]>
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 | |
100 | * always looks at the local rq data structures to find the most elegible task | |
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 | ||
90b5363a PZI |
323 | static inline void |
324 | rq_csd_init(struct rq *rq, call_single_data_t *csd, smp_call_func_t func) | |
325 | { | |
326 | csd->flags = 0; | |
327 | csd->func = func; | |
328 | csd->info = rq; | |
329 | } | |
535b9552 | 330 | |
8f4d37ec PZ |
331 | #ifdef CONFIG_SCHED_HRTICK |
332 | /* | |
333 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 334 | */ |
8f4d37ec | 335 | |
8f4d37ec PZ |
336 | static void hrtick_clear(struct rq *rq) |
337 | { | |
338 | if (hrtimer_active(&rq->hrtick_timer)) | |
339 | hrtimer_cancel(&rq->hrtick_timer); | |
340 | } | |
341 | ||
8f4d37ec PZ |
342 | /* |
343 | * High-resolution timer tick. | |
344 | * Runs from hardirq context with interrupts disabled. | |
345 | */ | |
346 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
347 | { | |
348 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 349 | struct rq_flags rf; |
8f4d37ec PZ |
350 | |
351 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
352 | ||
8a8c69c3 | 353 | rq_lock(rq, &rf); |
3e51f33f | 354 | update_rq_clock(rq); |
8f4d37ec | 355 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 356 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
357 | |
358 | return HRTIMER_NORESTART; | |
359 | } | |
360 | ||
95e904c7 | 361 | #ifdef CONFIG_SMP |
971ee28c | 362 | |
4961b6e1 | 363 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
364 | { |
365 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 366 | |
d5096aa6 | 367 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
368 | } |
369 | ||
31656519 PZ |
370 | /* |
371 | * called from hardirq (IPI) context | |
372 | */ | |
373 | static void __hrtick_start(void *arg) | |
b328ca18 | 374 | { |
31656519 | 375 | struct rq *rq = arg; |
8a8c69c3 | 376 | struct rq_flags rf; |
b328ca18 | 377 | |
8a8c69c3 | 378 | rq_lock(rq, &rf); |
971ee28c | 379 | __hrtick_restart(rq); |
8a8c69c3 | 380 | rq_unlock(rq, &rf); |
b328ca18 PZ |
381 | } |
382 | ||
31656519 PZ |
383 | /* |
384 | * Called to set the hrtick timer state. | |
385 | * | |
386 | * called with rq->lock held and irqs disabled | |
387 | */ | |
029632fb | 388 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 389 | { |
31656519 | 390 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 391 | ktime_t time; |
392 | s64 delta; | |
393 | ||
394 | /* | |
395 | * Don't schedule slices shorter than 10000ns, that just | |
396 | * doesn't make sense and can cause timer DoS. | |
397 | */ | |
398 | delta = max_t(s64, delay, 10000LL); | |
399 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 400 | |
cc584b21 | 401 | hrtimer_set_expires(timer, time); |
31656519 | 402 | |
fd3eafda | 403 | if (rq == this_rq()) |
971ee28c | 404 | __hrtick_restart(rq); |
fd3eafda | 405 | else |
c46fff2a | 406 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
407 | } |
408 | ||
31656519 PZ |
409 | #else |
410 | /* | |
411 | * Called to set the hrtick timer state. | |
412 | * | |
413 | * called with rq->lock held and irqs disabled | |
414 | */ | |
029632fb | 415 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 416 | { |
86893335 WL |
417 | /* |
418 | * Don't schedule slices shorter than 10000ns, that just | |
419 | * doesn't make sense. Rely on vruntime for fairness. | |
420 | */ | |
421 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 422 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 423 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 424 | } |
90b5363a | 425 | |
31656519 | 426 | #endif /* CONFIG_SMP */ |
8f4d37ec | 427 | |
77a021be | 428 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 429 | { |
31656519 | 430 | #ifdef CONFIG_SMP |
90b5363a | 431 | rq_csd_init(rq, &rq->hrtick_csd, __hrtick_start); |
31656519 | 432 | #endif |
d5096aa6 | 433 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 434 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 435 | } |
006c75f1 | 436 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
437 | static inline void hrtick_clear(struct rq *rq) |
438 | { | |
439 | } | |
440 | ||
77a021be | 441 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
442 | { |
443 | } | |
006c75f1 | 444 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 445 | |
5529578a FW |
446 | /* |
447 | * cmpxchg based fetch_or, macro so it works for different integer types | |
448 | */ | |
449 | #define fetch_or(ptr, mask) \ | |
450 | ({ \ | |
451 | typeof(ptr) _ptr = (ptr); \ | |
452 | typeof(mask) _mask = (mask); \ | |
453 | typeof(*_ptr) _old, _val = *_ptr; \ | |
454 | \ | |
455 | for (;;) { \ | |
456 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
457 | if (_old == _val) \ | |
458 | break; \ | |
459 | _val = _old; \ | |
460 | } \ | |
461 | _old; \ | |
462 | }) | |
463 | ||
e3baac47 | 464 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
465 | /* |
466 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
467 | * this avoids any races wrt polling state changes and thereby avoids | |
468 | * spurious IPIs. | |
469 | */ | |
470 | static bool set_nr_and_not_polling(struct task_struct *p) | |
471 | { | |
472 | struct thread_info *ti = task_thread_info(p); | |
473 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
474 | } | |
e3baac47 PZ |
475 | |
476 | /* | |
477 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
478 | * | |
479 | * If this returns true, then the idle task promises to call | |
480 | * sched_ttwu_pending() and reschedule soon. | |
481 | */ | |
482 | static bool set_nr_if_polling(struct task_struct *p) | |
483 | { | |
484 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 485 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
486 | |
487 | for (;;) { | |
488 | if (!(val & _TIF_POLLING_NRFLAG)) | |
489 | return false; | |
490 | if (val & _TIF_NEED_RESCHED) | |
491 | return true; | |
492 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
493 | if (old == val) | |
494 | break; | |
495 | val = old; | |
496 | } | |
497 | return true; | |
498 | } | |
499 | ||
fd99f91a PZ |
500 | #else |
501 | static bool set_nr_and_not_polling(struct task_struct *p) | |
502 | { | |
503 | set_tsk_need_resched(p); | |
504 | return true; | |
505 | } | |
e3baac47 PZ |
506 | |
507 | #ifdef CONFIG_SMP | |
508 | static bool set_nr_if_polling(struct task_struct *p) | |
509 | { | |
510 | return false; | |
511 | } | |
512 | #endif | |
fd99f91a PZ |
513 | #endif |
514 | ||
07879c6a | 515 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
516 | { |
517 | struct wake_q_node *node = &task->wake_q; | |
518 | ||
519 | /* | |
520 | * Atomically grab the task, if ->wake_q is !nil already it means | |
521 | * its already queued (either by us or someone else) and will get the | |
522 | * wakeup due to that. | |
523 | * | |
4c4e3731 PZ |
524 | * In order to ensure that a pending wakeup will observe our pending |
525 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 526 | */ |
4c4e3731 | 527 | smp_mb__before_atomic(); |
87ff19cb | 528 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 529 | return false; |
76751049 PZ |
530 | |
531 | /* | |
532 | * The head is context local, there can be no concurrency. | |
533 | */ | |
534 | *head->lastp = node; | |
535 | head->lastp = &node->next; | |
07879c6a DB |
536 | return true; |
537 | } | |
538 | ||
539 | /** | |
540 | * wake_q_add() - queue a wakeup for 'later' waking. | |
541 | * @head: the wake_q_head to add @task to | |
542 | * @task: the task to queue for 'later' wakeup | |
543 | * | |
544 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
545 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
546 | * instantly. | |
547 | * | |
548 | * This function must be used as-if it were wake_up_process(); IOW the task | |
549 | * must be ready to be woken at this location. | |
550 | */ | |
551 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
552 | { | |
553 | if (__wake_q_add(head, task)) | |
554 | get_task_struct(task); | |
555 | } | |
556 | ||
557 | /** | |
558 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
559 | * @head: the wake_q_head to add @task to | |
560 | * @task: the task to queue for 'later' wakeup | |
561 | * | |
562 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
563 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
564 | * instantly. | |
565 | * | |
566 | * This function must be used as-if it were wake_up_process(); IOW the task | |
567 | * must be ready to be woken at this location. | |
568 | * | |
569 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
570 | * that already hold reference to @task can call the 'safe' version and trust | |
571 | * wake_q to do the right thing depending whether or not the @task is already | |
572 | * queued for wakeup. | |
573 | */ | |
574 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
575 | { | |
576 | if (!__wake_q_add(head, task)) | |
577 | put_task_struct(task); | |
76751049 PZ |
578 | } |
579 | ||
580 | void wake_up_q(struct wake_q_head *head) | |
581 | { | |
582 | struct wake_q_node *node = head->first; | |
583 | ||
584 | while (node != WAKE_Q_TAIL) { | |
585 | struct task_struct *task; | |
586 | ||
587 | task = container_of(node, struct task_struct, wake_q); | |
588 | BUG_ON(!task); | |
d1ccc66d | 589 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
590 | node = node->next; |
591 | task->wake_q.next = NULL; | |
592 | ||
593 | /* | |
7696f991 AP |
594 | * wake_up_process() executes a full barrier, which pairs with |
595 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
596 | */ |
597 | wake_up_process(task); | |
598 | put_task_struct(task); | |
599 | } | |
600 | } | |
601 | ||
c24d20db | 602 | /* |
8875125e | 603 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
604 | * |
605 | * On UP this means the setting of the need_resched flag, on SMP it | |
606 | * might also involve a cross-CPU call to trigger the scheduler on | |
607 | * the target CPU. | |
608 | */ | |
8875125e | 609 | void resched_curr(struct rq *rq) |
c24d20db | 610 | { |
8875125e | 611 | struct task_struct *curr = rq->curr; |
c24d20db IM |
612 | int cpu; |
613 | ||
8875125e | 614 | lockdep_assert_held(&rq->lock); |
c24d20db | 615 | |
8875125e | 616 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
617 | return; |
618 | ||
8875125e | 619 | cpu = cpu_of(rq); |
fd99f91a | 620 | |
f27dde8d | 621 | if (cpu == smp_processor_id()) { |
8875125e | 622 | set_tsk_need_resched(curr); |
f27dde8d | 623 | set_preempt_need_resched(); |
c24d20db | 624 | return; |
f27dde8d | 625 | } |
c24d20db | 626 | |
8875125e | 627 | if (set_nr_and_not_polling(curr)) |
c24d20db | 628 | smp_send_reschedule(cpu); |
dfc68f29 AL |
629 | else |
630 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
631 | } |
632 | ||
029632fb | 633 | void resched_cpu(int cpu) |
c24d20db IM |
634 | { |
635 | struct rq *rq = cpu_rq(cpu); | |
636 | unsigned long flags; | |
637 | ||
7c2102e5 | 638 | raw_spin_lock_irqsave(&rq->lock, flags); |
a0982dfa PM |
639 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
640 | resched_curr(rq); | |
05fa785c | 641 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 642 | } |
06d8308c | 643 | |
b021fe3e | 644 | #ifdef CONFIG_SMP |
3451d024 | 645 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 646 | /* |
d1ccc66d IM |
647 | * In the semi idle case, use the nearest busy CPU for migrating timers |
648 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
649 | * |
650 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
651 | * selecting an idle CPU will add more delays to the timers than intended |
652 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 653 | */ |
bc7a34b8 | 654 | int get_nohz_timer_target(void) |
83cd4fe2 | 655 | { |
e938b9c9 | 656 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 VP |
657 | struct sched_domain *sd; |
658 | ||
e938b9c9 WL |
659 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
660 | if (!idle_cpu(cpu)) | |
661 | return cpu; | |
662 | default_cpu = cpu; | |
663 | } | |
6201b4d6 | 664 | |
057f3fad | 665 | rcu_read_lock(); |
83cd4fe2 | 666 | for_each_domain(cpu, sd) { |
e938b9c9 WL |
667 | for_each_cpu_and(i, sched_domain_span(sd), |
668 | housekeeping_cpumask(HK_FLAG_TIMER)) { | |
44496922 WL |
669 | if (cpu == i) |
670 | continue; | |
671 | ||
e938b9c9 | 672 | if (!idle_cpu(i)) { |
057f3fad PZ |
673 | cpu = i; |
674 | goto unlock; | |
675 | } | |
676 | } | |
83cd4fe2 | 677 | } |
9642d18e | 678 | |
e938b9c9 WL |
679 | if (default_cpu == -1) |
680 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
681 | cpu = default_cpu; | |
057f3fad PZ |
682 | unlock: |
683 | rcu_read_unlock(); | |
83cd4fe2 VP |
684 | return cpu; |
685 | } | |
d1ccc66d | 686 | |
06d8308c TG |
687 | /* |
688 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
689 | * idle CPU then this timer might expire before the next timer event | |
690 | * which is scheduled to wake up that CPU. In case of a completely | |
691 | * idle system the next event might even be infinite time into the | |
692 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
693 | * leaves the inner idle loop so the newly added timer is taken into | |
694 | * account when the CPU goes back to idle and evaluates the timer | |
695 | * wheel for the next timer event. | |
696 | */ | |
1c20091e | 697 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
698 | { |
699 | struct rq *rq = cpu_rq(cpu); | |
700 | ||
701 | if (cpu == smp_processor_id()) | |
702 | return; | |
703 | ||
67b9ca70 | 704 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 705 | smp_send_reschedule(cpu); |
dfc68f29 AL |
706 | else |
707 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
708 | } |
709 | ||
c5bfece2 | 710 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 711 | { |
53c5fa16 FW |
712 | /* |
713 | * We just need the target to call irq_exit() and re-evaluate | |
714 | * the next tick. The nohz full kick at least implies that. | |
715 | * If needed we can still optimize that later with an | |
716 | * empty IRQ. | |
717 | */ | |
379d9ecb PM |
718 | if (cpu_is_offline(cpu)) |
719 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 720 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
721 | if (cpu != smp_processor_id() || |
722 | tick_nohz_tick_stopped()) | |
53c5fa16 | 723 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
724 | return true; |
725 | } | |
726 | ||
727 | return false; | |
728 | } | |
729 | ||
379d9ecb PM |
730 | /* |
731 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
732 | * caller's responsibility to deal with the lost wakeup, for example, | |
733 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
734 | */ | |
1c20091e FW |
735 | void wake_up_nohz_cpu(int cpu) |
736 | { | |
c5bfece2 | 737 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
738 | wake_up_idle_cpu(cpu); |
739 | } | |
740 | ||
19a1f5ec | 741 | static void nohz_csd_func(void *info) |
45bf76df | 742 | { |
19a1f5ec PZ |
743 | struct rq *rq = info; |
744 | int cpu = cpu_of(rq); | |
745 | unsigned int flags; | |
873b4c65 VG |
746 | |
747 | /* | |
19a1f5ec | 748 | * Release the rq::nohz_csd. |
873b4c65 | 749 | */ |
19a1f5ec PZ |
750 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(cpu)); |
751 | WARN_ON(!(flags & NOHZ_KICK_MASK)); | |
45bf76df | 752 | |
19a1f5ec PZ |
753 | rq->idle_balance = idle_cpu(cpu); |
754 | if (rq->idle_balance && !need_resched()) { | |
755 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
756 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
757 | } | |
2069dd75 PZ |
758 | } |
759 | ||
3451d024 | 760 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 761 | |
ce831b38 | 762 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 763 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 764 | { |
76d92ac3 FW |
765 | int fifo_nr_running; |
766 | ||
767 | /* Deadline tasks, even if single, need the tick */ | |
768 | if (rq->dl.dl_nr_running) | |
769 | return false; | |
770 | ||
1e78cdbd | 771 | /* |
2548d546 PZ |
772 | * If there are more than one RR tasks, we need the tick to effect the |
773 | * actual RR behaviour. | |
1e78cdbd | 774 | */ |
76d92ac3 FW |
775 | if (rq->rt.rr_nr_running) { |
776 | if (rq->rt.rr_nr_running == 1) | |
777 | return true; | |
778 | else | |
779 | return false; | |
1e78cdbd RR |
780 | } |
781 | ||
2548d546 PZ |
782 | /* |
783 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
784 | * forced preemption between FIFO tasks. | |
785 | */ | |
786 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
787 | if (fifo_nr_running) | |
788 | return true; | |
789 | ||
790 | /* | |
791 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
792 | * if there's more than one we need the tick for involuntary | |
793 | * preemption. | |
794 | */ | |
795 | if (rq->nr_running > 1) | |
541b8264 | 796 | return false; |
ce831b38 | 797 | |
541b8264 | 798 | return true; |
ce831b38 FW |
799 | } |
800 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 801 | #endif /* CONFIG_SMP */ |
18d95a28 | 802 | |
a790de99 PT |
803 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
804 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 805 | /* |
8277434e PT |
806 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
807 | * node and @up when leaving it for the final time. | |
808 | * | |
809 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 810 | */ |
029632fb | 811 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 812 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
813 | { |
814 | struct task_group *parent, *child; | |
eb755805 | 815 | int ret; |
c09595f6 | 816 | |
8277434e PT |
817 | parent = from; |
818 | ||
c09595f6 | 819 | down: |
eb755805 PZ |
820 | ret = (*down)(parent, data); |
821 | if (ret) | |
8277434e | 822 | goto out; |
c09595f6 PZ |
823 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
824 | parent = child; | |
825 | goto down; | |
826 | ||
827 | up: | |
828 | continue; | |
829 | } | |
eb755805 | 830 | ret = (*up)(parent, data); |
8277434e PT |
831 | if (ret || parent == from) |
832 | goto out; | |
c09595f6 PZ |
833 | |
834 | child = parent; | |
835 | parent = parent->parent; | |
836 | if (parent) | |
837 | goto up; | |
8277434e | 838 | out: |
eb755805 | 839 | return ret; |
c09595f6 PZ |
840 | } |
841 | ||
029632fb | 842 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 843 | { |
e2b245f8 | 844 | return 0; |
eb755805 | 845 | } |
18d95a28 PZ |
846 | #endif |
847 | ||
9059393e | 848 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 849 | { |
f05998d4 NR |
850 | int prio = p->static_prio - MAX_RT_PRIO; |
851 | struct load_weight *load = &p->se.load; | |
852 | ||
dd41f596 IM |
853 | /* |
854 | * SCHED_IDLE tasks get minimal weight: | |
855 | */ | |
1da1843f | 856 | if (task_has_idle_policy(p)) { |
c8b28116 | 857 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 858 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
859 | return; |
860 | } | |
71f8bd46 | 861 | |
9059393e VG |
862 | /* |
863 | * SCHED_OTHER tasks have to update their load when changing their | |
864 | * weight | |
865 | */ | |
866 | if (update_load && p->sched_class == &fair_sched_class) { | |
867 | reweight_task(p, prio); | |
868 | } else { | |
869 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
870 | load->inv_weight = sched_prio_to_wmult[prio]; | |
871 | } | |
71f8bd46 IM |
872 | } |
873 | ||
69842cba | 874 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
875 | /* |
876 | * Serializes updates of utilization clamp values | |
877 | * | |
878 | * The (slow-path) user-space triggers utilization clamp value updates which | |
879 | * can require updates on (fast-path) scheduler's data structures used to | |
880 | * support enqueue/dequeue operations. | |
881 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
882 | * requests are serialized using a mutex to reduce the risk of conflicting | |
883 | * updates or API abuses. | |
884 | */ | |
885 | static DEFINE_MUTEX(uclamp_mutex); | |
886 | ||
e8f14172 PB |
887 | /* Max allowed minimum utilization */ |
888 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
889 | ||
890 | /* Max allowed maximum utilization */ | |
891 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
892 | ||
13685c4a QY |
893 | /* |
894 | * By default RT tasks run at the maximum performance point/capacity of the | |
895 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
896 | * SCHED_CAPACITY_SCALE. | |
897 | * | |
898 | * This knob allows admins to change the default behavior when uclamp is being | |
899 | * used. In battery powered devices, particularly, running at the maximum | |
900 | * capacity and frequency will increase energy consumption and shorten the | |
901 | * battery life. | |
902 | * | |
903 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
904 | * | |
905 | * This knob will not override the system default sched_util_clamp_min defined | |
906 | * above. | |
907 | */ | |
908 | unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; | |
909 | ||
e8f14172 PB |
910 | /* All clamps are required to be less or equal than these values */ |
911 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 912 | |
46609ce2 QY |
913 | /* |
914 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
915 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
916 | * enqueue/dequeue_task(). | |
917 | * | |
918 | * This allows users to continue to enable uclamp in their kernel config with | |
919 | * minimum uclamp overhead in the fast path. | |
920 | * | |
921 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
922 | * enabled, since we have an actual users that make use of uclamp | |
923 | * functionality. | |
924 | * | |
925 | * The knobs that would enable this static key are: | |
926 | * | |
927 | * * A task modifying its uclamp value with sched_setattr(). | |
928 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
929 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
930 | */ | |
931 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
932 | ||
69842cba PB |
933 | /* Integer rounded range for each bucket */ |
934 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
935 | ||
936 | #define for_each_clamp_id(clamp_id) \ | |
937 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
938 | ||
939 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
940 | { | |
941 | return clamp_value / UCLAMP_BUCKET_DELTA; | |
942 | } | |
943 | ||
7763baac | 944 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
945 | { |
946 | if (clamp_id == UCLAMP_MIN) | |
947 | return 0; | |
948 | return SCHED_CAPACITY_SCALE; | |
949 | } | |
950 | ||
a509a7cd PB |
951 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
952 | unsigned int value, bool user_defined) | |
69842cba PB |
953 | { |
954 | uc_se->value = value; | |
955 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 956 | uc_se->user_defined = user_defined; |
69842cba PB |
957 | } |
958 | ||
e496187d | 959 | static inline unsigned int |
0413d7f3 | 960 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
961 | unsigned int clamp_value) |
962 | { | |
963 | /* | |
964 | * Avoid blocked utilization pushing up the frequency when we go | |
965 | * idle (which drops the max-clamp) by retaining the last known | |
966 | * max-clamp. | |
967 | */ | |
968 | if (clamp_id == UCLAMP_MAX) { | |
969 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
970 | return clamp_value; | |
971 | } | |
972 | ||
973 | return uclamp_none(UCLAMP_MIN); | |
974 | } | |
975 | ||
0413d7f3 | 976 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
977 | unsigned int clamp_value) |
978 | { | |
979 | /* Reset max-clamp retention only on idle exit */ | |
980 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
981 | return; | |
982 | ||
983 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
984 | } | |
985 | ||
69842cba | 986 | static inline |
7763baac | 987 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 988 | unsigned int clamp_value) |
69842cba PB |
989 | { |
990 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
991 | int bucket_id = UCLAMP_BUCKETS - 1; | |
992 | ||
993 | /* | |
994 | * Since both min and max clamps are max aggregated, find the | |
995 | * top most bucket with tasks in. | |
996 | */ | |
997 | for ( ; bucket_id >= 0; bucket_id--) { | |
998 | if (!bucket[bucket_id].tasks) | |
999 | continue; | |
1000 | return bucket[bucket_id].value; | |
1001 | } | |
1002 | ||
1003 | /* No tasks -- default clamp values */ | |
e496187d | 1004 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
1005 | } |
1006 | ||
13685c4a QY |
1007 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
1008 | { | |
1009 | unsigned int default_util_min; | |
1010 | struct uclamp_se *uc_se; | |
1011 | ||
1012 | lockdep_assert_held(&p->pi_lock); | |
1013 | ||
1014 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1015 | ||
1016 | /* Only sync if user didn't override the default */ | |
1017 | if (uc_se->user_defined) | |
1018 | return; | |
1019 | ||
1020 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1021 | uclamp_se_set(uc_se, default_util_min, false); | |
1022 | } | |
1023 | ||
1024 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1025 | { | |
1026 | struct rq_flags rf; | |
1027 | struct rq *rq; | |
1028 | ||
1029 | if (!rt_task(p)) | |
1030 | return; | |
1031 | ||
1032 | /* Protect updates to p->uclamp_* */ | |
1033 | rq = task_rq_lock(p, &rf); | |
1034 | __uclamp_update_util_min_rt_default(p); | |
1035 | task_rq_unlock(rq, p, &rf); | |
1036 | } | |
1037 | ||
1038 | static void uclamp_sync_util_min_rt_default(void) | |
1039 | { | |
1040 | struct task_struct *g, *p; | |
1041 | ||
1042 | /* | |
1043 | * copy_process() sysctl_uclamp | |
1044 | * uclamp_min_rt = X; | |
1045 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1046 | * // link thread smp_mb__after_spinlock() | |
1047 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1048 | * sched_post_fork() for_each_process_thread() | |
1049 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1050 | * | |
1051 | * Ensures that either sched_post_fork() will observe the new | |
1052 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1053 | * task. | |
1054 | */ | |
1055 | read_lock(&tasklist_lock); | |
1056 | smp_mb__after_spinlock(); | |
1057 | read_unlock(&tasklist_lock); | |
1058 | ||
1059 | rcu_read_lock(); | |
1060 | for_each_process_thread(g, p) | |
1061 | uclamp_update_util_min_rt_default(p); | |
1062 | rcu_read_unlock(); | |
1063 | } | |
1064 | ||
3eac870a | 1065 | static inline struct uclamp_se |
0413d7f3 | 1066 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a PB |
1067 | { |
1068 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; | |
1069 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
1070 | struct uclamp_se uc_max; | |
1071 | ||
1072 | /* | |
1073 | * Tasks in autogroups or root task group will be | |
1074 | * restricted by system defaults. | |
1075 | */ | |
1076 | if (task_group_is_autogroup(task_group(p))) | |
1077 | return uc_req; | |
1078 | if (task_group(p) == &root_task_group) | |
1079 | return uc_req; | |
1080 | ||
1081 | uc_max = task_group(p)->uclamp[clamp_id]; | |
1082 | if (uc_req.value > uc_max.value || !uc_req.user_defined) | |
1083 | return uc_max; | |
1084 | #endif | |
1085 | ||
1086 | return uc_req; | |
1087 | } | |
1088 | ||
e8f14172 PB |
1089 | /* |
1090 | * The effective clamp bucket index of a task depends on, by increasing | |
1091 | * priority: | |
1092 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1093 | * - the task group effective clamp value, for tasks not either in the root |
1094 | * group or in an autogroup | |
e8f14172 PB |
1095 | * - the system default clamp value, defined by the sysadmin |
1096 | */ | |
1097 | static inline struct uclamp_se | |
0413d7f3 | 1098 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1099 | { |
3eac870a | 1100 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1101 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1102 | ||
1103 | /* System default restrictions always apply */ | |
1104 | if (unlikely(uc_req.value > uc_max.value)) | |
1105 | return uc_max; | |
1106 | ||
1107 | return uc_req; | |
1108 | } | |
1109 | ||
686516b5 | 1110 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1111 | { |
1112 | struct uclamp_se uc_eff; | |
1113 | ||
1114 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1115 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1116 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1117 | |
1118 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1119 | ||
686516b5 | 1120 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1121 | } |
1122 | ||
69842cba PB |
1123 | /* |
1124 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1125 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1126 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1127 | * |
1128 | * Tasks can have a task-specific value requested from user-space, track | |
1129 | * within each bucket the maximum value for tasks refcounted in it. | |
1130 | * This "local max aggregation" allows to track the exact "requested" value | |
1131 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1132 | */ |
1133 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1134 | enum uclamp_id clamp_id) |
69842cba PB |
1135 | { |
1136 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1137 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1138 | struct uclamp_bucket *bucket; | |
1139 | ||
1140 | lockdep_assert_held(&rq->lock); | |
1141 | ||
e8f14172 PB |
1142 | /* Update task effective clamp */ |
1143 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1144 | ||
69842cba PB |
1145 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1146 | bucket->tasks++; | |
e8f14172 | 1147 | uc_se->active = true; |
69842cba | 1148 | |
e496187d PB |
1149 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1150 | ||
60daf9c1 PB |
1151 | /* |
1152 | * Local max aggregation: rq buckets always track the max | |
1153 | * "requested" clamp value of its RUNNABLE tasks. | |
1154 | */ | |
1155 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1156 | bucket->value = uc_se->value; | |
1157 | ||
69842cba | 1158 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1159 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1160 | } |
1161 | ||
1162 | /* | |
1163 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1164 | * is released. If this is the last task reference counting the rq's max | |
1165 | * active clamp value, then the rq's clamp value is updated. | |
1166 | * | |
1167 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1168 | * always valid. If it's detected they are not, as defensive programming, | |
1169 | * enforce the expected state and warn. | |
1170 | */ | |
1171 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1172 | enum uclamp_id clamp_id) |
69842cba PB |
1173 | { |
1174 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1175 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1176 | struct uclamp_bucket *bucket; | |
e496187d | 1177 | unsigned int bkt_clamp; |
69842cba PB |
1178 | unsigned int rq_clamp; |
1179 | ||
1180 | lockdep_assert_held(&rq->lock); | |
1181 | ||
46609ce2 QY |
1182 | /* |
1183 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1184 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1185 | * | |
1186 | * In this case the uc_se->active flag should be false since no uclamp | |
1187 | * accounting was performed at enqueue time and we can just return | |
1188 | * here. | |
1189 | * | |
1190 | * Need to be careful of the following enqeueue/dequeue ordering | |
1191 | * problem too | |
1192 | * | |
1193 | * enqueue(taskA) | |
1194 | * // sched_uclamp_used gets enabled | |
1195 | * enqueue(taskB) | |
1196 | * dequeue(taskA) | |
1197 | * // Must not decrement bukcet->tasks here | |
1198 | * dequeue(taskB) | |
1199 | * | |
1200 | * where we could end up with stale data in uc_se and | |
1201 | * bucket[uc_se->bucket_id]. | |
1202 | * | |
1203 | * The following check here eliminates the possibility of such race. | |
1204 | */ | |
1205 | if (unlikely(!uc_se->active)) | |
1206 | return; | |
1207 | ||
69842cba | 1208 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1209 | |
69842cba PB |
1210 | SCHED_WARN_ON(!bucket->tasks); |
1211 | if (likely(bucket->tasks)) | |
1212 | bucket->tasks--; | |
46609ce2 | 1213 | |
e8f14172 | 1214 | uc_se->active = false; |
69842cba | 1215 | |
60daf9c1 PB |
1216 | /* |
1217 | * Keep "local max aggregation" simple and accept to (possibly) | |
1218 | * overboost some RUNNABLE tasks in the same bucket. | |
1219 | * The rq clamp bucket value is reset to its base value whenever | |
1220 | * there are no more RUNNABLE tasks refcounting it. | |
1221 | */ | |
69842cba PB |
1222 | if (likely(bucket->tasks)) |
1223 | return; | |
1224 | ||
1225 | rq_clamp = READ_ONCE(uc_rq->value); | |
1226 | /* | |
1227 | * Defensive programming: this should never happen. If it happens, | |
1228 | * e.g. due to future modification, warn and fixup the expected value. | |
1229 | */ | |
1230 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1231 | if (bucket->value >= rq_clamp) { |
1232 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1233 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1234 | } | |
69842cba PB |
1235 | } |
1236 | ||
1237 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1238 | { | |
0413d7f3 | 1239 | enum uclamp_id clamp_id; |
69842cba | 1240 | |
46609ce2 QY |
1241 | /* |
1242 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1243 | * | |
1244 | * The condition is constructed such that a NOP is generated when | |
1245 | * sched_uclamp_used is disabled. | |
1246 | */ | |
1247 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1248 | return; | |
1249 | ||
69842cba PB |
1250 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1251 | return; | |
1252 | ||
1253 | for_each_clamp_id(clamp_id) | |
1254 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1255 | |
1256 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1257 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1258 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1259 | } |
1260 | ||
1261 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1262 | { | |
0413d7f3 | 1263 | enum uclamp_id clamp_id; |
69842cba | 1264 | |
46609ce2 QY |
1265 | /* |
1266 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1267 | * | |
1268 | * The condition is constructed such that a NOP is generated when | |
1269 | * sched_uclamp_used is disabled. | |
1270 | */ | |
1271 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1272 | return; | |
1273 | ||
69842cba PB |
1274 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1275 | return; | |
1276 | ||
1277 | for_each_clamp_id(clamp_id) | |
1278 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1279 | } | |
1280 | ||
babbe170 | 1281 | static inline void |
0413d7f3 | 1282 | uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id) |
babbe170 PB |
1283 | { |
1284 | struct rq_flags rf; | |
1285 | struct rq *rq; | |
1286 | ||
1287 | /* | |
1288 | * Lock the task and the rq where the task is (or was) queued. | |
1289 | * | |
1290 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1291 | * price to pay to safely serialize util_{min,max} updates with | |
1292 | * enqueues, dequeues and migration operations. | |
1293 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1294 | */ | |
1295 | rq = task_rq_lock(p, &rf); | |
1296 | ||
1297 | /* | |
1298 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1299 | * If the task is not yet RUNNABLE and its task_struct is not | |
1300 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1301 | * it will already see the updated clamp bucket value. | |
1302 | */ | |
6e1ff077 | 1303 | if (p->uclamp[clamp_id].active) { |
babbe170 PB |
1304 | uclamp_rq_dec_id(rq, p, clamp_id); |
1305 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1306 | } | |
1307 | ||
1308 | task_rq_unlock(rq, p, &rf); | |
1309 | } | |
1310 | ||
e3b8b6a0 | 1311 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 PB |
1312 | static inline void |
1313 | uclamp_update_active_tasks(struct cgroup_subsys_state *css, | |
1314 | unsigned int clamps) | |
1315 | { | |
0413d7f3 | 1316 | enum uclamp_id clamp_id; |
babbe170 PB |
1317 | struct css_task_iter it; |
1318 | struct task_struct *p; | |
babbe170 PB |
1319 | |
1320 | css_task_iter_start(css, 0, &it); | |
1321 | while ((p = css_task_iter_next(&it))) { | |
1322 | for_each_clamp_id(clamp_id) { | |
1323 | if ((0x1 << clamp_id) & clamps) | |
1324 | uclamp_update_active(p, clamp_id); | |
1325 | } | |
1326 | } | |
1327 | css_task_iter_end(&it); | |
1328 | } | |
1329 | ||
7274a5c1 PB |
1330 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1331 | static void uclamp_update_root_tg(void) | |
1332 | { | |
1333 | struct task_group *tg = &root_task_group; | |
1334 | ||
1335 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1336 | sysctl_sched_uclamp_util_min, false); | |
1337 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1338 | sysctl_sched_uclamp_util_max, false); | |
1339 | ||
1340 | rcu_read_lock(); | |
1341 | cpu_util_update_eff(&root_task_group.css); | |
1342 | rcu_read_unlock(); | |
1343 | } | |
1344 | #else | |
1345 | static void uclamp_update_root_tg(void) { } | |
1346 | #endif | |
1347 | ||
e8f14172 | 1348 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1349 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1350 | { |
7274a5c1 | 1351 | bool update_root_tg = false; |
13685c4a | 1352 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1353 | int result; |
1354 | ||
2480c093 | 1355 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1356 | old_min = sysctl_sched_uclamp_util_min; |
1357 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1358 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1359 | |
1360 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1361 | if (result) | |
1362 | goto undo; | |
1363 | if (!write) | |
1364 | goto done; | |
1365 | ||
1366 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1367 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1368 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1369 | ||
e8f14172 PB |
1370 | result = -EINVAL; |
1371 | goto undo; | |
1372 | } | |
1373 | ||
1374 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1375 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1376 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1377 | update_root_tg = true; |
