]>
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 | */ |
9d246053 PA |
9 | #define CREATE_TRACE_POINTS |
10 | #include <trace/events/sched.h> | |
11 | #undef CREATE_TRACE_POINTS | |
12 | ||
325ea10c | 13 | #include "sched.h" |
1da177e4 | 14 | |
7281c8de | 15 | #include <linux/nospec.h> |
85f1abe0 | 16 | |
0ed557aa | 17 | #include <linux/kcov.h> |
d08b9f0c | 18 | #include <linux/scs.h> |
0ed557aa | 19 | |
96f951ed | 20 | #include <asm/switch_to.h> |
5517d86b | 21 | #include <asm/tlb.h> |
1da177e4 | 22 | |
ea138446 | 23 | #include "../workqueue_internal.h" |
771b53d0 | 24 | #include "../../fs/io-wq.h" |
29d5e047 | 25 | #include "../smpboot.h" |
6e0534f2 | 26 | |
91c27493 | 27 | #include "pelt.h" |
1f8db415 | 28 | #include "smp.h" |
91c27493 | 29 | |
a056a5be QY |
30 | /* |
31 | * Export tracepoints that act as a bare tracehook (ie: have no trace event | |
32 | * associated with them) to allow external modules to probe them. | |
33 | */ | |
34 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp); | |
35 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp); | |
36 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp); | |
37 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp); | |
38 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp); | |
51cf18c9 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp); |
a056a5be | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); |
4581bea8 VD |
41 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp); |
42 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); | |
9d246053 | 43 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); |
a056a5be | 44 | |
029632fb | 45 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 46 | |
a73f863a | 47 | #ifdef CONFIG_SCHED_DEBUG |
bf5c91ba IM |
48 | /* |
49 | * Debugging: various feature bits | |
765cc3a4 PB |
50 | * |
51 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
52 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
53 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 54 | */ |
f00b45c1 PZ |
55 | #define SCHED_FEAT(name, enabled) \ |
56 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 57 | const_debug unsigned int sysctl_sched_features = |
391e43da | 58 | #include "features.h" |
f00b45c1 | 59 | 0; |
f00b45c1 | 60 | #undef SCHED_FEAT |
765cc3a4 | 61 | #endif |
f00b45c1 | 62 | |
b82d9fdd PZ |
63 | /* |
64 | * Number of tasks to iterate in a single balance run. | |
65 | * Limited because this is done with IRQs disabled. | |
66 | */ | |
67 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
68 | ||
fa85ae24 | 69 | /* |
d1ccc66d | 70 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
71 | * default: 1s |
72 | */ | |
9f0c1e56 | 73 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 74 | |
029632fb | 75 | __read_mostly int scheduler_running; |
6892b75e | 76 | |
9f0c1e56 PZ |
77 | /* |
78 | * part of the period that we allow rt tasks to run in us. | |
79 | * default: 0.95s | |
80 | */ | |
81 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 82 | |
58877d34 PZ |
83 | |
84 | /* | |
85 | * Serialization rules: | |
86 | * | |
87 | * Lock order: | |
88 | * | |
89 | * p->pi_lock | |
90 | * rq->lock | |
91 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
92 | * | |
93 | * rq1->lock | |
94 | * rq2->lock where: rq1 < rq2 | |
95 | * | |
96 | * Regular state: | |
97 | * | |
98 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
99 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
b19a888c | 100 | * always looks at the local rq data structures to find the most eligible task |
58877d34 PZ |
101 | * to run next. |
102 | * | |
103 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
104 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
105 | * the local CPU to avoid bouncing the runqueue state around [ see | |
106 | * ttwu_queue_wakelist() ] | |
107 | * | |
108 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
109 | * complicated to avoid having to take two rq->locks. | |
110 | * | |
111 | * Special state: | |
112 | * | |
113 | * System-calls and anything external will use task_rq_lock() which acquires | |
114 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
115 | * stable while holding either lock: | |
116 | * | |
117 | * - sched_setaffinity()/ | |
118 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
119 | * - set_user_nice(): p->se.load, p->*prio | |
120 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
121 | * p->se.load, p->rt_priority, | |
122 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
123 | * - sched_setnuma(): p->numa_preferred_nid | |
124 | * - sched_move_task()/ | |
125 | * cpu_cgroup_fork(): p->sched_task_group | |
126 | * - uclamp_update_active() p->uclamp* | |
127 | * | |
128 | * p->state <- TASK_*: | |
129 | * | |
130 | * is changed locklessly using set_current_state(), __set_current_state() or | |
131 | * set_special_state(), see their respective comments, or by | |
132 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
133 | * concurrent self. | |
134 | * | |
135 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
136 | * | |
137 | * is set by activate_task() and cleared by deactivate_task(), under | |
138 | * rq->lock. Non-zero indicates the task is runnable, the special | |
139 | * ON_RQ_MIGRATING state is used for migration without holding both | |
140 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
141 | * | |
142 | * p->on_cpu <- { 0, 1 }: | |
143 | * | |
144 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
145 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
146 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
147 | * | |
148 | * [ The astute reader will observe that it is possible for two tasks on one | |
149 | * CPU to have ->on_cpu = 1 at the same time. ] | |
150 | * | |
151 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
152 | * | |
153 | * - Don't call set_task_cpu() on a blocked task: | |
154 | * | |
155 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
156 | * the CPU assignment of blocked tasks isn't required to be valid. | |
157 | * | |
158 | * - for try_to_wake_up(), called under p->pi_lock: | |
159 | * | |
160 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
161 | * | |
162 | * - for migration called under rq->lock: | |
163 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
164 | * | |
165 | * o move_queued_task() | |
166 | * o detach_task() | |
167 | * | |
168 | * - for migration called under double_rq_lock(): | |
169 | * | |
170 | * o __migrate_swap_task() | |
171 | * o push_rt_task() / pull_rt_task() | |
172 | * o push_dl_task() / pull_dl_task() | |
173 | * o dl_task_offline_migration() | |
174 | * | |
175 | */ | |
176 | ||
3e71a462 PZ |
177 | /* |
178 | * __task_rq_lock - lock the rq @p resides on. | |
179 | */ | |
eb580751 | 180 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
181 | __acquires(rq->lock) |
182 | { | |
183 | struct rq *rq; | |
184 | ||
185 | lockdep_assert_held(&p->pi_lock); | |
186 | ||
187 | for (;;) { | |
188 | rq = task_rq(p); | |
189 | raw_spin_lock(&rq->lock); | |
190 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 191 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
192 | return rq; |
193 | } | |
194 | raw_spin_unlock(&rq->lock); | |
195 | ||
196 | while (unlikely(task_on_rq_migrating(p))) | |
197 | cpu_relax(); | |
198 | } | |
199 | } | |
200 | ||
201 | /* | |
202 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
203 | */ | |
eb580751 | 204 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
205 | __acquires(p->pi_lock) |
206 | __acquires(rq->lock) | |
207 | { | |
208 | struct rq *rq; | |
209 | ||
210 | for (;;) { | |
eb580751 | 211 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
212 | rq = task_rq(p); |
213 | raw_spin_lock(&rq->lock); | |
214 | /* | |
215 | * move_queued_task() task_rq_lock() | |
216 | * | |
217 | * ACQUIRE (rq->lock) | |
218 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
219 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
220 | * [S] ->cpu = new_cpu [L] task_rq() | |
221 | * [L] ->on_rq | |
222 | * RELEASE (rq->lock) | |
223 | * | |
c546951d | 224 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
225 | * the old rq->lock will fully serialize against the stores. |
226 | * | |
c546951d AP |
227 | * If we observe the new CPU in task_rq_lock(), the address |
228 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
229 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
230 | */ |
231 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 232 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
233 | return rq; |
234 | } | |
235 | raw_spin_unlock(&rq->lock); | |
eb580751 | 236 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
237 | |
238 | while (unlikely(task_on_rq_migrating(p))) | |
239 | cpu_relax(); | |
240 | } | |
241 | } | |
242 | ||
535b9552 IM |
243 | /* |
244 | * RQ-clock updating methods: | |
245 | */ | |
246 | ||
247 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
248 | { | |
249 | /* | |
250 | * In theory, the compile should just see 0 here, and optimize out the call | |
251 | * to sched_rt_avg_update. But I don't trust it... | |
252 | */ | |
11d4afd4 VG |
253 | s64 __maybe_unused steal = 0, irq_delta = 0; |
254 | ||
535b9552 IM |
255 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
256 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
257 | ||
258 | /* | |
259 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
260 | * this case when a previous update_rq_clock() happened inside a | |
261 | * {soft,}irq region. | |
262 | * | |
263 | * When this happens, we stop ->clock_task and only update the | |
264 | * prev_irq_time stamp to account for the part that fit, so that a next | |
265 | * update will consume the rest. This ensures ->clock_task is | |
266 | * monotonic. | |
267 | * | |
268 | * It does however cause some slight miss-attribution of {soft,}irq | |
269 | * time, a more accurate solution would be to update the irq_time using | |
270 | * the current rq->clock timestamp, except that would require using | |
271 | * atomic ops. | |
272 | */ | |
273 | if (irq_delta > delta) | |
274 | irq_delta = delta; | |
275 | ||
276 | rq->prev_irq_time += irq_delta; | |
277 | delta -= irq_delta; | |
278 | #endif | |
279 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
280 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
281 | steal = paravirt_steal_clock(cpu_of(rq)); | |
282 | steal -= rq->prev_steal_time_rq; | |
283 | ||
284 | if (unlikely(steal > delta)) | |
285 | steal = delta; | |
286 | ||
287 | rq->prev_steal_time_rq += steal; | |
288 | delta -= steal; | |
289 | } | |
290 | #endif | |
291 | ||
292 | rq->clock_task += delta; | |
293 | ||
11d4afd4 | 294 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 295 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 296 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 297 | #endif |
23127296 | 298 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
299 | } |
300 | ||
301 | void update_rq_clock(struct rq *rq) | |
302 | { | |
303 | s64 delta; | |
304 | ||
305 | lockdep_assert_held(&rq->lock); | |
306 | ||
307 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
308 | return; | |
309 | ||
310 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
311 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
312 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
313 | rq->clock_update_flags |= RQCF_UPDATED; |
314 | #endif | |
26ae58d2 | 315 | |
535b9552 IM |
316 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
317 | if (delta < 0) | |
318 | return; | |
319 | rq->clock += delta; | |
320 | update_rq_clock_task(rq, delta); | |
321 | } | |
322 | ||
8f4d37ec PZ |
323 | #ifdef CONFIG_SCHED_HRTICK |
324 | /* | |
325 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 326 | */ |
8f4d37ec | 327 | |
8f4d37ec PZ |
328 | static void hrtick_clear(struct rq *rq) |
329 | { | |
330 | if (hrtimer_active(&rq->hrtick_timer)) | |
331 | hrtimer_cancel(&rq->hrtick_timer); | |
332 | } | |
333 | ||
8f4d37ec PZ |
334 | /* |
335 | * High-resolution timer tick. | |
336 | * Runs from hardirq context with interrupts disabled. | |
337 | */ | |
338 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
339 | { | |
340 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 341 | struct rq_flags rf; |
8f4d37ec PZ |
342 | |
343 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
344 | ||
8a8c69c3 | 345 | rq_lock(rq, &rf); |
3e51f33f | 346 | update_rq_clock(rq); |
8f4d37ec | 347 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 348 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
349 | |
350 | return HRTIMER_NORESTART; | |
351 | } | |
352 | ||
95e904c7 | 353 | #ifdef CONFIG_SMP |
971ee28c | 354 | |
4961b6e1 | 355 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
356 | { |
357 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 358 | |
d5096aa6 | 359 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
360 | } |
361 | ||
31656519 PZ |
362 | /* |
363 | * called from hardirq (IPI) context | |
364 | */ | |
365 | static void __hrtick_start(void *arg) | |
b328ca18 | 366 | { |
31656519 | 367 | struct rq *rq = arg; |
8a8c69c3 | 368 | struct rq_flags rf; |
b328ca18 | 369 | |
8a8c69c3 | 370 | rq_lock(rq, &rf); |
971ee28c | 371 | __hrtick_restart(rq); |
8a8c69c3 | 372 | rq_unlock(rq, &rf); |
b328ca18 PZ |
373 | } |
374 | ||
31656519 PZ |
375 | /* |
376 | * Called to set the hrtick timer state. | |
377 | * | |
378 | * called with rq->lock held and irqs disabled | |
379 | */ | |
029632fb | 380 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 381 | { |
31656519 | 382 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 383 | ktime_t time; |
384 | s64 delta; | |
385 | ||
386 | /* | |
387 | * Don't schedule slices shorter than 10000ns, that just | |
388 | * doesn't make sense and can cause timer DoS. | |
389 | */ | |
390 | delta = max_t(s64, delay, 10000LL); | |
391 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 392 | |
cc584b21 | 393 | hrtimer_set_expires(timer, time); |
31656519 | 394 | |
fd3eafda | 395 | if (rq == this_rq()) |
971ee28c | 396 | __hrtick_restart(rq); |
fd3eafda | 397 | else |
c46fff2a | 398 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
399 | } |
400 | ||
31656519 PZ |
401 | #else |
402 | /* | |
403 | * Called to set the hrtick timer state. | |
404 | * | |
405 | * called with rq->lock held and irqs disabled | |
406 | */ | |
029632fb | 407 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 408 | { |
86893335 WL |
409 | /* |
410 | * Don't schedule slices shorter than 10000ns, that just | |
411 | * doesn't make sense. Rely on vruntime for fairness. | |
412 | */ | |
413 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 414 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 415 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 416 | } |
90b5363a | 417 | |
31656519 | 418 | #endif /* CONFIG_SMP */ |
8f4d37ec | 419 | |
77a021be | 420 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 421 | { |
31656519 | 422 | #ifdef CONFIG_SMP |
545b8c8d | 423 | INIT_CSD(&rq->hrtick_csd, __hrtick_start, rq); |
31656519 | 424 | #endif |
d5096aa6 | 425 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 426 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 427 | } |
006c75f1 | 428 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
429 | static inline void hrtick_clear(struct rq *rq) |
430 | { | |
431 | } | |
432 | ||
77a021be | 433 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
434 | { |
435 | } | |
006c75f1 | 436 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 437 | |
5529578a FW |
438 | /* |
439 | * cmpxchg based fetch_or, macro so it works for different integer types | |
440 | */ | |
441 | #define fetch_or(ptr, mask) \ | |
442 | ({ \ | |
443 | typeof(ptr) _ptr = (ptr); \ | |
444 | typeof(mask) _mask = (mask); \ | |
445 | typeof(*_ptr) _old, _val = *_ptr; \ | |
446 | \ | |
447 | for (;;) { \ | |
448 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
449 | if (_old == _val) \ | |
450 | break; \ | |
451 | _val = _old; \ | |
452 | } \ | |
453 | _old; \ | |
454 | }) | |
455 | ||
e3baac47 | 456 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
457 | /* |
458 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
459 | * this avoids any races wrt polling state changes and thereby avoids | |
460 | * spurious IPIs. | |
461 | */ | |
462 | static bool set_nr_and_not_polling(struct task_struct *p) | |
463 | { | |
464 | struct thread_info *ti = task_thread_info(p); | |
465 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
466 | } | |
e3baac47 PZ |
467 | |
468 | /* | |
469 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
470 | * | |
471 | * If this returns true, then the idle task promises to call | |
472 | * sched_ttwu_pending() and reschedule soon. | |
473 | */ | |
474 | static bool set_nr_if_polling(struct task_struct *p) | |
475 | { | |
476 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 477 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
478 | |
479 | for (;;) { | |
480 | if (!(val & _TIF_POLLING_NRFLAG)) | |
481 | return false; | |
482 | if (val & _TIF_NEED_RESCHED) | |
483 | return true; | |
484 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
485 | if (old == val) | |
486 | break; | |
487 | val = old; | |
488 | } | |
489 | return true; | |
490 | } | |
491 | ||
fd99f91a PZ |
492 | #else |
493 | static bool set_nr_and_not_polling(struct task_struct *p) | |
494 | { | |
495 | set_tsk_need_resched(p); | |
496 | return true; | |
497 | } | |
e3baac47 PZ |
498 | |
499 | #ifdef CONFIG_SMP | |
500 | static bool set_nr_if_polling(struct task_struct *p) | |
501 | { | |
502 | return false; | |
503 | } | |
504 | #endif | |
fd99f91a PZ |
505 | #endif |
506 | ||
07879c6a | 507 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
508 | { |
509 | struct wake_q_node *node = &task->wake_q; | |
510 | ||
511 | /* | |
512 | * Atomically grab the task, if ->wake_q is !nil already it means | |
b19a888c | 513 | * it's already queued (either by us or someone else) and will get the |
76751049 PZ |
514 | * wakeup due to that. |
515 | * | |
4c4e3731 PZ |
516 | * In order to ensure that a pending wakeup will observe our pending |
517 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 518 | */ |
4c4e3731 | 519 | smp_mb__before_atomic(); |
87ff19cb | 520 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 521 | return false; |
76751049 PZ |
522 | |
523 | /* | |
524 | * The head is context local, there can be no concurrency. | |
525 | */ | |
526 | *head->lastp = node; | |
527 | head->lastp = &node->next; | |
07879c6a DB |
528 | return true; |
529 | } | |
530 | ||
531 | /** | |
532 | * wake_q_add() - queue a wakeup for 'later' waking. | |
533 | * @head: the wake_q_head to add @task to | |
534 | * @task: the task to queue for 'later' wakeup | |
535 | * | |
536 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
537 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
538 | * instantly. | |
539 | * | |
540 | * This function must be used as-if it were wake_up_process(); IOW the task | |
541 | * must be ready to be woken at this location. | |
542 | */ | |
543 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
544 | { | |
545 | if (__wake_q_add(head, task)) | |
546 | get_task_struct(task); | |
547 | } | |
548 | ||
549 | /** | |
550 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
551 | * @head: the wake_q_head to add @task to | |
552 | * @task: the task to queue for 'later' wakeup | |
553 | * | |
554 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
555 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
556 | * instantly. | |
557 | * | |
558 | * This function must be used as-if it were wake_up_process(); IOW the task | |
559 | * must be ready to be woken at this location. | |
560 | * | |
561 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
562 | * that already hold reference to @task can call the 'safe' version and trust | |
563 | * wake_q to do the right thing depending whether or not the @task is already | |
564 | * queued for wakeup. | |
565 | */ | |
566 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
567 | { | |
568 | if (!__wake_q_add(head, task)) | |
569 | put_task_struct(task); | |
76751049 PZ |
570 | } |
571 | ||
572 | void wake_up_q(struct wake_q_head *head) | |
573 | { | |
574 | struct wake_q_node *node = head->first; | |
575 | ||
576 | while (node != WAKE_Q_TAIL) { | |
577 | struct task_struct *task; | |
578 | ||
579 | task = container_of(node, struct task_struct, wake_q); | |
580 | BUG_ON(!task); | |
d1ccc66d | 581 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
582 | node = node->next; |
583 | task->wake_q.next = NULL; | |
584 | ||
585 | /* | |
7696f991 AP |
586 | * wake_up_process() executes a full barrier, which pairs with |
587 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
588 | */ |
589 | wake_up_process(task); | |
590 | put_task_struct(task); | |
591 | } | |
592 | } | |
593 | ||
c24d20db | 594 | /* |
8875125e | 595 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
596 | * |
597 | * On UP this means the setting of the need_resched flag, on SMP it | |
598 | * might also involve a cross-CPU call to trigger the scheduler on | |
599 | * the target CPU. | |
600 | */ | |
8875125e | 601 | void resched_curr(struct rq *rq) |
c24d20db | 602 | { |
8875125e | 603 | struct task_struct *curr = rq->curr; |
c24d20db IM |
604 | int cpu; |
605 | ||
8875125e | 606 | lockdep_assert_held(&rq->lock); |
c24d20db | 607 | |
8875125e | 608 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
609 | return; |
610 | ||
8875125e | 611 | cpu = cpu_of(rq); |
fd99f91a | 612 | |
f27dde8d | 613 | if (cpu == smp_processor_id()) { |
8875125e | 614 | set_tsk_need_resched(curr); |
f27dde8d | 615 | set_preempt_need_resched(); |
c24d20db | 616 | return; |
f27dde8d | 617 | } |
c24d20db | 618 | |
8875125e | 619 | if (set_nr_and_not_polling(curr)) |
c24d20db | 620 | smp_send_reschedule(cpu); |
dfc68f29 AL |
621 | else |
622 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
623 | } |
624 | ||
029632fb | 625 | void resched_cpu(int cpu) |
c24d20db IM |
626 | { |
627 | struct rq *rq = cpu_rq(cpu); | |
628 | unsigned long flags; | |
629 | ||
7c2102e5 | 630 | raw_spin_lock_irqsave(&rq->lock, flags); |
a0982dfa PM |
631 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
632 | resched_curr(rq); | |
05fa785c | 633 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 634 | } |
06d8308c | 635 | |
b021fe3e | 636 | #ifdef CONFIG_SMP |
3451d024 | 637 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 638 | /* |
d1ccc66d IM |
639 | * In the semi idle case, use the nearest busy CPU for migrating timers |
640 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
641 | * |
642 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
643 | * selecting an idle CPU will add more delays to the timers than intended |
644 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 645 | */ |
bc7a34b8 | 646 | int get_nohz_timer_target(void) |
83cd4fe2 | 647 | { |
e938b9c9 | 648 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 VP |
649 | struct sched_domain *sd; |
650 | ||
e938b9c9 WL |
651 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
652 | if (!idle_cpu(cpu)) | |
653 | return cpu; | |
654 | default_cpu = cpu; | |
655 | } | |
6201b4d6 | 656 | |
057f3fad | 657 | rcu_read_lock(); |
83cd4fe2 | 658 | for_each_domain(cpu, sd) { |
e938b9c9 WL |
659 | for_each_cpu_and(i, sched_domain_span(sd), |
660 | housekeeping_cpumask(HK_FLAG_TIMER)) { | |
44496922 WL |
661 | if (cpu == i) |
662 | continue; | |
663 | ||
e938b9c9 | 664 | if (!idle_cpu(i)) { |
057f3fad PZ |
665 | cpu = i; |
666 | goto unlock; | |
667 | } | |
668 | } | |
83cd4fe2 | 669 | } |
9642d18e | 670 | |
e938b9c9 WL |
671 | if (default_cpu == -1) |
672 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
673 | cpu = default_cpu; | |
057f3fad PZ |
674 | unlock: |
675 | rcu_read_unlock(); | |
83cd4fe2 VP |
676 | return cpu; |
677 | } | |
d1ccc66d | 678 | |
06d8308c TG |
679 | /* |
680 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
681 | * idle CPU then this timer might expire before the next timer event | |
682 | * which is scheduled to wake up that CPU. In case of a completely | |
683 | * idle system the next event might even be infinite time into the | |
684 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
685 | * leaves the inner idle loop so the newly added timer is taken into | |
686 | * account when the CPU goes back to idle and evaluates the timer | |
687 | * wheel for the next timer event. | |
688 | */ | |
1c20091e | 689 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
690 | { |
691 | struct rq *rq = cpu_rq(cpu); | |
692 | ||
693 | if (cpu == smp_processor_id()) | |
694 | return; | |
695 | ||
67b9ca70 | 696 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 697 | smp_send_reschedule(cpu); |
dfc68f29 AL |
698 | else |
699 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
700 | } |
701 | ||
c5bfece2 | 702 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 703 | { |
53c5fa16 FW |
704 | /* |
705 | * We just need the target to call irq_exit() and re-evaluate | |
706 | * the next tick. The nohz full kick at least implies that. | |
707 | * If needed we can still optimize that later with an | |
708 | * empty IRQ. | |
709 | */ | |
379d9ecb PM |
710 | if (cpu_is_offline(cpu)) |
711 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 712 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
713 | if (cpu != smp_processor_id() || |
714 | tick_nohz_tick_stopped()) | |
53c5fa16 | 715 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
716 | return true; |
717 | } | |
718 | ||
719 | return false; | |
720 | } | |
721 | ||
379d9ecb PM |
722 | /* |
723 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
724 | * caller's responsibility to deal with the lost wakeup, for example, | |
725 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
726 | */ | |
1c20091e FW |
727 | void wake_up_nohz_cpu(int cpu) |
728 | { | |
c5bfece2 | 729 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
730 | wake_up_idle_cpu(cpu); |
731 | } | |
732 | ||
19a1f5ec | 733 | static void nohz_csd_func(void *info) |
45bf76df | 734 | { |
19a1f5ec PZ |
735 | struct rq *rq = info; |
736 | int cpu = cpu_of(rq); | |
737 | unsigned int flags; | |
873b4c65 VG |
738 | |
739 | /* | |
19a1f5ec | 740 | * Release the rq::nohz_csd. |
873b4c65 | 741 | */ |
19a1f5ec PZ |
742 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(cpu)); |
743 | WARN_ON(!(flags & NOHZ_KICK_MASK)); | |
45bf76df | 744 | |
19a1f5ec PZ |
745 | rq->idle_balance = idle_cpu(cpu); |
746 | if (rq->idle_balance && !need_resched()) { | |
747 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
748 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
749 | } | |
2069dd75 PZ |
750 | } |
751 | ||
3451d024 | 752 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 753 | |
ce831b38 | 754 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 755 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 756 | { |
76d92ac3 FW |
757 | int fifo_nr_running; |
758 | ||
759 | /* Deadline tasks, even if single, need the tick */ | |
760 | if (rq->dl.dl_nr_running) | |
761 | return false; | |
762 | ||
1e78cdbd | 763 | /* |
b19a888c | 764 | * If there are more than one RR tasks, we need the tick to affect the |
2548d546 | 765 | * actual RR behaviour. |
1e78cdbd | 766 | */ |
76d92ac3 FW |
767 | if (rq->rt.rr_nr_running) { |
768 | if (rq->rt.rr_nr_running == 1) | |
769 | return true; | |
770 | else | |
771 | return false; | |
1e78cdbd RR |
772 | } |
773 | ||
2548d546 PZ |
774 | /* |
775 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
776 | * forced preemption between FIFO tasks. | |
777 | */ | |
778 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
779 | if (fifo_nr_running) | |
780 | return true; | |
781 | ||
782 | /* | |
783 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
784 | * if there's more than one we need the tick for involuntary | |
785 | * preemption. | |
786 | */ | |
787 | if (rq->nr_running > 1) | |
541b8264 | 788 | return false; |
ce831b38 | 789 | |
541b8264 | 790 | return true; |
ce831b38 FW |
791 | } |
792 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 793 | #endif /* CONFIG_SMP */ |
18d95a28 | 794 | |
a790de99 PT |
795 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
796 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 797 | /* |
8277434e PT |
798 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
799 | * node and @up when leaving it for the final time. | |
800 | * | |
801 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 802 | */ |
029632fb | 803 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 804 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
805 | { |
806 | struct task_group *parent, *child; | |
eb755805 | 807 | int ret; |
c09595f6 | 808 | |
8277434e PT |
809 | parent = from; |
810 | ||
c09595f6 | 811 | down: |
eb755805 PZ |
812 | ret = (*down)(parent, data); |
813 | if (ret) | |
8277434e | 814 | goto out; |
c09595f6 PZ |
815 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
816 | parent = child; | |
817 | goto down; | |
818 | ||
819 | up: | |
820 | continue; | |
821 | } | |
eb755805 | 822 | ret = (*up)(parent, data); |
8277434e PT |
823 | if (ret || parent == from) |
824 | goto out; | |
c09595f6 PZ |
825 | |
826 | child = parent; | |
827 | parent = parent->parent; | |
828 | if (parent) | |
829 | goto up; | |
8277434e | 830 | out: |
eb755805 | 831 | return ret; |
c09595f6 PZ |
832 | } |
833 | ||
029632fb | 834 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 835 | { |
e2b245f8 | 836 | return 0; |
eb755805 | 837 | } |
18d95a28 PZ |
838 | #endif |
839 | ||
9059393e | 840 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 841 | { |
f05998d4 NR |
842 | int prio = p->static_prio - MAX_RT_PRIO; |
843 | struct load_weight *load = &p->se.load; | |
844 | ||
dd41f596 IM |
845 | /* |
846 | * SCHED_IDLE tasks get minimal weight: | |
847 | */ | |
1da1843f | 848 | if (task_has_idle_policy(p)) { |
c8b28116 | 849 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 850 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
851 | return; |
852 | } | |
71f8bd46 | 853 | |
9059393e VG |
854 | /* |
855 | * SCHED_OTHER tasks have to update their load when changing their | |
856 | * weight | |
857 | */ | |
858 | if (update_load && p->sched_class == &fair_sched_class) { | |
859 | reweight_task(p, prio); | |
860 | } else { | |
861 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
862 | load->inv_weight = sched_prio_to_wmult[prio]; | |
863 | } | |
71f8bd46 IM |
864 | } |
865 | ||
69842cba | 866 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
867 | /* |
868 | * Serializes updates of utilization clamp values | |
869 | * | |
870 | * The (slow-path) user-space triggers utilization clamp value updates which | |
871 | * can require updates on (fast-path) scheduler's data structures used to | |
872 | * support enqueue/dequeue operations. | |
873 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
874 | * requests are serialized using a mutex to reduce the risk of conflicting | |
875 | * updates or API abuses. | |
876 | */ | |
877 | static DEFINE_MUTEX(uclamp_mutex); | |
878 | ||
e8f14172 PB |
879 | /* Max allowed minimum utilization */ |
880 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
881 | ||
882 | /* Max allowed maximum utilization */ | |
883 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
884 | ||
13685c4a QY |
885 | /* |
886 | * By default RT tasks run at the maximum performance point/capacity of the | |
887 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
888 | * SCHED_CAPACITY_SCALE. | |
889 | * | |
890 | * This knob allows admins to change the default behavior when uclamp is being | |
891 | * used. In battery powered devices, particularly, running at the maximum | |
892 | * capacity and frequency will increase energy consumption and shorten the | |
893 | * battery life. | |
894 | * | |
895 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
896 | * | |
897 | * This knob will not override the system default sched_util_clamp_min defined | |
898 | * above. | |
899 | */ | |
900 | unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; | |
901 | ||
e8f14172 PB |
902 | /* All clamps are required to be less or equal than these values */ |
903 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 904 | |
46609ce2 QY |
905 | /* |
906 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
907 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
908 | * enqueue/dequeue_task(). | |
909 | * | |
910 | * This allows users to continue to enable uclamp in their kernel config with | |
911 | * minimum uclamp overhead in the fast path. | |
912 | * | |
913 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
914 | * enabled, since we have an actual users that make use of uclamp | |
915 | * functionality. | |
916 | * | |
917 | * The knobs that would enable this static key are: | |
918 | * | |
919 | * * A task modifying its uclamp value with sched_setattr(). | |
920 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
921 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
922 | */ | |
923 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
924 | ||
69842cba PB |
925 | /* Integer rounded range for each bucket */ |
926 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
927 | ||
928 | #define for_each_clamp_id(clamp_id) \ | |
929 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
930 | ||
931 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
932 | { | |
933 | return clamp_value / UCLAMP_BUCKET_DELTA; | |
934 | } | |
935 | ||
7763baac | 936 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
937 | { |
938 | if (clamp_id == UCLAMP_MIN) | |
939 | return 0; | |
940 | return SCHED_CAPACITY_SCALE; | |
941 | } | |
942 | ||
a509a7cd PB |
943 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
944 | unsigned int value, bool user_defined) | |
69842cba PB |
945 | { |
946 | uc_se->value = value; | |
947 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 948 | uc_se->user_defined = user_defined; |
69842cba PB |
949 | } |
950 | ||
e496187d | 951 | static inline unsigned int |
0413d7f3 | 952 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
953 | unsigned int clamp_value) |
954 | { | |
955 | /* | |
956 | * Avoid blocked utilization pushing up the frequency when we go | |
957 | * idle (which drops the max-clamp) by retaining the last known | |
958 | * max-clamp. | |
959 | */ | |
960 | if (clamp_id == UCLAMP_MAX) { | |
961 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
962 | return clamp_value; | |
963 | } | |
964 | ||
965 | return uclamp_none(UCLAMP_MIN); | |
966 | } | |
967 | ||
0413d7f3 | 968 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
969 | unsigned int clamp_value) |
970 | { | |
971 | /* Reset max-clamp retention only on idle exit */ | |
972 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
973 | return; | |
974 | ||
975 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
976 | } | |
977 | ||
69842cba | 978 | static inline |
7763baac | 979 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 980 | unsigned int clamp_value) |
69842cba PB |
981 | { |
982 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
983 | int bucket_id = UCLAMP_BUCKETS - 1; | |
984 | ||
985 | /* | |
986 | * Since both min and max clamps are max aggregated, find the | |
987 | * top most bucket with tasks in. | |
988 | */ | |
989 | for ( ; bucket_id >= 0; bucket_id--) { | |
990 | if (!bucket[bucket_id].tasks) | |
991 | continue; | |
992 | return bucket[bucket_id].value; | |
993 | } | |
994 | ||
995 | /* No tasks -- default clamp values */ | |
e496187d | 996 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
997 | } |
998 | ||
13685c4a QY |
999 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
1000 | { | |
1001 | unsigned int default_util_min; | |
1002 | struct uclamp_se *uc_se; | |
1003 | ||
1004 | lockdep_assert_held(&p->pi_lock); | |
1005 | ||
1006 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1007 | ||
1008 | /* Only sync if user didn't override the default */ | |
1009 | if (uc_se->user_defined) | |
1010 | return; | |
1011 | ||
1012 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1013 | uclamp_se_set(uc_se, default_util_min, false); | |
1014 | } | |
1015 | ||
1016 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1017 | { | |
1018 | struct rq_flags rf; | |
1019 | struct rq *rq; | |
1020 | ||
1021 | if (!rt_task(p)) | |
1022 | return; | |
1023 | ||
1024 | /* Protect updates to p->uclamp_* */ | |
1025 | rq = task_rq_lock(p, &rf); | |
1026 | __uclamp_update_util_min_rt_default(p); | |
1027 | task_rq_unlock(rq, p, &rf); | |
1028 | } | |
1029 | ||
1030 | static void uclamp_sync_util_min_rt_default(void) | |
1031 | { | |
1032 | struct task_struct *g, *p; | |
1033 | ||
1034 | /* | |
1035 | * copy_process() sysctl_uclamp | |
1036 | * uclamp_min_rt = X; | |
1037 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1038 | * // link thread smp_mb__after_spinlock() | |
1039 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1040 | * sched_post_fork() for_each_process_thread() | |
1041 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1042 | * | |
1043 | * Ensures that either sched_post_fork() will observe the new | |
1044 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1045 | * task. | |
1046 | */ | |
1047 | read_lock(&tasklist_lock); | |
1048 | smp_mb__after_spinlock(); | |
1049 | read_unlock(&tasklist_lock); | |
1050 | ||
1051 | rcu_read_lock(); | |
1052 | for_each_process_thread(g, p) | |
1053 | uclamp_update_util_min_rt_default(p); | |
1054 | rcu_read_unlock(); | |
1055 | } | |
1056 | ||
3eac870a | 1057 | static inline struct uclamp_se |
0413d7f3 | 1058 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a PB |
1059 | { |
1060 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; | |
1061 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
1062 | struct uclamp_se uc_max; | |
1063 | ||
1064 | /* | |
1065 | * Tasks in autogroups or root task group will be | |
1066 | * restricted by system defaults. | |
1067 | */ | |
1068 | if (task_group_is_autogroup(task_group(p))) | |
1069 | return uc_req; | |
1070 | if (task_group(p) == &root_task_group) | |
1071 | return uc_req; | |
1072 | ||
1073 | uc_max = task_group(p)->uclamp[clamp_id]; | |
1074 | if (uc_req.value > uc_max.value || !uc_req.user_defined) | |
1075 | return uc_max; | |
1076 | #endif | |
1077 | ||
1078 | return uc_req; | |
1079 | } | |
1080 | ||
e8f14172 PB |
1081 | /* |
1082 | * The effective clamp bucket index of a task depends on, by increasing | |
1083 | * priority: | |
1084 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1085 | * - the task group effective clamp value, for tasks not either in the root |
1086 | * group or in an autogroup | |
e8f14172 PB |
1087 | * - the system default clamp value, defined by the sysadmin |
1088 | */ | |
1089 | static inline struct uclamp_se | |
0413d7f3 | 1090 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1091 | { |
3eac870a | 1092 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1093 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1094 | ||
1095 | /* System default restrictions always apply */ | |
1096 | if (unlikely(uc_req.value > uc_max.value)) | |
1097 | return uc_max; | |
1098 | ||
1099 | return uc_req; | |
1100 | } | |
1101 | ||
686516b5 | 1102 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1103 | { |
1104 | struct uclamp_se uc_eff; | |
1105 | ||
1106 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1107 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1108 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1109 | |
1110 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1111 | ||
686516b5 | 1112 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1113 | } |
1114 | ||
69842cba PB |
1115 | /* |
1116 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1117 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1118 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1119 | * |
1120 | * Tasks can have a task-specific value requested from user-space, track | |
1121 | * within each bucket the maximum value for tasks refcounted in it. | |
1122 | * This "local max aggregation" allows to track the exact "requested" value | |
1123 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1124 | */ |
1125 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1126 | enum uclamp_id clamp_id) |
69842cba PB |
1127 | { |
1128 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1129 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1130 | struct uclamp_bucket *bucket; | |
1131 | ||
1132 | lockdep_assert_held(&rq->lock); | |
1133 | ||
e8f14172 PB |
1134 | /* Update task effective clamp */ |
1135 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1136 | ||
69842cba PB |
1137 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1138 | bucket->tasks++; | |
e8f14172 | 1139 | uc_se->active = true; |
69842cba | 1140 | |
e496187d PB |
1141 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1142 | ||
60daf9c1 PB |
1143 | /* |
1144 | * Local max aggregation: rq buckets always track the max | |
1145 | * "requested" clamp value of its RUNNABLE tasks. | |
1146 | */ | |
1147 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1148 | bucket->value = uc_se->value; | |
1149 | ||
69842cba | 1150 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1151 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1152 | } |
1153 | ||
1154 | /* | |
1155 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1156 | * is released. If this is the last task reference counting the rq's max | |
1157 | * active clamp value, then the rq's clamp value is updated. | |
1158 | * | |
1159 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1160 | * always valid. If it's detected they are not, as defensive programming, | |
1161 | * enforce the expected state and warn. | |
1162 | */ | |
1163 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1164 | enum uclamp_id clamp_id) |
69842cba PB |
1165 | { |
1166 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1167 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1168 | struct uclamp_bucket *bucket; | |
e496187d | 1169 | unsigned int bkt_clamp; |
69842cba PB |
1170 | unsigned int rq_clamp; |
1171 | ||
1172 | lockdep_assert_held(&rq->lock); | |
1173 | ||
46609ce2 QY |
1174 | /* |
1175 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1176 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1177 | * | |
