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); | |
39 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); | |
4581bea8 VD |
40 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp); |
41 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); | |
9d246053 | 42 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); |
a056a5be | 43 | |
029632fb | 44 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 45 | |
e9666d10 | 46 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL) |
bf5c91ba IM |
47 | /* |
48 | * Debugging: various feature bits | |
765cc3a4 PB |
49 | * |
50 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
51 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
52 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 53 | */ |
f00b45c1 PZ |
54 | #define SCHED_FEAT(name, enabled) \ |
55 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 56 | const_debug unsigned int sysctl_sched_features = |
391e43da | 57 | #include "features.h" |
f00b45c1 | 58 | 0; |
f00b45c1 | 59 | #undef SCHED_FEAT |
765cc3a4 | 60 | #endif |
f00b45c1 | 61 | |
b82d9fdd PZ |
62 | /* |
63 | * Number of tasks to iterate in a single balance run. | |
64 | * Limited because this is done with IRQs disabled. | |
65 | */ | |
66 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
67 | ||
fa85ae24 | 68 | /* |
d1ccc66d | 69 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
70 | * default: 1s |
71 | */ | |
9f0c1e56 | 72 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 73 | |
029632fb | 74 | __read_mostly int scheduler_running; |
6892b75e | 75 | |
9f0c1e56 PZ |
76 | /* |
77 | * part of the period that we allow rt tasks to run in us. | |
78 | * default: 0.95s | |
79 | */ | |
80 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 81 | |
58877d34 PZ |
82 | |
83 | /* | |
84 | * Serialization rules: | |
85 | * | |
86 | * Lock order: | |
87 | * | |
88 | * p->pi_lock | |
89 | * rq->lock | |
90 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
91 | * | |
92 | * rq1->lock | |
93 | * rq2->lock where: rq1 < rq2 | |
94 | * | |
95 | * Regular state: | |
96 | * | |
97 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
98 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
99 | * always looks at the local rq data structures to find the most elegible task | |
100 | * to run next. | |
101 | * | |
102 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
103 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
104 | * the local CPU to avoid bouncing the runqueue state around [ see | |
105 | * ttwu_queue_wakelist() ] | |
106 | * | |
107 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
108 | * complicated to avoid having to take two rq->locks. | |
109 | * | |
110 | * Special state: | |
111 | * | |
112 | * System-calls and anything external will use task_rq_lock() which acquires | |
113 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
114 | * stable while holding either lock: | |
115 | * | |
116 | * - sched_setaffinity()/ | |
117 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
118 | * - set_user_nice(): p->se.load, p->*prio | |
119 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
120 | * p->se.load, p->rt_priority, | |
121 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
122 | * - sched_setnuma(): p->numa_preferred_nid | |
123 | * - sched_move_task()/ | |
124 | * cpu_cgroup_fork(): p->sched_task_group | |
125 | * - uclamp_update_active() p->uclamp* | |
126 | * | |
127 | * p->state <- TASK_*: | |
128 | * | |
129 | * is changed locklessly using set_current_state(), __set_current_state() or | |
130 | * set_special_state(), see their respective comments, or by | |
131 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
132 | * concurrent self. | |
133 | * | |
134 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
135 | * | |
136 | * is set by activate_task() and cleared by deactivate_task(), under | |
137 | * rq->lock. Non-zero indicates the task is runnable, the special | |
138 | * ON_RQ_MIGRATING state is used for migration without holding both | |
139 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
140 | * | |
141 | * p->on_cpu <- { 0, 1 }: | |
142 | * | |
143 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
144 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
145 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
146 | * | |
147 | * [ The astute reader will observe that it is possible for two tasks on one | |
148 | * CPU to have ->on_cpu = 1 at the same time. ] | |
149 | * | |
150 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
151 | * | |
152 | * - Don't call set_task_cpu() on a blocked task: | |
153 | * | |
154 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
155 | * the CPU assignment of blocked tasks isn't required to be valid. | |
156 | * | |
157 | * - for try_to_wake_up(), called under p->pi_lock: | |
158 | * | |
159 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
160 | * | |
161 | * - for migration called under rq->lock: | |
162 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
163 | * | |
164 | * o move_queued_task() | |
165 | * o detach_task() | |
166 | * | |
167 | * - for migration called under double_rq_lock(): | |
168 | * | |
169 | * o __migrate_swap_task() | |
170 | * o push_rt_task() / pull_rt_task() | |
171 | * o push_dl_task() / pull_dl_task() | |
172 | * o dl_task_offline_migration() | |
173 | * | |
174 | */ | |
175 | ||
3e71a462 PZ |
176 | /* |
177 | * __task_rq_lock - lock the rq @p resides on. | |
178 | */ | |
eb580751 | 179 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
180 | __acquires(rq->lock) |
181 | { | |
182 | struct rq *rq; | |
183 | ||
184 | lockdep_assert_held(&p->pi_lock); | |
185 | ||
186 | for (;;) { | |
187 | rq = task_rq(p); | |
188 | raw_spin_lock(&rq->lock); | |
189 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 190 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
191 | return rq; |
192 | } | |
193 | raw_spin_unlock(&rq->lock); | |
194 | ||
195 | while (unlikely(task_on_rq_migrating(p))) | |
196 | cpu_relax(); | |
197 | } | |
198 | } | |
199 | ||
200 | /* | |
201 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
202 | */ | |
eb580751 | 203 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
204 | __acquires(p->pi_lock) |
205 | __acquires(rq->lock) | |
206 | { | |
207 | struct rq *rq; | |
208 | ||
209 | for (;;) { | |
eb580751 | 210 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 PZ |
211 | rq = task_rq(p); |
212 | raw_spin_lock(&rq->lock); | |
213 | /* | |
214 | * move_queued_task() task_rq_lock() | |
215 | * | |
216 | * ACQUIRE (rq->lock) | |
217 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
218 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
219 | * [S] ->cpu = new_cpu [L] task_rq() | |
220 | * [L] ->on_rq | |
221 | * RELEASE (rq->lock) | |
222 | * | |
c546951d | 223 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
224 | * the old rq->lock will fully serialize against the stores. |
225 | * | |
c546951d AP |
226 | * If we observe the new CPU in task_rq_lock(), the address |
227 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
228 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
229 | */ |
230 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 231 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
232 | return rq; |
233 | } | |
234 | raw_spin_unlock(&rq->lock); | |
eb580751 | 235 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
236 | |
237 | while (unlikely(task_on_rq_migrating(p))) | |
238 | cpu_relax(); | |
239 | } | |
240 | } | |
241 | ||
535b9552 IM |
242 | /* |
243 | * RQ-clock updating methods: | |
244 | */ | |
245 | ||
246 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
247 | { | |
248 | /* | |
249 | * In theory, the compile should just see 0 here, and optimize out the call | |
250 | * to sched_rt_avg_update. But I don't trust it... | |
251 | */ | |
11d4afd4 VG |
252 | s64 __maybe_unused steal = 0, irq_delta = 0; |
253 | ||
535b9552 IM |
254 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
255 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
256 | ||
257 | /* | |
258 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
259 | * this case when a previous update_rq_clock() happened inside a | |
260 | * {soft,}irq region. | |
261 | * | |
262 | * When this happens, we stop ->clock_task and only update the | |
263 | * prev_irq_time stamp to account for the part that fit, so that a next | |
264 | * update will consume the rest. This ensures ->clock_task is | |
265 | * monotonic. | |
266 | * | |
267 | * It does however cause some slight miss-attribution of {soft,}irq | |
268 | * time, a more accurate solution would be to update the irq_time using | |
269 | * the current rq->clock timestamp, except that would require using | |
270 | * atomic ops. | |
271 | */ | |
272 | if (irq_delta > delta) | |
273 | irq_delta = delta; | |
274 | ||
275 | rq->prev_irq_time += irq_delta; | |
276 | delta -= irq_delta; | |
277 | #endif | |
278 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
279 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
280 | steal = paravirt_steal_clock(cpu_of(rq)); | |
281 | steal -= rq->prev_steal_time_rq; | |
282 | ||
283 | if (unlikely(steal > delta)) | |
284 | steal = delta; | |
285 | ||
286 | rq->prev_steal_time_rq += steal; | |
287 | delta -= steal; | |
288 | } | |
289 | #endif | |
290 | ||
291 | rq->clock_task += delta; | |
292 | ||
11d4afd4 | 293 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 294 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 295 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 296 | #endif |
23127296 | 297 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
298 | } |
299 | ||
300 | void update_rq_clock(struct rq *rq) | |
301 | { | |
302 | s64 delta; | |
303 | ||
304 | lockdep_assert_held(&rq->lock); | |
305 | ||
306 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
307 | return; | |
308 | ||
309 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
310 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
311 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
312 | rq->clock_update_flags |= RQCF_UPDATED; |
313 | #endif | |
26ae58d2 | 314 | |
535b9552 IM |
315 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
316 | if (delta < 0) | |
317 | return; | |
318 | rq->clock += delta; | |
319 | update_rq_clock_task(rq, delta); | |
320 | } | |
321 | ||
90b5363a PZI |
322 | static inline void |
323 | rq_csd_init(struct rq *rq, call_single_data_t *csd, smp_call_func_t func) | |
324 | { | |
325 | csd->flags = 0; | |
326 | csd->func = func; | |
327 | csd->info = rq; | |
328 | } | |
535b9552 | 329 | |
8f4d37ec PZ |
330 | #ifdef CONFIG_SCHED_HRTICK |
331 | /* | |
332 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 333 | */ |
8f4d37ec | 334 | |
8f4d37ec PZ |
335 | static void hrtick_clear(struct rq *rq) |
336 | { | |
337 | if (hrtimer_active(&rq->hrtick_timer)) | |
338 | hrtimer_cancel(&rq->hrtick_timer); | |
339 | } | |
340 | ||
8f4d37ec PZ |
341 | /* |
342 | * High-resolution timer tick. | |
343 | * Runs from hardirq context with interrupts disabled. | |
344 | */ | |
345 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
346 | { | |
347 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 348 | struct rq_flags rf; |
8f4d37ec PZ |
349 | |
350 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
351 | ||
8a8c69c3 | 352 | rq_lock(rq, &rf); |
3e51f33f | 353 | update_rq_clock(rq); |
8f4d37ec | 354 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 355 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
356 | |
357 | return HRTIMER_NORESTART; | |
358 | } | |
359 | ||
95e904c7 | 360 | #ifdef CONFIG_SMP |
971ee28c | 361 | |
4961b6e1 | 362 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
363 | { |
364 | struct hrtimer *timer = &rq->hrtick_timer; | |
971ee28c | 365 | |
d5096aa6 | 366 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
367 | } |
368 | ||
31656519 PZ |
369 | /* |
370 | * called from hardirq (IPI) context | |
371 | */ | |
372 | static void __hrtick_start(void *arg) | |
b328ca18 | 373 | { |
31656519 | 374 | struct rq *rq = arg; |
8a8c69c3 | 375 | struct rq_flags rf; |
b328ca18 | 376 | |
8a8c69c3 | 377 | rq_lock(rq, &rf); |
971ee28c | 378 | __hrtick_restart(rq); |
8a8c69c3 | 379 | rq_unlock(rq, &rf); |
b328ca18 PZ |
380 | } |
381 | ||
31656519 PZ |
382 | /* |
383 | * Called to set the hrtick timer state. | |
384 | * | |
385 | * called with rq->lock held and irqs disabled | |
386 | */ | |
029632fb | 387 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 388 | { |
31656519 | 389 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 390 | ktime_t time; |
391 | s64 delta; | |
392 | ||
393 | /* | |
394 | * Don't schedule slices shorter than 10000ns, that just | |
395 | * doesn't make sense and can cause timer DoS. | |
396 | */ | |
397 | delta = max_t(s64, delay, 10000LL); | |
398 | time = ktime_add_ns(timer->base->get_time(), delta); | |
b328ca18 | 399 | |
cc584b21 | 400 | hrtimer_set_expires(timer, time); |
31656519 | 401 | |
fd3eafda | 402 | if (rq == this_rq()) |
971ee28c | 403 | __hrtick_restart(rq); |
fd3eafda | 404 | else |
c46fff2a | 405 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
406 | } |
407 | ||
31656519 PZ |
408 | #else |
409 | /* | |
410 | * Called to set the hrtick timer state. | |
411 | * | |
412 | * called with rq->lock held and irqs disabled | |
413 | */ | |
029632fb | 414 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 415 | { |
86893335 WL |
416 | /* |
417 | * Don't schedule slices shorter than 10000ns, that just | |
418 | * doesn't make sense. Rely on vruntime for fairness. | |
419 | */ | |
420 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 421 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 422 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 423 | } |
90b5363a | 424 | |
31656519 | 425 | #endif /* CONFIG_SMP */ |
8f4d37ec | 426 | |
77a021be | 427 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 428 | { |
31656519 | 429 | #ifdef CONFIG_SMP |
90b5363a | 430 | rq_csd_init(rq, &rq->hrtick_csd, __hrtick_start); |
31656519 | 431 | #endif |
d5096aa6 | 432 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 433 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 434 | } |
006c75f1 | 435 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
436 | static inline void hrtick_clear(struct rq *rq) |
437 | { | |
438 | } | |
439 | ||
77a021be | 440 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
441 | { |
442 | } | |
006c75f1 | 443 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 444 | |
5529578a FW |
445 | /* |
446 | * cmpxchg based fetch_or, macro so it works for different integer types | |
447 | */ | |
448 | #define fetch_or(ptr, mask) \ | |
449 | ({ \ | |
450 | typeof(ptr) _ptr = (ptr); \ | |
451 | typeof(mask) _mask = (mask); \ | |
452 | typeof(*_ptr) _old, _val = *_ptr; \ | |
453 | \ | |
454 | for (;;) { \ | |
455 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
456 | if (_old == _val) \ | |
457 | break; \ | |
458 | _val = _old; \ | |
459 | } \ | |
460 | _old; \ | |
461 | }) | |
462 | ||
e3baac47 | 463 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
464 | /* |
465 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
466 | * this avoids any races wrt polling state changes and thereby avoids | |
467 | * spurious IPIs. | |
468 | */ | |
469 | static bool set_nr_and_not_polling(struct task_struct *p) | |
470 | { | |
471 | struct thread_info *ti = task_thread_info(p); | |
472 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
473 | } | |
e3baac47 PZ |
474 | |
475 | /* | |
476 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
477 | * | |
478 | * If this returns true, then the idle task promises to call | |
479 | * sched_ttwu_pending() and reschedule soon. | |
480 | */ | |
481 | static bool set_nr_if_polling(struct task_struct *p) | |
482 | { | |
483 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 484 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
485 | |
486 | for (;;) { | |
487 | if (!(val & _TIF_POLLING_NRFLAG)) | |
488 | return false; | |
489 | if (val & _TIF_NEED_RESCHED) | |
490 | return true; | |
491 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
492 | if (old == val) | |
493 | break; | |
494 | val = old; | |
495 | } | |
496 | return true; | |
497 | } | |
498 | ||
fd99f91a PZ |
499 | #else |
500 | static bool set_nr_and_not_polling(struct task_struct *p) | |
501 | { | |
502 | set_tsk_need_resched(p); | |
503 | return true; | |
504 | } | |
e3baac47 PZ |
505 | |
506 | #ifdef CONFIG_SMP | |
507 | static bool set_nr_if_polling(struct task_struct *p) | |
508 | { | |
509 | return false; | |
510 | } | |
511 | #endif | |
fd99f91a PZ |
512 | #endif |
513 | ||
07879c6a | 514 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
515 | { |
516 | struct wake_q_node *node = &task->wake_q; | |
517 | ||
518 | /* | |
519 | * Atomically grab the task, if ->wake_q is !nil already it means | |
520 | * its already queued (either by us or someone else) and will get the | |
521 | * wakeup due to that. | |
522 | * | |
4c4e3731 PZ |
523 | * In order to ensure that a pending wakeup will observe our pending |
524 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 525 | */ |
4c4e3731 | 526 | smp_mb__before_atomic(); |
87ff19cb | 527 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 528 | return false; |
76751049 PZ |
529 | |
530 | /* | |
531 | * The head is context local, there can be no concurrency. | |
532 | */ | |
533 | *head->lastp = node; | |
534 | head->lastp = &node->next; | |
07879c6a DB |
535 | return true; |
536 | } | |
537 | ||
538 | /** | |
539 | * wake_q_add() - queue a wakeup for 'later' waking. | |
540 | * @head: the wake_q_head to add @task to | |
541 | * @task: the task to queue for 'later' wakeup | |
542 | * | |
543 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
544 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
545 | * instantly. | |
546 | * | |
547 | * This function must be used as-if it were wake_up_process(); IOW the task | |
548 | * must be ready to be woken at this location. | |
549 | */ | |
550 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
551 | { | |
552 | if (__wake_q_add(head, task)) | |
553 | get_task_struct(task); | |
554 | } | |
555 | ||
556 | /** | |
557 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
558 | * @head: the wake_q_head to add @task to | |
559 | * @task: the task to queue for 'later' wakeup | |
560 | * | |
561 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
562 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
563 | * instantly. | |
564 | * | |
565 | * This function must be used as-if it were wake_up_process(); IOW the task | |
566 | * must be ready to be woken at this location. | |
567 | * | |
568 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
569 | * that already hold reference to @task can call the 'safe' version and trust | |
570 | * wake_q to do the right thing depending whether or not the @task is already | |
571 | * queued for wakeup. | |
572 | */ | |
573 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
574 | { | |
575 | if (!__wake_q_add(head, task)) | |
576 | put_task_struct(task); | |
76751049 PZ |
577 | } |
578 | ||
579 | void wake_up_q(struct wake_q_head *head) | |
580 | { | |
581 | struct wake_q_node *node = head->first; | |
582 | ||
583 | while (node != WAKE_Q_TAIL) { | |
584 | struct task_struct *task; | |
585 | ||
586 | task = container_of(node, struct task_struct, wake_q); | |
587 | BUG_ON(!task); | |
d1ccc66d | 588 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
589 | node = node->next; |
590 | task->wake_q.next = NULL; | |
591 | ||
592 | /* | |
7696f991 AP |
593 | * wake_up_process() executes a full barrier, which pairs with |
594 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
595 | */ |
596 | wake_up_process(task); | |
597 | put_task_struct(task); | |
598 | } | |
599 | } | |
600 | ||
c24d20db | 601 | /* |
8875125e | 602 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
603 | * |
604 | * On UP this means the setting of the need_resched flag, on SMP it | |
605 | * might also involve a cross-CPU call to trigger the scheduler on | |
606 | * the target CPU. | |
607 | */ | |
8875125e | 608 | void resched_curr(struct rq *rq) |
c24d20db | 609 | { |
8875125e | 610 | struct task_struct *curr = rq->curr; |
c24d20db IM |
611 | int cpu; |
612 | ||
8875125e | 613 | lockdep_assert_held(&rq->lock); |
c24d20db | 614 | |
8875125e | 615 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
616 | return; |
617 | ||
8875125e | 618 | cpu = cpu_of(rq); |
fd99f91a | 619 | |
f27dde8d | 620 | if (cpu == smp_processor_id()) { |
8875125e | 621 | set_tsk_need_resched(curr); |
f27dde8d | 622 | set_preempt_need_resched(); |
c24d20db | 623 | return; |
f27dde8d | 624 | } |
c24d20db | 625 | |
8875125e | 626 | if (set_nr_and_not_polling(curr)) |
c24d20db | 627 | smp_send_reschedule(cpu); |
dfc68f29 AL |
628 | else |
629 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
630 | } |
631 | ||
029632fb | 632 | void resched_cpu(int cpu) |
c24d20db IM |
633 | { |
634 | struct rq *rq = cpu_rq(cpu); | |
635 | unsigned long flags; | |
636 | ||
7c2102e5 | 637 | raw_spin_lock_irqsave(&rq->lock, flags); |
a0982dfa PM |
638 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
639 | resched_curr(rq); | |
05fa785c | 640 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 641 | } |
06d8308c | 642 | |
b021fe3e | 643 | #ifdef CONFIG_SMP |
3451d024 | 644 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 645 | /* |
d1ccc66d IM |
646 | * In the semi idle case, use the nearest busy CPU for migrating timers |
647 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
648 | * |
649 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
650 | * selecting an idle CPU will add more delays to the timers than intended |
651 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 652 | */ |
bc7a34b8 | 653 | int get_nohz_timer_target(void) |
83cd4fe2 | 654 | { |
e938b9c9 | 655 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 VP |
656 | struct sched_domain *sd; |
657 | ||
e938b9c9 WL |
658 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
659 | if (!idle_cpu(cpu)) | |
660 | return cpu; | |
661 | default_cpu = cpu; | |
662 | } | |
6201b4d6 | 663 | |
057f3fad | 664 | rcu_read_lock(); |
83cd4fe2 | 665 | for_each_domain(cpu, sd) { |
e938b9c9 WL |
666 | for_each_cpu_and(i, sched_domain_span(sd), |
667 | housekeeping_cpumask(HK_FLAG_TIMER)) { | |
44496922 WL |
668 | if (cpu == i) |
669 | continue; | |
670 | ||
e938b9c9 | 671 | if (!idle_cpu(i)) { |
057f3fad PZ |
672 | cpu = i; |
673 | goto unlock; | |
674 | } | |
675 | } | |
83cd4fe2 | 676 | } |
9642d18e | 677 | |
e938b9c9 WL |
678 | if (default_cpu == -1) |
679 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
680 | cpu = default_cpu; | |
057f3fad PZ |
681 | unlock: |
682 | rcu_read_unlock(); | |
83cd4fe2 VP |
683 | return cpu; |
684 | } | |
d1ccc66d | 685 | |
06d8308c TG |
686 | /* |
687 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
688 | * idle CPU then this timer might expire before the next timer event | |
689 | * which is scheduled to wake up that CPU. In case of a completely | |
690 | * idle system the next event might even be infinite time into the | |
691 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
692 | * leaves the inner idle loop so the newly added timer is taken into | |
693 | * account when the CPU goes back to idle and evaluates the timer | |
694 | * wheel for the next timer event. | |
695 | */ | |
1c20091e | 696 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
697 | { |
698 | struct rq *rq = cpu_rq(cpu); | |
699 | ||
700 | if (cpu == smp_processor_id()) | |
701 | return; | |
702 | ||
67b9ca70 | 703 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 704 | smp_send_reschedule(cpu); |
dfc68f29 AL |
705 | else |
706 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
707 | } |
708 | ||
c5bfece2 | 709 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 710 | { |
53c5fa16 FW |
711 | /* |
712 | * We just need the target to call irq_exit() and re-evaluate | |
713 | * the next tick. The nohz full kick at least implies that. | |
714 | * If needed we can still optimize that later with an | |
715 | * empty IRQ. | |
716 | */ | |
379d9ecb PM |
717 | if (cpu_is_offline(cpu)) |
718 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 719 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
720 | if (cpu != smp_processor_id() || |
721 | tick_nohz_tick_stopped()) | |
53c5fa16 | 722 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
723 | return true; |
724 | } | |
725 | ||
726 | return false; | |
727 | } | |
728 | ||
379d9ecb PM |
729 | /* |
730 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
731 | * caller's responsibility to deal with the lost wakeup, for example, | |
732 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
733 | */ | |
1c20091e FW |
734 | void wake_up_nohz_cpu(int cpu) |
735 | { | |
c5bfece2 | 736 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
737 | wake_up_idle_cpu(cpu); |
738 | } | |
739 | ||
19a1f5ec | 740 | static void nohz_csd_func(void *info) |
45bf76df | 741 | { |
19a1f5ec PZ |
742 | struct rq *rq = info; |
743 | int cpu = cpu_of(rq); | |
744 | unsigned int flags; | |
873b4c65 VG |
745 | |
746 | /* | |
19a1f5ec | 747 | * Release the rq::nohz_csd. |
873b4c65 | 748 | */ |
19a1f5ec PZ |
749 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK, nohz_flags(cpu)); |
750 | WARN_ON(!(flags & NOHZ_KICK_MASK)); | |
45bf76df | 751 | |
19a1f5ec PZ |
752 | rq->idle_balance = idle_cpu(cpu); |
753 | if (rq->idle_balance && !need_resched()) { | |
754 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
755 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
756 | } | |
2069dd75 PZ |
757 | } |
758 | ||
3451d024 | 759 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 760 | |
ce831b38 | 761 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 762 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 763 | { |
76d92ac3 FW |
764 | int fifo_nr_running; |
765 | ||
766 | /* Deadline tasks, even if single, need the tick */ | |
767 | if (rq->dl.dl_nr_running) | |
768 | return false; | |
769 | ||
1e78cdbd | 770 | /* |
2548d546 PZ |
771 | * If there are more than one RR tasks, we need the tick to effect the |
772 | * actual RR behaviour. | |
1e78cdbd | 773 | */ |
76d92ac3 FW |
774 | if (rq->rt.rr_nr_running) { |
775 | if (rq->rt.rr_nr_running == 1) | |
776 | return true; | |
777 | else | |
778 | return false; | |
1e78cdbd RR |
779 | } |
780 | ||
2548d546 PZ |
781 | /* |
782 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
783 | * forced preemption between FIFO tasks. | |
784 | */ | |
785 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
786 | if (fifo_nr_running) | |
787 | return true; | |
788 | ||
789 | /* | |
790 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
791 | * if there's more than one we need the tick for involuntary | |
792 | * preemption. | |
793 | */ | |
794 | if (rq->nr_running > 1) | |
541b8264 | 795 | return false; |
ce831b38 | 796 | |
541b8264 | 797 | return true; |
ce831b38 FW |
798 | } |
799 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 800 | #endif /* CONFIG_SMP */ |
18d95a28 | 801 | |
a790de99 PT |
802 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
803 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 804 | /* |
8277434e PT |
805 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
806 | * node and @up when leaving it for the final time. | |
807 | * | |
808 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 809 | */ |
029632fb | 810 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 811 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
812 | { |
813 | struct task_group *parent, *child; | |
eb755805 | 814 | int ret; |
c09595f6 | 815 | |
8277434e PT |
816 | parent = from; |
817 | ||
c09595f6 | 818 | down: |
eb755805 PZ |
819 | ret = (*down)(parent, data); |
820 | if (ret) | |
8277434e | 821 | goto out; |
c09595f6 PZ |
822 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
823 | parent = child; | |
824 | goto down; | |
825 | ||
826 | up: | |
827 | continue; | |
828 | } | |
eb755805 | 829 | ret = (*up)(parent, data); |
8277434e PT |
830 | if (ret || parent == from) |
831 | goto out; | |
c09595f6 PZ |
832 | |
833 | child = parent; | |
834 | parent = parent->parent; | |
835 | if (parent) | |
836 | goto up; | |
8277434e | 837 | out: |
eb755805 | 838 | return ret; |
c09595f6 PZ |
839 | } |
840 | ||
029632fb | 841 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 842 | { |
e2b245f8 | 843 | return 0; |
eb755805 | 844 | } |
18d95a28 PZ |
845 | #endif |
846 | ||
9059393e | 847 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 848 | { |
f05998d4 NR |
849 | int prio = p->static_prio - MAX_RT_PRIO; |
850 | struct load_weight *load = &p->se.load; | |
851 | ||
dd41f596 IM |
852 | /* |
853 | * SCHED_IDLE tasks get minimal weight: | |
854 | */ | |
1da1843f | 855 | if (task_has_idle_policy(p)) { |
c8b28116 | 856 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 857 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
858 | return; |
859 | } | |
71f8bd46 | 860 | |
9059393e VG |
861 | /* |
862 | * SCHED_OTHER tasks have to update their load when changing their | |
863 | * weight | |
864 | */ | |
865 | if (update_load && p->sched_class == &fair_sched_class) { | |
866 | reweight_task(p, prio); | |
867 | } else { | |
868 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
869 | load->inv_weight = sched_prio_to_wmult[prio]; | |
870 | } | |
71f8bd46 IM |
871 | } |
872 | ||
69842cba | 873 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
874 | /* |
875 | * Serializes updates of utilization clamp values | |
876 | * | |
877 | * The (slow-path) user-space triggers utilization clamp value updates which | |
878 | * can require updates on (fast-path) scheduler's data structures used to | |
879 | * support enqueue/dequeue operations. | |
880 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
881 | * requests are serialized using a mutex to reduce the risk of conflicting | |
882 | * updates or API abuses. | |
883 | */ | |
884 | static DEFINE_MUTEX(uclamp_mutex); | |
885 | ||
e8f14172 PB |
886 | /* Max allowed minimum utilization */ |
887 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
888 | ||
889 | /* Max allowed maximum utilization */ | |
890 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
891 | ||
13685c4a QY |
892 | /* |
893 | * By default RT tasks run at the maximum performance point/capacity of the | |
894 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
895 | * SCHED_CAPACITY_SCALE. | |
896 | * | |
897 | * This knob allows admins to change the default behavior when uclamp is being | |
898 | * used. In battery powered devices, particularly, running at the maximum | |
899 | * capacity and frequency will increase energy consumption and shorten the | |
900 | * battery life. | |
901 | * | |
902 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
903 | * | |
904 | * This knob will not override the system default sched_util_clamp_min defined | |
905 | * above. | |
906 | */ | |
907 | unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; | |
908 | ||
e8f14172 PB |
909 | /* All clamps are required to be less or equal than these values */ |
910 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 911 | |
46609ce2 QY |
912 | /* |
913 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
914 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
915 | * enqueue/dequeue_task(). | |
916 | * | |
917 | * This allows users to continue to enable uclamp in their kernel config with | |
918 | * minimum uclamp overhead in the fast path. | |
919 | * | |
920 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
921 | * enabled, since we have an actual users that make use of uclamp | |
922 | * functionality. | |
923 | * | |
924 | * The knobs that would enable this static key are: | |
925 | * | |
926 | * * A task modifying its uclamp value with sched_setattr(). | |
927 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
928 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
929 | */ | |
930 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
931 | ||
69842cba PB |
932 | /* Integer rounded range for each bucket */ |
933 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
934 | ||
935 | #define for_each_clamp_id(clamp_id) \ | |
936 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
937 | ||
938 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
939 | { | |
940 | return clamp_value / UCLAMP_BUCKET_DELTA; | |
941 | } | |
942 | ||
60daf9c1 PB |
943 | static inline unsigned int uclamp_bucket_base_value(unsigned int clamp_value) |
944 | { | |
945 | return UCLAMP_BUCKET_DELTA * uclamp_bucket_id(clamp_value); | |
946 | } | |
947 | ||
7763baac | 948 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
949 | { |
950 | if (clamp_id == UCLAMP_MIN) | |
951 | return 0; | |
952 | return SCHED_CAPACITY_SCALE; | |
953 | } | |
954 | ||
a509a7cd PB |
955 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
956 | unsigned int value, bool user_defined) | |
69842cba PB |
957 | { |
958 | uc_se->value = value; | |
959 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 960 | uc_se->user_defined = user_defined; |
69842cba PB |
961 | } |
962 | ||
e496187d | 963 | static inline unsigned int |
0413d7f3 | 964 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
965 | unsigned int clamp_value) |
966 | { | |
967 | /* | |
968 | * Avoid blocked utilization pushing up the frequency when we go | |
969 | * idle (which drops the max-clamp) by retaining the last known | |
970 | * max-clamp. | |
971 | */ | |
972 | if (clamp_id == UCLAMP_MAX) { | |
973 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
974 | return clamp_value; | |
975 | } | |
976 | ||
977 | return uclamp_none(UCLAMP_MIN); | |
978 | } | |
979 | ||
0413d7f3 | 980 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
981 | unsigned int clamp_value) |
982 | { | |
983 | /* Reset max-clamp retention only on idle exit */ | |
984 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
985 | return; | |
986 | ||
987 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
988 | } | |
989 | ||
69842cba | 990 | static inline |
7763baac | 991 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 992 | unsigned int clamp_value) |
69842cba PB |
993 | { |
994 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
995 | int bucket_id = UCLAMP_BUCKETS - 1; | |
996 | ||
997 | /* | |
998 | * Since both min and max clamps are max aggregated, find the | |
999 | * top most bucket with tasks in. | |
1000 | */ | |
1001 | for ( ; bucket_id >= 0; bucket_id--) { | |
1002 | if (!bucket[bucket_id].tasks) | |
1003 | continue; | |
1004 | return bucket[bucket_id].value; | |
1005 | } | |
1006 | ||
1007 | /* No tasks -- default clamp values */ | |
e496187d | 1008 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
1009 | } |
1010 | ||
13685c4a QY |
1011 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
1012 | { | |
1013 | unsigned int default_util_min; | |
1014 | struct uclamp_se *uc_se; | |
1015 | ||
1016 | lockdep_assert_held(&p->pi_lock); | |
1017 | ||
1018 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1019 | ||
1020 | /* Only sync if user didn't override the default */ | |
1021 | if (uc_se->user_defined) | |
1022 | return; | |
1023 | ||
1024 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1025 | uclamp_se_set(uc_se, default_util_min, false); | |
1026 | } | |
1027 | ||
1028 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1029 | { | |
1030 | struct rq_flags rf; | |
1031 | struct rq *rq; | |
1032 | ||
1033 | if (!rt_task(p)) | |
1034 | return; | |
1035 | ||
1036 | /* Protect updates to p->uclamp_* */ | |
1037 | rq = task_rq_lock(p, &rf); | |
1038 | __uclamp_update_util_min_rt_default(p); | |
1039 | task_rq_unlock(rq, p, &rf); | |
1040 | } | |
1041 | ||
1042 | static void uclamp_sync_util_min_rt_default(void) | |
1043 | { | |
1044 | struct task_struct *g, *p; | |
1045 | ||
1046 | /* | |
1047 | * copy_process() sysctl_uclamp | |
1048 | * uclamp_min_rt = X; | |
1049 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1050 | * // link thread smp_mb__after_spinlock() | |
1051 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1052 | * sched_post_fork() for_each_process_thread() | |
1053 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1054 | * | |
1055 | * Ensures that either sched_post_fork() will observe the new | |
1056 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1057 | * task. | |
1058 | */ | |
1059 | read_lock(&tasklist_lock); | |
1060 | smp_mb__after_spinlock(); | |
1061 | read_unlock(&tasklist_lock); | |
1062 | ||
1063 | rcu_read_lock(); | |
1064 | for_each_process_thread(g, p) | |
1065 | uclamp_update_util_min_rt_default(p); | |
1066 | rcu_read_unlock(); | |
1067 | } | |
1068 | ||
3eac870a | 1069 | static inline struct uclamp_se |
0413d7f3 | 1070 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a PB |
1071 | { |
1072 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; | |
1073 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
1074 | struct uclamp_se uc_max; | |
1075 | ||
1076 | /* | |
1077 | * Tasks in autogroups or root task group will be | |
1078 | * restricted by system defaults. | |
1079 | */ | |
1080 | if (task_group_is_autogroup(task_group(p))) | |
1081 | return uc_req; | |
1082 | if (task_group(p) == &root_task_group) | |
1083 | return uc_req; | |
1084 | ||
1085 | uc_max = task_group(p)->uclamp[clamp_id]; | |
1086 | if (uc_req.value > uc_max.value || !uc_req.user_defined) | |
1087 | return uc_max; | |
1088 | #endif | |
1089 | ||
1090 | return uc_req; | |
1091 | } | |
1092 | ||
e8f14172 PB |
1093 | /* |
1094 | * The effective clamp bucket index of a task depends on, by increasing | |
1095 | * priority: | |
1096 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1097 | * - the task group effective clamp value, for tasks not either in the root |
1098 | * group or in an autogroup | |
e8f14172 PB |
1099 | * - the system default clamp value, defined by the sysadmin |
1100 | */ | |
1101 | static inline struct uclamp_se | |
0413d7f3 | 1102 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1103 | { |
3eac870a | 1104 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1105 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1106 | ||
1107 | /* System default restrictions always apply */ | |
1108 | if (unlikely(uc_req.value > uc_max.value)) | |
1109 | return uc_max; | |
1110 | ||
1111 | return uc_req; | |
1112 | } | |
1113 | ||
686516b5 | 1114 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1115 | { |
1116 | struct uclamp_se uc_eff; | |
1117 | ||
1118 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1119 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1120 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1121 | |
1122 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1123 | ||
686516b5 | 1124 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1125 | } |
1126 | ||
69842cba PB |
1127 | /* |
1128 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1129 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1130 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1131 | * |
1132 | * Tasks can have a task-specific value requested from user-space, track | |
1133 | * within each bucket the maximum value for tasks refcounted in it. | |
1134 | * This "local max aggregation" allows to track the exact "requested" value | |
