]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
391e43da | 3 | * kernel/sched/core.c |
1da177e4 | 4 | * |
d1ccc66d | 5 | * Core kernel scheduler code and related syscalls |
1da177e4 LT |
6 | * |
7 | * Copyright (C) 1991-2002 Linus Torvalds | |
1da177e4 | 8 | */ |
325ea10c | 9 | #include "sched.h" |
1da177e4 | 10 | |
7281c8de | 11 | #include <linux/nospec.h> |
85f1abe0 | 12 | |
0ed557aa | 13 | #include <linux/kcov.h> |
d08b9f0c | 14 | #include <linux/scs.h> |
0ed557aa | 15 | |
96f951ed | 16 | #include <asm/switch_to.h> |
5517d86b | 17 | #include <asm/tlb.h> |
1da177e4 | 18 | |
ea138446 | 19 | #include "../workqueue_internal.h" |
771b53d0 | 20 | #include "../../fs/io-wq.h" |
29d5e047 | 21 | #include "../smpboot.h" |
6e0534f2 | 22 | |
91c27493 | 23 | #include "pelt.h" |
1f8db415 | 24 | #include "smp.h" |
91c27493 | 25 | |
a8d154b0 | 26 | #define CREATE_TRACE_POINTS |
ad8d75ff | 27 | #include <trace/events/sched.h> |
a8d154b0 | 28 | |
a056a5be QY |
29 | /* |
30 | * Export tracepoints that act as a bare tracehook (ie: have no trace event | |
31 | * associated with them) to allow external modules to probe them. | |
32 | */ | |
33 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp); | |
34 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp); | |
35 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp); | |
36 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp); | |
37 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp); | |
38 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); | |
39 | ||
029632fb | 40 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 41 | |
e9666d10 | 42 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL) |
bf5c91ba IM |
43 | /* |
44 | * Debugging: various feature bits | |
765cc3a4 PB |
45 | * |
46 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
47 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
48 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 49 | */ |
f00b45c1 PZ |
50 | #define SCHED_FEAT(name, enabled) \ |
51 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 52 | const_debug unsigned int sysctl_sched_features = |
391e43da | 53 | #include "features.h" |
f00b45c1 | 54 | 0; |
f00b45c1 | 55 | #undef SCHED_FEAT |
765cc3a4 | 56 | #endif |
f00b45c1 | 57 | |
b82d9fdd PZ |
58 | /* |
59 | * Number of tasks to iterate in a single balance run. | |
60 | * Limited because this is done with IRQs disabled. | |
61 | */ | |
62 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
63 | ||
fa85ae24 | 64 | /* |
d1ccc66d | 65 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
66 | * default: 1s |
67 | */ | |
9f0c1e56 | 68 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 69 | |
029632fb | 70 | __read_mostly int scheduler_running; |
6892b75e | 71 | |
9f0c1e56 PZ |
72 | /* |
73 | * part of the period that we allow rt tasks to run in us. | |
74 | * default: 0.95s | |
75 | */ | |
76 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 77 | |
3e71a462 PZ |
78 | /* |
79 | * __task_rq_lock - lock the rq @p resides on. | |
80 | */ | |
eb580751 | 81 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
82 | __acquires(rq->lock) |
83 | { | |
84 | struct rq *rq; | |
85 | ||
86 | lockdep_assert_held(&p->pi_lock); | |
87 | ||
88 | for (;;) { | |
89 | rq = task_rq(p); | |
90 | raw_spin_lock(&rq->lock); | |
91 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 92 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
93 | return rq; |
94 | } | |
95 | raw_spin_unlock(&rq->lock); | |
96 | ||
97 | while (unlikely(task_on_rq_migrating(p))) | |
98 | cpu_relax(); | |
99 | } | |
100 | } | |
101 | ||
102 | /* | |
103 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
104 | */ | |
eb580751 | 105 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
106 | __acquires(p->pi_lock) |
107 | __acquires(rq->lock) | |
108 | { | |
109 | struct rq *rq; | |
110 | ||
111 | for (;;) { | |
eb580751 | 112 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
113 | rq = task_rq(p); |
114 | raw_spin_lock(&rq->lock); | |
115 | /* | |
116 | * move_queued_task() task_rq_lock() | |
117 | * | |
118 | * ACQUIRE (rq->lock) | |
119 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
120 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
121 | * [S] ->cpu = new_cpu [L] task_rq() | |
122 | * [L] ->on_rq | |
123 | * RELEASE (rq->lock) | |
124 | * | |
c546951d | 125 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
126 | * the old rq->lock will fully serialize against the stores. |
127 | * | |
c546951d AP |
128 | * If we observe the new CPU in task_rq_lock(), the address |
129 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
130 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
131 | */ |
132 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 133 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
134 | return rq; |
135 | } | |
136 | raw_spin_unlock(&rq->lock); | |
eb580751 | 137 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
138 | |
139 | while (unlikely(task_on_rq_migrating(p))) | |
140 | cpu_relax(); | |
141 | } | |
142 | } | |
143 | ||
535b9552 IM |
144 | /* |
145 | * RQ-clock updating methods: | |
146 | */ | |
147 | ||
148 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
149 | { | |
150 | /* | |
151 | * In theory, the compile should just see 0 here, and optimize out the call | |
152 | * to sched_rt_avg_update. But I don't trust it... | |
153 | */ | |
11d4afd4 VG |
154 | s64 __maybe_unused steal = 0, irq_delta = 0; |
155 | ||
535b9552 IM |
156 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
157 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
158 | ||
159 | /* | |
160 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
161 | * this case when a previous update_rq_clock() happened inside a | |
162 | * {soft,}irq region. | |
163 | * | |
164 | * When this happens, we stop ->clock_task and only update the | |
165 | * prev_irq_time stamp to account for the part that fit, so that a next | |
166 | * update will consume the rest. This ensures ->clock_task is | |
167 | * monotonic. | |
168 | * | |
169 | * It does however cause some slight miss-attribution of {soft,}irq | |
170 | * time, a more accurate solution would be to update the irq_time using | |
171 | * the current rq->clock timestamp, except that would require using | |
172 | * atomic ops. | |
173 | */ | |
174 | if (irq_delta > delta) | |
175 | irq_delta = delta; | |
176 | ||
177 | rq->prev_irq_time += irq_delta; | |
178 | delta -= irq_delta; | |
179 | #endif | |
180 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
181 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
182 | steal = paravirt_steal_clock(cpu_of(rq)); | |
183 | steal -= rq->prev_steal_time_rq; | |
184 | ||
185 | if (unlikely(steal > delta)) | |
186 | steal = delta; | |
187 | ||
188 | rq->prev_steal_time_rq += steal; | |
189 | delta -= steal; | |
190 | } | |
191 | #endif | |
192 | ||
193 | rq->clock_task += delta; | |
194 | ||
11d4afd4 | 195 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 196 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 197 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 198 | #endif |
23127296 | 199 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
200 | } |
201 | ||
202 | void update_rq_clock(struct rq *rq) | |
203 | { | |
204 | s64 delta; | |
205 | ||
206 | lockdep_assert_held(&rq->lock); | |
207 | ||
208 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
209 | return; | |
210 | ||
211 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
212 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
213 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
214 | rq->clock_update_flags |= RQCF_UPDATED; |
215 | #endif | |
26ae58d2 | 216 | |
535b9552 IM |
217 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
218 | if (delta < 0) | |
219 | return; | |
220 | rq->clock += delta; | |
221 | update_rq_clock_task(rq, delta); | |
222 | } | |
223 | ||
90b5363a PZI |
224 | static inline void |
225 | rq_csd_init(struct rq *rq, call_single_data_t *csd, smp_call_func_t func) | |
226 | { | |
227 | csd->flags = 0; | |
228 | csd->func = func; | |
229 | csd->info = rq; | |
230 | } | |
535b9552 | 231 | |
8f4d37ec PZ |
232 | #ifdef CONFIG_SCHED_HRTICK |
233 | /* | |
234 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 235 | */ |
8f4d37ec | 236 | |
8f4d37ec PZ |
237 | static void hrtick_clear(struct rq *rq) |
238 | { | |
239 | if (hrtimer_active(&rq->hrtick_timer)) | |
240 | hrtimer_cancel(&rq->hrtick_timer); | |
241 | } | |
242 | ||
8f4d37ec PZ |
243 | /* |
244 | * High-resolution timer tick. | |
245 | * Runs from hardirq context with interrupts disabled. | |
246 | */ | |
247 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
248 | { | |
249 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 250 | struct rq_flags rf; |
8f4d37ec PZ |
251 | |
252 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
253 | ||
8a8c69c3 | 254 | rq_lock(rq, &rf); |
3e51f33f | 255 | update_rq_clock(rq); |
8f4d37ec | 256 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 257 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
258 | |
259 | return HRTIMER_NORESTART; | |
260 | } | |
261 | ||
95e904c7 | 262 | #ifdef CONFIG_SMP |
971ee28c | 263 | |
4961b6e1 | 264 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
265 | { |
266 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 267 | |
d5096aa6 | 268 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
269 | } |
270 | ||
31656519 PZ |
271 | /* |
272 | * called from hardirq (IPI) context | |
273 | */ | |
274 | static void __hrtick_start(void *arg) | |
b328ca18 | 275 | { |
31656519 | 276 | struct rq *rq = arg; |
8a8c69c3 | 277 | struct rq_flags rf; |
b328ca18 | 278 | |
8a8c69c3 | 279 | rq_lock(rq, &rf); |
971ee28c | 280 | __hrtick_restart(rq); |
8a8c69c3 | 281 | rq_unlock(rq, &rf); |
b328ca18 PZ |
282 | } |
283 | ||
31656519 PZ |
284 | /* |
285 | * Called to set the hrtick timer state. | |
286 | * | |
287 | * called with rq->lock held and irqs disabled | |
288 | */ | |
029632fb | 289 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 290 | { |
31656519 | 291 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 292 | ktime_t time; |
293 | s64 delta; | |
294 | ||
295 | /* | |
296 | * Don't schedule slices shorter than 10000ns, that just | |
297 | * doesn't make sense and can cause timer DoS. | |
298 | */ | |
299 | delta = max_t(s64, delay, 10000LL); | |
300 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 301 | |
cc584b21 | 302 | hrtimer_set_expires(timer, time); |
31656519 | 303 | |
fd3eafda | 304 | if (rq == this_rq()) |
971ee28c | 305 | __hrtick_restart(rq); |
fd3eafda | 306 | else |
c46fff2a | 307 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
308 | } |
309 | ||
31656519 PZ |
310 | #else |
311 | /* | |
312 | * Called to set the hrtick timer state. | |
313 | * | |
314 | * called with rq->lock held and irqs disabled | |
315 | */ | |
029632fb | 316 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 317 | { |
86893335 WL |
318 | /* |
319 | * Don't schedule slices shorter than 10000ns, that just | |
320 | * doesn't make sense. Rely on vruntime for fairness. | |
321 | */ | |
322 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 323 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 324 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 325 | } |
90b5363a | 326 | |
31656519 | 327 | #endif /* CONFIG_SMP */ |
8f4d37ec | 328 | |
77a021be | 329 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 330 | { |
31656519 | 331 | #ifdef CONFIG_SMP |
90b5363a | 332 | rq_csd_init(rq, &rq->hrtick_csd, __hrtick_start); |
31656519 | 333 | #endif |
d5096aa6 | 334 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 335 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 336 | } |
006c75f1 | 337 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
338 | static inline void hrtick_clear(struct rq *rq) |
339 | { | |
340 | } | |
341 | ||
77a021be | 342 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
343 | { |
344 | } | |
006c75f1 | 345 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 346 | |
5529578a FW |
347 | /* |
348 | * cmpxchg based fetch_or, macro so it works for different integer types | |
349 | */ | |
350 | #define fetch_or(ptr, mask) \ | |
351 | ({ \ | |
352 | typeof(ptr) _ptr = (ptr); \ | |
353 | typeof(mask) _mask = (mask); \ | |
354 | typeof(*_ptr) _old, _val = *_ptr; \ | |
355 | \ | |
356 | for (;;) { \ | |
357 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
358 | if (_old == _val) \ | |
359 | break; \ | |
360 | _val = _old; \ | |
361 | } \ | |
362 | _old; \ | |
363 | }) | |
364 | ||
e3baac47 | 365 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
366 | /* |
367 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
368 | * this avoids any races wrt polling state changes and thereby avoids | |
369 | * spurious IPIs. | |
370 | */ | |
371 | static bool set_nr_and_not_polling(struct task_struct *p) | |
372 | { | |
373 | struct thread_info *ti = task_thread_info(p); | |
374 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
375 | } | |
e3baac47 PZ |
376 | |
377 | /* | |
378 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
379 | * | |
380 | * If this returns true, then the idle task promises to call | |
381 | * sched_ttwu_pending() and reschedule soon. | |
382 | */ | |
383 | static bool set_nr_if_polling(struct task_struct *p) | |
384 | { | |
385 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 386 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
387 | |
388 | for (;;) { | |
389 | if (!(val & _TIF_POLLING_NRFLAG)) | |
390 | return false; | |
391 | if (val & _TIF_NEED_RESCHED) | |
392 | return true; | |
393 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
394 | if (old == val) | |
395 | break; | |
396 | val = old; | |
397 | } | |
398 | return true; | |
399 | } | |
400 | ||
fd99f91a PZ |
401 | #else |
402 | static bool set_nr_and_not_polling(struct task_struct *p) | |
403 | { | |
404 | set_tsk_need_resched(p); | |
405 | return true; | |
406 | } | |
e3baac47 PZ |
407 | |
408 | #ifdef CONFIG_SMP | |
409 | static bool set_nr_if_polling(struct task_struct *p) | |
410 | { | |
411 | return false; | |
412 | } | |
413 | #endif | |
fd99f91a PZ |
414 | #endif |
415 | ||
07879c6a | 416 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
417 | { |
418 | struct wake_q_node *node = &task->wake_q; | |
419 | ||
420 | /* | |
421 | * Atomically grab the task, if ->wake_q is !nil already it means | |
422 | * its already queued (either by us or someone else) and will get the | |
423 | * wakeup due to that. | |
424 | * | |
4c4e3731 PZ |
425 | * In order to ensure that a pending wakeup will observe our pending |
426 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 427 | */ |
4c4e3731 | 428 | smp_mb__before_atomic(); |
87ff19cb | 429 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 430 | return false; |
76751049 PZ |
431 | |
432 | /* | |
433 | * The head is context local, there can be no concurrency. | |
434 | */ | |
435 | *head->lastp = node; | |
436 | head->lastp = &node->next; | |
07879c6a DB |
437 | return true; |
438 | } | |
439 | ||
440 | /** | |
441 | * wake_q_add() - queue a wakeup for 'later' waking. | |
442 | * @head: the wake_q_head to add @task to | |
443 | * @task: the task to queue for 'later' wakeup | |
444 | * | |
445 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
446 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
447 | * instantly. | |
448 | * | |
449 | * This function must be used as-if it were wake_up_process(); IOW the task | |
450 | * must be ready to be woken at this location. | |
451 | */ | |
452 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
453 | { | |
454 | if (__wake_q_add(head, task)) | |
455 | get_task_struct(task); | |
456 | } | |
457 | ||
458 | /** | |
459 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
460 | * @head: the wake_q_head to add @task to | |
461 | * @task: the task to queue for 'later' wakeup | |
462 | * | |
463 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
464 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
465 | * instantly. | |
466 | * | |
467 | * This function must be used as-if it were wake_up_process(); IOW the task | |
468 | * must be ready to be woken at this location. | |
469 | * | |
470 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
471 | * that already hold reference to @task can call the 'safe' version and trust | |
472 | * wake_q to do the right thing depending whether or not the @task is already | |
473 | * queued for wakeup. | |
474 | */ | |
475 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
476 | { | |
477 | if (!__wake_q_add(head, task)) | |
478 | put_task_struct(task); | |
76751049 PZ |
479 | } |
480 | ||
481 | void wake_up_q(struct wake_q_head *head) | |
482 | { | |
483 | struct wake_q_node *node = head->first; | |
484 | ||
485 | while (node != WAKE_Q_TAIL) { | |
486 | struct task_struct *task; | |
487 | ||
488 | task = container_of(node, struct task_struct, wake_q); | |
489 | BUG_ON(!task); | |
d1ccc66d | 490 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
491 | node = node->next; |
492 | task->wake_q.next = NULL; | |
493 | ||
494 | /* | |
7696f991 AP |
495 | * wake_up_process() executes a full barrier, which pairs with |
496 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
497 | */ |
498 | wake_up_process(task); | |
499 | put_task_struct(task); | |
500 | } | |
501 | } | |
502 | ||
c24d20db | 503 | /* |
8875125e | 504 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
505 | * |
506 | * On UP this means the setting of the need_resched flag, on SMP it | |
507 | * might also involve a cross-CPU call to trigger the scheduler on | |
508 | * the target CPU. | |
509 | */ | |
8875125e | 510 | void resched_curr(struct rq *rq) |
c24d20db | 511 | { |
8875125e | 512 | struct task_struct *curr = rq->curr; |
c24d20db IM |
513 | int cpu; |
514 | ||
8875125e | 515 | lockdep_assert_held(&rq->lock); |
c24d20db | 516 | |
8875125e | 517 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
518 | return; |
519 | ||
8875125e | 520 | cpu = cpu_of(rq); |
fd99f91a | 521 | |
f27dde8d | 522 | if (cpu == smp_processor_id()) { |
8875125e | 523 | set_tsk_need_resched(curr); |
f27dde8d | 524 | set_preempt_need_resched(); |
c24d20db | 525 | return; |
f27dde8d | 526 | } |
c24d20db | 527 | |
8875125e | 528 | if (set_nr_and_not_polling(curr)) |
c24d20db | 529 | smp_send_reschedule(cpu); |
dfc68f29 AL |
530 | else |
531 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
532 | } |
533 | ||
029632fb | 534 | void resched_cpu(int cpu) |
c24d20db IM |
535 | { |
536 | struct rq *rq = cpu_rq(cpu); | |
537 | unsigned long flags; | |
538 | ||
7c2102e5 | 539 | raw_spin_lock_irqsave(&rq->lock, flags); |
a0982dfa PM |
540 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
541 | resched_curr(rq); | |
05fa785c | 542 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 543 | } |
06d8308c | 544 | |
b021fe3e | 545 | #ifdef CONFIG_SMP |
3451d024 | 546 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 547 | /* |
d1ccc66d IM |
548 | * In the semi idle case, use the nearest busy CPU for migrating timers |
549 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
550 | * |
551 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
552 | * selecting an idle CPU will add more delays to the timers than intended |
553 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 554 | */ |
bc7a34b8 | 555 | int get_nohz_timer_target(void) |
83cd4fe2 | 556 | { |
e938b9c9 | 557 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 VP |
558 | struct sched_domain *sd; |
559 | ||
e938b9c9 WL |
560 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
561 | if (!idle_cpu(cpu)) | |
562 | return cpu; | |
563 | default_cpu = cpu; | |
564 | } | |
6201b4d6 | 565 | |
057f3fad | 566 | rcu_read_lock(); |
83cd4fe2 | 567 | for_each_domain(cpu, sd) { |
e938b9c9 WL |
568 | for_each_cpu_and(i, sched_domain_span(sd), |
569 | housekeeping_cpumask(HK_FLAG_TIMER)) { | |
44496922 WL |
570 | if (cpu == i) |
571 | continue; | |
572 | ||
e938b9c9 | 573 | if (!idle_cpu(i)) { |
057f3fad PZ |
574 | cpu = i; |
575 | goto unlock; | |
576 | } | |
577 | } | |
83cd4fe2 | 578 | } |
9642d18e | 579 | |
e938b9c9 WL |
580 | if (default_cpu == -1) |
581 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
582 | cpu = default_cpu; | |
057f3fad PZ |
583 | unlock: |
584 | rcu_read_unlock(); | |
83cd4fe2 VP |
585 | return cpu; |
586 | } | |
d1ccc66d | 587 | |
06d8308c TG |
588 | /* |
589 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
590 | * idle CPU then this timer might expire before the next timer event | |
591 | * which is scheduled to wake up that CPU. In case of a completely | |
592 | * idle system the next event might even be infinite time into the | |
593 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
594 | * leaves the inner idle loop so the newly added timer is taken into | |
595 | * account when the CPU goes back to idle and evaluates the timer | |
596 | * wheel for the next timer event. | |
597 | */ | |
1c20091e | 598 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
599 | { |
600 | struct rq *rq = cpu_rq(cpu); | |
601 | ||
602 | if (cpu == smp_processor_id()) | |
603 | return; | |
604 | ||
67b9ca70 | 605 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 606 | smp_send_reschedule(cpu); |
dfc68f29 AL |
607 | else |
608 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
609 | } |
610 | ||
c5bfece2 | 611 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 612 | { |
53c5fa16 FW |
613 | /* |
614 | * We just need the target to call irq_exit() and re-evaluate | |
615 | * the next tick. The nohz full kick at least implies that. | |
616 | * If needed we can still optimize that later with an | |
617 | * empty IRQ. | |
618 | */ | |
379d9ecb PM |
619 | if (cpu_is_offline(cpu)) |
620 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 621 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
622 | if (cpu != smp_processor_id() || |
623 | tick_nohz_tick_stopped()) | |
53c5fa16 | 624 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
625 | return true; |
626 | } | |
627 | ||
628 | return false; | |
629 | } | |
630 | ||
379d9ecb PM |
631 | /* |
632 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
633 | * caller's responsibility to deal with the lost wakeup, for example, | |
634 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
635 | */ | |
1c20091e FW |
636 | void wake_up_nohz_cpu(int cpu) |
637 | { | |
c5bfece2 | 638 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
639 | wake_up_idle_cpu(cpu); |
640 | } | |
641 | ||
19a1f5ec | 642 | static void nohz_csd_func(void *info) |
45bf76df | 643 | { |
19a1f5ec PZ |
644 | struct rq *rq = info; |
645 | int cpu = cpu_of(rq); | |
646 | unsigned int flags; | |
873b4c65 VG |
647 | |
648 | /* | |
19a1f5ec | 649 | * Release the rq::nohz_csd. |
873b4c65 | 650 | */ |
19a1f5ec PZ |
651 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(cpu)); |
652 | WARN_ON(!(flags & NOHZ_KICK_MASK)); | |
45bf76df | 653 | |
19a1f5ec PZ |
654 | rq->idle_balance = idle_cpu(cpu); |
655 | if (rq->idle_balance && !need_resched()) { | |
656 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
657 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
658 | } | |
2069dd75 PZ |
659 | } |
660 | ||
3451d024 | 661 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 662 | |
ce831b38 | 663 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 664 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 665 | { |
76d92ac3 FW |
666 | int fifo_nr_running; |
667 | ||
668 | /* Deadline tasks, even if single, need the tick */ | |
669 | if (rq->dl.dl_nr_running) | |
670 | return false; | |
671 | ||
1e78cdbd | 672 | /* |
2548d546 PZ |
673 | * If there are more than one RR tasks, we need the tick to effect the |
674 | * actual RR behaviour. | |
1e78cdbd | 675 | */ |
76d92ac3 FW |
676 | if (rq->rt.rr_nr_running) { |
677 | if (rq->rt.rr_nr_running == 1) | |
678 | return true; | |
679 | else | |
680 | return false; | |
1e78cdbd RR |
681 | } |
682 | ||
2548d546 PZ |
683 | /* |
684 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
685 | * forced preemption between FIFO tasks. | |
686 | */ | |
687 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
688 | if (fifo_nr_running) | |
689 | return true; | |
690 | ||
691 | /* | |
692 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
693 | * if there's more than one we need the tick for involuntary | |
694 | * preemption. | |
695 | */ | |
696 | if (rq->nr_running > 1) | |
541b8264 | 697 | return false; |
ce831b38 | 698 | |
541b8264 | 699 | return true; |
ce831b38 FW |
700 | } |
701 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 702 | #endif /* CONFIG_SMP */ |
18d95a28 | 703 | |
a790de99 PT |
704 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
705 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 706 | /* |
8277434e PT |
707 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
708 | * node and @up when leaving it for the final time. | |
709 | * | |
710 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 711 | */ |
029632fb | 712 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 713 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
714 | { |
715 | struct task_group *parent, *child; | |
eb755805 | 716 | int ret; |
c09595f6 | 717 | |
8277434e PT |
718 | parent = from; |
719 | ||
c09595f6 | 720 | down: |
eb755805 PZ |
721 | ret = (*down)(parent, data); |
722 | if (ret) | |
8277434e | 723 | goto out; |
c09595f6 PZ |
724 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
725 | parent = child; | |
726 | goto down; | |
727 | ||
728 | up: | |
729 | continue; | |
730 | } | |
eb755805 | 731 | ret = (*up)(parent, data); |
8277434e PT |
732 | if (ret || parent == from) |
733 | goto out; | |
c09595f6 PZ |
734 | |
735 | child = parent; | |
736 | parent = parent->parent; | |
737 | if (parent) | |
738 | goto up; | |
8277434e | 739 | out: |
eb755805 | 740 | return ret; |
c09595f6 PZ |
741 | } |
742 | ||
029632fb | 743 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 744 | { |
e2b245f8 | 745 | return 0; |
eb755805 | 746 | } |
18d95a28 PZ |
747 | #endif |
748 | ||
9059393e | 749 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 750 | { |
f05998d4 NR |
751 | int prio = p->static_prio - MAX_RT_PRIO; |
752 | struct load_weight *load = &p->se.load; | |
753 | ||
dd41f596 IM |
754 | /* |
755 | * SCHED_IDLE tasks get minimal weight: | |
756 | */ | |
1da1843f | 757 | if (task_has_idle_policy(p)) { |
c8b28116 | 758 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 759 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
760 | return; |
761 | } | |
71f8bd46 | 762 | |
9059393e VG |
763 | /* |
764 | * SCHED_OTHER tasks have to update their load when changing their | |
765 | * weight | |
766 | */ | |
767 | if (update_load && p->sched_class == &fair_sched_class) { | |
768 | reweight_task(p, prio); | |
769 | } else { | |
770 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
771 | load->inv_weight = sched_prio_to_wmult[prio]; | |
772 | } | |
71f8bd46 IM |
773 | } |
774 | ||
69842cba | 775 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
776 | /* |
777 | * Serializes updates of utilization clamp values | |
778 | * | |
779 | * The (slow-path) user-space triggers utilization clamp value updates which | |
780 | * can require updates on (fast-path) scheduler's data structures used to | |
781 | * support enqueue/dequeue operations. | |
782 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
783 | * requests are serialized using a mutex to reduce the risk of conflicting | |
784 | * updates or API abuses. | |
785 | */ | |
786 | static DEFINE_MUTEX(uclamp_mutex); | |
787 | ||
e8f14172 PB |
788 | /* Max allowed minimum utilization */ |
789 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
790 | ||
791 | /* Max allowed maximum utilization */ | |
792 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
793 | ||
794 | /* All clamps are required to be less or equal than these values */ | |
795 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba PB |
796 | |
797 | /* Integer rounded range for each bucket */ | |
798 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
799 | ||
800 | #define for_each_clamp_id(clamp_id) \ | |
801 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
802 | ||
803 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
804 | { | |
805 | return clamp_value / UCLAMP_BUCKET_DELTA; | |
806 | } | |
807 | ||
60daf9c1 PB |
808 | static inline unsigned int uclamp_bucket_base_value(unsigned int clamp_value) |
809 | { | |
810 | return UCLAMP_BUCKET_DELTA * uclamp_bucket_id(clamp_value); | |
811 | } | |
812 | ||
7763baac | 813 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
814 | { |
815 | if (clamp_id == UCLAMP_MIN) | |
816 | return 0; | |
817 | return SCHED_CAPACITY_SCALE; | |
818 | } | |
819 | ||
a509a7cd PB |
820 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
821 | unsigned int value, bool user_defined) | |
69842cba PB |
822 | { |
823 | uc_se->value = value; | |
824 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 825 | uc_se->user_defined = user_defined; |
69842cba PB |
826 | } |
827 | ||
e496187d | 828 | static inline unsigned int |
0413d7f3 | 829 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
830 | unsigned int clamp_value) |
831 | { | |
832 | /* | |
833 | * Avoid blocked utilization pushing up the frequency when we go | |
834 | * idle (which drops the max-clamp) by retaining the last known | |
835 | * max-clamp. | |
836 | */ | |
837 | if (clamp_id == UCLAMP_MAX) { | |
838 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
839 | return clamp_value; | |
840 | } | |
841 | ||
842 | return uclamp_none(UCLAMP_MIN); | |
843 | } | |
844 | ||
0413d7f3 | 845 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
846 | unsigned int clamp_value) |
847 | { | |
848 | /* Reset max-clamp retention only on idle exit */ | |
849 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
850 | return; | |
851 | ||
852 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
853 | } | |
854 | ||
69842cba | 855 | static inline |
7763baac | 856 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 857 | unsigned int clamp_value) |
69842cba PB |
858 | { |
859 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
860 | int bucket_id = UCLAMP_BUCKETS - 1; | |
861 | ||
862 | /* | |
863 | * Since both min and max clamps are max aggregated, find the | |
864 | * top most bucket with tasks in. | |
865 | */ | |
866 | for ( ; bucket_id >= 0; bucket_id--) { | |
867 | if (!bucket[bucket_id].tasks) | |
868 | continue; | |
869 | return bucket[bucket_id].value; | |
870 | } | |
871 | ||
872 | /* No tasks -- default clamp values */ | |
e496187d | 873 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
874 | } |
875 | ||
3eac870a | 876 | static inline struct uclamp_se |
0413d7f3 | 877 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a PB |
878 | { |
879 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; | |
880 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
881 | struct uclamp_se uc_max; | |
882 | ||
883 | /* | |
884 | * Tasks in autogroups or root task group will be | |
885 | * restricted by system defaults. | |
886 | */ | |
887 | if (task_group_is_autogroup(task_group(p))) | |
888 | return uc_req; | |
889 | if (task_group(p) == &root_task_group) | |
890 | return uc_req; | |
891 | ||
892 | uc_max = task_group(p)->uclamp[clamp_id]; | |
893 | if (uc_req.value > uc_max.value || !uc_req.user_defined) | |
894 | return uc_max; | |
895 | #endif | |
896 | ||
897 | return uc_req; | |
898 | } | |
899 | ||
e8f14172 PB |
900 | /* |
901 | * The effective clamp bucket index of a task depends on, by increasing | |
902 | * priority: | |
903 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
904 | * - the task group effective clamp value, for tasks not either in the root |
905 | * group or in an autogroup | |
e8f14172 PB |
906 | * - the system default clamp value, defined by the sysadmin |
907 | */ | |
908 | static inline struct uclamp_se | |
0413d7f3 | 909 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 910 | { |
3eac870a | 911 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
912 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
913 | ||
914 | /* System default restrictions always apply */ | |
915 | if (unlikely(uc_req.value > uc_max.value)) | |
916 | return uc_max; | |
917 | ||
918 | return uc_req; | |
919 | } | |
920 | ||
686516b5 | 921 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
922 | { |
923 | struct uclamp_se uc_eff; | |
924 | ||
925 | /* Task currently refcounted: use back-annotated (effective) value */ | |
926 | if (p->uclamp[clamp_id].active) | |
686516b5 | 927 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
928 | |
929 | uc_eff = uclamp_eff_get(p, clamp_id); | |
930 | ||
686516b5 | 931 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
932 | } |
933 | ||
69842cba PB |
934 | /* |
935 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
936 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
937 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
938 | * |
939 | * Tasks can have a task-specific value requested from user-space, track | |
940 | * within each bucket the maximum value for tasks refcounted in it. | |
941 | * This "local max aggregation" allows to track the exact "requested" value | |
942 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
943 | */ |
944 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 945 | enum uclamp_id clamp_id) |
69842cba PB |
946 | { |
947 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
948 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
949 | struct uclamp_bucket *bucket; | |
950 | ||
951 | lockdep_assert_held(&rq->lock); | |
952 | ||
e8f14172 PB |
953 | /* Update task effective clamp */ |
954 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
955 | ||
69842cba PB |
956 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
957 | bucket->tasks++; | |
e8f14172 | 958 | uc_se->active = true; |
69842cba | 959 | |
e496187d PB |
960 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
961 | ||
60daf9c1 PB |
962 | /* |
963 | * Local max aggregation: rq buckets always track the max | |
964 | * "requested" clamp value of its RUNNABLE tasks. | |
965 | */ | |
966 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
967 | bucket->value = uc_se->value; | |
968 | ||
69842cba | 969 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 970 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
971 | } |
972 | ||
973 | /* | |
974 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
975 | * is released. If this is the last task reference counting the rq's max | |
976 | * active clamp value, then the rq's clamp value is updated. | |
977 | * | |
978 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
979 | * always valid. If it's detected they are not, as defensive programming, | |
980 | * enforce the expected state and warn. | |
981 | */ | |
982 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 983 | enum uclamp_id clamp_id) |
69842cba PB |
984 | { |
985 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
986 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
987 | struct uclamp_bucket *bucket; | |
e496187d | 988 | unsigned int bkt_clamp; |
69842cba PB |
989 | unsigned int rq_clamp; |
990 | ||
991 | lockdep_assert_held(&rq->lock); | |
992 | ||
993 | bucket = &uc_rq->bucket[uc_se->bucket_id]; | |
994 | SCHED_WARN_ON(!bucket->tasks); | |
995 | if (likely(bucket->tasks)) | |
996 | bucket->tasks--; | |
e8f14172 | 997 | uc_se->active = false; |
69842cba | 998 | |
60daf9c1 PB |
999 | /* |
1000 | * Keep "local max aggregation" simple and accept to (possibly) | |
1001 | * overboost some RUNNABLE tasks in the same bucket. | |
1002 | * The rq clamp bucket value is reset to its base value whenever | |
1003 | * there are no more RUNNABLE tasks refcounting it. | |
1004 | */ | |
69842cba PB |
1005 | if (likely(bucket->tasks)) |
1006 | return; | |
1007 | ||
1008 | rq_clamp = READ_ONCE(uc_rq->value); | |
1009 | /* | |
1010 | * Defensive programming: this should never happen. If it happens, | |
1011 | * e.g. due to future modification, warn and fixup the expected value. | |
1012 | */ | |
1013 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1014 | if (bucket->value >= rq_clamp) { |
1015 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1016 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1017 | } | |
69842cba PB |
1018 | } |
1019 | ||
1020 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1021 | { | |
0413d7f3 | 1022 | enum uclamp_id clamp_id; |
69842cba PB |
1023 | |
1024 | if (unlikely(!p->sched_class->uclamp_enabled)) | |
1025 | return; | |
1026 | ||
1027 | for_each_clamp_id(clamp_id) | |
1028 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1029 | |
1030 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1031 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1032 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1033 | } |
1034 | ||
1035 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1036 | { | |
0413d7f3 | 1037 | enum uclamp_id clamp_id; |
69842cba PB |
1038 | |
1039 | if (unlikely(!p->sched_class->uclamp_enabled)) | |
1040 | return; | |
1041 | ||
1042 | for_each_clamp_id(clamp_id) | |
1043 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1044 | } | |
1045 | ||
babbe170 | 1046 | static inline void |
