]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
391e43da | 3 | * kernel/sched/core.c |
1da177e4 | 4 | * |
d1ccc66d | 5 | * Core kernel scheduler code and related syscalls |
1da177e4 LT |
6 | * |
7 | * Copyright (C) 1991-2002 Linus Torvalds | |
1da177e4 | 8 | */ |
9d246053 PA |
9 | #define CREATE_TRACE_POINTS |
10 | #include <trace/events/sched.h> | |
11 | #undef CREATE_TRACE_POINTS | |
12 | ||
325ea10c | 13 | #include "sched.h" |
1da177e4 | 14 | |
7281c8de | 15 | #include <linux/nospec.h> |
85f1abe0 | 16 | |
0ed557aa | 17 | #include <linux/kcov.h> |
d08b9f0c | 18 | #include <linux/scs.h> |
0ed557aa | 19 | |
96f951ed | 20 | #include <asm/switch_to.h> |
5517d86b | 21 | #include <asm/tlb.h> |
1da177e4 | 22 | |
ea138446 | 23 | #include "../workqueue_internal.h" |
771b53d0 | 24 | #include "../../fs/io-wq.h" |
29d5e047 | 25 | #include "../smpboot.h" |
6e0534f2 | 26 | |
91c27493 | 27 | #include "pelt.h" |
1f8db415 | 28 | #include "smp.h" |
91c27493 | 29 | |
a056a5be QY |
30 | /* |
31 | * Export tracepoints that act as a bare tracehook (ie: have no trace event | |
32 | * associated with them) to allow external modules to probe them. | |
33 | */ | |
34 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp); | |
35 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp); | |
36 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp); | |
37 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp); | |
38 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp); | |
51cf18c9 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp); |
a056a5be | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); |
4581bea8 VD |
41 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp); |
42 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); | |
9d246053 | 43 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); |
a056a5be | 44 | |
029632fb | 45 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 46 | |
a73f863a | 47 | #ifdef CONFIG_SCHED_DEBUG |
bf5c91ba IM |
48 | /* |
49 | * Debugging: various feature bits | |
765cc3a4 PB |
50 | * |
51 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
52 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
53 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 54 | */ |
f00b45c1 PZ |
55 | #define SCHED_FEAT(name, enabled) \ |
56 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 57 | const_debug unsigned int sysctl_sched_features = |
391e43da | 58 | #include "features.h" |
f00b45c1 | 59 | 0; |
f00b45c1 | 60 | #undef SCHED_FEAT |
765cc3a4 | 61 | #endif |
f00b45c1 | 62 | |
b82d9fdd PZ |
63 | /* |
64 | * Number of tasks to iterate in a single balance run. | |
65 | * Limited because this is done with IRQs disabled. | |
66 | */ | |
67 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
68 | ||
fa85ae24 | 69 | /* |
d1ccc66d | 70 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
71 | * default: 1s |
72 | */ | |
9f0c1e56 | 73 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 74 | |
029632fb | 75 | __read_mostly int scheduler_running; |
6892b75e | 76 | |
9f0c1e56 PZ |
77 | /* |
78 | * part of the period that we allow rt tasks to run in us. | |
79 | * default: 0.95s | |
80 | */ | |
81 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 82 | |
58877d34 PZ |
83 | |
84 | /* | |
85 | * Serialization rules: | |
86 | * | |
87 | * Lock order: | |
88 | * | |
89 | * p->pi_lock | |
90 | * rq->lock | |
91 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
92 | * | |
93 | * rq1->lock | |
94 | * rq2->lock where: rq1 < rq2 | |
95 | * | |
96 | * Regular state: | |
97 | * | |
98 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
99 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
b19a888c | 100 | * always looks at the local rq data structures to find the most eligible task |
58877d34 PZ |
101 | * to run next. |
102 | * | |
103 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
104 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
105 | * the local CPU to avoid bouncing the runqueue state around [ see | |
106 | * ttwu_queue_wakelist() ] | |
107 | * | |
108 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
109 | * complicated to avoid having to take two rq->locks. | |
110 | * | |
111 | * Special state: | |
112 | * | |
113 | * System-calls and anything external will use task_rq_lock() which acquires | |
114 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
115 | * stable while holding either lock: | |
116 | * | |
117 | * - sched_setaffinity()/ | |
118 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
119 | * - set_user_nice(): p->se.load, p->*prio | |
120 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
121 | * p->se.load, p->rt_priority, | |
122 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
123 | * - sched_setnuma(): p->numa_preferred_nid | |
124 | * - sched_move_task()/ | |
125 | * cpu_cgroup_fork(): p->sched_task_group | |
126 | * - uclamp_update_active() p->uclamp* | |
127 | * | |
128 | * p->state <- TASK_*: | |
129 | * | |
130 | * is changed locklessly using set_current_state(), __set_current_state() or | |
131 | * set_special_state(), see their respective comments, or by | |
132 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
133 | * concurrent self. | |
134 | * | |
135 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
136 | * | |
137 | * is set by activate_task() and cleared by deactivate_task(), under | |
138 | * rq->lock. Non-zero indicates the task is runnable, the special | |
139 | * ON_RQ_MIGRATING state is used for migration without holding both | |
140 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
141 | * | |
142 | * p->on_cpu <- { 0, 1 }: | |
143 | * | |
144 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
145 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
146 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
147 | * | |
148 | * [ The astute reader will observe that it is possible for two tasks on one | |
149 | * CPU to have ->on_cpu = 1 at the same time. ] | |
150 | * | |
151 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
152 | * | |
153 | * - Don't call set_task_cpu() on a blocked task: | |
154 | * | |
155 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
156 | * the CPU assignment of blocked tasks isn't required to be valid. | |
157 | * | |
158 | * - for try_to_wake_up(), called under p->pi_lock: | |
159 | * | |
160 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
161 | * | |
162 | * - for migration called under rq->lock: | |
163 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
164 | * | |
165 | * o move_queued_task() | |
166 | * o detach_task() | |
167 | * | |
168 | * - for migration called under double_rq_lock(): | |
169 | * | |
170 | * o __migrate_swap_task() | |
171 | * o push_rt_task() / pull_rt_task() | |
172 | * o push_dl_task() / pull_dl_task() | |
173 | * o dl_task_offline_migration() | |
174 | * | |
175 | */ | |
176 | ||
3e71a462 PZ |
177 | /* |
178 | * __task_rq_lock - lock the rq @p resides on. | |
179 | */ | |
eb580751 | 180 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
181 | __acquires(rq->lock) |
182 | { | |
183 | struct rq *rq; | |
184 | ||
185 | lockdep_assert_held(&p->pi_lock); | |
186 | ||
187 | for (;;) { | |
188 | rq = task_rq(p); | |
189 | raw_spin_lock(&rq->lock); | |
190 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 191 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
192 | return rq; |
193 | } | |
194 | raw_spin_unlock(&rq->lock); | |
195 | ||
196 | while (unlikely(task_on_rq_migrating(p))) | |
197 | cpu_relax(); | |
198 | } | |
199 | } | |
200 | ||
201 | /* | |
202 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
203 | */ | |
eb580751 | 204 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
205 | __acquires(p->pi_lock) |
206 | __acquires(rq->lock) | |
207 | { | |
208 | struct rq *rq; | |
209 | ||
210 | for (;;) { | |
eb580751 | 211 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
212 | rq = task_rq(p); |
213 | raw_spin_lock(&rq->lock); | |
214 | /* | |
215 | * move_queued_task() task_rq_lock() | |
216 | * | |
217 | * ACQUIRE (rq->lock) | |
218 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
219 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
220 | * [S] ->cpu = new_cpu [L] task_rq() | |
221 | * [L] ->on_rq | |
222 | * RELEASE (rq->lock) | |
223 | * | |
c546951d | 224 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
225 | * the old rq->lock will fully serialize against the stores. |
226 | * | |
c546951d AP |
227 | * If we observe the new CPU in task_rq_lock(), the address |
228 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
229 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
230 | */ |
231 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 232 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
233 | return rq; |
234 | } | |
235 | raw_spin_unlock(&rq->lock); | |
eb580751 | 236 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
237 | |
238 | while (unlikely(task_on_rq_migrating(p))) | |
239 | cpu_relax(); | |
240 | } | |
241 | } | |
242 | ||
535b9552 IM |
243 | /* |
244 | * RQ-clock updating methods: | |
245 | */ | |
246 | ||
247 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
248 | { | |
249 | /* | |
250 | * In theory, the compile should just see 0 here, and optimize out the call | |
251 | * to sched_rt_avg_update. But I don't trust it... | |
252 | */ | |
11d4afd4 VG |
253 | s64 __maybe_unused steal = 0, irq_delta = 0; |
254 | ||
535b9552 IM |
255 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
256 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
257 | ||
258 | /* | |
259 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
260 | * this case when a previous update_rq_clock() happened inside a | |
261 | * {soft,}irq region. | |
262 | * | |
263 | * When this happens, we stop ->clock_task and only update the | |
264 | * prev_irq_time stamp to account for the part that fit, so that a next | |
265 | * update will consume the rest. This ensures ->clock_task is | |
266 | * monotonic. | |
267 | * | |
268 | * It does however cause some slight miss-attribution of {soft,}irq | |
269 | * time, a more accurate solution would be to update the irq_time using | |
270 | * the current rq->clock timestamp, except that would require using | |
271 | * atomic ops. | |
272 | */ | |
273 | if (irq_delta > delta) | |
274 | irq_delta = delta; | |
275 | ||
276 | rq->prev_irq_time += irq_delta; | |
277 | delta -= irq_delta; | |
278 | #endif | |
279 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
280 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
281 | steal = paravirt_steal_clock(cpu_of(rq)); | |
282 | steal -= rq->prev_steal_time_rq; | |
283 | ||
284 | if (unlikely(steal > delta)) | |
285 | steal = delta; | |
286 | ||
287 | rq->prev_steal_time_rq += steal; | |
288 | delta -= steal; | |
289 | } | |
290 | #endif | |
291 | ||
292 | rq->clock_task += delta; | |
293 | ||
11d4afd4 | 294 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 295 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 296 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 297 | #endif |
23127296 | 298 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
299 | } |
300 | ||
301 | void update_rq_clock(struct rq *rq) | |
302 | { | |
303 | s64 delta; | |
304 | ||
305 | lockdep_assert_held(&rq->lock); | |
306 | ||
307 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
308 | return; | |
309 | ||
310 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
311 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
312 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
313 | rq->clock_update_flags |= RQCF_UPDATED; |
314 | #endif | |
26ae58d2 | 315 | |
535b9552 IM |
316 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
317 | if (delta < 0) | |
318 | return; | |
319 | rq->clock += delta; | |
320 | update_rq_clock_task(rq, delta); | |
321 | } | |
322 | ||
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 | |
b19a888c | 521 | * it's already queued (either by us or someone else) and will get the |
76751049 PZ |
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 | /* |
b19a888c | 772 | * If there are more than one RR tasks, we need the tick to affect the |
2548d546 | 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 | * | |
b19a888c | 1190 | * Need to be careful of the following enqueue/dequeue ordering |
46609ce2 QY |
1191 | * problem too |
1192 | * | |
1193 | * enqueue(taskA) | |
1194 | * // sched_uclamp_used gets enabled | |
1195 | * enqueue(taskB) | |
1196 | * dequeue(taskA) | |
b19a888c | 1197 | * // Must not decrement bucket->tasks here |
46609ce2 QY |
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 | { | |
480a6ca2 DE |
1416 | int util_min = p->uclamp_req[UCLAMP_MIN].value; |
1417 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd | 1418 | |
480a6ca2 DE |
1419 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { |
1420 | util_min = attr->sched_util_min; | |
a509a7cd | 1421 | |
480a6ca2 DE |
1422 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) |
1423 | return -EINVAL; | |
1424 | } | |
1425 | ||
1426 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1427 | util_max = attr->sched_util_max; | |
1428 | ||
1429 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | |
1430 | return -EINVAL; | |
1431 | } | |
1432 | ||
1433 | if (util_min != -1 && util_max != -1 && util_min > util_max) | |
a509a7cd PB |
1434 | return -EINVAL; |
1435 | ||
e65855a5 QY |
1436 | /* |
1437 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1438 | * | |
1439 | * We need to do that here, because enabling static branches is a | |
1440 | * blocking operation which obviously cannot be done while holding | |
1441 | * scheduler locks. | |
1442 | */ | |
1443 | static_branch_enable(&sched_uclamp_used); | |
1444 | ||
a509a7cd PB |
1445 | return 0; |
1446 | } | |
1447 | ||
480a6ca2 DE |
1448 | static bool uclamp_reset(const struct sched_attr *attr, |
1449 | enum uclamp_id clamp_id, | |
1450 | struct uclamp_se *uc_se) | |
1451 | { | |
1452 | /* Reset on sched class change for a non user-defined clamp value. */ | |
1453 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | |
1454 | !uc_se->user_defined) | |
1455 | return true; | |
1456 | ||
1457 | /* Reset on sched_util_{min,max} == -1. */ | |
1458 | if (clamp_id == UCLAMP_MIN && | |
1459 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | |
1460 | attr->sched_util_min == -1) { | |
1461 | return true; | |
1462 | } | |
1463 | ||
1464 | if (clamp_id == UCLAMP_MAX && | |
1465 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | |
1466 | attr->sched_util_max == -1) { | |
1467 | return true; | |
1468 | } | |
1469 | ||
1470 | return false; | |
1471 | } | |
1472 | ||
a509a7cd PB |
1473 | static void __setscheduler_uclamp(struct task_struct *p, |
1474 | const struct sched_attr *attr) | |
1475 | { | |
0413d7f3 | 1476 | enum uclamp_id clamp_id; |
1a00d999 | 1477 | |
1a00d999 PB |
1478 | for_each_clamp_id(clamp_id) { |
1479 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
480a6ca2 | 1480 | unsigned int value; |
1a00d999 | 1481 | |
480a6ca2 | 1482 | if (!uclamp_reset(attr, clamp_id, uc_se)) |
1a00d999 PB |
1483 | continue; |
1484 | ||
13685c4a QY |
1485 | /* |
1486 | * RT by default have a 100% boost value that could be modified | |
1487 | * at runtime. | |
1488 | */ | |
1a00d999 | 1489 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
480a6ca2 | 1490 | value = sysctl_sched_uclamp_util_min_rt_default; |
13685c4a | 1491 | else |
480a6ca2 DE |
1492 | value = uclamp_none(clamp_id); |
1493 | ||
1494 | uclamp_se_set(uc_se, value, false); | |
1a00d999 | 1495 | |
1a00d999 PB |
1496 | } |
1497 | ||
a509a7cd PB |
1498 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1499 | return; | |
1500 | ||
480a6ca2 DE |
1501 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && |
1502 | attr->sched_util_min != -1) { | |
a509a7cd PB |
1503 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], |
1504 | attr->sched_util_min, true); | |
1505 | } | |
1506 | ||
480a6ca2 DE |
1507 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && |
1508 | attr->sched_util_max != -1) { | |
a509a7cd PB |
1509 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], |
1510 | attr->sched_util_max, true); | |
1511 | } | |
1512 | } | |
1513 | ||
e8f14172 PB |
1514 | static void uclamp_fork(struct task_struct *p) |
1515 | { | |
0413d7f3 | 1516 | enum uclamp_id clamp_id; |
e8f14172 | 1517 | |
13685c4a QY |
1518 | /* |
1519 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1520 | * as the task is still at its early fork stages. | |
1521 | */ | |
e8f14172 PB |
1522 | for_each_clamp_id(clamp_id) |
1523 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1524 | |
1525 | if (likely(!p->sched_reset_on_fork)) | |
1526 | return; | |
1527 | ||
1528 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1529 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1530 | uclamp_none(clamp_id), false); | |
a87498ac | 1531 | } |
e8f14172 PB |
1532 | } |
1533 | ||
13685c4a QY |
1534 | static void uclamp_post_fork(struct task_struct *p) |
1535 | { | |
1536 | uclamp_update_util_min_rt_default(p); | |
1537 | } | |
1538 | ||
d81ae8aa QY |
1539 | static void __init init_uclamp_rq(struct rq *rq) |
1540 | { | |
1541 | enum uclamp_id clamp_id; | |
1542 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1543 | ||
1544 | for_each_clamp_id(clamp_id) { | |
1545 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1546 | .value = uclamp_none(clamp_id) | |
1547 | }; | |
1548 | } | |
1549 | ||
1550 | rq->uclamp_flags = 0; | |
1551 | } | |
1552 | ||
69842cba PB |
1553 | static void __init init_uclamp(void) |
1554 | { | |
e8f14172 | 1555 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1556 | enum uclamp_id clamp_id; |
69842cba PB |
1557 | int cpu; |
1558 | ||
d81ae8aa QY |
1559 | for_each_possible_cpu(cpu) |
1560 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1561 | |
69842cba | 1562 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1563 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1564 | uclamp_none(clamp_id), false); |
69842cba | 1565 | } |
e8f14172 PB |
1566 | |
1567 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1568 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1569 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1570 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1571 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1572 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1573 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1574 | #endif |
1575 | } | |
69842cba PB |
1576 | } |
1577 | ||
1578 | #else /* CONFIG_UCLAMP_TASK */ | |
1579 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1580 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1581 | static inline int uclamp_validate(struct task_struct *p, |
1582 | const struct sched_attr *attr) | |
1583 | { | |
1584 | return -EOPNOTSUPP; | |
1585 | } | |
1586 | static void __setscheduler_uclamp(struct task_struct *p, | |
1587 | const struct sched_attr *attr) { } | |
e8f14172 | 1588 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1589 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1590 | static inline void init_uclamp(void) { } |
1591 | #endif /* CONFIG_UCLAMP_TASK */ | |
1592 | ||
1de64443 | 1593 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1594 | { |
0a67d1ee PZ |
1595 | if (!(flags & ENQUEUE_NOCLOCK)) |
1596 | update_rq_clock(rq); | |
1597 | ||
eb414681 | 1598 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 1599 | sched_info_queued(rq, p); |
eb414681 JW |
1600 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1601 | } | |
0a67d1ee | 1602 | |
69842cba | 1603 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1604 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1605 | } |
1606 | ||
1de64443 | 1607 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1608 | { |
0a67d1ee PZ |
1609 | if (!(flags & DEQUEUE_NOCLOCK)) |
1610 | update_rq_clock(rq); | |
1611 | ||
eb414681 | 1612 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 1613 | sched_info_dequeued(rq, p); |
eb414681 JW |
1614 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1615 | } | |
0a67d1ee | 1616 | |
69842cba | 1617 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1618 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1619 | } |
1620 | ||
029632fb | 1621 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1622 | { |
371fd7e7 | 1623 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
1624 | |
1625 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
1626 | } |
1627 | ||
029632fb | 1628 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1629 | { |
7dd77884 PZ |
1630 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
1631 | ||
371fd7e7 | 1632 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1633 | } |
1634 | ||
14531189 | 1635 | /* |
dd41f596 | 1636 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1637 | */ |
14531189 IM |
1638 | static inline int __normal_prio(struct task_struct *p) |
1639 | { | |
dd41f596 | 1640 | return p->static_prio; |
14531189 IM |
1641 | } |
1642 | ||
b29739f9 IM |
1643 | /* |
1644 | * Calculate the expected normal priority: i.e. priority | |
1645 | * without taking RT-inheritance into account. Might be | |
1646 | * boosted by interactivity modifiers. Changes upon fork, | |
1647 | * setprio syscalls, and whenever the interactivity | |
1648 | * estimator recalculates. | |
1649 | */ | |
36c8b586 | 1650 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1651 | { |
1652 | int prio; | |
1653 | ||
aab03e05 DF |
1654 | if (task_has_dl_policy(p)) |
1655 | prio = MAX_DL_PRIO-1; | |
1656 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
1657 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1658 | else | |
1659 | prio = __normal_prio(p); | |
1660 | return prio; | |
1661 | } | |
1662 | ||
1663 | /* | |
1664 | * Calculate the current priority, i.e. the priority | |
1665 | * taken into account by the scheduler. This value might | |
1666 | * be boosted by RT tasks, or might be boosted by | |
1667 | * interactivity modifiers. Will be RT if the task got | |
1668 | * RT-boosted. If not then it returns p->normal_prio. | |
1669 | */ | |
36c8b586 | 1670 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1671 | { |
1672 | p->normal_prio = normal_prio(p); | |
1673 | /* | |
1674 | * If we are RT tasks or we were boosted to RT priority, | |
1675 | * keep the priority unchanged. Otherwise, update priority | |
1676 | * to the normal priority: | |
1677 | */ | |
1678 | if (!rt_prio(p->prio)) | |
1679 | return p->normal_prio; | |
1680 | return p->prio; | |
1681 | } | |
1682 | ||
1da177e4 LT |
1683 | /** |
1684 | * task_curr - is this task currently executing on a CPU? | |
1685 | * @p: the task in question. | |
e69f6186 YB |
1686 | * |
1687 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1688 | */ |
36c8b586 | 1689 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1690 | { |
1691 | return cpu_curr(task_cpu(p)) == p; | |
1692 | } | |
1693 | ||
67dfa1b7 | 1694 | /* |
4c9a4bc8 PZ |
1695 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1696 | * use the balance_callback list if you want balancing. | |
1697 | * | |
1698 | * this means any call to check_class_changed() must be followed by a call to | |
1699 | * balance_callback(). | |
67dfa1b7 | 1700 | */ |
cb469845 SR |
1701 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1702 | const struct sched_class *prev_class, | |
da7a735e | 1703 | int oldprio) |
cb469845 SR |
1704 | { |
1705 | if (prev_class != p->sched_class) { | |
1706 | if (prev_class->switched_from) | |
da7a735e | 1707 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1708 | |
da7a735e | 1709 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1710 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1711 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1712 | } |
1713 | ||
029632fb | 1714 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 1715 | { |
aa93cd53 | 1716 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 1717 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
1718 | else if (p->sched_class > rq->curr->sched_class) |
1719 | resched_curr(rq); | |
1e5a7405 PZ |
1720 | |
1721 | /* | |
1722 | * A queue event has occurred, and we're going to schedule. In | |
1723 | * this case, we can save a useless back to back clock update. | |
1724 | */ | |
da0c1e65 | 1725 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 1726 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
1727 | } |
1728 | ||
1da177e4 | 1729 | #ifdef CONFIG_SMP |
175f0e25 | 1730 | |
af449901 PZ |
1731 | static void |
1732 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags); | |
1733 | ||
1734 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
1735 | const struct cpumask *new_mask, | |
1736 | u32 flags); | |
1737 | ||
1738 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
1739 | { | |
1740 | if (likely(!p->migration_disabled)) | |
1741 | return; | |
1742 | ||
1743 | if (p->cpus_ptr != &p->cpus_mask) | |
1744 | return; | |
1745 | ||
1746 | /* | |
1747 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
1748 | */ | |
1749 | __do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE); | |
1750 | } | |
1751 | ||
1752 | void migrate_disable(void) | |
1753 | { | |
3015ef4b TG |
1754 | struct task_struct *p = current; |
1755 | ||
1756 | if (p->migration_disabled) { | |
1757 | p->migration_disabled++; | |
af449901 | 1758 | return; |
3015ef4b | 1759 | } |
af449901 | 1760 | |
3015ef4b TG |
1761 | preempt_disable(); |
1762 | this_rq()->nr_pinned++; | |
1763 | p->migration_disabled = 1; | |
1764 | preempt_enable(); | |
af449901 PZ |
1765 | } |
1766 | EXPORT_SYMBOL_GPL(migrate_disable); | |
1767 | ||
1768 | void migrate_enable(void) | |
1769 | { | |
1770 | struct task_struct *p = current; | |
1771 | ||
6d337eab PZ |
1772 | if (p->migration_disabled > 1) { |
1773 | p->migration_disabled--; | |
af449901 | 1774 | return; |
6d337eab | 1775 | } |
af449901 | 1776 | |
6d337eab PZ |
1777 | /* |
1778 | * Ensure stop_task runs either before or after this, and that | |
1779 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
1780 | */ | |
1781 | preempt_disable(); | |
1782 | if (p->cpus_ptr != &p->cpus_mask) | |
1783 | __set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE); | |
1784 | /* | |
1785 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
1786 | * regular cpus_mask, otherwise things that race (eg. | |
1787 | * select_fallback_rq) get confused. | |
1788 | */ | |
af449901 | 1789 | barrier(); |
6d337eab | 1790 | p->migration_disabled = 0; |
3015ef4b | 1791 | this_rq()->nr_pinned--; |
6d337eab | 1792 | preempt_enable(); |
af449901 PZ |
1793 | } |
1794 | EXPORT_SYMBOL_GPL(migrate_enable); | |
1795 | ||
3015ef4b TG |
1796 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
1797 | { | |
1798 | return rq->nr_pinned; | |
1799 | } | |
1800 | ||
175f0e25 | 1801 | /* |
bee98539 | 1802 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
1803 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
1804 | */ | |
1805 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
1806 | { | |
3bd37062 | 1807 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
1808 | return false; |
1809 | ||
af449901 | 1810 | if (is_per_cpu_kthread(p) || is_migration_disabled(p)) |
175f0e25 PZ |
1811 | return cpu_online(cpu); |
1812 | ||
1813 | return cpu_active(cpu); | |
1814 | } | |
1815 | ||
5cc389bc PZ |
1816 | /* |
1817 | * This is how migration works: | |
1818 | * | |
1819 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1820 | * stop_one_cpu(). | |
1821 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1822 | * off the CPU) | |
1823 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1824 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1825 | * it and puts it into the right queue. | |
1826 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1827 | * is done. | |
1828 | */ | |
1829 | ||
1830 | /* | |
1831 | * move_queued_task - move a queued task to new rq. | |
1832 | * | |
1833 | * Returns (locked) new rq. Old rq's lock is released. | |
1834 | */ | |
8a8c69c3 PZ |
1835 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
1836 | struct task_struct *p, int new_cpu) | |
5cc389bc | 1837 | { |
5cc389bc PZ |
1838 | lockdep_assert_held(&rq->lock); |
1839 | ||
58877d34 | 1840 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 1841 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 1842 | rq_unlock(rq, rf); |
5cc389bc PZ |
1843 | |
1844 | rq = cpu_rq(new_cpu); | |
1845 | ||
8a8c69c3 | 1846 | rq_lock(rq, rf); |
5cc389bc | 1847 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 1848 | activate_task(rq, p, 0); |
5cc389bc PZ |
1849 | check_preempt_curr(rq, p, 0); |
1850 | ||
1851 | return rq; | |
1852 | } | |
1853 | ||
1854 | struct migration_arg { | |
6d337eab PZ |
1855 | struct task_struct *task; |
1856 | int dest_cpu; | |
1857 | struct set_affinity_pending *pending; | |
1858 | }; | |
1859 | ||
1860 | struct set_affinity_pending { | |
1861 | refcount_t refs; | |
1862 | struct completion done; | |
1863 | struct cpu_stop_work stop_work; | |
1864 | struct migration_arg arg; | |
5cc389bc PZ |
1865 | }; |
1866 | ||
1867 | /* | |
d1ccc66d | 1868 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
1869 | * this because either it can't run here any more (set_cpus_allowed() |
1870 | * away from this CPU, or CPU going down), or because we're | |
1871 | * attempting to rebalance this task on exec (sched_exec). | |
1872 | * | |
1873 | * So we race with normal scheduler movements, but that's OK, as long | |
1874 | * as the task is no longer on this CPU. | |
5cc389bc | 1875 | */ |
8a8c69c3 PZ |
1876 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
1877 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 1878 | { |
5cc389bc | 1879 | /* Affinity changed (again). */ |
175f0e25 | 1880 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 1881 | return rq; |
5cc389bc | 1882 | |
15ff991e | 1883 | update_rq_clock(rq); |
8a8c69c3 | 1884 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
1885 | |
1886 | return rq; | |
5cc389bc PZ |
1887 | } |
1888 | ||
1889 | /* | |
1890 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1891 | * and performs thread migration by bumping thread off CPU then | |
1892 | * 'pushing' onto another runqueue. | |
1893 | */ | |
1894 | static int migration_cpu_stop(void *data) | |
1895 | { | |
6d337eab | 1896 | struct set_affinity_pending *pending; |
5cc389bc | 1897 | struct migration_arg *arg = data; |
5e16bbc2 | 1898 | struct task_struct *p = arg->task; |
6d337eab | 1899 | int dest_cpu = arg->dest_cpu; |
5e16bbc2 | 1900 | struct rq *rq = this_rq(); |
6d337eab | 1901 | bool complete = false; |
8a8c69c3 | 1902 | struct rq_flags rf; |
5cc389bc PZ |
1903 | |
1904 | /* | |
d1ccc66d IM |
1905 | * The original target CPU might have gone down and we might |
1906 | * be on another CPU but it doesn't matter. | |
5cc389bc | 1907 | */ |
6d337eab | 1908 | local_irq_save(rf.flags); |
5cc389bc PZ |
1909 | /* |
1910 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 1911 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
1912 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
1913 | */ | |
a1488664 | 1914 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
1915 | |
1916 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1917 | rq_lock(rq, &rf); |
6d337eab PZ |
1918 | |
1919 | pending = p->migration_pending; | |
5e16bbc2 PZ |
1920 | /* |
1921 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1922 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1923 | * we're holding p->pi_lock. | |
1924 | */ | |
bf89a304 | 1925 | if (task_rq(p) == rq) { |
6d337eab PZ |
1926 | if (is_migration_disabled(p)) |
1927 | goto out; | |
1928 | ||
1929 | if (pending) { | |
1930 | p->migration_pending = NULL; | |
1931 | complete = true; | |
1932 | } | |
1933 | ||
1934 | /* migrate_enable() -- we must not race against SCA */ | |
1935 | if (dest_cpu < 0) { | |
1936 | /* | |
1937 | * When this was migrate_enable() but we no longer | |
1938 | * have a @pending, a concurrent SCA 'fixed' things | |
1939 | * and we should be valid again. Nothing to do. | |
1940 | */ | |
1941 | if (!pending) { | |
1293771e | 1942 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
1943 | goto out; |
1944 | } | |
1945 | ||
1946 | dest_cpu = cpumask_any_distribute(&p->cpus_mask); | |
1947 | } | |
1948 | ||
bf89a304 | 1949 | if (task_on_rq_queued(p)) |
6d337eab | 1950 | rq = __migrate_task(rq, &rf, p, dest_cpu); |
bf89a304 | 1951 | else |
6d337eab PZ |
1952 | p->wake_cpu = dest_cpu; |
1953 | ||
d707faa6 | 1954 | } else if (dest_cpu < 0 || pending) { |
6d337eab PZ |
1955 | /* |
1956 | * This happens when we get migrated between migrate_enable()'s | |
1957 | * preempt_enable() and scheduling the stopper task. At that | |
1958 | * point we're a regular task again and not current anymore. | |
1959 | * | |
1960 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
1961 | * more likely. | |
1962 | */ | |
1963 | ||
d707faa6 VS |
1964 | /* |
1965 | * The task moved before the stopper got to run. We're holding | |
1966 | * ->pi_lock, so the allowed mask is stable - if it got | |
1967 | * somewhere allowed, we're done. | |
1968 | */ | |
1969 | if (pending && cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { | |
1970 | p->migration_pending = NULL; | |
1971 | complete = true; | |
1972 | goto out; | |
1973 | } | |
1974 | ||
6d337eab PZ |
1975 | /* |
1976 | * When this was migrate_enable() but we no longer have an | |
1977 | * @pending, a concurrent SCA 'fixed' things and we should be | |
1978 | * valid again. Nothing to do. | |
1979 | */ | |
1980 | if (!pending) { | |
1293771e | 1981 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
1982 | goto out; |
1983 | } | |
1984 | ||
1985 | /* | |
1986 | * When migrate_enable() hits a rq mis-match we can't reliably | |
1987 | * determine is_migration_disabled() and so have to chase after | |
1988 | * it. | |
1989 | */ | |
1990 | task_rq_unlock(rq, p, &rf); | |
1991 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
1992 | &pending->arg, &pending->stop_work); | |
1993 | return 0; | |
bf89a304 | 1994 | } |
6d337eab PZ |
1995 | out: |
1996 | task_rq_unlock(rq, p, &rf); | |
1997 | ||
1998 | if (complete) | |
1999 | complete_all(&pending->done); | |
2000 | ||
2001 | /* For pending->{arg,stop_work} */ | |
2002 | pending = arg->pending; | |
2003 | if (pending && refcount_dec_and_test(&pending->refs)) | |
2004 | wake_up_var(&pending->refs); | |
5e16bbc2 | 2005 | |
5cc389bc PZ |
2006 | return 0; |
2007 | } | |
2008 | ||
a7c81556 PZ |
2009 | int push_cpu_stop(void *arg) |
2010 | { | |
2011 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2012 | struct task_struct *p = arg; | |
2013 | ||
2014 | raw_spin_lock_irq(&p->pi_lock); | |
2015 | raw_spin_lock(&rq->lock); | |
2016 | ||
2017 | if (task_rq(p) != rq) | |
2018 | goto out_unlock; | |
2019 | ||
2020 | if (is_migration_disabled(p)) { | |
2021 | p->migration_flags |= MDF_PUSH; | |
2022 | goto out_unlock; | |
2023 | } | |
2024 | ||
2025 | p->migration_flags &= ~MDF_PUSH; | |
2026 | ||
2027 | if (p->sched_class->find_lock_rq) | |
2028 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2029 | |
a7c81556 PZ |
2030 | if (!lowest_rq) |
2031 | goto out_unlock; | |
2032 | ||
2033 | // XXX validate p is still the highest prio task | |
2034 | if (task_rq(p) == rq) { | |
2035 | deactivate_task(rq, p, 0); | |
2036 | set_task_cpu(p, lowest_rq->cpu); | |
2037 | activate_task(lowest_rq, p, 0); | |
2038 | resched_curr(lowest_rq); | |
2039 | } | |
2040 | ||
2041 | double_unlock_balance(rq, lowest_rq); | |
2042 | ||
2043 | out_unlock: | |
2044 | rq->push_busy = false; | |
2045 | raw_spin_unlock(&rq->lock); | |
2046 | raw_spin_unlock_irq(&p->pi_lock); | |
2047 | ||
2048 | put_task_struct(p); | |
5cc389bc PZ |
2049 | return 0; |
2050 | } | |
2051 | ||
c5b28038 PZ |
2052 | /* |
2053 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2054 | * actually call this function. | |
2055 | */ | |
9cfc3e18 | 2056 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 2057 | { |
af449901 PZ |
2058 | if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2059 | p->cpus_ptr = new_mask; | |
2060 | return; | |
2061 | } | |
2062 | ||
3bd37062 | 2063 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
2064 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
2065 | } | |
2066 | ||
9cfc3e18 PZ |
2067 | static void |
2068 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 2069 | { |
6c37067e PZ |
2070 | struct rq *rq = task_rq(p); |
2071 | bool queued, running; | |
2072 | ||
af449901 PZ |
2073 | /* |
2074 | * This here violates the locking rules for affinity, since we're only | |
2075 | * supposed to change these variables while holding both rq->lock and | |
2076 | * p->pi_lock. | |
2077 | * | |
2078 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2079 | * accesses these variables under p->pi_lock and only does so after | |
2080 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2081 | * before finish_task(). | |
2082 | * | |
2083 | * XXX do further audits, this smells like something putrid. | |
2084 | */ | |
2085 | if (flags & SCA_MIGRATE_DISABLE) | |
2086 | SCHED_WARN_ON(!p->on_cpu); | |
2087 | else | |
2088 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2089 | |
2090 | queued = task_on_rq_queued(p); | |
2091 | running = task_current(rq, p); | |
