]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
391e43da | 3 | * kernel/sched/core.c |
1da177e4 | 4 | * |
d1ccc66d | 5 | * Core kernel scheduler code and related syscalls |
1da177e4 LT |
6 | * |
7 | * Copyright (C) 1991-2002 Linus Torvalds | |
1da177e4 | 8 | */ |
9d246053 PA |
9 | #define CREATE_TRACE_POINTS |
10 | #include <trace/events/sched.h> | |
11 | #undef CREATE_TRACE_POINTS | |
12 | ||
325ea10c | 13 | #include "sched.h" |
1da177e4 | 14 | |
7281c8de | 15 | #include <linux/nospec.h> |
85f1abe0 | 16 | |
0ed557aa | 17 | #include <linux/kcov.h> |
d08b9f0c | 18 | #include <linux/scs.h> |
0ed557aa | 19 | |
96f951ed | 20 | #include <asm/switch_to.h> |
5517d86b | 21 | #include <asm/tlb.h> |
1da177e4 | 22 | |
ea138446 | 23 | #include "../workqueue_internal.h" |
771b53d0 | 24 | #include "../../fs/io-wq.h" |
29d5e047 | 25 | #include "../smpboot.h" |
6e0534f2 | 26 | |
91c27493 | 27 | #include "pelt.h" |
1f8db415 | 28 | #include "smp.h" |
91c27493 | 29 | |
a056a5be QY |
30 | /* |
31 | * Export tracepoints that act as a bare tracehook (ie: have no trace event | |
32 | * associated with them) to allow external modules to probe them. | |
33 | */ | |
34 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp); | |
35 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp); | |
36 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp); | |
37 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp); | |
38 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp); | |
51cf18c9 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp); |
a056a5be | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); |
4581bea8 VD |
41 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp); |
42 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); | |
9d246053 | 43 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); |
a056a5be | 44 | |
029632fb | 45 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 46 | |
a73f863a | 47 | #ifdef CONFIG_SCHED_DEBUG |
bf5c91ba IM |
48 | /* |
49 | * Debugging: various feature bits | |
765cc3a4 PB |
50 | * |
51 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
52 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
53 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 54 | */ |
f00b45c1 PZ |
55 | #define SCHED_FEAT(name, enabled) \ |
56 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 57 | const_debug unsigned int sysctl_sched_features = |
391e43da | 58 | #include "features.h" |
f00b45c1 | 59 | 0; |
f00b45c1 | 60 | #undef SCHED_FEAT |
765cc3a4 | 61 | #endif |
f00b45c1 | 62 | |
b82d9fdd PZ |
63 | /* |
64 | * Number of tasks to iterate in a single balance run. | |
65 | * Limited because this is done with IRQs disabled. | |
66 | */ | |
67 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
68 | ||
fa85ae24 | 69 | /* |
d1ccc66d | 70 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
71 | * default: 1s |
72 | */ | |
9f0c1e56 | 73 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 74 | |
029632fb | 75 | __read_mostly int scheduler_running; |
6892b75e | 76 | |
9f0c1e56 PZ |
77 | /* |
78 | * part of the period that we allow rt tasks to run in us. | |
79 | * default: 0.95s | |
80 | */ | |
81 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 82 | |
58877d34 PZ |
83 | |
84 | /* | |
85 | * Serialization rules: | |
86 | * | |
87 | * Lock order: | |
88 | * | |
89 | * p->pi_lock | |
90 | * rq->lock | |
91 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
92 | * | |
93 | * rq1->lock | |
94 | * rq2->lock where: rq1 < rq2 | |
95 | * | |
96 | * Regular state: | |
97 | * | |
98 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
99 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
b19a888c | 100 | * always looks at the local rq data structures to find the most eligible task |
58877d34 PZ |
101 | * to run next. |
102 | * | |
103 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
104 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
105 | * the local CPU to avoid bouncing the runqueue state around [ see | |
106 | * ttwu_queue_wakelist() ] | |
107 | * | |
108 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
109 | * complicated to avoid having to take two rq->locks. | |
110 | * | |
111 | * Special state: | |
112 | * | |
113 | * System-calls and anything external will use task_rq_lock() which acquires | |
114 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
115 | * stable while holding either lock: | |
116 | * | |
117 | * - sched_setaffinity()/ | |
118 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
119 | * - set_user_nice(): p->se.load, p->*prio | |
120 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
121 | * p->se.load, p->rt_priority, | |
122 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
123 | * - sched_setnuma(): p->numa_preferred_nid | |
124 | * - sched_move_task()/ | |
125 | * cpu_cgroup_fork(): p->sched_task_group | |
126 | * - uclamp_update_active() p->uclamp* | |
127 | * | |
128 | * p->state <- TASK_*: | |
129 | * | |
130 | * is changed locklessly using set_current_state(), __set_current_state() or | |
131 | * set_special_state(), see their respective comments, or by | |
132 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
133 | * concurrent self. | |
134 | * | |
135 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
136 | * | |
137 | * is set by activate_task() and cleared by deactivate_task(), under | |
138 | * rq->lock. Non-zero indicates the task is runnable, the special | |
139 | * ON_RQ_MIGRATING state is used for migration without holding both | |
140 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
141 | * | |
142 | * p->on_cpu <- { 0, 1 }: | |
143 | * | |
144 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
145 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
146 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
147 | * | |
148 | * [ The astute reader will observe that it is possible for two tasks on one | |
149 | * CPU to have ->on_cpu = 1 at the same time. ] | |
150 | * | |
151 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
152 | * | |
153 | * - Don't call set_task_cpu() on a blocked task: | |
154 | * | |
155 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
156 | * the CPU assignment of blocked tasks isn't required to be valid. | |
157 | * | |
158 | * - for try_to_wake_up(), called under p->pi_lock: | |
159 | * | |
160 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
161 | * | |
162 | * - for migration called under rq->lock: | |
163 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
164 | * | |
165 | * o move_queued_task() | |
166 | * o detach_task() | |
167 | * | |
168 | * - for migration called under double_rq_lock(): | |
169 | * | |
170 | * o __migrate_swap_task() | |
171 | * o push_rt_task() / pull_rt_task() | |
172 | * o push_dl_task() / pull_dl_task() | |
173 | * o dl_task_offline_migration() | |
174 | * | |
175 | */ | |
176 | ||
3e71a462 PZ |
177 | /* |
178 | * __task_rq_lock - lock the rq @p resides on. | |
179 | */ | |
eb580751 | 180 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
181 | __acquires(rq->lock) |
182 | { | |
183 | struct rq *rq; | |
184 | ||
185 | lockdep_assert_held(&p->pi_lock); | |
186 | ||
187 | for (;;) { | |
188 | rq = task_rq(p); | |
189 | raw_spin_lock(&rq->lock); | |
190 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 191 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
192 | return rq; |
193 | } | |
194 | raw_spin_unlock(&rq->lock); | |
195 | ||
196 | while (unlikely(task_on_rq_migrating(p))) | |
197 | cpu_relax(); | |
198 | } | |
199 | } | |
200 | ||
201 | /* | |
202 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
203 | */ | |
eb580751 | 204 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
205 | __acquires(p->pi_lock) |
206 | __acquires(rq->lock) | |
207 | { | |
208 | struct rq *rq; | |
209 | ||
210 | for (;;) { | |
eb580751 | 211 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
212 | rq = task_rq(p); |
213 | raw_spin_lock(&rq->lock); | |
214 | /* | |
215 | * move_queued_task() task_rq_lock() | |
216 | * | |
217 | * ACQUIRE (rq->lock) | |
218 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
219 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
220 | * [S] ->cpu = new_cpu [L] task_rq() | |
221 | * [L] ->on_rq | |
222 | * RELEASE (rq->lock) | |
223 | * | |
c546951d | 224 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
225 | * the old rq->lock will fully serialize against the stores. |
226 | * | |
c546951d AP |
227 | * If we observe the new CPU in task_rq_lock(), the address |
228 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
229 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
230 | */ |
231 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 232 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
233 | return rq; |
234 | } | |
235 | raw_spin_unlock(&rq->lock); | |
eb580751 | 236 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
237 | |
238 | while (unlikely(task_on_rq_migrating(p))) | |
239 | cpu_relax(); | |
240 | } | |
241 | } | |
242 | ||
535b9552 IM |
243 | /* |
244 | * RQ-clock updating methods: | |
245 | */ | |
246 | ||
247 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
248 | { | |
249 | /* | |
250 | * In theory, the compile should just see 0 here, and optimize out the call | |
251 | * to sched_rt_avg_update. But I don't trust it... | |
252 | */ | |
11d4afd4 VG |
253 | s64 __maybe_unused steal = 0, irq_delta = 0; |
254 | ||
535b9552 IM |
255 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
256 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
257 | ||
258 | /* | |
259 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
260 | * this case when a previous update_rq_clock() happened inside a | |
261 | * {soft,}irq region. | |
262 | * | |
263 | * When this happens, we stop ->clock_task and only update the | |
264 | * prev_irq_time stamp to account for the part that fit, so that a next | |
265 | * update will consume the rest. This ensures ->clock_task is | |
266 | * monotonic. | |
267 | * | |
268 | * It does however cause some slight miss-attribution of {soft,}irq | |
269 | * time, a more accurate solution would be to update the irq_time using | |
270 | * the current rq->clock timestamp, except that would require using | |
271 | * atomic ops. | |
272 | */ | |
273 | if (irq_delta > delta) | |
274 | irq_delta = delta; | |
275 | ||
276 | rq->prev_irq_time += irq_delta; | |
277 | delta -= irq_delta; | |
278 | #endif | |
279 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
280 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
281 | steal = paravirt_steal_clock(cpu_of(rq)); | |
282 | steal -= rq->prev_steal_time_rq; | |
283 | ||
284 | if (unlikely(steal > delta)) | |
285 | steal = delta; | |
286 | ||
287 | rq->prev_steal_time_rq += steal; | |
288 | delta -= steal; | |
289 | } | |
290 | #endif | |
291 | ||
292 | rq->clock_task += delta; | |
293 | ||
11d4afd4 | 294 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 295 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 296 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 297 | #endif |
23127296 | 298 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
299 | } |
300 | ||
301 | void update_rq_clock(struct rq *rq) | |
302 | { | |
303 | s64 delta; | |
304 | ||
305 | lockdep_assert_held(&rq->lock); | |
306 | ||
307 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
308 | return; | |
309 | ||
310 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
311 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
312 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
313 | rq->clock_update_flags |= RQCF_UPDATED; |
314 | #endif | |
26ae58d2 | 315 | |
535b9552 IM |
316 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
317 | if (delta < 0) | |
318 | return; | |
319 | rq->clock += delta; | |
320 | update_rq_clock_task(rq, delta); | |
321 | } | |
322 | ||
8f4d37ec PZ |
323 | #ifdef CONFIG_SCHED_HRTICK |
324 | /* | |
325 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 326 | */ |
8f4d37ec | 327 | |
8f4d37ec PZ |
328 | static void hrtick_clear(struct rq *rq) |
329 | { | |
330 | if (hrtimer_active(&rq->hrtick_timer)) | |
331 | hrtimer_cancel(&rq->hrtick_timer); | |
332 | } | |
333 | ||
8f4d37ec PZ |
334 | /* |
335 | * High-resolution timer tick. | |
336 | * Runs from hardirq context with interrupts disabled. | |
337 | */ | |
338 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
339 | { | |
340 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 341 | struct rq_flags rf; |
8f4d37ec PZ |
342 | |
343 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
344 | ||
8a8c69c3 | 345 | rq_lock(rq, &rf); |
3e51f33f | 346 | update_rq_clock(rq); |
8f4d37ec | 347 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 348 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
349 | |
350 | return HRTIMER_NORESTART; | |
351 | } | |
352 | ||
95e904c7 | 353 | #ifdef CONFIG_SMP |
971ee28c | 354 | |
4961b6e1 | 355 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
356 | { |
357 | struct hrtimer *timer = &rq->hrtick_timer; | |
156ec6f4 | 358 | ktime_t time = rq->hrtick_time; |
971ee28c | 359 | |
156ec6f4 | 360 | hrtimer_start(timer, time, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
361 | } |
362 | ||
31656519 PZ |
363 | /* |
364 | * called from hardirq (IPI) context | |
365 | */ | |
366 | static void __hrtick_start(void *arg) | |
b328ca18 | 367 | { |
31656519 | 368 | struct rq *rq = arg; |
8a8c69c3 | 369 | struct rq_flags rf; |
b328ca18 | 370 | |
8a8c69c3 | 371 | rq_lock(rq, &rf); |
971ee28c | 372 | __hrtick_restart(rq); |
8a8c69c3 | 373 | rq_unlock(rq, &rf); |
b328ca18 PZ |
374 | } |
375 | ||
31656519 PZ |
376 | /* |
377 | * Called to set the hrtick timer state. | |
378 | * | |
379 | * called with rq->lock held and irqs disabled | |
380 | */ | |
029632fb | 381 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 382 | { |
31656519 | 383 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 384 | s64 delta; |
385 | ||
386 | /* | |
387 | * Don't schedule slices shorter than 10000ns, that just | |
388 | * doesn't make sense and can cause timer DoS. | |
389 | */ | |
390 | delta = max_t(s64, delay, 10000LL); | |
156ec6f4 | 391 | rq->hrtick_time = ktime_add_ns(timer->base->get_time(), delta); |
31656519 | 392 | |
fd3eafda | 393 | if (rq == this_rq()) |
971ee28c | 394 | __hrtick_restart(rq); |
fd3eafda | 395 | else |
c46fff2a | 396 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
397 | } |
398 | ||
31656519 PZ |
399 | #else |
400 | /* | |
401 | * Called to set the hrtick timer state. | |
402 | * | |
403 | * called with rq->lock held and irqs disabled | |
404 | */ | |
029632fb | 405 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 406 | { |
86893335 WL |
407 | /* |
408 | * Don't schedule slices shorter than 10000ns, that just | |
409 | * doesn't make sense. Rely on vruntime for fairness. | |
410 | */ | |
411 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 412 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 413 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 414 | } |
90b5363a | 415 | |
31656519 | 416 | #endif /* CONFIG_SMP */ |
8f4d37ec | 417 | |
77a021be | 418 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 419 | { |
31656519 | 420 | #ifdef CONFIG_SMP |
545b8c8d | 421 | INIT_CSD(&rq->hrtick_csd, __hrtick_start, rq); |
31656519 | 422 | #endif |
d5096aa6 | 423 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 424 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 425 | } |
006c75f1 | 426 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
427 | static inline void hrtick_clear(struct rq *rq) |
428 | { | |
429 | } | |
430 | ||
77a021be | 431 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
432 | { |
433 | } | |
006c75f1 | 434 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 435 | |
5529578a FW |
436 | /* |
437 | * cmpxchg based fetch_or, macro so it works for different integer types | |
438 | */ | |
439 | #define fetch_or(ptr, mask) \ | |
440 | ({ \ | |
441 | typeof(ptr) _ptr = (ptr); \ | |
442 | typeof(mask) _mask = (mask); \ | |
443 | typeof(*_ptr) _old, _val = *_ptr; \ | |
444 | \ | |
445 | for (;;) { \ | |
446 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
447 | if (_old == _val) \ | |
448 | break; \ | |
449 | _val = _old; \ | |
450 | } \ | |
451 | _old; \ | |
452 | }) | |
453 | ||
e3baac47 | 454 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
455 | /* |
456 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
457 | * this avoids any races wrt polling state changes and thereby avoids | |
458 | * spurious IPIs. | |
459 | */ | |
460 | static bool set_nr_and_not_polling(struct task_struct *p) | |
461 | { | |
462 | struct thread_info *ti = task_thread_info(p); | |
463 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
464 | } | |
e3baac47 PZ |
465 | |
466 | /* | |
467 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
468 | * | |
469 | * If this returns true, then the idle task promises to call | |
470 | * sched_ttwu_pending() and reschedule soon. | |
471 | */ | |
472 | static bool set_nr_if_polling(struct task_struct *p) | |
473 | { | |
474 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 475 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
476 | |
477 | for (;;) { | |
478 | if (!(val & _TIF_POLLING_NRFLAG)) | |
479 | return false; | |
480 | if (val & _TIF_NEED_RESCHED) | |
481 | return true; | |
482 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
483 | if (old == val) | |
484 | break; | |
485 | val = old; | |
486 | } | |
487 | return true; | |
488 | } | |
489 | ||
fd99f91a PZ |
490 | #else |
491 | static bool set_nr_and_not_polling(struct task_struct *p) | |
492 | { | |
493 | set_tsk_need_resched(p); | |
494 | return true; | |
495 | } | |
e3baac47 PZ |
496 | |
497 | #ifdef CONFIG_SMP | |
498 | static bool set_nr_if_polling(struct task_struct *p) | |
499 | { | |
500 | return false; | |
501 | } | |
502 | #endif | |
fd99f91a PZ |
503 | #endif |
504 | ||
07879c6a | 505 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
506 | { |
507 | struct wake_q_node *node = &task->wake_q; | |
508 | ||
509 | /* | |
510 | * Atomically grab the task, if ->wake_q is !nil already it means | |
b19a888c | 511 | * it's already queued (either by us or someone else) and will get the |
76751049 PZ |
512 | * wakeup due to that. |
513 | * | |
4c4e3731 PZ |
514 | * In order to ensure that a pending wakeup will observe our pending |
515 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 516 | */ |
4c4e3731 | 517 | smp_mb__before_atomic(); |
87ff19cb | 518 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 519 | return false; |
76751049 PZ |
520 | |
521 | /* | |
522 | * The head is context local, there can be no concurrency. | |
523 | */ | |
524 | *head->lastp = node; | |
525 | head->lastp = &node->next; | |
07879c6a DB |
526 | return true; |
527 | } | |
528 | ||
529 | /** | |
530 | * wake_q_add() - queue a wakeup for 'later' waking. | |
531 | * @head: the wake_q_head to add @task to | |
532 | * @task: the task to queue for 'later' wakeup | |
533 | * | |
534 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
535 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
536 | * instantly. | |
537 | * | |
538 | * This function must be used as-if it were wake_up_process(); IOW the task | |
539 | * must be ready to be woken at this location. | |
540 | */ | |
541 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
542 | { | |
543 | if (__wake_q_add(head, task)) | |
544 | get_task_struct(task); | |
545 | } | |
546 | ||
547 | /** | |
548 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
549 | * @head: the wake_q_head to add @task to | |
550 | * @task: the task to queue for 'later' wakeup | |
551 | * | |
552 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
553 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
554 | * instantly. | |
555 | * | |
556 | * This function must be used as-if it were wake_up_process(); IOW the task | |
557 | * must be ready to be woken at this location. | |
558 | * | |
559 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
560 | * that already hold reference to @task can call the 'safe' version and trust | |
561 | * wake_q to do the right thing depending whether or not the @task is already | |
562 | * queued for wakeup. | |
563 | */ | |
564 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
565 | { | |
566 | if (!__wake_q_add(head, task)) | |
567 | put_task_struct(task); | |
76751049 PZ |
568 | } |
569 | ||
570 | void wake_up_q(struct wake_q_head *head) | |
571 | { | |
572 | struct wake_q_node *node = head->first; | |
573 | ||
574 | while (node != WAKE_Q_TAIL) { | |
575 | struct task_struct *task; | |
576 | ||
577 | task = container_of(node, struct task_struct, wake_q); | |
578 | BUG_ON(!task); | |
d1ccc66d | 579 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
580 | node = node->next; |
581 | task->wake_q.next = NULL; | |
582 | ||
583 | /* | |
7696f991 AP |
584 | * wake_up_process() executes a full barrier, which pairs with |
585 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
586 | */ |
587 | wake_up_process(task); | |
588 | put_task_struct(task); | |
589 | } | |
590 | } | |
591 | ||
c24d20db | 592 | /* |
8875125e | 593 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
594 | * |
595 | * On UP this means the setting of the need_resched flag, on SMP it | |
596 | * might also involve a cross-CPU call to trigger the scheduler on | |
597 | * the target CPU. | |
598 | */ | |
8875125e | 599 | void resched_curr(struct rq *rq) |
c24d20db | 600 | { |
8875125e | 601 | struct task_struct *curr = rq->curr; |
c24d20db IM |
602 | int cpu; |
603 | ||
8875125e | 604 | lockdep_assert_held(&rq->lock); |
c24d20db | 605 | |
8875125e | 606 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
607 | return; |
608 | ||
8875125e | 609 | cpu = cpu_of(rq); |
fd99f91a | 610 | |
f27dde8d | 611 | if (cpu == smp_processor_id()) { |
8875125e | 612 | set_tsk_need_resched(curr); |
f27dde8d | 613 | set_preempt_need_resched(); |
c24d20db | 614 | return; |
f27dde8d | 615 | } |
c24d20db | 616 | |
8875125e | 617 | if (set_nr_and_not_polling(curr)) |
c24d20db | 618 | smp_send_reschedule(cpu); |
dfc68f29 AL |
619 | else |
620 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
621 | } |
622 | ||
029632fb | 623 | void resched_cpu(int cpu) |
c24d20db IM |
624 | { |
625 | struct rq *rq = cpu_rq(cpu); | |
626 | unsigned long flags; | |
627 | ||
7c2102e5 | 628 | raw_spin_lock_irqsave(&rq->lock, flags); |
a0982dfa PM |
629 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
630 | resched_curr(rq); | |
05fa785c | 631 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 632 | } |
06d8308c | 633 | |
b021fe3e | 634 | #ifdef CONFIG_SMP |
3451d024 | 635 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 636 | /* |
d1ccc66d IM |
637 | * In the semi idle case, use the nearest busy CPU for migrating timers |
638 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
639 | * |
640 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
641 | * selecting an idle CPU will add more delays to the timers than intended |
642 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 643 | */ |
bc7a34b8 | 644 | int get_nohz_timer_target(void) |
83cd4fe2 | 645 | { |
e938b9c9 | 646 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 VP |
647 | struct sched_domain *sd; |
648 | ||
e938b9c9 WL |
649 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
650 | if (!idle_cpu(cpu)) | |
651 | return cpu; | |
652 | default_cpu = cpu; | |
653 | } | |
6201b4d6 | 654 | |
057f3fad | 655 | rcu_read_lock(); |
83cd4fe2 | 656 | for_each_domain(cpu, sd) { |
e938b9c9 WL |
657 | for_each_cpu_and(i, sched_domain_span(sd), |
658 | housekeeping_cpumask(HK_FLAG_TIMER)) { | |
44496922 WL |
659 | if (cpu == i) |
660 | continue; | |
661 | ||
e938b9c9 | 662 | if (!idle_cpu(i)) { |
057f3fad PZ |
663 | cpu = i; |
664 | goto unlock; | |
665 | } | |
666 | } | |
83cd4fe2 | 667 | } |
9642d18e | 668 | |
e938b9c9 WL |
669 | if (default_cpu == -1) |
670 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
671 | cpu = default_cpu; | |
057f3fad PZ |
672 | unlock: |
673 | rcu_read_unlock(); | |
83cd4fe2 VP |
674 | return cpu; |
675 | } | |
d1ccc66d | 676 | |
06d8308c TG |
677 | /* |
678 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
679 | * idle CPU then this timer might expire before the next timer event | |
680 | * which is scheduled to wake up that CPU. In case of a completely | |
681 | * idle system the next event might even be infinite time into the | |
682 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
683 | * leaves the inner idle loop so the newly added timer is taken into | |
684 | * account when the CPU goes back to idle and evaluates the timer | |
685 | * wheel for the next timer event. | |
686 | */ | |
1c20091e | 687 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
688 | { |
689 | struct rq *rq = cpu_rq(cpu); | |
690 | ||
691 | if (cpu == smp_processor_id()) | |
692 | return; | |
693 | ||
67b9ca70 | 694 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 695 | smp_send_reschedule(cpu); |
dfc68f29 AL |
696 | else |
697 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
698 | } |
699 | ||
c5bfece2 | 700 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 701 | { |
53c5fa16 FW |
702 | /* |
703 | * We just need the target to call irq_exit() and re-evaluate | |
704 | * the next tick. The nohz full kick at least implies that. | |
705 | * If needed we can still optimize that later with an | |
706 | * empty IRQ. | |
707 | */ | |
379d9ecb PM |
708 | if (cpu_is_offline(cpu)) |
709 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 710 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
711 | if (cpu != smp_processor_id() || |
712 | tick_nohz_tick_stopped()) | |
53c5fa16 | 713 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
714 | return true; |
715 | } | |
716 | ||
717 | return false; | |
718 | } | |
719 | ||
379d9ecb PM |
720 | /* |
721 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
722 | * caller's responsibility to deal with the lost wakeup, for example, | |
723 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
724 | */ | |
1c20091e FW |
725 | void wake_up_nohz_cpu(int cpu) |
726 | { | |
c5bfece2 | 727 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
728 | wake_up_idle_cpu(cpu); |
729 | } | |
730 | ||
19a1f5ec | 731 | static void nohz_csd_func(void *info) |
45bf76df | 732 | { |
19a1f5ec PZ |
733 | struct rq *rq = info; |
734 | int cpu = cpu_of(rq); | |
735 | unsigned int flags; | |
873b4c65 VG |
736 | |
737 | /* | |
19a1f5ec | 738 | * Release the rq::nohz_csd. |
873b4c65 | 739 | */ |
19a1f5ec PZ |
740 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(cpu)); |
741 | WARN_ON(!(flags & NOHZ_KICK_MASK)); | |
45bf76df | 742 | |
19a1f5ec PZ |
743 | rq->idle_balance = idle_cpu(cpu); |
744 | if (rq->idle_balance && !need_resched()) { | |
745 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
746 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
747 | } | |
2069dd75 PZ |
748 | } |
749 | ||
3451d024 | 750 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 751 | |
ce831b38 | 752 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 753 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 754 | { |
76d92ac3 FW |
755 | int fifo_nr_running; |
756 | ||
757 | /* Deadline tasks, even if single, need the tick */ | |
758 | if (rq->dl.dl_nr_running) | |
759 | return false; | |
760 | ||
1e78cdbd | 761 | /* |
b19a888c | 762 | * If there are more than one RR tasks, we need the tick to affect the |
2548d546 | 763 | * actual RR behaviour. |
1e78cdbd | 764 | */ |
76d92ac3 FW |
765 | if (rq->rt.rr_nr_running) { |
766 | if (rq->rt.rr_nr_running == 1) | |
767 | return true; | |
768 | else | |
769 | return false; | |
1e78cdbd RR |
770 | } |
771 | ||
2548d546 PZ |
772 | /* |
773 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
774 | * forced preemption between FIFO tasks. | |
775 | */ | |
776 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
777 | if (fifo_nr_running) | |
778 | return true; | |
779 | ||
780 | /* | |
781 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
782 | * if there's more than one we need the tick for involuntary | |
783 | * preemption. | |
784 | */ | |
785 | if (rq->nr_running > 1) | |
541b8264 | 786 | return false; |
ce831b38 | 787 | |
541b8264 | 788 | return true; |
ce831b38 FW |
789 | } |
790 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 791 | #endif /* CONFIG_SMP */ |
18d95a28 | 792 | |
a790de99 PT |
793 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
794 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 795 | /* |
8277434e PT |
796 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
797 | * node and @up when leaving it for the final time. | |
798 | * | |
799 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 800 | */ |
029632fb | 801 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 802 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
803 | { |
804 | struct task_group *parent, *child; | |
eb755805 | 805 | int ret; |
c09595f6 | 806 | |
8277434e PT |
807 | parent = from; |
808 | ||
c09595f6 | 809 | down: |
eb755805 PZ |
810 | ret = (*down)(parent, data); |
811 | if (ret) | |
8277434e | 812 | goto out; |
c09595f6 PZ |
813 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
814 | parent = child; | |
815 | goto down; | |
816 | ||
817 | up: | |
818 | continue; | |
819 | } | |
eb755805 | 820 | ret = (*up)(parent, data); |
8277434e PT |
821 | if (ret || parent == from) |
822 | goto out; | |
c09595f6 PZ |
823 | |
824 | child = parent; | |
825 | parent = parent->parent; | |
826 | if (parent) | |
827 | goto up; | |
8277434e | 828 | out: |
eb755805 | 829 | return ret; |
c09595f6 PZ |
830 | } |
831 | ||
029632fb | 832 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 833 | { |
e2b245f8 | 834 | return 0; |
eb755805 | 835 | } |
18d95a28 PZ |
836 | #endif |
837 | ||
9059393e | 838 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 839 | { |
f05998d4 NR |
840 | int prio = p->static_prio - MAX_RT_PRIO; |
841 | struct load_weight *load = &p->se.load; | |
842 | ||
dd41f596 IM |
843 | /* |
844 | * SCHED_IDLE tasks get minimal weight: | |
845 | */ | |
1da1843f | 846 | if (task_has_idle_policy(p)) { |
c8b28116 | 847 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 848 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
849 | return; |
850 | } | |
71f8bd46 | 851 | |
9059393e VG |
852 | /* |
853 | * SCHED_OTHER tasks have to update their load when changing their | |
854 | * weight | |
855 | */ | |
856 | if (update_load && p->sched_class == &fair_sched_class) { | |
857 | reweight_task(p, prio); | |
858 | } else { | |
859 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
860 | load->inv_weight = sched_prio_to_wmult[prio]; | |
861 | } | |
71f8bd46 IM |
862 | } |
863 | ||
69842cba | 864 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
865 | /* |
866 | * Serializes updates of utilization clamp values | |
867 | * | |
868 | * The (slow-path) user-space triggers utilization clamp value updates which | |
869 | * can require updates on (fast-path) scheduler's data structures used to | |
870 | * support enqueue/dequeue operations. | |
871 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
872 | * requests are serialized using a mutex to reduce the risk of conflicting | |
873 | * updates or API abuses. | |
874 | */ | |
875 | static DEFINE_MUTEX(uclamp_mutex); | |
876 | ||
e8f14172 PB |
877 | /* Max allowed minimum utilization */ |
878 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
879 | ||
880 | /* Max allowed maximum utilization */ | |
881 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
882 | ||
13685c4a QY |
883 | /* |
884 | * By default RT tasks run at the maximum performance point/capacity of the | |
885 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
886 | * SCHED_CAPACITY_SCALE. | |
887 | * | |
888 | * This knob allows admins to change the default behavior when uclamp is being | |
889 | * used. In battery powered devices, particularly, running at the maximum | |
890 | * capacity and frequency will increase energy consumption and shorten the | |
891 | * battery life. | |
892 | * | |
893 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
894 | * | |
895 | * This knob will not override the system default sched_util_clamp_min defined | |
896 | * above. | |
897 | */ | |
898 | unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; | |
899 | ||
e8f14172 PB |
900 | /* All clamps are required to be less or equal than these values */ |
901 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 902 | |
46609ce2 QY |
903 | /* |
904 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
905 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
906 | * enqueue/dequeue_task(). | |
907 | * | |
908 | * This allows users to continue to enable uclamp in their kernel config with | |
909 | * minimum uclamp overhead in the fast path. | |
910 | * | |
911 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
912 | * enabled, since we have an actual users that make use of uclamp | |
913 | * functionality. | |
914 | * | |
915 | * The knobs that would enable this static key are: | |
916 | * | |
917 | * * A task modifying its uclamp value with sched_setattr(). | |
918 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
919 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
920 | */ | |
921 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
922 | ||
69842cba PB |
923 | /* Integer rounded range for each bucket */ |
924 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
925 | ||
926 | #define for_each_clamp_id(clamp_id) \ | |
927 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
928 | ||
929 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
930 | { | |
931 | return clamp_value / UCLAMP_BUCKET_DELTA; | |
932 | } | |
933 | ||
7763baac | 934 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
935 | { |
936 | if (clamp_id == UCLAMP_MIN) | |
937 | return 0; | |
938 | return SCHED_CAPACITY_SCALE; | |
939 | } | |
940 | ||
a509a7cd PB |
941 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
942 | unsigned int value, bool user_defined) | |
69842cba PB |
943 | { |
944 | uc_se->value = value; | |
945 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 946 | uc_se->user_defined = user_defined; |
69842cba PB |
947 | } |
948 | ||
e496187d | 949 | static inline unsigned int |
0413d7f3 | 950 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
951 | unsigned int clamp_value) |
952 | { | |
953 | /* | |
954 | * Avoid blocked utilization pushing up the frequency when we go | |
955 | * idle (which drops the max-clamp) by retaining the last known | |
956 | * max-clamp. | |
957 | */ | |
958 | if (clamp_id == UCLAMP_MAX) { | |
959 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
960 | return clamp_value; | |
961 | } | |
962 | ||
963 | return uclamp_none(UCLAMP_MIN); | |
964 | } | |
965 | ||
0413d7f3 | 966 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
967 | unsigned int clamp_value) |
968 | { | |
969 | /* Reset max-clamp retention only on idle exit */ | |
970 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
971 | return; | |
972 | ||
973 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
974 | } | |
975 | ||
69842cba | 976 | static inline |
7763baac | 977 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 978 | unsigned int clamp_value) |
69842cba PB |
979 | { |
980 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
981 | int bucket_id = UCLAMP_BUCKETS - 1; | |
982 | ||
983 | /* | |
984 | * Since both min and max clamps are max aggregated, find the | |
985 | * top most bucket with tasks in. | |
986 | */ | |
987 | for ( ; bucket_id >= 0; bucket_id--) { | |
988 | if (!bucket[bucket_id].tasks) | |
989 | continue; | |
990 | return bucket[bucket_id].value; | |
991 | } | |
992 | ||
993 | /* No tasks -- default clamp values */ | |
e496187d | 994 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
995 | } |
996 | ||
13685c4a QY |
997 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
998 | { | |
999 | unsigned int default_util_min; | |
1000 | struct uclamp_se *uc_se; | |
1001 | ||
1002 | lockdep_assert_held(&p->pi_lock); | |
1003 | ||
1004 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1005 | ||
1006 | /* Only sync if user didn't override the default */ | |
1007 | if (uc_se->user_defined) | |
1008 | return; | |
1009 | ||
1010 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1011 | uclamp_se_set(uc_se, default_util_min, false); | |
1012 | } | |
1013 | ||
1014 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1015 | { | |
1016 | struct rq_flags rf; | |
1017 | struct rq *rq; | |
1018 | ||
1019 | if (!rt_task(p)) | |
1020 | return; | |
1021 | ||
1022 | /* Protect updates to p->uclamp_* */ | |
1023 | rq = task_rq_lock(p, &rf); | |
1024 | __uclamp_update_util_min_rt_default(p); | |
1025 | task_rq_unlock(rq, p, &rf); | |
1026 | } | |
1027 | ||
1028 | static void uclamp_sync_util_min_rt_default(void) | |
1029 | { | |
1030 | struct task_struct *g, *p; | |
1031 | ||
1032 | /* | |
1033 | * copy_process() sysctl_uclamp | |
1034 | * uclamp_min_rt = X; | |
1035 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1036 | * // link thread smp_mb__after_spinlock() | |
1037 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1038 | * sched_post_fork() for_each_process_thread() | |
1039 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1040 | * | |
1041 | * Ensures that either sched_post_fork() will observe the new | |
1042 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1043 | * task. | |
1044 | */ | |
1045 | read_lock(&tasklist_lock); | |
1046 | smp_mb__after_spinlock(); | |
1047 | read_unlock(&tasklist_lock); | |
1048 | ||
1049 | rcu_read_lock(); | |
1050 | for_each_process_thread(g, p) | |
1051 | uclamp_update_util_min_rt_default(p); | |
1052 | rcu_read_unlock(); | |
1053 | } | |
1054 | ||
3eac870a | 1055 | static inline struct uclamp_se |
0413d7f3 | 1056 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a PB |
1057 | { |
1058 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; | |
1059 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
1060 | struct uclamp_se uc_max; | |
1061 | ||
1062 | /* | |
1063 | * Tasks in autogroups or root task group will be | |
1064 | * restricted by system defaults. | |
1065 | */ | |
1066 | if (task_group_is_autogroup(task_group(p))) | |
1067 | return uc_req; | |
1068 | if (task_group(p) == &root_task_group) | |
1069 | return uc_req; | |
1070 | ||
1071 | uc_max = task_group(p)->uclamp[clamp_id]; | |
1072 | if (uc_req.value > uc_max.value || !uc_req.user_defined) | |
1073 | return uc_max; | |
1074 | #endif | |
1075 | ||
1076 | return uc_req; | |
1077 | } | |
1078 | ||
e8f14172 PB |
1079 | /* |
1080 | * The effective clamp bucket index of a task depends on, by increasing | |
1081 | * priority: | |
1082 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1083 | * - the task group effective clamp value, for tasks not either in the root |
1084 | * group or in an autogroup | |
e8f14172 PB |
1085 | * - the system default clamp value, defined by the sysadmin |
1086 | */ | |
1087 | static inline struct uclamp_se | |
0413d7f3 | 1088 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1089 | { |
3eac870a | 1090 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1091 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1092 | ||
1093 | /* System default restrictions always apply */ | |
1094 | if (unlikely(uc_req.value > uc_max.value)) | |
1095 | return uc_max; | |
1096 | ||
1097 | return uc_req; | |
1098 | } | |
1099 | ||
686516b5 | 1100 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1101 | { |
1102 | struct uclamp_se uc_eff; | |
1103 | ||
1104 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1105 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1106 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1107 | |
1108 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1109 | ||
686516b5 | 1110 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1111 | } |
1112 | ||
69842cba PB |
1113 | /* |
1114 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1115 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1116 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1117 | * |
1118 | * Tasks can have a task-specific value requested from user-space, track | |
1119 | * within each bucket the maximum value for tasks refcounted in it. | |
1120 | * This "local max aggregation" allows to track the exact "requested" value | |
1121 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1122 | */ |
1123 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1124 | enum uclamp_id clamp_id) |
69842cba PB |
1125 | { |
1126 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1127 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1128 | struct uclamp_bucket *bucket; | |
1129 | ||
1130 | lockdep_assert_held(&rq->lock); | |
1131 | ||
e8f14172 PB |
1132 | /* Update task effective clamp */ |
1133 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1134 | ||
69842cba PB |
1135 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1136 | bucket->tasks++; | |
e8f14172 | 1137 | uc_se->active = true; |
69842cba | 1138 | |
e496187d PB |
1139 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1140 | ||
60daf9c1 PB |
1141 | /* |
1142 | * Local max aggregation: rq buckets always track the max | |
1143 | * "requested" clamp value of its RUNNABLE tasks. | |
1144 | */ | |
1145 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1146 | bucket->value = uc_se->value; | |
1147 | ||
69842cba | 1148 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1149 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1150 | } |
1151 | ||
1152 | /* | |
1153 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1154 | * is released. If this is the last task reference counting the rq's max | |
1155 | * active clamp value, then the rq's clamp value is updated. | |
1156 | * | |
1157 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1158 | * always valid. If it's detected they are not, as defensive programming, | |
1159 | * enforce the expected state and warn. | |
1160 | */ | |
1161 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1162 | enum uclamp_id clamp_id) |
69842cba PB |
1163 | { |
1164 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1165 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1166 | struct uclamp_bucket *bucket; | |
e496187d | 1167 | unsigned int bkt_clamp; |
69842cba PB |
1168 | unsigned int rq_clamp; |
1169 | ||
1170 | lockdep_assert_held(&rq->lock); | |
1171 | ||
46609ce2 QY |
1172 | /* |
1173 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1174 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1175 | * | |
1176 | * In this case the uc_se->active flag should be false since no uclamp | |
1177 | * accounting was performed at enqueue time and we can just return | |
1178 | * here. | |
1179 | * | |
b19a888c | 1180 | * Need to be careful of the following enqueue/dequeue ordering |
46609ce2 QY |
1181 | * problem too |
1182 | * | |
1183 | * enqueue(taskA) | |
1184 | * // sched_uclamp_used gets enabled | |
1185 | * enqueue(taskB) | |
1186 | * dequeue(taskA) | |
b19a888c | 1187 | * // Must not decrement bucket->tasks here |
46609ce2 QY |
1188 | * dequeue(taskB) |
1189 | * | |
1190 | * where we could end up with stale data in uc_se and | |
1191 | * bucket[uc_se->bucket_id]. | |
1192 | * | |
1193 | * The following check here eliminates the possibility of such race. | |
1194 | */ | |
1195 | if (unlikely(!uc_se->active)) | |
1196 | return; | |
1197 | ||
69842cba | 1198 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1199 | |
69842cba PB |
1200 | SCHED_WARN_ON(!bucket->tasks); |
1201 | if (likely(bucket->tasks)) | |
1202 | bucket->tasks--; | |
46609ce2 | 1203 | |
e8f14172 | 1204 | uc_se->active = false; |
69842cba | 1205 | |
60daf9c1 PB |
1206 | /* |
1207 | * Keep "local max aggregation" simple and accept to (possibly) | |
1208 | * overboost some RUNNABLE tasks in the same bucket. | |
1209 | * The rq clamp bucket value is reset to its base value whenever | |
1210 | * there are no more RUNNABLE tasks refcounting it. | |
1211 | */ | |
69842cba PB |
1212 | if (likely(bucket->tasks)) |
1213 | return; | |
1214 | ||
1215 | rq_clamp = READ_ONCE(uc_rq->value); | |
1216 | /* | |
1217 | * Defensive programming: this should never happen. If it happens, | |
1218 | * e.g. due to future modification, warn and fixup the expected value. | |
1219 | */ | |
