Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
391e43da | 3 | * kernel/sched/core.c |
1da177e4 | 4 | * |
d1ccc66d | 5 | * Core kernel scheduler code and related syscalls |
1da177e4 LT |
6 | * |
7 | * Copyright (C) 1991-2002 Linus Torvalds | |
1da177e4 | 8 | */ |
9d246053 PA |
9 | #define CREATE_TRACE_POINTS |
10 | #include <trace/events/sched.h> | |
11 | #undef CREATE_TRACE_POINTS | |
12 | ||
325ea10c | 13 | #include "sched.h" |
1da177e4 | 14 | |
7281c8de | 15 | #include <linux/nospec.h> |
85f1abe0 | 16 | |
0ed557aa | 17 | #include <linux/kcov.h> |
d08b9f0c | 18 | #include <linux/scs.h> |
0ed557aa | 19 | |
96f951ed | 20 | #include <asm/switch_to.h> |
5517d86b | 21 | #include <asm/tlb.h> |
1da177e4 | 22 | |
ea138446 | 23 | #include "../workqueue_internal.h" |
771b53d0 | 24 | #include "../../fs/io-wq.h" |
29d5e047 | 25 | #include "../smpboot.h" |
6e0534f2 | 26 | |
91c27493 | 27 | #include "pelt.h" |
1f8db415 | 28 | #include "smp.h" |
91c27493 | 29 | |
a056a5be QY |
30 | /* |
31 | * Export tracepoints that act as a bare tracehook (ie: have no trace event | |
32 | * associated with them) to allow external modules to probe them. | |
33 | */ | |
34 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp); | |
35 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp); | |
36 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp); | |
37 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp); | |
38 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp); | |
51cf18c9 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp); |
a056a5be | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); |
4581bea8 VD |
41 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp); |
42 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); | |
9d246053 | 43 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); |
a056a5be | 44 | |
029632fb | 45 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 46 | |
a73f863a | 47 | #ifdef CONFIG_SCHED_DEBUG |
bf5c91ba IM |
48 | /* |
49 | * Debugging: various feature bits | |
765cc3a4 PB |
50 | * |
51 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
52 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
53 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 54 | */ |
f00b45c1 PZ |
55 | #define SCHED_FEAT(name, enabled) \ |
56 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 57 | const_debug unsigned int sysctl_sched_features = |
391e43da | 58 | #include "features.h" |
f00b45c1 | 59 | 0; |
f00b45c1 | 60 | #undef SCHED_FEAT |
c006fac5 PT |
61 | |
62 | /* | |
63 | * Print a warning if need_resched is set for the given duration (if | |
64 | * LATENCY_WARN is enabled). | |
65 | * | |
66 | * If sysctl_resched_latency_warn_once is set, only one warning will be shown | |
67 | * per boot. | |
68 | */ | |
69 | __read_mostly int sysctl_resched_latency_warn_ms = 100; | |
70 | __read_mostly int sysctl_resched_latency_warn_once = 1; | |
71 | #endif /* CONFIG_SCHED_DEBUG */ | |
f00b45c1 | 72 | |
b82d9fdd PZ |
73 | /* |
74 | * Number of tasks to iterate in a single balance run. | |
75 | * Limited because this is done with IRQs disabled. | |
76 | */ | |
77 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
78 | ||
fa85ae24 | 79 | /* |
d1ccc66d | 80 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
81 | * default: 1s |
82 | */ | |
9f0c1e56 | 83 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 84 | |
029632fb | 85 | __read_mostly int scheduler_running; |
6892b75e | 86 | |
9edeaea1 PZ |
87 | #ifdef CONFIG_SCHED_CORE |
88 | ||
89 | DEFINE_STATIC_KEY_FALSE(__sched_core_enabled); | |
90 | ||
8a311c74 PZ |
91 | /* kernel prio, less is more */ |
92 | static inline int __task_prio(struct task_struct *p) | |
93 | { | |
94 | if (p->sched_class == &stop_sched_class) /* trumps deadline */ | |
95 | return -2; | |
96 | ||
97 | if (rt_prio(p->prio)) /* includes deadline */ | |
98 | return p->prio; /* [-1, 99] */ | |
99 | ||
100 | if (p->sched_class == &idle_sched_class) | |
101 | return MAX_RT_PRIO + NICE_WIDTH; /* 140 */ | |
102 | ||
103 | return MAX_RT_PRIO + MAX_NICE; /* 120, squash fair */ | |
104 | } | |
105 | ||
106 | /* | |
107 | * l(a,b) | |
108 | * le(a,b) := !l(b,a) | |
109 | * g(a,b) := l(b,a) | |
110 | * ge(a,b) := !l(a,b) | |
111 | */ | |
112 | ||
113 | /* real prio, less is less */ | |
c6047c2e | 114 | static inline bool prio_less(struct task_struct *a, struct task_struct *b, bool in_fi) |
8a311c74 PZ |
115 | { |
116 | ||
117 | int pa = __task_prio(a), pb = __task_prio(b); | |
118 | ||
119 | if (-pa < -pb) | |
120 | return true; | |
121 | ||
122 | if (-pb < -pa) | |
123 | return false; | |
124 | ||
125 | if (pa == -1) /* dl_prio() doesn't work because of stop_class above */ | |
126 | return !dl_time_before(a->dl.deadline, b->dl.deadline); | |
127 | ||
c6047c2e JFG |
128 | if (pa == MAX_RT_PRIO + MAX_NICE) /* fair */ |
129 | return cfs_prio_less(a, b, in_fi); | |
8a311c74 PZ |
130 | |
131 | return false; | |
132 | } | |
133 | ||
134 | static inline bool __sched_core_less(struct task_struct *a, struct task_struct *b) | |
135 | { | |
136 | if (a->core_cookie < b->core_cookie) | |
137 | return true; | |
138 | ||
139 | if (a->core_cookie > b->core_cookie) | |
140 | return false; | |
141 | ||
142 | /* flip prio, so high prio is leftmost */ | |
c6047c2e | 143 | if (prio_less(b, a, task_rq(a)->core->core_forceidle)) |
8a311c74 PZ |
144 | return true; |
145 | ||
146 | return false; | |
147 | } | |
148 | ||
149 | #define __node_2_sc(node) rb_entry((node), struct task_struct, core_node) | |
150 | ||
151 | static inline bool rb_sched_core_less(struct rb_node *a, const struct rb_node *b) | |
152 | { | |
153 | return __sched_core_less(__node_2_sc(a), __node_2_sc(b)); | |
154 | } | |
155 | ||
156 | static inline int rb_sched_core_cmp(const void *key, const struct rb_node *node) | |
157 | { | |
158 | const struct task_struct *p = __node_2_sc(node); | |
159 | unsigned long cookie = (unsigned long)key; | |
160 | ||
161 | if (cookie < p->core_cookie) | |
162 | return -1; | |
163 | ||
164 | if (cookie > p->core_cookie) | |
165 | return 1; | |
166 | ||
167 | return 0; | |
168 | } | |
169 | ||
6e33cad0 | 170 | void sched_core_enqueue(struct rq *rq, struct task_struct *p) |
8a311c74 PZ |
171 | { |
172 | rq->core->core_task_seq++; | |
173 | ||
174 | if (!p->core_cookie) | |
175 | return; | |
176 | ||
177 | rb_add(&p->core_node, &rq->core_tree, rb_sched_core_less); | |
178 | } | |
179 | ||
6e33cad0 | 180 | void sched_core_dequeue(struct rq *rq, struct task_struct *p) |
8a311c74 PZ |
181 | { |
182 | rq->core->core_task_seq++; | |
183 | ||
6e33cad0 | 184 | if (!sched_core_enqueued(p)) |
8a311c74 PZ |
185 | return; |
186 | ||
187 | rb_erase(&p->core_node, &rq->core_tree); | |
6e33cad0 | 188 | RB_CLEAR_NODE(&p->core_node); |
8a311c74 PZ |
189 | } |
190 | ||
191 | /* | |
192 | * Find left-most (aka, highest priority) task matching @cookie. | |
193 | */ | |
194 | static struct task_struct *sched_core_find(struct rq *rq, unsigned long cookie) | |
195 | { | |
196 | struct rb_node *node; | |
197 | ||
198 | node = rb_find_first((void *)cookie, &rq->core_tree, rb_sched_core_cmp); | |
199 | /* | |
200 | * The idle task always matches any cookie! | |
201 | */ | |
202 | if (!node) | |
203 | return idle_sched_class.pick_task(rq); | |
204 | ||
205 | return __node_2_sc(node); | |
206 | } | |
207 | ||
d2dfa17b PZ |
208 | static struct task_struct *sched_core_next(struct task_struct *p, unsigned long cookie) |
209 | { | |
210 | struct rb_node *node = &p->core_node; | |
211 | ||
212 | node = rb_next(node); | |
213 | if (!node) | |
214 | return NULL; | |
215 | ||
216 | p = container_of(node, struct task_struct, core_node); | |
217 | if (p->core_cookie != cookie) | |
218 | return NULL; | |
219 | ||
220 | return p; | |
221 | } | |
222 | ||
9edeaea1 PZ |
223 | /* |
224 | * Magic required such that: | |
225 | * | |
226 | * raw_spin_rq_lock(rq); | |
227 | * ... | |
228 | * raw_spin_rq_unlock(rq); | |
229 | * | |
230 | * ends up locking and unlocking the _same_ lock, and all CPUs | |
231 | * always agree on what rq has what lock. | |
232 | * | |
233 | * XXX entirely possible to selectively enable cores, don't bother for now. | |
234 | */ | |
235 | ||
236 | static DEFINE_MUTEX(sched_core_mutex); | |
875feb41 | 237 | static atomic_t sched_core_count; |
9edeaea1 PZ |
238 | static struct cpumask sched_core_mask; |
239 | ||
3c474b32 PZ |
240 | static void sched_core_lock(int cpu, unsigned long *flags) |
241 | { | |
242 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
243 | int t, i = 0; | |
244 | ||
245 | local_irq_save(*flags); | |
246 | for_each_cpu(t, smt_mask) | |
247 | raw_spin_lock_nested(&cpu_rq(t)->__lock, i++); | |
248 | } | |
249 | ||
250 | static void sched_core_unlock(int cpu, unsigned long *flags) | |
251 | { | |
252 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
253 | int t; | |
254 | ||
255 | for_each_cpu(t, smt_mask) | |
256 | raw_spin_unlock(&cpu_rq(t)->__lock); | |
257 | local_irq_restore(*flags); | |
258 | } | |
259 | ||
9edeaea1 PZ |
260 | static void __sched_core_flip(bool enabled) |
261 | { | |
3c474b32 PZ |
262 | unsigned long flags; |
263 | int cpu, t; | |
9edeaea1 PZ |
264 | |
265 | cpus_read_lock(); | |
266 | ||
267 | /* | |
268 | * Toggle the online cores, one by one. | |
269 | */ | |
270 | cpumask_copy(&sched_core_mask, cpu_online_mask); | |
271 | for_each_cpu(cpu, &sched_core_mask) { | |
272 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
273 | ||
3c474b32 | 274 | sched_core_lock(cpu, &flags); |
9edeaea1 PZ |
275 | |
276 | for_each_cpu(t, smt_mask) | |
277 | cpu_rq(t)->core_enabled = enabled; | |
278 | ||
3c474b32 | 279 | sched_core_unlock(cpu, &flags); |
9edeaea1 PZ |
280 | |
281 | cpumask_andnot(&sched_core_mask, &sched_core_mask, smt_mask); | |
282 | } | |
283 | ||
284 | /* | |
285 | * Toggle the offline CPUs. | |
286 | */ | |
287 | cpumask_copy(&sched_core_mask, cpu_possible_mask); | |
288 | cpumask_andnot(&sched_core_mask, &sched_core_mask, cpu_online_mask); | |
289 | ||
290 | for_each_cpu(cpu, &sched_core_mask) | |
291 | cpu_rq(cpu)->core_enabled = enabled; | |
292 | ||
293 | cpus_read_unlock(); | |
294 | } | |
295 | ||
8a311c74 | 296 | static void sched_core_assert_empty(void) |
9edeaea1 | 297 | { |
8a311c74 | 298 | int cpu; |
9edeaea1 | 299 | |
8a311c74 PZ |
300 | for_each_possible_cpu(cpu) |
301 | WARN_ON_ONCE(!RB_EMPTY_ROOT(&cpu_rq(cpu)->core_tree)); | |
302 | } | |
303 | ||
304 | static void __sched_core_enable(void) | |
305 | { | |
9edeaea1 PZ |
306 | static_branch_enable(&__sched_core_enabled); |
307 | /* | |
308 | * Ensure all previous instances of raw_spin_rq_*lock() have finished | |
309 | * and future ones will observe !sched_core_disabled(). | |
310 | */ | |
311 | synchronize_rcu(); | |
312 | __sched_core_flip(true); | |
8a311c74 | 313 | sched_core_assert_empty(); |
9edeaea1 PZ |
314 | } |
315 | ||
316 | static void __sched_core_disable(void) | |
317 | { | |
8a311c74 | 318 | sched_core_assert_empty(); |
9edeaea1 PZ |
319 | __sched_core_flip(false); |
320 | static_branch_disable(&__sched_core_enabled); | |
321 | } | |
322 | ||
323 | void sched_core_get(void) | |
324 | { | |
875feb41 PZ |
325 | if (atomic_inc_not_zero(&sched_core_count)) |
326 | return; | |
327 | ||
9edeaea1 | 328 | mutex_lock(&sched_core_mutex); |
875feb41 | 329 | if (!atomic_read(&sched_core_count)) |
9edeaea1 | 330 | __sched_core_enable(); |
875feb41 PZ |
331 | |
332 | smp_mb__before_atomic(); | |
333 | atomic_inc(&sched_core_count); | |
9edeaea1 PZ |
334 | mutex_unlock(&sched_core_mutex); |
335 | } | |
336 | ||
875feb41 | 337 | static void __sched_core_put(struct work_struct *work) |
9edeaea1 | 338 | { |
875feb41 | 339 | if (atomic_dec_and_mutex_lock(&sched_core_count, &sched_core_mutex)) { |
9edeaea1 | 340 | __sched_core_disable(); |
875feb41 PZ |
341 | mutex_unlock(&sched_core_mutex); |
342 | } | |
343 | } | |
344 | ||
345 | void sched_core_put(void) | |
346 | { | |
347 | static DECLARE_WORK(_work, __sched_core_put); | |
348 | ||
349 | /* | |
350 | * "There can be only one" | |
351 | * | |
352 | * Either this is the last one, or we don't actually need to do any | |
353 | * 'work'. If it is the last *again*, we rely on | |
354 | * WORK_STRUCT_PENDING_BIT. | |
355 | */ | |
356 | if (!atomic_add_unless(&sched_core_count, -1, 1)) | |
357 | schedule_work(&_work); | |
9edeaea1 PZ |
358 | } |
359 | ||
8a311c74 PZ |
360 | #else /* !CONFIG_SCHED_CORE */ |
361 | ||
362 | static inline void sched_core_enqueue(struct rq *rq, struct task_struct *p) { } | |
363 | static inline void sched_core_dequeue(struct rq *rq, struct task_struct *p) { } | |
364 | ||
9edeaea1 PZ |
365 | #endif /* CONFIG_SCHED_CORE */ |
366 | ||
9f0c1e56 PZ |
367 | /* |
368 | * part of the period that we allow rt tasks to run in us. | |
369 | * default: 0.95s | |
370 | */ | |
371 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 372 | |
58877d34 PZ |
373 | |
374 | /* | |
375 | * Serialization rules: | |
376 | * | |
377 | * Lock order: | |
378 | * | |
379 | * p->pi_lock | |
380 | * rq->lock | |
381 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
382 | * | |
383 | * rq1->lock | |
384 | * rq2->lock where: rq1 < rq2 | |
385 | * | |
386 | * Regular state: | |
387 | * | |
388 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
389 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
b19a888c | 390 | * always looks at the local rq data structures to find the most eligible task |
58877d34 PZ |
391 | * to run next. |
392 | * | |
393 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
394 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
395 | * the local CPU to avoid bouncing the runqueue state around [ see | |
396 | * ttwu_queue_wakelist() ] | |
397 | * | |
398 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
399 | * complicated to avoid having to take two rq->locks. | |
400 | * | |
401 | * Special state: | |
402 | * | |
403 | * System-calls and anything external will use task_rq_lock() which acquires | |
404 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
405 | * stable while holding either lock: | |
406 | * | |
407 | * - sched_setaffinity()/ | |
408 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
409 | * - set_user_nice(): p->se.load, p->*prio | |
410 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
411 | * p->se.load, p->rt_priority, | |
412 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
413 | * - sched_setnuma(): p->numa_preferred_nid | |
414 | * - sched_move_task()/ | |
415 | * cpu_cgroup_fork(): p->sched_task_group | |
416 | * - uclamp_update_active() p->uclamp* | |
417 | * | |
418 | * p->state <- TASK_*: | |
419 | * | |
420 | * is changed locklessly using set_current_state(), __set_current_state() or | |
421 | * set_special_state(), see their respective comments, or by | |
422 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
423 | * concurrent self. | |
424 | * | |
425 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
426 | * | |
427 | * is set by activate_task() and cleared by deactivate_task(), under | |
428 | * rq->lock. Non-zero indicates the task is runnable, the special | |
429 | * ON_RQ_MIGRATING state is used for migration without holding both | |
430 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
431 | * | |
432 | * p->on_cpu <- { 0, 1 }: | |
433 | * | |
434 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
435 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
436 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
437 | * | |
438 | * [ The astute reader will observe that it is possible for two tasks on one | |
439 | * CPU to have ->on_cpu = 1 at the same time. ] | |
440 | * | |
441 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
442 | * | |
443 | * - Don't call set_task_cpu() on a blocked task: | |
444 | * | |
445 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
446 | * the CPU assignment of blocked tasks isn't required to be valid. | |
447 | * | |
448 | * - for try_to_wake_up(), called under p->pi_lock: | |
449 | * | |
450 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
451 | * | |
452 | * - for migration called under rq->lock: | |
453 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
454 | * | |
455 | * o move_queued_task() | |
456 | * o detach_task() | |
457 | * | |
458 | * - for migration called under double_rq_lock(): | |
459 | * | |
460 | * o __migrate_swap_task() | |
461 | * o push_rt_task() / pull_rt_task() | |
462 | * o push_dl_task() / pull_dl_task() | |
463 | * o dl_task_offline_migration() | |
464 | * | |
465 | */ | |
466 | ||
39d371b7 PZ |
467 | void raw_spin_rq_lock_nested(struct rq *rq, int subclass) |
468 | { | |
d66f1b06 PZ |
469 | raw_spinlock_t *lock; |
470 | ||
9edeaea1 PZ |
471 | /* Matches synchronize_rcu() in __sched_core_enable() */ |
472 | preempt_disable(); | |
d66f1b06 PZ |
473 | if (sched_core_disabled()) { |
474 | raw_spin_lock_nested(&rq->__lock, subclass); | |
9edeaea1 PZ |
475 | /* preempt_count *MUST* be > 1 */ |
476 | preempt_enable_no_resched(); | |
d66f1b06 PZ |
477 | return; |
478 | } | |
479 | ||
480 | for (;;) { | |
9ef7e7e3 | 481 | lock = __rq_lockp(rq); |
d66f1b06 | 482 | raw_spin_lock_nested(lock, subclass); |
9ef7e7e3 | 483 | if (likely(lock == __rq_lockp(rq))) { |
9edeaea1 PZ |
484 | /* preempt_count *MUST* be > 1 */ |
485 | preempt_enable_no_resched(); | |
d66f1b06 | 486 | return; |
9edeaea1 | 487 | } |
d66f1b06 PZ |
488 | raw_spin_unlock(lock); |
489 | } | |
39d371b7 PZ |
490 | } |
491 | ||
492 | bool raw_spin_rq_trylock(struct rq *rq) | |
493 | { | |
d66f1b06 PZ |
494 | raw_spinlock_t *lock; |
495 | bool ret; | |
496 | ||
9edeaea1 PZ |
497 | /* Matches synchronize_rcu() in __sched_core_enable() */ |
498 | preempt_disable(); | |
499 | if (sched_core_disabled()) { | |
500 | ret = raw_spin_trylock(&rq->__lock); | |
