 | * if there's more than one we need the tick for involuntary | |
669 | * preemption. | |
670 | */ | |
671 | if (rq->nr_running > 1) | |
541b8264 | 672 | return false; |
ce831b38 | 673 | |
541b8264 | 674 | return true; |
ce831b38 FW |
675 | } |
676 | #endif /* CONFIG_NO_HZ_FULL */ | |
d842de87 | 677 | |
029632fb | 678 | void sched_avg_update(struct rq *rq) |
18d95a28 | 679 | { |
e9e9250b PZ |
680 | s64 period = sched_avg_period(); |
681 | ||
78becc27 | 682 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
0d98bb26 WD |
683 | /* |
684 | * Inline assembly required to prevent the compiler | |
685 | * optimising this loop into a divmod call. | |
686 | * See __iter_div_u64_rem() for another example of this. | |
687 | */ | |
688 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
689 | rq->age_stamp += period; |
690 | rq->rt_avg /= 2; | |
691 | } | |
18d95a28 PZ |
692 | } |
693 | ||
6d6bc0ad | 694 | #endif /* CONFIG_SMP */ |
18d95a28 | 695 | |
a790de99 PT |
696 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
697 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 698 | /* |
8277434e PT |
699 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
700 | * node and @up when leaving it for the final time. | |
701 | * | |
702 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 703 | */ |
029632fb | 704 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 705 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
706 | { |
707 | struct task_group *parent, *child; | |
eb755805 | 708 | int ret; |
c09595f6 | 709 | |
8277434e PT |
710 | parent = from; |
711 | ||
c09595f6 | 712 | down: |
eb755805 PZ |
713 | ret = (*down)(parent, data); |
714 | if (ret) | |
8277434e | 715 | goto out; |
c09595f6 PZ |
716 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
717 | parent = child; | |
718 | goto down; | |
719 | ||
720 | up: | |
721 | continue; | |
722 | } | |
eb755805 | 723 | ret = (*up)(parent, data); |
8277434e PT |
724 | if (ret || parent == from) |
725 | goto out; | |
c09595f6 PZ |
726 | |
727 | child = parent; | |
728 | parent = parent->parent; | |
729 | if (parent) | |
730 | goto up; | |
8277434e | 731 | out: |
eb755805 | 732 | return ret; |
c09595f6 PZ |
733 | } |
734 | ||
029632fb | 735 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 736 | { |
e2b245f8 | 737 | return 0; |
eb755805 | 738 | } |
18d95a28 PZ |
739 | #endif |
740 | ||
45bf76df IM |
741 | static void set_load_weight(struct task_struct *p) |
742 | { | |
f05998d4 NR |
743 | int prio = p->static_prio - MAX_RT_PRIO; |
744 | struct load_weight *load = &p->se.load; | |
745 | ||
dd41f596 IM |
746 | /* |
747 | * SCHED_IDLE tasks get minimal weight: | |
748 | */ | |
20f9cd2a | 749 | if (idle_policy(p->policy)) { |
c8b28116 | 750 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 751 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
752 | return; |
753 | } | |
71f8bd46 | 754 | |
ed82b8a1 AK |
755 | load->weight = scale_load(sched_prio_to_weight[prio]); |
756 | load->inv_weight = sched_prio_to_wmult[prio]; | |
71f8bd46 IM |
757 | } |
758 | ||
1de64443 | 759 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 760 | { |
a64692a3 | 761 | update_rq_clock(rq); |
1de64443 PZ |
762 | if (!(flags & ENQUEUE_RESTORE)) |
763 | sched_info_queued(rq, p); | |
371fd7e7 | 764 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
765 | } |
766 | ||
1de64443 | 767 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 768 | { |
a64692a3 | 769 | update_rq_clock(rq); |
1de64443 PZ |
770 | if (!(flags & DEQUEUE_SAVE)) |
771 | sched_info_dequeued(rq, p); | |
371fd7e7 | 772 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
773 | } |
774 | ||
029632fb | 775 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
776 | { |
777 | if (task_contributes_to_load(p)) | |
778 | rq->nr_uninterruptible--; | |
779 | ||
371fd7e7 | 780 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
781 | } |
782 | ||
029632fb | 783 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
784 | { |
785 | if (task_contributes_to_load(p)) | |
786 | rq->nr_uninterruptible++; | |
787 | ||
371fd7e7 | 788 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
789 | } |
790 | ||
34f971f6 PZ |
791 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
792 | { | |
793 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
794 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
795 | ||
796 | if (stop) { | |
797 | /* | |
798 | * Make it appear like a SCHED_FIFO task, its something | |
799 | * userspace knows about and won't get confused about. | |
800 | * | |
801 | * Also, it will make PI more or less work without too | |
802 | * much confusion -- but then, stop work should not | |
803 | * rely on PI working anyway. | |
804 | */ | |
805 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
806 | ||
807 | stop->sched_class = &stop_sched_class; | |
808 | } | |
809 | ||
810 | cpu_rq(cpu)->stop = stop; | |
811 | ||
812 | if (old_stop) { | |
813 | /* | |
814 | * Reset it back to a normal scheduling class so that | |
815 | * it can die in pieces. | |
816 | */ | |
817 | old_stop->sched_class = &rt_sched_class; | |
818 | } | |
819 | } | |
820 | ||
14531189 | 821 | /* |
dd41f596 | 822 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 823 | */ |
14531189 IM |
824 | static inline int __normal_prio(struct task_struct *p) |
825 | { | |
dd41f596 | 826 | return p->static_prio; |
14531189 IM |
827 | } |
828 | ||
b29739f9 IM |
829 | /* |
830 | * Calculate the expected normal priority: i.e. priority | |
831 | * without taking RT-inheritance into account. Might be | |
832 | * boosted by interactivity modifiers. Changes upon fork, | |
833 | * setprio syscalls, and whenever the interactivity | |
834 | * estimator recalculates. | |
835 | */ | |
36c8b586 | 836 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
837 | { |
838 | int prio; | |
839 | ||
aab03e05 DF |
840 | if (task_has_dl_policy(p)) |
841 | prio = MAX_DL_PRIO-1; | |
842 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
843 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
844 | else | |
845 | prio = __normal_prio(p); | |
846 | return prio; | |
847 | } | |
848 | ||
849 | /* | |
850 | * Calculate the current priority, i.e. the priority | |
851 | * taken into account by the scheduler. This value might | |
852 | * be boosted by RT tasks, or might be boosted by | |
853 | * interactivity modifiers. Will be RT if the task got | |
854 | * RT-boosted. If not then it returns p->normal_prio. | |
855 | */ | |
36c8b586 | 856 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
857 | { |
858 | p->normal_prio = normal_prio(p); | |
859 | /* | |
860 | * If we are RT tasks or we were boosted to RT priority, | |
861 | * keep the priority unchanged. Otherwise, update priority | |
862 | * to the normal priority: | |
863 | */ | |
864 | if (!rt_prio(p->prio)) | |
865 | return p->normal_prio; | |
866 | return p->prio; | |
867 | } | |
868 | ||
1da177e4 LT |
869 | /** |
870 | * task_curr - is this task currently executing on a CPU? | |
871 | * @p: the task in question. | |
e69f6186 YB |
872 | * |
873 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 874 | */ |
36c8b586 | 875 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
876 | { |
877 | return cpu_curr(task_cpu(p)) == p; | |
878 | } | |
879 | ||
67dfa1b7 | 880 | /* |
4c9a4bc8 PZ |
881 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
882 | * use the balance_callback list if you want balancing. | |
883 | * | |
884 | * this means any call to check_class_changed() must be followed by a call to | |
885 | * balance_callback(). | |
67dfa1b7 | 886 | */ |
cb469845 SR |
887 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
888 | const struct sched_class *prev_class, | |
da7a735e | 889 | int oldprio) |
cb469845 SR |
890 | { |
891 | if (prev_class != p->sched_class) { | |
892 | if (prev_class->switched_from) | |
da7a735e | 893 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 894 | |
da7a735e | 895 | p->sched_class->switched_to(rq, p); |
2d3d891d | 896 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 897 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
898 | } |
899 | ||
029632fb | 900 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
901 | { |
902 | const struct sched_class *class; | |
903 | ||
904 | if (p->sched_class == rq->curr->sched_class) { | |
905 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
906 | } else { | |
907 | for_each_class(class) { | |
908 | if (class == rq->curr->sched_class) | |
909 | break; | |
910 | if (class == p->sched_class) { | |
8875125e | 911 | resched_curr(rq); |
1e5a7405 PZ |
912 | break; |
913 | } | |
914 | } | |
915 | } | |
916 | ||
917 | /* | |
918 | * A queue event has occurred, and we're going to schedule. In | |
919 | * this case, we can save a useless back to back clock update. | |
920 | */ | |
da0c1e65 | 921 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
9edfbfed | 922 | rq_clock_skip_update(rq, true); |
1e5a7405 PZ |
923 | } |
924 | ||
1da177e4 | 925 | #ifdef CONFIG_SMP |
5cc389bc PZ |
926 | /* |
927 | * This is how migration works: | |
928 | * | |
929 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
930 | * stop_one_cpu(). | |
931 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
932 | * off the CPU) | |
933 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
934 | * 4) if it's in the wrong runqueue then the migration thread removes | |
935 | * it and puts it into the right queue. | |
936 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
937 | * is done. | |
938 | */ | |
939 | ||
940 | /* | |
941 | * move_queued_task - move a queued task to new rq. | |
942 | * | |
943 | * Returns (locked) new rq. Old rq's lock is released. | |
944 | */ | |
5e16bbc2 | 945 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
5cc389bc | 946 | { |
5cc389bc PZ |
947 | lockdep_assert_held(&rq->lock); |
948 | ||
5cc389bc | 949 | p->on_rq = TASK_ON_RQ_MIGRATING; |
3ea94de1 | 950 | dequeue_task(rq, p, 0); |
5cc389bc PZ |
951 | set_task_cpu(p, new_cpu); |
952 | raw_spin_unlock(&rq->lock); | |
953 | ||
954 | rq = cpu_rq(new_cpu); | |
955 | ||
956 | raw_spin_lock(&rq->lock); | |
957 | BUG_ON(task_cpu(p) != new_cpu); | |
5cc389bc | 958 | enqueue_task(rq, p, 0); |
3ea94de1 | 959 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
960 | check_preempt_curr(rq, p, 0); |
961 | ||
962 | return rq; | |
963 | } | |
964 | ||
965 | struct migration_arg { | |
966 | struct task_struct *task; | |
967 | int dest_cpu; | |
968 | }; | |
969 | ||
970 | /* | |
d1ccc66d | 971 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
972 | * this because either it can't run here any more (set_cpus_allowed() |
973 | * away from this CPU, or CPU going down), or because we're | |
974 | * attempting to rebalance this task on exec (sched_exec). | |
975 | * | |
976 | * So we race with normal scheduler movements, but that's OK, as long | |
977 | * as the task is no longer on this CPU. | |
5cc389bc | 978 | */ |
5e16bbc2 | 979 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
5cc389bc | 980 | { |
5cc389bc | 981 | if (unlikely(!cpu_active(dest_cpu))) |
5e16bbc2 | 982 | return rq; |
5cc389bc PZ |
983 | |
984 | /* Affinity changed (again). */ | |
0c98d344 | 985 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
5e16bbc2 | 986 | return rq; |
5cc389bc | 987 | |
5e16bbc2 PZ |
988 | rq = move_queued_task(rq, p, dest_cpu); |
989 | ||
990 | return rq; | |
5cc389bc PZ |
991 | } |
992 | ||
993 | /* | |
994 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
995 | * and performs thread migration by bumping thread off CPU then | |
996 | * 'pushing' onto another runqueue. | |
997 | */ | |
998 | static int migration_cpu_stop(void *data) | |
999 | { | |
1000 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1001 | struct task_struct *p = arg->task; |
1002 | struct rq *rq = this_rq(); | |
5cc389bc PZ |
1003 | |
1004 | /* | |
d1ccc66d IM |
1005 | * The original target CPU might have gone down and we might |
1006 | * be on another CPU but it doesn't matter. | |
5cc389bc PZ |
1007 | */ |
1008 | local_irq_disable(); | |
1009 | /* | |
1010 | * We need to explicitly wake pending tasks before running | |
1011 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
1012 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
1013 | */ | |
1014 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
1015 | |
1016 | raw_spin_lock(&p->pi_lock); | |
1017 | raw_spin_lock(&rq->lock); | |
1018 | /* | |
1019 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1020 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1021 | * we're holding p->pi_lock. | |
1022 | */ | |
bf89a304 CC |
1023 | if (task_rq(p) == rq) { |
1024 | if (task_on_rq_queued(p)) | |
1025 | rq = __migrate_task(rq, p, arg->dest_cpu); | |
1026 | else | |
1027 | p->wake_cpu = arg->dest_cpu; | |
1028 | } | |
5e16bbc2 PZ |
1029 | raw_spin_unlock(&rq->lock); |
1030 | raw_spin_unlock(&p->pi_lock); | |
1031 | ||
5cc389bc PZ |
1032 | local_irq_enable(); |
1033 | return 0; | |
1034 | } | |
1035 | ||
c5b28038 PZ |
1036 | /* |
1037 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1038 | * actually call this function. | |
1039 | */ | |
1040 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1041 | { |
5cc389bc PZ |
1042 | cpumask_copy(&p->cpus_allowed, new_mask); |
1043 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1044 | } | |
1045 | ||
c5b28038 PZ |
1046 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1047 | { | |
6c37067e PZ |
1048 | struct rq *rq = task_rq(p); |
1049 | bool queued, running; | |
1050 | ||
c5b28038 | 1051 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1052 | |
1053 | queued = task_on_rq_queued(p); | |
1054 | running = task_current(rq, p); | |
1055 | ||
1056 | if (queued) { | |
1057 | /* | |
1058 | * Because __kthread_bind() calls this on blocked tasks without | |
1059 | * holding rq->lock. | |
1060 | */ | |
1061 | lockdep_assert_held(&rq->lock); | |
1de64443 | 1062 | dequeue_task(rq, p, DEQUEUE_SAVE); |
6c37067e PZ |
1063 | } |
1064 | if (running) | |
1065 | put_prev_task(rq, p); | |
1066 | ||
c5b28038 | 1067 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e | 1068 | |
6c37067e | 1069 | if (queued) |
1de64443 | 1070 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
a399d233 | 1071 | if (running) |
b2bf6c31 | 1072 | set_curr_task(rq, p); |
c5b28038 PZ |
1073 | } |
1074 | ||
5cc389bc PZ |
1075 | /* |
1076 | * Change a given task's CPU affinity. Migrate the thread to a | |
1077 | * proper CPU and schedule it away if the CPU it's executing on | |
1078 | * is removed from the allowed bitmask. | |
1079 | * | |
1080 | * NOTE: the caller must have a valid reference to the task, the | |
1081 | * task must not exit() & deallocate itself prematurely. The | |
1082 | * call is not atomic; no spinlocks may be held. | |
1083 | */ | |
25834c73 PZ |
1084 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1085 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1086 | { |
e9d867a6 | 1087 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1088 | unsigned int dest_cpu; |
eb580751 PZ |
1089 | struct rq_flags rf; |
1090 | struct rq *rq; | |
5cc389bc PZ |
1091 | int ret = 0; |
1092 | ||
eb580751 | 1093 | rq = task_rq_lock(p, &rf); |
a499c3ea | 1094 | update_rq_clock(rq); |
5cc389bc | 1095 | |
e9d867a6 PZI |
1096 | if (p->flags & PF_KTHREAD) { |
1097 | /* | |
1098 | * Kernel threads are allowed on online && !active CPUs | |
1099 | */ | |
1100 | cpu_valid_mask = cpu_online_mask; | |
1101 | } | |
1102 | ||
25834c73 PZ |
1103 | /* |
1104 | * Must re-check here, to close a race against __kthread_bind(), | |
1105 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1106 | */ | |
1107 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1108 | ret = -EINVAL; | |
1109 | goto out; | |
1110 | } | |
1111 | ||
5cc389bc PZ |
1112 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1113 | goto out; | |
1114 | ||
e9d867a6 | 1115 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1116 | ret = -EINVAL; |
1117 | goto out; | |
1118 | } | |
1119 | ||
1120 | do_set_cpus_allowed(p, new_mask); | |
1121 | ||
e9d867a6 PZI |
1122 | if (p->flags & PF_KTHREAD) { |
1123 | /* | |
1124 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 1125 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
1126 | */ |
1127 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1128 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1129 | p->nr_cpus_allowed != 1); | |
1130 | } | |
1131 | ||
5cc389bc PZ |
1132 | /* Can the task run on the task's current CPU? If so, we're done */ |
1133 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1134 | goto out; | |
1135 | ||
e9d867a6 | 1136 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1137 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1138 | struct migration_arg arg = { p, dest_cpu }; | |
1139 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1140 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1141 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1142 | tlb_migrate_finish(p->mm); | |
1143 | return 0; | |
cbce1a68 PZ |
1144 | } else if (task_on_rq_queued(p)) { |
1145 | /* | |
1146 | * OK, since we're going to drop the lock immediately | |
1147 | * afterwards anyway. | |
1148 | */ | |
d8ac8971 | 1149 | rq_unpin_lock(rq, &rf); |
5e16bbc2 | 1150 | rq = move_queued_task(rq, p, dest_cpu); |
d8ac8971 | 1151 | rq_repin_lock(rq, &rf); |
cbce1a68 | 1152 | } |
5cc389bc | 1153 | out: |
eb580751 | 1154 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1155 | |
1156 | return ret; | |
1157 | } | |
25834c73 PZ |
1158 | |
1159 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1160 | { | |
1161 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1162 | } | |
5cc389bc PZ |
1163 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1164 | ||
dd41f596 | 1165 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1166 | { |
e2912009 PZ |
1167 | #ifdef CONFIG_SCHED_DEBUG |
1168 | /* | |
1169 | * We should never call set_task_cpu() on a blocked task, | |
1170 | * ttwu() will sort out the placement. | |
1171 | */ | |
077614ee | 1172 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1173 | !p->on_rq); |
0122ec5b | 1174 | |
3ea94de1 JP |
1175 | /* |
1176 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1177 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1178 | * time relying on p->on_rq. | |
1179 | */ | |
1180 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1181 | p->sched_class == &fair_sched_class && | |
1182 | (p->on_rq && !task_on_rq_migrating(p))); | |
1183 | ||
0122ec5b | 1184 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1185 | /* |
1186 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1187 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1188 | * | |
1189 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1190 | * see task_group(). |
6c6c54e1 PZ |
1191 | * |
1192 | * Furthermore, all task_rq users should acquire both locks, see | |
1193 | * task_rq_lock(). | |
1194 | */ | |
0122ec5b PZ |
1195 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1196 | lockdep_is_held(&task_rq(p)->lock))); | |
1197 | #endif | |
e2912009 PZ |
1198 | #endif |
1199 | ||
de1d7286 | 1200 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1201 | |
0c69774e | 1202 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1203 | if (p->sched_class->migrate_task_rq) |
5a4fd036 | 1204 | p->sched_class->migrate_task_rq(p); |
0c69774e | 1205 | p->se.nr_migrations++; |
ff303e66 | 1206 | perf_event_task_migrate(p); |
0c69774e | 1207 | } |
dd41f596 IM |
1208 | |
1209 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1210 | } |
1211 | ||
ac66f547 PZ |
1212 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1213 | { | |
da0c1e65 | 1214 | if (task_on_rq_queued(p)) { |
ac66f547 PZ |
1215 | struct rq *src_rq, *dst_rq; |
1216 | ||
1217 | src_rq = task_rq(p); | |
1218 | dst_rq = cpu_rq(cpu); | |
1219 | ||
3ea94de1 | 1220 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1221 | deactivate_task(src_rq, p, 0); |
1222 | set_task_cpu(p, cpu); | |
1223 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1224 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 PZ |
1225 | check_preempt_curr(dst_rq, p, 0); |
1226 | } else { | |
1227 | /* | |
1228 | * Task isn't running anymore; make it appear like we migrated | |
1229 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 1230 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
1231 | */ |
1232 | p->wake_cpu = cpu; | |
1233 | } | |
1234 | } | |
1235 | ||
1236 | struct migration_swap_arg { | |
1237 | struct task_struct *src_task, *dst_task; | |
1238 | int src_cpu, dst_cpu; | |
1239 | }; | |
1240 | ||
1241 | static int migrate_swap_stop(void *data) | |
1242 | { | |
1243 | struct migration_swap_arg *arg = data; | |
1244 | struct rq *src_rq, *dst_rq; | |
1245 | int ret = -EAGAIN; | |
1246 | ||
62694cd5 PZ |
1247 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1248 | return -EAGAIN; | |
1249 | ||
ac66f547 PZ |
1250 | src_rq = cpu_rq(arg->src_cpu); |
1251 | dst_rq = cpu_rq(arg->dst_cpu); | |
1252 | ||
74602315 PZ |
1253 | double_raw_lock(&arg->src_task->pi_lock, |
1254 | &arg->dst_task->pi_lock); | |
ac66f547 | 1255 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1256 | |
ac66f547 PZ |
1257 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1258 | goto unlock; | |
1259 | ||
1260 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1261 | goto unlock; | |
1262 | ||
0c98d344 | 1263 | if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed)) |
ac66f547 PZ |
1264 | goto unlock; |
1265 | ||
0c98d344 | 1266 | if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed)) |
ac66f547 PZ |
1267 | goto unlock; |
1268 | ||
1269 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1270 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1271 | ||
1272 | ret = 0; | |
1273 | ||
1274 | unlock: | |
1275 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1276 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1277 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1278 | |
1279 | return ret; | |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * Cross migrate two tasks | |
1284 | */ | |
1285 | int migrate_swap(struct task_struct *cur, struct task_struct *p) | |
1286 | { | |
1287 | struct migration_swap_arg arg; | |
1288 | int ret = -EINVAL; | |
1289 | ||
ac66f547 PZ |
1290 | arg = (struct migration_swap_arg){ |
1291 | .src_task = cur, | |
1292 | .src_cpu = task_cpu(cur), | |
1293 | .dst_task = p, | |
1294 | .dst_cpu = task_cpu(p), | |
1295 | }; | |
1296 | ||
1297 | if (arg.src_cpu == arg.dst_cpu) | |
1298 | goto out; | |
1299 | ||
6acce3ef PZ |
1300 | /* |
1301 | * These three tests are all lockless; this is OK since all of them | |
1302 | * will be re-checked with proper locks held further down the line. | |
1303 | */ | |
ac66f547 PZ |
1304 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1305 | goto out; | |
1306 | ||
0c98d344 | 1307 | if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed)) |
ac66f547 PZ |
1308 | goto out; |
1309 | ||
0c98d344 | 1310 | if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed)) |
ac66f547 PZ |
1311 | goto out; |
1312 | ||
286549dc | 1313 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1314 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1315 | ||
1316 | out: | |
ac66f547 PZ |
1317 | return ret; |
1318 | } | |
1319 | ||
1da177e4 LT |
1320 | /* |
1321 | * wait_task_inactive - wait for a thread to unschedule. | |
1322 | * | |
85ba2d86 RM |
1323 | * If @match_state is nonzero, it's the @p->state value just checked and |
1324 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1325 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1326 | * we return a positive number (its total switch count). If a second call | |
1327 | * a short while later returns the same number, the caller can be sure that | |
1328 | * @p has remained unscheduled the whole time. | |
1329 | * | |
1da177e4 LT |
1330 | * The caller must ensure that the task *will* unschedule sometime soon, |
1331 | * else this function might spin for a *long* time. This function can't | |
1332 | * be called with interrupts off, or it may introduce deadlock with | |
1333 | * smp_call_function() if an IPI is sent by the same process we are | |
1334 | * waiting to become inactive. | |
1335 | */ | |
85ba2d86 | 1336 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1337 | { |
da0c1e65 | 1338 | int running, queued; |
eb580751 | 1339 | struct rq_flags rf; |
85ba2d86 | 1340 | unsigned long ncsw; |
70b97a7f | 1341 | struct rq *rq; |
1da177e4 | 1342 | |
3a5c359a AK |
1343 | for (;;) { |
1344 | /* | |
1345 | * We do the initial early heuristics without holding | |
1346 | * any task-queue locks at all. We'll only try to get | |
1347 | * the runqueue lock when things look like they will | |
1348 | * work out! | |
1349 | */ | |
1350 | rq = task_rq(p); | |
fa490cfd | 1351 | |
3a5c359a AK |
1352 | /* |
1353 | * If the task is actively running on another CPU | |
1354 | * still, just relax and busy-wait without holding | |
1355 | * any locks. | |
1356 | * | |
1357 | * NOTE! Since we don't hold any locks, it's not | |
1358 | * even sure that "rq" stays as the right runqueue! | |
1359 | * But we don't care, since "task_running()" will | |
1360 | * return false if the runqueue has changed and p | |
1361 | * is actually now running somewhere else! | |
1362 | */ | |
85ba2d86 RM |
1363 | while (task_running(rq, p)) { |
1364 | if (match_state && unlikely(p->state != match_state)) | |
1365 | return 0; | |
3a5c359a | 1366 | cpu_relax(); |
85ba2d86 | 1367 | } |
fa490cfd | 1368 | |
3a5c359a AK |
1369 | /* |
1370 | * Ok, time to look more closely! We need the rq | |
1371 | * lock now, to be *sure*. If we're wrong, we'll | |
1372 | * just go back and repeat. | |
1373 | */ | |
eb580751 | 1374 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1375 | trace_sched_wait_task(p); |
3a5c359a | 1376 | running = task_running(rq, p); |
da0c1e65 | 1377 | queued = task_on_rq_queued(p); |
85ba2d86 | 1378 | ncsw = 0; |
f31e11d8 | 1379 | if (!match_state || p->state == match_state) |
93dcf55f | 1380 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1381 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1382 | |
85ba2d86 RM |
1383 | /* |
1384 | * If it changed from the expected state, bail out now. | |
1385 | */ | |
1386 | if (unlikely(!ncsw)) | |
1387 | break; | |
1388 | ||
3a5c359a AK |
1389 | /* |
1390 | * Was it really running after all now that we | |
1391 | * checked with the proper locks actually held? | |
1392 | * | |
1393 | * Oops. Go back and try again.. | |
1394 | */ | |
1395 | if (unlikely(running)) { | |
1396 | cpu_relax(); | |
1397 | continue; | |
1398 | } | |
fa490cfd | 1399 | |
3a5c359a AK |
1400 | /* |
1401 | * It's not enough that it's not actively running, | |
1402 | * it must be off the runqueue _entirely_, and not | |
1403 | * preempted! | |
1404 | * | |
80dd99b3 | 1405 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1406 | * running right now), it's preempted, and we should |
1407 | * yield - it could be a while. | |
1408 | */ | |
da0c1e65 | 1409 | if (unlikely(queued)) { |
8b0e1953 | 1410 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
1411 | |
1412 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1413 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1414 | continue; |
1415 | } | |
fa490cfd | 1416 | |
3a5c359a AK |
1417 | /* |
1418 | * Ahh, all good. It wasn't running, and it wasn't | |
1419 | * runnable, which means that it will never become | |
1420 | * running in the future either. We're all done! | |
1421 | */ | |
1422 | break; | |
1423 | } | |
85ba2d86 RM |
1424 | |
1425 | return ncsw; | |
1da177e4 LT |
1426 | } |
1427 | ||
1428 | /*** | |
1429 | * kick_process - kick a running thread to enter/exit the kernel | |
1430 | * @p: the to-be-kicked thread | |
1431 | * | |
1432 | * Cause a process which is running on another CPU to enter | |
1433 | * kernel-mode, without any delay. (to get signals handled.) | |
1434 | * | |
25985edc | 1435 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1436 | * because all it wants to ensure is that the remote task enters |
1437 | * the kernel. If the IPI races and the task has been migrated | |
1438 | * to another CPU then no harm is done and the purpose has been | |
1439 | * achieved as well. | |
1440 | */ | |
36c8b586 | 1441 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1442 | { |
1443 | int cpu; | |
1444 | ||
1445 | preempt_disable(); | |
1446 | cpu = task_cpu(p); | |
1447 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1448 | smp_send_reschedule(cpu); | |
1449 | preempt_enable(); | |
1450 | } | |
b43e3521 | 1451 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1452 | |
30da688e | 1453 | /* |
013fdb80 | 1454 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1455 | * |
1456 | * A few notes on cpu_active vs cpu_online: | |
1457 | * | |
1458 | * - cpu_active must be a subset of cpu_online | |
1459 | * | |
1460 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, | |
1461 | * see __set_cpus_allowed_ptr(). At this point the newly online | |
d1ccc66d | 1462 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
1463 | * see it. |
1464 | * | |
d1ccc66d | 1465 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 1466 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 1467 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
1468 | * off. |
1469 | * | |
1470 | * This means that fallback selection must not select !active CPUs. | |
1471 | * And can assume that any active CPU must be online. Conversely | |
1472 | * select_task_rq() below may allow selection of !active CPUs in order | |
1473 | * to satisfy the above rules. | |
30da688e | 1474 | */ |
5da9a0fb PZ |
1475 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1476 | { | |
aa00d89c TC |
1477 | int nid = cpu_to_node(cpu); |
1478 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1479 | enum { cpuset, possible, fail } state = cpuset; |
1480 | int dest_cpu; | |
5da9a0fb | 1481 | |
aa00d89c | 1482 | /* |
d1ccc66d IM |
1483 | * If the node that the CPU is on has been offlined, cpu_to_node() |
1484 | * will return -1. There is no CPU on the node, and we should | |
1485 | * select the CPU on the other node. | |
aa00d89c TC |
1486 | */ |
1487 | if (nid != -1) { | |
1488 | nodemask = cpumask_of_node(nid); | |
1489 | ||
1490 | /* Look for allowed, online CPU in same node. */ | |
1491 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1492 | if (!cpu_active(dest_cpu)) |
1493 | continue; | |
0c98d344 | 1494 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
aa00d89c TC |
1495 | return dest_cpu; |
1496 | } | |
2baab4e9 | 1497 | } |
5da9a0fb | 1498 | |
2baab4e9 PZ |
1499 | for (;;) { |
1500 | /* Any allowed, online CPU? */ | |
0c98d344 | 1501 | for_each_cpu(dest_cpu, &p->cpus_allowed) { |
feb245e3 TH |
1502 | if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu)) |
1503 | continue; | |
1504 | if (!cpu_online(dest_cpu)) | |
2baab4e9 PZ |
1505 | continue; |
1506 | goto out; | |
1507 | } | |
5da9a0fb | 1508 | |