1178 | * In this case the uc_se->active flag should be false since no uclamp | |
1179 | * accounting was performed at enqueue time and we can just return | |
1180 | * here. | |
1181 | * | |
b19a888c | 1182 | * Need to be careful of the following enqueue/dequeue ordering |
46609ce2 QY |
1183 | * problem too |
1184 | * | |
1185 | * enqueue(taskA) | |
1186 | * // sched_uclamp_used gets enabled | |
1187 | * enqueue(taskB) | |
1188 | * dequeue(taskA) | |
b19a888c | 1189 | * // Must not decrement bucket->tasks here |
46609ce2 QY |
1190 | * dequeue(taskB) |
1191 | * | |
1192 | * where we could end up with stale data in uc_se and | |
1193 | * bucket[uc_se->bucket_id]. | |
1194 | * | |
1195 | * The following check here eliminates the possibility of such race. | |
1196 | */ | |
1197 | if (unlikely(!uc_se->active)) | |
1198 | return; | |
1199 | ||
69842cba | 1200 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1201 | |
69842cba PB |
1202 | SCHED_WARN_ON(!bucket->tasks); |
1203 | if (likely(bucket->tasks)) | |
1204 | bucket->tasks--; | |
46609ce2 | 1205 | |
e8f14172 | 1206 | uc_se->active = false; |
69842cba | 1207 | |
60daf9c1 PB |
1208 | /* |
1209 | * Keep "local max aggregation" simple and accept to (possibly) | |
1210 | * overboost some RUNNABLE tasks in the same bucket. | |
1211 | * The rq clamp bucket value is reset to its base value whenever | |
1212 | * there are no more RUNNABLE tasks refcounting it. | |
1213 | */ | |
69842cba PB |
1214 | if (likely(bucket->tasks)) |
1215 | return; | |
1216 | ||
1217 | rq_clamp = READ_ONCE(uc_rq->value); | |
1218 | /* | |
1219 | * Defensive programming: this should never happen. If it happens, | |
1220 | * e.g. due to future modification, warn and fixup the expected value. | |
1221 | */ | |
1222 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1223 | if (bucket->value >= rq_clamp) { |
1224 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1225 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1226 | } | |
69842cba PB |
1227 | } |
1228 | ||
1229 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1230 | { | |
0413d7f3 | 1231 | enum uclamp_id clamp_id; |
69842cba | 1232 | |
46609ce2 QY |
1233 | /* |
1234 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1235 | * | |
1236 | * The condition is constructed such that a NOP is generated when | |
1237 | * sched_uclamp_used is disabled. | |
1238 | */ | |
1239 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1240 | return; | |
1241 | ||
69842cba PB |
1242 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1243 | return; | |
1244 | ||
1245 | for_each_clamp_id(clamp_id) | |
1246 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1247 | |
1248 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1249 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1250 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1251 | } |
1252 | ||
1253 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1254 | { | |
0413d7f3 | 1255 | enum uclamp_id clamp_id; |
69842cba | 1256 | |
46609ce2 QY |
1257 | /* |
1258 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1259 | * | |
1260 | * The condition is constructed such that a NOP is generated when | |
1261 | * sched_uclamp_used is disabled. | |
1262 | */ | |
1263 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1264 | return; | |
1265 | ||
69842cba PB |
1266 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1267 | return; | |
1268 | ||
1269 | for_each_clamp_id(clamp_id) | |
1270 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1271 | } | |
1272 | ||
babbe170 | 1273 | static inline void |
0413d7f3 | 1274 | uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id) |
babbe170 PB |
1275 | { |
1276 | struct rq_flags rf; | |
1277 | struct rq *rq; | |
1278 | ||
1279 | /* | |
1280 | * Lock the task and the rq where the task is (or was) queued. | |
1281 | * | |
1282 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1283 | * price to pay to safely serialize util_{min,max} updates with | |
1284 | * enqueues, dequeues and migration operations. | |
1285 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1286 | */ | |
1287 | rq = task_rq_lock(p, &rf); | |
1288 | ||
1289 | /* | |
1290 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1291 | * If the task is not yet RUNNABLE and its task_struct is not | |
1292 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1293 | * it will already see the updated clamp bucket value. | |
1294 | */ | |
6e1ff077 | 1295 | if (p->uclamp[clamp_id].active) { |
babbe170 PB |
1296 | uclamp_rq_dec_id(rq, p, clamp_id); |
1297 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1298 | } | |
1299 | ||
1300 | task_rq_unlock(rq, p, &rf); | |
1301 | } | |
1302 | ||
e3b8b6a0 | 1303 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 PB |
1304 | static inline void |
1305 | uclamp_update_active_tasks(struct cgroup_subsys_state *css, | |
1306 | unsigned int clamps) | |
1307 | { | |
0413d7f3 | 1308 | enum uclamp_id clamp_id; |
babbe170 PB |
1309 | struct css_task_iter it; |
1310 | struct task_struct *p; | |
babbe170 PB |
1311 | |
1312 | css_task_iter_start(css, 0, &it); | |
1313 | while ((p = css_task_iter_next(&it))) { | |
1314 | for_each_clamp_id(clamp_id) { | |
1315 | if ((0x1 << clamp_id) & clamps) | |
1316 | uclamp_update_active(p, clamp_id); | |
1317 | } | |
1318 | } | |
1319 | css_task_iter_end(&it); | |
1320 | } | |
1321 | ||
7274a5c1 PB |
1322 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1323 | static void uclamp_update_root_tg(void) | |
1324 | { | |
1325 | struct task_group *tg = &root_task_group; | |
1326 | ||
1327 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1328 | sysctl_sched_uclamp_util_min, false); | |
1329 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1330 | sysctl_sched_uclamp_util_max, false); | |
1331 | ||
1332 | rcu_read_lock(); | |
1333 | cpu_util_update_eff(&root_task_group.css); | |
1334 | rcu_read_unlock(); | |
1335 | } | |
1336 | #else | |
1337 | static void uclamp_update_root_tg(void) { } | |
1338 | #endif | |
1339 | ||
e8f14172 | 1340 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1341 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1342 | { |
7274a5c1 | 1343 | bool update_root_tg = false; |
13685c4a | 1344 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1345 | int result; |
1346 | ||
2480c093 | 1347 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1348 | old_min = sysctl_sched_uclamp_util_min; |
1349 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1350 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1351 | |
1352 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1353 | if (result) | |
1354 | goto undo; | |
1355 | if (!write) | |
1356 | goto done; | |
1357 | ||
1358 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1359 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1360 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1361 | ||
e8f14172 PB |
1362 | result = -EINVAL; |
1363 | goto undo; | |
1364 | } | |
1365 | ||
1366 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1367 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1368 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1369 | update_root_tg = true; |
e8f14172 PB |
1370 | } |
1371 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1372 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1373 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1374 | update_root_tg = true; |
e8f14172 PB |
1375 | } |
1376 | ||
46609ce2 QY |
1377 | if (update_root_tg) { |
1378 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1379 | uclamp_update_root_tg(); |
46609ce2 | 1380 | } |
7274a5c1 | 1381 | |
13685c4a QY |
1382 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1383 | static_branch_enable(&sched_uclamp_used); | |
1384 | uclamp_sync_util_min_rt_default(); | |
1385 | } | |
7274a5c1 | 1386 | |
e8f14172 | 1387 | /* |
7274a5c1 PB |
1388 | * We update all RUNNABLE tasks only when task groups are in use. |
1389 | * Otherwise, keep it simple and do just a lazy update at each next | |
1390 | * task enqueue time. | |
e8f14172 | 1391 | */ |
7274a5c1 | 1392 | |
e8f14172 PB |
1393 | goto done; |
1394 | ||
1395 | undo: | |
1396 | sysctl_sched_uclamp_util_min = old_min; | |
1397 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1398 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1399 | done: |
2480c093 | 1400 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1401 | |
1402 | return result; | |
1403 | } | |
1404 | ||
a509a7cd PB |
1405 | static int uclamp_validate(struct task_struct *p, |
1406 | const struct sched_attr *attr) | |
1407 | { | |
480a6ca2 DE |
1408 | int util_min = p->uclamp_req[UCLAMP_MIN].value; |
1409 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd | 1410 | |
480a6ca2 DE |
1411 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { |
1412 | util_min = attr->sched_util_min; | |
a509a7cd | 1413 | |
480a6ca2 DE |
1414 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) |
1415 | return -EINVAL; | |
1416 | } | |
1417 | ||
1418 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1419 | util_max = attr->sched_util_max; | |
1420 | ||
1421 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | |
1422 | return -EINVAL; | |
1423 | } | |
1424 | ||
1425 | if (util_min != -1 && util_max != -1 && util_min > util_max) | |
a509a7cd PB |
1426 | return -EINVAL; |
1427 | ||
e65855a5 QY |
1428 | /* |
1429 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1430 | * | |
1431 | * We need to do that here, because enabling static branches is a | |
1432 | * blocking operation which obviously cannot be done while holding | |
1433 | * scheduler locks. | |
1434 | */ | |
1435 | static_branch_enable(&sched_uclamp_used); | |
1436 | ||
a509a7cd PB |
1437 | return 0; |
1438 | } | |
1439 | ||
480a6ca2 DE |
1440 | static bool uclamp_reset(const struct sched_attr *attr, |
1441 | enum uclamp_id clamp_id, | |
1442 | struct uclamp_se *uc_se) | |
1443 | { | |
1444 | /* Reset on sched class change for a non user-defined clamp value. */ | |
1445 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | |
1446 | !uc_se->user_defined) | |
1447 | return true; | |
1448 | ||
1449 | /* Reset on sched_util_{min,max} == -1. */ | |
1450 | if (clamp_id == UCLAMP_MIN && | |
1451 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | |
1452 | attr->sched_util_min == -1) { | |
1453 | return true; | |
1454 | } | |
1455 | ||
1456 | if (clamp_id == UCLAMP_MAX && | |
1457 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | |
1458 | attr->sched_util_max == -1) { | |
1459 | return true; | |
1460 | } | |
1461 | ||
1462 | return false; | |
1463 | } | |
1464 | ||
a509a7cd PB |
1465 | static void __setscheduler_uclamp(struct task_struct *p, |
1466 | const struct sched_attr *attr) | |
1467 | { | |
0413d7f3 | 1468 | enum uclamp_id clamp_id; |
1a00d999 | 1469 | |
1a00d999 PB |
1470 | for_each_clamp_id(clamp_id) { |
1471 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
480a6ca2 | 1472 | unsigned int value; |
1a00d999 | 1473 | |
480a6ca2 | 1474 | if (!uclamp_reset(attr, clamp_id, uc_se)) |
1a00d999 PB |
1475 | continue; |
1476 | ||
13685c4a QY |
1477 | /* |
1478 | * RT by default have a 100% boost value that could be modified | |
1479 | * at runtime. | |
1480 | */ | |
1a00d999 | 1481 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
480a6ca2 | 1482 | value = sysctl_sched_uclamp_util_min_rt_default; |
13685c4a | 1483 | else |
480a6ca2 DE |
1484 | value = uclamp_none(clamp_id); |
1485 | ||
1486 | uclamp_se_set(uc_se, value, false); | |
1a00d999 | 1487 | |
1a00d999 PB |
1488 | } |
1489 | ||
a509a7cd PB |
1490 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1491 | return; | |
1492 | ||
480a6ca2 DE |
1493 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && |
1494 | attr->sched_util_min != -1) { | |
a509a7cd PB |
1495 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], |
1496 | attr->sched_util_min, true); | |
1497 | } | |
1498 | ||
480a6ca2 DE |
1499 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && |
1500 | attr->sched_util_max != -1) { | |
a509a7cd PB |
1501 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], |
1502 | attr->sched_util_max, true); | |
1503 | } | |
1504 | } | |
1505 | ||
e8f14172 PB |
1506 | static void uclamp_fork(struct task_struct *p) |
1507 | { | |
0413d7f3 | 1508 | enum uclamp_id clamp_id; |
e8f14172 | 1509 | |
13685c4a QY |
1510 | /* |
1511 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1512 | * as the task is still at its early fork stages. | |
1513 | */ | |
e8f14172 PB |
1514 | for_each_clamp_id(clamp_id) |
1515 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1516 | |
1517 | if (likely(!p->sched_reset_on_fork)) | |
1518 | return; | |
1519 | ||
1520 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1521 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1522 | uclamp_none(clamp_id), false); | |
a87498ac | 1523 | } |
e8f14172 PB |
1524 | } |
1525 | ||
13685c4a QY |
1526 | static void uclamp_post_fork(struct task_struct *p) |
1527 | { | |
1528 | uclamp_update_util_min_rt_default(p); | |
1529 | } | |
1530 | ||
d81ae8aa QY |
1531 | static void __init init_uclamp_rq(struct rq *rq) |
1532 | { | |
1533 | enum uclamp_id clamp_id; | |
1534 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1535 | ||
1536 | for_each_clamp_id(clamp_id) { | |
1537 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1538 | .value = uclamp_none(clamp_id) | |
1539 | }; | |
1540 | } | |
1541 | ||
1542 | rq->uclamp_flags = 0; | |
1543 | } | |
1544 | ||
69842cba PB |
1545 | static void __init init_uclamp(void) |
1546 | { | |
e8f14172 | 1547 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1548 | enum uclamp_id clamp_id; |
69842cba PB |
1549 | int cpu; |
1550 | ||
d81ae8aa QY |
1551 | for_each_possible_cpu(cpu) |
1552 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1553 | |
69842cba | 1554 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1555 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1556 | uclamp_none(clamp_id), false); |
69842cba | 1557 | } |
e8f14172 PB |
1558 | |
1559 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1560 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1561 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1562 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1563 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1564 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1565 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1566 | #endif |
1567 | } | |
69842cba PB |
1568 | } |
1569 | ||
1570 | #else /* CONFIG_UCLAMP_TASK */ | |
1571 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1572 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1573 | static inline int uclamp_validate(struct task_struct *p, |
1574 | const struct sched_attr *attr) | |
1575 | { | |
1576 | return -EOPNOTSUPP; | |
1577 | } | |
1578 | static void __setscheduler_uclamp(struct task_struct *p, | |
1579 | const struct sched_attr *attr) { } | |
e8f14172 | 1580 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1581 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1582 | static inline void init_uclamp(void) { } |
1583 | #endif /* CONFIG_UCLAMP_TASK */ | |
1584 | ||
1de64443 | 1585 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1586 | { |
0a67d1ee PZ |
1587 | if (!(flags & ENQUEUE_NOCLOCK)) |
1588 | update_rq_clock(rq); | |
1589 | ||
eb414681 | 1590 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 1591 | sched_info_queued(rq, p); |
eb414681 JW |
1592 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1593 | } | |
0a67d1ee | 1594 | |
69842cba | 1595 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1596 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1597 | } |
1598 | ||
1de64443 | 1599 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1600 | { |
0a67d1ee PZ |
1601 | if (!(flags & DEQUEUE_NOCLOCK)) |
1602 | update_rq_clock(rq); | |
1603 | ||
eb414681 | 1604 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 1605 | sched_info_dequeued(rq, p); |
eb414681 JW |
1606 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1607 | } | |
0a67d1ee | 1608 | |
69842cba | 1609 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1610 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1611 | } |
1612 | ||
029632fb | 1613 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1614 | { |
371fd7e7 | 1615 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
1616 | |
1617 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
1618 | } |
1619 | ||
029632fb | 1620 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1621 | { |
7dd77884 PZ |
1622 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
1623 | ||
371fd7e7 | 1624 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1625 | } |
1626 | ||
14531189 | 1627 | /* |
dd41f596 | 1628 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1629 | */ |
14531189 IM |
1630 | static inline int __normal_prio(struct task_struct *p) |
1631 | { | |
dd41f596 | 1632 | return p->static_prio; |
14531189 IM |
1633 | } |
1634 | ||
b29739f9 IM |
1635 | /* |
1636 | * Calculate the expected normal priority: i.e. priority | |
1637 | * without taking RT-inheritance into account. Might be | |
1638 | * boosted by interactivity modifiers. Changes upon fork, | |
1639 | * setprio syscalls, and whenever the interactivity | |
1640 | * estimator recalculates. | |
1641 | */ | |
36c8b586 | 1642 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1643 | { |
1644 | int prio; | |
1645 | ||
aab03e05 DF |
1646 | if (task_has_dl_policy(p)) |
1647 | prio = MAX_DL_PRIO-1; | |
1648 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
1649 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1650 | else | |
1651 | prio = __normal_prio(p); | |
1652 | return prio; | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * Calculate the current priority, i.e. the priority | |
1657 | * taken into account by the scheduler. This value might | |
1658 | * be boosted by RT tasks, or might be boosted by | |
1659 | * interactivity modifiers. Will be RT if the task got | |
1660 | * RT-boosted. If not then it returns p->normal_prio. | |
1661 | */ | |
36c8b586 | 1662 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1663 | { |
1664 | p->normal_prio = normal_prio(p); | |
1665 | /* | |
1666 | * If we are RT tasks or we were boosted to RT priority, | |
1667 | * keep the priority unchanged. Otherwise, update priority | |
1668 | * to the normal priority: | |
1669 | */ | |
1670 | if (!rt_prio(p->prio)) | |
1671 | return p->normal_prio; | |
1672 | return p->prio; | |
1673 | } | |
1674 | ||
1da177e4 LT |
1675 | /** |
1676 | * task_curr - is this task currently executing on a CPU? | |
1677 | * @p: the task in question. | |
e69f6186 YB |
1678 | * |
1679 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1680 | */ |
36c8b586 | 1681 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1682 | { |
1683 | return cpu_curr(task_cpu(p)) == p; | |
1684 | } | |
1685 | ||
67dfa1b7 | 1686 | /* |
4c9a4bc8 PZ |
1687 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1688 | * use the balance_callback list if you want balancing. | |
1689 | * | |
1690 | * this means any call to check_class_changed() must be followed by a call to | |
1691 | * balance_callback(). | |
67dfa1b7 | 1692 | */ |
cb469845 SR |
1693 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1694 | const struct sched_class *prev_class, | |
da7a735e | 1695 | int oldprio) |
cb469845 SR |
1696 | { |
1697 | if (prev_class != p->sched_class) { | |
1698 | if (prev_class->switched_from) | |
da7a735e | 1699 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1700 | |
da7a735e | 1701 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1702 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1703 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1704 | } |
1705 | ||
029632fb | 1706 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 1707 | { |
aa93cd53 | 1708 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 1709 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
1710 | else if (p->sched_class > rq->curr->sched_class) |
1711 | resched_curr(rq); | |
1e5a7405 PZ |
1712 | |
1713 | /* | |
1714 | * A queue event has occurred, and we're going to schedule. In | |
1715 | * this case, we can save a useless back to back clock update. | |
1716 | */ | |
da0c1e65 | 1717 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 1718 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
1719 | } |
1720 | ||
1da177e4 | 1721 | #ifdef CONFIG_SMP |
175f0e25 | 1722 | |
af449901 PZ |
1723 | static void |
1724 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags); | |
1725 | ||
1726 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
1727 | const struct cpumask *new_mask, | |
1728 | u32 flags); | |
1729 | ||
1730 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
1731 | { | |
1732 | if (likely(!p->migration_disabled)) | |
1733 | return; | |
1734 | ||
1735 | if (p->cpus_ptr != &p->cpus_mask) | |
1736 | return; | |
1737 | ||
1738 | /* | |
1739 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
1740 | */ | |
1741 | __do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE); | |
1742 | } | |
1743 | ||
1744 | void migrate_disable(void) | |
1745 | { | |
3015ef4b TG |
1746 | struct task_struct *p = current; |
1747 | ||
1748 | if (p->migration_disabled) { | |
1749 | p->migration_disabled++; | |
af449901 | 1750 | return; |
3015ef4b | 1751 | } |
af449901 | 1752 | |
3015ef4b TG |
1753 | preempt_disable(); |
1754 | this_rq()->nr_pinned++; | |
1755 | p->migration_disabled = 1; | |
1756 | preempt_enable(); | |
af449901 PZ |
1757 | } |
1758 | EXPORT_SYMBOL_GPL(migrate_disable); | |
1759 | ||
1760 | void migrate_enable(void) | |
1761 | { | |
1762 | struct task_struct *p = current; | |
1763 | ||
6d337eab PZ |
1764 | if (p->migration_disabled > 1) { |
1765 | p->migration_disabled--; | |
af449901 | 1766 | return; |
6d337eab | 1767 | } |
af449901 | 1768 | |
6d337eab PZ |
1769 | /* |
1770 | * Ensure stop_task runs either before or after this, and that | |
1771 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
1772 | */ | |
1773 | preempt_disable(); | |
1774 | if (p->cpus_ptr != &p->cpus_mask) | |
1775 | __set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE); | |
1776 | /* | |
1777 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
1778 | * regular cpus_mask, otherwise things that race (eg. | |
1779 | * select_fallback_rq) get confused. | |
1780 | */ | |
af449901 | 1781 | barrier(); |
6d337eab | 1782 | p->migration_disabled = 0; |
3015ef4b | 1783 | this_rq()->nr_pinned--; |
6d337eab | 1784 | preempt_enable(); |
af449901 PZ |
1785 | } |
1786 | EXPORT_SYMBOL_GPL(migrate_enable); | |
1787 | ||
3015ef4b TG |
1788 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
1789 | { | |
1790 | return rq->nr_pinned; | |
1791 | } | |
1792 | ||
175f0e25 | 1793 | /* |
bee98539 | 1794 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
1795 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
1796 | */ | |
1797 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
1798 | { | |
3bd37062 | 1799 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
1800 | return false; |
1801 | ||
af449901 | 1802 | if (is_per_cpu_kthread(p) || is_migration_disabled(p)) |
175f0e25 PZ |
1803 | return cpu_online(cpu); |
1804 | ||
1805 | return cpu_active(cpu); | |
1806 | } | |
1807 | ||
5cc389bc PZ |
1808 | /* |
1809 | * This is how migration works: | |
1810 | * | |
1811 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1812 | * stop_one_cpu(). | |
1813 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1814 | * off the CPU) | |
1815 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1816 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1817 | * it and puts it into the right queue. | |
1818 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1819 | * is done. | |
1820 | */ | |
1821 | ||
1822 | /* | |
1823 | * move_queued_task - move a queued task to new rq. | |
1824 | * | |
1825 | * Returns (locked) new rq. Old rq's lock is released. | |
1826 | */ | |
8a8c69c3 PZ |
1827 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
1828 | struct task_struct *p, int new_cpu) | |
5cc389bc | 1829 | { |
5cc389bc PZ |
1830 | lockdep_assert_held(&rq->lock); |
1831 | ||
58877d34 | 1832 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 1833 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 1834 | rq_unlock(rq, rf); |
5cc389bc PZ |
1835 | |
1836 | rq = cpu_rq(new_cpu); | |
1837 | ||
8a8c69c3 | 1838 | rq_lock(rq, rf); |
5cc389bc | 1839 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 1840 | activate_task(rq, p, 0); |
5cc389bc PZ |
1841 | check_preempt_curr(rq, p, 0); |
1842 | ||
1843 | return rq; | |
1844 | } | |
1845 | ||
1846 | struct migration_arg { | |
6d337eab PZ |
1847 | struct task_struct *task; |
1848 | int dest_cpu; | |
1849 | struct set_affinity_pending *pending; | |
1850 | }; | |
1851 | ||
1852 | struct set_affinity_pending { | |
1853 | refcount_t refs; | |
1854 | struct completion done; | |
1855 | struct cpu_stop_work stop_work; | |
1856 | struct migration_arg arg; | |
5cc389bc PZ |
1857 | }; |
1858 | ||
1859 | /* | |
d1ccc66d | 1860 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
1861 | * this because either it can't run here any more (set_cpus_allowed() |
1862 | * away from this CPU, or CPU going down), or because we're | |
1863 | * attempting to rebalance this task on exec (sched_exec). | |
1864 | * | |
1865 | * So we race with normal scheduler movements, but that's OK, as long | |
1866 | * as the task is no longer on this CPU. | |
5cc389bc | 1867 | */ |
8a8c69c3 PZ |
1868 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
1869 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 1870 | { |
5cc389bc | 1871 | /* Affinity changed (again). */ |
175f0e25 | 1872 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 1873 | return rq; |
5cc389bc | 1874 | |
15ff991e | 1875 | update_rq_clock(rq); |
8a8c69c3 | 1876 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
1877 | |
1878 | return rq; | |
5cc389bc PZ |
1879 | } |
1880 | ||
1881 | /* | |
1882 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1883 | * and performs thread migration by bumping thread off CPU then | |
1884 | * 'pushing' onto another runqueue. | |
1885 | */ | |
1886 | static int migration_cpu_stop(void *data) | |
1887 | { | |
6d337eab | 1888 | struct set_affinity_pending *pending; |
5cc389bc | 1889 | struct migration_arg *arg = data; |
5e16bbc2 | 1890 | struct task_struct *p = arg->task; |
6d337eab | 1891 | int dest_cpu = arg->dest_cpu; |
5e16bbc2 | 1892 | struct rq *rq = this_rq(); |
6d337eab | 1893 | bool complete = false; |
8a8c69c3 | 1894 | struct rq_flags rf; |
5cc389bc PZ |
1895 | |
1896 | /* | |
d1ccc66d IM |
1897 | * The original target CPU might have gone down and we might |
1898 | * be on another CPU but it doesn't matter. | |
5cc389bc | 1899 | */ |
6d337eab | 1900 | local_irq_save(rf.flags); |
5cc389bc PZ |
1901 | /* |
1902 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 1903 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
1904 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
1905 | */ | |
a1488664 | 1906 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
1907 | |
1908 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1909 | rq_lock(rq, &rf); |
6d337eab PZ |
1910 | |
1911 | pending = p->migration_pending; | |
5e16bbc2 PZ |
1912 | /* |
1913 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1914 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1915 | * we're holding p->pi_lock. | |
1916 | */ | |
bf89a304 | 1917 | if (task_rq(p) == rq) { |
6d337eab PZ |
1918 | if (is_migration_disabled(p)) |
1919 | goto out; | |
1920 | ||
1921 | if (pending) { | |
1922 | p->migration_pending = NULL; | |
1923 | complete = true; | |
1924 | } | |
1925 | ||
1926 | /* migrate_enable() -- we must not race against SCA */ | |
1927 | if (dest_cpu < 0) { | |
1928 | /* | |
1929 | * When this was migrate_enable() but we no longer | |
1930 | * have a @pending, a concurrent SCA 'fixed' things | |
1931 | * and we should be valid again. Nothing to do. | |
1932 | */ | |
1933 | if (!pending) { | |
1293771e | 1934 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
1935 | goto out; |
1936 | } | |
1937 | ||
1938 | dest_cpu = cpumask_any_distribute(&p->cpus_mask); | |
1939 | } | |
1940 | ||
bf89a304 | 1941 | if (task_on_rq_queued(p)) |
6d337eab | 1942 | rq = __migrate_task(rq, &rf, p, dest_cpu); |
bf89a304 | 1943 | else |
6d337eab PZ |
1944 | p->wake_cpu = dest_cpu; |
1945 | ||
d707faa6 | 1946 | } else if (dest_cpu < 0 || pending) { |
6d337eab PZ |
1947 | /* |
1948 | * This happens when we get migrated between migrate_enable()'s | |
1949 | * preempt_enable() and scheduling the stopper task. At that | |
1950 | * point we're a regular task again and not current anymore. | |
1951 | * | |
1952 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
1953 | * more likely. | |
1954 | */ | |
1955 | ||
d707faa6 VS |
1956 | /* |
1957 | * The task moved before the stopper got to run. We're holding | |
1958 | * ->pi_lock, so the allowed mask is stable - if it got | |
1959 | * somewhere allowed, we're done. | |
1960 | */ | |
1961 | if (pending && cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { | |
1962 | p->migration_pending = NULL; | |
1963 | complete = true; | |
1964 | goto out; | |
1965 | } | |
1966 | ||
6d337eab PZ |
1967 | /* |
1968 | * When this was migrate_enable() but we no longer have an | |
1969 | * @pending, a concurrent SCA 'fixed' things and we should be | |
1970 | * valid again. Nothing to do. | |
1971 | */ | |
1972 | if (!pending) { | |
1293771e | 1973 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
1974 | goto out; |
1975 | } | |
1976 | ||
1977 | /* | |
1978 | * When migrate_enable() hits a rq mis-match we can't reliably | |
1979 | * determine is_migration_disabled() and so have to chase after | |
1980 | * it. | |
1981 | */ | |
1982 | task_rq_unlock(rq, p, &rf); | |
1983 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
1984 | &pending->arg, &pending->stop_work); | |
1985 | return 0; | |
bf89a304 | 1986 | } |
6d337eab PZ |
1987 | out: |
1988 | task_rq_unlock(rq, p, &rf); | |
1989 | ||
1990 | if (complete) | |
1991 | complete_all(&pending->done); | |
1992 | ||
1993 | /* For pending->{arg,stop_work} */ | |
1994 | pending = arg->pending; | |
1995 | if (pending && refcount_dec_and_test(&pending->refs)) | |
1996 | wake_up_var(&pending->refs); | |
5e16bbc2 | 1997 | |
5cc389bc PZ |
1998 | return 0; |
1999 | } | |
2000 | ||
a7c81556 PZ |
2001 | int push_cpu_stop(void *arg) |
2002 | { | |
2003 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2004 | struct task_struct *p = arg; | |
2005 | ||
2006 | raw_spin_lock_irq(&p->pi_lock); | |
2007 | raw_spin_lock(&rq->lock); | |
2008 | ||
2009 | if (task_rq(p) != rq) | |
2010 | goto out_unlock; | |
2011 | ||
2012 | if (is_migration_disabled(p)) { | |
2013 | p->migration_flags |= MDF_PUSH; | |
2014 | goto out_unlock; | |
2015 | } | |
2016 | ||
2017 | p->migration_flags &= ~MDF_PUSH; | |
2018 | ||
2019 | if (p->sched_class->find_lock_rq) | |
2020 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2021 | |
a7c81556 PZ |
2022 | if (!lowest_rq) |
2023 | goto out_unlock; | |
2024 | ||
2025 | // XXX validate p is still the highest prio task | |
2026 | if (task_rq(p) == rq) { | |
2027 | deactivate_task(rq, p, 0); | |
2028 | set_task_cpu(p, lowest_rq->cpu); | |
2029 | activate_task(lowest_rq, p, 0); | |
2030 | resched_curr(lowest_rq); | |
2031 | } | |
2032 | ||
2033 | double_unlock_balance(rq, lowest_rq); | |
2034 | ||
2035 | out_unlock: | |
2036 | rq->push_busy = false; | |
2037 | raw_spin_unlock(&rq->lock); | |
2038 | raw_spin_unlock_irq(&p->pi_lock); | |
2039 | ||
2040 | put_task_struct(p); | |
5cc389bc PZ |
2041 | return 0; |
2042 | } | |
2043 | ||
c5b28038 PZ |
2044 | /* |
2045 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2046 | * actually call this function. | |
2047 | */ | |
9cfc3e18 | 2048 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 2049 | { |
af449901 PZ |
2050 | if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2051 | p->cpus_ptr = new_mask; | |
2052 | return; | |
2053 | } | |
2054 | ||
3bd37062 | 2055 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
2056 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
2057 | } | |
2058 | ||
9cfc3e18 PZ |
2059 | static void |
2060 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 2061 | { |
6c37067e PZ |
2062 | struct rq *rq = task_rq(p); |
2063 | bool queued, running; | |
2064 | ||
af449901 PZ |
2065 | /* |
2066 | * This here violates the locking rules for affinity, since we're only | |
2067 | * supposed to change these variables while holding both rq->lock and | |
2068 | * p->pi_lock. | |
2069 | * | |
2070 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2071 | * accesses these variables under p->pi_lock and only does so after | |
2072 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2073 | * before finish_task(). | |
2074 | * | |
2075 | * XXX do further audits, this smells like something putrid. | |
2076 | */ | |
2077 | if (flags & SCA_MIGRATE_DISABLE) | |
2078 | SCHED_WARN_ON(!p->on_cpu); | |
2079 | else | |
2080 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2081 | |
2082 | queued = task_on_rq_queued(p); | |
2083 | running = task_current(rq, p); | |
2084 | ||
2085 | if (queued) { | |
2086 | /* | |
2087 | * Because __kthread_bind() calls this on blocked tasks without | |
2088 | * holding rq->lock. | |
2089 | */ | |
2090 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 2091 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2092 | } |
2093 | if (running) | |
2094 | put_prev_task(rq, p); | |
2095 | ||
9cfc3e18 | 2096 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 2097 | |
6c37067e | 2098 | if (queued) |
7134b3e9 | 2099 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2100 | if (running) |
03b7fad1 | 2101 | set_next_task(rq, p); |
c5b28038 PZ |
2102 | } |
2103 | ||
9cfc3e18 PZ |
2104 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2105 | { | |
2106 | __do_set_cpus_allowed(p, new_mask, 0); | |
2107 | } | |
2108 | ||
6d337eab | 2109 | /* |
c777d847 VS |
2110 | * This function is wildly self concurrent; here be dragons. |
2111 | * | |
2112 | * | |
2113 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2114 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2115 | * running, we have to kick it out using the CPU stopper. | |
2116 | * | |
2117 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2118 | * Consider: | |
2119 | * | |
2120 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2121 | * | |
2122 | * P0@CPU0 P1 | |
2123 | * | |
2124 | * migrate_disable(); | |
2125 | * <preempted> | |
2126 | * set_cpus_allowed_ptr(P0, [1]); | |
2127 | * | |
2128 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2129 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2130 | * This means we need the following scheme: | |
2131 | * | |
2132 | * P0@CPU0 P1 | |
2133 | * | |
2134 | * migrate_disable(); | |
2135 | * <preempted> | |
2136 | * set_cpus_allowed_ptr(P0, [1]); | |
2137 | * <blocks> | |
2138 | * <resumes> | |
2139 | * migrate_enable(); | |
2140 | * __set_cpus_allowed_ptr(); | |
2141 | * <wakes local stopper> | |
2142 | * `--> <woken on migration completion> | |
2143 | * | |
2144 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2145 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2146 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2147 | * should come into effect at the end of the Migrate-Disable region is the last | |
2148 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2149 | * but we still need to properly signal those waiting tasks at the appropriate | |
2150 | * moment. | |
2151 | * | |
2152 | * This is implemented using struct set_affinity_pending. The first | |
2153 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2154 | * setup an instance of that struct and install it on the targeted task_struct. | |
2155 | * Any and all further callers will reuse that instance. Those then wait for | |
2156 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2157 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2158 | * | |
2159 | * | |
2160 | * (1) In the cases covered above. There is one more where the completion is | |
2161 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
2162 | * cancels the need for an active migration. Consider: | |
2163 | * | |
2164 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2165 | * | |
2166 | * P0@CPU0 P1 P2 | |
2167 | * | |
2168 | * migrate_disable(); | |
2169 | * <preempted> | |
2170 | * set_cpus_allowed_ptr(P0, [1]); | |
2171 | * <blocks> | |
2172 | * set_cpus_allowed_ptr(P0, [0, 1]); | |
2173 | * <signal completion> | |
2174 | * <awakes> | |
2175 | * | |
2176 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2177 | * pending affinity completion is preceded by an uninstallation of | |
2178 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2179 | */ |
2180 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2181 | int dest_cpu, unsigned int flags) | |
2182 | { | |
2183 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
2184 | struct migration_arg arg = { | |
2185 | .task = p, | |
2186 | .dest_cpu = dest_cpu, | |
2187 | }; | |
2188 | bool complete = false; | |
2189 | ||
2190 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2191 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2192 | struct task_struct *push_task = NULL; |
2193 | ||
2194 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2195 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2196 | rq->push_busy = true; | |
2197 | push_task = get_task_struct(p); | |
2198 | } | |
2199 | ||
6d337eab PZ |
2200 | pending = p->migration_pending; |
2201 | if (pending) { | |
2202 | refcount_inc(&pending->refs); | |
2203 | p->migration_pending = NULL; | |
2204 | complete = true; | |
2205 | } | |
2206 | task_rq_unlock(rq, p, rf); | |
2207 | ||
a7c81556 PZ |
2208 | if (push_task) { |
2209 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2210 | p, &rq->push_work); | |
2211 | } | |
2212 | ||
6d337eab PZ |
2213 | if (complete) |
2214 | goto do_complete; | |
2215 | ||
2216 | return 0; | |
2217 | } | |
2218 | ||
2219 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2220 | /* serialized by p->pi_lock */ | |
2221 | if (!p->migration_pending) { | |
c777d847 | 2222 | /* Install the request */ |
6d337eab PZ |
2223 | refcount_set(&my_pending.refs, 1); |
2224 | init_completion(&my_pending.done); | |
2225 | p->migration_pending = &my_pending; | |
2226 | } else { | |
2227 | pending = p->migration_pending; | |
2228 | refcount_inc(&pending->refs); | |
2229 | } | |
2230 | } | |
2231 | pending = p->migration_pending; | |
2232 | /* | |
2233 | * - !MIGRATE_ENABLE: | |
2234 | * we'll have installed a pending if there wasn't one already. | |
2235 | * | |
2236 | * - MIGRATE_ENABLE: | |
2237 | * we're here because the current CPU isn't matching anymore, | |
2238 | * the only way that can happen is because of a concurrent | |
2239 | * set_cpus_allowed_ptr() call, which should then still be | |
2240 | * pending completion. | |
2241 | * | |
2242 | * Either way, we really should have a @pending here. | |
2243 | */ | |
2244 | if (WARN_ON_ONCE(!pending)) { | |
2245 | task_rq_unlock(rq, p, rf); | |
2246 | return -EINVAL; | |
2247 | } | |
2248 | ||
2249 | if (flags & SCA_MIGRATE_ENABLE) { | |
2250 | ||
2251 | refcount_inc(&pending->refs); /* pending->{arg,stop_work} */ | |
a7c81556 | 2252 | p->migration_flags &= ~MDF_PUSH; |
6d337eab PZ |
2253 | task_rq_unlock(rq, p, rf); |
2254 | ||
2255 | pending->arg = (struct migration_arg) { | |
2256 | .task = p, | |
2257 | .dest_cpu = -1, | |
2258 | .pending = pending, | |
2259 | }; | |
2260 | ||
2261 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2262 | &pending->arg, &pending->stop_work); | |
2263 | ||
2264 | return 0; | |
2265 | } | |
2266 | ||
2267 | if (task_running(rq, p) || p->state == TASK_WAKING) { | |
c777d847 VS |
2268 | /* |
2269 | * Lessen races (and headaches) by delegating | |
2270 | * is_migration_disabled(p) checks to the stopper, which will | |
2271 | * run on the same CPU as said p. | |
2272 | */ | |
6d337eab PZ |
2273 | task_rq_unlock(rq, p, rf); |
2274 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | |
2275 | ||
2276 | } else { | |
2277 | ||
2278 | if (!is_migration_disabled(p)) { | |
2279 | if (task_on_rq_queued(p)) | |
2280 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2281 | ||
2282 | p->migration_pending = NULL; | |
2283 | complete = true; | |
2284 | } | |
2285 | task_rq_unlock(rq, p, rf); | |
2286 | ||
2287 | do_complete: | |
2288 | if (complete) | |
2289 | complete_all(&pending->done); | |
2290 | } | |
2291 | ||
2292 | wait_for_completion(&pending->done); | |
2293 | ||
2294 | if (refcount_dec_and_test(&pending->refs)) | |
2295 | wake_up_var(&pending->refs); | |
2296 | ||
c777d847 VS |
2297 | /* |
2298 | * Block the original owner of &pending until all subsequent callers | |
2299 | * have seen the completion and decremented the refcount | |
2300 | */ | |
6d337eab PZ |
2301 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2302 | ||
2303 | return 0; | |
2304 | } | |
2305 | ||
5cc389bc PZ |
2306 | /* |
2307 | * Change a given task's CPU affinity. Migrate the thread to a | |
2308 | * proper CPU and schedule it away if the CPU it's executing on | |
2309 | * is removed from the allowed bitmask. | |
2310 | * | |
2311 | * NOTE: the caller must have a valid reference to the task, the | |
2312 | * task must not exit() & deallocate itself prematurely. The | |
2313 | * call is not atomic; no spinlocks may be held. | |
2314 | */ | |
25834c73 | 2315 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
9cfc3e18 PZ |
2316 | const struct cpumask *new_mask, |
2317 | u32 flags) | |
5cc389bc | 2318 | { |
e9d867a6 | 2319 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 2320 | unsigned int dest_cpu; |
eb580751 PZ |
2321 | struct rq_flags rf; |
2322 | struct rq *rq; | |
5cc389bc PZ |
2323 | int ret = 0; |
2324 | ||
eb580751 | 2325 | rq = task_rq_lock(p, &rf); |
a499c3ea | 2326 | update_rq_clock(rq); |
5cc389bc | 2327 | |
af449901 | 2328 | if (p->flags & PF_KTHREAD || is_migration_disabled(p)) { |
e9d867a6 | 2329 | /* |
af449901 PZ |
2330 | * Kernel threads are allowed on online && !active CPUs. |
2331 | * | |
2332 | * Specifically, migration_disabled() tasks must not fail the | |
2333 | * cpumask_any_and_distribute() pick below, esp. so on | |
