]>
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 | */ |
325ea10c | 8 | #include "sched.h" |
1da177e4 | 9 | |
7281c8de | 10 | #include <linux/nospec.h> |
85f1abe0 | 11 | |
0ed557aa MR |
12 | #include <linux/kcov.h> |
13 | ||
96f951ed | 14 | #include <asm/switch_to.h> |
5517d86b | 15 | #include <asm/tlb.h> |
1da177e4 | 16 | |
ea138446 | 17 | #include "../workqueue_internal.h" |
29d5e047 | 18 | #include "../smpboot.h" |
6e0534f2 | 19 | |
91c27493 VG |
20 | #include "pelt.h" |
21 | ||
a8d154b0 | 22 | #define CREATE_TRACE_POINTS |
ad8d75ff | 23 | #include <trace/events/sched.h> |
a8d154b0 | 24 | |
029632fb | 25 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 26 | |
765cc3a4 | 27 | #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) |
bf5c91ba IM |
28 | /* |
29 | * Debugging: various feature bits | |
765cc3a4 PB |
30 | * |
31 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
32 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
33 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 34 | */ |
f00b45c1 PZ |
35 | #define SCHED_FEAT(name, enabled) \ |
36 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 37 | const_debug unsigned int sysctl_sched_features = |
391e43da | 38 | #include "features.h" |
f00b45c1 | 39 | 0; |
f00b45c1 | 40 | #undef SCHED_FEAT |
765cc3a4 | 41 | #endif |
f00b45c1 | 42 | |
b82d9fdd PZ |
43 | /* |
44 | * Number of tasks to iterate in a single balance run. | |
45 | * Limited because this is done with IRQs disabled. | |
46 | */ | |
47 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
48 | ||
fa85ae24 | 49 | /* |
d1ccc66d | 50 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
51 | * default: 1s |
52 | */ | |
9f0c1e56 | 53 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 54 | |
029632fb | 55 | __read_mostly int scheduler_running; |
6892b75e | 56 | |
9f0c1e56 PZ |
57 | /* |
58 | * part of the period that we allow rt tasks to run in us. | |
59 | * default: 0.95s | |
60 | */ | |
61 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 62 | |
3e71a462 PZ |
63 | /* |
64 | * __task_rq_lock - lock the rq @p resides on. | |
65 | */ | |
eb580751 | 66 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
67 | __acquires(rq->lock) |
68 | { | |
69 | struct rq *rq; | |
70 | ||
71 | lockdep_assert_held(&p->pi_lock); | |
72 | ||
73 | for (;;) { | |
74 | rq = task_rq(p); | |
75 | raw_spin_lock(&rq->lock); | |
76 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 77 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
78 | return rq; |
79 | } | |
80 | raw_spin_unlock(&rq->lock); | |
81 | ||
82 | while (unlikely(task_on_rq_migrating(p))) | |
83 | cpu_relax(); | |
84 | } | |
85 | } | |
86 | ||
87 | /* | |
88 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
89 | */ | |
eb580751 | 90 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
91 | __acquires(p->pi_lock) |
92 | __acquires(rq->lock) | |
93 | { | |
94 | struct rq *rq; | |
95 | ||
96 | for (;;) { | |
eb580751 | 97 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
98 | rq = task_rq(p); |
99 | raw_spin_lock(&rq->lock); | |
100 | /* | |
101 | * move_queued_task() task_rq_lock() | |
102 | * | |
103 | * ACQUIRE (rq->lock) | |
104 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
105 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
106 | * [S] ->cpu = new_cpu [L] task_rq() | |
107 | * [L] ->on_rq | |
108 | * RELEASE (rq->lock) | |
109 | * | |
97fb7a0a | 110 | * If we observe the old CPU in task_rq_lock, the acquire of |
3e71a462 PZ |
111 | * the old rq->lock will fully serialize against the stores. |
112 | * | |
d1ccc66d | 113 | * If we observe the new CPU in task_rq_lock, the acquire will |
3e71a462 PZ |
114 | * pair with the WMB to ensure we must then also see migrating. |
115 | */ | |
116 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 117 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
118 | return rq; |
119 | } | |
120 | raw_spin_unlock(&rq->lock); | |
eb580751 | 121 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
122 | |
123 | while (unlikely(task_on_rq_migrating(p))) | |
124 | cpu_relax(); | |
125 | } | |
126 | } | |
127 | ||
535b9552 IM |
128 | /* |
129 | * RQ-clock updating methods: | |
130 | */ | |
131 | ||
132 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
133 | { | |
134 | /* | |
135 | * In theory, the compile should just see 0 here, and optimize out the call | |
136 | * to sched_rt_avg_update. But I don't trust it... | |
137 | */ | |
11d4afd4 VG |
138 | s64 __maybe_unused steal = 0, irq_delta = 0; |
139 | ||
535b9552 IM |
140 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
141 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
142 | ||
143 | /* | |
144 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
145 | * this case when a previous update_rq_clock() happened inside a | |
146 | * {soft,}irq region. | |
147 | * | |
148 | * When this happens, we stop ->clock_task and only update the | |
149 | * prev_irq_time stamp to account for the part that fit, so that a next | |
150 | * update will consume the rest. This ensures ->clock_task is | |
151 | * monotonic. | |
152 | * | |
153 | * It does however cause some slight miss-attribution of {soft,}irq | |
154 | * time, a more accurate solution would be to update the irq_time using | |
155 | * the current rq->clock timestamp, except that would require using | |
156 | * atomic ops. | |
157 | */ | |
158 | if (irq_delta > delta) | |
159 | irq_delta = delta; | |
160 | ||
161 | rq->prev_irq_time += irq_delta; | |
162 | delta -= irq_delta; | |
163 | #endif | |
164 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
165 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
166 | steal = paravirt_steal_clock(cpu_of(rq)); | |
167 | steal -= rq->prev_steal_time_rq; | |
168 | ||
169 | if (unlikely(steal > delta)) | |
170 | steal = delta; | |
171 | ||
172 | rq->prev_steal_time_rq += steal; | |
173 | delta -= steal; | |
174 | } | |
175 | #endif | |
176 | ||
177 | rq->clock_task += delta; | |
178 | ||
11d4afd4 | 179 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 180 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 181 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 IM |
182 | #endif |
183 | } | |
184 | ||
185 | void update_rq_clock(struct rq *rq) | |
186 | { | |
187 | s64 delta; | |
188 | ||
189 | lockdep_assert_held(&rq->lock); | |
190 | ||
191 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
192 | return; | |
193 | ||
194 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
195 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
196 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
197 | rq->clock_update_flags |= RQCF_UPDATED; |
198 | #endif | |
26ae58d2 | 199 | |
535b9552 IM |
200 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
201 | if (delta < 0) | |
202 | return; | |
203 | rq->clock += delta; | |
204 | update_rq_clock_task(rq, delta); | |
205 | } | |
206 | ||
207 | ||
8f4d37ec PZ |
208 | #ifdef CONFIG_SCHED_HRTICK |
209 | /* | |
210 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 211 | */ |
8f4d37ec | 212 | |
8f4d37ec PZ |
213 | static void hrtick_clear(struct rq *rq) |
214 | { | |
215 | if (hrtimer_active(&rq->hrtick_timer)) | |
216 | hrtimer_cancel(&rq->hrtick_timer); | |
217 | } | |
218 | ||
8f4d37ec PZ |
219 | /* |
220 | * High-resolution timer tick. | |
221 | * Runs from hardirq context with interrupts disabled. | |
222 | */ | |
223 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
224 | { | |
225 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 226 | struct rq_flags rf; |
8f4d37ec PZ |
227 | |
228 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
229 | ||
8a8c69c3 | 230 | rq_lock(rq, &rf); |
3e51f33f | 231 | update_rq_clock(rq); |
8f4d37ec | 232 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 233 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
234 | |
235 | return HRTIMER_NORESTART; | |
236 | } | |
237 | ||
95e904c7 | 238 | #ifdef CONFIG_SMP |
971ee28c | 239 | |
4961b6e1 | 240 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
241 | { |
242 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 243 | |
4961b6e1 | 244 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
971ee28c PZ |
245 | } |
246 | ||
31656519 PZ |
247 | /* |
248 | * called from hardirq (IPI) context | |
249 | */ | |
250 | static void __hrtick_start(void *arg) | |
b328ca18 | 251 | { |
31656519 | 252 | struct rq *rq = arg; |
8a8c69c3 | 253 | struct rq_flags rf; |
b328ca18 | 254 | |
8a8c69c3 | 255 | rq_lock(rq, &rf); |
971ee28c | 256 | __hrtick_restart(rq); |
31656519 | 257 | rq->hrtick_csd_pending = 0; |
8a8c69c3 | 258 | rq_unlock(rq, &rf); |
b328ca18 PZ |
259 | } |
260 | ||
31656519 PZ |
261 | /* |
262 | * Called to set the hrtick timer state. | |
263 | * | |
264 | * called with rq->lock held and irqs disabled | |
265 | */ | |
029632fb | 266 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 267 | { |
31656519 | 268 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 269 | ktime_t time; |
270 | s64 delta; | |
271 | ||
272 | /* | |
273 | * Don't schedule slices shorter than 10000ns, that just | |
274 | * doesn't make sense and can cause timer DoS. | |
275 | */ | |
276 | delta = max_t(s64, delay, 10000LL); | |
277 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 278 | |
cc584b21 | 279 | hrtimer_set_expires(timer, time); |
31656519 PZ |
280 | |
281 | if (rq == this_rq()) { | |
971ee28c | 282 | __hrtick_restart(rq); |
31656519 | 283 | } else if (!rq->hrtick_csd_pending) { |
c46fff2a | 284 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
31656519 PZ |
285 | rq->hrtick_csd_pending = 1; |
286 | } | |
b328ca18 PZ |
287 | } |
288 | ||
31656519 PZ |
289 | #else |
290 | /* | |
291 | * Called to set the hrtick timer state. | |
292 | * | |
293 | * called with rq->lock held and irqs disabled | |
294 | */ | |
029632fb | 295 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 296 | { |
86893335 WL |
297 | /* |
298 | * Don't schedule slices shorter than 10000ns, that just | |
299 | * doesn't make sense. Rely on vruntime for fairness. | |
300 | */ | |
301 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 TG |
302 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
303 | HRTIMER_MODE_REL_PINNED); | |
31656519 | 304 | } |
31656519 | 305 | #endif /* CONFIG_SMP */ |
8f4d37ec | 306 | |
77a021be | 307 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 308 | { |
31656519 PZ |
309 | #ifdef CONFIG_SMP |
310 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 311 | |
31656519 PZ |
312 | rq->hrtick_csd.flags = 0; |
313 | rq->hrtick_csd.func = __hrtick_start; | |
314 | rq->hrtick_csd.info = rq; | |
315 | #endif | |
8f4d37ec | 316 | |
31656519 PZ |
317 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
318 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 319 | } |
006c75f1 | 320 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
321 | static inline void hrtick_clear(struct rq *rq) |
322 | { | |
323 | } | |
324 | ||
77a021be | 325 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
326 | { |
327 | } | |
006c75f1 | 328 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 329 | |
5529578a FW |
330 | /* |
331 | * cmpxchg based fetch_or, macro so it works for different integer types | |
332 | */ | |
333 | #define fetch_or(ptr, mask) \ | |
334 | ({ \ | |
335 | typeof(ptr) _ptr = (ptr); \ | |
336 | typeof(mask) _mask = (mask); \ | |
337 | typeof(*_ptr) _old, _val = *_ptr; \ | |
338 | \ | |
339 | for (;;) { \ | |
340 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
341 | if (_old == _val) \ | |
342 | break; \ | |
343 | _val = _old; \ | |
344 | } \ | |
345 | _old; \ | |
346 | }) | |
347 | ||
e3baac47 | 348 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
349 | /* |
350 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
351 | * this avoids any races wrt polling state changes and thereby avoids | |
352 | * spurious IPIs. | |
353 | */ | |
354 | static bool set_nr_and_not_polling(struct task_struct *p) | |
355 | { | |
356 | struct thread_info *ti = task_thread_info(p); | |
357 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
358 | } | |
e3baac47 PZ |
359 | |
360 | /* | |
361 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
362 | * | |
363 | * If this returns true, then the idle task promises to call | |
364 | * sched_ttwu_pending() and reschedule soon. | |
365 | */ | |
366 | static bool set_nr_if_polling(struct task_struct *p) | |
367 | { | |
368 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 369 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
370 | |
371 | for (;;) { | |
372 | if (!(val & _TIF_POLLING_NRFLAG)) | |
373 | return false; | |
374 | if (val & _TIF_NEED_RESCHED) | |
375 | return true; | |
376 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
377 | if (old == val) | |
378 | break; | |
379 | val = old; | |
380 | } | |
381 | return true; | |
382 | } | |
383 | ||
fd99f91a PZ |
384 | #else |
385 | static bool set_nr_and_not_polling(struct task_struct *p) | |
386 | { | |
387 | set_tsk_need_resched(p); | |
388 | return true; | |
389 | } | |
e3baac47 PZ |
390 | |
391 | #ifdef CONFIG_SMP | |
392 | static bool set_nr_if_polling(struct task_struct *p) | |
393 | { | |
394 | return false; | |
395 | } | |
396 | #endif | |
fd99f91a PZ |
397 | #endif |
398 | ||
76751049 PZ |
399 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
400 | { | |
401 | struct wake_q_node *node = &task->wake_q; | |
402 | ||
403 | /* | |
404 | * Atomically grab the task, if ->wake_q is !nil already it means | |
405 | * its already queued (either by us or someone else) and will get the | |
406 | * wakeup due to that. | |
407 | * | |
7696f991 AP |
408 | * This cmpxchg() executes a full barrier, which pairs with the full |
409 | * barrier executed by the wakeup in wake_up_q(). | |
76751049 PZ |
410 | */ |
411 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) | |
412 | return; | |
413 | ||
414 | get_task_struct(task); | |
415 | ||
416 | /* | |
417 | * The head is context local, there can be no concurrency. | |
418 | */ | |
419 | *head->lastp = node; | |
420 | head->lastp = &node->next; | |
421 | } | |
422 | ||
423 | void wake_up_q(struct wake_q_head *head) | |
424 | { | |
425 | struct wake_q_node *node = head->first; | |
426 | ||
427 | while (node != WAKE_Q_TAIL) { | |
428 | struct task_struct *task; | |
429 | ||
430 | task = container_of(node, struct task_struct, wake_q); | |
431 | BUG_ON(!task); | |
d1ccc66d | 432 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
433 | node = node->next; |
434 | task->wake_q.next = NULL; | |
435 | ||
436 | /* | |
7696f991 AP |
437 | * wake_up_process() executes a full barrier, which pairs with |
438 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
439 | */ |
440 | wake_up_process(task); | |
441 | put_task_struct(task); | |
442 | } | |
443 | } | |
444 | ||
c24d20db | 445 | /* |
8875125e | 446 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
447 | * |
448 | * On UP this means the setting of the need_resched flag, on SMP it | |
449 | * might also involve a cross-CPU call to trigger the scheduler on | |
450 | * the target CPU. | |
451 | */ | |
8875125e | 452 | void resched_curr(struct rq *rq) |
c24d20db | 453 | { |
8875125e | 454 | struct task_struct *curr = rq->curr; |
c24d20db IM |
455 | int cpu; |
456 | ||
8875125e | 457 | lockdep_assert_held(&rq->lock); |
c24d20db | 458 | |
8875125e | 459 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
460 | return; |
461 | ||
8875125e | 462 | cpu = cpu_of(rq); |
fd99f91a | 463 | |
f27dde8d | 464 | if (cpu == smp_processor_id()) { |
8875125e | 465 | set_tsk_need_resched(curr); |
f27dde8d | 466 | set_preempt_need_resched(); |
c24d20db | 467 | return; |
f27dde8d | 468 | } |
c24d20db | 469 | |
8875125e | 470 | if (set_nr_and_not_polling(curr)) |
c24d20db | 471 | smp_send_reschedule(cpu); |
dfc68f29 AL |
472 | else |
473 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
474 | } |
475 | ||
029632fb | 476 | void resched_cpu(int cpu) |
c24d20db IM |
477 | { |
478 | struct rq *rq = cpu_rq(cpu); | |
479 | unsigned long flags; | |
480 | ||
7c2102e5 | 481 | raw_spin_lock_irqsave(&rq->lock, flags); |
a0982dfa PM |
482 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
483 | resched_curr(rq); | |
05fa785c | 484 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 485 | } |
06d8308c | 486 | |
b021fe3e | 487 | #ifdef CONFIG_SMP |
3451d024 | 488 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 489 | /* |
d1ccc66d IM |
490 | * In the semi idle case, use the nearest busy CPU for migrating timers |
491 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
492 | * |
493 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
494 | * selecting an idle CPU will add more delays to the timers than intended |
495 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 496 | */ |
bc7a34b8 | 497 | int get_nohz_timer_target(void) |
83cd4fe2 | 498 | { |
bc7a34b8 | 499 | int i, cpu = smp_processor_id(); |
83cd4fe2 VP |
500 | struct sched_domain *sd; |
501 | ||
de201559 | 502 | if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER)) |
6201b4d6 VK |
503 | return cpu; |
504 | ||
057f3fad | 505 | rcu_read_lock(); |
83cd4fe2 | 506 | for_each_domain(cpu, sd) { |
057f3fad | 507 | for_each_cpu(i, sched_domain_span(sd)) { |
44496922 WL |
508 | if (cpu == i) |
509 | continue; | |
510 | ||
de201559 | 511 | if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) { |
057f3fad PZ |
512 | cpu = i; |
513 | goto unlock; | |
514 | } | |
515 | } | |
83cd4fe2 | 516 | } |
9642d18e | 517 | |
de201559 FW |
518 | if (!housekeeping_cpu(cpu, HK_FLAG_TIMER)) |
519 | cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
057f3fad PZ |
520 | unlock: |
521 | rcu_read_unlock(); | |
83cd4fe2 VP |
522 | return cpu; |
523 | } | |
d1ccc66d | 524 | |
06d8308c TG |
525 | /* |
526 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
527 | * idle CPU then this timer might expire before the next timer event | |
528 | * which is scheduled to wake up that CPU. In case of a completely | |
529 | * idle system the next event might even be infinite time into the | |
530 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
531 | * leaves the inner idle loop so the newly added timer is taken into | |
532 | * account when the CPU goes back to idle and evaluates the timer | |
533 | * wheel for the next timer event. | |
534 | */ | |
1c20091e | 535 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
536 | { |
537 | struct rq *rq = cpu_rq(cpu); | |
538 | ||
539 | if (cpu == smp_processor_id()) | |
540 | return; | |
541 | ||
67b9ca70 | 542 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 543 | smp_send_reschedule(cpu); |
dfc68f29 AL |
544 | else |
545 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
546 | } |
547 | ||
c5bfece2 | 548 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 549 | { |
53c5fa16 FW |
550 | /* |
551 | * We just need the target to call irq_exit() and re-evaluate | |
552 | * the next tick. The nohz full kick at least implies that. | |
553 | * If needed we can still optimize that later with an | |
554 | * empty IRQ. | |
555 | */ | |
379d9ecb PM |
556 | if (cpu_is_offline(cpu)) |
557 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 558 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
559 | if (cpu != smp_processor_id() || |
560 | tick_nohz_tick_stopped()) | |
53c5fa16 | 561 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
562 | return true; |
563 | } | |
564 | ||
565 | return false; | |
566 | } | |
567 | ||
379d9ecb PM |
568 | /* |
569 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
570 | * caller's responsibility to deal with the lost wakeup, for example, | |
571 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
572 | */ | |
1c20091e FW |
573 | void wake_up_nohz_cpu(int cpu) |
574 | { | |
c5bfece2 | 575 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
576 | wake_up_idle_cpu(cpu); |
577 | } | |
578 | ||
ca38062e | 579 | static inline bool got_nohz_idle_kick(void) |
45bf76df | 580 | { |
1c792db7 | 581 | int cpu = smp_processor_id(); |
873b4c65 | 582 | |
b7031a02 | 583 | if (!(atomic_read(nohz_flags(cpu)) & NOHZ_KICK_MASK)) |
873b4c65 VG |
584 | return false; |
585 | ||
586 | if (idle_cpu(cpu) && !need_resched()) | |
587 | return true; | |
588 | ||
589 | /* | |
590 | * We can't run Idle Load Balance on this CPU for this time so we | |
591 | * cancel it and clear NOHZ_BALANCE_KICK | |
592 | */ | |
b7031a02 | 593 | atomic_andnot(NOHZ_KICK_MASK, nohz_flags(cpu)); |
873b4c65 | 594 | return false; |
45bf76df IM |
595 | } |
596 | ||
3451d024 | 597 | #else /* CONFIG_NO_HZ_COMMON */ |
45bf76df | 598 | |
ca38062e | 599 | static inline bool got_nohz_idle_kick(void) |
2069dd75 | 600 | { |
ca38062e | 601 | return false; |
2069dd75 PZ |
602 | } |
603 | ||
3451d024 | 604 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 605 | |
ce831b38 | 606 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 607 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 608 | { |
76d92ac3 FW |
609 | int fifo_nr_running; |
610 | ||
611 | /* Deadline tasks, even if single, need the tick */ | |
612 | if (rq->dl.dl_nr_running) | |
613 | return false; | |
614 | ||
1e78cdbd | 615 | /* |
2548d546 PZ |
616 | * If there are more than one RR tasks, we need the tick to effect the |
617 | * actual RR behaviour. | |
1e78cdbd | 618 | */ |
76d92ac3 FW |
619 | if (rq->rt.rr_nr_running) { |
620 | if (rq->rt.rr_nr_running == 1) | |
621 | return true; | |
622 | else | |
623 | return false; | |
1e78cdbd RR |
624 | } |
625 | ||
2548d546 PZ |
626 | /* |
627 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
628 | * forced preemption between FIFO tasks. | |
629 | */ | |
630 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
631 | if (fifo_nr_running) | |
632 | return true; | |
633 | ||
634 | /* | |
635 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
636 | * if there's more than one we need the tick for involuntary | |
637 | * preemption. | |
638 | */ | |
639 | if (rq->nr_running > 1) | |
541b8264 | 640 | return false; |
ce831b38 | 641 | |
541b8264 | 642 | return true; |
ce831b38 FW |
643 | } |
644 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 645 | #endif /* CONFIG_SMP */ |
18d95a28 | 646 | |
a790de99 PT |
647 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
648 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 649 | /* |
8277434e PT |
650 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
651 | * node and @up when leaving it for the final time. | |
652 | * | |
653 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 654 | */ |
029632fb | 655 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 656 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
657 | { |
658 | struct task_group *parent, *child; | |
eb755805 | 659 | int ret; |
c09595f6 | 660 | |
8277434e PT |
661 | parent = from; |
662 | ||
c09595f6 | 663 | down: |
eb755805 PZ |
664 | ret = (*down)(parent, data); |
665 | if (ret) | |
8277434e | 666 | goto out; |
c09595f6 PZ |
667 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
668 | parent = child; | |
669 | goto down; | |
670 | ||
671 | up: | |
672 | continue; | |
673 | } | |
eb755805 | 674 | ret = (*up)(parent, data); |
8277434e PT |
675 | if (ret || parent == from) |
676 | goto out; | |
c09595f6 PZ |
677 | |
678 | child = parent; | |
679 | parent = parent->parent; | |
680 | if (parent) | |
681 | goto up; | |
8277434e | 682 | out: |
eb755805 | 683 | return ret; |
c09595f6 PZ |
684 | } |
685 | ||
029632fb | 686 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 687 | { |
e2b245f8 | 688 | return 0; |
eb755805 | 689 | } |
18d95a28 PZ |
690 | #endif |
691 | ||
9059393e | 692 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 693 | { |
f05998d4 NR |
694 | int prio = p->static_prio - MAX_RT_PRIO; |
695 | struct load_weight *load = &p->se.load; | |
696 | ||
dd41f596 IM |
697 | /* |
698 | * SCHED_IDLE tasks get minimal weight: | |
699 | */ | |
20f9cd2a | 700 | if (idle_policy(p->policy)) { |
c8b28116 | 701 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 702 | load->inv_weight = WMULT_IDLEPRIO; |
4a465e3e | 703 | p->se.runnable_weight = load->weight; |
dd41f596 IM |
704 | return; |
705 | } | |
71f8bd46 | 706 | |
9059393e VG |
707 | /* |
708 | * SCHED_OTHER tasks have to update their load when changing their | |
709 | * weight | |
710 | */ | |
711 | if (update_load && p->sched_class == &fair_sched_class) { | |
712 | reweight_task(p, prio); | |
713 | } else { | |
714 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
715 | load->inv_weight = sched_prio_to_wmult[prio]; | |
4a465e3e | 716 | p->se.runnable_weight = load->weight; |
9059393e | 717 | } |
71f8bd46 IM |
718 | } |
719 | ||
1de64443 | 720 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 721 | { |
0a67d1ee PZ |
722 | if (!(flags & ENQUEUE_NOCLOCK)) |
723 | update_rq_clock(rq); | |
724 | ||
eb414681 | 725 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 726 | sched_info_queued(rq, p); |
eb414681 JW |
727 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
728 | } | |
0a67d1ee | 729 | |
371fd7e7 | 730 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
731 | } |
732 | ||
1de64443 | 733 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 734 | { |
0a67d1ee PZ |
735 | if (!(flags & DEQUEUE_NOCLOCK)) |
736 | update_rq_clock(rq); | |
737 | ||
eb414681 | 738 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 739 | sched_info_dequeued(rq, p); |
eb414681 JW |
740 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
741 | } | |
0a67d1ee | 742 | |
371fd7e7 | 743 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
744 | } |
745 | ||
029632fb | 746 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
747 | { |
748 | if (task_contributes_to_load(p)) | |
749 | rq->nr_uninterruptible--; | |
750 | ||
371fd7e7 | 751 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
752 | } |
753 | ||
029632fb | 754 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
755 | { |
756 | if (task_contributes_to_load(p)) | |
757 | rq->nr_uninterruptible++; | |
758 | ||
371fd7e7 | 759 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
760 | } |
761 | ||
14531189 | 762 | /* |
dd41f596 | 763 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 764 | */ |
14531189 IM |
765 | static inline int __normal_prio(struct task_struct *p) |
766 | { | |
dd41f596 | 767 | return p->static_prio; |
14531189 IM |
768 | } |
769 | ||
b29739f9 IM |
770 | /* |
771 | * Calculate the expected normal priority: i.e. priority | |
772 | * without taking RT-inheritance into account. Might be | |
773 | * boosted by interactivity modifiers. Changes upon fork, | |
774 | * setprio syscalls, and whenever the interactivity | |
775 | * estimator recalculates. | |
776 | */ | |
36c8b586 | 777 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
778 | { |
779 | int prio; | |
780 | ||
aab03e05 DF |
781 | if (task_has_dl_policy(p)) |
782 | prio = MAX_DL_PRIO-1; | |
783 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
784 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
785 | else | |
786 | prio = __normal_prio(p); | |
787 | return prio; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Calculate the current priority, i.e. the priority | |
792 | * taken into account by the scheduler. This value might | |
793 | * be boosted by RT tasks, or might be boosted by | |
794 | * interactivity modifiers. Will be RT if the task got | |
795 | * RT-boosted. If not then it returns p->normal_prio. | |
796 | */ | |
36c8b586 | 797 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
798 | { |
799 | p->normal_prio = normal_prio(p); | |
800 | /* | |
801 | * If we are RT tasks or we were boosted to RT priority, | |
802 | * keep the priority unchanged. Otherwise, update priority | |
803 | * to the normal priority: | |
804 | */ | |
805 | if (!rt_prio(p->prio)) | |
806 | return p->normal_prio; | |
807 | return p->prio; | |
808 | } | |
809 | ||
1da177e4 LT |
810 | /** |
811 | * task_curr - is this task currently executing on a CPU? | |
812 | * @p: the task in question. | |
e69f6186 YB |
813 | * |
814 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 815 | */ |
36c8b586 | 816 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
817 | { |
818 | return cpu_curr(task_cpu(p)) == p; | |
819 | } | |
820 | ||
67dfa1b7 | 821 | /* |
4c9a4bc8 PZ |
822 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
823 | * use the balance_callback list if you want balancing. | |
824 | * | |
825 | * this means any call to check_class_changed() must be followed by a call to | |
826 | * balance_callback(). | |
67dfa1b7 | 827 | */ |
cb469845 SR |
828 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
829 | const struct sched_class *prev_class, | |
da7a735e | 830 | int oldprio) |
cb469845 SR |
831 | { |
832 | if (prev_class != p->sched_class) { | |
833 | if (prev_class->switched_from) | |
da7a735e | 834 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 835 | |
da7a735e | 836 | p->sched_class->switched_to(rq, p); |
2d3d891d | 837 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 838 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
839 | } |
840 | ||
029632fb | 841 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
842 | { |
843 | const struct sched_class *class; | |
844 | ||
845 | if (p->sched_class == rq->curr->sched_class) { | |
846 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
847 | } else { | |
848 | for_each_class(class) { | |
849 | if (class == rq->curr->sched_class) | |
850 | break; | |
851 | if (class == p->sched_class) { | |
8875125e | 852 | resched_curr(rq); |
1e5a7405 PZ |
853 | break; |
854 | } | |
855 | } | |
856 | } | |
857 | ||
858 | /* | |
859 | * A queue event has occurred, and we're going to schedule. In | |
860 | * this case, we can save a useless back to back clock update. | |
861 | */ | |
da0c1e65 | 862 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 863 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
864 | } |
865 | ||
1da177e4 | 866 | #ifdef CONFIG_SMP |
175f0e25 PZ |
867 | |
868 | static inline bool is_per_cpu_kthread(struct task_struct *p) | |
869 | { | |
870 | if (!(p->flags & PF_KTHREAD)) | |
871 | return false; | |
872 | ||
873 | if (p->nr_cpus_allowed != 1) | |
874 | return false; | |
875 | ||
876 | return true; | |
877 | } | |
878 | ||
879 | /* | |
880 | * Per-CPU kthreads are allowed to run on !actie && online CPUs, see | |
881 | * __set_cpus_allowed_ptr() and select_fallback_rq(). | |
882 | */ | |
883 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
884 | { | |
885 | if (!cpumask_test_cpu(cpu, &p->cpus_allowed)) | |
886 | return false; | |
887 | ||
888 | if (is_per_cpu_kthread(p)) | |
889 | return cpu_online(cpu); | |
890 | ||
891 | return cpu_active(cpu); | |
892 | } | |
893 | ||
5cc389bc PZ |
894 | /* |
895 | * This is how migration works: | |
896 | * | |
897 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
898 | * stop_one_cpu(). | |
899 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
900 | * off the CPU) | |
901 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
902 | * 4) if it's in the wrong runqueue then the migration thread removes | |
903 | * it and puts it into the right queue. | |
904 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
905 | * is done. | |
906 | */ | |
907 | ||
908 | /* | |
909 | * move_queued_task - move a queued task to new rq. | |
910 | * | |
911 | * Returns (locked) new rq. Old rq's lock is released. | |
912 | */ | |
8a8c69c3 PZ |
913 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
914 | struct task_struct *p, int new_cpu) | |
5cc389bc | 915 | { |
5cc389bc PZ |
916 | lockdep_assert_held(&rq->lock); |
917 | ||
5cc389bc | 918 | p->on_rq = TASK_ON_RQ_MIGRATING; |
15ff991e | 919 | dequeue_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 920 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 921 | rq_unlock(rq, rf); |
5cc389bc PZ |
922 | |
923 | rq = cpu_rq(new_cpu); | |
924 | ||
8a8c69c3 | 925 | rq_lock(rq, rf); |
5cc389bc | 926 | BUG_ON(task_cpu(p) != new_cpu); |
5cc389bc | 927 | enqueue_task(rq, p, 0); |
3ea94de1 | 928 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