0413d7f3 | 1047 | uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id) |
babbe170 PB |
1048 | { |
1049 | struct rq_flags rf; | |
1050 | struct rq *rq; | |
1051 | ||
1052 | /* | |
1053 | * Lock the task and the rq where the task is (or was) queued. | |
1054 | * | |
1055 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1056 | * price to pay to safely serialize util_{min,max} updates with | |
1057 | * enqueues, dequeues and migration operations. | |
1058 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1059 | */ | |
1060 | rq = task_rq_lock(p, &rf); | |
1061 | ||
1062 | /* | |
1063 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1064 | * If the task is not yet RUNNABLE and its task_struct is not | |
1065 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1066 | * it will already see the updated clamp bucket value. | |
1067 | */ | |
6e1ff077 | 1068 | if (p->uclamp[clamp_id].active) { |
babbe170 PB |
1069 | uclamp_rq_dec_id(rq, p, clamp_id); |
1070 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1071 | } | |
1072 | ||
1073 | task_rq_unlock(rq, p, &rf); | |
1074 | } | |
1075 | ||
e3b8b6a0 | 1076 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 PB |
1077 | static inline void |
1078 | uclamp_update_active_tasks(struct cgroup_subsys_state *css, | |
1079 | unsigned int clamps) | |
1080 | { | |
0413d7f3 | 1081 | enum uclamp_id clamp_id; |
babbe170 PB |
1082 | struct css_task_iter it; |
1083 | struct task_struct *p; | |
babbe170 PB |
1084 | |
1085 | css_task_iter_start(css, 0, &it); | |
1086 | while ((p = css_task_iter_next(&it))) { | |
1087 | for_each_clamp_id(clamp_id) { | |
1088 | if ((0x1 << clamp_id) & clamps) | |
1089 | uclamp_update_active(p, clamp_id); | |
1090 | } | |
1091 | } | |
1092 | css_task_iter_end(&it); | |
1093 | } | |
1094 | ||
7274a5c1 PB |
1095 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1096 | static void uclamp_update_root_tg(void) | |
1097 | { | |
1098 | struct task_group *tg = &root_task_group; | |
1099 | ||
1100 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1101 | sysctl_sched_uclamp_util_min, false); | |
1102 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1103 | sysctl_sched_uclamp_util_max, false); | |
1104 | ||
1105 | rcu_read_lock(); | |
1106 | cpu_util_update_eff(&root_task_group.css); | |
1107 | rcu_read_unlock(); | |
1108 | } | |
1109 | #else | |
1110 | static void uclamp_update_root_tg(void) { } | |
1111 | #endif | |
1112 | ||
e8f14172 | 1113 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1114 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1115 | { |
7274a5c1 | 1116 | bool update_root_tg = false; |
e8f14172 | 1117 | int old_min, old_max; |
e8f14172 PB |
1118 | int result; |
1119 | ||
2480c093 | 1120 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1121 | old_min = sysctl_sched_uclamp_util_min; |
1122 | old_max = sysctl_sched_uclamp_util_max; | |
1123 | ||
1124 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1125 | if (result) | |
1126 | goto undo; | |
1127 | if (!write) | |
1128 | goto done; | |
1129 | ||
1130 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
1131 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE) { | |
1132 | result = -EINVAL; | |
1133 | goto undo; | |
1134 | } | |
1135 | ||
1136 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1137 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1138 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1139 | update_root_tg = true; |
e8f14172 PB |
1140 | } |
1141 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1142 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1143 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1144 | update_root_tg = true; |
e8f14172 PB |
1145 | } |
1146 | ||
7274a5c1 PB |
1147 | if (update_root_tg) |
1148 | uclamp_update_root_tg(); | |
1149 | ||
e8f14172 | 1150 | /* |
7274a5c1 PB |
1151 | * We update all RUNNABLE tasks only when task groups are in use. |
1152 | * Otherwise, keep it simple and do just a lazy update at each next | |
1153 | * task enqueue time. | |
e8f14172 | 1154 | */ |
7274a5c1 | 1155 | |
e8f14172 PB |
1156 | goto done; |
1157 | ||
1158 | undo: | |
1159 | sysctl_sched_uclamp_util_min = old_min; | |
1160 | sysctl_sched_uclamp_util_max = old_max; | |
1161 | done: | |
2480c093 | 1162 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1163 | |
1164 | return result; | |
1165 | } | |
1166 | ||
a509a7cd PB |
1167 | static int uclamp_validate(struct task_struct *p, |
1168 | const struct sched_attr *attr) | |
1169 | { | |
1170 | unsigned int lower_bound = p->uclamp_req[UCLAMP_MIN].value; | |
1171 | unsigned int upper_bound = p->uclamp_req[UCLAMP_MAX].value; | |
1172 | ||
1173 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) | |
1174 | lower_bound = attr->sched_util_min; | |
1175 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) | |
1176 | upper_bound = attr->sched_util_max; | |
1177 | ||
1178 | if (lower_bound > upper_bound) | |
1179 | return -EINVAL; | |
1180 | if (upper_bound > SCHED_CAPACITY_SCALE) | |
1181 | return -EINVAL; | |
1182 | ||
1183 | return 0; | |
1184 | } | |
1185 | ||
1186 | static void __setscheduler_uclamp(struct task_struct *p, | |
1187 | const struct sched_attr *attr) | |
1188 | { | |
0413d7f3 | 1189 | enum uclamp_id clamp_id; |
1a00d999 PB |
1190 | |
1191 | /* | |
1192 | * On scheduling class change, reset to default clamps for tasks | |
1193 | * without a task-specific value. | |
1194 | */ | |
1195 | for_each_clamp_id(clamp_id) { | |
1196 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
1197 | unsigned int clamp_value = uclamp_none(clamp_id); | |
1198 | ||
1199 | /* Keep using defined clamps across class changes */ | |
1200 | if (uc_se->user_defined) | |
1201 | continue; | |
1202 | ||
1203 | /* By default, RT tasks always get 100% boost */ | |
1204 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) | |
1205 | clamp_value = uclamp_none(UCLAMP_MAX); | |
1206 | ||
1207 | uclamp_se_set(uc_se, clamp_value, false); | |
1208 | } | |
1209 | ||
a509a7cd PB |
1210 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1211 | return; | |
1212 | ||
1213 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { | |
1214 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], | |
1215 | attr->sched_util_min, true); | |
1216 | } | |
1217 | ||
1218 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1219 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], | |
1220 | attr->sched_util_max, true); | |
1221 | } | |
1222 | } | |
1223 | ||
e8f14172 PB |
1224 | static void uclamp_fork(struct task_struct *p) |
1225 | { | |
0413d7f3 | 1226 | enum uclamp_id clamp_id; |
e8f14172 PB |
1227 | |
1228 | for_each_clamp_id(clamp_id) | |
1229 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1230 | |
1231 | if (likely(!p->sched_reset_on_fork)) | |
1232 | return; | |
1233 | ||
1234 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1235 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1236 | uclamp_none(clamp_id), false); | |
a87498ac | 1237 | } |
e8f14172 PB |
1238 | } |
1239 | ||
69842cba PB |
1240 | static void __init init_uclamp(void) |
1241 | { | |
e8f14172 | 1242 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1243 | enum uclamp_id clamp_id; |
69842cba PB |
1244 | int cpu; |
1245 | ||
2480c093 PB |
1246 | mutex_init(&uclamp_mutex); |
1247 | ||
e496187d | 1248 | for_each_possible_cpu(cpu) { |
dcd6dffb LG |
1249 | memset(&cpu_rq(cpu)->uclamp, 0, |
1250 | sizeof(struct uclamp_rq)*UCLAMP_CNT); | |
e496187d PB |
1251 | cpu_rq(cpu)->uclamp_flags = 0; |
1252 | } | |
69842cba | 1253 | |
69842cba | 1254 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1255 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1256 | uclamp_none(clamp_id), false); |
69842cba | 1257 | } |
e8f14172 PB |
1258 | |
1259 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1260 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1261 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1262 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1263 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1264 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1265 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1266 | #endif |
1267 | } | |
69842cba PB |
1268 | } |
1269 | ||
1270 | #else /* CONFIG_UCLAMP_TASK */ | |
1271 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1272 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1273 | static inline int uclamp_validate(struct task_struct *p, |
1274 | const struct sched_attr *attr) | |
1275 | { | |
1276 | return -EOPNOTSUPP; | |
1277 | } | |
1278 | static void __setscheduler_uclamp(struct task_struct *p, | |
1279 | const struct sched_attr *attr) { } | |
e8f14172 | 1280 | static inline void uclamp_fork(struct task_struct *p) { } |
69842cba PB |
1281 | static inline void init_uclamp(void) { } |
1282 | #endif /* CONFIG_UCLAMP_TASK */ | |
1283 | ||
1de64443 | 1284 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1285 | { |
0a67d1ee PZ |
1286 | if (!(flags & ENQUEUE_NOCLOCK)) |
1287 | update_rq_clock(rq); | |
1288 | ||
eb414681 | 1289 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 1290 | sched_info_queued(rq, p); |
eb414681 JW |
1291 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1292 | } | |
0a67d1ee | 1293 | |
69842cba | 1294 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1295 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1296 | } |
1297 | ||
1de64443 | 1298 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1299 | { |
0a67d1ee PZ |
1300 | if (!(flags & DEQUEUE_NOCLOCK)) |
1301 | update_rq_clock(rq); | |
1302 | ||
eb414681 | 1303 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 1304 | sched_info_dequeued(rq, p); |
eb414681 JW |
1305 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1306 | } | |
0a67d1ee | 1307 | |
69842cba | 1308 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1309 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1310 | } |
1311 | ||
029632fb | 1312 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1313 | { |
1314 | if (task_contributes_to_load(p)) | |
1315 | rq->nr_uninterruptible--; | |
1316 | ||
371fd7e7 | 1317 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
1318 | |
1319 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
1320 | } |
1321 | ||
029632fb | 1322 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1323 | { |
7dd77884 PZ |
1324 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
1325 | ||
1e3c88bd PZ |
1326 | if (task_contributes_to_load(p)) |
1327 | rq->nr_uninterruptible++; | |
1328 | ||
371fd7e7 | 1329 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1330 | } |
1331 | ||
14531189 | 1332 | /* |
dd41f596 | 1333 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1334 | */ |
14531189 IM |
1335 | static inline int __normal_prio(struct task_struct *p) |
1336 | { | |
dd41f596 | 1337 | return p->static_prio; |
14531189 IM |
1338 | } |
1339 | ||
b29739f9 IM |
1340 | /* |
1341 | * Calculate the expected normal priority: i.e. priority | |
1342 | * without taking RT-inheritance into account. Might be | |
1343 | * boosted by interactivity modifiers. Changes upon fork, | |
1344 | * setprio syscalls, and whenever the interactivity | |
1345 | * estimator recalculates. | |
1346 | */ | |
36c8b586 | 1347 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1348 | { |
1349 | int prio; | |
1350 | ||
aab03e05 DF |
1351 | if (task_has_dl_policy(p)) |
1352 | prio = MAX_DL_PRIO-1; | |
1353 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
1354 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1355 | else | |
1356 | prio = __normal_prio(p); | |
1357 | return prio; | |
1358 | } | |
1359 | ||
1360 | /* | |
1361 | * Calculate the current priority, i.e. the priority | |
1362 | * taken into account by the scheduler. This value might | |
1363 | * be boosted by RT tasks, or might be boosted by | |
1364 | * interactivity modifiers. Will be RT if the task got | |
1365 | * RT-boosted. If not then it returns p->normal_prio. | |
1366 | */ | |
36c8b586 | 1367 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1368 | { |
1369 | p->normal_prio = normal_prio(p); | |
1370 | /* | |
1371 | * If we are RT tasks or we were boosted to RT priority, | |
1372 | * keep the priority unchanged. Otherwise, update priority | |
1373 | * to the normal priority: | |
1374 | */ | |
1375 | if (!rt_prio(p->prio)) | |
1376 | return p->normal_prio; | |
1377 | return p->prio; | |
1378 | } | |
1379 | ||
1da177e4 LT |
1380 | /** |
1381 | * task_curr - is this task currently executing on a CPU? | |
1382 | * @p: the task in question. | |
e69f6186 YB |
1383 | * |
1384 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1385 | */ |
36c8b586 | 1386 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1387 | { |
1388 | return cpu_curr(task_cpu(p)) == p; | |
1389 | } | |
1390 | ||
67dfa1b7 | 1391 | /* |
4c9a4bc8 PZ |
1392 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1393 | * use the balance_callback list if you want balancing. | |
1394 | * | |
1395 | * this means any call to check_class_changed() must be followed by a call to | |
1396 | * balance_callback(). | |
67dfa1b7 | 1397 | */ |
cb469845 SR |
1398 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1399 | const struct sched_class *prev_class, | |
da7a735e | 1400 | int oldprio) |
cb469845 SR |
1401 | { |
1402 | if (prev_class != p->sched_class) { | |
1403 | if (prev_class->switched_from) | |
da7a735e | 1404 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1405 | |
da7a735e | 1406 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1407 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1408 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1409 | } |
1410 | ||
029632fb | 1411 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 PZ |
1412 | { |
1413 | const struct sched_class *class; | |
1414 | ||
1415 | if (p->sched_class == rq->curr->sched_class) { | |
1416 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1417 | } else { | |
1418 | for_each_class(class) { | |
1419 | if (class == rq->curr->sched_class) | |
1420 | break; | |
1421 | if (class == p->sched_class) { | |
8875125e | 1422 | resched_curr(rq); |
1e5a7405 PZ |
1423 | break; |
1424 | } | |
1425 | } | |
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * A queue event has occurred, and we're going to schedule. In | |
1430 | * this case, we can save a useless back to back clock update. | |
1431 | */ | |
da0c1e65 | 1432 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 1433 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
1434 | } |
1435 | ||
1da177e4 | 1436 | #ifdef CONFIG_SMP |
175f0e25 | 1437 | |
175f0e25 | 1438 | /* |
bee98539 | 1439 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
1440 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
1441 | */ | |
1442 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
1443 | { | |
3bd37062 | 1444 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
1445 | return false; |
1446 | ||
1447 | if (is_per_cpu_kthread(p)) | |
1448 | return cpu_online(cpu); | |
1449 | ||
1450 | return cpu_active(cpu); | |
1451 | } | |
1452 | ||
5cc389bc PZ |
1453 | /* |
1454 | * This is how migration works: | |
1455 | * | |
1456 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1457 | * stop_one_cpu(). | |
1458 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1459 | * off the CPU) | |
1460 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1461 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1462 | * it and puts it into the right queue. | |
1463 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1464 | * is done. | |
1465 | */ | |
1466 | ||
1467 | /* | |
1468 | * move_queued_task - move a queued task to new rq. | |
1469 | * | |
1470 | * Returns (locked) new rq. Old rq's lock is released. | |
1471 | */ | |
8a8c69c3 PZ |
1472 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
1473 | struct task_struct *p, int new_cpu) | |
5cc389bc | 1474 | { |
5cc389bc PZ |
1475 | lockdep_assert_held(&rq->lock); |
1476 | ||
c546951d | 1477 | WRITE_ONCE(p->on_rq, TASK_ON_RQ_MIGRATING); |
15ff991e | 1478 | dequeue_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 1479 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 1480 | rq_unlock(rq, rf); |
5cc389bc PZ |
1481 | |
1482 | rq = cpu_rq(new_cpu); | |
1483 | ||
8a8c69c3 | 1484 | rq_lock(rq, rf); |
5cc389bc | 1485 | BUG_ON(task_cpu(p) != new_cpu); |
5cc389bc | 1486 | enqueue_task(rq, p, 0); |
3ea94de1 | 1487 | p->on_rq = TASK_ON_RQ_QUEUED; |
5cc389bc PZ |
1488 | check_preempt_curr(rq, p, 0); |
1489 | ||
1490 | return rq; | |
1491 | } | |
1492 | ||
1493 | struct migration_arg { | |
1494 | struct task_struct *task; | |
1495 | int dest_cpu; | |
1496 | }; | |
1497 | ||
1498 | /* | |
d1ccc66d | 1499 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
1500 | * this because either it can't run here any more (set_cpus_allowed() |
1501 | * away from this CPU, or CPU going down), or because we're | |
1502 | * attempting to rebalance this task on exec (sched_exec). | |
1503 | * | |
1504 | * So we race with normal scheduler movements, but that's OK, as long | |
1505 | * as the task is no longer on this CPU. | |
5cc389bc | 1506 | */ |
8a8c69c3 PZ |
1507 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
1508 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 1509 | { |
5cc389bc | 1510 | /* Affinity changed (again). */ |
175f0e25 | 1511 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 1512 | return rq; |
5cc389bc | 1513 | |
15ff991e | 1514 | update_rq_clock(rq); |
8a8c69c3 | 1515 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
1516 | |
1517 | return rq; | |
5cc389bc PZ |
1518 | } |
1519 | ||
1520 | /* | |
1521 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1522 | * and performs thread migration by bumping thread off CPU then | |
1523 | * 'pushing' onto another runqueue. | |
1524 | */ | |
1525 | static int migration_cpu_stop(void *data) | |
1526 | { | |
1527 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1528 | struct task_struct *p = arg->task; |
1529 | struct rq *rq = this_rq(); | |
8a8c69c3 | 1530 | struct rq_flags rf; |
5cc389bc PZ |
1531 | |
1532 | /* | |
d1ccc66d IM |
1533 | * The original target CPU might have gone down and we might |
1534 | * be on another CPU but it doesn't matter. | |
5cc389bc PZ |
1535 | */ |
1536 | local_irq_disable(); | |
1537 | /* | |
1538 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 1539 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
1540 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
1541 | */ | |
a1488664 | 1542 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
1543 | |
1544 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1545 | rq_lock(rq, &rf); |
5e16bbc2 PZ |
1546 | /* |
1547 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1548 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1549 | * we're holding p->pi_lock. | |
1550 | */ | |
bf89a304 CC |
1551 | if (task_rq(p) == rq) { |
1552 | if (task_on_rq_queued(p)) | |
8a8c69c3 | 1553 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 CC |
1554 | else |
1555 | p->wake_cpu = arg->dest_cpu; | |
1556 | } | |
8a8c69c3 | 1557 | rq_unlock(rq, &rf); |
5e16bbc2 PZ |
1558 | raw_spin_unlock(&p->pi_lock); |
1559 | ||
5cc389bc PZ |
1560 | local_irq_enable(); |
1561 | return 0; | |
1562 | } | |
1563 | ||
c5b28038 PZ |
1564 | /* |
1565 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1566 | * actually call this function. | |
1567 | */ | |
1568 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1569 | { |
3bd37062 | 1570 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
1571 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1572 | } | |
1573 | ||
c5b28038 PZ |
1574 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1575 | { | |
6c37067e PZ |
1576 | struct rq *rq = task_rq(p); |
1577 | bool queued, running; | |
1578 | ||
c5b28038 | 1579 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1580 | |
1581 | queued = task_on_rq_queued(p); | |
1582 | running = task_current(rq, p); | |
1583 | ||
1584 | if (queued) { | |
1585 | /* | |
1586 | * Because __kthread_bind() calls this on blocked tasks without | |
1587 | * holding rq->lock. | |
1588 | */ | |
1589 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 1590 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
1591 | } |
1592 | if (running) | |
1593 | put_prev_task(rq, p); | |
1594 | ||
c5b28038 | 1595 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e | 1596 | |
6c37067e | 1597 | if (queued) |
7134b3e9 | 1598 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 1599 | if (running) |
03b7fad1 | 1600 | set_next_task(rq, p); |
c5b28038 PZ |
1601 | } |
1602 | ||
5cc389bc PZ |
1603 | /* |
1604 | * Change a given task's CPU affinity. Migrate the thread to a | |
1605 | * proper CPU and schedule it away if the CPU it's executing on | |
1606 | * is removed from the allowed bitmask. | |
1607 | * | |
1608 | * NOTE: the caller must have a valid reference to the task, the | |
1609 | * task must not exit() & deallocate itself prematurely. The | |
1610 | * call is not atomic; no spinlocks may be held. | |
1611 | */ | |
25834c73 PZ |
1612 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1613 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1614 | { |
e9d867a6 | 1615 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1616 | unsigned int dest_cpu; |
eb580751 PZ |
1617 | struct rq_flags rf; |
1618 | struct rq *rq; | |
5cc389bc PZ |
1619 | int ret = 0; |
1620 | ||
eb580751 | 1621 | rq = task_rq_lock(p, &rf); |
a499c3ea | 1622 | update_rq_clock(rq); |
5cc389bc | 1623 | |
e9d867a6 PZI |
1624 | if (p->flags & PF_KTHREAD) { |
1625 | /* | |
1626 | * Kernel threads are allowed on online && !active CPUs | |
1627 | */ | |
1628 | cpu_valid_mask = cpu_online_mask; | |
1629 | } | |
1630 | ||
25834c73 PZ |
1631 | /* |
1632 | * Must re-check here, to close a race against __kthread_bind(), | |
1633 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1634 | */ | |
1635 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1636 | ret = -EINVAL; | |
1637 | goto out; | |
1638 | } | |
1639 | ||
3bd37062 | 1640 | if (cpumask_equal(p->cpus_ptr, new_mask)) |
5cc389bc PZ |
1641 | goto out; |
1642 | ||
46a87b38 PT |
1643 | /* |
1644 | * Picking a ~random cpu helps in cases where we are changing affinity | |
1645 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
1646 | * immediately required to distribute the tasks within their new mask. | |
1647 | */ | |
1648 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 1649 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
1650 | ret = -EINVAL; |
1651 | goto out; | |
1652 | } | |
1653 | ||
1654 | do_set_cpus_allowed(p, new_mask); | |
1655 | ||
e9d867a6 PZI |
1656 | if (p->flags & PF_KTHREAD) { |
1657 | /* | |
1658 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 1659 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
1660 | */ |
1661 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1662 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1663 | p->nr_cpus_allowed != 1); | |
1664 | } | |
1665 | ||
5cc389bc PZ |
1666 | /* Can the task run on the task's current CPU? If so, we're done */ |
1667 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1668 | goto out; | |
1669 | ||
5cc389bc PZ |
1670 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1671 | struct migration_arg arg = { p, dest_cpu }; | |
1672 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1673 | task_rq_unlock(rq, p, &rf); |
5cc389bc | 1674 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
5cc389bc | 1675 | return 0; |
cbce1a68 PZ |
1676 | } else if (task_on_rq_queued(p)) { |
1677 | /* | |
1678 | * OK, since we're going to drop the lock immediately | |
1679 | * afterwards anyway. | |
1680 | */ | |
8a8c69c3 | 1681 | rq = move_queued_task(rq, &rf, p, dest_cpu); |
cbce1a68 | 1682 | } |
5cc389bc | 1683 | out: |
eb580751 | 1684 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1685 | |
1686 | return ret; | |
1687 | } | |
25834c73 PZ |
1688 | |
1689 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1690 | { | |
1691 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1692 | } | |
5cc389bc PZ |
1693 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1694 | ||
dd41f596 | 1695 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1696 | { |
e2912009 PZ |
1697 | #ifdef CONFIG_SCHED_DEBUG |
1698 | /* | |
1699 | * We should never call set_task_cpu() on a blocked task, | |
1700 | * ttwu() will sort out the placement. | |
1701 | */ | |
077614ee | 1702 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1703 | !p->on_rq); |
0122ec5b | 1704 | |
3ea94de1 JP |
1705 | /* |
1706 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1707 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1708 | * time relying on p->on_rq. | |
1709 | */ | |
1710 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1711 | p->sched_class == &fair_sched_class && | |
1712 | (p->on_rq && !task_on_rq_migrating(p))); | |
1713 | ||
0122ec5b | 1714 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1715 | /* |
1716 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1717 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1718 | * | |
1719 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1720 | * see task_group(). |
6c6c54e1 PZ |
1721 | * |
1722 | * Furthermore, all task_rq users should acquire both locks, see | |
1723 | * task_rq_lock(). | |
1724 | */ | |
0122ec5b PZ |
1725 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1726 | lockdep_is_held(&task_rq(p)->lock))); | |
1727 | #endif | |
4ff9083b PZ |
1728 | /* |
1729 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
1730 | */ | |
1731 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
e2912009 PZ |
1732 | #endif |
1733 | ||
de1d7286 | 1734 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1735 | |
0c69774e | 1736 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 1737 | if (p->sched_class->migrate_task_rq) |
1327237a | 1738 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 1739 | p->se.nr_migrations++; |
d7822b1e | 1740 | rseq_migrate(p); |
ff303e66 | 1741 | perf_event_task_migrate(p); |
0c69774e | 1742 | } |
dd41f596 IM |
1743 | |
1744 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1745 | } |
1746 | ||
0ad4e3df | 1747 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
1748 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
1749 | { | |
da0c1e65 | 1750 | if (task_on_rq_queued(p)) { |
ac66f547 | 1751 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 1752 | struct rq_flags srf, drf; |
ac66f547 PZ |
1753 | |
1754 | src_rq = task_rq(p); | |
1755 | dst_rq = cpu_rq(cpu); | |
1756 | ||
8a8c69c3 PZ |
1757 | rq_pin_lock(src_rq, &srf); |
1758 | rq_pin_lock(dst_rq, &drf); | |
1759 | ||
ac66f547 PZ |
1760 | deactivate_task(src_rq, p, 0); |
1761 | set_task_cpu(p, cpu); | |
1762 | activate_task(dst_rq, p, 0); | |
1763 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
1764 | |
1765 | rq_unpin_lock(dst_rq, &drf); | |
1766 | rq_unpin_lock(src_rq, &srf); | |
1767 | ||
ac66f547 PZ |
1768 | } else { |
1769 | /* | |
1770 | * Task isn't running anymore; make it appear like we migrated | |
1771 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 1772 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
1773 | */ |
1774 | p->wake_cpu = cpu; | |
1775 | } | |
1776 | } | |
1777 | ||
1778 | struct migration_swap_arg { | |
1779 | struct task_struct *src_task, *dst_task; | |
1780 | int src_cpu, dst_cpu; | |
1781 | }; | |
1782 | ||
1783 | static int migrate_swap_stop(void *data) | |
1784 | { | |
1785 | struct migration_swap_arg *arg = data; | |
1786 | struct rq *src_rq, *dst_rq; | |
1787 | int ret = -EAGAIN; | |
1788 | ||
62694cd5 PZ |
1789 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
1790 | return -EAGAIN; | |
1791 | ||
ac66f547 PZ |
1792 | src_rq = cpu_rq(arg->src_cpu); |
1793 | dst_rq = cpu_rq(arg->dst_cpu); | |
1794 | ||
74602315 PZ |
1795 | double_raw_lock(&arg->src_task->pi_lock, |
1796 | &arg->dst_task->pi_lock); | |
ac66f547 | 1797 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 1798 | |
ac66f547 PZ |
1799 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
1800 | goto unlock; | |
1801 | ||
1802 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
1803 | goto unlock; | |
1804 | ||
3bd37062 | 1805 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
1806 | goto unlock; |
1807 | ||
3bd37062 | 1808 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
1809 | goto unlock; |
1810 | ||
1811 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
1812 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
1813 | ||
1814 | ret = 0; | |
1815 | ||
1816 | unlock: | |
1817 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
1818 | raw_spin_unlock(&arg->dst_task->pi_lock); |
1819 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
1820 | |
1821 | return ret; | |
1822 | } | |
1823 | ||
1824 | /* | |
1825 | * Cross migrate two tasks | |
1826 | */ | |
0ad4e3df SD |
1827 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
1828 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
1829 | { |
1830 | struct migration_swap_arg arg; | |
1831 | int ret = -EINVAL; | |
1832 | ||
ac66f547 PZ |
1833 | arg = (struct migration_swap_arg){ |
1834 | .src_task = cur, | |
0ad4e3df | 1835 | .src_cpu = curr_cpu, |
ac66f547 | 1836 | .dst_task = p, |
0ad4e3df | 1837 | .dst_cpu = target_cpu, |
ac66f547 PZ |
1838 | }; |
1839 | ||
1840 | if (arg.src_cpu == arg.dst_cpu) | |
1841 | goto out; | |
1842 | ||
6acce3ef PZ |
1843 | /* |
1844 | * These three tests are all lockless; this is OK since all of them | |
1845 | * will be re-checked with proper locks held further down the line. | |
1846 | */ | |
ac66f547 PZ |
1847 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
1848 | goto out; | |
1849 | ||
3bd37062 | 1850 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
1851 | goto out; |
1852 | ||
3bd37062 | 1853 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
1854 | goto out; |
1855 | ||
286549dc | 1856 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
1857 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
1858 | ||
1859 | out: | |
ac66f547 PZ |
1860 | return ret; |
1861 | } | |
0ad4e3df | 1862 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 1863 | |
1da177e4 LT |
1864 | /* |
1865 | * wait_task_inactive - wait for a thread to unschedule. | |
1866 | * | |
85ba2d86 RM |
1867 | * If @match_state is nonzero, it's the @p->state value just checked and |
1868 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1869 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1870 | * we return a positive number (its total switch count). If a second call | |
1871 | * a short while later returns the same number, the caller can be sure that | |
1872 | * @p has remained unscheduled the whole time. | |
1873 | * | |
1da177e4 LT |
1874 | * The caller must ensure that the task *will* unschedule sometime soon, |
1875 | * else this function might spin for a *long* time. This function can't | |
1876 | * be called with interrupts off, or it may introduce deadlock with | |
1877 | * smp_call_function() if an IPI is sent by the same process we are | |
1878 | * waiting to become inactive. | |
1879 | */ | |
85ba2d86 | 1880 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 1881 | { |
da0c1e65 | 1882 | int running, queued; |
eb580751 | 1883 | struct rq_flags rf; |
85ba2d86 | 1884 | unsigned long ncsw; |
70b97a7f | 1885 | struct rq *rq; |
1da177e4 | 1886 | |
3a5c359a AK |
1887 | for (;;) { |
1888 | /* | |
1889 | * We do the initial early heuristics without holding | |
1890 | * any task-queue locks at all. We'll only try to get | |
1891 | * the runqueue lock when things look like they will | |
1892 | * work out! | |
1893 | */ | |
1894 | rq = task_rq(p); | |
fa490cfd | 1895 | |
3a5c359a AK |
1896 | /* |
1897 | * If the task is actively running on another CPU | |
1898 | * still, just relax and busy-wait without holding | |
1899 | * any locks. | |
1900 | * | |
1901 | * NOTE! Since we don't hold any locks, it's not | |
1902 | * even sure that "rq" stays as the right runqueue! | |
1903 | * But we don't care, since "task_running()" will | |
1904 | * return false if the runqueue has changed and p | |
1905 | * is actually now running somewhere else! | |
1906 | */ | |
85ba2d86 RM |
1907 | while (task_running(rq, p)) { |
1908 | if (match_state && unlikely(p->state != match_state)) | |
1909 | return 0; | |
3a5c359a | 1910 | cpu_relax(); |
85ba2d86 | 1911 | } |
fa490cfd | 1912 | |
3a5c359a AK |
1913 | /* |
1914 | * Ok, time to look more closely! We need the rq | |
1915 | * lock now, to be *sure*. If we're wrong, we'll | |
1916 | * just go back and repeat. | |
1917 | */ | |
eb580751 | 1918 | rq = task_rq_lock(p, &rf); |
27a9da65 | 1919 | trace_sched_wait_task(p); |
3a5c359a | 1920 | running = task_running(rq, p); |
da0c1e65 | 1921 | queued = task_on_rq_queued(p); |
85ba2d86 | 1922 | ncsw = 0; |
f31e11d8 | 1923 | if (!match_state || p->state == match_state) |
93dcf55f | 1924 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 1925 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 1926 | |
85ba2d86 RM |
1927 | /* |
1928 | * If it changed from the expected state, bail out now. | |
1929 | */ | |
1930 | if (unlikely(!ncsw)) | |
1931 | break; | |
1932 | ||
3a5c359a AK |
1933 | /* |
1934 | * Was it really running after all now that we | |
1935 | * checked with the proper locks actually held? | |
1936 | * | |
1937 | * Oops. Go back and try again.. | |
1938 | */ | |
1939 | if (unlikely(running)) { | |
1940 | cpu_relax(); | |
1941 | continue; | |
1942 | } | |
fa490cfd | 1943 | |
3a5c359a AK |
1944 | /* |
1945 | * It's not enough that it's not actively running, | |
1946 | * it must be off the runqueue _entirely_, and not | |
1947 | * preempted! | |
1948 | * | |
80dd99b3 | 1949 | * So if it was still runnable (but just not actively |
3a5c359a AK |
1950 | * running right now), it's preempted, and we should |
1951 | * yield - it could be a while. | |
1952 | */ | |
da0c1e65 | 1953 | if (unlikely(queued)) { |
8b0e1953 | 1954 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
1955 | |
1956 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1957 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
1958 | continue; |
1959 | } | |
fa490cfd | 1960 | |
3a5c359a AK |
1961 | /* |
1962 | * Ahh, all good. It wasn't running, and it wasn't | |
1963 | * runnable, which means that it will never become | |
1964 | * running in the future either. We're all done! | |
1965 | */ | |
1966 | break; | |
1967 | } | |
85ba2d86 RM |
1968 | |
1969 | return ncsw; | |
1da177e4 LT |
1970 | } |
1971 | ||
1972 | /*** | |
1973 | * kick_process - kick a running thread to enter/exit the kernel | |
1974 | * @p: the to-be-kicked thread | |
1975 | * | |
1976 | * Cause a process which is running on another CPU to enter | |
1977 | * kernel-mode, without any delay. (to get signals handled.) | |
1978 | * | |
25985edc | 1979 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
1980 | * because all it wants to ensure is that the remote task enters |
1981 | * the kernel. If the IPI races and the task has been migrated | |
1982 | * to another CPU then no harm is done and the purpose has been | |
1983 | * achieved as well. | |
1984 | */ | |
36c8b586 | 1985 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1986 | { |
1987 | int cpu; | |
1988 | ||
1989 | preempt_disable(); | |
1990 | cpu = task_cpu(p); | |
1991 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1992 | smp_send_reschedule(cpu); | |
1993 | preempt_enable(); | |
1994 | } | |
b43e3521 | 1995 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 1996 | |
30da688e | 1997 | /* |
3bd37062 | 1998 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
1999 | * |
2000 | * A few notes on cpu_active vs cpu_online: | |
2001 | * | |
2002 | * - cpu_active must be a subset of cpu_online | |
2003 | * | |
97fb7a0a | 2004 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 2005 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 2006 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
2007 | * see it. |
2008 | * | |
d1ccc66d | 2009 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 2010 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 2011 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
2012 | * off. |
2013 | * | |
2014 | * This means that fallback selection must not select !active CPUs. | |
2015 | * And can assume that any active CPU must be online. Conversely | |
2016 | * select_task_rq() below may allow selection of !active CPUs in order | |
2017 | * to satisfy the above rules. | |
30da688e | 2018 | */ |
5da9a0fb PZ |
2019 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2020 | { | |
aa00d89c TC |
2021 | int nid = cpu_to_node(cpu); |
2022 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
2023 | enum { cpuset, possible, fail } state = cpuset; |
2024 | int dest_cpu; | |
5da9a0fb | 2025 | |
aa00d89c | 2026 | /* |
d1ccc66d IM |
2027 | * If the node that the CPU is on has been offlined, cpu_to_node() |
2028 | * will return -1. There is no CPU on the node, and we should | |
2029 | * select the CPU on the other node. | |
aa00d89c TC |
2030 | */ |
2031 | if (nid != -1) { | |
2032 | nodemask = cpumask_of_node(nid); | |
2033 | ||
2034 | /* Look for allowed, online CPU in same node. */ | |
2035 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
2036 | if (!cpu_active(dest_cpu)) |
2037 | continue; | |
3bd37062 | 2038 | if (cpumask_test_cpu(dest_cpu, p->cpus_ptr)) |