501 | preempt_enable(); | |
502 | return ret; | |
503 | } | |
d66f1b06 PZ |
504 | |
505 | for (;;) { | |
9ef7e7e3 | 506 | lock = __rq_lockp(rq); |
d66f1b06 | 507 | ret = raw_spin_trylock(lock); |
9ef7e7e3 | 508 | if (!ret || (likely(lock == __rq_lockp(rq)))) { |
9edeaea1 | 509 | preempt_enable(); |
d66f1b06 | 510 | return ret; |
9edeaea1 | 511 | } |
d66f1b06 PZ |
512 | raw_spin_unlock(lock); |
513 | } | |
39d371b7 PZ |
514 | } |
515 | ||
516 | void raw_spin_rq_unlock(struct rq *rq) | |
517 | { | |
518 | raw_spin_unlock(rq_lockp(rq)); | |
519 | } | |
520 | ||
d66f1b06 PZ |
521 | #ifdef CONFIG_SMP |
522 | /* | |
523 | * double_rq_lock - safely lock two runqueues | |
524 | */ | |
525 | void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
526 | { | |
527 | lockdep_assert_irqs_disabled(); | |
528 | ||
529 | if (rq_order_less(rq2, rq1)) | |
530 | swap(rq1, rq2); | |
531 | ||
532 | raw_spin_rq_lock(rq1); | |
9ef7e7e3 | 533 | if (__rq_lockp(rq1) == __rq_lockp(rq2)) |
d66f1b06 PZ |
534 | return; |
535 | ||
536 | raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING); | |
537 | } | |
538 | #endif | |
539 | ||
3e71a462 PZ |
540 | /* |
541 | * __task_rq_lock - lock the rq @p resides on. | |
542 | */ | |
eb580751 | 543 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
544 | __acquires(rq->lock) |
545 | { | |
546 | struct rq *rq; | |
547 | ||
548 | lockdep_assert_held(&p->pi_lock); | |
549 | ||
550 | for (;;) { | |
551 | rq = task_rq(p); | |
5cb9eaa3 | 552 | raw_spin_rq_lock(rq); |
3e71a462 | 553 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
d8ac8971 | 554 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
555 | return rq; |
556 | } | |
5cb9eaa3 | 557 | raw_spin_rq_unlock(rq); |
3e71a462 PZ |
558 | |
559 | while (unlikely(task_on_rq_migrating(p))) | |
560 | cpu_relax(); | |
561 | } | |
562 | } | |
563 | ||
564 | /* | |
565 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
566 | */ | |
eb580751 | 567 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
568 | __acquires(p->pi_lock) |
569 | __acquires(rq->lock) | |
570 | { | |
571 | struct rq *rq; | |
572 | ||
573 | for (;;) { | |
eb580751 | 574 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 | 575 | rq = task_rq(p); |
5cb9eaa3 | 576 | raw_spin_rq_lock(rq); |
3e71a462 PZ |
577 | /* |
578 | * move_queued_task() task_rq_lock() | |
579 | * | |
580 | * ACQUIRE (rq->lock) | |
581 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
582 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
583 | * [S] ->cpu = new_cpu [L] task_rq() | |
584 | * [L] ->on_rq | |
585 | * RELEASE (rq->lock) | |
586 | * | |
c546951d | 587 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
588 | * the old rq->lock will fully serialize against the stores. |
589 | * | |
c546951d AP |
590 | * If we observe the new CPU in task_rq_lock(), the address |
591 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
592 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
593 | */ |
594 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 595 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
596 | return rq; |
597 | } | |
5cb9eaa3 | 598 | raw_spin_rq_unlock(rq); |
eb580751 | 599 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
600 | |
601 | while (unlikely(task_on_rq_migrating(p))) | |
602 | cpu_relax(); | |
603 | } | |
604 | } | |
605 | ||
535b9552 IM |
606 | /* |
607 | * RQ-clock updating methods: | |
608 | */ | |
609 | ||
610 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
611 | { | |
612 | /* | |
613 | * In theory, the compile should just see 0 here, and optimize out the call | |
614 | * to sched_rt_avg_update. But I don't trust it... | |
615 | */ | |
11d4afd4 VG |
616 | s64 __maybe_unused steal = 0, irq_delta = 0; |
617 | ||
535b9552 IM |
618 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
619 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
620 | ||
621 | /* | |
622 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
623 | * this case when a previous update_rq_clock() happened inside a | |
624 | * {soft,}irq region. | |
625 | * | |
626 | * When this happens, we stop ->clock_task and only update the | |
627 | * prev_irq_time stamp to account for the part that fit, so that a next | |
628 | * update will consume the rest. This ensures ->clock_task is | |
629 | * monotonic. | |
630 | * | |
631 | * It does however cause some slight miss-attribution of {soft,}irq | |
632 | * time, a more accurate solution would be to update the irq_time using | |
633 | * the current rq->clock timestamp, except that would require using | |
634 | * atomic ops. | |
635 | */ | |
636 | if (irq_delta > delta) | |
637 | irq_delta = delta; | |
638 | ||
639 | rq->prev_irq_time += irq_delta; | |
640 | delta -= irq_delta; | |
641 | #endif | |
642 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
643 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
644 | steal = paravirt_steal_clock(cpu_of(rq)); | |
645 | steal -= rq->prev_steal_time_rq; | |
646 | ||
647 | if (unlikely(steal > delta)) | |
648 | steal = delta; | |
649 | ||
650 | rq->prev_steal_time_rq += steal; | |
651 | delta -= steal; | |
652 | } | |
653 | #endif | |
654 | ||
655 | rq->clock_task += delta; | |
656 | ||
11d4afd4 | 657 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 658 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 659 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 660 | #endif |
23127296 | 661 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
662 | } |
663 | ||
664 | void update_rq_clock(struct rq *rq) | |
665 | { | |
666 | s64 delta; | |
667 | ||
5cb9eaa3 | 668 | lockdep_assert_rq_held(rq); |
535b9552 IM |
669 | |
670 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
671 | return; | |
672 | ||
673 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
674 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
675 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
676 | rq->clock_update_flags |= RQCF_UPDATED; |
677 | #endif | |
26ae58d2 | 678 | |
535b9552 IM |
679 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
680 | if (delta < 0) | |
681 | return; | |
682 | rq->clock += delta; | |
683 | update_rq_clock_task(rq, delta); | |
684 | } | |
685 | ||
8f4d37ec PZ |
686 | #ifdef CONFIG_SCHED_HRTICK |
687 | /* | |
688 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 689 | */ |
8f4d37ec | 690 | |
8f4d37ec PZ |
691 | static void hrtick_clear(struct rq *rq) |
692 | { | |
693 | if (hrtimer_active(&rq->hrtick_timer)) | |
694 | hrtimer_cancel(&rq->hrtick_timer); | |
695 | } | |
696 | ||
8f4d37ec PZ |
697 | /* |
698 | * High-resolution timer tick. | |
699 | * Runs from hardirq context with interrupts disabled. | |
700 | */ | |
701 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
702 | { | |
703 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 704 | struct rq_flags rf; |
8f4d37ec PZ |
705 | |
706 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
707 | ||
8a8c69c3 | 708 | rq_lock(rq, &rf); |
3e51f33f | 709 | update_rq_clock(rq); |
8f4d37ec | 710 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 711 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
712 | |
713 | return HRTIMER_NORESTART; | |
714 | } | |
715 | ||
95e904c7 | 716 | #ifdef CONFIG_SMP |
971ee28c | 717 | |
4961b6e1 | 718 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
719 | { |
720 | struct hrtimer *timer = &rq->hrtick_timer; | |
156ec6f4 | 721 | ktime_t time = rq->hrtick_time; |
971ee28c | 722 | |
156ec6f4 | 723 | hrtimer_start(timer, time, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
724 | } |
725 | ||
31656519 PZ |
726 | /* |
727 | * called from hardirq (IPI) context | |
728 | */ | |
729 | static void __hrtick_start(void *arg) | |
b328ca18 | 730 | { |
31656519 | 731 | struct rq *rq = arg; |
8a8c69c3 | 732 | struct rq_flags rf; |
b328ca18 | 733 | |
8a8c69c3 | 734 | rq_lock(rq, &rf); |
971ee28c | 735 | __hrtick_restart(rq); |
8a8c69c3 | 736 | rq_unlock(rq, &rf); |
b328ca18 PZ |
737 | } |
738 | ||
31656519 PZ |
739 | /* |
740 | * Called to set the hrtick timer state. | |
741 | * | |
742 | * called with rq->lock held and irqs disabled | |
743 | */ | |
029632fb | 744 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 745 | { |
31656519 | 746 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 747 | s64 delta; |
748 | ||
749 | /* | |
750 | * Don't schedule slices shorter than 10000ns, that just | |
751 | * doesn't make sense and can cause timer DoS. | |
752 | */ | |
753 | delta = max_t(s64, delay, 10000LL); | |
156ec6f4 | 754 | rq->hrtick_time = ktime_add_ns(timer->base->get_time(), delta); |
31656519 | 755 | |
fd3eafda | 756 | if (rq == this_rq()) |
971ee28c | 757 | __hrtick_restart(rq); |
fd3eafda | 758 | else |
c46fff2a | 759 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
760 | } |
761 | ||
31656519 PZ |
762 | #else |
763 | /* | |
764 | * Called to set the hrtick timer state. | |
765 | * | |
766 | * called with rq->lock held and irqs disabled | |
767 | */ | |
029632fb | 768 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 769 | { |
86893335 WL |
770 | /* |
771 | * Don't schedule slices shorter than 10000ns, that just | |
772 | * doesn't make sense. Rely on vruntime for fairness. | |
773 | */ | |
774 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 775 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 776 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 777 | } |
90b5363a | 778 | |
31656519 | 779 | #endif /* CONFIG_SMP */ |
8f4d37ec | 780 | |
77a021be | 781 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 782 | { |
31656519 | 783 | #ifdef CONFIG_SMP |
545b8c8d | 784 | INIT_CSD(&rq->hrtick_csd, __hrtick_start, rq); |
31656519 | 785 | #endif |
d5096aa6 | 786 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 787 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 788 | } |
006c75f1 | 789 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
790 | static inline void hrtick_clear(struct rq *rq) |
791 | { | |
792 | } | |
793 | ||
77a021be | 794 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
795 | { |
796 | } | |
006c75f1 | 797 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 798 | |
5529578a FW |
799 | /* |
800 | * cmpxchg based fetch_or, macro so it works for different integer types | |
801 | */ | |
802 | #define fetch_or(ptr, mask) \ | |
803 | ({ \ | |
804 | typeof(ptr) _ptr = (ptr); \ | |
805 | typeof(mask) _mask = (mask); \ | |
806 | typeof(*_ptr) _old, _val = *_ptr; \ | |
807 | \ | |
808 | for (;;) { \ | |
809 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
810 | if (_old == _val) \ | |
811 | break; \ | |
812 | _val = _old; \ | |
813 | } \ | |
814 | _old; \ | |
815 | }) | |
816 | ||
e3baac47 | 817 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
818 | /* |
819 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
820 | * this avoids any races wrt polling state changes and thereby avoids | |
821 | * spurious IPIs. | |
822 | */ | |
823 | static bool set_nr_and_not_polling(struct task_struct *p) | |
824 | { | |
825 | struct thread_info *ti = task_thread_info(p); | |
826 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
827 | } | |
e3baac47 PZ |
828 | |
829 | /* | |
830 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
831 | * | |
832 | * If this returns true, then the idle task promises to call | |
833 | * sched_ttwu_pending() and reschedule soon. | |
834 | */ | |
835 | static bool set_nr_if_polling(struct task_struct *p) | |
836 | { | |
837 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 838 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
839 | |
840 | for (;;) { | |
841 | if (!(val & _TIF_POLLING_NRFLAG)) | |
842 | return false; | |
843 | if (val & _TIF_NEED_RESCHED) | |
844 | return true; | |
845 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
846 | if (old == val) | |
847 | break; | |
848 | val = old; | |
849 | } | |
850 | return true; | |
851 | } | |
852 | ||
fd99f91a PZ |
853 | #else |
854 | static bool set_nr_and_not_polling(struct task_struct *p) | |
855 | { | |
856 | set_tsk_need_resched(p); | |
857 | return true; | |
858 | } | |
e3baac47 PZ |
859 | |
860 | #ifdef CONFIG_SMP | |
861 | static bool set_nr_if_polling(struct task_struct *p) | |
862 | { | |
863 | return false; | |
864 | } | |
865 | #endif | |
fd99f91a PZ |
866 | #endif |
867 | ||
07879c6a | 868 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
869 | { |
870 | struct wake_q_node *node = &task->wake_q; | |
871 | ||
872 | /* | |
873 | * Atomically grab the task, if ->wake_q is !nil already it means | |
b19a888c | 874 | * it's already queued (either by us or someone else) and will get the |
76751049 PZ |
875 | * wakeup due to that. |
876 | * | |
4c4e3731 PZ |
877 | * In order to ensure that a pending wakeup will observe our pending |
878 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 879 | */ |
4c4e3731 | 880 | smp_mb__before_atomic(); |
87ff19cb | 881 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 882 | return false; |
76751049 PZ |
883 | |
884 | /* | |
885 | * The head is context local, there can be no concurrency. | |
886 | */ | |
887 | *head->lastp = node; | |
888 | head->lastp = &node->next; | |
07879c6a DB |
889 | return true; |
890 | } | |
891 | ||
892 | /** | |
893 | * wake_q_add() - queue a wakeup for 'later' waking. | |
894 | * @head: the wake_q_head to add @task to | |
895 | * @task: the task to queue for 'later' wakeup | |
896 | * | |
897 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
898 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
899 | * instantly. | |
900 | * | |
901 | * This function must be used as-if it were wake_up_process(); IOW the task | |
902 | * must be ready to be woken at this location. | |
903 | */ | |
904 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
905 | { | |
906 | if (__wake_q_add(head, task)) | |
907 | get_task_struct(task); | |
908 | } | |
909 | ||
910 | /** | |
911 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
912 | * @head: the wake_q_head to add @task to | |
913 | * @task: the task to queue for 'later' wakeup | |
914 | * | |
915 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
916 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
917 | * instantly. | |
918 | * | |
919 | * This function must be used as-if it were wake_up_process(); IOW the task | |
920 | * must be ready to be woken at this location. | |
921 | * | |
922 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
923 | * that already hold reference to @task can call the 'safe' version and trust | |
924 | * wake_q to do the right thing depending whether or not the @task is already | |
925 | * queued for wakeup. | |
926 | */ | |
927 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
928 | { | |
929 | if (!__wake_q_add(head, task)) | |
930 | put_task_struct(task); | |
76751049 PZ |
931 | } |
932 | ||
933 | void wake_up_q(struct wake_q_head *head) | |
934 | { | |
935 | struct wake_q_node *node = head->first; | |
936 | ||
937 | while (node != WAKE_Q_TAIL) { | |
938 | struct task_struct *task; | |
939 | ||
940 | task = container_of(node, struct task_struct, wake_q); | |
d1ccc66d | 941 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
942 | node = node->next; |
943 | task->wake_q.next = NULL; | |
944 | ||
945 | /* | |
7696f991 AP |
946 | * wake_up_process() executes a full barrier, which pairs with |
947 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
948 | */ |
949 | wake_up_process(task); | |
950 | put_task_struct(task); | |
951 | } | |
952 | } | |
953 | ||
c24d20db | 954 | /* |
8875125e | 955 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
956 | * |
957 | * On UP this means the setting of the need_resched flag, on SMP it | |
958 | * might also involve a cross-CPU call to trigger the scheduler on | |
959 | * the target CPU. | |
960 | */ | |
8875125e | 961 | void resched_curr(struct rq *rq) |
c24d20db | 962 | { |
8875125e | 963 | struct task_struct *curr = rq->curr; |
c24d20db IM |
964 | int cpu; |
965 | ||
5cb9eaa3 | 966 | lockdep_assert_rq_held(rq); |
c24d20db | 967 | |
8875125e | 968 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
969 | return; |
970 | ||
8875125e | 971 | cpu = cpu_of(rq); |
fd99f91a | 972 | |
f27dde8d | 973 | if (cpu == smp_processor_id()) { |
8875125e | 974 | set_tsk_need_resched(curr); |
f27dde8d | 975 | set_preempt_need_resched(); |
c24d20db | 976 | return; |
f27dde8d | 977 | } |
c24d20db | 978 | |
8875125e | 979 | if (set_nr_and_not_polling(curr)) |
c24d20db | 980 | smp_send_reschedule(cpu); |
dfc68f29 AL |
981 | else |
982 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
983 | } |
984 | ||
029632fb | 985 | void resched_cpu(int cpu) |
c24d20db IM |
986 | { |
987 | struct rq *rq = cpu_rq(cpu); | |
988 | unsigned long flags; | |
989 | ||
5cb9eaa3 | 990 | raw_spin_rq_lock_irqsave(rq, flags); |
a0982dfa PM |
991 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
992 | resched_curr(rq); | |
5cb9eaa3 | 993 | raw_spin_rq_unlock_irqrestore(rq, flags); |
c24d20db | 994 | } |
06d8308c | 995 | |
b021fe3e | 996 | #ifdef CONFIG_SMP |
3451d024 | 997 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 998 | /* |
d1ccc66d IM |
999 | * In the semi idle case, use the nearest busy CPU for migrating timers |
1000 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
1001 | * |
1002 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
1003 | * selecting an idle CPU will add more delays to the timers than intended |
1004 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 1005 | */ |
bc7a34b8 | 1006 | int get_nohz_timer_target(void) |
83cd4fe2 | 1007 | { |
e938b9c9 | 1008 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 | 1009 | struct sched_domain *sd; |
031e3bd8 | 1010 | const struct cpumask *hk_mask; |