2334 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2335 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2336 | */ |
2337 | cpu_valid_mask = cpu_online_mask; | |
2338 | } | |
2339 | ||
25834c73 PZ |
2340 | /* |
2341 | * Must re-check here, to close a race against __kthread_bind(), | |
2342 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2343 | */ | |
9cfc3e18 | 2344 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2345 | ret = -EINVAL; |
2346 | goto out; | |
2347 | } | |
2348 | ||
885b3ba4 VS |
2349 | if (!(flags & SCA_MIGRATE_ENABLE)) { |
2350 | if (cpumask_equal(&p->cpus_mask, new_mask)) | |
2351 | goto out; | |
2352 | ||
2353 | if (WARN_ON_ONCE(p == current && | |
2354 | is_migration_disabled(p) && | |
2355 | !cpumask_test_cpu(task_cpu(p), new_mask))) { | |
2356 | ret = -EBUSY; | |
2357 | goto out; | |
2358 | } | |
2359 | } | |
5cc389bc | 2360 | |
46a87b38 PT |
2361 | /* |
2362 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2363 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2364 | * immediately required to distribute the tasks within their new mask. | |
2365 | */ | |
2366 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 2367 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2368 | ret = -EINVAL; |
2369 | goto out; | |
2370 | } | |
2371 | ||
9cfc3e18 | 2372 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 2373 | |
e9d867a6 PZI |
2374 | if (p->flags & PF_KTHREAD) { |
2375 | /* | |
2376 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 2377 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
2378 | */ |
2379 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
2380 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
2381 | p->nr_cpus_allowed != 1); | |
2382 | } | |
2383 | ||
6d337eab | 2384 | return affine_move_task(rq, p, &rf, dest_cpu, flags); |
5cc389bc | 2385 | |
5cc389bc | 2386 | out: |
eb580751 | 2387 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
2388 | |
2389 | return ret; | |
2390 | } | |
25834c73 PZ |
2391 | |
2392 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
2393 | { | |
9cfc3e18 | 2394 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 2395 | } |
5cc389bc PZ |
2396 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2397 | ||
dd41f596 | 2398 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2399 | { |
e2912009 PZ |
2400 | #ifdef CONFIG_SCHED_DEBUG |
2401 | /* | |
2402 | * We should never call set_task_cpu() on a blocked task, | |
2403 | * ttwu() will sort out the placement. | |
2404 | */ | |
077614ee | 2405 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 2406 | !p->on_rq); |
0122ec5b | 2407 | |
3ea94de1 JP |
2408 | /* |
2409 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
2410 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
2411 | * time relying on p->on_rq. | |
2412 | */ | |
2413 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
2414 | p->sched_class == &fair_sched_class && | |
2415 | (p->on_rq && !task_on_rq_migrating(p))); | |
2416 | ||
0122ec5b | 2417 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
2418 | /* |
2419 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
2420 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
2421 | * | |
2422 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 2423 | * see task_group(). |
6c6c54e1 PZ |
2424 | * |
2425 | * Furthermore, all task_rq users should acquire both locks, see | |
2426 | * task_rq_lock(). | |
2427 | */ | |
0122ec5b PZ |
2428 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
2429 | lockdep_is_held(&task_rq(p)->lock))); | |
2430 | #endif | |
4ff9083b PZ |
2431 | /* |
2432 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
2433 | */ | |
2434 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
2435 | |
2436 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
2437 | #endif |
2438 | ||
de1d7286 | 2439 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2440 | |
0c69774e | 2441 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 2442 | if (p->sched_class->migrate_task_rq) |
1327237a | 2443 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 2444 | p->se.nr_migrations++; |
d7822b1e | 2445 | rseq_migrate(p); |
ff303e66 | 2446 | perf_event_task_migrate(p); |
0c69774e | 2447 | } |
dd41f596 IM |
2448 | |
2449 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2450 | } |
2451 | ||
0ad4e3df | 2452 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
2453 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
2454 | { | |
da0c1e65 | 2455 | if (task_on_rq_queued(p)) { |
ac66f547 | 2456 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 2457 | struct rq_flags srf, drf; |
ac66f547 PZ |
2458 | |
2459 | src_rq = task_rq(p); | |
2460 | dst_rq = cpu_rq(cpu); | |
2461 | ||
8a8c69c3 PZ |
2462 | rq_pin_lock(src_rq, &srf); |
2463 | rq_pin_lock(dst_rq, &drf); | |
2464 | ||
ac66f547 PZ |
2465 | deactivate_task(src_rq, p, 0); |
2466 | set_task_cpu(p, cpu); | |
2467 | activate_task(dst_rq, p, 0); | |
2468 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
2469 | |
2470 | rq_unpin_lock(dst_rq, &drf); | |
2471 | rq_unpin_lock(src_rq, &srf); | |
2472 | ||
ac66f547 PZ |
2473 | } else { |
2474 | /* | |
2475 | * Task isn't running anymore; make it appear like we migrated | |
2476 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 2477 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
2478 | */ |
2479 | p->wake_cpu = cpu; | |
2480 | } | |
2481 | } | |
2482 | ||
2483 | struct migration_swap_arg { | |
2484 | struct task_struct *src_task, *dst_task; | |
2485 | int src_cpu, dst_cpu; | |
2486 | }; | |
2487 | ||
2488 | static int migrate_swap_stop(void *data) | |
2489 | { | |
2490 | struct migration_swap_arg *arg = data; | |
2491 | struct rq *src_rq, *dst_rq; | |
2492 | int ret = -EAGAIN; | |
2493 | ||
62694cd5 PZ |
2494 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
2495 | return -EAGAIN; | |
2496 | ||
ac66f547 PZ |
2497 | src_rq = cpu_rq(arg->src_cpu); |
2498 | dst_rq = cpu_rq(arg->dst_cpu); | |
2499 | ||
74602315 PZ |
2500 | double_raw_lock(&arg->src_task->pi_lock, |
2501 | &arg->dst_task->pi_lock); | |
ac66f547 | 2502 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 2503 | |
ac66f547 PZ |
2504 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
2505 | goto unlock; | |
2506 | ||
2507 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
2508 | goto unlock; | |
2509 | ||
3bd37062 | 2510 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
2511 | goto unlock; |
2512 | ||
3bd37062 | 2513 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
2514 | goto unlock; |
2515 | ||
2516 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
2517 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
2518 | ||
2519 | ret = 0; | |
2520 | ||
2521 | unlock: | |
2522 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
2523 | raw_spin_unlock(&arg->dst_task->pi_lock); |
2524 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
2525 | |
2526 | return ret; | |
2527 | } | |
2528 | ||
2529 | /* | |
2530 | * Cross migrate two tasks | |
2531 | */ | |
0ad4e3df SD |
2532 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
2533 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
2534 | { |
2535 | struct migration_swap_arg arg; | |
2536 | int ret = -EINVAL; | |
2537 | ||
ac66f547 PZ |
2538 | arg = (struct migration_swap_arg){ |
2539 | .src_task = cur, | |
0ad4e3df | 2540 | .src_cpu = curr_cpu, |
ac66f547 | 2541 | .dst_task = p, |
0ad4e3df | 2542 | .dst_cpu = target_cpu, |
ac66f547 PZ |
2543 | }; |
2544 | ||
2545 | if (arg.src_cpu == arg.dst_cpu) | |
2546 | goto out; | |
2547 | ||
6acce3ef PZ |
2548 | /* |
2549 | * These three tests are all lockless; this is OK since all of them | |
2550 | * will be re-checked with proper locks held further down the line. | |
2551 | */ | |
ac66f547 PZ |
2552 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
2553 | goto out; | |
2554 | ||
3bd37062 | 2555 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
2556 | goto out; |
2557 | ||
3bd37062 | 2558 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
2559 | goto out; |
2560 | ||
286549dc | 2561 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
2562 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
2563 | ||
2564 | out: | |
ac66f547 PZ |
2565 | return ret; |
2566 | } | |
0ad4e3df | 2567 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 2568 | |
1da177e4 LT |
2569 | /* |
2570 | * wait_task_inactive - wait for a thread to unschedule. | |
2571 | * | |
85ba2d86 RM |
2572 | * If @match_state is nonzero, it's the @p->state value just checked and |
2573 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2574 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2575 | * we return a positive number (its total switch count). If a second call | |
2576 | * a short while later returns the same number, the caller can be sure that | |
2577 | * @p has remained unscheduled the whole time. | |
2578 | * | |
1da177e4 LT |
2579 | * The caller must ensure that the task *will* unschedule sometime soon, |
2580 | * else this function might spin for a *long* time. This function can't | |
2581 | * be called with interrupts off, or it may introduce deadlock with | |
2582 | * smp_call_function() if an IPI is sent by the same process we are | |
2583 | * waiting to become inactive. | |
2584 | */ | |
85ba2d86 | 2585 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 2586 | { |
da0c1e65 | 2587 | int running, queued; |
eb580751 | 2588 | struct rq_flags rf; |
85ba2d86 | 2589 | unsigned long ncsw; |
70b97a7f | 2590 | struct rq *rq; |
1da177e4 | 2591 | |
3a5c359a AK |
2592 | for (;;) { |
2593 | /* | |
2594 | * We do the initial early heuristics without holding | |
2595 | * any task-queue locks at all. We'll only try to get | |
2596 | * the runqueue lock when things look like they will | |
2597 | * work out! | |
2598 | */ | |
2599 | rq = task_rq(p); | |
fa490cfd | 2600 | |
3a5c359a AK |
2601 | /* |
2602 | * If the task is actively running on another CPU | |
2603 | * still, just relax and busy-wait without holding | |
2604 | * any locks. | |
2605 | * | |
2606 | * NOTE! Since we don't hold any locks, it's not | |
2607 | * even sure that "rq" stays as the right runqueue! | |
2608 | * But we don't care, since "task_running()" will | |
2609 | * return false if the runqueue has changed and p | |
2610 | * is actually now running somewhere else! | |
2611 | */ | |
85ba2d86 RM |
2612 | while (task_running(rq, p)) { |
2613 | if (match_state && unlikely(p->state != match_state)) | |
2614 | return 0; | |
3a5c359a | 2615 | cpu_relax(); |
85ba2d86 | 2616 | } |
fa490cfd | 2617 | |
3a5c359a AK |
2618 | /* |
2619 | * Ok, time to look more closely! We need the rq | |
2620 | * lock now, to be *sure*. If we're wrong, we'll | |
2621 | * just go back and repeat. | |
2622 | */ | |
eb580751 | 2623 | rq = task_rq_lock(p, &rf); |
27a9da65 | 2624 | trace_sched_wait_task(p); |
3a5c359a | 2625 | running = task_running(rq, p); |
da0c1e65 | 2626 | queued = task_on_rq_queued(p); |
85ba2d86 | 2627 | ncsw = 0; |
f31e11d8 | 2628 | if (!match_state || p->state == match_state) |
93dcf55f | 2629 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 2630 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 2631 | |
85ba2d86 RM |
2632 | /* |
2633 | * If it changed from the expected state, bail out now. | |
2634 | */ | |
2635 | if (unlikely(!ncsw)) | |
2636 | break; | |
2637 | ||
3a5c359a AK |
2638 | /* |
2639 | * Was it really running after all now that we | |
2640 | * checked with the proper locks actually held? | |
2641 | * | |
2642 | * Oops. Go back and try again.. | |
2643 | */ | |
2644 | if (unlikely(running)) { | |
2645 | cpu_relax(); | |
2646 | continue; | |
2647 | } | |
fa490cfd | 2648 | |
3a5c359a AK |
2649 | /* |
2650 | * It's not enough that it's not actively running, | |
2651 | * it must be off the runqueue _entirely_, and not | |
2652 | * preempted! | |
2653 | * | |
80dd99b3 | 2654 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2655 | * running right now), it's preempted, and we should |
2656 | * yield - it could be a while. | |
2657 | */ | |
da0c1e65 | 2658 | if (unlikely(queued)) { |
8b0e1953 | 2659 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
2660 | |
2661 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2662 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2663 | continue; |
2664 | } | |
fa490cfd | 2665 | |
3a5c359a AK |
2666 | /* |
2667 | * Ahh, all good. It wasn't running, and it wasn't | |
2668 | * runnable, which means that it will never become | |
2669 | * running in the future either. We're all done! | |
2670 | */ | |
2671 | break; | |
2672 | } | |
85ba2d86 RM |
2673 | |
2674 | return ncsw; | |
1da177e4 LT |
2675 | } |
2676 | ||
2677 | /*** | |
2678 | * kick_process - kick a running thread to enter/exit the kernel | |
2679 | * @p: the to-be-kicked thread | |
2680 | * | |
2681 | * Cause a process which is running on another CPU to enter | |
2682 | * kernel-mode, without any delay. (to get signals handled.) | |
2683 | * | |
25985edc | 2684 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2685 | * because all it wants to ensure is that the remote task enters |
2686 | * the kernel. If the IPI races and the task has been migrated | |
2687 | * to another CPU then no harm is done and the purpose has been | |
2688 | * achieved as well. | |
2689 | */ | |
36c8b586 | 2690 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2691 | { |
2692 | int cpu; | |
2693 | ||
2694 | preempt_disable(); | |
2695 | cpu = task_cpu(p); | |
2696 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2697 | smp_send_reschedule(cpu); | |
2698 | preempt_enable(); | |
2699 | } | |
b43e3521 | 2700 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 2701 | |
30da688e | 2702 | /* |
3bd37062 | 2703 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
2704 | * |
2705 | * A few notes on cpu_active vs cpu_online: | |
2706 | * | |
2707 | * - cpu_active must be a subset of cpu_online | |
2708 | * | |
97fb7a0a | 2709 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 2710 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 2711 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
2712 | * see it. |
2713 | * | |
d1ccc66d | 2714 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 2715 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 2716 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
2717 | * off. |
2718 | * | |
2719 | * This means that fallback selection must not select !active CPUs. | |
2720 | * And can assume that any active CPU must be online. Conversely | |
2721 | * select_task_rq() below may allow selection of !active CPUs in order | |
2722 | * to satisfy the above rules. | |
30da688e | 2723 | */ |
5da9a0fb PZ |
2724 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2725 | { | |
aa00d89c TC |
2726 | int nid = cpu_to_node(cpu); |
2727 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
2728 | enum { cpuset, possible, fail } state = cpuset; |
2729 | int dest_cpu; | |
5da9a0fb | 2730 | |
aa00d89c | 2731 | /* |
d1ccc66d IM |
2732 | * If the node that the CPU is on has been offlined, cpu_to_node() |
2733 | * will return -1. There is no CPU on the node, and we should | |
2734 | * select the CPU on the other node. | |
aa00d89c TC |
2735 | */ |
2736 | if (nid != -1) { | |
2737 | nodemask = cpumask_of_node(nid); | |
2738 | ||
2739 | /* Look for allowed, online CPU in same node. */ | |
2740 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
2741 | if (!cpu_active(dest_cpu)) |
2742 | continue; | |
3bd37062 | 2743 | if (cpumask_test_cpu(dest_cpu, p->cpus_ptr)) |
aa00d89c TC |
2744 | return dest_cpu; |
2745 | } | |
2baab4e9 | 2746 | } |
5da9a0fb | 2747 | |
2baab4e9 PZ |
2748 | for (;;) { |
2749 | /* Any allowed, online CPU? */ | |
3bd37062 | 2750 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 2751 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 2752 | continue; |
175f0e25 | 2753 | |
2baab4e9 PZ |
2754 | goto out; |
2755 | } | |
5da9a0fb | 2756 | |
e73e85f0 | 2757 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
2758 | switch (state) { |
2759 | case cpuset: | |
e73e85f0 ON |
2760 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
2761 | cpuset_cpus_allowed_fallback(p); | |
2762 | state = possible; | |
2763 | break; | |
2764 | } | |
df561f66 | 2765 | fallthrough; |
2baab4e9 | 2766 | case possible: |
af449901 PZ |
2767 | /* |
2768 | * XXX When called from select_task_rq() we only | |
2769 | * hold p->pi_lock and again violate locking order. | |
2770 | * | |
2771 | * More yuck to audit. | |
2772 | */ | |
2baab4e9 PZ |
2773 | do_set_cpus_allowed(p, cpu_possible_mask); |
2774 | state = fail; | |
2775 | break; | |
2776 | ||
2777 | case fail: | |
2778 | BUG(); | |
2779 | break; | |
2780 | } | |
2781 | } | |
2782 | ||
2783 | out: | |
2784 | if (state != cpuset) { | |
2785 | /* | |
2786 | * Don't tell them about moving exiting tasks or | |
2787 | * kernel threads (both mm NULL), since they never | |
2788 | * leave kernel. | |
2789 | */ | |
2790 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 2791 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
2792 | task_pid_nr(p), p->comm, cpu); |
2793 | } | |
5da9a0fb PZ |
2794 | } |
2795 | ||
2796 | return dest_cpu; | |
2797 | } | |
2798 | ||
e2912009 | 2799 | /* |
3bd37062 | 2800 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 2801 | */ |
970b13ba | 2802 | static inline |
3aef1551 | 2803 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 2804 | { |
cbce1a68 PZ |
2805 | lockdep_assert_held(&p->pi_lock); |
2806 | ||
af449901 | 2807 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 2808 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 2809 | else |
3bd37062 | 2810 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
2811 | |
2812 | /* | |
2813 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 2814 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 2815 | * CPU. |
e2912009 PZ |
2816 | * |
2817 | * Since this is common to all placement strategies, this lives here. | |
2818 | * | |
2819 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2820 | * not worry about this generic constraint ] | |
2821 | */ | |
7af443ee | 2822 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 2823 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2824 | |
2825 | return cpu; | |
970b13ba | 2826 | } |
09a40af5 | 2827 | |
f5832c19 NP |
2828 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2829 | { | |
ded467dc | 2830 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
2831 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
2832 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2833 | ||
2834 | if (stop) { | |
2835 | /* | |
2836 | * Make it appear like a SCHED_FIFO task, its something | |
2837 | * userspace knows about and won't get confused about. | |
2838 | * | |
2839 | * Also, it will make PI more or less work without too | |
2840 | * much confusion -- but then, stop work should not | |
2841 | * rely on PI working anyway. | |
2842 | */ | |
2843 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2844 | ||
2845 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
2846 | |
2847 | /* | |
2848 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
2849 | * adjust the effective priority of a task. As a result, | |
2850 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
2851 | * which can then trigger wakeups of the stop thread to push | |
2852 | * around the current task. | |
2853 | * | |
2854 | * The stop task itself will never be part of the PI-chain, it | |
2855 | * never blocks, therefore that ->pi_lock recursion is safe. | |
2856 | * Tell lockdep about this by placing the stop->pi_lock in its | |
2857 | * own class. | |
2858 | */ | |
2859 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
2860 | } |
2861 | ||
2862 | cpu_rq(cpu)->stop = stop; | |
2863 | ||
2864 | if (old_stop) { | |
2865 | /* | |
2866 | * Reset it back to a normal scheduling class so that | |
2867 | * it can die in pieces. | |
2868 | */ | |
2869 | old_stop->sched_class = &rt_sched_class; | |
2870 | } | |
2871 | } | |
2872 | ||
74d862b6 | 2873 | #else /* CONFIG_SMP */ |
25834c73 PZ |
2874 | |
2875 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
2876 | const struct cpumask *new_mask, |
2877 | u32 flags) | |
25834c73 PZ |
2878 | { |
2879 | return set_cpus_allowed_ptr(p, new_mask); | |
2880 | } | |
2881 | ||
af449901 PZ |
2882 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
2883 | ||
3015ef4b TG |
2884 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2885 | { | |
2886 | return false; | |
2887 | } | |
2888 | ||
74d862b6 | 2889 | #endif /* !CONFIG_SMP */ |
970b13ba | 2890 | |
d7c01d27 | 2891 | static void |
b84cb5df | 2892 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2893 | { |
4fa8d299 | 2894 | struct rq *rq; |
b84cb5df | 2895 | |
4fa8d299 JP |
2896 | if (!schedstat_enabled()) |
2897 | return; | |
2898 | ||
2899 | rq = this_rq(); | |
d7c01d27 | 2900 | |
4fa8d299 JP |
2901 | #ifdef CONFIG_SMP |
2902 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
2903 | __schedstat_inc(rq->ttwu_local); |
2904 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
2905 | } else { |
2906 | struct sched_domain *sd; | |
2907 | ||
b85c8b71 | 2908 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 2909 | rcu_read_lock(); |
4fa8d299 | 2910 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 2911 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 2912 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
2913 | break; |
2914 | } | |
2915 | } | |
057f3fad | 2916 | rcu_read_unlock(); |
d7c01d27 | 2917 | } |
f339b9dc PZ |
2918 | |
2919 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 2920 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
2921 | #endif /* CONFIG_SMP */ |
2922 | ||
b85c8b71 PZ |
2923 | __schedstat_inc(rq->ttwu_count); |
2924 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
2925 | |
2926 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 2927 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2928 | } |
2929 | ||
23f41eeb PZ |
2930 | /* |
2931 | * Mark the task runnable and perform wakeup-preemption. | |
2932 | */ | |
e7904a28 | 2933 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2934 | struct rq_flags *rf) |
9ed3811a | 2935 | { |
9ed3811a | 2936 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 2937 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
2938 | trace_sched_wakeup(p); |
2939 | ||
9ed3811a | 2940 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
2941 | if (p->sched_class->task_woken) { |
2942 | /* | |
b19a888c | 2943 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 2944 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 2945 | */ |
d8ac8971 | 2946 | rq_unpin_lock(rq, rf); |
9ed3811a | 2947 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2948 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 2949 | } |
9ed3811a | 2950 | |
e69c6341 | 2951 | if (rq->idle_stamp) { |
78becc27 | 2952 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 2953 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 2954 | |
abfafa54 JL |
2955 | update_avg(&rq->avg_idle, delta); |
2956 | ||
2957 | if (rq->avg_idle > max) | |
9ed3811a | 2958 | rq->avg_idle = max; |
abfafa54 | 2959 | |
9ed3811a TH |
2960 | rq->idle_stamp = 0; |
2961 | } | |
2962 | #endif | |
2963 | } | |
2964 | ||
c05fbafb | 2965 | static void |
e7904a28 | 2966 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2967 | struct rq_flags *rf) |
c05fbafb | 2968 | { |
77558e4d | 2969 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 2970 | |
cbce1a68 PZ |
2971 | lockdep_assert_held(&rq->lock); |
2972 | ||
c05fbafb PZ |
2973 | if (p->sched_contributes_to_load) |
2974 | rq->nr_uninterruptible--; | |
b5179ac7 | 2975 | |
dbfb089d | 2976 | #ifdef CONFIG_SMP |
b5179ac7 | 2977 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 2978 | en_flags |= ENQUEUE_MIGRATED; |
ec618b84 | 2979 | else |
c05fbafb | 2980 | #endif |
ec618b84 PZ |
2981 | if (p->in_iowait) { |
2982 | delayacct_blkio_end(p); | |
2983 | atomic_dec(&task_rq(p)->nr_iowait); | |
2984 | } | |
c05fbafb | 2985 | |
1b174a2c | 2986 | activate_task(rq, p, en_flags); |
d8ac8971 | 2987 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
2988 | } |
2989 | ||
2990 | /* | |
58877d34 PZ |
2991 | * Consider @p being inside a wait loop: |
2992 | * | |
2993 | * for (;;) { | |
2994 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
2995 | * | |
2996 | * if (CONDITION) | |
2997 | * break; | |
2998 | * | |
2999 | * schedule(); | |
3000 | * } | |
3001 | * __set_current_state(TASK_RUNNING); | |
3002 | * | |
3003 | * between set_current_state() and schedule(). In this case @p is still | |
3004 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3005 | * an atomic manner. | |
3006 | * | |
3007 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3008 | * then schedule() must still happen and p->state can be changed to | |
3009 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3010 | * need to do a full wakeup with enqueue. | |
3011 | * | |
3012 | * Returns: %true when the wakeup is done, | |
3013 | * %false otherwise. | |
c05fbafb | 3014 | */ |
58877d34 | 3015 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3016 | { |
eb580751 | 3017 | struct rq_flags rf; |
c05fbafb PZ |
3018 | struct rq *rq; |
3019 | int ret = 0; | |
3020 | ||
eb580751 | 3021 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3022 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3023 | /* check_preempt_curr() may use rq clock */ |
3024 | update_rq_clock(rq); | |
d8ac8971 | 3025 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3026 | ret = 1; |
3027 | } | |
eb580751 | 3028 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3029 | |
3030 | return ret; | |
3031 | } | |
3032 | ||
317f3941 | 3033 | #ifdef CONFIG_SMP |
a1488664 | 3034 | void sched_ttwu_pending(void *arg) |
317f3941 | 3035 | { |
a1488664 | 3036 | struct llist_node *llist = arg; |
317f3941 | 3037 | struct rq *rq = this_rq(); |
73215849 | 3038 | struct task_struct *p, *t; |
d8ac8971 | 3039 | struct rq_flags rf; |
317f3941 | 3040 | |
e3baac47 PZ |
3041 | if (!llist) |
3042 | return; | |
3043 | ||
126c2092 PZ |
3044 | /* |
3045 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
3046 | * Races such that false-negatives are possible, since they | |
3047 | * are shorter lived that false-positives would be. | |
3048 | */ | |
3049 | WRITE_ONCE(rq->ttwu_pending, 0); | |
3050 | ||
8a8c69c3 | 3051 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3052 | update_rq_clock(rq); |
317f3941 | 3053 | |
8c4890d1 | 3054 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3055 | if (WARN_ON_ONCE(p->on_cpu)) |
3056 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3057 | ||
3058 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3059 | set_task_cpu(p, cpu_of(rq)); | |
3060 | ||
73215849 | 3061 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3062 | } |
317f3941 | 3063 | |
8a8c69c3 | 3064 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3065 | } |
3066 | ||
b2a02fc4 | 3067 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3068 | { |
b2a02fc4 | 3069 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3070 | |
b2a02fc4 PZ |
3071 | if (!set_nr_if_polling(rq->idle)) |
3072 | arch_send_call_function_single_ipi(cpu); | |
3073 | else | |
3074 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3075 | } |
3076 | ||
2ebb1771 MG |
3077 | /* |
3078 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3079 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3080 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3081 | * of the wakeup instead of the waker. | |
3082 | */ | |
3083 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3084 | { |
e3baac47 PZ |
3085 | struct rq *rq = cpu_rq(cpu); |
3086 | ||
b7e7ade3 PZ |
3087 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3088 | ||
126c2092 | 3089 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3090 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3091 | } |
d6aa8f85 | 3092 | |
f6be8af1 CL |
3093 | void wake_up_if_idle(int cpu) |
3094 | { | |
3095 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3096 | struct rq_flags rf; |
f6be8af1 | 3097 | |
fd7de1e8 AL |
3098 | rcu_read_lock(); |
3099 | ||
3100 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3101 | goto out; | |
f6be8af1 CL |
3102 | |
3103 | if (set_nr_if_polling(rq->idle)) { | |
3104 | trace_sched_wake_idle_without_ipi(cpu); | |
3105 | } else { | |
8a8c69c3 | 3106 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
3107 | if (is_idle_task(rq->curr)) |
3108 | smp_send_reschedule(cpu); | |
d1ccc66d | 3109 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 3110 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 3111 | } |
fd7de1e8 AL |
3112 | |
3113 | out: | |
3114 | rcu_read_unlock(); | |
f6be8af1 CL |
3115 | } |
3116 | ||
39be3501 | 3117 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
3118 | { |
3119 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
3120 | } | |
c6e7bd7a | 3121 | |
2ebb1771 MG |
3122 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
3123 | { | |
3124 | /* | |
3125 | * If the CPU does not share cache, then queue the task on the | |
3126 | * remote rqs wakelist to avoid accessing remote data. | |
3127 | */ | |
3128 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3129 | return true; | |
3130 | ||
3131 | /* | |
3132 | * If the task is descheduling and the only running task on the | |
3133 | * CPU then use the wakelist to offload the task activation to | |
3134 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
3135 | * nr_running is checked to avoid unnecessary task stacking. | |
3136 | */ | |
739f70b4 | 3137 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
3138 | return true; |
3139 | ||
3140 | return false; | |
3141 | } | |
3142 | ||
3143 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3144 | { |
2ebb1771 | 3145 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
3146 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
3147 | return false; | |
3148 | ||
c6e7bd7a | 3149 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3150 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3151 | return true; |
3152 | } | |
3153 | ||
3154 | return false; | |
3155 | } | |
58877d34 PZ |
3156 | |
3157 | #else /* !CONFIG_SMP */ | |
3158 | ||
3159 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3160 | { | |
3161 | return false; | |
3162 | } | |
3163 | ||
d6aa8f85 | 3164 | #endif /* CONFIG_SMP */ |
317f3941 | 3165 | |
b5179ac7 | 3166 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3167 | { |
3168 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3169 | struct rq_flags rf; |
c05fbafb | 3170 | |
2ebb1771 | 3171 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3172 | return; |
317f3941 | 3173 | |
8a8c69c3 | 3174 | rq_lock(rq, &rf); |
77558e4d | 3175 | update_rq_clock(rq); |
d8ac8971 | 3176 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3177 | rq_unlock(rq, &rf); |
9ed3811a TH |
3178 | } |
3179 | ||
8643cda5 PZ |
3180 | /* |
3181 | * Notes on Program-Order guarantees on SMP systems. | |
3182 | * | |
3183 | * MIGRATION | |
3184 | * | |
3185 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3186 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3187 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3188 | * |
3189 | * For migration (of runnable tasks) this is provided by the following means: | |
3190 | * | |
3191 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3192 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3193 | * rq(c1)->lock (if not at the same time, then in that order). | |
3194 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3195 | * | |
7696f991 | 3196 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3197 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3198 | * |
3199 | * Example: | |
3200 | * | |
3201 | * CPU0 CPU1 CPU2 | |
3202 | * | |
3203 | * LOCK rq(0)->lock | |
3204 | * sched-out X | |
3205 | * sched-in Y | |
3206 | * UNLOCK rq(0)->lock | |
3207 | * | |
3208 | * LOCK rq(0)->lock // orders against CPU0 | |
3209 | * dequeue X | |
3210 | * UNLOCK rq(0)->lock | |
3211 | * | |
3212 | * LOCK rq(1)->lock | |
3213 | * enqueue X | |
3214 | * UNLOCK rq(1)->lock | |
3215 | * | |
3216 | * LOCK rq(1)->lock // orders against CPU2 | |
3217 | * sched-out Z | |
3218 | * sched-in X | |
3219 | * UNLOCK rq(1)->lock | |
3220 | * | |
3221 | * | |
3222 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3223 | * | |
3224 | * For blocking we (obviously) need to provide the same guarantee as for | |
3225 | * migration. However the means are completely different as there is no lock | |
3226 | * chain to provide order. Instead we do: | |
3227 | * | |
58877d34 PZ |
3228 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3229 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3230 | * |
3231 | * Example: | |
3232 | * | |
3233 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
3234 | * | |
3235 | * LOCK rq(0)->lock LOCK X->pi_lock | |
3236 | * dequeue X | |
3237 | * sched-out X | |
3238 | * smp_store_release(X->on_cpu, 0); | |
3239 | * | |
1f03e8d2 | 3240 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
3241 | * X->state = WAKING |
3242 | * set_task_cpu(X,2) | |
3243 | * | |
3244 | * LOCK rq(2)->lock | |
3245 | * enqueue X | |
3246 | * X->state = RUNNING | |
3247 | * UNLOCK rq(2)->lock | |
3248 | * | |
3249 | * LOCK rq(2)->lock // orders against CPU1 | |
3250 | * sched-out Z | |
3251 | * sched-in X | |
3252 | * UNLOCK rq(2)->lock | |
3253 | * | |
3254 | * UNLOCK X->pi_lock | |
3255 | * UNLOCK rq(0)->lock | |
3256 | * | |
3257 | * | |
7696f991 AP |
3258 | * However, for wakeups there is a second guarantee we must provide, namely we |
3259 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
3260 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
3261 | */ |
3262 | ||
9ed3811a | 3263 | /** |
1da177e4 | 3264 | * try_to_wake_up - wake up a thread |
9ed3811a | 3265 | * @p: the thread to be awakened |
1da177e4 | 3266 | * @state: the mask of task states that can be woken |
9ed3811a | 3267 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 3268 | * |
58877d34 PZ |
3269 | * Conceptually does: |
3270 | * | |
3271 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 3272 | * |
a2250238 PZ |
3273 | * If the task was not queued/runnable, also place it back on a runqueue. |
3274 | * | |
58877d34 PZ |
3275 | * This function is atomic against schedule() which would dequeue the task. |
3276 | * | |
3277 | * It issues a full memory barrier before accessing @p->state, see the comment | |
3278 | * with set_current_state(). | |
a2250238 | 3279 | * |
58877d34 | 3280 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 3281 | * |
58877d34 PZ |
3282 | * Relies on p->pi_lock stabilizing: |
3283 | * - p->sched_class | |
3284 | * - p->cpus_ptr | |
3285 | * - p->sched_task_group | |
3286 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
3287 | * | |
3288 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
3289 | * Takes rq->lock in: | |
3290 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
3291 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
3292 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
3293 | * | |
3294 | * As a consequence we race really badly with just about everything. See the | |
3295 | * many memory barriers and their comments for details. | |
7696f991 | 3296 | * |
a2250238 PZ |
3297 | * Return: %true if @p->state changes (an actual wakeup was done), |
3298 | * %false otherwise. | |
1da177e4 | 3299 | */ |
e4a52bcb PZ |
3300 | static int |
3301 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 3302 | { |
1da177e4 | 3303 | unsigned long flags; |
c05fbafb | 3304 | int cpu, success = 0; |
2398f2c6 | 3305 | |
e3d85487 | 3306 | preempt_disable(); |
aacedf26 PZ |
3307 | if (p == current) { |
3308 | /* | |
3309 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
3310 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 3311 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
3312 | * without taking any locks. |
3313 | * | |
3314 | * In particular: | |
3315 | * - we rely on Program-Order guarantees for all the ordering, | |
3316 | * - we're serialized against set_special_state() by virtue of | |
3317 | * it disabling IRQs (this allows not taking ->pi_lock). | |
3318 | */ | |
3319 | if (!(p->state & state)) | |
e3d85487 | 3320 | goto out; |
aacedf26 PZ |
3321 | |
3322 | success = 1; | |
aacedf26 PZ |
3323 | trace_sched_waking(p); |
3324 | p->state = TASK_RUNNING; | |
3325 | trace_sched_wakeup(p); | |
3326 | goto out; | |
3327 | } | |
3328 | ||
e0acd0a6 ON |
3329 | /* |
3330 | * If we are going to wake up a thread waiting for CONDITION we | |
3331 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
3332 | * reordered with p->state check below. This pairs with smp_store_mb() |
3333 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 3334 | */ |
013fdb80 | 3335 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 3336 | smp_mb__after_spinlock(); |
e9c84311 | 3337 | if (!(p->state & state)) |
aacedf26 | 3338 | goto unlock; |
1da177e4 | 3339 | |
fbd705a0 PZ |
3340 | trace_sched_waking(p); |
3341 | ||
d1ccc66d IM |
3342 | /* We're going to change ->state: */ |
3343 | success = 1; | |
1da177e4 | 3344 | |
135e8c92 BS |
3345 | /* |
3346 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
3347 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
3348 | * in smp_cond_load_acquire() below. | |
3349 | * | |
3d85b270 AP |
3350 | * sched_ttwu_pending() try_to_wake_up() |
3351 | * STORE p->on_rq = 1 LOAD p->state | |
3352 | * UNLOCK rq->lock | |
3353 | * | |
3354 | * __schedule() (switch to task 'p') | |
3355 | * LOCK rq->lock smp_rmb(); | |
3356 | * smp_mb__after_spinlock(); | |
3357 | * UNLOCK rq->lock | |
135e8c92 BS |
3358 | * |
3359 | * [task p] | |
3d85b270 | 3360 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 3361 | * |
3d85b270 AP |
3362 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3363 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
3364 | * |
3365 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
3366 | */ |
3367 | smp_rmb(); | |
58877d34 | 3368 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 3369 | goto unlock; |
1da177e4 | 3370 | |
1da177e4 | 3371 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
3372 | /* |
3373 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
3374 | * possible to, falsely, observe p->on_cpu == 0. | |
3375 | * | |
3376 | * One must be running (->on_cpu == 1) in order to remove oneself | |
3377 | * from the runqueue. | |
3378 | * | |
3d85b270 AP |
3379 | * __schedule() (switch to task 'p') try_to_wake_up() |
3380 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
3381 | * UNLOCK rq->lock | |
3382 | * | |
3383 | * __schedule() (put 'p' to sleep) | |
3384 | * LOCK rq->lock smp_rmb(); | |
3385 | * smp_mb__after_spinlock(); | |
3386 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 3387 | * |
3d85b270 AP |
3388 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3389 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
3390 | * |
3391 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
3392 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
3393 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 3394 | */ |
dbfb089d PZ |
3395 | smp_acquire__after_ctrl_dep(); |
3396 | ||
3397 | /* | |
3398 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
3399 | * == 0), which means we need to do an enqueue, change p->state to | |
3400 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
3401 | * enqueue, such as ttwu_queue_wakelist(). | |
3402 | */ | |
3403 | p->state = TASK_WAKING; | |
ecf7d01c | 3404 | |
c6e7bd7a PZ |
3405 | /* |
3406 | * If the owning (remote) CPU is still in the middle of schedule() with | |
3407 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
3408 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
3409 | * let the waker make forward progress. This is safe because IRQs are | |
3410 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
3411 | * |
3412 | * Ensure we load task_cpu(p) after p->on_cpu: | |
3413 | * | |
3414 | * set_task_cpu(p, cpu); | |
3415 | * STORE p->cpu = @cpu | |
3416 | * __schedule() (switch to task 'p') | |
3417 | * LOCK rq->lock | |
3418 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
3419 | * STORE p->on_cpu = 1 LOAD p->cpu | |
3420 | * | |
3421 | * to ensure we observe the correct CPU on which the task is currently | |
3422 | * scheduling. | |
c6e7bd7a | 3423 | */ |
b6e13e85 | 3424 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 3425 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
3426 | goto unlock; |
3427 | ||
e9c84311 | 3428 | /* |
d1ccc66d | 3429 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 3430 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 3431 | * |
31cb1bc0 | 3432 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
3433 | * |
3434 | * This ensures that tasks getting woken will be fully ordered against | |
3435 | * their previous state and preserve Program Order. | |
0970d299 | 3436 | */ |
1f03e8d2 | 3437 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 3438 | |
3aef1551 | 3439 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc | 3440 | if (task_cpu(p) != cpu) { |
ec618b84 PZ |
3441 | if (p->in_iowait) { |
3442 | delayacct_blkio_end(p); | |
3443 | atomic_dec(&task_rq(p)->nr_iowait); | |
3444 | } | |
3445 | ||
f339b9dc | 3446 | wake_flags |= WF_MIGRATED; |
eb414681 | 3447 | psi_ttwu_dequeue(p); |
e4a52bcb | 3448 | set_task_cpu(p, cpu); |
f339b9dc | 3449 | } |
b6e13e85 PZ |
3450 | #else |
3451 | cpu = task_cpu(p); | |
1da177e4 | 3452 | #endif /* CONFIG_SMP */ |
1da177e4 | 3453 | |
b5179ac7 | 3454 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 3455 | unlock: |
013fdb80 | 3456 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
3457 | out: |
3458 | if (success) | |
b6e13e85 | 3459 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 3460 | preempt_enable(); |
1da177e4 LT |
3461 | |
3462 | return success; | |
3463 | } | |
3464 | ||
2beaf328 PM |
3465 | /** |
3466 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
3467 | * @p: Process for which the function is to be invoked. | |
3468 | * @func: Function to invoke. | |
3469 | * @arg: Argument to function. | |
3470 | * | |
3471 | * If the specified task can be quickly locked into a definite state | |
3472 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