2092 | ||
2093 | if (queued) { | |
2094 | /* | |
2095 | * Because __kthread_bind() calls this on blocked tasks without | |
2096 | * holding rq->lock. | |
2097 | */ | |
2098 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 2099 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2100 | } |
2101 | if (running) | |
2102 | put_prev_task(rq, p); | |
2103 | ||
9cfc3e18 | 2104 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 2105 | |
6c37067e | 2106 | if (queued) |
7134b3e9 | 2107 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2108 | if (running) |
03b7fad1 | 2109 | set_next_task(rq, p); |
c5b28038 PZ |
2110 | } |
2111 | ||
9cfc3e18 PZ |
2112 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2113 | { | |
2114 | __do_set_cpus_allowed(p, new_mask, 0); | |
2115 | } | |
2116 | ||
6d337eab | 2117 | /* |
c777d847 VS |
2118 | * This function is wildly self concurrent; here be dragons. |
2119 | * | |
2120 | * | |
2121 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2122 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2123 | * running, we have to kick it out using the CPU stopper. | |
2124 | * | |
2125 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2126 | * Consider: | |
2127 | * | |
2128 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2129 | * | |
2130 | * P0@CPU0 P1 | |
2131 | * | |
2132 | * migrate_disable(); | |
2133 | * <preempted> | |
2134 | * set_cpus_allowed_ptr(P0, [1]); | |
2135 | * | |
2136 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2137 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2138 | * This means we need the following scheme: | |
2139 | * | |
2140 | * P0@CPU0 P1 | |
2141 | * | |
2142 | * migrate_disable(); | |
2143 | * <preempted> | |
2144 | * set_cpus_allowed_ptr(P0, [1]); | |
2145 | * <blocks> | |
2146 | * <resumes> | |
2147 | * migrate_enable(); | |
2148 | * __set_cpus_allowed_ptr(); | |
2149 | * <wakes local stopper> | |
2150 | * `--> <woken on migration completion> | |
2151 | * | |
2152 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2153 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2154 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2155 | * should come into effect at the end of the Migrate-Disable region is the last | |
2156 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2157 | * but we still need to properly signal those waiting tasks at the appropriate | |
2158 | * moment. | |
2159 | * | |
2160 | * This is implemented using struct set_affinity_pending. The first | |
2161 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2162 | * setup an instance of that struct and install it on the targeted task_struct. | |
2163 | * Any and all further callers will reuse that instance. Those then wait for | |
2164 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2165 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2166 | * | |
2167 | * | |
2168 | * (1) In the cases covered above. There is one more where the completion is | |
2169 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
2170 | * cancels the need for an active migration. Consider: | |
2171 | * | |
2172 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2173 | * | |
2174 | * P0@CPU0 P1 P2 | |
2175 | * | |
2176 | * migrate_disable(); | |
2177 | * <preempted> | |
2178 | * set_cpus_allowed_ptr(P0, [1]); | |
2179 | * <blocks> | |
2180 | * set_cpus_allowed_ptr(P0, [0, 1]); | |
2181 | * <signal completion> | |
2182 | * <awakes> | |
2183 | * | |
2184 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2185 | * pending affinity completion is preceded by an uninstallation of | |
2186 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2187 | */ |
2188 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2189 | int dest_cpu, unsigned int flags) | |
2190 | { | |
2191 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
2192 | struct migration_arg arg = { | |
2193 | .task = p, | |
2194 | .dest_cpu = dest_cpu, | |
2195 | }; | |
2196 | bool complete = false; | |
2197 | ||
2198 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2199 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2200 | struct task_struct *push_task = NULL; |
2201 | ||
2202 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2203 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2204 | rq->push_busy = true; | |
2205 | push_task = get_task_struct(p); | |
2206 | } | |
2207 | ||
6d337eab PZ |
2208 | pending = p->migration_pending; |
2209 | if (pending) { | |
2210 | refcount_inc(&pending->refs); | |
2211 | p->migration_pending = NULL; | |
2212 | complete = true; | |
2213 | } | |
2214 | task_rq_unlock(rq, p, rf); | |
2215 | ||
a7c81556 PZ |
2216 | if (push_task) { |
2217 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2218 | p, &rq->push_work); | |
2219 | } | |
2220 | ||
6d337eab PZ |
2221 | if (complete) |
2222 | goto do_complete; | |
2223 | ||
2224 | return 0; | |
2225 | } | |
2226 | ||
2227 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2228 | /* serialized by p->pi_lock */ | |
2229 | if (!p->migration_pending) { | |
c777d847 | 2230 | /* Install the request */ |
6d337eab PZ |
2231 | refcount_set(&my_pending.refs, 1); |
2232 | init_completion(&my_pending.done); | |
2233 | p->migration_pending = &my_pending; | |
2234 | } else { | |
2235 | pending = p->migration_pending; | |
2236 | refcount_inc(&pending->refs); | |
2237 | } | |
2238 | } | |
2239 | pending = p->migration_pending; | |
2240 | /* | |
2241 | * - !MIGRATE_ENABLE: | |
2242 | * we'll have installed a pending if there wasn't one already. | |
2243 | * | |
2244 | * - MIGRATE_ENABLE: | |
2245 | * we're here because the current CPU isn't matching anymore, | |
2246 | * the only way that can happen is because of a concurrent | |
2247 | * set_cpus_allowed_ptr() call, which should then still be | |
2248 | * pending completion. | |
2249 | * | |
2250 | * Either way, we really should have a @pending here. | |
2251 | */ | |
2252 | if (WARN_ON_ONCE(!pending)) { | |
2253 | task_rq_unlock(rq, p, rf); | |
2254 | return -EINVAL; | |
2255 | } | |
2256 | ||
2257 | if (flags & SCA_MIGRATE_ENABLE) { | |
2258 | ||
2259 | refcount_inc(&pending->refs); /* pending->{arg,stop_work} */ | |
a7c81556 | 2260 | p->migration_flags &= ~MDF_PUSH; |
6d337eab PZ |
2261 | task_rq_unlock(rq, p, rf); |
2262 | ||
2263 | pending->arg = (struct migration_arg) { | |
2264 | .task = p, | |
2265 | .dest_cpu = -1, | |
2266 | .pending = pending, | |
2267 | }; | |
2268 | ||
2269 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2270 | &pending->arg, &pending->stop_work); | |
2271 | ||
2272 | return 0; | |
2273 | } | |
2274 | ||
2275 | if (task_running(rq, p) || p->state == TASK_WAKING) { | |
c777d847 VS |
2276 | /* |
2277 | * Lessen races (and headaches) by delegating | |
2278 | * is_migration_disabled(p) checks to the stopper, which will | |
2279 | * run on the same CPU as said p. | |
2280 | */ | |
6d337eab PZ |
2281 | task_rq_unlock(rq, p, rf); |
2282 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | |
2283 | ||
2284 | } else { | |
2285 | ||
2286 | if (!is_migration_disabled(p)) { | |
2287 | if (task_on_rq_queued(p)) | |
2288 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2289 | ||
2290 | p->migration_pending = NULL; | |
2291 | complete = true; | |
2292 | } | |
2293 | task_rq_unlock(rq, p, rf); | |
2294 | ||
2295 | do_complete: | |
2296 | if (complete) | |
2297 | complete_all(&pending->done); | |
2298 | } | |
2299 | ||
2300 | wait_for_completion(&pending->done); | |
2301 | ||
2302 | if (refcount_dec_and_test(&pending->refs)) | |
2303 | wake_up_var(&pending->refs); | |
2304 | ||
c777d847 VS |
2305 | /* |
2306 | * Block the original owner of &pending until all subsequent callers | |
2307 | * have seen the completion and decremented the refcount | |
2308 | */ | |
6d337eab PZ |
2309 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2310 | ||
2311 | return 0; | |
2312 | } | |
2313 | ||
5cc389bc PZ |
2314 | /* |
2315 | * Change a given task's CPU affinity. Migrate the thread to a | |
2316 | * proper CPU and schedule it away if the CPU it's executing on | |
2317 | * is removed from the allowed bitmask. | |
2318 | * | |
2319 | * NOTE: the caller must have a valid reference to the task, the | |
2320 | * task must not exit() & deallocate itself prematurely. The | |
2321 | * call is not atomic; no spinlocks may be held. | |
2322 | */ | |
25834c73 | 2323 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
9cfc3e18 PZ |
2324 | const struct cpumask *new_mask, |
2325 | u32 flags) | |
5cc389bc | 2326 | { |
e9d867a6 | 2327 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 2328 | unsigned int dest_cpu; |
eb580751 PZ |
2329 | struct rq_flags rf; |
2330 | struct rq *rq; | |
5cc389bc PZ |
2331 | int ret = 0; |
2332 | ||
eb580751 | 2333 | rq = task_rq_lock(p, &rf); |
a499c3ea | 2334 | update_rq_clock(rq); |
5cc389bc | 2335 | |
af449901 | 2336 | if (p->flags & PF_KTHREAD || is_migration_disabled(p)) { |
e9d867a6 | 2337 | /* |
af449901 PZ |
2338 | * Kernel threads are allowed on online && !active CPUs. |
2339 | * | |
2340 | * Specifically, migration_disabled() tasks must not fail the | |
2341 | * cpumask_any_and_distribute() pick below, esp. so on | |
2342 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2343 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2344 | */ |
2345 | cpu_valid_mask = cpu_online_mask; | |
2346 | } | |
2347 | ||
25834c73 PZ |
2348 | /* |
2349 | * Must re-check here, to close a race against __kthread_bind(), | |
2350 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2351 | */ | |
9cfc3e18 | 2352 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2353 | ret = -EINVAL; |
2354 | goto out; | |
2355 | } | |
2356 | ||
885b3ba4 VS |
2357 | if (!(flags & SCA_MIGRATE_ENABLE)) { |
2358 | if (cpumask_equal(&p->cpus_mask, new_mask)) | |
2359 | goto out; | |
2360 | ||
2361 | if (WARN_ON_ONCE(p == current && | |
2362 | is_migration_disabled(p) && | |
2363 | !cpumask_test_cpu(task_cpu(p), new_mask))) { | |
2364 | ret = -EBUSY; | |
2365 | goto out; | |
2366 | } | |
2367 | } | |
5cc389bc | 2368 | |
46a87b38 PT |
2369 | /* |
2370 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2371 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2372 | * immediately required to distribute the tasks within their new mask. | |
2373 | */ | |
2374 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 2375 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2376 | ret = -EINVAL; |
2377 | goto out; | |
2378 | } | |
2379 | ||
9cfc3e18 | 2380 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 2381 | |
e9d867a6 PZI |
2382 | if (p->flags & PF_KTHREAD) { |
2383 | /* | |
2384 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 2385 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
2386 | */ |
2387 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
2388 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
2389 | p->nr_cpus_allowed != 1); | |
2390 | } | |
2391 | ||
6d337eab | 2392 | return affine_move_task(rq, p, &rf, dest_cpu, flags); |
5cc389bc | 2393 | |
5cc389bc | 2394 | out: |
eb580751 | 2395 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
2396 | |
2397 | return ret; | |
2398 | } | |
25834c73 PZ |
2399 | |
2400 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
2401 | { | |
9cfc3e18 | 2402 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 2403 | } |
5cc389bc PZ |
2404 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2405 | ||
dd41f596 | 2406 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2407 | { |
e2912009 PZ |
2408 | #ifdef CONFIG_SCHED_DEBUG |
2409 | /* | |
2410 | * We should never call set_task_cpu() on a blocked task, | |
2411 | * ttwu() will sort out the placement. | |
2412 | */ | |
077614ee | 2413 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 2414 | !p->on_rq); |
0122ec5b | 2415 | |
3ea94de1 JP |
2416 | /* |
2417 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
2418 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
2419 | * time relying on p->on_rq. | |
2420 | */ | |
2421 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
2422 | p->sched_class == &fair_sched_class && | |
2423 | (p->on_rq && !task_on_rq_migrating(p))); | |
2424 | ||
0122ec5b | 2425 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
2426 | /* |
2427 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
2428 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
2429 | * | |
2430 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 2431 | * see task_group(). |
6c6c54e1 PZ |
2432 | * |
2433 | * Furthermore, all task_rq users should acquire both locks, see | |
2434 | * task_rq_lock(). | |
2435 | */ | |
0122ec5b PZ |
2436 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
2437 | lockdep_is_held(&task_rq(p)->lock))); | |
2438 | #endif | |
4ff9083b PZ |
2439 | /* |
2440 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
2441 | */ | |
2442 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
2443 | |
2444 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
2445 | #endif |
2446 | ||
de1d7286 | 2447 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2448 | |
0c69774e | 2449 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 2450 | if (p->sched_class->migrate_task_rq) |
1327237a | 2451 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 2452 | p->se.nr_migrations++; |
d7822b1e | 2453 | rseq_migrate(p); |
ff303e66 | 2454 | perf_event_task_migrate(p); |
0c69774e | 2455 | } |
dd41f596 IM |
2456 | |
2457 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2458 | } |
2459 | ||
0ad4e3df | 2460 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
2461 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
2462 | { | |
da0c1e65 | 2463 | if (task_on_rq_queued(p)) { |
ac66f547 | 2464 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 2465 | struct rq_flags srf, drf; |
ac66f547 PZ |
2466 | |
2467 | src_rq = task_rq(p); | |
2468 | dst_rq = cpu_rq(cpu); | |
2469 | ||
8a8c69c3 PZ |
2470 | rq_pin_lock(src_rq, &srf); |
2471 | rq_pin_lock(dst_rq, &drf); | |
2472 | ||
ac66f547 PZ |
2473 | deactivate_task(src_rq, p, 0); |
2474 | set_task_cpu(p, cpu); | |
2475 | activate_task(dst_rq, p, 0); | |
2476 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
2477 | |
2478 | rq_unpin_lock(dst_rq, &drf); | |
2479 | rq_unpin_lock(src_rq, &srf); | |
2480 | ||
ac66f547 PZ |
2481 | } else { |
2482 | /* | |
2483 | * Task isn't running anymore; make it appear like we migrated | |
2484 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 2485 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
2486 | */ |
2487 | p->wake_cpu = cpu; | |
2488 | } | |
2489 | } | |
2490 | ||
2491 | struct migration_swap_arg { | |
2492 | struct task_struct *src_task, *dst_task; | |
2493 | int src_cpu, dst_cpu; | |
2494 | }; | |
2495 | ||
2496 | static int migrate_swap_stop(void *data) | |
2497 | { | |
2498 | struct migration_swap_arg *arg = data; | |
2499 | struct rq *src_rq, *dst_rq; | |
2500 | int ret = -EAGAIN; | |
2501 | ||
62694cd5 PZ |
2502 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
2503 | return -EAGAIN; | |
2504 | ||
ac66f547 PZ |
2505 | src_rq = cpu_rq(arg->src_cpu); |
2506 | dst_rq = cpu_rq(arg->dst_cpu); | |
2507 | ||
74602315 PZ |
2508 | double_raw_lock(&arg->src_task->pi_lock, |
2509 | &arg->dst_task->pi_lock); | |
ac66f547 | 2510 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 2511 | |
ac66f547 PZ |
2512 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
2513 | goto unlock; | |
2514 | ||
2515 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
2516 | goto unlock; | |
2517 | ||
3bd37062 | 2518 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
2519 | goto unlock; |
2520 | ||
3bd37062 | 2521 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
2522 | goto unlock; |
2523 | ||
2524 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
2525 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
2526 | ||
2527 | ret = 0; | |
2528 | ||
2529 | unlock: | |
2530 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
2531 | raw_spin_unlock(&arg->dst_task->pi_lock); |
2532 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
2533 | |
2534 | return ret; | |
2535 | } | |
2536 | ||
2537 | /* | |
2538 | * Cross migrate two tasks | |
2539 | */ | |
0ad4e3df SD |
2540 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
2541 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
2542 | { |
2543 | struct migration_swap_arg arg; | |
2544 | int ret = -EINVAL; | |
2545 | ||
ac66f547 PZ |
2546 | arg = (struct migration_swap_arg){ |
2547 | .src_task = cur, | |
0ad4e3df | 2548 | .src_cpu = curr_cpu, |
ac66f547 | 2549 | .dst_task = p, |
0ad4e3df | 2550 | .dst_cpu = target_cpu, |
ac66f547 PZ |
2551 | }; |
2552 | ||
2553 | if (arg.src_cpu == arg.dst_cpu) | |
2554 | goto out; | |
2555 | ||
6acce3ef PZ |
2556 | /* |
2557 | * These three tests are all lockless; this is OK since all of them | |
2558 | * will be re-checked with proper locks held further down the line. | |
2559 | */ | |
ac66f547 PZ |
2560 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
2561 | goto out; | |
2562 | ||
3bd37062 | 2563 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
2564 | goto out; |
2565 | ||
3bd37062 | 2566 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
2567 | goto out; |
2568 | ||
286549dc | 2569 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
2570 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
2571 | ||
2572 | out: | |
ac66f547 PZ |
2573 | return ret; |
2574 | } | |
0ad4e3df | 2575 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 2576 | |
1da177e4 LT |
2577 | /* |
2578 | * wait_task_inactive - wait for a thread to unschedule. | |
2579 | * | |
85ba2d86 RM |
2580 | * If @match_state is nonzero, it's the @p->state value just checked and |
2581 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2582 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2583 | * we return a positive number (its total switch count). If a second call | |
2584 | * a short while later returns the same number, the caller can be sure that | |
2585 | * @p has remained unscheduled the whole time. | |
2586 | * | |
1da177e4 LT |
2587 | * The caller must ensure that the task *will* unschedule sometime soon, |
2588 | * else this function might spin for a *long* time. This function can't | |
2589 | * be called with interrupts off, or it may introduce deadlock with | |
2590 | * smp_call_function() if an IPI is sent by the same process we are | |
2591 | * waiting to become inactive. | |
2592 | */ | |
85ba2d86 | 2593 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 2594 | { |
da0c1e65 | 2595 | int running, queued; |
eb580751 | 2596 | struct rq_flags rf; |
85ba2d86 | 2597 | unsigned long ncsw; |
70b97a7f | 2598 | struct rq *rq; |
1da177e4 | 2599 | |
3a5c359a AK |
2600 | for (;;) { |
2601 | /* | |
2602 | * We do the initial early heuristics without holding | |
2603 | * any task-queue locks at all. We'll only try to get | |
2604 | * the runqueue lock when things look like they will | |
2605 | * work out! | |
2606 | */ | |
2607 | rq = task_rq(p); | |
fa490cfd | 2608 | |
3a5c359a AK |
2609 | /* |
2610 | * If the task is actively running on another CPU | |
2611 | * still, just relax and busy-wait without holding | |
2612 | * any locks. | |
2613 | * | |
2614 | * NOTE! Since we don't hold any locks, it's not | |
2615 | * even sure that "rq" stays as the right runqueue! | |
2616 | * But we don't care, since "task_running()" will | |
2617 | * return false if the runqueue has changed and p | |
2618 | * is actually now running somewhere else! | |
2619 | */ | |
85ba2d86 RM |
2620 | while (task_running(rq, p)) { |
2621 | if (match_state && unlikely(p->state != match_state)) | |
2622 | return 0; | |
3a5c359a | 2623 | cpu_relax(); |
85ba2d86 | 2624 | } |
fa490cfd | 2625 | |
3a5c359a AK |
2626 | /* |
2627 | * Ok, time to look more closely! We need the rq | |
2628 | * lock now, to be *sure*. If we're wrong, we'll | |
2629 | * just go back and repeat. | |
2630 | */ | |
eb580751 | 2631 | rq = task_rq_lock(p, &rf); |
27a9da65 | 2632 | trace_sched_wait_task(p); |
3a5c359a | 2633 | running = task_running(rq, p); |
da0c1e65 | 2634 | queued = task_on_rq_queued(p); |
85ba2d86 | 2635 | ncsw = 0; |
f31e11d8 | 2636 | if (!match_state || p->state == match_state) |
93dcf55f | 2637 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 2638 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 2639 | |
85ba2d86 RM |
2640 | /* |
2641 | * If it changed from the expected state, bail out now. | |
2642 | */ | |
2643 | if (unlikely(!ncsw)) | |
2644 | break; | |
2645 | ||
3a5c359a AK |
2646 | /* |
2647 | * Was it really running after all now that we | |
2648 | * checked with the proper locks actually held? | |
2649 | * | |
2650 | * Oops. Go back and try again.. | |
2651 | */ | |
2652 | if (unlikely(running)) { | |
2653 | cpu_relax(); | |
2654 | continue; | |
2655 | } | |
fa490cfd | 2656 | |
3a5c359a AK |
2657 | /* |
2658 | * It's not enough that it's not actively running, | |
2659 | * it must be off the runqueue _entirely_, and not | |
2660 | * preempted! | |
2661 | * | |
80dd99b3 | 2662 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2663 | * running right now), it's preempted, and we should |
2664 | * yield - it could be a while. | |
2665 | */ | |
da0c1e65 | 2666 | if (unlikely(queued)) { |
8b0e1953 | 2667 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
2668 | |
2669 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2670 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2671 | continue; |
2672 | } | |
fa490cfd | 2673 | |
3a5c359a AK |
2674 | /* |
2675 | * Ahh, all good. It wasn't running, and it wasn't | |
2676 | * runnable, which means that it will never become | |
2677 | * running in the future either. We're all done! | |
2678 | */ | |
2679 | break; | |
2680 | } | |
85ba2d86 RM |
2681 | |
2682 | return ncsw; | |
1da177e4 LT |
2683 | } |
2684 | ||
2685 | /*** | |
2686 | * kick_process - kick a running thread to enter/exit the kernel | |
2687 | * @p: the to-be-kicked thread | |
2688 | * | |
2689 | * Cause a process which is running on another CPU to enter | |
2690 | * kernel-mode, without any delay. (to get signals handled.) | |
2691 | * | |
25985edc | 2692 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2693 | * because all it wants to ensure is that the remote task enters |
2694 | * the kernel. If the IPI races and the task has been migrated | |
2695 | * to another CPU then no harm is done and the purpose has been | |
2696 | * achieved as well. | |
2697 | */ | |
36c8b586 | 2698 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2699 | { |
2700 | int cpu; | |
2701 | ||
2702 | preempt_disable(); | |
2703 | cpu = task_cpu(p); | |
2704 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2705 | smp_send_reschedule(cpu); | |
2706 | preempt_enable(); | |
2707 | } | |
b43e3521 | 2708 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 2709 | |
30da688e | 2710 | /* |
3bd37062 | 2711 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
2712 | * |
2713 | * A few notes on cpu_active vs cpu_online: | |
2714 | * | |
2715 | * - cpu_active must be a subset of cpu_online | |
2716 | * | |
97fb7a0a | 2717 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 2718 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 2719 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
2720 | * see it. |
2721 | * | |
d1ccc66d | 2722 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 2723 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 2724 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
2725 | * off. |
2726 | * | |
2727 | * This means that fallback selection must not select !active CPUs. | |
2728 | * And can assume that any active CPU must be online. Conversely | |
2729 | * select_task_rq() below may allow selection of !active CPUs in order | |
2730 | * to satisfy the above rules. | |
30da688e | 2731 | */ |
5da9a0fb PZ |
2732 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2733 | { | |
aa00d89c TC |
2734 | int nid = cpu_to_node(cpu); |
2735 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
2736 | enum { cpuset, possible, fail } state = cpuset; |
2737 | int dest_cpu; | |
5da9a0fb | 2738 | |
aa00d89c | 2739 | /* |
d1ccc66d IM |
2740 | * If the node that the CPU is on has been offlined, cpu_to_node() |
2741 | * will return -1. There is no CPU on the node, and we should | |
2742 | * select the CPU on the other node. | |
aa00d89c TC |
2743 | */ |
2744 | if (nid != -1) { | |
2745 | nodemask = cpumask_of_node(nid); | |
2746 | ||
2747 | /* Look for allowed, online CPU in same node. */ | |
2748 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
2749 | if (!cpu_active(dest_cpu)) |
2750 | continue; | |
3bd37062 | 2751 | if (cpumask_test_cpu(dest_cpu, p->cpus_ptr)) |
aa00d89c TC |
2752 | return dest_cpu; |
2753 | } | |
2baab4e9 | 2754 | } |
5da9a0fb | 2755 | |
2baab4e9 PZ |
2756 | for (;;) { |
2757 | /* Any allowed, online CPU? */ | |
3bd37062 | 2758 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 2759 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 2760 | continue; |
175f0e25 | 2761 | |
2baab4e9 PZ |
2762 | goto out; |
2763 | } | |
5da9a0fb | 2764 | |
e73e85f0 | 2765 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
2766 | switch (state) { |
2767 | case cpuset: | |
e73e85f0 ON |
2768 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
2769 | cpuset_cpus_allowed_fallback(p); | |
2770 | state = possible; | |
2771 | break; | |
2772 | } | |
df561f66 | 2773 | fallthrough; |
2baab4e9 | 2774 | case possible: |
af449901 PZ |
2775 | /* |
2776 | * XXX When called from select_task_rq() we only | |
2777 | * hold p->pi_lock and again violate locking order. | |
2778 | * | |
2779 | * More yuck to audit. | |
2780 | */ | |
2baab4e9 PZ |
2781 | do_set_cpus_allowed(p, cpu_possible_mask); |
2782 | state = fail; | |
2783 | break; | |
2784 | ||
2785 | case fail: | |
2786 | BUG(); | |
2787 | break; | |
2788 | } | |
2789 | } | |
2790 | ||
2791 | out: | |
2792 | if (state != cpuset) { | |
2793 | /* | |
2794 | * Don't tell them about moving exiting tasks or | |
2795 | * kernel threads (both mm NULL), since they never | |
2796 | * leave kernel. | |
2797 | */ | |
2798 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 2799 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
2800 | task_pid_nr(p), p->comm, cpu); |
2801 | } | |
5da9a0fb PZ |
2802 | } |
2803 | ||
2804 | return dest_cpu; | |
2805 | } | |
2806 | ||
e2912009 | 2807 | /* |
3bd37062 | 2808 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 2809 | */ |
970b13ba | 2810 | static inline |
3aef1551 | 2811 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 2812 | { |
cbce1a68 PZ |
2813 | lockdep_assert_held(&p->pi_lock); |
2814 | ||
af449901 | 2815 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 2816 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 2817 | else |
3bd37062 | 2818 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
2819 | |
2820 | /* | |
2821 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 2822 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 2823 | * CPU. |
e2912009 PZ |
2824 | * |
2825 | * Since this is common to all placement strategies, this lives here. | |
2826 | * | |
2827 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2828 | * not worry about this generic constraint ] | |
2829 | */ | |
7af443ee | 2830 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 2831 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2832 | |
2833 | return cpu; | |
970b13ba | 2834 | } |
09a40af5 | 2835 | |
f5832c19 NP |
2836 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2837 | { | |
ded467dc | 2838 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
2839 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
2840 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2841 | ||
2842 | if (stop) { | |
2843 | /* | |
2844 | * Make it appear like a SCHED_FIFO task, its something | |
2845 | * userspace knows about and won't get confused about. | |
2846 | * | |
2847 | * Also, it will make PI more or less work without too | |
2848 | * much confusion -- but then, stop work should not | |
2849 | * rely on PI working anyway. | |
2850 | */ | |
2851 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2852 | ||
2853 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
2854 | |
2855 | /* | |
2856 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
2857 | * adjust the effective priority of a task. As a result, | |
2858 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
2859 | * which can then trigger wakeups of the stop thread to push | |
2860 | * around the current task. | |
2861 | * | |
2862 | * The stop task itself will never be part of the PI-chain, it | |
2863 | * never blocks, therefore that ->pi_lock recursion is safe. | |
2864 | * Tell lockdep about this by placing the stop->pi_lock in its | |
2865 | * own class. | |
2866 | */ | |
2867 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
2868 | } |
2869 | ||
2870 | cpu_rq(cpu)->stop = stop; | |
2871 | ||
2872 | if (old_stop) { | |
2873 | /* | |
2874 | * Reset it back to a normal scheduling class so that | |
2875 | * it can die in pieces. | |
2876 | */ | |
2877 | old_stop->sched_class = &rt_sched_class; | |
2878 | } | |
2879 | } | |
2880 | ||
74d862b6 | 2881 | #else /* CONFIG_SMP */ |
25834c73 PZ |
2882 | |
2883 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
2884 | const struct cpumask *new_mask, |
2885 | u32 flags) | |
25834c73 PZ |
2886 | { |
2887 | return set_cpus_allowed_ptr(p, new_mask); | |
2888 | } | |
2889 | ||
af449901 PZ |
2890 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
2891 | ||
3015ef4b TG |
2892 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2893 | { | |
2894 | return false; | |
2895 | } | |
2896 | ||
74d862b6 | 2897 | #endif /* !CONFIG_SMP */ |
af449901 | 2898 | |
d7c01d27 | 2899 | static void |
b84cb5df | 2900 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2901 | { |
4fa8d299 | 2902 | struct rq *rq; |
b84cb5df | 2903 | |
4fa8d299 JP |
2904 | if (!schedstat_enabled()) |
2905 | return; | |
2906 | ||
2907 | rq = this_rq(); | |
d7c01d27 | 2908 | |
4fa8d299 JP |
2909 | #ifdef CONFIG_SMP |
2910 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
2911 | __schedstat_inc(rq->ttwu_local); |
2912 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
2913 | } else { |
2914 | struct sched_domain *sd; | |
2915 | ||
b85c8b71 | 2916 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 2917 | rcu_read_lock(); |
4fa8d299 | 2918 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 2919 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 2920 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
2921 | break; |
2922 | } | |
2923 | } | |
057f3fad | 2924 | rcu_read_unlock(); |
d7c01d27 | 2925 | } |
f339b9dc PZ |
2926 | |
2927 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 2928 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
2929 | #endif /* CONFIG_SMP */ |
2930 | ||
b85c8b71 PZ |
2931 | __schedstat_inc(rq->ttwu_count); |
2932 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
2933 | |
2934 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 2935 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2936 | } |
2937 | ||
23f41eeb PZ |
2938 | /* |
2939 | * Mark the task runnable and perform wakeup-preemption. | |
2940 | */ | |
e7904a28 | 2941 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2942 | struct rq_flags *rf) |
9ed3811a | 2943 | { |
9ed3811a | 2944 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 2945 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
2946 | trace_sched_wakeup(p); |
2947 | ||
9ed3811a | 2948 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
2949 | if (p->sched_class->task_woken) { |
2950 | /* | |
b19a888c | 2951 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 2952 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 2953 | */ |
d8ac8971 | 2954 | rq_unpin_lock(rq, rf); |
9ed3811a | 2955 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2956 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 2957 | } |
9ed3811a | 2958 | |
e69c6341 | 2959 | if (rq->idle_stamp) { |
78becc27 | 2960 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 2961 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 2962 | |
abfafa54 JL |
2963 | update_avg(&rq->avg_idle, delta); |
2964 | ||
2965 | if (rq->avg_idle > max) | |
9ed3811a | 2966 | rq->avg_idle = max; |
abfafa54 | 2967 | |
9ed3811a TH |
2968 | rq->idle_stamp = 0; |
2969 | } | |
2970 | #endif | |
2971 | } | |
2972 | ||
c05fbafb | 2973 | static void |
e7904a28 | 2974 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2975 | struct rq_flags *rf) |
c05fbafb | 2976 | { |
77558e4d | 2977 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 2978 | |
cbce1a68 PZ |
2979 | lockdep_assert_held(&rq->lock); |
2980 | ||
c05fbafb PZ |
2981 | if (p->sched_contributes_to_load) |
2982 | rq->nr_uninterruptible--; | |
b5179ac7 | 2983 | |
dbfb089d | 2984 | #ifdef CONFIG_SMP |
b5179ac7 | 2985 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 2986 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
2987 | #endif |
2988 | ||
1b174a2c | 2989 | activate_task(rq, p, en_flags); |
d8ac8971 | 2990 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
2991 | } |
2992 | ||
2993 | /* | |
58877d34 PZ |
2994 | * Consider @p being inside a wait loop: |
2995 | * | |
2996 | * for (;;) { | |
2997 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
2998 | * | |
2999 | * if (CONDITION) | |
3000 | * break; | |
3001 | * | |
3002 | * schedule(); | |
3003 | * } | |
3004 | * __set_current_state(TASK_RUNNING); | |
3005 | * | |
3006 | * between set_current_state() and schedule(). In this case @p is still | |
3007 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3008 | * an atomic manner. | |
3009 | * | |
3010 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3011 | * then schedule() must still happen and p->state can be changed to | |
3012 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3013 | * need to do a full wakeup with enqueue. | |
3014 | * | |
3015 | * Returns: %true when the wakeup is done, | |
3016 | * %false otherwise. | |
c05fbafb | 3017 | */ |
58877d34 | 3018 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3019 | { |
eb580751 | 3020 | struct rq_flags rf; |
c05fbafb PZ |
3021 | struct rq *rq; |
3022 | int ret = 0; | |
3023 | ||
eb580751 | 3024 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3025 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3026 | /* check_preempt_curr() may use rq clock */ |
3027 | update_rq_clock(rq); | |
d8ac8971 | 3028 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3029 | ret = 1; |
3030 | } | |
eb580751 | 3031 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3032 | |
3033 | return ret; | |
3034 | } | |
3035 | ||
317f3941 | 3036 | #ifdef CONFIG_SMP |
a1488664 | 3037 | void sched_ttwu_pending(void *arg) |
317f3941 | 3038 | { |
a1488664 | 3039 | struct llist_node *llist = arg; |
317f3941 | 3040 | struct rq *rq = this_rq(); |
73215849 | 3041 | struct task_struct *p, *t; |
d8ac8971 | 3042 | struct rq_flags rf; |
317f3941 | 3043 | |
e3baac47 PZ |
3044 | if (!llist) |
3045 | return; | |
3046 | ||
126c2092 PZ |
3047 | /* |
3048 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
3049 | * Races such that false-negatives are possible, since they | |
3050 | * are shorter lived that false-positives would be. | |
3051 | */ | |
3052 | WRITE_ONCE(rq->ttwu_pending, 0); | |
3053 | ||
8a8c69c3 | 3054 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3055 | update_rq_clock(rq); |
317f3941 | 3056 | |
8c4890d1 | 3057 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3058 | if (WARN_ON_ONCE(p->on_cpu)) |
3059 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3060 | ||
3061 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3062 | set_task_cpu(p, cpu_of(rq)); | |
3063 | ||
73215849 | 3064 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3065 | } |
317f3941 | 3066 | |
8a8c69c3 | 3067 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3068 | } |
3069 | ||
b2a02fc4 | 3070 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3071 | { |
b2a02fc4 | 3072 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3073 | |
b2a02fc4 PZ |
3074 | if (!set_nr_if_polling(rq->idle)) |
3075 | arch_send_call_function_single_ipi(cpu); | |
3076 | else | |
3077 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3078 | } |
3079 | ||
2ebb1771 MG |
3080 | /* |
3081 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3082 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3083 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3084 | * of the wakeup instead of the waker. | |
3085 | */ | |
3086 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3087 | { |
e3baac47 PZ |
3088 | struct rq *rq = cpu_rq(cpu); |
3089 | ||
b7e7ade3 PZ |
3090 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3091 | ||
126c2092 | 3092 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3093 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3094 | } |
d6aa8f85 | 3095 | |
f6be8af1 CL |
3096 | void wake_up_if_idle(int cpu) |
3097 | { | |
3098 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3099 | struct rq_flags rf; |
f6be8af1 | 3100 | |
fd7de1e8 AL |
3101 | rcu_read_lock(); |
3102 | ||
3103 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3104 | goto out; | |
f6be8af1 CL |
3105 | |
3106 | if (set_nr_if_polling(rq->idle)) { | |