83cd4fe2 | 1011 | |
e938b9c9 WL |
1012 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
1013 | if (!idle_cpu(cpu)) | |
1014 | return cpu; | |
1015 | default_cpu = cpu; | |
1016 | } | |
6201b4d6 | 1017 | |
031e3bd8 YZ |
1018 | hk_mask = housekeeping_cpumask(HK_FLAG_TIMER); |
1019 | ||
057f3fad | 1020 | rcu_read_lock(); |
83cd4fe2 | 1021 | for_each_domain(cpu, sd) { |
031e3bd8 | 1022 | for_each_cpu_and(i, sched_domain_span(sd), hk_mask) { |
44496922 WL |
1023 | if (cpu == i) |
1024 | continue; | |
1025 | ||
e938b9c9 | 1026 | if (!idle_cpu(i)) { |
057f3fad PZ |
1027 | cpu = i; |
1028 | goto unlock; | |
1029 | } | |
1030 | } | |
83cd4fe2 | 1031 | } |
9642d18e | 1032 | |
e938b9c9 WL |
1033 | if (default_cpu == -1) |
1034 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
1035 | cpu = default_cpu; | |
057f3fad PZ |
1036 | unlock: |
1037 | rcu_read_unlock(); | |
83cd4fe2 VP |
1038 | return cpu; |
1039 | } | |
d1ccc66d | 1040 | |
06d8308c TG |
1041 | /* |
1042 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1043 | * idle CPU then this timer might expire before the next timer event | |
1044 | * which is scheduled to wake up that CPU. In case of a completely | |
1045 | * idle system the next event might even be infinite time into the | |
1046 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1047 | * leaves the inner idle loop so the newly added timer is taken into | |
1048 | * account when the CPU goes back to idle and evaluates the timer | |
1049 | * wheel for the next timer event. | |
1050 | */ | |
1c20091e | 1051 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
1052 | { |
1053 | struct rq *rq = cpu_rq(cpu); | |
1054 | ||
1055 | if (cpu == smp_processor_id()) | |
1056 | return; | |
1057 | ||
67b9ca70 | 1058 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 1059 | smp_send_reschedule(cpu); |
dfc68f29 AL |
1060 | else |
1061 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
1062 | } |
1063 | ||
c5bfece2 | 1064 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 1065 | { |
53c5fa16 FW |
1066 | /* |
1067 | * We just need the target to call irq_exit() and re-evaluate | |
1068 | * the next tick. The nohz full kick at least implies that. | |
1069 | * If needed we can still optimize that later with an | |
1070 | * empty IRQ. | |
1071 | */ | |
379d9ecb PM |
1072 | if (cpu_is_offline(cpu)) |
1073 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 1074 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
1075 | if (cpu != smp_processor_id() || |
1076 | tick_nohz_tick_stopped()) | |
53c5fa16 | 1077 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
1078 | return true; |
1079 | } | |
1080 | ||
1081 | return false; | |
1082 | } | |
1083 | ||
379d9ecb PM |
1084 | /* |
1085 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
1086 | * caller's responsibility to deal with the lost wakeup, for example, | |
1087 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
1088 | */ | |
1c20091e FW |
1089 | void wake_up_nohz_cpu(int cpu) |
1090 | { | |
c5bfece2 | 1091 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
1092 | wake_up_idle_cpu(cpu); |
1093 | } | |
1094 | ||
19a1f5ec | 1095 | static void nohz_csd_func(void *info) |
45bf76df | 1096 | { |
19a1f5ec PZ |
1097 | struct rq *rq = info; |
1098 | int cpu = cpu_of(rq); | |
1099 | unsigned int flags; | |
873b4c65 VG |
1100 | |
1101 | /* | |
19a1f5ec | 1102 | * Release the rq::nohz_csd. |
873b4c65 | 1103 | */ |
c6f88654 | 1104 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK | NOHZ_NEWILB_KICK, nohz_flags(cpu)); |
19a1f5ec | 1105 | WARN_ON(!(flags & NOHZ_KICK_MASK)); |
45bf76df | 1106 | |
19a1f5ec PZ |
1107 | rq->idle_balance = idle_cpu(cpu); |
1108 | if (rq->idle_balance && !need_resched()) { | |
1109 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
1110 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
1111 | } | |
2069dd75 PZ |
1112 | } |
1113 | ||
3451d024 | 1114 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 1115 | |
ce831b38 | 1116 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 1117 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 1118 | { |
76d92ac3 FW |
1119 | int fifo_nr_running; |
1120 | ||
1121 | /* Deadline tasks, even if single, need the tick */ | |
1122 | if (rq->dl.dl_nr_running) | |
1123 | return false; | |
1124 | ||
1e78cdbd | 1125 | /* |
b19a888c | 1126 | * If there are more than one RR tasks, we need the tick to affect the |
2548d546 | 1127 | * actual RR behaviour. |
1e78cdbd | 1128 | */ |
76d92ac3 FW |
1129 | if (rq->rt.rr_nr_running) { |
1130 | if (rq->rt.rr_nr_running == 1) | |
1131 | return true; | |
1132 | else | |
1133 | return false; | |
1e78cdbd RR |
1134 | } |
1135 | ||
2548d546 PZ |
1136 | /* |
1137 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
1138 | * forced preemption between FIFO tasks. | |
1139 | */ | |
1140 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
1141 | if (fifo_nr_running) | |
1142 | return true; | |
1143 | ||
1144 | /* | |
1145 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
1146 | * if there's more than one we need the tick for involuntary | |
1147 | * preemption. | |
1148 | */ | |
1149 | if (rq->nr_running > 1) | |
541b8264 | 1150 | return false; |
ce831b38 | 1151 | |
541b8264 | 1152 | return true; |
ce831b38 FW |
1153 | } |
1154 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 1155 | #endif /* CONFIG_SMP */ |
18d95a28 | 1156 | |
a790de99 PT |
1157 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
1158 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 1159 | /* |
8277434e PT |
1160 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
1161 | * node and @up when leaving it for the final time. | |
1162 | * | |
1163 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 1164 | */ |
029632fb | 1165 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 1166 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1167 | { |
1168 | struct task_group *parent, *child; | |
eb755805 | 1169 | int ret; |
c09595f6 | 1170 | |
8277434e PT |
1171 | parent = from; |
1172 | ||
c09595f6 | 1173 | down: |
eb755805 PZ |
1174 | ret = (*down)(parent, data); |
1175 | if (ret) | |
8277434e | 1176 | goto out; |
c09595f6 PZ |
1177 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1178 | parent = child; | |
1179 | goto down; | |
1180 | ||
1181 | up: | |
1182 | continue; | |
1183 | } | |
eb755805 | 1184 | ret = (*up)(parent, data); |
8277434e PT |
1185 | if (ret || parent == from) |
1186 | goto out; | |
c09595f6 PZ |
1187 | |
1188 | child = parent; | |
1189 | parent = parent->parent; | |
1190 | if (parent) | |
1191 | goto up; | |
8277434e | 1192 | out: |
eb755805 | 1193 | return ret; |
c09595f6 PZ |
1194 | } |
1195 | ||
029632fb | 1196 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 1197 | { |
e2b245f8 | 1198 | return 0; |
eb755805 | 1199 | } |
18d95a28 PZ |
1200 | #endif |
1201 | ||
9059393e | 1202 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 1203 | { |
f05998d4 NR |
1204 | int prio = p->static_prio - MAX_RT_PRIO; |
1205 | struct load_weight *load = &p->se.load; | |
1206 | ||
dd41f596 IM |
1207 | /* |
1208 | * SCHED_IDLE tasks get minimal weight: | |
1209 | */ | |
1da1843f | 1210 | if (task_has_idle_policy(p)) { |
c8b28116 | 1211 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 1212 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
1213 | return; |
1214 | } | |
71f8bd46 | 1215 | |
9059393e VG |
1216 | /* |
1217 | * SCHED_OTHER tasks have to update their load when changing their | |
1218 | * weight | |
1219 | */ | |
1220 | if (update_load && p->sched_class == &fair_sched_class) { | |
1221 | reweight_task(p, prio); | |
1222 | } else { | |
1223 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
1224 | load->inv_weight = sched_prio_to_wmult[prio]; | |
1225 | } | |
71f8bd46 IM |
1226 | } |
1227 | ||
69842cba | 1228 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
1229 | /* |
1230 | * Serializes updates of utilization clamp values | |
1231 | * | |
1232 | * The (slow-path) user-space triggers utilization clamp value updates which | |
1233 | * can require updates on (fast-path) scheduler's data structures used to | |
1234 | * support enqueue/dequeue operations. | |
1235 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
1236 | * requests are serialized using a mutex to reduce the risk of conflicting | |
1237 | * updates or API abuses. | |
1238 | */ | |
1239 | static DEFINE_MUTEX(uclamp_mutex); | |
1240 | ||
e8f14172 PB |
1241 | /* Max allowed minimum utilization */ |
1242 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
1243 | ||
1244 | /* Max allowed maximum utilization */ | |
1245 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
1246 | ||
13685c4a QY |
1247 | /* |
1248 | * By default RT tasks run at the maximum performance point/capacity of the | |
1249 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
1250 | * SCHED_CAPACITY_SCALE. | |
1251 | * | |
1252 | * This knob allows admins to change the default behavior when uclamp is being | |
1253 | * used. In battery powered devices, particularly, running at the maximum | |
1254 | * capacity and frequency will increase energy consumption and shorten the | |
1255 | * battery life. | |
1256 | * | |
1257 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
1258 | * | |
1259 | * This knob will not override the system default sched_util_clamp_min defined | |
1260 | * above. | |
1261 | */ | |
1262 | unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; | |
1263 | ||
e8f14172 PB |
1264 | /* All clamps are required to be less or equal than these values */ |
1265 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 1266 | |
46609ce2 QY |
1267 | /* |
1268 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
1269 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
1270 | * enqueue/dequeue_task(). | |
1271 | * | |
1272 | * This allows users to continue to enable uclamp in their kernel config with | |
1273 | * minimum uclamp overhead in the fast path. | |
1274 | * | |
1275 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
1276 | * enabled, since we have an actual users that make use of uclamp | |
1277 | * functionality. | |
1278 | * | |
1279 | * The knobs that would enable this static key are: | |
1280 | * | |
1281 | * * A task modifying its uclamp value with sched_setattr(). | |
1282 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
1283 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
1284 | */ | |
1285 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
1286 | ||
69842cba PB |
1287 | /* Integer rounded range for each bucket */ |
1288 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
1289 | ||
1290 | #define for_each_clamp_id(clamp_id) \ | |
1291 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
1292 | ||
1293 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
1294 | { | |
6d2f8909 | 1295 | return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1); |
69842cba PB |
1296 | } |
1297 | ||
7763baac | 1298 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
1299 | { |
1300 | if (clamp_id == UCLAMP_MIN) | |
1301 | return 0; | |
1302 | return SCHED_CAPACITY_SCALE; | |
1303 | } | |
1304 | ||
a509a7cd PB |
1305 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
1306 | unsigned int value, bool user_defined) | |
69842cba PB |
1307 | { |
1308 | uc_se->value = value; | |
1309 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 1310 | uc_se->user_defined = user_defined; |
69842cba PB |
1311 | } |
1312 | ||
e496187d | 1313 | static inline unsigned int |
0413d7f3 | 1314 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
1315 | unsigned int clamp_value) |
1316 | { | |
1317 | /* | |
1318 | * Avoid blocked utilization pushing up the frequency when we go | |
1319 | * idle (which drops the max-clamp) by retaining the last known | |
1320 | * max-clamp. | |
1321 | */ | |
1322 | if (clamp_id == UCLAMP_MAX) { | |
1323 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
1324 | return clamp_value; | |
1325 | } | |
1326 | ||
1327 | return uclamp_none(UCLAMP_MIN); | |
1328 | } | |
1329 | ||
0413d7f3 | 1330 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
1331 | unsigned int clamp_value) |
1332 | { | |
1333 | /* Reset max-clamp retention only on idle exit */ | |
1334 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
1335 | return; | |
1336 | ||
1337 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
1338 | } | |
1339 | ||
69842cba | 1340 | static inline |
7763baac | 1341 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 1342 | unsigned int clamp_value) |
69842cba PB |
1343 | { |
1344 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
1345 | int bucket_id = UCLAMP_BUCKETS - 1; | |
1346 | ||
1347 | /* | |
1348 | * Since both min and max clamps are max aggregated, find the | |
1349 | * top most bucket with tasks in. | |
1350 | */ | |
1351 | for ( ; bucket_id >= 0; bucket_id--) { | |
1352 | if (!bucket[bucket_id].tasks) | |
1353 | continue; | |
1354 | return bucket[bucket_id].value; | |
1355 | } | |
1356 | ||
1357 | /* No tasks -- default clamp values */ | |
e496187d | 1358 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
1359 | } |
1360 | ||
13685c4a QY |
1361 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
1362 | { | |
1363 | unsigned int default_util_min; | |
1364 | struct uclamp_se *uc_se; | |
1365 | ||
1366 | lockdep_assert_held(&p->pi_lock); | |
1367 | ||
1368 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1369 | ||
1370 | /* Only sync if user didn't override the default */ | |
1371 | if (uc_se->user_defined) | |
1372 | return; | |
1373 | ||
1374 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1375 | uclamp_se_set(uc_se, default_util_min, false); | |
1376 | } | |
1377 | ||
1378 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1379 | { | |
1380 | struct rq_flags rf; | |
1381 | struct rq *rq; | |
1382 | ||
1383 | if (!rt_task(p)) | |
1384 | return; | |
1385 | ||
1386 | /* Protect updates to p->uclamp_* */ | |
1387 | rq = task_rq_lock(p, &rf); | |
1388 | __uclamp_update_util_min_rt_default(p); | |
1389 | task_rq_unlock(rq, p, &rf); | |
1390 | } | |
1391 | ||
1392 | static void uclamp_sync_util_min_rt_default(void) | |
1393 | { | |
1394 | struct task_struct *g, *p; | |
1395 | ||
1396 | /* | |
1397 | * copy_process() sysctl_uclamp | |
1398 | * uclamp_min_rt = X; | |
1399 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1400 | * // link thread smp_mb__after_spinlock() | |
1401 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1402 | * sched_post_fork() for_each_process_thread() | |
1403 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1404 | * | |
1405 | * Ensures that either sched_post_fork() will observe the new | |
1406 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1407 | * task. | |
1408 | */ | |
1409 | read_lock(&tasklist_lock); | |
1410 | smp_mb__after_spinlock(); | |
1411 | read_unlock(&tasklist_lock); | |
1412 | ||
1413 | rcu_read_lock(); | |
1414 | for_each_process_thread(g, p) | |
1415 | uclamp_update_util_min_rt_default(p); | |
1416 | rcu_read_unlock(); | |
1417 | } | |
1418 | ||
3eac870a | 1419 | static inline struct uclamp_se |
0413d7f3 | 1420 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a | 1421 | { |
0213b708 | 1422 | /* Copy by value as we could modify it */ |
3eac870a PB |
1423 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; |
1424 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0213b708 | 1425 | unsigned int tg_min, tg_max, value; |
3eac870a PB |
1426 | |
1427 | /* | |
1428 | * Tasks in autogroups or root task group will be | |
1429 | * restricted by system defaults. | |
1430 | */ | |
1431 | if (task_group_is_autogroup(task_group(p))) | |
1432 | return uc_req; | |
1433 | if (task_group(p) == &root_task_group) | |
1434 | return uc_req; | |
1435 | ||
0213b708 QY |
1436 | tg_min = task_group(p)->uclamp[UCLAMP_MIN].value; |
1437 | tg_max = task_group(p)->uclamp[UCLAMP_MAX].value; | |
1438 | value = uc_req.value; | |
1439 | value = clamp(value, tg_min, tg_max); | |
1440 | uclamp_se_set(&uc_req, value, false); | |
3eac870a PB |
1441 | #endif |
1442 | ||
1443 | return uc_req; | |
1444 | } | |
1445 | ||
e8f14172 PB |
1446 | /* |
1447 | * The effective clamp bucket index of a task depends on, by increasing | |
1448 | * priority: | |
1449 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1450 | * - the task group effective clamp value, for tasks not either in the root |
1451 | * group or in an autogroup | |
e8f14172 PB |
1452 | * - the system default clamp value, defined by the sysadmin |
1453 | */ | |
1454 | static inline struct uclamp_se | |
0413d7f3 | 1455 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1456 | { |
3eac870a | 1457 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1458 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1459 | ||
1460 | /* System default restrictions always apply */ | |
1461 | if (unlikely(uc_req.value > uc_max.value)) | |
1462 | return uc_max; | |
1463 | ||
1464 | return uc_req; | |
1465 | } | |
1466 | ||
686516b5 | 1467 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1468 | { |
1469 | struct uclamp_se uc_eff; | |
1470 | ||
1471 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1472 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1473 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1474 | |
1475 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1476 | ||
686516b5 | 1477 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1478 | } |
1479 | ||
69842cba PB |
1480 | /* |
1481 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1482 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1483 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1484 | * |
1485 | * Tasks can have a task-specific value requested from user-space, track | |