e8f14172 PB |
1378 | } |
1379 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1380 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1381 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1382 | update_root_tg = true; |
e8f14172 PB |
1383 | } |
1384 | ||
46609ce2 QY |
1385 | if (update_root_tg) { |
1386 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1387 | uclamp_update_root_tg(); |
46609ce2 | 1388 | } |
7274a5c1 | 1389 | |
13685c4a QY |
1390 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1391 | static_branch_enable(&sched_uclamp_used); | |
1392 | uclamp_sync_util_min_rt_default(); | |
1393 | } | |
7274a5c1 | 1394 | |
e8f14172 | 1395 | /* |
7274a5c1 PB |
1396 | * We update all RUNNABLE tasks only when task groups are in use. |
1397 | * Otherwise, keep it simple and do just a lazy update at each next | |
1398 | * task enqueue time. | |
e8f14172 | 1399 | */ |
7274a5c1 | 1400 | |
e8f14172 PB |
1401 | goto done; |
1402 | ||
1403 | undo: | |
1404 | sysctl_sched_uclamp_util_min = old_min; | |
1405 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1406 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1407 | done: |
2480c093 | 1408 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1409 | |
1410 | return result; | |
1411 | } | |
1412 | ||
a509a7cd PB |
1413 | static int uclamp_validate(struct task_struct *p, |
1414 | const struct sched_attr *attr) | |
1415 | { | |
1416 | unsigned int lower_bound = p->uclamp_req[UCLAMP_MIN].value; | |
1417 | unsigned int upper_bound = p->uclamp_req[UCLAMP_MAX].value; | |
1418 | ||
1419 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) | |
1420 | lower_bound = attr->sched_util_min; | |
1421 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) | |
1422 | upper_bound = attr->sched_util_max; | |
1423 | ||
1424 | if (lower_bound > upper_bound) | |
1425 | return -EINVAL; | |
1426 | if (upper_bound > SCHED_CAPACITY_SCALE) | |
1427 | return -EINVAL; | |
1428 | ||
e65855a5 QY |
1429 | /* |
1430 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1431 | * | |
1432 | * We need to do that here, because enabling static branches is a | |
1433 | * blocking operation which obviously cannot be done while holding | |
1434 | * scheduler locks. | |
1435 | */ | |
1436 | static_branch_enable(&sched_uclamp_used); | |
1437 | ||
a509a7cd PB |
1438 | return 0; |
1439 | } | |
1440 | ||
1441 | static void __setscheduler_uclamp(struct task_struct *p, | |
1442 | const struct sched_attr *attr) | |
1443 | { | |
0413d7f3 | 1444 | enum uclamp_id clamp_id; |
1a00d999 PB |
1445 | |
1446 | /* | |
1447 | * On scheduling class change, reset to default clamps for tasks | |
1448 | * without a task-specific value. | |
1449 | */ | |
1450 | for_each_clamp_id(clamp_id) { | |
1451 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
1a00d999 PB |
1452 | |
1453 | /* Keep using defined clamps across class changes */ | |
1454 | if (uc_se->user_defined) | |
1455 | continue; | |
1456 | ||
13685c4a QY |
1457 | /* |
1458 | * RT by default have a 100% boost value that could be modified | |
1459 | * at runtime. | |
1460 | */ | |
1a00d999 | 1461 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
13685c4a QY |
1462 | __uclamp_update_util_min_rt_default(p); |
1463 | else | |
1464 | uclamp_se_set(uc_se, uclamp_none(clamp_id), false); | |
1a00d999 | 1465 | |
1a00d999 PB |
1466 | } |
1467 | ||
a509a7cd PB |
1468 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1469 | return; | |
1470 | ||
1471 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { | |
1472 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], | |
1473 | attr->sched_util_min, true); | |
1474 | } | |
1475 | ||
1476 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1477 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], | |
1478 | attr->sched_util_max, true); | |
1479 | } | |
1480 | } | |
1481 | ||
e8f14172 PB |
1482 | static void uclamp_fork(struct task_struct *p) |
1483 | { | |
0413d7f3 | 1484 | enum uclamp_id clamp_id; |
e8f14172 | 1485 | |
13685c4a QY |
1486 | /* |
1487 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1488 | * as the task is still at its early fork stages. | |
1489 | */ | |
e8f14172 PB |
1490 | for_each_clamp_id(clamp_id) |
1491 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1492 | |
1493 | if (likely(!p->sched_reset_on_fork)) | |
1494 | return; | |
1495 | ||
1496 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1497 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1498 | uclamp_none(clamp_id), false); | |
a87498ac | 1499 | } |
e8f14172 PB |
1500 | } |
1501 | ||
13685c4a QY |
1502 | static void uclamp_post_fork(struct task_struct *p) |
1503 | { | |
1504 | uclamp_update_util_min_rt_default(p); | |
1505 | } | |
1506 | ||
d81ae8aa QY |
1507 | static void __init init_uclamp_rq(struct rq *rq) |
1508 | { | |
1509 | enum uclamp_id clamp_id; | |
1510 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1511 | ||
1512 | for_each_clamp_id(clamp_id) { | |
1513 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1514 | .value = uclamp_none(clamp_id) | |
1515 | }; | |
1516 | } | |
1517 | ||
1518 | rq->uclamp_flags = 0; | |
1519 | } | |
1520 | ||
69842cba PB |
1521 | static void __init init_uclamp(void) |
1522 | { | |
e8f14172 | 1523 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1524 | enum uclamp_id clamp_id; |
69842cba PB |
1525 | int cpu; |
1526 | ||
d81ae8aa QY |
1527 | for_each_possible_cpu(cpu) |
1528 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1529 | |
69842cba | 1530 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1531 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1532 | uclamp_none(clamp_id), false); |
69842cba | 1533 | } |
e8f14172 PB |
1534 | |
1535 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1536 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1537 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1538 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1539 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1540 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1541 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1542 | #endif |
1543 | } | |
69842cba PB |
1544 | } |
1545 | ||
1546 | #else /* CONFIG_UCLAMP_TASK */ | |
1547 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1548 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1549 | static inline int uclamp_validate(struct task_struct *p, |
1550 | const struct sched_attr *attr) | |
1551 | { | |
1552 | return -EOPNOTSUPP; | |
1553 | } | |
1554 | static void __setscheduler_uclamp(struct task_struct *p, | |
1555 | const struct sched_attr *attr) { } | |
e8f14172 | 1556 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1557 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1558 | static inline void init_uclamp(void) { } |
1559 | #endif /* CONFIG_UCLAMP_TASK */ | |
1560 | ||
1de64443 | 1561 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1562 | { |
0a67d1ee PZ |
1563 | if (!(flags & ENQUEUE_NOCLOCK)) |
1564 | update_rq_clock(rq); | |
1565 | ||
eb414681 | 1566 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 1567 | sched_info_queued(rq, p); |
eb414681 JW |
1568 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1569 | } | |
0a67d1ee | 1570 | |
69842cba | 1571 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1572 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1573 | } |
1574 | ||
1de64443 | 1575 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1576 | { |
0a67d1ee PZ |
1577 | if (!(flags & DEQUEUE_NOCLOCK)) |
1578 | update_rq_clock(rq); | |
1579 | ||
eb414681 | 1580 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 1581 | sched_info_dequeued(rq, p); |
eb414681 JW |
1582 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1583 | } | |
0a67d1ee | 1584 | |
69842cba | 1585 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1586 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1587 | } |
1588 | ||
029632fb | 1589 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1590 | { |
371fd7e7 | 1591 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
1592 | |
1593 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
1594 | } |
1595 | ||
029632fb | 1596 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1597 | { |
7dd77884 PZ |
1598 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
1599 | ||
371fd7e7 | 1600 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1601 | } |
1602 | ||
14531189 | 1603 | /* |
dd41f596 | 1604 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1605 | */ |
14531189 IM |
1606 | static inline int __normal_prio(struct task_struct *p) |
1607 | { | |
dd41f596 | 1608 | return p->static_prio; |
14531189 IM |
1609 | } |
1610 | ||
b29739f9 IM |
1611 | /* |
1612 | * Calculate the expected normal priority: i.e. priority | |
1613 | * without taking RT-inheritance into account. Might be | |
1614 | * boosted by interactivity modifiers. Changes upon fork, | |
1615 | * setprio syscalls, and whenever the interactivity | |
1616 | * estimator recalculates. | |
1617 | */ | |
36c8b586 | 1618 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1619 | { |
1620 | int prio; | |
1621 | ||
aab03e05 DF |
1622 | if (task_has_dl_policy(p)) |
1623 | prio = MAX_DL_PRIO-1; | |
1624 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
1625 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1626 | else | |
1627 | prio = __normal_prio(p); | |
1628 | return prio; | |
1629 | } | |
1630 | ||
1631 | /* | |
1632 | * Calculate the current priority, i.e. the priority | |
1633 | * taken into account by the scheduler. This value might | |
1634 | * be boosted by RT tasks, or might be boosted by | |
1635 | * interactivity modifiers. Will be RT if the task got | |
1636 | * RT-boosted. If not then it returns p->normal_prio. | |
1637 | */ | |
36c8b586 | 1638 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1639 | { |
1640 | p->normal_prio = normal_prio(p); | |
1641 | /* | |
1642 | * If we are RT tasks or we were boosted to RT priority, | |
1643 | * keep the priority unchanged. Otherwise, update priority | |
1644 | * to the normal priority: | |
1645 | */ | |
1646 | if (!rt_prio(p->prio)) | |
1647 | return p->normal_prio; | |
1648 | return p->prio; | |
1649 | } | |
1650 | ||
1da177e4 LT |
1651 | /** |
1652 | * task_curr - is this task currently executing on a CPU? | |
1653 | * @p: the task in question. | |
e69f6186 YB |
1654 | * |
1655 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1656 | */ |
36c8b586 | 1657 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1658 | { |
1659 | return cpu_curr(task_cpu(p)) == p; | |
1660 | } | |
1661 | ||
67dfa1b7 | 1662 | /* |
4c9a4bc8 PZ |
1663 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1664 | * use the balance_callback list if you want balancing. | |
1665 | * | |
1666 | * this means any call to check_class_changed() must be followed by a call to | |
1667 | * balance_callback(). | |
67dfa1b7 | 1668 | */ |
cb469845 SR |
1669 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1670 | const struct sched_class *prev_class, | |
da7a735e | 1671 | int oldprio) |
cb469845 SR |
1672 | { |
1673 | if (prev_class != p->sched_class) { | |
1674 | if (prev_class->switched_from) | |
da7a735e | 1675 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1676 | |
da7a735e | 1677 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1678 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1679 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1680 | } |
1681 | ||
029632fb | 1682 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 1683 | { |
aa93cd53 | 1684 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 1685 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
1686 | else if (p->sched_class > rq->curr->sched_class) |
1687 | resched_curr(rq); | |
1e5a7405 PZ |
1688 | |
1689 | /* | |
1690 | * A queue event has occurred, and we're going to schedule. In | |
1691 | * this case, we can save a useless back to back clock update. | |
1692 | */ | |
da0c1e65 | 1693 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 1694 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
1695 | } |
1696 | ||
1da177e4 | 1697 | #ifdef CONFIG_SMP |
175f0e25 | 1698 | |
175f0e25 | 1699 | /* |
bee98539 | 1700 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
1701 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
1702 | */ | |
1703 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
1704 | { | |
3bd37062 | 1705 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
1706 | return false; |
1707 | ||
1708 | if (is_per_cpu_kthread(p)) | |
1709 | return cpu_online(cpu); | |
1710 | ||
1711 | return cpu_active(cpu); | |
1712 | } | |
1713 | ||
5cc389bc PZ |
1714 | /* |
1715 | * This is how migration works: | |
1716 | * | |
1717 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1718 | * stop_one_cpu(). | |
1719 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1720 | * off the CPU) | |
1721 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1722 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1723 | * it and puts it into the right queue. | |
1724 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1725 | * is done. | |
1726 | */ | |
1727 | ||
1728 | /* | |
1729 | * move_queued_task - move a queued task to new rq. | |
1730 | * | |
1731 | * Returns (locked) new rq. Old rq's lock is released. | |
1732 | */ | |
8a8c69c3 PZ |
1733 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
1734 | struct task_struct *p, int new_cpu) | |
5cc389bc | 1735 | { |
5cc389bc PZ |
1736 | lockdep_assert_held(&rq->lock); |
1737 | ||
58877d34 | 1738 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 1739 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 1740 | rq_unlock(rq, rf); |
5cc389bc PZ |
1741 | |
1742 | rq = cpu_rq(new_cpu); | |
1743 | ||
8a8c69c3 | 1744 | rq_lock(rq, rf); |
5cc389bc | 1745 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 1746 | activate_task(rq, p, 0); |
5cc389bc PZ |
1747 | check_preempt_curr(rq, p, 0); |
1748 | ||
1749 | return rq; | |
1750 | } | |
1751 | ||
1752 | struct migration_arg { | |
1753 | struct task_struct *task; | |
1754 | int dest_cpu; | |
1755 | }; | |
1756 | ||
1757 | /* | |
d1ccc66d | 1758 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
1759 | * this because either it can't run here any more (set_cpus_allowed() |
1760 | * away from this CPU, or CPU going down), or because we're | |
1761 | * attempting to rebalance this task on exec (sched_exec). | |
1762 | * | |
1763 | * So we race with normal scheduler movements, but that's OK, as long | |
1764 | * as the task is no longer on this CPU. | |
5cc389bc | 1765 | */ |
8a8c69c3 PZ |
1766 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
1767 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 1768 | { |
5cc389bc | 1769 | /* Affinity changed (again). */ |
175f0e25 | 1770 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 1771 | return rq; |
5cc389bc | 1772 | |
15ff991e | 1773 | update_rq_clock(rq); |
8a8c69c3 | 1774 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
1775 | |
1776 | return rq; | |
5cc389bc PZ |
1777 | } |
1778 | ||
1779 | /* | |
1780 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1781 | * and performs thread migration by bumping thread off CPU then | |
1782 | * 'pushing' onto another runqueue. | |
1783 | */ | |
1784 | static int migration_cpu_stop(void *data) | |
1785 | { | |
1786 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1787 | struct task_struct *p = arg->task; |
1788 | struct rq *rq = this_rq(); | |
8a8c69c3 | 1789 | struct rq_flags rf; |
5cc389bc PZ |
1790 | |
1791 | /* | |
d1ccc66d IM |
1792 | * The original target CPU might have gone down and we might |
1793 | * be on another CPU but it doesn't matter. | |
5cc389bc PZ |
1794 | */ |
1795 | local_irq_disable(); | |
1796 | /* | |
1797 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 1798 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
1799 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
1800 | */ | |
a1488664 | 1801 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
1802 | |
1803 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1804 | rq_lock(rq, &rf); |
5e16bbc2 PZ |
1805 | /* |
1806 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1807 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1808 | * we're holding p->pi_lock. | |
1809 | */ | |
bf89a304 CC |
1810 | if (task_rq(p) == rq) { |
1811 | if (task_on_rq_queued(p)) | |
8a8c69c3 | 1812 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 CC |
1813 | else |
1814 | p->wake_cpu = arg->dest_cpu; | |
1815 | } | |
8a8c69c3 | 1816 | rq_unlock(rq, &rf); |
5e16bbc2 PZ |
1817 | raw_spin_unlock(&p->pi_lock); |
1818 | ||
5cc389bc PZ |
1819 | local_irq_enable(); |
1820 | return 0; | |
1821 | } | |
1822 | ||
c5b28038 PZ |
1823 | /* |
1824 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1825 | * actually call this function. | |
1826 | */ | |
9cfc3e18 | 1827 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 1828 | { |
3bd37062 | 1829 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
1830 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1831 | } | |
1832 | ||
9cfc3e18 PZ |
1833 | static void |
1834 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 1835 | { |
6c37067e PZ |
1836 | struct rq *rq = task_rq(p); |
1837 | bool queued, running; | |
1838 | ||
c5b28038 | 1839 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1840 | |
1841 | queued = task_on_rq_queued(p); | |
1842 | running = task_current(rq, p); | |
1843 | ||
1844 | if (queued) { | |
1845 | /* | |
1846 | * Because __kthread_bind() calls this on blocked tasks without | |
1847 | * holding rq->lock. | |
1848 | */ | |
1849 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 1850 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
1851 | } |
1852 | if (running) | |
1853 | put_prev_task(rq, p); | |
1854 | ||
9cfc3e18 | 1855 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 1856 | |
6c37067e | 1857 | if (queued) |
7134b3e9 | 1858 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 1859 | if (running) |
03b7fad1 | 1860 | set_next_task(rq, p); |
c5b28038 PZ |
1861 | } |
1862 | ||
9cfc3e18 PZ |
1863 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1864 | { | |
1865 | __do_set_cpus_allowed(p, new_mask, 0); | |
1866 | } | |
1867 | ||
5cc389bc PZ |
1868 | /* |
1869 | * Change a given task's CPU affinity. Migrate the thread to a | |
1870 | * proper CPU and schedule it away if the CPU it's executing on | |
1871 | * is removed from the allowed bitmask. | |
1872 | * | |
1873 | * NOTE: the caller must have a valid reference to the task, the | |
1874 | * task must not exit() & deallocate itself prematurely. The | |
1875 | * call is not atomic; no spinlocks may be held. | |
1876 | */ | |
25834c73 | 1877 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
9cfc3e18 PZ |
1878 | const struct cpumask *new_mask, |
1879 | u32 flags) | |
5cc389bc | 1880 | { |
e9d867a6 | 1881 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1882 | unsigned int dest_cpu; |
eb580751 PZ |
1883 | struct rq_flags rf; |
1884 | struct rq *rq; | |
5cc389bc PZ |
1885 | int ret = 0; |
1886 | ||
eb580751 | 1887 | rq = task_rq_lock(p, &rf); |
a499c3ea | 1888 | update_rq_clock(rq); |
5cc389bc | 1889 | |
e9d867a6 PZI |
1890 | if (p->flags & PF_KTHREAD) { |
1891 | /* | |
1892 | * Kernel threads are allowed on online && !active CPUs | |
1893 | */ | |
1894 | cpu_valid_mask = cpu_online_mask; | |
1895 | } | |
1896 | ||
25834c73 PZ |
1897 | /* |
1898 | * Must re-check here, to close a race against __kthread_bind(), | |
1899 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1900 | */ | |
9cfc3e18 | 1901 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
1902 | ret = -EINVAL; |
1903 | goto out; | |
1904 | } | |
1905 | ||
fd844ba9 | 1906 | if (cpumask_equal(&p->cpus_mask, new_mask)) |
5cc389bc PZ |
1907 | goto out; |
1908 | ||
46a87b38 PT |
1909 | /* |
1910 | * Picking a ~random cpu helps in cases where we are changing affinity | |
1911 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
1912 | * immediately required to distribute the tasks within their new mask. | |
1913 | */ | |
1914 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 1915 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
1916 | ret = -EINVAL; |
1917 | goto out; | |
1918 | } | |
1919 | ||
9cfc3e18 | 1920 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 1921 | |
e9d867a6 PZI |
1922 | if (p->flags & PF_KTHREAD) { |
1923 | /* | |
1924 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 1925 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
1926 | */ |
1927 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1928 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1929 | p->nr_cpus_allowed != 1); | |
1930 | } | |
1931 | ||
5cc389bc PZ |
1932 | /* Can the task run on the task's current CPU? If so, we're done */ |
1933 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1934 | goto out; | |
1935 | ||
5cc389bc PZ |
1936 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1937 | struct migration_arg arg = { p, dest_cpu }; | |
1938 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1939 | task_rq_unlock(rq, p, &rf); |
5cc389bc | 1940 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
5cc389bc | 1941 | return 0; |
cbce1a68 PZ |
1942 | } else if (task_on_rq_queued(p)) { |
1943 | /* | |
1944 | * OK, since we're going to drop the lock immediately | |
1945 | * afterwards anyway. | |
1946 | */ | |
8a8c69c3 | 1947 | rq = move_queued_task(rq, &rf, p, dest_cpu); |
cbce1a68 | 1948 | } |
5cc389bc | 1949 | out: |
eb580751 | 1950 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1951 | |
1952 | return ret; | |
1953 | } | |
25834c73 PZ |
1954 | |
1955 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1956 | { | |
9cfc3e18 | 1957 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 1958 | } |
5cc389bc PZ |
1959 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1960 | ||
dd41f596 | 1961 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1962 | { |
e2912009 PZ |
1963 | #ifdef CONFIG_SCHED_DEBUG |
1964 | /* | |
1965 | * We should never call set_task_cpu() on a blocked task, | |
1966 | * ttwu() will sort out the placement. | |
1967 | */ | |
077614ee | 1968 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1969 | !p->on_rq); |
0122ec5b | 1970 | |
3ea94de1 JP |
1971 | /* |
1972 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1973 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1974 | * time relying on p->on_rq. | |
1975 | */ | |
1976 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1977 | p->sched_class == &fair_sched_class && | |
1978 | (p->on_rq && !task_on_rq_migrating(p))); | |
1979 | ||
0122ec5b | 1980 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1981 | /* |
1982 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1983 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1984 | * | |
1985 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1986 | * see task_group(). |
6c6c54e1 PZ |
1987 | * |
1988 | * Furthermore, all task_rq users should acquire both locks, see | |
1989 | * task_rq_lock(). | |
1990 | */ | |
0122ec5b PZ |
1991 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1992 | lockdep_is_held(&task_rq(p)->lock))); | |
1993 | #endif | |
4ff9083b PZ |
1994 | /* |
1995 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
1996 | */ | |
1997 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
e2912009 PZ |
1998 | #endif |
1999 | ||
de1d7286 | 2000 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2001 | |
0c69774e | 2002 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 2003 | if (p->sched_class->migrate_task_rq) |
1327237a | 2004 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 2005 | p->se.nr_migrations++; |
d7822b1e | 2006 | rseq_migrate(p); |
ff303e66 | 2007 | perf_event_task_migrate(p); |
0c69774e | 2008 | } |
dd41f596 IM |
2009 | |
2010 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2011 | } |
2012 | ||
0ad4e3df | 2013 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
2014 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
2015 | { | |
da0c1e65 | 2016 | if (task_on_rq_queued(p)) { |
ac66f547 | 2017 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 2018 | struct rq_flags srf, drf; |
ac66f547 PZ |
2019 | |
2020 | src_rq = task_rq(p); | |
2021 | dst_rq = cpu_rq(cpu); | |
2022 | ||
8a8c69c3 PZ |
2023 | rq_pin_lock(src_rq, &srf); |
2024 | rq_pin_lock(dst_rq, &drf); | |
2025 | ||
ac66f547 PZ |
2026 | deactivate_task(src_rq, p, 0); |
2027 | set_task_cpu(p, cpu); | |
2028 | activate_task(dst_rq, p, 0); | |
2029 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
2030 | |
2031 | rq_unpin_lock(dst_rq, &drf); | |
2032 | rq_unpin_lock(src_rq, &srf); | |
2033 | ||
ac66f547 PZ |
2034 | } else { |
2035 | /* | |
2036 | * Task isn't running anymore; make it appear like we migrated | |
2037 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 2038 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
2039 | */ |
2040 | p->wake_cpu = cpu; | |
2041 | } | |
2042 | } | |
2043 | ||
2044 | struct migration_swap_arg { | |
2045 | struct task_struct *src_task, *dst_task; | |
2046 | int src_cpu, dst_cpu; | |
2047 | }; | |
2048 | ||
2049 | static int migrate_swap_stop(void *data) | |
2050 | { | |
2051 | struct migration_swap_arg *arg = data; | |
2052 | struct rq *src_rq, *dst_rq; | |
2053 | int ret = -EAGAIN; | |
2054 | ||
62694cd5 PZ |
2055 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
2056 | return -EAGAIN; | |
2057 | ||
ac66f547 PZ |
2058 | src_rq = cpu_rq(arg->src_cpu); |
2059 | dst_rq = cpu_rq(arg->dst_cpu); | |
2060 | ||
74602315 PZ |
2061 | double_raw_lock(&arg->src_task->pi_lock, |
2062 | &arg->dst_task->pi_lock); | |
ac66f547 | 2063 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 2064 | |
ac66f547 PZ |
2065 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
2066 | goto unlock; | |
2067 | ||
2068 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
2069 | goto unlock; | |
2070 | ||
3bd37062 | 2071 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
2072 | goto unlock; |
2073 | ||
3bd37062 | 2074 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
2075 | goto unlock; |
2076 | ||
2077 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
2078 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
2079 | ||
2080 | ret = 0; | |
2081 | ||
2082 | unlock: | |
2083 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
2084 | raw_spin_unlock(&arg->dst_task->pi_lock); |
2085 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
2086 | |
2087 | return ret; | |
2088 | } | |
2089 | ||
2090 | /* | |
2091 | * Cross migrate two tasks | |
2092 | */ | |
0ad4e3df SD |
2093 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
2094 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
2095 | { |
2096 | struct migration_swap_arg arg; | |
2097 | int ret = -EINVAL; | |
2098 | ||
ac66f547 PZ |
2099 | arg = (struct migration_swap_arg){ |
2100 | .src_task = cur, | |
0ad4e3df | 2101 | .src_cpu = curr_cpu, |
ac66f547 | 2102 | .dst_task = p, |
0ad4e3df | 2103 | .dst_cpu = target_cpu, |
ac66f547 PZ |
2104 | }; |
2105 | ||
2106 | if (arg.src_cpu == arg.dst_cpu) | |
2107 | goto out; | |
2108 | ||
6acce3ef PZ |
2109 | /* |
2110 | * These three tests are all lockless; this is OK since all of them | |
2111 | * will be re-checked with proper locks held further down the line. | |
2112 | */ | |
ac66f547 PZ |
2113 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
2114 | goto out; | |
2115 | ||
3bd37062 | 2116 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
2117 | goto out; |
2118 | ||
3bd37062 | 2119 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
2120 | goto out; |
2121 | ||
286549dc | 2122 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
2123 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
2124 | ||
2125 | out: | |
ac66f547 PZ |
2126 | return ret; |
2127 | } | |
0ad4e3df | 2128 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 2129 | |
1da177e4 LT |
2130 | /* |
2131 | * wait_task_inactive - wait for a thread to unschedule. | |
2132 | * | |
85ba2d86 RM |
2133 | * If @match_state is nonzero, it's the @p->state value just checked and |
2134 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2135 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2136 | * we return a positive number (its total switch count). If a second call | |
2137 | * a short while later returns the same number, the caller can be sure that | |
2138 | * @p has remained unscheduled the whole time. | |
2139 | * | |
1da177e4 LT |
2140 | * The caller must ensure that the task *will* unschedule sometime soon, |
2141 | * else this function might spin for a *long* time. This function can't | |
2142 | * be called with interrupts off, or it may introduce deadlock with | |
2143 | * smp_call_function() if an IPI is sent by the same process we are | |
2144 | * waiting to become inactive. | |
2145 | */ | |
85ba2d86 | 2146 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 2147 | { |
da0c1e65 | 2148 | int running, queued; |
eb580751 | 2149 | struct rq_flags rf; |
85ba2d86 | 2150 | unsigned long ncsw; |
70b97a7f | 2151 | struct rq *rq; |
1da177e4 | 2152 | |
3a5c359a AK |
2153 | for (;;) { |
2154 | /* | |
2155 | * We do the initial early heuristics without holding | |
2156 | * any task-queue locks at all. We'll only try to get | |
2157 | * the runqueue lock when things look like they will | |
2158 | * work out! | |
2159 | */ | |
2160 | rq = task_rq(p); | |
fa490cfd | 2161 | |
3a5c359a AK |
2162 | /* |
2163 | * If the task is actively running on another CPU | |
2164 | * still, just relax and busy-wait without holding | |
2165 | * any locks. | |
2166 | * | |
2167 | * NOTE! Since we don't hold any locks, it's not | |
2168 | * even sure that "rq" stays as the right runqueue! | |
2169 | * But we don't care, since "task_running()" will | |
2170 | * return false if the runqueue has changed and p | |
2171 | * is actually now running somewhere else! | |
2172 | */ | |
85ba2d86 RM |
2173 | while (task_running(rq, p)) { |
2174 | if (match_state && unlikely(p->state != match_state)) | |
2175 | return 0; | |
3a5c359a | 2176 | cpu_relax(); |
85ba2d86 | 2177 | } |
fa490cfd | 2178 | |
3a5c359a AK |
2179 | /* |
2180 | * Ok, time to look more closely! We need the rq | |
2181 | * lock now, to be *sure*. If we're wrong, we'll | |
2182 | * just go back and repeat. | |
2183 | */ | |
eb580751 | 2184 | rq = task_rq_lock(p, &rf); |
27a9da65 | 2185 | trace_sched_wait_task(p); |
3a5c359a | 2186 | running = task_running(rq, p); |
da0c1e65 | 2187 | queued = task_on_rq_queued(p); |
85ba2d86 | 2188 | ncsw = 0; |
f31e11d8 | 2189 | if (!match_state || p->state == match_state) |
93dcf55f | 2190 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 2191 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 2192 | |
85ba2d86 RM |
2193 | /* |
2194 | * If it changed from the expected state, bail out now. | |
2195 | */ | |
2196 | if (unlikely(!ncsw)) | |
2197 | break; | |
2198 | ||
3a5c359a AK |
2199 | /* |
2200 | * Was it really running after all now that we | |
2201 | * checked with the proper locks actually held? | |
2202 | * | |
2203 | * Oops. Go back and try again.. | |
2204 | */ | |
2205 | if (unlikely(running)) { | |
2206 | cpu_relax(); | |
2207 | continue; | |
2208 | } | |
fa490cfd | 2209 | |
3a5c359a AK |
2210 | /* |
2211 | * It's not enough that it's not actively running, | |
2212 | * it must be off the runqueue _entirely_, and not | |
2213 | * preempted! | |
2214 | * | |
80dd99b3 | 2215 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2216 | * running right now), it's preempted, and we should |
2217 | * yield - it could be a while. | |
2218 | */ | |
da0c1e65 | 2219 | if (unlikely(queued)) { |
8b0e1953 | 2220 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
2221 | |
2222 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2223 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2224 | continue; |
2225 | } | |
fa490cfd | 2226 | |
3a5c359a AK |
2227 | /* |
2228 | * Ahh, all good. It wasn't running, and it wasn't | |
2229 | * runnable, which means that it will never become | |
2230 | * running in the future either. We're all done! | |
2231 | */ | |
2232 | break; | |
2233 | } | |
85ba2d86 RM |
2234 | |
2235 | return ncsw; | |
1da177e4 LT |
2236 | } |
2237 | ||
2238 | /*** | |
2239 | * kick_process - kick a running thread to enter/exit the kernel | |
2240 | * @p: the to-be-kicked thread | |
2241 | * | |
2242 | * Cause a process which is running on another CPU to enter | |
2243 | * kernel-mode, without any delay. (to get signals handled.) | |
2244 | * | |
25985edc | 2245 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2246 | * because all it wants to ensure is that the remote task enters |
2247 | * the kernel. If the IPI races and the task has been migrated | |
2248 | * to another CPU then no harm is done and the purpose has been | |
2249 | * achieved as well. | |
2250 | */ | |
36c8b586 | 2251 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2252 | { |
2253 | int cpu; | |
2254 | ||
2255 | preempt_disable(); | |
2256 | cpu = task_cpu(p); | |
2257 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2258 | smp_send_reschedule(cpu); | |
2259 | preempt_enable(); | |
2260 | } | |
b43e3521 | 2261 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 2262 | |
30da688e | 2263 | /* |
3bd37062 | 2264 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
2265 | * |
2266 | * A few notes on cpu_active vs cpu_online: | |
2267 | * | |
2268 | * - cpu_active must be a subset of cpu_online | |
2269 | * | |
97fb7a0a | 2270 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 2271 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 2272 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
2273 | * see it. |
2274 | * | |
d1ccc66d | 2275 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 2276 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 2277 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
2278 | * off. |
2279 | * | |
2280 | * This means that fallback selection must not select !active CPUs. | |
2281 | * And can assume that any active CPU must be online. Conversely | |
2282 | * select_task_rq() below may allow selection of !active CPUs in order | |
2283 | * to satisfy the above rules. | |
30da688e | 2284 | */ |
5da9a0fb PZ |
2285 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2286 | { | |
aa00d89c TC |
2287 | int nid = cpu_to_node(cpu); |
2288 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
2289 | enum { cpuset, possible, fail } state = cpuset; |
2290 | int dest_cpu; | |
5da9a0fb | 2291 | |
aa00d89c | 2292 | /* |
d1ccc66d IM |
2293 | * If the node that the CPU is on has been offlined, cpu_to_node() |
2294 | * will return -1. There is no CPU on the node, and we should | |
2295 | * select the CPU on the other node. | |
aa00d89c TC |
2296 | */ |
2297 | if (nid != -1) { | |
2298 | nodemask = cpumask_of_node(nid); | |
2299 | ||
2300 | /* Look for allowed, online CPU in same node. */ | |
2301 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
2302 | if (!cpu_active(dest_cpu)) |
2303 | continue; | |
3bd37062 | 2304 | if (cpumask_test_cpu(dest_cpu, p->cpus_ptr)) |
aa00d89c TC |
2305 | return dest_cpu; |
2306 | } | |
2baab4e9 | 2307 | } |
5da9a0fb | 2308 | |
2baab4e9 PZ |
2309 | for (;;) { |
2310 | /* Any allowed, online CPU? */ | |
3bd37062 | 2311 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 2312 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 2313 | continue; |
175f0e25 | 2314 | |
2baab4e9 PZ |
2315 | goto out; |
2316 | } | |
5da9a0fb | 2317 | |
e73e85f0 | 2318 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
2319 | switch (state) { |
2320 | case cpuset: | |
e73e85f0 ON |
2321 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
2322 | cpuset_cpus_allowed_fallback(p); | |
2323 | state = possible; | |
2324 | break; | |
2325 | } | |
df561f66 | 2326 | fallthrough; |
2baab4e9 PZ |
2327 | case possible: |
2328 | do_set_cpus_allowed(p, cpu_possible_mask); | |
2329 | state = fail; | |
2330 | break; | |
2331 | ||
2332 | case fail: | |
2333 | BUG(); | |
2334 | break; | |
2335 | } | |
2336 | } | |
2337 | ||
2338 | out: | |
2339 | if (state != cpuset) { | |
2340 | /* | |
2341 | * Don't tell them about moving exiting tasks or | |
2342 | * kernel threads (both mm NULL), since they never | |
2343 | * leave kernel. | |
2344 | */ | |
2345 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 2346 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
2347 | task_pid_nr(p), p->comm, cpu); |
2348 | } | |
5da9a0fb PZ |
2349 | } |
2350 | ||
2351 | return dest_cpu; | |
2352 | } | |
2353 | ||
e2912009 | 2354 | /* |
3bd37062 | 2355 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 2356 | */ |
970b13ba | 2357 | static inline |
ac66f547 | 2358 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 2359 | { |
cbce1a68 PZ |
2360 | lockdep_assert_held(&p->pi_lock); |
2361 | ||
4b53a341 | 2362 | if (p->nr_cpus_allowed > 1) |
6c1d9410 | 2363 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 | 2364 | else |
3bd37062 | 2365 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
2366 | |
2367 | /* | |
2368 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 2369 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 2370 | * CPU. |
e2912009 PZ |
2371 | * |
2372 | * Since this is common to all placement strategies, this lives here. | |
2373 | * | |
2374 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2375 | * not worry about this generic constraint ] | |
2376 | */ | |
7af443ee | 2377 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 2378 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2379 | |
2380 | return cpu; | |
970b13ba | 2381 | } |
09a40af5 | 2382 | |
f5832c19 NP |
2383 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2384 | { | |
2385 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2386 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2387 | ||
2388 | if (stop) { | |
2389 | /* | |
2390 | * Make it appear like a SCHED_FIFO task, its something | |
2391 | * userspace knows about and won't get confused about. | |
2392 | * | |
2393 | * Also, it will make PI more or less work without too | |
2394 | * much confusion -- but then, stop work should not | |
2395 | * rely on PI working anyway. | |
2396 | */ | |
2397 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2398 | ||
2399 | stop->sched_class = &stop_sched_class; | |
2400 | } | |
2401 | ||
2402 | cpu_rq(cpu)->stop = stop; | |