929 | check_preempt_curr(rq, p, 0); |
930 | ||
931 | return rq; | |
932 | } | |
933 | ||
934 | struct migration_arg { | |
935 | struct task_struct *task; | |
936 | int dest_cpu; | |
937 | }; | |
938 | ||
939 | /* | |
d1ccc66d | 940 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
941 | * this because either it can't run here any more (set_cpus_allowed() |
942 | * away from this CPU, or CPU going down), or because we're | |
943 | * attempting to rebalance this task on exec (sched_exec). | |
944 | * | |
945 | * So we race with normal scheduler movements, but that's OK, as long | |
946 | * as the task is no longer on this CPU. | |
5cc389bc | 947 | */ |
8a8c69c3 PZ |
948 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
949 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 950 | { |
5cc389bc | 951 | /* Affinity changed (again). */ |
175f0e25 | 952 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 953 | return rq; |
5cc389bc | 954 | |
15ff991e | 955 | update_rq_clock(rq); |
8a8c69c3 | 956 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
957 | |
958 | return rq; | |
5cc389bc PZ |
959 | } |
960 | ||
961 | /* | |
962 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
963 | * and performs thread migration by bumping thread off CPU then | |
964 | * 'pushing' onto another runqueue. | |
965 | */ | |
966 | static int migration_cpu_stop(void *data) | |
967 | { | |
968 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
969 | struct task_struct *p = arg->task; |
970 | struct rq *rq = this_rq(); | |
8a8c69c3 | 971 | struct rq_flags rf; |
5cc389bc PZ |
972 | |
973 | /* | |
d1ccc66d IM |
974 | * The original target CPU might have gone down and we might |
975 | * be on another CPU but it doesn't matter. | |
5cc389bc PZ |
976 | */ |
977 | local_irq_disable(); | |
978 | /* | |
979 | * We need to explicitly wake pending tasks before running | |
980 | * __migrate_task() such that we will not miss enforcing cpus_allowed | |
981 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. | |
982 | */ | |
983 | sched_ttwu_pending(); | |
5e16bbc2 PZ |
984 | |
985 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 986 | rq_lock(rq, &rf); |
5e16bbc2 PZ |
987 | /* |
988 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
989 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
990 | * we're holding p->pi_lock. | |
991 | */ | |
bf89a304 CC |
992 | if (task_rq(p) == rq) { |
993 | if (task_on_rq_queued(p)) | |
8a8c69c3 | 994 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 CC |
995 | else |
996 | p->wake_cpu = arg->dest_cpu; | |
997 | } | |
8a8c69c3 | 998 | rq_unlock(rq, &rf); |
5e16bbc2 PZ |
999 | raw_spin_unlock(&p->pi_lock); |
1000 | ||
5cc389bc PZ |
1001 | local_irq_enable(); |
1002 | return 0; | |
1003 | } | |
1004 | ||
c5b28038 PZ |
1005 | /* |
1006 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1007 | * actually call this function. | |
1008 | */ | |
1009 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1010 | { |
5cc389bc PZ |
1011 | cpumask_copy(&p->cpus_allowed, new_mask); |
1012 | p->nr_cpus_allowed = cpumask_weight(new_mask); | |
1013 | } | |
1014 | ||
c5b28038 PZ |
1015 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1016 | { | |
6c37067e PZ |
1017 | struct rq *rq = task_rq(p); |
1018 | bool queued, running; | |
1019 | ||
c5b28038 | 1020 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1021 | |
1022 | queued = task_on_rq_queued(p); | |
1023 | running = task_current(rq, p); | |
1024 | ||
1025 | if (queued) { | |
1026 | /* | |
1027 | * Because __kthread_bind() calls this on blocked tasks without | |
1028 | * holding rq->lock. | |
1029 | */ | |
1030 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 1031 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
1032 | } |
1033 | if (running) | |
1034 | put_prev_task(rq, p); | |
1035 | ||
c5b28038 | 1036 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e | 1037 | |
6c37067e | 1038 | if (queued) |
7134b3e9 | 1039 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 1040 | if (running) |
b2bf6c31 | 1041 | set_curr_task(rq, p); |
c5b28038 PZ |
1042 | } |
1043 | ||
5cc389bc PZ |
1044 | /* |
1045 | * Change a given task's CPU affinity. Migrate the thread to a | |
1046 | * proper CPU and schedule it away if the CPU it's executing on | |
1047 | * is removed from the allowed bitmask. | |
1048 | * | |
1049 | * NOTE: the caller must have a valid reference to the task, the | |
1050 | * task must not exit() & deallocate itself prematurely. The | |
1051 | * call is not atomic; no spinlocks may be held. | |
1052 | */ | |
25834c73 PZ |
1053 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1054 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1055 | { |
e9d867a6 | 1056 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1057 | unsigned int dest_cpu; |
eb580751 PZ |
1058 | struct rq_flags rf; |
1059 | struct rq *rq; | |
5cc389bc PZ |
1060 | int ret = 0; |
1061 | ||
eb580751 | 1062 | rq = task_rq_lock(p, &rf); |
a499c3ea | 1063 | update_rq_clock(rq); |
5cc389bc | 1064 | |
e9d867a6 PZI |
1065 | if (p->flags & PF_KTHREAD) { |
1066 | /* | |
1067 | * Kernel threads are allowed on online && !active CPUs | |
1068 | */ | |
1069 | cpu_valid_mask = cpu_online_mask; | |
1070 | } | |
1071 | ||
25834c73 PZ |
1072 | /* |
1073 | * Must re-check here, to close a race against __kthread_bind(), | |
1074 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1075 | */ | |
1076 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1077 | ret = -EINVAL; | |
1078 | goto out; | |
1079 | } | |
1080 | ||
5cc389bc PZ |
1081 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
1082 | goto out; | |
1083 | ||
e9d867a6 | 1084 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
5cc389bc PZ |
1085 | ret = -EINVAL; |
1086 | goto out; | |
1087 | } | |
1088 | ||
1089 | do_set_cpus_allowed(p, new_mask); | |
1090 | ||
e9d867a6 PZI |
1091 | if (p->flags & PF_KTHREAD) { |
1092 | /* | |
1093 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 1094 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
1095 | */ |
1096 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1097 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1098 | p->nr_cpus_allowed != 1); | |
1099 | } | |
1100 | ||
5cc389bc PZ |
1101 | /* Can the task run on the task's current CPU? If so, we're done */ |
1102 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1103 | goto out; | |
1104 | ||
e9d867a6 | 1105 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
5cc389bc PZ |
1106 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1107 | struct migration_arg arg = { p, dest_cpu }; | |
1108 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1109 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1110 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1111 | tlb_migrate_finish(p->mm); | |
1112 | return 0; | |
cbce1a68 PZ |
1113 | } else if (task_on_rq_queued(p)) { |
1114 | /* | |
1115 | * OK, since we're going to drop the lock immediately | |
1116 | * afterwards anyway. | |
1117 | */ | |
8a8c69c3 | 1118 | rq = move_queued_task(rq, &rf, p, dest_cpu); |
cbce1a68 | 1119 | } |
5cc389bc | 1120 | out: |
eb580751 | 1121 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1122 | |
1123 | return ret; | |
1124 | } | |
25834c73 PZ |
1125 | |
1126 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1127 | { | |
1128 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1129 | } | |
5cc389bc PZ |
1130 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1131 | ||
dd41f596 | 1132 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1133 | { |
e2912009 PZ |
1134 | #ifdef CONFIG_SCHED_DEBUG |
1135 | /* | |
1136 | * We should never call set_task_cpu() on a blocked task, | |
1137 | * ttwu() will sort out the placement. | |
1138 | */ | |
077614ee | 1139 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1140 | !p->on_rq); |
0122ec5b | 1141 | |
3ea94de1 JP |
1142 | /* |
1143 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1144 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1145 | * time relying on p->on_rq. | |
1146 | */ | |
1147 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1148 | p->sched_class == &fair_sched_class && | |
1149 | (p->on_rq && !task_on_rq_migrating(p))); | |
1150 | ||
0122ec5b | 1151 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1152 | /* |
1153 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1154 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1155 | * | |
1156 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1157 | * see task_group(). |
6c6c54e1 PZ |
1158 | * |
1159 | * Furthermore, all task_rq users should acquire both locks, see | |
1160 | * task_rq_lock(). | |
1161 | */ | |
0122ec5b PZ |
1162 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1163 | lockdep_is_held(&task_rq(p)->lock))); | |
1164 | #endif | |
4ff9083b PZ |
1165 | /* |
1166 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
1167 | */ | |
1168 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
e2912009 PZ |
1169 | #endif |
1170 | ||
de1d7286 | 1171 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1172 | |
0c69774e | 1173 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1174 | if (p->sched_class->migrate_task_rq) |
1327237a | 1175 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 1176 | p->se.nr_migrations++; |
d7822b1e | 1177 | rseq_migrate(p); |
ff303e66 | 1178 | perf_event_task_migrate(p); |
0c69774e | 1179 | } |
dd41f596 IM |
1180 | |
1181 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1182 | } |
1183 | ||
0ad4e3df | 1184 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
1185 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1186 | { | |
da0c1e65 | 1187 | if (task_on_rq_queued(p)) { |
ac66f547 | 1188 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 1189 | struct rq_flags srf, drf; |
ac66f547 PZ |
1190 | |
1191 | src_rq = task_rq(p); | |
1192 | dst_rq = cpu_rq(cpu); | |
1193 | ||
8a8c69c3 PZ |
1194 | rq_pin_lock(src_rq, &srf); |
1195 | rq_pin_lock(dst_rq, &drf); | |
1196 | ||
3ea94de1 | 1197 | p->on_rq = TASK_ON_RQ_MIGRATING; |
ac66f547 PZ |
1198 | deactivate_task(src_rq, p, 0); |
1199 | set_task_cpu(p, cpu); | |
1200 | activate_task(dst_rq, p, 0); | |
3ea94de1 | 1201 | p->on_rq = TASK_ON_RQ_QUEUED; |
ac66f547 | 1202 | check_preempt_curr(dst_rq, p, 0); |
8a8c69c3 PZ |
1203 | |
1204 | rq_unpin_lock(dst_rq, &drf); | |
1205 | rq_unpin_lock(src_rq, &srf); | |
1206 | ||
ac66f547 PZ |
1207 | } else { |
1208 | /* | |
1209 | * Task isn't running anymore; make it appear like we migrated | |
1210 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 1211 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
1212 | */ |
1213 | p->wake_cpu = cpu; | |
1214 | } | |
1215 | } | |
1216 | ||
1217 | struct migration_swap_arg { | |
1218 | struct task_struct *src_task, *dst_task; | |
1219 | int src_cpu, dst_cpu; | |
1220 | }; | |
1221 | ||
1222 | static int migrate_swap_stop(void *data) | |
1223 | { | |
1224 | struct migration_swap_arg *arg = data; | |
1225 | struct rq *src_rq, *dst_rq; | |
1226 | int ret = -EAGAIN; | |
1227 | ||
62694cd5 PZ |
1228 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1229 | return -EAGAIN; | |
1230 | ||
ac66f547 PZ |
1231 | src_rq = cpu_rq(arg->src_cpu); |
1232 | dst_rq = cpu_rq(arg->dst_cpu); | |
1233 | ||
74602315 PZ |
1234 | double_raw_lock(&arg->src_task->pi_lock, |
1235 | &arg->dst_task->pi_lock); | |
ac66f547 | 1236 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1237 | |
ac66f547 PZ |
1238 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1239 | goto unlock; | |
1240 | ||
1241 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1242 | goto unlock; | |
1243 | ||
0c98d344 | 1244 | if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed)) |
ac66f547 PZ |
1245 | goto unlock; |
1246 | ||
0c98d344 | 1247 | if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed)) |
ac66f547 PZ |
1248 | goto unlock; |
1249 | ||
1250 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1251 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1252 | ||
1253 | ret = 0; | |
1254 | ||
1255 | unlock: | |
1256 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1257 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1258 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1259 | |
1260 | return ret; | |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * Cross migrate two tasks | |
1265 | */ | |
0ad4e3df SD |
1266 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
1267 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
1268 | { |
1269 | struct migration_swap_arg arg; | |
1270 | int ret = -EINVAL; | |
1271 | ||
ac66f547 PZ |
1272 | arg = (struct migration_swap_arg){ |
1273 | .src_task = cur, | |
0ad4e3df | 1274 | .src_cpu = curr_cpu, |
ac66f547 | 1275 | .dst_task = p, |
0ad4e3df | 1276 | .dst_cpu = target_cpu, |
ac66f547 PZ |
1277 | }; |
1278 | ||
1279 | if (arg.src_cpu == arg.dst_cpu) | |
1280 | goto out; | |
1281 | ||
6acce3ef PZ |
1282 | /* |
1283 | * These three tests are all lockless; this is OK since all of them | |
1284 | * will be re-checked with proper locks held further down the line. | |
1285 | */ | |
ac66f547 PZ |
1286 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1287 | goto out; | |
1288 | ||
0c98d344 | 1289 | if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed)) |
ac66f547 PZ |
1290 | goto out; |
1291 | ||
0c98d344 | 1292 | if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed)) |
ac66f547 PZ |
1293 | goto out; |
1294 | ||
286549dc | 1295 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1296 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1297 | ||
1298 | out: | |
ac66f547 PZ |
1299 | return ret; |
1300 | } | |
0ad4e3df | 1301 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 1302 | |
1da177e4 LT |
1303 | /* |
1304 | * wait_task_inactive - wait for a thread to unschedule. | |
1305 | * | |
85ba2d86 RM |
1306 | * If @match_state is nonzero, it's the @p->state value just checked and |
1307 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1308 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1309 | * we return a positive number (its total switch count). If a second call | |
1310 | * a short while later returns the same number, the caller can be sure that | |
1311 | * @p has remained unscheduled the whole time. | |
1312 | * | |
1da177e4 LT |
1313 | * The caller must ensure that the task *will* unschedule sometime soon, |
1314 | * else this function might spin for a *long* time. This function can't | |
1315 | * be called with interrupts off, or it may introduce deadlock with | |
1316 | * smp_call_function() if an IPI is sent by the same process we are | |
1317 | * waiting to become inactive. | |
1318 | */ | |
85ba2d86 | 1319 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1320 | { |
da0c1e65 | 1321 | int running, queued; |
eb580751 | 1322 | struct rq_flags rf; |
85ba2d86 | 1323 | unsigned long ncsw; |
70b97a7f | 1324 | struct rq *rq; |
1da177e4 | 1325 | |
3a5c359a AK |
1326 | for (;;) { |
1327 | /* | |
1328 | * We do the initial early heuristics without holding | |
1329 | * any task-queue locks at all. We'll only try to get | |
1330 | * the runqueue lock when things look like they will | |
1331 | * work out! | |
1332 | */ | |
1333 | rq = task_rq(p); | |
fa490cfd | 1334 | |
3a5c359a AK |
1335 | /* |
1336 | * If the task is actively running on another CPU | |
1337 | * still, just relax and busy-wait without holding | |
1338 | * any locks. | |
1339 | * | |
1340 | * NOTE! Since we don't hold any locks, it's not | |
1341 | * even sure that "rq" stays as the right runqueue! | |
1342 | * But we don't care, since "task_running()" will | |
1343 | * return false if the runqueue has changed and p | |
1344 | * is actually now running somewhere else! | |
1345 | */ | |
85ba2d86 RM |
1346 | while (task_running(rq, p)) { |
1347 | if (match_state && unlikely(p->state != match_state)) | |
1348 | return 0; | |
3a5c359a | 1349 | cpu_relax(); |
85ba2d86 | 1350 | } |
fa490cfd | 1351 | |
3a5c359a AK |
1352 | /* |
1353 | * Ok, time to look more closely! We need the rq | |
1354 | * lock now, to be *sure*. If we're wrong, we'll | |
1355 | * just go back and repeat. | |
1356 | */ | |
eb580751 | 1357 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1358 | trace_sched_wait_task(p); |
3a5c359a | 1359 | running = task_running(rq, p); |
da0c1e65 | 1360 | queued = task_on_rq_queued(p); |
85ba2d86 | 1361 | ncsw = 0; |
f31e11d8 | 1362 | if (!match_state || p->state == match_state) |
93dcf55f | 1363 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1364 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1365 | |
85ba2d86 RM |
1366 | /* |
1367 | * If it changed from the expected state, bail out now. | |
1368 | */ | |
1369 | if (unlikely(!ncsw)) | |
1370 | break; | |
1371 | ||
3a5c359a AK |
1372 | /* |
1373 | * Was it really running after all now that we | |
1374 | * checked with the proper locks actually held? | |
1375 | * | |
1376 | * Oops. Go back and try again.. | |
1377 | */ | |
1378 | if (unlikely(running)) { | |
1379 | cpu_relax(); | |
1380 | continue; | |
1381 | } | |
fa490cfd | 1382 | |
3a5c359a AK |
1383 | /* |
1384 | * It's not enough that it's not actively running, | |
1385 | * it must be off the runqueue _entirely_, and not | |
1386 | * preempted! | |
1387 | * | |
80dd99b3 | 1388 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1389 | * running right now), it's preempted, and we should |
1390 | * yield - it could be a while. | |
1391 | */ | |
da0c1e65 | 1392 | if (unlikely(queued)) { |
8b0e1953 | 1393 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
1394 | |
1395 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1396 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1397 | continue; |
1398 | } | |
fa490cfd | 1399 | |
3a5c359a AK |
1400 | /* |
1401 | * Ahh, all good. It wasn't running, and it wasn't | |
1402 | * runnable, which means that it will never become | |
1403 | * running in the future either. We're all done! | |
1404 | */ | |
1405 | break; | |
1406 | } | |
85ba2d86 RM |
1407 | |
1408 | return ncsw; | |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | /*** | |
1412 | * kick_process - kick a running thread to enter/exit the kernel | |
1413 | * @p: the to-be-kicked thread | |
1414 | * | |
1415 | * Cause a process which is running on another CPU to enter | |
1416 | * kernel-mode, without any delay. (to get signals handled.) | |
1417 | * | |
25985edc | 1418 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1419 | * because all it wants to ensure is that the remote task enters |
1420 | * the kernel. If the IPI races and the task has been migrated | |
1421 | * to another CPU then no harm is done and the purpose has been | |
1422 | * achieved as well. | |
1423 | */ | |
36c8b586 | 1424 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1425 | { |
1426 | int cpu; | |
1427 | ||
1428 | preempt_disable(); | |
1429 | cpu = task_cpu(p); | |
1430 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1431 | smp_send_reschedule(cpu); | |
1432 | preempt_enable(); | |
1433 | } | |
b43e3521 | 1434 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1435 | |
30da688e | 1436 | /* |
013fdb80 | 1437 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1438 | * |
1439 | * A few notes on cpu_active vs cpu_online: | |
1440 | * | |
1441 | * - cpu_active must be a subset of cpu_online | |
1442 | * | |
97fb7a0a | 1443 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 1444 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 1445 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
1446 | * see it. |
1447 | * | |
d1ccc66d | 1448 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 1449 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 1450 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
1451 | * off. |
1452 | * | |
1453 | * This means that fallback selection must not select !active CPUs. | |
1454 | * And can assume that any active CPU must be online. Conversely | |
1455 | * select_task_rq() below may allow selection of !active CPUs in order | |
1456 | * to satisfy the above rules. | |
30da688e | 1457 | */ |
5da9a0fb PZ |
1458 | static int select_fallback_rq(int cpu, struct task_struct *p) |
1459 | { | |
aa00d89c TC |
1460 | int nid = cpu_to_node(cpu); |
1461 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
1462 | enum { cpuset, possible, fail } state = cpuset; |
1463 | int dest_cpu; | |
5da9a0fb | 1464 | |
aa00d89c | 1465 | /* |
d1ccc66d IM |
1466 | * If the node that the CPU is on has been offlined, cpu_to_node() |
1467 | * will return -1. There is no CPU on the node, and we should | |
1468 | * select the CPU on the other node. | |
aa00d89c TC |
1469 | */ |
1470 | if (nid != -1) { | |
1471 | nodemask = cpumask_of_node(nid); | |
1472 | ||
1473 | /* Look for allowed, online CPU in same node. */ | |
1474 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
1475 | if (!cpu_active(dest_cpu)) |
1476 | continue; | |
0c98d344 | 1477 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
aa00d89c TC |
1478 | return dest_cpu; |
1479 | } | |
2baab4e9 | 1480 | } |
5da9a0fb | 1481 | |
2baab4e9 PZ |
1482 | for (;;) { |
1483 | /* Any allowed, online CPU? */ | |
0c98d344 | 1484 | for_each_cpu(dest_cpu, &p->cpus_allowed) { |
175f0e25 | 1485 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 1486 | continue; |
175f0e25 | 1487 | |
2baab4e9 PZ |
1488 | goto out; |
1489 | } | |
5da9a0fb | 1490 | |
e73e85f0 | 1491 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
1492 | switch (state) { |
1493 | case cpuset: | |
e73e85f0 ON |
1494 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
1495 | cpuset_cpus_allowed_fallback(p); | |
1496 | state = possible; | |
1497 | break; | |
1498 | } | |
d1ccc66d | 1499 | /* Fall-through */ |
2baab4e9 PZ |
1500 | case possible: |
1501 | do_set_cpus_allowed(p, cpu_possible_mask); | |
1502 | state = fail; | |
1503 | break; | |
1504 | ||
1505 | case fail: | |
1506 | BUG(); | |
1507 | break; | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | out: | |
1512 | if (state != cpuset) { | |
1513 | /* | |
1514 | * Don't tell them about moving exiting tasks or | |
1515 | * kernel threads (both mm NULL), since they never | |
1516 | * leave kernel. | |
1517 | */ | |
1518 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 1519 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
1520 | task_pid_nr(p), p->comm, cpu); |
1521 | } | |
5da9a0fb PZ |
1522 | } |
1523 | ||
1524 | return dest_cpu; | |
1525 | } | |
1526 | ||
e2912009 | 1527 | /* |
013fdb80 | 1528 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 1529 | */ |
970b13ba | 1530 | static inline |
ac66f547 | 1531 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 1532 | { |
cbce1a68 PZ |
1533 | lockdep_assert_held(&p->pi_lock); |
1534 | ||
4b53a341 | 1535 | if (p->nr_cpus_allowed > 1) |
6c1d9410 | 1536 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 | 1537 | else |
0c98d344 | 1538 | cpu = cpumask_any(&p->cpus_allowed); |
e2912009 PZ |
1539 | |
1540 | /* | |
1541 | * In order not to call set_task_cpu() on a blocking task we need | |
1542 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
d1ccc66d | 1543 | * CPU. |
e2912009 PZ |
1544 | * |
1545 | * Since this is common to all placement strategies, this lives here. | |
1546 | * | |
1547 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
1548 | * not worry about this generic constraint ] | |
1549 | */ | |
7af443ee | 1550 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 1551 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
1552 | |
1553 | return cpu; | |
970b13ba | 1554 | } |
09a40af5 MG |
1555 | |
1556 | static void update_avg(u64 *avg, u64 sample) | |
1557 | { | |
1558 | s64 diff = sample - *avg; | |
1559 | *avg += diff >> 3; | |
1560 | } | |
25834c73 | 1561 | |
f5832c19 NP |
1562 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
1563 | { | |
1564 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
1565 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
1566 | ||
1567 | if (stop) { | |
1568 | /* | |
1569 | * Make it appear like a SCHED_FIFO task, its something | |
1570 | * userspace knows about and won't get confused about. | |
1571 | * | |
1572 | * Also, it will make PI more or less work without too | |
1573 | * much confusion -- but then, stop work should not | |
1574 | * rely on PI working anyway. | |
1575 | */ | |
1576 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
1577 | ||
1578 | stop->sched_class = &stop_sched_class; | |
1579 | } | |
1580 | ||
1581 | cpu_rq(cpu)->stop = stop; | |
1582 | ||
1583 | if (old_stop) { | |
1584 | /* | |
1585 | * Reset it back to a normal scheduling class so that | |
1586 | * it can die in pieces. | |
1587 | */ | |
1588 | old_stop->sched_class = &rt_sched_class; | |
1589 | } | |
1590 | } | |
1591 | ||
25834c73 PZ |
1592 | #else |
1593 | ||
1594 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
1595 | const struct cpumask *new_mask, bool check) | |
1596 | { | |
1597 | return set_cpus_allowed_ptr(p, new_mask); | |
1598 | } | |
1599 | ||
5cc389bc | 1600 | #endif /* CONFIG_SMP */ |
970b13ba | 1601 | |
d7c01d27 | 1602 | static void |
b84cb5df | 1603 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 1604 | { |
4fa8d299 | 1605 | struct rq *rq; |
b84cb5df | 1606 | |
4fa8d299 JP |
1607 | if (!schedstat_enabled()) |
1608 | return; | |
1609 | ||
1610 | rq = this_rq(); | |
d7c01d27 | 1611 | |
4fa8d299 JP |
1612 | #ifdef CONFIG_SMP |
1613 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
1614 | __schedstat_inc(rq->ttwu_local); |
1615 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
1616 | } else { |
1617 | struct sched_domain *sd; | |
1618 | ||
b85c8b71 | 1619 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 1620 | rcu_read_lock(); |
4fa8d299 | 1621 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 1622 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 1623 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
1624 | break; |
1625 | } | |
1626 | } | |
057f3fad | 1627 | rcu_read_unlock(); |
d7c01d27 | 1628 | } |
f339b9dc PZ |
1629 | |
1630 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 1631 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
1632 | #endif /* CONFIG_SMP */ |
1633 | ||
b85c8b71 PZ |
1634 | __schedstat_inc(rq->ttwu_count); |
1635 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
1636 | |
1637 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 1638 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
1639 | } |
1640 | ||
1de64443 | 1641 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
d7c01d27 | 1642 | { |
9ed3811a | 1643 | activate_task(rq, p, en_flags); |
da0c1e65 | 1644 | p->on_rq = TASK_ON_RQ_QUEUED; |
c2f7115e | 1645 | |
d1ccc66d | 1646 | /* If a worker is waking up, notify the workqueue: */ |
c2f7115e PZ |
1647 | if (p->flags & PF_WQ_WORKER) |
1648 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
1649 | } |
1650 | ||
23f41eeb PZ |
1651 | /* |
1652 | * Mark the task runnable and perform wakeup-preemption. | |
1653 | */ | |
e7904a28 | 1654 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1655 | struct rq_flags *rf) |
9ed3811a | 1656 | { |
9ed3811a | 1657 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 1658 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
1659 | trace_sched_wakeup(p); |
1660 | ||
9ed3811a | 1661 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
1662 | if (p->sched_class->task_woken) { |
1663 | /* | |
cbce1a68 PZ |
1664 | * Our task @p is fully woken up and running; so its safe to |
1665 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 1666 | */ |
d8ac8971 | 1667 | rq_unpin_lock(rq, rf); |
9ed3811a | 1668 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 1669 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 1670 | } |
9ed3811a | 1671 | |
e69c6341 | 1672 | if (rq->idle_stamp) { |
78becc27 | 1673 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 1674 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 1675 | |
abfafa54 JL |
1676 | update_avg(&rq->avg_idle, delta); |
1677 | ||
1678 | if (rq->avg_idle > max) | |
9ed3811a | 1679 | rq->avg_idle = max; |
abfafa54 | 1680 | |
9ed3811a TH |
1681 | rq->idle_stamp = 0; |
1682 | } | |
1683 | #endif | |
1684 | } | |
1685 | ||
c05fbafb | 1686 | static void |
e7904a28 | 1687 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 1688 | struct rq_flags *rf) |
c05fbafb | 1689 | { |
77558e4d | 1690 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 1691 | |
cbce1a68 PZ |
1692 | lockdep_assert_held(&rq->lock); |
1693 | ||
c05fbafb PZ |
1694 | #ifdef CONFIG_SMP |
1695 | if (p->sched_contributes_to_load) | |
1696 | rq->nr_uninterruptible--; | |
b5179ac7 | 1697 | |
b5179ac7 | 1698 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 1699 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
1700 | #endif |
1701 | ||
b5179ac7 | 1702 | ttwu_activate(rq, p, en_flags); |
d8ac8971 | 1703 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
1704 | } |
1705 | ||
1706 | /* | |
1707 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
1708 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
1709 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
1710 | * the task is still ->on_rq. | |
1711 | */ | |
1712 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
1713 | { | |
eb580751 | 1714 | struct rq_flags rf; |
c05fbafb PZ |
1715 | struct rq *rq; |
1716 | int ret = 0; | |
1717 | ||
eb580751 | 1718 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 1719 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
1720 | /* check_preempt_curr() may use rq clock */ |
1721 | update_rq_clock(rq); | |
d8ac8971 | 1722 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
1723 | ret = 1; |
1724 | } | |
eb580751 | 1725 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
1726 | |
1727 | return ret; | |
1728 | } | |
1729 | ||
317f3941 | 1730 | #ifdef CONFIG_SMP |
e3baac47 | 1731 | void sched_ttwu_pending(void) |
317f3941 PZ |
1732 | { |
1733 | struct rq *rq = this_rq(); | |
fa14ff4a | 1734 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
73215849 | 1735 | struct task_struct *p, *t; |
d8ac8971 | 1736 | struct rq_flags rf; |
317f3941 | 1737 | |
e3baac47 PZ |
1738 | if (!llist) |
1739 | return; | |
1740 | ||
8a8c69c3 | 1741 | rq_lock_irqsave(rq, &rf); |
77558e4d | 1742 | update_rq_clock(rq); |
317f3941 | 1743 | |
73215849 BP |
1744 | llist_for_each_entry_safe(p, t, llist, wake_entry) |
1745 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); | |
317f3941 | 1746 | |
8a8c69c3 | 1747 | rq_unlock_irqrestore(rq, &rf); |
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); | |
8a8c69c3 | 1805 | struct rq_flags rf; |
f6be8af1 | 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 { | |
8a8c69c3 | 1815 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
1816 | if (is_idle_task(rq->curr)) |
1817 | smp_send_reschedule(cpu); | |
d1ccc66d | 1818 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 1819 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 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 | ||
8a8c69c3 | 1845 | rq_lock(rq, &rf); |
77558e4d | 1846 | update_rq_clock(rq); |
d8ac8971 | 1847 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 1848 | rq_unlock(rq, &rf); |
9ed3811a TH |
1849 | } |
1850 | ||
8643cda5 PZ |
1851 | /* |
1852 | * Notes on Program-Order guarantees on SMP systems. | |
1853 | * | |
1854 | * MIGRATION | |
1855 | * | |
1856 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
1857 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
1858 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
1859 | * |
1860 | * For migration (of runnable tasks) this is provided by the following means: | |
1861 | * | |
1862 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
1863 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
1864 | * rq(c1)->lock (if not at the same time, then in that order). | |
1865 | * C) LOCK of the rq(c1)->lock scheduling in task | |
1866 | * | |
7696f991 | 1867 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 1868 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
1869 | * |
1870 | * Example: | |
1871 | * | |