e73e85f0 | 1509 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1510 | switch (state) { |
1511 | case cpuset: | |
e73e85f0 ON |
1512 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1513 | cpuset_cpus_allowed_fallback(p); | |
1514 | state = possible; | |
1515 | break; | |
1516 | } | |
d1ccc66d | 1517 | /* Fall-through */ |
2baab4e9 PZ |
1518 | case possible: |
1519 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1520 | state = fail; | |
1521 | break; | |
1522 | ||
1523 | case fail: | |
1524 | BUG(); | |
1525 | break; | |
1526 | } | |
1527 | } | |
1528 | ||
1529 | out: | |
1530 | if (state != cpuset) { | |
1531 | /* | |
1532 | * Don't tell them about moving exiting tasks or | |
1533 | * kernel threads (both mm NULL), since they never | |
1534 | * leave kernel. | |
1535 | */ | |
1536 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1537 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1538 | task_pid_nr(p), p->comm, cpu); |
1539 | } | |
5da9a0fb PZ |
1540 | } |
1541 | ||
1542 | return dest_cpu; | |
1543 | } | |
1544 | ||
e2912009 | 1545 | /* |
013fdb80 | 1546 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1547 | */ |
970b13ba | 1548 | static inline |
ac66f547 | 1549 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1550 | { |
cbce1a68 PZ |
1551 | lockdep_assert_held(&p->pi_lock); |
1552 | ||
4b53a341 | 1553 | if (p->nr_cpus_allowed > 1) |
6c1d9410 | 1554 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 | 1555 | else |
0c98d344 | 1556 | cpu = cpumask_any(&p->cpus_allowed); |
e2912009 PZ |
1557 | |
1558 | /* | |
1559 | * In order not to call set_task_cpu() on a blocking task we need | |
1560 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
d1ccc66d | 1561 | * CPU. |
e2912009 PZ |
1562 | * |
1563 | * Since this is common to all placement strategies, this lives here. | |
1564 | * | |
1565 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1566 | * not worry about this generic constraint ] | |
1567 | */ | |
0c98d344 | 1568 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || |
70f11205 | 1569 | !cpu_online(cpu))) |
5da9a0fb | 1570 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1571 | |
1572 | return cpu; | |
970b13ba | 1573 | } |
09a40af5 MG |
1574 | |
1575 | static void update_avg(u64 *avg, u64 sample) | |
1576 | { | |
1577 | s64 diff = sample - *avg; | |
1578 | *avg += diff >> 3; | |
1579 | } | |
25834c73 PZ |
1580 | |
1581 | #else | |
1582 | ||
1583 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1584 | const struct cpumask *new_mask, bool check) | |
1585 | { | |
1586 | return set_cpus_allowed_ptr(p, new_mask); | |
1587 | } | |
1588 | ||
5cc389bc | 1589 | #endif /* CONFIG_SMP */ |
970b13ba | 1590 | |
d7c01d27 | 1591 | static void |
b84cb5df | 1592 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1593 | { |
4fa8d299 | 1594 | struct rq *rq; |
b84cb5df | 1595 | |
4fa8d299 JP |
1596 | if (!schedstat_enabled()) |
1597 | return; | |
1598 | ||
1599 | rq = this_rq(); | |
d7c01d27 | 1600 | |
4fa8d299 JP |
1601 | #ifdef CONFIG_SMP |
1602 | if (cpu == rq->cpu) { | |
ae92882e JP |
1603 | schedstat_inc(rq->ttwu_local); |
1604 | schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
1605 | } else { |
1606 | struct sched_domain *sd; | |
1607 | ||
ae92882e | 1608 | schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 1609 | rcu_read_lock(); |
4fa8d299 | 1610 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 1611 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
ae92882e | 1612 | schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
1613 | break; |
1614 | } | |
1615 | } | |
057f3fad | 1616 | rcu_read_unlock(); |
d7c01d27 | 1617 | } |
f339b9dc PZ |
1618 | |
1619 | if (wake_flags & WF_MIGRATED) | |
ae92882e | 1620 | schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
1621 | #endif /* CONFIG_SMP */ |
1622 | ||
ae92882e JP |
1623 | schedstat_inc(rq->ttwu_count); |
1624 | schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
1625 | |
1626 | if (wake_flags & WF_SYNC) | |
ae92882e | 1627 | schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
1628 | } |
1629 | ||
1de64443 | 1630 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1631 | { |
9ed3811a | 1632 | activate_task(rq, p, en_flags); |
da0c1e65 | 1633 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e | 1634 | |
d1ccc66d | 1635 | /* If a worker is waking up, notify the workqueue: */ |
c2f7115e PZ |
1636 | if (p->flags & PF_WQ_WORKER) |
1637 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1638 | } |
1639 | ||
23f41eeb PZ |
1640 | /* |
1641 | * Mark the task runnable and perform wakeup-preemption. | |
1642 | */ | |
e7904a28 | 1643 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1644 | struct rq_flags *rf) |
9ed3811a | 1645 | { |
9ed3811a | 1646 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1647 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1648 | trace_sched_wakeup(p); |
1649 | ||
9ed3811a | 1650 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1651 | if (p->sched_class->task_woken) { |
1652 | /* | |
cbce1a68 PZ |
1653 | * Our task @p is fully woken up and running; so its safe to |
1654 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1655 | */ |
d8ac8971 | 1656 | rq_unpin_lock(rq, rf); |
9ed3811a | 1657 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 1658 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 1659 | } |
9ed3811a | 1660 | |
e69c6341 | 1661 | if (rq->idle_stamp) { |
78becc27 | 1662 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1663 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1664 | |
abfafa54 JL |
1665 | update_avg(&rq->avg_idle, delta); |
1666 | ||
1667 | if (rq->avg_idle > max) | |
9ed3811a | 1668 | rq->avg_idle = max; |
abfafa54 | 1669 | |
9ed3811a TH |
1670 | rq->idle_stamp = 0; |
1671 | } | |
1672 | #endif | |
1673 | } | |
1674 | ||
c05fbafb | 1675 | static void |
e7904a28 | 1676 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1677 | struct rq_flags *rf) |
c05fbafb | 1678 | { |
b5179ac7 PZ |
1679 | int en_flags = ENQUEUE_WAKEUP; |
1680 | ||
cbce1a68 PZ |
1681 | lockdep_assert_held(&rq->lock); |
1682 | ||
c05fbafb PZ |
1683 | #ifdef CONFIG_SMP |
1684 | if (p->sched_contributes_to_load) | |
1685 | rq->nr_uninterruptible--; | |
b5179ac7 | 1686 | |
b5179ac7 | 1687 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 1688 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
1689 | #endif |
1690 | ||
b5179ac7 | 1691 | ttwu_activate(rq, p, en_flags); |
d8ac8971 | 1692 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
1693 | } |
1694 | ||
1695 | /* | |
1696 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1697 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1698 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1699 | * the task is still ->on_rq. | |
1700 | */ | |
1701 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1702 | { | |
eb580751 | 1703 | struct rq_flags rf; |
c05fbafb PZ |
1704 | struct rq *rq; |
1705 | int ret = 0; | |
1706 | ||
eb580751 | 1707 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 1708 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1709 | /* check_preempt_curr() may use rq clock */ |
1710 | update_rq_clock(rq); | |
d8ac8971 | 1711 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
1712 | ret = 1; |
1713 | } | |
eb580751 | 1714 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
1715 | |
1716 | return ret; | |
1717 | } | |
1718 | ||
317f3941 | 1719 | #ifdef CONFIG_SMP |
e3baac47 | 1720 | void sched_ttwu_pending(void) |
317f3941 PZ |
1721 | { |
1722 | struct rq *rq = this_rq(); | |
fa14ff4a PZ |
1723 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
1724 | struct task_struct *p; | |
e3baac47 | 1725 | unsigned long flags; |
d8ac8971 | 1726 | struct rq_flags rf; |
317f3941 | 1727 | |
e3baac47 PZ |
1728 | if (!llist) |
1729 | return; | |
1730 | ||
1731 | raw_spin_lock_irqsave(&rq->lock, flags); | |
d8ac8971 | 1732 | rq_pin_lock(rq, &rf); |
317f3941 | 1733 | |
fa14ff4a | 1734 | while (llist) { |
b7e7ade3 PZ |
1735 | int wake_flags = 0; |
1736 | ||
fa14ff4a PZ |
1737 | p = llist_entry(llist, struct task_struct, wake_entry); |
1738 | llist = llist_next(llist); | |
b7e7ade3 PZ |
1739 | |
1740 | if (p->sched_remote_wakeup) | |
1741 | wake_flags = WF_MIGRATED; | |
1742 | ||
d8ac8971 | 1743 | ttwu_do_activate(rq, p, wake_flags, &rf); |
317f3941 PZ |
1744 | } |
1745 | ||
d8ac8971 | 1746 | rq_unpin_lock(rq, &rf); |
e3baac47 | 1747 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
317f3941 PZ |
1748 | } |
1749 | ||
1750 | void scheduler_ipi(void) | |
1751 | { | |
f27dde8d PZ |
1752 | /* |
1753 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting | |
1754 | * TIF_NEED_RESCHED remotely (for the first time) will also send | |
1755 | * this IPI. | |
1756 | */ | |
8cb75e0c | 1757 | preempt_fold_need_resched(); |
f27dde8d | 1758 | |
fd2ac4f4 | 1759 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
c5d753a5 PZ |
1760 | return; |
1761 | ||
1762 | /* | |
1763 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since | |
1764 | * traditionally all their work was done from the interrupt return | |
1765 | * path. Now that we actually do some work, we need to make sure | |
1766 | * we do call them. | |
1767 | * | |
1768 | * Some archs already do call them, luckily irq_enter/exit nest | |
1769 | * properly. | |
1770 | * | |
1771 | * Arguably we should visit all archs and update all handlers, | |
1772 | * however a fair share of IPIs are still resched only so this would | |
1773 | * somewhat pessimize the simple resched case. | |
1774 | */ | |
1775 | irq_enter(); | |
fa14ff4a | 1776 | sched_ttwu_pending(); |
ca38062e SS |
1777 | |
1778 | /* | |
1779 | * Check if someone kicked us for doing the nohz idle load balance. | |
1780 | */ | |
873b4c65 | 1781 | if (unlikely(got_nohz_idle_kick())) { |
6eb57e0d | 1782 | this_rq()->idle_balance = 1; |
ca38062e | 1783 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
6eb57e0d | 1784 | } |
c5d753a5 | 1785 | irq_exit(); |
317f3941 PZ |
1786 | } |
1787 | ||
b7e7ade3 | 1788 | static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags) |
317f3941 | 1789 | { |
e3baac47 PZ |
1790 | struct rq *rq = cpu_rq(cpu); |
1791 | ||
b7e7ade3 PZ |
1792 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
1793 | ||
e3baac47 PZ |
1794 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { |
1795 | if (!set_nr_if_polling(rq->idle)) | |
1796 | smp_send_reschedule(cpu); | |
1797 | else | |
1798 | trace_sched_wake_idle_without_ipi(cpu); | |
1799 | } | |
317f3941 | 1800 | } |
d6aa8f85 | 1801 | |
f6be8af1 CL |
1802 | void wake_up_if_idle(int cpu) |
1803 | { | |
1804 | struct rq *rq = cpu_rq(cpu); | |
1805 | unsigned long flags; | |
1806 | ||
fd7de1e8 AL |
1807 | rcu_read_lock(); |
1808 | ||
1809 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
1810 | goto out; | |
f6be8af1 CL |
1811 | |
1812 | if (set_nr_if_polling(rq->idle)) { | |
1813 | trace_sched_wake_idle_without_ipi(cpu); | |
1814 | } else { | |
1815 | raw_spin_lock_irqsave(&rq->lock, flags); | |
1816 | if (is_idle_task(rq->curr)) | |
1817 | smp_send_reschedule(cpu); | |
d1ccc66d | 1818 | /* Else CPU is not idle, do nothing here: */ |
f6be8af1 CL |
1819 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1820 | } | |
fd7de1e8 AL |
1821 | |
1822 | out: | |
1823 | rcu_read_unlock(); | |
f6be8af1 CL |
1824 | } |
1825 | ||
39be3501 | 1826 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
1827 | { |
1828 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
1829 | } | |
d6aa8f85 | 1830 | #endif /* CONFIG_SMP */ |
317f3941 | 1831 | |
b5179ac7 | 1832 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
1833 | { |
1834 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 1835 | struct rq_flags rf; |
c05fbafb | 1836 | |
17d9f311 | 1837 | #if defined(CONFIG_SMP) |
39be3501 | 1838 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
d1ccc66d | 1839 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
b7e7ade3 | 1840 | ttwu_queue_remote(p, cpu, wake_flags); |
317f3941 PZ |
1841 | return; |
1842 | } | |
1843 | #endif | |
1844 | ||
c05fbafb | 1845 | raw_spin_lock(&rq->lock); |
d8ac8971 MF |
1846 | rq_pin_lock(rq, &rf); |
1847 | ttwu_do_activate(rq, p, wake_flags, &rf); | |
1848 | rq_unpin_lock(rq, &rf); | |
c05fbafb | 1849 | raw_spin_unlock(&rq->lock); |
9ed3811a TH |
1850 | } |
1851 | ||
8643cda5 PZ |
1852 | /* |
1853 | * Notes on Program-Order guarantees on SMP systems. | |
1854 | * | |
1855 | * MIGRATION | |
1856 | * | |
1857 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
1858 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
1859 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
1860 | * |
1861 | * For migration (of runnable tasks) this is provided by the following means: | |
1862 | * | |
1863 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1864 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1865 | * rq(c1)->lock (if not at the same time, then in that order). | |
1866 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1867 | * | |
1868 | * Transitivity guarantees that B happens after A and C after B. | |
1869 | * Note: we only require RCpc transitivity. | |
d1ccc66d | 1870 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
1871 | * |
1872 | * Example: | |
1873 | * | |
1874 | * CPU0 CPU1 CPU2 | |
1875 | * | |
1876 | * LOCK rq(0)->lock | |
1877 | * sched-out X | |
1878 | * sched-in Y | |
1879 | * UNLOCK rq(0)->lock | |
1880 | * | |
1881 | * LOCK rq(0)->lock // orders against CPU0 | |
1882 | * dequeue X | |
1883 | * UNLOCK rq(0)->lock | |
1884 | * | |
1885 | * LOCK rq(1)->lock | |
1886 | * enqueue X | |
1887 | * UNLOCK rq(1)->lock | |
1888 | * | |
1889 | * LOCK rq(1)->lock // orders against CPU2 | |
1890 | * sched-out Z | |
1891 | * sched-in X | |
1892 | * UNLOCK rq(1)->lock | |
1893 | * | |
1894 | * | |
1895 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1896 | * | |
1897 | * For blocking we (obviously) need to provide the same guarantee as for | |
1898 | * migration. However the means are completely different as there is no lock | |
1899 | * chain to provide order. Instead we do: | |
1900 | * | |
1901 | * 1) smp_store_release(X->on_cpu, 0) | |
1f03e8d2 | 1902 | * 2) smp_cond_load_acquire(!X->on_cpu) |
8643cda5 PZ |
1903 | * |
1904 | * Example: | |
1905 | * | |
1906 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1907 | * | |
1908 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1909 | * dequeue X | |
1910 | * sched-out X | |
1911 | * smp_store_release(X->on_cpu, 0); | |
1912 | * | |
1f03e8d2 | 1913 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
1914 | * X->state = WAKING |
1915 | * set_task_cpu(X,2) | |
1916 | * | |
1917 | * LOCK rq(2)->lock | |
1918 | * enqueue X | |
1919 | * X->state = RUNNING | |
1920 | * UNLOCK rq(2)->lock | |
1921 | * | |
1922 | * LOCK rq(2)->lock // orders against CPU1 | |
1923 | * sched-out Z | |
1924 | * sched-in X | |
1925 | * UNLOCK rq(2)->lock | |
1926 | * | |
1927 | * UNLOCK X->pi_lock | |
1928 | * UNLOCK rq(0)->lock | |
1929 | * | |
1930 | * | |
1931 | * However; for wakeups there is a second guarantee we must provide, namely we | |
1932 | * must observe the state that lead to our wakeup. That is, not only must our | |
1933 | * task observe its own prior state, it must also observe the stores prior to | |
1934 | * its wakeup. | |
1935 | * | |
1936 | * This means that any means of doing remote wakeups must order the CPU doing | |
1937 | * the wakeup against the CPU the task is going to end up running on. This, | |
1938 | * however, is already required for the regular Program-Order guarantee above, | |
1f03e8d2 | 1939 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire). |
8643cda5 PZ |
1940 | * |
1941 | */ | |
1942 | ||
9ed3811a | 1943 | /** |
1da177e4 | 1944 | * try_to_wake_up - wake up a thread |
9ed3811a | 1945 | * @p: the thread to be awakened |
1da177e4 | 1946 | * @state: the mask of task states that can be woken |
9ed3811a | 1947 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 1948 | * |
a2250238 | 1949 | * If (@state & @p->state) @p->state = TASK_RUNNING. |
1da177e4 | 1950 | * |
a2250238 PZ |
1951 | * If the task was not queued/runnable, also place it back on a runqueue. |
1952 | * | |
1953 | * Atomic against schedule() which would dequeue a task, also see | |
1954 | * set_current_state(). | |
1955 | * | |
1956 | * Return: %true if @p->state changes (an actual wakeup was done), | |
1957 | * %false otherwise. | |
1da177e4 | 1958 | */ |
e4a52bcb PZ |
1959 | static int |
1960 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1961 | { |
1da177e4 | 1962 | unsigned long flags; |
c05fbafb | 1963 | int cpu, success = 0; |
2398f2c6 | 1964 | |
e0acd0a6 ON |
1965 | /* |
1966 | * If we are going to wake up a thread waiting for CONDITION we | |
1967 | * need to ensure that CONDITION=1 done by the caller can not be | |
1968 | * reordered with p->state check below. This pairs with mb() in | |
1969 | * set_current_state() the waiting thread does. | |
1970 | */ | |
1971 | smp_mb__before_spinlock(); | |
013fdb80 | 1972 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 1973 | if (!(p->state & state)) |
1da177e4 LT |
1974 | goto out; |
1975 | ||
fbd705a0 PZ |
1976 | trace_sched_waking(p); |
1977 | ||
d1ccc66d IM |
1978 | /* We're going to change ->state: */ |
1979 | success = 1; | |
1da177e4 | 1980 | cpu = task_cpu(p); |
1da177e4 | 1981 | |
135e8c92 BS |
1982 | /* |
1983 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
1984 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
1985 | * in smp_cond_load_acquire() below. | |
1986 | * | |
1987 | * sched_ttwu_pending() try_to_wake_up() | |
1988 | * [S] p->on_rq = 1; [L] P->state | |
1989 | * UNLOCK rq->lock -----. | |
1990 | * \ | |
1991 | * +--- RMB | |
1992 | * schedule() / | |
1993 | * LOCK rq->lock -----' | |
1994 | * UNLOCK rq->lock | |
1995 | * | |
1996 | * [task p] | |
1997 | * [S] p->state = UNINTERRUPTIBLE [L] p->on_rq | |
1998 | * | |
1999 | * Pairs with the UNLOCK+LOCK on rq->lock from the | |
2000 | * last wakeup of our task and the schedule that got our task | |
2001 | * current. | |
2002 | */ | |
2003 | smp_rmb(); | |
c05fbafb PZ |
2004 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2005 | goto stat; | |
1da177e4 | 2006 | |
1da177e4 | 2007 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2008 | /* |
2009 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2010 | * possible to, falsely, observe p->on_cpu == 0. | |
2011 | * | |
2012 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2013 | * from the runqueue. | |
2014 | * | |
2015 | * [S] ->on_cpu = 1; [L] ->on_rq | |
2016 | * UNLOCK rq->lock | |
2017 | * RMB | |
2018 | * LOCK rq->lock | |
2019 | * [S] ->on_rq = 0; [L] ->on_cpu | |
2020 | * | |
2021 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock | |
2022 | * from the consecutive calls to schedule(); the first switching to our | |
2023 | * task, the second putting it to sleep. | |
2024 | */ | |
2025 | smp_rmb(); | |
2026 | ||
e9c84311 | 2027 | /* |
d1ccc66d | 2028 | * If the owning (remote) CPU is still in the middle of schedule() with |
c05fbafb | 2029 | * this task as prev, wait until its done referencing the task. |
b75a2253 PZ |
2030 | * |
2031 | * Pairs with the smp_store_release() in finish_lock_switch(). | |
2032 | * | |
2033 | * This ensures that tasks getting woken will be fully ordered against | |
2034 | * their previous state and preserve Program Order. | |
0970d299 | 2035 | */ |
1f03e8d2 | 2036 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2037 | |
a8e4f2ea | 2038 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2039 | p->state = TASK_WAKING; |
e7693a36 | 2040 | |
e33a9bba TH |
2041 | if (p->in_iowait) { |
2042 | delayacct_blkio_end(); | |
2043 | atomic_dec(&task_rq(p)->nr_iowait); | |
2044 | } | |
2045 | ||
ac66f547 | 2046 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2047 | if (task_cpu(p) != cpu) { |
2048 | wake_flags |= WF_MIGRATED; | |
e4a52bcb | 2049 | set_task_cpu(p, cpu); |
f339b9dc | 2050 | } |
e33a9bba TH |
2051 | |
2052 | #else /* CONFIG_SMP */ | |
2053 | ||
2054 | if (p->in_iowait) { | |
2055 | delayacct_blkio_end(); | |
2056 | atomic_dec(&task_rq(p)->nr_iowait); | |
2057 | } | |
2058 | ||
1da177e4 | 2059 | #endif /* CONFIG_SMP */ |
1da177e4 | 2060 | |
b5179ac7 | 2061 | ttwu_queue(p, cpu, wake_flags); |
c05fbafb | 2062 | stat: |
4fa8d299 | 2063 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2064 | out: |
013fdb80 | 2065 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2066 | |
2067 | return success; | |
2068 | } | |
2069 | ||
21aa9af0 TH |
2070 | /** |
2071 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2072 | * @p: the thread to be awakened | |
9279e0d2 | 2073 | * @cookie: context's cookie for pinning |
21aa9af0 | 2074 | * |
2acca55e | 2075 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2076 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2077 | * the current task. |
21aa9af0 | 2078 | */ |
d8ac8971 | 2079 | static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf) |
21aa9af0 TH |
2080 | { |
2081 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2082 | |
383efcd0 TH |
2083 | if (WARN_ON_ONCE(rq != this_rq()) || |
2084 | WARN_ON_ONCE(p == current)) | |
2085 | return; | |
2086 | ||
21aa9af0 TH |
2087 | lockdep_assert_held(&rq->lock); |
2088 | ||
2acca55e | 2089 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2090 | /* |
2091 | * This is OK, because current is on_cpu, which avoids it being | |
2092 | * picked for load-balance and preemption/IRQs are still | |
2093 | * disabled avoiding further scheduler activity on it and we've | |
2094 | * not yet picked a replacement task. | |
2095 | */ | |
d8ac8971 | 2096 | rq_unpin_lock(rq, rf); |
2acca55e PZ |
2097 | raw_spin_unlock(&rq->lock); |
2098 | raw_spin_lock(&p->pi_lock); | |
2099 | raw_spin_lock(&rq->lock); | |
d8ac8971 | 2100 | rq_repin_lock(rq, rf); |
2acca55e PZ |
2101 | } |
2102 | ||
21aa9af0 | 2103 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2104 | goto out; |
21aa9af0 | 2105 | |
fbd705a0 PZ |
2106 | trace_sched_waking(p); |
2107 | ||
e33a9bba TH |
2108 | if (!task_on_rq_queued(p)) { |
2109 | if (p->in_iowait) { | |
2110 | delayacct_blkio_end(); | |
2111 | atomic_dec(&rq->nr_iowait); | |
2112 | } | |
d7c01d27 | 2113 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
e33a9bba | 2114 | } |
d7c01d27 | 2115 | |
d8ac8971 | 2116 | ttwu_do_wakeup(rq, p, 0, rf); |
4fa8d299 | 2117 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2118 | out: |
2119 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2120 | } |
2121 | ||
50fa610a DH |
2122 | /** |
2123 | * wake_up_process - Wake up a specific process | |
2124 | * @p: The process to be woken up. | |
2125 | * | |
2126 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2127 | * processes. |
2128 | * | |
2129 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a DH |
2130 | * |
2131 | * It may be assumed that this function implies a write memory barrier before | |
2132 | * changing the task state if and only if any tasks are woken up. | |
2133 | */ | |
7ad5b3a5 | 2134 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2135 | { |
9067ac85 | 2136 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2137 | } |
1da177e4 LT |
2138 | EXPORT_SYMBOL(wake_up_process); |
2139 | ||
7ad5b3a5 | 2140 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2141 | { |
2142 | return try_to_wake_up(p, state, 0); | |
2143 | } | |
2144 | ||
a5e7be3b JL |
2145 | /* |
2146 | * This function clears the sched_dl_entity static params. | |
2147 | */ | |
2148 | void __dl_clear_params(struct task_struct *p) | |
2149 | { | |
2150 | struct sched_dl_entity *dl_se = &p->dl; | |
2151 | ||
2152 | dl_se->dl_runtime = 0; | |
2153 | dl_se->dl_deadline = 0; | |
2154 | dl_se->dl_period = 0; | |
2155 | dl_se->flags = 0; | |
2156 | dl_se->dl_bw = 0; | |
40767b0d PZ |
2157 | |
2158 | dl_se->dl_throttled = 0; | |
40767b0d | 2159 | dl_se->dl_yielded = 0; |
a5e7be3b JL |
2160 | } |
2161 | ||
1da177e4 LT |
2162 | /* |
2163 | * Perform scheduler related setup for a newly forked process p. | |
2164 | * p is forked by current. | |
dd41f596 IM |
2165 | * |
2166 | * __sched_fork() is basic setup used by init_idle() too: | |
2167 | */ | |
5e1576ed | 2168 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2169 | { |
fd2f4419 PZ |
2170 | p->on_rq = 0; |
2171 | ||
2172 | p->se.on_rq = 0; | |
dd41f596 IM |
2173 | p->se.exec_start = 0; |
2174 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2175 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2176 | p->se.nr_migrations = 0; |
da7a735e | 2177 | p->se.vruntime = 0; |
fd2f4419 | 2178 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2179 | |
ad936d86 BP |
2180 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2181 | p->se.cfs_rq = NULL; | |
2182 | #endif | |
2183 | ||
6cfb0d5d | 2184 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2185 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2186 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2187 | #endif |
476d139c | 2188 | |
aab03e05 | 2189 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2190 | init_dl_task_timer(&p->dl); |
a5e7be3b | 2191 | __dl_clear_params(p); |
aab03e05 | 2192 | |
fa717060 | 2193 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2194 | p->rt.timeout = 0; |
2195 | p->rt.time_slice = sched_rr_timeslice; | |
2196 | p->rt.on_rq = 0; | |
2197 | p->rt.on_list = 0; | |
476d139c | 2198 | |
e107be36 AK |
2199 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2200 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2201 | #endif | |
cbee9f88 PZ |
2202 | |
2203 | #ifdef CONFIG_NUMA_BALANCING | |
2204 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { | |
7e8d16b6 | 2205 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
cbee9f88 PZ |
2206 | p->mm->numa_scan_seq = 0; |
2207 | } | |
2208 | ||
5e1576ed RR |
2209 | if (clone_flags & CLONE_VM) |
2210 | p->numa_preferred_nid = current->numa_preferred_nid; | |
2211 | else | |
2212 | p->numa_preferred_nid = -1; | |
2213 | ||
cbee9f88 PZ |
2214 | p->node_stamp = 0ULL; |
2215 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; | |
4b96a29b | 2216 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
cbee9f88 | 2217 | p->numa_work.next = &p->numa_work; |
44dba3d5 | 2218 | p->numa_faults = NULL; |
7e2703e6 RR |
2219 | p->last_task_numa_placement = 0; |
2220 | p->last_sum_exec_runtime = 0; | |
8c8a743c | 2221 | |
8c8a743c | 2222 | p->numa_group = NULL; |
cbee9f88 | 2223 | #endif /* CONFIG_NUMA_BALANCING */ |
dd41f596 IM |
2224 | } |
2225 | ||
2a595721 SD |
2226 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2227 | ||
1a687c2e | 2228 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2229 | |
1a687c2e MG |
2230 | void set_numabalancing_state(bool enabled) |
2231 | { | |
2232 | if (enabled) | |
2a595721 | 2233 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2234 | else |
2a595721 | 2235 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2236 | } |
54a43d54 AK |
2237 | |
2238 | #ifdef CONFIG_PROC_SYSCTL | |
2239 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2240 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2241 | { | |
2242 | struct ctl_table t; | |
2243 | int err; | |
2a595721 | 2244 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2245 | |
2246 | if (write && !capable(CAP_SYS_ADMIN)) | |
2247 | return -EPERM; | |
2248 | ||
2249 | t = *table; | |
2250 | t.data = &state; | |
2251 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2252 | if (err < 0) | |
2253 | return err; | |
2254 | if (write) | |
2255 | set_numabalancing_state(state); | |
2256 | return err; | |
2257 | } | |
2258 | #endif | |
2259 | #endif | |
dd41f596 | 2260 | |
4698f88c JP |
2261 | #ifdef CONFIG_SCHEDSTATS |
2262 | ||
cb251765 | 2263 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 2264 | static bool __initdata __sched_schedstats = false; |
cb251765 | 2265 | |
cb251765 MG |
2266 | static void set_schedstats(bool enabled) |
2267 | { | |
2268 | if (enabled) | |
2269 | static_branch_enable(&sched_schedstats); | |
2270 | else | |
2271 | static_branch_disable(&sched_schedstats); | |
2272 | } | |
2273 | ||
2274 | void force_schedstat_enabled(void) | |
2275 | { | |
2276 | if (!schedstat_enabled()) { | |
2277 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2278 | static_branch_enable(&sched_schedstats); | |
2279 | } | |
2280 | } | |
2281 | ||
2282 | static int __init setup_schedstats(char *str) | |
2283 | { | |
2284 | int ret = 0; | |
2285 | if (!str) | |
2286 | goto out; | |
2287 | ||
4698f88c JP |
2288 | /* |
2289 | * This code is called before jump labels have been set up, so we can't | |
2290 | * change the static branch directly just yet. Instead set a temporary | |
2291 | * variable so init_schedstats() can do it later. | |
2292 | */ | |
cb251765 | 2293 | if (!strcmp(str, "enable")) { |
4698f88c | 2294 | __sched_schedstats = true; |
cb251765 MG |
2295 | ret = 1; |
2296 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 2297 | __sched_schedstats = false; |
cb251765 MG |
2298 | ret = 1; |
2299 | } | |
2300 | out: | |
2301 | if (!ret) | |
2302 | pr_warn("Unable to parse schedstats=\n"); | |
2303 | ||
2304 | return ret; | |
2305 | } | |
2306 | __setup("schedstats=", setup_schedstats); | |
2307 | ||
4698f88c JP |
2308 | static void __init init_schedstats(void) |
2309 | { | |
2310 | set_schedstats(__sched_schedstats); | |
2311 | } | |
2312 | ||
cb251765 MG |
2313 | #ifdef CONFIG_PROC_SYSCTL |
2314 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2315 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2316 | { | |
2317 | struct ctl_table t; | |
2318 | int err; | |
2319 | int state = static_branch_likely(&sched_schedstats); | |
2320 | ||
2321 | if (write && !capable(CAP_SYS_ADMIN)) | |
2322 | return -EPERM; | |
2323 | ||
2324 | t = *table; | |
2325 | t.data = &state; | |
2326 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2327 | if (err < 0) | |
2328 | return err; | |
2329 | if (write) | |
2330 | set_schedstats(state); | |
2331 | return err; | |
2332 | } | |
4698f88c JP |
2333 | #endif /* CONFIG_PROC_SYSCTL */ |
2334 | #else /* !CONFIG_SCHEDSTATS */ | |
2335 | static inline void init_schedstats(void) {} | |
2336 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
2337 | |
2338 | /* | |
2339 | * fork()/clone()-time setup: | |
2340 | */ | |
aab03e05 | 2341 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2342 | { |
0122ec5b | 2343 | unsigned long flags; |
dd41f596 IM |
2344 | int cpu = get_cpu(); |
2345 | ||
5e1576ed | 2346 | __sched_fork(clone_flags, p); |
06b83b5f | 2347 | /* |
7dc603c9 | 2348 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
2349 | * nobody will actually run it, and a signal or other external |
2350 | * event cannot wake it up and insert it on the runqueue either. | |
2351 | */ | |
7dc603c9 | 2352 | p->state = TASK_NEW; |
dd41f596 | 2353 | |
c350a04e MG |
2354 | /* |
2355 | * Make sure we do not leak PI boosting priority to the child. | |
2356 | */ | |
2357 | p->prio = current->normal_prio; | |
2358 | ||
b9dc29e7 MG |
2359 | /* |