3107 | trace_sched_wake_idle_without_ipi(cpu); | |
3108 | } else { | |
8a8c69c3 | 3109 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
3110 | if (is_idle_task(rq->curr)) |
3111 | smp_send_reschedule(cpu); | |
d1ccc66d | 3112 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 3113 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 3114 | } |
fd7de1e8 AL |
3115 | |
3116 | out: | |
3117 | rcu_read_unlock(); | |
f6be8af1 CL |
3118 | } |
3119 | ||
39be3501 | 3120 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
3121 | { |
3122 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
3123 | } | |
c6e7bd7a | 3124 | |
2ebb1771 MG |
3125 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
3126 | { | |
3127 | /* | |
3128 | * If the CPU does not share cache, then queue the task on the | |
3129 | * remote rqs wakelist to avoid accessing remote data. | |
3130 | */ | |
3131 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3132 | return true; | |
3133 | ||
3134 | /* | |
3135 | * If the task is descheduling and the only running task on the | |
3136 | * CPU then use the wakelist to offload the task activation to | |
3137 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
3138 | * nr_running is checked to avoid unnecessary task stacking. | |
3139 | */ | |
739f70b4 | 3140 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
3141 | return true; |
3142 | ||
3143 | return false; | |
3144 | } | |
3145 | ||
3146 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3147 | { |
2ebb1771 | 3148 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
3149 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
3150 | return false; | |
3151 | ||
c6e7bd7a | 3152 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3153 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3154 | return true; |
3155 | } | |
3156 | ||
3157 | return false; | |
3158 | } | |
58877d34 PZ |
3159 | |
3160 | #else /* !CONFIG_SMP */ | |
3161 | ||
3162 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3163 | { | |
3164 | return false; | |
3165 | } | |
3166 | ||
d6aa8f85 | 3167 | #endif /* CONFIG_SMP */ |
317f3941 | 3168 | |
b5179ac7 | 3169 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3170 | { |
3171 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3172 | struct rq_flags rf; |
c05fbafb | 3173 | |
2ebb1771 | 3174 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3175 | return; |
317f3941 | 3176 | |
8a8c69c3 | 3177 | rq_lock(rq, &rf); |
77558e4d | 3178 | update_rq_clock(rq); |
d8ac8971 | 3179 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3180 | rq_unlock(rq, &rf); |
9ed3811a TH |
3181 | } |
3182 | ||
8643cda5 PZ |
3183 | /* |
3184 | * Notes on Program-Order guarantees on SMP systems. | |
3185 | * | |
3186 | * MIGRATION | |
3187 | * | |
3188 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3189 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3190 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3191 | * |
3192 | * For migration (of runnable tasks) this is provided by the following means: | |
3193 | * | |
3194 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3195 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3196 | * rq(c1)->lock (if not at the same time, then in that order). | |
3197 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3198 | * | |
7696f991 | 3199 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3200 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3201 | * |
3202 | * Example: | |
3203 | * | |
3204 | * CPU0 CPU1 CPU2 | |
3205 | * | |
3206 | * LOCK rq(0)->lock | |
3207 | * sched-out X | |
3208 | * sched-in Y | |
3209 | * UNLOCK rq(0)->lock | |
3210 | * | |
3211 | * LOCK rq(0)->lock // orders against CPU0 | |
3212 | * dequeue X | |
3213 | * UNLOCK rq(0)->lock | |
3214 | * | |
3215 | * LOCK rq(1)->lock | |
3216 | * enqueue X | |
3217 | * UNLOCK rq(1)->lock | |
3218 | * | |
3219 | * LOCK rq(1)->lock // orders against CPU2 | |
3220 | * sched-out Z | |
3221 | * sched-in X | |
3222 | * UNLOCK rq(1)->lock | |
3223 | * | |
3224 | * | |
3225 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3226 | * | |
3227 | * For blocking we (obviously) need to provide the same guarantee as for | |
3228 | * migration. However the means are completely different as there is no lock | |
3229 | * chain to provide order. Instead we do: | |
3230 | * | |
58877d34 PZ |
3231 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3232 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3233 | * |
3234 | * Example: | |
3235 | * | |
3236 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
3237 | * | |
3238 | * LOCK rq(0)->lock LOCK X->pi_lock | |
3239 | * dequeue X | |
3240 | * sched-out X | |
3241 | * smp_store_release(X->on_cpu, 0); | |
3242 | * | |
1f03e8d2 | 3243 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
3244 | * X->state = WAKING |
3245 | * set_task_cpu(X,2) | |
3246 | * | |
3247 | * LOCK rq(2)->lock | |
3248 | * enqueue X | |
3249 | * X->state = RUNNING | |
3250 | * UNLOCK rq(2)->lock | |
3251 | * | |
3252 | * LOCK rq(2)->lock // orders against CPU1 | |
3253 | * sched-out Z | |
3254 | * sched-in X | |
3255 | * UNLOCK rq(2)->lock | |
3256 | * | |
3257 | * UNLOCK X->pi_lock | |
3258 | * UNLOCK rq(0)->lock | |
3259 | * | |
3260 | * | |
7696f991 AP |
3261 | * However, for wakeups there is a second guarantee we must provide, namely we |
3262 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
3263 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
3264 | */ |
3265 | ||
9ed3811a | 3266 | /** |
1da177e4 | 3267 | * try_to_wake_up - wake up a thread |
9ed3811a | 3268 | * @p: the thread to be awakened |
1da177e4 | 3269 | * @state: the mask of task states that can be woken |
9ed3811a | 3270 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 3271 | * |
58877d34 PZ |
3272 | * Conceptually does: |
3273 | * | |
3274 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 3275 | * |
a2250238 PZ |
3276 | * If the task was not queued/runnable, also place it back on a runqueue. |
3277 | * | |
58877d34 PZ |
3278 | * This function is atomic against schedule() which would dequeue the task. |
3279 | * | |
3280 | * It issues a full memory barrier before accessing @p->state, see the comment | |
3281 | * with set_current_state(). | |
a2250238 | 3282 | * |
58877d34 | 3283 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 3284 | * |
58877d34 PZ |
3285 | * Relies on p->pi_lock stabilizing: |
3286 | * - p->sched_class | |
3287 | * - p->cpus_ptr | |
3288 | * - p->sched_task_group | |
3289 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
3290 | * | |
3291 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
3292 | * Takes rq->lock in: | |
3293 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
3294 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
3295 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
3296 | * | |
3297 | * As a consequence we race really badly with just about everything. See the | |
3298 | * many memory barriers and their comments for details. | |
7696f991 | 3299 | * |
a2250238 PZ |
3300 | * Return: %true if @p->state changes (an actual wakeup was done), |
3301 | * %false otherwise. | |
1da177e4 | 3302 | */ |
e4a52bcb PZ |
3303 | static int |
3304 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 3305 | { |
1da177e4 | 3306 | unsigned long flags; |
c05fbafb | 3307 | int cpu, success = 0; |
2398f2c6 | 3308 | |
e3d85487 | 3309 | preempt_disable(); |
aacedf26 PZ |
3310 | if (p == current) { |
3311 | /* | |
3312 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
3313 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 3314 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
3315 | * without taking any locks. |
3316 | * | |
3317 | * In particular: | |
3318 | * - we rely on Program-Order guarantees for all the ordering, | |
3319 | * - we're serialized against set_special_state() by virtue of | |
3320 | * it disabling IRQs (this allows not taking ->pi_lock). | |
3321 | */ | |
3322 | if (!(p->state & state)) | |
e3d85487 | 3323 | goto out; |
aacedf26 PZ |
3324 | |
3325 | success = 1; | |
aacedf26 PZ |
3326 | trace_sched_waking(p); |
3327 | p->state = TASK_RUNNING; | |
3328 | trace_sched_wakeup(p); | |
3329 | goto out; | |
3330 | } | |
3331 | ||
e0acd0a6 ON |
3332 | /* |
3333 | * If we are going to wake up a thread waiting for CONDITION we | |
3334 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
3335 | * reordered with p->state check below. This pairs with smp_store_mb() |
3336 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 3337 | */ |
013fdb80 | 3338 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 3339 | smp_mb__after_spinlock(); |
e9c84311 | 3340 | if (!(p->state & state)) |
aacedf26 | 3341 | goto unlock; |
1da177e4 | 3342 | |
fbd705a0 PZ |
3343 | trace_sched_waking(p); |
3344 | ||
d1ccc66d IM |
3345 | /* We're going to change ->state: */ |
3346 | success = 1; | |
1da177e4 | 3347 | |
135e8c92 BS |
3348 | /* |
3349 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
3350 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
3351 | * in smp_cond_load_acquire() below. | |
3352 | * | |
3d85b270 AP |
3353 | * sched_ttwu_pending() try_to_wake_up() |
3354 | * STORE p->on_rq = 1 LOAD p->state | |
3355 | * UNLOCK rq->lock | |
3356 | * | |
3357 | * __schedule() (switch to task 'p') | |
3358 | * LOCK rq->lock smp_rmb(); | |
3359 | * smp_mb__after_spinlock(); | |
3360 | * UNLOCK rq->lock | |
135e8c92 BS |
3361 | * |
3362 | * [task p] | |
3d85b270 | 3363 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 3364 | * |
3d85b270 AP |
3365 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3366 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
3367 | * |
3368 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
3369 | */ |
3370 | smp_rmb(); | |
58877d34 | 3371 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 3372 | goto unlock; |
1da177e4 | 3373 | |
c6e7bd7a PZ |
3374 | if (p->in_iowait) { |
3375 | delayacct_blkio_end(p); | |
3376 | atomic_dec(&task_rq(p)->nr_iowait); | |
3377 | } | |
3378 | ||
1da177e4 | 3379 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
3380 | /* |
3381 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
3382 | * possible to, falsely, observe p->on_cpu == 0. | |
3383 | * | |
3384 | * One must be running (->on_cpu == 1) in order to remove oneself | |
3385 | * from the runqueue. | |
3386 | * | |
3d85b270 AP |
3387 | * __schedule() (switch to task 'p') try_to_wake_up() |
3388 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
3389 | * UNLOCK rq->lock | |
3390 | * | |
3391 | * __schedule() (put 'p' to sleep) | |
3392 | * LOCK rq->lock smp_rmb(); | |
3393 | * smp_mb__after_spinlock(); | |
3394 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 3395 | * |
3d85b270 AP |
3396 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3397 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
3398 | * |
3399 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
3400 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
3401 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 3402 | */ |
dbfb089d PZ |
3403 | smp_acquire__after_ctrl_dep(); |
3404 | ||
3405 | /* | |
3406 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
3407 | * == 0), which means we need to do an enqueue, change p->state to | |
3408 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
3409 | * enqueue, such as ttwu_queue_wakelist(). | |
3410 | */ | |
3411 | p->state = TASK_WAKING; | |
ecf7d01c | 3412 | |
c6e7bd7a PZ |
3413 | /* |
3414 | * If the owning (remote) CPU is still in the middle of schedule() with | |
3415 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
3416 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
3417 | * let the waker make forward progress. This is safe because IRQs are | |
3418 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
3419 | * |
3420 | * Ensure we load task_cpu(p) after p->on_cpu: | |
3421 | * | |
3422 | * set_task_cpu(p, cpu); | |
3423 | * STORE p->cpu = @cpu | |
3424 | * __schedule() (switch to task 'p') | |
3425 | * LOCK rq->lock | |
3426 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
3427 | * STORE p->on_cpu = 1 LOAD p->cpu | |
3428 | * | |
3429 | * to ensure we observe the correct CPU on which the task is currently | |
3430 | * scheduling. | |
c6e7bd7a | 3431 | */ |
b6e13e85 | 3432 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 3433 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
3434 | goto unlock; |
3435 | ||
e9c84311 | 3436 | /* |
d1ccc66d | 3437 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 3438 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 3439 | * |
31cb1bc0 | 3440 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
3441 | * |
3442 | * This ensures that tasks getting woken will be fully ordered against | |
3443 | * their previous state and preserve Program Order. | |
0970d299 | 3444 | */ |
1f03e8d2 | 3445 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 3446 | |
3aef1551 | 3447 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc PZ |
3448 | if (task_cpu(p) != cpu) { |
3449 | wake_flags |= WF_MIGRATED; | |
eb414681 | 3450 | psi_ttwu_dequeue(p); |
e4a52bcb | 3451 | set_task_cpu(p, cpu); |
f339b9dc | 3452 | } |
b6e13e85 PZ |
3453 | #else |
3454 | cpu = task_cpu(p); | |
1da177e4 | 3455 | #endif /* CONFIG_SMP */ |
1da177e4 | 3456 | |
b5179ac7 | 3457 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 3458 | unlock: |
013fdb80 | 3459 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
3460 | out: |
3461 | if (success) | |
b6e13e85 | 3462 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 3463 | preempt_enable(); |
1da177e4 LT |
3464 | |
3465 | return success; | |
3466 | } | |
3467 | ||
2beaf328 PM |
3468 | /** |
3469 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
3470 | * @p: Process for which the function is to be invoked. | |
3471 | * @func: Function to invoke. | |
3472 | * @arg: Argument to function. | |
3473 | * | |
3474 | * If the specified task can be quickly locked into a definite state | |
3475 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
3476 | * state while invoking @func(@arg). This function can use ->on_rq and | |
3477 | * task_curr() to work out what the state is, if required. Given that | |
3478 | * @func can be invoked with a runqueue lock held, it had better be quite | |
3479 | * lightweight. | |
3480 | * | |
3481 | * Returns: | |
3482 | * @false if the task slipped out from under the locks. | |
3483 | * @true if the task was locked onto a runqueue or is sleeping. | |
3484 | * However, @func can override this by returning @false. | |
3485 | */ | |
3486 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
3487 | { | |
3488 | bool ret = false; | |
3489 | struct rq_flags rf; | |
3490 | struct rq *rq; | |
3491 | ||
3492 | lockdep_assert_irqs_enabled(); | |
3493 | raw_spin_lock_irq(&p->pi_lock); | |
3494 | if (p->on_rq) { | |
3495 | rq = __task_rq_lock(p, &rf); | |
3496 | if (task_rq(p) == rq) | |
3497 | ret = func(p, arg); | |
3498 | rq_unlock(rq, &rf); | |
3499 | } else { | |
3500 | switch (p->state) { | |
3501 | case TASK_RUNNING: | |
3502 | case TASK_WAKING: | |
3503 | break; | |
3504 | default: | |
3505 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
3506 | if (!p->on_rq) | |
3507 | ret = func(p, arg); | |
3508 | } | |
3509 | } | |
3510 | raw_spin_unlock_irq(&p->pi_lock); | |
3511 | return ret; | |
3512 | } | |
3513 | ||
50fa610a DH |
3514 | /** |
3515 | * wake_up_process - Wake up a specific process | |
3516 | * @p: The process to be woken up. | |
3517 | * | |
3518 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
3519 | * processes. |
3520 | * | |
3521 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 3522 | * |
7696f991 | 3523 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 3524 | */ |
7ad5b3a5 | 3525 | int wake_up_process(struct task_struct *p) |
1da177e4 | 3526 | { |
9067ac85 | 3527 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 3528 | } |
1da177e4 LT |
3529 | EXPORT_SYMBOL(wake_up_process); |
3530 | ||
7ad5b3a5 | 3531 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
3532 | { |
3533 | return try_to_wake_up(p, state, 0); | |
3534 | } | |
3535 | ||
1da177e4 LT |
3536 | /* |
3537 | * Perform scheduler related setup for a newly forked process p. | |
3538 | * p is forked by current. | |
dd41f596 IM |
3539 | * |
3540 | * __sched_fork() is basic setup used by init_idle() too: | |
3541 | */ | |
5e1576ed | 3542 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3543 | { |
fd2f4419 PZ |
3544 | p->on_rq = 0; |
3545 | ||
3546 | p->se.on_rq = 0; | |
dd41f596 IM |
3547 | p->se.exec_start = 0; |
3548 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 3549 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 3550 | p->se.nr_migrations = 0; |
da7a735e | 3551 | p->se.vruntime = 0; |
fd2f4419 | 3552 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 3553 | |
ad936d86 BP |
3554 | #ifdef CONFIG_FAIR_GROUP_SCHED |
3555 | p->se.cfs_rq = NULL; | |
3556 | #endif | |
3557 | ||
6cfb0d5d | 3558 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 3559 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 3560 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 3561 | #endif |
476d139c | 3562 | |
aab03e05 | 3563 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 3564 | init_dl_task_timer(&p->dl); |
209a0cbd | 3565 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 3566 | __dl_clear_params(p); |
aab03e05 | 3567 | |
fa717060 | 3568 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
3569 | p->rt.timeout = 0; |
3570 | p->rt.time_slice = sched_rr_timeslice; | |
3571 | p->rt.on_rq = 0; | |
3572 | p->rt.on_list = 0; | |
476d139c | 3573 | |
e107be36 AK |
3574 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3575 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
3576 | #endif | |
cbee9f88 | 3577 | |
5e1f0f09 MG |
3578 | #ifdef CONFIG_COMPACTION |
3579 | p->capture_control = NULL; | |
3580 | #endif | |
13784475 | 3581 | init_numa_balancing(clone_flags, p); |
a1488664 | 3582 | #ifdef CONFIG_SMP |
8c4890d1 | 3583 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 3584 | p->migration_pending = NULL; |
a1488664 | 3585 | #endif |
dd41f596 IM |
3586 | } |
3587 | ||
2a595721 SD |
3588 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
3589 | ||
1a687c2e | 3590 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 3591 | |
1a687c2e MG |
3592 | void set_numabalancing_state(bool enabled) |
3593 | { | |
3594 | if (enabled) | |
2a595721 | 3595 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 3596 | else |
2a595721 | 3597 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 3598 | } |
54a43d54 AK |
3599 | |
3600 | #ifdef CONFIG_PROC_SYSCTL | |
3601 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 3602 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
3603 | { |
3604 | struct ctl_table t; | |
3605 | int err; | |
2a595721 | 3606 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
3607 | |
3608 | if (write && !capable(CAP_SYS_ADMIN)) | |
3609 | return -EPERM; | |
3610 | ||
3611 | t = *table; | |
3612 | t.data = &state; | |
3613 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3614 | if (err < 0) | |
3615 | return err; | |
3616 | if (write) | |
3617 | set_numabalancing_state(state); | |
3618 | return err; | |
3619 | } | |
3620 | #endif | |
3621 | #endif | |
dd41f596 | 3622 | |
4698f88c JP |
3623 | #ifdef CONFIG_SCHEDSTATS |
3624 | ||
cb251765 | 3625 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 3626 | static bool __initdata __sched_schedstats = false; |
cb251765 | 3627 | |
cb251765 MG |
3628 | static void set_schedstats(bool enabled) |
3629 | { | |
3630 | if (enabled) | |
3631 | static_branch_enable(&sched_schedstats); | |
3632 | else | |
3633 | static_branch_disable(&sched_schedstats); | |
3634 | } | |
3635 | ||
3636 | void force_schedstat_enabled(void) | |
3637 | { | |
3638 | if (!schedstat_enabled()) { | |
3639 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
3640 | static_branch_enable(&sched_schedstats); | |
3641 | } | |
3642 | } | |
3643 | ||
3644 | static int __init setup_schedstats(char *str) | |
3645 | { | |
3646 | int ret = 0; | |
3647 | if (!str) | |
3648 | goto out; | |
3649 | ||
4698f88c JP |
3650 | /* |
3651 | * This code is called before jump labels have been set up, so we can't | |
3652 | * change the static branch directly just yet. Instead set a temporary | |
3653 | * variable so init_schedstats() can do it later. | |
3654 | */ | |
cb251765 | 3655 | if (!strcmp(str, "enable")) { |
4698f88c | 3656 | __sched_schedstats = true; |
cb251765 MG |
3657 | ret = 1; |
3658 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 3659 | __sched_schedstats = false; |
cb251765 MG |
3660 | ret = 1; |
3661 | } | |
3662 | out: | |
3663 | if (!ret) | |
3664 | pr_warn("Unable to parse schedstats=\n"); | |
3665 | ||
3666 | return ret; | |
3667 | } | |
3668 | __setup("schedstats=", setup_schedstats); | |
3669 | ||
4698f88c JP |
3670 | static void __init init_schedstats(void) |
3671 | { | |
3672 | set_schedstats(__sched_schedstats); | |
3673 | } | |
3674 | ||
cb251765 | 3675 | #ifdef CONFIG_PROC_SYSCTL |
32927393 CH |
3676 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
3677 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
3678 | { |
3679 | struct ctl_table t; | |
3680 | int err; | |
3681 | int state = static_branch_likely(&sched_schedstats); | |
3682 | ||
3683 | if (write && !capable(CAP_SYS_ADMIN)) | |
3684 | return -EPERM; | |
3685 | ||
3686 | t = *table; | |
3687 | t.data = &state; | |
3688 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3689 | if (err < 0) | |
3690 | return err; | |
3691 | if (write) | |
3692 | set_schedstats(state); | |
3693 | return err; | |
3694 | } | |
4698f88c JP |
3695 | #endif /* CONFIG_PROC_SYSCTL */ |
3696 | #else /* !CONFIG_SCHEDSTATS */ | |
3697 | static inline void init_schedstats(void) {} | |
3698 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
3699 | |
3700 | /* | |
3701 | * fork()/clone()-time setup: | |
3702 | */ | |
aab03e05 | 3703 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3704 | { |
0122ec5b | 3705 | unsigned long flags; |
dd41f596 | 3706 | |
5e1576ed | 3707 | __sched_fork(clone_flags, p); |
06b83b5f | 3708 | /* |
7dc603c9 | 3709 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
3710 | * nobody will actually run it, and a signal or other external |
3711 | * event cannot wake it up and insert it on the runqueue either. | |
3712 | */ | |
7dc603c9 | 3713 | p->state = TASK_NEW; |
dd41f596 | 3714 | |
c350a04e MG |
3715 | /* |
3716 | * Make sure we do not leak PI boosting priority to the child. | |
3717 | */ | |
3718 | p->prio = current->normal_prio; | |
3719 | ||
e8f14172 PB |
3720 | uclamp_fork(p); |
3721 | ||
b9dc29e7 MG |
3722 | /* |
3723 | * Revert to default priority/policy on fork if requested. | |
3724 | */ | |
3725 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 3726 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 3727 | p->policy = SCHED_NORMAL; |
6c697bdf | 3728 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
3729 | p->rt_priority = 0; |
3730 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
3731 | p->static_prio = NICE_TO_PRIO(0); | |
3732 | ||
3733 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 3734 | set_load_weight(p, false); |
6c697bdf | 3735 | |
b9dc29e7 MG |
3736 | /* |
3737 | * We don't need the reset flag anymore after the fork. It has | |
3738 | * fulfilled its duty: | |
3739 | */ | |
3740 | p->sched_reset_on_fork = 0; | |
3741 | } | |
ca94c442 | 3742 | |
af0fffd9 | 3743 | if (dl_prio(p->prio)) |
aab03e05 | 3744 | return -EAGAIN; |
af0fffd9 | 3745 | else if (rt_prio(p->prio)) |
aab03e05 | 3746 | p->sched_class = &rt_sched_class; |
af0fffd9 | 3747 | else |
2ddbf952 | 3748 | p->sched_class = &fair_sched_class; |
b29739f9 | 3749 | |
7dc603c9 | 3750 | init_entity_runnable_average(&p->se); |
cd29fe6f | 3751 | |
86951599 PZ |
3752 | /* |
3753 | * The child is not yet in the pid-hash so no cgroup attach races, | |
3754 | * and the cgroup is pinned to this child due to cgroup_fork() | |
3755 | * is ran before sched_fork(). | |
3756 | * | |
3757 | * Silence PROVE_RCU. | |
3758 | */ | |
0122ec5b | 3759 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ce3614da | 3760 | rseq_migrate(p); |
e210bffd | 3761 | /* |
d1ccc66d | 3762 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
3763 | * so use __set_task_cpu(). |
3764 | */ | |
af0fffd9 | 3765 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
3766 | if (p->sched_class->task_fork) |
3767 | p->sched_class->task_fork(p); | |
0122ec5b | 3768 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 3769 | |
f6db8347 | 3770 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 3771 | if (likely(sched_info_on())) |
52f17b6c | 3772 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 3773 | #endif |
3ca7a440 PZ |
3774 | #if defined(CONFIG_SMP) |
3775 | p->on_cpu = 0; | |
4866cde0 | 3776 | #endif |
01028747 | 3777 | init_task_preempt_count(p); |
806c09a7 | 3778 | #ifdef CONFIG_SMP |
917b627d | 3779 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 3780 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 3781 | #endif |
aab03e05 | 3782 | return 0; |
1da177e4 LT |
3783 | } |
3784 | ||
13685c4a QY |
3785 | void sched_post_fork(struct task_struct *p) |
3786 | { | |
3787 | uclamp_post_fork(p); | |
3788 | } | |
3789 | ||
332ac17e DF |
3790 | unsigned long to_ratio(u64 period, u64 runtime) |
3791 | { | |
3792 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 3793 | return BW_UNIT; |
332ac17e DF |
3794 | |
3795 | /* | |
3796 | * Doing this here saves a lot of checks in all | |
3797 | * the calling paths, and returning zero seems | |
3798 | * safe for them anyway. | |
3799 | */ | |
3800 | if (period == 0) | |
3801 | return 0; | |
3802 | ||
c52f14d3 | 3803 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
3804 | } |
3805 | ||
1da177e4 LT |
3806 | /* |
3807 | * wake_up_new_task - wake up a newly created task for the first time. | |
3808 | * | |
3809 | * This function will do some initial scheduler statistics housekeeping | |
3810 | * that must be done for every newly created context, then puts the task | |
3811 | * on the runqueue and wakes it. | |
3812 | */ | |
3e51e3ed | 3813 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 3814 | { |
eb580751 | 3815 | struct rq_flags rf; |
dd41f596 | 3816 | struct rq *rq; |
fabf318e | 3817 | |
eb580751 | 3818 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 3819 | p->state = TASK_RUNNING; |
fabf318e PZ |
3820 | #ifdef CONFIG_SMP |
3821 | /* | |
3822 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 3823 | * - cpus_ptr can change in the fork path |
d1ccc66d | 3824 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
3825 | * |
3826 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
3827 | * as we're not fully set-up yet. | |
fabf318e | 3828 | */ |
32e839dd | 3829 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 3830 | rseq_migrate(p); |
3aef1551 | 3831 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 3832 | #endif |
b7fa30c9 | 3833 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 3834 | update_rq_clock(rq); |
d0fe0b9c | 3835 | post_init_entity_util_avg(p); |
0017d735 | 3836 | |
7a57f32a | 3837 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 3838 | trace_sched_wakeup_new(p); |
a7558e01 | 3839 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3840 | #ifdef CONFIG_SMP |
0aaafaab PZ |
3841 | if (p->sched_class->task_woken) { |
3842 | /* | |
b19a888c | 3843 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
3844 | * drop it. |
3845 | */ | |
d8ac8971 | 3846 | rq_unpin_lock(rq, &rf); |
efbbd05a | 3847 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3848 | rq_repin_lock(rq, &rf); |
0aaafaab | 3849 | } |
9a897c5a | 3850 | #endif |
eb580751 | 3851 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
3852 | } |
3853 | ||
e107be36 AK |
3854 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3855 | ||
b7203428 | 3856 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 3857 | |
2ecd9d29 PZ |
3858 | void preempt_notifier_inc(void) |
3859 | { | |
b7203428 | 3860 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
3861 | } |
3862 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
3863 | ||
3864 | void preempt_notifier_dec(void) | |
3865 | { | |
b7203428 | 3866 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
3867 | } |
3868 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
3869 | ||
e107be36 | 3870 | /** |
80dd99b3 | 3871 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3872 | * @notifier: notifier struct to register |
e107be36 AK |
3873 | */ |
3874 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3875 | { | |
b7203428 | 3876 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
3877 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
3878 | ||
e107be36 AK |
3879 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
3880 | } | |
3881 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3882 | ||
3883 | /** | |
3884 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3885 | * @notifier: notifier struct to unregister |
e107be36 | 3886 | * |
d84525a8 | 3887 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
3888 | */ |
3889 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3890 | { | |
3891 | hlist_del(¬ifier->link); | |
3892 | } | |
3893 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3894 | ||
1cde2930 | 3895 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3896 | { |
3897 | struct preempt_notifier *notifier; | |
e107be36 | 3898 | |
b67bfe0d | 3899 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3900 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
3901 | } | |
3902 | ||
1cde2930 PZ |
3903 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
3904 | { | |
b7203428 | 3905 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3906 | __fire_sched_in_preempt_notifiers(curr); |
3907 | } | |
3908 | ||
e107be36 | 3909 | static void |
1cde2930 PZ |
3910 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3911 | struct task_struct *next) | |
e107be36 AK |
3912 | { |
3913 | struct preempt_notifier *notifier; | |
e107be36 | 3914 | |
b67bfe0d | 3915 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3916 | notifier->ops->sched_out(notifier, next); |
3917 | } | |
3918 | ||
1cde2930 PZ |
3919 | static __always_inline void |
3920 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3921 | struct task_struct *next) | |
3922 | { | |
b7203428 | 3923 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3924 | __fire_sched_out_preempt_notifiers(curr, next); |
3925 | } | |
3926 | ||
6d6bc0ad | 3927 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3928 | |
1cde2930 | 3929 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3930 | { |
3931 | } | |
3932 | ||
1cde2930 | 3933 | static inline void |
e107be36 AK |
3934 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3935 | struct task_struct *next) | |
3936 | { | |
3937 | } | |
3938 | ||
6d6bc0ad | 3939 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3940 | |
31cb1bc0 | 3941 | static inline void prepare_task(struct task_struct *next) |
3942 | { | |
3943 | #ifdef CONFIG_SMP | |
3944 | /* | |
3945 | * Claim the task as running, we do this before switching to it | |
3946 | * such that any running task will have this set. | |
58877d34 PZ |
3947 | * |
3948 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 3949 | */ |
58877d34 | 3950 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 3951 | #endif |
3952 | } | |
3953 | ||
3954 | static inline void finish_task(struct task_struct *prev) | |
3955 | { | |
3956 | #ifdef CONFIG_SMP | |
3957 | /* | |
58877d34 PZ |
3958 | * This must be the very last reference to @prev from this CPU. After |
3959 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
3960 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 3961 | * finished. |
3962 | * | |
3963 | * In particular, the load of prev->state in finish_task_switch() must | |
3964 | * happen before this. | |
3965 | * | |
3966 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
3967 | */ | |
3968 | smp_store_release(&prev->on_cpu, 0); | |
3969 | #endif | |
3970 | } | |
3971 | ||
565790d2 PZ |
3972 | #ifdef CONFIG_SMP |
3973 | ||
3974 | static void do_balance_callbacks(struct rq *rq, struct callback_head *head) | |
3975 | { | |
3976 | void (*func)(struct rq *rq); | |
3977 | struct callback_head *next; | |
3978 | ||
3979 | lockdep_assert_held(&rq->lock); | |
3980 | ||
3981 | while (head) { | |
3982 | func = (void (*)(struct rq *))head->func; | |
3983 | next = head->next; | |
3984 | head->next = NULL; | |
3985 | head = next; | |
3986 | ||
3987 | func(rq); | |
3988 | } | |
3989 | } | |
3990 | ||
3991 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
3992 | { | |
3993 | struct callback_head *head = rq->balance_callback; | |
3994 | ||
3995 | lockdep_assert_held(&rq->lock); | |
2558aacf | 3996 | if (head) { |
565790d2 | 3997 | rq->balance_callback = NULL; |
2558aacf PZ |
3998 | rq->balance_flags &= ~BALANCE_WORK; |
3999 | } | |
565790d2 PZ |
4000 | |
4001 | return head; | |
4002 | } | |
4003 | ||
4004 | static void __balance_callbacks(struct rq *rq) | |
4005 | { | |
4006 | do_balance_callbacks(rq, splice_balance_callbacks(rq)); | |
4007 | } | |
4008 | ||
4009 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4010 | { | |
4011 | unsigned long flags; | |
4012 | ||
4013 | if (unlikely(head)) { | |
4014 | raw_spin_lock_irqsave(&rq->lock, flags); | |
4015 | do_balance_callbacks(rq, head); | |
4016 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
4017 | } | |
4018 | } | |
4019 | ||
2558aacf PZ |
4020 | static void balance_push(struct rq *rq); |
4021 | ||
4022 | static inline void balance_switch(struct rq *rq) | |
4023 | { | |
4024 | if (likely(!rq->balance_flags)) | |
4025 | return; | |
4026 | ||
4027 | if (rq->balance_flags & BALANCE_PUSH) { | |
4028 | balance_push(rq); | |
4029 | return; | |
4030 | } | |
4031 | ||
4032 | __balance_callbacks(rq); | |
4033 | } | |
4034 | ||
565790d2 PZ |
4035 | #else |
4036 | ||
4037 | static inline void __balance_callbacks(struct rq *rq) | |
4038 | { | |
4039 | } | |
4040 | ||
4041 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
4042 | { | |
4043 | return NULL; | |
4044 | } | |
4045 | ||
4046 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4047 | { | |
4048 | } | |
4049 | ||
2558aacf PZ |
4050 | static inline void balance_switch(struct rq *rq) |
4051 | { | |
4052 | } | |
4053 | ||
565790d2 PZ |
4054 | #endif |
4055 | ||
269d5992 PZ |
4056 | static inline void |
4057 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4058 | { |
269d5992 PZ |
4059 | /* |
4060 | * Since the runqueue lock will be released by the next | |
4061 | * task (which is an invalid locking op but in the case | |
4062 | * of the scheduler it's an obvious special-case), so we | |
4063 | * do an early lockdep release here: | |
4064 | */ | |
4065 | rq_unpin_lock(rq, rf); | |
5facae4f | 4066 | spin_release(&rq->lock.dep_map, _THIS_IP_); |
31cb1bc0 | 4067 | #ifdef CONFIG_DEBUG_SPINLOCK |
4068 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 4069 | rq->lock.owner = next; |
31cb1bc0 | 4070 | #endif |
269d5992 PZ |
4071 | } |
4072 | ||
4073 | static inline void finish_lock_switch(struct rq *rq) | |
4074 | { | |
31cb1bc0 | 4075 | /* |
4076 | * If we are tracking spinlock dependencies then we have to | |
4077 | * fix up the runqueue lock - which gets 'carried over' from | |
4078 | * prev into current: | |
4079 | */ | |
4080 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
2558aacf | 4081 | balance_switch(rq); |
31cb1bc0 | 4082 | raw_spin_unlock_irq(&rq->lock); |
4083 | } | |
4084 | ||
325ea10c IM |
4085 | /* |
4086 | * NOP if the arch has not defined these: | |
4087 | */ | |
4088 | ||
4089 | #ifndef prepare_arch_switch | |
4090 | # define prepare_arch_switch(next) do { } while (0) | |
4091 | #endif | |
4092 | ||
4093 | #ifndef finish_arch_post_lock_switch | |
4094 | # define finish_arch_post_lock_switch() do { } while (0) | |
4095 | #endif | |
4096 | ||
4866cde0 NP |
4097 | /** |
4098 | * prepare_task_switch - prepare to switch tasks | |
4099 | * @rq: the runqueue preparing to switch | |
421cee29 | 4100 | * @prev: the current task that is being switched out |
4866cde0 NP |
4101 | * @next: the task we are going to switch to. |
4102 | * | |
4103 | * This is called with the rq lock held and interrupts off. It must | |
4104 | * be paired with a subsequent finish_task_switch after the context | |
4105 | * switch. | |
4106 | * | |
4107 | * prepare_task_switch sets up locking and calls architecture specific | |
4108 | * hooks. | |
4109 | */ | |
e107be36 AK |
4110 | static inline void |
4111 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
4112 | struct task_struct *next) | |
4866cde0 | 4113 | { |
0ed557aa | 4114 | kcov_prepare_switch(prev); |
43148951 | 4115 | sched_info_switch(rq, prev, next); |
fe4b04fa | 4116 | perf_event_task_sched_out(prev, next); |
d7822b1e | 4117 | rseq_preempt(prev); |
e107be36 | 4118 | fire_sched_out_preempt_notifiers(prev, next); |
31cb1bc0 | 4119 | prepare_task(next); |
4866cde0 NP |
4120 | prepare_arch_switch(next); |
4121 | } | |
4122 | ||
1da177e4 LT |
4123 | /** |
4124 | * finish_task_switch - clean up after a task-switch | |
4125 | * @prev: the thread we just switched away from. | |
4126 | * | |
4866cde0 NP |