2403 | ||
2404 | if (old_stop) { | |
2405 | /* | |
2406 | * Reset it back to a normal scheduling class so that | |
2407 | * it can die in pieces. | |
2408 | */ | |
2409 | old_stop->sched_class = &rt_sched_class; | |
2410 | } | |
2411 | } | |
2412 | ||
25834c73 PZ |
2413 | #else |
2414 | ||
2415 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
2416 | const struct cpumask *new_mask, |
2417 | u32 flags) | |
25834c73 PZ |
2418 | { |
2419 | return set_cpus_allowed_ptr(p, new_mask); | |
2420 | } | |
2421 | ||
5cc389bc | 2422 | #endif /* CONFIG_SMP */ |
970b13ba | 2423 | |
d7c01d27 | 2424 | static void |
b84cb5df | 2425 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2426 | { |
4fa8d299 | 2427 | struct rq *rq; |
b84cb5df | 2428 | |
4fa8d299 JP |
2429 | if (!schedstat_enabled()) |
2430 | return; | |
2431 | ||
2432 | rq = this_rq(); | |
d7c01d27 | 2433 | |
4fa8d299 JP |
2434 | #ifdef CONFIG_SMP |
2435 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
2436 | __schedstat_inc(rq->ttwu_local); |
2437 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
2438 | } else { |
2439 | struct sched_domain *sd; | |
2440 | ||
b85c8b71 | 2441 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 2442 | rcu_read_lock(); |
4fa8d299 | 2443 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 2444 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 2445 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
2446 | break; |
2447 | } | |
2448 | } | |
057f3fad | 2449 | rcu_read_unlock(); |
d7c01d27 | 2450 | } |
f339b9dc PZ |
2451 | |
2452 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 2453 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
2454 | #endif /* CONFIG_SMP */ |
2455 | ||
b85c8b71 PZ |
2456 | __schedstat_inc(rq->ttwu_count); |
2457 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
2458 | |
2459 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 2460 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2461 | } |
2462 | ||
23f41eeb PZ |
2463 | /* |
2464 | * Mark the task runnable and perform wakeup-preemption. | |
2465 | */ | |
e7904a28 | 2466 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2467 | struct rq_flags *rf) |
9ed3811a | 2468 | { |
9ed3811a | 2469 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 2470 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
2471 | trace_sched_wakeup(p); |
2472 | ||
9ed3811a | 2473 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
2474 | if (p->sched_class->task_woken) { |
2475 | /* | |
cbce1a68 PZ |
2476 | * Our task @p is fully woken up and running; so its safe to |
2477 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 2478 | */ |
d8ac8971 | 2479 | rq_unpin_lock(rq, rf); |
9ed3811a | 2480 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2481 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 2482 | } |
9ed3811a | 2483 | |
e69c6341 | 2484 | if (rq->idle_stamp) { |
78becc27 | 2485 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 2486 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 2487 | |
abfafa54 JL |
2488 | update_avg(&rq->avg_idle, delta); |
2489 | ||
2490 | if (rq->avg_idle > max) | |
9ed3811a | 2491 | rq->avg_idle = max; |
abfafa54 | 2492 | |
9ed3811a TH |
2493 | rq->idle_stamp = 0; |
2494 | } | |
2495 | #endif | |
2496 | } | |
2497 | ||
c05fbafb | 2498 | static void |
e7904a28 | 2499 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2500 | struct rq_flags *rf) |
c05fbafb | 2501 | { |
77558e4d | 2502 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 2503 | |
cbce1a68 PZ |
2504 | lockdep_assert_held(&rq->lock); |
2505 | ||
c05fbafb PZ |
2506 | if (p->sched_contributes_to_load) |
2507 | rq->nr_uninterruptible--; | |
b5179ac7 | 2508 | |
dbfb089d | 2509 | #ifdef CONFIG_SMP |
b5179ac7 | 2510 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 2511 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
2512 | #endif |
2513 | ||
1b174a2c | 2514 | activate_task(rq, p, en_flags); |
d8ac8971 | 2515 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
2516 | } |
2517 | ||
2518 | /* | |
58877d34 PZ |
2519 | * Consider @p being inside a wait loop: |
2520 | * | |
2521 | * for (;;) { | |
2522 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
2523 | * | |
2524 | * if (CONDITION) | |
2525 | * break; | |
2526 | * | |
2527 | * schedule(); | |
2528 | * } | |
2529 | * __set_current_state(TASK_RUNNING); | |
2530 | * | |
2531 | * between set_current_state() and schedule(). In this case @p is still | |
2532 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
2533 | * an atomic manner. | |
2534 | * | |
2535 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
2536 | * then schedule() must still happen and p->state can be changed to | |
2537 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
2538 | * need to do a full wakeup with enqueue. | |
2539 | * | |
2540 | * Returns: %true when the wakeup is done, | |
2541 | * %false otherwise. | |
c05fbafb | 2542 | */ |
58877d34 | 2543 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 2544 | { |
eb580751 | 2545 | struct rq_flags rf; |
c05fbafb PZ |
2546 | struct rq *rq; |
2547 | int ret = 0; | |
2548 | ||
eb580751 | 2549 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 2550 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
2551 | /* check_preempt_curr() may use rq clock */ |
2552 | update_rq_clock(rq); | |
d8ac8971 | 2553 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
2554 | ret = 1; |
2555 | } | |
eb580751 | 2556 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
2557 | |
2558 | return ret; | |
2559 | } | |
2560 | ||
317f3941 | 2561 | #ifdef CONFIG_SMP |
a1488664 | 2562 | void sched_ttwu_pending(void *arg) |
317f3941 | 2563 | { |
a1488664 | 2564 | struct llist_node *llist = arg; |
317f3941 | 2565 | struct rq *rq = this_rq(); |
73215849 | 2566 | struct task_struct *p, *t; |
d8ac8971 | 2567 | struct rq_flags rf; |
317f3941 | 2568 | |
e3baac47 PZ |
2569 | if (!llist) |
2570 | return; | |
2571 | ||
126c2092 PZ |
2572 | /* |
2573 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
2574 | * Races such that false-negatives are possible, since they | |
2575 | * are shorter lived that false-positives would be. | |
2576 | */ | |
2577 | WRITE_ONCE(rq->ttwu_pending, 0); | |
2578 | ||
8a8c69c3 | 2579 | rq_lock_irqsave(rq, &rf); |
77558e4d | 2580 | update_rq_clock(rq); |
317f3941 | 2581 | |
8c4890d1 | 2582 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
2583 | if (WARN_ON_ONCE(p->on_cpu)) |
2584 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
2585 | ||
2586 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
2587 | set_task_cpu(p, cpu_of(rq)); | |
2588 | ||
73215849 | 2589 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 2590 | } |
317f3941 | 2591 | |
8a8c69c3 | 2592 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
2593 | } |
2594 | ||
b2a02fc4 | 2595 | void send_call_function_single_ipi(int cpu) |
317f3941 | 2596 | { |
b2a02fc4 | 2597 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 2598 | |
b2a02fc4 PZ |
2599 | if (!set_nr_if_polling(rq->idle)) |
2600 | arch_send_call_function_single_ipi(cpu); | |
2601 | else | |
2602 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
2603 | } |
2604 | ||
2ebb1771 MG |
2605 | /* |
2606 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
2607 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
2608 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
2609 | * of the wakeup instead of the waker. | |
2610 | */ | |
2611 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 2612 | { |
e3baac47 PZ |
2613 | struct rq *rq = cpu_rq(cpu); |
2614 | ||
b7e7ade3 PZ |
2615 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
2616 | ||
126c2092 | 2617 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 2618 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 2619 | } |
d6aa8f85 | 2620 | |
f6be8af1 CL |
2621 | void wake_up_if_idle(int cpu) |
2622 | { | |
2623 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 2624 | struct rq_flags rf; |
f6be8af1 | 2625 | |
fd7de1e8 AL |
2626 | rcu_read_lock(); |
2627 | ||
2628 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
2629 | goto out; | |
f6be8af1 CL |
2630 | |
2631 | if (set_nr_if_polling(rq->idle)) { | |
2632 | trace_sched_wake_idle_without_ipi(cpu); | |
2633 | } else { | |
8a8c69c3 | 2634 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
2635 | if (is_idle_task(rq->curr)) |
2636 | smp_send_reschedule(cpu); | |
d1ccc66d | 2637 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 2638 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 2639 | } |
fd7de1e8 AL |
2640 | |
2641 | out: | |
2642 | rcu_read_unlock(); | |
f6be8af1 CL |
2643 | } |
2644 | ||
39be3501 | 2645 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
2646 | { |
2647 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
2648 | } | |
c6e7bd7a | 2649 | |
2ebb1771 MG |
2650 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
2651 | { | |
2652 | /* | |
2653 | * If the CPU does not share cache, then queue the task on the | |
2654 | * remote rqs wakelist to avoid accessing remote data. | |
2655 | */ | |
2656 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
2657 | return true; | |
2658 | ||
2659 | /* | |
2660 | * If the task is descheduling and the only running task on the | |
2661 | * CPU then use the wakelist to offload the task activation to | |
2662 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
2663 | * nr_running is checked to avoid unnecessary task stacking. | |
2664 | */ | |
739f70b4 | 2665 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
2666 | return true; |
2667 | ||
2668 | return false; | |
2669 | } | |
2670 | ||
2671 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 2672 | { |
2ebb1771 | 2673 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
2674 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
2675 | return false; | |
2676 | ||
c6e7bd7a | 2677 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 2678 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
2679 | return true; |
2680 | } | |
2681 | ||
2682 | return false; | |
2683 | } | |
58877d34 PZ |
2684 | |
2685 | #else /* !CONFIG_SMP */ | |
2686 | ||
2687 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
2688 | { | |
2689 | return false; | |
2690 | } | |
2691 | ||
d6aa8f85 | 2692 | #endif /* CONFIG_SMP */ |
317f3941 | 2693 | |
b5179ac7 | 2694 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
2695 | { |
2696 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 2697 | struct rq_flags rf; |
c05fbafb | 2698 | |
2ebb1771 | 2699 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 2700 | return; |
317f3941 | 2701 | |
8a8c69c3 | 2702 | rq_lock(rq, &rf); |
77558e4d | 2703 | update_rq_clock(rq); |
d8ac8971 | 2704 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 2705 | rq_unlock(rq, &rf); |
9ed3811a TH |
2706 | } |
2707 | ||
8643cda5 PZ |
2708 | /* |
2709 | * Notes on Program-Order guarantees on SMP systems. | |
2710 | * | |
2711 | * MIGRATION | |
2712 | * | |
2713 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
2714 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
2715 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
2716 | * |
2717 | * For migration (of runnable tasks) this is provided by the following means: | |
2718 | * | |
2719 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
2720 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
2721 | * rq(c1)->lock (if not at the same time, then in that order). | |
2722 | * C) LOCK of the rq(c1)->lock scheduling in task | |
2723 | * | |
7696f991 | 2724 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 2725 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
2726 | * |
2727 | * Example: | |
2728 | * | |
2729 | * CPU0 CPU1 CPU2 | |
2730 | * | |
2731 | * LOCK rq(0)->lock | |
2732 | * sched-out X | |
2733 | * sched-in Y | |
2734 | * UNLOCK rq(0)->lock | |
2735 | * | |
2736 | * LOCK rq(0)->lock // orders against CPU0 | |
2737 | * dequeue X | |
2738 | * UNLOCK rq(0)->lock | |
2739 | * | |
2740 | * LOCK rq(1)->lock | |
2741 | * enqueue X | |
2742 | * UNLOCK rq(1)->lock | |
2743 | * | |
2744 | * LOCK rq(1)->lock // orders against CPU2 | |
2745 | * sched-out Z | |
2746 | * sched-in X | |
2747 | * UNLOCK rq(1)->lock | |
2748 | * | |
2749 | * | |
2750 | * BLOCKING -- aka. SLEEP + WAKEUP | |
2751 | * | |
2752 | * For blocking we (obviously) need to provide the same guarantee as for | |
2753 | * migration. However the means are completely different as there is no lock | |
2754 | * chain to provide order. Instead we do: | |
2755 | * | |
58877d34 PZ |
2756 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
2757 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
2758 | * |
2759 | * Example: | |
2760 | * | |
2761 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
2762 | * | |
2763 | * LOCK rq(0)->lock LOCK X->pi_lock | |
2764 | * dequeue X | |
2765 | * sched-out X | |
2766 | * smp_store_release(X->on_cpu, 0); | |
2767 | * | |
1f03e8d2 | 2768 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
2769 | * X->state = WAKING |
2770 | * set_task_cpu(X,2) | |
2771 | * | |
2772 | * LOCK rq(2)->lock | |
2773 | * enqueue X | |
2774 | * X->state = RUNNING | |
2775 | * UNLOCK rq(2)->lock | |
2776 | * | |
2777 | * LOCK rq(2)->lock // orders against CPU1 | |
2778 | * sched-out Z | |
2779 | * sched-in X | |
2780 | * UNLOCK rq(2)->lock | |
2781 | * | |
2782 | * UNLOCK X->pi_lock | |
2783 | * UNLOCK rq(0)->lock | |
2784 | * | |
2785 | * | |
7696f991 AP |
2786 | * However, for wakeups there is a second guarantee we must provide, namely we |
2787 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
2788 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
2789 | */ |
2790 | ||
9ed3811a | 2791 | /** |
1da177e4 | 2792 | * try_to_wake_up - wake up a thread |
9ed3811a | 2793 | * @p: the thread to be awakened |
1da177e4 | 2794 | * @state: the mask of task states that can be woken |
9ed3811a | 2795 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 2796 | * |
58877d34 PZ |
2797 | * Conceptually does: |
2798 | * | |
2799 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 2800 | * |
a2250238 PZ |
2801 | * If the task was not queued/runnable, also place it back on a runqueue. |
2802 | * | |
58877d34 PZ |
2803 | * This function is atomic against schedule() which would dequeue the task. |
2804 | * | |
2805 | * It issues a full memory barrier before accessing @p->state, see the comment | |
2806 | * with set_current_state(). | |
a2250238 | 2807 | * |
58877d34 | 2808 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 2809 | * |
58877d34 PZ |
2810 | * Relies on p->pi_lock stabilizing: |
2811 | * - p->sched_class | |
2812 | * - p->cpus_ptr | |
2813 | * - p->sched_task_group | |
2814 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
2815 | * | |
2816 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
2817 | * Takes rq->lock in: | |
2818 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
2819 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
2820 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
2821 | * | |
2822 | * As a consequence we race really badly with just about everything. See the | |
2823 | * many memory barriers and their comments for details. | |
7696f991 | 2824 | * |
a2250238 PZ |
2825 | * Return: %true if @p->state changes (an actual wakeup was done), |
2826 | * %false otherwise. | |
1da177e4 | 2827 | */ |
e4a52bcb PZ |
2828 | static int |
2829 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2830 | { |
1da177e4 | 2831 | unsigned long flags; |
c05fbafb | 2832 | int cpu, success = 0; |
2398f2c6 | 2833 | |
e3d85487 | 2834 | preempt_disable(); |
aacedf26 PZ |
2835 | if (p == current) { |
2836 | /* | |
2837 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
2838 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 2839 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
2840 | * without taking any locks. |
2841 | * | |
2842 | * In particular: | |
2843 | * - we rely on Program-Order guarantees for all the ordering, | |
2844 | * - we're serialized against set_special_state() by virtue of | |
2845 | * it disabling IRQs (this allows not taking ->pi_lock). | |
2846 | */ | |
2847 | if (!(p->state & state)) | |
e3d85487 | 2848 | goto out; |
aacedf26 PZ |
2849 | |
2850 | success = 1; | |
aacedf26 PZ |
2851 | trace_sched_waking(p); |
2852 | p->state = TASK_RUNNING; | |
2853 | trace_sched_wakeup(p); | |
2854 | goto out; | |
2855 | } | |
2856 | ||
e0acd0a6 ON |
2857 | /* |
2858 | * If we are going to wake up a thread waiting for CONDITION we | |
2859 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
2860 | * reordered with p->state check below. This pairs with smp_store_mb() |
2861 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 2862 | */ |
013fdb80 | 2863 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 2864 | smp_mb__after_spinlock(); |
e9c84311 | 2865 | if (!(p->state & state)) |
aacedf26 | 2866 | goto unlock; |
1da177e4 | 2867 | |
fbd705a0 PZ |
2868 | trace_sched_waking(p); |
2869 | ||
d1ccc66d IM |
2870 | /* We're going to change ->state: */ |
2871 | success = 1; | |
1da177e4 | 2872 | |
135e8c92 BS |
2873 | /* |
2874 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
2875 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
2876 | * in smp_cond_load_acquire() below. | |
2877 | * | |
3d85b270 AP |
2878 | * sched_ttwu_pending() try_to_wake_up() |
2879 | * STORE p->on_rq = 1 LOAD p->state | |
2880 | * UNLOCK rq->lock | |
2881 | * | |
2882 | * __schedule() (switch to task 'p') | |
2883 | * LOCK rq->lock smp_rmb(); | |
2884 | * smp_mb__after_spinlock(); | |
2885 | * UNLOCK rq->lock | |
135e8c92 BS |
2886 | * |
2887 | * [task p] | |
3d85b270 | 2888 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 2889 | * |
3d85b270 AP |
2890 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
2891 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
2892 | * |
2893 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
2894 | */ |
2895 | smp_rmb(); | |
58877d34 | 2896 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 2897 | goto unlock; |
1da177e4 | 2898 | |
c6e7bd7a PZ |
2899 | if (p->in_iowait) { |
2900 | delayacct_blkio_end(p); | |
2901 | atomic_dec(&task_rq(p)->nr_iowait); | |
2902 | } | |
2903 | ||
1da177e4 | 2904 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2905 | /* |
2906 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2907 | * possible to, falsely, observe p->on_cpu == 0. | |
2908 | * | |
2909 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2910 | * from the runqueue. | |
2911 | * | |
3d85b270 AP |
2912 | * __schedule() (switch to task 'p') try_to_wake_up() |
2913 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
2914 | * UNLOCK rq->lock | |
2915 | * | |
2916 | * __schedule() (put 'p' to sleep) | |
2917 | * LOCK rq->lock smp_rmb(); | |
2918 | * smp_mb__after_spinlock(); | |
2919 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 2920 | * |
3d85b270 AP |
2921 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
2922 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
2923 | * |
2924 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
2925 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
2926 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 2927 | */ |
dbfb089d PZ |
2928 | smp_acquire__after_ctrl_dep(); |
2929 | ||
2930 | /* | |
2931 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
2932 | * == 0), which means we need to do an enqueue, change p->state to | |
2933 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
2934 | * enqueue, such as ttwu_queue_wakelist(). | |
2935 | */ | |
2936 | p->state = TASK_WAKING; | |
ecf7d01c | 2937 | |
c6e7bd7a PZ |
2938 | /* |
2939 | * If the owning (remote) CPU is still in the middle of schedule() with | |
2940 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
2941 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
2942 | * let the waker make forward progress. This is safe because IRQs are | |
2943 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
2944 | * |
2945 | * Ensure we load task_cpu(p) after p->on_cpu: | |
2946 | * | |
2947 | * set_task_cpu(p, cpu); | |
2948 | * STORE p->cpu = @cpu | |
2949 | * __schedule() (switch to task 'p') | |
2950 | * LOCK rq->lock | |
2951 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
2952 | * STORE p->on_cpu = 1 LOAD p->cpu | |
2953 | * | |
2954 | * to ensure we observe the correct CPU on which the task is currently | |
2955 | * scheduling. | |
c6e7bd7a | 2956 | */ |
b6e13e85 | 2957 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 2958 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
2959 | goto unlock; |
2960 | ||
e9c84311 | 2961 | /* |
d1ccc66d | 2962 | * If the owning (remote) CPU is still in the middle of schedule() with |
c05fbafb | 2963 | * this task as prev, wait until its done referencing the task. |
b75a2253 | 2964 | * |
31cb1bc0 | 2965 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
2966 | * |
2967 | * This ensures that tasks getting woken will be fully ordered against | |
2968 | * their previous state and preserve Program Order. | |
0970d299 | 2969 | */ |
1f03e8d2 | 2970 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2971 | |
ac66f547 | 2972 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2973 | if (task_cpu(p) != cpu) { |
2974 | wake_flags |= WF_MIGRATED; | |
eb414681 | 2975 | psi_ttwu_dequeue(p); |
e4a52bcb | 2976 | set_task_cpu(p, cpu); |
f339b9dc | 2977 | } |
b6e13e85 PZ |
2978 | #else |
2979 | cpu = task_cpu(p); | |
1da177e4 | 2980 | #endif /* CONFIG_SMP */ |
1da177e4 | 2981 | |
b5179ac7 | 2982 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 2983 | unlock: |
013fdb80 | 2984 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
2985 | out: |
2986 | if (success) | |
b6e13e85 | 2987 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 2988 | preempt_enable(); |
1da177e4 LT |
2989 | |
2990 | return success; | |
2991 | } | |
2992 | ||
2beaf328 PM |
2993 | /** |
2994 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
2995 | * @p: Process for which the function is to be invoked. | |
2996 | * @func: Function to invoke. | |
2997 | * @arg: Argument to function. | |
2998 | * | |
2999 | * If the specified task can be quickly locked into a definite state | |
3000 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
3001 | * state while invoking @func(@arg). This function can use ->on_rq and | |
3002 | * task_curr() to work out what the state is, if required. Given that | |
3003 | * @func can be invoked with a runqueue lock held, it had better be quite | |
3004 | * lightweight. | |
3005 | * | |
3006 | * Returns: | |
3007 | * @false if the task slipped out from under the locks. | |
3008 | * @true if the task was locked onto a runqueue or is sleeping. | |
3009 | * However, @func can override this by returning @false. | |
3010 | */ | |
3011 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
3012 | { | |
3013 | bool ret = false; | |
3014 | struct rq_flags rf; | |
3015 | struct rq *rq; | |
3016 | ||
3017 | lockdep_assert_irqs_enabled(); | |
3018 | raw_spin_lock_irq(&p->pi_lock); | |
3019 | if (p->on_rq) { | |
3020 | rq = __task_rq_lock(p, &rf); | |
3021 | if (task_rq(p) == rq) | |
3022 | ret = func(p, arg); | |
3023 | rq_unlock(rq, &rf); | |
3024 | } else { | |
3025 | switch (p->state) { | |
3026 | case TASK_RUNNING: | |
3027 | case TASK_WAKING: | |
3028 | break; | |
3029 | default: | |
3030 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
3031 | if (!p->on_rq) | |
3032 | ret = func(p, arg); | |
3033 | } | |
3034 | } | |
3035 | raw_spin_unlock_irq(&p->pi_lock); | |
3036 | return ret; | |
3037 | } | |
3038 | ||
50fa610a DH |
3039 | /** |
3040 | * wake_up_process - Wake up a specific process | |
3041 | * @p: The process to be woken up. | |
3042 | * | |
3043 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
3044 | * processes. |
3045 | * | |
3046 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 3047 | * |
7696f991 | 3048 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 3049 | */ |
7ad5b3a5 | 3050 | int wake_up_process(struct task_struct *p) |
1da177e4 | 3051 | { |
9067ac85 | 3052 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 3053 | } |
1da177e4 LT |
3054 | EXPORT_SYMBOL(wake_up_process); |
3055 | ||
7ad5b3a5 | 3056 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
3057 | { |
3058 | return try_to_wake_up(p, state, 0); | |
3059 | } | |
3060 | ||
1da177e4 LT |
3061 | /* |
3062 | * Perform scheduler related setup for a newly forked process p. | |
3063 | * p is forked by current. | |
dd41f596 IM |
3064 | * |
3065 | * __sched_fork() is basic setup used by init_idle() too: | |
3066 | */ | |
5e1576ed | 3067 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3068 | { |
fd2f4419 PZ |
3069 | p->on_rq = 0; |
3070 | ||
3071 | p->se.on_rq = 0; | |
dd41f596 IM |
3072 | p->se.exec_start = 0; |
3073 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 3074 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 3075 | p->se.nr_migrations = 0; |
da7a735e | 3076 | p->se.vruntime = 0; |
fd2f4419 | 3077 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 3078 | |
ad936d86 BP |
3079 | #ifdef CONFIG_FAIR_GROUP_SCHED |
3080 | p->se.cfs_rq = NULL; | |
3081 | #endif | |
3082 | ||
6cfb0d5d | 3083 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 3084 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 3085 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 3086 | #endif |
476d139c | 3087 | |
aab03e05 | 3088 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 3089 | init_dl_task_timer(&p->dl); |
209a0cbd | 3090 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 3091 | __dl_clear_params(p); |
aab03e05 | 3092 | |
fa717060 | 3093 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
3094 | p->rt.timeout = 0; |
3095 | p->rt.time_slice = sched_rr_timeslice; | |
3096 | p->rt.on_rq = 0; | |
3097 | p->rt.on_list = 0; | |
476d139c | 3098 | |
e107be36 AK |
3099 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3100 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
3101 | #endif | |
cbee9f88 | 3102 | |
5e1f0f09 MG |
3103 | #ifdef CONFIG_COMPACTION |
3104 | p->capture_control = NULL; | |
3105 | #endif | |
13784475 | 3106 | init_numa_balancing(clone_flags, p); |
a1488664 | 3107 | #ifdef CONFIG_SMP |
8c4890d1 | 3108 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
a1488664 | 3109 | #endif |
dd41f596 IM |
3110 | } |
3111 | ||
2a595721 SD |
3112 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
3113 | ||
1a687c2e | 3114 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 3115 | |
1a687c2e MG |
3116 | void set_numabalancing_state(bool enabled) |
3117 | { | |
3118 | if (enabled) | |
2a595721 | 3119 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 3120 | else |
2a595721 | 3121 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 3122 | } |
54a43d54 AK |
3123 | |
3124 | #ifdef CONFIG_PROC_SYSCTL | |
3125 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 3126 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
3127 | { |
3128 | struct ctl_table t; | |
3129 | int err; | |
2a595721 | 3130 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
3131 | |
3132 | if (write && !capable(CAP_SYS_ADMIN)) | |
3133 | return -EPERM; | |
3134 | ||
3135 | t = *table; | |
3136 | t.data = &state; | |
3137 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3138 | if (err < 0) | |
3139 | return err; | |
3140 | if (write) | |
3141 | set_numabalancing_state(state); | |
3142 | return err; | |
3143 | } | |
3144 | #endif | |
3145 | #endif | |
dd41f596 | 3146 | |
4698f88c JP |
3147 | #ifdef CONFIG_SCHEDSTATS |
3148 | ||
cb251765 | 3149 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 3150 | static bool __initdata __sched_schedstats = false; |
cb251765 | 3151 | |
cb251765 MG |
3152 | static void set_schedstats(bool enabled) |
3153 | { | |
3154 | if (enabled) | |
3155 | static_branch_enable(&sched_schedstats); | |
3156 | else | |
3157 | static_branch_disable(&sched_schedstats); | |
3158 | } | |
3159 | ||
3160 | void force_schedstat_enabled(void) | |
3161 | { | |
3162 | if (!schedstat_enabled()) { | |
3163 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
3164 | static_branch_enable(&sched_schedstats); | |
3165 | } | |
3166 | } | |
3167 | ||
3168 | static int __init setup_schedstats(char *str) | |
3169 | { | |
3170 | int ret = 0; | |
3171 | if (!str) | |
3172 | goto out; | |
3173 | ||
4698f88c JP |
3174 | /* |
3175 | * This code is called before jump labels have been set up, so we can't | |
3176 | * change the static branch directly just yet. Instead set a temporary | |
3177 | * variable so init_schedstats() can do it later. | |
3178 | */ | |
cb251765 | 3179 | if (!strcmp(str, "enable")) { |
4698f88c | 3180 | __sched_schedstats = true; |
cb251765 MG |
3181 | ret = 1; |
3182 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 3183 | __sched_schedstats = false; |
cb251765 MG |
3184 | ret = 1; |
3185 | } | |
3186 | out: | |
3187 | if (!ret) | |
3188 | pr_warn("Unable to parse schedstats=\n"); | |
3189 | ||
3190 | return ret; | |
3191 | } | |
3192 | __setup("schedstats=", setup_schedstats); | |
3193 | ||
4698f88c JP |
3194 | static void __init init_schedstats(void) |
3195 | { | |
3196 | set_schedstats(__sched_schedstats); | |
3197 | } | |
3198 | ||
cb251765 | 3199 | #ifdef CONFIG_PROC_SYSCTL |
32927393 CH |
3200 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
3201 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
3202 | { |
3203 | struct ctl_table t; | |
3204 | int err; | |
3205 | int state = static_branch_likely(&sched_schedstats); | |
3206 | ||
3207 | if (write && !capable(CAP_SYS_ADMIN)) | |
3208 | return -EPERM; | |
3209 | ||
3210 | t = *table; | |
3211 | t.data = &state; | |
3212 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3213 | if (err < 0) | |
3214 | return err; | |
3215 | if (write) | |
3216 | set_schedstats(state); | |
3217 | return err; | |
3218 | } | |
4698f88c JP |
3219 | #endif /* CONFIG_PROC_SYSCTL */ |
3220 | #else /* !CONFIG_SCHEDSTATS */ | |
3221 | static inline void init_schedstats(void) {} | |
3222 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
3223 | |
3224 | /* | |
3225 | * fork()/clone()-time setup: | |
3226 | */ | |
aab03e05 | 3227 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3228 | { |
0122ec5b | 3229 | unsigned long flags; |
dd41f596 | 3230 | |
5e1576ed | 3231 | __sched_fork(clone_flags, p); |
06b83b5f | 3232 | /* |
7dc603c9 | 3233 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
3234 | * nobody will actually run it, and a signal or other external |
3235 | * event cannot wake it up and insert it on the runqueue either. | |
3236 | */ | |
7dc603c9 | 3237 | p->state = TASK_NEW; |
dd41f596 | 3238 | |
c350a04e MG |
3239 | /* |
3240 | * Make sure we do not leak PI boosting priority to the child. | |
3241 | */ | |
3242 | p->prio = current->normal_prio; | |
3243 | ||
e8f14172 PB |
3244 | uclamp_fork(p); |
3245 | ||
b9dc29e7 MG |
3246 | /* |
3247 | * Revert to default priority/policy on fork if requested. | |
3248 | */ | |
3249 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 3250 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 3251 | p->policy = SCHED_NORMAL; |
6c697bdf | 3252 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
3253 | p->rt_priority = 0; |
3254 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
3255 | p->static_prio = NICE_TO_PRIO(0); | |
3256 | ||
3257 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 3258 | set_load_weight(p, false); |
6c697bdf | 3259 | |
b9dc29e7 MG |
3260 | /* |
3261 | * We don't need the reset flag anymore after the fork. It has | |
3262 | * fulfilled its duty: | |
3263 | */ | |
3264 | p->sched_reset_on_fork = 0; | |
3265 | } | |
ca94c442 | 3266 | |
af0fffd9 | 3267 | if (dl_prio(p->prio)) |
aab03e05 | 3268 | return -EAGAIN; |
af0fffd9 | 3269 | else if (rt_prio(p->prio)) |
aab03e05 | 3270 | p->sched_class = &rt_sched_class; |
af0fffd9 | 3271 | else |
2ddbf952 | 3272 | p->sched_class = &fair_sched_class; |
b29739f9 | 3273 | |
7dc603c9 | 3274 | init_entity_runnable_average(&p->se); |
cd29fe6f | 3275 | |
86951599 PZ |
3276 | /* |
3277 | * The child is not yet in the pid-hash so no cgroup attach races, | |
3278 | * and the cgroup is pinned to this child due to cgroup_fork() | |
3279 | * is ran before sched_fork(). | |
3280 | * | |
3281 | * Silence PROVE_RCU. | |
3282 | */ | |
0122ec5b | 3283 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ce3614da | 3284 | rseq_migrate(p); |
e210bffd | 3285 | /* |
d1ccc66d | 3286 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
3287 | * so use __set_task_cpu(). |
3288 | */ | |
af0fffd9 | 3289 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
3290 | if (p->sched_class->task_fork) |
3291 | p->sched_class->task_fork(p); | |
0122ec5b | 3292 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 3293 | |
f6db8347 | 3294 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 3295 | if (likely(sched_info_on())) |
52f17b6c | 3296 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 3297 | #endif |
3ca7a440 PZ |
3298 | #if defined(CONFIG_SMP) |
3299 | p->on_cpu = 0; | |
4866cde0 | 3300 | #endif |
01028747 | 3301 | init_task_preempt_count(p); |
806c09a7 | 3302 | #ifdef CONFIG_SMP |
917b627d | 3303 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 3304 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 3305 | #endif |
aab03e05 | 3306 | return 0; |
1da177e4 LT |
3307 | } |
3308 | ||
13685c4a QY |
3309 | void sched_post_fork(struct task_struct *p) |
3310 | { | |
3311 | uclamp_post_fork(p); | |
3312 | } | |
3313 | ||
332ac17e DF |
3314 | unsigned long to_ratio(u64 period, u64 runtime) |
3315 | { | |
3316 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 3317 | return BW_UNIT; |
332ac17e DF |
3318 | |
3319 | /* | |
3320 | * Doing this here saves a lot of checks in all | |
3321 | * the calling paths, and returning zero seems | |
3322 | * safe for them anyway. | |
3323 | */ | |
3324 | if (period == 0) | |
3325 | return 0; | |
3326 | ||
c52f14d3 | 3327 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
3328 | } |
3329 | ||
1da177e4 LT |
3330 | /* |
3331 | * wake_up_new_task - wake up a newly created task for the first time. | |
3332 | * | |
3333 | * This function will do some initial scheduler statistics housekeeping | |
3334 | * that must be done for every newly created context, then puts the task | |
3335 | * on the runqueue and wakes it. | |
3336 | */ | |
3e51e3ed | 3337 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 3338 | { |
eb580751 | 3339 | struct rq_flags rf; |
dd41f596 | 3340 | struct rq *rq; |
fabf318e | 3341 | |
eb580751 | 3342 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 3343 | p->state = TASK_RUNNING; |
fabf318e PZ |
3344 | #ifdef CONFIG_SMP |
3345 | /* | |
3346 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 3347 | * - cpus_ptr can change in the fork path |
d1ccc66d | 3348 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
3349 | * |
3350 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
3351 | * as we're not fully set-up yet. | |
fabf318e | 3352 | */ |
32e839dd | 3353 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 3354 | rseq_migrate(p); |
e210bffd | 3355 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 3356 | #endif |
b7fa30c9 | 3357 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 3358 | update_rq_clock(rq); |
d0fe0b9c | 3359 | post_init_entity_util_avg(p); |
0017d735 | 3360 | |
7a57f32a | 3361 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 3362 | trace_sched_wakeup_new(p); |
a7558e01 | 3363 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3364 | #ifdef CONFIG_SMP |
0aaafaab PZ |
3365 | if (p->sched_class->task_woken) { |
3366 | /* | |
3367 | * Nothing relies on rq->lock after this, so its fine to | |
3368 | * drop it. | |
3369 | */ | |
d8ac8971 | 3370 | rq_unpin_lock(rq, &rf); |
efbbd05a | 3371 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3372 | rq_repin_lock(rq, &rf); |
0aaafaab | 3373 | } |
9a897c5a | 3374 | #endif |
eb580751 | 3375 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
3376 | } |
3377 | ||
e107be36 AK |
3378 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3379 | ||
b7203428 | 3380 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 3381 | |
2ecd9d29 PZ |
3382 | void preempt_notifier_inc(void) |
3383 | { | |
b7203428 | 3384 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
3385 | } |
3386 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
3387 | ||
3388 | void preempt_notifier_dec(void) | |
3389 | { | |
b7203428 | 3390 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
3391 | } |
3392 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
3393 | ||
e107be36 | 3394 | /** |
80dd99b3 | 3395 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3396 | * @notifier: notifier struct to register |
e107be36 AK |
3397 | */ |
3398 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3399 | { | |
b7203428 | 3400 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
3401 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
3402 | ||
e107be36 AK |
3403 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
3404 | } | |
3405 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3406 | ||
3407 | /** | |
3408 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3409 | * @notifier: notifier struct to unregister |
e107be36 | 3410 | * |
d84525a8 | 3411 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
3412 | */ |
3413 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3414 | { | |
3415 | hlist_del(¬ifier->link); | |
3416 | } | |
3417 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3418 | ||
1cde2930 | 3419 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3420 | { |
3421 | struct preempt_notifier *notifier; | |
e107be36 | 3422 | |
b67bfe0d | 3423 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3424 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
3425 | } | |
3426 | ||
1cde2930 PZ |
3427 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
3428 | { | |
b7203428 | 3429 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3430 | __fire_sched_in_preempt_notifiers(curr); |
3431 | } | |
3432 | ||
e107be36 | 3433 | static void |
1cde2930 PZ |
3434 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3435 | struct task_struct *next) | |
e107be36 AK |
3436 | { |
3437 | struct preempt_notifier *notifier; | |
e107be36 | 3438 | |
b67bfe0d | 3439 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3440 | notifier->ops->sched_out(notifier, next); |