1220 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1221 | if (bucket->value >= rq_clamp) { |
1222 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1223 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1224 | } | |
69842cba PB |
1225 | } |
1226 | ||
1227 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1228 | { | |
0413d7f3 | 1229 | enum uclamp_id clamp_id; |
69842cba | 1230 | |
46609ce2 QY |
1231 | /* |
1232 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1233 | * | |
1234 | * The condition is constructed such that a NOP is generated when | |
1235 | * sched_uclamp_used is disabled. | |
1236 | */ | |
1237 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1238 | return; | |
1239 | ||
69842cba PB |
1240 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1241 | return; | |
1242 | ||
1243 | for_each_clamp_id(clamp_id) | |
1244 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1245 | |
1246 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1247 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1248 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1249 | } |
1250 | ||
1251 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1252 | { | |
0413d7f3 | 1253 | enum uclamp_id clamp_id; |
69842cba | 1254 | |
46609ce2 QY |
1255 | /* |
1256 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1257 | * | |
1258 | * The condition is constructed such that a NOP is generated when | |
1259 | * sched_uclamp_used is disabled. | |
1260 | */ | |
1261 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1262 | return; | |
1263 | ||
69842cba PB |
1264 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1265 | return; | |
1266 | ||
1267 | for_each_clamp_id(clamp_id) | |
1268 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1269 | } | |
1270 | ||
babbe170 | 1271 | static inline void |
0413d7f3 | 1272 | uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id) |
babbe170 PB |
1273 | { |
1274 | struct rq_flags rf; | |
1275 | struct rq *rq; | |
1276 | ||
1277 | /* | |
1278 | * Lock the task and the rq where the task is (or was) queued. | |
1279 | * | |
1280 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1281 | * price to pay to safely serialize util_{min,max} updates with | |
1282 | * enqueues, dequeues and migration operations. | |
1283 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1284 | */ | |
1285 | rq = task_rq_lock(p, &rf); | |
1286 | ||
1287 | /* | |
1288 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1289 | * If the task is not yet RUNNABLE and its task_struct is not | |
1290 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1291 | * it will already see the updated clamp bucket value. | |
1292 | */ | |
6e1ff077 | 1293 | if (p->uclamp[clamp_id].active) { |
babbe170 PB |
1294 | uclamp_rq_dec_id(rq, p, clamp_id); |
1295 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1296 | } | |
1297 | ||
1298 | task_rq_unlock(rq, p, &rf); | |
1299 | } | |
1300 | ||
e3b8b6a0 | 1301 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 PB |
1302 | static inline void |
1303 | uclamp_update_active_tasks(struct cgroup_subsys_state *css, | |
1304 | unsigned int clamps) | |
1305 | { | |
0413d7f3 | 1306 | enum uclamp_id clamp_id; |
babbe170 PB |
1307 | struct css_task_iter it; |
1308 | struct task_struct *p; | |
babbe170 PB |
1309 | |
1310 | css_task_iter_start(css, 0, &it); | |
1311 | while ((p = css_task_iter_next(&it))) { | |
1312 | for_each_clamp_id(clamp_id) { | |
1313 | if ((0x1 << clamp_id) & clamps) | |
1314 | uclamp_update_active(p, clamp_id); | |
1315 | } | |
1316 | } | |
1317 | css_task_iter_end(&it); | |
1318 | } | |
1319 | ||
7274a5c1 PB |
1320 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1321 | static void uclamp_update_root_tg(void) | |
1322 | { | |
1323 | struct task_group *tg = &root_task_group; | |
1324 | ||
1325 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1326 | sysctl_sched_uclamp_util_min, false); | |
1327 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1328 | sysctl_sched_uclamp_util_max, false); | |
1329 | ||
1330 | rcu_read_lock(); | |
1331 | cpu_util_update_eff(&root_task_group.css); | |
1332 | rcu_read_unlock(); | |
1333 | } | |
1334 | #else | |
1335 | static void uclamp_update_root_tg(void) { } | |
1336 | #endif | |
1337 | ||
e8f14172 | 1338 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1339 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1340 | { |
7274a5c1 | 1341 | bool update_root_tg = false; |
13685c4a | 1342 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1343 | int result; |
1344 | ||
2480c093 | 1345 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1346 | old_min = sysctl_sched_uclamp_util_min; |
1347 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1348 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1349 | |
1350 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1351 | if (result) | |
1352 | goto undo; | |
1353 | if (!write) | |
1354 | goto done; | |
1355 | ||
1356 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1357 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1358 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1359 | ||
e8f14172 PB |
1360 | result = -EINVAL; |
1361 | goto undo; | |
1362 | } | |
1363 | ||
1364 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1365 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1366 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1367 | update_root_tg = true; |
e8f14172 PB |
1368 | } |
1369 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1370 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1371 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1372 | update_root_tg = true; |
e8f14172 PB |
1373 | } |
1374 | ||
46609ce2 QY |
1375 | if (update_root_tg) { |
1376 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1377 | uclamp_update_root_tg(); |
46609ce2 | 1378 | } |
7274a5c1 | 1379 | |
13685c4a QY |
1380 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1381 | static_branch_enable(&sched_uclamp_used); | |
1382 | uclamp_sync_util_min_rt_default(); | |
1383 | } | |
7274a5c1 | 1384 | |
e8f14172 | 1385 | /* |
7274a5c1 PB |
1386 | * We update all RUNNABLE tasks only when task groups are in use. |
1387 | * Otherwise, keep it simple and do just a lazy update at each next | |
1388 | * task enqueue time. | |
e8f14172 | 1389 | */ |
7274a5c1 | 1390 | |
e8f14172 PB |
1391 | goto done; |
1392 | ||
1393 | undo: | |
1394 | sysctl_sched_uclamp_util_min = old_min; | |
1395 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1396 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1397 | done: |
2480c093 | 1398 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1399 | |
1400 | return result; | |
1401 | } | |
1402 | ||
a509a7cd PB |
1403 | static int uclamp_validate(struct task_struct *p, |
1404 | const struct sched_attr *attr) | |
1405 | { | |
480a6ca2 DE |
1406 | int util_min = p->uclamp_req[UCLAMP_MIN].value; |
1407 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd | 1408 | |
480a6ca2 DE |
1409 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { |
1410 | util_min = attr->sched_util_min; | |
a509a7cd | 1411 | |
480a6ca2 DE |
1412 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) |
1413 | return -EINVAL; | |
1414 | } | |
1415 | ||
1416 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1417 | util_max = attr->sched_util_max; | |
1418 | ||
1419 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | |
1420 | return -EINVAL; | |
1421 | } | |
1422 | ||
1423 | if (util_min != -1 && util_max != -1 && util_min > util_max) | |
a509a7cd PB |
1424 | return -EINVAL; |
1425 | ||
e65855a5 QY |
1426 | /* |
1427 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1428 | * | |
1429 | * We need to do that here, because enabling static branches is a | |
1430 | * blocking operation which obviously cannot be done while holding | |
1431 | * scheduler locks. | |
1432 | */ | |
1433 | static_branch_enable(&sched_uclamp_used); | |
1434 | ||
a509a7cd PB |
1435 | return 0; |
1436 | } | |
1437 | ||
480a6ca2 DE |
1438 | static bool uclamp_reset(const struct sched_attr *attr, |
1439 | enum uclamp_id clamp_id, | |
1440 | struct uclamp_se *uc_se) | |
1441 | { | |
1442 | /* Reset on sched class change for a non user-defined clamp value. */ | |
1443 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | |
1444 | !uc_se->user_defined) | |
1445 | return true; | |
1446 | ||
1447 | /* Reset on sched_util_{min,max} == -1. */ | |
1448 | if (clamp_id == UCLAMP_MIN && | |
1449 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | |
1450 | attr->sched_util_min == -1) { | |
1451 | return true; | |
1452 | } | |
1453 | ||
1454 | if (clamp_id == UCLAMP_MAX && | |
1455 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | |
1456 | attr->sched_util_max == -1) { | |
1457 | return true; | |
1458 | } | |
1459 | ||
1460 | return false; | |
1461 | } | |
1462 | ||
a509a7cd PB |
1463 | static void __setscheduler_uclamp(struct task_struct *p, |
1464 | const struct sched_attr *attr) | |
1465 | { | |
0413d7f3 | 1466 | enum uclamp_id clamp_id; |
1a00d999 | 1467 | |
1a00d999 PB |
1468 | for_each_clamp_id(clamp_id) { |
1469 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
480a6ca2 | 1470 | unsigned int value; |
1a00d999 | 1471 | |
480a6ca2 | 1472 | if (!uclamp_reset(attr, clamp_id, uc_se)) |
1a00d999 PB |
1473 | continue; |
1474 | ||
13685c4a QY |
1475 | /* |
1476 | * RT by default have a 100% boost value that could be modified | |
1477 | * at runtime. | |
1478 | */ | |
1a00d999 | 1479 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
480a6ca2 | 1480 | value = sysctl_sched_uclamp_util_min_rt_default; |
13685c4a | 1481 | else |
480a6ca2 DE |
1482 | value = uclamp_none(clamp_id); |
1483 | ||
1484 | uclamp_se_set(uc_se, value, false); | |
1a00d999 | 1485 | |
1a00d999 PB |
1486 | } |
1487 | ||
a509a7cd PB |
1488 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1489 | return; | |
1490 | ||
480a6ca2 DE |
1491 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && |
1492 | attr->sched_util_min != -1) { | |
a509a7cd PB |
1493 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], |
1494 | attr->sched_util_min, true); | |
1495 | } | |
1496 | ||
480a6ca2 DE |
1497 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && |
1498 | attr->sched_util_max != -1) { | |
a509a7cd PB |
1499 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], |
1500 | attr->sched_util_max, true); | |
1501 | } | |
1502 | } | |
1503 | ||
e8f14172 PB |
1504 | static void uclamp_fork(struct task_struct *p) |
1505 | { | |
0413d7f3 | 1506 | enum uclamp_id clamp_id; |
e8f14172 | 1507 | |
13685c4a QY |
1508 | /* |
1509 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1510 | * as the task is still at its early fork stages. | |
1511 | */ | |
e8f14172 PB |
1512 | for_each_clamp_id(clamp_id) |
1513 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1514 | |
1515 | if (likely(!p->sched_reset_on_fork)) | |
1516 | return; | |
1517 | ||
1518 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1519 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1520 | uclamp_none(clamp_id), false); | |
a87498ac | 1521 | } |
e8f14172 PB |
1522 | } |
1523 | ||
13685c4a QY |
1524 | static void uclamp_post_fork(struct task_struct *p) |
1525 | { | |
1526 | uclamp_update_util_min_rt_default(p); | |
1527 | } | |
1528 | ||
d81ae8aa QY |
1529 | static void __init init_uclamp_rq(struct rq *rq) |
1530 | { | |
1531 | enum uclamp_id clamp_id; | |
1532 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1533 | ||
1534 | for_each_clamp_id(clamp_id) { | |
1535 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1536 | .value = uclamp_none(clamp_id) | |
1537 | }; | |
1538 | } | |
1539 | ||
1540 | rq->uclamp_flags = 0; | |
1541 | } | |
1542 | ||
69842cba PB |
1543 | static void __init init_uclamp(void) |
1544 | { | |
e8f14172 | 1545 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1546 | enum uclamp_id clamp_id; |
69842cba PB |
1547 | int cpu; |
1548 | ||
d81ae8aa QY |
1549 | for_each_possible_cpu(cpu) |
1550 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1551 | |
69842cba | 1552 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1553 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1554 | uclamp_none(clamp_id), false); |
69842cba | 1555 | } |
e8f14172 PB |
1556 | |
1557 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1558 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1559 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1560 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1561 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1562 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1563 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1564 | #endif |
1565 | } | |
69842cba PB |
1566 | } |
1567 | ||
1568 | #else /* CONFIG_UCLAMP_TASK */ | |
1569 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1570 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1571 | static inline int uclamp_validate(struct task_struct *p, |
1572 | const struct sched_attr *attr) | |
1573 | { | |
1574 | return -EOPNOTSUPP; | |
1575 | } | |
1576 | static void __setscheduler_uclamp(struct task_struct *p, | |
1577 | const struct sched_attr *attr) { } | |
e8f14172 | 1578 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1579 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1580 | static inline void init_uclamp(void) { } |
1581 | #endif /* CONFIG_UCLAMP_TASK */ | |
1582 | ||
1de64443 | 1583 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1584 | { |
0a67d1ee PZ |
1585 | if (!(flags & ENQUEUE_NOCLOCK)) |
1586 | update_rq_clock(rq); | |
1587 | ||
eb414681 | 1588 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 1589 | sched_info_queued(rq, p); |
eb414681 JW |
1590 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1591 | } | |
0a67d1ee | 1592 | |
69842cba | 1593 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1594 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1595 | } |
1596 | ||
1de64443 | 1597 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1598 | { |
0a67d1ee PZ |
1599 | if (!(flags & DEQUEUE_NOCLOCK)) |
1600 | update_rq_clock(rq); | |
1601 | ||
eb414681 | 1602 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 1603 | sched_info_dequeued(rq, p); |
eb414681 JW |
1604 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1605 | } | |
0a67d1ee | 1606 | |
69842cba | 1607 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1608 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1609 | } |
1610 | ||
029632fb | 1611 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1612 | { |
371fd7e7 | 1613 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
1614 | |
1615 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
1616 | } |
1617 | ||
029632fb | 1618 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1619 | { |
7dd77884 PZ |
1620 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
1621 | ||
371fd7e7 | 1622 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1623 | } |
1624 | ||
14531189 | 1625 | /* |
dd41f596 | 1626 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1627 | */ |
14531189 IM |
1628 | static inline int __normal_prio(struct task_struct *p) |
1629 | { | |
dd41f596 | 1630 | return p->static_prio; |
14531189 IM |
1631 | } |
1632 | ||
b29739f9 IM |
1633 | /* |
1634 | * Calculate the expected normal priority: i.e. priority | |
1635 | * without taking RT-inheritance into account. Might be | |
1636 | * boosted by interactivity modifiers. Changes upon fork, | |
1637 | * setprio syscalls, and whenever the interactivity | |
1638 | * estimator recalculates. | |
1639 | */ | |
36c8b586 | 1640 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1641 | { |
1642 | int prio; | |
1643 | ||
aab03e05 DF |
1644 | if (task_has_dl_policy(p)) |
1645 | prio = MAX_DL_PRIO-1; | |
1646 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
1647 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1648 | else | |
1649 | prio = __normal_prio(p); | |
1650 | return prio; | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * Calculate the current priority, i.e. the priority | |
1655 | * taken into account by the scheduler. This value might | |
1656 | * be boosted by RT tasks, or might be boosted by | |
1657 | * interactivity modifiers. Will be RT if the task got | |
1658 | * RT-boosted. If not then it returns p->normal_prio. | |
1659 | */ | |
36c8b586 | 1660 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1661 | { |
1662 | p->normal_prio = normal_prio(p); | |
1663 | /* | |
1664 | * If we are RT tasks or we were boosted to RT priority, | |
1665 | * keep the priority unchanged. Otherwise, update priority | |
1666 | * to the normal priority: | |
1667 | */ | |
1668 | if (!rt_prio(p->prio)) | |
1669 | return p->normal_prio; | |
1670 | return p->prio; | |
1671 | } | |
1672 | ||
1da177e4 LT |
1673 | /** |
1674 | * task_curr - is this task currently executing on a CPU? | |
1675 | * @p: the task in question. | |
e69f6186 YB |
1676 | * |
1677 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1678 | */ |
36c8b586 | 1679 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1680 | { |
1681 | return cpu_curr(task_cpu(p)) == p; | |
1682 | } | |
1683 | ||
67dfa1b7 | 1684 | /* |
4c9a4bc8 PZ |
1685 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1686 | * use the balance_callback list if you want balancing. | |
1687 | * | |
1688 | * this means any call to check_class_changed() must be followed by a call to | |
1689 | * balance_callback(). | |
67dfa1b7 | 1690 | */ |
cb469845 SR |
1691 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1692 | const struct sched_class *prev_class, | |
da7a735e | 1693 | int oldprio) |
cb469845 SR |
1694 | { |
1695 | if (prev_class != p->sched_class) { | |
1696 | if (prev_class->switched_from) | |
da7a735e | 1697 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1698 | |
da7a735e | 1699 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1700 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1701 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1702 | } |
1703 | ||
029632fb | 1704 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 1705 | { |
aa93cd53 | 1706 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 1707 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
1708 | else if (p->sched_class > rq->curr->sched_class) |
1709 | resched_curr(rq); | |
1e5a7405 PZ |
1710 | |
1711 | /* | |
1712 | * A queue event has occurred, and we're going to schedule. In | |
1713 | * this case, we can save a useless back to back clock update. | |
1714 | */ | |
da0c1e65 | 1715 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 1716 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
1717 | } |
1718 | ||
1da177e4 | 1719 | #ifdef CONFIG_SMP |
175f0e25 | 1720 | |
af449901 PZ |
1721 | static void |
1722 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags); | |
1723 | ||
1724 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
1725 | const struct cpumask *new_mask, | |
1726 | u32 flags); | |
1727 | ||
1728 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
1729 | { | |
1730 | if (likely(!p->migration_disabled)) | |
1731 | return; | |
1732 | ||
1733 | if (p->cpus_ptr != &p->cpus_mask) | |
1734 | return; | |
1735 | ||
1736 | /* | |
1737 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
1738 | */ | |
1739 | __do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE); | |
1740 | } | |
1741 | ||
1742 | void migrate_disable(void) | |
1743 | { | |
3015ef4b TG |
1744 | struct task_struct *p = current; |
1745 | ||
1746 | if (p->migration_disabled) { | |
1747 | p->migration_disabled++; | |
af449901 | 1748 | return; |
3015ef4b | 1749 | } |
af449901 | 1750 | |
3015ef4b TG |
1751 | preempt_disable(); |
1752 | this_rq()->nr_pinned++; | |
1753 | p->migration_disabled = 1; | |
1754 | preempt_enable(); | |
af449901 PZ |
1755 | } |
1756 | EXPORT_SYMBOL_GPL(migrate_disable); | |
1757 | ||
1758 | void migrate_enable(void) | |
1759 | { | |
1760 | struct task_struct *p = current; | |
1761 | ||
6d337eab PZ |
1762 | if (p->migration_disabled > 1) { |
1763 | p->migration_disabled--; | |
af449901 | 1764 | return; |
6d337eab | 1765 | } |
af449901 | 1766 | |
6d337eab PZ |
1767 | /* |
1768 | * Ensure stop_task runs either before or after this, and that | |
1769 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
1770 | */ | |
1771 | preempt_disable(); | |
1772 | if (p->cpus_ptr != &p->cpus_mask) | |
1773 | __set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE); | |
1774 | /* | |
1775 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
1776 | * regular cpus_mask, otherwise things that race (eg. | |
1777 | * select_fallback_rq) get confused. | |
1778 | */ | |
af449901 | 1779 | barrier(); |
6d337eab | 1780 | p->migration_disabled = 0; |
3015ef4b | 1781 | this_rq()->nr_pinned--; |
6d337eab | 1782 | preempt_enable(); |
af449901 PZ |
1783 | } |
1784 | EXPORT_SYMBOL_GPL(migrate_enable); | |
1785 | ||
3015ef4b TG |
1786 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
1787 | { | |
1788 | return rq->nr_pinned; | |
1789 | } | |
1790 | ||
175f0e25 | 1791 | /* |
bee98539 | 1792 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
1793 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
1794 | */ | |
1795 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
1796 | { | |
5ba2ffba | 1797 | /* When not in the task's cpumask, no point in looking further. */ |
3bd37062 | 1798 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
1799 | return false; |
1800 | ||
5ba2ffba PZ |
1801 | /* migrate_disabled() must be allowed to finish. */ |
1802 | if (is_migration_disabled(p)) | |
175f0e25 PZ |
1803 | return cpu_online(cpu); |
1804 | ||
5ba2ffba PZ |
1805 | /* Non kernel threads are not allowed during either online or offline. */ |
1806 | if (!(p->flags & PF_KTHREAD)) | |
1807 | return cpu_active(cpu); | |
1808 | ||
1809 | /* KTHREAD_IS_PER_CPU is always allowed. */ | |
1810 | if (kthread_is_per_cpu(p)) | |
1811 | return cpu_online(cpu); | |
1812 | ||
1813 | /* Regular kernel threads don't get to stay during offline. */ | |
1814 | if (cpu_rq(cpu)->balance_push) | |
1815 | return false; | |
1816 | ||
1817 | /* But are allowed during online. */ | |
1818 | return cpu_online(cpu); | |
175f0e25 PZ |
1819 | } |
1820 | ||
5cc389bc PZ |
1821 | /* |
1822 | * This is how migration works: | |
1823 | * | |
1824 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1825 | * stop_one_cpu(). | |
1826 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1827 | * off the CPU) | |
1828 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1829 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1830 | * it and puts it into the right queue. | |
1831 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1832 | * is done. | |
1833 | */ | |
1834 | ||
1835 | /* | |
1836 | * move_queued_task - move a queued task to new rq. | |
1837 | * | |
1838 | * Returns (locked) new rq. Old rq's lock is released. | |
1839 | */ | |
8a8c69c3 PZ |
1840 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
1841 | struct task_struct *p, int new_cpu) | |
5cc389bc | 1842 | { |
5cc389bc PZ |
1843 | lockdep_assert_held(&rq->lock); |
1844 | ||
58877d34 | 1845 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 1846 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 1847 | rq_unlock(rq, rf); |
5cc389bc PZ |
1848 | |
1849 | rq = cpu_rq(new_cpu); | |
1850 | ||
8a8c69c3 | 1851 | rq_lock(rq, rf); |
5cc389bc | 1852 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 1853 | activate_task(rq, p, 0); |
5cc389bc PZ |
1854 | check_preempt_curr(rq, p, 0); |
1855 | ||
1856 | return rq; | |
1857 | } | |
1858 | ||
1859 | struct migration_arg { | |
6d337eab PZ |
1860 | struct task_struct *task; |
1861 | int dest_cpu; | |
1862 | struct set_affinity_pending *pending; | |
1863 | }; | |
1864 | ||
1865 | struct set_affinity_pending { | |
1866 | refcount_t refs; | |
1867 | struct completion done; | |
1868 | struct cpu_stop_work stop_work; | |
1869 | struct migration_arg arg; | |
5cc389bc PZ |
1870 | }; |
1871 | ||
1872 | /* | |
d1ccc66d | 1873 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
1874 | * this because either it can't run here any more (set_cpus_allowed() |
1875 | * away from this CPU, or CPU going down), or because we're | |
1876 | * attempting to rebalance this task on exec (sched_exec). | |
1877 | * | |
1878 | * So we race with normal scheduler movements, but that's OK, as long | |
1879 | * as the task is no longer on this CPU. | |
5cc389bc | 1880 | */ |
8a8c69c3 PZ |
1881 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
1882 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 1883 | { |
5cc389bc | 1884 | /* Affinity changed (again). */ |
175f0e25 | 1885 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 1886 | return rq; |
5cc389bc | 1887 | |
15ff991e | 1888 | update_rq_clock(rq); |
8a8c69c3 | 1889 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
1890 | |
1891 | return rq; | |
5cc389bc PZ |
1892 | } |
1893 | ||
1894 | /* | |
1895 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1896 | * and performs thread migration by bumping thread off CPU then | |
1897 | * 'pushing' onto another runqueue. | |
1898 | */ | |
1899 | static int migration_cpu_stop(void *data) | |
1900 | { | |
6d337eab | 1901 | struct set_affinity_pending *pending; |
5cc389bc | 1902 | struct migration_arg *arg = data; |
5e16bbc2 | 1903 | struct task_struct *p = arg->task; |
6d337eab | 1904 | int dest_cpu = arg->dest_cpu; |
5e16bbc2 | 1905 | struct rq *rq = this_rq(); |
6d337eab | 1906 | bool complete = false; |
8a8c69c3 | 1907 | struct rq_flags rf; |
5cc389bc PZ |
1908 | |
1909 | /* | |
d1ccc66d IM |
1910 | * The original target CPU might have gone down and we might |
1911 | * be on another CPU but it doesn't matter. | |
5cc389bc | 1912 | */ |
6d337eab | 1913 | local_irq_save(rf.flags); |
5cc389bc PZ |
1914 | /* |
1915 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 1916 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
1917 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
1918 | */ | |
a1488664 | 1919 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
1920 | |
1921 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1922 | rq_lock(rq, &rf); |
6d337eab PZ |
1923 | |
1924 | pending = p->migration_pending; | |
8a6edb52 PZ |
1925 | if (pending && !arg->pending) { |
1926 | /* | |
1927 | * This happens from sched_exec() and migrate_task_to(), | |
1928 | * neither of them care about pending and just want a task to | |
1929 | * maybe move about. | |
1930 | * | |
1931 | * Even if there is a pending, we can ignore it, since | |
1932 | * affine_move_task() will have it's own stop_work's in flight | |
1933 | * which will manage the completion. | |
1934 | * | |
1935 | * Notably, pending doesn't need to match arg->pending. This can | |
1936 | * happen when tripple concurrent affine_move_task() first sets | |
1937 | * pending, then clears pending and eventually sets another | |
1938 | * pending. | |
1939 | */ | |
1940 | pending = NULL; | |
1941 | } | |
1942 | ||
5e16bbc2 PZ |
1943 | /* |
1944 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1945 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1946 | * we're holding p->pi_lock. | |
1947 | */ | |
bf89a304 | 1948 | if (task_rq(p) == rq) { |
6d337eab PZ |
1949 | if (is_migration_disabled(p)) |
1950 | goto out; | |
1951 | ||
1952 | if (pending) { | |
1953 | p->migration_pending = NULL; | |
1954 | complete = true; | |
1955 | } | |
1956 | ||
1957 | /* migrate_enable() -- we must not race against SCA */ | |
1958 | if (dest_cpu < 0) { | |
1959 | /* | |
1960 | * When this was migrate_enable() but we no longer | |
1961 | * have a @pending, a concurrent SCA 'fixed' things | |
1962 | * and we should be valid again. Nothing to do. | |
1963 | */ | |
1964 | if (!pending) { | |
1293771e | 1965 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
1966 | goto out; |
1967 | } | |
1968 | ||
1969 | dest_cpu = cpumask_any_distribute(&p->cpus_mask); | |
1970 | } | |
1971 | ||
bf89a304 | 1972 | if (task_on_rq_queued(p)) |
6d337eab | 1973 | rq = __migrate_task(rq, &rf, p, dest_cpu); |
bf89a304 | 1974 | else |
6d337eab PZ |
1975 | p->wake_cpu = dest_cpu; |
1976 | ||
d707faa6 | 1977 | } else if (dest_cpu < 0 || pending) { |
6d337eab PZ |
1978 | /* |
1979 | * This happens when we get migrated between migrate_enable()'s | |
1980 | * preempt_enable() and scheduling the stopper task. At that | |
1981 | * point we're a regular task again and not current anymore. | |
1982 | * | |
1983 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
1984 | * more likely. | |
1985 | */ | |
1986 | ||
d707faa6 VS |
1987 | /* |
1988 | * The task moved before the stopper got to run. We're holding | |
1989 | * ->pi_lock, so the allowed mask is stable - if it got | |
1990 | * somewhere allowed, we're done. | |
1991 | */ | |
1992 | if (pending && cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { | |
1993 | p->migration_pending = NULL; | |
1994 | complete = true; | |
1995 | goto out; | |
1996 | } | |
1997 | ||
6d337eab PZ |
1998 | /* |
1999 | * When this was migrate_enable() but we no longer have an | |
2000 | * @pending, a concurrent SCA 'fixed' things and we should be | |
2001 | * valid again. Nothing to do. | |
2002 | */ | |
2003 | if (!pending) { | |
1293771e | 2004 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
2005 | goto out; |
2006 | } | |
2007 | ||
2008 | /* | |
2009 | * When migrate_enable() hits a rq mis-match we can't reliably | |
2010 | * determine is_migration_disabled() and so have to chase after | |
2011 | * it. | |
2012 | */ | |
2013 | task_rq_unlock(rq, p, &rf); | |
2014 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
2015 | &pending->arg, &pending->stop_work); | |
2016 | return 0; | |
bf89a304 | 2017 | } |
6d337eab PZ |
2018 | out: |
2019 | task_rq_unlock(rq, p, &rf); | |
2020 | ||
2021 | if (complete) | |
2022 | complete_all(&pending->done); | |
2023 | ||
2024 | /* For pending->{arg,stop_work} */ | |
2025 | pending = arg->pending; | |
2026 | if (pending && refcount_dec_and_test(&pending->refs)) | |
2027 | wake_up_var(&pending->refs); | |
5e16bbc2 | 2028 | |
5cc389bc PZ |
2029 | return 0; |
2030 | } | |
2031 | ||
a7c81556 PZ |
2032 | int push_cpu_stop(void *arg) |
2033 | { | |
2034 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2035 | struct task_struct *p = arg; | |
2036 | ||
2037 | raw_spin_lock_irq(&p->pi_lock); | |
2038 | raw_spin_lock(&rq->lock); | |
2039 | ||
2040 | if (task_rq(p) != rq) | |
2041 | goto out_unlock; | |
2042 | ||
2043 | if (is_migration_disabled(p)) { | |
2044 | p->migration_flags |= MDF_PUSH; | |
2045 | goto out_unlock; | |
2046 | } | |
2047 | ||
2048 | p->migration_flags &= ~MDF_PUSH; | |
2049 | ||
2050 | if (p->sched_class->find_lock_rq) | |
2051 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2052 | |
a7c81556 PZ |
2053 | if (!lowest_rq) |
2054 | goto out_unlock; | |
2055 | ||
2056 | // XXX validate p is still the highest prio task | |
2057 | if (task_rq(p) == rq) { | |
2058 | deactivate_task(rq, p, 0); | |
2059 | set_task_cpu(p, lowest_rq->cpu); | |
2060 | activate_task(lowest_rq, p, 0); | |
2061 | resched_curr(lowest_rq); | |
2062 | } | |
2063 | ||
2064 | double_unlock_balance(rq, lowest_rq); | |
2065 | ||
2066 | out_unlock: | |
2067 | rq->push_busy = false; | |
2068 | raw_spin_unlock(&rq->lock); | |
2069 | raw_spin_unlock_irq(&p->pi_lock); | |
2070 | ||
2071 | put_task_struct(p); | |
5cc389bc PZ |
2072 | return 0; |
2073 | } | |
2074 | ||
c5b28038 PZ |
2075 | /* |
2076 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2077 | * actually call this function. | |
2078 | */ | |
9cfc3e18 | 2079 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 2080 | { |
af449901 PZ |
2081 | if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2082 | p->cpus_ptr = new_mask; | |
2083 | return; | |
2084 | } | |
2085 | ||
3bd37062 | 2086 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
2087 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
2088 | } | |
2089 | ||
9cfc3e18 PZ |
2090 | static void |
2091 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 2092 | { |
6c37067e PZ |
2093 | struct rq *rq = task_rq(p); |
2094 | bool queued, running; | |
2095 | ||
af449901 PZ |
2096 | /* |
2097 | * This here violates the locking rules for affinity, since we're only | |
2098 | * supposed to change these variables while holding both rq->lock and | |
2099 | * p->pi_lock. | |
2100 | * | |
2101 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2102 | * accesses these variables under p->pi_lock and only does so after | |
2103 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2104 | * before finish_task(). | |
2105 | * | |
2106 | * XXX do further audits, this smells like something putrid. | |
2107 | */ | |
2108 | if (flags & SCA_MIGRATE_DISABLE) | |
2109 | SCHED_WARN_ON(!p->on_cpu); | |
2110 | else | |
2111 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2112 | |
2113 | queued = task_on_rq_queued(p); | |
2114 | running = task_current(rq, p); | |
2115 | ||
2116 | if (queued) { | |
2117 | /* | |
2118 | * Because __kthread_bind() calls this on blocked tasks without | |
2119 | * holding rq->lock. | |
2120 | */ | |
2121 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 2122 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2123 | } |
2124 | if (running) | |
2125 | put_prev_task(rq, p); | |
2126 | ||
9cfc3e18 | 2127 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 2128 | |
6c37067e | 2129 | if (queued) |
7134b3e9 | 2130 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2131 | if (running) |
03b7fad1 | 2132 | set_next_task(rq, p); |
c5b28038 PZ |
2133 | } |
2134 | ||
9cfc3e18 PZ |
2135 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2136 | { | |
2137 | __do_set_cpus_allowed(p, new_mask, 0); | |
2138 | } | |
2139 | ||
6d337eab | 2140 | /* |
c777d847 VS |
2141 | * This function is wildly self concurrent; here be dragons. |
2142 | * | |
2143 | * | |
2144 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2145 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2146 | * running, we have to kick it out using the CPU stopper. | |
2147 | * | |
2148 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2149 | * Consider: | |
2150 | * | |
2151 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2152 | * | |
2153 | * P0@CPU0 P1 | |
2154 | * | |
2155 | * migrate_disable(); | |
2156 | * <preempted> | |
2157 | * set_cpus_allowed_ptr(P0, [1]); | |
2158 | * | |
2159 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2160 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2161 | * This means we need the following scheme: | |
2162 | * | |
2163 | * P0@CPU0 P1 | |
2164 | * | |
2165 | * migrate_disable(); | |
2166 | * <preempted> | |
2167 | * set_cpus_allowed_ptr(P0, [1]); | |
2168 | * <blocks> | |
2169 | * <resumes> | |
2170 | * migrate_enable(); | |
2171 | * __set_cpus_allowed_ptr(); | |
2172 | * <wakes local stopper> | |
2173 | * `--> <woken on migration completion> | |
2174 | * | |
2175 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2176 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2177 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2178 | * should come into effect at the end of the Migrate-Disable region is the last | |
2179 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2180 | * but we still need to properly signal those waiting tasks at the appropriate | |
2181 | * moment. | |
2182 | * | |
2183 | * This is implemented using struct set_affinity_pending. The first | |
2184 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2185 | * setup an instance of that struct and install it on the targeted task_struct. | |
2186 | * Any and all further callers will reuse that instance. Those then wait for | |
2187 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2188 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2189 | * | |
2190 | * | |
2191 | * (1) In the cases covered above. There is one more where the completion is | |
2192 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
2193 | * cancels the need for an active migration. Consider: | |
2194 | * | |
2195 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2196 | * | |
2197 | * P0@CPU0 P1 P2 | |
2198 | * | |
2199 | * migrate_disable(); | |
2200 | * <preempted> | |
2201 | * set_cpus_allowed_ptr(P0, [1]); | |
2202 | * <blocks> | |
2203 | * set_cpus_allowed_ptr(P0, [0, 1]); | |
2204 | * <signal completion> | |
2205 | * <awakes> | |
2206 | * | |
2207 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2208 | * pending affinity completion is preceded by an uninstallation of | |
2209 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2210 | */ |
2211 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2212 | int dest_cpu, unsigned int flags) | |
2213 | { | |
2214 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
6d337eab PZ |
2215 | bool complete = false; |
2216 | ||
2217 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2218 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2219 | struct task_struct *push_task = NULL; |
2220 | ||
2221 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2222 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2223 | rq->push_busy = true; | |
2224 | push_task = get_task_struct(p); | |
2225 | } | |
2226 | ||
6d337eab PZ |
2227 | pending = p->migration_pending; |
2228 | if (pending) { | |
2229 | refcount_inc(&pending->refs); | |
2230 | p->migration_pending = NULL; | |
2231 | complete = true; | |
2232 | } | |
2233 | task_rq_unlock(rq, p, rf); | |
2234 | ||
a7c81556 PZ |
2235 | if (push_task) { |
2236 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2237 | p, &rq->push_work); | |
2238 | } | |
2239 | ||
6d337eab PZ |
2240 | if (complete) |
2241 | goto do_complete; | |
2242 | ||
2243 | return 0; | |
2244 | } | |
2245 | ||
2246 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2247 | /* serialized by p->pi_lock */ | |
2248 | if (!p->migration_pending) { | |
c777d847 | 2249 | /* Install the request */ |
6d337eab PZ |
2250 | refcount_set(&my_pending.refs, 1); |
2251 | init_completion(&my_pending.done); | |
8a6edb52 PZ |
2252 | my_pending.arg = (struct migration_arg) { |
2253 | .task = p, | |
2254 | .dest_cpu = -1, /* any */ | |
2255 | .pending = &my_pending, | |
2256 | }; | |
2257 | ||
6d337eab PZ |
2258 | p->migration_pending = &my_pending; |
2259 | } else { | |
2260 | pending = p->migration_pending; | |
2261 | refcount_inc(&pending->refs); | |
2262 | } | |
2263 | } | |
2264 | pending = p->migration_pending; | |
2265 | /* | |
2266 | * - !MIGRATE_ENABLE: | |
2267 | * we'll have installed a pending if there wasn't one already. | |
2268 | * | |
2269 | * - MIGRATE_ENABLE: | |
2270 | * we're here because the current CPU isn't matching anymore, | |
2271 | * the only way that can happen is because of a concurrent | |
2272 | * set_cpus_allowed_ptr() call, which should then still be | |
2273 | * pending completion. | |
2274 | * | |
2275 | * Either way, we really should have a @pending here. | |
2276 | */ | |
2277 | if (WARN_ON_ONCE(!pending)) { | |
2278 | task_rq_unlock(rq, p, rf); | |
2279 | return -EINVAL; | |
2280 | } | |
2281 | ||
2282 | if (flags & SCA_MIGRATE_ENABLE) { | |
2283 | ||
2284 | refcount_inc(&pending->refs); /* pending->{arg,stop_work} */ | |
a7c81556 | 2285 | p->migration_flags &= ~MDF_PUSH; |
6d337eab PZ |
2286 | task_rq_unlock(rq, p, rf); |
2287 | ||
6d337eab PZ |
2288 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, |
2289 | &pending->arg, &pending->stop_work); | |
2290 | ||
2291 | return 0; | |
2292 | } | |
2293 | ||
2294 | if (task_running(rq, p) || p->state == TASK_WAKING) { | |
c777d847 VS |
2295 | /* |
2296 | * Lessen races (and headaches) by delegating | |
2297 | * is_migration_disabled(p) checks to the stopper, which will | |
2298 | * run on the same CPU as said p. | |
2299 | */ | |
8a6edb52 | 2300 | refcount_inc(&pending->refs); /* pending->{arg,stop_work} */ |
6d337eab | 2301 | task_rq_unlock(rq, p, rf); |
8a6edb52 PZ |
2302 | |
2303 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2304 | &pending->arg, &pending->stop_work); | |
6d337eab PZ |
2305 | |
2306 | } else { | |
2307 | ||
2308 | if (!is_migration_disabled(p)) { | |
2309 | if (task_on_rq_queued(p)) | |
2310 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2311 | ||
2312 | p->migration_pending = NULL; | |
2313 | complete = true; | |
2314 | } | |
2315 | task_rq_unlock(rq, p, rf); | |
2316 | ||
2317 | do_complete: | |
2318 | if (complete) | |
2319 | complete_all(&pending->done); | |
2320 | } | |
2321 | ||
2322 | wait_for_completion(&pending->done); | |
2323 | ||
2324 | if (refcount_dec_and_test(&pending->refs)) | |
2325 | wake_up_var(&pending->refs); | |
2326 | ||
c777d847 VS |
2327 | /* |
2328 | * Block the original owner of &pending until all subsequent callers | |
2329 | * have seen the completion and decremented the refcount | |
2330 | */ | |
6d337eab PZ |
2331 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2332 | ||
2333 | return 0; | |
2334 | } | |
2335 | ||
5cc389bc PZ |
2336 | /* |
2337 | * Change a given task's CPU affinity. Migrate the thread to a | |
2338 | * proper CPU and schedule it away if the CPU it's executing on | |
2339 | * is removed from the allowed bitmask. | |
2340 | * | |
2341 | * NOTE: the caller must have a valid reference to the task, the | |
2342 | * task must not exit() & deallocate itself prematurely. The | |
2343 | * call is not atomic; no spinlocks may be held. | |
2344 | */ | |
25834c73 | 2345 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
9cfc3e18 PZ |
2346 | const struct cpumask *new_mask, |
2347 | u32 flags) | |
5cc389bc | 2348 | { |
e9d867a6 | 2349 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 2350 | unsigned int dest_cpu; |
eb580751 PZ |
2351 | struct rq_flags rf; |
2352 | struct rq *rq; | |
5cc389bc PZ |
2353 | int ret = 0; |
2354 | ||
eb580751 | 2355 | rq = task_rq_lock(p, &rf); |
a499c3ea | 2356 | update_rq_clock(rq); |
5cc389bc | 2357 | |
af449901 | 2358 | if (p->flags & PF_KTHREAD || is_migration_disabled(p)) { |
e9d867a6 | 2359 | /* |
741ba80f PZ |
2360 | * Kernel threads are allowed on online && !active CPUs, |
2361 | * however, during cpu-hot-unplug, even these might get pushed | |
2362 | * away if not KTHREAD_IS_PER_CPU. | |
af449901 PZ |
2363 | * |
2364 | * Specifically, migration_disabled() tasks must not fail the | |
2365 | * cpumask_any_and_distribute() pick below, esp. so on | |
2366 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2367 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2368 | */ |
2369 | cpu_valid_mask = cpu_online_mask; | |
2370 | } | |
2371 | ||
25834c73 PZ |
2372 | /* |
2373 | * Must re-check here, to close a race against __kthread_bind(), | |
2374 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2375 | */ | |
9cfc3e18 | 2376 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2377 | ret = -EINVAL; |
2378 | goto out; | |
2379 | } | |
2380 | ||
885b3ba4 VS |
2381 | if (!(flags & SCA_MIGRATE_ENABLE)) { |
2382 | if (cpumask_equal(&p->cpus_mask, new_mask)) | |
2383 | goto out; | |
2384 | ||
2385 | if (WARN_ON_ONCE(p == current && | |
2386 | is_migration_disabled(p) && | |
2387 | !cpumask_test_cpu(task_cpu(p), new_mask))) { | |
2388 | ret = -EBUSY; | |
2389 | goto out; | |
2390 | } | |
2391 | } | |
5cc389bc | 2392 | |
46a87b38 PT |
2393 | /* |
2394 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2395 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2396 | * immediately required to distribute the tasks within their new mask. | |
2397 | */ | |
2398 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 2399 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2400 | ret = -EINVAL; |
2401 | goto out; | |
2402 | } | |
2403 | ||
9cfc3e18 | 2404 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 2405 | |
6d337eab | 2406 | return affine_move_task(rq, p, &rf, dest_cpu, flags); |
5cc389bc | 2407 | |
5cc389bc | 2408 | out: |
eb580751 | 2409 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
2410 | |
2411 | return ret; | |
2412 | } | |
25834c73 PZ |
2413 | |
2414 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
2415 | { | |
9cfc3e18 | 2416 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 2417 | } |
5cc389bc PZ |
2418 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2419 | ||
dd41f596 | 2420 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2421 | { |
e2912009 PZ |
2422 | #ifdef CONFIG_SCHED_DEBUG |
2423 | /* | |
2424 | * We should never call set_task_cpu() on a blocked task, | |
2425 | * ttwu() will sort out the placement. | |
2426 | */ | |
077614ee | 2427 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 2428 | !p->on_rq); |
