]>
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 | |
e33a9bba TH |
2883 | /* |
2884 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
2885 | * | |
2886 | * The idea behind IO-wait account is to account the idle time that we could | |
2887 | * have spend running if it were not for IO. That is, if we were to improve the | |
2888 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
2889 | * | |
2890 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
2891 | * idle time as IO-wait, because if the storage were faster, it could've been | |
2892 | * running and we'd not be idle. | |
2893 | * | |
2894 | * This has been extended to SMP, by doing the same for each CPU. This however | |
2895 | * is broken. | |
2896 | * | |
2897 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
2898 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
2899 | * though, if the storage were faster, both could've ran at the same time, | |
2900 | * utilising both CPUs. | |
2901 | * | |
2902 | * This means, that when looking globally, the current IO-wait accounting on | |
2903 | * SMP is a lower bound, by reason of under accounting. | |
2904 | * | |
2905 | * Worse, since the numbers are provided per CPU, they are sometimes | |
2906 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
2907 | * associated with any one particular CPU, it can wake to another CPU than it | |
2908 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
2909 | * | |
2910 | * Task CPU affinities can make all that even more 'interesting'. | |
2911 | */ | |
2912 | ||
1da177e4 LT |
2913 | unsigned long nr_iowait(void) |
2914 | { | |
2915 | unsigned long i, sum = 0; | |
483b4ee6 | 2916 | |
0a945022 | 2917 | for_each_possible_cpu(i) |
1da177e4 | 2918 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2919 | |
1da177e4 LT |
2920 | return sum; |
2921 | } | |
483b4ee6 | 2922 | |
e33a9bba | 2923 | /* |
9c2298aa RW |
2924 | * Consumers of these two interfaces, like for example the cpuidle menu |
2925 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
2926 | * for a CPU that has IO-wait which might not even end up running the task when | |
2927 | * it does become runnable. | |
e33a9bba TH |
2928 | */ |
2929 | ||
8c215bd3 | 2930 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2931 | { |
8c215bd3 | 2932 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2933 | return atomic_read(&this->nr_iowait); |
2934 | } | |
46cb4b7c | 2935 | |
372ba8cb MG |
2936 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
2937 | { | |
3289bdb4 PZ |
2938 | struct rq *rq = this_rq(); |
2939 | *nr_waiters = atomic_read(&rq->nr_iowait); | |
2940 | *load = rq->load.weight; | |
372ba8cb MG |
2941 | } |
2942 | ||
dd41f596 | 2943 | #ifdef CONFIG_SMP |
8a0be9ef | 2944 | |
46cb4b7c | 2945 | /* |
38022906 PZ |
2946 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2947 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 2948 | */ |
38022906 | 2949 | void sched_exec(void) |
46cb4b7c | 2950 | { |
38022906 | 2951 | struct task_struct *p = current; |
1da177e4 | 2952 | unsigned long flags; |
0017d735 | 2953 | int dest_cpu; |
46cb4b7c | 2954 | |
8f42ced9 | 2955 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 2956 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
2957 | if (dest_cpu == smp_processor_id()) |
2958 | goto unlock; | |
38022906 | 2959 | |
8f42ced9 | 2960 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 2961 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 2962 | |
8f42ced9 PZ |
2963 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
2964 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
2965 | return; |
2966 | } | |
0017d735 | 2967 | unlock: |
8f42ced9 | 2968 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 2969 | } |
dd41f596 | 2970 | |
1da177e4 LT |
2971 | #endif |
2972 | ||
1da177e4 | 2973 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 2974 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
2975 | |
2976 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 2977 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 2978 | |
6075620b GG |
2979 | /* |
2980 | * The function fair_sched_class.update_curr accesses the struct curr | |
2981 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
2982 | * we observe a high rate of cache misses in practice. | |
2983 | * Prefetching this data results in improved performance. | |
2984 | */ | |
2985 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
2986 | { | |
2987 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
2988 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
2989 | #else | |
2990 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
2991 | #endif | |
2992 | prefetch(curr); | |
2993 | prefetch(&curr->exec_start); | |
2994 | } | |
2995 | ||
c5f8d995 HS |
2996 | /* |
2997 | * Return accounted runtime for the task. | |
2998 | * In case the task is currently running, return the runtime plus current's | |
2999 | * pending runtime that have not been accounted yet. | |
3000 | */ | |
3001 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3002 | { | |
eb580751 | 3003 | struct rq_flags rf; |
c5f8d995 | 3004 | struct rq *rq; |
6e998916 | 3005 | u64 ns; |
c5f8d995 | 3006 | |
911b2898 PZ |
3007 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3008 | /* | |
97fb7a0a | 3009 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
3010 | * So we have a optimization chance when the task's delta_exec is 0. |
3011 | * Reading ->on_cpu is racy, but this is ok. | |
3012 | * | |
d1ccc66d IM |
3013 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
3014 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 3015 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
3016 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3017 | * been accounted, so we're correct here as well. | |
911b2898 | 3018 | */ |
da0c1e65 | 3019 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3020 | return p->se.sum_exec_runtime; |
3021 | #endif | |
3022 | ||
eb580751 | 3023 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3024 | /* |
3025 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3026 | * project cycles that may never be accounted to this | |
3027 | * thread, breaking clock_gettime(). | |
3028 | */ | |
3029 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3030 | prefetch_curr_exec_start(p); |
6e998916 SG |
3031 | update_rq_clock(rq); |
3032 | p->sched_class->update_curr(rq); | |
3033 | } | |
3034 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3035 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3036 | |
3037 | return ns; | |
3038 | } | |
48f24c4d | 3039 | |
7835b98b CL |
3040 | /* |
3041 | * This function gets called by the timer code, with HZ frequency. | |
3042 | * We call it with interrupts disabled. | |
7835b98b CL |
3043 | */ |
3044 | void scheduler_tick(void) | |
3045 | { | |
7835b98b CL |
3046 | int cpu = smp_processor_id(); |
3047 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3048 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 3049 | struct rq_flags rf; |
3e51f33f PZ |
3050 | |
3051 | sched_clock_tick(); | |
dd41f596 | 3052 | |
8a8c69c3 PZ |
3053 | rq_lock(rq, &rf); |
3054 | ||
3e51f33f | 3055 | update_rq_clock(rq); |
fa85ae24 | 3056 | curr->sched_class->task_tick(rq, curr, 0); |
cee1afce | 3057 | cpu_load_update_active(rq); |
3289bdb4 | 3058 | calc_global_load_tick(rq); |
eb414681 | 3059 | psi_task_tick(rq); |
8a8c69c3 PZ |
3060 | |
3061 | rq_unlock(rq, &rf); | |
7835b98b | 3062 | |
e9d2b064 | 3063 | perf_event_task_tick(); |
e220d2dc | 3064 | |
e418e1c2 | 3065 | #ifdef CONFIG_SMP |
6eb57e0d | 3066 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3067 | trigger_load_balance(rq); |
e418e1c2 | 3068 | #endif |
1da177e4 LT |
3069 | } |
3070 | ||
265f22a9 | 3071 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
3072 | |
3073 | struct tick_work { | |
3074 | int cpu; | |
3075 | struct delayed_work work; | |
3076 | }; | |
3077 | ||
3078 | static struct tick_work __percpu *tick_work_cpu; | |
3079 | ||
3080 | static void sched_tick_remote(struct work_struct *work) | |
3081 | { | |
3082 | struct delayed_work *dwork = to_delayed_work(work); | |
3083 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
3084 | int cpu = twork->cpu; | |
3085 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 3086 | struct task_struct *curr; |
d84b3131 | 3087 | struct rq_flags rf; |
d9c0ffca | 3088 | u64 delta; |
d84b3131 FW |
3089 | |
3090 | /* | |
3091 | * Handle the tick only if it appears the remote CPU is running in full | |
3092 | * dynticks mode. The check is racy by nature, but missing a tick or | |
3093 | * having one too much is no big deal because the scheduler tick updates | |
3094 | * statistics and checks timeslices in a time-independent way, regardless | |
3095 | * of when exactly it is running. | |
3096 | */ | |
d9c0ffca FW |
3097 | if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu)) |
3098 | goto out_requeue; | |
d84b3131 | 3099 | |
d9c0ffca FW |
3100 | rq_lock_irq(rq, &rf); |
3101 | curr = rq->curr; | |
3102 | if (is_idle_task(curr)) | |
3103 | goto out_unlock; | |
d84b3131 | 3104 | |
d9c0ffca FW |
3105 | update_rq_clock(rq); |
3106 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
3107 | ||
3108 | /* | |
3109 | * Make sure the next tick runs within a reasonable | |
3110 | * amount of time. | |
3111 | */ | |
3112 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
3113 | curr->sched_class->task_tick(rq, curr, 0); | |
3114 | ||
3115 | out_unlock: | |
3116 | rq_unlock_irq(rq, &rf); | |
d84b3131 | 3117 | |
d9c0ffca | 3118 | out_requeue: |
d84b3131 FW |
3119 | /* |
3120 | * Run the remote tick once per second (1Hz). This arbitrary | |
3121 | * frequency is large enough to avoid overload but short enough | |
3122 | * to keep scheduler internal stats reasonably up to date. | |
3123 | */ | |
3124 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
3125 | } | |
3126 | ||
3127 | static void sched_tick_start(int cpu) | |
3128 | { | |
3129 | struct tick_work *twork; | |
3130 | ||
3131 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
3132 | return; | |
3133 | ||
3134 | WARN_ON_ONCE(!tick_work_cpu); | |
3135 | ||
3136 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
3137 | twork->cpu = cpu; | |
3138 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
3139 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
3140 | } | |
3141 | ||
3142 | #ifdef CONFIG_HOTPLUG_CPU | |
3143 | static void sched_tick_stop(int cpu) | |
3144 | { | |
3145 | struct tick_work *twork; | |
3146 | ||
3147 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
3148 | return; | |
3149 | ||
3150 | WARN_ON_ONCE(!tick_work_cpu); | |
3151 | ||
3152 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
3153 | cancel_delayed_work_sync(&twork->work); | |
3154 | } | |
3155 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3156 | ||
3157 | int __init sched_tick_offload_init(void) | |
3158 | { | |
3159 | tick_work_cpu = alloc_percpu(struct tick_work); | |
3160 | BUG_ON(!tick_work_cpu); | |
3161 | ||
3162 | return 0; | |
3163 | } | |
3164 | ||
3165 | #else /* !CONFIG_NO_HZ_FULL */ | |
3166 | static inline void sched_tick_start(int cpu) { } | |
3167 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 3168 | #endif |
1da177e4 | 3169 | |
7e49fcce | 3170 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 3171 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
3172 | /* |
3173 | * If the value passed in is equal to the current preempt count | |
3174 | * then we just disabled preemption. Start timing the latency. | |
3175 | */ | |
3176 | static inline void preempt_latency_start(int val) | |
3177 | { | |
3178 | if (preempt_count() == val) { | |
3179 | unsigned long ip = get_lock_parent_ip(); | |
3180 | #ifdef CONFIG_DEBUG_PREEMPT | |
3181 | current->preempt_disable_ip = ip; | |
3182 | #endif | |
3183 | trace_preempt_off(CALLER_ADDR0, ip); | |
3184 | } | |
3185 | } | |
7e49fcce | 3186 | |
edafe3a5 | 3187 | void preempt_count_add(int val) |
1da177e4 | 3188 | { |
6cd8a4bb | 3189 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3190 | /* |
3191 | * Underflow? | |
3192 | */ | |
9a11b49a IM |
3193 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3194 | return; | |
6cd8a4bb | 3195 | #endif |
bdb43806 | 3196 | __preempt_count_add(val); |
6cd8a4bb | 3197 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3198 | /* |
3199 | * Spinlock count overflowing soon? | |
3200 | */ | |
33859f7f MOS |
3201 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3202 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3203 | #endif |
47252cfb | 3204 | preempt_latency_start(val); |
1da177e4 | 3205 | } |
bdb43806 | 3206 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3207 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3208 | |
47252cfb SR |
3209 | /* |
3210 | * If the value passed in equals to the current preempt count | |
3211 | * then we just enabled preemption. Stop timing the latency. | |
3212 | */ | |
3213 | static inline void preempt_latency_stop(int val) | |
3214 | { | |
3215 | if (preempt_count() == val) | |
3216 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3217 | } | |
3218 | ||
edafe3a5 | 3219 | void preempt_count_sub(int val) |
1da177e4 | 3220 | { |
6cd8a4bb | 3221 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3222 | /* |
3223 | * Underflow? | |
3224 | */ | |
01e3eb82 | 3225 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3226 | return; |
1da177e4 LT |
3227 | /* |
3228 | * Is the spinlock portion underflowing? | |
3229 | */ | |
9a11b49a IM |
3230 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3231 | !(preempt_count() & PREEMPT_MASK))) | |
3232 | return; | |
6cd8a4bb | 3233 | #endif |
9a11b49a | 3234 | |
47252cfb | 3235 | preempt_latency_stop(val); |
bdb43806 | 3236 | __preempt_count_sub(val); |
1da177e4 | 3237 | } |
bdb43806 | 3238 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3239 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3240 | |
47252cfb SR |
3241 | #else |
3242 | static inline void preempt_latency_start(int val) { } | |
3243 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3244 | #endif |
3245 | ||
59ddbcb2 IM |
3246 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
3247 | { | |
3248 | #ifdef CONFIG_DEBUG_PREEMPT | |
3249 | return p->preempt_disable_ip; | |
3250 | #else | |
3251 | return 0; | |
3252 | #endif | |
3253 | } | |
3254 | ||
1da177e4 | 3255 | /* |
dd41f596 | 3256 | * Print scheduling while atomic bug: |
1da177e4 | 3257 | */ |
dd41f596 | 3258 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3259 | { |
d1c6d149 VN |
3260 | /* Save this before calling printk(), since that will clobber it */ |
3261 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3262 | ||
664dfa65 DJ |
3263 | if (oops_in_progress) |
3264 | return; | |
3265 | ||
3df0fc5b PZ |
3266 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3267 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3268 | |
dd41f596 | 3269 | debug_show_held_locks(prev); |
e21f5b15 | 3270 | print_modules(); |
dd41f596 IM |
3271 | if (irqs_disabled()) |
3272 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3273 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3274 | && in_atomic_preempt_off()) { | |
8f47b187 | 3275 | pr_err("Preemption disabled at:"); |
d1c6d149 | 3276 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
3277 | pr_cont("\n"); |
3278 | } | |
748c7201 DBO |
3279 | if (panic_on_warn) |
3280 | panic("scheduling while atomic\n"); | |
3281 | ||
6135fc1e | 3282 | dump_stack(); |
373d4d09 | 3283 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3284 | } |
1da177e4 | 3285 | |
dd41f596 IM |
3286 | /* |
3287 | * Various schedule()-time debugging checks and statistics: | |
3288 | */ | |
3289 | static inline void schedule_debug(struct task_struct *prev) | |
3290 | { | |
0d9e2632 | 3291 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3292 | if (task_stack_end_corrupted(prev)) |
3293 | panic("corrupted stack end detected inside scheduler\n"); | |
0d9e2632 | 3294 | #endif |
b99def8b | 3295 | |
1dc0fffc | 3296 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3297 | __schedule_bug(prev); |
1dc0fffc PZ |
3298 | preempt_count_set(PREEMPT_DISABLED); |
3299 | } | |
b3fbab05 | 3300 | rcu_sleep_check(); |
dd41f596 | 3301 | |
1da177e4 LT |
3302 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3303 | ||
ae92882e | 3304 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3305 | } |
3306 | ||
3307 | /* | |
3308 | * Pick up the highest-prio task: | |
3309 | */ | |
3310 | static inline struct task_struct * | |
d8ac8971 | 3311 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 3312 | { |
49ee5768 | 3313 | const struct sched_class *class; |
dd41f596 | 3314 | struct task_struct *p; |
1da177e4 LT |
3315 | |
3316 | /* | |
0ba87bb2 PZ |
3317 | * Optimization: we know that if all tasks are in the fair class we can |
3318 | * call that function directly, but only if the @prev task wasn't of a | |
3319 | * higher scheduling class, because otherwise those loose the | |
3320 | * opportunity to pull in more work from other CPUs. | |
1da177e4 | 3321 | */ |
0ba87bb2 PZ |
3322 | if (likely((prev->sched_class == &idle_sched_class || |
3323 | prev->sched_class == &fair_sched_class) && | |
3324 | rq->nr_running == rq->cfs.h_nr_running)) { | |
3325 | ||
d8ac8971 | 3326 | p = fair_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3327 | if (unlikely(p == RETRY_TASK)) |
3328 | goto again; | |
3329 | ||
d1ccc66d | 3330 | /* Assumes fair_sched_class->next == idle_sched_class */ |
6ccdc84b | 3331 | if (unlikely(!p)) |
d8ac8971 | 3332 | p = idle_sched_class.pick_next_task(rq, prev, rf); |
6ccdc84b PZ |
3333 | |
3334 | return p; | |
1da177e4 LT |
3335 | } |
3336 | ||
37e117c0 | 3337 | again: |
34f971f6 | 3338 | for_each_class(class) { |
d8ac8971 | 3339 | p = class->pick_next_task(rq, prev, rf); |
37e117c0 PZ |
3340 | if (p) { |
3341 | if (unlikely(p == RETRY_TASK)) | |
3342 | goto again; | |
dd41f596 | 3343 | return p; |
37e117c0 | 3344 | } |
dd41f596 | 3345 | } |
34f971f6 | 3346 | |
d1ccc66d IM |
3347 | /* The idle class should always have a runnable task: */ |
3348 | BUG(); | |
dd41f596 | 3349 | } |
1da177e4 | 3350 | |
dd41f596 | 3351 | /* |
c259e01a | 3352 | * __schedule() is the main scheduler function. |
edde96ea PE |
3353 | * |
3354 | * The main means of driving the scheduler and thus entering this function are: | |
3355 | * | |
3356 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
3357 | * | |
3358 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
3359 | * paths. For example, see arch/x86/entry_64.S. | |
3360 | * | |
3361 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
3362 | * interrupt handler scheduler_tick(). | |
3363 | * | |
3364 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
3365 | * task to the run-queue and that's it. | |
3366 | * | |
3367 | * Now, if the new task added to the run-queue preempts the current | |
3368 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
3369 | * called on the nearest possible occasion: | |
3370 | * | |
3371 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): | |
3372 | * | |
3373 | * - in syscall or exception context, at the next outmost | |
3374 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
3375 | * spin_unlock()!) | |
3376 | * | |
3377 | * - in IRQ context, return from interrupt-handler to | |
3378 | * preemptible context | |
3379 | * | |
3380 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) | |
3381 | * then at the next: | |
3382 | * | |
3383 | * - cond_resched() call | |
3384 | * - explicit schedule() call | |