1486 | * within each bucket the maximum value for tasks refcounted in it. | |
1487 | * This "local max aggregation" allows to track the exact "requested" value | |
1488 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1489 | */ |
1490 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1491 | enum uclamp_id clamp_id) |
69842cba PB |
1492 | { |
1493 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1494 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1495 | struct uclamp_bucket *bucket; | |
1496 | ||
5cb9eaa3 | 1497 | lockdep_assert_rq_held(rq); |
69842cba | 1498 | |
e8f14172 PB |
1499 | /* Update task effective clamp */ |
1500 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1501 | ||
69842cba PB |
1502 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1503 | bucket->tasks++; | |
e8f14172 | 1504 | uc_se->active = true; |
69842cba | 1505 | |
e496187d PB |
1506 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1507 | ||
60daf9c1 PB |
1508 | /* |
1509 | * Local max aggregation: rq buckets always track the max | |
1510 | * "requested" clamp value of its RUNNABLE tasks. | |
1511 | */ | |
1512 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1513 | bucket->value = uc_se->value; | |
1514 | ||
69842cba | 1515 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1516 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1517 | } |
1518 | ||
1519 | /* | |
1520 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1521 | * is released. If this is the last task reference counting the rq's max | |
1522 | * active clamp value, then the rq's clamp value is updated. | |
1523 | * | |
1524 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1525 | * always valid. If it's detected they are not, as defensive programming, | |
1526 | * enforce the expected state and warn. | |
1527 | */ | |
1528 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1529 | enum uclamp_id clamp_id) |
69842cba PB |
1530 | { |
1531 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1532 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1533 | struct uclamp_bucket *bucket; | |
e496187d | 1534 | unsigned int bkt_clamp; |
69842cba PB |
1535 | unsigned int rq_clamp; |
1536 | ||
5cb9eaa3 | 1537 | lockdep_assert_rq_held(rq); |
69842cba | 1538 | |
46609ce2 QY |
1539 | /* |
1540 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1541 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1542 | * | |
1543 | * In this case the uc_se->active flag should be false since no uclamp | |
1544 | * accounting was performed at enqueue time and we can just return | |
1545 | * here. | |
1546 | * | |
b19a888c | 1547 | * Need to be careful of the following enqueue/dequeue ordering |
46609ce2 QY |
1548 | * problem too |
1549 | * | |
1550 | * enqueue(taskA) | |
1551 | * // sched_uclamp_used gets enabled | |
1552 | * enqueue(taskB) | |
1553 | * dequeue(taskA) | |
b19a888c | 1554 | * // Must not decrement bucket->tasks here |
46609ce2 QY |
1555 | * dequeue(taskB) |
1556 | * | |
1557 | * where we could end up with stale data in uc_se and | |
1558 | * bucket[uc_se->bucket_id]. | |
1559 | * | |
1560 | * The following check here eliminates the possibility of such race. | |
1561 | */ | |
1562 | if (unlikely(!uc_se->active)) | |
1563 | return; | |
1564 | ||
69842cba | 1565 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1566 | |
69842cba PB |
1567 | SCHED_WARN_ON(!bucket->tasks); |
1568 | if (likely(bucket->tasks)) | |
1569 | bucket->tasks--; | |
46609ce2 | 1570 | |
e8f14172 | 1571 | uc_se->active = false; |
69842cba | 1572 | |
60daf9c1 PB |
1573 | /* |
1574 | * Keep "local max aggregation" simple and accept to (possibly) | |
1575 | * overboost some RUNNABLE tasks in the same bucket. | |
1576 | * The rq clamp bucket value is reset to its base value whenever | |
1577 | * there are no more RUNNABLE tasks refcounting it. | |
1578 | */ | |
69842cba PB |
1579 | if (likely(bucket->tasks)) |
1580 | return; | |
1581 | ||
1582 | rq_clamp = READ_ONCE(uc_rq->value); | |
1583 | /* | |
1584 | * Defensive programming: this should never happen. If it happens, | |
1585 | * e.g. due to future modification, warn and fixup the expected value. | |
1586 | */ | |
1587 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1588 | if (bucket->value >= rq_clamp) { |
1589 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1590 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1591 | } | |
69842cba PB |
1592 | } |
1593 | ||
1594 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1595 | { | |
0413d7f3 | 1596 | enum uclamp_id clamp_id; |
69842cba | 1597 | |
46609ce2 QY |
1598 | /* |
1599 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1600 | * | |
1601 | * The condition is constructed such that a NOP is generated when | |
1602 | * sched_uclamp_used is disabled. | |
1603 | */ | |
1604 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1605 | return; | |
1606 | ||
69842cba PB |
1607 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1608 | return; | |
1609 | ||
1610 | for_each_clamp_id(clamp_id) | |
1611 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1612 | |
1613 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1614 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1615 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1616 | } |
1617 | ||
1618 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1619 | { | |
0413d7f3 | 1620 | enum uclamp_id clamp_id; |
69842cba | 1621 | |
46609ce2 QY |
1622 | /* |
1623 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1624 | * | |
1625 | * The condition is constructed such that a NOP is generated when | |
1626 | * sched_uclamp_used is disabled. | |
1627 | */ | |
1628 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1629 | return; | |
1630 | ||
69842cba PB |
1631 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1632 | return; | |
1633 | ||
1634 | for_each_clamp_id(clamp_id) | |
1635 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1636 | } | |
1637 | ||
ca4984a7 QP |
1638 | static inline void uclamp_rq_reinc_id(struct rq *rq, struct task_struct *p, |
1639 | enum uclamp_id clamp_id) | |
1640 | { | |
1641 | if (!p->uclamp[clamp_id].active) | |
1642 | return; | |
1643 | ||
1644 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1645 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1646 | ||
1647 | /* | |
1648 | * Make sure to clear the idle flag if we've transiently reached 0 | |
1649 | * active tasks on rq. | |
1650 | */ | |
1651 | if (clamp_id == UCLAMP_MAX && (rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
1652 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
1653 | } | |
1654 | ||
babbe170 | 1655 | static inline void |
0213b708 | 1656 | uclamp_update_active(struct task_struct *p) |
babbe170 | 1657 | { |
0213b708 | 1658 | enum uclamp_id clamp_id; |
babbe170 PB |
1659 | struct rq_flags rf; |
1660 | struct rq *rq; | |
1661 | ||
1662 | /* | |
1663 | * Lock the task and the rq where the task is (or was) queued. | |
1664 | * | |
1665 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1666 | * price to pay to safely serialize util_{min,max} updates with | |
1667 | * enqueues, dequeues and migration operations. | |
1668 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1669 | */ | |
1670 | rq = task_rq_lock(p, &rf); | |
1671 | ||
1672 | /* | |
1673 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1674 | * If the task is not yet RUNNABLE and its task_struct is not | |
1675 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1676 | * it will already see the updated clamp bucket value. | |
1677 | */ | |
ca4984a7 QP |
1678 | for_each_clamp_id(clamp_id) |
1679 | uclamp_rq_reinc_id(rq, p, clamp_id); | |
babbe170 PB |
1680 | |
1681 | task_rq_unlock(rq, p, &rf); | |
1682 | } | |
1683 | ||
e3b8b6a0 | 1684 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 | 1685 | static inline void |
0213b708 | 1686 | uclamp_update_active_tasks(struct cgroup_subsys_state *css) |
babbe170 PB |
1687 | { |
1688 | struct css_task_iter it; | |
1689 | struct task_struct *p; | |
babbe170 PB |
1690 | |
1691 | css_task_iter_start(css, 0, &it); | |
0213b708 QY |
1692 | while ((p = css_task_iter_next(&it))) |
1693 | uclamp_update_active(p); | |
babbe170 PB |
1694 | css_task_iter_end(&it); |
1695 | } | |
1696 | ||
7274a5c1 PB |
1697 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1698 | static void uclamp_update_root_tg(void) | |
1699 | { | |
1700 | struct task_group *tg = &root_task_group; | |
1701 | ||
1702 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1703 | sysctl_sched_uclamp_util_min, false); | |
1704 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1705 | sysctl_sched_uclamp_util_max, false); | |
1706 | ||
1707 | rcu_read_lock(); | |
1708 | cpu_util_update_eff(&root_task_group.css); | |
1709 | rcu_read_unlock(); | |
1710 | } | |
1711 | #else | |
1712 | static void uclamp_update_root_tg(void) { } | |
1713 | #endif | |
1714 | ||
e8f14172 | 1715 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1716 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1717 | { |
7274a5c1 | 1718 | bool update_root_tg = false; |
13685c4a | 1719 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1720 | int result; |
1721 | ||
2480c093 | 1722 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1723 | old_min = sysctl_sched_uclamp_util_min; |
1724 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1725 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1726 | |
1727 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1728 | if (result) | |
1729 | goto undo; | |
1730 | if (!write) | |
1731 | goto done; | |
1732 | ||
1733 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1734 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1735 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1736 | ||
e8f14172 PB |
1737 | result = -EINVAL; |
1738 | goto undo; | |
1739 | } | |
1740 | ||
1741 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1742 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1743 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1744 | update_root_tg = true; |
e8f14172 PB |
1745 | } |
1746 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1747 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1748 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1749 | update_root_tg = true; |
e8f14172 PB |
1750 | } |
1751 | ||
46609ce2 QY |
1752 | if (update_root_tg) { |
1753 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1754 | uclamp_update_root_tg(); |
46609ce2 | 1755 | } |
7274a5c1 | 1756 | |
13685c4a QY |
1757 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1758 | static_branch_enable(&sched_uclamp_used); | |
1759 | uclamp_sync_util_min_rt_default(); | |
1760 | } | |
7274a5c1 | 1761 | |
e8f14172 | 1762 | /* |
7274a5c1 PB |
1763 | * We update all RUNNABLE tasks only when task groups are in use. |
1764 | * Otherwise, keep it simple and do just a lazy update at each next | |
1765 | * task enqueue time. | |
e8f14172 | 1766 | */ |
7274a5c1 | 1767 | |
e8f14172 PB |
1768 | goto done; |
1769 | ||
1770 | undo: | |
1771 | sysctl_sched_uclamp_util_min = old_min; | |
1772 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1773 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1774 | done: |
2480c093 | 1775 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1776 | |
1777 | return result; | |
1778 | } | |
1779 | ||
a509a7cd PB |
1780 | static int uclamp_validate(struct task_struct *p, |
1781 | const struct sched_attr *attr) | |
1782 | { | |
480a6ca2 DE |
1783 | int util_min = p->uclamp_req[UCLAMP_MIN].value; |
1784 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd | 1785 | |
480a6ca2 DE |
1786 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { |
1787 | util_min = attr->sched_util_min; | |
a509a7cd | 1788 | |
480a6ca2 DE |
1789 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) |
1790 | return -EINVAL; | |
1791 | } | |
1792 | ||
1793 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1794 | util_max = attr->sched_util_max; | |
1795 | ||
1796 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | |
1797 | return -EINVAL; | |
1798 | } | |
1799 | ||
1800 | if (util_min != -1 && util_max != -1 && util_min > util_max) | |
a509a7cd PB |
1801 | return -EINVAL; |
1802 | ||
e65855a5 QY |
1803 | /* |
1804 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1805 | * | |
1806 | * We need to do that here, because enabling static branches is a | |
1807 | * blocking operation which obviously cannot be done while holding | |
1808 | * scheduler locks. | |
1809 | */ | |
1810 | static_branch_enable(&sched_uclamp_used); | |
1811 | ||
a509a7cd PB |
1812 | return 0; |
1813 | } | |
1814 | ||
480a6ca2 DE |
1815 | static bool uclamp_reset(const struct sched_attr *attr, |
1816 | enum uclamp_id clamp_id, | |
1817 | struct uclamp_se *uc_se) | |
1818 | { | |
1819 | /* Reset on sched class change for a non user-defined clamp value. */ | |
1820 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | |
1821 | !uc_se->user_defined) | |
1822 | return true; | |
1823 | ||
1824 | /* Reset on sched_util_{min,max} == -1. */ | |
1825 | if (clamp_id == UCLAMP_MIN && | |
1826 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | |
1827 | attr->sched_util_min == -1) { | |
1828 | return true; | |
1829 | } | |
1830 | ||
1831 | if (clamp_id == UCLAMP_MAX && | |
1832 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | |
1833 | attr->sched_util_max == -1) { | |
1834 | return true; | |
1835 | } | |
1836 | ||
1837 | return false; | |
1838 | } | |
1839 | ||
a509a7cd PB |
1840 | static void __setscheduler_uclamp(struct task_struct *p, |
1841 | const struct sched_attr *attr) | |
1842 | { | |
0413d7f3 | 1843 | enum uclamp_id clamp_id; |
1a00d999 | 1844 | |
1a00d999 PB |
1845 | for_each_clamp_id(clamp_id) { |
1846 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
480a6ca2 | 1847 | unsigned int value; |
1a00d999 | 1848 | |
480a6ca2 | 1849 | if (!uclamp_reset(attr, clamp_id, uc_se)) |
1a00d999 PB |
1850 | continue; |
1851 | ||
13685c4a QY |
1852 | /* |
1853 | * RT by default have a 100% boost value that could be modified | |
1854 | * at runtime. | |
1855 | */ | |
1a00d999 | 1856 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
480a6ca2 | 1857 | value = sysctl_sched_uclamp_util_min_rt_default; |
13685c4a | 1858 | else |
480a6ca2 DE |
1859 | value = uclamp_none(clamp_id); |
1860 | ||
1861 | uclamp_se_set(uc_se, value, false); | |
1a00d999 | 1862 | |
1a00d999 PB |
1863 | } |
1864 | ||
a509a7cd PB |
1865 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1866 | return; | |
1867 | ||
480a6ca2 DE |
1868 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && |
1869 | attr->sched_util_min != -1) { | |
a509a7cd PB |
1870 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], |
1871 | attr->sched_util_min, true); | |
1872 | } | |
1873 | ||
480a6ca2 DE |
1874 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && |
1875 | attr->sched_util_max != -1) { | |
a509a7cd PB |
1876 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], |
1877 | attr->sched_util_max, true); | |
1878 | } | |
1879 | } | |
1880 | ||
e8f14172 PB |
1881 | static void uclamp_fork(struct task_struct *p) |
1882 | { | |
0413d7f3 | 1883 | enum uclamp_id clamp_id; |
e8f14172 | 1884 | |
13685c4a QY |
1885 | /* |
1886 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1887 | * as the task is still at its early fork stages. | |
1888 | */ | |
e8f14172 PB |
1889 | for_each_clamp_id(clamp_id) |
1890 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1891 | |
1892 | if (likely(!p->sched_reset_on_fork)) | |
1893 | return; | |
1894 | ||
1895 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1896 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1897 | uclamp_none(clamp_id), false); | |
a87498ac | 1898 | } |
e8f14172 PB |
1899 | } |
1900 | ||
13685c4a QY |
1901 | static void uclamp_post_fork(struct task_struct *p) |
1902 | { | |
1903 | uclamp_update_util_min_rt_default(p); | |
1904 | } | |
1905 | ||
d81ae8aa QY |
1906 | static void __init init_uclamp_rq(struct rq *rq) |
1907 | { | |
1908 | enum uclamp_id clamp_id; | |
1909 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1910 | ||
1911 | for_each_clamp_id(clamp_id) { | |
1912 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1913 | .value = uclamp_none(clamp_id) | |
1914 | }; | |
1915 | } | |
1916 | ||
1917 | rq->uclamp_flags = 0; | |
1918 | } | |
1919 | ||
69842cba PB |
1920 | static void __init init_uclamp(void) |
1921 | { | |
e8f14172 | 1922 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1923 | enum uclamp_id clamp_id; |
69842cba PB |
1924 | int cpu; |
1925 | ||
d81ae8aa QY |
1926 | for_each_possible_cpu(cpu) |
1927 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1928 | |
69842cba | 1929 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1930 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1931 | uclamp_none(clamp_id), false); |
69842cba | 1932 | } |
e8f14172 PB |
1933 | |
1934 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1935 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1936 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1937 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1938 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1939 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1940 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1941 | #endif |
1942 | } | |
69842cba PB |
1943 | } |
1944 | ||
1945 | #else /* CONFIG_UCLAMP_TASK */ | |
1946 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1947 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1948 | static inline int uclamp_validate(struct task_struct *p, |
1949 | const struct sched_attr *attr) | |
1950 | { | |
1951 | return -EOPNOTSUPP; | |
1952 | } | |