0122ec5b | 2429 | |
3ea94de1 JP |
2430 | /* |
2431 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
2432 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
2433 | * time relying on p->on_rq. | |
2434 | */ | |
2435 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
2436 | p->sched_class == &fair_sched_class && | |
2437 | (p->on_rq && !task_on_rq_migrating(p))); | |
2438 | ||
0122ec5b | 2439 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
2440 | /* |
2441 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
2442 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
2443 | * | |
2444 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 2445 | * see task_group(). |
6c6c54e1 PZ |
2446 | * |
2447 | * Furthermore, all task_rq users should acquire both locks, see | |
2448 | * task_rq_lock(). | |
2449 | */ | |
0122ec5b PZ |
2450 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
2451 | lockdep_is_held(&task_rq(p)->lock))); | |
2452 | #endif | |
4ff9083b PZ |
2453 | /* |
2454 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
2455 | */ | |
2456 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
2457 | |
2458 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
2459 | #endif |
2460 | ||
de1d7286 | 2461 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2462 | |
0c69774e | 2463 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 2464 | if (p->sched_class->migrate_task_rq) |
1327237a | 2465 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 2466 | p->se.nr_migrations++; |
d7822b1e | 2467 | rseq_migrate(p); |
ff303e66 | 2468 | perf_event_task_migrate(p); |
0c69774e | 2469 | } |
dd41f596 IM |
2470 | |
2471 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2472 | } |
2473 | ||
0ad4e3df | 2474 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
2475 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
2476 | { | |
da0c1e65 | 2477 | if (task_on_rq_queued(p)) { |
ac66f547 | 2478 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 2479 | struct rq_flags srf, drf; |
ac66f547 PZ |
2480 | |
2481 | src_rq = task_rq(p); | |
2482 | dst_rq = cpu_rq(cpu); | |
2483 | ||
8a8c69c3 PZ |
2484 | rq_pin_lock(src_rq, &srf); |
2485 | rq_pin_lock(dst_rq, &drf); | |
2486 | ||
ac66f547 PZ |
2487 | deactivate_task(src_rq, p, 0); |
2488 | set_task_cpu(p, cpu); | |
2489 | activate_task(dst_rq, p, 0); | |
2490 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
2491 | |
2492 | rq_unpin_lock(dst_rq, &drf); | |
2493 | rq_unpin_lock(src_rq, &srf); | |
2494 | ||
ac66f547 PZ |
2495 | } else { |
2496 | /* | |
2497 | * Task isn't running anymore; make it appear like we migrated | |
2498 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 2499 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
2500 | */ |
2501 | p->wake_cpu = cpu; | |
2502 | } | |
2503 | } | |
2504 | ||
2505 | struct migration_swap_arg { | |
2506 | struct task_struct *src_task, *dst_task; | |
2507 | int src_cpu, dst_cpu; | |
2508 | }; | |
2509 | ||
2510 | static int migrate_swap_stop(void *data) | |
2511 | { | |
2512 | struct migration_swap_arg *arg = data; | |
2513 | struct rq *src_rq, *dst_rq; | |
2514 | int ret = -EAGAIN; | |
2515 | ||
62694cd5 PZ |
2516 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
2517 | return -EAGAIN; | |
2518 | ||
ac66f547 PZ |
2519 | src_rq = cpu_rq(arg->src_cpu); |
2520 | dst_rq = cpu_rq(arg->dst_cpu); | |
2521 | ||
74602315 PZ |
2522 | double_raw_lock(&arg->src_task->pi_lock, |
2523 | &arg->dst_task->pi_lock); | |
ac66f547 | 2524 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 2525 | |
ac66f547 PZ |
2526 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
2527 | goto unlock; | |
2528 | ||
2529 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
2530 | goto unlock; | |
2531 | ||
3bd37062 | 2532 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
2533 | goto unlock; |
2534 | ||
3bd37062 | 2535 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
2536 | goto unlock; |
2537 | ||
2538 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
2539 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
2540 | ||
2541 | ret = 0; | |
2542 | ||
2543 | unlock: | |
2544 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
2545 | raw_spin_unlock(&arg->dst_task->pi_lock); |
2546 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
2547 | |
2548 | return ret; | |
2549 | } | |
2550 | ||
2551 | /* | |
2552 | * Cross migrate two tasks | |
2553 | */ | |
0ad4e3df SD |
2554 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
2555 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
2556 | { |
2557 | struct migration_swap_arg arg; | |
2558 | int ret = -EINVAL; | |
2559 | ||
ac66f547 PZ |
2560 | arg = (struct migration_swap_arg){ |
2561 | .src_task = cur, | |
0ad4e3df | 2562 | .src_cpu = curr_cpu, |
ac66f547 | 2563 | .dst_task = p, |
0ad4e3df | 2564 | .dst_cpu = target_cpu, |
ac66f547 PZ |
2565 | }; |
2566 | ||
2567 | if (arg.src_cpu == arg.dst_cpu) | |
2568 | goto out; | |
2569 | ||
6acce3ef PZ |
2570 | /* |
2571 | * These three tests are all lockless; this is OK since all of them | |
2572 | * will be re-checked with proper locks held further down the line. | |
2573 | */ | |
ac66f547 PZ |
2574 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
2575 | goto out; | |
2576 | ||
3bd37062 | 2577 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
2578 | goto out; |
2579 | ||
3bd37062 | 2580 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
2581 | goto out; |
2582 | ||
286549dc | 2583 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
2584 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
2585 | ||
2586 | out: | |
ac66f547 PZ |
2587 | return ret; |
2588 | } | |
0ad4e3df | 2589 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 2590 | |
1da177e4 LT |
2591 | /* |
2592 | * wait_task_inactive - wait for a thread to unschedule. | |
2593 | * | |
85ba2d86 RM |
2594 | * If @match_state is nonzero, it's the @p->state value just checked and |
2595 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2596 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2597 | * we return a positive number (its total switch count). If a second call | |
2598 | * a short while later returns the same number, the caller can be sure that | |
2599 | * @p has remained unscheduled the whole time. | |
2600 | * | |
1da177e4 LT |
2601 | * The caller must ensure that the task *will* unschedule sometime soon, |
2602 | * else this function might spin for a *long* time. This function can't | |
2603 | * be called with interrupts off, or it may introduce deadlock with | |
2604 | * smp_call_function() if an IPI is sent by the same process we are | |
2605 | * waiting to become inactive. | |
2606 | */ | |
85ba2d86 | 2607 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 2608 | { |
da0c1e65 | 2609 | int running, queued; |
eb580751 | 2610 | struct rq_flags rf; |
85ba2d86 | 2611 | unsigned long ncsw; |
70b97a7f | 2612 | struct rq *rq; |
1da177e4 | 2613 | |
3a5c359a AK |
2614 | for (;;) { |
2615 | /* | |
2616 | * We do the initial early heuristics without holding | |
2617 | * any task-queue locks at all. We'll only try to get | |
2618 | * the runqueue lock when things look like they will | |
2619 | * work out! | |
2620 | */ | |
2621 | rq = task_rq(p); | |
fa490cfd | 2622 | |
3a5c359a AK |
2623 | /* |
2624 | * If the task is actively running on another CPU | |
2625 | * still, just relax and busy-wait without holding | |
2626 | * any locks. | |
2627 | * | |
2628 | * NOTE! Since we don't hold any locks, it's not | |
2629 | * even sure that "rq" stays as the right runqueue! | |
2630 | * But we don't care, since "task_running()" will | |
2631 | * return false if the runqueue has changed and p | |
2632 | * is actually now running somewhere else! | |
2633 | */ | |
85ba2d86 RM |
2634 | while (task_running(rq, p)) { |
2635 | if (match_state && unlikely(p->state != match_state)) | |
2636 | return 0; | |
3a5c359a | 2637 | cpu_relax(); |
85ba2d86 | 2638 | } |
fa490cfd | 2639 | |
3a5c359a AK |
2640 | /* |
2641 | * Ok, time to look more closely! We need the rq | |
2642 | * lock now, to be *sure*. If we're wrong, we'll | |
2643 | * just go back and repeat. | |
2644 | */ | |
eb580751 | 2645 | rq = task_rq_lock(p, &rf); |
27a9da65 | 2646 | trace_sched_wait_task(p); |
3a5c359a | 2647 | running = task_running(rq, p); |
da0c1e65 | 2648 | queued = task_on_rq_queued(p); |
85ba2d86 | 2649 | ncsw = 0; |
f31e11d8 | 2650 | if (!match_state || p->state == match_state) |
93dcf55f | 2651 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 2652 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 2653 | |
85ba2d86 RM |
2654 | /* |
2655 | * If it changed from the expected state, bail out now. | |
2656 | */ | |
2657 | if (unlikely(!ncsw)) | |
2658 | break; | |
2659 | ||
3a5c359a AK |
2660 | /* |
2661 | * Was it really running after all now that we | |
2662 | * checked with the proper locks actually held? | |
2663 | * | |
2664 | * Oops. Go back and try again.. | |
2665 | */ | |
2666 | if (unlikely(running)) { | |
2667 | cpu_relax(); | |
2668 | continue; | |
2669 | } | |
fa490cfd | 2670 | |
3a5c359a AK |
2671 | /* |
2672 | * It's not enough that it's not actively running, | |
2673 | * it must be off the runqueue _entirely_, and not | |
2674 | * preempted! | |
2675 | * | |
80dd99b3 | 2676 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2677 | * running right now), it's preempted, and we should |
2678 | * yield - it could be a while. | |
2679 | */ | |
da0c1e65 | 2680 | if (unlikely(queued)) { |
8b0e1953 | 2681 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
2682 | |
2683 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2684 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2685 | continue; |
2686 | } | |
fa490cfd | 2687 | |
3a5c359a AK |
2688 | /* |
2689 | * Ahh, all good. It wasn't running, and it wasn't | |
2690 | * runnable, which means that it will never become | |
2691 | * running in the future either. We're all done! | |
2692 | */ | |
2693 | break; | |
2694 | } | |
85ba2d86 RM |
2695 | |
2696 | return ncsw; | |
1da177e4 LT |
2697 | } |
2698 | ||
2699 | /*** | |
2700 | * kick_process - kick a running thread to enter/exit the kernel | |
2701 | * @p: the to-be-kicked thread | |
2702 | * | |
2703 | * Cause a process which is running on another CPU to enter | |
2704 | * kernel-mode, without any delay. (to get signals handled.) | |
2705 | * | |
25985edc | 2706 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2707 | * because all it wants to ensure is that the remote task enters |
2708 | * the kernel. If the IPI races and the task has been migrated | |
2709 | * to another CPU then no harm is done and the purpose has been | |
2710 | * achieved as well. | |
2711 | */ | |
36c8b586 | 2712 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2713 | { |
2714 | int cpu; | |
2715 | ||
2716 | preempt_disable(); | |
2717 | cpu = task_cpu(p); | |
2718 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2719 | smp_send_reschedule(cpu); | |
2720 | preempt_enable(); | |
2721 | } | |
b43e3521 | 2722 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 2723 | |
30da688e | 2724 | /* |
3bd37062 | 2725 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
2726 | * |
2727 | * A few notes on cpu_active vs cpu_online: | |
2728 | * | |
2729 | * - cpu_active must be a subset of cpu_online | |
2730 | * | |
97fb7a0a | 2731 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 2732 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 2733 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
2734 | * see it. |
2735 | * | |
d1ccc66d | 2736 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 2737 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 2738 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
2739 | * off. |
2740 | * | |
2741 | * This means that fallback selection must not select !active CPUs. | |
2742 | * And can assume that any active CPU must be online. Conversely | |
2743 | * select_task_rq() below may allow selection of !active CPUs in order | |
2744 | * to satisfy the above rules. | |
30da688e | 2745 | */ |
5da9a0fb PZ |
2746 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2747 | { | |
aa00d89c TC |
2748 | int nid = cpu_to_node(cpu); |
2749 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
2750 | enum { cpuset, possible, fail } state = cpuset; |
2751 | int dest_cpu; | |
5da9a0fb | 2752 | |
aa00d89c | 2753 | /* |
d1ccc66d IM |
2754 | * If the node that the CPU is on has been offlined, cpu_to_node() |
2755 | * will return -1. There is no CPU on the node, and we should | |
2756 | * select the CPU on the other node. | |
aa00d89c TC |
2757 | */ |
2758 | if (nid != -1) { | |
2759 | nodemask = cpumask_of_node(nid); | |
2760 | ||
2761 | /* Look for allowed, online CPU in same node. */ | |
2762 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
2763 | if (!cpu_active(dest_cpu)) |
2764 | continue; | |
3bd37062 | 2765 | if (cpumask_test_cpu(dest_cpu, p->cpus_ptr)) |
aa00d89c TC |
2766 | return dest_cpu; |
2767 | } | |
2baab4e9 | 2768 | } |
5da9a0fb | 2769 | |
2baab4e9 PZ |
2770 | for (;;) { |
2771 | /* Any allowed, online CPU? */ | |
3bd37062 | 2772 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 2773 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 2774 | continue; |
175f0e25 | 2775 | |
2baab4e9 PZ |
2776 | goto out; |
2777 | } | |
5da9a0fb | 2778 | |
e73e85f0 | 2779 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
2780 | switch (state) { |
2781 | case cpuset: | |
e73e85f0 ON |
2782 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
2783 | cpuset_cpus_allowed_fallback(p); | |
2784 | state = possible; | |
2785 | break; | |
2786 | } | |
df561f66 | 2787 | fallthrough; |
2baab4e9 | 2788 | case possible: |
af449901 PZ |
2789 | /* |
2790 | * XXX When called from select_task_rq() we only | |
2791 | * hold p->pi_lock and again violate locking order. | |
2792 | * | |
2793 | * More yuck to audit. | |
2794 | */ | |
2baab4e9 PZ |
2795 | do_set_cpus_allowed(p, cpu_possible_mask); |
2796 | state = fail; | |
2797 | break; | |
2798 | ||
2799 | case fail: | |
2800 | BUG(); | |
2801 | break; | |
2802 | } | |
2803 | } | |
2804 | ||
2805 | out: | |
2806 | if (state != cpuset) { | |
2807 | /* | |
2808 | * Don't tell them about moving exiting tasks or | |
2809 | * kernel threads (both mm NULL), since they never | |
2810 | * leave kernel. | |
2811 | */ | |
2812 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 2813 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
2814 | task_pid_nr(p), p->comm, cpu); |
2815 | } | |
5da9a0fb PZ |
2816 | } |
2817 | ||
2818 | return dest_cpu; | |
2819 | } | |
2820 | ||
e2912009 | 2821 | /* |
3bd37062 | 2822 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 2823 | */ |
970b13ba | 2824 | static inline |
3aef1551 | 2825 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 2826 | { |
cbce1a68 PZ |
2827 | lockdep_assert_held(&p->pi_lock); |
2828 | ||
af449901 | 2829 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 2830 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 2831 | else |
3bd37062 | 2832 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
2833 | |
2834 | /* | |
2835 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 2836 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 2837 | * CPU. |
e2912009 PZ |
2838 | * |
2839 | * Since this is common to all placement strategies, this lives here. | |
2840 | * | |
2841 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2842 | * not worry about this generic constraint ] | |
2843 | */ | |
7af443ee | 2844 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 2845 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2846 | |
2847 | return cpu; | |
970b13ba | 2848 | } |
09a40af5 | 2849 | |
f5832c19 NP |
2850 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2851 | { | |
ded467dc | 2852 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
2853 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
2854 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2855 | ||
2856 | if (stop) { | |
2857 | /* | |
2858 | * Make it appear like a SCHED_FIFO task, its something | |
2859 | * userspace knows about and won't get confused about. | |
2860 | * | |
2861 | * Also, it will make PI more or less work without too | |
2862 | * much confusion -- but then, stop work should not | |
2863 | * rely on PI working anyway. | |
2864 | */ | |
2865 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2866 | ||
2867 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
2868 | |
2869 | /* | |
2870 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
2871 | * adjust the effective priority of a task. As a result, | |
2872 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
2873 | * which can then trigger wakeups of the stop thread to push | |
2874 | * around the current task. | |
2875 | * | |
2876 | * The stop task itself will never be part of the PI-chain, it | |
2877 | * never blocks, therefore that ->pi_lock recursion is safe. | |
2878 | * Tell lockdep about this by placing the stop->pi_lock in its | |
2879 | * own class. | |
2880 | */ | |
2881 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
2882 | } |
2883 | ||
2884 | cpu_rq(cpu)->stop = stop; | |
2885 | ||
2886 | if (old_stop) { | |
2887 | /* | |
2888 | * Reset it back to a normal scheduling class so that | |
2889 | * it can die in pieces. | |
2890 | */ | |
2891 | old_stop->sched_class = &rt_sched_class; | |
2892 | } | |
2893 | } | |
2894 | ||
74d862b6 | 2895 | #else /* CONFIG_SMP */ |
25834c73 PZ |
2896 | |
2897 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
2898 | const struct cpumask *new_mask, |
2899 | u32 flags) | |
25834c73 PZ |
2900 | { |
2901 | return set_cpus_allowed_ptr(p, new_mask); | |
2902 | } | |
2903 | ||
af449901 PZ |
2904 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
2905 | ||
3015ef4b TG |
2906 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2907 | { | |
2908 | return false; | |
2909 | } | |
2910 | ||
74d862b6 | 2911 | #endif /* !CONFIG_SMP */ |
970b13ba | 2912 | |
d7c01d27 | 2913 | static void |
b84cb5df | 2914 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2915 | { |
4fa8d299 | 2916 | struct rq *rq; |
b84cb5df | 2917 | |
4fa8d299 JP |
2918 | if (!schedstat_enabled()) |
2919 | return; | |
2920 | ||
2921 | rq = this_rq(); | |
d7c01d27 | 2922 | |
4fa8d299 JP |
2923 | #ifdef CONFIG_SMP |
2924 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
2925 | __schedstat_inc(rq->ttwu_local); |
2926 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
2927 | } else { |
2928 | struct sched_domain *sd; | |
2929 | ||
b85c8b71 | 2930 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 2931 | rcu_read_lock(); |
4fa8d299 | 2932 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 2933 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 2934 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
2935 | break; |
2936 | } | |
2937 | } | |
057f3fad | 2938 | rcu_read_unlock(); |
d7c01d27 | 2939 | } |
f339b9dc PZ |
2940 | |
2941 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 2942 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
2943 | #endif /* CONFIG_SMP */ |
2944 | ||
b85c8b71 PZ |
2945 | __schedstat_inc(rq->ttwu_count); |
2946 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
2947 | |
2948 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 2949 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2950 | } |
2951 | ||
23f41eeb PZ |
2952 | /* |
2953 | * Mark the task runnable and perform wakeup-preemption. | |
2954 | */ | |
e7904a28 | 2955 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2956 | struct rq_flags *rf) |
9ed3811a | 2957 | { |
9ed3811a | 2958 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 2959 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
2960 | trace_sched_wakeup(p); |
2961 | ||
9ed3811a | 2962 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
2963 | if (p->sched_class->task_woken) { |
2964 | /* | |
b19a888c | 2965 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 2966 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 2967 | */ |
d8ac8971 | 2968 | rq_unpin_lock(rq, rf); |
9ed3811a | 2969 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2970 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 2971 | } |
9ed3811a | 2972 | |
e69c6341 | 2973 | if (rq->idle_stamp) { |
78becc27 | 2974 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 2975 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 2976 | |
abfafa54 JL |
2977 | update_avg(&rq->avg_idle, delta); |
2978 | ||
2979 | if (rq->avg_idle > max) | |
9ed3811a | 2980 | rq->avg_idle = max; |
abfafa54 | 2981 | |
9ed3811a TH |
2982 | rq->idle_stamp = 0; |
2983 | } | |
2984 | #endif | |
2985 | } | |
2986 | ||
c05fbafb | 2987 | static void |
e7904a28 | 2988 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2989 | struct rq_flags *rf) |
c05fbafb | 2990 | { |
77558e4d | 2991 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 2992 | |
cbce1a68 PZ |
2993 | lockdep_assert_held(&rq->lock); |
2994 | ||
c05fbafb PZ |
2995 | if (p->sched_contributes_to_load) |
2996 | rq->nr_uninterruptible--; | |
b5179ac7 | 2997 | |
dbfb089d | 2998 | #ifdef CONFIG_SMP |
b5179ac7 | 2999 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 3000 | en_flags |= ENQUEUE_MIGRATED; |
ec618b84 | 3001 | else |
c05fbafb | 3002 | #endif |
ec618b84 PZ |
3003 | if (p->in_iowait) { |
3004 | delayacct_blkio_end(p); | |
3005 | atomic_dec(&task_rq(p)->nr_iowait); | |
3006 | } | |
c05fbafb | 3007 | |
1b174a2c | 3008 | activate_task(rq, p, en_flags); |
d8ac8971 | 3009 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
3010 | } |
3011 | ||
3012 | /* | |
58877d34 PZ |
3013 | * Consider @p being inside a wait loop: |
3014 | * | |
3015 | * for (;;) { | |
3016 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
3017 | * | |
3018 | * if (CONDITION) | |
3019 | * break; | |
3020 | * | |
3021 | * schedule(); | |
3022 | * } | |
3023 | * __set_current_state(TASK_RUNNING); | |
3024 | * | |
3025 | * between set_current_state() and schedule(). In this case @p is still | |
3026 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3027 | * an atomic manner. | |
3028 | * | |
3029 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3030 | * then schedule() must still happen and p->state can be changed to | |
3031 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3032 | * need to do a full wakeup with enqueue. | |
3033 | * | |
3034 | * Returns: %true when the wakeup is done, | |
3035 | * %false otherwise. | |
c05fbafb | 3036 | */ |
58877d34 | 3037 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3038 | { |
eb580751 | 3039 | struct rq_flags rf; |
c05fbafb PZ |
3040 | struct rq *rq; |
3041 | int ret = 0; | |
3042 | ||
eb580751 | 3043 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3044 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3045 | /* check_preempt_curr() may use rq clock */ |
3046 | update_rq_clock(rq); | |
d8ac8971 | 3047 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3048 | ret = 1; |
3049 | } | |
eb580751 | 3050 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3051 | |
3052 | return ret; | |
3053 | } | |
3054 | ||
317f3941 | 3055 | #ifdef CONFIG_SMP |
a1488664 | 3056 | void sched_ttwu_pending(void *arg) |
317f3941 | 3057 | { |
a1488664 | 3058 | struct llist_node *llist = arg; |
317f3941 | 3059 | struct rq *rq = this_rq(); |
73215849 | 3060 | struct task_struct *p, *t; |
d8ac8971 | 3061 | struct rq_flags rf; |
317f3941 | 3062 | |
e3baac47 PZ |
3063 | if (!llist) |
3064 | return; | |
3065 | ||
126c2092 PZ |
3066 | /* |
3067 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
3068 | * Races such that false-negatives are possible, since they | |
3069 | * are shorter lived that false-positives would be. | |
3070 | */ | |
3071 | WRITE_ONCE(rq->ttwu_pending, 0); | |
3072 | ||
8a8c69c3 | 3073 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3074 | update_rq_clock(rq); |
317f3941 | 3075 | |
8c4890d1 | 3076 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3077 | if (WARN_ON_ONCE(p->on_cpu)) |
3078 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3079 | ||
3080 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3081 | set_task_cpu(p, cpu_of(rq)); | |
3082 | ||
73215849 | 3083 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3084 | } |
317f3941 | 3085 | |
8a8c69c3 | 3086 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3087 | } |
3088 | ||
b2a02fc4 | 3089 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3090 | { |
b2a02fc4 | 3091 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3092 | |
b2a02fc4 PZ |
3093 | if (!set_nr_if_polling(rq->idle)) |
3094 | arch_send_call_function_single_ipi(cpu); | |
3095 | else | |
3096 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3097 | } |
3098 | ||
2ebb1771 MG |
3099 | /* |
3100 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3101 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3102 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3103 | * of the wakeup instead of the waker. | |
3104 | */ | |
3105 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3106 | { |
e3baac47 PZ |
3107 | struct rq *rq = cpu_rq(cpu); |
3108 | ||
b7e7ade3 PZ |
3109 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3110 | ||
126c2092 | 3111 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3112 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3113 | } |
d6aa8f85 | 3114 | |
f6be8af1 CL |
3115 | void wake_up_if_idle(int cpu) |
3116 | { | |
3117 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3118 | struct rq_flags rf; |
f6be8af1 | 3119 | |
fd7de1e8 AL |
3120 | rcu_read_lock(); |
3121 | ||
3122 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3123 | goto out; | |
f6be8af1 CL |
3124 | |
3125 | if (set_nr_if_polling(rq->idle)) { | |
3126 | trace_sched_wake_idle_without_ipi(cpu); | |
3127 | } else { | |
8a8c69c3 | 3128 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
3129 | if (is_idle_task(rq->curr)) |
3130 | smp_send_reschedule(cpu); | |
d1ccc66d | 3131 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 3132 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 3133 | } |
fd7de1e8 AL |
3134 | |
3135 | out: | |
3136 | rcu_read_unlock(); | |
f6be8af1 CL |
3137 | } |
3138 | ||
39be3501 | 3139 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
3140 | { |
3141 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
3142 | } | |
c6e7bd7a | 3143 | |
2ebb1771 MG |
3144 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
3145 | { | |
5ba2ffba PZ |
3146 | /* |
3147 | * Do not complicate things with the async wake_list while the CPU is | |
3148 | * in hotplug state. | |
3149 | */ | |
3150 | if (!cpu_active(cpu)) | |
3151 | return false; | |
3152 | ||
2ebb1771 MG |
3153 | /* |
3154 | * If the CPU does not share cache, then queue the task on the | |
3155 | * remote rqs wakelist to avoid accessing remote data. | |
3156 | */ | |
3157 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3158 | return true; | |
3159 | ||
3160 | /* | |
3161 | * If the task is descheduling and the only running task on the | |
3162 | * CPU then use the wakelist to offload the task activation to | |
3163 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
3164 | * nr_running is checked to avoid unnecessary task stacking. | |
3165 | */ | |
739f70b4 | 3166 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
3167 | return true; |
3168 | ||
3169 | return false; | |
3170 | } | |
3171 | ||
3172 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3173 | { |
2ebb1771 | 3174 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
3175 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
3176 | return false; | |
3177 | ||
c6e7bd7a | 3178 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3179 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3180 | return true; |
3181 | } | |
3182 | ||
3183 | return false; | |
3184 | } | |
58877d34 PZ |
3185 | |
3186 | #else /* !CONFIG_SMP */ | |
3187 | ||
3188 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3189 | { | |
3190 | return false; | |
3191 | } | |
3192 | ||
d6aa8f85 | 3193 | #endif /* CONFIG_SMP */ |
317f3941 | 3194 | |
b5179ac7 | 3195 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3196 | { |
3197 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3198 | struct rq_flags rf; |
c05fbafb | 3199 | |
2ebb1771 | 3200 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3201 | return; |
317f3941 | 3202 | |
8a8c69c3 | 3203 | rq_lock(rq, &rf); |
77558e4d | 3204 | update_rq_clock(rq); |
d8ac8971 | 3205 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3206 | rq_unlock(rq, &rf); |
9ed3811a TH |
3207 | } |
3208 | ||
8643cda5 PZ |
3209 | /* |
3210 | * Notes on Program-Order guarantees on SMP systems. | |
3211 | * | |
3212 | * MIGRATION | |
3213 | * | |
3214 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3215 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3216 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3217 | * |
3218 | * For migration (of runnable tasks) this is provided by the following means: | |
3219 | * | |
3220 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3221 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3222 | * rq(c1)->lock (if not at the same time, then in that order). | |
3223 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3224 | * | |
7696f991 | 3225 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3226 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3227 | * |
3228 | * Example: | |
3229 | * | |
3230 | * CPU0 CPU1 CPU2 | |
3231 | * | |
3232 | * LOCK rq(0)->lock | |
3233 | * sched-out X | |
3234 | * sched-in Y | |
3235 | * UNLOCK rq(0)->lock | |
3236 | * | |
3237 | * LOCK rq(0)->lock // orders against CPU0 | |
3238 | * dequeue X | |
3239 | * UNLOCK rq(0)->lock | |
3240 | * | |
3241 | * LOCK rq(1)->lock | |
3242 | * enqueue X | |
3243 | * UNLOCK rq(1)->lock | |
3244 | * | |
3245 | * LOCK rq(1)->lock // orders against CPU2 | |
3246 | * sched-out Z | |
3247 | * sched-in X | |
3248 | * UNLOCK rq(1)->lock | |
3249 | * | |
3250 | * | |
3251 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3252 | * | |
3253 | * For blocking we (obviously) need to provide the same guarantee as for | |
3254 | * migration. However the means are completely different as there is no lock | |
3255 | * chain to provide order. Instead we do: | |
3256 | * | |
58877d34 PZ |
3257 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3258 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3259 | * |
3260 | * Example: | |
3261 | * | |
3262 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
3263 | * | |
3264 | * LOCK rq(0)->lock LOCK X->pi_lock | |
3265 | * dequeue X | |
3266 | * sched-out X | |
3267 | * smp_store_release(X->on_cpu, 0); | |
3268 | * | |
1f03e8d2 | 3269 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
3270 | * X->state = WAKING |
3271 | * set_task_cpu(X,2) | |
3272 | * | |
3273 | * LOCK rq(2)->lock | |
3274 | * enqueue X | |
3275 | * X->state = RUNNING | |
3276 | * UNLOCK rq(2)->lock | |
3277 | * | |
3278 | * LOCK rq(2)->lock // orders against CPU1 | |
3279 | * sched-out Z | |
3280 | * sched-in X | |
3281 | * UNLOCK rq(2)->lock | |
3282 | * | |
3283 | * UNLOCK X->pi_lock | |
3284 | * UNLOCK rq(0)->lock | |
3285 | * | |
3286 | * | |
7696f991 AP |
3287 | * However, for wakeups there is a second guarantee we must provide, namely we |
3288 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
3289 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
3290 | */ |
3291 | ||
9ed3811a | 3292 | /** |
1da177e4 | 3293 | * try_to_wake_up - wake up a thread |
9ed3811a | 3294 | * @p: the thread to be awakened |
1da177e4 | 3295 | * @state: the mask of task states that can be woken |
9ed3811a | 3296 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 3297 | * |
58877d34 PZ |
3298 | * Conceptually does: |
3299 | * | |
3300 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 3301 | * |
a2250238 PZ |
3302 | * If the task was not queued/runnable, also place it back on a runqueue. |
3303 | * | |
58877d34 PZ |
3304 | * This function is atomic against schedule() which would dequeue the task. |
3305 | * | |
3306 | * It issues a full memory barrier before accessing @p->state, see the comment | |
3307 | * with set_current_state(). | |
a2250238 | 3308 | * |
58877d34 | 3309 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 3310 | * |
58877d34 PZ |
3311 | * Relies on p->pi_lock stabilizing: |
3312 | * - p->sched_class | |
3313 | * - p->cpus_ptr | |
3314 | * - p->sched_task_group | |
3315 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
3316 | * | |
3317 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
3318 | * Takes rq->lock in: | |
3319 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
3320 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
3321 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
3322 | * | |
3323 | * As a consequence we race really badly with just about everything. See the | |
3324 | * many memory barriers and their comments for details. | |
7696f991 | 3325 | * |
a2250238 PZ |
3326 | * Return: %true if @p->state changes (an actual wakeup was done), |
3327 | * %false otherwise. | |
1da177e4 | 3328 | */ |
e4a52bcb PZ |
3329 | static int |
3330 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 3331 | { |
1da177e4 | 3332 | unsigned long flags; |
c05fbafb | 3333 | int cpu, success = 0; |
2398f2c6 | 3334 | |
e3d85487 | 3335 | preempt_disable(); |
aacedf26 PZ |
3336 | if (p == current) { |
3337 | /* | |
3338 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
3339 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 3340 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
3341 | * without taking any locks. |
3342 | * | |
3343 | * In particular: | |
3344 | * - we rely on Program-Order guarantees for all the ordering, | |
3345 | * - we're serialized against set_special_state() by virtue of | |
3346 | * it disabling IRQs (this allows not taking ->pi_lock). | |
3347 | */ | |
3348 | if (!(p->state & state)) | |
e3d85487 | 3349 | goto out; |
aacedf26 PZ |
3350 | |
3351 | success = 1; | |
aacedf26 PZ |
3352 | trace_sched_waking(p); |
3353 | p->state = TASK_RUNNING; | |
3354 | trace_sched_wakeup(p); | |
3355 | goto out; | |
3356 | } | |
3357 | ||
e0acd0a6 ON |
3358 | /* |
3359 | * If we are going to wake up a thread waiting for CONDITION we | |
3360 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
3361 | * reordered with p->state check below. This pairs with smp_store_mb() |
3362 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 3363 | */ |
013fdb80 | 3364 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 3365 | smp_mb__after_spinlock(); |
e9c84311 | 3366 | if (!(p->state & state)) |
aacedf26 | 3367 | goto unlock; |
1da177e4 | 3368 | |
fbd705a0 PZ |
3369 | trace_sched_waking(p); |
3370 | ||
d1ccc66d IM |
3371 | /* We're going to change ->state: */ |
3372 | success = 1; | |
1da177e4 | 3373 | |
135e8c92 BS |
3374 | /* |
3375 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
3376 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
3377 | * in smp_cond_load_acquire() below. | |
3378 | * | |
3d85b270 AP |
3379 | * sched_ttwu_pending() try_to_wake_up() |
3380 | * STORE p->on_rq = 1 LOAD p->state | |
3381 | * UNLOCK rq->lock | |
3382 | * | |
3383 | * __schedule() (switch to task 'p') | |
3384 | * LOCK rq->lock smp_rmb(); | |
3385 | * smp_mb__after_spinlock(); | |
3386 | * UNLOCK rq->lock | |
135e8c92 BS |
3387 | * |
3388 | * [task p] | |
3d85b270 | 3389 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 3390 | * |
3d85b270 AP |
3391 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3392 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
3393 | * |
3394 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
3395 | */ |
3396 | smp_rmb(); | |
58877d34 | 3397 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 3398 | goto unlock; |
1da177e4 | 3399 | |
1da177e4 | 3400 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
3401 | /* |
3402 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
3403 | * possible to, falsely, observe p->on_cpu == 0. | |
3404 | * | |
3405 | * One must be running (->on_cpu == 1) in order to remove oneself | |
3406 | * from the runqueue. | |
3407 | * | |
3d85b270 AP |
3408 | * __schedule() (switch to task 'p') try_to_wake_up() |
3409 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
3410 | * UNLOCK rq->lock | |
3411 | * | |
3412 | * __schedule() (put 'p' to sleep) | |
3413 | * LOCK rq->lock smp_rmb(); | |
3414 | * smp_mb__after_spinlock(); | |
3415 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 3416 | * |
3d85b270 AP |
3417 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3418 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
3419 | * |
3420 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
3421 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
3422 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 3423 | */ |
dbfb089d PZ |
3424 | smp_acquire__after_ctrl_dep(); |
3425 | ||
3426 | /* | |
3427 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
3428 | * == 0), which means we need to do an enqueue, change p->state to | |
3429 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
3430 | * enqueue, such as ttwu_queue_wakelist(). | |
3431 | */ | |
3432 | p->state = TASK_WAKING; | |
ecf7d01c | 3433 | |
c6e7bd7a PZ |
3434 | /* |
3435 | * If the owning (remote) CPU is still in the middle of schedule() with | |
3436 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
3437 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
3438 | * let the waker make forward progress. This is safe because IRQs are | |
3439 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
3440 | * |
3441 | * Ensure we load task_cpu(p) after p->on_cpu: | |
3442 | * | |
3443 | * set_task_cpu(p, cpu); | |
3444 | * STORE p->cpu = @cpu | |
3445 | * __schedule() (switch to task 'p') | |
3446 | * LOCK rq->lock | |
3447 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
3448 | * STORE p->on_cpu = 1 LOAD p->cpu | |
3449 | * | |
3450 | * to ensure we observe the correct CPU on which the task is currently | |
3451 | * scheduling. | |
c6e7bd7a | 3452 | */ |
b6e13e85 | 3453 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 3454 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
3455 | goto unlock; |
3456 | ||
e9c84311 | 3457 | /* |
d1ccc66d | 3458 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 3459 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 3460 | * |
31cb1bc0 | 3461 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
3462 | * |
3463 | * This ensures that tasks getting woken will be fully ordered against | |
3464 | * their previous state and preserve Program Order. | |
0970d299 | 3465 | */ |
1f03e8d2 | 3466 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 3467 | |
3aef1551 | 3468 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc | 3469 | if (task_cpu(p) != cpu) { |
ec618b84 PZ |
3470 | if (p->in_iowait) { |
3471 | delayacct_blkio_end(p); | |
3472 | atomic_dec(&task_rq(p)->nr_iowait); | |
3473 | } | |
3474 | ||
f339b9dc | 3475 | wake_flags |= WF_MIGRATED; |
eb414681 | 3476 | psi_ttwu_dequeue(p); |
e4a52bcb | 3477 | set_task_cpu(p, cpu); |
f339b9dc | 3478 | } |
b6e13e85 PZ |
3479 | #else |
3480 | cpu = task_cpu(p); | |
1da177e4 | 3481 | #endif /* CONFIG_SMP */ |
1da177e4 | 3482 | |
b5179ac7 | 3483 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 3484 | unlock: |
013fdb80 | 3485 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
3486 | out: |
3487 | if (success) | |
b6e13e85 | 3488 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 3489 | preempt_enable(); |
1da177e4 LT |
3490 | |
3491 | return success; | |
3492 | } | |
3493 | ||
2beaf328 PM |
3494 | /** |
3495 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
1b7af295 | 3496 | * @p: Process for which the function is to be invoked, can be @current. |
2beaf328 PM |
3497 | * @func: Function to invoke. |
3498 | * @arg: Argument to function. | |
3499 | * | |
3500 | * If the specified task can be quickly locked into a definite state | |
3501 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
3502 | * state while invoking @func(@arg). This function can use ->on_rq and | |
3503 | * task_curr() to work out what the state is, if required. Given that | |
3504 | * @func can be invoked with a runqueue lock held, it had better be quite | |
3505 | * lightweight. | |
3506 | * | |
3507 | * Returns: | |
3508 | * @false if the task slipped out from under the locks. | |
3509 | * @true if the task was locked onto a runqueue or is sleeping. | |
3510 | * However, @func can override this by returning @false. | |
3511 | */ | |
3512 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
3513 | { | |
2beaf328 | 3514 | struct rq_flags rf; |
1b7af295 | 3515 | bool ret = false; |
2beaf328 PM |
3516 | struct rq *rq; |
3517 | ||
1b7af295 | 3518 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
2beaf328 PM |
3519 | if (p->on_rq) { |
3520 | rq = __task_rq_lock(p, &rf); | |
3521 | if (task_rq(p) == rq) | |
3522 | ret = func(p, arg); | |
3523 | rq_unlock(rq, &rf); | |
3524 | } else { | |
3525 | switch (p->state) { | |
3526 | case TASK_RUNNING: | |
3527 | case TASK_WAKING: | |
3528 | break; | |
3529 | default: | |
3530 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
3531 | if (!p->on_rq) | |
3532 | ret = func(p, arg); | |
3533 | } | |
3534 | } | |
1b7af295 | 3535 | raw_spin_unlock_irqrestore(&p->pi_lock, rf.flags); |
2beaf328 PM |
3536 | return ret; |
3537 | } | |
3538 | ||
50fa610a DH |
3539 | /** |
3540 | * wake_up_process - Wake up a specific process | |
3541 | * @p: The process to be woken up. | |
3542 | * | |
3543 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
3544 | * processes. |
3545 | * | |
3546 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 3547 | * |
7696f991 | 3548 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 3549 | */ |
7ad5b3a5 | 3550 | int wake_up_process(struct task_struct *p) |
1da177e4 | 3551 | { |
9067ac85 | 3552 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 3553 | } |
1da177e4 LT |
3554 | EXPORT_SYMBOL(wake_up_process); |
3555 | ||
7ad5b3a5 | 3556 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
3557 | { |
3558 | return try_to_wake_up(p, state, 0); | |
3559 | } | |
3560 | ||
1da177e4 LT |
3561 | /* |
3562 | * Perform scheduler related setup for a newly forked process p. | |
3563 | * p is forked by current. | |
dd41f596 IM |
3564 | * |
3565 | * __sched_fork() is basic setup used by init_idle() too: | |
3566 | */ | |
5e1576ed | 3567 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3568 | { |
fd2f4419 PZ |
3569 | p->on_rq = 0; |
3570 | ||
3571 | p->se.on_rq = 0; | |
dd41f596 IM |
3572 | p->se.exec_start = 0; |
3573 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 3574 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 3575 | p->se.nr_migrations = 0; |
da7a735e | 3576 | p->se.vruntime = 0; |
fd2f4419 | 3577 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 3578 | |
ad936d86 BP |
3579 | #ifdef CONFIG_FAIR_GROUP_SCHED |
3580 | p->se.cfs_rq = NULL; | |
3581 | #endif | |
3582 | ||
6cfb0d5d | 3583 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 3584 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 3585 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 3586 | #endif |
476d139c | 3587 | |
aab03e05 | 3588 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 3589 | init_dl_task_timer(&p->dl); |
209a0cbd | 3590 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 3591 | __dl_clear_params(p); |
aab03e05 | 3592 | |
fa717060 | 3593 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
3594 | p->rt.timeout = 0; |
3595 | p->rt.time_slice = sched_rr_timeslice; | |
3596 | p->rt.on_rq = 0; | |
3597 | p->rt.on_list = 0; | |
476d139c | 3598 | |
e107be36 AK |
3599 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3600 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
3601 | #endif | |
cbee9f88 | 3602 | |
5e1f0f09 MG |
3603 | #ifdef CONFIG_COMPACTION |
3604 | p->capture_control = NULL; | |
3605 | #endif | |
13784475 | 3606 | init_numa_balancing(clone_flags, p); |
a1488664 | 3607 | #ifdef CONFIG_SMP |
8c4890d1 | 3608 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 3609 | p->migration_pending = NULL; |
a1488664 | 3610 | #endif |
dd41f596 IM |
3611 | } |
3612 | ||
2a595721 SD |
3613 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
3614 | ||
1a687c2e | 3615 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 3616 | |
1a687c2e MG |
3617 | void set_numabalancing_state(bool enabled) |