aa00d89c TC |
2039 | return dest_cpu; |
2040 | } | |
2baab4e9 | 2041 | } |
5da9a0fb | 2042 | |
2baab4e9 PZ |
2043 | for (;;) { |
2044 | /* Any allowed, online CPU? */ | |
3bd37062 | 2045 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 2046 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 2047 | continue; |
175f0e25 | 2048 | |
2baab4e9 PZ |
2049 | goto out; |
2050 | } | |
5da9a0fb | 2051 | |
e73e85f0 | 2052 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
2053 | switch (state) { |
2054 | case cpuset: | |
e73e85f0 ON |
2055 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
2056 | cpuset_cpus_allowed_fallback(p); | |
2057 | state = possible; | |
2058 | break; | |
2059 | } | |
d1ccc66d | 2060 | /* Fall-through */ |
2baab4e9 PZ |
2061 | case possible: |
2062 | do_set_cpus_allowed(p, cpu_possible_mask); | |
2063 | state = fail; | |
2064 | break; | |
2065 | ||
2066 | case fail: | |
2067 | BUG(); | |
2068 | break; | |
2069 | } | |
2070 | } | |
2071 | ||
2072 | out: | |
2073 | if (state != cpuset) { | |
2074 | /* | |
2075 | * Don't tell them about moving exiting tasks or | |
2076 | * kernel threads (both mm NULL), since they never | |
2077 | * leave kernel. | |
2078 | */ | |
2079 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 2080 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
2081 | task_pid_nr(p), p->comm, cpu); |
2082 | } | |
5da9a0fb PZ |
2083 | } |
2084 | ||
2085 | return dest_cpu; | |
2086 | } | |
2087 | ||
e2912009 | 2088 | /* |
3bd37062 | 2089 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 2090 | */ |
970b13ba | 2091 | static inline |
ac66f547 | 2092 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 2093 | { |
cbce1a68 PZ |
2094 | lockdep_assert_held(&p->pi_lock); |
2095 | ||
4b53a341 | 2096 | if (p->nr_cpus_allowed > 1) |
6c1d9410 | 2097 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 | 2098 | else |
3bd37062 | 2099 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
2100 | |
2101 | /* | |
2102 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 2103 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 2104 | * CPU. |
e2912009 PZ |
2105 | * |
2106 | * Since this is common to all placement strategies, this lives here. | |
2107 | * | |
2108 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2109 | * not worry about this generic constraint ] | |
2110 | */ | |
7af443ee | 2111 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 2112 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2113 | |
2114 | return cpu; | |
970b13ba | 2115 | } |
09a40af5 | 2116 | |
f5832c19 NP |
2117 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2118 | { | |
2119 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2120 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2121 | ||
2122 | if (stop) { | |
2123 | /* | |
2124 | * Make it appear like a SCHED_FIFO task, its something | |
2125 | * userspace knows about and won't get confused about. | |
2126 | * | |
2127 | * Also, it will make PI more or less work without too | |
2128 | * much confusion -- but then, stop work should not | |
2129 | * rely on PI working anyway. | |
2130 | */ | |
2131 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2132 | ||
2133 | stop->sched_class = &stop_sched_class; | |
2134 | } | |
2135 | ||
2136 | cpu_rq(cpu)->stop = stop; | |
2137 | ||
2138 | if (old_stop) { | |
2139 | /* | |
2140 | * Reset it back to a normal scheduling class so that | |
2141 | * it can die in pieces. | |
2142 | */ | |
2143 | old_stop->sched_class = &rt_sched_class; | |
2144 | } | |
2145 | } | |
2146 | ||
25834c73 PZ |
2147 | #else |
2148 | ||
2149 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
2150 | const struct cpumask *new_mask, bool check) | |
2151 | { | |
2152 | return set_cpus_allowed_ptr(p, new_mask); | |
2153 | } | |
2154 | ||
5cc389bc | 2155 | #endif /* CONFIG_SMP */ |
970b13ba | 2156 | |
d7c01d27 | 2157 | static void |
b84cb5df | 2158 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2159 | { |
4fa8d299 | 2160 | struct rq *rq; |
b84cb5df | 2161 | |
4fa8d299 JP |
2162 | if (!schedstat_enabled()) |
2163 | return; | |
2164 | ||
2165 | rq = this_rq(); | |
d7c01d27 | 2166 | |
4fa8d299 JP |
2167 | #ifdef CONFIG_SMP |
2168 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
2169 | __schedstat_inc(rq->ttwu_local); |
2170 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
2171 | } else { |
2172 | struct sched_domain *sd; | |
2173 | ||
b85c8b71 | 2174 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 2175 | rcu_read_lock(); |
4fa8d299 | 2176 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 2177 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 2178 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
2179 | break; |
2180 | } | |
2181 | } | |
057f3fad | 2182 | rcu_read_unlock(); |
d7c01d27 | 2183 | } |
f339b9dc PZ |
2184 | |
2185 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 2186 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
2187 | #endif /* CONFIG_SMP */ |
2188 | ||
b85c8b71 PZ |
2189 | __schedstat_inc(rq->ttwu_count); |
2190 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
2191 | |
2192 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 2193 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2194 | } |
2195 | ||
23f41eeb PZ |
2196 | /* |
2197 | * Mark the task runnable and perform wakeup-preemption. | |
2198 | */ | |
e7904a28 | 2199 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2200 | struct rq_flags *rf) |
9ed3811a | 2201 | { |
9ed3811a | 2202 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 2203 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
2204 | trace_sched_wakeup(p); |
2205 | ||
9ed3811a | 2206 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
2207 | if (p->sched_class->task_woken) { |
2208 | /* | |
cbce1a68 PZ |
2209 | * Our task @p is fully woken up and running; so its safe to |
2210 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 2211 | */ |
d8ac8971 | 2212 | rq_unpin_lock(rq, rf); |
9ed3811a | 2213 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2214 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 2215 | } |
9ed3811a | 2216 | |
e69c6341 | 2217 | if (rq->idle_stamp) { |
78becc27 | 2218 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 2219 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 2220 | |
abfafa54 JL |
2221 | update_avg(&rq->avg_idle, delta); |
2222 | ||
2223 | if (rq->avg_idle > max) | |
9ed3811a | 2224 | rq->avg_idle = max; |
abfafa54 | 2225 | |
9ed3811a TH |
2226 | rq->idle_stamp = 0; |
2227 | } | |
2228 | #endif | |
2229 | } | |
2230 | ||
c05fbafb | 2231 | static void |
e7904a28 | 2232 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2233 | struct rq_flags *rf) |
c05fbafb | 2234 | { |
77558e4d | 2235 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 2236 | |
cbce1a68 PZ |
2237 | lockdep_assert_held(&rq->lock); |
2238 | ||
c05fbafb PZ |
2239 | #ifdef CONFIG_SMP |
2240 | if (p->sched_contributes_to_load) | |
2241 | rq->nr_uninterruptible--; | |
b5179ac7 | 2242 | |
b5179ac7 | 2243 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 2244 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
2245 | #endif |
2246 | ||
1b174a2c | 2247 | activate_task(rq, p, en_flags); |
d8ac8971 | 2248 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
2249 | } |
2250 | ||
2251 | /* | |
2252 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
2253 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
2254 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
2255 | * the task is still ->on_rq. | |
2256 | */ | |
2257 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
2258 | { | |
eb580751 | 2259 | struct rq_flags rf; |
c05fbafb PZ |
2260 | struct rq *rq; |
2261 | int ret = 0; | |
2262 | ||
eb580751 | 2263 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 2264 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
2265 | /* check_preempt_curr() may use rq clock */ |
2266 | update_rq_clock(rq); | |
d8ac8971 | 2267 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
2268 | ret = 1; |
2269 | } | |
eb580751 | 2270 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
2271 | |
2272 | return ret; | |
2273 | } | |
2274 | ||
317f3941 | 2275 | #ifdef CONFIG_SMP |
a1488664 | 2276 | void sched_ttwu_pending(void *arg) |
317f3941 | 2277 | { |
a1488664 | 2278 | struct llist_node *llist = arg; |
317f3941 | 2279 | struct rq *rq = this_rq(); |
73215849 | 2280 | struct task_struct *p, *t; |
d8ac8971 | 2281 | struct rq_flags rf; |
317f3941 | 2282 | |
e3baac47 PZ |
2283 | if (!llist) |
2284 | return; | |
2285 | ||
126c2092 PZ |
2286 | /* |
2287 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
2288 | * Races such that false-negatives are possible, since they | |
2289 | * are shorter lived that false-positives would be. | |
2290 | */ | |
2291 | WRITE_ONCE(rq->ttwu_pending, 0); | |
2292 | ||
8a8c69c3 | 2293 | rq_lock_irqsave(rq, &rf); |
77558e4d | 2294 | update_rq_clock(rq); |
317f3941 | 2295 | |
73215849 BP |
2296 | llist_for_each_entry_safe(p, t, llist, wake_entry) |
2297 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); | |
317f3941 | 2298 | |
8a8c69c3 | 2299 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
2300 | } |
2301 | ||
b2a02fc4 | 2302 | void send_call_function_single_ipi(int cpu) |
317f3941 | 2303 | { |
b2a02fc4 | 2304 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 2305 | |
b2a02fc4 PZ |
2306 | if (!set_nr_if_polling(rq->idle)) |
2307 | arch_send_call_function_single_ipi(cpu); | |
2308 | else | |
2309 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
2310 | } |
2311 | ||
2ebb1771 MG |
2312 | /* |
2313 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
2314 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
2315 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
2316 | * of the wakeup instead of the waker. | |
2317 | */ | |
2318 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 2319 | { |
e3baac47 PZ |
2320 | struct rq *rq = cpu_rq(cpu); |
2321 | ||
b7e7ade3 PZ |
2322 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
2323 | ||
126c2092 | 2324 | WRITE_ONCE(rq->ttwu_pending, 1); |
a1488664 | 2325 | __smp_call_single_queue(cpu, &p->wake_entry); |
317f3941 | 2326 | } |
d6aa8f85 | 2327 | |
f6be8af1 CL |
2328 | void wake_up_if_idle(int cpu) |
2329 | { | |
2330 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 2331 | struct rq_flags rf; |
f6be8af1 | 2332 | |
fd7de1e8 AL |
2333 | rcu_read_lock(); |
2334 | ||
2335 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
2336 | goto out; | |
f6be8af1 CL |
2337 | |
2338 | if (set_nr_if_polling(rq->idle)) { | |
2339 | trace_sched_wake_idle_without_ipi(cpu); | |
2340 | } else { | |
8a8c69c3 | 2341 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
2342 | if (is_idle_task(rq->curr)) |
2343 | smp_send_reschedule(cpu); | |
d1ccc66d | 2344 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 2345 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 2346 | } |
fd7de1e8 AL |
2347 | |
2348 | out: | |
2349 | rcu_read_unlock(); | |
f6be8af1 CL |
2350 | } |
2351 | ||
39be3501 | 2352 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
2353 | { |
2354 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
2355 | } | |
c6e7bd7a | 2356 | |
2ebb1771 MG |
2357 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
2358 | { | |
2359 | /* | |
2360 | * If the CPU does not share cache, then queue the task on the | |
2361 | * remote rqs wakelist to avoid accessing remote data. | |
2362 | */ | |
2363 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
2364 | return true; | |
2365 | ||
2366 | /* | |
2367 | * If the task is descheduling and the only running task on the | |
2368 | * CPU then use the wakelist to offload the task activation to | |
2369 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
2370 | * nr_running is checked to avoid unnecessary task stacking. | |
2371 | */ | |
2372 | if ((wake_flags & WF_ON_RQ) && cpu_rq(cpu)->nr_running <= 1) | |
2373 | return true; | |
2374 | ||
2375 | return false; | |
2376 | } | |
2377 | ||
2378 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 2379 | { |
2ebb1771 | 2380 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
c6e7bd7a | 2381 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 2382 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
2383 | return true; |
2384 | } | |
2385 | ||
2386 | return false; | |
2387 | } | |
d6aa8f85 | 2388 | #endif /* CONFIG_SMP */ |
317f3941 | 2389 | |
b5179ac7 | 2390 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
2391 | { |
2392 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 2393 | struct rq_flags rf; |
c05fbafb | 2394 | |
17d9f311 | 2395 | #if defined(CONFIG_SMP) |
2ebb1771 | 2396 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 2397 | return; |
317f3941 PZ |
2398 | #endif |
2399 | ||
8a8c69c3 | 2400 | rq_lock(rq, &rf); |
77558e4d | 2401 | update_rq_clock(rq); |
d8ac8971 | 2402 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 2403 | rq_unlock(rq, &rf); |
9ed3811a TH |
2404 | } |
2405 | ||
8643cda5 PZ |
2406 | /* |
2407 | * Notes on Program-Order guarantees on SMP systems. | |
2408 | * | |
2409 | * MIGRATION | |
2410 | * | |
2411 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
2412 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
2413 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
2414 | * |
2415 | * For migration (of runnable tasks) this is provided by the following means: | |
2416 | * | |
2417 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
2418 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
2419 | * rq(c1)->lock (if not at the same time, then in that order). | |
2420 | * C) LOCK of the rq(c1)->lock scheduling in task | |
2421 | * | |
7696f991 | 2422 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 2423 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
2424 | * |
2425 | * Example: | |
2426 | * | |
2427 | * CPU0 CPU1 CPU2 | |
2428 | * | |
2429 | * LOCK rq(0)->lock | |
2430 | * sched-out X | |
2431 | * sched-in Y | |
2432 | * UNLOCK rq(0)->lock | |
2433 | * | |
2434 | * LOCK rq(0)->lock // orders against CPU0 | |
2435 | * dequeue X | |
2436 | * UNLOCK rq(0)->lock | |
2437 | * | |
2438 | * LOCK rq(1)->lock | |
2439 | * enqueue X | |
2440 | * UNLOCK rq(1)->lock | |
2441 | * | |
2442 | * LOCK rq(1)->lock // orders against CPU2 | |
2443 | * sched-out Z | |
2444 | * sched-in X | |
2445 | * UNLOCK rq(1)->lock | |
2446 | * | |
2447 | * | |
2448 | * BLOCKING -- aka. SLEEP + WAKEUP | |
2449 | * | |
2450 | * For blocking we (obviously) need to provide the same guarantee as for | |
2451 | * migration. However the means are completely different as there is no lock | |
2452 | * chain to provide order. Instead we do: | |
2453 | * | |
2454 | * 1) smp_store_release(X->on_cpu, 0) | |
1f03e8d2 | 2455 | * 2) smp_cond_load_acquire(!X->on_cpu) |
8643cda5 PZ |
2456 | * |
2457 | * Example: | |
2458 | * | |
2459 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
2460 | * | |
2461 | * LOCK rq(0)->lock LOCK X->pi_lock | |
2462 | * dequeue X | |
2463 | * sched-out X | |
2464 | * smp_store_release(X->on_cpu, 0); | |
2465 | * | |
1f03e8d2 | 2466 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
2467 | * X->state = WAKING |
2468 | * set_task_cpu(X,2) | |
2469 | * | |
2470 | * LOCK rq(2)->lock | |
2471 | * enqueue X | |
2472 | * X->state = RUNNING | |
2473 | * UNLOCK rq(2)->lock | |
2474 | * | |
2475 | * LOCK rq(2)->lock // orders against CPU1 | |
2476 | * sched-out Z | |
2477 | * sched-in X | |
2478 | * UNLOCK rq(2)->lock | |
2479 | * | |
2480 | * UNLOCK X->pi_lock | |
2481 | * UNLOCK rq(0)->lock | |
2482 | * | |
2483 | * | |
7696f991 AP |
2484 | * However, for wakeups there is a second guarantee we must provide, namely we |
2485 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
2486 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
2487 | */ |
2488 | ||
9ed3811a | 2489 | /** |
1da177e4 | 2490 | * try_to_wake_up - wake up a thread |
9ed3811a | 2491 | * @p: the thread to be awakened |
1da177e4 | 2492 | * @state: the mask of task states that can be woken |
9ed3811a | 2493 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 2494 | * |
a2250238 | 2495 | * If (@state & @p->state) @p->state = TASK_RUNNING. |
1da177e4 | 2496 | * |
a2250238 PZ |
2497 | * If the task was not queued/runnable, also place it back on a runqueue. |
2498 | * | |
2499 | * Atomic against schedule() which would dequeue a task, also see | |
2500 | * set_current_state(). | |
2501 | * | |
7696f991 AP |
2502 | * This function executes a full memory barrier before accessing the task |
2503 | * state; see set_current_state(). | |
2504 | * | |
a2250238 PZ |
2505 | * Return: %true if @p->state changes (an actual wakeup was done), |
2506 | * %false otherwise. | |
1da177e4 | 2507 | */ |
e4a52bcb PZ |
2508 | static int |
2509 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2510 | { |
1da177e4 | 2511 | unsigned long flags; |
c05fbafb | 2512 | int cpu, success = 0; |
2398f2c6 | 2513 | |
e3d85487 | 2514 | preempt_disable(); |
aacedf26 PZ |
2515 | if (p == current) { |
2516 | /* | |
2517 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
2518 | * == smp_processor_id()'. Together this means we can special | |
2519 | * case the whole 'p->on_rq && ttwu_remote()' case below | |
2520 | * without taking any locks. | |
2521 | * | |
2522 | * In particular: | |
2523 | * - we rely on Program-Order guarantees for all the ordering, | |
2524 | * - we're serialized against set_special_state() by virtue of | |
2525 | * it disabling IRQs (this allows not taking ->pi_lock). | |
2526 | */ | |
2527 | if (!(p->state & state)) | |
e3d85487 | 2528 | goto out; |
aacedf26 PZ |
2529 | |
2530 | success = 1; | |
2531 | cpu = task_cpu(p); | |
2532 | trace_sched_waking(p); | |
2533 | p->state = TASK_RUNNING; | |
2534 | trace_sched_wakeup(p); | |
2535 | goto out; | |
2536 | } | |
2537 | ||
e0acd0a6 ON |
2538 | /* |
2539 | * If we are going to wake up a thread waiting for CONDITION we | |
2540 | * need to ensure that CONDITION=1 done by the caller can not be | |
2541 | * reordered with p->state check below. This pairs with mb() in | |
2542 | * set_current_state() the waiting thread does. | |
2543 | */ | |
013fdb80 | 2544 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 2545 | smp_mb__after_spinlock(); |
e9c84311 | 2546 | if (!(p->state & state)) |
aacedf26 | 2547 | goto unlock; |
1da177e4 | 2548 | |
fbd705a0 PZ |
2549 | trace_sched_waking(p); |
2550 | ||
d1ccc66d IM |
2551 | /* We're going to change ->state: */ |
2552 | success = 1; | |
1da177e4 | 2553 | cpu = task_cpu(p); |
1da177e4 | 2554 | |
135e8c92 BS |
2555 | /* |
2556 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
2557 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
2558 | * in smp_cond_load_acquire() below. | |
2559 | * | |
3d85b270 AP |
2560 | * sched_ttwu_pending() try_to_wake_up() |
2561 | * STORE p->on_rq = 1 LOAD p->state | |
2562 | * UNLOCK rq->lock | |
2563 | * | |
2564 | * __schedule() (switch to task 'p') | |
2565 | * LOCK rq->lock smp_rmb(); | |
2566 | * smp_mb__after_spinlock(); | |
2567 | * UNLOCK rq->lock | |
135e8c92 BS |
2568 | * |
2569 | * [task p] | |
3d85b270 | 2570 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 2571 | * |
3d85b270 AP |
2572 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
2573 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
2574 | * |
2575 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
2576 | */ |
2577 | smp_rmb(); | |
c05fbafb | 2578 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
aacedf26 | 2579 | goto unlock; |
1da177e4 | 2580 | |
c6e7bd7a PZ |
2581 | if (p->in_iowait) { |
2582 | delayacct_blkio_end(p); | |
2583 | atomic_dec(&task_rq(p)->nr_iowait); | |
2584 | } | |
2585 | ||
1da177e4 | 2586 | #ifdef CONFIG_SMP |
c6e7bd7a PZ |
2587 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
2588 | p->state = TASK_WAKING; | |
2589 | ||
ecf7d01c PZ |
2590 | /* |
2591 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2592 | * possible to, falsely, observe p->on_cpu == 0. | |
2593 | * | |
2594 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2595 | * from the runqueue. | |
2596 | * | |
3d85b270 AP |
2597 | * __schedule() (switch to task 'p') try_to_wake_up() |
2598 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
2599 | * UNLOCK rq->lock | |
2600 | * | |
2601 | * __schedule() (put 'p' to sleep) | |
2602 | * LOCK rq->lock smp_rmb(); | |
2603 | * smp_mb__after_spinlock(); | |
2604 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 2605 | * |
3d85b270 AP |
2606 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
2607 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
ecf7d01c PZ |
2608 | */ |
2609 | smp_rmb(); | |
2610 | ||
c6e7bd7a PZ |
2611 | /* |
2612 | * If the owning (remote) CPU is still in the middle of schedule() with | |
2613 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
2614 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
2615 | * let the waker make forward progress. This is safe because IRQs are | |
2616 | * disabled and the IPI will deliver after on_cpu is cleared. | |
2617 | */ | |
2ebb1771 | 2618 | if (READ_ONCE(p->on_cpu) && ttwu_queue_wakelist(p, cpu, wake_flags | WF_ON_RQ)) |
c6e7bd7a PZ |
2619 | goto unlock; |
2620 | ||
e9c84311 | 2621 | /* |
d1ccc66d | 2622 | * If the owning (remote) CPU is still in the middle of schedule() with |
c05fbafb | 2623 | * this task as prev, wait until its done referencing the task. |
b75a2253 | 2624 | * |
31cb1bc0 | 2625 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
2626 | * |
2627 | * This ensures that tasks getting woken will be fully ordered against | |
2628 | * their previous state and preserve Program Order. | |
0970d299 | 2629 | */ |
1f03e8d2 | 2630 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2631 | |
ac66f547 | 2632 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2633 | if (task_cpu(p) != cpu) { |
2634 | wake_flags |= WF_MIGRATED; | |
eb414681 | 2635 | psi_ttwu_dequeue(p); |
e4a52bcb | 2636 | set_task_cpu(p, cpu); |
f339b9dc | 2637 | } |
1da177e4 | 2638 | #endif /* CONFIG_SMP */ |
1da177e4 | 2639 | |
b5179ac7 | 2640 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 2641 | unlock: |
013fdb80 | 2642 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
2643 | out: |
2644 | if (success) | |
2645 | ttwu_stat(p, cpu, wake_flags); | |
e3d85487 | 2646 | preempt_enable(); |
1da177e4 LT |
2647 | |
2648 | return success; | |
2649 | } | |
2650 | ||
2beaf328 PM |
2651 | /** |
2652 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
2653 | * @p: Process for which the function is to be invoked. | |
2654 | * @func: Function to invoke. | |
2655 | * @arg: Argument to function. | |
2656 | * | |
2657 | * If the specified task can be quickly locked into a definite state | |
2658 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
2659 | * state while invoking @func(@arg). This function can use ->on_rq and | |
2660 | * task_curr() to work out what the state is, if required. Given that | |
2661 | * @func can be invoked with a runqueue lock held, it had better be quite | |
2662 | * lightweight. | |
2663 | * | |
2664 | * Returns: | |
2665 | * @false if the task slipped out from under the locks. | |
2666 | * @true if the task was locked onto a runqueue or is sleeping. | |
2667 | * However, @func can override this by returning @false. | |
2668 | */ | |
2669 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
2670 | { | |
2671 | bool ret = false; | |
2672 | struct rq_flags rf; | |
2673 | struct rq *rq; | |
2674 | ||
2675 | lockdep_assert_irqs_enabled(); | |
2676 | raw_spin_lock_irq(&p->pi_lock); | |
2677 | if (p->on_rq) { | |
2678 | rq = __task_rq_lock(p, &rf); | |
2679 | if (task_rq(p) == rq) | |
2680 | ret = func(p, arg); | |
2681 | rq_unlock(rq, &rf); | |
2682 | } else { | |
2683 | switch (p->state) { | |
2684 | case TASK_RUNNING: | |
2685 | case TASK_WAKING: | |
2686 | break; | |
2687 | default: | |
2688 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
2689 | if (!p->on_rq) | |
2690 | ret = func(p, arg); | |
2691 | } | |
2692 | } | |
2693 | raw_spin_unlock_irq(&p->pi_lock); | |
2694 | return ret; | |
2695 | } | |
2696 | ||
50fa610a DH |
2697 | /** |
2698 | * wake_up_process - Wake up a specific process | |
2699 | * @p: The process to be woken up. | |
2700 | * | |
2701 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
2702 | * processes. |
2703 | * | |
2704 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 2705 | * |
7696f991 | 2706 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 2707 | */ |
7ad5b3a5 | 2708 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2709 | { |
9067ac85 | 2710 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 2711 | } |
1da177e4 LT |
2712 | EXPORT_SYMBOL(wake_up_process); |
2713 | ||
7ad5b3a5 | 2714 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2715 | { |
2716 | return try_to_wake_up(p, state, 0); | |
2717 | } | |
2718 | ||
1da177e4 LT |
2719 | /* |
2720 | * Perform scheduler related setup for a newly forked process p. | |
2721 | * p is forked by current. | |
dd41f596 IM |
2722 | * |
2723 | * __sched_fork() is basic setup used by init_idle() too: | |
2724 | */ | |
5e1576ed | 2725 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2726 | { |
fd2f4419 PZ |
2727 | p->on_rq = 0; |
2728 | ||
2729 | p->se.on_rq = 0; | |
dd41f596 IM |
2730 | p->se.exec_start = 0; |
2731 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2732 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2733 | p->se.nr_migrations = 0; |
da7a735e | 2734 | p->se.vruntime = 0; |
fd2f4419 | 2735 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 2736 | |
ad936d86 BP |
2737 | #ifdef CONFIG_FAIR_GROUP_SCHED |
2738 | p->se.cfs_rq = NULL; | |
2739 | #endif | |
2740 | ||
6cfb0d5d | 2741 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 2742 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 2743 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2744 | #endif |
476d139c | 2745 | |
aab03e05 | 2746 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 2747 | init_dl_task_timer(&p->dl); |
209a0cbd | 2748 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 2749 | __dl_clear_params(p); |
aab03e05 | 2750 | |
fa717060 | 2751 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
2752 | p->rt.timeout = 0; |
2753 | p->rt.time_slice = sched_rr_timeslice; | |
2754 | p->rt.on_rq = 0; | |
2755 | p->rt.on_list = 0; | |
476d139c | 2756 | |
e107be36 AK |
2757 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2758 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2759 | #endif | |
cbee9f88 | 2760 | |
5e1f0f09 MG |
2761 | #ifdef CONFIG_COMPACTION |
2762 | p->capture_control = NULL; | |
2763 | #endif | |
13784475 | 2764 | init_numa_balancing(clone_flags, p); |
a1488664 PZ |
2765 | #ifdef CONFIG_SMP |
2766 | p->wake_entry_type = CSD_TYPE_TTWU; | |
2767 | #endif | |
dd41f596 IM |
2768 | } |
2769 | ||
2a595721 SD |
2770 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
2771 | ||
1a687c2e | 2772 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 2773 | |
1a687c2e MG |
2774 | void set_numabalancing_state(bool enabled) |
2775 | { | |
2776 | if (enabled) | |
2a595721 | 2777 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 2778 | else |
2a595721 | 2779 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 2780 | } |
54a43d54 AK |
2781 | |
2782 | #ifdef CONFIG_PROC_SYSCTL | |
2783 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 2784 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
2785 | { |
2786 | struct ctl_table t; | |
2787 | int err; | |
2a595721 | 2788 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
2789 | |
2790 | if (write && !capable(CAP_SYS_ADMIN)) | |
2791 | return -EPERM; | |
2792 | ||
2793 | t = *table; | |
2794 | t.data = &state; | |
2795 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2796 | if (err < 0) | |
2797 | return err; | |
2798 | if (write) | |
2799 | set_numabalancing_state(state); | |
2800 | return err; | |
2801 | } | |
2802 | #endif | |
2803 | #endif | |
dd41f596 | 2804 | |
4698f88c JP |
2805 | #ifdef CONFIG_SCHEDSTATS |
2806 | ||
cb251765 | 2807 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 2808 | static bool __initdata __sched_schedstats = false; |
cb251765 | 2809 | |
cb251765 MG |
2810 | static void set_schedstats(bool enabled) |
2811 | { | |
2812 | if (enabled) | |
2813 | static_branch_enable(&sched_schedstats); | |
2814 | else | |
2815 | static_branch_disable(&sched_schedstats); | |
2816 | } | |
2817 | ||
2818 | void force_schedstat_enabled(void) | |
2819 | { | |
2820 | if (!schedstat_enabled()) { | |
2821 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
2822 | static_branch_enable(&sched_schedstats); | |
2823 | } | |
2824 | } | |
2825 | ||
2826 | static int __init setup_schedstats(char *str) | |
2827 | { | |
2828 | int ret = 0; | |
2829 | if (!str) | |
2830 | goto out; | |
2831 | ||
4698f88c JP |
2832 | /* |
2833 | * This code is called before jump labels have been set up, so we can't | |
2834 | * change the static branch directly just yet. Instead set a temporary | |
2835 | * variable so init_schedstats() can do it later. | |
2836 | */ | |
cb251765 | 2837 | if (!strcmp(str, "enable")) { |
4698f88c | 2838 | __sched_schedstats = true; |
cb251765 MG |
2839 | ret = 1; |
2840 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 2841 | __sched_schedstats = false; |
cb251765 MG |
2842 | ret = 1; |
2843 | } | |
2844 | out: | |
2845 | if (!ret) | |
2846 | pr_warn("Unable to parse schedstats=\n"); | |
2847 | ||
2848 | return ret; | |
2849 | } | |
2850 | __setup("schedstats=", setup_schedstats); | |
2851 | ||
4698f88c JP |
2852 | static void __init init_schedstats(void) |
2853 | { | |
2854 | set_schedstats(__sched_schedstats); | |
2855 | } | |
2856 | ||
cb251765 | 2857 | #ifdef CONFIG_PROC_SYSCTL |
32927393 CH |
2858 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
2859 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
2860 | { |
2861 | struct ctl_table t; | |
2862 | int err; | |
2863 | int state = static_branch_likely(&sched_schedstats); | |
2864 | ||
2865 | if (write && !capable(CAP_SYS_ADMIN)) | |
2866 | return -EPERM; | |
2867 | ||
2868 | t = *table; | |
2869 | t.data = &state; | |
2870 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
2871 | if (err < 0) | |
2872 | return err; | |
2873 | if (write) | |
2874 | set_schedstats(state); | |
2875 | return err; | |
2876 | } | |
4698f88c JP |
2877 | #endif /* CONFIG_PROC_SYSCTL */ |
2878 | #else /* !CONFIG_SCHEDSTATS */ | |
2879 | static inline void init_schedstats(void) {} | |
2880 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
2881 | |
2882 | /* | |
2883 | * fork()/clone()-time setup: | |
2884 | */ | |
aab03e05 | 2885 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 2886 | { |
0122ec5b | 2887 | unsigned long flags; |
dd41f596 | 2888 | |
5e1576ed | 2889 | __sched_fork(clone_flags, p); |
06b83b5f | 2890 | /* |
7dc603c9 | 2891 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
2892 | * nobody will actually run it, and a signal or other external |
2893 | * event cannot wake it up and insert it on the runqueue either. | |
2894 | */ | |
7dc603c9 | 2895 | p->state = TASK_NEW; |
dd41f596 | 2896 | |
c350a04e MG |
2897 | /* |
2898 | * Make sure we do not leak PI boosting priority to the child. | |
2899 | */ | |
2900 | p->prio = current->normal_prio; | |
2901 | ||
e8f14172 PB |
2902 | uclamp_fork(p); |
2903 | ||
b9dc29e7 MG |
2904 | /* |
2905 | * Revert to default priority/policy on fork if requested. | |
2906 | */ | |
2907 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 2908 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 2909 | p->policy = SCHED_NORMAL; |
6c697bdf | 2910 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
2911 | p->rt_priority = 0; |
2912 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
2913 | p->static_prio = NICE_TO_PRIO(0); | |
2914 | ||
2915 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 2916 | set_load_weight(p, false); |
6c697bdf | 2917 | |
b9dc29e7 MG |
2918 | /* |
2919 | * We don't need the reset flag anymore after the fork. It has | |
2920 | * fulfilled its duty: | |
2921 | */ | |
2922 | p->sched_reset_on_fork = 0; | |
2923 | } | |
ca94c442 | 2924 | |
af0fffd9 | 2925 | if (dl_prio(p->prio)) |
aab03e05 | 2926 | return -EAGAIN; |
af0fffd9 | 2927 | else if (rt_prio(p->prio)) |
aab03e05 | 2928 | p->sched_class = &rt_sched_class; |
af0fffd9 | 2929 | else |
2ddbf952 | 2930 | p->sched_class = &fair_sched_class; |
b29739f9 | 2931 | |
7dc603c9 | 2932 | init_entity_runnable_average(&p->se); |
cd29fe6f | 2933 | |
86951599 PZ |
2934 | /* |
2935 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2936 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2937 | * is ran before sched_fork(). | |
2938 | * | |
2939 | * Silence PROVE_RCU. | |
2940 | */ | |
0122ec5b | 2941 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e210bffd | 2942 | /* |
d1ccc66d | 2943 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
2944 | * so use __set_task_cpu(). |
2945 | */ | |
af0fffd9 | 2946 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
2947 | if (p->sched_class->task_fork) |
2948 | p->sched_class->task_fork(p); | |
0122ec5b | 2949 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2950 | |
f6db8347 | 2951 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 2952 | if (likely(sched_info_on())) |
52f17b6c | 2953 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2954 | #endif |
3ca7a440 PZ |
2955 | #if defined(CONFIG_SMP) |
2956 | p->on_cpu = 0; | |
4866cde0 | 2957 | #endif |
01028747 | 2958 | init_task_preempt_count(p); |
806c09a7 | 2959 | #ifdef CONFIG_SMP |
917b627d | 2960 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 2961 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 2962 | #endif |
aab03e05 | 2963 | return 0; |
1da177e4 LT |
2964 | } |
2965 | ||
332ac17e DF |
2966 | unsigned long to_ratio(u64 period, u64 runtime) |
2967 | { | |
2968 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 2969 | return BW_UNIT; |
332ac17e DF |
2970 | |
2971 | /* | |
2972 | * Doing this here saves a lot of checks in all | |
2973 | * the calling paths, and returning zero seems | |
2974 | * safe for them anyway. | |
2975 | */ | |
2976 | if (period == 0) | |
2977 | return 0; | |
2978 | ||
c52f14d3 | 2979 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
2980 | } |
2981 | ||
1da177e4 LT |
2982 | /* |
2983 | * wake_up_new_task - wake up a newly created task for the first time. | |
2984 | * | |
2985 | * This function will do some initial scheduler statistics housekeeping | |
2986 | * that must be done for every newly created context, then puts the task | |
2987 | * on the runqueue and wakes it. | |
2988 | */ | |
3e51e3ed | 2989 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 2990 | { |
eb580751 | 2991 | struct rq_flags rf; |
dd41f596 | 2992 | struct rq *rq; |
fabf318e | 2993 | |
eb580751 | 2994 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 2995 | p->state = TASK_RUNNING; |
fabf318e PZ |
2996 | #ifdef CONFIG_SMP |
2997 | /* | |
2998 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 2999 | * - cpus_ptr can change in the fork path |
d1ccc66d | 3000 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
3001 | * |
3002 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
3003 | * as we're not fully set-up yet. | |
fabf318e | 3004 | */ |
32e839dd | 3005 | p->recent_used_cpu = task_cpu(p); |
e210bffd | 3006 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 3007 | #endif |
b7fa30c9 | 3008 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 3009 | update_rq_clock(rq); |
d0fe0b9c | 3010 | post_init_entity_util_avg(p); |
0017d735 | 3011 | |
7a57f32a | 3012 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 3013 | trace_sched_wakeup_new(p); |
a7558e01 | 3014 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3015 | #ifdef CONFIG_SMP |
0aaafaab PZ |
3016 | if (p->sched_class->task_woken) { |
3017 | /* | |
3018 | * Nothing relies on rq->lock after this, so its fine to | |
3019 | * drop it. | |
3020 | */ | |
d8ac8971 | 3021 | rq_unpin_lock(rq, &rf); |
efbbd05a | 3022 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3023 | rq_repin_lock(rq, &rf); |
0aaafaab | 3024 | } |
9a897c5a | 3025 | #endif |
eb580751 | 3026 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
3027 | } |
3028 | ||
e107be36 AK |
3029 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3030 | ||
b7203428 | 3031 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 3032 | |
2ecd9d29 PZ |
3033 | void preempt_notifier_inc(void) |
3034 | { | |
b7203428 | 3035 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
3036 | } |
3037 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
3038 | ||
3039 | void preempt_notifier_dec(void) | |
3040 | { | |
b7203428 | 3041 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
3042 | } |
3043 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
3044 | ||
e107be36 | 3045 | /** |
80dd99b3 | 3046 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3047 | * @notifier: notifier struct to register |
e107be36 AK |
3048 | */ |
3049 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3050 | { | |
b7203428 | 3051 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
3052 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
3053 | ||
e107be36 AK |
3054 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
3055 | } | |
3056 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3057 | ||
3058 | /** | |