1135 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1136 | */ |
1137 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1138 | enum uclamp_id clamp_id) |
69842cba PB |
1139 | { |
1140 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1141 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1142 | struct uclamp_bucket *bucket; | |
1143 | ||
1144 | lockdep_assert_held(&rq->lock); | |
1145 | ||
e8f14172 PB |
1146 | /* Update task effective clamp */ |
1147 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1148 | ||
69842cba PB |
1149 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1150 | bucket->tasks++; | |
e8f14172 | 1151 | uc_se->active = true; |
69842cba | 1152 | |
e496187d PB |
1153 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1154 | ||
60daf9c1 PB |
1155 | /* |
1156 | * Local max aggregation: rq buckets always track the max | |
1157 | * "requested" clamp value of its RUNNABLE tasks. | |
1158 | */ | |
1159 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1160 | bucket->value = uc_se->value; | |
1161 | ||
69842cba | 1162 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1163 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1164 | } |
1165 | ||
1166 | /* | |
1167 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1168 | * is released. If this is the last task reference counting the rq's max | |
1169 | * active clamp value, then the rq's clamp value is updated. | |
1170 | * | |
1171 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1172 | * always valid. If it's detected they are not, as defensive programming, | |
1173 | * enforce the expected state and warn. | |
1174 | */ | |
1175 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1176 | enum uclamp_id clamp_id) |
69842cba PB |
1177 | { |
1178 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1179 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1180 | struct uclamp_bucket *bucket; | |
e496187d | 1181 | unsigned int bkt_clamp; |
69842cba PB |
1182 | unsigned int rq_clamp; |
1183 | ||
1184 | lockdep_assert_held(&rq->lock); | |
1185 | ||
46609ce2 QY |
1186 | /* |
1187 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1188 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1189 | * | |
1190 | * In this case the uc_se->active flag should be false since no uclamp | |
1191 | * accounting was performed at enqueue time and we can just return | |
1192 | * here. | |
1193 | * | |
1194 | * Need to be careful of the following enqeueue/dequeue ordering | |
1195 | * problem too | |
1196 | * | |
1197 | * enqueue(taskA) | |
1198 | * // sched_uclamp_used gets enabled | |
1199 | * enqueue(taskB) | |
1200 | * dequeue(taskA) | |
1201 | * // Must not decrement bukcet->tasks here | |
1202 | * dequeue(taskB) | |
1203 | * | |
1204 | * where we could end up with stale data in uc_se and | |
1205 | * bucket[uc_se->bucket_id]. | |
1206 | * | |
1207 | * The following check here eliminates the possibility of such race. | |
1208 | */ | |
1209 | if (unlikely(!uc_se->active)) | |
1210 | return; | |
1211 | ||
69842cba | 1212 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1213 | |
69842cba PB |
1214 | SCHED_WARN_ON(!bucket->tasks); |
1215 | if (likely(bucket->tasks)) | |
1216 | bucket->tasks--; | |
46609ce2 | 1217 | |
e8f14172 | 1218 | uc_se->active = false; |
69842cba | 1219 | |
60daf9c1 PB |
1220 | /* |
1221 | * Keep "local max aggregation" simple and accept to (possibly) | |
1222 | * overboost some RUNNABLE tasks in the same bucket. | |
1223 | * The rq clamp bucket value is reset to its base value whenever | |
1224 | * there are no more RUNNABLE tasks refcounting it. | |
1225 | */ | |
69842cba PB |
1226 | if (likely(bucket->tasks)) |
1227 | return; | |
1228 | ||
1229 | rq_clamp = READ_ONCE(uc_rq->value); | |
1230 | /* | |
1231 | * Defensive programming: this should never happen. If it happens, | |
1232 | * e.g. due to future modification, warn and fixup the expected value. | |
1233 | */ | |
1234 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1235 | if (bucket->value >= rq_clamp) { |
1236 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1237 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1238 | } | |
69842cba PB |
1239 | } |
1240 | ||
1241 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1242 | { | |
0413d7f3 | 1243 | enum uclamp_id clamp_id; |
69842cba | 1244 | |
46609ce2 QY |
1245 | /* |
1246 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1247 | * | |
1248 | * The condition is constructed such that a NOP is generated when | |
1249 | * sched_uclamp_used is disabled. | |
1250 | */ | |
1251 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1252 | return; | |
1253 | ||
69842cba PB |
1254 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1255 | return; | |
1256 | ||
1257 | for_each_clamp_id(clamp_id) | |
1258 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1259 | |
1260 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1261 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1262 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1263 | } |
1264 | ||
1265 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1266 | { | |
0413d7f3 | 1267 | enum uclamp_id clamp_id; |
69842cba | 1268 | |
46609ce2 QY |
1269 | /* |
1270 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1271 | * | |
1272 | * The condition is constructed such that a NOP is generated when | |
1273 | * sched_uclamp_used is disabled. | |
1274 | */ | |
1275 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1276 | return; | |
1277 | ||
69842cba PB |
1278 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1279 | return; | |
1280 | ||
1281 | for_each_clamp_id(clamp_id) | |
1282 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1283 | } | |
1284 | ||
babbe170 | 1285 | static inline void |
0413d7f3 | 1286 | uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id) |
babbe170 PB |
1287 | { |
1288 | struct rq_flags rf; | |
1289 | struct rq *rq; | |
1290 | ||
1291 | /* | |
1292 | * Lock the task and the rq where the task is (or was) queued. | |
1293 | * | |
1294 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1295 | * price to pay to safely serialize util_{min,max} updates with | |
1296 | * enqueues, dequeues and migration operations. | |
1297 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1298 | */ | |
1299 | rq = task_rq_lock(p, &rf); | |
1300 | ||
1301 | /* | |
1302 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1303 | * If the task is not yet RUNNABLE and its task_struct is not | |
1304 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1305 | * it will already see the updated clamp bucket value. | |
1306 | */ | |
6e1ff077 | 1307 | if (p->uclamp[clamp_id].active) { |
babbe170 PB |
1308 | uclamp_rq_dec_id(rq, p, clamp_id); |
1309 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1310 | } | |
1311 | ||
1312 | task_rq_unlock(rq, p, &rf); | |
1313 | } | |
1314 | ||
e3b8b6a0 | 1315 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 PB |
1316 | static inline void |
1317 | uclamp_update_active_tasks(struct cgroup_subsys_state *css, | |
1318 | unsigned int clamps) | |
1319 | { | |
0413d7f3 | 1320 | enum uclamp_id clamp_id; |
babbe170 PB |
1321 | struct css_task_iter it; |
1322 | struct task_struct *p; | |
babbe170 PB |
1323 | |
1324 | css_task_iter_start(css, 0, &it); | |
1325 | while ((p = css_task_iter_next(&it))) { | |
1326 | for_each_clamp_id(clamp_id) { | |
1327 | if ((0x1 << clamp_id) & clamps) | |
1328 | uclamp_update_active(p, clamp_id); | |
1329 | } | |
1330 | } | |
1331 | css_task_iter_end(&it); | |
1332 | } | |
1333 | ||
7274a5c1 PB |
1334 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1335 | static void uclamp_update_root_tg(void) | |
1336 | { | |
1337 | struct task_group *tg = &root_task_group; | |
1338 | ||
1339 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1340 | sysctl_sched_uclamp_util_min, false); | |
1341 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1342 | sysctl_sched_uclamp_util_max, false); | |
1343 | ||
1344 | rcu_read_lock(); | |
1345 | cpu_util_update_eff(&root_task_group.css); | |
1346 | rcu_read_unlock(); | |
1347 | } | |
1348 | #else | |
1349 | static void uclamp_update_root_tg(void) { } | |
1350 | #endif | |
1351 | ||
e8f14172 | 1352 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1353 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1354 | { |
7274a5c1 | 1355 | bool update_root_tg = false; |
13685c4a | 1356 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1357 | int result; |
1358 | ||
2480c093 | 1359 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1360 | old_min = sysctl_sched_uclamp_util_min; |
1361 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1362 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1363 | |
1364 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1365 | if (result) | |
1366 | goto undo; | |
1367 | if (!write) | |
1368 | goto done; | |
1369 | ||
1370 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1371 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1372 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1373 | ||
e8f14172 PB |
1374 | result = -EINVAL; |
1375 | goto undo; | |
1376 | } | |
1377 | ||
1378 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1379 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1380 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1381 | update_root_tg = true; |
e8f14172 PB |
1382 | } |
1383 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1384 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1385 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1386 | update_root_tg = true; |
e8f14172 PB |
1387 | } |
1388 | ||
46609ce2 QY |
1389 | if (update_root_tg) { |
1390 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1391 | uclamp_update_root_tg(); |
46609ce2 | 1392 | } |
7274a5c1 | 1393 | |
13685c4a QY |
1394 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1395 | static_branch_enable(&sched_uclamp_used); | |
1396 | uclamp_sync_util_min_rt_default(); | |
1397 | } | |
7274a5c1 | 1398 | |
e8f14172 | 1399 | /* |
7274a5c1 PB |
1400 | * We update all RUNNABLE tasks only when task groups are in use. |
1401 | * Otherwise, keep it simple and do just a lazy update at each next | |
1402 | * task enqueue time. | |
e8f14172 | 1403 | */ |
7274a5c1 | 1404 | |
e8f14172 PB |
1405 | goto done; |
1406 | ||
1407 | undo: | |
1408 | sysctl_sched_uclamp_util_min = old_min; | |
1409 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1410 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1411 | done: |
2480c093 | 1412 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1413 | |
1414 | return result; | |
1415 | } | |
1416 | ||
a509a7cd PB |
1417 | static int uclamp_validate(struct task_struct *p, |
1418 | const struct sched_attr *attr) | |
1419 | { | |
1420 | unsigned int lower_bound = p->uclamp_req[UCLAMP_MIN].value; | |
1421 | unsigned int upper_bound = p->uclamp_req[UCLAMP_MAX].value; | |
1422 | ||
1423 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) | |
1424 | lower_bound = attr->sched_util_min; | |
1425 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) | |
1426 | upper_bound = attr->sched_util_max; | |
1427 | ||
1428 | if (lower_bound > upper_bound) | |
1429 | return -EINVAL; | |
1430 | if (upper_bound > SCHED_CAPACITY_SCALE) | |
1431 | return -EINVAL; | |
1432 | ||
e65855a5 QY |
1433 | /* |
1434 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1435 | * | |
1436 | * We need to do that here, because enabling static branches is a | |
1437 | * blocking operation which obviously cannot be done while holding | |
1438 | * scheduler locks. | |
1439 | */ | |
1440 | static_branch_enable(&sched_uclamp_used); | |
1441 | ||
a509a7cd PB |
1442 | return 0; |
1443 | } | |
1444 | ||
1445 | static void __setscheduler_uclamp(struct task_struct *p, | |
1446 | const struct sched_attr *attr) | |
1447 | { | |
0413d7f3 | 1448 | enum uclamp_id clamp_id; |
1a00d999 PB |
1449 | |
1450 | /* | |
1451 | * On scheduling class change, reset to default clamps for tasks | |
1452 | * without a task-specific value. | |
1453 | */ | |
1454 | for_each_clamp_id(clamp_id) { | |
1455 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
1a00d999 PB |
1456 | |
1457 | /* Keep using defined clamps across class changes */ | |
1458 | if (uc_se->user_defined) | |
1459 | continue; | |
1460 | ||
13685c4a QY |
1461 | /* |
1462 | * RT by default have a 100% boost value that could be modified | |
1463 | * at runtime. | |
1464 | */ | |
1a00d999 | 1465 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
13685c4a QY |
1466 | __uclamp_update_util_min_rt_default(p); |
1467 | else | |
1468 | uclamp_se_set(uc_se, uclamp_none(clamp_id), false); | |
1a00d999 | 1469 | |
1a00d999 PB |
1470 | } |
1471 | ||
a509a7cd PB |
1472 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1473 | return; | |
1474 | ||
1475 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { | |
1476 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], | |
1477 | attr->sched_util_min, true); | |
1478 | } | |
1479 | ||
1480 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1481 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], | |
1482 | attr->sched_util_max, true); | |
1483 | } | |
1484 | } | |
1485 | ||
e8f14172 PB |
1486 | static void uclamp_fork(struct task_struct *p) |
1487 | { | |
0413d7f3 | 1488 | enum uclamp_id clamp_id; |
e8f14172 | 1489 | |
13685c4a QY |
1490 | /* |
1491 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1492 | * as the task is still at its early fork stages. | |
1493 | */ | |
e8f14172 PB |
1494 | for_each_clamp_id(clamp_id) |
1495 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1496 | |
1497 | if (likely(!p->sched_reset_on_fork)) | |
1498 | return; | |
1499 | ||
1500 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1501 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1502 | uclamp_none(clamp_id), false); | |
a87498ac | 1503 | } |
e8f14172 PB |
1504 | } |
1505 | ||
13685c4a QY |
1506 | static void uclamp_post_fork(struct task_struct *p) |
1507 | { | |
1508 | uclamp_update_util_min_rt_default(p); | |
1509 | } | |
1510 | ||
d81ae8aa QY |
1511 | static void __init init_uclamp_rq(struct rq *rq) |
1512 | { | |
1513 | enum uclamp_id clamp_id; | |
1514 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1515 | ||
1516 | for_each_clamp_id(clamp_id) { | |
1517 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1518 | .value = uclamp_none(clamp_id) | |
1519 | }; | |
1520 | } | |
1521 | ||
1522 | rq->uclamp_flags = 0; | |
1523 | } | |
1524 | ||
69842cba PB |
1525 | static void __init init_uclamp(void) |
1526 | { | |
e8f14172 | 1527 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1528 | enum uclamp_id clamp_id; |
69842cba PB |
1529 | int cpu; |
1530 | ||
d81ae8aa QY |
1531 | for_each_possible_cpu(cpu) |
1532 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1533 | |
69842cba | 1534 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1535 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1536 | uclamp_none(clamp_id), false); |
69842cba | 1537 | } |
e8f14172 PB |
1538 | |
1539 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1540 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1541 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1542 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1543 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1544 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1545 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1546 | #endif |
1547 | } | |
69842cba PB |
1548 | } |
1549 | ||
1550 | #else /* CONFIG_UCLAMP_TASK */ | |
1551 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1552 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1553 | static inline int uclamp_validate(struct task_struct *p, |
1554 | const struct sched_attr *attr) | |
1555 | { | |
1556 | return -EOPNOTSUPP; | |
1557 | } | |
1558 | static void __setscheduler_uclamp(struct task_struct *p, | |
1559 | const struct sched_attr *attr) { } | |
e8f14172 | 1560 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1561 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1562 | static inline void init_uclamp(void) { } |
1563 | #endif /* CONFIG_UCLAMP_TASK */ | |
1564 | ||
1de64443 | 1565 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1566 | { |
0a67d1ee PZ |
1567 | if (!(flags & ENQUEUE_NOCLOCK)) |
1568 | update_rq_clock(rq); | |
1569 | ||
eb414681 | 1570 | if (!(flags & ENQUEUE_RESTORE)) { |
1de64443 | 1571 | sched_info_queued(rq, p); |
eb414681 JW |
1572 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1573 | } | |
0a67d1ee | 1574 | |
69842cba | 1575 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1576 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1577 | } |
1578 | ||
1de64443 | 1579 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1580 | { |
0a67d1ee PZ |
1581 | if (!(flags & DEQUEUE_NOCLOCK)) |
1582 | update_rq_clock(rq); | |
1583 | ||
eb414681 | 1584 | if (!(flags & DEQUEUE_SAVE)) { |
1de64443 | 1585 | sched_info_dequeued(rq, p); |
eb414681 JW |
1586 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1587 | } | |
0a67d1ee | 1588 | |
69842cba | 1589 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1590 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1591 | } |
1592 | ||
029632fb | 1593 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1594 | { |
371fd7e7 | 1595 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
1596 | |
1597 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
1598 | } |
1599 | ||
029632fb | 1600 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 1601 | { |
7dd77884 PZ |
1602 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
1603 | ||
371fd7e7 | 1604 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1605 | } |
1606 | ||
14531189 | 1607 | /* |
dd41f596 | 1608 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1609 | */ |
14531189 IM |
1610 | static inline int __normal_prio(struct task_struct *p) |
1611 | { | |
dd41f596 | 1612 | return p->static_prio; |
14531189 IM |
1613 | } |
1614 | ||
b29739f9 IM |
1615 | /* |
1616 | * Calculate the expected normal priority: i.e. priority | |
1617 | * without taking RT-inheritance into account. Might be | |
1618 | * boosted by interactivity modifiers. Changes upon fork, | |
1619 | * setprio syscalls, and whenever the interactivity | |
1620 | * estimator recalculates. | |
1621 | */ | |
36c8b586 | 1622 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1623 | { |
1624 | int prio; | |
1625 | ||
aab03e05 DF |
1626 | if (task_has_dl_policy(p)) |
1627 | prio = MAX_DL_PRIO-1; | |
1628 | else if (task_has_rt_policy(p)) | |
b29739f9 IM |
1629 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1630 | else | |
1631 | prio = __normal_prio(p); | |
1632 | return prio; | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * Calculate the current priority, i.e. the priority | |
1637 | * taken into account by the scheduler. This value might | |
1638 | * be boosted by RT tasks, or might be boosted by | |
1639 | * interactivity modifiers. Will be RT if the task got | |
1640 | * RT-boosted. If not then it returns p->normal_prio. | |
1641 | */ | |
36c8b586 | 1642 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1643 | { |
1644 | p->normal_prio = normal_prio(p); | |
1645 | /* | |
1646 | * If we are RT tasks or we were boosted to RT priority, | |
1647 | * keep the priority unchanged. Otherwise, update priority | |
1648 | * to the normal priority: | |
1649 | */ | |
1650 | if (!rt_prio(p->prio)) | |
1651 | return p->normal_prio; | |
1652 | return p->prio; | |
1653 | } | |
1654 | ||
1da177e4 LT |
1655 | /** |
1656 | * task_curr - is this task currently executing on a CPU? | |
1657 | * @p: the task in question. | |
e69f6186 YB |
1658 | * |
1659 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 1660 | */ |
36c8b586 | 1661 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1662 | { |
1663 | return cpu_curr(task_cpu(p)) == p; | |
1664 | } | |
1665 | ||
67dfa1b7 | 1666 | /* |
4c9a4bc8 PZ |
1667 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
1668 | * use the balance_callback list if you want balancing. | |
1669 | * | |
1670 | * this means any call to check_class_changed() must be followed by a call to | |
1671 | * balance_callback(). | |
67dfa1b7 | 1672 | */ |
cb469845 SR |
1673 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1674 | const struct sched_class *prev_class, | |
da7a735e | 1675 | int oldprio) |
cb469845 SR |
1676 | { |
1677 | if (prev_class != p->sched_class) { | |
1678 | if (prev_class->switched_from) | |
da7a735e | 1679 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 1680 | |
da7a735e | 1681 | p->sched_class->switched_to(rq, p); |
2d3d891d | 1682 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 1683 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
1684 | } |
1685 | ||
029632fb | 1686 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 1687 | { |
aa93cd53 | 1688 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 1689 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
1690 | else if (p->sched_class > rq->curr->sched_class) |
1691 | resched_curr(rq); | |
1e5a7405 PZ |
1692 | |
1693 | /* | |
1694 | * A queue event has occurred, and we're going to schedule. In | |
1695 | * this case, we can save a useless back to back clock update. | |
1696 | */ | |
da0c1e65 | 1697 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 1698 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
1699 | } |
1700 | ||
1da177e4 | 1701 | #ifdef CONFIG_SMP |
175f0e25 | 1702 | |
175f0e25 | 1703 | /* |
bee98539 | 1704 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
1705 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
1706 | */ | |
1707 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
1708 | { | |
3bd37062 | 1709 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
1710 | return false; |
1711 | ||
1712 | if (is_per_cpu_kthread(p)) | |
1713 | return cpu_online(cpu); | |
1714 | ||
1715 | return cpu_active(cpu); | |
1716 | } | |
1717 | ||
5cc389bc PZ |
1718 | /* |
1719 | * This is how migration works: | |
1720 | * | |
1721 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
1722 | * stop_one_cpu(). | |
1723 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
1724 | * off the CPU) | |
1725 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
1726 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1727 | * it and puts it into the right queue. | |
1728 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
1729 | * is done. | |
1730 | */ | |
1731 | ||
1732 | /* | |
1733 | * move_queued_task - move a queued task to new rq. | |
1734 | * | |
1735 | * Returns (locked) new rq. Old rq's lock is released. | |
1736 | */ | |
8a8c69c3 PZ |
1737 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
1738 | struct task_struct *p, int new_cpu) | |
5cc389bc | 1739 | { |
5cc389bc PZ |
1740 | lockdep_assert_held(&rq->lock); |
1741 | ||
58877d34 | 1742 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 1743 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 1744 | rq_unlock(rq, rf); |
5cc389bc PZ |
1745 | |
1746 | rq = cpu_rq(new_cpu); | |
1747 | ||
8a8c69c3 | 1748 | rq_lock(rq, rf); |
5cc389bc | 1749 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 1750 | activate_task(rq, p, 0); |
5cc389bc PZ |
1751 | check_preempt_curr(rq, p, 0); |
1752 | ||
1753 | return rq; | |
1754 | } | |
1755 | ||
1756 | struct migration_arg { | |
1757 | struct task_struct *task; | |
1758 | int dest_cpu; | |
1759 | }; | |
1760 | ||
1761 | /* | |
d1ccc66d | 1762 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
1763 | * this because either it can't run here any more (set_cpus_allowed() |
1764 | * away from this CPU, or CPU going down), or because we're | |
1765 | * attempting to rebalance this task on exec (sched_exec). | |
1766 | * | |
1767 | * So we race with normal scheduler movements, but that's OK, as long | |
1768 | * as the task is no longer on this CPU. | |
5cc389bc | 1769 | */ |
8a8c69c3 PZ |
1770 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
1771 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 1772 | { |
5cc389bc | 1773 | /* Affinity changed (again). */ |
175f0e25 | 1774 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 1775 | return rq; |
5cc389bc | 1776 | |
15ff991e | 1777 | update_rq_clock(rq); |
8a8c69c3 | 1778 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
1779 | |
1780 | return rq; | |
5cc389bc PZ |
1781 | } |
1782 | ||
1783 | /* | |
1784 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
1785 | * and performs thread migration by bumping thread off CPU then | |
1786 | * 'pushing' onto another runqueue. | |
1787 | */ | |
1788 | static int migration_cpu_stop(void *data) | |
1789 | { | |
1790 | struct migration_arg *arg = data; | |
5e16bbc2 PZ |
1791 | struct task_struct *p = arg->task; |
1792 | struct rq *rq = this_rq(); | |
8a8c69c3 | 1793 | struct rq_flags rf; |
5cc389bc PZ |
1794 | |
1795 | /* | |
d1ccc66d IM |
1796 | * The original target CPU might have gone down and we might |
1797 | * be on another CPU but it doesn't matter. | |
5cc389bc PZ |
1798 | */ |
1799 | local_irq_disable(); | |
1800 | /* | |
1801 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 1802 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
1803 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
1804 | */ | |
a1488664 | 1805 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
1806 | |
1807 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 1808 | rq_lock(rq, &rf); |
5e16bbc2 PZ |
1809 | /* |
1810 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
1811 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
1812 | * we're holding p->pi_lock. | |
1813 | */ | |
bf89a304 CC |
1814 | if (task_rq(p) == rq) { |
1815 | if (task_on_rq_queued(p)) | |
8a8c69c3 | 1816 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 CC |
1817 | else |
1818 | p->wake_cpu = arg->dest_cpu; | |
1819 | } | |
8a8c69c3 | 1820 | rq_unlock(rq, &rf); |
5e16bbc2 PZ |
1821 | raw_spin_unlock(&p->pi_lock); |
1822 | ||
5cc389bc PZ |
1823 | local_irq_enable(); |
1824 | return 0; | |
1825 | } | |
1826 | ||
c5b28038 PZ |
1827 | /* |
1828 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
1829 | * actually call this function. | |
1830 | */ | |
1831 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) | |
5cc389bc | 1832 | { |
3bd37062 | 1833 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
1834 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
1835 | } | |
1836 | ||
c5b28038 PZ |
1837 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
1838 | { | |
6c37067e PZ |
1839 | struct rq *rq = task_rq(p); |
1840 | bool queued, running; | |
1841 | ||
c5b28038 | 1842 | lockdep_assert_held(&p->pi_lock); |
6c37067e PZ |
1843 | |
1844 | queued = task_on_rq_queued(p); | |
1845 | running = task_current(rq, p); | |
1846 | ||
1847 | if (queued) { | |
1848 | /* | |
1849 | * Because __kthread_bind() calls this on blocked tasks without | |
1850 | * holding rq->lock. | |
1851 | */ | |
1852 | lockdep_assert_held(&rq->lock); | |
7a57f32a | 1853 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
1854 | } |
1855 | if (running) | |
1856 | put_prev_task(rq, p); | |
1857 | ||
c5b28038 | 1858 | p->sched_class->set_cpus_allowed(p, new_mask); |
6c37067e | 1859 | |
6c37067e | 1860 | if (queued) |
7134b3e9 | 1861 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 1862 | if (running) |
03b7fad1 | 1863 | set_next_task(rq, p); |
c5b28038 PZ |
1864 | } |
1865 | ||
5cc389bc PZ |
1866 | /* |
1867 | * Change a given task's CPU affinity. Migrate the thread to a | |
1868 | * proper CPU and schedule it away if the CPU it's executing on | |
1869 | * is removed from the allowed bitmask. | |
1870 | * | |
1871 | * NOTE: the caller must have a valid reference to the task, the | |
1872 | * task must not exit() & deallocate itself prematurely. The | |
1873 | * call is not atomic; no spinlocks may be held. | |
1874 | */ | |
25834c73 PZ |
1875 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
1876 | const struct cpumask *new_mask, bool check) | |
5cc389bc | 1877 | { |
e9d867a6 | 1878 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
5cc389bc | 1879 | unsigned int dest_cpu; |
eb580751 PZ |
1880 | struct rq_flags rf; |
1881 | struct rq *rq; | |
5cc389bc PZ |
1882 | int ret = 0; |
1883 | ||
eb580751 | 1884 | rq = task_rq_lock(p, &rf); |
a499c3ea | 1885 | update_rq_clock(rq); |
5cc389bc | 1886 | |
e9d867a6 PZI |
1887 | if (p->flags & PF_KTHREAD) { |
1888 | /* | |
1889 | * Kernel threads are allowed on online && !active CPUs | |
1890 | */ | |
1891 | cpu_valid_mask = cpu_online_mask; | |
1892 | } | |
1893 | ||
25834c73 PZ |
1894 | /* |
1895 | * Must re-check here, to close a race against __kthread_bind(), | |
1896 | * sched_setaffinity() is not guaranteed to observe the flag. | |
1897 | */ | |
1898 | if (check && (p->flags & PF_NO_SETAFFINITY)) { | |
1899 | ret = -EINVAL; | |
1900 | goto out; | |
1901 | } | |
1902 | ||
fd844ba9 | 1903 | if (cpumask_equal(&p->cpus_mask, new_mask)) |
5cc389bc PZ |
1904 | goto out; |
1905 | ||
46a87b38 PT |
1906 | /* |
1907 | * Picking a ~random cpu helps in cases where we are changing affinity | |
1908 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
1909 | * immediately required to distribute the tasks within their new mask. | |
1910 | */ | |
1911 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 1912 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
1913 | ret = -EINVAL; |
1914 | goto out; | |
1915 | } | |
1916 | ||
1917 | do_set_cpus_allowed(p, new_mask); | |
1918 | ||
e9d867a6 PZI |
1919 | if (p->flags & PF_KTHREAD) { |
1920 | /* | |
1921 | * For kernel threads that do indeed end up on online && | |
d1ccc66d | 1922 | * !active we want to ensure they are strict per-CPU threads. |
e9d867a6 PZI |
1923 | */ |
1924 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && | |
1925 | !cpumask_intersects(new_mask, cpu_active_mask) && | |
1926 | p->nr_cpus_allowed != 1); | |
1927 | } | |
1928 | ||
5cc389bc PZ |
1929 | /* Can the task run on the task's current CPU? If so, we're done */ |
1930 | if (cpumask_test_cpu(task_cpu(p), new_mask)) | |
1931 | goto out; | |
1932 | ||
5cc389bc PZ |
1933 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
1934 | struct migration_arg arg = { p, dest_cpu }; | |
1935 | /* Need help from migration thread: drop lock and wait. */ | |
eb580751 | 1936 | task_rq_unlock(rq, p, &rf); |
5cc389bc | 1937 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
5cc389bc | 1938 | return 0; |
cbce1a68 PZ |
1939 | } else if (task_on_rq_queued(p)) { |
1940 | /* | |
1941 | * OK, since we're going to drop the lock immediately | |
1942 | * afterwards anyway. | |
1943 | */ | |
8a8c69c3 | 1944 | rq = move_queued_task(rq, &rf, p, dest_cpu); |
cbce1a68 | 1945 | } |
5cc389bc | 1946 | out: |
eb580751 | 1947 | task_rq_unlock(rq, p, &rf); |
5cc389bc PZ |
1948 | |
1949 | return ret; | |
1950 | } | |
25834c73 PZ |
1951 | |
1952 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) | |
1953 | { | |
1954 | return __set_cpus_allowed_ptr(p, new_mask, false); | |
1955 | } | |
5cc389bc PZ |
1956 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1957 | ||
dd41f596 | 1958 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1959 | { |
e2912009 PZ |
1960 | #ifdef CONFIG_SCHED_DEBUG |
1961 | /* | |
1962 | * We should never call set_task_cpu() on a blocked task, | |
1963 | * ttwu() will sort out the placement. | |
1964 | */ | |
077614ee | 1965 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
e2336f6e | 1966 | !p->on_rq); |
0122ec5b | 1967 | |
3ea94de1 JP |
1968 | /* |
1969 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
1970 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
1971 | * time relying on p->on_rq. | |
1972 | */ | |
1973 | WARN_ON_ONCE(p->state == TASK_RUNNING && | |
1974 | p->sched_class == &fair_sched_class && | |
1975 | (p->on_rq && !task_on_rq_migrating(p))); | |
1976 | ||
0122ec5b | 1977 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
1978 | /* |
1979 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
1980 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
1981 | * | |
1982 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 1983 | * see task_group(). |
6c6c54e1 PZ |
1984 | * |
1985 | * Furthermore, all task_rq users should acquire both locks, see | |
1986 | * task_rq_lock(). | |
1987 | */ | |
0122ec5b PZ |
1988 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
1989 | lockdep_is_held(&task_rq(p)->lock))); | |
1990 | #endif | |
4ff9083b PZ |
1991 | /* |
1992 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
1993 | */ | |
1994 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
e2912009 PZ |
1995 | #endif |
1996 | ||
de1d7286 | 1997 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1998 | |
0c69774e | 1999 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 2000 | if (p->sched_class->migrate_task_rq) |
1327237a | 2001 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 2002 | p->se.nr_migrations++; |
d7822b1e | 2003 | rseq_migrate(p); |
ff303e66 | 2004 | perf_event_task_migrate(p); |
0c69774e | 2005 | } |
dd41f596 IM |
2006 | |
2007 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2008 | } |
2009 | ||
0ad4e3df | 2010 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
2011 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
2012 | { | |
da0c1e65 | 2013 | if (task_on_rq_queued(p)) { |
ac66f547 | 2014 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 2015 | struct rq_flags srf, drf; |
ac66f547 PZ |
2016 | |
2017 | src_rq = task_rq(p); | |
2018 | dst_rq = cpu_rq(cpu); | |
2019 | ||
8a8c69c3 PZ |
2020 | rq_pin_lock(src_rq, &srf); |
2021 | rq_pin_lock(dst_rq, &drf); | |
2022 | ||
ac66f547 PZ |
2023 | deactivate_task(src_rq, p, 0); |
2024 | set_task_cpu(p, cpu); | |
2025 | activate_task(dst_rq, p, 0); | |
2026 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
2027 | |
2028 | rq_unpin_lock(dst_rq, &drf); | |
2029 | rq_unpin_lock(src_rq, &srf); | |
2030 | ||
ac66f547 PZ |
2031 | } else { |
2032 | /* | |
2033 | * Task isn't running anymore; make it appear like we migrated | |
2034 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 2035 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
2036 | */ |
2037 | p->wake_cpu = cpu; | |
2038 | } | |
2039 | } | |
2040 | ||
2041 | struct migration_swap_arg { | |
2042 | struct task_struct *src_task, *dst_task; | |
2043 | int src_cpu, dst_cpu; | |
2044 | }; | |
2045 | ||
2046 | static int migrate_swap_stop(void *data) | |
2047 | { | |
2048 | struct migration_swap_arg *arg = data; | |
2049 | struct rq *src_rq, *dst_rq; | |
2050 | int ret = -EAGAIN; | |
2051 | ||
62694cd5 PZ |
2052 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
2053 | return -EAGAIN; | |
2054 | ||
ac66f547 PZ |
2055 | src_rq = cpu_rq(arg->src_cpu); |
2056 | dst_rq = cpu_rq(arg->dst_cpu); | |
2057 | ||
74602315 PZ |
2058 | double_raw_lock(&arg->src_task->pi_lock, |
2059 | &arg->dst_task->pi_lock); | |
ac66f547 | 2060 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 2061 | |
ac66f547 PZ |
2062 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
2063 | goto unlock; | |
2064 | ||
2065 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
2066 | goto unlock; | |
2067 | ||
3bd37062 | 2068 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
2069 | goto unlock; |
2070 | ||
3bd37062 | 2071 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
2072 | goto unlock; |
2073 | ||
2074 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
2075 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
2076 | ||
2077 | ret = 0; | |
2078 | ||
2079 | unlock: | |
2080 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
2081 | raw_spin_unlock(&arg->dst_task->pi_lock); |
2082 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
2083 | |
2084 | return ret; | |
2085 | } | |
2086 | ||
2087 | /* | |
2088 | * Cross migrate two tasks | |
2089 | */ | |
0ad4e3df SD |
2090 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
2091 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
2092 | { |
2093 | struct migration_swap_arg arg; | |
2094 | int ret = -EINVAL; | |
2095 | ||
ac66f547 PZ |
2096 | arg = (struct migration_swap_arg){ |
2097 | .src_task = cur, | |
0ad4e3df | 2098 | .src_cpu = curr_cpu, |
ac66f547 | 2099 | .dst_task = p, |
0ad4e3df | 2100 | .dst_cpu = target_cpu, |
ac66f547 PZ |
2101 | }; |
2102 | ||
2103 | if (arg.src_cpu == arg.dst_cpu) | |
2104 | goto out; | |
2105 | ||
6acce3ef PZ |
2106 | /* |
2107 | * These three tests are all lockless; this is OK since all of them | |
2108 | * will be re-checked with proper locks held further down the line. | |
2109 | */ | |
ac66f547 PZ |
2110 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
2111 | goto out; | |
2112 | ||
3bd37062 | 2113 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
2114 | goto out; |
2115 | ||
3bd37062 | 2116 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
2117 | goto out; |
2118 | ||
286549dc | 2119 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
2120 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
2121 | ||
2122 | out: | |
ac66f547 PZ |
2123 | return ret; |
2124 | } | |
0ad4e3df | 2125 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 2126 | |
1da177e4 LT |
2127 | /* |
2128 | * wait_task_inactive - wait for a thread to unschedule. | |
2129 | * | |
85ba2d86 RM |
2130 | * If @match_state is nonzero, it's the @p->state value just checked and |
2131 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2132 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2133 | * we return a positive number (its total switch count). If a second call | |
2134 | * a short while later returns the same number, the caller can be sure that | |
2135 | * @p has remained unscheduled the whole time. | |
2136 | * | |
1da177e4 LT |
2137 | * The caller must ensure that the task *will* unschedule sometime soon, |
2138 | * else this function might spin for a *long* time. This function can't | |
2139 | * be called with interrupts off, or it may introduce deadlock with | |
2140 | * smp_call_function() if an IPI is sent by the same process we are | |
2141 | * waiting to become inactive. | |
2142 | */ | |
85ba2d86 | 2143 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 | 2144 | { |
da0c1e65 | 2145 | int running, queued; |
eb580751 | 2146 | struct rq_flags rf; |
85ba2d86 | 2147 | unsigned long ncsw; |
70b97a7f | 2148 | struct rq *rq; |
1da177e4 | 2149 | |
3a5c359a AK |
2150 | for (;;) { |
2151 | /* | |
2152 | * We do the initial early heuristics without holding | |
2153 | * any task-queue locks at all. We'll only try to get | |
2154 | * the runqueue lock when things look like they will | |
2155 | * work out! | |
2156 | */ | |
2157 | rq = task_rq(p); | |
fa490cfd | 2158 | |
3a5c359a AK |
2159 | /* |
2160 | * If the task is actively running on another CPU | |
2161 | * still, just relax and busy-wait without holding | |
2162 | * any locks. | |
2163 | * | |
2164 | * NOTE! Since we don't hold any locks, it's not | |
2165 | * even sure that "rq" stays as the right runqueue! | |
2166 | * But we don't care, since "task_running()" will | |
2167 | * return false if the runqueue has changed and p | |
2168 | * is actually now running somewhere else! | |
2169 | */ | |
85ba2d86 RM |
2170 | while (task_running(rq, p)) { |
2171 | if (match_state && unlikely(p->state != match_state)) | |
2172 | return 0; | |
3a5c359a | 2173 | cpu_relax(); |
85ba2d86 | 2174 | } |
fa490cfd | 2175 | |
3a5c359a AK |
2176 | /* |