3618 | { | |
3619 | if (enabled) | |
2a595721 | 3620 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 3621 | else |
2a595721 | 3622 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 3623 | } |
54a43d54 AK |
3624 | |
3625 | #ifdef CONFIG_PROC_SYSCTL | |
3626 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 3627 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
3628 | { |
3629 | struct ctl_table t; | |
3630 | int err; | |
2a595721 | 3631 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
3632 | |
3633 | if (write && !capable(CAP_SYS_ADMIN)) | |
3634 | return -EPERM; | |
3635 | ||
3636 | t = *table; | |
3637 | t.data = &state; | |
3638 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3639 | if (err < 0) | |
3640 | return err; | |
3641 | if (write) | |
3642 | set_numabalancing_state(state); | |
3643 | return err; | |
3644 | } | |
3645 | #endif | |
3646 | #endif | |
dd41f596 | 3647 | |
4698f88c JP |
3648 | #ifdef CONFIG_SCHEDSTATS |
3649 | ||
cb251765 | 3650 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 3651 | static bool __initdata __sched_schedstats = false; |
cb251765 | 3652 | |
cb251765 MG |
3653 | static void set_schedstats(bool enabled) |
3654 | { | |
3655 | if (enabled) | |
3656 | static_branch_enable(&sched_schedstats); | |
3657 | else | |
3658 | static_branch_disable(&sched_schedstats); | |
3659 | } | |
3660 | ||
3661 | void force_schedstat_enabled(void) | |
3662 | { | |
3663 | if (!schedstat_enabled()) { | |
3664 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
3665 | static_branch_enable(&sched_schedstats); | |
3666 | } | |
3667 | } | |
3668 | ||
3669 | static int __init setup_schedstats(char *str) | |
3670 | { | |
3671 | int ret = 0; | |
3672 | if (!str) | |
3673 | goto out; | |
3674 | ||
4698f88c JP |
3675 | /* |
3676 | * This code is called before jump labels have been set up, so we can't | |
3677 | * change the static branch directly just yet. Instead set a temporary | |
3678 | * variable so init_schedstats() can do it later. | |
3679 | */ | |
cb251765 | 3680 | if (!strcmp(str, "enable")) { |
4698f88c | 3681 | __sched_schedstats = true; |
cb251765 MG |
3682 | ret = 1; |
3683 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 3684 | __sched_schedstats = false; |
cb251765 MG |
3685 | ret = 1; |
3686 | } | |
3687 | out: | |
3688 | if (!ret) | |
3689 | pr_warn("Unable to parse schedstats=\n"); | |
3690 | ||
3691 | return ret; | |
3692 | } | |
3693 | __setup("schedstats=", setup_schedstats); | |
3694 | ||
4698f88c JP |
3695 | static void __init init_schedstats(void) |
3696 | { | |
3697 | set_schedstats(__sched_schedstats); | |
3698 | } | |
3699 | ||
cb251765 | 3700 | #ifdef CONFIG_PROC_SYSCTL |
32927393 CH |
3701 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
3702 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
3703 | { |
3704 | struct ctl_table t; | |
3705 | int err; | |
3706 | int state = static_branch_likely(&sched_schedstats); | |
3707 | ||
3708 | if (write && !capable(CAP_SYS_ADMIN)) | |
3709 | return -EPERM; | |
3710 | ||
3711 | t = *table; | |
3712 | t.data = &state; | |
3713 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3714 | if (err < 0) | |
3715 | return err; | |
3716 | if (write) | |
3717 | set_schedstats(state); | |
3718 | return err; | |
3719 | } | |
4698f88c JP |
3720 | #endif /* CONFIG_PROC_SYSCTL */ |
3721 | #else /* !CONFIG_SCHEDSTATS */ | |
3722 | static inline void init_schedstats(void) {} | |
3723 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
3724 | |
3725 | /* | |
3726 | * fork()/clone()-time setup: | |
3727 | */ | |
aab03e05 | 3728 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3729 | { |
0122ec5b | 3730 | unsigned long flags; |
dd41f596 | 3731 | |
5e1576ed | 3732 | __sched_fork(clone_flags, p); |
06b83b5f | 3733 | /* |
7dc603c9 | 3734 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
3735 | * nobody will actually run it, and a signal or other external |
3736 | * event cannot wake it up and insert it on the runqueue either. | |
3737 | */ | |
7dc603c9 | 3738 | p->state = TASK_NEW; |
dd41f596 | 3739 | |
c350a04e MG |
3740 | /* |
3741 | * Make sure we do not leak PI boosting priority to the child. | |
3742 | */ | |
3743 | p->prio = current->normal_prio; | |
3744 | ||
e8f14172 PB |
3745 | uclamp_fork(p); |
3746 | ||
b9dc29e7 MG |
3747 | /* |
3748 | * Revert to default priority/policy on fork if requested. | |
3749 | */ | |
3750 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 3751 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 3752 | p->policy = SCHED_NORMAL; |
6c697bdf | 3753 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
3754 | p->rt_priority = 0; |
3755 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
3756 | p->static_prio = NICE_TO_PRIO(0); | |
3757 | ||
3758 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 3759 | set_load_weight(p, false); |
6c697bdf | 3760 | |
b9dc29e7 MG |
3761 | /* |
3762 | * We don't need the reset flag anymore after the fork. It has | |
3763 | * fulfilled its duty: | |
3764 | */ | |
3765 | p->sched_reset_on_fork = 0; | |
3766 | } | |
ca94c442 | 3767 | |
af0fffd9 | 3768 | if (dl_prio(p->prio)) |
aab03e05 | 3769 | return -EAGAIN; |
af0fffd9 | 3770 | else if (rt_prio(p->prio)) |
aab03e05 | 3771 | p->sched_class = &rt_sched_class; |
af0fffd9 | 3772 | else |
2ddbf952 | 3773 | p->sched_class = &fair_sched_class; |
b29739f9 | 3774 | |
7dc603c9 | 3775 | init_entity_runnable_average(&p->se); |
cd29fe6f | 3776 | |
86951599 PZ |
3777 | /* |
3778 | * The child is not yet in the pid-hash so no cgroup attach races, | |
3779 | * and the cgroup is pinned to this child due to cgroup_fork() | |
3780 | * is ran before sched_fork(). | |
3781 | * | |
3782 | * Silence PROVE_RCU. | |
3783 | */ | |
0122ec5b | 3784 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ce3614da | 3785 | rseq_migrate(p); |
e210bffd | 3786 | /* |
d1ccc66d | 3787 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
3788 | * so use __set_task_cpu(). |
3789 | */ | |
af0fffd9 | 3790 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
3791 | if (p->sched_class->task_fork) |
3792 | p->sched_class->task_fork(p); | |
0122ec5b | 3793 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 3794 | |
f6db8347 | 3795 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 3796 | if (likely(sched_info_on())) |
52f17b6c | 3797 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 3798 | #endif |
3ca7a440 PZ |
3799 | #if defined(CONFIG_SMP) |
3800 | p->on_cpu = 0; | |
4866cde0 | 3801 | #endif |
01028747 | 3802 | init_task_preempt_count(p); |
806c09a7 | 3803 | #ifdef CONFIG_SMP |
917b627d | 3804 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 3805 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 3806 | #endif |
aab03e05 | 3807 | return 0; |
1da177e4 LT |
3808 | } |
3809 | ||
13685c4a QY |
3810 | void sched_post_fork(struct task_struct *p) |
3811 | { | |
3812 | uclamp_post_fork(p); | |
3813 | } | |
3814 | ||
332ac17e DF |
3815 | unsigned long to_ratio(u64 period, u64 runtime) |
3816 | { | |
3817 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 3818 | return BW_UNIT; |
332ac17e DF |
3819 | |
3820 | /* | |
3821 | * Doing this here saves a lot of checks in all | |
3822 | * the calling paths, and returning zero seems | |
3823 | * safe for them anyway. | |
3824 | */ | |
3825 | if (period == 0) | |
3826 | return 0; | |
3827 | ||
c52f14d3 | 3828 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
3829 | } |
3830 | ||
1da177e4 LT |
3831 | /* |
3832 | * wake_up_new_task - wake up a newly created task for the first time. | |
3833 | * | |
3834 | * This function will do some initial scheduler statistics housekeeping | |
3835 | * that must be done for every newly created context, then puts the task | |
3836 | * on the runqueue and wakes it. | |
3837 | */ | |
3e51e3ed | 3838 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 3839 | { |
eb580751 | 3840 | struct rq_flags rf; |
dd41f596 | 3841 | struct rq *rq; |
fabf318e | 3842 | |
eb580751 | 3843 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 3844 | p->state = TASK_RUNNING; |
fabf318e PZ |
3845 | #ifdef CONFIG_SMP |
3846 | /* | |
3847 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 3848 | * - cpus_ptr can change in the fork path |
d1ccc66d | 3849 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
3850 | * |
3851 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
3852 | * as we're not fully set-up yet. | |
fabf318e | 3853 | */ |
32e839dd | 3854 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 3855 | rseq_migrate(p); |
3aef1551 | 3856 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 3857 | #endif |
b7fa30c9 | 3858 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 3859 | update_rq_clock(rq); |
d0fe0b9c | 3860 | post_init_entity_util_avg(p); |
0017d735 | 3861 | |
7a57f32a | 3862 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 3863 | trace_sched_wakeup_new(p); |
a7558e01 | 3864 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3865 | #ifdef CONFIG_SMP |
0aaafaab PZ |
3866 | if (p->sched_class->task_woken) { |
3867 | /* | |
b19a888c | 3868 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
3869 | * drop it. |
3870 | */ | |
d8ac8971 | 3871 | rq_unpin_lock(rq, &rf); |
efbbd05a | 3872 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3873 | rq_repin_lock(rq, &rf); |
0aaafaab | 3874 | } |
9a897c5a | 3875 | #endif |
eb580751 | 3876 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
3877 | } |
3878 | ||
e107be36 AK |
3879 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3880 | ||
b7203428 | 3881 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 3882 | |
2ecd9d29 PZ |
3883 | void preempt_notifier_inc(void) |
3884 | { | |
b7203428 | 3885 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
3886 | } |
3887 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
3888 | ||
3889 | void preempt_notifier_dec(void) | |
3890 | { | |
b7203428 | 3891 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
3892 | } |
3893 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
3894 | ||
e107be36 | 3895 | /** |
80dd99b3 | 3896 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3897 | * @notifier: notifier struct to register |
e107be36 AK |
3898 | */ |
3899 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3900 | { | |
b7203428 | 3901 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
3902 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
3903 | ||
e107be36 AK |
3904 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
3905 | } | |
3906 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3907 | ||
3908 | /** | |
3909 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3910 | * @notifier: notifier struct to unregister |
e107be36 | 3911 | * |
d84525a8 | 3912 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
3913 | */ |
3914 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3915 | { | |
3916 | hlist_del(¬ifier->link); | |
3917 | } | |
3918 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3919 | ||
1cde2930 | 3920 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3921 | { |
3922 | struct preempt_notifier *notifier; | |
e107be36 | 3923 | |
b67bfe0d | 3924 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3925 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
3926 | } | |
3927 | ||
1cde2930 PZ |
3928 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
3929 | { | |
b7203428 | 3930 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3931 | __fire_sched_in_preempt_notifiers(curr); |
3932 | } | |
3933 | ||
e107be36 | 3934 | static void |
1cde2930 PZ |
3935 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3936 | struct task_struct *next) | |
e107be36 AK |
3937 | { |
3938 | struct preempt_notifier *notifier; | |
e107be36 | 3939 | |
b67bfe0d | 3940 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3941 | notifier->ops->sched_out(notifier, next); |
3942 | } | |
3943 | ||
1cde2930 PZ |
3944 | static __always_inline void |
3945 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3946 | struct task_struct *next) | |
3947 | { | |
b7203428 | 3948 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3949 | __fire_sched_out_preempt_notifiers(curr, next); |
3950 | } | |
3951 | ||
6d6bc0ad | 3952 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3953 | |
1cde2930 | 3954 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3955 | { |
3956 | } | |
3957 | ||
1cde2930 | 3958 | static inline void |
e107be36 AK |
3959 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3960 | struct task_struct *next) | |
3961 | { | |
3962 | } | |
3963 | ||
6d6bc0ad | 3964 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3965 | |
31cb1bc0 | 3966 | static inline void prepare_task(struct task_struct *next) |
3967 | { | |
3968 | #ifdef CONFIG_SMP | |
3969 | /* | |
3970 | * Claim the task as running, we do this before switching to it | |
3971 | * such that any running task will have this set. | |
58877d34 PZ |
3972 | * |
3973 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 3974 | */ |
58877d34 | 3975 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 3976 | #endif |
3977 | } | |
3978 | ||
3979 | static inline void finish_task(struct task_struct *prev) | |
3980 | { | |
3981 | #ifdef CONFIG_SMP | |
3982 | /* | |
58877d34 PZ |
3983 | * This must be the very last reference to @prev from this CPU. After |
3984 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
3985 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 3986 | * finished. |
3987 | * | |
3988 | * In particular, the load of prev->state in finish_task_switch() must | |
3989 | * happen before this. | |
3990 | * | |
3991 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
3992 | */ | |
3993 | smp_store_release(&prev->on_cpu, 0); | |
3994 | #endif | |
3995 | } | |
3996 | ||
565790d2 PZ |
3997 | #ifdef CONFIG_SMP |
3998 | ||
3999 | static void do_balance_callbacks(struct rq *rq, struct callback_head *head) | |
4000 | { | |
4001 | void (*func)(struct rq *rq); | |
4002 | struct callback_head *next; | |
4003 | ||
4004 | lockdep_assert_held(&rq->lock); | |
4005 | ||
4006 | while (head) { | |
4007 | func = (void (*)(struct rq *))head->func; | |
4008 | next = head->next; | |
4009 | head->next = NULL; | |
4010 | head = next; | |
4011 | ||
4012 | func(rq); | |
4013 | } | |
4014 | } | |
4015 | ||
ae792702 PZ |
4016 | static void balance_push(struct rq *rq); |
4017 | ||
4018 | struct callback_head balance_push_callback = { | |
4019 | .next = NULL, | |
4020 | .func = (void (*)(struct callback_head *))balance_push, | |
4021 | }; | |
4022 | ||
565790d2 PZ |
4023 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) |
4024 | { | |
4025 | struct callback_head *head = rq->balance_callback; | |
4026 | ||
4027 | lockdep_assert_held(&rq->lock); | |
ae792702 | 4028 | if (head) |
565790d2 PZ |
4029 | rq->balance_callback = NULL; |
4030 | ||
4031 | return head; | |
4032 | } | |
4033 | ||
4034 | static void __balance_callbacks(struct rq *rq) | |
4035 | { | |
4036 | do_balance_callbacks(rq, splice_balance_callbacks(rq)); | |
4037 | } | |
4038 | ||
4039 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4040 | { | |
4041 | unsigned long flags; | |
4042 | ||
4043 | if (unlikely(head)) { | |
4044 | raw_spin_lock_irqsave(&rq->lock, flags); | |
4045 | do_balance_callbacks(rq, head); | |
4046 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
4047 | } | |
4048 | } | |
4049 | ||
4050 | #else | |
4051 | ||
4052 | static inline void __balance_callbacks(struct rq *rq) | |
4053 | { | |
4054 | } | |
4055 | ||
4056 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
4057 | { | |
4058 | return NULL; | |
4059 | } | |
4060 | ||
4061 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4062 | { | |
4063 | } | |
4064 | ||
4065 | #endif | |
4066 | ||
269d5992 PZ |
4067 | static inline void |
4068 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4069 | { |
269d5992 PZ |
4070 | /* |
4071 | * Since the runqueue lock will be released by the next | |
4072 | * task (which is an invalid locking op but in the case | |
4073 | * of the scheduler it's an obvious special-case), so we | |
4074 | * do an early lockdep release here: | |
4075 | */ | |
4076 | rq_unpin_lock(rq, rf); | |
5facae4f | 4077 | spin_release(&rq->lock.dep_map, _THIS_IP_); |
31cb1bc0 | 4078 | #ifdef CONFIG_DEBUG_SPINLOCK |
4079 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 4080 | rq->lock.owner = next; |
31cb1bc0 | 4081 | #endif |
269d5992 PZ |
4082 | } |
4083 | ||
4084 | static inline void finish_lock_switch(struct rq *rq) | |
4085 | { | |
31cb1bc0 | 4086 | /* |
4087 | * If we are tracking spinlock dependencies then we have to | |
4088 | * fix up the runqueue lock - which gets 'carried over' from | |
4089 | * prev into current: | |
4090 | */ | |
4091 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
ae792702 | 4092 | __balance_callbacks(rq); |
31cb1bc0 | 4093 | raw_spin_unlock_irq(&rq->lock); |
4094 | } | |
4095 | ||
325ea10c IM |
4096 | /* |
4097 | * NOP if the arch has not defined these: | |
4098 | */ | |
4099 | ||
4100 | #ifndef prepare_arch_switch | |
4101 | # define prepare_arch_switch(next) do { } while (0) | |
4102 | #endif | |
4103 | ||
4104 | #ifndef finish_arch_post_lock_switch | |
4105 | # define finish_arch_post_lock_switch() do { } while (0) | |
4106 | #endif | |
4107 | ||
5fbda3ec TG |
4108 | static inline void kmap_local_sched_out(void) |
4109 | { | |
4110 | #ifdef CONFIG_KMAP_LOCAL | |
4111 | if (unlikely(current->kmap_ctrl.idx)) | |
4112 | __kmap_local_sched_out(); | |
4113 | #endif | |
4114 | } | |
4115 | ||
4116 | static inline void kmap_local_sched_in(void) | |
4117 | { | |
4118 | #ifdef CONFIG_KMAP_LOCAL | |
4119 | if (unlikely(current->kmap_ctrl.idx)) | |
4120 | __kmap_local_sched_in(); | |
4121 | #endif | |
4122 | } | |
4123 | ||
4866cde0 NP |
4124 | /** |
4125 | * prepare_task_switch - prepare to switch tasks | |
4126 | * @rq: the runqueue preparing to switch | |
421cee29 | 4127 | * @prev: the current task that is being switched out |
4866cde0 NP |
4128 | * @next: the task we are going to switch to. |
4129 | * | |
4130 | * This is called with the rq lock held and interrupts off. It must | |
4131 | * be paired with a subsequent finish_task_switch after the context | |
4132 | * switch. | |
4133 | * | |
4134 | * prepare_task_switch sets up locking and calls architecture specific | |
4135 | * hooks. | |
4136 | */ | |
e107be36 AK |
4137 | static inline void |
4138 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
4139 | struct task_struct *next) | |
4866cde0 | 4140 | { |
0ed557aa | 4141 | kcov_prepare_switch(prev); |
43148951 | 4142 | sched_info_switch(rq, prev, next); |
fe4b04fa | 4143 | perf_event_task_sched_out(prev, next); |
d7822b1e | 4144 | rseq_preempt(prev); |
e107be36 | 4145 | fire_sched_out_preempt_notifiers(prev, next); |
5fbda3ec | 4146 | kmap_local_sched_out(); |
31cb1bc0 | 4147 | prepare_task(next); |
4866cde0 NP |
4148 | prepare_arch_switch(next); |
4149 | } | |
4150 | ||
1da177e4 LT |
4151 | /** |
4152 | * finish_task_switch - clean up after a task-switch | |
4153 | * @prev: the thread we just switched away from. | |
4154 | * | |
4866cde0 NP |
4155 | * finish_task_switch must be called after the context switch, paired |
4156 | * with a prepare_task_switch call before the context switch. | |
4157 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
4158 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
4159 | * |
4160 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 4161 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
4162 | * with the lock held can cause deadlocks; see schedule() for |
4163 | * details.) | |
dfa50b60 ON |
4164 | * |
4165 | * The context switch have flipped the stack from under us and restored the | |
4166 | * local variables which were saved when this task called schedule() in the | |
4167 | * past. prev == current is still correct but we need to recalculate this_rq | |
4168 | * because prev may have moved to another CPU. | |
1da177e4 | 4169 | */ |
dfa50b60 | 4170 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
4171 | __releases(rq->lock) |
4172 | { | |
dfa50b60 | 4173 | struct rq *rq = this_rq(); |
1da177e4 | 4174 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 4175 | long prev_state; |
1da177e4 | 4176 | |
609ca066 PZ |
4177 | /* |
4178 | * The previous task will have left us with a preempt_count of 2 | |
4179 | * because it left us after: | |
4180 | * | |
4181 | * schedule() | |
4182 | * preempt_disable(); // 1 | |
4183 | * __schedule() | |
4184 | * raw_spin_lock_irq(&rq->lock) // 2 | |
4185 | * | |
4186 | * Also, see FORK_PREEMPT_COUNT. | |
4187 | */ | |
e2bf1c4b PZ |
4188 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
4189 | "corrupted preempt_count: %s/%d/0x%x\n", | |
4190 | current->comm, current->pid, preempt_count())) | |
4191 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 4192 | |
1da177e4 LT |
4193 | rq->prev_mm = NULL; |
4194 | ||
4195 | /* | |
4196 | * A task struct has one reference for the use as "current". | |
c394cc9f | 4197 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
4198 | * schedule one last time. The schedule call will never return, and |
4199 | * the scheduled task must drop that reference. | |
95913d97 PZ |
4200 | * |
4201 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 4202 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
4203 | * running on another CPU and we could rave with its RUNNING -> DEAD |
4204 | * transition, resulting in a double drop. | |
1da177e4 | 4205 | */ |
55a101f8 | 4206 | prev_state = prev->state; |
bf9fae9f | 4207 | vtime_task_switch(prev); |
a8d757ef | 4208 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 4209 | finish_task(prev); |
4210 | finish_lock_switch(rq); | |
01f23e16 | 4211 | finish_arch_post_lock_switch(); |
0ed557aa | 4212 | kcov_finish_switch(current); |
5fbda3ec TG |
4213 | /* |
4214 | * kmap_local_sched_out() is invoked with rq::lock held and | |
4215 | * interrupts disabled. There is no requirement for that, but the | |
4216 | * sched out code does not have an interrupt enabled section. | |
4217 | * Restoring the maps on sched in does not require interrupts being | |
4218 | * disabled either. | |
4219 | */ | |
4220 | kmap_local_sched_in(); | |
e8fa1362 | 4221 | |
e107be36 | 4222 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 4223 | /* |
70216e18 MD |
4224 | * When switching through a kernel thread, the loop in |
4225 | * membarrier_{private,global}_expedited() may have observed that | |
4226 | * kernel thread and not issued an IPI. It is therefore possible to | |
4227 | * schedule between user->kernel->user threads without passing though | |
4228 | * switch_mm(). Membarrier requires a barrier after storing to | |
4229 | * rq->curr, before returning to userspace, so provide them here: | |
4230 | * | |
4231 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
4232 | * provided by mmdrop(), | |
4233 | * - a sync_core for SYNC_CORE. | |
306e0604 | 4234 | */ |
70216e18 MD |
4235 | if (mm) { |
4236 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 4237 | mmdrop(mm); |
70216e18 | 4238 | } |
1cef1150 PZ |
4239 | if (unlikely(prev_state == TASK_DEAD)) { |
4240 | if (prev->sched_class->task_dead) | |
4241 | prev->sched_class->task_dead(prev); | |
68f24b08 | 4242 | |
1cef1150 PZ |
4243 | /* |
4244 | * Remove function-return probe instances associated with this | |
4245 | * task and put them back on the free list. | |
4246 | */ | |
4247 | kprobe_flush_task(prev); | |
4248 | ||
4249 | /* Task is done with its stack. */ | |
4250 | put_task_stack(prev); | |
4251 | ||
0ff7b2cf | 4252 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 4253 | } |
99e5ada9 | 4254 | |
de734f89 | 4255 | tick_nohz_task_switch(); |
dfa50b60 | 4256 | return rq; |
1da177e4 LT |
4257 | } |
4258 | ||
4259 | /** | |
4260 | * schedule_tail - first thing a freshly forked thread must call. | |
4261 | * @prev: the thread we just switched away from. | |
4262 | */ | |
722a9f92 | 4263 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
4264 | __releases(rq->lock) |
4265 | { | |
1a43a14a | 4266 | struct rq *rq; |
da19ab51 | 4267 | |
609ca066 PZ |
4268 | /* |
4269 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
4270 | * finish_task_switch() for details. | |
4271 | * | |
4272 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
4273 | * and the preempt_enable() will end up enabling preemption (on | |
4274 | * PREEMPT_COUNT kernels). | |
4275 | */ | |
4276 | ||
dfa50b60 | 4277 | rq = finish_task_switch(prev); |
1a43a14a | 4278 | preempt_enable(); |
70b97a7f | 4279 | |
1da177e4 | 4280 | if (current->set_child_tid) |
b488893a | 4281 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
4282 | |
4283 | calculate_sigpending(); | |
1da177e4 LT |
4284 | } |
4285 | ||
4286 | /* | |
dfa50b60 | 4287 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 4288 | */ |
04936948 | 4289 | static __always_inline struct rq * |
70b97a7f | 4290 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 4291 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 4292 | { |
e107be36 | 4293 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 4294 | |
9226d125 ZA |
4295 | /* |
4296 | * For paravirt, this is coupled with an exit in switch_to to | |
4297 | * combine the page table reload and the switch backend into | |
4298 | * one hypercall. | |
4299 | */ | |
224101ed | 4300 | arch_start_context_switch(prev); |
9226d125 | 4301 | |
306e0604 | 4302 | /* |
139d025c PZ |
4303 | * kernel -> kernel lazy + transfer active |
4304 | * user -> kernel lazy + mmgrab() active | |
4305 | * | |
4306 | * kernel -> user switch + mmdrop() active | |
4307 | * user -> user switch | |
306e0604 | 4308 | */ |
139d025c PZ |
4309 | if (!next->mm) { // to kernel |
4310 | enter_lazy_tlb(prev->active_mm, next); | |
4311 | ||
4312 | next->active_mm = prev->active_mm; | |
4313 | if (prev->mm) // from user | |
4314 | mmgrab(prev->active_mm); | |
4315 | else | |
4316 | prev->active_mm = NULL; | |
4317 | } else { // to user | |
227a4aad | 4318 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
4319 | /* |
4320 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 4321 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
4322 | * |
4323 | * The below provides this either through switch_mm(), or in | |
4324 | * case 'prev->active_mm == next->mm' through | |
4325 | * finish_task_switch()'s mmdrop(). | |
4326 | */ | |
139d025c | 4327 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 4328 | |
139d025c PZ |
4329 | if (!prev->mm) { // from kernel |
4330 | /* will mmdrop() in finish_task_switch(). */ | |
4331 | rq->prev_mm = prev->active_mm; | |
4332 | prev->active_mm = NULL; | |
4333 | } | |
1da177e4 | 4334 | } |
92509b73 | 4335 | |
cb42c9a3 | 4336 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 4337 | |
269d5992 | 4338 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
4339 | |
4340 | /* Here we just switch the register state and the stack. */ | |
4341 | switch_to(prev, next, prev); | |
dd41f596 | 4342 | barrier(); |
dfa50b60 ON |
4343 | |
4344 | return finish_task_switch(prev); | |
1da177e4 LT |
4345 | } |
4346 | ||
4347 | /* | |
1c3e8264 | 4348 | * nr_running and nr_context_switches: |
1da177e4 LT |
4349 | * |
4350 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 4351 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
4352 | */ |
4353 | unsigned long nr_running(void) | |
4354 | { | |
4355 | unsigned long i, sum = 0; | |
4356 | ||
4357 | for_each_online_cpu(i) | |
4358 | sum += cpu_rq(i)->nr_running; | |
4359 | ||
4360 | return sum; | |
f711f609 | 4361 | } |
1da177e4 | 4362 | |
2ee507c4 | 4363 | /* |
d1ccc66d | 4364 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
4365 | * |
4366 | * Caution: this function does not check that the caller has disabled | |
4367 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
4368 | * race. The caller is responsible to use it correctly, for example: | |
4369 | * | |
dfcb245e | 4370 | * - from a non-preemptible section (of course) |
00cc1633 DD |
4371 | * |
4372 | * - from a thread that is bound to a single CPU | |
4373 | * | |
4374 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
4375 | */ |
4376 | bool single_task_running(void) | |
4377 | { | |
00cc1633 | 4378 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
4379 | } |
4380 | EXPORT_SYMBOL(single_task_running); | |
4381 | ||
1da177e4 | 4382 | unsigned long long nr_context_switches(void) |
46cb4b7c | 4383 | { |
cc94abfc SR |
4384 | int i; |
4385 | unsigned long long sum = 0; | |
46cb4b7c | 4386 | |
0a945022 | 4387 | for_each_possible_cpu(i) |
1da177e4 | 4388 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 4389 | |
1da177e4 LT |
4390 | return sum; |
4391 | } | |
483b4ee6 | 4392 | |
145d952a DL |
4393 | /* |
4394 | * Consumers of these two interfaces, like for example the cpuidle menu | |
4395 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
4396 | * for a CPU that has IO-wait which might not even end up running the task when | |
4397 | * it does become runnable. | |
4398 | */ | |
4399 | ||
4400 | unsigned long nr_iowait_cpu(int cpu) | |
4401 | { | |
4402 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
4403 | } | |
4404 | ||
e33a9bba | 4405 | /* |
b19a888c | 4406 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
4407 | * |
4408 | * The idea behind IO-wait account is to account the idle time that we could | |
4409 | * have spend running if it were not for IO. That is, if we were to improve the | |
4410 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
4411 | * | |
4412 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
4413 | * idle time as IO-wait, because if the storage were faster, it could've been | |
4414 | * running and we'd not be idle. | |
4415 | * | |
4416 | * This has been extended to SMP, by doing the same for each CPU. This however | |
4417 | * is broken. | |
4418 | * | |
4419 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
4420 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
4421 | * though, if the storage were faster, both could've ran at the same time, | |
4422 | * utilising both CPUs. | |
4423 | * | |
4424 | * This means, that when looking globally, the current IO-wait accounting on | |
4425 | * SMP is a lower bound, by reason of under accounting. | |
4426 | * | |
4427 | * Worse, since the numbers are provided per CPU, they are sometimes | |
4428 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
4429 | * associated with any one particular CPU, it can wake to another CPU than it | |
4430 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
4431 | * | |
4432 | * Task CPU affinities can make all that even more 'interesting'. | |
4433 | */ | |
4434 | ||
1da177e4 LT |
4435 | unsigned long nr_iowait(void) |
4436 | { | |
4437 | unsigned long i, sum = 0; | |
483b4ee6 | 4438 | |
0a945022 | 4439 | for_each_possible_cpu(i) |
145d952a | 4440 | sum += nr_iowait_cpu(i); |
46cb4b7c | 4441 | |
1da177e4 LT |
4442 | return sum; |
4443 | } | |
483b4ee6 | 4444 | |
dd41f596 | 4445 | #ifdef CONFIG_SMP |
8a0be9ef | 4446 | |
46cb4b7c | 4447 | /* |
38022906 PZ |
4448 | * sched_exec - execve() is a valuable balancing opportunity, because at |
4449 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 4450 | */ |
38022906 | 4451 | void sched_exec(void) |
46cb4b7c | 4452 | { |
38022906 | 4453 | struct task_struct *p = current; |
1da177e4 | 4454 | unsigned long flags; |
0017d735 | 4455 | int dest_cpu; |
46cb4b7c | 4456 | |
8f42ced9 | 4457 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 4458 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
4459 | if (dest_cpu == smp_processor_id()) |
4460 | goto unlock; | |
38022906 | 4461 | |
8f42ced9 | 4462 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 4463 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 4464 | |
8f42ced9 PZ |
4465 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
4466 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
4467 | return; |
4468 | } | |
0017d735 | 4469 | unlock: |
8f42ced9 | 4470 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 4471 | } |
dd41f596 | 4472 | |
1da177e4 LT |
4473 | #endif |
4474 | ||
1da177e4 | 4475 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 4476 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
4477 | |
4478 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 4479 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 4480 | |
6075620b GG |
4481 | /* |
4482 | * The function fair_sched_class.update_curr accesses the struct curr | |
4483 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
4484 | * we observe a high rate of cache misses in practice. | |
4485 | * Prefetching this data results in improved performance. | |
4486 | */ | |
4487 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
4488 | { | |
4489 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
4490 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
4491 | #else | |
4492 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
4493 | #endif | |
4494 | prefetch(curr); | |
4495 | prefetch(&curr->exec_start); | |
4496 | } | |
4497 | ||
c5f8d995 HS |
4498 | /* |
4499 | * Return accounted runtime for the task. | |
4500 | * In case the task is currently running, return the runtime plus current's | |
4501 | * pending runtime that have not been accounted yet. | |
4502 | */ | |
4503 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4504 | { | |
eb580751 | 4505 | struct rq_flags rf; |
c5f8d995 | 4506 | struct rq *rq; |
6e998916 | 4507 | u64 ns; |
c5f8d995 | 4508 | |
911b2898 PZ |
4509 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
4510 | /* | |
97fb7a0a | 4511 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
4512 | * So we have a optimization chance when the task's delta_exec is 0. |
4513 | * Reading ->on_cpu is racy, but this is ok. | |
4514 | * | |
d1ccc66d IM |
4515 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
4516 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 4517 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
4518 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
4519 | * been accounted, so we're correct here as well. | |
911b2898 | 4520 | */ |
da0c1e65 | 4521 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
4522 | return p->se.sum_exec_runtime; |
4523 | #endif | |
4524 | ||
eb580751 | 4525 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
4526 | /* |
4527 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
4528 | * project cycles that may never be accounted to this | |
4529 | * thread, breaking clock_gettime(). | |
4530 | */ | |
4531 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 4532 | prefetch_curr_exec_start(p); |
6e998916 SG |
4533 | update_rq_clock(rq); |
4534 | p->sched_class->update_curr(rq); | |
4535 | } | |
4536 | ns = p->se.sum_exec_runtime; | |
eb580751 | 4537 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
4538 | |
4539 | return ns; | |
4540 | } | |
48f24c4d | 4541 | |
7835b98b CL |
4542 | /* |
4543 | * This function gets called by the timer code, with HZ frequency. | |
4544 | * We call it with interrupts disabled. | |
7835b98b CL |
4545 | */ |
4546 | void scheduler_tick(void) | |
4547 | { | |
7835b98b CL |
4548 | int cpu = smp_processor_id(); |
4549 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4550 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 4551 | struct rq_flags rf; |
b4eccf5f | 4552 | unsigned long thermal_pressure; |
3e51f33f | 4553 | |
1567c3e3 | 4554 | arch_scale_freq_tick(); |
3e51f33f | 4555 | sched_clock_tick(); |
dd41f596 | 4556 | |
8a8c69c3 PZ |
4557 | rq_lock(rq, &rf); |
4558 | ||
3e51f33f | 4559 | update_rq_clock(rq); |
b4eccf5f | 4560 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 4561 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 4562 | curr->sched_class->task_tick(rq, curr, 0); |
3289bdb4 | 4563 | calc_global_load_tick(rq); |
eb414681 | 4564 | psi_task_tick(rq); |
8a8c69c3 PZ |
4565 | |
4566 | rq_unlock(rq, &rf); | |
7835b98b | 4567 | |
e9d2b064 | 4568 | perf_event_task_tick(); |
e220d2dc | 4569 | |
e418e1c2 | 4570 | #ifdef CONFIG_SMP |
6eb57e0d | 4571 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 4572 | trigger_load_balance(rq); |
e418e1c2 | 4573 | #endif |
1da177e4 LT |
4574 | } |
4575 | ||
265f22a9 | 4576 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
4577 | |
4578 | struct tick_work { | |
4579 | int cpu; | |
b55bd585 | 4580 | atomic_t state; |
d84b3131 FW |
4581 | struct delayed_work work; |
4582 | }; | |
b55bd585 PM |
4583 | /* Values for ->state, see diagram below. */ |
4584 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
4585 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
4586 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
4587 | ||
4588 | /* | |
4589 | * State diagram for ->state: | |
4590 | * | |
4591 | * | |
4592 | * TICK_SCHED_REMOTE_OFFLINE | |
4593 | * | ^ | |
4594 | * | | | |
4595 | * | | sched_tick_remote() | |
4596 | * | | | |
4597 | * | | | |
4598 | * +--TICK_SCHED_REMOTE_OFFLINING | |
4599 | * | ^ | |
4600 | * | | | |
4601 | * sched_tick_start() | | sched_tick_stop() | |
4602 | * | | | |
4603 | * V | | |
4604 | * TICK_SCHED_REMOTE_RUNNING | |
4605 | * | |
4606 | * | |
4607 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
4608 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
4609 | */ | |
d84b3131 FW |
4610 | |
4611 | static struct tick_work __percpu *tick_work_cpu; | |
4612 | ||
4613 | static void sched_tick_remote(struct work_struct *work) | |
4614 | { | |
4615 | struct delayed_work *dwork = to_delayed_work(work); | |
4616 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
4617 | int cpu = twork->cpu; | |
4618 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 4619 | struct task_struct *curr; |
d84b3131 | 4620 | struct rq_flags rf; |
d9c0ffca | 4621 | u64 delta; |
b55bd585 | 4622 | int os; |
d84b3131 FW |
4623 | |
4624 | /* | |
4625 | * Handle the tick only if it appears the remote CPU is running in full | |
4626 | * dynticks mode. The check is racy by nature, but missing a tick or | |
4627 | * having one too much is no big deal because the scheduler tick updates | |
4628 | * statistics and checks timeslices in a time-independent way, regardless | |
4629 | * of when exactly it is running. | |
4630 | */ | |
488603b8 | 4631 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 4632 | goto out_requeue; |
d84b3131 | 4633 | |
d9c0ffca FW |
4634 | rq_lock_irq(rq, &rf); |
4635 | curr = rq->curr; | |
488603b8 | 4636 | if (cpu_is_offline(cpu)) |
d9c0ffca | 4637 | goto out_unlock; |
d84b3131 | 4638 | |
d9c0ffca | 4639 | update_rq_clock(rq); |
d9c0ffca | 4640 | |
488603b8 SW |
4641 | if (!is_idle_task(curr)) { |
4642 | /* | |
4643 | * Make sure the next tick runs within a reasonable | |
4644 | * amount of time. | |
4645 | */ | |
4646 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
4647 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
4648 | } | |
d9c0ffca FW |
4649 | curr->sched_class->task_tick(rq, curr, 0); |
4650 | ||
ebc0f83c | 4651 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
4652 | out_unlock: |
4653 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 4654 | out_requeue: |
ebc0f83c | 4655 | |
d84b3131 FW |
4656 | /* |
4657 | * Run the remote tick once per second (1Hz). This arbitrary | |
4658 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
4659 | * to keep scheduler internal stats reasonably up to date. But |
4660 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 4661 | */ |
b55bd585 PM |
4662 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
4663 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
4664 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
4665 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
4666 | } |
4667 | ||
4668 | static void sched_tick_start(int cpu) | |
4669 | { | |
b55bd585 | 4670 | int os; |
d84b3131 FW |
4671 | struct tick_work *twork; |
4672 | ||
4673 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4674 | return; | |
4675 | ||
4676 | WARN_ON_ONCE(!tick_work_cpu); | |
4677 | ||
4678 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4679 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
4680 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
4681 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
4682 | twork->cpu = cpu; | |
4683 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
4684 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
4685 | } | |
d84b3131 FW |
4686 | } |
4687 | ||
4688 | #ifdef CONFIG_HOTPLUG_CPU | |
4689 | static void sched_tick_stop(int cpu) | |
4690 | { | |
4691 | struct tick_work *twork; | |
b55bd585 | 4692 | int os; |
d84b3131 FW |
4693 | |
4694 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4695 | return; | |
4696 | ||
4697 | WARN_ON_ONCE(!tick_work_cpu); | |
4698 | ||
4699 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4700 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
4701 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
4702 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
4703 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
4704 | } |
4705 | #endif /* CONFIG_HOTPLUG_CPU */ | |
4706 | ||
4707 | int __init sched_tick_offload_init(void) | |
4708 | { | |
4709 | tick_work_cpu = alloc_percpu(struct tick_work); | |
4710 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
4711 | return 0; |
4712 | } | |
4713 | ||
4714 | #else /* !CONFIG_NO_HZ_FULL */ | |
4715 | static inline void sched_tick_start(int cpu) { } | |
4716 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 4717 | #endif |
1da177e4 | 4718 | |
c1a280b6 | 4719 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 4720 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
4721 | /* |
4722 | * If the value passed in is equal to the current preempt count | |
4723 | * then we just disabled preemption. Start timing the latency. | |
4724 | */ | |
4725 | static inline void preempt_latency_start(int val) | |
4726 | { | |
4727 | if (preempt_count() == val) { | |
4728 | unsigned long ip = get_lock_parent_ip(); | |
4729 | #ifdef CONFIG_DEBUG_PREEMPT | |
4730 | current->preempt_disable_ip = ip; | |
4731 | #endif | |
4732 | trace_preempt_off(CALLER_ADDR0, ip); | |
4733 | } | |
4734 | } | |
7e49fcce | 4735 | |
edafe3a5 | 4736 | void preempt_count_add(int val) |
1da177e4 | 4737 | { |
6cd8a4bb | 4738 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4739 | /* |
4740 | * Underflow? | |
4741 | */ | |
9a11b49a IM |
4742 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4743 | return; | |
6cd8a4bb | 4744 | #endif |
bdb43806 | 4745 | __preempt_count_add(val); |
6cd8a4bb | 4746 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4747 | /* |
4748 | * Spinlock count overflowing soon? | |
4749 | */ | |
33859f7f MOS |
4750 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4751 | PREEMPT_MASK - 10); | |
6cd8a4bb | 4752 | #endif |
47252cfb | 4753 | preempt_latency_start(val); |
1da177e4 | 4754 | } |
bdb43806 | 4755 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 4756 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 4757 | |
47252cfb SR |
4758 | /* |
4759 | * If the value passed in equals to the current preempt count | |
4760 | * then we just enabled preemption. Stop timing the latency. | |
4761 | */ | |
4762 | static inline void preempt_latency_stop(int val) | |
4763 | { | |
4764 | if (preempt_count() == val) | |
4765 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
4766 | } | |
4767 | ||
edafe3a5 | 4768 | void preempt_count_sub(int val) |
1da177e4 | 4769 | { |
6cd8a4bb | 4770 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4771 | /* |
4772 | * Underflow? | |
4773 | */ | |
01e3eb82 | 4774 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4775 | return; |
1da177e4 LT |
4776 | /* |
4777 | * Is the spinlock portion underflowing? | |
4778 | */ | |
9a11b49a IM |
4779 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4780 | !(preempt_count() & PREEMPT_MASK))) | |
4781 | return; | |
6cd8a4bb | 4782 | #endif |
9a11b49a | 4783 | |
47252cfb | 4784 | preempt_latency_stop(val); |
bdb43806 | 4785 | __preempt_count_sub(val); |
1da177e4 | 4786 | } |
bdb43806 | 4787 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 4788 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 4789 | |
47252cfb SR |
4790 | #else |
4791 | static inline void preempt_latency_start(int val) { } | |
4792 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
4793 | #endif |
4794 | ||
59ddbcb2 IM |
4795 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
4796 | { | |
4797 | #ifdef CONFIG_DEBUG_PREEMPT | |
4798 | return p->preempt_disable_ip; | |
4799 | #else | |