1872 | * CPU0 CPU1 CPU2 | |
1873 | * | |
1874 | * LOCK rq(0)->lock | |
1875 | * sched-out X | |
1876 | * sched-in Y | |
1877 | * UNLOCK rq(0)->lock | |
1878 | * | |
1879 | * LOCK rq(0)->lock // orders against CPU0 | |
1880 | * dequeue X | |
1881 | * UNLOCK rq(0)->lock | |
1882 | * | |
1883 | * LOCK rq(1)->lock | |
1884 | * enqueue X | |
1885 | * UNLOCK rq(1)->lock | |
1886 | * | |
1887 | * LOCK rq(1)->lock // orders against CPU2 | |
1888 | * sched-out Z | |
1889 | * sched-in X | |
1890 | * UNLOCK rq(1)->lock | |
1891 | * | |
1892 | * | |
1893 | * BLOCKING -- aka. SLEEP + WAKEUP | |
1894 | * | |
1895 | * For blocking we (obviously) need to provide the same guarantee as for | |
1896 | * migration. However the means are completely different as there is no lock | |
1897 | * chain to provide order. Instead we do: | |
1898 | * | |
1899 | * 1) smp_store_release(X->on_cpu, 0) | |
1f03e8d2 | 1900 | * 2) smp_cond_load_acquire(!X->on_cpu) |
8643cda5 PZ |
1901 | * |
1902 | * Example: | |
1903 | * | |
1904 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
1905 | * | |
1906 | * LOCK rq(0)->lock LOCK X->pi_lock | |
1907 | * dequeue X | |
1908 | * sched-out X | |
1909 | * smp_store_release(X->on_cpu, 0); | |
1910 | * | |
1f03e8d2 | 1911 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
1912 | * X->state = WAKING |
1913 | * set_task_cpu(X,2) | |
1914 | * | |
1915 | * LOCK rq(2)->lock | |
1916 | * enqueue X | |
1917 | * X->state = RUNNING | |
1918 | * UNLOCK rq(2)->lock | |
1919 | * | |
1920 | * LOCK rq(2)->lock // orders against CPU1 | |
1921 | * sched-out Z | |
1922 | * sched-in X | |
1923 | * UNLOCK rq(2)->lock | |
1924 | * | |
1925 | * UNLOCK X->pi_lock | |
1926 | * UNLOCK rq(0)->lock | |
1927 | * | |
1928 | * | |
7696f991 AP |
1929 | * However, for wakeups there is a second guarantee we must provide, namely we |
1930 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
1931 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
1932 | */ |
1933 | ||
9ed3811a | 1934 | /** |
1da177e4 | 1935 | * try_to_wake_up - wake up a thread |
9ed3811a | 1936 | * @p: the thread to be awakened |
1da177e4 | 1937 | * @state: the mask of task states that can be woken |
9ed3811a | 1938 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 1939 | * |
a2250238 | 1940 | * If (@state & @p->state) @p->state = TASK_RUNNING. |
1da177e4 | 1941 | * |
a2250238 PZ |
1942 | * If the task was not queued/runnable, also place it back on a runqueue. |
1943 | * | |
1944 | * Atomic against schedule() which would dequeue a task, also see | |
1945 | * set_current_state(). | |
1946 | * | |
7696f991 AP |
1947 | * This function executes a full memory barrier before accessing the task |
1948 | * state; see set_current_state(). | |
1949 | * | |
a2250238 PZ |
1950 | * Return: %true if @p->state changes (an actual wakeup was done), |
1951 | * %false otherwise. | |
1da177e4 | 1952 | */ |
e4a52bcb PZ |
1953 | static int |
1954 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 1955 | { |
1da177e4 | 1956 | unsigned long flags; |
c05fbafb | 1957 | int cpu, success = 0; |
2398f2c6 | 1958 | |
e0acd0a6 ON |
1959 | /* |
1960 | * If we are going to wake up a thread waiting for CONDITION we | |
1961 | * need to ensure that CONDITION=1 done by the caller can not be | |
1962 | * reordered with p->state check below. This pairs with mb() in | |
1963 | * set_current_state() the waiting thread does. | |
1964 | */ | |
013fdb80 | 1965 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 1966 | smp_mb__after_spinlock(); |
e9c84311 | 1967 | if (!(p->state & state)) |
1da177e4 LT |
1968 | goto out; |
1969 | ||
fbd705a0 PZ |
1970 | trace_sched_waking(p); |
1971 | ||
d1ccc66d IM |
1972 | /* We're going to change ->state: */ |
1973 | success = 1; | |
1da177e4 | 1974 | cpu = task_cpu(p); |
1da177e4 | 1975 | |
135e8c92 BS |
1976 | /* |
1977 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
1978 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
1979 | * in smp_cond_load_acquire() below. | |
1980 | * | |
3d85b270 AP |
1981 | * sched_ttwu_pending() try_to_wake_up() |
1982 | * STORE p->on_rq = 1 LOAD p->state | |
1983 | * UNLOCK rq->lock | |
1984 | * | |
1985 | * __schedule() (switch to task 'p') | |
1986 | * LOCK rq->lock smp_rmb(); | |
1987 | * smp_mb__after_spinlock(); | |
1988 | * UNLOCK rq->lock | |
135e8c92 BS |
1989 | * |
1990 | * [task p] | |
3d85b270 | 1991 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 1992 | * |
3d85b270 AP |
1993 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
1994 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
135e8c92 BS |
1995 | */ |
1996 | smp_rmb(); | |
c05fbafb PZ |
1997 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
1998 | goto stat; | |
1da177e4 | 1999 | |
1da177e4 | 2000 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2001 | /* |
2002 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2003 | * possible to, falsely, observe p->on_cpu == 0. | |
2004 | * | |
2005 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2006 | * from the runqueue. | |
2007 | * | |
3d85b270 AP |
2008 | * __schedule() (switch to task 'p') try_to_wake_up() |
2009 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
2010 | * UNLOCK rq->lock | |
2011 | * | |
2012 | * __schedule() (put 'p' to sleep) | |
2013 | * LOCK rq->lock smp_rmb(); | |
2014 | * smp_mb__after_spinlock(); | |
2015 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 2016 | * |
3d85b270 AP |
2017 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
2018 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
ecf7d01c PZ |
2019 | */ |
2020 | smp_rmb(); | |
2021 | ||
e9c84311 | 2022 | /* |
d1ccc66d | 2023 | * If the owning (remote) CPU is still in the middle of schedule() with |
c05fbafb | 2024 | * this task as prev, wait until its done referencing the task. |
b75a2253 | 2025 | * |
31cb1bc0 | 2026 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
2027 | * |
2028 | * This ensures that tasks getting woken will be fully ordered against | |
2029 | * their previous state and preserve Program Order. | |
0970d299 | 2030 | */ |
1f03e8d2 | 2031 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2032 | |
a8e4f2ea | 2033 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2034 | p->state = TASK_WAKING; |
e7693a36 | 2035 | |
e33a9bba | 2036 | if (p->in_iowait) { |
c96f5471 | 2037 | delayacct_blkio_end(p); |
e33a9bba TH |
2038 | atomic_dec(&task_rq(p)->nr_iowait); |
2039 | } | |
2040 | ||
ac66f547 | 2041 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2042 | if (task_cpu(p) != cpu) { |
2043 | wake_flags |= WF_MIGRATED; | |
eb414681 | 2044 | psi_ttwu_dequeue(p); |
e4a52bcb | 2045 | set_task_cpu(p, cpu); |
f339b9dc | 2046 | } |
e33a9bba TH |
2047 | |
2048 | #else /* CONFIG_SMP */ | |
2049 | ||
2050 | if (p->in_iowait) { | |
c96f5471 | 2051 | delayacct_blkio_end(p); |
e33a9bba TH |
2052 | atomic_dec(&task_rq(p)->nr_iowait); |
2053 | } | |
2054 | ||
1da177e4 | 2055 | #endif /* CONFIG_SMP */ |
1da177e4 | 2056 | |
b5179ac7 | 2057 | ttwu_queue(p, cpu, wake_flags); |
c05fbafb | 2058 | stat: |
4fa8d299 | 2059 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2060 | out: |
013fdb80 | 2061 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2062 | |
2063 | return success; | |
2064 | } | |
2065 | ||
21aa9af0 TH |
2066 | /** |
2067 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2068 | * @p: the thread to be awakened | |
bf50f0e8 | 2069 | * @rf: request-queue flags for pinning |
21aa9af0 | 2070 | * |
2acca55e | 2071 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2072 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2073 | * the current task. |
21aa9af0 | 2074 | */ |
d8ac8971 | 2075 | static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf) |
21aa9af0 TH |
2076 | { |
2077 | struct rq *rq = task_rq(p); | |
21aa9af0 | 2078 | |
383efcd0 TH |
2079 | if (WARN_ON_ONCE(rq != this_rq()) || |
2080 | WARN_ON_ONCE(p == current)) | |
2081 | return; | |
2082 | ||
21aa9af0 TH |
2083 | lockdep_assert_held(&rq->lock); |
2084 | ||
2acca55e | 2085 | if (!raw_spin_trylock(&p->pi_lock)) { |
cbce1a68 PZ |
2086 | /* |
2087 | * This is OK, because current is on_cpu, which avoids it being | |
2088 | * picked for load-balance and preemption/IRQs are still | |
2089 | * disabled avoiding further scheduler activity on it and we've | |
2090 | * not yet picked a replacement task. | |
2091 | */ | |
8a8c69c3 | 2092 | rq_unlock(rq, rf); |
2acca55e | 2093 | raw_spin_lock(&p->pi_lock); |
8a8c69c3 | 2094 | rq_relock(rq, rf); |
2acca55e PZ |
2095 | } |
2096 | ||
21aa9af0 | 2097 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2098 | goto out; |
21aa9af0 | 2099 | |
fbd705a0 PZ |
2100 | trace_sched_waking(p); |
2101 | ||
e33a9bba TH |
2102 | if (!task_on_rq_queued(p)) { |
2103 | if (p->in_iowait) { | |
c96f5471 | 2104 | delayacct_blkio_end(p); |
e33a9bba TH |
2105 | atomic_dec(&rq->nr_iowait); |
2106 | } | |
bce4dc80 | 2107 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK); |
e33a9bba | 2108 | } |
d7c01d27 | 2109 | |
d8ac8971 | 2110 | ttwu_do_wakeup(rq, p, 0, rf); |
4fa8d299 | 2111 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2112 | out: |
2113 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2114 | } |
2115 | ||
50fa610a DH |
2116 | /** |
2117 | * wake_up_process - Wake up a specific process | |
2118 | * @p: The process to be woken up. | |
2119 | * | |
2120 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2121 | * processes. |
2122 | * | |
2123 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 2124 | * |
7696f991 | 2125 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 2126 | */ |
7ad5b3a5 | 2127 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2128 | { |
9067ac85 | 2129 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2130 | } |
1da177e4 LT |
2131 | EXPORT_SYMBOL(wake_up_process); |
2132 | ||
7ad5b3a5 | 2133 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2134 | { |
2135 | return try_to_wake_up(p, state, 0); | |
2136 | } | |
2137 | ||
1da177e4 LT |
2138 | /* |
2139 | * Perform scheduler related setup for a newly forked process p. | |
2140 | * p is forked by current. | |
dd41f596 IM |
2141 | * |
2142 | * __sched_fork() is basic setup used by init_idle() too: | |
2143 | */ | |
5e1576ed | 2144 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2145 | { |
fd2f4419 PZ |
2146 | p->on_rq = 0; |
2147 | ||
2148 | p->se.on_rq = 0; | |
dd41f596 IM |
2149 | p->se.exec_start = 0; |
2150 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2151 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2152 | p->se.nr_migrations = 0; |
da7a735e | 2153 | p->se.vruntime = 0; |
fd2f4419 | 2154 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2155 | |
ad936d86 BP |
2156 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2157 | p->se.cfs_rq = NULL; | |
2158 | #endif | |
2159 | ||
6cfb0d5d | 2160 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2161 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2162 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2163 | #endif |
476d139c | 2164 | |
aab03e05 | 2165 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2166 | init_dl_task_timer(&p->dl); |
209a0cbd | 2167 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 2168 | __dl_clear_params(p); |
aab03e05 | 2169 | |
fa717060 | 2170 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2171 | p->rt.timeout = 0; |
2172 | p->rt.time_slice = sched_rr_timeslice; | |
2173 | p->rt.on_rq = 0; | |
2174 | p->rt.on_list = 0; | |
476d139c | 2175 | |
e107be36 AK |
2176 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2177 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2178 | #endif | |
cbee9f88 | 2179 | |
13784475 | 2180 | init_numa_balancing(clone_flags, p); |
dd41f596 IM |
2181 | } |
2182 | ||
2a595721 SD |
2183 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2184 | ||
1a687c2e | 2185 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2186 | |
1a687c2e MG |
2187 | void set_numabalancing_state(bool enabled) |
2188 | { | |
2189 | if (enabled) | |
2a595721 | 2190 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2191 | else |
2a595721 | 2192 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2193 | } |
54a43d54 AK |
2194 | |
2195 | #ifdef CONFIG_PROC_SYSCTL | |
2196 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
2197 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2198 | { | |
2199 | struct ctl_table t; | |
2200 | int err; | |
2a595721 | 2201 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2202 | |
2203 | if (write && !capable(CAP_SYS_ADMIN)) | |
2204 | return -EPERM; | |
2205 | ||
2206 | t = *table; | |
2207 | t.data = &state; | |
2208 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2209 | if (err < 0) | |
2210 | return err; | |
2211 | if (write) | |
2212 | set_numabalancing_state(state); | |
2213 | return err; | |
2214 | } | |
2215 | #endif | |
2216 | #endif | |
dd41f596 | 2217 | |
4698f88c JP |
2218 | #ifdef CONFIG_SCHEDSTATS |
2219 | ||
cb251765 | 2220 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 2221 | static bool __initdata __sched_schedstats = false; |
cb251765 | 2222 | |
cb251765 MG |
2223 | static void set_schedstats(bool enabled) |
2224 | { | |
2225 | if (enabled) | |
2226 | static_branch_enable(&sched_schedstats); | |
2227 | else | |
2228 | static_branch_disable(&sched_schedstats); | |
2229 | } | |
2230 | ||
2231 | void force_schedstat_enabled(void) | |
2232 | { | |
2233 | if (!schedstat_enabled()) { | |
2234 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2235 | static_branch_enable(&sched_schedstats); | |
2236 | } | |
2237 | } | |
2238 | ||
2239 | static int __init setup_schedstats(char *str) | |
2240 | { | |
2241 | int ret = 0; | |
2242 | if (!str) | |
2243 | goto out; | |
2244 | ||
4698f88c JP |
2245 | /* |
2246 | * This code is called before jump labels have been set up, so we can't | |
2247 | * change the static branch directly just yet. Instead set a temporary | |
2248 | * variable so init_schedstats() can do it later. | |
2249 | */ | |
cb251765 | 2250 | if (!strcmp(str, "enable")) { |
4698f88c | 2251 | __sched_schedstats = true; |
cb251765 MG |
2252 | ret = 1; |
2253 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 2254 | __sched_schedstats = false; |
cb251765 MG |
2255 | ret = 1; |
2256 | } | |
2257 | out: | |
2258 | if (!ret) | |
2259 | pr_warn("Unable to parse schedstats=\n"); | |
2260 | ||
2261 | return ret; | |
2262 | } | |
2263 | __setup("schedstats=", setup_schedstats); | |
2264 | ||
4698f88c JP |
2265 | static void __init init_schedstats(void) |
2266 | { | |
2267 | set_schedstats(__sched_schedstats); | |
2268 | } | |
2269 | ||
cb251765 MG |
2270 | #ifdef CONFIG_PROC_SYSCTL |
2271 | int sysctl_schedstats(struct ctl_table *table, int write, | |
2272 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2273 | { | |
2274 | struct ctl_table t; | |
2275 | int err; | |
2276 | int state = static_branch_likely(&sched_schedstats); | |
2277 | ||
2278 | if (write && !capable(CAP_SYS_ADMIN)) | |
2279 | return -EPERM; | |
2280 | ||
2281 | t = *table; | |
2282 | t.data = &state; | |
2283 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2284 | if (err < 0) | |
2285 | return err; | |
2286 | if (write) | |
2287 | set_schedstats(state); | |
2288 | return err; | |
2289 | } | |
4698f88c JP |
2290 | #endif /* CONFIG_PROC_SYSCTL */ |
2291 | #else /* !CONFIG_SCHEDSTATS */ | |
2292 | static inline void init_schedstats(void) {} | |
2293 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
2294 | |
2295 | /* | |
2296 | * fork()/clone()-time setup: | |
2297 | */ | |
aab03e05 | 2298 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2299 | { |
0122ec5b | 2300 | unsigned long flags; |
dd41f596 | 2301 | |
5e1576ed | 2302 | __sched_fork(clone_flags, p); |
06b83b5f | 2303 | /* |
7dc603c9 | 2304 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
2305 | * nobody will actually run it, and a signal or other external |
2306 | * event cannot wake it up and insert it on the runqueue either. | |
2307 | */ | |
7dc603c9 | 2308 | p->state = TASK_NEW; |
dd41f596 | 2309 | |
c350a04e MG |
2310 | /* |
2311 | * Make sure we do not leak PI boosting priority to the child. | |
2312 | */ | |
2313 | p->prio = current->normal_prio; | |
2314 | ||
b9dc29e7 MG |
2315 | /* |
2316 | * Revert to default priority/policy on fork if requested. | |
2317 | */ | |
2318 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2319 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2320 | p->policy = SCHED_NORMAL; |
6c697bdf | 2321 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2322 | p->rt_priority = 0; |
2323 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2324 | p->static_prio = NICE_TO_PRIO(0); | |
2325 | ||
2326 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 2327 | set_load_weight(p, false); |
6c697bdf | 2328 | |
b9dc29e7 MG |
2329 | /* |
2330 | * We don't need the reset flag anymore after the fork. It has | |
2331 | * fulfilled its duty: | |
2332 | */ | |
2333 | p->sched_reset_on_fork = 0; | |
2334 | } | |
ca94c442 | 2335 | |
af0fffd9 | 2336 | if (dl_prio(p->prio)) |
aab03e05 | 2337 | return -EAGAIN; |
af0fffd9 | 2338 | else if (rt_prio(p->prio)) |
aab03e05 | 2339 | p->sched_class = &rt_sched_class; |
af0fffd9 | 2340 | else |
2ddbf952 | 2341 | p->sched_class = &fair_sched_class; |
b29739f9 | 2342 | |
7dc603c9 | 2343 | init_entity_runnable_average(&p->se); |
cd29fe6f | 2344 | |
86951599 PZ |
2345 | /* |
2346 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2347 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2348 | * is ran before sched_fork(). | |
2349 | * | |
2350 | * Silence PROVE_RCU. | |
2351 | */ | |
0122ec5b | 2352 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e210bffd | 2353 | /* |
d1ccc66d | 2354 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
2355 | * so use __set_task_cpu(). |
2356 | */ | |
af0fffd9 | 2357 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
2358 | if (p->sched_class->task_fork) |
2359 | p->sched_class->task_fork(p); | |
0122ec5b | 2360 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2361 | |
f6db8347 | 2362 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2363 | if (likely(sched_info_on())) |
52f17b6c | 2364 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2365 | #endif |
3ca7a440 PZ |
2366 | #if defined(CONFIG_SMP) |
2367 | p->on_cpu = 0; | |
4866cde0 | 2368 | #endif |
01028747 | 2369 | init_task_preempt_count(p); |
806c09a7 | 2370 | #ifdef CONFIG_SMP |
917b627d | 2371 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2372 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2373 | #endif |
aab03e05 | 2374 | return 0; |
1da177e4 LT |
2375 | } |
2376 | ||
332ac17e DF |
2377 | unsigned long to_ratio(u64 period, u64 runtime) |
2378 | { | |
2379 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 2380 | return BW_UNIT; |
332ac17e DF |
2381 | |
2382 | /* | |
2383 | * Doing this here saves a lot of checks in all | |
2384 | * the calling paths, and returning zero seems | |
2385 | * safe for them anyway. | |
2386 | */ | |
2387 | if (period == 0) | |
2388 | return 0; | |
2389 | ||
c52f14d3 | 2390 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
2391 | } |
2392 | ||
1da177e4 LT |
2393 | /* |
2394 | * wake_up_new_task - wake up a newly created task for the first time. | |
2395 | * | |
2396 | * This function will do some initial scheduler statistics housekeeping | |
2397 | * that must be done for every newly created context, then puts the task | |
2398 | * on the runqueue and wakes it. | |
2399 | */ | |
3e51e3ed | 2400 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2401 | { |
eb580751 | 2402 | struct rq_flags rf; |
dd41f596 | 2403 | struct rq *rq; |
fabf318e | 2404 | |
eb580751 | 2405 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 2406 | p->state = TASK_RUNNING; |
fabf318e PZ |
2407 | #ifdef CONFIG_SMP |
2408 | /* | |
2409 | * Fork balancing, do it here and not earlier because: | |
2410 | * - cpus_allowed can change in the fork path | |
d1ccc66d | 2411 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
2412 | * |
2413 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
2414 | * as we're not fully set-up yet. | |
fabf318e | 2415 | */ |
32e839dd | 2416 | p->recent_used_cpu = task_cpu(p); |
e210bffd | 2417 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 2418 | #endif |
b7fa30c9 | 2419 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 2420 | update_rq_clock(rq); |
2b8c41da | 2421 | post_init_entity_util_avg(&p->se); |
0017d735 | 2422 | |
7a57f32a | 2423 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
da0c1e65 | 2424 | p->on_rq = TASK_ON_RQ_QUEUED; |
fbd705a0 | 2425 | trace_sched_wakeup_new(p); |
a7558e01 | 2426 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2427 | #ifdef CONFIG_SMP |
0aaafaab PZ |
2428 | if (p->sched_class->task_woken) { |
2429 | /* | |
2430 | * Nothing relies on rq->lock after this, so its fine to | |
2431 | * drop it. | |
2432 | */ | |
d8ac8971 | 2433 | rq_unpin_lock(rq, &rf); |
efbbd05a | 2434 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2435 | rq_repin_lock(rq, &rf); |
0aaafaab | 2436 | } |
9a897c5a | 2437 | #endif |
eb580751 | 2438 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
2439 | } |
2440 | ||
e107be36 AK |
2441 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2442 | ||
b7203428 | 2443 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 2444 | |
2ecd9d29 PZ |
2445 | void preempt_notifier_inc(void) |
2446 | { | |
b7203428 | 2447 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
2448 | } |
2449 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
2450 | ||
2451 | void preempt_notifier_dec(void) | |
2452 | { | |
b7203428 | 2453 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
2454 | } |
2455 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
2456 | ||
e107be36 | 2457 | /** |
80dd99b3 | 2458 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2459 | * @notifier: notifier struct to register |
e107be36 AK |
2460 | */ |
2461 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2462 | { | |
b7203428 | 2463 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
2464 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
2465 | ||
e107be36 AK |
2466 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
2467 | } | |
2468 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2469 | ||
2470 | /** | |
2471 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2472 | * @notifier: notifier struct to unregister |
e107be36 | 2473 | * |
d84525a8 | 2474 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
2475 | */ |
2476 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2477 | { | |
2478 | hlist_del(¬ifier->link); | |
2479 | } | |
2480 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2481 | ||
1cde2930 | 2482 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2483 | { |
2484 | struct preempt_notifier *notifier; | |
e107be36 | 2485 | |
b67bfe0d | 2486 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2487 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
2488 | } | |
2489 | ||
1cde2930 PZ |
2490 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
2491 | { | |
b7203428 | 2492 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
2493 | __fire_sched_in_preempt_notifiers(curr); |
2494 | } | |
2495 | ||
e107be36 | 2496 | static void |
1cde2930 PZ |
2497 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2498 | struct task_struct *next) | |
e107be36 AK |
2499 | { |
2500 | struct preempt_notifier *notifier; | |
e107be36 | 2501 | |
b67bfe0d | 2502 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
2503 | notifier->ops->sched_out(notifier, next); |
2504 | } | |
2505 | ||
1cde2930 PZ |
2506 | static __always_inline void |
2507 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2508 | struct task_struct *next) | |
2509 | { | |
b7203428 | 2510 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
2511 | __fire_sched_out_preempt_notifiers(curr, next); |
2512 | } | |
2513 | ||
6d6bc0ad | 2514 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2515 | |
1cde2930 | 2516 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
2517 | { |
2518 | } | |
2519 | ||
1cde2930 | 2520 | static inline void |
e107be36 AK |
2521 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
2522 | struct task_struct *next) | |
2523 | { | |
2524 | } | |
2525 | ||
6d6bc0ad | 2526 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2527 | |
31cb1bc0 | 2528 | static inline void prepare_task(struct task_struct *next) |
2529 | { | |
2530 | #ifdef CONFIG_SMP | |
2531 | /* | |
2532 | * Claim the task as running, we do this before switching to it | |
2533 | * such that any running task will have this set. | |
2534 | */ | |
2535 | next->on_cpu = 1; | |
2536 | #endif | |
2537 | } | |
2538 | ||
2539 | static inline void finish_task(struct task_struct *prev) | |
2540 | { | |
2541 | #ifdef CONFIG_SMP | |
2542 | /* | |
2543 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
2544 | * We must ensure this doesn't happen until the switch is completely | |
2545 | * finished. | |
2546 | * | |
2547 | * In particular, the load of prev->state in finish_task_switch() must | |
2548 | * happen before this. | |
2549 | * | |
2550 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
2551 | */ | |
2552 | smp_store_release(&prev->on_cpu, 0); | |
2553 | #endif | |
2554 | } | |
2555 | ||
269d5992 PZ |
2556 | static inline void |
2557 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 2558 | { |
269d5992 PZ |
2559 | /* |
2560 | * Since the runqueue lock will be released by the next | |
2561 | * task (which is an invalid locking op but in the case | |
2562 | * of the scheduler it's an obvious special-case), so we | |
2563 | * do an early lockdep release here: | |
2564 | */ | |
2565 | rq_unpin_lock(rq, rf); | |
2566 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); | |
31cb1bc0 | 2567 | #ifdef CONFIG_DEBUG_SPINLOCK |
2568 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 2569 | rq->lock.owner = next; |
31cb1bc0 | 2570 | #endif |
269d5992 PZ |
2571 | } |
2572 | ||
2573 | static inline void finish_lock_switch(struct rq *rq) | |
2574 | { | |
31cb1bc0 | 2575 | /* |
2576 | * If we are tracking spinlock dependencies then we have to | |
2577 | * fix up the runqueue lock - which gets 'carried over' from | |
2578 | * prev into current: | |
2579 | */ | |
2580 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
31cb1bc0 | 2581 | raw_spin_unlock_irq(&rq->lock); |
2582 | } | |
2583 | ||
325ea10c IM |
2584 | /* |
2585 | * NOP if the arch has not defined these: | |
2586 | */ | |
2587 | ||
2588 | #ifndef prepare_arch_switch | |
2589 | # define prepare_arch_switch(next) do { } while (0) | |
2590 | #endif | |
2591 | ||
2592 | #ifndef finish_arch_post_lock_switch | |
2593 | # define finish_arch_post_lock_switch() do { } while (0) | |
2594 | #endif | |
2595 | ||
4866cde0 NP |
2596 | /** |
2597 | * prepare_task_switch - prepare to switch tasks | |
2598 | * @rq: the runqueue preparing to switch | |
421cee29 | 2599 | * @prev: the current task that is being switched out |
4866cde0 NP |
2600 | * @next: the task we are going to switch to. |
2601 | * | |
2602 | * This is called with the rq lock held and interrupts off. It must | |
2603 | * be paired with a subsequent finish_task_switch after the context | |
2604 | * switch. | |
2605 | * | |
2606 | * prepare_task_switch sets up locking and calls architecture specific | |
2607 | * hooks. | |
2608 | */ | |
e107be36 AK |
2609 | static inline void |
2610 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2611 | struct task_struct *next) | |
4866cde0 | 2612 | { |
0ed557aa | 2613 | kcov_prepare_switch(prev); |
43148951 | 2614 | sched_info_switch(rq, prev, next); |
fe4b04fa | 2615 | perf_event_task_sched_out(prev, next); |
d7822b1e | 2616 | rseq_preempt(prev); |
e107be36 | 2617 | fire_sched_out_preempt_notifiers(prev, next); |
31cb1bc0 | 2618 | prepare_task(next); |
4866cde0 NP |
2619 | prepare_arch_switch(next); |
2620 | } | |
2621 | ||
1da177e4 LT |
2622 | /** |
2623 | * finish_task_switch - clean up after a task-switch | |
2624 | * @prev: the thread we just switched away from. | |
2625 | * | |
4866cde0 NP |
2626 | * finish_task_switch must be called after the context switch, paired |
2627 | * with a prepare_task_switch call before the context switch. | |
2628 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2629 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2630 | * |
2631 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2632 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2633 | * with the lock held can cause deadlocks; see schedule() for |
2634 | * details.) | |
dfa50b60 ON |
2635 | * |
2636 | * The context switch have flipped the stack from under us and restored the | |
2637 | * local variables which were saved when this task called schedule() in the | |
2638 | * past. prev == current is still correct but we need to recalculate this_rq | |
2639 | * because prev may have moved to another CPU. | |
1da177e4 | 2640 | */ |
dfa50b60 | 2641 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
2642 | __releases(rq->lock) |
2643 | { | |
dfa50b60 | 2644 | struct rq *rq = this_rq(); |
1da177e4 | 2645 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2646 | long prev_state; |
1da177e4 | 2647 | |
609ca066 PZ |
2648 | /* |
2649 | * The previous task will have left us with a preempt_count of 2 | |
2650 | * because it left us after: | |
2651 | * | |
2652 | * schedule() | |
2653 | * preempt_disable(); // 1 | |
2654 | * __schedule() | |
2655 | * raw_spin_lock_irq(&rq->lock) // 2 | |
2656 | * | |
2657 | * Also, see FORK_PREEMPT_COUNT. | |
2658 | */ | |
e2bf1c4b PZ |
2659 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
2660 | "corrupted preempt_count: %s/%d/0x%x\n", | |
2661 | current->comm, current->pid, preempt_count())) | |
2662 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 2663 | |
1da177e4 LT |
2664 | rq->prev_mm = NULL; |
2665 | ||
2666 | /* | |
2667 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2668 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2669 | * schedule one last time. The schedule call will never return, and |
2670 | * the scheduled task must drop that reference. | |
95913d97 PZ |
2671 | * |
2672 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 2673 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
2674 | * running on another CPU and we could rave with its RUNNING -> DEAD |
2675 | * transition, resulting in a double drop. | |
1da177e4 | 2676 | */ |
55a101f8 | 2677 | prev_state = prev->state; |
bf9fae9f | 2678 | vtime_task_switch(prev); |
a8d757ef | 2679 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 2680 | finish_task(prev); |
2681 | finish_lock_switch(rq); | |
01f23e16 | 2682 | finish_arch_post_lock_switch(); |
0ed557aa | 2683 | kcov_finish_switch(current); |
e8fa1362 | 2684 | |
e107be36 | 2685 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 2686 | /* |
70216e18 MD |
2687 | * When switching through a kernel thread, the loop in |
2688 | * membarrier_{private,global}_expedited() may have observed that | |
2689 | * kernel thread and not issued an IPI. It is therefore possible to | |
2690 | * schedule between user->kernel->user threads without passing though | |
2691 | * switch_mm(). Membarrier requires a barrier after storing to | |
2692 | * rq->curr, before returning to userspace, so provide them here: | |
2693 | * | |
2694 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
2695 | * provided by mmdrop(), | |
2696 | * - a sync_core for SYNC_CORE. | |
306e0604 | 2697 | */ |
70216e18 MD |
2698 | if (mm) { |
2699 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 2700 | mmdrop(mm); |
70216e18 | 2701 | } |
1cef1150 PZ |
2702 | if (unlikely(prev_state == TASK_DEAD)) { |
2703 | if (prev->sched_class->task_dead) | |
2704 | prev->sched_class->task_dead(prev); | |
68f24b08 | 2705 | |
1cef1150 PZ |
2706 | /* |
2707 | * Remove function-return probe instances associated with this | |
2708 | * task and put them back on the free list. | |
2709 | */ | |
2710 | kprobe_flush_task(prev); | |
2711 | ||
2712 | /* Task is done with its stack. */ | |