3385 | * - return from syscall or exception to user-space | |
3386 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 3387 | * |
b30f0e3f | 3388 | * WARNING: must be called with preemption disabled! |
dd41f596 | 3389 | */ |
499d7955 | 3390 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
3391 | { |
3392 | struct task_struct *prev, *next; | |
67ca7bde | 3393 | unsigned long *switch_count; |
d8ac8971 | 3394 | struct rq_flags rf; |
dd41f596 | 3395 | struct rq *rq; |
31656519 | 3396 | int cpu; |
dd41f596 | 3397 | |
dd41f596 IM |
3398 | cpu = smp_processor_id(); |
3399 | rq = cpu_rq(cpu); | |
dd41f596 | 3400 | prev = rq->curr; |
dd41f596 | 3401 | |
dd41f596 | 3402 | schedule_debug(prev); |
1da177e4 | 3403 | |
31656519 | 3404 | if (sched_feat(HRTICK)) |
f333fdc9 | 3405 | hrtick_clear(rq); |
8f4d37ec | 3406 | |
46a5d164 | 3407 | local_irq_disable(); |
bcbfdd01 | 3408 | rcu_note_context_switch(preempt); |
46a5d164 | 3409 | |
e0acd0a6 ON |
3410 | /* |
3411 | * Make sure that signal_pending_state()->signal_pending() below | |
3412 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
3413 | * done by the caller to avoid the race with signal_wake_up(). | |
306e0604 MD |
3414 | * |
3415 | * The membarrier system call requires a full memory barrier | |
3416 | * after coming from user-space, before storing to rq->curr. | |
e0acd0a6 | 3417 | */ |
8a8c69c3 | 3418 | rq_lock(rq, &rf); |
d89e588c | 3419 | smp_mb__after_spinlock(); |
1da177e4 | 3420 | |
d1ccc66d IM |
3421 | /* Promote REQ to ACT */ |
3422 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 3423 | update_rq_clock(rq); |
9edfbfed | 3424 | |
246d86b5 | 3425 | switch_count = &prev->nivcsw; |
fc13aeba | 3426 | if (!preempt && prev->state) { |
21aa9af0 | 3427 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3428 | prev->state = TASK_RUNNING; |
21aa9af0 | 3429 | } else { |
bce4dc80 | 3430 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e PZ |
3431 | prev->on_rq = 0; |
3432 | ||
e33a9bba TH |
3433 | if (prev->in_iowait) { |
3434 | atomic_inc(&rq->nr_iowait); | |
3435 | delayacct_blkio_start(); | |
3436 | } | |
3437 | ||
21aa9af0 | 3438 | /* |
2acca55e PZ |
3439 | * If a worker went to sleep, notify and ask workqueue |
3440 | * whether it wants to wake up a task to maintain | |
3441 | * concurrency. | |
21aa9af0 TH |
3442 | */ |
3443 | if (prev->flags & PF_WQ_WORKER) { | |
3444 | struct task_struct *to_wakeup; | |
3445 | ||
9b7f6597 | 3446 | to_wakeup = wq_worker_sleeping(prev); |
21aa9af0 | 3447 | if (to_wakeup) |
d8ac8971 | 3448 | try_to_wake_up_local(to_wakeup, &rf); |
21aa9af0 | 3449 | } |
21aa9af0 | 3450 | } |
dd41f596 | 3451 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3452 | } |
3453 | ||
d8ac8971 | 3454 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 3455 | clear_tsk_need_resched(prev); |
f27dde8d | 3456 | clear_preempt_need_resched(); |
1da177e4 | 3457 | |
1da177e4 | 3458 | if (likely(prev != next)) { |
1da177e4 LT |
3459 | rq->nr_switches++; |
3460 | rq->curr = next; | |
22e4ebb9 MD |
3461 | /* |
3462 | * The membarrier system call requires each architecture | |
3463 | * to have a full memory barrier after updating | |
306e0604 MD |
3464 | * rq->curr, before returning to user-space. |
3465 | * | |
3466 | * Here are the schemes providing that barrier on the | |
3467 | * various architectures: | |
3468 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
3469 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
3470 | * - finish_lock_switch() for weakly-ordered | |
3471 | * architectures where spin_unlock is a full barrier, | |
3472 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
3473 | * is a RELEASE barrier), | |
22e4ebb9 | 3474 | */ |
1da177e4 LT |
3475 | ++*switch_count; |
3476 | ||
c73464b1 | 3477 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
3478 | |
3479 | /* Also unlocks the rq: */ | |
3480 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 3481 | } else { |
cb42c9a3 | 3482 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
8a8c69c3 | 3483 | rq_unlock_irq(rq, &rf); |
cbce1a68 | 3484 | } |
1da177e4 | 3485 | |
e3fca9e7 | 3486 | balance_callback(rq); |
1da177e4 | 3487 | } |
c259e01a | 3488 | |
9af6528e PZ |
3489 | void __noreturn do_task_dead(void) |
3490 | { | |
d1ccc66d | 3491 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 3492 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
3493 | |
3494 | /* Tell freezer to ignore us: */ | |
3495 | current->flags |= PF_NOFREEZE; | |
3496 | ||
9af6528e PZ |
3497 | __schedule(false); |
3498 | BUG(); | |
d1ccc66d IM |
3499 | |
3500 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 3501 | for (;;) |
d1ccc66d | 3502 | cpu_relax(); |
9af6528e PZ |
3503 | } |
3504 | ||
9c40cef2 TG |
3505 | static inline void sched_submit_work(struct task_struct *tsk) |
3506 | { | |
3c7d5184 | 3507 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
9c40cef2 TG |
3508 | return; |
3509 | /* | |
3510 | * If we are going to sleep and we have plugged IO queued, | |
3511 | * make sure to submit it to avoid deadlocks. | |
3512 | */ | |
3513 | if (blk_needs_flush_plug(tsk)) | |
3514 | blk_schedule_flush_plug(tsk); | |
3515 | } | |
3516 | ||
722a9f92 | 3517 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 3518 | { |
9c40cef2 TG |
3519 | struct task_struct *tsk = current; |
3520 | ||
3521 | sched_submit_work(tsk); | |
bfd9b2b5 | 3522 | do { |
b30f0e3f | 3523 | preempt_disable(); |
fc13aeba | 3524 | __schedule(false); |
b30f0e3f | 3525 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 3526 | } while (need_resched()); |
c259e01a | 3527 | } |
1da177e4 LT |
3528 | EXPORT_SYMBOL(schedule); |
3529 | ||
8663effb SRV |
3530 | /* |
3531 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
3532 | * state (have scheduled out non-voluntarily) by making sure that all | |
3533 | * tasks have either left the run queue or have gone into user space. | |
3534 | * As idle tasks do not do either, they must not ever be preempted | |
3535 | * (schedule out non-voluntarily). | |
3536 | * | |
3537 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
3538 | * never enables preemption because it does not call sched_submit_work(). | |
3539 | */ | |
3540 | void __sched schedule_idle(void) | |
3541 | { | |
3542 | /* | |
3543 | * As this skips calling sched_submit_work(), which the idle task does | |
3544 | * regardless because that function is a nop when the task is in a | |
3545 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
3546 | * current task can be in any other state. Note, idle is always in the | |
3547 | * TASK_RUNNING state. | |
3548 | */ | |
3549 | WARN_ON_ONCE(current->state); | |
3550 | do { | |
3551 | __schedule(false); | |
3552 | } while (need_resched()); | |
3553 | } | |
3554 | ||
91d1aa43 | 3555 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 3556 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
3557 | { |
3558 | /* | |
3559 | * If we come here after a random call to set_need_resched(), | |
3560 | * or we have been woken up remotely but the IPI has not yet arrived, | |
3561 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
3562 | * we find a better solution. | |
7cc78f8f AL |
3563 | * |
3564 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 3565 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 3566 | * too frequently to make sense yet. |
20ab65e3 | 3567 | */ |
7cc78f8f | 3568 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 3569 | schedule(); |
7cc78f8f | 3570 | exception_exit(prev_state); |
20ab65e3 FW |
3571 | } |
3572 | #endif | |
3573 | ||
c5491ea7 TG |
3574 | /** |
3575 | * schedule_preempt_disabled - called with preemption disabled | |
3576 | * | |
3577 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
3578 | */ | |
3579 | void __sched schedule_preempt_disabled(void) | |
3580 | { | |
ba74c144 | 3581 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
3582 | schedule(); |
3583 | preempt_disable(); | |
3584 | } | |
3585 | ||
06b1f808 | 3586 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
3587 | { |
3588 | do { | |
47252cfb SR |
3589 | /* |
3590 | * Because the function tracer can trace preempt_count_sub() | |
3591 | * and it also uses preempt_enable/disable_notrace(), if | |
3592 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3593 | * by the function tracer will call this function again and | |
3594 | * cause infinite recursion. | |
3595 | * | |
3596 | * Preemption must be disabled here before the function | |
3597 | * tracer can trace. Break up preempt_disable() into two | |
3598 | * calls. One to disable preemption without fear of being | |
3599 | * traced. The other to still record the preemption latency, | |
3600 | * which can also be traced by the function tracer. | |
3601 | */ | |
499d7955 | 3602 | preempt_disable_notrace(); |
47252cfb | 3603 | preempt_latency_start(1); |
fc13aeba | 3604 | __schedule(true); |
47252cfb | 3605 | preempt_latency_stop(1); |
499d7955 | 3606 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
3607 | |
3608 | /* | |
3609 | * Check again in case we missed a preemption opportunity | |
3610 | * between schedule and now. | |
3611 | */ | |
a18b5d01 FW |
3612 | } while (need_resched()); |
3613 | } | |
3614 | ||
1da177e4 LT |
3615 | #ifdef CONFIG_PREEMPT |
3616 | /* | |
2ed6e34f | 3617 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3618 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3619 | * occur there and call schedule directly. |
3620 | */ | |
722a9f92 | 3621 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 3622 | { |
1da177e4 LT |
3623 | /* |
3624 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3625 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3626 | */ |
fbb00b56 | 3627 | if (likely(!preemptible())) |
1da177e4 LT |
3628 | return; |
3629 | ||
a18b5d01 | 3630 | preempt_schedule_common(); |
1da177e4 | 3631 | } |
376e2424 | 3632 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 3633 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 3634 | |
009f60e2 | 3635 | /** |
4eaca0a8 | 3636 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
3637 | * |
3638 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
3639 | * recursion and tracing preempt enabling caused by the tracing | |
3640 | * infrastructure itself. But as tracing can happen in areas coming | |
3641 | * from userspace or just about to enter userspace, a preempt enable | |
3642 | * can occur before user_exit() is called. This will cause the scheduler | |
3643 | * to be called when the system is still in usermode. | |
3644 | * | |
3645 | * To prevent this, the preempt_enable_notrace will use this function | |
3646 | * instead of preempt_schedule() to exit user context if needed before | |
3647 | * calling the scheduler. | |
3648 | */ | |
4eaca0a8 | 3649 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
3650 | { |
3651 | enum ctx_state prev_ctx; | |
3652 | ||
3653 | if (likely(!preemptible())) | |
3654 | return; | |
3655 | ||
3656 | do { | |
47252cfb SR |
3657 | /* |
3658 | * Because the function tracer can trace preempt_count_sub() | |
3659 | * and it also uses preempt_enable/disable_notrace(), if | |
3660 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
3661 | * by the function tracer will call this function again and | |
3662 | * cause infinite recursion. | |
3663 | * | |
3664 | * Preemption must be disabled here before the function | |
3665 | * tracer can trace. Break up preempt_disable() into two | |
3666 | * calls. One to disable preemption without fear of being | |
3667 | * traced. The other to still record the preemption latency, | |
3668 | * which can also be traced by the function tracer. | |
3669 | */ | |
3d8f74dd | 3670 | preempt_disable_notrace(); |
47252cfb | 3671 | preempt_latency_start(1); |
009f60e2 ON |
3672 | /* |
3673 | * Needs preempt disabled in case user_exit() is traced | |
3674 | * and the tracer calls preempt_enable_notrace() causing | |
3675 | * an infinite recursion. | |
3676 | */ | |
3677 | prev_ctx = exception_enter(); | |
fc13aeba | 3678 | __schedule(true); |
009f60e2 ON |
3679 | exception_exit(prev_ctx); |
3680 | ||
47252cfb | 3681 | preempt_latency_stop(1); |
3d8f74dd | 3682 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
3683 | } while (need_resched()); |
3684 | } | |
4eaca0a8 | 3685 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 3686 | |
32e475d7 | 3687 | #endif /* CONFIG_PREEMPT */ |
1da177e4 LT |
3688 | |
3689 | /* | |
2ed6e34f | 3690 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3691 | * off of irq context. |
3692 | * Note, that this is called and return with irqs disabled. This will | |
3693 | * protect us against recursive calling from irq. | |
3694 | */ | |
722a9f92 | 3695 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 3696 | { |
b22366cd | 3697 | enum ctx_state prev_state; |
6478d880 | 3698 | |
2ed6e34f | 3699 | /* Catch callers which need to be fixed */ |
f27dde8d | 3700 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 3701 | |
b22366cd FW |
3702 | prev_state = exception_enter(); |
3703 | ||
3a5c359a | 3704 | do { |
3d8f74dd | 3705 | preempt_disable(); |
3a5c359a | 3706 | local_irq_enable(); |
fc13aeba | 3707 | __schedule(true); |
3a5c359a | 3708 | local_irq_disable(); |
3d8f74dd | 3709 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 3710 | } while (need_resched()); |
b22366cd FW |
3711 | |
3712 | exception_exit(prev_state); | |
1da177e4 LT |
3713 | } |
3714 | ||
ac6424b9 | 3715 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3716 | void *key) |
1da177e4 | 3717 | { |
63859d4f | 3718 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3719 | } |
1da177e4 LT |
3720 | EXPORT_SYMBOL(default_wake_function); |
3721 | ||
b29739f9 IM |
3722 | #ifdef CONFIG_RT_MUTEXES |
3723 | ||
acd58620 PZ |
3724 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
3725 | { | |
3726 | if (pi_task) | |
3727 | prio = min(prio, pi_task->prio); | |
3728 | ||
3729 | return prio; | |
3730 | } | |
3731 | ||
3732 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
3733 | { | |
3734 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
3735 | ||
3736 | return __rt_effective_prio(pi_task, prio); | |
3737 | } | |
3738 | ||
b29739f9 IM |
3739 | /* |
3740 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
3741 | * @p: task to boost |
3742 | * @pi_task: donor task | |
b29739f9 IM |
3743 | * |
3744 | * This function changes the 'effective' priority of a task. It does | |
3745 | * not touch ->normal_prio like __setscheduler(). | |
3746 | * | |
c365c292 TG |
3747 | * Used by the rt_mutex code to implement priority inheritance |
3748 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 3749 | */ |
acd58620 | 3750 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 3751 | { |
acd58620 | 3752 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 3753 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 3754 | const struct sched_class *prev_class; |
eb580751 PZ |
3755 | struct rq_flags rf; |
3756 | struct rq *rq; | |
b29739f9 | 3757 | |
acd58620 PZ |
3758 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
3759 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
3760 | ||
3761 | /* | |
3762 | * If nothing changed; bail early. | |
3763 | */ | |
3764 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
3765 | return; | |
b29739f9 | 3766 | |
eb580751 | 3767 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 3768 | update_rq_clock(rq); |
acd58620 PZ |
3769 | /* |
3770 | * Set under pi_lock && rq->lock, such that the value can be used under | |
3771 | * either lock. | |
3772 | * | |
3773 | * Note that there is loads of tricky to make this pointer cache work | |
3774 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
3775 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
3776 | * task is allowed to run again (and can exit). This ensures the pointer | |
3777 | * points to a blocked task -- which guaratees the task is present. | |
3778 | */ | |
3779 | p->pi_top_task = pi_task; | |
3780 | ||
3781 | /* | |
3782 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
3783 | */ | |
3784 | if (prio == p->prio && !dl_prio(prio)) | |
3785 | goto out_unlock; | |
b29739f9 | 3786 | |
1c4dd99b TG |
3787 | /* |
3788 | * Idle task boosting is a nono in general. There is one | |
3789 | * exception, when PREEMPT_RT and NOHZ is active: | |
3790 | * | |
3791 | * The idle task calls get_next_timer_interrupt() and holds | |
3792 | * the timer wheel base->lock on the CPU and another CPU wants | |
3793 | * to access the timer (probably to cancel it). We can safely | |
3794 | * ignore the boosting request, as the idle CPU runs this code | |
3795 | * with interrupts disabled and will complete the lock | |
3796 | * protected section without being interrupted. So there is no | |
3797 | * real need to boost. | |
3798 | */ | |
3799 | if (unlikely(p == rq->idle)) { | |
3800 | WARN_ON(p != rq->curr); | |
3801 | WARN_ON(p->pi_blocked_on); | |
3802 | goto out_unlock; | |
3803 | } | |
3804 | ||
b91473ff | 3805 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 3806 | oldprio = p->prio; |
ff77e468 PZ |
3807 | |
3808 | if (oldprio == prio) | |
3809 | queue_flag &= ~DEQUEUE_MOVE; | |
3810 | ||
83ab0aa0 | 3811 | prev_class = p->sched_class; |
da0c1e65 | 3812 | queued = task_on_rq_queued(p); |
051a1d1a | 3813 | running = task_current(rq, p); |
da0c1e65 | 3814 | if (queued) |
ff77e468 | 3815 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 3816 | if (running) |
f3cd1c4e | 3817 | put_prev_task(rq, p); |
dd41f596 | 3818 | |
2d3d891d DF |
3819 | /* |
3820 | * Boosting condition are: | |
3821 | * 1. -rt task is running and holds mutex A | |
3822 | * --> -dl task blocks on mutex A | |
3823 | * | |
3824 | * 2. -dl task is running and holds mutex A | |
3825 | * --> -dl task blocks on mutex A and could preempt the | |
3826 | * running task | |
3827 | */ | |
3828 | if (dl_prio(prio)) { | |
466af29b ON |
3829 | if (!dl_prio(p->normal_prio) || |
3830 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 3831 | p->dl.dl_boosted = 1; |
ff77e468 | 3832 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
3833 | } else |
3834 | p->dl.dl_boosted = 0; | |
aab03e05 | 3835 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
3836 | } else if (rt_prio(prio)) { |
3837 | if (dl_prio(oldprio)) | |
3838 | p->dl.dl_boosted = 0; | |
3839 | if (oldprio < prio) | |
ff77e468 | 3840 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 3841 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
3842 | } else { |
3843 | if (dl_prio(oldprio)) | |
3844 | p->dl.dl_boosted = 0; | |
746db944 BS |
3845 | if (rt_prio(oldprio)) |
3846 | p->rt.timeout = 0; | |
dd41f596 | 3847 | p->sched_class = &fair_sched_class; |
2d3d891d | 3848 | } |
dd41f596 | 3849 | |
b29739f9 IM |
3850 | p->prio = prio; |
3851 | ||
da0c1e65 | 3852 | if (queued) |
ff77e468 | 3853 | enqueue_task(rq, p, queue_flag); |
a399d233 | 3854 | if (running) |
b2bf6c31 | 3855 | set_curr_task(rq, p); |
cb469845 | 3856 | |
da7a735e | 3857 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 3858 | out_unlock: |
d1ccc66d IM |
3859 | /* Avoid rq from going away on us: */ |
3860 | preempt_disable(); | |
eb580751 | 3861 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
3862 | |
3863 | balance_callback(rq); | |
3864 | preempt_enable(); | |
b29739f9 | 3865 | } |
acd58620 PZ |
3866 | #else |
3867 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
3868 | { | |
3869 | return prio; | |
3870 | } | |
b29739f9 | 3871 | #endif |
d50dde5a | 3872 | |
36c8b586 | 3873 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3874 | { |
49bd21ef PZ |
3875 | bool queued, running; |