4800 | return 0; | |
4801 | #endif | |
4802 | } | |
4803 | ||
1da177e4 | 4804 | /* |
dd41f596 | 4805 | * Print scheduling while atomic bug: |
1da177e4 | 4806 | */ |
dd41f596 | 4807 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4808 | { |
d1c6d149 VN |
4809 | /* Save this before calling printk(), since that will clobber it */ |
4810 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
4811 | ||
664dfa65 DJ |
4812 | if (oops_in_progress) |
4813 | return; | |
4814 | ||
3df0fc5b PZ |
4815 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4816 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4817 | |
dd41f596 | 4818 | debug_show_held_locks(prev); |
e21f5b15 | 4819 | print_modules(); |
dd41f596 IM |
4820 | if (irqs_disabled()) |
4821 | print_irqtrace_events(prev); | |
d1c6d149 VN |
4822 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
4823 | && in_atomic_preempt_off()) { | |
8f47b187 | 4824 | pr_err("Preemption disabled at:"); |
2062a4e8 | 4825 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 4826 | } |
748c7201 DBO |
4827 | if (panic_on_warn) |
4828 | panic("scheduling while atomic\n"); | |
4829 | ||
6135fc1e | 4830 | dump_stack(); |
373d4d09 | 4831 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 4832 | } |
1da177e4 | 4833 | |
dd41f596 IM |
4834 | /* |
4835 | * Various schedule()-time debugging checks and statistics: | |
4836 | */ | |
312364f3 | 4837 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 4838 | { |
0d9e2632 | 4839 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
4840 | if (task_stack_end_corrupted(prev)) |
4841 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
4842 | |
4843 | if (task_scs_end_corrupted(prev)) | |
4844 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 4845 | #endif |
b99def8b | 4846 | |
312364f3 DV |
4847 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
4848 | if (!preempt && prev->state && prev->non_block_count) { | |
4849 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", | |
4850 | prev->comm, prev->pid, prev->non_block_count); | |
4851 | dump_stack(); | |
4852 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
4853 | } | |
4854 | #endif | |
4855 | ||
1dc0fffc | 4856 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 4857 | __schedule_bug(prev); |
1dc0fffc PZ |
4858 | preempt_count_set(PREEMPT_DISABLED); |
4859 | } | |
b3fbab05 | 4860 | rcu_sleep_check(); |
9f68b5b7 | 4861 | SCHED_WARN_ON(ct_state() == CONTEXT_USER); |
dd41f596 | 4862 | |
1da177e4 LT |
4863 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4864 | ||
ae92882e | 4865 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
4866 | } |
4867 | ||
457d1f46 CY |
4868 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
4869 | struct rq_flags *rf) | |
4870 | { | |
4871 | #ifdef CONFIG_SMP | |
4872 | const struct sched_class *class; | |
4873 | /* | |
4874 | * We must do the balancing pass before put_prev_task(), such | |
4875 | * that when we release the rq->lock the task is in the same | |
4876 | * state as before we took rq->lock. | |
4877 | * | |
4878 | * We can terminate the balance pass as soon as we know there is | |
4879 | * a runnable task of @class priority or higher. | |
4880 | */ | |
4881 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
4882 | if (class->balance(rq, prev, rf)) | |
4883 | break; | |
4884 | } | |
4885 | #endif | |
4886 | ||
4887 | put_prev_task(rq, prev); | |
4888 | } | |
4889 | ||
dd41f596 IM |
4890 | /* |
4891 | * Pick up the highest-prio task: | |
4892 | */ | |
4893 | static inline struct task_struct * | |
d8ac8971 | 4894 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 4895 | { |
49ee5768 | 4896 | const struct sched_class *class; |
dd41f596 | 4897 | struct task_struct *p; |
1da177e4 LT |
4898 | |
4899 | /* | |
0ba87bb2 PZ |
4900 | * Optimization: we know that if all tasks are in the fair class we can |
4901 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 4902 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 4903 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 4904 | */ |
aa93cd53 | 4905 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
4906 | rq->nr_running == rq->cfs.h_nr_running)) { |
4907 | ||
5d7d6056 | 4908 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 4909 | if (unlikely(p == RETRY_TASK)) |
67692435 | 4910 | goto restart; |
6ccdc84b | 4911 | |
d1ccc66d | 4912 | /* Assumes fair_sched_class->next == idle_sched_class */ |
5d7d6056 | 4913 | if (!p) { |
f488e105 | 4914 | put_prev_task(rq, prev); |
98c2f700 | 4915 | p = pick_next_task_idle(rq); |
f488e105 | 4916 | } |
6ccdc84b PZ |
4917 | |
4918 | return p; | |
1da177e4 LT |
4919 | } |
4920 | ||
67692435 | 4921 | restart: |
457d1f46 | 4922 | put_prev_task_balance(rq, prev, rf); |
67692435 | 4923 | |
34f971f6 | 4924 | for_each_class(class) { |
98c2f700 | 4925 | p = class->pick_next_task(rq); |
67692435 | 4926 | if (p) |
dd41f596 | 4927 | return p; |
dd41f596 | 4928 | } |
34f971f6 | 4929 | |
d1ccc66d IM |
4930 | /* The idle class should always have a runnable task: */ |
4931 | BUG(); | |
dd41f596 | 4932 | } |
1da177e4 | 4933 | |
dd41f596 | 4934 | /* |
c259e01a | 4935 | * __schedule() is the main scheduler function. |
edde96ea PE |
4936 | * |
4937 | * The main means of driving the scheduler and thus entering this function are: | |
4938 | * | |
4939 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
4940 | * | |
4941 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
4942 | * paths. For example, see arch/x86/entry_64.S. | |
4943 | * | |
4944 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
4945 | * interrupt handler scheduler_tick(). | |
4946 | * | |
4947 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
4948 | * task to the run-queue and that's it. | |
4949 | * | |
4950 | * Now, if the new task added to the run-queue preempts the current | |
4951 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
4952 | * called on the nearest possible occasion: | |
4953 | * | |
c1a280b6 | 4954 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
4955 | * |
4956 | * - in syscall or exception context, at the next outmost | |
4957 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
4958 | * spin_unlock()!) | |
4959 | * | |
4960 | * - in IRQ context, return from interrupt-handler to | |
4961 | * preemptible context | |
4962 | * | |
c1a280b6 | 4963 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
4964 | * then at the next: |
4965 | * | |
4966 | * - cond_resched() call | |
4967 | * - explicit schedule() call | |
4968 | * - return from syscall or exception to user-space | |
4969 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 4970 | * |
b30f0e3f | 4971 | * WARNING: must be called with preemption disabled! |
dd41f596 | 4972 | */ |
499d7955 | 4973 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
4974 | { |
4975 | struct task_struct *prev, *next; | |
67ca7bde | 4976 | unsigned long *switch_count; |
dbfb089d | 4977 | unsigned long prev_state; |
d8ac8971 | 4978 | struct rq_flags rf; |
dd41f596 | 4979 | struct rq *rq; |
31656519 | 4980 | int cpu; |
dd41f596 | 4981 | |
dd41f596 IM |
4982 | cpu = smp_processor_id(); |
4983 | rq = cpu_rq(cpu); | |
dd41f596 | 4984 | prev = rq->curr; |
dd41f596 | 4985 | |
312364f3 | 4986 | schedule_debug(prev, preempt); |
1da177e4 | 4987 | |
e0ee463c | 4988 | if (sched_feat(HRTICK) || sched_feat(HRTICK_DL)) |
f333fdc9 | 4989 | hrtick_clear(rq); |
8f4d37ec | 4990 | |
46a5d164 | 4991 | local_irq_disable(); |
bcbfdd01 | 4992 | rcu_note_context_switch(preempt); |
46a5d164 | 4993 | |
e0acd0a6 ON |
4994 | /* |
4995 | * Make sure that signal_pending_state()->signal_pending() below | |
4996 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
4997 | * done by the caller to avoid the race with signal_wake_up(): |
4998 | * | |
4999 | * __set_current_state(@state) signal_wake_up() | |
5000 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
5001 | * wake_up_state(p, state) | |
5002 | * LOCK rq->lock LOCK p->pi_state | |
5003 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
5004 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 5005 | * |
dbfb089d | 5006 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 5007 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 5008 | */ |
8a8c69c3 | 5009 | rq_lock(rq, &rf); |
d89e588c | 5010 | smp_mb__after_spinlock(); |
1da177e4 | 5011 | |
d1ccc66d IM |
5012 | /* Promote REQ to ACT */ |
5013 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 5014 | update_rq_clock(rq); |
9edfbfed | 5015 | |
246d86b5 | 5016 | switch_count = &prev->nivcsw; |
d136122f | 5017 | |
dbfb089d | 5018 | /* |
d136122f PZ |
5019 | * We must load prev->state once (task_struct::state is volatile), such |
5020 | * that: | |
5021 | * | |
5022 | * - we form a control dependency vs deactivate_task() below. | |
5023 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 5024 | */ |
d136122f PZ |
5025 | prev_state = prev->state; |
5026 | if (!preempt && prev_state) { | |
dbfb089d | 5027 | if (signal_pending_state(prev_state, prev)) { |
1da177e4 | 5028 | prev->state = TASK_RUNNING; |
21aa9af0 | 5029 | } else { |
dbfb089d PZ |
5030 | prev->sched_contributes_to_load = |
5031 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
5032 | !(prev_state & TASK_NOLOAD) && | |
5033 | !(prev->flags & PF_FROZEN); | |
5034 | ||
5035 | if (prev->sched_contributes_to_load) | |
5036 | rq->nr_uninterruptible++; | |
5037 | ||
5038 | /* | |
5039 | * __schedule() ttwu() | |
d136122f PZ |
5040 | * prev_state = prev->state; if (p->on_rq && ...) |
5041 | * if (prev_state) goto out; | |
5042 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
5043 | * p->state = TASK_WAKING | |
5044 | * | |
5045 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
5046 | * |
5047 | * After this, schedule() must not care about p->state any more. | |
5048 | */ | |
bce4dc80 | 5049 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 5050 | |
e33a9bba TH |
5051 | if (prev->in_iowait) { |
5052 | atomic_inc(&rq->nr_iowait); | |
5053 | delayacct_blkio_start(); | |
5054 | } | |
21aa9af0 | 5055 | } |
dd41f596 | 5056 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5057 | } |
5058 | ||
d8ac8971 | 5059 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 5060 | clear_tsk_need_resched(prev); |
f27dde8d | 5061 | clear_preempt_need_resched(); |
1da177e4 | 5062 | |
1da177e4 | 5063 | if (likely(prev != next)) { |
1da177e4 | 5064 | rq->nr_switches++; |
5311a98f EB |
5065 | /* |
5066 | * RCU users of rcu_dereference(rq->curr) may not see | |
5067 | * changes to task_struct made by pick_next_task(). | |
5068 | */ | |
5069 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
5070 | /* |
5071 | * The membarrier system call requires each architecture | |
5072 | * to have a full memory barrier after updating | |
306e0604 MD |
5073 | * rq->curr, before returning to user-space. |
5074 | * | |
5075 | * Here are the schemes providing that barrier on the | |
5076 | * various architectures: | |
5077 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
5078 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
5079 | * - finish_lock_switch() for weakly-ordered | |
5080 | * architectures where spin_unlock is a full barrier, | |
5081 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
5082 | * is a RELEASE barrier), | |
22e4ebb9 | 5083 | */ |
1da177e4 LT |
5084 | ++*switch_count; |
5085 | ||
af449901 | 5086 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
5087 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
5088 | ||
c73464b1 | 5089 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
5090 | |
5091 | /* Also unlocks the rq: */ | |
5092 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 5093 | } else { |
cb42c9a3 | 5094 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 5095 | |
565790d2 PZ |
5096 | rq_unpin_lock(rq, &rf); |
5097 | __balance_callbacks(rq); | |
5098 | raw_spin_unlock_irq(&rq->lock); | |
5099 | } | |
1da177e4 | 5100 | } |
c259e01a | 5101 | |
9af6528e PZ |
5102 | void __noreturn do_task_dead(void) |
5103 | { | |
d1ccc66d | 5104 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 5105 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
5106 | |
5107 | /* Tell freezer to ignore us: */ | |
5108 | current->flags |= PF_NOFREEZE; | |
5109 | ||
9af6528e PZ |
5110 | __schedule(false); |
5111 | BUG(); | |
d1ccc66d IM |
5112 | |
5113 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 5114 | for (;;) |
d1ccc66d | 5115 | cpu_relax(); |
9af6528e PZ |
5116 | } |
5117 | ||
9c40cef2 TG |
5118 | static inline void sched_submit_work(struct task_struct *tsk) |
5119 | { | |
c1cecf88 SAS |
5120 | unsigned int task_flags; |
5121 | ||
b0fdc013 | 5122 | if (!tsk->state) |
9c40cef2 | 5123 | return; |
6d25be57 | 5124 | |
c1cecf88 | 5125 | task_flags = tsk->flags; |
6d25be57 TG |
5126 | /* |
5127 | * If a worker went to sleep, notify and ask workqueue whether | |
5128 | * it wants to wake up a task to maintain concurrency. | |
5129 | * As this function is called inside the schedule() context, | |
5130 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
5131 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
5132 | * requires it. | |
6d25be57 | 5133 | */ |
c1cecf88 | 5134 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 5135 | preempt_disable(); |
c1cecf88 | 5136 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
5137 | wq_worker_sleeping(tsk); |
5138 | else | |
5139 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
5140 | preempt_enable_no_resched(); |
5141 | } | |
5142 | ||
b0fdc013 SAS |
5143 | if (tsk_is_pi_blocked(tsk)) |
5144 | return; | |
5145 | ||
9c40cef2 TG |
5146 | /* |
5147 | * If we are going to sleep and we have plugged IO queued, | |
5148 | * make sure to submit it to avoid deadlocks. | |
5149 | */ | |
5150 | if (blk_needs_flush_plug(tsk)) | |
5151 | blk_schedule_flush_plug(tsk); | |
5152 | } | |
5153 | ||
6d25be57 TG |
5154 | static void sched_update_worker(struct task_struct *tsk) |
5155 | { | |
771b53d0 JA |
5156 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
5157 | if (tsk->flags & PF_WQ_WORKER) | |
5158 | wq_worker_running(tsk); | |
5159 | else | |
5160 | io_wq_worker_running(tsk); | |
5161 | } | |
6d25be57 TG |
5162 | } |
5163 | ||
722a9f92 | 5164 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 5165 | { |
9c40cef2 TG |
5166 | struct task_struct *tsk = current; |
5167 | ||
5168 | sched_submit_work(tsk); | |
bfd9b2b5 | 5169 | do { |
b30f0e3f | 5170 | preempt_disable(); |
fc13aeba | 5171 | __schedule(false); |
b30f0e3f | 5172 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 5173 | } while (need_resched()); |
6d25be57 | 5174 | sched_update_worker(tsk); |
c259e01a | 5175 | } |
1da177e4 LT |
5176 | EXPORT_SYMBOL(schedule); |
5177 | ||
8663effb SRV |
5178 | /* |
5179 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
5180 | * state (have scheduled out non-voluntarily) by making sure that all | |
5181 | * tasks have either left the run queue or have gone into user space. | |
5182 | * As idle tasks do not do either, they must not ever be preempted | |
5183 | * (schedule out non-voluntarily). | |
5184 | * | |
5185 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
5186 | * never enables preemption because it does not call sched_submit_work(). | |
5187 | */ | |
5188 | void __sched schedule_idle(void) | |
5189 | { | |
5190 | /* | |
5191 | * As this skips calling sched_submit_work(), which the idle task does | |
5192 | * regardless because that function is a nop when the task is in a | |
5193 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
5194 | * current task can be in any other state. Note, idle is always in the | |
5195 | * TASK_RUNNING state. | |
5196 | */ | |
5197 | WARN_ON_ONCE(current->state); | |
5198 | do { | |
5199 | __schedule(false); | |
5200 | } while (need_resched()); | |
5201 | } | |
5202 | ||
6775de49 | 5203 | #if defined(CONFIG_CONTEXT_TRACKING) && !defined(CONFIG_HAVE_CONTEXT_TRACKING_OFFSTACK) |
722a9f92 | 5204 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
5205 | { |
5206 | /* | |
5207 | * If we come here after a random call to set_need_resched(), | |
5208 | * or we have been woken up remotely but the IPI has not yet arrived, | |
5209 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
5210 | * we find a better solution. | |
7cc78f8f AL |
5211 | * |
5212 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 5213 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 5214 | * too frequently to make sense yet. |
20ab65e3 | 5215 | */ |
7cc78f8f | 5216 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 5217 | schedule(); |
7cc78f8f | 5218 | exception_exit(prev_state); |
20ab65e3 FW |
5219 | } |
5220 | #endif | |
5221 | ||
c5491ea7 TG |
5222 | /** |
5223 | * schedule_preempt_disabled - called with preemption disabled | |
5224 | * | |
5225 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
5226 | */ | |
5227 | void __sched schedule_preempt_disabled(void) | |
5228 | { | |
ba74c144 | 5229 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
5230 | schedule(); |
5231 | preempt_disable(); | |
5232 | } | |
5233 | ||
06b1f808 | 5234 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
5235 | { |
5236 | do { | |
47252cfb SR |
5237 | /* |
5238 | * Because the function tracer can trace preempt_count_sub() | |
5239 | * and it also uses preempt_enable/disable_notrace(), if | |
5240 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5241 | * by the function tracer will call this function again and | |
5242 | * cause infinite recursion. | |
5243 | * | |
5244 | * Preemption must be disabled here before the function | |
5245 | * tracer can trace. Break up preempt_disable() into two | |
5246 | * calls. One to disable preemption without fear of being | |
5247 | * traced. The other to still record the preemption latency, | |
5248 | * which can also be traced by the function tracer. | |
5249 | */ | |
499d7955 | 5250 | preempt_disable_notrace(); |
47252cfb | 5251 | preempt_latency_start(1); |
fc13aeba | 5252 | __schedule(true); |
47252cfb | 5253 | preempt_latency_stop(1); |
499d7955 | 5254 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
5255 | |
5256 | /* | |
5257 | * Check again in case we missed a preemption opportunity | |
5258 | * between schedule and now. | |
5259 | */ | |
a18b5d01 FW |
5260 | } while (need_resched()); |
5261 | } | |
5262 | ||
c1a280b6 | 5263 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 5264 | /* |
a49b4f40 VS |
5265 | * This is the entry point to schedule() from in-kernel preemption |
5266 | * off of preempt_enable. | |
1da177e4 | 5267 | */ |
722a9f92 | 5268 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 5269 | { |
1da177e4 LT |
5270 | /* |
5271 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5272 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5273 | */ |
fbb00b56 | 5274 | if (likely(!preemptible())) |
1da177e4 LT |
5275 | return; |
5276 | ||
a18b5d01 | 5277 | preempt_schedule_common(); |
1da177e4 | 5278 | } |
376e2424 | 5279 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 5280 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 5281 | |
2c9a98d3 PZI |
5282 | #ifdef CONFIG_PREEMPT_DYNAMIC |
5283 | DEFINE_STATIC_CALL(preempt_schedule, __preempt_schedule_func); | |
ef72661e | 5284 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule); |
2c9a98d3 PZI |
5285 | #endif |
5286 | ||
5287 | ||
009f60e2 | 5288 | /** |
4eaca0a8 | 5289 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
5290 | * |
5291 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
5292 | * recursion and tracing preempt enabling caused by the tracing | |
5293 | * infrastructure itself. But as tracing can happen in areas coming | |
5294 | * from userspace or just about to enter userspace, a preempt enable | |
5295 | * can occur before user_exit() is called. This will cause the scheduler | |
5296 | * to be called when the system is still in usermode. | |
5297 | * | |
5298 | * To prevent this, the preempt_enable_notrace will use this function | |
5299 | * instead of preempt_schedule() to exit user context if needed before | |
5300 | * calling the scheduler. | |
5301 | */ | |
4eaca0a8 | 5302 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
5303 | { |
5304 | enum ctx_state prev_ctx; | |
5305 | ||
5306 | if (likely(!preemptible())) | |
5307 | return; | |
5308 | ||
5309 | do { | |
47252cfb SR |
5310 | /* |
5311 | * Because the function tracer can trace preempt_count_sub() | |
5312 | * and it also uses preempt_enable/disable_notrace(), if | |
5313 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5314 | * by the function tracer will call this function again and | |
5315 | * cause infinite recursion. | |
5316 | * | |
5317 | * Preemption must be disabled here before the function | |
5318 | * tracer can trace. Break up preempt_disable() into two | |
5319 | * calls. One to disable preemption without fear of being | |
5320 | * traced. The other to still record the preemption latency, | |
5321 | * which can also be traced by the function tracer. | |
5322 | */ | |
3d8f74dd | 5323 | preempt_disable_notrace(); |
47252cfb | 5324 | preempt_latency_start(1); |
009f60e2 ON |
5325 | /* |
5326 | * Needs preempt disabled in case user_exit() is traced | |
5327 | * and the tracer calls preempt_enable_notrace() causing | |
5328 | * an infinite recursion. | |
5329 | */ | |
5330 | prev_ctx = exception_enter(); | |
fc13aeba | 5331 | __schedule(true); |
009f60e2 ON |
5332 | exception_exit(prev_ctx); |
5333 | ||
47252cfb | 5334 | preempt_latency_stop(1); |
3d8f74dd | 5335 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
5336 | } while (need_resched()); |
5337 | } | |
4eaca0a8 | 5338 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 5339 | |
2c9a98d3 PZI |
5340 | #ifdef CONFIG_PREEMPT_DYNAMIC |
5341 | DEFINE_STATIC_CALL(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
ef72661e | 5342 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule_notrace); |
2c9a98d3 PZI |
5343 | #endif |
5344 | ||
c1a280b6 | 5345 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 | 5346 | |
826bfeb3 PZI |
5347 | #ifdef CONFIG_PREEMPT_DYNAMIC |
5348 | ||
5349 | #include <linux/entry-common.h> | |
5350 | ||
5351 | /* | |
5352 | * SC:cond_resched | |
5353 | * SC:might_resched | |
5354 | * SC:preempt_schedule | |
5355 | * SC:preempt_schedule_notrace | |
5356 | * SC:irqentry_exit_cond_resched | |
5357 | * | |
5358 | * | |
5359 | * NONE: | |
5360 | * cond_resched <- __cond_resched | |
5361 | * might_resched <- RET0 | |
5362 | * preempt_schedule <- NOP | |
5363 | * preempt_schedule_notrace <- NOP | |
5364 | * irqentry_exit_cond_resched <- NOP | |
5365 | * | |
5366 | * VOLUNTARY: | |
5367 | * cond_resched <- __cond_resched | |
5368 | * might_resched <- __cond_resched | |
5369 | * preempt_schedule <- NOP | |
5370 | * preempt_schedule_notrace <- NOP | |
5371 | * irqentry_exit_cond_resched <- NOP | |
5372 | * | |
5373 | * FULL: | |
5374 | * cond_resched <- RET0 | |
5375 | * might_resched <- RET0 | |
5376 | * preempt_schedule <- preempt_schedule | |
5377 | * preempt_schedule_notrace <- preempt_schedule_notrace | |
5378 | * irqentry_exit_cond_resched <- irqentry_exit_cond_resched | |
5379 | */ | |
e59e10f8 PZ |
5380 | |
5381 | enum { | |
5382 | preempt_dynamic_none = 0, | |
5383 | preempt_dynamic_voluntary, | |
5384 | preempt_dynamic_full, | |
5385 | }; | |
5386 | ||
5387 | static int preempt_dynamic_mode = preempt_dynamic_full; | |
5388 | ||
5389 | static int sched_dynamic_mode(const char *str) | |
826bfeb3 | 5390 | { |
e59e10f8 PZ |
5391 | if (!strcmp(str, "none")) |
5392 | return 0; | |
5393 | ||
5394 | if (!strcmp(str, "voluntary")) | |
5395 | return 1; | |
5396 | ||
5397 | if (!strcmp(str, "full")) | |
5398 | return 2; | |
5399 | ||
5400 | return -1; | |
5401 | } | |
5402 | ||
5403 | static void sched_dynamic_update(int mode) | |
5404 | { | |
5405 | /* | |
5406 | * Avoid {NONE,VOLUNTARY} -> FULL transitions from ever ending up in | |
5407 | * the ZERO state, which is invalid. | |
5408 | */ | |
5409 | static_call_update(cond_resched, __cond_resched); | |
5410 | static_call_update(might_resched, __cond_resched); | |
5411 | static_call_update(preempt_schedule, __preempt_schedule_func); | |
5412 | static_call_update(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
5413 | static_call_update(irqentry_exit_cond_resched, irqentry_exit_cond_resched); | |
5414 | ||
5415 | switch (mode) { | |
5416 | case preempt_dynamic_none: | |
826bfeb3 PZI |
5417 | static_call_update(cond_resched, __cond_resched); |
5418 | static_call_update(might_resched, (typeof(&__cond_resched)) __static_call_return0); | |
5419 | static_call_update(preempt_schedule, (typeof(&preempt_schedule)) NULL); | |
5420 | static_call_update(preempt_schedule_notrace, (typeof(&preempt_schedule_notrace)) NULL); | |
5421 | static_call_update(irqentry_exit_cond_resched, (typeof(&irqentry_exit_cond_resched)) NULL); | |
e59e10f8 PZ |
5422 | pr_info("Dynamic Preempt: none\n"); |
5423 | break; | |
5424 | ||
5425 | case preempt_dynamic_voluntary: | |
826bfeb3 PZI |
5426 | static_call_update(cond_resched, __cond_resched); |
5427 | static_call_update(might_resched, __cond_resched); | |
5428 | static_call_update(preempt_schedule, (typeof(&preempt_schedule)) NULL); | |
5429 | static_call_update(preempt_schedule_notrace, (typeof(&preempt_schedule_notrace)) NULL); | |
5430 | static_call_update(irqentry_exit_cond_resched, (typeof(&irqentry_exit_cond_resched)) NULL); | |
e59e10f8 PZ |
5431 | pr_info("Dynamic Preempt: voluntary\n"); |
5432 | break; | |
5433 | ||
5434 | case preempt_dynamic_full: | |
826bfeb3 PZI |
5435 | static_call_update(cond_resched, (typeof(&__cond_resched)) __static_call_return0); |
5436 | static_call_update(might_resched, (typeof(&__cond_resched)) __static_call_return0); | |
5437 | static_call_update(preempt_schedule, __preempt_schedule_func); | |
5438 | static_call_update(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
5439 | static_call_update(irqentry_exit_cond_resched, irqentry_exit_cond_resched); | |
e59e10f8 PZ |
5440 | pr_info("Dynamic Preempt: full\n"); |
5441 | break; | |
5442 | } | |
5443 | ||
5444 | preempt_dynamic_mode = mode; | |
5445 | } | |
5446 | ||
5447 | static int __init setup_preempt_mode(char *str) | |
5448 | { | |
5449 | int mode = sched_dynamic_mode(str); | |
5450 | if (mode < 0) { | |
5451 | pr_warn("Dynamic Preempt: unsupported mode: %s\n", str); | |
826bfeb3 PZI |
5452 | return 1; |
5453 | } | |
e59e10f8 PZ |
5454 | |
5455 | sched_dynamic_update(mode); | |
826bfeb3 PZI |
5456 | return 0; |
5457 | } | |
5458 | __setup("preempt=", setup_preempt_mode); | |
5459 | ||
e59e10f8 PZ |
5460 | #ifdef CONFIG_SCHED_DEBUG |
5461 | ||
5462 | static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf, | |
5463 | size_t cnt, loff_t *ppos) | |
5464 | { | |
5465 | char buf[16]; | |
5466 | int mode; | |
5467 | ||
5468 | if (cnt > 15) | |
5469 | cnt = 15; | |
5470 | ||
5471 | if (copy_from_user(&buf, ubuf, cnt)) | |
5472 | return -EFAULT; | |
5473 | ||
5474 | buf[cnt] = 0; | |
5475 | mode = sched_dynamic_mode(strstrip(buf)); | |
5476 | if (mode < 0) | |
5477 | return mode; | |
5478 | ||
5479 | sched_dynamic_update(mode); | |
5480 | ||
5481 | *ppos += cnt; | |
5482 | ||
5483 | return cnt; | |
5484 | } | |
5485 | ||
5486 | static int sched_dynamic_show(struct seq_file *m, void *v) | |
5487 | { | |
5488 | static const char * preempt_modes[] = { | |
5489 | "none", "voluntary", "full" | |
5490 | }; | |
5491 | int i; | |
5492 | ||
5493 | for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) { | |
5494 | if (preempt_dynamic_mode == i) | |
5495 | seq_puts(m, "("); | |
5496 | seq_puts(m, preempt_modes[i]); | |
5497 | if (preempt_dynamic_mode == i) | |
5498 | seq_puts(m, ")"); | |
5499 | ||
5500 | seq_puts(m, " "); | |
5501 | } | |
5502 | ||
5503 | seq_puts(m, "\n"); | |
5504 | return 0; | |
5505 | } | |
5506 | ||
5507 | static int sched_dynamic_open(struct inode *inode, struct file *filp) | |
5508 | { | |
5509 | return single_open(filp, sched_dynamic_show, NULL); | |
5510 | } | |
5511 | ||
5512 | static const struct file_operations sched_dynamic_fops = { | |
5513 | .open = sched_dynamic_open, | |
5514 | .write = sched_dynamic_write, | |
5515 | .read = seq_read, | |
5516 | .llseek = seq_lseek, | |
5517 | .release = single_release, | |
5518 | }; | |
5519 | ||
5520 | static __init int sched_init_debug_dynamic(void) | |
5521 | { | |
5522 | debugfs_create_file("sched_preempt", 0644, NULL, NULL, &sched_dynamic_fops); | |
5523 | return 0; | |
5524 | } | |
5525 | late_initcall(sched_init_debug_dynamic); | |
5526 | ||
5527 | #endif /* CONFIG_SCHED_DEBUG */ | |
826bfeb3 PZI |
5528 | #endif /* CONFIG_PREEMPT_DYNAMIC */ |
5529 | ||
5530 | ||
1da177e4 | 5531 | /* |
a49b4f40 | 5532 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5533 | * off of irq context. |
5534 | * Note, that this is called and return with irqs disabled. This will | |
5535 | * protect us against recursive calling from irq. | |
5536 | */ | |
722a9f92 | 5537 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 5538 | { |
b22366cd | 5539 | enum ctx_state prev_state; |
6478d880 | 5540 | |
2ed6e34f | 5541 | /* Catch callers which need to be fixed */ |
f27dde8d | 5542 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 5543 | |
b22366cd FW |
5544 | prev_state = exception_enter(); |
5545 | ||
3a5c359a | 5546 | do { |
3d8f74dd | 5547 | preempt_disable(); |
3a5c359a | 5548 | local_irq_enable(); |
fc13aeba | 5549 | __schedule(true); |
3a5c359a | 5550 | local_irq_disable(); |
3d8f74dd | 5551 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 5552 | } while (need_resched()); |
b22366cd FW |
5553 | |
5554 | exception_exit(prev_state); | |
1da177e4 LT |
5555 | } |
5556 | ||
ac6424b9 | 5557 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 5558 | void *key) |
1da177e4 | 5559 | { |
062d3f95 | 5560 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 5561 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 5562 | } |
1da177e4 LT |
5563 | EXPORT_SYMBOL(default_wake_function); |
5564 | ||
b29739f9 IM |
5565 | #ifdef CONFIG_RT_MUTEXES |
5566 | ||
acd58620 PZ |
5567 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
5568 | { | |
5569 | if (pi_task) | |
5570 | prio = min(prio, pi_task->prio); | |
5571 | ||
5572 | return prio; | |
5573 | } | |
5574 | ||
5575 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5576 | { | |
5577 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
5578 | ||
5579 | return __rt_effective_prio(pi_task, prio); | |
5580 | } | |
5581 | ||
b29739f9 IM |
5582 | /* |
5583 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
5584 | * @p: task to boost |
5585 | * @pi_task: donor task | |
b29739f9 IM |
5586 | * |
5587 | * This function changes the 'effective' priority of a task. It does | |
5588 | * not touch ->normal_prio like __setscheduler(). | |
5589 | * | |
c365c292 TG |
5590 | * Used by the rt_mutex code to implement priority inheritance |
5591 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 5592 | */ |
acd58620 | 5593 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 5594 | { |
acd58620 | 5595 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 5596 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 5597 | const struct sched_class *prev_class; |
eb580751 PZ |
5598 | struct rq_flags rf; |
5599 | struct rq *rq; | |
b29739f9 | 5600 | |
acd58620 PZ |
5601 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
5602 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
5603 | ||
5604 | /* | |
5605 | * If nothing changed; bail early. | |
5606 | */ | |
5607 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
5608 | return; | |
b29739f9 | 5609 | |
eb580751 | 5610 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 5611 | update_rq_clock(rq); |
acd58620 PZ |
5612 | /* |
5613 | * Set under pi_lock && rq->lock, such that the value can be used under | |
5614 | * either lock. | |
5615 | * | |
5616 | * Note that there is loads of tricky to make this pointer cache work | |
5617 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
5618 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
5619 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 5620 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
5621 | */ |
5622 | p->pi_top_task = pi_task; | |
5623 | ||
5624 | /* | |
5625 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
5626 | */ | |
5627 | if (prio == p->prio && !dl_prio(prio)) | |
5628 | goto out_unlock; | |
b29739f9 | 5629 | |
1c4dd99b TG |
5630 | /* |
5631 | * Idle task boosting is a nono in general. There is one | |
5632 | * exception, when PREEMPT_RT and NOHZ is active: | |
5633 | * | |
5634 | * The idle task calls get_next_timer_interrupt() and holds | |
5635 | * the timer wheel base->lock on the CPU and another CPU wants | |
5636 | * to access the timer (probably to cancel it). We can safely | |
5637 | * ignore the boosting request, as the idle CPU runs this code | |
5638 | * with interrupts disabled and will complete the lock | |
5639 | * protected section without being interrupted. So there is no | |
5640 | * real need to boost. | |
5641 | */ | |
5642 | if (unlikely(p == rq->idle)) { | |
5643 | WARN_ON(p != rq->curr); | |
5644 | WARN_ON(p->pi_blocked_on); | |
5645 | goto out_unlock; | |
5646 | } | |
5647 | ||
b91473ff | 5648 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 5649 | oldprio = p->prio; |
ff77e468 PZ |
5650 | |
5651 | if (oldprio == prio) | |
5652 | queue_flag &= ~DEQUEUE_MOVE; | |
5653 | ||
83ab0aa0 | 5654 | prev_class = p->sched_class; |
da0c1e65 | 5655 | queued = task_on_rq_queued(p); |
051a1d1a | 5656 | running = task_current(rq, p); |
da0c1e65 | 5657 | if (queued) |
ff77e468 | 5658 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 5659 | if (running) |
f3cd1c4e | 5660 | put_prev_task(rq, p); |
dd41f596 | 5661 | |
2d3d891d DF |
5662 | /* |
5663 | * Boosting condition are: | |
5664 | * 1. -rt task is running and holds mutex A | |
5665 | * --> -dl task blocks on mutex A | |
5666 | * | |
5667 | * 2. -dl task is running and holds mutex A | |
5668 | * --> -dl task blocks on mutex A and could preempt the | |
5669 | * running task | |
5670 | */ | |
5671 | if (dl_prio(prio)) { | |
466af29b | 5672 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
5673 | (pi_task && dl_prio(pi_task->prio) && |
5674 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2279f540 | 5675 | p->dl.pi_se = pi_task->dl.pi_se; |
ff77e468 | 5676 | queue_flag |= ENQUEUE_REPLENISH; |
2279f540 JL |
5677 | } else { |
5678 | p->dl.pi_se = &p->dl; | |
5679 | } | |
aab03e05 | 5680 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
5681 | } else if (rt_prio(prio)) { |
5682 | if (dl_prio(oldprio)) | |
2279f540 | 5683 | p->dl.pi_se = &p->dl; |
2d3d891d | 5684 | if (oldprio < prio) |
ff77e468 | 5685 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 5686 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
5687 | } else { |
5688 | if (dl_prio(oldprio)) | |
2279f540 | 5689 | p->dl.pi_se = &p->dl; |
746db944 BS |
5690 | if (rt_prio(oldprio)) |
5691 | p->rt.timeout = 0; | |
dd41f596 | 5692 | p->sched_class = &fair_sched_class; |
2d3d891d | 5693 | } |
dd41f596 | 5694 | |
b29739f9 IM |
5695 | p->prio = prio; |
5696 | ||
da0c1e65 | 5697 | if (queued) |
ff77e468 | 5698 | enqueue_task(rq, p, queue_flag); |
a399d233 | 5699 | if (running) |
03b7fad1 | 5700 | set_next_task(rq, p); |
cb469845 | 5701 | |
da7a735e | 5702 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 5703 | out_unlock: |
d1ccc66d IM |
5704 | /* Avoid rq from going away on us: */ |
5705 | preempt_disable(); | |
4c9a4bc8 | 5706 | |
565790d2 PZ |
5707 | rq_unpin_lock(rq, &rf); |
5708 | __balance_callbacks(rq); | |
5709 | raw_spin_unlock(&rq->lock); | |
5710 | ||
4c9a4bc8 | 5711 | preempt_enable(); |
b29739f9 | 5712 | } |
acd58620 PZ |
5713 | #else |
5714 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5715 | { | |
5716 | return prio; | |
5717 | } | |
b29739f9 | 5718 | #endif |
d50dde5a | 5719 | |
36c8b586 | 5720 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5721 | { |
49bd21ef | 5722 | bool queued, running; |
53a23364 | 5723 | int old_prio; |
eb580751 | 5724 | struct rq_flags rf; |
70b97a7f | 5725 | struct rq *rq; |
1da177e4 | 5726 | |
75e45d51 | 5727 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
5728 | return; |
5729 | /* | |
5730 | * We have to be careful, if called from sys_setpriority(), | |
5731 | * the task might be in the middle of scheduling on another CPU. | |
5732 | */ | |
eb580751 | 5733 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
5734 | update_rq_clock(rq); |
5735 | ||
1da177e4 LT |
5736 | /* |
5737 | * The RT priorities are set via sched_setscheduler(), but we still | |
5738 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 5739 | * it won't have any effect on scheduling until the task is |
aab03e05 | 5740 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 5741 | */ |
aab03e05 | 5742 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
5743 | p->static_prio = NICE_TO_PRIO(nice); |
5744 | goto out_unlock; | |
5745 | } | |
da0c1e65 | 5746 | queued = task_on_rq_queued(p); |
49bd21ef | 5747 | running = task_current(rq, p); |
da0c1e65 | 5748 | if (queued) |
7a57f32a | 5749 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
5750 | if (running) |
5751 | put_prev_task(rq, p); | |
1da177e4 | 5752 | |
1da177e4 | 5753 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 5754 | set_load_weight(p, true); |
b29739f9 IM |
5755 | old_prio = p->prio; |
5756 | p->prio = effective_prio(p); | |
1da177e4 | 5757 | |
5443a0be | 5758 | if (queued) |
7134b3e9 | 5759 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 5760 | if (running) |
03b7fad1 | 5761 | set_next_task(rq, p); |
5443a0be FW |
5762 | |
5763 | /* | |
5764 | * If the task increased its priority or is running and | |
5765 | * lowered its priority, then reschedule its CPU: | |
5766 | */ | |
5767 | p->sched_class->prio_changed(rq, p, old_prio); | |
5768 | ||
1da177e4 | 5769 | out_unlock: |
eb580751 | 5770 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 5771 | } |
1da177e4 LT |
5772 | EXPORT_SYMBOL(set_user_nice); |
5773 | ||
e43379f1 MM |
5774 | /* |
5775 | * can_nice - check if a task can reduce its nice value | |
5776 | * @p: task | |
5777 | * @nice: nice value | |
5778 | */ | |
36c8b586 | 5779 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5780 | { |
d1ccc66d | 5781 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 5782 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 5783 | |
78d7d407 | 5784 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5785 | capable(CAP_SYS_NICE)); |
5786 | } | |
5787 | ||
1da177e4 LT |
5788 | #ifdef __ARCH_WANT_SYS_NICE |
5789 | ||
5790 | /* | |
5791 | * sys_nice - change the priority of the current process. | |
5792 | * @increment: priority increment | |
5793 | * | |
5794 | * sys_setpriority is a more generic, but much slower function that | |
5795 | * does similar things. | |
5796 | */ | |
5add95d4 | 5797 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5798 | { |
48f24c4d | 5799 | long nice, retval; |
1da177e4 LT |
5800 | |
5801 | /* | |
5802 | * Setpriority might change our priority at the same moment. | |
5803 | * We don't have to worry. Conceptually one call occurs first | |
5804 | * and we have a single winner. | |
5805 | */ | |
a9467fa3 | 5806 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 5807 | nice = task_nice(current) + increment; |
1da177e4 | 5808 | |
a9467fa3 | 5809 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
5810 | if (increment < 0 && !can_nice(current, nice)) |
5811 | return -EPERM; | |
5812 | ||
1da177e4 LT |
5813 | retval = security_task_setnice(current, nice); |
5814 | if (retval) | |
5815 | return retval; | |
5816 | ||
5817 | set_user_nice(current, nice); | |
5818 | return 0; | |
5819 | } | |
5820 | ||
5821 | #endif | |
5822 | ||
5823 | /** | |
5824 | * task_prio - return the priority value of a given task. | |
5825 | * @p: the task in question. | |
5826 | * | |
e69f6186 | 5827 | * Return: The priority value as seen by users in /proc. |
c541bb78 DE |
5828 | * |
5829 | * sched policy return value kernel prio user prio/nice | |
5830 | * | |
5831 | * normal, batch, idle [0 ... 39] [100 ... 139] 0/[-20 ... 19] | |
5832 | * fifo, rr [-2 ... -100] [98 ... 0] [1 ... 99] | |
5833 | * deadline -101 -1 0 | |
1da177e4 | 5834 | */ |
36c8b586 | 5835 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5836 | { |
5837 | return p->prio - MAX_RT_PRIO; | |
5838 | } | |
5839 | ||
1da177e4 | 5840 | /** |
d1ccc66d | 5841 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 5842 | * @cpu: the processor in question. |
e69f6186 YB |
5843 | * |
5844 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
5845 | */ |
5846 | int idle_cpu(int cpu) | |
5847 | { | |
908a3283 TG |
5848 | struct rq *rq = cpu_rq(cpu); |
5849 | ||
5850 | if (rq->curr != rq->idle) | |
5851 | return 0; | |
5852 | ||
5853 | if (rq->nr_running) | |
5854 | return 0; | |
5855 | ||
5856 | #ifdef CONFIG_SMP | |
126c2092 | 5857 | if (rq->ttwu_pending) |
908a3283 TG |
5858 | return 0; |
5859 | #endif | |
5860 | ||
5861 | return 1; | |
1da177e4 LT |
5862 | } |
5863 | ||
943d355d RJ |
5864 | /** |
5865 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
5866 | * @cpu: the CPU in question. | |
5867 | * | |
5868 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
5869 | */ | |
5870 | int available_idle_cpu(int cpu) | |
5871 | { | |
5872 | if (!idle_cpu(cpu)) | |
5873 | return 0; | |
5874 | ||
247f2f6f RJ |
5875 | if (vcpu_is_preempted(cpu)) |
5876 | return 0; | |
5877 | ||
908a3283 | 5878 | return 1; |
1da177e4 LT |
5879 | } |
5880 | ||
1da177e4 | 5881 | /** |
d1ccc66d | 5882 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 5883 | * @cpu: the processor in question. |
e69f6186 | 5884 | * |
d1ccc66d | 5885 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 5886 | */ |
36c8b586 | 5887 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5888 | { |
5889 | return cpu_rq(cpu)->idle; | |
5890 | } | |
5891 | ||
7d6a905f VK |
5892 | #ifdef CONFIG_SMP |
5893 | /* | |
5894 | * This function computes an effective utilization for the given CPU, to be | |
5895 | * used for frequency selection given the linear relation: f = u * f_max. | |
5896 | * | |
5897 | * The scheduler tracks the following metrics: | |
5898 | * | |
5899 | * cpu_util_{cfs,rt,dl,irq}() | |
5900 | * cpu_bw_dl() | |
5901 | * | |
5902 | * Where the cfs,rt and dl util numbers are tracked with the same metric and | |
5903 | * synchronized windows and are thus directly comparable. | |
5904 | * | |
5905 | * The cfs,rt,dl utilization are the running times measured with rq->clock_task | |
5906 | * which excludes things like IRQ and steal-time. These latter are then accrued | |
5907 | * in the irq utilization. | |
5908 | * | |
5909 | * The DL bandwidth number otoh is not a measured metric but a value computed | |
5910 | * based on the task model parameters and gives the minimal utilization | |
5911 | * required to meet deadlines. | |
5912 | */ | |
a5418be9 VK |
5913 | unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, |
5914 | unsigned long max, enum cpu_util_type type, | |
7d6a905f VK |
5915 | struct task_struct *p) |
5916 | { | |
5917 | unsigned long dl_util, util, irq; | |
5918 | struct rq *rq = cpu_rq(cpu); | |
5919 | ||
5920 | if (!uclamp_is_used() && | |
5921 | type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) { | |
5922 | return max; | |
5923 | } | |
5924 | ||
5925 | /* | |
5926 | * Early check to see if IRQ/steal time saturates the CPU, can be | |
5927 | * because of inaccuracies in how we track these -- see | |
5928 | * update_irq_load_avg(). | |
5929 | */ | |
5930 | irq = cpu_util_irq(rq); | |
5931 | if (unlikely(irq >= max)) | |
5932 | return max; | |
5933 | ||
5934 | /* | |
5935 | * Because the time spend on RT/DL tasks is visible as 'lost' time to | |
5936 | * CFS tasks and we use the same metric to track the effective | |
5937 | * utilization (PELT windows are synchronized) we can directly add them | |
5938 | * to obtain the CPU's actual utilization. | |
5939 | * | |
5940 | * CFS and RT utilization can be boosted or capped, depending on | |
5941 | * utilization clamp constraints requested by currently RUNNABLE | |
5942 | * tasks. | |
5943 | * When there are no CFS RUNNABLE tasks, clamps are released and | |
5944 | * frequency will be gracefully reduced with the utilization decay. | |
5945 | */ | |
5946 | util = util_cfs + cpu_util_rt(rq); | |
5947 | if (type == FREQUENCY_UTIL) | |
5948 | util = uclamp_rq_util_with(rq, util, p); | |
5949 | ||
5950 | dl_util = cpu_util_dl(rq); | |
5951 | ||
5952 | /* | |
