]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
391e43da | 3 | * kernel/sched/core.c |
1da177e4 | 4 | * |
d1ccc66d | 5 | * Core kernel scheduler code and related syscalls |
1da177e4 LT |
6 | * |
7 | * Copyright (C) 1991-2002 Linus Torvalds | |
1da177e4 | 8 | */ |
9d246053 PA |
9 | #define CREATE_TRACE_POINTS |
10 | #include <trace/events/sched.h> | |
11 | #undef CREATE_TRACE_POINTS | |
12 | ||
325ea10c | 13 | #include "sched.h" |
1da177e4 | 14 | |
7281c8de | 15 | #include <linux/nospec.h> |
85f1abe0 | 16 | |
0ed557aa | 17 | #include <linux/kcov.h> |
d08b9f0c | 18 | #include <linux/scs.h> |
0ed557aa | 19 | |
96f951ed | 20 | #include <asm/switch_to.h> |
5517d86b | 21 | #include <asm/tlb.h> |
1da177e4 | 22 | |
ea138446 | 23 | #include "../workqueue_internal.h" |
771b53d0 | 24 | #include "../../fs/io-wq.h" |
29d5e047 | 25 | #include "../smpboot.h" |
6e0534f2 | 26 | |
91c27493 | 27 | #include "pelt.h" |
1f8db415 | 28 | #include "smp.h" |
91c27493 | 29 | |
a056a5be QY |
30 | /* |
31 | * Export tracepoints that act as a bare tracehook (ie: have no trace event | |
32 | * associated with them) to allow external modules to probe them. | |
33 | */ | |
34 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp); | |
35 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp); | |
36 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp); | |
37 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp); | |
38 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp); | |
51cf18c9 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp); |
a056a5be | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); |
4581bea8 VD |
41 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp); |
42 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); | |
9d246053 | 43 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); |
a056a5be | 44 | |
029632fb | 45 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 46 | |
a73f863a | 47 | #ifdef CONFIG_SCHED_DEBUG |
bf5c91ba IM |
48 | /* |
49 | * Debugging: various feature bits | |
765cc3a4 PB |
50 | * |
51 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
52 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
53 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 54 | */ |
f00b45c1 PZ |
55 | #define SCHED_FEAT(name, enabled) \ |
56 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 57 | const_debug unsigned int sysctl_sched_features = |
391e43da | 58 | #include "features.h" |
f00b45c1 | 59 | 0; |
f00b45c1 | 60 | #undef SCHED_FEAT |
765cc3a4 | 61 | #endif |
f00b45c1 | 62 | |
b82d9fdd PZ |
63 | /* |
64 | * Number of tasks to iterate in a single balance run. | |
65 | * Limited because this is done with IRQs disabled. | |
66 | */ | |
67 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
68 | ||
fa85ae24 | 69 | /* |
d1ccc66d | 70 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
71 | * default: 1s |
72 | */ | |
9f0c1e56 | 73 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 74 | |
029632fb | 75 | __read_mostly int scheduler_running; |
6892b75e | 76 | |
9f0c1e56 PZ |
77 | /* |
78 | * part of the period that we allow rt tasks to run in us. | |
79 | * default: 0.95s | |
80 | */ | |
81 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 82 | |
58877d34 PZ |
83 | |
84 | /* | |
85 | * Serialization rules: | |
86 | * | |
87 | * Lock order: | |
88 | * | |
89 | * p->pi_lock | |
90 | * rq->lock | |
91 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
92 | * | |
93 | * rq1->lock | |
94 | * rq2->lock where: rq1 < rq2 | |
95 | * | |
96 | * Regular state: | |
97 | * | |
98 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
99 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
b19a888c | 100 | * always looks at the local rq data structures to find the most eligible task |
58877d34 PZ |
101 | * to run next. |
102 | * | |
103 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
104 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
105 | * the local CPU to avoid bouncing the runqueue state around [ see | |
106 | * ttwu_queue_wakelist() ] | |
107 | * | |
108 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
109 | * complicated to avoid having to take two rq->locks. | |
110 | * | |
111 | * Special state: | |
112 | * | |
113 | * System-calls and anything external will use task_rq_lock() which acquires | |
114 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
115 | * stable while holding either lock: | |
116 | * | |
117 | * - sched_setaffinity()/ | |
118 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
119 | * - set_user_nice(): p->se.load, p->*prio | |
120 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
121 | * p->se.load, p->rt_priority, | |
122 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
123 | * - sched_setnuma(): p->numa_preferred_nid | |
124 | * - sched_move_task()/ | |
125 | * cpu_cgroup_fork(): p->sched_task_group | |
126 | * - uclamp_update_active() p->uclamp* | |
127 | * | |
128 | * p->state <- TASK_*: | |
129 | * | |
130 | * is changed locklessly using set_current_state(), __set_current_state() or | |
131 | * set_special_state(), see their respective comments, or by | |
132 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
133 | * concurrent self. | |
134 | * | |
135 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
136 | * | |
137 | * is set by activate_task() and cleared by deactivate_task(), under | |
138 | * rq->lock. Non-zero indicates the task is runnable, the special | |
139 | * ON_RQ_MIGRATING state is used for migration without holding both | |
140 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
141 | * | |
142 | * p->on_cpu <- { 0, 1 }: | |
143 | * | |
144 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
145 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
146 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
147 | * | |
148 | * [ The astute reader will observe that it is possible for two tasks on one | |
149 | * CPU to have ->on_cpu = 1 at the same time. ] | |
150 | * | |
151 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
152 | * | |
153 | * - Don't call set_task_cpu() on a blocked task: | |
154 | * | |
155 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
156 | * the CPU assignment of blocked tasks isn't required to be valid. | |
157 | * | |
158 | * - for try_to_wake_up(), called under p->pi_lock: | |
159 | * | |
160 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
161 | * | |
162 | * - for migration called under rq->lock: | |
163 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
164 | * | |
165 | * o move_queued_task() | |
166 | * o detach_task() | |
167 | * | |
168 | * - for migration called under double_rq_lock(): | |
169 | * | |
170 | * o __migrate_swap_task() | |
171 | * o push_rt_task() / pull_rt_task() | |
172 | * o push_dl_task() / pull_dl_task() | |
173 | * o dl_task_offline_migration() | |
174 | * | |
175 | */ | |
176 | ||
3e71a462 PZ |
177 | /* |
178 | * __task_rq_lock - lock the rq @p resides on. | |
179 | */ | |
eb580751 | 180 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
181 | __acquires(rq->lock) |
182 | { | |
183 | struct rq *rq; | |
184 | ||
185 | lockdep_assert_held(&p->pi_lock); | |
186 | ||
187 | for (;;) { | |
188 | rq = task_rq(p); | |
189 | raw_spin_lock(&rq->lock); | |
190 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 191 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
192 | return rq; |
193 | } | |
194 | raw_spin_unlock(&rq->lock); | |
195 | ||
196 | while (unlikely(task_on_rq_migrating(p))) | |
197 | cpu_relax(); | |
198 | } | |
199 | } | |
200 | ||
201 | /* | |
202 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
203 | */ | |
eb580751 | 204 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
205 | __acquires(p->pi_lock) |
206 | __acquires(rq->lock) | |
207 | { | |
208 | struct rq *rq; | |
209 | ||
210 | for (;;) { | |
eb580751 | 211 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
212 | rq = task_rq(p); |
213 | raw_spin_lock(&rq->lock); | |
214 | /* | |
215 | * move_queued_task() task_rq_lock() | |
216 | * | |
217 | * ACQUIRE (rq->lock) | |
218 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
219 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
220 | * [S] ->cpu = new_cpu [L] task_rq() | |
221 | * [L] ->on_rq | |
222 | * RELEASE (rq->lock) | |
223 | * | |
c546951d | 224 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
225 | * the old rq->lock will fully serialize against the stores. |
226 | * | |
c546951d AP |
227 | * If we observe the new CPU in task_rq_lock(), the address |
228 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
229 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
230 | */ |
231 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 232 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
233 | return rq; |
234 | } | |
235 | raw_spin_unlock(&rq->lock); | |
eb580751 | 236 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
237 | |
238 | while (unlikely(task_on_rq_migrating(p))) | |
239 | cpu_relax(); | |
240 | } | |
241 | } | |
242 | ||
535b9552 IM |
243 | /* |
244 | * RQ-clock updating methods: | |
245 | */ | |
246 | ||
247 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
248 | { | |
249 | /* | |
250 | * In theory, the compile should just see 0 here, and optimize out the call | |
251 | * to sched_rt_avg_update. But I don't trust it... | |
252 | */ | |
11d4afd4 VG |
253 | s64 __maybe_unused steal = 0, irq_delta = 0; |
254 | ||
535b9552 IM |
255 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
256 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
257 | ||
258 | /* | |
259 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
260 | * this case when a previous update_rq_clock() happened inside a | |
261 | * {soft,}irq region. | |
262 | * | |
263 | * When this happens, we stop ->clock_task and only update the | |
264 | * prev_irq_time stamp to account for the part that fit, so that a next | |
265 | * update will consume the rest. This ensures ->clock_task is | |
266 | * monotonic. | |
267 | * | |
268 | * It does however cause some slight miss-attribution of {soft,}irq | |
269 | * time, a more accurate solution would be to update the irq_time using | |
270 | * the current rq->clock timestamp, except that would require using | |
271 | * atomic ops. | |
272 | */ | |
273 | if (irq_delta > delta) | |
274 | irq_delta = delta; | |
275 | ||
276 | rq->prev_irq_time += irq_delta; | |
277 | delta -= irq_delta; | |
278 | #endif | |
279 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
280 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
281 | steal = paravirt_steal_clock(cpu_of(rq)); | |
282 | steal -= rq->prev_steal_time_rq; | |
283 | ||
284 | if (unlikely(steal > delta)) | |
285 | steal = delta; | |
286 | ||
287 | rq->prev_steal_time_rq += steal; | |
288 | delta -= steal; | |
289 | } | |
290 | #endif | |
291 | ||
292 | rq->clock_task += delta; | |
293 | ||
11d4afd4 | 294 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 295 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 296 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 297 | #endif |
23127296 | 298 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
299 | } |
300 | ||
301 | void update_rq_clock(struct rq *rq) | |
302 | { | |
303 | s64 delta; | |
304 | ||
305 | lockdep_assert_held(&rq->lock); | |
306 | ||
307 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
308 | return; | |
309 | ||
310 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
311 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
312 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
313 | rq->clock_update_flags |= RQCF_UPDATED; |
314 | #endif | |
26ae58d2 | 315 | |
535b9552 IM |
316 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
317 | if (delta < 0) | |
318 | return; | |
319 | rq->clock += delta; | |
320 | update_rq_clock_task(rq, delta); | |
321 | } | |
322 | ||
8f4d37ec PZ |
323 | #ifdef CONFIG_SCHED_HRTICK |
324 | /* | |
325 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 326 | */ |
8f4d37ec | 327 | |
8f4d37ec PZ |
328 | static void hrtick_clear(struct rq *rq) |
329 | { | |
330 | if (hrtimer_active(&rq->hrtick_timer)) | |
331 | hrtimer_cancel(&rq->hrtick_timer); | |
332 | } | |
333 | ||
8f4d37ec PZ |
334 | /* |
335 | * High-resolution timer tick. | |
336 | * Runs from hardirq context with interrupts disabled. | |
337 | */ | |
338 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
339 | { | |
340 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 341 | struct rq_flags rf; |
8f4d37ec PZ |
342 | |
343 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
344 | ||
8a8c69c3 | 345 | rq_lock(rq, &rf); |
3e51f33f | 346 | update_rq_clock(rq); |
8f4d37ec | 347 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 348 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
349 | |
350 | return HRTIMER_NORESTART; | |
351 | } | |
352 | ||
95e904c7 | 353 | #ifdef CONFIG_SMP |
971ee28c | 354 | |
4961b6e1 | 355 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
356 | { |
357 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 358 | |
d5096aa6 | 359 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
360 | } |
361 | ||
31656519 PZ |
362 | /* |
363 | * called from hardirq (IPI) context | |
364 | */ | |
365 | static void __hrtick_start(void *arg) | |
b328ca18 | 366 | { |
31656519 | 367 | struct rq *rq = arg; |
8a8c69c3 | 368 | struct rq_flags rf; |
b328ca18 | 369 | |
8a8c69c3 | 370 | rq_lock(rq, &rf); |
971ee28c | 371 | __hrtick_restart(rq); |
8a8c69c3 | 372 | rq_unlock(rq, &rf); |
b328ca18 PZ |
373 | } |
374 | ||
31656519 PZ |
375 | /* |
376 | * Called to set the hrtick timer state. | |
377 | * | |
378 | * called with rq->lock held and irqs disabled | |
379 | */ | |
029632fb | 380 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 381 | { |
31656519 | 382 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 383 | ktime_t time; |
384 | s64 delta; | |
385 | ||
386 | /* | |
387 | * Don't schedule slices shorter than 10000ns, that just | |
388 | * doesn't make sense and can cause timer DoS. | |
389 | */ | |
390 | delta = max_t(s64, delay, 10000LL); | |
391 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 392 | |
cc584b21 | 393 | hrtimer_set_expires(timer, time); |
31656519 | 394 | |
fd3eafda | 395 | if (rq == this_rq()) |
971ee28c | 396 | __hrtick_restart(rq); |
fd3eafda | 397 | else |
c46fff2a | 398 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
399 | } |
400 | ||
31656519 PZ |
401 | #else |
402 | /* | |
403 | * Called to set the hrtick timer state. | |
404 | * | |
405 | * called with rq->lock held and irqs disabled | |
406 | */ | |
029632fb | 407 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 408 | { |
86893335 WL |
409 | /* |
410 | * Don't schedule slices shorter than 10000ns, that just | |
411 | * doesn't make sense. Rely on vruntime for fairness. | |
412 | */ | |
413 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 414 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 415 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 416 | } |
90b5363a | 417 | |
31656519 | 418 | #endif /* CONFIG_SMP */ |
8f4d37ec | 419 | |
77a021be | 420 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 421 | { |
31656519 | 422 | #ifdef CONFIG_SMP |
545b8c8d | 423 | INIT_CSD(&rq->hrtick_csd, __hrtick_start, rq); |
31656519 | 424 | #endif |
d5096aa6 | 425 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 426 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 427 | } |
006c75f1 | 428 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
429 | static inline void hrtick_clear(struct rq *rq) |
430 | { | |
431 | } | |
432 | ||
77a021be | 433 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
434 | { |
435 | } | |
006c75f1 | 436 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 437 | |
5529578a FW |
438 | /* |
439 | * cmpxchg based fetch_or, macro so it works for different integer types | |
440 | */ | |
441 | #define fetch_or(ptr, mask) \ | |
442 | ({ \ | |
443 | typeof(ptr) _ptr = (ptr); \ | |
444 | typeof(mask) _mask = (mask); \ | |
445 | typeof(*_ptr) _old, _val = *_ptr; \ | |
446 | \ | |
447 | for (;;) { \ | |
448 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
449 | if (_old == _val) \ | |
450 | break; \ | |
451 | _val = _old; \ | |
452 | } \ | |
453 | _old; \ | |
454 | }) | |
455 | ||
e3baac47 | 456 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
457 | /* |
458 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
459 | * this avoids any races wrt polling state changes and thereby avoids | |
460 | * spurious IPIs. | |
461 | */ | |
462 | static bool set_nr_and_not_polling(struct task_struct *p) | |
463 | { | |
464 | struct thread_info *ti = task_thread_info(p); | |
465 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
466 | } | |
e3baac47 PZ |
467 | |
468 | /* | |
469 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
470 | * | |
471 | * If this returns true, then the idle task promises to call | |
472 | * sched_ttwu_pending() and reschedule soon. | |
473 | */ | |
474 | static bool set_nr_if_polling(struct task_struct *p) | |
475 | { | |
476 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 477 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
478 | |
479 | for (;;) { | |
480 | if (!(val & _TIF_POLLING_NRFLAG)) | |
481 | return false; | |
482 | if (val & _TIF_NEED_RESCHED) | |
483 | return true; | |
484 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
485 | if (old == val) | |
486 | break; | |
487 | val = old; | |
488 | } | |
489 | return true; | |
490 | } | |
491 | ||
fd99f91a PZ |
492 | #else |
493 | static bool set_nr_and_not_polling(struct task_struct *p) | |
494 | { | |
495 | set_tsk_need_resched(p); | |
496 | return true; | |
497 | } | |
e3baac47 PZ |
498 | |
499 | #ifdef CONFIG_SMP | |
500 | static bool set_nr_if_polling(struct task_struct *p) | |
501 | { | |
502 | return false; | |
503 | } | |
504 | #endif | |
fd99f91a PZ |
505 | #endif |
506 | ||
07879c6a | 507 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
508 | { |
509 | struct wake_q_node *node = &task->wake_q; | |
510 | ||
511 | /* | |
512 | * Atomically grab the task, if ->wake_q is !nil already it means | |
b19a888c | 513 | * it's already queued (either by us or someone else) and will get the |
76751049 PZ |
514 | * wakeup due to that. |
515 | * | |
4c4e3731 PZ |
516 | * In order to ensure that a pending wakeup will observe our pending |
517 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 518 | */ |
4c4e3731 | 519 | smp_mb__before_atomic(); |
87ff19cb | 520 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 521 | return false; |
76751049 PZ |
522 | |
523 | /* | |
524 | * The head is context local, there can be no concurrency. | |
525 | */ | |
526 | *head->lastp = node; | |
527 | head->lastp = &node->next; | |
07879c6a DB |
528 | return true; |
529 | } | |
530 | ||
531 | /** | |
532 | * wake_q_add() - queue a wakeup for 'later' waking. | |
533 | * @head: the wake_q_head to add @task to | |
534 | * @task: the task to queue for 'later' wakeup | |
535 | * | |
536 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
537 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
538 | * instantly. | |
539 | * | |
540 | * This function must be used as-if it were wake_up_process(); IOW the task | |
541 | * must be ready to be woken at this location. | |
542 | */ | |
543 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
544 | { | |
545 | if (__wake_q_add(head, task)) | |
546 | get_task_struct(task); | |
547 | } | |
548 | ||
549 | /** | |
550 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
551 | * @head: the wake_q_head to add @task to | |
552 | * @task: the task to queue for 'later' wakeup | |
553 | * | |
554 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
555 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
556 | * instantly. | |
557 | * | |
558 | * This function must be used as-if it were wake_up_process(); IOW the task | |
559 | * must be ready to be woken at this location. | |
560 | * | |
561 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
562 | * that already hold reference to @task can call the 'safe' version and trust | |
563 | * wake_q to do the right thing depending whether or not the @task is already | |
564 | * queued for wakeup. | |
565 | */ | |
566 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
567 | { | |
568 | if (!__wake_q_add(head, task)) | |
569 | put_task_struct(task); | |
76751049 PZ |
570 | } |
571 | ||
572 | void wake_up_q(struct wake_q_head *head) | |
573 | { | |
574 | struct wake_q_node *node = head->first; | |
575 | ||
576 | while (node != WAKE_Q_TAIL) { | |
577 | struct task_struct *task; | |
578 | ||
579 | task = container_of(node, struct task_struct, wake_q); | |
580 | BUG_ON(!task); | |
d1ccc66d | 581 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
582 | node = node->next; |
583 | task->wake_q.next = NULL; | |
584 | ||
585 | /* | |
7696f991 AP |
586 | * wake_up_process() executes a full barrier, which pairs with |
587 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
588 | */ |
589 | wake_up_process(task); | |
590 | put_task_struct(task); | |
591 | } | |
592 | } | |
593 | ||
c24d20db | 594 | /* |
8875125e | 595 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
596 | * |
597 | * On UP this means the setting of the need_resched flag, on SMP it | |
598 | * might also involve a cross-CPU call to trigger the scheduler on | |
599 | * the target CPU. | |
600 | */ | |
8875125e | 601 | void resched_curr(struct rq *rq) |
c24d20db | 602 | { |
8875125e | 603 | struct task_struct *curr = rq->curr; |
c24d20db IM |
604 | int cpu; |
605 | ||
8875125e | 606 | lockdep_assert_held(&rq->lock); |
c24d20db | 607 | |
8875125e | 608 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
609 | return; |
610 | ||
8875125e | 611 | cpu = cpu_of(rq); |
fd99f91a | 612 | |
f27dde8d | 613 | if (cpu == smp_processor_id()) { |
8875125e | 614 | set_tsk_need_resched(curr); |
f27dde8d | 615 | set_preempt_need_resched(); |
c24d20db | 616 | return; |
f27dde8d | 617 | } |
c24d20db | 618 | |
8875125e | 619 | if (set_nr_and_not_polling(curr)) |
c24d20db | 620 | smp_send_reschedule(cpu); |
dfc68f29 AL |
621 | else |
622 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
623 | } |
624 | ||
029632fb | 625 | void resched_cpu(int cpu) |
c24d20db IM |
626 | { |
627 | struct rq *rq = cpu_rq(cpu); | |
628 | unsigned long flags; | |
629 | ||
7c2102e5 | 630 | raw_spin_lock_irqsave(&rq->lock, flags); |
a0982dfa PM |
631 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
632 | resched_curr(rq); | |
05fa785c | 633 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 634 | } |
06d8308c | 635 | |
b021fe3e | 636 | #ifdef CONFIG_SMP |
3451d024 | 637 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 638 | /* |
d1ccc66d IM |
639 | * In the semi idle case, use the nearest busy CPU for migrating timers |
640 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
641 | * |
642 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
643 | * selecting an idle CPU will add more delays to the timers than intended |
644 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 645 | */ |
bc7a34b8 | 646 | int get_nohz_timer_target(void) |
83cd4fe2 | 647 | { |
e938b9c9 | 648 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 VP |
649 | struct sched_domain *sd; |
650 | ||
e938b9c9 WL |
651 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
652 | if (!idle_cpu(cpu)) | |
653 | return cpu; | |
654 | default_cpu = cpu; | |
655 | } | |
6201b4d6 | 656 | |
057f3fad | 657 | rcu_read_lock(); |
83cd4fe2 | 658 | for_each_domain(cpu, sd) { |
e938b9c9 WL |
659 | for_each_cpu_and(i, sched_domain_span(sd), |
660 | housekeeping_cpumask(HK_FLAG_TIMER)) { | |
44496922 WL |
661 | if (cpu == i) |
662 | continue; | |
663 | ||
e938b9c9 | 664 | if (!idle_cpu(i)) { |
057f3fad PZ |
665 | cpu = i; |
666 | goto unlock; | |
667 | } | |
668 | } | |
83cd4fe2 | 669 | } |
9642d18e | 670 | |
e938b9c9 WL |
671 | if (default_cpu == -1) |
672 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
673 | cpu = default_cpu; | |
057f3fad PZ |
674 | unlock: |
675 | rcu_read_unlock(); | |
83cd4fe2 VP |
676 | return cpu; |
677 | } | |
d1ccc66d | 678 | |
06d8308c TG |
679 | /* |
680 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
681 | * idle CPU then this timer might expire before the next timer event | |
682 | * which is scheduled to wake up that CPU. In case of a completely | |
683 | * idle system the next event might even be infinite time into the | |
684 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
685 | * leaves the inner idle loop so the newly added timer is taken into | |
686 | * account when the CPU goes back to idle and evaluates the timer | |
687 | * wheel for the next timer event. | |
688 | */ | |
1c20091e | 689 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
690 | { |
691 | struct rq *rq = cpu_rq(cpu); | |
692 | ||
693 | if (cpu == smp_processor_id()) | |
694 | return; | |
695 | ||
67b9ca70 | 696 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 697 | smp_send_reschedule(cpu); |
dfc68f29 AL |
698 | else |
699 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
700 | } |
701 | ||
c5bfece2 | 702 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 703 | { |
53c5fa16 FW |
704 | /* |
705 | * We just need the target to call irq_exit() and re-evaluate | |
706 | * the next tick. The nohz full kick at least implies that. | |
707 | * If needed we can still optimize that later with an | |
708 | * empty IRQ. | |
709 | */ | |
379d9ecb PM |
710 | if (cpu_is_offline(cpu)) |
711 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 712 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
713 | if (cpu != smp_processor_id() || |
714 | tick_nohz_tick_stopped()) | |
53c5fa16 | 715 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
716 | return true; |
717 | } | |
718 | ||
719 | return false; | |
720 | } | |
721 | ||
379d9ecb PM |
722 | /* |
723 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
724 | * caller's responsibility to deal with the lost wakeup, for example, | |
725 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
726 | */ | |
1c20091e FW |
727 | void wake_up_nohz_cpu(int cpu) |
728 | { | |
c5bfece2 | 729 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
730 | wake_up_idle_cpu(cpu); |
731 | } | |
732 | ||
19a1f5ec | 733 | static void nohz_csd_func(void *info) |
45bf76df | 734 | { |
19a1f5ec PZ |
735 | struct rq *rq = info; |
736 | int cpu = cpu_of(rq); | |
737 | unsigned int flags; | |
873b4c65 VG |
738 | |
739 | /* | |
19a1f5ec | 740 | * Release the rq::nohz_csd. |
873b4c65 | 741 | */ |
19a1f5ec PZ |
742 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(cpu)); |
743 | WARN_ON(!(flags & NOHZ_KICK_MASK)); | |
45bf76df | 744 | |
19a1f5ec PZ |
745 | rq->idle_balance = idle_cpu(cpu); |
746 | if (rq->idle_balance && !need_resched()) { | |
747 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
748 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
749 | } | |
2069dd75 PZ |
750 | } |
751 | ||
3451d024 | 752 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 753 | |
ce831b38 | 754 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 755 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 756 | { |
76d92ac3 FW |
757 | int fifo_nr_running; |
758 | ||
759 | /* Deadline tasks, even if single, need the tick */ | |
760 | if (rq->dl.dl_nr_running) | |
761 | return false; | |
762 | ||
1e78cdbd | 763 | /* |
b19a888c | 764 | * If there are more than one RR tasks, we need the tick to affect the |
2548d546 | 765 | * actual RR behaviour. |
1e78cdbd | 766 | */ |
76d92ac3 FW |
767 | if (rq->rt.rr_nr_running) { |
768 | if (rq->rt.rr_nr_running == 1) | |
769 | return true; | |
770 | else | |
771 | return false; | |
1e78cdbd RR |
772 | } |
773 | ||
2548d546 PZ |
774 | /* |
775 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
776 | * forced preemption between FIFO tasks. | |
777 | */ | |
778 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
779 | if (fifo_nr_running) | |
780 | return true; | |
781 | ||
782 | /* | |
783 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
784 | * if there's more than one we need the tick for involuntary | |
785 | * preemption. | |
786 | */ | |
787 | if (rq->nr_running > 1) | |
541b8264 | 788 | return false; |
ce831b38 | 789 | |
541b8264 | 790 | return true; |
ce831b38 FW |
791 | } |
792 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 793 | #endif /* CONFIG_SMP */ |
18d95a28 | 794 | |
a790de99 PT |
795 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
796 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 797 | /* |
8277434e PT |
798 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
799 | * node and @up when leaving it for the final time. | |
800 | * | |
801 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 802 | */ |
029632fb | 803 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 804 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
805 | { |
806 | struct task_group *parent, *child; | |
eb755805 | 807 | int ret; |
c09595f6 | 808 | |
8277434e PT |
809 | parent = from; |
810 | ||
c09595f6 | 811 | down: |
eb755805 PZ |
812 | ret = (*down)(parent, data); |
813 | if (ret) | |
8277434e | 814 | goto out; |
c09595f6 PZ |
815 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
816 | parent = child; | |
817 | goto down; | |
818 | ||
819 | up: | |
820 | continue; | |
821 | } | |
eb755805 | 822 | ret = (*up)(parent, data); |
8277434e PT |
823 | if (ret || parent == from) |
824 | goto out; | |
c09595f6 PZ |
825 | |
826 | child = parent; | |
827 | parent = parent->parent; | |
828 | if (parent) | |
829 | goto up; | |
8277434e | 830 | out: |
eb755805 | 831 | return ret; |
c09595f6 PZ |
832 | } |
833 | ||
029632fb | 834 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 835 | { |
e2b245f8 | 836 | return 0; |
eb755805 | 837 | } |
18d95a28 PZ |
838 | #endif |
839 | ||
9059393e | 840 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 841 | { |
f05998d4 NR |
842 | int prio = p->static_prio - MAX_RT_PRIO; |
843 | struct load_weight *load = &p->se.load; | |
844 | ||
dd41f596 IM |
845 | /* |
846 | * SCHED_IDLE tasks get minimal weight: | |
847 | */ | |
1da1843f | 848 | if (task_has_idle_policy(p)) { |
c8b28116 | 849 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 850 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
851 | return; |
852 | } | |
71f8bd46 | 853 | |
9059393e VG |
854 | /* |
855 | * SCHED_OTHER tasks have to update their load when changing their | |
856 | * weight | |
857 | */ | |
858 | if (update_load && p->sched_class == &fair_sched_class) { | |
859 | reweight_task(p, prio); | |
860 | } else { | |
861 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
862 | load->inv_weight = sched_prio_to_wmult[prio]; | |
863 | } | |
71f8bd46 IM |
864 | } |
865 | ||
69842cba | 866 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
867 | /* |
868 | * Serializes updates of utilization clamp values | |
869 | * | |
870 | * The (slow-path) user-space triggers utilization clamp value updates which | |
871 | * can require updates on (fast-path) scheduler's data structures used to | |
872 | * support enqueue/dequeue operations. | |
873 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
874 | * requests are serialized using a mutex to reduce the risk of conflicting | |
875 | * updates or API abuses. | |
876 | */ | |
877 | static DEFINE_MUTEX(uclamp_mutex); | |
878 | ||
e8f14172 PB |
879 | /* Max allowed minimum utilization */ |
880 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
881 | ||
882 | /* Max allowed maximum utilization */ | |
883 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
884 | ||
13685c4a QY |
885 | /* |
886 | * By default RT tasks run at the maximum performance point/capacity of the | |
887 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
888 | * SCHED_CAPACITY_SCALE. | |
889 | * | |
890 | * This knob allows admins to change the default behavior when uclamp is being | |
891 | * used. In battery powered devices, particularly, running at the maximum | |
892 | * capacity and frequency will increase energy consumption and shorten the | |
893 | * battery life. | |
894 | * | |
895 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
896 | * | |
897 | * This knob will not override the system default sched_util_clamp_min defined | |
898 | * above. | |
899 | */ | |
900 | unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; | |
901 | ||
e8f14172 PB |
902 | /* All clamps are required to be less or equal than these values */ |
903 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 904 | |
46609ce2 QY |
905 | /* |
906 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
907 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
908 | * enqueue/dequeue_task(). | |
909 | * | |
910 | * This allows users to continue to enable uclamp in their kernel config with | |
911 | * minimum uclamp overhead in the fast path. | |
912 | * | |
913 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
914 | * enabled, since we have an actual users that make use of uclamp | |
915 | * functionality. | |
916 | * | |
917 | * The knobs that would enable this static key are: | |
918 | * | |
919 | * * A task modifying its uclamp value with sched_setattr(). | |
920 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
921 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
922 | */ | |
923 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
924 | ||
69842cba PB |
925 | /* Integer rounded range for each bucket */ |
926 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
927 | ||
928 | #define for_each_clamp_id(clamp_id) \ | |
929 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
930 | ||
931 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
932 | { | |
933 | return clamp_value / UCLAMP_BUCKET_DELTA; | |
934 | } | |
935 | ||
7763baac | 936 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
937 | { |
938 | if (clamp_id == UCLAMP_MIN) | |
939 | return 0; | |
940 | return SCHED_CAPACITY_SCALE; | |
941 | } | |
942 | ||
a509a7cd PB |
943 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
944 | unsigned int value, bool user_defined) | |
69842cba PB |
945 | { |
946 | uc_se->value = value; | |
947 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 948 | uc_se->user_defined = user_defined; |
69842cba PB |
949 | } |
950 | ||
e496187d | 951 | static inline unsigned int |
0413d7f3 | 952 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
953 | unsigned int clamp_value) |
954 | { | |
955 | /* | |
956 | * Avoid blocked utilization pushing up the frequency when we go | |
957 | * idle (which drops the max-clamp) by retaining the last known | |
958 | * max-clamp. | |
959 | */ | |
960 | if (clamp_id == UCLAMP_MAX) { | |
961 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
962 | return clamp_value; | |
963 | } | |
964 | ||
965 | return uclamp_none(UCLAMP_MIN); | |
966 | } | |
967 | ||
0413d7f3 | 968 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
969 | unsigned int clamp_value) |
970 | { | |
971 | /* Reset max-clamp retention only on idle exit */ | |
972 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
973 | return; | |
974 | ||
975 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
976 | } | |
977 | ||
69842cba | 978 | static inline |
7763baac | 979 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 980 | unsigned int clamp_value) |
69842cba PB |
981 | { |
982 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
983 | int bucket_id = UCLAMP_BUCKETS - 1; | |
984 | ||
985 | /* | |
986 | * Since both min and max clamps are max aggregated, find the | |
987 | * top most bucket with tasks in. | |
988 | */ | |
989 | for ( ; bucket_id >= 0; bucket_id--) { | |
990 | if (!bucket[bucket_id].tasks) | |
991 | continue; | |
992 | return bucket[bucket_id].value; | |
993 | } | |
994 | ||
995 | /* No tasks -- default clamp values */ | |
e496187d | 996 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
997 | } |
998 | ||
13685c4a QY |
999 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
1000 | { | |
1001 | unsigned int default_util_min; | |
1002 | struct uclamp_se *uc_se; | |
1003 | ||
1004 | lockdep_assert_held(&p->pi_lock); | |
1005 | ||
1006 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1007 | ||
1008 | /* Only sync if user didn't override the default */ | |
1009 | if (uc_se->user_defined) | |
1010 | return; | |
1011 | ||
1012 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1013 | uclamp_se_set(uc_se, default_util_min, false); | |
1014 | } | |
1015 | ||
1016 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1017 | { | |
1018 | struct rq_flags rf; | |
1019 | struct rq *rq; | |
1020 | ||
1021 | if (!rt_task(p)) | |
1022 | return; | |
1023 | ||
1024 | /* Protect updates to p->uclamp_* */ | |
1025 | rq = task_rq_lock(p, &rf); | |
1026 | __uclamp_update_util_min_rt_default(p); | |
1027 | task_rq_unlock(rq, p, &rf); | |
1028 | } | |
1029 | ||
1030 | static void uclamp_sync_util_min_rt_default(void) | |
1031 | { | |
1032 | struct task_struct *g, *p; | |
1033 | ||
1034 | /* | |
1035 | * copy_process() sysctl_uclamp | |
1036 | * uclamp_min_rt = X; | |
1037 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1038 | * // link thread smp_mb__after_spinlock() | |
1039 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1040 | * sched_post_fork() for_each_process_thread() | |
1041 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1042 | * | |
1043 | * Ensures that either sched_post_fork() will observe the new | |
1044 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1045 | * task. | |
1046 | */ | |
1047 | read_lock(&tasklist_lock); | |
1048 | smp_mb__after_spinlock(); | |
1049 | read_unlock(&tasklist_lock); | |
1050 | ||
1051 | rcu_read_lock(); | |
1052 | for_each_process_thread(g, p) | |
1053 | uclamp_update_util_min_rt_default(p); | |
1054 | rcu_read_unlock(); | |
1055 | } | |
1056 | ||
3eac870a | 1057 | static inline struct uclamp_se |
0413d7f3 | 1058 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a PB |
1059 | { |
1060 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; | |
1061 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
1062 | struct uclamp_se uc_max; | |
1063 | ||
1064 | /* | |
1065 | * Tasks in autogroups or root task group will be | |
1066 | * restricted by system defaults. | |
1067 | */ | |
1068 | if (task_group_is_autogroup(task_group(p))) | |
1069 | return uc_req; | |
1070 | if (task_group(p) == &root_task_group) | |
1071 | return uc_req; | |
1072 | ||
1073 | uc_max = task_group(p)->uclamp[clamp_id]; | |
1074 | if (uc_req.value > uc_max.value || !uc_req.user_defined) | |
1075 | return uc_max; | |
1076 | #endif | |
1077 | ||
1078 | return uc_req; | |
1079 | } | |
1080 | ||
e8f14172 PB |
1081 | /* |
1082 | * The effective clamp bucket index of a task depends on, by increasing | |
1083 | * priority: | |
1084 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1085 | * - the task group effective clamp value, for tasks not either in the root |
1086 | * group or in an autogroup | |
e8f14172 PB |
1087 | * - the system default clamp value, defined by the sysadmin |
1088 | */ | |
1089 | static inline struct uclamp_se | |
0413d7f3 | 1090 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1091 | { |
3eac870a | 1092 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1093 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1094 | ||
1095 | /* System default restrictions always apply */ | |
1096 | if (unlikely(uc_req.value > uc_max.value)) | |
1097 | return uc_max; | |
1098 | ||
1099 | return uc_req; | |
1100 | } | |
1101 | ||
686516b5 | 1102 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1103 | { |
1104 | struct uclamp_se uc_eff; | |
1105 | ||
1106 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1107 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1108 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1109 | |
1110 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1111 | ||
686516b5 | 1112 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1113 | } |
1114 | ||
69842cba PB |
1115 | /* |
1116 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1117 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1118 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1119 | * |
1120 | * Tasks can have a task-specific value requested from user-space, track | |
1121 | * within each bucket the maximum value for tasks refcounted in it. | |
1122 | * This "local max aggregation" allows to track the exact "requested" value | |
1123 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1124 | */ |
1125 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1126 | enum uclamp_id clamp_id) |
69842cba PB |
1127 | { |
1128 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1129 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1130 | struct uclamp_bucket *bucket; | |
1131 | ||
1132 | lockdep_assert_held(&rq->lock); | |
1133 | ||
e8f14172 PB |
1134 | /* Update task effective clamp */ |
1135 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1136 | ||
69842cba PB |
1137 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1138 | bucket->tasks++; | |
e8f14172 | 1139 | uc_se->active = true; |
69842cba | 1140 | |
e496187d PB |
1141 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1142 | ||
60daf9c1 PB |
1143 | /* |
1144 | * Local max aggregation: rq buckets always track the max | |
1145 | * "requested" clamp value of its RUNNABLE tasks. | |
1146 | */ | |
1147 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1148 | bucket->value = uc_se->value; | |
1149 | ||
69842cba | 1150 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1151 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1152 | } |
1153 | ||
1154 | /* | |
1155 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1156 | * is released. If this is the last task reference counting the rq's max | |
1157 | * active clamp value, then the rq's clamp value is updated. | |
1158 | * | |
1159 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1160 | * always valid. If it's detected they are not, as defensive programming, | |
1161 | * enforce the expected state and warn. | |
1162 | */ | |
1163 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1164 | enum uclamp_id clamp_id) |
69842cba PB |
1165 | { |
1166 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1167 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1168 | struct uclamp_bucket *bucket; | |
e496187d | 1169 | unsigned int bkt_clamp; |
69842cba PB |
1170 | unsigned int rq_clamp; |
1171 | ||
1172 | lockdep_assert_held(&rq->lock); | |
1173 | ||
46609ce2 QY |
1174 | /* |
1175 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1176 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1177 | * | |
1178 | * In this case the uc_se->active flag should be false since no uclamp | |
1179 | * accounting was performed at enqueue time and we can just return | |
1180 | * here. | |
1181 | * | |
b19a888c | 1182 | * Need to be careful of the following enqueue/dequeue ordering |
46609ce2 QY |
1183 | * problem too |
1184 | * | |
1185 | * enqueue(taskA) | |
1186 | * // sched_uclamp_used gets enabled | |
1187 | * enqueue(taskB) | |
1188 | * dequeue(taskA) | |
b19a888c | 1189 | * // Must not decrement bucket->tasks here |
46609ce2 QY |
1190 | * dequeue(taskB) |
1191 | * | |