3441 | } | |
3442 | ||
1cde2930 PZ |
3443 | static __always_inline void |
3444 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3445 | struct task_struct *next) | |
3446 | { | |
b7203428 | 3447 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3448 | __fire_sched_out_preempt_notifiers(curr, next); |
3449 | } | |
3450 | ||
6d6bc0ad | 3451 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3452 | |
1cde2930 | 3453 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3454 | { |
3455 | } | |
3456 | ||
1cde2930 | 3457 | static inline void |
e107be36 AK |
3458 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3459 | struct task_struct *next) | |
3460 | { | |
3461 | } | |
3462 | ||
6d6bc0ad | 3463 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3464 | |
31cb1bc0 | 3465 | static inline void prepare_task(struct task_struct *next) |
3466 | { | |
3467 | #ifdef CONFIG_SMP | |
3468 | /* | |
3469 | * Claim the task as running, we do this before switching to it | |
3470 | * such that any running task will have this set. | |
58877d34 PZ |
3471 | * |
3472 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 3473 | */ |
58877d34 | 3474 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 3475 | #endif |
3476 | } | |
3477 | ||
3478 | static inline void finish_task(struct task_struct *prev) | |
3479 | { | |
3480 | #ifdef CONFIG_SMP | |
3481 | /* | |
58877d34 PZ |
3482 | * This must be the very last reference to @prev from this CPU. After |
3483 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
3484 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 3485 | * finished. |
3486 | * | |
3487 | * In particular, the load of prev->state in finish_task_switch() must | |
3488 | * happen before this. | |
3489 | * | |
3490 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
3491 | */ | |
3492 | smp_store_release(&prev->on_cpu, 0); | |
3493 | #endif | |
3494 | } | |
3495 | ||
565790d2 PZ |
3496 | #ifdef CONFIG_SMP |
3497 | ||
3498 | static void do_balance_callbacks(struct rq *rq, struct callback_head *head) | |
3499 | { | |
3500 | void (*func)(struct rq *rq); | |
3501 | struct callback_head *next; | |
3502 | ||
3503 | lockdep_assert_held(&rq->lock); | |
3504 | ||
3505 | while (head) { | |
3506 | func = (void (*)(struct rq *))head->func; | |
3507 | next = head->next; | |
3508 | head->next = NULL; | |
3509 | head = next; | |
3510 | ||
3511 | func(rq); | |
3512 | } | |
3513 | } | |
3514 | ||
3515 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
3516 | { | |
3517 | struct callback_head *head = rq->balance_callback; | |
3518 | ||
3519 | lockdep_assert_held(&rq->lock); | |
2558aacf | 3520 | if (head) { |
565790d2 | 3521 | rq->balance_callback = NULL; |
2558aacf PZ |
3522 | rq->balance_flags &= ~BALANCE_WORK; |
3523 | } | |
565790d2 PZ |
3524 | |
3525 | return head; | |
3526 | } | |
3527 | ||
3528 | static void __balance_callbacks(struct rq *rq) | |
3529 | { | |
3530 | do_balance_callbacks(rq, splice_balance_callbacks(rq)); | |
3531 | } | |
3532 | ||
3533 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
3534 | { | |
3535 | unsigned long flags; | |
3536 | ||
3537 | if (unlikely(head)) { | |
3538 | raw_spin_lock_irqsave(&rq->lock, flags); | |
3539 | do_balance_callbacks(rq, head); | |
3540 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
3541 | } | |
3542 | } | |
3543 | ||
2558aacf PZ |
3544 | static void balance_push(struct rq *rq); |
3545 | ||
3546 | static inline void balance_switch(struct rq *rq) | |
3547 | { | |
3548 | if (likely(!rq->balance_flags)) | |
3549 | return; | |
3550 | ||
3551 | if (rq->balance_flags & BALANCE_PUSH) { | |
3552 | balance_push(rq); | |
3553 | return; | |
3554 | } | |
3555 | ||
3556 | __balance_callbacks(rq); | |
3557 | } | |
3558 | ||
565790d2 PZ |
3559 | #else |
3560 | ||
3561 | static inline void __balance_callbacks(struct rq *rq) | |
3562 | { | |
3563 | } | |
3564 | ||
3565 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
3566 | { | |
3567 | return NULL; | |
3568 | } | |
3569 | ||
3570 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
3571 | { | |
3572 | } | |
3573 | ||
2558aacf PZ |
3574 | static inline void balance_switch(struct rq *rq) |
3575 | { | |
3576 | } | |
3577 | ||
565790d2 PZ |
3578 | #endif |
3579 | ||
269d5992 PZ |
3580 | static inline void |
3581 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 3582 | { |
269d5992 PZ |
3583 | /* |
3584 | * Since the runqueue lock will be released by the next | |
3585 | * task (which is an invalid locking op but in the case | |
3586 | * of the scheduler it's an obvious special-case), so we | |
3587 | * do an early lockdep release here: | |
3588 | */ | |
3589 | rq_unpin_lock(rq, rf); | |
5facae4f | 3590 | spin_release(&rq->lock.dep_map, _THIS_IP_); |
31cb1bc0 | 3591 | #ifdef CONFIG_DEBUG_SPINLOCK |
3592 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 3593 | rq->lock.owner = next; |
31cb1bc0 | 3594 | #endif |
269d5992 PZ |
3595 | } |
3596 | ||
3597 | static inline void finish_lock_switch(struct rq *rq) | |
3598 | { | |
31cb1bc0 | 3599 | /* |
3600 | * If we are tracking spinlock dependencies then we have to | |
3601 | * fix up the runqueue lock - which gets 'carried over' from | |
3602 | * prev into current: | |
3603 | */ | |
3604 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
2558aacf | 3605 | balance_switch(rq); |
31cb1bc0 | 3606 | raw_spin_unlock_irq(&rq->lock); |
3607 | } | |
3608 | ||
325ea10c IM |
3609 | /* |
3610 | * NOP if the arch has not defined these: | |
3611 | */ | |
3612 | ||
3613 | #ifndef prepare_arch_switch | |
3614 | # define prepare_arch_switch(next) do { } while (0) | |
3615 | #endif | |
3616 | ||
3617 | #ifndef finish_arch_post_lock_switch | |
3618 | # define finish_arch_post_lock_switch() do { } while (0) | |
3619 | #endif | |
3620 | ||
4866cde0 NP |
3621 | /** |
3622 | * prepare_task_switch - prepare to switch tasks | |
3623 | * @rq: the runqueue preparing to switch | |
421cee29 | 3624 | * @prev: the current task that is being switched out |
4866cde0 NP |
3625 | * @next: the task we are going to switch to. |
3626 | * | |
3627 | * This is called with the rq lock held and interrupts off. It must | |
3628 | * be paired with a subsequent finish_task_switch after the context | |
3629 | * switch. | |
3630 | * | |
3631 | * prepare_task_switch sets up locking and calls architecture specific | |
3632 | * hooks. | |
3633 | */ | |
e107be36 AK |
3634 | static inline void |
3635 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
3636 | struct task_struct *next) | |
4866cde0 | 3637 | { |
0ed557aa | 3638 | kcov_prepare_switch(prev); |
43148951 | 3639 | sched_info_switch(rq, prev, next); |
fe4b04fa | 3640 | perf_event_task_sched_out(prev, next); |
d7822b1e | 3641 | rseq_preempt(prev); |
e107be36 | 3642 | fire_sched_out_preempt_notifiers(prev, next); |
31cb1bc0 | 3643 | prepare_task(next); |
4866cde0 NP |
3644 | prepare_arch_switch(next); |
3645 | } | |
3646 | ||
1da177e4 LT |
3647 | /** |
3648 | * finish_task_switch - clean up after a task-switch | |
3649 | * @prev: the thread we just switched away from. | |
3650 | * | |
4866cde0 NP |
3651 | * finish_task_switch must be called after the context switch, paired |
3652 | * with a prepare_task_switch call before the context switch. | |
3653 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
3654 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
3655 | * |
3656 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 3657 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
3658 | * with the lock held can cause deadlocks; see schedule() for |
3659 | * details.) | |
dfa50b60 ON |
3660 | * |
3661 | * The context switch have flipped the stack from under us and restored the | |
3662 | * local variables which were saved when this task called schedule() in the | |
3663 | * past. prev == current is still correct but we need to recalculate this_rq | |
3664 | * because prev may have moved to another CPU. | |
1da177e4 | 3665 | */ |
dfa50b60 | 3666 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
3667 | __releases(rq->lock) |
3668 | { | |
dfa50b60 | 3669 | struct rq *rq = this_rq(); |
1da177e4 | 3670 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 3671 | long prev_state; |
1da177e4 | 3672 | |
609ca066 PZ |
3673 | /* |
3674 | * The previous task will have left us with a preempt_count of 2 | |
3675 | * because it left us after: | |
3676 | * | |
3677 | * schedule() | |
3678 | * preempt_disable(); // 1 | |
3679 | * __schedule() | |
3680 | * raw_spin_lock_irq(&rq->lock) // 2 | |
3681 | * | |
3682 | * Also, see FORK_PREEMPT_COUNT. | |
3683 | */ | |
e2bf1c4b PZ |
3684 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
3685 | "corrupted preempt_count: %s/%d/0x%x\n", | |
3686 | current->comm, current->pid, preempt_count())) | |
3687 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 3688 | |
1da177e4 LT |
3689 | rq->prev_mm = NULL; |
3690 | ||
3691 | /* | |
3692 | * A task struct has one reference for the use as "current". | |
c394cc9f | 3693 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
3694 | * schedule one last time. The schedule call will never return, and |
3695 | * the scheduled task must drop that reference. | |
95913d97 PZ |
3696 | * |
3697 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 3698 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
3699 | * running on another CPU and we could rave with its RUNNING -> DEAD |
3700 | * transition, resulting in a double drop. | |
1da177e4 | 3701 | */ |
55a101f8 | 3702 | prev_state = prev->state; |
bf9fae9f | 3703 | vtime_task_switch(prev); |
a8d757ef | 3704 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 3705 | finish_task(prev); |
3706 | finish_lock_switch(rq); | |
01f23e16 | 3707 | finish_arch_post_lock_switch(); |
0ed557aa | 3708 | kcov_finish_switch(current); |
e8fa1362 | 3709 | |
e107be36 | 3710 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 3711 | /* |
70216e18 MD |
3712 | * When switching through a kernel thread, the loop in |
3713 | * membarrier_{private,global}_expedited() may have observed that | |
3714 | * kernel thread and not issued an IPI. It is therefore possible to | |
3715 | * schedule between user->kernel->user threads without passing though | |
3716 | * switch_mm(). Membarrier requires a barrier after storing to | |
3717 | * rq->curr, before returning to userspace, so provide them here: | |
3718 | * | |
3719 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
3720 | * provided by mmdrop(), | |
3721 | * - a sync_core for SYNC_CORE. | |
306e0604 | 3722 | */ |
70216e18 MD |
3723 | if (mm) { |
3724 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 3725 | mmdrop(mm); |
70216e18 | 3726 | } |
1cef1150 PZ |
3727 | if (unlikely(prev_state == TASK_DEAD)) { |
3728 | if (prev->sched_class->task_dead) | |
3729 | prev->sched_class->task_dead(prev); | |
68f24b08 | 3730 | |
1cef1150 PZ |
3731 | /* |
3732 | * Remove function-return probe instances associated with this | |
3733 | * task and put them back on the free list. | |
3734 | */ | |
3735 | kprobe_flush_task(prev); | |
3736 | ||
3737 | /* Task is done with its stack. */ | |
3738 | put_task_stack(prev); | |
3739 | ||
0ff7b2cf | 3740 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 3741 | } |
99e5ada9 | 3742 | |
de734f89 | 3743 | tick_nohz_task_switch(); |
dfa50b60 | 3744 | return rq; |
1da177e4 LT |
3745 | } |
3746 | ||
3747 | /** | |
3748 | * schedule_tail - first thing a freshly forked thread must call. | |
3749 | * @prev: the thread we just switched away from. | |
3750 | */ | |
722a9f92 | 3751 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3752 | __releases(rq->lock) |
3753 | { | |
1a43a14a | 3754 | struct rq *rq; |
da19ab51 | 3755 | |
609ca066 PZ |
3756 | /* |
3757 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
3758 | * finish_task_switch() for details. | |
3759 | * | |
3760 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
3761 | * and the preempt_enable() will end up enabling preemption (on | |
3762 | * PREEMPT_COUNT kernels). | |
3763 | */ | |
3764 | ||
dfa50b60 | 3765 | rq = finish_task_switch(prev); |
1a43a14a | 3766 | preempt_enable(); |
70b97a7f | 3767 | |
1da177e4 | 3768 | if (current->set_child_tid) |
b488893a | 3769 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
3770 | |
3771 | calculate_sigpending(); | |
1da177e4 LT |
3772 | } |
3773 | ||
3774 | /* | |
dfa50b60 | 3775 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 3776 | */ |
04936948 | 3777 | static __always_inline struct rq * |
70b97a7f | 3778 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 3779 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 3780 | { |
e107be36 | 3781 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3782 | |
9226d125 ZA |
3783 | /* |
3784 | * For paravirt, this is coupled with an exit in switch_to to | |
3785 | * combine the page table reload and the switch backend into | |
3786 | * one hypercall. | |
3787 | */ | |
224101ed | 3788 | arch_start_context_switch(prev); |
9226d125 | 3789 | |
306e0604 | 3790 | /* |
139d025c PZ |
3791 | * kernel -> kernel lazy + transfer active |
3792 | * user -> kernel lazy + mmgrab() active | |
3793 | * | |
3794 | * kernel -> user switch + mmdrop() active | |
3795 | * user -> user switch | |
306e0604 | 3796 | */ |
139d025c PZ |
3797 | if (!next->mm) { // to kernel |
3798 | enter_lazy_tlb(prev->active_mm, next); | |
3799 | ||
3800 | next->active_mm = prev->active_mm; | |
3801 | if (prev->mm) // from user | |
3802 | mmgrab(prev->active_mm); | |
3803 | else | |
3804 | prev->active_mm = NULL; | |
3805 | } else { // to user | |
227a4aad | 3806 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
3807 | /* |
3808 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 3809 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
3810 | * |
3811 | * The below provides this either through switch_mm(), or in | |
3812 | * case 'prev->active_mm == next->mm' through | |
3813 | * finish_task_switch()'s mmdrop(). | |
3814 | */ | |
139d025c | 3815 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 3816 | |
139d025c PZ |
3817 | if (!prev->mm) { // from kernel |
3818 | /* will mmdrop() in finish_task_switch(). */ | |
3819 | rq->prev_mm = prev->active_mm; | |
3820 | prev->active_mm = NULL; | |
3821 | } | |
1da177e4 | 3822 | } |
92509b73 | 3823 | |
cb42c9a3 | 3824 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 3825 | |
269d5992 | 3826 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
3827 | |
3828 | /* Here we just switch the register state and the stack. */ | |
3829 | switch_to(prev, next, prev); | |
dd41f596 | 3830 | barrier(); |
dfa50b60 ON |
3831 | |
3832 | return finish_task_switch(prev); | |
1da177e4 LT |
3833 | } |
3834 | ||
3835 | /* | |
1c3e8264 | 3836 | * nr_running and nr_context_switches: |
1da177e4 LT |
3837 | * |
3838 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 3839 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
3840 | */ |
3841 | unsigned long nr_running(void) | |
3842 | { | |
3843 | unsigned long i, sum = 0; | |
3844 | ||
3845 | for_each_online_cpu(i) | |
3846 | sum += cpu_rq(i)->nr_running; | |
3847 | ||
3848 | return sum; | |
f711f609 | 3849 | } |
1da177e4 | 3850 | |
2ee507c4 | 3851 | /* |
d1ccc66d | 3852 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
3853 | * |
3854 | * Caution: this function does not check that the caller has disabled | |
3855 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
3856 | * race. The caller is responsible to use it correctly, for example: | |
3857 | * | |
dfcb245e | 3858 | * - from a non-preemptible section (of course) |
00cc1633 DD |
3859 | * |
3860 | * - from a thread that is bound to a single CPU | |
3861 | * | |
3862 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
3863 | */ |
3864 | bool single_task_running(void) | |
3865 | { | |
00cc1633 | 3866 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
3867 | } |
3868 | EXPORT_SYMBOL(single_task_running); | |
3869 | ||
1da177e4 | 3870 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3871 | { |
cc94abfc SR |
3872 | int i; |
3873 | unsigned long long sum = 0; | |
46cb4b7c | 3874 | |
0a945022 | 3875 | for_each_possible_cpu(i) |
1da177e4 | 3876 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3877 | |
1da177e4 LT |
3878 | return sum; |
3879 | } | |
483b4ee6 | 3880 | |
145d952a DL |
3881 | /* |
3882 | * Consumers of these two interfaces, like for example the cpuidle menu | |
3883 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
3884 | * for a CPU that has IO-wait which might not even end up running the task when | |
3885 | * it does become runnable. | |
3886 | */ | |
3887 | ||
3888 | unsigned long nr_iowait_cpu(int cpu) | |
3889 | { | |
3890 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
3891 | } | |
3892 | ||
e33a9bba TH |
3893 | /* |
3894 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
3895 | * | |
3896 | * The idea behind IO-wait account is to account the idle time that we could | |
3897 | * have spend running if it were not for IO. That is, if we were to improve the | |
3898 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
3899 | * | |
3900 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
3901 | * idle time as IO-wait, because if the storage were faster, it could've been | |
3902 | * running and we'd not be idle. | |
3903 | * | |
3904 | * This has been extended to SMP, by doing the same for each CPU. This however | |
3905 | * is broken. | |
3906 | * | |
3907 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
3908 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
3909 | * though, if the storage were faster, both could've ran at the same time, | |
3910 | * utilising both CPUs. | |
3911 | * | |
3912 | * This means, that when looking globally, the current IO-wait accounting on | |
3913 | * SMP is a lower bound, by reason of under accounting. | |
3914 | * | |
3915 | * Worse, since the numbers are provided per CPU, they are sometimes | |
3916 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
3917 | * associated with any one particular CPU, it can wake to another CPU than it | |
3918 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
3919 | * | |
3920 | * Task CPU affinities can make all that even more 'interesting'. | |
3921 | */ | |
3922 | ||
1da177e4 LT |
3923 | unsigned long nr_iowait(void) |
3924 | { | |
3925 | unsigned long i, sum = 0; | |
483b4ee6 | 3926 | |
0a945022 | 3927 | for_each_possible_cpu(i) |
145d952a | 3928 | sum += nr_iowait_cpu(i); |
46cb4b7c | 3929 | |
1da177e4 LT |
3930 | return sum; |
3931 | } | |
483b4ee6 | 3932 | |
dd41f596 | 3933 | #ifdef CONFIG_SMP |
8a0be9ef | 3934 | |
46cb4b7c | 3935 | /* |
38022906 PZ |
3936 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3937 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3938 | */ |
38022906 | 3939 | void sched_exec(void) |
46cb4b7c | 3940 | { |
38022906 | 3941 | struct task_struct *p = current; |
1da177e4 | 3942 | unsigned long flags; |
0017d735 | 3943 | int dest_cpu; |
46cb4b7c | 3944 | |
8f42ced9 | 3945 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 3946 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3947 | if (dest_cpu == smp_processor_id()) |
3948 | goto unlock; | |
38022906 | 3949 | |
8f42ced9 | 3950 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3951 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3952 | |
8f42ced9 PZ |
3953 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3954 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3955 | return; |
3956 | } | |
0017d735 | 3957 | unlock: |
8f42ced9 | 3958 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3959 | } |
dd41f596 | 3960 | |
1da177e4 LT |
3961 | #endif |
3962 | ||
1da177e4 | 3963 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 3964 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
3965 | |
3966 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 3967 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 3968 | |
6075620b GG |
3969 | /* |
3970 | * The function fair_sched_class.update_curr accesses the struct curr | |
3971 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
3972 | * we observe a high rate of cache misses in practice. | |
3973 | * Prefetching this data results in improved performance. | |
3974 | */ | |
3975 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3976 | { | |
3977 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3978 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3979 | #else | |
3980 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3981 | #endif | |
3982 | prefetch(curr); | |
3983 | prefetch(&curr->exec_start); | |
3984 | } | |
3985 | ||
c5f8d995 HS |
3986 | /* |
3987 | * Return accounted runtime for the task. | |
3988 | * In case the task is currently running, return the runtime plus current's | |
3989 | * pending runtime that have not been accounted yet. | |
3990 | */ | |
3991 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3992 | { | |
eb580751 | 3993 | struct rq_flags rf; |
c5f8d995 | 3994 | struct rq *rq; |
6e998916 | 3995 | u64 ns; |
c5f8d995 | 3996 | |
911b2898 PZ |
3997 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3998 | /* | |
97fb7a0a | 3999 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
4000 | * So we have a optimization chance when the task's delta_exec is 0. |
4001 | * Reading ->on_cpu is racy, but this is ok. | |
4002 | * | |
d1ccc66d IM |
4003 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
4004 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 4005 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
4006 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
4007 | * been accounted, so we're correct here as well. | |
911b2898 | 4008 | */ |
da0c1e65 | 4009 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
4010 | return p->se.sum_exec_runtime; |
4011 | #endif | |
4012 | ||
eb580751 | 4013 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
4014 | /* |
4015 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
4016 | * project cycles that may never be accounted to this | |
4017 | * thread, breaking clock_gettime(). | |
4018 | */ | |
4019 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 4020 | prefetch_curr_exec_start(p); |
6e998916 SG |
4021 | update_rq_clock(rq); |
4022 | p->sched_class->update_curr(rq); | |
4023 | } | |
4024 | ns = p->se.sum_exec_runtime; | |
eb580751 | 4025 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
4026 | |
4027 | return ns; | |
4028 | } | |
48f24c4d | 4029 | |
7835b98b CL |
4030 | /* |
4031 | * This function gets called by the timer code, with HZ frequency. | |
4032 | * We call it with interrupts disabled. | |
7835b98b CL |
4033 | */ |
4034 | void scheduler_tick(void) | |
4035 | { | |
7835b98b CL |
4036 | int cpu = smp_processor_id(); |
4037 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4038 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 4039 | struct rq_flags rf; |
b4eccf5f | 4040 | unsigned long thermal_pressure; |
3e51f33f | 4041 | |
1567c3e3 | 4042 | arch_scale_freq_tick(); |
3e51f33f | 4043 | sched_clock_tick(); |
dd41f596 | 4044 | |
8a8c69c3 PZ |
4045 | rq_lock(rq, &rf); |
4046 | ||
3e51f33f | 4047 | update_rq_clock(rq); |
b4eccf5f | 4048 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 4049 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 4050 | curr->sched_class->task_tick(rq, curr, 0); |
3289bdb4 | 4051 | calc_global_load_tick(rq); |
eb414681 | 4052 | psi_task_tick(rq); |
8a8c69c3 PZ |
4053 | |
4054 | rq_unlock(rq, &rf); | |
7835b98b | 4055 | |
e9d2b064 | 4056 | perf_event_task_tick(); |
e220d2dc | 4057 | |
e418e1c2 | 4058 | #ifdef CONFIG_SMP |
6eb57e0d | 4059 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 4060 | trigger_load_balance(rq); |
e418e1c2 | 4061 | #endif |
1da177e4 LT |
4062 | } |
4063 | ||
265f22a9 | 4064 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
4065 | |
4066 | struct tick_work { | |
4067 | int cpu; | |
b55bd585 | 4068 | atomic_t state; |
d84b3131 FW |
4069 | struct delayed_work work; |
4070 | }; | |
b55bd585 PM |
4071 | /* Values for ->state, see diagram below. */ |
4072 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
4073 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
4074 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
4075 | ||
4076 | /* | |
4077 | * State diagram for ->state: | |
4078 | * | |
4079 | * | |
4080 | * TICK_SCHED_REMOTE_OFFLINE | |
4081 | * | ^ | |
4082 | * | | | |
4083 | * | | sched_tick_remote() | |
4084 | * | | | |
4085 | * | | | |
4086 | * +--TICK_SCHED_REMOTE_OFFLINING | |
4087 | * | ^ | |
4088 | * | | | |
4089 | * sched_tick_start() | | sched_tick_stop() | |
4090 | * | | | |
4091 | * V | | |
4092 | * TICK_SCHED_REMOTE_RUNNING | |
4093 | * | |
4094 | * | |
4095 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
4096 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
4097 | */ | |
d84b3131 FW |
4098 | |
4099 | static struct tick_work __percpu *tick_work_cpu; | |
4100 | ||
4101 | static void sched_tick_remote(struct work_struct *work) | |
4102 | { | |
4103 | struct delayed_work *dwork = to_delayed_work(work); | |
4104 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
4105 | int cpu = twork->cpu; | |
4106 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 4107 | struct task_struct *curr; |
d84b3131 | 4108 | struct rq_flags rf; |
d9c0ffca | 4109 | u64 delta; |
b55bd585 | 4110 | int os; |
d84b3131 FW |
4111 | |
4112 | /* | |
4113 | * Handle the tick only if it appears the remote CPU is running in full | |
4114 | * dynticks mode. The check is racy by nature, but missing a tick or | |
4115 | * having one too much is no big deal because the scheduler tick updates | |
4116 | * statistics and checks timeslices in a time-independent way, regardless | |
4117 | * of when exactly it is running. | |
4118 | */ | |
488603b8 | 4119 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 4120 | goto out_requeue; |
d84b3131 | 4121 | |
d9c0ffca FW |
4122 | rq_lock_irq(rq, &rf); |
4123 | curr = rq->curr; | |
488603b8 | 4124 | if (cpu_is_offline(cpu)) |
d9c0ffca | 4125 | goto out_unlock; |
d84b3131 | 4126 | |
d9c0ffca | 4127 | update_rq_clock(rq); |
d9c0ffca | 4128 | |
488603b8 SW |
4129 | if (!is_idle_task(curr)) { |
4130 | /* | |
4131 | * Make sure the next tick runs within a reasonable | |
4132 | * amount of time. | |
4133 | */ | |
4134 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
4135 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
4136 | } | |
d9c0ffca FW |
4137 | curr->sched_class->task_tick(rq, curr, 0); |
4138 | ||
ebc0f83c | 4139 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
4140 | out_unlock: |
4141 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 4142 | out_requeue: |
ebc0f83c | 4143 | |
d84b3131 FW |
4144 | /* |
4145 | * Run the remote tick once per second (1Hz). This arbitrary | |
4146 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
4147 | * to keep scheduler internal stats reasonably up to date. But |
4148 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 4149 | */ |
b55bd585 PM |
4150 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
4151 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
4152 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
4153 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
4154 | } |
4155 | ||
4156 | static void sched_tick_start(int cpu) | |
4157 | { | |
b55bd585 | 4158 | int os; |
d84b3131 FW |
4159 | struct tick_work *twork; |
4160 | ||
4161 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4162 | return; | |
4163 | ||
4164 | WARN_ON_ONCE(!tick_work_cpu); | |
4165 | ||
4166 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4167 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
4168 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
4169 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
4170 | twork->cpu = cpu; | |
4171 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
4172 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
4173 | } | |
d84b3131 FW |
4174 | } |
4175 | ||
4176 | #ifdef CONFIG_HOTPLUG_CPU | |
4177 | static void sched_tick_stop(int cpu) | |
4178 | { | |
4179 | struct tick_work *twork; | |
b55bd585 | 4180 | int os; |
d84b3131 FW |
4181 | |
4182 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4183 | return; | |
4184 | ||
4185 | WARN_ON_ONCE(!tick_work_cpu); | |
4186 | ||
4187 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4188 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
4189 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
4190 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
4191 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
4192 | } |
4193 | #endif /* CONFIG_HOTPLUG_CPU */ | |
4194 | ||
4195 | int __init sched_tick_offload_init(void) | |
4196 | { | |
4197 | tick_work_cpu = alloc_percpu(struct tick_work); | |
4198 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
4199 | return 0; |
4200 | } | |
4201 | ||
4202 | #else /* !CONFIG_NO_HZ_FULL */ | |
4203 | static inline void sched_tick_start(int cpu) { } | |
4204 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 4205 | #endif |
1da177e4 | 4206 | |
c1a280b6 | 4207 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 4208 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
4209 | /* |
4210 | * If the value passed in is equal to the current preempt count | |
4211 | * then we just disabled preemption. Start timing the latency. | |
4212 | */ | |
4213 | static inline void preempt_latency_start(int val) | |
4214 | { | |
4215 | if (preempt_count() == val) { | |
4216 | unsigned long ip = get_lock_parent_ip(); | |
4217 | #ifdef CONFIG_DEBUG_PREEMPT | |
4218 | current->preempt_disable_ip = ip; | |
4219 | #endif | |
4220 | trace_preempt_off(CALLER_ADDR0, ip); | |
4221 | } | |
4222 | } | |
7e49fcce | 4223 | |
edafe3a5 | 4224 | void preempt_count_add(int val) |
1da177e4 | 4225 | { |
6cd8a4bb | 4226 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4227 | /* |
4228 | * Underflow? | |
4229 | */ | |
9a11b49a IM |
4230 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4231 | return; | |
6cd8a4bb | 4232 | #endif |
bdb43806 | 4233 | __preempt_count_add(val); |
6cd8a4bb | 4234 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4235 | /* |
4236 | * Spinlock count overflowing soon? | |
4237 | */ | |
33859f7f MOS |
4238 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4239 | PREEMPT_MASK - 10); | |
6cd8a4bb | 4240 | #endif |
47252cfb | 4241 | preempt_latency_start(val); |
1da177e4 | 4242 | } |
bdb43806 | 4243 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 4244 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 4245 | |
47252cfb SR |
4246 | /* |
4247 | * If the value passed in equals to the current preempt count | |
4248 | * then we just enabled preemption. Stop timing the latency. | |
4249 | */ | |
4250 | static inline void preempt_latency_stop(int val) | |
4251 | { | |
4252 | if (preempt_count() == val) | |
4253 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
4254 | } | |
4255 | ||
edafe3a5 | 4256 | void preempt_count_sub(int val) |
1da177e4 | 4257 | { |
6cd8a4bb | 4258 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4259 | /* |
4260 | * Underflow? | |
4261 | */ | |
01e3eb82 | 4262 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4263 | return; |
1da177e4 LT |
4264 | /* |
4265 | * Is the spinlock portion underflowing? | |
4266 | */ | |
9a11b49a IM |
4267 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4268 | !(preempt_count() & PREEMPT_MASK))) | |
4269 | return; | |
6cd8a4bb | 4270 | #endif |
9a11b49a | 4271 | |
47252cfb | 4272 | preempt_latency_stop(val); |
bdb43806 | 4273 | __preempt_count_sub(val); |
1da177e4 | 4274 | } |
bdb43806 | 4275 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 4276 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 4277 | |
47252cfb SR |
4278 | #else |
4279 | static inline void preempt_latency_start(int val) { } | |
4280 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
4281 | #endif |
4282 | ||
59ddbcb2 IM |
4283 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
4284 | { | |
4285 | #ifdef CONFIG_DEBUG_PREEMPT | |
4286 | return p->preempt_disable_ip; | |
4287 | #else | |
4288 | return 0; | |
4289 | #endif | |
4290 | } | |
4291 | ||
1da177e4 | 4292 | /* |
dd41f596 | 4293 | * Print scheduling while atomic bug: |
1da177e4 | 4294 | */ |
dd41f596 | 4295 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4296 | { |
d1c6d149 VN |
4297 | /* Save this before calling printk(), since that will clobber it */ |
4298 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
4299 | ||
664dfa65 DJ |
4300 | if (oops_in_progress) |
4301 | return; | |
4302 | ||
3df0fc5b PZ |
4303 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4304 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4305 | |
dd41f596 | 4306 | debug_show_held_locks(prev); |
e21f5b15 | 4307 | print_modules(); |
dd41f596 IM |
4308 | if (irqs_disabled()) |
4309 | print_irqtrace_events(prev); | |
d1c6d149 VN |
4310 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
4311 | && in_atomic_preempt_off()) { | |
8f47b187 | 4312 | pr_err("Preemption disabled at:"); |
2062a4e8 | 4313 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 4314 | } |
748c7201 DBO |
4315 | if (panic_on_warn) |
4316 | panic("scheduling while atomic\n"); | |
4317 | ||
6135fc1e | 4318 | dump_stack(); |
373d4d09 | 4319 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 4320 | } |
1da177e4 | 4321 | |
dd41f596 IM |
4322 | /* |
4323 | * Various schedule()-time debugging checks and statistics: | |
4324 | */ | |
312364f3 | 4325 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 4326 | { |
0d9e2632 | 4327 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
4328 | if (task_stack_end_corrupted(prev)) |
4329 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
4330 | |
4331 | if (task_scs_end_corrupted(prev)) | |
4332 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 4333 | #endif |
b99def8b | 4334 | |
312364f3 DV |
4335 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
4336 | if (!preempt && prev->state && prev->non_block_count) { | |
4337 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", | |
4338 | prev->comm, prev->pid, prev->non_block_count); | |
4339 | dump_stack(); | |
4340 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
4341 | } | |
4342 | #endif | |
4343 | ||
1dc0fffc | 4344 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 4345 | __schedule_bug(prev); |
1dc0fffc PZ |
4346 | preempt_count_set(PREEMPT_DISABLED); |
4347 | } | |
b3fbab05 | 4348 | rcu_sleep_check(); |
dd41f596 | 4349 | |
1da177e4 LT |
4350 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4351 | ||
ae92882e | 4352 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
4353 | } |
4354 | ||
457d1f46 CY |
4355 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
4356 | struct rq_flags *rf) | |
4357 | { | |
4358 | #ifdef CONFIG_SMP | |
4359 | const struct sched_class *class; | |
4360 | /* | |
4361 | * We must do the balancing pass before put_prev_task(), such | |
4362 | * that when we release the rq->lock the task is in the same | |
4363 | * state as before we took rq->lock. | |
4364 | * | |
4365 | * We can terminate the balance pass as soon as we know there is | |
4366 | * a runnable task of @class priority or higher. | |
4367 | */ | |
4368 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
4369 | if (class->balance(rq, prev, rf)) | |
4370 | break; | |
4371 | } | |
4372 | #endif | |
4373 | ||
4374 | put_prev_task(rq, prev); | |
4375 | } | |
4376 | ||
dd41f596 IM |
4377 | /* |
4378 | * Pick up the highest-prio task: | |
4379 | */ | |
4380 | static inline struct task_struct * | |
d8ac8971 | 4381 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 4382 | { |
49ee5768 | 4383 | const struct sched_class *class; |
dd41f596 | 4384 | struct task_struct *p; |
1da177e4 LT |
4385 | |
4386 | /* | |
0ba87bb2 PZ |
4387 | * Optimization: we know that if all tasks are in the fair class we can |
4388 | * call that function directly, but only if the @prev task wasn't of a | |
4389 | * higher scheduling class, because otherwise those loose the | |
4390 | * opportunity to pull in more work from other CPUs. | |
1da177e4 | 4391 | */ |
aa93cd53 | 4392 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
4393 | rq->nr_running == rq->cfs.h_nr_running)) { |
4394 | ||
5d7d6056 | 4395 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 4396 | if (unlikely(p == RETRY_TASK)) |
67692435 | 4397 | goto restart; |
6ccdc84b | 4398 | |
d1ccc66d | 4399 | /* Assumes fair_sched_class->next == idle_sched_class */ |
5d7d6056 | 4400 | if (!p) { |
f488e105 | 4401 | put_prev_task(rq, prev); |
98c2f700 | 4402 | p = pick_next_task_idle(rq); |
f488e105 | 4403 | } |
6ccdc84b PZ |
4404 | |
4405 | return p; | |
1da177e4 LT |
4406 | } |
4407 | ||
67692435 | 4408 | restart: |
457d1f46 | 4409 | put_prev_task_balance(rq, prev, rf); |
67692435 | 4410 | |
34f971f6 | 4411 | for_each_class(class) { |
98c2f700 | 4412 | p = class->pick_next_task(rq); |
67692435 | 4413 | if (p) |
dd41f596 | 4414 | return p; |
dd41f596 | 4415 | } |
34f971f6 | 4416 | |
d1ccc66d IM |
4417 | /* The idle class should always have a runnable task: */ |
4418 | BUG(); | |
dd41f596 | 4419 | } |
1da177e4 | 4420 | |
dd41f596 | 4421 | /* |
c259e01a | 4422 | * __schedule() is the main scheduler function. |
edde96ea PE |
4423 | * |
4424 | * The main means of driving the scheduler and thus entering this function are: | |
4425 | * | |
4426 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
4427 | * | |
4428 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
4429 | * paths. For example, see arch/x86/entry_64.S. | |
4430 | * | |
4431 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
4432 | * interrupt handler scheduler_tick(). | |
4433 | * | |
4434 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
4435 | * task to the run-queue and that's it. | |
4436 | * | |
4437 | * Now, if the new task added to the run-queue preempts the current | |
4438 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
4439 | * called on the nearest possible occasion: | |
4440 | * | |
c1a280b6 | 4441 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
4442 | * |
4443 | * - in syscall or exception context, at the next outmost | |
4444 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
4445 | * spin_unlock()!) | |
4446 | * | |
4447 | * - in IRQ context, return from interrupt-handler to | |
4448 | * preemptible context | |
4449 | * | |
c1a280b6 | 4450 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