3059 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3060 | * @notifier: notifier struct to unregister |
e107be36 | 3061 | * |
d84525a8 | 3062 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
3063 | */ |
3064 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3065 | { | |
3066 | hlist_del(¬ifier->link); | |
3067 | } | |
3068 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3069 | ||
1cde2930 | 3070 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3071 | { |
3072 | struct preempt_notifier *notifier; | |
e107be36 | 3073 | |
b67bfe0d | 3074 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3075 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
3076 | } | |
3077 | ||
1cde2930 PZ |
3078 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
3079 | { | |
b7203428 | 3080 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3081 | __fire_sched_in_preempt_notifiers(curr); |
3082 | } | |
3083 | ||
e107be36 | 3084 | static void |
1cde2930 PZ |
3085 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3086 | struct task_struct *next) | |
e107be36 AK |
3087 | { |
3088 | struct preempt_notifier *notifier; | |
e107be36 | 3089 | |
b67bfe0d | 3090 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3091 | notifier->ops->sched_out(notifier, next); |
3092 | } | |
3093 | ||
1cde2930 PZ |
3094 | static __always_inline void |
3095 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3096 | struct task_struct *next) | |
3097 | { | |
b7203428 | 3098 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3099 | __fire_sched_out_preempt_notifiers(curr, next); |
3100 | } | |
3101 | ||
6d6bc0ad | 3102 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3103 | |
1cde2930 | 3104 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3105 | { |
3106 | } | |
3107 | ||
1cde2930 | 3108 | static inline void |
e107be36 AK |
3109 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3110 | struct task_struct *next) | |
3111 | { | |
3112 | } | |
3113 | ||
6d6bc0ad | 3114 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3115 | |
31cb1bc0 | 3116 | static inline void prepare_task(struct task_struct *next) |
3117 | { | |
3118 | #ifdef CONFIG_SMP | |
3119 | /* | |
3120 | * Claim the task as running, we do this before switching to it | |
3121 | * such that any running task will have this set. | |
3122 | */ | |
3123 | next->on_cpu = 1; | |
3124 | #endif | |
3125 | } | |
3126 | ||
3127 | static inline void finish_task(struct task_struct *prev) | |
3128 | { | |
3129 | #ifdef CONFIG_SMP | |
3130 | /* | |
3131 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
3132 | * We must ensure this doesn't happen until the switch is completely | |
3133 | * finished. | |
3134 | * | |
3135 | * In particular, the load of prev->state in finish_task_switch() must | |
3136 | * happen before this. | |
3137 | * | |
3138 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
3139 | */ | |
3140 | smp_store_release(&prev->on_cpu, 0); | |
3141 | #endif | |
3142 | } | |
3143 | ||
269d5992 PZ |
3144 | static inline void |
3145 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 3146 | { |
269d5992 PZ |
3147 | /* |
3148 | * Since the runqueue lock will be released by the next | |
3149 | * task (which is an invalid locking op but in the case | |
3150 | * of the scheduler it's an obvious special-case), so we | |
3151 | * do an early lockdep release here: | |
3152 | */ | |
3153 | rq_unpin_lock(rq, rf); | |
5facae4f | 3154 | spin_release(&rq->lock.dep_map, _THIS_IP_); |
31cb1bc0 | 3155 | #ifdef CONFIG_DEBUG_SPINLOCK |
3156 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 3157 | rq->lock.owner = next; |
31cb1bc0 | 3158 | #endif |
269d5992 PZ |
3159 | } |
3160 | ||
3161 | static inline void finish_lock_switch(struct rq *rq) | |
3162 | { | |
31cb1bc0 | 3163 | /* |
3164 | * If we are tracking spinlock dependencies then we have to | |
3165 | * fix up the runqueue lock - which gets 'carried over' from | |
3166 | * prev into current: | |
3167 | */ | |
3168 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
31cb1bc0 | 3169 | raw_spin_unlock_irq(&rq->lock); |
3170 | } | |
3171 | ||
325ea10c IM |
3172 | /* |
3173 | * NOP if the arch has not defined these: | |
3174 | */ | |
3175 | ||
3176 | #ifndef prepare_arch_switch | |
3177 | # define prepare_arch_switch(next) do { } while (0) | |
3178 | #endif | |
3179 | ||
3180 | #ifndef finish_arch_post_lock_switch | |
3181 | # define finish_arch_post_lock_switch() do { } while (0) | |
3182 | #endif | |
3183 | ||
4866cde0 NP |
3184 | /** |
3185 | * prepare_task_switch - prepare to switch tasks | |
3186 | * @rq: the runqueue preparing to switch | |
421cee29 | 3187 | * @prev: the current task that is being switched out |
4866cde0 NP |
3188 | * @next: the task we are going to switch to. |
3189 | * | |
3190 | * This is called with the rq lock held and interrupts off. It must | |
3191 | * be paired with a subsequent finish_task_switch after the context | |
3192 | * switch. | |
3193 | * | |
3194 | * prepare_task_switch sets up locking and calls architecture specific | |
3195 | * hooks. | |
3196 | */ | |
e107be36 AK |
3197 | static inline void |
3198 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
3199 | struct task_struct *next) | |
4866cde0 | 3200 | { |
0ed557aa | 3201 | kcov_prepare_switch(prev); |
43148951 | 3202 | sched_info_switch(rq, prev, next); |
fe4b04fa | 3203 | perf_event_task_sched_out(prev, next); |
d7822b1e | 3204 | rseq_preempt(prev); |
e107be36 | 3205 | fire_sched_out_preempt_notifiers(prev, next); |
31cb1bc0 | 3206 | prepare_task(next); |
4866cde0 NP |
3207 | prepare_arch_switch(next); |
3208 | } | |
3209 | ||
1da177e4 LT |
3210 | /** |
3211 | * finish_task_switch - clean up after a task-switch | |
3212 | * @prev: the thread we just switched away from. | |
3213 | * | |
4866cde0 NP |
3214 | * finish_task_switch must be called after the context switch, paired |
3215 | * with a prepare_task_switch call before the context switch. | |
3216 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
3217 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
3218 | * |
3219 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 3220 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
3221 | * with the lock held can cause deadlocks; see schedule() for |
3222 | * details.) | |
dfa50b60 ON |
3223 | * |
3224 | * The context switch have flipped the stack from under us and restored the | |
3225 | * local variables which were saved when this task called schedule() in the | |
3226 | * past. prev == current is still correct but we need to recalculate this_rq | |
3227 | * because prev may have moved to another CPU. | |
1da177e4 | 3228 | */ |
dfa50b60 | 3229 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
3230 | __releases(rq->lock) |
3231 | { | |
dfa50b60 | 3232 | struct rq *rq = this_rq(); |
1da177e4 | 3233 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 3234 | long prev_state; |
1da177e4 | 3235 | |
609ca066 PZ |
3236 | /* |
3237 | * The previous task will have left us with a preempt_count of 2 | |
3238 | * because it left us after: | |
3239 | * | |
3240 | * schedule() | |
3241 | * preempt_disable(); // 1 | |
3242 | * __schedule() | |
3243 | * raw_spin_lock_irq(&rq->lock) // 2 | |
3244 | * | |
3245 | * Also, see FORK_PREEMPT_COUNT. | |
3246 | */ | |
e2bf1c4b PZ |
3247 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
3248 | "corrupted preempt_count: %s/%d/0x%x\n", | |
3249 | current->comm, current->pid, preempt_count())) | |
3250 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 3251 | |
1da177e4 LT |
3252 | rq->prev_mm = NULL; |
3253 | ||
3254 | /* | |
3255 | * A task struct has one reference for the use as "current". | |
c394cc9f | 3256 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
3257 | * schedule one last time. The schedule call will never return, and |
3258 | * the scheduled task must drop that reference. | |
95913d97 PZ |
3259 | * |
3260 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 3261 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
3262 | * running on another CPU and we could rave with its RUNNING -> DEAD |
3263 | * transition, resulting in a double drop. | |
1da177e4 | 3264 | */ |
55a101f8 | 3265 | prev_state = prev->state; |
bf9fae9f | 3266 | vtime_task_switch(prev); |
a8d757ef | 3267 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 3268 | finish_task(prev); |
3269 | finish_lock_switch(rq); | |
01f23e16 | 3270 | finish_arch_post_lock_switch(); |
0ed557aa | 3271 | kcov_finish_switch(current); |
e8fa1362 | 3272 | |
e107be36 | 3273 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 3274 | /* |
70216e18 MD |
3275 | * When switching through a kernel thread, the loop in |
3276 | * membarrier_{private,global}_expedited() may have observed that | |
3277 | * kernel thread and not issued an IPI. It is therefore possible to | |
3278 | * schedule between user->kernel->user threads without passing though | |
3279 | * switch_mm(). Membarrier requires a barrier after storing to | |
3280 | * rq->curr, before returning to userspace, so provide them here: | |
3281 | * | |
3282 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
3283 | * provided by mmdrop(), | |
3284 | * - a sync_core for SYNC_CORE. | |
306e0604 | 3285 | */ |
70216e18 MD |
3286 | if (mm) { |
3287 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 3288 | mmdrop(mm); |
70216e18 | 3289 | } |
1cef1150 PZ |
3290 | if (unlikely(prev_state == TASK_DEAD)) { |
3291 | if (prev->sched_class->task_dead) | |
3292 | prev->sched_class->task_dead(prev); | |
68f24b08 | 3293 | |
1cef1150 PZ |
3294 | /* |
3295 | * Remove function-return probe instances associated with this | |
3296 | * task and put them back on the free list. | |
3297 | */ | |
3298 | kprobe_flush_task(prev); | |
3299 | ||
3300 | /* Task is done with its stack. */ | |
3301 | put_task_stack(prev); | |
3302 | ||
0ff7b2cf | 3303 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 3304 | } |
99e5ada9 | 3305 | |
de734f89 | 3306 | tick_nohz_task_switch(); |
dfa50b60 | 3307 | return rq; |
1da177e4 LT |
3308 | } |
3309 | ||
3f029d3c GH |
3310 | #ifdef CONFIG_SMP |
3311 | ||
3f029d3c | 3312 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 3313 | static void __balance_callback(struct rq *rq) |
3f029d3c | 3314 | { |
e3fca9e7 PZ |
3315 | struct callback_head *head, *next; |
3316 | void (*func)(struct rq *rq); | |
3317 | unsigned long flags; | |
3f029d3c | 3318 | |
e3fca9e7 PZ |
3319 | raw_spin_lock_irqsave(&rq->lock, flags); |
3320 | head = rq->balance_callback; | |
3321 | rq->balance_callback = NULL; | |
3322 | while (head) { | |
3323 | func = (void (*)(struct rq *))head->func; | |
3324 | next = head->next; | |
3325 | head->next = NULL; | |
3326 | head = next; | |
3f029d3c | 3327 | |
e3fca9e7 | 3328 | func(rq); |
3f029d3c | 3329 | } |
e3fca9e7 PZ |
3330 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3331 | } | |
3332 | ||
3333 | static inline void balance_callback(struct rq *rq) | |
3334 | { | |
3335 | if (unlikely(rq->balance_callback)) | |
3336 | __balance_callback(rq); | |
3f029d3c GH |
3337 | } |
3338 | ||
3339 | #else | |
da19ab51 | 3340 | |
e3fca9e7 | 3341 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 3342 | { |
1da177e4 LT |
3343 | } |
3344 | ||
3f029d3c GH |
3345 | #endif |
3346 | ||
1da177e4 LT |
3347 | /** |
3348 | * schedule_tail - first thing a freshly forked thread must call. | |
3349 | * @prev: the thread we just switched away from. | |
3350 | */ | |
722a9f92 | 3351 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3352 | __releases(rq->lock) |
3353 | { | |
1a43a14a | 3354 | struct rq *rq; |
da19ab51 | 3355 | |
609ca066 PZ |
3356 | /* |
3357 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
3358 | * finish_task_switch() for details. | |
3359 | * | |
3360 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
3361 | * and the preempt_enable() will end up enabling preemption (on | |
3362 | * PREEMPT_COUNT kernels). | |
3363 | */ | |
3364 | ||
dfa50b60 | 3365 | rq = finish_task_switch(prev); |
e3fca9e7 | 3366 | balance_callback(rq); |
1a43a14a | 3367 | preempt_enable(); |
70b97a7f | 3368 | |
1da177e4 | 3369 | if (current->set_child_tid) |
b488893a | 3370 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
3371 | |
3372 | calculate_sigpending(); | |
1da177e4 LT |
3373 | } |
3374 | ||
3375 | /* | |
dfa50b60 | 3376 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 3377 | */ |
04936948 | 3378 | static __always_inline struct rq * |
70b97a7f | 3379 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 3380 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 3381 | { |
e107be36 | 3382 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3383 | |
9226d125 ZA |
3384 | /* |
3385 | * For paravirt, this is coupled with an exit in switch_to to | |
3386 | * combine the page table reload and the switch backend into | |
3387 | * one hypercall. | |
3388 | */ | |
224101ed | 3389 | arch_start_context_switch(prev); |
9226d125 | 3390 | |
306e0604 | 3391 | /* |
139d025c PZ |
3392 | * kernel -> kernel lazy + transfer active |
3393 | * user -> kernel lazy + mmgrab() active | |
3394 | * | |
3395 | * kernel -> user switch + mmdrop() active | |
3396 | * user -> user switch | |
306e0604 | 3397 | */ |
139d025c PZ |
3398 | if (!next->mm) { // to kernel |
3399 | enter_lazy_tlb(prev->active_mm, next); | |
3400 | ||
3401 | next->active_mm = prev->active_mm; | |
3402 | if (prev->mm) // from user | |
3403 | mmgrab(prev->active_mm); | |
3404 | else | |
3405 | prev->active_mm = NULL; | |
3406 | } else { // to user | |
227a4aad | 3407 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
3408 | /* |
3409 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 3410 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
3411 | * |
3412 | * The below provides this either through switch_mm(), or in | |
3413 | * case 'prev->active_mm == next->mm' through | |
3414 | * finish_task_switch()'s mmdrop(). | |
3415 | */ | |
139d025c | 3416 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 3417 | |
139d025c PZ |
3418 | if (!prev->mm) { // from kernel |
3419 | /* will mmdrop() in finish_task_switch(). */ | |
3420 | rq->prev_mm = prev->active_mm; | |
3421 | prev->active_mm = NULL; | |
3422 | } | |
1da177e4 | 3423 | } |
92509b73 | 3424 | |
cb42c9a3 | 3425 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 3426 | |
269d5992 | 3427 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
3428 | |
3429 | /* Here we just switch the register state and the stack. */ | |
3430 | switch_to(prev, next, prev); | |
dd41f596 | 3431 | barrier(); |
dfa50b60 ON |
3432 | |
3433 | return finish_task_switch(prev); | |
1da177e4 LT |
3434 | } |
3435 | ||
3436 | /* | |
1c3e8264 | 3437 | * nr_running and nr_context_switches: |
1da177e4 LT |
3438 | * |
3439 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 3440 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
3441 | */ |
3442 | unsigned long nr_running(void) | |
3443 | { | |
3444 | unsigned long i, sum = 0; | |
3445 | ||
3446 | for_each_online_cpu(i) | |
3447 | sum += cpu_rq(i)->nr_running; | |
3448 | ||
3449 | return sum; | |
f711f609 | 3450 | } |
1da177e4 | 3451 | |
2ee507c4 | 3452 | /* |
d1ccc66d | 3453 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
3454 | * |
3455 | * Caution: this function does not check that the caller has disabled | |
3456 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
3457 | * race. The caller is responsible to use it correctly, for example: | |
3458 | * | |
dfcb245e | 3459 | * - from a non-preemptible section (of course) |
00cc1633 DD |
3460 | * |
3461 | * - from a thread that is bound to a single CPU | |
3462 | * | |
3463 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
3464 | */ |
3465 | bool single_task_running(void) | |
3466 | { | |
00cc1633 | 3467 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
3468 | } |
3469 | EXPORT_SYMBOL(single_task_running); | |
3470 | ||
1da177e4 | 3471 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3472 | { |
cc94abfc SR |
3473 | int i; |
3474 | unsigned long long sum = 0; | |
46cb4b7c | 3475 | |
0a945022 | 3476 | for_each_possible_cpu(i) |
1da177e4 | 3477 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3478 | |
1da177e4 LT |
3479 | return sum; |
3480 | } | |
483b4ee6 | 3481 | |
145d952a DL |
3482 | /* |
3483 | * Consumers of these two interfaces, like for example the cpuidle menu | |
3484 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
3485 | * for a CPU that has IO-wait which might not even end up running the task when | |
3486 | * it does become runnable. | |
3487 | */ | |
3488 | ||
3489 | unsigned long nr_iowait_cpu(int cpu) | |
3490 | { | |
3491 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
3492 | } | |
3493 | ||
e33a9bba TH |
3494 | /* |
3495 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
3496 | * | |
3497 | * The idea behind IO-wait account is to account the idle time that we could | |
3498 | * have spend running if it were not for IO. That is, if we were to improve the | |
3499 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
3500 | * | |
3501 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
3502 | * idle time as IO-wait, because if the storage were faster, it could've been | |
3503 | * running and we'd not be idle. | |
3504 | * | |
3505 | * This has been extended to SMP, by doing the same for each CPU. This however | |
3506 | * is broken. | |
3507 | * | |
3508 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
3509 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
3510 | * though, if the storage were faster, both could've ran at the same time, | |
3511 | * utilising both CPUs. | |
3512 | * | |
3513 | * This means, that when looking globally, the current IO-wait accounting on | |
3514 | * SMP is a lower bound, by reason of under accounting. | |
3515 | * | |
3516 | * Worse, since the numbers are provided per CPU, they are sometimes | |
3517 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
3518 | * associated with any one particular CPU, it can wake to another CPU than it | |
3519 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
3520 | * | |
3521 | * Task CPU affinities can make all that even more 'interesting'. | |
3522 | */ | |
3523 | ||
1da177e4 LT |
3524 | unsigned long nr_iowait(void) |
3525 | { | |
3526 | unsigned long i, sum = 0; | |
483b4ee6 | 3527 | |
0a945022 | 3528 | for_each_possible_cpu(i) |
145d952a | 3529 | sum += nr_iowait_cpu(i); |
46cb4b7c | 3530 | |
1da177e4 LT |
3531 | return sum; |
3532 | } | |
483b4ee6 | 3533 | |
dd41f596 | 3534 | #ifdef CONFIG_SMP |
8a0be9ef | 3535 | |
46cb4b7c | 3536 | /* |
38022906 PZ |
3537 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3538 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3539 | */ |
38022906 | 3540 | void sched_exec(void) |
46cb4b7c | 3541 | { |
38022906 | 3542 | struct task_struct *p = current; |
1da177e4 | 3543 | unsigned long flags; |
0017d735 | 3544 | int dest_cpu; |
46cb4b7c | 3545 | |
8f42ced9 | 3546 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 3547 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3548 | if (dest_cpu == smp_processor_id()) |
3549 | goto unlock; | |
38022906 | 3550 | |
8f42ced9 | 3551 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3552 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3553 | |
8f42ced9 PZ |
3554 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3555 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3556 | return; |
3557 | } | |
0017d735 | 3558 | unlock: |
8f42ced9 | 3559 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3560 | } |
dd41f596 | 3561 | |
1da177e4 LT |
3562 | #endif |
3563 | ||
1da177e4 | 3564 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 3565 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
3566 | |
3567 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 3568 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 3569 | |
6075620b GG |
3570 | /* |
3571 | * The function fair_sched_class.update_curr accesses the struct curr | |
3572 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
3573 | * we observe a high rate of cache misses in practice. | |
3574 | * Prefetching this data results in improved performance. | |
3575 | */ | |
3576 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3577 | { | |
3578 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3579 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3580 | #else | |
3581 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3582 | #endif | |
3583 | prefetch(curr); | |
3584 | prefetch(&curr->exec_start); | |
3585 | } | |
3586 | ||
c5f8d995 HS |
3587 | /* |
3588 | * Return accounted runtime for the task. | |
3589 | * In case the task is currently running, return the runtime plus current's | |
3590 | * pending runtime that have not been accounted yet. | |
3591 | */ | |
3592 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3593 | { | |
eb580751 | 3594 | struct rq_flags rf; |
c5f8d995 | 3595 | struct rq *rq; |
6e998916 | 3596 | u64 ns; |
c5f8d995 | 3597 | |
911b2898 PZ |
3598 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3599 | /* | |
97fb7a0a | 3600 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
3601 | * So we have a optimization chance when the task's delta_exec is 0. |
3602 | * Reading ->on_cpu is racy, but this is ok. | |
3603 | * | |
d1ccc66d IM |
3604 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
3605 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 3606 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
3607 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3608 | * been accounted, so we're correct here as well. | |
911b2898 | 3609 | */ |
da0c1e65 | 3610 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3611 | return p->se.sum_exec_runtime; |
3612 | #endif | |
3613 | ||
eb580751 | 3614 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3615 | /* |
3616 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3617 | * project cycles that may never be accounted to this | |
3618 | * thread, breaking clock_gettime(). | |
3619 | */ | |
3620 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3621 | prefetch_curr_exec_start(p); |
6e998916 SG |
3622 | update_rq_clock(rq); |
3623 | p->sched_class->update_curr(rq); | |
3624 | } | |
3625 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3626 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3627 | |
3628 | return ns; | |
3629 | } | |
48f24c4d | 3630 | |
14533a16 IM |
3631 | DEFINE_PER_CPU(unsigned long, thermal_pressure); |
3632 | ||
3633 | void arch_set_thermal_pressure(struct cpumask *cpus, | |
3634 | unsigned long th_pressure) | |
3635 | { | |
3636 | int cpu; | |
3637 | ||
3638 | for_each_cpu(cpu, cpus) | |
3639 | WRITE_ONCE(per_cpu(thermal_pressure, cpu), th_pressure); | |
3640 | } | |
3641 | ||
7835b98b CL |
3642 | /* |
3643 | * This function gets called by the timer code, with HZ frequency. | |
3644 | * We call it with interrupts disabled. | |
7835b98b CL |
3645 | */ |
3646 | void scheduler_tick(void) | |
3647 | { | |
7835b98b CL |
3648 | int cpu = smp_processor_id(); |
3649 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3650 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 3651 | struct rq_flags rf; |
b4eccf5f | 3652 | unsigned long thermal_pressure; |
3e51f33f | 3653 | |
1567c3e3 | 3654 | arch_scale_freq_tick(); |
3e51f33f | 3655 | sched_clock_tick(); |
dd41f596 | 3656 | |
8a8c69c3 PZ |
3657 | rq_lock(rq, &rf); |
3658 | ||
3e51f33f | 3659 | update_rq_clock(rq); |
b4eccf5f | 3660 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 3661 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 3662 | curr->sched_class->task_tick(rq, curr, 0); |
3289bdb4 | 3663 | calc_global_load_tick(rq); |
eb414681 | 3664 | psi_task_tick(rq); |
8a8c69c3 PZ |
3665 | |
3666 | rq_unlock(rq, &rf); | |
7835b98b | 3667 | |
e9d2b064 | 3668 | perf_event_task_tick(); |
e220d2dc | 3669 | |
e418e1c2 | 3670 | #ifdef CONFIG_SMP |
6eb57e0d | 3671 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 3672 | trigger_load_balance(rq); |
e418e1c2 | 3673 | #endif |
1da177e4 LT |
3674 | } |
3675 | ||
265f22a9 | 3676 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
3677 | |
3678 | struct tick_work { | |
3679 | int cpu; | |
b55bd585 | 3680 | atomic_t state; |
d84b3131 FW |
3681 | struct delayed_work work; |
3682 | }; | |
b55bd585 PM |
3683 | /* Values for ->state, see diagram below. */ |
3684 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
3685 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
3686 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
3687 | ||
3688 | /* | |
3689 | * State diagram for ->state: | |
3690 | * | |
3691 | * | |
3692 | * TICK_SCHED_REMOTE_OFFLINE | |
3693 | * | ^ | |
3694 | * | | | |
3695 | * | | sched_tick_remote() | |
3696 | * | | | |
3697 | * | | | |
3698 | * +--TICK_SCHED_REMOTE_OFFLINING | |
3699 | * | ^ | |
3700 | * | | | |
3701 | * sched_tick_start() | | sched_tick_stop() | |
3702 | * | | | |
3703 | * V | | |
3704 | * TICK_SCHED_REMOTE_RUNNING | |
3705 | * | |
3706 | * | |
3707 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
3708 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
3709 | */ | |
d84b3131 FW |
3710 | |
3711 | static struct tick_work __percpu *tick_work_cpu; | |
3712 | ||
3713 | static void sched_tick_remote(struct work_struct *work) | |
3714 | { | |
3715 | struct delayed_work *dwork = to_delayed_work(work); | |
3716 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
3717 | int cpu = twork->cpu; | |
3718 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 3719 | struct task_struct *curr; |
d84b3131 | 3720 | struct rq_flags rf; |
d9c0ffca | 3721 | u64 delta; |
b55bd585 | 3722 | int os; |
d84b3131 FW |
3723 | |
3724 | /* | |
3725 | * Handle the tick only if it appears the remote CPU is running in full | |
3726 | * dynticks mode. The check is racy by nature, but missing a tick or | |
3727 | * having one too much is no big deal because the scheduler tick updates | |
3728 | * statistics and checks timeslices in a time-independent way, regardless | |
3729 | * of when exactly it is running. | |
3730 | */ | |
488603b8 | 3731 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 3732 | goto out_requeue; |
d84b3131 | 3733 | |
d9c0ffca FW |
3734 | rq_lock_irq(rq, &rf); |
3735 | curr = rq->curr; | |
488603b8 | 3736 | if (cpu_is_offline(cpu)) |
d9c0ffca | 3737 | goto out_unlock; |
d84b3131 | 3738 | |
d9c0ffca | 3739 | update_rq_clock(rq); |
d9c0ffca | 3740 | |
488603b8 SW |
3741 | if (!is_idle_task(curr)) { |
3742 | /* | |
3743 | * Make sure the next tick runs within a reasonable | |
3744 | * amount of time. | |
3745 | */ | |
3746 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
3747 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
3748 | } | |
d9c0ffca FW |
3749 | curr->sched_class->task_tick(rq, curr, 0); |
3750 | ||
ebc0f83c | 3751 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
3752 | out_unlock: |
3753 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 3754 | out_requeue: |
ebc0f83c | 3755 | |
d84b3131 FW |
3756 | /* |
3757 | * Run the remote tick once per second (1Hz). This arbitrary | |
3758 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
3759 | * to keep scheduler internal stats reasonably up to date. But |
3760 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 3761 | */ |
b55bd585 PM |
3762 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
3763 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
3764 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
3765 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
3766 | } |
3767 | ||
3768 | static void sched_tick_start(int cpu) | |
3769 | { | |
b55bd585 | 3770 | int os; |
d84b3131 FW |
3771 | struct tick_work *twork; |
3772 | ||
3773 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
3774 | return; | |
3775 | ||
3776 | WARN_ON_ONCE(!tick_work_cpu); | |
3777 | ||
3778 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
3779 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
3780 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
3781 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
3782 | twork->cpu = cpu; | |
3783 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
3784 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
3785 | } | |
d84b3131 FW |
3786 | } |
3787 | ||
3788 | #ifdef CONFIG_HOTPLUG_CPU | |
3789 | static void sched_tick_stop(int cpu) | |
3790 | { | |
3791 | struct tick_work *twork; | |
b55bd585 | 3792 | int os; |
d84b3131 FW |
3793 | |
3794 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
3795 | return; | |
3796 | ||
3797 | WARN_ON_ONCE(!tick_work_cpu); | |
3798 | ||
3799 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
3800 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
3801 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
3802 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
3803 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
3804 | } |
3805 | #endif /* CONFIG_HOTPLUG_CPU */ | |
3806 | ||
3807 | int __init sched_tick_offload_init(void) | |
3808 | { | |
3809 | tick_work_cpu = alloc_percpu(struct tick_work); | |
3810 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
3811 | return 0; |
3812 | } | |
3813 | ||
3814 | #else /* !CONFIG_NO_HZ_FULL */ | |
3815 | static inline void sched_tick_start(int cpu) { } | |
3816 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 3817 | #endif |
1da177e4 | 3818 | |
c1a280b6 | 3819 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 3820 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
3821 | /* |
3822 | * If the value passed in is equal to the current preempt count | |
3823 | * then we just disabled preemption. Start timing the latency. | |
3824 | */ | |
3825 | static inline void preempt_latency_start(int val) | |
3826 | { | |
3827 | if (preempt_count() == val) { | |
3828 | unsigned long ip = get_lock_parent_ip(); | |
3829 | #ifdef CONFIG_DEBUG_PREEMPT | |
3830 | current->preempt_disable_ip = ip; | |
3831 | #endif | |
3832 | trace_preempt_off(CALLER_ADDR0, ip); | |
3833 | } | |
3834 | } | |
7e49fcce | 3835 | |
edafe3a5 | 3836 | void preempt_count_add(int val) |
1da177e4 | 3837 | { |
6cd8a4bb | 3838 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3839 | /* |
3840 | * Underflow? | |
3841 | */ | |
9a11b49a IM |
3842 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3843 | return; | |
6cd8a4bb | 3844 | #endif |
bdb43806 | 3845 | __preempt_count_add(val); |
6cd8a4bb | 3846 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3847 | /* |
3848 | * Spinlock count overflowing soon? | |
3849 | */ | |
33859f7f MOS |
3850 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3851 | PREEMPT_MASK - 10); | |
6cd8a4bb | 3852 | #endif |
47252cfb | 3853 | preempt_latency_start(val); |
1da177e4 | 3854 | } |
bdb43806 | 3855 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 3856 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 3857 | |
47252cfb SR |
3858 | /* |
3859 | * If the value passed in equals to the current preempt count | |
3860 | * then we just enabled preemption. Stop timing the latency. | |
3861 | */ | |
3862 | static inline void preempt_latency_stop(int val) | |
3863 | { | |
3864 | if (preempt_count() == val) | |
3865 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
3866 | } | |
3867 | ||
edafe3a5 | 3868 | void preempt_count_sub(int val) |
1da177e4 | 3869 | { |
6cd8a4bb | 3870 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3871 | /* |
3872 | * Underflow? | |
3873 | */ | |
01e3eb82 | 3874 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3875 | return; |
1da177e4 LT |
3876 | /* |
3877 | * Is the spinlock portion underflowing? | |
3878 | */ | |
9a11b49a IM |
3879 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3880 | !(preempt_count() & PREEMPT_MASK))) | |
3881 | return; | |
6cd8a4bb | 3882 | #endif |
9a11b49a | 3883 | |
47252cfb | 3884 | preempt_latency_stop(val); |
bdb43806 | 3885 | __preempt_count_sub(val); |
1da177e4 | 3886 | } |
bdb43806 | 3887 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 3888 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 3889 | |
47252cfb SR |
3890 | #else |
3891 | static inline void preempt_latency_start(int val) { } | |
3892 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
3893 | #endif |
3894 | ||
59ddbcb2 IM |
3895 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
3896 | { | |
3897 | #ifdef CONFIG_DEBUG_PREEMPT | |
3898 | return p->preempt_disable_ip; | |
3899 | #else | |
3900 | return 0; | |
3901 | #endif | |
3902 | } | |
3903 | ||
1da177e4 | 3904 | /* |
dd41f596 | 3905 | * Print scheduling while atomic bug: |
1da177e4 | 3906 | */ |
dd41f596 | 3907 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3908 | { |
d1c6d149 VN |
3909 | /* Save this before calling printk(), since that will clobber it */ |
3910 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
3911 | ||
664dfa65 DJ |
3912 | if (oops_in_progress) |
3913 | return; | |
3914 | ||
3df0fc5b PZ |
3915 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3916 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3917 | |
dd41f596 | 3918 | debug_show_held_locks(prev); |
e21f5b15 | 3919 | print_modules(); |
dd41f596 IM |
3920 | if (irqs_disabled()) |
3921 | print_irqtrace_events(prev); | |
d1c6d149 VN |
3922 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
3923 | && in_atomic_preempt_off()) { | |
8f47b187 | 3924 | pr_err("Preemption disabled at:"); |
2062a4e8 | 3925 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 3926 | } |
748c7201 DBO |
3927 | if (panic_on_warn) |
3928 | panic("scheduling while atomic\n"); | |
3929 | ||
6135fc1e | 3930 | dump_stack(); |
373d4d09 | 3931 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 3932 | } |
1da177e4 | 3933 | |
dd41f596 IM |
3934 | /* |
3935 | * Various schedule()-time debugging checks and statistics: | |
3936 | */ | |
312364f3 | 3937 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 3938 | { |
0d9e2632 | 3939 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
3940 | if (task_stack_end_corrupted(prev)) |
3941 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
3942 | |
3943 | if (task_scs_end_corrupted(prev)) | |
3944 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 3945 | #endif |
b99def8b | 3946 | |
312364f3 DV |
3947 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
3948 | if (!preempt && prev->state && prev->non_block_count) { | |
3949 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", | |
3950 | prev->comm, prev->pid, prev->non_block_count); | |
3951 | dump_stack(); | |
3952 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
3953 | } | |
3954 | #endif | |
3955 | ||
1dc0fffc | 3956 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 3957 | __schedule_bug(prev); |
1dc0fffc PZ |
3958 | preempt_count_set(PREEMPT_DISABLED); |
3959 | } | |
b3fbab05 | 3960 | rcu_sleep_check(); |
dd41f596 | 3961 | |
1da177e4 LT |
3962 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3963 | ||
ae92882e | 3964 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
3965 | } |
3966 | ||
457d1f46 CY |
3967 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
3968 | struct rq_flags *rf) | |
3969 | { | |
3970 | #ifdef CONFIG_SMP | |
3971 | const struct sched_class *class; | |
3972 | /* | |
3973 | * We must do the balancing pass before put_prev_task(), such | |
3974 | * that when we release the rq->lock the task is in the same | |
3975 | * state as before we took rq->lock. | |
3976 | * | |
3977 | * We can terminate the balance pass as soon as we know there is | |
3978 | * a runnable task of @class priority or higher. | |
3979 | */ | |
3980 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
3981 | if (class->balance(rq, prev, rf)) | |
3982 | break; | |
3983 | } | |
3984 | #endif | |
3985 | ||
3986 | put_prev_task(rq, prev); | |
3987 | } | |
3988 | ||
dd41f596 IM |
3989 | /* |
3990 | * Pick up the highest-prio task: | |
3991 | */ | |
3992 | static inline struct task_struct * | |
d8ac8971 | 3993 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 3994 | { |
49ee5768 | 3995 | const struct sched_class *class; |
dd41f596 | 3996 | struct task_struct *p; |
1da177e4 LT |
3997 | |
3998 | /* | |
0ba87bb2 PZ |
3999 | * Optimization: we know that if all tasks are in the fair class we can |
4000 | * call that function directly, but only if the @prev task wasn't of a | |
4001 | * higher scheduling class, because otherwise those loose the | |
4002 | * opportunity to pull in more work from other CPUs. | |
1da177e4 | 4003 | */ |
0ba87bb2 PZ |
4004 | if (likely((prev->sched_class == &idle_sched_class || |
4005 | prev->sched_class == &fair_sched_class) && | |
4006 | rq->nr_running == rq->cfs.h_nr_running)) { | |
4007 | ||
5d7d6056 | 4008 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 4009 | if (unlikely(p == RETRY_TASK)) |
67692435 | 4010 | goto restart; |
6ccdc84b | 4011 | |
d1ccc66d | 4012 | /* Assumes fair_sched_class->next == idle_sched_class */ |
5d7d6056 | 4013 | if (!p) { |
f488e105 | 4014 | put_prev_task(rq, prev); |
98c2f700 | 4015 | p = pick_next_task_idle(rq); |
f488e105 | 4016 | } |
6ccdc84b PZ |
4017 | |
4018 | return p; | |
1da177e4 LT |
4019 | } |
4020 | ||
67692435 | 4021 | restart: |
457d1f46 | 4022 | put_prev_task_balance(rq, prev, rf); |
67692435 | 4023 | |
34f971f6 | 4024 | for_each_class(class) { |
98c2f700 | 4025 | p = class->pick_next_task(rq); |
67692435 | 4026 | if (p) |
dd41f596 | 4027 | return p; |
dd41f596 | 4028 | } |
34f971f6 | 4029 | |
d1ccc66d IM |
4030 | /* The idle class should always have a runnable task: */ |