1192 | * where we could end up with stale data in uc_se and | |
1193 | * bucket[uc_se->bucket_id]. | |
1194 | * | |
1195 | * The following check here eliminates the possibility of such race. | |
1196 | */ | |
1197 | if (unlikely(!uc_se->active)) | |
1198 | return; | |
1199 | ||
69842cba | 1200 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1201 | |
69842cba PB |
1202 | SCHED_WARN_ON(!bucket->tasks); |
1203 | if (likely(bucket->tasks)) | |
1204 | bucket->tasks--; | |
46609ce2 | 1205 | |
e8f14172 | 1206 | uc_se->active = false; |
69842cba | 1207 | |
60daf9c1 PB |
1208 | /* |
1209 | * Keep "local max aggregation" simple and accept to (possibly) | |
1210 | * overboost some RUNNABLE tasks in the same bucket. | |
1211 | * The rq clamp bucket value is reset to its base value whenever | |
1212 | * there are no more RUNNABLE tasks refcounting it. | |
1213 | */ | |
69842cba PB |
1214 | if (likely(bucket->tasks)) |
1215 | return; | |
1216 | ||
1217 | rq_clamp = READ_ONCE(uc_rq->value); | |
1218 | /* | |
1219 | * Defensive programming: this should never happen. If it happens, | |
1220 | * e.g. due to future modification, warn and fixup the expected value. | |
1221 | */ | |
1222 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1223 | if (bucket->value >= rq_clamp) { |
1224 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1225 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1226 | } | |
69842cba PB |
1227 | } |
1228 | ||
1229 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1230 | { | |
0413d7f3 | 1231 | enum uclamp_id clamp_id; |
69842cba | 1232 | |
46609ce2 QY |
1233 | /* |
1234 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1235 | * | |
1236 | * The condition is constructed such that a NOP is generated when | |
1237 | * sched_uclamp_used is disabled. | |
1238 | */ | |
1239 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1240 | return; | |
1241 | ||
69842cba PB |
1242 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1243 | return; | |
1244 | ||
1245 | for_each_clamp_id(clamp_id) | |
1246 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1247 | |
1248 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1249 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1250 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1251 | } |
1252 | ||
1253 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1254 | { | |
0413d7f3 | 1255 | enum uclamp_id clamp_id; |
69842cba | 1256 | |
46609ce2 QY |
1257 | /* |
1258 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1259 | * | |
1260 | * The condition is constructed such that a NOP is generated when | |
1261 | * sched_uclamp_used is disabled. | |
1262 | */ | |
1263 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1264 | return; | |
1265 | ||
69842cba PB |
1266 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1267 | return; | |
1268 | ||
1269 | for_each_clamp_id(clamp_id) | |
1270 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1271 | } | |
1272 | ||
babbe170 | 1273 | static inline void |
0413d7f3 | 1274 | uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id) |
babbe170 PB |
1275 | { |
1276 | struct rq_flags rf; | |
1277 | struct rq *rq; | |
1278 | ||
1279 | /* | |
1280 | * Lock the task and the rq where the task is (or was) queued. | |
1281 | * | |
1282 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1283 | * price to pay to safely serialize util_{min,max} updates with | |
1284 | * enqueues, dequeues and migration operations. | |
1285 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1286 | */ | |
1287 | rq = task_rq_lock(p, &rf); | |
1288 | ||
1289 | /* | |
1290 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1291 | * If the task is not yet RUNNABLE and its task_struct is not | |
1292 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1293 | * it will already see the updated clamp bucket value. | |
1294 | */ | |
6e1ff077 | 1295 | if (p->uclamp[clamp_id].active) { |
babbe170 PB |
1296 | uclamp_rq_dec_id(rq, p, clamp_id); |
1297 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1298 | } | |
1299 | ||
1300 | task_rq_unlock(rq, p, &rf); | |
1301 | } | |
1302 | ||
e3b8b6a0 | 1303 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 PB |
1304 | static inline void |
1305 | uclamp_update_active_tasks(struct cgroup_subsys_state *css, | |
1306 | unsigned int clamps) | |
1307 | { | |
0413d7f3 | 1308 | enum uclamp_id clamp_id; |
babbe170 PB |
1309 | struct css_task_iter it; |
1310 | struct task_struct *p; | |
babbe170 PB |
1311 | |
1312 | css_task_iter_start(css, 0, &it); | |
1313 | while ((p = css_task_iter_next(&it))) { | |
1314 | for_each_clamp_id(clamp_id) { | |
1315 | if ((0x1 << clamp_id) & clamps) | |
1316 | uclamp_update_active(p, clamp_id); | |
1317 | } | |
1318 | } | |
1319 | css_task_iter_end(&it); | |
1320 | } | |
1321 | ||
7274a5c1 PB |
1322 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1323 | static void uclamp_update_root_tg(void) | |
1324 | { | |
1325 | struct task_group *tg = &root_task_group; | |
1326 | ||
1327 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1328 | sysctl_sched_uclamp_util_min, false); | |
1329 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1330 | sysctl_sched_uclamp_util_max, false); | |
1331 | ||
1332 | rcu_read_lock(); | |
1333 | cpu_util_update_eff(&root_task_group.css); | |
1334 | rcu_read_unlock(); | |
1335 | } | |
1336 | #else | |
1337 | static void uclamp_update_root_tg(void) { } | |
1338 | #endif | |
1339 | ||
e8f14172 | 1340 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1341 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1342 | { |
7274a5c1 | 1343 | bool update_root_tg = false; |
13685c4a | 1344 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1345 | int result; |
1346 | ||
2480c093 | 1347 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1348 | old_min = sysctl_sched_uclamp_util_min; |
1349 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1350 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1351 | |
1352 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1353 | if (result) | |
1354 | goto undo; | |
1355 | if (!write) | |
1356 | goto done; | |
1357 | ||
1358 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1359 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1360 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1361 | ||
e8f14172 PB |
1362 | result = -EINVAL; |
1363 | goto undo; | |
1364 | } | |
1365 | ||
1366 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1367 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1368 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1369 | update_root_tg = true; |
e8f14172 PB |
1370 | } |
1371 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1372 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1373 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1374 | update_root_tg = true; |
e8f14172 PB |
1375 | } |
1376 | ||
46609ce2 QY |
1377 | if (update_root_tg) { |
1378 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1379 | uclamp_update_root_tg(); |
46609ce2 | 1380 | } |
7274a5c1 | 1381 | |
13685c4a QY |
1382 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1383 | static_branch_enable(&sched_uclamp_used); | |
1384 | uclamp_sync_util_min_rt_default(); | |
1385 | } | |
7274a5c1 | 1386 | |
e8f14172 | 1387 | /* |
7274a5c1 PB |
1388 | * We update all RUNNABLE tasks only when task groups are in use. |
1389 | * Otherwise, keep it simple and do just a lazy update at each next | |
1390 | * task enqueue time. | |
e8f14172 | 1391 | */ |
7274a5c1 | 1392 | |
e8f14172 PB |
1393 | goto done; |
1394 | ||
1395 | undo: | |
1396 | sysctl_sched_uclamp_util_min = old_min; | |
1397 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1398 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1399 | done: |
2480c093 | 1400 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1401 | |
1402 | return result; | |
1403 | } | |
1404 | ||
a509a7cd PB |
1405 | static int uclamp_validate(struct task_struct *p, |
1406 | const struct sched_attr *attr) | |
1407 | { | |
480a6ca2 DE |
1408 | int util_min = p->uclamp_req[UCLAMP_MIN].value; |
1409 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd | 1410 | |
480a6ca2 DE |
1411 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { |
1412 | util_min = attr->sched_util_min; | |
a509a7cd | 1413 | |
480a6ca2 DE |
1414 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) |
1415 | return -EINVAL; | |
1416 | } | |
1417 | ||
1418 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1419 | util_max = attr->sched_util_max; | |
1420 | ||
1421 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | |
1422 | return -EINVAL; | |
1423 | } | |
1424 | ||
1425 | if (util_min != -1 && util_max != -1 && util_min > util_max) | |
a509a7cd PB |
1426 | return -EINVAL; |
1427 | ||
e65855a5 QY |
1428 | /* |
1429 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1430 | * | |
1431 | * We need to do that here, because enabling static branches is a | |
1432 | * blocking operation which obviously cannot be done while holding | |
1433 | * scheduler locks. | |
1434 | */ | |
1435 | static_branch_enable(&sched_uclamp_used); | |
1436 | ||
a509a7cd PB |
1437 | return 0; |
1438 | } | |
1439 | ||
480a6ca2 DE |
1440 | static bool uclamp_reset(const struct sched_attr *attr, |
1441 | enum uclamp_id clamp_id, | |
1442 | struct uclamp_se *uc_se) | |
1443 | { | |
1444 | /* Reset on sched class change for a non user-defined clamp value. */ | |
1445 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | |
1446 | !uc_se->user_defined) | |
1447 | return true; | |
1448 | ||
1449 | /* Reset on sched_util_{min,max} == -1. */ | |
1450 | if (clamp_id == UCLAMP_MIN && | |
1451 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | |
1452 | attr->sched_util_min == -1) { | |
1453 | return true; | |
1454 | } | |
1455 | ||
1456 | if (clamp_id == UCLAMP_MAX && | |
1457 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | |
1458 | attr->sched_util_max == -1) { | |
1459 | return true; | |
1460 | } | |
1461 | ||
1462 | return false; | |
1463 | } | |
1464 | ||
a509a7cd PB |
1465 | static void __setscheduler_uclamp(struct task_struct *p, |
1466 | const struct sched_attr *attr) | |
1467 | { | |
0413d7f3 | 1468 | enum uclamp_id clamp_id; |
1a00d999 | 1469 | |
1a00d999 PB |
1470 | for_each_clamp_id(clamp_id) { |
1471 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
480a6ca2 | 1472 | unsigned int value; |
1a00d999 | 1473 | |
480a6ca2 | 1474 | if (!uclamp_reset(attr, clamp_id, uc_se)) |
1a00d999 PB |
1475 | continue; |
1476 | ||
13685c4a QY |
1477 | /* |
1478 | * RT by default have a 100% boost value that could be modified | |
1479 | * at runtime. | |
1480 | */ | |
1a00d999 | 1481 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
480a6ca2 | 1482 | value = sysctl_sched_uclamp_util_min_rt_default; |
13685c4a | 1483 | else |
480a6ca2 DE |
1484 | value = uclamp_none(clamp_id); |
1485 | ||
1486 | uclamp_se_set(uc_se, value, false); | |
1a00d999 | 1487 | |
1a00d999 PB |
1488 | } |
1489 | ||
a509a7cd PB |
1490 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1491 | return; | |
1492 | ||
480a6ca2 DE |
1493 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && |
1494 | attr->sched_util_min != -1) { | |
a509a7cd PB |
1495 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], |
1496 | attr->sched_util_min, true); | |
1497 | } | |
1498 | ||
480a6ca2 DE |
1499 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && |
1500 | attr->sched_util_max != -1) { | |
a509a7cd PB |
1501 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], |
1502 | attr->sched_util_max, true); | |
1503 | } | |
1504 | } | |
1505 | ||
e8f14172 PB |
1506 | static void uclamp_fork(struct task_struct *p) |
1507 | { | |
0413d7f3 | 1508 | enum uclamp_id clamp_id; |
e8f14172 | 1509 | |
13685c4a QY |
1510 | /* |
1511 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1512 | * as the task is still at its early fork stages. | |
1513 | */ | |
e8f14172 PB |
1514 | for_each_clamp_id(clamp_id) |
1515 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1516 | |
1517 | if (likely(!p->sched_reset_on_fork)) | |
1518 | return; | |
1519 | ||
1520 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1521 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1522 | uclamp_none(clamp_id), false); | |
a87498ac | 1523 | } |
e8f14172 PB |
1524 | } |
1525 | ||
13685c4a QY |
1526 | static void uclamp_post_fork(struct task_struct *p) |
1527 | { | |
1528 | uclamp_update_util_min_rt_default(p); | |
1529 | } | |
1530 | ||
d81ae8aa QY |
1531 | static void __init init_uclamp_rq(struct rq *rq) |
1532 | { | |
1533 | enum uclamp_id clamp_id; | |
1534 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1535 | ||
1536 | for_each_clamp_id(clamp_id) { | |
1537 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1538 | .value = uclamp_none(clamp_id) | |
1539 | }; | |
1540 | } | |
1541 | ||
1542 | rq->uclamp_flags = 0; | |
1543 | } | |
1544 | ||
69842cba PB |
1545 | static void __init init_uclamp(void) |
1546 | { | |
e8f14172 | 1547 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1548 | enum uclamp_id clamp_id; |
69842cba PB |
1549 | int cpu; |
1550 | ||
d81ae8aa QY |
1551 | for_each_possible_cpu(cpu) |
1552 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1553 | |
69842cba | 1554 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1555 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1556 | uclamp_none(clamp_id), false); |
69842cba | 1557 | } |
e8f14172 PB |
1558 | |
1559 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1560 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1561 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1562 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1563 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1564 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1565 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1566 | #endif |
1567 | } | |
69842cba PB |
1568 | } |
1569 | ||
1570 | #else /* CONFIG_UCLAMP_TASK */ | |
1571 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1572 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1573 | static inline int uclamp_validate(struct task_struct *p, |
1574 | const struct sched_attr *attr) | |
1575 | { | |
1576 | return -EOPNOTSUPP; | |
1577 | } | |
1578 | static void __setscheduler_uclamp(struct task_struct *p, | |
1579 | const struct sched_attr *attr) { } | |
e8f14172 | 1580 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1581 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1582 | static inline void init_uclamp(void) { } |
1583 | #endif /* CONFIG_UCLAMP_TASK */ | |
1584 | ||
1de64443 | 1585 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1586 | { |
0a67d1ee PZ |
1587 | if (!(flags & ENQUEUE_NOCLOCK)) |
1588 | update_rq_clock(rq); | |
1589 | ||
eb414681 | 1590 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 1591 | sched_info_queued(rq, p); |
eb414681 JW |
1592 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1593 | } | |
0a67d1ee | 1594 | |
69842cba | 1595 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1596 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1597 | } |
1598 | ||
1de64443 | 1599 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1600 | { |
0a67d1ee PZ |
1601 | if (!(flags & DEQUEUE_NOCLOCK)) |
1602 | update_rq_clock(rq); | |
1603 | ||
eb414681 | 1604 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 1605 | sched_info_dequeued(rq, p); |
eb414681 JW |
1606 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1607 | } | |
0a67d1ee | 1608 | |
69842cba | 1609 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1610 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1611 | } |
1612 | ||
029632fb | 1613 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1614 | { |
371fd7e7 | 1615 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
1616 | |
1617 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
1618 | } |
1619 | ||
029632fb | 1620 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1621 | { |
7dd77884 PZ |
1622 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
1623 | ||
371fd7e7 | 1624 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1625 | } |
1626 | ||
14531189 | 1627 | /* |
dd41f596 | 1628 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1629 | */ |
14531189 IM |
1630 | static inline int __normal_prio(struct task_struct *p) |
1631 | { | |
dd41f596 | 1632 | return p->static_prio; |
14531189 IM |
1633 | } |
1634 | ||
b29739f9 IM |
1635 | /* |
1636 | * Calculate the expected normal priority: i.e. priority | |
1637 | * without taking RT-inheritance into account. Might be | |
1638 | * boosted by interactivity modifiers. Changes upon fork, | |
1639 | * setprio syscalls, and whenever the interactivity | |
1640 | * estimator recalculates. | |
1641 | */ | |
36c8b586 | 1642 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1643 | { |
1644 | int prio; | |
1645 | ||
aab03e05 DF |
1646 | if (task_has_dl_policy(p)) |
1647 | prio = MAX_DL_PRIO-1; | |
1648 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
1649 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1650 | else | |
1651 | prio = __normal_prio(p); | |
1652 | return prio; | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * Calculate the current priority, i.e. the priority | |
1657 | * taken into account by the scheduler. This value might | |
1658 | * be boosted by RT tasks, or might be boosted by | |
1659 | * interactivity modifiers. Will be RT if the task got | |
1660 | * RT-boosted. If not then it returns p->normal_prio. | |
1661 | */ | |
36c8b586 | 1662 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1663 | { |
1664 | p->normal_prio = normal_prio(p); | |
1665 | /* | |
1666 | * If we are RT tasks or we were boosted to RT priority, | |
1667 | * keep the priority unchanged. Otherwise, update priority | |
1668 | * to the normal priority: | |
1669 | */ | |
1670 | if (!rt_prio(p->prio)) | |
1671 | return p->normal_prio; | |
1672 | return p->prio; | |
1673 | } | |
1674 | ||
1da177e4 LT |
1675 | /** |
1676 | * task_curr - is this task currently executing on a CPU? | |
1677 | * @p: the task in question. | |
e69f6186 YB |
1678 | * |
1679 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1680 | */ |
36c8b586 | 1681 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1682 | { |
1683 | return cpu_curr(task_cpu(p)) == p; | |
1684 | } | |
1685 | ||
67dfa1b7 | 1686 | /* |
4c9a4bc8 PZ |
1687 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1688 | * use the balance_callback list if you want balancing. | |
1689 | * | |
1690 | * this means any call to check_class_changed() must be followed by a call to | |
1691 | * balance_callback(). | |
67dfa1b7 | 1692 | */ |
cb469845 SR |
1693 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1694 | const struct sched_class *prev_class, | |
da7a735e | 1695 | int oldprio) |
cb469845 SR |
1696 | { |
1697 | if (prev_class != p->sched_class) { | |
1698 | if (prev_class->switched_from) | |
da7a735e | 1699 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1700 | |
da7a735e | 1701 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1702 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1703 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1704 | } |
1705 | ||
029632fb | 1706 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 1707 | { |
aa93cd53 | 1708 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 1709 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
1710 | else if (p->sched_class > rq->curr->sched_class) |
1711 | resched_curr(rq); | |
1e5a7405 PZ |
1712 | |
1713 | /* | |
1714 | * A queue event has occurred, and we're going to schedule. In | |
1715 | * this case, we can save a useless back to back clock update. | |
1716 | */ | |
da0c1e65 | 1717 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 1718 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
1719 | } |
1720 | ||
1da177e4 | 1721 | #ifdef CONFIG_SMP |
175f0e25 | 1722 | |
af449901 PZ |
1723 | static void |
1724 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags); | |
1725 | ||
1726 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
1727 | const struct cpumask *new_mask, | |
1728 | u32 flags); | |
1729 | ||
1730 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
1731 | { | |
1732 | if (likely(!p->migration_disabled)) | |
1733 | return; | |
1734 | ||
1735 | if (p->cpus_ptr != &p->cpus_mask) | |
1736 | return; | |
1737 | ||
1738 | /* | |
1739 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
1740 | */ | |
1741 | __do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE); | |
1742 | } | |
1743 | ||
1744 | void migrate_disable(void) | |
1745 | { | |
3015ef4b TG |
1746 | struct task_struct *p = current; |
1747 | ||
1748 | if (p->migration_disabled) { | |
1749 | p->migration_disabled++; | |
af449901 | 1750 | return; |
3015ef4b | 1751 | } |
af449901 | 1752 | |
3015ef4b TG |
1753 | preempt_disable(); |
1754 | this_rq()->nr_pinned++; | |
1755 | p->migration_disabled = 1; | |
1756 | preempt_enable(); | |
af449901 PZ |
1757 | } |
1758 | EXPORT_SYMBOL_GPL(migrate_disable); | |
1759 | ||
1760 | void migrate_enable(void) | |
1761 | { | |
1762 | struct task_struct *p = current; | |
1763 | ||
6d337eab PZ |
1764 | if (p->migration_disabled > 1) { |
1765 | p->migration_disabled--; | |
af449901 | 1766 | return; |
6d337eab | 1767 | } |
af449901 | 1768 | |
6d337eab PZ |
1769 | /* |
1770 | * Ensure stop_task runs either before or after this, and that | |
1771 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
1772 | */ | |
1773 | preempt_disable(); | |
1774 | if (p->cpus_ptr != &p->cpus_mask) | |
1775 | __set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE); | |
1776 | /* | |
1777 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
1778 | * regular cpus_mask, otherwise things that race (eg. | |
1779 | * select_fallback_rq) get confused. | |
1780 | */ | |
af449901 | 1781 | barrier(); |
6d337eab | 1782 | p->migration_disabled = 0; |
3015ef4b | 1783 | this_rq()->nr_pinned--; |
6d337eab | 1784 | preempt_enable(); |
af449901 PZ |
1785 | } |
1786 | EXPORT_SYMBOL_GPL(migrate_enable); | |
1787 | ||
3015ef4b TG |
1788 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
1789 | { | |
1790 | return rq->nr_pinned; | |
1791 | } | |
1792 | ||
175f0e25 | 1793 | /* |
bee98539 | 1794 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
1795 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
1796 | */ | |
1797 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
1798 | { | |
5ba2ffba | 1799 | /* When not in the task's cpumask, no point in looking further. */ |
3bd37062 | 1800 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
1801 | return false; |
1802 | ||
5ba2ffba PZ |
1803 | /* migrate_disabled() must be allowed to finish. */ |
1804 | if (is_migration_disabled(p)) | |
175f0e25 PZ |
1805 | return cpu_online(cpu); |
1806 | ||
5ba2ffba PZ |
1807 | /* Non kernel threads are not allowed during either online or offline. */ |
1808 | if (!(p->flags & PF_KTHREAD)) | |
1809 | return cpu_active(cpu); | |
1810 | ||
1811 | /* KTHREAD_IS_PER_CPU is always allowed. */ | |
1812 | if (kthread_is_per_cpu(p)) | |
1813 | return cpu_online(cpu); | |
1814 | ||
1815 | /* Regular kernel threads don't get to stay during offline. */ | |
1816 | if (cpu_rq(cpu)->balance_push) | |
1817 | return false; | |
1818 | ||
1819 | /* But are allowed during online. */ | |
1820 | return cpu_online(cpu); | |
175f0e25 PZ |
1821 | } |
1822 | ||
5cc389bc PZ |
1823 | /* |
1824 | * This is how migration works: | |
1825 | * | |
1826 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1827 | * stop_one_cpu(). | |
1828 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1829 | * off the CPU) | |
1830 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1831 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1832 | * it and puts it into the right queue. | |
1833 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1834 | * is done. | |
1835 | */ | |
1836 | ||
1837 | /* | |
1838 | * move_queued_task - move a queued task to new rq. | |
1839 | * | |
1840 | * Returns (locked) new rq. Old rq's lock is released. | |
1841 | */ | |
8a8c69c3 PZ |
1842 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
1843 | struct task_struct *p, int new_cpu) | |
5cc389bc | 1844 | { |
5cc389bc PZ |
1845 | lockdep_assert_held(&rq->lock); |
1846 | ||
58877d34 | 1847 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 1848 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 1849 | rq_unlock(rq, rf); |
5cc389bc PZ |
1850 | |
1851 | rq = cpu_rq(new_cpu); | |
1852 | ||
8a8c69c3 | 1853 | rq_lock(rq, rf); |
5cc389bc | 1854 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 1855 | activate_task(rq, p, 0); |
5cc389bc PZ |
1856 | check_preempt_curr(rq, p, 0); |
1857 | ||
1858 | return rq; | |
1859 | } | |
1860 | ||
1861 | struct migration_arg { | |
6d337eab PZ |
1862 | struct task_struct *task; |
1863 | int dest_cpu; | |
1864 | struct set_affinity_pending *pending; | |
1865 | }; | |
1866 | ||
1867 | struct set_affinity_pending { | |
1868 | refcount_t refs; | |
1869 | struct completion done; | |
1870 | struct cpu_stop_work stop_work; | |
1871 | struct migration_arg arg; | |
5cc389bc PZ |
1872 | }; |
1873 | ||
1874 | /* | |
d1ccc66d | 1875 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
1876 | * this because either it can't run here any more (set_cpus_allowed() |
1877 | * away from this CPU, or CPU going down), or because we're | |
1878 | * attempting to rebalance this task on exec (sched_exec). | |
1879 | * | |
1880 | * So we race with normal scheduler movements, but that's OK, as long | |
1881 | * as the task is no longer on this CPU. | |
5cc389bc | 1882 | */ |
8a8c69c3 PZ |
1883 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
1884 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 1885 | { |
5cc389bc | 1886 | /* Affinity changed (again). */ |
175f0e25 | 1887 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 1888 | return rq; |
5cc389bc | 1889 | |
15ff991e | 1890 | update_rq_clock(rq); |
8a8c69c3 | 1891 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
1892 | |
1893 | return rq; | |
5cc389bc PZ |
1894 | } |
1895 | ||
1896 | /* | |
1897 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1898 | * and performs thread migration by bumping thread off CPU then | |
1899 | * 'pushing' onto another runqueue. | |
1900 | */ | |
1901 | static int migration_cpu_stop(void *data) | |
1902 | { | |
6d337eab | 1903 | struct set_affinity_pending *pending; |
5cc389bc | 1904 | struct migration_arg *arg = data; |
5e16bbc2 | 1905 | struct task_struct *p = arg->task; |
6d337eab | 1906 | int dest_cpu = arg->dest_cpu; |
5e16bbc2 | 1907 | struct rq *rq = this_rq(); |
6d337eab | 1908 | bool complete = false; |
8a8c69c3 | 1909 | struct rq_flags rf; |
5cc389bc PZ |
1910 | |
1911 | /* | |
d1ccc66d IM |
1912 | * The original target CPU might have gone down and we might |
1913 | * be on another CPU but it doesn't matter. | |
5cc389bc | 1914 | */ |
6d337eab | 1915 | local_irq_save(rf.flags); |
5cc389bc PZ |
1916 | /* |
1917 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 1918 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
1919 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
1920 | */ | |
a1488664 | 1921 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
1922 | |
1923 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1924 | rq_lock(rq, &rf); |
6d337eab PZ |
1925 | |
1926 | pending = p->migration_pending; | |
5e16bbc2 PZ |
1927 | /* |
1928 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1929 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1930 | * we're holding p->pi_lock. | |
1931 | */ | |
bf89a304 | 1932 | if (task_rq(p) == rq) { |
6d337eab PZ |
1933 | if (is_migration_disabled(p)) |
1934 | goto out; | |
1935 | ||
1936 | if (pending) { | |
1937 | p->migration_pending = NULL; | |
1938 | complete = true; | |
1939 | } | |
1940 | ||
1941 | /* migrate_enable() -- we must not race against SCA */ | |
1942 | if (dest_cpu < 0) { | |
1943 | /* | |
1944 | * When this was migrate_enable() but we no longer | |
1945 | * have a @pending, a concurrent SCA 'fixed' things | |
1946 | * and we should be valid again. Nothing to do. | |
1947 | */ | |
1948 | if (!pending) { | |
1293771e | 1949 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
1950 | goto out; |
1951 | } | |
1952 | ||
1953 | dest_cpu = cpumask_any_distribute(&p->cpus_mask); | |
1954 | } | |
1955 | ||
bf89a304 | 1956 | if (task_on_rq_queued(p)) |
6d337eab | 1957 | rq = __migrate_task(rq, &rf, p, dest_cpu); |
bf89a304 | 1958 | else |
6d337eab PZ |
1959 | p->wake_cpu = dest_cpu; |
1960 | ||
d707faa6 | 1961 | } else if (dest_cpu < 0 || pending) { |
6d337eab PZ |
1962 | /* |
1963 | * This happens when we get migrated between migrate_enable()'s | |
1964 | * preempt_enable() and scheduling the stopper task. At that | |
1965 | * point we're a regular task again and not current anymore. | |
1966 | * | |
1967 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
1968 | * more likely. | |
1969 | */ | |
1970 | ||
d707faa6 VS |
1971 | /* |
1972 | * The task moved before the stopper got to run. We're holding | |
1973 | * ->pi_lock, so the allowed mask is stable - if it got | |
1974 | * somewhere allowed, we're done. | |
1975 | */ | |
1976 | if (pending && cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { | |
1977 | p->migration_pending = NULL; | |
1978 | complete = true; | |
1979 | goto out; | |
1980 | } | |
1981 | ||
6d337eab PZ |
1982 | /* |
1983 | * When this was migrate_enable() but we no longer have an | |
1984 | * @pending, a concurrent SCA 'fixed' things and we should be | |
1985 | * valid again. Nothing to do. | |
1986 | */ | |
1987 | if (!pending) { | |
1293771e | 1988 | WARN_ON_ONCE(!cpumask_test_cpu(task_cpu(p), &p->cpus_mask)); |
6d337eab PZ |
1989 | goto out; |
1990 | } | |
1991 | ||
1992 | /* | |
1993 | * When migrate_enable() hits a rq mis-match we can't reliably | |
1994 | * determine is_migration_disabled() and so have to chase after | |
1995 | * it. | |
1996 | */ | |
1997 | task_rq_unlock(rq, p, &rf); | |
1998 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
1999 | &pending->arg, &pending->stop_work); | |
2000 | return 0; | |
bf89a304 | 2001 | } |
6d337eab PZ |
2002 | out: |
2003 | task_rq_unlock(rq, p, &rf); | |
2004 | ||
2005 | if (complete) | |
2006 | complete_all(&pending->done); | |
2007 | ||
2008 | /* For pending->{arg,stop_work} */ | |
2009 | pending = arg->pending; | |
2010 | if (pending && refcount_dec_and_test(&pending->refs)) | |
2011 | wake_up_var(&pending->refs); | |
5e16bbc2 | 2012 | |
5cc389bc PZ |
2013 | return 0; |
2014 | } | |
2015 | ||
a7c81556 PZ |
2016 | int push_cpu_stop(void *arg) |
2017 | { | |
2018 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2019 | struct task_struct *p = arg; | |
2020 | ||
2021 | raw_spin_lock_irq(&p->pi_lock); | |
2022 | raw_spin_lock(&rq->lock); | |
2023 | ||
2024 | if (task_rq(p) != rq) | |
2025 | goto out_unlock; | |
2026 | ||
2027 | if (is_migration_disabled(p)) { | |
2028 | p->migration_flags |= MDF_PUSH; | |
2029 | goto out_unlock; | |
2030 | } | |
2031 | ||
2032 | p->migration_flags &= ~MDF_PUSH; | |
2033 | ||
2034 | if (p->sched_class->find_lock_rq) | |
2035 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2036 | |
a7c81556 PZ |
2037 | if (!lowest_rq) |
2038 | goto out_unlock; | |
2039 | ||
2040 | // XXX validate p is still the highest prio task | |
2041 | if (task_rq(p) == rq) { | |
2042 | deactivate_task(rq, p, 0); | |
2043 | set_task_cpu(p, lowest_rq->cpu); | |
2044 | activate_task(lowest_rq, p, 0); | |
2045 | resched_curr(lowest_rq); | |
2046 | } | |
2047 | ||
2048 | double_unlock_balance(rq, lowest_rq); | |
2049 | ||
2050 | out_unlock: | |
2051 | rq->push_busy = false; | |
2052 | raw_spin_unlock(&rq->lock); | |
2053 | raw_spin_unlock_irq(&p->pi_lock); | |
2054 | ||
2055 | put_task_struct(p); | |
5cc389bc PZ |
2056 | return 0; |
2057 | } | |
2058 | ||
c5b28038 PZ |
2059 | /* |
2060 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2061 | * actually call this function. | |
2062 | */ | |
9cfc3e18 | 2063 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 2064 | { |
af449901 PZ |
2065 | if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2066 | p->cpus_ptr = new_mask; | |
2067 | return; | |
2068 | } | |
2069 | ||
3bd37062 | 2070 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
2071 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
2072 | } | |
2073 | ||
9cfc3e18 PZ |
2074 | static void |
2075 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 2076 | { |
6c37067e PZ |
2077 | struct rq *rq = task_rq(p); |
2078 | bool queued, running; | |
2079 | ||
af449901 PZ |
2080 | /* |
2081 | * This here violates the locking rules for affinity, since we're only | |
2082 | * supposed to change these variables while holding both rq->lock and | |
2083 | * p->pi_lock. | |
2084 | * | |
2085 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2086 | * accesses these variables under p->pi_lock and only does so after | |
2087 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2088 | * before finish_task(). | |
2089 | * | |
2090 | * XXX do further audits, this smells like something putrid. | |
2091 | */ | |
2092 | if (flags & SCA_MIGRATE_DISABLE) | |
2093 | SCHED_WARN_ON(!p->on_cpu); | |
2094 | else | |
2095 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2096 | |
2097 | queued = task_on_rq_queued(p); | |
2098 | running = task_current(rq, p); | |
2099 | ||
2100 | if (queued) { | |
2101 | /* | |
2102 | * Because __kthread_bind() calls this on blocked tasks without | |
2103 | * holding rq->lock. | |
2104 | */ | |
2105 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 2106 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2107 | } |
2108 | if (running) | |
2109 | put_prev_task(rq, p); | |
2110 | ||
9cfc3e18 | 2111 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 2112 | |
6c37067e | 2113 | if (queued) |
7134b3e9 | 2114 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2115 | if (running) |
03b7fad1 | 2116 | set_next_task(rq, p); |
c5b28038 PZ |
2117 | } |
2118 | ||
9cfc3e18 PZ |
2119 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2120 | { | |
2121 | __do_set_cpus_allowed(p, new_mask, 0); | |
2122 | } | |
2123 | ||
6d337eab | 2124 | /* |
c777d847 VS |
2125 | * This function is wildly self concurrent; here be dragons. |
2126 | * | |
2127 | * | |
2128 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2129 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2130 | * running, we have to kick it out using the CPU stopper. | |
2131 | * | |
2132 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2133 | * Consider: | |
2134 | * | |
2135 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2136 | * | |
2137 | * P0@CPU0 P1 | |
2138 | * | |
2139 | * migrate_disable(); | |
2140 | * <preempted> | |
2141 | * set_cpus_allowed_ptr(P0, [1]); | |
2142 | * | |
2143 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2144 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2145 | * This means we need the following scheme: | |
2146 | * | |
2147 | * P0@CPU0 P1 | |
2148 | * | |
2149 | * migrate_disable(); | |
2150 | * <preempted> | |
2151 | * set_cpus_allowed_ptr(P0, [1]); | |
2152 | * <blocks> | |
2153 | * <resumes> | |
2154 | * migrate_enable(); | |
2155 | * __set_cpus_allowed_ptr(); | |
2156 | * <wakes local stopper> | |
2157 | * `--> <woken on migration completion> | |
2158 | * | |
2159 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2160 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2161 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2162 | * should come into effect at the end of the Migrate-Disable region is the last | |
2163 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2164 | * but we still need to properly signal those waiting tasks at the appropriate | |
2165 | * moment. | |
2166 | * | |
2167 | * This is implemented using struct set_affinity_pending. The first | |
2168 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2169 | * setup an instance of that struct and install it on the targeted task_struct. | |
2170 | * Any and all further callers will reuse that instance. Those then wait for | |
2171 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2172 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2173 | * | |
2174 | * | |
2175 | * (1) In the cases covered above. There is one more where the completion is | |
2176 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
2177 | * cancels the need for an active migration. Consider: | |
2178 | * | |
2179 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2180 | * | |
2181 | * P0@CPU0 P1 P2 | |
2182 | * | |
2183 | * migrate_disable(); | |
2184 | * <preempted> | |
2185 | * set_cpus_allowed_ptr(P0, [1]); | |
2186 | * <blocks> | |
2187 | * set_cpus_allowed_ptr(P0, [0, 1]); | |
2188 | * <signal completion> | |
2189 | * <awakes> | |
2190 | * | |
2191 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2192 | * pending affinity completion is preceded by an uninstallation of | |
2193 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2194 | */ |
2195 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2196 | int dest_cpu, unsigned int flags) | |
2197 | { | |
2198 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
2199 | struct migration_arg arg = { | |
2200 | .task = p, | |
2201 | .dest_cpu = dest_cpu, | |
2202 | }; | |
2203 | bool complete = false; | |
2204 | ||
2205 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2206 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2207 | struct task_struct *push_task = NULL; |
2208 | ||
2209 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2210 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2211 | rq->push_busy = true; | |
2212 | push_task = get_task_struct(p); | |
2213 | } | |
2214 | ||
6d337eab PZ |
2215 | pending = p->migration_pending; |
2216 | if (pending) { | |
2217 | refcount_inc(&pending->refs); | |
2218 | p->migration_pending = NULL; | |
2219 | complete = true; | |
2220 | } | |
2221 | task_rq_unlock(rq, p, rf); | |
2222 | ||
a7c81556 PZ |
2223 | if (push_task) { |
2224 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2225 | p, &rq->push_work); | |
2226 | } | |
2227 | ||
6d337eab PZ |
2228 | if (complete) |
2229 | goto do_complete; | |
2230 | ||
2231 | return 0; | |
2232 | } | |
2233 | ||
2234 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2235 | /* serialized by p->pi_lock */ | |
2236 | if (!p->migration_pending) { | |
c777d847 | 2237 | /* Install the request */ |
6d337eab PZ |
2238 | refcount_set(&my_pending.refs, 1); |
2239 | init_completion(&my_pending.done); | |
2240 | p->migration_pending = &my_pending; | |
2241 | } else { | |
2242 | pending = p->migration_pending; | |
2243 | refcount_inc(&pending->refs); | |
2244 | } | |
2245 | } | |
2246 | pending = p->migration_pending; | |
2247 | /* | |
2248 | * - !MIGRATE_ENABLE: | |
2249 | * we'll have installed a pending if there wasn't one already. | |
2250 | * | |
2251 | * - MIGRATE_ENABLE: | |
2252 | * we're here because the current CPU isn't matching anymore, | |
2253 | * the only way that can happen is because of a concurrent | |
2254 | * set_cpus_allowed_ptr() call, which should then still be | |
2255 | * pending completion. | |
2256 | * | |
2257 | * Either way, we really should have a @pending here. | |
2258 | */ | |
2259 | if (WARN_ON_ONCE(!pending)) { | |
2260 | task_rq_unlock(rq, p, rf); | |
2261 | return -EINVAL; | |
2262 | } | |
2263 | ||
2264 | if (flags & SCA_MIGRATE_ENABLE) { | |
2265 | ||
2266 | refcount_inc(&pending->refs); /* pending->{arg,stop_work} */ | |
a7c81556 | 2267 | p->migration_flags &= ~MDF_PUSH; |
6d337eab PZ |
2268 | task_rq_unlock(rq, p, rf); |
2269 | ||
2270 | pending->arg = (struct migration_arg) { | |
2271 | .task = p, | |
2272 | .dest_cpu = -1, | |
2273 | .pending = pending, | |
2274 | }; | |
2275 | ||
2276 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2277 | &pending->arg, &pending->stop_work); | |
2278 | ||
2279 | return 0; | |
2280 | } | |
2281 | ||
2282 | if (task_running(rq, p) || p->state == TASK_WAKING) { | |
c777d847 VS |
2283 | /* |
2284 | * Lessen races (and headaches) by delegating | |
2285 | * is_migration_disabled(p) checks to the stopper, which will | |
2286 | * run on the same CPU as said p. | |
2287 | */ | |
6d337eab PZ |
2288 | task_rq_unlock(rq, p, rf); |
2289 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); | |
2290 | ||
2291 | } else { | |
2292 | ||
2293 | if (!is_migration_disabled(p)) { | |
2294 | if (task_on_rq_queued(p)) | |
2295 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2296 | ||
2297 | p->migration_pending = NULL; | |
2298 | complete = true; | |
2299 | } | |
2300 | task_rq_unlock(rq, p, rf); | |
2301 | ||
2302 | do_complete: | |
2303 | if (complete) | |
2304 | complete_all(&pending->done); | |
2305 | } | |
2306 | ||
2307 | wait_for_completion(&pending->done); | |
2308 | ||
2309 | if (refcount_dec_and_test(&pending->refs)) | |
2310 | wake_up_var(&pending->refs); | |
2311 | ||
c777d847 VS |
2312 | /* |
2313 | * Block the original owner of &pending until all subsequent callers | |
2314 | * have seen the completion and decremented the refcount | |
2315 | */ | |
6d337eab PZ |
2316 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2317 | ||
2318 | return 0; | |
2319 | } | |
2320 | ||
5cc389bc PZ |
2321 | /* |
2322 | * Change a given task's CPU affinity. Migrate the thread to a | |
2323 | * proper CPU and schedule it away if the CPU it's executing on | |
2324 | * is removed from the allowed bitmask. | |
2325 | * | |
2326 | * NOTE: the caller must have a valid reference to the task, the | |
2327 | * task must not exit() & deallocate itself prematurely. The | |
2328 | * call is not atomic; no spinlocks may be held. | |
2329 | */ | |
25834c73 | 2330 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
9cfc3e18 PZ |
2331 | const struct cpumask *new_mask, |
2332 | u32 flags) | |
5cc389bc | 2333 | { |
e9d867a6 | 2334 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 2335 | unsigned int dest_cpu; |
eb580751 PZ |
2336 | struct rq_flags rf; |
2337 | struct rq *rq; | |
5cc389bc PZ |
2338 | int ret = 0; |
2339 | ||
eb580751 | 2340 | rq = task_rq_lock(p, &rf); |
a499c3ea | 2341 | update_rq_clock(rq); |
5cc389bc | 2342 | |
af449901 | 2343 | if (p->flags & PF_KTHREAD || is_migration_disabled(p)) { |
e9d867a6 | 2344 | /* |
741ba80f PZ |
2345 | * Kernel threads are allowed on online && !active CPUs, |
2346 | * however, during cpu-hot-unplug, even these might get pushed | |
2347 | * away if not KTHREAD_IS_PER_CPU. | |
af449901 PZ |
2348 | * |
2349 | * Specifically, migration_disabled() tasks must not fail the | |
2350 | * cpumask_any_and_distribute() pick below, esp. so on | |
2351 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2352 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2353 | */ |
2354 | cpu_valid_mask = cpu_online_mask; | |
2355 | } | |
2356 | ||
25834c73 PZ |
2357 | /* |
2358 | * Must re-check here, to close a race against __kthread_bind(), | |
2359 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2360 | */ | |
9cfc3e18 | 2361 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2362 | ret = -EINVAL; |
2363 | goto out; | |
2364 | } | |
2365 | ||
885b3ba4 VS |
2366 | if (!(flags & SCA_MIGRATE_ENABLE)) { |
2367 | if (cpumask_equal(&p->cpus_mask, new_mask)) | |
2368 | goto out; | |