5953 | * For frequency selection we do not make cpu_util_dl() a permanent part | |
5954 | * of this sum because we want to use cpu_bw_dl() later on, but we need | |
5955 | * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such | |
5956 | * that we select f_max when there is no idle time. | |
5957 | * | |
5958 | * NOTE: numerical errors or stop class might cause us to not quite hit | |
5959 | * saturation when we should -- something for later. | |
5960 | */ | |
5961 | if (util + dl_util >= max) | |
5962 | return max; | |
5963 | ||
5964 | /* | |
5965 | * OTOH, for energy computation we need the estimated running time, so | |
5966 | * include util_dl and ignore dl_bw. | |
5967 | */ | |
5968 | if (type == ENERGY_UTIL) | |
5969 | util += dl_util; | |
5970 | ||
5971 | /* | |
5972 | * There is still idle time; further improve the number by using the | |
5973 | * irq metric. Because IRQ/steal time is hidden from the task clock we | |
5974 | * need to scale the task numbers: | |
5975 | * | |
5976 | * max - irq | |
5977 | * U' = irq + --------- * U | |
5978 | * max | |
5979 | */ | |
5980 | util = scale_irq_capacity(util, irq, max); | |
5981 | util += irq; | |
5982 | ||
5983 | /* | |
5984 | * Bandwidth required by DEADLINE must always be granted while, for | |
5985 | * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism | |
5986 | * to gracefully reduce the frequency when no tasks show up for longer | |
5987 | * periods of time. | |
5988 | * | |
5989 | * Ideally we would like to set bw_dl as min/guaranteed freq and util + | |
5990 | * bw_dl as requested freq. However, cpufreq is not yet ready for such | |
5991 | * an interface. So, we only do the latter for now. | |
5992 | */ | |
5993 | if (type == FREQUENCY_UTIL) | |
5994 | util += cpu_bw_dl(rq); | |
5995 | ||
5996 | return min(max, util); | |
5997 | } | |
a5418be9 VK |
5998 | |
5999 | unsigned long sched_cpu_util(int cpu, unsigned long max) | |
6000 | { | |
6001 | return effective_cpu_util(cpu, cpu_util_cfs(cpu_rq(cpu)), max, | |
6002 | ENERGY_UTIL, NULL); | |
6003 | } | |
7d6a905f VK |
6004 | #endif /* CONFIG_SMP */ |
6005 | ||
1da177e4 LT |
6006 | /** |
6007 | * find_process_by_pid - find a process with a matching PID value. | |
6008 | * @pid: the pid in question. | |
e69f6186 YB |
6009 | * |
6010 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 6011 | */ |
a9957449 | 6012 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6013 | { |
228ebcbe | 6014 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6015 | } |
6016 | ||
c13db6b1 SR |
6017 | /* |
6018 | * sched_setparam() passes in -1 for its policy, to let the functions | |
6019 | * it calls know not to change it. | |
6020 | */ | |
6021 | #define SETPARAM_POLICY -1 | |
6022 | ||
c365c292 TG |
6023 | static void __setscheduler_params(struct task_struct *p, |
6024 | const struct sched_attr *attr) | |
1da177e4 | 6025 | { |
d50dde5a DF |
6026 | int policy = attr->sched_policy; |
6027 | ||
c13db6b1 | 6028 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
6029 | policy = p->policy; |
6030 | ||
1da177e4 | 6031 | p->policy = policy; |
d50dde5a | 6032 | |
aab03e05 DF |
6033 | if (dl_policy(policy)) |
6034 | __setparam_dl(p, attr); | |
39fd8fd2 | 6035 | else if (fair_policy(policy)) |
d50dde5a DF |
6036 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
6037 | ||
39fd8fd2 PZ |
6038 | /* |
6039 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
6040 | * !rt_policy. Always setting this ensures that things like | |
6041 | * getparam()/getattr() don't report silly values for !rt tasks. | |
6042 | */ | |
6043 | p->rt_priority = attr->sched_priority; | |
383afd09 | 6044 | p->normal_prio = normal_prio(p); |
9059393e | 6045 | set_load_weight(p, true); |
c365c292 | 6046 | } |
39fd8fd2 | 6047 | |
c365c292 TG |
6048 | /* Actually do priority change: must hold pi & rq lock. */ |
6049 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 6050 | const struct sched_attr *attr, bool keep_boost) |
c365c292 | 6051 | { |
a509a7cd PB |
6052 | /* |
6053 | * If params can't change scheduling class changes aren't allowed | |
6054 | * either. | |
6055 | */ | |
6056 | if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) | |
6057 | return; | |
6058 | ||
c365c292 | 6059 | __setscheduler_params(p, attr); |
d50dde5a | 6060 | |
383afd09 | 6061 | /* |
0782e63b TG |
6062 | * Keep a potential priority boosting if called from |
6063 | * sched_setscheduler(). | |
383afd09 | 6064 | */ |
acd58620 | 6065 | p->prio = normal_prio(p); |
0782e63b | 6066 | if (keep_boost) |
acd58620 | 6067 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 6068 | |
aab03e05 DF |
6069 | if (dl_prio(p->prio)) |
6070 | p->sched_class = &dl_sched_class; | |
6071 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
6072 | p->sched_class = &rt_sched_class; |
6073 | else | |
6074 | p->sched_class = &fair_sched_class; | |
1da177e4 | 6075 | } |
aab03e05 | 6076 | |
c69e8d9c | 6077 | /* |
d1ccc66d | 6078 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
6079 | */ |
6080 | static bool check_same_owner(struct task_struct *p) | |
6081 | { | |
6082 | const struct cred *cred = current_cred(), *pcred; | |
6083 | bool match; | |
6084 | ||
6085 | rcu_read_lock(); | |
6086 | pcred = __task_cred(p); | |
9c806aa0 EB |
6087 | match = (uid_eq(cred->euid, pcred->euid) || |
6088 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
6089 | rcu_read_unlock(); |
6090 | return match; | |
6091 | } | |
6092 | ||
d50dde5a DF |
6093 | static int __sched_setscheduler(struct task_struct *p, |
6094 | const struct sched_attr *attr, | |
dbc7f069 | 6095 | bool user, bool pi) |
1da177e4 | 6096 | { |
383afd09 SR |
6097 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
6098 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 6099 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 6100 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 6101 | const struct sched_class *prev_class; |
565790d2 | 6102 | struct callback_head *head; |
eb580751 | 6103 | struct rq_flags rf; |
ca94c442 | 6104 | int reset_on_fork; |
7a57f32a | 6105 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 6106 | struct rq *rq; |
1da177e4 | 6107 | |
896bbb25 SRV |
6108 | /* The pi code expects interrupts enabled */ |
6109 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 6110 | recheck: |
d1ccc66d | 6111 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
6112 | if (policy < 0) { |
6113 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6114 | policy = oldpolicy = p->policy; |
ca94c442 | 6115 | } else { |
7479f3c9 | 6116 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 6117 | |
20f9cd2a | 6118 | if (!valid_policy(policy)) |
ca94c442 LP |
6119 | return -EINVAL; |
6120 | } | |
6121 | ||
794a56eb | 6122 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
6123 | return -EINVAL; |
6124 | ||
1da177e4 LT |
6125 | /* |
6126 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
ae18ad28 | 6127 | * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL, |
dd41f596 | 6128 | * SCHED_BATCH and SCHED_IDLE is 0. |
1da177e4 | 6129 | */ |
ae18ad28 | 6130 | if (attr->sched_priority > MAX_RT_PRIO-1) |
1da177e4 | 6131 | return -EINVAL; |
aab03e05 DF |
6132 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
6133 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
6134 | return -EINVAL; |
6135 | ||
37e4ab3f OC |
6136 | /* |
6137 | * Allow unprivileged RT tasks to decrease priority: | |
6138 | */ | |
961ccddd | 6139 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 6140 | if (fair_policy(policy)) { |
d0ea0268 | 6141 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 6142 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
6143 | return -EPERM; |
6144 | } | |
6145 | ||
e05606d3 | 6146 | if (rt_policy(policy)) { |
a44702e8 ON |
6147 | unsigned long rlim_rtprio = |
6148 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 6149 | |
d1ccc66d | 6150 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
6151 | if (policy != p->policy && !rlim_rtprio) |
6152 | return -EPERM; | |
6153 | ||
d1ccc66d | 6154 | /* Can't increase priority: */ |
d50dde5a DF |
6155 | if (attr->sched_priority > p->rt_priority && |
6156 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
6157 | return -EPERM; |
6158 | } | |
c02aa73b | 6159 | |
d44753b8 JL |
6160 | /* |
6161 | * Can't set/change SCHED_DEADLINE policy at all for now | |
6162 | * (safest behavior); in the future we would like to allow | |
6163 | * unprivileged DL tasks to increase their relative deadline | |
6164 | * or reduce their runtime (both ways reducing utilization) | |
6165 | */ | |
6166 | if (dl_policy(policy)) | |
6167 | return -EPERM; | |
6168 | ||
dd41f596 | 6169 | /* |
c02aa73b DH |
6170 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
6171 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 6172 | */ |
1da1843f | 6173 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 6174 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
6175 | return -EPERM; |
6176 | } | |
5fe1d75f | 6177 | |
d1ccc66d | 6178 | /* Can't change other user's priorities: */ |
c69e8d9c | 6179 | if (!check_same_owner(p)) |
37e4ab3f | 6180 | return -EPERM; |
ca94c442 | 6181 | |
d1ccc66d | 6182 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
6183 | if (p->sched_reset_on_fork && !reset_on_fork) |
6184 | return -EPERM; | |
37e4ab3f | 6185 | } |
1da177e4 | 6186 | |
725aad24 | 6187 | if (user) { |
794a56eb JL |
6188 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
6189 | return -EINVAL; | |
6190 | ||
b0ae1981 | 6191 | retval = security_task_setscheduler(p); |
725aad24 JF |
6192 | if (retval) |
6193 | return retval; | |
6194 | } | |
6195 | ||
a509a7cd PB |
6196 | /* Update task specific "requested" clamps */ |
6197 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
6198 | retval = uclamp_validate(p, attr); | |
6199 | if (retval) | |
6200 | return retval; | |
6201 | } | |
6202 | ||
710da3c8 JL |
6203 | if (pi) |
6204 | cpuset_read_lock(); | |
6205 | ||
b29739f9 | 6206 | /* |
d1ccc66d | 6207 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 6208 | * changing the priority of the task: |
0122ec5b | 6209 | * |
25985edc | 6210 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
6211 | * runqueue lock must be held. |
6212 | */ | |
eb580751 | 6213 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 6214 | update_rq_clock(rq); |
dc61b1d6 | 6215 | |
34f971f6 | 6216 | /* |
d1ccc66d | 6217 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
6218 | */ |
6219 | if (p == rq->stop) { | |
4b211f2b MP |
6220 | retval = -EINVAL; |
6221 | goto unlock; | |
34f971f6 PZ |
6222 | } |
6223 | ||
a51e9198 | 6224 | /* |
d6b1e911 TG |
6225 | * If not changing anything there's no need to proceed further, |
6226 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 6227 | */ |
d50dde5a | 6228 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 6229 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
6230 | goto change; |
6231 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
6232 | goto change; | |
75381608 | 6233 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 6234 | goto change; |
a509a7cd PB |
6235 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
6236 | goto change; | |
d50dde5a | 6237 | |
d6b1e911 | 6238 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
6239 | retval = 0; |
6240 | goto unlock; | |
a51e9198 | 6241 | } |
d50dde5a | 6242 | change: |
a51e9198 | 6243 | |
dc61b1d6 | 6244 | if (user) { |
332ac17e | 6245 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
6246 | /* |
6247 | * Do not allow realtime tasks into groups that have no runtime | |
6248 | * assigned. | |
6249 | */ | |
6250 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
6251 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
6252 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
6253 | retval = -EPERM; |
6254 | goto unlock; | |
dc61b1d6 | 6255 | } |
dc61b1d6 | 6256 | #endif |
332ac17e | 6257 | #ifdef CONFIG_SMP |
794a56eb JL |
6258 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
6259 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 6260 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
6261 | |
6262 | /* | |
6263 | * Don't allow tasks with an affinity mask smaller than | |
6264 | * the entire root_domain to become SCHED_DEADLINE. We | |
6265 | * will also fail if there's no bandwidth available. | |
6266 | */ | |
3bd37062 | 6267 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 6268 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
6269 | retval = -EPERM; |
6270 | goto unlock; | |
332ac17e DF |
6271 | } |
6272 | } | |
6273 | #endif | |
6274 | } | |
dc61b1d6 | 6275 | |
d1ccc66d | 6276 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
6277 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
6278 | policy = oldpolicy = -1; | |
eb580751 | 6279 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
6280 | if (pi) |
6281 | cpuset_read_unlock(); | |
1da177e4 LT |
6282 | goto recheck; |
6283 | } | |
332ac17e DF |
6284 | |
6285 | /* | |
6286 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
6287 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
6288 | * is available. | |
6289 | */ | |
06a76fe0 | 6290 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
6291 | retval = -EBUSY; |
6292 | goto unlock; | |
332ac17e DF |
6293 | } |
6294 | ||
c365c292 TG |
6295 | p->sched_reset_on_fork = reset_on_fork; |
6296 | oldprio = p->prio; | |
6297 | ||
dbc7f069 PZ |
6298 | if (pi) { |
6299 | /* | |
6300 | * Take priority boosted tasks into account. If the new | |
6301 | * effective priority is unchanged, we just store the new | |
6302 | * normal parameters and do not touch the scheduler class and | |
6303 | * the runqueue. This will be done when the task deboost | |
6304 | * itself. | |
6305 | */ | |
acd58620 | 6306 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
6307 | if (new_effective_prio == oldprio) |
6308 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
6309 | } |
6310 | ||
da0c1e65 | 6311 | queued = task_on_rq_queued(p); |
051a1d1a | 6312 | running = task_current(rq, p); |
da0c1e65 | 6313 | if (queued) |
ff77e468 | 6314 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 6315 | if (running) |
f3cd1c4e | 6316 | put_prev_task(rq, p); |
f6b53205 | 6317 | |
83ab0aa0 | 6318 | prev_class = p->sched_class; |
a509a7cd | 6319 | |
dbc7f069 | 6320 | __setscheduler(rq, p, attr, pi); |
a509a7cd | 6321 | __setscheduler_uclamp(p, attr); |
f6b53205 | 6322 | |
da0c1e65 | 6323 | if (queued) { |
81a44c54 TG |
6324 | /* |
6325 | * We enqueue to tail when the priority of a task is | |
6326 | * increased (user space view). | |
6327 | */ | |
ff77e468 PZ |
6328 | if (oldprio < p->prio) |
6329 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 6330 | |
ff77e468 | 6331 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 6332 | } |
a399d233 | 6333 | if (running) |
03b7fad1 | 6334 | set_next_task(rq, p); |
cb469845 | 6335 | |
da7a735e | 6336 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
6337 | |
6338 | /* Avoid rq from going away on us: */ | |
6339 | preempt_disable(); | |
565790d2 | 6340 | head = splice_balance_callbacks(rq); |
eb580751 | 6341 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 6342 | |
710da3c8 JL |
6343 | if (pi) { |
6344 | cpuset_read_unlock(); | |
dbc7f069 | 6345 | rt_mutex_adjust_pi(p); |
710da3c8 | 6346 | } |
95e02ca9 | 6347 | |
d1ccc66d | 6348 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 6349 | balance_callbacks(rq, head); |
4c9a4bc8 | 6350 | preempt_enable(); |
95e02ca9 | 6351 | |
1da177e4 | 6352 | return 0; |
4b211f2b MP |
6353 | |
6354 | unlock: | |
6355 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
6356 | if (pi) |
6357 | cpuset_read_unlock(); | |
4b211f2b | 6358 | return retval; |
1da177e4 | 6359 | } |
961ccddd | 6360 | |
7479f3c9 PZ |
6361 | static int _sched_setscheduler(struct task_struct *p, int policy, |
6362 | const struct sched_param *param, bool check) | |
6363 | { | |
6364 | struct sched_attr attr = { | |
6365 | .sched_policy = policy, | |
6366 | .sched_priority = param->sched_priority, | |
6367 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
6368 | }; | |
6369 | ||
c13db6b1 SR |
6370 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
6371 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
6372 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
6373 | policy &= ~SCHED_RESET_ON_FORK; | |
6374 | attr.sched_policy = policy; | |
6375 | } | |
6376 | ||
dbc7f069 | 6377 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 6378 | } |
961ccddd RR |
6379 | /** |
6380 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6381 | * @p: the task in question. | |
6382 | * @policy: new policy. | |
6383 | * @param: structure containing the new RT priority. | |
6384 | * | |
7318d4cc PZ |
6385 | * Use sched_set_fifo(), read its comment. |
6386 | * | |
e69f6186 YB |
6387 | * Return: 0 on success. An error code otherwise. |
6388 | * | |
961ccddd RR |
6389 | * NOTE that the task may be already dead. |
6390 | */ | |
6391 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 6392 | const struct sched_param *param) |
961ccddd | 6393 | { |
7479f3c9 | 6394 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 6395 | } |
1da177e4 | 6396 | |
d50dde5a DF |
6397 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
6398 | { | |
dbc7f069 | 6399 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 6400 | } |
d50dde5a | 6401 | |
794a56eb JL |
6402 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
6403 | { | |
6404 | return __sched_setscheduler(p, attr, false, true); | |
6405 | } | |
6406 | ||
961ccddd RR |
6407 | /** |
6408 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6409 | * @p: the task in question. | |
6410 | * @policy: new policy. | |
6411 | * @param: structure containing the new RT priority. | |
6412 | * | |
6413 | * Just like sched_setscheduler, only don't bother checking if the | |
6414 | * current context has permission. For example, this is needed in | |
6415 | * stop_machine(): we create temporary high priority worker threads, | |
6416 | * but our caller might not have that capability. | |
e69f6186 YB |
6417 | * |
6418 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
6419 | */ |
6420 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 6421 | const struct sched_param *param) |
961ccddd | 6422 | { |
7479f3c9 | 6423 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
6424 | } |
6425 | ||
7318d4cc PZ |
6426 | /* |
6427 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
6428 | * incapable of resource management, which is the one thing an OS really should | |
6429 | * be doing. | |
6430 | * | |
6431 | * This is of course the reason it is limited to privileged users only. | |
6432 | * | |
6433 | * Worse still; it is fundamentally impossible to compose static priority | |
6434 | * workloads. You cannot take two correctly working static prio workloads | |
6435 | * and smash them together and still expect them to work. | |
6436 | * | |
6437 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
6438 | * | |
6439 | * MAX_RT_PRIO / 2 | |
6440 | * | |
6441 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
6442 | * know enough information to make a sensible choice. | |
6443 | */ | |
8b700983 | 6444 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
6445 | { |
6446 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 6447 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6448 | } |
6449 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
6450 | ||
6451 | /* | |
6452 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
6453 | */ | |
8b700983 | 6454 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
6455 | { |
6456 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 6457 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6458 | } |
6459 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
6460 | ||
8b700983 | 6461 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
6462 | { |
6463 | struct sched_attr attr = { | |
6464 | .sched_policy = SCHED_NORMAL, | |
6465 | .sched_nice = nice, | |
6466 | }; | |
8b700983 | 6467 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
6468 | } |
6469 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 6470 | |
95cdf3b7 IM |
6471 | static int |
6472 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6473 | { |
1da177e4 LT |
6474 | struct sched_param lparam; |
6475 | struct task_struct *p; | |
36c8b586 | 6476 | int retval; |
1da177e4 LT |
6477 | |
6478 | if (!param || pid < 0) | |
6479 | return -EINVAL; | |
6480 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6481 | return -EFAULT; | |
5fe1d75f ON |
6482 | |
6483 | rcu_read_lock(); | |
6484 | retval = -ESRCH; | |
1da177e4 | 6485 | p = find_process_by_pid(pid); |
710da3c8 JL |
6486 | if (likely(p)) |
6487 | get_task_struct(p); | |
5fe1d75f | 6488 | rcu_read_unlock(); |
36c8b586 | 6489 | |
710da3c8 JL |
6490 | if (likely(p)) { |
6491 | retval = sched_setscheduler(p, policy, &lparam); | |
6492 | put_task_struct(p); | |
6493 | } | |
6494 | ||
1da177e4 LT |
6495 | return retval; |
6496 | } | |
6497 | ||
d50dde5a DF |
6498 | /* |
6499 | * Mimics kernel/events/core.c perf_copy_attr(). | |
6500 | */ | |
d1ccc66d | 6501 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
6502 | { |
6503 | u32 size; | |
6504 | int ret; | |
6505 | ||
d1ccc66d | 6506 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
6507 | memset(attr, 0, sizeof(*attr)); |
6508 | ||
6509 | ret = get_user(size, &uattr->size); | |
6510 | if (ret) | |
6511 | return ret; | |
6512 | ||
d1ccc66d IM |
6513 | /* ABI compatibility quirk: */ |
6514 | if (!size) | |
d50dde5a | 6515 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 6516 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
6517 | goto err_size; |
6518 | ||
dff3a85f AS |
6519 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
6520 | if (ret) { | |
6521 | if (ret == -E2BIG) | |
6522 | goto err_size; | |
6523 | return ret; | |
d50dde5a DF |
6524 | } |
6525 | ||
a509a7cd PB |
6526 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
6527 | size < SCHED_ATTR_SIZE_VER1) | |
6528 | return -EINVAL; | |
6529 | ||
d50dde5a | 6530 | /* |
d1ccc66d | 6531 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
6532 | * to be strict and return an error on out-of-bounds values? |
6533 | */ | |
75e45d51 | 6534 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 6535 | |
e78c7bca | 6536 | return 0; |
d50dde5a DF |
6537 | |
6538 | err_size: | |
6539 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 6540 | return -E2BIG; |
d50dde5a DF |
6541 | } |
6542 | ||
1da177e4 LT |
6543 | /** |
6544 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6545 | * @pid: the pid in question. | |
6546 | * @policy: new policy. | |
6547 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6548 | * |
6549 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6550 | */ |
d1ccc66d | 6551 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 6552 | { |
c21761f1 JB |
6553 | if (policy < 0) |
6554 | return -EINVAL; | |
6555 | ||
1da177e4 LT |
6556 | return do_sched_setscheduler(pid, policy, param); |
6557 | } | |
6558 | ||
6559 | /** | |
6560 | * sys_sched_setparam - set/change the RT priority of a thread | |
6561 | * @pid: the pid in question. | |
6562 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6563 | * |
6564 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6565 | */ |
5add95d4 | 6566 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6567 | { |
c13db6b1 | 6568 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
6569 | } |
6570 | ||
d50dde5a DF |
6571 | /** |
6572 | * sys_sched_setattr - same as above, but with extended sched_attr | |
6573 | * @pid: the pid in question. | |
5778fccf | 6574 | * @uattr: structure containing the extended parameters. |
db66d756 | 6575 | * @flags: for future extension. |
d50dde5a | 6576 | */ |
6d35ab48 PZ |
6577 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
6578 | unsigned int, flags) | |
d50dde5a DF |
6579 | { |
6580 | struct sched_attr attr; | |
6581 | struct task_struct *p; | |
6582 | int retval; | |
6583 | ||
6d35ab48 | 6584 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
6585 | return -EINVAL; |
6586 | ||
143cf23d MK |
6587 | retval = sched_copy_attr(uattr, &attr); |
6588 | if (retval) | |
6589 | return retval; | |
d50dde5a | 6590 | |
b14ed2c2 | 6591 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 6592 | return -EINVAL; |
1d6362fa PB |
6593 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
6594 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
6595 | |
6596 | rcu_read_lock(); | |
6597 | retval = -ESRCH; | |
6598 | p = find_process_by_pid(pid); | |
a509a7cd PB |
6599 | if (likely(p)) |
6600 | get_task_struct(p); | |
d50dde5a DF |
6601 | rcu_read_unlock(); |
6602 | ||
a509a7cd PB |
6603 | if (likely(p)) { |
6604 | retval = sched_setattr(p, &attr); | |
6605 | put_task_struct(p); | |
6606 | } | |
6607 | ||
d50dde5a DF |
6608 | return retval; |
6609 | } | |
6610 | ||
1da177e4 LT |
6611 | /** |
6612 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6613 | * @pid: the pid in question. | |
e69f6186 YB |
6614 | * |
6615 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
6616 | * code. | |
1da177e4 | 6617 | */ |
5add95d4 | 6618 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6619 | { |
36c8b586 | 6620 | struct task_struct *p; |
3a5c359a | 6621 | int retval; |
1da177e4 LT |
6622 | |
6623 | if (pid < 0) | |
3a5c359a | 6624 | return -EINVAL; |
1da177e4 LT |
6625 | |
6626 | retval = -ESRCH; | |
5fe85be0 | 6627 | rcu_read_lock(); |
1da177e4 LT |
6628 | p = find_process_by_pid(pid); |
6629 | if (p) { | |
6630 | retval = security_task_getscheduler(p); | |
6631 | if (!retval) | |
ca94c442 LP |
6632 | retval = p->policy |
6633 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 6634 | } |
5fe85be0 | 6635 | rcu_read_unlock(); |
1da177e4 LT |
6636 | return retval; |
6637 | } | |
6638 | ||
6639 | /** | |
ca94c442 | 6640 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6641 | * @pid: the pid in question. |
6642 | * @param: structure containing the RT priority. | |
e69f6186 YB |
6643 | * |
6644 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
6645 | * code. | |
1da177e4 | 6646 | */ |
5add95d4 | 6647 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6648 | { |
ce5f7f82 | 6649 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 6650 | struct task_struct *p; |
3a5c359a | 6651 | int retval; |
1da177e4 LT |
6652 | |
6653 | if (!param || pid < 0) | |
3a5c359a | 6654 | return -EINVAL; |
1da177e4 | 6655 | |
5fe85be0 | 6656 | rcu_read_lock(); |
1da177e4 LT |
6657 | p = find_process_by_pid(pid); |
6658 | retval = -ESRCH; | |
6659 | if (!p) | |
6660 | goto out_unlock; | |
6661 | ||
6662 | retval = security_task_getscheduler(p); | |
6663 | if (retval) | |
6664 | goto out_unlock; | |
6665 | ||
ce5f7f82 PZ |
6666 | if (task_has_rt_policy(p)) |
6667 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 6668 | rcu_read_unlock(); |
1da177e4 LT |
6669 | |
6670 | /* | |
6671 | * This one might sleep, we cannot do it with a spinlock held ... | |
6672 | */ | |
6673 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6674 | ||
1da177e4 LT |
6675 | return retval; |
6676 | ||
6677 | out_unlock: | |
5fe85be0 | 6678 | rcu_read_unlock(); |
1da177e4 LT |
6679 | return retval; |
6680 | } | |
6681 | ||
1251201c IM |
6682 | /* |
6683 | * Copy the kernel size attribute structure (which might be larger | |
6684 | * than what user-space knows about) to user-space. | |
6685 | * | |
6686 | * Note that all cases are valid: user-space buffer can be larger or | |
6687 | * smaller than the kernel-space buffer. The usual case is that both | |
6688 | * have the same size. | |
6689 | */ | |
6690 | static int | |
6691 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
6692 | struct sched_attr *kattr, | |
6693 | unsigned int usize) | |
d50dde5a | 6694 | { |
1251201c | 6695 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 6696 | |
96d4f267 | 6697 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
6698 | return -EFAULT; |
6699 | ||
6700 | /* | |
1251201c IM |
6701 | * sched_getattr() ABI forwards and backwards compatibility: |
6702 | * | |
6703 | * If usize == ksize then we just copy everything to user-space and all is good. | |
6704 | * | |
6705 | * If usize < ksize then we only copy as much as user-space has space for, | |
6706 | * this keeps ABI compatibility as well. We skip the rest. | |
6707 | * | |
6708 | * If usize > ksize then user-space is using a newer version of the ABI, | |
6709 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
6710 | * detect the kernel's knowledge of attributes from the attr->size value | |
6711 | * which is set to ksize in this case. | |
d50dde5a | 6712 | */ |
1251201c | 6713 | kattr->size = min(usize, ksize); |
d50dde5a | 6714 | |
1251201c | 6715 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
6716 | return -EFAULT; |
6717 | ||
22400674 | 6718 | return 0; |
d50dde5a DF |
6719 | } |
6720 | ||
6721 | /** | |
aab03e05 | 6722 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 6723 | * @pid: the pid in question. |
5778fccf | 6724 | * @uattr: structure containing the extended parameters. |
dff3a85f | 6725 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 6726 | * @flags: for future extension. |
d50dde5a | 6727 | */ |
6d35ab48 | 6728 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 6729 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 6730 | { |
1251201c | 6731 | struct sched_attr kattr = { }; |
d50dde5a DF |
6732 | struct task_struct *p; |
6733 | int retval; | |
6734 | ||
1251201c IM |
6735 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
6736 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
6737 | return -EINVAL; |
6738 | ||
6739 | rcu_read_lock(); | |
6740 | p = find_process_by_pid(pid); | |
6741 | retval = -ESRCH; | |
6742 | if (!p) | |
6743 | goto out_unlock; | |
6744 | ||
6745 | retval = security_task_getscheduler(p); | |
6746 | if (retval) | |
6747 | goto out_unlock; | |
6748 | ||
1251201c | 6749 | kattr.sched_policy = p->policy; |
7479f3c9 | 6750 | if (p->sched_reset_on_fork) |
1251201c | 6751 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
aab03e05 | 6752 | if (task_has_dl_policy(p)) |
1251201c | 6753 | __getparam_dl(p, &kattr); |
aab03e05 | 6754 | else if (task_has_rt_policy(p)) |
1251201c | 6755 | kattr.sched_priority = p->rt_priority; |
d50dde5a | 6756 | else |
1251201c | 6757 | kattr.sched_nice = task_nice(p); |
d50dde5a | 6758 | |
a509a7cd | 6759 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
6760 | /* |
6761 | * This could race with another potential updater, but this is fine | |
6762 | * because it'll correctly read the old or the new value. We don't need | |
6763 | * to guarantee who wins the race as long as it doesn't return garbage. | |
6764 | */ | |
1251201c IM |
6765 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
6766 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
6767 | #endif |
6768 | ||
d50dde5a DF |
6769 | rcu_read_unlock(); |
6770 | ||
1251201c | 6771 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
6772 | |
6773 | out_unlock: | |
6774 | rcu_read_unlock(); | |
6775 | return retval; | |
6776 | } | |
6777 | ||
96f874e2 | 6778 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6779 | { |
5a16f3d3 | 6780 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6781 | struct task_struct *p; |
6782 | int retval; | |
1da177e4 | 6783 | |
23f5d142 | 6784 | rcu_read_lock(); |
1da177e4 LT |
6785 | |
6786 | p = find_process_by_pid(pid); | |
6787 | if (!p) { | |
23f5d142 | 6788 | rcu_read_unlock(); |
1da177e4 LT |
6789 | return -ESRCH; |
6790 | } | |
6791 | ||
23f5d142 | 6792 | /* Prevent p going away */ |
1da177e4 | 6793 | get_task_struct(p); |
23f5d142 | 6794 | rcu_read_unlock(); |
1da177e4 | 6795 | |
14a40ffc TH |
6796 | if (p->flags & PF_NO_SETAFFINITY) { |
6797 | retval = -EINVAL; | |
6798 | goto out_put_task; | |
6799 | } | |
5a16f3d3 RR |
6800 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6801 | retval = -ENOMEM; | |
6802 | goto out_put_task; | |
6803 | } | |
6804 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6805 | retval = -ENOMEM; | |
6806 | goto out_free_cpus_allowed; | |
6807 | } | |
1da177e4 | 6808 | retval = -EPERM; |
4c44aaaf EB |
6809 | if (!check_same_owner(p)) { |
6810 | rcu_read_lock(); | |
6811 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
6812 | rcu_read_unlock(); | |
16303ab2 | 6813 | goto out_free_new_mask; |
4c44aaaf EB |
6814 | } |
6815 | rcu_read_unlock(); | |
6816 | } | |
1da177e4 | 6817 | |
b0ae1981 | 6818 | retval = security_task_setscheduler(p); |
e7834f8f | 6819 | if (retval) |
16303ab2 | 6820 | goto out_free_new_mask; |
e7834f8f | 6821 | |
e4099a5e PZ |
6822 | |
6823 | cpuset_cpus_allowed(p, cpus_allowed); | |
6824 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
6825 | ||
332ac17e DF |
6826 | /* |
6827 | * Since bandwidth control happens on root_domain basis, | |
6828 | * if admission test is enabled, we only admit -deadline | |
6829 | * tasks allowed to run on all the CPUs in the task's | |
6830 | * root_domain. | |
6831 | */ | |
6832 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
6833 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
6834 | rcu_read_lock(); | |
6835 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 6836 | retval = -EBUSY; |
f1e3a093 | 6837 | rcu_read_unlock(); |
16303ab2 | 6838 | goto out_free_new_mask; |
332ac17e | 6839 | } |
f1e3a093 | 6840 | rcu_read_unlock(); |
332ac17e DF |
6841 | } |
6842 | #endif | |
49246274 | 6843 | again: |
9cfc3e18 | 6844 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK); |
1da177e4 | 6845 | |
8707d8b8 | 6846 | if (!retval) { |
5a16f3d3 RR |
6847 | cpuset_cpus_allowed(p, cpus_allowed); |
6848 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6849 | /* |
6850 | * We must have raced with a concurrent cpuset | |
6851 | * update. Just reset the cpus_allowed to the | |
6852 | * cpuset's cpus_allowed | |
6853 | */ | |
5a16f3d3 | 6854 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6855 | goto again; |
6856 | } | |
6857 | } | |
16303ab2 | 6858 | out_free_new_mask: |
5a16f3d3 RR |
6859 | free_cpumask_var(new_mask); |
6860 | out_free_cpus_allowed: | |
6861 | free_cpumask_var(cpus_allowed); | |
6862 | out_put_task: | |
1da177e4 | 6863 | put_task_struct(p); |
1da177e4 LT |
6864 | return retval; |
6865 | } | |
6866 | ||
6867 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6868 | struct cpumask *new_mask) |
1da177e4 | 6869 | { |
96f874e2 RR |
6870 | if (len < cpumask_size()) |
6871 | cpumask_clear(new_mask); | |
6872 | else if (len > cpumask_size()) | |
6873 | len = cpumask_size(); | |
6874 | ||
1da177e4 LT |
6875 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6876 | } | |
6877 | ||
6878 | /** | |
d1ccc66d | 6879 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
6880 | * @pid: pid of the process |
6881 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6882 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
6883 | * |
6884 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6885 | */ |
5add95d4 HC |
6886 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6887 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6888 | { |
5a16f3d3 | 6889 | cpumask_var_t new_mask; |
1da177e4 LT |
6890 | int retval; |
6891 | ||
5a16f3d3 RR |
6892 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6893 | return -ENOMEM; | |
1da177e4 | 6894 | |
5a16f3d3 RR |
6895 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6896 | if (retval == 0) | |
6897 | retval = sched_setaffinity(pid, new_mask); | |
6898 | free_cpumask_var(new_mask); | |
6899 | return retval; | |
1da177e4 LT |
6900 | } |
6901 | ||
96f874e2 | 6902 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6903 | { |
36c8b586 | 6904 | struct task_struct *p; |
31605683 | 6905 | unsigned long flags; |
1da177e4 | 6906 | int retval; |
1da177e4 | 6907 | |
23f5d142 | 6908 | rcu_read_lock(); |
1da177e4 LT |
6909 | |
6910 | retval = -ESRCH; | |
6911 | p = find_process_by_pid(pid); | |
6912 | if (!p) | |
6913 | goto out_unlock; | |
6914 | ||
e7834f8f DQ |
6915 | retval = security_task_getscheduler(p); |
6916 | if (retval) | |
6917 | goto out_unlock; | |
6918 | ||
013fdb80 | 6919 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 6920 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 6921 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
6922 | |
6923 | out_unlock: | |
23f5d142 | 6924 | rcu_read_unlock(); |
1da177e4 | 6925 | |
9531b62f | 6926 | return retval; |
1da177e4 LT |
6927 | } |
6928 | ||
6929 | /** | |
d1ccc66d | 6930 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
6931 | * @pid: pid of the process |
6932 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6933 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 6934 | * |
599b4840 ZW |
6935 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
6936 | * error code otherwise. | |
1da177e4 | 6937 | */ |
5add95d4 HC |
6938 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6939 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6940 | { |
6941 | int ret; | |
f17c8607 | 6942 | cpumask_var_t mask; |
1da177e4 | 6943 | |
84fba5ec | 6944 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
6945 | return -EINVAL; |
6946 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
6947 | return -EINVAL; |
6948 | ||
f17c8607 RR |
6949 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6950 | return -ENOMEM; | |
1da177e4 | 6951 | |
f17c8607 RR |
6952 | ret = sched_getaffinity(pid, mask); |
6953 | if (ret == 0) { | |
4de373a1 | 6954 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
6955 | |
6956 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
6957 | ret = -EFAULT; |
6958 | else | |
cd3d8031 | 6959 | ret = retlen; |
f17c8607 RR |
6960 | } |
6961 | free_cpumask_var(mask); | |
1da177e4 | 6962 | |
f17c8607 | 6963 | return ret; |
1da177e4 LT |
6964 | } |
6965 | ||
7d4dd4f1 | 6966 | static void do_sched_yield(void) |
1da177e4 | 6967 | { |
8a8c69c3 PZ |
6968 | struct rq_flags rf; |
6969 | struct rq *rq; | |
6970 | ||
246b3b33 | 6971 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 6972 | |
ae92882e | 6973 | schedstat_inc(rq->yld_count); |
4530d7ab | 6974 | current->sched_class->yield_task(rq); |
1da177e4 | 6975 | |
8a8c69c3 | 6976 | preempt_disable(); |
345a957f | 6977 | rq_unlock_irq(rq, &rf); |
ba74c144 | 6978 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
6979 | |
6980 | schedule(); | |
7d4dd4f1 | 6981 | } |
1da177e4 | 6982 | |
59a74b15 MCC |
6983 | /** |
6984 | * sys_sched_yield - yield the current processor to other threads. | |
6985 | * | |
6986 | * This function yields the current CPU to other tasks. If there are no | |
6987 | * other threads running on this CPU then this function will return. | |
6988 | * | |
6989 | * Return: 0. | |
6990 | */ | |
7d4dd4f1 DB |
6991 | SYSCALL_DEFINE0(sched_yield) |
6992 | { | |
6993 | do_sched_yield(); | |
1da177e4 LT |
6994 | return 0; |
6995 | } | |
6996 | ||
b965f1dd PZI |
6997 | #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) |
6998 | int __sched __cond_resched(void) | |
1da177e4 | 6999 | { |
fe32d3cd | 7000 | if (should_resched(0)) { |
a18b5d01 | 7001 | preempt_schedule_common(); |
1da177e4 LT |
7002 | return 1; |
7003 | } | |
b965f1dd | 7004 | #ifndef CONFIG_PREEMPT_RCU |
f79c3ad6 | 7005 | rcu_all_qs(); |
b965f1dd | 7006 | #endif |
1da177e4 LT |
7007 | return 0; |
7008 | } | |
b965f1dd PZI |
7009 | EXPORT_SYMBOL(__cond_resched); |
7010 | #endif | |
7011 | ||
7012 | #ifdef CONFIG_PREEMPT_DYNAMIC | |
7013 | DEFINE_STATIC_CALL_RET0(cond_resched, __cond_resched); | |
ef72661e | 7014 | EXPORT_STATIC_CALL_TRAMP(cond_resched); |
b965f1dd PZI |
7015 | |
7016 | DEFINE_STATIC_CALL_RET0(might_resched, __cond_resched); | |
ef72661e | 7017 | EXPORT_STATIC_CALL_TRAMP(might_resched); |
35a773a0 | 7018 | #endif |
1da177e4 LT |
7019 | |
7020 | /* | |
613afbf8 | 7021 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
7022 | * call schedule, and on return reacquire the lock. |
7023 | * | |
c1a280b6 | 7024 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
7025 | * operations here to prevent schedule() from being called twice (once via |
7026 | * spin_unlock(), once by hand). | |
7027 | */ | |
613afbf8 | 7028 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 7029 | { |
fe32d3cd | 7030 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
7031 | int ret = 0; |
7032 | ||
f607c668 PZ |
7033 | lockdep_assert_held(lock); |
7034 | ||
4a81e832 | 7035 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 7036 | spin_unlock(lock); |
d86ee480 | 7037 | if (resched) |
a18b5d01 | 7038 | preempt_schedule_common(); |
95c354fe NP |
7039 | else |
7040 | cpu_relax(); | |
6df3cecb | 7041 | ret = 1; |
1da177e4 | 7042 | spin_lock(lock); |
1da177e4 | 7043 | } |
6df3cecb | 7044 | return ret; |
1da177e4 | 7045 | } |
613afbf8 | 7046 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 7047 | |
f3d4b4b1 BG |
7048 | int __cond_resched_rwlock_read(rwlock_t *lock) |
7049 | { | |
7050 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
7051 | int ret = 0; | |
7052 | ||
7053 | lockdep_assert_held_read(lock); | |
7054 | ||
7055 | if (rwlock_needbreak(lock) || resched) { | |
7056 | read_unlock(lock); | |
7057 | if (resched) | |
7058 | preempt_schedule_common(); | |
7059 | else | |
7060 | cpu_relax(); | |
7061 | ret = 1; | |
7062 | read_lock(lock); | |
7063 | } | |
7064 | return ret; | |
7065 | } | |
7066 | EXPORT_SYMBOL(__cond_resched_rwlock_read); | |
7067 | ||
7068 | int __cond_resched_rwlock_write(rwlock_t *lock) | |
7069 | { | |
7070 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
7071 | int ret = 0; | |
7072 | ||
7073 | lockdep_assert_held_write(lock); | |
7074 | ||
7075 | if (rwlock_needbreak(lock) || resched) { | |
7076 | write_unlock(lock); | |
7077 | if (resched) | |
7078 | preempt_schedule_common(); | |
7079 | else | |
7080 | cpu_relax(); | |
7081 | ret = 1; | |
7082 | write_lock(lock); | |
7083 | } | |
7084 | return ret; | |
7085 | } | |
7086 | EXPORT_SYMBOL(__cond_resched_rwlock_write); | |
7087 | ||
1da177e4 LT |
7088 | /** |
7089 | * yield - yield the current processor to other threads. | |
7090 | * | |
8e3fabfd PZ |
7091 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
7092 | * | |
7093 | * The scheduler is at all times free to pick the calling task as the most | |
7094 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 7095 | * it, it's already broken. |
8e3fabfd PZ |
7096 | * |
7097 | * Typical broken usage is: | |
7098 | * | |
7099 | * while (!event) | |
d1ccc66d | 7100 | * yield(); |
8e3fabfd PZ |
7101 | * |
7102 | * where one assumes that yield() will let 'the other' process run that will | |
7103 | * make event true. If the current task is a SCHED_FIFO task that will never | |
7104 | * happen. Never use yield() as a progress guarantee!! | |
7105 | * | |
7106 | * If you want to use yield() to wait for something, use wait_event(). | |
7107 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
7108 | * If you still want to use yield(), do not! | |
1da177e4 LT |
7109 | */ |
7110 | void __sched yield(void) | |
7111 | { | |
7112 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 7113 | do_sched_yield(); |
1da177e4 | 7114 | } |
1da177e4 LT |
7115 | EXPORT_SYMBOL(yield); |
7116 | ||
d95f4122 MG |
7117 | /** |
7118 | * yield_to - yield the current processor to another thread in | |
7119 | * your thread group, or accelerate that thread toward the | |
7120 | * processor it's on. | |
16addf95 RD |
7121 | * @p: target task |
7122 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
7123 | * |
7124 | * It's the caller's job to ensure that the target task struct | |
7125 | * can't go away on us before we can do any checks. | |
7126 | * | |
e69f6186 | 7127 | * Return: |
7b270f60 PZ |
7128 | * true (>0) if we indeed boosted the target task. |
7129 | * false (0) if we failed to boost the target. | |
7130 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 7131 | */ |
fa93384f | 7132 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
7133 | { |
7134 | struct task_struct *curr = current; | |
7135 | struct rq *rq, *p_rq; | |
7136 | unsigned long flags; | |
c3c18640 | 7137 | int yielded = 0; |
d95f4122 MG |
7138 | |
7139 | local_irq_save(flags); | |
7140 | rq = this_rq(); | |
7141 | ||
7142 | again: | |
7143 | p_rq = task_rq(p); | |
7b270f60 PZ |
7144 | /* |
7145 | * If we're the only runnable task on the rq and target rq also | |
7146 | * has only one task, there's absolutely no point in yielding. | |