3473 | * state while invoking @func(@arg). This function can use ->on_rq and | |
3474 | * task_curr() to work out what the state is, if required. Given that | |
3475 | * @func can be invoked with a runqueue lock held, it had better be quite | |
3476 | * lightweight. | |
3477 | * | |
3478 | * Returns: | |
3479 | * @false if the task slipped out from under the locks. | |
3480 | * @true if the task was locked onto a runqueue or is sleeping. | |
3481 | * However, @func can override this by returning @false. | |
3482 | */ | |
3483 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
3484 | { | |
3485 | bool ret = false; | |
3486 | struct rq_flags rf; | |
3487 | struct rq *rq; | |
3488 | ||
3489 | lockdep_assert_irqs_enabled(); | |
3490 | raw_spin_lock_irq(&p->pi_lock); | |
3491 | if (p->on_rq) { | |
3492 | rq = __task_rq_lock(p, &rf); | |
3493 | if (task_rq(p) == rq) | |
3494 | ret = func(p, arg); | |
3495 | rq_unlock(rq, &rf); | |
3496 | } else { | |
3497 | switch (p->state) { | |
3498 | case TASK_RUNNING: | |
3499 | case TASK_WAKING: | |
3500 | break; | |
3501 | default: | |
3502 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
3503 | if (!p->on_rq) | |
3504 | ret = func(p, arg); | |
3505 | } | |
3506 | } | |
3507 | raw_spin_unlock_irq(&p->pi_lock); | |
3508 | return ret; | |
3509 | } | |
3510 | ||
50fa610a DH |
3511 | /** |
3512 | * wake_up_process - Wake up a specific process | |
3513 | * @p: The process to be woken up. | |
3514 | * | |
3515 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
3516 | * processes. |
3517 | * | |
3518 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 3519 | * |
7696f991 | 3520 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 3521 | */ |
7ad5b3a5 | 3522 | int wake_up_process(struct task_struct *p) |
1da177e4 | 3523 | { |
9067ac85 | 3524 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 3525 | } |
1da177e4 LT |
3526 | EXPORT_SYMBOL(wake_up_process); |
3527 | ||
7ad5b3a5 | 3528 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
3529 | { |
3530 | return try_to_wake_up(p, state, 0); | |
3531 | } | |
3532 | ||
1da177e4 LT |
3533 | /* |
3534 | * Perform scheduler related setup for a newly forked process p. | |
3535 | * p is forked by current. | |
dd41f596 IM |
3536 | * |
3537 | * __sched_fork() is basic setup used by init_idle() too: | |
3538 | */ | |
5e1576ed | 3539 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3540 | { |
fd2f4419 PZ |
3541 | p->on_rq = 0; |
3542 | ||
3543 | p->se.on_rq = 0; | |
dd41f596 IM |
3544 | p->se.exec_start = 0; |
3545 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 3546 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 3547 | p->se.nr_migrations = 0; |
da7a735e | 3548 | p->se.vruntime = 0; |
fd2f4419 | 3549 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 3550 | |
ad936d86 BP |
3551 | #ifdef CONFIG_FAIR_GROUP_SCHED |
3552 | p->se.cfs_rq = NULL; | |
3553 | #endif | |
3554 | ||
6cfb0d5d | 3555 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 3556 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 3557 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 3558 | #endif |
476d139c | 3559 | |
aab03e05 | 3560 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 3561 | init_dl_task_timer(&p->dl); |
209a0cbd | 3562 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 3563 | __dl_clear_params(p); |
aab03e05 | 3564 | |
fa717060 | 3565 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
3566 | p->rt.timeout = 0; |
3567 | p->rt.time_slice = sched_rr_timeslice; | |
3568 | p->rt.on_rq = 0; | |
3569 | p->rt.on_list = 0; | |
476d139c | 3570 | |
e107be36 AK |
3571 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3572 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
3573 | #endif | |
cbee9f88 | 3574 | |
5e1f0f09 MG |
3575 | #ifdef CONFIG_COMPACTION |
3576 | p->capture_control = NULL; | |
3577 | #endif | |
13784475 | 3578 | init_numa_balancing(clone_flags, p); |
a1488664 | 3579 | #ifdef CONFIG_SMP |
8c4890d1 | 3580 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 3581 | p->migration_pending = NULL; |
a1488664 | 3582 | #endif |
dd41f596 IM |
3583 | } |
3584 | ||
2a595721 SD |
3585 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
3586 | ||
1a687c2e | 3587 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 3588 | |
1a687c2e MG |
3589 | void set_numabalancing_state(bool enabled) |
3590 | { | |
3591 | if (enabled) | |
2a595721 | 3592 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 3593 | else |
2a595721 | 3594 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 3595 | } |
54a43d54 AK |
3596 | |
3597 | #ifdef CONFIG_PROC_SYSCTL | |
3598 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 3599 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
3600 | { |
3601 | struct ctl_table t; | |
3602 | int err; | |
2a595721 | 3603 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
3604 | |
3605 | if (write && !capable(CAP_SYS_ADMIN)) | |
3606 | return -EPERM; | |
3607 | ||
3608 | t = *table; | |
3609 | t.data = &state; | |
3610 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3611 | if (err < 0) | |
3612 | return err; | |
3613 | if (write) | |
3614 | set_numabalancing_state(state); | |
3615 | return err; | |
3616 | } | |
3617 | #endif | |
3618 | #endif | |
dd41f596 | 3619 | |
4698f88c JP |
3620 | #ifdef CONFIG_SCHEDSTATS |
3621 | ||
cb251765 | 3622 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 3623 | static bool __initdata __sched_schedstats = false; |
cb251765 | 3624 | |
cb251765 MG |
3625 | static void set_schedstats(bool enabled) |
3626 | { | |
3627 | if (enabled) | |
3628 | static_branch_enable(&sched_schedstats); | |
3629 | else | |
3630 | static_branch_disable(&sched_schedstats); | |
3631 | } | |
3632 | ||
3633 | void force_schedstat_enabled(void) | |
3634 | { | |
3635 | if (!schedstat_enabled()) { | |
3636 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
3637 | static_branch_enable(&sched_schedstats); | |
3638 | } | |
3639 | } | |
3640 | ||
3641 | static int __init setup_schedstats(char *str) | |
3642 | { | |
3643 | int ret = 0; | |
3644 | if (!str) | |
3645 | goto out; | |
3646 | ||
4698f88c JP |
3647 | /* |
3648 | * This code is called before jump labels have been set up, so we can't | |
3649 | * change the static branch directly just yet. Instead set a temporary | |
3650 | * variable so init_schedstats() can do it later. | |
3651 | */ | |
cb251765 | 3652 | if (!strcmp(str, "enable")) { |
4698f88c | 3653 | __sched_schedstats = true; |
cb251765 MG |
3654 | ret = 1; |
3655 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 3656 | __sched_schedstats = false; |
cb251765 MG |
3657 | ret = 1; |
3658 | } | |
3659 | out: | |
3660 | if (!ret) | |
3661 | pr_warn("Unable to parse schedstats=\n"); | |
3662 | ||
3663 | return ret; | |
3664 | } | |
3665 | __setup("schedstats=", setup_schedstats); | |
3666 | ||
4698f88c JP |
3667 | static void __init init_schedstats(void) |
3668 | { | |
3669 | set_schedstats(__sched_schedstats); | |
3670 | } | |
3671 | ||
cb251765 | 3672 | #ifdef CONFIG_PROC_SYSCTL |
32927393 CH |
3673 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
3674 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
3675 | { |
3676 | struct ctl_table t; | |
3677 | int err; | |
3678 | int state = static_branch_likely(&sched_schedstats); | |
3679 | ||
3680 | if (write && !capable(CAP_SYS_ADMIN)) | |
3681 | return -EPERM; | |
3682 | ||
3683 | t = *table; | |
3684 | t.data = &state; | |
3685 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3686 | if (err < 0) | |
3687 | return err; | |
3688 | if (write) | |
3689 | set_schedstats(state); | |
3690 | return err; | |
3691 | } | |
4698f88c JP |
3692 | #endif /* CONFIG_PROC_SYSCTL */ |
3693 | #else /* !CONFIG_SCHEDSTATS */ | |
3694 | static inline void init_schedstats(void) {} | |
3695 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
3696 | |
3697 | /* | |
3698 | * fork()/clone()-time setup: | |
3699 | */ | |
aab03e05 | 3700 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3701 | { |
0122ec5b | 3702 | unsigned long flags; |
dd41f596 | 3703 | |
5e1576ed | 3704 | __sched_fork(clone_flags, p); |
06b83b5f | 3705 | /* |
7dc603c9 | 3706 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
3707 | * nobody will actually run it, and a signal or other external |
3708 | * event cannot wake it up and insert it on the runqueue either. | |
3709 | */ | |
7dc603c9 | 3710 | p->state = TASK_NEW; |
dd41f596 | 3711 | |
c350a04e MG |
3712 | /* |
3713 | * Make sure we do not leak PI boosting priority to the child. | |
3714 | */ | |
3715 | p->prio = current->normal_prio; | |
3716 | ||
e8f14172 PB |
3717 | uclamp_fork(p); |
3718 | ||
b9dc29e7 MG |
3719 | /* |
3720 | * Revert to default priority/policy on fork if requested. | |
3721 | */ | |
3722 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 3723 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 3724 | p->policy = SCHED_NORMAL; |
6c697bdf | 3725 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
3726 | p->rt_priority = 0; |
3727 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
3728 | p->static_prio = NICE_TO_PRIO(0); | |
3729 | ||
3730 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 3731 | set_load_weight(p, false); |
6c697bdf | 3732 | |
b9dc29e7 MG |
3733 | /* |
3734 | * We don't need the reset flag anymore after the fork. It has | |
3735 | * fulfilled its duty: | |
3736 | */ | |
3737 | p->sched_reset_on_fork = 0; | |
3738 | } | |
ca94c442 | 3739 | |
af0fffd9 | 3740 | if (dl_prio(p->prio)) |
aab03e05 | 3741 | return -EAGAIN; |
af0fffd9 | 3742 | else if (rt_prio(p->prio)) |
aab03e05 | 3743 | p->sched_class = &rt_sched_class; |
af0fffd9 | 3744 | else |
2ddbf952 | 3745 | p->sched_class = &fair_sched_class; |
b29739f9 | 3746 | |
7dc603c9 | 3747 | init_entity_runnable_average(&p->se); |
cd29fe6f | 3748 | |
86951599 PZ |
3749 | /* |
3750 | * The child is not yet in the pid-hash so no cgroup attach races, | |
3751 | * and the cgroup is pinned to this child due to cgroup_fork() | |
3752 | * is ran before sched_fork(). | |
3753 | * | |
3754 | * Silence PROVE_RCU. | |
3755 | */ | |
0122ec5b | 3756 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ce3614da | 3757 | rseq_migrate(p); |
e210bffd | 3758 | /* |
d1ccc66d | 3759 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
3760 | * so use __set_task_cpu(). |
3761 | */ | |
af0fffd9 | 3762 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
3763 | if (p->sched_class->task_fork) |
3764 | p->sched_class->task_fork(p); | |
0122ec5b | 3765 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 3766 | |
f6db8347 | 3767 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 3768 | if (likely(sched_info_on())) |
52f17b6c | 3769 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 3770 | #endif |
3ca7a440 PZ |
3771 | #if defined(CONFIG_SMP) |
3772 | p->on_cpu = 0; | |
4866cde0 | 3773 | #endif |
01028747 | 3774 | init_task_preempt_count(p); |
806c09a7 | 3775 | #ifdef CONFIG_SMP |
917b627d | 3776 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 3777 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 3778 | #endif |
aab03e05 | 3779 | return 0; |
1da177e4 LT |
3780 | } |
3781 | ||
13685c4a QY |
3782 | void sched_post_fork(struct task_struct *p) |
3783 | { | |
3784 | uclamp_post_fork(p); | |
3785 | } | |
3786 | ||
332ac17e DF |
3787 | unsigned long to_ratio(u64 period, u64 runtime) |
3788 | { | |
3789 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 3790 | return BW_UNIT; |
332ac17e DF |
3791 | |
3792 | /* | |
3793 | * Doing this here saves a lot of checks in all | |
3794 | * the calling paths, and returning zero seems | |
3795 | * safe for them anyway. | |
3796 | */ | |
3797 | if (period == 0) | |
3798 | return 0; | |
3799 | ||
c52f14d3 | 3800 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
3801 | } |
3802 | ||
1da177e4 LT |
3803 | /* |
3804 | * wake_up_new_task - wake up a newly created task for the first time. | |
3805 | * | |
3806 | * This function will do some initial scheduler statistics housekeeping | |
3807 | * that must be done for every newly created context, then puts the task | |
3808 | * on the runqueue and wakes it. | |
3809 | */ | |
3e51e3ed | 3810 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 3811 | { |
eb580751 | 3812 | struct rq_flags rf; |
dd41f596 | 3813 | struct rq *rq; |
fabf318e | 3814 | |
eb580751 | 3815 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 3816 | p->state = TASK_RUNNING; |
fabf318e PZ |
3817 | #ifdef CONFIG_SMP |
3818 | /* | |
3819 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 3820 | * - cpus_ptr can change in the fork path |
d1ccc66d | 3821 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
3822 | * |
3823 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
3824 | * as we're not fully set-up yet. | |
fabf318e | 3825 | */ |
32e839dd | 3826 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 3827 | rseq_migrate(p); |
3aef1551 | 3828 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 3829 | #endif |
b7fa30c9 | 3830 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 3831 | update_rq_clock(rq); |
d0fe0b9c | 3832 | post_init_entity_util_avg(p); |
0017d735 | 3833 | |
7a57f32a | 3834 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 3835 | trace_sched_wakeup_new(p); |
a7558e01 | 3836 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3837 | #ifdef CONFIG_SMP |
0aaafaab PZ |
3838 | if (p->sched_class->task_woken) { |
3839 | /* | |
b19a888c | 3840 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
3841 | * drop it. |
3842 | */ | |
d8ac8971 | 3843 | rq_unpin_lock(rq, &rf); |
efbbd05a | 3844 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3845 | rq_repin_lock(rq, &rf); |
0aaafaab | 3846 | } |
9a897c5a | 3847 | #endif |
eb580751 | 3848 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
3849 | } |
3850 | ||
e107be36 AK |
3851 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3852 | ||
b7203428 | 3853 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 3854 | |
2ecd9d29 PZ |
3855 | void preempt_notifier_inc(void) |
3856 | { | |
b7203428 | 3857 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
3858 | } |
3859 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
3860 | ||
3861 | void preempt_notifier_dec(void) | |
3862 | { | |
b7203428 | 3863 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
3864 | } |
3865 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
3866 | ||
e107be36 | 3867 | /** |
80dd99b3 | 3868 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3869 | * @notifier: notifier struct to register |
e107be36 AK |
3870 | */ |
3871 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3872 | { | |
b7203428 | 3873 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
3874 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
3875 | ||
e107be36 AK |
3876 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
3877 | } | |
3878 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3879 | ||
3880 | /** | |
3881 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3882 | * @notifier: notifier struct to unregister |
e107be36 | 3883 | * |
d84525a8 | 3884 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
3885 | */ |
3886 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3887 | { | |
3888 | hlist_del(¬ifier->link); | |
3889 | } | |
3890 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3891 | ||
1cde2930 | 3892 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3893 | { |
3894 | struct preempt_notifier *notifier; | |
e107be36 | 3895 | |
b67bfe0d | 3896 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3897 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
3898 | } | |
3899 | ||
1cde2930 PZ |
3900 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
3901 | { | |
b7203428 | 3902 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3903 | __fire_sched_in_preempt_notifiers(curr); |
3904 | } | |
3905 | ||
e107be36 | 3906 | static void |
1cde2930 PZ |
3907 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3908 | struct task_struct *next) | |
e107be36 AK |
3909 | { |
3910 | struct preempt_notifier *notifier; | |
e107be36 | 3911 | |
b67bfe0d | 3912 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3913 | notifier->ops->sched_out(notifier, next); |
3914 | } | |
3915 | ||
1cde2930 PZ |
3916 | static __always_inline void |
3917 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3918 | struct task_struct *next) | |
3919 | { | |
b7203428 | 3920 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3921 | __fire_sched_out_preempt_notifiers(curr, next); |
3922 | } | |
3923 | ||
6d6bc0ad | 3924 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3925 | |
1cde2930 | 3926 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3927 | { |
3928 | } | |
3929 | ||
1cde2930 | 3930 | static inline void |
e107be36 AK |
3931 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3932 | struct task_struct *next) | |
3933 | { | |
3934 | } | |
3935 | ||
6d6bc0ad | 3936 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3937 | |
31cb1bc0 | 3938 | static inline void prepare_task(struct task_struct *next) |
3939 | { | |
3940 | #ifdef CONFIG_SMP | |
3941 | /* | |
3942 | * Claim the task as running, we do this before switching to it | |
3943 | * such that any running task will have this set. | |
58877d34 PZ |
3944 | * |
3945 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 3946 | */ |
58877d34 | 3947 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 3948 | #endif |
3949 | } | |
3950 | ||
3951 | static inline void finish_task(struct task_struct *prev) | |
3952 | { | |
3953 | #ifdef CONFIG_SMP | |
3954 | /* | |
58877d34 PZ |
3955 | * This must be the very last reference to @prev from this CPU. After |
3956 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
3957 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 3958 | * finished. |
3959 | * | |
3960 | * In particular, the load of prev->state in finish_task_switch() must | |
3961 | * happen before this. | |
3962 | * | |
3963 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
3964 | */ | |
3965 | smp_store_release(&prev->on_cpu, 0); | |
3966 | #endif | |
3967 | } | |
3968 | ||
565790d2 PZ |
3969 | #ifdef CONFIG_SMP |
3970 | ||
3971 | static void do_balance_callbacks(struct rq *rq, struct callback_head *head) | |
3972 | { | |
3973 | void (*func)(struct rq *rq); | |
3974 | struct callback_head *next; | |
3975 | ||
3976 | lockdep_assert_held(&rq->lock); | |
3977 | ||
3978 | while (head) { | |
3979 | func = (void (*)(struct rq *))head->func; | |
3980 | next = head->next; | |
3981 | head->next = NULL; | |
3982 | head = next; | |
3983 | ||
3984 | func(rq); | |
3985 | } | |
3986 | } | |
3987 | ||
ae792702 PZ |
3988 | static void balance_push(struct rq *rq); |
3989 | ||
3990 | struct callback_head balance_push_callback = { | |
3991 | .next = NULL, | |
3992 | .func = (void (*)(struct callback_head *))balance_push, | |
3993 | }; | |
3994 | ||
565790d2 PZ |
3995 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) |
3996 | { | |
3997 | struct callback_head *head = rq->balance_callback; | |
3998 | ||
3999 | lockdep_assert_held(&rq->lock); | |
ae792702 | 4000 | if (head) |
565790d2 PZ |
4001 | rq->balance_callback = NULL; |
4002 | ||
4003 | return head; | |
4004 | } | |
4005 | ||
4006 | static void __balance_callbacks(struct rq *rq) | |
4007 | { | |
4008 | do_balance_callbacks(rq, splice_balance_callbacks(rq)); | |
4009 | } | |
4010 | ||
4011 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4012 | { | |
4013 | unsigned long flags; | |
4014 | ||
4015 | if (unlikely(head)) { | |
4016 | raw_spin_lock_irqsave(&rq->lock, flags); | |
4017 | do_balance_callbacks(rq, head); | |
4018 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
4019 | } | |
4020 | } | |
4021 | ||
4022 | #else | |
4023 | ||
4024 | static inline void __balance_callbacks(struct rq *rq) | |
4025 | { | |
4026 | } | |
4027 | ||
4028 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
4029 | { | |
4030 | return NULL; | |
4031 | } | |
4032 | ||
4033 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4034 | { | |
4035 | } | |
4036 | ||
4037 | #endif | |
4038 | ||
269d5992 PZ |
4039 | static inline void |
4040 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4041 | { |
269d5992 PZ |
4042 | /* |
4043 | * Since the runqueue lock will be released by the next | |
4044 | * task (which is an invalid locking op but in the case | |
4045 | * of the scheduler it's an obvious special-case), so we | |
4046 | * do an early lockdep release here: | |
4047 | */ | |
4048 | rq_unpin_lock(rq, rf); | |
5facae4f | 4049 | spin_release(&rq->lock.dep_map, _THIS_IP_); |
31cb1bc0 | 4050 | #ifdef CONFIG_DEBUG_SPINLOCK |
4051 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 4052 | rq->lock.owner = next; |
31cb1bc0 | 4053 | #endif |
269d5992 PZ |
4054 | } |
4055 | ||
4056 | static inline void finish_lock_switch(struct rq *rq) | |
4057 | { | |
31cb1bc0 | 4058 | /* |
4059 | * If we are tracking spinlock dependencies then we have to | |
4060 | * fix up the runqueue lock - which gets 'carried over' from | |
4061 | * prev into current: | |
4062 | */ | |
4063 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
ae792702 | 4064 | __balance_callbacks(rq); |
31cb1bc0 | 4065 | raw_spin_unlock_irq(&rq->lock); |
4066 | } | |
4067 | ||
325ea10c IM |
4068 | /* |
4069 | * NOP if the arch has not defined these: | |
4070 | */ | |
4071 | ||
4072 | #ifndef prepare_arch_switch | |
4073 | # define prepare_arch_switch(next) do { } while (0) | |
4074 | #endif | |
4075 | ||
4076 | #ifndef finish_arch_post_lock_switch | |
4077 | # define finish_arch_post_lock_switch() do { } while (0) | |
4078 | #endif | |
4079 | ||
5fbda3ec TG |
4080 | static inline void kmap_local_sched_out(void) |
4081 | { | |
4082 | #ifdef CONFIG_KMAP_LOCAL | |
4083 | if (unlikely(current->kmap_ctrl.idx)) | |
4084 | __kmap_local_sched_out(); | |
4085 | #endif | |
4086 | } | |
4087 | ||
4088 | static inline void kmap_local_sched_in(void) | |
4089 | { | |
4090 | #ifdef CONFIG_KMAP_LOCAL | |
4091 | if (unlikely(current->kmap_ctrl.idx)) | |
4092 | __kmap_local_sched_in(); | |
4093 | #endif | |
4094 | } | |
4095 | ||
4866cde0 NP |
4096 | /** |
4097 | * prepare_task_switch - prepare to switch tasks | |
4098 | * @rq: the runqueue preparing to switch | |
421cee29 | 4099 | * @prev: the current task that is being switched out |
4866cde0 NP |
4100 | * @next: the task we are going to switch to. |
4101 | * | |
4102 | * This is called with the rq lock held and interrupts off. It must | |
4103 | * be paired with a subsequent finish_task_switch after the context | |
4104 | * switch. | |
4105 | * | |
4106 | * prepare_task_switch sets up locking and calls architecture specific | |
4107 | * hooks. | |
4108 | */ | |
e107be36 AK |
4109 | static inline void |
4110 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
4111 | struct task_struct *next) | |
4866cde0 | 4112 | { |
0ed557aa | 4113 | kcov_prepare_switch(prev); |
43148951 | 4114 | sched_info_switch(rq, prev, next); |
fe4b04fa | 4115 | perf_event_task_sched_out(prev, next); |
d7822b1e | 4116 | rseq_preempt(prev); |
e107be36 | 4117 | fire_sched_out_preempt_notifiers(prev, next); |
5fbda3ec | 4118 | kmap_local_sched_out(); |
31cb1bc0 | 4119 | prepare_task(next); |
4866cde0 NP |
4120 | prepare_arch_switch(next); |
4121 | } | |
4122 | ||
1da177e4 LT |
4123 | /** |
4124 | * finish_task_switch - clean up after a task-switch | |
4125 | * @prev: the thread we just switched away from. | |
4126 | * | |
4866cde0 NP |
4127 | * finish_task_switch must be called after the context switch, paired |
4128 | * with a prepare_task_switch call before the context switch. | |
4129 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
4130 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
4131 | * |
4132 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 4133 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
4134 | * with the lock held can cause deadlocks; see schedule() for |
4135 | * details.) | |
dfa50b60 ON |
4136 | * |
4137 | * The context switch have flipped the stack from under us and restored the | |
4138 | * local variables which were saved when this task called schedule() in the | |
4139 | * past. prev == current is still correct but we need to recalculate this_rq | |
4140 | * because prev may have moved to another CPU. | |
1da177e4 | 4141 | */ |
dfa50b60 | 4142 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
4143 | __releases(rq->lock) |
4144 | { | |
dfa50b60 | 4145 | struct rq *rq = this_rq(); |
1da177e4 | 4146 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 4147 | long prev_state; |
1da177e4 | 4148 | |
609ca066 PZ |
4149 | /* |
4150 | * The previous task will have left us with a preempt_count of 2 | |
4151 | * because it left us after: | |
4152 | * | |
4153 | * schedule() | |
4154 | * preempt_disable(); // 1 | |
4155 | * __schedule() | |
4156 | * raw_spin_lock_irq(&rq->lock) // 2 | |
4157 | * | |
4158 | * Also, see FORK_PREEMPT_COUNT. | |
4159 | */ | |
e2bf1c4b PZ |
4160 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
4161 | "corrupted preempt_count: %s/%d/0x%x\n", | |
4162 | current->comm, current->pid, preempt_count())) | |
4163 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 4164 | |
1da177e4 LT |
4165 | rq->prev_mm = NULL; |
4166 | ||
4167 | /* | |
4168 | * A task struct has one reference for the use as "current". | |
c394cc9f | 4169 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
4170 | * schedule one last time. The schedule call will never return, and |
4171 | * the scheduled task must drop that reference. | |
95913d97 PZ |
4172 | * |
4173 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 4174 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
4175 | * running on another CPU and we could rave with its RUNNING -> DEAD |
4176 | * transition, resulting in a double drop. | |
1da177e4 | 4177 | */ |
55a101f8 | 4178 | prev_state = prev->state; |
bf9fae9f | 4179 | vtime_task_switch(prev); |
a8d757ef | 4180 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 4181 | finish_task(prev); |
4182 | finish_lock_switch(rq); | |
01f23e16 | 4183 | finish_arch_post_lock_switch(); |
0ed557aa | 4184 | kcov_finish_switch(current); |
5fbda3ec TG |
4185 | /* |
4186 | * kmap_local_sched_out() is invoked with rq::lock held and | |
4187 | * interrupts disabled. There is no requirement for that, but the | |
4188 | * sched out code does not have an interrupt enabled section. | |
4189 | * Restoring the maps on sched in does not require interrupts being | |
4190 | * disabled either. | |
4191 | */ | |
4192 | kmap_local_sched_in(); | |
e8fa1362 | 4193 | |
e107be36 | 4194 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 4195 | /* |
70216e18 MD |
4196 | * When switching through a kernel thread, the loop in |
4197 | * membarrier_{private,global}_expedited() may have observed that | |
4198 | * kernel thread and not issued an IPI. It is therefore possible to | |
4199 | * schedule between user->kernel->user threads without passing though | |
4200 | * switch_mm(). Membarrier requires a barrier after storing to | |
4201 | * rq->curr, before returning to userspace, so provide them here: | |
4202 | * | |
4203 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
4204 | * provided by mmdrop(), | |
4205 | * - a sync_core for SYNC_CORE. | |
306e0604 | 4206 | */ |
70216e18 MD |
4207 | if (mm) { |
4208 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 4209 | mmdrop(mm); |
70216e18 | 4210 | } |
1cef1150 PZ |
4211 | if (unlikely(prev_state == TASK_DEAD)) { |
4212 | if (prev->sched_class->task_dead) | |
4213 | prev->sched_class->task_dead(prev); | |
68f24b08 | 4214 | |
1cef1150 PZ |
4215 | /* |
4216 | * Remove function-return probe instances associated with this | |
4217 | * task and put them back on the free list. | |
4218 | */ | |
4219 | kprobe_flush_task(prev); | |
4220 | ||
4221 | /* Task is done with its stack. */ | |
4222 | put_task_stack(prev); | |
4223 | ||
0ff7b2cf | 4224 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 4225 | } |
99e5ada9 | 4226 | |
de734f89 | 4227 | tick_nohz_task_switch(); |
dfa50b60 | 4228 | return rq; |
1da177e4 LT |
4229 | } |
4230 | ||
4231 | /** | |
4232 | * schedule_tail - first thing a freshly forked thread must call. | |
4233 | * @prev: the thread we just switched away from. | |
4234 | */ | |
722a9f92 | 4235 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
4236 | __releases(rq->lock) |
4237 | { | |
1a43a14a | 4238 | struct rq *rq; |
da19ab51 | 4239 | |
609ca066 PZ |
4240 | /* |
4241 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
4242 | * finish_task_switch() for details. | |
4243 | * | |
4244 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
4245 | * and the preempt_enable() will end up enabling preemption (on | |
4246 | * PREEMPT_COUNT kernels). | |
4247 | */ | |
4248 | ||
dfa50b60 | 4249 | rq = finish_task_switch(prev); |
1a43a14a | 4250 | preempt_enable(); |
70b97a7f | 4251 | |
1da177e4 | 4252 | if (current->set_child_tid) |
b488893a | 4253 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
4254 | |
4255 | calculate_sigpending(); | |
1da177e4 LT |
4256 | } |
4257 | ||
4258 | /* | |
dfa50b60 | 4259 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 4260 | */ |
04936948 | 4261 | static __always_inline struct rq * |
70b97a7f | 4262 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 4263 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 4264 | { |
e107be36 | 4265 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 4266 | |
9226d125 ZA |
4267 | /* |
4268 | * For paravirt, this is coupled with an exit in switch_to to | |
4269 | * combine the page table reload and the switch backend into | |
4270 | * one hypercall. | |
4271 | */ | |
224101ed | 4272 | arch_start_context_switch(prev); |
9226d125 | 4273 | |
306e0604 | 4274 | /* |
139d025c PZ |
4275 | * kernel -> kernel lazy + transfer active |
4276 | * user -> kernel lazy + mmgrab() active | |
4277 | * | |
4278 | * kernel -> user switch + mmdrop() active | |
4279 | * user -> user switch | |
306e0604 | 4280 | */ |
139d025c PZ |
4281 | if (!next->mm) { // to kernel |
4282 | enter_lazy_tlb(prev->active_mm, next); | |
4283 | ||
4284 | next->active_mm = prev->active_mm; | |
4285 | if (prev->mm) // from user | |
4286 | mmgrab(prev->active_mm); | |
4287 | else | |
4288 | prev->active_mm = NULL; | |
4289 | } else { // to user | |
227a4aad | 4290 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
4291 | /* |
4292 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 4293 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
4294 | * |
4295 | * The below provides this either through switch_mm(), or in | |
4296 | * case 'prev->active_mm == next->mm' through | |
4297 | * finish_task_switch()'s mmdrop(). | |
4298 | */ | |
139d025c | 4299 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 4300 | |
139d025c PZ |
4301 | if (!prev->mm) { // from kernel |
4302 | /* will mmdrop() in finish_task_switch(). */ | |
4303 | rq->prev_mm = prev->active_mm; | |
4304 | prev->active_mm = NULL; | |
4305 | } | |
1da177e4 | 4306 | } |
92509b73 | 4307 | |
cb42c9a3 | 4308 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 4309 | |
269d5992 | 4310 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
4311 | |
4312 | /* Here we just switch the register state and the stack. */ | |
4313 | switch_to(prev, next, prev); | |
dd41f596 | 4314 | barrier(); |
dfa50b60 ON |
4315 | |
4316 | return finish_task_switch(prev); | |
1da177e4 LT |
4317 | } |
4318 | ||
4319 | /* | |
1c3e8264 | 4320 | * nr_running and nr_context_switches: |
1da177e4 LT |
4321 | * |
4322 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 4323 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
4324 | */ |
4325 | unsigned long nr_running(void) | |
4326 | { | |
4327 | unsigned long i, sum = 0; | |
4328 | ||
4329 | for_each_online_cpu(i) | |
4330 | sum += cpu_rq(i)->nr_running; | |
4331 | ||
4332 | return sum; | |
f711f609 | 4333 | } |
1da177e4 | 4334 | |
2ee507c4 | 4335 | /* |
d1ccc66d | 4336 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
4337 | * |
4338 | * Caution: this function does not check that the caller has disabled | |
4339 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
4340 | * race. The caller is responsible to use it correctly, for example: | |
4341 | * | |
dfcb245e | 4342 | * - from a non-preemptible section (of course) |
00cc1633 DD |
4343 | * |
4344 | * - from a thread that is bound to a single CPU | |
4345 | * | |
4346 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
4347 | */ |
4348 | bool single_task_running(void) | |
4349 | { | |
00cc1633 | 4350 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
4351 | } |
4352 | EXPORT_SYMBOL(single_task_running); | |
4353 | ||
1da177e4 | 4354 | unsigned long long nr_context_switches(void) |
46cb4b7c | 4355 | { |
cc94abfc SR |
4356 | int i; |
4357 | unsigned long long sum = 0; | |
46cb4b7c | 4358 | |
0a945022 | 4359 | for_each_possible_cpu(i) |
1da177e4 | 4360 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 4361 | |
1da177e4 LT |
4362 | return sum; |
4363 | } | |
483b4ee6 | 4364 | |
145d952a DL |
4365 | /* |
4366 | * Consumers of these two interfaces, like for example the cpuidle menu | |
4367 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
4368 | * for a CPU that has IO-wait which might not even end up running the task when | |
4369 | * it does become runnable. | |
4370 | */ | |
4371 | ||
4372 | unsigned long nr_iowait_cpu(int cpu) | |
4373 | { | |
4374 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
4375 | } | |
4376 | ||
e33a9bba | 4377 | /* |
b19a888c | 4378 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
4379 | * |
4380 | * The idea behind IO-wait account is to account the idle time that we could | |
4381 | * have spend running if it were not for IO. That is, if we were to improve the | |
4382 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
4383 | * | |
4384 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
4385 | * idle time as IO-wait, because if the storage were faster, it could've been | |
4386 | * running and we'd not be idle. | |
4387 | * | |
4388 | * This has been extended to SMP, by doing the same for each CPU. This however | |
4389 | * is broken. | |
4390 | * | |
4391 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
4392 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
4393 | * though, if the storage were faster, both could've ran at the same time, | |
4394 | * utilising both CPUs. | |
4395 | * | |
4396 | * This means, that when looking globally, the current IO-wait accounting on | |
4397 | * SMP is a lower bound, by reason of under accounting. | |
4398 | * | |
4399 | * Worse, since the numbers are provided per CPU, they are sometimes | |
4400 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
4401 | * associated with any one particular CPU, it can wake to another CPU than it | |
4402 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
4403 | * | |
4404 | * Task CPU affinities can make all that even more 'interesting'. | |
4405 | */ | |
4406 | ||
1da177e4 LT |
4407 | unsigned long nr_iowait(void) |
4408 | { | |
4409 | unsigned long i, sum = 0; | |
483b4ee6 | 4410 | |
0a945022 | 4411 | for_each_possible_cpu(i) |
145d952a | 4412 | sum += nr_iowait_cpu(i); |
46cb4b7c | 4413 | |
1da177e4 LT |
4414 | return sum; |
4415 | } | |
483b4ee6 | 4416 | |
dd41f596 | 4417 | #ifdef CONFIG_SMP |
8a0be9ef | 4418 | |
46cb4b7c | 4419 | /* |
38022906 PZ |
4420 | * sched_exec - execve() is a valuable balancing opportunity, because at |
4421 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 4422 | */ |
38022906 | 4423 | void sched_exec(void) |
46cb4b7c | 4424 | { |
38022906 | 4425 | struct task_struct *p = current; |
1da177e4 | 4426 | unsigned long flags; |
0017d735 | 4427 | int dest_cpu; |
46cb4b7c | 4428 | |
8f42ced9 | 4429 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 4430 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
4431 | if (dest_cpu == smp_processor_id()) |
4432 | goto unlock; | |
38022906 | 4433 | |
8f42ced9 | 4434 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 4435 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 4436 | |
8f42ced9 PZ |
4437 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
4438 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
4439 | return; |
4440 | } | |
0017d735 | 4441 | unlock: |
8f42ced9 | 4442 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 4443 | } |
dd41f596 | 4444 | |
1da177e4 LT |
4445 | #endif |
4446 | ||
1da177e4 | 4447 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 4448 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
4449 | |
4450 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 4451 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 4452 | |
6075620b GG |
4453 | /* |
4454 | * The function fair_sched_class.update_curr accesses the struct curr | |
4455 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
4456 | * we observe a high rate of cache misses in practice. | |
4457 | * Prefetching this data results in improved performance. | |
4458 | */ | |
4459 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
4460 | { | |
4461 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
4462 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
4463 | #else | |
4464 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
4465 | #endif | |
4466 | prefetch(curr); | |
4467 | prefetch(&curr->exec_start); | |
4468 | } | |
4469 | ||
c5f8d995 HS |
4470 | /* |
4471 | * Return accounted runtime for the task. | |
4472 | * In case the task is currently running, return the runtime plus current's | |
4473 | * pending runtime that have not been accounted yet. | |
4474 | */ | |
4475 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4476 | { | |
eb580751 | 4477 | struct rq_flags rf; |
c5f8d995 | 4478 | struct rq *rq; |
6e998916 | 4479 | u64 ns; |
c5f8d995 | 4480 | |
911b2898 PZ |
4481 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
4482 | /* | |
97fb7a0a | 4483 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
4484 | * So we have a optimization chance when the task's delta_exec is 0. |
4485 | * Reading ->on_cpu is racy, but this is ok. | |
4486 | * | |
d1ccc66d IM |
4487 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
4488 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 4489 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
4490 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
4491 | * been accounted, so we're correct here as well. | |
911b2898 | 4492 | */ |
da0c1e65 | 4493 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
4494 | return p->se.sum_exec_runtime; |
4495 | #endif | |
4496 | ||
eb580751 | 4497 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
4498 | /* |
4499 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
4500 | * project cycles that may never be accounted to this | |
4501 | * thread, breaking clock_gettime(). | |
4502 | */ | |
4503 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 4504 | prefetch_curr_exec_start(p); |
6e998916 SG |
4505 | update_rq_clock(rq); |
4506 | p->sched_class->update_curr(rq); | |
4507 | } | |
4508 | ns = p->se.sum_exec_runtime; | |
eb580751 | 4509 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
4510 | |
4511 | return ns; | |
4512 | } | |
48f24c4d | 4513 | |
7835b98b CL |
4514 | /* |
4515 | * This function gets called by the timer code, with HZ frequency. | |
4516 | * We call it with interrupts disabled. | |
7835b98b CL |
4517 | */ |
4518 | void scheduler_tick(void) | |
4519 | { | |
7835b98b CL |
4520 | int cpu = smp_processor_id(); |
4521 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4522 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 4523 | struct rq_flags rf; |
b4eccf5f | 4524 | unsigned long thermal_pressure; |
3e51f33f | 4525 | |
1567c3e3 | 4526 | arch_scale_freq_tick(); |
3e51f33f | 4527 | sched_clock_tick(); |
dd41f596 | 4528 | |
8a8c69c3 PZ |
4529 | rq_lock(rq, &rf); |
4530 | ||
3e51f33f | 4531 | update_rq_clock(rq); |
b4eccf5f | 4532 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 4533 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 4534 | curr->sched_class->task_tick(rq, curr, 0); |
3289bdb4 | 4535 | calc_global_load_tick(rq); |
eb414681 | 4536 | psi_task_tick(rq); |
8a8c69c3 PZ |
4537 | |
4538 | rq_unlock(rq, &rf); | |
7835b98b | 4539 | |
e9d2b064 | 4540 | perf_event_task_tick(); |
e220d2dc | 4541 | |
e418e1c2 | 4542 | #ifdef CONFIG_SMP |
6eb57e0d | 4543 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 4544 | trigger_load_balance(rq); |
e418e1c2 | 4545 | #endif |
1da177e4 LT |
4546 | } |
4547 | ||
265f22a9 | 4548 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
4549 | |
4550 | struct tick_work { | |
4551 | int cpu; | |
b55bd585 | 4552 | atomic_t state; |
d84b3131 FW |
4553 | struct delayed_work work; |
4554 | }; | |
b55bd585 PM |
4555 | /* Values for ->state, see diagram below. */ |
4556 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
4557 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
4558 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
4559 | ||
4560 | /* | |
4561 | * State diagram for ->state: | |
4562 | * | |
4563 | * | |
4564 | * TICK_SCHED_REMOTE_OFFLINE | |
4565 | * | ^ | |
4566 | * | | | |
4567 | * | | sched_tick_remote() | |
4568 | * | | | |
4569 | * | | | |
4570 | * +--TICK_SCHED_REMOTE_OFFLINING | |
4571 | * | ^ | |
4572 | * | | | |
4573 | * sched_tick_start() | | sched_tick_stop() | |
4574 | * | | | |
4575 | * V | | |
4576 | * TICK_SCHED_REMOTE_RUNNING | |
4577 | * | |
4578 | * | |
4579 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
4580 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
4581 | */ | |
d84b3131 FW |
4582 | |
4583 | static struct tick_work __percpu *tick_work_cpu; | |
4584 | ||
4585 | static void sched_tick_remote(struct work_struct *work) | |
4586 | { | |
4587 | struct delayed_work *dwork = to_delayed_work(work); | |
4588 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
4589 | int cpu = twork->cpu; | |
4590 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 4591 | struct task_struct *curr; |
d84b3131 | 4592 | struct rq_flags rf; |
d9c0ffca | 4593 | u64 delta; |
b55bd585 | 4594 | int os; |
d84b3131 FW |
4595 | |
4596 | /* | |
4597 | * Handle the tick only if it appears the remote CPU is running in full | |
4598 | * dynticks mode. The check is racy by nature, but missing a tick or | |
4599 | * having one too much is no big deal because the scheduler tick updates | |
4600 | * statistics and checks timeslices in a time-independent way, regardless | |