2369 | ||
2370 | if (WARN_ON_ONCE(p == current && | |
2371 | is_migration_disabled(p) && | |
2372 | !cpumask_test_cpu(task_cpu(p), new_mask))) { | |
2373 | ret = -EBUSY; | |
2374 | goto out; | |
2375 | } | |
2376 | } | |
5cc389bc | 2377 | |
46a87b38 PT |
2378 | /* |
2379 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2380 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2381 | * immediately required to distribute the tasks within their new mask. | |
2382 | */ | |
2383 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 2384 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2385 | ret = -EINVAL; |
2386 | goto out; | |
2387 | } | |
2388 | ||
9cfc3e18 | 2389 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 2390 | |
6d337eab | 2391 | return affine_move_task(rq, p, &rf, dest_cpu, flags); |
5cc389bc | 2392 | |
5cc389bc | 2393 | out: |
eb580751 | 2394 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
2395 | |
2396 | return ret; | |
2397 | } | |
25834c73 PZ |
2398 | |
2399 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
2400 | { | |
9cfc3e18 | 2401 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 2402 | } |
5cc389bc PZ |
2403 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2404 | ||
dd41f596 | 2405 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2406 | { |
e2912009 PZ |
2407 | #ifdef CONFIG_SCHED_DEBUG |
2408 | /* | |
2409 | * We should never call set_task_cpu() on a blocked task, | |
2410 | * ttwu() will sort out the placement. | |
2411 | */ | |
077614ee | 2412 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 2413 | !p->on_rq); |
0122ec5b | 2414 | |
3ea94de1 JP |
2415 | /* |
2416 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
2417 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
2418 | * time relying on p->on_rq. | |
2419 | */ | |
2420 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
2421 | p->sched_class == &fair_sched_class && | |
2422 | (p->on_rq && !task_on_rq_migrating(p))); | |
2423 | ||
0122ec5b | 2424 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
2425 | /* |
2426 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
2427 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
2428 | * | |
2429 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 2430 | * see task_group(). |
6c6c54e1 PZ |
2431 | * |
2432 | * Furthermore, all task_rq users should acquire both locks, see | |
2433 | * task_rq_lock(). | |
2434 | */ | |
0122ec5b PZ |
2435 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
2436 | lockdep_is_held(&task_rq(p)->lock))); | |
2437 | #endif | |
4ff9083b PZ |
2438 | /* |
2439 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
2440 | */ | |
2441 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
2442 | |
2443 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
2444 | #endif |
2445 | ||
de1d7286 | 2446 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2447 | |
0c69774e | 2448 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 2449 | if (p->sched_class->migrate_task_rq) |
1327237a | 2450 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 2451 | p->se.nr_migrations++; |
d7822b1e | 2452 | rseq_migrate(p); |
ff303e66 | 2453 | perf_event_task_migrate(p); |
0c69774e | 2454 | } |
dd41f596 IM |
2455 | |
2456 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2457 | } |
2458 | ||
0ad4e3df | 2459 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
2460 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
2461 | { | |
da0c1e65 | 2462 | if (task_on_rq_queued(p)) { |
ac66f547 | 2463 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 2464 | struct rq_flags srf, drf; |
ac66f547 PZ |
2465 | |
2466 | src_rq = task_rq(p); | |
2467 | dst_rq = cpu_rq(cpu); | |
2468 | ||
8a8c69c3 PZ |
2469 | rq_pin_lock(src_rq, &srf); |
2470 | rq_pin_lock(dst_rq, &drf); | |
2471 | ||
ac66f547 PZ |
2472 | deactivate_task(src_rq, p, 0); |
2473 | set_task_cpu(p, cpu); | |
2474 | activate_task(dst_rq, p, 0); | |
2475 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
2476 | |
2477 | rq_unpin_lock(dst_rq, &drf); | |
2478 | rq_unpin_lock(src_rq, &srf); | |
2479 | ||
ac66f547 PZ |
2480 | } else { |
2481 | /* | |
2482 | * Task isn't running anymore; make it appear like we migrated | |
2483 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 2484 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
2485 | */ |
2486 | p->wake_cpu = cpu; | |
2487 | } | |
2488 | } | |
2489 | ||
2490 | struct migration_swap_arg { | |
2491 | struct task_struct *src_task, *dst_task; | |
2492 | int src_cpu, dst_cpu; | |
2493 | }; | |
2494 | ||
2495 | static int migrate_swap_stop(void *data) | |
2496 | { | |
2497 | struct migration_swap_arg *arg = data; | |
2498 | struct rq *src_rq, *dst_rq; | |
2499 | int ret = -EAGAIN; | |
2500 | ||
62694cd5 PZ |
2501 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
2502 | return -EAGAIN; | |
2503 | ||
ac66f547 PZ |
2504 | src_rq = cpu_rq(arg->src_cpu); |
2505 | dst_rq = cpu_rq(arg->dst_cpu); | |
2506 | ||
74602315 PZ |
2507 | double_raw_lock(&arg->src_task->pi_lock, |
2508 | &arg->dst_task->pi_lock); | |
ac66f547 | 2509 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 2510 | |
ac66f547 PZ |
2511 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
2512 | goto unlock; | |
2513 | ||
2514 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
2515 | goto unlock; | |
2516 | ||
3bd37062 | 2517 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
2518 | goto unlock; |
2519 | ||
3bd37062 | 2520 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
2521 | goto unlock; |
2522 | ||
2523 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
2524 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
2525 | ||
2526 | ret = 0; | |
2527 | ||
2528 | unlock: | |
2529 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
2530 | raw_spin_unlock(&arg->dst_task->pi_lock); |
2531 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
2532 | |
2533 | return ret; | |
2534 | } | |
2535 | ||
2536 | /* | |
2537 | * Cross migrate two tasks | |
2538 | */ | |
0ad4e3df SD |
2539 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
2540 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
2541 | { |
2542 | struct migration_swap_arg arg; | |
2543 | int ret = -EINVAL; | |
2544 | ||
ac66f547 PZ |
2545 | arg = (struct migration_swap_arg){ |
2546 | .src_task = cur, | |
0ad4e3df | 2547 | .src_cpu = curr_cpu, |
ac66f547 | 2548 | .dst_task = p, |
0ad4e3df | 2549 | .dst_cpu = target_cpu, |
ac66f547 PZ |
2550 | }; |
2551 | ||
2552 | if (arg.src_cpu == arg.dst_cpu) | |
2553 | goto out; | |
2554 | ||
6acce3ef PZ |
2555 | /* |
2556 | * These three tests are all lockless; this is OK since all of them | |
2557 | * will be re-checked with proper locks held further down the line. | |
2558 | */ | |
ac66f547 PZ |
2559 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
2560 | goto out; | |
2561 | ||
3bd37062 | 2562 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
2563 | goto out; |
2564 | ||
3bd37062 | 2565 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
2566 | goto out; |
2567 | ||
286549dc | 2568 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
2569 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
2570 | ||
2571 | out: | |
ac66f547 PZ |
2572 | return ret; |
2573 | } | |
0ad4e3df | 2574 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 2575 | |
1da177e4 LT |
2576 | /* |
2577 | * wait_task_inactive - wait for a thread to unschedule. | |
2578 | * | |
85ba2d86 RM |
2579 | * If @match_state is nonzero, it's the @p->state value just checked and |
2580 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2581 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2582 | * we return a positive number (its total switch count). If a second call | |
2583 | * a short while later returns the same number, the caller can be sure that | |
2584 | * @p has remained unscheduled the whole time. | |
2585 | * | |
1da177e4 LT |
2586 | * The caller must ensure that the task *will* unschedule sometime soon, |
2587 | * else this function might spin for a *long* time. This function can't | |
2588 | * be called with interrupts off, or it may introduce deadlock with | |
2589 | * smp_call_function() if an IPI is sent by the same process we are | |
2590 | * waiting to become inactive. | |
2591 | */ | |
85ba2d86 | 2592 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 2593 | { |
da0c1e65 | 2594 | int running, queued; |
eb580751 | 2595 | struct rq_flags rf; |
85ba2d86 | 2596 | unsigned long ncsw; |
70b97a7f | 2597 | struct rq *rq; |
1da177e4 | 2598 | |
3a5c359a AK |
2599 | for (;;) { |
2600 | /* | |
2601 | * We do the initial early heuristics without holding | |
2602 | * any task-queue locks at all. We'll only try to get | |
2603 | * the runqueue lock when things look like they will | |
2604 | * work out! | |
2605 | */ | |
2606 | rq = task_rq(p); | |
fa490cfd | 2607 | |
3a5c359a AK |
2608 | /* |
2609 | * If the task is actively running on another CPU | |
2610 | * still, just relax and busy-wait without holding | |
2611 | * any locks. | |
2612 | * | |
2613 | * NOTE! Since we don't hold any locks, it's not | |
2614 | * even sure that "rq" stays as the right runqueue! | |
2615 | * But we don't care, since "task_running()" will | |
2616 | * return false if the runqueue has changed and p | |
2617 | * is actually now running somewhere else! | |
2618 | */ | |
85ba2d86 RM |
2619 | while (task_running(rq, p)) { |
2620 | if (match_state && unlikely(p->state != match_state)) | |
2621 | return 0; | |
3a5c359a | 2622 | cpu_relax(); |
85ba2d86 | 2623 | } |
fa490cfd | 2624 | |
3a5c359a AK |
2625 | /* |
2626 | * Ok, time to look more closely! We need the rq | |
2627 | * lock now, to be *sure*. If we're wrong, we'll | |
2628 | * just go back and repeat. | |
2629 | */ | |
eb580751 | 2630 | rq = task_rq_lock(p, &rf); |
27a9da65 | 2631 | trace_sched_wait_task(p); |
3a5c359a | 2632 | running = task_running(rq, p); |
da0c1e65 | 2633 | queued = task_on_rq_queued(p); |
85ba2d86 | 2634 | ncsw = 0; |
f31e11d8 | 2635 | if (!match_state || p->state == match_state) |
93dcf55f | 2636 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 2637 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 2638 | |
85ba2d86 RM |
2639 | /* |
2640 | * If it changed from the expected state, bail out now. | |
2641 | */ | |
2642 | if (unlikely(!ncsw)) | |
2643 | break; | |
2644 | ||
3a5c359a AK |
2645 | /* |
2646 | * Was it really running after all now that we | |
2647 | * checked with the proper locks actually held? | |
2648 | * | |
2649 | * Oops. Go back and try again.. | |
2650 | */ | |
2651 | if (unlikely(running)) { | |
2652 | cpu_relax(); | |
2653 | continue; | |
2654 | } | |
fa490cfd | 2655 | |
3a5c359a AK |
2656 | /* |
2657 | * It's not enough that it's not actively running, | |
2658 | * it must be off the runqueue _entirely_, and not | |
2659 | * preempted! | |
2660 | * | |
80dd99b3 | 2661 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2662 | * running right now), it's preempted, and we should |
2663 | * yield - it could be a while. | |
2664 | */ | |
da0c1e65 | 2665 | if (unlikely(queued)) { |
8b0e1953 | 2666 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
2667 | |
2668 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2669 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2670 | continue; |
2671 | } | |
fa490cfd | 2672 | |
3a5c359a AK |
2673 | /* |
2674 | * Ahh, all good. It wasn't running, and it wasn't | |
2675 | * runnable, which means that it will never become | |
2676 | * running in the future either. We're all done! | |
2677 | */ | |
2678 | break; | |
2679 | } | |
85ba2d86 RM |
2680 | |
2681 | return ncsw; | |
1da177e4 LT |
2682 | } |
2683 | ||
2684 | /*** | |
2685 | * kick_process - kick a running thread to enter/exit the kernel | |
2686 | * @p: the to-be-kicked thread | |
2687 | * | |
2688 | * Cause a process which is running on another CPU to enter | |
2689 | * kernel-mode, without any delay. (to get signals handled.) | |
2690 | * | |
25985edc | 2691 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2692 | * because all it wants to ensure is that the remote task enters |
2693 | * the kernel. If the IPI races and the task has been migrated | |
2694 | * to another CPU then no harm is done and the purpose has been | |
2695 | * achieved as well. | |
2696 | */ | |
36c8b586 | 2697 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2698 | { |
2699 | int cpu; | |
2700 | ||
2701 | preempt_disable(); | |
2702 | cpu = task_cpu(p); | |
2703 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2704 | smp_send_reschedule(cpu); | |
2705 | preempt_enable(); | |
2706 | } | |
b43e3521 | 2707 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 2708 | |
30da688e | 2709 | /* |
3bd37062 | 2710 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
2711 | * |
2712 | * A few notes on cpu_active vs cpu_online: | |
2713 | * | |
2714 | * - cpu_active must be a subset of cpu_online | |
2715 | * | |
97fb7a0a | 2716 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 2717 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 2718 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
2719 | * see it. |
2720 | * | |
d1ccc66d | 2721 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 2722 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 2723 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
2724 | * off. |
2725 | * | |
2726 | * This means that fallback selection must not select !active CPUs. | |
2727 | * And can assume that any active CPU must be online. Conversely | |
2728 | * select_task_rq() below may allow selection of !active CPUs in order | |
2729 | * to satisfy the above rules. | |
30da688e | 2730 | */ |
5da9a0fb PZ |
2731 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2732 | { | |
aa00d89c TC |
2733 | int nid = cpu_to_node(cpu); |
2734 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
2735 | enum { cpuset, possible, fail } state = cpuset; |
2736 | int dest_cpu; | |
5da9a0fb | 2737 | |
aa00d89c | 2738 | /* |
d1ccc66d IM |
2739 | * If the node that the CPU is on has been offlined, cpu_to_node() |
2740 | * will return -1. There is no CPU on the node, and we should | |
2741 | * select the CPU on the other node. | |
aa00d89c TC |
2742 | */ |
2743 | if (nid != -1) { | |
2744 | nodemask = cpumask_of_node(nid); | |
2745 | ||
2746 | /* Look for allowed, online CPU in same node. */ | |
2747 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
2748 | if (!cpu_active(dest_cpu)) |
2749 | continue; | |
3bd37062 | 2750 | if (cpumask_test_cpu(dest_cpu, p->cpus_ptr)) |
aa00d89c TC |
2751 | return dest_cpu; |
2752 | } | |
2baab4e9 | 2753 | } |
5da9a0fb | 2754 | |
2baab4e9 PZ |
2755 | for (;;) { |
2756 | /* Any allowed, online CPU? */ | |
3bd37062 | 2757 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 2758 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 2759 | continue; |
175f0e25 | 2760 | |
2baab4e9 PZ |
2761 | goto out; |
2762 | } | |
5da9a0fb | 2763 | |
e73e85f0 | 2764 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
2765 | switch (state) { |
2766 | case cpuset: | |
e73e85f0 ON |
2767 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
2768 | cpuset_cpus_allowed_fallback(p); | |
2769 | state = possible; | |
2770 | break; | |
2771 | } | |
df561f66 | 2772 | fallthrough; |
2baab4e9 | 2773 | case possible: |
af449901 PZ |
2774 | /* |
2775 | * XXX When called from select_task_rq() we only | |
2776 | * hold p->pi_lock and again violate locking order. | |
2777 | * | |
2778 | * More yuck to audit. | |
2779 | */ | |
2baab4e9 PZ |
2780 | do_set_cpus_allowed(p, cpu_possible_mask); |
2781 | state = fail; | |
2782 | break; | |
2783 | ||
2784 | case fail: | |
2785 | BUG(); | |
2786 | break; | |
2787 | } | |
2788 | } | |
2789 | ||
2790 | out: | |
2791 | if (state != cpuset) { | |
2792 | /* | |
2793 | * Don't tell them about moving exiting tasks or | |
2794 | * kernel threads (both mm NULL), since they never | |
2795 | * leave kernel. | |
2796 | */ | |
2797 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 2798 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
2799 | task_pid_nr(p), p->comm, cpu); |
2800 | } | |
5da9a0fb PZ |
2801 | } |
2802 | ||
2803 | return dest_cpu; | |
2804 | } | |
2805 | ||
e2912009 | 2806 | /* |
3bd37062 | 2807 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 2808 | */ |
970b13ba | 2809 | static inline |
3aef1551 | 2810 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 2811 | { |
cbce1a68 PZ |
2812 | lockdep_assert_held(&p->pi_lock); |
2813 | ||
af449901 | 2814 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 2815 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 2816 | else |
3bd37062 | 2817 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
2818 | |
2819 | /* | |
2820 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 2821 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 2822 | * CPU. |
e2912009 PZ |
2823 | * |
2824 | * Since this is common to all placement strategies, this lives here. | |
2825 | * | |
2826 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2827 | * not worry about this generic constraint ] | |
2828 | */ | |
7af443ee | 2829 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 2830 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2831 | |
2832 | return cpu; | |
970b13ba | 2833 | } |
09a40af5 | 2834 | |
f5832c19 NP |
2835 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2836 | { | |
ded467dc | 2837 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
2838 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
2839 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2840 | ||
2841 | if (stop) { | |
2842 | /* | |
2843 | * Make it appear like a SCHED_FIFO task, its something | |
2844 | * userspace knows about and won't get confused about. | |
2845 | * | |
2846 | * Also, it will make PI more or less work without too | |
2847 | * much confusion -- but then, stop work should not | |
2848 | * rely on PI working anyway. | |
2849 | */ | |
2850 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2851 | ||
2852 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
2853 | |
2854 | /* | |
2855 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
2856 | * adjust the effective priority of a task. As a result, | |
2857 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
2858 | * which can then trigger wakeups of the stop thread to push | |
2859 | * around the current task. | |
2860 | * | |
2861 | * The stop task itself will never be part of the PI-chain, it | |
2862 | * never blocks, therefore that ->pi_lock recursion is safe. | |
2863 | * Tell lockdep about this by placing the stop->pi_lock in its | |
2864 | * own class. | |
2865 | */ | |
2866 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
2867 | } |
2868 | ||
2869 | cpu_rq(cpu)->stop = stop; | |
2870 | ||
2871 | if (old_stop) { | |
2872 | /* | |
2873 | * Reset it back to a normal scheduling class so that | |
2874 | * it can die in pieces. | |
2875 | */ | |
2876 | old_stop->sched_class = &rt_sched_class; | |
2877 | } | |
2878 | } | |
2879 | ||
74d862b6 | 2880 | #else /* CONFIG_SMP */ |
25834c73 PZ |
2881 | |
2882 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
2883 | const struct cpumask *new_mask, |
2884 | u32 flags) | |
25834c73 PZ |
2885 | { |
2886 | return set_cpus_allowed_ptr(p, new_mask); | |
2887 | } | |
2888 | ||
af449901 PZ |
2889 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
2890 | ||
3015ef4b TG |
2891 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2892 | { | |
2893 | return false; | |
2894 | } | |
2895 | ||
74d862b6 | 2896 | #endif /* !CONFIG_SMP */ |
970b13ba | 2897 | |
d7c01d27 | 2898 | static void |
b84cb5df | 2899 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2900 | { |
4fa8d299 | 2901 | struct rq *rq; |
b84cb5df | 2902 | |
4fa8d299 JP |
2903 | if (!schedstat_enabled()) |
2904 | return; | |
2905 | ||
2906 | rq = this_rq(); | |
d7c01d27 | 2907 | |
4fa8d299 JP |
2908 | #ifdef CONFIG_SMP |
2909 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
2910 | __schedstat_inc(rq->ttwu_local); |
2911 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
2912 | } else { |
2913 | struct sched_domain *sd; | |
2914 | ||
b85c8b71 | 2915 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 2916 | rcu_read_lock(); |
4fa8d299 | 2917 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 2918 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 2919 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
2920 | break; |
2921 | } | |
2922 | } | |
057f3fad | 2923 | rcu_read_unlock(); |
d7c01d27 | 2924 | } |
f339b9dc PZ |
2925 | |
2926 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 2927 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
2928 | #endif /* CONFIG_SMP */ |
2929 | ||
b85c8b71 PZ |
2930 | __schedstat_inc(rq->ttwu_count); |
2931 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
2932 | |
2933 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 2934 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2935 | } |
2936 | ||
23f41eeb PZ |
2937 | /* |
2938 | * Mark the task runnable and perform wakeup-preemption. | |
2939 | */ | |
e7904a28 | 2940 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2941 | struct rq_flags *rf) |
9ed3811a | 2942 | { |
9ed3811a | 2943 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 2944 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
2945 | trace_sched_wakeup(p); |
2946 | ||
9ed3811a | 2947 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
2948 | if (p->sched_class->task_woken) { |
2949 | /* | |
b19a888c | 2950 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 2951 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 2952 | */ |
d8ac8971 | 2953 | rq_unpin_lock(rq, rf); |
9ed3811a | 2954 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2955 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 2956 | } |
9ed3811a | 2957 | |
e69c6341 | 2958 | if (rq->idle_stamp) { |
78becc27 | 2959 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 2960 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 2961 | |
abfafa54 JL |
2962 | update_avg(&rq->avg_idle, delta); |
2963 | ||
2964 | if (rq->avg_idle > max) | |
9ed3811a | 2965 | rq->avg_idle = max; |
abfafa54 | 2966 | |
9ed3811a TH |
2967 | rq->idle_stamp = 0; |
2968 | } | |
2969 | #endif | |
2970 | } | |
2971 | ||
c05fbafb | 2972 | static void |
e7904a28 | 2973 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2974 | struct rq_flags *rf) |
c05fbafb | 2975 | { |
77558e4d | 2976 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 2977 | |
cbce1a68 PZ |
2978 | lockdep_assert_held(&rq->lock); |
2979 | ||
c05fbafb PZ |
2980 | if (p->sched_contributes_to_load) |
2981 | rq->nr_uninterruptible--; | |
b5179ac7 | 2982 | |
dbfb089d | 2983 | #ifdef CONFIG_SMP |
b5179ac7 | 2984 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 2985 | en_flags |= ENQUEUE_MIGRATED; |
ec618b84 | 2986 | else |
c05fbafb | 2987 | #endif |
ec618b84 PZ |
2988 | if (p->in_iowait) { |
2989 | delayacct_blkio_end(p); | |
2990 | atomic_dec(&task_rq(p)->nr_iowait); | |
2991 | } | |
c05fbafb | 2992 | |
1b174a2c | 2993 | activate_task(rq, p, en_flags); |
d8ac8971 | 2994 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
2995 | } |
2996 | ||
2997 | /* | |
58877d34 PZ |
2998 | * Consider @p being inside a wait loop: |
2999 | * | |
3000 | * for (;;) { | |
3001 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
3002 | * | |
3003 | * if (CONDITION) | |
3004 | * break; | |
3005 | * | |
3006 | * schedule(); | |
3007 | * } | |
3008 | * __set_current_state(TASK_RUNNING); | |
3009 | * | |
3010 | * between set_current_state() and schedule(). In this case @p is still | |
3011 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3012 | * an atomic manner. | |
3013 | * | |
3014 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3015 | * then schedule() must still happen and p->state can be changed to | |
3016 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3017 | * need to do a full wakeup with enqueue. | |
3018 | * | |
3019 | * Returns: %true when the wakeup is done, | |
3020 | * %false otherwise. | |
c05fbafb | 3021 | */ |
58877d34 | 3022 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3023 | { |
eb580751 | 3024 | struct rq_flags rf; |
c05fbafb PZ |
3025 | struct rq *rq; |
3026 | int ret = 0; | |
3027 | ||
eb580751 | 3028 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3029 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3030 | /* check_preempt_curr() may use rq clock */ |
3031 | update_rq_clock(rq); | |
d8ac8971 | 3032 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3033 | ret = 1; |
3034 | } | |
eb580751 | 3035 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3036 | |
3037 | return ret; | |
3038 | } | |
3039 | ||
317f3941 | 3040 | #ifdef CONFIG_SMP |
a1488664 | 3041 | void sched_ttwu_pending(void *arg) |
317f3941 | 3042 | { |
a1488664 | 3043 | struct llist_node *llist = arg; |
317f3941 | 3044 | struct rq *rq = this_rq(); |
73215849 | 3045 | struct task_struct *p, *t; |
d8ac8971 | 3046 | struct rq_flags rf; |
317f3941 | 3047 | |
e3baac47 PZ |
3048 | if (!llist) |
3049 | return; | |
3050 | ||
126c2092 PZ |
3051 | /* |
3052 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
3053 | * Races such that false-negatives are possible, since they | |
3054 | * are shorter lived that false-positives would be. | |
3055 | */ | |
3056 | WRITE_ONCE(rq->ttwu_pending, 0); | |
3057 | ||
8a8c69c3 | 3058 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3059 | update_rq_clock(rq); |
317f3941 | 3060 | |
8c4890d1 | 3061 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3062 | if (WARN_ON_ONCE(p->on_cpu)) |
3063 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3064 | ||
3065 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3066 | set_task_cpu(p, cpu_of(rq)); | |
3067 | ||
73215849 | 3068 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3069 | } |
317f3941 | 3070 | |
8a8c69c3 | 3071 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3072 | } |
3073 | ||
b2a02fc4 | 3074 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3075 | { |
b2a02fc4 | 3076 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3077 | |
b2a02fc4 PZ |
3078 | if (!set_nr_if_polling(rq->idle)) |
3079 | arch_send_call_function_single_ipi(cpu); | |
3080 | else | |
3081 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3082 | } |
3083 | ||
2ebb1771 MG |
3084 | /* |
3085 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3086 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3087 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3088 | * of the wakeup instead of the waker. | |
3089 | */ | |
3090 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3091 | { |
e3baac47 PZ |
3092 | struct rq *rq = cpu_rq(cpu); |
3093 | ||
b7e7ade3 PZ |
3094 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3095 | ||
126c2092 | 3096 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3097 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3098 | } |
d6aa8f85 | 3099 | |
f6be8af1 CL |
3100 | void wake_up_if_idle(int cpu) |
3101 | { | |
3102 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3103 | struct rq_flags rf; |
f6be8af1 | 3104 | |
fd7de1e8 AL |
3105 | rcu_read_lock(); |
3106 | ||
3107 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3108 | goto out; | |
f6be8af1 CL |
3109 | |
3110 | if (set_nr_if_polling(rq->idle)) { | |
3111 | trace_sched_wake_idle_without_ipi(cpu); | |
3112 | } else { | |
8a8c69c3 | 3113 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
3114 | if (is_idle_task(rq->curr)) |
3115 | smp_send_reschedule(cpu); | |
d1ccc66d | 3116 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 3117 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 3118 | } |
fd7de1e8 AL |
3119 | |
3120 | out: | |
3121 | rcu_read_unlock(); | |
f6be8af1 CL |
3122 | } |
3123 | ||
39be3501 | 3124 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
3125 | { |
3126 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
3127 | } | |
c6e7bd7a | 3128 | |
2ebb1771 MG |
3129 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
3130 | { | |
5ba2ffba PZ |
3131 | /* |
3132 | * Do not complicate things with the async wake_list while the CPU is | |
3133 | * in hotplug state. | |
3134 | */ | |
3135 | if (!cpu_active(cpu)) | |
3136 | return false; | |
3137 | ||
2ebb1771 MG |
3138 | /* |
3139 | * If the CPU does not share cache, then queue the task on the | |
3140 | * remote rqs wakelist to avoid accessing remote data. | |
3141 | */ | |
3142 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3143 | return true; | |
3144 | ||
3145 | /* | |
3146 | * If the task is descheduling and the only running task on the | |
3147 | * CPU then use the wakelist to offload the task activation to | |
3148 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
3149 | * nr_running is checked to avoid unnecessary task stacking. | |
3150 | */ | |
739f70b4 | 3151 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
3152 | return true; |
3153 | ||
3154 | return false; | |
3155 | } | |
3156 | ||
3157 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3158 | { |
2ebb1771 | 3159 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
3160 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
3161 | return false; | |
3162 | ||
c6e7bd7a | 3163 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3164 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3165 | return true; |
3166 | } | |
3167 | ||
3168 | return false; | |
3169 | } | |
58877d34 PZ |
3170 | |
3171 | #else /* !CONFIG_SMP */ | |
3172 | ||
3173 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3174 | { | |
3175 | return false; | |
3176 | } | |
3177 | ||
d6aa8f85 | 3178 | #endif /* CONFIG_SMP */ |
317f3941 | 3179 | |
b5179ac7 | 3180 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3181 | { |
3182 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3183 | struct rq_flags rf; |
c05fbafb | 3184 | |
2ebb1771 | 3185 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3186 | return; |
317f3941 | 3187 | |
8a8c69c3 | 3188 | rq_lock(rq, &rf); |
77558e4d | 3189 | update_rq_clock(rq); |
d8ac8971 | 3190 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3191 | rq_unlock(rq, &rf); |
9ed3811a TH |
3192 | } |
3193 | ||
8643cda5 PZ |
3194 | /* |
3195 | * Notes on Program-Order guarantees on SMP systems. | |
3196 | * | |
3197 | * MIGRATION | |
3198 | * | |
3199 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3200 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3201 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3202 | * |
3203 | * For migration (of runnable tasks) this is provided by the following means: | |
3204 | * | |
3205 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3206 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3207 | * rq(c1)->lock (if not at the same time, then in that order). | |
3208 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3209 | * | |
7696f991 | 3210 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3211 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3212 | * |
3213 | * Example: | |
3214 | * | |
3215 | * CPU0 CPU1 CPU2 | |
3216 | * | |
3217 | * LOCK rq(0)->lock | |
3218 | * sched-out X | |
3219 | * sched-in Y | |
3220 | * UNLOCK rq(0)->lock | |
3221 | * | |
3222 | * LOCK rq(0)->lock // orders against CPU0 | |
3223 | * dequeue X | |
3224 | * UNLOCK rq(0)->lock | |
3225 | * | |
3226 | * LOCK rq(1)->lock | |
3227 | * enqueue X | |
3228 | * UNLOCK rq(1)->lock | |
3229 | * | |
3230 | * LOCK rq(1)->lock // orders against CPU2 | |
3231 | * sched-out Z | |
3232 | * sched-in X | |
3233 | * UNLOCK rq(1)->lock | |
3234 | * | |
3235 | * | |
3236 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3237 | * | |
3238 | * For blocking we (obviously) need to provide the same guarantee as for | |
3239 | * migration. However the means are completely different as there is no lock | |
3240 | * chain to provide order. Instead we do: | |
3241 | * | |
58877d34 PZ |
3242 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3243 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3244 | * |
3245 | * Example: | |
3246 | * | |
3247 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
3248 | * | |
3249 | * LOCK rq(0)->lock LOCK X->pi_lock | |
3250 | * dequeue X | |
3251 | * sched-out X | |
3252 | * smp_store_release(X->on_cpu, 0); | |
3253 | * | |
1f03e8d2 | 3254 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
3255 | * X->state = WAKING |
3256 | * set_task_cpu(X,2) | |
3257 | * | |
3258 | * LOCK rq(2)->lock | |
3259 | * enqueue X | |
3260 | * X->state = RUNNING | |
3261 | * UNLOCK rq(2)->lock | |
3262 | * | |
3263 | * LOCK rq(2)->lock // orders against CPU1 | |
3264 | * sched-out Z | |
3265 | * sched-in X | |
3266 | * UNLOCK rq(2)->lock | |
3267 | * | |
3268 | * UNLOCK X->pi_lock | |
3269 | * UNLOCK rq(0)->lock | |
3270 | * | |
3271 | * | |
7696f991 AP |
3272 | * However, for wakeups there is a second guarantee we must provide, namely we |
3273 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
3274 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
3275 | */ |
3276 | ||
9ed3811a | 3277 | /** |
1da177e4 | 3278 | * try_to_wake_up - wake up a thread |
9ed3811a | 3279 | * @p: the thread to be awakened |
1da177e4 | 3280 | * @state: the mask of task states that can be woken |
9ed3811a | 3281 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 3282 | * |
58877d34 PZ |
3283 | * Conceptually does: |
3284 | * | |
3285 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 3286 | * |
a2250238 PZ |
3287 | * If the task was not queued/runnable, also place it back on a runqueue. |
3288 | * | |
58877d34 PZ |
3289 | * This function is atomic against schedule() which would dequeue the task. |
3290 | * | |
3291 | * It issues a full memory barrier before accessing @p->state, see the comment | |
3292 | * with set_current_state(). | |
a2250238 | 3293 | * |
58877d34 | 3294 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 3295 | * |
58877d34 PZ |
3296 | * Relies on p->pi_lock stabilizing: |
3297 | * - p->sched_class | |
3298 | * - p->cpus_ptr | |
3299 | * - p->sched_task_group | |
3300 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
3301 | * | |
3302 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
3303 | * Takes rq->lock in: | |
3304 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
3305 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
3306 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
3307 | * | |
3308 | * As a consequence we race really badly with just about everything. See the | |
3309 | * many memory barriers and their comments for details. | |
7696f991 | 3310 | * |
a2250238 PZ |
3311 | * Return: %true if @p->state changes (an actual wakeup was done), |
3312 | * %false otherwise. | |
1da177e4 | 3313 | */ |
e4a52bcb PZ |
3314 | static int |
3315 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 3316 | { |
1da177e4 | 3317 | unsigned long flags; |
c05fbafb | 3318 | int cpu, success = 0; |
2398f2c6 | 3319 | |
e3d85487 | 3320 | preempt_disable(); |
aacedf26 PZ |
3321 | if (p == current) { |
3322 | /* | |
3323 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
3324 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 3325 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
3326 | * without taking any locks. |
3327 | * | |
3328 | * In particular: | |
3329 | * - we rely on Program-Order guarantees for all the ordering, | |
3330 | * - we're serialized against set_special_state() by virtue of | |
3331 | * it disabling IRQs (this allows not taking ->pi_lock). | |
3332 | */ | |
3333 | if (!(p->state & state)) | |
e3d85487 | 3334 | goto out; |
aacedf26 PZ |
3335 | |
3336 | success = 1; | |
aacedf26 PZ |
3337 | trace_sched_waking(p); |
3338 | p->state = TASK_RUNNING; | |
3339 | trace_sched_wakeup(p); | |
3340 | goto out; | |
3341 | } | |
3342 | ||
e0acd0a6 ON |
3343 | /* |
3344 | * If we are going to wake up a thread waiting for CONDITION we | |
3345 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
3346 | * reordered with p->state check below. This pairs with smp_store_mb() |
3347 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 3348 | */ |
013fdb80 | 3349 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 3350 | smp_mb__after_spinlock(); |
e9c84311 | 3351 | if (!(p->state & state)) |
aacedf26 | 3352 | goto unlock; |
1da177e4 | 3353 | |
fbd705a0 PZ |
3354 | trace_sched_waking(p); |
3355 | ||
d1ccc66d IM |
3356 | /* We're going to change ->state: */ |
3357 | success = 1; | |
1da177e4 | 3358 | |
135e8c92 BS |
3359 | /* |
3360 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
3361 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
3362 | * in smp_cond_load_acquire() below. | |
3363 | * | |
3d85b270 AP |
3364 | * sched_ttwu_pending() try_to_wake_up() |
3365 | * STORE p->on_rq = 1 LOAD p->state | |
3366 | * UNLOCK rq->lock | |
3367 | * | |
3368 | * __schedule() (switch to task 'p') | |
3369 | * LOCK rq->lock smp_rmb(); | |
3370 | * smp_mb__after_spinlock(); | |
3371 | * UNLOCK rq->lock | |
135e8c92 BS |
3372 | * |
3373 | * [task p] | |
3d85b270 | 3374 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 3375 | * |
3d85b270 AP |
3376 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3377 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
3378 | * |
3379 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
3380 | */ |
3381 | smp_rmb(); | |
58877d34 | 3382 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 3383 | goto unlock; |
1da177e4 | 3384 | |
1da177e4 | 3385 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
3386 | /* |
3387 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
3388 | * possible to, falsely, observe p->on_cpu == 0. | |
3389 | * | |
3390 | * One must be running (->on_cpu == 1) in order to remove oneself | |
3391 | * from the runqueue. | |
3392 | * | |
3d85b270 AP |
3393 | * __schedule() (switch to task 'p') try_to_wake_up() |
3394 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
3395 | * UNLOCK rq->lock | |
3396 | * | |
3397 | * __schedule() (put 'p' to sleep) | |
3398 | * LOCK rq->lock smp_rmb(); | |
3399 | * smp_mb__after_spinlock(); | |
3400 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 3401 | * |
3d85b270 AP |
3402 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
3403 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
3404 | * |
3405 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
3406 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
3407 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 3408 | */ |
dbfb089d PZ |
3409 | smp_acquire__after_ctrl_dep(); |
3410 | ||
3411 | /* | |
3412 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
3413 | * == 0), which means we need to do an enqueue, change p->state to | |
3414 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
3415 | * enqueue, such as ttwu_queue_wakelist(). | |
3416 | */ | |
3417 | p->state = TASK_WAKING; | |
ecf7d01c | 3418 | |
c6e7bd7a PZ |
3419 | /* |
3420 | * If the owning (remote) CPU is still in the middle of schedule() with | |
3421 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
3422 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
3423 | * let the waker make forward progress. This is safe because IRQs are | |
3424 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
3425 | * |
3426 | * Ensure we load task_cpu(p) after p->on_cpu: | |
3427 | * | |
3428 | * set_task_cpu(p, cpu); | |
3429 | * STORE p->cpu = @cpu | |
3430 | * __schedule() (switch to task 'p') | |
3431 | * LOCK rq->lock | |
3432 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
3433 | * STORE p->on_cpu = 1 LOAD p->cpu | |
3434 | * | |
3435 | * to ensure we observe the correct CPU on which the task is currently | |
3436 | * scheduling. | |
c6e7bd7a | 3437 | */ |
b6e13e85 | 3438 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 3439 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
3440 | goto unlock; |
3441 | ||
e9c84311 | 3442 | /* |
d1ccc66d | 3443 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 3444 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 3445 | * |
31cb1bc0 | 3446 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
3447 | * |
3448 | * This ensures that tasks getting woken will be fully ordered against | |
3449 | * their previous state and preserve Program Order. | |
0970d299 | 3450 | */ |
1f03e8d2 | 3451 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 3452 | |
3aef1551 | 3453 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc | 3454 | if (task_cpu(p) != cpu) { |
ec618b84 PZ |
3455 | if (p->in_iowait) { |
3456 | delayacct_blkio_end(p); | |
3457 | atomic_dec(&task_rq(p)->nr_iowait); | |
3458 | } | |
3459 | ||
f339b9dc | 3460 | wake_flags |= WF_MIGRATED; |
eb414681 | 3461 | psi_ttwu_dequeue(p); |
e4a52bcb | 3462 | set_task_cpu(p, cpu); |
f339b9dc | 3463 | } |
b6e13e85 PZ |
3464 | #else |
3465 | cpu = task_cpu(p); | |
1da177e4 | 3466 | #endif /* CONFIG_SMP */ |
1da177e4 | 3467 | |
b5179ac7 | 3468 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 3469 | unlock: |
013fdb80 | 3470 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
3471 | out: |
3472 | if (success) | |
b6e13e85 | 3473 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 3474 | preempt_enable(); |
1da177e4 LT |
3475 | |
3476 | return success; | |
3477 | } | |
3478 | ||
2beaf328 PM |
3479 | /** |
3480 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
3481 | * @p: Process for which the function is to be invoked. | |
3482 | * @func: Function to invoke. | |
3483 | * @arg: Argument to function. | |
3484 | * | |
3485 | * If the specified task can be quickly locked into a definite state | |
3486 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
3487 | * state while invoking @func(@arg). This function can use ->on_rq and | |
3488 | * task_curr() to work out what the state is, if required. Given that | |
3489 | * @func can be invoked with a runqueue lock held, it had better be quite | |
3490 | * lightweight. | |
3491 | * | |
3492 | * Returns: | |
3493 | * @false if the task slipped out from under the locks. | |
3494 | * @true if the task was locked onto a runqueue or is sleeping. | |
3495 | * However, @func can override this by returning @false. | |
3496 | */ | |
3497 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
3498 | { | |
3499 | bool ret = false; | |
3500 | struct rq_flags rf; | |
3501 | struct rq *rq; | |
3502 | ||
3503 | lockdep_assert_irqs_enabled(); | |
3504 | raw_spin_lock_irq(&p->pi_lock); | |
3505 | if (p->on_rq) { | |
3506 | rq = __task_rq_lock(p, &rf); | |
3507 | if (task_rq(p) == rq) | |
3508 | ret = func(p, arg); | |
3509 | rq_unlock(rq, &rf); | |
3510 | } else { | |
3511 | switch (p->state) { | |
3512 | case TASK_RUNNING: | |
3513 | case TASK_WAKING: | |
3514 | break; | |
3515 | default: | |
3516 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
3517 | if (!p->on_rq) | |
3518 | ret = func(p, arg); | |
3519 | } | |
3520 | } | |
3521 | raw_spin_unlock_irq(&p->pi_lock); | |
3522 | return ret; | |
3523 | } | |
3524 | ||
50fa610a DH |
3525 | /** |
3526 | * wake_up_process - Wake up a specific process | |
3527 | * @p: The process to be woken up. | |
3528 | * | |
3529 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
3530 | * processes. |
3531 | * | |
3532 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 3533 | * |
7696f991 | 3534 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 3535 | */ |