4451 | * then at the next: |
4452 | * | |
4453 | * - cond_resched() call | |
4454 | * - explicit schedule() call | |
4455 | * - return from syscall or exception to user-space | |
4456 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 4457 | * |
b30f0e3f | 4458 | * WARNING: must be called with preemption disabled! |
dd41f596 | 4459 | */ |
499d7955 | 4460 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
4461 | { |
4462 | struct task_struct *prev, *next; | |
67ca7bde | 4463 | unsigned long *switch_count; |
dbfb089d | 4464 | unsigned long prev_state; |
d8ac8971 | 4465 | struct rq_flags rf; |
dd41f596 | 4466 | struct rq *rq; |
31656519 | 4467 | int cpu; |
dd41f596 | 4468 | |
dd41f596 IM |
4469 | cpu = smp_processor_id(); |
4470 | rq = cpu_rq(cpu); | |
dd41f596 | 4471 | prev = rq->curr; |
dd41f596 | 4472 | |
312364f3 | 4473 | schedule_debug(prev, preempt); |
1da177e4 | 4474 | |
31656519 | 4475 | if (sched_feat(HRTICK)) |
f333fdc9 | 4476 | hrtick_clear(rq); |
8f4d37ec | 4477 | |
46a5d164 | 4478 | local_irq_disable(); |
bcbfdd01 | 4479 | rcu_note_context_switch(preempt); |
46a5d164 | 4480 | |
e0acd0a6 ON |
4481 | /* |
4482 | * Make sure that signal_pending_state()->signal_pending() below | |
4483 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
4484 | * done by the caller to avoid the race with signal_wake_up(): |
4485 | * | |
4486 | * __set_current_state(@state) signal_wake_up() | |
4487 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
4488 | * wake_up_state(p, state) | |
4489 | * LOCK rq->lock LOCK p->pi_state | |
4490 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
4491 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 4492 | * |
dbfb089d | 4493 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 4494 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 4495 | */ |
8a8c69c3 | 4496 | rq_lock(rq, &rf); |
d89e588c | 4497 | smp_mb__after_spinlock(); |
1da177e4 | 4498 | |
d1ccc66d IM |
4499 | /* Promote REQ to ACT */ |
4500 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 4501 | update_rq_clock(rq); |
9edfbfed | 4502 | |
246d86b5 | 4503 | switch_count = &prev->nivcsw; |
d136122f | 4504 | |
dbfb089d | 4505 | /* |
d136122f PZ |
4506 | * We must load prev->state once (task_struct::state is volatile), such |
4507 | * that: | |
4508 | * | |
4509 | * - we form a control dependency vs deactivate_task() below. | |
4510 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 4511 | */ |
d136122f PZ |
4512 | prev_state = prev->state; |
4513 | if (!preempt && prev_state) { | |
dbfb089d | 4514 | if (signal_pending_state(prev_state, prev)) { |
1da177e4 | 4515 | prev->state = TASK_RUNNING; |
21aa9af0 | 4516 | } else { |
dbfb089d PZ |
4517 | prev->sched_contributes_to_load = |
4518 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
4519 | !(prev_state & TASK_NOLOAD) && | |
4520 | !(prev->flags & PF_FROZEN); | |
4521 | ||
4522 | if (prev->sched_contributes_to_load) | |
4523 | rq->nr_uninterruptible++; | |
4524 | ||
4525 | /* | |
4526 | * __schedule() ttwu() | |
d136122f PZ |
4527 | * prev_state = prev->state; if (p->on_rq && ...) |
4528 | * if (prev_state) goto out; | |
4529 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
4530 | * p->state = TASK_WAKING | |
4531 | * | |
4532 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
4533 | * |
4534 | * After this, schedule() must not care about p->state any more. | |
4535 | */ | |
bce4dc80 | 4536 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 4537 | |
e33a9bba TH |
4538 | if (prev->in_iowait) { |
4539 | atomic_inc(&rq->nr_iowait); | |
4540 | delayacct_blkio_start(); | |
4541 | } | |
21aa9af0 | 4542 | } |
dd41f596 | 4543 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4544 | } |
4545 | ||
d8ac8971 | 4546 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 4547 | clear_tsk_need_resched(prev); |
f27dde8d | 4548 | clear_preempt_need_resched(); |
1da177e4 | 4549 | |
1da177e4 | 4550 | if (likely(prev != next)) { |
1da177e4 | 4551 | rq->nr_switches++; |
5311a98f EB |
4552 | /* |
4553 | * RCU users of rcu_dereference(rq->curr) may not see | |
4554 | * changes to task_struct made by pick_next_task(). | |
4555 | */ | |
4556 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
4557 | /* |
4558 | * The membarrier system call requires each architecture | |
4559 | * to have a full memory barrier after updating | |
306e0604 MD |
4560 | * rq->curr, before returning to user-space. |
4561 | * | |
4562 | * Here are the schemes providing that barrier on the | |
4563 | * various architectures: | |
4564 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
4565 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
4566 | * - finish_lock_switch() for weakly-ordered | |
4567 | * architectures where spin_unlock is a full barrier, | |
4568 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
4569 | * is a RELEASE barrier), | |
22e4ebb9 | 4570 | */ |
1da177e4 LT |
4571 | ++*switch_count; |
4572 | ||
b05e75d6 JW |
4573 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
4574 | ||
c73464b1 | 4575 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
4576 | |
4577 | /* Also unlocks the rq: */ | |
4578 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 4579 | } else { |
cb42c9a3 | 4580 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 4581 | |
565790d2 PZ |
4582 | rq_unpin_lock(rq, &rf); |
4583 | __balance_callbacks(rq); | |
4584 | raw_spin_unlock_irq(&rq->lock); | |
4585 | } | |
1da177e4 | 4586 | } |
c259e01a | 4587 | |
9af6528e PZ |
4588 | void __noreturn do_task_dead(void) |
4589 | { | |
d1ccc66d | 4590 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 4591 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
4592 | |
4593 | /* Tell freezer to ignore us: */ | |
4594 | current->flags |= PF_NOFREEZE; | |
4595 | ||
9af6528e PZ |
4596 | __schedule(false); |
4597 | BUG(); | |
d1ccc66d IM |
4598 | |
4599 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 4600 | for (;;) |
d1ccc66d | 4601 | cpu_relax(); |
9af6528e PZ |
4602 | } |
4603 | ||
9c40cef2 TG |
4604 | static inline void sched_submit_work(struct task_struct *tsk) |
4605 | { | |
c1cecf88 SAS |
4606 | unsigned int task_flags; |
4607 | ||
b0fdc013 | 4608 | if (!tsk->state) |
9c40cef2 | 4609 | return; |
6d25be57 | 4610 | |
c1cecf88 | 4611 | task_flags = tsk->flags; |
6d25be57 TG |
4612 | /* |
4613 | * If a worker went to sleep, notify and ask workqueue whether | |
4614 | * it wants to wake up a task to maintain concurrency. | |
4615 | * As this function is called inside the schedule() context, | |
4616 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
4617 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
4618 | * requires it. | |
6d25be57 | 4619 | */ |
c1cecf88 | 4620 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 4621 | preempt_disable(); |
c1cecf88 | 4622 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
4623 | wq_worker_sleeping(tsk); |
4624 | else | |
4625 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
4626 | preempt_enable_no_resched(); |
4627 | } | |
4628 | ||
b0fdc013 SAS |
4629 | if (tsk_is_pi_blocked(tsk)) |
4630 | return; | |
4631 | ||
9c40cef2 TG |
4632 | /* |
4633 | * If we are going to sleep and we have plugged IO queued, | |
4634 | * make sure to submit it to avoid deadlocks. | |
4635 | */ | |
4636 | if (blk_needs_flush_plug(tsk)) | |
4637 | blk_schedule_flush_plug(tsk); | |
4638 | } | |
4639 | ||
6d25be57 TG |
4640 | static void sched_update_worker(struct task_struct *tsk) |
4641 | { | |
771b53d0 JA |
4642 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
4643 | if (tsk->flags & PF_WQ_WORKER) | |
4644 | wq_worker_running(tsk); | |
4645 | else | |
4646 | io_wq_worker_running(tsk); | |
4647 | } | |
6d25be57 TG |
4648 | } |
4649 | ||
722a9f92 | 4650 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 4651 | { |
9c40cef2 TG |
4652 | struct task_struct *tsk = current; |
4653 | ||
4654 | sched_submit_work(tsk); | |
bfd9b2b5 | 4655 | do { |
b30f0e3f | 4656 | preempt_disable(); |
fc13aeba | 4657 | __schedule(false); |
b30f0e3f | 4658 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 4659 | } while (need_resched()); |
6d25be57 | 4660 | sched_update_worker(tsk); |
c259e01a | 4661 | } |
1da177e4 LT |
4662 | EXPORT_SYMBOL(schedule); |
4663 | ||
8663effb SRV |
4664 | /* |
4665 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
4666 | * state (have scheduled out non-voluntarily) by making sure that all | |
4667 | * tasks have either left the run queue or have gone into user space. | |
4668 | * As idle tasks do not do either, they must not ever be preempted | |
4669 | * (schedule out non-voluntarily). | |
4670 | * | |
4671 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
4672 | * never enables preemption because it does not call sched_submit_work(). | |
4673 | */ | |
4674 | void __sched schedule_idle(void) | |
4675 | { | |
4676 | /* | |
4677 | * As this skips calling sched_submit_work(), which the idle task does | |
4678 | * regardless because that function is a nop when the task is in a | |
4679 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
4680 | * current task can be in any other state. Note, idle is always in the | |
4681 | * TASK_RUNNING state. | |
4682 | */ | |
4683 | WARN_ON_ONCE(current->state); | |
4684 | do { | |
4685 | __schedule(false); | |
4686 | } while (need_resched()); | |
4687 | } | |
4688 | ||
91d1aa43 | 4689 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 4690 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
4691 | { |
4692 | /* | |
4693 | * If we come here after a random call to set_need_resched(), | |
4694 | * or we have been woken up remotely but the IPI has not yet arrived, | |
4695 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
4696 | * we find a better solution. | |
7cc78f8f AL |
4697 | * |
4698 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 4699 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 4700 | * too frequently to make sense yet. |
20ab65e3 | 4701 | */ |
7cc78f8f | 4702 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 4703 | schedule(); |
7cc78f8f | 4704 | exception_exit(prev_state); |
20ab65e3 FW |
4705 | } |
4706 | #endif | |
4707 | ||
c5491ea7 TG |
4708 | /** |
4709 | * schedule_preempt_disabled - called with preemption disabled | |
4710 | * | |
4711 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
4712 | */ | |
4713 | void __sched schedule_preempt_disabled(void) | |
4714 | { | |
ba74c144 | 4715 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
4716 | schedule(); |
4717 | preempt_disable(); | |
4718 | } | |
4719 | ||
06b1f808 | 4720 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
4721 | { |
4722 | do { | |
47252cfb SR |
4723 | /* |
4724 | * Because the function tracer can trace preempt_count_sub() | |
4725 | * and it also uses preempt_enable/disable_notrace(), if | |
4726 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
4727 | * by the function tracer will call this function again and | |
4728 | * cause infinite recursion. | |
4729 | * | |
4730 | * Preemption must be disabled here before the function | |
4731 | * tracer can trace. Break up preempt_disable() into two | |
4732 | * calls. One to disable preemption without fear of being | |
4733 | * traced. The other to still record the preemption latency, | |
4734 | * which can also be traced by the function tracer. | |
4735 | */ | |
499d7955 | 4736 | preempt_disable_notrace(); |
47252cfb | 4737 | preempt_latency_start(1); |
fc13aeba | 4738 | __schedule(true); |
47252cfb | 4739 | preempt_latency_stop(1); |
499d7955 | 4740 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
4741 | |
4742 | /* | |
4743 | * Check again in case we missed a preemption opportunity | |
4744 | * between schedule and now. | |
4745 | */ | |
a18b5d01 FW |
4746 | } while (need_resched()); |
4747 | } | |
4748 | ||
c1a280b6 | 4749 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 4750 | /* |
a49b4f40 VS |
4751 | * This is the entry point to schedule() from in-kernel preemption |
4752 | * off of preempt_enable. | |
1da177e4 | 4753 | */ |
722a9f92 | 4754 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 4755 | { |
1da177e4 LT |
4756 | /* |
4757 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4758 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4759 | */ |
fbb00b56 | 4760 | if (likely(!preemptible())) |
1da177e4 LT |
4761 | return; |
4762 | ||
a18b5d01 | 4763 | preempt_schedule_common(); |
1da177e4 | 4764 | } |
376e2424 | 4765 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 4766 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 4767 | |
009f60e2 | 4768 | /** |
4eaca0a8 | 4769 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
4770 | * |
4771 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
4772 | * recursion and tracing preempt enabling caused by the tracing | |
4773 | * infrastructure itself. But as tracing can happen in areas coming | |
4774 | * from userspace or just about to enter userspace, a preempt enable | |
4775 | * can occur before user_exit() is called. This will cause the scheduler | |
4776 | * to be called when the system is still in usermode. | |
4777 | * | |
4778 | * To prevent this, the preempt_enable_notrace will use this function | |
4779 | * instead of preempt_schedule() to exit user context if needed before | |
4780 | * calling the scheduler. | |
4781 | */ | |
4eaca0a8 | 4782 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
4783 | { |
4784 | enum ctx_state prev_ctx; | |
4785 | ||
4786 | if (likely(!preemptible())) | |
4787 | return; | |
4788 | ||
4789 | do { | |
47252cfb SR |
4790 | /* |
4791 | * Because the function tracer can trace preempt_count_sub() | |
4792 | * and it also uses preempt_enable/disable_notrace(), if | |
4793 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
4794 | * by the function tracer will call this function again and | |
4795 | * cause infinite recursion. | |
4796 | * | |
4797 | * Preemption must be disabled here before the function | |
4798 | * tracer can trace. Break up preempt_disable() into two | |
4799 | * calls. One to disable preemption without fear of being | |
4800 | * traced. The other to still record the preemption latency, | |
4801 | * which can also be traced by the function tracer. | |
4802 | */ | |
3d8f74dd | 4803 | preempt_disable_notrace(); |
47252cfb | 4804 | preempt_latency_start(1); |
009f60e2 ON |
4805 | /* |
4806 | * Needs preempt disabled in case user_exit() is traced | |
4807 | * and the tracer calls preempt_enable_notrace() causing | |
4808 | * an infinite recursion. | |
4809 | */ | |
4810 | prev_ctx = exception_enter(); | |
fc13aeba | 4811 | __schedule(true); |
009f60e2 ON |
4812 | exception_exit(prev_ctx); |
4813 | ||
47252cfb | 4814 | preempt_latency_stop(1); |
3d8f74dd | 4815 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
4816 | } while (need_resched()); |
4817 | } | |
4eaca0a8 | 4818 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 4819 | |
c1a280b6 | 4820 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 LT |
4821 | |
4822 | /* | |
a49b4f40 | 4823 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4824 | * off of irq context. |
4825 | * Note, that this is called and return with irqs disabled. This will | |
4826 | * protect us against recursive calling from irq. | |
4827 | */ | |
722a9f92 | 4828 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 4829 | { |
b22366cd | 4830 | enum ctx_state prev_state; |
6478d880 | 4831 | |
2ed6e34f | 4832 | /* Catch callers which need to be fixed */ |
f27dde8d | 4833 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 4834 | |
b22366cd FW |
4835 | prev_state = exception_enter(); |
4836 | ||
3a5c359a | 4837 | do { |
3d8f74dd | 4838 | preempt_disable(); |
3a5c359a | 4839 | local_irq_enable(); |
fc13aeba | 4840 | __schedule(true); |
3a5c359a | 4841 | local_irq_disable(); |
3d8f74dd | 4842 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 4843 | } while (need_resched()); |
b22366cd FW |
4844 | |
4845 | exception_exit(prev_state); | |
1da177e4 LT |
4846 | } |
4847 | ||
ac6424b9 | 4848 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4849 | void *key) |
1da177e4 | 4850 | { |
062d3f95 | 4851 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 4852 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4853 | } |
1da177e4 LT |
4854 | EXPORT_SYMBOL(default_wake_function); |
4855 | ||
b29739f9 IM |
4856 | #ifdef CONFIG_RT_MUTEXES |
4857 | ||
acd58620 PZ |
4858 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
4859 | { | |
4860 | if (pi_task) | |
4861 | prio = min(prio, pi_task->prio); | |
4862 | ||
4863 | return prio; | |
4864 | } | |
4865 | ||
4866 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
4867 | { | |
4868 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
4869 | ||
4870 | return __rt_effective_prio(pi_task, prio); | |
4871 | } | |
4872 | ||
b29739f9 IM |
4873 | /* |
4874 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
4875 | * @p: task to boost |
4876 | * @pi_task: donor task | |
b29739f9 IM |
4877 | * |
4878 | * This function changes the 'effective' priority of a task. It does | |
4879 | * not touch ->normal_prio like __setscheduler(). | |
4880 | * | |
c365c292 TG |
4881 | * Used by the rt_mutex code to implement priority inheritance |
4882 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 4883 | */ |
acd58620 | 4884 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 4885 | { |
acd58620 | 4886 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 4887 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 4888 | const struct sched_class *prev_class; |
eb580751 PZ |
4889 | struct rq_flags rf; |
4890 | struct rq *rq; | |
b29739f9 | 4891 | |
acd58620 PZ |
4892 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
4893 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
4894 | ||
4895 | /* | |
4896 | * If nothing changed; bail early. | |
4897 | */ | |
4898 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
4899 | return; | |
b29739f9 | 4900 | |
eb580751 | 4901 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 4902 | update_rq_clock(rq); |
acd58620 PZ |
4903 | /* |
4904 | * Set under pi_lock && rq->lock, such that the value can be used under | |
4905 | * either lock. | |
4906 | * | |
4907 | * Note that there is loads of tricky to make this pointer cache work | |
4908 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
4909 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
4910 | * task is allowed to run again (and can exit). This ensures the pointer | |
4911 | * points to a blocked task -- which guaratees the task is present. | |
4912 | */ | |
4913 | p->pi_top_task = pi_task; | |
4914 | ||
4915 | /* | |
4916 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
4917 | */ | |
4918 | if (prio == p->prio && !dl_prio(prio)) | |
4919 | goto out_unlock; | |
b29739f9 | 4920 | |
1c4dd99b TG |
4921 | /* |
4922 | * Idle task boosting is a nono in general. There is one | |
4923 | * exception, when PREEMPT_RT and NOHZ is active: | |
4924 | * | |
4925 | * The idle task calls get_next_timer_interrupt() and holds | |
4926 | * the timer wheel base->lock on the CPU and another CPU wants | |
4927 | * to access the timer (probably to cancel it). We can safely | |
4928 | * ignore the boosting request, as the idle CPU runs this code | |
4929 | * with interrupts disabled and will complete the lock | |
4930 | * protected section without being interrupted. So there is no | |
4931 | * real need to boost. | |
4932 | */ | |
4933 | if (unlikely(p == rq->idle)) { | |
4934 | WARN_ON(p != rq->curr); | |
4935 | WARN_ON(p->pi_blocked_on); | |
4936 | goto out_unlock; | |
4937 | } | |
4938 | ||
b91473ff | 4939 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 4940 | oldprio = p->prio; |
ff77e468 PZ |
4941 | |
4942 | if (oldprio == prio) | |
4943 | queue_flag &= ~DEQUEUE_MOVE; | |
4944 | ||
83ab0aa0 | 4945 | prev_class = p->sched_class; |
da0c1e65 | 4946 | queued = task_on_rq_queued(p); |
051a1d1a | 4947 | running = task_current(rq, p); |
da0c1e65 | 4948 | if (queued) |
ff77e468 | 4949 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 4950 | if (running) |
f3cd1c4e | 4951 | put_prev_task(rq, p); |
dd41f596 | 4952 | |
2d3d891d DF |
4953 | /* |
4954 | * Boosting condition are: | |
4955 | * 1. -rt task is running and holds mutex A | |
4956 | * --> -dl task blocks on mutex A | |
4957 | * | |
4958 | * 2. -dl task is running and holds mutex A | |
4959 | * --> -dl task blocks on mutex A and could preempt the | |
4960 | * running task | |
4961 | */ | |
4962 | if (dl_prio(prio)) { | |
466af29b | 4963 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
4964 | (pi_task && dl_prio(pi_task->prio) && |
4965 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 4966 | p->dl.dl_boosted = 1; |
ff77e468 | 4967 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
4968 | } else |
4969 | p->dl.dl_boosted = 0; | |
aab03e05 | 4970 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
4971 | } else if (rt_prio(prio)) { |
4972 | if (dl_prio(oldprio)) | |
4973 | p->dl.dl_boosted = 0; | |
4974 | if (oldprio < prio) | |
ff77e468 | 4975 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 4976 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
4977 | } else { |
4978 | if (dl_prio(oldprio)) | |
4979 | p->dl.dl_boosted = 0; | |
746db944 BS |
4980 | if (rt_prio(oldprio)) |
4981 | p->rt.timeout = 0; | |
dd41f596 | 4982 | p->sched_class = &fair_sched_class; |
2d3d891d | 4983 | } |
dd41f596 | 4984 | |
b29739f9 IM |
4985 | p->prio = prio; |
4986 | ||
da0c1e65 | 4987 | if (queued) |
ff77e468 | 4988 | enqueue_task(rq, p, queue_flag); |
a399d233 | 4989 | if (running) |
03b7fad1 | 4990 | set_next_task(rq, p); |
cb469845 | 4991 | |
da7a735e | 4992 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 4993 | out_unlock: |
d1ccc66d IM |
4994 | /* Avoid rq from going away on us: */ |
4995 | preempt_disable(); | |
4c9a4bc8 | 4996 | |
565790d2 PZ |
4997 | rq_unpin_lock(rq, &rf); |
4998 | __balance_callbacks(rq); | |
4999 | raw_spin_unlock(&rq->lock); | |
5000 | ||
4c9a4bc8 | 5001 | preempt_enable(); |
b29739f9 | 5002 | } |
acd58620 PZ |
5003 | #else |
5004 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5005 | { | |
5006 | return prio; | |
5007 | } | |
b29739f9 | 5008 | #endif |
d50dde5a | 5009 | |
36c8b586 | 5010 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5011 | { |
49bd21ef | 5012 | bool queued, running; |
53a23364 | 5013 | int old_prio; |
eb580751 | 5014 | struct rq_flags rf; |
70b97a7f | 5015 | struct rq *rq; |
1da177e4 | 5016 | |
75e45d51 | 5017 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
5018 | return; |
5019 | /* | |
5020 | * We have to be careful, if called from sys_setpriority(), | |
5021 | * the task might be in the middle of scheduling on another CPU. | |
5022 | */ | |
eb580751 | 5023 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
5024 | update_rq_clock(rq); |
5025 | ||
1da177e4 LT |
5026 | /* |
5027 | * The RT priorities are set via sched_setscheduler(), but we still | |
5028 | * allow the 'normal' nice value to be set - but as expected | |
5029 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 5030 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 5031 | */ |
aab03e05 | 5032 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
5033 | p->static_prio = NICE_TO_PRIO(nice); |
5034 | goto out_unlock; | |
5035 | } | |
da0c1e65 | 5036 | queued = task_on_rq_queued(p); |
49bd21ef | 5037 | running = task_current(rq, p); |
da0c1e65 | 5038 | if (queued) |
7a57f32a | 5039 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
5040 | if (running) |
5041 | put_prev_task(rq, p); | |
1da177e4 | 5042 | |
1da177e4 | 5043 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 5044 | set_load_weight(p, true); |
b29739f9 IM |
5045 | old_prio = p->prio; |
5046 | p->prio = effective_prio(p); | |
1da177e4 | 5047 | |
5443a0be | 5048 | if (queued) |
7134b3e9 | 5049 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 5050 | if (running) |
03b7fad1 | 5051 | set_next_task(rq, p); |
5443a0be FW |
5052 | |
5053 | /* | |
5054 | * If the task increased its priority or is running and | |
5055 | * lowered its priority, then reschedule its CPU: | |
5056 | */ | |
5057 | p->sched_class->prio_changed(rq, p, old_prio); | |
5058 | ||
1da177e4 | 5059 | out_unlock: |
eb580751 | 5060 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 5061 | } |
1da177e4 LT |
5062 | EXPORT_SYMBOL(set_user_nice); |
5063 | ||
e43379f1 MM |
5064 | /* |
5065 | * can_nice - check if a task can reduce its nice value | |
5066 | * @p: task | |
5067 | * @nice: nice value | |
5068 | */ | |
36c8b586 | 5069 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5070 | { |
d1ccc66d | 5071 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 5072 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 5073 | |
78d7d407 | 5074 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5075 | capable(CAP_SYS_NICE)); |
5076 | } | |
5077 | ||
1da177e4 LT |
5078 | #ifdef __ARCH_WANT_SYS_NICE |
5079 | ||
5080 | /* | |
5081 | * sys_nice - change the priority of the current process. | |
5082 | * @increment: priority increment | |
5083 | * | |
5084 | * sys_setpriority is a more generic, but much slower function that | |
5085 | * does similar things. | |
5086 | */ | |
5add95d4 | 5087 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5088 | { |
48f24c4d | 5089 | long nice, retval; |
1da177e4 LT |
5090 | |
5091 | /* | |
5092 | * Setpriority might change our priority at the same moment. | |
5093 | * We don't have to worry. Conceptually one call occurs first | |
5094 | * and we have a single winner. | |
5095 | */ | |
a9467fa3 | 5096 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 5097 | nice = task_nice(current) + increment; |
1da177e4 | 5098 | |
a9467fa3 | 5099 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
5100 | if (increment < 0 && !can_nice(current, nice)) |
5101 | return -EPERM; | |
5102 | ||
1da177e4 LT |
5103 | retval = security_task_setnice(current, nice); |
5104 | if (retval) | |
5105 | return retval; | |
5106 | ||
5107 | set_user_nice(current, nice); | |
5108 | return 0; | |
5109 | } | |
5110 | ||
5111 | #endif | |
5112 | ||
5113 | /** | |
5114 | * task_prio - return the priority value of a given task. | |
5115 | * @p: the task in question. | |
5116 | * | |
e69f6186 | 5117 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
5118 | * RT tasks are offset by -200. Normal tasks are centered |
5119 | * around 0, value goes from -16 to +15. | |
5120 | */ | |
36c8b586 | 5121 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5122 | { |
5123 | return p->prio - MAX_RT_PRIO; | |
5124 | } | |
5125 | ||
1da177e4 | 5126 | /** |
d1ccc66d | 5127 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 5128 | * @cpu: the processor in question. |
e69f6186 YB |
5129 | * |
5130 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
5131 | */ |
5132 | int idle_cpu(int cpu) | |
5133 | { | |
908a3283 TG |
5134 | struct rq *rq = cpu_rq(cpu); |
5135 | ||
5136 | if (rq->curr != rq->idle) | |
5137 | return 0; | |
5138 | ||
5139 | if (rq->nr_running) | |
5140 | return 0; | |
5141 | ||
5142 | #ifdef CONFIG_SMP | |
126c2092 | 5143 | if (rq->ttwu_pending) |
908a3283 TG |
5144 | return 0; |
5145 | #endif | |
5146 | ||
5147 | return 1; | |
1da177e4 LT |
5148 | } |
5149 | ||
943d355d RJ |
5150 | /** |
5151 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
5152 | * @cpu: the CPU in question. | |
5153 | * | |
5154 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
5155 | */ | |
5156 | int available_idle_cpu(int cpu) | |
5157 | { | |
5158 | if (!idle_cpu(cpu)) | |
5159 | return 0; | |
5160 | ||
247f2f6f RJ |
5161 | if (vcpu_is_preempted(cpu)) |
5162 | return 0; | |
5163 | ||
908a3283 | 5164 | return 1; |
1da177e4 LT |
5165 | } |
5166 | ||
1da177e4 | 5167 | /** |
d1ccc66d | 5168 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 5169 | * @cpu: the processor in question. |
e69f6186 | 5170 | * |
d1ccc66d | 5171 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 5172 | */ |
36c8b586 | 5173 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5174 | { |
5175 | return cpu_rq(cpu)->idle; | |
5176 | } | |
5177 | ||
5178 | /** | |
5179 | * find_process_by_pid - find a process with a matching PID value. | |
5180 | * @pid: the pid in question. | |
e69f6186 YB |
5181 | * |
5182 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 5183 | */ |
a9957449 | 5184 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5185 | { |
228ebcbe | 5186 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5187 | } |
5188 | ||
c13db6b1 SR |
5189 | /* |
5190 | * sched_setparam() passes in -1 for its policy, to let the functions | |
5191 | * it calls know not to change it. | |
5192 | */ | |
5193 | #define SETPARAM_POLICY -1 | |
5194 | ||
c365c292 TG |
5195 | static void __setscheduler_params(struct task_struct *p, |
5196 | const struct sched_attr *attr) | |
1da177e4 | 5197 | { |
d50dde5a DF |
5198 | int policy = attr->sched_policy; |
5199 | ||
c13db6b1 | 5200 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
5201 | policy = p->policy; |
5202 | ||
1da177e4 | 5203 | p->policy = policy; |
d50dde5a | 5204 | |
aab03e05 DF |
5205 | if (dl_policy(policy)) |
5206 | __setparam_dl(p, attr); | |
39fd8fd2 | 5207 | else if (fair_policy(policy)) |
d50dde5a DF |
5208 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
5209 | ||
39fd8fd2 PZ |
5210 | /* |
5211 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
5212 | * !rt_policy. Always setting this ensures that things like | |
5213 | * getparam()/getattr() don't report silly values for !rt tasks. | |
5214 | */ | |
5215 | p->rt_priority = attr->sched_priority; | |
383afd09 | 5216 | p->normal_prio = normal_prio(p); |
9059393e | 5217 | set_load_weight(p, true); |
c365c292 | 5218 | } |
39fd8fd2 | 5219 | |
c365c292 TG |
5220 | /* Actually do priority change: must hold pi & rq lock. */ |
5221 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 5222 | const struct sched_attr *attr, bool keep_boost) |
c365c292 | 5223 | { |
a509a7cd PB |
5224 | /* |
5225 | * If params can't change scheduling class changes aren't allowed | |
5226 | * either. | |
5227 | */ | |
5228 | if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) | |
5229 | return; | |
5230 | ||
c365c292 | 5231 | __setscheduler_params(p, attr); |
d50dde5a | 5232 | |
383afd09 | 5233 | /* |
0782e63b TG |
5234 | * Keep a potential priority boosting if called from |
5235 | * sched_setscheduler(). | |
383afd09 | 5236 | */ |
acd58620 | 5237 | p->prio = normal_prio(p); |
0782e63b | 5238 | if (keep_boost) |
acd58620 | 5239 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 5240 | |
aab03e05 DF |
5241 | if (dl_prio(p->prio)) |
5242 | p->sched_class = &dl_sched_class; | |
5243 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
5244 | p->sched_class = &rt_sched_class; |
5245 | else | |
5246 | p->sched_class = &fair_sched_class; | |
1da177e4 | 5247 | } |
aab03e05 | 5248 | |
c69e8d9c | 5249 | /* |
d1ccc66d | 5250 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
5251 | */ |
5252 | static bool check_same_owner(struct task_struct *p) | |
5253 | { | |
5254 | const struct cred *cred = current_cred(), *pcred; | |
5255 | bool match; | |
5256 | ||
5257 | rcu_read_lock(); | |
5258 | pcred = __task_cred(p); | |
9c806aa0 EB |
5259 | match = (uid_eq(cred->euid, pcred->euid) || |
5260 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
5261 | rcu_read_unlock(); |
5262 | return match; | |
5263 | } | |
5264 | ||
d50dde5a DF |
5265 | static int __sched_setscheduler(struct task_struct *p, |
5266 | const struct sched_attr *attr, | |
dbc7f069 | 5267 | bool user, bool pi) |
1da177e4 | 5268 | { |
383afd09 SR |
5269 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
5270 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 5271 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 5272 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 5273 | const struct sched_class *prev_class; |
565790d2 | 5274 | struct callback_head *head; |
eb580751 | 5275 | struct rq_flags rf; |
ca94c442 | 5276 | int reset_on_fork; |
7a57f32a | 5277 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 5278 | struct rq *rq; |
1da177e4 | 5279 | |
896bbb25 SRV |
5280 | /* The pi code expects interrupts enabled */ |
5281 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 5282 | recheck: |
d1ccc66d | 5283 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
5284 | if (policy < 0) { |
5285 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5286 | policy = oldpolicy = p->policy; |
ca94c442 | 5287 | } else { |
7479f3c9 | 5288 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 5289 | |
20f9cd2a | 5290 | if (!valid_policy(policy)) |
ca94c442 LP |
5291 | return -EINVAL; |
5292 | } | |
5293 | ||
794a56eb | 5294 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
5295 | return -EINVAL; |
5296 | ||
1da177e4 LT |
5297 | /* |
5298 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5299 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5300 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 5301 | */ |
0bb040a4 | 5302 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 5303 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5304 | return -EINVAL; |
aab03e05 DF |
5305 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
5306 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
5307 | return -EINVAL; |
5308 | ||
37e4ab3f OC |
5309 | /* |
5310 | * Allow unprivileged RT tasks to decrease priority: | |
5311 | */ | |
961ccddd | 5312 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 5313 | if (fair_policy(policy)) { |
d0ea0268 | 5314 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 5315 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
5316 | return -EPERM; |
5317 | } | |
5318 | ||
e05606d3 | 5319 | if (rt_policy(policy)) { |
a44702e8 ON |
5320 | unsigned long rlim_rtprio = |
5321 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 5322 | |
d1ccc66d | 5323 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
5324 | if (policy != p->policy && !rlim_rtprio) |
5325 | return -EPERM; | |
5326 | ||
d1ccc66d | 5327 | /* Can't increase priority: */ |
d50dde5a DF |
5328 | if (attr->sched_priority > p->rt_priority && |
5329 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
5330 | return -EPERM; |
5331 | } | |
c02aa73b | 5332 | |
d44753b8 JL |
5333 | /* |
5334 | * Can't set/change SCHED_DEADLINE policy at all for now | |
5335 | * (safest behavior); in the future we would like to allow | |
5336 | * unprivileged DL tasks to increase their relative deadline | |
5337 | * or reduce their runtime (both ways reducing utilization) | |
5338 | */ | |
5339 | if (dl_policy(policy)) | |
5340 | return -EPERM; | |
5341 | ||
dd41f596 | 5342 | /* |
c02aa73b DH |
5343 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5344 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5345 | */ |
1da1843f | 5346 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 5347 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
5348 | return -EPERM; |
5349 | } | |
5fe1d75f | 5350 | |
d1ccc66d | 5351 | /* Can't change other user's priorities: */ |
c69e8d9c | 5352 | if (!check_same_owner(p)) |
37e4ab3f | 5353 | return -EPERM; |
ca94c442 | 5354 | |
d1ccc66d | 5355 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
5356 | if (p->sched_reset_on_fork && !reset_on_fork) |
5357 | return -EPERM; | |
37e4ab3f | 5358 | } |
1da177e4 | 5359 | |
725aad24 | 5360 | if (user) { |
794a56eb JL |
5361 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
5362 | return -EINVAL; | |
5363 | ||
b0ae1981 | 5364 | retval = security_task_setscheduler(p); |
725aad24 JF |
5365 | if (retval) |
5366 | return retval; | |
5367 | } | |
5368 | ||
a509a7cd PB |
5369 | /* Update task specific "requested" clamps */ |
5370 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
5371 | retval = uclamp_validate(p, attr); | |
5372 | if (retval) | |
5373 | return retval; | |
5374 | } | |
5375 | ||
710da3c8 JL |
5376 | if (pi) |
5377 | cpuset_read_lock(); | |
5378 | ||
b29739f9 | 5379 | /* |
d1ccc66d | 5380 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 5381 | * changing the priority of the task: |
0122ec5b | 5382 | * |
25985edc | 5383 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5384 | * runqueue lock must be held. |
5385 | */ | |
eb580751 | 5386 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 5387 | update_rq_clock(rq); |
dc61b1d6 | 5388 | |
34f971f6 | 5389 | /* |
d1ccc66d | 5390 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
5391 | */ |
5392 | if (p == rq->stop) { | |
4b211f2b MP |
5393 | retval = -EINVAL; |
5394 | goto unlock; | |
34f971f6 PZ |
5395 | } |
5396 | ||
a51e9198 | 5397 | /* |
d6b1e911 TG |
5398 | * If not changing anything there's no need to proceed further, |
5399 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 5400 | */ |
d50dde5a | 5401 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 5402 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
5403 | goto change; |
5404 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
5405 | goto change; | |
75381608 | 5406 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 5407 | goto change; |
a509a7cd PB |
5408 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
5409 | goto change; | |
d50dde5a | 5410 | |
d6b1e911 | 5411 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
5412 | retval = 0; |
5413 | goto unlock; | |
a51e9198 | 5414 | } |
d50dde5a | 5415 | change: |
a51e9198 | 5416 | |
dc61b1d6 | 5417 | if (user) { |
332ac17e | 5418 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
5419 | /* |
5420 | * Do not allow realtime tasks into groups that have no runtime | |
5421 | * assigned. | |
5422 | */ | |
5423 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5424 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5425 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
5426 | retval = -EPERM; |
5427 | goto unlock; | |
dc61b1d6 | 5428 | } |
dc61b1d6 | 5429 | #endif |
332ac17e | 5430 | #ifdef CONFIG_SMP |
794a56eb JL |
5431 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
5432 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 5433 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
5434 | |
5435 | /* | |
5436 | * Don't allow tasks with an affinity mask smaller than | |
5437 | * the entire root_domain to become SCHED_DEADLINE. We | |
5438 | * will also fail if there's no bandwidth available. | |
5439 | */ | |