4127 | * finish_task_switch must be called after the context switch, paired |
4128 | * with a prepare_task_switch call before the context switch. | |
4129 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
4130 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
4131 | * |
4132 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 4133 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
4134 | * with the lock held can cause deadlocks; see schedule() for |
4135 | * details.) | |
dfa50b60 ON |
4136 | * |
4137 | * The context switch have flipped the stack from under us and restored the | |
4138 | * local variables which were saved when this task called schedule() in the | |
4139 | * past. prev == current is still correct but we need to recalculate this_rq | |
4140 | * because prev may have moved to another CPU. | |
1da177e4 | 4141 | */ |
dfa50b60 | 4142 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
4143 | __releases(rq->lock) |
4144 | { | |
dfa50b60 | 4145 | struct rq *rq = this_rq(); |
1da177e4 | 4146 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 4147 | long prev_state; |
1da177e4 | 4148 | |
609ca066 PZ |
4149 | /* |
4150 | * The previous task will have left us with a preempt_count of 2 | |
4151 | * because it left us after: | |
4152 | * | |
4153 | * schedule() | |
4154 | * preempt_disable(); // 1 | |
4155 | * __schedule() | |
4156 | * raw_spin_lock_irq(&rq->lock) // 2 | |
4157 | * | |
4158 | * Also, see FORK_PREEMPT_COUNT. | |
4159 | */ | |
e2bf1c4b PZ |
4160 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
4161 | "corrupted preempt_count: %s/%d/0x%x\n", | |
4162 | current->comm, current->pid, preempt_count())) | |
4163 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 4164 | |
1da177e4 LT |
4165 | rq->prev_mm = NULL; |
4166 | ||
4167 | /* | |
4168 | * A task struct has one reference for the use as "current". | |
c394cc9f | 4169 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
4170 | * schedule one last time. The schedule call will never return, and |
4171 | * the scheduled task must drop that reference. | |
95913d97 PZ |
4172 | * |
4173 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 4174 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
4175 | * running on another CPU and we could rave with its RUNNING -> DEAD |
4176 | * transition, resulting in a double drop. | |
1da177e4 | 4177 | */ |
55a101f8 | 4178 | prev_state = prev->state; |
bf9fae9f | 4179 | vtime_task_switch(prev); |
a8d757ef | 4180 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 4181 | finish_task(prev); |
4182 | finish_lock_switch(rq); | |
01f23e16 | 4183 | finish_arch_post_lock_switch(); |
0ed557aa | 4184 | kcov_finish_switch(current); |
e8fa1362 | 4185 | |
e107be36 | 4186 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 4187 | /* |
70216e18 MD |
4188 | * When switching through a kernel thread, the loop in |
4189 | * membarrier_{private,global}_expedited() may have observed that | |
4190 | * kernel thread and not issued an IPI. It is therefore possible to | |
4191 | * schedule between user->kernel->user threads without passing though | |
4192 | * switch_mm(). Membarrier requires a barrier after storing to | |
4193 | * rq->curr, before returning to userspace, so provide them here: | |
4194 | * | |
4195 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
4196 | * provided by mmdrop(), | |
4197 | * - a sync_core for SYNC_CORE. | |
306e0604 | 4198 | */ |
70216e18 MD |
4199 | if (mm) { |
4200 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 4201 | mmdrop(mm); |
70216e18 | 4202 | } |
1cef1150 PZ |
4203 | if (unlikely(prev_state == TASK_DEAD)) { |
4204 | if (prev->sched_class->task_dead) | |
4205 | prev->sched_class->task_dead(prev); | |
68f24b08 | 4206 | |
1cef1150 PZ |
4207 | /* |
4208 | * Remove function-return probe instances associated with this | |
4209 | * task and put them back on the free list. | |
4210 | */ | |
4211 | kprobe_flush_task(prev); | |
4212 | ||
4213 | /* Task is done with its stack. */ | |
4214 | put_task_stack(prev); | |
4215 | ||
0ff7b2cf | 4216 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 4217 | } |
99e5ada9 | 4218 | |
de734f89 | 4219 | tick_nohz_task_switch(); |
dfa50b60 | 4220 | return rq; |
1da177e4 LT |
4221 | } |
4222 | ||
4223 | /** | |
4224 | * schedule_tail - first thing a freshly forked thread must call. | |
4225 | * @prev: the thread we just switched away from. | |
4226 | */ | |
722a9f92 | 4227 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
4228 | __releases(rq->lock) |
4229 | { | |
1a43a14a | 4230 | struct rq *rq; |
da19ab51 | 4231 | |
609ca066 PZ |
4232 | /* |
4233 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
4234 | * finish_task_switch() for details. | |
4235 | * | |
4236 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
4237 | * and the preempt_enable() will end up enabling preemption (on | |
4238 | * PREEMPT_COUNT kernels). | |
4239 | */ | |
4240 | ||
dfa50b60 | 4241 | rq = finish_task_switch(prev); |
1a43a14a | 4242 | preempt_enable(); |
70b97a7f | 4243 | |
1da177e4 | 4244 | if (current->set_child_tid) |
b488893a | 4245 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
4246 | |
4247 | calculate_sigpending(); | |
1da177e4 LT |
4248 | } |
4249 | ||
4250 | /* | |
dfa50b60 | 4251 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 4252 | */ |
04936948 | 4253 | static __always_inline struct rq * |
70b97a7f | 4254 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 4255 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 4256 | { |
e107be36 | 4257 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 4258 | |
9226d125 ZA |
4259 | /* |
4260 | * For paravirt, this is coupled with an exit in switch_to to | |
4261 | * combine the page table reload and the switch backend into | |
4262 | * one hypercall. | |
4263 | */ | |
224101ed | 4264 | arch_start_context_switch(prev); |
9226d125 | 4265 | |
306e0604 | 4266 | /* |
139d025c PZ |
4267 | * kernel -> kernel lazy + transfer active |
4268 | * user -> kernel lazy + mmgrab() active | |
4269 | * | |
4270 | * kernel -> user switch + mmdrop() active | |
4271 | * user -> user switch | |
306e0604 | 4272 | */ |
139d025c PZ |
4273 | if (!next->mm) { // to kernel |
4274 | enter_lazy_tlb(prev->active_mm, next); | |
4275 | ||
4276 | next->active_mm = prev->active_mm; | |
4277 | if (prev->mm) // from user | |
4278 | mmgrab(prev->active_mm); | |
4279 | else | |
4280 | prev->active_mm = NULL; | |
4281 | } else { // to user | |
227a4aad | 4282 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
4283 | /* |
4284 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 4285 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
4286 | * |
4287 | * The below provides this either through switch_mm(), or in | |
4288 | * case 'prev->active_mm == next->mm' through | |
4289 | * finish_task_switch()'s mmdrop(). | |
4290 | */ | |
139d025c | 4291 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 4292 | |
139d025c PZ |
4293 | if (!prev->mm) { // from kernel |
4294 | /* will mmdrop() in finish_task_switch(). */ | |
4295 | rq->prev_mm = prev->active_mm; | |
4296 | prev->active_mm = NULL; | |
4297 | } | |
1da177e4 | 4298 | } |
92509b73 | 4299 | |
cb42c9a3 | 4300 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 4301 | |
269d5992 | 4302 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
4303 | |
4304 | /* Here we just switch the register state and the stack. */ | |
4305 | switch_to(prev, next, prev); | |
dd41f596 | 4306 | barrier(); |
dfa50b60 ON |
4307 | |
4308 | return finish_task_switch(prev); | |
1da177e4 LT |
4309 | } |
4310 | ||
4311 | /* | |
1c3e8264 | 4312 | * nr_running and nr_context_switches: |
1da177e4 LT |
4313 | * |
4314 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 4315 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
4316 | */ |
4317 | unsigned long nr_running(void) | |
4318 | { | |
4319 | unsigned long i, sum = 0; | |
4320 | ||
4321 | for_each_online_cpu(i) | |
4322 | sum += cpu_rq(i)->nr_running; | |
4323 | ||
4324 | return sum; | |
f711f609 | 4325 | } |
1da177e4 | 4326 | |
2ee507c4 | 4327 | /* |
d1ccc66d | 4328 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
4329 | * |
4330 | * Caution: this function does not check that the caller has disabled | |
4331 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
4332 | * race. The caller is responsible to use it correctly, for example: | |
4333 | * | |
dfcb245e | 4334 | * - from a non-preemptible section (of course) |
00cc1633 DD |
4335 | * |
4336 | * - from a thread that is bound to a single CPU | |
4337 | * | |
4338 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
4339 | */ |
4340 | bool single_task_running(void) | |
4341 | { | |
00cc1633 | 4342 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
4343 | } |
4344 | EXPORT_SYMBOL(single_task_running); | |
4345 | ||
1da177e4 | 4346 | unsigned long long nr_context_switches(void) |
46cb4b7c | 4347 | { |
cc94abfc SR |
4348 | int i; |
4349 | unsigned long long sum = 0; | |
46cb4b7c | 4350 | |
0a945022 | 4351 | for_each_possible_cpu(i) |
1da177e4 | 4352 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 4353 | |
1da177e4 LT |
4354 | return sum; |
4355 | } | |
483b4ee6 | 4356 | |
145d952a DL |
4357 | /* |
4358 | * Consumers of these two interfaces, like for example the cpuidle menu | |
4359 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
4360 | * for a CPU that has IO-wait which might not even end up running the task when | |
4361 | * it does become runnable. | |
4362 | */ | |
4363 | ||
4364 | unsigned long nr_iowait_cpu(int cpu) | |
4365 | { | |
4366 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
4367 | } | |
4368 | ||
e33a9bba | 4369 | /* |
b19a888c | 4370 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
4371 | * |
4372 | * The idea behind IO-wait account is to account the idle time that we could | |
4373 | * have spend running if it were not for IO. That is, if we were to improve the | |
4374 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
4375 | * | |
4376 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
4377 | * idle time as IO-wait, because if the storage were faster, it could've been | |
4378 | * running and we'd not be idle. | |
4379 | * | |
4380 | * This has been extended to SMP, by doing the same for each CPU. This however | |
4381 | * is broken. | |
4382 | * | |
4383 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
4384 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
4385 | * though, if the storage were faster, both could've ran at the same time, | |
4386 | * utilising both CPUs. | |
4387 | * | |
4388 | * This means, that when looking globally, the current IO-wait accounting on | |
4389 | * SMP is a lower bound, by reason of under accounting. | |
4390 | * | |
4391 | * Worse, since the numbers are provided per CPU, they are sometimes | |
4392 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
4393 | * associated with any one particular CPU, it can wake to another CPU than it | |
4394 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
4395 | * | |
4396 | * Task CPU affinities can make all that even more 'interesting'. | |
4397 | */ | |
4398 | ||
1da177e4 LT |
4399 | unsigned long nr_iowait(void) |
4400 | { | |
4401 | unsigned long i, sum = 0; | |
483b4ee6 | 4402 | |
0a945022 | 4403 | for_each_possible_cpu(i) |
145d952a | 4404 | sum += nr_iowait_cpu(i); |
46cb4b7c | 4405 | |
1da177e4 LT |
4406 | return sum; |
4407 | } | |
483b4ee6 | 4408 | |
dd41f596 | 4409 | #ifdef CONFIG_SMP |
8a0be9ef | 4410 | |
46cb4b7c | 4411 | /* |
38022906 PZ |
4412 | * sched_exec - execve() is a valuable balancing opportunity, because at |
4413 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 4414 | */ |
38022906 | 4415 | void sched_exec(void) |
46cb4b7c | 4416 | { |
38022906 | 4417 | struct task_struct *p = current; |
1da177e4 | 4418 | unsigned long flags; |
0017d735 | 4419 | int dest_cpu; |
46cb4b7c | 4420 | |
8f42ced9 | 4421 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 4422 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
4423 | if (dest_cpu == smp_processor_id()) |
4424 | goto unlock; | |
38022906 | 4425 | |
8f42ced9 | 4426 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 4427 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 4428 | |
8f42ced9 PZ |
4429 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
4430 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
4431 | return; |
4432 | } | |
0017d735 | 4433 | unlock: |
8f42ced9 | 4434 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 4435 | } |
dd41f596 | 4436 | |
1da177e4 LT |
4437 | #endif |
4438 | ||
1da177e4 | 4439 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 4440 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
4441 | |
4442 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 4443 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 4444 | |
6075620b GG |
4445 | /* |
4446 | * The function fair_sched_class.update_curr accesses the struct curr | |
4447 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
4448 | * we observe a high rate of cache misses in practice. | |
4449 | * Prefetching this data results in improved performance. | |
4450 | */ | |
4451 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
4452 | { | |
4453 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
4454 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
4455 | #else | |
4456 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
4457 | #endif | |
4458 | prefetch(curr); | |
4459 | prefetch(&curr->exec_start); | |
4460 | } | |
4461 | ||
c5f8d995 HS |
4462 | /* |
4463 | * Return accounted runtime for the task. | |
4464 | * In case the task is currently running, return the runtime plus current's | |
4465 | * pending runtime that have not been accounted yet. | |
4466 | */ | |
4467 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4468 | { | |
eb580751 | 4469 | struct rq_flags rf; |
c5f8d995 | 4470 | struct rq *rq; |
6e998916 | 4471 | u64 ns; |
c5f8d995 | 4472 | |
911b2898 PZ |
4473 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
4474 | /* | |
97fb7a0a | 4475 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
4476 | * So we have a optimization chance when the task's delta_exec is 0. |
4477 | * Reading ->on_cpu is racy, but this is ok. | |
4478 | * | |
d1ccc66d IM |
4479 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
4480 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 4481 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
4482 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
4483 | * been accounted, so we're correct here as well. | |
911b2898 | 4484 | */ |
da0c1e65 | 4485 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
4486 | return p->se.sum_exec_runtime; |
4487 | #endif | |
4488 | ||
eb580751 | 4489 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
4490 | /* |
4491 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
4492 | * project cycles that may never be accounted to this | |
4493 | * thread, breaking clock_gettime(). | |
4494 | */ | |
4495 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 4496 | prefetch_curr_exec_start(p); |
6e998916 SG |
4497 | update_rq_clock(rq); |
4498 | p->sched_class->update_curr(rq); | |
4499 | } | |
4500 | ns = p->se.sum_exec_runtime; | |
eb580751 | 4501 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
4502 | |
4503 | return ns; | |
4504 | } | |
48f24c4d | 4505 | |
7835b98b CL |
4506 | /* |
4507 | * This function gets called by the timer code, with HZ frequency. | |
4508 | * We call it with interrupts disabled. | |
7835b98b CL |
4509 | */ |
4510 | void scheduler_tick(void) | |
4511 | { | |
7835b98b CL |
4512 | int cpu = smp_processor_id(); |
4513 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4514 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 4515 | struct rq_flags rf; |
b4eccf5f | 4516 | unsigned long thermal_pressure; |
3e51f33f | 4517 | |
1567c3e3 | 4518 | arch_scale_freq_tick(); |
3e51f33f | 4519 | sched_clock_tick(); |
dd41f596 | 4520 | |
8a8c69c3 PZ |
4521 | rq_lock(rq, &rf); |
4522 | ||
3e51f33f | 4523 | update_rq_clock(rq); |
b4eccf5f | 4524 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 4525 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 4526 | curr->sched_class->task_tick(rq, curr, 0); |
3289bdb4 | 4527 | calc_global_load_tick(rq); |
eb414681 | 4528 | psi_task_tick(rq); |
8a8c69c3 PZ |
4529 | |
4530 | rq_unlock(rq, &rf); | |
7835b98b | 4531 | |
e9d2b064 | 4532 | perf_event_task_tick(); |
e220d2dc | 4533 | |
e418e1c2 | 4534 | #ifdef CONFIG_SMP |
6eb57e0d | 4535 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 4536 | trigger_load_balance(rq); |
e418e1c2 | 4537 | #endif |
1da177e4 LT |
4538 | } |
4539 | ||
265f22a9 | 4540 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
4541 | |
4542 | struct tick_work { | |
4543 | int cpu; | |
b55bd585 | 4544 | atomic_t state; |
d84b3131 FW |
4545 | struct delayed_work work; |
4546 | }; | |
b55bd585 PM |
4547 | /* Values for ->state, see diagram below. */ |
4548 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
4549 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
4550 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
4551 | ||
4552 | /* | |
4553 | * State diagram for ->state: | |
4554 | * | |
4555 | * | |
4556 | * TICK_SCHED_REMOTE_OFFLINE | |
4557 | * | ^ | |
4558 | * | | | |
4559 | * | | sched_tick_remote() | |
4560 | * | | | |
4561 | * | | | |
4562 | * +--TICK_SCHED_REMOTE_OFFLINING | |
4563 | * | ^ | |
4564 | * | | | |
4565 | * sched_tick_start() | | sched_tick_stop() | |
4566 | * | | | |
4567 | * V | | |
4568 | * TICK_SCHED_REMOTE_RUNNING | |
4569 | * | |
4570 | * | |
4571 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
4572 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
4573 | */ | |
d84b3131 FW |
4574 | |
4575 | static struct tick_work __percpu *tick_work_cpu; | |
4576 | ||
4577 | static void sched_tick_remote(struct work_struct *work) | |
4578 | { | |
4579 | struct delayed_work *dwork = to_delayed_work(work); | |
4580 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
4581 | int cpu = twork->cpu; | |
4582 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 4583 | struct task_struct *curr; |
d84b3131 | 4584 | struct rq_flags rf; |
d9c0ffca | 4585 | u64 delta; |
b55bd585 | 4586 | int os; |
d84b3131 FW |
4587 | |
4588 | /* | |
4589 | * Handle the tick only if it appears the remote CPU is running in full | |
4590 | * dynticks mode. The check is racy by nature, but missing a tick or | |
4591 | * having one too much is no big deal because the scheduler tick updates | |
4592 | * statistics and checks timeslices in a time-independent way, regardless | |
4593 | * of when exactly it is running. | |
4594 | */ | |
488603b8 | 4595 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 4596 | goto out_requeue; |
d84b3131 | 4597 | |
d9c0ffca FW |
4598 | rq_lock_irq(rq, &rf); |
4599 | curr = rq->curr; | |
488603b8 | 4600 | if (cpu_is_offline(cpu)) |
d9c0ffca | 4601 | goto out_unlock; |
d84b3131 | 4602 | |
d9c0ffca | 4603 | update_rq_clock(rq); |
d9c0ffca | 4604 | |
488603b8 SW |
4605 | if (!is_idle_task(curr)) { |
4606 | /* | |
4607 | * Make sure the next tick runs within a reasonable | |
4608 | * amount of time. | |
4609 | */ | |
4610 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
4611 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
4612 | } | |
d9c0ffca FW |
4613 | curr->sched_class->task_tick(rq, curr, 0); |
4614 | ||
ebc0f83c | 4615 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
4616 | out_unlock: |
4617 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 4618 | out_requeue: |
ebc0f83c | 4619 | |
d84b3131 FW |
4620 | /* |
4621 | * Run the remote tick once per second (1Hz). This arbitrary | |
4622 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
4623 | * to keep scheduler internal stats reasonably up to date. But |
4624 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 4625 | */ |
b55bd585 PM |
4626 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
4627 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
4628 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
4629 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
4630 | } |
4631 | ||
4632 | static void sched_tick_start(int cpu) | |
4633 | { | |
b55bd585 | 4634 | int os; |
d84b3131 FW |
4635 | struct tick_work *twork; |
4636 | ||
4637 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4638 | return; | |
4639 | ||
4640 | WARN_ON_ONCE(!tick_work_cpu); | |
4641 | ||
4642 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4643 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
4644 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
4645 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
4646 | twork->cpu = cpu; | |
4647 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
4648 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
4649 | } | |
d84b3131 FW |
4650 | } |
4651 | ||
4652 | #ifdef CONFIG_HOTPLUG_CPU | |
4653 | static void sched_tick_stop(int cpu) | |
4654 | { | |
4655 | struct tick_work *twork; | |
b55bd585 | 4656 | int os; |
d84b3131 FW |
4657 | |
4658 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4659 | return; | |
4660 | ||
4661 | WARN_ON_ONCE(!tick_work_cpu); | |
4662 | ||
4663 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4664 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
4665 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
4666 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
4667 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
4668 | } |
4669 | #endif /* CONFIG_HOTPLUG_CPU */ | |
4670 | ||
4671 | int __init sched_tick_offload_init(void) | |
4672 | { | |
4673 | tick_work_cpu = alloc_percpu(struct tick_work); | |
4674 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
4675 | return 0; |
4676 | } | |
4677 | ||
4678 | #else /* !CONFIG_NO_HZ_FULL */ | |
4679 | static inline void sched_tick_start(int cpu) { } | |
4680 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 4681 | #endif |
1da177e4 | 4682 | |
c1a280b6 | 4683 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 4684 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
4685 | /* |
4686 | * If the value passed in is equal to the current preempt count | |
4687 | * then we just disabled preemption. Start timing the latency. | |
4688 | */ | |
4689 | static inline void preempt_latency_start(int val) | |
4690 | { | |
4691 | if (preempt_count() == val) { | |
4692 | unsigned long ip = get_lock_parent_ip(); | |
4693 | #ifdef CONFIG_DEBUG_PREEMPT | |
4694 | current->preempt_disable_ip = ip; | |
4695 | #endif | |
4696 | trace_preempt_off(CALLER_ADDR0, ip); | |
4697 | } | |
4698 | } | |
7e49fcce | 4699 | |
edafe3a5 | 4700 | void preempt_count_add(int val) |
1da177e4 | 4701 | { |
6cd8a4bb | 4702 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4703 | /* |
4704 | * Underflow? | |
4705 | */ | |
9a11b49a IM |
4706 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4707 | return; | |
6cd8a4bb | 4708 | #endif |
bdb43806 | 4709 | __preempt_count_add(val); |
6cd8a4bb | 4710 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4711 | /* |
4712 | * Spinlock count overflowing soon? | |
4713 | */ | |
33859f7f MOS |
4714 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4715 | PREEMPT_MASK - 10); | |
6cd8a4bb | 4716 | #endif |
47252cfb | 4717 | preempt_latency_start(val); |
1da177e4 | 4718 | } |
bdb43806 | 4719 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 4720 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 4721 | |
47252cfb SR |
4722 | /* |
4723 | * If the value passed in equals to the current preempt count | |
4724 | * then we just enabled preemption. Stop timing the latency. | |
4725 | */ | |
4726 | static inline void preempt_latency_stop(int val) | |
4727 | { | |
4728 | if (preempt_count() == val) | |
4729 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
4730 | } | |
4731 | ||
edafe3a5 | 4732 | void preempt_count_sub(int val) |
1da177e4 | 4733 | { |
6cd8a4bb | 4734 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4735 | /* |
4736 | * Underflow? | |
4737 | */ | |
01e3eb82 | 4738 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4739 | return; |
1da177e4 LT |
4740 | /* |
4741 | * Is the spinlock portion underflowing? | |
4742 | */ | |
9a11b49a IM |
4743 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4744 | !(preempt_count() & PREEMPT_MASK))) | |
4745 | return; | |
6cd8a4bb | 4746 | #endif |
9a11b49a | 4747 | |
47252cfb | 4748 | preempt_latency_stop(val); |
bdb43806 | 4749 | __preempt_count_sub(val); |
1da177e4 | 4750 | } |
bdb43806 | 4751 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 4752 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 4753 | |
47252cfb SR |
4754 | #else |
4755 | static inline void preempt_latency_start(int val) { } | |
4756 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
4757 | #endif |
4758 | ||
59ddbcb2 IM |
4759 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
4760 | { | |
4761 | #ifdef CONFIG_DEBUG_PREEMPT | |
4762 | return p->preempt_disable_ip; | |
4763 | #else | |
4764 | return 0; | |
4765 | #endif | |
4766 | } | |
4767 | ||
1da177e4 | 4768 | /* |
dd41f596 | 4769 | * Print scheduling while atomic bug: |
1da177e4 | 4770 | */ |
dd41f596 | 4771 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4772 | { |
d1c6d149 VN |
4773 | /* Save this before calling printk(), since that will clobber it */ |
4774 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
4775 | ||
664dfa65 DJ |
4776 | if (oops_in_progress) |
4777 | return; | |
4778 | ||
3df0fc5b PZ |
4779 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4780 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4781 | |
dd41f596 | 4782 | debug_show_held_locks(prev); |
e21f5b15 | 4783 | print_modules(); |
dd41f596 IM |
4784 | if (irqs_disabled()) |
4785 | print_irqtrace_events(prev); | |
d1c6d149 VN |
4786 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
4787 | && in_atomic_preempt_off()) { | |
8f47b187 | 4788 | pr_err("Preemption disabled at:"); |
2062a4e8 | 4789 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 4790 | } |
748c7201 DBO |
4791 | if (panic_on_warn) |
4792 | panic("scheduling while atomic\n"); | |
4793 | ||
6135fc1e | 4794 | dump_stack(); |
373d4d09 | 4795 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 4796 | } |
1da177e4 | 4797 | |
dd41f596 IM |
4798 | /* |
4799 | * Various schedule()-time debugging checks and statistics: | |
4800 | */ | |
312364f3 | 4801 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 4802 | { |
0d9e2632 | 4803 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
4804 | if (task_stack_end_corrupted(prev)) |
4805 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
4806 | |
4807 | if (task_scs_end_corrupted(prev)) | |
4808 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 4809 | #endif |
b99def8b | 4810 | |
312364f3 DV |
4811 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
4812 | if (!preempt && prev->state && prev->non_block_count) { | |
4813 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", | |
4814 | prev->comm, prev->pid, prev->non_block_count); | |
4815 | dump_stack(); | |
4816 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
4817 | } | |
4818 | #endif | |
4819 | ||
1dc0fffc | 4820 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 4821 | __schedule_bug(prev); |
1dc0fffc PZ |
4822 | preempt_count_set(PREEMPT_DISABLED); |
4823 | } | |
b3fbab05 | 4824 | rcu_sleep_check(); |
dd41f596 | 4825 | |
1da177e4 LT |
4826 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4827 | ||
ae92882e | 4828 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
4829 | } |
4830 | ||
457d1f46 CY |
4831 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
4832 | struct rq_flags *rf) | |
4833 | { | |
4834 | #ifdef CONFIG_SMP | |
4835 | const struct sched_class *class; | |
4836 | /* | |
4837 | * We must do the balancing pass before put_prev_task(), such | |
4838 | * that when we release the rq->lock the task is in the same | |
4839 | * state as before we took rq->lock. | |
4840 | * | |
4841 | * We can terminate the balance pass as soon as we know there is | |
4842 | * a runnable task of @class priority or higher. | |
4843 | */ | |
4844 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
4845 | if (class->balance(rq, prev, rf)) | |
4846 | break; | |
4847 | } | |
4848 | #endif | |
4849 | ||
4850 | put_prev_task(rq, prev); | |
4851 | } | |
4852 | ||
dd41f596 IM |
4853 | /* |
4854 | * Pick up the highest-prio task: | |
4855 | */ | |
4856 | static inline struct task_struct * | |
d8ac8971 | 4857 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 4858 | { |
49ee5768 | 4859 | const struct sched_class *class; |
dd41f596 | 4860 | struct task_struct *p; |
1da177e4 LT |
4861 | |
4862 | /* | |
0ba87bb2 PZ |
4863 | * Optimization: we know that if all tasks are in the fair class we can |
4864 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 4865 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 4866 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 4867 | */ |
aa93cd53 | 4868 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
4869 | rq->nr_running == rq->cfs.h_nr_running)) { |
4870 | ||
5d7d6056 | 4871 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 4872 | if (unlikely(p == RETRY_TASK)) |
67692435 | 4873 | goto restart; |
6ccdc84b | 4874 | |
d1ccc66d | 4875 | /* Assumes fair_sched_class->next == idle_sched_class */ |
5d7d6056 | 4876 | if (!p) { |
f488e105 | 4877 | put_prev_task(rq, prev); |
98c2f700 | 4878 | p = pick_next_task_idle(rq); |
f488e105 | 4879 | } |
6ccdc84b PZ |
4880 | |
4881 | return p; | |
1da177e4 LT |
4882 | } |
4883 | ||
67692435 | 4884 | restart: |
457d1f46 | 4885 | put_prev_task_balance(rq, prev, rf); |
67692435 | 4886 | |
34f971f6 | 4887 | for_each_class(class) { |
98c2f700 | 4888 | p = class->pick_next_task(rq); |
67692435 | 4889 | if (p) |
dd41f596 | 4890 | return p; |
dd41f596 | 4891 | } |
34f971f6 | 4892 | |
d1ccc66d IM |
4893 | /* The idle class should always have a runnable task: */ |
4894 | BUG(); | |
dd41f596 | 4895 | } |
1da177e4 | 4896 | |
dd41f596 | 4897 | /* |
c259e01a | 4898 | * __schedule() is the main scheduler function. |
edde96ea PE |
4899 | * |
4900 | * The main means of driving the scheduler and thus entering this function are: | |
4901 | * | |
4902 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
4903 | * | |
4904 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
4905 | * paths. For example, see arch/x86/entry_64.S. | |
4906 | * | |
4907 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
4908 | * interrupt handler scheduler_tick(). | |
4909 | * | |
4910 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
4911 | * task to the run-queue and that's it. | |
4912 | * | |
4913 | * Now, if the new task added to the run-queue preempts the current | |
4914 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
4915 | * called on the nearest possible occasion: | |
4916 | * | |
c1a280b6 | 4917 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
4918 | * |
4919 | * - in syscall or exception context, at the next outmost | |
4920 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
4921 | * spin_unlock()!) | |
4922 | * | |
4923 | * - in IRQ context, return from interrupt-handler to | |
4924 | * preemptible context | |
4925 | * | |
c1a280b6 | 4926 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
4927 | * then at the next: |
4928 | * | |
4929 | * - cond_resched() call | |
4930 | * - explicit schedule() call | |
4931 | * - return from syscall or exception to user-space | |
4932 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 4933 | * |
b30f0e3f | 4934 | * WARNING: must be called with preemption disabled! |
dd41f596 | 4935 | */ |
499d7955 | 4936 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
4937 | { |
4938 | struct task_struct *prev, *next; | |
67ca7bde | 4939 | unsigned long *switch_count; |
dbfb089d | 4940 | unsigned long prev_state; |
d8ac8971 | 4941 | struct rq_flags rf; |
dd41f596 | 4942 | struct rq *rq; |
31656519 | 4943 | int cpu; |
dd41f596 | 4944 | |
dd41f596 IM |
4945 | cpu = smp_processor_id(); |
4946 | rq = cpu_rq(cpu); | |
dd41f596 | 4947 | prev = rq->curr; |
dd41f596 | 4948 | |
312364f3 | 4949 | schedule_debug(prev, preempt); |
1da177e4 | 4950 | |
31656519 | 4951 | if (sched_feat(HRTICK)) |
f333fdc9 | 4952 | hrtick_clear(rq); |
8f4d37ec | 4953 | |
46a5d164 | 4954 | local_irq_disable(); |
bcbfdd01 | 4955 | rcu_note_context_switch(preempt); |
46a5d164 | 4956 | |
e0acd0a6 ON |
4957 | /* |
4958 | * Make sure that signal_pending_state()->signal_pending() below | |
4959 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
4960 | * done by the caller to avoid the race with signal_wake_up(): |
4961 | * | |
4962 | * __set_current_state(@state) signal_wake_up() | |
4963 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
4964 | * wake_up_state(p, state) | |
4965 | * LOCK rq->lock LOCK p->pi_state | |
4966 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
4967 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 4968 | * |
dbfb089d | 4969 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 4970 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 4971 | */ |
8a8c69c3 | 4972 | rq_lock(rq, &rf); |
d89e588c | 4973 | smp_mb__after_spinlock(); |
1da177e4 | 4974 | |
d1ccc66d IM |
4975 | /* Promote REQ to ACT */ |
4976 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 4977 | update_rq_clock(rq); |
9edfbfed | 4978 | |
246d86b5 | 4979 | switch_count = &prev->nivcsw; |
d136122f | 4980 | |
dbfb089d | 4981 | /* |
d136122f PZ |
4982 | * We must load prev->state once (task_struct::state is volatile), such |
4983 | * that: | |
4984 | * | |
4985 | * - we form a control dependency vs deactivate_task() below. | |
4986 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 4987 | */ |
d136122f PZ |
4988 | prev_state = prev->state; |
4989 | if (!preempt && prev_state) { | |
dbfb089d | 4990 | if (signal_pending_state(prev_state, prev)) { |
1da177e4 | 4991 | prev->state = TASK_RUNNING; |
21aa9af0 | 4992 | } else { |
dbfb089d PZ |
4993 | prev->sched_contributes_to_load = |
4994 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
4995 | !(prev_state & TASK_NOLOAD) && | |
4996 | !(prev->flags & PF_FROZEN); | |
4997 | ||
4998 | if (prev->sched_contributes_to_load) | |
4999 | rq->nr_uninterruptible++; | |
5000 | ||
5001 | /* | |
5002 | * __schedule() ttwu() | |
d136122f PZ |
5003 | * prev_state = prev->state; if (p->on_rq && ...) |
5004 | * if (prev_state) goto out; | |
5005 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
5006 | * p->state = TASK_WAKING | |
5007 | * | |
5008 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
5009 | * |
5010 | * After this, schedule() must not care about p->state any more. | |
5011 | */ | |
bce4dc80 | 5012 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 5013 | |
e33a9bba TH |
5014 | if (prev->in_iowait) { |
5015 | atomic_inc(&rq->nr_iowait); | |
5016 | delayacct_blkio_start(); | |
5017 | } | |
21aa9af0 | 5018 | } |
dd41f596 | 5019 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5020 | } |
5021 | ||
d8ac8971 | 5022 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 5023 | clear_tsk_need_resched(prev); |
f27dde8d | 5024 | clear_preempt_need_resched(); |
1da177e4 | 5025 | |
1da177e4 | 5026 | if (likely(prev != next)) { |
1da177e4 | 5027 | rq->nr_switches++; |
5311a98f EB |
5028 | /* |
5029 | * RCU users of rcu_dereference(rq->curr) may not see | |
5030 | * changes to task_struct made by pick_next_task(). | |
5031 | */ | |
5032 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
5033 | /* |
5034 | * The membarrier system call requires each architecture | |
5035 | * to have a full memory barrier after updating | |
306e0604 MD |
5036 | * rq->curr, before returning to user-space. |
5037 | * | |
5038 | * Here are the schemes providing that barrier on the | |
5039 | * various architectures: | |
5040 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
5041 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
5042 | * - finish_lock_switch() for weakly-ordered | |
5043 | * architectures where spin_unlock is a full barrier, | |
5044 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
5045 | * is a RELEASE barrier), | |
22e4ebb9 | 5046 | */ |
1da177e4 LT |
5047 | ++*switch_count; |
5048 | ||
af449901 | 5049 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
5050 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
5051 | ||
c73464b1 | 5052 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
5053 | |
5054 | /* Also unlocks the rq: */ | |
5055 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 5056 | } else { |
cb42c9a3 | 5057 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 5058 | |
565790d2 PZ |
5059 | rq_unpin_lock(rq, &rf); |
5060 | __balance_callbacks(rq); | |
5061 | raw_spin_unlock_irq(&rq->lock); | |
5062 | } | |
1da177e4 | 5063 | } |
c259e01a | 5064 | |
9af6528e PZ |
5065 | void __noreturn do_task_dead(void) |
5066 | { | |
d1ccc66d | 5067 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 5068 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
5069 | |
5070 | /* Tell freezer to ignore us: */ | |
5071 | current->flags |= PF_NOFREEZE; | |
5072 | ||
9af6528e PZ |
5073 | __schedule(false); |
5074 | BUG(); | |
d1ccc66d IM |
5075 | |
5076 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 5077 | for (;;) |