7ad5b3a5 | 3536 | int wake_up_process(struct task_struct *p) |
1da177e4 | 3537 | { |
9067ac85 | 3538 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 3539 | } |
1da177e4 LT |
3540 | EXPORT_SYMBOL(wake_up_process); |
3541 | ||
7ad5b3a5 | 3542 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
3543 | { |
3544 | return try_to_wake_up(p, state, 0); | |
3545 | } | |
3546 | ||
1da177e4 LT |
3547 | /* |
3548 | * Perform scheduler related setup for a newly forked process p. | |
3549 | * p is forked by current. | |
dd41f596 IM |
3550 | * |
3551 | * __sched_fork() is basic setup used by init_idle() too: | |
3552 | */ | |
5e1576ed | 3553 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3554 | { |
fd2f4419 PZ |
3555 | p->on_rq = 0; |
3556 | ||
3557 | p->se.on_rq = 0; | |
dd41f596 IM |
3558 | p->se.exec_start = 0; |
3559 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 3560 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 3561 | p->se.nr_migrations = 0; |
da7a735e | 3562 | p->se.vruntime = 0; |
fd2f4419 | 3563 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 3564 | |
ad936d86 BP |
3565 | #ifdef CONFIG_FAIR_GROUP_SCHED |
3566 | p->se.cfs_rq = NULL; | |
3567 | #endif | |
3568 | ||
6cfb0d5d | 3569 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 3570 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 3571 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 3572 | #endif |
476d139c | 3573 | |
aab03e05 | 3574 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 3575 | init_dl_task_timer(&p->dl); |
209a0cbd | 3576 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 3577 | __dl_clear_params(p); |
aab03e05 | 3578 | |
fa717060 | 3579 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
3580 | p->rt.timeout = 0; |
3581 | p->rt.time_slice = sched_rr_timeslice; | |
3582 | p->rt.on_rq = 0; | |
3583 | p->rt.on_list = 0; | |
476d139c | 3584 | |
e107be36 AK |
3585 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3586 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
3587 | #endif | |
cbee9f88 | 3588 | |
5e1f0f09 MG |
3589 | #ifdef CONFIG_COMPACTION |
3590 | p->capture_control = NULL; | |
3591 | #endif | |
13784475 | 3592 | init_numa_balancing(clone_flags, p); |
a1488664 | 3593 | #ifdef CONFIG_SMP |
8c4890d1 | 3594 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 3595 | p->migration_pending = NULL; |
a1488664 | 3596 | #endif |
dd41f596 IM |
3597 | } |
3598 | ||
2a595721 SD |
3599 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
3600 | ||
1a687c2e | 3601 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 3602 | |
1a687c2e MG |
3603 | void set_numabalancing_state(bool enabled) |
3604 | { | |
3605 | if (enabled) | |
2a595721 | 3606 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 3607 | else |
2a595721 | 3608 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 3609 | } |
54a43d54 AK |
3610 | |
3611 | #ifdef CONFIG_PROC_SYSCTL | |
3612 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 3613 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
3614 | { |
3615 | struct ctl_table t; | |
3616 | int err; | |
2a595721 | 3617 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
3618 | |
3619 | if (write && !capable(CAP_SYS_ADMIN)) | |
3620 | return -EPERM; | |
3621 | ||
3622 | t = *table; | |
3623 | t.data = &state; | |
3624 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3625 | if (err < 0) | |
3626 | return err; | |
3627 | if (write) | |
3628 | set_numabalancing_state(state); | |
3629 | return err; | |
3630 | } | |
3631 | #endif | |
3632 | #endif | |
dd41f596 | 3633 | |
4698f88c JP |
3634 | #ifdef CONFIG_SCHEDSTATS |
3635 | ||
cb251765 | 3636 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 3637 | static bool __initdata __sched_schedstats = false; |
cb251765 | 3638 | |
cb251765 MG |
3639 | static void set_schedstats(bool enabled) |
3640 | { | |
3641 | if (enabled) | |
3642 | static_branch_enable(&sched_schedstats); | |
3643 | else | |
3644 | static_branch_disable(&sched_schedstats); | |
3645 | } | |
3646 | ||
3647 | void force_schedstat_enabled(void) | |
3648 | { | |
3649 | if (!schedstat_enabled()) { | |
3650 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
3651 | static_branch_enable(&sched_schedstats); | |
3652 | } | |
3653 | } | |
3654 | ||
3655 | static int __init setup_schedstats(char *str) | |
3656 | { | |
3657 | int ret = 0; | |
3658 | if (!str) | |
3659 | goto out; | |
3660 | ||
4698f88c JP |
3661 | /* |
3662 | * This code is called before jump labels have been set up, so we can't | |
3663 | * change the static branch directly just yet. Instead set a temporary | |
3664 | * variable so init_schedstats() can do it later. | |
3665 | */ | |
cb251765 | 3666 | if (!strcmp(str, "enable")) { |
4698f88c | 3667 | __sched_schedstats = true; |
cb251765 MG |
3668 | ret = 1; |
3669 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 3670 | __sched_schedstats = false; |
cb251765 MG |
3671 | ret = 1; |
3672 | } | |
3673 | out: | |
3674 | if (!ret) | |
3675 | pr_warn("Unable to parse schedstats=\n"); | |
3676 | ||
3677 | return ret; | |
3678 | } | |
3679 | __setup("schedstats=", setup_schedstats); | |
3680 | ||
4698f88c JP |
3681 | static void __init init_schedstats(void) |
3682 | { | |
3683 | set_schedstats(__sched_schedstats); | |
3684 | } | |
3685 | ||
cb251765 | 3686 | #ifdef CONFIG_PROC_SYSCTL |
32927393 CH |
3687 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
3688 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
3689 | { |
3690 | struct ctl_table t; | |
3691 | int err; | |
3692 | int state = static_branch_likely(&sched_schedstats); | |
3693 | ||
3694 | if (write && !capable(CAP_SYS_ADMIN)) | |
3695 | return -EPERM; | |
3696 | ||
3697 | t = *table; | |
3698 | t.data = &state; | |
3699 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3700 | if (err < 0) | |
3701 | return err; | |
3702 | if (write) | |
3703 | set_schedstats(state); | |
3704 | return err; | |
3705 | } | |
4698f88c JP |
3706 | #endif /* CONFIG_PROC_SYSCTL */ |
3707 | #else /* !CONFIG_SCHEDSTATS */ | |
3708 | static inline void init_schedstats(void) {} | |
3709 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
3710 | |
3711 | /* | |
3712 | * fork()/clone()-time setup: | |
3713 | */ | |
aab03e05 | 3714 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3715 | { |
0122ec5b | 3716 | unsigned long flags; |
dd41f596 | 3717 | |
5e1576ed | 3718 | __sched_fork(clone_flags, p); |
06b83b5f | 3719 | /* |
7dc603c9 | 3720 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
3721 | * nobody will actually run it, and a signal or other external |
3722 | * event cannot wake it up and insert it on the runqueue either. | |
3723 | */ | |
7dc603c9 | 3724 | p->state = TASK_NEW; |
dd41f596 | 3725 | |
c350a04e MG |
3726 | /* |
3727 | * Make sure we do not leak PI boosting priority to the child. | |
3728 | */ | |
3729 | p->prio = current->normal_prio; | |
3730 | ||
e8f14172 PB |
3731 | uclamp_fork(p); |
3732 | ||
b9dc29e7 MG |
3733 | /* |
3734 | * Revert to default priority/policy on fork if requested. | |
3735 | */ | |
3736 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 3737 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 3738 | p->policy = SCHED_NORMAL; |
6c697bdf | 3739 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
3740 | p->rt_priority = 0; |
3741 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
3742 | p->static_prio = NICE_TO_PRIO(0); | |
3743 | ||
3744 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 3745 | set_load_weight(p, false); |
6c697bdf | 3746 | |
b9dc29e7 MG |
3747 | /* |
3748 | * We don't need the reset flag anymore after the fork. It has | |
3749 | * fulfilled its duty: | |
3750 | */ | |
3751 | p->sched_reset_on_fork = 0; | |
3752 | } | |
ca94c442 | 3753 | |
af0fffd9 | 3754 | if (dl_prio(p->prio)) |
aab03e05 | 3755 | return -EAGAIN; |
af0fffd9 | 3756 | else if (rt_prio(p->prio)) |
aab03e05 | 3757 | p->sched_class = &rt_sched_class; |
af0fffd9 | 3758 | else |
2ddbf952 | 3759 | p->sched_class = &fair_sched_class; |
b29739f9 | 3760 | |
7dc603c9 | 3761 | init_entity_runnable_average(&p->se); |
cd29fe6f | 3762 | |
86951599 PZ |
3763 | /* |
3764 | * The child is not yet in the pid-hash so no cgroup attach races, | |
3765 | * and the cgroup is pinned to this child due to cgroup_fork() | |
3766 | * is ran before sched_fork(). | |
3767 | * | |
3768 | * Silence PROVE_RCU. | |
3769 | */ | |
0122ec5b | 3770 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ce3614da | 3771 | rseq_migrate(p); |
e210bffd | 3772 | /* |
d1ccc66d | 3773 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
3774 | * so use __set_task_cpu(). |
3775 | */ | |
af0fffd9 | 3776 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
3777 | if (p->sched_class->task_fork) |
3778 | p->sched_class->task_fork(p); | |
0122ec5b | 3779 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 3780 | |
f6db8347 | 3781 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 3782 | if (likely(sched_info_on())) |
52f17b6c | 3783 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 3784 | #endif |
3ca7a440 PZ |
3785 | #if defined(CONFIG_SMP) |
3786 | p->on_cpu = 0; | |
4866cde0 | 3787 | #endif |
01028747 | 3788 | init_task_preempt_count(p); |
806c09a7 | 3789 | #ifdef CONFIG_SMP |
917b627d | 3790 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 3791 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 3792 | #endif |
aab03e05 | 3793 | return 0; |
1da177e4 LT |
3794 | } |
3795 | ||
13685c4a QY |
3796 | void sched_post_fork(struct task_struct *p) |
3797 | { | |
3798 | uclamp_post_fork(p); | |
3799 | } | |
3800 | ||
332ac17e DF |
3801 | unsigned long to_ratio(u64 period, u64 runtime) |
3802 | { | |
3803 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 3804 | return BW_UNIT; |
332ac17e DF |
3805 | |
3806 | /* | |
3807 | * Doing this here saves a lot of checks in all | |
3808 | * the calling paths, and returning zero seems | |
3809 | * safe for them anyway. | |
3810 | */ | |
3811 | if (period == 0) | |
3812 | return 0; | |
3813 | ||
c52f14d3 | 3814 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
3815 | } |
3816 | ||
1da177e4 LT |
3817 | /* |
3818 | * wake_up_new_task - wake up a newly created task for the first time. | |
3819 | * | |
3820 | * This function will do some initial scheduler statistics housekeeping | |
3821 | * that must be done for every newly created context, then puts the task | |
3822 | * on the runqueue and wakes it. | |
3823 | */ | |
3e51e3ed | 3824 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 3825 | { |
eb580751 | 3826 | struct rq_flags rf; |
dd41f596 | 3827 | struct rq *rq; |
fabf318e | 3828 | |
eb580751 | 3829 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 3830 | p->state = TASK_RUNNING; |
fabf318e PZ |
3831 | #ifdef CONFIG_SMP |
3832 | /* | |
3833 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 3834 | * - cpus_ptr can change in the fork path |
d1ccc66d | 3835 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
3836 | * |
3837 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
3838 | * as we're not fully set-up yet. | |
fabf318e | 3839 | */ |
32e839dd | 3840 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 3841 | rseq_migrate(p); |
3aef1551 | 3842 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 3843 | #endif |
b7fa30c9 | 3844 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 3845 | update_rq_clock(rq); |
d0fe0b9c | 3846 | post_init_entity_util_avg(p); |
0017d735 | 3847 | |
7a57f32a | 3848 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 3849 | trace_sched_wakeup_new(p); |
a7558e01 | 3850 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3851 | #ifdef CONFIG_SMP |
0aaafaab PZ |
3852 | if (p->sched_class->task_woken) { |
3853 | /* | |
b19a888c | 3854 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
3855 | * drop it. |
3856 | */ | |
d8ac8971 | 3857 | rq_unpin_lock(rq, &rf); |
efbbd05a | 3858 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3859 | rq_repin_lock(rq, &rf); |
0aaafaab | 3860 | } |
9a897c5a | 3861 | #endif |
eb580751 | 3862 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
3863 | } |
3864 | ||
e107be36 AK |
3865 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3866 | ||
b7203428 | 3867 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 3868 | |
2ecd9d29 PZ |
3869 | void preempt_notifier_inc(void) |
3870 | { | |
b7203428 | 3871 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
3872 | } |
3873 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
3874 | ||
3875 | void preempt_notifier_dec(void) | |
3876 | { | |
b7203428 | 3877 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
3878 | } |
3879 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
3880 | ||
e107be36 | 3881 | /** |
80dd99b3 | 3882 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3883 | * @notifier: notifier struct to register |
e107be36 AK |
3884 | */ |
3885 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3886 | { | |
b7203428 | 3887 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
3888 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
3889 | ||
e107be36 AK |
3890 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
3891 | } | |
3892 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3893 | ||
3894 | /** | |
3895 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3896 | * @notifier: notifier struct to unregister |
e107be36 | 3897 | * |
d84525a8 | 3898 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
3899 | */ |
3900 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3901 | { | |
3902 | hlist_del(¬ifier->link); | |
3903 | } | |
3904 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3905 | ||
1cde2930 | 3906 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3907 | { |
3908 | struct preempt_notifier *notifier; | |
e107be36 | 3909 | |
b67bfe0d | 3910 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3911 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
3912 | } | |
3913 | ||
1cde2930 PZ |
3914 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
3915 | { | |
b7203428 | 3916 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3917 | __fire_sched_in_preempt_notifiers(curr); |
3918 | } | |
3919 | ||
e107be36 | 3920 | static void |
1cde2930 PZ |
3921 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3922 | struct task_struct *next) | |
e107be36 AK |
3923 | { |
3924 | struct preempt_notifier *notifier; | |
e107be36 | 3925 | |
b67bfe0d | 3926 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3927 | notifier->ops->sched_out(notifier, next); |
3928 | } | |
3929 | ||
1cde2930 PZ |
3930 | static __always_inline void |
3931 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3932 | struct task_struct *next) | |
3933 | { | |
b7203428 | 3934 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3935 | __fire_sched_out_preempt_notifiers(curr, next); |
3936 | } | |
3937 | ||
6d6bc0ad | 3938 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3939 | |
1cde2930 | 3940 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3941 | { |
3942 | } | |
3943 | ||
1cde2930 | 3944 | static inline void |
e107be36 AK |
3945 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3946 | struct task_struct *next) | |
3947 | { | |
3948 | } | |
3949 | ||
6d6bc0ad | 3950 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3951 | |
31cb1bc0 | 3952 | static inline void prepare_task(struct task_struct *next) |
3953 | { | |
3954 | #ifdef CONFIG_SMP | |
3955 | /* | |
3956 | * Claim the task as running, we do this before switching to it | |
3957 | * such that any running task will have this set. | |
58877d34 PZ |
3958 | * |
3959 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 3960 | */ |
58877d34 | 3961 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 3962 | #endif |
3963 | } | |
3964 | ||
3965 | static inline void finish_task(struct task_struct *prev) | |
3966 | { | |
3967 | #ifdef CONFIG_SMP | |
3968 | /* | |
58877d34 PZ |
3969 | * This must be the very last reference to @prev from this CPU. After |
3970 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
3971 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 3972 | * finished. |
3973 | * | |
3974 | * In particular, the load of prev->state in finish_task_switch() must | |
3975 | * happen before this. | |
3976 | * | |
3977 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
3978 | */ | |
3979 | smp_store_release(&prev->on_cpu, 0); | |
3980 | #endif | |
3981 | } | |
3982 | ||
565790d2 PZ |
3983 | #ifdef CONFIG_SMP |
3984 | ||
3985 | static void do_balance_callbacks(struct rq *rq, struct callback_head *head) | |
3986 | { | |
3987 | void (*func)(struct rq *rq); | |
3988 | struct callback_head *next; | |
3989 | ||
3990 | lockdep_assert_held(&rq->lock); | |
3991 | ||
3992 | while (head) { | |
3993 | func = (void (*)(struct rq *))head->func; | |
3994 | next = head->next; | |
3995 | head->next = NULL; | |
3996 | head = next; | |
3997 | ||
3998 | func(rq); | |
3999 | } | |
4000 | } | |
4001 | ||
ae792702 PZ |
4002 | static void balance_push(struct rq *rq); |
4003 | ||
4004 | struct callback_head balance_push_callback = { | |
4005 | .next = NULL, | |
4006 | .func = (void (*)(struct callback_head *))balance_push, | |
4007 | }; | |
4008 | ||
565790d2 PZ |
4009 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) |
4010 | { | |
4011 | struct callback_head *head = rq->balance_callback; | |
4012 | ||
4013 | lockdep_assert_held(&rq->lock); | |
ae792702 | 4014 | if (head) |
565790d2 PZ |
4015 | rq->balance_callback = NULL; |
4016 | ||
4017 | return head; | |
4018 | } | |
4019 | ||
4020 | static void __balance_callbacks(struct rq *rq) | |
4021 | { | |
4022 | do_balance_callbacks(rq, splice_balance_callbacks(rq)); | |
4023 | } | |
4024 | ||
4025 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4026 | { | |
4027 | unsigned long flags; | |
4028 | ||
4029 | if (unlikely(head)) { | |
4030 | raw_spin_lock_irqsave(&rq->lock, flags); | |
4031 | do_balance_callbacks(rq, head); | |
4032 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
4033 | } | |
4034 | } | |
4035 | ||
4036 | #else | |
4037 | ||
4038 | static inline void __balance_callbacks(struct rq *rq) | |
4039 | { | |
4040 | } | |
4041 | ||
4042 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
4043 | { | |
4044 | return NULL; | |
4045 | } | |
4046 | ||
4047 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4048 | { | |
4049 | } | |
4050 | ||
4051 | #endif | |
4052 | ||
269d5992 PZ |
4053 | static inline void |
4054 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4055 | { |
269d5992 PZ |
4056 | /* |
4057 | * Since the runqueue lock will be released by the next | |
4058 | * task (which is an invalid locking op but in the case | |
4059 | * of the scheduler it's an obvious special-case), so we | |
4060 | * do an early lockdep release here: | |
4061 | */ | |
4062 | rq_unpin_lock(rq, rf); | |
5facae4f | 4063 | spin_release(&rq->lock.dep_map, _THIS_IP_); |
31cb1bc0 | 4064 | #ifdef CONFIG_DEBUG_SPINLOCK |
4065 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 4066 | rq->lock.owner = next; |
31cb1bc0 | 4067 | #endif |
269d5992 PZ |
4068 | } |
4069 | ||
4070 | static inline void finish_lock_switch(struct rq *rq) | |
4071 | { | |
31cb1bc0 | 4072 | /* |
4073 | * If we are tracking spinlock dependencies then we have to | |
4074 | * fix up the runqueue lock - which gets 'carried over' from | |
4075 | * prev into current: | |
4076 | */ | |
4077 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
ae792702 | 4078 | __balance_callbacks(rq); |
31cb1bc0 | 4079 | raw_spin_unlock_irq(&rq->lock); |
4080 | } | |
4081 | ||
325ea10c IM |
4082 | /* |
4083 | * NOP if the arch has not defined these: | |
4084 | */ | |
4085 | ||
4086 | #ifndef prepare_arch_switch | |
4087 | # define prepare_arch_switch(next) do { } while (0) | |
4088 | #endif | |
4089 | ||
4090 | #ifndef finish_arch_post_lock_switch | |
4091 | # define finish_arch_post_lock_switch() do { } while (0) | |
4092 | #endif | |
4093 | ||
5fbda3ec TG |
4094 | static inline void kmap_local_sched_out(void) |
4095 | { | |
4096 | #ifdef CONFIG_KMAP_LOCAL | |
4097 | if (unlikely(current->kmap_ctrl.idx)) | |
4098 | __kmap_local_sched_out(); | |
4099 | #endif | |
4100 | } | |
4101 | ||
4102 | static inline void kmap_local_sched_in(void) | |
4103 | { | |
4104 | #ifdef CONFIG_KMAP_LOCAL | |
4105 | if (unlikely(current->kmap_ctrl.idx)) | |
4106 | __kmap_local_sched_in(); | |
4107 | #endif | |
4108 | } | |
4109 | ||
4866cde0 NP |
4110 | /** |
4111 | * prepare_task_switch - prepare to switch tasks | |
4112 | * @rq: the runqueue preparing to switch | |
421cee29 | 4113 | * @prev: the current task that is being switched out |
4866cde0 NP |
4114 | * @next: the task we are going to switch to. |
4115 | * | |
4116 | * This is called with the rq lock held and interrupts off. It must | |
4117 | * be paired with a subsequent finish_task_switch after the context | |
4118 | * switch. | |
4119 | * | |
4120 | * prepare_task_switch sets up locking and calls architecture specific | |
4121 | * hooks. | |
4122 | */ | |
e107be36 AK |
4123 | static inline void |
4124 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
4125 | struct task_struct *next) | |
4866cde0 | 4126 | { |
0ed557aa | 4127 | kcov_prepare_switch(prev); |
43148951 | 4128 | sched_info_switch(rq, prev, next); |
fe4b04fa | 4129 | perf_event_task_sched_out(prev, next); |
d7822b1e | 4130 | rseq_preempt(prev); |
e107be36 | 4131 | fire_sched_out_preempt_notifiers(prev, next); |
5fbda3ec | 4132 | kmap_local_sched_out(); |
31cb1bc0 | 4133 | prepare_task(next); |
4866cde0 NP |
4134 | prepare_arch_switch(next); |
4135 | } | |
4136 | ||
1da177e4 LT |
4137 | /** |
4138 | * finish_task_switch - clean up after a task-switch | |
4139 | * @prev: the thread we just switched away from. | |
4140 | * | |
4866cde0 NP |
4141 | * finish_task_switch must be called after the context switch, paired |
4142 | * with a prepare_task_switch call before the context switch. | |
4143 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
4144 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
4145 | * |
4146 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 4147 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
4148 | * with the lock held can cause deadlocks; see schedule() for |
4149 | * details.) | |
dfa50b60 ON |
4150 | * |
4151 | * The context switch have flipped the stack from under us and restored the | |
4152 | * local variables which were saved when this task called schedule() in the | |
4153 | * past. prev == current is still correct but we need to recalculate this_rq | |
4154 | * because prev may have moved to another CPU. | |
1da177e4 | 4155 | */ |
dfa50b60 | 4156 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
4157 | __releases(rq->lock) |
4158 | { | |
dfa50b60 | 4159 | struct rq *rq = this_rq(); |
1da177e4 | 4160 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 4161 | long prev_state; |
1da177e4 | 4162 | |
609ca066 PZ |
4163 | /* |
4164 | * The previous task will have left us with a preempt_count of 2 | |
4165 | * because it left us after: | |
4166 | * | |
4167 | * schedule() | |
4168 | * preempt_disable(); // 1 | |
4169 | * __schedule() | |
4170 | * raw_spin_lock_irq(&rq->lock) // 2 | |
4171 | * | |
4172 | * Also, see FORK_PREEMPT_COUNT. | |
4173 | */ | |
e2bf1c4b PZ |
4174 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
4175 | "corrupted preempt_count: %s/%d/0x%x\n", | |
4176 | current->comm, current->pid, preempt_count())) | |
4177 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 4178 | |
1da177e4 LT |
4179 | rq->prev_mm = NULL; |
4180 | ||
4181 | /* | |
4182 | * A task struct has one reference for the use as "current". | |
c394cc9f | 4183 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
4184 | * schedule one last time. The schedule call will never return, and |
4185 | * the scheduled task must drop that reference. | |
95913d97 PZ |
4186 | * |
4187 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 4188 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
4189 | * running on another CPU and we could rave with its RUNNING -> DEAD |
4190 | * transition, resulting in a double drop. | |
1da177e4 | 4191 | */ |
55a101f8 | 4192 | prev_state = prev->state; |
bf9fae9f | 4193 | vtime_task_switch(prev); |
a8d757ef | 4194 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 4195 | finish_task(prev); |
4196 | finish_lock_switch(rq); | |
01f23e16 | 4197 | finish_arch_post_lock_switch(); |
0ed557aa | 4198 | kcov_finish_switch(current); |
5fbda3ec TG |
4199 | /* |
4200 | * kmap_local_sched_out() is invoked with rq::lock held and | |
4201 | * interrupts disabled. There is no requirement for that, but the | |
4202 | * sched out code does not have an interrupt enabled section. | |
4203 | * Restoring the maps on sched in does not require interrupts being | |
4204 | * disabled either. | |
4205 | */ | |
4206 | kmap_local_sched_in(); | |
e8fa1362 | 4207 | |
e107be36 | 4208 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 4209 | /* |
70216e18 MD |
4210 | * When switching through a kernel thread, the loop in |
4211 | * membarrier_{private,global}_expedited() may have observed that | |
4212 | * kernel thread and not issued an IPI. It is therefore possible to | |
4213 | * schedule between user->kernel->user threads without passing though | |
4214 | * switch_mm(). Membarrier requires a barrier after storing to | |
4215 | * rq->curr, before returning to userspace, so provide them here: | |
4216 | * | |
4217 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
4218 | * provided by mmdrop(), | |
4219 | * - a sync_core for SYNC_CORE. | |
306e0604 | 4220 | */ |
70216e18 MD |
4221 | if (mm) { |
4222 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 4223 | mmdrop(mm); |
70216e18 | 4224 | } |
1cef1150 PZ |
4225 | if (unlikely(prev_state == TASK_DEAD)) { |
4226 | if (prev->sched_class->task_dead) | |
4227 | prev->sched_class->task_dead(prev); | |
68f24b08 | 4228 | |
1cef1150 PZ |
4229 | /* |
4230 | * Remove function-return probe instances associated with this | |
4231 | * task and put them back on the free list. | |
4232 | */ | |
4233 | kprobe_flush_task(prev); | |
4234 | ||
4235 | /* Task is done with its stack. */ | |
4236 | put_task_stack(prev); | |
4237 | ||
0ff7b2cf | 4238 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 4239 | } |
99e5ada9 | 4240 | |
de734f89 | 4241 | tick_nohz_task_switch(); |
dfa50b60 | 4242 | return rq; |
1da177e4 LT |
4243 | } |
4244 | ||
4245 | /** | |
4246 | * schedule_tail - first thing a freshly forked thread must call. | |
4247 | * @prev: the thread we just switched away from. | |
4248 | */ | |
722a9f92 | 4249 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
4250 | __releases(rq->lock) |
4251 | { | |
1a43a14a | 4252 | struct rq *rq; |
da19ab51 | 4253 | |
609ca066 PZ |
4254 | /* |
4255 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
4256 | * finish_task_switch() for details. | |
4257 | * | |
4258 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
4259 | * and the preempt_enable() will end up enabling preemption (on | |
4260 | * PREEMPT_COUNT kernels). | |
4261 | */ | |
4262 | ||
dfa50b60 | 4263 | rq = finish_task_switch(prev); |
1a43a14a | 4264 | preempt_enable(); |
70b97a7f | 4265 | |
1da177e4 | 4266 | if (current->set_child_tid) |
b488893a | 4267 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
4268 | |
4269 | calculate_sigpending(); | |
1da177e4 LT |
4270 | } |
4271 | ||
4272 | /* | |
dfa50b60 | 4273 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 4274 | */ |
04936948 | 4275 | static __always_inline struct rq * |
70b97a7f | 4276 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 4277 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 4278 | { |
e107be36 | 4279 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 4280 | |
9226d125 ZA |
4281 | /* |
4282 | * For paravirt, this is coupled with an exit in switch_to to | |
4283 | * combine the page table reload and the switch backend into | |
4284 | * one hypercall. | |
4285 | */ | |
224101ed | 4286 | arch_start_context_switch(prev); |
9226d125 | 4287 | |
306e0604 | 4288 | /* |
139d025c PZ |
4289 | * kernel -> kernel lazy + transfer active |
4290 | * user -> kernel lazy + mmgrab() active | |
4291 | * | |
4292 | * kernel -> user switch + mmdrop() active | |
4293 | * user -> user switch | |
306e0604 | 4294 | */ |
139d025c PZ |
4295 | if (!next->mm) { // to kernel |
4296 | enter_lazy_tlb(prev->active_mm, next); | |
4297 | ||
4298 | next->active_mm = prev->active_mm; | |
4299 | if (prev->mm) // from user | |
4300 | mmgrab(prev->active_mm); | |
4301 | else | |
4302 | prev->active_mm = NULL; | |
4303 | } else { // to user | |
227a4aad | 4304 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
4305 | /* |
4306 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 4307 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
4308 | * |
4309 | * The below provides this either through switch_mm(), or in | |
4310 | * case 'prev->active_mm == next->mm' through | |
4311 | * finish_task_switch()'s mmdrop(). | |
4312 | */ | |
139d025c | 4313 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 4314 | |
139d025c PZ |
4315 | if (!prev->mm) { // from kernel |
4316 | /* will mmdrop() in finish_task_switch(). */ | |
4317 | rq->prev_mm = prev->active_mm; | |
4318 | prev->active_mm = NULL; | |
4319 | } | |
1da177e4 | 4320 | } |
92509b73 | 4321 | |
cb42c9a3 | 4322 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 4323 | |
269d5992 | 4324 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
4325 | |
4326 | /* Here we just switch the register state and the stack. */ | |
4327 | switch_to(prev, next, prev); | |
dd41f596 | 4328 | barrier(); |
dfa50b60 ON |
4329 | |
4330 | return finish_task_switch(prev); | |
1da177e4 LT |
4331 | } |
4332 | ||
4333 | /* | |
1c3e8264 | 4334 | * nr_running and nr_context_switches: |
1da177e4 LT |
4335 | * |
4336 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 4337 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
4338 | */ |
4339 | unsigned long nr_running(void) | |
4340 | { | |
4341 | unsigned long i, sum = 0; | |
4342 | ||
4343 | for_each_online_cpu(i) | |
4344 | sum += cpu_rq(i)->nr_running; | |
4345 | ||
4346 | return sum; | |
f711f609 | 4347 | } |
1da177e4 | 4348 | |
2ee507c4 | 4349 | /* |
d1ccc66d | 4350 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
4351 | * |
4352 | * Caution: this function does not check that the caller has disabled | |
4353 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
4354 | * race. The caller is responsible to use it correctly, for example: | |
4355 | * | |
dfcb245e | 4356 | * - from a non-preemptible section (of course) |
00cc1633 DD |
4357 | * |
4358 | * - from a thread that is bound to a single CPU | |
4359 | * | |
4360 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
4361 | */ |
4362 | bool single_task_running(void) | |
4363 | { | |
00cc1633 | 4364 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
4365 | } |
4366 | EXPORT_SYMBOL(single_task_running); | |
4367 | ||
1da177e4 | 4368 | unsigned long long nr_context_switches(void) |
46cb4b7c | 4369 | { |
cc94abfc SR |
4370 | int i; |
4371 | unsigned long long sum = 0; | |
46cb4b7c | 4372 | |
0a945022 | 4373 | for_each_possible_cpu(i) |
1da177e4 | 4374 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 4375 | |
1da177e4 LT |
4376 | return sum; |
4377 | } | |
483b4ee6 | 4378 | |
145d952a DL |
4379 | /* |
4380 | * Consumers of these two interfaces, like for example the cpuidle menu | |
4381 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
4382 | * for a CPU that has IO-wait which might not even end up running the task when | |
4383 | * it does become runnable. | |
4384 | */ | |
4385 | ||
4386 | unsigned long nr_iowait_cpu(int cpu) | |
4387 | { | |
4388 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
4389 | } | |
4390 | ||
e33a9bba | 4391 | /* |
b19a888c | 4392 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
4393 | * |
4394 | * The idea behind IO-wait account is to account the idle time that we could | |
4395 | * have spend running if it were not for IO. That is, if we were to improve the | |
4396 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
4397 | * | |
4398 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
4399 | * idle time as IO-wait, because if the storage were faster, it could've been | |
4400 | * running and we'd not be idle. | |
4401 | * | |
4402 | * This has been extended to SMP, by doing the same for each CPU. This however | |
4403 | * is broken. | |
4404 | * | |
4405 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
4406 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
4407 | * though, if the storage were faster, both could've ran at the same time, | |
4408 | * utilising both CPUs. | |
4409 | * | |
4410 | * This means, that when looking globally, the current IO-wait accounting on | |
4411 | * SMP is a lower bound, by reason of under accounting. | |
4412 | * | |
4413 | * Worse, since the numbers are provided per CPU, they are sometimes | |
4414 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
4415 | * associated with any one particular CPU, it can wake to another CPU than it | |
4416 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
4417 | * | |
4418 | * Task CPU affinities can make all that even more 'interesting'. | |
4419 | */ | |
4420 | ||
1da177e4 LT |
4421 | unsigned long nr_iowait(void) |
4422 | { | |
4423 | unsigned long i, sum = 0; | |
483b4ee6 | 4424 | |
0a945022 | 4425 | for_each_possible_cpu(i) |
145d952a | 4426 | sum += nr_iowait_cpu(i); |
46cb4b7c | 4427 | |
1da177e4 LT |
4428 | return sum; |
4429 | } | |
483b4ee6 | 4430 | |
dd41f596 | 4431 | #ifdef CONFIG_SMP |
8a0be9ef | 4432 | |
46cb4b7c | 4433 | /* |
38022906 PZ |
4434 | * sched_exec - execve() is a valuable balancing opportunity, because at |
4435 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 4436 | */ |
38022906 | 4437 | void sched_exec(void) |
46cb4b7c | 4438 | { |
38022906 | 4439 | struct task_struct *p = current; |
1da177e4 | 4440 | unsigned long flags; |
0017d735 | 4441 | int dest_cpu; |
46cb4b7c | 4442 | |
8f42ced9 | 4443 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 4444 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
4445 | if (dest_cpu == smp_processor_id()) |
4446 | goto unlock; | |
38022906 | 4447 | |
8f42ced9 | 4448 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 4449 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 4450 | |
8f42ced9 PZ |
4451 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
4452 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
4453 | return; |
4454 | } | |
0017d735 | 4455 | unlock: |
8f42ced9 | 4456 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 4457 | } |
dd41f596 | 4458 | |
1da177e4 LT |
4459 | #endif |
4460 | ||
1da177e4 | 4461 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 4462 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
4463 | |
4464 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 4465 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 4466 | |
6075620b GG |
4467 | /* |
4468 | * The function fair_sched_class.update_curr accesses the struct curr | |
4469 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
4470 | * we observe a high rate of cache misses in practice. | |
4471 | * Prefetching this data results in improved performance. | |
4472 | */ | |
4473 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
4474 | { | |
4475 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
4476 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
4477 | #else | |
4478 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
4479 | #endif | |
4480 | prefetch(curr); | |
4481 | prefetch(&curr->exec_start); | |
4482 | } | |
4483 | ||
c5f8d995 HS |
4484 | /* |
4485 | * Return accounted runtime for the task. | |
4486 | * In case the task is currently running, return the runtime plus current's | |
4487 | * pending runtime that have not been accounted yet. | |
4488 | */ | |
4489 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4490 | { | |
eb580751 | 4491 | struct rq_flags rf; |
c5f8d995 | 4492 | struct rq *rq; |
6e998916 | 4493 | u64 ns; |
c5f8d995 | 4494 | |
911b2898 PZ |
4495 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
4496 | /* | |
97fb7a0a | 4497 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
4498 | * So we have a optimization chance when the task's delta_exec is 0. |
4499 | * Reading ->on_cpu is racy, but this is ok. | |
4500 | * | |
d1ccc66d IM |
4501 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
4502 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 4503 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
4504 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
4505 | * been accounted, so we're correct here as well. | |
911b2898 | 4506 | */ |
da0c1e65 | 4507 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
4508 | return p->se.sum_exec_runtime; |
4509 | #endif | |
4510 | ||
eb580751 | 4511 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
4512 | /* |
4513 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
4514 | * project cycles that may never be accounted to this | |
4515 | * thread, breaking clock_gettime(). | |
4516 | */ | |
4517 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 4518 | prefetch_curr_exec_start(p); |
6e998916 SG |
4519 | update_rq_clock(rq); |
4520 | p->sched_class->update_curr(rq); | |
4521 | } | |
4522 | ns = p->se.sum_exec_runtime; | |
eb580751 | 4523 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
4524 | |
4525 | return ns; | |
4526 | } | |
48f24c4d | 4527 | |
7835b98b CL |
4528 | /* |
4529 | * This function gets called by the timer code, with HZ frequency. | |
4530 | * We call it with interrupts disabled. | |
7835b98b CL |
4531 | */ |
4532 | void scheduler_tick(void) | |
4533 | { | |
7835b98b CL |
4534 | int cpu = smp_processor_id(); |
4535 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4536 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 4537 | struct rq_flags rf; |
b4eccf5f | 4538 | unsigned long thermal_pressure; |
3e51f33f | 4539 | |
1567c3e3 | 4540 | arch_scale_freq_tick(); |
3e51f33f | 4541 | sched_clock_tick(); |
dd41f596 | 4542 | |
8a8c69c3 PZ |
4543 | rq_lock(rq, &rf); |
4544 | ||
3e51f33f | 4545 | update_rq_clock(rq); |
b4eccf5f | 4546 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 4547 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 4548 | curr->sched_class->task_tick(rq, curr, 0); |
3289bdb4 | 4549 | calc_global_load_tick(rq); |
eb414681 | 4550 | psi_task_tick(rq); |
8a8c69c3 PZ |
4551 | |
4552 | rq_unlock(rq, &rf); | |
7835b98b | 4553 | |
e9d2b064 | 4554 | perf_event_task_tick(); |
e220d2dc | 4555 | |
e418e1c2 | 4556 | #ifdef CONFIG_SMP |
6eb57e0d | 4557 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 4558 | trigger_load_balance(rq); |
e418e1c2 | 4559 | #endif |
1da177e4 LT |
4560 | } |
4561 | ||
265f22a9 | 4562 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
4563 | |
4564 | struct tick_work { | |
4565 | int cpu; | |
b55bd585 | 4566 | atomic_t state; |
d84b3131 FW |
4567 | struct delayed_work work; |
4568 | }; | |
b55bd585 PM |
4569 | /* Values for ->state, see diagram below. */ |
4570 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
4571 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
4572 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
4573 | ||
4574 | /* | |
4575 | * State diagram for ->state: | |
4576 | * | |
4577 | * | |
4578 | * TICK_SCHED_REMOTE_OFFLINE | |
4579 | * | ^ | |
4580 | * | | | |
4581 | * | | sched_tick_remote() | |
4582 | * | | | |
4583 | * | | | |
4584 | * +--TICK_SCHED_REMOTE_OFFLINING | |
4585 | * | ^ | |
4586 | * | | | |
4587 | * sched_tick_start() | | sched_tick_stop() | |
4588 | * | | | |
4589 | * V | | |
4590 | * TICK_SCHED_REMOTE_RUNNING | |
4591 | * | |
4592 | * | |
4593 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
4594 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
4595 | */ | |
d84b3131 FW |
4596 | |
4597 | static struct tick_work __percpu *tick_work_cpu; | |
4598 | ||
4599 | static void sched_tick_remote(struct work_struct *work) | |
4600 | { | |
4601 | struct delayed_work *dwork = to_delayed_work(work); | |
4602 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
4603 | int cpu = twork->cpu; | |
4604 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 4605 | struct task_struct *curr; |
d84b3131 | 4606 | struct rq_flags rf; |
d9c0ffca | 4607 | u64 delta; |
b55bd585 | 4608 | int os; |
d84b3131 FW |
4609 | |
4610 | /* | |
4611 | * Handle the tick only if it appears the remote CPU is running in full | |
4612 | * dynticks mode. The check is racy by nature, but missing a tick or | |
4613 | * having one too much is no big deal because the scheduler tick updates | |
4614 | * statistics and checks timeslices in a time-independent way, regardless | |
4615 | * of when exactly it is running. | |
4616 | */ | |
488603b8 | 4617 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 4618 | goto out_requeue; |
d84b3131 | 4619 | |
d9c0ffca FW |
4620 | rq_lock_irq(rq, &rf); |
4621 | curr = rq->curr; | |
488603b8 | 4622 | if (cpu_is_offline(cpu)) |
d9c0ffca | 4623 | goto out_unlock; |
d84b3131 | 4624 | |
d9c0ffca | 4625 | update_rq_clock(rq); |
d9c0ffca | 4626 | |
488603b8 SW |
4627 | if (!is_idle_task(curr)) { |
4628 | /* | |
4629 | * Make sure the next tick runs within a reasonable | |
4630 | * amount of time. | |
4631 | */ | |
4632 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
4633 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
4634 | } | |
d9c0ffca FW |
4635 | curr->sched_class->task_tick(rq, curr, 0); |
4636 | ||
ebc0f83c | 4637 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
4638 | out_unlock: |
4639 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 4640 | out_requeue: |
ebc0f83c | 4641 | |
d84b3131 FW |
4642 | /* |
4643 | * Run the remote tick once per second (1Hz). This arbitrary | |
4644 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
4645 | * to keep scheduler internal stats reasonably up to date. But |
4646 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 4647 | */ |
b55bd585 PM |
4648 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
4649 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
4650 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
4651 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
4652 | } |
4653 | ||
4654 | static void sched_tick_start(int cpu) | |
4655 | { | |
b55bd585 | 4656 | int os; |
d84b3131 FW |
4657 | struct tick_work *twork; |
4658 | ||
4659 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4660 | return; | |
4661 | ||
4662 | WARN_ON_ONCE(!tick_work_cpu); | |
4663 | ||
4664 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4665 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
4666 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
4667 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
4668 | twork->cpu = cpu; | |
4669 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
4670 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
4671 | } | |
d84b3131 FW |
4672 | } |
4673 | ||
4674 | #ifdef CONFIG_HOTPLUG_CPU | |
4675 | static void sched_tick_stop(int cpu) | |
4676 | { | |
4677 | struct tick_work *twork; | |
b55bd585 | 4678 | int os; |
d84b3131 FW |
4679 | |
4680 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4681 | return; | |