1953 | static void __setscheduler_uclamp(struct task_struct *p, | |
1954 | const struct sched_attr *attr) { } | |
e8f14172 | 1955 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1956 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1957 | static inline void init_uclamp(void) { } |
1958 | #endif /* CONFIG_UCLAMP_TASK */ | |
1959 | ||
a1dfb631 MT |
1960 | bool sched_task_on_rq(struct task_struct *p) |
1961 | { | |
1962 | return task_on_rq_queued(p); | |
1963 | } | |
1964 | ||
1de64443 | 1965 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1966 | { |
0a67d1ee PZ |
1967 | if (!(flags & ENQUEUE_NOCLOCK)) |
1968 | update_rq_clock(rq); | |
1969 | ||
eb414681 | 1970 | if (!(flags & ENQUEUE_RESTORE)) { |
4e29fb70 | 1971 | sched_info_enqueue(rq, p); |
eb414681 JW |
1972 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1973 | } | |
0a67d1ee | 1974 | |
69842cba | 1975 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1976 | p->sched_class->enqueue_task(rq, p, flags); |
8a311c74 PZ |
1977 | |
1978 | if (sched_core_enabled(rq)) | |
1979 | sched_core_enqueue(rq, p); | |
71f8bd46 IM |
1980 | } |
1981 | ||
1de64443 | 1982 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1983 | { |
8a311c74 PZ |
1984 | if (sched_core_enabled(rq)) |
1985 | sched_core_dequeue(rq, p); | |
1986 | ||
0a67d1ee PZ |
1987 | if (!(flags & DEQUEUE_NOCLOCK)) |
1988 | update_rq_clock(rq); | |
1989 | ||
eb414681 | 1990 | if (!(flags & DEQUEUE_SAVE)) { |
4e29fb70 | 1991 | sched_info_dequeue(rq, p); |
eb414681 JW |
1992 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
1993 | } | |
0a67d1ee | 1994 | |
69842cba | 1995 | uclamp_rq_dec(rq, p); |
371fd7e7 | 1996 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1997 | } |
1998 | ||
029632fb | 1999 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 2000 | { |
371fd7e7 | 2001 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
2002 | |
2003 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
2004 | } |
2005 | ||
029632fb | 2006 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 2007 | { |
7dd77884 PZ |
2008 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
2009 | ||
371fd7e7 | 2010 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
2011 | } |
2012 | ||
f558c2b8 | 2013 | static inline int __normal_prio(int policy, int rt_prio, int nice) |
14531189 | 2014 | { |
f558c2b8 PZ |
2015 | int prio; |
2016 | ||
2017 | if (dl_policy(policy)) | |
2018 | prio = MAX_DL_PRIO - 1; | |
2019 | else if (rt_policy(policy)) | |
2020 | prio = MAX_RT_PRIO - 1 - rt_prio; | |
2021 | else | |
2022 | prio = NICE_TO_PRIO(nice); | |
2023 | ||
2024 | return prio; | |
14531189 IM |
2025 | } |
2026 | ||
b29739f9 IM |
2027 | /* |
2028 | * Calculate the expected normal priority: i.e. priority | |
2029 | * without taking RT-inheritance into account. Might be | |
2030 | * boosted by interactivity modifiers. Changes upon fork, | |
2031 | * setprio syscalls, and whenever the interactivity | |
2032 | * estimator recalculates. | |
2033 | */ | |
36c8b586 | 2034 | static inline int normal_prio(struct task_struct *p) |
b29739f9 | 2035 | { |
f558c2b8 | 2036 | return __normal_prio(p->policy, p->rt_priority, PRIO_TO_NICE(p->static_prio)); |
b29739f9 IM |
2037 | } |
2038 | ||
2039 | /* | |
2040 | * Calculate the current priority, i.e. the priority | |
2041 | * taken into account by the scheduler. This value might | |
2042 | * be boosted by RT tasks, or might be boosted by | |
2043 | * interactivity modifiers. Will be RT if the task got | |
2044 | * RT-boosted. If not then it returns p->normal_prio. | |
2045 | */ | |
36c8b586 | 2046 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2047 | { |
2048 | p->normal_prio = normal_prio(p); | |
2049 | /* | |
2050 | * If we are RT tasks or we were boosted to RT priority, | |
2051 | * keep the priority unchanged. Otherwise, update priority | |
2052 | * to the normal priority: | |
2053 | */ | |
2054 | if (!rt_prio(p->prio)) | |
2055 | return p->normal_prio; | |
2056 | return p->prio; | |
2057 | } | |
2058 | ||
1da177e4 LT |
2059 | /** |
2060 | * task_curr - is this task currently executing on a CPU? | |
2061 | * @p: the task in question. | |
e69f6186 YB |
2062 | * |
2063 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 2064 | */ |
36c8b586 | 2065 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2066 | { |
2067 | return cpu_curr(task_cpu(p)) == p; | |
2068 | } | |
2069 | ||
67dfa1b7 | 2070 | /* |
4c9a4bc8 PZ |
2071 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
2072 | * use the balance_callback list if you want balancing. | |
2073 | * | |
2074 | * this means any call to check_class_changed() must be followed by a call to | |
2075 | * balance_callback(). | |
67dfa1b7 | 2076 | */ |
cb469845 SR |
2077 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2078 | const struct sched_class *prev_class, | |
da7a735e | 2079 | int oldprio) |
cb469845 SR |
2080 | { |
2081 | if (prev_class != p->sched_class) { | |
2082 | if (prev_class->switched_from) | |
da7a735e | 2083 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 2084 | |
da7a735e | 2085 | p->sched_class->switched_to(rq, p); |
2d3d891d | 2086 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 2087 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
2088 | } |
2089 | ||
029632fb | 2090 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 2091 | { |
aa93cd53 | 2092 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 2093 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
2094 | else if (p->sched_class > rq->curr->sched_class) |
2095 | resched_curr(rq); | |
1e5a7405 PZ |
2096 | |
2097 | /* | |
2098 | * A queue event has occurred, and we're going to schedule. In | |
2099 | * this case, we can save a useless back to back clock update. | |
2100 | */ | |
da0c1e65 | 2101 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 2102 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
2103 | } |
2104 | ||
1da177e4 | 2105 | #ifdef CONFIG_SMP |
175f0e25 | 2106 | |
af449901 PZ |
2107 | static void |
2108 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags); | |
2109 | ||
2110 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
2111 | const struct cpumask *new_mask, | |
2112 | u32 flags); | |
2113 | ||
2114 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
2115 | { | |
2116 | if (likely(!p->migration_disabled)) | |
2117 | return; | |
2118 | ||
2119 | if (p->cpus_ptr != &p->cpus_mask) | |
2120 | return; | |
2121 | ||
2122 | /* | |
2123 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
2124 | */ | |
2125 | __do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE); | |
2126 | } | |
2127 | ||
2128 | void migrate_disable(void) | |
2129 | { | |
3015ef4b TG |
2130 | struct task_struct *p = current; |
2131 | ||
2132 | if (p->migration_disabled) { | |
2133 | p->migration_disabled++; | |
af449901 | 2134 | return; |
3015ef4b | 2135 | } |
af449901 | 2136 | |
3015ef4b TG |
2137 | preempt_disable(); |
2138 | this_rq()->nr_pinned++; | |
2139 | p->migration_disabled = 1; | |
2140 | preempt_enable(); | |
af449901 PZ |
2141 | } |
2142 | EXPORT_SYMBOL_GPL(migrate_disable); | |
2143 | ||
2144 | void migrate_enable(void) | |
2145 | { | |
2146 | struct task_struct *p = current; | |
2147 | ||
6d337eab PZ |
2148 | if (p->migration_disabled > 1) { |
2149 | p->migration_disabled--; | |
af449901 | 2150 | return; |
6d337eab | 2151 | } |
af449901 | 2152 | |
6d337eab PZ |
2153 | /* |
2154 | * Ensure stop_task runs either before or after this, and that | |
2155 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
2156 | */ | |
2157 | preempt_disable(); | |
2158 | if (p->cpus_ptr != &p->cpus_mask) | |
2159 | __set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE); | |
2160 | /* | |
2161 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
2162 | * regular cpus_mask, otherwise things that race (eg. | |
2163 | * select_fallback_rq) get confused. | |
2164 | */ | |
af449901 | 2165 | barrier(); |
6d337eab | 2166 | p->migration_disabled = 0; |
3015ef4b | 2167 | this_rq()->nr_pinned--; |
6d337eab | 2168 | preempt_enable(); |
af449901 PZ |
2169 | } |
2170 | EXPORT_SYMBOL_GPL(migrate_enable); | |
2171 | ||
3015ef4b TG |
2172 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2173 | { | |
2174 | return rq->nr_pinned; | |
2175 | } | |
2176 | ||
175f0e25 | 2177 | /* |
bee98539 | 2178 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
2179 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
2180 | */ | |
2181 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
2182 | { | |
5ba2ffba | 2183 | /* When not in the task's cpumask, no point in looking further. */ |
3bd37062 | 2184 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
2185 | return false; |
2186 | ||
5ba2ffba PZ |
2187 | /* migrate_disabled() must be allowed to finish. */ |
2188 | if (is_migration_disabled(p)) | |
175f0e25 PZ |
2189 | return cpu_online(cpu); |
2190 | ||
5ba2ffba PZ |
2191 | /* Non kernel threads are not allowed during either online or offline. */ |
2192 | if (!(p->flags & PF_KTHREAD)) | |
9ae606bc | 2193 | return cpu_active(cpu) && task_cpu_possible(cpu, p); |
5ba2ffba PZ |
2194 | |
2195 | /* KTHREAD_IS_PER_CPU is always allowed. */ | |
2196 | if (kthread_is_per_cpu(p)) | |
2197 | return cpu_online(cpu); | |
2198 | ||
2199 | /* Regular kernel threads don't get to stay during offline. */ | |
b5c44773 | 2200 | if (cpu_dying(cpu)) |
5ba2ffba PZ |
2201 | return false; |
2202 | ||
2203 | /* But are allowed during online. */ | |
2204 | return cpu_online(cpu); | |
175f0e25 PZ |
2205 | } |
2206 | ||
5cc389bc PZ |
2207 | /* |
2208 | * This is how migration works: | |
2209 | * | |
2210 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
2211 | * stop_one_cpu(). | |
2212 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
2213 | * off the CPU) | |
2214 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
2215 | * 4) if it's in the wrong runqueue then the migration thread removes | |
2216 | * it and puts it into the right queue. | |
2217 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
2218 | * is done. | |
2219 | */ | |
2220 | ||
2221 | /* | |
2222 | * move_queued_task - move a queued task to new rq. | |
2223 | * | |
2224 | * Returns (locked) new rq. Old rq's lock is released. | |
2225 | */ | |
8a8c69c3 PZ |
2226 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
2227 | struct task_struct *p, int new_cpu) | |
5cc389bc | 2228 | { |
5cb9eaa3 | 2229 | lockdep_assert_rq_held(rq); |
5cc389bc | 2230 | |
58877d34 | 2231 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 2232 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 2233 | rq_unlock(rq, rf); |
5cc389bc PZ |
2234 | |
2235 | rq = cpu_rq(new_cpu); | |
2236 | ||
8a8c69c3 | 2237 | rq_lock(rq, rf); |
5cc389bc | 2238 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 2239 | activate_task(rq, p, 0); |
5cc389bc PZ |
2240 | check_preempt_curr(rq, p, 0); |
2241 | ||
2242 | return rq; | |
2243 | } | |
2244 | ||
2245 | struct migration_arg { | |
6d337eab PZ |
2246 | struct task_struct *task; |
2247 | int dest_cpu; | |
2248 | struct set_affinity_pending *pending; | |
2249 | }; | |
2250 | ||
50caf9c1 PZ |
2251 | /* |
2252 | * @refs: number of wait_for_completion() | |
2253 | * @stop_pending: is @stop_work in use | |
2254 | */ | |
6d337eab PZ |
2255 | struct set_affinity_pending { |
2256 | refcount_t refs; | |
9e81889c | 2257 | unsigned int stop_pending; |
6d337eab PZ |
2258 | struct completion done; |
2259 | struct cpu_stop_work stop_work; | |
2260 | struct migration_arg arg; | |
5cc389bc PZ |
2261 | }; |
2262 | ||
2263 | /* | |
d1ccc66d | 2264 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
2265 | * this because either it can't run here any more (set_cpus_allowed() |
2266 | * away from this CPU, or CPU going down), or because we're | |
2267 | * attempting to rebalance this task on exec (sched_exec). | |
2268 | * | |
2269 | * So we race with normal scheduler movements, but that's OK, as long | |
2270 | * as the task is no longer on this CPU. | |
5cc389bc | 2271 | */ |
8a8c69c3 PZ |
2272 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
2273 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 2274 | { |
5cc389bc | 2275 | /* Affinity changed (again). */ |
175f0e25 | 2276 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 2277 | return rq; |
5cc389bc | 2278 | |
15ff991e | 2279 | update_rq_clock(rq); |
8a8c69c3 | 2280 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
2281 | |
2282 | return rq; | |
5cc389bc PZ |
2283 | } |
2284 | ||
2285 | /* | |
2286 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
2287 | * and performs thread migration by bumping thread off CPU then | |
2288 | * 'pushing' onto another runqueue. | |
2289 | */ | |
2290 | static int migration_cpu_stop(void *data) | |
2291 | { | |
2292 | struct migration_arg *arg = data; | |
c20cf065 | 2293 | struct set_affinity_pending *pending = arg->pending; |
5e16bbc2 PZ |
2294 | struct task_struct *p = arg->task; |
2295 | struct rq *rq = this_rq(); | |
6d337eab | 2296 | bool complete = false; |
8a8c69c3 | 2297 | struct rq_flags rf; |
5cc389bc PZ |
2298 | |
2299 | /* | |
d1ccc66d IM |
2300 | * The original target CPU might have gone down and we might |
2301 | * be on another CPU but it doesn't matter. | |
5cc389bc | 2302 | */ |
6d337eab | 2303 | local_irq_save(rf.flags); |
5cc389bc PZ |
2304 | /* |
2305 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 2306 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
2307 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
2308 | */ | |
a1488664 | 2309 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
2310 | |
2311 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 2312 | rq_lock(rq, &rf); |
6d337eab | 2313 | |
e140749c VS |
2314 | /* |
2315 | * If we were passed a pending, then ->stop_pending was set, thus | |
2316 | * p->migration_pending must have remained stable. | |
2317 | */ | |
2318 | WARN_ON_ONCE(pending && pending != p->migration_pending); | |
2319 | ||
5e16bbc2 PZ |
2320 | /* |
2321 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
2322 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
2323 | * we're holding p->pi_lock. | |
2324 | */ | |
bf89a304 | 2325 | if (task_rq(p) == rq) { |
6d337eab PZ |
2326 | if (is_migration_disabled(p)) |
2327 | goto out; | |
2328 | ||
2329 | if (pending) { | |
e140749c | 2330 | p->migration_pending = NULL; |
6d337eab | 2331 | complete = true; |
6d337eab | 2332 | |
3f1bc119 PZ |
2333 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) |
2334 | goto out; | |
3f1bc119 | 2335 | } |
6d337eab | 2336 | |
bf89a304 | 2337 | if (task_on_rq_queued(p)) |
475ea6c6 | 2338 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 | 2339 | else |
475ea6c6 | 2340 | p->wake_cpu = arg->dest_cpu; |
6d337eab | 2341 | |
3f1bc119 PZ |
2342 | /* |
2343 | * XXX __migrate_task() can fail, at which point we might end | |
2344 | * up running on a dodgy CPU, AFAICT this can only happen | |
2345 | * during CPU hotplug, at which point we'll get pushed out | |
2346 | * anyway, so it's probably not a big deal. | |
2347 | */ | |
2348 | ||
c20cf065 | 2349 | } else if (pending) { |
6d337eab PZ |
2350 | /* |
2351 | * This happens when we get migrated between migrate_enable()'s | |
2352 | * preempt_enable() and scheduling the stopper task. At that | |
2353 | * point we're a regular task again and not current anymore. | |
2354 | * | |
2355 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
2356 | * more likely. | |
2357 | */ | |
2358 | ||
d707faa6 VS |
2359 | /* |
2360 | * The task moved before the stopper got to run. We're holding | |
2361 | * ->pi_lock, so the allowed mask is stable - if it got | |
2362 | * somewhere allowed, we're done. | |
2363 | */ | |
c20cf065 | 2364 | if (cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { |
e140749c | 2365 | p->migration_pending = NULL; |
d707faa6 VS |
2366 | complete = true; |
2367 | goto out; | |
2368 | } | |
2369 | ||
6d337eab PZ |
2370 | /* |
2371 | * When migrate_enable() hits a rq mis-match we can't reliably | |
2372 | * determine is_migration_disabled() and so have to chase after | |
2373 | * it. | |
2374 | */ | |
9e81889c | 2375 | WARN_ON_ONCE(!pending->stop_pending); |
6d337eab PZ |
2376 | task_rq_unlock(rq, p, &rf); |
2377 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
2378 | &pending->arg, &pending->stop_work); | |
2379 | return 0; | |
bf89a304 | 2380 | } |
6d337eab | 2381 | out: |
9e81889c PZ |
2382 | if (pending) |
2383 | pending->stop_pending = false; | |
6d337eab PZ |
2384 | task_rq_unlock(rq, p, &rf); |
2385 | ||
2386 | if (complete) | |
2387 | complete_all(&pending->done); | |
2388 | ||
5cc389bc PZ |
2389 | return 0; |
2390 | } | |
2391 | ||
a7c81556 PZ |
2392 | int push_cpu_stop(void *arg) |
2393 | { | |
2394 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2395 | struct task_struct *p = arg; | |
2396 | ||
2397 | raw_spin_lock_irq(&p->pi_lock); | |
5cb9eaa3 | 2398 | raw_spin_rq_lock(rq); |
a7c81556 PZ |
2399 | |
2400 | if (task_rq(p) != rq) | |
2401 | goto out_unlock; | |
2402 | ||
2403 | if (is_migration_disabled(p)) { | |
2404 | p->migration_flags |= MDF_PUSH; | |
2405 | goto out_unlock; | |
2406 | } | |
2407 | ||
2408 | p->migration_flags &= ~MDF_PUSH; | |
2409 | ||
2410 | if (p->sched_class->find_lock_rq) | |
2411 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2412 | |
a7c81556 PZ |
2413 | if (!lowest_rq) |
2414 | goto out_unlock; | |
2415 | ||
2416 | // XXX validate p is still the highest prio task | |
2417 | if (task_rq(p) == rq) { | |
2418 | deactivate_task(rq, p, 0); | |
2419 | set_task_cpu(p, lowest_rq->cpu); | |
2420 | activate_task(lowest_rq, p, 0); | |
2421 | resched_curr(lowest_rq); | |
2422 | } | |
2423 | ||
2424 | double_unlock_balance(rq, lowest_rq); | |
2425 | ||