2713 | put_task_stack(prev); | |
2714 | ||
2715 | put_task_struct(prev); | |
c6fd91f0 | 2716 | } |
99e5ada9 | 2717 | |
de734f89 | 2718 | tick_nohz_task_switch(); |
dfa50b60 | 2719 | return rq; |
1da177e4 LT |
2720 | } |
2721 | ||
3f029d3c GH |
2722 | #ifdef CONFIG_SMP |
2723 | ||
3f029d3c | 2724 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 2725 | static void __balance_callback(struct rq *rq) |
3f029d3c | 2726 | { |
e3fca9e7 PZ |
2727 | struct callback_head *head, *next; |
2728 | void (*func)(struct rq *rq); | |
2729 | unsigned long flags; | |
3f029d3c | 2730 | |
e3fca9e7 PZ |
2731 | raw_spin_lock_irqsave(&rq->lock, flags); |
2732 | head = rq->balance_callback; | |
2733 | rq->balance_callback = NULL; | |
2734 | while (head) { | |
2735 | func = (void (*)(struct rq *))head->func; | |
2736 | next = head->next; | |
2737 | head->next = NULL; | |
2738 | head = next; | |
3f029d3c | 2739 | |
e3fca9e7 | 2740 | func(rq); |
3f029d3c | 2741 | } |
e3fca9e7 PZ |
2742 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
2743 | } | |
2744 | ||
2745 | static inline void balance_callback(struct rq *rq) | |
2746 | { | |
2747 | if (unlikely(rq->balance_callback)) | |
2748 | __balance_callback(rq); | |
3f029d3c GH |
2749 | } |
2750 | ||
2751 | #else | |
da19ab51 | 2752 | |
e3fca9e7 | 2753 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 2754 | { |
1da177e4 LT |
2755 | } |
2756 | ||
3f029d3c GH |
2757 | #endif |
2758 | ||
1da177e4 LT |
2759 | /** |
2760 | * schedule_tail - first thing a freshly forked thread must call. | |
2761 | * @prev: the thread we just switched away from. | |
2762 | */ | |
722a9f92 | 2763 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2764 | __releases(rq->lock) |
2765 | { | |
1a43a14a | 2766 | struct rq *rq; |
da19ab51 | 2767 | |
609ca066 PZ |
2768 | /* |
2769 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
2770 | * finish_task_switch() for details. | |
2771 | * | |
2772 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
2773 | * and the preempt_enable() will end up enabling preemption (on | |
2774 | * PREEMPT_COUNT kernels). | |
2775 | */ | |
2776 | ||
dfa50b60 | 2777 | rq = finish_task_switch(prev); |
e3fca9e7 | 2778 | balance_callback(rq); |
1a43a14a | 2779 | preempt_enable(); |
70b97a7f | 2780 | |
1da177e4 | 2781 | if (current->set_child_tid) |
b488893a | 2782 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
2783 | |
2784 | calculate_sigpending(); | |
1da177e4 LT |
2785 | } |
2786 | ||
2787 | /* | |
dfa50b60 | 2788 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 2789 | */ |
04936948 | 2790 | static __always_inline struct rq * |
70b97a7f | 2791 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 2792 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 2793 | { |
dd41f596 | 2794 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2795 | |
e107be36 | 2796 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2797 | |
dd41f596 IM |
2798 | mm = next->mm; |
2799 | oldmm = prev->active_mm; | |
9226d125 ZA |
2800 | /* |
2801 | * For paravirt, this is coupled with an exit in switch_to to | |
2802 | * combine the page table reload and the switch backend into | |
2803 | * one hypercall. | |
2804 | */ | |
224101ed | 2805 | arch_start_context_switch(prev); |
9226d125 | 2806 | |
306e0604 MD |
2807 | /* |
2808 | * If mm is non-NULL, we pass through switch_mm(). If mm is | |
2809 | * NULL, we will pass through mmdrop() in finish_task_switch(). | |
2810 | * Both of these contain the full memory barrier required by | |
2811 | * membarrier after storing to rq->curr, before returning to | |
2812 | * user-space. | |
2813 | */ | |
31915ab4 | 2814 | if (!mm) { |
1da177e4 | 2815 | next->active_mm = oldmm; |
f1f10076 | 2816 | mmgrab(oldmm); |
1da177e4 LT |
2817 | enter_lazy_tlb(oldmm, next); |
2818 | } else | |
f98db601 | 2819 | switch_mm_irqs_off(oldmm, mm, next); |
1da177e4 | 2820 | |
31915ab4 | 2821 | if (!prev->mm) { |
1da177e4 | 2822 | prev->active_mm = NULL; |
1da177e4 LT |
2823 | rq->prev_mm = oldmm; |
2824 | } | |
92509b73 | 2825 | |
cb42c9a3 | 2826 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 2827 | |
269d5992 | 2828 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
2829 | |
2830 | /* Here we just switch the register state and the stack. */ | |
2831 | switch_to(prev, next, prev); | |
dd41f596 | 2832 | barrier(); |
dfa50b60 ON |
2833 | |
2834 | return finish_task_switch(prev); | |
1da177e4 LT |
2835 | } |
2836 | ||
2837 | /* | |
1c3e8264 | 2838 | * nr_running and nr_context_switches: |
1da177e4 LT |
2839 | * |
2840 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 2841 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
2842 | */ |
2843 | unsigned long nr_running(void) | |
2844 | { | |
2845 | unsigned long i, sum = 0; | |
2846 | ||
2847 | for_each_online_cpu(i) | |
2848 | sum += cpu_rq(i)->nr_running; | |
2849 | ||
2850 | return sum; | |
f711f609 | 2851 | } |
1da177e4 | 2852 | |
2ee507c4 | 2853 | /* |
d1ccc66d | 2854 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
2855 | * |
2856 | * Caution: this function does not check that the caller has disabled | |
2857 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
2858 | * race. The caller is responsible to use it correctly, for example: | |
2859 | * | |
2860 | * - from a non-preemptable section (of course) | |
2861 | * | |
2862 | * - from a thread that is bound to a single CPU | |
2863 | * | |
2864 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
2865 | */ |
2866 | bool single_task_running(void) | |
2867 | { | |
00cc1633 | 2868 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
2869 | } |
2870 | EXPORT_SYMBOL(single_task_running); | |
2871 | ||
1da177e4 | 2872 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2873 | { |
cc94abfc SR |
2874 | int i; |
2875 | unsigned long long sum = 0; | |
46cb4b7c | 2876 | |
0a945022 | 2877 | for_each_possible_cpu(i) |
1da177e4 | 2878 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2879 | |
1da177e4 LT |
2880 | return sum; |
2881 | } | |
483b4ee6 | 2882 | |
145d952a DL |
2883 | /* |
2884 | * Consumers of these two interfaces, like for example the cpuidle menu | |
2885 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
2886 | * for a CPU that has IO-wait which might not even end up running the task when | |
2887 | * it does become runnable. | |
2888 | */ | |
2889 | ||
2890 | unsigned long nr_iowait_cpu(int cpu) | |
2891 | { | |
2892 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
2893 | } | |
2894 | ||
e33a9bba TH |
2895 | /* |
2896 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
2897 | * | |
2898 | * The idea behind IO-wait account is to account the idle time that we could | |
2899 | * have spend running if it were not for IO. That is, if we were to improve the | |
2900 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
2901 | * | |
2902 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
2903 | * idle time as IO-wait, because if the storage were faster, it could've been | |
2904 | * running and we'd not be idle. | |
2905 | * | |
2906 | * This has been extended to SMP, by doing the same for each CPU. This however | |
2907 | * is broken. | |
2908 | * | |
2909 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
2910 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
2911 | * though, if the storage were faster, both could've ran at the same time, | |
2912 | * utilising both CPUs. | |
2913 | * | |
2914 | * This means, that when looking globally, the current IO-wait accounting on | |
2915 | * SMP is a lower bound, by reason of under accounting. | |
2916 | * | |
2917 | * Worse, since the numbers are provided per CPU, they are sometimes | |
2918 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
2919 | * associated with any one particular CPU, it can wake to another CPU than it | |
2920 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
2921 | * | |
2922 | * Task CPU affinities can make all that even more 'interesting'. | |
2923 | */ | |
2924 | ||
1da177e4 LT |
2925 | unsigned long nr_iowait(void) |
2926 | { | |
2927 | unsigned long i, sum = 0; | |
483b4ee6 | 2928 | |
0a945022 | 2929 | for_each_possible_cpu(i) |
145d952a | 2930 | sum += nr_iowait_cpu(i); |
46cb4b7c | 2931 | |
1da177e4 LT |
2932 | return sum; |
2933 | } | |
483b4ee6 | 2934 | |
dd41f596 | 2935 | #ifdef CONFIG_SMP |
8a0be9ef | 2936 | |
46cb4b7c | 2937 | /* |
38022906 PZ |
2938 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2939 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2940 | */ |
38022906 | 2941 | void sched_exec(void) |
46cb4b7c | 2942 | { |
38022906 | 2943 | struct task_struct *p = current; |
1da177e4 | 2944 | unsigned long flags; |
0017d735 | 2945 | int dest_cpu; |
46cb4b7c | 2946 | |
8f42ced9 | 2947 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2948 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2949 | if (dest_cpu == smp_processor_id()) |
2950 | goto unlock; | |
38022906 | 2951 | |
8f42ced9 | 2952 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2953 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2954 | |
8f42ced9 PZ |
2955 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2956 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2957 | return; |
2958 | } | |
0017d735 | 2959 | unlock: |
8f42ced9 | 2960 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2961 | } |
dd41f596 | 2962 | |
1da177e4 LT |
2963 | #endif |
2964 | ||
1da177e4 | 2965 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2966 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2967 | |
2968 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2969 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2970 | |
6075620b GG |
2971 | /* |
2972 | * The function fair_sched_class.update_curr accesses the struct curr | |
2973 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
2974 | * we observe a high rate of cache misses in practice. | |
2975 | * Prefetching this data results in improved performance. | |
2976 | */ | |
2977 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
2978 | { | |
2979 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
2980 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
2981 | #else | |
2982 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
2983 | #endif | |
2984 | prefetch(curr); | |
2985 | prefetch(&curr->exec_start); | |
2986 | } | |
2987 | ||
c5f8d995 HS |
2988 | /* |
2989 | * Return accounted runtime for the task. | |
2990 | * In case the task is currently running, return the runtime plus current's | |
2991 | * pending runtime that have not been accounted yet. | |
2992 | */ | |
2993 | unsigned long long task_sched_runtime(struct task_struct *p) | |
2994 | { | |
eb580751 | 2995 | struct rq_flags rf; |
c5f8d995 | 2996 | struct rq *rq; |
6e998916 | 2997 | u64 ns; |
c5f8d995 | 2998 | |
911b2898 PZ |
2999 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3000 | /* | |
97fb7a0a | 3001 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
3002 | * So we have a optimization chance when the task's delta_exec is 0. |
3003 | * Reading ->on_cpu is racy, but this is ok. | |
3004 | * | |
d1ccc66d IM |
3005 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
3006 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 3007 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
3008 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3009 | * been accounted, so we're correct here as well. | |
911b2898 | 3010 | */ |
da0c1e65 | 3011 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3012 | return p->se.sum_exec_runtime; |
3013 | #endif | |
3014 | ||
eb580751 | 3015 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3016 | /* |
3017 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3018 | * project cycles that may never be accounted to this | |
3019 | * thread, breaking clock_gettime(). | |
3020 | */ | |
3021 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3022 | prefetch_curr_exec_start(p); |
6e998916 SG |
3023 | update_rq_clock(rq); |
3024 | p->sched_class->update_curr(rq); | |
3025 | } | |
3026 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3027 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3028 | |
3029 | return ns; | |
3030 | } | |
48f24c4d | 3031 | |
7835b98b CL |
3032 | /* |
3033 | * This function gets called by the timer code, with HZ frequency. | |
3034 | * We call it with interrupts disabled. | |
7835b98b CL |
3035 | */ |
3036 | void scheduler_tick(void) | |
3037 | { | |
7835b98b CL |
3038 | int cpu = smp_processor_id(); |
3039 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3040 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 3041 | struct rq_flags rf; |
3e51f33f PZ |
3042 | |
3043 | sched_clock_tick(); | |
dd41f596 | 3044 | |
8a8c69c3 PZ |
3045 | rq_lock(rq, &rf); |
3046 | ||
3e51f33f | 3047 | update_rq_clock(rq); |
fa85ae24 | 3048 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3049 | cpu_load_update_active(rq); |
3289bdb4 | 3050 | calc_global_load_tick(rq); |
eb414681 | 3051 | psi_task_tick(rq); |
8a8c69c3 PZ |
3052 | |
3053 | rq_unlock(rq, &rf); | |
7835b98b | 3054 | |
e9d2b064 | 3055 | perf_event_task_tick(); |
e220d2dc | 3056 | |
e418e1c2 | 3057 | #ifdef CONFIG_SMP |
6eb57e0d | 3058 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3059 | trigger_load_balance(rq); |
e418e1c2 | 3060 | #endif |
1da177e4 LT |
3061 | } |
3062 | ||
265f22a9 | 3063 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
3064 | |
3065 | struct tick_work { | |
3066 | int cpu; | |
3067 | struct delayed_work work; | |
3068 | }; | |
3069 | ||
3070 | static struct tick_work __percpu *tick_work_cpu; | |
3071 | ||
3072 | static void sched_tick_remote(struct work_struct *work) | |
3073 | { | |
3074 | struct delayed_work *dwork = to_delayed_work(work); | |
3075 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
3076 | int cpu = twork->cpu; | |
3077 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 3078 | struct task_struct *curr; |
d84b3131 | 3079 | struct rq_flags rf; |
d9c0ffca | 3080 | u64 delta; |
d84b3131 FW |
3081 | |
3082 | /* | |
3083 | * Handle the tick only if it appears the remote CPU is running in full | |
3084 | * dynticks mode. The check is racy by nature, but missing a tick or | |
3085 | * having one too much is no big deal because the scheduler tick updates | |
3086 | * statistics and checks timeslices in a time-independent way, regardless | |
3087 | * of when exactly it is running. | |
3088 | */ | |
d9c0ffca FW |
3089 | if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu)) |
3090 | goto out_requeue; | |
d84b3131 | 3091 | |
d9c0ffca FW |
3092 | rq_lock_irq(rq, &rf); |
3093 | curr = rq->curr; | |
3094 | if (is_idle_task(curr)) | |
3095 | goto out_unlock; | |
d84b3131 | 3096 | |
d9c0ffca FW |
3097 | update_rq_clock(rq); |
3098 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
3099 | ||
3100 | /* | |
3101 | * Make sure the next tick runs within a reasonable | |
3102 | * amount of time. | |
3103 | */ | |
3104 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
3105 | curr->sched_class->task_tick(rq, curr, 0); | |
3106 | ||
3107 | out_unlock: | |
3108 | rq_unlock_irq(rq, &rf); | |
d84b3131 | 3109 | |
d9c0ffca | 3110 | out_requeue: |
d84b3131 FW |
3111 | /* |
3112 | * Run the remote tick once per second (1Hz). This arbitrary | |
3113 | * frequency is large enough to avoid overload but short enough | |
3114 | * to keep scheduler internal stats reasonably up to date. | |
3115 | */ | |
3116 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
3117 | } | |
3118 | ||
3119 | static void sched_tick_start(int cpu) | |
3120 | { | |
3121 | struct tick_work *twork; | |
3122 | ||
3123 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
3124 | return; | |
3125 | ||
3126 | WARN_ON_ONCE(!tick_work_cpu); | |
3127 | ||
3128 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
3129 | twork->cpu = cpu; | |
3130 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
3131 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
3132 | } | |
3133 | ||
3134 | #ifdef CONFIG_HOTPLUG_CPU | |
3135 | static void sched_tick_stop(int cpu) | |
3136 | { | |
3137 | struct tick_work *twork; | |
3138 | ||
3139 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
3140 | return; | |
3141 | ||
3142 | WARN_ON_ONCE(!tick_work_cpu); | |
3143 | ||
3144 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
3145 | cancel_delayed_work_sync(&twork->work); | |
3146 | } | |
3147 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3148 | ||
3149 | int __init sched_tick_offload_init(void) | |
3150 | { | |
3151 | tick_work_cpu = alloc_percpu(struct tick_work); | |
3152 | BUG_ON(!tick_work_cpu); | |
3153 | ||
3154 | return 0; | |
3155 | } | |
3156 | ||
3157 | #else /* !CONFIG_NO_HZ_FULL */ | |
3158 | static inline void sched_tick_start(int cpu) { } | |
3159 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 3160 | #endif |
1da177e4 | 3161 | |
7e49fcce | 3162 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 3163 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
3164 | /* |
3165 | * If the value passed in is equal to the current preempt count | |
3166 | * then we just disabled preemption. Start timing the latency. | |
3167 | */ | |
3168 | static inline void preempt_latency_start(int val) | |
3169 | { | |
3170 | if (preempt_count() == val) { | |
3171 | unsigned long ip = get_lock_parent_ip(); | |
3172 | #ifdef CONFIG_DEBUG_PREEMPT | |
3173 | current->preempt_disable_ip = ip; | |
3174 | #endif | |
3175 | trace_preempt_off(CALLER_ADDR0, ip); | |
3176 | } | |
3177 | } | |
7e49fcce | 3178 | |
edafe3a5 | 3179 | void preempt_count_add(int val) |
1da177e4 | 3180 | { |
6cd8a4bb | 3181 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3182 | /* |
3183 | * Underflow? | |
3184 | */ | |
9a11b49a IM |
3185 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3186 | return; | |
6cd8a4bb | 3187 | #endif |
bdb43806 | 3188 | __preempt_count_add(val); |
6cd8a4bb | 3189 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3190 | /* |
3191 | * Spinlock count overflowing soon? | |
3192 | */ | |
33859f7f MOS |
3193 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3194 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3195 | #endif |
47252cfb | 3196 | preempt_latency_start(val); |
1da177e4 | 3197 | } |
bdb43806 | 3198 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3199 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3200 | |
47252cfb SR |
3201 | /* |
3202 | * If the value passed in equals to the current preempt count | |
3203 | * then we just enabled preemption. Stop timing the latency. | |
3204 | */ | |
3205 | static inline void preempt_latency_stop(int val) | |
3206 | { | |
3207 | if (preempt_count() == val) | |
3208 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3209 | } | |
3210 | ||
edafe3a5 | 3211 | void preempt_count_sub(int val) |
1da177e4 | 3212 | { |
6cd8a4bb | 3213 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3214 | /* |
3215 | * Underflow? | |
3216 | */ | |
01e3eb82 | 3217 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3218 | return; |
1da177e4 LT |
3219 | /* |
3220 | * Is the spinlock portion underflowing? | |
3221 | */ | |
9a11b49a IM |
3222 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3223 | !(preempt_count() & PREEMPT_MASK))) | |
3224 | return; | |
6cd8a4bb | 3225 | #endif |
9a11b49a | 3226 | |
47252cfb | 3227 | preempt_latency_stop(val); |
bdb43806 | 3228 | __preempt_count_sub(val); |
1da177e4 | 3229 | } |
bdb43806 | 3230 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3231 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3232 | |
47252cfb SR |
3233 | #else |
3234 | static inline void preempt_latency_start(int val) { } | |
3235 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3236 | #endif |
3237 | ||
59ddbcb2 IM |
3238 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
3239 | { | |
3240 | #ifdef CONFIG_DEBUG_PREEMPT | |
3241 | return p->preempt_disable_ip; | |
3242 | #else | |
3243 | return 0; | |
3244 | #endif | |
3245 | } | |
3246 | ||
1da177e4 | 3247 | /* |
dd41f596 | 3248 | * Print scheduling while atomic bug: |
1da177e4 | 3249 | */ |
dd41f596 | 3250 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3251 | { |
d1c6d149 VN |
3252 | /* Save this before calling printk(), since that will clobber it */ |
3253 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3254 | ||
664dfa65 DJ |
3255 | if (oops_in_progress) |
3256 | return; | |
3257 | ||
3df0fc5b PZ |
3258 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3259 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3260 | |
dd41f596 | 3261 | debug_show_held_locks(prev); |
e21f5b15 | 3262 | print_modules(); |
dd41f596 IM |
3263 | if (irqs_disabled()) |
3264 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3265 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3266 | && in_atomic_preempt_off()) { | |
8f47b187 | 3267 | pr_err("Preemption disabled at:"); |
d1c6d149 | 3268 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
3269 | pr_cont("\n"); |
3270 | } | |
748c7201 DBO |
3271 | if (panic_on_warn) |
3272 | panic("scheduling while atomic\n"); | |
3273 | ||
6135fc1e | 3274 | dump_stack(); |
373d4d09 | 3275 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3276 | } |
1da177e4 | 3277 | |
dd41f596 IM |
3278 | /* |
3279 | * Various schedule()-time debugging checks and statistics: | |
3280 | */ | |
3281 | static inline void schedule_debug(struct task_struct *prev) | |
3282 | { | |
0d9e2632 | 3283 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3284 | if (task_stack_end_corrupted(prev)) |
3285 | panic("corrupted stack end detected inside scheduler\n"); | |
0d9e2632 | 3286 | #endif |
b99def8b | 3287 | |
1dc0fffc | 3288 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3289 | __schedule_bug(prev); |
1dc0fffc PZ |
3290 | preempt_count_set(PREEMPT_DISABLED); |
3291 | } | |
b3fbab05 | 3292 | rcu_sleep_check(); |
dd41f596 | 3293 | |
1da177e4 LT |
3294 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3295 | ||
ae92882e | 3296 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3297 | } |
3298 | ||
3299 | /* | |
3300 | * Pick up the highest-prio task: | |
3301 | */ | |
3302 | static inline struct task_struct * | |
d8ac8971 | 3303 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 3304 | { |
49ee5768 | 3305 | const struct sched_class *class; |
dd41f596 | 3306 | struct task_struct *p; |
1da177e4 LT |
3307 | |
3308 | /* | |
0ba87bb2 PZ |
3309 | * Optimization: we know that if all tasks are in the fair class we can |
3310 | * call that function directly, but only if the @prev task wasn't of a | |
3311 | * higher scheduling class, because otherwise those loose the | |
3312 | * opportunity to pull in more work from other CPUs. | |
1da177e4 | 3313 | */ |
0ba87bb2 PZ |
3314 | if (likely((prev->sched_class == &idle_sched_class || |
3315 | prev->sched_class == &fair_sched_class) && | |
3316 | rq->nr_running == rq->cfs.h_nr_running)) { | |
3317 | ||
d8ac8971 | 3318 | p = fair_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3319 | if (unlikely(p == RETRY_TASK)) |
3320 | goto again; | |
3321 | ||
d1ccc66d | 3322 | /* Assumes fair_sched_class->next == idle_sched_class */ |
6ccdc84b | 3323 | if (unlikely(!p)) |
d8ac8971 | 3324 | p = idle_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3325 | |
3326 | return p; | |
1da177e4 LT |
3327 | } |
3328 | ||
37e117c0 | 3329 | again: |
34f971f6 | 3330 | for_each_class(class) { |
d8ac8971 | 3331 | p = class->pick_next_task(rq, prev, rf); |
37e117c0 PZ |
3332 | if (p) { |
3333 | if (unlikely(p == RETRY_TASK)) | |
3334 | goto again; | |
dd41f596 | 3335 | return p; |
37e117c0 | 3336 | } |
dd41f596 | 3337 | } |
34f971f6 | 3338 | |
d1ccc66d IM |
3339 | /* The idle class should always have a runnable task: */ |
3340 | BUG(); | |
dd41f596 | 3341 | } |
1da177e4 | 3342 | |
dd41f596 | 3343 | /* |
c259e01a | 3344 | * __schedule() is the main scheduler function. |
edde96ea PE |
3345 | * |
3346 | * The main means of driving the scheduler and thus entering this function are: | |
3347 | * | |
3348 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3349 | * | |
3350 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3351 | * paths. For example, see arch/x86/entry_64.S. | |
3352 | * | |
3353 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3354 | * interrupt handler scheduler_tick(). | |
3355 | * | |
3356 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3357 | * task to the run-queue and that's it. | |
3358 | * | |
3359 | * Now, if the new task added to the run-queue preempts the current | |
3360 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3361 | * called on the nearest possible occasion: | |
3362 | * | |
3363 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3364 | * | |
3365 | * - in syscall or exception context, at the next outmost | |
3366 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3367 | * spin_unlock()!) | |
3368 | * | |
3369 | * - in IRQ context, return from interrupt-handler to | |
3370 | * preemptible context | |
3371 | * | |
3372 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3373 | * then at the next: | |
3374 | * | |
3375 | * - cond_resched() call | |
3376 | * - explicit schedule() call | |
3377 | * - return from syscall or exception to user-space | |
3378 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3379 | * |
b30f0e3f | 3380 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3381 | */ |
499d7955 | 3382 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3383 | { |
3384 | struct task_struct *prev, *next; | |
67ca7bde | 3385 | unsigned long *switch_count; |
d8ac8971 | 3386 | struct rq_flags rf; |
dd41f596 | 3387 | struct rq *rq; |
31656519 | 3388 | int cpu; |
dd41f596 | 3389 | |
dd41f596 IM |
3390 | cpu = smp_processor_id(); |
3391 | rq = cpu_rq(cpu); | |
dd41f596 | 3392 | prev = rq->curr; |
dd41f596 | 3393 | |
dd41f596 | 3394 | schedule_debug(prev); |
1da177e4 | 3395 | |
31656519 | 3396 | if (sched_feat(HRTICK)) |
f333fdc9 | 3397 | hrtick_clear(rq); |
8f4d37ec | 3398 | |
46a5d164 | 3399 | local_irq_disable(); |
bcbfdd01 | 3400 | rcu_note_context_switch(preempt); |
46a5d164 | 3401 | |
e0acd0a6 ON |
3402 | /* |
3403 | * Make sure that signal_pending_state()->signal_pending() below | |
3404 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3405 | * done by the caller to avoid the race with signal_wake_up(). | |
306e0604 MD |
3406 | * |
3407 | * The membarrier system call requires a full memory barrier | |
3408 | * after coming from user-space, before storing to rq->curr. | |
e0acd0a6 | 3409 | */ |
8a8c69c3 | 3410 | rq_lock(rq, &rf); |
d89e588c | 3411 | smp_mb__after_spinlock(); |
1da177e4 | 3412 | |
d1ccc66d IM |
3413 | /* Promote REQ to ACT */ |
3414 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 3415 | update_rq_clock(rq); |
9edfbfed | 3416 | |
246d86b5 | 3417 | switch_count = &prev->nivcsw; |
fc13aeba | 3418 | if (!preempt && prev->state) { |
21aa9af0 | 3419 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3420 | prev->state = TASK_RUNNING; |
21aa9af0 | 3421 | } else { |
bce4dc80 | 3422 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e PZ |
3423 | prev->on_rq = 0; |
3424 | ||
e33a9bba TH |
3425 | if (prev->in_iowait) { |
3426 | atomic_inc(&rq->nr_iowait); | |
3427 | delayacct_blkio_start(); | |
3428 | } | |
3429 | ||
21aa9af0 | 3430 | /* |
2acca55e PZ |
3431 | * If a worker went to sleep, notify and ask workqueue |
3432 | * whether it wants to wake up a task to maintain | |
3433 | * concurrency. | |
21aa9af0 TH |
3434 | */ |
3435 | if (prev->flags & PF_WQ_WORKER) { | |
3436 | struct task_struct *to_wakeup; | |
3437 | ||
9b7f6597 | 3438 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3439 | if (to_wakeup) |
d8ac8971 | 3440 | try_to_wake_up_local(to_wakeup, &rf); |
21aa9af0 | 3441 | } |
21aa9af0 | 3442 | } |
dd41f596 | 3443 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3444 | } |
3445 | ||
d8ac8971 | 3446 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 3447 | clear_tsk_need_resched(prev); |
f27dde8d | 3448 | clear_preempt_need_resched(); |
1da177e4 | 3449 | |
1da177e4 | 3450 | if (likely(prev != next)) { |
1da177e4 LT |
3451 | rq->nr_switches++; |
3452 | rq->curr = next; | |
22e4ebb9 MD |
3453 | /* |
3454 | * The membarrier system call requires each architecture | |
3455 | * to have a full memory barrier after updating | |
306e0604 MD |
3456 | * rq->curr, before returning to user-space. |
3457 | * | |
3458 | * Here are the schemes providing that barrier on the | |
3459 | * various architectures: | |
3460 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
3461 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
3462 | * - finish_lock_switch() for weakly-ordered | |
3463 | * architectures where spin_unlock is a full barrier, | |
3464 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
3465 | * is a RELEASE barrier), | |
22e4ebb9 | 3466 | */ |
1da177e4 LT |
3467 | ++*switch_count; |
3468 | ||
c73464b1 | 3469 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
3470 | |
3471 | /* Also unlocks the rq: */ | |
3472 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 3473 | } else { |
cb42c9a3 | 3474 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
8a8c69c3 | 3475 | rq_unlock_irq(rq, &rf); |
cbce1a68 | 3476 | } |
1da177e4 | 3477 | |
e3fca9e7 | 3478 | balance_callback(rq); |
1da177e4 | 3479 | } |
c259e01a | 3480 | |
9af6528e PZ |
3481 | void __noreturn do_task_dead(void) |
3482 | { | |
d1ccc66d | 3483 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 3484 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
3485 | |
3486 | /* Tell freezer to ignore us: */ | |
3487 | current->flags |= PF_NOFREEZE; | |
3488 | ||
9af6528e PZ |
3489 | __schedule(false); |
3490 | BUG(); | |
d1ccc66d IM |
3491 | |
3492 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 3493 | for (;;) |
d1ccc66d | 3494 | cpu_relax(); |
9af6528e PZ |
3495 | } |
3496 | ||
9c40cef2 TG |
3497 | static inline void sched_submit_work(struct task_struct *tsk) |
3498 | { | |
3c7d5184 | 3499 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3500 | return; |
3501 | /* | |
3502 | * If we are going to sleep and we have plugged IO queued, | |
3503 | * make sure to submit it to avoid deadlocks. | |
3504 | */ | |
3505 | if (blk_needs_flush_plug(tsk)) | |
3506 | blk_schedule_flush_plug(tsk); | |
3507 | } | |
3508 | ||
722a9f92 | 3509 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3510 | { |
9c40cef2 TG |
3511 | struct task_struct *tsk = current; |
3512 | ||
3513 | sched_submit_work(tsk); | |
bfd9b2b5 | 3514 | do { |
b30f0e3f | 3515 | preempt_disable(); |
fc13aeba | 3516 | __schedule(false); |
b30f0e3f | 3517 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3518 | } while (need_resched()); |
c259e01a | 3519 | } |
1da177e4 LT |
3520 | EXPORT_SYMBOL(schedule); |
3521 | ||
8663effb SRV |
3522 | /* |
3523 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
3524 | * state (have scheduled out non-voluntarily) by making sure that all | |
3525 | * tasks have either left the run queue or have gone into user space. | |
3526 | * As idle tasks do not do either, they must not ever be preempted | |
3527 | * (schedule out non-voluntarily). | |
3528 | * | |
3529 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
3530 | * never enables preemption because it does not call sched_submit_work(). | |
3531 | */ | |
3532 | void __sched schedule_idle(void) | |
3533 | { | |
3534 | /* | |
3535 | * As this skips calling sched_submit_work(), which the idle task does | |
3536 | * regardless because that function is a nop when the task is in a | |
3537 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
3538 | * current task can be in any other state. Note, idle is always in the | |
3539 | * TASK_RUNNING state. | |
3540 | */ | |
3541 | WARN_ON_ONCE(current->state); | |
3542 | do { | |
3543 | __schedule(false); | |
3544 | } while (need_resched()); | |
3545 | } | |
3546 | ||
91d1aa43 | 3547 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3548 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3549 | { |
3550 | /* | |
3551 | * If we come here after a random call to set_need_resched(), | |