3876 | int old_prio, delta; | |
eb580751 | 3877 | struct rq_flags rf; |
70b97a7f | 3878 | struct rq *rq; |
1da177e4 | 3879 | |
75e45d51 | 3880 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
3881 | return; |
3882 | /* | |
3883 | * We have to be careful, if called from sys_setpriority(), | |
3884 | * the task might be in the middle of scheduling on another CPU. | |
3885 | */ | |
eb580751 | 3886 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
3887 | update_rq_clock(rq); |
3888 | ||
1da177e4 LT |
3889 | /* |
3890 | * The RT priorities are set via sched_setscheduler(), but we still | |
3891 | * allow the 'normal' nice value to be set - but as expected | |
3892 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 3893 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 3894 | */ |
aab03e05 | 3895 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
3896 | p->static_prio = NICE_TO_PRIO(nice); |
3897 | goto out_unlock; | |
3898 | } | |
da0c1e65 | 3899 | queued = task_on_rq_queued(p); |
49bd21ef | 3900 | running = task_current(rq, p); |
da0c1e65 | 3901 | if (queued) |
7a57f32a | 3902 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
3903 | if (running) |
3904 | put_prev_task(rq, p); | |
1da177e4 | 3905 | |
1da177e4 | 3906 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 3907 | set_load_weight(p, true); |
b29739f9 IM |
3908 | old_prio = p->prio; |
3909 | p->prio = effective_prio(p); | |
3910 | delta = p->prio - old_prio; | |
1da177e4 | 3911 | |
da0c1e65 | 3912 | if (queued) { |
7134b3e9 | 3913 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
1da177e4 | 3914 | /* |
d5f9f942 AM |
3915 | * If the task increased its priority or is running and |
3916 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3917 | */ |
d5f9f942 | 3918 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
8875125e | 3919 | resched_curr(rq); |
1da177e4 | 3920 | } |
49bd21ef PZ |
3921 | if (running) |
3922 | set_curr_task(rq, p); | |
1da177e4 | 3923 | out_unlock: |
eb580751 | 3924 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 3925 | } |
1da177e4 LT |
3926 | EXPORT_SYMBOL(set_user_nice); |
3927 | ||
e43379f1 MM |
3928 | /* |
3929 | * can_nice - check if a task can reduce its nice value | |
3930 | * @p: task | |
3931 | * @nice: nice value | |
3932 | */ | |
36c8b586 | 3933 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3934 | { |
d1ccc66d | 3935 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 3936 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 3937 | |
78d7d407 | 3938 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
3939 | capable(CAP_SYS_NICE)); |
3940 | } | |
3941 | ||
1da177e4 LT |
3942 | #ifdef __ARCH_WANT_SYS_NICE |
3943 | ||
3944 | /* | |
3945 | * sys_nice - change the priority of the current process. | |
3946 | * @increment: priority increment | |
3947 | * | |
3948 | * sys_setpriority is a more generic, but much slower function that | |
3949 | * does similar things. | |
3950 | */ | |
5add95d4 | 3951 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 3952 | { |
48f24c4d | 3953 | long nice, retval; |
1da177e4 LT |
3954 | |
3955 | /* | |
3956 | * Setpriority might change our priority at the same moment. | |
3957 | * We don't have to worry. Conceptually one call occurs first | |
3958 | * and we have a single winner. | |
3959 | */ | |
a9467fa3 | 3960 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 3961 | nice = task_nice(current) + increment; |
1da177e4 | 3962 | |
a9467fa3 | 3963 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
3964 | if (increment < 0 && !can_nice(current, nice)) |
3965 | return -EPERM; | |
3966 | ||
1da177e4 LT |
3967 | retval = security_task_setnice(current, nice); |
3968 | if (retval) | |
3969 | return retval; | |
3970 | ||
3971 | set_user_nice(current, nice); | |
3972 | return 0; | |
3973 | } | |
3974 | ||
3975 | #endif | |
3976 | ||
3977 | /** | |
3978 | * task_prio - return the priority value of a given task. | |
3979 | * @p: the task in question. | |
3980 | * | |
e69f6186 | 3981 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
3982 | * RT tasks are offset by -200. Normal tasks are centered |
3983 | * around 0, value goes from -16 to +15. | |
3984 | */ | |
36c8b586 | 3985 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3986 | { |
3987 | return p->prio - MAX_RT_PRIO; | |
3988 | } | |
3989 | ||
1da177e4 | 3990 | /** |
d1ccc66d | 3991 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 3992 | * @cpu: the processor in question. |
e69f6186 YB |
3993 | * |
3994 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
3995 | */ |
3996 | int idle_cpu(int cpu) | |
3997 | { | |
908a3283 TG |
3998 | struct rq *rq = cpu_rq(cpu); |
3999 | ||
4000 | if (rq->curr != rq->idle) | |
4001 | return 0; | |
4002 | ||
4003 | if (rq->nr_running) | |
4004 | return 0; | |
4005 | ||
4006 | #ifdef CONFIG_SMP | |
4007 | if (!llist_empty(&rq->wake_list)) | |
4008 | return 0; | |
4009 | #endif | |
4010 | ||
4011 | return 1; | |
1da177e4 LT |
4012 | } |
4013 | ||
943d355d RJ |
4014 | /** |
4015 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
4016 | * @cpu: the CPU in question. | |
4017 | * | |
4018 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
4019 | */ | |
4020 | int available_idle_cpu(int cpu) | |
4021 | { | |
4022 | if (!idle_cpu(cpu)) | |
4023 | return 0; | |
4024 | ||
247f2f6f RJ |
4025 | if (vcpu_is_preempted(cpu)) |
4026 | return 0; | |
4027 | ||
908a3283 | 4028 | return 1; |
1da177e4 LT |
4029 | } |
4030 | ||
1da177e4 | 4031 | /** |
d1ccc66d | 4032 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 4033 | * @cpu: the processor in question. |
e69f6186 | 4034 | * |
d1ccc66d | 4035 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 4036 | */ |
36c8b586 | 4037 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4038 | { |
4039 | return cpu_rq(cpu)->idle; | |
4040 | } | |
4041 | ||
4042 | /** | |
4043 | * find_process_by_pid - find a process with a matching PID value. | |
4044 | * @pid: the pid in question. | |
e69f6186 YB |
4045 | * |
4046 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 4047 | */ |
a9957449 | 4048 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4049 | { |
228ebcbe | 4050 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4051 | } |
4052 | ||
c13db6b1 SR |
4053 | /* |
4054 | * sched_setparam() passes in -1 for its policy, to let the functions | |
4055 | * it calls know not to change it. | |
4056 | */ | |
4057 | #define SETPARAM_POLICY -1 | |
4058 | ||
c365c292 TG |
4059 | static void __setscheduler_params(struct task_struct *p, |
4060 | const struct sched_attr *attr) | |
1da177e4 | 4061 | { |
d50dde5a DF |
4062 | int policy = attr->sched_policy; |
4063 | ||
c13db6b1 | 4064 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
4065 | policy = p->policy; |
4066 | ||
1da177e4 | 4067 | p->policy = policy; |
d50dde5a | 4068 | |
aab03e05 DF |
4069 | if (dl_policy(policy)) |
4070 | __setparam_dl(p, attr); | |
39fd8fd2 | 4071 | else if (fair_policy(policy)) |
d50dde5a DF |
4072 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
4073 | ||
39fd8fd2 PZ |
4074 | /* |
4075 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
4076 | * !rt_policy. Always setting this ensures that things like | |
4077 | * getparam()/getattr() don't report silly values for !rt tasks. | |
4078 | */ | |
4079 | p->rt_priority = attr->sched_priority; | |
383afd09 | 4080 | p->normal_prio = normal_prio(p); |
9059393e | 4081 | set_load_weight(p, true); |
c365c292 | 4082 | } |
39fd8fd2 | 4083 | |
c365c292 TG |
4084 | /* Actually do priority change: must hold pi & rq lock. */ |
4085 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 4086 | const struct sched_attr *attr, bool keep_boost) |
c365c292 TG |
4087 | { |
4088 | __setscheduler_params(p, attr); | |
d50dde5a | 4089 | |
383afd09 | 4090 | /* |
0782e63b TG |
4091 | * Keep a potential priority boosting if called from |
4092 | * sched_setscheduler(). | |
383afd09 | 4093 | */ |
acd58620 | 4094 | p->prio = normal_prio(p); |
0782e63b | 4095 | if (keep_boost) |
acd58620 | 4096 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 4097 | |
aab03e05 DF |
4098 | if (dl_prio(p->prio)) |
4099 | p->sched_class = &dl_sched_class; | |
4100 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
4101 | p->sched_class = &rt_sched_class; |
4102 | else | |
4103 | p->sched_class = &fair_sched_class; | |
1da177e4 | 4104 | } |
aab03e05 | 4105 | |
c69e8d9c | 4106 | /* |
d1ccc66d | 4107 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
4108 | */ |
4109 | static bool check_same_owner(struct task_struct *p) | |
4110 | { | |
4111 | const struct cred *cred = current_cred(), *pcred; | |
4112 | bool match; | |
4113 | ||
4114 | rcu_read_lock(); | |
4115 | pcred = __task_cred(p); | |
9c806aa0 EB |
4116 | match = (uid_eq(cred->euid, pcred->euid) || |
4117 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4118 | rcu_read_unlock(); |
4119 | return match; | |
4120 | } | |
4121 | ||
d50dde5a DF |
4122 | static int __sched_setscheduler(struct task_struct *p, |
4123 | const struct sched_attr *attr, | |
dbc7f069 | 4124 | bool user, bool pi) |
1da177e4 | 4125 | { |
383afd09 SR |
4126 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4127 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4128 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4129 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4130 | const struct sched_class *prev_class; |
eb580751 | 4131 | struct rq_flags rf; |
ca94c442 | 4132 | int reset_on_fork; |
7a57f32a | 4133 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 4134 | struct rq *rq; |
1da177e4 | 4135 | |
896bbb25 SRV |
4136 | /* The pi code expects interrupts enabled */ |
4137 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 4138 | recheck: |
d1ccc66d | 4139 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
4140 | if (policy < 0) { |
4141 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4142 | policy = oldpolicy = p->policy; |
ca94c442 | 4143 | } else { |
7479f3c9 | 4144 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4145 | |
20f9cd2a | 4146 | if (!valid_policy(policy)) |
ca94c442 LP |
4147 | return -EINVAL; |
4148 | } | |
4149 | ||
794a56eb | 4150 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
4151 | return -EINVAL; |
4152 | ||
1da177e4 LT |
4153 | /* |
4154 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4155 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4156 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4157 | */ |
0bb040a4 | 4158 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4159 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4160 | return -EINVAL; |
aab03e05 DF |
4161 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4162 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4163 | return -EINVAL; |
4164 | ||
37e4ab3f OC |
4165 | /* |
4166 | * Allow unprivileged RT tasks to decrease priority: | |
4167 | */ | |
961ccddd | 4168 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4169 | if (fair_policy(policy)) { |
d0ea0268 | 4170 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4171 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4172 | return -EPERM; |
4173 | } | |
4174 | ||
e05606d3 | 4175 | if (rt_policy(policy)) { |
a44702e8 ON |
4176 | unsigned long rlim_rtprio = |
4177 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 4178 | |
d1ccc66d | 4179 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
4180 | if (policy != p->policy && !rlim_rtprio) |
4181 | return -EPERM; | |
4182 | ||
d1ccc66d | 4183 | /* Can't increase priority: */ |
d50dde5a DF |
4184 | if (attr->sched_priority > p->rt_priority && |
4185 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4186 | return -EPERM; |
4187 | } | |
c02aa73b | 4188 | |
d44753b8 JL |
4189 | /* |
4190 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4191 | * (safest behavior); in the future we would like to allow | |
4192 | * unprivileged DL tasks to increase their relative deadline | |
4193 | * or reduce their runtime (both ways reducing utilization) | |
4194 | */ | |
4195 | if (dl_policy(policy)) | |
4196 | return -EPERM; | |
4197 | ||
dd41f596 | 4198 | /* |
c02aa73b DH |
4199 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4200 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4201 | */ |
20f9cd2a | 4202 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
d0ea0268 | 4203 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4204 | return -EPERM; |
4205 | } | |
5fe1d75f | 4206 | |
d1ccc66d | 4207 | /* Can't change other user's priorities: */ |
c69e8d9c | 4208 | if (!check_same_owner(p)) |
37e4ab3f | 4209 | return -EPERM; |
ca94c442 | 4210 | |
d1ccc66d | 4211 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
4212 | if (p->sched_reset_on_fork && !reset_on_fork) |
4213 | return -EPERM; | |
37e4ab3f | 4214 | } |
1da177e4 | 4215 | |
725aad24 | 4216 | if (user) { |
794a56eb JL |
4217 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
4218 | return -EINVAL; | |
4219 | ||
b0ae1981 | 4220 | retval = security_task_setscheduler(p); |
725aad24 JF |
4221 | if (retval) |
4222 | return retval; | |
4223 | } | |
4224 | ||
b29739f9 | 4225 | /* |
d1ccc66d | 4226 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 4227 | * changing the priority of the task: |
0122ec5b | 4228 | * |
25985edc | 4229 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4230 | * runqueue lock must be held. |
4231 | */ | |
eb580751 | 4232 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 4233 | update_rq_clock(rq); |
dc61b1d6 | 4234 | |
34f971f6 | 4235 | /* |
d1ccc66d | 4236 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
4237 | */ |
4238 | if (p == rq->stop) { | |
eb580751 | 4239 | task_rq_unlock(rq, p, &rf); |
34f971f6 PZ |
4240 | return -EINVAL; |
4241 | } | |
4242 | ||
a51e9198 | 4243 | /* |
d6b1e911 TG |
4244 | * If not changing anything there's no need to proceed further, |
4245 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4246 | */ |
d50dde5a | 4247 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4248 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4249 | goto change; |
4250 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4251 | goto change; | |
75381608 | 4252 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4253 | goto change; |
d50dde5a | 4254 | |
d6b1e911 | 4255 | p->sched_reset_on_fork = reset_on_fork; |
eb580751 | 4256 | task_rq_unlock(rq, p, &rf); |
a51e9198 DF |
4257 | return 0; |
4258 | } | |
d50dde5a | 4259 | change: |
a51e9198 | 4260 | |
dc61b1d6 | 4261 | if (user) { |
332ac17e | 4262 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4263 | /* |
4264 | * Do not allow realtime tasks into groups that have no runtime | |
4265 | * assigned. | |
4266 | */ | |
4267 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4268 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4269 | !task_group_is_autogroup(task_group(p))) { | |
eb580751 | 4270 | task_rq_unlock(rq, p, &rf); |
dc61b1d6 PZ |
4271 | return -EPERM; |
4272 | } | |
dc61b1d6 | 4273 | #endif |
332ac17e | 4274 | #ifdef CONFIG_SMP |
794a56eb JL |
4275 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
4276 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 4277 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
4278 | |
4279 | /* | |
4280 | * Don't allow tasks with an affinity mask smaller than | |
4281 | * the entire root_domain to become SCHED_DEADLINE. We | |
4282 | * will also fail if there's no bandwidth available. | |
4283 | */ | |
e4099a5e PZ |
4284 | if (!cpumask_subset(span, &p->cpus_allowed) || |
4285 | rq->rd->dl_bw.bw == 0) { | |
eb580751 | 4286 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4287 | return -EPERM; |
4288 | } | |
4289 | } | |
4290 | #endif | |
4291 | } | |
dc61b1d6 | 4292 | |
d1ccc66d | 4293 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
4294 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
4295 | policy = oldpolicy = -1; | |
eb580751 | 4296 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4297 | goto recheck; |
4298 | } | |
332ac17e DF |
4299 | |
4300 | /* | |
4301 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
4302 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
4303 | * is available. | |
4304 | */ | |
06a76fe0 | 4305 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
eb580751 | 4306 | task_rq_unlock(rq, p, &rf); |
332ac17e DF |
4307 | return -EBUSY; |
4308 | } | |
4309 | ||
c365c292 TG |
4310 | p->sched_reset_on_fork = reset_on_fork; |
4311 | oldprio = p->prio; | |
4312 | ||
dbc7f069 PZ |
4313 | if (pi) { |
4314 | /* | |
4315 | * Take priority boosted tasks into account. If the new | |
4316 | * effective priority is unchanged, we just store the new | |
4317 | * normal parameters and do not touch the scheduler class and | |
4318 | * the runqueue. This will be done when the task deboost | |
4319 | * itself. | |
4320 | */ | |
acd58620 | 4321 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
4322 | if (new_effective_prio == oldprio) |
4323 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
4324 | } |
4325 | ||
da0c1e65 | 4326 | queued = task_on_rq_queued(p); |
051a1d1a | 4327 | running = task_current(rq, p); |
da0c1e65 | 4328 | if (queued) |
ff77e468 | 4329 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 4330 | if (running) |
f3cd1c4e | 4331 | put_prev_task(rq, p); |
f6b53205 | 4332 | |
83ab0aa0 | 4333 | prev_class = p->sched_class; |
dbc7f069 | 4334 | __setscheduler(rq, p, attr, pi); |
f6b53205 | 4335 | |
da0c1e65 | 4336 | if (queued) { |
81a44c54 TG |
4337 | /* |
4338 | * We enqueue to tail when the priority of a task is | |
4339 | * increased (user space view). | |
4340 | */ | |
ff77e468 PZ |
4341 | if (oldprio < p->prio) |
4342 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 4343 | |
ff77e468 | 4344 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 4345 | } |
a399d233 | 4346 | if (running) |
b2bf6c31 | 4347 | set_curr_task(rq, p); |
cb469845 | 4348 | |
da7a735e | 4349 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
4350 | |
4351 | /* Avoid rq from going away on us: */ | |
4352 | preempt_disable(); | |
eb580751 | 4353 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 4354 | |
dbc7f069 PZ |
4355 | if (pi) |
4356 | rt_mutex_adjust_pi(p); | |
95e02ca9 | 4357 | |
d1ccc66d | 4358 | /* Run balance callbacks after we've adjusted the PI chain: */ |
4c9a4bc8 PZ |
4359 | balance_callback(rq); |
4360 | preempt_enable(); | |
95e02ca9 | 4361 | |
1da177e4 LT |
4362 | return 0; |
4363 | } | |
961ccddd | 4364 | |
7479f3c9 PZ |
4365 | static int _sched_setscheduler(struct task_struct *p, int policy, |
4366 | const struct sched_param *param, bool check) | |
4367 | { | |
4368 | struct sched_attr attr = { | |
4369 | .sched_policy = policy, | |
4370 | .sched_priority = param->sched_priority, | |
4371 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
4372 | }; | |
4373 | ||
c13db6b1 SR |
4374 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
4375 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
4376 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
4377 | policy &= ~SCHED_RESET_ON_FORK; | |
4378 | attr.sched_policy = policy; | |
4379 | } | |
4380 | ||
dbc7f069 | 4381 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 4382 | } |
961ccddd RR |
4383 | /** |
4384 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4385 | * @p: the task in question. | |
4386 | * @policy: new policy. | |
4387 | * @param: structure containing the new RT priority. | |
4388 | * | |
e69f6186 YB |
4389 | * Return: 0 on success. An error code otherwise. |
4390 | * | |
961ccddd RR |
4391 | * NOTE that the task may be already dead. |
4392 | */ | |