2426 | out_unlock: | |
2427 | rq->push_busy = false; | |
5cb9eaa3 | 2428 | raw_spin_rq_unlock(rq); |
a7c81556 PZ |
2429 | raw_spin_unlock_irq(&p->pi_lock); |
2430 | ||
2431 | put_task_struct(p); | |
5cc389bc PZ |
2432 | return 0; |
2433 | } | |
2434 | ||
c5b28038 PZ |
2435 | /* |
2436 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2437 | * actually call this function. | |
2438 | */ | |
9cfc3e18 | 2439 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 2440 | { |
af449901 PZ |
2441 | if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2442 | p->cpus_ptr = new_mask; | |
2443 | return; | |
2444 | } | |
2445 | ||
3bd37062 | 2446 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
2447 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
2448 | } | |
2449 | ||
9cfc3e18 PZ |
2450 | static void |
2451 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 2452 | { |
6c37067e PZ |
2453 | struct rq *rq = task_rq(p); |
2454 | bool queued, running; | |
2455 | ||
af449901 PZ |
2456 | /* |
2457 | * This here violates the locking rules for affinity, since we're only | |
2458 | * supposed to change these variables while holding both rq->lock and | |
2459 | * p->pi_lock. | |
2460 | * | |
2461 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2462 | * accesses these variables under p->pi_lock and only does so after | |
2463 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2464 | * before finish_task(). | |
2465 | * | |
2466 | * XXX do further audits, this smells like something putrid. | |
2467 | */ | |
2468 | if (flags & SCA_MIGRATE_DISABLE) | |
2469 | SCHED_WARN_ON(!p->on_cpu); | |
2470 | else | |
2471 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2472 | |
2473 | queued = task_on_rq_queued(p); | |
2474 | running = task_current(rq, p); | |
2475 | ||
2476 | if (queued) { | |
2477 | /* | |
2478 | * Because __kthread_bind() calls this on blocked tasks without | |
2479 | * holding rq->lock. | |
2480 | */ | |
5cb9eaa3 | 2481 | lockdep_assert_rq_held(rq); |
7a57f32a | 2482 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2483 | } |
2484 | if (running) | |
2485 | put_prev_task(rq, p); | |
2486 | ||
9cfc3e18 | 2487 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 2488 | |
6c37067e | 2489 | if (queued) |
7134b3e9 | 2490 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2491 | if (running) |
03b7fad1 | 2492 | set_next_task(rq, p); |
c5b28038 PZ |
2493 | } |
2494 | ||
9cfc3e18 PZ |
2495 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2496 | { | |
2497 | __do_set_cpus_allowed(p, new_mask, 0); | |
2498 | } | |
2499 | ||
b90ca8ba WD |
2500 | int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, |
2501 | int node) | |
2502 | { | |
2503 | if (!src->user_cpus_ptr) | |
2504 | return 0; | |
2505 | ||
2506 | dst->user_cpus_ptr = kmalloc_node(cpumask_size(), GFP_KERNEL, node); | |
2507 | if (!dst->user_cpus_ptr) | |
2508 | return -ENOMEM; | |
2509 | ||
2510 | cpumask_copy(dst->user_cpus_ptr, src->user_cpus_ptr); | |
2511 | return 0; | |
2512 | } | |
2513 | ||
07ec77a1 WD |
2514 | static inline struct cpumask *clear_user_cpus_ptr(struct task_struct *p) |
2515 | { | |
2516 | struct cpumask *user_mask = NULL; | |
2517 | ||
2518 | swap(p->user_cpus_ptr, user_mask); | |
2519 | ||
2520 | return user_mask; | |
2521 | } | |
2522 | ||
b90ca8ba WD |
2523 | void release_user_cpus_ptr(struct task_struct *p) |
2524 | { | |
07ec77a1 | 2525 | kfree(clear_user_cpus_ptr(p)); |
b90ca8ba WD |
2526 | } |
2527 | ||
6d337eab | 2528 | /* |
c777d847 VS |
2529 | * This function is wildly self concurrent; here be dragons. |
2530 | * | |
2531 | * | |
2532 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2533 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2534 | * running, we have to kick it out using the CPU stopper. | |
2535 | * | |
2536 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2537 | * Consider: | |
2538 | * | |
2539 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2540 | * | |
2541 | * P0@CPU0 P1 | |
2542 | * | |
2543 | * migrate_disable(); | |
2544 | * <preempted> | |
2545 | * set_cpus_allowed_ptr(P0, [1]); | |
2546 | * | |
2547 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2548 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2549 | * This means we need the following scheme: | |
2550 | * | |
2551 | * P0@CPU0 P1 | |
2552 | * | |
2553 | * migrate_disable(); | |
2554 | * <preempted> | |
2555 | * set_cpus_allowed_ptr(P0, [1]); | |
2556 | * <blocks> | |
2557 | * <resumes> | |
2558 | * migrate_enable(); | |
2559 | * __set_cpus_allowed_ptr(); | |
2560 | * <wakes local stopper> | |
2561 | * `--> <woken on migration completion> | |
2562 | * | |
2563 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2564 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2565 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2566 | * should come into effect at the end of the Migrate-Disable region is the last | |
2567 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2568 | * but we still need to properly signal those waiting tasks at the appropriate | |
2569 | * moment. | |
2570 | * | |
2571 | * This is implemented using struct set_affinity_pending. The first | |
2572 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2573 | * setup an instance of that struct and install it on the targeted task_struct. | |
2574 | * Any and all further callers will reuse that instance. Those then wait for | |
2575 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2576 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2577 | * | |
2578 | * | |
2579 | * (1) In the cases covered above. There is one more where the completion is | |
2580 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
e140749c VS |
2581 | * occurs after the stopper bailed out due to the targeted task still being |
2582 | * Migrate-Disable. Consider: | |
c777d847 VS |
2583 | * |
2584 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2585 | * | |
e140749c VS |
2586 | * CPU0 P1 P2 |
2587 | * <P0> | |
2588 | * migrate_disable(); | |
2589 | * <preempted> | |
c777d847 VS |
2590 | * set_cpus_allowed_ptr(P0, [1]); |
2591 | * <blocks> | |
e140749c VS |
2592 | * <migration/0> |
2593 | * migration_cpu_stop() | |
2594 | * is_migration_disabled() | |
2595 | * <bails> | |
c777d847 VS |
2596 | * set_cpus_allowed_ptr(P0, [0, 1]); |
2597 | * <signal completion> | |
2598 | * <awakes> | |
2599 | * | |
2600 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2601 | * pending affinity completion is preceded by an uninstallation of | |
2602 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2603 | */ |
2604 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2605 | int dest_cpu, unsigned int flags) | |
2606 | { | |
2607 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
9e81889c | 2608 | bool stop_pending, complete = false; |
6d337eab PZ |
2609 | |
2610 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2611 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2612 | struct task_struct *push_task = NULL; |
2613 | ||
2614 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2615 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2616 | rq->push_busy = true; | |
2617 | push_task = get_task_struct(p); | |
2618 | } | |
2619 | ||
50caf9c1 PZ |
2620 | /* |
2621 | * If there are pending waiters, but no pending stop_work, | |
2622 | * then complete now. | |
2623 | */ | |
6d337eab | 2624 | pending = p->migration_pending; |
50caf9c1 | 2625 | if (pending && !pending->stop_pending) { |
6d337eab PZ |
2626 | p->migration_pending = NULL; |
2627 | complete = true; | |
2628 | } | |
50caf9c1 | 2629 | |
6d337eab PZ |
2630 | task_rq_unlock(rq, p, rf); |
2631 | ||
a7c81556 PZ |
2632 | if (push_task) { |
2633 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2634 | p, &rq->push_work); | |
2635 | } | |
2636 | ||
6d337eab | 2637 | if (complete) |
50caf9c1 | 2638 | complete_all(&pending->done); |
6d337eab PZ |
2639 | |
2640 | return 0; | |
2641 | } | |
2642 | ||
2643 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2644 | /* serialized by p->pi_lock */ | |
2645 | if (!p->migration_pending) { | |
c777d847 | 2646 | /* Install the request */ |
6d337eab PZ |
2647 | refcount_set(&my_pending.refs, 1); |
2648 | init_completion(&my_pending.done); | |
8a6edb52 PZ |
2649 | my_pending.arg = (struct migration_arg) { |
2650 | .task = p, | |
475ea6c6 | 2651 | .dest_cpu = dest_cpu, |
8a6edb52 PZ |
2652 | .pending = &my_pending, |
2653 | }; | |
2654 | ||
6d337eab PZ |
2655 | p->migration_pending = &my_pending; |
2656 | } else { | |
2657 | pending = p->migration_pending; | |
2658 | refcount_inc(&pending->refs); | |
475ea6c6 VS |
2659 | /* |
2660 | * Affinity has changed, but we've already installed a | |
2661 | * pending. migration_cpu_stop() *must* see this, else | |
2662 | * we risk a completion of the pending despite having a | |
2663 | * task on a disallowed CPU. | |
2664 | * | |
2665 | * Serialized by p->pi_lock, so this is safe. | |
2666 | */ | |
2667 | pending->arg.dest_cpu = dest_cpu; | |
6d337eab PZ |
2668 | } |
2669 | } | |
2670 | pending = p->migration_pending; | |
2671 | /* | |
2672 | * - !MIGRATE_ENABLE: | |
2673 | * we'll have installed a pending if there wasn't one already. | |
2674 | * | |
2675 | * - MIGRATE_ENABLE: | |
2676 | * we're here because the current CPU isn't matching anymore, | |
2677 | * the only way that can happen is because of a concurrent | |
2678 | * set_cpus_allowed_ptr() call, which should then still be | |
2679 | * pending completion. | |
2680 | * | |
2681 | * Either way, we really should have a @pending here. | |
2682 | */ | |
2683 | if (WARN_ON_ONCE(!pending)) { | |
2684 | task_rq_unlock(rq, p, rf); | |
2685 | return -EINVAL; | |
2686 | } | |
2687 | ||
2f064a59 | 2688 | if (task_running(rq, p) || READ_ONCE(p->__state) == TASK_WAKING) { |
c777d847 | 2689 | /* |
58b1a450 PZ |
2690 | * MIGRATE_ENABLE gets here because 'p == current', but for |
2691 | * anything else we cannot do is_migration_disabled(), punt | |
2692 | * and have the stopper function handle it all race-free. | |
c777d847 | 2693 | */ |
9e81889c PZ |
2694 | stop_pending = pending->stop_pending; |
2695 | if (!stop_pending) | |
2696 | pending->stop_pending = true; | |
58b1a450 | 2697 | |
58b1a450 PZ |
2698 | if (flags & SCA_MIGRATE_ENABLE) |
2699 | p->migration_flags &= ~MDF_PUSH; | |
50caf9c1 | 2700 | |
6d337eab | 2701 | task_rq_unlock(rq, p, rf); |
8a6edb52 | 2702 | |
9e81889c PZ |
2703 | if (!stop_pending) { |
2704 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2705 | &pending->arg, &pending->stop_work); | |
2706 | } | |
6d337eab | 2707 | |
58b1a450 PZ |
2708 | if (flags & SCA_MIGRATE_ENABLE) |
2709 | return 0; | |
6d337eab PZ |
2710 | } else { |
2711 | ||
2712 | if (!is_migration_disabled(p)) { | |
2713 | if (task_on_rq_queued(p)) | |
2714 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2715 | ||
50caf9c1 PZ |
2716 | if (!pending->stop_pending) { |
2717 | p->migration_pending = NULL; | |
2718 | complete = true; | |
2719 | } | |
6d337eab PZ |
2720 | } |
2721 | task_rq_unlock(rq, p, rf); | |
2722 | ||
6d337eab PZ |
2723 | if (complete) |
2724 | complete_all(&pending->done); | |
2725 | } | |
2726 | ||
2727 | wait_for_completion(&pending->done); | |
2728 | ||
2729 | if (refcount_dec_and_test(&pending->refs)) | |
50caf9c1 | 2730 | wake_up_var(&pending->refs); /* No UaF, just an address */ |
6d337eab | 2731 | |
c777d847 VS |
2732 | /* |
2733 | * Block the original owner of &pending until all subsequent callers | |
2734 | * have seen the completion and decremented the refcount | |
2735 | */ | |
6d337eab PZ |
2736 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2737 | ||
50caf9c1 PZ |
2738 | /* ARGH */ |
2739 | WARN_ON_ONCE(my_pending.stop_pending); | |
2740 | ||
6d337eab PZ |
2741 | return 0; |
2742 | } | |
2743 | ||
5cc389bc | 2744 | /* |
07ec77a1 | 2745 | * Called with both p->pi_lock and rq->lock held; drops both before returning. |
5cc389bc | 2746 | */ |
07ec77a1 WD |
2747 | static int __set_cpus_allowed_ptr_locked(struct task_struct *p, |
2748 | const struct cpumask *new_mask, | |
2749 | u32 flags, | |
2750 | struct rq *rq, | |
2751 | struct rq_flags *rf) | |
2752 | __releases(rq->lock) | |
2753 | __releases(p->pi_lock) | |
5cc389bc | 2754 | { |
234a503e | 2755 | const struct cpumask *cpu_allowed_mask = task_cpu_possible_mask(p); |
e9d867a6 | 2756 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
234a503e | 2757 | bool kthread = p->flags & PF_KTHREAD; |
07ec77a1 | 2758 | struct cpumask *user_mask = NULL; |
5cc389bc PZ |
2759 | unsigned int dest_cpu; |
2760 | int ret = 0; | |
2761 | ||
a499c3ea | 2762 | update_rq_clock(rq); |
5cc389bc | 2763 | |
234a503e | 2764 | if (kthread || is_migration_disabled(p)) { |
e9d867a6 | 2765 | /* |
741ba80f PZ |
2766 | * Kernel threads are allowed on online && !active CPUs, |
2767 | * however, during cpu-hot-unplug, even these might get pushed | |
2768 | * away if not KTHREAD_IS_PER_CPU. | |
af449901 PZ |
2769 | * |
2770 | * Specifically, migration_disabled() tasks must not fail the | |
2771 | * cpumask_any_and_distribute() pick below, esp. so on | |
2772 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2773 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2774 | */ |
2775 | cpu_valid_mask = cpu_online_mask; | |
2776 | } | |
2777 | ||
234a503e WD |
2778 | if (!kthread && !cpumask_subset(new_mask, cpu_allowed_mask)) { |
2779 | ret = -EINVAL; | |
2780 | goto out; | |
2781 | } | |
2782 | ||
25834c73 PZ |
2783 | /* |
2784 | * Must re-check here, to close a race against __kthread_bind(), | |
2785 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2786 | */ | |
9cfc3e18 | 2787 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2788 | ret = -EINVAL; |
2789 | goto out; | |
2790 | } | |
2791 | ||
885b3ba4 VS |
2792 | if (!(flags & SCA_MIGRATE_ENABLE)) { |
2793 | if (cpumask_equal(&p->cpus_mask, new_mask)) | |
2794 | goto out; | |
2795 | ||
2796 | if (WARN_ON_ONCE(p == current && | |
2797 | is_migration_disabled(p) && | |
2798 | !cpumask_test_cpu(task_cpu(p), new_mask))) { | |
2799 | ret = -EBUSY; | |
2800 | goto out; | |
2801 | } | |
2802 | } | |
5cc389bc | 2803 | |
46a87b38 PT |
2804 | /* |
2805 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2806 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2807 | * immediately required to distribute the tasks within their new mask. | |
2808 | */ | |
2809 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 2810 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2811 | ret = -EINVAL; |
2812 | goto out; | |
2813 | } | |
2814 | ||
9cfc3e18 | 2815 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 2816 | |
07ec77a1 WD |
2817 | if (flags & SCA_USER) |
2818 | user_mask = clear_user_cpus_ptr(p); | |
2819 | ||
2820 | ret = affine_move_task(rq, p, rf, dest_cpu, flags); | |
2821 | ||
2822 | kfree(user_mask); | |
2823 | ||
2824 | return ret; | |
5cc389bc | 2825 | |
5cc389bc | 2826 | out: |
07ec77a1 | 2827 | task_rq_unlock(rq, p, rf); |
5cc389bc PZ |
2828 | |
2829 | return ret; | |
2830 | } | |
25834c73 | 2831 | |
07ec77a1 WD |
2832 | /* |
2833 | * Change a given task's CPU affinity. Migrate the thread to a | |
2834 | * proper CPU and schedule it away if the CPU it's executing on | |
2835 | * is removed from the allowed bitmask. | |
2836 | * | |
2837 | * NOTE: the caller must have a valid reference to the task, the | |
2838 | * task must not exit() & deallocate itself prematurely. The | |
2839 | * call is not atomic; no spinlocks may be held. | |
2840 | */ | |
2841 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
2842 | const struct cpumask *new_mask, u32 flags) | |
2843 | { | |
2844 | struct rq_flags rf; | |
2845 | struct rq *rq; | |
2846 | ||
2847 | rq = task_rq_lock(p, &rf); | |
2848 | return __set_cpus_allowed_ptr_locked(p, new_mask, flags, rq, &rf); | |
2849 | } | |
2850 | ||
25834c73 PZ |
2851 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
2852 | { | |
9cfc3e18 | 2853 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 2854 | } |
5cc389bc PZ |
2855 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2856 | ||
07ec77a1 WD |
2857 | /* |
2858 | * Change a given task's CPU affinity to the intersection of its current | |
2859 | * affinity mask and @subset_mask, writing the resulting mask to @new_mask | |
2860 | * and pointing @p->user_cpus_ptr to a copy of the old mask. | |
2861 | * If the resulting mask is empty, leave the affinity unchanged and return | |
2862 | * -EINVAL. | |
2863 | */ | |
2864 | static int restrict_cpus_allowed_ptr(struct task_struct *p, | |
2865 | struct cpumask *new_mask, | |
2866 | const struct cpumask *subset_mask) | |
2867 | { | |
2868 | struct cpumask *user_mask = NULL; | |
2869 | struct rq_flags rf; | |
2870 | struct rq *rq; | |
2871 | int err; | |
2872 | ||
2873 | if (!p->user_cpus_ptr) { | |
2874 | user_mask = kmalloc(cpumask_size(), GFP_KERNEL); | |
2875 | if (!user_mask) | |
2876 | return -ENOMEM; | |
2877 | } | |
2878 | ||
2879 | rq = task_rq_lock(p, &rf); | |
2880 | ||
2881 | /* | |
2882 | * Forcefully restricting the affinity of a deadline task is | |
2883 | * likely to cause problems, so fail and noisily override the | |
2884 | * mask entirely. | |
2885 | */ | |
2886 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { | |
2887 | err = -EPERM; | |
2888 | goto err_unlock; | |
2889 | } | |
2890 | ||
2891 | if (!cpumask_and(new_mask, &p->cpus_mask, subset_mask)) { | |
2892 | err = -EINVAL; | |
2893 | goto err_unlock; | |
2894 | } | |
2895 | ||
2896 | /* | |
2897 | * We're about to butcher the task affinity, so keep track of what | |
2898 | * the user asked for in case we're able to restore it later on. | |
2899 | */ | |
2900 | if (user_mask) { | |
2901 | cpumask_copy(user_mask, p->cpus_ptr); | |
2902 | p->user_cpus_ptr = user_mask; | |