4031 | BUG(); | |
dd41f596 | 4032 | } |
1da177e4 | 4033 | |
dd41f596 | 4034 | /* |
c259e01a | 4035 | * __schedule() is the main scheduler function. |
edde96ea PE |
4036 | * |
4037 | * The main means of driving the scheduler and thus entering this function are: | |
4038 | * | |
4039 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
4040 | * | |
4041 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
4042 | * paths. For example, see arch/x86/entry_64.S. | |
4043 | * | |
4044 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
4045 | * interrupt handler scheduler_tick(). | |
4046 | * | |
4047 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
4048 | * task to the run-queue and that's it. | |
4049 | * | |
4050 | * Now, if the new task added to the run-queue preempts the current | |
4051 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
4052 | * called on the nearest possible occasion: | |
4053 | * | |
c1a280b6 | 4054 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
4055 | * |
4056 | * - in syscall or exception context, at the next outmost | |
4057 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
4058 | * spin_unlock()!) | |
4059 | * | |
4060 | * - in IRQ context, return from interrupt-handler to | |
4061 | * preemptible context | |
4062 | * | |
c1a280b6 | 4063 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
4064 | * then at the next: |
4065 | * | |
4066 | * - cond_resched() call | |
4067 | * - explicit schedule() call | |
4068 | * - return from syscall or exception to user-space | |
4069 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 4070 | * |
b30f0e3f | 4071 | * WARNING: must be called with preemption disabled! |
dd41f596 | 4072 | */ |
499d7955 | 4073 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
4074 | { |
4075 | struct task_struct *prev, *next; | |
67ca7bde | 4076 | unsigned long *switch_count; |
d8ac8971 | 4077 | struct rq_flags rf; |
dd41f596 | 4078 | struct rq *rq; |
31656519 | 4079 | int cpu; |
dd41f596 | 4080 | |
dd41f596 IM |
4081 | cpu = smp_processor_id(); |
4082 | rq = cpu_rq(cpu); | |
dd41f596 | 4083 | prev = rq->curr; |
dd41f596 | 4084 | |
312364f3 | 4085 | schedule_debug(prev, preempt); |
1da177e4 | 4086 | |
31656519 | 4087 | if (sched_feat(HRTICK)) |
f333fdc9 | 4088 | hrtick_clear(rq); |
8f4d37ec | 4089 | |
46a5d164 | 4090 | local_irq_disable(); |
bcbfdd01 | 4091 | rcu_note_context_switch(preempt); |
46a5d164 | 4092 | |
e0acd0a6 ON |
4093 | /* |
4094 | * Make sure that signal_pending_state()->signal_pending() below | |
4095 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
4096 | * done by the caller to avoid the race with signal_wake_up(). | |
306e0604 MD |
4097 | * |
4098 | * The membarrier system call requires a full memory barrier | |
4099 | * after coming from user-space, before storing to rq->curr. | |
e0acd0a6 | 4100 | */ |
8a8c69c3 | 4101 | rq_lock(rq, &rf); |
d89e588c | 4102 | smp_mb__after_spinlock(); |
1da177e4 | 4103 | |
d1ccc66d IM |
4104 | /* Promote REQ to ACT */ |
4105 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 4106 | update_rq_clock(rq); |
9edfbfed | 4107 | |
246d86b5 | 4108 | switch_count = &prev->nivcsw; |
fc13aeba | 4109 | if (!preempt && prev->state) { |
34ec35ad | 4110 | if (signal_pending_state(prev->state, prev)) { |
1da177e4 | 4111 | prev->state = TASK_RUNNING; |
21aa9af0 | 4112 | } else { |
bce4dc80 | 4113 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 4114 | |
e33a9bba TH |
4115 | if (prev->in_iowait) { |
4116 | atomic_inc(&rq->nr_iowait); | |
4117 | delayacct_blkio_start(); | |
4118 | } | |
21aa9af0 | 4119 | } |
dd41f596 | 4120 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4121 | } |
4122 | ||
d8ac8971 | 4123 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 4124 | clear_tsk_need_resched(prev); |
f27dde8d | 4125 | clear_preempt_need_resched(); |
1da177e4 | 4126 | |
1da177e4 | 4127 | if (likely(prev != next)) { |
1da177e4 | 4128 | rq->nr_switches++; |
5311a98f EB |
4129 | /* |
4130 | * RCU users of rcu_dereference(rq->curr) may not see | |
4131 | * changes to task_struct made by pick_next_task(). | |
4132 | */ | |
4133 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
4134 | /* |
4135 | * The membarrier system call requires each architecture | |
4136 | * to have a full memory barrier after updating | |
306e0604 MD |
4137 | * rq->curr, before returning to user-space. |
4138 | * | |
4139 | * Here are the schemes providing that barrier on the | |
4140 | * various architectures: | |
4141 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
4142 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
4143 | * - finish_lock_switch() for weakly-ordered | |
4144 | * architectures where spin_unlock is a full barrier, | |
4145 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
4146 | * is a RELEASE barrier), | |
22e4ebb9 | 4147 | */ |
1da177e4 LT |
4148 | ++*switch_count; |
4149 | ||
b05e75d6 JW |
4150 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
4151 | ||
c73464b1 | 4152 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
4153 | |
4154 | /* Also unlocks the rq: */ | |
4155 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 4156 | } else { |
cb42c9a3 | 4157 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
8a8c69c3 | 4158 | rq_unlock_irq(rq, &rf); |
cbce1a68 | 4159 | } |
1da177e4 | 4160 | |
e3fca9e7 | 4161 | balance_callback(rq); |
1da177e4 | 4162 | } |
c259e01a | 4163 | |
9af6528e PZ |
4164 | void __noreturn do_task_dead(void) |
4165 | { | |
d1ccc66d | 4166 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 4167 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
4168 | |
4169 | /* Tell freezer to ignore us: */ | |
4170 | current->flags |= PF_NOFREEZE; | |
4171 | ||
9af6528e PZ |
4172 | __schedule(false); |
4173 | BUG(); | |
d1ccc66d IM |
4174 | |
4175 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 4176 | for (;;) |
d1ccc66d | 4177 | cpu_relax(); |
9af6528e PZ |
4178 | } |
4179 | ||
9c40cef2 TG |
4180 | static inline void sched_submit_work(struct task_struct *tsk) |
4181 | { | |
b0fdc013 | 4182 | if (!tsk->state) |
9c40cef2 | 4183 | return; |
6d25be57 TG |
4184 | |
4185 | /* | |
4186 | * If a worker went to sleep, notify and ask workqueue whether | |
4187 | * it wants to wake up a task to maintain concurrency. | |
4188 | * As this function is called inside the schedule() context, | |
4189 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
4190 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
4191 | * requires it. | |
6d25be57 | 4192 | */ |
771b53d0 | 4193 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 4194 | preempt_disable(); |
771b53d0 JA |
4195 | if (tsk->flags & PF_WQ_WORKER) |
4196 | wq_worker_sleeping(tsk); | |
4197 | else | |
4198 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
4199 | preempt_enable_no_resched(); |
4200 | } | |
4201 | ||
b0fdc013 SAS |
4202 | if (tsk_is_pi_blocked(tsk)) |
4203 | return; | |
4204 | ||
9c40cef2 TG |
4205 | /* |
4206 | * If we are going to sleep and we have plugged IO queued, | |
4207 | * make sure to submit it to avoid deadlocks. | |
4208 | */ | |
4209 | if (blk_needs_flush_plug(tsk)) | |
4210 | blk_schedule_flush_plug(tsk); | |
4211 | } | |
4212 | ||
6d25be57 TG |
4213 | static void sched_update_worker(struct task_struct *tsk) |
4214 | { | |
771b53d0 JA |
4215 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
4216 | if (tsk->flags & PF_WQ_WORKER) | |
4217 | wq_worker_running(tsk); | |
4218 | else | |
4219 | io_wq_worker_running(tsk); | |
4220 | } | |
6d25be57 TG |
4221 | } |
4222 | ||
722a9f92 | 4223 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 4224 | { |
9c40cef2 TG |
4225 | struct task_struct *tsk = current; |
4226 | ||
4227 | sched_submit_work(tsk); | |
bfd9b2b5 | 4228 | do { |
b30f0e3f | 4229 | preempt_disable(); |
fc13aeba | 4230 | __schedule(false); |
b30f0e3f | 4231 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 4232 | } while (need_resched()); |
6d25be57 | 4233 | sched_update_worker(tsk); |
c259e01a | 4234 | } |
1da177e4 LT |
4235 | EXPORT_SYMBOL(schedule); |
4236 | ||
8663effb SRV |
4237 | /* |
4238 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
4239 | * state (have scheduled out non-voluntarily) by making sure that all | |
4240 | * tasks have either left the run queue or have gone into user space. | |
4241 | * As idle tasks do not do either, they must not ever be preempted | |
4242 | * (schedule out non-voluntarily). | |
4243 | * | |
4244 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
4245 | * never enables preemption because it does not call sched_submit_work(). | |
4246 | */ | |
4247 | void __sched schedule_idle(void) | |
4248 | { | |
4249 | /* | |
4250 | * As this skips calling sched_submit_work(), which the idle task does | |
4251 | * regardless because that function is a nop when the task is in a | |
4252 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
4253 | * current task can be in any other state. Note, idle is always in the | |
4254 | * TASK_RUNNING state. | |
4255 | */ | |
4256 | WARN_ON_ONCE(current->state); | |
4257 | do { | |
4258 | __schedule(false); | |
4259 | } while (need_resched()); | |
4260 | } | |
4261 | ||
91d1aa43 | 4262 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 4263 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
4264 | { |
4265 | /* | |
4266 | * If we come here after a random call to set_need_resched(), | |
4267 | * or we have been woken up remotely but the IPI has not yet arrived, | |
4268 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
4269 | * we find a better solution. | |
7cc78f8f AL |
4270 | * |
4271 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 4272 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 4273 | * too frequently to make sense yet. |
20ab65e3 | 4274 | */ |
7cc78f8f | 4275 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 4276 | schedule(); |
7cc78f8f | 4277 | exception_exit(prev_state); |
20ab65e3 FW |
4278 | } |
4279 | #endif | |
4280 | ||
c5491ea7 TG |
4281 | /** |
4282 | * schedule_preempt_disabled - called with preemption disabled | |
4283 | * | |
4284 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
4285 | */ | |
4286 | void __sched schedule_preempt_disabled(void) | |
4287 | { | |
ba74c144 | 4288 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
4289 | schedule(); |
4290 | preempt_disable(); | |
4291 | } | |
4292 | ||
06b1f808 | 4293 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
4294 | { |
4295 | do { | |
47252cfb SR |
4296 | /* |
4297 | * Because the function tracer can trace preempt_count_sub() | |
4298 | * and it also uses preempt_enable/disable_notrace(), if | |
4299 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
4300 | * by the function tracer will call this function again and | |
4301 | * cause infinite recursion. | |
4302 | * | |
4303 | * Preemption must be disabled here before the function | |
4304 | * tracer can trace. Break up preempt_disable() into two | |
4305 | * calls. One to disable preemption without fear of being | |
4306 | * traced. The other to still record the preemption latency, | |
4307 | * which can also be traced by the function tracer. | |
4308 | */ | |
499d7955 | 4309 | preempt_disable_notrace(); |
47252cfb | 4310 | preempt_latency_start(1); |
fc13aeba | 4311 | __schedule(true); |
47252cfb | 4312 | preempt_latency_stop(1); |
499d7955 | 4313 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
4314 | |
4315 | /* | |
4316 | * Check again in case we missed a preemption opportunity | |
4317 | * between schedule and now. | |
4318 | */ | |
a18b5d01 FW |
4319 | } while (need_resched()); |
4320 | } | |
4321 | ||
c1a280b6 | 4322 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 4323 | /* |
a49b4f40 VS |
4324 | * This is the entry point to schedule() from in-kernel preemption |
4325 | * off of preempt_enable. | |
1da177e4 | 4326 | */ |
722a9f92 | 4327 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 4328 | { |
1da177e4 LT |
4329 | /* |
4330 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4331 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4332 | */ |
fbb00b56 | 4333 | if (likely(!preemptible())) |
1da177e4 LT |
4334 | return; |
4335 | ||
a18b5d01 | 4336 | preempt_schedule_common(); |
1da177e4 | 4337 | } |
376e2424 | 4338 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 4339 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 4340 | |
009f60e2 | 4341 | /** |
4eaca0a8 | 4342 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
4343 | * |
4344 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
4345 | * recursion and tracing preempt enabling caused by the tracing | |
4346 | * infrastructure itself. But as tracing can happen in areas coming | |
4347 | * from userspace or just about to enter userspace, a preempt enable | |
4348 | * can occur before user_exit() is called. This will cause the scheduler | |
4349 | * to be called when the system is still in usermode. | |
4350 | * | |
4351 | * To prevent this, the preempt_enable_notrace will use this function | |
4352 | * instead of preempt_schedule() to exit user context if needed before | |
4353 | * calling the scheduler. | |
4354 | */ | |
4eaca0a8 | 4355 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
4356 | { |
4357 | enum ctx_state prev_ctx; | |
4358 | ||
4359 | if (likely(!preemptible())) | |
4360 | return; | |
4361 | ||
4362 | do { | |
47252cfb SR |
4363 | /* |
4364 | * Because the function tracer can trace preempt_count_sub() | |
4365 | * and it also uses preempt_enable/disable_notrace(), if | |
4366 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
4367 | * by the function tracer will call this function again and | |
4368 | * cause infinite recursion. | |
4369 | * | |
4370 | * Preemption must be disabled here before the function | |
4371 | * tracer can trace. Break up preempt_disable() into two | |
4372 | * calls. One to disable preemption without fear of being | |
4373 | * traced. The other to still record the preemption latency, | |
4374 | * which can also be traced by the function tracer. | |
4375 | */ | |
3d8f74dd | 4376 | preempt_disable_notrace(); |
47252cfb | 4377 | preempt_latency_start(1); |
009f60e2 ON |
4378 | /* |
4379 | * Needs preempt disabled in case user_exit() is traced | |
4380 | * and the tracer calls preempt_enable_notrace() causing | |
4381 | * an infinite recursion. | |
4382 | */ | |
4383 | prev_ctx = exception_enter(); | |
fc13aeba | 4384 | __schedule(true); |
009f60e2 ON |
4385 | exception_exit(prev_ctx); |
4386 | ||
47252cfb | 4387 | preempt_latency_stop(1); |
3d8f74dd | 4388 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
4389 | } while (need_resched()); |
4390 | } | |
4eaca0a8 | 4391 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 4392 | |
c1a280b6 | 4393 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 LT |
4394 | |
4395 | /* | |
a49b4f40 | 4396 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4397 | * off of irq context. |
4398 | * Note, that this is called and return with irqs disabled. This will | |
4399 | * protect us against recursive calling from irq. | |
4400 | */ | |
722a9f92 | 4401 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 4402 | { |
b22366cd | 4403 | enum ctx_state prev_state; |
6478d880 | 4404 | |
2ed6e34f | 4405 | /* Catch callers which need to be fixed */ |
f27dde8d | 4406 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 4407 | |
b22366cd FW |
4408 | prev_state = exception_enter(); |
4409 | ||
3a5c359a | 4410 | do { |
3d8f74dd | 4411 | preempt_disable(); |
3a5c359a | 4412 | local_irq_enable(); |
fc13aeba | 4413 | __schedule(true); |
3a5c359a | 4414 | local_irq_disable(); |
3d8f74dd | 4415 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 4416 | } while (need_resched()); |
b22366cd FW |
4417 | |
4418 | exception_exit(prev_state); | |
1da177e4 LT |
4419 | } |
4420 | ||
ac6424b9 | 4421 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4422 | void *key) |
1da177e4 | 4423 | { |
63859d4f | 4424 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4425 | } |
1da177e4 LT |
4426 | EXPORT_SYMBOL(default_wake_function); |
4427 | ||
b29739f9 IM |
4428 | #ifdef CONFIG_RT_MUTEXES |
4429 | ||
acd58620 PZ |
4430 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
4431 | { | |
4432 | if (pi_task) | |
4433 | prio = min(prio, pi_task->prio); | |
4434 | ||
4435 | return prio; | |
4436 | } | |
4437 | ||
4438 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
4439 | { | |
4440 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
4441 | ||
4442 | return __rt_effective_prio(pi_task, prio); | |
4443 | } | |
4444 | ||
b29739f9 IM |
4445 | /* |
4446 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
4447 | * @p: task to boost |
4448 | * @pi_task: donor task | |
b29739f9 IM |
4449 | * |
4450 | * This function changes the 'effective' priority of a task. It does | |
4451 | * not touch ->normal_prio like __setscheduler(). | |
4452 | * | |
c365c292 TG |
4453 | * Used by the rt_mutex code to implement priority inheritance |
4454 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 4455 | */ |
acd58620 | 4456 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 4457 | { |
acd58620 | 4458 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 4459 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 4460 | const struct sched_class *prev_class; |
eb580751 PZ |
4461 | struct rq_flags rf; |
4462 | struct rq *rq; | |
b29739f9 | 4463 | |
acd58620 PZ |
4464 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
4465 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
4466 | ||
4467 | /* | |
4468 | * If nothing changed; bail early. | |
4469 | */ | |
4470 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
4471 | return; | |
b29739f9 | 4472 | |
eb580751 | 4473 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 4474 | update_rq_clock(rq); |
acd58620 PZ |
4475 | /* |
4476 | * Set under pi_lock && rq->lock, such that the value can be used under | |
4477 | * either lock. | |
4478 | * | |
4479 | * Note that there is loads of tricky to make this pointer cache work | |
4480 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
4481 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
4482 | * task is allowed to run again (and can exit). This ensures the pointer | |
4483 | * points to a blocked task -- which guaratees the task is present. | |
4484 | */ | |
4485 | p->pi_top_task = pi_task; | |
4486 | ||
4487 | /* | |
4488 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
4489 | */ | |
4490 | if (prio == p->prio && !dl_prio(prio)) | |
4491 | goto out_unlock; | |
b29739f9 | 4492 | |
1c4dd99b TG |
4493 | /* |
4494 | * Idle task boosting is a nono in general. There is one | |
4495 | * exception, when PREEMPT_RT and NOHZ is active: | |
4496 | * | |
4497 | * The idle task calls get_next_timer_interrupt() and holds | |
4498 | * the timer wheel base->lock on the CPU and another CPU wants | |
4499 | * to access the timer (probably to cancel it). We can safely | |
4500 | * ignore the boosting request, as the idle CPU runs this code | |
4501 | * with interrupts disabled and will complete the lock | |
4502 | * protected section without being interrupted. So there is no | |
4503 | * real need to boost. | |
4504 | */ | |
4505 | if (unlikely(p == rq->idle)) { | |
4506 | WARN_ON(p != rq->curr); | |
4507 | WARN_ON(p->pi_blocked_on); | |
4508 | goto out_unlock; | |
4509 | } | |
4510 | ||
b91473ff | 4511 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 4512 | oldprio = p->prio; |
ff77e468 PZ |
4513 | |
4514 | if (oldprio == prio) | |
4515 | queue_flag &= ~DEQUEUE_MOVE; | |
4516 | ||
83ab0aa0 | 4517 | prev_class = p->sched_class; |
da0c1e65 | 4518 | queued = task_on_rq_queued(p); |
051a1d1a | 4519 | running = task_current(rq, p); |
da0c1e65 | 4520 | if (queued) |
ff77e468 | 4521 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 4522 | if (running) |
f3cd1c4e | 4523 | put_prev_task(rq, p); |
dd41f596 | 4524 | |
2d3d891d DF |
4525 | /* |
4526 | * Boosting condition are: | |
4527 | * 1. -rt task is running and holds mutex A | |
4528 | * --> -dl task blocks on mutex A | |
4529 | * | |
4530 | * 2. -dl task is running and holds mutex A | |
4531 | * --> -dl task blocks on mutex A and could preempt the | |
4532 | * running task | |
4533 | */ | |
4534 | if (dl_prio(prio)) { | |
466af29b ON |
4535 | if (!dl_prio(p->normal_prio) || |
4536 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 4537 | p->dl.dl_boosted = 1; |
ff77e468 | 4538 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
4539 | } else |
4540 | p->dl.dl_boosted = 0; | |
aab03e05 | 4541 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
4542 | } else if (rt_prio(prio)) { |
4543 | if (dl_prio(oldprio)) | |
4544 | p->dl.dl_boosted = 0; | |
4545 | if (oldprio < prio) | |
ff77e468 | 4546 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 4547 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
4548 | } else { |
4549 | if (dl_prio(oldprio)) | |
4550 | p->dl.dl_boosted = 0; | |
746db944 BS |
4551 | if (rt_prio(oldprio)) |
4552 | p->rt.timeout = 0; | |
dd41f596 | 4553 | p->sched_class = &fair_sched_class; |
2d3d891d | 4554 | } |
dd41f596 | 4555 | |
b29739f9 IM |
4556 | p->prio = prio; |
4557 | ||
da0c1e65 | 4558 | if (queued) |
ff77e468 | 4559 | enqueue_task(rq, p, queue_flag); |
a399d233 | 4560 | if (running) |
03b7fad1 | 4561 | set_next_task(rq, p); |
cb469845 | 4562 | |
da7a735e | 4563 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 4564 | out_unlock: |
d1ccc66d IM |
4565 | /* Avoid rq from going away on us: */ |
4566 | preempt_disable(); | |
eb580751 | 4567 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
4568 | |
4569 | balance_callback(rq); | |
4570 | preempt_enable(); | |
b29739f9 | 4571 | } |
acd58620 PZ |
4572 | #else |
4573 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
4574 | { | |
4575 | return prio; | |
4576 | } | |
b29739f9 | 4577 | #endif |
d50dde5a | 4578 | |
36c8b586 | 4579 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4580 | { |
49bd21ef | 4581 | bool queued, running; |
53a23364 | 4582 | int old_prio; |
eb580751 | 4583 | struct rq_flags rf; |
70b97a7f | 4584 | struct rq *rq; |
1da177e4 | 4585 | |
75e45d51 | 4586 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
4587 | return; |
4588 | /* | |
4589 | * We have to be careful, if called from sys_setpriority(), | |
4590 | * the task might be in the middle of scheduling on another CPU. | |
4591 | */ | |
eb580751 | 4592 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
4593 | update_rq_clock(rq); |
4594 | ||
1da177e4 LT |
4595 | /* |
4596 | * The RT priorities are set via sched_setscheduler(), but we still | |
4597 | * allow the 'normal' nice value to be set - but as expected | |
4598 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 4599 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 4600 | */ |
aab03e05 | 4601 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
4602 | p->static_prio = NICE_TO_PRIO(nice); |
4603 | goto out_unlock; | |
4604 | } | |
da0c1e65 | 4605 | queued = task_on_rq_queued(p); |
49bd21ef | 4606 | running = task_current(rq, p); |
da0c1e65 | 4607 | if (queued) |
7a57f32a | 4608 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
4609 | if (running) |
4610 | put_prev_task(rq, p); | |
1da177e4 | 4611 | |
1da177e4 | 4612 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 4613 | set_load_weight(p, true); |
b29739f9 IM |
4614 | old_prio = p->prio; |
4615 | p->prio = effective_prio(p); | |
1da177e4 | 4616 | |
5443a0be | 4617 | if (queued) |
7134b3e9 | 4618 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 4619 | if (running) |
03b7fad1 | 4620 | set_next_task(rq, p); |
5443a0be FW |
4621 | |
4622 | /* | |
4623 | * If the task increased its priority or is running and | |
4624 | * lowered its priority, then reschedule its CPU: | |
4625 | */ | |
4626 | p->sched_class->prio_changed(rq, p, old_prio); | |
4627 | ||
1da177e4 | 4628 | out_unlock: |
eb580751 | 4629 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 4630 | } |
1da177e4 LT |
4631 | EXPORT_SYMBOL(set_user_nice); |
4632 | ||
e43379f1 MM |
4633 | /* |
4634 | * can_nice - check if a task can reduce its nice value | |
4635 | * @p: task | |
4636 | * @nice: nice value | |
4637 | */ | |
36c8b586 | 4638 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4639 | { |
d1ccc66d | 4640 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 4641 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 4642 | |
78d7d407 | 4643 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4644 | capable(CAP_SYS_NICE)); |
4645 | } | |
4646 | ||
1da177e4 LT |
4647 | #ifdef __ARCH_WANT_SYS_NICE |
4648 | ||
4649 | /* | |
4650 | * sys_nice - change the priority of the current process. | |
4651 | * @increment: priority increment | |
4652 | * | |
4653 | * sys_setpriority is a more generic, but much slower function that | |
4654 | * does similar things. | |
4655 | */ | |
5add95d4 | 4656 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4657 | { |
48f24c4d | 4658 | long nice, retval; |
1da177e4 LT |
4659 | |
4660 | /* | |
4661 | * Setpriority might change our priority at the same moment. | |
4662 | * We don't have to worry. Conceptually one call occurs first | |
4663 | * and we have a single winner. | |
4664 | */ | |
a9467fa3 | 4665 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 4666 | nice = task_nice(current) + increment; |
1da177e4 | 4667 | |
a9467fa3 | 4668 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
4669 | if (increment < 0 && !can_nice(current, nice)) |
4670 | return -EPERM; | |
4671 | ||
1da177e4 LT |
4672 | retval = security_task_setnice(current, nice); |
4673 | if (retval) | |
4674 | return retval; | |
4675 | ||
4676 | set_user_nice(current, nice); | |
4677 | return 0; | |
4678 | } | |
4679 | ||
4680 | #endif | |
4681 | ||
4682 | /** | |
4683 | * task_prio - return the priority value of a given task. | |
4684 | * @p: the task in question. | |
4685 | * | |
e69f6186 | 4686 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
4687 | * RT tasks are offset by -200. Normal tasks are centered |
4688 | * around 0, value goes from -16 to +15. | |
4689 | */ | |
36c8b586 | 4690 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4691 | { |
4692 | return p->prio - MAX_RT_PRIO; | |
4693 | } | |
4694 | ||
1da177e4 | 4695 | /** |
d1ccc66d | 4696 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 4697 | * @cpu: the processor in question. |
e69f6186 YB |
4698 | * |
4699 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
4700 | */ |
4701 | int idle_cpu(int cpu) | |
4702 | { | |
908a3283 TG |
4703 | struct rq *rq = cpu_rq(cpu); |
4704 | ||
4705 | if (rq->curr != rq->idle) | |
4706 | return 0; | |
4707 | ||
4708 | if (rq->nr_running) | |
4709 | return 0; | |
4710 | ||
4711 | #ifdef CONFIG_SMP | |
126c2092 | 4712 | if (rq->ttwu_pending) |
908a3283 TG |
4713 | return 0; |
4714 | #endif | |
4715 | ||
4716 | return 1; | |
1da177e4 LT |
4717 | } |
4718 | ||
943d355d RJ |
4719 | /** |
4720 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
4721 | * @cpu: the CPU in question. | |
4722 | * | |
4723 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
4724 | */ | |
4725 | int available_idle_cpu(int cpu) | |
4726 | { | |
4727 | if (!idle_cpu(cpu)) | |
4728 | return 0; | |
4729 | ||
247f2f6f RJ |
4730 | if (vcpu_is_preempted(cpu)) |
4731 | return 0; | |
4732 | ||
908a3283 | 4733 | return 1; |
1da177e4 LT |
4734 | } |
4735 | ||
1da177e4 | 4736 | /** |
d1ccc66d | 4737 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 4738 | * @cpu: the processor in question. |
e69f6186 | 4739 | * |
d1ccc66d | 4740 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 4741 | */ |
36c8b586 | 4742 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4743 | { |
4744 | return cpu_rq(cpu)->idle; | |
4745 | } | |
4746 | ||
4747 | /** | |
4748 | * find_process_by_pid - find a process with a matching PID value. | |
4749 | * @pid: the pid in question. | |
e69f6186 YB |
4750 | * |
4751 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 4752 | */ |
a9957449 | 4753 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4754 | { |
228ebcbe | 4755 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4756 | } |
4757 | ||
c13db6b1 SR |
4758 | /* |
4759 | * sched_setparam() passes in -1 for its policy, to let the functions | |
4760 | * it calls know not to change it. | |
4761 | */ | |
4762 | #define SETPARAM_POLICY -1 | |
4763 | ||
c365c292 TG |
4764 | static void __setscheduler_params(struct task_struct *p, |
4765 | const struct sched_attr *attr) | |
1da177e4 | 4766 | { |
d50dde5a DF |
4767 | int policy = attr->sched_policy; |
4768 | ||
c13db6b1 | 4769 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
4770 | policy = p->policy; |
4771 | ||
1da177e4 | 4772 | p->policy = policy; |
d50dde5a | 4773 | |
aab03e05 DF |
4774 | if (dl_policy(policy)) |
4775 | __setparam_dl(p, attr); | |
39fd8fd2 | 4776 | else if (fair_policy(policy)) |
d50dde5a DF |
4777 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
4778 | ||
39fd8fd2 PZ |
4779 | /* |
4780 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
4781 | * !rt_policy. Always setting this ensures that things like | |
4782 | * getparam()/getattr() don't report silly values for !rt tasks. | |
4783 | */ | |
4784 | p->rt_priority = attr->sched_priority; | |
383afd09 | 4785 | p->normal_prio = normal_prio(p); |
9059393e | 4786 | set_load_weight(p, true); |
c365c292 | 4787 | } |
39fd8fd2 | 4788 | |
c365c292 TG |
4789 | /* Actually do priority change: must hold pi & rq lock. */ |
4790 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 4791 | const struct sched_attr *attr, bool keep_boost) |
c365c292 | 4792 | { |
a509a7cd PB |
4793 | /* |
4794 | * If params can't change scheduling class changes aren't allowed | |
4795 | * either. | |
4796 | */ | |
4797 | if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) | |
4798 | return; | |
4799 | ||
c365c292 | 4800 | __setscheduler_params(p, attr); |
d50dde5a | 4801 | |
383afd09 | 4802 | /* |
0782e63b TG |
4803 | * Keep a potential priority boosting if called from |
4804 | * sched_setscheduler(). | |
383afd09 | 4805 | */ |
acd58620 | 4806 | p->prio = normal_prio(p); |
0782e63b | 4807 | if (keep_boost) |
acd58620 | 4808 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 4809 | |
aab03e05 DF |
4810 | if (dl_prio(p->prio)) |
4811 | p->sched_class = &dl_sched_class; | |
4812 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
4813 | p->sched_class = &rt_sched_class; |
4814 | else | |
4815 | p->sched_class = &fair_sched_class; | |
1da177e4 | 4816 | } |
aab03e05 | 4817 | |
c69e8d9c | 4818 | /* |
d1ccc66d | 4819 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
4820 | */ |
4821 | static bool check_same_owner(struct task_struct *p) | |
4822 | { | |
4823 | const struct cred *cred = current_cred(), *pcred; | |
4824 | bool match; | |
4825 | ||
4826 | rcu_read_lock(); | |
4827 | pcred = __task_cred(p); | |
9c806aa0 EB |
4828 | match = (uid_eq(cred->euid, pcred->euid) || |
4829 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
4830 | rcu_read_unlock(); |
4831 | return match; | |
4832 | } | |
4833 | ||
d50dde5a DF |
4834 | static int __sched_setscheduler(struct task_struct *p, |
4835 | const struct sched_attr *attr, | |
dbc7f069 | 4836 | bool user, bool pi) |
1da177e4 | 4837 | { |
383afd09 SR |
4838 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
4839 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 4840 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 4841 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 4842 | const struct sched_class *prev_class; |
eb580751 | 4843 | struct rq_flags rf; |
ca94c442 | 4844 | int reset_on_fork; |
7a57f32a | 4845 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 4846 | struct rq *rq; |
1da177e4 | 4847 | |
896bbb25 SRV |
4848 | /* The pi code expects interrupts enabled */ |
4849 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 4850 | recheck: |
d1ccc66d | 4851 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
4852 | if (policy < 0) { |
4853 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4854 | policy = oldpolicy = p->policy; |
ca94c442 | 4855 | } else { |
7479f3c9 | 4856 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 4857 | |
20f9cd2a | 4858 | if (!valid_policy(policy)) |
ca94c442 LP |
4859 | return -EINVAL; |
4860 | } | |
4861 | ||
794a56eb | 4862 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
4863 | return -EINVAL; |
4864 | ||
1da177e4 LT |
4865 | /* |
4866 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4867 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4868 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 4869 | */ |
0bb040a4 | 4870 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 4871 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4872 | return -EINVAL; |
aab03e05 DF |
4873 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
4874 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
4875 | return -EINVAL; |
4876 | ||
37e4ab3f OC |
4877 | /* |
4878 | * Allow unprivileged RT tasks to decrease priority: | |
4879 | */ | |
961ccddd | 4880 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 4881 | if (fair_policy(policy)) { |
d0ea0268 | 4882 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 4883 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
4884 | return -EPERM; |
4885 | } | |
4886 | ||
e05606d3 | 4887 | if (rt_policy(policy)) { |
a44702e8 ON |
4888 | unsigned long rlim_rtprio = |
4889 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 4890 | |
d1ccc66d | 4891 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
4892 | if (policy != p->policy && !rlim_rtprio) |
4893 | return -EPERM; | |
4894 | ||
d1ccc66d | 4895 | /* Can't increase priority: */ |
d50dde5a DF |
4896 | if (attr->sched_priority > p->rt_priority && |
4897 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
4898 | return -EPERM; |
4899 | } | |
c02aa73b | 4900 | |
d44753b8 JL |
4901 | /* |
4902 | * Can't set/change SCHED_DEADLINE policy at all for now | |
4903 | * (safest behavior); in the future we would like to allow | |
4904 | * unprivileged DL tasks to increase their relative deadline | |
4905 | * or reduce their runtime (both ways reducing utilization) | |
4906 | */ | |
4907 | if (dl_policy(policy)) | |
4908 | return -EPERM; | |
4909 | ||
dd41f596 | 4910 | /* |
c02aa73b DH |
4911 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
4912 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 4913 | */ |
1da1843f | 4914 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 4915 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
4916 | return -EPERM; |
4917 | } | |
5fe1d75f | 4918 | |
d1ccc66d | 4919 | /* Can't change other user's priorities: */ |
c69e8d9c | 4920 | if (!check_same_owner(p)) |
37e4ab3f | 4921 | return -EPERM; |
ca94c442 | 4922 | |
d1ccc66d | 4923 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
4924 | if (p->sched_reset_on_fork && !reset_on_fork) |
4925 | return -EPERM; | |
37e4ab3f | 4926 | } |
1da177e4 | 4927 | |
725aad24 | 4928 | if (user) { |
794a56eb JL |
4929 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
4930 | return -EINVAL; | |
4931 | ||
b0ae1981 | 4932 | retval = security_task_setscheduler(p); |
725aad24 JF |
4933 | if (retval) |
4934 | return retval; | |
4935 | } | |
4936 | ||
a509a7cd PB |
4937 | /* Update task specific "requested" clamps */ |
4938 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
4939 | retval = uclamp_validate(p, attr); | |
4940 | if (retval) | |
4941 | return retval; | |
4942 | } | |
4943 | ||
710da3c8 JL |
4944 | if (pi) |
4945 | cpuset_read_lock(); | |
4946 | ||
b29739f9 | 4947 | /* |
d1ccc66d | 4948 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 4949 | * changing the priority of the task: |
0122ec5b | 4950 | * |
25985edc | 4951 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
4952 | * runqueue lock must be held. |
4953 | */ | |
eb580751 | 4954 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 4955 | update_rq_clock(rq); |
dc61b1d6 | 4956 | |
34f971f6 | 4957 | /* |
d1ccc66d | 4958 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
4959 | */ |
4960 | if (p == rq->stop) { | |
4b211f2b MP |
4961 | retval = -EINVAL; |
4962 | goto unlock; | |
34f971f6 PZ |
4963 | } |
4964 | ||
a51e9198 | 4965 | /* |
d6b1e911 TG |
4966 | * If not changing anything there's no need to proceed further, |
4967 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 4968 | */ |
d50dde5a | 4969 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 4970 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
4971 | goto change; |
4972 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
4973 | goto change; | |
75381608 | 4974 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 4975 | goto change; |
a509a7cd PB |
4976 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
4977 | goto change; | |
d50dde5a | 4978 | |
d6b1e911 | 4979 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
4980 | retval = 0; |
4981 | goto unlock; | |
a51e9198 | 4982 | } |
d50dde5a | 4983 | change: |
a51e9198 | 4984 | |
dc61b1d6 | 4985 | if (user) { |
332ac17e | 4986 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
4987 | /* |
4988 | * Do not allow realtime tasks into groups that have no runtime | |
4989 | * assigned. | |
4990 | */ | |
4991 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4992 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4993 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
4994 | retval = -EPERM; |
4995 | goto unlock; | |