4601 | * of when exactly it is running. | |
4602 | */ | |
488603b8 | 4603 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 4604 | goto out_requeue; |
d84b3131 | 4605 | |
d9c0ffca FW |
4606 | rq_lock_irq(rq, &rf); |
4607 | curr = rq->curr; | |
488603b8 | 4608 | if (cpu_is_offline(cpu)) |
d9c0ffca | 4609 | goto out_unlock; |
d84b3131 | 4610 | |
d9c0ffca | 4611 | update_rq_clock(rq); |
d9c0ffca | 4612 | |
488603b8 SW |
4613 | if (!is_idle_task(curr)) { |
4614 | /* | |
4615 | * Make sure the next tick runs within a reasonable | |
4616 | * amount of time. | |
4617 | */ | |
4618 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
4619 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
4620 | } | |
d9c0ffca FW |
4621 | curr->sched_class->task_tick(rq, curr, 0); |
4622 | ||
ebc0f83c | 4623 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
4624 | out_unlock: |
4625 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 4626 | out_requeue: |
ebc0f83c | 4627 | |
d84b3131 FW |
4628 | /* |
4629 | * Run the remote tick once per second (1Hz). This arbitrary | |
4630 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
4631 | * to keep scheduler internal stats reasonably up to date. But |
4632 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 4633 | */ |
b55bd585 PM |
4634 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
4635 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
4636 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
4637 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
4638 | } |
4639 | ||
4640 | static void sched_tick_start(int cpu) | |
4641 | { | |
b55bd585 | 4642 | int os; |
d84b3131 FW |
4643 | struct tick_work *twork; |
4644 | ||
4645 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4646 | return; | |
4647 | ||
4648 | WARN_ON_ONCE(!tick_work_cpu); | |
4649 | ||
4650 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4651 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
4652 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
4653 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
4654 | twork->cpu = cpu; | |
4655 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
4656 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
4657 | } | |
d84b3131 FW |
4658 | } |
4659 | ||
4660 | #ifdef CONFIG_HOTPLUG_CPU | |
4661 | static void sched_tick_stop(int cpu) | |
4662 | { | |
4663 | struct tick_work *twork; | |
b55bd585 | 4664 | int os; |
d84b3131 FW |
4665 | |
4666 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4667 | return; | |
4668 | ||
4669 | WARN_ON_ONCE(!tick_work_cpu); | |
4670 | ||
4671 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4672 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
4673 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
4674 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
4675 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
4676 | } |
4677 | #endif /* CONFIG_HOTPLUG_CPU */ | |
4678 | ||
4679 | int __init sched_tick_offload_init(void) | |
4680 | { | |
4681 | tick_work_cpu = alloc_percpu(struct tick_work); | |
4682 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
4683 | return 0; |
4684 | } | |
4685 | ||
4686 | #else /* !CONFIG_NO_HZ_FULL */ | |
4687 | static inline void sched_tick_start(int cpu) { } | |
4688 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 4689 | #endif |
1da177e4 | 4690 | |
c1a280b6 | 4691 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 4692 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
4693 | /* |
4694 | * If the value passed in is equal to the current preempt count | |
4695 | * then we just disabled preemption. Start timing the latency. | |
4696 | */ | |
4697 | static inline void preempt_latency_start(int val) | |
4698 | { | |
4699 | if (preempt_count() == val) { | |
4700 | unsigned long ip = get_lock_parent_ip(); | |
4701 | #ifdef CONFIG_DEBUG_PREEMPT | |
4702 | current->preempt_disable_ip = ip; | |
4703 | #endif | |
4704 | trace_preempt_off(CALLER_ADDR0, ip); | |
4705 | } | |
4706 | } | |
7e49fcce | 4707 | |
edafe3a5 | 4708 | void preempt_count_add(int val) |
1da177e4 | 4709 | { |
6cd8a4bb | 4710 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4711 | /* |
4712 | * Underflow? | |
4713 | */ | |
9a11b49a IM |
4714 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4715 | return; | |
6cd8a4bb | 4716 | #endif |
bdb43806 | 4717 | __preempt_count_add(val); |
6cd8a4bb | 4718 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4719 | /* |
4720 | * Spinlock count overflowing soon? | |
4721 | */ | |
33859f7f MOS |
4722 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4723 | PREEMPT_MASK - 10); | |
6cd8a4bb | 4724 | #endif |
47252cfb | 4725 | preempt_latency_start(val); |
1da177e4 | 4726 | } |
bdb43806 | 4727 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 4728 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 4729 | |
47252cfb SR |
4730 | /* |
4731 | * If the value passed in equals to the current preempt count | |
4732 | * then we just enabled preemption. Stop timing the latency. | |
4733 | */ | |
4734 | static inline void preempt_latency_stop(int val) | |
4735 | { | |
4736 | if (preempt_count() == val) | |
4737 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
4738 | } | |
4739 | ||
edafe3a5 | 4740 | void preempt_count_sub(int val) |
1da177e4 | 4741 | { |
6cd8a4bb | 4742 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4743 | /* |
4744 | * Underflow? | |
4745 | */ | |
01e3eb82 | 4746 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4747 | return; |
1da177e4 LT |
4748 | /* |
4749 | * Is the spinlock portion underflowing? | |
4750 | */ | |
9a11b49a IM |
4751 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4752 | !(preempt_count() & PREEMPT_MASK))) | |
4753 | return; | |
6cd8a4bb | 4754 | #endif |
9a11b49a | 4755 | |
47252cfb | 4756 | preempt_latency_stop(val); |
bdb43806 | 4757 | __preempt_count_sub(val); |
1da177e4 | 4758 | } |
bdb43806 | 4759 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 4760 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 4761 | |
47252cfb SR |
4762 | #else |
4763 | static inline void preempt_latency_start(int val) { } | |
4764 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
4765 | #endif |
4766 | ||
59ddbcb2 IM |
4767 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
4768 | { | |
4769 | #ifdef CONFIG_DEBUG_PREEMPT | |
4770 | return p->preempt_disable_ip; | |
4771 | #else | |
4772 | return 0; | |
4773 | #endif | |
4774 | } | |
4775 | ||
1da177e4 | 4776 | /* |
dd41f596 | 4777 | * Print scheduling while atomic bug: |
1da177e4 | 4778 | */ |
dd41f596 | 4779 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4780 | { |
d1c6d149 VN |
4781 | /* Save this before calling printk(), since that will clobber it */ |
4782 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
4783 | ||
664dfa65 DJ |
4784 | if (oops_in_progress) |
4785 | return; | |
4786 | ||
3df0fc5b PZ |
4787 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4788 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4789 | |
dd41f596 | 4790 | debug_show_held_locks(prev); |
e21f5b15 | 4791 | print_modules(); |
dd41f596 IM |
4792 | if (irqs_disabled()) |
4793 | print_irqtrace_events(prev); | |
d1c6d149 VN |
4794 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
4795 | && in_atomic_preempt_off()) { | |
8f47b187 | 4796 | pr_err("Preemption disabled at:"); |
2062a4e8 | 4797 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 4798 | } |
748c7201 DBO |
4799 | if (panic_on_warn) |
4800 | panic("scheduling while atomic\n"); | |
4801 | ||
6135fc1e | 4802 | dump_stack(); |
373d4d09 | 4803 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 4804 | } |
1da177e4 | 4805 | |
dd41f596 IM |
4806 | /* |
4807 | * Various schedule()-time debugging checks and statistics: | |
4808 | */ | |
312364f3 | 4809 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 4810 | { |
0d9e2632 | 4811 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
4812 | if (task_stack_end_corrupted(prev)) |
4813 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
4814 | |
4815 | if (task_scs_end_corrupted(prev)) | |
4816 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 4817 | #endif |
b99def8b | 4818 | |
312364f3 DV |
4819 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
4820 | if (!preempt && prev->state && prev->non_block_count) { | |
4821 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", | |
4822 | prev->comm, prev->pid, prev->non_block_count); | |
4823 | dump_stack(); | |
4824 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
4825 | } | |
4826 | #endif | |
4827 | ||
1dc0fffc | 4828 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 4829 | __schedule_bug(prev); |
1dc0fffc PZ |
4830 | preempt_count_set(PREEMPT_DISABLED); |
4831 | } | |
b3fbab05 | 4832 | rcu_sleep_check(); |
9f68b5b7 | 4833 | SCHED_WARN_ON(ct_state() == CONTEXT_USER); |
dd41f596 | 4834 | |
1da177e4 LT |
4835 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4836 | ||
ae92882e | 4837 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
4838 | } |
4839 | ||
457d1f46 CY |
4840 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
4841 | struct rq_flags *rf) | |
4842 | { | |
4843 | #ifdef CONFIG_SMP | |
4844 | const struct sched_class *class; | |
4845 | /* | |
4846 | * We must do the balancing pass before put_prev_task(), such | |
4847 | * that when we release the rq->lock the task is in the same | |
4848 | * state as before we took rq->lock. | |
4849 | * | |
4850 | * We can terminate the balance pass as soon as we know there is | |
4851 | * a runnable task of @class priority or higher. | |
4852 | */ | |
4853 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
4854 | if (class->balance(rq, prev, rf)) | |
4855 | break; | |
4856 | } | |
4857 | #endif | |
4858 | ||
4859 | put_prev_task(rq, prev); | |
4860 | } | |
4861 | ||
dd41f596 IM |
4862 | /* |
4863 | * Pick up the highest-prio task: | |
4864 | */ | |
4865 | static inline struct task_struct * | |
d8ac8971 | 4866 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 4867 | { |
49ee5768 | 4868 | const struct sched_class *class; |
dd41f596 | 4869 | struct task_struct *p; |
1da177e4 LT |
4870 | |
4871 | /* | |
0ba87bb2 PZ |
4872 | * Optimization: we know that if all tasks are in the fair class we can |
4873 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 4874 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 4875 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 4876 | */ |
aa93cd53 | 4877 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
4878 | rq->nr_running == rq->cfs.h_nr_running)) { |
4879 | ||
5d7d6056 | 4880 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 4881 | if (unlikely(p == RETRY_TASK)) |
67692435 | 4882 | goto restart; |
6ccdc84b | 4883 | |
d1ccc66d | 4884 | /* Assumes fair_sched_class->next == idle_sched_class */ |
5d7d6056 | 4885 | if (!p) { |
f488e105 | 4886 | put_prev_task(rq, prev); |
98c2f700 | 4887 | p = pick_next_task_idle(rq); |
f488e105 | 4888 | } |
6ccdc84b PZ |
4889 | |
4890 | return p; | |
1da177e4 LT |
4891 | } |
4892 | ||
67692435 | 4893 | restart: |
457d1f46 | 4894 | put_prev_task_balance(rq, prev, rf); |
67692435 | 4895 | |
34f971f6 | 4896 | for_each_class(class) { |
98c2f700 | 4897 | p = class->pick_next_task(rq); |
67692435 | 4898 | if (p) |
dd41f596 | 4899 | return p; |
dd41f596 | 4900 | } |
34f971f6 | 4901 | |
d1ccc66d IM |
4902 | /* The idle class should always have a runnable task: */ |
4903 | BUG(); | |
dd41f596 | 4904 | } |
1da177e4 | 4905 | |
dd41f596 | 4906 | /* |
c259e01a | 4907 | * __schedule() is the main scheduler function. |
edde96ea PE |
4908 | * |
4909 | * The main means of driving the scheduler and thus entering this function are: | |
4910 | * | |
4911 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
4912 | * | |
4913 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
4914 | * paths. For example, see arch/x86/entry_64.S. | |
4915 | * | |
4916 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
4917 | * interrupt handler scheduler_tick(). | |
4918 | * | |
4919 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
4920 | * task to the run-queue and that's it. | |
4921 | * | |
4922 | * Now, if the new task added to the run-queue preempts the current | |
4923 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
4924 | * called on the nearest possible occasion: | |
4925 | * | |
c1a280b6 | 4926 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
4927 | * |
4928 | * - in syscall or exception context, at the next outmost | |
4929 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
4930 | * spin_unlock()!) | |
4931 | * | |
4932 | * - in IRQ context, return from interrupt-handler to | |
4933 | * preemptible context | |
4934 | * | |
c1a280b6 | 4935 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
4936 | * then at the next: |
4937 | * | |
4938 | * - cond_resched() call | |
4939 | * - explicit schedule() call | |
4940 | * - return from syscall or exception to user-space | |
4941 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 4942 | * |
b30f0e3f | 4943 | * WARNING: must be called with preemption disabled! |
dd41f596 | 4944 | */ |
499d7955 | 4945 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
4946 | { |
4947 | struct task_struct *prev, *next; | |
67ca7bde | 4948 | unsigned long *switch_count; |
dbfb089d | 4949 | unsigned long prev_state; |
d8ac8971 | 4950 | struct rq_flags rf; |
dd41f596 | 4951 | struct rq *rq; |
31656519 | 4952 | int cpu; |
dd41f596 | 4953 | |
dd41f596 IM |
4954 | cpu = smp_processor_id(); |
4955 | rq = cpu_rq(cpu); | |
dd41f596 | 4956 | prev = rq->curr; |
dd41f596 | 4957 | |
312364f3 | 4958 | schedule_debug(prev, preempt); |
1da177e4 | 4959 | |
31656519 | 4960 | if (sched_feat(HRTICK)) |
f333fdc9 | 4961 | hrtick_clear(rq); |
8f4d37ec | 4962 | |
46a5d164 | 4963 | local_irq_disable(); |
bcbfdd01 | 4964 | rcu_note_context_switch(preempt); |
46a5d164 | 4965 | |
e0acd0a6 ON |
4966 | /* |
4967 | * Make sure that signal_pending_state()->signal_pending() below | |
4968 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
4969 | * done by the caller to avoid the race with signal_wake_up(): |
4970 | * | |
4971 | * __set_current_state(@state) signal_wake_up() | |
4972 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
4973 | * wake_up_state(p, state) | |
4974 | * LOCK rq->lock LOCK p->pi_state | |
4975 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
4976 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 4977 | * |
dbfb089d | 4978 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 4979 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 4980 | */ |
8a8c69c3 | 4981 | rq_lock(rq, &rf); |
d89e588c | 4982 | smp_mb__after_spinlock(); |
1da177e4 | 4983 | |
d1ccc66d IM |
4984 | /* Promote REQ to ACT */ |
4985 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 4986 | update_rq_clock(rq); |
9edfbfed | 4987 | |
246d86b5 | 4988 | switch_count = &prev->nivcsw; |
d136122f | 4989 | |
dbfb089d | 4990 | /* |
d136122f PZ |
4991 | * We must load prev->state once (task_struct::state is volatile), such |
4992 | * that: | |
4993 | * | |
4994 | * - we form a control dependency vs deactivate_task() below. | |
4995 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 4996 | */ |
d136122f PZ |
4997 | prev_state = prev->state; |
4998 | if (!preempt && prev_state) { | |
dbfb089d | 4999 | if (signal_pending_state(prev_state, prev)) { |
1da177e4 | 5000 | prev->state = TASK_RUNNING; |
21aa9af0 | 5001 | } else { |
dbfb089d PZ |
5002 | prev->sched_contributes_to_load = |
5003 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
5004 | !(prev_state & TASK_NOLOAD) && | |
5005 | !(prev->flags & PF_FROZEN); | |
5006 | ||
5007 | if (prev->sched_contributes_to_load) | |
5008 | rq->nr_uninterruptible++; | |
5009 | ||
5010 | /* | |
5011 | * __schedule() ttwu() | |
d136122f PZ |
5012 | * prev_state = prev->state; if (p->on_rq && ...) |
5013 | * if (prev_state) goto out; | |
5014 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
5015 | * p->state = TASK_WAKING | |
5016 | * | |
5017 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
5018 | * |
5019 | * After this, schedule() must not care about p->state any more. | |
5020 | */ | |
bce4dc80 | 5021 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 5022 | |
e33a9bba TH |
5023 | if (prev->in_iowait) { |
5024 | atomic_inc(&rq->nr_iowait); | |
5025 | delayacct_blkio_start(); | |
5026 | } | |
21aa9af0 | 5027 | } |
dd41f596 | 5028 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5029 | } |
5030 | ||
d8ac8971 | 5031 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 5032 | clear_tsk_need_resched(prev); |
f27dde8d | 5033 | clear_preempt_need_resched(); |
1da177e4 | 5034 | |
1da177e4 | 5035 | if (likely(prev != next)) { |
1da177e4 | 5036 | rq->nr_switches++; |
5311a98f EB |
5037 | /* |
5038 | * RCU users of rcu_dereference(rq->curr) may not see | |
5039 | * changes to task_struct made by pick_next_task(). | |
5040 | */ | |
5041 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
5042 | /* |
5043 | * The membarrier system call requires each architecture | |
5044 | * to have a full memory barrier after updating | |
306e0604 MD |
5045 | * rq->curr, before returning to user-space. |
5046 | * | |
5047 | * Here are the schemes providing that barrier on the | |
5048 | * various architectures: | |
5049 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
5050 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
5051 | * - finish_lock_switch() for weakly-ordered | |
5052 | * architectures where spin_unlock is a full barrier, | |
5053 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
5054 | * is a RELEASE barrier), | |
22e4ebb9 | 5055 | */ |
1da177e4 LT |
5056 | ++*switch_count; |
5057 | ||
af449901 | 5058 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
5059 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
5060 | ||
c73464b1 | 5061 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
5062 | |
5063 | /* Also unlocks the rq: */ | |
5064 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 5065 | } else { |
cb42c9a3 | 5066 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 5067 | |
565790d2 PZ |
5068 | rq_unpin_lock(rq, &rf); |
5069 | __balance_callbacks(rq); | |
5070 | raw_spin_unlock_irq(&rq->lock); | |
5071 | } | |
1da177e4 | 5072 | } |
c259e01a | 5073 | |
9af6528e PZ |
5074 | void __noreturn do_task_dead(void) |
5075 | { | |
d1ccc66d | 5076 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 5077 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
5078 | |
5079 | /* Tell freezer to ignore us: */ | |
5080 | current->flags |= PF_NOFREEZE; | |
5081 | ||
9af6528e PZ |
5082 | __schedule(false); |
5083 | BUG(); | |
d1ccc66d IM |
5084 | |
5085 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 5086 | for (;;) |
d1ccc66d | 5087 | cpu_relax(); |
9af6528e PZ |
5088 | } |
5089 | ||
9c40cef2 TG |
5090 | static inline void sched_submit_work(struct task_struct *tsk) |
5091 | { | |
c1cecf88 SAS |
5092 | unsigned int task_flags; |
5093 | ||
b0fdc013 | 5094 | if (!tsk->state) |
9c40cef2 | 5095 | return; |
6d25be57 | 5096 | |
c1cecf88 | 5097 | task_flags = tsk->flags; |
6d25be57 TG |
5098 | /* |
5099 | * If a worker went to sleep, notify and ask workqueue whether | |
5100 | * it wants to wake up a task to maintain concurrency. | |
5101 | * As this function is called inside the schedule() context, | |
5102 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
5103 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
5104 | * requires it. | |
6d25be57 | 5105 | */ |
c1cecf88 | 5106 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 5107 | preempt_disable(); |
c1cecf88 | 5108 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
5109 | wq_worker_sleeping(tsk); |
5110 | else | |
5111 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
5112 | preempt_enable_no_resched(); |
5113 | } | |
5114 | ||
b0fdc013 SAS |
5115 | if (tsk_is_pi_blocked(tsk)) |
5116 | return; | |
5117 | ||
9c40cef2 TG |
5118 | /* |
5119 | * If we are going to sleep and we have plugged IO queued, | |
5120 | * make sure to submit it to avoid deadlocks. | |
5121 | */ | |
5122 | if (blk_needs_flush_plug(tsk)) | |
5123 | blk_schedule_flush_plug(tsk); | |
5124 | } | |
5125 | ||
6d25be57 TG |
5126 | static void sched_update_worker(struct task_struct *tsk) |
5127 | { | |
771b53d0 JA |
5128 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
5129 | if (tsk->flags & PF_WQ_WORKER) | |
5130 | wq_worker_running(tsk); | |
5131 | else | |
5132 | io_wq_worker_running(tsk); | |
5133 | } | |
6d25be57 TG |
5134 | } |
5135 | ||
722a9f92 | 5136 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 5137 | { |
9c40cef2 TG |
5138 | struct task_struct *tsk = current; |
5139 | ||
5140 | sched_submit_work(tsk); | |
bfd9b2b5 | 5141 | do { |
b30f0e3f | 5142 | preempt_disable(); |
fc13aeba | 5143 | __schedule(false); |
b30f0e3f | 5144 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 5145 | } while (need_resched()); |
6d25be57 | 5146 | sched_update_worker(tsk); |
c259e01a | 5147 | } |
1da177e4 LT |
5148 | EXPORT_SYMBOL(schedule); |
5149 | ||
8663effb SRV |
5150 | /* |
5151 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
5152 | * state (have scheduled out non-voluntarily) by making sure that all | |
5153 | * tasks have either left the run queue or have gone into user space. | |
5154 | * As idle tasks do not do either, they must not ever be preempted | |
5155 | * (schedule out non-voluntarily). | |
5156 | * | |
5157 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
5158 | * never enables preemption because it does not call sched_submit_work(). | |
5159 | */ | |
5160 | void __sched schedule_idle(void) | |
5161 | { | |
5162 | /* | |
5163 | * As this skips calling sched_submit_work(), which the idle task does | |
5164 | * regardless because that function is a nop when the task is in a | |
5165 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
5166 | * current task can be in any other state. Note, idle is always in the | |
5167 | * TASK_RUNNING state. | |
5168 | */ | |
5169 | WARN_ON_ONCE(current->state); | |
5170 | do { | |
5171 | __schedule(false); | |
5172 | } while (need_resched()); | |
5173 | } | |
5174 | ||
6775de49 | 5175 | #if defined(CONFIG_CONTEXT_TRACKING) && !defined(CONFIG_HAVE_CONTEXT_TRACKING_OFFSTACK) |
722a9f92 | 5176 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
5177 | { |
5178 | /* | |
5179 | * If we come here after a random call to set_need_resched(), | |
5180 | * or we have been woken up remotely but the IPI has not yet arrived, | |
5181 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
5182 | * we find a better solution. | |
7cc78f8f AL |
5183 | * |
5184 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 5185 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 5186 | * too frequently to make sense yet. |
20ab65e3 | 5187 | */ |
7cc78f8f | 5188 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 5189 | schedule(); |
7cc78f8f | 5190 | exception_exit(prev_state); |
20ab65e3 FW |
5191 | } |
5192 | #endif | |
5193 | ||
c5491ea7 TG |
5194 | /** |
5195 | * schedule_preempt_disabled - called with preemption disabled | |
5196 | * | |
5197 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
5198 | */ | |
5199 | void __sched schedule_preempt_disabled(void) | |
5200 | { | |
ba74c144 | 5201 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
5202 | schedule(); |
5203 | preempt_disable(); | |
5204 | } | |
5205 | ||
06b1f808 | 5206 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
5207 | { |
5208 | do { | |
47252cfb SR |
5209 | /* |
5210 | * Because the function tracer can trace preempt_count_sub() | |
5211 | * and it also uses preempt_enable/disable_notrace(), if | |
5212 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5213 | * by the function tracer will call this function again and | |
5214 | * cause infinite recursion. | |
5215 | * | |
5216 | * Preemption must be disabled here before the function | |
5217 | * tracer can trace. Break up preempt_disable() into two | |
5218 | * calls. One to disable preemption without fear of being | |
5219 | * traced. The other to still record the preemption latency, | |
5220 | * which can also be traced by the function tracer. | |
5221 | */ | |
499d7955 | 5222 | preempt_disable_notrace(); |
47252cfb | 5223 | preempt_latency_start(1); |
fc13aeba | 5224 | __schedule(true); |
47252cfb | 5225 | preempt_latency_stop(1); |
499d7955 | 5226 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
5227 | |
5228 | /* | |
5229 | * Check again in case we missed a preemption opportunity | |
5230 | * between schedule and now. | |
5231 | */ | |
a18b5d01 FW |
5232 | } while (need_resched()); |
5233 | } | |
5234 | ||
c1a280b6 | 5235 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 5236 | /* |
a49b4f40 VS |
5237 | * This is the entry point to schedule() from in-kernel preemption |
5238 | * off of preempt_enable. | |
1da177e4 | 5239 | */ |
722a9f92 | 5240 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 5241 | { |
1da177e4 LT |
5242 | /* |
5243 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5244 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5245 | */ |
fbb00b56 | 5246 | if (likely(!preemptible())) |
1da177e4 LT |
5247 | return; |
5248 | ||
a18b5d01 | 5249 | preempt_schedule_common(); |
1da177e4 | 5250 | } |
376e2424 | 5251 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 5252 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 5253 | |
009f60e2 | 5254 | /** |
4eaca0a8 | 5255 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
5256 | * |
5257 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
5258 | * recursion and tracing preempt enabling caused by the tracing | |
5259 | * infrastructure itself. But as tracing can happen in areas coming | |
5260 | * from userspace or just about to enter userspace, a preempt enable | |
5261 | * can occur before user_exit() is called. This will cause the scheduler | |
5262 | * to be called when the system is still in usermode. | |
5263 | * | |
5264 | * To prevent this, the preempt_enable_notrace will use this function | |
5265 | * instead of preempt_schedule() to exit user context if needed before | |
5266 | * calling the scheduler. | |
5267 | */ | |
4eaca0a8 | 5268 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
5269 | { |
5270 | enum ctx_state prev_ctx; | |
5271 | ||
5272 | if (likely(!preemptible())) | |
5273 | return; | |
5274 | ||
5275 | do { | |
47252cfb SR |
5276 | /* |
5277 | * Because the function tracer can trace preempt_count_sub() | |
5278 | * and it also uses preempt_enable/disable_notrace(), if | |
5279 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5280 | * by the function tracer will call this function again and | |
5281 | * cause infinite recursion. | |
5282 | * | |
5283 | * Preemption must be disabled here before the function | |
5284 | * tracer can trace. Break up preempt_disable() into two | |
5285 | * calls. One to disable preemption without fear of being | |
5286 | * traced. The other to still record the preemption latency, | |
5287 | * which can also be traced by the function tracer. | |
5288 | */ | |
3d8f74dd | 5289 | preempt_disable_notrace(); |
47252cfb | 5290 | preempt_latency_start(1); |
009f60e2 ON |
5291 | /* |
5292 | * Needs preempt disabled in case user_exit() is traced | |
5293 | * and the tracer calls preempt_enable_notrace() causing | |
5294 | * an infinite recursion. | |
5295 | */ | |
5296 | prev_ctx = exception_enter(); | |
fc13aeba | 5297 | __schedule(true); |
009f60e2 ON |
5298 | exception_exit(prev_ctx); |
5299 | ||
47252cfb | 5300 | preempt_latency_stop(1); |
3d8f74dd | 5301 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
5302 | } while (need_resched()); |
5303 | } | |
4eaca0a8 | 5304 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 5305 | |
c1a280b6 | 5306 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 LT |
5307 | |
5308 | /* | |
a49b4f40 | 5309 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5310 | * off of irq context. |
5311 | * Note, that this is called and return with irqs disabled. This will | |
5312 | * protect us against recursive calling from irq. | |
5313 | */ | |
722a9f92 | 5314 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 5315 | { |
b22366cd | 5316 | enum ctx_state prev_state; |
6478d880 | 5317 | |
2ed6e34f | 5318 | /* Catch callers which need to be fixed */ |
f27dde8d | 5319 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 5320 | |
b22366cd FW |
5321 | prev_state = exception_enter(); |
5322 | ||
3a5c359a | 5323 | do { |
3d8f74dd | 5324 | preempt_disable(); |
3a5c359a | 5325 | local_irq_enable(); |
fc13aeba | 5326 | __schedule(true); |
3a5c359a | 5327 | local_irq_disable(); |
3d8f74dd | 5328 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 5329 | } while (need_resched()); |
b22366cd FW |
5330 | |
5331 | exception_exit(prev_state); | |
1da177e4 LT |
5332 | } |
5333 | ||
ac6424b9 | 5334 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 5335 | void *key) |
1da177e4 | 5336 | { |
062d3f95 | 5337 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 5338 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 5339 | } |
1da177e4 LT |
5340 | EXPORT_SYMBOL(default_wake_function); |
5341 | ||
b29739f9 IM |
5342 | #ifdef CONFIG_RT_MUTEXES |
5343 | ||
acd58620 PZ |
5344 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
5345 | { | |
5346 | if (pi_task) | |
5347 | prio = min(prio, pi_task->prio); | |
5348 | ||
5349 | return prio; | |
5350 | } | |
5351 | ||
5352 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5353 | { | |
5354 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
5355 | ||
5356 | return __rt_effective_prio(pi_task, prio); | |
5357 | } | |
5358 | ||
b29739f9 IM |
5359 | /* |
5360 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
5361 | * @p: task to boost |
5362 | * @pi_task: donor task | |
b29739f9 IM |
5363 | * |
5364 | * This function changes the 'effective' priority of a task. It does | |
5365 | * not touch ->normal_prio like __setscheduler(). | |
5366 | * | |
c365c292 TG |
5367 | * Used by the rt_mutex code to implement priority inheritance |
5368 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 5369 | */ |
acd58620 | 5370 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 5371 | { |
acd58620 | 5372 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 5373 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 5374 | const struct sched_class *prev_class; |
eb580751 PZ |
5375 | struct rq_flags rf; |
5376 | struct rq *rq; | |
b29739f9 | 5377 | |
acd58620 PZ |
5378 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
5379 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
5380 | ||
5381 | /* | |
5382 | * If nothing changed; bail early. | |
5383 | */ | |
5384 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
5385 | return; | |
b29739f9 | 5386 | |
eb580751 | 5387 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 5388 | update_rq_clock(rq); |
acd58620 PZ |
5389 | /* |
5390 | * Set under pi_lock && rq->lock, such that the value can be used under | |
5391 | * either lock. | |
5392 | * | |
5393 | * Note that there is loads of tricky to make this pointer cache work | |
5394 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
5395 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
5396 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 5397 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
5398 | */ |
5399 | p->pi_top_task = pi_task; | |
5400 | ||
5401 | /* | |
5402 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
5403 | */ | |
5404 | if (prio == p->prio && !dl_prio(prio)) | |
5405 | goto out_unlock; | |
b29739f9 | 5406 | |
1c4dd99b TG |
5407 | /* |
5408 | * Idle task boosting is a nono in general. There is one | |
5409 | * exception, when PREEMPT_RT and NOHZ is active: | |
5410 | * | |
5411 | * The idle task calls get_next_timer_interrupt() and holds | |
5412 | * the timer wheel base->lock on the CPU and another CPU wants | |
5413 | * to access the timer (probably to cancel it). We can safely | |
5414 | * ignore the boosting request, as the idle CPU runs this code | |
5415 | * with interrupts disabled and will complete the lock | |
5416 | * protected section without being interrupted. So there is no | |
5417 | * real need to boost. | |
5418 | */ | |
5419 | if (unlikely(p == rq->idle)) { | |
5420 | WARN_ON(p != rq->curr); | |
5421 | WARN_ON(p->pi_blocked_on); | |
5422 | goto out_unlock; | |
5423 | } | |
5424 | ||
b91473ff | 5425 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 5426 | oldprio = p->prio; |
ff77e468 PZ |
5427 | |
5428 | if (oldprio == prio) | |
5429 | queue_flag &= ~DEQUEUE_MOVE; | |
5430 | ||
83ab0aa0 | 5431 | prev_class = p->sched_class; |
da0c1e65 | 5432 | queued = task_on_rq_queued(p); |
051a1d1a | 5433 | running = task_current(rq, p); |
da0c1e65 | 5434 | if (queued) |
ff77e468 | 5435 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 5436 | if (running) |
f3cd1c4e | 5437 | put_prev_task(rq, p); |
dd41f596 | 5438 | |
2d3d891d DF |
5439 | /* |
5440 | * Boosting condition are: | |
5441 | * 1. -rt task is running and holds mutex A | |
5442 | * --> -dl task blocks on mutex A | |
5443 | * | |
5444 | * 2. -dl task is running and holds mutex A | |
5445 | * --> -dl task blocks on mutex A and could preempt the | |
5446 | * running task | |
5447 | */ | |
5448 | if (dl_prio(prio)) { | |
466af29b | 5449 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
5450 | (pi_task && dl_prio(pi_task->prio) && |
5451 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2279f540 | 5452 | p->dl.pi_se = pi_task->dl.pi_se; |
ff77e468 | 5453 | queue_flag |= ENQUEUE_REPLENISH; |
2279f540 JL |
5454 | } else { |
5455 | p->dl.pi_se = &p->dl; | |
5456 | } | |
aab03e05 | 5457 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
5458 | } else if (rt_prio(prio)) { |
5459 | if (dl_prio(oldprio)) | |
2279f540 | 5460 | p->dl.pi_se = &p->dl; |
2d3d891d | 5461 | if (oldprio < prio) |
ff77e468 | 5462 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 5463 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
5464 | } else { |
5465 | if (dl_prio(oldprio)) | |
2279f540 | 5466 | p->dl.pi_se = &p->dl; |
746db944 BS |
5467 | if (rt_prio(oldprio)) |
5468 | p->rt.timeout = 0; | |
dd41f596 | 5469 | p->sched_class = &fair_sched_class; |
2d3d891d | 5470 | } |
dd41f596 | 5471 | |
b29739f9 IM |
5472 | p->prio = prio; |
5473 | ||
da0c1e65 | 5474 | if (queued) |
ff77e468 | 5475 | enqueue_task(rq, p, queue_flag); |
a399d233 | 5476 | if (running) |
03b7fad1 | 5477 | set_next_task(rq, p); |
cb469845 | 5478 | |
da7a735e | 5479 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 5480 | out_unlock: |
d1ccc66d IM |
5481 | /* Avoid rq from going away on us: */ |
5482 | preempt_disable(); | |
4c9a4bc8 | 5483 | |
565790d2 PZ |
5484 | rq_unpin_lock(rq, &rf); |
5485 | __balance_callbacks(rq); | |
5486 | raw_spin_unlock(&rq->lock); | |
5487 | ||
4c9a4bc8 | 5488 | preempt_enable(); |
b29739f9 | 5489 | } |
acd58620 PZ |
5490 | #else |
5491 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5492 | { | |
5493 | return prio; | |
5494 | } | |
b29739f9 | 5495 | #endif |
d50dde5a | 5496 | |
36c8b586 | 5497 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5498 | { |
49bd21ef | 5499 | bool queued, running; |
53a23364 | 5500 | int old_prio; |
eb580751 | 5501 | struct rq_flags rf; |
70b97a7f | 5502 | struct rq *rq; |
1da177e4 | 5503 | |
75e45d51 | 5504 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
5505 | return; |
5506 | /* | |
5507 | * We have to be careful, if called from sys_setpriority(), | |
5508 | * the task might be in the middle of scheduling on another CPU. | |
5509 | */ | |
eb580751 | 5510 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
5511 | update_rq_clock(rq); |
5512 | ||
1da177e4 LT |
5513 | /* |
5514 | * The RT priorities are set via sched_setscheduler(), but we still | |
5515 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 5516 | * it won't have any effect on scheduling until the task is |
aab03e05 | 5517 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 5518 | */ |
aab03e05 | 5519 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
5520 | p->static_prio = NICE_TO_PRIO(nice); |
5521 | goto out_unlock; | |
5522 | } | |
da0c1e65 | 5523 | queued = task_on_rq_queued(p); |
49bd21ef | 5524 | running = task_current(rq, p); |
da0c1e65 | 5525 | if (queued) |
7a57f32a | 5526 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
5527 | if (running) |
5528 | put_prev_task(rq, p); | |
1da177e4 | 5529 | |
1da177e4 | 5530 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 5531 | set_load_weight(p, true); |
b29739f9 IM |
5532 | old_prio = p->prio; |
5533 | p->prio = effective_prio(p); | |
1da177e4 | 5534 | |
5443a0be | 5535 | if (queued) |
7134b3e9 | 5536 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 5537 | if (running) |
03b7fad1 | 5538 | set_next_task(rq, p); |
5443a0be FW |
5539 | |
5540 | /* | |
5541 | * If the task increased its priority or is running and | |
5542 | * lowered its priority, then reschedule its CPU: | |
5543 | */ | |
5544 | p->sched_class->prio_changed(rq, p, old_prio); | |
5545 | ||
1da177e4 | 5546 | out_unlock: |
eb580751 | 5547 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 5548 | } |
1da177e4 LT |
5549 | EXPORT_SYMBOL(set_user_nice); |
5550 | ||
e43379f1 MM |
5551 | /* |
5552 | * can_nice - check if a task can reduce its nice value | |
5553 | * @p: task | |
5554 | * @nice: nice value | |
5555 | */ | |
36c8b586 | 5556 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5557 | { |
d1ccc66d | 5558 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 5559 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 5560 | |
78d7d407 | 5561 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5562 | capable(CAP_SYS_NICE)); |
5563 | } | |
5564 | ||
1da177e4 LT |
5565 | #ifdef __ARCH_WANT_SYS_NICE |
5566 | ||
5567 | /* | |
5568 | * sys_nice - change the priority of the current process. | |
5569 | * @increment: priority increment | |
5570 | * | |
5571 | * sys_setpriority is a more generic, but much slower function that | |
5572 | * does similar things. | |
5573 | */ | |
5add95d4 | 5574 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5575 | { |
48f24c4d | 5576 | long nice, retval; |
1da177e4 LT |
5577 | |
5578 | /* | |
5579 | * Setpriority might change our priority at the same moment. | |
5580 | * We don't have to worry. Conceptually one call occurs first | |
5581 | * and we have a single winner. | |
5582 | */ | |
a9467fa3 | 5583 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 5584 | nice = task_nice(current) + increment; |
1da177e4 | 5585 | |
a9467fa3 | 5586 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
5587 | if (increment < 0 && !can_nice(current, nice)) |
5588 | return -EPERM; | |
5589 | ||
1da177e4 LT |
5590 | retval = security_task_setnice(current, nice); |
5591 | if (retval) | |
5592 | return retval; | |
5593 | ||
5594 | set_user_nice(current, nice); | |
5595 | return 0; | |
5596 | } | |
5597 | ||
5598 | #endif | |
5599 | ||
5600 | /** | |
5601 | * task_prio - return the priority value of a given task. | |
5602 | * @p: the task in question. | |
5603 | * | |
e69f6186 | 5604 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
5605 | * RT tasks are offset by -200. Normal tasks are centered |
5606 | * around 0, value goes from -16 to +15. | |
5607 | */ | |
36c8b586 | 5608 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5609 | { |
5610 | return p->prio - MAX_RT_PRIO; | |
5611 | } | |
5612 | ||
1da177e4 | 5613 | /** |
d1ccc66d | 5614 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 5615 | * @cpu: the processor in question. |
e69f6186 YB |
5616 | * |
5617 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
5618 | */ |
5619 | int idle_cpu(int cpu) | |
5620 | { | |
908a3283 TG |
5621 | struct rq *rq = cpu_rq(cpu); |
5622 | ||
5623 | if (rq->curr != rq->idle) | |
5624 | return 0; | |
5625 | ||
5626 | if (rq->nr_running) | |
5627 | return 0; | |
5628 | ||
5629 | #ifdef CONFIG_SMP | |
126c2092 | 5630 | if (rq->ttwu_pending) |
908a3283 TG |
5631 | return 0; |
5632 | #endif | |
5633 | ||
5634 | return 1; | |
1da177e4 LT |
5635 | } |
5636 | ||
943d355d RJ |
5637 | /** |
5638 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
5639 | * @cpu: the CPU in question. | |
5640 | * | |
5641 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
5642 | */ | |
5643 | int available_idle_cpu(int cpu) | |
5644 | { | |
5645 | if (!idle_cpu(cpu)) | |
5646 | return 0; | |
5647 | ||
247f2f6f RJ |
5648 | if (vcpu_is_preempted(cpu)) |
5649 | return 0; | |
5650 | ||
908a3283 | 5651 | return 1; |
1da177e4 LT |
5652 | } |
5653 | ||
1da177e4 | 5654 | /** |
d1ccc66d | 5655 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 5656 | * @cpu: the processor in question. |
e69f6186 | 5657 | * |
d1ccc66d | 5658 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 5659 | */ |
36c8b586 | 5660 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5661 | { |
5662 | return cpu_rq(cpu)->idle; | |
5663 | } | |
5664 | ||
5665 | /** | |
5666 | * find_process_by_pid - find a process with a matching PID value. | |
5667 | * @pid: the pid in question. | |
e69f6186 YB |
5668 | * |
5669 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 5670 | */ |
a9957449 | 5671 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5672 | { |
228ebcbe | 5673 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5674 | } |
5675 | ||
c13db6b1 SR |
5676 | /* |
5677 | * sched_setparam() passes in -1 for its policy, to let the functions | |
5678 | * it calls know not to change it. | |
5679 | */ | |
5680 | #define SETPARAM_POLICY -1 | |
5681 | ||
c365c292 TG |
5682 | static void __setscheduler_params(struct task_struct *p, |
5683 | const struct sched_attr *attr) | |
1da177e4 | 5684 | { |
d50dde5a DF |
5685 | int policy = attr->sched_policy; |
5686 | ||
c13db6b1 | 5687 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
5688 | policy = p->policy; |
5689 | ||
1da177e4 | 5690 | p->policy = policy; |
d50dde5a | 5691 | |
aab03e05 DF |
5692 | if (dl_policy(policy)) |
5693 | __setparam_dl(p, attr); | |
39fd8fd2 | 5694 | else if (fair_policy(policy)) |
d50dde5a DF |
5695 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
5696 | ||
39fd8fd2 PZ |
5697 | /* |
5698 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
5699 | * !rt_policy. Always setting this ensures that things like | |
5700 | * getparam()/getattr() don't report silly values for !rt tasks. | |
5701 | */ | |
5702 | p->rt_priority = attr->sched_priority; | |