2177 | * Ok, time to look more closely! We need the rq | |
2178 | * lock now, to be *sure*. If we're wrong, we'll | |
2179 | * just go back and repeat. | |
2180 | */ | |
eb580751 | 2181 | rq = task_rq_lock(p, &rf); |
27a9da65 | 2182 | trace_sched_wait_task(p); |
3a5c359a | 2183 | running = task_running(rq, p); |
da0c1e65 | 2184 | queued = task_on_rq_queued(p); |
85ba2d86 | 2185 | ncsw = 0; |
f31e11d8 | 2186 | if (!match_state || p->state == match_state) |
93dcf55f | 2187 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 2188 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 2189 | |
85ba2d86 RM |
2190 | /* |
2191 | * If it changed from the expected state, bail out now. | |
2192 | */ | |
2193 | if (unlikely(!ncsw)) | |
2194 | break; | |
2195 | ||
3a5c359a AK |
2196 | /* |
2197 | * Was it really running after all now that we | |
2198 | * checked with the proper locks actually held? | |
2199 | * | |
2200 | * Oops. Go back and try again.. | |
2201 | */ | |
2202 | if (unlikely(running)) { | |
2203 | cpu_relax(); | |
2204 | continue; | |
2205 | } | |
fa490cfd | 2206 | |
3a5c359a AK |
2207 | /* |
2208 | * It's not enough that it's not actively running, | |
2209 | * it must be off the runqueue _entirely_, and not | |
2210 | * preempted! | |
2211 | * | |
80dd99b3 | 2212 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2213 | * running right now), it's preempted, and we should |
2214 | * yield - it could be a while. | |
2215 | */ | |
da0c1e65 | 2216 | if (unlikely(queued)) { |
8b0e1953 | 2217 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
2218 | |
2219 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2220 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2221 | continue; |
2222 | } | |
fa490cfd | 2223 | |
3a5c359a AK |
2224 | /* |
2225 | * Ahh, all good. It wasn't running, and it wasn't | |
2226 | * runnable, which means that it will never become | |
2227 | * running in the future either. We're all done! | |
2228 | */ | |
2229 | break; | |
2230 | } | |
85ba2d86 RM |
2231 | |
2232 | return ncsw; | |
1da177e4 LT |
2233 | } |
2234 | ||
2235 | /*** | |
2236 | * kick_process - kick a running thread to enter/exit the kernel | |
2237 | * @p: the to-be-kicked thread | |
2238 | * | |
2239 | * Cause a process which is running on another CPU to enter | |
2240 | * kernel-mode, without any delay. (to get signals handled.) | |
2241 | * | |
25985edc | 2242 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2243 | * because all it wants to ensure is that the remote task enters |
2244 | * the kernel. If the IPI races and the task has been migrated | |
2245 | * to another CPU then no harm is done and the purpose has been | |
2246 | * achieved as well. | |
2247 | */ | |
36c8b586 | 2248 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2249 | { |
2250 | int cpu; | |
2251 | ||
2252 | preempt_disable(); | |
2253 | cpu = task_cpu(p); | |
2254 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2255 | smp_send_reschedule(cpu); | |
2256 | preempt_enable(); | |
2257 | } | |
b43e3521 | 2258 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 2259 | |
30da688e | 2260 | /* |
3bd37062 | 2261 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
2262 | * |
2263 | * A few notes on cpu_active vs cpu_online: | |
2264 | * | |
2265 | * - cpu_active must be a subset of cpu_online | |
2266 | * | |
97fb7a0a | 2267 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 2268 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 2269 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
2270 | * see it. |
2271 | * | |
d1ccc66d | 2272 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 2273 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 2274 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
2275 | * off. |
2276 | * | |
2277 | * This means that fallback selection must not select !active CPUs. | |
2278 | * And can assume that any active CPU must be online. Conversely | |
2279 | * select_task_rq() below may allow selection of !active CPUs in order | |
2280 | * to satisfy the above rules. | |
30da688e | 2281 | */ |
5da9a0fb PZ |
2282 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2283 | { | |
aa00d89c TC |
2284 | int nid = cpu_to_node(cpu); |
2285 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
2286 | enum { cpuset, possible, fail } state = cpuset; |
2287 | int dest_cpu; | |
5da9a0fb | 2288 | |
aa00d89c | 2289 | /* |
d1ccc66d IM |
2290 | * If the node that the CPU is on has been offlined, cpu_to_node() |
2291 | * will return -1. There is no CPU on the node, and we should | |
2292 | * select the CPU on the other node. | |
aa00d89c TC |
2293 | */ |
2294 | if (nid != -1) { | |
2295 | nodemask = cpumask_of_node(nid); | |
2296 | ||
2297 | /* Look for allowed, online CPU in same node. */ | |
2298 | for_each_cpu(dest_cpu, nodemask) { | |
aa00d89c TC |
2299 | if (!cpu_active(dest_cpu)) |
2300 | continue; | |
3bd37062 | 2301 | if (cpumask_test_cpu(dest_cpu, p->cpus_ptr)) |
aa00d89c TC |
2302 | return dest_cpu; |
2303 | } | |
2baab4e9 | 2304 | } |
5da9a0fb | 2305 | |
2baab4e9 PZ |
2306 | for (;;) { |
2307 | /* Any allowed, online CPU? */ | |
3bd37062 | 2308 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 2309 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 2310 | continue; |
175f0e25 | 2311 | |
2baab4e9 PZ |
2312 | goto out; |
2313 | } | |
5da9a0fb | 2314 | |
e73e85f0 | 2315 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
2316 | switch (state) { |
2317 | case cpuset: | |
e73e85f0 ON |
2318 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
2319 | cpuset_cpus_allowed_fallback(p); | |
2320 | state = possible; | |
2321 | break; | |
2322 | } | |
df561f66 | 2323 | fallthrough; |
2baab4e9 PZ |
2324 | case possible: |
2325 | do_set_cpus_allowed(p, cpu_possible_mask); | |
2326 | state = fail; | |
2327 | break; | |
2328 | ||
2329 | case fail: | |
2330 | BUG(); | |
2331 | break; | |
2332 | } | |
2333 | } | |
2334 | ||
2335 | out: | |
2336 | if (state != cpuset) { | |
2337 | /* | |
2338 | * Don't tell them about moving exiting tasks or | |
2339 | * kernel threads (both mm NULL), since they never | |
2340 | * leave kernel. | |
2341 | */ | |
2342 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 2343 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
2344 | task_pid_nr(p), p->comm, cpu); |
2345 | } | |
5da9a0fb PZ |
2346 | } |
2347 | ||
2348 | return dest_cpu; | |
2349 | } | |
2350 | ||
e2912009 | 2351 | /* |
3bd37062 | 2352 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 2353 | */ |
970b13ba | 2354 | static inline |
ac66f547 | 2355 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
970b13ba | 2356 | { |
cbce1a68 PZ |
2357 | lockdep_assert_held(&p->pi_lock); |
2358 | ||
4b53a341 | 2359 | if (p->nr_cpus_allowed > 1) |
6c1d9410 | 2360 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
e9d867a6 | 2361 | else |
3bd37062 | 2362 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
2363 | |
2364 | /* | |
2365 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 2366 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 2367 | * CPU. |
e2912009 PZ |
2368 | * |
2369 | * Since this is common to all placement strategies, this lives here. | |
2370 | * | |
2371 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2372 | * not worry about this generic constraint ] | |
2373 | */ | |
7af443ee | 2374 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 2375 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2376 | |
2377 | return cpu; | |
970b13ba | 2378 | } |
09a40af5 | 2379 | |
f5832c19 NP |
2380 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2381 | { | |
2382 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2383 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2384 | ||
2385 | if (stop) { | |
2386 | /* | |
2387 | * Make it appear like a SCHED_FIFO task, its something | |
2388 | * userspace knows about and won't get confused about. | |
2389 | * | |
2390 | * Also, it will make PI more or less work without too | |
2391 | * much confusion -- but then, stop work should not | |
2392 | * rely on PI working anyway. | |
2393 | */ | |
2394 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2395 | ||
2396 | stop->sched_class = &stop_sched_class; | |
2397 | } | |
2398 | ||
2399 | cpu_rq(cpu)->stop = stop; | |
2400 | ||
2401 | if (old_stop) { | |
2402 | /* | |
2403 | * Reset it back to a normal scheduling class so that | |
2404 | * it can die in pieces. | |
2405 | */ | |
2406 | old_stop->sched_class = &rt_sched_class; | |
2407 | } | |
2408 | } | |
2409 | ||
25834c73 PZ |
2410 | #else |
2411 | ||
2412 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
2413 | const struct cpumask *new_mask, bool check) | |
2414 | { | |
2415 | return set_cpus_allowed_ptr(p, new_mask); | |
2416 | } | |
2417 | ||
5cc389bc | 2418 | #endif /* CONFIG_SMP */ |
970b13ba | 2419 | |
d7c01d27 | 2420 | static void |
b84cb5df | 2421 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2422 | { |
4fa8d299 | 2423 | struct rq *rq; |
b84cb5df | 2424 | |
4fa8d299 JP |
2425 | if (!schedstat_enabled()) |
2426 | return; | |
2427 | ||
2428 | rq = this_rq(); | |
d7c01d27 | 2429 | |
4fa8d299 JP |
2430 | #ifdef CONFIG_SMP |
2431 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
2432 | __schedstat_inc(rq->ttwu_local); |
2433 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
2434 | } else { |
2435 | struct sched_domain *sd; | |
2436 | ||
b85c8b71 | 2437 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 2438 | rcu_read_lock(); |
4fa8d299 | 2439 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 2440 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 2441 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
2442 | break; |
2443 | } | |
2444 | } | |
057f3fad | 2445 | rcu_read_unlock(); |
d7c01d27 | 2446 | } |
f339b9dc PZ |
2447 | |
2448 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 2449 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
2450 | #endif /* CONFIG_SMP */ |
2451 | ||
b85c8b71 PZ |
2452 | __schedstat_inc(rq->ttwu_count); |
2453 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
2454 | |
2455 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 2456 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2457 | } |
2458 | ||
23f41eeb PZ |
2459 | /* |
2460 | * Mark the task runnable and perform wakeup-preemption. | |
2461 | */ | |
e7904a28 | 2462 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2463 | struct rq_flags *rf) |
9ed3811a | 2464 | { |
9ed3811a | 2465 | check_preempt_curr(rq, p, wake_flags); |
9ed3811a | 2466 | p->state = TASK_RUNNING; |
fbd705a0 PZ |
2467 | trace_sched_wakeup(p); |
2468 | ||
9ed3811a | 2469 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
2470 | if (p->sched_class->task_woken) { |
2471 | /* | |
cbce1a68 PZ |
2472 | * Our task @p is fully woken up and running; so its safe to |
2473 | * drop the rq->lock, hereafter rq is only used for statistics. | |
4c9a4bc8 | 2474 | */ |
d8ac8971 | 2475 | rq_unpin_lock(rq, rf); |
9ed3811a | 2476 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 2477 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 2478 | } |
9ed3811a | 2479 | |
e69c6341 | 2480 | if (rq->idle_stamp) { |
78becc27 | 2481 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 2482 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 2483 | |
abfafa54 JL |
2484 | update_avg(&rq->avg_idle, delta); |
2485 | ||
2486 | if (rq->avg_idle > max) | |
9ed3811a | 2487 | rq->avg_idle = max; |
abfafa54 | 2488 | |
9ed3811a TH |
2489 | rq->idle_stamp = 0; |
2490 | } | |
2491 | #endif | |
2492 | } | |
2493 | ||
c05fbafb | 2494 | static void |
e7904a28 | 2495 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 2496 | struct rq_flags *rf) |
c05fbafb | 2497 | { |
77558e4d | 2498 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 2499 | |
cbce1a68 PZ |
2500 | lockdep_assert_held(&rq->lock); |
2501 | ||
c05fbafb PZ |
2502 | if (p->sched_contributes_to_load) |
2503 | rq->nr_uninterruptible--; | |
b5179ac7 | 2504 | |
dbfb089d | 2505 | #ifdef CONFIG_SMP |
b5179ac7 | 2506 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 2507 | en_flags |= ENQUEUE_MIGRATED; |
c05fbafb PZ |
2508 | #endif |
2509 | ||
1b174a2c | 2510 | activate_task(rq, p, en_flags); |
d8ac8971 | 2511 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
2512 | } |
2513 | ||
2514 | /* | |
58877d34 PZ |
2515 | * Consider @p being inside a wait loop: |
2516 | * | |
2517 | * for (;;) { | |
2518 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
2519 | * | |
2520 | * if (CONDITION) | |
2521 | * break; | |
2522 | * | |
2523 | * schedule(); | |
2524 | * } | |
2525 | * __set_current_state(TASK_RUNNING); | |
2526 | * | |
2527 | * between set_current_state() and schedule(). In this case @p is still | |
2528 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
2529 | * an atomic manner. | |
2530 | * | |
2531 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
2532 | * then schedule() must still happen and p->state can be changed to | |
2533 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
2534 | * need to do a full wakeup with enqueue. | |
2535 | * | |
2536 | * Returns: %true when the wakeup is done, | |
2537 | * %false otherwise. | |
c05fbafb | 2538 | */ |
58877d34 | 2539 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 2540 | { |
eb580751 | 2541 | struct rq_flags rf; |
c05fbafb PZ |
2542 | struct rq *rq; |
2543 | int ret = 0; | |
2544 | ||
eb580751 | 2545 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 2546 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
2547 | /* check_preempt_curr() may use rq clock */ |
2548 | update_rq_clock(rq); | |
d8ac8971 | 2549 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
2550 | ret = 1; |
2551 | } | |
eb580751 | 2552 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
2553 | |
2554 | return ret; | |
2555 | } | |
2556 | ||
317f3941 | 2557 | #ifdef CONFIG_SMP |
a1488664 | 2558 | void sched_ttwu_pending(void *arg) |
317f3941 | 2559 | { |
a1488664 | 2560 | struct llist_node *llist = arg; |
317f3941 | 2561 | struct rq *rq = this_rq(); |
73215849 | 2562 | struct task_struct *p, *t; |
d8ac8971 | 2563 | struct rq_flags rf; |
317f3941 | 2564 | |
e3baac47 PZ |
2565 | if (!llist) |
2566 | return; | |
2567 | ||
126c2092 PZ |
2568 | /* |
2569 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
2570 | * Races such that false-negatives are possible, since they | |
2571 | * are shorter lived that false-positives would be. | |
2572 | */ | |
2573 | WRITE_ONCE(rq->ttwu_pending, 0); | |
2574 | ||
8a8c69c3 | 2575 | rq_lock_irqsave(rq, &rf); |
77558e4d | 2576 | update_rq_clock(rq); |
317f3941 | 2577 | |
8c4890d1 | 2578 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
2579 | if (WARN_ON_ONCE(p->on_cpu)) |
2580 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
2581 | ||
2582 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
2583 | set_task_cpu(p, cpu_of(rq)); | |
2584 | ||
73215849 | 2585 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 2586 | } |
317f3941 | 2587 | |
8a8c69c3 | 2588 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
2589 | } |
2590 | ||
b2a02fc4 | 2591 | void send_call_function_single_ipi(int cpu) |
317f3941 | 2592 | { |
b2a02fc4 | 2593 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 2594 | |
b2a02fc4 PZ |
2595 | if (!set_nr_if_polling(rq->idle)) |
2596 | arch_send_call_function_single_ipi(cpu); | |
2597 | else | |
2598 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
2599 | } |
2600 | ||
2ebb1771 MG |
2601 | /* |
2602 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
2603 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
2604 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
2605 | * of the wakeup instead of the waker. | |
2606 | */ | |
2607 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 2608 | { |
e3baac47 PZ |
2609 | struct rq *rq = cpu_rq(cpu); |
2610 | ||
b7e7ade3 PZ |
2611 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
2612 | ||
126c2092 | 2613 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 2614 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 2615 | } |
d6aa8f85 | 2616 | |
f6be8af1 CL |
2617 | void wake_up_if_idle(int cpu) |
2618 | { | |
2619 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 2620 | struct rq_flags rf; |
f6be8af1 | 2621 | |
fd7de1e8 AL |
2622 | rcu_read_lock(); |
2623 | ||
2624 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
2625 | goto out; | |
f6be8af1 CL |
2626 | |
2627 | if (set_nr_if_polling(rq->idle)) { | |
2628 | trace_sched_wake_idle_without_ipi(cpu); | |
2629 | } else { | |
8a8c69c3 | 2630 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
2631 | if (is_idle_task(rq->curr)) |
2632 | smp_send_reschedule(cpu); | |
d1ccc66d | 2633 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 2634 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 2635 | } |
fd7de1e8 AL |
2636 | |
2637 | out: | |
2638 | rcu_read_unlock(); | |
f6be8af1 CL |
2639 | } |
2640 | ||
39be3501 | 2641 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
2642 | { |
2643 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
2644 | } | |
c6e7bd7a | 2645 | |
2ebb1771 MG |
2646 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
2647 | { | |
2648 | /* | |
2649 | * If the CPU does not share cache, then queue the task on the | |
2650 | * remote rqs wakelist to avoid accessing remote data. | |
2651 | */ | |
2652 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
2653 | return true; | |
2654 | ||
2655 | /* | |
2656 | * If the task is descheduling and the only running task on the | |
2657 | * CPU then use the wakelist to offload the task activation to | |
2658 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
2659 | * nr_running is checked to avoid unnecessary task stacking. | |
2660 | */ | |
739f70b4 | 2661 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
2662 | return true; |
2663 | ||
2664 | return false; | |
2665 | } | |
2666 | ||
2667 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 2668 | { |
2ebb1771 | 2669 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
2670 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
2671 | return false; | |
2672 | ||
c6e7bd7a | 2673 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 2674 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
2675 | return true; |
2676 | } | |
2677 | ||
2678 | return false; | |
2679 | } | |
58877d34 PZ |
2680 | |
2681 | #else /* !CONFIG_SMP */ | |
2682 | ||
2683 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
2684 | { | |
2685 | return false; | |
2686 | } | |
2687 | ||
d6aa8f85 | 2688 | #endif /* CONFIG_SMP */ |
317f3941 | 2689 | |
b5179ac7 | 2690 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
2691 | { |
2692 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 2693 | struct rq_flags rf; |
c05fbafb | 2694 | |
2ebb1771 | 2695 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 2696 | return; |
317f3941 | 2697 | |
8a8c69c3 | 2698 | rq_lock(rq, &rf); |
77558e4d | 2699 | update_rq_clock(rq); |
d8ac8971 | 2700 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 2701 | rq_unlock(rq, &rf); |
9ed3811a TH |
2702 | } |
2703 | ||
8643cda5 PZ |
2704 | /* |
2705 | * Notes on Program-Order guarantees on SMP systems. | |
2706 | * | |
2707 | * MIGRATION | |
2708 | * | |
2709 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
2710 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
2711 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
2712 | * |
2713 | * For migration (of runnable tasks) this is provided by the following means: | |
2714 | * | |
2715 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
2716 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
2717 | * rq(c1)->lock (if not at the same time, then in that order). | |
2718 | * C) LOCK of the rq(c1)->lock scheduling in task | |
2719 | * | |
7696f991 | 2720 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 2721 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
2722 | * |
2723 | * Example: | |
2724 | * | |
2725 | * CPU0 CPU1 CPU2 | |
2726 | * | |
2727 | * LOCK rq(0)->lock | |
2728 | * sched-out X | |
2729 | * sched-in Y | |
2730 | * UNLOCK rq(0)->lock | |
2731 | * | |
2732 | * LOCK rq(0)->lock // orders against CPU0 | |
2733 | * dequeue X | |
2734 | * UNLOCK rq(0)->lock | |
2735 | * | |
2736 | * LOCK rq(1)->lock | |
2737 | * enqueue X | |
2738 | * UNLOCK rq(1)->lock | |
2739 | * | |
2740 | * LOCK rq(1)->lock // orders against CPU2 | |
2741 | * sched-out Z | |
2742 | * sched-in X | |
2743 | * UNLOCK rq(1)->lock | |
2744 | * | |
2745 | * | |
2746 | * BLOCKING -- aka. SLEEP + WAKEUP | |
2747 | * | |
2748 | * For blocking we (obviously) need to provide the same guarantee as for | |
2749 | * migration. However the means are completely different as there is no lock | |
2750 | * chain to provide order. Instead we do: | |
2751 | * | |
58877d34 PZ |
2752 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
2753 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
2754 | * |
2755 | * Example: | |
2756 | * | |
2757 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
2758 | * | |
2759 | * LOCK rq(0)->lock LOCK X->pi_lock | |
2760 | * dequeue X | |
2761 | * sched-out X | |
2762 | * smp_store_release(X->on_cpu, 0); | |
2763 | * | |
1f03e8d2 | 2764 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
2765 | * X->state = WAKING |
2766 | * set_task_cpu(X,2) | |
2767 | * | |
2768 | * LOCK rq(2)->lock | |
2769 | * enqueue X | |
2770 | * X->state = RUNNING | |
2771 | * UNLOCK rq(2)->lock | |
2772 | * | |
2773 | * LOCK rq(2)->lock // orders against CPU1 | |
2774 | * sched-out Z | |
2775 | * sched-in X | |
2776 | * UNLOCK rq(2)->lock | |
2777 | * | |
2778 | * UNLOCK X->pi_lock | |
2779 | * UNLOCK rq(0)->lock | |
2780 | * | |
2781 | * | |
7696f991 AP |
2782 | * However, for wakeups there is a second guarantee we must provide, namely we |
2783 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
2784 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
2785 | */ |
2786 | ||
9ed3811a | 2787 | /** |
1da177e4 | 2788 | * try_to_wake_up - wake up a thread |
9ed3811a | 2789 | * @p: the thread to be awakened |
1da177e4 | 2790 | * @state: the mask of task states that can be woken |
9ed3811a | 2791 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 2792 | * |
58877d34 PZ |
2793 | * Conceptually does: |
2794 | * | |
2795 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 2796 | * |
a2250238 PZ |
2797 | * If the task was not queued/runnable, also place it back on a runqueue. |
2798 | * | |
58877d34 PZ |
2799 | * This function is atomic against schedule() which would dequeue the task. |
2800 | * | |
2801 | * It issues a full memory barrier before accessing @p->state, see the comment | |
2802 | * with set_current_state(). | |
a2250238 | 2803 | * |
58877d34 | 2804 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 2805 | * |
58877d34 PZ |
2806 | * Relies on p->pi_lock stabilizing: |
2807 | * - p->sched_class | |
2808 | * - p->cpus_ptr | |
2809 | * - p->sched_task_group | |
2810 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
2811 | * | |
2812 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
2813 | * Takes rq->lock in: | |
2814 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
2815 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
2816 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
2817 | * | |
2818 | * As a consequence we race really badly with just about everything. See the | |
2819 | * many memory barriers and their comments for details. | |
7696f991 | 2820 | * |
a2250238 PZ |
2821 | * Return: %true if @p->state changes (an actual wakeup was done), |
2822 | * %false otherwise. | |
1da177e4 | 2823 | */ |
e4a52bcb PZ |
2824 | static int |
2825 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2826 | { |
1da177e4 | 2827 | unsigned long flags; |
c05fbafb | 2828 | int cpu, success = 0; |
2398f2c6 | 2829 | |
e3d85487 | 2830 | preempt_disable(); |
aacedf26 PZ |
2831 | if (p == current) { |
2832 | /* | |
2833 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
2834 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 2835 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
2836 | * without taking any locks. |
2837 | * | |
2838 | * In particular: | |
2839 | * - we rely on Program-Order guarantees for all the ordering, | |
2840 | * - we're serialized against set_special_state() by virtue of | |
2841 | * it disabling IRQs (this allows not taking ->pi_lock). | |
2842 | */ | |
2843 | if (!(p->state & state)) | |
e3d85487 | 2844 | goto out; |
aacedf26 PZ |
2845 | |
2846 | success = 1; | |
aacedf26 PZ |
2847 | trace_sched_waking(p); |
2848 | p->state = TASK_RUNNING; | |
2849 | trace_sched_wakeup(p); | |
2850 | goto out; | |
2851 | } | |
2852 | ||
e0acd0a6 ON |
2853 | /* |
2854 | * If we are going to wake up a thread waiting for CONDITION we | |
2855 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
2856 | * reordered with p->state check below. This pairs with smp_store_mb() |
2857 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 2858 | */ |
013fdb80 | 2859 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 2860 | smp_mb__after_spinlock(); |
e9c84311 | 2861 | if (!(p->state & state)) |
aacedf26 | 2862 | goto unlock; |
1da177e4 | 2863 | |
fbd705a0 PZ |
2864 | trace_sched_waking(p); |
2865 | ||
d1ccc66d IM |
2866 | /* We're going to change ->state: */ |
2867 | success = 1; | |
1da177e4 | 2868 | |
135e8c92 BS |
2869 | /* |
2870 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
2871 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
2872 | * in smp_cond_load_acquire() below. | |
2873 | * | |
3d85b270 AP |
2874 | * sched_ttwu_pending() try_to_wake_up() |
2875 | * STORE p->on_rq = 1 LOAD p->state | |
2876 | * UNLOCK rq->lock | |
2877 | * | |
2878 | * __schedule() (switch to task 'p') | |
2879 | * LOCK rq->lock smp_rmb(); | |
2880 | * smp_mb__after_spinlock(); | |
2881 | * UNLOCK rq->lock | |
135e8c92 BS |
2882 | * |
2883 | * [task p] | |
3d85b270 | 2884 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 2885 | * |
3d85b270 AP |
2886 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
2887 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
2888 | * |
2889 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
2890 | */ |
2891 | smp_rmb(); | |
58877d34 | 2892 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 2893 | goto unlock; |
1da177e4 | 2894 | |
c6e7bd7a PZ |
2895 | if (p->in_iowait) { |
2896 | delayacct_blkio_end(p); | |
2897 | atomic_dec(&task_rq(p)->nr_iowait); | |
2898 | } | |
2899 | ||
1da177e4 | 2900 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
2901 | /* |
2902 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
2903 | * possible to, falsely, observe p->on_cpu == 0. | |
2904 | * | |
2905 | * One must be running (->on_cpu == 1) in order to remove oneself | |
2906 | * from the runqueue. | |
2907 | * | |
3d85b270 AP |
2908 | * __schedule() (switch to task 'p') try_to_wake_up() |
2909 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
2910 | * UNLOCK rq->lock | |
2911 | * | |
2912 | * __schedule() (put 'p' to sleep) | |
2913 | * LOCK rq->lock smp_rmb(); | |
2914 | * smp_mb__after_spinlock(); | |
2915 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 2916 | * |
3d85b270 AP |
2917 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
2918 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
2919 | * |
2920 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
2921 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
2922 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 2923 | */ |
dbfb089d PZ |
2924 | smp_acquire__after_ctrl_dep(); |
2925 | ||
2926 | /* | |
2927 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
2928 | * == 0), which means we need to do an enqueue, change p->state to | |
2929 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
2930 | * enqueue, such as ttwu_queue_wakelist(). | |
2931 | */ | |
2932 | p->state = TASK_WAKING; | |
ecf7d01c | 2933 | |
c6e7bd7a PZ |
2934 | /* |
2935 | * If the owning (remote) CPU is still in the middle of schedule() with | |
2936 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
2937 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
2938 | * let the waker make forward progress. This is safe because IRQs are | |
2939 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
2940 | * |
2941 | * Ensure we load task_cpu(p) after p->on_cpu: | |
2942 | * | |
2943 | * set_task_cpu(p, cpu); | |
2944 | * STORE p->cpu = @cpu | |
2945 | * __schedule() (switch to task 'p') | |
2946 | * LOCK rq->lock | |
2947 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
2948 | * STORE p->on_cpu = 1 LOAD p->cpu | |
2949 | * | |
2950 | * to ensure we observe the correct CPU on which the task is currently | |
2951 | * scheduling. | |
c6e7bd7a | 2952 | */ |
b6e13e85 | 2953 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 2954 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
2955 | goto unlock; |
2956 | ||
e9c84311 | 2957 | /* |
d1ccc66d | 2958 | * If the owning (remote) CPU is still in the middle of schedule() with |
c05fbafb | 2959 | * this task as prev, wait until its done referencing the task. |
b75a2253 | 2960 | * |
31cb1bc0 | 2961 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
2962 | * |
2963 | * This ensures that tasks getting woken will be fully ordered against | |
2964 | * their previous state and preserve Program Order. | |
0970d299 | 2965 | */ |
1f03e8d2 | 2966 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 2967 | |
ac66f547 | 2968 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
f339b9dc PZ |
2969 | if (task_cpu(p) != cpu) { |
2970 | wake_flags |= WF_MIGRATED; | |
eb414681 | 2971 | psi_ttwu_dequeue(p); |
e4a52bcb | 2972 | set_task_cpu(p, cpu); |
f339b9dc | 2973 | } |
b6e13e85 PZ |
2974 | #else |
2975 | cpu = task_cpu(p); | |
1da177e4 | 2976 | #endif /* CONFIG_SMP */ |
1da177e4 | 2977 | |
b5179ac7 | 2978 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 2979 | unlock: |
013fdb80 | 2980 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
2981 | out: |
2982 | if (success) | |
b6e13e85 | 2983 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 2984 | preempt_enable(); |
1da177e4 LT |
2985 | |
2986 | return success; | |
2987 | } | |
2988 | ||
2beaf328 PM |
2989 | /** |
2990 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
2991 | * @p: Process for which the function is to be invoked. | |
2992 | * @func: Function to invoke. | |
2993 | * @arg: Argument to function. | |
2994 | * | |
2995 | * If the specified task can be quickly locked into a definite state | |
2996 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
2997 | * state while invoking @func(@arg). This function can use ->on_rq and | |
2998 | * task_curr() to work out what the state is, if required. Given that | |
2999 | * @func can be invoked with a runqueue lock held, it had better be quite | |
3000 | * lightweight. | |
3001 | * | |
3002 | * Returns: | |
3003 | * @false if the task slipped out from under the locks. | |
3004 | * @true if the task was locked onto a runqueue or is sleeping. | |
3005 | * However, @func can override this by returning @false. | |
3006 | */ | |
3007 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
3008 | { | |
3009 | bool ret = false; | |
3010 | struct rq_flags rf; | |
3011 | struct rq *rq; | |
3012 | ||
3013 | lockdep_assert_irqs_enabled(); | |
3014 | raw_spin_lock_irq(&p->pi_lock); | |
3015 | if (p->on_rq) { | |
3016 | rq = __task_rq_lock(p, &rf); | |
3017 | if (task_rq(p) == rq) | |
3018 | ret = func(p, arg); | |
3019 | rq_unlock(rq, &rf); | |
3020 | } else { | |
3021 | switch (p->state) { | |
3022 | case TASK_RUNNING: | |
3023 | case TASK_WAKING: | |
3024 | break; | |
3025 | default: | |
3026 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
3027 | if (!p->on_rq) | |
3028 | ret = func(p, arg); | |
3029 | } | |
3030 | } | |
3031 | raw_spin_unlock_irq(&p->pi_lock); | |
3032 | return ret; | |
3033 | } | |
3034 | ||
50fa610a DH |
3035 | /** |
3036 | * wake_up_process - Wake up a specific process | |
3037 | * @p: The process to be woken up. | |
3038 | * | |
3039 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
3040 | * processes. |
3041 | * | |
3042 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 3043 | * |
7696f991 | 3044 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 3045 | */ |
7ad5b3a5 | 3046 | int wake_up_process(struct task_struct *p) |
1da177e4 | 3047 | { |
9067ac85 | 3048 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 3049 | } |
1da177e4 LT |
3050 | EXPORT_SYMBOL(wake_up_process); |
3051 | ||
7ad5b3a5 | 3052 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
3053 | { |
3054 | return try_to_wake_up(p, state, 0); | |
3055 | } | |
3056 | ||
1da177e4 LT |
3057 | /* |
3058 | * Perform scheduler related setup for a newly forked process p. | |
3059 | * p is forked by current. | |
dd41f596 IM |
3060 | * |
3061 | * __sched_fork() is basic setup used by init_idle() too: | |
3062 | */ | |
5e1576ed | 3063 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3064 | { |
fd2f4419 PZ |
3065 | p->on_rq = 0; |
3066 | ||
3067 | p->se.on_rq = 0; | |
dd41f596 IM |
3068 | p->se.exec_start = 0; |
3069 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 3070 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 3071 | p->se.nr_migrations = 0; |
da7a735e | 3072 | p->se.vruntime = 0; |
fd2f4419 | 3073 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 3074 | |
ad936d86 BP |
3075 | #ifdef CONFIG_FAIR_GROUP_SCHED |
3076 | p->se.cfs_rq = NULL; | |
3077 | #endif | |
3078 | ||
6cfb0d5d | 3079 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 3080 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 3081 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 3082 | #endif |
476d139c | 3083 | |
aab03e05 | 3084 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 3085 | init_dl_task_timer(&p->dl); |
209a0cbd | 3086 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 3087 | __dl_clear_params(p); |
aab03e05 | 3088 | |
fa717060 | 3089 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
3090 | p->rt.timeout = 0; |
3091 | p->rt.time_slice = sched_rr_timeslice; | |
3092 | p->rt.on_rq = 0; | |
3093 | p->rt.on_list = 0; | |
476d139c | 3094 | |
e107be36 AK |
3095 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3096 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
3097 | #endif | |
cbee9f88 | 3098 | |
5e1f0f09 MG |
3099 | #ifdef CONFIG_COMPACTION |
3100 | p->capture_control = NULL; | |
3101 | #endif | |
13784475 | 3102 | init_numa_balancing(clone_flags, p); |
a1488664 | 3103 | #ifdef CONFIG_SMP |
8c4890d1 | 3104 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
a1488664 | 3105 | #endif |
dd41f596 IM |
3106 | } |
3107 | ||
2a595721 SD |
3108 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
3109 | ||
1a687c2e | 3110 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 3111 | |
1a687c2e MG |
3112 | void set_numabalancing_state(bool enabled) |
3113 | { | |
3114 | if (enabled) | |
2a595721 | 3115 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 3116 | else |
2a595721 | 3117 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 3118 | } |
54a43d54 AK |
3119 | |
3120 | #ifdef CONFIG_PROC_SYSCTL | |
3121 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 3122 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
3123 | { |
3124 | struct ctl_table t; | |
3125 | int err; | |
2a595721 | 3126 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
3127 | |
3128 | if (write && !capable(CAP_SYS_ADMIN)) | |
3129 | return -EPERM; | |
3130 | ||
3131 | t = *table; | |
3132 | t.data = &state; | |
3133 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3134 | if (err < 0) | |
3135 | return err; | |
3136 | if (write) | |
3137 | set_numabalancing_state(state); | |
3138 | return err; | |
3139 | } | |
3140 | #endif | |
3141 | #endif | |
dd41f596 | 3142 | |
4698f88c JP |
3143 | #ifdef CONFIG_SCHEDSTATS |
3144 | ||
cb251765 | 3145 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4698f88c | 3146 | static bool __initdata __sched_schedstats = false; |
cb251765 | 3147 | |
cb251765 MG |
3148 | static void set_schedstats(bool enabled) |
3149 | { | |
3150 | if (enabled) | |
3151 | static_branch_enable(&sched_schedstats); | |
3152 | else | |
3153 | static_branch_disable(&sched_schedstats); | |
3154 | } | |
3155 | ||
3156 | void force_schedstat_enabled(void) | |
3157 | { | |
3158 | if (!schedstat_enabled()) { | |
3159 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
3160 | static_branch_enable(&sched_schedstats); | |
3161 | } | |
3162 | } | |
3163 | ||
3164 | static int __init setup_schedstats(char *str) | |
3165 | { | |
3166 | int ret = 0; | |
3167 | if (!str) | |
3168 | goto out; | |
3169 | ||
4698f88c JP |
3170 | /* |
3171 | * This code is called before jump labels have been set up, so we can't | |
3172 | * change the static branch directly just yet. Instead set a temporary | |
3173 | * variable so init_schedstats() can do it later. | |
3174 | */ | |
cb251765 | 3175 | if (!strcmp(str, "enable")) { |
4698f88c | 3176 | __sched_schedstats = true; |
cb251765 MG |
3177 | ret = 1; |
3178 | } else if (!strcmp(str, "disable")) { | |
4698f88c | 3179 | __sched_schedstats = false; |
cb251765 MG |
3180 | ret = 1; |
3181 | } | |
3182 | out: | |
3183 | if (!ret) | |
3184 | pr_warn("Unable to parse schedstats=\n"); | |
3185 | ||
3186 | return ret; | |
3187 | } | |
3188 | __setup("schedstats=", setup_schedstats); | |
3189 | ||
4698f88c JP |
3190 | static void __init init_schedstats(void) |
3191 | { | |
3192 | set_schedstats(__sched_schedstats); | |
3193 | } | |
3194 | ||
cb251765 | 3195 | #ifdef CONFIG_PROC_SYSCTL |
32927393 CH |
3196 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
3197 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
3198 | { |
3199 | struct ctl_table t; | |
3200 | int err; | |
3201 | int state = static_branch_likely(&sched_schedstats); | |
3202 | ||
3203 | if (write && !capable(CAP_SYS_ADMIN)) | |
3204 | return -EPERM; | |
3205 | ||
3206 | t = *table; | |
3207 | t.data = &state; | |
3208 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
3209 | if (err < 0) | |
3210 | return err; | |
3211 | if (write) | |
3212 | set_schedstats(state); | |
3213 | return err; | |
3214 | } | |
4698f88c JP |
3215 | #endif /* CONFIG_PROC_SYSCTL */ |
3216 | #else /* !CONFIG_SCHEDSTATS */ | |
3217 | static inline void init_schedstats(void) {} | |
3218 | #endif /* CONFIG_SCHEDSTATS */ | |
dd41f596 IM |
3219 | |
3220 | /* | |
3221 | * fork()/clone()-time setup: | |
3222 | */ | |
aab03e05 | 3223 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 3224 | { |
0122ec5b | 3225 | unsigned long flags; |
dd41f596 | 3226 | |
5e1576ed | 3227 | __sched_fork(clone_flags, p); |
06b83b5f | 3228 | /* |
7dc603c9 | 3229 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
3230 | * nobody will actually run it, and a signal or other external |
3231 | * event cannot wake it up and insert it on the runqueue either. | |
3232 | */ | |
7dc603c9 | 3233 | p->state = TASK_NEW; |
dd41f596 | 3234 | |
c350a04e MG |
3235 | /* |
3236 | * Make sure we do not leak PI boosting priority to the child. | |
3237 | */ | |
3238 | p->prio = current->normal_prio; | |
3239 | ||
e8f14172 PB |
3240 | uclamp_fork(p); |
3241 | ||
b9dc29e7 MG |
3242 | /* |
3243 | * Revert to default priority/policy on fork if requested. | |
3244 | */ | |
3245 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 3246 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 3247 | p->policy = SCHED_NORMAL; |