dc61b1d6 | 4996 | } |
dc61b1d6 | 4997 | #endif |
332ac17e | 4998 | #ifdef CONFIG_SMP |
794a56eb JL |
4999 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
5000 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 5001 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
5002 | |
5003 | /* | |
5004 | * Don't allow tasks with an affinity mask smaller than | |
5005 | * the entire root_domain to become SCHED_DEADLINE. We | |
5006 | * will also fail if there's no bandwidth available. | |
5007 | */ | |
3bd37062 | 5008 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 5009 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
5010 | retval = -EPERM; |
5011 | goto unlock; | |
332ac17e DF |
5012 | } |
5013 | } | |
5014 | #endif | |
5015 | } | |
dc61b1d6 | 5016 | |
d1ccc66d | 5017 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
5018 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
5019 | policy = oldpolicy = -1; | |
eb580751 | 5020 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
5021 | if (pi) |
5022 | cpuset_read_unlock(); | |
1da177e4 LT |
5023 | goto recheck; |
5024 | } | |
332ac17e DF |
5025 | |
5026 | /* | |
5027 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
5028 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
5029 | * is available. | |
5030 | */ | |
06a76fe0 | 5031 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
5032 | retval = -EBUSY; |
5033 | goto unlock; | |
332ac17e DF |
5034 | } |
5035 | ||
c365c292 TG |
5036 | p->sched_reset_on_fork = reset_on_fork; |
5037 | oldprio = p->prio; | |
5038 | ||
dbc7f069 PZ |
5039 | if (pi) { |
5040 | /* | |
5041 | * Take priority boosted tasks into account. If the new | |
5042 | * effective priority is unchanged, we just store the new | |
5043 | * normal parameters and do not touch the scheduler class and | |
5044 | * the runqueue. This will be done when the task deboost | |
5045 | * itself. | |
5046 | */ | |
acd58620 | 5047 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
5048 | if (new_effective_prio == oldprio) |
5049 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
5050 | } |
5051 | ||
da0c1e65 | 5052 | queued = task_on_rq_queued(p); |
051a1d1a | 5053 | running = task_current(rq, p); |
da0c1e65 | 5054 | if (queued) |
ff77e468 | 5055 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 5056 | if (running) |
f3cd1c4e | 5057 | put_prev_task(rq, p); |
f6b53205 | 5058 | |
83ab0aa0 | 5059 | prev_class = p->sched_class; |
a509a7cd | 5060 | |
dbc7f069 | 5061 | __setscheduler(rq, p, attr, pi); |
a509a7cd | 5062 | __setscheduler_uclamp(p, attr); |
f6b53205 | 5063 | |
da0c1e65 | 5064 | if (queued) { |
81a44c54 TG |
5065 | /* |
5066 | * We enqueue to tail when the priority of a task is | |
5067 | * increased (user space view). | |
5068 | */ | |
ff77e468 PZ |
5069 | if (oldprio < p->prio) |
5070 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 5071 | |
ff77e468 | 5072 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 5073 | } |
a399d233 | 5074 | if (running) |
03b7fad1 | 5075 | set_next_task(rq, p); |
cb469845 | 5076 | |
da7a735e | 5077 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
5078 | |
5079 | /* Avoid rq from going away on us: */ | |
5080 | preempt_disable(); | |
eb580751 | 5081 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 5082 | |
710da3c8 JL |
5083 | if (pi) { |
5084 | cpuset_read_unlock(); | |
dbc7f069 | 5085 | rt_mutex_adjust_pi(p); |
710da3c8 | 5086 | } |
95e02ca9 | 5087 | |
d1ccc66d | 5088 | /* Run balance callbacks after we've adjusted the PI chain: */ |
4c9a4bc8 PZ |
5089 | balance_callback(rq); |
5090 | preempt_enable(); | |
95e02ca9 | 5091 | |
1da177e4 | 5092 | return 0; |
4b211f2b MP |
5093 | |
5094 | unlock: | |
5095 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
5096 | if (pi) |
5097 | cpuset_read_unlock(); | |
4b211f2b | 5098 | return retval; |
1da177e4 | 5099 | } |
961ccddd | 5100 | |
7479f3c9 PZ |
5101 | static int _sched_setscheduler(struct task_struct *p, int policy, |
5102 | const struct sched_param *param, bool check) | |
5103 | { | |
5104 | struct sched_attr attr = { | |
5105 | .sched_policy = policy, | |
5106 | .sched_priority = param->sched_priority, | |
5107 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
5108 | }; | |
5109 | ||
c13db6b1 SR |
5110 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
5111 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
5112 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
5113 | policy &= ~SCHED_RESET_ON_FORK; | |
5114 | attr.sched_policy = policy; | |
5115 | } | |
5116 | ||
dbc7f069 | 5117 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 5118 | } |
961ccddd RR |
5119 | /** |
5120 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5121 | * @p: the task in question. | |
5122 | * @policy: new policy. | |
5123 | * @param: structure containing the new RT priority. | |
5124 | * | |
e69f6186 YB |
5125 | * Return: 0 on success. An error code otherwise. |
5126 | * | |
961ccddd RR |
5127 | * NOTE that the task may be already dead. |
5128 | */ | |
5129 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5130 | const struct sched_param *param) |
961ccddd | 5131 | { |
7479f3c9 | 5132 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 5133 | } |
1da177e4 LT |
5134 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5135 | ||
d50dde5a DF |
5136 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
5137 | { | |
dbc7f069 | 5138 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a DF |
5139 | } |
5140 | EXPORT_SYMBOL_GPL(sched_setattr); | |
5141 | ||
794a56eb JL |
5142 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
5143 | { | |
5144 | return __sched_setscheduler(p, attr, false, true); | |
5145 | } | |
5146 | ||
961ccddd RR |
5147 | /** |
5148 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5149 | * @p: the task in question. | |
5150 | * @policy: new policy. | |
5151 | * @param: structure containing the new RT priority. | |
5152 | * | |
5153 | * Just like sched_setscheduler, only don't bother checking if the | |
5154 | * current context has permission. For example, this is needed in | |
5155 | * stop_machine(): we create temporary high priority worker threads, | |
5156 | * but our caller might not have that capability. | |
e69f6186 YB |
5157 | * |
5158 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
5159 | */ |
5160 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5161 | const struct sched_param *param) |
961ccddd | 5162 | { |
7479f3c9 | 5163 | return _sched_setscheduler(p, policy, param, false); |
961ccddd | 5164 | } |
84778472 | 5165 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
961ccddd | 5166 | |
95cdf3b7 IM |
5167 | static int |
5168 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5169 | { |
1da177e4 LT |
5170 | struct sched_param lparam; |
5171 | struct task_struct *p; | |
36c8b586 | 5172 | int retval; |
1da177e4 LT |
5173 | |
5174 | if (!param || pid < 0) | |
5175 | return -EINVAL; | |
5176 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5177 | return -EFAULT; | |
5fe1d75f ON |
5178 | |
5179 | rcu_read_lock(); | |
5180 | retval = -ESRCH; | |
1da177e4 | 5181 | p = find_process_by_pid(pid); |
710da3c8 JL |
5182 | if (likely(p)) |
5183 | get_task_struct(p); | |
5fe1d75f | 5184 | rcu_read_unlock(); |
36c8b586 | 5185 | |
710da3c8 JL |
5186 | if (likely(p)) { |
5187 | retval = sched_setscheduler(p, policy, &lparam); | |
5188 | put_task_struct(p); | |
5189 | } | |
5190 | ||
1da177e4 LT |
5191 | return retval; |
5192 | } | |
5193 | ||
d50dde5a DF |
5194 | /* |
5195 | * Mimics kernel/events/core.c perf_copy_attr(). | |
5196 | */ | |
d1ccc66d | 5197 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
5198 | { |
5199 | u32 size; | |
5200 | int ret; | |
5201 | ||
d1ccc66d | 5202 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
5203 | memset(attr, 0, sizeof(*attr)); |
5204 | ||
5205 | ret = get_user(size, &uattr->size); | |
5206 | if (ret) | |
5207 | return ret; | |
5208 | ||
d1ccc66d IM |
5209 | /* ABI compatibility quirk: */ |
5210 | if (!size) | |
d50dde5a | 5211 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 5212 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
5213 | goto err_size; |
5214 | ||
dff3a85f AS |
5215 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
5216 | if (ret) { | |
5217 | if (ret == -E2BIG) | |
5218 | goto err_size; | |
5219 | return ret; | |
d50dde5a DF |
5220 | } |
5221 | ||
a509a7cd PB |
5222 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
5223 | size < SCHED_ATTR_SIZE_VER1) | |
5224 | return -EINVAL; | |
5225 | ||
d50dde5a | 5226 | /* |
d1ccc66d | 5227 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
5228 | * to be strict and return an error on out-of-bounds values? |
5229 | */ | |
75e45d51 | 5230 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 5231 | |
e78c7bca | 5232 | return 0; |
d50dde5a DF |
5233 | |
5234 | err_size: | |
5235 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 5236 | return -E2BIG; |
d50dde5a DF |
5237 | } |
5238 | ||
1da177e4 LT |
5239 | /** |
5240 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5241 | * @pid: the pid in question. | |
5242 | * @policy: new policy. | |
5243 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
5244 | * |
5245 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5246 | */ |
d1ccc66d | 5247 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 5248 | { |
c21761f1 JB |
5249 | if (policy < 0) |
5250 | return -EINVAL; | |
5251 | ||
1da177e4 LT |
5252 | return do_sched_setscheduler(pid, policy, param); |
5253 | } | |
5254 | ||
5255 | /** | |
5256 | * sys_sched_setparam - set/change the RT priority of a thread | |
5257 | * @pid: the pid in question. | |
5258 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
5259 | * |
5260 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5261 | */ |
5add95d4 | 5262 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 5263 | { |
c13db6b1 | 5264 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
5265 | } |
5266 | ||
d50dde5a DF |
5267 | /** |
5268 | * sys_sched_setattr - same as above, but with extended sched_attr | |
5269 | * @pid: the pid in question. | |
5778fccf | 5270 | * @uattr: structure containing the extended parameters. |
db66d756 | 5271 | * @flags: for future extension. |
d50dde5a | 5272 | */ |
6d35ab48 PZ |
5273 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
5274 | unsigned int, flags) | |
d50dde5a DF |
5275 | { |
5276 | struct sched_attr attr; | |
5277 | struct task_struct *p; | |
5278 | int retval; | |
5279 | ||
6d35ab48 | 5280 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
5281 | return -EINVAL; |
5282 | ||
143cf23d MK |
5283 | retval = sched_copy_attr(uattr, &attr); |
5284 | if (retval) | |
5285 | return retval; | |
d50dde5a | 5286 | |
b14ed2c2 | 5287 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 5288 | return -EINVAL; |
1d6362fa PB |
5289 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
5290 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
5291 | |
5292 | rcu_read_lock(); | |
5293 | retval = -ESRCH; | |
5294 | p = find_process_by_pid(pid); | |
a509a7cd PB |
5295 | if (likely(p)) |
5296 | get_task_struct(p); | |
d50dde5a DF |
5297 | rcu_read_unlock(); |
5298 | ||
a509a7cd PB |
5299 | if (likely(p)) { |
5300 | retval = sched_setattr(p, &attr); | |
5301 | put_task_struct(p); | |
5302 | } | |
5303 | ||
d50dde5a DF |
5304 | return retval; |
5305 | } | |
5306 | ||
1da177e4 LT |
5307 | /** |
5308 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5309 | * @pid: the pid in question. | |
e69f6186 YB |
5310 | * |
5311 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
5312 | * code. | |
1da177e4 | 5313 | */ |
5add95d4 | 5314 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5315 | { |
36c8b586 | 5316 | struct task_struct *p; |
3a5c359a | 5317 | int retval; |
1da177e4 LT |
5318 | |
5319 | if (pid < 0) | |
3a5c359a | 5320 | return -EINVAL; |
1da177e4 LT |
5321 | |
5322 | retval = -ESRCH; | |
5fe85be0 | 5323 | rcu_read_lock(); |
1da177e4 LT |
5324 | p = find_process_by_pid(pid); |
5325 | if (p) { | |
5326 | retval = security_task_getscheduler(p); | |
5327 | if (!retval) | |
ca94c442 LP |
5328 | retval = p->policy |
5329 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5330 | } |
5fe85be0 | 5331 | rcu_read_unlock(); |
1da177e4 LT |
5332 | return retval; |
5333 | } | |
5334 | ||
5335 | /** | |
ca94c442 | 5336 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5337 | * @pid: the pid in question. |
5338 | * @param: structure containing the RT priority. | |
e69f6186 YB |
5339 | * |
5340 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
5341 | * code. | |
1da177e4 | 5342 | */ |
5add95d4 | 5343 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 5344 | { |
ce5f7f82 | 5345 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 5346 | struct task_struct *p; |
3a5c359a | 5347 | int retval; |
1da177e4 LT |
5348 | |
5349 | if (!param || pid < 0) | |
3a5c359a | 5350 | return -EINVAL; |
1da177e4 | 5351 | |
5fe85be0 | 5352 | rcu_read_lock(); |
1da177e4 LT |
5353 | p = find_process_by_pid(pid); |
5354 | retval = -ESRCH; | |
5355 | if (!p) | |
5356 | goto out_unlock; | |
5357 | ||
5358 | retval = security_task_getscheduler(p); | |
5359 | if (retval) | |
5360 | goto out_unlock; | |
5361 | ||
ce5f7f82 PZ |
5362 | if (task_has_rt_policy(p)) |
5363 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5364 | rcu_read_unlock(); |
1da177e4 LT |
5365 | |
5366 | /* | |
5367 | * This one might sleep, we cannot do it with a spinlock held ... | |
5368 | */ | |
5369 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5370 | ||
1da177e4 LT |
5371 | return retval; |
5372 | ||
5373 | out_unlock: | |
5fe85be0 | 5374 | rcu_read_unlock(); |
1da177e4 LT |
5375 | return retval; |
5376 | } | |
5377 | ||
1251201c IM |
5378 | /* |
5379 | * Copy the kernel size attribute structure (which might be larger | |
5380 | * than what user-space knows about) to user-space. | |
5381 | * | |
5382 | * Note that all cases are valid: user-space buffer can be larger or | |
5383 | * smaller than the kernel-space buffer. The usual case is that both | |
5384 | * have the same size. | |
5385 | */ | |
5386 | static int | |
5387 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
5388 | struct sched_attr *kattr, | |
5389 | unsigned int usize) | |
d50dde5a | 5390 | { |
1251201c | 5391 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 5392 | |
96d4f267 | 5393 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
5394 | return -EFAULT; |
5395 | ||
5396 | /* | |
1251201c IM |
5397 | * sched_getattr() ABI forwards and backwards compatibility: |
5398 | * | |
5399 | * If usize == ksize then we just copy everything to user-space and all is good. | |
5400 | * | |
5401 | * If usize < ksize then we only copy as much as user-space has space for, | |
5402 | * this keeps ABI compatibility as well. We skip the rest. | |
5403 | * | |
5404 | * If usize > ksize then user-space is using a newer version of the ABI, | |
5405 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
5406 | * detect the kernel's knowledge of attributes from the attr->size value | |
5407 | * which is set to ksize in this case. | |
d50dde5a | 5408 | */ |
1251201c | 5409 | kattr->size = min(usize, ksize); |
d50dde5a | 5410 | |
1251201c | 5411 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
5412 | return -EFAULT; |
5413 | ||
22400674 | 5414 | return 0; |
d50dde5a DF |
5415 | } |
5416 | ||
5417 | /** | |
aab03e05 | 5418 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 5419 | * @pid: the pid in question. |
5778fccf | 5420 | * @uattr: structure containing the extended parameters. |
dff3a85f | 5421 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 5422 | * @flags: for future extension. |
d50dde5a | 5423 | */ |
6d35ab48 | 5424 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 5425 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 5426 | { |
1251201c | 5427 | struct sched_attr kattr = { }; |
d50dde5a DF |
5428 | struct task_struct *p; |
5429 | int retval; | |
5430 | ||
1251201c IM |
5431 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
5432 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
5433 | return -EINVAL; |
5434 | ||
5435 | rcu_read_lock(); | |
5436 | p = find_process_by_pid(pid); | |
5437 | retval = -ESRCH; | |
5438 | if (!p) | |
5439 | goto out_unlock; | |
5440 | ||
5441 | retval = security_task_getscheduler(p); | |
5442 | if (retval) | |
5443 | goto out_unlock; | |
5444 | ||
1251201c | 5445 | kattr.sched_policy = p->policy; |
7479f3c9 | 5446 | if (p->sched_reset_on_fork) |
1251201c | 5447 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
aab03e05 | 5448 | if (task_has_dl_policy(p)) |
1251201c | 5449 | __getparam_dl(p, &kattr); |
aab03e05 | 5450 | else if (task_has_rt_policy(p)) |
1251201c | 5451 | kattr.sched_priority = p->rt_priority; |
d50dde5a | 5452 | else |
1251201c | 5453 | kattr.sched_nice = task_nice(p); |
d50dde5a | 5454 | |
a509a7cd | 5455 | #ifdef CONFIG_UCLAMP_TASK |
1251201c IM |
5456 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
5457 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
5458 | #endif |
5459 | ||
d50dde5a DF |
5460 | rcu_read_unlock(); |
5461 | ||
1251201c | 5462 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
5463 | |
5464 | out_unlock: | |
5465 | rcu_read_unlock(); | |
5466 | return retval; | |
5467 | } | |
5468 | ||
96f874e2 | 5469 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5470 | { |
5a16f3d3 | 5471 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5472 | struct task_struct *p; |
5473 | int retval; | |
1da177e4 | 5474 | |
23f5d142 | 5475 | rcu_read_lock(); |
1da177e4 LT |
5476 | |
5477 | p = find_process_by_pid(pid); | |
5478 | if (!p) { | |
23f5d142 | 5479 | rcu_read_unlock(); |
1da177e4 LT |
5480 | return -ESRCH; |
5481 | } | |
5482 | ||
23f5d142 | 5483 | /* Prevent p going away */ |
1da177e4 | 5484 | get_task_struct(p); |
23f5d142 | 5485 | rcu_read_unlock(); |
1da177e4 | 5486 | |
14a40ffc TH |
5487 | if (p->flags & PF_NO_SETAFFINITY) { |
5488 | retval = -EINVAL; | |
5489 | goto out_put_task; | |
5490 | } | |
5a16f3d3 RR |
5491 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5492 | retval = -ENOMEM; | |
5493 | goto out_put_task; | |
5494 | } | |
5495 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5496 | retval = -ENOMEM; | |
5497 | goto out_free_cpus_allowed; | |
5498 | } | |
1da177e4 | 5499 | retval = -EPERM; |
4c44aaaf EB |
5500 | if (!check_same_owner(p)) { |
5501 | rcu_read_lock(); | |
5502 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
5503 | rcu_read_unlock(); | |
16303ab2 | 5504 | goto out_free_new_mask; |
4c44aaaf EB |
5505 | } |
5506 | rcu_read_unlock(); | |
5507 | } | |
1da177e4 | 5508 | |
b0ae1981 | 5509 | retval = security_task_setscheduler(p); |
e7834f8f | 5510 | if (retval) |
16303ab2 | 5511 | goto out_free_new_mask; |
e7834f8f | 5512 | |
e4099a5e PZ |
5513 | |
5514 | cpuset_cpus_allowed(p, cpus_allowed); | |
5515 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
5516 | ||
332ac17e DF |
5517 | /* |
5518 | * Since bandwidth control happens on root_domain basis, | |
5519 | * if admission test is enabled, we only admit -deadline | |
5520 | * tasks allowed to run on all the CPUs in the task's | |
5521 | * root_domain. | |
5522 | */ | |
5523 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
5524 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
5525 | rcu_read_lock(); | |
5526 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 5527 | retval = -EBUSY; |
f1e3a093 | 5528 | rcu_read_unlock(); |
16303ab2 | 5529 | goto out_free_new_mask; |
332ac17e | 5530 | } |
f1e3a093 | 5531 | rcu_read_unlock(); |
332ac17e DF |
5532 | } |
5533 | #endif | |
49246274 | 5534 | again: |
25834c73 | 5535 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 5536 | |
8707d8b8 | 5537 | if (!retval) { |
5a16f3d3 RR |
5538 | cpuset_cpus_allowed(p, cpus_allowed); |
5539 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5540 | /* |
5541 | * We must have raced with a concurrent cpuset | |
5542 | * update. Just reset the cpus_allowed to the | |
5543 | * cpuset's cpus_allowed | |
5544 | */ | |
5a16f3d3 | 5545 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5546 | goto again; |
5547 | } | |
5548 | } | |
16303ab2 | 5549 | out_free_new_mask: |
5a16f3d3 RR |
5550 | free_cpumask_var(new_mask); |
5551 | out_free_cpus_allowed: | |
5552 | free_cpumask_var(cpus_allowed); | |
5553 | out_put_task: | |
1da177e4 | 5554 | put_task_struct(p); |
1da177e4 LT |
5555 | return retval; |
5556 | } | |
5557 | ||
5558 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5559 | struct cpumask *new_mask) |
1da177e4 | 5560 | { |
96f874e2 RR |
5561 | if (len < cpumask_size()) |
5562 | cpumask_clear(new_mask); | |
5563 | else if (len > cpumask_size()) | |
5564 | len = cpumask_size(); | |
5565 | ||
1da177e4 LT |
5566 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5567 | } | |
5568 | ||
5569 | /** | |
d1ccc66d | 5570 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
5571 | * @pid: pid of the process |
5572 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 5573 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
5574 | * |
5575 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5576 | */ |
5add95d4 HC |
5577 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5578 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5579 | { |
5a16f3d3 | 5580 | cpumask_var_t new_mask; |
1da177e4 LT |
5581 | int retval; |
5582 | ||
5a16f3d3 RR |
5583 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5584 | return -ENOMEM; | |
1da177e4 | 5585 | |
5a16f3d3 RR |
5586 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5587 | if (retval == 0) | |
5588 | retval = sched_setaffinity(pid, new_mask); | |
5589 | free_cpumask_var(new_mask); | |
5590 | return retval; | |
1da177e4 LT |
5591 | } |
5592 | ||
96f874e2 | 5593 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5594 | { |
36c8b586 | 5595 | struct task_struct *p; |
31605683 | 5596 | unsigned long flags; |
1da177e4 | 5597 | int retval; |
1da177e4 | 5598 | |
23f5d142 | 5599 | rcu_read_lock(); |
1da177e4 LT |
5600 | |
5601 | retval = -ESRCH; | |
5602 | p = find_process_by_pid(pid); | |
5603 | if (!p) | |
5604 | goto out_unlock; | |
5605 | ||
e7834f8f DQ |
5606 | retval = security_task_getscheduler(p); |
5607 | if (retval) | |
5608 | goto out_unlock; | |
5609 | ||
013fdb80 | 5610 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 5611 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 5612 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5613 | |
5614 | out_unlock: | |
23f5d142 | 5615 | rcu_read_unlock(); |
1da177e4 | 5616 | |
9531b62f | 5617 | return retval; |
1da177e4 LT |
5618 | } |
5619 | ||
5620 | /** | |
d1ccc66d | 5621 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
5622 | * @pid: pid of the process |
5623 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 5624 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 5625 | * |
599b4840 ZW |
5626 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
5627 | * error code otherwise. | |
1da177e4 | 5628 | */ |
5add95d4 HC |
5629 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5630 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5631 | { |
5632 | int ret; | |
f17c8607 | 5633 | cpumask_var_t mask; |
1da177e4 | 5634 | |
84fba5ec | 5635 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5636 | return -EINVAL; |
5637 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5638 | return -EINVAL; |
5639 | ||
f17c8607 RR |
5640 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5641 | return -ENOMEM; | |
1da177e4 | 5642 | |
f17c8607 RR |
5643 | ret = sched_getaffinity(pid, mask); |
5644 | if (ret == 0) { | |
4de373a1 | 5645 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
5646 | |
5647 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5648 | ret = -EFAULT; |
5649 | else | |
cd3d8031 | 5650 | ret = retlen; |
f17c8607 RR |
5651 | } |
5652 | free_cpumask_var(mask); | |
1da177e4 | 5653 | |
f17c8607 | 5654 | return ret; |
1da177e4 LT |
5655 | } |
5656 | ||
5657 | /** | |
5658 | * sys_sched_yield - yield the current processor to other threads. | |
5659 | * | |
dd41f596 IM |
5660 | * This function yields the current CPU to other tasks. If there are no |
5661 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
5662 | * |
5663 | * Return: 0. | |
1da177e4 | 5664 | */ |
7d4dd4f1 | 5665 | static void do_sched_yield(void) |
1da177e4 | 5666 | { |
8a8c69c3 PZ |
5667 | struct rq_flags rf; |
5668 | struct rq *rq; | |
5669 | ||
246b3b33 | 5670 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 5671 | |
ae92882e | 5672 | schedstat_inc(rq->yld_count); |
4530d7ab | 5673 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5674 | |
5675 | /* | |
5676 | * Since we are going to call schedule() anyway, there's | |
5677 | * no need to preempt or enable interrupts: | |
5678 | */ | |
8a8c69c3 PZ |
5679 | preempt_disable(); |
5680 | rq_unlock(rq, &rf); | |
ba74c144 | 5681 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
5682 | |
5683 | schedule(); | |
7d4dd4f1 | 5684 | } |
1da177e4 | 5685 | |
7d4dd4f1 DB |
5686 | SYSCALL_DEFINE0(sched_yield) |
5687 | { | |
5688 | do_sched_yield(); | |
1da177e4 LT |
5689 | return 0; |
5690 | } | |
5691 | ||
c1a280b6 | 5692 | #ifndef CONFIG_PREEMPTION |
02b67cc3 | 5693 | int __sched _cond_resched(void) |
1da177e4 | 5694 | { |
fe32d3cd | 5695 | if (should_resched(0)) { |
a18b5d01 | 5696 | preempt_schedule_common(); |
1da177e4 LT |
5697 | return 1; |
5698 | } | |
f79c3ad6 | 5699 | rcu_all_qs(); |
1da177e4 LT |
5700 | return 0; |
5701 | } | |
02b67cc3 | 5702 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 5703 | #endif |
1da177e4 LT |
5704 | |
5705 | /* | |
613afbf8 | 5706 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5707 | * call schedule, and on return reacquire the lock. |
5708 | * | |
c1a280b6 | 5709 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
5710 | * operations here to prevent schedule() from being called twice (once via |
5711 | * spin_unlock(), once by hand). | |
5712 | */ | |
613afbf8 | 5713 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5714 | { |
fe32d3cd | 5715 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
5716 | int ret = 0; |
5717 | ||
f607c668 PZ |
5718 | lockdep_assert_held(lock); |
5719 | ||
4a81e832 | 5720 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5721 | spin_unlock(lock); |
d86ee480 | 5722 | if (resched) |
a18b5d01 | 5723 | preempt_schedule_common(); |
95c354fe NP |
5724 | else |
5725 | cpu_relax(); | |
6df3cecb | 5726 | ret = 1; |
1da177e4 | 5727 | spin_lock(lock); |
1da177e4 | 5728 | } |
6df3cecb | 5729 | return ret; |
1da177e4 | 5730 | } |
613afbf8 | 5731 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5732 | |
1da177e4 LT |
5733 | /** |
5734 | * yield - yield the current processor to other threads. | |
5735 | * | |
8e3fabfd PZ |
5736 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
5737 | * | |
5738 | * The scheduler is at all times free to pick the calling task as the most | |
5739 | * eligible task to run, if removing the yield() call from your code breaks | |
5740 | * it, its already broken. | |
5741 | * | |
5742 | * Typical broken usage is: | |
5743 | * | |
5744 | * while (!event) | |
d1ccc66d | 5745 | * yield(); |
8e3fabfd PZ |
5746 | * |
5747 | * where one assumes that yield() will let 'the other' process run that will | |
5748 | * make event true. If the current task is a SCHED_FIFO task that will never | |
5749 | * happen. Never use yield() as a progress guarantee!! | |
5750 | * | |
5751 | * If you want to use yield() to wait for something, use wait_event(). | |
5752 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
5753 | * If you still want to use yield(), do not! | |
1da177e4 LT |
5754 | */ |
5755 | void __sched yield(void) | |
5756 | { | |
5757 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 5758 | do_sched_yield(); |
1da177e4 | 5759 | } |
1da177e4 LT |
5760 | EXPORT_SYMBOL(yield); |
5761 | ||
d95f4122 MG |
5762 | /** |
5763 | * yield_to - yield the current processor to another thread in | |
5764 | * your thread group, or accelerate that thread toward the | |
5765 | * processor it's on. | |
16addf95 RD |
5766 | * @p: target task |
5767 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5768 | * |
5769 | * It's the caller's job to ensure that the target task struct | |
5770 | * can't go away on us before we can do any checks. | |
5771 | * | |
e69f6186 | 5772 | * Return: |
7b270f60 PZ |
5773 | * true (>0) if we indeed boosted the target task. |
5774 | * false (0) if we failed to boost the target. | |
5775 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 5776 | */ |
fa93384f | 5777 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
5778 | { |
5779 | struct task_struct *curr = current; | |
5780 | struct rq *rq, *p_rq; | |
5781 | unsigned long flags; | |
c3c18640 | 5782 | int yielded = 0; |
d95f4122 MG |
5783 | |
5784 | local_irq_save(flags); | |
5785 | rq = this_rq(); | |
5786 | ||
5787 | again: | |
5788 | p_rq = task_rq(p); | |
7b270f60 PZ |
5789 | /* |
5790 | * If we're the only runnable task on the rq and target rq also | |
5791 | * has only one task, there's absolutely no point in yielding. | |
5792 | */ | |
5793 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
5794 | yielded = -ESRCH; | |
5795 | goto out_irq; | |
5796 | } | |
5797 | ||
d95f4122 | 5798 | double_rq_lock(rq, p_rq); |
39e24d8f | 5799 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
5800 | double_rq_unlock(rq, p_rq); |
5801 | goto again; | |
5802 | } | |
5803 | ||
5804 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 5805 | goto out_unlock; |
d95f4122 MG |
5806 | |
5807 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 5808 | goto out_unlock; |
d95f4122 MG |
5809 | |
5810 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 5811 | goto out_unlock; |
d95f4122 MG |
5812 | |
5813 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5814 | if (yielded) { |
ae92882e | 5815 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
5816 | /* |
5817 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5818 | * fairness. | |
5819 | */ | |
5820 | if (preempt && rq != p_rq) | |
8875125e | 5821 | resched_curr(p_rq); |
6d1cafd8 | 5822 | } |
d95f4122 | 5823 | |
7b270f60 | 5824 | out_unlock: |
d95f4122 | 5825 | double_rq_unlock(rq, p_rq); |
7b270f60 | 5826 | out_irq: |
d95f4122 MG |
5827 | local_irq_restore(flags); |
5828 | ||
7b270f60 | 5829 | if (yielded > 0) |
d95f4122 MG |
5830 | schedule(); |
5831 | ||
5832 | return yielded; | |
5833 | } | |
5834 | EXPORT_SYMBOL_GPL(yield_to); | |
5835 | ||
10ab5643 TH |
5836 | int io_schedule_prepare(void) |
5837 | { | |
5838 | int old_iowait = current->in_iowait; | |
5839 | ||
5840 | current->in_iowait = 1; | |
5841 | blk_schedule_flush_plug(current); | |
5842 | ||
5843 | return old_iowait; | |
5844 | } | |
5845 | ||
5846 | void io_schedule_finish(int token) | |
5847 | { | |
5848 | current->in_iowait = token; | |
5849 | } | |
5850 | ||
1da177e4 | 5851 | /* |
41a2d6cf | 5852 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5853 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 5854 | */ |
1da177e4 LT |
5855 | long __sched io_schedule_timeout(long timeout) |
5856 | { | |
10ab5643 | 5857 | int token; |
1da177e4 LT |
5858 | long ret; |
5859 | ||
10ab5643 | 5860 | token = io_schedule_prepare(); |
1da177e4 | 5861 | ret = schedule_timeout(timeout); |
10ab5643 | 5862 | io_schedule_finish(token); |
9cff8ade | 5863 | |
1da177e4 LT |
5864 | return ret; |
5865 | } | |
9cff8ade | 5866 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 5867 | |
e3b929b0 | 5868 | void __sched io_schedule(void) |
10ab5643 TH |
5869 | { |
5870 | int token; | |
5871 | ||
5872 | token = io_schedule_prepare(); | |
5873 | schedule(); | |
5874 | io_schedule_finish(token); | |
5875 | } | |
5876 | EXPORT_SYMBOL(io_schedule); | |
5877 | ||
1da177e4 LT |
5878 | /** |
5879 | * sys_sched_get_priority_max - return maximum RT priority. | |
5880 | * @policy: scheduling class. | |
5881 | * | |
e69f6186 YB |
5882 | * Return: On success, this syscall returns the maximum |
5883 | * rt_priority that can be used by a given scheduling class. | |
5884 | * On failure, a negative error code is returned. | |
1da177e4 | 5885 | */ |
5add95d4 | 5886 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5887 | { |
5888 | int ret = -EINVAL; | |
5889 | ||
5890 | switch (policy) { | |
5891 | case SCHED_FIFO: | |
5892 | case SCHED_RR: | |
5893 | ret = MAX_USER_RT_PRIO-1; | |
5894 | break; | |
aab03e05 | 5895 | case SCHED_DEADLINE: |
1da177e4 | 5896 | case SCHED_NORMAL: |
b0a9499c | 5897 | case SCHED_BATCH: |
dd41f596 | 5898 | case SCHED_IDLE: |
1da177e4 LT |
5899 | ret = 0; |
5900 | break; | |
5901 | } | |
5902 | return ret; | |
5903 | } | |
5904 | ||
5905 | /** | |
5906 | * sys_sched_get_priority_min - return minimum RT priority. | |
5907 | * @policy: scheduling class. | |
5908 | * | |
e69f6186 YB |
5909 | * Return: On success, this syscall returns the minimum |
5910 | * rt_priority that can be used by a given scheduling class. | |
5911 | * On failure, a negative error code is returned. | |
1da177e4 | 5912 | */ |
5add95d4 | 5913 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5914 | { |
5915 | int ret = -EINVAL; | |
5916 | ||
5917 | switch (policy) { | |
5918 | case SCHED_FIFO: | |
5919 | case SCHED_RR: | |
5920 | ret = 1; | |
5921 | break; | |
aab03e05 | 5922 | case SCHED_DEADLINE: |
1da177e4 | 5923 | case SCHED_NORMAL: |
b0a9499c | 5924 | case SCHED_BATCH: |
dd41f596 | 5925 | case SCHED_IDLE: |
1da177e4 LT |
5926 | ret = 0; |
5927 | } | |
5928 | return ret; | |
5929 | } | |
5930 | ||
abca5fc5 | 5931 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 5932 | { |
36c8b586 | 5933 | struct task_struct *p; |
a4ec24b4 | 5934 | unsigned int time_slice; |
eb580751 | 5935 | struct rq_flags rf; |
dba091b9 | 5936 | struct rq *rq; |
3a5c359a | 5937 | int retval; |
1da177e4 LT |
5938 | |
5939 | if (pid < 0) | |
3a5c359a | 5940 | return -EINVAL; |
1da177e4 LT |
5941 | |
5942 | retval = -ESRCH; | |
1a551ae7 | 5943 | rcu_read_lock(); |
1da177e4 LT |
5944 | p = find_process_by_pid(pid); |
5945 | if (!p) | |
5946 | goto out_unlock; | |
5947 | ||
5948 | retval = security_task_getscheduler(p); | |
5949 | if (retval) | |
5950 | goto out_unlock; | |
5951 | ||
eb580751 | 5952 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
5953 | time_slice = 0; |
5954 | if (p->sched_class->get_rr_interval) | |
5955 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 5956 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 5957 | |
1a551ae7 | 5958 | rcu_read_unlock(); |
abca5fc5 AV |
5959 | jiffies_to_timespec64(time_slice, t); |
5960 | return 0; | |
3a5c359a | 5961 | |
1da177e4 | 5962 | out_unlock: |
1a551ae7 | 5963 | rcu_read_unlock(); |
1da177e4 LT |
5964 | return retval; |
5965 | } | |
5966 | ||
2064a5ab RD |
5967 | /** |
5968 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5969 | * @pid: pid of the process. | |
5970 | * @interval: userspace pointer to the timeslice value. | |
5971 | * | |
5972 | * this syscall writes the default timeslice value of a given process | |
5973 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5974 | * | |
5975 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
5976 | * an error code. | |
5977 | */ | |
abca5fc5 | 5978 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 5979 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
5980 | { |
5981 | struct timespec64 t; | |
5982 | int retval = sched_rr_get_interval(pid, &t); | |
5983 | ||
5984 | if (retval == 0) | |
5985 | retval = put_timespec64(&t, interval); | |
5986 | ||
5987 | return retval; | |
5988 | } | |
5989 | ||
474b9c77 | 5990 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
5991 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
5992 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
5993 | { |
5994 | struct timespec64 t; | |
5995 | int retval = sched_rr_get_interval(pid, &t); | |
5996 | ||
5997 | if (retval == 0) | |
9afc5eee | 5998 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
5999 | return retval; |
6000 | } | |
6001 | #endif | |
6002 | ||
82a1fcb9 | 6003 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6004 | { |
1da177e4 | 6005 | unsigned long free = 0; |
4e79752c | 6006 | int ppid; |
c930b2c0 | 6007 | |
38200502 TH |
6008 | if (!try_get_task_stack(p)) |
6009 | return; | |
20435d84 XX |
6010 | |
6011 | printk(KERN_INFO "%-15.15s %c", p->comm, task_state_to_char(p)); | |
6012 | ||
6013 | if (p->state == TASK_RUNNING) | |
3df0fc5b | 6014 | printk(KERN_CONT " running task "); |
1da177e4 | 6015 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 6016 | free = stack_not_used(p); |
1da177e4 | 6017 | #endif |
a90e984c | 6018 | ppid = 0; |
4e79752c | 6019 | rcu_read_lock(); |
a90e984c ON |
6020 | if (pid_alive(p)) |
6021 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 6022 | rcu_read_unlock(); |
3df0fc5b | 6023 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
4e79752c | 6024 | task_pid_nr(p), ppid, |
aa47b7e0 | 6025 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 6026 | |
3d1cb205 | 6027 | print_worker_info(KERN_INFO, p); |
9cb8f069 | 6028 | show_stack(p, NULL, KERN_INFO); |
38200502 | 6029 | put_task_stack(p); |
1da177e4 | 6030 | } |
0032f4e8 | 6031 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 6032 | |
5d68cc95 PZ |
6033 | static inline bool |
6034 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
6035 | { | |