3bd37062 | 5440 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 5441 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
5442 | retval = -EPERM; |
5443 | goto unlock; | |
332ac17e DF |
5444 | } |
5445 | } | |
5446 | #endif | |
5447 | } | |
dc61b1d6 | 5448 | |
d1ccc66d | 5449 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
5450 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
5451 | policy = oldpolicy = -1; | |
eb580751 | 5452 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
5453 | if (pi) |
5454 | cpuset_read_unlock(); | |
1da177e4 LT |
5455 | goto recheck; |
5456 | } | |
332ac17e DF |
5457 | |
5458 | /* | |
5459 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
5460 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
5461 | * is available. | |
5462 | */ | |
06a76fe0 | 5463 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
5464 | retval = -EBUSY; |
5465 | goto unlock; | |
332ac17e DF |
5466 | } |
5467 | ||
c365c292 TG |
5468 | p->sched_reset_on_fork = reset_on_fork; |
5469 | oldprio = p->prio; | |
5470 | ||
dbc7f069 PZ |
5471 | if (pi) { |
5472 | /* | |
5473 | * Take priority boosted tasks into account. If the new | |
5474 | * effective priority is unchanged, we just store the new | |
5475 | * normal parameters and do not touch the scheduler class and | |
5476 | * the runqueue. This will be done when the task deboost | |
5477 | * itself. | |
5478 | */ | |
acd58620 | 5479 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
5480 | if (new_effective_prio == oldprio) |
5481 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
5482 | } |
5483 | ||
da0c1e65 | 5484 | queued = task_on_rq_queued(p); |
051a1d1a | 5485 | running = task_current(rq, p); |
da0c1e65 | 5486 | if (queued) |
ff77e468 | 5487 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 5488 | if (running) |
f3cd1c4e | 5489 | put_prev_task(rq, p); |
f6b53205 | 5490 | |
83ab0aa0 | 5491 | prev_class = p->sched_class; |
a509a7cd | 5492 | |
dbc7f069 | 5493 | __setscheduler(rq, p, attr, pi); |
a509a7cd | 5494 | __setscheduler_uclamp(p, attr); |
f6b53205 | 5495 | |
da0c1e65 | 5496 | if (queued) { |
81a44c54 TG |
5497 | /* |
5498 | * We enqueue to tail when the priority of a task is | |
5499 | * increased (user space view). | |
5500 | */ | |
ff77e468 PZ |
5501 | if (oldprio < p->prio) |
5502 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 5503 | |
ff77e468 | 5504 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 5505 | } |
a399d233 | 5506 | if (running) |
03b7fad1 | 5507 | set_next_task(rq, p); |
cb469845 | 5508 | |
da7a735e | 5509 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
5510 | |
5511 | /* Avoid rq from going away on us: */ | |
5512 | preempt_disable(); | |
565790d2 | 5513 | head = splice_balance_callbacks(rq); |
eb580751 | 5514 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 5515 | |
710da3c8 JL |
5516 | if (pi) { |
5517 | cpuset_read_unlock(); | |
dbc7f069 | 5518 | rt_mutex_adjust_pi(p); |
710da3c8 | 5519 | } |
95e02ca9 | 5520 | |
d1ccc66d | 5521 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 5522 | balance_callbacks(rq, head); |
4c9a4bc8 | 5523 | preempt_enable(); |
95e02ca9 | 5524 | |
1da177e4 | 5525 | return 0; |
4b211f2b MP |
5526 | |
5527 | unlock: | |
5528 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
5529 | if (pi) |
5530 | cpuset_read_unlock(); | |
4b211f2b | 5531 | return retval; |
1da177e4 | 5532 | } |
961ccddd | 5533 | |
7479f3c9 PZ |
5534 | static int _sched_setscheduler(struct task_struct *p, int policy, |
5535 | const struct sched_param *param, bool check) | |
5536 | { | |
5537 | struct sched_attr attr = { | |
5538 | .sched_policy = policy, | |
5539 | .sched_priority = param->sched_priority, | |
5540 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
5541 | }; | |
5542 | ||
c13db6b1 SR |
5543 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
5544 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
5545 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
5546 | policy &= ~SCHED_RESET_ON_FORK; | |
5547 | attr.sched_policy = policy; | |
5548 | } | |
5549 | ||
dbc7f069 | 5550 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 5551 | } |
961ccddd RR |
5552 | /** |
5553 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5554 | * @p: the task in question. | |
5555 | * @policy: new policy. | |
5556 | * @param: structure containing the new RT priority. | |
5557 | * | |
7318d4cc PZ |
5558 | * Use sched_set_fifo(), read its comment. |
5559 | * | |
e69f6186 YB |
5560 | * Return: 0 on success. An error code otherwise. |
5561 | * | |
961ccddd RR |
5562 | * NOTE that the task may be already dead. |
5563 | */ | |
5564 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5565 | const struct sched_param *param) |
961ccddd | 5566 | { |
7479f3c9 | 5567 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 5568 | } |
1da177e4 | 5569 | |
d50dde5a DF |
5570 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
5571 | { | |
dbc7f069 | 5572 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 5573 | } |
d50dde5a | 5574 | |
794a56eb JL |
5575 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
5576 | { | |
5577 | return __sched_setscheduler(p, attr, false, true); | |
5578 | } | |
5579 | ||
961ccddd RR |
5580 | /** |
5581 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5582 | * @p: the task in question. | |
5583 | * @policy: new policy. | |
5584 | * @param: structure containing the new RT priority. | |
5585 | * | |
5586 | * Just like sched_setscheduler, only don't bother checking if the | |
5587 | * current context has permission. For example, this is needed in | |
5588 | * stop_machine(): we create temporary high priority worker threads, | |
5589 | * but our caller might not have that capability. | |
e69f6186 YB |
5590 | * |
5591 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
5592 | */ |
5593 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5594 | const struct sched_param *param) |
961ccddd | 5595 | { |
7479f3c9 | 5596 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
5597 | } |
5598 | ||
7318d4cc PZ |
5599 | /* |
5600 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
5601 | * incapable of resource management, which is the one thing an OS really should | |
5602 | * be doing. | |
5603 | * | |
5604 | * This is of course the reason it is limited to privileged users only. | |
5605 | * | |
5606 | * Worse still; it is fundamentally impossible to compose static priority | |
5607 | * workloads. You cannot take two correctly working static prio workloads | |
5608 | * and smash them together and still expect them to work. | |
5609 | * | |
5610 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
5611 | * | |
5612 | * MAX_RT_PRIO / 2 | |
5613 | * | |
5614 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
5615 | * know enough information to make a sensible choice. | |
5616 | */ | |
8b700983 | 5617 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
5618 | { |
5619 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 5620 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
5621 | } |
5622 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
5623 | ||
5624 | /* | |
5625 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
5626 | */ | |
8b700983 | 5627 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
5628 | { |
5629 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 5630 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
5631 | } |
5632 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
5633 | ||
8b700983 | 5634 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
5635 | { |
5636 | struct sched_attr attr = { | |
5637 | .sched_policy = SCHED_NORMAL, | |
5638 | .sched_nice = nice, | |
5639 | }; | |
8b700983 | 5640 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
5641 | } |
5642 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 5643 | |
95cdf3b7 IM |
5644 | static int |
5645 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5646 | { |
1da177e4 LT |
5647 | struct sched_param lparam; |
5648 | struct task_struct *p; | |
36c8b586 | 5649 | int retval; |
1da177e4 LT |
5650 | |
5651 | if (!param || pid < 0) | |
5652 | return -EINVAL; | |
5653 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5654 | return -EFAULT; | |
5fe1d75f ON |
5655 | |
5656 | rcu_read_lock(); | |
5657 | retval = -ESRCH; | |
1da177e4 | 5658 | p = find_process_by_pid(pid); |
710da3c8 JL |
5659 | if (likely(p)) |
5660 | get_task_struct(p); | |
5fe1d75f | 5661 | rcu_read_unlock(); |
36c8b586 | 5662 | |
710da3c8 JL |
5663 | if (likely(p)) { |
5664 | retval = sched_setscheduler(p, policy, &lparam); | |
5665 | put_task_struct(p); | |
5666 | } | |
5667 | ||
1da177e4 LT |
5668 | return retval; |
5669 | } | |
5670 | ||
d50dde5a DF |
5671 | /* |
5672 | * Mimics kernel/events/core.c perf_copy_attr(). | |
5673 | */ | |
d1ccc66d | 5674 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
5675 | { |
5676 | u32 size; | |
5677 | int ret; | |
5678 | ||
d1ccc66d | 5679 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
5680 | memset(attr, 0, sizeof(*attr)); |
5681 | ||
5682 | ret = get_user(size, &uattr->size); | |
5683 | if (ret) | |
5684 | return ret; | |
5685 | ||
d1ccc66d IM |
5686 | /* ABI compatibility quirk: */ |
5687 | if (!size) | |
d50dde5a | 5688 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 5689 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
5690 | goto err_size; |
5691 | ||
dff3a85f AS |
5692 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
5693 | if (ret) { | |
5694 | if (ret == -E2BIG) | |
5695 | goto err_size; | |
5696 | return ret; | |
d50dde5a DF |
5697 | } |
5698 | ||
a509a7cd PB |
5699 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
5700 | size < SCHED_ATTR_SIZE_VER1) | |
5701 | return -EINVAL; | |
5702 | ||
d50dde5a | 5703 | /* |
d1ccc66d | 5704 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
5705 | * to be strict and return an error on out-of-bounds values? |
5706 | */ | |
75e45d51 | 5707 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 5708 | |
e78c7bca | 5709 | return 0; |
d50dde5a DF |
5710 | |
5711 | err_size: | |
5712 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 5713 | return -E2BIG; |
d50dde5a DF |
5714 | } |
5715 | ||
1da177e4 LT |
5716 | /** |
5717 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5718 | * @pid: the pid in question. | |
5719 | * @policy: new policy. | |
5720 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
5721 | * |
5722 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5723 | */ |
d1ccc66d | 5724 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 5725 | { |
c21761f1 JB |
5726 | if (policy < 0) |
5727 | return -EINVAL; | |
5728 | ||
1da177e4 LT |
5729 | return do_sched_setscheduler(pid, policy, param); |
5730 | } | |
5731 | ||
5732 | /** | |
5733 | * sys_sched_setparam - set/change the RT priority of a thread | |
5734 | * @pid: the pid in question. | |
5735 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
5736 | * |
5737 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5738 | */ |
5add95d4 | 5739 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 5740 | { |
c13db6b1 | 5741 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
5742 | } |
5743 | ||
d50dde5a DF |
5744 | /** |
5745 | * sys_sched_setattr - same as above, but with extended sched_attr | |
5746 | * @pid: the pid in question. | |
5778fccf | 5747 | * @uattr: structure containing the extended parameters. |
db66d756 | 5748 | * @flags: for future extension. |
d50dde5a | 5749 | */ |
6d35ab48 PZ |
5750 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
5751 | unsigned int, flags) | |
d50dde5a DF |
5752 | { |
5753 | struct sched_attr attr; | |
5754 | struct task_struct *p; | |
5755 | int retval; | |
5756 | ||
6d35ab48 | 5757 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
5758 | return -EINVAL; |
5759 | ||
143cf23d MK |
5760 | retval = sched_copy_attr(uattr, &attr); |
5761 | if (retval) | |
5762 | return retval; | |
d50dde5a | 5763 | |
b14ed2c2 | 5764 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 5765 | return -EINVAL; |
1d6362fa PB |
5766 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
5767 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
5768 | |
5769 | rcu_read_lock(); | |
5770 | retval = -ESRCH; | |
5771 | p = find_process_by_pid(pid); | |
a509a7cd PB |
5772 | if (likely(p)) |
5773 | get_task_struct(p); | |
d50dde5a DF |
5774 | rcu_read_unlock(); |
5775 | ||
a509a7cd PB |
5776 | if (likely(p)) { |
5777 | retval = sched_setattr(p, &attr); | |
5778 | put_task_struct(p); | |
5779 | } | |
5780 | ||
d50dde5a DF |
5781 | return retval; |
5782 | } | |
5783 | ||
1da177e4 LT |
5784 | /** |
5785 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5786 | * @pid: the pid in question. | |
e69f6186 YB |
5787 | * |
5788 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
5789 | * code. | |
1da177e4 | 5790 | */ |
5add95d4 | 5791 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5792 | { |
36c8b586 | 5793 | struct task_struct *p; |
3a5c359a | 5794 | int retval; |
1da177e4 LT |
5795 | |
5796 | if (pid < 0) | |
3a5c359a | 5797 | return -EINVAL; |
1da177e4 LT |
5798 | |
5799 | retval = -ESRCH; | |
5fe85be0 | 5800 | rcu_read_lock(); |
1da177e4 LT |
5801 | p = find_process_by_pid(pid); |
5802 | if (p) { | |
5803 | retval = security_task_getscheduler(p); | |
5804 | if (!retval) | |
ca94c442 LP |
5805 | retval = p->policy |
5806 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5807 | } |
5fe85be0 | 5808 | rcu_read_unlock(); |
1da177e4 LT |
5809 | return retval; |
5810 | } | |
5811 | ||
5812 | /** | |
ca94c442 | 5813 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5814 | * @pid: the pid in question. |
5815 | * @param: structure containing the RT priority. | |
e69f6186 YB |
5816 | * |
5817 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
5818 | * code. | |
1da177e4 | 5819 | */ |
5add95d4 | 5820 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 5821 | { |
ce5f7f82 | 5822 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 5823 | struct task_struct *p; |
3a5c359a | 5824 | int retval; |
1da177e4 LT |
5825 | |
5826 | if (!param || pid < 0) | |
3a5c359a | 5827 | return -EINVAL; |
1da177e4 | 5828 | |
5fe85be0 | 5829 | rcu_read_lock(); |
1da177e4 LT |
5830 | p = find_process_by_pid(pid); |
5831 | retval = -ESRCH; | |
5832 | if (!p) | |
5833 | goto out_unlock; | |
5834 | ||
5835 | retval = security_task_getscheduler(p); | |
5836 | if (retval) | |
5837 | goto out_unlock; | |
5838 | ||
ce5f7f82 PZ |
5839 | if (task_has_rt_policy(p)) |
5840 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5841 | rcu_read_unlock(); |
1da177e4 LT |
5842 | |
5843 | /* | |
5844 | * This one might sleep, we cannot do it with a spinlock held ... | |
5845 | */ | |
5846 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5847 | ||
1da177e4 LT |
5848 | return retval; |
5849 | ||
5850 | out_unlock: | |
5fe85be0 | 5851 | rcu_read_unlock(); |
1da177e4 LT |
5852 | return retval; |
5853 | } | |
5854 | ||
1251201c IM |
5855 | /* |
5856 | * Copy the kernel size attribute structure (which might be larger | |
5857 | * than what user-space knows about) to user-space. | |
5858 | * | |
5859 | * Note that all cases are valid: user-space buffer can be larger or | |
5860 | * smaller than the kernel-space buffer. The usual case is that both | |
5861 | * have the same size. | |
5862 | */ | |
5863 | static int | |
5864 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
5865 | struct sched_attr *kattr, | |
5866 | unsigned int usize) | |
d50dde5a | 5867 | { |
1251201c | 5868 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 5869 | |
96d4f267 | 5870 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
5871 | return -EFAULT; |
5872 | ||
5873 | /* | |
1251201c IM |
5874 | * sched_getattr() ABI forwards and backwards compatibility: |
5875 | * | |
5876 | * If usize == ksize then we just copy everything to user-space and all is good. | |
5877 | * | |
5878 | * If usize < ksize then we only copy as much as user-space has space for, | |
5879 | * this keeps ABI compatibility as well. We skip the rest. | |
5880 | * | |
5881 | * If usize > ksize then user-space is using a newer version of the ABI, | |
5882 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
5883 | * detect the kernel's knowledge of attributes from the attr->size value | |
5884 | * which is set to ksize in this case. | |
d50dde5a | 5885 | */ |
1251201c | 5886 | kattr->size = min(usize, ksize); |
d50dde5a | 5887 | |
1251201c | 5888 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
5889 | return -EFAULT; |
5890 | ||
22400674 | 5891 | return 0; |
d50dde5a DF |
5892 | } |
5893 | ||
5894 | /** | |
aab03e05 | 5895 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 5896 | * @pid: the pid in question. |
5778fccf | 5897 | * @uattr: structure containing the extended parameters. |
dff3a85f | 5898 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 5899 | * @flags: for future extension. |
d50dde5a | 5900 | */ |
6d35ab48 | 5901 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 5902 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 5903 | { |
1251201c | 5904 | struct sched_attr kattr = { }; |
d50dde5a DF |
5905 | struct task_struct *p; |
5906 | int retval; | |
5907 | ||
1251201c IM |
5908 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
5909 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
5910 | return -EINVAL; |
5911 | ||
5912 | rcu_read_lock(); | |
5913 | p = find_process_by_pid(pid); | |
5914 | retval = -ESRCH; | |
5915 | if (!p) | |
5916 | goto out_unlock; | |
5917 | ||
5918 | retval = security_task_getscheduler(p); | |
5919 | if (retval) | |
5920 | goto out_unlock; | |
5921 | ||
1251201c | 5922 | kattr.sched_policy = p->policy; |
7479f3c9 | 5923 | if (p->sched_reset_on_fork) |
1251201c | 5924 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
aab03e05 | 5925 | if (task_has_dl_policy(p)) |
1251201c | 5926 | __getparam_dl(p, &kattr); |
aab03e05 | 5927 | else if (task_has_rt_policy(p)) |
1251201c | 5928 | kattr.sched_priority = p->rt_priority; |
d50dde5a | 5929 | else |
1251201c | 5930 | kattr.sched_nice = task_nice(p); |
d50dde5a | 5931 | |
a509a7cd | 5932 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
5933 | /* |
5934 | * This could race with another potential updater, but this is fine | |
5935 | * because it'll correctly read the old or the new value. We don't need | |
5936 | * to guarantee who wins the race as long as it doesn't return garbage. | |
5937 | */ | |
1251201c IM |
5938 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
5939 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
5940 | #endif |
5941 | ||
d50dde5a DF |
5942 | rcu_read_unlock(); |
5943 | ||
1251201c | 5944 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
5945 | |
5946 | out_unlock: | |
5947 | rcu_read_unlock(); | |
5948 | return retval; | |
5949 | } | |
5950 | ||
96f874e2 | 5951 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5952 | { |
5a16f3d3 | 5953 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5954 | struct task_struct *p; |
5955 | int retval; | |
1da177e4 | 5956 | |
23f5d142 | 5957 | rcu_read_lock(); |
1da177e4 LT |
5958 | |
5959 | p = find_process_by_pid(pid); | |
5960 | if (!p) { | |
23f5d142 | 5961 | rcu_read_unlock(); |
1da177e4 LT |
5962 | return -ESRCH; |
5963 | } | |
5964 | ||
23f5d142 | 5965 | /* Prevent p going away */ |
1da177e4 | 5966 | get_task_struct(p); |
23f5d142 | 5967 | rcu_read_unlock(); |
1da177e4 | 5968 | |
14a40ffc TH |
5969 | if (p->flags & PF_NO_SETAFFINITY) { |
5970 | retval = -EINVAL; | |
5971 | goto out_put_task; | |
5972 | } | |
5a16f3d3 RR |
5973 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5974 | retval = -ENOMEM; | |
5975 | goto out_put_task; | |
5976 | } | |
5977 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5978 | retval = -ENOMEM; | |
5979 | goto out_free_cpus_allowed; | |
5980 | } | |
1da177e4 | 5981 | retval = -EPERM; |
4c44aaaf EB |
5982 | if (!check_same_owner(p)) { |
5983 | rcu_read_lock(); | |
5984 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
5985 | rcu_read_unlock(); | |
16303ab2 | 5986 | goto out_free_new_mask; |
4c44aaaf EB |
5987 | } |
5988 | rcu_read_unlock(); | |
5989 | } | |
1da177e4 | 5990 | |
b0ae1981 | 5991 | retval = security_task_setscheduler(p); |
e7834f8f | 5992 | if (retval) |
16303ab2 | 5993 | goto out_free_new_mask; |
e7834f8f | 5994 | |
e4099a5e PZ |
5995 | |
5996 | cpuset_cpus_allowed(p, cpus_allowed); | |
5997 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
5998 | ||
332ac17e DF |
5999 | /* |
6000 | * Since bandwidth control happens on root_domain basis, | |
6001 | * if admission test is enabled, we only admit -deadline | |
6002 | * tasks allowed to run on all the CPUs in the task's | |
6003 | * root_domain. | |
6004 | */ | |
6005 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
6006 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
6007 | rcu_read_lock(); | |
6008 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 6009 | retval = -EBUSY; |
f1e3a093 | 6010 | rcu_read_unlock(); |
16303ab2 | 6011 | goto out_free_new_mask; |
332ac17e | 6012 | } |
f1e3a093 | 6013 | rcu_read_unlock(); |
332ac17e DF |
6014 | } |
6015 | #endif | |
49246274 | 6016 | again: |
9cfc3e18 | 6017 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK); |
1da177e4 | 6018 | |
8707d8b8 | 6019 | if (!retval) { |
5a16f3d3 RR |
6020 | cpuset_cpus_allowed(p, cpus_allowed); |
6021 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6022 | /* |
6023 | * We must have raced with a concurrent cpuset | |
6024 | * update. Just reset the cpus_allowed to the | |
6025 | * cpuset's cpus_allowed | |
6026 | */ | |
5a16f3d3 | 6027 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6028 | goto again; |
6029 | } | |
6030 | } | |
16303ab2 | 6031 | out_free_new_mask: |
5a16f3d3 RR |
6032 | free_cpumask_var(new_mask); |
6033 | out_free_cpus_allowed: | |
6034 | free_cpumask_var(cpus_allowed); | |
6035 | out_put_task: | |
1da177e4 | 6036 | put_task_struct(p); |
1da177e4 LT |
6037 | return retval; |
6038 | } | |
6039 | ||
6040 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6041 | struct cpumask *new_mask) |
1da177e4 | 6042 | { |
96f874e2 RR |
6043 | if (len < cpumask_size()) |
6044 | cpumask_clear(new_mask); | |
6045 | else if (len > cpumask_size()) | |
6046 | len = cpumask_size(); | |
6047 | ||
1da177e4 LT |
6048 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6049 | } | |
6050 | ||
6051 | /** | |
d1ccc66d | 6052 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
6053 | * @pid: pid of the process |
6054 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6055 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
6056 | * |
6057 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6058 | */ |
5add95d4 HC |
6059 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6060 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6061 | { |
5a16f3d3 | 6062 | cpumask_var_t new_mask; |
1da177e4 LT |
6063 | int retval; |
6064 | ||
5a16f3d3 RR |
6065 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6066 | return -ENOMEM; | |
1da177e4 | 6067 | |
5a16f3d3 RR |
6068 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6069 | if (retval == 0) | |
6070 | retval = sched_setaffinity(pid, new_mask); | |
6071 | free_cpumask_var(new_mask); | |
6072 | return retval; | |
1da177e4 LT |
6073 | } |
6074 | ||
96f874e2 | 6075 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6076 | { |
36c8b586 | 6077 | struct task_struct *p; |
31605683 | 6078 | unsigned long flags; |
1da177e4 | 6079 | int retval; |
1da177e4 | 6080 | |
23f5d142 | 6081 | rcu_read_lock(); |
1da177e4 LT |
6082 | |
6083 | retval = -ESRCH; | |
6084 | p = find_process_by_pid(pid); | |
6085 | if (!p) | |
6086 | goto out_unlock; | |
6087 | ||
e7834f8f DQ |
6088 | retval = security_task_getscheduler(p); |
6089 | if (retval) | |
6090 | goto out_unlock; | |
6091 | ||
013fdb80 | 6092 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 6093 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 6094 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
6095 | |
6096 | out_unlock: | |
23f5d142 | 6097 | rcu_read_unlock(); |
1da177e4 | 6098 | |
9531b62f | 6099 | return retval; |
1da177e4 LT |
6100 | } |
6101 | ||
6102 | /** | |
d1ccc66d | 6103 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
6104 | * @pid: pid of the process |
6105 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6106 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 6107 | * |
599b4840 ZW |
6108 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
6109 | * error code otherwise. | |
1da177e4 | 6110 | */ |
5add95d4 HC |
6111 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6112 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6113 | { |
6114 | int ret; | |
f17c8607 | 6115 | cpumask_var_t mask; |
1da177e4 | 6116 | |
84fba5ec | 6117 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
6118 | return -EINVAL; |
6119 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
6120 | return -EINVAL; |
6121 | ||
f17c8607 RR |
6122 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6123 | return -ENOMEM; | |
1da177e4 | 6124 | |
f17c8607 RR |
6125 | ret = sched_getaffinity(pid, mask); |
6126 | if (ret == 0) { | |
4de373a1 | 6127 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
6128 | |
6129 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
6130 | ret = -EFAULT; |
6131 | else | |
cd3d8031 | 6132 | ret = retlen; |
f17c8607 RR |
6133 | } |
6134 | free_cpumask_var(mask); | |
1da177e4 | 6135 | |
f17c8607 | 6136 | return ret; |
1da177e4 LT |
6137 | } |
6138 | ||
6139 | /** | |
6140 | * sys_sched_yield - yield the current processor to other threads. | |
6141 | * | |
dd41f596 IM |
6142 | * This function yields the current CPU to other tasks. If there are no |
6143 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
6144 | * |
6145 | * Return: 0. | |
1da177e4 | 6146 | */ |
7d4dd4f1 | 6147 | static void do_sched_yield(void) |
1da177e4 | 6148 | { |
8a8c69c3 PZ |
6149 | struct rq_flags rf; |
6150 | struct rq *rq; | |
6151 | ||
246b3b33 | 6152 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 6153 | |
ae92882e | 6154 | schedstat_inc(rq->yld_count); |
4530d7ab | 6155 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6156 | |
6157 | /* | |
6158 | * Since we are going to call schedule() anyway, there's | |
6159 | * no need to preempt or enable interrupts: | |
6160 | */ | |
8a8c69c3 PZ |
6161 | preempt_disable(); |
6162 | rq_unlock(rq, &rf); | |
ba74c144 | 6163 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
6164 | |
6165 | schedule(); | |
7d4dd4f1 | 6166 | } |
1da177e4 | 6167 | |
7d4dd4f1 DB |
6168 | SYSCALL_DEFINE0(sched_yield) |
6169 | { | |
6170 | do_sched_yield(); | |
1da177e4 LT |
6171 | return 0; |
6172 | } | |
6173 | ||
c1a280b6 | 6174 | #ifndef CONFIG_PREEMPTION |
02b67cc3 | 6175 | int __sched _cond_resched(void) |
1da177e4 | 6176 | { |
fe32d3cd | 6177 | if (should_resched(0)) { |
a18b5d01 | 6178 | preempt_schedule_common(); |
1da177e4 LT |
6179 | return 1; |
6180 | } | |
f79c3ad6 | 6181 | rcu_all_qs(); |
1da177e4 LT |
6182 | return 0; |
6183 | } | |
02b67cc3 | 6184 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 6185 | #endif |
1da177e4 LT |
6186 | |
6187 | /* | |
613afbf8 | 6188 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6189 | * call schedule, and on return reacquire the lock. |
6190 | * | |
c1a280b6 | 6191 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
6192 | * operations here to prevent schedule() from being called twice (once via |
6193 | * spin_unlock(), once by hand). | |
6194 | */ | |
613afbf8 | 6195 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6196 | { |
fe32d3cd | 6197 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
6198 | int ret = 0; |
6199 | ||
f607c668 PZ |
6200 | lockdep_assert_held(lock); |
6201 | ||
4a81e832 | 6202 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6203 | spin_unlock(lock); |
d86ee480 | 6204 | if (resched) |
a18b5d01 | 6205 | preempt_schedule_common(); |
95c354fe NP |
6206 | else |
6207 | cpu_relax(); | |
6df3cecb | 6208 | ret = 1; |
1da177e4 | 6209 | spin_lock(lock); |
1da177e4 | 6210 | } |
6df3cecb | 6211 | return ret; |
1da177e4 | 6212 | } |
613afbf8 | 6213 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6214 | |
1da177e4 LT |
6215 | /** |
6216 | * yield - yield the current processor to other threads. | |
6217 | * | |
8e3fabfd PZ |
6218 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
6219 | * | |
6220 | * The scheduler is at all times free to pick the calling task as the most | |
6221 | * eligible task to run, if removing the yield() call from your code breaks | |
6222 | * it, its already broken. | |
6223 | * | |
6224 | * Typical broken usage is: | |
6225 | * | |
6226 | * while (!event) | |
d1ccc66d | 6227 | * yield(); |
8e3fabfd PZ |
6228 | * |
6229 | * where one assumes that yield() will let 'the other' process run that will | |
6230 | * make event true. If the current task is a SCHED_FIFO task that will never | |
6231 | * happen. Never use yield() as a progress guarantee!! | |
6232 | * | |
6233 | * If you want to use yield() to wait for something, use wait_event(). | |
6234 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
6235 | * If you still want to use yield(), do not! | |
1da177e4 LT |
6236 | */ |
6237 | void __sched yield(void) | |
6238 | { | |
6239 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 6240 | do_sched_yield(); |
1da177e4 | 6241 | } |
1da177e4 LT |
6242 | EXPORT_SYMBOL(yield); |
6243 | ||
d95f4122 MG |
6244 | /** |
6245 | * yield_to - yield the current processor to another thread in | |
6246 | * your thread group, or accelerate that thread toward the | |
6247 | * processor it's on. | |
16addf95 RD |
6248 | * @p: target task |
6249 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
6250 | * |
6251 | * It's the caller's job to ensure that the target task struct | |
6252 | * can't go away on us before we can do any checks. | |
6253 | * | |
e69f6186 | 6254 | * Return: |
7b270f60 PZ |
6255 | * true (>0) if we indeed boosted the target task. |
6256 | * false (0) if we failed to boost the target. | |
6257 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 6258 | */ |
fa93384f | 6259 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
6260 | { |
6261 | struct task_struct *curr = current; | |
6262 | struct rq *rq, *p_rq; | |
6263 | unsigned long flags; | |
c3c18640 | 6264 | int yielded = 0; |
d95f4122 MG |
6265 | |
6266 | local_irq_save(flags); | |
6267 | rq = this_rq(); | |
6268 | ||
6269 | again: | |
6270 | p_rq = task_rq(p); | |
7b270f60 PZ |
6271 | /* |
6272 | * If we're the only runnable task on the rq and target rq also | |
6273 | * has only one task, there's absolutely no point in yielding. | |
6274 | */ | |
6275 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
6276 | yielded = -ESRCH; | |
6277 | goto out_irq; | |
6278 | } | |
6279 | ||
d95f4122 | 6280 | double_rq_lock(rq, p_rq); |
39e24d8f | 6281 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
6282 | double_rq_unlock(rq, p_rq); |
6283 | goto again; | |
6284 | } | |
6285 | ||
6286 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 6287 | goto out_unlock; |
d95f4122 MG |
6288 | |
6289 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 6290 | goto out_unlock; |
d95f4122 MG |
6291 | |
6292 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 6293 | goto out_unlock; |
d95f4122 | 6294 | |
0900acf2 | 6295 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 6296 | if (yielded) { |
ae92882e | 6297 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
6298 | /* |
6299 | * Make p's CPU reschedule; pick_next_entity takes care of | |
6300 | * fairness. | |
6301 | */ | |
6302 | if (preempt && rq != p_rq) | |
8875125e | 6303 | resched_curr(p_rq); |
6d1cafd8 | 6304 | } |
d95f4122 | 6305 | |
7b270f60 | 6306 | out_unlock: |
d95f4122 | 6307 | double_rq_unlock(rq, p_rq); |
7b270f60 | 6308 | out_irq: |
d95f4122 MG |
6309 | local_irq_restore(flags); |
6310 | ||
7b270f60 | 6311 | if (yielded > 0) |
d95f4122 MG |
6312 | schedule(); |
6313 | ||
6314 | return yielded; | |
6315 | } | |
6316 | EXPORT_SYMBOL_GPL(yield_to); | |
6317 | ||
10ab5643 TH |
6318 | int io_schedule_prepare(void) |
6319 | { | |
6320 | int old_iowait = current->in_iowait; | |
6321 | ||
6322 | current->in_iowait = 1; | |
6323 | blk_schedule_flush_plug(current); | |
6324 | ||
6325 | return old_iowait; | |
6326 | } | |
6327 | ||
6328 | void io_schedule_finish(int token) | |
6329 | { | |
6330 | current->in_iowait = token; | |
6331 | } | |
6332 | ||
1da177e4 | 6333 | /* |
41a2d6cf | 6334 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6335 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 6336 | */ |
1da177e4 LT |
6337 | long __sched io_schedule_timeout(long timeout) |
6338 | { | |
10ab5643 | 6339 | int token; |
1da177e4 LT |
6340 | long ret; |
6341 | ||
10ab5643 | 6342 | token = io_schedule_prepare(); |
1da177e4 | 6343 | ret = schedule_timeout(timeout); |
10ab5643 | 6344 | io_schedule_finish(token); |
9cff8ade | 6345 | |
1da177e4 LT |
6346 | return ret; |
6347 | } | |
9cff8ade | 6348 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 6349 | |
e3b929b0 | 6350 | void __sched io_schedule(void) |
10ab5643 TH |
6351 | { |
6352 | int token; | |
6353 | ||
6354 | token = io_schedule_prepare(); | |
6355 | schedule(); | |
6356 | io_schedule_finish(token); | |
6357 | } | |
6358 | EXPORT_SYMBOL(io_schedule); | |
6359 | ||
1da177e4 LT |
6360 | /** |
6361 | * sys_sched_get_priority_max - return maximum RT priority. | |
6362 | * @policy: scheduling class. | |
6363 | * | |
e69f6186 YB |
6364 | * Return: On success, this syscall returns the maximum |
6365 | * rt_priority that can be used by a given scheduling class. | |
6366 | * On failure, a negative error code is returned. | |
1da177e4 | 6367 | */ |
5add95d4 | 6368 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6369 | { |
6370 | int ret = -EINVAL; | |
6371 | ||
6372 | switch (policy) { | |
6373 | case SCHED_FIFO: | |
6374 | case SCHED_RR: | |
6375 | ret = MAX_USER_RT_PRIO-1; | |
6376 | break; | |
aab03e05 | 6377 | case SCHED_DEADLINE: |
1da177e4 | 6378 | case SCHED_NORMAL: |
b0a9499c | 6379 | case SCHED_BATCH: |
dd41f596 | 6380 | case SCHED_IDLE: |
1da177e4 LT |
6381 | ret = 0; |
6382 | break; | |
6383 | } | |
6384 | return ret; | |
6385 | } | |
6386 | ||
6387 | /** | |
6388 | * sys_sched_get_priority_min - return minimum RT priority. | |
6389 | * @policy: scheduling class. | |
6390 | * | |
e69f6186 YB |
6391 | * Return: On success, this syscall returns the minimum |
6392 | * rt_priority that can be used by a given scheduling class. | |
6393 | * On failure, a negative error code is returned. | |
1da177e4 | 6394 | */ |
5add95d4 | 6395 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6396 | { |
6397 | int ret = -EINVAL; | |
6398 | ||
6399 | switch (policy) { | |
6400 | case SCHED_FIFO: | |
6401 | case SCHED_RR: | |
6402 | ret = 1; | |
6403 | break; | |
aab03e05 | 6404 | case SCHED_DEADLINE: |
1da177e4 | 6405 | case SCHED_NORMAL: |
b0a9499c | 6406 | case SCHED_BATCH: |
dd41f596 | 6407 | case SCHED_IDLE: |
1da177e4 LT |
6408 | ret = 0; |
6409 | } | |
6410 | return ret; | |
6411 | } | |
6412 | ||
abca5fc5 | 6413 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 6414 | { |
36c8b586 | 6415 | struct task_struct *p; |
a4ec24b4 | 6416 | unsigned int time_slice; |
eb580751 | 6417 | struct rq_flags rf; |
dba091b9 | 6418 | struct rq *rq; |
3a5c359a | 6419 | int retval; |
1da177e4 LT |
6420 | |
6421 | if (pid < 0) | |
3a5c359a | 6422 | return -EINVAL; |
1da177e4 LT |
6423 | |
6424 | retval = -ESRCH; | |
1a551ae7 | 6425 | rcu_read_lock(); |
1da177e4 LT |
6426 | p = find_process_by_pid(pid); |
6427 | if (!p) | |
6428 | goto out_unlock; | |
6429 | ||
6430 | retval = security_task_getscheduler(p); | |
6431 | if (retval) | |
6432 | goto out_unlock; | |
6433 | ||
eb580751 | 6434 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
6435 | time_slice = 0; |
6436 | if (p->sched_class->get_rr_interval) | |
6437 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 6438 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 6439 | |
1a551ae7 | 6440 | rcu_read_unlock(); |
abca5fc5 AV |
6441 | jiffies_to_timespec64(time_slice, t); |
6442 | return 0; | |
3a5c359a | 6443 | |
1da177e4 | 6444 | out_unlock: |
1a551ae7 | 6445 | rcu_read_unlock(); |
1da177e4 LT |
6446 | return retval; |
6447 | } | |
6448 | ||
2064a5ab RD |
6449 | /** |
6450 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6451 | * @pid: pid of the process. | |
6452 | * @interval: userspace pointer to the timeslice value. | |
6453 | * | |
6454 | * this syscall writes the default timeslice value of a given process | |
6455 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6456 | * | |
6457 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
6458 | * an error code. | |
6459 | */ | |
abca5fc5 | 6460 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 6461 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
6462 | { |
6463 | struct timespec64 t; | |
6464 | int retval = sched_rr_get_interval(pid, &t); | |
6465 | ||
6466 | if (retval == 0) | |
6467 | retval = put_timespec64(&t, interval); | |
6468 | ||
6469 | return retval; | |
6470 | } | |
6471 | ||
474b9c77 | 6472 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
6473 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
6474 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
6475 | { |
6476 | struct timespec64 t; | |
6477 | int retval = sched_rr_get_interval(pid, &t); | |
6478 | ||
6479 | if (retval == 0) | |
9afc5eee | 6480 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
6481 | return retval; |
6482 | } | |
6483 | #endif | |
6484 | ||
82a1fcb9 | 6485 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6486 | { |
1da177e4 | 6487 | unsigned long free = 0; |
4e79752c | 6488 | int ppid; |
c930b2c0 | 6489 | |
38200502 TH |
6490 | if (!try_get_task_stack(p)) |
6491 | return; | |
20435d84 | 6492 | |
cc172ff3 | 6493 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 XX |
6494 | |
6495 | if (p->state == TASK_RUNNING) | |
cc172ff3 | 6496 | pr_cont(" running task "); |
1da177e4 | 6497 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 6498 | free = stack_not_used(p); |
1da177e4 | 6499 | #endif |
a90e984c | 6500 | ppid = 0; |
4e79752c | 6501 | rcu_read_lock(); |
a90e984c ON |
6502 | if (pid_alive(p)) |
6503 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 6504 | rcu_read_unlock(); |
cc172ff3 LZ |
6505 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