2903 | } | |
2904 | ||
2905 | return __set_cpus_allowed_ptr_locked(p, new_mask, 0, rq, &rf); | |
2906 | ||
2907 | err_unlock: | |
2908 | task_rq_unlock(rq, p, &rf); | |
2909 | kfree(user_mask); | |
2910 | return err; | |
2911 | } | |
2912 | ||
2913 | /* | |
2914 | * Restrict the CPU affinity of task @p so that it is a subset of | |
2915 | * task_cpu_possible_mask() and point @p->user_cpu_ptr to a copy of the | |
2916 | * old affinity mask. If the resulting mask is empty, we warn and walk | |
2917 | * up the cpuset hierarchy until we find a suitable mask. | |
2918 | */ | |
2919 | void force_compatible_cpus_allowed_ptr(struct task_struct *p) | |
2920 | { | |
2921 | cpumask_var_t new_mask; | |
2922 | const struct cpumask *override_mask = task_cpu_possible_mask(p); | |
2923 | ||
2924 | alloc_cpumask_var(&new_mask, GFP_KERNEL); | |
2925 | ||
2926 | /* | |
2927 | * __migrate_task() can fail silently in the face of concurrent | |
2928 | * offlining of the chosen destination CPU, so take the hotplug | |
2929 | * lock to ensure that the migration succeeds. | |
2930 | */ | |
2931 | cpus_read_lock(); | |
2932 | if (!cpumask_available(new_mask)) | |
2933 | goto out_set_mask; | |
2934 | ||
2935 | if (!restrict_cpus_allowed_ptr(p, new_mask, override_mask)) | |
2936 | goto out_free_mask; | |
2937 | ||
2938 | /* | |
2939 | * We failed to find a valid subset of the affinity mask for the | |
2940 | * task, so override it based on its cpuset hierarchy. | |
2941 | */ | |
2942 | cpuset_cpus_allowed(p, new_mask); | |
2943 | override_mask = new_mask; | |
2944 | ||
2945 | out_set_mask: | |
2946 | if (printk_ratelimit()) { | |
2947 | printk_deferred("Overriding affinity for process %d (%s) to CPUs %*pbl\n", | |
2948 | task_pid_nr(p), p->comm, | |
2949 | cpumask_pr_args(override_mask)); | |
2950 | } | |
2951 | ||
2952 | WARN_ON(set_cpus_allowed_ptr(p, override_mask)); | |
2953 | out_free_mask: | |
2954 | cpus_read_unlock(); | |
2955 | free_cpumask_var(new_mask); | |
2956 | } | |
2957 | ||
2958 | static int | |
2959 | __sched_setaffinity(struct task_struct *p, const struct cpumask *mask); | |
2960 | ||
2961 | /* | |
2962 | * Restore the affinity of a task @p which was previously restricted by a | |
2963 | * call to force_compatible_cpus_allowed_ptr(). This will clear (and free) | |
2964 | * @p->user_cpus_ptr. | |
2965 | * | |
2966 | * It is the caller's responsibility to serialise this with any calls to | |
2967 | * force_compatible_cpus_allowed_ptr(@p). | |
2968 | */ | |
2969 | void relax_compatible_cpus_allowed_ptr(struct task_struct *p) | |
2970 | { | |
2971 | struct cpumask *user_mask = p->user_cpus_ptr; | |
2972 | unsigned long flags; | |
2973 | ||
2974 | /* | |
2975 | * Try to restore the old affinity mask. If this fails, then | |
2976 | * we free the mask explicitly to avoid it being inherited across | |
2977 | * a subsequent fork(). | |
2978 | */ | |
2979 | if (!user_mask || !__sched_setaffinity(p, user_mask)) | |
2980 | return; | |
2981 | ||
2982 | raw_spin_lock_irqsave(&p->pi_lock, flags); | |
2983 | user_mask = clear_user_cpus_ptr(p); | |
2984 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
2985 | ||
2986 | kfree(user_mask); | |
2987 | } | |
2988 | ||
dd41f596 | 2989 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2990 | { |
e2912009 | 2991 | #ifdef CONFIG_SCHED_DEBUG |
2f064a59 PZ |
2992 | unsigned int state = READ_ONCE(p->__state); |
2993 | ||
e2912009 PZ |
2994 | /* |
2995 | * We should never call set_task_cpu() on a blocked task, | |
2996 | * ttwu() will sort out the placement. | |
2997 | */ | |
2f064a59 | 2998 | WARN_ON_ONCE(state != TASK_RUNNING && state != TASK_WAKING && !p->on_rq); |
0122ec5b | 2999 | |
3ea94de1 JP |
3000 | /* |
3001 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
3002 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
3003 | * time relying on p->on_rq. | |
3004 | */ | |
2f064a59 | 3005 | WARN_ON_ONCE(state == TASK_RUNNING && |
3ea94de1 JP |
3006 | p->sched_class == &fair_sched_class && |
3007 | (p->on_rq && !task_on_rq_migrating(p))); | |
3008 | ||
0122ec5b | 3009 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
3010 | /* |
3011 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
3012 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
3013 | * | |
3014 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 3015 | * see task_group(). |
6c6c54e1 PZ |
3016 | * |
3017 | * Furthermore, all task_rq users should acquire both locks, see | |
3018 | * task_rq_lock(). | |
3019 | */ | |
0122ec5b | 3020 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
9ef7e7e3 | 3021 | lockdep_is_held(__rq_lockp(task_rq(p))))); |
0122ec5b | 3022 | #endif |
4ff9083b PZ |
3023 | /* |
3024 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
3025 | */ | |
3026 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
3027 | |
3028 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
3029 | #endif |
3030 | ||
de1d7286 | 3031 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 3032 | |
0c69774e | 3033 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 3034 | if (p->sched_class->migrate_task_rq) |
1327237a | 3035 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 3036 | p->se.nr_migrations++; |
d7822b1e | 3037 | rseq_migrate(p); |
ff303e66 | 3038 | perf_event_task_migrate(p); |
0c69774e | 3039 | } |
dd41f596 IM |
3040 | |
3041 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
3042 | } |
3043 | ||
0ad4e3df | 3044 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
3045 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
3046 | { | |
da0c1e65 | 3047 | if (task_on_rq_queued(p)) { |
ac66f547 | 3048 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 3049 | struct rq_flags srf, drf; |
ac66f547 PZ |
3050 | |
3051 | src_rq = task_rq(p); | |
3052 | dst_rq = cpu_rq(cpu); | |
3053 | ||
8a8c69c3 PZ |
3054 | rq_pin_lock(src_rq, &srf); |
3055 | rq_pin_lock(dst_rq, &drf); | |
3056 | ||
ac66f547 PZ |
3057 | deactivate_task(src_rq, p, 0); |
3058 | set_task_cpu(p, cpu); | |
3059 | activate_task(dst_rq, p, 0); | |
3060 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
3061 | |
3062 | rq_unpin_lock(dst_rq, &drf); | |
3063 | rq_unpin_lock(src_rq, &srf); | |
3064 | ||
ac66f547 PZ |
3065 | } else { |
3066 | /* | |
3067 | * Task isn't running anymore; make it appear like we migrated | |
3068 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 3069 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
3070 | */ |
3071 | p->wake_cpu = cpu; | |
3072 | } | |
3073 | } | |
3074 | ||
3075 | struct migration_swap_arg { | |
3076 | struct task_struct *src_task, *dst_task; | |
3077 | int src_cpu, dst_cpu; | |
3078 | }; | |
3079 | ||
3080 | static int migrate_swap_stop(void *data) | |
3081 | { | |
3082 | struct migration_swap_arg *arg = data; | |
3083 | struct rq *src_rq, *dst_rq; | |
3084 | int ret = -EAGAIN; | |
3085 | ||
62694cd5 PZ |
3086 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
3087 | return -EAGAIN; | |
3088 | ||
ac66f547 PZ |
3089 | src_rq = cpu_rq(arg->src_cpu); |
3090 | dst_rq = cpu_rq(arg->dst_cpu); | |
3091 | ||
74602315 PZ |
3092 | double_raw_lock(&arg->src_task->pi_lock, |
3093 | &arg->dst_task->pi_lock); | |
ac66f547 | 3094 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 3095 | |
ac66f547 PZ |
3096 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
3097 | goto unlock; | |
3098 | ||
3099 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
3100 | goto unlock; | |
3101 | ||
3bd37062 | 3102 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
3103 | goto unlock; |
3104 | ||
3bd37062 | 3105 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
3106 | goto unlock; |
3107 | ||
3108 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
3109 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
3110 | ||
3111 | ret = 0; | |
3112 | ||
3113 | unlock: | |
3114 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
3115 | raw_spin_unlock(&arg->dst_task->pi_lock); |
3116 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
3117 | |
3118 | return ret; | |
3119 | } | |
3120 | ||
3121 | /* | |
3122 | * Cross migrate two tasks | |
3123 | */ | |
0ad4e3df SD |
3124 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
3125 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
3126 | { |
3127 | struct migration_swap_arg arg; | |
3128 | int ret = -EINVAL; | |
3129 | ||
ac66f547 PZ |
3130 | arg = (struct migration_swap_arg){ |
3131 | .src_task = cur, | |
0ad4e3df | 3132 | .src_cpu = curr_cpu, |
ac66f547 | 3133 | .dst_task = p, |
0ad4e3df | 3134 | .dst_cpu = target_cpu, |
ac66f547 PZ |
3135 | }; |
3136 | ||
3137 | if (arg.src_cpu == arg.dst_cpu) | |
3138 | goto out; | |
3139 | ||
6acce3ef PZ |
3140 | /* |
3141 | * These three tests are all lockless; this is OK since all of them | |
3142 | * will be re-checked with proper locks held further down the line. | |
3143 | */ | |
ac66f547 PZ |
3144 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
3145 | goto out; | |
3146 | ||
3bd37062 | 3147 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
3148 | goto out; |
3149 | ||
3bd37062 | 3150 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
3151 | goto out; |
3152 | ||
286549dc | 3153 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
3154 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
3155 | ||
3156 | out: | |
ac66f547 PZ |
3157 | return ret; |
3158 | } | |
0ad4e3df | 3159 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 3160 | |
1da177e4 LT |
3161 | /* |
3162 | * wait_task_inactive - wait for a thread to unschedule. | |
3163 | * | |
85ba2d86 RM |
3164 | * If @match_state is nonzero, it's the @p->state value just checked and |
3165 | * not expected to change. If it changes, i.e. @p might have woken up, | |
3166 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
3167 | * we return a positive number (its total switch count). If a second call | |
3168 | * a short while later returns the same number, the caller can be sure that | |
3169 | * @p has remained unscheduled the whole time. | |
3170 | * | |
1da177e4 LT |
3171 | * The caller must ensure that the task *will* unschedule sometime soon, |
3172 | * else this function might spin for a *long* time. This function can't | |
3173 | * be called with interrupts off, or it may introduce deadlock with | |
3174 | * smp_call_function() if an IPI is sent by the same process we are | |
3175 | * waiting to become inactive. | |
3176 | */ | |
2f064a59 | 3177 | unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state) |
1da177e4 | 3178 | { |
da0c1e65 | 3179 | int running, queued; |
eb580751 | 3180 | struct rq_flags rf; |
85ba2d86 | 3181 | unsigned long ncsw; |
70b97a7f | 3182 | struct rq *rq; |
1da177e4 | 3183 | |
3a5c359a AK |
3184 | for (;;) { |
3185 | /* | |
3186 | * We do the initial early heuristics without holding | |
3187 | * any task-queue locks at all. We'll only try to get | |
3188 | * the runqueue lock when things look like they will | |
3189 | * work out! | |
3190 | */ | |
3191 | rq = task_rq(p); | |
fa490cfd | 3192 | |
3a5c359a AK |
3193 | /* |
3194 | * If the task is actively running on another CPU | |
3195 | * still, just relax and busy-wait without holding | |
3196 | * any locks. | |
3197 | * | |
3198 | * NOTE! Since we don't hold any locks, it's not | |
3199 | * even sure that "rq" stays as the right runqueue! | |
3200 | * But we don't care, since "task_running()" will | |
3201 | * return false if the runqueue has changed and p | |
3202 | * is actually now running somewhere else! | |
3203 | */ | |
85ba2d86 | 3204 | while (task_running(rq, p)) { |
2f064a59 | 3205 | if (match_state && unlikely(READ_ONCE(p->__state) != match_state)) |
85ba2d86 | 3206 | return 0; |
3a5c359a | 3207 | cpu_relax(); |
85ba2d86 | 3208 | } |
fa490cfd | 3209 | |
3a5c359a AK |
3210 | /* |
3211 | * Ok, time to look more closely! We need the rq | |
3212 | * lock now, to be *sure*. If we're wrong, we'll | |
3213 | * just go back and repeat. | |
3214 | */ | |
eb580751 | 3215 | rq = task_rq_lock(p, &rf); |
27a9da65 | 3216 | trace_sched_wait_task(p); |
3a5c359a | 3217 | running = task_running(rq, p); |
da0c1e65 | 3218 | queued = task_on_rq_queued(p); |
85ba2d86 | 3219 | ncsw = 0; |
2f064a59 | 3220 | if (!match_state || READ_ONCE(p->__state) == match_state) |
93dcf55f | 3221 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 3222 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 3223 | |
85ba2d86 RM |
3224 | /* |
3225 | * If it changed from the expected state, bail out now. | |
3226 | */ | |
3227 | if (unlikely(!ncsw)) | |
3228 | break; | |
3229 | ||
3a5c359a AK |
3230 | /* |
3231 | * Was it really running after all now that we | |
3232 | * checked with the proper locks actually held? | |
3233 | * | |
3234 | * Oops. Go back and try again.. | |
3235 | */ | |
3236 | if (unlikely(running)) { | |
3237 | cpu_relax(); | |
3238 | continue; | |
3239 | } | |
fa490cfd | 3240 | |
3a5c359a AK |
3241 | /* |
3242 | * It's not enough that it's not actively running, | |
3243 | * it must be off the runqueue _entirely_, and not | |
3244 | * preempted! | |
3245 | * | |
80dd99b3 | 3246 | * So if it was still runnable (but just not actively |
3a5c359a AK |
3247 | * running right now), it's preempted, and we should |
3248 | * yield - it could be a while. | |
3249 | */ | |
da0c1e65 | 3250 | if (unlikely(queued)) { |
8b0e1953 | 3251 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
3252 | |
3253 | set_current_state(TASK_UNINTERRUPTIBLE); | |
3254 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
3255 | continue; |
3256 | } | |
fa490cfd | 3257 | |
3a5c359a AK |
3258 | /* |
3259 | * Ahh, all good. It wasn't running, and it wasn't | |
3260 | * runnable, which means that it will never become | |
3261 | * running in the future either. We're all done! | |
3262 | */ | |
3263 | break; | |
3264 | } | |
85ba2d86 RM |
3265 | |
3266 | return ncsw; | |
1da177e4 LT |
3267 | } |
3268 | ||
3269 | /*** | |
3270 | * kick_process - kick a running thread to enter/exit the kernel | |
3271 | * @p: the to-be-kicked thread | |
3272 | * | |
3273 | * Cause a process which is running on another CPU to enter | |
3274 | * kernel-mode, without any delay. (to get signals handled.) | |
3275 | * | |
25985edc | 3276 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
3277 | * because all it wants to ensure is that the remote task enters |
3278 | * the kernel. If the IPI races and the task has been migrated | |
3279 | * to another CPU then no harm is done and the purpose has been | |
3280 | * achieved as well. | |
3281 | */ | |
36c8b586 | 3282 | void kick_process(struct task_struct *p) |
1da177e4 LT |
3283 | { |
3284 | int cpu; | |
3285 | ||
3286 | preempt_disable(); | |
3287 | cpu = task_cpu(p); | |
3288 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
3289 | smp_send_reschedule(cpu); | |
3290 | preempt_enable(); | |
3291 | } | |
b43e3521 | 3292 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 3293 | |
30da688e | 3294 | /* |
3bd37062 | 3295 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
3296 | * |
3297 | * A few notes on cpu_active vs cpu_online: | |
3298 | * | |
3299 | * - cpu_active must be a subset of cpu_online | |
3300 | * | |
97fb7a0a | 3301 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 3302 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 3303 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
3304 | * see it. |
3305 | * | |
d1ccc66d | 3306 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 3307 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 3308 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
3309 | * off. |
3310 | * | |
3311 | * This means that fallback selection must not select !active CPUs. | |
3312 | * And can assume that any active CPU must be online. Conversely | |
3313 | * select_task_rq() below may allow selection of !active CPUs in order | |
3314 | * to satisfy the above rules. | |
30da688e | 3315 | */ |
5da9a0fb PZ |
3316 | static int select_fallback_rq(int cpu, struct task_struct *p) |
3317 | { | |
aa00d89c TC |
3318 | int nid = cpu_to_node(cpu); |
3319 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
3320 | enum { cpuset, possible, fail } state = cpuset; |
3321 | int dest_cpu; | |
5da9a0fb | 3322 | |
aa00d89c | 3323 | /* |
d1ccc66d IM |
3324 | * If the node that the CPU is on has been offlined, cpu_to_node() |
3325 | * will return -1. There is no CPU on the node, and we should | |
3326 | * select the CPU on the other node. | |
aa00d89c TC |
3327 | */ |
3328 | if (nid != -1) { | |
3329 | nodemask = cpumask_of_node(nid); | |
3330 | ||
3331 | /* Look for allowed, online CPU in same node. */ | |
3332 | for_each_cpu(dest_cpu, nodemask) { | |
9ae606bc | 3333 | if (is_cpu_allowed(p, dest_cpu)) |
aa00d89c TC |
3334 | return dest_cpu; |
3335 | } | |
2baab4e9 | 3336 | } |
5da9a0fb | 3337 | |
2baab4e9 PZ |
3338 | for (;;) { |
3339 | /* Any allowed, online CPU? */ | |
3bd37062 | 3340 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 3341 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 3342 | continue; |
175f0e25 | 3343 | |
2baab4e9 PZ |
3344 | goto out; |
3345 | } | |
5da9a0fb | 3346 | |
e73e85f0 | 3347 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
3348 | switch (state) { |
3349 | case cpuset: | |
97c0054d | 3350 | if (cpuset_cpus_allowed_fallback(p)) { |
e73e85f0 ON |
3351 | state = possible; |
3352 | break; | |
3353 | } | |
df561f66 | 3354 | fallthrough; |
2baab4e9 | 3355 | case possible: |
af449901 PZ |
3356 | /* |
3357 | * XXX When called from select_task_rq() we only | |
3358 | * hold p->pi_lock and again violate locking order. | |
3359 | * | |
3360 | * More yuck to audit. | |
3361 | */ | |
9ae606bc | 3362 | do_set_cpus_allowed(p, task_cpu_possible_mask(p)); |
2baab4e9 PZ |
3363 | state = fail; |
3364 | break; | |