6036 | /* no filter, everything matches */ | |
6037 | if (!state_filter) | |
6038 | return true; | |
6039 | ||
6040 | /* filter, but doesn't match */ | |
6041 | if (!(p->state & state_filter)) | |
6042 | return false; | |
6043 | ||
6044 | /* | |
6045 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
6046 | * TASK_KILLABLE). | |
6047 | */ | |
6048 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
6049 | return false; | |
6050 | ||
6051 | return true; | |
6052 | } | |
6053 | ||
6054 | ||
e59e2ae2 | 6055 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6056 | { |
36c8b586 | 6057 | struct task_struct *g, *p; |
1da177e4 | 6058 | |
4bd77321 | 6059 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
6060 | printk(KERN_INFO |
6061 | " task PC stack pid father\n"); | |
1da177e4 | 6062 | #else |
3df0fc5b PZ |
6063 | printk(KERN_INFO |
6064 | " task PC stack pid father\n"); | |
1da177e4 | 6065 | #endif |
510f5acc | 6066 | rcu_read_lock(); |
5d07f420 | 6067 | for_each_process_thread(g, p) { |
1da177e4 LT |
6068 | /* |
6069 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 6070 | * console might take a lot of time: |
57675cb9 AR |
6071 | * Also, reset softlockup watchdogs on all CPUs, because |
6072 | * another CPU might be blocked waiting for us to process | |
6073 | * an IPI. | |
1da177e4 LT |
6074 | */ |
6075 | touch_nmi_watchdog(); | |
57675cb9 | 6076 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 6077 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 6078 | sched_show_task(p); |
5d07f420 | 6079 | } |
1da177e4 | 6080 | |
dd41f596 | 6081 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
6082 | if (!state_filter) |
6083 | sysrq_sched_debug_show(); | |
dd41f596 | 6084 | #endif |
510f5acc | 6085 | rcu_read_unlock(); |
e59e2ae2 IM |
6086 | /* |
6087 | * Only show locks if all tasks are dumped: | |
6088 | */ | |
93335a21 | 6089 | if (!state_filter) |
e59e2ae2 | 6090 | debug_show_all_locks(); |
1da177e4 LT |
6091 | } |
6092 | ||
f340c0d1 IM |
6093 | /** |
6094 | * init_idle - set up an idle thread for a given CPU | |
6095 | * @idle: task in question | |
d1ccc66d | 6096 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
6097 | * |
6098 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6099 | * flag, to make booting more robust. | |
6100 | */ | |
0db0628d | 6101 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6102 | { |
70b97a7f | 6103 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6104 | unsigned long flags; |
6105 | ||
ff51ff84 PZ |
6106 | __sched_fork(0, idle); |
6107 | ||
25834c73 PZ |
6108 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
6109 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 6110 | |
06b83b5f | 6111 | idle->state = TASK_RUNNING; |
dd41f596 | 6112 | idle->se.exec_start = sched_clock(); |
c1de45ca | 6113 | idle->flags |= PF_IDLE; |
dd41f596 | 6114 | |
d08b9f0c | 6115 | scs_task_reset(idle); |
e1b77c92 MR |
6116 | kasan_unpoison_task_stack(idle); |
6117 | ||
de9b8f5d PZ |
6118 | #ifdef CONFIG_SMP |
6119 | /* | |
6120 | * Its possible that init_idle() gets called multiple times on a task, | |
6121 | * in that case do_set_cpus_allowed() will not do the right thing. | |
6122 | * | |
6123 | * And since this is boot we can forgo the serialization. | |
6124 | */ | |
6125 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
6126 | #endif | |
6506cf6c PZ |
6127 | /* |
6128 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 6129 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
6130 | * lockdep check in task_group() will fail. |
6131 | * | |
6132 | * Similar case to sched_fork(). / Alternatively we could | |
6133 | * use task_rq_lock() here and obtain the other rq->lock. | |
6134 | * | |
6135 | * Silence PROVE_RCU | |
6136 | */ | |
6137 | rcu_read_lock(); | |
dd41f596 | 6138 | __set_task_cpu(idle, cpu); |
6506cf6c | 6139 | rcu_read_unlock(); |
1da177e4 | 6140 | |
5311a98f EB |
6141 | rq->idle = idle; |
6142 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 6143 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 6144 | #ifdef CONFIG_SMP |
3ca7a440 | 6145 | idle->on_cpu = 1; |
4866cde0 | 6146 | #endif |
25834c73 PZ |
6147 | raw_spin_unlock(&rq->lock); |
6148 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
6149 | |
6150 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 6151 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 6152 | |
dd41f596 IM |
6153 | /* |
6154 | * The idle tasks have their own, simple scheduling class: | |
6155 | */ | |
6156 | idle->sched_class = &idle_sched_class; | |
868baf07 | 6157 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 6158 | vtime_init_idle(idle, cpu); |
de9b8f5d | 6159 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
6160 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
6161 | #endif | |
19978ca6 IM |
6162 | } |
6163 | ||
e1d4eeec NP |
6164 | #ifdef CONFIG_SMP |
6165 | ||
f82f8042 JL |
6166 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
6167 | const struct cpumask *trial) | |
6168 | { | |
06a76fe0 | 6169 | int ret = 1; |
f82f8042 | 6170 | |
bb2bc55a MG |
6171 | if (!cpumask_weight(cur)) |
6172 | return ret; | |
6173 | ||
06a76fe0 | 6174 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
6175 | |
6176 | return ret; | |
6177 | } | |
6178 | ||
7f51412a JL |
6179 | int task_can_attach(struct task_struct *p, |
6180 | const struct cpumask *cs_cpus_allowed) | |
6181 | { | |
6182 | int ret = 0; | |
6183 | ||
6184 | /* | |
6185 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 6186 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
6187 | * affinity and isolating such threads by their set of |
6188 | * allowed nodes is unnecessary. Thus, cpusets are not | |
6189 | * applicable for such threads. This prevents checking for | |
6190 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 6191 | * before cpus_mask may be changed. |
7f51412a JL |
6192 | */ |
6193 | if (p->flags & PF_NO_SETAFFINITY) { | |
6194 | ret = -EINVAL; | |
6195 | goto out; | |
6196 | } | |
6197 | ||
7f51412a | 6198 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
6199 | cs_cpus_allowed)) |
6200 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 6201 | |
7f51412a JL |
6202 | out: |
6203 | return ret; | |
6204 | } | |
6205 | ||
f2cb1360 | 6206 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 6207 | |
e6628d5b MG |
6208 | #ifdef CONFIG_NUMA_BALANCING |
6209 | /* Migrate current task p to target_cpu */ | |
6210 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
6211 | { | |
6212 | struct migration_arg arg = { p, target_cpu }; | |
6213 | int curr_cpu = task_cpu(p); | |
6214 | ||
6215 | if (curr_cpu == target_cpu) | |
6216 | return 0; | |
6217 | ||
3bd37062 | 6218 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
6219 | return -EINVAL; |
6220 | ||
6221 | /* TODO: This is not properly updating schedstats */ | |
6222 | ||
286549dc | 6223 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
6224 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
6225 | } | |
0ec8aa00 PZ |
6226 | |
6227 | /* | |
6228 | * Requeue a task on a given node and accurately track the number of NUMA | |
6229 | * tasks on the runqueues | |
6230 | */ | |
6231 | void sched_setnuma(struct task_struct *p, int nid) | |
6232 | { | |
da0c1e65 | 6233 | bool queued, running; |
eb580751 PZ |
6234 | struct rq_flags rf; |
6235 | struct rq *rq; | |
0ec8aa00 | 6236 | |
eb580751 | 6237 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 6238 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
6239 | running = task_current(rq, p); |
6240 | ||
da0c1e65 | 6241 | if (queued) |
1de64443 | 6242 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 6243 | if (running) |
f3cd1c4e | 6244 | put_prev_task(rq, p); |
0ec8aa00 PZ |
6245 | |
6246 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 6247 | |
da0c1e65 | 6248 | if (queued) |
7134b3e9 | 6249 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 6250 | if (running) |
03b7fad1 | 6251 | set_next_task(rq, p); |
eb580751 | 6252 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 6253 | } |
5cc389bc | 6254 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 6255 | |
1da177e4 | 6256 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 6257 | /* |
d1ccc66d | 6258 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 6259 | * offline. |
054b9108 | 6260 | */ |
48c5ccae | 6261 | void idle_task_exit(void) |
1da177e4 | 6262 | { |
48c5ccae | 6263 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6264 | |
48c5ccae | 6265 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 6266 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 6267 | |
a53efe5f | 6268 | if (mm != &init_mm) { |
252d2a41 | 6269 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
6270 | finish_arch_post_lock_switch(); |
6271 | } | |
bf2c59fc PZ |
6272 | |
6273 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
6274 | } |
6275 | ||
6276 | /* | |
5d180232 PZ |
6277 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
6278 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
6279 | * nr_active count is stable. We need to take the teardown thread which |
6280 | * is calling this into account, so we hand in adjust = 1 to the load | |
6281 | * calculation. | |
5d180232 PZ |
6282 | * |
6283 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 6284 | */ |
5d180232 | 6285 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 6286 | { |
d60585c5 | 6287 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
6288 | if (delta) |
6289 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
6290 | } |
6291 | ||
10e7071b | 6292 | static struct task_struct *__pick_migrate_task(struct rq *rq) |
3f1d2a31 | 6293 | { |
10e7071b PZ |
6294 | const struct sched_class *class; |
6295 | struct task_struct *next; | |
3f1d2a31 | 6296 | |
10e7071b | 6297 | for_each_class(class) { |
98c2f700 | 6298 | next = class->pick_next_task(rq); |
10e7071b | 6299 | if (next) { |
6e2df058 | 6300 | next->sched_class->put_prev_task(rq, next); |
10e7071b PZ |
6301 | return next; |
6302 | } | |
6303 | } | |
3f1d2a31 | 6304 | |
10e7071b PZ |
6305 | /* The idle class should always have a runnable task */ |
6306 | BUG(); | |
6307 | } | |
3f1d2a31 | 6308 | |
48f24c4d | 6309 | /* |
48c5ccae PZ |
6310 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6311 | * try_to_wake_up()->select_task_rq(). | |
6312 | * | |
6313 | * Called with rq->lock held even though we'er in stop_machine() and | |
6314 | * there's no concurrency possible, we hold the required locks anyway | |
6315 | * because of lock validation efforts. | |
1da177e4 | 6316 | */ |
8a8c69c3 | 6317 | static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf) |
1da177e4 | 6318 | { |
5e16bbc2 | 6319 | struct rq *rq = dead_rq; |
48c5ccae | 6320 | struct task_struct *next, *stop = rq->stop; |
8a8c69c3 | 6321 | struct rq_flags orf = *rf; |
48c5ccae | 6322 | int dest_cpu; |
1da177e4 LT |
6323 | |
6324 | /* | |
48c5ccae PZ |
6325 | * Fudge the rq selection such that the below task selection loop |
6326 | * doesn't get stuck on the currently eligible stop task. | |
6327 | * | |
6328 | * We're currently inside stop_machine() and the rq is either stuck | |
6329 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6330 | * either way we should never end up calling schedule() until we're | |
6331 | * done here. | |
1da177e4 | 6332 | */ |
48c5ccae | 6333 | rq->stop = NULL; |
48f24c4d | 6334 | |
77bd3970 FW |
6335 | /* |
6336 | * put_prev_task() and pick_next_task() sched | |
6337 | * class method both need to have an up-to-date | |
6338 | * value of rq->clock[_task] | |
6339 | */ | |
6340 | update_rq_clock(rq); | |
6341 | ||
5e16bbc2 | 6342 | for (;;) { |
48c5ccae PZ |
6343 | /* |
6344 | * There's this thread running, bail when that's the only | |
d1ccc66d | 6345 | * remaining thread: |
48c5ccae PZ |
6346 | */ |
6347 | if (rq->nr_running == 1) | |
dd41f596 | 6348 | break; |
48c5ccae | 6349 | |
10e7071b | 6350 | next = __pick_migrate_task(rq); |
e692ab53 | 6351 | |
5473e0cc | 6352 | /* |
3bd37062 | 6353 | * Rules for changing task_struct::cpus_mask are holding |
5473e0cc WL |
6354 | * both pi_lock and rq->lock, such that holding either |
6355 | * stabilizes the mask. | |
6356 | * | |
6357 | * Drop rq->lock is not quite as disastrous as it usually is | |
6358 | * because !cpu_active at this point, which means load-balance | |
6359 | * will not interfere. Also, stop-machine. | |
6360 | */ | |
8a8c69c3 | 6361 | rq_unlock(rq, rf); |
5473e0cc | 6362 | raw_spin_lock(&next->pi_lock); |
8a8c69c3 | 6363 | rq_relock(rq, rf); |
5473e0cc WL |
6364 | |
6365 | /* | |
6366 | * Since we're inside stop-machine, _nothing_ should have | |
6367 | * changed the task, WARN if weird stuff happened, because in | |
6368 | * that case the above rq->lock drop is a fail too. | |
6369 | */ | |
6370 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
6371 | raw_spin_unlock(&next->pi_lock); | |
6372 | continue; | |
6373 | } | |
6374 | ||
48c5ccae | 6375 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 6376 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
8a8c69c3 | 6377 | rq = __migrate_task(rq, rf, next, dest_cpu); |
5e16bbc2 | 6378 | if (rq != dead_rq) { |
8a8c69c3 | 6379 | rq_unlock(rq, rf); |
5e16bbc2 | 6380 | rq = dead_rq; |
8a8c69c3 PZ |
6381 | *rf = orf; |
6382 | rq_relock(rq, rf); | |
5e16bbc2 | 6383 | } |
5473e0cc | 6384 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 6385 | } |
dce48a84 | 6386 | |
48c5ccae | 6387 | rq->stop = stop; |
dce48a84 | 6388 | } |
1da177e4 LT |
6389 | #endif /* CONFIG_HOTPLUG_CPU */ |
6390 | ||
f2cb1360 | 6391 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
6392 | { |
6393 | if (!rq->online) { | |
6394 | const struct sched_class *class; | |
6395 | ||
c6c4927b | 6396 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6397 | rq->online = 1; |
6398 | ||
6399 | for_each_class(class) { | |
6400 | if (class->rq_online) | |
6401 | class->rq_online(rq); | |
6402 | } | |
6403 | } | |
6404 | } | |
6405 | ||
f2cb1360 | 6406 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
6407 | { |
6408 | if (rq->online) { | |
6409 | const struct sched_class *class; | |
6410 | ||
6411 | for_each_class(class) { | |
6412 | if (class->rq_offline) | |
6413 | class->rq_offline(rq); | |
6414 | } | |
6415 | ||
c6c4927b | 6416 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6417 | rq->online = 0; |
6418 | } | |
6419 | } | |
6420 | ||
d1ccc66d IM |
6421 | /* |
6422 | * used to mark begin/end of suspend/resume: | |
6423 | */ | |
6424 | static int num_cpus_frozen; | |
d35be8ba | 6425 | |
1da177e4 | 6426 | /* |
3a101d05 TH |
6427 | * Update cpusets according to cpu_active mask. If cpusets are |
6428 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6429 | * around partition_sched_domains(). | |
d35be8ba SB |
6430 | * |
6431 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6432 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6433 | */ |
40190a78 | 6434 | static void cpuset_cpu_active(void) |
e761b772 | 6435 | { |
40190a78 | 6436 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
6437 | /* |
6438 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6439 | * resume sequence. As long as this is not the last online | |
6440 | * operation in the resume sequence, just build a single sched | |
6441 | * domain, ignoring cpusets. | |
6442 | */ | |
50e76632 PZ |
6443 | partition_sched_domains(1, NULL, NULL); |
6444 | if (--num_cpus_frozen) | |
135fb3e1 | 6445 | return; |
d35be8ba SB |
6446 | /* |
6447 | * This is the last CPU online operation. So fall through and | |
6448 | * restore the original sched domains by considering the | |
6449 | * cpuset configurations. | |
6450 | */ | |
50e76632 | 6451 | cpuset_force_rebuild(); |
3a101d05 | 6452 | } |
30e03acd | 6453 | cpuset_update_active_cpus(); |
3a101d05 | 6454 | } |
e761b772 | 6455 | |
40190a78 | 6456 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 6457 | { |
40190a78 | 6458 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 6459 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 6460 | return -EBUSY; |
30e03acd | 6461 | cpuset_update_active_cpus(); |
135fb3e1 | 6462 | } else { |
d35be8ba SB |
6463 | num_cpus_frozen++; |
6464 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 6465 | } |
135fb3e1 | 6466 | return 0; |
e761b772 | 6467 | } |
e761b772 | 6468 | |
40190a78 | 6469 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 6470 | { |
7d976699 | 6471 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 6472 | struct rq_flags rf; |
7d976699 | 6473 | |
ba2591a5 PZ |
6474 | #ifdef CONFIG_SCHED_SMT |
6475 | /* | |
c5511d03 | 6476 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 6477 | */ |
c5511d03 PZI |
6478 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
6479 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 6480 | #endif |
40190a78 | 6481 | set_cpu_active(cpu, true); |
135fb3e1 | 6482 | |
40190a78 | 6483 | if (sched_smp_initialized) { |
135fb3e1 | 6484 | sched_domains_numa_masks_set(cpu); |
40190a78 | 6485 | cpuset_cpu_active(); |
e761b772 | 6486 | } |
7d976699 TG |
6487 | |
6488 | /* | |
6489 | * Put the rq online, if not already. This happens: | |
6490 | * | |
6491 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 6492 | * after all CPUs have been brought up. |
7d976699 TG |
6493 | * |
6494 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
6495 | * domains. | |
6496 | */ | |
8a8c69c3 | 6497 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
6498 | if (rq->rd) { |
6499 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
6500 | set_rq_online(rq); | |
6501 | } | |
8a8c69c3 | 6502 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 6503 | |
40190a78 | 6504 | return 0; |
135fb3e1 TG |
6505 | } |
6506 | ||
40190a78 | 6507 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 6508 | { |
135fb3e1 TG |
6509 | int ret; |
6510 | ||
40190a78 | 6511 | set_cpu_active(cpu, false); |
b2454caa PZ |
6512 | /* |
6513 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
6514 | * users of this state to go away such that all new such users will | |
6515 | * observe it. | |
6516 | * | |
b2454caa PZ |
6517 | * Do sync before park smpboot threads to take care the rcu boost case. |
6518 | */ | |
309ba859 | 6519 | synchronize_rcu(); |
40190a78 | 6520 | |
c5511d03 PZI |
6521 | #ifdef CONFIG_SCHED_SMT |
6522 | /* | |
6523 | * When going down, decrement the number of cores with SMT present. | |
6524 | */ | |
6525 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
6526 | static_branch_dec_cpuslocked(&sched_smt_present); | |
6527 | #endif | |
6528 | ||
40190a78 TG |
6529 | if (!sched_smp_initialized) |
6530 | return 0; | |
6531 | ||
6532 | ret = cpuset_cpu_inactive(cpu); | |
6533 | if (ret) { | |
6534 | set_cpu_active(cpu, true); | |
6535 | return ret; | |
135fb3e1 | 6536 | } |
40190a78 TG |
6537 | sched_domains_numa_masks_clear(cpu); |
6538 | return 0; | |
135fb3e1 TG |
6539 | } |
6540 | ||
94baf7a5 TG |
6541 | static void sched_rq_cpu_starting(unsigned int cpu) |
6542 | { | |
6543 | struct rq *rq = cpu_rq(cpu); | |
6544 | ||
6545 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
6546 | update_max_interval(); |
6547 | } | |
6548 | ||
135fb3e1 TG |
6549 | int sched_cpu_starting(unsigned int cpu) |
6550 | { | |
94baf7a5 | 6551 | sched_rq_cpu_starting(cpu); |
d84b3131 | 6552 | sched_tick_start(cpu); |
135fb3e1 | 6553 | return 0; |
e761b772 | 6554 | } |
e761b772 | 6555 | |
f2785ddb TG |
6556 | #ifdef CONFIG_HOTPLUG_CPU |
6557 | int sched_cpu_dying(unsigned int cpu) | |
6558 | { | |
6559 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 6560 | struct rq_flags rf; |
f2785ddb TG |
6561 | |
6562 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 6563 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
6564 | |
6565 | rq_lock_irqsave(rq, &rf); | |
f2785ddb TG |
6566 | if (rq->rd) { |
6567 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
6568 | set_rq_offline(rq); | |
6569 | } | |
8a8c69c3 | 6570 | migrate_tasks(rq, &rf); |
f2785ddb | 6571 | BUG_ON(rq->nr_running != 1); |
8a8c69c3 PZ |
6572 | rq_unlock_irqrestore(rq, &rf); |
6573 | ||
f2785ddb TG |
6574 | calc_load_migrate(rq); |
6575 | update_max_interval(); | |
00357f5e | 6576 | nohz_balance_exit_idle(rq); |
e5ef27d0 | 6577 | hrtick_clear(rq); |
f2785ddb TG |
6578 | return 0; |
6579 | } | |
6580 | #endif | |
6581 | ||
1da177e4 LT |
6582 | void __init sched_init_smp(void) |
6583 | { | |
cb83b629 PZ |
6584 | sched_init_numa(); |
6585 | ||
6acce3ef PZ |
6586 | /* |
6587 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 6588 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 6589 | * happen. |
6acce3ef | 6590 | */ |
712555ee | 6591 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 6592 | sched_init_domains(cpu_active_mask); |
712555ee | 6593 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 6594 | |
5c1e1767 | 6595 | /* Move init over to a non-isolated CPU */ |
edb93821 | 6596 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 6597 | BUG(); |
19978ca6 | 6598 | sched_init_granularity(); |
4212823f | 6599 | |
0e3900e6 | 6600 | init_sched_rt_class(); |
1baca4ce | 6601 | init_sched_dl_class(); |
1b568f0a | 6602 | |
e26fbffd | 6603 | sched_smp_initialized = true; |
1da177e4 | 6604 | } |
e26fbffd TG |
6605 | |
6606 | static int __init migration_init(void) | |
6607 | { | |
77a5352b | 6608 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 6609 | return 0; |
1da177e4 | 6610 | } |
e26fbffd TG |
6611 | early_initcall(migration_init); |
6612 | ||
1da177e4 LT |
6613 | #else |
6614 | void __init sched_init_smp(void) | |
6615 | { | |
19978ca6 | 6616 | sched_init_granularity(); |
1da177e4 LT |
6617 | } |
6618 | #endif /* CONFIG_SMP */ | |
6619 | ||
6620 | int in_sched_functions(unsigned long addr) | |
6621 | { | |
1da177e4 LT |
6622 | return in_lock_functions(addr) || |
6623 | (addr >= (unsigned long)__sched_text_start | |
6624 | && addr < (unsigned long)__sched_text_end); | |
6625 | } | |
6626 | ||
029632fb | 6627 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
6628 | /* |
6629 | * Default task group. | |
6630 | * Every task in system belongs to this group at bootup. | |
6631 | */ | |
029632fb | 6632 | struct task_group root_task_group; |
35cf4e50 | 6633 | LIST_HEAD(task_groups); |
b0367629 WL |
6634 | |
6635 | /* Cacheline aligned slab cache for task_group */ | |
6636 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 6637 | #endif |
6f505b16 | 6638 | |
e6252c3e | 6639 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 6640 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 6641 | |
1da177e4 LT |
6642 | void __init sched_init(void) |
6643 | { | |
a1dc0446 | 6644 | unsigned long ptr = 0; |
55627e3c | 6645 | int i; |
434d53b0 | 6646 | |
5822a454 | 6647 | wait_bit_init(); |
9dcb8b68 | 6648 | |
434d53b0 | 6649 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 6650 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
6651 | #endif |
6652 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 6653 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 6654 | #endif |
a1dc0446 QC |
6655 | if (ptr) { |
6656 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
6657 | |
6658 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 6659 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
6660 | ptr += nr_cpu_ids * sizeof(void **); |
6661 | ||
07e06b01 | 6662 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 6663 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 6664 | |
b1d1779e WY |
6665 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
6666 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 6667 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 6668 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6669 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
6670 | ptr += nr_cpu_ids * sizeof(void **); |
6671 | ||
07e06b01 | 6672 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
6673 | ptr += nr_cpu_ids * sizeof(void **); |
6674 | ||
6d6bc0ad | 6675 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 6676 | } |
df7c8e84 | 6677 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
6678 | for_each_possible_cpu(i) { |
6679 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
6680 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
6681 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
6682 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 6683 | } |
b74e6278 | 6684 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 6685 | |
d1ccc66d IM |
6686 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
6687 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 6688 | |
57d885fe GH |
6689 | #ifdef CONFIG_SMP |
6690 | init_defrootdomain(); | |
6691 | #endif | |
6692 | ||
d0b27fa7 | 6693 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6694 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 6695 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 6696 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 6697 | |
7c941438 | 6698 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
6699 | task_group_cache = KMEM_CACHE(task_group, 0); |
6700 | ||
07e06b01 YZ |
6701 | list_add(&root_task_group.list, &task_groups); |
6702 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 6703 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 6704 | autogroup_init(&init_task); |
7c941438 | 6705 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 6706 | |
0a945022 | 6707 | for_each_possible_cpu(i) { |
70b97a7f | 6708 | struct rq *rq; |
1da177e4 LT |
6709 | |
6710 | rq = cpu_rq(i); | |
05fa785c | 6711 | raw_spin_lock_init(&rq->lock); |
7897986b | 6712 | rq->nr_running = 0; |
dce48a84 TG |
6713 | rq->calc_load_active = 0; |
6714 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 6715 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
6716 | init_rt_rq(&rq->rt); |
6717 | init_dl_rq(&rq->dl); | |
dd41f596 | 6718 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 6719 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 6720 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 6721 | /* |
d1ccc66d | 6722 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
6723 | * |
6724 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
6725 | * gets 100% of the CPU resources in the system. This overall |
6726 | * system CPU resource is divided among the tasks of | |
07e06b01 | 6727 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
6728 | * based on each entity's (task or task-group's) weight |
6729 | * (se->load.weight). | |
6730 | * | |
07e06b01 | 6731 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 6732 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 6733 | * then A0's share of the CPU resource is: |
354d60c2 | 6734 | * |
0d905bca | 6735 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 6736 | * |
07e06b01 YZ |
6737 | * We achieve this by letting root_task_group's tasks sit |
6738 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 6739 | */ |
07e06b01 | 6740 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
6741 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6742 | ||
6743 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 6744 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 6745 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 6746 | #endif |
1da177e4 | 6747 | #ifdef CONFIG_SMP |
41c7ce9a | 6748 | rq->sd = NULL; |
57d885fe | 6749 | rq->rd = NULL; |
ca6d75e6 | 6750 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 6751 | rq->balance_callback = NULL; |
1da177e4 | 6752 | rq->active_balance = 0; |
dd41f596 | 6753 | rq->next_balance = jiffies; |
1da177e4 | 6754 | rq->push_cpu = 0; |
0a2966b4 | 6755 | rq->cpu = i; |
1f11eb6a | 6756 | rq->online = 0; |
eae0c9df MG |
6757 | rq->idle_stamp = 0; |
6758 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 6759 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
6760 | |
6761 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
6762 | ||
dc938520 | 6763 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 6764 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 6765 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 6766 | atomic_set(&rq->nohz_flags, 0); |
90b5363a PZI |
6767 | |
6768 | rq_csd_init(rq, &rq->nohz_csd, nohz_csd_func); | |
83cd4fe2 | 6769 | #endif |
9fd81dd5 | 6770 | #endif /* CONFIG_SMP */ |
77a021be | 6771 | hrtick_rq_init(rq); |
1da177e4 | 6772 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
6773 | } |
6774 | ||
9059393e | 6775 | set_load_weight(&init_task, false); |
b50f60ce | 6776 | |
1da177e4 LT |
6777 | /* |
6778 | * The boot idle thread does lazy MMU switching as well: | |
6779 | */ | |
f1f10076 | 6780 | mmgrab(&init_mm); |
1da177e4 LT |
6781 | enter_lazy_tlb(&init_mm, current); |
6782 | ||
6783 | /* | |
6784 | * Make us the idle thread. Technically, schedule() should not be | |
6785 | * called from this thread, however somewhere below it might be, | |
6786 | * but because we are the idle thread, we just pick up running again | |
6787 | * when this runqueue becomes "idle". | |
6788 | */ | |
6789 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
6790 | |
6791 | calc_load_update = jiffies + LOAD_FREQ; | |
6792 | ||
bf4d83f6 | 6793 | #ifdef CONFIG_SMP |
29d5e047 | 6794 | idle_thread_set_boot_cpu(); |
029632fb PZ |
6795 | #endif |
6796 | init_sched_fair_class(); | |
6a7b3dc3 | 6797 | |
4698f88c JP |
6798 | init_schedstats(); |
6799 | ||
eb414681 JW |
6800 | psi_init(); |
6801 | ||
69842cba PB |
6802 | init_uclamp(); |
6803 | ||
6892b75e | 6804 | scheduler_running = 1; |
1da177e4 LT |
6805 | } |
6806 | ||
d902db1e | 6807 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
6808 | static inline int preempt_count_equals(int preempt_offset) |
6809 | { | |
da7142e2 | 6810 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 6811 | |
4ba8216c | 6812 | return (nested == preempt_offset); |
e4aafea2 FW |
6813 | } |
6814 | ||
d894837f | 6815 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 6816 | { |
8eb23b9f PZ |
6817 | /* |
6818 | * Blocking primitives will set (and therefore destroy) current->state, | |
6819 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
6820 | * otherwise we will destroy state. | |
6821 | */ | |
00845eb9 | 6822 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
6823 | "do not call blocking ops when !TASK_RUNNING; " |
6824 | "state=%lx set at [<%p>] %pS\n", | |
6825 | current->state, | |
6826 | (void *)current->task_state_change, | |
00845eb9 | 6827 | (void *)current->task_state_change); |
8eb23b9f | 6828 | |
3427445a PZ |
6829 | ___might_sleep(file, line, preempt_offset); |
6830 | } | |
6831 | EXPORT_SYMBOL(__might_sleep); | |
6832 | ||
6833 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 6834 | { |
d1ccc66d IM |
6835 | /* Ratelimiting timestamp: */ |
6836 | static unsigned long prev_jiffy; | |
6837 | ||
d1c6d149 | 6838 | unsigned long preempt_disable_ip; |
1da177e4 | 6839 | |
d1ccc66d IM |
6840 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
6841 | rcu_sleep_check(); | |
6842 | ||
db273be2 | 6843 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 6844 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
6845 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
6846 | oops_in_progress) | |
aef745fc | 6847 | return; |
1c3c5eab | 6848 | |
aef745fc IM |
6849 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
6850 | return; | |
6851 | prev_jiffy = jiffies; | |
6852 | ||
d1ccc66d | 6853 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
6854 | preempt_disable_ip = get_preempt_disable_ip(current); |
6855 | ||
3df0fc5b PZ |
6856 | printk(KERN_ERR |
6857 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
6858 | file, line); | |
6859 | printk(KERN_ERR | |
312364f3 DV |
6860 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
6861 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 6862 | current->pid, current->comm); |
aef745fc | 6863 | |
a8b686b3 ES |
6864 | if (task_stack_end_corrupted(current)) |
6865 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
6866 | ||
aef745fc IM |
6867 | debug_show_held_locks(current); |
6868 | if (irqs_disabled()) | |
6869 | print_irqtrace_events(current); | |
d1c6d149 VN |
6870 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
6871 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 6872 | pr_err("Preemption disabled at:"); |
2062a4e8 | 6873 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 6874 | } |
aef745fc | 6875 | dump_stack(); |
f0b22e39 | 6876 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 6877 | } |
3427445a | 6878 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
6879 | |
6880 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
6881 | { | |
6882 | static unsigned long prev_jiffy; | |
6883 | ||
6884 | if (irqs_disabled()) | |
6885 | return; | |
6886 | ||
6887 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
6888 | return; | |
6889 | ||
6890 | if (preempt_count() > preempt_offset) | |
6891 | return; | |
6892 | ||
6893 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6894 | return; | |
6895 | prev_jiffy = jiffies; | |
6896 | ||
6897 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
6898 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
6899 | in_atomic(), irqs_disabled(), | |
6900 | current->pid, current->comm); | |
6901 | ||
6902 | debug_show_held_locks(current); | |
6903 | dump_stack(); | |
6904 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
6905 | } | |
6906 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
1da177e4 LT |
6907 | #endif |
6908 | ||
6909 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 6910 | void normalize_rt_tasks(void) |
3a5e4dc1 | 6911 | { |
dbc7f069 | 6912 | struct task_struct *g, *p; |
d50dde5a DF |
6913 | struct sched_attr attr = { |
6914 | .sched_policy = SCHED_NORMAL, | |
6915 | }; | |
1da177e4 | 6916 | |
3472eaa1 | 6917 | read_lock(&tasklist_lock); |
5d07f420 | 6918 | for_each_process_thread(g, p) { |
178be793 IM |
6919 | /* |
6920 | * Only normalize user tasks: | |
6921 | */ | |
3472eaa1 | 6922 | if (p->flags & PF_KTHREAD) |
178be793 IM |
6923 | continue; |
6924 | ||
4fa8d299 JP |
6925 | p->se.exec_start = 0; |
6926 | schedstat_set(p->se.statistics.wait_start, 0); | |
6927 | schedstat_set(p->se.statistics.sleep_start, 0); | |
6928 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 6929 | |
aab03e05 | 6930 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
6931 | /* |
6932 | * Renice negative nice level userspace | |
6933 | * tasks back to 0: | |
6934 | */ | |
3472eaa1 | 6935 | if (task_nice(p) < 0) |
dd41f596 | 6936 | set_user_nice(p, 0); |
1da177e4 | 6937 | continue; |
dd41f596 | 6938 | } |
1da177e4 | 6939 | |
dbc7f069 | 6940 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 6941 | } |
3472eaa1 | 6942 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6943 | } |
6944 | ||
6945 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 6946 | |
67fc4e0c | 6947 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 6948 | /* |
67fc4e0c | 6949 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
6950 | * |
6951 | * They can only be called when the whole system has been | |
6952 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6953 | * activity can take place. Using them for anything else would | |
6954 | * be a serious bug, and as a result, they aren't even visible | |
6955 | * under any other configuration. | |
6956 | */ | |
6957 | ||
6958 | /** | |
d1ccc66d | 6959 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
6960 | * @cpu: the processor in question. |
6961 | * | |
6962 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
6963 | * |
6964 | * Return: The current task for @cpu. | |
1df5c10a | 6965 | */ |
36c8b586 | 6966 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6967 | { |
6968 | return cpu_curr(cpu); | |
6969 | } | |
6970 | ||
67fc4e0c JW |
6971 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
6972 | ||
6973 | #ifdef CONFIG_IA64 | |
1df5c10a | 6974 | /** |
5feeb783 | 6975 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
6976 | * @cpu: the processor in question. |
6977 | * @p: the task pointer to set. | |
6978 | * | |
6979 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 6980 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 6981 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
6982 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6983 | * and caller must save the original value of the current task (see | |
6984 | * curr_task() above) and restore that value before reenabling interrupts and | |
6985 | * re-starting the system. | |
6986 | * | |
6987 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6988 | */ | |
a458ae2e | 6989 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6990 | { |
6991 | cpu_curr(cpu) = p; | |
6992 | } | |
6993 | ||
6994 | #endif | |
29f59db3 | 6995 | |
7c941438 | 6996 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
6997 | /* task_group_lock serializes the addition/removal of task groups */ |
6998 | static DEFINE_SPINLOCK(task_group_lock); | |
6999 | ||
2480c093 PB |
7000 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
7001 | struct task_group *parent) | |
7002 | { | |
7003 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 7004 | enum uclamp_id clamp_id; |
2480c093 PB |
7005 | |
7006 | for_each_clamp_id(clamp_id) { | |
7007 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
7008 | uclamp_none(clamp_id), false); | |
0b60ba2d | 7009 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
7010 | } |
7011 | #endif | |