2360 | * Revert to default priority/policy on fork if requested. | |
2361 | */ | |
2362 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2363 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2364 | p->policy = SCHED_NORMAL; |
6c697bdf | 2365 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2366 | p->rt_priority = 0; |
2367 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2368 | p->static_prio = NICE_TO_PRIO(0); | |
2369 | ||
2370 | p->prio = p->normal_prio = __normal_prio(p); | |
2371 | set_load_weight(p); | |
6c697bdf | 2372 | |
b9dc29e7 MG |
2373 | /* |
2374 | * We don't need the reset flag anymore after the fork. It has | |
2375 | * fulfilled its duty: | |
2376 | */ | |
2377 | p->sched_reset_on_fork = 0; | |
2378 | } | |
ca94c442 | 2379 | |
aab03e05 DF |
2380 | if (dl_prio(p->prio)) { |
2381 | put_cpu(); | |
2382 | return -EAGAIN; | |
2383 | } else if (rt_prio(p->prio)) { | |
2384 | p->sched_class = &rt_sched_class; | |
2385 | } else { | |
2ddbf952 | 2386 | p->sched_class = &fair_sched_class; |
aab03e05 | 2387 | } |
b29739f9 | 2388 | |
7dc603c9 | 2389 | init_entity_runnable_average(&p->se); |
cd29fe6f | 2390 | |
86951599 PZ |
2391 | /* |
2392 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2393 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2394 | * is ran before sched_fork(). | |
2395 | * | |
2396 | * Silence PROVE_RCU. | |
2397 | */ | |
0122ec5b | 2398 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e210bffd | 2399 | /* |
d1ccc66d | 2400 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
2401 | * so use __set_task_cpu(). |
2402 | */ | |
2403 | __set_task_cpu(p, cpu); | |
2404 | if (p->sched_class->task_fork) | |
2405 | p->sched_class->task_fork(p); | |
0122ec5b | 2406 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2407 | |
f6db8347 | 2408 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2409 | if (likely(sched_info_on())) |
52f17b6c | 2410 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2411 | #endif |
3ca7a440 PZ |
2412 | #if defined(CONFIG_SMP) |
2413 | p->on_cpu = 0; | |
4866cde0 | 2414 | #endif |
01028747 | 2415 | init_task_preempt_count(p); |
806c09a7 | 2416 | #ifdef CONFIG_SMP |
917b627d | 2417 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2418 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2419 | #endif |
917b627d | 2420 | |
476d139c | 2421 | put_cpu(); |
aab03e05 | 2422 | return 0; |
1da177e4 LT |
2423 | } |
2424 | ||
332ac17e DF |
2425 | unsigned long to_ratio(u64 period, u64 runtime) |
2426 | { | |
2427 | if (runtime == RUNTIME_INF) | |
2428 | return 1ULL << 20; | |
2429 | ||
2430 | /* | |
2431 | * Doing this here saves a lot of checks in all | |
2432 | * the calling paths, and returning zero seems | |
2433 | * safe for them anyway. | |
2434 | */ | |
2435 | if (period == 0) | |
2436 | return 0; | |
2437 | ||
2438 | return div64_u64(runtime << 20, period); | |
2439 | } | |
2440 | ||
2441 | #ifdef CONFIG_SMP | |
2442 | inline struct dl_bw *dl_bw_of(int i) | |
2443 | { | |
f78f5b90 PM |
2444 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2445 | "sched RCU must be held"); | |
332ac17e DF |
2446 | return &cpu_rq(i)->rd->dl_bw; |
2447 | } | |
2448 | ||
de212f18 | 2449 | static inline int dl_bw_cpus(int i) |
332ac17e | 2450 | { |
de212f18 PZ |
2451 | struct root_domain *rd = cpu_rq(i)->rd; |
2452 | int cpus = 0; | |
2453 | ||
f78f5b90 PM |
2454 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
2455 | "sched RCU must be held"); | |
de212f18 PZ |
2456 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
2457 | cpus++; | |
2458 | ||
2459 | return cpus; | |
332ac17e DF |
2460 | } |
2461 | #else | |
2462 | inline struct dl_bw *dl_bw_of(int i) | |
2463 | { | |
2464 | return &cpu_rq(i)->dl.dl_bw; | |
2465 | } | |
2466 | ||
de212f18 | 2467 | static inline int dl_bw_cpus(int i) |
332ac17e DF |
2468 | { |
2469 | return 1; | |
2470 | } | |
2471 | #endif | |
2472 | ||
332ac17e DF |
2473 | /* |
2474 | * We must be sure that accepting a new task (or allowing changing the | |
2475 | * parameters of an existing one) is consistent with the bandwidth | |
2476 | * constraints. If yes, this function also accordingly updates the currently | |
2477 | * allocated bandwidth to reflect the new situation. | |
2478 | * | |
2479 | * This function is called while holding p's rq->lock. | |
40767b0d PZ |
2480 | * |
2481 | * XXX we should delay bw change until the task's 0-lag point, see | |
2482 | * __setparam_dl(). | |
332ac17e DF |
2483 | */ |
2484 | static int dl_overflow(struct task_struct *p, int policy, | |
2485 | const struct sched_attr *attr) | |
2486 | { | |
2487 | ||
2488 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); | |
4df1638c | 2489 | u64 period = attr->sched_period ?: attr->sched_deadline; |
332ac17e DF |
2490 | u64 runtime = attr->sched_runtime; |
2491 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; | |
de212f18 | 2492 | int cpus, err = -1; |
332ac17e | 2493 | |
fec148c0 XP |
2494 | /* !deadline task may carry old deadline bandwidth */ |
2495 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) | |
332ac17e DF |
2496 | return 0; |
2497 | ||
2498 | /* | |
2499 | * Either if a task, enters, leave, or stays -deadline but changes | |
2500 | * its parameters, we may need to update accordingly the total | |
2501 | * allocated bandwidth of the container. | |
2502 | */ | |
2503 | raw_spin_lock(&dl_b->lock); | |
de212f18 | 2504 | cpus = dl_bw_cpus(task_cpu(p)); |
332ac17e DF |
2505 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
2506 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { | |
2507 | __dl_add(dl_b, new_bw); | |
2508 | err = 0; | |
2509 | } else if (dl_policy(policy) && task_has_dl_policy(p) && | |
2510 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { | |
2511 | __dl_clear(dl_b, p->dl.dl_bw); | |
2512 | __dl_add(dl_b, new_bw); | |
2513 | err = 0; | |
2514 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { | |
2515 | __dl_clear(dl_b, p->dl.dl_bw); | |
2516 | err = 0; | |
2517 | } | |
2518 | raw_spin_unlock(&dl_b->lock); | |
2519 | ||
2520 | return err; | |
2521 | } | |
2522 | ||
2523 | extern void init_dl_bw(struct dl_bw *dl_b); | |
2524 | ||
1da177e4 LT |
2525 | /* |
2526 | * wake_up_new_task - wake up a newly created task for the first time. | |
2527 | * | |
2528 | * This function will do some initial scheduler statistics housekeeping | |
2529 | * that must be done for every newly created context, then puts the task | |
2530 | * on the runqueue and wakes it. | |
2531 | */ | |
3e51e3ed | 2532 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2533 | { |
eb580751 | 2534 | struct rq_flags rf; |
dd41f596 | 2535 | struct rq *rq; |
fabf318e | 2536 | |
eb580751 | 2537 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 2538 | p->state = TASK_RUNNING; |
fabf318e PZ |
2539 | #ifdef CONFIG_SMP |
2540 | /* | |
2541 | * Fork balancing, do it here and not earlier because: | |
2542 | * - cpus_allowed can change in the fork path | |
d1ccc66d | 2543 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
2544 | * |
2545 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
2546 | * as we're not fully set-up yet. | |
fabf318e | 2547 | */ |
e210bffd | 2548 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2549 | #endif |
b7fa30c9 | 2550 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 2551 | update_rq_clock(rq); |
2b8c41da | 2552 | post_init_entity_util_avg(&p->se); |
0017d735 | 2553 | |
cd29fe6f | 2554 | activate_task(rq, p, 0); |
da0c1e65 | 2555 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2556 | trace_sched_wakeup_new(p); |
a7558e01 | 2557 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2558 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2559 | if (p->sched_class->task_woken) { |
2560 | /* | |
2561 | * Nothing relies on rq->lock after this, so its fine to | |
2562 | * drop it. | |
2563 | */ | |
d8ac8971 | 2564 | rq_unpin_lock(rq, &rf); |
efbbd05a | 2565 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2566 | rq_repin_lock(rq, &rf); |
0aaafaab | 2567 | } |
9a897c5a | 2568 | #endif |
eb580751 | 2569 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
2570 | } |
2571 | ||
e107be36 AK |
2572 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2573 | ||
1cde2930 PZ |
2574 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
2575 | ||
2ecd9d29 PZ |
2576 | void preempt_notifier_inc(void) |
2577 | { | |
2578 | static_key_slow_inc(&preempt_notifier_key); | |
2579 | } | |
2580 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2581 | ||
2582 | void preempt_notifier_dec(void) | |
2583 | { | |
2584 | static_key_slow_dec(&preempt_notifier_key); | |
2585 | } | |
2586 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2587 | ||
e107be36 | 2588 | /** |
80dd99b3 | 2589 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2590 | * @notifier: notifier struct to register |
e107be36 AK |
2591 | */ |
2592 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2593 | { | |
2ecd9d29 PZ |
2594 | if (!static_key_false(&preempt_notifier_key)) |
2595 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); | |
2596 | ||
e107be36 AK |
2597 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2598 | } | |
2599 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2600 | ||
2601 | /** | |
2602 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2603 | * @notifier: notifier struct to unregister |
e107be36 | 2604 | * |
d84525a8 | 2605 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2606 | */ |
2607 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2608 | { | |
2609 | hlist_del(¬ifier->link); | |
2610 | } | |
2611 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2612 | ||
1cde2930 | 2613 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2614 | { |
2615 | struct preempt_notifier *notifier; | |
e107be36 | 2616 | |
b67bfe0d | 2617 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2618 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2619 | } | |
2620 | ||
1cde2930 PZ |
2621 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2622 | { | |
2623 | if (static_key_false(&preempt_notifier_key)) | |
2624 | __fire_sched_in_preempt_notifiers(curr); | |
2625 | } | |
2626 | ||
e107be36 | 2627 | static void |
1cde2930 PZ |
2628 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2629 | struct task_struct *next) | |
e107be36 AK |
2630 | { |
2631 | struct preempt_notifier *notifier; | |
e107be36 | 2632 | |
b67bfe0d | 2633 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2634 | notifier->ops->sched_out(notifier, next); |
2635 | } | |
2636 | ||
1cde2930 PZ |
2637 | static __always_inline void |
2638 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2639 | struct task_struct *next) | |
2640 | { | |
2641 | if (static_key_false(&preempt_notifier_key)) | |
2642 | __fire_sched_out_preempt_notifiers(curr, next); | |
2643 | } | |
2644 | ||
6d6bc0ad | 2645 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2646 | |
1cde2930 | 2647 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2648 | { |
2649 | } | |
2650 | ||
1cde2930 | 2651 | static inline void |
e107be36 AK |
2652 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2653 | struct task_struct *next) | |
2654 | { | |
2655 | } | |
2656 | ||
6d6bc0ad | 2657 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2658 | |
4866cde0 NP |
2659 | /** |
2660 | * prepare_task_switch - prepare to switch tasks | |
2661 | * @rq: the runqueue preparing to switch | |
421cee29 | 2662 | * @prev: the current task that is being switched out |
4866cde0 NP |
2663 | * @next: the task we are going to switch to. |
2664 | * | |
2665 | * This is called with the rq lock held and interrupts off. It must | |
2666 | * be paired with a subsequent finish_task_switch after the context | |
2667 | * switch. | |
2668 | * | |
2669 | * prepare_task_switch sets up locking and calls architecture specific | |
2670 | * hooks. | |
2671 | */ | |
e107be36 AK |
2672 | static inline void |
2673 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2674 | struct task_struct *next) | |
4866cde0 | 2675 | { |
43148951 | 2676 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2677 | perf_event_task_sched_out(prev, next); |
e107be36 | 2678 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2679 | prepare_lock_switch(rq, next); |
2680 | prepare_arch_switch(next); | |
2681 | } | |
2682 | ||
1da177e4 LT |
2683 | /** |
2684 | * finish_task_switch - clean up after a task-switch | |
2685 | * @prev: the thread we just switched away from. | |
2686 | * | |
4866cde0 NP |
2687 | * finish_task_switch must be called after the context switch, paired |
2688 | * with a prepare_task_switch call before the context switch. | |
2689 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2690 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2691 | * |
2692 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2693 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2694 | * with the lock held can cause deadlocks; see schedule() for |
2695 | * details.) | |
dfa50b60 ON |
2696 | * |
2697 | * The context switch have flipped the stack from under us and restored the | |
2698 | * local variables which were saved when this task called schedule() in the | |
2699 | * past. prev == current is still correct but we need to recalculate this_rq | |
2700 | * because prev may have moved to another CPU. | |
1da177e4 | 2701 | */ |
dfa50b60 | 2702 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2703 | __releases(rq->lock) |
2704 | { | |
dfa50b60 | 2705 | struct rq *rq = this_rq(); |
1da177e4 | 2706 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2707 | long prev_state; |
1da177e4 | 2708 | |
609ca066 PZ |
2709 | /* |
2710 | * The previous task will have left us with a preempt_count of 2 | |
2711 | * because it left us after: | |
2712 | * | |
2713 | * schedule() | |
2714 | * preempt_disable(); // 1 | |
2715 | * __schedule() | |
2716 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2717 | * | |
2718 | * Also, see FORK_PREEMPT_COUNT. | |
2719 | */ | |
e2bf1c4b PZ |
2720 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2721 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2722 | current->comm, current->pid, preempt_count())) | |
2723 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2724 | |
1da177e4 LT |
2725 | rq->prev_mm = NULL; |
2726 | ||
2727 | /* | |
2728 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2729 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2730 | * schedule one last time. The schedule call will never return, and |
2731 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2732 | * |
2733 | * We must observe prev->state before clearing prev->on_cpu (in | |
2734 | * finish_lock_switch), otherwise a concurrent wakeup can get prev | |
2735 | * running on another CPU and we could rave with its RUNNING -> DEAD | |
2736 | * transition, resulting in a double drop. | |
1da177e4 | 2737 | */ |
55a101f8 | 2738 | prev_state = prev->state; |
bf9fae9f | 2739 | vtime_task_switch(prev); |
a8d757ef | 2740 | perf_event_task_sched_in(prev, current); |
4866cde0 | 2741 | finish_lock_switch(rq, prev); |
01f23e16 | 2742 | finish_arch_post_lock_switch(); |
e8fa1362 | 2743 | |
e107be36 | 2744 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2745 | if (mm) |
2746 | mmdrop(mm); | |
c394cc9f | 2747 | if (unlikely(prev_state == TASK_DEAD)) { |
e6c390f2 DF |
2748 | if (prev->sched_class->task_dead) |
2749 | prev->sched_class->task_dead(prev); | |
2750 | ||
c6fd91f0 | 2751 | /* |
2752 | * Remove function-return probe instances associated with this | |
2753 | * task and put them back on the free list. | |
9761eea8 | 2754 | */ |
c6fd91f0 | 2755 | kprobe_flush_task(prev); |
68f24b08 AL |
2756 | |
2757 | /* Task is done with its stack. */ | |
2758 | put_task_stack(prev); | |
2759 | ||
1da177e4 | 2760 | put_task_struct(prev); |
c6fd91f0 | 2761 | } |
99e5ada9 | 2762 | |
de734f89 | 2763 | tick_nohz_task_switch(); |
dfa50b60 | 2764 | return rq; |
1da177e4 LT |
2765 | } |
2766 | ||
3f029d3c GH |
2767 | #ifdef CONFIG_SMP |
2768 | ||
3f029d3c | 2769 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2770 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2771 | { |
e3fca9e7 PZ |
2772 | struct callback_head *head, *next; |
2773 | void (*func)(struct rq *rq); | |
2774 | unsigned long flags; | |
3f029d3c | 2775 | |
e3fca9e7 PZ |
2776 | raw_spin_lock_irqsave(&rq->lock, flags); |
2777 | head = rq->balance_callback; | |
2778 | rq->balance_callback = NULL; | |
2779 | while (head) { | |
2780 | func = (void (*)(struct rq *))head->func; | |
2781 | next = head->next; | |
2782 | head->next = NULL; | |
2783 | head = next; | |
3f029d3c | 2784 | |
e3fca9e7 | 2785 | func(rq); |
3f029d3c | 2786 | } |
e3fca9e7 PZ |
2787 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2788 | } | |
2789 | ||
2790 | static inline void balance_callback(struct rq *rq) | |
2791 | { | |
2792 | if (unlikely(rq->balance_callback)) | |
2793 | __balance_callback(rq); | |
3f029d3c GH |
2794 | } |
2795 | ||
2796 | #else | |
da19ab51 | 2797 | |
e3fca9e7 | 2798 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2799 | { |
1da177e4 LT |
2800 | } |
2801 | ||
3f029d3c GH |
2802 | #endif |
2803 | ||
1da177e4 LT |
2804 | /** |
2805 | * schedule_tail - first thing a freshly forked thread must call. | |
2806 | * @prev: the thread we just switched away from. | |
2807 | */ | |
722a9f92 | 2808 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2809 | __releases(rq->lock) |
2810 | { | |
1a43a14a | 2811 | struct rq *rq; |
da19ab51 | 2812 | |
609ca066 PZ |
2813 | /* |
2814 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2815 | * finish_task_switch() for details. | |
2816 | * | |
2817 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2818 | * and the preempt_enable() will end up enabling preemption (on | |
2819 | * PREEMPT_COUNT kernels). | |
2820 | */ | |
2821 | ||
dfa50b60 | 2822 | rq = finish_task_switch(prev); |
e3fca9e7 | 2823 | balance_callback(rq); |
1a43a14a | 2824 | preempt_enable(); |
70b97a7f | 2825 | |
1da177e4 | 2826 | if (current->set_child_tid) |
b488893a | 2827 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2828 | } |
2829 | ||
2830 | /* | |
dfa50b60 | 2831 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2832 | */ |
04936948 | 2833 | static __always_inline struct rq * |
70b97a7f | 2834 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 2835 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 2836 | { |
dd41f596 | 2837 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2838 | |
e107be36 | 2839 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2840 | |
dd41f596 IM |
2841 | mm = next->mm; |
2842 | oldmm = prev->active_mm; | |
9226d125 ZA |
2843 | /* |
2844 | * For paravirt, this is coupled with an exit in switch_to to | |
2845 | * combine the page table reload and the switch backend into | |
2846 | * one hypercall. | |
2847 | */ | |
224101ed | 2848 | arch_start_context_switch(prev); |
9226d125 | 2849 | |
31915ab4 | 2850 | if (!mm) { |
1da177e4 | 2851 | next->active_mm = oldmm; |
f1f10076 | 2852 | mmgrab(oldmm); |
1da177e4 LT |
2853 | enter_lazy_tlb(oldmm, next); |
2854 | } else | |
f98db601 | 2855 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2856 | |
31915ab4 | 2857 | if (!prev->mm) { |
1da177e4 | 2858 | prev->active_mm = NULL; |
1da177e4 LT |
2859 | rq->prev_mm = oldmm; |
2860 | } | |
92509b73 | 2861 | |
cb42c9a3 | 2862 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 2863 | |
3a5f5e48 IM |
2864 | /* |
2865 | * Since the runqueue lock will be released by the next | |
2866 | * task (which is an invalid locking op but in the case | |
2867 | * of the scheduler it's an obvious special-case), so we | |
2868 | * do an early lockdep release here: | |
2869 | */ | |
d8ac8971 | 2870 | rq_unpin_lock(rq, rf); |
8a25d5de | 2871 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
2872 | |
2873 | /* Here we just switch the register state and the stack. */ | |
2874 | switch_to(prev, next, prev); | |
dd41f596 | 2875 | barrier(); |
dfa50b60 ON |
2876 | |
2877 | return finish_task_switch(prev); | |
1da177e4 LT |
2878 | } |
2879 | ||
2880 | /* | |
1c3e8264 | 2881 | * nr_running and nr_context_switches: |
1da177e4 LT |
2882 | * |
2883 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2884 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2885 | */ |
2886 | unsigned long nr_running(void) | |
2887 | { | |
2888 | unsigned long i, sum = 0; | |
2889 | ||
2890 | for_each_online_cpu(i) | |
2891 | sum += cpu_rq(i)->nr_running; | |
2892 | ||
2893 | return sum; | |
f711f609 | 2894 | } |
1da177e4 | 2895 | |
2ee507c4 | 2896 | /* |
d1ccc66d | 2897 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
2898 | * |
2899 | * Caution: this function does not check that the caller has disabled | |
2900 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2901 | * race. The caller is responsible to use it correctly, for example: | |
2902 | * | |
2903 | * - from a non-preemptable section (of course) | |
2904 | * | |
2905 | * - from a thread that is bound to a single CPU | |
2906 | * | |
2907 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2908 | */ |
2909 | bool single_task_running(void) | |
2910 | { | |
00cc1633 | 2911 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2912 | } |
2913 | EXPORT_SYMBOL(single_task_running); | |
2914 | ||
1da177e4 | 2915 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2916 | { |
cc94abfc SR |
2917 | int i; |
2918 | unsigned long long sum = 0; | |
46cb4b7c | 2919 | |
0a945022 | 2920 | for_each_possible_cpu(i) |
1da177e4 | 2921 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2922 | |
1da177e4 LT |
2923 | return sum; |
2924 | } | |
483b4ee6 | 2925 | |
e33a9bba TH |
2926 | /* |
2927 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
2928 | * | |
2929 | * The idea behind IO-wait account is to account the idle time that we could | |
2930 | * have spend running if it were not for IO. That is, if we were to improve the | |
2931 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
2932 | * | |
2933 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
2934 | * idle time as IO-wait, because if the storage were faster, it could've been | |
2935 | * running and we'd not be idle. | |
2936 | * | |
2937 | * This has been extended to SMP, by doing the same for each CPU. This however | |
2938 | * is broken. | |
2939 | * | |
2940 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
2941 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
2942 | * though, if the storage were faster, both could've ran at the same time, | |
2943 | * utilising both CPUs. | |
2944 | * | |
2945 | * This means, that when looking globally, the current IO-wait accounting on | |
2946 | * SMP is a lower bound, by reason of under accounting. | |
2947 | * | |
2948 | * Worse, since the numbers are provided per CPU, they are sometimes | |
2949 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
2950 | * associated with any one particular CPU, it can wake to another CPU than it | |
2951 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
2952 | * | |
2953 | * Task CPU affinities can make all that even more 'interesting'. | |
2954 | */ | |
2955 | ||
1da177e4 LT |
2956 | unsigned long nr_iowait(void) |
2957 | { | |
2958 | unsigned long i, sum = 0; | |
483b4ee6 | 2959 | |
0a945022 | 2960 | for_each_possible_cpu(i) |
1da177e4 | 2961 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2962 | |
1da177e4 LT |
2963 | return sum; |
2964 | } | |
483b4ee6 | 2965 | |
e33a9bba TH |
2966 | /* |
2967 | * Consumers of these two interfaces, like for example the cpufreq menu | |
2968 | * governor are using nonsensical data. Boosting frequency for a CPU that has | |
2969 | * IO-wait which might not even end up running the task when it does become | |
2970 | * runnable. | |
2971 | */ | |
2972 | ||
8c215bd3 | 2973 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2974 | { |
8c215bd3 | 2975 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2976 | return atomic_read(&this->nr_iowait); |
2977 | } | |
46cb4b7c | 2978 | |
372ba8cb MG |
2979 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2980 | { | |
3289bdb4 PZ |
2981 | struct rq *rq = this_rq(); |
2982 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2983 | *load = rq->load.weight; | |
372ba8cb MG |
2984 | } |
2985 | ||
dd41f596 | 2986 | #ifdef CONFIG_SMP |
8a0be9ef | 2987 | |
46cb4b7c | 2988 | /* |
38022906 PZ |
2989 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2990 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2991 | */ |
38022906 | 2992 | void sched_exec(void) |
46cb4b7c | 2993 | { |
38022906 | 2994 | struct task_struct *p = current; |
1da177e4 | 2995 | unsigned long flags; |
0017d735 | 2996 | int dest_cpu; |
46cb4b7c | 2997 | |
8f42ced9 | 2998 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2999 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3000 | if (dest_cpu == smp_processor_id()) |
3001 | goto unlock; | |
38022906 | 3002 | |
8f42ced9 | 3003 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3004 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3005 | |
8f42ced9 PZ |
3006 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3007 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3008 | return; |
3009 | } | |
0017d735 | 3010 | unlock: |
8f42ced9 | 3011 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3012 | } |
dd41f596 | 3013 | |
1da177e4 LT |
3014 | #endif |
3015 | ||
1da177e4 | 3016 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 3017 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
3018 | |
3019 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 3020 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 3021 | |
6075620b GG |
3022 | /* |
3023 | * The function fair_sched_class.update_curr accesses the struct curr | |
3024 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
3025 | * we observe a high rate of cache misses in practice. | |
3026 | * Prefetching this data results in improved performance. | |
3027 | */ | |
3028 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3029 | { | |
3030 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3031 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3032 | #else | |
3033 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3034 | #endif | |
3035 | prefetch(curr); | |
3036 | prefetch(&curr->exec_start); | |
3037 | } | |
3038 | ||
c5f8d995 HS |
3039 | /* |
3040 | * Return accounted runtime for the task. | |
3041 | * In case the task is currently running, return the runtime plus current's | |
3042 | * pending runtime that have not been accounted yet. | |
3043 | */ | |
3044 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3045 | { | |
eb580751 | 3046 | struct rq_flags rf; |
c5f8d995 | 3047 | struct rq *rq; |
6e998916 | 3048 | u64 ns; |
c5f8d995 | 3049 | |
911b2898 PZ |
3050 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3051 | /* | |
3052 | * 64-bit doesn't need locks to atomically read a 64bit value. | |
3053 | * So we have a optimization chance when the task's delta_exec is 0. | |
3054 | * Reading ->on_cpu is racy, but this is ok. | |
3055 | * | |
d1ccc66d IM |
3056 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
3057 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 3058 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
3059 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3060 | * been accounted, so we're correct here as well. | |
911b2898 | 3061 | */ |
da0c1e65 | 3062 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3063 | return p->se.sum_exec_runtime; |
3064 | #endif | |
3065 | ||
eb580751 | 3066 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3067 | /* |
3068 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3069 | * project cycles that may never be accounted to this | |
3070 | * thread, breaking clock_gettime(). | |
3071 | */ | |
3072 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3073 | prefetch_curr_exec_start(p); |
6e998916 SG |
3074 | update_rq_clock(rq); |
3075 | p->sched_class->update_curr(rq); | |
3076 | } | |
3077 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3078 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3079 | |
3080 | return ns; | |
3081 | } | |
48f24c4d | 3082 | |
7835b98b CL |
3083 | /* |
3084 | * This function gets called by the timer code, with HZ frequency. | |
3085 | * We call it with interrupts disabled. | |
7835b98b CL |
3086 | */ |
3087 | void scheduler_tick(void) | |
3088 | { | |
7835b98b CL |
3089 | int cpu = smp_processor_id(); |
3090 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3091 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3092 | |
3093 | sched_clock_tick(); | |
dd41f596 | 3094 | |
05fa785c | 3095 | raw_spin_lock(&rq->lock); |
3e51f33f | 3096 | update_rq_clock(rq); |
fa85ae24 | 3097 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3098 | cpu_load_update_active(rq); |
3289bdb4 | 3099 | calc_global_load_tick(rq); |
05fa785c | 3100 | raw_spin_unlock(&rq->lock); |
7835b98b | 3101 | |
e9d2b064 | 3102 | perf_event_task_tick(); |
e220d2dc | 3103 | |
e418e1c2 | 3104 | #ifdef CONFIG_SMP |
6eb57e0d | 3105 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3106 | trigger_load_balance(rq); |
e418e1c2 | 3107 | #endif |
265f22a9 | 3108 | rq_last_tick_reset(rq); |
1da177e4 LT |
3109 | } |
3110 | ||
265f22a9 FW |
3111 | #ifdef CONFIG_NO_HZ_FULL |
3112 | /** | |
3113 | * scheduler_tick_max_deferment | |
3114 | * | |
3115 | * Keep at least one tick per second when a single | |
3116 | * active task is running because the scheduler doesn't | |
3117 | * yet completely support full dynticks environment. | |
3118 | * | |
3119 | * This makes sure that uptime, CFS vruntime, load | |
3120 | * balancing, etc... continue to move forward, even | |
3121 | * with a very low granularity. | |
e69f6186 YB |
3122 | * |
3123 | * Return: Maximum deferment in nanoseconds. | |
265f22a9 FW |
3124 | */ |
3125 | u64 scheduler_tick_max_deferment(void) | |
3126 | { | |
3127 | struct rq *rq = this_rq(); | |
316c1608 | 3128 | unsigned long next, now = READ_ONCE(jiffies); |
265f22a9 FW |
3129 | |
3130 | next = rq->last_sched_tick + HZ; | |
3131 | ||
3132 | if (time_before_eq(next, now)) | |
3133 | return 0; | |
3134 | ||
8fe8ff09 | 3135 | return jiffies_to_nsecs(next - now); |
1da177e4 | 3136 | } |
265f22a9 | 3137 | #endif |
1da177e4 | 3138 | |
7e49fcce SR |
3139 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3140 | defined(CONFIG_PREEMPT_TRACER)) | |
47252cfb SR |
3141 | /* |
3142 | * If the value passed in is equal to the current preempt count | |
3143 | * then we just disabled preemption. Start timing the latency. | |
3144 | */ | |
3145 | static inline void preempt_latency_start(int val) | |
3146 | { | |
3147 | if (preempt_count() == val) { | |
3148 | unsigned long ip = get_lock_parent_ip(); | |
3149 | #ifdef CONFIG_DEBUG_PREEMPT | |
3150 | current->preempt_disable_ip = ip; | |
3151 | #endif | |
3152 | trace_preempt_off(CALLER_ADDR0, ip); | |
3153 | } | |
3154 | } | |
7e49fcce | 3155 | |
edafe3a5 | 3156 | void preempt_count_add(int val) |
1da177e4 | 3157 | { |
6cd8a4bb | 3158 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3159 | /* |