4682 | ||
4683 | WARN_ON_ONCE(!tick_work_cpu); | |
4684 | ||
4685 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4686 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
4687 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
4688 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
4689 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
4690 | } |
4691 | #endif /* CONFIG_HOTPLUG_CPU */ | |
4692 | ||
4693 | int __init sched_tick_offload_init(void) | |
4694 | { | |
4695 | tick_work_cpu = alloc_percpu(struct tick_work); | |
4696 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
4697 | return 0; |
4698 | } | |
4699 | ||
4700 | #else /* !CONFIG_NO_HZ_FULL */ | |
4701 | static inline void sched_tick_start(int cpu) { } | |
4702 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 4703 | #endif |
1da177e4 | 4704 | |
c1a280b6 | 4705 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 4706 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
4707 | /* |
4708 | * If the value passed in is equal to the current preempt count | |
4709 | * then we just disabled preemption. Start timing the latency. | |
4710 | */ | |
4711 | static inline void preempt_latency_start(int val) | |
4712 | { | |
4713 | if (preempt_count() == val) { | |
4714 | unsigned long ip = get_lock_parent_ip(); | |
4715 | #ifdef CONFIG_DEBUG_PREEMPT | |
4716 | current->preempt_disable_ip = ip; | |
4717 | #endif | |
4718 | trace_preempt_off(CALLER_ADDR0, ip); | |
4719 | } | |
4720 | } | |
7e49fcce | 4721 | |
edafe3a5 | 4722 | void preempt_count_add(int val) |
1da177e4 | 4723 | { |
6cd8a4bb | 4724 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4725 | /* |
4726 | * Underflow? | |
4727 | */ | |
9a11b49a IM |
4728 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4729 | return; | |
6cd8a4bb | 4730 | #endif |
bdb43806 | 4731 | __preempt_count_add(val); |
6cd8a4bb | 4732 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4733 | /* |
4734 | * Spinlock count overflowing soon? | |
4735 | */ | |
33859f7f MOS |
4736 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4737 | PREEMPT_MASK - 10); | |
6cd8a4bb | 4738 | #endif |
47252cfb | 4739 | preempt_latency_start(val); |
1da177e4 | 4740 | } |
bdb43806 | 4741 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 4742 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 4743 | |
47252cfb SR |
4744 | /* |
4745 | * If the value passed in equals to the current preempt count | |
4746 | * then we just enabled preemption. Stop timing the latency. | |
4747 | */ | |
4748 | static inline void preempt_latency_stop(int val) | |
4749 | { | |
4750 | if (preempt_count() == val) | |
4751 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
4752 | } | |
4753 | ||
edafe3a5 | 4754 | void preempt_count_sub(int val) |
1da177e4 | 4755 | { |
6cd8a4bb | 4756 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4757 | /* |
4758 | * Underflow? | |
4759 | */ | |
01e3eb82 | 4760 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4761 | return; |
1da177e4 LT |
4762 | /* |
4763 | * Is the spinlock portion underflowing? | |
4764 | */ | |
9a11b49a IM |
4765 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4766 | !(preempt_count() & PREEMPT_MASK))) | |
4767 | return; | |
6cd8a4bb | 4768 | #endif |
9a11b49a | 4769 | |
47252cfb | 4770 | preempt_latency_stop(val); |
bdb43806 | 4771 | __preempt_count_sub(val); |
1da177e4 | 4772 | } |
bdb43806 | 4773 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 4774 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 4775 | |
47252cfb SR |
4776 | #else |
4777 | static inline void preempt_latency_start(int val) { } | |
4778 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
4779 | #endif |
4780 | ||
59ddbcb2 IM |
4781 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
4782 | { | |
4783 | #ifdef CONFIG_DEBUG_PREEMPT | |
4784 | return p->preempt_disable_ip; | |
4785 | #else | |
4786 | return 0; | |
4787 | #endif | |
4788 | } | |
4789 | ||
1da177e4 | 4790 | /* |
dd41f596 | 4791 | * Print scheduling while atomic bug: |
1da177e4 | 4792 | */ |
dd41f596 | 4793 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4794 | { |
d1c6d149 VN |
4795 | /* Save this before calling printk(), since that will clobber it */ |
4796 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
4797 | ||
664dfa65 DJ |
4798 | if (oops_in_progress) |
4799 | return; | |
4800 | ||
3df0fc5b PZ |
4801 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4802 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4803 | |
dd41f596 | 4804 | debug_show_held_locks(prev); |
e21f5b15 | 4805 | print_modules(); |
dd41f596 IM |
4806 | if (irqs_disabled()) |
4807 | print_irqtrace_events(prev); | |
d1c6d149 VN |
4808 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
4809 | && in_atomic_preempt_off()) { | |
8f47b187 | 4810 | pr_err("Preemption disabled at:"); |
2062a4e8 | 4811 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 4812 | } |
748c7201 DBO |
4813 | if (panic_on_warn) |
4814 | panic("scheduling while atomic\n"); | |
4815 | ||
6135fc1e | 4816 | dump_stack(); |
373d4d09 | 4817 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 4818 | } |
1da177e4 | 4819 | |
dd41f596 IM |
4820 | /* |
4821 | * Various schedule()-time debugging checks and statistics: | |
4822 | */ | |
312364f3 | 4823 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 4824 | { |
0d9e2632 | 4825 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
4826 | if (task_stack_end_corrupted(prev)) |
4827 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
4828 | |
4829 | if (task_scs_end_corrupted(prev)) | |
4830 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 4831 | #endif |
b99def8b | 4832 | |
312364f3 DV |
4833 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
4834 | if (!preempt && prev->state && prev->non_block_count) { | |
4835 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", | |
4836 | prev->comm, prev->pid, prev->non_block_count); | |
4837 | dump_stack(); | |
4838 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
4839 | } | |
4840 | #endif | |
4841 | ||
1dc0fffc | 4842 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 4843 | __schedule_bug(prev); |
1dc0fffc PZ |
4844 | preempt_count_set(PREEMPT_DISABLED); |
4845 | } | |
b3fbab05 | 4846 | rcu_sleep_check(); |
9f68b5b7 | 4847 | SCHED_WARN_ON(ct_state() == CONTEXT_USER); |
dd41f596 | 4848 | |
1da177e4 LT |
4849 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4850 | ||
ae92882e | 4851 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
4852 | } |
4853 | ||
457d1f46 CY |
4854 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
4855 | struct rq_flags *rf) | |
4856 | { | |
4857 | #ifdef CONFIG_SMP | |
4858 | const struct sched_class *class; | |
4859 | /* | |
4860 | * We must do the balancing pass before put_prev_task(), such | |
4861 | * that when we release the rq->lock the task is in the same | |
4862 | * state as before we took rq->lock. | |
4863 | * | |
4864 | * We can terminate the balance pass as soon as we know there is | |
4865 | * a runnable task of @class priority or higher. | |
4866 | */ | |
4867 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
4868 | if (class->balance(rq, prev, rf)) | |
4869 | break; | |
4870 | } | |
4871 | #endif | |
4872 | ||
4873 | put_prev_task(rq, prev); | |
4874 | } | |
4875 | ||
dd41f596 IM |
4876 | /* |
4877 | * Pick up the highest-prio task: | |
4878 | */ | |
4879 | static inline struct task_struct * | |
d8ac8971 | 4880 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 4881 | { |
49ee5768 | 4882 | const struct sched_class *class; |
dd41f596 | 4883 | struct task_struct *p; |
1da177e4 LT |
4884 | |
4885 | /* | |
0ba87bb2 PZ |
4886 | * Optimization: we know that if all tasks are in the fair class we can |
4887 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 4888 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 4889 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 4890 | */ |
aa93cd53 | 4891 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
4892 | rq->nr_running == rq->cfs.h_nr_running)) { |
4893 | ||
5d7d6056 | 4894 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 4895 | if (unlikely(p == RETRY_TASK)) |
67692435 | 4896 | goto restart; |
6ccdc84b | 4897 | |
d1ccc66d | 4898 | /* Assumes fair_sched_class->next == idle_sched_class */ |
5d7d6056 | 4899 | if (!p) { |
f488e105 | 4900 | put_prev_task(rq, prev); |
98c2f700 | 4901 | p = pick_next_task_idle(rq); |
f488e105 | 4902 | } |
6ccdc84b PZ |
4903 | |
4904 | return p; | |
1da177e4 LT |
4905 | } |
4906 | ||
67692435 | 4907 | restart: |
457d1f46 | 4908 | put_prev_task_balance(rq, prev, rf); |
67692435 | 4909 | |
34f971f6 | 4910 | for_each_class(class) { |
98c2f700 | 4911 | p = class->pick_next_task(rq); |
67692435 | 4912 | if (p) |
dd41f596 | 4913 | return p; |
dd41f596 | 4914 | } |
34f971f6 | 4915 | |
d1ccc66d IM |
4916 | /* The idle class should always have a runnable task: */ |
4917 | BUG(); | |
dd41f596 | 4918 | } |
1da177e4 | 4919 | |
dd41f596 | 4920 | /* |
c259e01a | 4921 | * __schedule() is the main scheduler function. |
edde96ea PE |
4922 | * |
4923 | * The main means of driving the scheduler and thus entering this function are: | |
4924 | * | |
4925 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
4926 | * | |
4927 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
4928 | * paths. For example, see arch/x86/entry_64.S. | |
4929 | * | |
4930 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
4931 | * interrupt handler scheduler_tick(). | |
4932 | * | |
4933 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
4934 | * task to the run-queue and that's it. | |
4935 | * | |
4936 | * Now, if the new task added to the run-queue preempts the current | |
4937 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
4938 | * called on the nearest possible occasion: | |
4939 | * | |
c1a280b6 | 4940 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
4941 | * |
4942 | * - in syscall or exception context, at the next outmost | |
4943 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
4944 | * spin_unlock()!) | |
4945 | * | |
4946 | * - in IRQ context, return from interrupt-handler to | |
4947 | * preemptible context | |
4948 | * | |
c1a280b6 | 4949 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
4950 | * then at the next: |
4951 | * | |
4952 | * - cond_resched() call | |
4953 | * - explicit schedule() call | |
4954 | * - return from syscall or exception to user-space | |
4955 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 4956 | * |
b30f0e3f | 4957 | * WARNING: must be called with preemption disabled! |
dd41f596 | 4958 | */ |
499d7955 | 4959 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
4960 | { |
4961 | struct task_struct *prev, *next; | |
67ca7bde | 4962 | unsigned long *switch_count; |
dbfb089d | 4963 | unsigned long prev_state; |
d8ac8971 | 4964 | struct rq_flags rf; |
dd41f596 | 4965 | struct rq *rq; |
31656519 | 4966 | int cpu; |
dd41f596 | 4967 | |
dd41f596 IM |
4968 | cpu = smp_processor_id(); |
4969 | rq = cpu_rq(cpu); | |
dd41f596 | 4970 | prev = rq->curr; |
dd41f596 | 4971 | |
312364f3 | 4972 | schedule_debug(prev, preempt); |
1da177e4 | 4973 | |
31656519 | 4974 | if (sched_feat(HRTICK)) |
f333fdc9 | 4975 | hrtick_clear(rq); |
8f4d37ec | 4976 | |
46a5d164 | 4977 | local_irq_disable(); |
bcbfdd01 | 4978 | rcu_note_context_switch(preempt); |
46a5d164 | 4979 | |
e0acd0a6 ON |
4980 | /* |
4981 | * Make sure that signal_pending_state()->signal_pending() below | |
4982 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
4983 | * done by the caller to avoid the race with signal_wake_up(): |
4984 | * | |
4985 | * __set_current_state(@state) signal_wake_up() | |
4986 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
4987 | * wake_up_state(p, state) | |
4988 | * LOCK rq->lock LOCK p->pi_state | |
4989 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
4990 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 4991 | * |
dbfb089d | 4992 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 4993 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 4994 | */ |
8a8c69c3 | 4995 | rq_lock(rq, &rf); |
d89e588c | 4996 | smp_mb__after_spinlock(); |
1da177e4 | 4997 | |
d1ccc66d IM |
4998 | /* Promote REQ to ACT */ |
4999 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 5000 | update_rq_clock(rq); |
9edfbfed | 5001 | |
246d86b5 | 5002 | switch_count = &prev->nivcsw; |
d136122f | 5003 | |
dbfb089d | 5004 | /* |
d136122f PZ |
5005 | * We must load prev->state once (task_struct::state is volatile), such |
5006 | * that: | |
5007 | * | |
5008 | * - we form a control dependency vs deactivate_task() below. | |
5009 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 5010 | */ |
d136122f PZ |
5011 | prev_state = prev->state; |
5012 | if (!preempt && prev_state) { | |
dbfb089d | 5013 | if (signal_pending_state(prev_state, prev)) { |
1da177e4 | 5014 | prev->state = TASK_RUNNING; |
21aa9af0 | 5015 | } else { |
dbfb089d PZ |
5016 | prev->sched_contributes_to_load = |
5017 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
5018 | !(prev_state & TASK_NOLOAD) && | |
5019 | !(prev->flags & PF_FROZEN); | |
5020 | ||
5021 | if (prev->sched_contributes_to_load) | |
5022 | rq->nr_uninterruptible++; | |
5023 | ||
5024 | /* | |
5025 | * __schedule() ttwu() | |
d136122f PZ |
5026 | * prev_state = prev->state; if (p->on_rq && ...) |
5027 | * if (prev_state) goto out; | |
5028 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
5029 | * p->state = TASK_WAKING | |
5030 | * | |
5031 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
5032 | * |
5033 | * After this, schedule() must not care about p->state any more. | |
5034 | */ | |
bce4dc80 | 5035 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 5036 | |
e33a9bba TH |
5037 | if (prev->in_iowait) { |
5038 | atomic_inc(&rq->nr_iowait); | |
5039 | delayacct_blkio_start(); | |
5040 | } | |
21aa9af0 | 5041 | } |
dd41f596 | 5042 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5043 | } |
5044 | ||
d8ac8971 | 5045 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 5046 | clear_tsk_need_resched(prev); |
f27dde8d | 5047 | clear_preempt_need_resched(); |
1da177e4 | 5048 | |
1da177e4 | 5049 | if (likely(prev != next)) { |
1da177e4 | 5050 | rq->nr_switches++; |
5311a98f EB |
5051 | /* |
5052 | * RCU users of rcu_dereference(rq->curr) may not see | |
5053 | * changes to task_struct made by pick_next_task(). | |
5054 | */ | |
5055 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
5056 | /* |
5057 | * The membarrier system call requires each architecture | |
5058 | * to have a full memory barrier after updating | |
306e0604 MD |
5059 | * rq->curr, before returning to user-space. |
5060 | * | |
5061 | * Here are the schemes providing that barrier on the | |
5062 | * various architectures: | |
5063 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
5064 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
5065 | * - finish_lock_switch() for weakly-ordered | |
5066 | * architectures where spin_unlock is a full barrier, | |
5067 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
5068 | * is a RELEASE barrier), | |
22e4ebb9 | 5069 | */ |
1da177e4 LT |
5070 | ++*switch_count; |
5071 | ||
af449901 | 5072 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
5073 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
5074 | ||
c73464b1 | 5075 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
5076 | |
5077 | /* Also unlocks the rq: */ | |
5078 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 5079 | } else { |
cb42c9a3 | 5080 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 5081 | |
565790d2 PZ |
5082 | rq_unpin_lock(rq, &rf); |
5083 | __balance_callbacks(rq); | |
5084 | raw_spin_unlock_irq(&rq->lock); | |
5085 | } | |
1da177e4 | 5086 | } |
c259e01a | 5087 | |
9af6528e PZ |
5088 | void __noreturn do_task_dead(void) |
5089 | { | |
d1ccc66d | 5090 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 5091 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
5092 | |
5093 | /* Tell freezer to ignore us: */ | |
5094 | current->flags |= PF_NOFREEZE; | |
5095 | ||
9af6528e PZ |
5096 | __schedule(false); |
5097 | BUG(); | |
d1ccc66d IM |
5098 | |
5099 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 5100 | for (;;) |
d1ccc66d | 5101 | cpu_relax(); |
9af6528e PZ |
5102 | } |
5103 | ||
9c40cef2 TG |
5104 | static inline void sched_submit_work(struct task_struct *tsk) |
5105 | { | |
c1cecf88 SAS |
5106 | unsigned int task_flags; |
5107 | ||
b0fdc013 | 5108 | if (!tsk->state) |
9c40cef2 | 5109 | return; |
6d25be57 | 5110 | |
c1cecf88 | 5111 | task_flags = tsk->flags; |
6d25be57 TG |
5112 | /* |
5113 | * If a worker went to sleep, notify and ask workqueue whether | |
5114 | * it wants to wake up a task to maintain concurrency. | |
5115 | * As this function is called inside the schedule() context, | |
5116 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
5117 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
5118 | * requires it. | |
6d25be57 | 5119 | */ |
c1cecf88 | 5120 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 5121 | preempt_disable(); |
c1cecf88 | 5122 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
5123 | wq_worker_sleeping(tsk); |
5124 | else | |
5125 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
5126 | preempt_enable_no_resched(); |
5127 | } | |
5128 | ||
b0fdc013 SAS |
5129 | if (tsk_is_pi_blocked(tsk)) |
5130 | return; | |
5131 | ||
9c40cef2 TG |
5132 | /* |
5133 | * If we are going to sleep and we have plugged IO queued, | |
5134 | * make sure to submit it to avoid deadlocks. | |
5135 | */ | |
5136 | if (blk_needs_flush_plug(tsk)) | |
5137 | blk_schedule_flush_plug(tsk); | |
5138 | } | |
5139 | ||
6d25be57 TG |
5140 | static void sched_update_worker(struct task_struct *tsk) |
5141 | { | |
771b53d0 JA |
5142 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
5143 | if (tsk->flags & PF_WQ_WORKER) | |
5144 | wq_worker_running(tsk); | |
5145 | else | |
5146 | io_wq_worker_running(tsk); | |
5147 | } | |
6d25be57 TG |
5148 | } |
5149 | ||
722a9f92 | 5150 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 5151 | { |
9c40cef2 TG |
5152 | struct task_struct *tsk = current; |
5153 | ||
5154 | sched_submit_work(tsk); | |
bfd9b2b5 | 5155 | do { |
b30f0e3f | 5156 | preempt_disable(); |
fc13aeba | 5157 | __schedule(false); |
b30f0e3f | 5158 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 5159 | } while (need_resched()); |
6d25be57 | 5160 | sched_update_worker(tsk); |
c259e01a | 5161 | } |
1da177e4 LT |
5162 | EXPORT_SYMBOL(schedule); |
5163 | ||
8663effb SRV |
5164 | /* |
5165 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
5166 | * state (have scheduled out non-voluntarily) by making sure that all | |
5167 | * tasks have either left the run queue or have gone into user space. | |
5168 | * As idle tasks do not do either, they must not ever be preempted | |
5169 | * (schedule out non-voluntarily). | |
5170 | * | |
5171 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
5172 | * never enables preemption because it does not call sched_submit_work(). | |
5173 | */ | |
5174 | void __sched schedule_idle(void) | |
5175 | { | |
5176 | /* | |
5177 | * As this skips calling sched_submit_work(), which the idle task does | |
5178 | * regardless because that function is a nop when the task is in a | |
5179 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
5180 | * current task can be in any other state. Note, idle is always in the | |
5181 | * TASK_RUNNING state. | |
5182 | */ | |
5183 | WARN_ON_ONCE(current->state); | |
5184 | do { | |
5185 | __schedule(false); | |
5186 | } while (need_resched()); | |
5187 | } | |
5188 | ||
6775de49 | 5189 | #if defined(CONFIG_CONTEXT_TRACKING) && !defined(CONFIG_HAVE_CONTEXT_TRACKING_OFFSTACK) |
722a9f92 | 5190 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
5191 | { |
5192 | /* | |
5193 | * If we come here after a random call to set_need_resched(), | |
5194 | * or we have been woken up remotely but the IPI has not yet arrived, | |
5195 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
5196 | * we find a better solution. | |
7cc78f8f AL |
5197 | * |
5198 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 5199 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 5200 | * too frequently to make sense yet. |
20ab65e3 | 5201 | */ |
7cc78f8f | 5202 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 5203 | schedule(); |
7cc78f8f | 5204 | exception_exit(prev_state); |
20ab65e3 FW |
5205 | } |
5206 | #endif | |
5207 | ||
c5491ea7 TG |
5208 | /** |
5209 | * schedule_preempt_disabled - called with preemption disabled | |
5210 | * | |
5211 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
5212 | */ | |
5213 | void __sched schedule_preempt_disabled(void) | |
5214 | { | |
ba74c144 | 5215 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
5216 | schedule(); |
5217 | preempt_disable(); | |
5218 | } | |
5219 | ||
06b1f808 | 5220 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
5221 | { |
5222 | do { | |
47252cfb SR |
5223 | /* |
5224 | * Because the function tracer can trace preempt_count_sub() | |
5225 | * and it also uses preempt_enable/disable_notrace(), if | |
5226 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5227 | * by the function tracer will call this function again and | |
5228 | * cause infinite recursion. | |
5229 | * | |
5230 | * Preemption must be disabled here before the function | |
5231 | * tracer can trace. Break up preempt_disable() into two | |
5232 | * calls. One to disable preemption without fear of being | |
5233 | * traced. The other to still record the preemption latency, | |
5234 | * which can also be traced by the function tracer. | |
5235 | */ | |
499d7955 | 5236 | preempt_disable_notrace(); |
47252cfb | 5237 | preempt_latency_start(1); |
fc13aeba | 5238 | __schedule(true); |
47252cfb | 5239 | preempt_latency_stop(1); |
499d7955 | 5240 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
5241 | |
5242 | /* | |
5243 | * Check again in case we missed a preemption opportunity | |
5244 | * between schedule and now. | |
5245 | */ | |
a18b5d01 FW |
5246 | } while (need_resched()); |
5247 | } | |
5248 | ||
c1a280b6 | 5249 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 5250 | /* |
a49b4f40 VS |
5251 | * This is the entry point to schedule() from in-kernel preemption |
5252 | * off of preempt_enable. | |
1da177e4 | 5253 | */ |
722a9f92 | 5254 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 5255 | { |
1da177e4 LT |
5256 | /* |
5257 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5258 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5259 | */ |
fbb00b56 | 5260 | if (likely(!preemptible())) |
1da177e4 LT |
5261 | return; |
5262 | ||
a18b5d01 | 5263 | preempt_schedule_common(); |
1da177e4 | 5264 | } |
376e2424 | 5265 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 5266 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 5267 | |
009f60e2 | 5268 | /** |
4eaca0a8 | 5269 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
5270 | * |
5271 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
5272 | * recursion and tracing preempt enabling caused by the tracing | |
5273 | * infrastructure itself. But as tracing can happen in areas coming | |
5274 | * from userspace or just about to enter userspace, a preempt enable | |
5275 | * can occur before user_exit() is called. This will cause the scheduler | |
5276 | * to be called when the system is still in usermode. | |
5277 | * | |
5278 | * To prevent this, the preempt_enable_notrace will use this function | |
5279 | * instead of preempt_schedule() to exit user context if needed before | |
5280 | * calling the scheduler. | |
5281 | */ | |
4eaca0a8 | 5282 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
5283 | { |
5284 | enum ctx_state prev_ctx; | |
5285 | ||
5286 | if (likely(!preemptible())) | |
5287 | return; | |
5288 | ||
5289 | do { | |
47252cfb SR |
5290 | /* |
5291 | * Because the function tracer can trace preempt_count_sub() | |
5292 | * and it also uses preempt_enable/disable_notrace(), if | |
5293 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
5294 | * by the function tracer will call this function again and | |
5295 | * cause infinite recursion. | |
5296 | * | |
5297 | * Preemption must be disabled here before the function | |
5298 | * tracer can trace. Break up preempt_disable() into two | |
5299 | * calls. One to disable preemption without fear of being | |
5300 | * traced. The other to still record the preemption latency, | |
5301 | * which can also be traced by the function tracer. | |
5302 | */ | |
3d8f74dd | 5303 | preempt_disable_notrace(); |
47252cfb | 5304 | preempt_latency_start(1); |
009f60e2 ON |
5305 | /* |
5306 | * Needs preempt disabled in case user_exit() is traced | |
5307 | * and the tracer calls preempt_enable_notrace() causing | |
5308 | * an infinite recursion. | |
5309 | */ | |
5310 | prev_ctx = exception_enter(); | |
fc13aeba | 5311 | __schedule(true); |
009f60e2 ON |
5312 | exception_exit(prev_ctx); |
5313 | ||
47252cfb | 5314 | preempt_latency_stop(1); |
3d8f74dd | 5315 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
5316 | } while (need_resched()); |
5317 | } | |
4eaca0a8 | 5318 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 5319 | |
c1a280b6 | 5320 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 LT |
5321 | |
5322 | /* | |
a49b4f40 | 5323 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5324 | * off of irq context. |
5325 | * Note, that this is called and return with irqs disabled. This will | |
5326 | * protect us against recursive calling from irq. | |
5327 | */ | |
722a9f92 | 5328 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 5329 | { |
b22366cd | 5330 | enum ctx_state prev_state; |
6478d880 | 5331 | |
2ed6e34f | 5332 | /* Catch callers which need to be fixed */ |
f27dde8d | 5333 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 5334 | |
b22366cd FW |
5335 | prev_state = exception_enter(); |
5336 | ||
3a5c359a | 5337 | do { |
3d8f74dd | 5338 | preempt_disable(); |
3a5c359a | 5339 | local_irq_enable(); |
fc13aeba | 5340 | __schedule(true); |
3a5c359a | 5341 | local_irq_disable(); |
3d8f74dd | 5342 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 5343 | } while (need_resched()); |
b22366cd FW |
5344 | |
5345 | exception_exit(prev_state); | |
1da177e4 LT |
5346 | } |
5347 | ||
ac6424b9 | 5348 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 5349 | void *key) |
1da177e4 | 5350 | { |
062d3f95 | 5351 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 5352 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 5353 | } |
1da177e4 LT |
5354 | EXPORT_SYMBOL(default_wake_function); |
5355 | ||
b29739f9 IM |
5356 | #ifdef CONFIG_RT_MUTEXES |
5357 | ||
acd58620 PZ |
5358 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
5359 | { | |
5360 | if (pi_task) | |
5361 | prio = min(prio, pi_task->prio); | |
5362 | ||
5363 | return prio; | |
5364 | } | |
5365 | ||
5366 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5367 | { | |
5368 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
5369 | ||
5370 | return __rt_effective_prio(pi_task, prio); | |
5371 | } | |
5372 | ||
b29739f9 IM |
5373 | /* |
5374 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
5375 | * @p: task to boost |
5376 | * @pi_task: donor task | |
b29739f9 IM |
5377 | * |
5378 | * This function changes the 'effective' priority of a task. It does | |
5379 | * not touch ->normal_prio like __setscheduler(). | |
5380 | * | |
c365c292 TG |
5381 | * Used by the rt_mutex code to implement priority inheritance |
5382 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 5383 | */ |
acd58620 | 5384 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 5385 | { |
acd58620 | 5386 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 5387 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 5388 | const struct sched_class *prev_class; |
eb580751 PZ |
5389 | struct rq_flags rf; |
5390 | struct rq *rq; | |
b29739f9 | 5391 | |
acd58620 PZ |
5392 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
5393 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
5394 | ||
5395 | /* | |
5396 | * If nothing changed; bail early. | |
5397 | */ | |
5398 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
5399 | return; | |
b29739f9 | 5400 | |
eb580751 | 5401 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 5402 | update_rq_clock(rq); |
acd58620 PZ |
5403 | /* |
5404 | * Set under pi_lock && rq->lock, such that the value can be used under | |
5405 | * either lock. | |
5406 | * | |
5407 | * Note that there is loads of tricky to make this pointer cache work | |
5408 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
5409 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
5410 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 5411 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
5412 | */ |
5413 | p->pi_top_task = pi_task; | |
5414 | ||
5415 | /* | |
5416 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
5417 | */ | |
5418 | if (prio == p->prio && !dl_prio(prio)) | |
5419 | goto out_unlock; | |
b29739f9 | 5420 | |
1c4dd99b TG |
5421 | /* |
5422 | * Idle task boosting is a nono in general. There is one | |
5423 | * exception, when PREEMPT_RT and NOHZ is active: | |
5424 | * | |
5425 | * The idle task calls get_next_timer_interrupt() and holds | |
5426 | * the timer wheel base->lock on the CPU and another CPU wants | |
5427 | * to access the timer (probably to cancel it). We can safely | |
5428 | * ignore the boosting request, as the idle CPU runs this code | |
5429 | * with interrupts disabled and will complete the lock | |
5430 | * protected section without being interrupted. So there is no | |
5431 | * real need to boost. | |
5432 | */ | |
5433 | if (unlikely(p == rq->idle)) { | |
5434 | WARN_ON(p != rq->curr); | |
5435 | WARN_ON(p->pi_blocked_on); | |
5436 | goto out_unlock; | |
5437 | } | |
5438 | ||
b91473ff | 5439 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 5440 | oldprio = p->prio; |
ff77e468 PZ |
5441 | |
5442 | if (oldprio == prio) | |
5443 | queue_flag &= ~DEQUEUE_MOVE; | |
5444 | ||
83ab0aa0 | 5445 | prev_class = p->sched_class; |
da0c1e65 | 5446 | queued = task_on_rq_queued(p); |
051a1d1a | 5447 | running = task_current(rq, p); |
da0c1e65 | 5448 | if (queued) |
ff77e468 | 5449 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 5450 | if (running) |
f3cd1c4e | 5451 | put_prev_task(rq, p); |
dd41f596 | 5452 | |
2d3d891d DF |
5453 | /* |
5454 | * Boosting condition are: | |
5455 | * 1. -rt task is running and holds mutex A | |
5456 | * --> -dl task blocks on mutex A | |
5457 | * | |
5458 | * 2. -dl task is running and holds mutex A | |
5459 | * --> -dl task blocks on mutex A and could preempt the | |
5460 | * running task | |
5461 | */ | |
5462 | if (dl_prio(prio)) { | |
466af29b | 5463 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
5464 | (pi_task && dl_prio(pi_task->prio) && |
5465 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2279f540 | 5466 | p->dl.pi_se = pi_task->dl.pi_se; |
ff77e468 | 5467 | queue_flag |= ENQUEUE_REPLENISH; |
2279f540 JL |
5468 | } else { |
5469 | p->dl.pi_se = &p->dl; | |
5470 | } | |
aab03e05 | 5471 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
5472 | } else if (rt_prio(prio)) { |
5473 | if (dl_prio(oldprio)) | |
2279f540 | 5474 | p->dl.pi_se = &p->dl; |
2d3d891d | 5475 | if (oldprio < prio) |
ff77e468 | 5476 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 5477 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
5478 | } else { |
5479 | if (dl_prio(oldprio)) | |
2279f540 | 5480 | p->dl.pi_se = &p->dl; |
746db944 BS |
5481 | if (rt_prio(oldprio)) |
5482 | p->rt.timeout = 0; | |
dd41f596 | 5483 | p->sched_class = &fair_sched_class; |
2d3d891d | 5484 | } |
dd41f596 | 5485 | |
b29739f9 IM |
5486 | p->prio = prio; |
5487 | ||
da0c1e65 | 5488 | if (queued) |
ff77e468 | 5489 | enqueue_task(rq, p, queue_flag); |
a399d233 | 5490 | if (running) |
03b7fad1 | 5491 | set_next_task(rq, p); |
cb469845 | 5492 | |
da7a735e | 5493 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 5494 | out_unlock: |
d1ccc66d IM |
5495 | /* Avoid rq from going away on us: */ |
5496 | preempt_disable(); | |
4c9a4bc8 | 5497 | |
565790d2 PZ |
5498 | rq_unpin_lock(rq, &rf); |
5499 | __balance_callbacks(rq); | |
5500 | raw_spin_unlock(&rq->lock); | |
5501 | ||
4c9a4bc8 | 5502 | preempt_enable(); |
b29739f9 | 5503 | } |
acd58620 PZ |
5504 | #else |
5505 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
5506 | { | |
5507 | return prio; | |
5508 | } | |
b29739f9 | 5509 | #endif |
d50dde5a | 5510 | |
36c8b586 | 5511 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5512 | { |
49bd21ef | 5513 | bool queued, running; |
53a23364 | 5514 | int old_prio; |
eb580751 | 5515 | struct rq_flags rf; |
70b97a7f | 5516 | struct rq *rq; |
1da177e4 | 5517 | |
75e45d51 | 5518 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
5519 | return; |
5520 | /* | |
5521 | * We have to be careful, if called from sys_setpriority(), | |
5522 | * the task might be in the middle of scheduling on another CPU. | |
5523 | */ | |
eb580751 | 5524 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
5525 | update_rq_clock(rq); |
5526 | ||
1da177e4 LT |
5527 | /* |
5528 | * The RT priorities are set via sched_setscheduler(), but we still | |
5529 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 5530 | * it won't have any effect on scheduling until the task is |
aab03e05 | 5531 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 5532 | */ |
aab03e05 | 5533 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
5534 | p->static_prio = NICE_TO_PRIO(nice); |
5535 | goto out_unlock; | |
5536 | } | |
da0c1e65 | 5537 | queued = task_on_rq_queued(p); |
49bd21ef | 5538 | running = task_current(rq, p); |
da0c1e65 | 5539 | if (queued) |
7a57f32a | 5540 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
5541 | if (running) |
5542 | put_prev_task(rq, p); | |
1da177e4 | 5543 | |
1da177e4 | 5544 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 5545 | set_load_weight(p, true); |
b29739f9 IM |
5546 | old_prio = p->prio; |
5547 | p->prio = effective_prio(p); | |
1da177e4 | 5548 | |
5443a0be | 5549 | if (queued) |
7134b3e9 | 5550 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 5551 | if (running) |
03b7fad1 | 5552 | set_next_task(rq, p); |
5443a0be FW |
5553 | |
5554 | /* | |
5555 | * If the task increased its priority or is running and | |
5556 | * lowered its priority, then reschedule its CPU: | |
5557 | */ | |
5558 | p->sched_class->prio_changed(rq, p, old_prio); | |
5559 | ||
1da177e4 | 5560 | out_unlock: |
eb580751 | 5561 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 5562 | } |
1da177e4 LT |
5563 | EXPORT_SYMBOL(set_user_nice); |
5564 | ||
e43379f1 MM |
5565 | /* |
5566 | * can_nice - check if a task can reduce its nice value | |
5567 | * @p: task | |
5568 | * @nice: nice value | |
5569 | */ | |
36c8b586 | 5570 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5571 | { |
d1ccc66d | 5572 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 5573 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 5574 | |
78d7d407 | 5575 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5576 | capable(CAP_SYS_NICE)); |
5577 | } | |
5578 | ||
1da177e4 LT |
5579 | #ifdef __ARCH_WANT_SYS_NICE |
5580 | ||
5581 | /* | |
5582 | * sys_nice - change the priority of the current process. | |
5583 | * @increment: priority increment | |
5584 | * | |
5585 | * sys_setpriority is a more generic, but much slower function that | |
5586 | * does similar things. | |
5587 | */ | |
5add95d4 | 5588 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5589 | { |
48f24c4d | 5590 | long nice, retval; |
1da177e4 LT |
5591 | |
5592 | /* | |
5593 | * Setpriority might change our priority at the same moment. | |
5594 | * We don't have to worry. Conceptually one call occurs first | |
5595 | * and we have a single winner. | |
5596 | */ | |
a9467fa3 | 5597 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 5598 | nice = task_nice(current) + increment; |
1da177e4 | 5599 | |
a9467fa3 | 5600 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
5601 | if (increment < 0 && !can_nice(current, nice)) |
5602 | return -EPERM; | |
5603 | ||
1da177e4 LT |
5604 | retval = security_task_setnice(current, nice); |
5605 | if (retval) | |
5606 | return retval; | |
5607 | ||
5608 | set_user_nice(current, nice); | |
5609 | return 0; | |
5610 | } | |
5611 | ||
5612 | #endif | |
5613 | ||
5614 | /** | |
5615 | * task_prio - return the priority value of a given task. | |
5616 | * @p: the task in question. | |
5617 | * | |
e69f6186 | 5618 | * Return: The priority value as seen by users in /proc. |
c541bb78 DE |
5619 | * |
5620 | * sched policy return value kernel prio user prio/nice | |
5621 | * | |
5622 | * normal, batch, idle [0 ... 39] [100 ... 139] 0/[-20 ... 19] | |
5623 | * fifo, rr [-2 ... -100] [98 ... 0] [1 ... 99] | |
5624 | * deadline -101 -1 0 | |
1da177e4 | 5625 | */ |
36c8b586 | 5626 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5627 | { |
5628 | return p->prio - MAX_RT_PRIO; | |
5629 | } | |
5630 | ||
1da177e4 | 5631 | /** |
d1ccc66d | 5632 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 5633 | * @cpu: the processor in question. |
e69f6186 YB |
5634 | * |
5635 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
5636 | */ |
5637 | int idle_cpu(int cpu) | |
5638 | { | |
908a3283 TG |
5639 | struct rq *rq = cpu_rq(cpu); |
5640 | ||
5641 | if (rq->curr != rq->idle) | |
5642 | return 0; | |
5643 | ||
5644 | if (rq->nr_running) | |
5645 | return 0; | |
5646 | ||
5647 | #ifdef CONFIG_SMP | |
126c2092 | 5648 | if (rq->ttwu_pending) |
908a3283 TG |
5649 | return 0; |
5650 | #endif | |
5651 | ||
5652 | return 1; | |
1da177e4 LT |
5653 | } |
5654 | ||
943d355d RJ |
5655 | /** |
5656 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
5657 | * @cpu: the CPU in question. | |
5658 | * | |
5659 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
5660 | */ | |
5661 | int available_idle_cpu(int cpu) | |
5662 | { | |
5663 | if (!idle_cpu(cpu)) | |
5664 | return 0; | |
5665 | ||
247f2f6f RJ |
5666 | if (vcpu_is_preempted(cpu)) |
5667 | return 0; | |
5668 | ||
908a3283 | 5669 | return 1; |
1da177e4 LT |
5670 | } |
5671 | ||
1da177e4 | 5672 | /** |
d1ccc66d | 5673 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 5674 | * @cpu: the processor in question. |
e69f6186 | 5675 | * |
d1ccc66d | 5676 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 5677 | */ |
36c8b586 | 5678 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5679 | { |
5680 | return cpu_rq(cpu)->idle; | |
5681 | } | |
5682 | ||
7d6a905f VK |
5683 | #ifdef CONFIG_SMP |
5684 | /* | |
5685 | * This function computes an effective utilization for the given CPU, to be | |
5686 | * used for frequency selection given the linear relation: f = u * f_max. | |
5687 | * | |
5688 | * The scheduler tracks the following metrics: | |
5689 | * | |
5690 | * cpu_util_{cfs,rt,dl,irq}() | |
5691 | * cpu_bw_dl() | |
5692 | * | |
5693 | * Where the cfs,rt and dl util numbers are tracked with the same metric and | |
5694 | * synchronized windows and are thus directly comparable. | |
5695 | * | |
5696 | * The cfs,rt,dl utilization are the running times measured with rq->clock_task | |
5697 | * which excludes things like IRQ and steal-time. These latter are then accrued | |
5698 | * in the irq utilization. | |
5699 | * | |
5700 | * The DL bandwidth number otoh is not a measured metric but a value computed | |
5701 | * based on the task model parameters and gives the minimal utilization | |
5702 | * required to meet deadlines. | |
5703 | */ | |
a5418be9 VK |
5704 | unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, |
5705 | unsigned long max, enum cpu_util_type type, | |
7d6a905f VK |
5706 | struct task_struct *p) |
5707 | { | |
5708 | unsigned long dl_util, util, irq; | |
5709 | struct rq *rq = cpu_rq(cpu); | |
5710 | ||
5711 | if (!uclamp_is_used() && | |
5712 | type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) { | |
5713 | return max; | |
5714 | } | |
5715 | ||
5716 | /* | |
5717 | * Early check to see if IRQ/steal time saturates the CPU, can be | |
5718 | * because of inaccuracies in how we track these -- see | |
5719 | * update_irq_load_avg(). | |
5720 | */ | |
5721 | irq = cpu_util_irq(rq); | |
5722 | if (unlikely(irq >= max)) | |
5723 | return max; | |
5724 | ||
5725 | /* | |
5726 | * Because the time spend on RT/DL tasks is visible as 'lost' time to | |
5727 | * CFS tasks and we use the same metric to track the effective | |
5728 | * utilization (PELT windows are synchronized) we can directly add them | |
5729 | * to obtain the CPU's actual utilization. | |
5730 | * | |
5731 | * CFS and RT utilization can be boosted or capped, depending on | |
5732 | * utilization clamp constraints requested by currently RUNNABLE | |
5733 | * tasks. | |
5734 | * When there are no CFS RUNNABLE tasks, clamps are released and | |
5735 | * frequency will be gracefully reduced with the utilization decay. | |
5736 | */ | |
5737 | util = util_cfs + cpu_util_rt(rq); | |
5738 | if (type == FREQUENCY_UTIL) | |
5739 | util = uclamp_rq_util_with(rq, util, p); | |
5740 | ||
5741 | dl_util = cpu_util_dl(rq); | |
5742 | ||
5743 | /* | |
5744 | * For frequency selection we do not make cpu_util_dl() a permanent part | |
5745 | * of this sum because we want to use cpu_bw_dl() later on, but we need | |
5746 | * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such | |
5747 | * that we select f_max when there is no idle time. | |
5748 | * | |
5749 | * NOTE: numerical errors or stop class might cause us to not quite hit | |
5750 | * saturation when we should -- something for later. | |
5751 | */ | |
5752 | if (util + dl_util >= max) | |
5753 | return max; | |
5754 | ||
5755 | /* | |
5756 | * OTOH, for energy computation we need the estimated running time, so | |
5757 | * include util_dl and ignore dl_bw. | |
5758 | */ | |
5759 | if (type == ENERGY_UTIL) | |
5760 | util += dl_util; | |
5761 | ||
5762 | /* | |
5763 | * There is still idle time; further improve the number by using the | |
5764 | * irq metric. Because IRQ/steal time is hidden from the task clock we | |
5765 | * need to scale the task numbers: | |
5766 | * | |
5767 | * max - irq | |
5768 | * U' = irq + --------- * U | |
5769 | * max | |
5770 | */ | |
5771 | util = scale_irq_capacity(util, irq, max); | |
5772 | util += irq; | |
5773 | ||
5774 | /* | |
5775 | * Bandwidth required by DEADLINE must always be granted while, for | |
5776 | * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism | |
5777 | * to gracefully reduce the frequency when no tasks show up for longer | |
5778 | * periods of time. | |
5779 | * | |
5780 | * Ideally we would like to set bw_dl as min/guaranteed freq and util + | |
5781 | * bw_dl as requested freq. However, cpufreq is not yet ready for such | |
5782 | * an interface. So, we only do the latter for now. | |
5783 | */ | |
5784 | if (type == FREQUENCY_UTIL) | |
5785 | util += cpu_bw_dl(rq); | |
5786 | ||
5787 | return min(max, util); | |
5788 | } | |
a5418be9 VK |
5789 | |
5790 | unsigned long sched_cpu_util(int cpu, unsigned long max) | |
5791 | { | |
5792 | return effective_cpu_util(cpu, cpu_util_cfs(cpu_rq(cpu)), max, | |