7012 | } | |
7013 | ||
2f5177f0 | 7014 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7015 | { |
7016 | free_fair_sched_group(tg); | |
7017 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7018 | autogroup_free(tg); |
b0367629 | 7019 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7020 | } |
7021 | ||
7022 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7023 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7024 | { |
7025 | struct task_group *tg; | |
bccbe08a | 7026 | |
b0367629 | 7027 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7028 | if (!tg) |
7029 | return ERR_PTR(-ENOMEM); | |
7030 | ||
ec7dc8ac | 7031 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7032 | goto err; |
7033 | ||
ec7dc8ac | 7034 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7035 | goto err; |
7036 | ||
2480c093 PB |
7037 | alloc_uclamp_sched_group(tg, parent); |
7038 | ||
ace783b9 LZ |
7039 | return tg; |
7040 | ||
7041 | err: | |
2f5177f0 | 7042 | sched_free_group(tg); |
ace783b9 LZ |
7043 | return ERR_PTR(-ENOMEM); |
7044 | } | |
7045 | ||
7046 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7047 | { | |
7048 | unsigned long flags; | |
7049 | ||
8ed36996 | 7050 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7051 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 7052 | |
d1ccc66d IM |
7053 | /* Root should already exist: */ |
7054 | WARN_ON(!parent); | |
f473aa5e PZ |
7055 | |
7056 | tg->parent = parent; | |
f473aa5e | 7057 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7058 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7059 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
7060 | |
7061 | online_fair_sched_group(tg); | |
29f59db3 SV |
7062 | } |
7063 | ||
9b5b7751 | 7064 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7065 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7066 | { |
d1ccc66d | 7067 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 7068 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7069 | } |
7070 | ||
4cf86d77 | 7071 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 7072 | { |
d1ccc66d | 7073 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 7074 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7075 | } |
7076 | ||
7077 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7078 | { |
8ed36996 | 7079 | unsigned long flags; |
29f59db3 | 7080 | |
d1ccc66d | 7081 | /* End participation in shares distribution: */ |
6fe1f348 | 7082 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7083 | |
7084 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7085 | list_del_rcu(&tg->list); |
f473aa5e | 7086 | list_del_rcu(&tg->siblings); |
8ed36996 | 7087 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7088 | } |
7089 | ||
ea86cb4b | 7090 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 7091 | { |
8323f26c | 7092 | struct task_group *tg; |
29f59db3 | 7093 | |
f7b8a47d KT |
7094 | /* |
7095 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7096 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7097 | * to prevent lockdep warnings. | |
7098 | */ | |
7099 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7100 | struct task_group, css); |
7101 | tg = autogroup_task_group(tsk, tg); | |
7102 | tsk->sched_task_group = tg; | |
7103 | ||
810b3817 | 7104 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
7105 | if (tsk->sched_class->task_change_group) |
7106 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 7107 | else |
810b3817 | 7108 | #endif |
b2b5ce02 | 7109 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
7110 | } |
7111 | ||
7112 | /* | |
7113 | * Change task's runqueue when it moves between groups. | |
7114 | * | |
7115 | * The caller of this function should have put the task in its new group by | |
7116 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
7117 | * its new group. | |
7118 | */ | |
7119 | void sched_move_task(struct task_struct *tsk) | |
7120 | { | |
7a57f32a PZ |
7121 | int queued, running, queue_flags = |
7122 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
7123 | struct rq_flags rf; |
7124 | struct rq *rq; | |
7125 | ||
7126 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 7127 | update_rq_clock(rq); |
ea86cb4b VG |
7128 | |
7129 | running = task_current(rq, tsk); | |
7130 | queued = task_on_rq_queued(tsk); | |
7131 | ||
7132 | if (queued) | |
7a57f32a | 7133 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 7134 | if (running) |
ea86cb4b VG |
7135 | put_prev_task(rq, tsk); |
7136 | ||
7137 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 7138 | |
da0c1e65 | 7139 | if (queued) |
7a57f32a | 7140 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 7141 | if (running) { |
03b7fad1 | 7142 | set_next_task(rq, tsk); |
2a4b03ff VG |
7143 | /* |
7144 | * After changing group, the running task may have joined a | |
7145 | * throttled one but it's still the running task. Trigger a | |
7146 | * resched to make sure that task can still run. | |
7147 | */ | |
7148 | resched_curr(rq); | |
7149 | } | |
29f59db3 | 7150 | |
eb580751 | 7151 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 7152 | } |
68318b8e | 7153 | |
a7c6d554 | 7154 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7155 | { |
a7c6d554 | 7156 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7157 | } |
7158 | ||
eb95419b TH |
7159 | static struct cgroup_subsys_state * |
7160 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7161 | { |
eb95419b TH |
7162 | struct task_group *parent = css_tg(parent_css); |
7163 | struct task_group *tg; | |
68318b8e | 7164 | |
eb95419b | 7165 | if (!parent) { |
68318b8e | 7166 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7167 | return &root_task_group.css; |
68318b8e SV |
7168 | } |
7169 | ||
ec7dc8ac | 7170 | tg = sched_create_group(parent); |
68318b8e SV |
7171 | if (IS_ERR(tg)) |
7172 | return ERR_PTR(-ENOMEM); | |
7173 | ||
68318b8e SV |
7174 | return &tg->css; |
7175 | } | |
7176 | ||
96b77745 KK |
7177 | /* Expose task group only after completing cgroup initialization */ |
7178 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
7179 | { | |
7180 | struct task_group *tg = css_tg(css); | |
7181 | struct task_group *parent = css_tg(css->parent); | |
7182 | ||
7183 | if (parent) | |
7184 | sched_online_group(tg, parent); | |
7226017a QY |
7185 | |
7186 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
7187 | /* Propagate the effective uclamp value for the new group */ | |
7188 | cpu_util_update_eff(css); | |
7189 | #endif | |
7190 | ||
96b77745 KK |
7191 | return 0; |
7192 | } | |
7193 | ||
2f5177f0 | 7194 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 7195 | { |
eb95419b | 7196 | struct task_group *tg = css_tg(css); |
ace783b9 | 7197 | |
2f5177f0 | 7198 | sched_offline_group(tg); |
ace783b9 LZ |
7199 | } |
7200 | ||
eb95419b | 7201 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7202 | { |
eb95419b | 7203 | struct task_group *tg = css_tg(css); |
68318b8e | 7204 | |
2f5177f0 PZ |
7205 | /* |
7206 | * Relies on the RCU grace period between css_released() and this. | |
7207 | */ | |
7208 | sched_free_group(tg); | |
ace783b9 LZ |
7209 | } |
7210 | ||
ea86cb4b VG |
7211 | /* |
7212 | * This is called before wake_up_new_task(), therefore we really only | |
7213 | * have to set its group bits, all the other stuff does not apply. | |
7214 | */ | |
b53202e6 | 7215 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 7216 | { |
ea86cb4b VG |
7217 | struct rq_flags rf; |
7218 | struct rq *rq; | |
7219 | ||
7220 | rq = task_rq_lock(task, &rf); | |
7221 | ||
80f5c1b8 | 7222 | update_rq_clock(rq); |
ea86cb4b VG |
7223 | sched_change_group(task, TASK_SET_GROUP); |
7224 | ||
7225 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
7226 | } |
7227 | ||
1f7dd3e5 | 7228 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 7229 | { |
bb9d97b6 | 7230 | struct task_struct *task; |
1f7dd3e5 | 7231 | struct cgroup_subsys_state *css; |
7dc603c9 | 7232 | int ret = 0; |
bb9d97b6 | 7233 | |
1f7dd3e5 | 7234 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 7235 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7236 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7237 | return -EINVAL; |
b68aa230 | 7238 | #endif |
7dc603c9 PZ |
7239 | /* |
7240 | * Serialize against wake_up_new_task() such that if its | |
7241 | * running, we're sure to observe its full state. | |
7242 | */ | |
7243 | raw_spin_lock_irq(&task->pi_lock); | |
7244 | /* | |
7245 | * Avoid calling sched_move_task() before wake_up_new_task() | |
7246 | * has happened. This would lead to problems with PELT, due to | |
7247 | * move wanting to detach+attach while we're not attached yet. | |
7248 | */ | |
7249 | if (task->state == TASK_NEW) | |
7250 | ret = -EINVAL; | |
7251 | raw_spin_unlock_irq(&task->pi_lock); | |
7252 | ||
7253 | if (ret) | |
7254 | break; | |
bb9d97b6 | 7255 | } |
7dc603c9 | 7256 | return ret; |
be367d09 | 7257 | } |
68318b8e | 7258 | |
1f7dd3e5 | 7259 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 7260 | { |
bb9d97b6 | 7261 | struct task_struct *task; |
1f7dd3e5 | 7262 | struct cgroup_subsys_state *css; |
bb9d97b6 | 7263 | |
1f7dd3e5 | 7264 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 7265 | sched_move_task(task); |
68318b8e SV |
7266 | } |
7267 | ||
2480c093 | 7268 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
7269 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
7270 | { | |
7271 | struct cgroup_subsys_state *top_css = css; | |
7272 | struct uclamp_se *uc_parent = NULL; | |
7273 | struct uclamp_se *uc_se = NULL; | |
7274 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 7275 | enum uclamp_id clamp_id; |
0b60ba2d PB |
7276 | unsigned int clamps; |
7277 | ||
7278 | css_for_each_descendant_pre(css, top_css) { | |
7279 | uc_parent = css_tg(css)->parent | |
7280 | ? css_tg(css)->parent->uclamp : NULL; | |
7281 | ||
7282 | for_each_clamp_id(clamp_id) { | |
7283 | /* Assume effective clamps matches requested clamps */ | |
7284 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
7285 | /* Cap effective clamps with parent's effective clamps */ | |
7286 | if (uc_parent && | |
7287 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
7288 | eff[clamp_id] = uc_parent[clamp_id].value; | |
7289 | } | |
7290 | } | |
7291 | /* Ensure protection is always capped by limit */ | |
7292 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
7293 | ||
7294 | /* Propagate most restrictive effective clamps */ | |
7295 | clamps = 0x0; | |
7296 | uc_se = css_tg(css)->uclamp; | |
7297 | for_each_clamp_id(clamp_id) { | |
7298 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
7299 | continue; | |
7300 | uc_se[clamp_id].value = eff[clamp_id]; | |
7301 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
7302 | clamps |= (0x1 << clamp_id); | |
7303 | } | |
babbe170 | 7304 | if (!clamps) { |
0b60ba2d | 7305 | css = css_rightmost_descendant(css); |
babbe170 PB |
7306 | continue; |
7307 | } | |
7308 | ||
7309 | /* Immediately update descendants RUNNABLE tasks */ | |
7310 | uclamp_update_active_tasks(css, clamps); | |
0b60ba2d PB |
7311 | } |
7312 | } | |
2480c093 PB |
7313 | |
7314 | /* | |
7315 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
7316 | * C expression. Since there is no way to convert a macro argument (N) into a | |
7317 | * character constant, use two levels of macros. | |
7318 | */ | |
7319 | #define _POW10(exp) ((unsigned int)1e##exp) | |
7320 | #define POW10(exp) _POW10(exp) | |
7321 | ||
7322 | struct uclamp_request { | |
7323 | #define UCLAMP_PERCENT_SHIFT 2 | |
7324 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
7325 | s64 percent; | |
7326 | u64 util; | |
7327 | int ret; | |
7328 | }; | |
7329 | ||
7330 | static inline struct uclamp_request | |
7331 | capacity_from_percent(char *buf) | |
7332 | { | |
7333 | struct uclamp_request req = { | |
7334 | .percent = UCLAMP_PERCENT_SCALE, | |
7335 | .util = SCHED_CAPACITY_SCALE, | |
7336 | .ret = 0, | |
7337 | }; | |
7338 | ||
7339 | buf = strim(buf); | |
7340 | if (strcmp(buf, "max")) { | |
7341 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
7342 | &req.percent); | |
7343 | if (req.ret) | |
7344 | return req; | |
b562d140 | 7345 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
7346 | req.ret = -ERANGE; |
7347 | return req; | |
7348 | } | |
7349 | ||
7350 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
7351 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
7352 | } | |
7353 | ||
7354 | return req; | |
7355 | } | |
7356 | ||
7357 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
7358 | size_t nbytes, loff_t off, | |
7359 | enum uclamp_id clamp_id) | |
7360 | { | |
7361 | struct uclamp_request req; | |
7362 | struct task_group *tg; | |
7363 | ||
7364 | req = capacity_from_percent(buf); | |
7365 | if (req.ret) | |
7366 | return req.ret; | |
7367 | ||
7368 | mutex_lock(&uclamp_mutex); | |
7369 | rcu_read_lock(); | |
7370 | ||
7371 | tg = css_tg(of_css(of)); | |
7372 | if (tg->uclamp_req[clamp_id].value != req.util) | |
7373 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
7374 | ||
7375 | /* | |
7376 | * Because of not recoverable conversion rounding we keep track of the | |
7377 | * exact requested value | |
7378 | */ | |
7379 | tg->uclamp_pct[clamp_id] = req.percent; | |
7380 | ||
0b60ba2d PB |
7381 | /* Update effective clamps to track the most restrictive value */ |
7382 | cpu_util_update_eff(of_css(of)); | |
7383 | ||
2480c093 PB |
7384 | rcu_read_unlock(); |
7385 | mutex_unlock(&uclamp_mutex); | |
7386 | ||
7387 | return nbytes; | |
7388 | } | |
7389 | ||
7390 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
7391 | char *buf, size_t nbytes, | |
7392 | loff_t off) | |
7393 | { | |
7394 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
7395 | } | |
7396 | ||
7397 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
7398 | char *buf, size_t nbytes, | |
7399 | loff_t off) | |
7400 | { | |
7401 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
7402 | } | |
7403 | ||
7404 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
7405 | enum uclamp_id clamp_id) | |
7406 | { | |
7407 | struct task_group *tg; | |
7408 | u64 util_clamp; | |
7409 | u64 percent; | |
7410 | u32 rem; | |
7411 | ||
7412 | rcu_read_lock(); | |
7413 | tg = css_tg(seq_css(sf)); | |
7414 | util_clamp = tg->uclamp_req[clamp_id].value; | |
7415 | rcu_read_unlock(); | |
7416 | ||
7417 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
7418 | seq_puts(sf, "max\n"); | |
7419 | return; | |
7420 | } | |
7421 | ||
7422 | percent = tg->uclamp_pct[clamp_id]; | |
7423 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
7424 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
7425 | } | |
7426 | ||
7427 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
7428 | { | |
7429 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
7430 | return 0; | |
7431 | } | |
7432 | ||
7433 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
7434 | { | |
7435 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
7436 | return 0; | |
7437 | } | |
7438 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
7439 | ||
052f1dc7 | 7440 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7441 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7442 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7443 | { |
5b61d50a KK |
7444 | if (shareval > scale_load_down(ULONG_MAX)) |
7445 | shareval = MAX_SHARES; | |
182446d0 | 7446 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7447 | } |
7448 | ||
182446d0 TH |
7449 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7450 | struct cftype *cft) | |
68318b8e | 7451 | { |
182446d0 | 7452 | struct task_group *tg = css_tg(css); |
68318b8e | 7453 | |
c8b28116 | 7454 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7455 | } |
ab84d31e PT |
7456 | |
7457 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7458 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7459 | ||
ab84d31e | 7460 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 7461 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
7462 | /* More than 203 days if BW_SHIFT equals 20. */ |
7463 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 7464 | |
a790de99 PT |
7465 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7466 | ||
ab84d31e PT |
7467 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7468 | { | |
56f570e5 | 7469 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7470 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7471 | |
7472 | if (tg == &root_task_group) | |
7473 | return -EINVAL; | |
7474 | ||
7475 | /* | |
7476 | * Ensure we have at some amount of bandwidth every period. This is | |
7477 | * to prevent reaching a state of large arrears when throttled via | |
7478 | * entity_tick() resulting in prolonged exit starvation. | |
7479 | */ | |
7480 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7481 | return -EINVAL; | |
7482 | ||
7483 | /* | |
7484 | * Likewise, bound things on the otherside by preventing insane quota | |
7485 | * periods. This also allows us to normalize in computing quota | |
7486 | * feasibility. | |
7487 | */ | |
7488 | if (period > max_cfs_quota_period) | |
7489 | return -EINVAL; | |
7490 | ||
d505b8af HC |
7491 | /* |
7492 | * Bound quota to defend quota against overflow during bandwidth shift. | |
7493 | */ | |
7494 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
7495 | return -EINVAL; | |
7496 | ||
0e59bdae KT |
7497 | /* |
7498 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
7499 | * unthrottle_offline_cfs_rqs(). | |
7500 | */ | |
7501 | get_online_cpus(); | |
a790de99 PT |
7502 | mutex_lock(&cfs_constraints_mutex); |
7503 | ret = __cfs_schedulable(tg, period, quota); | |
7504 | if (ret) | |
7505 | goto out_unlock; | |
7506 | ||
58088ad0 | 7507 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7508 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7509 | /* |
7510 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7511 | * before making related changes, and on->off must occur afterwards | |
7512 | */ | |
7513 | if (runtime_enabled && !runtime_was_enabled) | |
7514 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7515 | raw_spin_lock_irq(&cfs_b->lock); |
7516 | cfs_b->period = ns_to_ktime(period); | |
7517 | cfs_b->quota = quota; | |
58088ad0 | 7518 | |
a9cf55b2 | 7519 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
7520 | |
7521 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
7522 | if (runtime_enabled) |
7523 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 7524 | |
ab84d31e PT |
7525 | raw_spin_unlock_irq(&cfs_b->lock); |
7526 | ||
0e59bdae | 7527 | for_each_online_cpu(i) { |
ab84d31e | 7528 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 7529 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 7530 | struct rq_flags rf; |
ab84d31e | 7531 | |
8a8c69c3 | 7532 | rq_lock_irq(rq, &rf); |
58088ad0 | 7533 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7534 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7535 | |
029632fb | 7536 | if (cfs_rq->throttled) |
671fd9da | 7537 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 7538 | rq_unlock_irq(rq, &rf); |
ab84d31e | 7539 | } |
1ee14e6c BS |
7540 | if (runtime_was_enabled && !runtime_enabled) |
7541 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7542 | out_unlock: |
7543 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 7544 | put_online_cpus(); |
ab84d31e | 7545 | |
a790de99 | 7546 | return ret; |
ab84d31e PT |
7547 | } |
7548 | ||
b1546edc | 7549 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e PT |
7550 | { |
7551 | u64 quota, period; | |
7552 | ||
029632fb | 7553 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7554 | if (cfs_quota_us < 0) |
7555 | quota = RUNTIME_INF; | |
1a8b4540 | 7556 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 7557 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
7558 | else |
7559 | return -EINVAL; | |
ab84d31e PT |
7560 | |
7561 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7562 | } | |
7563 | ||
b1546edc | 7564 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
7565 | { |
7566 | u64 quota_us; | |
7567 | ||
029632fb | 7568 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7569 | return -1; |
7570 | ||
029632fb | 7571 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7572 | do_div(quota_us, NSEC_PER_USEC); |
7573 | ||
7574 | return quota_us; | |
7575 | } | |
7576 | ||
b1546edc | 7577 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e PT |
7578 | { |
7579 | u64 quota, period; | |
7580 | ||
1a8b4540 KK |
7581 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
7582 | return -EINVAL; | |
7583 | ||
ab84d31e | 7584 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 7585 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 7586 | |
ab84d31e PT |
7587 | return tg_set_cfs_bandwidth(tg, period, quota); |
7588 | } | |
7589 | ||
b1546edc | 7590 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
7591 | { |
7592 | u64 cfs_period_us; | |
7593 | ||
029632fb | 7594 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7595 | do_div(cfs_period_us, NSEC_PER_USEC); |
7596 | ||
7597 | return cfs_period_us; | |
7598 | } | |
7599 | ||
182446d0 TH |
7600 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
7601 | struct cftype *cft) | |
ab84d31e | 7602 | { |
182446d0 | 7603 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
7604 | } |
7605 | ||
182446d0 TH |
7606 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
7607 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 7608 | { |
182446d0 | 7609 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
7610 | } |
7611 | ||
182446d0 TH |
7612 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
7613 | struct cftype *cft) | |
ab84d31e | 7614 | { |
182446d0 | 7615 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
7616 | } |
7617 | ||
182446d0 TH |
7618 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
7619 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 7620 | { |
182446d0 | 7621 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
7622 | } |
7623 | ||
a790de99 PT |
7624 | struct cfs_schedulable_data { |
7625 | struct task_group *tg; | |
7626 | u64 period, quota; | |
7627 | }; | |
7628 | ||
7629 | /* | |
7630 | * normalize group quota/period to be quota/max_period | |
7631 | * note: units are usecs | |
7632 | */ | |
7633 | static u64 normalize_cfs_quota(struct task_group *tg, | |
7634 | struct cfs_schedulable_data *d) | |
7635 | { | |
7636 | u64 quota, period; | |
7637 | ||
7638 | if (tg == d->tg) { | |
7639 | period = d->period; | |
7640 | quota = d->quota; | |
7641 | } else { | |
7642 | period = tg_get_cfs_period(tg); | |
7643 | quota = tg_get_cfs_quota(tg); | |
7644 | } | |
7645 | ||
7646 | /* note: these should typically be equivalent */ | |
7647 | if (quota == RUNTIME_INF || quota == -1) | |
7648 | return RUNTIME_INF; | |
7649 | ||
7650 | return to_ratio(period, quota); | |
7651 | } | |
7652 | ||
7653 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
7654 | { | |
7655 | struct cfs_schedulable_data *d = data; | |
029632fb | 7656 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
7657 | s64 quota = 0, parent_quota = -1; |
7658 | ||
7659 | if (!tg->parent) { | |
7660 | quota = RUNTIME_INF; | |
7661 | } else { | |
029632fb | 7662 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
7663 | |
7664 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 7665 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
7666 | |
7667 | /* | |
c53593e5 TH |
7668 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
7669 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 7670 | * limit is set: |
a790de99 | 7671 | */ |
c53593e5 TH |
7672 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
7673 | quota = min(quota, parent_quota); | |
7674 | } else { | |
7675 | if (quota == RUNTIME_INF) | |
7676 | quota = parent_quota; | |
7677 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
7678 | return -EINVAL; | |
7679 | } | |
a790de99 | 7680 | } |
9c58c79a | 7681 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
7682 | |
7683 | return 0; | |
7684 | } | |
7685 | ||
7686 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
7687 | { | |
8277434e | 7688 | int ret; |
a790de99 PT |
7689 | struct cfs_schedulable_data data = { |
7690 | .tg = tg, | |
7691 | .period = period, | |
7692 | .quota = quota, | |
7693 | }; | |
7694 | ||
7695 | if (quota != RUNTIME_INF) { | |
7696 | do_div(data.period, NSEC_PER_USEC); | |
7697 | do_div(data.quota, NSEC_PER_USEC); | |
7698 | } | |
7699 | ||
8277434e PT |
7700 | rcu_read_lock(); |
7701 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
7702 | rcu_read_unlock(); | |
7703 | ||
7704 | return ret; | |
a790de99 | 7705 | } |
e8da1b18 | 7706 | |
a1f7164c | 7707 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 7708 | { |
2da8ca82 | 7709 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 7710 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 7711 | |
44ffc75b TH |
7712 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
7713 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
7714 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 7715 | |
3d6c50c2 YW |
7716 | if (schedstat_enabled() && tg != &root_task_group) { |
7717 | u64 ws = 0; | |
7718 | int i; | |
7719 | ||
7720 | for_each_possible_cpu(i) | |
7721 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
7722 | ||
7723 | seq_printf(sf, "wait_sum %llu\n", ws); | |
7724 | } | |
7725 | ||
e8da1b18 NR |
7726 | return 0; |
7727 | } | |
ab84d31e | 7728 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 7729 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 7730 | |
052f1dc7 | 7731 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
7732 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
7733 | struct cftype *cft, s64 val) | |
6f505b16 | 7734 | { |
182446d0 | 7735 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
7736 | } |
7737 | ||
182446d0 TH |
7738 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
7739 | struct cftype *cft) | |
6f505b16 | 7740 | { |
182446d0 | 7741 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 7742 | } |
d0b27fa7 | 7743 | |
182446d0 TH |
7744 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
7745 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 7746 | { |
182446d0 | 7747 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
7748 | } |
7749 | ||
182446d0 TH |
7750 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
7751 | struct cftype *cft) | |
d0b27fa7 | 7752 | { |
182446d0 | 7753 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 7754 | } |
6d6bc0ad | 7755 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 7756 | |
a1f7164c | 7757 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 7758 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
7759 | { |
7760 | .name = "shares", | |
f4c753b7 PM |
7761 | .read_u64 = cpu_shares_read_u64, |
7762 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 7763 | }, |
052f1dc7 | 7764 | #endif |
ab84d31e PT |
7765 | #ifdef CONFIG_CFS_BANDWIDTH |
7766 | { | |
7767 | .name = "cfs_quota_us", | |
7768 | .read_s64 = cpu_cfs_quota_read_s64, | |
7769 | .write_s64 = cpu_cfs_quota_write_s64, | |
7770 | }, | |
7771 | { | |
7772 | .name = "cfs_period_us", | |
7773 | .read_u64 = cpu_cfs_period_read_u64, | |
7774 | .write_u64 = cpu_cfs_period_write_u64, | |
7775 | }, | |
e8da1b18 NR |
7776 | { |
7777 | .name = "stat", | |
a1f7164c | 7778 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 7779 | }, |
ab84d31e | 7780 | #endif |
052f1dc7 | 7781 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7782 | { |
9f0c1e56 | 7783 | .name = "rt_runtime_us", |
06ecb27c PM |
7784 | .read_s64 = cpu_rt_runtime_read, |
7785 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 7786 | }, |
d0b27fa7 PZ |
7787 | { |
7788 | .name = "rt_period_us", | |
f4c753b7 PM |
7789 | .read_u64 = cpu_rt_period_read_uint, |
7790 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 7791 | }, |
2480c093 PB |
7792 | #endif |
7793 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
7794 | { | |
7795 | .name = "uclamp.min", | |
7796 | .flags = CFTYPE_NOT_ON_ROOT, | |
7797 | .seq_show = cpu_uclamp_min_show, | |
7798 | .write = cpu_uclamp_min_write, | |
7799 | }, | |
7800 | { | |
7801 | .name = "uclamp.max", | |
7802 | .flags = CFTYPE_NOT_ON_ROOT, | |
7803 | .seq_show = cpu_uclamp_max_show, | |
7804 | .write = cpu_uclamp_max_write, | |
7805 | }, | |
052f1dc7 | 7806 | #endif |
d1ccc66d | 7807 | { } /* Terminate */ |
68318b8e SV |
7808 | }; |
7809 | ||
d41bf8c9 TH |
7810 | static int cpu_extra_stat_show(struct seq_file *sf, |
7811 | struct cgroup_subsys_state *css) | |
0d593634 | 7812 | { |
0d593634 TH |
7813 | #ifdef CONFIG_CFS_BANDWIDTH |
7814 | { | |
d41bf8c9 | 7815 | struct task_group *tg = css_tg(css); |
0d593634 TH |
7816 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
7817 | u64 throttled_usec; | |
7818 | ||
7819 | throttled_usec = cfs_b->throttled_time; | |
7820 | do_div(throttled_usec, NSEC_PER_USEC); | |
7821 | ||
7822 | seq_printf(sf, "nr_periods %d\n" | |
7823 | "nr_throttled %d\n" | |
7824 | "throttled_usec %llu\n", | |
7825 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
7826 | throttled_usec); | |
7827 | } | |
7828 | #endif | |
7829 | return 0; | |
7830 | } | |
7831 | ||
7832 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7833 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
7834 | struct cftype *cft) | |
7835 | { | |
7836 | struct task_group *tg = css_tg(css); | |
7837 | u64 weight = scale_load_down(tg->shares); | |
7838 | ||
7839 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
7840 | } | |
7841 | ||
7842 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
7843 | struct cftype *cft, u64 weight) | |
7844 | { | |
7845 | /* | |
7846 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
7847 | * values which are 1, 100 and 10000 respectively. While it loses | |
7848 | * a bit of range on both ends, it maps pretty well onto the shares | |
7849 | * value used by scheduler and the round-trip conversions preserve | |
7850 | * the original value over the entire range. | |
7851 | */ | |
7852 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
7853 | return -ERANGE; | |
7854 | ||
7855 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
7856 | ||
7857 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
7858 | } | |
7859 | ||
7860 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
7861 | struct cftype *cft) | |
7862 | { | |
7863 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
7864 | int last_delta = INT_MAX; | |
7865 | int prio, delta; | |
7866 | ||
7867 | /* find the closest nice value to the current weight */ | |
7868 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
7869 | delta = abs(sched_prio_to_weight[prio] - weight); | |
7870 | if (delta >= last_delta) | |
7871 | break; | |
7872 | last_delta = delta; | |
7873 | } | |
7874 | ||
7875 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
7876 | } | |
7877 | ||
7878 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
7879 | struct cftype *cft, s64 nice) | |
7880 | { | |
7881 | unsigned long weight; | |
7281c8de | 7882 | int idx; |
0d593634 TH |
7883 | |
7884 | if (nice < MIN_NICE || nice > MAX_NICE) | |
7885 | return -ERANGE; | |
7886 | ||
7281c8de PZ |
7887 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
7888 | idx = array_index_nospec(idx, 40); | |
7889 | weight = sched_prio_to_weight[idx]; | |
7890 | ||
0d593634 TH |
7891 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
7892 | } | |
7893 | #endif | |
7894 | ||
7895 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
7896 | long period, long quota) | |
7897 | { | |
7898 | if (quota < 0) | |
7899 | seq_puts(sf, "max"); | |
7900 | else | |
7901 | seq_printf(sf, "%ld", quota); | |
7902 | ||
7903 | seq_printf(sf, " %ld\n", period); | |
7904 | } | |
7905 | ||
7906 | /* caller should put the current value in *@periodp before calling */ | |
7907 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
7908 | u64 *periodp, u64 *quotap) | |
7909 | { | |
7910 | char tok[21]; /* U64_MAX */ | |
7911 | ||
4c47acd8 | 7912 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
7913 | return -EINVAL; |
7914 | ||
7915 | *periodp *= NSEC_PER_USEC; | |
7916 | ||
7917 | if (sscanf(tok, "%llu", quotap)) | |
7918 | *quotap *= NSEC_PER_USEC; | |
7919 | else if (!strcmp(tok, "max")) | |
7920 | *quotap = RUNTIME_INF; | |
7921 | else | |
7922 | return -EINVAL; | |
7923 | ||
7924 | return 0; | |
7925 | } | |
7926 | ||
7927 | #ifdef CONFIG_CFS_BANDWIDTH | |
7928 | static int cpu_max_show(struct seq_file *sf, void *v) | |
7929 | { | |
7930 | struct task_group *tg = css_tg(seq_css(sf)); | |
7931 | ||
7932 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
7933 | return 0; | |
7934 | } | |
7935 | ||
7936 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
7937 | char *buf, size_t nbytes, loff_t off) | |
7938 | { | |
7939 | struct task_group *tg = css_tg(of_css(of)); | |
7940 | u64 period = tg_get_cfs_period(tg); | |
7941 | u64 quota; | |
7942 | int ret; | |
7943 | ||
7944 | ret = cpu_period_quota_parse(buf, &period, "a); | |
7945 | if (!ret) | |
7946 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
7947 | return ret ?: nbytes; | |
7948 | } | |
7949 | #endif | |
7950 | ||
7951 | static struct cftype cpu_files[] = { | |
0d593634 TH |
7952 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7953 | { | |
7954 | .name = "weight", | |
7955 | .flags = CFTYPE_NOT_ON_ROOT, | |
7956 | .read_u64 = cpu_weight_read_u64, | |
7957 | .write_u64 = cpu_weight_write_u64, | |
7958 | }, | |
7959 | { | |
7960 | .name = "weight.nice", | |
7961 | .flags = CFTYPE_NOT_ON_ROOT, | |
7962 | .read_s64 = cpu_weight_nice_read_s64, | |
7963 | .write_s64 = cpu_weight_nice_write_s64, | |
7964 | }, | |
7965 | #endif | |
7966 | #ifdef CONFIG_CFS_BANDWIDTH | |
7967 | { | |
7968 | .name = "max", | |
7969 | .flags = CFTYPE_NOT_ON_ROOT, | |
7970 | .seq_show = cpu_max_show, | |
7971 | .write = cpu_max_write, | |
7972 | }, | |
2480c093 PB |
7973 | #endif |
7974 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
7975 | { | |
7976 | .name = "uclamp.min", | |
7977 | .flags = CFTYPE_NOT_ON_ROOT, | |
7978 | .seq_show = cpu_uclamp_min_show, | |
7979 | .write = cpu_uclamp_min_write, | |
7980 | }, | |
7981 | { | |
7982 | .name = "uclamp.max", | |
7983 | .flags = CFTYPE_NOT_ON_ROOT, | |
7984 | .seq_show = cpu_uclamp_max_show, | |
7985 | .write = cpu_uclamp_max_write, | |
7986 | }, | |
0d593634 TH |
7987 | #endif |
7988 | { } /* terminate */ | |
7989 | }; | |
7990 | ||
073219e9 | 7991 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 7992 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 7993 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 7994 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 7995 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 7996 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 7997 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
7998 | .can_attach = cpu_cgroup_can_attach, |
7999 | .attach = cpu_cgroup_attach, | |
a1f7164c | 8000 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 8001 | .dfl_cftypes = cpu_files, |
b38e42e9 | 8002 | .early_init = true, |
0d593634 | 8003 | .threaded = true, |
68318b8e SV |
8004 | }; |
8005 | ||
052f1dc7 | 8006 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8007 | |
b637a328 PM |
8008 | void dump_cpu_task(int cpu) |
8009 | { | |
8010 | pr_info("Task dump for CPU %d:\n", cpu); | |
8011 | sched_show_task(cpu_curr(cpu)); | |
8012 | } | |
ed82b8a1 AK |
8013 | |
8014 | /* | |
8015 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8016 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8017 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8018 | * that remained on nice 0. | |
8019 | * | |
8020 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8021 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8022 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8023 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8024 | * the relative distance between them is ~25%.) | |
8025 | */ | |
8026 | const int sched_prio_to_weight[40] = { | |
8027 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8028 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8029 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8030 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8031 | /* 0 */ 1024, 820, 655, 526, 423, | |
8032 | /* 5 */ 335, 272, 215, 172, 137, | |
8033 | /* 10 */ 110, 87, 70, 56, 45, | |
8034 | /* 15 */ 36, 29, 23, 18, 15, | |
8035 | }; | |
8036 | ||
8037 | /* | |
8038 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8039 | * | |
8040 | * In cases where the weight does not change often, we can use the | |
8041 | * precalculated inverse to speed up arithmetics by turning divisions | |
8042 | * into multiplications: | |
8043 | */ | |
8044 | const u32 sched_prio_to_wmult[40] = { | |
8045 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8046 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8047 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8048 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8049 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8050 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8051 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8052 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8053 | }; | |
14a7405b IM |
8054 | |
8055 | #undef CREATE_TRACE_POINTS |