4393 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4394 | const struct sched_param *param) |
961ccddd | 4395 | { |
7479f3c9 | 4396 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 4397 | } |
1da177e4 LT |
4398 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4399 | ||
d50dde5a DF |
4400 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
4401 | { | |
dbc7f069 | 4402 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
4403 | } |
4404 | EXPORT_SYMBOL_GPL(sched_setattr); | |
4405 | ||
794a56eb JL |
4406 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
4407 | { | |
4408 | return __sched_setscheduler(p, attr, false, true); | |
4409 | } | |
4410 | ||
961ccddd RR |
4411 | /** |
4412 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4413 | * @p: the task in question. | |
4414 | * @policy: new policy. | |
4415 | * @param: structure containing the new RT priority. | |
4416 | * | |
4417 | * Just like sched_setscheduler, only don't bother checking if the | |
4418 | * current context has permission. For example, this is needed in | |
4419 | * stop_machine(): we create temporary high priority worker threads, | |
4420 | * but our caller might not have that capability. | |
e69f6186 YB |
4421 | * |
4422 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
4423 | */ |
4424 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4425 | const struct sched_param *param) |
961ccddd | 4426 | { |
7479f3c9 | 4427 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 4428 | } |
84778472 | 4429 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 4430 | |
95cdf3b7 IM |
4431 | static int |
4432 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4433 | { |
1da177e4 LT |
4434 | struct sched_param lparam; |
4435 | struct task_struct *p; | |
36c8b586 | 4436 | int retval; |
1da177e4 LT |
4437 | |
4438 | if (!param || pid < 0) | |
4439 | return -EINVAL; | |
4440 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4441 | return -EFAULT; | |
5fe1d75f ON |
4442 | |
4443 | rcu_read_lock(); | |
4444 | retval = -ESRCH; | |
1da177e4 | 4445 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4446 | if (p != NULL) |
4447 | retval = sched_setscheduler(p, policy, &lparam); | |
4448 | rcu_read_unlock(); | |
36c8b586 | 4449 | |
1da177e4 LT |
4450 | return retval; |
4451 | } | |
4452 | ||
d50dde5a DF |
4453 | /* |
4454 | * Mimics kernel/events/core.c perf_copy_attr(). | |
4455 | */ | |
d1ccc66d | 4456 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
4457 | { |
4458 | u32 size; | |
4459 | int ret; | |
4460 | ||
4461 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) | |
4462 | return -EFAULT; | |
4463 | ||
d1ccc66d | 4464 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
4465 | memset(attr, 0, sizeof(*attr)); |
4466 | ||
4467 | ret = get_user(size, &uattr->size); | |
4468 | if (ret) | |
4469 | return ret; | |
4470 | ||
d1ccc66d IM |
4471 | /* Bail out on silly large: */ |
4472 | if (size > PAGE_SIZE) | |
d50dde5a DF |
4473 | goto err_size; |
4474 | ||
d1ccc66d IM |
4475 | /* ABI compatibility quirk: */ |
4476 | if (!size) | |
d50dde5a DF |
4477 | size = SCHED_ATTR_SIZE_VER0; |
4478 | ||
4479 | if (size < SCHED_ATTR_SIZE_VER0) | |
4480 | goto err_size; | |
4481 | ||
4482 | /* | |
4483 | * If we're handed a bigger struct than we know of, | |
4484 | * ensure all the unknown bits are 0 - i.e. new | |
4485 | * user-space does not rely on any kernel feature | |
4486 | * extensions we dont know about yet. | |
4487 | */ | |
4488 | if (size > sizeof(*attr)) { | |
4489 | unsigned char __user *addr; | |
4490 | unsigned char __user *end; | |
4491 | unsigned char val; | |
4492 | ||
4493 | addr = (void __user *)uattr + sizeof(*attr); | |
4494 | end = (void __user *)uattr + size; | |
4495 | ||
4496 | for (; addr < end; addr++) { | |
4497 | ret = get_user(val, addr); | |
4498 | if (ret) | |
4499 | return ret; | |
4500 | if (val) | |
4501 | goto err_size; | |
4502 | } | |
4503 | size = sizeof(*attr); | |
4504 | } | |
4505 | ||
4506 | ret = copy_from_user(attr, uattr, size); | |
4507 | if (ret) | |
4508 | return -EFAULT; | |
4509 | ||
4510 | /* | |
d1ccc66d | 4511 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
4512 | * to be strict and return an error on out-of-bounds values? |
4513 | */ | |
75e45d51 | 4514 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 4515 | |
e78c7bca | 4516 | return 0; |
d50dde5a DF |
4517 | |
4518 | err_size: | |
4519 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 4520 | return -E2BIG; |
d50dde5a DF |
4521 | } |
4522 | ||
1da177e4 LT |
4523 | /** |
4524 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4525 | * @pid: the pid in question. | |
4526 | * @policy: new policy. | |
4527 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4528 | * |
4529 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4530 | */ |
d1ccc66d | 4531 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 4532 | { |
c21761f1 JB |
4533 | if (policy < 0) |
4534 | return -EINVAL; | |
4535 | ||
1da177e4 LT |
4536 | return do_sched_setscheduler(pid, policy, param); |
4537 | } | |
4538 | ||
4539 | /** | |
4540 | * sys_sched_setparam - set/change the RT priority of a thread | |
4541 | * @pid: the pid in question. | |
4542 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
4543 | * |
4544 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4545 | */ |
5add95d4 | 4546 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4547 | { |
c13db6b1 | 4548 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
4549 | } |
4550 | ||
d50dde5a DF |
4551 | /** |
4552 | * sys_sched_setattr - same as above, but with extended sched_attr | |
4553 | * @pid: the pid in question. | |
5778fccf | 4554 | * @uattr: structure containing the extended parameters. |
db66d756 | 4555 | * @flags: for future extension. |
d50dde5a | 4556 | */ |
6d35ab48 PZ |
4557 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
4558 | unsigned int, flags) | |
d50dde5a DF |
4559 | { |
4560 | struct sched_attr attr; | |
4561 | struct task_struct *p; | |
4562 | int retval; | |
4563 | ||
6d35ab48 | 4564 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
4565 | return -EINVAL; |
4566 | ||
143cf23d MK |
4567 | retval = sched_copy_attr(uattr, &attr); |
4568 | if (retval) | |
4569 | return retval; | |
d50dde5a | 4570 | |
b14ed2c2 | 4571 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 4572 | return -EINVAL; |
d50dde5a DF |
4573 | |
4574 | rcu_read_lock(); | |
4575 | retval = -ESRCH; | |
4576 | p = find_process_by_pid(pid); | |
4577 | if (p != NULL) | |
4578 | retval = sched_setattr(p, &attr); | |
4579 | rcu_read_unlock(); | |
4580 | ||
4581 | return retval; | |
4582 | } | |
4583 | ||
1da177e4 LT |
4584 | /** |
4585 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4586 | * @pid: the pid in question. | |
e69f6186 YB |
4587 | * |
4588 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
4589 | * code. | |
1da177e4 | 4590 | */ |
5add95d4 | 4591 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4592 | { |
36c8b586 | 4593 | struct task_struct *p; |
3a5c359a | 4594 | int retval; |
1da177e4 LT |
4595 | |
4596 | if (pid < 0) | |
3a5c359a | 4597 | return -EINVAL; |
1da177e4 LT |
4598 | |
4599 | retval = -ESRCH; | |
5fe85be0 | 4600 | rcu_read_lock(); |
1da177e4 LT |
4601 | p = find_process_by_pid(pid); |
4602 | if (p) { | |
4603 | retval = security_task_getscheduler(p); | |
4604 | if (!retval) | |
ca94c442 LP |
4605 | retval = p->policy |
4606 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4607 | } |
5fe85be0 | 4608 | rcu_read_unlock(); |
1da177e4 LT |
4609 | return retval; |
4610 | } | |
4611 | ||
4612 | /** | |
ca94c442 | 4613 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4614 | * @pid: the pid in question. |
4615 | * @param: structure containing the RT priority. | |
e69f6186 YB |
4616 | * |
4617 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
4618 | * code. | |
1da177e4 | 4619 | */ |
5add95d4 | 4620 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 4621 | { |
ce5f7f82 | 4622 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 4623 | struct task_struct *p; |
3a5c359a | 4624 | int retval; |
1da177e4 LT |
4625 | |
4626 | if (!param || pid < 0) | |
3a5c359a | 4627 | return -EINVAL; |
1da177e4 | 4628 | |
5fe85be0 | 4629 | rcu_read_lock(); |
1da177e4 LT |
4630 | p = find_process_by_pid(pid); |
4631 | retval = -ESRCH; | |
4632 | if (!p) | |
4633 | goto out_unlock; | |
4634 | ||
4635 | retval = security_task_getscheduler(p); | |
4636 | if (retval) | |
4637 | goto out_unlock; | |
4638 | ||
ce5f7f82 PZ |
4639 | if (task_has_rt_policy(p)) |
4640 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4641 | rcu_read_unlock(); |
1da177e4 LT |
4642 | |
4643 | /* | |
4644 | * This one might sleep, we cannot do it with a spinlock held ... | |
4645 | */ | |
4646 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4647 | ||
1da177e4 LT |
4648 | return retval; |
4649 | ||
4650 | out_unlock: | |
5fe85be0 | 4651 | rcu_read_unlock(); |
1da177e4 LT |
4652 | return retval; |
4653 | } | |
4654 | ||
d50dde5a DF |
4655 | static int sched_read_attr(struct sched_attr __user *uattr, |
4656 | struct sched_attr *attr, | |
4657 | unsigned int usize) | |
4658 | { | |
4659 | int ret; | |
4660 | ||
4661 | if (!access_ok(VERIFY_WRITE, uattr, usize)) | |
4662 | return -EFAULT; | |
4663 | ||
4664 | /* | |
4665 | * If we're handed a smaller struct than we know of, | |
4666 | * ensure all the unknown bits are 0 - i.e. old | |
4667 | * user-space does not get uncomplete information. | |
4668 | */ | |
4669 | if (usize < sizeof(*attr)) { | |
4670 | unsigned char *addr; | |
4671 | unsigned char *end; | |
4672 | ||
4673 | addr = (void *)attr + usize; | |
4674 | end = (void *)attr + sizeof(*attr); | |
4675 | ||
4676 | for (; addr < end; addr++) { | |
4677 | if (*addr) | |
22400674 | 4678 | return -EFBIG; |
d50dde5a DF |
4679 | } |
4680 | ||
4681 | attr->size = usize; | |
4682 | } | |
4683 | ||
4efbc454 | 4684 | ret = copy_to_user(uattr, attr, attr->size); |
d50dde5a DF |
4685 | if (ret) |
4686 | return -EFAULT; | |
4687 | ||
22400674 | 4688 | return 0; |
d50dde5a DF |
4689 | } |
4690 | ||
4691 | /** | |
aab03e05 | 4692 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 4693 | * @pid: the pid in question. |
5778fccf | 4694 | * @uattr: structure containing the extended parameters. |
d50dde5a | 4695 | * @size: sizeof(attr) for fwd/bwd comp. |
db66d756 | 4696 | * @flags: for future extension. |
d50dde5a | 4697 | */ |
6d35ab48 PZ |
4698 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
4699 | unsigned int, size, unsigned int, flags) | |
d50dde5a DF |
4700 | { |
4701 | struct sched_attr attr = { | |
4702 | .size = sizeof(struct sched_attr), | |
4703 | }; | |
4704 | struct task_struct *p; | |
4705 | int retval; | |
4706 | ||
4707 | if (!uattr || pid < 0 || size > PAGE_SIZE || | |
6d35ab48 | 4708 | size < SCHED_ATTR_SIZE_VER0 || flags) |
d50dde5a DF |
4709 | return -EINVAL; |
4710 | ||
4711 | rcu_read_lock(); | |
4712 | p = find_process_by_pid(pid); | |
4713 | retval = -ESRCH; | |
4714 | if (!p) | |
4715 | goto out_unlock; | |
4716 | ||
4717 | retval = security_task_getscheduler(p); | |
4718 | if (retval) | |
4719 | goto out_unlock; | |
4720 | ||
4721 | attr.sched_policy = p->policy; | |
7479f3c9 PZ |
4722 | if (p->sched_reset_on_fork) |
4723 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | |
aab03e05 DF |
4724 | if (task_has_dl_policy(p)) |
4725 | __getparam_dl(p, &attr); | |
4726 | else if (task_has_rt_policy(p)) | |
d50dde5a DF |
4727 | attr.sched_priority = p->rt_priority; |
4728 | else | |
d0ea0268 | 4729 | attr.sched_nice = task_nice(p); |
d50dde5a DF |
4730 | |
4731 | rcu_read_unlock(); | |
4732 | ||
4733 | retval = sched_read_attr(uattr, &attr, size); | |
4734 | return retval; | |
4735 | ||
4736 | out_unlock: | |
4737 | rcu_read_unlock(); | |
4738 | return retval; | |
4739 | } | |
4740 | ||
96f874e2 | 4741 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4742 | { |
5a16f3d3 | 4743 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4744 | struct task_struct *p; |
4745 | int retval; | |
1da177e4 | 4746 | |
23f5d142 | 4747 | rcu_read_lock(); |
1da177e4 LT |
4748 | |
4749 | p = find_process_by_pid(pid); | |
4750 | if (!p) { | |
23f5d142 | 4751 | rcu_read_unlock(); |
1da177e4 LT |
4752 | return -ESRCH; |
4753 | } | |
4754 | ||
23f5d142 | 4755 | /* Prevent p going away */ |
1da177e4 | 4756 | get_task_struct(p); |
23f5d142 | 4757 | rcu_read_unlock(); |
1da177e4 | 4758 | |
14a40ffc TH |
4759 | if (p->flags & PF_NO_SETAFFINITY) { |
4760 | retval = -EINVAL; | |
4761 | goto out_put_task; | |
4762 | } | |
5a16f3d3 RR |
4763 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4764 | retval = -ENOMEM; | |
4765 | goto out_put_task; | |
4766 | } | |
4767 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4768 | retval = -ENOMEM; | |
4769 | goto out_free_cpus_allowed; | |
4770 | } | |
1da177e4 | 4771 | retval = -EPERM; |
4c44aaaf EB |
4772 | if (!check_same_owner(p)) { |
4773 | rcu_read_lock(); | |
4774 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
4775 | rcu_read_unlock(); | |
16303ab2 | 4776 | goto out_free_new_mask; |
4c44aaaf EB |
4777 | } |
4778 | rcu_read_unlock(); | |
4779 | } | |
1da177e4 | 4780 | |
b0ae1981 | 4781 | retval = security_task_setscheduler(p); |
e7834f8f | 4782 | if (retval) |
16303ab2 | 4783 | goto out_free_new_mask; |
e7834f8f | 4784 | |
e4099a5e PZ |
4785 | |
4786 | cpuset_cpus_allowed(p, cpus_allowed); | |
4787 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
4788 | ||
332ac17e DF |
4789 | /* |
4790 | * Since bandwidth control happens on root_domain basis, | |
4791 | * if admission test is enabled, we only admit -deadline | |
4792 | * tasks allowed to run on all the CPUs in the task's | |
4793 | * root_domain. | |
4794 | */ | |
4795 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
4796 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
4797 | rcu_read_lock(); | |
4798 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 4799 | retval = -EBUSY; |
f1e3a093 | 4800 | rcu_read_unlock(); |
16303ab2 | 4801 | goto out_free_new_mask; |
332ac17e | 4802 | } |
f1e3a093 | 4803 | rcu_read_unlock(); |
332ac17e DF |
4804 | } |
4805 | #endif | |
49246274 | 4806 | again: |
25834c73 | 4807 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 4808 | |
8707d8b8 | 4809 | if (!retval) { |
5a16f3d3 RR |
4810 | cpuset_cpus_allowed(p, cpus_allowed); |
4811 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4812 | /* |
4813 | * We must have raced with a concurrent cpuset | |
4814 | * update. Just reset the cpus_allowed to the | |
4815 | * cpuset's cpus_allowed | |
4816 | */ | |
5a16f3d3 | 4817 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4818 | goto again; |
4819 | } | |
4820 | } | |
16303ab2 | 4821 | out_free_new_mask: |
5a16f3d3 RR |
4822 | free_cpumask_var(new_mask); |
4823 | out_free_cpus_allowed: | |
4824 | free_cpumask_var(cpus_allowed); | |
4825 | out_put_task: | |
1da177e4 | 4826 | put_task_struct(p); |
1da177e4 LT |
4827 | return retval; |
4828 | } | |
4829 | ||
4830 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4831 | struct cpumask *new_mask) |
1da177e4 | 4832 | { |
96f874e2 RR |
4833 | if (len < cpumask_size()) |
4834 | cpumask_clear(new_mask); | |
4835 | else if (len > cpumask_size()) | |
4836 | len = cpumask_size(); | |
4837 | ||
1da177e4 LT |
4838 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4839 | } | |
4840 | ||
4841 | /** | |
d1ccc66d | 4842 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
4843 | * @pid: pid of the process |
4844 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4845 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
4846 | * |
4847 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 4848 | */ |
5add95d4 HC |
4849 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4850 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4851 | { |
5a16f3d3 | 4852 | cpumask_var_t new_mask; |
1da177e4 LT |
4853 | int retval; |
4854 | ||
5a16f3d3 RR |
4855 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4856 | return -ENOMEM; | |
1da177e4 | 4857 | |
5a16f3d3 RR |
4858 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4859 | if (retval == 0) | |
4860 | retval = sched_setaffinity(pid, new_mask); | |
4861 | free_cpumask_var(new_mask); | |
4862 | return retval; | |
1da177e4 LT |
4863 | } |
4864 | ||
96f874e2 | 4865 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4866 | { |
36c8b586 | 4867 | struct task_struct *p; |
31605683 | 4868 | unsigned long flags; |
1da177e4 | 4869 | int retval; |
1da177e4 | 4870 | |
23f5d142 | 4871 | rcu_read_lock(); |
1da177e4 LT |
4872 | |
4873 | retval = -ESRCH; | |
4874 | p = find_process_by_pid(pid); | |
4875 | if (!p) | |
4876 | goto out_unlock; | |
4877 | ||
e7834f8f DQ |
4878 | retval = security_task_getscheduler(p); |
4879 | if (retval) | |
4880 | goto out_unlock; | |
4881 | ||
013fdb80 | 4882 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
6acce3ef | 4883 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
013fdb80 | 4884 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4885 | |
4886 | out_unlock: | |
23f5d142 | 4887 | rcu_read_unlock(); |
1da177e4 | 4888 | |
9531b62f | 4889 | return retval; |
1da177e4 LT |
4890 | } |
4891 | ||
4892 | /** | |
d1ccc66d | 4893 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
4894 | * @pid: pid of the process |
4895 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 4896 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 4897 | * |
599b4840 ZW |
4898 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
4899 | * error code otherwise. | |
1da177e4 | 4900 | */ |
5add95d4 HC |
4901 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4902 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4903 | { |
4904 | int ret; | |
f17c8607 | 4905 | cpumask_var_t mask; |
1da177e4 | 4906 | |
84fba5ec | 4907 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4908 | return -EINVAL; |
4909 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4910 | return -EINVAL; |
4911 | ||
f17c8607 RR |
4912 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4913 | return -ENOMEM; | |
1da177e4 | 4914 | |
f17c8607 RR |
4915 | ret = sched_getaffinity(pid, mask); |
4916 | if (ret == 0) { | |
4de373a1 | 4917 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
4918 | |
4919 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4920 | ret = -EFAULT; |
4921 | else | |
cd3d8031 | 4922 | ret = retlen; |
f17c8607 RR |
4923 | } |
4924 | free_cpumask_var(mask); | |
1da177e4 | 4925 | |
f17c8607 | 4926 | return ret; |
1da177e4 LT |
4927 | } |
4928 | ||
4929 | /** | |
4930 | * sys_sched_yield - yield the current processor to other threads. | |
4931 | * | |
dd41f596 IM |
4932 | * This function yields the current CPU to other tasks. If there are no |
4933 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
4934 | * |
4935 | * Return: 0. | |
1da177e4 | 4936 | */ |
7d4dd4f1 | 4937 | static void do_sched_yield(void) |
1da177e4 | 4938 | { |
8a8c69c3 PZ |
4939 | struct rq_flags rf; |
4940 | struct rq *rq; | |
4941 | ||
246b3b33 | 4942 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 4943 | |
ae92882e | 4944 | schedstat_inc(rq->yld_count); |
4530d7ab | 4945 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4946 | |
4947 | /* | |
4948 | * Since we are going to call schedule() anyway, there's | |
4949 | * no need to preempt or enable interrupts: | |
4950 | */ | |
8a8c69c3 PZ |
4951 | preempt_disable(); |
4952 | rq_unlock(rq, &rf); | |
ba74c144 | 4953 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
4954 | |
4955 | schedule(); | |
7d4dd4f1 | 4956 | } |
1da177e4 | 4957 | |
7d4dd4f1 DB |