6c697bdf | 3248 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
3249 | p->rt_priority = 0; |
3250 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
3251 | p->static_prio = NICE_TO_PRIO(0); | |
3252 | ||
3253 | p->prio = p->normal_prio = __normal_prio(p); | |
9059393e | 3254 | set_load_weight(p, false); |
6c697bdf | 3255 | |
b9dc29e7 MG |
3256 | /* |
3257 | * We don't need the reset flag anymore after the fork. It has | |
3258 | * fulfilled its duty: | |
3259 | */ | |
3260 | p->sched_reset_on_fork = 0; | |
3261 | } | |
ca94c442 | 3262 | |
af0fffd9 | 3263 | if (dl_prio(p->prio)) |
aab03e05 | 3264 | return -EAGAIN; |
af0fffd9 | 3265 | else if (rt_prio(p->prio)) |
aab03e05 | 3266 | p->sched_class = &rt_sched_class; |
af0fffd9 | 3267 | else |
2ddbf952 | 3268 | p->sched_class = &fair_sched_class; |
b29739f9 | 3269 | |
7dc603c9 | 3270 | init_entity_runnable_average(&p->se); |
cd29fe6f | 3271 | |
86951599 PZ |
3272 | /* |
3273 | * The child is not yet in the pid-hash so no cgroup attach races, | |
3274 | * and the cgroup is pinned to this child due to cgroup_fork() | |
3275 | * is ran before sched_fork(). | |
3276 | * | |
3277 | * Silence PROVE_RCU. | |
3278 | */ | |
0122ec5b | 3279 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ce3614da | 3280 | rseq_migrate(p); |
e210bffd | 3281 | /* |
d1ccc66d | 3282 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
3283 | * so use __set_task_cpu(). |
3284 | */ | |
af0fffd9 | 3285 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
3286 | if (p->sched_class->task_fork) |
3287 | p->sched_class->task_fork(p); | |
0122ec5b | 3288 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 3289 | |
f6db8347 | 3290 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 3291 | if (likely(sched_info_on())) |
52f17b6c | 3292 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 3293 | #endif |
3ca7a440 PZ |
3294 | #if defined(CONFIG_SMP) |
3295 | p->on_cpu = 0; | |
4866cde0 | 3296 | #endif |
01028747 | 3297 | init_task_preempt_count(p); |
806c09a7 | 3298 | #ifdef CONFIG_SMP |
917b627d | 3299 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 3300 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 3301 | #endif |
aab03e05 | 3302 | return 0; |
1da177e4 LT |
3303 | } |
3304 | ||
13685c4a QY |
3305 | void sched_post_fork(struct task_struct *p) |
3306 | { | |
3307 | uclamp_post_fork(p); | |
3308 | } | |
3309 | ||
332ac17e DF |
3310 | unsigned long to_ratio(u64 period, u64 runtime) |
3311 | { | |
3312 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 3313 | return BW_UNIT; |
332ac17e DF |
3314 | |
3315 | /* | |
3316 | * Doing this here saves a lot of checks in all | |
3317 | * the calling paths, and returning zero seems | |
3318 | * safe for them anyway. | |
3319 | */ | |
3320 | if (period == 0) | |
3321 | return 0; | |
3322 | ||
c52f14d3 | 3323 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
3324 | } |
3325 | ||
1da177e4 LT |
3326 | /* |
3327 | * wake_up_new_task - wake up a newly created task for the first time. | |
3328 | * | |
3329 | * This function will do some initial scheduler statistics housekeeping | |
3330 | * that must be done for every newly created context, then puts the task | |
3331 | * on the runqueue and wakes it. | |
3332 | */ | |
3e51e3ed | 3333 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 3334 | { |
eb580751 | 3335 | struct rq_flags rf; |
dd41f596 | 3336 | struct rq *rq; |
fabf318e | 3337 | |
eb580751 | 3338 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
7dc603c9 | 3339 | p->state = TASK_RUNNING; |
fabf318e PZ |
3340 | #ifdef CONFIG_SMP |
3341 | /* | |
3342 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 3343 | * - cpus_ptr can change in the fork path |
d1ccc66d | 3344 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
3345 | * |
3346 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
3347 | * as we're not fully set-up yet. | |
fabf318e | 3348 | */ |
32e839dd | 3349 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 3350 | rseq_migrate(p); |
e210bffd | 3351 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
0017d735 | 3352 | #endif |
b7fa30c9 | 3353 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 3354 | update_rq_clock(rq); |
d0fe0b9c | 3355 | post_init_entity_util_avg(p); |
0017d735 | 3356 | |
7a57f32a | 3357 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 3358 | trace_sched_wakeup_new(p); |
a7558e01 | 3359 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 3360 | #ifdef CONFIG_SMP |
0aaafaab PZ |
3361 | if (p->sched_class->task_woken) { |
3362 | /* | |
3363 | * Nothing relies on rq->lock after this, so its fine to | |
3364 | * drop it. | |
3365 | */ | |
d8ac8971 | 3366 | rq_unpin_lock(rq, &rf); |
efbbd05a | 3367 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3368 | rq_repin_lock(rq, &rf); |
0aaafaab | 3369 | } |
9a897c5a | 3370 | #endif |
eb580751 | 3371 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
3372 | } |
3373 | ||
e107be36 AK |
3374 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
3375 | ||
b7203428 | 3376 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 3377 | |
2ecd9d29 PZ |
3378 | void preempt_notifier_inc(void) |
3379 | { | |
b7203428 | 3380 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
3381 | } |
3382 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
3383 | ||
3384 | void preempt_notifier_dec(void) | |
3385 | { | |
b7203428 | 3386 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
3387 | } |
3388 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
3389 | ||
e107be36 | 3390 | /** |
80dd99b3 | 3391 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 3392 | * @notifier: notifier struct to register |
e107be36 AK |
3393 | */ |
3394 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
3395 | { | |
b7203428 | 3396 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
3397 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
3398 | ||
e107be36 AK |
3399 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
3400 | } | |
3401 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
3402 | ||
3403 | /** | |
3404 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 3405 | * @notifier: notifier struct to unregister |
e107be36 | 3406 | * |
d84525a8 | 3407 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
3408 | */ |
3409 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
3410 | { | |
3411 | hlist_del(¬ifier->link); | |
3412 | } | |
3413 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
3414 | ||
1cde2930 | 3415 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3416 | { |
3417 | struct preempt_notifier *notifier; | |
e107be36 | 3418 | |
b67bfe0d | 3419 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3420 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
3421 | } | |
3422 | ||
1cde2930 PZ |
3423 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
3424 | { | |
b7203428 | 3425 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3426 | __fire_sched_in_preempt_notifiers(curr); |
3427 | } | |
3428 | ||
e107be36 | 3429 | static void |
1cde2930 PZ |
3430 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3431 | struct task_struct *next) | |
e107be36 AK |
3432 | { |
3433 | struct preempt_notifier *notifier; | |
e107be36 | 3434 | |
b67bfe0d | 3435 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
3436 | notifier->ops->sched_out(notifier, next); |
3437 | } | |
3438 | ||
1cde2930 PZ |
3439 | static __always_inline void |
3440 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
3441 | struct task_struct *next) | |
3442 | { | |
b7203428 | 3443 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
3444 | __fire_sched_out_preempt_notifiers(curr, next); |
3445 | } | |
3446 | ||
6d6bc0ad | 3447 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3448 | |
1cde2930 | 3449 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
3450 | { |
3451 | } | |
3452 | ||
1cde2930 | 3453 | static inline void |
e107be36 AK |
3454 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
3455 | struct task_struct *next) | |
3456 | { | |
3457 | } | |
3458 | ||
6d6bc0ad | 3459 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 3460 | |
31cb1bc0 | 3461 | static inline void prepare_task(struct task_struct *next) |
3462 | { | |
3463 | #ifdef CONFIG_SMP | |
3464 | /* | |
3465 | * Claim the task as running, we do this before switching to it | |
3466 | * such that any running task will have this set. | |
58877d34 PZ |
3467 | * |
3468 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 3469 | */ |
58877d34 | 3470 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 3471 | #endif |
3472 | } | |
3473 | ||
3474 | static inline void finish_task(struct task_struct *prev) | |
3475 | { | |
3476 | #ifdef CONFIG_SMP | |
3477 | /* | |
58877d34 PZ |
3478 | * This must be the very last reference to @prev from this CPU. After |
3479 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
3480 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 3481 | * finished. |
3482 | * | |
3483 | * In particular, the load of prev->state in finish_task_switch() must | |
3484 | * happen before this. | |
3485 | * | |
3486 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
3487 | */ | |
3488 | smp_store_release(&prev->on_cpu, 0); | |
3489 | #endif | |
3490 | } | |
3491 | ||
269d5992 PZ |
3492 | static inline void |
3493 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 3494 | { |
269d5992 PZ |
3495 | /* |
3496 | * Since the runqueue lock will be released by the next | |
3497 | * task (which is an invalid locking op but in the case | |
3498 | * of the scheduler it's an obvious special-case), so we | |
3499 | * do an early lockdep release here: | |
3500 | */ | |
3501 | rq_unpin_lock(rq, rf); | |
5facae4f | 3502 | spin_release(&rq->lock.dep_map, _THIS_IP_); |
31cb1bc0 | 3503 | #ifdef CONFIG_DEBUG_SPINLOCK |
3504 | /* this is a valid case when another task releases the spinlock */ | |
269d5992 | 3505 | rq->lock.owner = next; |
31cb1bc0 | 3506 | #endif |
269d5992 PZ |
3507 | } |
3508 | ||
3509 | static inline void finish_lock_switch(struct rq *rq) | |
3510 | { | |
31cb1bc0 | 3511 | /* |
3512 | * If we are tracking spinlock dependencies then we have to | |
3513 | * fix up the runqueue lock - which gets 'carried over' from | |
3514 | * prev into current: | |
3515 | */ | |
3516 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
31cb1bc0 | 3517 | raw_spin_unlock_irq(&rq->lock); |
3518 | } | |
3519 | ||
325ea10c IM |
3520 | /* |
3521 | * NOP if the arch has not defined these: | |
3522 | */ | |
3523 | ||
3524 | #ifndef prepare_arch_switch | |
3525 | # define prepare_arch_switch(next) do { } while (0) | |
3526 | #endif | |
3527 | ||
3528 | #ifndef finish_arch_post_lock_switch | |
3529 | # define finish_arch_post_lock_switch() do { } while (0) | |
3530 | #endif | |
3531 | ||
4866cde0 NP |
3532 | /** |
3533 | * prepare_task_switch - prepare to switch tasks | |
3534 | * @rq: the runqueue preparing to switch | |
421cee29 | 3535 | * @prev: the current task that is being switched out |
4866cde0 NP |
3536 | * @next: the task we are going to switch to. |
3537 | * | |
3538 | * This is called with the rq lock held and interrupts off. It must | |
3539 | * be paired with a subsequent finish_task_switch after the context | |
3540 | * switch. | |
3541 | * | |
3542 | * prepare_task_switch sets up locking and calls architecture specific | |
3543 | * hooks. | |
3544 | */ | |
e107be36 AK |
3545 | static inline void |
3546 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
3547 | struct task_struct *next) | |
4866cde0 | 3548 | { |
0ed557aa | 3549 | kcov_prepare_switch(prev); |
43148951 | 3550 | sched_info_switch(rq, prev, next); |
fe4b04fa | 3551 | perf_event_task_sched_out(prev, next); |
d7822b1e | 3552 | rseq_preempt(prev); |
e107be36 | 3553 | fire_sched_out_preempt_notifiers(prev, next); |
31cb1bc0 | 3554 | prepare_task(next); |
4866cde0 NP |
3555 | prepare_arch_switch(next); |
3556 | } | |
3557 | ||
1da177e4 LT |
3558 | /** |
3559 | * finish_task_switch - clean up after a task-switch | |
3560 | * @prev: the thread we just switched away from. | |
3561 | * | |
4866cde0 NP |
3562 | * finish_task_switch must be called after the context switch, paired |
3563 | * with a prepare_task_switch call before the context switch. | |
3564 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
3565 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
3566 | * |
3567 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 3568 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
3569 | * with the lock held can cause deadlocks; see schedule() for |
3570 | * details.) | |
dfa50b60 ON |
3571 | * |
3572 | * The context switch have flipped the stack from under us and restored the | |
3573 | * local variables which were saved when this task called schedule() in the | |
3574 | * past. prev == current is still correct but we need to recalculate this_rq | |
3575 | * because prev may have moved to another CPU. | |
1da177e4 | 3576 | */ |
dfa50b60 | 3577 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
3578 | __releases(rq->lock) |
3579 | { | |
dfa50b60 | 3580 | struct rq *rq = this_rq(); |
1da177e4 | 3581 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 3582 | long prev_state; |
1da177e4 | 3583 | |
609ca066 PZ |
3584 | /* |
3585 | * The previous task will have left us with a preempt_count of 2 | |
3586 | * because it left us after: | |
3587 | * | |
3588 | * schedule() | |
3589 | * preempt_disable(); // 1 | |
3590 | * __schedule() | |
3591 | * raw_spin_lock_irq(&rq->lock) // 2 | |
3592 | * | |
3593 | * Also, see FORK_PREEMPT_COUNT. | |
3594 | */ | |
e2bf1c4b PZ |
3595 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
3596 | "corrupted preempt_count: %s/%d/0x%x\n", | |
3597 | current->comm, current->pid, preempt_count())) | |
3598 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 3599 | |
1da177e4 LT |
3600 | rq->prev_mm = NULL; |
3601 | ||
3602 | /* | |
3603 | * A task struct has one reference for the use as "current". | |
c394cc9f | 3604 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
3605 | * schedule one last time. The schedule call will never return, and |
3606 | * the scheduled task must drop that reference. | |
95913d97 PZ |
3607 | * |
3608 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 3609 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
3610 | * running on another CPU and we could rave with its RUNNING -> DEAD |
3611 | * transition, resulting in a double drop. | |
1da177e4 | 3612 | */ |
55a101f8 | 3613 | prev_state = prev->state; |
bf9fae9f | 3614 | vtime_task_switch(prev); |
a8d757ef | 3615 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 3616 | finish_task(prev); |
3617 | finish_lock_switch(rq); | |
01f23e16 | 3618 | finish_arch_post_lock_switch(); |
0ed557aa | 3619 | kcov_finish_switch(current); |
e8fa1362 | 3620 | |
e107be36 | 3621 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 3622 | /* |
70216e18 MD |
3623 | * When switching through a kernel thread, the loop in |
3624 | * membarrier_{private,global}_expedited() may have observed that | |
3625 | * kernel thread and not issued an IPI. It is therefore possible to | |
3626 | * schedule between user->kernel->user threads without passing though | |
3627 | * switch_mm(). Membarrier requires a barrier after storing to | |
3628 | * rq->curr, before returning to userspace, so provide them here: | |
3629 | * | |
3630 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
3631 | * provided by mmdrop(), | |
3632 | * - a sync_core for SYNC_CORE. | |
306e0604 | 3633 | */ |
70216e18 MD |
3634 | if (mm) { |
3635 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 3636 | mmdrop(mm); |
70216e18 | 3637 | } |
1cef1150 PZ |
3638 | if (unlikely(prev_state == TASK_DEAD)) { |
3639 | if (prev->sched_class->task_dead) | |
3640 | prev->sched_class->task_dead(prev); | |
68f24b08 | 3641 | |
1cef1150 PZ |
3642 | /* |
3643 | * Remove function-return probe instances associated with this | |
3644 | * task and put them back on the free list. | |
3645 | */ | |
3646 | kprobe_flush_task(prev); | |
3647 | ||
3648 | /* Task is done with its stack. */ | |
3649 | put_task_stack(prev); | |
3650 | ||
0ff7b2cf | 3651 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 3652 | } |
99e5ada9 | 3653 | |
de734f89 | 3654 | tick_nohz_task_switch(); |
dfa50b60 | 3655 | return rq; |
1da177e4 LT |
3656 | } |
3657 | ||
3f029d3c GH |
3658 | #ifdef CONFIG_SMP |
3659 | ||
3f029d3c | 3660 | /* rq->lock is NOT held, but preemption is disabled */ |
e3fca9e7 | 3661 | static void __balance_callback(struct rq *rq) |
3f029d3c | 3662 | { |
e3fca9e7 PZ |
3663 | struct callback_head *head, *next; |
3664 | void (*func)(struct rq *rq); | |
3665 | unsigned long flags; | |
3f029d3c | 3666 | |
e3fca9e7 PZ |
3667 | raw_spin_lock_irqsave(&rq->lock, flags); |
3668 | head = rq->balance_callback; | |
3669 | rq->balance_callback = NULL; | |
3670 | while (head) { | |
3671 | func = (void (*)(struct rq *))head->func; | |
3672 | next = head->next; | |
3673 | head->next = NULL; | |
3674 | head = next; | |
3f029d3c | 3675 | |
e3fca9e7 | 3676 | func(rq); |
3f029d3c | 3677 | } |
e3fca9e7 PZ |
3678 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3679 | } | |
3680 | ||
3681 | static inline void balance_callback(struct rq *rq) | |
3682 | { | |
3683 | if (unlikely(rq->balance_callback)) | |
3684 | __balance_callback(rq); | |
3f029d3c GH |
3685 | } |
3686 | ||
3687 | #else | |
da19ab51 | 3688 | |
e3fca9e7 | 3689 | static inline void balance_callback(struct rq *rq) |
3f029d3c | 3690 | { |
1da177e4 LT |
3691 | } |
3692 | ||
3f029d3c GH |
3693 | #endif |
3694 | ||
1da177e4 LT |
3695 | /** |
3696 | * schedule_tail - first thing a freshly forked thread must call. | |
3697 | * @prev: the thread we just switched away from. | |
3698 | */ | |
722a9f92 | 3699 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3700 | __releases(rq->lock) |
3701 | { | |
1a43a14a | 3702 | struct rq *rq; |
da19ab51 | 3703 | |
609ca066 PZ |
3704 | /* |
3705 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
3706 | * finish_task_switch() for details. | |
3707 | * | |
3708 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
3709 | * and the preempt_enable() will end up enabling preemption (on | |
3710 | * PREEMPT_COUNT kernels). | |
3711 | */ | |
3712 | ||
dfa50b60 | 3713 | rq = finish_task_switch(prev); |
e3fca9e7 | 3714 | balance_callback(rq); |
1a43a14a | 3715 | preempt_enable(); |
70b97a7f | 3716 | |
1da177e4 | 3717 | if (current->set_child_tid) |
b488893a | 3718 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
3719 | |
3720 | calculate_sigpending(); | |
1da177e4 LT |
3721 | } |
3722 | ||
3723 | /* | |
dfa50b60 | 3724 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 3725 | */ |
04936948 | 3726 | static __always_inline struct rq * |
70b97a7f | 3727 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 3728 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 3729 | { |
e107be36 | 3730 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3731 | |
9226d125 ZA |
3732 | /* |
3733 | * For paravirt, this is coupled with an exit in switch_to to | |
3734 | * combine the page table reload and the switch backend into | |
3735 | * one hypercall. | |
3736 | */ | |
224101ed | 3737 | arch_start_context_switch(prev); |
9226d125 | 3738 | |
306e0604 | 3739 | /* |
139d025c PZ |
3740 | * kernel -> kernel lazy + transfer active |
3741 | * user -> kernel lazy + mmgrab() active | |
3742 | * | |
3743 | * kernel -> user switch + mmdrop() active | |
3744 | * user -> user switch | |
306e0604 | 3745 | */ |
139d025c PZ |
3746 | if (!next->mm) { // to kernel |
3747 | enter_lazy_tlb(prev->active_mm, next); | |
3748 | ||
3749 | next->active_mm = prev->active_mm; | |
3750 | if (prev->mm) // from user | |
3751 | mmgrab(prev->active_mm); | |
3752 | else | |
3753 | prev->active_mm = NULL; | |
3754 | } else { // to user | |
227a4aad | 3755 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
3756 | /* |
3757 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 3758 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
3759 | * |
3760 | * The below provides this either through switch_mm(), or in | |
3761 | * case 'prev->active_mm == next->mm' through | |
3762 | * finish_task_switch()'s mmdrop(). | |
3763 | */ | |
139d025c | 3764 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 3765 | |
139d025c PZ |
3766 | if (!prev->mm) { // from kernel |
3767 | /* will mmdrop() in finish_task_switch(). */ | |
3768 | rq->prev_mm = prev->active_mm; | |
3769 | prev->active_mm = NULL; | |
3770 | } | |
1da177e4 | 3771 | } |
92509b73 | 3772 | |
cb42c9a3 | 3773 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 3774 | |
269d5992 | 3775 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
3776 | |
3777 | /* Here we just switch the register state and the stack. */ | |
3778 | switch_to(prev, next, prev); | |
dd41f596 | 3779 | barrier(); |
dfa50b60 ON |
3780 | |
3781 | return finish_task_switch(prev); | |
1da177e4 LT |
3782 | } |
3783 | ||
3784 | /* | |
1c3e8264 | 3785 | * nr_running and nr_context_switches: |
1da177e4 LT |
3786 | * |
3787 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 3788 | * threads, total number of context switches performed since bootup. |
1da177e4 LT |
3789 | */ |
3790 | unsigned long nr_running(void) | |
3791 | { | |
3792 | unsigned long i, sum = 0; | |
3793 | ||
3794 | for_each_online_cpu(i) | |
3795 | sum += cpu_rq(i)->nr_running; | |
3796 | ||
3797 | return sum; | |
f711f609 | 3798 | } |
1da177e4 | 3799 | |
2ee507c4 | 3800 | /* |
d1ccc66d | 3801 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
3802 | * |
3803 | * Caution: this function does not check that the caller has disabled | |
3804 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
3805 | * race. The caller is responsible to use it correctly, for example: | |
3806 | * | |
dfcb245e | 3807 | * - from a non-preemptible section (of course) |
00cc1633 DD |
3808 | * |
3809 | * - from a thread that is bound to a single CPU | |
3810 | * | |
3811 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
3812 | */ |
3813 | bool single_task_running(void) | |
3814 | { | |
00cc1633 | 3815 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
3816 | } |
3817 | EXPORT_SYMBOL(single_task_running); | |
3818 | ||
1da177e4 | 3819 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3820 | { |
cc94abfc SR |
3821 | int i; |
3822 | unsigned long long sum = 0; | |
46cb4b7c | 3823 | |
0a945022 | 3824 | for_each_possible_cpu(i) |
1da177e4 | 3825 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3826 | |
1da177e4 LT |
3827 | return sum; |
3828 | } | |
483b4ee6 | 3829 | |
145d952a DL |
3830 | /* |
3831 | * Consumers of these two interfaces, like for example the cpuidle menu | |
3832 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
3833 | * for a CPU that has IO-wait which might not even end up running the task when | |
3834 | * it does become runnable. | |
3835 | */ | |
3836 | ||
3837 | unsigned long nr_iowait_cpu(int cpu) | |
3838 | { | |
3839 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
3840 | } | |
3841 | ||
e33a9bba TH |
3842 | /* |
3843 | * IO-wait accounting, and how its mostly bollocks (on SMP). | |
3844 | * | |
3845 | * The idea behind IO-wait account is to account the idle time that we could | |
3846 | * have spend running if it were not for IO. That is, if we were to improve the | |
3847 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
3848 | * | |
3849 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
3850 | * idle time as IO-wait, because if the storage were faster, it could've been | |
3851 | * running and we'd not be idle. | |
3852 | * | |
3853 | * This has been extended to SMP, by doing the same for each CPU. This however | |
3854 | * is broken. | |
3855 | * | |
3856 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
3857 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
3858 | * though, if the storage were faster, both could've ran at the same time, | |
3859 | * utilising both CPUs. | |
3860 | * | |
3861 | * This means, that when looking globally, the current IO-wait accounting on | |
3862 | * SMP is a lower bound, by reason of under accounting. | |
3863 | * | |
3864 | * Worse, since the numbers are provided per CPU, they are sometimes | |
3865 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
3866 | * associated with any one particular CPU, it can wake to another CPU than it | |
3867 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
3868 | * | |
3869 | * Task CPU affinities can make all that even more 'interesting'. | |
3870 | */ | |
3871 | ||
1da177e4 LT |
3872 | unsigned long nr_iowait(void) |
3873 | { | |
3874 | unsigned long i, sum = 0; | |
483b4ee6 | 3875 | |
0a945022 | 3876 | for_each_possible_cpu(i) |
145d952a | 3877 | sum += nr_iowait_cpu(i); |
46cb4b7c | 3878 | |
1da177e4 LT |
3879 | return sum; |
3880 | } | |
483b4ee6 | 3881 | |
dd41f596 | 3882 | #ifdef CONFIG_SMP |
8a0be9ef | 3883 | |
46cb4b7c | 3884 | /* |
38022906 PZ |
3885 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3886 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3887 | */ |
38022906 | 3888 | void sched_exec(void) |
46cb4b7c | 3889 | { |
38022906 | 3890 | struct task_struct *p = current; |
1da177e4 | 3891 | unsigned long flags; |
0017d735 | 3892 | int dest_cpu; |
46cb4b7c | 3893 | |
8f42ced9 | 3894 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ac66f547 | 3895 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3896 | if (dest_cpu == smp_processor_id()) |
3897 | goto unlock; | |
38022906 | 3898 | |
8f42ced9 | 3899 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3900 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3901 | |
8f42ced9 PZ |
3902 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3903 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3904 | return; |
3905 | } | |
0017d735 | 3906 | unlock: |
8f42ced9 | 3907 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3908 | } |
dd41f596 | 3909 | |
1da177e4 LT |
3910 | #endif |
3911 | ||
1da177e4 | 3912 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 3913 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
3914 | |
3915 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 3916 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 3917 | |
6075620b GG |
3918 | /* |
3919 | * The function fair_sched_class.update_curr accesses the struct curr | |
3920 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
3921 | * we observe a high rate of cache misses in practice. | |
3922 | * Prefetching this data results in improved performance. | |
3923 | */ | |
3924 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
3925 | { | |
3926 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
3927 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
3928 | #else | |
3929 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
3930 | #endif | |
3931 | prefetch(curr); | |
3932 | prefetch(&curr->exec_start); | |
3933 | } | |
3934 | ||
c5f8d995 HS |
3935 | /* |
3936 | * Return accounted runtime for the task. | |
3937 | * In case the task is currently running, return the runtime plus current's | |
3938 | * pending runtime that have not been accounted yet. | |
3939 | */ | |
3940 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3941 | { | |
eb580751 | 3942 | struct rq_flags rf; |
c5f8d995 | 3943 | struct rq *rq; |
6e998916 | 3944 | u64 ns; |
c5f8d995 | 3945 | |
911b2898 PZ |
3946 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
3947 | /* | |
97fb7a0a | 3948 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
3949 | * So we have a optimization chance when the task's delta_exec is 0. |
3950 | * Reading ->on_cpu is racy, but this is ok. | |
3951 | * | |
d1ccc66d IM |
3952 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
3953 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 3954 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
3955 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
3956 | * been accounted, so we're correct here as well. | |
911b2898 | 3957 | */ |
da0c1e65 | 3958 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
3959 | return p->se.sum_exec_runtime; |
3960 | #endif | |
3961 | ||
eb580751 | 3962 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
3963 | /* |
3964 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
3965 | * project cycles that may never be accounted to this | |
3966 | * thread, breaking clock_gettime(). | |
3967 | */ | |
3968 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 3969 | prefetch_curr_exec_start(p); |
6e998916 SG |
3970 | update_rq_clock(rq); |
3971 | p->sched_class->update_curr(rq); | |
3972 | } | |
3973 | ns = p->se.sum_exec_runtime; | |
eb580751 | 3974 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
3975 | |
3976 | return ns; | |
3977 | } | |
48f24c4d | 3978 | |
7835b98b CL |
3979 | /* |
3980 | * This function gets called by the timer code, with HZ frequency. | |
3981 | * We call it with interrupts disabled. | |
7835b98b CL |
3982 | */ |
3983 | void scheduler_tick(void) | |
3984 | { | |
7835b98b CL |
3985 | int cpu = smp_processor_id(); |
3986 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3987 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 3988 | struct rq_flags rf; |
b4eccf5f | 3989 | unsigned long thermal_pressure; |
3e51f33f | 3990 | |
1567c3e3 | 3991 | arch_scale_freq_tick(); |
3e51f33f | 3992 | sched_clock_tick(); |
dd41f596 | 3993 | |
8a8c69c3 PZ |
3994 | rq_lock(rq, &rf); |
3995 | ||
3e51f33f | 3996 | update_rq_clock(rq); |
b4eccf5f | 3997 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 3998 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 3999 | curr->sched_class->task_tick(rq, curr, 0); |
3289bdb4 | 4000 | calc_global_load_tick(rq); |
eb414681 | 4001 | psi_task_tick(rq); |
8a8c69c3 PZ |
4002 | |
4003 | rq_unlock(rq, &rf); | |
7835b98b | 4004 | |
e9d2b064 | 4005 | perf_event_task_tick(); |
e220d2dc | 4006 | |
e418e1c2 | 4007 | #ifdef CONFIG_SMP |
6eb57e0d | 4008 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 4009 | trigger_load_balance(rq); |
e418e1c2 | 4010 | #endif |
1da177e4 LT |
4011 | } |
4012 | ||
265f22a9 | 4013 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
4014 | |
4015 | struct tick_work { | |
4016 | int cpu; | |
b55bd585 | 4017 | atomic_t state; |
d84b3131 FW |
4018 | struct delayed_work work; |
4019 | }; | |
b55bd585 PM |
4020 | /* Values for ->state, see diagram below. */ |
4021 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
4022 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
4023 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
4024 | ||
4025 | /* | |
4026 | * State diagram for ->state: | |
4027 | * | |
4028 | * | |
4029 | * TICK_SCHED_REMOTE_OFFLINE | |
4030 | * | ^ | |
4031 | * | | | |
4032 | * | | sched_tick_remote() | |
4033 | * | | | |
4034 | * | | | |
4035 | * +--TICK_SCHED_REMOTE_OFFLINING | |
4036 | * | ^ | |
4037 | * | | | |
4038 | * sched_tick_start() | | sched_tick_stop() | |
4039 | * | | | |
4040 | * V | | |
4041 | * TICK_SCHED_REMOTE_RUNNING | |
4042 | * | |
4043 | * | |
4044 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
4045 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
4046 | */ | |
d84b3131 FW |
4047 | |
4048 | static struct tick_work __percpu *tick_work_cpu; | |
4049 | ||
4050 | static void sched_tick_remote(struct work_struct *work) | |
4051 | { | |
4052 | struct delayed_work *dwork = to_delayed_work(work); | |
4053 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
4054 | int cpu = twork->cpu; | |
4055 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 4056 | struct task_struct *curr; |
d84b3131 | 4057 | struct rq_flags rf; |
d9c0ffca | 4058 | u64 delta; |
b55bd585 | 4059 | int os; |
d84b3131 FW |
4060 | |
4061 | /* | |
4062 | * Handle the tick only if it appears the remote CPU is running in full | |
4063 | * dynticks mode. The check is racy by nature, but missing a tick or | |
4064 | * having one too much is no big deal because the scheduler tick updates | |
4065 | * statistics and checks timeslices in a time-independent way, regardless | |
4066 | * of when exactly it is running. | |
4067 | */ | |
488603b8 | 4068 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 4069 | goto out_requeue; |
d84b3131 | 4070 | |
d9c0ffca FW |
4071 | rq_lock_irq(rq, &rf); |
4072 | curr = rq->curr; | |
488603b8 | 4073 | if (cpu_is_offline(cpu)) |
d9c0ffca | 4074 | goto out_unlock; |
d84b3131 | 4075 | |
d9c0ffca | 4076 | update_rq_clock(rq); |
d9c0ffca | 4077 | |
488603b8 SW |
4078 | if (!is_idle_task(curr)) { |
4079 | /* | |
4080 | * Make sure the next tick runs within a reasonable | |
4081 | * amount of time. | |
4082 | */ | |
4083 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
4084 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
4085 | } | |
d9c0ffca FW |
4086 | curr->sched_class->task_tick(rq, curr, 0); |
4087 | ||
ebc0f83c | 4088 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
4089 | out_unlock: |
4090 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 4091 | out_requeue: |
ebc0f83c | 4092 | |
d84b3131 FW |
4093 | /* |
4094 | * Run the remote tick once per second (1Hz). This arbitrary | |
4095 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
4096 | * to keep scheduler internal stats reasonably up to date. But |
4097 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 4098 | */ |
b55bd585 PM |
4099 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
4100 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
4101 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
4102 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
4103 | } |
4104 | ||
4105 | static void sched_tick_start(int cpu) | |
4106 | { | |
b55bd585 | 4107 | int os; |
d84b3131 FW |
4108 | struct tick_work *twork; |
4109 | ||
4110 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4111 | return; | |
4112 | ||
4113 | WARN_ON_ONCE(!tick_work_cpu); | |
4114 | ||
4115 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4116 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
4117 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
4118 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
4119 | twork->cpu = cpu; | |
4120 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
4121 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
4122 | } | |
d84b3131 FW |
4123 | } |
4124 | ||
4125 | #ifdef CONFIG_HOTPLUG_CPU | |
4126 | static void sched_tick_stop(int cpu) | |
4127 | { | |
4128 | struct tick_work *twork; | |
b55bd585 | 4129 | int os; |
d84b3131 FW |
4130 | |
4131 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
4132 | return; | |
4133 | ||
4134 | WARN_ON_ONCE(!tick_work_cpu); | |
4135 | ||
4136 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
4137 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
4138 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
4139 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
4140 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
4141 | } |
4142 | #endif /* CONFIG_HOTPLUG_CPU */ | |
4143 | ||
4144 | int __init sched_tick_offload_init(void) | |
4145 | { | |
4146 | tick_work_cpu = alloc_percpu(struct tick_work); | |
4147 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
4148 | return 0; |
4149 | } | |
4150 | ||
4151 | #else /* !CONFIG_NO_HZ_FULL */ | |
4152 | static inline void sched_tick_start(int cpu) { } | |
4153 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 4154 | #endif |
1da177e4 | 4155 | |
c1a280b6 | 4156 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 4157 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
4158 | /* |
4159 | * If the value passed in is equal to the current preempt count | |
4160 | * then we just disabled preemption. Start timing the latency. | |
4161 | */ | |
4162 | static inline void preempt_latency_start(int val) | |
4163 | { | |
4164 | if (preempt_count() == val) { | |
4165 | unsigned long ip = get_lock_parent_ip(); | |
4166 | #ifdef CONFIG_DEBUG_PREEMPT | |
4167 | current->preempt_disable_ip = ip; | |
4168 | #endif | |
4169 | trace_preempt_off(CALLER_ADDR0, ip); | |
4170 | } | |
4171 | } | |
7e49fcce | 4172 | |
edafe3a5 | 4173 | void preempt_count_add(int val) |
1da177e4 | 4174 | { |
6cd8a4bb | 4175 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4176 | /* |
4177 | * Underflow? | |
4178 | */ | |
9a11b49a IM |
4179 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4180 | return; | |
6cd8a4bb | 4181 | #endif |
bdb43806 | 4182 | __preempt_count_add(val); |
6cd8a4bb | 4183 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4184 | /* |
4185 | * Spinlock count overflowing soon? | |
4186 | */ | |
33859f7f MOS |
4187 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4188 | PREEMPT_MASK - 10); | |
6cd8a4bb | 4189 | #endif |
47252cfb | 4190 | preempt_latency_start(val); |
1da177e4 | 4191 | } |
bdb43806 | 4192 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 4193 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 4194 | |
47252cfb SR |
4195 | /* |
4196 | * If the value passed in equals to the current preempt count | |
4197 | * then we just enabled preemption. Stop timing the latency. | |
4198 | */ | |
4199 | static inline void preempt_latency_stop(int val) | |
4200 | { | |
4201 | if (preempt_count() == val) | |
4202 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
4203 | } | |
4204 | ||
edafe3a5 | 4205 | void preempt_count_sub(int val) |
1da177e4 | 4206 | { |
6cd8a4bb | 4207 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4208 | /* |
4209 | * Underflow? | |
4210 | */ | |
01e3eb82 | 4211 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4212 | return; |
1da177e4 LT |
4213 | /* |
4214 | * Is the spinlock portion underflowing? | |
4215 | */ | |
9a11b49a IM |
4216 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4217 | !(preempt_count() & PREEMPT_MASK))) | |
4218 | return; | |
6cd8a4bb | 4219 | #endif |
9a11b49a | 4220 | |
47252cfb | 4221 | preempt_latency_stop(val); |
bdb43806 | 4222 | __preempt_count_sub(val); |
1da177e4 | 4223 | } |
bdb43806 | 4224 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 4225 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 4226 | |
47252cfb SR |
4227 | #else |
4228 | static inline void preempt_latency_start(int val) { } | |
4229 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
4230 | #endif |
4231 | ||
59ddbcb2 IM |
4232 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
4233 | { | |
4234 | #ifdef CONFIG_DEBUG_PREEMPT | |
4235 | return p->preempt_disable_ip; | |
4236 | #else | |
4237 | return 0; | |
4238 | #endif | |
4239 | } | |
4240 | ||
1da177e4 | 4241 | /* |
dd41f596 | 4242 | * Print scheduling while atomic bug: |
1da177e4 | 4243 | */ |
dd41f596 | 4244 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4245 | { |
d1c6d149 VN |
4246 | /* Save this before calling printk(), since that will clobber it */ |
4247 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
4248 | ||
664dfa65 DJ |
4249 | if (oops_in_progress) |
4250 | return; | |
4251 | ||
3df0fc5b PZ |
4252 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4253 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4254 | |
dd41f596 | 4255 | debug_show_held_locks(prev); |
e21f5b15 | 4256 | print_modules(); |
dd41f596 IM |
4257 | if (irqs_disabled()) |
4258 | print_irqtrace_events(prev); | |
d1c6d149 VN |
4259 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
4260 | && in_atomic_preempt_off()) { | |
8f47b187 | 4261 | pr_err("Preemption disabled at:"); |
2062a4e8 | 4262 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 4263 | } |
748c7201 DBO |