7147 | */ | |
7148 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
7149 | yielded = -ESRCH; | |
7150 | goto out_irq; | |
7151 | } | |
7152 | ||
d95f4122 | 7153 | double_rq_lock(rq, p_rq); |
39e24d8f | 7154 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
7155 | double_rq_unlock(rq, p_rq); |
7156 | goto again; | |
7157 | } | |
7158 | ||
7159 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 7160 | goto out_unlock; |
d95f4122 MG |
7161 | |
7162 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 7163 | goto out_unlock; |
d95f4122 MG |
7164 | |
7165 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 7166 | goto out_unlock; |
d95f4122 | 7167 | |
0900acf2 | 7168 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 7169 | if (yielded) { |
ae92882e | 7170 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
7171 | /* |
7172 | * Make p's CPU reschedule; pick_next_entity takes care of | |
7173 | * fairness. | |
7174 | */ | |
7175 | if (preempt && rq != p_rq) | |
8875125e | 7176 | resched_curr(p_rq); |
6d1cafd8 | 7177 | } |
d95f4122 | 7178 | |
7b270f60 | 7179 | out_unlock: |
d95f4122 | 7180 | double_rq_unlock(rq, p_rq); |
7b270f60 | 7181 | out_irq: |
d95f4122 MG |
7182 | local_irq_restore(flags); |
7183 | ||
7b270f60 | 7184 | if (yielded > 0) |
d95f4122 MG |
7185 | schedule(); |
7186 | ||
7187 | return yielded; | |
7188 | } | |
7189 | EXPORT_SYMBOL_GPL(yield_to); | |
7190 | ||
10ab5643 TH |
7191 | int io_schedule_prepare(void) |
7192 | { | |
7193 | int old_iowait = current->in_iowait; | |
7194 | ||
7195 | current->in_iowait = 1; | |
7196 | blk_schedule_flush_plug(current); | |
7197 | ||
7198 | return old_iowait; | |
7199 | } | |
7200 | ||
7201 | void io_schedule_finish(int token) | |
7202 | { | |
7203 | current->in_iowait = token; | |
7204 | } | |
7205 | ||
1da177e4 | 7206 | /* |
41a2d6cf | 7207 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 7208 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 7209 | */ |
1da177e4 LT |
7210 | long __sched io_schedule_timeout(long timeout) |
7211 | { | |
10ab5643 | 7212 | int token; |
1da177e4 LT |
7213 | long ret; |
7214 | ||
10ab5643 | 7215 | token = io_schedule_prepare(); |
1da177e4 | 7216 | ret = schedule_timeout(timeout); |
10ab5643 | 7217 | io_schedule_finish(token); |
9cff8ade | 7218 | |
1da177e4 LT |
7219 | return ret; |
7220 | } | |
9cff8ade | 7221 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 7222 | |
e3b929b0 | 7223 | void __sched io_schedule(void) |
10ab5643 TH |
7224 | { |
7225 | int token; | |
7226 | ||
7227 | token = io_schedule_prepare(); | |
7228 | schedule(); | |
7229 | io_schedule_finish(token); | |
7230 | } | |
7231 | EXPORT_SYMBOL(io_schedule); | |
7232 | ||
1da177e4 LT |
7233 | /** |
7234 | * sys_sched_get_priority_max - return maximum RT priority. | |
7235 | * @policy: scheduling class. | |
7236 | * | |
e69f6186 YB |
7237 | * Return: On success, this syscall returns the maximum |
7238 | * rt_priority that can be used by a given scheduling class. | |
7239 | * On failure, a negative error code is returned. | |
1da177e4 | 7240 | */ |
5add95d4 | 7241 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
7242 | { |
7243 | int ret = -EINVAL; | |
7244 | ||
7245 | switch (policy) { | |
7246 | case SCHED_FIFO: | |
7247 | case SCHED_RR: | |
ae18ad28 | 7248 | ret = MAX_RT_PRIO-1; |
1da177e4 | 7249 | break; |
aab03e05 | 7250 | case SCHED_DEADLINE: |
1da177e4 | 7251 | case SCHED_NORMAL: |
b0a9499c | 7252 | case SCHED_BATCH: |
dd41f596 | 7253 | case SCHED_IDLE: |
1da177e4 LT |
7254 | ret = 0; |
7255 | break; | |
7256 | } | |
7257 | return ret; | |
7258 | } | |
7259 | ||
7260 | /** | |
7261 | * sys_sched_get_priority_min - return minimum RT priority. | |
7262 | * @policy: scheduling class. | |
7263 | * | |
e69f6186 YB |
7264 | * Return: On success, this syscall returns the minimum |
7265 | * rt_priority that can be used by a given scheduling class. | |
7266 | * On failure, a negative error code is returned. | |
1da177e4 | 7267 | */ |
5add95d4 | 7268 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
7269 | { |
7270 | int ret = -EINVAL; | |
7271 | ||
7272 | switch (policy) { | |
7273 | case SCHED_FIFO: | |
7274 | case SCHED_RR: | |
7275 | ret = 1; | |
7276 | break; | |
aab03e05 | 7277 | case SCHED_DEADLINE: |
1da177e4 | 7278 | case SCHED_NORMAL: |
b0a9499c | 7279 | case SCHED_BATCH: |
dd41f596 | 7280 | case SCHED_IDLE: |
1da177e4 LT |
7281 | ret = 0; |
7282 | } | |
7283 | return ret; | |
7284 | } | |
7285 | ||
abca5fc5 | 7286 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 7287 | { |
36c8b586 | 7288 | struct task_struct *p; |
a4ec24b4 | 7289 | unsigned int time_slice; |
eb580751 | 7290 | struct rq_flags rf; |
dba091b9 | 7291 | struct rq *rq; |
3a5c359a | 7292 | int retval; |
1da177e4 LT |
7293 | |
7294 | if (pid < 0) | |
3a5c359a | 7295 | return -EINVAL; |
1da177e4 LT |
7296 | |
7297 | retval = -ESRCH; | |
1a551ae7 | 7298 | rcu_read_lock(); |
1da177e4 LT |
7299 | p = find_process_by_pid(pid); |
7300 | if (!p) | |
7301 | goto out_unlock; | |
7302 | ||
7303 | retval = security_task_getscheduler(p); | |
7304 | if (retval) | |
7305 | goto out_unlock; | |
7306 | ||
eb580751 | 7307 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
7308 | time_slice = 0; |
7309 | if (p->sched_class->get_rr_interval) | |
7310 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 7311 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 7312 | |
1a551ae7 | 7313 | rcu_read_unlock(); |
abca5fc5 AV |
7314 | jiffies_to_timespec64(time_slice, t); |
7315 | return 0; | |
3a5c359a | 7316 | |
1da177e4 | 7317 | out_unlock: |
1a551ae7 | 7318 | rcu_read_unlock(); |
1da177e4 LT |
7319 | return retval; |
7320 | } | |
7321 | ||
2064a5ab RD |
7322 | /** |
7323 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
7324 | * @pid: pid of the process. | |
7325 | * @interval: userspace pointer to the timeslice value. | |
7326 | * | |
7327 | * this syscall writes the default timeslice value of a given process | |
7328 | * into the user-space timespec buffer. A value of '0' means infinity. | |
7329 | * | |
7330 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
7331 | * an error code. | |
7332 | */ | |
abca5fc5 | 7333 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 7334 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
7335 | { |
7336 | struct timespec64 t; | |
7337 | int retval = sched_rr_get_interval(pid, &t); | |
7338 | ||
7339 | if (retval == 0) | |
7340 | retval = put_timespec64(&t, interval); | |
7341 | ||
7342 | return retval; | |
7343 | } | |
7344 | ||
474b9c77 | 7345 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
7346 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
7347 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
7348 | { |
7349 | struct timespec64 t; | |
7350 | int retval = sched_rr_get_interval(pid, &t); | |
7351 | ||
7352 | if (retval == 0) | |
9afc5eee | 7353 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
7354 | return retval; |
7355 | } | |
7356 | #endif | |
7357 | ||
82a1fcb9 | 7358 | void sched_show_task(struct task_struct *p) |
1da177e4 | 7359 | { |
1da177e4 | 7360 | unsigned long free = 0; |
4e79752c | 7361 | int ppid; |
c930b2c0 | 7362 | |
38200502 TH |
7363 | if (!try_get_task_stack(p)) |
7364 | return; | |
20435d84 | 7365 | |
cc172ff3 | 7366 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 XX |
7367 | |
7368 | if (p->state == TASK_RUNNING) | |
cc172ff3 | 7369 | pr_cont(" running task "); |
1da177e4 | 7370 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 7371 | free = stack_not_used(p); |
1da177e4 | 7372 | #endif |
a90e984c | 7373 | ppid = 0; |
4e79752c | 7374 | rcu_read_lock(); |
a90e984c ON |
7375 | if (pid_alive(p)) |
7376 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 7377 | rcu_read_unlock(); |
cc172ff3 LZ |
7378 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
7379 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 7380 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 7381 | |
3d1cb205 | 7382 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 7383 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 7384 | show_stack(p, NULL, KERN_INFO); |
38200502 | 7385 | put_task_stack(p); |
1da177e4 | 7386 | } |
0032f4e8 | 7387 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 7388 | |
5d68cc95 PZ |
7389 | static inline bool |
7390 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
7391 | { | |
7392 | /* no filter, everything matches */ | |
7393 | if (!state_filter) | |
7394 | return true; | |
7395 | ||
7396 | /* filter, but doesn't match */ | |
7397 | if (!(p->state & state_filter)) | |
7398 | return false; | |
7399 | ||
7400 | /* | |
7401 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
7402 | * TASK_KILLABLE). | |
7403 | */ | |
7404 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
7405 | return false; | |
7406 | ||
7407 | return true; | |
7408 | } | |
7409 | ||
7410 | ||
e59e2ae2 | 7411 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 7412 | { |
36c8b586 | 7413 | struct task_struct *g, *p; |
1da177e4 | 7414 | |
510f5acc | 7415 | rcu_read_lock(); |
5d07f420 | 7416 | for_each_process_thread(g, p) { |
1da177e4 LT |
7417 | /* |
7418 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 7419 | * console might take a lot of time: |
57675cb9 AR |
7420 | * Also, reset softlockup watchdogs on all CPUs, because |
7421 | * another CPU might be blocked waiting for us to process | |
7422 | * an IPI. | |
1da177e4 LT |
7423 | */ |
7424 | touch_nmi_watchdog(); | |
57675cb9 | 7425 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 7426 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 7427 | sched_show_task(p); |
5d07f420 | 7428 | } |
1da177e4 | 7429 | |
dd41f596 | 7430 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
7431 | if (!state_filter) |
7432 | sysrq_sched_debug_show(); | |
dd41f596 | 7433 | #endif |
510f5acc | 7434 | rcu_read_unlock(); |
e59e2ae2 IM |
7435 | /* |
7436 | * Only show locks if all tasks are dumped: | |
7437 | */ | |
93335a21 | 7438 | if (!state_filter) |
e59e2ae2 | 7439 | debug_show_all_locks(); |
1da177e4 LT |
7440 | } |
7441 | ||
f340c0d1 IM |
7442 | /** |
7443 | * init_idle - set up an idle thread for a given CPU | |
7444 | * @idle: task in question | |
d1ccc66d | 7445 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
7446 | * |
7447 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
7448 | * flag, to make booting more robust. | |
7449 | */ | |
0db0628d | 7450 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 7451 | { |
70b97a7f | 7452 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
7453 | unsigned long flags; |
7454 | ||
ff51ff84 PZ |
7455 | __sched_fork(0, idle); |
7456 | ||
25834c73 PZ |
7457 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
7458 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 7459 | |
06b83b5f | 7460 | idle->state = TASK_RUNNING; |
dd41f596 | 7461 | idle->se.exec_start = sched_clock(); |
c1de45ca | 7462 | idle->flags |= PF_IDLE; |
dd41f596 | 7463 | |
d08b9f0c | 7464 | scs_task_reset(idle); |
e1b77c92 MR |
7465 | kasan_unpoison_task_stack(idle); |
7466 | ||
de9b8f5d PZ |
7467 | #ifdef CONFIG_SMP |
7468 | /* | |
b19a888c | 7469 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
7470 | * in that case do_set_cpus_allowed() will not do the right thing. |
7471 | * | |
7472 | * And since this is boot we can forgo the serialization. | |
7473 | */ | |
9cfc3e18 | 7474 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 7475 | #endif |
6506cf6c PZ |
7476 | /* |
7477 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 7478 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
7479 | * lockdep check in task_group() will fail. |
7480 | * | |
7481 | * Similar case to sched_fork(). / Alternatively we could | |
7482 | * use task_rq_lock() here and obtain the other rq->lock. | |
7483 | * | |
7484 | * Silence PROVE_RCU | |
7485 | */ | |
7486 | rcu_read_lock(); | |
dd41f596 | 7487 | __set_task_cpu(idle, cpu); |
6506cf6c | 7488 | rcu_read_unlock(); |
1da177e4 | 7489 | |
5311a98f EB |
7490 | rq->idle = idle; |
7491 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 7492 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 7493 | #ifdef CONFIG_SMP |
3ca7a440 | 7494 | idle->on_cpu = 1; |
4866cde0 | 7495 | #endif |
25834c73 PZ |
7496 | raw_spin_unlock(&rq->lock); |
7497 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
7498 | |
7499 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 7500 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 7501 | |
dd41f596 IM |
7502 | /* |
7503 | * The idle tasks have their own, simple scheduling class: | |
7504 | */ | |
7505 | idle->sched_class = &idle_sched_class; | |
868baf07 | 7506 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 7507 | vtime_init_idle(idle, cpu); |
de9b8f5d | 7508 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
7509 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
7510 | #endif | |
19978ca6 IM |
7511 | } |
7512 | ||
e1d4eeec NP |
7513 | #ifdef CONFIG_SMP |
7514 | ||
f82f8042 JL |
7515 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
7516 | const struct cpumask *trial) | |
7517 | { | |
06a76fe0 | 7518 | int ret = 1; |
f82f8042 | 7519 | |
bb2bc55a MG |
7520 | if (!cpumask_weight(cur)) |
7521 | return ret; | |
7522 | ||
06a76fe0 | 7523 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
7524 | |
7525 | return ret; | |
7526 | } | |
7527 | ||
7f51412a JL |
7528 | int task_can_attach(struct task_struct *p, |
7529 | const struct cpumask *cs_cpus_allowed) | |
7530 | { | |
7531 | int ret = 0; | |
7532 | ||
7533 | /* | |
7534 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 7535 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
7536 | * affinity and isolating such threads by their set of |
7537 | * allowed nodes is unnecessary. Thus, cpusets are not | |
7538 | * applicable for such threads. This prevents checking for | |
7539 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 7540 | * before cpus_mask may be changed. |
7f51412a JL |
7541 | */ |
7542 | if (p->flags & PF_NO_SETAFFINITY) { | |
7543 | ret = -EINVAL; | |
7544 | goto out; | |
7545 | } | |
7546 | ||
7f51412a | 7547 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
7548 | cs_cpus_allowed)) |
7549 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 7550 | |
7f51412a JL |
7551 | out: |
7552 | return ret; | |
7553 | } | |
7554 | ||
f2cb1360 | 7555 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 7556 | |
e6628d5b MG |
7557 | #ifdef CONFIG_NUMA_BALANCING |
7558 | /* Migrate current task p to target_cpu */ | |
7559 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
7560 | { | |
7561 | struct migration_arg arg = { p, target_cpu }; | |
7562 | int curr_cpu = task_cpu(p); | |
7563 | ||
7564 | if (curr_cpu == target_cpu) | |
7565 | return 0; | |
7566 | ||
3bd37062 | 7567 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
7568 | return -EINVAL; |
7569 | ||
7570 | /* TODO: This is not properly updating schedstats */ | |
7571 | ||
286549dc | 7572 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
7573 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
7574 | } | |
0ec8aa00 PZ |
7575 | |
7576 | /* | |
7577 | * Requeue a task on a given node and accurately track the number of NUMA | |
7578 | * tasks on the runqueues | |
7579 | */ | |
7580 | void sched_setnuma(struct task_struct *p, int nid) | |
7581 | { | |
da0c1e65 | 7582 | bool queued, running; |
eb580751 PZ |
7583 | struct rq_flags rf; |
7584 | struct rq *rq; | |
0ec8aa00 | 7585 | |
eb580751 | 7586 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 7587 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
7588 | running = task_current(rq, p); |
7589 | ||
da0c1e65 | 7590 | if (queued) |
1de64443 | 7591 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 7592 | if (running) |
f3cd1c4e | 7593 | put_prev_task(rq, p); |
0ec8aa00 PZ |
7594 | |
7595 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 7596 | |
da0c1e65 | 7597 | if (queued) |
7134b3e9 | 7598 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 7599 | if (running) |
03b7fad1 | 7600 | set_next_task(rq, p); |
eb580751 | 7601 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 7602 | } |
5cc389bc | 7603 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 7604 | |
1da177e4 | 7605 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 7606 | /* |
d1ccc66d | 7607 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 7608 | * offline. |
054b9108 | 7609 | */ |
48c5ccae | 7610 | void idle_task_exit(void) |
1da177e4 | 7611 | { |
48c5ccae | 7612 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 7613 | |
48c5ccae | 7614 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 7615 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 7616 | |
a53efe5f | 7617 | if (mm != &init_mm) { |
252d2a41 | 7618 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
7619 | finish_arch_post_lock_switch(); |
7620 | } | |
bf2c59fc PZ |
7621 | |
7622 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
7623 | } |
7624 | ||
2558aacf | 7625 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 7626 | { |
2558aacf PZ |
7627 | struct task_struct *p = arg; |
7628 | struct rq *rq = this_rq(); | |
7629 | struct rq_flags rf; | |
7630 | int cpu; | |
1da177e4 | 7631 | |
2558aacf PZ |
7632 | raw_spin_lock_irq(&p->pi_lock); |
7633 | rq_lock(rq, &rf); | |
3f1d2a31 | 7634 | |
2558aacf PZ |
7635 | update_rq_clock(rq); |
7636 | ||
7637 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
7638 | cpu = select_fallback_rq(rq->cpu, p); | |
7639 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 7640 | } |
3f1d2a31 | 7641 | |
2558aacf PZ |
7642 | rq_unlock(rq, &rf); |
7643 | raw_spin_unlock_irq(&p->pi_lock); | |
7644 | ||
7645 | put_task_struct(p); | |
7646 | ||
7647 | return 0; | |
10e7071b | 7648 | } |
3f1d2a31 | 7649 | |
2558aacf PZ |
7650 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
7651 | ||
48f24c4d | 7652 | /* |
2558aacf | 7653 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
1da177e4 | 7654 | */ |
2558aacf | 7655 | static void balance_push(struct rq *rq) |
1da177e4 | 7656 | { |
2558aacf PZ |
7657 | struct task_struct *push_task = rq->curr; |
7658 | ||
7659 | lockdep_assert_held(&rq->lock); | |
7660 | SCHED_WARN_ON(rq->cpu != smp_processor_id()); | |
ae792702 PZ |
7661 | /* |
7662 | * Ensure the thing is persistent until balance_push_set(.on = false); | |
7663 | */ | |
7664 | rq->balance_callback = &balance_push_callback; | |
1da177e4 LT |
7665 | |
7666 | /* | |
2558aacf PZ |
7667 | * Both the cpu-hotplug and stop task are in this case and are |
7668 | * required to complete the hotplug process. | |
5ba2ffba PZ |
7669 | * |
7670 | * XXX: the idle task does not match kthread_is_per_cpu() due to | |
7671 | * histerical raisins. | |
1da177e4 | 7672 | */ |
5ba2ffba PZ |
7673 | if (rq->idle == push_task || |
7674 | ((push_task->flags & PF_KTHREAD) && kthread_is_per_cpu(push_task)) || | |
7675 | is_migration_disabled(push_task)) { | |
7676 | ||
f2469a1f TG |
7677 | /* |
7678 | * If this is the idle task on the outgoing CPU try to wake | |
7679 | * up the hotplug control thread which might wait for the | |
7680 | * last task to vanish. The rcuwait_active() check is | |
7681 | * accurate here because the waiter is pinned on this CPU | |
7682 | * and can't obviously be running in parallel. | |
3015ef4b TG |
7683 | * |
7684 | * On RT kernels this also has to check whether there are | |
7685 | * pinned and scheduled out tasks on the runqueue. They | |
7686 | * need to leave the migrate disabled section first. | |
f2469a1f | 7687 | */ |
3015ef4b TG |
7688 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
7689 | rcuwait_active(&rq->hotplug_wait)) { | |
f2469a1f TG |
7690 | raw_spin_unlock(&rq->lock); |
7691 | rcuwait_wake_up(&rq->hotplug_wait); | |
7692 | raw_spin_lock(&rq->lock); | |
7693 | } | |
2558aacf | 7694 | return; |
f2469a1f | 7695 | } |
48f24c4d | 7696 | |
2558aacf | 7697 | get_task_struct(push_task); |
77bd3970 | 7698 | /* |
2558aacf PZ |
7699 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
7700 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 7701 | */ |
2558aacf PZ |
7702 | raw_spin_unlock(&rq->lock); |
7703 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, | |
7704 | this_cpu_ptr(&push_work)); | |
7705 | /* | |
7706 | * At this point need_resched() is true and we'll take the loop in | |
7707 | * schedule(). The next pick is obviously going to be the stop task | |
5ba2ffba | 7708 | * which kthread_is_per_cpu() and will push this task away. |
2558aacf PZ |
7709 | */ |
7710 | raw_spin_lock(&rq->lock); | |
7711 | } | |
77bd3970 | 7712 | |
2558aacf PZ |
7713 | static void balance_push_set(int cpu, bool on) |
7714 | { | |
7715 | struct rq *rq = cpu_rq(cpu); | |
7716 | struct rq_flags rf; | |
48c5ccae | 7717 | |
2558aacf | 7718 | rq_lock_irqsave(rq, &rf); |
975707f2 | 7719 | rq->balance_push = on; |
22f667c9 PZ |
7720 | if (on) { |
7721 | WARN_ON_ONCE(rq->balance_callback); | |
ae792702 | 7722 | rq->balance_callback = &balance_push_callback; |
22f667c9 | 7723 | } else if (rq->balance_callback == &balance_push_callback) { |
ae792702 | 7724 | rq->balance_callback = NULL; |
22f667c9 | 7725 | } |
2558aacf PZ |
7726 | rq_unlock_irqrestore(rq, &rf); |
7727 | } | |
e692ab53 | 7728 | |
f2469a1f TG |
7729 | /* |
7730 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
7731 | * inactive. All tasks which are not per CPU kernel threads are either | |
7732 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
7733 | * during wakeup. Wait until the CPU is quiescent. | |
7734 | */ | |
7735 | static void balance_hotplug_wait(void) | |
7736 | { | |
7737 | struct rq *rq = this_rq(); | |
5473e0cc | 7738 | |
3015ef4b TG |
7739 | rcuwait_wait_event(&rq->hotplug_wait, |
7740 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
7741 | TASK_UNINTERRUPTIBLE); |
7742 | } | |
5473e0cc | 7743 | |
2558aacf | 7744 | #else |
dce48a84 | 7745 | |
2558aacf PZ |
7746 | static inline void balance_push(struct rq *rq) |
7747 | { | |
dce48a84 | 7748 | } |
dce48a84 | 7749 | |
2558aacf PZ |
7750 | static inline void balance_push_set(int cpu, bool on) |
7751 | { | |
7752 | } | |
7753 | ||
f2469a1f TG |
7754 | static inline void balance_hotplug_wait(void) |
7755 | { | |
dce48a84 | 7756 | } |
f2469a1f | 7757 | |
1da177e4 LT |
7758 | #endif /* CONFIG_HOTPLUG_CPU */ |
7759 | ||
f2cb1360 | 7760 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
7761 | { |
7762 | if (!rq->online) { | |
7763 | const struct sched_class *class; | |
7764 | ||
c6c4927b | 7765 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7766 | rq->online = 1; |
7767 | ||
7768 | for_each_class(class) { | |
7769 | if (class->rq_online) | |
7770 | class->rq_online(rq); | |
7771 | } | |
7772 | } | |
7773 | } | |
7774 | ||
f2cb1360 | 7775 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
7776 | { |
7777 | if (rq->online) { | |
7778 | const struct sched_class *class; | |
7779 | ||
7780 | for_each_class(class) { | |
7781 | if (class->rq_offline) | |
7782 | class->rq_offline(rq); | |
7783 | } | |
7784 | ||
c6c4927b | 7785 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7786 | rq->online = 0; |
7787 | } | |
7788 | } | |
7789 | ||
d1ccc66d IM |
7790 | /* |
7791 | * used to mark begin/end of suspend/resume: | |
7792 | */ | |
7793 | static int num_cpus_frozen; | |
d35be8ba | 7794 | |
1da177e4 | 7795 | /* |
3a101d05 TH |
7796 | * Update cpusets according to cpu_active mask. If cpusets are |
7797 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7798 | * around partition_sched_domains(). | |
d35be8ba SB |
7799 | * |
7800 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7801 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7802 | */ |
40190a78 | 7803 | static void cpuset_cpu_active(void) |
e761b772 | 7804 | { |
40190a78 | 7805 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7806 | /* |
7807 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7808 | * resume sequence. As long as this is not the last online | |
7809 | * operation in the resume sequence, just build a single sched | |
7810 | * domain, ignoring cpusets. | |
7811 | */ | |
50e76632 PZ |
7812 | partition_sched_domains(1, NULL, NULL); |
7813 | if (--num_cpus_frozen) | |
135fb3e1 | 7814 | return; |
d35be8ba SB |
7815 | /* |
7816 | * This is the last CPU online operation. So fall through and | |
7817 | * restore the original sched domains by considering the | |
7818 | * cpuset configurations. | |
7819 | */ | |
50e76632 | 7820 | cpuset_force_rebuild(); |
3a101d05 | 7821 | } |
30e03acd | 7822 | cpuset_update_active_cpus(); |
3a101d05 | 7823 | } |
e761b772 | 7824 | |
40190a78 | 7825 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7826 | { |
40190a78 | 7827 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 7828 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 7829 | return -EBUSY; |
30e03acd | 7830 | cpuset_update_active_cpus(); |
135fb3e1 | 7831 | } else { |
d35be8ba SB |
7832 | num_cpus_frozen++; |
7833 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7834 | } |
135fb3e1 | 7835 | return 0; |
e761b772 | 7836 | } |
e761b772 | 7837 | |
40190a78 | 7838 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7839 | { |
7d976699 | 7840 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 7841 | struct rq_flags rf; |
7d976699 | 7842 | |
22f667c9 PZ |
7843 | /* |
7844 | * Make sure that when the hotplug state machine does a roll-back | |
7845 | * we clear balance_push. Ideally that would happen earlier... | |
7846 | */ | |
2558aacf PZ |
7847 | balance_push_set(cpu, false); |
7848 | ||
ba2591a5 PZ |
7849 | #ifdef CONFIG_SCHED_SMT |
7850 | /* | |
c5511d03 | 7851 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 7852 | */ |
c5511d03 PZI |
7853 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
7854 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 7855 | #endif |
40190a78 | 7856 | set_cpu_active(cpu, true); |
135fb3e1 | 7857 | |
40190a78 | 7858 | if (sched_smp_initialized) { |
135fb3e1 | 7859 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7860 | cpuset_cpu_active(); |
e761b772 | 7861 | } |
7d976699 TG |
7862 | |
7863 | /* | |
7864 | * Put the rq online, if not already. This happens: | |
7865 | * | |
7866 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 7867 | * after all CPUs have been brought up. |
7d976699 TG |
7868 | * |
7869 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7870 | * domains. | |
7871 | */ | |
8a8c69c3 | 7872 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
7873 | if (rq->rd) { |
7874 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7875 | set_rq_online(rq); | |
7876 | } | |
8a8c69c3 | 7877 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 7878 | |
40190a78 | 7879 | return 0; |
135fb3e1 TG |
7880 | } |
7881 | ||
40190a78 | 7882 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7883 | { |
120455c5 PZ |
7884 | struct rq *rq = cpu_rq(cpu); |
7885 | struct rq_flags rf; | |
135fb3e1 TG |
7886 | int ret; |
7887 | ||
e0b257c3 AMB |
7888 | /* |
7889 | * Remove CPU from nohz.idle_cpus_mask to prevent participating in | |
7890 | * load balancing when not active | |
7891 | */ | |
7892 | nohz_balance_exit_idle(rq); | |
7893 | ||
40190a78 | 7894 | set_cpu_active(cpu, false); |
741ba80f PZ |
7895 | |
7896 | /* | |
7897 | * From this point forward, this CPU will refuse to run any task that | |
7898 | * is not: migrate_disable() or KTHREAD_IS_PER_CPU, and will actively | |
7899 | * push those tasks away until this gets cleared, see | |
7900 | * sched_cpu_dying(). | |
7901 | */ | |
975707f2 PZ |
7902 | balance_push_set(cpu, true); |
7903 | ||
b2454caa | 7904 | /* |
975707f2 PZ |
7905 | * We've cleared cpu_active_mask / set balance_push, wait for all |
7906 | * preempt-disabled and RCU users of this state to go away such that | |
7907 | * all new such users will observe it. | |
b2454caa | 7908 | * |
5ba2ffba PZ |
7909 | * Specifically, we rely on ttwu to no longer target this CPU, see |
7910 | * ttwu_queue_cond() and is_cpu_allowed(). | |
7911 | * | |
b2454caa PZ |
7912 | * Do sync before park smpboot threads to take care the rcu boost case. |
7913 | */ | |
309ba859 | 7914 | synchronize_rcu(); |
40190a78 | 7915 | |
120455c5 PZ |
7916 | rq_lock_irqsave(rq, &rf); |
7917 | if (rq->rd) { | |
7918 | update_rq_clock(rq); | |
7919 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7920 | set_rq_offline(rq); | |
7921 | } | |
7922 | rq_unlock_irqrestore(rq, &rf); | |
7923 | ||
c5511d03 PZI |
7924 | #ifdef CONFIG_SCHED_SMT |
7925 | /* | |
7926 | * When going down, decrement the number of cores with SMT present. | |
7927 | */ | |
7928 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
7929 | static_branch_dec_cpuslocked(&sched_smt_present); | |
7930 | #endif | |
7931 | ||
40190a78 TG |
7932 | if (!sched_smp_initialized) |
7933 | return 0; | |
7934 | ||
7935 | ret = cpuset_cpu_inactive(cpu); | |
7936 | if (ret) { | |
2558aacf | 7937 | balance_push_set(cpu, false); |
40190a78 TG |
7938 | set_cpu_active(cpu, true); |
7939 | return ret; | |
135fb3e1 | 7940 | } |
40190a78 TG |
7941 | sched_domains_numa_masks_clear(cpu); |
7942 | return 0; | |
135fb3e1 TG |
7943 | } |
7944 | ||
94baf7a5 TG |
7945 | static void sched_rq_cpu_starting(unsigned int cpu) |
7946 | { | |
7947 | struct rq *rq = cpu_rq(cpu); | |
7948 | ||
7949 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7950 | update_max_interval(); |
7951 | } | |
7952 | ||
135fb3e1 TG |
7953 | int sched_cpu_starting(unsigned int cpu) |
7954 | { | |
94baf7a5 | 7955 | sched_rq_cpu_starting(cpu); |
d84b3131 | 7956 | sched_tick_start(cpu); |
135fb3e1 | 7957 | return 0; |
e761b772 | 7958 | } |
e761b772 | 7959 | |
f2785ddb | 7960 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
7961 | |
7962 | /* | |
7963 | * Invoked immediately before the stopper thread is invoked to bring the | |
7964 | * CPU down completely. At this point all per CPU kthreads except the | |
7965 | * hotplug thread (current) and the stopper thread (inactive) have been | |
7966 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
7967 | * any of those which might be on the way out are gone. | |
7968 | * | |
7969 | * If after this point a bound task is being woken on this CPU then the | |
7970 | * responsible hotplug callback has failed to do it's job. | |
7971 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
7972 | */ | |
7973 | int sched_cpu_wait_empty(unsigned int cpu) | |
7974 | { | |
7975 | balance_hotplug_wait(); | |
7976 | return 0; | |
7977 | } | |
7978 | ||
7979 | /* | |
7980 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
7981 | * might have. Called from the CPU stopper task after ensuring that the | |
7982 | * stopper is the last running task on the CPU, so nr_active count is | |
7983 | * stable. We need to take the teardown thread which is calling this into | |
7984 | * account, so we hand in adjust = 1 to the load calculation. | |
7985 | * | |
7986 | * Also see the comment "Global load-average calculations". | |
7987 | */ | |
7988 | static void calc_load_migrate(struct rq *rq) | |
7989 | { | |
7990 | long delta = calc_load_fold_active(rq, 1); | |
7991 | ||
7992 | if (delta) | |
7993 | atomic_long_add(delta, &calc_load_tasks); | |
7994 | } | |
7995 | ||
36c6e17b VS |
7996 | static void dump_rq_tasks(struct rq *rq, const char *loglvl) |
7997 | { | |
7998 | struct task_struct *g, *p; | |
7999 | int cpu = cpu_of(rq); | |
8000 | ||
8001 | lockdep_assert_held(&rq->lock); | |
8002 | ||
8003 | printk("%sCPU%d enqueued tasks (%u total):\n", loglvl, cpu, rq->nr_running); | |
8004 | for_each_process_thread(g, p) { | |
8005 | if (task_cpu(p) != cpu) | |
8006 | continue; | |
8007 | ||
8008 | if (!task_on_rq_queued(p)) | |
8009 | continue; | |
8010 | ||
8011 | printk("%s\tpid: %d, name: %s\n", loglvl, p->pid, p->comm); | |
8012 | } | |
8013 | } | |
8014 | ||
f2785ddb TG |
8015 | int sched_cpu_dying(unsigned int cpu) |
8016 | { | |
8017 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 8018 | struct rq_flags rf; |
f2785ddb TG |
8019 | |
8020 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 8021 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
8022 | |
8023 | rq_lock_irqsave(rq, &rf); | |
36c6e17b VS |
8024 | if (rq->nr_running != 1 || rq_has_pinned_tasks(rq)) { |
8025 | WARN(true, "Dying CPU not properly vacated!"); | |
8026 | dump_rq_tasks(rq, KERN_WARNING); | |
8027 | } | |
8a8c69c3 PZ |
8028 | rq_unlock_irqrestore(rq, &rf); |
8029 | ||
22f667c9 PZ |
8030 | /* |
8031 | * Now that the CPU is offline, make sure we're welcome | |
8032 | * to new tasks once we come back up. | |
8033 | */ | |
8034 | balance_push_set(cpu, false); | |
8035 | ||
f2785ddb TG |
8036 | calc_load_migrate(rq); |
8037 | update_max_interval(); | |
e5ef27d0 | 8038 | hrtick_clear(rq); |
f2785ddb TG |
8039 | return 0; |
8040 | } | |
8041 | #endif | |
8042 | ||
1da177e4 LT |
8043 | void __init sched_init_smp(void) |
8044 | { | |
cb83b629 PZ |
8045 | sched_init_numa(); |
8046 | ||
6acce3ef PZ |
8047 | /* |
8048 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 8049 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 8050 | * happen. |
6acce3ef | 8051 | */ |
712555ee | 8052 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 8053 | sched_init_domains(cpu_active_mask); |
712555ee | 8054 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 8055 | |
5c1e1767 | 8056 | /* Move init over to a non-isolated CPU */ |
edb93821 | 8057 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 8058 | BUG(); |
19978ca6 | 8059 | sched_init_granularity(); |
4212823f | 8060 | |
0e3900e6 | 8061 | init_sched_rt_class(); |
1baca4ce | 8062 | init_sched_dl_class(); |
1b568f0a | 8063 | |
e26fbffd | 8064 | sched_smp_initialized = true; |
1da177e4 | 8065 | } |
e26fbffd TG |
8066 | |
8067 | static int __init migration_init(void) | |
8068 | { | |
77a5352b | 8069 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 8070 | return 0; |
1da177e4 | 8071 | } |
e26fbffd TG |
8072 | early_initcall(migration_init); |
8073 | ||
1da177e4 LT |
8074 | #else |
8075 | void __init sched_init_smp(void) | |
8076 | { | |
19978ca6 | 8077 | sched_init_granularity(); |
1da177e4 LT |
8078 | } |
8079 | #endif /* CONFIG_SMP */ | |
8080 | ||
8081 | int in_sched_functions(unsigned long addr) | |
8082 | { | |
1da177e4 LT |
8083 | return in_lock_functions(addr) || |
8084 | (addr >= (unsigned long)__sched_text_start | |
8085 | && addr < (unsigned long)__sched_text_end); | |
8086 | } | |
8087 | ||
029632fb | 8088 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
8089 | /* |
8090 | * Default task group. | |
8091 | * Every task in system belongs to this group at bootup. | |
8092 | */ | |
029632fb | 8093 | struct task_group root_task_group; |
35cf4e50 | 8094 | LIST_HEAD(task_groups); |
b0367629 WL |
8095 | |
8096 | /* Cacheline aligned slab cache for task_group */ | |
8097 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 8098 | #endif |
6f505b16 | 8099 | |
e6252c3e | 8100 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 8101 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 8102 | |
1da177e4 LT |
8103 | void __init sched_init(void) |
8104 | { | |
a1dc0446 | 8105 | unsigned long ptr = 0; |
55627e3c | 8106 | int i; |
434d53b0 | 8107 | |
c3a340f7 SRV |
8108 | /* Make sure the linker didn't screw up */ |
8109 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
8110 | &fair_sched_class + 1 != &rt_sched_class || | |
8111 | &rt_sched_class + 1 != &dl_sched_class); | |
8112 | #ifdef CONFIG_SMP | |
8113 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
8114 | #endif | |
8115 | ||
5822a454 | 8116 | wait_bit_init(); |
9dcb8b68 | 8117 | |
434d53b0 | 8118 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 8119 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
8120 | #endif |
8121 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 8122 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 8123 | #endif |
a1dc0446 QC |
8124 | if (ptr) { |
8125 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
8126 | |
8127 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 8128 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
8129 | ptr += nr_cpu_ids * sizeof(void **); |
8130 | ||
07e06b01 | 8131 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 8132 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 8133 | |
b1d1779e WY |
8134 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
8135 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 8136 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 8137 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8138 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
8139 | ptr += nr_cpu_ids * sizeof(void **); |
8140 | ||
07e06b01 | 8141 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
8142 | ptr += nr_cpu_ids * sizeof(void **); |
8143 | ||
6d6bc0ad | 8144 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 8145 | } |
df7c8e84 | 8146 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
8147 | for_each_possible_cpu(i) { |
8148 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
8149 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
8150 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
8151 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 8152 | } |
b74e6278 | 8153 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 8154 | |
d1ccc66d IM |
8155 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
8156 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 8157 | |
57d885fe GH |
8158 | #ifdef CONFIG_SMP |
8159 | init_defrootdomain(); | |
8160 | #endif | |
8161 | ||
d0b27fa7 | 8162 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8163 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 8164 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 8165 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8166 | |
7c941438 | 8167 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
8168 | task_group_cache = KMEM_CACHE(task_group, 0); |
8169 | ||
07e06b01 YZ |
8170 | list_add(&root_task_group.list, &task_groups); |
8171 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 8172 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 8173 | autogroup_init(&init_task); |
7c941438 | 8174 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8175 | |
0a945022 | 8176 | for_each_possible_cpu(i) { |
70b97a7f | 8177 | struct rq *rq; |
1da177e4 LT |
8178 | |
8179 | rq = cpu_rq(i); | |
05fa785c | 8180 | raw_spin_lock_init(&rq->lock); |
7897986b | 8181 | rq->nr_running = 0; |
dce48a84 TG |
8182 | rq->calc_load_active = 0; |
8183 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 8184 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
8185 | init_rt_rq(&rq->rt); |
8186 | init_dl_rq(&rq->dl); | |
dd41f596 | 8187 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 8188 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 8189 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 8190 | /* |
d1ccc66d | 8191 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
8192 | * |
8193 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
8194 | * gets 100% of the CPU resources in the system. This overall |
8195 | * system CPU resource is divided among the tasks of | |
07e06b01 | 8196 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
8197 | * based on each entity's (task or task-group's) weight |
8198 | * (se->load.weight). | |
8199 | * | |
07e06b01 | 8200 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 8201 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 8202 | * then A0's share of the CPU resource is: |
354d60c2 | 8203 | * |
0d905bca | 8204 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 8205 | * |
07e06b01 YZ |
8206 | * We achieve this by letting root_task_group's tasks sit |
8207 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 8208 | */ |
07e06b01 | 8209 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
8210 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8211 | ||
8212 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8213 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 8214 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 8215 | #endif |
1da177e4 | 8216 | #ifdef CONFIG_SMP |
41c7ce9a | 8217 | rq->sd = NULL; |
57d885fe | 8218 | rq->rd = NULL; |
ca6d75e6 | 8219 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 8220 | rq->balance_callback = NULL; |
1da177e4 | 8221 | rq->active_balance = 0; |
dd41f596 | 8222 | rq->next_balance = jiffies; |
1da177e4 | 8223 | rq->push_cpu = 0; |
0a2966b4 | 8224 | rq->cpu = i; |
1f11eb6a | 8225 | rq->online = 0; |
eae0c9df MG |
8226 | rq->idle_stamp = 0; |
8227 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 8228 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
8229 | |
8230 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
8231 | ||
dc938520 | 8232 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 8233 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 8234 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 8235 | atomic_set(&rq->nohz_flags, 0); |
90b5363a | 8236 | |
545b8c8d | 8237 | INIT_CSD(&rq->nohz_csd, nohz_csd_func, rq); |
83cd4fe2 | 8238 | #endif |
f2469a1f TG |
8239 | #ifdef CONFIG_HOTPLUG_CPU |
8240 | rcuwait_init(&rq->hotplug_wait); | |
83cd4fe2 | 8241 | #endif |
9fd81dd5 | 8242 | #endif /* CONFIG_SMP */ |
77a021be | 8243 | hrtick_rq_init(rq); |
1da177e4 | 8244 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8245 | } |
8246 | ||
9059393e | 8247 | set_load_weight(&init_task, false); |
b50f60ce | 8248 | |
1da177e4 LT |
8249 | /* |
8250 | * The boot idle thread does lazy MMU switching as well: | |
8251 | */ | |
f1f10076 | 8252 | mmgrab(&init_mm); |
1da177e4 LT |
8253 | enter_lazy_tlb(&init_mm, current); |
8254 | ||
8255 | /* | |
8256 | * Make us the idle thread. Technically, schedule() should not be | |
8257 | * called from this thread, however somewhere below it might be, | |
8258 | * but because we are the idle thread, we just pick up running again | |
8259 | * when this runqueue becomes "idle". | |
8260 | */ | |
8261 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8262 | |
8263 | calc_load_update = jiffies + LOAD_FREQ; | |
8264 | ||
bf4d83f6 | 8265 | #ifdef CONFIG_SMP |
29d5e047 | 8266 | idle_thread_set_boot_cpu(); |
029632fb PZ |
8267 | #endif |
8268 | init_sched_fair_class(); | |
6a7b3dc3 | 8269 | |
4698f88c JP |
8270 | init_schedstats(); |
8271 | ||
eb414681 JW |
8272 | psi_init(); |
8273 | ||
69842cba PB |
8274 | init_uclamp(); |
8275 | ||
6892b75e | 8276 | scheduler_running = 1; |
1da177e4 LT |
8277 | } |
8278 | ||
d902db1e | 8279 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
8280 | static inline int preempt_count_equals(int preempt_offset) |
8281 | { | |
da7142e2 | 8282 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 8283 | |
4ba8216c | 8284 | return (nested == preempt_offset); |
e4aafea2 FW |
8285 | } |
8286 | ||
d894837f | 8287 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8288 | { |
8eb23b9f PZ |
8289 | /* |
8290 | * Blocking primitives will set (and therefore destroy) current->state, | |
8291 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
8292 | * otherwise we will destroy state. | |
8293 | */ | |
00845eb9 | 8294 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
8295 | "do not call blocking ops when !TASK_RUNNING; " |
8296 | "state=%lx set at [<%p>] %pS\n", | |
8297 | current->state, | |
8298 | (void *)current->task_state_change, | |
00845eb9 | 8299 | (void *)current->task_state_change); |
8eb23b9f | 8300 | |
3427445a PZ |
8301 | ___might_sleep(file, line, preempt_offset); |
8302 | } | |
8303 | EXPORT_SYMBOL(__might_sleep); | |
8304 | ||
8305 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 8306 | { |
d1ccc66d IM |
8307 | /* Ratelimiting timestamp: */ |
8308 | static unsigned long prev_jiffy; | |
8309 | ||
d1c6d149 | 8310 | unsigned long preempt_disable_ip; |
1da177e4 | 8311 | |
d1ccc66d IM |
8312 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
8313 | rcu_sleep_check(); | |
8314 | ||
db273be2 | 8315 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 8316 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
8317 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
8318 | oops_in_progress) | |
aef745fc | 8319 | return; |
1c3c5eab | 8320 | |
aef745fc IM |
8321 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
8322 | return; | |
8323 | prev_jiffy = jiffies; | |
8324 | ||
d1ccc66d | 8325 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
8326 | preempt_disable_ip = get_preempt_disable_ip(current); |
8327 | ||
3df0fc5b PZ |
8328 | printk(KERN_ERR |
8329 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8330 | file, line); | |