3160 | * Underflow? | |
3161 | */ | |
9a11b49a IM |
3162 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3163 | return; | |
6cd8a4bb | 3164 | #endif |
bdb43806 | 3165 | __preempt_count_add(val); |
6cd8a4bb | 3166 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3167 | /* |
3168 | * Spinlock count overflowing soon? | |
3169 | */ | |
33859f7f MOS |
3170 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3171 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3172 | #endif |
47252cfb | 3173 | preempt_latency_start(val); |
1da177e4 | 3174 | } |
bdb43806 | 3175 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3176 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3177 | |
47252cfb SR |
3178 | /* |
3179 | * If the value passed in equals to the current preempt count | |
3180 | * then we just enabled preemption. Stop timing the latency. | |
3181 | */ | |
3182 | static inline void preempt_latency_stop(int val) | |
3183 | { | |
3184 | if (preempt_count() == val) | |
3185 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3186 | } | |
3187 | ||
edafe3a5 | 3188 | void preempt_count_sub(int val) |
1da177e4 | 3189 | { |
6cd8a4bb | 3190 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3191 | /* |
3192 | * Underflow? | |
3193 | */ | |
01e3eb82 | 3194 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3195 | return; |
1da177e4 LT |
3196 | /* |
3197 | * Is the spinlock portion underflowing? | |
3198 | */ | |
9a11b49a IM |
3199 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3200 | !(preempt_count() & PREEMPT_MASK))) | |
3201 | return; | |
6cd8a4bb | 3202 | #endif |
9a11b49a | 3203 | |
47252cfb | 3204 | preempt_latency_stop(val); |
bdb43806 | 3205 | __preempt_count_sub(val); |
1da177e4 | 3206 | } |
bdb43806 | 3207 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3208 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3209 | |
47252cfb SR |
3210 | #else |
3211 | static inline void preempt_latency_start(int val) { } | |
3212 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3213 | #endif |
3214 | ||
59ddbcb2 IM |
3215 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
3216 | { | |
3217 | #ifdef CONFIG_DEBUG_PREEMPT | |
3218 | return p->preempt_disable_ip; | |
3219 | #else | |
3220 | return 0; | |
3221 | #endif | |
3222 | } | |
3223 | ||
1da177e4 | 3224 | /* |
dd41f596 | 3225 | * Print scheduling while atomic bug: |
1da177e4 | 3226 | */ |
dd41f596 | 3227 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3228 | { |
d1c6d149 VN |
3229 | /* Save this before calling printk(), since that will clobber it */ |
3230 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3231 | ||
664dfa65 DJ |
3232 | if (oops_in_progress) |
3233 | return; | |
3234 | ||
3df0fc5b PZ |
3235 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3236 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3237 | |
dd41f596 | 3238 | debug_show_held_locks(prev); |
e21f5b15 | 3239 | print_modules(); |
dd41f596 IM |
3240 | if (irqs_disabled()) |
3241 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3242 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3243 | && in_atomic_preempt_off()) { | |
8f47b187 | 3244 | pr_err("Preemption disabled at:"); |
d1c6d149 | 3245 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
3246 | pr_cont("\n"); |
3247 | } | |
748c7201 DBO |
3248 | if (panic_on_warn) |
3249 | panic("scheduling while atomic\n"); | |
3250 | ||
6135fc1e | 3251 | dump_stack(); |
373d4d09 | 3252 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3253 | } |
1da177e4 | 3254 | |
dd41f596 IM |
3255 | /* |
3256 | * Various schedule()-time debugging checks and statistics: | |
3257 | */ | |
3258 | static inline void schedule_debug(struct task_struct *prev) | |
3259 | { | |
0d9e2632 | 3260 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3261 | if (task_stack_end_corrupted(prev)) |
3262 | panic("corrupted stack end detected inside scheduler\n"); | |
0d9e2632 | 3263 | #endif |
b99def8b | 3264 | |
1dc0fffc | 3265 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3266 | __schedule_bug(prev); |
1dc0fffc PZ |
3267 | preempt_count_set(PREEMPT_DISABLED); |
3268 | } | |
b3fbab05 | 3269 | rcu_sleep_check(); |
dd41f596 | 3270 | |
1da177e4 LT |
3271 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3272 | ||
ae92882e | 3273 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3274 | } |
3275 | ||
3276 | /* | |
3277 | * Pick up the highest-prio task: | |
3278 | */ | |
3279 | static inline struct task_struct * | |
d8ac8971 | 3280 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 3281 | { |
49ee5768 | 3282 | const struct sched_class *class; |
dd41f596 | 3283 | struct task_struct *p; |
1da177e4 LT |
3284 | |
3285 | /* | |
dd41f596 IM |
3286 | * Optimization: we know that if all tasks are in |
3287 | * the fair class we can call that function directly: | |
1da177e4 | 3288 | */ |
49ee5768 | 3289 | if (likely(rq->nr_running == rq->cfs.h_nr_running)) { |
d8ac8971 | 3290 | p = fair_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3291 | if (unlikely(p == RETRY_TASK)) |
3292 | goto again; | |
3293 | ||
d1ccc66d | 3294 | /* Assumes fair_sched_class->next == idle_sched_class */ |
6ccdc84b | 3295 | if (unlikely(!p)) |
d8ac8971 | 3296 | p = idle_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3297 | |
3298 | return p; | |
1da177e4 LT |
3299 | } |
3300 | ||
37e117c0 | 3301 | again: |
34f971f6 | 3302 | for_each_class(class) { |
d8ac8971 | 3303 | p = class->pick_next_task(rq, prev, rf); |
37e117c0 PZ |
3304 | if (p) { |
3305 | if (unlikely(p == RETRY_TASK)) | |
3306 | goto again; | |
dd41f596 | 3307 | return p; |
37e117c0 | 3308 | } |
dd41f596 | 3309 | } |
34f971f6 | 3310 | |
d1ccc66d IM |
3311 | /* The idle class should always have a runnable task: */ |
3312 | BUG(); | |
dd41f596 | 3313 | } |
1da177e4 | 3314 | |
dd41f596 | 3315 | /* |
c259e01a | 3316 | * __schedule() is the main scheduler function. |
edde96ea PE |
3317 | * |
3318 | * The main means of driving the scheduler and thus entering this function are: | |
3319 | * | |
3320 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3321 | * | |
3322 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3323 | * paths. For example, see arch/x86/entry_64.S. | |
3324 | * | |
3325 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3326 | * interrupt handler scheduler_tick(). | |
3327 | * | |
3328 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3329 | * task to the run-queue and that's it. | |
3330 | * | |
3331 | * Now, if the new task added to the run-queue preempts the current | |
3332 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3333 | * called on the nearest possible occasion: | |
3334 | * | |
3335 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3336 | * | |
3337 | * - in syscall or exception context, at the next outmost | |
3338 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3339 | * spin_unlock()!) | |
3340 | * | |
3341 | * - in IRQ context, return from interrupt-handler to | |
3342 | * preemptible context | |
3343 | * | |
3344 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3345 | * then at the next: | |
3346 | * | |
3347 | * - cond_resched() call | |
3348 | * - explicit schedule() call | |
3349 | * - return from syscall or exception to user-space | |
3350 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3351 | * |
b30f0e3f | 3352 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3353 | */ |
499d7955 | 3354 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3355 | { |
3356 | struct task_struct *prev, *next; | |
67ca7bde | 3357 | unsigned long *switch_count; |
d8ac8971 | 3358 | struct rq_flags rf; |
dd41f596 | 3359 | struct rq *rq; |
31656519 | 3360 | int cpu; |
dd41f596 | 3361 | |
dd41f596 IM |
3362 | cpu = smp_processor_id(); |
3363 | rq = cpu_rq(cpu); | |
dd41f596 | 3364 | prev = rq->curr; |
dd41f596 | 3365 | |
dd41f596 | 3366 | schedule_debug(prev); |
1da177e4 | 3367 | |
31656519 | 3368 | if (sched_feat(HRTICK)) |
f333fdc9 | 3369 | hrtick_clear(rq); |
8f4d37ec | 3370 | |
46a5d164 PM |
3371 | local_irq_disable(); |
3372 | rcu_note_context_switch(); | |
3373 | ||
e0acd0a6 ON |
3374 | /* |
3375 | * Make sure that signal_pending_state()->signal_pending() below | |
3376 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3377 | * done by the caller to avoid the race with signal_wake_up(). | |
3378 | */ | |
3379 | smp_mb__before_spinlock(); | |
46a5d164 | 3380 | raw_spin_lock(&rq->lock); |
d8ac8971 | 3381 | rq_pin_lock(rq, &rf); |
1da177e4 | 3382 | |
d1ccc66d IM |
3383 | /* Promote REQ to ACT */ |
3384 | rq->clock_update_flags <<= 1; | |
9edfbfed | 3385 | |
246d86b5 | 3386 | switch_count = &prev->nivcsw; |
fc13aeba | 3387 | if (!preempt && prev->state) { |
21aa9af0 | 3388 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3389 | prev->state = TASK_RUNNING; |
21aa9af0 | 3390 | } else { |
2acca55e PZ |
3391 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
3392 | prev->on_rq = 0; | |
3393 | ||
e33a9bba TH |
3394 | if (prev->in_iowait) { |
3395 | atomic_inc(&rq->nr_iowait); | |
3396 | delayacct_blkio_start(); | |
3397 | } | |
3398 | ||
21aa9af0 | 3399 | /* |
2acca55e PZ |
3400 | * If a worker went to sleep, notify and ask workqueue |
3401 | * whether it wants to wake up a task to maintain | |
3402 | * concurrency. | |
21aa9af0 TH |
3403 | */ |
3404 | if (prev->flags & PF_WQ_WORKER) { | |
3405 | struct task_struct *to_wakeup; | |
3406 | ||
9b7f6597 | 3407 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3408 | if (to_wakeup) |
d8ac8971 | 3409 | try_to_wake_up_local(to_wakeup, &rf); |
21aa9af0 | 3410 | } |
21aa9af0 | 3411 | } |
dd41f596 | 3412 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3413 | } |
3414 | ||
9edfbfed | 3415 | if (task_on_rq_queued(prev)) |
606dba2e PZ |
3416 | update_rq_clock(rq); |
3417 | ||
d8ac8971 | 3418 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 3419 | clear_tsk_need_resched(prev); |
f27dde8d | 3420 | clear_preempt_need_resched(); |
1da177e4 | 3421 | |
1da177e4 | 3422 | if (likely(prev != next)) { |
1da177e4 LT |
3423 | rq->nr_switches++; |
3424 | rq->curr = next; | |
3425 | ++*switch_count; | |
3426 | ||
c73464b1 | 3427 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
3428 | |
3429 | /* Also unlocks the rq: */ | |
3430 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 3431 | } else { |
cb42c9a3 | 3432 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
d8ac8971 | 3433 | rq_unpin_lock(rq, &rf); |
05fa785c | 3434 | raw_spin_unlock_irq(&rq->lock); |
cbce1a68 | 3435 | } |
1da177e4 | 3436 | |
e3fca9e7 | 3437 | balance_callback(rq); |
1da177e4 | 3438 | } |
c259e01a | 3439 | |
9af6528e PZ |
3440 | void __noreturn do_task_dead(void) |
3441 | { | |
3442 | /* | |
3443 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed | |
3444 | * when the following two conditions become true. | |
3445 | * - There is race condition of mmap_sem (It is acquired by | |
3446 | * exit_mm()), and | |
3447 | * - SMI occurs before setting TASK_RUNINNG. | |
3448 | * (or hypervisor of virtual machine switches to other guest) | |
3449 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD | |
3450 | * | |
3451 | * To avoid it, we have to wait for releasing tsk->pi_lock which | |
3452 | * is held by try_to_wake_up() | |
3453 | */ | |
3454 | smp_mb(); | |
3455 | raw_spin_unlock_wait(¤t->pi_lock); | |
3456 | ||
d1ccc66d | 3457 | /* Causes final put_task_struct in finish_task_switch(): */ |
9af6528e | 3458 | __set_current_state(TASK_DEAD); |
d1ccc66d IM |
3459 | |
3460 | /* Tell freezer to ignore us: */ | |
3461 | current->flags |= PF_NOFREEZE; | |
3462 | ||
9af6528e PZ |
3463 | __schedule(false); |
3464 | BUG(); | |
d1ccc66d IM |
3465 | |
3466 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 3467 | for (;;) |
d1ccc66d | 3468 | cpu_relax(); |
9af6528e PZ |
3469 | } |
3470 | ||
9c40cef2 TG |
3471 | static inline void sched_submit_work(struct task_struct *tsk) |
3472 | { | |
3c7d5184 | 3473 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3474 | return; |
3475 | /* | |
3476 | * If we are going to sleep and we have plugged IO queued, | |
3477 | * make sure to submit it to avoid deadlocks. | |
3478 | */ | |
3479 | if (blk_needs_flush_plug(tsk)) | |
3480 | blk_schedule_flush_plug(tsk); | |
3481 | } | |
3482 | ||
722a9f92 | 3483 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3484 | { |
9c40cef2 TG |
3485 | struct task_struct *tsk = current; |
3486 | ||
3487 | sched_submit_work(tsk); | |
bfd9b2b5 | 3488 | do { |
b30f0e3f | 3489 | preempt_disable(); |
fc13aeba | 3490 | __schedule(false); |
b30f0e3f | 3491 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3492 | } while (need_resched()); |
c259e01a | 3493 | } |
1da177e4 LT |
3494 | EXPORT_SYMBOL(schedule); |
3495 | ||
91d1aa43 | 3496 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3497 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3498 | { |
3499 | /* | |
3500 | * If we come here after a random call to set_need_resched(), | |
3501 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3502 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3503 | * we find a better solution. | |
7cc78f8f AL |
3504 | * |
3505 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3506 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3507 | * too frequently to make sense yet. |
20ab65e3 | 3508 | */ |
7cc78f8f | 3509 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3510 | schedule(); |
7cc78f8f | 3511 | exception_exit(prev_state); |
20ab65e3 FW |
3512 | } |
3513 | #endif | |
3514 | ||
c5491ea7 TG |
3515 | /** |
3516 | * schedule_preempt_disabled - called with preemption disabled | |
3517 | * | |
3518 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3519 | */ | |
3520 | void __sched schedule_preempt_disabled(void) | |
3521 | { | |
ba74c144 | 3522 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3523 | schedule(); |
3524 | preempt_disable(); | |
3525 | } | |
3526 | ||
06b1f808 | 3527 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3528 | { |
3529 | do { | |
47252cfb SR |
3530 | /* |
3531 | * Because the function tracer can trace preempt_count_sub() | |
3532 | * and it also uses preempt_enable/disable_notrace(), if | |
3533 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3534 | * by the function tracer will call this function again and | |
3535 | * cause infinite recursion. | |
3536 | * | |
3537 | * Preemption must be disabled here before the function | |
3538 | * tracer can trace. Break up preempt_disable() into two | |
3539 | * calls. One to disable preemption without fear of being | |
3540 | * traced. The other to still record the preemption latency, | |
3541 | * which can also be traced by the function tracer. | |
3542 | */ | |
499d7955 | 3543 | preempt_disable_notrace(); |
47252cfb | 3544 | preempt_latency_start(1); |
fc13aeba | 3545 | __schedule(true); |
47252cfb | 3546 | preempt_latency_stop(1); |
499d7955 | 3547 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3548 | |
3549 | /* | |
3550 | * Check again in case we missed a preemption opportunity | |
3551 | * between schedule and now. | |
3552 | */ | |
a18b5d01 FW |
3553 | } while (need_resched()); |
3554 | } | |
3555 | ||
1da177e4 LT |
3556 | #ifdef CONFIG_PREEMPT |
3557 | /* | |
2ed6e34f | 3558 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3559 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3560 | * occur there and call schedule directly. |
3561 | */ | |
722a9f92 | 3562 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3563 | { |
1da177e4 LT |
3564 | /* |
3565 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3566 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3567 | */ |
fbb00b56 | 3568 | if (likely(!preemptible())) |
1da177e4 LT |
3569 | return; |
3570 | ||
a18b5d01 | 3571 | preempt_schedule_common(); |
1da177e4 | 3572 | } |
376e2424 | 3573 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3574 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3575 | |
009f60e2 | 3576 | /** |
4eaca0a8 | 3577 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3578 | * |
3579 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3580 | * recursion and tracing preempt enabling caused by the tracing | |
3581 | * infrastructure itself. But as tracing can happen in areas coming | |
3582 | * from userspace or just about to enter userspace, a preempt enable | |
3583 | * can occur before user_exit() is called. This will cause the scheduler | |
3584 | * to be called when the system is still in usermode. | |
3585 | * | |
3586 | * To prevent this, the preempt_enable_notrace will use this function | |
3587 | * instead of preempt_schedule() to exit user context if needed before | |
3588 | * calling the scheduler. | |
3589 | */ | |
4eaca0a8 | 3590 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3591 | { |
3592 | enum ctx_state prev_ctx; | |
3593 | ||
3594 | if (likely(!preemptible())) | |
3595 | return; | |
3596 | ||
3597 | do { | |
47252cfb SR |
3598 | /* |
3599 | * Because the function tracer can trace preempt_count_sub() | |
3600 | * and it also uses preempt_enable/disable_notrace(), if | |
3601 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3602 | * by the function tracer will call this function again and | |
3603 | * cause infinite recursion. | |
3604 | * | |
3605 | * Preemption must be disabled here before the function | |
3606 | * tracer can trace. Break up preempt_disable() into two | |
3607 | * calls. One to disable preemption without fear of being | |
3608 | * traced. The other to still record the preemption latency, | |
3609 | * which can also be traced by the function tracer. | |
3610 | */ | |
3d8f74dd | 3611 | preempt_disable_notrace(); |
47252cfb | 3612 | preempt_latency_start(1); |
009f60e2 ON |
3613 | /* |
3614 | * Needs preempt disabled in case user_exit() is traced | |
3615 | * and the tracer calls preempt_enable_notrace() causing | |
3616 | * an infinite recursion. | |
3617 | */ | |
3618 | prev_ctx = exception_enter(); | |
fc13aeba | 3619 | __schedule(true); |
009f60e2 ON |
3620 | exception_exit(prev_ctx); |
3621 | ||
47252cfb | 3622 | preempt_latency_stop(1); |
3d8f74dd | 3623 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3624 | } while (need_resched()); |
3625 | } | |
4eaca0a8 | 3626 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3627 | |
32e475d7 | 3628 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3629 | |
3630 | /* | |
2ed6e34f | 3631 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3632 | * off of irq context. |
3633 | * Note, that this is called and return with irqs disabled. This will | |
3634 | * protect us against recursive calling from irq. | |
3635 | */ | |
722a9f92 | 3636 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3637 | { |
b22366cd | 3638 | enum ctx_state prev_state; |
6478d880 | 3639 | |
2ed6e34f | 3640 | /* Catch callers which need to be fixed */ |
f27dde8d | 3641 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3642 | |
b22366cd FW |
3643 | prev_state = exception_enter(); |
3644 | ||
3a5c359a | 3645 | do { |
3d8f74dd | 3646 | preempt_disable(); |
3a5c359a | 3647 | local_irq_enable(); |
fc13aeba | 3648 | __schedule(true); |
3a5c359a | 3649 | local_irq_disable(); |
3d8f74dd | 3650 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3651 | } while (need_resched()); |
b22366cd FW |
3652 | |
3653 | exception_exit(prev_state); | |
1da177e4 LT |
3654 | } |
3655 | ||
63859d4f | 3656 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3657 | void *key) |
1da177e4 | 3658 | { |
63859d4f | 3659 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3660 | } |
1da177e4 LT |
3661 | EXPORT_SYMBOL(default_wake_function); |
3662 | ||
b29739f9 IM |
3663 | #ifdef CONFIG_RT_MUTEXES |
3664 | ||
3665 | /* | |
3666 | * rt_mutex_setprio - set the current priority of a task | |
3667 | * @p: task | |
3668 | * @prio: prio value (kernel-internal form) | |
3669 | * | |
3670 | * This function changes the 'effective' priority of a task. It does | |
3671 | * not touch ->normal_prio like __setscheduler(). | |
3672 | * | |
c365c292 TG |
3673 | * Used by the rt_mutex code to implement priority inheritance |
3674 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3675 | */ |
36c8b586 | 3676 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3677 | { |
ff77e468 | 3678 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
83ab0aa0 | 3679 | const struct sched_class *prev_class; |
eb580751 PZ |
3680 | struct rq_flags rf; |
3681 | struct rq *rq; | |
b29739f9 | 3682 | |
aab03e05 | 3683 | BUG_ON(prio > MAX_PRIO); |
b29739f9 | 3684 | |
eb580751 | 3685 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 3686 | update_rq_clock(rq); |
b29739f9 | 3687 | |
1c4dd99b TG |
3688 | /* |
3689 | * Idle task boosting is a nono in general. There is one | |
3690 | * exception, when PREEMPT_RT and NOHZ is active: | |
3691 | * | |
3692 | * The idle task calls get_next_timer_interrupt() and holds | |
3693 | * the timer wheel base->lock on the CPU and another CPU wants | |
3694 | * to access the timer (probably to cancel it). We can safely | |
3695 | * ignore the boosting request, as the idle CPU runs this code | |
3696 | * with interrupts disabled and will complete the lock | |
3697 | * protected section without being interrupted. So there is no | |
3698 | * real need to boost. | |
3699 | */ | |
3700 | if (unlikely(p == rq->idle)) { | |
3701 | WARN_ON(p != rq->curr); | |
3702 | WARN_ON(p->pi_blocked_on); | |
3703 | goto out_unlock; | |
3704 | } | |
3705 | ||
a8027073 | 3706 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 3707 | oldprio = p->prio; |
ff77e468 PZ |
3708 | |
3709 | if (oldprio == prio) | |
3710 | queue_flag &= ~DEQUEUE_MOVE; | |
3711 | ||
83ab0aa0 | 3712 | prev_class = p->sched_class; |
da0c1e65 | 3713 | queued = task_on_rq_queued(p); |
051a1d1a | 3714 | running = task_current(rq, p); |
da0c1e65 | 3715 | if (queued) |
ff77e468 | 3716 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3717 | if (running) |
f3cd1c4e | 3718 | put_prev_task(rq, p); |
dd41f596 | 3719 | |
2d3d891d DF |
3720 | /* |
3721 | * Boosting condition are: | |
3722 | * 1. -rt task is running and holds mutex A | |
3723 | * --> -dl task blocks on mutex A | |
3724 | * | |
3725 | * 2. -dl task is running and holds mutex A | |
3726 | * --> -dl task blocks on mutex A and could preempt the | |
3727 | * running task | |
3728 | */ | |
3729 | if (dl_prio(prio)) { | |
466af29b ON |
3730 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
3731 | if (!dl_prio(p->normal_prio) || | |
3732 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3733 | p->dl.dl_boosted = 1; |
ff77e468 | 3734 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3735 | } else |
3736 | p->dl.dl_boosted = 0; | |
aab03e05 | 3737 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3738 | } else if (rt_prio(prio)) { |
3739 | if (dl_prio(oldprio)) | |
3740 | p->dl.dl_boosted = 0; | |
3741 | if (oldprio < prio) | |
ff77e468 | 3742 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3743 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3744 | } else { |
3745 | if (dl_prio(oldprio)) | |
3746 | p->dl.dl_boosted = 0; | |
746db944 BS |
3747 | if (rt_prio(oldprio)) |
3748 | p->rt.timeout = 0; | |
dd41f596 | 3749 | p->sched_class = &fair_sched_class; |
2d3d891d | 3750 | } |
dd41f596 | 3751 | |
b29739f9 IM |
3752 | p->prio = prio; |
3753 | ||
da0c1e65 | 3754 | if (queued) |
ff77e468 | 3755 | enqueue_task(rq, p, queue_flag); |
a399d233 | 3756 | if (running) |
b2bf6c31 | 3757 | set_curr_task(rq, p); |
cb469845 | 3758 | |
da7a735e | 3759 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3760 | out_unlock: |
d1ccc66d IM |
3761 | /* Avoid rq from going away on us: */ |
3762 | preempt_disable(); | |
eb580751 | 3763 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3764 | |
3765 | balance_callback(rq); | |
3766 | preempt_enable(); | |
b29739f9 | 3767 | } |
b29739f9 | 3768 | #endif |
d50dde5a | 3769 | |
36c8b586 | 3770 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3771 | { |
49bd21ef PZ |
3772 | bool queued, running; |
3773 | int old_prio, delta; | |
eb580751 | 3774 | struct rq_flags rf; |
70b97a7f | 3775 | struct rq *rq; |
1da177e4 | 3776 | |
75e45d51 | 3777 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3778 | return; |
3779 | /* | |
3780 | * We have to be careful, if called from sys_setpriority(), | |
3781 | * the task might be in the middle of scheduling on another CPU. | |
3782 | */ | |
eb580751 | 3783 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
3784 | update_rq_clock(rq); |
3785 | ||
1da177e4 LT |
3786 | /* |
3787 | * The RT priorities are set via sched_setscheduler(), but we still | |
3788 | * allow the 'normal' nice value to be set - but as expected | |
3789 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3790 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3791 | */ |
aab03e05 | 3792 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3793 | p->static_prio = NICE_TO_PRIO(nice); |
3794 | goto out_unlock; | |
3795 | } | |
da0c1e65 | 3796 | queued = task_on_rq_queued(p); |
49bd21ef | 3797 | running = task_current(rq, p); |
da0c1e65 | 3798 | if (queued) |
1de64443 | 3799 | dequeue_task(rq, p, DEQUEUE_SAVE); |
49bd21ef PZ |
3800 | if (running) |
3801 | put_prev_task(rq, p); | |
1da177e4 | 3802 | |
1da177e4 | 3803 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3804 | set_load_weight(p); |
b29739f9 IM |
3805 | old_prio = p->prio; |
3806 | p->prio = effective_prio(p); | |
3807 | delta = p->prio - old_prio; | |
1da177e4 | 3808 | |
da0c1e65 | 3809 | if (queued) { |
1de64443 | 3810 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
1da177e4 | 3811 | /* |
d5f9f942 AM |
3812 | * If the task increased its priority or is running and |
3813 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3814 | */ |
d5f9f942 | 3815 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3816 | resched_curr(rq); |
1da177e4 | 3817 | } |
49bd21ef PZ |
3818 | if (running) |
3819 | set_curr_task(rq, p); | |
1da177e4 | 3820 | out_unlock: |
eb580751 | 3821 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3822 | } |
1da177e4 LT |
3823 | EXPORT_SYMBOL(set_user_nice); |
3824 | ||
e43379f1 MM |
3825 | /* |
3826 | * can_nice - check if a task can reduce its nice value | |
3827 | * @p: task | |
3828 | * @nice: nice value | |
3829 | */ | |
36c8b586 | 3830 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3831 | { |
d1ccc66d | 3832 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 3833 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3834 | |
78d7d407 | 3835 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3836 | capable(CAP_SYS_NICE)); |
3837 | } | |
3838 | ||
1da177e4 LT |
3839 | #ifdef __ARCH_WANT_SYS_NICE |
3840 | ||
3841 | /* | |
3842 | * sys_nice - change the priority of the current process. | |
3843 | * @increment: priority increment | |
3844 | * | |
3845 | * sys_setpriority is a more generic, but much slower function that | |
3846 | * does similar things. | |
3847 | */ | |
5add95d4 | 3848 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3849 | { |
48f24c4d | 3850 | long nice, retval; |
1da177e4 LT |
3851 | |
3852 | /* | |
3853 | * Setpriority might change our priority at the same moment. | |
3854 | * We don't have to worry. Conceptually one call occurs first | |
3855 | * and we have a single winner. | |
3856 | */ | |
a9467fa3 | 3857 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3858 | nice = task_nice(current) + increment; |
1da177e4 | 3859 | |
a9467fa3 | 3860 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3861 | if (increment < 0 && !can_nice(current, nice)) |
3862 | return -EPERM; | |
3863 | ||
1da177e4 LT |
3864 | retval = security_task_setnice(current, nice); |
3865 | if (retval) | |
3866 | return retval; | |
3867 | ||
3868 | set_user_nice(current, nice); | |
3869 | return 0; | |
3870 | } | |
3871 | ||
3872 | #endif | |
3873 | ||
3874 | /** | |
3875 | * task_prio - return the priority value of a given task. | |
3876 | * @p: the task in question. | |
3877 | * | |
e69f6186 | 3878 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3879 | * RT tasks are offset by -200. Normal tasks are centered |
3880 | * around 0, value goes from -16 to +15. | |
3881 | */ | |
36c8b586 | 3882 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3883 | { |
3884 | return p->prio - MAX_RT_PRIO; | |
3885 | } | |
3886 | ||
1da177e4 | 3887 | /** |
d1ccc66d | 3888 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 3889 | * @cpu: the processor in question. |
e69f6186 YB |
3890 | * |
3891 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3892 | */ |
3893 | int idle_cpu(int cpu) | |
3894 | { | |
908a3283 TG |
3895 | struct rq *rq = cpu_rq(cpu); |
3896 | ||
3897 | if (rq->curr != rq->idle) | |
3898 | return 0; | |
3899 | ||
3900 | if (rq->nr_running) | |
3901 | return 0; | |
3902 | ||
3903 | #ifdef CONFIG_SMP | |
3904 | if (!llist_empty(&rq->wake_list)) | |
3905 | return 0; | |
3906 | #endif | |
3907 | ||
3908 | return 1; | |
1da177e4 LT |
3909 | } |
3910 | ||
1da177e4 | 3911 | /** |
d1ccc66d | 3912 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 3913 | * @cpu: the processor in question. |
e69f6186 | 3914 | * |
d1ccc66d | 3915 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 3916 | */ |
36c8b586 | 3917 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3918 | { |
3919 | return cpu_rq(cpu)->idle; | |
3920 | } | |
3921 | ||
3922 | /** | |
3923 | * find_process_by_pid - find a process with a matching PID value. | |
3924 | * @pid: the pid in question. | |
e69f6186 YB |
3925 | * |
3926 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 3927 | */ |
a9957449 | 3928 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 3929 | { |
228ebcbe | 3930 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
3931 | } |
3932 | ||
aab03e05 DF |
3933 | /* |
3934 | * This function initializes the sched_dl_entity of a newly becoming | |
3935 | * SCHED_DEADLINE task. | |
3936 | * | |
3937 | * Only the static values are considered here, the actual runtime and the | |
3938 | * absolute deadline will be properly calculated when the task is enqueued | |
3939 | * for the first time with its new policy. | |
3940 | */ | |
3941 | static void | |
3942 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) | |
3943 | { | |
3944 | struct sched_dl_entity *dl_se = &p->dl; | |
3945 | ||
aab03e05 DF |
3946 | dl_se->dl_runtime = attr->sched_runtime; |
3947 | dl_se->dl_deadline = attr->sched_deadline; | |
755378a4 | 3948 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
aab03e05 | 3949 | dl_se->flags = attr->sched_flags; |
332ac17e | 3950 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
40767b0d PZ |
3951 | |
3952 | /* | |
3953 | * Changing the parameters of a task is 'tricky' and we're not doing | |
3954 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). | |
3955 | * | |
3956 | * What we SHOULD do is delay the bandwidth release until the 0-lag | |
3957 | * point. This would include retaining the task_struct until that time | |
3958 | * and change dl_overflow() to not immediately decrement the current | |
3959 | * amount. | |
3960 | * | |