5793 | ENERGY_UTIL, NULL); | |
5794 | } | |
7d6a905f VK |
5795 | #endif /* CONFIG_SMP */ |
5796 | ||
1da177e4 LT |
5797 | /** |
5798 | * find_process_by_pid - find a process with a matching PID value. | |
5799 | * @pid: the pid in question. | |
e69f6186 YB |
5800 | * |
5801 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 5802 | */ |
a9957449 | 5803 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5804 | { |
228ebcbe | 5805 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5806 | } |
5807 | ||
c13db6b1 SR |
5808 | /* |
5809 | * sched_setparam() passes in -1 for its policy, to let the functions | |
5810 | * it calls know not to change it. | |
5811 | */ | |
5812 | #define SETPARAM_POLICY -1 | |
5813 | ||
c365c292 TG |
5814 | static void __setscheduler_params(struct task_struct *p, |
5815 | const struct sched_attr *attr) | |
1da177e4 | 5816 | { |
d50dde5a DF |
5817 | int policy = attr->sched_policy; |
5818 | ||
c13db6b1 | 5819 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
5820 | policy = p->policy; |
5821 | ||
1da177e4 | 5822 | p->policy = policy; |
d50dde5a | 5823 | |
aab03e05 DF |
5824 | if (dl_policy(policy)) |
5825 | __setparam_dl(p, attr); | |
39fd8fd2 | 5826 | else if (fair_policy(policy)) |
d50dde5a DF |
5827 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
5828 | ||
39fd8fd2 PZ |
5829 | /* |
5830 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
5831 | * !rt_policy. Always setting this ensures that things like | |
5832 | * getparam()/getattr() don't report silly values for !rt tasks. | |
5833 | */ | |
5834 | p->rt_priority = attr->sched_priority; | |
383afd09 | 5835 | p->normal_prio = normal_prio(p); |
9059393e | 5836 | set_load_weight(p, true); |
c365c292 | 5837 | } |
39fd8fd2 | 5838 | |
c365c292 TG |
5839 | /* Actually do priority change: must hold pi & rq lock. */ |
5840 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 5841 | const struct sched_attr *attr, bool keep_boost) |
c365c292 | 5842 | { |
a509a7cd PB |
5843 | /* |
5844 | * If params can't change scheduling class changes aren't allowed | |
5845 | * either. | |
5846 | */ | |
5847 | if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) | |
5848 | return; | |
5849 | ||
c365c292 | 5850 | __setscheduler_params(p, attr); |
d50dde5a | 5851 | |
383afd09 | 5852 | /* |
0782e63b TG |
5853 | * Keep a potential priority boosting if called from |
5854 | * sched_setscheduler(). | |
383afd09 | 5855 | */ |
acd58620 | 5856 | p->prio = normal_prio(p); |
0782e63b | 5857 | if (keep_boost) |
acd58620 | 5858 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 5859 | |
aab03e05 DF |
5860 | if (dl_prio(p->prio)) |
5861 | p->sched_class = &dl_sched_class; | |
5862 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
5863 | p->sched_class = &rt_sched_class; |
5864 | else | |
5865 | p->sched_class = &fair_sched_class; | |
1da177e4 | 5866 | } |
aab03e05 | 5867 | |
c69e8d9c | 5868 | /* |
d1ccc66d | 5869 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
5870 | */ |
5871 | static bool check_same_owner(struct task_struct *p) | |
5872 | { | |
5873 | const struct cred *cred = current_cred(), *pcred; | |
5874 | bool match; | |
5875 | ||
5876 | rcu_read_lock(); | |
5877 | pcred = __task_cred(p); | |
9c806aa0 EB |
5878 | match = (uid_eq(cred->euid, pcred->euid) || |
5879 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
5880 | rcu_read_unlock(); |
5881 | return match; | |
5882 | } | |
5883 | ||
d50dde5a DF |
5884 | static int __sched_setscheduler(struct task_struct *p, |
5885 | const struct sched_attr *attr, | |
dbc7f069 | 5886 | bool user, bool pi) |
1da177e4 | 5887 | { |
383afd09 SR |
5888 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
5889 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 5890 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 5891 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 5892 | const struct sched_class *prev_class; |
565790d2 | 5893 | struct callback_head *head; |
eb580751 | 5894 | struct rq_flags rf; |
ca94c442 | 5895 | int reset_on_fork; |
7a57f32a | 5896 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 5897 | struct rq *rq; |
1da177e4 | 5898 | |
896bbb25 SRV |
5899 | /* The pi code expects interrupts enabled */ |
5900 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 5901 | recheck: |
d1ccc66d | 5902 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
5903 | if (policy < 0) { |
5904 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5905 | policy = oldpolicy = p->policy; |
ca94c442 | 5906 | } else { |
7479f3c9 | 5907 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 5908 | |
20f9cd2a | 5909 | if (!valid_policy(policy)) |
ca94c442 LP |
5910 | return -EINVAL; |
5911 | } | |
5912 | ||
794a56eb | 5913 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
5914 | return -EINVAL; |
5915 | ||
1da177e4 LT |
5916 | /* |
5917 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
ae18ad28 | 5918 | * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL, |
dd41f596 | 5919 | * SCHED_BATCH and SCHED_IDLE is 0. |
1da177e4 | 5920 | */ |
ae18ad28 | 5921 | if (attr->sched_priority > MAX_RT_PRIO-1) |
1da177e4 | 5922 | return -EINVAL; |
aab03e05 DF |
5923 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
5924 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
5925 | return -EINVAL; |
5926 | ||
37e4ab3f OC |
5927 | /* |
5928 | * Allow unprivileged RT tasks to decrease priority: | |
5929 | */ | |
961ccddd | 5930 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 5931 | if (fair_policy(policy)) { |
d0ea0268 | 5932 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 5933 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
5934 | return -EPERM; |
5935 | } | |
5936 | ||
e05606d3 | 5937 | if (rt_policy(policy)) { |
a44702e8 ON |
5938 | unsigned long rlim_rtprio = |
5939 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 5940 | |
d1ccc66d | 5941 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
5942 | if (policy != p->policy && !rlim_rtprio) |
5943 | return -EPERM; | |
5944 | ||
d1ccc66d | 5945 | /* Can't increase priority: */ |
d50dde5a DF |
5946 | if (attr->sched_priority > p->rt_priority && |
5947 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
5948 | return -EPERM; |
5949 | } | |
c02aa73b | 5950 | |
d44753b8 JL |
5951 | /* |
5952 | * Can't set/change SCHED_DEADLINE policy at all for now | |
5953 | * (safest behavior); in the future we would like to allow | |
5954 | * unprivileged DL tasks to increase their relative deadline | |
5955 | * or reduce their runtime (both ways reducing utilization) | |
5956 | */ | |
5957 | if (dl_policy(policy)) | |
5958 | return -EPERM; | |
5959 | ||
dd41f596 | 5960 | /* |
c02aa73b DH |
5961 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5962 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5963 | */ |
1da1843f | 5964 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 5965 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
5966 | return -EPERM; |
5967 | } | |
5fe1d75f | 5968 | |
d1ccc66d | 5969 | /* Can't change other user's priorities: */ |
c69e8d9c | 5970 | if (!check_same_owner(p)) |
37e4ab3f | 5971 | return -EPERM; |
ca94c442 | 5972 | |
d1ccc66d | 5973 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
5974 | if (p->sched_reset_on_fork && !reset_on_fork) |
5975 | return -EPERM; | |
37e4ab3f | 5976 | } |
1da177e4 | 5977 | |
725aad24 | 5978 | if (user) { |
794a56eb JL |
5979 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
5980 | return -EINVAL; | |
5981 | ||
b0ae1981 | 5982 | retval = security_task_setscheduler(p); |
725aad24 JF |
5983 | if (retval) |
5984 | return retval; | |
5985 | } | |
5986 | ||
a509a7cd PB |
5987 | /* Update task specific "requested" clamps */ |
5988 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
5989 | retval = uclamp_validate(p, attr); | |
5990 | if (retval) | |
5991 | return retval; | |
5992 | } | |
5993 | ||
710da3c8 JL |
5994 | if (pi) |
5995 | cpuset_read_lock(); | |
5996 | ||
b29739f9 | 5997 | /* |
d1ccc66d | 5998 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 5999 | * changing the priority of the task: |
0122ec5b | 6000 | * |
25985edc | 6001 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
6002 | * runqueue lock must be held. |
6003 | */ | |
eb580751 | 6004 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 6005 | update_rq_clock(rq); |
dc61b1d6 | 6006 | |
34f971f6 | 6007 | /* |
d1ccc66d | 6008 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
6009 | */ |
6010 | if (p == rq->stop) { | |
4b211f2b MP |
6011 | retval = -EINVAL; |
6012 | goto unlock; | |
34f971f6 PZ |
6013 | } |
6014 | ||
a51e9198 | 6015 | /* |
d6b1e911 TG |
6016 | * If not changing anything there's no need to proceed further, |
6017 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 6018 | */ |
d50dde5a | 6019 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 6020 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
6021 | goto change; |
6022 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
6023 | goto change; | |
75381608 | 6024 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 6025 | goto change; |
a509a7cd PB |
6026 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
6027 | goto change; | |
d50dde5a | 6028 | |
d6b1e911 | 6029 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
6030 | retval = 0; |
6031 | goto unlock; | |
a51e9198 | 6032 | } |
d50dde5a | 6033 | change: |
a51e9198 | 6034 | |
dc61b1d6 | 6035 | if (user) { |
332ac17e | 6036 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
6037 | /* |
6038 | * Do not allow realtime tasks into groups that have no runtime | |
6039 | * assigned. | |
6040 | */ | |
6041 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
6042 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
6043 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
6044 | retval = -EPERM; |
6045 | goto unlock; | |
dc61b1d6 | 6046 | } |
dc61b1d6 | 6047 | #endif |
332ac17e | 6048 | #ifdef CONFIG_SMP |
794a56eb JL |
6049 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
6050 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 6051 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
6052 | |
6053 | /* | |
6054 | * Don't allow tasks with an affinity mask smaller than | |
6055 | * the entire root_domain to become SCHED_DEADLINE. We | |
6056 | * will also fail if there's no bandwidth available. | |
6057 | */ | |
3bd37062 | 6058 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 6059 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
6060 | retval = -EPERM; |
6061 | goto unlock; | |
332ac17e DF |
6062 | } |
6063 | } | |
6064 | #endif | |
6065 | } | |
dc61b1d6 | 6066 | |
d1ccc66d | 6067 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
6068 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
6069 | policy = oldpolicy = -1; | |
eb580751 | 6070 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
6071 | if (pi) |
6072 | cpuset_read_unlock(); | |
1da177e4 LT |
6073 | goto recheck; |
6074 | } | |
332ac17e DF |
6075 | |
6076 | /* | |
6077 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
6078 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
6079 | * is available. | |
6080 | */ | |
06a76fe0 | 6081 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
6082 | retval = -EBUSY; |
6083 | goto unlock; | |
332ac17e DF |
6084 | } |
6085 | ||
c365c292 TG |
6086 | p->sched_reset_on_fork = reset_on_fork; |
6087 | oldprio = p->prio; | |
6088 | ||
dbc7f069 PZ |
6089 | if (pi) { |
6090 | /* | |
6091 | * Take priority boosted tasks into account. If the new | |
6092 | * effective priority is unchanged, we just store the new | |
6093 | * normal parameters and do not touch the scheduler class and | |
6094 | * the runqueue. This will be done when the task deboost | |
6095 | * itself. | |
6096 | */ | |
acd58620 | 6097 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
6098 | if (new_effective_prio == oldprio) |
6099 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
6100 | } |
6101 | ||
da0c1e65 | 6102 | queued = task_on_rq_queued(p); |
051a1d1a | 6103 | running = task_current(rq, p); |
da0c1e65 | 6104 | if (queued) |
ff77e468 | 6105 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 6106 | if (running) |
f3cd1c4e | 6107 | put_prev_task(rq, p); |
f6b53205 | 6108 | |
83ab0aa0 | 6109 | prev_class = p->sched_class; |
a509a7cd | 6110 | |
dbc7f069 | 6111 | __setscheduler(rq, p, attr, pi); |
a509a7cd | 6112 | __setscheduler_uclamp(p, attr); |
f6b53205 | 6113 | |
da0c1e65 | 6114 | if (queued) { |
81a44c54 TG |
6115 | /* |
6116 | * We enqueue to tail when the priority of a task is | |
6117 | * increased (user space view). | |
6118 | */ | |
ff77e468 PZ |
6119 | if (oldprio < p->prio) |
6120 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 6121 | |
ff77e468 | 6122 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 6123 | } |
a399d233 | 6124 | if (running) |
03b7fad1 | 6125 | set_next_task(rq, p); |
cb469845 | 6126 | |
da7a735e | 6127 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
6128 | |
6129 | /* Avoid rq from going away on us: */ | |
6130 | preempt_disable(); | |
565790d2 | 6131 | head = splice_balance_callbacks(rq); |
eb580751 | 6132 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 6133 | |
710da3c8 JL |
6134 | if (pi) { |
6135 | cpuset_read_unlock(); | |
dbc7f069 | 6136 | rt_mutex_adjust_pi(p); |
710da3c8 | 6137 | } |
95e02ca9 | 6138 | |
d1ccc66d | 6139 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 6140 | balance_callbacks(rq, head); |
4c9a4bc8 | 6141 | preempt_enable(); |
95e02ca9 | 6142 | |
1da177e4 | 6143 | return 0; |
4b211f2b MP |
6144 | |
6145 | unlock: | |
6146 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
6147 | if (pi) |
6148 | cpuset_read_unlock(); | |
4b211f2b | 6149 | return retval; |
1da177e4 | 6150 | } |
961ccddd | 6151 | |
7479f3c9 PZ |
6152 | static int _sched_setscheduler(struct task_struct *p, int policy, |
6153 | const struct sched_param *param, bool check) | |
6154 | { | |
6155 | struct sched_attr attr = { | |
6156 | .sched_policy = policy, | |
6157 | .sched_priority = param->sched_priority, | |
6158 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
6159 | }; | |
6160 | ||
c13db6b1 SR |
6161 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
6162 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
6163 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
6164 | policy &= ~SCHED_RESET_ON_FORK; | |
6165 | attr.sched_policy = policy; | |
6166 | } | |
6167 | ||
dbc7f069 | 6168 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 6169 | } |
961ccddd RR |
6170 | /** |
6171 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6172 | * @p: the task in question. | |
6173 | * @policy: new policy. | |
6174 | * @param: structure containing the new RT priority. | |
6175 | * | |
7318d4cc PZ |
6176 | * Use sched_set_fifo(), read its comment. |
6177 | * | |
e69f6186 YB |
6178 | * Return: 0 on success. An error code otherwise. |
6179 | * | |
961ccddd RR |
6180 | * NOTE that the task may be already dead. |
6181 | */ | |
6182 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 6183 | const struct sched_param *param) |
961ccddd | 6184 | { |
7479f3c9 | 6185 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 6186 | } |
1da177e4 | 6187 | |
d50dde5a DF |
6188 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
6189 | { | |
dbc7f069 | 6190 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 6191 | } |
d50dde5a | 6192 | |
794a56eb JL |
6193 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
6194 | { | |
6195 | return __sched_setscheduler(p, attr, false, true); | |
6196 | } | |
6197 | ||
961ccddd RR |
6198 | /** |
6199 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6200 | * @p: the task in question. | |
6201 | * @policy: new policy. | |
6202 | * @param: structure containing the new RT priority. | |
6203 | * | |
6204 | * Just like sched_setscheduler, only don't bother checking if the | |
6205 | * current context has permission. For example, this is needed in | |
6206 | * stop_machine(): we create temporary high priority worker threads, | |
6207 | * but our caller might not have that capability. | |
e69f6186 YB |
6208 | * |
6209 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
6210 | */ |
6211 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 6212 | const struct sched_param *param) |
961ccddd | 6213 | { |
7479f3c9 | 6214 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
6215 | } |
6216 | ||
7318d4cc PZ |
6217 | /* |
6218 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
6219 | * incapable of resource management, which is the one thing an OS really should | |
6220 | * be doing. | |
6221 | * | |
6222 | * This is of course the reason it is limited to privileged users only. | |
6223 | * | |
6224 | * Worse still; it is fundamentally impossible to compose static priority | |
6225 | * workloads. You cannot take two correctly working static prio workloads | |
6226 | * and smash them together and still expect them to work. | |
6227 | * | |
6228 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
6229 | * | |
6230 | * MAX_RT_PRIO / 2 | |
6231 | * | |
6232 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
6233 | * know enough information to make a sensible choice. | |
6234 | */ | |
8b700983 | 6235 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
6236 | { |
6237 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 6238 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6239 | } |
6240 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
6241 | ||
6242 | /* | |
6243 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
6244 | */ | |
8b700983 | 6245 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
6246 | { |
6247 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 6248 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
6249 | } |
6250 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
6251 | ||
8b700983 | 6252 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
6253 | { |
6254 | struct sched_attr attr = { | |
6255 | .sched_policy = SCHED_NORMAL, | |
6256 | .sched_nice = nice, | |
6257 | }; | |
8b700983 | 6258 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
6259 | } |
6260 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 6261 | |
95cdf3b7 IM |
6262 | static int |
6263 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6264 | { |
1da177e4 LT |
6265 | struct sched_param lparam; |
6266 | struct task_struct *p; | |
36c8b586 | 6267 | int retval; |
1da177e4 LT |
6268 | |
6269 | if (!param || pid < 0) | |
6270 | return -EINVAL; | |
6271 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6272 | return -EFAULT; | |
5fe1d75f ON |
6273 | |
6274 | rcu_read_lock(); | |
6275 | retval = -ESRCH; | |
1da177e4 | 6276 | p = find_process_by_pid(pid); |
710da3c8 JL |
6277 | if (likely(p)) |
6278 | get_task_struct(p); | |
5fe1d75f | 6279 | rcu_read_unlock(); |
36c8b586 | 6280 | |
710da3c8 JL |
6281 | if (likely(p)) { |
6282 | retval = sched_setscheduler(p, policy, &lparam); | |
6283 | put_task_struct(p); | |
6284 | } | |
6285 | ||
1da177e4 LT |
6286 | return retval; |
6287 | } | |
6288 | ||
d50dde5a DF |
6289 | /* |
6290 | * Mimics kernel/events/core.c perf_copy_attr(). | |
6291 | */ | |
d1ccc66d | 6292 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
6293 | { |
6294 | u32 size; | |
6295 | int ret; | |
6296 | ||
d1ccc66d | 6297 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
6298 | memset(attr, 0, sizeof(*attr)); |
6299 | ||
6300 | ret = get_user(size, &uattr->size); | |
6301 | if (ret) | |
6302 | return ret; | |
6303 | ||
d1ccc66d IM |
6304 | /* ABI compatibility quirk: */ |
6305 | if (!size) | |
d50dde5a | 6306 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 6307 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
6308 | goto err_size; |
6309 | ||
dff3a85f AS |
6310 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
6311 | if (ret) { | |
6312 | if (ret == -E2BIG) | |
6313 | goto err_size; | |
6314 | return ret; | |
d50dde5a DF |
6315 | } |
6316 | ||
a509a7cd PB |
6317 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
6318 | size < SCHED_ATTR_SIZE_VER1) | |
6319 | return -EINVAL; | |
6320 | ||
d50dde5a | 6321 | /* |
d1ccc66d | 6322 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
6323 | * to be strict and return an error on out-of-bounds values? |
6324 | */ | |
75e45d51 | 6325 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 6326 | |
e78c7bca | 6327 | return 0; |
d50dde5a DF |
6328 | |
6329 | err_size: | |
6330 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 6331 | return -E2BIG; |
d50dde5a DF |
6332 | } |
6333 | ||
1da177e4 LT |
6334 | /** |
6335 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6336 | * @pid: the pid in question. | |
6337 | * @policy: new policy. | |
6338 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6339 | * |
6340 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6341 | */ |
d1ccc66d | 6342 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 6343 | { |
c21761f1 JB |
6344 | if (policy < 0) |
6345 | return -EINVAL; | |
6346 | ||
1da177e4 LT |
6347 | return do_sched_setscheduler(pid, policy, param); |
6348 | } | |
6349 | ||
6350 | /** | |
6351 | * sys_sched_setparam - set/change the RT priority of a thread | |
6352 | * @pid: the pid in question. | |
6353 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
6354 | * |
6355 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6356 | */ |
5add95d4 | 6357 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6358 | { |
c13db6b1 | 6359 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
6360 | } |
6361 | ||
d50dde5a DF |
6362 | /** |
6363 | * sys_sched_setattr - same as above, but with extended sched_attr | |
6364 | * @pid: the pid in question. | |
5778fccf | 6365 | * @uattr: structure containing the extended parameters. |
db66d756 | 6366 | * @flags: for future extension. |
d50dde5a | 6367 | */ |
6d35ab48 PZ |
6368 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
6369 | unsigned int, flags) | |
d50dde5a DF |
6370 | { |
6371 | struct sched_attr attr; | |
6372 | struct task_struct *p; | |
6373 | int retval; | |
6374 | ||
6d35ab48 | 6375 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
6376 | return -EINVAL; |
6377 | ||
143cf23d MK |
6378 | retval = sched_copy_attr(uattr, &attr); |
6379 | if (retval) | |
6380 | return retval; | |
d50dde5a | 6381 | |
b14ed2c2 | 6382 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 6383 | return -EINVAL; |
1d6362fa PB |
6384 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
6385 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
6386 | |
6387 | rcu_read_lock(); | |
6388 | retval = -ESRCH; | |
6389 | p = find_process_by_pid(pid); | |
a509a7cd PB |
6390 | if (likely(p)) |
6391 | get_task_struct(p); | |
d50dde5a DF |
6392 | rcu_read_unlock(); |
6393 | ||
a509a7cd PB |
6394 | if (likely(p)) { |
6395 | retval = sched_setattr(p, &attr); | |
6396 | put_task_struct(p); | |
6397 | } | |
6398 | ||
d50dde5a DF |
6399 | return retval; |
6400 | } | |
6401 | ||
1da177e4 LT |
6402 | /** |
6403 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6404 | * @pid: the pid in question. | |
e69f6186 YB |
6405 | * |
6406 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
6407 | * code. | |
1da177e4 | 6408 | */ |
5add95d4 | 6409 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6410 | { |
36c8b586 | 6411 | struct task_struct *p; |
3a5c359a | 6412 | int retval; |
1da177e4 LT |
6413 | |
6414 | if (pid < 0) | |
3a5c359a | 6415 | return -EINVAL; |
1da177e4 LT |
6416 | |
6417 | retval = -ESRCH; | |
5fe85be0 | 6418 | rcu_read_lock(); |
1da177e4 LT |
6419 | p = find_process_by_pid(pid); |
6420 | if (p) { | |
6421 | retval = security_task_getscheduler(p); | |
6422 | if (!retval) | |
ca94c442 LP |
6423 | retval = p->policy |
6424 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 6425 | } |
5fe85be0 | 6426 | rcu_read_unlock(); |
1da177e4 LT |
6427 | return retval; |
6428 | } | |
6429 | ||
6430 | /** | |
ca94c442 | 6431 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6432 | * @pid: the pid in question. |
6433 | * @param: structure containing the RT priority. | |
e69f6186 YB |
6434 | * |
6435 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
6436 | * code. | |
1da177e4 | 6437 | */ |
5add95d4 | 6438 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 6439 | { |
ce5f7f82 | 6440 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 6441 | struct task_struct *p; |
3a5c359a | 6442 | int retval; |
1da177e4 LT |
6443 | |
6444 | if (!param || pid < 0) | |
3a5c359a | 6445 | return -EINVAL; |
1da177e4 | 6446 | |
5fe85be0 | 6447 | rcu_read_lock(); |
1da177e4 LT |
6448 | p = find_process_by_pid(pid); |
6449 | retval = -ESRCH; | |
6450 | if (!p) | |
6451 | goto out_unlock; | |
6452 | ||
6453 | retval = security_task_getscheduler(p); | |
6454 | if (retval) | |
6455 | goto out_unlock; | |
6456 | ||
ce5f7f82 PZ |
6457 | if (task_has_rt_policy(p)) |
6458 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 6459 | rcu_read_unlock(); |
1da177e4 LT |
6460 | |
6461 | /* | |
6462 | * This one might sleep, we cannot do it with a spinlock held ... | |
6463 | */ | |
6464 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6465 | ||
1da177e4 LT |
6466 | return retval; |
6467 | ||
6468 | out_unlock: | |
5fe85be0 | 6469 | rcu_read_unlock(); |
1da177e4 LT |
6470 | return retval; |
6471 | } | |
6472 | ||
1251201c IM |
6473 | /* |
6474 | * Copy the kernel size attribute structure (which might be larger | |
6475 | * than what user-space knows about) to user-space. | |
6476 | * | |
6477 | * Note that all cases are valid: user-space buffer can be larger or | |
6478 | * smaller than the kernel-space buffer. The usual case is that both | |
6479 | * have the same size. | |
6480 | */ | |
6481 | static int | |
6482 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
6483 | struct sched_attr *kattr, | |
6484 | unsigned int usize) | |
d50dde5a | 6485 | { |
1251201c | 6486 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 6487 | |
96d4f267 | 6488 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
6489 | return -EFAULT; |
6490 | ||
6491 | /* | |
1251201c IM |
6492 | * sched_getattr() ABI forwards and backwards compatibility: |
6493 | * | |
6494 | * If usize == ksize then we just copy everything to user-space and all is good. | |
6495 | * | |
6496 | * If usize < ksize then we only copy as much as user-space has space for, | |
6497 | * this keeps ABI compatibility as well. We skip the rest. | |
6498 | * | |
6499 | * If usize > ksize then user-space is using a newer version of the ABI, | |
6500 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
6501 | * detect the kernel's knowledge of attributes from the attr->size value | |
6502 | * which is set to ksize in this case. | |
d50dde5a | 6503 | */ |
1251201c | 6504 | kattr->size = min(usize, ksize); |
d50dde5a | 6505 | |
1251201c | 6506 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
6507 | return -EFAULT; |
6508 | ||
22400674 | 6509 | return 0; |
d50dde5a DF |
6510 | } |
6511 | ||
6512 | /** | |
aab03e05 | 6513 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 6514 | * @pid: the pid in question. |
5778fccf | 6515 | * @uattr: structure containing the extended parameters. |
dff3a85f | 6516 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 6517 | * @flags: for future extension. |
d50dde5a | 6518 | */ |
6d35ab48 | 6519 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 6520 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 6521 | { |
1251201c | 6522 | struct sched_attr kattr = { }; |
d50dde5a DF |
6523 | struct task_struct *p; |
6524 | int retval; | |
6525 | ||
1251201c IM |
6526 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
6527 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
6528 | return -EINVAL; |
6529 | ||
6530 | rcu_read_lock(); | |
6531 | p = find_process_by_pid(pid); | |
6532 | retval = -ESRCH; | |
6533 | if (!p) | |
6534 | goto out_unlock; | |
6535 | ||
6536 | retval = security_task_getscheduler(p); | |
6537 | if (retval) | |
6538 | goto out_unlock; | |
6539 | ||
1251201c | 6540 | kattr.sched_policy = p->policy; |
7479f3c9 | 6541 | if (p->sched_reset_on_fork) |
1251201c | 6542 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
aab03e05 | 6543 | if (task_has_dl_policy(p)) |
1251201c | 6544 | __getparam_dl(p, &kattr); |
aab03e05 | 6545 | else if (task_has_rt_policy(p)) |
1251201c | 6546 | kattr.sched_priority = p->rt_priority; |
d50dde5a | 6547 | else |
1251201c | 6548 | kattr.sched_nice = task_nice(p); |
d50dde5a | 6549 | |
a509a7cd | 6550 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
6551 | /* |
6552 | * This could race with another potential updater, but this is fine | |
6553 | * because it'll correctly read the old or the new value. We don't need | |
6554 | * to guarantee who wins the race as long as it doesn't return garbage. | |
6555 | */ | |
1251201c IM |
6556 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
6557 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
6558 | #endif |
6559 | ||
d50dde5a DF |
6560 | rcu_read_unlock(); |
6561 | ||
1251201c | 6562 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
6563 | |
6564 | out_unlock: | |
6565 | rcu_read_unlock(); | |
6566 | return retval; | |
6567 | } | |
6568 | ||
96f874e2 | 6569 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6570 | { |
5a16f3d3 | 6571 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6572 | struct task_struct *p; |
6573 | int retval; | |
1da177e4 | 6574 | |
23f5d142 | 6575 | rcu_read_lock(); |
1da177e4 LT |
6576 | |
6577 | p = find_process_by_pid(pid); | |
6578 | if (!p) { | |
23f5d142 | 6579 | rcu_read_unlock(); |
1da177e4 LT |
6580 | return -ESRCH; |
6581 | } | |
6582 | ||
23f5d142 | 6583 | /* Prevent p going away */ |
1da177e4 | 6584 | get_task_struct(p); |
23f5d142 | 6585 | rcu_read_unlock(); |
1da177e4 | 6586 | |
14a40ffc TH |
6587 | if (p->flags & PF_NO_SETAFFINITY) { |
6588 | retval = -EINVAL; | |
6589 | goto out_put_task; | |
6590 | } | |
5a16f3d3 RR |
6591 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6592 | retval = -ENOMEM; | |
6593 | goto out_put_task; | |
6594 | } | |
6595 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6596 | retval = -ENOMEM; | |
6597 | goto out_free_cpus_allowed; | |
6598 | } | |
1da177e4 | 6599 | retval = -EPERM; |
4c44aaaf EB |
6600 | if (!check_same_owner(p)) { |
6601 | rcu_read_lock(); | |
6602 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
6603 | rcu_read_unlock(); | |
16303ab2 | 6604 | goto out_free_new_mask; |
4c44aaaf EB |
6605 | } |
6606 | rcu_read_unlock(); | |
6607 | } | |
1da177e4 | 6608 | |
b0ae1981 | 6609 | retval = security_task_setscheduler(p); |
e7834f8f | 6610 | if (retval) |
16303ab2 | 6611 | goto out_free_new_mask; |
e7834f8f | 6612 | |
e4099a5e PZ |
6613 | |
6614 | cpuset_cpus_allowed(p, cpus_allowed); | |
6615 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
6616 | ||
332ac17e DF |
6617 | /* |
6618 | * Since bandwidth control happens on root_domain basis, | |
6619 | * if admission test is enabled, we only admit -deadline | |
6620 | * tasks allowed to run on all the CPUs in the task's | |
6621 | * root_domain. | |
6622 | */ | |
6623 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
6624 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
6625 | rcu_read_lock(); | |
6626 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 6627 | retval = -EBUSY; |
f1e3a093 | 6628 | rcu_read_unlock(); |
16303ab2 | 6629 | goto out_free_new_mask; |
332ac17e | 6630 | } |
f1e3a093 | 6631 | rcu_read_unlock(); |
332ac17e DF |
6632 | } |
6633 | #endif | |
49246274 | 6634 | again: |
9cfc3e18 | 6635 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK); |
1da177e4 | 6636 | |
8707d8b8 | 6637 | if (!retval) { |
5a16f3d3 RR |
6638 | cpuset_cpus_allowed(p, cpus_allowed); |
6639 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6640 | /* |
6641 | * We must have raced with a concurrent cpuset | |
6642 | * update. Just reset the cpus_allowed to the | |
6643 | * cpuset's cpus_allowed | |
6644 | */ | |
5a16f3d3 | 6645 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6646 | goto again; |
6647 | } | |
6648 | } | |
16303ab2 | 6649 | out_free_new_mask: |
5a16f3d3 RR |
6650 | free_cpumask_var(new_mask); |
6651 | out_free_cpus_allowed: | |
6652 | free_cpumask_var(cpus_allowed); | |
6653 | out_put_task: | |
1da177e4 | 6654 | put_task_struct(p); |
1da177e4 LT |
6655 | return retval; |
6656 | } | |
6657 | ||
6658 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6659 | struct cpumask *new_mask) |
1da177e4 | 6660 | { |
96f874e2 RR |
6661 | if (len < cpumask_size()) |
6662 | cpumask_clear(new_mask); | |
6663 | else if (len > cpumask_size()) | |
6664 | len = cpumask_size(); | |
6665 | ||
1da177e4 LT |
6666 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6667 | } | |
6668 | ||
6669 | /** | |
d1ccc66d | 6670 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
6671 | * @pid: pid of the process |
6672 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6673 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
6674 | * |
6675 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 6676 | */ |
5add95d4 HC |
6677 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6678 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6679 | { |
5a16f3d3 | 6680 | cpumask_var_t new_mask; |
1da177e4 LT |
6681 | int retval; |
6682 | ||
5a16f3d3 RR |
6683 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6684 | return -ENOMEM; | |
1da177e4 | 6685 | |
5a16f3d3 RR |
6686 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6687 | if (retval == 0) | |
6688 | retval = sched_setaffinity(pid, new_mask); | |
6689 | free_cpumask_var(new_mask); | |
6690 | return retval; | |
1da177e4 LT |
6691 | } |
6692 | ||
96f874e2 | 6693 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6694 | { |
36c8b586 | 6695 | struct task_struct *p; |
31605683 | 6696 | unsigned long flags; |
1da177e4 | 6697 | int retval; |
1da177e4 | 6698 | |
23f5d142 | 6699 | rcu_read_lock(); |
1da177e4 LT |
6700 | |
6701 | retval = -ESRCH; | |
6702 | p = find_process_by_pid(pid); | |
6703 | if (!p) | |
6704 | goto out_unlock; | |
6705 | ||
e7834f8f DQ |
6706 | retval = security_task_getscheduler(p); |
6707 | if (retval) | |
6708 | goto out_unlock; | |
6709 | ||
013fdb80 | 6710 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 6711 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 6712 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
6713 | |
6714 | out_unlock: | |
23f5d142 | 6715 | rcu_read_unlock(); |
1da177e4 | 6716 | |
9531b62f | 6717 | return retval; |
1da177e4 LT |
6718 | } |
6719 | ||
6720 | /** | |
d1ccc66d | 6721 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
6722 | * @pid: pid of the process |
6723 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6724 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 6725 | * |
599b4840 ZW |
6726 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
6727 | * error code otherwise. | |
1da177e4 | 6728 | */ |
5add95d4 HC |
6729 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6730 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6731 | { |
6732 | int ret; | |
f17c8607 | 6733 | cpumask_var_t mask; |
1da177e4 | 6734 | |
84fba5ec | 6735 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
6736 | return -EINVAL; |
6737 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
6738 | return -EINVAL; |
6739 | ||
f17c8607 RR |
6740 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6741 | return -ENOMEM; | |
1da177e4 | 6742 | |
f17c8607 RR |
6743 | ret = sched_getaffinity(pid, mask); |
6744 | if (ret == 0) { | |
4de373a1 | 6745 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
6746 | |
6747 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
6748 | ret = -EFAULT; |
6749 | else | |
cd3d8031 | 6750 | ret = retlen; |
f17c8607 RR |
6751 | } |
6752 | free_cpumask_var(mask); | |
1da177e4 | 6753 | |
f17c8607 | 6754 | return ret; |
1da177e4 LT |
6755 | } |
6756 | ||
7d4dd4f1 | 6757 | static void do_sched_yield(void) |
1da177e4 | 6758 | { |
8a8c69c3 PZ |
6759 | struct rq_flags rf; |
6760 | struct rq *rq; | |
6761 | ||
246b3b33 | 6762 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 6763 | |
ae92882e | 6764 | schedstat_inc(rq->yld_count); |
4530d7ab | 6765 | current->sched_class->yield_task(rq); |
1da177e4 | 6766 | |
8a8c69c3 | 6767 | preempt_disable(); |
345a957f | 6768 | rq_unlock_irq(rq, &rf); |
ba74c144 | 6769 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
6770 | |
6771 | schedule(); | |
7d4dd4f1 | 6772 | } |
1da177e4 | 6773 | |
59a74b15 MCC |
6774 | /** |
6775 | * sys_sched_yield - yield the current processor to other threads. | |
6776 | * | |
6777 | * This function yields the current CPU to other tasks. If there are no | |
6778 | * other threads running on this CPU then this function will return. | |
6779 | * | |
6780 | * Return: 0. | |
6781 | */ | |
7d4dd4f1 DB |
6782 | SYSCALL_DEFINE0(sched_yield) |
6783 | { | |
6784 | do_sched_yield(); | |
1da177e4 LT |
6785 | return 0; |
6786 | } | |
6787 | ||
c1a280b6 | 6788 | #ifndef CONFIG_PREEMPTION |
02b67cc3 | 6789 | int __sched _cond_resched(void) |
1da177e4 | 6790 | { |
fe32d3cd | 6791 | if (should_resched(0)) { |
a18b5d01 | 6792 | preempt_schedule_common(); |
1da177e4 LT |
6793 | return 1; |
6794 | } | |
f79c3ad6 | 6795 | rcu_all_qs(); |
1da177e4 LT |
6796 | return 0; |
6797 | } | |
02b67cc3 | 6798 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 6799 | #endif |
1da177e4 LT |
6800 | |
6801 | /* | |
613afbf8 | 6802 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6803 | * call schedule, and on return reacquire the lock. |
6804 | * | |
c1a280b6 | 6805 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
6806 | * operations here to prevent schedule() from being called twice (once via |
6807 | * spin_unlock(), once by hand). | |
6808 | */ | |
613afbf8 | 6809 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6810 | { |
fe32d3cd | 6811 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
6812 | int ret = 0; |
6813 | ||
f607c668 PZ |
6814 | lockdep_assert_held(lock); |
6815 | ||
4a81e832 | 6816 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6817 | spin_unlock(lock); |
d86ee480 | 6818 | if (resched) |
a18b5d01 | 6819 | preempt_schedule_common(); |
95c354fe NP |
6820 | else |
6821 | cpu_relax(); | |
6df3cecb | 6822 | ret = 1; |
1da177e4 | 6823 | spin_lock(lock); |
1da177e4 | 6824 | } |
6df3cecb | 6825 | return ret; |
1da177e4 | 6826 | } |
613afbf8 | 6827 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6828 | |
1da177e4 LT |
6829 | /** |
6830 | * yield - yield the current processor to other threads. | |
6831 | * | |
8e3fabfd PZ |
6832 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
6833 | * | |
6834 | * The scheduler is at all times free to pick the calling task as the most | |
6835 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 6836 | * it, it's already broken. |
8e3fabfd PZ |
6837 | * |
6838 | * Typical broken usage is: | |
6839 | * | |
6840 | * while (!event) | |
d1ccc66d | 6841 | * yield(); |
8e3fabfd PZ |
6842 | * |
6843 | * where one assumes that yield() will let 'the other' process run that will | |
6844 | * make event true. If the current task is a SCHED_FIFO task that will never | |
6845 | * happen. Never use yield() as a progress guarantee!! | |
6846 | * | |
6847 | * If you want to use yield() to wait for something, use wait_event(). | |
6848 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
6849 | * If you still want to use yield(), do not! | |
1da177e4 LT |
6850 | */ |
6851 | void __sched yield(void) | |
6852 | { | |
6853 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 6854 | do_sched_yield(); |
1da177e4 | 6855 | } |
1da177e4 LT |
6856 | EXPORT_SYMBOL(yield); |
6857 | ||
d95f4122 MG |
6858 | /** |
6859 | * yield_to - yield the current processor to another thread in | |
6860 | * your thread group, or accelerate that thread toward the | |
6861 | * processor it's on. | |
16addf95 RD |
6862 | * @p: target task |
6863 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
6864 | * |
6865 | * It's the caller's job to ensure that the target task struct | |
6866 | * can't go away on us before we can do any checks. | |
6867 | * | |
e69f6186 | 6868 | * Return: |
7b270f60 PZ |
6869 | * true (>0) if we indeed boosted the target task. |
6870 | * false (0) if we failed to boost the target. | |
6871 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 6872 | */ |
fa93384f | 6873 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
6874 | { |
6875 | struct task_struct *curr = current; | |
6876 | struct rq *rq, *p_rq; | |
6877 | unsigned long flags; | |
c3c18640 | 6878 | int yielded = 0; |
d95f4122 MG |
6879 | |
6880 | local_irq_save(flags); | |
6881 | rq = this_rq(); | |
6882 | ||
6883 | again: | |
6884 | p_rq = task_rq(p); | |
7b270f60 PZ |
6885 | /* |
6886 | * If we're the only runnable task on the rq and target rq also | |
6887 | * has only one task, there's absolutely no point in yielding. | |
6888 | */ | |
6889 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
6890 | yielded = -ESRCH; | |
6891 | goto out_irq; | |
6892 | } | |
6893 | ||
d95f4122 | 6894 | double_rq_lock(rq, p_rq); |
39e24d8f | 6895 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
6896 | double_rq_unlock(rq, p_rq); |
6897 | goto again; | |
6898 | } | |
6899 | ||
6900 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 6901 | goto out_unlock; |
d95f4122 MG |
6902 | |
6903 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 6904 | goto out_unlock; |
d95f4122 MG |
6905 | |
6906 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 6907 | goto out_unlock; |
d95f4122 | 6908 | |
0900acf2 | 6909 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 6910 | if (yielded) { |
ae92882e | 6911 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
6912 | /* |
6913 | * Make p's CPU reschedule; pick_next_entity takes care of | |
6914 | * fairness. | |
6915 | */ | |
6916 | if (preempt && rq != p_rq) | |
8875125e | 6917 | resched_curr(p_rq); |
6d1cafd8 | 6918 | } |
d95f4122 | 6919 | |
7b270f60 | 6920 | out_unlock: |
d95f4122 | 6921 | double_rq_unlock(rq, p_rq); |
7b270f60 | 6922 | out_irq: |
d95f4122 MG |
6923 | local_irq_restore(flags); |
6924 | ||
7b270f60 | 6925 | if (yielded > 0) |
d95f4122 MG |
6926 | schedule(); |
6927 | ||
6928 | return yielded; | |
6929 | } | |
6930 | EXPORT_SYMBOL_GPL(yield_to); | |
6931 | ||
10ab5643 TH |
6932 | int io_schedule_prepare(void) |
6933 | { | |
6934 | int old_iowait = current->in_iowait; | |
6935 | ||
6936 | current->in_iowait = 1; | |
6937 | blk_schedule_flush_plug(current); | |
6938 | ||
6939 | return old_iowait; | |
6940 | } | |
6941 | ||
6942 | void io_schedule_finish(int token) | |
6943 | { | |
6944 | current->in_iowait = token; | |
6945 | } | |
6946 | ||
1da177e4 | 6947 | /* |
41a2d6cf | 6948 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6949 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 6950 | */ |
1da177e4 LT |
6951 | long __sched io_schedule_timeout(long timeout) |
6952 | { | |
10ab5643 | 6953 | int token; |
1da177e4 LT |
6954 | long ret; |
6955 | ||
10ab5643 | 6956 | token = io_schedule_prepare(); |
1da177e4 | 6957 | ret = schedule_timeout(timeout); |
10ab5643 | 6958 | io_schedule_finish(token); |
9cff8ade | 6959 | |
1da177e4 LT |
6960 | return ret; |
6961 | } | |
9cff8ade | 6962 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 6963 | |