383afd09 | 5703 | p->normal_prio = normal_prio(p); |
9059393e | 5704 | set_load_weight(p, true); |
c365c292 | 5705 | } |
39fd8fd2 | 5706 | |
c365c292 TG |
5707 | /* Actually do priority change: must hold pi & rq lock. */ |
5708 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 5709 | const struct sched_attr *attr, bool keep_boost) |
c365c292 | 5710 | { |
a509a7cd PB |
5711 | /* |
5712 | * If params can't change scheduling class changes aren't allowed | |
5713 | * either. | |
5714 | */ | |
5715 | if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) | |
5716 | return; | |
5717 | ||
c365c292 | 5718 | __setscheduler_params(p, attr); |
d50dde5a | 5719 | |
383afd09 | 5720 | /* |
0782e63b TG |
5721 | * Keep a potential priority boosting if called from |
5722 | * sched_setscheduler(). | |
383afd09 | 5723 | */ |
acd58620 | 5724 | p->prio = normal_prio(p); |
0782e63b | 5725 | if (keep_boost) |
acd58620 | 5726 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 5727 | |
aab03e05 DF |
5728 | if (dl_prio(p->prio)) |
5729 | p->sched_class = &dl_sched_class; | |
5730 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
5731 | p->sched_class = &rt_sched_class; |
5732 | else | |
5733 | p->sched_class = &fair_sched_class; | |
1da177e4 | 5734 | } |
aab03e05 | 5735 | |
c69e8d9c | 5736 | /* |
d1ccc66d | 5737 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
5738 | */ |
5739 | static bool check_same_owner(struct task_struct *p) | |
5740 | { | |
5741 | const struct cred *cred = current_cred(), *pcred; | |
5742 | bool match; | |
5743 | ||
5744 | rcu_read_lock(); | |
5745 | pcred = __task_cred(p); | |
9c806aa0 EB |
5746 | match = (uid_eq(cred->euid, pcred->euid) || |
5747 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
5748 | rcu_read_unlock(); |
5749 | return match; | |
5750 | } | |
5751 | ||
d50dde5a DF |
5752 | static int __sched_setscheduler(struct task_struct *p, |
5753 | const struct sched_attr *attr, | |
dbc7f069 | 5754 | bool user, bool pi) |
1da177e4 | 5755 | { |
383afd09 SR |
5756 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
5757 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 5758 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 5759 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 5760 | const struct sched_class *prev_class; |
565790d2 | 5761 | struct callback_head *head; |
eb580751 | 5762 | struct rq_flags rf; |
ca94c442 | 5763 | int reset_on_fork; |
7a57f32a | 5764 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 5765 | struct rq *rq; |
1da177e4 | 5766 | |
896bbb25 SRV |
5767 | /* The pi code expects interrupts enabled */ |
5768 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 5769 | recheck: |
d1ccc66d | 5770 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
5771 | if (policy < 0) { |
5772 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5773 | policy = oldpolicy = p->policy; |
ca94c442 | 5774 | } else { |
7479f3c9 | 5775 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 5776 | |
20f9cd2a | 5777 | if (!valid_policy(policy)) |
ca94c442 LP |
5778 | return -EINVAL; |
5779 | } | |
5780 | ||
794a56eb | 5781 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
5782 | return -EINVAL; |
5783 | ||
1da177e4 LT |
5784 | /* |
5785 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5786 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5787 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 5788 | */ |
0bb040a4 | 5789 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 5790 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5791 | return -EINVAL; |
aab03e05 DF |
5792 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
5793 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
5794 | return -EINVAL; |
5795 | ||
37e4ab3f OC |
5796 | /* |
5797 | * Allow unprivileged RT tasks to decrease priority: | |
5798 | */ | |
961ccddd | 5799 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 5800 | if (fair_policy(policy)) { |
d0ea0268 | 5801 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 5802 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
5803 | return -EPERM; |
5804 | } | |
5805 | ||
e05606d3 | 5806 | if (rt_policy(policy)) { |
a44702e8 ON |
5807 | unsigned long rlim_rtprio = |
5808 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 5809 | |
d1ccc66d | 5810 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
5811 | if (policy != p->policy && !rlim_rtprio) |
5812 | return -EPERM; | |
5813 | ||
d1ccc66d | 5814 | /* Can't increase priority: */ |
d50dde5a DF |
5815 | if (attr->sched_priority > p->rt_priority && |
5816 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
5817 | return -EPERM; |
5818 | } | |
c02aa73b | 5819 | |
d44753b8 JL |
5820 | /* |
5821 | * Can't set/change SCHED_DEADLINE policy at all for now | |
5822 | * (safest behavior); in the future we would like to allow | |
5823 | * unprivileged DL tasks to increase their relative deadline | |
5824 | * or reduce their runtime (both ways reducing utilization) | |
5825 | */ | |
5826 | if (dl_policy(policy)) | |
5827 | return -EPERM; | |
5828 | ||
dd41f596 | 5829 | /* |
c02aa73b DH |
5830 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5831 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5832 | */ |
1da1843f | 5833 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 5834 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
5835 | return -EPERM; |
5836 | } | |
5fe1d75f | 5837 | |
d1ccc66d | 5838 | /* Can't change other user's priorities: */ |
c69e8d9c | 5839 | if (!check_same_owner(p)) |
37e4ab3f | 5840 | return -EPERM; |
ca94c442 | 5841 | |
d1ccc66d | 5842 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
5843 | if (p->sched_reset_on_fork && !reset_on_fork) |
5844 | return -EPERM; | |
37e4ab3f | 5845 | } |
1da177e4 | 5846 | |
725aad24 | 5847 | if (user) { |
794a56eb JL |
5848 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
5849 | return -EINVAL; | |
5850 | ||
b0ae1981 | 5851 | retval = security_task_setscheduler(p); |
725aad24 JF |
5852 | if (retval) |
5853 | return retval; | |
5854 | } | |
5855 | ||
a509a7cd PB |
5856 | /* Update task specific "requested" clamps */ |
5857 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
5858 | retval = uclamp_validate(p, attr); | |
5859 | if (retval) | |
5860 | return retval; | |
5861 | } | |
5862 | ||
710da3c8 JL |
5863 | if (pi) |
5864 | cpuset_read_lock(); | |
5865 | ||
b29739f9 | 5866 | /* |
d1ccc66d | 5867 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 5868 | * changing the priority of the task: |
0122ec5b | 5869 | * |
25985edc | 5870 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5871 | * runqueue lock must be held. |
5872 | */ | |
eb580751 | 5873 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 5874 | update_rq_clock(rq); |
dc61b1d6 | 5875 | |
34f971f6 | 5876 | /* |
d1ccc66d | 5877 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
5878 | */ |
5879 | if (p == rq->stop) { | |
4b211f2b MP |
5880 | retval = -EINVAL; |
5881 | goto unlock; | |
34f971f6 PZ |
5882 | } |
5883 | ||
a51e9198 | 5884 | /* |
d6b1e911 TG |
5885 | * If not changing anything there's no need to proceed further, |
5886 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 5887 | */ |
d50dde5a | 5888 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 5889 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
5890 | goto change; |
5891 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
5892 | goto change; | |
75381608 | 5893 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 5894 | goto change; |
a509a7cd PB |
5895 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
5896 | goto change; | |
d50dde5a | 5897 | |
d6b1e911 | 5898 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
5899 | retval = 0; |
5900 | goto unlock; | |
a51e9198 | 5901 | } |
d50dde5a | 5902 | change: |
a51e9198 | 5903 | |
dc61b1d6 | 5904 | if (user) { |
332ac17e | 5905 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
5906 | /* |
5907 | * Do not allow realtime tasks into groups that have no runtime | |
5908 | * assigned. | |
5909 | */ | |
5910 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5911 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5912 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
5913 | retval = -EPERM; |
5914 | goto unlock; | |
dc61b1d6 | 5915 | } |
dc61b1d6 | 5916 | #endif |
332ac17e | 5917 | #ifdef CONFIG_SMP |
794a56eb JL |
5918 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
5919 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 5920 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
5921 | |
5922 | /* | |
5923 | * Don't allow tasks with an affinity mask smaller than | |
5924 | * the entire root_domain to become SCHED_DEADLINE. We | |
5925 | * will also fail if there's no bandwidth available. | |
5926 | */ | |
3bd37062 | 5927 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 5928 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
5929 | retval = -EPERM; |
5930 | goto unlock; | |
332ac17e DF |
5931 | } |
5932 | } | |
5933 | #endif | |
5934 | } | |
dc61b1d6 | 5935 | |
d1ccc66d | 5936 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
5937 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
5938 | policy = oldpolicy = -1; | |
eb580751 | 5939 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
5940 | if (pi) |
5941 | cpuset_read_unlock(); | |
1da177e4 LT |
5942 | goto recheck; |
5943 | } | |
332ac17e DF |
5944 | |
5945 | /* | |
5946 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
5947 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
5948 | * is available. | |
5949 | */ | |
06a76fe0 | 5950 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
5951 | retval = -EBUSY; |
5952 | goto unlock; | |
332ac17e DF |
5953 | } |
5954 | ||
c365c292 TG |
5955 | p->sched_reset_on_fork = reset_on_fork; |
5956 | oldprio = p->prio; | |
5957 | ||
dbc7f069 PZ |
5958 | if (pi) { |
5959 | /* | |
5960 | * Take priority boosted tasks into account. If the new | |
5961 | * effective priority is unchanged, we just store the new | |
5962 | * normal parameters and do not touch the scheduler class and | |
5963 | * the runqueue. This will be done when the task deboost | |
5964 | * itself. | |
5965 | */ | |
acd58620 | 5966 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
5967 | if (new_effective_prio == oldprio) |
5968 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
5969 | } |
5970 | ||
da0c1e65 | 5971 | queued = task_on_rq_queued(p); |
051a1d1a | 5972 | running = task_current(rq, p); |
da0c1e65 | 5973 | if (queued) |
ff77e468 | 5974 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 5975 | if (running) |
f3cd1c4e | 5976 | put_prev_task(rq, p); |
f6b53205 | 5977 | |
83ab0aa0 | 5978 | prev_class = p->sched_class; |
a509a7cd | 5979 | |
dbc7f069 | 5980 | __setscheduler(rq, p, attr, pi); |
a509a7cd | 5981 | __setscheduler_uclamp(p, attr); |
f6b53205 | 5982 | |
da0c1e65 | 5983 | if (queued) { |
81a44c54 TG |
5984 | /* |
5985 | * We enqueue to tail when the priority of a task is | |
5986 | * increased (user space view). | |
5987 | */ | |
ff77e468 PZ |
5988 | if (oldprio < p->prio) |
5989 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 5990 | |
ff77e468 | 5991 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 5992 | } |
a399d233 | 5993 | if (running) |
03b7fad1 | 5994 | set_next_task(rq, p); |
cb469845 | 5995 | |
da7a735e | 5996 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
5997 | |
5998 | /* Avoid rq from going away on us: */ | |
5999 | preempt_disable(); | |
565790d2 | 6000 | head = splice_balance_callbacks(rq); |
eb580751 | 6001 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 6002 | |
710da3c8 JL |
6003 | if (pi) { |
6004 | cpuset_read_unlock(); | |
dbc7f069 | 6005 | rt_mutex_adjust_pi(p); |
710da3c8 | 6006 | } |
95e02ca9 | 6007 | |
d1ccc66d | 6008 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 6009 | balance_callbacks(rq, head); |
4c9a4bc8 | 6010 | preempt_enable(); |
95e02ca9 | 6011 | |
1da177e4 | 6012 | return 0; |
4b211f2b MP |
6013 | |
6014 | unlock: | |
6015 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
6016 | if (pi) |
6017 | cpuset_read_unlock(); | |
4b211f2b | 6018 | return retval; |
1da177e4 | 6019 | } |
961ccddd | 6020 | |
7479f3c9 PZ |
6021 | static int _sched_setscheduler(struct task_struct *p, int policy, |
6022 | const struct sched_param *param, bool check) | |
6023 | { | |
6024 | struct sched_attr attr = { | |
6025 | .sched_policy = policy, | |
6026 | .sched_priority = param->sched_priority, | |
6027 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
6028 | }; | |
6029 | ||
c13db6b1 SR |
6030 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
6031 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
6032 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
6033 | policy &= ~SCHED_RESET_ON_FORK; | |
6034 | attr.sched_policy = policy; | |
6035 | } | |
6036 | ||
dbc7f069 | 6037 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 6038 | } |
961ccddd RR |
6039 | /** |
6040 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6041 | * @p: the task in question. | |
6042 | * @policy: new policy. | |
6043 | * @param: structure containing the new RT priority. | |
6044 | * | |
7318d4cc PZ |
6045 | * Use sched_set_fifo(), read its comment. |
6046 | * | |
e69f6186 YB |
6047 | * Return: 0 on success. An error code otherwise. |
6048 | * | |
961ccddd RR |
6049 | * NOTE that the task may be already dead. |
6050 | */ | |
6051 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 6052 | const struct sched_param *param) |
961ccddd | 6053 | { |
7479f3c9 | 6054 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 6055 | } |
1da177e4 | 6056 | |
d50dde5a DF |
6057 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
6058 | { | |
dbc7f069 | 6059 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 6060 | } |
d50dde5a | 6061 | |
794a56eb JL |
6062 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
6063 | { | |
6064 | return __sched_setscheduler(p, attr, false, true); | |
6065 | } | |
6066 | ||
961ccddd RR |
6067 | /** |
6068 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6069 | * @p: the task in question. | |
6070 | * @policy: new policy. | |
6071 | * @param: structure containing the new RT priority. | |
6072 | * | |
6073 | * Just like sched_setscheduler, only don't bother checking if the | |
6074 | * current context has permission. For example, this is needed in | |
6075 | * stop_machine(): we create temporary high priority worker threads, | |
6076 | * but our caller might not have that capability. | |
e69f6186 YB |
6077 | * |
6078 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
6079 | */ |
6080 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 6081 | const struct sched_param *param) |
961ccddd | 6082 | { |
7479f3c9 | 6083 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
6084 | } |
6085 | ||
7318d4cc PZ |
6086 | /* |
6087 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
6088 | * incapable of resource management, which is the one thing an OS really should | |
6089 | * be doing. | |
6090 | * | |
6091 | * This is of course the reason it is limited to privileged users only. | |
6092 | * | |
6093 | * Worse still; it is fundamentally impossible to compose static priority | |
6094 | * workloads. You cannot take two correctly working static prio workloads | |
6095 | * and smash them together and still expect them to work. | |
6096 | * | |
6097 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
6098 | * | |
6099 | * MAX_RT_PRIO / 2 | |
6100 | * | |
6101 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
6102 | * know enough information to make a sensible choice. | |
6103 | */ | |
8b700983 | 6104 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
6105 | { |
6106 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 6107 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6108 | } |
6109 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
6110 | ||
6111 | /* | |
6112 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
6113 | */ | |
8b700983 | 6114 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
6115 | { |
6116 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 6117 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6118 | } |
6119 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
6120 | ||
8b700983 | 6121 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
6122 | { |
6123 | struct sched_attr attr = { | |
6124 | .sched_policy = SCHED_NORMAL, | |
6125 | .sched_nice = nice, | |
6126 | }; | |
8b700983 | 6127 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
6128 | } |
6129 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 6130 | |
95cdf3b7 IM |
6131 | static int |
6132 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6133 | { |
1da177e4 LT |
6134 | struct sched_param lparam; |
6135 | struct task_struct *p; | |
36c8b586 | 6136 | int retval; |
1da177e4 LT |
6137 | |
6138 | if (!param || pid < 0) | |
6139 | return -EINVAL; | |
6140 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6141 | return -EFAULT; | |
5fe1d75f ON |
6142 | |
6143 | rcu_read_lock(); | |
6144 | retval = -ESRCH; | |
1da177e4 | 6145 | p = find_process_by_pid(pid); |
710da3c8 JL |
6146 | if (likely(p)) |
6147 | get_task_struct(p); | |
5fe1d75f | 6148 | rcu_read_unlock(); |
36c8b586 | 6149 | |
710da3c8 JL |
6150 | if (likely(p)) { |
6151 | retval = sched_setscheduler(p, policy, &lparam); | |
6152 | put_task_struct(p); | |
6153 | } | |
6154 | ||
1da177e4 LT |
6155 | return retval; |
6156 | } | |
6157 | ||
d50dde5a DF |
6158 | /* |
6159 | * Mimics kernel/events/core.c perf_copy_attr(). | |
6160 | */ | |
d1ccc66d | 6161 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
6162 | { |
6163 | u32 size; | |
6164 | int ret; | |
6165 | ||
d1ccc66d | 6166 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
6167 | memset(attr, 0, sizeof(*attr)); |
6168 | ||
6169 | ret = get_user(size, &uattr->size); | |
6170 | if (ret) | |
6171 | return ret; | |
6172 | ||
d1ccc66d IM |
6173 | /* ABI compatibility quirk: */ |
6174 | if (!size) | |
d50dde5a | 6175 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 6176 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
6177 | goto err_size; |
6178 | ||
dff3a85f AS |
6179 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
6180 | if (ret) { | |
6181 | if (ret == -E2BIG) | |
6182 | goto err_size; | |
6183 | return ret; | |
d50dde5a DF |
6184 | } |
6185 | ||
a509a7cd PB |
6186 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
6187 | size < SCHED_ATTR_SIZE_VER1) | |
6188 | return -EINVAL; | |
6189 | ||
d50dde5a | 6190 | /* |
d1ccc66d | 6191 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
6192 | * to be strict and return an error on out-of-bounds values? |
6193 | */ | |
75e45d51 | 6194 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 6195 | |
e78c7bca | 6196 | return 0; |
d50dde5a DF |
6197 | |
6198 | err_size: | |
6199 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 6200 | return -E2BIG; |
d50dde5a DF |
6201 | } |
6202 | ||
1da177e4 LT |
6203 | /** |
6204 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6205 | * @pid: the pid in question. | |
6206 | * @policy: new policy. | |
6207 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6208 | * |
6209 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6210 | */ |
d1ccc66d | 6211 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 6212 | { |
c21761f1 JB |
6213 | if (policy < 0) |
6214 | return -EINVAL; | |
6215 | ||
1da177e4 LT |
6216 | return do_sched_setscheduler(pid, policy, param); |
6217 | } | |
6218 | ||
6219 | /** | |
6220 | * sys_sched_setparam - set/change the RT priority of a thread | |
6221 | * @pid: the pid in question. | |
6222 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6223 | * |
6224 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6225 | */ |
5add95d4 | 6226 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6227 | { |
c13db6b1 | 6228 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
6229 | } |
6230 | ||
d50dde5a DF |
6231 | /** |
6232 | * sys_sched_setattr - same as above, but with extended sched_attr | |
6233 | * @pid: the pid in question. | |
5778fccf | 6234 | * @uattr: structure containing the extended parameters. |
db66d756 | 6235 | * @flags: for future extension. |
d50dde5a | 6236 | */ |
6d35ab48 PZ |
6237 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
6238 | unsigned int, flags) | |
d50dde5a DF |
6239 | { |
6240 | struct sched_attr attr; | |
6241 | struct task_struct *p; | |
6242 | int retval; | |
6243 | ||
6d35ab48 | 6244 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
6245 | return -EINVAL; |
6246 | ||
143cf23d MK |
6247 | retval = sched_copy_attr(uattr, &attr); |
6248 | if (retval) | |
6249 | return retval; | |
d50dde5a | 6250 | |
b14ed2c2 | 6251 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 6252 | return -EINVAL; |
1d6362fa PB |
6253 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
6254 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
6255 | |
6256 | rcu_read_lock(); | |
6257 | retval = -ESRCH; | |
6258 | p = find_process_by_pid(pid); | |
a509a7cd PB |
6259 | if (likely(p)) |
6260 | get_task_struct(p); | |
d50dde5a DF |
6261 | rcu_read_unlock(); |
6262 | ||
a509a7cd PB |
6263 | if (likely(p)) { |
6264 | retval = sched_setattr(p, &attr); | |
6265 | put_task_struct(p); | |
6266 | } | |
6267 | ||
d50dde5a DF |
6268 | return retval; |
6269 | } | |
6270 | ||
1da177e4 LT |
6271 | /** |
6272 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6273 | * @pid: the pid in question. | |
e69f6186 YB |
6274 | * |
6275 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
6276 | * code. | |
1da177e4 | 6277 | */ |
5add95d4 | 6278 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6279 | { |
36c8b586 | 6280 | struct task_struct *p; |
3a5c359a | 6281 | int retval; |
1da177e4 LT |
6282 | |
6283 | if (pid < 0) | |
3a5c359a | 6284 | return -EINVAL; |
1da177e4 LT |
6285 | |
6286 | retval = -ESRCH; | |
5fe85be0 | 6287 | rcu_read_lock(); |
1da177e4 LT |
6288 | p = find_process_by_pid(pid); |
6289 | if (p) { | |
6290 | retval = security_task_getscheduler(p); | |
6291 | if (!retval) | |
ca94c442 LP |
6292 | retval = p->policy |
6293 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 6294 | } |
5fe85be0 | 6295 | rcu_read_unlock(); |
1da177e4 LT |
6296 | return retval; |
6297 | } | |
6298 | ||
6299 | /** | |
ca94c442 | 6300 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6301 | * @pid: the pid in question. |
6302 | * @param: structure containing the RT priority. | |
e69f6186 YB |
6303 | * |
6304 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
6305 | * code. | |
1da177e4 | 6306 | */ |
5add95d4 | 6307 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6308 | { |
ce5f7f82 | 6309 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 6310 | struct task_struct *p; |
3a5c359a | 6311 | int retval; |
1da177e4 LT |
6312 | |
6313 | if (!param || pid < 0) | |
3a5c359a | 6314 | return -EINVAL; |
1da177e4 | 6315 | |
5fe85be0 | 6316 | rcu_read_lock(); |
1da177e4 LT |
6317 | p = find_process_by_pid(pid); |
6318 | retval = -ESRCH; | |
6319 | if (!p) | |
6320 | goto out_unlock; | |
6321 | ||
6322 | retval = security_task_getscheduler(p); | |
6323 | if (retval) | |
6324 | goto out_unlock; | |
6325 | ||
ce5f7f82 PZ |
6326 | if (task_has_rt_policy(p)) |
6327 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 6328 | rcu_read_unlock(); |
1da177e4 LT |
6329 | |
6330 | /* | |
6331 | * This one might sleep, we cannot do it with a spinlock held ... | |
6332 | */ | |
6333 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6334 | ||
1da177e4 LT |
6335 | return retval; |
6336 | ||
6337 | out_unlock: | |
5fe85be0 | 6338 | rcu_read_unlock(); |
1da177e4 LT |
6339 | return retval; |
6340 | } | |
6341 | ||
1251201c IM |
6342 | /* |
6343 | * Copy the kernel size attribute structure (which might be larger | |
6344 | * than what user-space knows about) to user-space. | |
6345 | * | |
6346 | * Note that all cases are valid: user-space buffer can be larger or | |
6347 | * smaller than the kernel-space buffer. The usual case is that both | |
6348 | * have the same size. | |
6349 | */ | |
6350 | static int | |
6351 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
6352 | struct sched_attr *kattr, | |
6353 | unsigned int usize) | |
d50dde5a | 6354 | { |
1251201c | 6355 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 6356 | |
96d4f267 | 6357 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
6358 | return -EFAULT; |
6359 | ||
6360 | /* | |
1251201c IM |
6361 | * sched_getattr() ABI forwards and backwards compatibility: |
6362 | * | |
6363 | * If usize == ksize then we just copy everything to user-space and all is good. | |
6364 | * | |
6365 | * If usize < ksize then we only copy as much as user-space has space for, | |
6366 | * this keeps ABI compatibility as well. We skip the rest. | |
6367 | * | |
6368 | * If usize > ksize then user-space is using a newer version of the ABI, | |
6369 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
6370 | * detect the kernel's knowledge of attributes from the attr->size value | |
6371 | * which is set to ksize in this case. | |
d50dde5a | 6372 | */ |
1251201c | 6373 | kattr->size = min(usize, ksize); |
d50dde5a | 6374 | |
1251201c | 6375 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
6376 | return -EFAULT; |
6377 | ||
22400674 | 6378 | return 0; |
d50dde5a DF |
6379 | } |
6380 | ||
6381 | /** | |
aab03e05 | 6382 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 6383 | * @pid: the pid in question. |
5778fccf | 6384 | * @uattr: structure containing the extended parameters. |
dff3a85f | 6385 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 6386 | * @flags: for future extension. |
d50dde5a | 6387 | */ |
6d35ab48 | 6388 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 6389 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 6390 | { |
1251201c | 6391 | struct sched_attr kattr = { }; |
d50dde5a DF |
6392 | struct task_struct *p; |
6393 | int retval; | |
6394 | ||
1251201c IM |
6395 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
6396 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
6397 | return -EINVAL; |
6398 | ||
6399 | rcu_read_lock(); | |
6400 | p = find_process_by_pid(pid); | |
6401 | retval = -ESRCH; | |
6402 | if (!p) | |
6403 | goto out_unlock; | |
6404 | ||
6405 | retval = security_task_getscheduler(p); | |
6406 | if (retval) | |
6407 | goto out_unlock; | |
6408 | ||
1251201c | 6409 | kattr.sched_policy = p->policy; |
7479f3c9 | 6410 | if (p->sched_reset_on_fork) |
1251201c | 6411 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
aab03e05 | 6412 | if (task_has_dl_policy(p)) |
1251201c | 6413 | __getparam_dl(p, &kattr); |
aab03e05 | 6414 | else if (task_has_rt_policy(p)) |
1251201c | 6415 | kattr.sched_priority = p->rt_priority; |
d50dde5a | 6416 | else |
1251201c | 6417 | kattr.sched_nice = task_nice(p); |
d50dde5a | 6418 | |
a509a7cd | 6419 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
6420 | /* |
6421 | * This could race with another potential updater, but this is fine | |
6422 | * because it'll correctly read the old or the new value. We don't need | |
6423 | * to guarantee who wins the race as long as it doesn't return garbage. | |
6424 | */ | |
1251201c IM |
6425 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
6426 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
6427 | #endif |
6428 | ||
d50dde5a DF |
6429 | rcu_read_unlock(); |
6430 | ||
1251201c | 6431 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
6432 | |
6433 | out_unlock: | |
6434 | rcu_read_unlock(); | |
6435 | return retval; | |
6436 | } | |
6437 | ||
96f874e2 | 6438 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6439 | { |
5a16f3d3 | 6440 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6441 | struct task_struct *p; |
6442 | int retval; | |
1da177e4 | 6443 | |
23f5d142 | 6444 | rcu_read_lock(); |
1da177e4 LT |
6445 | |
6446 | p = find_process_by_pid(pid); | |
6447 | if (!p) { | |
23f5d142 | 6448 | rcu_read_unlock(); |
1da177e4 LT |
6449 | return -ESRCH; |
6450 | } | |
6451 | ||
23f5d142 | 6452 | /* Prevent p going away */ |
1da177e4 | 6453 | get_task_struct(p); |
23f5d142 | 6454 | rcu_read_unlock(); |
1da177e4 | 6455 | |
14a40ffc TH |
6456 | if (p->flags & PF_NO_SETAFFINITY) { |
6457 | retval = -EINVAL; | |
6458 | goto out_put_task; | |
6459 | } | |
5a16f3d3 RR |
6460 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6461 | retval = -ENOMEM; | |
6462 | goto out_put_task; | |
6463 | } | |
6464 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6465 | retval = -ENOMEM; | |
6466 | goto out_free_cpus_allowed; | |
6467 | } | |
1da177e4 | 6468 | retval = -EPERM; |
4c44aaaf EB |
6469 | if (!check_same_owner(p)) { |
6470 | rcu_read_lock(); | |
6471 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
6472 | rcu_read_unlock(); | |
16303ab2 | 6473 | goto out_free_new_mask; |
4c44aaaf EB |
6474 | } |
6475 | rcu_read_unlock(); | |
6476 | } | |
1da177e4 | 6477 | |
b0ae1981 | 6478 | retval = security_task_setscheduler(p); |
e7834f8f | 6479 | if (retval) |
16303ab2 | 6480 | goto out_free_new_mask; |
e7834f8f | 6481 | |
e4099a5e PZ |
6482 | |
6483 | cpuset_cpus_allowed(p, cpus_allowed); | |
6484 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
6485 | ||
332ac17e DF |
6486 | /* |
6487 | * Since bandwidth control happens on root_domain basis, | |
6488 | * if admission test is enabled, we only admit -deadline | |
6489 | * tasks allowed to run on all the CPUs in the task's | |
6490 | * root_domain. | |
6491 | */ | |
6492 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
6493 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
6494 | rcu_read_lock(); | |
6495 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 6496 | retval = -EBUSY; |
f1e3a093 | 6497 | rcu_read_unlock(); |
16303ab2 | 6498 | goto out_free_new_mask; |
332ac17e | 6499 | } |
f1e3a093 | 6500 | rcu_read_unlock(); |
332ac17e DF |
6501 | } |
6502 | #endif | |
49246274 | 6503 | again: |
9cfc3e18 | 6504 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK); |
1da177e4 | 6505 | |
8707d8b8 | 6506 | if (!retval) { |
5a16f3d3 RR |
6507 | cpuset_cpus_allowed(p, cpus_allowed); |
6508 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6509 | /* |
6510 | * We must have raced with a concurrent cpuset | |
6511 | * update. Just reset the cpus_allowed to the | |
6512 | * cpuset's cpus_allowed | |
6513 | */ | |
5a16f3d3 | 6514 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6515 | goto again; |
6516 | } | |
6517 | } | |
16303ab2 | 6518 | out_free_new_mask: |
5a16f3d3 RR |
6519 | free_cpumask_var(new_mask); |
6520 | out_free_cpus_allowed: | |
6521 | free_cpumask_var(cpus_allowed); | |
6522 | out_put_task: | |
1da177e4 | 6523 | put_task_struct(p); |
1da177e4 LT |
6524 | return retval; |
6525 | } | |
6526 | ||
6527 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6528 | struct cpumask *new_mask) |
1da177e4 | 6529 | { |
96f874e2 RR |
6530 | if (len < cpumask_size()) |
6531 | cpumask_clear(new_mask); | |
6532 | else if (len > cpumask_size()) | |
6533 | len = cpumask_size(); | |
6534 | ||
1da177e4 LT |
6535 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6536 | } | |
6537 | ||
6538 | /** | |
d1ccc66d | 6539 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
6540 | * @pid: pid of the process |
6541 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6542 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
6543 | * |
6544 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6545 | */ |
5add95d4 HC |
6546 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6547 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6548 | { |
5a16f3d3 | 6549 | cpumask_var_t new_mask; |
1da177e4 LT |
6550 | int retval; |
6551 | ||
5a16f3d3 RR |
6552 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6553 | return -ENOMEM; | |
1da177e4 | 6554 | |
5a16f3d3 RR |
6555 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6556 | if (retval == 0) | |
6557 | retval = sched_setaffinity(pid, new_mask); | |
6558 | free_cpumask_var(new_mask); | |
6559 | return retval; | |
1da177e4 LT |
6560 | } |
6561 | ||
96f874e2 | 6562 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6563 | { |
36c8b586 | 6564 | struct task_struct *p; |
31605683 | 6565 | unsigned long flags; |
1da177e4 | 6566 | int retval; |
1da177e4 | 6567 | |
23f5d142 | 6568 | rcu_read_lock(); |
1da177e4 LT |
6569 | |
6570 | retval = -ESRCH; | |
6571 | p = find_process_by_pid(pid); | |
6572 | if (!p) | |
6573 | goto out_unlock; | |
6574 | ||
e7834f8f DQ |
6575 | retval = security_task_getscheduler(p); |
6576 | if (retval) | |
6577 | goto out_unlock; | |
6578 | ||
013fdb80 | 6579 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 6580 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 6581 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
6582 | |
6583 | out_unlock: | |
23f5d142 | 6584 | rcu_read_unlock(); |
1da177e4 | 6585 | |
9531b62f | 6586 | return retval; |
1da177e4 LT |
6587 | } |
6588 | ||
6589 | /** | |
d1ccc66d | 6590 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
6591 | * @pid: pid of the process |
6592 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6593 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 6594 | * |
599b4840 ZW |
6595 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
6596 | * error code otherwise. | |
1da177e4 | 6597 | */ |
5add95d4 HC |
6598 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6599 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6600 | { |
6601 | int ret; | |
f17c8607 | 6602 | cpumask_var_t mask; |
1da177e4 | 6603 | |
84fba5ec | 6604 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
6605 | return -EINVAL; |
6606 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
6607 | return -EINVAL; |
6608 | ||
f17c8607 RR |
6609 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6610 | return -ENOMEM; | |
1da177e4 | 6611 | |
f17c8607 RR |
6612 | ret = sched_getaffinity(pid, mask); |
6613 | if (ret == 0) { | |
4de373a1 | 6614 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
6615 | |
6616 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
6617 | ret = -EFAULT; |
6618 | else | |
cd3d8031 | 6619 | ret = retlen; |
f17c8607 RR |
6620 | } |
6621 | free_cpumask_var(mask); | |
1da177e4 | 6622 | |
f17c8607 | 6623 | return ret; |
1da177e4 LT |
6624 | } |
6625 | ||
7d4dd4f1 | 6626 | static void do_sched_yield(void) |
1da177e4 | 6627 | { |
8a8c69c3 PZ |
6628 | struct rq_flags rf; |
6629 | struct rq *rq; | |
6630 | ||
246b3b33 | 6631 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 6632 | |
ae92882e | 6633 | schedstat_inc(rq->yld_count); |
4530d7ab | 6634 | current->sched_class->yield_task(rq); |
1da177e4 | 6635 | |
8a8c69c3 | 6636 | preempt_disable(); |
345a957f | 6637 | rq_unlock_irq(rq, &rf); |
ba74c144 | 6638 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
6639 | |
6640 | schedule(); | |
7d4dd4f1 | 6641 | } |
1da177e4 | 6642 | |
59a74b15 MCC |
6643 | /** |
6644 | * sys_sched_yield - yield the current processor to other threads. | |
6645 | * | |
6646 | * This function yields the current CPU to other tasks. If there are no | |
6647 | * other threads running on this CPU then this function will return. | |
6648 | * | |
6649 | * Return: 0. | |
6650 | */ | |
7d4dd4f1 DB |
6651 | SYSCALL_DEFINE0(sched_yield) |
6652 | { | |
6653 | do_sched_yield(); | |
1da177e4 LT |
6654 | return 0; |
6655 | } | |
6656 | ||
c1a280b6 | 6657 | #ifndef CONFIG_PREEMPTION |
02b67cc3 | 6658 | int __sched _cond_resched(void) |
1da177e4 | 6659 | { |
fe32d3cd | 6660 | if (should_resched(0)) { |
a18b5d01 | 6661 | preempt_schedule_common(); |
1da177e4 LT |
6662 | return 1; |
6663 | } | |
f79c3ad6 | 6664 | rcu_all_qs(); |
1da177e4 LT |
6665 | return 0; |
6666 | } | |
02b67cc3 | 6667 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 6668 | #endif |
1da177e4 LT |
6669 | |
6670 | /* | |
613afbf8 | 6671 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6672 | * call schedule, and on return reacquire the lock. |
6673 | * | |
c1a280b6 | 6674 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
6675 | * operations here to prevent schedule() from being called twice (once via |
6676 | * spin_unlock(), once by hand). | |
6677 | */ | |
613afbf8 | 6678 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6679 | { |
fe32d3cd | 6680 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
6681 | int ret = 0; |
6682 | ||
f607c668 PZ |
6683 | lockdep_assert_held(lock); |
6684 | ||
4a81e832 | 6685 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6686 | spin_unlock(lock); |
d86ee480 | 6687 | if (resched) |
a18b5d01 | 6688 | preempt_schedule_common(); |
95c354fe NP |
6689 | else |
6690 | cpu_relax(); | |
6df3cecb | 6691 | ret = 1; |
1da177e4 | 6692 | spin_lock(lock); |
1da177e4 | 6693 | } |
6df3cecb | 6694 | return ret; |
1da177e4 | 6695 | } |
613afbf8 | 6696 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6697 | |
1da177e4 LT |
6698 | /** |
6699 | * yield - yield the current processor to other threads. | |
6700 | * | |
8e3fabfd PZ |
6701 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
6702 | * | |
6703 | * The scheduler is at all times free to pick the calling task as the most | |
6704 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 6705 | * it, it's already broken. |
8e3fabfd PZ |
6706 | * |
6707 | * Typical broken usage is: | |
6708 | * | |
6709 | * while (!event) | |
d1ccc66d | 6710 | * yield(); |
8e3fabfd PZ |
6711 | * |
6712 | * where one assumes that yield() will let 'the other' process run that will | |
6713 | * make event true. If the current task is a SCHED_FIFO task that will never | |
6714 | * happen. Never use yield() as a progress guarantee!! | |
6715 | * | |
6716 | * If you want to use yield() to wait for something, use wait_event(). | |
6717 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
6718 | * If you still want to use yield(), do not! | |
1da177e4 LT |
6719 | */ |
6720 | void __sched yield(void) | |
6721 | { | |
6722 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 6723 | do_sched_yield(); |
1da177e4 | 6724 | } |
1da177e4 LT |
6725 | EXPORT_SYMBOL(yield); |
6726 | ||
d95f4122 MG |
6727 | /** |
6728 | * yield_to - yield the current processor to another thread in | |
6729 | * your thread group, or accelerate that thread toward the | |
6730 | * processor it's on. | |
16addf95 RD |
6731 | * @p: target task |
6732 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
6733 | * |
6734 | * It's the caller's job to ensure that the target task struct | |
6735 | * can't go away on us before we can do any checks. | |
6736 | * | |
e69f6186 | 6737 | * Return: |
7b270f60 PZ |
6738 | * true (>0) if we indeed boosted the target task. |
6739 | * false (0) if we failed to boost the target. | |
6740 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 6741 | */ |
fa93384f | 6742 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
6743 | { |
6744 | struct task_struct *curr = current; | |
6745 | struct rq *rq, *p_rq; | |
6746 | unsigned long flags; | |
c3c18640 | 6747 | int yielded = 0; |
d95f4122 MG |
6748 | |
6749 | local_irq_save(flags); | |
6750 | rq = this_rq(); | |
6751 | ||
6752 | again: | |
6753 | p_rq = task_rq(p); | |
7b270f60 PZ |
6754 | /* |
6755 | * If we're the only runnable task on the rq and target rq also | |
6756 | * has only one task, there's absolutely no point in yielding. | |
6757 | */ | |
6758 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
6759 | yielded = -ESRCH; | |
6760 | goto out_irq; | |
6761 | } | |
6762 | ||
d95f4122 | 6763 | double_rq_lock(rq, p_rq); |
39e24d8f | 6764 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
6765 | double_rq_unlock(rq, p_rq); |
6766 | goto again; | |
6767 | } | |
6768 | ||
6769 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 6770 | goto out_unlock; |
d95f4122 MG |
6771 | |
6772 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 6773 | goto out_unlock; |
d95f4122 MG |
6774 | |
6775 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 6776 | goto out_unlock; |
d95f4122 | 6777 | |
0900acf2 | 6778 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 6779 | if (yielded) { |
ae92882e | 6780 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
6781 | /* |
6782 | * Make p's CPU reschedule; pick_next_entity takes care of | |
6783 | * fairness. | |
6784 | */ | |
6785 | if (preempt && rq != p_rq) | |
8875125e | 6786 | resched_curr(p_rq); |
6d1cafd8 | 6787 | } |
d95f4122 | 6788 | |
7b270f60 | 6789 | out_unlock: |
d95f4122 | 6790 | double_rq_unlock(rq, p_rq); |
7b270f60 | 6791 | out_irq: |
d95f4122 MG |
6792 | local_irq_restore(flags); |
6793 | ||
7b270f60 | 6794 | if (yielded > 0) |
d95f4122 MG |
6795 | schedule(); |
6796 | ||
6797 | return yielded; | |
6798 | } | |
6799 | EXPORT_SYMBOL_GPL(yield_to); | |
6800 | ||
10ab5643 TH |
6801 | int io_schedule_prepare(void) |
6802 | { | |
6803 | int old_iowait = current->in_iowait; | |
6804 | ||
6805 | current->in_iowait = 1; | |
6806 | blk_schedule_flush_plug(current); | |
6807 | ||
6808 | return old_iowait; | |
6809 | } | |
6810 | ||
6811 | void io_schedule_finish(int token) | |
6812 | { | |
6813 | current->in_iowait = token; | |
6814 | } | |
6815 | ||
1da177e4 | 6816 | /* |
41a2d6cf | 6817 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6818 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 6819 | */ |
1da177e4 LT |
6820 | long __sched io_schedule_timeout(long timeout) |
6821 | { | |
10ab5643 | 6822 | int token; |
1da177e4 LT |
6823 | long ret; |
6824 | ||
10ab5643 | 6825 | token = io_schedule_prepare(); |
1da177e4 | 6826 | ret = schedule_timeout(timeout); |
10ab5643 | 6827 | io_schedule_finish(token); |
9cff8ade | 6828 | |
1da177e4 LT |
6829 | return ret; |
6830 | } | |
9cff8ade | 6831 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 6832 | |
e3b929b0 | 6833 | void __sched io_schedule(void) |
10ab5643 TH |
6834 | { |
6835 | int token; | |
6836 | ||
6837 | token = io_schedule_prepare(); | |
6838 | schedule(); | |
6839 | io_schedule_finish(token); | |
6840 | } | |
6841 | EXPORT_SYMBOL(io_schedule); | |
6842 | ||
1da177e4 LT |
6843 | /** |
6844 | * sys_sched_get_priority_max - return maximum RT priority. | |
6845 | * @policy: scheduling class. | |
6846 | * | |
e69f6186 YB |
6847 | * Return: On success, this syscall returns the maximum |
6848 | * rt_priority that can be used by a given scheduling class. | |
6849 | * On failure, a negative error code is returned. | |
1da177e4 | 6850 | */ |
5add95d4 | 6851 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6852 | { |