3961 | * Instead we retain the current runtime/deadline and let the new | |
3962 | * parameters take effect after the current reservation period lapses. | |
3963 | * This is safe (albeit pessimistic) because the 0-lag point is always | |
3964 | * before the current scheduling deadline. | |
3965 | * | |
3966 | * We can still have temporary overloads because we do not delay the | |
3967 | * change in bandwidth until that time; so admission control is | |
3968 | * not on the safe side. It does however guarantee tasks will never | |
3969 | * consume more than promised. | |
3970 | */ | |
aab03e05 DF |
3971 | } |
3972 | ||
c13db6b1 SR |
3973 | /* |
3974 | * sched_setparam() passes in -1 for its policy, to let the functions | |
3975 | * it calls know not to change it. | |
3976 | */ | |
3977 | #define SETPARAM_POLICY -1 | |
3978 | ||
c365c292 TG |
3979 | static void __setscheduler_params(struct task_struct *p, |
3980 | const struct sched_attr *attr) | |
1da177e4 | 3981 | { |
d50dde5a DF |
3982 | int policy = attr->sched_policy; |
3983 | ||
c13db6b1 | 3984 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
3985 | policy = p->policy; |
3986 | ||
1da177e4 | 3987 | p->policy = policy; |
d50dde5a | 3988 | |
aab03e05 DF |
3989 | if (dl_policy(policy)) |
3990 | __setparam_dl(p, attr); | |
39fd8fd2 | 3991 | else if (fair_policy(policy)) |
d50dde5a DF |
3992 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
3993 | ||
39fd8fd2 PZ |
3994 | /* |
3995 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
3996 | * !rt_policy. Always setting this ensures that things like | |
3997 | * getparam()/getattr() don't report silly values for !rt tasks. | |
3998 | */ | |
3999 | p->rt_priority = attr->sched_priority; | |
383afd09 | 4000 | p->normal_prio = normal_prio(p); |
c365c292 TG |
4001 | set_load_weight(p); |
4002 | } | |
39fd8fd2 | 4003 | |
c365c292 TG |
4004 | /* Actually do priority change: must hold pi & rq lock. */ |
4005 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 4006 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
4007 | { |
4008 | __setscheduler_params(p, attr); | |
d50dde5a | 4009 | |
383afd09 | 4010 | /* |
0782e63b TG |
4011 | * Keep a potential priority boosting if called from |
4012 | * sched_setscheduler(). | |
383afd09 | 4013 | */ |
0782e63b TG |
4014 | if (keep_boost) |
4015 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); | |
4016 | else | |
4017 | p->prio = normal_prio(p); | |
383afd09 | 4018 | |
aab03e05 DF |
4019 | if (dl_prio(p->prio)) |
4020 | p->sched_class = &dl_sched_class; | |
4021 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
4022 | p->sched_class = &rt_sched_class; |
4023 | else | |
4024 | p->sched_class = &fair_sched_class; | |
1da177e4 | 4025 | } |
aab03e05 DF |
4026 | |
4027 | static void | |
4028 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) | |
4029 | { | |
4030 | struct sched_dl_entity *dl_se = &p->dl; | |
4031 | ||
4032 | attr->sched_priority = p->rt_priority; | |
4033 | attr->sched_runtime = dl_se->dl_runtime; | |
4034 | attr->sched_deadline = dl_se->dl_deadline; | |
755378a4 | 4035 | attr->sched_period = dl_se->dl_period; |
aab03e05 DF |
4036 | attr->sched_flags = dl_se->flags; |
4037 | } | |
4038 | ||
4039 | /* | |
4040 | * This function validates the new parameters of a -deadline task. | |
4041 | * We ask for the deadline not being zero, and greater or equal | |
755378a4 | 4042 | * than the runtime, as well as the period of being zero or |
332ac17e | 4043 | * greater than deadline. Furthermore, we have to be sure that |
b0827819 JL |
4044 | * user parameters are above the internal resolution of 1us (we |
4045 | * check sched_runtime only since it is always the smaller one) and | |
4046 | * below 2^63 ns (we have to check both sched_deadline and | |
4047 | * sched_period, as the latter can be zero). | |
aab03e05 DF |
4048 | */ |
4049 | static bool | |
4050 | __checkparam_dl(const struct sched_attr *attr) | |
4051 | { | |
b0827819 JL |
4052 | /* deadline != 0 */ |
4053 | if (attr->sched_deadline == 0) | |
4054 | return false; | |
4055 | ||
4056 | /* | |
4057 | * Since we truncate DL_SCALE bits, make sure we're at least | |
4058 | * that big. | |
4059 | */ | |
4060 | if (attr->sched_runtime < (1ULL << DL_SCALE)) | |
4061 | return false; | |
4062 | ||
4063 | /* | |
4064 | * Since we use the MSB for wrap-around and sign issues, make | |
4065 | * sure it's not set (mind that period can be equal to zero). | |
4066 | */ | |
4067 | if (attr->sched_deadline & (1ULL << 63) || | |
4068 | attr->sched_period & (1ULL << 63)) | |
4069 | return false; | |
4070 | ||
4071 | /* runtime <= deadline <= period (if period != 0) */ | |
4072 | if ((attr->sched_period != 0 && | |
4073 | attr->sched_period < attr->sched_deadline) || | |
4074 | attr->sched_deadline < attr->sched_runtime) | |
4075 | return false; | |
4076 | ||
4077 | return true; | |
aab03e05 DF |
4078 | } |
4079 | ||
c69e8d9c | 4080 | /* |
d1ccc66d | 4081 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
4082 | */ |
4083 | static bool check_same_owner(struct task_struct *p) | |
4084 | { | |
4085 | const struct cred *cred = current_cred(), *pcred; | |
4086 | bool match; | |
4087 | ||
4088 | rcu_read_lock(); | |
4089 | pcred = __task_cred(p); | |
9c806aa0 EB |
4090 | match = (uid_eq(cred->euid, pcred->euid) || |
4091 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4092 | rcu_read_unlock(); |
4093 | return match; | |
4094 | } | |
4095 | ||
d1ccc66d | 4096 | static bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr) |
75381608 WL |
4097 | { |
4098 | struct sched_dl_entity *dl_se = &p->dl; | |
4099 | ||
4100 | if (dl_se->dl_runtime != attr->sched_runtime || | |
4101 | dl_se->dl_deadline != attr->sched_deadline || | |
4102 | dl_se->dl_period != attr->sched_period || | |
4103 | dl_se->flags != attr->sched_flags) | |
4104 | return true; | |
4105 | ||
4106 | return false; | |
4107 | } | |
4108 | ||
d50dde5a DF |
4109 | static int __sched_setscheduler(struct task_struct *p, |
4110 | const struct sched_attr *attr, | |
dbc7f069 | 4111 | bool user, bool pi) |
1da177e4 | 4112 | { |
383afd09 SR |
4113 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4114 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4115 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4116 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4117 | const struct sched_class *prev_class; |
eb580751 | 4118 | struct rq_flags rf; |
ca94c442 | 4119 | int reset_on_fork; |
ff77e468 | 4120 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
eb580751 | 4121 | struct rq *rq; |
1da177e4 | 4122 | |
d1ccc66d | 4123 | /* May grab non-irq protected spin_locks: */ |
66e5393a | 4124 | BUG_ON(in_interrupt()); |
1da177e4 | 4125 | recheck: |
d1ccc66d | 4126 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
4127 | if (policy < 0) { |
4128 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4129 | policy = oldpolicy = p->policy; |
ca94c442 | 4130 | } else { |
7479f3c9 | 4131 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4132 | |
20f9cd2a | 4133 | if (!valid_policy(policy)) |
ca94c442 LP |
4134 | return -EINVAL; |
4135 | } | |
4136 | ||
7479f3c9 PZ |
4137 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
4138 | return -EINVAL; | |
4139 | ||
1da177e4 LT |
4140 | /* |
4141 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4142 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4143 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4144 | */ |
0bb040a4 | 4145 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4146 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4147 | return -EINVAL; |
aab03e05 DF |
4148 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4149 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4150 | return -EINVAL; |
4151 | ||
37e4ab3f OC |
4152 | /* |
4153 | * Allow unprivileged RT tasks to decrease priority: | |
4154 | */ | |
961ccddd | 4155 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4156 | if (fair_policy(policy)) { |
d0ea0268 | 4157 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4158 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4159 | return -EPERM; |
4160 | } | |
4161 | ||
e05606d3 | 4162 | if (rt_policy(policy)) { |
a44702e8 ON |
4163 | unsigned long rlim_rtprio = |
4164 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 4165 | |
d1ccc66d | 4166 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
4167 | if (policy != p->policy && !rlim_rtprio) |
4168 | return -EPERM; | |
4169 | ||
d1ccc66d | 4170 | /* Can't increase priority: */ |
d50dde5a DF |
4171 | if (attr->sched_priority > p->rt_priority && |
4172 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4173 | return -EPERM; |
4174 | } | |
c02aa73b | 4175 | |
d44753b8 JL |
4176 | /* |
4177 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4178 | * (safest behavior); in the future we would like to allow | |
4179 | * unprivileged DL tasks to increase their relative deadline | |
4180 | * or reduce their runtime (both ways reducing utilization) | |
4181 | */ | |
4182 | if (dl_policy(policy)) | |
4183 | return -EPERM; | |
4184 | ||
dd41f596 | 4185 | /* |
c02aa73b DH |
4186 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4187 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4188 | */ |
20f9cd2a | 4189 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4190 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4191 | return -EPERM; |
4192 | } | |
5fe1d75f | 4193 | |
d1ccc66d | 4194 | /* Can't change other user's priorities: */ |
c69e8d9c | 4195 | if (!check_same_owner(p)) |
37e4ab3f | 4196 | return -EPERM; |
ca94c442 | 4197 | |
d1ccc66d | 4198 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
4199 | if (p->sched_reset_on_fork && !reset_on_fork) |
4200 | return -EPERM; | |
37e4ab3f | 4201 | } |
1da177e4 | 4202 | |
725aad24 | 4203 | if (user) { |
b0ae1981 | 4204 | retval = security_task_setscheduler(p); |
725aad24 JF |
4205 | if (retval) |
4206 | return retval; | |
4207 | } | |
4208 | ||
b29739f9 | 4209 | /* |
d1ccc66d | 4210 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 4211 | * changing the priority of the task: |
0122ec5b | 4212 | * |
25985edc | 4213 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4214 | * runqueue lock must be held. |
4215 | */ | |
eb580751 | 4216 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 4217 | update_rq_clock(rq); |
dc61b1d6 | 4218 | |
34f971f6 | 4219 | /* |
d1ccc66d | 4220 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
4221 | */ |
4222 | if (p == rq->stop) { | |
eb580751 | 4223 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4224 | return -EINVAL; |
4225 | } | |
4226 | ||
a51e9198 | 4227 | /* |
d6b1e911 TG |
4228 | * If not changing anything there's no need to proceed further, |
4229 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4230 | */ |
d50dde5a | 4231 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4232 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4233 | goto change; |
4234 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4235 | goto change; | |
75381608 | 4236 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4237 | goto change; |
d50dde5a | 4238 | |
d6b1e911 | 4239 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4240 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4241 | return 0; |
4242 | } | |
d50dde5a | 4243 | change: |
a51e9198 | 4244 | |
dc61b1d6 | 4245 | if (user) { |
332ac17e | 4246 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4247 | /* |
4248 | * Do not allow realtime tasks into groups that have no runtime | |
4249 | * assigned. | |
4250 | */ | |
4251 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4252 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4253 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4254 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4255 | return -EPERM; |
4256 | } | |
dc61b1d6 | 4257 | #endif |
332ac17e DF |
4258 | #ifdef CONFIG_SMP |
4259 | if (dl_bandwidth_enabled() && dl_policy(policy)) { | |
4260 | cpumask_t *span = rq->rd->span; | |
332ac17e DF |
4261 | |
4262 | /* | |
4263 | * Don't allow tasks with an affinity mask smaller than | |
4264 | * the entire root_domain to become SCHED_DEADLINE. We | |
4265 | * will also fail if there's no bandwidth available. | |
4266 | */ | |
e4099a5e PZ |
4267 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4268 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4269 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4270 | return -EPERM; |
4271 | } | |
4272 | } | |
4273 | #endif | |
4274 | } | |
dc61b1d6 | 4275 | |
d1ccc66d | 4276 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
4277 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
4278 | policy = oldpolicy = -1; | |
eb580751 | 4279 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4280 | goto recheck; |
4281 | } | |
332ac17e DF |
4282 | |
4283 | /* | |
4284 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4285 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4286 | * is available. | |
4287 | */ | |
e4099a5e | 4288 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
eb580751 | 4289 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4290 | return -EBUSY; |
4291 | } | |
4292 | ||
c365c292 TG |
4293 | p->sched_reset_on_fork = reset_on_fork; |
4294 | oldprio = p->prio; | |
4295 | ||
dbc7f069 PZ |
4296 | if (pi) { |
4297 | /* | |
4298 | * Take priority boosted tasks into account. If the new | |
4299 | * effective priority is unchanged, we just store the new | |
4300 | * normal parameters and do not touch the scheduler class and | |
4301 | * the runqueue. This will be done when the task deboost | |
4302 | * itself. | |
4303 | */ | |
4304 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); | |
ff77e468 PZ |
4305 | if (new_effective_prio == oldprio) |
4306 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4307 | } |
4308 | ||
da0c1e65 | 4309 | queued = task_on_rq_queued(p); |
051a1d1a | 4310 | running = task_current(rq, p); |
da0c1e65 | 4311 | if (queued) |
ff77e468 | 4312 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4313 | if (running) |
f3cd1c4e | 4314 | put_prev_task(rq, p); |
f6b53205 | 4315 | |
83ab0aa0 | 4316 | prev_class = p->sched_class; |
dbc7f069 | 4317 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4318 | |
da0c1e65 | 4319 | if (queued) { |
81a44c54 TG |
4320 | /* |
4321 | * We enqueue to tail when the priority of a task is | |
4322 | * increased (user space view). | |
4323 | */ | |
ff77e468 PZ |
4324 | if (oldprio < p->prio) |
4325 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4326 | |
ff77e468 | 4327 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4328 | } |
a399d233 | 4329 | if (running) |
b2bf6c31 | 4330 | set_curr_task(rq, p); |
cb469845 | 4331 | |
da7a735e | 4332 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
4333 | |
4334 | /* Avoid rq from going away on us: */ | |
4335 | preempt_disable(); | |
eb580751 | 4336 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4337 | |
dbc7f069 PZ |
4338 | if (pi) |
4339 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4340 | |
d1ccc66d | 4341 | /* Run balance callbacks after we've adjusted the PI chain: */ |
4c9a4bc8 PZ |
4342 | balance_callback(rq); |
4343 | preempt_enable(); | |
95e02ca9 | 4344 | |
1da177e4 LT |
4345 | return 0; |
4346 | } | |
961ccddd | 4347 | |
7479f3c9 PZ |
4348 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4349 | const struct sched_param *param, bool check) | |
4350 | { | |
4351 | struct sched_attr attr = { | |
4352 | .sched_policy = policy, | |
4353 | .sched_priority = param->sched_priority, | |
4354 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4355 | }; | |
4356 | ||
c13db6b1 SR |
4357 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4358 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4359 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4360 | policy &= ~SCHED_RESET_ON_FORK; | |
4361 | attr.sched_policy = policy; | |
4362 | } | |
4363 | ||
dbc7f069 | 4364 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4365 | } |
961ccddd RR |
4366 | /** |
4367 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4368 | * @p: the task in question. | |
4369 | * @policy: new policy. | |
4370 | * @param: structure containing the new RT priority. | |
4371 | * | |
e69f6186 YB |
4372 | * Return: 0 on success. An error code otherwise. |
4373 | * | |
961ccddd RR |
4374 | * NOTE that the task may be already dead. |
4375 | */ | |
4376 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4377 | const struct sched_param *param) |
961ccddd | 4378 | { |
7479f3c9 | 4379 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4380 | } |
1da177e4 LT |
4381 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4382 | ||
d50dde5a DF |
4383 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4384 | { | |
dbc7f069 | 4385 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4386 | } |
4387 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4388 | ||
961ccddd RR |
4389 | /** |
4390 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4391 | * @p: the task in question. | |
4392 | * @policy: new policy. | |
4393 | * @param: structure containing the new RT priority. | |
4394 | * | |
4395 | * Just like sched_setscheduler, only don't bother checking if the | |
4396 | * current context has permission. For example, this is needed in | |
4397 | * stop_machine(): we create temporary high priority worker threads, | |
4398 | * but our caller might not have that capability. | |
e69f6186 YB |
4399 | * |
4400 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4401 | */ |
4402 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4403 | const struct sched_param *param) |
961ccddd | 4404 | { |
7479f3c9 | 4405 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4406 | } |
84778472 | 4407 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4408 | |
95cdf3b7 IM |
4409 | static int |
4410 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4411 | { |
1da177e4 LT |
4412 | struct sched_param lparam; |
4413 | struct task_struct *p; | |
36c8b586 | 4414 | int retval; |
1da177e4 LT |
4415 | |
4416 | if (!param || pid < 0) | |
4417 | return -EINVAL; | |
4418 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4419 | return -EFAULT; | |
5fe1d75f ON |
4420 | |
4421 | rcu_read_lock(); | |
4422 | retval = -ESRCH; | |
1da177e4 | 4423 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4424 | if (p != NULL) |
4425 | retval = sched_setscheduler(p, policy, &lparam); | |
4426 | rcu_read_unlock(); | |
36c8b586 | 4427 | |
1da177e4 LT |
4428 | return retval; |
4429 | } | |
4430 | ||
d50dde5a DF |
4431 | /* |
4432 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4433 | */ | |
d1ccc66d | 4434 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
4435 | { |
4436 | u32 size; | |
4437 | int ret; | |
4438 | ||
4439 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4440 | return -EFAULT; | |
4441 | ||
d1ccc66d | 4442 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
4443 | memset(attr, 0, sizeof(*attr)); |
4444 | ||
4445 | ret = get_user(size, &uattr->size); | |
4446 | if (ret) | |
4447 | return ret; | |
4448 | ||
d1ccc66d IM |
4449 | /* Bail out on silly large: */ |
4450 | if (size > PAGE_SIZE) | |
d50dde5a DF |
4451 | goto err_size; |
4452 | ||
d1ccc66d IM |
4453 | /* ABI compatibility quirk: */ |
4454 | if (!size) | |
d50dde5a DF |
4455 | size = SCHED_ATTR_SIZE_VER0; |
4456 | ||
4457 | if (size < SCHED_ATTR_SIZE_VER0) | |
4458 | goto err_size; | |
4459 | ||
4460 | /* | |
4461 | * If we're handed a bigger struct than we know of, | |
4462 | * ensure all the unknown bits are 0 - i.e. new | |
4463 | * user-space does not rely on any kernel feature | |
4464 | * extensions we dont know about yet. | |
4465 | */ | |
4466 | if (size > sizeof(*attr)) { | |
4467 | unsigned char __user *addr; | |
4468 | unsigned char __user *end; | |
4469 | unsigned char val; | |
4470 | ||
4471 | addr = (void __user *)uattr + sizeof(*attr); | |
4472 | end = (void __user *)uattr + size; | |
4473 | ||
4474 | for (; addr < end; addr++) { | |
4475 | ret = get_user(val, addr); | |
4476 | if (ret) | |
4477 | return ret; | |
4478 | if (val) | |
4479 | goto err_size; | |
4480 | } | |
4481 | size = sizeof(*attr); | |
4482 | } | |
4483 | ||
4484 | ret = copy_from_user(attr, uattr, size); | |
4485 | if (ret) | |
4486 | return -EFAULT; | |
4487 | ||
4488 | /* | |
d1ccc66d | 4489 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
4490 | * to be strict and return an error on out-of-bounds values? |
4491 | */ | |
75e45d51 | 4492 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4493 | |
e78c7bca | 4494 | return 0; |
d50dde5a DF |
4495 | |
4496 | err_size: | |
4497 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4498 | return -E2BIG; |
d50dde5a DF |
4499 | } |
4500 | ||
1da177e4 LT |
4501 | /** |
4502 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4503 | * @pid: the pid in question. | |
4504 | * @policy: new policy. | |
4505 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4506 | * |
4507 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4508 | */ |
d1ccc66d | 4509 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 4510 | { |
c21761f1 JB |
4511 | if (policy < 0) |
4512 | return -EINVAL; | |
4513 | ||
1da177e4 LT |
4514 | return do_sched_setscheduler(pid, policy, param); |
4515 | } | |
4516 | ||
4517 | /** | |
4518 | * sys_sched_setparam - set/change the RT priority of a thread | |
4519 | * @pid: the pid in question. | |
4520 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4521 | * |
4522 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4523 | */ |
5add95d4 | 4524 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4525 | { |
c13db6b1 | 4526 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4527 | } |
4528 | ||
d50dde5a DF |
4529 | /** |
4530 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4531 | * @pid: the pid in question. | |
5778fccf | 4532 | * @uattr: structure containing the extended parameters. |
db66d756 | 4533 | * @flags: for future extension. |
d50dde5a | 4534 | */ |
6d35ab48 PZ |
4535 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4536 | unsigned int, flags) | |
d50dde5a DF |
4537 | { |
4538 | struct sched_attr attr; | |
4539 | struct task_struct *p; | |
4540 | int retval; | |
4541 | ||
6d35ab48 | 4542 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4543 | return -EINVAL; |
4544 | ||
143cf23d MK |
4545 | retval = sched_copy_attr(uattr, &attr); |
4546 | if (retval) | |
4547 | return retval; | |
d50dde5a | 4548 | |
b14ed2c2 | 4549 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4550 | return -EINVAL; |
d50dde5a DF |
4551 | |
4552 | rcu_read_lock(); | |
4553 | retval = -ESRCH; | |
4554 | p = find_process_by_pid(pid); | |
4555 | if (p != NULL) | |
4556 | retval = sched_setattr(p, &attr); | |
4557 | rcu_read_unlock(); | |
4558 | ||
4559 | return retval; | |
4560 | } | |
4561 | ||
1da177e4 LT |
4562 | /** |
4563 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4564 | * @pid: the pid in question. | |
e69f6186 YB |
4565 | * |
4566 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4567 | * code. | |
1da177e4 | 4568 | */ |
5add95d4 | 4569 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4570 | { |
36c8b586 | 4571 | struct task_struct *p; |
3a5c359a | 4572 | int retval; |
1da177e4 LT |
4573 | |
4574 | if (pid < 0) | |
3a5c359a | 4575 | return -EINVAL; |
1da177e4 LT |
4576 | |
4577 | retval = -ESRCH; | |
5fe85be0 | 4578 | rcu_read_lock(); |
1da177e4 LT |
4579 | p = find_process_by_pid(pid); |
4580 | if (p) { | |
4581 | retval = security_task_getscheduler(p); | |
4582 | if (!retval) | |
ca94c442 LP |
4583 | retval = p->policy |
4584 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4585 | } |
5fe85be0 | 4586 | rcu_read_unlock(); |
1da177e4 LT |
4587 | return retval; |
4588 | } | |
4589 | ||
4590 | /** | |
ca94c442 | 4591 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4592 | * @pid: the pid in question. |
4593 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4594 | * |
4595 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4596 | * code. | |
1da177e4 | 4597 | */ |
5add95d4 | 4598 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4599 | { |
ce5f7f82 | 4600 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4601 | struct task_struct *p; |
3a5c359a | 4602 | int retval; |
1da177e4 LT |
4603 | |
4604 | if (!param || pid < 0) | |
3a5c359a | 4605 | return -EINVAL; |
1da177e4 | 4606 | |
5fe85be0 | 4607 | rcu_read_lock(); |
1da177e4 LT |
4608 | p = find_process_by_pid(pid); |
4609 | retval = -ESRCH; | |
4610 | if (!p) | |
4611 | goto out_unlock; | |
4612 | ||
4613 | retval = security_task_getscheduler(p); | |
4614 | if (retval) | |
4615 | goto out_unlock; | |
4616 | ||
ce5f7f82 PZ |
4617 | if (task_has_rt_policy(p)) |
4618 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4619 | rcu_read_unlock(); |
1da177e4 LT |
4620 | |
4621 | /* | |
4622 | * This one might sleep, we cannot do it with a spinlock held ... | |
4623 | */ | |
4624 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4625 | ||
1da177e4 LT |
4626 | return retval; |
4627 | ||
4628 | out_unlock: | |
5fe85be0 | 4629 | rcu_read_unlock(); |
1da177e4 LT |
4630 | return retval; |
4631 | } | |
4632 | ||
d50dde5a DF |
4633 | static int sched_read_attr(struct sched_attr __user *uattr, |
4634 | struct sched_attr *attr, | |
4635 | unsigned int usize) | |
4636 | { | |
4637 | int ret; | |
4638 | ||
4639 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4640 | return -EFAULT; | |
4641 | ||
4642 | /* | |
4643 | * If we're handed a smaller struct than we know of, | |
4644 | * ensure all the unknown bits are 0 - i.e. old | |
4645 | * user-space does not get uncomplete information. | |
4646 | */ | |
4647 | if (usize < sizeof(*attr)) { | |
4648 | unsigned char *addr; | |
4649 | unsigned char *end; | |
4650 | ||
4651 | addr = (void *)attr + usize; | |
4652 | end = (void *)attr + sizeof(*attr); | |
4653 | ||
4654 | for (; addr < end; addr++) { | |
4655 | if (*addr) | |
22400674 | 4656 | return -EFBIG; |
d50dde5a DF |
4657 | } |
4658 | ||
4659 | attr->size = usize; | |
4660 | } | |
4661 | ||
4efbc454 | 4662 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4663 | if (ret) |
4664 | return -EFAULT; | |
4665 | ||
22400674 | 4666 | return 0; |
d50dde5a DF |
4667 | } |
4668 | ||
4669 | /** | |
aab03e05 | 4670 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4671 | * @pid: the pid in question. |
5778fccf | 4672 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4673 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4674 | * @flags: for future extension. |
d50dde5a | 4675 | */ |
6d35ab48 PZ |
4676 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4677 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4678 | { |
4679 | struct sched_attr attr = { | |
4680 | .size = sizeof(struct sched_attr), | |
4681 | }; | |
4682 | struct task_struct *p; | |
4683 | int retval; | |
4684 | ||
4685 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4686 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4687 | return -EINVAL; |
4688 | ||
4689 | rcu_read_lock(); | |
4690 | p = find_process_by_pid(pid); | |
4691 | retval = -ESRCH; | |
4692 | if (!p) | |
4693 | goto out_unlock; | |
4694 | ||
4695 | retval = security_task_getscheduler(p); | |
4696 | if (retval) | |
4697 | goto out_unlock; | |
4698 | ||
4699 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4700 | if (p->sched_reset_on_fork) |
4701 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4702 | if (task_has_dl_policy(p)) |
4703 | __getparam_dl(p, &attr); | |
4704 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4705 | attr.sched_priority = p->rt_priority; |
4706 | else | |
d0ea0268 | 4707 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4708 | |
4709 | rcu_read_unlock(); | |
4710 | ||
4711 | retval = sched_read_attr(uattr, &attr, size); | |
4712 | return retval; | |
4713 | ||
4714 | out_unlock: | |
4715 | rcu_read_unlock(); | |
4716 | return retval; | |
4717 | } | |
4718 | ||
96f874e2 | 4719 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4720 | { |
5a16f3d3 | 4721 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4722 | struct task_struct *p; |
4723 | int retval; | |
1da177e4 | 4724 | |
23f5d142 | 4725 | rcu_read_lock(); |
1da177e4 LT |
4726 | |
4727 | p = find_process_by_pid(pid); | |
4728 | if (!p) { | |
23f5d142 | 4729 | rcu_read_unlock(); |
1da177e4 LT |
4730 | return -ESRCH; |
4731 | } | |
4732 | ||
23f5d142 | 4733 | /* Prevent p going away */ |
1da177e4 | 4734 | get_task_struct(p); |
23f5d142 | 4735 | rcu_read_unlock(); |
1da177e4 | 4736 | |
14a40ffc TH |
4737 | if (p->flags & PF_NO_SETAFFINITY) { |
4738 | retval = -EINVAL; | |
4739 | goto out_put_task; | |
4740 | } | |
5a16f3d3 RR |
4741 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4742 | retval = -ENOMEM; | |
4743 | goto out_put_task; | |
4744 | } | |
4745 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4746 | retval = -ENOMEM; | |
4747 | goto out_free_cpus_allowed; | |
4748 | } | |
1da177e4 | 4749 | retval = -EPERM; |
4c44aaaf EB |
4750 | if (!check_same_owner(p)) { |
4751 | rcu_read_lock(); | |
4752 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4753 | rcu_read_unlock(); | |
16303ab2 | 4754 | goto out_free_new_mask; |
4c44aaaf EB |
4755 | } |
4756 | rcu_read_unlock(); | |
4757 | } | |
1da177e4 | 4758 | |
b0ae1981 | 4759 | retval = security_task_setscheduler(p); |
e7834f8f | 4760 | if (retval) |
16303ab2 | 4761 | goto out_free_new_mask; |
e7834f8f | 4762 | |
e4099a5e PZ |
4763 | |
4764 | cpuset_cpus_allowed(p, cpus_allowed); | |
4765 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4766 | ||
332ac17e DF |
4767 | /* |
4768 | * Since bandwidth control happens on root_domain basis, | |
4769 | * if admission test is enabled, we only admit -deadline | |
4770 | * tasks allowed to run on all the CPUs in the task's | |
4771 | * root_domain. | |
4772 | */ | |
4773 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4774 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4775 | rcu_read_lock(); | |
4776 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4777 | retval = -EBUSY; |
f1e3a093 | 4778 | rcu_read_unlock(); |
16303ab2 | 4779 | goto out_free_new_mask; |
332ac17e | 4780 | } |
f1e3a093 | 4781 | rcu_read_unlock(); |
332ac17e DF |
4782 | } |
4783 | #endif | |
49246274 | 4784 | again: |
25834c73 | 4785 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4786 | |
8707d8b8 | 4787 | if (!retval) { |
5a16f3d3 RR |
4788 | cpuset_cpus_allowed(p, cpus_allowed); |
4789 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4790 | /* |
4791 | * We must have raced with a concurrent cpuset | |
4792 | * update. Just reset the cpus_allowed to the | |
4793 | * cpuset's cpus_allowed | |
4794 | */ | |
5a16f3d3 | 4795 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4796 | goto again; |
4797 | } | |
4798 | } | |
16303ab2 | 4799 | out_free_new_mask: |
5a16f3d3 RR |
4800 | free_cpumask_var(new_mask); |
4801 | out_free_cpus_allowed: | |
4802 | free_cpumask_var(cpus_allowed); | |
4803 | out_put_task: | |