3552 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3553 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3554 | * we find a better solution. | |
7cc78f8f AL |
3555 | * |
3556 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3557 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3558 | * too frequently to make sense yet. |
20ab65e3 | 3559 | */ |
7cc78f8f | 3560 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3561 | schedule(); |
7cc78f8f | 3562 | exception_exit(prev_state); |
20ab65e3 FW |
3563 | } |
3564 | #endif | |
3565 | ||
c5491ea7 TG |
3566 | /** |
3567 | * schedule_preempt_disabled - called with preemption disabled | |
3568 | * | |
3569 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3570 | */ | |
3571 | void __sched schedule_preempt_disabled(void) | |
3572 | { | |
ba74c144 | 3573 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3574 | schedule(); |
3575 | preempt_disable(); | |
3576 | } | |
3577 | ||
06b1f808 | 3578 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3579 | { |
3580 | do { | |
47252cfb SR |
3581 | /* |
3582 | * Because the function tracer can trace preempt_count_sub() | |
3583 | * and it also uses preempt_enable/disable_notrace(), if | |
3584 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3585 | * by the function tracer will call this function again and | |
3586 | * cause infinite recursion. | |
3587 | * | |
3588 | * Preemption must be disabled here before the function | |
3589 | * tracer can trace. Break up preempt_disable() into two | |
3590 | * calls. One to disable preemption without fear of being | |
3591 | * traced. The other to still record the preemption latency, | |
3592 | * which can also be traced by the function tracer. | |
3593 | */ | |
499d7955 | 3594 | preempt_disable_notrace(); |
47252cfb | 3595 | preempt_latency_start(1); |
fc13aeba | 3596 | __schedule(true); |
47252cfb | 3597 | preempt_latency_stop(1); |
499d7955 | 3598 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3599 | |
3600 | /* | |
3601 | * Check again in case we missed a preemption opportunity | |
3602 | * between schedule and now. | |
3603 | */ | |
a18b5d01 FW |
3604 | } while (need_resched()); |
3605 | } | |
3606 | ||
1da177e4 LT |
3607 | #ifdef CONFIG_PREEMPT |
3608 | /* | |
2ed6e34f | 3609 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3610 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3611 | * occur there and call schedule directly. |
3612 | */ | |
722a9f92 | 3613 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3614 | { |
1da177e4 LT |
3615 | /* |
3616 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3617 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3618 | */ |
fbb00b56 | 3619 | if (likely(!preemptible())) |
1da177e4 LT |
3620 | return; |
3621 | ||
a18b5d01 | 3622 | preempt_schedule_common(); |
1da177e4 | 3623 | } |
376e2424 | 3624 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3625 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3626 | |
009f60e2 | 3627 | /** |
4eaca0a8 | 3628 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3629 | * |
3630 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3631 | * recursion and tracing preempt enabling caused by the tracing | |
3632 | * infrastructure itself. But as tracing can happen in areas coming | |
3633 | * from userspace or just about to enter userspace, a preempt enable | |
3634 | * can occur before user_exit() is called. This will cause the scheduler | |
3635 | * to be called when the system is still in usermode. | |
3636 | * | |
3637 | * To prevent this, the preempt_enable_notrace will use this function | |
3638 | * instead of preempt_schedule() to exit user context if needed before | |
3639 | * calling the scheduler. | |
3640 | */ | |
4eaca0a8 | 3641 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3642 | { |
3643 | enum ctx_state prev_ctx; | |
3644 | ||
3645 | if (likely(!preemptible())) | |
3646 | return; | |
3647 | ||
3648 | do { | |
47252cfb SR |
3649 | /* |
3650 | * Because the function tracer can trace preempt_count_sub() | |
3651 | * and it also uses preempt_enable/disable_notrace(), if | |
3652 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3653 | * by the function tracer will call this function again and | |
3654 | * cause infinite recursion. | |
3655 | * | |
3656 | * Preemption must be disabled here before the function | |
3657 | * tracer can trace. Break up preempt_disable() into two | |
3658 | * calls. One to disable preemption without fear of being | |
3659 | * traced. The other to still record the preemption latency, | |
3660 | * which can also be traced by the function tracer. | |
3661 | */ | |
3d8f74dd | 3662 | preempt_disable_notrace(); |
47252cfb | 3663 | preempt_latency_start(1); |
009f60e2 ON |
3664 | /* |
3665 | * Needs preempt disabled in case user_exit() is traced | |
3666 | * and the tracer calls preempt_enable_notrace() causing | |
3667 | * an infinite recursion. | |
3668 | */ | |
3669 | prev_ctx = exception_enter(); | |
fc13aeba | 3670 | __schedule(true); |
009f60e2 ON |
3671 | exception_exit(prev_ctx); |
3672 | ||
47252cfb | 3673 | preempt_latency_stop(1); |
3d8f74dd | 3674 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3675 | } while (need_resched()); |
3676 | } | |
4eaca0a8 | 3677 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3678 | |
32e475d7 | 3679 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3680 | |
3681 | /* | |
2ed6e34f | 3682 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3683 | * off of irq context. |
3684 | * Note, that this is called and return with irqs disabled. This will | |
3685 | * protect us against recursive calling from irq. | |
3686 | */ | |
722a9f92 | 3687 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3688 | { |
b22366cd | 3689 | enum ctx_state prev_state; |
6478d880 | 3690 | |
2ed6e34f | 3691 | /* Catch callers which need to be fixed */ |
f27dde8d | 3692 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3693 | |
b22366cd FW |
3694 | prev_state = exception_enter(); |
3695 | ||
3a5c359a | 3696 | do { |
3d8f74dd | 3697 | preempt_disable(); |
3a5c359a | 3698 | local_irq_enable(); |
fc13aeba | 3699 | __schedule(true); |
3a5c359a | 3700 | local_irq_disable(); |
3d8f74dd | 3701 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3702 | } while (need_resched()); |
b22366cd FW |
3703 | |
3704 | exception_exit(prev_state); | |
1da177e4 LT |
3705 | } |
3706 | ||
ac6424b9 | 3707 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3708 | void *key) |
1da177e4 | 3709 | { |
63859d4f | 3710 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3711 | } |
1da177e4 LT |
3712 | EXPORT_SYMBOL(default_wake_function); |
3713 | ||
b29739f9 IM |
3714 | #ifdef CONFIG_RT_MUTEXES |
3715 | ||
acd58620 PZ |
3716 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
3717 | { | |
3718 | if (pi_task) | |
3719 | prio = min(prio, pi_task->prio); | |
3720 | ||
3721 | return prio; | |
3722 | } | |
3723 | ||
3724 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
3725 | { | |
3726 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
3727 | ||
3728 | return __rt_effective_prio(pi_task, prio); | |
3729 | } | |
3730 | ||
b29739f9 IM |
3731 | /* |
3732 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
3733 | * @p: task to boost |
3734 | * @pi_task: donor task | |
b29739f9 IM |
3735 | * |
3736 | * This function changes the 'effective' priority of a task. It does | |
3737 | * not touch ->normal_prio like __setscheduler(). | |
3738 | * | |
c365c292 TG |
3739 | * Used by the rt_mutex code to implement priority inheritance |
3740 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3741 | */ |
acd58620 | 3742 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 3743 | { |
acd58620 | 3744 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 3745 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 3746 | const struct sched_class *prev_class; |
eb580751 PZ |
3747 | struct rq_flags rf; |
3748 | struct rq *rq; | |
b29739f9 | 3749 | |
acd58620 PZ |
3750 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
3751 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
3752 | ||
3753 | /* | |
3754 | * If nothing changed; bail early. | |
3755 | */ | |
3756 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
3757 | return; | |
b29739f9 | 3758 | |
eb580751 | 3759 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 3760 | update_rq_clock(rq); |
acd58620 PZ |
3761 | /* |
3762 | * Set under pi_lock && rq->lock, such that the value can be used under | |
3763 | * either lock. | |
3764 | * | |
3765 | * Note that there is loads of tricky to make this pointer cache work | |
3766 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
3767 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
3768 | * task is allowed to run again (and can exit). This ensures the pointer | |
3769 | * points to a blocked task -- which guaratees the task is present. | |
3770 | */ | |
3771 | p->pi_top_task = pi_task; | |
3772 | ||
3773 | /* | |
3774 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
3775 | */ | |
3776 | if (prio == p->prio && !dl_prio(prio)) | |
3777 | goto out_unlock; | |
b29739f9 | 3778 | |
1c4dd99b TG |
3779 | /* |
3780 | * Idle task boosting is a nono in general. There is one | |
3781 | * exception, when PREEMPT_RT and NOHZ is active: | |
3782 | * | |
3783 | * The idle task calls get_next_timer_interrupt() and holds | |
3784 | * the timer wheel base->lock on the CPU and another CPU wants | |
3785 | * to access the timer (probably to cancel it). We can safely | |
3786 | * ignore the boosting request, as the idle CPU runs this code | |
3787 | * with interrupts disabled and will complete the lock | |
3788 | * protected section without being interrupted. So there is no | |
3789 | * real need to boost. | |
3790 | */ | |
3791 | if (unlikely(p == rq->idle)) { | |
3792 | WARN_ON(p != rq->curr); | |
3793 | WARN_ON(p->pi_blocked_on); | |
3794 | goto out_unlock; | |
3795 | } | |
3796 | ||
b91473ff | 3797 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 3798 | oldprio = p->prio; |
ff77e468 PZ |
3799 | |
3800 | if (oldprio == prio) | |
3801 | queue_flag &= ~DEQUEUE_MOVE; | |
3802 | ||
83ab0aa0 | 3803 | prev_class = p->sched_class; |
da0c1e65 | 3804 | queued = task_on_rq_queued(p); |
051a1d1a | 3805 | running = task_current(rq, p); |
da0c1e65 | 3806 | if (queued) |
ff77e468 | 3807 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3808 | if (running) |
f3cd1c4e | 3809 | put_prev_task(rq, p); |
dd41f596 | 3810 | |
2d3d891d DF |
3811 | /* |
3812 | * Boosting condition are: | |
3813 | * 1. -rt task is running and holds mutex A | |
3814 | * --> -dl task blocks on mutex A | |
3815 | * | |
3816 | * 2. -dl task is running and holds mutex A | |
3817 | * --> -dl task blocks on mutex A and could preempt the | |
3818 | * running task | |
3819 | */ | |
3820 | if (dl_prio(prio)) { | |
466af29b ON |
3821 | if (!dl_prio(p->normal_prio) || |
3822 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3823 | p->dl.dl_boosted = 1; |
ff77e468 | 3824 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3825 | } else |
3826 | p->dl.dl_boosted = 0; | |
aab03e05 | 3827 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3828 | } else if (rt_prio(prio)) { |
3829 | if (dl_prio(oldprio)) | |
3830 | p->dl.dl_boosted = 0; | |
3831 | if (oldprio < prio) | |
ff77e468 | 3832 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3833 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3834 | } else { |
3835 | if (dl_prio(oldprio)) | |
3836 | p->dl.dl_boosted = 0; | |
746db944 BS |
3837 | if (rt_prio(oldprio)) |
3838 | p->rt.timeout = 0; | |
dd41f596 | 3839 | p->sched_class = &fair_sched_class; |
2d3d891d | 3840 | } |
dd41f596 | 3841 | |
b29739f9 IM |
3842 | p->prio = prio; |
3843 | ||
da0c1e65 | 3844 | if (queued) |
ff77e468 | 3845 | enqueue_task(rq, p, queue_flag); |
a399d233 | 3846 | if (running) |
b2bf6c31 | 3847 | set_curr_task(rq, p); |
cb469845 | 3848 | |
da7a735e | 3849 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3850 | out_unlock: |
d1ccc66d IM |
3851 | /* Avoid rq from going away on us: */ |
3852 | preempt_disable(); | |
eb580751 | 3853 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3854 | |
3855 | balance_callback(rq); | |
3856 | preempt_enable(); | |
b29739f9 | 3857 | } |
acd58620 PZ |
3858 | #else |
3859 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
3860 | { | |
3861 | return prio; | |
3862 | } | |
b29739f9 | 3863 | #endif |
d50dde5a | 3864 | |
36c8b586 | 3865 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3866 | { |
49bd21ef PZ |
3867 | bool queued, running; |
3868 | int old_prio, delta; | |
eb580751 | 3869 | struct rq_flags rf; |
70b97a7f | 3870 | struct rq *rq; |
1da177e4 | 3871 | |
75e45d51 | 3872 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3873 | return; |
3874 | /* | |
3875 | * We have to be careful, if called from sys_setpriority(), | |
3876 | * the task might be in the middle of scheduling on another CPU. | |
3877 | */ | |
eb580751 | 3878 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
3879 | update_rq_clock(rq); |
3880 | ||
1da177e4 LT |
3881 | /* |
3882 | * The RT priorities are set via sched_setscheduler(), but we still | |
3883 | * allow the 'normal' nice value to be set - but as expected | |
3884 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3885 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3886 | */ |
aab03e05 | 3887 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3888 | p->static_prio = NICE_TO_PRIO(nice); |
3889 | goto out_unlock; | |
3890 | } | |
da0c1e65 | 3891 | queued = task_on_rq_queued(p); |
49bd21ef | 3892 | running = task_current(rq, p); |
da0c1e65 | 3893 | if (queued) |
7a57f32a | 3894 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
3895 | if (running) |
3896 | put_prev_task(rq, p); | |
1da177e4 | 3897 | |
1da177e4 | 3898 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 3899 | set_load_weight(p, true); |
b29739f9 IM |
3900 | old_prio = p->prio; |
3901 | p->prio = effective_prio(p); | |
3902 | delta = p->prio - old_prio; | |
1da177e4 | 3903 | |
da0c1e65 | 3904 | if (queued) { |
7134b3e9 | 3905 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
1da177e4 | 3906 | /* |
d5f9f942 AM |
3907 | * If the task increased its priority or is running and |
3908 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3909 | */ |
d5f9f942 | 3910 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3911 | resched_curr(rq); |
1da177e4 | 3912 | } |
49bd21ef PZ |
3913 | if (running) |
3914 | set_curr_task(rq, p); | |
1da177e4 | 3915 | out_unlock: |
eb580751 | 3916 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3917 | } |
1da177e4 LT |
3918 | EXPORT_SYMBOL(set_user_nice); |
3919 | ||
e43379f1 MM |
3920 | /* |
3921 | * can_nice - check if a task can reduce its nice value | |
3922 | * @p: task | |
3923 | * @nice: nice value | |
3924 | */ | |
36c8b586 | 3925 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3926 | { |
d1ccc66d | 3927 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 3928 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3929 | |
78d7d407 | 3930 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3931 | capable(CAP_SYS_NICE)); |
3932 | } | |
3933 | ||
1da177e4 LT |
3934 | #ifdef __ARCH_WANT_SYS_NICE |
3935 | ||
3936 | /* | |
3937 | * sys_nice - change the priority of the current process. | |
3938 | * @increment: priority increment | |
3939 | * | |
3940 | * sys_setpriority is a more generic, but much slower function that | |
3941 | * does similar things. | |
3942 | */ | |
5add95d4 | 3943 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3944 | { |
48f24c4d | 3945 | long nice, retval; |
1da177e4 LT |
3946 | |
3947 | /* | |
3948 | * Setpriority might change our priority at the same moment. | |
3949 | * We don't have to worry. Conceptually one call occurs first | |
3950 | * and we have a single winner. | |
3951 | */ | |
a9467fa3 | 3952 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3953 | nice = task_nice(current) + increment; |
1da177e4 | 3954 | |
a9467fa3 | 3955 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3956 | if (increment < 0 && !can_nice(current, nice)) |
3957 | return -EPERM; | |
3958 | ||
1da177e4 LT |
3959 | retval = security_task_setnice(current, nice); |
3960 | if (retval) | |
3961 | return retval; | |
3962 | ||
3963 | set_user_nice(current, nice); | |
3964 | return 0; | |
3965 | } | |
3966 | ||
3967 | #endif | |
3968 | ||
3969 | /** | |
3970 | * task_prio - return the priority value of a given task. | |
3971 | * @p: the task in question. | |
3972 | * | |
e69f6186 | 3973 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3974 | * RT tasks are offset by -200. Normal tasks are centered |
3975 | * around 0, value goes from -16 to +15. | |
3976 | */ | |
36c8b586 | 3977 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3978 | { |
3979 | return p->prio - MAX_RT_PRIO; | |
3980 | } | |
3981 | ||
1da177e4 | 3982 | /** |
d1ccc66d | 3983 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 3984 | * @cpu: the processor in question. |
e69f6186 YB |
3985 | * |
3986 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3987 | */ |
3988 | int idle_cpu(int cpu) | |
3989 | { | |
908a3283 TG |
3990 | struct rq *rq = cpu_rq(cpu); |
3991 | ||
3992 | if (rq->curr != rq->idle) | |
3993 | return 0; | |
3994 | ||
3995 | if (rq->nr_running) | |
3996 | return 0; | |
3997 | ||
3998 | #ifdef CONFIG_SMP | |
3999 | if (!llist_empty(&rq->wake_list)) | |
4000 | return 0; | |
4001 | #endif | |
4002 | ||
4003 | return 1; | |
1da177e4 LT |
4004 | } |
4005 | ||
943d355d RJ |
4006 | /** |
4007 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
4008 | * @cpu: the CPU in question. | |
4009 | * | |
4010 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
4011 | */ | |
4012 | int available_idle_cpu(int cpu) | |
4013 | { | |
4014 | if (!idle_cpu(cpu)) | |
4015 | return 0; | |
4016 | ||
247f2f6f RJ |
4017 | if (vcpu_is_preempted(cpu)) |
4018 | return 0; | |
4019 | ||
908a3283 | 4020 | return 1; |
1da177e4 LT |
4021 | } |
4022 | ||
1da177e4 | 4023 | /** |
d1ccc66d | 4024 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 4025 | * @cpu: the processor in question. |
e69f6186 | 4026 | * |
d1ccc66d | 4027 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 4028 | */ |
36c8b586 | 4029 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4030 | { |
4031 | return cpu_rq(cpu)->idle; | |
4032 | } | |
4033 | ||
4034 | /** | |
4035 | * find_process_by_pid - find a process with a matching PID value. | |
4036 | * @pid: the pid in question. | |
e69f6186 YB |
4037 | * |
4038 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 4039 | */ |
a9957449 | 4040 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4041 | { |
228ebcbe | 4042 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4043 | } |
4044 | ||
c13db6b1 SR |
4045 | /* |
4046 | * sched_setparam() passes in -1 for its policy, to let the functions | |
4047 | * it calls know not to change it. | |
4048 | */ | |
4049 | #define SETPARAM_POLICY -1 | |
4050 | ||
c365c292 TG |
4051 | static void __setscheduler_params(struct task_struct *p, |
4052 | const struct sched_attr *attr) | |
1da177e4 | 4053 | { |
d50dde5a DF |
4054 | int policy = attr->sched_policy; |
4055 | ||
c13db6b1 | 4056 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
4057 | policy = p->policy; |
4058 | ||
1da177e4 | 4059 | p->policy = policy; |
d50dde5a | 4060 | |
aab03e05 DF |
4061 | if (dl_policy(policy)) |
4062 | __setparam_dl(p, attr); | |
39fd8fd2 | 4063 | else if (fair_policy(policy)) |
d50dde5a DF |
4064 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
4065 | ||
39fd8fd2 PZ |
4066 | /* |
4067 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
4068 | * !rt_policy. Always setting this ensures that things like | |
4069 | * getparam()/getattr() don't report silly values for !rt tasks. | |
4070 | */ | |
4071 | p->rt_priority = attr->sched_priority; | |
383afd09 | 4072 | p->normal_prio = normal_prio(p); |
9059393e | 4073 | set_load_weight(p, true); |
c365c292 | 4074 | } |
39fd8fd2 | 4075 | |
c365c292 TG |
4076 | /* Actually do priority change: must hold pi & rq lock. */ |
4077 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 4078 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
4079 | { |
4080 | __setscheduler_params(p, attr); | |
d50dde5a | 4081 | |
383afd09 | 4082 | /* |
0782e63b TG |
4083 | * Keep a potential priority boosting if called from |
4084 | * sched_setscheduler(). | |
383afd09 | 4085 | */ |
acd58620 | 4086 | p->prio = normal_prio(p); |
0782e63b | 4087 | if (keep_boost) |
acd58620 | 4088 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 4089 | |
aab03e05 DF |
4090 | if (dl_prio(p->prio)) |
4091 | p->sched_class = &dl_sched_class; | |
4092 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
4093 | p->sched_class = &rt_sched_class; |
4094 | else | |
4095 | p->sched_class = &fair_sched_class; | |
1da177e4 | 4096 | } |
aab03e05 | 4097 | |
c69e8d9c | 4098 | /* |
d1ccc66d | 4099 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
4100 | */ |
4101 | static bool check_same_owner(struct task_struct *p) | |
4102 | { | |
4103 | const struct cred *cred = current_cred(), *pcred; | |
4104 | bool match; | |
4105 | ||
4106 | rcu_read_lock(); | |
4107 | pcred = __task_cred(p); | |
9c806aa0 EB |
4108 | match = (uid_eq(cred->euid, pcred->euid) || |
4109 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4110 | rcu_read_unlock(); |
4111 | return match; | |
4112 | } | |
4113 | ||
d50dde5a DF |
4114 | static int __sched_setscheduler(struct task_struct *p, |
4115 | const struct sched_attr *attr, | |
dbc7f069 | 4116 | bool user, bool pi) |
1da177e4 | 4117 | { |
383afd09 SR |
4118 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4119 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4120 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4121 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4122 | const struct sched_class *prev_class; |
eb580751 | 4123 | struct rq_flags rf; |
ca94c442 | 4124 | int reset_on_fork; |
7a57f32a | 4125 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 4126 | struct rq *rq; |
1da177e4 | 4127 | |
896bbb25 SRV |
4128 | /* The pi code expects interrupts enabled */ |
4129 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 4130 | recheck: |
d1ccc66d | 4131 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
4132 | if (policy < 0) { |
4133 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4134 | policy = oldpolicy = p->policy; |
ca94c442 | 4135 | } else { |
7479f3c9 | 4136 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4137 | |
20f9cd2a | 4138 | if (!valid_policy(policy)) |
ca94c442 LP |
4139 | return -EINVAL; |
4140 | } | |
4141 | ||
794a56eb | 4142 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
4143 | return -EINVAL; |
4144 | ||
1da177e4 LT |
4145 | /* |
4146 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4147 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4148 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4149 | */ |
0bb040a4 | 4150 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4151 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4152 | return -EINVAL; |
aab03e05 DF |
4153 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4154 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4155 | return -EINVAL; |
4156 | ||
37e4ab3f OC |
4157 | /* |
4158 | * Allow unprivileged RT tasks to decrease priority: | |
4159 | */ | |
961ccddd | 4160 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4161 | if (fair_policy(policy)) { |
d0ea0268 | 4162 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4163 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4164 | return -EPERM; |
4165 | } | |
4166 | ||
e05606d3 | 4167 | if (rt_policy(policy)) { |
a44702e8 ON |
4168 | unsigned long rlim_rtprio = |
4169 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 4170 | |
d1ccc66d | 4171 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
4172 | if (policy != p->policy && !rlim_rtprio) |
4173 | return -EPERM; | |
4174 | ||
d1ccc66d | 4175 | /* Can't increase priority: */ |
d50dde5a DF |
4176 | if (attr->sched_priority > p->rt_priority && |
4177 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4178 | return -EPERM; |
4179 | } | |
c02aa73b | 4180 | |
d44753b8 JL |
4181 | /* |
4182 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4183 | * (safest behavior); in the future we would like to allow | |
4184 | * unprivileged DL tasks to increase their relative deadline | |
4185 | * or reduce their runtime (both ways reducing utilization) | |
4186 | */ | |
4187 | if (dl_policy(policy)) | |
4188 | return -EPERM; | |
4189 | ||
dd41f596 | 4190 | /* |
c02aa73b DH |
4191 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4192 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4193 | */ |
20f9cd2a | 4194 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4195 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4196 | return -EPERM; |
4197 | } | |
5fe1d75f | 4198 | |
d1ccc66d | 4199 | /* Can't change other user's priorities: */ |
c69e8d9c | 4200 | if (!check_same_owner(p)) |
37e4ab3f | 4201 | return -EPERM; |
ca94c442 | 4202 | |
d1ccc66d | 4203 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
4204 | if (p->sched_reset_on_fork && !reset_on_fork) |
4205 | return -EPERM; | |
37e4ab3f | 4206 | } |
1da177e4 | 4207 | |
725aad24 | 4208 | if (user) { |
794a56eb JL |
4209 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
4210 | return -EINVAL; | |
4211 | ||
b0ae1981 | 4212 | retval = security_task_setscheduler(p); |
725aad24 JF |
4213 | if (retval) |
4214 | return retval; | |
4215 | } | |
4216 | ||
b29739f9 | 4217 | /* |
d1ccc66d | 4218 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 4219 | * changing the priority of the task: |
0122ec5b | 4220 | * |
25985edc | 4221 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4222 | * runqueue lock must be held. |
4223 | */ | |
eb580751 | 4224 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 4225 | update_rq_clock(rq); |
dc61b1d6 | 4226 | |
34f971f6 | 4227 | /* |
d1ccc66d | 4228 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
4229 | */ |
4230 | if (p == rq->stop) { | |
eb580751 | 4231 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4232 | return -EINVAL; |
4233 | } | |
4234 | ||
a51e9198 | 4235 | /* |
d6b1e911 TG |
4236 | * If not changing anything there's no need to proceed further, |
4237 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4238 | */ |
d50dde5a | 4239 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4240 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4241 | goto change; |
4242 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4243 | goto change; | |
75381608 | 4244 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4245 | goto change; |
d50dde5a | 4246 | |
d6b1e911 | 4247 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4248 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4249 | return 0; |
4250 | } | |
d50dde5a | 4251 | change: |
a51e9198 | 4252 | |
dc61b1d6 | 4253 | if (user) { |
332ac17e | 4254 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4255 | /* |
4256 | * Do not allow realtime tasks into groups that have no runtime | |
4257 | * assigned. | |
4258 | */ | |
4259 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4260 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4261 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4262 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4263 | return -EPERM; |
4264 | } | |
dc61b1d6 | 4265 | #endif |
332ac17e | 4266 | #ifdef CONFIG_SMP |
794a56eb JL |
4267 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
4268 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 4269 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
4270 | |
4271 | /* | |
4272 | * Don't allow tasks with an affinity mask smaller than | |
4273 | * the entire root_domain to become SCHED_DEADLINE. We | |
4274 | * will also fail if there's no bandwidth available. | |
4275 | */ | |
e4099a5e PZ |
4276 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4277 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4278 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4279 | return -EPERM; |
4280 | } | |
4281 | } | |
4282 | #endif | |
4283 | } | |
dc61b1d6 | 4284 | |
d1ccc66d | 4285 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
4286 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
4287 | policy = oldpolicy = -1; | |
eb580751 | 4288 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4289 | goto recheck; |
4290 | } | |
332ac17e DF |
4291 | |
4292 | /* | |
4293 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4294 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4295 | * is available. | |
4296 | */ | |
06a76fe0 | 4297 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
eb580751 | 4298 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4299 | return -EBUSY; |
4300 | } | |
4301 | ||
c365c292 TG |
4302 | p->sched_reset_on_fork = reset_on_fork; |
4303 | oldprio = p->prio; | |
4304 | ||
dbc7f069 PZ |
4305 | if (pi) { |
4306 | /* | |
4307 | * Take priority boosted tasks into account. If the new | |
4308 | * effective priority is unchanged, we just store the new | |
4309 | * normal parameters and do not touch the scheduler class and | |
4310 | * the runqueue. This will be done when the task deboost | |
4311 | * itself. | |
4312 | */ | |
acd58620 | 4313 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
4314 | if (new_effective_prio == oldprio) |
4315 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4316 | } |
4317 | ||
da0c1e65 | 4318 | queued = task_on_rq_queued(p); |
051a1d1a | 4319 | running = task_current(rq, p); |
da0c1e65 | 4320 | if (queued) |
ff77e468 | 4321 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4322 | if (running) |
f3cd1c4e | 4323 | put_prev_task(rq, p); |
f6b53205 | 4324 | |
83ab0aa0 | 4325 | prev_class = p->sched_class; |
dbc7f069 | 4326 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4327 | |
da0c1e65 | 4328 | if (queued) { |
81a44c54 TG |
4329 | /* |
4330 | * We enqueue to tail when the priority of a task is | |
4331 | * increased (user space view). | |
4332 | */ | |
ff77e468 PZ |
4333 | if (oldprio < p->prio) |
4334 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4335 | |
ff77e468 | 4336 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4337 | } |
a399d233 | 4338 | if (running) |
b2bf6c31 | 4339 | set_curr_task(rq, p); |
cb469845 | 4340 | |
da7a735e | 4341 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
4342 | |
4343 | /* Avoid rq from going away on us: */ | |
4344 | preempt_disable(); | |
eb580751 | 4345 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4346 | |
dbc7f069 PZ |
4347 | if (pi) |
4348 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4349 | |
d1ccc66d | 4350 | /* Run balance callbacks after we've adjusted the PI chain: */ |
4c9a4bc8 PZ |
4351 | balance_callback(rq); |
4352 | preempt_enable(); | |
95e02ca9 | 4353 | |
1da177e4 LT |
4354 | return 0; |
4355 | } | |
961ccddd | 4356 | |
7479f3c9 PZ |
4357 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4358 | const struct sched_param *param, bool check) | |
4359 | { | |
4360 | struct sched_attr attr = { | |
4361 | .sched_policy = policy, | |
4362 | .sched_priority = param->sched_priority, | |
4363 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4364 | }; | |
4365 | ||
c13db6b1 SR |
4366 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4367 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4368 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4369 | policy &= ~SCHED_RESET_ON_FORK; | |
4370 | attr.sched_policy = policy; | |
4371 | } | |
4372 | ||
dbc7f069 | 4373 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4374 | } |
961ccddd RR |
4375 | /** |
4376 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4377 | * @p: the task in question. | |
4378 | * @policy: new policy. | |
4379 | * @param: structure containing the new RT priority. | |
4380 | * | |
e69f6186 YB |
4381 | * Return: 0 on success. An error code otherwise. |
4382 | * | |
961ccddd RR |
4383 | * NOTE that the task may be already dead. |
4384 | */ | |