4958 | SYSCALL_DEFINE0(sched_yield) |
4959 | { | |
4960 | do_sched_yield(); | |
1da177e4 LT |
4961 | return 0; |
4962 | } | |
4963 | ||
35a773a0 | 4964 | #ifndef CONFIG_PREEMPT |
02b67cc3 | 4965 | int __sched _cond_resched(void) |
1da177e4 | 4966 | { |
fe32d3cd | 4967 | if (should_resched(0)) { |
a18b5d01 | 4968 | preempt_schedule_common(); |
1da177e4 LT |
4969 | return 1; |
4970 | } | |
f79c3ad6 | 4971 | rcu_all_qs(); |
1da177e4 LT |
4972 | return 0; |
4973 | } | |
02b67cc3 | 4974 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 4975 | #endif |
1da177e4 LT |
4976 | |
4977 | /* | |
613afbf8 | 4978 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4979 | * call schedule, and on return reacquire the lock. |
4980 | * | |
41a2d6cf | 4981 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4982 | * operations here to prevent schedule() from being called twice (once via |
4983 | * spin_unlock(), once by hand). | |
4984 | */ | |
613afbf8 | 4985 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4986 | { |
fe32d3cd | 4987 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
4988 | int ret = 0; |
4989 | ||
f607c668 PZ |
4990 | lockdep_assert_held(lock); |
4991 | ||
4a81e832 | 4992 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4993 | spin_unlock(lock); |
d86ee480 | 4994 | if (resched) |
a18b5d01 | 4995 | preempt_schedule_common(); |
95c354fe NP |
4996 | else |
4997 | cpu_relax(); | |
6df3cecb | 4998 | ret = 1; |
1da177e4 | 4999 | spin_lock(lock); |
1da177e4 | 5000 | } |
6df3cecb | 5001 | return ret; |
1da177e4 | 5002 | } |
613afbf8 | 5003 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5004 | |
1da177e4 LT |
5005 | /** |
5006 | * yield - yield the current processor to other threads. | |
5007 | * | |
8e3fabfd PZ |
5008 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
5009 | * | |
5010 | * The scheduler is at all times free to pick the calling task as the most | |
5011 | * eligible task to run, if removing the yield() call from your code breaks | |
5012 | * it, its already broken. | |
5013 | * | |
5014 | * Typical broken usage is: | |
5015 | * | |
5016 | * while (!event) | |
d1ccc66d | 5017 | * yield(); |
8e3fabfd PZ |
5018 | * |
5019 | * where one assumes that yield() will let 'the other' process run that will | |
5020 | * make event true. If the current task is a SCHED_FIFO task that will never | |
5021 | * happen. Never use yield() as a progress guarantee!! | |
5022 | * | |
5023 | * If you want to use yield() to wait for something, use wait_event(). | |
5024 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
5025 | * If you still want to use yield(), do not! | |
1da177e4 LT |
5026 | */ |
5027 | void __sched yield(void) | |
5028 | { | |
5029 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 5030 | do_sched_yield(); |
1da177e4 | 5031 | } |
1da177e4 LT |
5032 | EXPORT_SYMBOL(yield); |
5033 | ||
d95f4122 MG |
5034 | /** |
5035 | * yield_to - yield the current processor to another thread in | |
5036 | * your thread group, or accelerate that thread toward the | |
5037 | * processor it's on. | |
16addf95 RD |
5038 | * @p: target task |
5039 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5040 | * |
5041 | * It's the caller's job to ensure that the target task struct | |
5042 | * can't go away on us before we can do any checks. | |
5043 | * | |
e69f6186 | 5044 | * Return: |
7b270f60 PZ |
5045 | * true (>0) if we indeed boosted the target task. |
5046 | * false (0) if we failed to boost the target. | |
5047 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 5048 | */ |
fa93384f | 5049 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
5050 | { |
5051 | struct task_struct *curr = current; | |
5052 | struct rq *rq, *p_rq; | |
5053 | unsigned long flags; | |
c3c18640 | 5054 | int yielded = 0; |
d95f4122 MG |
5055 | |
5056 | local_irq_save(flags); | |
5057 | rq = this_rq(); | |
5058 | ||
5059 | again: | |
5060 | p_rq = task_rq(p); | |
7b270f60 PZ |
5061 | /* |
5062 | * If we're the only runnable task on the rq and target rq also | |
5063 | * has only one task, there's absolutely no point in yielding. | |
5064 | */ | |
5065 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
5066 | yielded = -ESRCH; | |
5067 | goto out_irq; | |
5068 | } | |
5069 | ||
d95f4122 | 5070 | double_rq_lock(rq, p_rq); |
39e24d8f | 5071 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
5072 | double_rq_unlock(rq, p_rq); |
5073 | goto again; | |
5074 | } | |
5075 | ||
5076 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 5077 | goto out_unlock; |
d95f4122 MG |
5078 | |
5079 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 5080 | goto out_unlock; |
d95f4122 MG |
5081 | |
5082 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 5083 | goto out_unlock; |
d95f4122 MG |
5084 | |
5085 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5086 | if (yielded) { |
ae92882e | 5087 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5088 | /* |
5089 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5090 | * fairness. | |
5091 | */ | |
5092 | if (preempt && rq != p_rq) | |
8875125e | 5093 | resched_curr(p_rq); |
6d1cafd8 | 5094 | } |
d95f4122 | 5095 | |
7b270f60 | 5096 | out_unlock: |
d95f4122 | 5097 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5098 | out_irq: |
d95f4122 MG |
5099 | local_irq_restore(flags); |
5100 | ||
7b270f60 | 5101 | if (yielded > 0) |
d95f4122 MG |
5102 | schedule(); |
5103 | ||
5104 | return yielded; | |
5105 | } | |
5106 | EXPORT_SYMBOL_GPL(yield_to); | |
5107 | ||
10ab5643 TH |
5108 | int io_schedule_prepare(void) |
5109 | { | |
5110 | int old_iowait = current->in_iowait; | |
5111 | ||
5112 | current->in_iowait = 1; | |
5113 | blk_schedule_flush_plug(current); | |
5114 | ||
5115 | return old_iowait; | |
5116 | } | |
5117 | ||
5118 | void io_schedule_finish(int token) | |
5119 | { | |
5120 | current->in_iowait = token; | |
5121 | } | |
5122 | ||
1da177e4 | 5123 | /* |
41a2d6cf | 5124 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5125 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5126 | */ |
1da177e4 LT |
5127 | long __sched io_schedule_timeout(long timeout) |
5128 | { | |
10ab5643 | 5129 | int token; |
1da177e4 LT |
5130 | long ret; |
5131 | ||
10ab5643 | 5132 | token = io_schedule_prepare(); |
1da177e4 | 5133 | ret = schedule_timeout(timeout); |
10ab5643 | 5134 | io_schedule_finish(token); |
9cff8ade | 5135 | |
1da177e4 LT |
5136 | return ret; |
5137 | } | |
9cff8ade | 5138 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 5139 | |
10ab5643 TH |
5140 | void io_schedule(void) |
5141 | { | |
5142 | int token; | |
5143 | ||
5144 | token = io_schedule_prepare(); | |
5145 | schedule(); | |
5146 | io_schedule_finish(token); | |
5147 | } | |
5148 | EXPORT_SYMBOL(io_schedule); | |
5149 | ||
1da177e4 LT |
5150 | /** |
5151 | * sys_sched_get_priority_max - return maximum RT priority. | |
5152 | * @policy: scheduling class. | |
5153 | * | |
e69f6186 YB |
5154 | * Return: On success, this syscall returns the maximum |
5155 | * rt_priority that can be used by a given scheduling class. | |
5156 | * On failure, a negative error code is returned. | |
1da177e4 | 5157 | */ |
5add95d4 | 5158 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5159 | { |
5160 | int ret = -EINVAL; | |
5161 | ||
5162 | switch (policy) { | |
5163 | case SCHED_FIFO: | |
5164 | case SCHED_RR: | |
5165 | ret = MAX_USER_RT_PRIO-1; | |
5166 | break; | |
aab03e05 | 5167 | case SCHED_DEADLINE: |
1da177e4 | 5168 | case SCHED_NORMAL: |
b0a9499c | 5169 | case SCHED_BATCH: |
dd41f596 | 5170 | case SCHED_IDLE: |
1da177e4 LT |
5171 | ret = 0; |
5172 | break; | |
5173 | } | |
5174 | return ret; | |
5175 | } | |
5176 | ||
5177 | /** | |
5178 | * sys_sched_get_priority_min - return minimum RT priority. | |
5179 | * @policy: scheduling class. | |
5180 | * | |
e69f6186 YB |
5181 | * Return: On success, this syscall returns the minimum |
5182 | * rt_priority that can be used by a given scheduling class. | |
5183 | * On failure, a negative error code is returned. | |
1da177e4 | 5184 | */ |
5add95d4 | 5185 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5186 | { |
5187 | int ret = -EINVAL; | |
5188 | ||
5189 | switch (policy) { | |
5190 | case SCHED_FIFO: | |
5191 | case SCHED_RR: | |
5192 | ret = 1; | |
5193 | break; | |
aab03e05 | 5194 | case SCHED_DEADLINE: |
1da177e4 | 5195 | case SCHED_NORMAL: |
b0a9499c | 5196 | case SCHED_BATCH: |
dd41f596 | 5197 | case SCHED_IDLE: |
1da177e4 LT |
5198 | ret = 0; |
5199 | } | |
5200 | return ret; | |
5201 | } | |
5202 | ||
abca5fc5 | 5203 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 5204 | { |
36c8b586 | 5205 | struct task_struct *p; |
a4ec24b4 | 5206 | unsigned int time_slice; |
eb580751 | 5207 | struct rq_flags rf; |
dba091b9 | 5208 | struct rq *rq; |
3a5c359a | 5209 | int retval; |
1da177e4 LT |
5210 | |
5211 | if (pid < 0) | |
3a5c359a | 5212 | return -EINVAL; |
1da177e4 LT |
5213 | |
5214 | retval = -ESRCH; | |
1a551ae7 | 5215 | rcu_read_lock(); |
1da177e4 LT |
5216 | p = find_process_by_pid(pid); |
5217 | if (!p) | |
5218 | goto out_unlock; | |
5219 | ||
5220 | retval = security_task_getscheduler(p); | |
5221 | if (retval) | |
5222 | goto out_unlock; | |
5223 | ||
eb580751 | 5224 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5225 | time_slice = 0; |
5226 | if (p->sched_class->get_rr_interval) | |
5227 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5228 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5229 | |
1a551ae7 | 5230 | rcu_read_unlock(); |
abca5fc5 AV |
5231 | jiffies_to_timespec64(time_slice, t); |
5232 | return 0; | |
3a5c359a | 5233 | |
1da177e4 | 5234 | out_unlock: |
1a551ae7 | 5235 | rcu_read_unlock(); |
1da177e4 LT |
5236 | return retval; |
5237 | } | |
5238 | ||
2064a5ab RD |
5239 | /** |
5240 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5241 | * @pid: pid of the process. | |
5242 | * @interval: userspace pointer to the timeslice value. | |
5243 | * | |
5244 | * this syscall writes the default timeslice value of a given process | |
5245 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5246 | * | |
5247 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5248 | * an error code. | |
5249 | */ | |
abca5fc5 | 5250 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 5251 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
5252 | { |
5253 | struct timespec64 t; | |
5254 | int retval = sched_rr_get_interval(pid, &t); | |
5255 | ||
5256 | if (retval == 0) | |
5257 | retval = put_timespec64(&t, interval); | |
5258 | ||
5259 | return retval; | |
5260 | } | |
5261 | ||
474b9c77 | 5262 | #ifdef CONFIG_COMPAT_32BIT_TIME |
abca5fc5 AV |
5263 | COMPAT_SYSCALL_DEFINE2(sched_rr_get_interval, |
5264 | compat_pid_t, pid, | |
9afc5eee | 5265 | struct old_timespec32 __user *, interval) |
abca5fc5 AV |
5266 | { |
5267 | struct timespec64 t; | |
5268 | int retval = sched_rr_get_interval(pid, &t); | |
5269 | ||
5270 | if (retval == 0) | |
9afc5eee | 5271 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
5272 | return retval; |
5273 | } | |
5274 | #endif | |
5275 | ||
82a1fcb9 | 5276 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5277 | { |
1da177e4 | 5278 | unsigned long free = 0; |
4e79752c | 5279 | int ppid; |
c930b2c0 | 5280 | |
38200502 TH |
5281 | if (!try_get_task_stack(p)) |
5282 | return; | |
20435d84 XX |
5283 | |
5284 | printk(KERN_INFO "%-15.15s %c", p->comm, task_state_to_char(p)); | |
5285 | ||
5286 | if (p->state == TASK_RUNNING) | |
3df0fc5b | 5287 | printk(KERN_CONT " running task "); |
1da177e4 | 5288 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 5289 | free = stack_not_used(p); |
1da177e4 | 5290 | #endif |
a90e984c | 5291 | ppid = 0; |
4e79752c | 5292 | rcu_read_lock(); |
a90e984c ON |
5293 | if (pid_alive(p)) |
5294 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 5295 | rcu_read_unlock(); |
3df0fc5b | 5296 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 5297 | task_pid_nr(p), ppid, |
aa47b7e0 | 5298 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 5299 | |
3d1cb205 | 5300 | print_worker_info(KERN_INFO, p); |
5fb5e6de | 5301 | show_stack(p, NULL); |
38200502 | 5302 | put_task_stack(p); |
1da177e4 | 5303 | } |
0032f4e8 | 5304 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 5305 | |
5d68cc95 PZ |
5306 | static inline bool |
5307 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
5308 | { | |
5309 | /* no filter, everything matches */ | |
5310 | if (!state_filter) | |
5311 | return true; | |
5312 | ||
5313 | /* filter, but doesn't match */ | |
5314 | if (!(p->state & state_filter)) | |
5315 | return false; | |
5316 | ||
5317 | /* | |
5318 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
5319 | * TASK_KILLABLE). | |
5320 | */ | |
5321 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
5322 | return false; | |
5323 | ||
5324 | return true; | |
5325 | } | |
5326 | ||
5327 | ||
e59e2ae2 | 5328 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5329 | { |
36c8b586 | 5330 | struct task_struct *g, *p; |
1da177e4 | 5331 | |
4bd77321 | 5332 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5333 | printk(KERN_INFO |
5334 | " task PC stack pid father\n"); | |
1da177e4 | 5335 | #else |
3df0fc5b PZ |
5336 | printk(KERN_INFO |
5337 | " task PC stack pid father\n"); | |
1da177e4 | 5338 | #endif |
510f5acc | 5339 | rcu_read_lock(); |
5d07f420 | 5340 | for_each_process_thread(g, p) { |
1da177e4 LT |
5341 | /* |
5342 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5343 | * console might take a lot of time: |
57675cb9 AR |
5344 | * Also, reset softlockup watchdogs on all CPUs, because |
5345 | * another CPU might be blocked waiting for us to process | |
5346 | * an IPI. | |
1da177e4 LT |
5347 | */ |
5348 | touch_nmi_watchdog(); | |
57675cb9 | 5349 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 5350 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 5351 | sched_show_task(p); |
5d07f420 | 5352 | } |
1da177e4 | 5353 | |
dd41f596 | 5354 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
5355 | if (!state_filter) |
5356 | sysrq_sched_debug_show(); | |
dd41f596 | 5357 | #endif |
510f5acc | 5358 | rcu_read_unlock(); |
e59e2ae2 IM |
5359 | /* |
5360 | * Only show locks if all tasks are dumped: | |
5361 | */ | |
93335a21 | 5362 | if (!state_filter) |
e59e2ae2 | 5363 | debug_show_all_locks(); |
1da177e4 LT |
5364 | } |
5365 | ||
f340c0d1 IM |
5366 | /** |
5367 | * init_idle - set up an idle thread for a given CPU | |
5368 | * @idle: task in question | |
d1ccc66d | 5369 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
5370 | * |
5371 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5372 | * flag, to make booting more robust. | |
5373 | */ | |
0db0628d | 5374 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5375 | { |
70b97a7f | 5376 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5377 | unsigned long flags; |
5378 | ||
25834c73 PZ |
5379 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5380 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 5381 | |
5e1576ed | 5382 | __sched_fork(0, idle); |
06b83b5f | 5383 | idle->state = TASK_RUNNING; |
dd41f596 | 5384 | idle->se.exec_start = sched_clock(); |
c1de45ca | 5385 | idle->flags |= PF_IDLE; |
dd41f596 | 5386 | |
e1b77c92 MR |
5387 | kasan_unpoison_task_stack(idle); |
5388 | ||
de9b8f5d PZ |
5389 | #ifdef CONFIG_SMP |
5390 | /* | |
5391 | * Its possible that init_idle() gets called multiple times on a task, | |
5392 | * in that case do_set_cpus_allowed() will not do the right thing. | |
5393 | * | |
5394 | * And since this is boot we can forgo the serialization. | |
5395 | */ | |
5396 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
5397 | #endif | |
6506cf6c PZ |
5398 | /* |
5399 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 5400 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
5401 | * lockdep check in task_group() will fail. |
5402 | * | |
5403 | * Similar case to sched_fork(). / Alternatively we could | |
5404 | * use task_rq_lock() here and obtain the other rq->lock. | |
5405 | * | |
5406 | * Silence PROVE_RCU | |
5407 | */ | |
5408 | rcu_read_lock(); | |
dd41f596 | 5409 | __set_task_cpu(idle, cpu); |
6506cf6c | 5410 | rcu_read_unlock(); |
1da177e4 | 5411 | |
1da177e4 | 5412 | rq->curr = rq->idle = idle; |
da0c1e65 | 5413 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 5414 | #ifdef CONFIG_SMP |
3ca7a440 | 5415 | idle->on_cpu = 1; |
4866cde0 | 5416 | #endif |
25834c73 PZ |
5417 | raw_spin_unlock(&rq->lock); |
5418 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
5419 | |
5420 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 5421 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 5422 | |
dd41f596 IM |
5423 | /* |
5424 | * The idle tasks have their own, simple scheduling class: | |
5425 | */ | |
5426 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5427 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 5428 | vtime_init_idle(idle, cpu); |
de9b8f5d | 5429 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
5430 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
5431 | #endif | |
19978ca6 IM |
5432 | } |
5433 | ||
e1d4eeec NP |
5434 | #ifdef CONFIG_SMP |
5435 | ||
f82f8042 JL |
5436 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
5437 | const struct cpumask *trial) | |
5438 | { | |
06a76fe0 | 5439 | int ret = 1; |
f82f8042 | 5440 | |
bb2bc55a MG |
5441 | if (!cpumask_weight(cur)) |
5442 | return ret; | |
5443 | ||
06a76fe0 | 5444 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
5445 | |
5446 | return ret; | |
5447 | } | |
5448 | ||
7f51412a JL |
5449 | int task_can_attach(struct task_struct *p, |
5450 | const struct cpumask *cs_cpus_allowed) | |
5451 | { | |
5452 | int ret = 0; | |
5453 | ||
5454 | /* | |
5455 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 5456 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
5457 | * affinity and isolating such threads by their set of |
5458 | * allowed nodes is unnecessary. Thus, cpusets are not | |
5459 | * applicable for such threads. This prevents checking for | |
5460 | * success of set_cpus_allowed_ptr() on all attached tasks | |
5461 | * before cpus_allowed may be changed. | |
5462 | */ | |
5463 | if (p->flags & PF_NO_SETAFFINITY) { | |
5464 | ret = -EINVAL; | |
5465 | goto out; | |
5466 | } | |
5467 | ||
7f51412a | 5468 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
5469 | cs_cpus_allowed)) |
5470 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 5471 | |
7f51412a JL |
5472 | out: |
5473 | return ret; | |
5474 | } | |
5475 | ||
f2cb1360 | 5476 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 5477 | |
e6628d5b MG |
5478 | #ifdef CONFIG_NUMA_BALANCING |
5479 | /* Migrate current task p to target_cpu */ | |
5480 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
5481 | { | |
5482 | struct migration_arg arg = { p, target_cpu }; | |
5483 | int curr_cpu = task_cpu(p); | |
5484 | ||
5485 | if (curr_cpu == target_cpu) | |
5486 | return 0; | |
5487 | ||
0c98d344 | 5488 | if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed)) |
e6628d5b MG |
5489 | return -EINVAL; |
5490 | ||
5491 | /* TODO: This is not properly updating schedstats */ | |
5492 | ||
286549dc | 5493 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
5494 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
5495 | } | |
0ec8aa00 PZ |
5496 | |
5497 | /* | |
5498 | * Requeue a task on a given node and accurately track the number of NUMA | |
5499 | * tasks on the runqueues | |
5500 | */ | |
5501 | void sched_setnuma(struct task_struct *p, int nid) | |
5502 | { | |
da0c1e65 | 5503 | bool queued, running; |
eb580751 PZ |