1da177e4 | 4804 | put_task_struct(p); |
1da177e4 LT |
4805 | return retval; |
4806 | } | |
4807 | ||
4808 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4809 | struct cpumask *new_mask) |
1da177e4 | 4810 | { |
96f874e2 RR |
4811 | if (len < cpumask_size()) |
4812 | cpumask_clear(new_mask); | |
4813 | else if (len > cpumask_size()) | |
4814 | len = cpumask_size(); | |
4815 | ||
1da177e4 LT |
4816 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4817 | } | |
4818 | ||
4819 | /** | |
d1ccc66d | 4820 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
4821 | * @pid: pid of the process |
4822 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4823 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
4824 | * |
4825 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4826 | */ |
5add95d4 HC |
4827 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4828 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4829 | { |
5a16f3d3 | 4830 | cpumask_var_t new_mask; |
1da177e4 LT |
4831 | int retval; |
4832 | ||
5a16f3d3 RR |
4833 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4834 | return -ENOMEM; | |
1da177e4 | 4835 | |
5a16f3d3 RR |
4836 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4837 | if (retval == 0) | |
4838 | retval = sched_setaffinity(pid, new_mask); | |
4839 | free_cpumask_var(new_mask); | |
4840 | return retval; | |
1da177e4 LT |
4841 | } |
4842 | ||
96f874e2 | 4843 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4844 | { |
36c8b586 | 4845 | struct task_struct *p; |
31605683 | 4846 | unsigned long flags; |
1da177e4 | 4847 | int retval; |
1da177e4 | 4848 | |
23f5d142 | 4849 | rcu_read_lock(); |
1da177e4 LT |
4850 | |
4851 | retval = -ESRCH; | |
4852 | p = find_process_by_pid(pid); | |
4853 | if (!p) | |
4854 | goto out_unlock; | |
4855 | ||
e7834f8f DQ |
4856 | retval = security_task_getscheduler(p); |
4857 | if (retval) | |
4858 | goto out_unlock; | |
4859 | ||
013fdb80 | 4860 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4861 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4862 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4863 | |
4864 | out_unlock: | |
23f5d142 | 4865 | rcu_read_unlock(); |
1da177e4 | 4866 | |
9531b62f | 4867 | return retval; |
1da177e4 LT |
4868 | } |
4869 | ||
4870 | /** | |
d1ccc66d | 4871 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
4872 | * @pid: pid of the process |
4873 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4874 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 4875 | * |
599b4840 ZW |
4876 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
4877 | * error code otherwise. | |
1da177e4 | 4878 | */ |
5add95d4 HC |
4879 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4880 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4881 | { |
4882 | int ret; | |
f17c8607 | 4883 | cpumask_var_t mask; |
1da177e4 | 4884 | |
84fba5ec | 4885 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4886 | return -EINVAL; |
4887 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4888 | return -EINVAL; |
4889 | ||
f17c8607 RR |
4890 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4891 | return -ENOMEM; | |
1da177e4 | 4892 | |
f17c8607 RR |
4893 | ret = sched_getaffinity(pid, mask); |
4894 | if (ret == 0) { | |
8bc037fb | 4895 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4896 | |
4897 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4898 | ret = -EFAULT; |
4899 | else | |
cd3d8031 | 4900 | ret = retlen; |
f17c8607 RR |
4901 | } |
4902 | free_cpumask_var(mask); | |
1da177e4 | 4903 | |
f17c8607 | 4904 | return ret; |
1da177e4 LT |
4905 | } |
4906 | ||
4907 | /** | |
4908 | * sys_sched_yield - yield the current processor to other threads. | |
4909 | * | |
dd41f596 IM |
4910 | * This function yields the current CPU to other tasks. If there are no |
4911 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4912 | * |
4913 | * Return: 0. | |
1da177e4 | 4914 | */ |
5add95d4 | 4915 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4916 | { |
70b97a7f | 4917 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4918 | |
ae92882e | 4919 | schedstat_inc(rq->yld_count); |
4530d7ab | 4920 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4921 | |
4922 | /* | |
4923 | * Since we are going to call schedule() anyway, there's | |
4924 | * no need to preempt or enable interrupts: | |
4925 | */ | |
4926 | __release(rq->lock); | |
8a25d5de | 4927 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4928 | do_raw_spin_unlock(&rq->lock); |
ba74c144 | 4929 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4930 | |
4931 | schedule(); | |
4932 | ||
4933 | return 0; | |
4934 | } | |
4935 | ||
35a773a0 | 4936 | #ifndef CONFIG_PREEMPT |
02b67cc3 | 4937 | int __sched _cond_resched(void) |
1da177e4 | 4938 | { |
fe32d3cd | 4939 | if (should_resched(0)) { |
a18b5d01 | 4940 | preempt_schedule_common(); |
1da177e4 LT |
4941 | return 1; |
4942 | } | |
4943 | return 0; | |
4944 | } | |
02b67cc3 | 4945 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 4946 | #endif |
1da177e4 LT |
4947 | |
4948 | /* | |
613afbf8 | 4949 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4950 | * call schedule, and on return reacquire the lock. |
4951 | * | |
41a2d6cf | 4952 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4953 | * operations here to prevent schedule() from being called twice (once via |
4954 | * spin_unlock(), once by hand). | |
4955 | */ | |
613afbf8 | 4956 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4957 | { |
fe32d3cd | 4958 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4959 | int ret = 0; |
4960 | ||
f607c668 PZ |
4961 | lockdep_assert_held(lock); |
4962 | ||
4a81e832 | 4963 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4964 | spin_unlock(lock); |
d86ee480 | 4965 | if (resched) |
a18b5d01 | 4966 | preempt_schedule_common(); |
95c354fe NP |
4967 | else |
4968 | cpu_relax(); | |
6df3cecb | 4969 | ret = 1; |
1da177e4 | 4970 | spin_lock(lock); |
1da177e4 | 4971 | } |
6df3cecb | 4972 | return ret; |
1da177e4 | 4973 | } |
613afbf8 | 4974 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4975 | |
613afbf8 | 4976 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4977 | { |
4978 | BUG_ON(!in_softirq()); | |
4979 | ||
fe32d3cd | 4980 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
98d82567 | 4981 | local_bh_enable(); |
a18b5d01 | 4982 | preempt_schedule_common(); |
1da177e4 LT |
4983 | local_bh_disable(); |
4984 | return 1; | |
4985 | } | |
4986 | return 0; | |
4987 | } | |
613afbf8 | 4988 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4989 | |
1da177e4 LT |
4990 | /** |
4991 | * yield - yield the current processor to other threads. | |
4992 | * | |
8e3fabfd PZ |
4993 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
4994 | * | |
4995 | * The scheduler is at all times free to pick the calling task as the most | |
4996 | * eligible task to run, if removing the yield() call from your code breaks | |
4997 | * it, its already broken. | |
4998 | * | |
4999 | * Typical broken usage is: | |
5000 | * | |
5001 | * while (!event) | |
d1ccc66d | 5002 | * yield(); |
8e3fabfd PZ |
5003 | * |
5004 | * where one assumes that yield() will let 'the other' process run that will | |
5005 | * make event true. If the current task is a SCHED_FIFO task that will never | |
5006 | * happen. Never use yield() as a progress guarantee!! | |
5007 | * | |
5008 | * If you want to use yield() to wait for something, use wait_event(). | |
5009 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
5010 | * If you still want to use yield(), do not! | |
1da177e4 LT |
5011 | */ |
5012 | void __sched yield(void) | |
5013 | { | |
5014 | set_current_state(TASK_RUNNING); | |
5015 | sys_sched_yield(); | |
5016 | } | |
1da177e4 LT |
5017 | EXPORT_SYMBOL(yield); |
5018 | ||
d95f4122 MG |
5019 | /** |
5020 | * yield_to - yield the current processor to another thread in | |
5021 | * your thread group, or accelerate that thread toward the | |
5022 | * processor it's on. | |
16addf95 RD |
5023 | * @p: target task |
5024 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5025 | * |
5026 | * It's the caller's job to ensure that the target task struct | |
5027 | * can't go away on us before we can do any checks. | |
5028 | * | |
e69f6186 | 5029 | * Return: |
7b270f60 PZ |
5030 | * true (>0) if we indeed boosted the target task. |
5031 | * false (0) if we failed to boost the target. | |
5032 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 5033 | */ |
fa93384f | 5034 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
5035 | { |
5036 | struct task_struct *curr = current; | |
5037 | struct rq *rq, *p_rq; | |
5038 | unsigned long flags; | |
c3c18640 | 5039 | int yielded = 0; |
d95f4122 MG |
5040 | |
5041 | local_irq_save(flags); | |
5042 | rq = this_rq(); | |
5043 | ||
5044 | again: | |
5045 | p_rq = task_rq(p); | |
7b270f60 PZ |
5046 | /* |
5047 | * If we're the only runnable task on the rq and target rq also | |
5048 | * has only one task, there's absolutely no point in yielding. | |
5049 | */ | |
5050 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
5051 | yielded = -ESRCH; | |
5052 | goto out_irq; | |
5053 | } | |
5054 | ||
d95f4122 | 5055 | double_rq_lock(rq, p_rq); |
39e24d8f | 5056 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
5057 | double_rq_unlock(rq, p_rq); |
5058 | goto again; | |
5059 | } | |
5060 | ||
5061 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 5062 | goto out_unlock; |
d95f4122 MG |
5063 | |
5064 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 5065 | goto out_unlock; |
d95f4122 MG |
5066 | |
5067 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 5068 | goto out_unlock; |
d95f4122 MG |
5069 | |
5070 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5071 | if (yielded) { |
ae92882e | 5072 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5073 | /* |
5074 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5075 | * fairness. | |
5076 | */ | |
5077 | if (preempt && rq != p_rq) | |
8875125e | 5078 | resched_curr(p_rq); |
6d1cafd8 | 5079 | } |
d95f4122 | 5080 | |
7b270f60 | 5081 | out_unlock: |
d95f4122 | 5082 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5083 | out_irq: |
d95f4122 MG |
5084 | local_irq_restore(flags); |
5085 | ||
7b270f60 | 5086 | if (yielded > 0) |
d95f4122 MG |
5087 | schedule(); |
5088 | ||
5089 | return yielded; | |
5090 | } | |
5091 | EXPORT_SYMBOL_GPL(yield_to); | |
5092 | ||
10ab5643 TH |
5093 | int io_schedule_prepare(void) |
5094 | { | |
5095 | int old_iowait = current->in_iowait; | |
5096 | ||
5097 | current->in_iowait = 1; | |
5098 | blk_schedule_flush_plug(current); | |
5099 | ||
5100 | return old_iowait; | |
5101 | } | |
5102 | ||
5103 | void io_schedule_finish(int token) | |
5104 | { | |
5105 | current->in_iowait = token; | |
5106 | } | |
5107 | ||
1da177e4 | 5108 | /* |
41a2d6cf | 5109 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5110 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5111 | */ |
1da177e4 LT |
5112 | long __sched io_schedule_timeout(long timeout) |
5113 | { | |
10ab5643 | 5114 | int token; |
1da177e4 LT |
5115 | long ret; |
5116 | ||
10ab5643 | 5117 | token = io_schedule_prepare(); |
1da177e4 | 5118 | ret = schedule_timeout(timeout); |
10ab5643 | 5119 | io_schedule_finish(token); |
9cff8ade | 5120 | |
1da177e4 LT |
5121 | return ret; |
5122 | } | |
9cff8ade | 5123 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 5124 | |
10ab5643 TH |
5125 | void io_schedule(void) |
5126 | { | |
5127 | int token; | |
5128 | ||
5129 | token = io_schedule_prepare(); | |
5130 | schedule(); | |
5131 | io_schedule_finish(token); | |
5132 | } | |
5133 | EXPORT_SYMBOL(io_schedule); | |
5134 | ||
1da177e4 LT |
5135 | /** |
5136 | * sys_sched_get_priority_max - return maximum RT priority. | |
5137 | * @policy: scheduling class. | |
5138 | * | |
e69f6186 YB |
5139 | * Return: On success, this syscall returns the maximum |
5140 | * rt_priority that can be used by a given scheduling class. | |
5141 | * On failure, a negative error code is returned. | |
1da177e4 | 5142 | */ |
5add95d4 | 5143 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5144 | { |
5145 | int ret = -EINVAL; | |
5146 | ||
5147 | switch (policy) { | |
5148 | case SCHED_FIFO: | |
5149 | case SCHED_RR: | |
5150 | ret = MAX_USER_RT_PRIO-1; | |
5151 | break; | |
aab03e05 | 5152 | case SCHED_DEADLINE: |
1da177e4 | 5153 | case SCHED_NORMAL: |
b0a9499c | 5154 | case SCHED_BATCH: |
dd41f596 | 5155 | case SCHED_IDLE: |
1da177e4 LT |
5156 | ret = 0; |
5157 | break; | |
5158 | } | |
5159 | return ret; | |
5160 | } | |
5161 | ||
5162 | /** | |
5163 | * sys_sched_get_priority_min - return minimum RT priority. | |
5164 | * @policy: scheduling class. | |
5165 | * | |
e69f6186 YB |
5166 | * Return: On success, this syscall returns the minimum |
5167 | * rt_priority that can be used by a given scheduling class. | |
5168 | * On failure, a negative error code is returned. | |
1da177e4 | 5169 | */ |
5add95d4 | 5170 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5171 | { |
5172 | int ret = -EINVAL; | |
5173 | ||
5174 | switch (policy) { | |
5175 | case SCHED_FIFO: | |
5176 | case SCHED_RR: | |
5177 | ret = 1; | |
5178 | break; | |
aab03e05 | 5179 | case SCHED_DEADLINE: |
1da177e4 | 5180 | case SCHED_NORMAL: |
b0a9499c | 5181 | case SCHED_BATCH: |
dd41f596 | 5182 | case SCHED_IDLE: |
1da177e4 LT |
5183 | ret = 0; |
5184 | } | |
5185 | return ret; | |
5186 | } | |
5187 | ||
5188 | /** | |
5189 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5190 | * @pid: pid of the process. | |
5191 | * @interval: userspace pointer to the timeslice value. | |
5192 | * | |
5193 | * this syscall writes the default timeslice value of a given process | |
5194 | * into the user-space timespec buffer. A value of '0' means infinity. | |
e69f6186 YB |
5195 | * |
5196 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5197 | * an error code. | |
1da177e4 | 5198 | */ |
17da2bd9 | 5199 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5200 | struct timespec __user *, interval) |
1da177e4 | 5201 | { |
36c8b586 | 5202 | struct task_struct *p; |
a4ec24b4 | 5203 | unsigned int time_slice; |
eb580751 PZ |
5204 | struct rq_flags rf; |
5205 | struct timespec t; | |
dba091b9 | 5206 | struct rq *rq; |
3a5c359a | 5207 | int retval; |
1da177e4 LT |
5208 | |
5209 | if (pid < 0) | |
3a5c359a | 5210 | return -EINVAL; |
1da177e4 LT |
5211 | |
5212 | retval = -ESRCH; | |
1a551ae7 | 5213 | rcu_read_lock(); |
1da177e4 LT |
5214 | p = find_process_by_pid(pid); |
5215 | if (!p) | |
5216 | goto out_unlock; | |
5217 | ||
5218 | retval = security_task_getscheduler(p); | |
5219 | if (retval) | |
5220 | goto out_unlock; | |
5221 | ||
eb580751 | 5222 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5223 | time_slice = 0; |
5224 | if (p->sched_class->get_rr_interval) | |
5225 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5226 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5227 | |
1a551ae7 | 5228 | rcu_read_unlock(); |
a4ec24b4 | 5229 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5230 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5231 | return retval; |
3a5c359a | 5232 | |
1da177e4 | 5233 | out_unlock: |
1a551ae7 | 5234 | rcu_read_unlock(); |
1da177e4 LT |
5235 | return retval; |
5236 | } | |
5237 | ||
7c731e0a | 5238 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5239 | |
82a1fcb9 | 5240 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5241 | { |
1da177e4 | 5242 | unsigned long free = 0; |
4e79752c | 5243 | int ppid; |
1f8a7633 | 5244 | unsigned long state = p->state; |
1da177e4 | 5245 | |
c930b2c0 IM |
5246 | /* Make sure the string lines up properly with the number of task states: */ |
5247 | BUILD_BUG_ON(sizeof(TASK_STATE_TO_CHAR_STR)-1 != ilog2(TASK_STATE_MAX)+1); | |
5248 | ||
38200502 TH |
5249 | if (!try_get_task_stack(p)) |
5250 | return; | |
1f8a7633 TH |
5251 | if (state) |
5252 | state = __ffs(state) + 1; | |
28d0686c | 5253 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5254 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
1da177e4 | 5255 | if (state == TASK_RUNNING) |
3df0fc5b | 5256 | printk(KERN_CONT " running task "); |
1da177e4 | 5257 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 5258 | free = stack_not_used(p); |
1da177e4 | 5259 | #endif |
a90e984c | 5260 | ppid = 0; |
4e79752c | 5261 | rcu_read_lock(); |
a90e984c ON |
5262 | if (pid_alive(p)) |
5263 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5264 | rcu_read_unlock(); |
3df0fc5b | 5265 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5266 | task_pid_nr(p), ppid, |
aa47b7e0 | 5267 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5268 | |
3d1cb205 | 5269 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5270 | show_stack(p, NULL); |
38200502 | 5271 | put_task_stack(p); |
1da177e4 LT |
5272 | } |
5273 | ||
e59e2ae2 | 5274 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5275 | { |
36c8b586 | 5276 | struct task_struct *g, *p; |
1da177e4 | 5277 | |
4bd77321 | 5278 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5279 | printk(KERN_INFO |
5280 | " task PC stack pid father\n"); | |
1da177e4 | 5281 | #else |
3df0fc5b PZ |
5282 | printk(KERN_INFO |
5283 | " task PC stack pid father\n"); | |
1da177e4 | 5284 | #endif |
510f5acc | 5285 | rcu_read_lock(); |
5d07f420 | 5286 | for_each_process_thread(g, p) { |
1da177e4 LT |
5287 | /* |
5288 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5289 | * console might take a lot of time: |
57675cb9 AR |
5290 | * Also, reset softlockup watchdogs on all CPUs, because |
5291 | * another CPU might be blocked waiting for us to process | |
5292 | * an IPI. | |
1da177e4 LT |
5293 | */ |
5294 | touch_nmi_watchdog(); | |
57675cb9 | 5295 | touch_all_softlockup_watchdogs(); |
39bc89fd | 5296 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5297 | sched_show_task(p); |
5d07f420 | 5298 | } |
1da177e4 | 5299 | |
dd41f596 | 5300 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5301 | if (!state_filter) |
5302 | sysrq_sched_debug_show(); | |
dd41f596 | 5303 | #endif |
510f5acc | 5304 | rcu_read_unlock(); |
e59e2ae2 IM |
5305 | /* |
5306 | * Only show locks if all tasks are dumped: | |
5307 | */ | |
93335a21 | 5308 | if (!state_filter) |
e59e2ae2 | 5309 | debug_show_all_locks(); |
1da177e4 LT |
5310 | } |
5311 | ||
0db0628d | 5312 | void init_idle_bootup_task(struct task_struct *idle) |
1df21055 | 5313 | { |
dd41f596 | 5314 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5315 | } |
5316 | ||
f340c0d1 IM |
5317 | /** |
5318 | * init_idle - set up an idle thread for a given CPU | |
5319 | * @idle: task in question | |
d1ccc66d | 5320 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
5321 | * |
5322 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5323 | * flag, to make booting more robust. | |
5324 | */ | |
0db0628d | 5325 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5326 | { |
70b97a7f | 5327 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5328 | unsigned long flags; |
5329 | ||
25834c73 PZ |
5330 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5331 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5332 | |
5e1576ed | 5333 | __sched_fork(0, idle); |
06b83b5f | 5334 | idle->state = TASK_RUNNING; |
dd41f596 | 5335 | idle->se.exec_start = sched_clock(); |
c1de45ca | 5336 | idle->flags |= PF_IDLE; |
dd41f596 | 5337 | |
e1b77c92 MR |
5338 | kasan_unpoison_task_stack(idle); |
5339 | ||
de9b8f5d PZ |
5340 | #ifdef CONFIG_SMP |
5341 | /* | |
5342 | * Its possible that init_idle() gets called multiple times on a task, | |
5343 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5344 | * | |
5345 | * And since this is boot we can forgo the serialization. | |
5346 | */ | |
5347 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5348 | #endif | |
6506cf6c PZ |
5349 | /* |
5350 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 5351 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
5352 | * lockdep check in task_group() will fail. |
5353 | * | |
5354 | * Similar case to sched_fork(). / Alternatively we could | |
5355 | * use task_rq_lock() here and obtain the other rq->lock. | |
5356 | * | |
5357 | * Silence PROVE_RCU | |
5358 | */ | |
5359 | rcu_read_lock(); | |
dd41f596 | 5360 | __set_task_cpu(idle, cpu); |
6506cf6c | 5361 | rcu_read_unlock(); |
1da177e4 | 5362 | |
1da177e4 | 5363 | rq->curr = rq->idle = idle; |
da0c1e65 | 5364 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5365 | #ifdef CONFIG_SMP |
3ca7a440 | 5366 | idle->on_cpu = 1; |
4866cde0 | 5367 | #endif |
25834c73 PZ |
5368 | raw_spin_unlock(&rq->lock); |
5369 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5370 | |
5371 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5372 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5373 | |
dd41f596 IM |
5374 | /* |
5375 | * The idle tasks have their own, simple scheduling class: | |
5376 | */ | |
5377 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5378 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5379 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5380 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5381 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5382 | #endif | |
19978ca6 IM |
5383 | } |
5384 | ||
f82f8042 JL |
5385 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5386 | const struct cpumask *trial) | |
5387 | { | |
5388 | int ret = 1, trial_cpus; | |
5389 | struct dl_bw *cur_dl_b; | |
5390 | unsigned long flags; | |
5391 | ||
bb2bc55a MG |
5392 | if (!cpumask_weight(cur)) |
5393 | return ret; | |
5394 | ||
75e23e49 | 5395 | rcu_read_lock_sched(); |
f82f8042 JL |
5396 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
5397 | trial_cpus = cpumask_weight(trial); | |
5398 | ||
5399 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); | |
5400 | if (cur_dl_b->bw != -1 && | |
5401 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) | |
5402 | ret = 0; | |
5403 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); | |
75e23e49 | 5404 | rcu_read_unlock_sched(); |
f82f8042 JL |
5405 | |
5406 | return ret; | |
5407 | } | |
5408 | ||
7f51412a JL |
5409 | int task_can_attach(struct task_struct *p, |
5410 | const struct cpumask *cs_cpus_allowed) | |
5411 | { | |
5412 | int ret = 0; | |
5413 | ||
5414 | /* | |
5415 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 5416 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
5417 | * affinity and isolating such threads by their set of |
5418 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5419 | * applicable for such threads. This prevents checking for | |
5420 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5421 | * before cpus_allowed may be changed. | |
5422 | */ | |
5423 | if (p->flags & PF_NO_SETAFFINITY) { | |
5424 | ret = -EINVAL; | |
5425 | goto out; | |
5426 | } | |
5427 | ||
5428 | #ifdef CONFIG_SMP | |
5429 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, | |
5430 | cs_cpus_allowed)) { | |
5431 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, | |
5432 | cs_cpus_allowed); | |
75e23e49 | 5433 | struct dl_bw *dl_b; |
7f51412a JL |
5434 | bool overflow; |
5435 | int cpus; | |
5436 | unsigned long flags; | |
5437 | ||
75e23e49 JL |
5438 | rcu_read_lock_sched(); |
5439 | dl_b = dl_bw_of(dest_cpu); | |
7f51412a JL |
5440 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5441 | cpus = dl_bw_cpus(dest_cpu); | |
5442 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); | |
5443 | if (overflow) | |
5444 | ret = -EBUSY; | |
5445 | else { | |
5446 | /* | |
5447 | * We reserve space for this task in the destination | |
5448 | * root_domain, as we can't fail after this point. | |
5449 | * We will free resources in the source root_domain | |
5450 | * later on (see set_cpus_allowed_dl()). | |
5451 | */ | |
5452 | __dl_add(dl_b, p->dl.dl_bw); | |
5453 | } | |
5454 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
75e23e49 | 5455 | rcu_read_unlock_sched(); |
7f51412a JL |
5456 | |
5457 | } | |
5458 | #endif | |
5459 | out: | |
5460 | return ret; | |
5461 | } | |
5462 | ||
1da177e4 | 5463 | #ifdef CONFIG_SMP |
1da177e4 | 5464 | |
f2cb1360 | 5465 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 5466 | |
e6628d5b MG |
5467 | #ifdef CONFIG_NUMA_BALANCING |
5468 | /* Migrate current task p to target_cpu */ | |
5469 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5470 | { | |
5471 | struct migration_arg arg = { p, target_cpu }; | |
5472 | int curr_cpu = task_cpu(p); | |
5473 | ||
5474 | if (curr_cpu == target_cpu) | |
5475 | return 0; | |
5476 | ||
0c98d344 | 5477 | if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed)) |
e6628d5b MG |
5478 | return -EINVAL; |
5479 | ||
5480 | /* TODO: This is not properly updating schedstats */ | |
5481 | ||
286549dc | 5482 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5483 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5484 | } | |
0ec8aa00 PZ |
5485 | |
5486 | /* | |
5487 | * Requeue a task on a given node and accurately track the number of NUMA | |
5488 | * tasks on the runqueues | |
5489 | */ | |
5490 | void sched_setnuma(struct task_struct *p, int nid) | |
5491 | { | |
da0c1e65 | 5492 | bool queued, running; |
eb580751 PZ |
5493 | struct rq_flags rf; |
5494 | struct rq *rq; | |
0ec8aa00 | 5495 | |
eb580751 | 5496 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5497 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5498 | running = task_current(rq, p); |
5499 | ||
da0c1e65 | 5500 | if (queued) |
1de64443 | 5501 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5502 | if (running) |
f3cd1c4e | 5503 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5504 | |
5505 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 5506 | |
da0c1e65 | 5507 | if (queued) |
1de64443 | 5508 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
a399d233 | 5509 | if (running) |
b2bf6c31 | 5510 | set_curr_task(rq, p); |
eb580751 | 5511 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5512 | } |
5cc389bc | 5513 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5514 | |
1da177e4 | 5515 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5516 | /* |
d1ccc66d | 5517 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 5518 | * offline. |
054b9108 | 5519 | */ |
48c5ccae | 5520 | void idle_task_exit(void) |
1da177e4 | 5521 | { |
48c5ccae | 5522 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5523 | |
48c5ccae | 5524 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5525 | |
a53efe5f | 5526 | if (mm != &init_mm) { |
f98db601 | 5527 | switch_mm_irqs_off(mm, &init_mm, current); |
a53efe5f MS |
5528 | finish_arch_post_lock_switch(); |
5529 | } | |
48c5ccae | 5530 | mmdrop(mm); |
1da177e4 LT |
5531 | } |
5532 | ||
5533 | /* | |
5d180232 PZ |
5534 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5535 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
5536 | * nr_active count is stable. We need to take the teardown thread which |
5537 | * is calling this into account, so we hand in adjust = 1 to the load | |
5538 | * calculation. | |
5d180232 PZ |
5539 | * |
5540 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5541 | */ |
5d180232 | 5542 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5543 | { |
d60585c5 | 5544 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
5545 | if (delta) |
5546 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5547 | } |
5548 | ||
3f1d2a31 PZ |
5549 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5550 | { | |
5551 | } | |
5552 | ||
5553 | static const struct sched_class fake_sched_class = { | |
5554 | .put_prev_task = put_prev_task_fake, | |
5555 | }; | |
5556 | ||
5557 | static struct task_struct fake_task = { | |
5558 | /* | |
5559 | * Avoid pull_{rt,dl}_task() | |
5560 | */ | |
5561 | .prio = MAX_PRIO + 1, | |
5562 | .sched_class = &fake_sched_class, | |
5563 | }; | |
5564 | ||
48f24c4d | 5565 | /* |
48c5ccae PZ |
5566 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5567 | * try_to_wake_up()->select_task_rq(). | |
5568 | * | |
5569 | * Called with rq->lock held even though we'er in stop_machine() and | |
5570 | * there's no concurrency possible, we hold the required locks anyway | |
5571 | * because of lock validation efforts. | |
1da177e4 | 5572 | */ |
5e16bbc2 | 5573 | static void migrate_tasks(struct rq *dead_rq) |
1da177e4 | 5574 | { |
5e16bbc2 | 5575 | struct rq *rq = dead_rq; |
48c5ccae | 5576 | struct task_struct *next, *stop = rq->stop; |
8cb68b34 | 5577 | struct rq_flags rf; |
48c5ccae | 5578 | int dest_cpu; |
1da177e4 LT |
5579 | |
5580 | /* | |
48c5ccae PZ |
5581 | * Fudge the rq selection such that the below task selection loop |
5582 | * doesn't get stuck on the currently eligible stop task. | |
5583 | * | |
5584 | * We're currently inside stop_machine() and the rq is either stuck | |
5585 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5586 | * either way we should never end up calling schedule() until we're | |
5587 | * done here. | |
1da177e4 | 5588 | */ |
48c5ccae | 5589 | rq->stop = NULL; |
48f24c4d | 5590 | |
77bd3970 FW |
5591 | /* |
5592 | * put_prev_task() and pick_next_task() sched | |
5593 | * class method both need to have an up-to-date | |
5594 | * value of rq->clock[_task] | |
5595 | */ | |
8cb68b34 | 5596 | rq_pin_lock(rq, &rf); |
77bd3970 | 5597 | update_rq_clock(rq); |
8cb68b34 | 5598 | rq_unpin_lock(rq, &rf); |
77bd3970 | 5599 | |
5e16bbc2 | 5600 | for (;;) { |
48c5ccae PZ |
5601 | /* |
5602 | * There's this thread running, bail when that's the only | |
d1ccc66d | 5603 | * remaining thread: |
48c5ccae PZ |
5604 | */ |
5605 | if (rq->nr_running == 1) | |
dd41f596 | 5606 | break; |
48c5ccae | 5607 | |
cbce1a68 | 5608 | /* |
d1ccc66d | 5609 | * pick_next_task() assumes pinned rq->lock: |
cbce1a68 | 5610 | */ |
8cb68b34 | 5611 | rq_repin_lock(rq, &rf); |
d8ac8971 | 5612 | next = pick_next_task(rq, &fake_task, &rf); |
48c5ccae | 5613 | BUG_ON(!next); |
79c53799 | 5614 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5615 | |
5473e0cc WL |
5616 | /* |
5617 | * Rules for changing task_struct::cpus_allowed are holding | |
5618 | * both pi_lock and rq->lock, such that holding either | |
5619 | * stabilizes the mask. | |
5620 | * | |
5621 | * Drop rq->lock is not quite as disastrous as it usually is | |
5622 | * because !cpu_active at this point, which means load-balance | |
5623 | * will not interfere. Also, stop-machine. | |
5624 | */ | |
d8ac8971 | 5625 | rq_unpin_lock(rq, &rf); |
5473e0cc WL |
5626 | raw_spin_unlock(&rq->lock); |
5627 | raw_spin_lock(&next->pi_lock); | |
5628 | raw_spin_lock(&rq->lock); | |
5629 | ||
5630 | /* | |
5631 | * Since we're inside stop-machine, _nothing_ should have | |
5632 | * changed the task, WARN if weird stuff happened, because in | |
5633 | * that case the above rq->lock drop is a fail too. | |
5634 | */ | |
5635 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5636 | raw_spin_unlock(&next->pi_lock); | |
5637 | continue; | |
5638 | } | |
5639 | ||
48c5ccae | 5640 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5641 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
48c5ccae | 5642 | |