4385 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4386 | const struct sched_param *param) |
961ccddd | 4387 | { |
7479f3c9 | 4388 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4389 | } |
1da177e4 LT |
4390 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4391 | ||
d50dde5a DF |
4392 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4393 | { | |
dbc7f069 | 4394 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4395 | } |
4396 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4397 | ||
794a56eb JL |
4398 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
4399 | { | |
4400 | return __sched_setscheduler(p, attr, false, true); | |
4401 | } | |
4402 | ||
961ccddd RR |
4403 | /** |
4404 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4405 | * @p: the task in question. | |
4406 | * @policy: new policy. | |
4407 | * @param: structure containing the new RT priority. | |
4408 | * | |
4409 | * Just like sched_setscheduler, only don't bother checking if the | |
4410 | * current context has permission. For example, this is needed in | |
4411 | * stop_machine(): we create temporary high priority worker threads, | |
4412 | * but our caller might not have that capability. | |
e69f6186 YB |
4413 | * |
4414 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4415 | */ |
4416 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4417 | const struct sched_param *param) |
961ccddd | 4418 | { |
7479f3c9 | 4419 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4420 | } |
84778472 | 4421 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4422 | |
95cdf3b7 IM |
4423 | static int |
4424 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4425 | { |
1da177e4 LT |
4426 | struct sched_param lparam; |
4427 | struct task_struct *p; | |
36c8b586 | 4428 | int retval; |
1da177e4 LT |
4429 | |
4430 | if (!param || pid < 0) | |
4431 | return -EINVAL; | |
4432 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4433 | return -EFAULT; | |
5fe1d75f ON |
4434 | |
4435 | rcu_read_lock(); | |
4436 | retval = -ESRCH; | |
1da177e4 | 4437 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4438 | if (p != NULL) |
4439 | retval = sched_setscheduler(p, policy, &lparam); | |
4440 | rcu_read_unlock(); | |
36c8b586 | 4441 | |
1da177e4 LT |
4442 | return retval; |
4443 | } | |
4444 | ||
d50dde5a DF |
4445 | /* |
4446 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4447 | */ | |
d1ccc66d | 4448 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
4449 | { |
4450 | u32 size; | |
4451 | int ret; | |
4452 | ||
4453 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4454 | return -EFAULT; | |
4455 | ||
d1ccc66d | 4456 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
4457 | memset(attr, 0, sizeof(*attr)); |
4458 | ||
4459 | ret = get_user(size, &uattr->size); | |
4460 | if (ret) | |
4461 | return ret; | |
4462 | ||
d1ccc66d IM |
4463 | /* Bail out on silly large: */ |
4464 | if (size > PAGE_SIZE) | |
d50dde5a DF |
4465 | goto err_size; |
4466 | ||
d1ccc66d IM |
4467 | /* ABI compatibility quirk: */ |
4468 | if (!size) | |
d50dde5a DF |
4469 | size = SCHED_ATTR_SIZE_VER0; |
4470 | ||
4471 | if (size < SCHED_ATTR_SIZE_VER0) | |
4472 | goto err_size; | |
4473 | ||
4474 | /* | |
4475 | * If we're handed a bigger struct than we know of, | |
4476 | * ensure all the unknown bits are 0 - i.e. new | |
4477 | * user-space does not rely on any kernel feature | |
4478 | * extensions we dont know about yet. | |
4479 | */ | |
4480 | if (size > sizeof(*attr)) { | |
4481 | unsigned char __user *addr; | |
4482 | unsigned char __user *end; | |
4483 | unsigned char val; | |
4484 | ||
4485 | addr = (void __user *)uattr + sizeof(*attr); | |
4486 | end = (void __user *)uattr + size; | |
4487 | ||
4488 | for (; addr < end; addr++) { | |
4489 | ret = get_user(val, addr); | |
4490 | if (ret) | |
4491 | return ret; | |
4492 | if (val) | |
4493 | goto err_size; | |
4494 | } | |
4495 | size = sizeof(*attr); | |
4496 | } | |
4497 | ||
4498 | ret = copy_from_user(attr, uattr, size); | |
4499 | if (ret) | |
4500 | return -EFAULT; | |
4501 | ||
4502 | /* | |
d1ccc66d | 4503 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
4504 | * to be strict and return an error on out-of-bounds values? |
4505 | */ | |
75e45d51 | 4506 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4507 | |
e78c7bca | 4508 | return 0; |
d50dde5a DF |
4509 | |
4510 | err_size: | |
4511 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4512 | return -E2BIG; |
d50dde5a DF |
4513 | } |
4514 | ||
1da177e4 LT |
4515 | /** |
4516 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4517 | * @pid: the pid in question. | |
4518 | * @policy: new policy. | |
4519 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4520 | * |
4521 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4522 | */ |
d1ccc66d | 4523 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 4524 | { |
c21761f1 JB |
4525 | if (policy < 0) |
4526 | return -EINVAL; | |
4527 | ||
1da177e4 LT |
4528 | return do_sched_setscheduler(pid, policy, param); |
4529 | } | |
4530 | ||
4531 | /** | |
4532 | * sys_sched_setparam - set/change the RT priority of a thread | |
4533 | * @pid: the pid in question. | |
4534 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4535 | * |
4536 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4537 | */ |
5add95d4 | 4538 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4539 | { |
c13db6b1 | 4540 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4541 | } |
4542 | ||
d50dde5a DF |
4543 | /** |
4544 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4545 | * @pid: the pid in question. | |
5778fccf | 4546 | * @uattr: structure containing the extended parameters. |
db66d756 | 4547 | * @flags: for future extension. |
d50dde5a | 4548 | */ |
6d35ab48 PZ |
4549 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4550 | unsigned int, flags) | |
d50dde5a DF |
4551 | { |
4552 | struct sched_attr attr; | |
4553 | struct task_struct *p; | |
4554 | int retval; | |
4555 | ||
6d35ab48 | 4556 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4557 | return -EINVAL; |
4558 | ||
143cf23d MK |
4559 | retval = sched_copy_attr(uattr, &attr); |
4560 | if (retval) | |
4561 | return retval; | |
d50dde5a | 4562 | |
b14ed2c2 | 4563 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4564 | return -EINVAL; |
d50dde5a DF |
4565 | |
4566 | rcu_read_lock(); | |
4567 | retval = -ESRCH; | |
4568 | p = find_process_by_pid(pid); | |
4569 | if (p != NULL) | |
4570 | retval = sched_setattr(p, &attr); | |
4571 | rcu_read_unlock(); | |
4572 | ||
4573 | return retval; | |
4574 | } | |
4575 | ||
1da177e4 LT |
4576 | /** |
4577 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4578 | * @pid: the pid in question. | |
e69f6186 YB |
4579 | * |
4580 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4581 | * code. | |
1da177e4 | 4582 | */ |
5add95d4 | 4583 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4584 | { |
36c8b586 | 4585 | struct task_struct *p; |
3a5c359a | 4586 | int retval; |
1da177e4 LT |
4587 | |
4588 | if (pid < 0) | |
3a5c359a | 4589 | return -EINVAL; |
1da177e4 LT |
4590 | |
4591 | retval = -ESRCH; | |
5fe85be0 | 4592 | rcu_read_lock(); |
1da177e4 LT |
4593 | p = find_process_by_pid(pid); |
4594 | if (p) { | |
4595 | retval = security_task_getscheduler(p); | |
4596 | if (!retval) | |
ca94c442 LP |
4597 | retval = p->policy |
4598 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4599 | } |
5fe85be0 | 4600 | rcu_read_unlock(); |
1da177e4 LT |
4601 | return retval; |
4602 | } | |
4603 | ||
4604 | /** | |
ca94c442 | 4605 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4606 | * @pid: the pid in question. |
4607 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4608 | * |
4609 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4610 | * code. | |
1da177e4 | 4611 | */ |
5add95d4 | 4612 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4613 | { |
ce5f7f82 | 4614 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4615 | struct task_struct *p; |
3a5c359a | 4616 | int retval; |
1da177e4 LT |
4617 | |
4618 | if (!param || pid < 0) | |
3a5c359a | 4619 | return -EINVAL; |
1da177e4 | 4620 | |
5fe85be0 | 4621 | rcu_read_lock(); |
1da177e4 LT |
4622 | p = find_process_by_pid(pid); |
4623 | retval = -ESRCH; | |
4624 | if (!p) | |
4625 | goto out_unlock; | |
4626 | ||
4627 | retval = security_task_getscheduler(p); | |
4628 | if (retval) | |
4629 | goto out_unlock; | |
4630 | ||
ce5f7f82 PZ |
4631 | if (task_has_rt_policy(p)) |
4632 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4633 | rcu_read_unlock(); |
1da177e4 LT |
4634 | |
4635 | /* | |
4636 | * This one might sleep, we cannot do it with a spinlock held ... | |
4637 | */ | |
4638 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4639 | ||
1da177e4 LT |
4640 | return retval; |
4641 | ||
4642 | out_unlock: | |
5fe85be0 | 4643 | rcu_read_unlock(); |
1da177e4 LT |
4644 | return retval; |
4645 | } | |
4646 | ||
d50dde5a DF |
4647 | static int sched_read_attr(struct sched_attr __user *uattr, |
4648 | struct sched_attr *attr, | |
4649 | unsigned int usize) | |
4650 | { | |
4651 | int ret; | |
4652 | ||
4653 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4654 | return -EFAULT; | |
4655 | ||
4656 | /* | |
4657 | * If we're handed a smaller struct than we know of, | |
4658 | * ensure all the unknown bits are 0 - i.e. old | |
4659 | * user-space does not get uncomplete information. | |
4660 | */ | |
4661 | if (usize < sizeof(*attr)) { | |
4662 | unsigned char *addr; | |
4663 | unsigned char *end; | |
4664 | ||
4665 | addr = (void *)attr + usize; | |
4666 | end = (void *)attr + sizeof(*attr); | |
4667 | ||
4668 | for (; addr < end; addr++) { | |
4669 | if (*addr) | |
22400674 | 4670 | return -EFBIG; |
d50dde5a DF |
4671 | } |
4672 | ||
4673 | attr->size = usize; | |
4674 | } | |
4675 | ||
4efbc454 | 4676 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4677 | if (ret) |
4678 | return -EFAULT; | |
4679 | ||
22400674 | 4680 | return 0; |
d50dde5a DF |
4681 | } |
4682 | ||
4683 | /** | |
aab03e05 | 4684 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4685 | * @pid: the pid in question. |
5778fccf | 4686 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4687 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4688 | * @flags: for future extension. |
d50dde5a | 4689 | */ |
6d35ab48 PZ |
4690 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4691 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4692 | { |
4693 | struct sched_attr attr = { | |
4694 | .size = sizeof(struct sched_attr), | |
4695 | }; | |
4696 | struct task_struct *p; | |
4697 | int retval; | |
4698 | ||
4699 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4700 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4701 | return -EINVAL; |
4702 | ||
4703 | rcu_read_lock(); | |
4704 | p = find_process_by_pid(pid); | |
4705 | retval = -ESRCH; | |
4706 | if (!p) | |
4707 | goto out_unlock; | |
4708 | ||
4709 | retval = security_task_getscheduler(p); | |
4710 | if (retval) | |
4711 | goto out_unlock; | |
4712 | ||
4713 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4714 | if (p->sched_reset_on_fork) |
4715 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4716 | if (task_has_dl_policy(p)) |
4717 | __getparam_dl(p, &attr); | |
4718 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4719 | attr.sched_priority = p->rt_priority; |
4720 | else | |
d0ea0268 | 4721 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4722 | |
4723 | rcu_read_unlock(); | |
4724 | ||
4725 | retval = sched_read_attr(uattr, &attr, size); | |
4726 | return retval; | |
4727 | ||
4728 | out_unlock: | |
4729 | rcu_read_unlock(); | |
4730 | return retval; | |
4731 | } | |
4732 | ||
96f874e2 | 4733 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4734 | { |
5a16f3d3 | 4735 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4736 | struct task_struct *p; |
4737 | int retval; | |
1da177e4 | 4738 | |
23f5d142 | 4739 | rcu_read_lock(); |
1da177e4 LT |
4740 | |
4741 | p = find_process_by_pid(pid); | |
4742 | if (!p) { | |
23f5d142 | 4743 | rcu_read_unlock(); |
1da177e4 LT |
4744 | return -ESRCH; |
4745 | } | |
4746 | ||
23f5d142 | 4747 | /* Prevent p going away */ |
1da177e4 | 4748 | get_task_struct(p); |
23f5d142 | 4749 | rcu_read_unlock(); |
1da177e4 | 4750 | |
14a40ffc TH |
4751 | if (p->flags & PF_NO_SETAFFINITY) { |
4752 | retval = -EINVAL; | |
4753 | goto out_put_task; | |
4754 | } | |
5a16f3d3 RR |
4755 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4756 | retval = -ENOMEM; | |
4757 | goto out_put_task; | |
4758 | } | |
4759 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4760 | retval = -ENOMEM; | |
4761 | goto out_free_cpus_allowed; | |
4762 | } | |
1da177e4 | 4763 | retval = -EPERM; |
4c44aaaf EB |
4764 | if (!check_same_owner(p)) { |
4765 | rcu_read_lock(); | |
4766 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4767 | rcu_read_unlock(); | |
16303ab2 | 4768 | goto out_free_new_mask; |
4c44aaaf EB |
4769 | } |
4770 | rcu_read_unlock(); | |
4771 | } | |
1da177e4 | 4772 | |
b0ae1981 | 4773 | retval = security_task_setscheduler(p); |
e7834f8f | 4774 | if (retval) |
16303ab2 | 4775 | goto out_free_new_mask; |
e7834f8f | 4776 | |
e4099a5e PZ |
4777 | |
4778 | cpuset_cpus_allowed(p, cpus_allowed); | |
4779 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4780 | ||
332ac17e DF |
4781 | /* |
4782 | * Since bandwidth control happens on root_domain basis, | |
4783 | * if admission test is enabled, we only admit -deadline | |
4784 | * tasks allowed to run on all the CPUs in the task's | |
4785 | * root_domain. | |
4786 | */ | |
4787 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4788 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4789 | rcu_read_lock(); | |
4790 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4791 | retval = -EBUSY; |
f1e3a093 | 4792 | rcu_read_unlock(); |
16303ab2 | 4793 | goto out_free_new_mask; |
332ac17e | 4794 | } |
f1e3a093 | 4795 | rcu_read_unlock(); |
332ac17e DF |
4796 | } |
4797 | #endif | |
49246274 | 4798 | again: |
25834c73 | 4799 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4800 | |
8707d8b8 | 4801 | if (!retval) { |
5a16f3d3 RR |
4802 | cpuset_cpus_allowed(p, cpus_allowed); |
4803 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4804 | /* |
4805 | * We must have raced with a concurrent cpuset | |
4806 | * update. Just reset the cpus_allowed to the | |
4807 | * cpuset's cpus_allowed | |
4808 | */ | |
5a16f3d3 | 4809 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4810 | goto again; |
4811 | } | |
4812 | } | |
16303ab2 | 4813 | out_free_new_mask: |
5a16f3d3 RR |
4814 | free_cpumask_var(new_mask); |
4815 | out_free_cpus_allowed: | |
4816 | free_cpumask_var(cpus_allowed); | |
4817 | out_put_task: | |
1da177e4 | 4818 | put_task_struct(p); |
1da177e4 LT |
4819 | return retval; |
4820 | } | |
4821 | ||
4822 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4823 | struct cpumask *new_mask) |
1da177e4 | 4824 | { |
96f874e2 RR |
4825 | if (len < cpumask_size()) |
4826 | cpumask_clear(new_mask); | |
4827 | else if (len > cpumask_size()) | |
4828 | len = cpumask_size(); | |
4829 | ||
1da177e4 LT |
4830 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4831 | } | |
4832 | ||
4833 | /** | |
d1ccc66d | 4834 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
4835 | * @pid: pid of the process |
4836 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4837 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
4838 | * |
4839 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4840 | */ |
5add95d4 HC |
4841 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4842 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4843 | { |
5a16f3d3 | 4844 | cpumask_var_t new_mask; |
1da177e4 LT |
4845 | int retval; |
4846 | ||
5a16f3d3 RR |
4847 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4848 | return -ENOMEM; | |
1da177e4 | 4849 | |
5a16f3d3 RR |
4850 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4851 | if (retval == 0) | |
4852 | retval = sched_setaffinity(pid, new_mask); | |
4853 | free_cpumask_var(new_mask); | |
4854 | return retval; | |
1da177e4 LT |
4855 | } |
4856 | ||
96f874e2 | 4857 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4858 | { |
36c8b586 | 4859 | struct task_struct *p; |
31605683 | 4860 | unsigned long flags; |
1da177e4 | 4861 | int retval; |
1da177e4 | 4862 | |
23f5d142 | 4863 | rcu_read_lock(); |
1da177e4 LT |
4864 | |
4865 | retval = -ESRCH; | |
4866 | p = find_process_by_pid(pid); | |
4867 | if (!p) | |
4868 | goto out_unlock; | |
4869 | ||
e7834f8f DQ |
4870 | retval = security_task_getscheduler(p); |
4871 | if (retval) | |
4872 | goto out_unlock; | |
4873 | ||
013fdb80 | 4874 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4875 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4876 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4877 | |
4878 | out_unlock: | |
23f5d142 | 4879 | rcu_read_unlock(); |
1da177e4 | 4880 | |
9531b62f | 4881 | return retval; |
1da177e4 LT |
4882 | } |
4883 | ||
4884 | /** | |
d1ccc66d | 4885 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
4886 | * @pid: pid of the process |
4887 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4888 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 4889 | * |
599b4840 ZW |
4890 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
4891 | * error code otherwise. | |
1da177e4 | 4892 | */ |
5add95d4 HC |
4893 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4894 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4895 | { |
4896 | int ret; | |
f17c8607 | 4897 | cpumask_var_t mask; |
1da177e4 | 4898 | |
84fba5ec | 4899 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4900 | return -EINVAL; |
4901 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4902 | return -EINVAL; |
4903 | ||
f17c8607 RR |
4904 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4905 | return -ENOMEM; | |
1da177e4 | 4906 | |
f17c8607 RR |
4907 | ret = sched_getaffinity(pid, mask); |
4908 | if (ret == 0) { | |
4de373a1 | 4909 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
4910 | |
4911 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4912 | ret = -EFAULT; |
4913 | else | |
cd3d8031 | 4914 | ret = retlen; |
f17c8607 RR |
4915 | } |
4916 | free_cpumask_var(mask); | |
1da177e4 | 4917 | |
f17c8607 | 4918 | return ret; |
1da177e4 LT |
4919 | } |
4920 | ||
4921 | /** | |
4922 | * sys_sched_yield - yield the current processor to other threads. | |
4923 | * | |
dd41f596 IM |
4924 | * This function yields the current CPU to other tasks. If there are no |
4925 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4926 | * |
4927 | * Return: 0. | |
1da177e4 | 4928 | */ |
7d4dd4f1 | 4929 | static void do_sched_yield(void) |
1da177e4 | 4930 | { |
8a8c69c3 PZ |
4931 | struct rq_flags rf; |
4932 | struct rq *rq; | |
4933 | ||
246b3b33 | 4934 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 4935 | |
ae92882e | 4936 | schedstat_inc(rq->yld_count); |
4530d7ab | 4937 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4938 | |
4939 | /* | |
4940 | * Since we are going to call schedule() anyway, there's | |
4941 | * no need to preempt or enable interrupts: | |
4942 | */ | |
8a8c69c3 PZ |
4943 | preempt_disable(); |
4944 | rq_unlock(rq, &rf); | |
ba74c144 | 4945 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4946 | |
4947 | schedule(); | |
7d4dd4f1 | 4948 | } |
1da177e4 | 4949 | |
7d4dd4f1 DB |
4950 | SYSCALL_DEFINE0(sched_yield) |
4951 | { | |
4952 | do_sched_yield(); | |
1da177e4 LT |
4953 | return 0; |
4954 | } | |
4955 | ||
35a773a0 | 4956 | #ifndef CONFIG_PREEMPT |
02b67cc3 | 4957 | int __sched _cond_resched(void) |
1da177e4 | 4958 | { |
fe32d3cd | 4959 | if (should_resched(0)) { |
a18b5d01 | 4960 | preempt_schedule_common(); |
1da177e4 LT |
4961 | return 1; |
4962 | } | |
f79c3ad6 | 4963 | rcu_all_qs(); |
1da177e4 LT |
4964 | return 0; |
4965 | } | |
02b67cc3 | 4966 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 4967 | #endif |
1da177e4 LT |
4968 | |
4969 | /* | |
613afbf8 | 4970 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4971 | * call schedule, and on return reacquire the lock. |
4972 | * | |
41a2d6cf | 4973 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4974 | * operations here to prevent schedule() from being called twice (once via |
4975 | * spin_unlock(), once by hand). | |
4976 | */ | |
613afbf8 | 4977 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4978 | { |
fe32d3cd | 4979 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4980 | int ret = 0; |
4981 | ||
f607c668 PZ |
4982 | lockdep_assert_held(lock); |
4983 | ||
4a81e832 | 4984 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4985 | spin_unlock(lock); |
d86ee480 | 4986 | if (resched) |
a18b5d01 | 4987 | preempt_schedule_common(); |
95c354fe NP |
4988 | else |
4989 | cpu_relax(); | |
6df3cecb | 4990 | ret = 1; |
1da177e4 | 4991 | spin_lock(lock); |
1da177e4 | 4992 | } |
6df3cecb | 4993 | return ret; |
1da177e4 | 4994 | } |
613afbf8 | 4995 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4996 | |
1da177e4 LT |
4997 | /** |
4998 | * yield - yield the current processor to other threads. | |
4999 | * | |
8e3fabfd PZ |
5000 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
5001 | * | |
5002 | * The scheduler is at all times free to pick the calling task as the most | |
5003 | * eligible task to run, if removing the yield() call from your code breaks | |
5004 | * it, its already broken. | |
5005 | * | |
5006 | * Typical broken usage is: | |
5007 | * | |
5008 | * while (!event) | |
d1ccc66d | 5009 | * yield(); |
8e3fabfd PZ |
5010 | * |
5011 | * where one assumes that yield() will let 'the other' process run that will | |
5012 | * make event true. If the current task is a SCHED_FIFO task that will never | |
5013 | * happen. Never use yield() as a progress guarantee!! | |
5014 | * | |
5015 | * If you want to use yield() to wait for something, use wait_event(). | |
5016 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
5017 | * If you still want to use yield(), do not! | |
1da177e4 LT |
5018 | */ |
5019 | void __sched yield(void) | |
5020 | { | |
5021 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 5022 | do_sched_yield(); |
1da177e4 | 5023 | } |
1da177e4 LT |
5024 | EXPORT_SYMBOL(yield); |
5025 | ||
d95f4122 MG |
5026 | /** |
5027 | * yield_to - yield the current processor to another thread in | |
5028 | * your thread group, or accelerate that thread toward the | |
5029 | * processor it's on. | |
16addf95 RD |
5030 | * @p: target task |
5031 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5032 | * |
5033 | * It's the caller's job to ensure that the target task struct | |
5034 | * can't go away on us before we can do any checks. | |
5035 | * | |
e69f6186 | 5036 | * Return: |
7b270f60 PZ |
5037 | * true (>0) if we indeed boosted the target task. |
5038 | * false (0) if we failed to boost the target. | |
5039 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 5040 | */ |
fa93384f | 5041 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
5042 | { |
5043 | struct task_struct *curr = current; | |
5044 | struct rq *rq, *p_rq; | |
5045 | unsigned long flags; | |
c3c18640 | 5046 | int yielded = 0; |
d95f4122 MG |
5047 | |
5048 | local_irq_save(flags); | |
5049 | rq = this_rq(); | |
5050 | ||
5051 | again: | |
5052 | p_rq = task_rq(p); | |
7b270f60 PZ |
5053 | /* |
5054 | * If we're the only runnable task on the rq and target rq also | |
5055 | * has only one task, there's absolutely no point in yielding. | |
5056 | */ | |
5057 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
5058 | yielded = -ESRCH; | |
5059 | goto out_irq; | |
5060 | } | |
5061 | ||
d95f4122 | 5062 | double_rq_lock(rq, p_rq); |
39e24d8f | 5063 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
5064 | double_rq_unlock(rq, p_rq); |
5065 | goto again; | |
5066 | } | |
5067 | ||
5068 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 5069 | goto out_unlock; |
d95f4122 MG |
5070 | |
5071 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 5072 | goto out_unlock; |
d95f4122 MG |
5073 | |
5074 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 5075 | goto out_unlock; |
d95f4122 MG |
5076 | |
5077 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5078 | if (yielded) { |
ae92882e | 5079 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5080 | /* |
5081 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5082 | * fairness. | |
5083 | */ | |
5084 | if (preempt && rq != p_rq) | |
8875125e | 5085 | resched_curr(p_rq); |
6d1cafd8 | 5086 | } |
d95f4122 | 5087 | |
7b270f60 | 5088 | out_unlock: |
d95f4122 | 5089 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5090 | out_irq: |
d95f4122 MG |
5091 | local_irq_restore(flags); |
5092 | ||
7b270f60 | 5093 | if (yielded > 0) |
d95f4122 MG |
5094 | schedule(); |
5095 | ||
5096 | return yielded; | |
5097 | } | |
5098 | EXPORT_SYMBOL_GPL(yield_to); | |
5099 | ||
10ab5643 TH |
5100 | int io_schedule_prepare(void) |
5101 | { | |
5102 | int old_iowait = current->in_iowait; | |
5103 | ||
5104 | current->in_iowait = 1; | |
5105 | blk_schedule_flush_plug(current); | |
5106 | ||
5107 | return old_iowait; | |
5108 | } | |
5109 | ||
5110 | void io_schedule_finish(int token) | |
5111 | { | |
5112 | current->in_iowait = token; | |
5113 | } | |
5114 | ||
1da177e4 | 5115 | /* |
41a2d6cf | 5116 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5117 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5118 | */ |
1da177e4 LT |
5119 | long __sched io_schedule_timeout(long timeout) |
5120 | { | |
10ab5643 | 5121 | int token; |
1da177e4 LT |
5122 | long ret; |
5123 | ||
10ab5643 | 5124 | token = io_schedule_prepare(); |
1da177e4 | 5125 | ret = schedule_timeout(timeout); |
10ab5643 | 5126 | io_schedule_finish(token); |
9cff8ade | 5127 | |
1da177e4 LT |
5128 | return ret; |
5129 | } | |
9cff8ade | 5130 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 5131 | |
10ab5643 TH |
5132 | void io_schedule(void) |
5133 | { | |
5134 | int token; | |
5135 | ||
5136 | token = io_schedule_prepare(); | |
5137 | schedule(); | |
5138 | io_schedule_finish(token); | |
5139 | } | |
5140 | EXPORT_SYMBOL(io_schedule); | |
5141 | ||
1da177e4 LT |
5142 | /** |
5143 | * sys_sched_get_priority_max - return maximum RT priority. | |
5144 | * @policy: scheduling class. | |
5145 | * | |
e69f6186 YB |
5146 | * Return: On success, this syscall returns the maximum |
5147 | * rt_priority that can be used by a given scheduling class. | |
5148 | * On failure, a negative error code is returned. | |
1da177e4 | 5149 | */ |
5add95d4 | 5150 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5151 | { |
5152 | int ret = -EINVAL; | |
5153 | ||
5154 | switch (policy) { | |
5155 | case SCHED_FIFO: | |
5156 | case SCHED_RR: | |
5157 | ret = MAX_USER_RT_PRIO-1; | |
5158 | break; | |
aab03e05 | 5159 | case SCHED_DEADLINE: |
1da177e4 | 5160 | case SCHED_NORMAL: |
b0a9499c | 5161 | case SCHED_BATCH: |
dd41f596 | 5162 | case SCHED_IDLE: |
1da177e4 LT |
5163 | ret = 0; |
5164 | break; | |
5165 | } | |
5166 | return ret; | |
5167 | } | |
5168 | ||
5169 | /** | |
5170 | * sys_sched_get_priority_min - return minimum RT priority. | |
5171 | * @policy: scheduling class. | |
5172 | * | |
e69f6186 YB |
5173 | * Return: On success, this syscall returns the minimum |
5174 | * rt_priority that can be used by a given scheduling class. | |
5175 | * On failure, a negative error code is returned. | |
1da177e4 | 5176 | */ |
5add95d4 | 5177 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5178 | { |
5179 | int ret = -EINVAL; | |
5180 | ||
5181 | switch (policy) { | |
5182 | case SCHED_FIFO: | |
5183 | case SCHED_RR: | |
5184 | ret = 1; | |
5185 | break; | |
aab03e05 | 5186 | case SCHED_DEADLINE: |
1da177e4 | 5187 | case SCHED_NORMAL: |
b0a9499c | 5188 | case SCHED_BATCH: |
dd41f596 | 5189 | case SCHED_IDLE: |
1da177e4 LT |
5190 | ret = 0; |
5191 | } | |
5192 | return ret; | |
5193 | } | |
5194 | ||
abca5fc5 | 5195 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 5196 | { |
36c8b586 | 5197 | struct task_struct *p; |
a4ec24b4 | 5198 | unsigned int time_slice; |
eb580751 | 5199 | struct rq_flags rf; |
dba091b9 | 5200 | struct rq *rq; |
3a5c359a | 5201 | int retval; |
1da177e4 LT |
5202 | |
5203 | if (pid < 0) | |
3a5c359a | 5204 | return -EINVAL; |
1da177e4 LT |
5205 | |
5206 | retval = -ESRCH; | |
1a551ae7 | 5207 | rcu_read_lock(); |
1da177e4 LT |
5208 | p = find_process_by_pid(pid); |
5209 | if (!p) | |
5210 | goto out_unlock; | |
5211 | ||
5212 | retval = security_task_getscheduler(p); | |
5213 | if (retval) | |
5214 | goto out_unlock; | |
5215 | ||
eb580751 | 5216 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5217 | time_slice = 0; |
5218 | if (p->sched_class->get_rr_interval) | |
5219 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5220 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5221 | |
1a551ae7 | 5222 | rcu_read_unlock(); |
abca5fc5 AV |
5223 | jiffies_to_timespec64(time_slice, t); |
5224 | return 0; | |
3a5c359a | 5225 | |
1da177e4 | 5226 | out_unlock: |
1a551ae7 | 5227 | rcu_read_unlock(); |
1da177e4 LT |
5228 | return retval; |
5229 | } | |
5230 | ||
2064a5ab RD |
5231 | /** |
5232 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5233 | * @pid: pid of the process. | |
5234 | * @interval: userspace pointer to the timeslice value. | |
5235 | * | |
5236 | * this syscall writes the default timeslice value of a given process | |
5237 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5238 | * | |
5239 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5240 | * an error code. | |
5241 | */ | |
abca5fc5 | 5242 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 5243 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
5244 | { |
5245 | struct timespec64 t; | |
5246 | int retval = sched_rr_get_interval(pid, &t); | |
5247 | ||
5248 | if (retval == 0) | |
5249 | retval = put_timespec64(&t, interval); | |
5250 | ||
5251 | return retval; | |
5252 | } | |
5253 | ||
474b9c77 | 5254 | #ifdef CONFIG_COMPAT_32BIT_TIME |
abca5fc5 AV |
5255 | COMPAT_SYSCALL_DEFINE2(sched_rr_get_interval, |
5256 | compat_pid_t, pid, | |
9afc5eee | 5257 | struct old_timespec32 __user *, interval) |
abca5fc5 AV |
5258 | { |
5259 | struct timespec64 t; | |
5260 | int retval = sched_rr_get_interval(pid, &t); | |
5261 | ||
5262 | if (retval == 0) | |
9afc5eee | 5263 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
5264 | return retval; |
5265 | } | |
5266 | #endif | |
5267 | ||
82a1fcb9 | 5268 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5269 | { |
1da177e4 | 5270 | unsigned long free = 0; |
4e79752c | 5271 | int ppid; |
c930b2c0 | 5272 | |
38200502 TH |
5273 | if (!try_get_task_stack(p)) |
5274 | return; | |
20435d84 XX |
5275 | |
5276 | printk(KERN_INFO "%-15.15s %c", p->comm, task_state_to_char(p)); | |
5277 | ||
5278 | if (p->state == TASK_RUNNING) | |
3df0fc5b | 5279 | printk(KERN_CONT " running task "); |
1da177e4 | 5280 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 5281 | free = stack_not_used(p); |
1da177e4 | 5282 | #endif |
a90e984c | 5283 | ppid = 0; |
4e79752c | 5284 | rcu_read_lock(); |
a90e984c ON |
5285 | if (pid_alive(p)) |
5286 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5287 | rcu_read_unlock(); |
3df0fc5b | 5288 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5289 | task_pid_nr(p), ppid, |
aa47b7e0 | 5290 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5291 | |
3d1cb205 | 5292 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5293 | show_stack(p, NULL); |
38200502 | 5294 | put_task_stack(p); |
1da177e4 | 5295 | } |
0032f4e8 | 5296 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 5297 | |
5d68cc95 PZ |
5298 | static inline bool |
5299 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
5300 | { | |
5301 | /* no filter, everything matches */ | |
5302 | if (!state_filter) | |
5303 | return true; | |
5304 | ||
5305 | /* filter, but doesn't match */ | |
5306 | if (!(p->state & state_filter)) | |
5307 | return false; | |
5308 | ||
5309 | /* | |
5310 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