d1ccc66d | 5078 | cpu_relax(); |
9af6528e PZ |
5079 | } |
5080 | ||
9c40cef2 TG |
5081 | static inline void sched_submit_work(struct task_struct *tsk) |
5082 | { | |
c1cecf88 SAS |
5083 | unsigned int task_flags; |
5084 | ||
b0fdc013 | 5085 | if (!tsk->state) |
9c40cef2 | 5086 | return; |
6d25be57 | 5087 | |
c1cecf88 | 5088 | task_flags = tsk->flags; |
6d25be57 TG |
5089 | /* |
5090 | * If a worker went to sleep, notify and ask workqueue whether | |
5091 | * it wants to wake up a task to maintain concurrency. | |
5092 | * As this function is called inside the schedule() context, | |
5093 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
5094 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
5095 | * requires it. | |
6d25be57 | 5096 | */ |
c1cecf88 | 5097 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 5098 | preempt_disable(); |
c1cecf88 | 5099 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
5100 | wq_worker_sleeping(tsk); |
5101 | else | |
5102 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
5103 | preempt_enable_no_resched(); |
5104 | } | |
5105 | ||
b0fdc013 SAS |
5106 | if (tsk_is_pi_blocked(tsk)) |
5107 | return; | |
5108 | ||
9c40cef2 TG |
5109 | /* |
5110 | * If we are going to sleep and we have plugged IO queued, | |
5111 | * make sure to submit it to avoid deadlocks. | |
5112 | */ | |
5113 | if (blk_needs_flush_plug(tsk)) | |
5114 | blk_schedule_flush_plug(tsk); | |
5115 | } | |
5116 | ||
6d25be57 TG |
5117 | static void sched_update_worker(struct task_struct *tsk) |
5118 | { | |
771b53d0 JA |
5119 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
5120 | if (tsk->flags & PF_WQ_WORKER) | |
5121 | wq_worker_running(tsk); | |
5122 | else | |
5123 | io_wq_worker_running(tsk); | |
5124 | } | |
6d25be57 TG |
5125 | } |
5126 | ||
722a9f92 | 5127 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 5128 | { |
9c40cef2 TG |
5129 | struct task_struct *tsk = current; |
5130 | ||
5131 | sched_submit_work(tsk); | |
bfd9b2b5 | 5132 | do { |
b30f0e3f | 5133 | preempt_disable(); |
fc13aeba | 5134 | __schedule(false); |
b30f0e3f | 5135 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 5136 | } while (need_resched()); |
6d25be57 | 5137 | sched_update_worker(tsk); |
c259e01a | 5138 | } |
1da177e4 LT |
5139 | EXPORT_SYMBOL(schedule); |
5140 | ||
8663effb SRV |
5141 | /* |
5142 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
5143 | * state (have scheduled out non-voluntarily) by making sure that all | |
5144 | * tasks have either left the run queue or have gone into user space. | |
5145 | * As idle tasks do not do either, they must not ever be preempted | |
5146 | * (schedule out non-voluntarily). | |
5147 | * | |
5148 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
5149 | * never enables preemption because it does not call sched_submit_work(). | |
5150 | */ | |
5151 | void __sched schedule_idle(void) | |
5152 | { | |
5153 | /* | |
5154 | * As this skips calling sched_submit_work(), which the idle task does | |
5155 | * regardless because that function is a nop when the task is in a | |
5156 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
5157 | * current task can be in any other state. Note, idle is always in the | |
5158 | * TASK_RUNNING state. | |
5159 | */ | |
5160 | WARN_ON_ONCE(current->state); | |
5161 | do { | |
5162 | __schedule(false); | |
5163 | } while (need_resched()); | |
5164 | } | |
5165 | ||
91d1aa43 | 5166 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 5167 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
5168 | { |
5169 | /* | |
5170 | * If we come here after a random call to set_need_resched(), | |
5171 | * or we have been woken up remotely but the IPI has not yet arrived, | |
5172 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
5173 | * we find a better solution. | |
7cc78f8f AL |
5174 | * |
5175 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 5176 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 5177 | * too frequently to make sense yet. |
20ab65e3 | 5178 | */ |
7cc78f8f | 5179 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 5180 | schedule(); |
7cc78f8f | 5181 | exception_exit(prev_state); |
20ab65e3 FW |
5182 | } |
5183 | #endif | |
5184 | ||
c5491ea7 TG |
5185 | /** |
5186 | * schedule_preempt_disabled - called with preemption disabled | |
5187 | * | |
5188 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
5189 | */ | |
5190 | void __sched schedule_preempt_disabled(void) | |
5191 | { | |
ba74c144 | 5192 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
5193 | schedule(); |
5194 | preempt_disable(); | |
5195 | } | |
5196 | ||
06b1f808 | 5197 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
5198 | { |
5199 | do { | |
47252cfb SR |
5200 | /* |
5201 | * Because the function tracer can trace preempt_count_sub() | |
5202 | * and it also uses preempt_enable/disable_notrace(), if | |
5203 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5204 | * by the function tracer will call this function again and | |
5205 | * cause infinite recursion. | |
5206 | * | |
5207 | * Preemption must be disabled here before the function | |
5208 | * tracer can trace. Break up preempt_disable() into two | |
5209 | * calls. One to disable preemption without fear of being | |
5210 | * traced. The other to still record the preemption latency, | |
5211 | * which can also be traced by the function tracer. | |
5212 | */ | |
499d7955 | 5213 | preempt_disable_notrace(); |
47252cfb | 5214 | preempt_latency_start(1); |
fc13aeba | 5215 | __schedule(true); |
47252cfb | 5216 | preempt_latency_stop(1); |
499d7955 | 5217 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
5218 | |
5219 | /* | |
5220 | * Check again in case we missed a preemption opportunity | |
5221 | * between schedule and now. | |
5222 | */ | |
a18b5d01 FW |
5223 | } while (need_resched()); |
5224 | } | |
5225 | ||
c1a280b6 | 5226 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 5227 | /* |
a49b4f40 VS |
5228 | * This is the entry point to schedule() from in-kernel preemption |
5229 | * off of preempt_enable. | |
1da177e4 | 5230 | */ |
722a9f92 | 5231 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 5232 | { |
1da177e4 LT |
5233 | /* |
5234 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5235 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5236 | */ |
fbb00b56 | 5237 | if (likely(!preemptible())) |
1da177e4 LT |
5238 | return; |
5239 | ||
a18b5d01 | 5240 | preempt_schedule_common(); |
1da177e4 | 5241 | } |
376e2424 | 5242 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 5243 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 5244 | |
009f60e2 | 5245 | /** |
4eaca0a8 | 5246 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
5247 | * |
5248 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
5249 | * recursion and tracing preempt enabling caused by the tracing | |
5250 | * infrastructure itself. But as tracing can happen in areas coming | |
5251 | * from userspace or just about to enter userspace, a preempt enable | |
5252 | * can occur before user_exit() is called. This will cause the scheduler | |
5253 | * to be called when the system is still in usermode. | |
5254 | * | |
5255 | * To prevent this, the preempt_enable_notrace will use this function | |
5256 | * instead of preempt_schedule() to exit user context if needed before | |
5257 | * calling the scheduler. | |
5258 | */ | |
4eaca0a8 | 5259 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
5260 | { |
5261 | enum ctx_state prev_ctx; | |
5262 | ||
5263 | if (likely(!preemptible())) | |
5264 | return; | |
5265 | ||
5266 | do { | |
47252cfb SR |
5267 | /* |
5268 | * Because the function tracer can trace preempt_count_sub() | |
5269 | * and it also uses preempt_enable/disable_notrace(), if | |
5270 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5271 | * by the function tracer will call this function again and | |
5272 | * cause infinite recursion. | |
5273 | * | |
5274 | * Preemption must be disabled here before the function | |
5275 | * tracer can trace. Break up preempt_disable() into two | |
5276 | * calls. One to disable preemption without fear of being | |
5277 | * traced. The other to still record the preemption latency, | |
5278 | * which can also be traced by the function tracer. | |
5279 | */ | |
3d8f74dd | 5280 | preempt_disable_notrace(); |
47252cfb | 5281 | preempt_latency_start(1); |
009f60e2 ON |
5282 | /* |
5283 | * Needs preempt disabled in case user_exit() is traced | |
5284 | * and the tracer calls preempt_enable_notrace() causing | |
5285 | * an infinite recursion. | |
5286 | */ | |
5287 | prev_ctx = exception_enter(); | |
fc13aeba | 5288 | __schedule(true); |
009f60e2 ON |
5289 | exception_exit(prev_ctx); |
5290 | ||
47252cfb | 5291 | preempt_latency_stop(1); |
3d8f74dd | 5292 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
5293 | } while (need_resched()); |
5294 | } | |
4eaca0a8 | 5295 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 5296 | |
c1a280b6 | 5297 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 LT |
5298 | |
5299 | /* | |
a49b4f40 | 5300 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5301 | * off of irq context. |
5302 | * Note, that this is called and return with irqs disabled. This will | |
5303 | * protect us against recursive calling from irq. | |
5304 | */ | |
722a9f92 | 5305 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 5306 | { |
b22366cd | 5307 | enum ctx_state prev_state; |
6478d880 | 5308 | |
2ed6e34f | 5309 | /* Catch callers which need to be fixed */ |
f27dde8d | 5310 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 5311 | |
b22366cd FW |
5312 | prev_state = exception_enter(); |
5313 | ||
3a5c359a | 5314 | do { |
3d8f74dd | 5315 | preempt_disable(); |
3a5c359a | 5316 | local_irq_enable(); |
fc13aeba | 5317 | __schedule(true); |
3a5c359a | 5318 | local_irq_disable(); |
3d8f74dd | 5319 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 5320 | } while (need_resched()); |
b22366cd FW |
5321 | |
5322 | exception_exit(prev_state); | |
1da177e4 LT |
5323 | } |
5324 | ||
ac6424b9 | 5325 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 5326 | void *key) |
1da177e4 | 5327 | { |
062d3f95 | 5328 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 5329 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 5330 | } |
1da177e4 LT |
5331 | EXPORT_SYMBOL(default_wake_function); |
5332 | ||
b29739f9 IM |
5333 | #ifdef CONFIG_RT_MUTEXES |
5334 | ||
acd58620 PZ |
5335 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
5336 | { | |
5337 | if (pi_task) | |
5338 | prio = min(prio, pi_task->prio); | |
5339 | ||
5340 | return prio; | |
5341 | } | |
5342 | ||
5343 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5344 | { | |
5345 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
5346 | ||
5347 | return __rt_effective_prio(pi_task, prio); | |
5348 | } | |
5349 | ||
b29739f9 IM |
5350 | /* |
5351 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
5352 | * @p: task to boost |
5353 | * @pi_task: donor task | |
b29739f9 IM |
5354 | * |
5355 | * This function changes the 'effective' priority of a task. It does | |
5356 | * not touch ->normal_prio like __setscheduler(). | |
5357 | * | |
c365c292 TG |
5358 | * Used by the rt_mutex code to implement priority inheritance |
5359 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 5360 | */ |
acd58620 | 5361 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 5362 | { |
acd58620 | 5363 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 5364 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 5365 | const struct sched_class *prev_class; |
eb580751 PZ |
5366 | struct rq_flags rf; |
5367 | struct rq *rq; | |
b29739f9 | 5368 | |
acd58620 PZ |
5369 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
5370 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
5371 | ||
5372 | /* | |
5373 | * If nothing changed; bail early. | |
5374 | */ | |
5375 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
5376 | return; | |
b29739f9 | 5377 | |
eb580751 | 5378 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 5379 | update_rq_clock(rq); |
acd58620 PZ |
5380 | /* |
5381 | * Set under pi_lock && rq->lock, such that the value can be used under | |
5382 | * either lock. | |
5383 | * | |
5384 | * Note that there is loads of tricky to make this pointer cache work | |
5385 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
5386 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
5387 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 5388 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
5389 | */ |
5390 | p->pi_top_task = pi_task; | |
5391 | ||
5392 | /* | |
5393 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
5394 | */ | |
5395 | if (prio == p->prio && !dl_prio(prio)) | |
5396 | goto out_unlock; | |
b29739f9 | 5397 | |
1c4dd99b TG |
5398 | /* |
5399 | * Idle task boosting is a nono in general. There is one | |
5400 | * exception, when PREEMPT_RT and NOHZ is active: | |
5401 | * | |
5402 | * The idle task calls get_next_timer_interrupt() and holds | |
5403 | * the timer wheel base->lock on the CPU and another CPU wants | |
5404 | * to access the timer (probably to cancel it). We can safely | |
5405 | * ignore the boosting request, as the idle CPU runs this code | |
5406 | * with interrupts disabled and will complete the lock | |
5407 | * protected section without being interrupted. So there is no | |
5408 | * real need to boost. | |
5409 | */ | |
5410 | if (unlikely(p == rq->idle)) { | |
5411 | WARN_ON(p != rq->curr); | |
5412 | WARN_ON(p->pi_blocked_on); | |
5413 | goto out_unlock; | |
5414 | } | |
5415 | ||
b91473ff | 5416 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 5417 | oldprio = p->prio; |
ff77e468 PZ |
5418 | |
5419 | if (oldprio == prio) | |
5420 | queue_flag &= ~DEQUEUE_MOVE; | |
5421 | ||
83ab0aa0 | 5422 | prev_class = p->sched_class; |
da0c1e65 | 5423 | queued = task_on_rq_queued(p); |
051a1d1a | 5424 | running = task_current(rq, p); |
da0c1e65 | 5425 | if (queued) |
ff77e468 | 5426 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 5427 | if (running) |
f3cd1c4e | 5428 | put_prev_task(rq, p); |
dd41f596 | 5429 | |
2d3d891d DF |
5430 | /* |
5431 | * Boosting condition are: | |
5432 | * 1. -rt task is running and holds mutex A | |
5433 | * --> -dl task blocks on mutex A | |
5434 | * | |
5435 | * 2. -dl task is running and holds mutex A | |
5436 | * --> -dl task blocks on mutex A and could preempt the | |
5437 | * running task | |
5438 | */ | |
5439 | if (dl_prio(prio)) { | |
466af29b | 5440 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
5441 | (pi_task && dl_prio(pi_task->prio) && |
5442 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 5443 | p->dl.dl_boosted = 1; |
ff77e468 | 5444 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
5445 | } else |
5446 | p->dl.dl_boosted = 0; | |
aab03e05 | 5447 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
5448 | } else if (rt_prio(prio)) { |
5449 | if (dl_prio(oldprio)) | |
5450 | p->dl.dl_boosted = 0; | |
5451 | if (oldprio < prio) | |
ff77e468 | 5452 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 5453 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
5454 | } else { |
5455 | if (dl_prio(oldprio)) | |
5456 | p->dl.dl_boosted = 0; | |
746db944 BS |
5457 | if (rt_prio(oldprio)) |
5458 | p->rt.timeout = 0; | |
dd41f596 | 5459 | p->sched_class = &fair_sched_class; |
2d3d891d | 5460 | } |
dd41f596 | 5461 | |
b29739f9 IM |
5462 | p->prio = prio; |
5463 | ||
da0c1e65 | 5464 | if (queued) |
ff77e468 | 5465 | enqueue_task(rq, p, queue_flag); |
a399d233 | 5466 | if (running) |
03b7fad1 | 5467 | set_next_task(rq, p); |
cb469845 | 5468 | |
da7a735e | 5469 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 5470 | out_unlock: |
d1ccc66d IM |
5471 | /* Avoid rq from going away on us: */ |
5472 | preempt_disable(); | |
4c9a4bc8 | 5473 | |
565790d2 PZ |
5474 | rq_unpin_lock(rq, &rf); |
5475 | __balance_callbacks(rq); | |
5476 | raw_spin_unlock(&rq->lock); | |
5477 | ||
4c9a4bc8 | 5478 | preempt_enable(); |
b29739f9 | 5479 | } |
acd58620 PZ |
5480 | #else |
5481 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5482 | { | |
5483 | return prio; | |
5484 | } | |
b29739f9 | 5485 | #endif |
d50dde5a | 5486 | |
36c8b586 | 5487 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5488 | { |
49bd21ef | 5489 | bool queued, running; |
53a23364 | 5490 | int old_prio; |
eb580751 | 5491 | struct rq_flags rf; |
70b97a7f | 5492 | struct rq *rq; |
1da177e4 | 5493 | |
75e45d51 | 5494 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
5495 | return; |
5496 | /* | |
5497 | * We have to be careful, if called from sys_setpriority(), | |
5498 | * the task might be in the middle of scheduling on another CPU. | |
5499 | */ | |
eb580751 | 5500 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
5501 | update_rq_clock(rq); |
5502 | ||
1da177e4 LT |
5503 | /* |
5504 | * The RT priorities are set via sched_setscheduler(), but we still | |
5505 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 5506 | * it won't have any effect on scheduling until the task is |
aab03e05 | 5507 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 5508 | */ |
aab03e05 | 5509 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
5510 | p->static_prio = NICE_TO_PRIO(nice); |
5511 | goto out_unlock; | |
5512 | } | |
da0c1e65 | 5513 | queued = task_on_rq_queued(p); |
49bd21ef | 5514 | running = task_current(rq, p); |
da0c1e65 | 5515 | if (queued) |
7a57f32a | 5516 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
5517 | if (running) |
5518 | put_prev_task(rq, p); | |
1da177e4 | 5519 | |
1da177e4 | 5520 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 5521 | set_load_weight(p, true); |
b29739f9 IM |
5522 | old_prio = p->prio; |
5523 | p->prio = effective_prio(p); | |
1da177e4 | 5524 | |
5443a0be | 5525 | if (queued) |
7134b3e9 | 5526 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 5527 | if (running) |
03b7fad1 | 5528 | set_next_task(rq, p); |
5443a0be FW |
5529 | |
5530 | /* | |
5531 | * If the task increased its priority or is running and | |
5532 | * lowered its priority, then reschedule its CPU: | |
5533 | */ | |
5534 | p->sched_class->prio_changed(rq, p, old_prio); | |
5535 | ||
1da177e4 | 5536 | out_unlock: |
eb580751 | 5537 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 5538 | } |
1da177e4 LT |
5539 | EXPORT_SYMBOL(set_user_nice); |
5540 | ||
e43379f1 MM |
5541 | /* |
5542 | * can_nice - check if a task can reduce its nice value | |
5543 | * @p: task | |
5544 | * @nice: nice value | |
5545 | */ | |
36c8b586 | 5546 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5547 | { |
d1ccc66d | 5548 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 5549 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 5550 | |
78d7d407 | 5551 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5552 | capable(CAP_SYS_NICE)); |
5553 | } | |
5554 | ||
1da177e4 LT |
5555 | #ifdef __ARCH_WANT_SYS_NICE |
5556 | ||
5557 | /* | |
5558 | * sys_nice - change the priority of the current process. | |
5559 | * @increment: priority increment | |
5560 | * | |
5561 | * sys_setpriority is a more generic, but much slower function that | |
5562 | * does similar things. | |
5563 | */ | |
5add95d4 | 5564 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5565 | { |
48f24c4d | 5566 | long nice, retval; |
1da177e4 LT |
5567 | |
5568 | /* | |
5569 | * Setpriority might change our priority at the same moment. | |
5570 | * We don't have to worry. Conceptually one call occurs first | |
5571 | * and we have a single winner. | |
5572 | */ | |
a9467fa3 | 5573 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 5574 | nice = task_nice(current) + increment; |
1da177e4 | 5575 | |
a9467fa3 | 5576 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
5577 | if (increment < 0 && !can_nice(current, nice)) |
5578 | return -EPERM; | |
5579 | ||
1da177e4 LT |
5580 | retval = security_task_setnice(current, nice); |
5581 | if (retval) | |
5582 | return retval; | |
5583 | ||
5584 | set_user_nice(current, nice); | |
5585 | return 0; | |
5586 | } | |
5587 | ||
5588 | #endif | |
5589 | ||
5590 | /** | |
5591 | * task_prio - return the priority value of a given task. | |
5592 | * @p: the task in question. | |
5593 | * | |
e69f6186 | 5594 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
5595 | * RT tasks are offset by -200. Normal tasks are centered |
5596 | * around 0, value goes from -16 to +15. | |
5597 | */ | |
36c8b586 | 5598 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5599 | { |
5600 | return p->prio - MAX_RT_PRIO; | |
5601 | } | |
5602 | ||
1da177e4 | 5603 | /** |
d1ccc66d | 5604 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 5605 | * @cpu: the processor in question. |
e69f6186 YB |
5606 | * |
5607 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
5608 | */ |
5609 | int idle_cpu(int cpu) | |
5610 | { | |
908a3283 TG |
5611 | struct rq *rq = cpu_rq(cpu); |
5612 | ||
5613 | if (rq->curr != rq->idle) | |
5614 | return 0; | |
5615 | ||
5616 | if (rq->nr_running) | |
5617 | return 0; | |
5618 | ||
5619 | #ifdef CONFIG_SMP | |
126c2092 | 5620 | if (rq->ttwu_pending) |
908a3283 TG |
5621 | return 0; |
5622 | #endif | |
5623 | ||
5624 | return 1; | |
1da177e4 LT |
5625 | } |
5626 | ||
943d355d RJ |
5627 | /** |
5628 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
5629 | * @cpu: the CPU in question. | |
5630 | * | |
5631 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
5632 | */ | |
5633 | int available_idle_cpu(int cpu) | |
5634 | { | |
5635 | if (!idle_cpu(cpu)) | |
5636 | return 0; | |
5637 | ||
247f2f6f RJ |
5638 | if (vcpu_is_preempted(cpu)) |
5639 | return 0; | |
5640 | ||
908a3283 | 5641 | return 1; |
1da177e4 LT |
5642 | } |
5643 | ||
1da177e4 | 5644 | /** |
d1ccc66d | 5645 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 5646 | * @cpu: the processor in question. |
e69f6186 | 5647 | * |
d1ccc66d | 5648 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 5649 | */ |
36c8b586 | 5650 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5651 | { |
5652 | return cpu_rq(cpu)->idle; | |
5653 | } | |
5654 | ||
5655 | /** | |
5656 | * find_process_by_pid - find a process with a matching PID value. | |
5657 | * @pid: the pid in question. | |
e69f6186 YB |
5658 | * |
5659 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 5660 | */ |
a9957449 | 5661 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5662 | { |
228ebcbe | 5663 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5664 | } |
5665 | ||
c13db6b1 SR |
5666 | /* |
5667 | * sched_setparam() passes in -1 for its policy, to let the functions | |
5668 | * it calls know not to change it. | |
5669 | */ | |
5670 | #define SETPARAM_POLICY -1 | |
5671 | ||
c365c292 TG |
5672 | static void __setscheduler_params(struct task_struct *p, |
5673 | const struct sched_attr *attr) | |
1da177e4 | 5674 | { |
d50dde5a DF |
5675 | int policy = attr->sched_policy; |
5676 | ||
c13db6b1 | 5677 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
5678 | policy = p->policy; |
5679 | ||
1da177e4 | 5680 | p->policy = policy; |
d50dde5a | 5681 | |
aab03e05 DF |
5682 | if (dl_policy(policy)) |
5683 | __setparam_dl(p, attr); | |
39fd8fd2 | 5684 | else if (fair_policy(policy)) |
d50dde5a DF |
5685 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
5686 | ||
39fd8fd2 PZ |
5687 | /* |
5688 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
5689 | * !rt_policy. Always setting this ensures that things like | |
5690 | * getparam()/getattr() don't report silly values for !rt tasks. | |
5691 | */ | |
5692 | p->rt_priority = attr->sched_priority; | |
383afd09 | 5693 | p->normal_prio = normal_prio(p); |
9059393e | 5694 | set_load_weight(p, true); |
c365c292 | 5695 | } |
39fd8fd2 | 5696 | |
c365c292 TG |
5697 | /* Actually do priority change: must hold pi & rq lock. */ |
5698 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 5699 | const struct sched_attr *attr, bool keep_boost) |
c365c292 | 5700 | { |
a509a7cd PB |
5701 | /* |
5702 | * If params can't change scheduling class changes aren't allowed | |
5703 | * either. | |
5704 | */ | |
5705 | if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) | |
5706 | return; | |
5707 | ||
c365c292 | 5708 | __setscheduler_params(p, attr); |
d50dde5a | 5709 | |
383afd09 | 5710 | /* |
0782e63b TG |
5711 | * Keep a potential priority boosting if called from |
5712 | * sched_setscheduler(). | |
383afd09 | 5713 | */ |
acd58620 | 5714 | p->prio = normal_prio(p); |
0782e63b | 5715 | if (keep_boost) |
acd58620 | 5716 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 5717 | |
aab03e05 DF |
5718 | if (dl_prio(p->prio)) |
5719 | p->sched_class = &dl_sched_class; | |
5720 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
5721 | p->sched_class = &rt_sched_class; |
5722 | else | |
5723 | p->sched_class = &fair_sched_class; | |
1da177e4 | 5724 | } |
aab03e05 | 5725 | |
c69e8d9c | 5726 | /* |
d1ccc66d | 5727 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
5728 | */ |
5729 | static bool check_same_owner(struct task_struct *p) | |
5730 | { | |
5731 | const struct cred *cred = current_cred(), *pcred; | |
5732 | bool match; | |
5733 | ||
5734 | rcu_read_lock(); | |
5735 | pcred = __task_cred(p); | |
9c806aa0 EB |
5736 | match = (uid_eq(cred->euid, pcred->euid) || |
5737 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
5738 | rcu_read_unlock(); |
5739 | return match; | |
5740 | } | |
5741 | ||
d50dde5a DF |
5742 | static int __sched_setscheduler(struct task_struct *p, |
5743 | const struct sched_attr *attr, | |
dbc7f069 | 5744 | bool user, bool pi) |
1da177e4 | 5745 | { |
383afd09 SR |
5746 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
5747 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 5748 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 5749 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 5750 | const struct sched_class *prev_class; |
565790d2 | 5751 | struct callback_head *head; |
eb580751 | 5752 | struct rq_flags rf; |
ca94c442 | 5753 | int reset_on_fork; |
7a57f32a | 5754 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 5755 | struct rq *rq; |
1da177e4 | 5756 | |
896bbb25 SRV |
5757 | /* The pi code expects interrupts enabled */ |
5758 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 5759 | recheck: |
d1ccc66d | 5760 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
5761 | if (policy < 0) { |
5762 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5763 | policy = oldpolicy = p->policy; |
ca94c442 | 5764 | } else { |
7479f3c9 | 5765 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 5766 | |
20f9cd2a | 5767 | if (!valid_policy(policy)) |
ca94c442 LP |
5768 | return -EINVAL; |
5769 | } | |
5770 | ||
794a56eb | 5771 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
5772 | return -EINVAL; |
5773 | ||
1da177e4 LT |
5774 | /* |
5775 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5776 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5777 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 5778 | */ |
0bb040a4 | 5779 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 5780 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5781 | return -EINVAL; |
aab03e05 DF |
5782 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
5783 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
5784 | return -EINVAL; |
5785 | ||
37e4ab3f OC |
5786 | /* |
5787 | * Allow unprivileged RT tasks to decrease priority: | |
5788 | */ | |
961ccddd | 5789 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 5790 | if (fair_policy(policy)) { |
d0ea0268 | 5791 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 5792 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
5793 | return -EPERM; |
5794 | } | |
5795 | ||
e05606d3 | 5796 | if (rt_policy(policy)) { |
a44702e8 ON |
5797 | unsigned long rlim_rtprio = |
5798 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 5799 | |
d1ccc66d | 5800 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
5801 | if (policy != p->policy && !rlim_rtprio) |
5802 | return -EPERM; | |
5803 | ||
d1ccc66d | 5804 | /* Can't increase priority: */ |
d50dde5a DF |
5805 | if (attr->sched_priority > p->rt_priority && |
5806 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
5807 | return -EPERM; |
5808 | } | |
c02aa73b | 5809 | |
d44753b8 JL |
5810 | /* |
5811 | * Can't set/change SCHED_DEADLINE policy at all for now | |
5812 | * (safest behavior); in the future we would like to allow | |
5813 | * unprivileged DL tasks to increase their relative deadline | |
5814 | * or reduce their runtime (both ways reducing utilization) | |
5815 | */ | |
5816 | if (dl_policy(policy)) | |
5817 | return -EPERM; | |
5818 | ||
dd41f596 | 5819 | /* |
c02aa73b DH |
5820 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5821 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5822 | */ |
1da1843f | 5823 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 5824 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
5825 | return -EPERM; |
5826 | } | |
5fe1d75f | 5827 | |
d1ccc66d | 5828 | /* Can't change other user's priorities: */ |
c69e8d9c | 5829 | if (!check_same_owner(p)) |
37e4ab3f | 5830 | return -EPERM; |
ca94c442 | 5831 | |
d1ccc66d | 5832 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
5833 | if (p->sched_reset_on_fork && !reset_on_fork) |
5834 | return -EPERM; | |
37e4ab3f | 5835 | } |
1da177e4 | 5836 | |
725aad24 | 5837 | if (user) { |
794a56eb JL |
5838 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
5839 | return -EINVAL; | |
5840 | ||
b0ae1981 | 5841 | retval = security_task_setscheduler(p); |
725aad24 JF |
5842 | if (retval) |
5843 | return retval; | |
5844 | } | |
5845 | ||
a509a7cd PB |
5846 | /* Update task specific "requested" clamps */ |
5847 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
5848 | retval = uclamp_validate(p, attr); | |
5849 | if (retval) | |
5850 | return retval; | |
5851 | } | |
5852 | ||
710da3c8 JL |
5853 | if (pi) |
5854 | cpuset_read_lock(); | |
5855 | ||
b29739f9 | 5856 | /* |
d1ccc66d | 5857 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 5858 | * changing the priority of the task: |
0122ec5b | 5859 | * |
25985edc | 5860 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5861 | * runqueue lock must be held. |
5862 | */ | |
eb580751 | 5863 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 5864 | update_rq_clock(rq); |
dc61b1d6 | 5865 | |
34f971f6 | 5866 | /* |
d1ccc66d | 5867 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
5868 | */ |
5869 | if (p == rq->stop) { | |
4b211f2b MP |
5870 | retval = -EINVAL; |
5871 | goto unlock; | |
34f971f6 PZ |
5872 | } |
5873 | ||
a51e9198 | 5874 | /* |
d6b1e911 TG |
5875 | * If not changing anything there's no need to proceed further, |
5876 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 5877 | */ |
d50dde5a | 5878 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 5879 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
5880 | goto change; |
5881 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
5882 | goto change; | |
75381608 | 5883 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 5884 | goto change; |
a509a7cd PB |
5885 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
5886 | goto change; | |
d50dde5a | 5887 | |
d6b1e911 | 5888 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
5889 | retval = 0; |
5890 | goto unlock; | |
a51e9198 | 5891 | } |
d50dde5a | 5892 | change: |
a51e9198 | 5893 | |
dc61b1d6 | 5894 | if (user) { |
332ac17e | 5895 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
5896 | /* |
5897 | * Do not allow realtime tasks into groups that have no runtime | |
5898 | * assigned. | |
5899 | */ | |
5900 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5901 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5902 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
5903 | retval = -EPERM; |
5904 | goto unlock; | |
dc61b1d6 | 5905 | } |
dc61b1d6 | 5906 | #endif |
332ac17e | 5907 | #ifdef CONFIG_SMP |
794a56eb JL |
5908 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
5909 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 5910 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
5911 | |
5912 | /* | |
5913 | * Don't allow tasks with an affinity mask smaller than | |
5914 | * the entire root_domain to become SCHED_DEADLINE. We | |
5915 | * will also fail if there's no bandwidth available. | |
5916 | */ | |
3bd37062 | 5917 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 5918 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
5919 | retval = -EPERM; |
5920 | goto unlock; | |
332ac17e DF |
5921 | } |
5922 | } | |
5923 | #endif | |
5924 | } | |
dc61b1d6 | 5925 | |
d1ccc66d | 5926 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
5927 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
5928 | policy = oldpolicy = -1; | |
eb580751 | 5929 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
5930 | if (pi) |
5931 | cpuset_read_unlock(); | |
1da177e4 LT |
5932 | goto recheck; |
5933 | } | |
332ac17e DF |
5934 | |
5935 | /* | |
5936 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
5937 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
5938 | * is available. | |
5939 | */ | |
06a76fe0 | 5940 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
5941 | retval = -EBUSY; |
5942 | goto unlock; | |
332ac17e DF |
5943 | } |
5944 | ||
c365c292 TG |
5945 | p->sched_reset_on_fork = reset_on_fork; |
5946 | oldprio = p->prio; | |
5947 | ||
dbc7f069 PZ |
5948 | if (pi) { |
5949 | /* | |
5950 | * Take priority boosted tasks into account. If the new | |
5951 | * effective priority is unchanged, we just store the new | |
5952 | * normal parameters and do not touch the scheduler class and | |
5953 | * the runqueue. This will be done when the task deboost | |
5954 | * itself. | |
5955 | */ | |
acd58620 | 5956 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
5957 | if (new_effective_prio == oldprio) |
5958 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
5959 | } |
5960 | ||
da0c1e65 | 5961 | queued = task_on_rq_queued(p); |
051a1d1a | 5962 | running = task_current(rq, p); |
da0c1e65 | 5963 | if (queued) |
ff77e468 | 5964 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 5965 | if (running) |
f3cd1c4e | 5966 | put_prev_task(rq, p); |
f6b53205 | 5967 | |
83ab0aa0 | 5968 | prev_class = p->sched_class; |
a509a7cd | 5969 | |
dbc7f069 | 5970 | __setscheduler(rq, p, attr, pi); |
a509a7cd | 5971 | __setscheduler_uclamp(p, attr); |
f6b53205 | 5972 | |
da0c1e65 | 5973 | if (queued) { |
81a44c54 TG |
5974 | /* |
5975 | * We enqueue to tail when the priority of a task is | |
5976 | * increased (user space view). | |
5977 | */ | |
ff77e468 PZ |
5978 | if (oldprio < p->prio) |
5979 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 5980 | |
ff77e468 | 5981 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 5982 | } |
a399d233 | 5983 | if (running) |
03b7fad1 | 5984 | set_next_task(rq, p); |
cb469845 | 5985 | |
da7a735e | 5986 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
5987 | |
5988 | /* Avoid rq from going away on us: */ | |
5989 | preempt_disable(); | |
565790d2 | 5990 | head = splice_balance_callbacks(rq); |
eb580751 | 5991 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 5992 | |
710da3c8 JL |
5993 | if (pi) { |
5994 | cpuset_read_unlock(); | |
dbc7f069 | 5995 | rt_mutex_adjust_pi(p); |
710da3c8 | 5996 | } |
95e02ca9 | 5997 | |
d1ccc66d | 5998 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 5999 | balance_callbacks(rq, head); |
4c9a4bc8 | 6000 | preempt_enable(); |
95e02ca9 | 6001 | |
1da177e4 | 6002 | return 0; |
4b211f2b MP |
6003 | |
6004 | unlock: | |
6005 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
6006 | if (pi) |
6007 | cpuset_read_unlock(); | |
4b211f2b | 6008 | return retval; |
1da177e4 | 6009 | } |
961ccddd | 6010 | |
7479f3c9 PZ |
6011 | static int _sched_setscheduler(struct task_struct *p, int policy, |
6012 | const struct sched_param *param, bool check) | |
6013 | { | |
6014 | struct sched_attr attr = { | |
6015 | .sched_policy = policy, | |
6016 | .sched_priority = param->sched_priority, | |
6017 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
6018 | }; | |
6019 | ||
c13db6b1 SR |
6020 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
6021 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
6022 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
6023 | policy &= ~SCHED_RESET_ON_FORK; | |
6024 | attr.sched_policy = policy; | |
6025 | } | |
6026 | ||
dbc7f069 | 6027 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 6028 | } |
961ccddd RR |
6029 | /** |
6030 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6031 | * @p: the task in question. | |
6032 | * @policy: new policy. | |
6033 | * @param: structure containing the new RT priority. | |
6034 | * | |
7318d4cc PZ |
6035 | * Use sched_set_fifo(), read its comment. |
6036 | * | |
e69f6186 YB |
6037 | * Return: 0 on success. An error code otherwise. |
6038 | * | |
961ccddd RR |
6039 | * NOTE that the task may be already dead. |
6040 | */ | |
6041 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 6042 | const struct sched_param *param) |
961ccddd | 6043 | { |
7479f3c9 | 6044 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 6045 | } |
1da177e4 | 6046 | |
d50dde5a DF |
6047 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
6048 | { | |