5e16bbc2 PZ |
5643 | rq = __migrate_task(rq, next, dest_cpu); |
5644 | if (rq != dead_rq) { | |
5645 | raw_spin_unlock(&rq->lock); | |
5646 | rq = dead_rq; | |
5647 | raw_spin_lock(&rq->lock); | |
5648 | } | |
5473e0cc | 5649 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5650 | } |
dce48a84 | 5651 | |
48c5ccae | 5652 | rq->stop = stop; |
dce48a84 | 5653 | } |
1da177e4 LT |
5654 | #endif /* CONFIG_HOTPLUG_CPU */ |
5655 | ||
f2cb1360 | 5656 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
5657 | { |
5658 | if (!rq->online) { | |
5659 | const struct sched_class *class; | |
5660 | ||
c6c4927b | 5661 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5662 | rq->online = 1; |
5663 | ||
5664 | for_each_class(class) { | |
5665 | if (class->rq_online) | |
5666 | class->rq_online(rq); | |
5667 | } | |
5668 | } | |
5669 | } | |
5670 | ||
f2cb1360 | 5671 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
5672 | { |
5673 | if (rq->online) { | |
5674 | const struct sched_class *class; | |
5675 | ||
5676 | for_each_class(class) { | |
5677 | if (class->rq_offline) | |
5678 | class->rq_offline(rq); | |
5679 | } | |
5680 | ||
c6c4927b | 5681 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5682 | rq->online = 0; |
5683 | } | |
5684 | } | |
5685 | ||
9cf7243d | 5686 | static void set_cpu_rq_start_time(unsigned int cpu) |
1da177e4 | 5687 | { |
969c7921 | 5688 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5689 | |
a803f026 CM |
5690 | rq->age_stamp = sched_clock_cpu(cpu); |
5691 | } | |
5692 | ||
d1ccc66d IM |
5693 | /* |
5694 | * used to mark begin/end of suspend/resume: | |
5695 | */ | |
5696 | static int num_cpus_frozen; | |
d35be8ba | 5697 | |
1da177e4 | 5698 | /* |
3a101d05 TH |
5699 | * Update cpusets according to cpu_active mask. If cpusets are |
5700 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
5701 | * around partition_sched_domains(). | |
d35be8ba SB |
5702 | * |
5703 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
5704 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 5705 | */ |
40190a78 | 5706 | static void cpuset_cpu_active(void) |
e761b772 | 5707 | { |
40190a78 | 5708 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
5709 | /* |
5710 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
5711 | * resume sequence. As long as this is not the last online | |
5712 | * operation in the resume sequence, just build a single sched | |
5713 | * domain, ignoring cpusets. | |
5714 | */ | |
5715 | num_cpus_frozen--; | |
5716 | if (likely(num_cpus_frozen)) { | |
5717 | partition_sched_domains(1, NULL, NULL); | |
135fb3e1 | 5718 | return; |
d35be8ba | 5719 | } |
d35be8ba SB |
5720 | /* |
5721 | * This is the last CPU online operation. So fall through and | |
5722 | * restore the original sched domains by considering the | |
5723 | * cpuset configurations. | |
5724 | */ | |
3a101d05 | 5725 | } |
135fb3e1 | 5726 | cpuset_update_active_cpus(true); |
3a101d05 | 5727 | } |
e761b772 | 5728 | |
40190a78 | 5729 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 5730 | { |
3c18d447 | 5731 | unsigned long flags; |
3c18d447 | 5732 | struct dl_bw *dl_b; |
533445c6 OS |
5733 | bool overflow; |
5734 | int cpus; | |
3c18d447 | 5735 | |
40190a78 | 5736 | if (!cpuhp_tasks_frozen) { |
533445c6 OS |
5737 | rcu_read_lock_sched(); |
5738 | dl_b = dl_bw_of(cpu); | |
3c18d447 | 5739 | |
533445c6 OS |
5740 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
5741 | cpus = dl_bw_cpus(cpu); | |
5742 | overflow = __dl_overflow(dl_b, cpus, 0, 0); | |
5743 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); | |
3c18d447 | 5744 | |
533445c6 | 5745 | rcu_read_unlock_sched(); |
3c18d447 | 5746 | |
533445c6 | 5747 | if (overflow) |
135fb3e1 | 5748 | return -EBUSY; |
7ddf96b0 | 5749 | cpuset_update_active_cpus(false); |
135fb3e1 | 5750 | } else { |
d35be8ba SB |
5751 | num_cpus_frozen++; |
5752 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 5753 | } |
135fb3e1 | 5754 | return 0; |
e761b772 | 5755 | } |
e761b772 | 5756 | |
40190a78 | 5757 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 5758 | { |
7d976699 TG |
5759 | struct rq *rq = cpu_rq(cpu); |
5760 | unsigned long flags; | |
5761 | ||
40190a78 | 5762 | set_cpu_active(cpu, true); |
135fb3e1 | 5763 | |
40190a78 | 5764 | if (sched_smp_initialized) { |
135fb3e1 | 5765 | sched_domains_numa_masks_set(cpu); |
40190a78 | 5766 | cpuset_cpu_active(); |
e761b772 | 5767 | } |
7d976699 TG |
5768 | |
5769 | /* | |
5770 | * Put the rq online, if not already. This happens: | |
5771 | * | |
5772 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 5773 | * after all CPUs have been brought up. |
7d976699 TG |
5774 | * |
5775 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
5776 | * domains. | |
5777 | */ | |
5778 | raw_spin_lock_irqsave(&rq->lock, flags); | |
5779 | if (rq->rd) { | |
5780 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5781 | set_rq_online(rq); | |
5782 | } | |
5783 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
5784 | ||
5785 | update_max_interval(); | |
5786 | ||
40190a78 | 5787 | return 0; |
135fb3e1 TG |
5788 | } |
5789 | ||
40190a78 | 5790 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 5791 | { |
135fb3e1 TG |
5792 | int ret; |
5793 | ||
40190a78 | 5794 | set_cpu_active(cpu, false); |
b2454caa PZ |
5795 | /* |
5796 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
5797 | * users of this state to go away such that all new such users will | |
5798 | * observe it. | |
5799 | * | |
5800 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
5801 | * not imply sync_sched(), so wait for both. | |
5802 | * | |
5803 | * Do sync before park smpboot threads to take care the rcu boost case. | |
5804 | */ | |
5805 | if (IS_ENABLED(CONFIG_PREEMPT)) | |
5806 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
5807 | else | |
5808 | synchronize_rcu(); | |
40190a78 TG |
5809 | |
5810 | if (!sched_smp_initialized) | |
5811 | return 0; | |
5812 | ||
5813 | ret = cpuset_cpu_inactive(cpu); | |
5814 | if (ret) { | |
5815 | set_cpu_active(cpu, true); | |
5816 | return ret; | |
135fb3e1 | 5817 | } |
40190a78 TG |
5818 | sched_domains_numa_masks_clear(cpu); |
5819 | return 0; | |
135fb3e1 TG |
5820 | } |
5821 | ||
94baf7a5 TG |
5822 | static void sched_rq_cpu_starting(unsigned int cpu) |
5823 | { | |
5824 | struct rq *rq = cpu_rq(cpu); | |
5825 | ||
5826 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
5827 | update_max_interval(); |
5828 | } | |
5829 | ||
135fb3e1 TG |
5830 | int sched_cpu_starting(unsigned int cpu) |
5831 | { | |
5832 | set_cpu_rq_start_time(cpu); | |
94baf7a5 | 5833 | sched_rq_cpu_starting(cpu); |
135fb3e1 | 5834 | return 0; |
e761b772 | 5835 | } |
e761b772 | 5836 | |
f2785ddb TG |
5837 | #ifdef CONFIG_HOTPLUG_CPU |
5838 | int sched_cpu_dying(unsigned int cpu) | |
5839 | { | |
5840 | struct rq *rq = cpu_rq(cpu); | |
5841 | unsigned long flags; | |
5842 | ||
5843 | /* Handle pending wakeups and then migrate everything off */ | |
5844 | sched_ttwu_pending(); | |
5845 | raw_spin_lock_irqsave(&rq->lock, flags); | |
5846 | if (rq->rd) { | |
5847 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5848 | set_rq_offline(rq); | |
5849 | } | |
5850 | migrate_tasks(rq); | |
5851 | BUG_ON(rq->nr_running != 1); | |
5852 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
5853 | calc_load_migrate(rq); | |
5854 | update_max_interval(); | |
20a5c8cc | 5855 | nohz_balance_exit_idle(cpu); |
e5ef27d0 | 5856 | hrtick_clear(rq); |
f2785ddb TG |
5857 | return 0; |
5858 | } | |
5859 | #endif | |
5860 | ||
1b568f0a PZ |
5861 | #ifdef CONFIG_SCHED_SMT |
5862 | DEFINE_STATIC_KEY_FALSE(sched_smt_present); | |
5863 | ||
5864 | static void sched_init_smt(void) | |
5865 | { | |
5866 | /* | |
5867 | * We've enumerated all CPUs and will assume that if any CPU | |
5868 | * has SMT siblings, CPU0 will too. | |
5869 | */ | |
5870 | if (cpumask_weight(cpu_smt_mask(0)) > 1) | |
5871 | static_branch_enable(&sched_smt_present); | |
5872 | } | |
5873 | #else | |
5874 | static inline void sched_init_smt(void) { } | |
5875 | #endif | |
5876 | ||
1da177e4 LT |
5877 | void __init sched_init_smp(void) |
5878 | { | |
dcc30a35 RR |
5879 | cpumask_var_t non_isolated_cpus; |
5880 | ||
5881 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 5882 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 5883 | |
cb83b629 PZ |
5884 | sched_init_numa(); |
5885 | ||
6acce3ef PZ |
5886 | /* |
5887 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 5888 | * CPU masks are stable and all blatant races in the below code cannot |
6acce3ef PZ |
5889 | * happen. |
5890 | */ | |
712555ee | 5891 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 5892 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
5893 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
5894 | if (cpumask_empty(non_isolated_cpus)) | |
5895 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 5896 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 5897 | |
5c1e1767 | 5898 | /* Move init over to a non-isolated CPU */ |
dcc30a35 | 5899 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 5900 | BUG(); |
19978ca6 | 5901 | sched_init_granularity(); |
dcc30a35 | 5902 | free_cpumask_var(non_isolated_cpus); |
4212823f | 5903 | |
0e3900e6 | 5904 | init_sched_rt_class(); |
1baca4ce | 5905 | init_sched_dl_class(); |
1b568f0a PZ |
5906 | |
5907 | sched_init_smt(); | |
9881b024 | 5908 | sched_clock_init_late(); |
1b568f0a | 5909 | |
e26fbffd | 5910 | sched_smp_initialized = true; |
1da177e4 | 5911 | } |
e26fbffd TG |
5912 | |
5913 | static int __init migration_init(void) | |
5914 | { | |
94baf7a5 | 5915 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 5916 | return 0; |
1da177e4 | 5917 | } |
e26fbffd TG |
5918 | early_initcall(migration_init); |
5919 | ||
1da177e4 LT |
5920 | #else |
5921 | void __init sched_init_smp(void) | |
5922 | { | |
19978ca6 | 5923 | sched_init_granularity(); |
9881b024 | 5924 | sched_clock_init_late(); |
1da177e4 LT |
5925 | } |
5926 | #endif /* CONFIG_SMP */ | |
5927 | ||
5928 | int in_sched_functions(unsigned long addr) | |
5929 | { | |
1da177e4 LT |
5930 | return in_lock_functions(addr) || |
5931 | (addr >= (unsigned long)__sched_text_start | |
5932 | && addr < (unsigned long)__sched_text_end); | |
5933 | } | |
5934 | ||
029632fb | 5935 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
5936 | /* |
5937 | * Default task group. | |
5938 | * Every task in system belongs to this group at bootup. | |
5939 | */ | |
029632fb | 5940 | struct task_group root_task_group; |
35cf4e50 | 5941 | LIST_HEAD(task_groups); |
b0367629 WL |
5942 | |
5943 | /* Cacheline aligned slab cache for task_group */ | |
5944 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 5945 | #endif |
6f505b16 | 5946 | |
e6252c3e | 5947 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 5948 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 5949 | |
9dcb8b68 LT |
5950 | #define WAIT_TABLE_BITS 8 |
5951 | #define WAIT_TABLE_SIZE (1 << WAIT_TABLE_BITS) | |
5952 | static wait_queue_head_t bit_wait_table[WAIT_TABLE_SIZE] __cacheline_aligned; | |
5953 | ||
5954 | wait_queue_head_t *bit_waitqueue(void *word, int bit) | |
5955 | { | |
5956 | const int shift = BITS_PER_LONG == 32 ? 5 : 6; | |
5957 | unsigned long val = (unsigned long)word << shift | bit; | |
5958 | ||
5959 | return bit_wait_table + hash_long(val, WAIT_TABLE_BITS); | |
5960 | } | |
5961 | EXPORT_SYMBOL(bit_waitqueue); | |
5962 | ||
1da177e4 LT |
5963 | void __init sched_init(void) |
5964 | { | |
dd41f596 | 5965 | int i, j; |
434d53b0 MT |
5966 | unsigned long alloc_size = 0, ptr; |
5967 | ||
9881b024 PZ |
5968 | sched_clock_init(); |
5969 | ||
9dcb8b68 LT |
5970 | for (i = 0; i < WAIT_TABLE_SIZE; i++) |
5971 | init_waitqueue_head(bit_wait_table + i); | |
5972 | ||
434d53b0 MT |
5973 | #ifdef CONFIG_FAIR_GROUP_SCHED |
5974 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5975 | #endif | |
5976 | #ifdef CONFIG_RT_GROUP_SCHED | |
5977 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5978 | #endif | |
434d53b0 | 5979 | if (alloc_size) { |
36b7b6d4 | 5980 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
5981 | |
5982 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 5983 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
5984 | ptr += nr_cpu_ids * sizeof(void **); |
5985 | ||
07e06b01 | 5986 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 5987 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 5988 | |
6d6bc0ad | 5989 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 5990 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 5991 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
5992 | ptr += nr_cpu_ids * sizeof(void **); |
5993 | ||
07e06b01 | 5994 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
5995 | ptr += nr_cpu_ids * sizeof(void **); |
5996 | ||
6d6bc0ad | 5997 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 5998 | } |
df7c8e84 | 5999 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
6000 | for_each_possible_cpu(i) { |
6001 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
6002 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
6003 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
6004 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 6005 | } |
b74e6278 | 6006 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 6007 | |
d1ccc66d IM |
6008 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
6009 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 6010 | |
57d885fe GH |
6011 | #ifdef CONFIG_SMP |
6012 | init_defrootdomain(); | |
6013 | #endif | |
6014 | ||
d0b27fa7 | 6015 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6016 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6017 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6018 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6019 | |
7c941438 | 6020 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
6021 | task_group_cache = KMEM_CACHE(task_group, 0); |
6022 | ||
07e06b01 YZ |
6023 | list_add(&root_task_group.list, &task_groups); |
6024 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6025 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6026 | autogroup_init(&init_task); |
7c941438 | 6027 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6028 | |
0a945022 | 6029 | for_each_possible_cpu(i) { |
70b97a7f | 6030 | struct rq *rq; |
1da177e4 LT |
6031 | |
6032 | rq = cpu_rq(i); | |
05fa785c | 6033 | raw_spin_lock_init(&rq->lock); |
7897986b | 6034 | rq->nr_running = 0; |
dce48a84 TG |
6035 | rq->calc_load_active = 0; |
6036 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6037 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
6038 | init_rt_rq(&rq->rt); |
6039 | init_dl_rq(&rq->dl); | |
dd41f596 | 6040 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 6041 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 6042 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 6043 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 6044 | /* |
d1ccc66d | 6045 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
6046 | * |
6047 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
6048 | * gets 100% of the CPU resources in the system. This overall |
6049 | * system CPU resource is divided among the tasks of | |
07e06b01 | 6050 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6051 | * based on each entity's (task or task-group's) weight |
6052 | * (se->load.weight). | |
6053 | * | |
07e06b01 | 6054 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 6055 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 6056 | * then A0's share of the CPU resource is: |
354d60c2 | 6057 | * |
0d905bca | 6058 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6059 | * |
07e06b01 YZ |
6060 | * We achieve this by letting root_task_group's tasks sit |
6061 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6062 | */ |
ab84d31e | 6063 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6064 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6065 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6066 | ||
6067 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6068 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6069 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6070 | #endif |
1da177e4 | 6071 | |
dd41f596 IM |
6072 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6073 | rq->cpu_load[j] = 0; | |
fdf3e95d | 6074 | |
1da177e4 | 6075 | #ifdef CONFIG_SMP |
41c7ce9a | 6076 | rq->sd = NULL; |
57d885fe | 6077 | rq->rd = NULL; |
ca6d75e6 | 6078 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 6079 | rq->balance_callback = NULL; |
1da177e4 | 6080 | rq->active_balance = 0; |
dd41f596 | 6081 | rq->next_balance = jiffies; |
1da177e4 | 6082 | rq->push_cpu = 0; |
0a2966b4 | 6083 | rq->cpu = i; |
1f11eb6a | 6084 | rq->online = 0; |
eae0c9df MG |
6085 | rq->idle_stamp = 0; |
6086 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6087 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6088 | |
6089 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6090 | ||
dc938520 | 6091 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6092 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 6093 | rq->last_load_update_tick = jiffies; |
1c792db7 | 6094 | rq->nohz_flags = 0; |
83cd4fe2 | 6095 | #endif |
265f22a9 FW |
6096 | #ifdef CONFIG_NO_HZ_FULL |
6097 | rq->last_sched_tick = 0; | |
6098 | #endif | |
9fd81dd5 | 6099 | #endif /* CONFIG_SMP */ |
8f4d37ec | 6100 | init_rq_hrtick(rq); |
1da177e4 | 6101 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6102 | } |
6103 | ||
2dd73a4f | 6104 | set_load_weight(&init_task); |
b50f60ce | 6105 | |
1da177e4 LT |
6106 | /* |
6107 | * The boot idle thread does lazy MMU switching as well: | |
6108 | */ | |
f1f10076 | 6109 | mmgrab(&init_mm); |
1da177e4 LT |
6110 | enter_lazy_tlb(&init_mm, current); |
6111 | ||
6112 | /* | |
6113 | * Make us the idle thread. Technically, schedule() should not be | |
6114 | * called from this thread, however somewhere below it might be, | |
6115 | * but because we are the idle thread, we just pick up running again | |
6116 | * when this runqueue becomes "idle". | |
6117 | */ | |
6118 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6119 | |
6120 | calc_load_update = jiffies + LOAD_FREQ; | |
6121 | ||
bf4d83f6 | 6122 | #ifdef CONFIG_SMP |
4cb98839 | 6123 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
bdddd296 RR |
6124 | /* May be allocated at isolcpus cmdline parse time */ |
6125 | if (cpu_isolated_map == NULL) | |
6126 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
29d5e047 | 6127 | idle_thread_set_boot_cpu(); |
9cf7243d | 6128 | set_cpu_rq_start_time(smp_processor_id()); |
029632fb PZ |
6129 | #endif |
6130 | init_sched_fair_class(); | |
6a7b3dc3 | 6131 | |
4698f88c JP |
6132 | init_schedstats(); |
6133 | ||
6892b75e | 6134 | scheduler_running = 1; |
1da177e4 LT |
6135 | } |
6136 | ||
d902db1e | 6137 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6138 | static inline int preempt_count_equals(int preempt_offset) |
6139 | { | |
da7142e2 | 6140 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 6141 | |
4ba8216c | 6142 | return (nested == preempt_offset); |
e4aafea2 FW |
6143 | } |
6144 | ||
d894837f | 6145 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6146 | { |
8eb23b9f PZ |
6147 | /* |
6148 | * Blocking primitives will set (and therefore destroy) current->state, | |
6149 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
6150 | * otherwise we will destroy state. | |
6151 | */ | |
00845eb9 | 6152 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
6153 | "do not call blocking ops when !TASK_RUNNING; " |
6154 | "state=%lx set at [<%p>] %pS\n", | |
6155 | current->state, | |
6156 | (void *)current->task_state_change, | |
00845eb9 | 6157 | (void *)current->task_state_change); |
8eb23b9f | 6158 | |
3427445a PZ |
6159 | ___might_sleep(file, line, preempt_offset); |
6160 | } | |
6161 | EXPORT_SYMBOL(__might_sleep); | |
6162 | ||
6163 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 6164 | { |
d1ccc66d IM |
6165 | /* Ratelimiting timestamp: */ |
6166 | static unsigned long prev_jiffy; | |
6167 | ||
d1c6d149 | 6168 | unsigned long preempt_disable_ip; |
1da177e4 | 6169 | |
d1ccc66d IM |
6170 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
6171 | rcu_sleep_check(); | |
6172 | ||
db273be2 TG |
6173 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
6174 | !is_idle_task(current)) || | |
e4aafea2 | 6175 | system_state != SYSTEM_RUNNING || oops_in_progress) |
aef745fc IM |
6176 | return; |
6177 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6178 | return; | |
6179 | prev_jiffy = jiffies; | |
6180 | ||
d1ccc66d | 6181 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
6182 | preempt_disable_ip = get_preempt_disable_ip(current); |
6183 | ||
3df0fc5b PZ |
6184 | printk(KERN_ERR |
6185 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6186 | file, line); | |
6187 | printk(KERN_ERR | |
6188 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6189 | in_atomic(), irqs_disabled(), | |
6190 | current->pid, current->comm); | |
aef745fc | 6191 | |
a8b686b3 ES |
6192 | if (task_stack_end_corrupted(current)) |
6193 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
6194 | ||
aef745fc IM |
6195 | debug_show_held_locks(current); |
6196 | if (irqs_disabled()) | |
6197 | print_irqtrace_events(current); | |
d1c6d149 VN |
6198 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
6199 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 6200 | pr_err("Preemption disabled at:"); |
d1c6d149 | 6201 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
6202 | pr_cont("\n"); |
6203 | } | |
aef745fc | 6204 | dump_stack(); |
f0b22e39 | 6205 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 6206 | } |
3427445a | 6207 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
6208 | #endif |
6209 | ||
6210 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 6211 | void normalize_rt_tasks(void) |
3a5e4dc1 | 6212 | { |
dbc7f069 | 6213 | struct task_struct *g, *p; |
d50dde5a DF |
6214 | struct sched_attr attr = { |
6215 | .sched_policy = SCHED_NORMAL, | |
6216 | }; | |
1da177e4 | 6217 | |
3472eaa1 | 6218 | read_lock(&tasklist_lock); |
5d07f420 | 6219 | for_each_process_thread(g, p) { |
178be793 IM |
6220 | /* |
6221 | * Only normalize user tasks: | |
6222 | */ | |
3472eaa1 | 6223 | if (p->flags & PF_KTHREAD) |
178be793 IM |
6224 | continue; |
6225 | ||
4fa8d299 JP |
6226 | p->se.exec_start = 0; |
6227 | schedstat_set(p->se.statistics.wait_start, 0); | |
6228 | schedstat_set(p->se.statistics.sleep_start, 0); | |
6229 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 6230 | |
aab03e05 | 6231 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
6232 | /* |
6233 | * Renice negative nice level userspace | |
6234 | * tasks back to 0: | |
6235 | */ | |
3472eaa1 | 6236 | if (task_nice(p) < 0) |
dd41f596 | 6237 | set_user_nice(p, 0); |
1da177e4 | 6238 | continue; |
dd41f596 | 6239 | } |
1da177e4 | 6240 | |
dbc7f069 | 6241 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 6242 | } |
3472eaa1 | 6243 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6244 | } |
6245 | ||
6246 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6247 | |
67fc4e0c | 6248 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6249 | /* |
67fc4e0c | 6250 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6251 | * |
6252 | * They can only be called when the whole system has been | |
6253 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6254 | * activity can take place. Using them for anything else would | |
6255 | * be a serious bug, and as a result, they aren't even visible | |
6256 | * under any other configuration. | |
6257 | */ | |
6258 | ||
6259 | /** | |
d1ccc66d | 6260 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
6261 | * @cpu: the processor in question. |
6262 | * | |
6263 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
6264 | * |
6265 | * Return: The current task for @cpu. | |
1df5c10a | 6266 | */ |
36c8b586 | 6267 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6268 | { |
6269 | return cpu_curr(cpu); | |
6270 | } | |
6271 | ||
67fc4e0c JW |
6272 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6273 | ||
6274 | #ifdef CONFIG_IA64 | |
1df5c10a | 6275 | /** |
d1ccc66d | 6276 | * set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
6277 | * @cpu: the processor in question. |
6278 | * @p: the task pointer to set. | |
6279 | * | |
6280 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 6281 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 6282 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
6283 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6284 | * and caller must save the original value of the current task (see | |
6285 | * curr_task() above) and restore that value before reenabling interrupts and | |
6286 | * re-starting the system. | |
6287 | * | |
6288 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6289 | */ | |
a458ae2e | 6290 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6291 | { |
6292 | cpu_curr(cpu) = p; | |
6293 | } | |
6294 | ||
6295 | #endif | |
29f59db3 | 6296 | |
7c941438 | 6297 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6298 | /* task_group_lock serializes the addition/removal of task groups */ |
6299 | static DEFINE_SPINLOCK(task_group_lock); | |
6300 | ||
2f5177f0 | 6301 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
6302 | { |
6303 | free_fair_sched_group(tg); | |
6304 | free_rt_sched_group(tg); | |
e9aa1dd1 | 6305 | autogroup_free(tg); |
b0367629 | 6306 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
6307 | } |
6308 | ||
6309 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 6310 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
6311 | { |
6312 | struct task_group *tg; | |
bccbe08a | 6313 | |
b0367629 | 6314 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
6315 | if (!tg) |
6316 | return ERR_PTR(-ENOMEM); | |
6317 | ||
ec7dc8ac | 6318 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
6319 | goto err; |
6320 | ||
ec7dc8ac | 6321 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
6322 | goto err; |
6323 | ||
ace783b9 LZ |
6324 | return tg; |
6325 | ||
6326 | err: | |
2f5177f0 | 6327 | sched_free_group(tg); |
ace783b9 LZ |
6328 | return ERR_PTR(-ENOMEM); |
6329 | } | |
6330 | ||
6331 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
6332 | { | |
6333 | unsigned long flags; | |
6334 | ||
8ed36996 | 6335 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 6336 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 6337 | |
d1ccc66d IM |
6338 | /* Root should already exist: */ |
6339 | WARN_ON(!parent); | |
f473aa5e PZ |
6340 | |
6341 | tg->parent = parent; | |
f473aa5e | 6342 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 6343 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 6344 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
6345 | |
6346 | online_fair_sched_group(tg); | |
29f59db3 SV |
6347 | } |
6348 | ||
9b5b7751 | 6349 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 6350 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 6351 | { |
d1ccc66d | 6352 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 6353 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
6354 | } |
6355 | ||
4cf86d77 | 6356 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 6357 | { |
d1ccc66d | 6358 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 6359 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
6360 | } |
6361 | ||
6362 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 6363 | { |
8ed36996 | 6364 | unsigned long flags; |
29f59db3 | 6365 | |
d1ccc66d | 6366 | /* End participation in shares distribution: */ |
6fe1f348 | 6367 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
6368 | |
6369 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 6370 | list_del_rcu(&tg->list); |
f473aa5e | 6371 | list_del_rcu(&tg->siblings); |
8ed36996 | 6372 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6373 | } |
6374 | ||
ea86cb4b | 6375 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 6376 | { |
8323f26c | 6377 | struct task_group *tg; |
29f59db3 | 6378 | |
f7b8a47d KT |
6379 | /* |
6380 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
6381 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
6382 | * to prevent lockdep warnings. | |
6383 | */ | |
6384 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
6385 | struct task_group, css); |
6386 | tg = autogroup_task_group(tsk, tg); | |
6387 | tsk->sched_task_group = tg; | |
6388 | ||
810b3817 | 6389 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
6390 | if (tsk->sched_class->task_change_group) |
6391 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 6392 | else |
810b3817 | 6393 | #endif |
b2b5ce02 | 6394 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
6395 | } |
6396 | ||
6397 | /* | |
6398 | * Change task's runqueue when it moves between groups. | |
6399 | * | |
6400 | * The caller of this function should have put the task in its new group by | |
6401 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
6402 | * its new group. | |
6403 | */ | |
6404 | void sched_move_task(struct task_struct *tsk) | |
6405 | { | |
6406 | int queued, running; | |
6407 | struct rq_flags rf; | |
6408 | struct rq *rq; | |
6409 | ||
6410 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 6411 | update_rq_clock(rq); |
ea86cb4b VG |
6412 | |
6413 | running = task_current(rq, tsk); | |
6414 | queued = task_on_rq_queued(tsk); | |
6415 | ||
6416 | if (queued) | |
6417 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); | |
bb3bac2c | 6418 | if (running) |
ea86cb4b VG |
6419 | put_prev_task(rq, tsk); |
6420 | ||
6421 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 6422 | |
da0c1e65 | 6423 | if (queued) |
ff77e468 | 6424 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
bb3bac2c | 6425 | if (running) |
b2bf6c31 | 6426 | set_curr_task(rq, tsk); |
29f59db3 | 6427 | |
eb580751 | 6428 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 6429 | } |
7c941438 | 6430 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 6431 | |
a790de99 PT |
6432 | #ifdef CONFIG_RT_GROUP_SCHED |
6433 | /* | |
6434 | * Ensure that the real time constraints are schedulable. | |
6435 | */ | |
6436 | static DEFINE_MUTEX(rt_constraints_mutex); | |
9f0c1e56 | 6437 | |
9a7e0b18 PZ |
6438 | /* Must be called with tasklist_lock held */ |
6439 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 6440 | { |
9a7e0b18 | 6441 | struct task_struct *g, *p; |
b40b2e8e | 6442 | |
1fe89e1b PZ |
6443 | /* |
6444 | * Autogroups do not have RT tasks; see autogroup_create(). | |
6445 | */ | |
6446 | if (task_group_is_autogroup(tg)) | |
6447 | return 0; | |
6448 | ||
5d07f420 | 6449 | for_each_process_thread(g, p) { |
8651c658 | 6450 | if (rt_task(p) && task_group(p) == tg) |
9a7e0b18 | 6451 | return 1; |
5d07f420 | 6452 | } |
b40b2e8e | 6453 | |
9a7e0b18 PZ |
6454 | return 0; |
6455 | } | |
b40b2e8e | 6456 | |
9a7e0b18 PZ |
6457 | struct rt_schedulable_data { |
6458 | struct task_group *tg; | |