6506 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 6507 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 6508 | |
3d1cb205 | 6509 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 6510 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 6511 | show_stack(p, NULL, KERN_INFO); |
38200502 | 6512 | put_task_stack(p); |
1da177e4 | 6513 | } |
0032f4e8 | 6514 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 6515 | |
5d68cc95 PZ |
6516 | static inline bool |
6517 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
6518 | { | |
6519 | /* no filter, everything matches */ | |
6520 | if (!state_filter) | |
6521 | return true; | |
6522 | ||
6523 | /* filter, but doesn't match */ | |
6524 | if (!(p->state & state_filter)) | |
6525 | return false; | |
6526 | ||
6527 | /* | |
6528 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
6529 | * TASK_KILLABLE). | |
6530 | */ | |
6531 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
6532 | return false; | |
6533 | ||
6534 | return true; | |
6535 | } | |
6536 | ||
6537 | ||
e59e2ae2 | 6538 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6539 | { |
36c8b586 | 6540 | struct task_struct *g, *p; |
1da177e4 | 6541 | |
510f5acc | 6542 | rcu_read_lock(); |
5d07f420 | 6543 | for_each_process_thread(g, p) { |
1da177e4 LT |
6544 | /* |
6545 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 6546 | * console might take a lot of time: |
57675cb9 AR |
6547 | * Also, reset softlockup watchdogs on all CPUs, because |
6548 | * another CPU might be blocked waiting for us to process | |
6549 | * an IPI. | |
1da177e4 LT |
6550 | */ |
6551 | touch_nmi_watchdog(); | |
57675cb9 | 6552 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 6553 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 6554 | sched_show_task(p); |
5d07f420 | 6555 | } |
1da177e4 | 6556 | |
dd41f596 | 6557 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
6558 | if (!state_filter) |
6559 | sysrq_sched_debug_show(); | |
dd41f596 | 6560 | #endif |
510f5acc | 6561 | rcu_read_unlock(); |
e59e2ae2 IM |
6562 | /* |
6563 | * Only show locks if all tasks are dumped: | |
6564 | */ | |
93335a21 | 6565 | if (!state_filter) |
e59e2ae2 | 6566 | debug_show_all_locks(); |
1da177e4 LT |
6567 | } |
6568 | ||
f340c0d1 IM |
6569 | /** |
6570 | * init_idle - set up an idle thread for a given CPU | |
6571 | * @idle: task in question | |
d1ccc66d | 6572 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
6573 | * |
6574 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6575 | * flag, to make booting more robust. | |
6576 | */ | |
0db0628d | 6577 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6578 | { |
70b97a7f | 6579 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6580 | unsigned long flags; |
6581 | ||
ff51ff84 PZ |
6582 | __sched_fork(0, idle); |
6583 | ||
25834c73 PZ |
6584 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
6585 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 6586 | |
06b83b5f | 6587 | idle->state = TASK_RUNNING; |
dd41f596 | 6588 | idle->se.exec_start = sched_clock(); |
c1de45ca | 6589 | idle->flags |= PF_IDLE; |
dd41f596 | 6590 | |
d08b9f0c | 6591 | scs_task_reset(idle); |
e1b77c92 MR |
6592 | kasan_unpoison_task_stack(idle); |
6593 | ||
de9b8f5d PZ |
6594 | #ifdef CONFIG_SMP |
6595 | /* | |
6596 | * Its possible that init_idle() gets called multiple times on a task, | |
6597 | * in that case do_set_cpus_allowed() will not do the right thing. | |
6598 | * | |
6599 | * And since this is boot we can forgo the serialization. | |
6600 | */ | |
9cfc3e18 | 6601 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 6602 | #endif |
6506cf6c PZ |
6603 | /* |
6604 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 6605 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
6606 | * lockdep check in task_group() will fail. |
6607 | * | |
6608 | * Similar case to sched_fork(). / Alternatively we could | |
6609 | * use task_rq_lock() here and obtain the other rq->lock. | |
6610 | * | |
6611 | * Silence PROVE_RCU | |
6612 | */ | |
6613 | rcu_read_lock(); | |
dd41f596 | 6614 | __set_task_cpu(idle, cpu); |
6506cf6c | 6615 | rcu_read_unlock(); |
1da177e4 | 6616 | |
5311a98f EB |
6617 | rq->idle = idle; |
6618 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 6619 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 6620 | #ifdef CONFIG_SMP |
3ca7a440 | 6621 | idle->on_cpu = 1; |
4866cde0 | 6622 | #endif |
25834c73 PZ |
6623 | raw_spin_unlock(&rq->lock); |
6624 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
6625 | |
6626 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 6627 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 6628 | |
dd41f596 IM |
6629 | /* |
6630 | * The idle tasks have their own, simple scheduling class: | |
6631 | */ | |
6632 | idle->sched_class = &idle_sched_class; | |
868baf07 | 6633 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 6634 | vtime_init_idle(idle, cpu); |
de9b8f5d | 6635 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
6636 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
6637 | #endif | |
19978ca6 IM |
6638 | } |
6639 | ||
e1d4eeec NP |
6640 | #ifdef CONFIG_SMP |
6641 | ||
f82f8042 JL |
6642 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
6643 | const struct cpumask *trial) | |
6644 | { | |
06a76fe0 | 6645 | int ret = 1; |
f82f8042 | 6646 | |
bb2bc55a MG |
6647 | if (!cpumask_weight(cur)) |
6648 | return ret; | |
6649 | ||
06a76fe0 | 6650 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
6651 | |
6652 | return ret; | |
6653 | } | |
6654 | ||
7f51412a JL |
6655 | int task_can_attach(struct task_struct *p, |
6656 | const struct cpumask *cs_cpus_allowed) | |
6657 | { | |
6658 | int ret = 0; | |
6659 | ||
6660 | /* | |
6661 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 6662 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
6663 | * affinity and isolating such threads by their set of |
6664 | * allowed nodes is unnecessary. Thus, cpusets are not | |
6665 | * applicable for such threads. This prevents checking for | |
6666 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 6667 | * before cpus_mask may be changed. |
7f51412a JL |
6668 | */ |
6669 | if (p->flags & PF_NO_SETAFFINITY) { | |
6670 | ret = -EINVAL; | |
6671 | goto out; | |
6672 | } | |
6673 | ||
7f51412a | 6674 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
6675 | cs_cpus_allowed)) |
6676 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 6677 | |
7f51412a JL |
6678 | out: |
6679 | return ret; | |
6680 | } | |
6681 | ||
f2cb1360 | 6682 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 6683 | |
e6628d5b MG |
6684 | #ifdef CONFIG_NUMA_BALANCING |
6685 | /* Migrate current task p to target_cpu */ | |
6686 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
6687 | { | |
6688 | struct migration_arg arg = { p, target_cpu }; | |
6689 | int curr_cpu = task_cpu(p); | |
6690 | ||
6691 | if (curr_cpu == target_cpu) | |
6692 | return 0; | |
6693 | ||
3bd37062 | 6694 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
6695 | return -EINVAL; |
6696 | ||
6697 | /* TODO: This is not properly updating schedstats */ | |
6698 | ||
286549dc | 6699 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
6700 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
6701 | } | |
0ec8aa00 PZ |
6702 | |
6703 | /* | |
6704 | * Requeue a task on a given node and accurately track the number of NUMA | |
6705 | * tasks on the runqueues | |
6706 | */ | |
6707 | void sched_setnuma(struct task_struct *p, int nid) | |
6708 | { | |
da0c1e65 | 6709 | bool queued, running; |
eb580751 PZ |
6710 | struct rq_flags rf; |
6711 | struct rq *rq; | |
0ec8aa00 | 6712 | |
eb580751 | 6713 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 6714 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
6715 | running = task_current(rq, p); |
6716 | ||
da0c1e65 | 6717 | if (queued) |
1de64443 | 6718 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 6719 | if (running) |
f3cd1c4e | 6720 | put_prev_task(rq, p); |
0ec8aa00 PZ |
6721 | |
6722 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 6723 | |
da0c1e65 | 6724 | if (queued) |
7134b3e9 | 6725 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 6726 | if (running) |
03b7fad1 | 6727 | set_next_task(rq, p); |
eb580751 | 6728 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 6729 | } |
5cc389bc | 6730 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 6731 | |
1da177e4 | 6732 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 6733 | /* |
d1ccc66d | 6734 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 6735 | * offline. |
054b9108 | 6736 | */ |
48c5ccae | 6737 | void idle_task_exit(void) |
1da177e4 | 6738 | { |
48c5ccae | 6739 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6740 | |
48c5ccae | 6741 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 6742 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 6743 | |
a53efe5f | 6744 | if (mm != &init_mm) { |
252d2a41 | 6745 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
6746 | finish_arch_post_lock_switch(); |
6747 | } | |
bf2c59fc PZ |
6748 | |
6749 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
6750 | } |
6751 | ||
2558aacf PZ |
6752 | static int __balance_push_cpu_stop(void *arg) |
6753 | { | |
6754 | struct task_struct *p = arg; | |
6755 | struct rq *rq = this_rq(); | |
6756 | struct rq_flags rf; | |
6757 | int cpu; | |
6758 | ||
6759 | raw_spin_lock_irq(&p->pi_lock); | |
6760 | rq_lock(rq, &rf); | |
6761 | ||
6762 | update_rq_clock(rq); | |
6763 | ||
6764 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
6765 | cpu = select_fallback_rq(rq->cpu, p); | |
6766 | rq = __migrate_task(rq, &rf, p, cpu); | |
6767 | } | |
6768 | ||
6769 | rq_unlock(rq, &rf); | |
6770 | raw_spin_unlock_irq(&p->pi_lock); | |
6771 | ||
6772 | put_task_struct(p); | |
6773 | ||
6774 | return 0; | |
6775 | } | |
6776 | ||
6777 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); | |
6778 | ||
6779 | /* | |
6780 | * Ensure we only run per-cpu kthreads once the CPU goes !active. | |
6781 | */ | |
6782 | static void balance_push(struct rq *rq) | |
6783 | { | |
6784 | struct task_struct *push_task = rq->curr; | |
6785 | ||
6786 | lockdep_assert_held(&rq->lock); | |
6787 | SCHED_WARN_ON(rq->cpu != smp_processor_id()); | |
6788 | ||
6789 | /* | |
6790 | * Both the cpu-hotplug and stop task are in this case and are | |
6791 | * required to complete the hotplug process. | |
6792 | */ | |
f2469a1f TG |
6793 | if (is_per_cpu_kthread(push_task)) { |
6794 | /* | |
6795 | * If this is the idle task on the outgoing CPU try to wake | |
6796 | * up the hotplug control thread which might wait for the | |
6797 | * last task to vanish. The rcuwait_active() check is | |
6798 | * accurate here because the waiter is pinned on this CPU | |
6799 | * and can't obviously be running in parallel. | |
6800 | */ | |
6801 | if (!rq->nr_running && rcuwait_active(&rq->hotplug_wait)) { | |
6802 | raw_spin_unlock(&rq->lock); | |
6803 | rcuwait_wake_up(&rq->hotplug_wait); | |
6804 | raw_spin_lock(&rq->lock); | |
6805 | } | |
2558aacf | 6806 | return; |
f2469a1f | 6807 | } |
2558aacf PZ |
6808 | |
6809 | get_task_struct(push_task); | |
6810 | /* | |
6811 | * Temporarily drop rq->lock such that we can wake-up the stop task. | |
6812 | * Both preemption and IRQs are still disabled. | |
6813 | */ | |
6814 | raw_spin_unlock(&rq->lock); | |
6815 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, | |
6816 | this_cpu_ptr(&push_work)); | |
6817 | /* | |
6818 | * At this point need_resched() is true and we'll take the loop in | |
6819 | * schedule(). The next pick is obviously going to be the stop task | |
6820 | * which is_per_cpu_kthread() and will push this task away. | |
6821 | */ | |
6822 | raw_spin_lock(&rq->lock); | |
6823 | } | |
6824 | ||
6825 | static void balance_push_set(int cpu, bool on) | |
6826 | { | |
6827 | struct rq *rq = cpu_rq(cpu); | |
6828 | struct rq_flags rf; | |
6829 | ||
6830 | rq_lock_irqsave(rq, &rf); | |
6831 | if (on) | |
6832 | rq->balance_flags |= BALANCE_PUSH; | |
6833 | else | |
6834 | rq->balance_flags &= ~BALANCE_PUSH; | |
6835 | rq_unlock_irqrestore(rq, &rf); | |
6836 | } | |
6837 | ||
f2469a1f TG |
6838 | /* |
6839 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
6840 | * inactive. All tasks which are not per CPU kernel threads are either | |
6841 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
6842 | * during wakeup. Wait until the CPU is quiescent. | |
6843 | */ | |
6844 | static void balance_hotplug_wait(void) | |
6845 | { | |
6846 | struct rq *rq = this_rq(); | |
6847 | ||
6848 | rcuwait_wait_event(&rq->hotplug_wait, rq->nr_running == 1, | |
6849 | TASK_UNINTERRUPTIBLE); | |
6850 | } | |
6851 | ||
2558aacf PZ |
6852 | #else |
6853 | ||
6854 | static inline void balance_push(struct rq *rq) | |
6855 | { | |
6856 | } | |
6857 | ||
6858 | static inline void balance_push_set(int cpu, bool on) | |
6859 | { | |
6860 | } | |
6861 | ||
f2469a1f TG |
6862 | static inline void balance_hotplug_wait(void) |
6863 | { | |
6864 | } | |
6865 | ||
1da177e4 LT |
6866 | #endif /* CONFIG_HOTPLUG_CPU */ |
6867 | ||
f2cb1360 | 6868 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
6869 | { |
6870 | if (!rq->online) { | |
6871 | const struct sched_class *class; | |
6872 | ||
c6c4927b | 6873 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6874 | rq->online = 1; |
6875 | ||
6876 | for_each_class(class) { | |
6877 | if (class->rq_online) | |
6878 | class->rq_online(rq); | |
6879 | } | |
6880 | } | |
6881 | } | |
6882 | ||
f2cb1360 | 6883 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
6884 | { |
6885 | if (rq->online) { | |
6886 | const struct sched_class *class; | |
6887 | ||
6888 | for_each_class(class) { | |
6889 | if (class->rq_offline) | |
6890 | class->rq_offline(rq); | |
6891 | } | |
6892 | ||
c6c4927b | 6893 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6894 | rq->online = 0; |
6895 | } | |
6896 | } | |
6897 | ||
d1ccc66d IM |
6898 | /* |
6899 | * used to mark begin/end of suspend/resume: | |
6900 | */ | |
6901 | static int num_cpus_frozen; | |
d35be8ba | 6902 | |
1da177e4 | 6903 | /* |
3a101d05 TH |
6904 | * Update cpusets according to cpu_active mask. If cpusets are |
6905 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6906 | * around partition_sched_domains(). | |
d35be8ba SB |
6907 | * |
6908 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6909 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6910 | */ |
40190a78 | 6911 | static void cpuset_cpu_active(void) |
e761b772 | 6912 | { |
40190a78 | 6913 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
6914 | /* |
6915 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6916 | * resume sequence. As long as this is not the last online | |
6917 | * operation in the resume sequence, just build a single sched | |
6918 | * domain, ignoring cpusets. | |
6919 | */ | |
50e76632 PZ |
6920 | partition_sched_domains(1, NULL, NULL); |
6921 | if (--num_cpus_frozen) | |
135fb3e1 | 6922 | return; |
d35be8ba SB |
6923 | /* |
6924 | * This is the last CPU online operation. So fall through and | |
6925 | * restore the original sched domains by considering the | |
6926 | * cpuset configurations. | |
6927 | */ | |
50e76632 | 6928 | cpuset_force_rebuild(); |
3a101d05 | 6929 | } |
30e03acd | 6930 | cpuset_update_active_cpus(); |
3a101d05 | 6931 | } |
e761b772 | 6932 | |
40190a78 | 6933 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 6934 | { |
40190a78 | 6935 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 6936 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 6937 | return -EBUSY; |
30e03acd | 6938 | cpuset_update_active_cpus(); |
135fb3e1 | 6939 | } else { |
d35be8ba SB |
6940 | num_cpus_frozen++; |
6941 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 6942 | } |
135fb3e1 | 6943 | return 0; |
e761b772 | 6944 | } |
e761b772 | 6945 | |
40190a78 | 6946 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 6947 | { |
7d976699 | 6948 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 6949 | struct rq_flags rf; |
7d976699 | 6950 | |
2558aacf PZ |
6951 | balance_push_set(cpu, false); |
6952 | ||
ba2591a5 PZ |
6953 | #ifdef CONFIG_SCHED_SMT |
6954 | /* | |
c5511d03 | 6955 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 6956 | */ |
c5511d03 PZI |
6957 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
6958 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 6959 | #endif |
40190a78 | 6960 | set_cpu_active(cpu, true); |
135fb3e1 | 6961 | |
40190a78 | 6962 | if (sched_smp_initialized) { |
135fb3e1 | 6963 | sched_domains_numa_masks_set(cpu); |
40190a78 | 6964 | cpuset_cpu_active(); |
e761b772 | 6965 | } |
7d976699 TG |
6966 | |
6967 | /* | |
6968 | * Put the rq online, if not already. This happens: | |
6969 | * | |
6970 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 6971 | * after all CPUs have been brought up. |
7d976699 TG |
6972 | * |
6973 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
6974 | * domains. | |
6975 | */ | |
8a8c69c3 | 6976 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
6977 | if (rq->rd) { |
6978 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
6979 | set_rq_online(rq); | |
6980 | } | |
8a8c69c3 | 6981 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 6982 | |
40190a78 | 6983 | return 0; |
135fb3e1 TG |
6984 | } |
6985 | ||
40190a78 | 6986 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 6987 | { |
120455c5 PZ |
6988 | struct rq *rq = cpu_rq(cpu); |
6989 | struct rq_flags rf; | |
135fb3e1 TG |
6990 | int ret; |
6991 | ||
40190a78 | 6992 | set_cpu_active(cpu, false); |
b2454caa PZ |
6993 | /* |
6994 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
6995 | * users of this state to go away such that all new such users will | |
6996 | * observe it. | |
6997 | * | |
b2454caa PZ |
6998 | * Do sync before park smpboot threads to take care the rcu boost case. |
6999 | */ | |
309ba859 | 7000 | synchronize_rcu(); |
40190a78 | 7001 | |
2558aacf PZ |
7002 | balance_push_set(cpu, true); |
7003 | ||
120455c5 PZ |
7004 | rq_lock_irqsave(rq, &rf); |
7005 | if (rq->rd) { | |
7006 | update_rq_clock(rq); | |
7007 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7008 | set_rq_offline(rq); | |
7009 | } | |
7010 | rq_unlock_irqrestore(rq, &rf); | |
7011 | ||
c5511d03 PZI |
7012 | #ifdef CONFIG_SCHED_SMT |
7013 | /* | |
7014 | * When going down, decrement the number of cores with SMT present. | |
7015 | */ | |
7016 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
7017 | static_branch_dec_cpuslocked(&sched_smt_present); | |
7018 | #endif | |
7019 | ||
40190a78 TG |
7020 | if (!sched_smp_initialized) |
7021 | return 0; | |
7022 | ||
7023 | ret = cpuset_cpu_inactive(cpu); | |
7024 | if (ret) { | |
2558aacf | 7025 | balance_push_set(cpu, false); |
40190a78 TG |
7026 | set_cpu_active(cpu, true); |
7027 | return ret; | |
135fb3e1 | 7028 | } |
40190a78 TG |
7029 | sched_domains_numa_masks_clear(cpu); |
7030 | return 0; | |
135fb3e1 TG |
7031 | } |
7032 | ||
94baf7a5 TG |
7033 | static void sched_rq_cpu_starting(unsigned int cpu) |
7034 | { | |
7035 | struct rq *rq = cpu_rq(cpu); | |
7036 | ||
7037 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7038 | update_max_interval(); |
7039 | } | |
7040 | ||
135fb3e1 TG |
7041 | int sched_cpu_starting(unsigned int cpu) |
7042 | { | |
94baf7a5 | 7043 | sched_rq_cpu_starting(cpu); |
d84b3131 | 7044 | sched_tick_start(cpu); |
135fb3e1 | 7045 | return 0; |
e761b772 | 7046 | } |
e761b772 | 7047 | |
f2785ddb | 7048 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
7049 | |
7050 | /* | |
7051 | * Invoked immediately before the stopper thread is invoked to bring the | |
7052 | * CPU down completely. At this point all per CPU kthreads except the | |
7053 | * hotplug thread (current) and the stopper thread (inactive) have been | |
7054 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
7055 | * any of those which might be on the way out are gone. | |
7056 | * | |
7057 | * If after this point a bound task is being woken on this CPU then the | |
7058 | * responsible hotplug callback has failed to do it's job. | |
7059 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
7060 | */ | |
7061 | int sched_cpu_wait_empty(unsigned int cpu) | |
7062 | { | |
7063 | balance_hotplug_wait(); | |
7064 | return 0; | |
7065 | } | |
7066 | ||
7067 | /* | |
7068 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
7069 | * might have. Called from the CPU stopper task after ensuring that the | |
7070 | * stopper is the last running task on the CPU, so nr_active count is | |
7071 | * stable. We need to take the teardown thread which is calling this into | |
7072 | * account, so we hand in adjust = 1 to the load calculation. | |
7073 | * | |
7074 | * Also see the comment "Global load-average calculations". | |
7075 | */ | |
7076 | static void calc_load_migrate(struct rq *rq) | |
7077 | { | |
7078 | long delta = calc_load_fold_active(rq, 1); | |
7079 | ||
7080 | if (delta) | |
7081 | atomic_long_add(delta, &calc_load_tasks); | |
7082 | } | |
7083 | ||
f2785ddb TG |
7084 | int sched_cpu_dying(unsigned int cpu) |
7085 | { | |
7086 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 7087 | struct rq_flags rf; |
f2785ddb TG |
7088 | |
7089 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 7090 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
7091 | |
7092 | rq_lock_irqsave(rq, &rf); | |
f2785ddb | 7093 | BUG_ON(rq->nr_running != 1); |
8a8c69c3 PZ |
7094 | rq_unlock_irqrestore(rq, &rf); |
7095 | ||
f2785ddb TG |
7096 | calc_load_migrate(rq); |
7097 | update_max_interval(); | |
00357f5e | 7098 | nohz_balance_exit_idle(rq); |
e5ef27d0 | 7099 | hrtick_clear(rq); |
f2785ddb TG |
7100 | return 0; |
7101 | } | |
7102 | #endif | |
7103 | ||
1da177e4 LT |
7104 | void __init sched_init_smp(void) |
7105 | { | |
cb83b629 PZ |
7106 | sched_init_numa(); |
7107 | ||
6acce3ef PZ |
7108 | /* |
7109 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 7110 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 7111 | * happen. |
6acce3ef | 7112 | */ |
712555ee | 7113 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 7114 | sched_init_domains(cpu_active_mask); |
712555ee | 7115 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7116 | |
5c1e1767 | 7117 | /* Move init over to a non-isolated CPU */ |
edb93821 | 7118 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 7119 | BUG(); |
19978ca6 | 7120 | sched_init_granularity(); |
4212823f | 7121 | |
0e3900e6 | 7122 | init_sched_rt_class(); |
1baca4ce | 7123 | init_sched_dl_class(); |
1b568f0a | 7124 | |
e26fbffd | 7125 | sched_smp_initialized = true; |
1da177e4 | 7126 | } |
e26fbffd TG |
7127 | |
7128 | static int __init migration_init(void) | |
7129 | { | |
77a5352b | 7130 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 7131 | return 0; |
1da177e4 | 7132 | } |
e26fbffd TG |
7133 | early_initcall(migration_init); |
7134 | ||
1da177e4 LT |
7135 | #else |
7136 | void __init sched_init_smp(void) | |
7137 | { | |
19978ca6 | 7138 | sched_init_granularity(); |
1da177e4 LT |
7139 | } |
7140 | #endif /* CONFIG_SMP */ | |
7141 | ||
7142 | int in_sched_functions(unsigned long addr) | |
7143 | { | |
1da177e4 LT |
7144 | return in_lock_functions(addr) || |
7145 | (addr >= (unsigned long)__sched_text_start | |
7146 | && addr < (unsigned long)__sched_text_end); | |
7147 | } | |
7148 | ||
029632fb | 7149 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7150 | /* |
7151 | * Default task group. | |
7152 | * Every task in system belongs to this group at bootup. | |
7153 | */ | |
029632fb | 7154 | struct task_group root_task_group; |
35cf4e50 | 7155 | LIST_HEAD(task_groups); |
b0367629 WL |
7156 | |
7157 | /* Cacheline aligned slab cache for task_group */ | |
7158 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7159 | #endif |
6f505b16 | 7160 | |
e6252c3e | 7161 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 7162 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 7163 | |
1da177e4 LT |
7164 | void __init sched_init(void) |
7165 | { | |
a1dc0446 | 7166 | unsigned long ptr = 0; |
55627e3c | 7167 | int i; |
434d53b0 | 7168 | |
c3a340f7 SRV |
7169 | /* Make sure the linker didn't screw up */ |
7170 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
7171 | &fair_sched_class + 1 != &rt_sched_class || | |
7172 | &rt_sched_class + 1 != &dl_sched_class); | |
7173 | #ifdef CONFIG_SMP | |
7174 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
7175 | #endif | |
7176 | ||
5822a454 | 7177 | wait_bit_init(); |
9dcb8b68 | 7178 | |
434d53b0 | 7179 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 7180 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
7181 | #endif |
7182 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 7183 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 7184 | #endif |
a1dc0446 QC |
7185 | if (ptr) { |
7186 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
7187 | |
7188 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7189 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7190 | ptr += nr_cpu_ids * sizeof(void **); |
7191 | ||
07e06b01 | 7192 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7193 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7194 | |
b1d1779e WY |
7195 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
7196 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 7197 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7198 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7199 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7200 | ptr += nr_cpu_ids * sizeof(void **); |
7201 | ||
07e06b01 | 7202 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7203 | ptr += nr_cpu_ids * sizeof(void **); |
7204 | ||
6d6bc0ad | 7205 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7206 | } |
df7c8e84 | 7207 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7208 | for_each_possible_cpu(i) { |
7209 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7210 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
7211 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
7212 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7213 | } |
b74e6278 | 7214 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7215 | |
d1ccc66d IM |
7216 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
7217 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 7218 | |
57d885fe GH |
7219 | #ifdef CONFIG_SMP |
7220 | init_defrootdomain(); | |
7221 | #endif | |
7222 | ||
d0b27fa7 | 7223 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7224 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7225 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7226 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7227 | |
7c941438 | 7228 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7229 | task_group_cache = KMEM_CACHE(task_group, 0); |
7230 | ||
07e06b01 YZ |
7231 | list_add(&root_task_group.list, &task_groups); |
7232 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7233 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7234 | autogroup_init(&init_task); |
7c941438 | 7235 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7236 | |
0a945022 | 7237 | for_each_possible_cpu(i) { |
70b97a7f | 7238 | struct rq *rq; |
1da177e4 LT |
7239 | |
7240 | rq = cpu_rq(i); | |
05fa785c | 7241 | raw_spin_lock_init(&rq->lock); |
7897986b | 7242 | rq->nr_running = 0; |
dce48a84 TG |
7243 | rq->calc_load_active = 0; |
7244 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7245 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7246 | init_rt_rq(&rq->rt); |
7247 | init_dl_rq(&rq->dl); | |
dd41f596 | 7248 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 7249 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 7250 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 7251 | /* |
d1ccc66d | 7252 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
7253 | * |
7254 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
7255 | * gets 100% of the CPU resources in the system. This overall |
7256 | * system CPU resource is divided among the tasks of | |
07e06b01 | 7257 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7258 | * based on each entity's (task or task-group's) weight |
7259 | * (se->load.weight). | |
7260 | * | |
07e06b01 | 7261 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 7262 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 7263 | * then A0's share of the CPU resource is: |
354d60c2 | 7264 | * |
0d905bca | 7265 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7266 | * |
07e06b01 YZ |
7267 | * We achieve this by letting root_task_group's tasks sit |
7268 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7269 | */ |
07e06b01 | 7270 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7271 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7272 | ||
7273 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7274 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7275 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7276 | #endif |
1da177e4 | 7277 | #ifdef CONFIG_SMP |
41c7ce9a | 7278 | rq->sd = NULL; |
57d885fe | 7279 | rq->rd = NULL; |
ca6d75e6 | 7280 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7281 | rq->balance_callback = NULL; |
1da177e4 | 7282 | rq->active_balance = 0; |
dd41f596 | 7283 | rq->next_balance = jiffies; |
1da177e4 | 7284 | rq->push_cpu = 0; |
0a2966b4 | 7285 | rq->cpu = i; |
1f11eb6a | 7286 | rq->online = 0; |
eae0c9df MG |
7287 | rq->idle_stamp = 0; |
7288 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7289 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7290 | |
7291 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7292 | ||
dc938520 | 7293 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7294 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 7295 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 7296 | atomic_set(&rq->nohz_flags, 0); |
90b5363a PZI |
7297 | |
7298 | rq_csd_init(rq, &rq->nohz_csd, nohz_csd_func); | |
83cd4fe2 | 7299 | #endif |
f2469a1f TG |
7300 | #ifdef CONFIG_HOTPLUG_CPU |
7301 | rcuwait_init(&rq->hotplug_wait); | |
7302 | #endif | |
9fd81dd5 | 7303 | #endif /* CONFIG_SMP */ |
77a021be | 7304 | hrtick_rq_init(rq); |
1da177e4 | 7305 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7306 | } |
7307 | ||
9059393e | 7308 | set_load_weight(&init_task, false); |
b50f60ce | 7309 | |
1da177e4 LT |
7310 | /* |
7311 | * The boot idle thread does lazy MMU switching as well: | |
7312 | */ | |
f1f10076 | 7313 | mmgrab(&init_mm); |
1da177e4 LT |
7314 | enter_lazy_tlb(&init_mm, current); |
7315 | ||
7316 | /* | |
7317 | * Make us the idle thread. Technically, schedule() should not be | |
7318 | * called from this thread, however somewhere below it might be, | |
7319 | * but because we are the idle thread, we just pick up running again | |
7320 | * when this runqueue becomes "idle". | |
7321 | */ | |
7322 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7323 | |
7324 | calc_load_update = jiffies + LOAD_FREQ; | |
7325 | ||
bf4d83f6 | 7326 | #ifdef CONFIG_SMP |
29d5e047 | 7327 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7328 | #endif |
7329 | init_sched_fair_class(); | |
6a7b3dc3 | 7330 | |
4698f88c JP |
7331 | init_schedstats(); |
7332 | ||
eb414681 JW |
7333 | psi_init(); |
7334 | ||
69842cba PB |
7335 | init_uclamp(); |
7336 | ||
6892b75e | 7337 | scheduler_running = 1; |
1da177e4 LT |
7338 | } |
7339 | ||
d902db1e | 7340 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7341 | static inline int preempt_count_equals(int preempt_offset) |
7342 | { | |
da7142e2 | 7343 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7344 | |
4ba8216c | 7345 | return (nested == preempt_offset); |
e4aafea2 FW |
7346 | } |
7347 | ||
d894837f | 7348 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7349 | { |
8eb23b9f PZ |
7350 | /* |
7351 | * Blocking primitives will set (and therefore destroy) current->state, | |
7352 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7353 | * otherwise we will destroy state. | |
7354 | */ | |
00845eb9 | 7355 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7356 | "do not call blocking ops when !TASK_RUNNING; " |
7357 | "state=%lx set at [<%p>] %pS\n", | |
7358 | current->state, | |
7359 | (void *)current->task_state_change, | |
00845eb9 | 7360 | (void *)current->task_state_change); |
8eb23b9f | 7361 | |
3427445a PZ |
7362 | ___might_sleep(file, line, preempt_offset); |
7363 | } | |
7364 | EXPORT_SYMBOL(__might_sleep); | |
7365 | ||
7366 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7367 | { |
d1ccc66d IM |
7368 | /* Ratelimiting timestamp: */ |
7369 | static unsigned long prev_jiffy; | |
7370 | ||
d1c6d149 | 7371 | unsigned long preempt_disable_ip; |
1da177e4 | 7372 | |
d1ccc66d IM |
7373 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
7374 | rcu_sleep_check(); | |
7375 | ||
db273be2 | 7376 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 7377 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
7378 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
7379 | oops_in_progress) | |
aef745fc | 7380 | return; |
1c3c5eab | 7381 | |
aef745fc IM |
7382 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
7383 | return; | |
7384 | prev_jiffy = jiffies; | |
7385 | ||
d1ccc66d | 7386 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
7387 | preempt_disable_ip = get_preempt_disable_ip(current); |
7388 | ||
3df0fc5b PZ |
7389 | printk(KERN_ERR |
7390 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7391 | file, line); | |
7392 | printk(KERN_ERR | |
312364f3 DV |
7393 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
7394 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 7395 | current->pid, current->comm); |
aef745fc | 7396 | |
a8b686b3 ES |
7397 | if (task_stack_end_corrupted(current)) |
7398 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7399 | ||
aef745fc IM |
7400 | debug_show_held_locks(current); |
7401 | if (irqs_disabled()) | |
7402 | print_irqtrace_events(current); | |
d1c6d149 VN |
7403 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
7404 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 7405 | pr_err("Preemption disabled at:"); |
2062a4e8 | 7406 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 7407 | } |
aef745fc | 7408 | dump_stack(); |
f0b22e39 | 7409 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 7410 | } |
3427445a | 7411 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
7412 | |
7413 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
7414 | { | |
7415 | static unsigned long prev_jiffy; | |
7416 | ||
7417 | if (irqs_disabled()) | |
7418 | return; | |
7419 | ||
7420 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
7421 | return; | |
7422 | ||
7423 | if (preempt_count() > preempt_offset) | |
7424 | return; | |
7425 | ||
7426 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7427 | return; | |
7428 | prev_jiffy = jiffies; | |
7429 | ||
7430 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
7431 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7432 | in_atomic(), irqs_disabled(), | |
7433 | current->pid, current->comm); | |
7434 | ||
7435 | debug_show_held_locks(current); | |
7436 | dump_stack(); | |
7437 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
7438 | } | |
7439 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
1da177e4 LT |
7440 | #endif |
7441 | ||
7442 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7443 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7444 | { |
dbc7f069 | 7445 | struct task_struct *g, *p; |
d50dde5a DF |
7446 | struct sched_attr attr = { |
7447 | .sched_policy = SCHED_NORMAL, | |
7448 | }; | |
1da177e4 | 7449 | |
3472eaa1 | 7450 | read_lock(&tasklist_lock); |
5d07f420 | 7451 | for_each_process_thread(g, p) { |
178be793 IM |
7452 | /* |
7453 | * Only normalize user tasks: | |
7454 | */ | |
3472eaa1 | 7455 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7456 | continue; |
7457 | ||
4fa8d299 JP |
7458 | p->se.exec_start = 0; |
7459 | schedstat_set(p->se.statistics.wait_start, 0); | |
7460 | schedstat_set(p->se.statistics.sleep_start, 0); | |
7461 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 7462 | |
aab03e05 | 7463 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7464 | /* |
7465 | * Renice negative nice level userspace | |
7466 | * tasks back to 0: | |
7467 | */ | |
3472eaa1 | 7468 | if (task_nice(p) < 0) |
dd41f596 | 7469 | set_user_nice(p, 0); |
1da177e4 | 7470 | continue; |
dd41f596 | 7471 | } |
1da177e4 | 7472 | |
dbc7f069 | 7473 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7474 | } |
3472eaa1 | 7475 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7476 | } |
7477 | ||
7478 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7479 | |
67fc4e0c | 7480 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7481 | /* |
67fc4e0c | 7482 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7483 | * |
7484 | * They can only be called when the whole system has been | |
7485 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7486 | * activity can take place. Using them for anything else would | |
7487 | * be a serious bug, and as a result, they aren't even visible | |
7488 | * under any other configuration. | |
7489 | */ | |
7490 | ||
7491 | /** | |
d1ccc66d | 7492 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
7493 | * @cpu: the processor in question. |
7494 | * | |
7495 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7496 | * |
7497 | * Return: The current task for @cpu. | |
1df5c10a | 7498 | */ |
36c8b586 | 7499 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7500 | { |
7501 | return cpu_curr(cpu); | |
7502 | } | |
7503 | ||
67fc4e0c JW |
7504 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7505 | ||
7506 | #ifdef CONFIG_IA64 | |
1df5c10a | 7507 | /** |
5feeb783 | 7508 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
7509 | * @cpu: the processor in question. |
7510 | * @p: the task pointer to set. | |
7511 | * | |
7512 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 7513 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 7514 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
7515 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7516 | * and caller must save the original value of the current task (see | |
7517 | * curr_task() above) and restore that value before reenabling interrupts and | |
7518 | * re-starting the system. | |
7519 | * | |
7520 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7521 | */ | |