5504 | struct rq_flags rf; |
5505 | struct rq *rq; | |
0ec8aa00 | 5506 | |
eb580751 | 5507 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 5508 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
5509 | running = task_current(rq, p); |
5510 | ||
da0c1e65 | 5511 | if (queued) |
1de64443 | 5512 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 5513 | if (running) |
f3cd1c4e | 5514 | put_prev_task(rq, p); |
0ec8aa00 PZ |
5515 | |
5516 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 5517 | |
da0c1e65 | 5518 | if (queued) |
7134b3e9 | 5519 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 5520 | if (running) |
b2bf6c31 | 5521 | set_curr_task(rq, p); |
eb580751 | 5522 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 5523 | } |
5cc389bc | 5524 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 5525 | |
1da177e4 | 5526 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5527 | /* |
d1ccc66d | 5528 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 5529 | * offline. |
054b9108 | 5530 | */ |
48c5ccae | 5531 | void idle_task_exit(void) |
1da177e4 | 5532 | { |
48c5ccae | 5533 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5534 | |
48c5ccae | 5535 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5536 | |
a53efe5f | 5537 | if (mm != &init_mm) { |
252d2a41 | 5538 | switch_mm(mm, &init_mm, current); |
3eda69c9 | 5539 | current->active_mm = &init_mm; |
a53efe5f MS |
5540 | finish_arch_post_lock_switch(); |
5541 | } | |
48c5ccae | 5542 | mmdrop(mm); |
1da177e4 LT |
5543 | } |
5544 | ||
5545 | /* | |
5d180232 PZ |
5546 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
5547 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
5548 | * nr_active count is stable. We need to take the teardown thread which |
5549 | * is calling this into account, so we hand in adjust = 1 to the load | |
5550 | * calculation. | |
5d180232 PZ |
5551 | * |
5552 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 5553 | */ |
5d180232 | 5554 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 5555 | { |
d60585c5 | 5556 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
5557 | if (delta) |
5558 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
5559 | } |
5560 | ||
3f1d2a31 PZ |
5561 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
5562 | { | |
5563 | } | |
5564 | ||
5565 | static const struct sched_class fake_sched_class = { | |
5566 | .put_prev_task = put_prev_task_fake, | |
5567 | }; | |
5568 | ||
5569 | static struct task_struct fake_task = { | |
5570 | /* | |
5571 | * Avoid pull_{rt,dl}_task() | |
5572 | */ | |
5573 | .prio = MAX_PRIO + 1, | |
5574 | .sched_class = &fake_sched_class, | |
5575 | }; | |
5576 | ||
48f24c4d | 5577 | /* |
48c5ccae PZ |
5578 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5579 | * try_to_wake_up()->select_task_rq(). | |
5580 | * | |
5581 | * Called with rq->lock held even though we'er in stop_machine() and | |
5582 | * there's no concurrency possible, we hold the required locks anyway | |
5583 | * because of lock validation efforts. | |
1da177e4 | 5584 | */ |
8a8c69c3 | 5585 | static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf) |
1da177e4 | 5586 | { |
5e16bbc2 | 5587 | struct rq *rq = dead_rq; |
48c5ccae | 5588 | struct task_struct *next, *stop = rq->stop; |
8a8c69c3 | 5589 | struct rq_flags orf = *rf; |
48c5ccae | 5590 | int dest_cpu; |
1da177e4 LT |
5591 | |
5592 | /* | |
48c5ccae PZ |
5593 | * Fudge the rq selection such that the below task selection loop |
5594 | * doesn't get stuck on the currently eligible stop task. | |
5595 | * | |
5596 | * We're currently inside stop_machine() and the rq is either stuck | |
5597 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5598 | * either way we should never end up calling schedule() until we're | |
5599 | * done here. | |
1da177e4 | 5600 | */ |
48c5ccae | 5601 | rq->stop = NULL; |
48f24c4d | 5602 | |
77bd3970 FW |
5603 | /* |
5604 | * put_prev_task() and pick_next_task() sched | |
5605 | * class method both need to have an up-to-date | |
5606 | * value of rq->clock[_task] | |
5607 | */ | |
5608 | update_rq_clock(rq); | |
5609 | ||
5e16bbc2 | 5610 | for (;;) { |
48c5ccae PZ |
5611 | /* |
5612 | * There's this thread running, bail when that's the only | |
d1ccc66d | 5613 | * remaining thread: |
48c5ccae PZ |
5614 | */ |
5615 | if (rq->nr_running == 1) | |
dd41f596 | 5616 | break; |
48c5ccae | 5617 | |
cbce1a68 | 5618 | /* |
d1ccc66d | 5619 | * pick_next_task() assumes pinned rq->lock: |
cbce1a68 | 5620 | */ |
8a8c69c3 | 5621 | next = pick_next_task(rq, &fake_task, rf); |
48c5ccae | 5622 | BUG_ON(!next); |
5b713a3d | 5623 | put_prev_task(rq, next); |
e692ab53 | 5624 | |
5473e0cc WL |
5625 | /* |
5626 | * Rules for changing task_struct::cpus_allowed are holding | |
5627 | * both pi_lock and rq->lock, such that holding either | |
5628 | * stabilizes the mask. | |
5629 | * | |
5630 | * Drop rq->lock is not quite as disastrous as it usually is | |
5631 | * because !cpu_active at this point, which means load-balance | |
5632 | * will not interfere. Also, stop-machine. | |
5633 | */ | |
8a8c69c3 | 5634 | rq_unlock(rq, rf); |
5473e0cc | 5635 | raw_spin_lock(&next->pi_lock); |
8a8c69c3 | 5636 | rq_relock(rq, rf); |
5473e0cc WL |
5637 | |
5638 | /* | |
5639 | * Since we're inside stop-machine, _nothing_ should have | |
5640 | * changed the task, WARN if weird stuff happened, because in | |
5641 | * that case the above rq->lock drop is a fail too. | |
5642 | */ | |
5643 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
5644 | raw_spin_unlock(&next->pi_lock); | |
5645 | continue; | |
5646 | } | |
5647 | ||
48c5ccae | 5648 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 5649 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
8a8c69c3 | 5650 | rq = __migrate_task(rq, rf, next, dest_cpu); |
5e16bbc2 | 5651 | if (rq != dead_rq) { |
8a8c69c3 | 5652 | rq_unlock(rq, rf); |
5e16bbc2 | 5653 | rq = dead_rq; |
8a8c69c3 PZ |
5654 | *rf = orf; |
5655 | rq_relock(rq, rf); | |
5e16bbc2 | 5656 | } |
5473e0cc | 5657 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 5658 | } |
dce48a84 | 5659 | |
48c5ccae | 5660 | rq->stop = stop; |
dce48a84 | 5661 | } |
1da177e4 LT |
5662 | #endif /* CONFIG_HOTPLUG_CPU */ |
5663 | ||
f2cb1360 | 5664 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
5665 | { |
5666 | if (!rq->online) { | |
5667 | const struct sched_class *class; | |
5668 | ||
c6c4927b | 5669 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5670 | rq->online = 1; |
5671 | ||
5672 | for_each_class(class) { | |
5673 | if (class->rq_online) | |
5674 | class->rq_online(rq); | |
5675 | } | |
5676 | } | |
5677 | } | |
5678 | ||
f2cb1360 | 5679 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
5680 | { |
5681 | if (rq->online) { | |
5682 | const struct sched_class *class; | |
5683 | ||
5684 | for_each_class(class) { | |
5685 | if (class->rq_offline) | |
5686 | class->rq_offline(rq); | |
5687 | } | |
5688 | ||
c6c4927b | 5689 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5690 | rq->online = 0; |
5691 | } | |
5692 | } | |
5693 | ||
d1ccc66d IM |
5694 | /* |
5695 | * used to mark begin/end of suspend/resume: | |
5696 | */ | |
5697 | static int num_cpus_frozen; | |
d35be8ba | 5698 | |
1da177e4 | 5699 | /* |
3a101d05 TH |
5700 | * Update cpusets according to cpu_active mask. If cpusets are |
5701 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
5702 | * around partition_sched_domains(). | |
d35be8ba SB |
5703 | * |
5704 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
5705 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 5706 | */ |
40190a78 | 5707 | static void cpuset_cpu_active(void) |
e761b772 | 5708 | { |
40190a78 | 5709 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
5710 | /* |
5711 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
5712 | * resume sequence. As long as this is not the last online | |
5713 | * operation in the resume sequence, just build a single sched | |
5714 | * domain, ignoring cpusets. | |
5715 | */ | |
50e76632 PZ |
5716 | partition_sched_domains(1, NULL, NULL); |
5717 | if (--num_cpus_frozen) | |
135fb3e1 | 5718 | return; |
d35be8ba SB |
5719 | /* |
5720 | * This is the last CPU online operation. So fall through and | |
5721 | * restore the original sched domains by considering the | |
5722 | * cpuset configurations. | |
5723 | */ | |
50e76632 | 5724 | cpuset_force_rebuild(); |
3a101d05 | 5725 | } |
30e03acd | 5726 | cpuset_update_active_cpus(); |
3a101d05 | 5727 | } |
e761b772 | 5728 | |
40190a78 | 5729 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 5730 | { |
40190a78 | 5731 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 5732 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 5733 | return -EBUSY; |
30e03acd | 5734 | cpuset_update_active_cpus(); |
135fb3e1 | 5735 | } else { |
d35be8ba SB |
5736 | num_cpus_frozen++; |
5737 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 5738 | } |
135fb3e1 | 5739 | return 0; |
e761b772 | 5740 | } |
e761b772 | 5741 | |
40190a78 | 5742 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 5743 | { |
7d976699 | 5744 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 5745 | struct rq_flags rf; |
7d976699 | 5746 | |
ba2591a5 PZ |
5747 | #ifdef CONFIG_SCHED_SMT |
5748 | /* | |
5749 | * The sched_smt_present static key needs to be evaluated on every | |
5750 | * hotplug event because at boot time SMT might be disabled when | |
5751 | * the number of booted CPUs is limited. | |
5752 | * | |
5753 | * If then later a sibling gets hotplugged, then the key would stay | |
5754 | * off and SMT scheduling would never be functional. | |
5755 | */ | |
5756 | if (cpumask_weight(cpu_smt_mask(cpu)) > 1) | |
5757 | static_branch_enable_cpuslocked(&sched_smt_present); | |
5758 | #endif | |
40190a78 | 5759 | set_cpu_active(cpu, true); |
135fb3e1 | 5760 | |
40190a78 | 5761 | if (sched_smp_initialized) { |
135fb3e1 | 5762 | sched_domains_numa_masks_set(cpu); |
40190a78 | 5763 | cpuset_cpu_active(); |
e761b772 | 5764 | } |
7d976699 TG |
5765 | |
5766 | /* | |
5767 | * Put the rq online, if not already. This happens: | |
5768 | * | |
5769 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 5770 | * after all CPUs have been brought up. |
7d976699 TG |
5771 | * |
5772 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
5773 | * domains. | |
5774 | */ | |
8a8c69c3 | 5775 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
5776 | if (rq->rd) { |
5777 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5778 | set_rq_online(rq); | |
5779 | } | |
8a8c69c3 | 5780 | rq_unlock_irqrestore(rq, &rf); |
7d976699 TG |
5781 | |
5782 | update_max_interval(); | |
5783 | ||
40190a78 | 5784 | return 0; |
135fb3e1 TG |
5785 | } |
5786 | ||
40190a78 | 5787 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 5788 | { |
135fb3e1 TG |
5789 | int ret; |
5790 | ||
40190a78 | 5791 | set_cpu_active(cpu, false); |
b2454caa PZ |
5792 | /* |
5793 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
5794 | * users of this state to go away such that all new such users will | |
5795 | * observe it. | |
5796 | * | |
b2454caa PZ |
5797 | * Do sync before park smpboot threads to take care the rcu boost case. |
5798 | */ | |
d7d34d5e | 5799 | synchronize_rcu_mult(call_rcu, call_rcu_sched); |
40190a78 TG |
5800 | |
5801 | if (!sched_smp_initialized) | |
5802 | return 0; | |
5803 | ||
5804 | ret = cpuset_cpu_inactive(cpu); | |
5805 | if (ret) { | |
5806 | set_cpu_active(cpu, true); | |
5807 | return ret; | |
135fb3e1 | 5808 | } |
40190a78 TG |
5809 | sched_domains_numa_masks_clear(cpu); |
5810 | return 0; | |
135fb3e1 TG |
5811 | } |
5812 | ||
94baf7a5 TG |
5813 | static void sched_rq_cpu_starting(unsigned int cpu) |
5814 | { | |
5815 | struct rq *rq = cpu_rq(cpu); | |
5816 | ||
5817 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
5818 | update_max_interval(); |
5819 | } | |
5820 | ||
135fb3e1 TG |
5821 | int sched_cpu_starting(unsigned int cpu) |
5822 | { | |
94baf7a5 | 5823 | sched_rq_cpu_starting(cpu); |
d84b3131 | 5824 | sched_tick_start(cpu); |
135fb3e1 | 5825 | return 0; |
e761b772 | 5826 | } |
e761b772 | 5827 | |
f2785ddb TG |
5828 | #ifdef CONFIG_HOTPLUG_CPU |
5829 | int sched_cpu_dying(unsigned int cpu) | |
5830 | { | |
5831 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 5832 | struct rq_flags rf; |
f2785ddb TG |
5833 | |
5834 | /* Handle pending wakeups and then migrate everything off */ | |
5835 | sched_ttwu_pending(); | |
d84b3131 | 5836 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
5837 | |
5838 | rq_lock_irqsave(rq, &rf); | |
f2785ddb TG |
5839 | if (rq->rd) { |
5840 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
5841 | set_rq_offline(rq); | |
5842 | } | |
8a8c69c3 | 5843 | migrate_tasks(rq, &rf); |
f2785ddb | 5844 | BUG_ON(rq->nr_running != 1); |
8a8c69c3 PZ |
5845 | rq_unlock_irqrestore(rq, &rf); |
5846 | ||
f2785ddb TG |
5847 | calc_load_migrate(rq); |
5848 | update_max_interval(); | |
00357f5e | 5849 | nohz_balance_exit_idle(rq); |
e5ef27d0 | 5850 | hrtick_clear(rq); |
f2785ddb TG |
5851 | return 0; |
5852 | } | |
5853 | #endif | |
5854 | ||
1da177e4 LT |
5855 | void __init sched_init_smp(void) |
5856 | { | |
cb83b629 PZ |
5857 | sched_init_numa(); |
5858 | ||
6acce3ef PZ |
5859 | /* |
5860 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 5861 | * CPU masks are stable and all blatant races in the below code cannot |
40fa3780 VS |
5862 | * happen. The hotplug lock is nevertheless taken to satisfy lockdep, |
5863 | * but there won't be any contention on it. | |
6acce3ef | 5864 | */ |
40fa3780 | 5865 | cpus_read_lock(); |
712555ee | 5866 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 5867 | sched_init_domains(cpu_active_mask); |
712555ee | 5868 | mutex_unlock(&sched_domains_mutex); |
40fa3780 | 5869 | cpus_read_unlock(); |
e761b772 | 5870 | |
5c1e1767 | 5871 | /* Move init over to a non-isolated CPU */ |
edb93821 | 5872 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 5873 | BUG(); |
19978ca6 | 5874 | sched_init_granularity(); |
4212823f | 5875 | |
0e3900e6 | 5876 | init_sched_rt_class(); |
1baca4ce | 5877 | init_sched_dl_class(); |
1b568f0a | 5878 | |
e26fbffd | 5879 | sched_smp_initialized = true; |
1da177e4 | 5880 | } |
e26fbffd TG |
5881 | |
5882 | static int __init migration_init(void) | |
5883 | { | |
94baf7a5 | 5884 | sched_rq_cpu_starting(smp_processor_id()); |
e26fbffd | 5885 | return 0; |
1da177e4 | 5886 | } |
e26fbffd TG |
5887 | early_initcall(migration_init); |
5888 | ||
1da177e4 LT |
5889 | #else |
5890 | void __init sched_init_smp(void) | |
5891 | { | |
19978ca6 | 5892 | sched_init_granularity(); |
1da177e4 LT |
5893 | } |
5894 | #endif /* CONFIG_SMP */ | |
5895 | ||
5896 | int in_sched_functions(unsigned long addr) | |
5897 | { | |
1da177e4 LT |
5898 | return in_lock_functions(addr) || |
5899 | (addr >= (unsigned long)__sched_text_start | |
5900 | && addr < (unsigned long)__sched_text_end); | |
5901 | } | |
5902 | ||
029632fb | 5903 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
5904 | /* |
5905 | * Default task group. | |
5906 | * Every task in system belongs to this group at bootup. | |
5907 | */ | |
029632fb | 5908 | struct task_group root_task_group; |
35cf4e50 | 5909 | LIST_HEAD(task_groups); |
b0367629 WL |
5910 | |
5911 | /* Cacheline aligned slab cache for task_group */ | |
5912 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 5913 | #endif |
6f505b16 | 5914 | |
e6252c3e | 5915 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 5916 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 5917 | |
1da177e4 LT |
5918 | void __init sched_init(void) |
5919 | { | |
dd41f596 | 5920 | int i, j; |
434d53b0 MT |
5921 | unsigned long alloc_size = 0, ptr; |
5922 | ||
5822a454 | 5923 | wait_bit_init(); |
9dcb8b68 | 5924 | |
434d53b0 MT |
5925 | #ifdef CONFIG_FAIR_GROUP_SCHED |
5926 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5927 | #endif | |
5928 | #ifdef CONFIG_RT_GROUP_SCHED | |
5929 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
5930 | #endif | |
434d53b0 | 5931 | if (alloc_size) { |
36b7b6d4 | 5932 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
5933 | |
5934 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 5935 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
5936 | ptr += nr_cpu_ids * sizeof(void **); |
5937 | ||
07e06b01 | 5938 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 5939 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 5940 | |
6d6bc0ad | 5941 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 5942 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 5943 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
5944 | ptr += nr_cpu_ids * sizeof(void **); |
5945 | ||
07e06b01 | 5946 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
5947 | ptr += nr_cpu_ids * sizeof(void **); |
5948 | ||
6d6bc0ad | 5949 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 5950 | } |
df7c8e84 | 5951 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
5952 | for_each_possible_cpu(i) { |
5953 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
5954 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
5955 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
5956 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 5957 | } |
b74e6278 | 5958 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 5959 | |
d1ccc66d IM |
5960 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
5961 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 5962 | |
57d885fe GH |
5963 | #ifdef CONFIG_SMP |
5964 | init_defrootdomain(); | |
5965 | #endif | |
5966 | ||
d0b27fa7 | 5967 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 5968 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 5969 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 5970 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 5971 | |
7c941438 | 5972 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
5973 | task_group_cache = KMEM_CACHE(task_group, 0); |
5974 | ||
07e06b01 YZ |
5975 | list_add(&root_task_group.list, &task_groups); |
5976 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 5977 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 5978 | autogroup_init(&init_task); |
7c941438 | 5979 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 5980 | |
0a945022 | 5981 | for_each_possible_cpu(i) { |
70b97a7f | 5982 | struct rq *rq; |
1da177e4 LT |
5983 | |
5984 | rq = cpu_rq(i); | |
05fa785c | 5985 | raw_spin_lock_init(&rq->lock); |
7897986b | 5986 | rq->nr_running = 0; |
dce48a84 TG |
5987 | rq->calc_load_active = 0; |
5988 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 5989 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
5990 | init_rt_rq(&rq->rt); |
5991 | init_dl_rq(&rq->dl); | |
dd41f596 | 5992 | #ifdef CONFIG_FAIR_GROUP_SCHED |
029632fb | 5993 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6f505b16 | 5994 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 5995 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 5996 | /* |
d1ccc66d | 5997 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
5998 | * |
5999 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
6000 | * gets 100% of the CPU resources in the system. This overall |
6001 | * system CPU resource is divided among the tasks of | |
07e06b01 | 6002 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6003 | * based on each entity's (task or task-group's) weight |
6004 | * (se->load.weight). | |
6005 | * | |
07e06b01 | 6006 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 6007 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 6008 | * then A0's share of the CPU resource is: |
354d60c2 | 6009 | * |