6459 | u64 rt_period; | |
6460 | u64 rt_runtime; | |
6461 | }; | |
b40b2e8e | 6462 | |
a790de99 | 6463 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
9a7e0b18 PZ |
6464 | { |
6465 | struct rt_schedulable_data *d = data; | |
6466 | struct task_group *child; | |
6467 | unsigned long total, sum = 0; | |
6468 | u64 period, runtime; | |
b40b2e8e | 6469 | |
9a7e0b18 PZ |
6470 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
6471 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 6472 | |
9a7e0b18 PZ |
6473 | if (tg == d->tg) { |
6474 | period = d->rt_period; | |
6475 | runtime = d->rt_runtime; | |
b40b2e8e | 6476 | } |
b40b2e8e | 6477 | |
4653f803 PZ |
6478 | /* |
6479 | * Cannot have more runtime than the period. | |
6480 | */ | |
6481 | if (runtime > period && runtime != RUNTIME_INF) | |
6482 | return -EINVAL; | |
6f505b16 | 6483 | |
4653f803 PZ |
6484 | /* |
6485 | * Ensure we don't starve existing RT tasks. | |
6486 | */ | |
9a7e0b18 PZ |
6487 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
6488 | return -EBUSY; | |
6f505b16 | 6489 | |
9a7e0b18 | 6490 | total = to_ratio(period, runtime); |
6f505b16 | 6491 | |
4653f803 PZ |
6492 | /* |
6493 | * Nobody can have more than the global setting allows. | |
6494 | */ | |
6495 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
6496 | return -EINVAL; | |
6f505b16 | 6497 | |
4653f803 PZ |
6498 | /* |
6499 | * The sum of our children's runtime should not exceed our own. | |
6500 | */ | |
9a7e0b18 PZ |
6501 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
6502 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
6503 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 6504 | |
9a7e0b18 PZ |
6505 | if (child == d->tg) { |
6506 | period = d->rt_period; | |
6507 | runtime = d->rt_runtime; | |
6508 | } | |
6f505b16 | 6509 | |
9a7e0b18 | 6510 | sum += to_ratio(period, runtime); |
9f0c1e56 | 6511 | } |
6f505b16 | 6512 | |
9a7e0b18 PZ |
6513 | if (sum > total) |
6514 | return -EINVAL; | |
6515 | ||
6516 | return 0; | |
6f505b16 PZ |
6517 | } |
6518 | ||
9a7e0b18 | 6519 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 6520 | { |
8277434e PT |
6521 | int ret; |
6522 | ||
9a7e0b18 PZ |
6523 | struct rt_schedulable_data data = { |
6524 | .tg = tg, | |
6525 | .rt_period = period, | |
6526 | .rt_runtime = runtime, | |
6527 | }; | |
6528 | ||
8277434e PT |
6529 | rcu_read_lock(); |
6530 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); | |
6531 | rcu_read_unlock(); | |
6532 | ||
6533 | return ret; | |
521f1a24 DG |
6534 | } |
6535 | ||
ab84d31e | 6536 | static int tg_set_rt_bandwidth(struct task_group *tg, |
d0b27fa7 | 6537 | u64 rt_period, u64 rt_runtime) |
6f505b16 | 6538 | { |
ac086bc2 | 6539 | int i, err = 0; |
9f0c1e56 | 6540 | |
2636ed5f PZ |
6541 | /* |
6542 | * Disallowing the root group RT runtime is BAD, it would disallow the | |
6543 | * kernel creating (and or operating) RT threads. | |
6544 | */ | |
6545 | if (tg == &root_task_group && rt_runtime == 0) | |
6546 | return -EINVAL; | |
6547 | ||
6548 | /* No period doesn't make any sense. */ | |
6549 | if (rt_period == 0) | |
6550 | return -EINVAL; | |
6551 | ||
9f0c1e56 | 6552 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 6553 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
6554 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
6555 | if (err) | |
9f0c1e56 | 6556 | goto unlock; |
ac086bc2 | 6557 | |
0986b11b | 6558 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
6559 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
6560 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
6561 | |
6562 | for_each_possible_cpu(i) { | |
6563 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
6564 | ||
0986b11b | 6565 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6566 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 6567 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6568 | } |
0986b11b | 6569 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 6570 | unlock: |
521f1a24 | 6571 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
6572 | mutex_unlock(&rt_constraints_mutex); |
6573 | ||
6574 | return err; | |
6f505b16 PZ |
6575 | } |
6576 | ||
25cc7da7 | 6577 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
d0b27fa7 PZ |
6578 | { |
6579 | u64 rt_runtime, rt_period; | |
6580 | ||
6581 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
6582 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
6583 | if (rt_runtime_us < 0) | |
6584 | rt_runtime = RUNTIME_INF; | |
6585 | ||
ab84d31e | 6586 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
6587 | } |
6588 | ||
25cc7da7 | 6589 | static long sched_group_rt_runtime(struct task_group *tg) |
9f0c1e56 PZ |
6590 | { |
6591 | u64 rt_runtime_us; | |
6592 | ||
d0b27fa7 | 6593 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
6594 | return -1; |
6595 | ||
d0b27fa7 | 6596 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
6597 | do_div(rt_runtime_us, NSEC_PER_USEC); |
6598 | return rt_runtime_us; | |
6599 | } | |
d0b27fa7 | 6600 | |
ce2f5fe4 | 6601 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
d0b27fa7 PZ |
6602 | { |
6603 | u64 rt_runtime, rt_period; | |
6604 | ||
ce2f5fe4 | 6605 | rt_period = rt_period_us * NSEC_PER_USEC; |
d0b27fa7 PZ |
6606 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
6607 | ||
ab84d31e | 6608 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
d0b27fa7 PZ |
6609 | } |
6610 | ||
25cc7da7 | 6611 | static long sched_group_rt_period(struct task_group *tg) |
d0b27fa7 PZ |
6612 | { |
6613 | u64 rt_period_us; | |
6614 | ||
6615 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
6616 | do_div(rt_period_us, NSEC_PER_USEC); | |
6617 | return rt_period_us; | |
6618 | } | |
332ac17e | 6619 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6620 | |
332ac17e | 6621 | #ifdef CONFIG_RT_GROUP_SCHED |
d0b27fa7 PZ |
6622 | static int sched_rt_global_constraints(void) |
6623 | { | |
6624 | int ret = 0; | |
6625 | ||
6626 | mutex_lock(&rt_constraints_mutex); | |
9a7e0b18 | 6627 | read_lock(&tasklist_lock); |
4653f803 | 6628 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 6629 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
6630 | mutex_unlock(&rt_constraints_mutex); |
6631 | ||
6632 | return ret; | |
6633 | } | |
54e99124 | 6634 | |
25cc7da7 | 6635 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
54e99124 DG |
6636 | { |
6637 | /* Don't accept realtime tasks when there is no way for them to run */ | |
6638 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
6639 | return 0; | |
6640 | ||
6641 | return 1; | |
6642 | } | |
6643 | ||
6d6bc0ad | 6644 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
6645 | static int sched_rt_global_constraints(void) |
6646 | { | |
ac086bc2 | 6647 | unsigned long flags; |
8c5e9554 | 6648 | int i; |
ec5d4989 | 6649 | |
0986b11b | 6650 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
6651 | for_each_possible_cpu(i) { |
6652 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
6653 | ||
0986b11b | 6654 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6655 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 6656 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 6657 | } |
0986b11b | 6658 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 6659 | |
8c5e9554 | 6660 | return 0; |
d0b27fa7 | 6661 | } |
6d6bc0ad | 6662 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6663 | |
a1963b81 | 6664 | static int sched_dl_global_validate(void) |
332ac17e | 6665 | { |
1724813d PZ |
6666 | u64 runtime = global_rt_runtime(); |
6667 | u64 period = global_rt_period(); | |
332ac17e | 6668 | u64 new_bw = to_ratio(period, runtime); |
f10e00f4 | 6669 | struct dl_bw *dl_b; |
1724813d | 6670 | int cpu, ret = 0; |
49516342 | 6671 | unsigned long flags; |
332ac17e DF |
6672 | |
6673 | /* | |
6674 | * Here we want to check the bandwidth not being set to some | |
6675 | * value smaller than the currently allocated bandwidth in | |
6676 | * any of the root_domains. | |
6677 | * | |
6678 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than | |
6679 | * cycling on root_domains... Discussion on different/better | |
6680 | * solutions is welcome! | |
6681 | */ | |
1724813d | 6682 | for_each_possible_cpu(cpu) { |
f10e00f4 KT |
6683 | rcu_read_lock_sched(); |
6684 | dl_b = dl_bw_of(cpu); | |
332ac17e | 6685 | |
49516342 | 6686 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d PZ |
6687 | if (new_bw < dl_b->total_bw) |
6688 | ret = -EBUSY; | |
49516342 | 6689 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
1724813d | 6690 | |
f10e00f4 KT |
6691 | rcu_read_unlock_sched(); |
6692 | ||
1724813d PZ |
6693 | if (ret) |
6694 | break; | |
332ac17e DF |
6695 | } |
6696 | ||
1724813d | 6697 | return ret; |
332ac17e DF |
6698 | } |
6699 | ||
1724813d | 6700 | static void sched_dl_do_global(void) |
ce0dbbbb | 6701 | { |
1724813d | 6702 | u64 new_bw = -1; |
f10e00f4 | 6703 | struct dl_bw *dl_b; |
1724813d | 6704 | int cpu; |
49516342 | 6705 | unsigned long flags; |
ce0dbbbb | 6706 | |
1724813d PZ |
6707 | def_dl_bandwidth.dl_period = global_rt_period(); |
6708 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); | |
6709 | ||
6710 | if (global_rt_runtime() != RUNTIME_INF) | |
6711 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); | |
6712 | ||
6713 | /* | |
6714 | * FIXME: As above... | |
6715 | */ | |
6716 | for_each_possible_cpu(cpu) { | |
f10e00f4 KT |
6717 | rcu_read_lock_sched(); |
6718 | dl_b = dl_bw_of(cpu); | |
1724813d | 6719 | |
49516342 | 6720 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
1724813d | 6721 | dl_b->bw = new_bw; |
49516342 | 6722 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
f10e00f4 KT |
6723 | |
6724 | rcu_read_unlock_sched(); | |
ce0dbbbb | 6725 | } |
1724813d PZ |
6726 | } |
6727 | ||
6728 | static int sched_rt_global_validate(void) | |
6729 | { | |
6730 | if (sysctl_sched_rt_period <= 0) | |
6731 | return -EINVAL; | |
6732 | ||
e9e7cb38 JL |
6733 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
6734 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) | |
1724813d PZ |
6735 | return -EINVAL; |
6736 | ||
6737 | return 0; | |
6738 | } | |
6739 | ||
6740 | static void sched_rt_do_global(void) | |
6741 | { | |
6742 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
6743 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); | |
ce0dbbbb CW |
6744 | } |
6745 | ||
d0b27fa7 | 6746 | int sched_rt_handler(struct ctl_table *table, int write, |
8d65af78 | 6747 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
6748 | loff_t *ppos) |
6749 | { | |
d0b27fa7 PZ |
6750 | int old_period, old_runtime; |
6751 | static DEFINE_MUTEX(mutex); | |
1724813d | 6752 | int ret; |
d0b27fa7 PZ |
6753 | |
6754 | mutex_lock(&mutex); | |
6755 | old_period = sysctl_sched_rt_period; | |
6756 | old_runtime = sysctl_sched_rt_runtime; | |
6757 | ||
8d65af78 | 6758 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
6759 | |
6760 | if (!ret && write) { | |
1724813d PZ |
6761 | ret = sched_rt_global_validate(); |
6762 | if (ret) | |
6763 | goto undo; | |
6764 | ||
a1963b81 | 6765 | ret = sched_dl_global_validate(); |
1724813d PZ |
6766 | if (ret) |
6767 | goto undo; | |
6768 | ||
a1963b81 | 6769 | ret = sched_rt_global_constraints(); |
1724813d PZ |
6770 | if (ret) |
6771 | goto undo; | |
6772 | ||
6773 | sched_rt_do_global(); | |
6774 | sched_dl_do_global(); | |
6775 | } | |
6776 | if (0) { | |
6777 | undo: | |
6778 | sysctl_sched_rt_period = old_period; | |
6779 | sysctl_sched_rt_runtime = old_runtime; | |
d0b27fa7 PZ |
6780 | } |
6781 | mutex_unlock(&mutex); | |
6782 | ||
6783 | return ret; | |
6784 | } | |
68318b8e | 6785 | |
1724813d | 6786 | int sched_rr_handler(struct ctl_table *table, int write, |
332ac17e DF |
6787 | void __user *buffer, size_t *lenp, |
6788 | loff_t *ppos) | |
6789 | { | |
6790 | int ret; | |
332ac17e | 6791 | static DEFINE_MUTEX(mutex); |
332ac17e DF |
6792 | |
6793 | mutex_lock(&mutex); | |
332ac17e | 6794 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d1ccc66d IM |
6795 | /* |
6796 | * Make sure that internally we keep jiffies. | |
6797 | * Also, writing zero resets the timeslice to default: | |
6798 | */ | |
332ac17e | 6799 | if (!ret && write) { |
975e155e SZ |
6800 | sched_rr_timeslice = |
6801 | sysctl_sched_rr_timeslice <= 0 ? RR_TIMESLICE : | |
6802 | msecs_to_jiffies(sysctl_sched_rr_timeslice); | |
332ac17e DF |
6803 | } |
6804 | mutex_unlock(&mutex); | |
332ac17e DF |
6805 | return ret; |
6806 | } | |
6807 | ||
052f1dc7 | 6808 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e | 6809 | |
a7c6d554 | 6810 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 6811 | { |
a7c6d554 | 6812 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
6813 | } |
6814 | ||
eb95419b TH |
6815 | static struct cgroup_subsys_state * |
6816 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 6817 | { |
eb95419b TH |
6818 | struct task_group *parent = css_tg(parent_css); |
6819 | struct task_group *tg; | |
68318b8e | 6820 | |
eb95419b | 6821 | if (!parent) { |
68318b8e | 6822 | /* This is early initialization for the top cgroup */ |
07e06b01 | 6823 | return &root_task_group.css; |
68318b8e SV |
6824 | } |
6825 | ||
ec7dc8ac | 6826 | tg = sched_create_group(parent); |
68318b8e SV |
6827 | if (IS_ERR(tg)) |
6828 | return ERR_PTR(-ENOMEM); | |
6829 | ||
68318b8e SV |
6830 | return &tg->css; |
6831 | } | |
6832 | ||
96b77745 KK |
6833 | /* Expose task group only after completing cgroup initialization */ |
6834 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
6835 | { | |
6836 | struct task_group *tg = css_tg(css); | |
6837 | struct task_group *parent = css_tg(css->parent); | |
6838 | ||
6839 | if (parent) | |
6840 | sched_online_group(tg, parent); | |
6841 | return 0; | |
6842 | } | |
6843 | ||
2f5177f0 | 6844 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 6845 | { |
eb95419b | 6846 | struct task_group *tg = css_tg(css); |
ace783b9 | 6847 | |
2f5177f0 | 6848 | sched_offline_group(tg); |
ace783b9 LZ |
6849 | } |
6850 | ||
eb95419b | 6851 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 6852 | { |
eb95419b | 6853 | struct task_group *tg = css_tg(css); |
68318b8e | 6854 | |
2f5177f0 PZ |
6855 | /* |
6856 | * Relies on the RCU grace period between css_released() and this. | |
6857 | */ | |
6858 | sched_free_group(tg); | |
ace783b9 LZ |
6859 | } |
6860 | ||
ea86cb4b VG |
6861 | /* |
6862 | * This is called before wake_up_new_task(), therefore we really only | |
6863 | * have to set its group bits, all the other stuff does not apply. | |
6864 | */ | |
b53202e6 | 6865 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 6866 | { |
ea86cb4b VG |
6867 | struct rq_flags rf; |
6868 | struct rq *rq; | |
6869 | ||
6870 | rq = task_rq_lock(task, &rf); | |
6871 | ||
80f5c1b8 | 6872 | update_rq_clock(rq); |
ea86cb4b VG |
6873 | sched_change_group(task, TASK_SET_GROUP); |
6874 | ||
6875 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
6876 | } |
6877 | ||
1f7dd3e5 | 6878 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 6879 | { |
bb9d97b6 | 6880 | struct task_struct *task; |
1f7dd3e5 | 6881 | struct cgroup_subsys_state *css; |
7dc603c9 | 6882 | int ret = 0; |
bb9d97b6 | 6883 | |
1f7dd3e5 | 6884 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 6885 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 6886 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 6887 | return -EINVAL; |
b68aa230 | 6888 | #else |
bb9d97b6 TH |
6889 | /* We don't support RT-tasks being in separate groups */ |
6890 | if (task->sched_class != &fair_sched_class) | |
6891 | return -EINVAL; | |
b68aa230 | 6892 | #endif |
7dc603c9 PZ |
6893 | /* |
6894 | * Serialize against wake_up_new_task() such that if its | |
6895 | * running, we're sure to observe its full state. | |
6896 | */ | |
6897 | raw_spin_lock_irq(&task->pi_lock); | |
6898 | /* | |
6899 | * Avoid calling sched_move_task() before wake_up_new_task() | |
6900 | * has happened. This would lead to problems with PELT, due to | |
6901 | * move wanting to detach+attach while we're not attached yet. | |
6902 | */ | |
6903 | if (task->state == TASK_NEW) | |
6904 | ret = -EINVAL; | |
6905 | raw_spin_unlock_irq(&task->pi_lock); | |
6906 | ||
6907 | if (ret) | |
6908 | break; | |
bb9d97b6 | 6909 | } |
7dc603c9 | 6910 | return ret; |
be367d09 | 6911 | } |
68318b8e | 6912 | |
1f7dd3e5 | 6913 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 6914 | { |
bb9d97b6 | 6915 | struct task_struct *task; |
1f7dd3e5 | 6916 | struct cgroup_subsys_state *css; |
bb9d97b6 | 6917 | |
1f7dd3e5 | 6918 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 6919 | sched_move_task(task); |
68318b8e SV |
6920 | } |
6921 | ||
052f1dc7 | 6922 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
6923 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
6924 | struct cftype *cftype, u64 shareval) | |
68318b8e | 6925 | { |
182446d0 | 6926 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
6927 | } |
6928 | ||
182446d0 TH |
6929 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
6930 | struct cftype *cft) | |
68318b8e | 6931 | { |
182446d0 | 6932 | struct task_group *tg = css_tg(css); |
68318b8e | 6933 | |
c8b28116 | 6934 | return (u64) scale_load_down(tg->shares); |
68318b8e | 6935 | } |
ab84d31e PT |
6936 | |
6937 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
6938 | static DEFINE_MUTEX(cfs_constraints_mutex); |
6939 | ||
ab84d31e PT |
6940 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
6941 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
6942 | ||
a790de99 PT |
6943 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
6944 | ||
ab84d31e PT |
6945 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
6946 | { | |
56f570e5 | 6947 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 6948 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
6949 | |
6950 | if (tg == &root_task_group) | |
6951 | return -EINVAL; | |
6952 | ||
6953 | /* | |
6954 | * Ensure we have at some amount of bandwidth every period. This is | |
6955 | * to prevent reaching a state of large arrears when throttled via | |
6956 | * entity_tick() resulting in prolonged exit starvation. | |
6957 | */ | |
6958 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
6959 | return -EINVAL; | |
6960 | ||
6961 | /* | |
6962 | * Likewise, bound things on the otherside by preventing insane quota | |
6963 | * periods. This also allows us to normalize in computing quota | |
6964 | * feasibility. | |
6965 | */ | |
6966 | if (period > max_cfs_quota_period) | |
6967 | return -EINVAL; | |
6968 | ||
0e59bdae KT |
6969 | /* |
6970 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
6971 | * unthrottle_offline_cfs_rqs(). | |
6972 | */ | |
6973 | get_online_cpus(); | |
a790de99 PT |
6974 | mutex_lock(&cfs_constraints_mutex); |
6975 | ret = __cfs_schedulable(tg, period, quota); | |
6976 | if (ret) | |
6977 | goto out_unlock; | |
6978 | ||
58088ad0 | 6979 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 6980 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
6981 | /* |
6982 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
6983 | * before making related changes, and on->off must occur afterwards | |
6984 | */ | |
6985 | if (runtime_enabled && !runtime_was_enabled) | |
6986 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
6987 | raw_spin_lock_irq(&cfs_b->lock); |
6988 | cfs_b->period = ns_to_ktime(period); | |
6989 | cfs_b->quota = quota; | |
58088ad0 | 6990 | |
a9cf55b2 | 6991 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
6992 | |
6993 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
6994 | if (runtime_enabled) |
6995 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 6996 | |
ab84d31e PT |
6997 | raw_spin_unlock_irq(&cfs_b->lock); |
6998 | ||
0e59bdae | 6999 | for_each_online_cpu(i) { |
ab84d31e | 7000 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 7001 | struct rq *rq = cfs_rq->rq; |
ab84d31e PT |
7002 | |
7003 | raw_spin_lock_irq(&rq->lock); | |
58088ad0 | 7004 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7005 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7006 | |
029632fb | 7007 | if (cfs_rq->throttled) |
671fd9da | 7008 | unthrottle_cfs_rq(cfs_rq); |
ab84d31e PT |
7009 | raw_spin_unlock_irq(&rq->lock); |
7010 | } | |
1ee14e6c BS |
7011 | if (runtime_was_enabled && !runtime_enabled) |
7012 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7013 | out_unlock: |
7014 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 7015 | put_online_cpus(); |
ab84d31e | 7016 | |
a790de99 | 7017 | return ret; |
ab84d31e PT |
7018 | } |
7019 | ||
7020 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
7021 | { | |
7022 | u64 quota, period; | |
7023 | ||
029632fb | 7024 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7025 | if (cfs_quota_us < 0) |
7026 | quota = RUNTIME_INF; | |
7027 | else | |
7028 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
7029 | ||
7030 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7031 | } | |
7032 | ||
7033 | long tg_get_cfs_quota(struct task_group *tg) | |
7034 | { | |
7035 | u64 quota_us; | |
7036 | ||
029632fb | 7037 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7038 | return -1; |
7039 | ||
029632fb | 7040 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7041 | do_div(quota_us, NSEC_PER_USEC); |
7042 | ||
7043 | return quota_us; | |
7044 | } | |
7045 | ||
7046 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
7047 | { | |
7048 | u64 quota, period; | |
7049 | ||
7050 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 7051 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7052 | |
ab84d31e PT |
7053 | return tg_set_cfs_bandwidth(tg, period, quota); |
7054 | } | |
7055 | ||
7056 | long tg_get_cfs_period(struct task_group *tg) | |
7057 | { | |
7058 | u64 cfs_period_us; | |
7059 | ||
029632fb | 7060 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7061 | do_div(cfs_period_us, NSEC_PER_USEC); |
7062 | ||
7063 | return cfs_period_us; | |
7064 | } | |
7065 | ||
182446d0 TH |
7066 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7067 | struct cftype *cft) | |
ab84d31e | 7068 | { |
182446d0 | 7069 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7070 | } |
7071 | ||
182446d0 TH |
7072 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7073 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7074 | { |
182446d0 | 7075 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7076 | } |
7077 | ||
182446d0 TH |
7078 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7079 | struct cftype *cft) | |
ab84d31e | 7080 | { |
182446d0 | 7081 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7082 | } |
7083 | ||
182446d0 TH |
7084 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7085 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7086 | { |
182446d0 | 7087 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7088 | } |
7089 | ||
a790de99 PT |
7090 | struct cfs_schedulable_data { |
7091 | struct task_group *tg; | |
7092 | u64 period, quota; | |
7093 | }; | |
7094 | ||
7095 | /* | |
7096 | * normalize group quota/period to be quota/max_period | |
7097 | * note: units are usecs | |
7098 | */ | |
7099 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7100 | struct cfs_schedulable_data *d) | |
7101 | { | |
7102 | u64 quota, period; | |
7103 | ||
7104 | if (tg == d->tg) { | |
7105 | period = d->period; | |
7106 | quota = d->quota; | |
7107 | } else { | |
7108 | period = tg_get_cfs_period(tg); | |
7109 | quota = tg_get_cfs_quota(tg); | |
7110 | } | |
7111 | ||
7112 | /* note: these should typically be equivalent */ | |
7113 | if (quota == RUNTIME_INF || quota == -1) | |
7114 | return RUNTIME_INF; | |
7115 | ||
7116 | return to_ratio(period, quota); | |
7117 | } | |
7118 | ||
7119 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7120 | { | |
7121 | struct cfs_schedulable_data *d = data; | |
029632fb | 7122 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7123 | s64 quota = 0, parent_quota = -1; |
7124 | ||
7125 | if (!tg->parent) { | |
7126 | quota = RUNTIME_INF; | |
7127 | } else { | |
029632fb | 7128 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7129 | |
7130 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 7131 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
7132 | |
7133 | /* | |
d1ccc66d IM |
7134 | * Ensure max(child_quota) <= parent_quota, inherit when no |
7135 | * limit is set: | |
a790de99 PT |
7136 | */ |
7137 | if (quota == RUNTIME_INF) | |
7138 | quota = parent_quota; | |
7139 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7140 | return -EINVAL; | |
7141 | } | |
9c58c79a | 7142 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
7143 | |
7144 | return 0; | |
7145 | } | |
7146 | ||
7147 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7148 | { | |
8277434e | 7149 | int ret; |
a790de99 PT |
7150 | struct cfs_schedulable_data data = { |
7151 | .tg = tg, | |
7152 | .period = period, | |
7153 | .quota = quota, | |
7154 | }; | |
7155 | ||
7156 | if (quota != RUNTIME_INF) { | |
7157 | do_div(data.period, NSEC_PER_USEC); | |
7158 | do_div(data.quota, NSEC_PER_USEC); | |
7159 | } | |
7160 | ||
8277434e PT |
7161 | rcu_read_lock(); |
7162 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7163 | rcu_read_unlock(); | |
7164 | ||
7165 | return ret; | |
a790de99 | 7166 | } |
e8da1b18 | 7167 | |
2da8ca82 | 7168 | static int cpu_stats_show(struct seq_file *sf, void *v) |
e8da1b18 | 7169 | { |
2da8ca82 | 7170 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7171 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7172 | |
44ffc75b TH |
7173 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7174 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7175 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 NR |
7176 | |
7177 | return 0; | |
7178 | } | |
ab84d31e | 7179 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7180 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7181 | |
052f1dc7 | 7182 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7183 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7184 | struct cftype *cft, s64 val) | |
6f505b16 | 7185 | { |
182446d0 | 7186 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7187 | } |
7188 | ||
182446d0 TH |
7189 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7190 | struct cftype *cft) | |
6f505b16 | 7191 | { |
182446d0 | 7192 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7193 | } |
d0b27fa7 | 7194 | |
182446d0 TH |
7195 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7196 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7197 | { |
182446d0 | 7198 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7199 | } |
7200 | ||
182446d0 TH |
7201 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7202 | struct cftype *cft) | |
d0b27fa7 | 7203 | { |
182446d0 | 7204 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7205 | } |
6d6bc0ad | 7206 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7207 | |
fe5c7cc2 | 7208 | static struct cftype cpu_files[] = { |
052f1dc7 | 7209 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7210 | { |
7211 | .name = "shares", | |
f4c753b7 PM |
7212 | .read_u64 = cpu_shares_read_u64, |
7213 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7214 | }, |
052f1dc7 | 7215 | #endif |
ab84d31e PT |
7216 | #ifdef CONFIG_CFS_BANDWIDTH |
7217 | { | |
7218 | .name = "cfs_quota_us", | |
7219 | .read_s64 = cpu_cfs_quota_read_s64, | |
7220 | .write_s64 = cpu_cfs_quota_write_s64, | |
7221 | }, | |
7222 | { | |
7223 | .name = "cfs_period_us", | |
7224 | .read_u64 = cpu_cfs_period_read_u64, | |
7225 | .write_u64 = cpu_cfs_period_write_u64, | |
7226 | }, | |
e8da1b18 NR |
7227 | { |
7228 | .name = "stat", | |
2da8ca82 | 7229 | .seq_show = cpu_stats_show, |
e8da1b18 | 7230 | }, |
ab84d31e | 7231 | #endif |
052f1dc7 | 7232 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7233 | { |
9f0c1e56 | 7234 | .name = "rt_runtime_us", |
06ecb27c PM |
7235 | .read_s64 = cpu_rt_runtime_read, |
7236 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7237 | }, |
d0b27fa7 PZ |
7238 | { |
7239 | .name = "rt_period_us", | |
f4c753b7 PM |
7240 | .read_u64 = cpu_rt_period_read_uint, |
7241 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7242 | }, |
052f1dc7 | 7243 | #endif |
d1ccc66d | 7244 | { } /* Terminate */ |
68318b8e SV |
7245 | }; |
7246 | ||
073219e9 | 7247 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 7248 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 7249 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 7250 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 7251 | .css_free = cpu_cgroup_css_free, |
eeb61e53 | 7252 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
7253 | .can_attach = cpu_cgroup_can_attach, |
7254 | .attach = cpu_cgroup_attach, | |
5577964e | 7255 | .legacy_cftypes = cpu_files, |
b38e42e9 | 7256 | .early_init = true, |
68318b8e SV |
7257 | }; |
7258 | ||
052f1dc7 | 7259 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 7260 | |
b637a328 PM |
7261 | void dump_cpu_task(int cpu) |
7262 | { | |
7263 | pr_info("Task dump for CPU %d:\n", cpu); | |
7264 | sched_show_task(cpu_curr(cpu)); | |
7265 | } | |
ed82b8a1 AK |
7266 | |
7267 | /* | |
7268 | * Nice levels are multiplicative, with a gentle 10% change for every | |
7269 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
7270 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
7271 | * that remained on nice 0. | |
7272 | * | |
7273 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
7274 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
7275 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
7276 | * If a task goes up by ~10% and another task goes down by ~10% then | |
7277 | * the relative distance between them is ~25%.) | |
7278 | */ | |
7279 | const int sched_prio_to_weight[40] = { | |
7280 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
7281 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
7282 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
7283 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
7284 | /* 0 */ 1024, 820, 655, 526, 423, | |
7285 | /* 5 */ 335, 272, 215, 172, 137, | |
7286 | /* 10 */ 110, 87, 70, 56, 45, | |
7287 | /* 15 */ 36, 29, 23, 18, 15, | |
7288 | }; | |
7289 | ||
7290 | /* | |
7291 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
7292 | * | |
7293 | * In cases where the weight does not change often, we can use the | |
7294 | * precalculated inverse to speed up arithmetics by turning divisions | |
7295 | * into multiplications: | |
7296 | */ | |
7297 | const u32 sched_prio_to_wmult[40] = { | |
7298 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
7299 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
7300 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
7301 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
7302 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
7303 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
7304 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
7305 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
7306 | }; |