5311 | * TASK_KILLABLE). | |
5312 | */ | |
5313 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
5314 | return false; | |
5315 | ||
5316 | return true; | |
5317 | } | |
5318 | ||
5319 | ||
e59e2ae2 | 5320 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5321 | { |
36c8b586 | 5322 | struct task_struct *g, *p; |
1da177e4 | 5323 | |
4bd77321 | 5324 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5325 | printk(KERN_INFO |
5326 | " task PC stack pid father\n"); | |
1da177e4 | 5327 | #else |
3df0fc5b PZ |
5328 | printk(KERN_INFO |
5329 | " task PC stack pid father\n"); | |
1da177e4 | 5330 | #endif |
510f5acc | 5331 | rcu_read_lock(); |
5d07f420 | 5332 | for_each_process_thread(g, p) { |
1da177e4 LT |
5333 | /* |
5334 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5335 | * console might take a lot of time: |
57675cb9 AR |
5336 | * Also, reset softlockup watchdogs on all CPUs, because |
5337 | * another CPU might be blocked waiting for us to process | |
5338 | * an IPI. | |
1da177e4 LT |
5339 | */ |
5340 | touch_nmi_watchdog(); | |
57675cb9 | 5341 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 5342 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 5343 | sched_show_task(p); |
5d07f420 | 5344 | } |
1da177e4 | 5345 | |
dd41f596 | 5346 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5347 | if (!state_filter) |
5348 | sysrq_sched_debug_show(); | |
dd41f596 | 5349 | #endif |
510f5acc | 5350 | rcu_read_unlock(); |
e59e2ae2 IM |
5351 | /* |
5352 | * Only show locks if all tasks are dumped: | |
5353 | */ | |
93335a21 | 5354 | if (!state_filter) |
e59e2ae2 | 5355 | debug_show_all_locks(); |
1da177e4 LT |
5356 | } |
5357 | ||
f340c0d1 IM |
5358 | /** |
5359 | * init_idle - set up an idle thread for a given CPU | |
5360 | * @idle: task in question | |
d1ccc66d | 5361 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
5362 | * |
5363 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5364 | * flag, to make booting more robust. | |
5365 | */ | |
0db0628d | 5366 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5367 | { |
70b97a7f | 5368 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5369 | unsigned long flags; |
5370 | ||
25834c73 PZ |
5371 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5372 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5373 | |
5e1576ed | 5374 | __sched_fork(0, idle); |
06b83b5f | 5375 | idle->state = TASK_RUNNING; |
dd41f596 | 5376 | idle->se.exec_start = sched_clock(); |
c1de45ca | 5377 | idle->flags |= PF_IDLE; |
dd41f596 | 5378 | |
e1b77c92 MR |
5379 | kasan_unpoison_task_stack(idle); |
5380 | ||
de9b8f5d PZ |
5381 | #ifdef CONFIG_SMP |
5382 | /* | |
5383 | * Its possible that init_idle() gets called multiple times on a task, | |
5384 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5385 | * | |
5386 | * And since this is boot we can forgo the serialization. | |
5387 | */ | |
5388 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5389 | #endif | |
6506cf6c PZ |
5390 | /* |
5391 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 5392 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
5393 | * lockdep check in task_group() will fail. |
5394 | * | |
5395 | * Similar case to sched_fork(). / Alternatively we could | |
5396 | * use task_rq_lock() here and obtain the other rq->lock. | |
5397 | * | |
5398 | * Silence PROVE_RCU | |
5399 | */ | |
5400 | rcu_read_lock(); | |
dd41f596 | 5401 | __set_task_cpu(idle, cpu); |
6506cf6c | 5402 | rcu_read_unlock(); |
1da177e4 | 5403 | |
1da177e4 | 5404 | rq->curr = rq->idle = idle; |
da0c1e65 | 5405 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5406 | #ifdef CONFIG_SMP |
3ca7a440 | 5407 | idle->on_cpu = 1; |
4866cde0 | 5408 | #endif |
25834c73 PZ |
5409 | raw_spin_unlock(&rq->lock); |
5410 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5411 | |
5412 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5413 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5414 | |
dd41f596 IM |
5415 | /* |
5416 | * The idle tasks have their own, simple scheduling class: | |
5417 | */ | |
5418 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5419 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5420 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5421 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5422 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5423 | #endif | |
19978ca6 IM |
5424 | } |
5425 | ||
e1d4eeec NP |
5426 | #ifdef CONFIG_SMP |
5427 | ||
f82f8042 JL |
5428 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5429 | const struct cpumask *trial) | |
5430 | { | |
06a76fe0 | 5431 | int ret = 1; |
f82f8042 | 5432 | |
bb2bc55a MG |
5433 | if (!cpumask_weight(cur)) |
5434 | return ret; | |
5435 | ||
06a76fe0 | 5436 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
5437 | |
5438 | return ret; | |
5439 | } | |
5440 | ||
7f51412a JL |
5441 | int task_can_attach(struct task_struct *p, |
5442 | const struct cpumask *cs_cpus_allowed) | |
5443 | { | |
5444 | int ret = 0; | |
5445 | ||
5446 | /* | |
5447 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 5448 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
5449 | * affinity and isolating such threads by their set of |
5450 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5451 | * applicable for such threads. This prevents checking for | |
5452 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5453 | * before cpus_allowed may be changed. | |
5454 | */ | |
5455 | if (p->flags & PF_NO_SETAFFINITY) { | |
5456 | ret = -EINVAL; | |
5457 | goto out; | |
5458 | } | |
5459 | ||
7f51412a | 5460 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
5461 | cs_cpus_allowed)) |
5462 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 5463 | |
7f51412a JL |
5464 | out: |
5465 | return ret; | |
5466 | } | |
5467 | ||
f2cb1360 | 5468 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 5469 | |
e6628d5b MG |
5470 | #ifdef CONFIG_NUMA_BALANCING |
5471 | /* Migrate current task p to target_cpu */ | |
5472 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5473 | { | |
5474 | struct migration_arg arg = { p, target_cpu }; | |
5475 | int curr_cpu = task_cpu(p); | |
5476 | ||
5477 | if (curr_cpu == target_cpu) | |
5478 | return 0; | |
5479 | ||
0c98d344 | 5480 | if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed)) |
e6628d5b MG |
5481 | return -EINVAL; |
5482 | ||
5483 | /* TODO: This is not properly updating schedstats */ | |
5484 | ||
286549dc | 5485 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5486 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5487 | } | |
0ec8aa00 PZ |
5488 | |
5489 | /* | |
5490 | * Requeue a task on a given node and accurately track the number of NUMA | |
5491 | * tasks on the runqueues | |
5492 | */ | |
5493 | void sched_setnuma(struct task_struct *p, int nid) | |
5494 | { | |
da0c1e65 | 5495 | bool queued, running; |
eb580751 PZ |
5496 | struct rq_flags rf; |
5497 | struct rq *rq; | |
0ec8aa00 | 5498 | |
eb580751 | 5499 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5500 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5501 | running = task_current(rq, p); |
5502 | ||
da0c1e65 | 5503 | if (queued) |
1de64443 | 5504 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5505 | if (running) |
f3cd1c4e | 5506 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5507 | |
5508 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 5509 | |
da0c1e65 | 5510 | if (queued) |
7134b3e9 | 5511 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 5512 | if (running) |
b2bf6c31 | 5513 | set_curr_task(rq, p); |
eb580751 | 5514 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5515 | } |
5cc389bc | 5516 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5517 | |
1da177e4 | 5518 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5519 | /* |
d1ccc66d | 5520 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 5521 | * offline. |
054b9108 | 5522 | */ |
48c5ccae | 5523 | void idle_task_exit(void) |
1da177e4 | 5524 | { |
48c5ccae | 5525 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5526 | |
48c5ccae | 5527 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5528 | |
a53efe5f | 5529 | if (mm != &init_mm) { |
252d2a41 | 5530 | switch_mm(mm, &init_mm, current); |
3eda69c9 | 5531 | current->active_mm = &init_mm; |
a53efe5f MS |
5532 | finish_arch_post_lock_switch(); |
5533 | } | |
48c5ccae | 5534 | mmdrop(mm); |
1da177e4 LT |
5535 | } |
5536 | ||
5537 | /* | |
5d180232 PZ |
5538 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5539 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
5540 | * nr_active count is stable. We need to take the teardown thread which |
5541 | * is calling this into account, so we hand in adjust = 1 to the load | |
5542 | * calculation. | |
5d180232 PZ |
5543 | * |
5544 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5545 | */ |
5d180232 | 5546 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5547 | { |
d60585c5 | 5548 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
5549 | if (delta) |
5550 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5551 | } |
5552 | ||
3f1d2a31 PZ |
5553 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5554 | { | |
5555 | } | |
5556 | ||
5557 | static const struct sched_class fake_sched_class = { | |
5558 | .put_prev_task = put_prev_task_fake, | |
5559 | }; | |
5560 | ||
5561 | static struct task_struct fake_task = { | |
5562 | /* | |
5563 | * Avoid pull_{rt,dl}_task() | |
5564 | */ | |
5565 | .prio = MAX_PRIO + 1, | |
5566 | .sched_class = &fake_sched_class, | |
5567 | }; | |
5568 | ||
48f24c4d | 5569 | /* |
48c5ccae PZ |
5570 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5571 | * try_to_wake_up()->select_task_rq(). | |
5572 | * | |
5573 | * Called with rq->lock held even though we'er in stop_machine() and | |
5574 | * there's no concurrency possible, we hold the required locks anyway | |
5575 | * because of lock validation efforts. | |
1da177e4 | 5576 | */ |
8a8c69c3 | 5577 | static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf) |
1da177e4 | 5578 | { |
5e16bbc2 | 5579 | struct rq *rq = dead_rq; |
48c5ccae | 5580 | struct task_struct *next, *stop = rq->stop; |
8a8c69c3 | 5581 | struct rq_flags orf = *rf; |
48c5ccae | 5582 | int dest_cpu; |
1da177e4 LT |
5583 | |
5584 | /* | |
48c5ccae PZ |
5585 | * Fudge the rq selection such that the below task selection loop |
5586 | * doesn't get stuck on the currently eligible stop task. | |
5587 | * | |
5588 | * We're currently inside stop_machine() and the rq is either stuck | |
5589 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5590 | * either way we should never end up calling schedule() until we're | |
5591 | * done here. | |
1da177e4 | 5592 | */ |
48c5ccae | 5593 | rq->stop = NULL; |
48f24c4d | 5594 | |
77bd3970 FW |
5595 | /* |
5596 | * put_prev_task() and pick_next_task() sched | |
5597 | * class method both need to have an up-to-date | |
5598 | * value of rq->clock[_task] | |
5599 | */ | |
5600 | update_rq_clock(rq); | |
5601 | ||
5e16bbc2 | 5602 | for (;;) { |
48c5ccae PZ |
5603 | /* |
5604 | * There's this thread running, bail when that's the only | |
d1ccc66d | 5605 | * remaining thread: |
48c5ccae PZ |
5606 | */ |
5607 | if (rq->nr_running == 1) | |
dd41f596 | 5608 | break; |
48c5ccae | 5609 | |
cbce1a68 | 5610 | /* |
d1ccc66d | 5611 | * pick_next_task() assumes pinned rq->lock: |
cbce1a68 | 5612 | */ |
8a8c69c3 | 5613 | next = pick_next_task(rq, &fake_task, rf); |
48c5ccae | 5614 | BUG_ON(!next); |
5b713a3d | 5615 | put_prev_task(rq, next); |
e692ab53 | 5616 | |
5473e0cc WL |
5617 | /* |
5618 | * Rules for changing task_struct::cpus_allowed are holding | |
5619 | * both pi_lock and rq->lock, such that holding either | |
5620 | * stabilizes the mask. | |
5621 | * | |
5622 | * Drop rq->lock is not quite as disastrous as it usually is | |
5623 | * because !cpu_active at this point, which means load-balance | |
5624 | * will not interfere. Also, stop-machine. | |
5625 | */ | |
8a8c69c3 | 5626 | rq_unlock(rq, rf); |
5473e0cc | 5627 | raw_spin_lock(&next->pi_lock); |
8a8c69c3 | 5628 | rq_relock(rq, rf); |
5473e0cc WL |
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); |
8a8c69c3 | 5642 | rq = __migrate_task(rq, rf, next, dest_cpu); |
5e16bbc2 | 5643 | if (rq != dead_rq) { |
8a8c69c3 | 5644 | rq_unlock(rq, rf); |
5e16bbc2 | 5645 | rq = dead_rq; |
8a8c69c3 PZ |
5646 | *rf = orf; |
5647 | rq_relock(rq, rf); | |
5e16bbc2 | 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 | ||
d1ccc66d IM |
5686 | /* |
5687 | * used to mark begin/end of suspend/resume: | |
5688 | */ | |
5689 | static int num_cpus_frozen; | |
d35be8ba | 5690 | |
1da177e4 | 5691 | /* |
3a101d05 TH |
5692 | * Update cpusets according to cpu_active mask. If cpusets are |
5693 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
5694 | * around partition_sched_domains(). | |
d35be8ba SB |
5695 | * |
5696 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
5697 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 5698 | */ |
40190a78 | 5699 | static void cpuset_cpu_active(void) |
e761b772 | 5700 | { |
40190a78 | 5701 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
5702 | /* |
5703 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
5704 | * resume sequence. As long as this is not the last online | |
5705 | * operation in the resume sequence, just build a single sched | |
5706 | * domain, ignoring cpusets. | |
5707 | */ | |
50e76632 PZ |
5708 | partition_sched_domains(1, NULL, NULL); |
5709 | if (--num_cpus_frozen) | |
135fb3e1 | 5710 | return; |
d35be8ba SB |
5711 | /* |
5712 | * This is the last CPU online operation. So fall through and | |
5713 | * restore the original sched domains by considering the | |
5714 | * cpuset configurations. | |
5715 | */ | |
50e76632 | 5716 | cpuset_force_rebuild(); |
3a101d05 | 5717 | } |
30e03acd | 5718 | cpuset_update_active_cpus(); |
3a101d05 | 5719 | } |
e761b772 | 5720 | |
40190a78 | 5721 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 5722 | { |
40190a78 | 5723 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 5724 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 5725 | return -EBUSY; |
30e03acd | 5726 | cpuset_update_active_cpus(); |
135fb3e1 | 5727 | } else { |
d35be8ba SB |
5728 | num_cpus_frozen++; |
5729 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 5730 | } |
135fb3e1 | 5731 | return 0; |
e761b772 | 5732 | } |
e761b772 | 5733 | |
40190a78 | 5734 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 5735 | { |
7d976699 | 5736 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 5737 | struct rq_flags rf; |
7d976699 | 5738 | |
ba2591a5 PZ |
5739 | #ifdef CONFIG_SCHED_SMT |
5740 | /* | |
5741 | * The sched_smt_present static key needs to be evaluated on every | |
5742 | * hotplug event because at boot time SMT might be disabled when | |
5743 | * the number of booted CPUs is limited. | |
5744 | * | |
5745 | * If then later a sibling gets hotplugged, then the key would stay | |
5746 | * off and SMT scheduling would never be functional. | |
5747 | */ | |
5748 | if (cpumask_weight(cpu_smt_mask(cpu)) > 1) | |
5749 | static_branch_enable_cpuslocked(&sched_smt_present); | |
5750 | #endif | |
40190a78 | 5751 | set_cpu_active(cpu, true); |
135fb3e1 | 5752 | |
40190a78 | 5753 | if (sched_smp_initialized) { |
135fb3e1 | 5754 | sched_domains_numa_masks_set(cpu); |
40190a78 | 5755 | cpuset_cpu_active(); |
e761b772 | 5756 | } |
7d976699 TG |
5757 | |
5758 | /* | |
5759 | * Put the rq online, if not already. This happens: | |
5760 | * | |
5761 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 5762 | * after all CPUs have been brought up. |
7d976699 TG |
5763 | * |
5764 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
5765 | * domains. | |
5766 | */ | |
8a8c69c3 | 5767 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
5768 | if (rq->rd) { |
5769 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5770 | set_rq_online(rq); | |
5771 | } | |
8a8c69c3 | 5772 | rq_unlock_irqrestore(rq, &rf); |
7d976699 TG |
5773 | |
5774 | update_max_interval(); | |
5775 | ||
40190a78 | 5776 | return 0; |
135fb3e1 TG |
5777 | } |
5778 | ||
40190a78 | 5779 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 5780 | { |
135fb3e1 TG |
5781 | int ret; |
5782 | ||
40190a78 | 5783 | set_cpu_active(cpu, false); |
b2454caa PZ |
5784 | /* |
5785 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
5786 | * users of this state to go away such that all new such users will | |
5787 | * observe it. | |
5788 | * | |
b2454caa PZ |
5789 | * Do sync before park smpboot threads to take care the rcu boost case. |
5790 | */ | |
d7d34d5e | 5791 | synchronize_rcu_mult(call_rcu, call_rcu_sched); |
40190a78 TG |
5792 | |
5793 | if (!sched_smp_initialized) | |
5794 | return 0; | |
5795 | ||
5796 | ret = cpuset_cpu_inactive(cpu); | |
5797 | if (ret) { | |
5798 | set_cpu_active(cpu, true); | |
5799 | return ret; | |
135fb3e1 | 5800 | } |
40190a78 TG |
5801 | sched_domains_numa_masks_clear(cpu); |
5802 | return 0; | |
135fb3e1 TG |
5803 | } |
5804 | ||
94baf7a5 TG |
5805 | static void sched_rq_cpu_starting(unsigned int cpu) |
5806 | { | |
5807 | struct rq *rq = cpu_rq(cpu); | |
5808 | ||
5809 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
5810 | update_max_interval(); |
5811 | } | |
5812 | ||
135fb3e1 TG |
5813 | int sched_cpu_starting(unsigned int cpu) |
5814 | { | |
94baf7a5 | 5815 | sched_rq_cpu_starting(cpu); |
d84b3131 | 5816 | sched_tick_start(cpu); |
135fb3e1 | 5817 | return 0; |
e761b772 | 5818 | } |
e761b772 | 5819 | |
f2785ddb TG |
5820 | #ifdef CONFIG_HOTPLUG_CPU |
5821 | int sched_cpu_dying(unsigned int cpu) | |
5822 | { | |
5823 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 5824 | struct rq_flags rf; |
f2785ddb TG |
5825 | |
5826 | /* Handle pending wakeups and then migrate everything off */ | |
5827 | sched_ttwu_pending(); | |
d84b3131 | 5828 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
5829 | |
5830 | rq_lock_irqsave(rq, &rf); | |
f2785ddb TG |
5831 | if (rq->rd) { |
5832 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5833 | set_rq_offline(rq); | |
5834 | } | |
8a8c69c3 | 5835 | migrate_tasks(rq, &rf); |
f2785ddb | 5836 | BUG_ON(rq->nr_running != 1); |
8a8c69c3 PZ |
5837 | rq_unlock_irqrestore(rq, &rf); |
5838 | ||
f2785ddb TG |
5839 | calc_load_migrate(rq); |
5840 | update_max_interval(); | |
00357f5e | 5841 | nohz_balance_exit_idle(rq); |
e5ef27d0 | 5842 | hrtick_clear(rq); |
f2785ddb TG |
5843 | return 0; |
5844 | } | |
5845 | #endif | |
5846 | ||
1da177e4 LT |
5847 | void __init sched_init_smp(void) |
5848 | { | |
cb83b629 PZ |
5849 | sched_init_numa(); |
5850 | ||
6acce3ef PZ |
5851 | /* |
5852 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 5853 | * CPU masks are stable and all blatant races in the below code cannot |
40fa3780 VS |
5854 | * happen. The hotplug lock is nevertheless taken to satisfy lockdep, |
5855 | * but there won't be any contention on it. | |
6acce3ef | 5856 | */ |
40fa3780 | 5857 | cpus_read_lock(); |
712555ee | 5858 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 5859 | sched_init_domains(cpu_active_mask); |
712555ee | 5860 | mutex_unlock(&sched_domains_mutex); |
40fa3780 | 5861 | cpus_read_unlock(); |
e761b772 | 5862 | |
5c1e1767 | 5863 | /* Move init over to a non-isolated CPU */ |
edb93821 | 5864 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 5865 | BUG(); |
19978ca6 | 5866 | sched_init_granularity(); |
4212823f | 5867 | |
0e3900e6 | 5868 | init_sched_rt_class(); |
1baca4ce | 5869 | init_sched_dl_class(); |
1b568f0a | 5870 | |
e26fbffd | 5871 | sched_smp_initialized = true; |
1da177e4 | 5872 | } |
e26fbffd TG |
5873 | |
5874 | static int __init migration_init(void) | |
5875 | { | |
94baf7a5 | 5876 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 5877 | return 0; |
1da177e4 | 5878 | } |
e26fbffd TG |
5879 | early_initcall(migration_init); |
5880 | ||
1da177e4 LT |
5881 | #else |
5882 | void __init sched_init_smp(void) | |
5883 | { | |
19978ca6 | 5884 | sched_init_granularity(); |
1da177e4 LT |
5885 | } |
5886 | #endif /* CONFIG_SMP */ | |
5887 | ||
5888 | int in_sched_functions(unsigned long addr) | |
5889 | { | |
1da177e4 LT |
5890 | return in_lock_functions(addr) || |
5891 | (addr >= (unsigned long)__sched_text_start | |
5892 | && addr < (unsigned long)__sched_text_end); | |
5893 | } | |
5894 | ||
029632fb | 5895 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
5896 | /* |
5897 | * Default task group. | |
5898 | * Every task in system belongs to this group at bootup. | |
5899 | */ | |
029632fb | 5900 | struct task_group root_task_group; |
35cf4e50 | 5901 | LIST_HEAD(task_groups); |
b0367629 WL |
5902 | |
5903 | /* Cacheline aligned slab cache for task_group */ | |
5904 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 5905 | #endif |
6f505b16 | 5906 | |
e6252c3e | 5907 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 5908 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 5909 | |
1da177e4 LT |
5910 | void __init sched_init(void) |
5911 | { | |
dd41f596 | 5912 | int i, j; |
434d53b0 MT |
5913 | unsigned long alloc_size = 0, ptr; |
5914 | ||
5822a454 | 5915 | wait_bit_init(); |
9dcb8b68 | 5916 | |
434d53b0 MT |
5917 | #ifdef CONFIG_FAIR_GROUP_SCHED |
5918 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5919 | #endif | |
5920 | #ifdef CONFIG_RT_GROUP_SCHED | |
5921 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5922 | #endif | |
434d53b0 | 5923 | if (alloc_size) { |
36b7b6d4 | 5924 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
5925 | |
5926 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 5927 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
5928 | ptr += nr_cpu_ids * sizeof(void **); |
5929 | ||
07e06b01 | 5930 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 5931 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 5932 | |
6d6bc0ad | 5933 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 5934 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 5935 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
5936 | ptr += nr_cpu_ids * sizeof(void **); |
5937 | ||
07e06b01 | 5938 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
5939 | ptr += nr_cpu_ids * sizeof(void **); |
5940 | ||
6d6bc0ad | 5941 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 5942 | } |
df7c8e84 | 5943 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
5944 | for_each_possible_cpu(i) { |
5945 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
5946 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
5947 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
5948 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 5949 | } |
b74e6278 | 5950 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 5951 | |
d1ccc66d IM |
5952 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
5953 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 5954 | |
57d885fe GH |
5955 | #ifdef CONFIG_SMP |
5956 | init_defrootdomain(); | |
5957 | #endif | |
5958 | ||
d0b27fa7 | 5959 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 5960 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 5961 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 5962 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 5963 | |
7c941438 | 5964 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
5965 | task_group_cache = KMEM_CACHE(task_group, 0); |
5966 | ||
07e06b01 YZ |
5967 | list_add(&root_task_group.list, &task_groups); |
5968 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 5969 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 5970 | autogroup_init(&init_task); |
7c941438 | 5971 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 5972 | |
0a945022 | 5973 | for_each_possible_cpu(i) { |
70b97a7f | 5974 | struct rq *rq; |
1da177e4 LT |
5975 | |
5976 | rq = cpu_rq(i); | |
05fa785c | 5977 | raw_spin_lock_init(&rq->lock); |
7897986b | 5978 | rq->nr_running = 0; |
dce48a84 TG |
5979 | rq->calc_load_active = 0; |
5980 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 5981 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
5982 | init_rt_rq(&rq->rt); |
5983 | init_dl_rq(&rq->dl); | |
dd41f596 | 5984 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 5985 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 5986 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 5987 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 5988 | /* |
d1ccc66d | 5989 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
5990 | * |
5991 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
5992 | * gets 100% of the CPU resources in the system. This overall |
5993 | * system CPU resource is divided among the tasks of | |
07e06b01 | 5994 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
5995 | * based on each entity's (task or task-group's) weight |
5996 | * (se->load.weight). | |
5997 | * | |
07e06b01 | 5998 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 5999 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 6000 | * then A0's share of the CPU resource is: |
354d60c2 | 6001 | * |
0d905bca | 6002 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6003 | * |
07e06b01 YZ |
6004 | * We achieve this by letting root_task_group's tasks sit |
6005 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6006 | */ |
ab84d31e | 6007 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6008 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6009 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6010 | ||
6011 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6012 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6013 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6014 | #endif |
1da177e4 | 6015 | |
dd41f596 IM |
6016 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6017 | rq->cpu_load[j] = 0; | |
fdf3e95d | 6018 | |
1da177e4 | 6019 | #ifdef CONFIG_SMP |
41c7ce9a | 6020 | rq->sd = NULL; |
57d885fe | 6021 | rq->rd = NULL; |
ca6d75e6 | 6022 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 6023 | rq->balance_callback = NULL; |
1da177e4 | 6024 | rq->active_balance = 0; |
dd41f596 | 6025 | rq->next_balance = jiffies; |
1da177e4 | 6026 | rq->push_cpu = 0; |
0a2966b4 | 6027 | rq->cpu = i; |
1f11eb6a | 6028 | rq->online = 0; |
eae0c9df MG |
6029 | rq->idle_stamp = 0; |
6030 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6031 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6032 | |
6033 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6034 | ||
dc938520 | 6035 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6036 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 6037 | rq->last_load_update_tick = jiffies; |
e022e0d3 | 6038 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 6039 | atomic_set(&rq->nohz_flags, 0); |
83cd4fe2 | 6040 | #endif |
9fd81dd5 | 6041 | #endif /* CONFIG_SMP */ |
77a021be | 6042 | hrtick_rq_init(rq); |
1da177e4 | 6043 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6044 | } |
6045 | ||
9059393e | 6046 | set_load_weight(&init_task, false); |
b50f60ce | 6047 | |
1da177e4 LT |
6048 | /* |
6049 | * The boot idle thread does lazy MMU switching as well: | |
6050 | */ | |
f1f10076 | 6051 | mmgrab(&init_mm); |
1da177e4 LT |
6052 | enter_lazy_tlb(&init_mm, current); |
6053 | ||
6054 | /* | |
6055 | * Make us the idle thread. Technically, schedule() should not be | |
6056 | * called from this thread, however somewhere below it might be, | |
6057 | * but because we are the idle thread, we just pick up running again | |
6058 | * when this runqueue becomes "idle". | |
6059 | */ | |
6060 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6061 | |
6062 | calc_load_update = jiffies + LOAD_FREQ; | |
6063 | ||
bf4d83f6 | 6064 | #ifdef CONFIG_SMP |
29d5e047 | 6065 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6066 | #endif |
6067 | init_sched_fair_class(); | |
6a7b3dc3 | 6068 | |
4698f88c JP |
6069 | init_schedstats(); |
6070 | ||
eb414681 JW |
6071 | psi_init(); |
6072 | ||
6892b75e | 6073 | scheduler_running = 1; |
1da177e4 LT |
6074 | } |
6075 | ||
d902db1e | 6076 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6077 | static inline int preempt_count_equals(int preempt_offset) |
6078 | { | |
da7142e2 | 6079 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 6080 | |
4ba8216c | 6081 | return (nested == preempt_offset); |
e4aafea2 FW |
6082 | } |
6083 | ||
d894837f | 6084 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6085 | { |
8eb23b9f PZ |
6086 | /* |
6087 | * Blocking primitives will set (and therefore destroy) current->state, | |
6088 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
6089 | * otherwise we will destroy state. | |
6090 | */ | |
00845eb9 | 6091 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
6092 | "do not call blocking ops when !TASK_RUNNING; " |
6093 | "state=%lx set at [<%p>] %pS\n", | |
6094 | current->state, | |
6095 | (void *)current->task_state_change, | |
00845eb9 | 6096 | (void *)current->task_state_change); |
8eb23b9f | 6097 | |
3427445a PZ |
6098 | ___might_sleep(file, line, preempt_offset); |
6099 | } | |
6100 | EXPORT_SYMBOL(__might_sleep); | |
6101 | ||
6102 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 6103 | { |
d1ccc66d IM |
6104 | /* Ratelimiting timestamp: */ |
6105 | static unsigned long prev_jiffy; | |
6106 | ||
d1c6d149 | 6107 | unsigned long preempt_disable_ip; |
1da177e4 | 6108 | |
d1ccc66d IM |
6109 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
6110 | rcu_sleep_check(); | |
6111 | ||
db273be2 TG |
6112 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
6113 | !is_idle_task(current)) || | |
1c3c5eab TG |
6114 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
6115 | oops_in_progress) | |
aef745fc | 6116 | return; |
1c3c5eab | 6117 | |
aef745fc IM |
6118 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
6119 | return; | |
6120 | prev_jiffy = jiffies; | |
6121 | ||
d1ccc66d | 6122 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
6123 | preempt_disable_ip = get_preempt_disable_ip(current); |
6124 | ||
3df0fc5b PZ |
6125 | printk(KERN_ERR |
6126 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6127 | file, line); | |
6128 | printk(KERN_ERR | |
6129 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6130 | in_atomic(), irqs_disabled(), | |
6131 | current->pid, current->comm); | |
aef745fc | 6132 | |
a8b686b3 ES |
6133 | if (task_stack_end_corrupted(current)) |
6134 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
6135 | ||
aef745fc IM |
6136 | debug_show_held_locks(current); |
6137 | if (irqs_disabled()) | |
6138 | print_irqtrace_events(current); | |
d1c6d149 VN |
6139 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
6140 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 6141 | pr_err("Preemption disabled at:"); |
d1c6d149 | 6142 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
6143 | pr_cont("\n"); |
6144 | } | |
aef745fc | 6145 | dump_stack(); |
f0b22e39 | 6146 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 6147 | } |
3427445a | 6148 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
6149 | #endif |
6150 | ||
6151 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 6152 | void normalize_rt_tasks(void) |
3a5e4dc1 | 6153 | { |
dbc7f069 | 6154 | struct task_struct *g, *p; |
d50dde5a DF |
6155 | struct sched_attr attr = { |
6156 | .sched_policy = SCHED_NORMAL, | |
6157 | }; | |
1da177e4 | 6158 | |
3472eaa1 | 6159 | read_lock(&tasklist_lock); |
5d07f420 | 6160 | for_each_process_thread(g, p) { |
178be793 IM |
6161 | /* |
6162 | * Only normalize user tasks: | |
6163 | */ | |
3472eaa1 | 6164 | if (p->flags & PF_KTHREAD) |
178be793 IM |
6165 | continue; |
6166 | ||
4fa8d299 JP |
6167 | p->se.exec_start = 0; |
6168 | schedstat_set(p->se.statistics.wait_start, 0); | |
6169 | schedstat_set(p->se.statistics.sleep_start, 0); | |
6170 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 6171 | |
aab03e05 | 6172 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
6173 | /* |
6174 | * Renice negative nice level userspace | |
6175 | * tasks back to 0: | |
6176 | */ | |
3472eaa1 | 6177 | if (task_nice(p) < 0) |
dd41f596 | 6178 | set_user_nice(p, 0); |