6853 | int ret = -EINVAL; | |
6854 | ||
6855 | switch (policy) { | |
6856 | case SCHED_FIFO: | |
6857 | case SCHED_RR: | |
6858 | ret = MAX_USER_RT_PRIO-1; | |
6859 | break; | |
aab03e05 | 6860 | case SCHED_DEADLINE: |
1da177e4 | 6861 | case SCHED_NORMAL: |
b0a9499c | 6862 | case SCHED_BATCH: |
dd41f596 | 6863 | case SCHED_IDLE: |
1da177e4 LT |
6864 | ret = 0; |
6865 | break; | |
6866 | } | |
6867 | return ret; | |
6868 | } | |
6869 | ||
6870 | /** | |
6871 | * sys_sched_get_priority_min - return minimum RT priority. | |
6872 | * @policy: scheduling class. | |
6873 | * | |
e69f6186 YB |
6874 | * Return: On success, this syscall returns the minimum |
6875 | * rt_priority that can be used by a given scheduling class. | |
6876 | * On failure, a negative error code is returned. | |
1da177e4 | 6877 | */ |
5add95d4 | 6878 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6879 | { |
6880 | int ret = -EINVAL; | |
6881 | ||
6882 | switch (policy) { | |
6883 | case SCHED_FIFO: | |
6884 | case SCHED_RR: | |
6885 | ret = 1; | |
6886 | break; | |
aab03e05 | 6887 | case SCHED_DEADLINE: |
1da177e4 | 6888 | case SCHED_NORMAL: |
b0a9499c | 6889 | case SCHED_BATCH: |
dd41f596 | 6890 | case SCHED_IDLE: |
1da177e4 LT |
6891 | ret = 0; |
6892 | } | |
6893 | return ret; | |
6894 | } | |
6895 | ||
abca5fc5 | 6896 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 6897 | { |
36c8b586 | 6898 | struct task_struct *p; |
a4ec24b4 | 6899 | unsigned int time_slice; |
eb580751 | 6900 | struct rq_flags rf; |
dba091b9 | 6901 | struct rq *rq; |
3a5c359a | 6902 | int retval; |
1da177e4 LT |
6903 | |
6904 | if (pid < 0) | |
3a5c359a | 6905 | return -EINVAL; |
1da177e4 LT |
6906 | |
6907 | retval = -ESRCH; | |
1a551ae7 | 6908 | rcu_read_lock(); |
1da177e4 LT |
6909 | p = find_process_by_pid(pid); |
6910 | if (!p) | |
6911 | goto out_unlock; | |
6912 | ||
6913 | retval = security_task_getscheduler(p); | |
6914 | if (retval) | |
6915 | goto out_unlock; | |
6916 | ||
eb580751 | 6917 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
6918 | time_slice = 0; |
6919 | if (p->sched_class->get_rr_interval) | |
6920 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 6921 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 6922 | |
1a551ae7 | 6923 | rcu_read_unlock(); |
abca5fc5 AV |
6924 | jiffies_to_timespec64(time_slice, t); |
6925 | return 0; | |
3a5c359a | 6926 | |
1da177e4 | 6927 | out_unlock: |
1a551ae7 | 6928 | rcu_read_unlock(); |
1da177e4 LT |
6929 | return retval; |
6930 | } | |
6931 | ||
2064a5ab RD |
6932 | /** |
6933 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6934 | * @pid: pid of the process. | |
6935 | * @interval: userspace pointer to the timeslice value. | |
6936 | * | |
6937 | * this syscall writes the default timeslice value of a given process | |
6938 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6939 | * | |
6940 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
6941 | * an error code. | |
6942 | */ | |
abca5fc5 | 6943 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 6944 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
6945 | { |
6946 | struct timespec64 t; | |
6947 | int retval = sched_rr_get_interval(pid, &t); | |
6948 | ||
6949 | if (retval == 0) | |
6950 | retval = put_timespec64(&t, interval); | |
6951 | ||
6952 | return retval; | |
6953 | } | |
6954 | ||
474b9c77 | 6955 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
6956 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
6957 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
6958 | { |
6959 | struct timespec64 t; | |
6960 | int retval = sched_rr_get_interval(pid, &t); | |
6961 | ||
6962 | if (retval == 0) | |
9afc5eee | 6963 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
6964 | return retval; |
6965 | } | |
6966 | #endif | |
6967 | ||
82a1fcb9 | 6968 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6969 | { |
1da177e4 | 6970 | unsigned long free = 0; |
4e79752c | 6971 | int ppid; |
c930b2c0 | 6972 | |
38200502 TH |
6973 | if (!try_get_task_stack(p)) |
6974 | return; | |
20435d84 | 6975 | |
cc172ff3 | 6976 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 XX |
6977 | |
6978 | if (p->state == TASK_RUNNING) | |
cc172ff3 | 6979 | pr_cont(" running task "); |
1da177e4 | 6980 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 6981 | free = stack_not_used(p); |
1da177e4 | 6982 | #endif |
a90e984c | 6983 | ppid = 0; |
4e79752c | 6984 | rcu_read_lock(); |
a90e984c ON |
6985 | if (pid_alive(p)) |
6986 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 6987 | rcu_read_unlock(); |
cc172ff3 LZ |
6988 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
6989 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 6990 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 6991 | |
3d1cb205 | 6992 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 6993 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 6994 | show_stack(p, NULL, KERN_INFO); |
38200502 | 6995 | put_task_stack(p); |
1da177e4 | 6996 | } |
0032f4e8 | 6997 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 6998 | |
5d68cc95 PZ |
6999 | static inline bool |
7000 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
7001 | { | |
7002 | /* no filter, everything matches */ | |
7003 | if (!state_filter) | |
7004 | return true; | |
7005 | ||
7006 | /* filter, but doesn't match */ | |
7007 | if (!(p->state & state_filter)) | |
7008 | return false; | |
7009 | ||
7010 | /* | |
7011 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
7012 | * TASK_KILLABLE). | |
7013 | */ | |
7014 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
7015 | return false; | |
7016 | ||
7017 | return true; | |
7018 | } | |
7019 | ||
7020 | ||
e59e2ae2 | 7021 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 7022 | { |
36c8b586 | 7023 | struct task_struct *g, *p; |
1da177e4 | 7024 | |
510f5acc | 7025 | rcu_read_lock(); |
5d07f420 | 7026 | for_each_process_thread(g, p) { |
1da177e4 LT |
7027 | /* |
7028 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 7029 | * console might take a lot of time: |
57675cb9 AR |
7030 | * Also, reset softlockup watchdogs on all CPUs, because |
7031 | * another CPU might be blocked waiting for us to process | |
7032 | * an IPI. | |
1da177e4 LT |
7033 | */ |
7034 | touch_nmi_watchdog(); | |
57675cb9 | 7035 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 7036 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 7037 | sched_show_task(p); |
5d07f420 | 7038 | } |
1da177e4 | 7039 | |
dd41f596 | 7040 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
7041 | if (!state_filter) |
7042 | sysrq_sched_debug_show(); | |
dd41f596 | 7043 | #endif |
510f5acc | 7044 | rcu_read_unlock(); |
e59e2ae2 IM |
7045 | /* |
7046 | * Only show locks if all tasks are dumped: | |
7047 | */ | |
93335a21 | 7048 | if (!state_filter) |
e59e2ae2 | 7049 | debug_show_all_locks(); |
1da177e4 LT |
7050 | } |
7051 | ||
f340c0d1 IM |
7052 | /** |
7053 | * init_idle - set up an idle thread for a given CPU | |
7054 | * @idle: task in question | |
d1ccc66d | 7055 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
7056 | * |
7057 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
7058 | * flag, to make booting more robust. | |
7059 | */ | |
0db0628d | 7060 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 7061 | { |
70b97a7f | 7062 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
7063 | unsigned long flags; |
7064 | ||
ff51ff84 PZ |
7065 | __sched_fork(0, idle); |
7066 | ||
25834c73 PZ |
7067 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
7068 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 7069 | |
06b83b5f | 7070 | idle->state = TASK_RUNNING; |
dd41f596 | 7071 | idle->se.exec_start = sched_clock(); |
c1de45ca | 7072 | idle->flags |= PF_IDLE; |
dd41f596 | 7073 | |
d08b9f0c | 7074 | scs_task_reset(idle); |
e1b77c92 MR |
7075 | kasan_unpoison_task_stack(idle); |
7076 | ||
de9b8f5d PZ |
7077 | #ifdef CONFIG_SMP |
7078 | /* | |
b19a888c | 7079 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
7080 | * in that case do_set_cpus_allowed() will not do the right thing. |
7081 | * | |
7082 | * And since this is boot we can forgo the serialization. | |
7083 | */ | |
9cfc3e18 | 7084 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 7085 | #endif |
6506cf6c PZ |
7086 | /* |
7087 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 7088 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
7089 | * lockdep check in task_group() will fail. |
7090 | * | |
7091 | * Similar case to sched_fork(). / Alternatively we could | |
7092 | * use task_rq_lock() here and obtain the other rq->lock. | |
7093 | * | |
7094 | * Silence PROVE_RCU | |
7095 | */ | |
7096 | rcu_read_lock(); | |
dd41f596 | 7097 | __set_task_cpu(idle, cpu); |
6506cf6c | 7098 | rcu_read_unlock(); |
1da177e4 | 7099 | |
5311a98f EB |
7100 | rq->idle = idle; |
7101 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 7102 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 7103 | #ifdef CONFIG_SMP |
3ca7a440 | 7104 | idle->on_cpu = 1; |
4866cde0 | 7105 | #endif |
25834c73 PZ |
7106 | raw_spin_unlock(&rq->lock); |
7107 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
7108 | |
7109 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 7110 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 7111 | |
dd41f596 IM |
7112 | /* |
7113 | * The idle tasks have their own, simple scheduling class: | |
7114 | */ | |
7115 | idle->sched_class = &idle_sched_class; | |
868baf07 | 7116 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 7117 | vtime_init_idle(idle, cpu); |
de9b8f5d | 7118 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
7119 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
7120 | #endif | |
19978ca6 IM |
7121 | } |
7122 | ||
e1d4eeec NP |
7123 | #ifdef CONFIG_SMP |
7124 | ||
f82f8042 JL |
7125 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
7126 | const struct cpumask *trial) | |
7127 | { | |
06a76fe0 | 7128 | int ret = 1; |
f82f8042 | 7129 | |
bb2bc55a MG |
7130 | if (!cpumask_weight(cur)) |
7131 | return ret; | |
7132 | ||
06a76fe0 | 7133 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
7134 | |
7135 | return ret; | |
7136 | } | |
7137 | ||
7f51412a JL |
7138 | int task_can_attach(struct task_struct *p, |
7139 | const struct cpumask *cs_cpus_allowed) | |
7140 | { | |
7141 | int ret = 0; | |
7142 | ||
7143 | /* | |
7144 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 7145 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
7146 | * affinity and isolating such threads by their set of |
7147 | * allowed nodes is unnecessary. Thus, cpusets are not | |
7148 | * applicable for such threads. This prevents checking for | |
7149 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 7150 | * before cpus_mask may be changed. |
7f51412a JL |
7151 | */ |
7152 | if (p->flags & PF_NO_SETAFFINITY) { | |
7153 | ret = -EINVAL; | |
7154 | goto out; | |
7155 | } | |
7156 | ||
7f51412a | 7157 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
7158 | cs_cpus_allowed)) |
7159 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 7160 | |
7f51412a JL |
7161 | out: |
7162 | return ret; | |
7163 | } | |
7164 | ||
f2cb1360 | 7165 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 7166 | |
e6628d5b MG |
7167 | #ifdef CONFIG_NUMA_BALANCING |
7168 | /* Migrate current task p to target_cpu */ | |
7169 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
7170 | { | |
7171 | struct migration_arg arg = { p, target_cpu }; | |
7172 | int curr_cpu = task_cpu(p); | |
7173 | ||
7174 | if (curr_cpu == target_cpu) | |
7175 | return 0; | |
7176 | ||
3bd37062 | 7177 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
7178 | return -EINVAL; |
7179 | ||
7180 | /* TODO: This is not properly updating schedstats */ | |
7181 | ||
286549dc | 7182 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
7183 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
7184 | } | |
0ec8aa00 PZ |
7185 | |
7186 | /* | |
7187 | * Requeue a task on a given node and accurately track the number of NUMA | |
7188 | * tasks on the runqueues | |
7189 | */ | |
7190 | void sched_setnuma(struct task_struct *p, int nid) | |
7191 | { | |
da0c1e65 | 7192 | bool queued, running; |
eb580751 PZ |
7193 | struct rq_flags rf; |
7194 | struct rq *rq; | |
0ec8aa00 | 7195 | |
eb580751 | 7196 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 7197 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
7198 | running = task_current(rq, p); |
7199 | ||
da0c1e65 | 7200 | if (queued) |
1de64443 | 7201 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 7202 | if (running) |
f3cd1c4e | 7203 | put_prev_task(rq, p); |
0ec8aa00 PZ |
7204 | |
7205 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 7206 | |
da0c1e65 | 7207 | if (queued) |
7134b3e9 | 7208 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 7209 | if (running) |
03b7fad1 | 7210 | set_next_task(rq, p); |
eb580751 | 7211 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 7212 | } |
5cc389bc | 7213 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 7214 | |
1da177e4 | 7215 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 7216 | /* |
d1ccc66d | 7217 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 7218 | * offline. |
054b9108 | 7219 | */ |
48c5ccae | 7220 | void idle_task_exit(void) |
1da177e4 | 7221 | { |
48c5ccae | 7222 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 7223 | |
48c5ccae | 7224 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 7225 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 7226 | |
a53efe5f | 7227 | if (mm != &init_mm) { |
252d2a41 | 7228 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
7229 | finish_arch_post_lock_switch(); |
7230 | } | |
bf2c59fc PZ |
7231 | |
7232 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
7233 | } |
7234 | ||
2558aacf | 7235 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 7236 | { |
2558aacf PZ |
7237 | struct task_struct *p = arg; |
7238 | struct rq *rq = this_rq(); | |
7239 | struct rq_flags rf; | |
7240 | int cpu; | |
1da177e4 | 7241 | |
2558aacf PZ |
7242 | raw_spin_lock_irq(&p->pi_lock); |
7243 | rq_lock(rq, &rf); | |
3f1d2a31 | 7244 | |
2558aacf PZ |
7245 | update_rq_clock(rq); |
7246 | ||
7247 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
7248 | cpu = select_fallback_rq(rq->cpu, p); | |
7249 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 7250 | } |
3f1d2a31 | 7251 | |
2558aacf PZ |
7252 | rq_unlock(rq, &rf); |
7253 | raw_spin_unlock_irq(&p->pi_lock); | |
7254 | ||
7255 | put_task_struct(p); | |
7256 | ||
7257 | return 0; | |
10e7071b | 7258 | } |
3f1d2a31 | 7259 | |
2558aacf PZ |
7260 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
7261 | ||
48f24c4d | 7262 | /* |
2558aacf | 7263 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
1da177e4 | 7264 | */ |
2558aacf | 7265 | static void balance_push(struct rq *rq) |
1da177e4 | 7266 | { |
2558aacf PZ |
7267 | struct task_struct *push_task = rq->curr; |
7268 | ||
7269 | lockdep_assert_held(&rq->lock); | |
7270 | SCHED_WARN_ON(rq->cpu != smp_processor_id()); | |
ae792702 PZ |
7271 | /* |
7272 | * Ensure the thing is persistent until balance_push_set(.on = false); | |
7273 | */ | |
7274 | rq->balance_callback = &balance_push_callback; | |
1da177e4 LT |
7275 | |
7276 | /* | |
2558aacf PZ |
7277 | * Both the cpu-hotplug and stop task are in this case and are |
7278 | * required to complete the hotplug process. | |
1da177e4 | 7279 | */ |
3015ef4b | 7280 | if (is_per_cpu_kthread(push_task) || is_migration_disabled(push_task)) { |
f2469a1f TG |
7281 | /* |
7282 | * If this is the idle task on the outgoing CPU try to wake | |
7283 | * up the hotplug control thread which might wait for the | |
7284 | * last task to vanish. The rcuwait_active() check is | |
7285 | * accurate here because the waiter is pinned on this CPU | |
7286 | * and can't obviously be running in parallel. | |
3015ef4b TG |
7287 | * |
7288 | * On RT kernels this also has to check whether there are | |
7289 | * pinned and scheduled out tasks on the runqueue. They | |
7290 | * need to leave the migrate disabled section first. | |
f2469a1f | 7291 | */ |
3015ef4b TG |
7292 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
7293 | rcuwait_active(&rq->hotplug_wait)) { | |
f2469a1f TG |
7294 | raw_spin_unlock(&rq->lock); |
7295 | rcuwait_wake_up(&rq->hotplug_wait); | |
7296 | raw_spin_lock(&rq->lock); | |
7297 | } | |
2558aacf | 7298 | return; |
f2469a1f | 7299 | } |
48f24c4d | 7300 | |
2558aacf | 7301 | get_task_struct(push_task); |
77bd3970 | 7302 | /* |
2558aacf PZ |
7303 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
7304 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 7305 | */ |
2558aacf PZ |
7306 | raw_spin_unlock(&rq->lock); |
7307 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, | |
7308 | this_cpu_ptr(&push_work)); | |
7309 | /* | |
7310 | * At this point need_resched() is true and we'll take the loop in | |
7311 | * schedule(). The next pick is obviously going to be the stop task | |
7312 | * which is_per_cpu_kthread() and will push this task away. | |
7313 | */ | |
7314 | raw_spin_lock(&rq->lock); | |
7315 | } | |
77bd3970 | 7316 | |
2558aacf PZ |
7317 | static void balance_push_set(int cpu, bool on) |
7318 | { | |
7319 | struct rq *rq = cpu_rq(cpu); | |
7320 | struct rq_flags rf; | |
48c5ccae | 7321 | |
2558aacf | 7322 | rq_lock_irqsave(rq, &rf); |
975707f2 | 7323 | rq->balance_push = on; |
22f667c9 PZ |
7324 | if (on) { |
7325 | WARN_ON_ONCE(rq->balance_callback); | |
ae792702 | 7326 | rq->balance_callback = &balance_push_callback; |
22f667c9 | 7327 | } else if (rq->balance_callback == &balance_push_callback) { |
ae792702 | 7328 | rq->balance_callback = NULL; |
22f667c9 | 7329 | } |
2558aacf PZ |
7330 | rq_unlock_irqrestore(rq, &rf); |
7331 | } | |
e692ab53 | 7332 | |
f2469a1f TG |
7333 | /* |
7334 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
7335 | * inactive. All tasks which are not per CPU kernel threads are either | |
7336 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
7337 | * during wakeup. Wait until the CPU is quiescent. | |
7338 | */ | |
7339 | static void balance_hotplug_wait(void) | |
7340 | { | |
7341 | struct rq *rq = this_rq(); | |
5473e0cc | 7342 | |
3015ef4b TG |
7343 | rcuwait_wait_event(&rq->hotplug_wait, |
7344 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
7345 | TASK_UNINTERRUPTIBLE); |
7346 | } | |
5473e0cc | 7347 | |
2558aacf | 7348 | #else |
dce48a84 | 7349 | |
2558aacf PZ |
7350 | static inline void balance_push(struct rq *rq) |
7351 | { | |
dce48a84 | 7352 | } |
dce48a84 | 7353 | |
2558aacf PZ |
7354 | static inline void balance_push_set(int cpu, bool on) |
7355 | { | |
7356 | } | |
7357 | ||
f2469a1f TG |
7358 | static inline void balance_hotplug_wait(void) |
7359 | { | |
dce48a84 | 7360 | } |
f2469a1f | 7361 | |
1da177e4 LT |
7362 | #endif /* CONFIG_HOTPLUG_CPU */ |
7363 | ||
f2cb1360 | 7364 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
7365 | { |
7366 | if (!rq->online) { | |
7367 | const struct sched_class *class; | |
7368 | ||
c6c4927b | 7369 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7370 | rq->online = 1; |
7371 | ||
7372 | for_each_class(class) { | |
7373 | if (class->rq_online) | |
7374 | class->rq_online(rq); | |
7375 | } | |
7376 | } | |
7377 | } | |
7378 | ||
f2cb1360 | 7379 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
7380 | { |
7381 | if (rq->online) { | |
7382 | const struct sched_class *class; | |
7383 | ||
7384 | for_each_class(class) { | |
7385 | if (class->rq_offline) | |
7386 | class->rq_offline(rq); | |
7387 | } | |
7388 | ||
c6c4927b | 7389 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7390 | rq->online = 0; |
7391 | } | |
7392 | } | |
7393 | ||
d1ccc66d IM |
7394 | /* |
7395 | * used to mark begin/end of suspend/resume: | |
7396 | */ | |
7397 | static int num_cpus_frozen; | |
d35be8ba | 7398 | |
1da177e4 | 7399 | /* |
3a101d05 TH |
7400 | * Update cpusets according to cpu_active mask. If cpusets are |
7401 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7402 | * around partition_sched_domains(). | |
d35be8ba SB |
7403 | * |
7404 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7405 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7406 | */ |
40190a78 | 7407 | static void cpuset_cpu_active(void) |
e761b772 | 7408 | { |
40190a78 | 7409 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7410 | /* |
7411 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7412 | * resume sequence. As long as this is not the last online | |
7413 | * operation in the resume sequence, just build a single sched | |
7414 | * domain, ignoring cpusets. | |
7415 | */ | |
50e76632 PZ |
7416 | partition_sched_domains(1, NULL, NULL); |
7417 | if (--num_cpus_frozen) | |
135fb3e1 | 7418 | return; |
d35be8ba SB |
7419 | /* |
7420 | * This is the last CPU online operation. So fall through and | |
7421 | * restore the original sched domains by considering the | |
7422 | * cpuset configurations. | |
7423 | */ | |
50e76632 | 7424 | cpuset_force_rebuild(); |
3a101d05 | 7425 | } |
30e03acd | 7426 | cpuset_update_active_cpus(); |
3a101d05 | 7427 | } |
e761b772 | 7428 | |
40190a78 | 7429 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7430 | { |
40190a78 | 7431 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 7432 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 7433 | return -EBUSY; |
30e03acd | 7434 | cpuset_update_active_cpus(); |
135fb3e1 | 7435 | } else { |
d35be8ba SB |
7436 | num_cpus_frozen++; |
7437 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7438 | } |
135fb3e1 | 7439 | return 0; |
e761b772 | 7440 | } |
e761b772 | 7441 | |
40190a78 | 7442 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7443 | { |
7d976699 | 7444 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 7445 | struct rq_flags rf; |
7d976699 | 7446 | |
22f667c9 PZ |
7447 | /* |
7448 | * Make sure that when the hotplug state machine does a roll-back | |
7449 | * we clear balance_push. Ideally that would happen earlier... | |
7450 | */ | |
2558aacf PZ |
7451 | balance_push_set(cpu, false); |
7452 | ||
ba2591a5 PZ |
7453 | #ifdef CONFIG_SCHED_SMT |
7454 | /* | |
c5511d03 | 7455 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 7456 | */ |
c5511d03 PZI |
7457 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
7458 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 7459 | #endif |
40190a78 | 7460 | set_cpu_active(cpu, true); |
135fb3e1 | 7461 | |
40190a78 | 7462 | if (sched_smp_initialized) { |
135fb3e1 | 7463 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7464 | cpuset_cpu_active(); |
e761b772 | 7465 | } |
7d976699 TG |
7466 | |
7467 | /* | |
7468 | * Put the rq online, if not already. This happens: | |
7469 | * | |
7470 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 7471 | * after all CPUs have been brought up. |
7d976699 TG |
7472 | * |
7473 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7474 | * domains. | |
7475 | */ | |
8a8c69c3 | 7476 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
7477 | if (rq->rd) { |
7478 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7479 | set_rq_online(rq); | |
7480 | } | |
8a8c69c3 | 7481 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 7482 | |
40190a78 | 7483 | return 0; |
135fb3e1 TG |
7484 | } |
7485 | ||
40190a78 | 7486 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7487 | { |
120455c5 PZ |
7488 | struct rq *rq = cpu_rq(cpu); |
7489 | struct rq_flags rf; | |
135fb3e1 TG |
7490 | int ret; |
7491 | ||
40190a78 | 7492 | set_cpu_active(cpu, false); |
975707f2 PZ |
7493 | balance_push_set(cpu, true); |
7494 | ||
b2454caa | 7495 | /* |
975707f2 PZ |
7496 | * We've cleared cpu_active_mask / set balance_push, wait for all |
7497 | * preempt-disabled and RCU users of this state to go away such that | |
7498 | * all new such users will observe it. | |
b2454caa | 7499 | * |
b2454caa PZ |
7500 | * Do sync before park smpboot threads to take care the rcu boost case. |
7501 | */ | |
309ba859 | 7502 | synchronize_rcu(); |
40190a78 | 7503 | |
120455c5 PZ |
7504 | rq_lock_irqsave(rq, &rf); |
7505 | if (rq->rd) { | |
7506 | update_rq_clock(rq); | |
7507 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7508 | set_rq_offline(rq); | |
7509 | } | |
7510 | rq_unlock_irqrestore(rq, &rf); | |
7511 | ||
c5511d03 PZI |
7512 | #ifdef CONFIG_SCHED_SMT |
7513 | /* | |
7514 | * When going down, decrement the number of cores with SMT present. | |
7515 | */ | |
7516 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
7517 | static_branch_dec_cpuslocked(&sched_smt_present); | |
7518 | #endif | |
7519 | ||
40190a78 TG |
7520 | if (!sched_smp_initialized) |
7521 | return 0; | |
7522 | ||
7523 | ret = cpuset_cpu_inactive(cpu); | |
7524 | if (ret) { | |
2558aacf | 7525 | balance_push_set(cpu, false); |
40190a78 TG |
7526 | set_cpu_active(cpu, true); |
7527 | return ret; | |
135fb3e1 | 7528 | } |
40190a78 TG |
7529 | sched_domains_numa_masks_clear(cpu); |
7530 | return 0; | |
135fb3e1 TG |
7531 | } |
7532 | ||
94baf7a5 TG |
7533 | static void sched_rq_cpu_starting(unsigned int cpu) |
7534 | { | |
7535 | struct rq *rq = cpu_rq(cpu); | |
7536 | ||
7537 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7538 | update_max_interval(); |
7539 | } | |
7540 | ||
135fb3e1 TG |
7541 | int sched_cpu_starting(unsigned int cpu) |
7542 | { | |
94baf7a5 | 7543 | sched_rq_cpu_starting(cpu); |
d84b3131 | 7544 | sched_tick_start(cpu); |
135fb3e1 | 7545 | return 0; |
e761b772 | 7546 | } |
e761b772 | 7547 | |
f2785ddb | 7548 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
7549 | |
7550 | /* | |
7551 | * Invoked immediately before the stopper thread is invoked to bring the | |
7552 | * CPU down completely. At this point all per CPU kthreads except the | |
7553 | * hotplug thread (current) and the stopper thread (inactive) have been | |
7554 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
7555 | * any of those which might be on the way out are gone. | |
7556 | * | |
7557 | * If after this point a bound task is being woken on this CPU then the | |
7558 | * responsible hotplug callback has failed to do it's job. | |
7559 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
7560 | */ | |
7561 | int sched_cpu_wait_empty(unsigned int cpu) | |
7562 | { | |
7563 | balance_hotplug_wait(); | |
7564 | return 0; | |
7565 | } | |
7566 | ||
7567 | /* | |
7568 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
7569 | * might have. Called from the CPU stopper task after ensuring that the | |
7570 | * stopper is the last running task on the CPU, so nr_active count is | |
7571 | * stable. We need to take the teardown thread which is calling this into | |
7572 | * account, so we hand in adjust = 1 to the load calculation. | |
7573 | * | |
7574 | * Also see the comment "Global load-average calculations". | |
7575 | */ | |
7576 | static void calc_load_migrate(struct rq *rq) | |
7577 | { | |
7578 | long delta = calc_load_fold_active(rq, 1); | |
7579 | ||
7580 | if (delta) | |
7581 | atomic_long_add(delta, &calc_load_tasks); | |
7582 | } | |
7583 | ||
36c6e17b VS |
7584 | static void dump_rq_tasks(struct rq *rq, const char *loglvl) |
7585 | { | |
7586 | struct task_struct *g, *p; | |
7587 | int cpu = cpu_of(rq); | |
7588 | ||
7589 | lockdep_assert_held(&rq->lock); | |
7590 | ||
7591 | printk("%sCPU%d enqueued tasks (%u total):\n", loglvl, cpu, rq->nr_running); | |
7592 | for_each_process_thread(g, p) { | |
7593 | if (task_cpu(p) != cpu) | |
7594 | continue; | |
7595 | ||
7596 | if (!task_on_rq_queued(p)) | |
7597 | continue; | |
7598 | ||
7599 | printk("%s\tpid: %d, name: %s\n", loglvl, p->pid, p->comm); | |
7600 | } | |
7601 | } | |
7602 | ||
f2785ddb TG |
7603 | int sched_cpu_dying(unsigned int cpu) |
7604 | { | |
7605 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 7606 | struct rq_flags rf; |
f2785ddb TG |
7607 | |
7608 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 7609 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
7610 | |
7611 | rq_lock_irqsave(rq, &rf); | |
36c6e17b VS |
7612 | if (rq->nr_running != 1 || rq_has_pinned_tasks(rq)) { |
7613 | WARN(true, "Dying CPU not properly vacated!"); | |
7614 | dump_rq_tasks(rq, KERN_WARNING); | |
7615 | } | |
8a8c69c3 PZ |
7616 | rq_unlock_irqrestore(rq, &rf); |
7617 | ||
22f667c9 PZ |
7618 | /* |
7619 | * Now that the CPU is offline, make sure we're welcome | |
7620 | * to new tasks once we come back up. | |
7621 | */ | |
7622 | balance_push_set(cpu, false); | |
7623 | ||
f2785ddb TG |
7624 | calc_load_migrate(rq); |
7625 | update_max_interval(); | |
00357f5e | 7626 | nohz_balance_exit_idle(rq); |
e5ef27d0 | 7627 | hrtick_clear(rq); |
f2785ddb TG |
7628 | return 0; |
7629 | } | |
7630 | #endif | |
7631 | ||
1da177e4 LT |
7632 | void __init sched_init_smp(void) |
7633 | { | |
cb83b629 PZ |
7634 | sched_init_numa(); |
7635 | ||
6acce3ef PZ |
7636 | /* |
7637 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 7638 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 7639 | * happen. |
6acce3ef | 7640 | */ |
712555ee | 7641 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 7642 | sched_init_domains(cpu_active_mask); |
712555ee | 7643 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7644 | |
5c1e1767 | 7645 | /* Move init over to a non-isolated CPU */ |
edb93821 | 7646 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 7647 | BUG(); |
19978ca6 | 7648 | sched_init_granularity(); |
4212823f | 7649 | |
0e3900e6 | 7650 | init_sched_rt_class(); |
1baca4ce | 7651 | init_sched_dl_class(); |
1b568f0a | 7652 | |
e26fbffd | 7653 | sched_smp_initialized = true; |
1da177e4 | 7654 | } |
e26fbffd TG |
7655 | |
7656 | static int __init migration_init(void) | |
7657 | { | |
77a5352b | 7658 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 7659 | return 0; |
1da177e4 | 7660 | } |
e26fbffd TG |
7661 | early_initcall(migration_init); |
7662 | ||
1da177e4 LT |
7663 | #else |
7664 | void __init sched_init_smp(void) | |
7665 | { | |
19978ca6 | 7666 | sched_init_granularity(); |
1da177e4 LT |
7667 | } |
7668 | #endif /* CONFIG_SMP */ | |
7669 | ||
7670 | int in_sched_functions(unsigned long addr) | |
7671 | { | |
1da177e4 LT |
7672 | return in_lock_functions(addr) || |
7673 | (addr >= (unsigned long)__sched_text_start | |
7674 | && addr < (unsigned long)__sched_text_end); | |
7675 | } | |
7676 | ||
029632fb | 7677 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7678 | /* |
7679 | * Default task group. | |
7680 | * Every task in system belongs to this group at bootup. | |
7681 | */ | |
029632fb | 7682 | struct task_group root_task_group; |
35cf4e50 | 7683 | LIST_HEAD(task_groups); |
b0367629 WL |
7684 | |
7685 | /* Cacheline aligned slab cache for task_group */ | |
7686 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7687 | #endif |
6f505b16 | 7688 | |
e6252c3e | 7689 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 7690 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 7691 | |
1da177e4 LT |
7692 | void __init sched_init(void) |
7693 | { | |
a1dc0446 | 7694 | unsigned long ptr = 0; |
55627e3c | 7695 | int i; |
434d53b0 | 7696 | |
c3a340f7 SRV |
7697 | /* Make sure the linker didn't screw up */ |
7698 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
7699 | &fair_sched_class + 1 != &rt_sched_class || | |
7700 | &rt_sched_class + 1 != &dl_sched_class); | |
7701 | #ifdef CONFIG_SMP | |
7702 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
7703 | #endif | |
7704 | ||
5822a454 | 7705 | wait_bit_init(); |
9dcb8b68 | 7706 | |
434d53b0 | 7707 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 7708 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
7709 | #endif |
7710 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 7711 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 7712 | #endif |
a1dc0446 QC |
7713 | if (ptr) { |
7714 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
7715 | |
7716 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7717 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7718 | ptr += nr_cpu_ids * sizeof(void **); |
7719 | ||
07e06b01 | 7720 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7721 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7722 | |
b1d1779e WY |
7723 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
7724 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 7725 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7726 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7727 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7728 | ptr += nr_cpu_ids * sizeof(void **); |
7729 | ||
07e06b01 | 7730 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7731 | ptr += nr_cpu_ids * sizeof(void **); |
7732 | ||
6d6bc0ad | 7733 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7734 | } |
df7c8e84 | 7735 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7736 | for_each_possible_cpu(i) { |
7737 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7738 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
7739 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
7740 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7741 | } |
b74e6278 | 7742 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7743 | |
d1ccc66d IM |
7744 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
7745 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 7746 | |
57d885fe GH |
7747 | #ifdef CONFIG_SMP |
7748 | init_defrootdomain(); | |
7749 | #endif | |
7750 | ||
d0b27fa7 | 7751 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7752 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7753 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7754 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7755 | |
7c941438 | 7756 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7757 | task_group_cache = KMEM_CACHE(task_group, 0); |
7758 | ||
07e06b01 YZ |
7759 | list_add(&root_task_group.list, &task_groups); |
7760 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7761 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7762 | autogroup_init(&init_task); |
7c941438 | 7763 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7764 | |
0a945022 | 7765 | for_each_possible_cpu(i) { |
70b97a7f | 7766 | struct rq *rq; |
1da177e4 LT |
7767 | |
7768 | rq = cpu_rq(i); | |
05fa785c | 7769 | raw_spin_lock_init(&rq->lock); |
7897986b | 7770 | rq->nr_running = 0; |
dce48a84 TG |
7771 | rq->calc_load_active = 0; |
7772 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7773 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7774 | init_rt_rq(&rq->rt); |
7775 | init_dl_rq(&rq->dl); | |
dd41f596 | 7776 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 7777 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 7778 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 7779 | /* |
d1ccc66d | 7780 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
7781 | * |
7782 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
7783 | * gets 100% of the CPU resources in the system. This overall |
7784 | * system CPU resource is divided among the tasks of | |
07e06b01 | 7785 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7786 | * based on each entity's (task or task-group's) weight |
7787 | * (se->load.weight). | |
7788 | * | |
07e06b01 | 7789 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 7790 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 7791 | * then A0's share of the CPU resource is: |
354d60c2 | 7792 | * |
0d905bca | 7793 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7794 | * |
07e06b01 YZ |
7795 | * We achieve this by letting root_task_group's tasks sit |
7796 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7797 | */ |
07e06b01 | 7798 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7799 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7800 | ||
7801 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7802 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7803 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7804 | #endif |
1da177e4 | 7805 | #ifdef CONFIG_SMP |
41c7ce9a | 7806 | rq->sd = NULL; |
57d885fe | 7807 | rq->rd = NULL; |
ca6d75e6 | 7808 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7809 | rq->balance_callback = NULL; |
1da177e4 | 7810 | rq->active_balance = 0; |
dd41f596 | 7811 | rq->next_balance = jiffies; |
1da177e4 | 7812 | rq->push_cpu = 0; |
0a2966b4 | 7813 | rq->cpu = i; |
1f11eb6a | 7814 | rq->online = 0; |
eae0c9df MG |
7815 | rq->idle_stamp = 0; |
7816 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7817 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7818 | |
7819 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7820 | ||
dc938520 | 7821 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7822 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 7823 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 7824 | atomic_set(&rq->nohz_flags, 0); |
90b5363a | 7825 | |
545b8c8d | 7826 | INIT_CSD(&rq->nohz_csd, nohz_csd_func, rq); |
83cd4fe2 | 7827 | #endif |
f2469a1f TG |
7828 | #ifdef CONFIG_HOTPLUG_CPU |
7829 | rcuwait_init(&rq->hotplug_wait); | |
83cd4fe2 | 7830 | #endif |
9fd81dd5 | 7831 | #endif /* CONFIG_SMP */ |
77a021be | 7832 | hrtick_rq_init(rq); |
1da177e4 | 7833 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7834 | } |
7835 | ||
9059393e | 7836 | set_load_weight(&init_task, false); |
b50f60ce | 7837 | |
1da177e4 LT |
7838 | /* |
7839 | * The boot idle thread does lazy MMU switching as well: | |
7840 | */ | |
f1f10076 | 7841 | mmgrab(&init_mm); |
1da177e4 LT |
7842 | enter_lazy_tlb(&init_mm, current); |
7843 | ||
7844 | /* | |
7845 | * Make us the idle thread. Technically, schedule() should not be | |
7846 | * called from this thread, however somewhere below it might be, | |
7847 | * but because we are the idle thread, we just pick up running again | |
7848 | * when this runqueue becomes "idle". | |
7849 | */ | |
7850 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7851 | |
7852 | calc_load_update = jiffies + LOAD_FREQ; | |
7853 | ||
bf4d83f6 | 7854 | #ifdef CONFIG_SMP |
29d5e047 | 7855 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7856 | #endif |
7857 | init_sched_fair_class(); | |
6a7b3dc3 | 7858 | |
4698f88c JP |
7859 | init_schedstats(); |
7860 | ||
eb414681 JW |
7861 | psi_init(); |
7862 | ||
69842cba PB |
7863 | init_uclamp(); |
7864 | ||
6892b75e | 7865 | scheduler_running = 1; |
1da177e4 LT |
7866 | } |
7867 | ||
d902db1e | 7868 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7869 | static inline int preempt_count_equals(int preempt_offset) |
7870 | { | |
da7142e2 | 7871 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7872 | |
4ba8216c | 7873 | return (nested == preempt_offset); |
e4aafea2 FW |
7874 | } |
7875 | ||
d894837f | 7876 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7877 | { |
8eb23b9f PZ |
7878 | /* |
7879 | * Blocking primitives will set (and therefore destroy) current->state, | |
7880 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7881 | * otherwise we will destroy state. | |
7882 | */ | |
00845eb9 | 7883 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7884 | "do not call blocking ops when !TASK_RUNNING; " |
7885 | "state=%lx set at [<%p>] %pS\n", | |
7886 | current->state, | |
7887 | (void *)current->task_state_change, | |
00845eb9 | 7888 | (void *)current->task_state_change); |
8eb23b9f | 7889 | |
3427445a PZ |
7890 | ___might_sleep(file, line, preempt_offset); |
7891 | } | |
7892 | EXPORT_SYMBOL(__might_sleep); | |
7893 | ||
7894 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7895 | { |
d1ccc66d IM |
7896 | /* Ratelimiting timestamp: */ |
7897 | static unsigned long prev_jiffy; | |
7898 | ||
d1c6d149 | 7899 | unsigned long preempt_disable_ip; |
1da177e4 | 7900 | |
d1ccc66d IM |
7901 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
7902 | rcu_sleep_check(); | |
7903 | ||
db273be2 | 7904 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 7905 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
7906 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
7907 | oops_in_progress) | |
aef745fc | 7908 | return; |
1c3c5eab | 7909 | |
aef745fc IM |
7910 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
7911 | return; | |
7912 | prev_jiffy = jiffies; | |
7913 | ||
d1ccc66d | 7914 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
7915 | preempt_disable_ip = get_preempt_disable_ip(current); |
7916 | ||
3df0fc5b PZ |
7917 | printk(KERN_ERR |
7918 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7919 | file, line); | |
7920 | printk(KERN_ERR | |
312364f3 DV |
7921 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
7922 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 7923 | current->pid, current->comm); |
aef745fc | 7924 | |
a8b686b3 ES |
7925 | if (task_stack_end_corrupted(current)) |
7926 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7927 | ||
aef745fc IM |
7928 | debug_show_held_locks(current); |
7929 | if (irqs_disabled()) | |
7930 | print_irqtrace_events(current); | |
d1c6d149 VN |
7931 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
7932 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 7933 | pr_err("Preemption disabled at:"); |
2062a4e8 | 7934 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 7935 | } |
aef745fc | 7936 | dump_stack(); |
f0b22e39 | 7937 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 7938 | } |
3427445a | 7939 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
7940 | |
7941 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
7942 | { | |
7943 | static unsigned long prev_jiffy; | |
7944 | ||
7945 | if (irqs_disabled()) | |
7946 | return; | |
7947 | ||
7948 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
7949 | return; | |
7950 | ||
7951 | if (preempt_count() > preempt_offset) | |
7952 | return; | |
7953 | ||
7954 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7955 | return; | |
7956 | prev_jiffy = jiffies; | |
7957 | ||
7958 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
7959 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7960 | in_atomic(), irqs_disabled(), | |
7961 | current->pid, current->comm); | |
7962 | ||
7963 | debug_show_held_locks(current); | |
7964 | dump_stack(); | |
7965 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
7966 | } | |
7967 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
7968 | |
7969 | #ifdef CONFIG_SMP | |
7970 | void __cant_migrate(const char *file, int line) | |
7971 | { | |
7972 | static unsigned long prev_jiffy; | |
7973 | ||
7974 | if (irqs_disabled()) | |
7975 | return; | |