4264 | if (panic_on_warn) |
4265 | panic("scheduling while atomic\n"); | |
4266 | ||
6135fc1e | 4267 | dump_stack(); |
373d4d09 | 4268 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 4269 | } |
1da177e4 | 4270 | |
dd41f596 IM |
4271 | /* |
4272 | * Various schedule()-time debugging checks and statistics: | |
4273 | */ | |
312364f3 | 4274 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 4275 | { |
0d9e2632 | 4276 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
4277 | if (task_stack_end_corrupted(prev)) |
4278 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
4279 | |
4280 | if (task_scs_end_corrupted(prev)) | |
4281 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 4282 | #endif |
b99def8b | 4283 | |
312364f3 DV |
4284 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
4285 | if (!preempt && prev->state && prev->non_block_count) { | |
4286 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", | |
4287 | prev->comm, prev->pid, prev->non_block_count); | |
4288 | dump_stack(); | |
4289 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
4290 | } | |
4291 | #endif | |
4292 | ||
1dc0fffc | 4293 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 4294 | __schedule_bug(prev); |
1dc0fffc PZ |
4295 | preempt_count_set(PREEMPT_DISABLED); |
4296 | } | |
b3fbab05 | 4297 | rcu_sleep_check(); |
dd41f596 | 4298 | |
1da177e4 LT |
4299 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4300 | ||
ae92882e | 4301 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
4302 | } |
4303 | ||
457d1f46 CY |
4304 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
4305 | struct rq_flags *rf) | |
4306 | { | |
4307 | #ifdef CONFIG_SMP | |
4308 | const struct sched_class *class; | |
4309 | /* | |
4310 | * We must do the balancing pass before put_prev_task(), such | |
4311 | * that when we release the rq->lock the task is in the same | |
4312 | * state as before we took rq->lock. | |
4313 | * | |
4314 | * We can terminate the balance pass as soon as we know there is | |
4315 | * a runnable task of @class priority or higher. | |
4316 | */ | |
4317 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
4318 | if (class->balance(rq, prev, rf)) | |
4319 | break; | |
4320 | } | |
4321 | #endif | |
4322 | ||
4323 | put_prev_task(rq, prev); | |
4324 | } | |
4325 | ||
dd41f596 IM |
4326 | /* |
4327 | * Pick up the highest-prio task: | |
4328 | */ | |
4329 | static inline struct task_struct * | |
d8ac8971 | 4330 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 4331 | { |
49ee5768 | 4332 | const struct sched_class *class; |
dd41f596 | 4333 | struct task_struct *p; |
1da177e4 LT |
4334 | |
4335 | /* | |
0ba87bb2 PZ |
4336 | * Optimization: we know that if all tasks are in the fair class we can |
4337 | * call that function directly, but only if the @prev task wasn't of a | |
4338 | * higher scheduling class, because otherwise those loose the | |
4339 | * opportunity to pull in more work from other CPUs. | |
1da177e4 | 4340 | */ |
aa93cd53 | 4341 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
4342 | rq->nr_running == rq->cfs.h_nr_running)) { |
4343 | ||
5d7d6056 | 4344 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 4345 | if (unlikely(p == RETRY_TASK)) |
67692435 | 4346 | goto restart; |
6ccdc84b | 4347 | |
d1ccc66d | 4348 | /* Assumes fair_sched_class->next == idle_sched_class */ |
5d7d6056 | 4349 | if (!p) { |
f488e105 | 4350 | put_prev_task(rq, prev); |
98c2f700 | 4351 | p = pick_next_task_idle(rq); |
f488e105 | 4352 | } |
6ccdc84b PZ |
4353 | |
4354 | return p; | |
1da177e4 LT |
4355 | } |
4356 | ||
67692435 | 4357 | restart: |
457d1f46 | 4358 | put_prev_task_balance(rq, prev, rf); |
67692435 | 4359 | |
34f971f6 | 4360 | for_each_class(class) { |
98c2f700 | 4361 | p = class->pick_next_task(rq); |
67692435 | 4362 | if (p) |
dd41f596 | 4363 | return p; |
dd41f596 | 4364 | } |
34f971f6 | 4365 | |
d1ccc66d IM |
4366 | /* The idle class should always have a runnable task: */ |
4367 | BUG(); | |
dd41f596 | 4368 | } |
1da177e4 | 4369 | |
dd41f596 | 4370 | /* |
c259e01a | 4371 | * __schedule() is the main scheduler function. |
edde96ea PE |
4372 | * |
4373 | * The main means of driving the scheduler and thus entering this function are: | |
4374 | * | |
4375 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
4376 | * | |
4377 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
4378 | * paths. For example, see arch/x86/entry_64.S. | |
4379 | * | |
4380 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
4381 | * interrupt handler scheduler_tick(). | |
4382 | * | |
4383 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
4384 | * task to the run-queue and that's it. | |
4385 | * | |
4386 | * Now, if the new task added to the run-queue preempts the current | |
4387 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
4388 | * called on the nearest possible occasion: | |
4389 | * | |
c1a280b6 | 4390 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
4391 | * |
4392 | * - in syscall or exception context, at the next outmost | |
4393 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
4394 | * spin_unlock()!) | |
4395 | * | |
4396 | * - in IRQ context, return from interrupt-handler to | |
4397 | * preemptible context | |
4398 | * | |
c1a280b6 | 4399 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
4400 | * then at the next: |
4401 | * | |
4402 | * - cond_resched() call | |
4403 | * - explicit schedule() call | |
4404 | * - return from syscall or exception to user-space | |
4405 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 4406 | * |
b30f0e3f | 4407 | * WARNING: must be called with preemption disabled! |
dd41f596 | 4408 | */ |
499d7955 | 4409 | static void __sched notrace __schedule(bool preempt) |
dd41f596 IM |
4410 | { |
4411 | struct task_struct *prev, *next; | |
67ca7bde | 4412 | unsigned long *switch_count; |
dbfb089d | 4413 | unsigned long prev_state; |
d8ac8971 | 4414 | struct rq_flags rf; |
dd41f596 | 4415 | struct rq *rq; |
31656519 | 4416 | int cpu; |
dd41f596 | 4417 | |
dd41f596 IM |
4418 | cpu = smp_processor_id(); |
4419 | rq = cpu_rq(cpu); | |
dd41f596 | 4420 | prev = rq->curr; |
dd41f596 | 4421 | |
312364f3 | 4422 | schedule_debug(prev, preempt); |
1da177e4 | 4423 | |
31656519 | 4424 | if (sched_feat(HRTICK)) |
f333fdc9 | 4425 | hrtick_clear(rq); |
8f4d37ec | 4426 | |
46a5d164 | 4427 | local_irq_disable(); |
bcbfdd01 | 4428 | rcu_note_context_switch(preempt); |
46a5d164 | 4429 | |
e0acd0a6 ON |
4430 | /* |
4431 | * Make sure that signal_pending_state()->signal_pending() below | |
4432 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
4433 | * done by the caller to avoid the race with signal_wake_up(): |
4434 | * | |
4435 | * __set_current_state(@state) signal_wake_up() | |
4436 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
4437 | * wake_up_state(p, state) | |
4438 | * LOCK rq->lock LOCK p->pi_state | |
4439 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
4440 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 4441 | * |
dbfb089d | 4442 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 4443 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 4444 | */ |
8a8c69c3 | 4445 | rq_lock(rq, &rf); |
d89e588c | 4446 | smp_mb__after_spinlock(); |
1da177e4 | 4447 | |
d1ccc66d IM |
4448 | /* Promote REQ to ACT */ |
4449 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 4450 | update_rq_clock(rq); |
9edfbfed | 4451 | |
246d86b5 | 4452 | switch_count = &prev->nivcsw; |
d136122f | 4453 | |
dbfb089d | 4454 | /* |
d136122f PZ |
4455 | * We must load prev->state once (task_struct::state is volatile), such |
4456 | * that: | |
4457 | * | |
4458 | * - we form a control dependency vs deactivate_task() below. | |
4459 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 4460 | */ |
d136122f PZ |
4461 | prev_state = prev->state; |
4462 | if (!preempt && prev_state) { | |
dbfb089d | 4463 | if (signal_pending_state(prev_state, prev)) { |
1da177e4 | 4464 | prev->state = TASK_RUNNING; |
21aa9af0 | 4465 | } else { |
dbfb089d PZ |
4466 | prev->sched_contributes_to_load = |
4467 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
4468 | !(prev_state & TASK_NOLOAD) && | |
4469 | !(prev->flags & PF_FROZEN); | |
4470 | ||
4471 | if (prev->sched_contributes_to_load) | |
4472 | rq->nr_uninterruptible++; | |
4473 | ||
4474 | /* | |
4475 | * __schedule() ttwu() | |
d136122f PZ |
4476 | * prev_state = prev->state; if (p->on_rq && ...) |
4477 | * if (prev_state) goto out; | |
4478 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
4479 | * p->state = TASK_WAKING | |
4480 | * | |
4481 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
4482 | * |
4483 | * After this, schedule() must not care about p->state any more. | |
4484 | */ | |
bce4dc80 | 4485 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 4486 | |
e33a9bba TH |
4487 | if (prev->in_iowait) { |
4488 | atomic_inc(&rq->nr_iowait); | |
4489 | delayacct_blkio_start(); | |
4490 | } | |
21aa9af0 | 4491 | } |
dd41f596 | 4492 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4493 | } |
4494 | ||
d8ac8971 | 4495 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 4496 | clear_tsk_need_resched(prev); |
f27dde8d | 4497 | clear_preempt_need_resched(); |
1da177e4 | 4498 | |
1da177e4 | 4499 | if (likely(prev != next)) { |
1da177e4 | 4500 | rq->nr_switches++; |
5311a98f EB |
4501 | /* |
4502 | * RCU users of rcu_dereference(rq->curr) may not see | |
4503 | * changes to task_struct made by pick_next_task(). | |
4504 | */ | |
4505 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
4506 | /* |
4507 | * The membarrier system call requires each architecture | |
4508 | * to have a full memory barrier after updating | |
306e0604 MD |
4509 | * rq->curr, before returning to user-space. |
4510 | * | |
4511 | * Here are the schemes providing that barrier on the | |
4512 | * various architectures: | |
4513 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
4514 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
4515 | * - finish_lock_switch() for weakly-ordered | |
4516 | * architectures where spin_unlock is a full barrier, | |
4517 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
4518 | * is a RELEASE barrier), | |
22e4ebb9 | 4519 | */ |
1da177e4 LT |
4520 | ++*switch_count; |
4521 | ||
b05e75d6 JW |
4522 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
4523 | ||
c73464b1 | 4524 | trace_sched_switch(preempt, prev, next); |
d1ccc66d IM |
4525 | |
4526 | /* Also unlocks the rq: */ | |
4527 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 4528 | } else { |
cb42c9a3 | 4529 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
8a8c69c3 | 4530 | rq_unlock_irq(rq, &rf); |
cbce1a68 | 4531 | } |
1da177e4 | 4532 | |
e3fca9e7 | 4533 | balance_callback(rq); |
1da177e4 | 4534 | } |
c259e01a | 4535 | |
9af6528e PZ |
4536 | void __noreturn do_task_dead(void) |
4537 | { | |
d1ccc66d | 4538 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 4539 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
4540 | |
4541 | /* Tell freezer to ignore us: */ | |
4542 | current->flags |= PF_NOFREEZE; | |
4543 | ||
9af6528e PZ |
4544 | __schedule(false); |
4545 | BUG(); | |
d1ccc66d IM |
4546 | |
4547 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 4548 | for (;;) |
d1ccc66d | 4549 | cpu_relax(); |
9af6528e PZ |
4550 | } |
4551 | ||
9c40cef2 TG |
4552 | static inline void sched_submit_work(struct task_struct *tsk) |
4553 | { | |
b0fdc013 | 4554 | if (!tsk->state) |
9c40cef2 | 4555 | return; |
6d25be57 TG |
4556 | |
4557 | /* | |
4558 | * If a worker went to sleep, notify and ask workqueue whether | |
4559 | * it wants to wake up a task to maintain concurrency. | |
4560 | * As this function is called inside the schedule() context, | |
4561 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
4562 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
4563 | * requires it. | |
6d25be57 | 4564 | */ |
771b53d0 | 4565 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 4566 | preempt_disable(); |
771b53d0 JA |
4567 | if (tsk->flags & PF_WQ_WORKER) |
4568 | wq_worker_sleeping(tsk); | |
4569 | else | |
4570 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
4571 | preempt_enable_no_resched(); |
4572 | } | |
4573 | ||
b0fdc013 SAS |
4574 | if (tsk_is_pi_blocked(tsk)) |
4575 | return; | |
4576 | ||
9c40cef2 TG |
4577 | /* |
4578 | * If we are going to sleep and we have plugged IO queued, | |
4579 | * make sure to submit it to avoid deadlocks. | |
4580 | */ | |
4581 | if (blk_needs_flush_plug(tsk)) | |
4582 | blk_schedule_flush_plug(tsk); | |
4583 | } | |
4584 | ||
6d25be57 TG |
4585 | static void sched_update_worker(struct task_struct *tsk) |
4586 | { | |
771b53d0 JA |
4587 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
4588 | if (tsk->flags & PF_WQ_WORKER) | |
4589 | wq_worker_running(tsk); | |
4590 | else | |
4591 | io_wq_worker_running(tsk); | |
4592 | } | |
6d25be57 TG |
4593 | } |
4594 | ||
722a9f92 | 4595 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 4596 | { |
9c40cef2 TG |
4597 | struct task_struct *tsk = current; |
4598 | ||
4599 | sched_submit_work(tsk); | |
bfd9b2b5 | 4600 | do { |
b30f0e3f | 4601 | preempt_disable(); |
fc13aeba | 4602 | __schedule(false); |
b30f0e3f | 4603 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 4604 | } while (need_resched()); |
6d25be57 | 4605 | sched_update_worker(tsk); |
c259e01a | 4606 | } |
1da177e4 LT |
4607 | EXPORT_SYMBOL(schedule); |
4608 | ||
8663effb SRV |
4609 | /* |
4610 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
4611 | * state (have scheduled out non-voluntarily) by making sure that all | |
4612 | * tasks have either left the run queue or have gone into user space. | |
4613 | * As idle tasks do not do either, they must not ever be preempted | |
4614 | * (schedule out non-voluntarily). | |
4615 | * | |
4616 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
4617 | * never enables preemption because it does not call sched_submit_work(). | |
4618 | */ | |
4619 | void __sched schedule_idle(void) | |
4620 | { | |
4621 | /* | |
4622 | * As this skips calling sched_submit_work(), which the idle task does | |
4623 | * regardless because that function is a nop when the task is in a | |
4624 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
4625 | * current task can be in any other state. Note, idle is always in the | |
4626 | * TASK_RUNNING state. | |
4627 | */ | |
4628 | WARN_ON_ONCE(current->state); | |
4629 | do { | |
4630 | __schedule(false); | |
4631 | } while (need_resched()); | |
4632 | } | |
4633 | ||
91d1aa43 | 4634 | #ifdef CONFIG_CONTEXT_TRACKING |
722a9f92 | 4635 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
4636 | { |
4637 | /* | |
4638 | * If we come here after a random call to set_need_resched(), | |
4639 | * or we have been woken up remotely but the IPI has not yet arrived, | |
4640 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
4641 | * we find a better solution. | |
7cc78f8f AL |
4642 | * |
4643 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 4644 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 4645 | * too frequently to make sense yet. |
20ab65e3 | 4646 | */ |
7cc78f8f | 4647 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 4648 | schedule(); |
7cc78f8f | 4649 | exception_exit(prev_state); |
20ab65e3 FW |
4650 | } |
4651 | #endif | |
4652 | ||
c5491ea7 TG |
4653 | /** |
4654 | * schedule_preempt_disabled - called with preemption disabled | |
4655 | * | |
4656 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
4657 | */ | |
4658 | void __sched schedule_preempt_disabled(void) | |
4659 | { | |
ba74c144 | 4660 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
4661 | schedule(); |
4662 | preempt_disable(); | |
4663 | } | |
4664 | ||
06b1f808 | 4665 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
4666 | { |
4667 | do { | |
47252cfb SR |
4668 | /* |
4669 | * Because the function tracer can trace preempt_count_sub() | |
4670 | * and it also uses preempt_enable/disable_notrace(), if | |
4671 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
4672 | * by the function tracer will call this function again and | |
4673 | * cause infinite recursion. | |
4674 | * | |
4675 | * Preemption must be disabled here before the function | |
4676 | * tracer can trace. Break up preempt_disable() into two | |
4677 | * calls. One to disable preemption without fear of being | |
4678 | * traced. The other to still record the preemption latency, | |
4679 | * which can also be traced by the function tracer. | |
4680 | */ | |
499d7955 | 4681 | preempt_disable_notrace(); |
47252cfb | 4682 | preempt_latency_start(1); |
fc13aeba | 4683 | __schedule(true); |
47252cfb | 4684 | preempt_latency_stop(1); |
499d7955 | 4685 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
4686 | |
4687 | /* | |
4688 | * Check again in case we missed a preemption opportunity | |
4689 | * between schedule and now. | |
4690 | */ | |
a18b5d01 FW |
4691 | } while (need_resched()); |
4692 | } | |
4693 | ||
c1a280b6 | 4694 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 4695 | /* |
a49b4f40 VS |
4696 | * This is the entry point to schedule() from in-kernel preemption |
4697 | * off of preempt_enable. | |
1da177e4 | 4698 | */ |
722a9f92 | 4699 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 4700 | { |
1da177e4 LT |
4701 | /* |
4702 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4703 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4704 | */ |
fbb00b56 | 4705 | if (likely(!preemptible())) |
1da177e4 LT |
4706 | return; |
4707 | ||
a18b5d01 | 4708 | preempt_schedule_common(); |
1da177e4 | 4709 | } |
376e2424 | 4710 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 4711 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 4712 | |
009f60e2 | 4713 | /** |
4eaca0a8 | 4714 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
4715 | * |
4716 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
4717 | * recursion and tracing preempt enabling caused by the tracing | |
4718 | * infrastructure itself. But as tracing can happen in areas coming | |
4719 | * from userspace or just about to enter userspace, a preempt enable | |
4720 | * can occur before user_exit() is called. This will cause the scheduler | |
4721 | * to be called when the system is still in usermode. | |
4722 | * | |
4723 | * To prevent this, the preempt_enable_notrace will use this function | |
4724 | * instead of preempt_schedule() to exit user context if needed before | |
4725 | * calling the scheduler. | |
4726 | */ | |
4eaca0a8 | 4727 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
4728 | { |
4729 | enum ctx_state prev_ctx; | |
4730 | ||
4731 | if (likely(!preemptible())) | |
4732 | return; | |
4733 | ||
4734 | do { | |
47252cfb SR |
4735 | /* |
4736 | * Because the function tracer can trace preempt_count_sub() | |
4737 | * and it also uses preempt_enable/disable_notrace(), if | |
4738 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
4739 | * by the function tracer will call this function again and | |
4740 | * cause infinite recursion. | |
4741 | * | |
4742 | * Preemption must be disabled here before the function | |
4743 | * tracer can trace. Break up preempt_disable() into two | |
4744 | * calls. One to disable preemption without fear of being | |
4745 | * traced. The other to still record the preemption latency, | |
4746 | * which can also be traced by the function tracer. | |
4747 | */ | |
3d8f74dd | 4748 | preempt_disable_notrace(); |
47252cfb | 4749 | preempt_latency_start(1); |
009f60e2 ON |
4750 | /* |
4751 | * Needs preempt disabled in case user_exit() is traced | |
4752 | * and the tracer calls preempt_enable_notrace() causing | |
4753 | * an infinite recursion. | |
4754 | */ | |
4755 | prev_ctx = exception_enter(); | |
fc13aeba | 4756 | __schedule(true); |
009f60e2 ON |
4757 | exception_exit(prev_ctx); |
4758 | ||
47252cfb | 4759 | preempt_latency_stop(1); |
3d8f74dd | 4760 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
4761 | } while (need_resched()); |
4762 | } | |
4eaca0a8 | 4763 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 4764 | |
c1a280b6 | 4765 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 LT |
4766 | |
4767 | /* | |
a49b4f40 | 4768 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4769 | * off of irq context. |
4770 | * Note, that this is called and return with irqs disabled. This will | |
4771 | * protect us against recursive calling from irq. | |
4772 | */ | |
722a9f92 | 4773 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 4774 | { |
b22366cd | 4775 | enum ctx_state prev_state; |
6478d880 | 4776 | |
2ed6e34f | 4777 | /* Catch callers which need to be fixed */ |
f27dde8d | 4778 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 4779 | |
b22366cd FW |
4780 | prev_state = exception_enter(); |
4781 | ||
3a5c359a | 4782 | do { |
3d8f74dd | 4783 | preempt_disable(); |
3a5c359a | 4784 | local_irq_enable(); |
fc13aeba | 4785 | __schedule(true); |
3a5c359a | 4786 | local_irq_disable(); |
3d8f74dd | 4787 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 4788 | } while (need_resched()); |
b22366cd FW |
4789 | |
4790 | exception_exit(prev_state); | |
1da177e4 LT |
4791 | } |
4792 | ||
ac6424b9 | 4793 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4794 | void *key) |
1da177e4 | 4795 | { |
062d3f95 | 4796 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 4797 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4798 | } |
1da177e4 LT |
4799 | EXPORT_SYMBOL(default_wake_function); |
4800 | ||
b29739f9 IM |
4801 | #ifdef CONFIG_RT_MUTEXES |
4802 | ||
acd58620 PZ |
4803 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
4804 | { | |
4805 | if (pi_task) | |
4806 | prio = min(prio, pi_task->prio); | |
4807 | ||
4808 | return prio; | |
4809 | } | |
4810 | ||
4811 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
4812 | { | |
4813 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
4814 | ||
4815 | return __rt_effective_prio(pi_task, prio); | |
4816 | } | |
4817 | ||
b29739f9 IM |
4818 | /* |
4819 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
4820 | * @p: task to boost |
4821 | * @pi_task: donor task | |
b29739f9 IM |
4822 | * |
4823 | * This function changes the 'effective' priority of a task. It does | |
4824 | * not touch ->normal_prio like __setscheduler(). | |
4825 | * | |
c365c292 TG |
4826 | * Used by the rt_mutex code to implement priority inheritance |
4827 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 4828 | */ |
acd58620 | 4829 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 4830 | { |
acd58620 | 4831 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 4832 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 4833 | const struct sched_class *prev_class; |
eb580751 PZ |
4834 | struct rq_flags rf; |
4835 | struct rq *rq; | |
b29739f9 | 4836 | |
acd58620 PZ |
4837 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
4838 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
4839 | ||
4840 | /* | |
4841 | * If nothing changed; bail early. | |
4842 | */ | |
4843 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
4844 | return; | |
b29739f9 | 4845 | |
eb580751 | 4846 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 4847 | update_rq_clock(rq); |
acd58620 PZ |
4848 | /* |
4849 | * Set under pi_lock && rq->lock, such that the value can be used under | |
4850 | * either lock. | |
4851 | * | |
4852 | * Note that there is loads of tricky to make this pointer cache work | |
4853 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
4854 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
4855 | * task is allowed to run again (and can exit). This ensures the pointer | |
4856 | * points to a blocked task -- which guaratees the task is present. | |
4857 | */ | |
4858 | p->pi_top_task = pi_task; | |
4859 | ||
4860 | /* | |
4861 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
4862 | */ | |
4863 | if (prio == p->prio && !dl_prio(prio)) | |
4864 | goto out_unlock; | |
b29739f9 | 4865 | |
1c4dd99b TG |
4866 | /* |
4867 | * Idle task boosting is a nono in general. There is one | |
4868 | * exception, when PREEMPT_RT and NOHZ is active: | |
4869 | * | |
4870 | * The idle task calls get_next_timer_interrupt() and holds | |
4871 | * the timer wheel base->lock on the CPU and another CPU wants | |
4872 | * to access the timer (probably to cancel it). We can safely | |
4873 | * ignore the boosting request, as the idle CPU runs this code | |
4874 | * with interrupts disabled and will complete the lock | |
4875 | * protected section without being interrupted. So there is no | |
4876 | * real need to boost. | |
4877 | */ | |
4878 | if (unlikely(p == rq->idle)) { | |
4879 | WARN_ON(p != rq->curr); | |
4880 | WARN_ON(p->pi_blocked_on); | |
4881 | goto out_unlock; | |
4882 | } | |
4883 | ||
b91473ff | 4884 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 4885 | oldprio = p->prio; |
ff77e468 PZ |
4886 | |
4887 | if (oldprio == prio) | |
4888 | queue_flag &= ~DEQUEUE_MOVE; | |
4889 | ||
83ab0aa0 | 4890 | prev_class = p->sched_class; |
da0c1e65 | 4891 | queued = task_on_rq_queued(p); |
051a1d1a | 4892 | running = task_current(rq, p); |
da0c1e65 | 4893 | if (queued) |
ff77e468 | 4894 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 4895 | if (running) |
f3cd1c4e | 4896 | put_prev_task(rq, p); |
dd41f596 | 4897 | |
2d3d891d DF |
4898 | /* |
4899 | * Boosting condition are: | |
4900 | * 1. -rt task is running and holds mutex A | |
4901 | * --> -dl task blocks on mutex A | |
4902 | * | |
4903 | * 2. -dl task is running and holds mutex A | |
4904 | * --> -dl task blocks on mutex A and could preempt the | |
4905 | * running task | |
4906 | */ | |
4907 | if (dl_prio(prio)) { | |
466af29b | 4908 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
4909 | (pi_task && dl_prio(pi_task->prio) && |
4910 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2d3d891d | 4911 | p->dl.dl_boosted = 1; |
ff77e468 | 4912 | queue_flag |= ENQUEUE_REPLENISH; |
2d3d891d DF |
4913 | } else |
4914 | p->dl.dl_boosted = 0; | |
aab03e05 | 4915 | p->sched_class = &dl_sched_class; |
2d3d891d DF |
4916 | } else if (rt_prio(prio)) { |
4917 | if (dl_prio(oldprio)) | |
4918 | p->dl.dl_boosted = 0; | |
4919 | if (oldprio < prio) | |
ff77e468 | 4920 | queue_flag |= ENQUEUE_HEAD; |
dd41f596 | 4921 | p->sched_class = &rt_sched_class; |
2d3d891d DF |
4922 | } else { |
4923 | if (dl_prio(oldprio)) | |
4924 | p->dl.dl_boosted = 0; | |
746db944 BS |
4925 | if (rt_prio(oldprio)) |
4926 | p->rt.timeout = 0; | |
dd41f596 | 4927 | p->sched_class = &fair_sched_class; |
2d3d891d | 4928 | } |
dd41f596 | 4929 | |
b29739f9 IM |
4930 | p->prio = prio; |
4931 | ||
da0c1e65 | 4932 | if (queued) |
ff77e468 | 4933 | enqueue_task(rq, p, queue_flag); |
a399d233 | 4934 | if (running) |
03b7fad1 | 4935 | set_next_task(rq, p); |
cb469845 | 4936 | |
da7a735e | 4937 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 4938 | out_unlock: |
d1ccc66d IM |
4939 | /* Avoid rq from going away on us: */ |
4940 | preempt_disable(); | |
eb580751 | 4941 | __task_rq_unlock(rq, &rf); |
4c9a4bc8 PZ |
4942 | |
4943 | balance_callback(rq); | |
4944 | preempt_enable(); | |
b29739f9 | 4945 | } |
acd58620 PZ |
4946 | #else |
4947 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
4948 | { | |
4949 | return prio; | |
4950 | } | |
b29739f9 | 4951 | #endif |
d50dde5a | 4952 | |
36c8b586 | 4953 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4954 | { |
49bd21ef | 4955 | bool queued, running; |
53a23364 | 4956 | int old_prio; |
eb580751 | 4957 | struct rq_flags rf; |
70b97a7f | 4958 | struct rq *rq; |
1da177e4 | 4959 | |
75e45d51 | 4960 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
4961 | return; |
4962 | /* | |
4963 | * We have to be careful, if called from sys_setpriority(), | |
4964 | * the task might be in the middle of scheduling on another CPU. | |
4965 | */ | |
eb580751 | 4966 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
4967 | update_rq_clock(rq); |
4968 | ||
1da177e4 LT |
4969 | /* |
4970 | * The RT priorities are set via sched_setscheduler(), but we still | |
4971 | * allow the 'normal' nice value to be set - but as expected | |
4972 | * it wont have any effect on scheduling until the task is | |
aab03e05 | 4973 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 4974 | */ |
aab03e05 | 4975 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
4976 | p->static_prio = NICE_TO_PRIO(nice); |
4977 | goto out_unlock; | |
4978 | } | |
da0c1e65 | 4979 | queued = task_on_rq_queued(p); |
49bd21ef | 4980 | running = task_current(rq, p); |
da0c1e65 | 4981 | if (queued) |
7a57f32a | 4982 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
4983 | if (running) |
4984 | put_prev_task(rq, p); | |
1da177e4 | 4985 | |
1da177e4 | 4986 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 4987 | set_load_weight(p, true); |
b29739f9 IM |
4988 | old_prio = p->prio; |
4989 | p->prio = effective_prio(p); | |
1da177e4 | 4990 | |
5443a0be | 4991 | if (queued) |
7134b3e9 | 4992 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 4993 | if (running) |
03b7fad1 | 4994 | set_next_task(rq, p); |
5443a0be FW |
4995 | |
4996 | /* | |
4997 | * If the task increased its priority or is running and | |
4998 | * lowered its priority, then reschedule its CPU: | |
4999 | */ | |
5000 | p->sched_class->prio_changed(rq, p, old_prio); | |
5001 | ||
1da177e4 | 5002 | out_unlock: |
eb580751 | 5003 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 5004 | } |
1da177e4 LT |
5005 | EXPORT_SYMBOL(set_user_nice); |
5006 | ||
e43379f1 MM |
5007 | /* |
5008 | * can_nice - check if a task can reduce its nice value | |
5009 | * @p: task | |
5010 | * @nice: nice value | |
5011 | */ | |
36c8b586 | 5012 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5013 | { |
d1ccc66d | 5014 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 5015 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 5016 | |
78d7d407 | 5017 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
5018 | capable(CAP_SYS_NICE)); |
5019 | } | |
5020 | ||
1da177e4 LT |
5021 | #ifdef __ARCH_WANT_SYS_NICE |
5022 | ||
5023 | /* | |
5024 | * sys_nice - change the priority of the current process. | |
5025 | * @increment: priority increment | |
5026 | * | |
5027 | * sys_setpriority is a more generic, but much slower function that | |
5028 | * does similar things. | |
5029 | */ | |
5add95d4 | 5030 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5031 | { |
48f24c4d | 5032 | long nice, retval; |
1da177e4 LT |
5033 | |
5034 | /* | |
5035 | * Setpriority might change our priority at the same moment. | |
5036 | * We don't have to worry. Conceptually one call occurs first | |
5037 | * and we have a single winner. | |
5038 | */ | |
a9467fa3 | 5039 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 5040 | nice = task_nice(current) + increment; |
1da177e4 | 5041 | |
a9467fa3 | 5042 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
5043 | if (increment < 0 && !can_nice(current, nice)) |
5044 | return -EPERM; | |
5045 | ||
1da177e4 LT |
5046 | retval = security_task_setnice(current, nice); |
5047 | if (retval) | |
5048 | return retval; | |
5049 | ||
5050 | set_user_nice(current, nice); | |
5051 | return 0; | |
5052 | } | |
5053 | ||
5054 | #endif | |
5055 | ||
5056 | /** | |
5057 | * task_prio - return the priority value of a given task. | |
5058 | * @p: the task in question. | |
5059 | * | |
e69f6186 | 5060 | * Return: The priority value as seen by users in /proc. |
1da177e4 LT |
5061 | * RT tasks are offset by -200. Normal tasks are centered |
5062 | * around 0, value goes from -16 to +15. | |
5063 | */ | |
36c8b586 | 5064 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5065 | { |
5066 | return p->prio - MAX_RT_PRIO; | |
5067 | } | |
5068 | ||
1da177e4 | 5069 | /** |
d1ccc66d | 5070 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 5071 | * @cpu: the processor in question. |
e69f6186 YB |
5072 | * |
5073 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
5074 | */ |
5075 | int idle_cpu(int cpu) | |
5076 | { | |
908a3283 TG |
5077 | struct rq *rq = cpu_rq(cpu); |
5078 | ||
5079 | if (rq->curr != rq->idle) | |
5080 | return 0; | |
5081 | ||
5082 | if (rq->nr_running) | |
5083 | return 0; | |
5084 | ||
5085 | #ifdef CONFIG_SMP | |
126c2092 | 5086 | if (rq->ttwu_pending) |
908a3283 TG |
5087 | return 0; |
5088 | #endif | |
5089 | ||
5090 | return 1; | |
1da177e4 LT |
5091 | } |
5092 | ||
943d355d RJ |
5093 | /** |
5094 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
5095 | * @cpu: the CPU in question. | |
5096 | * | |
5097 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
5098 | */ | |
5099 | int available_idle_cpu(int cpu) | |
5100 | { | |
5101 | if (!idle_cpu(cpu)) | |
5102 | return 0; | |
5103 | ||
247f2f6f RJ |
5104 | if (vcpu_is_preempted(cpu)) |
5105 | return 0; | |
5106 | ||
908a3283 | 5107 | return 1; |
1da177e4 LT |
5108 | } |
5109 | ||
1da177e4 | 5110 | /** |
d1ccc66d | 5111 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 5112 | * @cpu: the processor in question. |
e69f6186 | 5113 | * |
d1ccc66d | 5114 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 5115 | */ |
36c8b586 | 5116 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5117 | { |
5118 | return cpu_rq(cpu)->idle; | |
5119 | } | |
5120 | ||
5121 | /** | |
5122 | * find_process_by_pid - find a process with a matching PID value. | |
5123 | * @pid: the pid in question. | |
e69f6186 YB |
5124 | * |
5125 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 5126 | */ |
a9957449 | 5127 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5128 | { |
228ebcbe | 5129 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5130 | } |
5131 | ||
c13db6b1 SR |
5132 | /* |
5133 | * sched_setparam() passes in -1 for its policy, to let the functions | |
5134 | * it calls know not to change it. | |
5135 | */ | |
5136 | #define SETPARAM_POLICY -1 | |
5137 | ||
c365c292 TG |
5138 | static void __setscheduler_params(struct task_struct *p, |
5139 | const struct sched_attr *attr) | |
1da177e4 | 5140 | { |
d50dde5a DF |
5141 | int policy = attr->sched_policy; |
5142 | ||
c13db6b1 | 5143 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
5144 | policy = p->policy; |
5145 | ||
1da177e4 | 5146 | p->policy = policy; |
d50dde5a | 5147 | |
aab03e05 DF |
5148 | if (dl_policy(policy)) |
5149 | __setparam_dl(p, attr); | |
39fd8fd2 | 5150 | else if (fair_policy(policy)) |
d50dde5a DF |
5151 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
5152 | ||
39fd8fd2 PZ |
5153 | /* |
5154 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
5155 | * !rt_policy. Always setting this ensures that things like | |
5156 | * getparam()/getattr() don't report silly values for !rt tasks. | |
5157 | */ | |
5158 | p->rt_priority = attr->sched_priority; | |
383afd09 | 5159 | p->normal_prio = normal_prio(p); |
9059393e | 5160 | set_load_weight(p, true); |
c365c292 | 5161 | } |
39fd8fd2 | 5162 | |
c365c292 TG |
5163 | /* Actually do priority change: must hold pi & rq lock. */ |
5164 | static void __setscheduler(struct rq *rq, struct task_struct *p, | |
0782e63b | 5165 | const struct sched_attr *attr, bool keep_boost) |
c365c292 | 5166 | { |
a509a7cd PB |
5167 | /* |
5168 | * If params can't change scheduling class changes aren't allowed | |
5169 | * either. | |
5170 | */ | |
5171 | if (attr->sched_flags & SCHED_FLAG_KEEP_PARAMS) | |
5172 | return; | |
5173 | ||
c365c292 | 5174 | __setscheduler_params(p, attr); |
d50dde5a | 5175 | |
383afd09 | 5176 | /* |
0782e63b TG |
5177 | * Keep a potential priority boosting if called from |
5178 | * sched_setscheduler(). | |
383afd09 | 5179 | */ |
acd58620 | 5180 | p->prio = normal_prio(p); |
0782e63b | 5181 | if (keep_boost) |
acd58620 | 5182 | p->prio = rt_effective_prio(p, p->prio); |
383afd09 | 5183 | |
aab03e05 DF |
5184 | if (dl_prio(p->prio)) |
5185 | p->sched_class = &dl_sched_class; | |
5186 | else if (rt_prio(p->prio)) | |
ffd44db5 PZ |
5187 | p->sched_class = &rt_sched_class; |
5188 | else | |
5189 | p->sched_class = &fair_sched_class; | |
1da177e4 | 5190 | } |
aab03e05 | 5191 | |
c69e8d9c | 5192 | /* |
d1ccc66d | 5193 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
5194 | */ |
5195 | static bool check_same_owner(struct task_struct *p) | |
5196 | { | |
5197 | const struct cred *cred = current_cred(), *pcred; | |
5198 | bool match; | |
5199 | ||
5200 | rcu_read_lock(); | |
5201 | pcred = __task_cred(p); | |
9c806aa0 EB |
5202 | match = (uid_eq(cred->euid, pcred->euid) || |
5203 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
5204 | rcu_read_unlock(); |
5205 | return match; | |
5206 | } | |
5207 | ||
d50dde5a DF |
5208 | static int __sched_setscheduler(struct task_struct *p, |
5209 | const struct sched_attr *attr, | |
dbc7f069 | 5210 | bool user, bool pi) |
1da177e4 | 5211 | { |
383afd09 SR |
5212 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
5213 | MAX_RT_PRIO - 1 - attr->sched_priority; | |
da0c1e65 | 5214 | int retval, oldprio, oldpolicy = -1, queued, running; |
0782e63b | 5215 | int new_effective_prio, policy = attr->sched_policy; |
83ab0aa0 | 5216 | const struct sched_class *prev_class; |
eb580751 | 5217 | struct rq_flags rf; |
ca94c442 | 5218 | int reset_on_fork; |
7a57f32a | 5219 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 5220 | struct rq *rq; |
1da177e4 | 5221 | |
896bbb25 SRV |
5222 | /* The pi code expects interrupts enabled */ |
5223 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 5224 | recheck: |
d1ccc66d | 5225 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
5226 | if (policy < 0) { |
5227 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5228 | policy = oldpolicy = p->policy; |
ca94c442 | 5229 | } else { |
7479f3c9 | 5230 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 5231 | |
20f9cd2a | 5232 | if (!valid_policy(policy)) |
ca94c442 LP |
5233 | return -EINVAL; |
5234 | } | |
5235 | ||
794a56eb | 5236 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
5237 | return -EINVAL; |
5238 | ||
1da177e4 LT |
5239 | /* |
5240 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5241 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5242 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 | 5243 | */ |
0bb040a4 | 5244 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
d50dde5a | 5245 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5246 | return -EINVAL; |
aab03e05 DF |
5247 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
5248 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
5249 | return -EINVAL; |
5250 | ||
37e4ab3f OC |
5251 | /* |
5252 | * Allow unprivileged RT tasks to decrease priority: | |
5253 | */ | |
961ccddd | 5254 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 5255 | if (fair_policy(policy)) { |
d0ea0268 | 5256 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 5257 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
5258 | return -EPERM; |
5259 | } | |
5260 | ||
e05606d3 | 5261 | if (rt_policy(policy)) { |
a44702e8 ON |
5262 | unsigned long rlim_rtprio = |
5263 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 5264 | |
d1ccc66d | 5265 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
5266 | if (policy != p->policy && !rlim_rtprio) |
5267 | return -EPERM; | |
5268 | ||
d1ccc66d | 5269 | /* Can't increase priority: */ |
d50dde5a DF |
5270 | if (attr->sched_priority > p->rt_priority && |
5271 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
5272 | return -EPERM; |
5273 | } | |
c02aa73b | 5274 | |
d44753b8 JL |
5275 | /* |