1da177e4 | 6179 | continue; |
dd41f596 | 6180 | } |
1da177e4 | 6181 | |
dbc7f069 | 6182 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 6183 | } |
3472eaa1 | 6184 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6185 | } |
6186 | ||
6187 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6188 | |
67fc4e0c | 6189 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6190 | /* |
67fc4e0c | 6191 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6192 | * |
6193 | * They can only be called when the whole system has been | |
6194 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6195 | * activity can take place. Using them for anything else would | |
6196 | * be a serious bug, and as a result, they aren't even visible | |
6197 | * under any other configuration. | |
6198 | */ | |
6199 | ||
6200 | /** | |
d1ccc66d | 6201 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
6202 | * @cpu: the processor in question. |
6203 | * | |
6204 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
6205 | * |
6206 | * Return: The current task for @cpu. | |
1df5c10a | 6207 | */ |
36c8b586 | 6208 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6209 | { |
6210 | return cpu_curr(cpu); | |
6211 | } | |
6212 | ||
67fc4e0c JW |
6213 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6214 | ||
6215 | #ifdef CONFIG_IA64 | |
1df5c10a | 6216 | /** |
d1ccc66d | 6217 | * set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
6218 | * @cpu: the processor in question. |
6219 | * @p: the task pointer to set. | |
6220 | * | |
6221 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 6222 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 6223 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
6224 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6225 | * and caller must save the original value of the current task (see | |
6226 | * curr_task() above) and restore that value before reenabling interrupts and | |
6227 | * re-starting the system. | |
6228 | * | |
6229 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6230 | */ | |
a458ae2e | 6231 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6232 | { |
6233 | cpu_curr(cpu) = p; | |
6234 | } | |
6235 | ||
6236 | #endif | |
29f59db3 | 6237 | |
7c941438 | 6238 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6239 | /* task_group_lock serializes the addition/removal of task groups */ |
6240 | static DEFINE_SPINLOCK(task_group_lock); | |
6241 | ||
2f5177f0 | 6242 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
6243 | { |
6244 | free_fair_sched_group(tg); | |
6245 | free_rt_sched_group(tg); | |
e9aa1dd1 | 6246 | autogroup_free(tg); |
b0367629 | 6247 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
6248 | } |
6249 | ||
6250 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 6251 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
6252 | { |
6253 | struct task_group *tg; | |
bccbe08a | 6254 | |
b0367629 | 6255 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
6256 | if (!tg) |
6257 | return ERR_PTR(-ENOMEM); | |
6258 | ||
ec7dc8ac | 6259 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
6260 | goto err; |
6261 | ||
ec7dc8ac | 6262 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
6263 | goto err; |
6264 | ||
ace783b9 LZ |
6265 | return tg; |
6266 | ||
6267 | err: | |
2f5177f0 | 6268 | sched_free_group(tg); |
ace783b9 LZ |
6269 | return ERR_PTR(-ENOMEM); |
6270 | } | |
6271 | ||
6272 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
6273 | { | |
6274 | unsigned long flags; | |
6275 | ||
8ed36996 | 6276 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 6277 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 6278 | |
d1ccc66d IM |
6279 | /* Root should already exist: */ |
6280 | WARN_ON(!parent); | |
f473aa5e PZ |
6281 | |
6282 | tg->parent = parent; | |
f473aa5e | 6283 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 6284 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 6285 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
6286 | |
6287 | online_fair_sched_group(tg); | |
29f59db3 SV |
6288 | } |
6289 | ||
9b5b7751 | 6290 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 6291 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 6292 | { |
d1ccc66d | 6293 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 6294 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
6295 | } |
6296 | ||
4cf86d77 | 6297 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 6298 | { |
d1ccc66d | 6299 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 6300 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
6301 | } |
6302 | ||
6303 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 6304 | { |
8ed36996 | 6305 | unsigned long flags; |
29f59db3 | 6306 | |
d1ccc66d | 6307 | /* End participation in shares distribution: */ |
6fe1f348 | 6308 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
6309 | |
6310 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 6311 | list_del_rcu(&tg->list); |
f473aa5e | 6312 | list_del_rcu(&tg->siblings); |
8ed36996 | 6313 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6314 | } |
6315 | ||
ea86cb4b | 6316 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 6317 | { |
8323f26c | 6318 | struct task_group *tg; |
29f59db3 | 6319 | |
f7b8a47d KT |
6320 | /* |
6321 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
6322 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
6323 | * to prevent lockdep warnings. | |
6324 | */ | |
6325 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
6326 | struct task_group, css); |
6327 | tg = autogroup_task_group(tsk, tg); | |
6328 | tsk->sched_task_group = tg; | |
6329 | ||
810b3817 | 6330 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
6331 | if (tsk->sched_class->task_change_group) |
6332 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 6333 | else |
810b3817 | 6334 | #endif |
b2b5ce02 | 6335 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
6336 | } |
6337 | ||
6338 | /* | |
6339 | * Change task's runqueue when it moves between groups. | |
6340 | * | |
6341 | * The caller of this function should have put the task in its new group by | |
6342 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
6343 | * its new group. | |
6344 | */ | |
6345 | void sched_move_task(struct task_struct *tsk) | |
6346 | { | |
7a57f32a PZ |
6347 | int queued, running, queue_flags = |
6348 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
6349 | struct rq_flags rf; |
6350 | struct rq *rq; | |
6351 | ||
6352 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 6353 | update_rq_clock(rq); |
ea86cb4b VG |
6354 | |
6355 | running = task_current(rq, tsk); | |
6356 | queued = task_on_rq_queued(tsk); | |
6357 | ||
6358 | if (queued) | |
7a57f32a | 6359 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 6360 | if (running) |
ea86cb4b VG |
6361 | put_prev_task(rq, tsk); |
6362 | ||
6363 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 6364 | |
da0c1e65 | 6365 | if (queued) |
7a57f32a | 6366 | enqueue_task(rq, tsk, queue_flags); |
bb3bac2c | 6367 | if (running) |
b2bf6c31 | 6368 | set_curr_task(rq, tsk); |
29f59db3 | 6369 | |
eb580751 | 6370 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 6371 | } |
68318b8e | 6372 | |
a7c6d554 | 6373 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 6374 | { |
a7c6d554 | 6375 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
6376 | } |
6377 | ||
eb95419b TH |
6378 | static struct cgroup_subsys_state * |
6379 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 6380 | { |
eb95419b TH |
6381 | struct task_group *parent = css_tg(parent_css); |
6382 | struct task_group *tg; | |
68318b8e | 6383 | |
eb95419b | 6384 | if (!parent) { |
68318b8e | 6385 | /* This is early initialization for the top cgroup */ |
07e06b01 | 6386 | return &root_task_group.css; |
68318b8e SV |
6387 | } |
6388 | ||
ec7dc8ac | 6389 | tg = sched_create_group(parent); |
68318b8e SV |
6390 | if (IS_ERR(tg)) |
6391 | return ERR_PTR(-ENOMEM); | |
6392 | ||
68318b8e SV |
6393 | return &tg->css; |
6394 | } | |
6395 | ||
96b77745 KK |
6396 | /* Expose task group only after completing cgroup initialization */ |
6397 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
6398 | { | |
6399 | struct task_group *tg = css_tg(css); | |
6400 | struct task_group *parent = css_tg(css->parent); | |
6401 | ||
6402 | if (parent) | |
6403 | sched_online_group(tg, parent); | |
6404 | return 0; | |
6405 | } | |
6406 | ||
2f5177f0 | 6407 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 6408 | { |
eb95419b | 6409 | struct task_group *tg = css_tg(css); |
ace783b9 | 6410 | |
2f5177f0 | 6411 | sched_offline_group(tg); |
ace783b9 LZ |
6412 | } |
6413 | ||
eb95419b | 6414 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 6415 | { |
eb95419b | 6416 | struct task_group *tg = css_tg(css); |
68318b8e | 6417 | |
2f5177f0 PZ |
6418 | /* |
6419 | * Relies on the RCU grace period between css_released() and this. | |
6420 | */ | |
6421 | sched_free_group(tg); | |
ace783b9 LZ |
6422 | } |
6423 | ||
ea86cb4b VG |
6424 | /* |
6425 | * This is called before wake_up_new_task(), therefore we really only | |
6426 | * have to set its group bits, all the other stuff does not apply. | |
6427 | */ | |
b53202e6 | 6428 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 6429 | { |
ea86cb4b VG |
6430 | struct rq_flags rf; |
6431 | struct rq *rq; | |
6432 | ||
6433 | rq = task_rq_lock(task, &rf); | |
6434 | ||
80f5c1b8 | 6435 | update_rq_clock(rq); |
ea86cb4b VG |
6436 | sched_change_group(task, TASK_SET_GROUP); |
6437 | ||
6438 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
6439 | } |
6440 | ||
1f7dd3e5 | 6441 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 6442 | { |
bb9d97b6 | 6443 | struct task_struct *task; |
1f7dd3e5 | 6444 | struct cgroup_subsys_state *css; |
7dc603c9 | 6445 | int ret = 0; |
bb9d97b6 | 6446 | |
1f7dd3e5 | 6447 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 6448 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 6449 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 6450 | return -EINVAL; |
b68aa230 | 6451 | #else |
bb9d97b6 TH |
6452 | /* We don't support RT-tasks being in separate groups */ |
6453 | if (task->sched_class != &fair_sched_class) | |
6454 | return -EINVAL; | |
b68aa230 | 6455 | #endif |
7dc603c9 PZ |
6456 | /* |
6457 | * Serialize against wake_up_new_task() such that if its | |
6458 | * running, we're sure to observe its full state. | |
6459 | */ | |
6460 | raw_spin_lock_irq(&task->pi_lock); | |
6461 | /* | |
6462 | * Avoid calling sched_move_task() before wake_up_new_task() | |
6463 | * has happened. This would lead to problems with PELT, due to | |
6464 | * move wanting to detach+attach while we're not attached yet. | |
6465 | */ | |
6466 | if (task->state == TASK_NEW) | |
6467 | ret = -EINVAL; | |
6468 | raw_spin_unlock_irq(&task->pi_lock); | |
6469 | ||
6470 | if (ret) | |
6471 | break; | |
bb9d97b6 | 6472 | } |
7dc603c9 | 6473 | return ret; |
be367d09 | 6474 | } |
68318b8e | 6475 | |
1f7dd3e5 | 6476 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 6477 | { |
bb9d97b6 | 6478 | struct task_struct *task; |
1f7dd3e5 | 6479 | struct cgroup_subsys_state *css; |
bb9d97b6 | 6480 | |
1f7dd3e5 | 6481 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 6482 | sched_move_task(task); |
68318b8e SV |
6483 | } |
6484 | ||
052f1dc7 | 6485 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
6486 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
6487 | struct cftype *cftype, u64 shareval) | |
68318b8e | 6488 | { |
182446d0 | 6489 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
6490 | } |
6491 | ||
182446d0 TH |
6492 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
6493 | struct cftype *cft) | |
68318b8e | 6494 | { |
182446d0 | 6495 | struct task_group *tg = css_tg(css); |
68318b8e | 6496 | |
c8b28116 | 6497 | return (u64) scale_load_down(tg->shares); |
68318b8e | 6498 | } |
ab84d31e PT |
6499 | |
6500 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
6501 | static DEFINE_MUTEX(cfs_constraints_mutex); |
6502 | ||
ab84d31e PT |
6503 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
6504 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
6505 | ||
a790de99 PT |
6506 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
6507 | ||
ab84d31e PT |
6508 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
6509 | { | |
56f570e5 | 6510 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 6511 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
6512 | |
6513 | if (tg == &root_task_group) | |
6514 | return -EINVAL; | |
6515 | ||
6516 | /* | |
6517 | * Ensure we have at some amount of bandwidth every period. This is | |
6518 | * to prevent reaching a state of large arrears when throttled via | |
6519 | * entity_tick() resulting in prolonged exit starvation. | |
6520 | */ | |
6521 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
6522 | return -EINVAL; | |
6523 | ||
6524 | /* | |
6525 | * Likewise, bound things on the otherside by preventing insane quota | |
6526 | * periods. This also allows us to normalize in computing quota | |
6527 | * feasibility. | |
6528 | */ | |
6529 | if (period > max_cfs_quota_period) | |
6530 | return -EINVAL; | |
6531 | ||
0e59bdae KT |
6532 | /* |
6533 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
6534 | * unthrottle_offline_cfs_rqs(). | |
6535 | */ | |
6536 | get_online_cpus(); | |
a790de99 PT |
6537 | mutex_lock(&cfs_constraints_mutex); |
6538 | ret = __cfs_schedulable(tg, period, quota); | |
6539 | if (ret) | |
6540 | goto out_unlock; | |
6541 | ||
58088ad0 | 6542 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 6543 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
6544 | /* |
6545 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
6546 | * before making related changes, and on->off must occur afterwards | |
6547 | */ | |
6548 | if (runtime_enabled && !runtime_was_enabled) | |
6549 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
6550 | raw_spin_lock_irq(&cfs_b->lock); |
6551 | cfs_b->period = ns_to_ktime(period); | |
6552 | cfs_b->quota = quota; | |
58088ad0 | 6553 | |
a9cf55b2 | 6554 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
6555 | |
6556 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
6557 | if (runtime_enabled) |
6558 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 6559 | |
ab84d31e PT |
6560 | raw_spin_unlock_irq(&cfs_b->lock); |
6561 | ||
0e59bdae | 6562 | for_each_online_cpu(i) { |
ab84d31e | 6563 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 6564 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 6565 | struct rq_flags rf; |
ab84d31e | 6566 | |
8a8c69c3 | 6567 | rq_lock_irq(rq, &rf); |
58088ad0 | 6568 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 6569 | cfs_rq->runtime_remaining = 0; |
671fd9da | 6570 | |
029632fb | 6571 | if (cfs_rq->throttled) |
671fd9da | 6572 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 6573 | rq_unlock_irq(rq, &rf); |
ab84d31e | 6574 | } |
1ee14e6c BS |
6575 | if (runtime_was_enabled && !runtime_enabled) |
6576 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
6577 | out_unlock: |
6578 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 6579 | put_online_cpus(); |
ab84d31e | 6580 | |
a790de99 | 6581 | return ret; |
ab84d31e PT |
6582 | } |
6583 | ||
6584 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
6585 | { | |
6586 | u64 quota, period; | |
6587 | ||
029632fb | 6588 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
6589 | if (cfs_quota_us < 0) |
6590 | quota = RUNTIME_INF; | |
6591 | else | |
6592 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
6593 | ||
6594 | return tg_set_cfs_bandwidth(tg, period, quota); | |
6595 | } | |
6596 | ||
6597 | long tg_get_cfs_quota(struct task_group *tg) | |
6598 | { | |
6599 | u64 quota_us; | |
6600 | ||
029632fb | 6601 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
6602 | return -1; |
6603 | ||
029632fb | 6604 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
6605 | do_div(quota_us, NSEC_PER_USEC); |
6606 | ||
6607 | return quota_us; | |
6608 | } | |
6609 | ||
6610 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
6611 | { | |
6612 | u64 quota, period; | |
6613 | ||
6614 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 6615 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 6616 | |
ab84d31e PT |
6617 | return tg_set_cfs_bandwidth(tg, period, quota); |
6618 | } | |
6619 | ||
6620 | long tg_get_cfs_period(struct task_group *tg) | |
6621 | { | |
6622 | u64 cfs_period_us; | |
6623 | ||
029632fb | 6624 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
6625 | do_div(cfs_period_us, NSEC_PER_USEC); |
6626 | ||
6627 | return cfs_period_us; | |
6628 | } | |
6629 | ||
182446d0 TH |
6630 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
6631 | struct cftype *cft) | |
ab84d31e | 6632 | { |
182446d0 | 6633 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
6634 | } |
6635 | ||
182446d0 TH |
6636 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
6637 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 6638 | { |
182446d0 | 6639 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
6640 | } |
6641 | ||
182446d0 TH |
6642 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
6643 | struct cftype *cft) | |
ab84d31e | 6644 | { |
182446d0 | 6645 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
6646 | } |
6647 | ||
182446d0 TH |
6648 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
6649 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 6650 | { |
182446d0 | 6651 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
6652 | } |
6653 | ||
a790de99 PT |
6654 | struct cfs_schedulable_data { |
6655 | struct task_group *tg; | |
6656 | u64 period, quota; | |
6657 | }; | |
6658 | ||
6659 | /* | |
6660 | * normalize group quota/period to be quota/max_period | |
6661 | * note: units are usecs | |
6662 | */ | |
6663 | static u64 normalize_cfs_quota(struct task_group *tg, | |
6664 | struct cfs_schedulable_data *d) | |
6665 | { | |
6666 | u64 quota, period; | |
6667 | ||
6668 | if (tg == d->tg) { | |
6669 | period = d->period; | |
6670 | quota = d->quota; | |
6671 | } else { | |
6672 | period = tg_get_cfs_period(tg); | |
6673 | quota = tg_get_cfs_quota(tg); | |
6674 | } | |
6675 | ||
6676 | /* note: these should typically be equivalent */ | |
6677 | if (quota == RUNTIME_INF || quota == -1) | |
6678 | return RUNTIME_INF; | |
6679 | ||
6680 | return to_ratio(period, quota); | |
6681 | } | |
6682 | ||
6683 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
6684 | { | |
6685 | struct cfs_schedulable_data *d = data; | |
029632fb | 6686 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
6687 | s64 quota = 0, parent_quota = -1; |
6688 | ||
6689 | if (!tg->parent) { | |
6690 | quota = RUNTIME_INF; | |
6691 | } else { | |
029632fb | 6692 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
6693 | |
6694 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 6695 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
6696 | |
6697 | /* | |
c53593e5 TH |
6698 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
6699 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 6700 | * limit is set: |
a790de99 | 6701 | */ |
c53593e5 TH |
6702 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
6703 | quota = min(quota, parent_quota); | |
6704 | } else { | |
6705 | if (quota == RUNTIME_INF) | |
6706 | quota = parent_quota; | |
6707 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
6708 | return -EINVAL; | |
6709 | } | |
a790de99 | 6710 | } |
9c58c79a | 6711 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
6712 | |
6713 | return 0; | |
6714 | } | |
6715 | ||
6716 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
6717 | { | |
8277434e | 6718 | int ret; |
a790de99 PT |
6719 | struct cfs_schedulable_data data = { |
6720 | .tg = tg, | |
6721 | .period = period, | |
6722 | .quota = quota, | |
6723 | }; | |
6724 | ||
6725 | if (quota != RUNTIME_INF) { | |
6726 | do_div(data.period, NSEC_PER_USEC); | |
6727 | do_div(data.quota, NSEC_PER_USEC); | |
6728 | } | |
6729 | ||
8277434e PT |
6730 | rcu_read_lock(); |
6731 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
6732 | rcu_read_unlock(); | |
6733 | ||
6734 | return ret; | |
a790de99 | 6735 | } |
e8da1b18 | 6736 | |
a1f7164c | 6737 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 6738 | { |
2da8ca82 | 6739 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 6740 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 6741 | |
44ffc75b TH |
6742 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
6743 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
6744 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 6745 | |
3d6c50c2 YW |
6746 | if (schedstat_enabled() && tg != &root_task_group) { |
6747 | u64 ws = 0; | |
6748 | int i; | |
6749 | ||
6750 | for_each_possible_cpu(i) | |
6751 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
6752 | ||
6753 | seq_printf(sf, "wait_sum %llu\n", ws); | |
6754 | } | |
6755 | ||
e8da1b18 NR |
6756 | return 0; |
6757 | } | |
ab84d31e | 6758 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 6759 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 6760 | |
052f1dc7 | 6761 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
6762 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
6763 | struct cftype *cft, s64 val) | |
6f505b16 | 6764 | { |
182446d0 | 6765 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
6766 | } |
6767 | ||
182446d0 TH |
6768 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
6769 | struct cftype *cft) | |
6f505b16 | 6770 | { |
182446d0 | 6771 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 6772 | } |
d0b27fa7 | 6773 | |
182446d0 TH |
6774 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
6775 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 6776 | { |
182446d0 | 6777 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
6778 | } |
6779 | ||
182446d0 TH |
6780 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
6781 | struct cftype *cft) | |
d0b27fa7 | 6782 | { |
182446d0 | 6783 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 6784 | } |
6d6bc0ad | 6785 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 6786 | |
a1f7164c | 6787 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 6788 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
6789 | { |
6790 | .name = "shares", | |
f4c753b7 PM |
6791 | .read_u64 = cpu_shares_read_u64, |
6792 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 6793 | }, |
052f1dc7 | 6794 | #endif |
ab84d31e PT |
6795 | #ifdef CONFIG_CFS_BANDWIDTH |
6796 | { | |
6797 | .name = "cfs_quota_us", | |
6798 | .read_s64 = cpu_cfs_quota_read_s64, | |
6799 | .write_s64 = cpu_cfs_quota_write_s64, | |
6800 | }, | |
6801 | { | |
6802 | .name = "cfs_period_us", | |
6803 | .read_u64 = cpu_cfs_period_read_u64, | |
6804 | .write_u64 = cpu_cfs_period_write_u64, | |
6805 | }, | |
e8da1b18 NR |
6806 | { |
6807 | .name = "stat", | |
a1f7164c | 6808 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 6809 | }, |
ab84d31e | 6810 | #endif |
052f1dc7 | 6811 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6812 | { |
9f0c1e56 | 6813 | .name = "rt_runtime_us", |
06ecb27c PM |
6814 | .read_s64 = cpu_rt_runtime_read, |
6815 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 6816 | }, |
d0b27fa7 PZ |
6817 | { |
6818 | .name = "rt_period_us", | |
f4c753b7 PM |
6819 | .read_u64 = cpu_rt_period_read_uint, |
6820 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 6821 | }, |
052f1dc7 | 6822 | #endif |
d1ccc66d | 6823 | { } /* Terminate */ |
68318b8e SV |
6824 | }; |
6825 | ||
d41bf8c9 TH |
6826 | static int cpu_extra_stat_show(struct seq_file *sf, |
6827 | struct cgroup_subsys_state *css) | |
0d593634 | 6828 | { |
0d593634 TH |
6829 | #ifdef CONFIG_CFS_BANDWIDTH |
6830 | { | |
d41bf8c9 | 6831 | struct task_group *tg = css_tg(css); |
0d593634 TH |
6832 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
6833 | u64 throttled_usec; | |
6834 | ||
6835 | throttled_usec = cfs_b->throttled_time; | |
6836 | do_div(throttled_usec, NSEC_PER_USEC); | |
6837 | ||
6838 | seq_printf(sf, "nr_periods %d\n" | |
6839 | "nr_throttled %d\n" | |
6840 | "throttled_usec %llu\n", | |
6841 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
6842 | throttled_usec); | |
6843 | } | |
6844 | #endif | |
6845 | return 0; | |
6846 | } | |
6847 | ||
6848 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6849 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
6850 | struct cftype *cft) | |
6851 | { | |
6852 | struct task_group *tg = css_tg(css); | |
6853 | u64 weight = scale_load_down(tg->shares); | |
6854 | ||
6855 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
6856 | } | |
6857 | ||
6858 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
6859 | struct cftype *cft, u64 weight) | |
6860 | { | |
6861 | /* | |
6862 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
6863 | * values which are 1, 100 and 10000 respectively. While it loses | |
6864 | * a bit of range on both ends, it maps pretty well onto the shares | |
6865 | * value used by scheduler and the round-trip conversions preserve | |
6866 | * the original value over the entire range. | |
6867 | */ | |
6868 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
6869 | return -ERANGE; | |
6870 | ||
6871 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
6872 | ||
6873 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
6874 | } | |
6875 | ||
6876 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
6877 | struct cftype *cft) | |
6878 | { | |
6879 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
6880 | int last_delta = INT_MAX; | |
6881 | int prio, delta; | |
6882 | ||
6883 | /* find the closest nice value to the current weight */ | |
6884 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
6885 | delta = abs(sched_prio_to_weight[prio] - weight); | |
6886 | if (delta >= last_delta) | |
6887 | break; | |
6888 | last_delta = delta; | |
6889 | } | |
6890 | ||
6891 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
6892 | } | |
6893 | ||
6894 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
6895 | struct cftype *cft, s64 nice) | |
6896 | { | |
6897 | unsigned long weight; | |
7281c8de | 6898 | int idx; |
0d593634 TH |
6899 | |
6900 | if (nice < MIN_NICE || nice > MAX_NICE) | |
6901 | return -ERANGE; | |
6902 | ||
7281c8de PZ |
6903 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
6904 | idx = array_index_nospec(idx, 40); | |
6905 | weight = sched_prio_to_weight[idx]; | |
6906 | ||
0d593634 TH |
6907 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
6908 | } | |
6909 | #endif | |
6910 | ||
6911 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
6912 | long period, long quota) | |
6913 | { | |
6914 | if (quota < 0) | |
6915 | seq_puts(sf, "max"); | |
6916 | else | |
6917 | seq_printf(sf, "%ld", quota); | |
6918 | ||
6919 | seq_printf(sf, " %ld\n", period); | |
6920 | } | |
6921 | ||
6922 | /* caller should put the current value in *@periodp before calling */ | |
6923 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
6924 | u64 *periodp, u64 *quotap) | |
6925 | { | |
6926 | char tok[21]; /* U64_MAX */ | |
6927 | ||
6928 | if (!sscanf(buf, "%s %llu", tok, periodp)) | |
6929 | return -EINVAL; | |
6930 | ||
6931 | *periodp *= NSEC_PER_USEC; | |
6932 | ||
6933 | if (sscanf(tok, "%llu", quotap)) | |
6934 | *quotap *= NSEC_PER_USEC; | |
6935 | else if (!strcmp(tok, "max")) | |
6936 | *quotap = RUNTIME_INF; | |
6937 | else | |
6938 | return -EINVAL; | |
6939 | ||
6940 | return 0; | |
6941 | } | |
6942 | ||
6943 | #ifdef CONFIG_CFS_BANDWIDTH | |
6944 | static int cpu_max_show(struct seq_file *sf, void *v) | |
6945 | { | |
6946 | struct task_group *tg = css_tg(seq_css(sf)); | |
6947 | ||
6948 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
6949 | return 0; | |
6950 | } | |
6951 | ||
6952 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
6953 | char *buf, size_t nbytes, loff_t off) | |
6954 | { | |
6955 | struct task_group *tg = css_tg(of_css(of)); | |
6956 | u64 period = tg_get_cfs_period(tg); | |
6957 | u64 quota; | |
6958 | int ret; | |
6959 | ||
6960 | ret = cpu_period_quota_parse(buf, &period, "a); | |
6961 | if (!ret) | |
6962 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
6963 | return ret ?: nbytes; | |
6964 | } | |
6965 | #endif | |
6966 | ||
6967 | static struct cftype cpu_files[] = { | |
0d593634 TH |
6968 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6969 | { | |
6970 | .name = "weight", | |
6971 | .flags = CFTYPE_NOT_ON_ROOT, | |
6972 | .read_u64 = cpu_weight_read_u64, | |
6973 | .write_u64 = cpu_weight_write_u64, | |
6974 | }, | |
6975 | { | |
6976 | .name = "weight.nice", | |
6977 | .flags = CFTYPE_NOT_ON_ROOT, | |
6978 | .read_s64 = cpu_weight_nice_read_s64, | |
6979 | .write_s64 = cpu_weight_nice_write_s64, | |
6980 | }, | |
6981 | #endif | |
6982 | #ifdef CONFIG_CFS_BANDWIDTH | |
6983 | { | |
6984 | .name = "max", | |
6985 | .flags = CFTYPE_NOT_ON_ROOT, | |
6986 | .seq_show = cpu_max_show, | |
6987 | .write = cpu_max_write, | |
6988 | }, | |
6989 | #endif | |
6990 | { } /* terminate */ | |
6991 | }; | |
6992 | ||
073219e9 | 6993 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 6994 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 6995 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 6996 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 6997 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 6998 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 6999 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
7000 | .can_attach = cpu_cgroup_can_attach, |
7001 | .attach = cpu_cgroup_attach, | |
a1f7164c | 7002 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 7003 | .dfl_cftypes = cpu_files, |
b38e42e9 | 7004 | .early_init = true, |
0d593634 | 7005 | .threaded = true, |
68318b8e SV |
7006 | }; |
7007 | ||
052f1dc7 | 7008 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 7009 | |
b637a328 PM |
7010 | void dump_cpu_task(int cpu) |
7011 | { | |
7012 | pr_info("Task dump for CPU %d:\n", cpu); | |
7013 | sched_show_task(cpu_curr(cpu)); | |
7014 | } | |
ed82b8a1 AK |
7015 | |
7016 | /* | |
7017 | * Nice levels are multiplicative, with a gentle 10% change for every | |
7018 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
7019 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
7020 | * that remained on nice 0. | |
7021 | * | |
7022 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
7023 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
7024 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
7025 | * If a task goes up by ~10% and another task goes down by ~10% then | |
7026 | * the relative distance between them is ~25%.) | |
7027 | */ | |
7028 | const int sched_prio_to_weight[40] = { | |
7029 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
7030 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
7031 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
7032 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
7033 | /* 0 */ 1024, 820, 655, 526, 423, | |
7034 | /* 5 */ 335, 272, 215, 172, 137, | |
7035 | /* 10 */ 110, 87, 70, 56, 45, | |
7036 | /* 15 */ 36, 29, 23, 18, 15, | |
7037 | }; | |
7038 | ||
7039 | /* | |
7040 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
7041 | * | |
7042 | * In cases where the weight does not change often, we can use the | |
7043 | * precalculated inverse to speed up arithmetics by turning divisions | |
7044 | * into multiplications: | |
7045 | */ | |
7046 | const u32 sched_prio_to_wmult[40] = { | |
7047 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
7048 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
7049 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
7050 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
7051 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
7052 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
7053 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
7054 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
7055 | }; | |
14a7405b IM |
7056 | |
7057 | #undef CREATE_TRACE_POINTS |