e3b929b0 | 6964 | void __sched io_schedule(void) |
10ab5643 TH |
6965 | { |
6966 | int token; | |
6967 | ||
6968 | token = io_schedule_prepare(); | |
6969 | schedule(); | |
6970 | io_schedule_finish(token); | |
6971 | } | |
6972 | EXPORT_SYMBOL(io_schedule); | |
6973 | ||
1da177e4 LT |
6974 | /** |
6975 | * sys_sched_get_priority_max - return maximum RT priority. | |
6976 | * @policy: scheduling class. | |
6977 | * | |
e69f6186 YB |
6978 | * Return: On success, this syscall returns the maximum |
6979 | * rt_priority that can be used by a given scheduling class. | |
6980 | * On failure, a negative error code is returned. | |
1da177e4 | 6981 | */ |
5add95d4 | 6982 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6983 | { |
6984 | int ret = -EINVAL; | |
6985 | ||
6986 | switch (policy) { | |
6987 | case SCHED_FIFO: | |
6988 | case SCHED_RR: | |
ae18ad28 | 6989 | ret = MAX_RT_PRIO-1; |
1da177e4 | 6990 | break; |
aab03e05 | 6991 | case SCHED_DEADLINE: |
1da177e4 | 6992 | case SCHED_NORMAL: |
b0a9499c | 6993 | case SCHED_BATCH: |
dd41f596 | 6994 | case SCHED_IDLE: |
1da177e4 LT |
6995 | ret = 0; |
6996 | break; | |
6997 | } | |
6998 | return ret; | |
6999 | } | |
7000 | ||
7001 | /** | |
7002 | * sys_sched_get_priority_min - return minimum RT priority. | |
7003 | * @policy: scheduling class. | |
7004 | * | |
e69f6186 YB |
7005 | * Return: On success, this syscall returns the minimum |
7006 | * rt_priority that can be used by a given scheduling class. | |
7007 | * On failure, a negative error code is returned. | |
1da177e4 | 7008 | */ |
5add95d4 | 7009 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
7010 | { |
7011 | int ret = -EINVAL; | |
7012 | ||
7013 | switch (policy) { | |
7014 | case SCHED_FIFO: | |
7015 | case SCHED_RR: | |
7016 | ret = 1; | |
7017 | break; | |
aab03e05 | 7018 | case SCHED_DEADLINE: |
1da177e4 | 7019 | case SCHED_NORMAL: |
b0a9499c | 7020 | case SCHED_BATCH: |
dd41f596 | 7021 | case SCHED_IDLE: |
1da177e4 LT |
7022 | ret = 0; |
7023 | } | |
7024 | return ret; | |
7025 | } | |
7026 | ||
abca5fc5 | 7027 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 7028 | { |
36c8b586 | 7029 | struct task_struct *p; |
a4ec24b4 | 7030 | unsigned int time_slice; |
eb580751 | 7031 | struct rq_flags rf; |
dba091b9 | 7032 | struct rq *rq; |
3a5c359a | 7033 | int retval; |
1da177e4 LT |
7034 | |
7035 | if (pid < 0) | |
3a5c359a | 7036 | return -EINVAL; |
1da177e4 LT |
7037 | |
7038 | retval = -ESRCH; | |
1a551ae7 | 7039 | rcu_read_lock(); |
1da177e4 LT |
7040 | p = find_process_by_pid(pid); |
7041 | if (!p) | |
7042 | goto out_unlock; | |
7043 | ||
7044 | retval = security_task_getscheduler(p); | |
7045 | if (retval) | |
7046 | goto out_unlock; | |
7047 | ||
eb580751 | 7048 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
7049 | time_slice = 0; |
7050 | if (p->sched_class->get_rr_interval) | |
7051 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 7052 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 7053 | |
1a551ae7 | 7054 | rcu_read_unlock(); |
abca5fc5 AV |
7055 | jiffies_to_timespec64(time_slice, t); |
7056 | return 0; | |
3a5c359a | 7057 | |
1da177e4 | 7058 | out_unlock: |
1a551ae7 | 7059 | rcu_read_unlock(); |
1da177e4 LT |
7060 | return retval; |
7061 | } | |
7062 | ||
2064a5ab RD |
7063 | /** |
7064 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
7065 | * @pid: pid of the process. | |
7066 | * @interval: userspace pointer to the timeslice value. | |
7067 | * | |
7068 | * this syscall writes the default timeslice value of a given process | |
7069 | * into the user-space timespec buffer. A value of '0' means infinity. | |
7070 | * | |
7071 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
7072 | * an error code. | |
7073 | */ | |
abca5fc5 | 7074 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 7075 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
7076 | { |
7077 | struct timespec64 t; | |
7078 | int retval = sched_rr_get_interval(pid, &t); | |
7079 | ||
7080 | if (retval == 0) | |
7081 | retval = put_timespec64(&t, interval); | |
7082 | ||
7083 | return retval; | |
7084 | } | |
7085 | ||
474b9c77 | 7086 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
7087 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
7088 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
7089 | { |
7090 | struct timespec64 t; | |
7091 | int retval = sched_rr_get_interval(pid, &t); | |
7092 | ||
7093 | if (retval == 0) | |
9afc5eee | 7094 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
7095 | return retval; |
7096 | } | |
7097 | #endif | |
7098 | ||
82a1fcb9 | 7099 | void sched_show_task(struct task_struct *p) |
1da177e4 | 7100 | { |
1da177e4 | 7101 | unsigned long free = 0; |
4e79752c | 7102 | int ppid; |
c930b2c0 | 7103 | |
38200502 TH |
7104 | if (!try_get_task_stack(p)) |
7105 | return; | |
20435d84 | 7106 | |
cc172ff3 | 7107 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 XX |
7108 | |
7109 | if (p->state == TASK_RUNNING) | |
cc172ff3 | 7110 | pr_cont(" running task "); |
1da177e4 | 7111 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 7112 | free = stack_not_used(p); |
1da177e4 | 7113 | #endif |
a90e984c | 7114 | ppid = 0; |
4e79752c | 7115 | rcu_read_lock(); |
a90e984c ON |
7116 | if (pid_alive(p)) |
7117 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 7118 | rcu_read_unlock(); |
cc172ff3 LZ |
7119 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
7120 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 7121 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 7122 | |
3d1cb205 | 7123 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 7124 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 7125 | show_stack(p, NULL, KERN_INFO); |
38200502 | 7126 | put_task_stack(p); |
1da177e4 | 7127 | } |
0032f4e8 | 7128 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 7129 | |
5d68cc95 PZ |
7130 | static inline bool |
7131 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
7132 | { | |
7133 | /* no filter, everything matches */ | |
7134 | if (!state_filter) | |
7135 | return true; | |
7136 | ||
7137 | /* filter, but doesn't match */ | |
7138 | if (!(p->state & state_filter)) | |
7139 | return false; | |
7140 | ||
7141 | /* | |
7142 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
7143 | * TASK_KILLABLE). | |
7144 | */ | |
7145 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
7146 | return false; | |
7147 | ||
7148 | return true; | |
7149 | } | |
7150 | ||
7151 | ||
e59e2ae2 | 7152 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 7153 | { |
36c8b586 | 7154 | struct task_struct *g, *p; |
1da177e4 | 7155 | |
510f5acc | 7156 | rcu_read_lock(); |
5d07f420 | 7157 | for_each_process_thread(g, p) { |
1da177e4 LT |
7158 | /* |
7159 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 7160 | * console might take a lot of time: |
57675cb9 AR |
7161 | * Also, reset softlockup watchdogs on all CPUs, because |
7162 | * another CPU might be blocked waiting for us to process | |
7163 | * an IPI. | |
1da177e4 LT |
7164 | */ |
7165 | touch_nmi_watchdog(); | |
57675cb9 | 7166 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 7167 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 7168 | sched_show_task(p); |
5d07f420 | 7169 | } |
1da177e4 | 7170 | |
dd41f596 | 7171 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
7172 | if (!state_filter) |
7173 | sysrq_sched_debug_show(); | |
dd41f596 | 7174 | #endif |
510f5acc | 7175 | rcu_read_unlock(); |
e59e2ae2 IM |
7176 | /* |
7177 | * Only show locks if all tasks are dumped: | |
7178 | */ | |
93335a21 | 7179 | if (!state_filter) |
e59e2ae2 | 7180 | debug_show_all_locks(); |
1da177e4 LT |
7181 | } |
7182 | ||
f340c0d1 IM |
7183 | /** |
7184 | * init_idle - set up an idle thread for a given CPU | |
7185 | * @idle: task in question | |
d1ccc66d | 7186 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
7187 | * |
7188 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
7189 | * flag, to make booting more robust. | |
7190 | */ | |
0db0628d | 7191 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 7192 | { |
70b97a7f | 7193 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
7194 | unsigned long flags; |
7195 | ||
ff51ff84 PZ |
7196 | __sched_fork(0, idle); |
7197 | ||
25834c73 PZ |
7198 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
7199 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 7200 | |
06b83b5f | 7201 | idle->state = TASK_RUNNING; |
dd41f596 | 7202 | idle->se.exec_start = sched_clock(); |
c1de45ca | 7203 | idle->flags |= PF_IDLE; |
dd41f596 | 7204 | |
d08b9f0c | 7205 | scs_task_reset(idle); |
e1b77c92 MR |
7206 | kasan_unpoison_task_stack(idle); |
7207 | ||
de9b8f5d PZ |
7208 | #ifdef CONFIG_SMP |
7209 | /* | |
b19a888c | 7210 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
7211 | * in that case do_set_cpus_allowed() will not do the right thing. |
7212 | * | |
7213 | * And since this is boot we can forgo the serialization. | |
7214 | */ | |
9cfc3e18 | 7215 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 7216 | #endif |
6506cf6c PZ |
7217 | /* |
7218 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 7219 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
7220 | * lockdep check in task_group() will fail. |
7221 | * | |
7222 | * Similar case to sched_fork(). / Alternatively we could | |
7223 | * use task_rq_lock() here and obtain the other rq->lock. | |
7224 | * | |
7225 | * Silence PROVE_RCU | |
7226 | */ | |
7227 | rcu_read_lock(); | |
dd41f596 | 7228 | __set_task_cpu(idle, cpu); |
6506cf6c | 7229 | rcu_read_unlock(); |
1da177e4 | 7230 | |
5311a98f EB |
7231 | rq->idle = idle; |
7232 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 7233 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 7234 | #ifdef CONFIG_SMP |
3ca7a440 | 7235 | idle->on_cpu = 1; |
4866cde0 | 7236 | #endif |
25834c73 PZ |
7237 | raw_spin_unlock(&rq->lock); |
7238 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
7239 | |
7240 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 7241 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 7242 | |
dd41f596 IM |
7243 | /* |
7244 | * The idle tasks have their own, simple scheduling class: | |
7245 | */ | |
7246 | idle->sched_class = &idle_sched_class; | |
868baf07 | 7247 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 7248 | vtime_init_idle(idle, cpu); |
de9b8f5d | 7249 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
7250 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
7251 | #endif | |
19978ca6 IM |
7252 | } |
7253 | ||
e1d4eeec NP |
7254 | #ifdef CONFIG_SMP |
7255 | ||
f82f8042 JL |
7256 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
7257 | const struct cpumask *trial) | |
7258 | { | |
06a76fe0 | 7259 | int ret = 1; |
f82f8042 | 7260 | |
bb2bc55a MG |
7261 | if (!cpumask_weight(cur)) |
7262 | return ret; | |
7263 | ||
06a76fe0 | 7264 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
7265 | |
7266 | return ret; | |
7267 | } | |
7268 | ||
7f51412a JL |
7269 | int task_can_attach(struct task_struct *p, |
7270 | const struct cpumask *cs_cpus_allowed) | |
7271 | { | |
7272 | int ret = 0; | |
7273 | ||
7274 | /* | |
7275 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 7276 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
7277 | * affinity and isolating such threads by their set of |
7278 | * allowed nodes is unnecessary. Thus, cpusets are not | |
7279 | * applicable for such threads. This prevents checking for | |
7280 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 7281 | * before cpus_mask may be changed. |
7f51412a JL |
7282 | */ |
7283 | if (p->flags & PF_NO_SETAFFINITY) { | |
7284 | ret = -EINVAL; | |
7285 | goto out; | |
7286 | } | |
7287 | ||
7f51412a | 7288 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
7289 | cs_cpus_allowed)) |
7290 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 7291 | |
7f51412a JL |
7292 | out: |
7293 | return ret; | |
7294 | } | |
7295 | ||
f2cb1360 | 7296 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 7297 | |
e6628d5b MG |
7298 | #ifdef CONFIG_NUMA_BALANCING |
7299 | /* Migrate current task p to target_cpu */ | |
7300 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
7301 | { | |
7302 | struct migration_arg arg = { p, target_cpu }; | |
7303 | int curr_cpu = task_cpu(p); | |
7304 | ||
7305 | if (curr_cpu == target_cpu) | |
7306 | return 0; | |
7307 | ||
3bd37062 | 7308 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
7309 | return -EINVAL; |
7310 | ||
7311 | /* TODO: This is not properly updating schedstats */ | |
7312 | ||
286549dc | 7313 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
7314 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
7315 | } | |
0ec8aa00 PZ |
7316 | |
7317 | /* | |
7318 | * Requeue a task on a given node and accurately track the number of NUMA | |
7319 | * tasks on the runqueues | |
7320 | */ | |
7321 | void sched_setnuma(struct task_struct *p, int nid) | |
7322 | { | |
da0c1e65 | 7323 | bool queued, running; |
eb580751 PZ |
7324 | struct rq_flags rf; |
7325 | struct rq *rq; | |
0ec8aa00 | 7326 | |
eb580751 | 7327 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 7328 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
7329 | running = task_current(rq, p); |
7330 | ||
da0c1e65 | 7331 | if (queued) |
1de64443 | 7332 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 7333 | if (running) |
f3cd1c4e | 7334 | put_prev_task(rq, p); |
0ec8aa00 PZ |
7335 | |
7336 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 7337 | |
da0c1e65 | 7338 | if (queued) |
7134b3e9 | 7339 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 7340 | if (running) |
03b7fad1 | 7341 | set_next_task(rq, p); |
eb580751 | 7342 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 7343 | } |
5cc389bc | 7344 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 7345 | |
1da177e4 | 7346 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 7347 | /* |
d1ccc66d | 7348 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 7349 | * offline. |
054b9108 | 7350 | */ |
48c5ccae | 7351 | void idle_task_exit(void) |
1da177e4 | 7352 | { |
48c5ccae | 7353 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 7354 | |
48c5ccae | 7355 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 7356 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 7357 | |
a53efe5f | 7358 | if (mm != &init_mm) { |
252d2a41 | 7359 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
7360 | finish_arch_post_lock_switch(); |
7361 | } | |
bf2c59fc PZ |
7362 | |
7363 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
7364 | } |
7365 | ||
2558aacf | 7366 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 7367 | { |
2558aacf PZ |
7368 | struct task_struct *p = arg; |
7369 | struct rq *rq = this_rq(); | |
7370 | struct rq_flags rf; | |
7371 | int cpu; | |
1da177e4 | 7372 | |
2558aacf PZ |
7373 | raw_spin_lock_irq(&p->pi_lock); |
7374 | rq_lock(rq, &rf); | |
3f1d2a31 | 7375 | |
2558aacf PZ |
7376 | update_rq_clock(rq); |
7377 | ||
7378 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
7379 | cpu = select_fallback_rq(rq->cpu, p); | |
7380 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 7381 | } |
3f1d2a31 | 7382 | |
2558aacf PZ |
7383 | rq_unlock(rq, &rf); |
7384 | raw_spin_unlock_irq(&p->pi_lock); | |
7385 | ||
7386 | put_task_struct(p); | |
7387 | ||
7388 | return 0; | |
10e7071b | 7389 | } |
3f1d2a31 | 7390 | |
2558aacf PZ |
7391 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
7392 | ||
48f24c4d | 7393 | /* |
2558aacf | 7394 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
1da177e4 | 7395 | */ |
2558aacf | 7396 | static void balance_push(struct rq *rq) |
1da177e4 | 7397 | { |
2558aacf PZ |
7398 | struct task_struct *push_task = rq->curr; |
7399 | ||
7400 | lockdep_assert_held(&rq->lock); | |
7401 | SCHED_WARN_ON(rq->cpu != smp_processor_id()); | |
ae792702 PZ |
7402 | /* |
7403 | * Ensure the thing is persistent until balance_push_set(.on = false); | |
7404 | */ | |
7405 | rq->balance_callback = &balance_push_callback; | |
1da177e4 LT |
7406 | |
7407 | /* | |
2558aacf PZ |
7408 | * Both the cpu-hotplug and stop task are in this case and are |
7409 | * required to complete the hotplug process. | |
5ba2ffba PZ |
7410 | * |
7411 | * XXX: the idle task does not match kthread_is_per_cpu() due to | |
7412 | * histerical raisins. | |
1da177e4 | 7413 | */ |
5ba2ffba PZ |
7414 | if (rq->idle == push_task || |
7415 | ((push_task->flags & PF_KTHREAD) && kthread_is_per_cpu(push_task)) || | |
7416 | is_migration_disabled(push_task)) { | |
7417 | ||
f2469a1f TG |
7418 | /* |
7419 | * If this is the idle task on the outgoing CPU try to wake | |
7420 | * up the hotplug control thread which might wait for the | |
7421 | * last task to vanish. The rcuwait_active() check is | |
7422 | * accurate here because the waiter is pinned on this CPU | |
7423 | * and can't obviously be running in parallel. | |
3015ef4b TG |
7424 | * |
7425 | * On RT kernels this also has to check whether there are | |
7426 | * pinned and scheduled out tasks on the runqueue. They | |
7427 | * need to leave the migrate disabled section first. | |
f2469a1f | 7428 | */ |
3015ef4b TG |
7429 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
7430 | rcuwait_active(&rq->hotplug_wait)) { | |
f2469a1f TG |
7431 | raw_spin_unlock(&rq->lock); |
7432 | rcuwait_wake_up(&rq->hotplug_wait); | |
7433 | raw_spin_lock(&rq->lock); | |
7434 | } | |
2558aacf | 7435 | return; |
f2469a1f | 7436 | } |
48f24c4d | 7437 | |
2558aacf | 7438 | get_task_struct(push_task); |
77bd3970 | 7439 | /* |
2558aacf PZ |
7440 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
7441 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 7442 | */ |
2558aacf PZ |
7443 | raw_spin_unlock(&rq->lock); |
7444 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, | |
7445 | this_cpu_ptr(&push_work)); | |
7446 | /* | |
7447 | * At this point need_resched() is true and we'll take the loop in | |
7448 | * schedule(). The next pick is obviously going to be the stop task | |
5ba2ffba | 7449 | * which kthread_is_per_cpu() and will push this task away. |
2558aacf PZ |
7450 | */ |
7451 | raw_spin_lock(&rq->lock); | |
7452 | } | |
77bd3970 | 7453 | |
2558aacf PZ |
7454 | static void balance_push_set(int cpu, bool on) |
7455 | { | |
7456 | struct rq *rq = cpu_rq(cpu); | |
7457 | struct rq_flags rf; | |
48c5ccae | 7458 | |
2558aacf | 7459 | rq_lock_irqsave(rq, &rf); |
975707f2 | 7460 | rq->balance_push = on; |
22f667c9 PZ |
7461 | if (on) { |
7462 | WARN_ON_ONCE(rq->balance_callback); | |
ae792702 | 7463 | rq->balance_callback = &balance_push_callback; |
22f667c9 | 7464 | } else if (rq->balance_callback == &balance_push_callback) { |
ae792702 | 7465 | rq->balance_callback = NULL; |
22f667c9 | 7466 | } |
2558aacf PZ |
7467 | rq_unlock_irqrestore(rq, &rf); |
7468 | } | |
e692ab53 | 7469 | |
f2469a1f TG |
7470 | /* |
7471 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
7472 | * inactive. All tasks which are not per CPU kernel threads are either | |
7473 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
7474 | * during wakeup. Wait until the CPU is quiescent. | |
7475 | */ | |
7476 | static void balance_hotplug_wait(void) | |
7477 | { | |
7478 | struct rq *rq = this_rq(); | |
5473e0cc | 7479 | |
3015ef4b TG |
7480 | rcuwait_wait_event(&rq->hotplug_wait, |
7481 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
7482 | TASK_UNINTERRUPTIBLE); |
7483 | } | |
5473e0cc | 7484 | |
2558aacf | 7485 | #else |
dce48a84 | 7486 | |
2558aacf PZ |
7487 | static inline void balance_push(struct rq *rq) |
7488 | { | |
dce48a84 | 7489 | } |
dce48a84 | 7490 | |
2558aacf PZ |
7491 | static inline void balance_push_set(int cpu, bool on) |
7492 | { | |
7493 | } | |
7494 | ||
f2469a1f TG |
7495 | static inline void balance_hotplug_wait(void) |
7496 | { | |
dce48a84 | 7497 | } |
f2469a1f | 7498 | |
1da177e4 LT |
7499 | #endif /* CONFIG_HOTPLUG_CPU */ |
7500 | ||
f2cb1360 | 7501 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
7502 | { |
7503 | if (!rq->online) { | |
7504 | const struct sched_class *class; | |
7505 | ||
c6c4927b | 7506 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7507 | rq->online = 1; |
7508 | ||
7509 | for_each_class(class) { | |
7510 | if (class->rq_online) | |
7511 | class->rq_online(rq); | |
7512 | } | |
7513 | } | |
7514 | } | |
7515 | ||
f2cb1360 | 7516 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
7517 | { |
7518 | if (rq->online) { | |
7519 | const struct sched_class *class; | |
7520 | ||
7521 | for_each_class(class) { | |
7522 | if (class->rq_offline) | |
7523 | class->rq_offline(rq); | |
7524 | } | |
7525 | ||
c6c4927b | 7526 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7527 | rq->online = 0; |
7528 | } | |
7529 | } | |
7530 | ||
d1ccc66d IM |
7531 | /* |
7532 | * used to mark begin/end of suspend/resume: | |
7533 | */ | |
7534 | static int num_cpus_frozen; | |
d35be8ba | 7535 | |
1da177e4 | 7536 | /* |
3a101d05 TH |
7537 | * Update cpusets according to cpu_active mask. If cpusets are |
7538 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7539 | * around partition_sched_domains(). | |
d35be8ba SB |
7540 | * |
7541 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
7542 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 7543 | */ |
40190a78 | 7544 | static void cpuset_cpu_active(void) |
e761b772 | 7545 | { |
40190a78 | 7546 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
7547 | /* |
7548 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
7549 | * resume sequence. As long as this is not the last online | |
7550 | * operation in the resume sequence, just build a single sched | |
7551 | * domain, ignoring cpusets. | |
7552 | */ | |
50e76632 PZ |
7553 | partition_sched_domains(1, NULL, NULL); |
7554 | if (--num_cpus_frozen) | |
135fb3e1 | 7555 | return; |
d35be8ba SB |
7556 | /* |
7557 | * This is the last CPU online operation. So fall through and | |
7558 | * restore the original sched domains by considering the | |
7559 | * cpuset configurations. | |
7560 | */ | |
50e76632 | 7561 | cpuset_force_rebuild(); |
3a101d05 | 7562 | } |
30e03acd | 7563 | cpuset_update_active_cpus(); |
3a101d05 | 7564 | } |
e761b772 | 7565 | |
40190a78 | 7566 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 7567 | { |
40190a78 | 7568 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 7569 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 7570 | return -EBUSY; |
30e03acd | 7571 | cpuset_update_active_cpus(); |
135fb3e1 | 7572 | } else { |
d35be8ba SB |
7573 | num_cpus_frozen++; |
7574 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 7575 | } |
135fb3e1 | 7576 | return 0; |
e761b772 | 7577 | } |
e761b772 | 7578 | |
40190a78 | 7579 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 7580 | { |
7d976699 | 7581 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 7582 | struct rq_flags rf; |
7d976699 | 7583 | |
22f667c9 PZ |
7584 | /* |
7585 | * Make sure that when the hotplug state machine does a roll-back | |
7586 | * we clear balance_push. Ideally that would happen earlier... | |
7587 | */ | |
2558aacf PZ |
7588 | balance_push_set(cpu, false); |
7589 | ||
ba2591a5 PZ |
7590 | #ifdef CONFIG_SCHED_SMT |
7591 | /* | |
c5511d03 | 7592 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 7593 | */ |
c5511d03 PZI |
7594 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
7595 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 7596 | #endif |
40190a78 | 7597 | set_cpu_active(cpu, true); |
135fb3e1 | 7598 | |
40190a78 | 7599 | if (sched_smp_initialized) { |
135fb3e1 | 7600 | sched_domains_numa_masks_set(cpu); |
40190a78 | 7601 | cpuset_cpu_active(); |
e761b772 | 7602 | } |
7d976699 TG |
7603 | |
7604 | /* | |
7605 | * Put the rq online, if not already. This happens: | |
7606 | * | |
7607 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 7608 | * after all CPUs have been brought up. |
7d976699 TG |
7609 | * |
7610 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
7611 | * domains. | |
7612 | */ | |
8a8c69c3 | 7613 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
7614 | if (rq->rd) { |
7615 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7616 | set_rq_online(rq); | |
7617 | } | |
8a8c69c3 | 7618 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 7619 | |
40190a78 | 7620 | return 0; |
135fb3e1 TG |
7621 | } |
7622 | ||
40190a78 | 7623 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 7624 | { |
120455c5 PZ |
7625 | struct rq *rq = cpu_rq(cpu); |
7626 | struct rq_flags rf; | |
135fb3e1 TG |
7627 | int ret; |
7628 | ||
e0b257c3 AMB |
7629 | /* |
7630 | * Remove CPU from nohz.idle_cpus_mask to prevent participating in | |
7631 | * load balancing when not active | |
7632 | */ | |
7633 | nohz_balance_exit_idle(rq); | |
7634 | ||
40190a78 | 7635 | set_cpu_active(cpu, false); |
741ba80f PZ |
7636 | |
7637 | /* | |
7638 | * From this point forward, this CPU will refuse to run any task that | |
7639 | * is not: migrate_disable() or KTHREAD_IS_PER_CPU, and will actively | |
7640 | * push those tasks away until this gets cleared, see | |
7641 | * sched_cpu_dying(). | |
7642 | */ | |
975707f2 PZ |
7643 | balance_push_set(cpu, true); |
7644 | ||
b2454caa | 7645 | /* |
975707f2 PZ |
7646 | * We've cleared cpu_active_mask / set balance_push, wait for all |
7647 | * preempt-disabled and RCU users of this state to go away such that | |
7648 | * all new such users will observe it. | |
b2454caa | 7649 | * |
5ba2ffba PZ |
7650 | * Specifically, we rely on ttwu to no longer target this CPU, see |
7651 | * ttwu_queue_cond() and is_cpu_allowed(). | |
b2454caa | 7652 | * |
b2454caa PZ |
7653 | * Do sync before park smpboot threads to take care the rcu boost case. |
7654 | */ | |
309ba859 | 7655 | synchronize_rcu(); |
40190a78 | 7656 | |
120455c5 PZ |
7657 | rq_lock_irqsave(rq, &rf); |
7658 | if (rq->rd) { | |
7659 | update_rq_clock(rq); | |
7660 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
7661 | set_rq_offline(rq); | |
7662 | } | |
7663 | rq_unlock_irqrestore(rq, &rf); | |
7664 | ||
c5511d03 PZI |
7665 | #ifdef CONFIG_SCHED_SMT |
7666 | /* | |
7667 | * When going down, decrement the number of cores with SMT present. | |
7668 | */ | |
7669 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
7670 | static_branch_dec_cpuslocked(&sched_smt_present); | |
7671 | #endif | |
7672 | ||
40190a78 TG |
7673 | if (!sched_smp_initialized) |
7674 | return 0; | |
7675 | ||
7676 | ret = cpuset_cpu_inactive(cpu); | |
7677 | if (ret) { | |
2558aacf | 7678 | balance_push_set(cpu, false); |
40190a78 TG |
7679 | set_cpu_active(cpu, true); |
7680 | return ret; | |
135fb3e1 | 7681 | } |
40190a78 TG |
7682 | sched_domains_numa_masks_clear(cpu); |
7683 | return 0; | |
135fb3e1 TG |
7684 | } |
7685 | ||
94baf7a5 TG |
7686 | static void sched_rq_cpu_starting(unsigned int cpu) |
7687 | { | |
7688 | struct rq *rq = cpu_rq(cpu); | |
7689 | ||
7690 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
7691 | update_max_interval(); |
7692 | } | |
7693 | ||
135fb3e1 TG |
7694 | int sched_cpu_starting(unsigned int cpu) |
7695 | { | |
94baf7a5 | 7696 | sched_rq_cpu_starting(cpu); |
d84b3131 | 7697 | sched_tick_start(cpu); |
135fb3e1 | 7698 | return 0; |
e761b772 | 7699 | } |
e761b772 | 7700 | |
f2785ddb | 7701 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
7702 | |
7703 | /* | |
7704 | * Invoked immediately before the stopper thread is invoked to bring the | |
7705 | * CPU down completely. At this point all per CPU kthreads except the | |
7706 | * hotplug thread (current) and the stopper thread (inactive) have been | |
7707 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
7708 | * any of those which might be on the way out are gone. | |
7709 | * | |
7710 | * If after this point a bound task is being woken on this CPU then the | |
7711 | * responsible hotplug callback has failed to do it's job. | |
7712 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
7713 | */ | |
7714 | int sched_cpu_wait_empty(unsigned int cpu) | |
7715 | { | |
7716 | balance_hotplug_wait(); | |
7717 | return 0; | |
7718 | } | |
7719 | ||
7720 | /* | |
7721 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
7722 | * might have. Called from the CPU stopper task after ensuring that the | |
7723 | * stopper is the last running task on the CPU, so nr_active count is | |
7724 | * stable. We need to take the teardown thread which is calling this into | |
7725 | * account, so we hand in adjust = 1 to the load calculation. | |
7726 | * | |
7727 | * Also see the comment "Global load-average calculations". | |
7728 | */ | |
7729 | static void calc_load_migrate(struct rq *rq) | |
7730 | { | |
7731 | long delta = calc_load_fold_active(rq, 1); | |
7732 | ||
7733 | if (delta) | |
7734 | atomic_long_add(delta, &calc_load_tasks); | |
7735 | } | |
7736 | ||
36c6e17b VS |
7737 | static void dump_rq_tasks(struct rq *rq, const char *loglvl) |
7738 | { | |
7739 | struct task_struct *g, *p; | |
7740 | int cpu = cpu_of(rq); | |
7741 | ||
7742 | lockdep_assert_held(&rq->lock); | |
7743 | ||
7744 | printk("%sCPU%d enqueued tasks (%u total):\n", loglvl, cpu, rq->nr_running); | |
7745 | for_each_process_thread(g, p) { | |
7746 | if (task_cpu(p) != cpu) | |
7747 | continue; | |
7748 | ||
7749 | if (!task_on_rq_queued(p)) | |
7750 | continue; | |
7751 | ||
7752 | printk("%s\tpid: %d, name: %s\n", loglvl, p->pid, p->comm); | |
7753 | } | |
7754 | } | |
7755 | ||
f2785ddb TG |
7756 | int sched_cpu_dying(unsigned int cpu) |
7757 | { | |
7758 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 7759 | struct rq_flags rf; |
f2785ddb TG |
7760 | |
7761 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 7762 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
7763 | |
7764 | rq_lock_irqsave(rq, &rf); | |
36c6e17b VS |
7765 | if (rq->nr_running != 1 || rq_has_pinned_tasks(rq)) { |
7766 | WARN(true, "Dying CPU not properly vacated!"); | |
7767 | dump_rq_tasks(rq, KERN_WARNING); | |
7768 | } | |
8a8c69c3 PZ |
7769 | rq_unlock_irqrestore(rq, &rf); |
7770 | ||
22f667c9 PZ |
7771 | /* |
7772 | * Now that the CPU is offline, make sure we're welcome | |
7773 | * to new tasks once we come back up. | |
7774 | */ | |
7775 | balance_push_set(cpu, false); | |
7776 | ||
f2785ddb TG |
7777 | calc_load_migrate(rq); |
7778 | update_max_interval(); | |
e5ef27d0 | 7779 | hrtick_clear(rq); |
f2785ddb TG |
7780 | return 0; |
7781 | } | |
7782 | #endif | |
7783 | ||
1da177e4 LT |
7784 | void __init sched_init_smp(void) |
7785 | { | |
cb83b629 PZ |
7786 | sched_init_numa(); |
7787 | ||
6acce3ef PZ |
7788 | /* |
7789 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 7790 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 7791 | * happen. |
6acce3ef | 7792 | */ |
712555ee | 7793 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 7794 | sched_init_domains(cpu_active_mask); |
712555ee | 7795 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7796 | |
5c1e1767 | 7797 | /* Move init over to a non-isolated CPU */ |
edb93821 | 7798 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 7799 | BUG(); |
19978ca6 | 7800 | sched_init_granularity(); |
4212823f | 7801 | |
0e3900e6 | 7802 | init_sched_rt_class(); |
1baca4ce | 7803 | init_sched_dl_class(); |
1b568f0a | 7804 | |
e26fbffd | 7805 | sched_smp_initialized = true; |
1da177e4 | 7806 | } |
e26fbffd TG |
7807 | |
7808 | static int __init migration_init(void) | |
7809 | { | |
77a5352b | 7810 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 7811 | return 0; |
1da177e4 | 7812 | } |
e26fbffd TG |
7813 | early_initcall(migration_init); |
7814 | ||
1da177e4 LT |
7815 | #else |
7816 | void __init sched_init_smp(void) | |
7817 | { | |
19978ca6 | 7818 | sched_init_granularity(); |
1da177e4 LT |
7819 | } |
7820 | #endif /* CONFIG_SMP */ | |
7821 | ||
7822 | int in_sched_functions(unsigned long addr) | |
7823 | { | |
1da177e4 LT |
7824 | return in_lock_functions(addr) || |
7825 | (addr >= (unsigned long)__sched_text_start | |
7826 | && addr < (unsigned long)__sched_text_end); | |
7827 | } | |
7828 | ||
029632fb | 7829 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7830 | /* |
7831 | * Default task group. | |
7832 | * Every task in system belongs to this group at bootup. | |
7833 | */ | |
029632fb | 7834 | struct task_group root_task_group; |
35cf4e50 | 7835 | LIST_HEAD(task_groups); |
b0367629 WL |
7836 | |
7837 | /* Cacheline aligned slab cache for task_group */ | |
7838 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7839 | #endif |
6f505b16 | 7840 | |
e6252c3e | 7841 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 7842 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 7843 | |
1da177e4 LT |
7844 | void __init sched_init(void) |
7845 | { | |
a1dc0446 | 7846 | unsigned long ptr = 0; |
55627e3c | 7847 | int i; |
434d53b0 | 7848 | |
c3a340f7 SRV |
7849 | /* Make sure the linker didn't screw up */ |
7850 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
7851 | &fair_sched_class + 1 != &rt_sched_class || | |
7852 | &rt_sched_class + 1 != &dl_sched_class); | |
7853 | #ifdef CONFIG_SMP | |
7854 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
7855 | #endif | |
7856 | ||
5822a454 | 7857 | wait_bit_init(); |
9dcb8b68 | 7858 | |
434d53b0 | 7859 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 7860 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
7861 | #endif |
7862 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 7863 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 7864 | #endif |
a1dc0446 QC |
7865 | if (ptr) { |
7866 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
7867 | |
7868 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7869 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7870 | ptr += nr_cpu_ids * sizeof(void **); |
7871 | ||
07e06b01 | 7872 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7873 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7874 | |
b1d1779e WY |
7875 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
7876 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 7877 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7878 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7879 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7880 | ptr += nr_cpu_ids * sizeof(void **); |
7881 | ||
07e06b01 | 7882 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7883 | ptr += nr_cpu_ids * sizeof(void **); |
7884 | ||
6d6bc0ad | 7885 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7886 | } |
df7c8e84 | 7887 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7888 | for_each_possible_cpu(i) { |
7889 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7890 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
7891 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
7892 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7893 | } |
b74e6278 | 7894 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7895 | |
d1ccc66d IM |
7896 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
7897 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 7898 | |
57d885fe GH |
7899 | #ifdef CONFIG_SMP |
7900 | init_defrootdomain(); | |
7901 | #endif | |
7902 | ||
d0b27fa7 | 7903 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7904 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7905 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7906 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7907 | |
7c941438 | 7908 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7909 | task_group_cache = KMEM_CACHE(task_group, 0); |
7910 | ||
07e06b01 YZ |
7911 | list_add(&root_task_group.list, &task_groups); |
7912 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7913 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7914 | autogroup_init(&init_task); |
7c941438 | 7915 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7916 | |
0a945022 | 7917 | for_each_possible_cpu(i) { |
70b97a7f | 7918 | struct rq *rq; |
1da177e4 LT |
7919 | |
7920 | rq = cpu_rq(i); | |
05fa785c | 7921 | raw_spin_lock_init(&rq->lock); |
7897986b | 7922 | rq->nr_running = 0; |
dce48a84 TG |
7923 | rq->calc_load_active = 0; |
7924 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7925 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7926 | init_rt_rq(&rq->rt); |
7927 | init_dl_rq(&rq->dl); | |
dd41f596 | 7928 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 7929 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 7930 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 7931 | /* |
d1ccc66d | 7932 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
7933 | * |
7934 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
7935 | * gets 100% of the CPU resources in the system. This overall |
7936 | * system CPU resource is divided among the tasks of | |
07e06b01 | 7937 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7938 | * based on each entity's (task or task-group's) weight |
7939 | * (se->load.weight). | |
7940 | * | |
07e06b01 | 7941 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 7942 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 7943 | * then A0's share of the CPU resource is: |
354d60c2 | 7944 | * |
0d905bca | 7945 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7946 | * |
07e06b01 YZ |
7947 | * We achieve this by letting root_task_group's tasks sit |
7948 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7949 | */ |
07e06b01 | 7950 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7951 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7952 | ||
7953 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7954 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7955 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7956 | #endif |
1da177e4 | 7957 | #ifdef CONFIG_SMP |
41c7ce9a | 7958 | rq->sd = NULL; |
57d885fe | 7959 | rq->rd = NULL; |
ca6d75e6 | 7960 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7961 | rq->balance_callback = NULL; |
1da177e4 | 7962 | rq->active_balance = 0; |
dd41f596 | 7963 | rq->next_balance = jiffies; |
1da177e4 | 7964 | rq->push_cpu = 0; |
0a2966b4 | 7965 | rq->cpu = i; |
1f11eb6a | 7966 | rq->online = 0; |
eae0c9df MG |
7967 | rq->idle_stamp = 0; |
7968 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7969 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7970 | |
7971 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7972 | ||
dc938520 | 7973 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7974 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 7975 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 7976 | atomic_set(&rq->nohz_flags, 0); |
90b5363a | 7977 | |
545b8c8d | 7978 | INIT_CSD(&rq->nohz_csd, nohz_csd_func, rq); |
83cd4fe2 | 7979 | #endif |
f2469a1f TG |
7980 | #ifdef CONFIG_HOTPLUG_CPU |
7981 | rcuwait_init(&rq->hotplug_wait); | |
83cd4fe2 | 7982 | #endif |
9fd81dd5 | 7983 | #endif /* CONFIG_SMP */ |
77a021be | 7984 | hrtick_rq_init(rq); |
1da177e4 | 7985 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7986 | } |
7987 | ||
9059393e | 7988 | set_load_weight(&init_task, false); |
b50f60ce | 7989 | |
1da177e4 LT |
7990 | /* |
7991 | * The boot idle thread does lazy MMU switching as well: | |
7992 | */ | |
f1f10076 | 7993 | mmgrab(&init_mm); |
1da177e4 LT |
7994 | enter_lazy_tlb(&init_mm, current); |
7995 | ||
7996 | /* | |
7997 | * Make us the idle thread. Technically, schedule() should not be | |
7998 | * called from this thread, however somewhere below it might be, | |
7999 | * but because we are the idle thread, we just pick up running again | |
8000 | * when this runqueue becomes "idle". | |
8001 | */ | |
8002 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8003 | |
8004 | calc_load_update = jiffies + LOAD_FREQ; | |
8005 | ||
bf4d83f6 | 8006 | #ifdef CONFIG_SMP |
29d5e047 | 8007 | idle_thread_set_boot_cpu(); |
029632fb PZ |
8008 | #endif |
8009 | init_sched_fair_class(); | |
6a7b3dc3 | 8010 | |
4698f88c JP |
8011 | init_schedstats(); |
8012 | ||
eb414681 JW |
8013 | psi_init(); |
8014 | ||
69842cba PB |
8015 | init_uclamp(); |
8016 | ||
6892b75e | 8017 | scheduler_running = 1; |
1da177e4 LT |
8018 | } |
8019 | ||
d902db1e | 8020 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
8021 | static inline int preempt_count_equals(int preempt_offset) |
8022 | { | |
da7142e2 | 8023 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 8024 | |
4ba8216c | 8025 | return (nested == preempt_offset); |
e4aafea2 FW |
8026 | } |
8027 | ||
d894837f | 8028 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8029 | { |
8eb23b9f PZ |
8030 | /* |
8031 | * Blocking primitives will set (and therefore destroy) current->state, | |
8032 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
8033 | * otherwise we will destroy state. | |
8034 | */ | |
00845eb9 | 8035 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
8036 | "do not call blocking ops when !TASK_RUNNING; " |
8037 | "state=%lx set at [<%p>] %pS\n", | |
8038 | current->state, | |
8039 | (void *)current->task_state_change, | |
00845eb9 | 8040 | (void *)current->task_state_change); |
8eb23b9f | 8041 | |
3427445a PZ |
8042 | ___might_sleep(file, line, preempt_offset); |
8043 | } | |
8044 | EXPORT_SYMBOL(__might_sleep); | |
8045 | ||
8046 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 8047 | { |
d1ccc66d IM |
8048 | /* Ratelimiting timestamp: */ |
8049 | static unsigned long prev_jiffy; | |
8050 | ||
d1c6d149 | 8051 | unsigned long preempt_disable_ip; |
1da177e4 | 8052 | |
d1ccc66d IM |
8053 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
8054 | rcu_sleep_check(); | |
8055 | ||
db273be2 | 8056 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 8057 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
8058 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
8059 | oops_in_progress) | |
aef745fc | 8060 | return; |
1c3c5eab | 8061 | |
aef745fc IM |
8062 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
8063 | return; | |
8064 | prev_jiffy = jiffies; | |
8065 | ||
d1ccc66d | 8066 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
8067 | preempt_disable_ip = get_preempt_disable_ip(current); |
8068 | ||
3df0fc5b PZ |
8069 | printk(KERN_ERR |
8070 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8071 | file, line); | |
8072 | printk(KERN_ERR | |
312364f3 DV |
8073 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
8074 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 8075 | current->pid, current->comm); |
aef745fc | 8076 | |
a8b686b3 ES |
8077 | if (task_stack_end_corrupted(current)) |
8078 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
8079 | ||
aef745fc IM |
8080 | debug_show_held_locks(current); |
8081 | if (irqs_disabled()) | |
8082 | print_irqtrace_events(current); | |
d1c6d149 VN |
8083 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
8084 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 8085 | pr_err("Preemption disabled at:"); |
2062a4e8 | 8086 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 8087 | } |
aef745fc | 8088 | dump_stack(); |
f0b22e39 | 8089 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 8090 | } |
3427445a | 8091 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
8092 | |
8093 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
8094 | { | |
8095 | static unsigned long prev_jiffy; | |
8096 | ||
8097 | if (irqs_disabled()) | |
8098 | return; | |
8099 | ||
8100 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
8101 | return; | |