a458ae2e | 7522 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7523 | { |
7524 | cpu_curr(cpu) = p; | |
7525 | } | |
7526 | ||
7527 | #endif | |
29f59db3 | 7528 | |
7c941438 | 7529 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7530 | /* task_group_lock serializes the addition/removal of task groups */ |
7531 | static DEFINE_SPINLOCK(task_group_lock); | |
7532 | ||
2480c093 PB |
7533 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
7534 | struct task_group *parent) | |
7535 | { | |
7536 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 7537 | enum uclamp_id clamp_id; |
2480c093 PB |
7538 | |
7539 | for_each_clamp_id(clamp_id) { | |
7540 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
7541 | uclamp_none(clamp_id), false); | |
0b60ba2d | 7542 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
7543 | } |
7544 | #endif | |
7545 | } | |
7546 | ||
2f5177f0 | 7547 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7548 | { |
7549 | free_fair_sched_group(tg); | |
7550 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7551 | autogroup_free(tg); |
b0367629 | 7552 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7553 | } |
7554 | ||
7555 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7556 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7557 | { |
7558 | struct task_group *tg; | |
bccbe08a | 7559 | |
b0367629 | 7560 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7561 | if (!tg) |
7562 | return ERR_PTR(-ENOMEM); | |
7563 | ||
ec7dc8ac | 7564 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7565 | goto err; |
7566 | ||
ec7dc8ac | 7567 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7568 | goto err; |
7569 | ||
2480c093 PB |
7570 | alloc_uclamp_sched_group(tg, parent); |
7571 | ||
ace783b9 LZ |
7572 | return tg; |
7573 | ||
7574 | err: | |
2f5177f0 | 7575 | sched_free_group(tg); |
ace783b9 LZ |
7576 | return ERR_PTR(-ENOMEM); |
7577 | } | |
7578 | ||
7579 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7580 | { | |
7581 | unsigned long flags; | |
7582 | ||
8ed36996 | 7583 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7584 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 7585 | |
d1ccc66d IM |
7586 | /* Root should already exist: */ |
7587 | WARN_ON(!parent); | |
f473aa5e PZ |
7588 | |
7589 | tg->parent = parent; | |
f473aa5e | 7590 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7591 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7592 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
7593 | |
7594 | online_fair_sched_group(tg); | |
29f59db3 SV |
7595 | } |
7596 | ||
9b5b7751 | 7597 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7598 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7599 | { |
d1ccc66d | 7600 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 7601 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7602 | } |
7603 | ||
4cf86d77 | 7604 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 7605 | { |
d1ccc66d | 7606 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 7607 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7608 | } |
7609 | ||
7610 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7611 | { |
8ed36996 | 7612 | unsigned long flags; |
29f59db3 | 7613 | |
d1ccc66d | 7614 | /* End participation in shares distribution: */ |
6fe1f348 | 7615 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7616 | |
7617 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7618 | list_del_rcu(&tg->list); |
f473aa5e | 7619 | list_del_rcu(&tg->siblings); |
8ed36996 | 7620 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7621 | } |
7622 | ||
ea86cb4b | 7623 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 7624 | { |
8323f26c | 7625 | struct task_group *tg; |
29f59db3 | 7626 | |
f7b8a47d KT |
7627 | /* |
7628 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7629 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7630 | * to prevent lockdep warnings. | |
7631 | */ | |
7632 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7633 | struct task_group, css); |
7634 | tg = autogroup_task_group(tsk, tg); | |
7635 | tsk->sched_task_group = tg; | |
7636 | ||
810b3817 | 7637 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
7638 | if (tsk->sched_class->task_change_group) |
7639 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 7640 | else |
810b3817 | 7641 | #endif |
b2b5ce02 | 7642 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
7643 | } |
7644 | ||
7645 | /* | |
7646 | * Change task's runqueue when it moves between groups. | |
7647 | * | |
7648 | * The caller of this function should have put the task in its new group by | |
7649 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
7650 | * its new group. | |
7651 | */ | |
7652 | void sched_move_task(struct task_struct *tsk) | |
7653 | { | |
7a57f32a PZ |
7654 | int queued, running, queue_flags = |
7655 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
7656 | struct rq_flags rf; |
7657 | struct rq *rq; | |
7658 | ||
7659 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 7660 | update_rq_clock(rq); |
ea86cb4b VG |
7661 | |
7662 | running = task_current(rq, tsk); | |
7663 | queued = task_on_rq_queued(tsk); | |
7664 | ||
7665 | if (queued) | |
7a57f32a | 7666 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 7667 | if (running) |
ea86cb4b VG |
7668 | put_prev_task(rq, tsk); |
7669 | ||
7670 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 7671 | |
da0c1e65 | 7672 | if (queued) |
7a57f32a | 7673 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 7674 | if (running) { |
03b7fad1 | 7675 | set_next_task(rq, tsk); |
2a4b03ff VG |
7676 | /* |
7677 | * After changing group, the running task may have joined a | |
7678 | * throttled one but it's still the running task. Trigger a | |
7679 | * resched to make sure that task can still run. | |
7680 | */ | |
7681 | resched_curr(rq); | |
7682 | } | |
29f59db3 | 7683 | |
eb580751 | 7684 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 7685 | } |
68318b8e | 7686 | |
a7c6d554 | 7687 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7688 | { |
a7c6d554 | 7689 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7690 | } |
7691 | ||
eb95419b TH |
7692 | static struct cgroup_subsys_state * |
7693 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7694 | { |
eb95419b TH |
7695 | struct task_group *parent = css_tg(parent_css); |
7696 | struct task_group *tg; | |
68318b8e | 7697 | |
eb95419b | 7698 | if (!parent) { |
68318b8e | 7699 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7700 | return &root_task_group.css; |
68318b8e SV |
7701 | } |
7702 | ||
ec7dc8ac | 7703 | tg = sched_create_group(parent); |
68318b8e SV |
7704 | if (IS_ERR(tg)) |
7705 | return ERR_PTR(-ENOMEM); | |
7706 | ||
68318b8e SV |
7707 | return &tg->css; |
7708 | } | |
7709 | ||
96b77745 KK |
7710 | /* Expose task group only after completing cgroup initialization */ |
7711 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
7712 | { | |
7713 | struct task_group *tg = css_tg(css); | |
7714 | struct task_group *parent = css_tg(css->parent); | |
7715 | ||
7716 | if (parent) | |
7717 | sched_online_group(tg, parent); | |
7226017a QY |
7718 | |
7719 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
7720 | /* Propagate the effective uclamp value for the new group */ | |
7721 | cpu_util_update_eff(css); | |
7722 | #endif | |
7723 | ||
96b77745 KK |
7724 | return 0; |
7725 | } | |
7726 | ||
2f5177f0 | 7727 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 7728 | { |
eb95419b | 7729 | struct task_group *tg = css_tg(css); |
ace783b9 | 7730 | |
2f5177f0 | 7731 | sched_offline_group(tg); |
ace783b9 LZ |
7732 | } |
7733 | ||
eb95419b | 7734 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7735 | { |
eb95419b | 7736 | struct task_group *tg = css_tg(css); |
68318b8e | 7737 | |
2f5177f0 PZ |
7738 | /* |
7739 | * Relies on the RCU grace period between css_released() and this. | |
7740 | */ | |
7741 | sched_free_group(tg); | |
ace783b9 LZ |
7742 | } |
7743 | ||
ea86cb4b VG |
7744 | /* |
7745 | * This is called before wake_up_new_task(), therefore we really only | |
7746 | * have to set its group bits, all the other stuff does not apply. | |
7747 | */ | |
b53202e6 | 7748 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 7749 | { |
ea86cb4b VG |
7750 | struct rq_flags rf; |
7751 | struct rq *rq; | |
7752 | ||
7753 | rq = task_rq_lock(task, &rf); | |
7754 | ||
80f5c1b8 | 7755 | update_rq_clock(rq); |
ea86cb4b VG |
7756 | sched_change_group(task, TASK_SET_GROUP); |
7757 | ||
7758 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
7759 | } |
7760 | ||
1f7dd3e5 | 7761 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 7762 | { |
bb9d97b6 | 7763 | struct task_struct *task; |
1f7dd3e5 | 7764 | struct cgroup_subsys_state *css; |
7dc603c9 | 7765 | int ret = 0; |
bb9d97b6 | 7766 | |
1f7dd3e5 | 7767 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 7768 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7769 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7770 | return -EINVAL; |
b68aa230 | 7771 | #endif |
7dc603c9 PZ |
7772 | /* |
7773 | * Serialize against wake_up_new_task() such that if its | |
7774 | * running, we're sure to observe its full state. | |
7775 | */ | |
7776 | raw_spin_lock_irq(&task->pi_lock); | |
7777 | /* | |
7778 | * Avoid calling sched_move_task() before wake_up_new_task() | |
7779 | * has happened. This would lead to problems with PELT, due to | |
7780 | * move wanting to detach+attach while we're not attached yet. | |
7781 | */ | |
7782 | if (task->state == TASK_NEW) | |
7783 | ret = -EINVAL; | |
7784 | raw_spin_unlock_irq(&task->pi_lock); | |
7785 | ||
7786 | if (ret) | |
7787 | break; | |
bb9d97b6 | 7788 | } |
7dc603c9 | 7789 | return ret; |
be367d09 | 7790 | } |
68318b8e | 7791 | |
1f7dd3e5 | 7792 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 7793 | { |
bb9d97b6 | 7794 | struct task_struct *task; |
1f7dd3e5 | 7795 | struct cgroup_subsys_state *css; |
bb9d97b6 | 7796 | |
1f7dd3e5 | 7797 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 7798 | sched_move_task(task); |
68318b8e SV |
7799 | } |
7800 | ||
2480c093 | 7801 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
7802 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
7803 | { | |
7804 | struct cgroup_subsys_state *top_css = css; | |
7805 | struct uclamp_se *uc_parent = NULL; | |
7806 | struct uclamp_se *uc_se = NULL; | |
7807 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 7808 | enum uclamp_id clamp_id; |
0b60ba2d PB |
7809 | unsigned int clamps; |
7810 | ||
7811 | css_for_each_descendant_pre(css, top_css) { | |
7812 | uc_parent = css_tg(css)->parent | |
7813 | ? css_tg(css)->parent->uclamp : NULL; | |
7814 | ||
7815 | for_each_clamp_id(clamp_id) { | |
7816 | /* Assume effective clamps matches requested clamps */ | |
7817 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
7818 | /* Cap effective clamps with parent's effective clamps */ | |
7819 | if (uc_parent && | |
7820 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
7821 | eff[clamp_id] = uc_parent[clamp_id].value; | |
7822 | } | |
7823 | } | |
7824 | /* Ensure protection is always capped by limit */ | |
7825 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
7826 | ||
7827 | /* Propagate most restrictive effective clamps */ | |
7828 | clamps = 0x0; | |
7829 | uc_se = css_tg(css)->uclamp; | |
7830 | for_each_clamp_id(clamp_id) { | |
7831 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
7832 | continue; | |
7833 | uc_se[clamp_id].value = eff[clamp_id]; | |
7834 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
7835 | clamps |= (0x1 << clamp_id); | |
7836 | } | |
babbe170 | 7837 | if (!clamps) { |
0b60ba2d | 7838 | css = css_rightmost_descendant(css); |
babbe170 PB |
7839 | continue; |
7840 | } | |
7841 | ||
7842 | /* Immediately update descendants RUNNABLE tasks */ | |
7843 | uclamp_update_active_tasks(css, clamps); | |
0b60ba2d PB |
7844 | } |
7845 | } | |
2480c093 PB |
7846 | |
7847 | /* | |
7848 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
7849 | * C expression. Since there is no way to convert a macro argument (N) into a | |
7850 | * character constant, use two levels of macros. | |
7851 | */ | |
7852 | #define _POW10(exp) ((unsigned int)1e##exp) | |
7853 | #define POW10(exp) _POW10(exp) | |
7854 | ||
7855 | struct uclamp_request { | |
7856 | #define UCLAMP_PERCENT_SHIFT 2 | |
7857 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
7858 | s64 percent; | |
7859 | u64 util; | |
7860 | int ret; | |
7861 | }; | |
7862 | ||
7863 | static inline struct uclamp_request | |
7864 | capacity_from_percent(char *buf) | |
7865 | { | |
7866 | struct uclamp_request req = { | |
7867 | .percent = UCLAMP_PERCENT_SCALE, | |
7868 | .util = SCHED_CAPACITY_SCALE, | |
7869 | .ret = 0, | |
7870 | }; | |
7871 | ||
7872 | buf = strim(buf); | |
7873 | if (strcmp(buf, "max")) { | |
7874 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
7875 | &req.percent); | |
7876 | if (req.ret) | |
7877 | return req; | |
b562d140 | 7878 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
7879 | req.ret = -ERANGE; |
7880 | return req; | |
7881 | } | |
7882 | ||
7883 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
7884 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
7885 | } | |
7886 | ||
7887 | return req; | |
7888 | } | |
7889 | ||
7890 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
7891 | size_t nbytes, loff_t off, | |
7892 | enum uclamp_id clamp_id) | |
7893 | { | |
7894 | struct uclamp_request req; | |
7895 | struct task_group *tg; | |
7896 | ||
7897 | req = capacity_from_percent(buf); | |
7898 | if (req.ret) | |
7899 | return req.ret; | |
7900 | ||
46609ce2 QY |
7901 | static_branch_enable(&sched_uclamp_used); |
7902 | ||
2480c093 PB |
7903 | mutex_lock(&uclamp_mutex); |
7904 | rcu_read_lock(); | |
7905 | ||
7906 | tg = css_tg(of_css(of)); | |
7907 | if (tg->uclamp_req[clamp_id].value != req.util) | |
7908 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
7909 | ||
7910 | /* | |
7911 | * Because of not recoverable conversion rounding we keep track of the | |
7912 | * exact requested value | |
7913 | */ | |
7914 | tg->uclamp_pct[clamp_id] = req.percent; | |
7915 | ||
0b60ba2d PB |
7916 | /* Update effective clamps to track the most restrictive value */ |
7917 | cpu_util_update_eff(of_css(of)); | |
7918 | ||
2480c093 PB |
7919 | rcu_read_unlock(); |
7920 | mutex_unlock(&uclamp_mutex); | |
7921 | ||
7922 | return nbytes; | |
7923 | } | |
7924 | ||
7925 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
7926 | char *buf, size_t nbytes, | |
7927 | loff_t off) | |
7928 | { | |
7929 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
7930 | } | |
7931 | ||
7932 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
7933 | char *buf, size_t nbytes, | |
7934 | loff_t off) | |
7935 | { | |
7936 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
7937 | } | |
7938 | ||
7939 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
7940 | enum uclamp_id clamp_id) | |
7941 | { | |
7942 | struct task_group *tg; | |
7943 | u64 util_clamp; | |
7944 | u64 percent; | |
7945 | u32 rem; | |
7946 | ||
7947 | rcu_read_lock(); | |
7948 | tg = css_tg(seq_css(sf)); | |
7949 | util_clamp = tg->uclamp_req[clamp_id].value; | |
7950 | rcu_read_unlock(); | |
7951 | ||
7952 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
7953 | seq_puts(sf, "max\n"); | |
7954 | return; | |
7955 | } | |
7956 | ||
7957 | percent = tg->uclamp_pct[clamp_id]; | |
7958 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
7959 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
7960 | } | |
7961 | ||
7962 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
7963 | { | |
7964 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
7965 | return 0; | |
7966 | } | |
7967 | ||
7968 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
7969 | { | |
7970 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
7971 | return 0; | |
7972 | } | |
7973 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
7974 | ||
052f1dc7 | 7975 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7976 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7977 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7978 | { |
5b61d50a KK |
7979 | if (shareval > scale_load_down(ULONG_MAX)) |
7980 | shareval = MAX_SHARES; | |
182446d0 | 7981 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7982 | } |
7983 | ||
182446d0 TH |
7984 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7985 | struct cftype *cft) | |
68318b8e | 7986 | { |
182446d0 | 7987 | struct task_group *tg = css_tg(css); |
68318b8e | 7988 | |
c8b28116 | 7989 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7990 | } |
ab84d31e PT |
7991 | |
7992 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7993 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7994 | ||
ab84d31e | 7995 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 7996 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
7997 | /* More than 203 days if BW_SHIFT equals 20. */ |
7998 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 7999 | |
a790de99 PT |
8000 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8001 | ||
ab84d31e PT |
8002 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8003 | { | |
56f570e5 | 8004 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8005 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8006 | |
8007 | if (tg == &root_task_group) | |
8008 | return -EINVAL; | |
8009 | ||
8010 | /* | |
8011 | * Ensure we have at some amount of bandwidth every period. This is | |
8012 | * to prevent reaching a state of large arrears when throttled via | |
8013 | * entity_tick() resulting in prolonged exit starvation. | |
8014 | */ | |
8015 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8016 | return -EINVAL; | |
8017 | ||
8018 | /* | |
8019 | * Likewise, bound things on the otherside by preventing insane quota | |
8020 | * periods. This also allows us to normalize in computing quota | |
8021 | * feasibility. | |
8022 | */ | |
8023 | if (period > max_cfs_quota_period) | |
8024 | return -EINVAL; | |
8025 | ||
d505b8af HC |
8026 | /* |
8027 | * Bound quota to defend quota against overflow during bandwidth shift. | |
8028 | */ | |
8029 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
8030 | return -EINVAL; | |
8031 | ||
0e59bdae KT |
8032 | /* |
8033 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8034 | * unthrottle_offline_cfs_rqs(). | |
8035 | */ | |
8036 | get_online_cpus(); | |
a790de99 PT |
8037 | mutex_lock(&cfs_constraints_mutex); |
8038 | ret = __cfs_schedulable(tg, period, quota); | |
8039 | if (ret) | |
8040 | goto out_unlock; | |
8041 | ||
58088ad0 | 8042 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8043 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8044 | /* |
8045 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8046 | * before making related changes, and on->off must occur afterwards | |
8047 | */ | |
8048 | if (runtime_enabled && !runtime_was_enabled) | |
8049 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8050 | raw_spin_lock_irq(&cfs_b->lock); |
8051 | cfs_b->period = ns_to_ktime(period); | |
8052 | cfs_b->quota = quota; | |
58088ad0 | 8053 | |
a9cf55b2 | 8054 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
8055 | |
8056 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
8057 | if (runtime_enabled) |
8058 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 8059 | |
ab84d31e PT |
8060 | raw_spin_unlock_irq(&cfs_b->lock); |
8061 | ||
0e59bdae | 8062 | for_each_online_cpu(i) { |
ab84d31e | 8063 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8064 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 8065 | struct rq_flags rf; |
ab84d31e | 8066 | |
8a8c69c3 | 8067 | rq_lock_irq(rq, &rf); |
58088ad0 | 8068 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8069 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8070 | |
029632fb | 8071 | if (cfs_rq->throttled) |
671fd9da | 8072 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 8073 | rq_unlock_irq(rq, &rf); |
ab84d31e | 8074 | } |
1ee14e6c BS |
8075 | if (runtime_was_enabled && !runtime_enabled) |
8076 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8077 | out_unlock: |
8078 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8079 | put_online_cpus(); |
ab84d31e | 8080 | |
a790de99 | 8081 | return ret; |
ab84d31e PT |
8082 | } |
8083 | ||
b1546edc | 8084 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e PT |
8085 | { |
8086 | u64 quota, period; | |
8087 | ||
029632fb | 8088 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8089 | if (cfs_quota_us < 0) |
8090 | quota = RUNTIME_INF; | |
1a8b4540 | 8091 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 8092 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
8093 | else |
8094 | return -EINVAL; | |
ab84d31e PT |
8095 | |
8096 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8097 | } | |
8098 | ||
b1546edc | 8099 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
8100 | { |
8101 | u64 quota_us; | |
8102 | ||
029632fb | 8103 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8104 | return -1; |
8105 | ||
029632fb | 8106 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8107 | do_div(quota_us, NSEC_PER_USEC); |
8108 | ||
8109 | return quota_us; | |
8110 | } | |
8111 | ||
b1546edc | 8112 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e PT |
8113 | { |
8114 | u64 quota, period; | |
8115 | ||
1a8b4540 KK |
8116 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
8117 | return -EINVAL; | |
8118 | ||
ab84d31e | 8119 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 8120 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8121 | |
ab84d31e PT |
8122 | return tg_set_cfs_bandwidth(tg, period, quota); |
8123 | } | |
8124 | ||
b1546edc | 8125 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
8126 | { |
8127 | u64 cfs_period_us; | |
8128 | ||
029632fb | 8129 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8130 | do_div(cfs_period_us, NSEC_PER_USEC); |
8131 | ||
8132 | return cfs_period_us; | |
8133 | } | |
8134 | ||
182446d0 TH |
8135 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8136 | struct cftype *cft) | |
ab84d31e | 8137 | { |
182446d0 | 8138 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8139 | } |
8140 | ||
182446d0 TH |
8141 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8142 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8143 | { |
182446d0 | 8144 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8145 | } |
8146 | ||
182446d0 TH |
8147 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8148 | struct cftype *cft) | |
ab84d31e | 8149 | { |
182446d0 | 8150 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8151 | } |
8152 | ||
182446d0 TH |
8153 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8154 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8155 | { |
182446d0 | 8156 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8157 | } |
8158 | ||
a790de99 PT |
8159 | struct cfs_schedulable_data { |
8160 | struct task_group *tg; | |
8161 | u64 period, quota; | |
8162 | }; | |
8163 | ||
8164 | /* | |
8165 | * normalize group quota/period to be quota/max_period | |
8166 | * note: units are usecs | |
8167 | */ | |
8168 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8169 | struct cfs_schedulable_data *d) | |
8170 | { | |
8171 | u64 quota, period; | |
8172 | ||
8173 | if (tg == d->tg) { | |
8174 | period = d->period; | |
8175 | quota = d->quota; | |
8176 | } else { | |
8177 | period = tg_get_cfs_period(tg); | |
8178 | quota = tg_get_cfs_quota(tg); | |
8179 | } | |
8180 | ||
8181 | /* note: these should typically be equivalent */ | |
8182 | if (quota == RUNTIME_INF || quota == -1) | |
8183 | return RUNTIME_INF; | |
8184 | ||
8185 | return to_ratio(period, quota); | |
8186 | } | |
8187 | ||
8188 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8189 | { | |
8190 | struct cfs_schedulable_data *d = data; | |
029632fb | 8191 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8192 | s64 quota = 0, parent_quota = -1; |
8193 | ||
8194 | if (!tg->parent) { | |
8195 | quota = RUNTIME_INF; | |
8196 | } else { | |
029632fb | 8197 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8198 | |
8199 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8200 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8201 | |
8202 | /* | |
c53593e5 TH |
8203 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
8204 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 8205 | * limit is set: |
a790de99 | 8206 | */ |
c53593e5 TH |
8207 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
8208 | quota = min(quota, parent_quota); | |
8209 | } else { | |
8210 | if (quota == RUNTIME_INF) | |
8211 | quota = parent_quota; | |
8212 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8213 | return -EINVAL; | |
8214 | } | |
a790de99 | 8215 | } |
9c58c79a | 8216 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8217 | |
8218 | return 0; | |
8219 | } | |
8220 | ||
8221 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8222 | { | |
8277434e | 8223 | int ret; |
a790de99 PT |
8224 | struct cfs_schedulable_data data = { |
8225 | .tg = tg, | |
8226 | .period = period, | |
8227 | .quota = quota, | |
8228 | }; | |
8229 | ||
8230 | if (quota != RUNTIME_INF) { | |
8231 | do_div(data.period, NSEC_PER_USEC); | |
8232 | do_div(data.quota, NSEC_PER_USEC); | |
8233 | } | |
8234 | ||
8277434e PT |
8235 | rcu_read_lock(); |
8236 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8237 | rcu_read_unlock(); | |
8238 | ||
8239 | return ret; | |
a790de99 | 8240 | } |
e8da1b18 | 8241 | |
a1f7164c | 8242 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 8243 | { |
2da8ca82 | 8244 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8245 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8246 | |
44ffc75b TH |
8247 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8248 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8249 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 8250 | |
3d6c50c2 YW |
8251 | if (schedstat_enabled() && tg != &root_task_group) { |
8252 | u64 ws = 0; | |
8253 | int i; | |
8254 | ||
8255 | for_each_possible_cpu(i) | |
8256 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
8257 | ||
8258 | seq_printf(sf, "wait_sum %llu\n", ws); | |
8259 | } | |
8260 | ||
e8da1b18 NR |
8261 | return 0; |
8262 | } | |
ab84d31e | 8263 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8264 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8265 | |
052f1dc7 | 8266 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8267 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8268 | struct cftype *cft, s64 val) | |
6f505b16 | 8269 | { |
182446d0 | 8270 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8271 | } |
8272 | ||
182446d0 TH |
8273 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8274 | struct cftype *cft) | |
6f505b16 | 8275 | { |
182446d0 | 8276 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8277 | } |
d0b27fa7 | 8278 | |
182446d0 TH |
8279 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8280 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8281 | { |
182446d0 | 8282 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8283 | } |
8284 | ||
182446d0 TH |
8285 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8286 | struct cftype *cft) | |
d0b27fa7 | 8287 | { |
182446d0 | 8288 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8289 | } |
6d6bc0ad | 8290 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8291 | |
a1f7164c | 8292 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 8293 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8294 | { |
8295 | .name = "shares", | |
f4c753b7 PM |
8296 | .read_u64 = cpu_shares_read_u64, |
8297 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8298 | }, |
052f1dc7 | 8299 | #endif |
ab84d31e PT |
8300 | #ifdef CONFIG_CFS_BANDWIDTH |
8301 | { | |
8302 | .name = "cfs_quota_us", | |
8303 | .read_s64 = cpu_cfs_quota_read_s64, | |
8304 | .write_s64 = cpu_cfs_quota_write_s64, | |
8305 | }, | |
8306 | { | |
8307 | .name = "cfs_period_us", | |
8308 | .read_u64 = cpu_cfs_period_read_u64, | |
8309 | .write_u64 = cpu_cfs_period_write_u64, | |
8310 | }, | |
e8da1b18 NR |
8311 | { |
8312 | .name = "stat", | |
a1f7164c | 8313 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 8314 | }, |
ab84d31e | 8315 | #endif |
052f1dc7 | 8316 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8317 | { |
9f0c1e56 | 8318 | .name = "rt_runtime_us", |
06ecb27c PM |
8319 | .read_s64 = cpu_rt_runtime_read, |
8320 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8321 | }, |
d0b27fa7 PZ |
8322 | { |
8323 | .name = "rt_period_us", | |
f4c753b7 PM |
8324 | .read_u64 = cpu_rt_period_read_uint, |
8325 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8326 | }, |
2480c093 PB |
8327 | #endif |
8328 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8329 | { | |
8330 | .name = "uclamp.min", | |
8331 | .flags = CFTYPE_NOT_ON_ROOT, | |
8332 | .seq_show = cpu_uclamp_min_show, | |
8333 | .write = cpu_uclamp_min_write, | |
8334 | }, | |
8335 | { | |
8336 | .name = "uclamp.max", | |
8337 | .flags = CFTYPE_NOT_ON_ROOT, | |
8338 | .seq_show = cpu_uclamp_max_show, | |
8339 | .write = cpu_uclamp_max_write, | |
8340 | }, | |
052f1dc7 | 8341 | #endif |
d1ccc66d | 8342 | { } /* Terminate */ |
68318b8e SV |
8343 | }; |
8344 | ||
d41bf8c9 TH |
8345 | static int cpu_extra_stat_show(struct seq_file *sf, |
8346 | struct cgroup_subsys_state *css) | |
0d593634 | 8347 | { |
0d593634 TH |
8348 | #ifdef CONFIG_CFS_BANDWIDTH |
8349 | { | |
d41bf8c9 | 8350 | struct task_group *tg = css_tg(css); |
0d593634 TH |
8351 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
8352 | u64 throttled_usec; | |
8353 | ||
8354 | throttled_usec = cfs_b->throttled_time; | |
8355 | do_div(throttled_usec, NSEC_PER_USEC); | |
8356 | ||
8357 | seq_printf(sf, "nr_periods %d\n" | |
8358 | "nr_throttled %d\n" | |
8359 | "throttled_usec %llu\n", | |
8360 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
8361 | throttled_usec); | |
8362 | } | |
8363 | #endif | |
8364 | return 0; | |
8365 | } | |
8366 | ||
8367 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8368 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
8369 | struct cftype *cft) | |
8370 | { | |
8371 | struct task_group *tg = css_tg(css); | |
8372 | u64 weight = scale_load_down(tg->shares); | |
8373 | ||
8374 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
8375 | } | |
8376 | ||
8377 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
8378 | struct cftype *cft, u64 weight) | |
8379 | { | |
8380 | /* | |
8381 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
8382 | * values which are 1, 100 and 10000 respectively. While it loses | |
8383 | * a bit of range on both ends, it maps pretty well onto the shares | |
8384 | * value used by scheduler and the round-trip conversions preserve | |
8385 | * the original value over the entire range. | |
8386 | */ | |
8387 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
8388 | return -ERANGE; | |
8389 | ||
8390 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
8391 | ||
8392 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
8393 | } | |
8394 | ||
8395 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
8396 | struct cftype *cft) | |
8397 | { | |
8398 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
8399 | int last_delta = INT_MAX; | |
8400 | int prio, delta; | |
8401 | ||
8402 | /* find the closest nice value to the current weight */ | |
8403 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
8404 | delta = abs(sched_prio_to_weight[prio] - weight); | |
8405 | if (delta >= last_delta) | |
8406 | break; | |
8407 | last_delta = delta; | |
8408 | } | |
8409 | ||
8410 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
8411 | } | |
8412 | ||
8413 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
8414 | struct cftype *cft, s64 nice) | |
8415 | { | |
8416 | unsigned long weight; | |
7281c8de | 8417 | int idx; |
0d593634 TH |
8418 | |
8419 | if (nice < MIN_NICE || nice > MAX_NICE) | |
8420 | return -ERANGE; | |
8421 | ||
7281c8de PZ |
8422 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
8423 | idx = array_index_nospec(idx, 40); | |
8424 | weight = sched_prio_to_weight[idx]; | |
8425 | ||
0d593634 TH |
8426 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
8427 | } | |
8428 | #endif | |
8429 | ||
8430 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
8431 | long period, long quota) | |
8432 | { | |
8433 | if (quota < 0) | |
8434 | seq_puts(sf, "max"); | |
8435 | else | |
8436 | seq_printf(sf, "%ld", quota); | |
8437 | ||
8438 | seq_printf(sf, " %ld\n", period); | |
8439 | } | |
8440 | ||
8441 | /* caller should put the current value in *@periodp before calling */ | |
8442 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
8443 | u64 *periodp, u64 *quotap) | |
8444 | { | |
8445 | char tok[21]; /* U64_MAX */ | |
8446 | ||
4c47acd8 | 8447 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
8448 | return -EINVAL; |
8449 | ||
8450 | *periodp *= NSEC_PER_USEC; | |
8451 | ||
8452 | if (sscanf(tok, "%llu", quotap)) | |
8453 | *quotap *= NSEC_PER_USEC; | |
8454 | else if (!strcmp(tok, "max")) | |
8455 | *quotap = RUNTIME_INF; | |
8456 | else | |
8457 | return -EINVAL; | |
8458 | ||
8459 | return 0; | |
8460 | } | |
8461 | ||
8462 | #ifdef CONFIG_CFS_BANDWIDTH | |
8463 | static int cpu_max_show(struct seq_file *sf, void *v) | |
8464 | { | |
8465 | struct task_group *tg = css_tg(seq_css(sf)); | |
8466 | ||
8467 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
8468 | return 0; | |
8469 | } | |
8470 | ||
8471 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
8472 | char *buf, size_t nbytes, loff_t off) | |
8473 | { | |
8474 | struct task_group *tg = css_tg(of_css(of)); | |
8475 | u64 period = tg_get_cfs_period(tg); | |
8476 | u64 quota; | |
8477 | int ret; | |
8478 | ||
8479 | ret = cpu_period_quota_parse(buf, &period, "a); | |
8480 | if (!ret) | |
8481 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
8482 | return ret ?: nbytes; | |
8483 | } | |
8484 | #endif | |
8485 | ||
8486 | static struct cftype cpu_files[] = { | |
0d593634 TH |
8487 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8488 | { | |
8489 | .name = "weight", | |
8490 | .flags = CFTYPE_NOT_ON_ROOT, | |
8491 | .read_u64 = cpu_weight_read_u64, | |
8492 | .write_u64 = cpu_weight_write_u64, | |
8493 | }, | |
8494 | { | |
8495 | .name = "weight.nice", | |
8496 | .flags = CFTYPE_NOT_ON_ROOT, | |
8497 | .read_s64 = cpu_weight_nice_read_s64, | |
8498 | .write_s64 = cpu_weight_nice_write_s64, | |
8499 | }, | |
8500 | #endif | |
8501 | #ifdef CONFIG_CFS_BANDWIDTH | |
8502 | { | |
8503 | .name = "max", | |
8504 | .flags = CFTYPE_NOT_ON_ROOT, | |
8505 | .seq_show = cpu_max_show, | |
8506 | .write = cpu_max_write, | |
8507 | }, | |
2480c093 PB |
8508 | #endif |
8509 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8510 | { | |
8511 | .name = "uclamp.min", | |
8512 | .flags = CFTYPE_NOT_ON_ROOT, | |
8513 | .seq_show = cpu_uclamp_min_show, | |
8514 | .write = cpu_uclamp_min_write, | |
8515 | }, | |
8516 | { | |
8517 | .name = "uclamp.max", | |
8518 | .flags = CFTYPE_NOT_ON_ROOT, | |
8519 | .seq_show = cpu_uclamp_max_show, | |
8520 | .write = cpu_uclamp_max_write, | |
8521 | }, | |
0d593634 TH |
8522 | #endif |
8523 | { } /* terminate */ | |
8524 | }; | |
8525 | ||
073219e9 | 8526 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8527 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 8528 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 8529 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8530 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 8531 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 8532 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8533 | .can_attach = cpu_cgroup_can_attach, |
8534 | .attach = cpu_cgroup_attach, | |
a1f7164c | 8535 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 8536 | .dfl_cftypes = cpu_files, |
b38e42e9 | 8537 | .early_init = true, |
0d593634 | 8538 | .threaded = true, |
68318b8e SV |
8539 | }; |
8540 | ||
052f1dc7 | 8541 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8542 | |
b637a328 PM |
8543 | void dump_cpu_task(int cpu) |
8544 | { | |
8545 | pr_info("Task dump for CPU %d:\n", cpu); | |
8546 | sched_show_task(cpu_curr(cpu)); | |
8547 | } | |
ed82b8a1 AK |
8548 | |
8549 | /* | |
8550 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8551 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8552 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8553 | * that remained on nice 0. | |
8554 | * | |
8555 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8556 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8557 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8558 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8559 | * the relative distance between them is ~25%.) | |
8560 | */ | |
8561 | const int sched_prio_to_weight[40] = { | |
8562 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8563 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8564 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8565 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8566 | /* 0 */ 1024, 820, 655, 526, 423, | |
8567 | /* 5 */ 335, 272, 215, 172, 137, | |
8568 | /* 10 */ 110, 87, 70, 56, 45, | |
8569 | /* 15 */ 36, 29, 23, 18, 15, | |
8570 | }; | |
8571 | ||
8572 | /* | |
8573 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8574 | * | |
8575 | * In cases where the weight does not change often, we can use the | |
8576 | * precalculated inverse to speed up arithmetics by turning divisions | |
8577 | * into multiplications: | |
8578 | */ | |
8579 | const u32 sched_prio_to_wmult[40] = { | |
8580 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8581 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8582 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8583 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8584 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8585 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8586 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8587 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8588 | }; | |
14a7405b | 8589 | |
9d246053 PA |
8590 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
8591 | { | |
8592 | trace_sched_update_nr_running_tp(rq, count); | |
8593 | } |