0d905bca | 6010 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6011 | * |
07e06b01 YZ |
6012 | * We achieve this by letting root_task_group's tasks sit |
6013 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6014 | */ |
ab84d31e | 6015 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
07e06b01 | 6016 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6017 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6018 | ||
6019 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6020 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6021 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6022 | #endif |
1da177e4 | 6023 | |
dd41f596 IM |
6024 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6025 | rq->cpu_load[j] = 0; | |
fdf3e95d | 6026 | |
1da177e4 | 6027 | #ifdef CONFIG_SMP |
41c7ce9a | 6028 | rq->sd = NULL; |
57d885fe | 6029 | rq->rd = NULL; |
ca6d75e6 | 6030 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 6031 | rq->balance_callback = NULL; |
1da177e4 | 6032 | rq->active_balance = 0; |
dd41f596 | 6033 | rq->next_balance = jiffies; |
1da177e4 | 6034 | rq->push_cpu = 0; |
0a2966b4 | 6035 | rq->cpu = i; |
1f11eb6a | 6036 | rq->online = 0; |
eae0c9df MG |
6037 | rq->idle_stamp = 0; |
6038 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6039 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6040 | |
6041 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6042 | ||
dc938520 | 6043 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6044 | #ifdef CONFIG_NO_HZ_COMMON |
9fd81dd5 | 6045 | rq->last_load_update_tick = jiffies; |
e022e0d3 | 6046 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 6047 | atomic_set(&rq->nohz_flags, 0); |
83cd4fe2 | 6048 | #endif |
9fd81dd5 | 6049 | #endif /* CONFIG_SMP */ |
77a021be | 6050 | hrtick_rq_init(rq); |
1da177e4 | 6051 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6052 | } |
6053 | ||
9059393e | 6054 | set_load_weight(&init_task, false); |
b50f60ce | 6055 | |
1da177e4 LT |
6056 | /* |
6057 | * The boot idle thread does lazy MMU switching as well: | |
6058 | */ | |
f1f10076 | 6059 | mmgrab(&init_mm); |
1da177e4 LT |
6060 | enter_lazy_tlb(&init_mm, current); |
6061 | ||
6062 | /* | |
6063 | * Make us the idle thread. Technically, schedule() should not be | |
6064 | * called from this thread, however somewhere below it might be, | |
6065 | * but because we are the idle thread, we just pick up running again | |
6066 | * when this runqueue becomes "idle". | |
6067 | */ | |
6068 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6069 | |
6070 | calc_load_update = jiffies + LOAD_FREQ; | |
6071 | ||
bf4d83f6 | 6072 | #ifdef CONFIG_SMP |
29d5e047 | 6073 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6074 | #endif |
6075 | init_sched_fair_class(); | |
6a7b3dc3 | 6076 | |
4698f88c JP |
6077 | init_schedstats(); |
6078 | ||
eb414681 JW |
6079 | psi_init(); |
6080 | ||
6892b75e | 6081 | scheduler_running = 1; |
1da177e4 LT |
6082 | } |
6083 | ||
d902db1e | 6084 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6085 | static inline int preempt_count_equals(int preempt_offset) |
6086 | { | |
da7142e2 | 6087 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 6088 | |
4ba8216c | 6089 | return (nested == preempt_offset); |
e4aafea2 FW |
6090 | } |
6091 | ||
d894837f | 6092 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6093 | { |
8eb23b9f PZ |
6094 | /* |
6095 | * Blocking primitives will set (and therefore destroy) current->state, | |
6096 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
6097 | * otherwise we will destroy state. | |
6098 | */ | |
00845eb9 | 6099 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
6100 | "do not call blocking ops when !TASK_RUNNING; " |
6101 | "state=%lx set at [<%p>] %pS\n", | |
6102 | current->state, | |
6103 | (void *)current->task_state_change, | |
00845eb9 | 6104 | (void *)current->task_state_change); |
8eb23b9f | 6105 | |
3427445a PZ |
6106 | ___might_sleep(file, line, preempt_offset); |
6107 | } | |
6108 | EXPORT_SYMBOL(__might_sleep); | |
6109 | ||
6110 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 6111 | { |
d1ccc66d IM |
6112 | /* Ratelimiting timestamp: */ |
6113 | static unsigned long prev_jiffy; | |
6114 | ||
d1c6d149 | 6115 | unsigned long preempt_disable_ip; |
1da177e4 | 6116 | |
d1ccc66d IM |
6117 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
6118 | rcu_sleep_check(); | |
6119 | ||
db273be2 TG |
6120 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
6121 | !is_idle_task(current)) || | |
1c3c5eab TG |
6122 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
6123 | oops_in_progress) | |
aef745fc | 6124 | return; |
1c3c5eab | 6125 | |
aef745fc IM |
6126 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
6127 | return; | |
6128 | prev_jiffy = jiffies; | |
6129 | ||
d1ccc66d | 6130 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
6131 | preempt_disable_ip = get_preempt_disable_ip(current); |
6132 | ||
3df0fc5b PZ |
6133 | printk(KERN_ERR |
6134 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6135 | file, line); | |
6136 | printk(KERN_ERR | |
6137 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6138 | in_atomic(), irqs_disabled(), | |
6139 | current->pid, current->comm); | |
aef745fc | 6140 | |
a8b686b3 ES |
6141 | if (task_stack_end_corrupted(current)) |
6142 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
6143 | ||
aef745fc IM |
6144 | debug_show_held_locks(current); |
6145 | if (irqs_disabled()) | |
6146 | print_irqtrace_events(current); | |
d1c6d149 VN |
6147 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
6148 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 6149 | pr_err("Preemption disabled at:"); |
d1c6d149 | 6150 | print_ip_sym(preempt_disable_ip); |
8f47b187 TG |
6151 | pr_cont("\n"); |
6152 | } | |
aef745fc | 6153 | dump_stack(); |
f0b22e39 | 6154 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 6155 | } |
3427445a | 6156 | EXPORT_SYMBOL(___might_sleep); |
1da177e4 LT |
6157 | #endif |
6158 | ||
6159 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 6160 | void normalize_rt_tasks(void) |
3a5e4dc1 | 6161 | { |
dbc7f069 | 6162 | struct task_struct *g, *p; |
d50dde5a DF |
6163 | struct sched_attr attr = { |
6164 | .sched_policy = SCHED_NORMAL, | |
6165 | }; | |
1da177e4 | 6166 | |
3472eaa1 | 6167 | read_lock(&tasklist_lock); |
5d07f420 | 6168 | for_each_process_thread(g, p) { |
178be793 IM |
6169 | /* |
6170 | * Only normalize user tasks: | |
6171 | */ | |
3472eaa1 | 6172 | if (p->flags & PF_KTHREAD) |
178be793 IM |
6173 | continue; |
6174 | ||
4fa8d299 JP |
6175 | p->se.exec_start = 0; |
6176 | schedstat_set(p->se.statistics.wait_start, 0); | |
6177 | schedstat_set(p->se.statistics.sleep_start, 0); | |
6178 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 6179 | |
aab03e05 | 6180 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
6181 | /* |
6182 | * Renice negative nice level userspace | |
6183 | * tasks back to 0: | |
6184 | */ | |
3472eaa1 | 6185 | if (task_nice(p) < 0) |
dd41f596 | 6186 | set_user_nice(p, 0); |
1da177e4 | 6187 | continue; |
dd41f596 | 6188 | } |
1da177e4 | 6189 | |
dbc7f069 | 6190 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 6191 | } |
3472eaa1 | 6192 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6193 | } |
6194 | ||
6195 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6196 | |
67fc4e0c | 6197 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6198 | /* |
67fc4e0c | 6199 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6200 | * |
6201 | * They can only be called when the whole system has been | |
6202 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6203 | * activity can take place. Using them for anything else would | |
6204 | * be a serious bug, and as a result, they aren't even visible | |
6205 | * under any other configuration. | |
6206 | */ | |
6207 | ||
6208 | /** | |
d1ccc66d | 6209 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
6210 | * @cpu: the processor in question. |
6211 | * | |
6212 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
6213 | * |
6214 | * Return: The current task for @cpu. | |
1df5c10a | 6215 | */ |
36c8b586 | 6216 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6217 | { |
6218 | return cpu_curr(cpu); | |
6219 | } | |
6220 | ||
67fc4e0c JW |
6221 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6222 | ||
6223 | #ifdef CONFIG_IA64 | |
1df5c10a | 6224 | /** |
d1ccc66d | 6225 | * set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
6226 | * @cpu: the processor in question. |
6227 | * @p: the task pointer to set. | |
6228 | * | |
6229 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 6230 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 6231 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
6232 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6233 | * and caller must save the original value of the current task (see | |
6234 | * curr_task() above) and restore that value before reenabling interrupts and | |
6235 | * re-starting the system. | |
6236 | * | |
6237 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6238 | */ | |
a458ae2e | 6239 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6240 | { |
6241 | cpu_curr(cpu) = p; | |
6242 | } | |
6243 | ||
6244 | #endif | |
29f59db3 | 6245 | |
7c941438 | 6246 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6247 | /* task_group_lock serializes the addition/removal of task groups */ |
6248 | static DEFINE_SPINLOCK(task_group_lock); | |
6249 | ||
2f5177f0 | 6250 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
6251 | { |
6252 | free_fair_sched_group(tg); | |
6253 | free_rt_sched_group(tg); | |
e9aa1dd1 | 6254 | autogroup_free(tg); |
b0367629 | 6255 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
6256 | } |
6257 | ||
6258 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 6259 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
6260 | { |
6261 | struct task_group *tg; | |
bccbe08a | 6262 | |
b0367629 | 6263 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
6264 | if (!tg) |
6265 | return ERR_PTR(-ENOMEM); | |
6266 | ||
ec7dc8ac | 6267 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
6268 | goto err; |
6269 | ||
ec7dc8ac | 6270 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
6271 | goto err; |
6272 | ||
ace783b9 LZ |
6273 | return tg; |
6274 | ||
6275 | err: | |
2f5177f0 | 6276 | sched_free_group(tg); |
ace783b9 LZ |
6277 | return ERR_PTR(-ENOMEM); |
6278 | } | |
6279 | ||
6280 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
6281 | { | |
6282 | unsigned long flags; | |
6283 | ||
8ed36996 | 6284 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 6285 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 6286 | |
d1ccc66d IM |
6287 | /* Root should already exist: */ |
6288 | WARN_ON(!parent); | |
f473aa5e PZ |
6289 | |
6290 | tg->parent = parent; | |
f473aa5e | 6291 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 6292 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 6293 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
6294 | |
6295 | online_fair_sched_group(tg); | |
29f59db3 SV |
6296 | } |
6297 | ||
9b5b7751 | 6298 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 6299 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 6300 | { |
d1ccc66d | 6301 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 6302 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
6303 | } |
6304 | ||
4cf86d77 | 6305 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 6306 | { |
d1ccc66d | 6307 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 6308 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
6309 | } |
6310 | ||
6311 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 6312 | { |
8ed36996 | 6313 | unsigned long flags; |
29f59db3 | 6314 | |
d1ccc66d | 6315 | /* End participation in shares distribution: */ |
6fe1f348 | 6316 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
6317 | |
6318 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 6319 | list_del_rcu(&tg->list); |
f473aa5e | 6320 | list_del_rcu(&tg->siblings); |
8ed36996 | 6321 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
6322 | } |
6323 | ||
ea86cb4b | 6324 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 6325 | { |
8323f26c | 6326 | struct task_group *tg; |
29f59db3 | 6327 | |
f7b8a47d KT |
6328 | /* |
6329 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
6330 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
6331 | * to prevent lockdep warnings. | |
6332 | */ | |
6333 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
6334 | struct task_group, css); |
6335 | tg = autogroup_task_group(tsk, tg); | |
6336 | tsk->sched_task_group = tg; | |
6337 | ||
810b3817 | 6338 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
6339 | if (tsk->sched_class->task_change_group) |
6340 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 6341 | else |
810b3817 | 6342 | #endif |
b2b5ce02 | 6343 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
6344 | } |
6345 | ||
6346 | /* | |
6347 | * Change task's runqueue when it moves between groups. | |
6348 | * | |
6349 | * The caller of this function should have put the task in its new group by | |
6350 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
6351 | * its new group. | |
6352 | */ | |
6353 | void sched_move_task(struct task_struct *tsk) | |
6354 | { | |
7a57f32a PZ |
6355 | int queued, running, queue_flags = |
6356 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
6357 | struct rq_flags rf; |
6358 | struct rq *rq; | |
6359 | ||
6360 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 6361 | update_rq_clock(rq); |
ea86cb4b VG |
6362 | |
6363 | running = task_current(rq, tsk); | |
6364 | queued = task_on_rq_queued(tsk); | |
6365 | ||
6366 | if (queued) | |
7a57f32a | 6367 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 6368 | if (running) |
ea86cb4b VG |
6369 | put_prev_task(rq, tsk); |
6370 | ||
6371 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 6372 | |
da0c1e65 | 6373 | if (queued) |
7a57f32a | 6374 | enqueue_task(rq, tsk, queue_flags); |
bb3bac2c | 6375 | if (running) |
b2bf6c31 | 6376 | set_curr_task(rq, tsk); |
29f59db3 | 6377 | |
eb580751 | 6378 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 6379 | } |
68318b8e | 6380 | |
a7c6d554 | 6381 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 6382 | { |
a7c6d554 | 6383 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
6384 | } |
6385 | ||
eb95419b TH |
6386 | static struct cgroup_subsys_state * |
6387 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 6388 | { |
eb95419b TH |
6389 | struct task_group *parent = css_tg(parent_css); |
6390 | struct task_group *tg; | |
68318b8e | 6391 | |
eb95419b | 6392 | if (!parent) { |
68318b8e | 6393 | /* This is early initialization for the top cgroup */ |
07e06b01 | 6394 | return &root_task_group.css; |
68318b8e SV |
6395 | } |
6396 | ||
ec7dc8ac | 6397 | tg = sched_create_group(parent); |
68318b8e SV |
6398 | if (IS_ERR(tg)) |
6399 | return ERR_PTR(-ENOMEM); | |
6400 | ||
68318b8e SV |
6401 | return &tg->css; |
6402 | } | |
6403 | ||
96b77745 KK |
6404 | /* Expose task group only after completing cgroup initialization */ |
6405 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
6406 | { | |
6407 | struct task_group *tg = css_tg(css); | |
6408 | struct task_group *parent = css_tg(css->parent); | |
6409 | ||
6410 | if (parent) | |
6411 | sched_online_group(tg, parent); | |
6412 | return 0; | |
6413 | } | |
6414 | ||
2f5177f0 | 6415 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 6416 | { |
eb95419b | 6417 | struct task_group *tg = css_tg(css); |
ace783b9 | 6418 | |
2f5177f0 | 6419 | sched_offline_group(tg); |
ace783b9 LZ |
6420 | } |
6421 | ||
eb95419b | 6422 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 6423 | { |
eb95419b | 6424 | struct task_group *tg = css_tg(css); |
68318b8e | 6425 | |
2f5177f0 PZ |
6426 | /* |
6427 | * Relies on the RCU grace period between css_released() and this. | |
6428 | */ | |
6429 | sched_free_group(tg); | |
ace783b9 LZ |
6430 | } |
6431 | ||
ea86cb4b VG |
6432 | /* |
6433 | * This is called before wake_up_new_task(), therefore we really only | |
6434 | * have to set its group bits, all the other stuff does not apply. | |
6435 | */ | |
b53202e6 | 6436 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 6437 | { |
ea86cb4b VG |
6438 | struct rq_flags rf; |
6439 | struct rq *rq; | |
6440 | ||
6441 | rq = task_rq_lock(task, &rf); | |
6442 | ||
80f5c1b8 | 6443 | update_rq_clock(rq); |
ea86cb4b VG |
6444 | sched_change_group(task, TASK_SET_GROUP); |
6445 | ||
6446 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
6447 | } |
6448 | ||
1f7dd3e5 | 6449 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 6450 | { |
bb9d97b6 | 6451 | struct task_struct *task; |
1f7dd3e5 | 6452 | struct cgroup_subsys_state *css; |
7dc603c9 | 6453 | int ret = 0; |
bb9d97b6 | 6454 | |
1f7dd3e5 | 6455 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 6456 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 6457 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 6458 | return -EINVAL; |
b68aa230 | 6459 | #else |
bb9d97b6 TH |
6460 | /* We don't support RT-tasks being in separate groups */ |
6461 | if (task->sched_class != &fair_sched_class) | |
6462 | return -EINVAL; | |
b68aa230 | 6463 | #endif |
7dc603c9 PZ |
6464 | /* |
6465 | * Serialize against wake_up_new_task() such that if its | |
6466 | * running, we're sure to observe its full state. | |
6467 | */ | |
6468 | raw_spin_lock_irq(&task->pi_lock); | |
6469 | /* | |
6470 | * Avoid calling sched_move_task() before wake_up_new_task() | |
6471 | * has happened. This would lead to problems with PELT, due to | |
6472 | * move wanting to detach+attach while we're not attached yet. | |
6473 | */ | |
6474 | if (task->state == TASK_NEW) | |
6475 | ret = -EINVAL; | |
6476 | raw_spin_unlock_irq(&task->pi_lock); | |
6477 | ||
6478 | if (ret) | |
6479 | break; | |
bb9d97b6 | 6480 | } |
7dc603c9 | 6481 | return ret; |
be367d09 | 6482 | } |
68318b8e | 6483 | |
1f7dd3e5 | 6484 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 6485 | { |
bb9d97b6 | 6486 | struct task_struct *task; |
1f7dd3e5 | 6487 | struct cgroup_subsys_state *css; |
bb9d97b6 | 6488 | |
1f7dd3e5 | 6489 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 6490 | sched_move_task(task); |
68318b8e SV |
6491 | } |
6492 | ||
052f1dc7 | 6493 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
6494 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
6495 | struct cftype *cftype, u64 shareval) | |
68318b8e | 6496 | { |
182446d0 | 6497 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
6498 | } |
6499 | ||
182446d0 TH |
6500 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
6501 | struct cftype *cft) | |
68318b8e | 6502 | { |
182446d0 | 6503 | struct task_group *tg = css_tg(css); |
68318b8e | 6504 | |
c8b28116 | 6505 | return (u64) scale_load_down(tg->shares); |
68318b8e | 6506 | } |
ab84d31e PT |
6507 | |
6508 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
6509 | static DEFINE_MUTEX(cfs_constraints_mutex); |
6510 | ||
ab84d31e PT |
6511 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
6512 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ | |
6513 | ||
a790de99 PT |
6514 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
6515 | ||
ab84d31e PT |
6516 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
6517 | { | |
56f570e5 | 6518 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 6519 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
6520 | |
6521 | if (tg == &root_task_group) | |