5276 | * Can't set/change SCHED_DEADLINE policy at all for now | |
5277 | * (safest behavior); in the future we would like to allow | |
5278 | * unprivileged DL tasks to increase their relative deadline | |
5279 | * or reduce their runtime (both ways reducing utilization) | |
5280 | */ | |
5281 | if (dl_policy(policy)) | |
5282 | return -EPERM; | |
5283 | ||
dd41f596 | 5284 | /* |
c02aa73b DH |
5285 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5286 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5287 | */ |
1da1843f | 5288 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 5289 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
5290 | return -EPERM; |
5291 | } | |
5fe1d75f | 5292 | |
d1ccc66d | 5293 | /* Can't change other user's priorities: */ |
c69e8d9c | 5294 | if (!check_same_owner(p)) |
37e4ab3f | 5295 | return -EPERM; |
ca94c442 | 5296 | |
d1ccc66d | 5297 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
5298 | if (p->sched_reset_on_fork && !reset_on_fork) |
5299 | return -EPERM; | |
37e4ab3f | 5300 | } |
1da177e4 | 5301 | |
725aad24 | 5302 | if (user) { |
794a56eb JL |
5303 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
5304 | return -EINVAL; | |
5305 | ||
b0ae1981 | 5306 | retval = security_task_setscheduler(p); |
725aad24 JF |
5307 | if (retval) |
5308 | return retval; | |
5309 | } | |
5310 | ||
a509a7cd PB |
5311 | /* Update task specific "requested" clamps */ |
5312 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
5313 | retval = uclamp_validate(p, attr); | |
5314 | if (retval) | |
5315 | return retval; | |
5316 | } | |
5317 | ||
710da3c8 JL |
5318 | if (pi) |
5319 | cpuset_read_lock(); | |
5320 | ||
b29739f9 | 5321 | /* |
d1ccc66d | 5322 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 5323 | * changing the priority of the task: |
0122ec5b | 5324 | * |
25985edc | 5325 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5326 | * runqueue lock must be held. |
5327 | */ | |
eb580751 | 5328 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 5329 | update_rq_clock(rq); |
dc61b1d6 | 5330 | |
34f971f6 | 5331 | /* |
d1ccc66d | 5332 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
5333 | */ |
5334 | if (p == rq->stop) { | |
4b211f2b MP |
5335 | retval = -EINVAL; |
5336 | goto unlock; | |
34f971f6 PZ |
5337 | } |
5338 | ||
a51e9198 | 5339 | /* |
d6b1e911 TG |
5340 | * If not changing anything there's no need to proceed further, |
5341 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 5342 | */ |
d50dde5a | 5343 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 5344 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
5345 | goto change; |
5346 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
5347 | goto change; | |
75381608 | 5348 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 5349 | goto change; |
a509a7cd PB |
5350 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
5351 | goto change; | |
d50dde5a | 5352 | |
d6b1e911 | 5353 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
5354 | retval = 0; |
5355 | goto unlock; | |
a51e9198 | 5356 | } |
d50dde5a | 5357 | change: |
a51e9198 | 5358 | |
dc61b1d6 | 5359 | if (user) { |
332ac17e | 5360 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
5361 | /* |
5362 | * Do not allow realtime tasks into groups that have no runtime | |
5363 | * assigned. | |
5364 | */ | |
5365 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5366 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5367 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
5368 | retval = -EPERM; |
5369 | goto unlock; | |
dc61b1d6 | 5370 | } |
dc61b1d6 | 5371 | #endif |
332ac17e | 5372 | #ifdef CONFIG_SMP |
794a56eb JL |
5373 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
5374 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 5375 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
5376 | |
5377 | /* | |
5378 | * Don't allow tasks with an affinity mask smaller than | |
5379 | * the entire root_domain to become SCHED_DEADLINE. We | |
5380 | * will also fail if there's no bandwidth available. | |
5381 | */ | |
3bd37062 | 5382 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 5383 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
5384 | retval = -EPERM; |
5385 | goto unlock; | |
332ac17e DF |
5386 | } |
5387 | } | |
5388 | #endif | |
5389 | } | |
dc61b1d6 | 5390 | |
d1ccc66d | 5391 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
5392 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
5393 | policy = oldpolicy = -1; | |
eb580751 | 5394 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
5395 | if (pi) |
5396 | cpuset_read_unlock(); | |
1da177e4 LT |
5397 | goto recheck; |
5398 | } | |
332ac17e DF |
5399 | |
5400 | /* | |
5401 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
5402 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
5403 | * is available. | |
5404 | */ | |
06a76fe0 | 5405 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
5406 | retval = -EBUSY; |
5407 | goto unlock; | |
332ac17e DF |
5408 | } |
5409 | ||
c365c292 TG |
5410 | p->sched_reset_on_fork = reset_on_fork; |
5411 | oldprio = p->prio; | |
5412 | ||
dbc7f069 PZ |
5413 | if (pi) { |
5414 | /* | |
5415 | * Take priority boosted tasks into account. If the new | |
5416 | * effective priority is unchanged, we just store the new | |
5417 | * normal parameters and do not touch the scheduler class and | |
5418 | * the runqueue. This will be done when the task deboost | |
5419 | * itself. | |
5420 | */ | |
acd58620 | 5421 | new_effective_prio = rt_effective_prio(p, newprio); |
ff77e468 PZ |
5422 | if (new_effective_prio == oldprio) |
5423 | queue_flags &= ~DEQUEUE_MOVE; | |
c365c292 TG |
5424 | } |
5425 | ||
da0c1e65 | 5426 | queued = task_on_rq_queued(p); |
051a1d1a | 5427 | running = task_current(rq, p); |
da0c1e65 | 5428 | if (queued) |
ff77e468 | 5429 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 5430 | if (running) |
f3cd1c4e | 5431 | put_prev_task(rq, p); |
f6b53205 | 5432 | |
83ab0aa0 | 5433 | prev_class = p->sched_class; |
a509a7cd | 5434 | |
dbc7f069 | 5435 | __setscheduler(rq, p, attr, pi); |
a509a7cd | 5436 | __setscheduler_uclamp(p, attr); |
f6b53205 | 5437 | |
da0c1e65 | 5438 | if (queued) { |
81a44c54 TG |
5439 | /* |
5440 | * We enqueue to tail when the priority of a task is | |
5441 | * increased (user space view). | |
5442 | */ | |
ff77e468 PZ |
5443 | if (oldprio < p->prio) |
5444 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 5445 | |
ff77e468 | 5446 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 5447 | } |
a399d233 | 5448 | if (running) |
03b7fad1 | 5449 | set_next_task(rq, p); |
cb469845 | 5450 | |
da7a735e | 5451 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
5452 | |
5453 | /* Avoid rq from going away on us: */ | |
5454 | preempt_disable(); | |
eb580751 | 5455 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 5456 | |
710da3c8 JL |
5457 | if (pi) { |
5458 | cpuset_read_unlock(); | |
dbc7f069 | 5459 | rt_mutex_adjust_pi(p); |
710da3c8 | 5460 | } |
95e02ca9 | 5461 | |
d1ccc66d | 5462 | /* Run balance callbacks after we've adjusted the PI chain: */ |
4c9a4bc8 PZ |
5463 | balance_callback(rq); |
5464 | preempt_enable(); | |
95e02ca9 | 5465 | |
1da177e4 | 5466 | return 0; |
4b211f2b MP |
5467 | |
5468 | unlock: | |
5469 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
5470 | if (pi) |
5471 | cpuset_read_unlock(); | |
4b211f2b | 5472 | return retval; |
1da177e4 | 5473 | } |
961ccddd | 5474 | |
7479f3c9 PZ |
5475 | static int _sched_setscheduler(struct task_struct *p, int policy, |
5476 | const struct sched_param *param, bool check) | |
5477 | { | |
5478 | struct sched_attr attr = { | |
5479 | .sched_policy = policy, | |
5480 | .sched_priority = param->sched_priority, | |
5481 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
5482 | }; | |
5483 | ||
c13db6b1 SR |
5484 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
5485 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
5486 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
5487 | policy &= ~SCHED_RESET_ON_FORK; | |
5488 | attr.sched_policy = policy; | |
5489 | } | |
5490 | ||
dbc7f069 | 5491 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 5492 | } |
961ccddd RR |
5493 | /** |
5494 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5495 | * @p: the task in question. | |
5496 | * @policy: new policy. | |
5497 | * @param: structure containing the new RT priority. | |
5498 | * | |
7318d4cc PZ |
5499 | * Use sched_set_fifo(), read its comment. |
5500 | * | |
e69f6186 YB |
5501 | * Return: 0 on success. An error code otherwise. |
5502 | * | |
961ccddd RR |
5503 | * NOTE that the task may be already dead. |
5504 | */ | |
5505 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5506 | const struct sched_param *param) |
961ccddd | 5507 | { |
7479f3c9 | 5508 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 5509 | } |
1da177e4 | 5510 | |
d50dde5a DF |
5511 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
5512 | { | |
dbc7f069 | 5513 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 5514 | } |
d50dde5a | 5515 | |
794a56eb JL |
5516 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
5517 | { | |
5518 | return __sched_setscheduler(p, attr, false, true); | |
5519 | } | |
5520 | ||
961ccddd RR |
5521 | /** |
5522 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5523 | * @p: the task in question. | |
5524 | * @policy: new policy. | |
5525 | * @param: structure containing the new RT priority. | |
5526 | * | |
5527 | * Just like sched_setscheduler, only don't bother checking if the | |
5528 | * current context has permission. For example, this is needed in | |
5529 | * stop_machine(): we create temporary high priority worker threads, | |
5530 | * but our caller might not have that capability. | |
e69f6186 YB |
5531 | * |
5532 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
5533 | */ |
5534 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5535 | const struct sched_param *param) |
961ccddd | 5536 | { |
7479f3c9 | 5537 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
5538 | } |
5539 | ||
7318d4cc PZ |
5540 | /* |
5541 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
5542 | * incapable of resource management, which is the one thing an OS really should | |
5543 | * be doing. | |
5544 | * | |
5545 | * This is of course the reason it is limited to privileged users only. | |
5546 | * | |
5547 | * Worse still; it is fundamentally impossible to compose static priority | |
5548 | * workloads. You cannot take two correctly working static prio workloads | |
5549 | * and smash them together and still expect them to work. | |
5550 | * | |
5551 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
5552 | * | |
5553 | * MAX_RT_PRIO / 2 | |
5554 | * | |
5555 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
5556 | * know enough information to make a sensible choice. | |
5557 | */ | |
8b700983 | 5558 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
5559 | { |
5560 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 5561 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
5562 | } |
5563 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
5564 | ||
5565 | /* | |
5566 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
5567 | */ | |
8b700983 | 5568 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
5569 | { |
5570 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 5571 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
5572 | } |
5573 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
5574 | ||
8b700983 | 5575 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
5576 | { |
5577 | struct sched_attr attr = { | |
5578 | .sched_policy = SCHED_NORMAL, | |
5579 | .sched_nice = nice, | |
5580 | }; | |
8b700983 | 5581 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
5582 | } |
5583 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 5584 | |
95cdf3b7 IM |
5585 | static int |
5586 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5587 | { |
1da177e4 LT |
5588 | struct sched_param lparam; |
5589 | struct task_struct *p; | |
36c8b586 | 5590 | int retval; |
1da177e4 LT |
5591 | |
5592 | if (!param || pid < 0) | |
5593 | return -EINVAL; | |
5594 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5595 | return -EFAULT; | |
5fe1d75f ON |
5596 | |
5597 | rcu_read_lock(); | |
5598 | retval = -ESRCH; | |
1da177e4 | 5599 | p = find_process_by_pid(pid); |
710da3c8 JL |
5600 | if (likely(p)) |
5601 | get_task_struct(p); | |
5fe1d75f | 5602 | rcu_read_unlock(); |
36c8b586 | 5603 | |
710da3c8 JL |
5604 | if (likely(p)) { |
5605 | retval = sched_setscheduler(p, policy, &lparam); | |
5606 | put_task_struct(p); | |
5607 | } | |
5608 | ||
1da177e4 LT |
5609 | return retval; |
5610 | } | |
5611 | ||
d50dde5a DF |
5612 | /* |
5613 | * Mimics kernel/events/core.c perf_copy_attr(). | |
5614 | */ | |
d1ccc66d | 5615 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
5616 | { |
5617 | u32 size; | |
5618 | int ret; | |
5619 | ||
d1ccc66d | 5620 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
5621 | memset(attr, 0, sizeof(*attr)); |
5622 | ||
5623 | ret = get_user(size, &uattr->size); | |
5624 | if (ret) | |
5625 | return ret; | |
5626 | ||
d1ccc66d IM |
5627 | /* ABI compatibility quirk: */ |
5628 | if (!size) | |
d50dde5a | 5629 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 5630 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
5631 | goto err_size; |
5632 | ||
dff3a85f AS |
5633 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
5634 | if (ret) { | |
5635 | if (ret == -E2BIG) | |
5636 | goto err_size; | |
5637 | return ret; | |
d50dde5a DF |
5638 | } |
5639 | ||
a509a7cd PB |
5640 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
5641 | size < SCHED_ATTR_SIZE_VER1) | |
5642 | return -EINVAL; | |
5643 | ||
d50dde5a | 5644 | /* |
d1ccc66d | 5645 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
5646 | * to be strict and return an error on out-of-bounds values? |
5647 | */ | |
75e45d51 | 5648 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 5649 | |
e78c7bca | 5650 | return 0; |
d50dde5a DF |
5651 | |
5652 | err_size: | |
5653 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 5654 | return -E2BIG; |
d50dde5a DF |
5655 | } |
5656 | ||
1da177e4 LT |
5657 | /** |
5658 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5659 | * @pid: the pid in question. | |
5660 | * @policy: new policy. | |
5661 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
5662 | * |
5663 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5664 | */ |
d1ccc66d | 5665 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 5666 | { |
c21761f1 JB |
5667 | if (policy < 0) |
5668 | return -EINVAL; | |
5669 | ||
1da177e4 LT |
5670 | return do_sched_setscheduler(pid, policy, param); |
5671 | } | |
5672 | ||
5673 | /** | |
5674 | * sys_sched_setparam - set/change the RT priority of a thread | |
5675 | * @pid: the pid in question. | |
5676 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
5677 | * |
5678 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5679 | */ |
5add95d4 | 5680 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 5681 | { |
c13db6b1 | 5682 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
5683 | } |
5684 | ||
d50dde5a DF |
5685 | /** |
5686 | * sys_sched_setattr - same as above, but with extended sched_attr | |
5687 | * @pid: the pid in question. | |
5778fccf | 5688 | * @uattr: structure containing the extended parameters. |
db66d756 | 5689 | * @flags: for future extension. |
d50dde5a | 5690 | */ |
6d35ab48 PZ |
5691 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
5692 | unsigned int, flags) | |
d50dde5a DF |
5693 | { |
5694 | struct sched_attr attr; | |
5695 | struct task_struct *p; | |
5696 | int retval; | |
5697 | ||
6d35ab48 | 5698 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
5699 | return -EINVAL; |
5700 | ||
143cf23d MK |
5701 | retval = sched_copy_attr(uattr, &attr); |
5702 | if (retval) | |
5703 | return retval; | |
d50dde5a | 5704 | |
b14ed2c2 | 5705 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 5706 | return -EINVAL; |
1d6362fa PB |
5707 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
5708 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
5709 | |
5710 | rcu_read_lock(); | |
5711 | retval = -ESRCH; | |
5712 | p = find_process_by_pid(pid); | |
a509a7cd PB |
5713 | if (likely(p)) |
5714 | get_task_struct(p); | |
d50dde5a DF |
5715 | rcu_read_unlock(); |
5716 | ||
a509a7cd PB |
5717 | if (likely(p)) { |
5718 | retval = sched_setattr(p, &attr); | |
5719 | put_task_struct(p); | |
5720 | } | |
5721 | ||
d50dde5a DF |
5722 | return retval; |
5723 | } | |
5724 | ||
1da177e4 LT |
5725 | /** |
5726 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5727 | * @pid: the pid in question. | |
e69f6186 YB |
5728 | * |
5729 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
5730 | * code. | |
1da177e4 | 5731 | */ |
5add95d4 | 5732 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5733 | { |
36c8b586 | 5734 | struct task_struct *p; |
3a5c359a | 5735 | int retval; |
1da177e4 LT |
5736 | |
5737 | if (pid < 0) | |
3a5c359a | 5738 | return -EINVAL; |
1da177e4 LT |
5739 | |
5740 | retval = -ESRCH; | |
5fe85be0 | 5741 | rcu_read_lock(); |
1da177e4 LT |
5742 | p = find_process_by_pid(pid); |
5743 | if (p) { | |
5744 | retval = security_task_getscheduler(p); | |
5745 | if (!retval) | |
ca94c442 LP |
5746 | retval = p->policy |
5747 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5748 | } |
5fe85be0 | 5749 | rcu_read_unlock(); |
1da177e4 LT |
5750 | return retval; |
5751 | } | |
5752 | ||
5753 | /** | |
ca94c442 | 5754 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5755 | * @pid: the pid in question. |
5756 | * @param: structure containing the RT priority. | |
e69f6186 YB |
5757 | * |
5758 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
5759 | * code. | |
1da177e4 | 5760 | */ |
5add95d4 | 5761 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 5762 | { |
ce5f7f82 | 5763 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 5764 | struct task_struct *p; |
3a5c359a | 5765 | int retval; |
1da177e4 LT |
5766 | |
5767 | if (!param || pid < 0) | |
3a5c359a | 5768 | return -EINVAL; |
1da177e4 | 5769 | |
5fe85be0 | 5770 | rcu_read_lock(); |
1da177e4 LT |
5771 | p = find_process_by_pid(pid); |
5772 | retval = -ESRCH; | |
5773 | if (!p) | |
5774 | goto out_unlock; | |
5775 | ||
5776 | retval = security_task_getscheduler(p); | |
5777 | if (retval) | |
5778 | goto out_unlock; | |
5779 | ||
ce5f7f82 PZ |
5780 | if (task_has_rt_policy(p)) |
5781 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5782 | rcu_read_unlock(); |
1da177e4 LT |
5783 | |
5784 | /* | |
5785 | * This one might sleep, we cannot do it with a spinlock held ... | |
5786 | */ | |
5787 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5788 | ||
1da177e4 LT |
5789 | return retval; |
5790 | ||
5791 | out_unlock: | |
5fe85be0 | 5792 | rcu_read_unlock(); |
1da177e4 LT |
5793 | return retval; |
5794 | } | |
5795 | ||
1251201c IM |
5796 | /* |
5797 | * Copy the kernel size attribute structure (which might be larger | |
5798 | * than what user-space knows about) to user-space. | |
5799 | * | |
5800 | * Note that all cases are valid: user-space buffer can be larger or | |
5801 | * smaller than the kernel-space buffer. The usual case is that both | |
5802 | * have the same size. | |
5803 | */ | |
5804 | static int | |
5805 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
5806 | struct sched_attr *kattr, | |
5807 | unsigned int usize) | |
d50dde5a | 5808 | { |
1251201c | 5809 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 5810 | |
96d4f267 | 5811 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
5812 | return -EFAULT; |
5813 | ||
5814 | /* | |
1251201c IM |
5815 | * sched_getattr() ABI forwards and backwards compatibility: |
5816 | * | |
5817 | * If usize == ksize then we just copy everything to user-space and all is good. | |
5818 | * | |
5819 | * If usize < ksize then we only copy as much as user-space has space for, | |
5820 | * this keeps ABI compatibility as well. We skip the rest. | |
5821 | * | |
5822 | * If usize > ksize then user-space is using a newer version of the ABI, | |
5823 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
5824 | * detect the kernel's knowledge of attributes from the attr->size value | |
5825 | * which is set to ksize in this case. | |
d50dde5a | 5826 | */ |
1251201c | 5827 | kattr->size = min(usize, ksize); |
d50dde5a | 5828 | |
1251201c | 5829 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
5830 | return -EFAULT; |
5831 | ||
22400674 | 5832 | return 0; |
d50dde5a DF |
5833 | } |
5834 | ||
5835 | /** | |
aab03e05 | 5836 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 5837 | * @pid: the pid in question. |
5778fccf | 5838 | * @uattr: structure containing the extended parameters. |
dff3a85f | 5839 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 5840 | * @flags: for future extension. |
d50dde5a | 5841 | */ |
6d35ab48 | 5842 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 5843 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 5844 | { |
1251201c | 5845 | struct sched_attr kattr = { }; |
d50dde5a DF |
5846 | struct task_struct *p; |
5847 | int retval; | |
5848 | ||
1251201c IM |
5849 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
5850 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
5851 | return -EINVAL; |
5852 | ||
5853 | rcu_read_lock(); | |
5854 | p = find_process_by_pid(pid); | |
5855 | retval = -ESRCH; | |
5856 | if (!p) | |
5857 | goto out_unlock; | |
5858 | ||
5859 | retval = security_task_getscheduler(p); | |
5860 | if (retval) | |
5861 | goto out_unlock; | |
5862 | ||
1251201c | 5863 | kattr.sched_policy = p->policy; |
7479f3c9 | 5864 | if (p->sched_reset_on_fork) |
1251201c | 5865 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
aab03e05 | 5866 | if (task_has_dl_policy(p)) |
1251201c | 5867 | __getparam_dl(p, &kattr); |
aab03e05 | 5868 | else if (task_has_rt_policy(p)) |
1251201c | 5869 | kattr.sched_priority = p->rt_priority; |
d50dde5a | 5870 | else |
1251201c | 5871 | kattr.sched_nice = task_nice(p); |
d50dde5a | 5872 | |
a509a7cd | 5873 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
5874 | /* |
5875 | * This could race with another potential updater, but this is fine | |
5876 | * because it'll correctly read the old or the new value. We don't need | |
5877 | * to guarantee who wins the race as long as it doesn't return garbage. | |
5878 | */ | |
1251201c IM |
5879 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
5880 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
5881 | #endif |
5882 | ||
d50dde5a DF |
5883 | rcu_read_unlock(); |
5884 | ||
1251201c | 5885 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
5886 | |
5887 | out_unlock: | |
5888 | rcu_read_unlock(); | |
5889 | return retval; | |
5890 | } | |
5891 | ||
96f874e2 | 5892 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5893 | { |
5a16f3d3 | 5894 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5895 | struct task_struct *p; |
5896 | int retval; | |
1da177e4 | 5897 | |
23f5d142 | 5898 | rcu_read_lock(); |
1da177e4 LT |
5899 | |
5900 | p = find_process_by_pid(pid); | |
5901 | if (!p) { | |
23f5d142 | 5902 | rcu_read_unlock(); |
1da177e4 LT |
5903 | return -ESRCH; |
5904 | } | |
5905 | ||
23f5d142 | 5906 | /* Prevent p going away */ |
1da177e4 | 5907 | get_task_struct(p); |
23f5d142 | 5908 | rcu_read_unlock(); |
1da177e4 | 5909 | |
14a40ffc TH |
5910 | if (p->flags & PF_NO_SETAFFINITY) { |
5911 | retval = -EINVAL; | |
5912 | goto out_put_task; | |
5913 | } | |
5a16f3d3 RR |
5914 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5915 | retval = -ENOMEM; | |
5916 | goto out_put_task; | |
5917 | } | |
5918 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5919 | retval = -ENOMEM; | |
5920 | goto out_free_cpus_allowed; | |
5921 | } | |
1da177e4 | 5922 | retval = -EPERM; |
4c44aaaf EB |
5923 | if (!check_same_owner(p)) { |
5924 | rcu_read_lock(); | |
5925 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
5926 | rcu_read_unlock(); | |
16303ab2 | 5927 | goto out_free_new_mask; |
4c44aaaf EB |
5928 | } |
5929 | rcu_read_unlock(); | |
5930 | } | |
1da177e4 | 5931 | |
b0ae1981 | 5932 | retval = security_task_setscheduler(p); |
e7834f8f | 5933 | if (retval) |
16303ab2 | 5934 | goto out_free_new_mask; |
e7834f8f | 5935 | |
e4099a5e PZ |
5936 | |
5937 | cpuset_cpus_allowed(p, cpus_allowed); | |
5938 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
5939 | ||
332ac17e DF |
5940 | /* |
5941 | * Since bandwidth control happens on root_domain basis, | |
5942 | * if admission test is enabled, we only admit -deadline | |
5943 | * tasks allowed to run on all the CPUs in the task's | |
5944 | * root_domain. | |
5945 | */ | |
5946 | #ifdef CONFIG_SMP | |
f1e3a093 KT |
5947 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
5948 | rcu_read_lock(); | |
5949 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { | |
332ac17e | 5950 | retval = -EBUSY; |
f1e3a093 | 5951 | rcu_read_unlock(); |
16303ab2 | 5952 | goto out_free_new_mask; |
332ac17e | 5953 | } |
f1e3a093 | 5954 | rcu_read_unlock(); |
332ac17e DF |
5955 | } |
5956 | #endif | |
49246274 | 5957 | again: |
25834c73 | 5958 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
1da177e4 | 5959 | |
8707d8b8 | 5960 | if (!retval) { |
5a16f3d3 RR |
5961 | cpuset_cpus_allowed(p, cpus_allowed); |
5962 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5963 | /* |
5964 | * We must have raced with a concurrent cpuset | |
5965 | * update. Just reset the cpus_allowed to the | |
5966 | * cpuset's cpus_allowed | |
5967 | */ | |
5a16f3d3 | 5968 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5969 | goto again; |
5970 | } | |
5971 | } | |
16303ab2 | 5972 | out_free_new_mask: |
5a16f3d3 RR |
5973 | free_cpumask_var(new_mask); |
5974 | out_free_cpus_allowed: | |
5975 | free_cpumask_var(cpus_allowed); | |
5976 | out_put_task: | |
1da177e4 | 5977 | put_task_struct(p); |
1da177e4 LT |
5978 | return retval; |
5979 | } | |
5980 | ||
5981 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5982 | struct cpumask *new_mask) |
1da177e4 | 5983 | { |
96f874e2 RR |
5984 | if (len < cpumask_size()) |
5985 | cpumask_clear(new_mask); | |
5986 | else if (len > cpumask_size()) | |
5987 | len = cpumask_size(); | |
5988 | ||
1da177e4 LT |
5989 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5990 | } | |
5991 | ||
5992 | /** | |
d1ccc66d | 5993 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
5994 | * @pid: pid of the process |
5995 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 5996 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
5997 | * |
5998 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 5999 | */ |
5add95d4 HC |
6000 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6001 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6002 | { |
5a16f3d3 | 6003 | cpumask_var_t new_mask; |
1da177e4 LT |
6004 | int retval; |
6005 | ||
5a16f3d3 RR |
6006 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6007 | return -ENOMEM; | |
1da177e4 | 6008 | |
5a16f3d3 RR |
6009 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6010 | if (retval == 0) | |
6011 | retval = sched_setaffinity(pid, new_mask); | |
6012 | free_cpumask_var(new_mask); | |
6013 | return retval; | |
1da177e4 LT |
6014 | } |
6015 | ||
96f874e2 | 6016 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6017 | { |
36c8b586 | 6018 | struct task_struct *p; |
31605683 | 6019 | unsigned long flags; |
1da177e4 | 6020 | int retval; |
1da177e4 | 6021 | |
23f5d142 | 6022 | rcu_read_lock(); |
1da177e4 LT |
6023 | |
6024 | retval = -ESRCH; | |
6025 | p = find_process_by_pid(pid); | |
6026 | if (!p) | |
6027 | goto out_unlock; | |
6028 | ||
e7834f8f DQ |
6029 | retval = security_task_getscheduler(p); |
6030 | if (retval) | |
6031 | goto out_unlock; | |
6032 | ||
013fdb80 | 6033 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 6034 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 6035 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
6036 | |
6037 | out_unlock: | |
23f5d142 | 6038 | rcu_read_unlock(); |
1da177e4 | 6039 | |
9531b62f | 6040 | return retval; |
1da177e4 LT |
6041 | } |
6042 | ||
6043 | /** | |
d1ccc66d | 6044 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
6045 | * @pid: pid of the process |
6046 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 6047 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 6048 | * |
599b4840 ZW |
6049 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
6050 | * error code otherwise. | |
1da177e4 | 6051 | */ |
5add95d4 HC |
6052 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6053 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6054 | { |
6055 | int ret; | |
f17c8607 | 6056 | cpumask_var_t mask; |
1da177e4 | 6057 | |
84fba5ec | 6058 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
6059 | return -EINVAL; |
6060 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
6061 | return -EINVAL; |
6062 | ||
f17c8607 RR |
6063 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6064 | return -ENOMEM; | |
1da177e4 | 6065 | |
f17c8607 RR |
6066 | ret = sched_getaffinity(pid, mask); |
6067 | if (ret == 0) { | |
4de373a1 | 6068 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
6069 | |
6070 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
6071 | ret = -EFAULT; |
6072 | else | |
cd3d8031 | 6073 | ret = retlen; |
f17c8607 RR |
6074 | } |
6075 | free_cpumask_var(mask); | |
1da177e4 | 6076 | |
f17c8607 | 6077 | return ret; |
1da177e4 LT |
6078 | } |
6079 | ||
6080 | /** | |
6081 | * sys_sched_yield - yield the current processor to other threads. | |
6082 | * | |
dd41f596 IM |
6083 | * This function yields the current CPU to other tasks. If there are no |
6084 | * other threads running on this CPU then this function will return. | |
e69f6186 YB |
6085 | * |
6086 | * Return: 0. | |
1da177e4 | 6087 | */ |
7d4dd4f1 | 6088 | static void do_sched_yield(void) |
1da177e4 | 6089 | { |
8a8c69c3 PZ |
6090 | struct rq_flags rf; |
6091 | struct rq *rq; | |
6092 | ||
246b3b33 | 6093 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 6094 | |
ae92882e | 6095 | schedstat_inc(rq->yld_count); |
4530d7ab | 6096 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6097 | |
6098 | /* | |
6099 | * Since we are going to call schedule() anyway, there's | |
6100 | * no need to preempt or enable interrupts: | |
6101 | */ | |
8a8c69c3 PZ |
6102 | preempt_disable(); |
6103 | rq_unlock(rq, &rf); | |
ba74c144 | 6104 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
6105 | |
6106 | schedule(); | |
7d4dd4f1 | 6107 | } |
1da177e4 | 6108 | |
7d4dd4f1 DB |
6109 | SYSCALL_DEFINE0(sched_yield) |
6110 | { | |
6111 | do_sched_yield(); | |
1da177e4 LT |
6112 | return 0; |
6113 | } | |
6114 | ||
c1a280b6 | 6115 | #ifndef CONFIG_PREEMPTION |
02b67cc3 | 6116 | int __sched _cond_resched(void) |
1da177e4 | 6117 | { |
fe32d3cd | 6118 | if (should_resched(0)) { |
a18b5d01 | 6119 | preempt_schedule_common(); |
1da177e4 LT |
6120 | return 1; |
6121 | } | |
f79c3ad6 | 6122 | rcu_all_qs(); |
1da177e4 LT |
6123 | return 0; |
6124 | } | |
02b67cc3 | 6125 | EXPORT_SYMBOL(_cond_resched); |
35a773a0 | 6126 | #endif |
1da177e4 LT |
6127 | |
6128 | /* | |
613afbf8 | 6129 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6130 | * call schedule, and on return reacquire the lock. |
6131 | * | |
c1a280b6 | 6132 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
6133 | * operations here to prevent schedule() from being called twice (once via |
6134 | * spin_unlock(), once by hand). | |
6135 | */ | |
613afbf8 | 6136 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6137 | { |
fe32d3cd | 6138 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
6139 | int ret = 0; |
6140 | ||
f607c668 PZ |
6141 | lockdep_assert_held(lock); |
6142 | ||
4a81e832 | 6143 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6144 | spin_unlock(lock); |
d86ee480 | 6145 | if (resched) |
a18b5d01 | 6146 | preempt_schedule_common(); |
95c354fe NP |
6147 | else |
6148 | cpu_relax(); | |
6df3cecb | 6149 | ret = 1; |
1da177e4 | 6150 | spin_lock(lock); |
1da177e4 | 6151 | } |
6df3cecb | 6152 | return ret; |
1da177e4 | 6153 | } |
613afbf8 | 6154 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6155 | |
1da177e4 LT |
6156 | /** |
6157 | * yield - yield the current processor to other threads. | |
6158 | * | |
8e3fabfd PZ |
6159 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
6160 | * | |
6161 | * The scheduler is at all times free to pick the calling task as the most | |
6162 | * eligible task to run, if removing the yield() call from your code breaks | |
6163 | * it, its already broken. | |
6164 | * | |
6165 | * Typical broken usage is: | |
6166 | * | |
6167 | * while (!event) | |
d1ccc66d | 6168 | * yield(); |
8e3fabfd PZ |
6169 | * |
6170 | * where one assumes that yield() will let 'the other' process run that will | |
6171 | * make event true. If the current task is a SCHED_FIFO task that will never | |
6172 | * happen. Never use yield() as a progress guarantee!! | |
6173 | * | |
6174 | * If you want to use yield() to wait for something, use wait_event(). | |
6175 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
6176 | * If you still want to use yield(), do not! | |
1da177e4 LT |
6177 | */ |
6178 | void __sched yield(void) | |
6179 | { | |
6180 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 6181 | do_sched_yield(); |
1da177e4 | 6182 | } |
1da177e4 LT |
6183 | EXPORT_SYMBOL(yield); |
6184 | ||
d95f4122 MG |
6185 | /** |
6186 | * yield_to - yield the current processor to another thread in | |
6187 | * your thread group, or accelerate that thread toward the | |
6188 | * processor it's on. | |
16addf95 RD |
6189 | * @p: target task |
6190 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
6191 | * |
6192 | * It's the caller's job to ensure that the target task struct | |
6193 | * can't go away on us before we can do any checks. | |
6194 | * | |
e69f6186 | 6195 | * Return: |
7b270f60 PZ |
6196 | * true (>0) if we indeed boosted the target task. |
6197 | * false (0) if we failed to boost the target. | |
6198 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 6199 | */ |
fa93384f | 6200 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
6201 | { |
6202 | struct task_struct *curr = current; | |
6203 | struct rq *rq, *p_rq; | |
6204 | unsigned long flags; | |
c3c18640 | 6205 | int yielded = 0; |
d95f4122 MG |
6206 | |
6207 | local_irq_save(flags); | |
6208 | rq = this_rq(); | |
6209 | ||
6210 | again: | |
6211 | p_rq = task_rq(p); | |
7b270f60 PZ |
6212 | /* |
6213 | * If we're the only runnable task on the rq and target rq also | |
6214 | * has only one task, there's absolutely no point in yielding. | |
6215 | */ | |
6216 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
6217 | yielded = -ESRCH; | |
6218 | goto out_irq; | |
6219 | } | |
6220 | ||
d95f4122 | 6221 | double_rq_lock(rq, p_rq); |
39e24d8f | 6222 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
6223 | double_rq_unlock(rq, p_rq); |
6224 | goto again; | |
6225 | } | |
6226 | ||
6227 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 6228 | goto out_unlock; |
d95f4122 MG |
6229 | |
6230 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 6231 | goto out_unlock; |
d95f4122 MG |
6232 | |
6233 | if (task_running(p_rq, p) || p->state) | |
7b270f60 | 6234 | goto out_unlock; |
d95f4122 | 6235 | |
0900acf2 | 6236 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 6237 | if (yielded) { |
ae92882e | 6238 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
6239 | /* |
6240 | * Make p's CPU reschedule; pick_next_entity takes care of | |
6241 | * fairness. | |
6242 | */ | |
6243 | if (preempt && rq != p_rq) | |
8875125e | 6244 | resched_curr(p_rq); |
6d1cafd8 | 6245 | } |
d95f4122 | 6246 | |
7b270f60 | 6247 | out_unlock: |
d95f4122 | 6248 | double_rq_unlock(rq, p_rq); |
7b270f60 | 6249 | out_irq: |
d95f4122 MG |
6250 | local_irq_restore(flags); |
6251 | ||
7b270f60 | 6252 | if (yielded > 0) |
d95f4122 MG |
6253 | schedule(); |
6254 | ||
6255 | return yielded; | |
6256 | } | |
6257 | EXPORT_SYMBOL_GPL(yield_to); | |
6258 | ||
10ab5643 TH |
6259 | int io_schedule_prepare(void) |
6260 | { | |
6261 | int old_iowait = current->in_iowait; | |
6262 | ||
6263 | current->in_iowait = 1; | |
6264 | blk_schedule_flush_plug(current); | |
6265 | ||
6266 | return old_iowait; | |
6267 | } | |
6268 | ||
6269 | void io_schedule_finish(int token) | |
6270 | { | |
6271 | current->in_iowait = token; | |
6272 | } | |
6273 | ||
1da177e4 | 6274 | /* |
41a2d6cf | 6275 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 6276 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 6277 | */ |
1da177e4 LT |
6278 | long __sched io_schedule_timeout(long timeout) |
6279 | { | |
10ab5643 | 6280 | int token; |
1da177e4 LT |
6281 | long ret; |
6282 | ||
10ab5643 | 6283 | token = io_schedule_prepare(); |
1da177e4 | 6284 | ret = schedule_timeout(timeout); |
10ab5643 | 6285 | io_schedule_finish(token); |
9cff8ade | 6286 | |
1da177e4 LT |
6287 | return ret; |
6288 | } | |
9cff8ade | 6289 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 6290 | |
e3b929b0 | 6291 | void __sched io_schedule(void) |
10ab5643 TH |
6292 | { |
6293 | int token; | |
6294 | ||
6295 | token = io_schedule_prepare(); | |
6296 | schedule(); | |
6297 | io_schedule_finish(token); | |
6298 | } | |
6299 | EXPORT_SYMBOL(io_schedule); | |
6300 | ||
1da177e4 LT |
6301 | /** |
6302 | * sys_sched_get_priority_max - return maximum RT priority. | |
6303 | * @policy: scheduling class. | |
6304 | * | |
e69f6186 YB |
6305 | * Return: On success, this syscall returns the maximum |
6306 | * rt_priority that can be used by a given scheduling class. | |
6307 | * On failure, a negative error code is returned. | |
1da177e4 | 6308 | */ |
5add95d4 | 6309 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6310 | { |
6311 | int ret = -EINVAL; | |
6312 | ||
6313 | switch (policy) { | |
6314 | case SCHED_FIFO: | |
6315 | case SCHED_RR: | |
6316 | ret = MAX_USER_RT_PRIO-1; | |
6317 | break; | |
aab03e05 | 6318 | case SCHED_DEADLINE: |
1da177e4 | 6319 | case SCHED_NORMAL: |
b0a9499c | 6320 | case SCHED_BATCH: |
dd41f596 | 6321 | case SCHED_IDLE: |
1da177e4 LT |
6322 | ret = 0; |
6323 | break; | |
6324 | } | |
6325 | return ret; | |
6326 | } | |
6327 | ||
6328 | /** | |
6329 | * sys_sched_get_priority_min - return minimum RT priority. | |
6330 | * @policy: scheduling class. | |
6331 | * | |
e69f6186 YB |
6332 | * Return: On success, this syscall returns the minimum |
6333 | * rt_priority that can be used by a given scheduling class. | |
6334 | * On failure, a negative error code is returned. | |
1da177e4 | 6335 | */ |
5add95d4 | 6336 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6337 | { |
6338 | int ret = -EINVAL; | |
6339 | ||
6340 | switch (policy) { | |
6341 | case SCHED_FIFO: | |
6342 | case SCHED_RR: | |
6343 | ret = 1; | |
6344 | break; | |
aab03e05 | 6345 | case SCHED_DEADLINE: |
1da177e4 | 6346 | case SCHED_NORMAL: |
b0a9499c | 6347 | case SCHED_BATCH: |
dd41f596 | 6348 | case SCHED_IDLE: |
1da177e4 LT |
6349 | ret = 0; |
6350 | } | |
6351 | return ret; | |
6352 | } | |
6353 | ||
abca5fc5 | 6354 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 6355 | { |
36c8b586 | 6356 | struct task_struct *p; |
a4ec24b4 | 6357 | unsigned int time_slice; |
eb580751 | 6358 | struct rq_flags rf; |
dba091b9 | 6359 | struct rq *rq; |
3a5c359a | 6360 | int retval; |