8102 | ||
8103 | if (preempt_count() > preempt_offset) | |
8104 | return; | |
8105 | ||
8106 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8107 | return; | |
8108 | prev_jiffy = jiffies; | |
8109 | ||
8110 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
8111 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8112 | in_atomic(), irqs_disabled(), | |
8113 | current->pid, current->comm); | |
8114 | ||
8115 | debug_show_held_locks(current); | |
8116 | dump_stack(); | |
8117 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
8118 | } | |
8119 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
8120 | |
8121 | #ifdef CONFIG_SMP | |
8122 | void __cant_migrate(const char *file, int line) | |
8123 | { | |
8124 | static unsigned long prev_jiffy; | |
8125 | ||
8126 | if (irqs_disabled()) | |
8127 | return; | |
8128 | ||
8129 | if (is_migration_disabled(current)) | |
8130 | return; | |
8131 | ||
8132 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
8133 | return; | |
8134 | ||
8135 | if (preempt_count() > 0) | |
8136 | return; | |
8137 | ||
8138 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8139 | return; | |
8140 | prev_jiffy = jiffies; | |
8141 | ||
8142 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
8143 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
8144 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
8145 | current->pid, current->comm); | |
8146 | ||
8147 | debug_show_held_locks(current); | |
8148 | dump_stack(); | |
8149 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
8150 | } | |
8151 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
8152 | #endif | |
1da177e4 LT |
8153 | #endif |
8154 | ||
8155 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 8156 | void normalize_rt_tasks(void) |
3a5e4dc1 | 8157 | { |
dbc7f069 | 8158 | struct task_struct *g, *p; |
d50dde5a DF |
8159 | struct sched_attr attr = { |
8160 | .sched_policy = SCHED_NORMAL, | |
8161 | }; | |
1da177e4 | 8162 | |
3472eaa1 | 8163 | read_lock(&tasklist_lock); |
5d07f420 | 8164 | for_each_process_thread(g, p) { |
178be793 IM |
8165 | /* |
8166 | * Only normalize user tasks: | |
8167 | */ | |
3472eaa1 | 8168 | if (p->flags & PF_KTHREAD) |
178be793 IM |
8169 | continue; |
8170 | ||
4fa8d299 JP |
8171 | p->se.exec_start = 0; |
8172 | schedstat_set(p->se.statistics.wait_start, 0); | |
8173 | schedstat_set(p->se.statistics.sleep_start, 0); | |
8174 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 8175 | |
aab03e05 | 8176 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
8177 | /* |
8178 | * Renice negative nice level userspace | |
8179 | * tasks back to 0: | |
8180 | */ | |
3472eaa1 | 8181 | if (task_nice(p) < 0) |
dd41f596 | 8182 | set_user_nice(p, 0); |
1da177e4 | 8183 | continue; |
dd41f596 | 8184 | } |
1da177e4 | 8185 | |
dbc7f069 | 8186 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 8187 | } |
3472eaa1 | 8188 | read_unlock(&tasklist_lock); |
1da177e4 LT |
8189 | } |
8190 | ||
8191 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8192 | |
67fc4e0c | 8193 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8194 | /* |
67fc4e0c | 8195 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8196 | * |
8197 | * They can only be called when the whole system has been | |
8198 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8199 | * activity can take place. Using them for anything else would | |
8200 | * be a serious bug, and as a result, they aren't even visible | |
8201 | * under any other configuration. | |
8202 | */ | |
8203 | ||
8204 | /** | |
d1ccc66d | 8205 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
8206 | * @cpu: the processor in question. |
8207 | * | |
8208 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
8209 | * |
8210 | * Return: The current task for @cpu. | |
1df5c10a | 8211 | */ |
36c8b586 | 8212 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8213 | { |
8214 | return cpu_curr(cpu); | |
8215 | } | |
8216 | ||
67fc4e0c JW |
8217 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8218 | ||
8219 | #ifdef CONFIG_IA64 | |
1df5c10a | 8220 | /** |
5feeb783 | 8221 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
8222 | * @cpu: the processor in question. |
8223 | * @p: the task pointer to set. | |
8224 | * | |
8225 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 8226 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 8227 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
8228 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8229 | * and caller must save the original value of the current task (see | |
8230 | * curr_task() above) and restore that value before reenabling interrupts and | |
8231 | * re-starting the system. | |
8232 | * | |
8233 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8234 | */ | |
a458ae2e | 8235 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8236 | { |
8237 | cpu_curr(cpu) = p; | |
8238 | } | |
8239 | ||
8240 | #endif | |
29f59db3 | 8241 | |
7c941438 | 8242 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
8243 | /* task_group_lock serializes the addition/removal of task groups */ |
8244 | static DEFINE_SPINLOCK(task_group_lock); | |
8245 | ||
2480c093 PB |
8246 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
8247 | struct task_group *parent) | |
8248 | { | |
8249 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 8250 | enum uclamp_id clamp_id; |
2480c093 PB |
8251 | |
8252 | for_each_clamp_id(clamp_id) { | |
8253 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
8254 | uclamp_none(clamp_id), false); | |
0b60ba2d | 8255 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
8256 | } |
8257 | #endif | |
8258 | } | |
8259 | ||
2f5177f0 | 8260 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
8261 | { |
8262 | free_fair_sched_group(tg); | |
8263 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8264 | autogroup_free(tg); |
b0367629 | 8265 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
8266 | } |
8267 | ||
8268 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8269 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8270 | { |
8271 | struct task_group *tg; | |
bccbe08a | 8272 | |
b0367629 | 8273 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
8274 | if (!tg) |
8275 | return ERR_PTR(-ENOMEM); | |
8276 | ||
ec7dc8ac | 8277 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8278 | goto err; |
8279 | ||
ec7dc8ac | 8280 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8281 | goto err; |
8282 | ||
2480c093 PB |
8283 | alloc_uclamp_sched_group(tg, parent); |
8284 | ||
ace783b9 LZ |
8285 | return tg; |
8286 | ||
8287 | err: | |
2f5177f0 | 8288 | sched_free_group(tg); |
ace783b9 LZ |
8289 | return ERR_PTR(-ENOMEM); |
8290 | } | |
8291 | ||
8292 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
8293 | { | |
8294 | unsigned long flags; | |
8295 | ||
8ed36996 | 8296 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8297 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 8298 | |
d1ccc66d IM |
8299 | /* Root should already exist: */ |
8300 | WARN_ON(!parent); | |
f473aa5e PZ |
8301 | |
8302 | tg->parent = parent; | |
f473aa5e | 8303 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8304 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8305 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
8306 | |
8307 | online_fair_sched_group(tg); | |
29f59db3 SV |
8308 | } |
8309 | ||
9b5b7751 | 8310 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 8311 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8312 | { |
d1ccc66d | 8313 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 8314 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8315 | } |
8316 | ||
4cf86d77 | 8317 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 8318 | { |
d1ccc66d | 8319 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 8320 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
8321 | } |
8322 | ||
8323 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 8324 | { |
8ed36996 | 8325 | unsigned long flags; |
29f59db3 | 8326 | |
d1ccc66d | 8327 | /* End participation in shares distribution: */ |
6fe1f348 | 8328 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
8329 | |
8330 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8331 | list_del_rcu(&tg->list); |
f473aa5e | 8332 | list_del_rcu(&tg->siblings); |
8ed36996 | 8333 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
8334 | } |
8335 | ||
ea86cb4b | 8336 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 8337 | { |
8323f26c | 8338 | struct task_group *tg; |
29f59db3 | 8339 | |
f7b8a47d KT |
8340 | /* |
8341 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
8342 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
8343 | * to prevent lockdep warnings. | |
8344 | */ | |
8345 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
8346 | struct task_group, css); |
8347 | tg = autogroup_task_group(tsk, tg); | |
8348 | tsk->sched_task_group = tg; | |
8349 | ||
810b3817 | 8350 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
8351 | if (tsk->sched_class->task_change_group) |
8352 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 8353 | else |
810b3817 | 8354 | #endif |
b2b5ce02 | 8355 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
8356 | } |
8357 | ||
8358 | /* | |
8359 | * Change task's runqueue when it moves between groups. | |
8360 | * | |
8361 | * The caller of this function should have put the task in its new group by | |
8362 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
8363 | * its new group. | |
8364 | */ | |
8365 | void sched_move_task(struct task_struct *tsk) | |
8366 | { | |
7a57f32a PZ |
8367 | int queued, running, queue_flags = |
8368 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
8369 | struct rq_flags rf; |
8370 | struct rq *rq; | |
8371 | ||
8372 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 8373 | update_rq_clock(rq); |
ea86cb4b VG |
8374 | |
8375 | running = task_current(rq, tsk); | |
8376 | queued = task_on_rq_queued(tsk); | |
8377 | ||
8378 | if (queued) | |
7a57f32a | 8379 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 8380 | if (running) |
ea86cb4b VG |
8381 | put_prev_task(rq, tsk); |
8382 | ||
8383 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 8384 | |
da0c1e65 | 8385 | if (queued) |
7a57f32a | 8386 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 8387 | if (running) { |
03b7fad1 | 8388 | set_next_task(rq, tsk); |
2a4b03ff VG |
8389 | /* |
8390 | * After changing group, the running task may have joined a | |
8391 | * throttled one but it's still the running task. Trigger a | |
8392 | * resched to make sure that task can still run. | |
8393 | */ | |
8394 | resched_curr(rq); | |
8395 | } | |
29f59db3 | 8396 | |
eb580751 | 8397 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 8398 | } |
68318b8e | 8399 | |
a7c6d554 | 8400 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 8401 | { |
a7c6d554 | 8402 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
8403 | } |
8404 | ||
eb95419b TH |
8405 | static struct cgroup_subsys_state * |
8406 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 8407 | { |
eb95419b TH |
8408 | struct task_group *parent = css_tg(parent_css); |
8409 | struct task_group *tg; | |
68318b8e | 8410 | |
eb95419b | 8411 | if (!parent) { |
68318b8e | 8412 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8413 | return &root_task_group.css; |
68318b8e SV |
8414 | } |
8415 | ||
ec7dc8ac | 8416 | tg = sched_create_group(parent); |
68318b8e SV |
8417 | if (IS_ERR(tg)) |
8418 | return ERR_PTR(-ENOMEM); | |
8419 | ||
68318b8e SV |
8420 | return &tg->css; |
8421 | } | |
8422 | ||
96b77745 KK |
8423 | /* Expose task group only after completing cgroup initialization */ |
8424 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
8425 | { | |
8426 | struct task_group *tg = css_tg(css); | |
8427 | struct task_group *parent = css_tg(css->parent); | |
8428 | ||
8429 | if (parent) | |
8430 | sched_online_group(tg, parent); | |
7226017a QY |
8431 | |
8432 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8433 | /* Propagate the effective uclamp value for the new group */ | |
8434 | cpu_util_update_eff(css); | |
8435 | #endif | |
8436 | ||
96b77745 KK |
8437 | return 0; |
8438 | } | |
8439 | ||
2f5177f0 | 8440 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 8441 | { |
eb95419b | 8442 | struct task_group *tg = css_tg(css); |
ace783b9 | 8443 | |
2f5177f0 | 8444 | sched_offline_group(tg); |
ace783b9 LZ |
8445 | } |
8446 | ||
eb95419b | 8447 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 8448 | { |
eb95419b | 8449 | struct task_group *tg = css_tg(css); |
68318b8e | 8450 | |
2f5177f0 PZ |
8451 | /* |
8452 | * Relies on the RCU grace period between css_released() and this. | |
8453 | */ | |
8454 | sched_free_group(tg); | |
ace783b9 LZ |
8455 | } |
8456 | ||
ea86cb4b VG |
8457 | /* |
8458 | * This is called before wake_up_new_task(), therefore we really only | |
8459 | * have to set its group bits, all the other stuff does not apply. | |
8460 | */ | |
b53202e6 | 8461 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 8462 | { |
ea86cb4b VG |
8463 | struct rq_flags rf; |
8464 | struct rq *rq; | |
8465 | ||
8466 | rq = task_rq_lock(task, &rf); | |
8467 | ||
80f5c1b8 | 8468 | update_rq_clock(rq); |
ea86cb4b VG |
8469 | sched_change_group(task, TASK_SET_GROUP); |
8470 | ||
8471 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
8472 | } |
8473 | ||
1f7dd3e5 | 8474 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 8475 | { |
bb9d97b6 | 8476 | struct task_struct *task; |
1f7dd3e5 | 8477 | struct cgroup_subsys_state *css; |
7dc603c9 | 8478 | int ret = 0; |
bb9d97b6 | 8479 | |
1f7dd3e5 | 8480 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 8481 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 8482 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 8483 | return -EINVAL; |
b68aa230 | 8484 | #endif |
7dc603c9 | 8485 | /* |
b19a888c | 8486 | * Serialize against wake_up_new_task() such that if it's |
7dc603c9 PZ |
8487 | * running, we're sure to observe its full state. |
8488 | */ | |
8489 | raw_spin_lock_irq(&task->pi_lock); | |
8490 | /* | |
8491 | * Avoid calling sched_move_task() before wake_up_new_task() | |
8492 | * has happened. This would lead to problems with PELT, due to | |
8493 | * move wanting to detach+attach while we're not attached yet. | |
8494 | */ | |
8495 | if (task->state == TASK_NEW) | |
8496 | ret = -EINVAL; | |
8497 | raw_spin_unlock_irq(&task->pi_lock); | |
8498 | ||
8499 | if (ret) | |
8500 | break; | |
bb9d97b6 | 8501 | } |
7dc603c9 | 8502 | return ret; |
be367d09 | 8503 | } |
68318b8e | 8504 | |
1f7dd3e5 | 8505 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 8506 | { |
bb9d97b6 | 8507 | struct task_struct *task; |
1f7dd3e5 | 8508 | struct cgroup_subsys_state *css; |
bb9d97b6 | 8509 | |
1f7dd3e5 | 8510 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 8511 | sched_move_task(task); |
68318b8e SV |
8512 | } |
8513 | ||
2480c093 | 8514 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
8515 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
8516 | { | |
8517 | struct cgroup_subsys_state *top_css = css; | |
8518 | struct uclamp_se *uc_parent = NULL; | |
8519 | struct uclamp_se *uc_se = NULL; | |
8520 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 8521 | enum uclamp_id clamp_id; |
0b60ba2d PB |
8522 | unsigned int clamps; |
8523 | ||
8524 | css_for_each_descendant_pre(css, top_css) { | |
8525 | uc_parent = css_tg(css)->parent | |
8526 | ? css_tg(css)->parent->uclamp : NULL; | |
8527 | ||
8528 | for_each_clamp_id(clamp_id) { | |
8529 | /* Assume effective clamps matches requested clamps */ | |
8530 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
8531 | /* Cap effective clamps with parent's effective clamps */ | |
8532 | if (uc_parent && | |
8533 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
8534 | eff[clamp_id] = uc_parent[clamp_id].value; | |
8535 | } | |
8536 | } | |
8537 | /* Ensure protection is always capped by limit */ | |
8538 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
8539 | ||
8540 | /* Propagate most restrictive effective clamps */ | |
8541 | clamps = 0x0; | |
8542 | uc_se = css_tg(css)->uclamp; | |
8543 | for_each_clamp_id(clamp_id) { | |
8544 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
8545 | continue; | |
8546 | uc_se[clamp_id].value = eff[clamp_id]; | |
8547 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
8548 | clamps |= (0x1 << clamp_id); | |
8549 | } | |
babbe170 | 8550 | if (!clamps) { |
0b60ba2d | 8551 | css = css_rightmost_descendant(css); |
babbe170 PB |
8552 | continue; |
8553 | } | |
8554 | ||
8555 | /* Immediately update descendants RUNNABLE tasks */ | |
8556 | uclamp_update_active_tasks(css, clamps); | |
0b60ba2d PB |
8557 | } |
8558 | } | |
2480c093 PB |
8559 | |
8560 | /* | |
8561 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
8562 | * C expression. Since there is no way to convert a macro argument (N) into a | |
8563 | * character constant, use two levels of macros. | |
8564 | */ | |
8565 | #define _POW10(exp) ((unsigned int)1e##exp) | |
8566 | #define POW10(exp) _POW10(exp) | |
8567 | ||
8568 | struct uclamp_request { | |
8569 | #define UCLAMP_PERCENT_SHIFT 2 | |
8570 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
8571 | s64 percent; | |
8572 | u64 util; | |
8573 | int ret; | |
8574 | }; | |
8575 | ||
8576 | static inline struct uclamp_request | |
8577 | capacity_from_percent(char *buf) | |
8578 | { | |
8579 | struct uclamp_request req = { | |
8580 | .percent = UCLAMP_PERCENT_SCALE, | |
8581 | .util = SCHED_CAPACITY_SCALE, | |
8582 | .ret = 0, | |
8583 | }; | |
8584 | ||
8585 | buf = strim(buf); | |
8586 | if (strcmp(buf, "max")) { | |
8587 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
8588 | &req.percent); | |
8589 | if (req.ret) | |
8590 | return req; | |
b562d140 | 8591 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
8592 | req.ret = -ERANGE; |
8593 | return req; | |
8594 | } | |
8595 | ||
8596 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
8597 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
8598 | } | |
8599 | ||
8600 | return req; | |
8601 | } | |
8602 | ||
8603 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
8604 | size_t nbytes, loff_t off, | |
8605 | enum uclamp_id clamp_id) | |
8606 | { | |
8607 | struct uclamp_request req; | |
8608 | struct task_group *tg; | |
8609 | ||
8610 | req = capacity_from_percent(buf); | |
8611 | if (req.ret) | |
8612 | return req.ret; | |
8613 | ||
46609ce2 QY |
8614 | static_branch_enable(&sched_uclamp_used); |
8615 | ||
2480c093 PB |
8616 | mutex_lock(&uclamp_mutex); |
8617 | rcu_read_lock(); | |
8618 | ||
8619 | tg = css_tg(of_css(of)); | |
8620 | if (tg->uclamp_req[clamp_id].value != req.util) | |
8621 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
8622 | ||
8623 | /* | |
8624 | * Because of not recoverable conversion rounding we keep track of the | |
8625 | * exact requested value | |
8626 | */ | |
8627 | tg->uclamp_pct[clamp_id] = req.percent; | |
8628 | ||
0b60ba2d PB |
8629 | /* Update effective clamps to track the most restrictive value */ |
8630 | cpu_util_update_eff(of_css(of)); | |
8631 | ||
2480c093 PB |
8632 | rcu_read_unlock(); |
8633 | mutex_unlock(&uclamp_mutex); | |
8634 | ||
8635 | return nbytes; | |
8636 | } | |
8637 | ||
8638 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
8639 | char *buf, size_t nbytes, | |
8640 | loff_t off) | |
8641 | { | |
8642 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
8643 | } | |
8644 | ||
8645 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
8646 | char *buf, size_t nbytes, | |
8647 | loff_t off) | |
8648 | { | |
8649 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
8650 | } | |
8651 | ||
8652 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
8653 | enum uclamp_id clamp_id) | |
8654 | { | |
8655 | struct task_group *tg; | |
8656 | u64 util_clamp; | |
8657 | u64 percent; | |
8658 | u32 rem; | |
8659 | ||
8660 | rcu_read_lock(); | |
8661 | tg = css_tg(seq_css(sf)); | |
8662 | util_clamp = tg->uclamp_req[clamp_id].value; | |
8663 | rcu_read_unlock(); | |
8664 | ||
8665 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
8666 | seq_puts(sf, "max\n"); | |
8667 | return; | |
8668 | } | |
8669 | ||
8670 | percent = tg->uclamp_pct[clamp_id]; | |
8671 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
8672 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
8673 | } | |
8674 | ||
8675 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
8676 | { | |
8677 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
8678 | return 0; | |
8679 | } | |
8680 | ||
8681 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
8682 | { | |
8683 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
8684 | return 0; | |
8685 | } | |
8686 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
8687 | ||
052f1dc7 | 8688 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
8689 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
8690 | struct cftype *cftype, u64 shareval) | |
68318b8e | 8691 | { |
5b61d50a KK |
8692 | if (shareval > scale_load_down(ULONG_MAX)) |
8693 | shareval = MAX_SHARES; | |
182446d0 | 8694 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
8695 | } |
8696 | ||
182446d0 TH |
8697 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
8698 | struct cftype *cft) | |
68318b8e | 8699 | { |
182446d0 | 8700 | struct task_group *tg = css_tg(css); |
68318b8e | 8701 | |
c8b28116 | 8702 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8703 | } |
ab84d31e PT |
8704 | |
8705 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
8706 | static DEFINE_MUTEX(cfs_constraints_mutex); |
8707 | ||
ab84d31e | 8708 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 8709 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
8710 | /* More than 203 days if BW_SHIFT equals 20. */ |
8711 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 8712 | |
a790de99 PT |
8713 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
8714 | ||
ab84d31e PT |
8715 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
8716 | { | |
56f570e5 | 8717 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 8718 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
8719 | |
8720 | if (tg == &root_task_group) | |
8721 | return -EINVAL; | |
8722 | ||
8723 | /* | |
8724 | * Ensure we have at some amount of bandwidth every period. This is | |
8725 | * to prevent reaching a state of large arrears when throttled via | |
8726 | * entity_tick() resulting in prolonged exit starvation. | |
8727 | */ | |
8728 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
8729 | return -EINVAL; | |
8730 | ||
8731 | /* | |
8732 | * Likewise, bound things on the otherside by preventing insane quota | |
8733 | * periods. This also allows us to normalize in computing quota | |
8734 | * feasibility. | |
8735 | */ | |
8736 | if (period > max_cfs_quota_period) | |
8737 | return -EINVAL; | |
8738 | ||
d505b8af HC |
8739 | /* |
8740 | * Bound quota to defend quota against overflow during bandwidth shift. | |
8741 | */ | |
8742 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
8743 | return -EINVAL; | |
8744 | ||
0e59bdae KT |
8745 | /* |
8746 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
8747 | * unthrottle_offline_cfs_rqs(). | |
8748 | */ | |
8749 | get_online_cpus(); | |
a790de99 PT |
8750 | mutex_lock(&cfs_constraints_mutex); |
8751 | ret = __cfs_schedulable(tg, period, quota); | |
8752 | if (ret) | |
8753 | goto out_unlock; | |
8754 | ||
58088ad0 | 8755 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 8756 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
8757 | /* |
8758 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
8759 | * before making related changes, and on->off must occur afterwards | |
8760 | */ | |
8761 | if (runtime_enabled && !runtime_was_enabled) | |
8762 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
8763 | raw_spin_lock_irq(&cfs_b->lock); |
8764 | cfs_b->period = ns_to_ktime(period); | |
8765 | cfs_b->quota = quota; | |
58088ad0 | 8766 | |
a9cf55b2 | 8767 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
8768 | |
8769 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
8770 | if (runtime_enabled) |
8771 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 8772 | |
ab84d31e PT |
8773 | raw_spin_unlock_irq(&cfs_b->lock); |
8774 | ||
0e59bdae | 8775 | for_each_online_cpu(i) { |
ab84d31e | 8776 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 8777 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 8778 | struct rq_flags rf; |
ab84d31e | 8779 | |
8a8c69c3 | 8780 | rq_lock_irq(rq, &rf); |
58088ad0 | 8781 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 8782 | cfs_rq->runtime_remaining = 0; |
671fd9da | 8783 | |
029632fb | 8784 | if (cfs_rq->throttled) |
671fd9da | 8785 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 8786 | rq_unlock_irq(rq, &rf); |
ab84d31e | 8787 | } |
1ee14e6c BS |
8788 | if (runtime_was_enabled && !runtime_enabled) |
8789 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
8790 | out_unlock: |
8791 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 8792 | put_online_cpus(); |
ab84d31e | 8793 | |
a790de99 | 8794 | return ret; |
ab84d31e PT |
8795 | } |
8796 | ||
b1546edc | 8797 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e PT |
8798 | { |
8799 | u64 quota, period; | |
8800 | ||
029632fb | 8801 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8802 | if (cfs_quota_us < 0) |
8803 | quota = RUNTIME_INF; | |
1a8b4540 | 8804 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 8805 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
8806 | else |
8807 | return -EINVAL; | |
ab84d31e PT |
8808 | |
8809 | return tg_set_cfs_bandwidth(tg, period, quota); | |
8810 | } | |
8811 | ||
b1546edc | 8812 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
8813 | { |
8814 | u64 quota_us; | |
8815 | ||
029632fb | 8816 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
8817 | return -1; |
8818 | ||
029632fb | 8819 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
8820 | do_div(quota_us, NSEC_PER_USEC); |
8821 | ||
8822 | return quota_us; | |
8823 | } | |
8824 | ||
b1546edc | 8825 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e PT |
8826 | { |
8827 | u64 quota, period; | |
8828 | ||
1a8b4540 KK |
8829 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
8830 | return -EINVAL; | |
8831 | ||
ab84d31e | 8832 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 8833 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8834 | |
ab84d31e PT |
8835 | return tg_set_cfs_bandwidth(tg, period, quota); |
8836 | } | |
8837 | ||
b1546edc | 8838 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
8839 | { |
8840 | u64 cfs_period_us; | |
8841 | ||
029632fb | 8842 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8843 | do_div(cfs_period_us, NSEC_PER_USEC); |
8844 | ||
8845 | return cfs_period_us; | |
8846 | } | |
8847 | ||
182446d0 TH |
8848 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8849 | struct cftype *cft) | |
ab84d31e | 8850 | { |
182446d0 | 8851 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8852 | } |
8853 | ||
182446d0 TH |
8854 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8855 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8856 | { |
182446d0 | 8857 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8858 | } |
8859 | ||
182446d0 TH |
8860 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8861 | struct cftype *cft) | |
ab84d31e | 8862 | { |
182446d0 | 8863 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8864 | } |
8865 | ||
182446d0 TH |
8866 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8867 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8868 | { |
182446d0 | 8869 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8870 | } |
8871 | ||
a790de99 PT |
8872 | struct cfs_schedulable_data { |
8873 | struct task_group *tg; | |
8874 | u64 period, quota; | |
8875 | }; | |
8876 | ||
8877 | /* | |
8878 | * normalize group quota/period to be quota/max_period | |
8879 | * note: units are usecs | |
8880 | */ | |
8881 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8882 | struct cfs_schedulable_data *d) | |
8883 | { | |
8884 | u64 quota, period; | |
8885 | ||
8886 | if (tg == d->tg) { | |
8887 | period = d->period; | |
8888 | quota = d->quota; | |
8889 | } else { | |
8890 | period = tg_get_cfs_period(tg); | |
8891 | quota = tg_get_cfs_quota(tg); | |
8892 | } | |
8893 | ||
8894 | /* note: these should typically be equivalent */ | |
8895 | if (quota == RUNTIME_INF || quota == -1) | |
8896 | return RUNTIME_INF; | |
8897 | ||
8898 | return to_ratio(period, quota); | |
8899 | } | |
8900 | ||
8901 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8902 | { | |
8903 | struct cfs_schedulable_data *d = data; | |
029632fb | 8904 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8905 | s64 quota = 0, parent_quota = -1; |
8906 | ||
8907 | if (!tg->parent) { | |
8908 | quota = RUNTIME_INF; | |
8909 | } else { | |
029632fb | 8910 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8911 | |
8912 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8913 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8914 | |
8915 | /* | |
c53593e5 TH |
8916 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
8917 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 8918 | * limit is set: |
a790de99 | 8919 | */ |
c53593e5 TH |
8920 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
8921 | quota = min(quota, parent_quota); | |
8922 | } else { | |
8923 | if (quota == RUNTIME_INF) | |
8924 | quota = parent_quota; | |
8925 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8926 | return -EINVAL; | |
8927 | } | |
a790de99 | 8928 | } |
9c58c79a | 8929 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8930 | |
8931 | return 0; | |
8932 | } | |
8933 | ||
8934 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8935 | { | |
8277434e | 8936 | int ret; |
a790de99 PT |
8937 | struct cfs_schedulable_data data = { |
8938 | .tg = tg, | |
8939 | .period = period, | |
8940 | .quota = quota, | |
8941 | }; | |
8942 | ||
8943 | if (quota != RUNTIME_INF) { | |
8944 | do_div(data.period, NSEC_PER_USEC); | |
8945 | do_div(data.quota, NSEC_PER_USEC); | |
8946 | } | |
8947 | ||
8277434e PT |
8948 | rcu_read_lock(); |
8949 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8950 | rcu_read_unlock(); | |
8951 | ||
8952 | return ret; | |
a790de99 | 8953 | } |
e8da1b18 | 8954 | |
a1f7164c | 8955 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 8956 | { |
2da8ca82 | 8957 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8958 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8959 | |
44ffc75b TH |
8960 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8961 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8962 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 8963 | |
3d6c50c2 YW |
8964 | if (schedstat_enabled() && tg != &root_task_group) { |
8965 | u64 ws = 0; | |
8966 | int i; | |
8967 | ||
8968 | for_each_possible_cpu(i) | |
8969 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
8970 | ||
8971 | seq_printf(sf, "wait_sum %llu\n", ws); | |
8972 | } | |
8973 | ||
e8da1b18 NR |
8974 | return 0; |
8975 | } | |
ab84d31e | 8976 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8977 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8978 | |
052f1dc7 | 8979 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8980 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8981 | struct cftype *cft, s64 val) | |
6f505b16 | 8982 | { |
182446d0 | 8983 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8984 | } |
8985 | ||
182446d0 TH |
8986 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8987 | struct cftype *cft) | |
6f505b16 | 8988 | { |
182446d0 | 8989 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8990 | } |
d0b27fa7 | 8991 | |
182446d0 TH |
8992 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8993 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8994 | { |
182446d0 | 8995 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8996 | } |
8997 | ||
182446d0 TH |
8998 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8999 | struct cftype *cft) | |
d0b27fa7 | 9000 | { |
182446d0 | 9001 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 9002 | } |
6d6bc0ad | 9003 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9004 | |
a1f7164c | 9005 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 9006 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9007 | { |
9008 | .name = "shares", | |
f4c753b7 PM |
9009 | .read_u64 = cpu_shares_read_u64, |
9010 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9011 | }, |
052f1dc7 | 9012 | #endif |
ab84d31e PT |
9013 | #ifdef CONFIG_CFS_BANDWIDTH |
9014 | { | |
9015 | .name = "cfs_quota_us", | |
9016 | .read_s64 = cpu_cfs_quota_read_s64, | |
9017 | .write_s64 = cpu_cfs_quota_write_s64, | |
9018 | }, | |
9019 | { | |
9020 | .name = "cfs_period_us", | |
9021 | .read_u64 = cpu_cfs_period_read_u64, | |
9022 | .write_u64 = cpu_cfs_period_write_u64, | |
9023 | }, | |
e8da1b18 NR |
9024 | { |
9025 | .name = "stat", | |
a1f7164c | 9026 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 9027 | }, |
ab84d31e | 9028 | #endif |
052f1dc7 | 9029 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9030 | { |
9f0c1e56 | 9031 | .name = "rt_runtime_us", |
06ecb27c PM |
9032 | .read_s64 = cpu_rt_runtime_read, |
9033 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9034 | }, |
d0b27fa7 PZ |
9035 | { |
9036 | .name = "rt_period_us", | |
f4c753b7 PM |
9037 | .read_u64 = cpu_rt_period_read_uint, |
9038 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9039 | }, |
2480c093 PB |
9040 | #endif |
9041 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9042 | { | |
9043 | .name = "uclamp.min", | |
9044 | .flags = CFTYPE_NOT_ON_ROOT, | |
9045 | .seq_show = cpu_uclamp_min_show, | |
9046 | .write = cpu_uclamp_min_write, | |
9047 | }, | |
9048 | { | |
9049 | .name = "uclamp.max", | |
9050 | .flags = CFTYPE_NOT_ON_ROOT, | |
9051 | .seq_show = cpu_uclamp_max_show, | |
9052 | .write = cpu_uclamp_max_write, | |
9053 | }, | |
052f1dc7 | 9054 | #endif |
d1ccc66d | 9055 | { } /* Terminate */ |
68318b8e SV |
9056 | }; |
9057 | ||
d41bf8c9 TH |
9058 | static int cpu_extra_stat_show(struct seq_file *sf, |
9059 | struct cgroup_subsys_state *css) | |
0d593634 | 9060 | { |
0d593634 TH |
9061 | #ifdef CONFIG_CFS_BANDWIDTH |
9062 | { | |
d41bf8c9 | 9063 | struct task_group *tg = css_tg(css); |
0d593634 TH |
9064 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
9065 | u64 throttled_usec; | |
9066 | ||
9067 | throttled_usec = cfs_b->throttled_time; | |
9068 | do_div(throttled_usec, NSEC_PER_USEC); | |
9069 | ||
9070 | seq_printf(sf, "nr_periods %d\n" | |
9071 | "nr_throttled %d\n" | |
9072 | "throttled_usec %llu\n", | |
9073 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
9074 | throttled_usec); | |
9075 | } | |
9076 | #endif | |
9077 | return 0; | |
9078 | } | |
9079 | ||
9080 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9081 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
9082 | struct cftype *cft) | |
9083 | { | |
9084 | struct task_group *tg = css_tg(css); | |
9085 | u64 weight = scale_load_down(tg->shares); | |
9086 | ||
9087 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
9088 | } | |
9089 | ||
9090 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
9091 | struct cftype *cft, u64 weight) | |
9092 | { | |
9093 | /* | |
9094 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
9095 | * values which are 1, 100 and 10000 respectively. While it loses | |
9096 | * a bit of range on both ends, it maps pretty well onto the shares | |
9097 | * value used by scheduler and the round-trip conversions preserve | |
9098 | * the original value over the entire range. | |
9099 | */ | |
9100 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
9101 | return -ERANGE; | |
9102 | ||
9103 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
9104 | ||
9105 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
9106 | } | |
9107 | ||
9108 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
9109 | struct cftype *cft) | |
9110 | { | |
9111 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
9112 | int last_delta = INT_MAX; | |
9113 | int prio, delta; | |
9114 | ||
9115 | /* find the closest nice value to the current weight */ | |
9116 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
9117 | delta = abs(sched_prio_to_weight[prio] - weight); | |
9118 | if (delta >= last_delta) | |
9119 | break; | |
9120 | last_delta = delta; | |
9121 | } | |
9122 | ||
9123 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
9124 | } | |
9125 | ||
9126 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
9127 | struct cftype *cft, s64 nice) | |
9128 | { | |
9129 | unsigned long weight; | |
7281c8de | 9130 | int idx; |
0d593634 TH |
9131 | |
9132 | if (nice < MIN_NICE || nice > MAX_NICE) | |
9133 | return -ERANGE; | |
9134 | ||
7281c8de PZ |
9135 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
9136 | idx = array_index_nospec(idx, 40); | |
9137 | weight = sched_prio_to_weight[idx]; | |
9138 | ||
0d593634 TH |
9139 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
9140 | } | |
9141 | #endif | |
9142 | ||
9143 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
9144 | long period, long quota) | |
9145 | { | |
9146 | if (quota < 0) | |
9147 | seq_puts(sf, "max"); | |
9148 | else | |
9149 | seq_printf(sf, "%ld", quota); | |
9150 | ||
9151 | seq_printf(sf, " %ld\n", period); | |
9152 | } | |
9153 | ||
9154 | /* caller should put the current value in *@periodp before calling */ | |
9155 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
9156 | u64 *periodp, u64 *quotap) | |
9157 | { | |
9158 | char tok[21]; /* U64_MAX */ | |
9159 | ||
4c47acd8 | 9160 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
9161 | return -EINVAL; |
9162 | ||
9163 | *periodp *= NSEC_PER_USEC; | |
9164 | ||
9165 | if (sscanf(tok, "%llu", quotap)) | |
9166 | *quotap *= NSEC_PER_USEC; | |
9167 | else if (!strcmp(tok, "max")) | |
9168 | *quotap = RUNTIME_INF; | |
9169 | else | |
9170 | return -EINVAL; | |
9171 | ||
9172 | return 0; | |
9173 | } | |
9174 | ||
9175 | #ifdef CONFIG_CFS_BANDWIDTH | |
9176 | static int cpu_max_show(struct seq_file *sf, void *v) | |
9177 | { | |
9178 | struct task_group *tg = css_tg(seq_css(sf)); | |
9179 | ||
9180 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
9181 | return 0; | |
9182 | } | |
9183 | ||
9184 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
9185 | char *buf, size_t nbytes, loff_t off) | |
9186 | { | |
9187 | struct task_group *tg = css_tg(of_css(of)); | |
9188 | u64 period = tg_get_cfs_period(tg); | |
9189 | u64 quota; | |
9190 | int ret; | |
9191 | ||
9192 | ret = cpu_period_quota_parse(buf, &period, "a); | |
9193 | if (!ret) | |
9194 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
9195 | return ret ?: nbytes; | |
9196 | } | |
9197 | #endif | |
9198 | ||
9199 | static struct cftype cpu_files[] = { | |
0d593634 TH |
9200 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9201 | { | |
9202 | .name = "weight", | |
9203 | .flags = CFTYPE_NOT_ON_ROOT, | |
9204 | .read_u64 = cpu_weight_read_u64, | |
9205 | .write_u64 = cpu_weight_write_u64, | |
9206 | }, | |
9207 | { | |
9208 | .name = "weight.nice", | |
9209 | .flags = CFTYPE_NOT_ON_ROOT, | |
9210 | .read_s64 = cpu_weight_nice_read_s64, | |
9211 | .write_s64 = cpu_weight_nice_write_s64, | |
9212 | }, | |
9213 | #endif | |
9214 | #ifdef CONFIG_CFS_BANDWIDTH | |
9215 | { | |
9216 | .name = "max", | |
9217 | .flags = CFTYPE_NOT_ON_ROOT, | |
9218 | .seq_show = cpu_max_show, | |
9219 | .write = cpu_max_write, | |
9220 | }, | |
2480c093 PB |
9221 | #endif |
9222 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9223 | { | |
9224 | .name = "uclamp.min", | |
9225 | .flags = CFTYPE_NOT_ON_ROOT, | |
9226 | .seq_show = cpu_uclamp_min_show, | |
9227 | .write = cpu_uclamp_min_write, | |
9228 | }, | |
9229 | { | |
9230 | .name = "uclamp.max", | |
9231 | .flags = CFTYPE_NOT_ON_ROOT, | |
9232 | .seq_show = cpu_uclamp_max_show, | |
9233 | .write = cpu_uclamp_max_write, | |
9234 | }, | |
0d593634 TH |
9235 | #endif |
9236 | { } /* terminate */ | |
9237 | }; | |
9238 | ||
073219e9 | 9239 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 9240 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 9241 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 9242 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 9243 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 9244 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 9245 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
9246 | .can_attach = cpu_cgroup_can_attach, |
9247 | .attach = cpu_cgroup_attach, | |
a1f7164c | 9248 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 9249 | .dfl_cftypes = cpu_files, |
b38e42e9 | 9250 | .early_init = true, |
0d593634 | 9251 | .threaded = true, |
68318b8e SV |
9252 | }; |
9253 | ||
052f1dc7 | 9254 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 9255 | |
b637a328 PM |
9256 | void dump_cpu_task(int cpu) |
9257 | { | |
9258 | pr_info("Task dump for CPU %d:\n", cpu); | |
9259 | sched_show_task(cpu_curr(cpu)); | |
9260 | } | |
ed82b8a1 AK |
9261 | |
9262 | /* | |
9263 | * Nice levels are multiplicative, with a gentle 10% change for every | |
9264 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
9265 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
9266 | * that remained on nice 0. | |
9267 | * | |
9268 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
9269 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
9270 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
9271 | * If a task goes up by ~10% and another task goes down by ~10% then | |
9272 | * the relative distance between them is ~25%.) | |
9273 | */ | |
9274 | const int sched_prio_to_weight[40] = { | |
9275 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
9276 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
9277 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
9278 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
9279 | /* 0 */ 1024, 820, 655, 526, 423, | |
9280 | /* 5 */ 335, 272, 215, 172, 137, | |
9281 | /* 10 */ 110, 87, 70, 56, 45, | |
9282 | /* 15 */ 36, 29, 23, 18, 15, | |
9283 | }; | |
9284 | ||
9285 | /* | |
9286 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
9287 | * | |
9288 | * In cases where the weight does not change often, we can use the | |
9289 | * precalculated inverse to speed up arithmetics by turning divisions | |
9290 | * into multiplications: | |
9291 | */ | |
9292 | const u32 sched_prio_to_wmult[40] = { | |
9293 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
9294 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
9295 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
9296 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
9297 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
9298 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
9299 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
9300 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
9301 | }; | |
14a7405b | 9302 | |
9d246053 PA |
9303 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
9304 | { | |
9305 | trace_sched_update_nr_running_tp(rq, count); | |
9306 | } |