dbc7f069 | 6049 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 6050 | } |
d50dde5a | 6051 | |
794a56eb JL |
6052 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
6053 | { | |
6054 | return __sched_setscheduler(p, attr, false, true); | |
6055 | } | |
6056 | ||
961ccddd RR |
6057 | /** |
6058 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6059 | * @p: the task in question. | |
6060 | * @policy: new policy. | |
6061 | * @param: structure containing the new RT priority. | |
6062 | * | |
6063 | * Just like sched_setscheduler, only don't bother checking if the | |
6064 | * current context has permission. For example, this is needed in | |
6065 | * stop_machine(): we create temporary high priority worker threads, | |
6066 | * but our caller might not have that capability. | |
e69f6186 YB |
6067 | * |
6068 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
6069 | */ |
6070 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 6071 | const struct sched_param *param) |
961ccddd | 6072 | { |
7479f3c9 | 6073 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
6074 | } |
6075 | ||
7318d4cc PZ |
6076 | /* |
6077 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
6078 | * incapable of resource management, which is the one thing an OS really should | |
6079 | * be doing. | |
6080 | * | |
6081 | * This is of course the reason it is limited to privileged users only. | |
6082 | * | |
6083 | * Worse still; it is fundamentally impossible to compose static priority | |
6084 | * workloads. You cannot take two correctly working static prio workloads | |
6085 | * and smash them together and still expect them to work. | |
6086 | * | |
6087 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
6088 | * | |
6089 | * MAX_RT_PRIO / 2 | |
6090 | * | |
6091 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
6092 | * know enough information to make a sensible choice. | |
6093 | */ | |
8b700983 | 6094 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
6095 | { |
6096 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 6097 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6098 | } |
6099 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
6100 | ||
6101 | /* | |
6102 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
6103 | */ | |
8b700983 | 6104 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
6105 | { |
6106 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 6107 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6108 | } |
6109 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
6110 | ||
8b700983 | 6111 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
6112 | { |
6113 | struct sched_attr attr = { | |
6114 | .sched_policy = SCHED_NORMAL, | |
6115 | .sched_nice = nice, | |
6116 | }; | |
8b700983 | 6117 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
6118 | } |
6119 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 6120 | |
95cdf3b7 IM |
6121 | static int |
6122 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6123 | { |
1da177e4 LT |
6124 | struct sched_param lparam; |
6125 | struct task_struct *p; | |
36c8b586 | 6126 | int retval; |
1da177e4 LT |
6127 | |
6128 | if (!param || pid < 0) | |
6129 | return -EINVAL; | |
6130 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6131 | return -EFAULT; | |
5fe1d75f ON |
6132 | |
6133 | rcu_read_lock(); | |
6134 | retval = -ESRCH; | |
1da177e4 | 6135 | p = find_process_by_pid(pid); |
710da3c8 JL |
6136 | if (likely(p)) |
6137 | get_task_struct(p); | |
5fe1d75f | 6138 | rcu_read_unlock(); |
36c8b586 | 6139 | |
710da3c8 JL |
6140 | if (likely(p)) { |
6141 | retval = sched_setscheduler(p, policy, &lparam); | |
6142 | put_task_struct(p); | |
6143 | } | |
6144 | ||
1da177e4 LT |
6145 | return retval; |
6146 | } | |
6147 | ||
d50dde5a DF |
6148 | /* |
6149 | * Mimics kernel/events/core.c perf_copy_attr(). | |
6150 | */ | |
d1ccc66d | 6151 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
6152 | { |
6153 | u32 size; | |
6154 | int ret; | |
6155 | ||
d1ccc66d | 6156 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
6157 | memset(attr, 0, sizeof(*attr)); |
6158 | ||
6159 | ret = get_user(size, &uattr->size); | |
6160 | if (ret) | |
6161 | return ret; | |
6162 | ||
d1ccc66d IM |
6163 | /* ABI compatibility quirk: */ |
6164 | if (!size) | |
d50dde5a | 6165 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 6166 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
6167 | goto err_size; |
6168 | ||
dff3a85f AS |
6169 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
6170 | if (ret) { | |
6171 | if (ret == -E2BIG) | |
6172 | goto err_size; | |
6173 | return ret; | |
d50dde5a DF |
6174 | } |
6175 | ||
a509a7cd PB |
6176 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
6177 | size < SCHED_ATTR_SIZE_VER1) | |
6178 | return -EINVAL; | |
6179 | ||
d50dde5a | 6180 | /* |
d1ccc66d | 6181 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
6182 | * to be strict and return an error on out-of-bounds values? |
6183 | */ | |
75e45d51 | 6184 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 6185 | |
e78c7bca | 6186 | return 0; |
d50dde5a DF |
6187 | |
6188 | err_size: | |
6189 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 6190 | return -E2BIG; |
d50dde5a DF |
6191 | } |
6192 | ||
1da177e4 LT |
6193 | /** |
6194 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6195 | * @pid: the pid in question. | |
6196 | * @policy: new policy. | |
6197 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6198 | * |
6199 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6200 | */ |
d1ccc66d | 6201 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 6202 | { |
c21761f1 JB |
6203 | if (policy < 0) |
6204 | return -EINVAL; | |
6205 | ||
1da177e4 LT |
6206 | return do_sched_setscheduler(pid, policy, param); |
6207 | } | |
6208 | ||
6209 | /** | |
6210 | * sys_sched_setparam - set/change the RT priority of a thread | |
6211 | * @pid: the pid in question. | |
6212 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6213 | * |
6214 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6215 | */ |
5add95d4 | 6216 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6217 | { |
c13db6b1 | 6218 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
6219 | } |
6220 | ||
d50dde5a DF |
6221 | /** |
6222 | * sys_sched_setattr - same as above, but with extended sched_attr | |
6223 | * @pid: the pid in question. | |
5778fccf | 6224 | * @uattr: structure containing the extended parameters. |
db66d756 | 6225 | * @flags: for future extension. |
d50dde5a | 6226 | */ |
6d35ab48 PZ |
6227 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
6228 | unsigned int, flags) | |
d50dde5a DF |
6229 | { |
6230 | struct sched_attr attr; | |
6231 | struct task_struct *p; | |
6232 | int retval; | |
6233 | ||
6d35ab48 | 6234 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
6235 | return -EINVAL; |
6236 | ||
143cf23d MK |
6237 | retval = sched_copy_attr(uattr, &attr); |
6238 | if (retval) | |
6239 | return retval; | |
d50dde5a | 6240 | |
b14ed2c2 | 6241 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 6242 | return -EINVAL; |
1d6362fa PB |
6243 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
6244 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
6245 | |
6246 | rcu_read_lock(); | |
6247 | retval = -ESRCH; | |
6248 | p = find_process_by_pid(pid); | |
a509a7cd PB |
6249 | if (likely(p)) |
6250 | get_task_struct(p); | |
d50dde5a DF |
6251 | rcu_read_unlock(); |
6252 | ||
a509a7cd PB |
6253 | if (likely(p)) { |
6254 | retval = sched_setattr(p, &attr); | |
6255 | put_task_struct(p); | |
6256 | } | |
6257 | ||
d50dde5a DF |
6258 | return retval; |
6259 | } | |
6260 | ||
1da177e4 LT |
6261 | /** |
6262 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6263 | * @pid: the pid in question. | |
e69f6186 YB |
6264 | * |
6265 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
6266 | * code. | |
1da177e4 | 6267 | */ |
5add95d4 | 6268 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6269 | { |
36c8b586 | 6270 | struct task_struct *p; |
3a5c359a | 6271 | int retval; |
1da177e4 LT |
6272 | |
6273 | if (pid < 0) | |
3a5c359a | 6274 | return -EINVAL; |
1da177e4 LT |
6275 | |
6276 | retval = -ESRCH; | |
5fe85be0 | 6277 | rcu_read_lock(); |
1da177e4 LT |
6278 | p = find_process_by_pid(pid); |
6279 | if (p) { | |
6280 | retval = security_task_getscheduler(p); | |
6281 | if (!retval) | |
ca94c442 LP |
6282 | retval = p->policy |
6283 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 6284 | } |
5fe85be0 | 6285 | rcu_read_unlock(); |
1da177e4 LT |
6286 | return retval; |
6287 | } | |
6288 | ||
6289 | /** | |
ca94c442 | 6290 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6291 | * @pid: the pid in question. |
6292 | * @param: structure containing the RT priority. | |
e69f6186 YB |
6293 | * |
6294 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
6295 | * code. | |
1da177e4 | 6296 | */ |
5add95d4 | 6297 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6298 | { |
ce5f7f82 | 6299 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 6300 | struct task_struct *p; |
3a5c359a | 6301 | int retval; |
1da177e4 LT |
6302 | |
6303 | if (!param || pid < 0) | |
3a5c359a | 6304 | return -EINVAL; |
1da177e4 | 6305 | |
5fe85be0 | 6306 | rcu_read_lock(); |
1da177e4 LT |
6307 | p = find_process_by_pid(pid); |
6308 | retval = -ESRCH; | |
6309 | if (!p) | |
6310 | goto out_unlock; | |
6311 | ||
6312 | retval = security_task_getscheduler(p); | |
6313 | if (retval) | |
6314 | goto out_unlock; | |
6315 | ||
ce5f7f82 PZ |
6316 | if (task_has_rt_policy(p)) |
6317 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 6318 | rcu_read_unlock(); |
1da177e4 LT |
6319 | |
6320 | /* | |
6321 | * This one might sleep, we cannot do it with a spinlock held ... | |
6322 | */ | |
6323 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6324 | ||
1da177e4 LT |
6325 | return retval; |
6326 | ||
6327 | out_unlock: | |
5fe85be0 | 6328 | rcu_read_unlock(); |
1da177e4 LT |
6329 | return retval; |
6330 | } | |
6331 | ||
1251201c IM |
6332 | /* |
6333 | * Copy the kernel size attribute structure (which might be larger | |
6334 | * than what user-space knows about) to user-space. | |
6335 | * | |
6336 | * Note that all cases are valid: user-space buffer can be larger or | |
6337 | * smaller than the kernel-space buffer. The usual case is that both | |
6338 | * have the same size. | |
6339 | */ | |
6340 | static int | |
6341 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
6342 | struct sched_attr *kattr, | |
6343 | unsigned int usize) | |
d50dde5a | 6344 | { |
1251201c | 6345 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 6346 | |
96d4f267 | 6347 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
6348 | return -EFAULT; |
6349 | ||
6350 | /* | |
1251201c IM |
6351 | * sched_getattr() ABI forwards and backwards compatibility: |
6352 | * | |
6353 | * If usize == ksize then we just copy everything to user-space and all is good. | |
6354 | * | |
6355 | * If usize < ksize then we only copy as much as user-space has space for, | |
6356 | * this keeps ABI compatibility as well. We skip the rest. | |
6357 | * | |
6358 | * If usize > ksize then user-space is using a newer version of the ABI, | |
6359 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
6360 | * detect the kernel's knowledge of attributes from the attr->size value | |
6361 | * which is set to ksize in this case. | |
d50dde5a | 6362 | */ |
1251201c | 6363 | kattr->size = min(usize, ksize); |
d50dde5a | 6364 | |
1251201c | 6365 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
6366 | return -EFAULT; |
6367 | ||
22400674 | 6368 | return 0; |
d50dde5a DF |
6369 | } |
6370 | ||
6371 | /** | |
aab03e05 | 6372 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 6373 | * @pid: the pid in question. |
5778fccf | 6374 | * @uattr: structure containing the extended parameters. |
dff3a85f | 6375 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 6376 | * @flags: for future extension. |
d50dde5a | 6377 | */ |
6d35ab48 | 6378 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 6379 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 6380 | { |
1251201c | 6381 | struct sched_attr kattr = { }; |
d50dde5a DF |
6382 | struct task_struct *p; |
6383 | int retval; | |
6384 | ||
1251201c IM |
6385 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
6386 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
6387 | return -EINVAL; |
6388 | ||
6389 | rcu_read_lock(); | |
6390 | p = find_process_by_pid(pid); | |
6391 | retval = -ESRCH; | |
6392 | if (!p) | |
6393 | goto out_unlock; | |
6394 | ||
6395 | retval = security_task_getscheduler(p); | |
6396 | if (retval) | |
6397 | goto out_unlock; | |
6398 | ||
1251201c | 6399 | kattr.sched_policy = p->policy; |
7479f3c9 | 6400 | if (p->sched_reset_on_fork) |
1251201c | 6401 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
aab03e05 | 6402 | if (task_has_dl_policy(p)) |
1251201c | 6403 | __getparam_dl(p, &kattr); |
aab03e05 | 6404 | else if (task_has_rt_policy(p)) |
1251201c | 6405 | kattr.sched_priority = p->rt_priority; |
d50dde5a | 6406 | else |
1251201c | 6407 | kattr.sched_nice = task_nice(p); |
d50dde5a | 6408 | |
a509a7cd | 6409 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
6410 | /* |
6411 | * This could race with another potential updater, but this is fine | |
6412 | * because it'll correctly read the old or the new value. We don't need | |
6413 | * to guarantee who wins the race as long as it doesn't return garbage. | |
6414 | */ | |
1251201c IM |
6415 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
6416 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
6417 | #endif |
6418 | ||
d50dde5a DF |
6419 | rcu_read_unlock(); |
6420 | ||
1251201c | 6421 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
6422 | |
6423 | out_unlock: | |
6424 | rcu_read_unlock(); | |
6425 | return retval; | |
6426 | } | |
6427 | ||
96f874e2 | 6428 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6429 | { |
5a16f3d3 | 6430 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6431 | struct task_struct *p; |
6432 | int retval; | |
1da177e4 | 6433 | |
23f5d142 | 6434 | rcu_read_lock(); |
1da177e4 LT |
6435 | |
6436 | p = find_process_by_pid(pid); | |
6437 | if (!p) { | |
23f5d142 | 6438 | rcu_read_unlock(); |
1da177e4 LT |
6439 | return -ESRCH; |
6440 | } | |
6441 | ||
23f5d142 | 6442 | /* Prevent p going away */ |
1da177e4 | 6443 | get_task_struct(p); |
23f5d142 | 6444 | rcu_read_unlock(); |
1da177e4 | 6445 | |
14a40ffc TH |
6446 | if (p->flags & PF_NO_SETAFFINITY) { |
6447 | retval = -EINVAL; | |
6448 | goto out_put_task; | |
6449 | } | |
5a16f3d3 RR |
6450 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6451 | retval = -ENOMEM; | |
6452 | goto out_put_task; | |
6453 | } | |
6454 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6455 | retval = -ENOMEM; | |
6456 | goto out_free_cpus_allowed; | |
6457 | } | |
1da177e4 | 6458 | retval = -EPERM; |
4c44aaaf EB |
6459 | if (!check_same_owner(p)) { |
6460 | rcu_read_lock(); | |
6461 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
6462 | rcu_read_unlock(); | |
16303ab2 | 6463 | goto out_free_new_mask; |
4c44aaaf EB |
6464 | } |
6465 | rcu_read_unlock(); | |
6466 | } | |
1da177e4 | 6467 | |
b0ae1981 | 6468 | retval = security_task_setscheduler(p); |
e7834f8f | 6469 | if (retval) |
16303ab2 | 6470 | goto out_free_new_mask; |
e7834f8f | 6471 | |
e4099a5e PZ |
6472 | |
6473 | cpuset_cpus_allowed(p, cpus_allowed); | |
6474 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
6475 | ||
332ac17e DF |
6476 | /* |
6477 | * Since bandwidth control happens on root_domain basis, | |
6478 | * if admission test is enabled, we only admit -deadline | |
6479 | * tasks allowed to run on all the CPUs in the task's | |
6480 | * root_domain. | |
6481 | */ | |
6482 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
6483 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
6484 | rcu_read_lock(); | |
6485 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 6486 | retval = -EBUSY; |
f1e3a093 | 6487 | rcu_read_unlock(); |
16303ab2 | 6488 | goto out_free_new_mask; |
332ac17e | 6489 | } |
f1e3a093 | 6490 | rcu_read_unlock(); |
332ac17e DF |
6491 | } |
6492 | #endif | |
49246274 | 6493 | again: |
9cfc3e18 | 6494 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK); |
1da177e4 | 6495 | |
8707d8b8 | 6496 | if (!retval) { |
5a16f3d3 RR |
6497 | cpuset_cpus_allowed(p, cpus_allowed); |
6498 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6499 | /* |
6500 | * We must have raced with a concurrent cpuset | |
6501 | * update. Just reset the cpus_allowed to the | |
6502 | * cpuset's cpus_allowed | |
6503 | */ | |
5a16f3d3 | 6504 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6505 | goto again; |
6506 | } | |
6507 | } | |
16303ab2 | 6508 | out_free_new_mask: |
5a16f3d3 RR |
6509 | free_cpumask_var(new_mask); |
6510 | out_free_cpus_allowed: | |
6511 | free_cpumask_var(cpus_allowed); | |
6512 | out_put_task: | |
1da177e4 | 6513 | put_task_struct(p); |
1da177e4 LT |
6514 | return retval; |
6515 | } | |
6516 | ||
6517 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6518 | struct cpumask *new_mask) |
1da177e4 | 6519 | { |
96f874e2 RR |
6520 | if (len < cpumask_size()) |
6521 | cpumask_clear(new_mask); | |
6522 | else if (len > cpumask_size()) | |
6523 | len = cpumask_size(); | |
6524 | ||
1da177e4 LT |
6525 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6526 | } | |
6527 | ||
6528 | /** | |
d1ccc66d | 6529 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
6530 | * @pid: pid of the process |
6531 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6532 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
6533 | * |
6534 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6535 | */ |
5add95d4 HC |
6536 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6537 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6538 | { |
5a16f3d3 | 6539 | cpumask_var_t new_mask; |
1da177e4 LT |
6540 | int retval; |
6541 | ||
5a16f3d3 RR |
6542 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6543 | return -ENOMEM; | |
1da177e4 | 6544 | |
5a16f3d3 RR |
6545 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6546 | if (retval == 0) | |
6547 | retval = sched_setaffinity(pid, new_mask); | |
6548 | free_cpumask_var(new_mask); | |
6549 | return retval; | |
1da177e4 LT |
6550 | } |
6551 | ||
96f874e2 | 6552 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6553 | { |
36c8b586 | 6554 | struct task_struct *p; |
31605683 | 6555 | unsigned long flags; |
1da177e4 | 6556 | int retval; |
1da177e4 | 6557 | |
23f5d142 | 6558 | rcu_read_lock(); |
1da177e4 LT |
6559 | |
6560 | retval = -ESRCH; | |
6561 | p = find_process_by_pid(pid); | |
6562 | if (!p) | |
6563 | goto out_unlock; | |
6564 | ||
e7834f8f DQ |
6565 | retval = security_task_getscheduler(p); |
6566 | if (retval) | |
6567 | goto out_unlock; | |
6568 | ||
013fdb80 | 6569 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 6570 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 6571 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
6572 | |
6573 | out_unlock: | |
23f5d142 | 6574 | rcu_read_unlock(); |
1da177e4 | 6575 | |
9531b62f | 6576 | return retval; |
1da177e4 LT |
6577 | } |
6578 | ||
6579 | /** | |
d1ccc66d | 6580 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
6581 | * @pid: pid of the process |
6582 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6583 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 6584 | * |
599b4840 ZW |
6585 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
6586 | * error code otherwise. | |
1da177e4 | 6587 | */ |
5add95d4 HC |
6588 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6589 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6590 | { |
6591 | int ret; | |
f17c8607 | 6592 | cpumask_var_t mask; |
1da177e4 | 6593 | |
84fba5ec | 6594 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
6595 | return -EINVAL; |
6596 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
6597 | return -EINVAL; |
6598 | ||
f17c8607 RR |
6599 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6600 | return -ENOMEM; | |
1da177e4 | 6601 | |
f17c8607 RR |
6602 | ret = sched_getaffinity(pid, mask); |
6603 | if (ret == 0) { | |
4de373a1 | 6604 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
6605 | |
6606 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
6607 | ret = -EFAULT; |
6608 | else | |
cd3d8031 | 6609 | ret = retlen; |
f17c8607 RR |
6610 | } |
6611 | free_cpumask_var(mask); | |
1da177e4 | 6612 | |
f17c8607 | 6613 | return ret; |
1da177e4 LT |
6614 | } |
6615 | ||
6616 | /** | |
6617 | * sys_sched_yield - yield the current processor to other threads. | |
6618 | * | |
dd41f596 IM |
6619 | * This function yields the current CPU to other tasks. If there are no |
6620 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
6621 | * |
6622 | * Return: 0. | |
1da177e4 | 6623 | */ |
7d4dd4f1 | 6624 | static void do_sched_yield(void) |
1da177e4 | 6625 | { |
8a8c69c3 PZ |
6626 | struct rq_flags rf; |
6627 | struct rq *rq; | |
6628 | ||
246b3b33 | 6629 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 6630 | |
ae92882e | 6631 | schedstat_inc(rq->yld_count); |
4530d7ab | 6632 | current->sched_class->yield_task(rq); |
1da177e4 | 6633 | |
8a8c69c3 | 6634 | preempt_disable(); |
345a957f | 6635 | rq_unlock_irq(rq, &rf); |
ba74c144 | 6636 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
6637 | |
6638 | schedule(); | |
7d4dd4f1 | 6639 | } |
1da177e4 | 6640 | |
7d4dd4f1 DB |
6641 | SYSCALL_DEFINE0(sched_yield) |
6642 | { | |
6643 | do_sched_yield(); | |
1da177e4 LT |
6644 | return 0; |
6645 | } | |
6646 | ||
c1a280b6 | 6647 | #ifndef CONFIG_PREEMPTION |
02b67cc3 | 6648 | int __sched _cond_resched(void) |
1da177e4 | 6649 | { |
fe32d3cd | 6650 | if (should_resched(0)) { |
a18b5d01 | 6651 | preempt_schedule_common(); |
1da177e4 LT |
6652 | return 1; |
6653 | } | |
f79c3ad6 | 6654 | rcu_all_qs(); |
1da177e4 LT |
6655 | return 0; |
6656 | } | |
02b67cc3 | 6657 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 6658 | #endif |
1da177e4 LT |
6659 | |
6660 | /* | |
613afbf8 | 6661 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6662 | * call schedule, and on return reacquire the lock. |
6663 | * | |
c1a280b6 | 6664 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
6665 | * operations here to prevent schedule() from being called twice (once via |
6666 | * spin_unlock(), once by hand). | |
6667 | */ | |
613afbf8 | 6668 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6669 | { |
fe32d3cd | 6670 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
6671 | int ret = 0; |
6672 | ||
f607c668 PZ |
6673 | lockdep_assert_held(lock); |
6674 | ||
4a81e832 | 6675 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6676 | spin_unlock(lock); |
d86ee480 | 6677 | if (resched) |
a18b5d01 | 6678 | preempt_schedule_common(); |
95c354fe NP |
6679 | else |
6680 | cpu_relax(); | |
6df3cecb | 6681 | ret = 1; |
1da177e4 | 6682 | spin_lock(lock); |
1da177e4 | 6683 | } |
6df3cecb | 6684 | return ret; |
1da177e4 | 6685 | } |
613afbf8 | 6686 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6687 | |
1da177e4 LT |
6688 | /** |
6689 | * yield - yield the current processor to other threads. | |
6690 | * | |
8e3fabfd PZ |
6691 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
6692 | * | |
6693 | * The scheduler is at all times free to pick the calling task as the most | |
6694 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 6695 | * it, it's already broken. |
8e3fabfd PZ |
6696 | * |
6697 | * Typical broken usage is: | |
6698 | * | |
6699 | * while (!event) | |
d1ccc66d | 6700 | * yield(); |
8e3fabfd PZ |
6701 | * |
6702 | * where one assumes that yield() will let 'the other' process run that will | |
6703 | * make event true. If the current task is a SCHED_FIFO task that will never | |
6704 | * happen. Never use yield() as a progress guarantee!! | |
6705 | * | |
6706 | * If you want to use yield() to wait for something, use wait_event(). | |
6707 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
6708 | * If you still want to use yield(), do not! | |
1da177e4 LT |
6709 | */ |
6710 | void __sched yield(void) | |
6711 | { | |
6712 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 6713 | do_sched_yield(); |
1da177e4 | 6714 | } |
1da177e4 LT |
6715 | EXPORT_SYMBOL(yield); |
6716 | ||
d95f4122 MG |
6717 | /** |
6718 | * yield_to - yield the current processor to another thread in | |
6719 | * your thread group, or accelerate that thread toward the | |
6720 | * processor it's on. | |
16addf95 RD |
6721 | * @p: target task |
6722 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
6723 | * |
6724 | * It's the caller's job to ensure that the target task struct | |
6725 | * can't go away on us before we can do any checks. | |
6726 | * | |
e69f6186 | 6727 | * Return: |
7b270f60 PZ |
6728 | * true (>0) if we indeed boosted the target task. |
6729 | * false (0) if we failed to boost the target. | |
6730 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 6731 | */ |
fa93384f | 6732 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
6733 | { |
6734 | struct task_struct *curr = current; | |
6735 | struct rq *rq, *p_rq; | |
6736 | unsigned long flags; | |
c3c18640 | 6737 | int yielded = 0; |
d95f4122 MG |
6738 | |
6739 | local_irq_save(flags); | |
6740 | rq = this_rq(); | |
6741 | ||
6742 | again: | |
6743 | p_rq = task_rq(p); | |
7b270f60 PZ |
6744 | /* |
6745 | * If we're the only runnable task on the rq and target rq also | |
6746 | * has only one task, there's absolutely no point in yielding. | |
6747 | */ | |
6748 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
6749 | yielded = -ESRCH; | |
6750 | goto out_irq; | |
6751 | } | |
6752 | ||
d95f4122 | 6753 | double_rq_lock(rq, p_rq); |
39e24d8f | 6754 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
6755 | double_rq_unlock(rq, p_rq); |
6756 | goto again; | |
6757 | } | |
6758 | ||
6759 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 6760 | goto out_unlock; |
d95f4122 MG |
6761 | |
6762 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 6763 | goto out_unlock; |
d95f4122 MG |
6764 | |
6765 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 6766 | goto out_unlock; |
d95f4122 | 6767 | |
0900acf2 | 6768 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 6769 | if (yielded) { |
ae92882e | 6770 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
6771 | /* |
6772 | * Make p's CPU reschedule; pick_next_entity takes care of | |
6773 | * fairness. | |
6774 | */ | |
6775 | if (preempt && rq != p_rq) | |
8875125e | 6776 | resched_curr(p_rq); |
6d1cafd8 | 6777 | } |
d95f4122 | 6778 | |
7b270f60 | 6779 | out_unlock: |
d95f4122 | 6780 | double_rq_unlock(rq, p_rq); |
7b270f60 | 6781 | out_irq: |
d95f4122 MG |
6782 | local_irq_restore(flags); |
6783 | ||
7b270f60 | 6784 | if (yielded > 0) |
d95f4122 MG |
6785 | schedule(); |
6786 | ||
6787 | return yielded; | |
6788 | } | |
6789 | EXPORT_SYMBOL_GPL(yield_to); | |
6790 | ||
10ab5643 TH |
6791 | int io_schedule_prepare(void) |
6792 | { | |
6793 | int old_iowait = current->in_iowait; | |
6794 | ||
6795 | current->in_iowait = 1; | |
6796 | blk_schedule_flush_plug(current); | |
6797 | ||
6798 | return old_iowait; | |
6799 | } | |
6800 | ||
6801 | void io_schedule_finish(int token) | |
6802 | { | |
6803 | current->in_iowait = token; | |
6804 | } | |
6805 | ||
1da177e4 | 6806 | /* |
41a2d6cf | 6807 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6808 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 6809 | */ |
1da177e4 LT |
6810 | long __sched io_schedule_timeout(long timeout) |
6811 | { | |
10ab5643 | 6812 | int token; |
1da177e4 LT |
6813 | long ret; |
6814 | ||
10ab5643 | 6815 | token = io_schedule_prepare(); |
1da177e4 | 6816 | ret = schedule_timeout(timeout); |
10ab5643 | 6817 | io_schedule_finish(token); |
9cff8ade | 6818 | |
1da177e4 LT |
6819 | return ret; |
6820 | } | |
9cff8ade | 6821 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 6822 | |
e3b929b0 | 6823 | void __sched io_schedule(void) |
10ab5643 TH |
6824 | { |
6825 | int token; | |
6826 | ||
6827 | token = io_schedule_prepare(); | |
6828 | schedule(); | |
6829 | io_schedule_finish(token); | |
6830 | } | |
6831 | EXPORT_SYMBOL(io_schedule); | |
6832 | ||
1da177e4 LT |
6833 | /** |
6834 | * sys_sched_get_priority_max - return maximum RT priority. | |
6835 | * @policy: scheduling class. | |
6836 | * | |
e69f6186 YB |
6837 | * Return: On success, this syscall returns the maximum |
6838 | * rt_priority that can be used by a given scheduling class. | |
6839 | * On failure, a negative error code is returned. | |
1da177e4 | 6840 | */ |
5add95d4 | 6841 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6842 | { |
6843 | int ret = -EINVAL; | |
6844 | ||
6845 | switch (policy) { | |
6846 | case SCHED_FIFO: | |
6847 | case SCHED_RR: | |
6848 | ret = MAX_USER_RT_PRIO-1; | |
6849 | break; | |
aab03e05 | 6850 | case SCHED_DEADLINE: |
1da177e4 | 6851 | case SCHED_NORMAL: |
b0a9499c | 6852 | case SCHED_BATCH: |
dd41f596 | 6853 | case SCHED_IDLE: |
1da177e4 LT |
6854 | ret = 0; |
6855 | break; | |
6856 | } | |
6857 | return ret; | |
6858 | } | |
6859 | ||
6860 | /** | |
6861 | * sys_sched_get_priority_min - return minimum RT priority. | |
6862 | * @policy: scheduling class. | |
6863 | * | |
e69f6186 YB |
6864 | * Return: On success, this syscall returns the minimum |
6865 | * rt_priority that can be used by a given scheduling class. | |
6866 | * On failure, a negative error code is returned. | |
1da177e4 | 6867 | */ |
5add95d4 | 6868 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6869 | { |
6870 | int ret = -EINVAL; | |
6871 | ||
6872 | switch (policy) { | |
6873 | case SCHED_FIFO: | |
6874 | case SCHED_RR: | |
6875 | ret = 1; | |
6876 | break; | |
aab03e05 | 6877 | case SCHED_DEADLINE: |
1da177e4 | 6878 | case SCHED_NORMAL: |
b0a9499c | 6879 | case SCHED_BATCH: |
dd41f596 | 6880 | case SCHED_IDLE: |
1da177e4 LT |
6881 | ret = 0; |
6882 | } | |
6883 | return ret; | |
6884 | } | |
6885 | ||
abca5fc5 | 6886 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 6887 | { |
36c8b586 | 6888 | struct task_struct *p; |
a4ec24b4 | 6889 | unsigned int time_slice; |
eb580751 | 6890 | struct rq_flags rf; |
dba091b9 | 6891 | struct rq *rq; |
3a5c359a | 6892 | int retval; |
1da177e4 LT |
6893 | |
6894 | if (pid < 0) | |
3a5c359a | 6895 | return -EINVAL; |
1da177e4 LT |
6896 | |
6897 | retval = -ESRCH; | |
1a551ae7 | 6898 | rcu_read_lock(); |
1da177e4 LT |
6899 | p = find_process_by_pid(pid); |
6900 | if (!p) | |
6901 | goto out_unlock; | |
6902 | ||
6903 | retval = security_task_getscheduler(p); | |
6904 | if (retval) | |
6905 | goto out_unlock; | |
6906 | ||
eb580751 | 6907 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
6908 | time_slice = 0; |
6909 | if (p->sched_class->get_rr_interval) | |
6910 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 6911 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 6912 | |
1a551ae7 | 6913 | rcu_read_unlock(); |
abca5fc5 AV |
6914 | jiffies_to_timespec64(time_slice, t); |
6915 | return 0; | |
3a5c359a | 6916 | |
1da177e4 | 6917 | out_unlock: |
1a551ae7 | 6918 | rcu_read_unlock(); |
1da177e4 LT |
6919 | return retval; |
6920 | } | |
6921 | ||
2064a5ab RD |
6922 | /** |
6923 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6924 | * @pid: pid of the process. | |
6925 | * @interval: userspace pointer to the timeslice value. | |
6926 | * | |
6927 | * this syscall writes the default timeslice value of a given process | |
6928 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6929 | * | |
6930 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
6931 | * an error code. | |
6932 | */ | |
abca5fc5 | 6933 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 6934 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
6935 | { |
6936 | struct timespec64 t; | |
6937 | int retval = sched_rr_get_interval(pid, &t); | |
6938 | ||
6939 | if (retval == 0) | |
6940 | retval = put_timespec64(&t, interval); | |
6941 | ||
6942 | return retval; | |
6943 | } | |
6944 | ||
474b9c77 | 6945 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
6946 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
6947 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
6948 | { |
6949 | struct timespec64 t; | |
6950 | int retval = sched_rr_get_interval(pid, &t); | |
6951 | ||
6952 | if (retval == 0) | |
9afc5eee | 6953 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
6954 | return retval; |
6955 | } | |
6956 | #endif | |
6957 | ||
82a1fcb9 | 6958 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6959 | { |
1da177e4 | 6960 | unsigned long free = 0; |
4e79752c | 6961 | int ppid; |
c930b2c0 | 6962 | |
38200502 TH |
6963 | if (!try_get_task_stack(p)) |
6964 | return; | |
20435d84 | 6965 | |
cc172ff3 | 6966 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 XX |
6967 | |
6968 | if (p->state == TASK_RUNNING) | |
cc172ff3 | 6969 | pr_cont(" running task "); |
1da177e4 | 6970 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 6971 | free = stack_not_used(p); |
1da177e4 | 6972 | #endif |
a90e984c | 6973 | ppid = 0; |
4e79752c | 6974 | rcu_read_lock(); |
a90e984c ON |
6975 | if (pid_alive(p)) |
6976 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 6977 | rcu_read_unlock(); |
cc172ff3 LZ |
6978 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
6979 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 6980 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 6981 | |
3d1cb205 | 6982 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 6983 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 6984 | show_stack(p, NULL, KERN_INFO); |
38200502 | 6985 | put_task_stack(p); |
1da177e4 | 6986 | } |
0032f4e8 | 6987 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 6988 | |
5d68cc95 PZ |
6989 | static inline bool |
6990 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
6991 | { | |
6992 | /* no filter, everything matches */ | |
6993 | if (!state_filter) | |
6994 | return true; | |
6995 | ||
6996 | /* filter, but doesn't match */ | |
6997 | if (!(p->state & state_filter)) | |
6998 | return false; | |
6999 | ||
7000 | /* | |
7001 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
7002 | * TASK_KILLABLE). | |
7003 | */ | |
7004 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
7005 | return false; | |
7006 | ||
7007 | return true; | |
7008 | } | |
7009 | ||
7010 | ||
e59e2ae2 | 7011 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 7012 | { |
36c8b586 | 7013 | struct task_struct *g, *p; |
1da177e4 | 7014 | |
510f5acc | 7015 | rcu_read_lock(); |
5d07f420 | 7016 | for_each_process_thread(g, p) { |
1da177e4 LT |
7017 | /* |
7018 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 7019 | * console might take a lot of time: |
57675cb9 AR |
7020 | * Also, reset softlockup watchdogs on all CPUs, because |
7021 | * another CPU might be blocked waiting for us to process | |
7022 | * an IPI. | |
1da177e4 LT |
7023 | */ |
7024 | touch_nmi_watchdog(); | |
57675cb9 | 7025 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 7026 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 7027 | sched_show_task(p); |
5d07f420 | 7028 | } |
1da177e4 | 7029 | |
dd41f596 | 7030 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
7031 | if (!state_filter) |
7032 | sysrq_sched_debug_show(); | |
dd41f596 | 7033 | #endif |
510f5acc | 7034 | rcu_read_unlock(); |
e59e2ae2 IM |
7035 | /* |
7036 | * Only show locks if all tasks are dumped: | |
7037 | */ | |
93335a21 | 7038 | if (!state_filter) |
e59e2ae2 | 7039 | debug_show_all_locks(); |
1da177e4 LT |
7040 | } |
7041 | ||
f340c0d1 IM |
7042 | /** |
7043 | * init_idle - set up an idle thread for a given CPU | |
7044 | * @idle: task in question | |
d1ccc66d | 7045 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
7046 | * |
7047 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
7048 | * flag, to make booting more robust. | |
7049 | */ | |
0db0628d | 7050 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 7051 | { |
70b97a7f | 7052 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
7053 | unsigned long flags; |
7054 | ||
ff51ff84 PZ |
7055 | __sched_fork(0, idle); |
7056 | ||
25834c73 PZ |
7057 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
7058 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 7059 | |
06b83b5f | 7060 | idle->state = TASK_RUNNING; |
dd41f596 | 7061 | idle->se.exec_start = sched_clock(); |
c1de45ca | 7062 | idle->flags |= PF_IDLE; |
dd41f596 | 7063 | |
d08b9f0c | 7064 | scs_task_reset(idle); |
e1b77c92 MR |
7065 | kasan_unpoison_task_stack(idle); |
7066 | ||
de9b8f5d PZ |
7067 | #ifdef CONFIG_SMP |
7068 | /* | |
b19a888c | 7069 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
7070 | * in that case do_set_cpus_allowed() will not do the right thing. |
7071 | * | |
7072 | * And since this is boot we can forgo the serialization. | |
7073 | */ | |
9cfc3e18 | 7074 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 7075 | #endif |
6506cf6c PZ |
7076 | /* |
7077 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 7078 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
7079 | * lockdep check in task_group() will fail. |
7080 | * | |
7081 | * Similar case to sched_fork(). / Alternatively we could | |
7082 | * use task_rq_lock() here and obtain the other rq->lock. | |
7083 | * | |
7084 | * Silence PROVE_RCU | |