1da177e4 LT |
6361 | |
6362 | if (pid < 0) | |
3a5c359a | 6363 | return -EINVAL; |
1da177e4 LT |
6364 | |
6365 | retval = -ESRCH; | |
1a551ae7 | 6366 | rcu_read_lock(); |
1da177e4 LT |
6367 | p = find_process_by_pid(pid); |
6368 | if (!p) | |
6369 | goto out_unlock; | |
6370 | ||
6371 | retval = security_task_getscheduler(p); | |
6372 | if (retval) | |
6373 | goto out_unlock; | |
6374 | ||
eb580751 | 6375 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
6376 | time_slice = 0; |
6377 | if (p->sched_class->get_rr_interval) | |
6378 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 6379 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 6380 | |
1a551ae7 | 6381 | rcu_read_unlock(); |
abca5fc5 AV |
6382 | jiffies_to_timespec64(time_slice, t); |
6383 | return 0; | |
3a5c359a | 6384 | |
1da177e4 | 6385 | out_unlock: |
1a551ae7 | 6386 | rcu_read_unlock(); |
1da177e4 LT |
6387 | return retval; |
6388 | } | |
6389 | ||
2064a5ab RD |
6390 | /** |
6391 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6392 | * @pid: pid of the process. | |
6393 | * @interval: userspace pointer to the timeslice value. | |
6394 | * | |
6395 | * this syscall writes the default timeslice value of a given process | |
6396 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6397 | * | |
6398 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
6399 | * an error code. | |
6400 | */ | |
abca5fc5 | 6401 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 6402 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
6403 | { |
6404 | struct timespec64 t; | |
6405 | int retval = sched_rr_get_interval(pid, &t); | |
6406 | ||
6407 | if (retval == 0) | |
6408 | retval = put_timespec64(&t, interval); | |
6409 | ||
6410 | return retval; | |
6411 | } | |
6412 | ||
474b9c77 | 6413 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
6414 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
6415 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
6416 | { |
6417 | struct timespec64 t; | |
6418 | int retval = sched_rr_get_interval(pid, &t); | |
6419 | ||
6420 | if (retval == 0) | |
9afc5eee | 6421 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
6422 | return retval; |
6423 | } | |
6424 | #endif | |
6425 | ||
82a1fcb9 | 6426 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6427 | { |
1da177e4 | 6428 | unsigned long free = 0; |
4e79752c | 6429 | int ppid; |
c930b2c0 | 6430 | |
38200502 TH |
6431 | if (!try_get_task_stack(p)) |
6432 | return; | |
20435d84 | 6433 | |
cc172ff3 | 6434 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 XX |
6435 | |
6436 | if (p->state == TASK_RUNNING) | |
cc172ff3 | 6437 | pr_cont(" running task "); |
1da177e4 | 6438 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 6439 | free = stack_not_used(p); |
1da177e4 | 6440 | #endif |
a90e984c | 6441 | ppid = 0; |
4e79752c | 6442 | rcu_read_lock(); |
a90e984c ON |
6443 | if (pid_alive(p)) |
6444 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 6445 | rcu_read_unlock(); |
cc172ff3 LZ |
6446 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
6447 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 6448 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 6449 | |
3d1cb205 | 6450 | print_worker_info(KERN_INFO, p); |
9cb8f069 | 6451 | show_stack(p, NULL, KERN_INFO); |
38200502 | 6452 | put_task_stack(p); |
1da177e4 | 6453 | } |
0032f4e8 | 6454 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 6455 | |
5d68cc95 PZ |
6456 | static inline bool |
6457 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
6458 | { | |
6459 | /* no filter, everything matches */ | |
6460 | if (!state_filter) | |
6461 | return true; | |
6462 | ||
6463 | /* filter, but doesn't match */ | |
6464 | if (!(p->state & state_filter)) | |
6465 | return false; | |
6466 | ||
6467 | /* | |
6468 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
6469 | * TASK_KILLABLE). | |
6470 | */ | |
6471 | if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE) | |
6472 | return false; | |
6473 | ||
6474 | return true; | |
6475 | } | |
6476 | ||
6477 | ||
e59e2ae2 | 6478 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6479 | { |
36c8b586 | 6480 | struct task_struct *g, *p; |
1da177e4 | 6481 | |
510f5acc | 6482 | rcu_read_lock(); |
5d07f420 | 6483 | for_each_process_thread(g, p) { |
1da177e4 LT |
6484 | /* |
6485 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 6486 | * console might take a lot of time: |
57675cb9 AR |
6487 | * Also, reset softlockup watchdogs on all CPUs, because |
6488 | * another CPU might be blocked waiting for us to process | |
6489 | * an IPI. | |
1da177e4 LT |
6490 | */ |
6491 | touch_nmi_watchdog(); | |
57675cb9 | 6492 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 6493 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 6494 | sched_show_task(p); |
5d07f420 | 6495 | } |
1da177e4 | 6496 | |
dd41f596 | 6497 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
6498 | if (!state_filter) |
6499 | sysrq_sched_debug_show(); | |
dd41f596 | 6500 | #endif |
510f5acc | 6501 | rcu_read_unlock(); |
e59e2ae2 IM |
6502 | /* |
6503 | * Only show locks if all tasks are dumped: | |
6504 | */ | |
93335a21 | 6505 | if (!state_filter) |
e59e2ae2 | 6506 | debug_show_all_locks(); |
1da177e4 LT |
6507 | } |
6508 | ||
f340c0d1 IM |
6509 | /** |
6510 | * init_idle - set up an idle thread for a given CPU | |
6511 | * @idle: task in question | |
d1ccc66d | 6512 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
6513 | * |
6514 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6515 | * flag, to make booting more robust. | |
6516 | */ | |
0db0628d | 6517 | void init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6518 | { |
70b97a7f | 6519 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6520 | unsigned long flags; |
6521 | ||
ff51ff84 PZ |
6522 | __sched_fork(0, idle); |
6523 | ||
25834c73 PZ |
6524 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
6525 | raw_spin_lock(&rq->lock); | |
5cbd54ef | 6526 | |
06b83b5f | 6527 | idle->state = TASK_RUNNING; |
dd41f596 | 6528 | idle->se.exec_start = sched_clock(); |
c1de45ca | 6529 | idle->flags |= PF_IDLE; |
dd41f596 | 6530 | |
d08b9f0c | 6531 | scs_task_reset(idle); |
e1b77c92 MR |
6532 | kasan_unpoison_task_stack(idle); |
6533 | ||
de9b8f5d PZ |
6534 | #ifdef CONFIG_SMP |
6535 | /* | |
6536 | * Its possible that init_idle() gets called multiple times on a task, | |
6537 | * in that case do_set_cpus_allowed() will not do the right thing. | |
6538 | * | |
6539 | * And since this is boot we can forgo the serialization. | |
6540 | */ | |
6541 | set_cpus_allowed_common(idle, cpumask_of(cpu)); | |
6542 | #endif | |
6506cf6c PZ |
6543 | /* |
6544 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 6545 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
6546 | * lockdep check in task_group() will fail. |
6547 | * | |
6548 | * Similar case to sched_fork(). / Alternatively we could | |
6549 | * use task_rq_lock() here and obtain the other rq->lock. | |
6550 | * | |
6551 | * Silence PROVE_RCU | |
6552 | */ | |
6553 | rcu_read_lock(); | |
dd41f596 | 6554 | __set_task_cpu(idle, cpu); |
6506cf6c | 6555 | rcu_read_unlock(); |
1da177e4 | 6556 | |
5311a98f EB |
6557 | rq->idle = idle; |
6558 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 6559 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 6560 | #ifdef CONFIG_SMP |
3ca7a440 | 6561 | idle->on_cpu = 1; |
4866cde0 | 6562 | #endif |
25834c73 PZ |
6563 | raw_spin_unlock(&rq->lock); |
6564 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); | |
1da177e4 LT |
6565 | |
6566 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 6567 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 6568 | |
dd41f596 IM |
6569 | /* |
6570 | * The idle tasks have their own, simple scheduling class: | |
6571 | */ | |
6572 | idle->sched_class = &idle_sched_class; | |
868baf07 | 6573 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 6574 | vtime_init_idle(idle, cpu); |
de9b8f5d | 6575 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
6576 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
6577 | #endif | |
19978ca6 IM |
6578 | } |
6579 | ||
e1d4eeec NP |
6580 | #ifdef CONFIG_SMP |
6581 | ||
f82f8042 JL |
6582 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
6583 | const struct cpumask *trial) | |
6584 | { | |
06a76fe0 | 6585 | int ret = 1; |
f82f8042 | 6586 | |
bb2bc55a MG |
6587 | if (!cpumask_weight(cur)) |
6588 | return ret; | |
6589 | ||
06a76fe0 | 6590 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
6591 | |
6592 | return ret; | |
6593 | } | |
6594 | ||
7f51412a JL |
6595 | int task_can_attach(struct task_struct *p, |
6596 | const struct cpumask *cs_cpus_allowed) | |
6597 | { | |
6598 | int ret = 0; | |
6599 | ||
6600 | /* | |
6601 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 6602 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
6603 | * affinity and isolating such threads by their set of |
6604 | * allowed nodes is unnecessary. Thus, cpusets are not | |
6605 | * applicable for such threads. This prevents checking for | |
6606 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 6607 | * before cpus_mask may be changed. |
7f51412a JL |
6608 | */ |
6609 | if (p->flags & PF_NO_SETAFFINITY) { | |
6610 | ret = -EINVAL; | |
6611 | goto out; | |
6612 | } | |
6613 | ||
7f51412a | 6614 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
6615 | cs_cpus_allowed)) |
6616 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 6617 | |
7f51412a JL |
6618 | out: |
6619 | return ret; | |
6620 | } | |
6621 | ||
f2cb1360 | 6622 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 6623 | |
e6628d5b MG |
6624 | #ifdef CONFIG_NUMA_BALANCING |
6625 | /* Migrate current task p to target_cpu */ | |
6626 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
6627 | { | |
6628 | struct migration_arg arg = { p, target_cpu }; | |
6629 | int curr_cpu = task_cpu(p); | |
6630 | ||
6631 | if (curr_cpu == target_cpu) | |
6632 | return 0; | |
6633 | ||
3bd37062 | 6634 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
6635 | return -EINVAL; |
6636 | ||
6637 | /* TODO: This is not properly updating schedstats */ | |
6638 | ||
286549dc | 6639 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
6640 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
6641 | } | |
0ec8aa00 PZ |
6642 | |
6643 | /* | |
6644 | * Requeue a task on a given node and accurately track the number of NUMA | |
6645 | * tasks on the runqueues | |
6646 | */ | |
6647 | void sched_setnuma(struct task_struct *p, int nid) | |
6648 | { | |
da0c1e65 | 6649 | bool queued, running; |
eb580751 PZ |
6650 | struct rq_flags rf; |
6651 | struct rq *rq; | |
0ec8aa00 | 6652 | |
eb580751 | 6653 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 6654 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
6655 | running = task_current(rq, p); |
6656 | ||
da0c1e65 | 6657 | if (queued) |
1de64443 | 6658 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 6659 | if (running) |
f3cd1c4e | 6660 | put_prev_task(rq, p); |
0ec8aa00 PZ |
6661 | |
6662 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 6663 | |
da0c1e65 | 6664 | if (queued) |
7134b3e9 | 6665 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 6666 | if (running) |
03b7fad1 | 6667 | set_next_task(rq, p); |
eb580751 | 6668 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 6669 | } |
5cc389bc | 6670 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 6671 | |
1da177e4 | 6672 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 6673 | /* |
d1ccc66d | 6674 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 6675 | * offline. |
054b9108 | 6676 | */ |
48c5ccae | 6677 | void idle_task_exit(void) |
1da177e4 | 6678 | { |
48c5ccae | 6679 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6680 | |
48c5ccae | 6681 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 6682 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 6683 | |
a53efe5f | 6684 | if (mm != &init_mm) { |
252d2a41 | 6685 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
6686 | finish_arch_post_lock_switch(); |
6687 | } | |
bf2c59fc PZ |
6688 | |
6689 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
6690 | } |
6691 | ||
6692 | /* | |
5d180232 PZ |
6693 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
6694 | * we might have. Assumes we're called after migrate_tasks() so that the | |
d60585c5 TG |
6695 | * nr_active count is stable. We need to take the teardown thread which |
6696 | * is calling this into account, so we hand in adjust = 1 to the load | |
6697 | * calculation. | |
5d180232 PZ |
6698 | * |
6699 | * Also see the comment "Global load-average calculations". | |
1da177e4 | 6700 | */ |
5d180232 | 6701 | static void calc_load_migrate(struct rq *rq) |
1da177e4 | 6702 | { |
d60585c5 | 6703 | long delta = calc_load_fold_active(rq, 1); |
5d180232 PZ |
6704 | if (delta) |
6705 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 LT |
6706 | } |
6707 | ||
10e7071b | 6708 | static struct task_struct *__pick_migrate_task(struct rq *rq) |
3f1d2a31 | 6709 | { |
10e7071b PZ |
6710 | const struct sched_class *class; |
6711 | struct task_struct *next; | |
3f1d2a31 | 6712 | |
10e7071b | 6713 | for_each_class(class) { |
98c2f700 | 6714 | next = class->pick_next_task(rq); |
10e7071b | 6715 | if (next) { |
6e2df058 | 6716 | next->sched_class->put_prev_task(rq, next); |
10e7071b PZ |
6717 | return next; |
6718 | } | |
6719 | } | |
3f1d2a31 | 6720 | |
10e7071b PZ |
6721 | /* The idle class should always have a runnable task */ |
6722 | BUG(); | |
6723 | } | |
3f1d2a31 | 6724 | |
48f24c4d | 6725 | /* |
48c5ccae PZ |
6726 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6727 | * try_to_wake_up()->select_task_rq(). | |
6728 | * | |
6729 | * Called with rq->lock held even though we'er in stop_machine() and | |
6730 | * there's no concurrency possible, we hold the required locks anyway | |
6731 | * because of lock validation efforts. | |
1da177e4 | 6732 | */ |
8a8c69c3 | 6733 | static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf) |
1da177e4 | 6734 | { |
5e16bbc2 | 6735 | struct rq *rq = dead_rq; |
48c5ccae | 6736 | struct task_struct *next, *stop = rq->stop; |
8a8c69c3 | 6737 | struct rq_flags orf = *rf; |
48c5ccae | 6738 | int dest_cpu; |
1da177e4 LT |
6739 | |
6740 | /* | |
48c5ccae PZ |
6741 | * Fudge the rq selection such that the below task selection loop |
6742 | * doesn't get stuck on the currently eligible stop task. | |
6743 | * | |
6744 | * We're currently inside stop_machine() and the rq is either stuck | |
6745 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6746 | * either way we should never end up calling schedule() until we're | |
6747 | * done here. | |
1da177e4 | 6748 | */ |
48c5ccae | 6749 | rq->stop = NULL; |
48f24c4d | 6750 | |
77bd3970 FW |
6751 | /* |
6752 | * put_prev_task() and pick_next_task() sched | |
6753 | * class method both need to have an up-to-date | |
6754 | * value of rq->clock[_task] | |
6755 | */ | |
6756 | update_rq_clock(rq); | |
6757 | ||
5e16bbc2 | 6758 | for (;;) { |
48c5ccae PZ |
6759 | /* |
6760 | * There's this thread running, bail when that's the only | |
d1ccc66d | 6761 | * remaining thread: |
48c5ccae PZ |
6762 | */ |
6763 | if (rq->nr_running == 1) | |
dd41f596 | 6764 | break; |
48c5ccae | 6765 | |
10e7071b | 6766 | next = __pick_migrate_task(rq); |
e692ab53 | 6767 | |
5473e0cc | 6768 | /* |
3bd37062 | 6769 | * Rules for changing task_struct::cpus_mask are holding |
5473e0cc WL |
6770 | * both pi_lock and rq->lock, such that holding either |
6771 | * stabilizes the mask. | |
6772 | * | |
6773 | * Drop rq->lock is not quite as disastrous as it usually is | |
6774 | * because !cpu_active at this point, which means load-balance | |
6775 | * will not interfere. Also, stop-machine. | |
6776 | */ | |
8a8c69c3 | 6777 | rq_unlock(rq, rf); |
5473e0cc | 6778 | raw_spin_lock(&next->pi_lock); |
8a8c69c3 | 6779 | rq_relock(rq, rf); |
5473e0cc WL |
6780 | |
6781 | /* | |
6782 | * Since we're inside stop-machine, _nothing_ should have | |
6783 | * changed the task, WARN if weird stuff happened, because in | |
6784 | * that case the above rq->lock drop is a fail too. | |
6785 | */ | |
6786 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { | |
6787 | raw_spin_unlock(&next->pi_lock); | |
6788 | continue; | |
6789 | } | |
6790 | ||
48c5ccae | 6791 | /* Find suitable destination for @next, with force if needed. */ |
5e16bbc2 | 6792 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
8a8c69c3 | 6793 | rq = __migrate_task(rq, rf, next, dest_cpu); |
5e16bbc2 | 6794 | if (rq != dead_rq) { |
8a8c69c3 | 6795 | rq_unlock(rq, rf); |
5e16bbc2 | 6796 | rq = dead_rq; |
8a8c69c3 PZ |
6797 | *rf = orf; |
6798 | rq_relock(rq, rf); | |
5e16bbc2 | 6799 | } |
5473e0cc | 6800 | raw_spin_unlock(&next->pi_lock); |
1da177e4 | 6801 | } |
dce48a84 | 6802 | |
48c5ccae | 6803 | rq->stop = stop; |
dce48a84 | 6804 | } |
1da177e4 LT |
6805 | #endif /* CONFIG_HOTPLUG_CPU */ |
6806 | ||
f2cb1360 | 6807 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
6808 | { |
6809 | if (!rq->online) { | |
6810 | const struct sched_class *class; | |
6811 | ||
c6c4927b | 6812 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6813 | rq->online = 1; |
6814 | ||
6815 | for_each_class(class) { | |
6816 | if (class->rq_online) | |
6817 | class->rq_online(rq); | |
6818 | } | |
6819 | } | |
6820 | } | |
6821 | ||
f2cb1360 | 6822 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
6823 | { |
6824 | if (rq->online) { | |
6825 | const struct sched_class *class; | |
6826 | ||
6827 | for_each_class(class) { | |
6828 | if (class->rq_offline) | |
6829 | class->rq_offline(rq); | |
6830 | } | |
6831 | ||
c6c4927b | 6832 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6833 | rq->online = 0; |
6834 | } | |
6835 | } | |
6836 | ||
d1ccc66d IM |
6837 | /* |
6838 | * used to mark begin/end of suspend/resume: | |
6839 | */ | |
6840 | static int num_cpus_frozen; | |
d35be8ba | 6841 | |
1da177e4 | 6842 | /* |
3a101d05 TH |
6843 | * Update cpusets according to cpu_active mask. If cpusets are |
6844 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
6845 | * around partition_sched_domains(). | |
d35be8ba SB |
6846 | * |
6847 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
6848 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 6849 | */ |
40190a78 | 6850 | static void cpuset_cpu_active(void) |
e761b772 | 6851 | { |
40190a78 | 6852 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
6853 | /* |
6854 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
6855 | * resume sequence. As long as this is not the last online | |
6856 | * operation in the resume sequence, just build a single sched | |
6857 | * domain, ignoring cpusets. | |
6858 | */ | |
50e76632 PZ |
6859 | partition_sched_domains(1, NULL, NULL); |
6860 | if (--num_cpus_frozen) | |
135fb3e1 | 6861 | return; |
d35be8ba SB |
6862 | /* |
6863 | * This is the last CPU online operation. So fall through and | |
6864 | * restore the original sched domains by considering the | |
6865 | * cpuset configurations. | |
6866 | */ | |
50e76632 | 6867 | cpuset_force_rebuild(); |
3a101d05 | 6868 | } |
30e03acd | 6869 | cpuset_update_active_cpus(); |
3a101d05 | 6870 | } |
e761b772 | 6871 | |
40190a78 | 6872 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 6873 | { |
40190a78 | 6874 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 6875 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 6876 | return -EBUSY; |
30e03acd | 6877 | cpuset_update_active_cpus(); |
135fb3e1 | 6878 | } else { |
d35be8ba SB |
6879 | num_cpus_frozen++; |
6880 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 6881 | } |
135fb3e1 | 6882 | return 0; |
e761b772 | 6883 | } |
e761b772 | 6884 | |
40190a78 | 6885 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 6886 | { |
7d976699 | 6887 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 6888 | struct rq_flags rf; |
7d976699 | 6889 | |
ba2591a5 PZ |
6890 | #ifdef CONFIG_SCHED_SMT |
6891 | /* | |
c5511d03 | 6892 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 6893 | */ |
c5511d03 PZI |
6894 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
6895 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 6896 | #endif |
40190a78 | 6897 | set_cpu_active(cpu, true); |
135fb3e1 | 6898 | |
40190a78 | 6899 | if (sched_smp_initialized) { |
135fb3e1 | 6900 | sched_domains_numa_masks_set(cpu); |
40190a78 | 6901 | cpuset_cpu_active(); |
e761b772 | 6902 | } |
7d976699 TG |
6903 | |
6904 | /* | |
6905 | * Put the rq online, if not already. This happens: | |
6906 | * | |
6907 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 6908 | * after all CPUs have been brought up. |
7d976699 TG |
6909 | * |
6910 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
6911 | * domains. | |
6912 | */ | |
8a8c69c3 | 6913 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
6914 | if (rq->rd) { |
6915 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
6916 | set_rq_online(rq); | |
6917 | } | |
8a8c69c3 | 6918 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 6919 | |
40190a78 | 6920 | return 0; |
135fb3e1 TG |
6921 | } |
6922 | ||
40190a78 | 6923 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 6924 | { |
135fb3e1 TG |
6925 | int ret; |
6926 | ||
40190a78 | 6927 | set_cpu_active(cpu, false); |
b2454caa PZ |
6928 | /* |
6929 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU | |
6930 | * users of this state to go away such that all new such users will | |
6931 | * observe it. | |
6932 | * | |
b2454caa PZ |
6933 | * Do sync before park smpboot threads to take care the rcu boost case. |
6934 | */ | |
309ba859 | 6935 | synchronize_rcu(); |
40190a78 | 6936 | |
c5511d03 PZI |
6937 | #ifdef CONFIG_SCHED_SMT |
6938 | /* | |
6939 | * When going down, decrement the number of cores with SMT present. | |
6940 | */ | |
6941 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
6942 | static_branch_dec_cpuslocked(&sched_smt_present); | |
6943 | #endif | |
6944 | ||
40190a78 TG |
6945 | if (!sched_smp_initialized) |
6946 | return 0; | |
6947 | ||
6948 | ret = cpuset_cpu_inactive(cpu); | |
6949 | if (ret) { | |
6950 | set_cpu_active(cpu, true); | |
6951 | return ret; | |
135fb3e1 | 6952 | } |
40190a78 TG |
6953 | sched_domains_numa_masks_clear(cpu); |
6954 | return 0; | |
135fb3e1 TG |
6955 | } |
6956 | ||
94baf7a5 TG |
6957 | static void sched_rq_cpu_starting(unsigned int cpu) |
6958 | { | |
6959 | struct rq *rq = cpu_rq(cpu); | |
6960 | ||
6961 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
6962 | update_max_interval(); |
6963 | } | |
6964 | ||
135fb3e1 TG |
6965 | int sched_cpu_starting(unsigned int cpu) |
6966 | { | |
94baf7a5 | 6967 | sched_rq_cpu_starting(cpu); |
d84b3131 | 6968 | sched_tick_start(cpu); |
135fb3e1 | 6969 | return 0; |
e761b772 | 6970 | } |
e761b772 | 6971 | |
f2785ddb TG |
6972 | #ifdef CONFIG_HOTPLUG_CPU |
6973 | int sched_cpu_dying(unsigned int cpu) | |
6974 | { | |
6975 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 6976 | struct rq_flags rf; |
f2785ddb TG |
6977 | |
6978 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 6979 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
6980 | |
6981 | rq_lock_irqsave(rq, &rf); | |
f2785ddb TG |
6982 | if (rq->rd) { |
6983 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
6984 | set_rq_offline(rq); | |
6985 | } | |
8a8c69c3 | 6986 | migrate_tasks(rq, &rf); |
f2785ddb | 6987 | BUG_ON(rq->nr_running != 1); |
8a8c69c3 PZ |
6988 | rq_unlock_irqrestore(rq, &rf); |
6989 | ||
f2785ddb TG |
6990 | calc_load_migrate(rq); |
6991 | update_max_interval(); | |
00357f5e | 6992 | nohz_balance_exit_idle(rq); |
e5ef27d0 | 6993 | hrtick_clear(rq); |
f2785ddb TG |
6994 | return 0; |
6995 | } | |
6996 | #endif | |
6997 | ||
1da177e4 LT |
6998 | void __init sched_init_smp(void) |
6999 | { | |
cb83b629 PZ |
7000 | sched_init_numa(); |
7001 | ||
6acce3ef PZ |
7002 | /* |
7003 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 7004 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 7005 | * happen. |
6acce3ef | 7006 | */ |
712555ee | 7007 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 7008 | sched_init_domains(cpu_active_mask); |
712555ee | 7009 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 7010 | |
5c1e1767 | 7011 | /* Move init over to a non-isolated CPU */ |
edb93821 | 7012 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 7013 | BUG(); |
19978ca6 | 7014 | sched_init_granularity(); |
4212823f | 7015 | |
0e3900e6 | 7016 | init_sched_rt_class(); |
1baca4ce | 7017 | init_sched_dl_class(); |
1b568f0a | 7018 | |
e26fbffd | 7019 | sched_smp_initialized = true; |
1da177e4 | 7020 | } |
e26fbffd TG |
7021 | |
7022 | static int __init migration_init(void) | |
7023 | { | |
77a5352b | 7024 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 7025 | return 0; |
1da177e4 | 7026 | } |
e26fbffd TG |
7027 | early_initcall(migration_init); |
7028 | ||
1da177e4 LT |
7029 | #else |
7030 | void __init sched_init_smp(void) | |
7031 | { | |
19978ca6 | 7032 | sched_init_granularity(); |
1da177e4 LT |
7033 | } |
7034 | #endif /* CONFIG_SMP */ | |
7035 | ||
7036 | int in_sched_functions(unsigned long addr) | |
7037 | { | |
1da177e4 LT |
7038 | return in_lock_functions(addr) || |
7039 | (addr >= (unsigned long)__sched_text_start | |
7040 | && addr < (unsigned long)__sched_text_end); | |
7041 | } | |
7042 | ||
029632fb | 7043 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
7044 | /* |
7045 | * Default task group. | |
7046 | * Every task in system belongs to this group at bootup. | |
7047 | */ | |
029632fb | 7048 | struct task_group root_task_group; |
35cf4e50 | 7049 | LIST_HEAD(task_groups); |
b0367629 WL |
7050 | |
7051 | /* Cacheline aligned slab cache for task_group */ | |
7052 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 7053 | #endif |
6f505b16 | 7054 | |
e6252c3e | 7055 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 7056 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 7057 | |
1da177e4 LT |
7058 | void __init sched_init(void) |
7059 | { | |
a1dc0446 | 7060 | unsigned long ptr = 0; |
55627e3c | 7061 | int i; |
434d53b0 | 7062 | |
c3a340f7 SRV |
7063 | /* Make sure the linker didn't screw up */ |
7064 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
7065 | &fair_sched_class + 1 != &rt_sched_class || | |
7066 | &rt_sched_class + 1 != &dl_sched_class); | |
7067 | #ifdef CONFIG_SMP | |
7068 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
7069 | #endif | |
7070 | ||
5822a454 | 7071 | wait_bit_init(); |
9dcb8b68 | 7072 | |
434d53b0 | 7073 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 7074 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
7075 | #endif |
7076 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 7077 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 7078 | #endif |
a1dc0446 QC |
7079 | if (ptr) { |
7080 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
7081 | |
7082 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7083 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7084 | ptr += nr_cpu_ids * sizeof(void **); |
7085 | ||
07e06b01 | 7086 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7087 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7088 | |
b1d1779e WY |
7089 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
7090 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 7091 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7092 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7093 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7094 | ptr += nr_cpu_ids * sizeof(void **); |
7095 | ||
07e06b01 | 7096 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7097 | ptr += nr_cpu_ids * sizeof(void **); |
7098 | ||
6d6bc0ad | 7099 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 7100 | } |
df7c8e84 | 7101 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
7102 | for_each_possible_cpu(i) { |
7103 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
7104 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
7105 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
7106 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 7107 | } |
b74e6278 | 7108 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 7109 | |
d1ccc66d IM |
7110 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
7111 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 7112 | |
57d885fe GH |
7113 | #ifdef CONFIG_SMP |
7114 | init_defrootdomain(); | |
7115 | #endif | |
7116 | ||
d0b27fa7 | 7117 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7118 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7119 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7120 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7121 | |
7c941438 | 7122 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
7123 | task_group_cache = KMEM_CACHE(task_group, 0); |
7124 | ||
07e06b01 YZ |
7125 | list_add(&root_task_group.list, &task_groups); |
7126 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 7127 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 7128 | autogroup_init(&init_task); |
7c941438 | 7129 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7130 | |
0a945022 | 7131 | for_each_possible_cpu(i) { |
70b97a7f | 7132 | struct rq *rq; |
1da177e4 LT |
7133 | |
7134 | rq = cpu_rq(i); | |
05fa785c | 7135 | raw_spin_lock_init(&rq->lock); |
7897986b | 7136 | rq->nr_running = 0; |
dce48a84 TG |
7137 | rq->calc_load_active = 0; |
7138 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 7139 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
7140 | init_rt_rq(&rq->rt); |
7141 | init_dl_rq(&rq->dl); | |
dd41f596 | 7142 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 7143 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 7144 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 7145 | /* |
d1ccc66d | 7146 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
7147 | * |
7148 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
7149 | * gets 100% of the CPU resources in the system. This overall |
7150 | * system CPU resource is divided among the tasks of | |
07e06b01 | 7151 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7152 | * based on each entity's (task or task-group's) weight |
7153 | * (se->load.weight). | |
7154 | * | |
07e06b01 | 7155 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 7156 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 7157 | * then A0's share of the CPU resource is: |
354d60c2 | 7158 | * |
0d905bca | 7159 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7160 | * |
07e06b01 YZ |
7161 | * We achieve this by letting root_task_group's tasks sit |
7162 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7163 | */ |
07e06b01 | 7164 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7165 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7166 | ||
7167 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7168 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7169 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7170 | #endif |
1da177e4 | 7171 | #ifdef CONFIG_SMP |
41c7ce9a | 7172 | rq->sd = NULL; |
57d885fe | 7173 | rq->rd = NULL; |
ca6d75e6 | 7174 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
e3fca9e7 | 7175 | rq->balance_callback = NULL; |
1da177e4 | 7176 | rq->active_balance = 0; |
dd41f596 | 7177 | rq->next_balance = jiffies; |
1da177e4 | 7178 | rq->push_cpu = 0; |
0a2966b4 | 7179 | rq->cpu = i; |
1f11eb6a | 7180 | rq->online = 0; |
eae0c9df MG |
7181 | rq->idle_stamp = 0; |
7182 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
9bd721c5 | 7183 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
7184 | |
7185 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
7186 | ||
dc938520 | 7187 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 7188 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 7189 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 7190 | atomic_set(&rq->nohz_flags, 0); |
90b5363a PZI |
7191 | |
7192 | rq_csd_init(rq, &rq->nohz_csd, nohz_csd_func); | |
83cd4fe2 | 7193 | #endif |
9fd81dd5 | 7194 | #endif /* CONFIG_SMP */ |
77a021be | 7195 | hrtick_rq_init(rq); |
1da177e4 | 7196 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7197 | } |
7198 | ||
9059393e | 7199 | set_load_weight(&init_task, false); |
b50f60ce | 7200 | |
1da177e4 LT |
7201 | /* |
7202 | * The boot idle thread does lazy MMU switching as well: | |
7203 | */ | |
f1f10076 | 7204 | mmgrab(&init_mm); |
1da177e4 LT |
7205 | enter_lazy_tlb(&init_mm, current); |
7206 | ||
7207 | /* | |
7208 | * Make us the idle thread. Technically, schedule() should not be | |
7209 | * called from this thread, however somewhere below it might be, | |
7210 | * but because we are the idle thread, we just pick up running again | |
7211 | * when this runqueue becomes "idle". | |
7212 | */ | |
7213 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7214 | |
7215 | calc_load_update = jiffies + LOAD_FREQ; | |
7216 | ||
bf4d83f6 | 7217 | #ifdef CONFIG_SMP |
29d5e047 | 7218 | idle_thread_set_boot_cpu(); |
029632fb PZ |
7219 | #endif |
7220 | init_sched_fair_class(); | |
6a7b3dc3 | 7221 | |
4698f88c JP |
7222 | init_schedstats(); |
7223 | ||
eb414681 JW |
7224 | psi_init(); |
7225 | ||
69842cba PB |
7226 | init_uclamp(); |
7227 | ||
6892b75e | 7228 | scheduler_running = 1; |
1da177e4 LT |
7229 | } |
7230 | ||
d902db1e | 7231 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
7232 | static inline int preempt_count_equals(int preempt_offset) |
7233 | { | |
da7142e2 | 7234 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 7235 | |
4ba8216c | 7236 | return (nested == preempt_offset); |
e4aafea2 FW |
7237 | } |
7238 | ||
d894837f | 7239 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7240 | { |
8eb23b9f PZ |
7241 | /* |
7242 | * Blocking primitives will set (and therefore destroy) current->state, | |
7243 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
7244 | * otherwise we will destroy state. | |
7245 | */ | |
00845eb9 | 7246 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
8eb23b9f PZ |
7247 | "do not call blocking ops when !TASK_RUNNING; " |
7248 | "state=%lx set at [<%p>] %pS\n", | |
7249 | current->state, | |
7250 | (void *)current->task_state_change, | |
00845eb9 | 7251 | (void *)current->task_state_change); |
8eb23b9f | 7252 | |
3427445a PZ |
7253 | ___might_sleep(file, line, preempt_offset); |
7254 | } | |
7255 | EXPORT_SYMBOL(__might_sleep); | |
7256 | ||
7257 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 7258 | { |
d1ccc66d IM |
7259 | /* Ratelimiting timestamp: */ |
7260 | static unsigned long prev_jiffy; | |
7261 | ||
d1c6d149 | 7262 | unsigned long preempt_disable_ip; |
1da177e4 | 7263 | |
d1ccc66d IM |
7264 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
7265 | rcu_sleep_check(); | |
7266 | ||
db273be2 | 7267 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 7268 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
7269 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
7270 | oops_in_progress) | |
aef745fc | 7271 | return; |
1c3c5eab | 7272 | |
aef745fc IM |
7273 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
7274 | return; | |
7275 | prev_jiffy = jiffies; | |
7276 | ||
d1ccc66d | 7277 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
7278 | preempt_disable_ip = get_preempt_disable_ip(current); |
7279 | ||
3df0fc5b PZ |
7280 | printk(KERN_ERR |
7281 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7282 | file, line); | |
7283 | printk(KERN_ERR | |
312364f3 DV |
7284 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
7285 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 7286 | current->pid, current->comm); |
aef745fc | 7287 | |
a8b686b3 ES |
7288 | if (task_stack_end_corrupted(current)) |
7289 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
7290 | ||
aef745fc IM |
7291 | debug_show_held_locks(current); |
7292 | if (irqs_disabled()) | |
7293 | print_irqtrace_events(current); | |
d1c6d149 VN |
7294 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
7295 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 7296 | pr_err("Preemption disabled at:"); |
2062a4e8 | 7297 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 7298 | } |
aef745fc | 7299 | dump_stack(); |
f0b22e39 | 7300 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 7301 | } |
3427445a | 7302 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
7303 | |
7304 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
7305 | { | |
7306 | static unsigned long prev_jiffy; | |
7307 | ||
7308 | if (irqs_disabled()) | |
7309 | return; | |
7310 | ||
7311 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
7312 | return; | |
7313 | ||
7314 | if (preempt_count() > preempt_offset) | |
7315 | return; | |
7316 | ||
7317 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7318 | return; | |
7319 | prev_jiffy = jiffies; | |
7320 | ||
7321 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
7322 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7323 | in_atomic(), irqs_disabled(), | |
7324 | current->pid, current->comm); | |
7325 | ||
7326 | debug_show_held_locks(current); | |
7327 | dump_stack(); | |
7328 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
7329 | } | |
7330 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
1da177e4 LT |
7331 | #endif |
7332 | ||
7333 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 7334 | void normalize_rt_tasks(void) |
3a5e4dc1 | 7335 | { |
dbc7f069 | 7336 | struct task_struct *g, *p; |
d50dde5a DF |
7337 | struct sched_attr attr = { |
7338 | .sched_policy = SCHED_NORMAL, | |
7339 | }; | |
1da177e4 | 7340 | |
3472eaa1 | 7341 | read_lock(&tasklist_lock); |
5d07f420 | 7342 | for_each_process_thread(g, p) { |
178be793 IM |
7343 | /* |
7344 | * Only normalize user tasks: | |
7345 | */ | |
3472eaa1 | 7346 | if (p->flags & PF_KTHREAD) |
178be793 IM |
7347 | continue; |
7348 | ||
4fa8d299 JP |
7349 | p->se.exec_start = 0; |
7350 | schedstat_set(p->se.statistics.wait_start, 0); | |
7351 | schedstat_set(p->se.statistics.sleep_start, 0); | |
7352 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 7353 | |
aab03e05 | 7354 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
7355 | /* |
7356 | * Renice negative nice level userspace | |
7357 | * tasks back to 0: | |
7358 | */ | |
3472eaa1 | 7359 | if (task_nice(p) < 0) |
dd41f596 | 7360 | set_user_nice(p, 0); |
1da177e4 | 7361 | continue; |
dd41f596 | 7362 | } |
1da177e4 | 7363 | |
dbc7f069 | 7364 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 7365 | } |
3472eaa1 | 7366 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7367 | } |
7368 | ||
7369 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7370 | |
67fc4e0c | 7371 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7372 | /* |
67fc4e0c | 7373 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7374 | * |
7375 | * They can only be called when the whole system has been | |
7376 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7377 | * activity can take place. Using them for anything else would | |
7378 | * be a serious bug, and as a result, they aren't even visible | |
7379 | * under any other configuration. | |
7380 | */ | |
7381 | ||
7382 | /** | |
d1ccc66d | 7383 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
7384 | * @cpu: the processor in question. |
7385 | * | |
7386 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
7387 | * |
7388 | * Return: The current task for @cpu. | |
1df5c10a | 7389 | */ |
36c8b586 | 7390 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7391 | { |