8331 | printk(KERN_ERR | |
312364f3 DV |
8332 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
8333 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 8334 | current->pid, current->comm); |
aef745fc | 8335 | |
a8b686b3 ES |
8336 | if (task_stack_end_corrupted(current)) |
8337 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
8338 | ||
aef745fc IM |
8339 | debug_show_held_locks(current); |
8340 | if (irqs_disabled()) | |
8341 | print_irqtrace_events(current); | |
d1c6d149 VN |
8342 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
8343 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 8344 | pr_err("Preemption disabled at:"); |
2062a4e8 | 8345 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 8346 | } |
aef745fc | 8347 | dump_stack(); |
f0b22e39 | 8348 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 8349 | } |
3427445a | 8350 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
8351 | |
8352 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
8353 | { | |
8354 | static unsigned long prev_jiffy; | |
8355 | ||
8356 | if (irqs_disabled()) | |
8357 | return; | |
8358 | ||
8359 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
8360 | return; | |
8361 | ||
8362 | if (preempt_count() > preempt_offset) | |
8363 | return; | |
8364 | ||
8365 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8366 | return; | |
8367 | prev_jiffy = jiffies; | |
8368 | ||
8369 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
8370 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8371 | in_atomic(), irqs_disabled(), | |
8372 | current->pid, current->comm); | |
8373 | ||
8374 | debug_show_held_locks(current); | |
8375 | dump_stack(); | |
8376 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
8377 | } | |
8378 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
8379 | |
8380 | #ifdef CONFIG_SMP | |
8381 | void __cant_migrate(const char *file, int line) | |
8382 | { | |
8383 | static unsigned long prev_jiffy; | |
8384 | ||
8385 | if (irqs_disabled()) | |
8386 | return; | |
8387 | ||
8388 | if (is_migration_disabled(current)) | |
8389 | return; | |
8390 | ||
8391 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
8392 | return; | |
8393 | ||
8394 | if (preempt_count() > 0) | |
8395 | return; | |
8396 | ||
8397 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8398 | return; | |
8399 | prev_jiffy = jiffies; | |
8400 | ||
8401 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
8402 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
8403 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
8404 | current->pid, current->comm); | |
8405 | ||
8406 | debug_show_held_locks(current); | |
8407 | dump_stack(); | |
8408 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
8409 | } | |
8410 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
8411 | #endif | |
1da177e4 LT |
8412 | #endif |
8413 | ||
8414 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 8415 | void normalize_rt_tasks(void) |
3a5e4dc1 | 8416 | { |
dbc7f069 | 8417 | struct task_struct *g, *p; |
d50dde5a DF |
8418 | struct sched_attr attr = { |
8419 | .sched_policy = SCHED_NORMAL, | |
8420 | }; | |
1da177e4 | 8421 | |
3472eaa1 | 8422 | read_lock(&tasklist_lock); |
5d07f420 | 8423 | for_each_process_thread(g, p) { |
178be793 IM |
8424 | /* |
8425 | * Only normalize user tasks: | |
8426 | */ | |
3472eaa1 | 8427 | if (p->flags & PF_KTHREAD) |
178be793 IM |
8428 | continue; |
8429 | ||
4fa8d299 JP |
8430 | p->se.exec_start = 0; |
8431 | schedstat_set(p->se.statistics.wait_start, 0); | |
8432 | schedstat_set(p->se.statistics.sleep_start, 0); | |
8433 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 8434 | |
aab03e05 | 8435 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
8436 | /* |
8437 | * Renice negative nice level userspace | |
8438 | * tasks back to 0: | |
8439 | */ | |
3472eaa1 | 8440 | if (task_nice(p) < 0) |
dd41f596 | 8441 | set_user_nice(p, 0); |
1da177e4 | 8442 | continue; |
dd41f596 | 8443 | } |
1da177e4 | 8444 | |
dbc7f069 | 8445 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 8446 | } |
3472eaa1 | 8447 | read_unlock(&tasklist_lock); |
1da177e4 LT |
8448 | } |
8449 | ||
8450 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8451 | |
67fc4e0c | 8452 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8453 | /* |
67fc4e0c | 8454 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8455 | * |
8456 | * They can only be called when the whole system has been | |
8457 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8458 | * activity can take place. Using them for anything else would | |
8459 | * be a serious bug, and as a result, they aren't even visible | |
8460 | * under any other configuration. | |
8461 | */ | |
8462 | ||
8463 | /** | |
d1ccc66d | 8464 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
8465 | * @cpu: the processor in question. |
8466 | * | |
8467 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
8468 | * |
8469 | * Return: The current task for @cpu. | |
1df5c10a | 8470 | */ |
36c8b586 | 8471 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8472 | { |
8473 | return cpu_curr(cpu); | |
8474 | } | |
8475 | ||
67fc4e0c JW |
8476 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8477 | ||
8478 | #ifdef CONFIG_IA64 | |
1df5c10a | 8479 | /** |
5feeb783 | 8480 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
8481 | * @cpu: the processor in question. |
8482 | * @p: the task pointer to set. | |
8483 | * | |
8484 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 8485 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 8486 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
8487 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8488 | * and caller must save the original value of the current task (see | |
8489 | * curr_task() above) and restore that value before reenabling interrupts and | |
8490 | * re-starting the system. | |
8491 | * | |
8492 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8493 | */ | |
a458ae2e | 8494 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8495 | { |
8496 | cpu_curr(cpu) = p; | |
8497 | } | |
8498 | ||
8499 | #endif | |
29f59db3 | 8500 | |
7c941438 | 8501 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
8502 | /* task_group_lock serializes the addition/removal of task groups */ |
8503 | static DEFINE_SPINLOCK(task_group_lock); | |
8504 | ||
2480c093 PB |
8505 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
8506 | struct task_group *parent) | |
8507 | { | |
8508 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 8509 | enum uclamp_id clamp_id; |
2480c093 PB |
8510 | |
8511 | for_each_clamp_id(clamp_id) { | |
8512 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
8513 | uclamp_none(clamp_id), false); | |
0b60ba2d | 8514 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
8515 | } |
8516 | #endif | |
8517 | } | |
8518 | ||
2f5177f0 | 8519 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
8520 | { |
8521 | free_fair_sched_group(tg); | |
8522 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8523 | autogroup_free(tg); |
b0367629 | 8524 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
8525 | } |
8526 | ||
8527 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8528 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8529 | { |
8530 | struct task_group *tg; | |
bccbe08a | 8531 | |
b0367629 | 8532 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
8533 | if (!tg) |
8534 | return ERR_PTR(-ENOMEM); | |
8535 | ||
ec7dc8ac | 8536 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8537 | goto err; |
8538 | ||
ec7dc8ac | 8539 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8540 | goto err; |
8541 | ||
2480c093 PB |
8542 | alloc_uclamp_sched_group(tg, parent); |
8543 | ||
ace783b9 LZ |
8544 | return tg; |
8545 | ||
8546 | err: | |
2f5177f0 | 8547 | sched_free_group(tg); |
ace783b9 LZ |
8548 | return ERR_PTR(-ENOMEM); |
8549 | } | |
8550 | ||
8551 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
8552 | { | |
8553 | unsigned long flags; | |
8554 | ||
8ed36996 | 8555 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8556 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 8557 | |
d1ccc66d IM |
8558 | /* Root should already exist: */ |
8559 | WARN_ON(!parent); | |
f473aa5e PZ |
8560 | |
8561 | tg->parent = parent; | |
f473aa5e | 8562 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8563 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8564 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
8565 | |
8566 | online_fair_sched_group(tg); | |
29f59db3 SV |
8567 | } |
8568 | ||
9b5b7751 | 8569 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 8570 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8571 | { |
d1ccc66d | 8572 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 8573 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8574 | } |
8575 | ||
4cf86d77 | 8576 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 8577 | { |
d1ccc66d | 8578 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 8579 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
8580 | } |
8581 | ||
8582 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 8583 | { |
8ed36996 | 8584 | unsigned long flags; |
29f59db3 | 8585 | |
d1ccc66d | 8586 | /* End participation in shares distribution: */ |
6fe1f348 | 8587 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
8588 | |
8589 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8590 | list_del_rcu(&tg->list); |
f473aa5e | 8591 | list_del_rcu(&tg->siblings); |
8ed36996 | 8592 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
8593 | } |
8594 | ||
ea86cb4b | 8595 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 8596 | { |
8323f26c | 8597 | struct task_group *tg; |
29f59db3 | 8598 | |
f7b8a47d KT |
8599 | /* |
8600 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
8601 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
8602 | * to prevent lockdep warnings. | |
8603 | */ | |
8604 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
8605 | struct task_group, css); |
8606 | tg = autogroup_task_group(tsk, tg); | |
8607 | tsk->sched_task_group = tg; | |
8608 | ||
810b3817 | 8609 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
8610 | if (tsk->sched_class->task_change_group) |
8611 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 8612 | else |
810b3817 | 8613 | #endif |
b2b5ce02 | 8614 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
8615 | } |
8616 | ||
8617 | /* | |
8618 | * Change task's runqueue when it moves between groups. | |
8619 | * | |
8620 | * The caller of this function should have put the task in its new group by | |
8621 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
8622 | * its new group. | |
8623 | */ | |
8624 | void sched_move_task(struct task_struct *tsk) | |
8625 | { | |
7a57f32a PZ |
8626 | int queued, running, queue_flags = |
8627 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
8628 | struct rq_flags rf; |
8629 | struct rq *rq; | |
8630 | ||
8631 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 8632 | update_rq_clock(rq); |
ea86cb4b VG |
8633 | |
8634 | running = task_current(rq, tsk); | |
8635 | queued = task_on_rq_queued(tsk); | |
8636 | ||
8637 | if (queued) | |
7a57f32a | 8638 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 8639 | if (running) |
ea86cb4b VG |
8640 | put_prev_task(rq, tsk); |
8641 | ||
8642 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 8643 | |
da0c1e65 | 8644 | if (queued) |
7a57f32a | 8645 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 8646 | if (running) { |
03b7fad1 | 8647 | set_next_task(rq, tsk); |
2a4b03ff VG |
8648 | /* |
8649 | * After changing group, the running task may have joined a | |
8650 | * throttled one but it's still the running task. Trigger a | |
8651 | * resched to make sure that task can still run. | |
8652 | */ | |
8653 | resched_curr(rq); | |
8654 | } | |
29f59db3 | 8655 | |
eb580751 | 8656 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 8657 | } |
68318b8e | 8658 | |
a7c6d554 | 8659 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8660 | { |
a7c6d554 | 8661 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8662 | } |
8663 | ||
eb95419b TH |
8664 | static struct cgroup_subsys_state * |
8665 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8666 | { |
eb95419b TH |
8667 | struct task_group *parent = css_tg(parent_css); |
8668 | struct task_group *tg; | |
68318b8e | 8669 | |
eb95419b | 8670 | if (!parent) { |
68318b8e | 8671 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8672 | return &root_task_group.css; |
68318b8e SV |
8673 | } |
8674 | ||
ec7dc8ac | 8675 | tg = sched_create_group(parent); |
68318b8e SV |
8676 | if (IS_ERR(tg)) |
8677 | return ERR_PTR(-ENOMEM); | |
8678 | ||
68318b8e SV |
8679 | return &tg->css; |
8680 | } | |
8681 | ||
96b77745 KK |
8682 | /* Expose task group only after completing cgroup initialization */ |
8683 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
8684 | { | |
8685 | struct task_group *tg = css_tg(css); | |
8686 | struct task_group *parent = css_tg(css->parent); | |
8687 | ||
8688 | if (parent) | |
8689 | sched_online_group(tg, parent); | |
7226017a QY |
8690 | |
8691 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8692 | /* Propagate the effective uclamp value for the new group */ | |
8693 | cpu_util_update_eff(css); | |
8694 | #endif | |
8695 | ||
96b77745 KK |
8696 | return 0; |
8697 | } | |
8698 | ||
2f5177f0 | 8699 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8700 | { |
eb95419b | 8701 | struct task_group *tg = css_tg(css); |
ace783b9 | 8702 | |
2f5177f0 | 8703 | sched_offline_group(tg); |
ace783b9 LZ |
8704 | } |
8705 | ||
eb95419b | 8706 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8707 | { |
eb95419b | 8708 | struct task_group *tg = css_tg(css); |
68318b8e | 8709 | |
2f5177f0 PZ |
8710 | /* |
8711 | * Relies on the RCU grace period between css_released() and this. | |
8712 | */ | |
8713 | sched_free_group(tg); | |
ace783b9 LZ |
8714 | } |
8715 | ||
ea86cb4b VG |
8716 | /* |
8717 | * This is called before wake_up_new_task(), therefore we really only | |
8718 | * have to set its group bits, all the other stuff does not apply. | |
8719 | */ | |
b53202e6 | 8720 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 8721 | { |
ea86cb4b VG |
8722 | struct rq_flags rf; |
8723 | struct rq *rq; | |
8724 | ||
8725 | rq = task_rq_lock(task, &rf); | |
8726 | ||
80f5c1b8 | 8727 | update_rq_clock(rq); |
ea86cb4b VG |
8728 | sched_change_group(task, TASK_SET_GROUP); |
8729 | ||
8730 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
8731 | } |
8732 | ||
1f7dd3e5 | 8733 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8734 | { |
bb9d97b6 | 8735 | struct task_struct *task; |
1f7dd3e5 | 8736 | struct cgroup_subsys_state *css; |
7dc603c9 | 8737 | int ret = 0; |
bb9d97b6 | 8738 | |
1f7dd3e5 | 8739 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8740 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8741 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8742 | return -EINVAL; |
b68aa230 | 8743 | #endif |
7dc603c9 | 8744 | /* |
b19a888c | 8745 | * Serialize against wake_up_new_task() such that if it's |
7dc603c9 PZ |
8746 | * running, we're sure to observe its full state. |
8747 | */ | |
8748 | raw_spin_lock_irq(&task->pi_lock); | |
8749 | /* | |
8750 | * Avoid calling sched_move_task() before wake_up_new_task() | |
8751 | * has happened. This would lead to problems with PELT, due to | |
8752 | * move wanting to detach+attach while we're not attached yet. | |
8753 | */ | |
8754 | if (task->state == TASK_NEW) | |
8755 | ret = -EINVAL; | |
8756 | raw_spin_unlock_irq(&task->pi_lock); | |
8757 | ||
8758 | if (ret) | |
8759 | break; | |
bb9d97b6 | 8760 | } |
7dc603c9 | 8761 | return ret; |
be367d09 | 8762 | } |
68318b8e | 8763 | |
1f7dd3e5 | 8764 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8765 | { |
bb9d97b6 | 8766 | struct task_struct *task; |
1f7dd3e5 | 8767 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8768 | |
1f7dd3e5 | 8769 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8770 | sched_move_task(task); |
68318b8e SV |
8771 | } |
8772 | ||
2480c093 | 8773 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
8774 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
8775 | { | |
8776 | struct cgroup_subsys_state *top_css = css; | |
8777 | struct uclamp_se *uc_parent = NULL; | |
8778 | struct uclamp_se *uc_se = NULL; | |
8779 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 8780 | enum uclamp_id clamp_id; |
0b60ba2d PB |
8781 | unsigned int clamps; |
8782 | ||
8783 | css_for_each_descendant_pre(css, top_css) { | |
8784 | uc_parent = css_tg(css)->parent | |
8785 | ? css_tg(css)->parent->uclamp : NULL; | |
8786 | ||
8787 | for_each_clamp_id(clamp_id) { | |
8788 | /* Assume effective clamps matches requested clamps */ | |
8789 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
8790 | /* Cap effective clamps with parent's effective clamps */ | |
8791 | if (uc_parent && | |
8792 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
8793 | eff[clamp_id] = uc_parent[clamp_id].value; | |
8794 | } | |
8795 | } | |
8796 | /* Ensure protection is always capped by limit */ | |
8797 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
8798 | ||
8799 | /* Propagate most restrictive effective clamps */ | |
8800 | clamps = 0x0; | |
8801 | uc_se = css_tg(css)->uclamp; | |
8802 | for_each_clamp_id(clamp_id) { | |
8803 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
8804 | continue; | |
8805 | uc_se[clamp_id].value = eff[clamp_id]; | |
8806 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
8807 | clamps |= (0x1 << clamp_id); | |
8808 | } | |
babbe170 | 8809 | if (!clamps) { |
0b60ba2d | 8810 | css = css_rightmost_descendant(css); |
babbe170 PB |
8811 | continue; |
8812 | } | |
8813 | ||
8814 | /* Immediately update descendants RUNNABLE tasks */ | |
8815 | uclamp_update_active_tasks(css, clamps); | |
0b60ba2d PB |
8816 | } |
8817 | } | |
2480c093 PB |
8818 | |
8819 | /* | |
8820 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
8821 | * C expression. Since there is no way to convert a macro argument (N) into a | |
8822 | * character constant, use two levels of macros. | |
8823 | */ | |
8824 | #define _POW10(exp) ((unsigned int)1e##exp) | |
8825 | #define POW10(exp) _POW10(exp) | |
8826 | ||
8827 | struct uclamp_request { | |
8828 | #define UCLAMP_PERCENT_SHIFT 2 | |
8829 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
8830 | s64 percent; | |
8831 | u64 util; | |
8832 | int ret; | |
8833 | }; | |
8834 | ||
8835 | static inline struct uclamp_request | |
8836 | capacity_from_percent(char *buf) | |
8837 | { | |
8838 | struct uclamp_request req = { | |
8839 | .percent = UCLAMP_PERCENT_SCALE, | |
8840 | .util = SCHED_CAPACITY_SCALE, | |
8841 | .ret = 0, | |
8842 | }; | |
8843 | ||
8844 | buf = strim(buf); | |
8845 | if (strcmp(buf, "max")) { | |
8846 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
8847 | &req.percent); | |
8848 | if (req.ret) | |
8849 | return req; | |
b562d140 | 8850 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
8851 | req.ret = -ERANGE; |
8852 | return req; | |
8853 | } | |
8854 | ||
8855 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
8856 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
8857 | } | |
8858 | ||
8859 | return req; | |
8860 | } | |
8861 | ||
8862 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
8863 | size_t nbytes, loff_t off, | |
8864 | enum uclamp_id clamp_id) | |
8865 | { | |
8866 | struct uclamp_request req; | |
8867 | struct task_group *tg; | |
8868 | ||
8869 | req = capacity_from_percent(buf); | |
8870 | if (req.ret) | |
8871 | return req.ret; | |
8872 | ||
46609ce2 QY |
8873 | static_branch_enable(&sched_uclamp_used); |
8874 | ||
2480c093 PB |
8875 | mutex_lock(&uclamp_mutex); |
8876 | rcu_read_lock(); | |
8877 | ||
8878 | tg = css_tg(of_css(of)); | |
8879 | if (tg->uclamp_req[clamp_id].value != req.util) | |
8880 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
8881 | ||
8882 | /* | |
8883 | * Because of not recoverable conversion rounding we keep track of the | |
8884 | * exact requested value | |
8885 | */ | |
8886 | tg->uclamp_pct[clamp_id] = req.percent; | |
8887 | ||
0b60ba2d PB |
8888 | /* Update effective clamps to track the most restrictive value */ |
8889 | cpu_util_update_eff(of_css(of)); | |
8890 | ||
2480c093 PB |
8891 | rcu_read_unlock(); |
8892 | mutex_unlock(&uclamp_mutex); | |
8893 | ||
8894 | return nbytes; | |
8895 | } | |
8896 | ||
8897 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
8898 | char *buf, size_t nbytes, | |
8899 | loff_t off) | |
8900 | { | |
8901 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
8902 | } | |
8903 | ||
8904 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
8905 | char *buf, size_t nbytes, | |
8906 | loff_t off) | |
8907 | { | |
8908 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
8909 | } | |
8910 | ||
8911 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
8912 | enum uclamp_id clamp_id) | |
8913 | { | |
8914 | struct task_group *tg; | |
8915 | u64 util_clamp; | |
8916 | u64 percent; | |
8917 | u32 rem; | |
8918 | ||
8919 | rcu_read_lock(); | |
8920 | tg = css_tg(seq_css(sf)); | |
8921 | util_clamp = tg->uclamp_req[clamp_id].value; | |
8922 | rcu_read_unlock(); | |
8923 | ||
8924 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
8925 | seq_puts(sf, "max\n"); | |
8926 | return; | |
8927 | } | |
8928 | ||
8929 | percent = tg->uclamp_pct[clamp_id]; | |
8930 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
8931 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
8932 | } | |
8933 | ||
8934 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
8935 | { | |
8936 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
8937 | return 0; | |
8938 | } | |
8939 | ||
8940 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
8941 | { | |
8942 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
8943 | return 0; | |
8944 | } | |
8945 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
8946 | ||
052f1dc7 | 8947 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8948 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8949 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8950 | { |
5b61d50a KK |
8951 | if (shareval > scale_load_down(ULONG_MAX)) |
8952 | shareval = MAX_SHARES; | |
182446d0 | 8953 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8954 | } |
8955 | ||
182446d0 TH |
8956 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8957 | struct cftype *cft) | |
68318b8e | 8958 | { |
182446d0 | 8959 | struct task_group *tg = css_tg(css); |
68318b8e | 8960 | |
c8b28116 | 8961 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8962 | } |
ab84d31e PT |
8963 | |
8964 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8965 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8966 | ||
ab84d31e | 8967 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 8968 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
8969 | /* More than 203 days if BW_SHIFT equals 20. */ |
8970 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 8971 | |
a790de99 PT |
8972 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8973 | ||
ab84d31e PT |
8974 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8975 | { | |
56f570e5 | 8976 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8977 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8978 | |
8979 | if (tg == &root_task_group) | |
8980 | return -EINVAL; | |
8981 | ||
8982 | /* | |
8983 | * Ensure we have at some amount of bandwidth every period. This is | |
8984 | * to prevent reaching a state of large arrears when throttled via | |
8985 | * entity_tick() resulting in prolonged exit starvation. | |
8986 | */ | |
8987 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8988 | return -EINVAL; | |
8989 | ||
8990 | /* | |
8991 | * Likewise, bound things on the otherside by preventing insane quota | |
8992 | * periods. This also allows us to normalize in computing quota | |
8993 | * feasibility. | |
8994 | */ | |
8995 | if (period > max_cfs_quota_period) | |
8996 | return -EINVAL; | |
8997 | ||
d505b8af HC |
8998 | /* |
8999 | * Bound quota to defend quota against overflow during bandwidth shift. | |
9000 | */ | |
9001 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
9002 | return -EINVAL; | |
9003 | ||
0e59bdae KT |
9004 | /* |
9005 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
9006 | * unthrottle_offline_cfs_rqs(). | |
9007 | */ | |
9008 | get_online_cpus(); | |
a790de99 PT |
9009 | mutex_lock(&cfs_constraints_mutex); |
9010 | ret = __cfs_schedulable(tg, period, quota); | |
9011 | if (ret) | |
9012 | goto out_unlock; | |
9013 | ||
58088ad0 | 9014 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 9015 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
9016 | /* |
9017 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
9018 | * before making related changes, and on->off must occur afterwards | |
9019 | */ | |
9020 | if (runtime_enabled && !runtime_was_enabled) | |
9021 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
9022 | raw_spin_lock_irq(&cfs_b->lock); |
9023 | cfs_b->period = ns_to_ktime(period); | |
9024 | cfs_b->quota = quota; | |
58088ad0 | 9025 | |
a9cf55b2 | 9026 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
9027 | |
9028 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
9029 | if (runtime_enabled) |
9030 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 9031 | |
ab84d31e PT |
9032 | raw_spin_unlock_irq(&cfs_b->lock); |
9033 | ||
0e59bdae | 9034 | for_each_online_cpu(i) { |
ab84d31e | 9035 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 9036 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 9037 | struct rq_flags rf; |
ab84d31e | 9038 | |
8a8c69c3 | 9039 | rq_lock_irq(rq, &rf); |
58088ad0 | 9040 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 9041 | cfs_rq->runtime_remaining = 0; |
671fd9da | 9042 | |
029632fb | 9043 | if (cfs_rq->throttled) |
671fd9da | 9044 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 9045 | rq_unlock_irq(rq, &rf); |
ab84d31e | 9046 | } |
1ee14e6c BS |
9047 | if (runtime_was_enabled && !runtime_enabled) |
9048 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
9049 | out_unlock: |
9050 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 9051 | put_online_cpus(); |
ab84d31e | 9052 | |
a790de99 | 9053 | return ret; |
ab84d31e PT |
9054 | } |
9055 | ||
b1546edc | 9056 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e PT |
9057 | { |
9058 | u64 quota, period; | |
9059 | ||
029632fb | 9060 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
9061 | if (cfs_quota_us < 0) |
9062 | quota = RUNTIME_INF; | |
1a8b4540 | 9063 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 9064 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
9065 | else |
9066 | return -EINVAL; | |
ab84d31e PT |
9067 | |
9068 | return tg_set_cfs_bandwidth(tg, period, quota); | |
9069 | } | |
9070 | ||
b1546edc | 9071 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
9072 | { |
9073 | u64 quota_us; | |
9074 | ||
029632fb | 9075 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
9076 | return -1; |
9077 | ||
029632fb | 9078 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
9079 | do_div(quota_us, NSEC_PER_USEC); |
9080 | ||
9081 | return quota_us; | |
9082 | } | |
9083 | ||
b1546edc | 9084 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e PT |
9085 | { |
9086 | u64 quota, period; | |
9087 | ||
1a8b4540 KK |
9088 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
9089 | return -EINVAL; | |
9090 | ||
ab84d31e | 9091 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 9092 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 9093 | |
ab84d31e PT |
9094 | return tg_set_cfs_bandwidth(tg, period, quota); |
9095 | } | |
9096 | ||
b1546edc | 9097 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
9098 | { |
9099 | u64 cfs_period_us; | |
9100 | ||
029632fb | 9101 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
9102 | do_div(cfs_period_us, NSEC_PER_USEC); |
9103 | ||
9104 | return cfs_period_us; | |
9105 | } | |
9106 | ||
182446d0 TH |
9107 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
9108 | struct cftype *cft) | |
ab84d31e | 9109 | { |
182446d0 | 9110 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
9111 | } |
9112 | ||
182446d0 TH |
9113 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
9114 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 9115 | { |
182446d0 | 9116 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
9117 | } |
9118 | ||
182446d0 TH |
9119 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
9120 | struct cftype *cft) | |
ab84d31e | 9121 | { |
182446d0 | 9122 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
9123 | } |
9124 | ||
182446d0 TH |
9125 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
9126 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 9127 | { |
182446d0 | 9128 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
9129 | } |
9130 | ||
a790de99 PT |
9131 | struct cfs_schedulable_data { |
9132 | struct task_group *tg; | |
9133 | u64 period, quota; | |
9134 | }; | |
9135 | ||
9136 | /* | |
9137 | * normalize group quota/period to be quota/max_period | |
9138 | * note: units are usecs | |
9139 | */ | |
9140 | static u64 normalize_cfs_quota(struct task_group *tg, | |
9141 | struct cfs_schedulable_data *d) | |
9142 | { | |
9143 | u64 quota, period; | |
9144 | ||
9145 | if (tg == d->tg) { | |
9146 | period = d->period; | |
9147 | quota = d->quota; | |
9148 | } else { | |
9149 | period = tg_get_cfs_period(tg); | |
9150 | quota = tg_get_cfs_quota(tg); | |
9151 | } | |
9152 | ||
9153 | /* note: these should typically be equivalent */ | |
9154 | if (quota == RUNTIME_INF || quota == -1) | |
9155 | return RUNTIME_INF; | |
9156 | ||
9157 | return to_ratio(period, quota); | |
9158 | } | |
9159 | ||
9160 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
9161 | { | |
9162 | struct cfs_schedulable_data *d = data; | |
029632fb | 9163 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
9164 | s64 quota = 0, parent_quota = -1; |
9165 | ||
9166 | if (!tg->parent) { | |
9167 | quota = RUNTIME_INF; | |
9168 | } else { | |
029632fb | 9169 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
9170 | |
9171 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 9172 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
9173 | |
9174 | /* | |
c53593e5 TH |
9175 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
9176 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 9177 | * limit is set: |
a790de99 | 9178 | */ |
c53593e5 TH |
9179 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
9180 | quota = min(quota, parent_quota); | |
9181 | } else { | |
9182 | if (quota == RUNTIME_INF) | |
9183 | quota = parent_quota; | |
9184 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
9185 | return -EINVAL; | |
9186 | } | |
a790de99 | 9187 | } |
9c58c79a | 9188 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
9189 | |
9190 | return 0; | |
9191 | } | |
9192 | ||
9193 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
9194 | { | |
8277434e | 9195 | int ret; |
a790de99 PT |
9196 | struct cfs_schedulable_data data = { |
9197 | .tg = tg, | |
9198 | .period = period, | |
9199 | .quota = quota, | |
9200 | }; | |
9201 | ||
9202 | if (quota != RUNTIME_INF) { | |
9203 | do_div(data.period, NSEC_PER_USEC); | |
9204 | do_div(data.quota, NSEC_PER_USEC); | |
9205 | } | |
9206 | ||
8277434e PT |
9207 | rcu_read_lock(); |
9208 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
9209 | rcu_read_unlock(); | |
9210 | ||
9211 | return ret; | |
a790de99 | 9212 | } |
e8da1b18 | 9213 | |
a1f7164c | 9214 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 9215 | { |
2da8ca82 | 9216 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 9217 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 9218 | |
44ffc75b TH |
9219 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
9220 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
9221 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 9222 | |
3d6c50c2 YW |
9223 | if (schedstat_enabled() && tg != &root_task_group) { |
9224 | u64 ws = 0; | |
9225 | int i; | |
9226 | ||
9227 | for_each_possible_cpu(i) | |
9228 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
9229 | ||
9230 | seq_printf(sf, "wait_sum %llu\n", ws); | |
9231 | } | |
9232 | ||
e8da1b18 NR |
9233 | return 0; |
9234 | } | |
ab84d31e | 9235 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 9236 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9237 | |
052f1dc7 | 9238 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
9239 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
9240 | struct cftype *cft, s64 val) | |
6f505b16 | 9241 | { |
182446d0 | 9242 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
9243 | } |
9244 | ||
182446d0 TH |
9245 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
9246 | struct cftype *cft) | |
6f505b16 | 9247 | { |
182446d0 | 9248 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 9249 | } |
d0b27fa7 | 9250 | |
182446d0 TH |
9251 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
9252 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 9253 | { |
182446d0 | 9254 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
9255 | } |
9256 | ||
182446d0 TH |
9257 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
9258 | struct cftype *cft) | |
d0b27fa7 | 9259 | { |
182446d0 | 9260 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 9261 | } |
6d6bc0ad | 9262 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9263 | |
a1f7164c | 9264 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 9265 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9266 | { |
9267 | .name = "shares", | |
f4c753b7 PM |
9268 | .read_u64 = cpu_shares_read_u64, |
9269 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9270 | }, |
052f1dc7 | 9271 | #endif |
ab84d31e PT |
9272 | #ifdef CONFIG_CFS_BANDWIDTH |
9273 | { | |
9274 | .name = "cfs_quota_us", | |
9275 | .read_s64 = cpu_cfs_quota_read_s64, | |
9276 | .write_s64 = cpu_cfs_quota_write_s64, | |
9277 | }, | |
9278 | { | |
9279 | .name = "cfs_period_us", | |
9280 | .read_u64 = cpu_cfs_period_read_u64, | |
9281 | .write_u64 = cpu_cfs_period_write_u64, | |
9282 | }, | |
e8da1b18 NR |
9283 | { |
9284 | .name = "stat", | |
a1f7164c | 9285 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 9286 | }, |
ab84d31e | 9287 | #endif |
052f1dc7 | 9288 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9289 | { |
9f0c1e56 | 9290 | .name = "rt_runtime_us", |
06ecb27c PM |
9291 | .read_s64 = cpu_rt_runtime_read, |
9292 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9293 | }, |
d0b27fa7 PZ |
9294 | { |
9295 | .name = "rt_period_us", | |
f4c753b7 PM |
9296 | .read_u64 = cpu_rt_period_read_uint, |
9297 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9298 | }, |
2480c093 PB |
9299 | #endif |
9300 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9301 | { | |
9302 | .name = "uclamp.min", | |
9303 | .flags = CFTYPE_NOT_ON_ROOT, | |
9304 | .seq_show = cpu_uclamp_min_show, | |
9305 | .write = cpu_uclamp_min_write, | |
9306 | }, | |
9307 | { | |
9308 | .name = "uclamp.max", | |
9309 | .flags = CFTYPE_NOT_ON_ROOT, | |
9310 | .seq_show = cpu_uclamp_max_show, | |
9311 | .write = cpu_uclamp_max_write, | |
9312 | }, | |
052f1dc7 | 9313 | #endif |
d1ccc66d | 9314 | { } /* Terminate */ |
68318b8e SV |
9315 | }; |
9316 | ||
d41bf8c9 TH |
9317 | static int cpu_extra_stat_show(struct seq_file *sf, |
9318 | struct cgroup_subsys_state *css) | |
0d593634 | 9319 | { |
0d593634 TH |
9320 | #ifdef CONFIG_CFS_BANDWIDTH |
9321 | { | |
d41bf8c9 | 9322 | struct task_group *tg = css_tg(css); |
0d593634 TH |
9323 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
9324 | u64 throttled_usec; | |
9325 | ||
9326 | throttled_usec = cfs_b->throttled_time; | |
9327 | do_div(throttled_usec, NSEC_PER_USEC); | |
9328 | ||
9329 | seq_printf(sf, "nr_periods %d\n" | |
9330 | "nr_throttled %d\n" | |
9331 | "throttled_usec %llu\n", | |
9332 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
9333 | throttled_usec); | |
9334 | } | |
9335 | #endif | |
9336 | return 0; | |
9337 | } | |
9338 | ||
9339 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9340 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
9341 | struct cftype *cft) | |
9342 | { | |
9343 | struct task_group *tg = css_tg(css); | |
9344 | u64 weight = scale_load_down(tg->shares); | |
9345 | ||
9346 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
9347 | } | |
9348 | ||
9349 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
9350 | struct cftype *cft, u64 weight) | |
9351 | { | |
9352 | /* | |
9353 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
9354 | * values which are 1, 100 and 10000 respectively. While it loses | |
9355 | * a bit of range on both ends, it maps pretty well onto the shares | |
9356 | * value used by scheduler and the round-trip conversions preserve | |
9357 | * the original value over the entire range. | |
9358 | */ | |
9359 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
9360 | return -ERANGE; | |
9361 | ||
9362 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
9363 | ||
9364 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
9365 | } | |
9366 | ||
9367 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
9368 | struct cftype *cft) | |
9369 | { | |
9370 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
9371 | int last_delta = INT_MAX; | |
9372 | int prio, delta; | |
9373 | ||
9374 | /* find the closest nice value to the current weight */ | |
9375 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
9376 | delta = abs(sched_prio_to_weight[prio] - weight); | |
9377 | if (delta >= last_delta) | |
9378 | break; | |
9379 | last_delta = delta; | |
9380 | } | |
9381 | ||
9382 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
9383 | } | |
9384 | ||
9385 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
9386 | struct cftype *cft, s64 nice) | |
9387 | { | |
9388 | unsigned long weight; | |
7281c8de | 9389 | int idx; |
0d593634 TH |
9390 | |
9391 | if (nice < MIN_NICE || nice > MAX_NICE) | |
9392 | return -ERANGE; | |
9393 | ||
7281c8de PZ |
9394 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
9395 | idx = array_index_nospec(idx, 40); | |
9396 | weight = sched_prio_to_weight[idx]; | |
9397 | ||
0d593634 TH |
9398 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
9399 | } | |
9400 | #endif | |
9401 | ||
9402 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
9403 | long period, long quota) | |
9404 | { | |
9405 | if (quota < 0) | |
9406 | seq_puts(sf, "max"); | |
9407 | else | |
9408 | seq_printf(sf, "%ld", quota); | |
9409 | ||
9410 | seq_printf(sf, " %ld\n", period); | |
9411 | } | |
9412 | ||
9413 | /* caller should put the current value in *@periodp before calling */ | |
9414 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
9415 | u64 *periodp, u64 *quotap) | |
9416 | { | |
9417 | char tok[21]; /* U64_MAX */ | |
9418 | ||
4c47acd8 | 9419 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
9420 | return -EINVAL; |
9421 | ||
9422 | *periodp *= NSEC_PER_USEC; | |
9423 | ||
9424 | if (sscanf(tok, "%llu", quotap)) | |
9425 | *quotap *= NSEC_PER_USEC; | |
9426 | else if (!strcmp(tok, "max")) | |
9427 | *quotap = RUNTIME_INF; | |
9428 | else | |
9429 | return -EINVAL; | |
9430 | ||
9431 | return 0; | |
9432 | } | |
9433 | ||
9434 | #ifdef CONFIG_CFS_BANDWIDTH | |
9435 | static int cpu_max_show(struct seq_file *sf, void *v) | |
9436 | { | |
9437 | struct task_group *tg = css_tg(seq_css(sf)); | |
9438 | ||
9439 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
9440 | return 0; | |
9441 | } | |
9442 | ||
9443 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
9444 | char *buf, size_t nbytes, loff_t off) | |
9445 | { | |
9446 | struct task_group *tg = css_tg(of_css(of)); | |
9447 | u64 period = tg_get_cfs_period(tg); | |
9448 | u64 quota; | |
9449 | int ret; | |
9450 | ||
9451 | ret = cpu_period_quota_parse(buf, &period, "a); | |
9452 | if (!ret) | |
9453 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
9454 | return ret ?: nbytes; | |
9455 | } | |
9456 | #endif | |
9457 | ||
9458 | static struct cftype cpu_files[] = { | |
0d593634 TH |
9459 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9460 | { | |
9461 | .name = "weight", | |
9462 | .flags = CFTYPE_NOT_ON_ROOT, | |
9463 | .read_u64 = cpu_weight_read_u64, | |
9464 | .write_u64 = cpu_weight_write_u64, | |
9465 | }, | |
9466 | { | |
9467 | .name = "weight.nice", | |
9468 | .flags = CFTYPE_NOT_ON_ROOT, | |
9469 | .read_s64 = cpu_weight_nice_read_s64, | |
9470 | .write_s64 = cpu_weight_nice_write_s64, | |
9471 | }, | |
9472 | #endif | |
9473 | #ifdef CONFIG_CFS_BANDWIDTH | |
9474 | { | |
9475 | .name = "max", | |
9476 | .flags = CFTYPE_NOT_ON_ROOT, | |
9477 | .seq_show = cpu_max_show, | |
9478 | .write = cpu_max_write, | |
9479 | }, | |
2480c093 PB |
9480 | #endif |
9481 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9482 | { | |
9483 | .name = "uclamp.min", | |
9484 | .flags = CFTYPE_NOT_ON_ROOT, | |
9485 | .seq_show = cpu_uclamp_min_show, | |
9486 | .write = cpu_uclamp_min_write, | |
9487 | }, | |
9488 | { | |
9489 | .name = "uclamp.max", | |
9490 | .flags = CFTYPE_NOT_ON_ROOT, | |
9491 | .seq_show = cpu_uclamp_max_show, | |
9492 | .write = cpu_uclamp_max_write, | |
9493 | }, | |
0d593634 TH |
9494 | #endif |
9495 | { } /* terminate */ | |
9496 | }; | |
9497 | ||
073219e9 | 9498 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 9499 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 9500 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 9501 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 9502 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 9503 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 9504 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
9505 | .can_attach = cpu_cgroup_can_attach, |
9506 | .attach = cpu_cgroup_attach, | |
a1f7164c | 9507 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 9508 | .dfl_cftypes = cpu_files, |
b38e42e9 | 9509 | .early_init = true, |
0d593634 | 9510 | .threaded = true, |
68318b8e SV |
9511 | }; |
9512 | ||
052f1dc7 | 9513 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 9514 | |
b637a328 PM |
9515 | void dump_cpu_task(int cpu) |
9516 | { | |
9517 | pr_info("Task dump for CPU %d:\n", cpu); | |
9518 | sched_show_task(cpu_curr(cpu)); | |
9519 | } | |
ed82b8a1 AK |
9520 | |
9521 | /* | |
9522 | * Nice levels are multiplicative, with a gentle 10% change for every | |
9523 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
9524 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
9525 | * that remained on nice 0. | |
9526 | * | |
9527 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
9528 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
9529 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
9530 | * If a task goes up by ~10% and another task goes down by ~10% then | |
9531 | * the relative distance between them is ~25%.) | |
9532 | */ | |
9533 | const int sched_prio_to_weight[40] = { | |
9534 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
9535 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
9536 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
9537 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
9538 | /* 0 */ 1024, 820, 655, 526, 423, | |
9539 | /* 5 */ 335, 272, 215, 172, 137, | |
9540 | /* 10 */ 110, 87, 70, 56, 45, | |
9541 | /* 15 */ 36, 29, 23, 18, 15, | |
9542 | }; | |
9543 | ||
9544 | /* | |
9545 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
9546 | * | |
9547 | * In cases where the weight does not change often, we can use the | |
9548 | * precalculated inverse to speed up arithmetics by turning divisions | |
9549 | * into multiplications: | |
9550 | */ | |
9551 | const u32 sched_prio_to_wmult[40] = { | |
9552 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
9553 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
9554 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
9555 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
9556 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
9557 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
9558 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
9559 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
9560 | }; | |
14a7405b | 9561 | |
9d246053 PA |
9562 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
9563 | { | |
9564 | trace_sched_update_nr_running_tp(rq, count); | |
9565 | } |