7085 | */ | |
7086 | rcu_read_lock(); | |
dd41f596 | 7087 | __set_task_cpu(idle, cpu); |
6506cf6c | 7088 | rcu_read_unlock(); |
1da177e4 | 7089 | |
5311a98f EB |
7090 | rq->idle = idle; |
7091 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 7092 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 7093 | #ifdef CONFIG_SMP |
3ca7a440 | 7094 | idle->on_cpu = 1; |
4866cde0 | 7095 | #endif |
25834c73 PZ |
7096 | raw_spin_unlock(&rq->lock); |
7097 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
7098 | |
7099 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 7100 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 7101 | |
dd41f596 IM |
7102 | /* |
7103 | * The idle tasks have their own, simple scheduling class: | |
7104 | */ | |
7105 | idle->sched_class = &idle_sched_class; | |
868baf07 | 7106 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 7107 | vtime_init_idle(idle, cpu); |
de9b8f5d | 7108 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
7109 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
7110 | #endif | |
19978ca6 IM |
7111 | } |
7112 | ||
e1d4eeec NP |
7113 | #ifdef CONFIG_SMP |
7114 | ||
f82f8042 JL |
7115 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
7116 | const struct cpumask *trial) | |
7117 | { | |
06a76fe0 | 7118 | int ret = 1; |
f82f8042 | 7119 | |
bb2bc55a MG |
7120 | if (!cpumask_weight(cur)) |
7121 | return ret; | |
7122 | ||
06a76fe0 | 7123 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
7124 | |
7125 | return ret; | |
7126 | } | |
7127 | ||
7f51412a JL |
7128 | int task_can_attach(struct task_struct *p, |
7129 | const struct cpumask *cs_cpus_allowed) | |
7130 | { | |
7131 | int ret = 0; | |
7132 | ||
7133 | /* | |
7134 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 7135 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
7136 | * affinity and isolating such threads by their set of |
7137 | * allowed nodes is unnecessary. Thus, cpusets are not | |
7138 | * applicable for such threads. This prevents checking for | |
7139 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 7140 | * before cpus_mask may be changed. |
7f51412a JL |
7141 | */ |
7142 | if (p->flags & PF_NO_SETAFFINITY) { | |
7143 | ret = -EINVAL; | |
7144 | goto out; | |
7145 | } | |
7146 | ||
7f51412a | 7147 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
7148 | cs_cpus_allowed)) |
7149 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 7150 | |
7f51412a JL |
7151 | out: |
7152 | return ret; | |
7153 | } | |
7154 | ||
f2cb1360 | 7155 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 7156 | |
e6628d5b MG |
7157 | #ifdef CONFIG_NUMA_BALANCING |
7158 | /* Migrate current task p to target_cpu */ | |
7159 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
7160 | { | |
7161 | struct migration_arg arg = { p, target_cpu }; | |
7162 | int curr_cpu = task_cpu(p); | |
7163 | ||
7164 | if (curr_cpu == target_cpu) | |
7165 | return 0; | |
7166 | ||
3bd37062 | 7167 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
7168 | return -EINVAL; |
7169 | ||
7170 | /* TODO: This is not properly updating schedstats */ | |
7171 | ||
286549dc | 7172 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
7173 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
7174 | } | |
0ec8aa00 PZ |
7175 | |
7176 | /* | |
7177 | * Requeue a task on a given node and accurately track the number of NUMA | |
7178 | * tasks on the runqueues | |
7179 | */ | |
7180 | void sched_setnuma(struct task_struct *p, int nid) | |
7181 | { | |
da0c1e65 | 7182 | bool queued, running; |
eb580751 PZ |
7183 | struct rq_flags rf; |
7184 | struct rq *rq; | |
0ec8aa00 | 7185 | |
eb580751 | 7186 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 7187 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
7188 | running = task_current(rq, p); |
7189 | ||
da0c1e65 | 7190 | if (queued) |
1de64443 | 7191 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 7192 | if (running) |
f3cd1c4e | 7193 | put_prev_task(rq, p); |
0ec8aa00 PZ |
7194 | |
7195 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 7196 | |
da0c1e65 | 7197 | if (queued) |
7134b3e9 | 7198 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 7199 | if (running) |
03b7fad1 | 7200 | set_next_task(rq, p); |
eb580751 | 7201 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 7202 | } |
5cc389bc | 7203 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 7204 | |
1da177e4 | 7205 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 7206 | /* |
d1ccc66d | 7207 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 7208 | * offline. |
054b9108 | 7209 | */ |
48c5ccae | 7210 | void idle_task_exit(void) |
1da177e4 | 7211 | { |
48c5ccae | 7212 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 7213 | |
48c5ccae | 7214 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 7215 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 7216 | |
a53efe5f | 7217 | if (mm != &init_mm) { |
252d2a41 | 7218 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
7219 | finish_arch_post_lock_switch(); |
7220 | } | |
bf2c59fc PZ |
7221 | |
7222 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
7223 | } |
7224 | ||
2558aacf | 7225 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 7226 | { |
2558aacf PZ |
7227 | struct task_struct *p = arg; |
7228 | struct rq *rq = this_rq(); | |
7229 | struct rq_flags rf; | |
7230 | int cpu; | |
1da177e4 | 7231 | |
2558aacf PZ |
7232 | raw_spin_lock_irq(&p->pi_lock); |
7233 | rq_lock(rq, &rf); | |
3f1d2a31 | 7234 | |
2558aacf PZ |
7235 | update_rq_clock(rq); |
7236 | ||
7237 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
7238 | cpu = select_fallback_rq(rq->cpu, p); | |
7239 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 7240 | } |
3f1d2a31 | 7241 | |
2558aacf PZ |
7242 | rq_unlock(rq, &rf); |
7243 | raw_spin_unlock_irq(&p->pi_lock); | |
7244 | ||
7245 | put_task_struct(p); | |
7246 | ||
7247 | return 0; | |
10e7071b | 7248 | } |
3f1d2a31 | 7249 | |
2558aacf PZ |
7250 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
7251 | ||
48f24c4d | 7252 | /* |
2558aacf | 7253 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
1da177e4 | 7254 | */ |
2558aacf | 7255 | static void balance_push(struct rq *rq) |
1da177e4 | 7256 | { |
2558aacf PZ |
7257 | struct task_struct *push_task = rq->curr; |
7258 | ||
7259 | lockdep_assert_held(&rq->lock); | |
7260 | SCHED_WARN_ON(rq->cpu != smp_processor_id()); | |
1da177e4 LT |
7261 | |
7262 | /* | |
2558aacf PZ |
7263 | * Both the cpu-hotplug and stop task are in this case and are |
7264 | * required to complete the hotplug process. | |
1da177e4 | 7265 | */ |
3015ef4b | 7266 | if (is_per_cpu_kthread(push_task) || is_migration_disabled(push_task)) { |
f2469a1f TG |
7267 | /* |
7268 | * If this is the idle task on the outgoing CPU try to wake | |
7269 | * up the hotplug control thread which might wait for the | |
7270 | * last task to vanish. The rcuwait_active() check is | |
7271 | * accurate here because the waiter is pinned on this CPU | |
7272 | * and can't obviously be running in parallel. | |
3015ef4b TG |
7273 | * |
7274 | * On RT kernels this also has to check whether there are | |
7275 | * pinned and scheduled out tasks on the runqueue. They | |
7276 | * need to leave the migrate disabled section first. | |
f2469a1f | 7277 | */ |
3015ef4b TG |
7278 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
7279 | rcuwait_active(&rq->hotplug_wait)) { | |
f2469a1f TG |
7280 | raw_spin_unlock(&rq->lock); |
7281 | rcuwait_wake_up(&rq->hotplug_wait); | |
7282 | raw_spin_lock(&rq->lock); | |
7283 | } | |
2558aacf | 7284 | return; |
f2469a1f | 7285 | } |
48f24c4d | 7286 | |
2558aacf | 7287 | get_task_struct(push_task); |
77bd3970 | 7288 | /* |
2558aacf PZ |
7289 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
7290 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 7291 | */ |
2558aacf PZ |
7292 | raw_spin_unlock(&rq->lock); |
7293 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, | |
7294 | this_cpu_ptr(&push_work)); | |
7295 | /* | |
7296 | * At this point need_resched() is true and we'll take the loop in | |
7297 | * schedule(). The next pick is obviously going to be the stop task | |
7298 | * which is_per_cpu_kthread() and will push this task away. | |
7299 | */ | |
7300 | raw_spin_lock(&rq->lock); | |
7301 | } | |
77bd3970 | 7302 | |
2558aacf PZ |
7303 | static void balance_push_set(int cpu, bool on) |
7304 | { | |
7305 | struct rq *rq = cpu_rq(cpu); | |
7306 | struct rq_flags rf; | |
48c5ccae | 7307 | |
2558aacf PZ |
7308 | rq_lock_irqsave(rq, &rf); |
7309 | if (on) | |
7310 | rq->balance_flags |= BALANCE_PUSH; | |
7311 | else | |
7312 | rq->balance_flags &= ~BALANCE_PUSH; | |
7313 | rq_unlock_irqrestore(rq, &rf); | |
7314 | } | |
e692ab53 | 7315 | |
f2469a1f TG |
7316 | /* |
7317 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
7318 | * inactive. All tasks which are not per CPU kernel threads are either | |
7319 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
7320 | * during wakeup. Wait until the CPU is quiescent. | |
7321 | */ | |
7322 | static void balance_hotplug_wait(void) | |
7323 | { | |
7324 | struct rq *rq = this_rq(); | |
5473e0cc | 7325 | |
3015ef4b TG |
7326 | rcuwait_wait_event(&rq->hotplug_wait, |
7327 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
7328 | TASK_UNINTERRUPTIBLE); |
7329 | } | |
5473e0cc | 7330 | |
2558aacf | 7331 | #else |
dce48a84 | 7332 | |
2558aacf PZ |
7333 | static inline void balance_push(struct rq *rq) |
7334 | { | |
dce48a84 | 7335 | } |
2558aacf PZ |
7336 | |
7337 | static inline void balance_push_set(int cpu, bool on) | |
7338 | { | |
7339 | } | |
7340 | ||
f2469a1f TG |
7341 | static inline void balance_hotplug_wait(void) |
7342 | { | |
7343 | } | |
7344 | ||
1da177e4 LT |
7345 | #endif /* CONFIG_HOTPLUG_CPU */ |
7346 | ||
f2cb1360 | 7347 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
7348 | { |
7349 | if (!rq->online) { | |
7350 | const struct sched_class *class; | |
7351 | ||
c6c4927b | 7352 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7353 | rq->online = 1; |
7354 | ||
7355 | for_each_class(class) { | |
7356 | if (class->rq_online) | |
7357 | class->rq_online(rq); | |
7358 | } | |
7359 | } | |
7360 | } | |
7361 | ||
f2cb1360 | 7362 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
7363 | { |
7364 | if (rq->online) { | |
7365 | const struct sched_class *class; | |
7366 | ||
7367 | for_each_class(class) { | |
7368 | if (class->rq_offline) | |
7369 | class->rq_offline(rq); | |
7370 | } | |
7371 | ||
c6c4927b | 7372 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7373 | rq->online = 0; |
7374 | } | |
7375 | } | |
7376 | ||
d1ccc66d IM |
7377 | /* |
7378 | * used to mark begin/end of suspend/resume: | |
7379 | */ | |
7380 | static int num_cpus_frozen; | |
d35be8ba | 7381 | |
1da177e4 | 7382 | /* |
3a101d05 TH |
7383 | * Update cpusets according to cpu_active mask. If cpusets are |
7384 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7385 | * around partition_sched_domains(). | |
d35be8ba SB |
7386 | * |
7387 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7388 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7389 | */ |
40190a78 | 7390 | static void cpuset_cpu_active(void) |
e761b772 | 7391 | { |
40190a78 | 7392 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7393 | /* |
7394 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7395 | * resume sequence. As long as this is not the last online | |
7396 | * operation in the resume sequence, just build a single sched | |
7397 | * domain, ignoring cpusets. | |
7398 | */ | |
50e76632 PZ |
7399 | partition_sched_domains(1, NULL, NULL); |
7400 | if (--num_cpus_frozen) | |
135fb3e1 | 7401 | return; |
d35be8ba SB |
7402 | /* |
7403 | * This is the last CPU online operation. So fall through and | |
7404 | * restore the original sched domains by considering the | |
7405 | * cpuset configurations. | |
7406 | */ | |
50e76632 | 7407 | cpuset_force_rebuild(); |
3a101d05 | 7408 | } |
30e03acd | 7409 | cpuset_update_active_cpus(); |
3a101d05 | 7410 | } |
e761b772 | 7411 | |
40190a78 | 7412 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7413 | { |
40190a78 | 7414 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 7415 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 7416 | return -EBUSY; |
30e03acd | 7417 | cpuset_update_active_cpus(); |
135fb3e1 | 7418 | } else { |
d35be8ba SB |
7419 | num_cpus_frozen++; |
7420 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7421 | } |
135fb3e1 | 7422 | return 0; |
e761b772 | 7423 | } |
e761b772 | 7424 | |
40190a78 | 7425 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7426 | { |
7d976699 | 7427 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 7428 | struct rq_flags rf; |
7d976699 | 7429 | |
2558aacf PZ |
7430 | balance_push_set(cpu, false); |
7431 | ||
ba2591a5 PZ |
7432 | #ifdef CONFIG_SCHED_SMT |
7433 | /* | |
c5511d03 | 7434 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 7435 | */ |
c5511d03 PZI |
7436 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
7437 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 7438 | #endif |
40190a78 | 7439 | set_cpu_active(cpu, true); |
135fb3e1 | 7440 | |
40190a78 | 7441 | if (sched_smp_initialized) { |
135fb3e1 | 7442 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7443 | cpuset_cpu_active(); |
e761b772 | 7444 | } |
7d976699 TG |
7445 | |
7446 | /* | |
7447 | * Put the rq online, if not already. This happens: | |
7448 | * | |
7449 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 7450 | * after all CPUs have been brought up. |
7d976699 TG |
7451 | * |
7452 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7453 | * domains. | |
7454 | */ | |
8a8c69c3 | 7455 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
7456 | if (rq->rd) { |
7457 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7458 | set_rq_online(rq); | |
7459 | } | |
8a8c69c3 | 7460 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 7461 | |
40190a78 | 7462 | return 0; |
135fb3e1 TG |
7463 | } |
7464 | ||
40190a78 | 7465 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7466 | { |
120455c5 PZ |
7467 | struct rq *rq = cpu_rq(cpu); |
7468 | struct rq_flags rf; | |
135fb3e1 TG |
7469 | int ret; |
7470 | ||
40190a78 | 7471 | set_cpu_active(cpu, false); |
b2454caa PZ |
7472 | /* |
7473 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
7474 | * users of this state to go away such that all new such users will | |
7475 | * observe it. | |
7476 | * | |
b2454caa PZ |
7477 | * Do sync before park smpboot threads to take care the rcu boost case. |
7478 | */ | |
309ba859 | 7479 | synchronize_rcu(); |
40190a78 | 7480 | |
2558aacf PZ |
7481 | balance_push_set(cpu, true); |
7482 | ||
120455c5 PZ |
7483 | rq_lock_irqsave(rq, &rf); |
7484 | if (rq->rd) { | |
7485 | update_rq_clock(rq); | |
7486 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7487 | set_rq_offline(rq); | |
7488 | } | |
7489 | rq_unlock_irqrestore(rq, &rf); | |
7490 | ||
c5511d03 PZI |
7491 | #ifdef CONFIG_SCHED_SMT |
7492 | /* | |
7493 | * When going down, decrement the number of cores with SMT present. | |
7494 | */ | |
7495 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
7496 | static_branch_dec_cpuslocked(&sched_smt_present); | |
7497 | #endif | |
7498 | ||
40190a78 TG |
7499 | if (!sched_smp_initialized) |
7500 | return 0; | |
7501 | ||
7502 | ret = cpuset_cpu_inactive(cpu); | |
7503 | if (ret) { | |
2558aacf | 7504 | balance_push_set(cpu, false); |
40190a78 TG |
7505 | set_cpu_active(cpu, true); |
7506 | return ret; | |
135fb3e1 | 7507 | } |
40190a78 TG |
7508 | sched_domains_numa_masks_clear(cpu); |
7509 | return 0; | |
135fb3e1 TG |
7510 | } |
7511 | ||
94baf7a5 TG |
7512 | static void sched_rq_cpu_starting(unsigned int cpu) |
7513 | { | |
7514 | struct rq *rq = cpu_rq(cpu); | |
7515 | ||
7516 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7517 | update_max_interval(); |
7518 | } | |
7519 | ||
135fb3e1 TG |
7520 | int sched_cpu_starting(unsigned int cpu) |
7521 | { | |
94baf7a5 | 7522 | sched_rq_cpu_starting(cpu); |
d84b3131 | 7523 | sched_tick_start(cpu); |
135fb3e1 | 7524 | return 0; |
e761b772 | 7525 | } |
e761b772 | 7526 | |
f2785ddb | 7527 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
7528 | |
7529 | /* | |
7530 | * Invoked immediately before the stopper thread is invoked to bring the | |
7531 | * CPU down completely. At this point all per CPU kthreads except the | |
7532 | * hotplug thread (current) and the stopper thread (inactive) have been | |
7533 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
7534 | * any of those which might be on the way out are gone. | |
7535 | * | |
7536 | * If after this point a bound task is being woken on this CPU then the | |
7537 | * responsible hotplug callback has failed to do it's job. | |
7538 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
7539 | */ | |
7540 | int sched_cpu_wait_empty(unsigned int cpu) | |
7541 | { | |
7542 | balance_hotplug_wait(); | |
7543 | return 0; | |
7544 | } | |
7545 | ||
7546 | /* | |
7547 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
7548 | * might have. Called from the CPU stopper task after ensuring that the | |
7549 | * stopper is the last running task on the CPU, so nr_active count is | |
7550 | * stable. We need to take the teardown thread which is calling this into | |
7551 | * account, so we hand in adjust = 1 to the load calculation. | |
7552 | * | |
7553 | * Also see the comment "Global load-average calculations". | |
7554 | */ | |
7555 | static void calc_load_migrate(struct rq *rq) | |
7556 | { | |
7557 | long delta = calc_load_fold_active(rq, 1); | |
7558 | ||
7559 | if (delta) | |
7560 | atomic_long_add(delta, &calc_load_tasks); | |
7561 | } | |
7562 | ||
f2785ddb TG |
7563 | int sched_cpu_dying(unsigned int cpu) |
7564 | { | |
7565 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 7566 | struct rq_flags rf; |
f2785ddb TG |
7567 | |
7568 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 7569 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
7570 | |
7571 | rq_lock_irqsave(rq, &rf); | |
3015ef4b | 7572 | BUG_ON(rq->nr_running != 1 || rq_has_pinned_tasks(rq)); |
8a8c69c3 PZ |
7573 | rq_unlock_irqrestore(rq, &rf); |
7574 | ||
f2785ddb TG |
7575 | calc_load_migrate(rq); |
7576 | update_max_interval(); | |
00357f5e | 7577 | nohz_balance_exit_idle(rq); |
e5ef27d0 | 7578 | hrtick_clear(rq); |
f2785ddb TG |
7579 | return 0; |
7580 | } | |
7581 | #endif | |
7582 | ||
1da177e4 LT |
7583 | void __init sched_init_smp(void) |
7584 | { | |
cb83b629 PZ |
7585 | sched_init_numa(); |
7586 | ||
6acce3ef PZ |
7587 | /* |
7588 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 7589 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 7590 | * happen. |
6acce3ef | 7591 | */ |
712555ee | 7592 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 7593 | sched_init_domains(cpu_active_mask); |
712555ee | 7594 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7595 | |
5c1e1767 | 7596 | /* Move init over to a non-isolated CPU */ |
edb93821 | 7597 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 7598 | BUG(); |
19978ca6 | 7599 | sched_init_granularity(); |
4212823f | 7600 | |
0e3900e6 | 7601 | init_sched_rt_class(); |
1baca4ce | 7602 | init_sched_dl_class(); |
1b568f0a | 7603 | |
e26fbffd | 7604 | sched_smp_initialized = true; |
1da177e4 | 7605 | } |
e26fbffd TG |
7606 | |
7607 | static int __init migration_init(void) | |
7608 | { | |
77a5352b | 7609 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 7610 | return 0; |
1da177e4 | 7611 | } |
e26fbffd TG |
7612 | early_initcall(migration_init); |
7613 | ||
1da177e4 LT |
7614 | #else |
7615 | void __init sched_init_smp(void) | |
7616 | { | |
19978ca6 | 7617 | sched_init_granularity(); |
1da177e4 LT |
7618 | } |
7619 | #endif /* CONFIG_SMP */ | |
7620 | ||
7621 | int in_sched_functions(unsigned long addr) | |
7622 | { | |
1da177e4 LT |
7623 | return in_lock_functions(addr) || |
7624 | (addr >= (unsigned long)__sched_text_start | |
7625 | && addr < (unsigned long)__sched_text_end); | |
7626 | } | |
7627 | ||
029632fb | 7628 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7629 | /* |
7630 | * Default task group. | |
7631 | * Every task in system belongs to this group at bootup. | |
7632 | */ | |
029632fb | 7633 | struct task_group root_task_group; |
35cf4e50 | 7634 | LIST_HEAD(task_groups); |
b0367629 WL |
7635 | |
7636 | /* Cacheline aligned slab cache for task_group */ | |
7637 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7638 | #endif |
6f505b16 | 7639 | |
e6252c3e | 7640 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 7641 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 7642 | |
1da177e4 LT |
7643 | void __init sched_init(void) |
7644 | { | |
a1dc0446 | 7645 | unsigned long ptr = 0; |
55627e3c | 7646 | int i; |
434d53b0 | 7647 | |
c3a340f7 SRV |
7648 | /* Make sure the linker didn't screw up */ |
7649 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
7650 | &fair_sched_class + 1 != &rt_sched_class || | |
7651 | &rt_sched_class + 1 != &dl_sched_class); | |
7652 | #ifdef CONFIG_SMP | |
7653 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
7654 | #endif | |
7655 | ||
5822a454 | 7656 | wait_bit_init(); |
9dcb8b68 | 7657 | |
434d53b0 | 7658 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 7659 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
7660 | #endif |
7661 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 7662 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 7663 | #endif |
a1dc0446 QC |
7664 | if (ptr) { |
7665 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
7666 | |
7667 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7668 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7669 | ptr += nr_cpu_ids * sizeof(void **); |
7670 | ||
07e06b01 | 7671 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7672 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7673 | |
b1d1779e WY |
7674 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
7675 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 7676 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7677 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7678 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7679 | ptr += nr_cpu_ids * sizeof(void **); |
7680 | ||
07e06b01 | 7681 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7682 | ptr += nr_cpu_ids * sizeof(void **); |
7683 | ||
6d6bc0ad | 7684 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7685 | } |
df7c8e84 | 7686 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7687 | for_each_possible_cpu(i) { |
7688 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7689 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
7690 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
7691 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7692 | } |
b74e6278 | 7693 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7694 | |
d1ccc66d IM |
7695 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
7696 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 7697 | |
57d885fe GH |
7698 | #ifdef CONFIG_SMP |
7699 | init_defrootdomain(); | |
7700 | #endif | |
7701 | ||
d0b27fa7 | 7702 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7703 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7704 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7705 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7706 | |
7c941438 | 7707 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7708 | task_group_cache = KMEM_CACHE(task_group, 0); |
7709 | ||
07e06b01 YZ |
7710 | list_add(&root_task_group.list, &task_groups); |
7711 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7712 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7713 | autogroup_init(&init_task); |
7c941438 | 7714 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7715 | |
0a945022 | 7716 | for_each_possible_cpu(i) { |
70b97a7f | 7717 | struct rq *rq; |
1da177e4 LT |
7718 | |
7719 | rq = cpu_rq(i); | |
05fa785c | 7720 | raw_spin_lock_init(&rq->lock); |
7897986b | 7721 | rq->nr_running = 0; |
dce48a84 TG |
7722 | rq->calc_load_active = 0; |
7723 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7724 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7725 | init_rt_rq(&rq->rt); |
7726 | init_dl_rq(&rq->dl); | |
dd41f596 | 7727 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 7728 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 7729 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 7730 | /* |
d1ccc66d | 7731 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
7732 | * |
7733 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
7734 | * gets 100% of the CPU resources in the system. This overall |
7735 | * system CPU resource is divided among the tasks of | |
07e06b01 | 7736 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7737 | * based on each entity's (task or task-group's) weight |
7738 | * (se->load.weight). | |
7739 | * | |
07e06b01 | 7740 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 7741 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 7742 | * then A0's share of the CPU resource is: |
354d60c2 | 7743 | * |
0d905bca | 7744 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7745 | * |
07e06b01 YZ |
7746 | * We achieve this by letting root_task_group's tasks sit |
7747 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7748 | */ |
07e06b01 | 7749 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7750 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7751 | ||
7752 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7753 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7754 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7755 | #endif |
1da177e4 | 7756 | #ifdef CONFIG_SMP |
41c7ce9a | 7757 | rq->sd = NULL; |
57d885fe | 7758 | rq->rd = NULL; |
ca6d75e6 | 7759 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7760 | rq->balance_callback = NULL; |
1da177e4 | 7761 | rq->active_balance = 0; |
dd41f596 | 7762 | rq->next_balance = jiffies; |
1da177e4 | 7763 | rq->push_cpu = 0; |
0a2966b4 | 7764 | rq->cpu = i; |
1f11eb6a | 7765 | rq->online = 0; |
eae0c9df MG |
7766 | rq->idle_stamp = 0; |
7767 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7768 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7769 | |
7770 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7771 | ||
dc938520 | 7772 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7773 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 7774 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 7775 | atomic_set(&rq->nohz_flags, 0); |
90b5363a PZI |
7776 | |
7777 | rq_csd_init(rq, &rq->nohz_csd, nohz_csd_func); | |
83cd4fe2 | 7778 | #endif |
f2469a1f TG |
7779 | #ifdef CONFIG_HOTPLUG_CPU |
7780 | rcuwait_init(&rq->hotplug_wait); | |
7781 | #endif | |
9fd81dd5 | 7782 | #endif /* CONFIG_SMP */ |
77a021be | 7783 | hrtick_rq_init(rq); |
1da177e4 | 7784 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7785 | } |
7786 | ||
9059393e | 7787 | set_load_weight(&init_task, false); |
b50f60ce | 7788 | |
1da177e4 LT |
7789 | /* |
7790 | * The boot idle thread does lazy MMU switching as well: | |
7791 | */ | |
f1f10076 | 7792 | mmgrab(&init_mm); |
1da177e4 LT |
7793 | enter_lazy_tlb(&init_mm, current); |
7794 | ||
7795 | /* | |
7796 | * Make us the idle thread. Technically, schedule() should not be | |
7797 | * called from this thread, however somewhere below it might be, | |
7798 | * but because we are the idle thread, we just pick up running again | |
7799 | * when this runqueue becomes "idle". | |
7800 | */ | |
7801 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7802 | |
7803 | calc_load_update = jiffies + LOAD_FREQ; | |
7804 | ||
bf4d83f6 | 7805 | #ifdef CONFIG_SMP |
29d5e047 | 7806 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7807 | #endif |
7808 | init_sched_fair_class(); | |
6a7b3dc3 | 7809 | |
4698f88c JP |
7810 | init_schedstats(); |
7811 | ||
eb414681 JW |
7812 | psi_init(); |
7813 | ||
69842cba PB |
7814 | init_uclamp(); |
7815 | ||
6892b75e | 7816 | scheduler_running = 1; |
1da177e4 LT |
7817 | } |
7818 | ||
d902db1e | 7819 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7820 | static inline int preempt_count_equals(int preempt_offset) |
7821 | { | |
da7142e2 | 7822 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7823 | |
4ba8216c | 7824 | return (nested == preempt_offset); |
e4aafea2 FW |
7825 | } |
7826 | ||
d894837f | 7827 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7828 | { |
8eb23b9f PZ |
7829 | /* |
7830 | * Blocking primitives will set (and therefore destroy) current->state, | |
7831 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7832 | * otherwise we will destroy state. | |
7833 | */ | |
00845eb9 | 7834 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7835 | "do not call blocking ops when !TASK_RUNNING; " |
7836 | "state=%lx set at [<%p>] %pS\n", | |
7837 | current->state, | |
7838 | (void *)current->task_state_change, | |
00845eb9 | 7839 | (void *)current->task_state_change); |
8eb23b9f | 7840 | |
3427445a PZ |
7841 | ___might_sleep(file, line, preempt_offset); |
7842 | } | |
7843 | EXPORT_SYMBOL(__might_sleep); | |
7844 | ||
7845 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7846 | { |
d1ccc66d IM |
7847 | /* Ratelimiting timestamp: */ |
7848 | static unsigned long prev_jiffy; | |
7849 | ||
d1c6d149 | 7850 | unsigned long preempt_disable_ip; |
1da177e4 | 7851 | |
d1ccc66d IM |
7852 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
7853 | rcu_sleep_check(); | |
7854 | ||
db273be2 | 7855 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 7856 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
7857 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
7858 | oops_in_progress) | |
aef745fc | 7859 | return; |
1c3c5eab | 7860 | |
aef745fc IM |
7861 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
7862 | return; | |
7863 | prev_jiffy = jiffies; | |
7864 | ||
d1ccc66d | 7865 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
7866 | preempt_disable_ip = get_preempt_disable_ip(current); |
7867 | ||
3df0fc5b PZ |
7868 | printk(KERN_ERR |
7869 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7870 | file, line); | |
7871 | printk(KERN_ERR | |
312364f3 DV |
7872 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
7873 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 7874 | current->pid, current->comm); |
aef745fc | 7875 | |
a8b686b3 ES |
7876 | if (task_stack_end_corrupted(current)) |
7877 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7878 | ||
aef745fc IM |
7879 | debug_show_held_locks(current); |
7880 | if (irqs_disabled()) | |
7881 | print_irqtrace_events(current); | |
d1c6d149 VN |
7882 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
7883 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 7884 | pr_err("Preemption disabled at:"); |
2062a4e8 | 7885 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 7886 | } |
aef745fc | 7887 | dump_stack(); |
f0b22e39 | 7888 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 7889 | } |
3427445a | 7890 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
7891 | |
7892 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
7893 | { | |
7894 | static unsigned long prev_jiffy; | |
7895 | ||
7896 | if (irqs_disabled()) | |
7897 | return; | |
7898 | ||
7899 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
7900 | return; | |
7901 | ||
7902 | if (preempt_count() > preempt_offset) | |
7903 | return; | |
7904 | ||
7905 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7906 | return; | |
7907 | prev_jiffy = jiffies; | |
7908 | ||
7909 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
7910 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7911 | in_atomic(), irqs_disabled(), | |
7912 | current->pid, current->comm); | |
7913 | ||
7914 | debug_show_held_locks(current); | |
7915 | dump_stack(); | |
7916 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
7917 | } | |
7918 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
7919 | |
7920 | #ifdef CONFIG_SMP | |
7921 | void __cant_migrate(const char *file, int line) | |
7922 | { | |
7923 | static unsigned long prev_jiffy; | |
7924 | ||
7925 | if (irqs_disabled()) | |
7926 | return; | |
7927 | ||
7928 | if (is_migration_disabled(current)) | |
7929 | return; | |
7930 | ||
7931 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
7932 | return; | |
7933 | ||
7934 | if (preempt_count() > 0) | |
7935 | return; | |
7936 | ||
7937 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7938 | return; | |
7939 | prev_jiffy = jiffies; | |
7940 | ||
7941 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
7942 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
7943 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
7944 | current->pid, current->comm); | |
7945 | ||
7946 | debug_show_held_locks(current); | |
7947 | dump_stack(); | |
7948 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
7949 | } | |
7950 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
7951 | #endif | |
1da177e4 LT |
7952 | #endif |
7953 | ||
7954 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7955 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7956 | { |
dbc7f069 | 7957 | struct task_struct *g, *p; |
d50dde5a DF |
7958 | struct sched_attr attr = { |
7959 | .sched_policy = SCHED_NORMAL, | |
7960 | }; | |
1da177e4 | 7961 | |
3472eaa1 | 7962 | read_lock(&tasklist_lock); |
5d07f420 | 7963 | for_each_process_thread(g, p) { |
178be793 IM |
7964 | /* |
7965 | * Only normalize user tasks: | |
7966 | */ | |
3472eaa1 | 7967 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7968 | continue; |
7969 | ||
4fa8d299 JP |
7970 | p->se.exec_start = 0; |
7971 | schedstat_set(p->se.statistics.wait_start, 0); | |
7972 | schedstat_set(p->se.statistics.sleep_start, 0); | |
7973 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 7974 | |
aab03e05 | 7975 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7976 | /* |
7977 | * Renice negative nice level userspace | |
7978 | * tasks back to 0: | |
7979 | */ | |
3472eaa1 | 7980 | if (task_nice(p) < 0) |
dd41f596 | 7981 | set_user_nice(p, 0); |
1da177e4 | 7982 | continue; |
dd41f596 | 7983 | } |
1da177e4 | 7984 | |
dbc7f069 | 7985 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7986 | } |
3472eaa1 | 7987 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7988 | } |
7989 | ||
7990 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7991 | |
67fc4e0c | 7992 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7993 | /* |
67fc4e0c | 7994 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7995 | * |
7996 | * They can only be called when the whole system has been | |
7997 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7998 | * activity can take place. Using them for anything else would | |
7999 | * be a serious bug, and as a result, they aren't even visible | |
8000 | * under any other configuration. | |
8001 | */ | |
8002 | ||
8003 | /** | |
d1ccc66d | 8004 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
8005 | * @cpu: the processor in question. |
8006 | * | |
8007 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
8008 | * |
8009 | * Return: The current task for @cpu. | |
1df5c10a | 8010 | */ |
36c8b586 | 8011 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8012 | { |
8013 | return cpu_curr(cpu); | |
8014 | } | |
8015 | ||
67fc4e0c JW |
8016 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8017 | ||
8018 | #ifdef CONFIG_IA64 | |
1df5c10a | 8019 | /** |
5feeb783 | 8020 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
8021 | * @cpu: the processor in question. |
8022 | * @p: the task pointer to set. | |
8023 | * | |
8024 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 8025 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 8026 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
8027 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8028 | * and caller must save the original value of the current task (see | |
8029 | * curr_task() above) and restore that value before reenabling interrupts and | |
8030 | * re-starting the system. | |
8031 | * | |
8032 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8033 | */ | |
a458ae2e | 8034 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8035 | { |
8036 | cpu_curr(cpu) = p; | |
8037 | } | |
8038 | ||
8039 | #endif | |
29f59db3 | 8040 | |
7c941438 | 8041 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
8042 | /* task_group_lock serializes the addition/removal of task groups */ |
8043 | static DEFINE_SPINLOCK(task_group_lock); | |
8044 | ||
2480c093 PB |
8045 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
8046 | struct task_group *parent) | |
8047 | { | |
8048 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 8049 | enum uclamp_id clamp_id; |
2480c093 PB |
8050 | |
8051 | for_each_clamp_id(clamp_id) { | |
8052 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
8053 | uclamp_none(clamp_id), false); | |
0b60ba2d | 8054 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
8055 | } |
8056 | #endif | |
8057 | } | |
8058 | ||
2f5177f0 | 8059 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
8060 | { |
8061 | free_fair_sched_group(tg); | |
8062 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8063 | autogroup_free(tg); |
b0367629 | 8064 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
8065 | } |
8066 | ||