6522 | return -EINVAL; | |
6523 | ||
6524 | /* | |
6525 | * Ensure we have at some amount of bandwidth every period. This is | |
6526 | * to prevent reaching a state of large arrears when throttled via | |
6527 | * entity_tick() resulting in prolonged exit starvation. | |
6528 | */ | |
6529 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
6530 | return -EINVAL; | |
6531 | ||
6532 | /* | |
6533 | * Likewise, bound things on the otherside by preventing insane quota | |
6534 | * periods. This also allows us to normalize in computing quota | |
6535 | * feasibility. | |
6536 | */ | |
6537 | if (period > max_cfs_quota_period) | |
6538 | return -EINVAL; | |
6539 | ||
0e59bdae KT |
6540 | /* |
6541 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
6542 | * unthrottle_offline_cfs_rqs(). | |
6543 | */ | |
6544 | get_online_cpus(); | |
a790de99 PT |
6545 | mutex_lock(&cfs_constraints_mutex); |
6546 | ret = __cfs_schedulable(tg, period, quota); | |
6547 | if (ret) | |
6548 | goto out_unlock; | |
6549 | ||
58088ad0 | 6550 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 6551 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
6552 | /* |
6553 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
6554 | * before making related changes, and on->off must occur afterwards | |
6555 | */ | |
6556 | if (runtime_enabled && !runtime_was_enabled) | |
6557 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
6558 | raw_spin_lock_irq(&cfs_b->lock); |
6559 | cfs_b->period = ns_to_ktime(period); | |
6560 | cfs_b->quota = quota; | |
58088ad0 | 6561 | |
a9cf55b2 | 6562 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
6563 | |
6564 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
6565 | if (runtime_enabled) |
6566 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 6567 | |
ab84d31e PT |
6568 | raw_spin_unlock_irq(&cfs_b->lock); |
6569 | ||
0e59bdae | 6570 | for_each_online_cpu(i) { |
ab84d31e | 6571 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 6572 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 6573 | struct rq_flags rf; |
ab84d31e | 6574 | |
8a8c69c3 | 6575 | rq_lock_irq(rq, &rf); |
58088ad0 | 6576 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 6577 | cfs_rq->runtime_remaining = 0; |
671fd9da | 6578 | |
029632fb | 6579 | if (cfs_rq->throttled) |
671fd9da | 6580 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 6581 | rq_unlock_irq(rq, &rf); |
ab84d31e | 6582 | } |
1ee14e6c BS |
6583 | if (runtime_was_enabled && !runtime_enabled) |
6584 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
6585 | out_unlock: |
6586 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 6587 | put_online_cpus(); |
ab84d31e | 6588 | |
a790de99 | 6589 | return ret; |
ab84d31e PT |
6590 | } |
6591 | ||
6592 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) | |
6593 | { | |
6594 | u64 quota, period; | |
6595 | ||
029632fb | 6596 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
6597 | if (cfs_quota_us < 0) |
6598 | quota = RUNTIME_INF; | |
6599 | else | |
6600 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; | |
6601 | ||
6602 | return tg_set_cfs_bandwidth(tg, period, quota); | |
6603 | } | |
6604 | ||
6605 | long tg_get_cfs_quota(struct task_group *tg) | |
6606 | { | |
6607 | u64 quota_us; | |
6608 | ||
029632fb | 6609 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
6610 | return -1; |
6611 | ||
029632fb | 6612 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
6613 | do_div(quota_us, NSEC_PER_USEC); |
6614 | ||
6615 | return quota_us; | |
6616 | } | |
6617 | ||
6618 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) | |
6619 | { | |
6620 | u64 quota, period; | |
6621 | ||
6622 | period = (u64)cfs_period_us * NSEC_PER_USEC; | |
029632fb | 6623 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 6624 | |
ab84d31e PT |
6625 | return tg_set_cfs_bandwidth(tg, period, quota); |
6626 | } | |
6627 | ||
6628 | long tg_get_cfs_period(struct task_group *tg) | |
6629 | { | |
6630 | u64 cfs_period_us; | |
6631 | ||
029632fb | 6632 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
6633 | do_div(cfs_period_us, NSEC_PER_USEC); |
6634 | ||
6635 | return cfs_period_us; | |
6636 | } | |
6637 | ||
182446d0 TH |
6638 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
6639 | struct cftype *cft) | |
ab84d31e | 6640 | { |
182446d0 | 6641 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
6642 | } |
6643 | ||
182446d0 TH |
6644 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
6645 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 6646 | { |
182446d0 | 6647 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
6648 | } |
6649 | ||
182446d0 TH |
6650 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
6651 | struct cftype *cft) | |
ab84d31e | 6652 | { |
182446d0 | 6653 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
6654 | } |
6655 | ||
182446d0 TH |
6656 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
6657 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 6658 | { |
182446d0 | 6659 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
6660 | } |
6661 | ||
a790de99 PT |
6662 | struct cfs_schedulable_data { |
6663 | struct task_group *tg; | |
6664 | u64 period, quota; | |
6665 | }; | |
6666 | ||
6667 | /* | |
6668 | * normalize group quota/period to be quota/max_period | |
6669 | * note: units are usecs | |
6670 | */ | |
6671 | static u64 normalize_cfs_quota(struct task_group *tg, | |
6672 | struct cfs_schedulable_data *d) | |
6673 | { | |
6674 | u64 quota, period; | |
6675 | ||
6676 | if (tg == d->tg) { | |
6677 | period = d->period; | |
6678 | quota = d->quota; | |
6679 | } else { | |
6680 | period = tg_get_cfs_period(tg); | |
6681 | quota = tg_get_cfs_quota(tg); | |
6682 | } | |
6683 | ||
6684 | /* note: these should typically be equivalent */ | |
6685 | if (quota == RUNTIME_INF || quota == -1) | |
6686 | return RUNTIME_INF; | |
6687 | ||
6688 | return to_ratio(period, quota); | |
6689 | } | |
6690 | ||
6691 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
6692 | { | |
6693 | struct cfs_schedulable_data *d = data; | |
029632fb | 6694 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
6695 | s64 quota = 0, parent_quota = -1; |
6696 | ||
6697 | if (!tg->parent) { | |
6698 | quota = RUNTIME_INF; | |
6699 | } else { | |
029632fb | 6700 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
6701 | |
6702 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 6703 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
6704 | |
6705 | /* | |
c53593e5 TH |
6706 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
6707 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 6708 | * limit is set: |
a790de99 | 6709 | */ |
c53593e5 TH |
6710 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
6711 | quota = min(quota, parent_quota); | |
6712 | } else { | |
6713 | if (quota == RUNTIME_INF) | |
6714 | quota = parent_quota; | |
6715 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
6716 | return -EINVAL; | |
6717 | } | |
a790de99 | 6718 | } |
9c58c79a | 6719 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
6720 | |
6721 | return 0; | |
6722 | } | |
6723 | ||
6724 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
6725 | { | |
8277434e | 6726 | int ret; |
a790de99 PT |
6727 | struct cfs_schedulable_data data = { |
6728 | .tg = tg, | |
6729 | .period = period, | |
6730 | .quota = quota, | |
6731 | }; | |
6732 | ||
6733 | if (quota != RUNTIME_INF) { | |
6734 | do_div(data.period, NSEC_PER_USEC); | |
6735 | do_div(data.quota, NSEC_PER_USEC); | |
6736 | } | |
6737 | ||
8277434e PT |
6738 | rcu_read_lock(); |
6739 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
6740 | rcu_read_unlock(); | |
6741 | ||
6742 | return ret; | |
a790de99 | 6743 | } |
e8da1b18 | 6744 | |
a1f7164c | 6745 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 6746 | { |
2da8ca82 | 6747 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 6748 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 6749 | |
44ffc75b TH |
6750 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
6751 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
6752 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 6753 | |
3d6c50c2 YW |
6754 | if (schedstat_enabled() && tg != &root_task_group) { |
6755 | u64 ws = 0; | |
6756 | int i; | |
6757 | ||
6758 | for_each_possible_cpu(i) | |
6759 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
6760 | ||
6761 | seq_printf(sf, "wait_sum %llu\n", ws); | |
6762 | } | |
6763 | ||
e8da1b18 NR |
6764 | return 0; |
6765 | } | |
ab84d31e | 6766 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 6767 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 6768 | |
052f1dc7 | 6769 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
6770 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
6771 | struct cftype *cft, s64 val) | |
6f505b16 | 6772 | { |
182446d0 | 6773 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
6774 | } |
6775 | ||
182446d0 TH |
6776 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
6777 | struct cftype *cft) | |
6f505b16 | 6778 | { |
182446d0 | 6779 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 6780 | } |
d0b27fa7 | 6781 | |
182446d0 TH |
6782 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
6783 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 6784 | { |
182446d0 | 6785 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
6786 | } |
6787 | ||
182446d0 TH |
6788 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
6789 | struct cftype *cft) | |
d0b27fa7 | 6790 | { |
182446d0 | 6791 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 6792 | } |
6d6bc0ad | 6793 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 6794 | |
a1f7164c | 6795 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 6796 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
6797 | { |
6798 | .name = "shares", | |
f4c753b7 PM |
6799 | .read_u64 = cpu_shares_read_u64, |
6800 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 6801 | }, |
052f1dc7 | 6802 | #endif |
ab84d31e PT |
6803 | #ifdef CONFIG_CFS_BANDWIDTH |
6804 | { | |
6805 | .name = "cfs_quota_us", | |
6806 | .read_s64 = cpu_cfs_quota_read_s64, | |
6807 | .write_s64 = cpu_cfs_quota_write_s64, | |
6808 | }, | |
6809 | { | |
6810 | .name = "cfs_period_us", | |
6811 | .read_u64 = cpu_cfs_period_read_u64, | |
6812 | .write_u64 = cpu_cfs_period_write_u64, | |
6813 | }, | |
e8da1b18 NR |
6814 | { |
6815 | .name = "stat", | |
a1f7164c | 6816 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 6817 | }, |
ab84d31e | 6818 | #endif |
052f1dc7 | 6819 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 6820 | { |
9f0c1e56 | 6821 | .name = "rt_runtime_us", |
06ecb27c PM |
6822 | .read_s64 = cpu_rt_runtime_read, |
6823 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 6824 | }, |
d0b27fa7 PZ |
6825 | { |
6826 | .name = "rt_period_us", | |
f4c753b7 PM |
6827 | .read_u64 = cpu_rt_period_read_uint, |
6828 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 6829 | }, |
052f1dc7 | 6830 | #endif |
d1ccc66d | 6831 | { } /* Terminate */ |
68318b8e SV |
6832 | }; |
6833 | ||
d41bf8c9 TH |
6834 | static int cpu_extra_stat_show(struct seq_file *sf, |
6835 | struct cgroup_subsys_state *css) | |
0d593634 | 6836 | { |
0d593634 TH |
6837 | #ifdef CONFIG_CFS_BANDWIDTH |
6838 | { | |
d41bf8c9 | 6839 | struct task_group *tg = css_tg(css); |
0d593634 TH |
6840 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
6841 | u64 throttled_usec; | |
6842 | ||
6843 | throttled_usec = cfs_b->throttled_time; | |
6844 | do_div(throttled_usec, NSEC_PER_USEC); | |
6845 | ||
6846 | seq_printf(sf, "nr_periods %d\n" | |
6847 | "nr_throttled %d\n" | |
6848 | "throttled_usec %llu\n", | |
6849 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
6850 | throttled_usec); | |
6851 | } | |
6852 | #endif | |
6853 | return 0; | |
6854 | } | |
6855 | ||
6856 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6857 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
6858 | struct cftype *cft) | |
6859 | { | |
6860 | struct task_group *tg = css_tg(css); | |
6861 | u64 weight = scale_load_down(tg->shares); | |
6862 | ||
6863 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
6864 | } | |
6865 | ||
6866 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
6867 | struct cftype *cft, u64 weight) | |
6868 | { | |
6869 | /* | |
6870 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
6871 | * values which are 1, 100 and 10000 respectively. While it loses | |
6872 | * a bit of range on both ends, it maps pretty well onto the shares | |
6873 | * value used by scheduler and the round-trip conversions preserve | |
6874 | * the original value over the entire range. | |
6875 | */ | |
6876 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
6877 | return -ERANGE; | |
6878 | ||
6879 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
6880 | ||
6881 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
6882 | } | |
6883 | ||
6884 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
6885 | struct cftype *cft) | |
6886 | { | |
6887 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
6888 | int last_delta = INT_MAX; | |
6889 | int prio, delta; | |
6890 | ||
6891 | /* find the closest nice value to the current weight */ | |
6892 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
6893 | delta = abs(sched_prio_to_weight[prio] - weight); | |
6894 | if (delta >= last_delta) | |
6895 | break; | |
6896 | last_delta = delta; | |
6897 | } | |
6898 | ||
6899 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
6900 | } | |
6901 | ||
6902 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
6903 | struct cftype *cft, s64 nice) | |
6904 | { | |
6905 | unsigned long weight; | |
7281c8de | 6906 | int idx; |
0d593634 TH |
6907 | |
6908 | if (nice < MIN_NICE || nice > MAX_NICE) | |
6909 | return -ERANGE; | |
6910 | ||
7281c8de PZ |
6911 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
6912 | idx = array_index_nospec(idx, 40); | |
6913 | weight = sched_prio_to_weight[idx]; | |
6914 | ||
0d593634 TH |
6915 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
6916 | } | |
6917 | #endif | |
6918 | ||
6919 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
6920 | long period, long quota) | |
6921 | { | |
6922 | if (quota < 0) | |
6923 | seq_puts(sf, "max"); | |
6924 | else | |
6925 | seq_printf(sf, "%ld", quota); | |
6926 | ||
6927 | seq_printf(sf, " %ld\n", period); | |
6928 | } | |
6929 | ||
6930 | /* caller should put the current value in *@periodp before calling */ | |
6931 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
6932 | u64 *periodp, u64 *quotap) | |
6933 | { | |
6934 | char tok[21]; /* U64_MAX */ | |
6935 | ||
6936 | if (!sscanf(buf, "%s %llu", tok, periodp)) | |
6937 | return -EINVAL; | |
6938 | ||
6939 | *periodp *= NSEC_PER_USEC; | |
6940 | ||
6941 | if (sscanf(tok, "%llu", quotap)) | |
6942 | *quotap *= NSEC_PER_USEC; | |
6943 | else if (!strcmp(tok, "max")) | |
6944 | *quotap = RUNTIME_INF; | |
6945 | else | |
6946 | return -EINVAL; | |
6947 | ||
6948 | return 0; | |
6949 | } | |
6950 | ||
6951 | #ifdef CONFIG_CFS_BANDWIDTH | |
6952 | static int cpu_max_show(struct seq_file *sf, void *v) | |
6953 | { | |
6954 | struct task_group *tg = css_tg(seq_css(sf)); | |
6955 | ||
6956 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
6957 | return 0; | |
6958 | } | |
6959 | ||
6960 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
6961 | char *buf, size_t nbytes, loff_t off) | |
6962 | { | |
6963 | struct task_group *tg = css_tg(of_css(of)); | |
6964 | u64 period = tg_get_cfs_period(tg); | |
6965 | u64 quota; | |
6966 | int ret; | |
6967 | ||
6968 | ret = cpu_period_quota_parse(buf, &period, "a); | |
6969 | if (!ret) | |
6970 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
6971 | return ret ?: nbytes; | |
6972 | } | |
6973 | #endif | |
6974 | ||
6975 | static struct cftype cpu_files[] = { | |
0d593634 TH |
6976 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6977 | { | |
6978 | .name = "weight", | |
6979 | .flags = CFTYPE_NOT_ON_ROOT, | |
6980 | .read_u64 = cpu_weight_read_u64, | |
6981 | .write_u64 = cpu_weight_write_u64, | |
6982 | }, | |
6983 | { | |
6984 | .name = "weight.nice", | |
6985 | .flags = CFTYPE_NOT_ON_ROOT, | |
6986 | .read_s64 = cpu_weight_nice_read_s64, | |
6987 | .write_s64 = cpu_weight_nice_write_s64, | |
6988 | }, | |
6989 | #endif | |
6990 | #ifdef CONFIG_CFS_BANDWIDTH | |
6991 | { | |
6992 | .name = "max", | |
6993 | .flags = CFTYPE_NOT_ON_ROOT, | |
6994 | .seq_show = cpu_max_show, | |
6995 | .write = cpu_max_write, | |
6996 | }, | |
6997 | #endif | |
6998 | { } /* terminate */ | |
6999 | }; | |
7000 | ||
073219e9 | 7001 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 7002 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 7003 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 7004 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 7005 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 7006 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 7007 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
7008 | .can_attach = cpu_cgroup_can_attach, |
7009 | .attach = cpu_cgroup_attach, | |
a1f7164c | 7010 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 7011 | .dfl_cftypes = cpu_files, |
b38e42e9 | 7012 | .early_init = true, |
0d593634 | 7013 | .threaded = true, |
68318b8e SV |
7014 | }; |
7015 | ||
052f1dc7 | 7016 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 7017 | |
b637a328 PM |
7018 | void dump_cpu_task(int cpu) |
7019 | { | |
7020 | pr_info("Task dump for CPU %d:\n", cpu); | |
7021 | sched_show_task(cpu_curr(cpu)); | |
7022 | } | |
ed82b8a1 AK |
7023 | |
7024 | /* | |
7025 | * Nice levels are multiplicative, with a gentle 10% change for every | |
7026 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
7027 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
7028 | * that remained on nice 0. | |
7029 | * | |
7030 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
7031 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
7032 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
7033 | * If a task goes up by ~10% and another task goes down by ~10% then | |
7034 | * the relative distance between them is ~25%.) | |
7035 | */ | |
7036 | const int sched_prio_to_weight[40] = { | |
7037 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
7038 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
7039 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
7040 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
7041 | /* 0 */ 1024, 820, 655, 526, 423, | |
7042 | /* 5 */ 335, 272, 215, 172, 137, | |
7043 | /* 10 */ 110, 87, 70, 56, 45, | |
7044 | /* 15 */ 36, 29, 23, 18, 15, | |
7045 | }; | |
7046 | ||
7047 | /* | |
7048 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
7049 | * | |
7050 | * In cases where the weight does not change often, we can use the | |
7051 | * precalculated inverse to speed up arithmetics by turning divisions | |
7052 | * into multiplications: | |
7053 | */ | |
7054 | const u32 sched_prio_to_wmult[40] = { | |
7055 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
7056 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
7057 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
7058 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
7059 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
7060 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
7061 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
7062 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
7063 | }; | |
14a7405b IM |
7064 | |
7065 | #undef CREATE_TRACE_POINTS |