7392 | return cpu_curr(cpu); | |
7393 | } | |
7394 | ||
67fc4e0c JW |
7395 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7396 | ||
7397 | #ifdef CONFIG_IA64 | |
1df5c10a | 7398 | /** |
5feeb783 | 7399 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
7400 | * @cpu: the processor in question. |
7401 | * @p: the task pointer to set. | |
7402 | * | |
7403 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 7404 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 7405 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
7406 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7407 | * and caller must save the original value of the current task (see | |
7408 | * curr_task() above) and restore that value before reenabling interrupts and | |
7409 | * re-starting the system. | |
7410 | * | |
7411 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7412 | */ | |
a458ae2e | 7413 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7414 | { |
7415 | cpu_curr(cpu) = p; | |
7416 | } | |
7417 | ||
7418 | #endif | |
29f59db3 | 7419 | |
7c941438 | 7420 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
7421 | /* task_group_lock serializes the addition/removal of task groups */ |
7422 | static DEFINE_SPINLOCK(task_group_lock); | |
7423 | ||
2480c093 PB |
7424 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
7425 | struct task_group *parent) | |
7426 | { | |
7427 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 7428 | enum uclamp_id clamp_id; |
2480c093 PB |
7429 | |
7430 | for_each_clamp_id(clamp_id) { | |
7431 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
7432 | uclamp_none(clamp_id), false); | |
0b60ba2d | 7433 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
7434 | } |
7435 | #endif | |
7436 | } | |
7437 | ||
2f5177f0 | 7438 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
7439 | { |
7440 | free_fair_sched_group(tg); | |
7441 | free_rt_sched_group(tg); | |
e9aa1dd1 | 7442 | autogroup_free(tg); |
b0367629 | 7443 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
7444 | } |
7445 | ||
7446 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7447 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7448 | { |
7449 | struct task_group *tg; | |
bccbe08a | 7450 | |
b0367629 | 7451 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
7452 | if (!tg) |
7453 | return ERR_PTR(-ENOMEM); | |
7454 | ||
ec7dc8ac | 7455 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
7456 | goto err; |
7457 | ||
ec7dc8ac | 7458 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
7459 | goto err; |
7460 | ||
2480c093 PB |
7461 | alloc_uclamp_sched_group(tg, parent); |
7462 | ||
ace783b9 LZ |
7463 | return tg; |
7464 | ||
7465 | err: | |
2f5177f0 | 7466 | sched_free_group(tg); |
ace783b9 LZ |
7467 | return ERR_PTR(-ENOMEM); |
7468 | } | |
7469 | ||
7470 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
7471 | { | |
7472 | unsigned long flags; | |
7473 | ||
8ed36996 | 7474 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 7475 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 7476 | |
d1ccc66d IM |
7477 | /* Root should already exist: */ |
7478 | WARN_ON(!parent); | |
f473aa5e PZ |
7479 | |
7480 | tg->parent = parent; | |
f473aa5e | 7481 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 7482 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 7483 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
7484 | |
7485 | online_fair_sched_group(tg); | |
29f59db3 SV |
7486 | } |
7487 | ||
9b5b7751 | 7488 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 7489 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 7490 | { |
d1ccc66d | 7491 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 7492 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
7493 | } |
7494 | ||
4cf86d77 | 7495 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 7496 | { |
d1ccc66d | 7497 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 7498 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
7499 | } |
7500 | ||
7501 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 7502 | { |
8ed36996 | 7503 | unsigned long flags; |
29f59db3 | 7504 | |
d1ccc66d | 7505 | /* End participation in shares distribution: */ |
6fe1f348 | 7506 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
7507 | |
7508 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 7509 | list_del_rcu(&tg->list); |
f473aa5e | 7510 | list_del_rcu(&tg->siblings); |
8ed36996 | 7511 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
7512 | } |
7513 | ||
ea86cb4b | 7514 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 7515 | { |
8323f26c | 7516 | struct task_group *tg; |
29f59db3 | 7517 | |
f7b8a47d KT |
7518 | /* |
7519 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
7520 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
7521 | * to prevent lockdep warnings. | |
7522 | */ | |
7523 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
7524 | struct task_group, css); |
7525 | tg = autogroup_task_group(tsk, tg); | |
7526 | tsk->sched_task_group = tg; | |
7527 | ||
810b3817 | 7528 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
7529 | if (tsk->sched_class->task_change_group) |
7530 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 7531 | else |
810b3817 | 7532 | #endif |
b2b5ce02 | 7533 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
7534 | } |
7535 | ||
7536 | /* | |
7537 | * Change task's runqueue when it moves between groups. | |
7538 | * | |
7539 | * The caller of this function should have put the task in its new group by | |
7540 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
7541 | * its new group. | |
7542 | */ | |
7543 | void sched_move_task(struct task_struct *tsk) | |
7544 | { | |
7a57f32a PZ |
7545 | int queued, running, queue_flags = |
7546 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
7547 | struct rq_flags rf; |
7548 | struct rq *rq; | |
7549 | ||
7550 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 7551 | update_rq_clock(rq); |
ea86cb4b VG |
7552 | |
7553 | running = task_current(rq, tsk); | |
7554 | queued = task_on_rq_queued(tsk); | |
7555 | ||
7556 | if (queued) | |
7a57f32a | 7557 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 7558 | if (running) |
ea86cb4b VG |
7559 | put_prev_task(rq, tsk); |
7560 | ||
7561 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 7562 | |
da0c1e65 | 7563 | if (queued) |
7a57f32a | 7564 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 7565 | if (running) { |
03b7fad1 | 7566 | set_next_task(rq, tsk); |
2a4b03ff VG |
7567 | /* |
7568 | * After changing group, the running task may have joined a | |
7569 | * throttled one but it's still the running task. Trigger a | |
7570 | * resched to make sure that task can still run. | |
7571 | */ | |
7572 | resched_curr(rq); | |
7573 | } | |
29f59db3 | 7574 | |
eb580751 | 7575 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 7576 | } |
68318b8e | 7577 | |
a7c6d554 | 7578 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 7579 | { |
a7c6d554 | 7580 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
7581 | } |
7582 | ||
eb95419b TH |
7583 | static struct cgroup_subsys_state * |
7584 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 7585 | { |
eb95419b TH |
7586 | struct task_group *parent = css_tg(parent_css); |
7587 | struct task_group *tg; | |
68318b8e | 7588 | |
eb95419b | 7589 | if (!parent) { |
68318b8e | 7590 | /* This is early initialization for the top cgroup */ |
07e06b01 | 7591 | return &root_task_group.css; |
68318b8e SV |
7592 | } |
7593 | ||
ec7dc8ac | 7594 | tg = sched_create_group(parent); |
68318b8e SV |
7595 | if (IS_ERR(tg)) |
7596 | return ERR_PTR(-ENOMEM); | |
7597 | ||
68318b8e SV |
7598 | return &tg->css; |
7599 | } | |
7600 | ||
96b77745 KK |
7601 | /* Expose task group only after completing cgroup initialization */ |
7602 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
7603 | { | |
7604 | struct task_group *tg = css_tg(css); | |
7605 | struct task_group *parent = css_tg(css->parent); | |
7606 | ||
7607 | if (parent) | |
7608 | sched_online_group(tg, parent); | |
7226017a QY |
7609 | |
7610 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
7611 | /* Propagate the effective uclamp value for the new group */ | |
7612 | cpu_util_update_eff(css); | |
7613 | #endif | |
7614 | ||
96b77745 KK |
7615 | return 0; |
7616 | } | |
7617 | ||
2f5177f0 | 7618 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 7619 | { |
eb95419b | 7620 | struct task_group *tg = css_tg(css); |
ace783b9 | 7621 | |
2f5177f0 | 7622 | sched_offline_group(tg); |
ace783b9 LZ |
7623 | } |
7624 | ||
eb95419b | 7625 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 7626 | { |
eb95419b | 7627 | struct task_group *tg = css_tg(css); |
68318b8e | 7628 | |
2f5177f0 PZ |
7629 | /* |
7630 | * Relies on the RCU grace period between css_released() and this. | |
7631 | */ | |
7632 | sched_free_group(tg); | |
ace783b9 LZ |
7633 | } |
7634 | ||
ea86cb4b VG |
7635 | /* |
7636 | * This is called before wake_up_new_task(), therefore we really only | |
7637 | * have to set its group bits, all the other stuff does not apply. | |
7638 | */ | |
b53202e6 | 7639 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 7640 | { |
ea86cb4b VG |
7641 | struct rq_flags rf; |
7642 | struct rq *rq; | |
7643 | ||
7644 | rq = task_rq_lock(task, &rf); | |
7645 | ||
80f5c1b8 | 7646 | update_rq_clock(rq); |
ea86cb4b VG |
7647 | sched_change_group(task, TASK_SET_GROUP); |
7648 | ||
7649 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
7650 | } |
7651 | ||
1f7dd3e5 | 7652 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 7653 | { |
bb9d97b6 | 7654 | struct task_struct *task; |
1f7dd3e5 | 7655 | struct cgroup_subsys_state *css; |
7dc603c9 | 7656 | int ret = 0; |
bb9d97b6 | 7657 | |
1f7dd3e5 | 7658 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 7659 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 7660 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 7661 | return -EINVAL; |
b68aa230 | 7662 | #endif |
7dc603c9 PZ |
7663 | /* |
7664 | * Serialize against wake_up_new_task() such that if its | |
7665 | * running, we're sure to observe its full state. | |
7666 | */ | |
7667 | raw_spin_lock_irq(&task->pi_lock); | |
7668 | /* | |
7669 | * Avoid calling sched_move_task() before wake_up_new_task() | |
7670 | * has happened. This would lead to problems with PELT, due to | |
7671 | * move wanting to detach+attach while we're not attached yet. | |
7672 | */ | |
7673 | if (task->state == TASK_NEW) | |
7674 | ret = -EINVAL; | |
7675 | raw_spin_unlock_irq(&task->pi_lock); | |
7676 | ||
7677 | if (ret) | |
7678 | break; | |
bb9d97b6 | 7679 | } |
7dc603c9 | 7680 | return ret; |
be367d09 | 7681 | } |
68318b8e | 7682 | |
1f7dd3e5 | 7683 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 7684 | { |
bb9d97b6 | 7685 | struct task_struct *task; |
1f7dd3e5 | 7686 | struct cgroup_subsys_state *css; |
bb9d97b6 | 7687 | |
1f7dd3e5 | 7688 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 7689 | sched_move_task(task); |
68318b8e SV |
7690 | } |
7691 | ||
2480c093 | 7692 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
7693 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
7694 | { | |
7695 | struct cgroup_subsys_state *top_css = css; | |
7696 | struct uclamp_se *uc_parent = NULL; | |
7697 | struct uclamp_se *uc_se = NULL; | |
7698 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 7699 | enum uclamp_id clamp_id; |
0b60ba2d PB |
7700 | unsigned int clamps; |
7701 | ||
7702 | css_for_each_descendant_pre(css, top_css) { | |
7703 | uc_parent = css_tg(css)->parent | |
7704 | ? css_tg(css)->parent->uclamp : NULL; | |
7705 | ||
7706 | for_each_clamp_id(clamp_id) { | |
7707 | /* Assume effective clamps matches requested clamps */ | |
7708 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
7709 | /* Cap effective clamps with parent's effective clamps */ | |
7710 | if (uc_parent && | |
7711 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
7712 | eff[clamp_id] = uc_parent[clamp_id].value; | |
7713 | } | |
7714 | } | |
7715 | /* Ensure protection is always capped by limit */ | |
7716 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
7717 | ||
7718 | /* Propagate most restrictive effective clamps */ | |
7719 | clamps = 0x0; | |
7720 | uc_se = css_tg(css)->uclamp; | |
7721 | for_each_clamp_id(clamp_id) { | |
7722 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
7723 | continue; | |
7724 | uc_se[clamp_id].value = eff[clamp_id]; | |
7725 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
7726 | clamps |= (0x1 << clamp_id); | |
7727 | } | |
babbe170 | 7728 | if (!clamps) { |
0b60ba2d | 7729 | css = css_rightmost_descendant(css); |
babbe170 PB |
7730 | continue; |
7731 | } | |
7732 | ||
7733 | /* Immediately update descendants RUNNABLE tasks */ | |
7734 | uclamp_update_active_tasks(css, clamps); | |
0b60ba2d PB |
7735 | } |
7736 | } | |
2480c093 PB |
7737 | |
7738 | /* | |
7739 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
7740 | * C expression. Since there is no way to convert a macro argument (N) into a | |
7741 | * character constant, use two levels of macros. | |
7742 | */ | |
7743 | #define _POW10(exp) ((unsigned int)1e##exp) | |
7744 | #define POW10(exp) _POW10(exp) | |
7745 | ||
7746 | struct uclamp_request { | |
7747 | #define UCLAMP_PERCENT_SHIFT 2 | |
7748 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
7749 | s64 percent; | |
7750 | u64 util; | |
7751 | int ret; | |
7752 | }; | |
7753 | ||
7754 | static inline struct uclamp_request | |
7755 | capacity_from_percent(char *buf) | |
7756 | { | |
7757 | struct uclamp_request req = { | |
7758 | .percent = UCLAMP_PERCENT_SCALE, | |
7759 | .util = SCHED_CAPACITY_SCALE, | |
7760 | .ret = 0, | |
7761 | }; | |
7762 | ||
7763 | buf = strim(buf); | |
7764 | if (strcmp(buf, "max")) { | |
7765 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
7766 | &req.percent); | |
7767 | if (req.ret) | |
7768 | return req; | |
b562d140 | 7769 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
7770 | req.ret = -ERANGE; |
7771 | return req; | |
7772 | } | |
7773 | ||
7774 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
7775 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
7776 | } | |
7777 | ||
7778 | return req; | |
7779 | } | |
7780 | ||
7781 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
7782 | size_t nbytes, loff_t off, | |
7783 | enum uclamp_id clamp_id) | |
7784 | { | |
7785 | struct uclamp_request req; | |
7786 | struct task_group *tg; | |
7787 | ||
7788 | req = capacity_from_percent(buf); | |
7789 | if (req.ret) | |
7790 | return req.ret; | |
7791 | ||
46609ce2 QY |
7792 | static_branch_enable(&sched_uclamp_used); |
7793 | ||
2480c093 PB |
7794 | mutex_lock(&uclamp_mutex); |
7795 | rcu_read_lock(); | |
7796 | ||
7797 | tg = css_tg(of_css(of)); | |
7798 | if (tg->uclamp_req[clamp_id].value != req.util) | |
7799 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
7800 | ||
7801 | /* | |
7802 | * Because of not recoverable conversion rounding we keep track of the | |
7803 | * exact requested value | |
7804 | */ | |
7805 | tg->uclamp_pct[clamp_id] = req.percent; | |
7806 | ||
0b60ba2d PB |
7807 | /* Update effective clamps to track the most restrictive value */ |
7808 | cpu_util_update_eff(of_css(of)); | |
7809 | ||
2480c093 PB |
7810 | rcu_read_unlock(); |
7811 | mutex_unlock(&uclamp_mutex); | |
7812 | ||
7813 | return nbytes; | |
7814 | } | |
7815 | ||
7816 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
7817 | char *buf, size_t nbytes, | |
7818 | loff_t off) | |
7819 | { | |
7820 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
7821 | } | |
7822 | ||
7823 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
7824 | char *buf, size_t nbytes, | |
7825 | loff_t off) | |
7826 | { | |
7827 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
7828 | } | |
7829 | ||
7830 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
7831 | enum uclamp_id clamp_id) | |
7832 | { | |
7833 | struct task_group *tg; | |
7834 | u64 util_clamp; | |
7835 | u64 percent; | |
7836 | u32 rem; | |
7837 | ||
7838 | rcu_read_lock(); | |
7839 | tg = css_tg(seq_css(sf)); | |
7840 | util_clamp = tg->uclamp_req[clamp_id].value; | |
7841 | rcu_read_unlock(); | |
7842 | ||
7843 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
7844 | seq_puts(sf, "max\n"); | |
7845 | return; | |
7846 | } | |
7847 | ||
7848 | percent = tg->uclamp_pct[clamp_id]; | |
7849 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
7850 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
7851 | } | |
7852 | ||
7853 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
7854 | { | |
7855 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
7856 | return 0; | |
7857 | } | |
7858 | ||
7859 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
7860 | { | |
7861 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
7862 | return 0; | |
7863 | } | |
7864 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
7865 | ||
052f1dc7 | 7866 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
7867 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
7868 | struct cftype *cftype, u64 shareval) | |
68318b8e | 7869 | { |
5b61d50a KK |
7870 | if (shareval > scale_load_down(ULONG_MAX)) |
7871 | shareval = MAX_SHARES; | |
182446d0 | 7872 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
7873 | } |
7874 | ||
182446d0 TH |
7875 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
7876 | struct cftype *cft) | |
68318b8e | 7877 | { |
182446d0 | 7878 | struct task_group *tg = css_tg(css); |
68318b8e | 7879 | |
c8b28116 | 7880 | return (u64) scale_load_down(tg->shares); |
68318b8e | 7881 | } |
ab84d31e PT |
7882 | |
7883 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
7884 | static DEFINE_MUTEX(cfs_constraints_mutex); |
7885 | ||
ab84d31e | 7886 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 7887 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
7888 | /* More than 203 days if BW_SHIFT equals 20. */ |
7889 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 7890 | |
a790de99 PT |
7891 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
7892 | ||
ab84d31e PT |
7893 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
7894 | { | |
56f570e5 | 7895 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 7896 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
7897 | |
7898 | if (tg == &root_task_group) | |
7899 | return -EINVAL; | |
7900 | ||
7901 | /* | |
7902 | * Ensure we have at some amount of bandwidth every period. This is | |
7903 | * to prevent reaching a state of large arrears when throttled via | |
7904 | * entity_tick() resulting in prolonged exit starvation. | |
7905 | */ | |
7906 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
7907 | return -EINVAL; | |
7908 | ||
7909 | /* | |
7910 | * Likewise, bound things on the otherside by preventing insane quota | |
7911 | * periods. This also allows us to normalize in computing quota | |
7912 | * feasibility. | |
7913 | */ | |
7914 | if (period > max_cfs_quota_period) | |
7915 | return -EINVAL; | |
7916 | ||
d505b8af HC |
7917 | /* |
7918 | * Bound quota to defend quota against overflow during bandwidth shift. | |
7919 | */ | |
7920 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
7921 | return -EINVAL; | |
7922 | ||
0e59bdae KT |
7923 | /* |
7924 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
7925 | * unthrottle_offline_cfs_rqs(). | |
7926 | */ | |
7927 | get_online_cpus(); | |
a790de99 PT |
7928 | mutex_lock(&cfs_constraints_mutex); |
7929 | ret = __cfs_schedulable(tg, period, quota); | |
7930 | if (ret) | |
7931 | goto out_unlock; | |
7932 | ||
58088ad0 | 7933 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 7934 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
7935 | /* |
7936 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
7937 | * before making related changes, and on->off must occur afterwards | |
7938 | */ | |
7939 | if (runtime_enabled && !runtime_was_enabled) | |
7940 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
7941 | raw_spin_lock_irq(&cfs_b->lock); |
7942 | cfs_b->period = ns_to_ktime(period); | |
7943 | cfs_b->quota = quota; | |
58088ad0 | 7944 | |
a9cf55b2 | 7945 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
7946 | |
7947 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
7948 | if (runtime_enabled) |
7949 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 7950 | |
ab84d31e PT |
7951 | raw_spin_unlock_irq(&cfs_b->lock); |
7952 | ||
0e59bdae | 7953 | for_each_online_cpu(i) { |
ab84d31e | 7954 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 7955 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 7956 | struct rq_flags rf; |
ab84d31e | 7957 | |
8a8c69c3 | 7958 | rq_lock_irq(rq, &rf); |
58088ad0 | 7959 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 7960 | cfs_rq->runtime_remaining = 0; |
671fd9da | 7961 | |
029632fb | 7962 | if (cfs_rq->throttled) |
671fd9da | 7963 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 7964 | rq_unlock_irq(rq, &rf); |
ab84d31e | 7965 | } |
1ee14e6c BS |
7966 | if (runtime_was_enabled && !runtime_enabled) |
7967 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
7968 | out_unlock: |
7969 | mutex_unlock(&cfs_constraints_mutex); | |
0e59bdae | 7970 | put_online_cpus(); |
ab84d31e | 7971 | |
a790de99 | 7972 | return ret; |
ab84d31e PT |
7973 | } |
7974 | ||
b1546edc | 7975 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e PT |
7976 | { |
7977 | u64 quota, period; | |
7978 | ||
029632fb | 7979 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
7980 | if (cfs_quota_us < 0) |
7981 | quota = RUNTIME_INF; | |
1a8b4540 | 7982 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 7983 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
7984 | else |
7985 | return -EINVAL; | |
ab84d31e PT |
7986 | |
7987 | return tg_set_cfs_bandwidth(tg, period, quota); | |
7988 | } | |
7989 | ||
b1546edc | 7990 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
7991 | { |
7992 | u64 quota_us; | |
7993 | ||
029632fb | 7994 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
7995 | return -1; |
7996 | ||
029632fb | 7997 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
7998 | do_div(quota_us, NSEC_PER_USEC); |
7999 | ||
8000 | return quota_us; | |
8001 | } | |
8002 | ||
b1546edc | 8003 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e PT |
8004 | { |
8005 | u64 quota, period; | |
8006 | ||
1a8b4540 KK |
8007 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
8008 | return -EINVAL; | |
8009 | ||
ab84d31e | 8010 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 8011 | quota = tg->cfs_bandwidth.quota; |
ab84d31e | 8012 | |
ab84d31e PT |
8013 | return tg_set_cfs_bandwidth(tg, period, quota); |
8014 | } | |
8015 | ||
b1546edc | 8016 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
8017 | { |
8018 | u64 cfs_period_us; | |
8019 | ||
029632fb | 8020 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
8021 | do_div(cfs_period_us, NSEC_PER_USEC); |
8022 | ||
8023 | return cfs_period_us; | |
8024 | } | |
8025 | ||
182446d0 TH |
8026 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
8027 | struct cftype *cft) | |
ab84d31e | 8028 | { |
182446d0 | 8029 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
8030 | } |
8031 | ||
182446d0 TH |
8032 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
8033 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 8034 | { |
182446d0 | 8035 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
8036 | } |
8037 | ||
182446d0 TH |
8038 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
8039 | struct cftype *cft) | |
ab84d31e | 8040 | { |
182446d0 | 8041 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
8042 | } |
8043 | ||
182446d0 TH |
8044 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
8045 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 8046 | { |
182446d0 | 8047 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
8048 | } |
8049 | ||
a790de99 PT |
8050 | struct cfs_schedulable_data { |
8051 | struct task_group *tg; | |
8052 | u64 period, quota; | |
8053 | }; | |
8054 | ||
8055 | /* | |
8056 | * normalize group quota/period to be quota/max_period | |
8057 | * note: units are usecs | |
8058 | */ | |
8059 | static u64 normalize_cfs_quota(struct task_group *tg, | |
8060 | struct cfs_schedulable_data *d) | |
8061 | { | |
8062 | u64 quota, period; | |
8063 | ||
8064 | if (tg == d->tg) { | |
8065 | period = d->period; | |
8066 | quota = d->quota; | |
8067 | } else { | |
8068 | period = tg_get_cfs_period(tg); | |
8069 | quota = tg_get_cfs_quota(tg); | |
8070 | } | |
8071 | ||
8072 | /* note: these should typically be equivalent */ | |
8073 | if (quota == RUNTIME_INF || quota == -1) | |
8074 | return RUNTIME_INF; | |
8075 | ||
8076 | return to_ratio(period, quota); | |
8077 | } | |
8078 | ||
8079 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
8080 | { | |
8081 | struct cfs_schedulable_data *d = data; | |
029632fb | 8082 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
8083 | s64 quota = 0, parent_quota = -1; |
8084 | ||
8085 | if (!tg->parent) { | |
8086 | quota = RUNTIME_INF; | |
8087 | } else { | |
029632fb | 8088 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
8089 | |
8090 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 8091 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
8092 | |
8093 | /* | |
c53593e5 TH |
8094 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
8095 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 8096 | * limit is set: |
a790de99 | 8097 | */ |
c53593e5 TH |
8098 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
8099 | quota = min(quota, parent_quota); | |
8100 | } else { | |
8101 | if (quota == RUNTIME_INF) | |
8102 | quota = parent_quota; | |
8103 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
8104 | return -EINVAL; | |
8105 | } | |
a790de99 | 8106 | } |
9c58c79a | 8107 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
8108 | |
8109 | return 0; | |
8110 | } | |
8111 | ||
8112 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
8113 | { | |
8277434e | 8114 | int ret; |
a790de99 PT |
8115 | struct cfs_schedulable_data data = { |
8116 | .tg = tg, | |
8117 | .period = period, | |
8118 | .quota = quota, | |
8119 | }; | |
8120 | ||
8121 | if (quota != RUNTIME_INF) { | |
8122 | do_div(data.period, NSEC_PER_USEC); | |
8123 | do_div(data.quota, NSEC_PER_USEC); | |
8124 | } | |
8125 | ||
8277434e PT |
8126 | rcu_read_lock(); |
8127 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
8128 | rcu_read_unlock(); | |
8129 | ||
8130 | return ret; | |
a790de99 | 8131 | } |
e8da1b18 | 8132 | |
a1f7164c | 8133 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 8134 | { |
2da8ca82 | 8135 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 8136 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 8137 | |
44ffc75b TH |
8138 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
8139 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
8140 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 8141 | |
3d6c50c2 YW |
8142 | if (schedstat_enabled() && tg != &root_task_group) { |
8143 | u64 ws = 0; | |
8144 | int i; | |
8145 | ||
8146 | for_each_possible_cpu(i) | |
8147 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
8148 | ||
8149 | seq_printf(sf, "wait_sum %llu\n", ws); | |
8150 | } | |
8151 | ||
e8da1b18 NR |
8152 | return 0; |
8153 | } | |
ab84d31e | 8154 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 8155 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8156 | |
052f1dc7 | 8157 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
8158 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
8159 | struct cftype *cft, s64 val) | |
6f505b16 | 8160 | { |
182446d0 | 8161 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
8162 | } |
8163 | ||
182446d0 TH |
8164 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
8165 | struct cftype *cft) | |
6f505b16 | 8166 | { |
182446d0 | 8167 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 8168 | } |
d0b27fa7 | 8169 | |
182446d0 TH |
8170 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
8171 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 8172 | { |
182446d0 | 8173 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
8174 | } |
8175 | ||
182446d0 TH |
8176 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
8177 | struct cftype *cft) | |
d0b27fa7 | 8178 | { |
182446d0 | 8179 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 8180 | } |
6d6bc0ad | 8181 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8182 | |
a1f7164c | 8183 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 8184 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8185 | { |
8186 | .name = "shares", | |
f4c753b7 PM |
8187 | .read_u64 = cpu_shares_read_u64, |
8188 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8189 | }, |
052f1dc7 | 8190 | #endif |
ab84d31e PT |
8191 | #ifdef CONFIG_CFS_BANDWIDTH |
8192 | { | |
8193 | .name = "cfs_quota_us", | |
8194 | .read_s64 = cpu_cfs_quota_read_s64, | |
8195 | .write_s64 = cpu_cfs_quota_write_s64, | |
8196 | }, | |
8197 | { | |
8198 | .name = "cfs_period_us", | |
8199 | .read_u64 = cpu_cfs_period_read_u64, | |
8200 | .write_u64 = cpu_cfs_period_write_u64, | |
8201 | }, | |
e8da1b18 NR |
8202 | { |
8203 | .name = "stat", | |
a1f7164c | 8204 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 8205 | }, |
ab84d31e | 8206 | #endif |
052f1dc7 | 8207 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8208 | { |
9f0c1e56 | 8209 | .name = "rt_runtime_us", |
06ecb27c PM |
8210 | .read_s64 = cpu_rt_runtime_read, |
8211 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8212 | }, |
d0b27fa7 PZ |
8213 | { |
8214 | .name = "rt_period_us", | |
f4c753b7 PM |
8215 | .read_u64 = cpu_rt_period_read_uint, |
8216 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8217 | }, |
2480c093 PB |
8218 | #endif |
8219 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8220 | { | |
8221 | .name = "uclamp.min", | |
8222 | .flags = CFTYPE_NOT_ON_ROOT, | |
8223 | .seq_show = cpu_uclamp_min_show, | |
8224 | .write = cpu_uclamp_min_write, | |
8225 | }, | |
8226 | { | |
8227 | .name = "uclamp.max", | |
8228 | .flags = CFTYPE_NOT_ON_ROOT, | |
8229 | .seq_show = cpu_uclamp_max_show, | |
8230 | .write = cpu_uclamp_max_write, | |
8231 | }, | |
052f1dc7 | 8232 | #endif |
d1ccc66d | 8233 | { } /* Terminate */ |
68318b8e SV |
8234 | }; |
8235 | ||
d41bf8c9 TH |
8236 | static int cpu_extra_stat_show(struct seq_file *sf, |
8237 | struct cgroup_subsys_state *css) | |
0d593634 | 8238 | { |
0d593634 TH |
8239 | #ifdef CONFIG_CFS_BANDWIDTH |
8240 | { | |
d41bf8c9 | 8241 | struct task_group *tg = css_tg(css); |
0d593634 TH |
8242 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
8243 | u64 throttled_usec; | |
8244 | ||
8245 | throttled_usec = cfs_b->throttled_time; | |
8246 | do_div(throttled_usec, NSEC_PER_USEC); | |
8247 | ||
8248 | seq_printf(sf, "nr_periods %d\n" | |
8249 | "nr_throttled %d\n" | |
8250 | "throttled_usec %llu\n", | |
8251 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
8252 | throttled_usec); | |
8253 | } | |
8254 | #endif | |
8255 | return 0; | |
8256 | } | |
8257 | ||
8258 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8259 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
8260 | struct cftype *cft) | |
8261 | { | |
8262 | struct task_group *tg = css_tg(css); | |
8263 | u64 weight = scale_load_down(tg->shares); | |
8264 | ||
8265 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
8266 | } | |
8267 | ||
8268 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
8269 | struct cftype *cft, u64 weight) | |
8270 | { | |
8271 | /* | |
8272 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
8273 | * values which are 1, 100 and 10000 respectively. While it loses | |
8274 | * a bit of range on both ends, it maps pretty well onto the shares | |
8275 | * value used by scheduler and the round-trip conversions preserve | |
8276 | * the original value over the entire range. | |
8277 | */ | |
8278 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
8279 | return -ERANGE; | |
8280 | ||
8281 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
8282 | ||
8283 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
8284 | } | |
8285 | ||
8286 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
8287 | struct cftype *cft) | |
8288 | { | |
8289 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
8290 | int last_delta = INT_MAX; | |
8291 | int prio, delta; | |
8292 | ||
8293 | /* find the closest nice value to the current weight */ | |
8294 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
8295 | delta = abs(sched_prio_to_weight[prio] - weight); | |
8296 | if (delta >= last_delta) | |
8297 | break; | |
8298 | last_delta = delta; | |
8299 | } | |
8300 | ||
8301 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
8302 | } | |
8303 | ||
8304 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
8305 | struct cftype *cft, s64 nice) | |
8306 | { | |
8307 | unsigned long weight; | |
7281c8de | 8308 | int idx; |
0d593634 TH |
8309 | |
8310 | if (nice < MIN_NICE || nice > MAX_NICE) | |
8311 | return -ERANGE; | |
8312 | ||
7281c8de PZ |
8313 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
8314 | idx = array_index_nospec(idx, 40); | |
8315 | weight = sched_prio_to_weight[idx]; | |
8316 | ||
0d593634 TH |
8317 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
8318 | } | |
8319 | #endif | |
8320 | ||
8321 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
8322 | long period, long quota) | |
8323 | { | |
8324 | if (quota < 0) | |
8325 | seq_puts(sf, "max"); | |
8326 | else | |
8327 | seq_printf(sf, "%ld", quota); | |
8328 | ||
8329 | seq_printf(sf, " %ld\n", period); | |
8330 | } | |
8331 | ||
8332 | /* caller should put the current value in *@periodp before calling */ | |
8333 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
8334 | u64 *periodp, u64 *quotap) | |
8335 | { | |
8336 | char tok[21]; /* U64_MAX */ | |
8337 | ||
4c47acd8 | 8338 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
8339 | return -EINVAL; |
8340 | ||
8341 | *periodp *= NSEC_PER_USEC; | |
8342 | ||
8343 | if (sscanf(tok, "%llu", quotap)) | |
8344 | *quotap *= NSEC_PER_USEC; | |
8345 | else if (!strcmp(tok, "max")) | |
8346 | *quotap = RUNTIME_INF; | |
8347 | else | |
8348 | return -EINVAL; | |
8349 | ||
8350 | return 0; | |
8351 | } | |
8352 | ||
8353 | #ifdef CONFIG_CFS_BANDWIDTH | |
8354 | static int cpu_max_show(struct seq_file *sf, void *v) | |
8355 | { | |
8356 | struct task_group *tg = css_tg(seq_css(sf)); | |
8357 | ||
8358 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
8359 | return 0; | |
8360 | } | |
8361 | ||
8362 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
8363 | char *buf, size_t nbytes, loff_t off) | |
8364 | { | |
8365 | struct task_group *tg = css_tg(of_css(of)); | |
8366 | u64 period = tg_get_cfs_period(tg); | |
8367 | u64 quota; | |
8368 | int ret; | |
8369 | ||
8370 | ret = cpu_period_quota_parse(buf, &period, "a); | |
8371 | if (!ret) | |
8372 | ret = tg_set_cfs_bandwidth(tg, period, quota); | |
8373 | return ret ?: nbytes; | |
8374 | } | |
8375 | #endif | |
8376 | ||
8377 | static struct cftype cpu_files[] = { | |
0d593634 TH |
8378 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8379 | { | |
8380 | .name = "weight", | |
8381 | .flags = CFTYPE_NOT_ON_ROOT, | |
8382 | .read_u64 = cpu_weight_read_u64, | |
8383 | .write_u64 = cpu_weight_write_u64, | |
8384 | }, | |
8385 | { | |
8386 | .name = "weight.nice", | |
8387 | .flags = CFTYPE_NOT_ON_ROOT, | |
8388 | .read_s64 = cpu_weight_nice_read_s64, | |
8389 | .write_s64 = cpu_weight_nice_write_s64, | |
8390 | }, | |
8391 | #endif | |
8392 | #ifdef CONFIG_CFS_BANDWIDTH | |
8393 | { | |
8394 | .name = "max", | |
8395 | .flags = CFTYPE_NOT_ON_ROOT, | |
8396 | .seq_show = cpu_max_show, | |
8397 | .write = cpu_max_write, | |
8398 | }, | |
2480c093 PB |
8399 | #endif |
8400 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
8401 | { | |
8402 | .name = "uclamp.min", | |
8403 | .flags = CFTYPE_NOT_ON_ROOT, | |
8404 | .seq_show = cpu_uclamp_min_show, | |
8405 | .write = cpu_uclamp_min_write, | |
8406 | }, | |
8407 | { | |
8408 | .name = "uclamp.max", | |
8409 | .flags = CFTYPE_NOT_ON_ROOT, | |
8410 | .seq_show = cpu_uclamp_max_show, | |
8411 | .write = cpu_uclamp_max_write, | |
8412 | }, | |
0d593634 TH |
8413 | #endif |
8414 | { } /* terminate */ | |
8415 | }; | |
8416 | ||
073219e9 | 8417 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 8418 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 8419 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 8420 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 8421 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 8422 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 8423 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
8424 | .can_attach = cpu_cgroup_can_attach, |
8425 | .attach = cpu_cgroup_attach, | |
a1f7164c | 8426 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 8427 | .dfl_cftypes = cpu_files, |
b38e42e9 | 8428 | .early_init = true, |
0d593634 | 8429 | .threaded = true, |
68318b8e SV |
8430 | }; |
8431 | ||
052f1dc7 | 8432 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 8433 | |
b637a328 PM |
8434 | void dump_cpu_task(int cpu) |
8435 | { | |
8436 | pr_info("Task dump for CPU %d:\n", cpu); | |
8437 | sched_show_task(cpu_curr(cpu)); | |
8438 | } | |
ed82b8a1 AK |
8439 | |
8440 | /* | |
8441 | * Nice levels are multiplicative, with a gentle 10% change for every | |
8442 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
8443 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
8444 | * that remained on nice 0. | |
8445 | * | |
8446 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
8447 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
8448 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
8449 | * If a task goes up by ~10% and another task goes down by ~10% then | |
8450 | * the relative distance between them is ~25%.) | |
8451 | */ | |
8452 | const int sched_prio_to_weight[40] = { | |
8453 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
8454 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
8455 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
8456 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
8457 | /* 0 */ 1024, 820, 655, 526, 423, | |
8458 | /* 5 */ 335, 272, 215, 172, 137, | |
8459 | /* 10 */ 110, 87, 70, 56, 45, | |
8460 | /* 15 */ 36, 29, 23, 18, 15, | |
8461 | }; | |
8462 | ||
8463 | /* | |
8464 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
8465 | * | |
8466 | * In cases where the weight does not change often, we can use the | |
8467 | * precalculated inverse to speed up arithmetics by turning divisions | |
8468 | * into multiplications: | |
8469 | */ | |
8470 | const u32 sched_prio_to_wmult[40] = { | |
8471 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
8472 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
8473 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
8474 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
8475 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
8476 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
8477 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
8478 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
8479 | }; | |
14a7405b | 8480 | |
9d246053 PA |
8481 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
8482 | { | |
8483 | trace_sched_update_nr_running_tp(rq, count); | |
8484 | } |