7976 | ||
7977 | if (is_migration_disabled(current)) | |
7978 | return; | |
7979 | ||
7980 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
7981 | return; | |
7982 | ||
7983 | if (preempt_count() > 0) | |
7984 | return; | |
7985 | ||
7986 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7987 | return; | |
7988 | prev_jiffy = jiffies; | |
7989 | ||
7990 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
7991 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
7992 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
7993 | current->pid, current->comm); | |
7994 | ||
7995 | debug_show_held_locks(current); | |
7996 | dump_stack(); | |
7997 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
7998 | } | |
7999 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
8000 | #endif | |
1da177e4 LT |
8001 | #endif |
8002 | ||
8003 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 8004 | void normalize_rt_tasks(void) |
3a5e4dc1 | 8005 | { |
dbc7f069 | 8006 | struct task_struct *g, *p; |
d50dde5a DF |
8007 | struct sched_attr attr = { |
8008 | .sched_policy = SCHED_NORMAL, | |
8009 | }; | |
1da177e4 | 8010 | |
3472eaa1 | 8011 | read_lock(&tasklist_lock); |
5d07f420 | 8012 | for_each_process_thread(g, p) { |
178be793 IM |
8013 | /* |
8014 | * Only normalize user tasks: | |
8015 | */ | |
3472eaa1 | 8016 | if (p->flags & PF_KTHREAD) |
178be793 IM |
8017 | continue; |
8018 | ||
4fa8d299 JP |
8019 | p->se.exec_start = 0; |
8020 | schedstat_set(p->se.statistics.wait_start, 0); | |
8021 | schedstat_set(p->se.statistics.sleep_start, 0); | |
8022 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 8023 | |
aab03e05 | 8024 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
8025 | /* |
8026 | * Renice negative nice level userspace | |
8027 | * tasks back to 0: | |
8028 | */ | |
3472eaa1 | 8029 | if (task_nice(p) < 0) |
dd41f596 | 8030 | set_user_nice(p, 0); |
1da177e4 | 8031 | continue; |
dd41f596 | 8032 | } |
1da177e4 | 8033 | |
dbc7f069 | 8034 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 8035 | } |
3472eaa1 | 8036 | read_unlock(&tasklist_lock); |
1da177e4 LT |
8037 | } |
8038 | ||
8039 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8040 | |
67fc4e0c | 8041 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8042 | /* |
67fc4e0c | 8043 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8044 | * |
8045 | * They can only be called when the whole system has been | |
8046 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8047 | * activity can take place. Using them for anything else would | |
8048 | * be a serious bug, and as a result, they aren't even visible | |
8049 | * under any other configuration. | |
8050 | */ | |
8051 | ||
8052 | /** | |
d1ccc66d | 8053 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
8054 | * @cpu: the processor in question. |
8055 | * | |
8056 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
8057 | * |
8058 | * Return: The current task for @cpu. | |
1df5c10a | 8059 | */ |
36c8b586 | 8060 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8061 | { |
8062 | return cpu_curr(cpu); | |
8063 | } | |
8064 | ||
67fc4e0c JW |
8065 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8066 | ||
8067 | #ifdef CONFIG_IA64 | |
1df5c10a | 8068 | /** |
5feeb783 | 8069 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
8070 | * @cpu: the processor in question. |
8071 | * @p: the task pointer to set. | |
8072 | * | |
8073 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 8074 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 8075 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
8076 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8077 | * and caller must save the original value of the current task (see | |
8078 | * curr_task() above) and restore that value before reenabling interrupts and | |
8079 | * re-starting the system. | |
8080 | * | |
8081 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8082 | */ | |
a458ae2e | 8083 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8084 | { |
8085 | cpu_curr(cpu) = p; | |
8086 | } | |
8087 | ||
8088 | #endif | |
29f59db3 | 8089 | |
7c941438 | 8090 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
8091 | /* task_group_lock serializes the addition/removal of task groups */ |
8092 | static DEFINE_SPINLOCK(task_group_lock); | |
8093 | ||
2480c093 PB |
8094 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
8095 | struct task_group *parent) | |
8096 | { | |
8097 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 8098 | enum uclamp_id clamp_id; |
2480c093 PB |
8099 | |
8100 | for_each_clamp_id(clamp_id) { | |
8101 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
8102 | uclamp_none(clamp_id), false); | |
0b60ba2d | 8103 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
8104 | } |
8105 | #endif | |
8106 | } | |
8107 | ||
2f5177f0 | 8108 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
8109 | { |
8110 | free_fair_sched_group(tg); | |
8111 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8112 | autogroup_free(tg); |
b0367629 | 8113 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
8114 | } |
8115 | ||
8116 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8117 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8118 | { |
8119 | struct task_group *tg; | |
bccbe08a | 8120 | |
b0367629 | 8121 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
8122 | if (!tg) |
8123 | return ERR_PTR(-ENOMEM); | |
8124 | ||
ec7dc8ac | 8125 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8126 | goto err; |
8127 | ||
ec7dc8ac | 8128 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8129 | goto err; |
8130 | ||
2480c093 PB |
8131 | alloc_uclamp_sched_group(tg, parent); |
8132 | ||
ace783b9 LZ |
8133 | return tg; |
8134 | ||
8135 | err: | |
2f5177f0 | 8136 | sched_free_group(tg); |
ace783b9 LZ |
8137 | return ERR_PTR(-ENOMEM); |
8138 | } | |
8139 | ||
8140 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
8141 | { | |
8142 | unsigned long flags; | |
8143 | ||
8ed36996 | 8144 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8145 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 8146 | |
d1ccc66d IM |
8147 | /* Root should already exist: */ |
8148 | WARN_ON(!parent); | |
f473aa5e PZ |
8149 | |
8150 | tg->parent = parent; | |
f473aa5e | 8151 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8152 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8153 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
8154 | |
8155 | online_fair_sched_group(tg); | |
29f59db3 SV |
8156 | } |
8157 | ||
9b5b7751 | 8158 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 8159 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8160 | { |
d1ccc66d | 8161 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 8162 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8163 | } |
8164 | ||
4cf86d77 | 8165 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 8166 | { |
d1ccc66d | 8167 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 8168 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
8169 | } |
8170 | ||
8171 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 8172 | { |
8ed36996 | 8173 | unsigned long flags; |
29f59db3 | 8174 | |
d1ccc66d | 8175 | /* End participation in shares distribution: */ |
6fe1f348 | 8176 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
8177 | |
8178 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8179 | list_del_rcu(&tg->list); |
f473aa5e | 8180 | list_del_rcu(&tg->siblings); |
8ed36996 | 8181 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
8182 | } |
8183 | ||
ea86cb4b | 8184 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 8185 | { |
8323f26c | 8186 | struct task_group *tg; |
29f59db3 | 8187 | |
f7b8a47d KT |
8188 | /* |
8189 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
8190 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
8191 | * to prevent lockdep warnings. | |
8192 | */ | |
8193 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
8194 | struct task_group, css); |
8195 | tg = autogroup_task_group(tsk, tg); | |
8196 | tsk->sched_task_group = tg; | |
8197 | ||
810b3817 | 8198 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
8199 | if (tsk->sched_class->task_change_group) |
8200 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 8201 | else |
810b3817 | 8202 | #endif |
b2b5ce02 | 8203 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
8204 | } |
8205 | ||
8206 | /* | |
8207 | * Change task's runqueue when it moves between groups. | |
8208 | * | |
8209 | * The caller of this function should have put the task in its new group by | |
8210 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
8211 | * its new group. | |
8212 | */ | |
8213 | void sched_move_task(struct task_struct *tsk) | |
8214 | { | |
7a57f32a PZ |
8215 | int queued, running, queue_flags = |
8216 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
8217 | struct rq_flags rf; |
8218 | struct rq *rq; | |
8219 | ||
8220 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 8221 | update_rq_clock(rq); |
ea86cb4b VG |
8222 | |
8223 | running = task_current(rq, tsk); | |
8224 | queued = task_on_rq_queued(tsk); | |
8225 | ||
8226 | if (queued) | |
7a57f32a | 8227 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 8228 | if (running) |
ea86cb4b VG |
8229 | put_prev_task(rq, tsk); |
8230 | ||
8231 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 8232 | |
da0c1e65 | 8233 | if (queued) |
7a57f32a | 8234 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 8235 | if (running) { |
03b7fad1 | 8236 | set_next_task(rq, tsk); |
2a4b03ff VG |
8237 | /* |
8238 | * After changing group, the running task may have joined a | |
8239 | * throttled one but it's still the running task. Trigger a | |
8240 | * resched to make sure that task can still run. | |
8241 | */ | |
8242 | resched_curr(rq); | |
8243 | } | |
29f59db3 | 8244 | |
eb580751 | 8245 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 8246 | } |
68318b8e | 8247 | |
a7c6d554 | 8248 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8249 | { |
a7c6d554 | 8250 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8251 | } |
8252 | ||
eb95419b TH |
8253 | static struct cgroup_subsys_state * |
8254 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8255 | { |
eb95419b TH |
8256 | struct task_group *parent = css_tg(parent_css); |
8257 | struct task_group *tg; | |
68318b8e | 8258 | |
eb95419b | 8259 | if (!parent) { |
68318b8e | 8260 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8261 | return &root_task_group.css; |
68318b8e SV |
8262 | } |
8263 | ||
ec7dc8ac | 8264 | tg = sched_create_group(parent); |
68318b8e SV |
8265 | if (IS_ERR(tg)) |
8266 | return ERR_PTR(-ENOMEM); | |
8267 | ||
68318b8e SV |
8268 | return &tg->css; |
8269 | } | |
8270 | ||
96b77745 KK |
8271 | /* Expose task group only after completing cgroup initialization */ |
8272 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
8273 | { | |
8274 | struct task_group *tg = css_tg(css); | |
8275 | struct task_group *parent = css_tg(css->parent); | |
8276 | ||
8277 | if (parent) | |
8278 | sched_online_group(tg, parent); | |
7226017a QY |
8279 | |
8280 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8281 | /* Propagate the effective uclamp value for the new group */ | |
8282 | cpu_util_update_eff(css); | |
8283 | #endif | |
8284 | ||
96b77745 KK |
8285 | return 0; |
8286 | } | |
8287 | ||
2f5177f0 | 8288 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8289 | { |
eb95419b | 8290 | struct task_group *tg = css_tg(css); |
ace783b9 | 8291 | |
2f5177f0 | 8292 | sched_offline_group(tg); |
ace783b9 LZ |
8293 | } |
8294 | ||
eb95419b | 8295 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8296 | { |
eb95419b | 8297 | struct task_group *tg = css_tg(css); |
68318b8e | 8298 | |
2f5177f0 PZ |
8299 | /* |
8300 | * Relies on the RCU grace period between css_released() and this. | |
8301 | */ | |
8302 | sched_free_group(tg); | |
ace783b9 LZ |
8303 | } |
8304 | ||
ea86cb4b VG |
8305 | /* |
8306 | * This is called before wake_up_new_task(), therefore we really only | |
8307 | * have to set its group bits, all the other stuff does not apply. | |
8308 | */ | |
b53202e6 | 8309 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 8310 | { |
ea86cb4b VG |
8311 | struct rq_flags rf; |
8312 | struct rq *rq; | |
8313 | ||
8314 | rq = task_rq_lock(task, &rf); | |
8315 | ||
80f5c1b8 | 8316 | update_rq_clock(rq); |
ea86cb4b VG |
8317 | sched_change_group(task, TASK_SET_GROUP); |
8318 | ||
8319 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
8320 | } |
8321 | ||
1f7dd3e5 | 8322 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8323 | { |
bb9d97b6 | 8324 | struct task_struct *task; |
1f7dd3e5 | 8325 | struct cgroup_subsys_state *css; |
7dc603c9 | 8326 | int ret = 0; |
bb9d97b6 | 8327 | |
1f7dd3e5 | 8328 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8329 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8330 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8331 | return -EINVAL; |
b68aa230 | 8332 | #endif |
7dc603c9 | 8333 | /* |
b19a888c | 8334 | * Serialize against wake_up_new_task() such that if it's |
7dc603c9 PZ |
8335 | * running, we're sure to observe its full state. |
8336 | */ | |
8337 | raw_spin_lock_irq(&task->pi_lock); | |
8338 | /* | |
8339 | * Avoid calling sched_move_task() before wake_up_new_task() | |
8340 | * has happened. This would lead to problems with PELT, due to | |
8341 | * move wanting to detach+attach while we're not attached yet. | |
8342 | */ | |
8343 | if (task->state == TASK_NEW) | |
8344 | ret = -EINVAL; | |
8345 | raw_spin_unlock_irq(&task->pi_lock); | |
8346 | ||
8347 | if (ret) | |
8348 | break; | |
bb9d97b6 | 8349 | } |
7dc603c9 | 8350 | return ret; |
be367d09 | 8351 | } |
68318b8e | 8352 | |
1f7dd3e5 | 8353 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8354 | { |
bb9d97b6 | 8355 | struct task_struct *task; |
1f7dd3e5 | 8356 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8357 | |
1f7dd3e5 | 8358 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8359 | sched_move_task(task); |
68318b8e SV |
8360 | } |
8361 | ||
2480c093 | 8362 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
8363 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
8364 | { | |
8365 | struct cgroup_subsys_state *top_css = css; | |
8366 | struct uclamp_se *uc_parent = NULL; | |
8367 | struct uclamp_se *uc_se = NULL; | |
8368 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 8369 | enum uclamp_id clamp_id; |
0b60ba2d PB |
8370 | unsigned int clamps; |
8371 | ||
8372 | css_for_each_descendant_pre(css, top_css) { | |
8373 | uc_parent = css_tg(css)->parent | |
8374 | ? css_tg(css)->parent->uclamp : NULL; | |
8375 | ||
8376 | for_each_clamp_id(clamp_id) { | |
8377 | /* Assume effective clamps matches requested clamps */ | |
8378 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
8379 | /* Cap effective clamps with parent's effective clamps */ | |
8380 | if (uc_parent && | |
8381 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
8382 | eff[clamp_id] = uc_parent[clamp_id].value; | |
8383 | } | |
8384 | } | |
8385 | /* Ensure protection is always capped by limit */ | |
8386 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
8387 | ||
8388 | /* Propagate most restrictive effective clamps */ | |
8389 | clamps = 0x0; | |
8390 | uc_se = css_tg(css)->uclamp; | |
8391 | for_each_clamp_id(clamp_id) { | |
8392 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
8393 | continue; | |
8394 | uc_se[clamp_id].value = eff[clamp_id]; | |
8395 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
8396 | clamps |= (0x1 << clamp_id); | |
8397 | } | |
babbe170 | 8398 | if (!clamps) { |
0b60ba2d | 8399 | css = css_rightmost_descendant(css); |
babbe170 PB |
8400 | continue; |
8401 | } | |
8402 | ||
8403 | /* Immediately update descendants RUNNABLE tasks */ | |
8404 | uclamp_update_active_tasks(css, clamps); | |
0b60ba2d PB |
8405 | } |
8406 | } | |
2480c093 PB |
8407 | |
8408 | /* | |
8409 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
8410 | * C expression. Since there is no way to convert a macro argument (N) into a | |
8411 | * character constant, use two levels of macros. | |
8412 | */ | |
8413 | #define _POW10(exp) ((unsigned int)1e##exp) | |
8414 | #define POW10(exp) _POW10(exp) | |
8415 | ||
8416 | struct uclamp_request { | |
8417 | #define UCLAMP_PERCENT_SHIFT 2 | |
8418 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
8419 | s64 percent; | |
8420 | u64 util; | |
8421 | int ret; | |
8422 | }; | |
8423 | ||
8424 | static inline struct uclamp_request | |
8425 | capacity_from_percent(char *buf) | |
8426 | { | |
8427 | struct uclamp_request req = { | |
8428 | .percent = UCLAMP_PERCENT_SCALE, | |
8429 | .util = SCHED_CAPACITY_SCALE, | |
8430 | .ret = 0, | |
8431 | }; | |
8432 | ||
8433 | buf = strim(buf); | |
8434 | if (strcmp(buf, "max")) { | |
8435 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
8436 | &req.percent); | |
8437 | if (req.ret) | |
8438 | return req; | |
b562d140 | 8439 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
8440 | req.ret = -ERANGE; |
8441 | return req; | |
8442 | } | |
8443 | ||
8444 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
8445 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
8446 | } | |
8447 | ||
8448 | return req; | |
8449 | } | |
8450 | ||
8451 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
8452 | size_t nbytes, loff_t off, | |
8453 | enum uclamp_id clamp_id) | |
8454 | { | |
8455 | struct uclamp_request req; | |
8456 | struct task_group *tg; | |
8457 | ||
8458 | req = capacity_from_percent(buf); | |
8459 | if (req.ret) | |
8460 | return req.ret; | |
8461 | ||
46609ce2 QY |
8462 | static_branch_enable(&sched_uclamp_used); |
8463 | ||
2480c093 PB |
8464 | mutex_lock(&uclamp_mutex); |
8465 | rcu_read_lock(); | |
8466 | ||
8467 | tg = css_tg(of_css(of)); | |
8468 | if (tg->uclamp_req[clamp_id].value != req.util) | |
8469 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
8470 | ||
8471 | /* | |
8472 | * Because of not recoverable conversion rounding we keep track of the | |
8473 | * exact requested value | |
8474 | */ | |
8475 | tg->uclamp_pct[clamp_id] = req.percent; | |
8476 | ||
0b60ba2d PB |
8477 | /* Update effective clamps to track the most restrictive value */ |
8478 | cpu_util_update_eff(of_css(of)); | |
8479 | ||
2480c093 PB |
8480 | rcu_read_unlock(); |
8481 | mutex_unlock(&uclamp_mutex); | |
8482 | ||
8483 | return nbytes; | |
8484 | } | |
8485 | ||
8486 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
8487 | char *buf, size_t nbytes, | |
8488 | loff_t off) | |
8489 | { | |
8490 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
8491 | } | |
8492 | ||
8493 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
8494 | char *buf, size_t nbytes, | |
8495 | loff_t off) | |
8496 | { | |
8497 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
8498 | } | |
8499 | ||
8500 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
8501 | enum uclamp_id clamp_id) | |
8502 | { | |
8503 | struct task_group *tg; | |
8504 | u64 util_clamp; | |
8505 | u64 percent; | |
8506 | u32 rem; | |
8507 | ||
8508 | rcu_read_lock(); | |
8509 | tg = css_tg(seq_css(sf)); | |
8510 | util_clamp = tg->uclamp_req[clamp_id].value; | |
8511 | rcu_read_unlock(); | |
8512 | ||
8513 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
8514 | seq_puts(sf, "max\n"); | |
8515 | return; | |
8516 | } | |
8517 | ||
8518 | percent = tg->uclamp_pct[clamp_id]; | |
8519 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
8520 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
8521 | } | |
8522 | ||
8523 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
8524 | { | |
8525 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
8526 | return 0; | |
8527 | } | |
8528 | ||
8529 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
8530 | { | |
8531 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
8532 | return 0; | |
8533 | } | |
8534 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
8535 | ||
052f1dc7 | 8536 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8537 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8538 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8539 | { |
5b61d50a KK |
8540 | if (shareval > scale_load_down(ULONG_MAX)) |
8541 | shareval = MAX_SHARES; | |
182446d0 | 8542 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8543 | } |
8544 | ||
182446d0 TH |
8545 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8546 | struct cftype *cft) | |
68318b8e | 8547 | { |
182446d0 | 8548 | struct task_group *tg = css_tg(css); |
68318b8e | 8549 | |
c8b28116 | 8550 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8551 | } |
ab84d31e PT |
8552 | |
8553 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8554 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8555 | ||
ab84d31e | 8556 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 8557 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
8558 | /* More than 203 days if BW_SHIFT equals 20. */ |
8559 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 8560 | |
a790de99 PT |
8561 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8562 | ||
ab84d31e PT |
8563 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8564 | { | |
56f570e5 | 8565 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8566 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8567 | |
8568 | if (tg == &root_task_group) | |
8569 | return -EINVAL; | |
8570 | ||
8571 | /* | |
8572 | * Ensure we have at some amount of bandwidth every period. This is | |
8573 | * to prevent reaching a state of large arrears when throttled via | |
8574 | * entity_tick() resulting in prolonged exit starvation. | |
8575 | */ | |
8576 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8577 | return -EINVAL; | |
8578 | ||
8579 | /* | |
8580 | * Likewise, bound things on the otherside by preventing insane quota | |
8581 | * periods. This also allows us to normalize in computing quota | |
8582 | * feasibility. | |
8583 | */ | |
8584 | if (period > max_cfs_quota_period) | |
8585 | return -EINVAL; | |
8586 | ||
d505b8af HC |
8587 | /* |
8588 | * Bound quota to defend quota against overflow during bandwidth shift. | |
8589 | */ | |
8590 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
8591 | return -EINVAL; | |
8592 | ||
0e59bdae KT |
8593 | /* |
8594 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8595 | * unthrottle_offline_cfs_rqs(). | |
8596 | */ | |
8597 | get_online_cpus(); | |
a790de99 PT |
8598 | mutex_lock(&cfs_constraints_mutex); |
8599 | ret = __cfs_schedulable(tg, period, quota); | |
8600 | if (ret) | |
8601 | goto out_unlock; | |
8602 | ||
58088ad0 | 8603 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8604 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8605 | /* |
8606 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8607 | * before making related changes, and on->off must occur afterwards | |
8608 | */ | |
8609 | if (runtime_enabled && !runtime_was_enabled) | |
8610 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8611 | raw_spin_lock_irq(&cfs_b->lock); |
8612 | cfs_b->period = ns_to_ktime(period); | |
8613 | cfs_b->quota = quota; | |
58088ad0 | 8614 | |
a9cf55b2 | 8615 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
8616 | |
8617 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
8618 | if (runtime_enabled) |
8619 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 8620 | |
ab84d31e PT |
8621 | raw_spin_unlock_irq(&cfs_b->lock); |
8622 | ||
0e59bdae | 8623 | for_each_online_cpu(i) { |
ab84d31e | 8624 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8625 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 8626 | struct rq_flags rf; |
ab84d31e | 8627 | |
8a8c69c3 | 8628 | rq_lock_irq(rq, &rf); |
58088ad0 | 8629 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8630 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8631 | |
029632fb | 8632 | if (cfs_rq->throttled) |
671fd9da | 8633 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 8634 | rq_unlock_irq(rq, &rf); |
ab84d31e | 8635 | } |
1ee14e6c BS |
8636 | if (runtime_was_enabled && !runtime_enabled) |
8637 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8638 | out_unlock: |
8639 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8640 | put_online_cpus(); |
ab84d31e | 8641 | |
a790de99 | 8642 | return ret; |
ab84d31e PT |
8643 | } |
8644 | ||
b1546edc | 8645 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e PT |
8646 | { |
8647 | u64 quota, period; | |
8648 | ||
029632fb | 8649 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8650 | if (cfs_quota_us < 0) |
8651 | quota = RUNTIME_INF; | |
1a8b4540 | 8652 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 8653 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
8654 | else |
8655 | return -EINVAL; | |
ab84d31e PT |
8656 | |
8657 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8658 | } | |
8659 | ||
b1546edc | 8660 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
8661 | { |
8662 | u64 quota_us; | |
8663 | ||
029632fb | 8664 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8665 | return -1; |
8666 | ||
029632fb | 8667 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8668 | do_div(quota_us, NSEC_PER_USEC); |
8669 | ||
8670 | return quota_us; | |
8671 | } | |
8672 | ||
b1546edc | 8673 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e PT |
8674 | { |
8675 | u64 quota, period; | |
8676 | ||
1a8b4540 KK |
8677 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
8678 | return -EINVAL; | |
8679 | ||
ab84d31e | 8680 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 8681 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8682 | |
ab84d31e PT |
8683 | return tg_set_cfs_bandwidth(tg, period, quota); |
8684 | } | |
8685 | ||
b1546edc | 8686 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
8687 | { |
8688 | u64 cfs_period_us; | |
8689 | ||
029632fb | 8690 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8691 | do_div(cfs_period_us, NSEC_PER_USEC); |
8692 | ||
8693 | return cfs_period_us; | |
8694 | } | |
8695 | ||
182446d0 TH |
8696 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8697 | struct cftype *cft) | |
ab84d31e | 8698 | { |
182446d0 | 8699 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8700 | } |
8701 | ||
182446d0 TH |
8702 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8703 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8704 | { |
182446d0 | 8705 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8706 | } |
8707 | ||
182446d0 TH |
8708 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8709 | struct cftype *cft) | |
ab84d31e | 8710 | { |
182446d0 | 8711 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8712 | } |
8713 | ||
182446d0 TH |
8714 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8715 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8716 | { |
182446d0 | 8717 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8718 | } |
8719 | ||
a790de99 PT |
8720 | struct cfs_schedulable_data { |
8721 | struct task_group *tg; | |
8722 | u64 period, quota; | |
8723 | }; | |
8724 | ||
8725 | /* | |
8726 | * normalize group quota/period to be quota/max_period | |
8727 | * note: units are usecs | |
8728 | */ | |
8729 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8730 | struct cfs_schedulable_data *d) | |
8731 | { | |
8732 | u64 quota, period; | |
8733 | ||
8734 | if (tg == d->tg) { | |
8735 | period = d->period; | |
8736 | quota = d->quota; | |
8737 | } else { | |
8738 | period = tg_get_cfs_period(tg); | |
8739 | quota = tg_get_cfs_quota(tg); | |
8740 | } | |
8741 | ||
8742 | /* note: these should typically be equivalent */ | |
8743 | if (quota == RUNTIME_INF || quota == -1) | |
8744 | return RUNTIME_INF; | |
8745 | ||
8746 | return to_ratio(period, quota); | |
8747 | } | |
8748 | ||
8749 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8750 | { | |
8751 | struct cfs_schedulable_data *d = data; | |
029632fb | 8752 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8753 | s64 quota = 0, parent_quota = -1; |
8754 | ||
8755 | if (!tg->parent) { | |
8756 | quota = RUNTIME_INF; | |
8757 | } else { | |
029632fb | 8758 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8759 | |
8760 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8761 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8762 | |
8763 | /* | |
c53593e5 TH |
8764 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
8765 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 8766 | * limit is set: |
a790de99 | 8767 | */ |
c53593e5 TH |
8768 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
8769 | quota = min(quota, parent_quota); | |
8770 | } else { | |
8771 | if (quota == RUNTIME_INF) | |
8772 | quota = parent_quota; | |
8773 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8774 | return -EINVAL; | |
8775 | } | |
a790de99 | 8776 | } |
9c58c79a | 8777 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8778 | |
8779 | return 0; | |
8780 | } | |
8781 | ||
8782 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8783 | { | |
8277434e | 8784 | int ret; |
a790de99 PT |
8785 | struct cfs_schedulable_data data = { |
8786 | .tg = tg, | |
8787 | .period = period, | |
8788 | .quota = quota, | |
8789 | }; | |
8790 | ||
8791 | if (quota != RUNTIME_INF) { | |
8792 | do_div(data.period, NSEC_PER_USEC); | |
8793 | do_div(data.quota, NSEC_PER_USEC); | |
8794 | } | |
8795 | ||
8277434e PT |
8796 | rcu_read_lock(); |
8797 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8798 | rcu_read_unlock(); | |
8799 | ||
8800 | return ret; | |
a790de99 | 8801 | } |
e8da1b18 | 8802 | |
a1f7164c | 8803 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 8804 | { |
2da8ca82 | 8805 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8806 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8807 | |
44ffc75b TH |
8808 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8809 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8810 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 8811 | |
3d6c50c2 YW |
8812 | if (schedstat_enabled() && tg != &root_task_group) { |
8813 | u64 ws = 0; | |
8814 | int i; | |
8815 | ||
8816 | for_each_possible_cpu(i) | |
8817 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
8818 | ||
8819 | seq_printf(sf, "wait_sum %llu\n", ws); | |
8820 | } | |
8821 | ||
e8da1b18 NR |
8822 | return 0; |
8823 | } | |
ab84d31e | 8824 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8825 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8826 | |
052f1dc7 | 8827 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8828 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8829 | struct cftype *cft, s64 val) | |
6f505b16 | 8830 | { |
182446d0 | 8831 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8832 | } |
8833 | ||
182446d0 TH |
8834 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8835 | struct cftype *cft) | |
6f505b16 | 8836 | { |
182446d0 | 8837 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8838 | } |
d0b27fa7 | 8839 | |
182446d0 TH |
8840 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8841 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8842 | { |
182446d0 | 8843 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8844 | } |
8845 | ||
182446d0 TH |
8846 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8847 | struct cftype *cft) | |
d0b27fa7 | 8848 | { |
182446d0 | 8849 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8850 | } |
6d6bc0ad | 8851 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8852 | |
a1f7164c | 8853 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 8854 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8855 | { |
8856 | .name = "shares", | |
f4c753b7 PM |
8857 | .read_u64 = cpu_shares_read_u64, |
8858 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8859 | }, |
052f1dc7 | 8860 | #endif |
ab84d31e PT |
8861 | #ifdef CONFIG_CFS_BANDWIDTH |
8862 | { | |
8863 | .name = "cfs_quota_us", | |
8864 | .read_s64 = cpu_cfs_quota_read_s64, | |
8865 | .write_s64 = cpu_cfs_quota_write_s64, | |
8866 | }, | |
8867 | { | |
8868 | .name = "cfs_period_us", | |
8869 | .read_u64 = cpu_cfs_period_read_u64, | |
8870 | .write_u64 = cpu_cfs_period_write_u64, | |
8871 | }, | |
e8da1b18 NR |
8872 | { |
8873 | .name = "stat", | |
a1f7164c | 8874 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 8875 | }, |
ab84d31e | 8876 | #endif |
052f1dc7 | 8877 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8878 | { |
9f0c1e56 | 8879 | .name = "rt_runtime_us", |
06ecb27c PM |
8880 | .read_s64 = cpu_rt_runtime_read, |
8881 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8882 | }, |
d0b27fa7 PZ |
8883 | { |
8884 | .name = "rt_period_us", | |
f4c753b7 PM |
8885 | .read_u64 = cpu_rt_period_read_uint, |
8886 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8887 | }, |
2480c093 PB |
8888 | #endif |
8889 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8890 | { | |
8891 | .name = "uclamp.min", | |
8892 | .flags = CFTYPE_NOT_ON_ROOT, | |
8893 | .seq_show = cpu_uclamp_min_show, | |
8894 | .write = cpu_uclamp_min_write, | |
8895 | }, | |
8896 | { | |
8897 | .name = "uclamp.max", | |
8898 | .flags = CFTYPE_NOT_ON_ROOT, | |
8899 | .seq_show = cpu_uclamp_max_show, | |
8900 | .write = cpu_uclamp_max_write, | |
8901 | }, | |
052f1dc7 | 8902 | #endif |
d1ccc66d | 8903 | { } /* Terminate */ |
68318b8e SV |
8904 | }; |
8905 | ||
d41bf8c9 TH |
8906 | static int cpu_extra_stat_show(struct seq_file *sf, |
8907 | struct cgroup_subsys_state *css) | |
0d593634 | 8908 | { |
0d593634 TH |
8909 | #ifdef CONFIG_CFS_BANDWIDTH |
8910 | { | |
d41bf8c9 | 8911 | struct task_group *tg = css_tg(css); |
0d593634 TH |
8912 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
8913 | u64 throttled_usec; | |
8914 | ||
8915 | throttled_usec = cfs_b->throttled_time; | |
8916 | do_div(throttled_usec, NSEC_PER_USEC); | |
8917 | ||
8918 | seq_printf(sf, "nr_periods %d\n" | |
8919 | "nr_throttled %d\n" | |
8920 | "throttled_usec %llu\n", | |
8921 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
8922 | throttled_usec); | |
8923 | } | |
8924 | #endif | |
8925 | return 0; | |
8926 | } | |
8927 | ||
8928 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8929 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
8930 | struct cftype *cft) | |
8931 | { | |
8932 | struct task_group *tg = css_tg(css); | |
8933 | u64 weight = scale_load_down(tg->shares); | |
8934 | ||
8935 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
8936 | } | |
8937 | ||
8938 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
8939 | struct cftype *cft, u64 weight) | |
8940 | { | |
8941 | /* | |
8942 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
8943 | * values which are 1, 100 and 10000 respectively. While it loses | |
8944 | * a bit of range on both ends, it maps pretty well onto the shares | |
8945 | * value used by scheduler and the round-trip conversions preserve | |
8946 | * the original value over the entire range. | |
8947 | */ | |
8948 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
8949 | return -ERANGE; | |
8950 | ||
8951 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
8952 | ||
8953 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
8954 | } | |
8955 | ||
8956 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
8957 | struct cftype *cft) | |
8958 | { | |
8959 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
8960 | int last_delta = INT_MAX; | |
8961 | int prio, delta; | |
8962 | ||
8963 | /* find the closest nice value to the current weight */ | |
8964 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
8965 | delta = abs(sched_prio_to_weight[prio] - weight); | |
8966 | if (delta >= last_delta) | |
8967 | break; | |
8968 | last_delta = delta; | |
8969 | } | |
8970 | ||
8971 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
8972 | } | |
8973 | ||
8974 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
8975 | struct cftype *cft, s64 nice) | |
8976 | { | |
8977 | unsigned long weight; | |
7281c8de | 8978 | int idx; |
0d593634 TH |
8979 | |
8980 | if (nice < MIN_NICE || nice > MAX_NICE) | |
8981 | return -ERANGE; | |
8982 | ||
7281c8de PZ |
8983 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
8984 | idx = array_index_nospec(idx, 40); | |
8985 | weight = sched_prio_to_weight[idx]; | |
8986 | ||
0d593634 TH |
8987 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
8988 | } | |
8989 | #endif | |
8990 | ||
8991 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
8992 | long period, long quota) | |
8993 | { | |
8994 | if (quota < 0) | |
8995 | seq_puts(sf, "max"); | |
8996 | else | |
8997 | seq_printf(sf, "%ld", quota); | |
8998 | ||
8999 | seq_printf(sf, " %ld\n", period); | |
9000 | } | |
9001 | ||
9002 | /* caller should put the current value in *@periodp before calling */ | |
9003 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
9004 | u64 *periodp, u64 *quotap) | |
9005 | { | |
9006 | char tok[21]; /* U64_MAX */ | |
9007 | ||
4c47acd8 | 9008 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
9009 | return -EINVAL; |
9010 | ||
9011 | *periodp *= NSEC_PER_USEC; | |
9012 | ||
9013 | if (sscanf(tok, "%llu", quotap)) | |
9014 | *quotap *= NSEC_PER_USEC; | |
9015 | else if (!strcmp(tok, "max")) | |
9016 | *quotap = RUNTIME_INF; | |
9017 | else | |
9018 | return -EINVAL; | |
9019 | ||
9020 | return 0; | |
9021 | } | |
9022 | ||
9023 | #ifdef CONFIG_CFS_BANDWIDTH | |
9024 | static int cpu_max_show(struct seq_file *sf, void *v) | |
9025 | { | |
9026 | struct task_group *tg = css_tg(seq_css(sf)); | |
9027 | ||
9028 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
9029 | return 0; | |
9030 | } | |
9031 | ||
9032 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
9033 | char *buf, size_t nbytes, loff_t off) | |
9034 | { | |
9035 | struct task_group *tg = css_tg(of_css(of)); | |
9036 | u64 period = tg_get_cfs_period(tg); | |
9037 | u64 quota; | |
9038 | int ret; | |
9039 | ||
9040 | ret = cpu_period_quota_parse(buf, &period, "a); | |
9041 | if (!ret) | |
9042 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
9043 | return ret ?: nbytes; | |
9044 | } | |
9045 | #endif | |
9046 | ||
9047 | static struct cftype cpu_files[] = { | |
0d593634 TH |
9048 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9049 | { | |
9050 | .name = "weight", | |
9051 | .flags = CFTYPE_NOT_ON_ROOT, | |
9052 | .read_u64 = cpu_weight_read_u64, | |
9053 | .write_u64 = cpu_weight_write_u64, | |
9054 | }, | |
9055 | { | |
9056 | .name = "weight.nice", | |
9057 | .flags = CFTYPE_NOT_ON_ROOT, | |
9058 | .read_s64 = cpu_weight_nice_read_s64, | |
9059 | .write_s64 = cpu_weight_nice_write_s64, | |
9060 | }, | |
9061 | #endif | |
9062 | #ifdef CONFIG_CFS_BANDWIDTH | |
9063 | { | |
9064 | .name = "max", | |
9065 | .flags = CFTYPE_NOT_ON_ROOT, | |
9066 | .seq_show = cpu_max_show, | |
9067 | .write = cpu_max_write, | |
9068 | }, | |
2480c093 PB |
9069 | #endif |
9070 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9071 | { | |
9072 | .name = "uclamp.min", | |
9073 | .flags = CFTYPE_NOT_ON_ROOT, | |
9074 | .seq_show = cpu_uclamp_min_show, | |
9075 | .write = cpu_uclamp_min_write, | |
9076 | }, | |
9077 | { | |
9078 | .name = "uclamp.max", | |
9079 | .flags = CFTYPE_NOT_ON_ROOT, | |
9080 | .seq_show = cpu_uclamp_max_show, | |
9081 | .write = cpu_uclamp_max_write, | |
9082 | }, | |
0d593634 TH |
9083 | #endif |
9084 | { } /* terminate */ | |
9085 | }; | |
9086 | ||
073219e9 | 9087 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 9088 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 9089 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 9090 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 9091 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 9092 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 9093 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
9094 | .can_attach = cpu_cgroup_can_attach, |
9095 | .attach = cpu_cgroup_attach, | |
a1f7164c | 9096 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 9097 | .dfl_cftypes = cpu_files, |
b38e42e9 | 9098 | .early_init = true, |
0d593634 | 9099 | .threaded = true, |
68318b8e SV |
9100 | }; |
9101 | ||
052f1dc7 | 9102 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 9103 | |
b637a328 PM |
9104 | void dump_cpu_task(int cpu) |
9105 | { | |
9106 | pr_info("Task dump for CPU %d:\n", cpu); | |
9107 | sched_show_task(cpu_curr(cpu)); | |
9108 | } | |
ed82b8a1 AK |
9109 | |
9110 | /* | |
9111 | * Nice levels are multiplicative, with a gentle 10% change for every | |
9112 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
9113 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
9114 | * that remained on nice 0. | |
9115 | * | |
9116 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
9117 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
9118 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
9119 | * If a task goes up by ~10% and another task goes down by ~10% then | |
9120 | * the relative distance between them is ~25%.) | |
9121 | */ | |
9122 | const int sched_prio_to_weight[40] = { | |
9123 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
9124 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
9125 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
9126 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
9127 | /* 0 */ 1024, 820, 655, 526, 423, | |
9128 | /* 5 */ 335, 272, 215, 172, 137, | |
9129 | /* 10 */ 110, 87, 70, 56, 45, | |
9130 | /* 15 */ 36, 29, 23, 18, 15, | |
9131 | }; | |
9132 | ||
9133 | /* | |
9134 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
9135 | * | |
9136 | * In cases where the weight does not change often, we can use the | |
9137 | * precalculated inverse to speed up arithmetics by turning divisions | |
9138 | * into multiplications: | |
9139 | */ | |
9140 | const u32 sched_prio_to_wmult[40] = { | |
9141 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
9142 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
9143 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
9144 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
9145 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
9146 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
9147 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
9148 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
9149 | }; | |
14a7405b | 9150 | |
9d246053 PA |
9151 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
9152 | { | |
9153 | trace_sched_update_nr_running_tp(rq, count); | |
9154 | } |