8067 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8068 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8069 | { |
8070 | struct task_group *tg; | |
bccbe08a | 8071 | |
b0367629 | 8072 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
8073 | if (!tg) |
8074 | return ERR_PTR(-ENOMEM); | |
8075 | ||
ec7dc8ac | 8076 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8077 | goto err; |
8078 | ||
ec7dc8ac | 8079 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8080 | goto err; |
8081 | ||
2480c093 PB |
8082 | alloc_uclamp_sched_group(tg, parent); |
8083 | ||
ace783b9 LZ |
8084 | return tg; |
8085 | ||
8086 | err: | |
2f5177f0 | 8087 | sched_free_group(tg); |
ace783b9 LZ |
8088 | return ERR_PTR(-ENOMEM); |
8089 | } | |
8090 | ||
8091 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
8092 | { | |
8093 | unsigned long flags; | |
8094 | ||
8ed36996 | 8095 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8096 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 8097 | |
d1ccc66d IM |
8098 | /* Root should already exist: */ |
8099 | WARN_ON(!parent); | |
f473aa5e PZ |
8100 | |
8101 | tg->parent = parent; | |
f473aa5e | 8102 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8103 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8104 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
8105 | |
8106 | online_fair_sched_group(tg); | |
29f59db3 SV |
8107 | } |
8108 | ||
9b5b7751 | 8109 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 8110 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8111 | { |
d1ccc66d | 8112 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 8113 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8114 | } |
8115 | ||
4cf86d77 | 8116 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 8117 | { |
d1ccc66d | 8118 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 8119 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
8120 | } |
8121 | ||
8122 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 8123 | { |
8ed36996 | 8124 | unsigned long flags; |
29f59db3 | 8125 | |
d1ccc66d | 8126 | /* End participation in shares distribution: */ |
6fe1f348 | 8127 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
8128 | |
8129 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8130 | list_del_rcu(&tg->list); |
f473aa5e | 8131 | list_del_rcu(&tg->siblings); |
8ed36996 | 8132 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
8133 | } |
8134 | ||
ea86cb4b | 8135 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 8136 | { |
8323f26c | 8137 | struct task_group *tg; |
29f59db3 | 8138 | |
f7b8a47d KT |
8139 | /* |
8140 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
8141 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
8142 | * to prevent lockdep warnings. | |
8143 | */ | |
8144 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
8145 | struct task_group, css); |
8146 | tg = autogroup_task_group(tsk, tg); | |
8147 | tsk->sched_task_group = tg; | |
8148 | ||
810b3817 | 8149 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
8150 | if (tsk->sched_class->task_change_group) |
8151 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 8152 | else |
810b3817 | 8153 | #endif |
b2b5ce02 | 8154 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
8155 | } |
8156 | ||
8157 | /* | |
8158 | * Change task's runqueue when it moves between groups. | |
8159 | * | |
8160 | * The caller of this function should have put the task in its new group by | |
8161 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
8162 | * its new group. | |
8163 | */ | |
8164 | void sched_move_task(struct task_struct *tsk) | |
8165 | { | |
7a57f32a PZ |
8166 | int queued, running, queue_flags = |
8167 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
8168 | struct rq_flags rf; |
8169 | struct rq *rq; | |
8170 | ||
8171 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 8172 | update_rq_clock(rq); |
ea86cb4b VG |
8173 | |
8174 | running = task_current(rq, tsk); | |
8175 | queued = task_on_rq_queued(tsk); | |
8176 | ||
8177 | if (queued) | |
7a57f32a | 8178 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 8179 | if (running) |
ea86cb4b VG |
8180 | put_prev_task(rq, tsk); |
8181 | ||
8182 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 8183 | |
da0c1e65 | 8184 | if (queued) |
7a57f32a | 8185 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 8186 | if (running) { |
03b7fad1 | 8187 | set_next_task(rq, tsk); |
2a4b03ff VG |
8188 | /* |
8189 | * After changing group, the running task may have joined a | |
8190 | * throttled one but it's still the running task. Trigger a | |
8191 | * resched to make sure that task can still run. | |
8192 | */ | |
8193 | resched_curr(rq); | |
8194 | } | |
29f59db3 | 8195 | |
eb580751 | 8196 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 8197 | } |
68318b8e | 8198 | |
a7c6d554 | 8199 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8200 | { |
a7c6d554 | 8201 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8202 | } |
8203 | ||
eb95419b TH |
8204 | static struct cgroup_subsys_state * |
8205 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8206 | { |
eb95419b TH |
8207 | struct task_group *parent = css_tg(parent_css); |
8208 | struct task_group *tg; | |
68318b8e | 8209 | |
eb95419b | 8210 | if (!parent) { |
68318b8e | 8211 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8212 | return &root_task_group.css; |
68318b8e SV |
8213 | } |
8214 | ||
ec7dc8ac | 8215 | tg = sched_create_group(parent); |
68318b8e SV |
8216 | if (IS_ERR(tg)) |
8217 | return ERR_PTR(-ENOMEM); | |
8218 | ||
68318b8e SV |
8219 | return &tg->css; |
8220 | } | |
8221 | ||
96b77745 KK |
8222 | /* Expose task group only after completing cgroup initialization */ |
8223 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
8224 | { | |
8225 | struct task_group *tg = css_tg(css); | |
8226 | struct task_group *parent = css_tg(css->parent); | |
8227 | ||
8228 | if (parent) | |
8229 | sched_online_group(tg, parent); | |
7226017a QY |
8230 | |
8231 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8232 | /* Propagate the effective uclamp value for the new group */ | |
8233 | cpu_util_update_eff(css); | |
8234 | #endif | |
8235 | ||
96b77745 KK |
8236 | return 0; |
8237 | } | |
8238 | ||
2f5177f0 | 8239 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8240 | { |
eb95419b | 8241 | struct task_group *tg = css_tg(css); |
ace783b9 | 8242 | |
2f5177f0 | 8243 | sched_offline_group(tg); |
ace783b9 LZ |
8244 | } |
8245 | ||
eb95419b | 8246 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8247 | { |
eb95419b | 8248 | struct task_group *tg = css_tg(css); |
68318b8e | 8249 | |
2f5177f0 PZ |
8250 | /* |
8251 | * Relies on the RCU grace period between css_released() and this. | |
8252 | */ | |
8253 | sched_free_group(tg); | |
ace783b9 LZ |
8254 | } |
8255 | ||
ea86cb4b VG |
8256 | /* |
8257 | * This is called before wake_up_new_task(), therefore we really only | |
8258 | * have to set its group bits, all the other stuff does not apply. | |
8259 | */ | |
b53202e6 | 8260 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 8261 | { |
ea86cb4b VG |
8262 | struct rq_flags rf; |
8263 | struct rq *rq; | |
8264 | ||
8265 | rq = task_rq_lock(task, &rf); | |
8266 | ||
80f5c1b8 | 8267 | update_rq_clock(rq); |
ea86cb4b VG |
8268 | sched_change_group(task, TASK_SET_GROUP); |
8269 | ||
8270 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
8271 | } |
8272 | ||
1f7dd3e5 | 8273 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8274 | { |
bb9d97b6 | 8275 | struct task_struct *task; |
1f7dd3e5 | 8276 | struct cgroup_subsys_state *css; |
7dc603c9 | 8277 | int ret = 0; |
bb9d97b6 | 8278 | |
1f7dd3e5 | 8279 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8280 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8281 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8282 | return -EINVAL; |
b68aa230 | 8283 | #endif |
7dc603c9 | 8284 | /* |
b19a888c | 8285 | * Serialize against wake_up_new_task() such that if it's |
7dc603c9 PZ |
8286 | * running, we're sure to observe its full state. |
8287 | */ | |
8288 | raw_spin_lock_irq(&task->pi_lock); | |
8289 | /* | |
8290 | * Avoid calling sched_move_task() before wake_up_new_task() | |
8291 | * has happened. This would lead to problems with PELT, due to | |
8292 | * move wanting to detach+attach while we're not attached yet. | |
8293 | */ | |
8294 | if (task->state == TASK_NEW) | |
8295 | ret = -EINVAL; | |
8296 | raw_spin_unlock_irq(&task->pi_lock); | |
8297 | ||
8298 | if (ret) | |
8299 | break; | |
bb9d97b6 | 8300 | } |
7dc603c9 | 8301 | return ret; |
be367d09 | 8302 | } |
68318b8e | 8303 | |
1f7dd3e5 | 8304 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8305 | { |
bb9d97b6 | 8306 | struct task_struct *task; |
1f7dd3e5 | 8307 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8308 | |
1f7dd3e5 | 8309 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8310 | sched_move_task(task); |
68318b8e SV |
8311 | } |
8312 | ||
2480c093 | 8313 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
8314 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
8315 | { | |
8316 | struct cgroup_subsys_state *top_css = css; | |
8317 | struct uclamp_se *uc_parent = NULL; | |
8318 | struct uclamp_se *uc_se = NULL; | |
8319 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 8320 | enum uclamp_id clamp_id; |
0b60ba2d PB |
8321 | unsigned int clamps; |
8322 | ||
8323 | css_for_each_descendant_pre(css, top_css) { | |
8324 | uc_parent = css_tg(css)->parent | |
8325 | ? css_tg(css)->parent->uclamp : NULL; | |
8326 | ||
8327 | for_each_clamp_id(clamp_id) { | |
8328 | /* Assume effective clamps matches requested clamps */ | |
8329 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
8330 | /* Cap effective clamps with parent's effective clamps */ | |
8331 | if (uc_parent && | |
8332 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
8333 | eff[clamp_id] = uc_parent[clamp_id].value; | |
8334 | } | |
8335 | } | |
8336 | /* Ensure protection is always capped by limit */ | |
8337 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
8338 | ||
8339 | /* Propagate most restrictive effective clamps */ | |
8340 | clamps = 0x0; | |
8341 | uc_se = css_tg(css)->uclamp; | |
8342 | for_each_clamp_id(clamp_id) { | |
8343 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
8344 | continue; | |
8345 | uc_se[clamp_id].value = eff[clamp_id]; | |
8346 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
8347 | clamps |= (0x1 << clamp_id); | |
8348 | } | |
babbe170 | 8349 | if (!clamps) { |
0b60ba2d | 8350 | css = css_rightmost_descendant(css); |
babbe170 PB |
8351 | continue; |
8352 | } | |
8353 | ||
8354 | /* Immediately update descendants RUNNABLE tasks */ | |
8355 | uclamp_update_active_tasks(css, clamps); | |
0b60ba2d PB |
8356 | } |
8357 | } | |
2480c093 PB |
8358 | |
8359 | /* | |
8360 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
8361 | * C expression. Since there is no way to convert a macro argument (N) into a | |
8362 | * character constant, use two levels of macros. | |
8363 | */ | |
8364 | #define _POW10(exp) ((unsigned int)1e##exp) | |
8365 | #define POW10(exp) _POW10(exp) | |
8366 | ||
8367 | struct uclamp_request { | |
8368 | #define UCLAMP_PERCENT_SHIFT 2 | |
8369 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
8370 | s64 percent; | |
8371 | u64 util; | |
8372 | int ret; | |
8373 | }; | |
8374 | ||
8375 | static inline struct uclamp_request | |
8376 | capacity_from_percent(char *buf) | |
8377 | { | |
8378 | struct uclamp_request req = { | |
8379 | .percent = UCLAMP_PERCENT_SCALE, | |
8380 | .util = SCHED_CAPACITY_SCALE, | |
8381 | .ret = 0, | |
8382 | }; | |
8383 | ||
8384 | buf = strim(buf); | |
8385 | if (strcmp(buf, "max")) { | |
8386 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
8387 | &req.percent); | |
8388 | if (req.ret) | |
8389 | return req; | |
b562d140 | 8390 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
8391 | req.ret = -ERANGE; |
8392 | return req; | |
8393 | } | |
8394 | ||
8395 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
8396 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
8397 | } | |
8398 | ||
8399 | return req; | |
8400 | } | |
8401 | ||
8402 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
8403 | size_t nbytes, loff_t off, | |
8404 | enum uclamp_id clamp_id) | |
8405 | { | |
8406 | struct uclamp_request req; | |
8407 | struct task_group *tg; | |
8408 | ||
8409 | req = capacity_from_percent(buf); | |
8410 | if (req.ret) | |
8411 | return req.ret; | |
8412 | ||
46609ce2 QY |
8413 | static_branch_enable(&sched_uclamp_used); |
8414 | ||
2480c093 PB |
8415 | mutex_lock(&uclamp_mutex); |
8416 | rcu_read_lock(); | |
8417 | ||
8418 | tg = css_tg(of_css(of)); | |
8419 | if (tg->uclamp_req[clamp_id].value != req.util) | |
8420 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
8421 | ||
8422 | /* | |
8423 | * Because of not recoverable conversion rounding we keep track of the | |
8424 | * exact requested value | |
8425 | */ | |
8426 | tg->uclamp_pct[clamp_id] = req.percent; | |
8427 | ||
0b60ba2d PB |
8428 | /* Update effective clamps to track the most restrictive value */ |
8429 | cpu_util_update_eff(of_css(of)); | |
8430 | ||
2480c093 PB |
8431 | rcu_read_unlock(); |
8432 | mutex_unlock(&uclamp_mutex); | |
8433 | ||
8434 | return nbytes; | |
8435 | } | |
8436 | ||
8437 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
8438 | char *buf, size_t nbytes, | |
8439 | loff_t off) | |
8440 | { | |
8441 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
8442 | } | |
8443 | ||
8444 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
8445 | char *buf, size_t nbytes, | |
8446 | loff_t off) | |
8447 | { | |
8448 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
8449 | } | |
8450 | ||
8451 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
8452 | enum uclamp_id clamp_id) | |
8453 | { | |
8454 | struct task_group *tg; | |
8455 | u64 util_clamp; | |
8456 | u64 percent; | |
8457 | u32 rem; | |
8458 | ||
8459 | rcu_read_lock(); | |
8460 | tg = css_tg(seq_css(sf)); | |
8461 | util_clamp = tg->uclamp_req[clamp_id].value; | |
8462 | rcu_read_unlock(); | |
8463 | ||
8464 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
8465 | seq_puts(sf, "max\n"); | |
8466 | return; | |
8467 | } | |
8468 | ||
8469 | percent = tg->uclamp_pct[clamp_id]; | |
8470 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
8471 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
8472 | } | |
8473 | ||
8474 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
8475 | { | |
8476 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
8477 | return 0; | |
8478 | } | |
8479 | ||
8480 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
8481 | { | |
8482 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
8483 | return 0; | |
8484 | } | |
8485 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
8486 | ||
052f1dc7 | 8487 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8488 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8489 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8490 | { |
5b61d50a KK |
8491 | if (shareval > scale_load_down(ULONG_MAX)) |
8492 | shareval = MAX_SHARES; | |
182446d0 | 8493 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8494 | } |
8495 | ||
182446d0 TH |
8496 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8497 | struct cftype *cft) | |
68318b8e | 8498 | { |
182446d0 | 8499 | struct task_group *tg = css_tg(css); |
68318b8e | 8500 | |
c8b28116 | 8501 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8502 | } |
ab84d31e PT |
8503 | |
8504 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8505 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8506 | ||
ab84d31e | 8507 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 8508 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
8509 | /* More than 203 days if BW_SHIFT equals 20. */ |
8510 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 8511 | |
a790de99 PT |
8512 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8513 | ||
ab84d31e PT |
8514 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8515 | { | |
56f570e5 | 8516 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8517 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8518 | |
8519 | if (tg == &root_task_group) | |
8520 | return -EINVAL; | |
8521 | ||
8522 | /* | |
8523 | * Ensure we have at some amount of bandwidth every period. This is | |
8524 | * to prevent reaching a state of large arrears when throttled via | |
8525 | * entity_tick() resulting in prolonged exit starvation. | |
8526 | */ | |
8527 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8528 | return -EINVAL; | |
8529 | ||
8530 | /* | |
8531 | * Likewise, bound things on the otherside by preventing insane quota | |
8532 | * periods. This also allows us to normalize in computing quota | |
8533 | * feasibility. | |
8534 | */ | |
8535 | if (period > max_cfs_quota_period) | |
8536 | return -EINVAL; | |
8537 | ||
d505b8af HC |
8538 | /* |
8539 | * Bound quota to defend quota against overflow during bandwidth shift. | |
8540 | */ | |
8541 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
8542 | return -EINVAL; | |
8543 | ||
0e59bdae KT |
8544 | /* |
8545 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8546 | * unthrottle_offline_cfs_rqs(). | |
8547 | */ | |
8548 | get_online_cpus(); | |
a790de99 PT |
8549 | mutex_lock(&cfs_constraints_mutex); |
8550 | ret = __cfs_schedulable(tg, period, quota); | |
8551 | if (ret) | |
8552 | goto out_unlock; | |
8553 | ||
58088ad0 | 8554 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8555 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8556 | /* |
8557 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8558 | * before making related changes, and on->off must occur afterwards | |
8559 | */ | |
8560 | if (runtime_enabled && !runtime_was_enabled) | |
8561 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8562 | raw_spin_lock_irq(&cfs_b->lock); |
8563 | cfs_b->period = ns_to_ktime(period); | |
8564 | cfs_b->quota = quota; | |
58088ad0 | 8565 | |
a9cf55b2 | 8566 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
8567 | |
8568 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
8569 | if (runtime_enabled) |
8570 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 8571 | |
ab84d31e PT |
8572 | raw_spin_unlock_irq(&cfs_b->lock); |
8573 | ||
0e59bdae | 8574 | for_each_online_cpu(i) { |
ab84d31e | 8575 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8576 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 8577 | struct rq_flags rf; |
ab84d31e | 8578 | |
8a8c69c3 | 8579 | rq_lock_irq(rq, &rf); |
58088ad0 | 8580 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8581 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8582 | |
029632fb | 8583 | if (cfs_rq->throttled) |
671fd9da | 8584 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 8585 | rq_unlock_irq(rq, &rf); |
ab84d31e | 8586 | } |
1ee14e6c BS |
8587 | if (runtime_was_enabled && !runtime_enabled) |
8588 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8589 | out_unlock: |
8590 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8591 | put_online_cpus(); |
ab84d31e | 8592 | |
a790de99 | 8593 | return ret; |
ab84d31e PT |
8594 | } |
8595 | ||
b1546edc | 8596 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e PT |
8597 | { |
8598 | u64 quota, period; | |
8599 | ||
029632fb | 8600 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8601 | if (cfs_quota_us < 0) |
8602 | quota = RUNTIME_INF; | |
1a8b4540 | 8603 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 8604 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
8605 | else |
8606 | return -EINVAL; | |
ab84d31e PT |
8607 | |
8608 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8609 | } | |
8610 | ||
b1546edc | 8611 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
8612 | { |
8613 | u64 quota_us; | |
8614 | ||
029632fb | 8615 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8616 | return -1; |
8617 | ||
029632fb | 8618 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8619 | do_div(quota_us, NSEC_PER_USEC); |
8620 | ||
8621 | return quota_us; | |
8622 | } | |
8623 | ||
b1546edc | 8624 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e PT |
8625 | { |
8626 | u64 quota, period; | |
8627 | ||
1a8b4540 KK |
8628 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
8629 | return -EINVAL; | |
8630 | ||
ab84d31e | 8631 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 8632 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8633 | |
ab84d31e PT |
8634 | return tg_set_cfs_bandwidth(tg, period, quota); |
8635 | } | |
8636 | ||
b1546edc | 8637 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
8638 | { |
8639 | u64 cfs_period_us; | |
8640 | ||
029632fb | 8641 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8642 | do_div(cfs_period_us, NSEC_PER_USEC); |
8643 | ||
8644 | return cfs_period_us; | |
8645 | } | |
8646 | ||
182446d0 TH |
8647 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8648 | struct cftype *cft) | |
ab84d31e | 8649 | { |
182446d0 | 8650 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8651 | } |
8652 | ||
182446d0 TH |
8653 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8654 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8655 | { |
182446d0 | 8656 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8657 | } |
8658 | ||
182446d0 TH |
8659 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8660 | struct cftype *cft) | |
ab84d31e | 8661 | { |
182446d0 | 8662 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8663 | } |
8664 | ||
182446d0 TH |
8665 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8666 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8667 | { |
182446d0 | 8668 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8669 | } |
8670 | ||
a790de99 PT |
8671 | struct cfs_schedulable_data { |
8672 | struct task_group *tg; | |
8673 | u64 period, quota; | |
8674 | }; | |
8675 | ||
8676 | /* | |
8677 | * normalize group quota/period to be quota/max_period | |
8678 | * note: units are usecs | |
8679 | */ | |
8680 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8681 | struct cfs_schedulable_data *d) | |
8682 | { | |
8683 | u64 quota, period; | |
8684 | ||
8685 | if (tg == d->tg) { | |
8686 | period = d->period; | |
8687 | quota = d->quota; | |
8688 | } else { | |
8689 | period = tg_get_cfs_period(tg); | |
8690 | quota = tg_get_cfs_quota(tg); | |
8691 | } | |
8692 | ||
8693 | /* note: these should typically be equivalent */ | |
8694 | if (quota == RUNTIME_INF || quota == -1) | |
8695 | return RUNTIME_INF; | |
8696 | ||
8697 | return to_ratio(period, quota); | |
8698 | } | |
8699 | ||
8700 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8701 | { | |
8702 | struct cfs_schedulable_data *d = data; | |
029632fb | 8703 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8704 | s64 quota = 0, parent_quota = -1; |
8705 | ||
8706 | if (!tg->parent) { | |
8707 | quota = RUNTIME_INF; | |
8708 | } else { | |
029632fb | 8709 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8710 | |
8711 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8712 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8713 | |
8714 | /* | |
c53593e5 TH |
8715 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
8716 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 8717 | * limit is set: |
a790de99 | 8718 | */ |
c53593e5 TH |
8719 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
8720 | quota = min(quota, parent_quota); | |
8721 | } else { | |
8722 | if (quota == RUNTIME_INF) | |
8723 | quota = parent_quota; | |
8724 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8725 | return -EINVAL; | |
8726 | } | |
a790de99 | 8727 | } |
9c58c79a | 8728 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8729 | |
8730 | return 0; | |
8731 | } | |
8732 | ||
8733 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8734 | { | |
8277434e | 8735 | int ret; |
a790de99 PT |
8736 | struct cfs_schedulable_data data = { |
8737 | .tg = tg, | |
8738 | .period = period, | |
8739 | .quota = quota, | |
8740 | }; | |
8741 | ||
8742 | if (quota != RUNTIME_INF) { | |
8743 | do_div(data.period, NSEC_PER_USEC); | |
8744 | do_div(data.quota, NSEC_PER_USEC); | |
8745 | } | |
8746 | ||
8277434e PT |
8747 | rcu_read_lock(); |
8748 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8749 | rcu_read_unlock(); | |
8750 | ||
8751 | return ret; | |
a790de99 | 8752 | } |
e8da1b18 | 8753 | |
a1f7164c | 8754 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 8755 | { |
2da8ca82 | 8756 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8757 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8758 | |
44ffc75b TH |
8759 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8760 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8761 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 8762 | |
3d6c50c2 YW |
8763 | if (schedstat_enabled() && tg != &root_task_group) { |
8764 | u64 ws = 0; | |
8765 | int i; | |
8766 | ||
8767 | for_each_possible_cpu(i) | |
8768 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
8769 | ||
8770 | seq_printf(sf, "wait_sum %llu\n", ws); | |
8771 | } | |
8772 | ||
e8da1b18 NR |
8773 | return 0; |
8774 | } | |
ab84d31e | 8775 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8776 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8777 | |
052f1dc7 | 8778 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8779 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8780 | struct cftype *cft, s64 val) | |
6f505b16 | 8781 | { |
182446d0 | 8782 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8783 | } |
8784 | ||
182446d0 TH |
8785 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8786 | struct cftype *cft) | |
6f505b16 | 8787 | { |
182446d0 | 8788 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8789 | } |
d0b27fa7 | 8790 | |
182446d0 TH |
8791 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8792 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8793 | { |
182446d0 | 8794 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8795 | } |
8796 | ||
182446d0 TH |
8797 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8798 | struct cftype *cft) | |
d0b27fa7 | 8799 | { |
182446d0 | 8800 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8801 | } |
6d6bc0ad | 8802 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8803 | |
a1f7164c | 8804 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 8805 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8806 | { |
8807 | .name = "shares", | |
f4c753b7 PM |
8808 | .read_u64 = cpu_shares_read_u64, |
8809 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8810 | }, |
052f1dc7 | 8811 | #endif |
ab84d31e PT |
8812 | #ifdef CONFIG_CFS_BANDWIDTH |
8813 | { | |
8814 | .name = "cfs_quota_us", | |
8815 | .read_s64 = cpu_cfs_quota_read_s64, | |
8816 | .write_s64 = cpu_cfs_quota_write_s64, | |
8817 | }, | |
8818 | { | |
8819 | .name = "cfs_period_us", | |
8820 | .read_u64 = cpu_cfs_period_read_u64, | |
8821 | .write_u64 = cpu_cfs_period_write_u64, | |
8822 | }, | |
e8da1b18 NR |
8823 | { |
8824 | .name = "stat", | |
a1f7164c | 8825 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 8826 | }, |
ab84d31e | 8827 | #endif |
052f1dc7 | 8828 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8829 | { |
9f0c1e56 | 8830 | .name = "rt_runtime_us", |
06ecb27c PM |
8831 | .read_s64 = cpu_rt_runtime_read, |
8832 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8833 | }, |
d0b27fa7 PZ |
8834 | { |
8835 | .name = "rt_period_us", | |
f4c753b7 PM |
8836 | .read_u64 = cpu_rt_period_read_uint, |
8837 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8838 | }, |
2480c093 PB |
8839 | #endif |
8840 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8841 | { | |
8842 | .name = "uclamp.min", | |
8843 | .flags = CFTYPE_NOT_ON_ROOT, | |
8844 | .seq_show = cpu_uclamp_min_show, | |
8845 | .write = cpu_uclamp_min_write, | |
8846 | }, | |
8847 | { | |
8848 | .name = "uclamp.max", | |
8849 | .flags = CFTYPE_NOT_ON_ROOT, | |
8850 | .seq_show = cpu_uclamp_max_show, | |
8851 | .write = cpu_uclamp_max_write, | |
8852 | }, | |
052f1dc7 | 8853 | #endif |
d1ccc66d | 8854 | { } /* Terminate */ |
68318b8e SV |
8855 | }; |
8856 | ||
d41bf8c9 TH |
8857 | static int cpu_extra_stat_show(struct seq_file *sf, |
8858 | struct cgroup_subsys_state *css) | |
0d593634 | 8859 | { |
0d593634 TH |
8860 | #ifdef CONFIG_CFS_BANDWIDTH |
8861 | { | |
d41bf8c9 | 8862 | struct task_group *tg = css_tg(css); |
0d593634 TH |
8863 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
8864 | u64 throttled_usec; | |
8865 | ||
8866 | throttled_usec = cfs_b->throttled_time; | |
8867 | do_div(throttled_usec, NSEC_PER_USEC); | |
8868 | ||
8869 | seq_printf(sf, "nr_periods %d\n" | |
8870 | "nr_throttled %d\n" | |
8871 | "throttled_usec %llu\n", | |
8872 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
8873 | throttled_usec); | |
8874 | } | |
8875 | #endif | |
8876 | return 0; | |
8877 | } | |
8878 | ||
8879 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8880 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
8881 | struct cftype *cft) | |
8882 | { | |
8883 | struct task_group *tg = css_tg(css); | |
8884 | u64 weight = scale_load_down(tg->shares); | |
8885 | ||
8886 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
8887 | } | |
8888 | ||
8889 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
8890 | struct cftype *cft, u64 weight) | |
8891 | { | |
8892 | /* | |
8893 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
8894 | * values which are 1, 100 and 10000 respectively. While it loses | |
8895 | * a bit of range on both ends, it maps pretty well onto the shares | |
8896 | * value used by scheduler and the round-trip conversions preserve | |
8897 | * the original value over the entire range. | |
8898 | */ | |
8899 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
8900 | return -ERANGE; | |
8901 | ||
8902 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
8903 | ||
8904 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
8905 | } | |
8906 | ||
8907 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
8908 | struct cftype *cft) | |
8909 | { | |
8910 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
8911 | int last_delta = INT_MAX; | |
8912 | int prio, delta; | |
8913 | ||
8914 | /* find the closest nice value to the current weight */ | |
8915 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
8916 | delta = abs(sched_prio_to_weight[prio] - weight); | |
8917 | if (delta >= last_delta) | |
8918 | break; | |
8919 | last_delta = delta; | |
8920 | } | |
8921 | ||
8922 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
8923 | } | |
8924 | ||
8925 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
8926 | struct cftype *cft, s64 nice) | |
8927 | { | |
8928 | unsigned long weight; | |
7281c8de | 8929 | int idx; |
0d593634 TH |
8930 | |
8931 | if (nice < MIN_NICE || nice > MAX_NICE) | |
8932 | return -ERANGE; | |
8933 | ||
7281c8de PZ |
8934 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
8935 | idx = array_index_nospec(idx, 40); | |
8936 | weight = sched_prio_to_weight[idx]; | |
8937 | ||
0d593634 TH |
8938 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
8939 | } | |
8940 | #endif | |
8941 | ||
8942 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
8943 | long period, long quota) | |
8944 | { | |
8945 | if (quota < 0) | |
8946 | seq_puts(sf, "max"); | |
8947 | else | |
8948 | seq_printf(sf, "%ld", quota); | |
8949 | ||
8950 | seq_printf(sf, " %ld\n", period); | |
8951 | } | |
8952 | ||
8953 | /* caller should put the current value in *@periodp before calling */ | |
8954 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
8955 | u64 *periodp, u64 *quotap) | |
8956 | { | |
8957 | char tok[21]; /* U64_MAX */ | |
8958 | ||
4c47acd8 | 8959 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
8960 | return -EINVAL; |
8961 | ||
8962 | *periodp *= NSEC_PER_USEC; | |
8963 | ||
8964 | if (sscanf(tok, "%llu", quotap)) | |
8965 | *quotap *= NSEC_PER_USEC; | |
8966 | else if (!strcmp(tok, "max")) | |
8967 | *quotap = RUNTIME_INF; | |
8968 | else | |
8969 | return -EINVAL; | |
8970 | ||
8971 | return 0; | |
8972 | } | |
8973 | ||
8974 | #ifdef CONFIG_CFS_BANDWIDTH | |
8975 | static int cpu_max_show(struct seq_file *sf, void *v) | |
8976 | { | |
8977 | struct task_group *tg = css_tg(seq_css(sf)); | |
8978 | ||
8979 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
8980 | return 0; | |
8981 | } | |
8982 | ||
8983 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
8984 | char *buf, size_t nbytes, loff_t off) | |
8985 | { | |
8986 | struct task_group *tg = css_tg(of_css(of)); | |
8987 | u64 period = tg_get_cfs_period(tg); | |
8988 | u64 quota; | |
8989 | int ret; | |
8990 | ||
8991 | ret = cpu_period_quota_parse(buf, &period, "a); | |
8992 | if (!ret) | |
8993 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
8994 | return ret ?: nbytes; | |
8995 | } | |
8996 | #endif | |
8997 | ||
8998 | static struct cftype cpu_files[] = { | |
0d593634 TH |
8999 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9000 | { | |
9001 | .name = "weight", | |
9002 | .flags = CFTYPE_NOT_ON_ROOT, | |
9003 | .read_u64 = cpu_weight_read_u64, | |
9004 | .write_u64 = cpu_weight_write_u64, | |
9005 | }, | |
9006 | { | |
9007 | .name = "weight.nice", | |
9008 | .flags = CFTYPE_NOT_ON_ROOT, | |
9009 | .read_s64 = cpu_weight_nice_read_s64, | |
9010 | .write_s64 = cpu_weight_nice_write_s64, | |
9011 | }, | |
9012 | #endif | |
9013 | #ifdef CONFIG_CFS_BANDWIDTH | |
9014 | { | |
9015 | .name = "max", | |
9016 | .flags = CFTYPE_NOT_ON_ROOT, | |
9017 | .seq_show = cpu_max_show, | |
9018 | .write = cpu_max_write, | |
9019 | }, | |
2480c093 PB |
9020 | #endif |
9021 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9022 | { | |
9023 | .name = "uclamp.min", | |
9024 | .flags = CFTYPE_NOT_ON_ROOT, | |
9025 | .seq_show = cpu_uclamp_min_show, | |
9026 | .write = cpu_uclamp_min_write, | |
9027 | }, | |
9028 | { | |
9029 | .name = "uclamp.max", | |
9030 | .flags = CFTYPE_NOT_ON_ROOT, | |
9031 | .seq_show = cpu_uclamp_max_show, | |
9032 | .write = cpu_uclamp_max_write, | |
9033 | }, | |
0d593634 TH |
9034 | #endif |
9035 | { } /* terminate */ | |
9036 | }; | |
9037 | ||
073219e9 | 9038 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 9039 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 9040 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 9041 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 9042 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 9043 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 9044 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
9045 | .can_attach = cpu_cgroup_can_attach, |
9046 | .attach = cpu_cgroup_attach, | |
a1f7164c | 9047 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 9048 | .dfl_cftypes = cpu_files, |
b38e42e9 | 9049 | .early_init = true, |
0d593634 | 9050 | .threaded = true, |
68318b8e SV |
9051 | }; |
9052 | ||
052f1dc7 | 9053 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 9054 | |
b637a328 PM |
9055 | void dump_cpu_task(int cpu) |
9056 | { | |
9057 | pr_info("Task dump for CPU %d:\n", cpu); | |
9058 | sched_show_task(cpu_curr(cpu)); | |
9059 | } | |
ed82b8a1 AK |
9060 | |
9061 | /* | |
9062 | * Nice levels are multiplicative, with a gentle 10% change for every | |
9063 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
9064 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
9065 | * that remained on nice 0. | |
9066 | * | |
9067 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
9068 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
9069 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
9070 | * If a task goes up by ~10% and another task goes down by ~10% then | |
9071 | * the relative distance between them is ~25%.) | |
9072 | */ | |
9073 | const int sched_prio_to_weight[40] = { | |
9074 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
9075 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
9076 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
9077 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
9078 | /* 0 */ 1024, 820, 655, 526, 423, | |
9079 | /* 5 */ 335, 272, 215, 172, 137, | |
9080 | /* 10 */ 110, 87, 70, 56, 45, | |
9081 | /* 15 */ 36, 29, 23, 18, 15, | |
9082 | }; | |
9083 | ||
9084 | /* | |
9085 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
9086 | * | |
9087 | * In cases where the weight does not change often, we can use the | |
9088 | * precalculated inverse to speed up arithmetics by turning divisions | |
9089 | * into multiplications: | |
9090 | */ | |
9091 | const u32 sched_prio_to_wmult[40] = { | |
9092 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
9093 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
9094 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
9095 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
9096 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
9097 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
9098 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
9099 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
9100 | }; | |
14a7405b | 9101 | |
9d246053 PA |
9102 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
9103 | { | |
9104 | trace_sched_update_nr_running_tp(rq, count); | |
9105 | } |