2baab4e9 PZ |
3365 | case fail: |
3366 | BUG(); | |
3367 | break; | |
3368 | } | |
3369 | } | |
3370 | ||
3371 | out: | |
3372 | if (state != cpuset) { | |
3373 | /* | |
3374 | * Don't tell them about moving exiting tasks or | |
3375 | * kernel threads (both mm NULL), since they never | |
3376 | * leave kernel. | |
3377 | */ | |
3378 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 3379 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
3380 | task_pid_nr(p), p->comm, cpu); |
3381 | } | |
5da9a0fb PZ |
3382 | } |
3383 | ||
3384 | return dest_cpu; | |
3385 | } | |
3386 | ||
e2912009 | 3387 | /* |
3bd37062 | 3388 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 3389 | */ |
970b13ba | 3390 | static inline |
3aef1551 | 3391 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 3392 | { |
cbce1a68 PZ |
3393 | lockdep_assert_held(&p->pi_lock); |
3394 | ||
af449901 | 3395 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 3396 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 3397 | else |
3bd37062 | 3398 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
3399 | |
3400 | /* | |
3401 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 3402 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 3403 | * CPU. |
e2912009 PZ |
3404 | * |
3405 | * Since this is common to all placement strategies, this lives here. | |
3406 | * | |
3407 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
3408 | * not worry about this generic constraint ] | |
3409 | */ | |
7af443ee | 3410 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 3411 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
3412 | |
3413 | return cpu; | |
970b13ba | 3414 | } |
09a40af5 | 3415 | |
f5832c19 NP |
3416 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
3417 | { | |
ded467dc | 3418 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
3419 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
3420 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
3421 | ||
3422 | if (stop) { | |
3423 | /* | |
3424 | * Make it appear like a SCHED_FIFO task, its something | |
3425 | * userspace knows about and won't get confused about. | |
3426 | * | |
3427 | * Also, it will make PI more or less work without too | |
3428 | * much confusion -- but then, stop work should not | |
3429 | * rely on PI working anyway. | |
3430 | */ | |
3431 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
3432 | ||
3433 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
3434 | |
3435 | /* | |
3436 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
3437 | * adjust the effective priority of a task. As a result, | |
3438 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
3439 | * which can then trigger wakeups of the stop thread to push | |
3440 | * around the current task. | |
3441 | * | |
3442 | * The stop task itself will never be part of the PI-chain, it | |
3443 | * never blocks, therefore that ->pi_lock recursion is safe. | |
3444 | * Tell lockdep about this by placing the stop->pi_lock in its | |
3445 | * own class. | |
3446 | */ | |
3447 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
3448 | } |
3449 | ||
3450 | cpu_rq(cpu)->stop = stop; | |
3451 | ||
3452 | if (old_stop) { | |
3453 | /* | |
3454 | * Reset it back to a normal scheduling class so that | |
3455 | * it can die in pieces. | |
3456 | */ | |
3457 | old_stop->sched_class = &rt_sched_class; | |
3458 | } | |
3459 | } | |
3460 | ||
74d862b6 | 3461 | #else /* CONFIG_SMP */ |
25834c73 PZ |
3462 | |
3463 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
3464 | const struct cpumask *new_mask, |
3465 | u32 flags) | |
25834c73 PZ |
3466 | { |
3467 | return set_cpus_allowed_ptr(p, new_mask); | |
3468 | } | |
3469 | ||
af449901 PZ |
3470 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
3471 | ||
3015ef4b TG |
3472 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
3473 | { | |
3474 | return false; | |
3475 | } | |
3476 | ||
74d862b6 | 3477 | #endif /* !CONFIG_SMP */ |
970b13ba | 3478 | |
d7c01d27 | 3479 | static void |
b84cb5df | 3480 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 3481 | { |
4fa8d299 | 3482 | struct rq *rq; |
b84cb5df | 3483 | |
4fa8d299 JP |
3484 | if (!schedstat_enabled()) |
3485 | return; | |
3486 | ||
3487 | rq = this_rq(); | |
d7c01d27 | 3488 | |
4fa8d299 JP |
3489 | #ifdef CONFIG_SMP |
3490 | if (cpu == rq->cpu) { | |
b85c8b71 PZ |
3491 | __schedstat_inc(rq->ttwu_local); |
3492 | __schedstat_inc(p->se.statistics.nr_wakeups_local); | |
d7c01d27 PZ |
3493 | } else { |
3494 | struct sched_domain *sd; | |
3495 | ||
b85c8b71 | 3496 | __schedstat_inc(p->se.statistics.nr_wakeups_remote); |
057f3fad | 3497 | rcu_read_lock(); |
4fa8d299 | 3498 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 3499 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 3500 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
3501 | break; |
3502 | } | |
3503 | } | |
057f3fad | 3504 | rcu_read_unlock(); |
d7c01d27 | 3505 | } |
f339b9dc PZ |
3506 | |
3507 | if (wake_flags & WF_MIGRATED) | |
b85c8b71 | 3508 | __schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
d7c01d27 PZ |
3509 | #endif /* CONFIG_SMP */ |
3510 | ||
b85c8b71 PZ |
3511 | __schedstat_inc(rq->ttwu_count); |
3512 | __schedstat_inc(p->se.statistics.nr_wakeups); | |
d7c01d27 PZ |
3513 | |
3514 | if (wake_flags & WF_SYNC) | |
b85c8b71 | 3515 | __schedstat_inc(p->se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
3516 | } |
3517 | ||
23f41eeb PZ |
3518 | /* |
3519 | * Mark the task runnable and perform wakeup-preemption. | |
3520 | */ | |
e7904a28 | 3521 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3522 | struct rq_flags *rf) |
9ed3811a | 3523 | { |
9ed3811a | 3524 | check_preempt_curr(rq, p, wake_flags); |
2f064a59 | 3525 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fbd705a0 PZ |
3526 | trace_sched_wakeup(p); |
3527 | ||
9ed3811a | 3528 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
3529 | if (p->sched_class->task_woken) { |
3530 | /* | |
b19a888c | 3531 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 3532 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 3533 | */ |
d8ac8971 | 3534 | rq_unpin_lock(rq, rf); |
9ed3811a | 3535 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3536 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 3537 | } |
9ed3811a | 3538 | |
e69c6341 | 3539 | if (rq->idle_stamp) { |
78becc27 | 3540 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 3541 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 3542 | |
abfafa54 JL |
3543 | update_avg(&rq->avg_idle, delta); |
3544 | ||
3545 | if (rq->avg_idle > max) | |
9ed3811a | 3546 | rq->avg_idle = max; |
abfafa54 | 3547 | |
94aafc3e PZ |
3548 | rq->wake_stamp = jiffies; |
3549 | rq->wake_avg_idle = rq->avg_idle / 2; | |
3550 | ||
9ed3811a TH |
3551 | rq->idle_stamp = 0; |
3552 | } | |
3553 | #endif | |
3554 | } | |
3555 | ||
c05fbafb | 3556 | static void |
e7904a28 | 3557 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3558 | struct rq_flags *rf) |
c05fbafb | 3559 | { |
77558e4d | 3560 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 3561 | |
5cb9eaa3 | 3562 | lockdep_assert_rq_held(rq); |
cbce1a68 | 3563 | |
c05fbafb PZ |
3564 | if (p->sched_contributes_to_load) |
3565 | rq->nr_uninterruptible--; | |
b5179ac7 | 3566 | |
dbfb089d | 3567 | #ifdef CONFIG_SMP |
b5179ac7 | 3568 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 3569 | en_flags |= ENQUEUE_MIGRATED; |
ec618b84 | 3570 | else |
c05fbafb | 3571 | #endif |
ec618b84 PZ |
3572 | if (p->in_iowait) { |
3573 | delayacct_blkio_end(p); | |
3574 | atomic_dec(&task_rq(p)->nr_iowait); | |
3575 | } | |
c05fbafb | 3576 | |
1b174a2c | 3577 | activate_task(rq, p, en_flags); |
d8ac8971 | 3578 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
3579 | } |
3580 | ||
3581 | /* | |
58877d34 PZ |
3582 | * Consider @p being inside a wait loop: |
3583 | * | |
3584 | * for (;;) { | |
3585 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
3586 | * | |
3587 | * if (CONDITION) | |
3588 | * break; | |
3589 | * | |
3590 | * schedule(); | |
3591 | * } | |
3592 | * __set_current_state(TASK_RUNNING); | |
3593 | * | |
3594 | * between set_current_state() and schedule(). In this case @p is still | |
3595 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3596 | * an atomic manner. | |
3597 | * | |
3598 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3599 | * then schedule() must still happen and p->state can be changed to | |
3600 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3601 | * need to do a full wakeup with enqueue. | |
3602 | * | |
3603 | * Returns: %true when the wakeup is done, | |
3604 | * %false otherwise. | |
c05fbafb | 3605 | */ |
58877d34 | 3606 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3607 | { |
eb580751 | 3608 | struct rq_flags rf; |
c05fbafb PZ |
3609 | struct rq *rq; |
3610 | int ret = 0; | |
3611 | ||
eb580751 | 3612 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3613 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3614 | /* check_preempt_curr() may use rq clock */ |
3615 | update_rq_clock(rq); | |
d8ac8971 | 3616 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3617 | ret = 1; |
3618 | } | |
eb580751 | 3619 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3620 | |
3621 | return ret; | |
3622 | } | |
3623 | ||
317f3941 | 3624 | #ifdef CONFIG_SMP |
a1488664 | 3625 | void sched_ttwu_pending(void *arg) |
317f3941 | 3626 | { |
a1488664 | 3627 | struct llist_node *llist = arg; |
317f3941 | 3628 | struct rq *rq = this_rq(); |
73215849 | 3629 | struct task_struct *p, *t; |
d8ac8971 | 3630 | struct rq_flags rf; |
317f3941 | 3631 | |
e3baac47 PZ |
3632 | if (!llist) |
3633 | return; | |
3634 | ||
126c2092 PZ |
3635 | /* |
3636 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
3637 | * Races such that false-negatives are possible, since they | |
3638 | * are shorter lived that false-positives would be. | |
3639 | */ | |
3640 | WRITE_ONCE(rq->ttwu_pending, 0); | |
3641 | ||
8a8c69c3 | 3642 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3643 | update_rq_clock(rq); |
317f3941 | 3644 | |
8c4890d1 | 3645 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3646 | if (WARN_ON_ONCE(p->on_cpu)) |
3647 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3648 | ||
3649 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3650 | set_task_cpu(p, cpu_of(rq)); | |
3651 | ||
73215849 | 3652 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3653 | } |
317f3941 | 3654 | |
8a8c69c3 | 3655 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3656 | } |
3657 | ||
b2a02fc4 | 3658 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3659 | { |
b2a02fc4 | 3660 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3661 | |
b2a02fc4 PZ |
3662 | if (!set_nr_if_polling(rq->idle)) |
3663 | arch_send_call_function_single_ipi(cpu); | |
3664 | else | |
3665 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3666 | } |
3667 | ||
2ebb1771 MG |
3668 | /* |
3669 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3670 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3671 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3672 | * of the wakeup instead of the waker. | |
3673 | */ | |
3674 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3675 | { |
e3baac47 PZ |
3676 | struct rq *rq = cpu_rq(cpu); |
3677 | ||
b7e7ade3 PZ |
3678 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3679 | ||
126c2092 | 3680 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3681 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3682 | } |
d6aa8f85 | 3683 | |
f6be8af1 CL |
3684 | void wake_up_if_idle(int cpu) |
3685 | { | |
3686 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3687 | struct rq_flags rf; |
f6be8af1 | 3688 | |
fd7de1e8 AL |
3689 | rcu_read_lock(); |
3690 | ||
3691 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3692 | goto out; | |
f6be8af1 CL |
3693 | |
3694 | if (set_nr_if_polling(rq->idle)) { | |
3695 | trace_sched_wake_idle_without_ipi(cpu); | |
3696 | } else { | |
8a8c69c3 | 3697 | rq_lock_irqsave(rq, &rf); |
f6be8af1 CL |
3698 | if (is_idle_task(rq->curr)) |
3699 | smp_send_reschedule(cpu); | |
d1ccc66d | 3700 | /* Else CPU is not idle, do nothing here: */ |
8a8c69c3 | 3701 | rq_unlock_irqrestore(rq, &rf); |
f6be8af1 | 3702 | } |
fd7de1e8 AL |
3703 | |
3704 | out: | |
3705 | rcu_read_unlock(); | |
f6be8af1 CL |
3706 | } |
3707 | ||
39be3501 | 3708 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 PZ |
3709 | { |
3710 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); | |
3711 | } | |
c6e7bd7a | 3712 | |
2ebb1771 MG |
3713 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
3714 | { | |
5ba2ffba PZ |
3715 | /* |
3716 | * Do not complicate things with the async wake_list while the CPU is | |
3717 | * in hotplug state. | |
3718 | */ | |
3719 | if (!cpu_active(cpu)) | |
3720 | return false; | |
3721 | ||
2ebb1771 MG |
3722 | /* |
3723 | * If the CPU does not share cache, then queue the task on the | |
3724 | * remote rqs wakelist to avoid accessing remote data. | |
3725 | */ | |
3726 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3727 | return true; | |
3728 | ||
3729 | /* | |
3730 | * If the task is descheduling and the only running task on the | |
3731 | * CPU then use the wakelist to offload the task activation to | |
3732 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
3733 | * nr_running is checked to avoid unnecessary task stacking. | |
3734 | */ | |
739f70b4 | 3735 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
3736 | return true; |
3737 | ||
3738 | return false; | |
3739 | } | |
3740 | ||
3741 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3742 | { |
2ebb1771 | 3743 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
3744 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
3745 | return false; | |
3746 | ||
c6e7bd7a | 3747 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3748 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3749 | return true; |
3750 | } | |
3751 | ||
3752 | return false; | |
3753 | } | |
58877d34 PZ |
3754 | |
3755 | #else /* !CONFIG_SMP */ | |
3756 | ||
3757 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3758 | { | |
3759 | return false; | |
3760 | } | |
3761 | ||
d6aa8f85 | 3762 | #endif /* CONFIG_SMP */ |
317f3941 | 3763 | |
b5179ac7 | 3764 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3765 | { |
3766 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3767 | struct rq_flags rf; |
c05fbafb | 3768 | |
2ebb1771 | 3769 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3770 | return; |
317f3941 | 3771 | |
8a8c69c3 | 3772 | rq_lock(rq, &rf); |
77558e4d | 3773 | update_rq_clock(rq); |
d8ac8971 | 3774 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3775 | rq_unlock(rq, &rf); |
9ed3811a TH |
3776 | } |
3777 | ||
43295d73 TG |
3778 | /* |
3779 | * Invoked from try_to_wake_up() to check whether the task can be woken up. | |
3780 | * | |
3781 | * The caller holds p::pi_lock if p != current or has preemption | |
3782 | * disabled when p == current. | |
5f220be2 TG |
3783 | * |
3784 | * The rules of PREEMPT_RT saved_state: | |
3785 | * | |
3786 | * The related locking code always holds p::pi_lock when updating | |
3787 | * p::saved_state, which means the code is fully serialized in both cases. | |
3788 | * | |
3789 | * The lock wait and lock wakeups happen via TASK_RTLOCK_WAIT. No other | |
3790 | * bits set. This allows to distinguish all wakeup scenarios. | |
43295d73 TG |
3791 | */ |
3792 | static __always_inline | |
3793 | bool ttwu_state_match(struct task_struct *p, unsigned int state, int *success) | |
3794 | { | |
5f220be2 TG |
3795 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)) { |
3796 | WARN_ON_ONCE((state & TASK_RTLOCK_WAIT) && | |
3797 | state != TASK_RTLOCK_WAIT); | |
3798 | } | |
3799 | ||
43295d73 TG |
3800 | if (READ_ONCE(p->__state) & state) { |
3801 | *success = 1; | |
3802 | return true; | |
3803 | } | |
5f220be2 TG |
3804 | |
3805 | #ifdef CONFIG_PREEMPT_RT | |
3806 | /* | |
3807 | * Saved state preserves the task state across blocking on | |
3808 | * an RT lock. If the state matches, set p::saved_state to | |
3809 | * TASK_RUNNING, but do not wake the task because it waits | |
3810 | * for a lock wakeup. Also indicate success because from | |
3811 | * the regular waker's point of view this has succeeded. | |
3812 | * | |
3813 | * After acquiring the lock the task will restore p::__state | |
3814 | * from p::saved_state which ensures that the regular | |
3815 | * wakeup is not lost. The restore will also set | |
3816 | * p::saved_state to TASK_RUNNING so any further tests will | |
3817 | * not result in false positives vs. @success | |
3818 | */ | |
3819 | if (p->saved_state & state) { | |
3820 | p->saved_state = TASK_RUNNING; | |
3821 | *success = 1; | |
3822 | } | |
3823 | #endif | |
43295d73 TG |
3824 | return false; |
3825 | } | |
3826 | ||
8643cda5 PZ |
3827 | /* |
3828 | * Notes on Program-Order guarantees on SMP systems. | |
3829 | * | |
3830 | * MIGRATION | |
3831 | * | |
3832 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3833 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3834 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3835 | * |
3836 | * For migration (of runnable tasks) this is provided by the following means: | |
3837 | * | |
3838 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3839 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3840 | * rq(c1)->lock (if not at the same time, then in that order). | |
3841 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3842 | * | |
7696f991 | 3843 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3844 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3845 | * |
3846 | * Example: | |
3847 | * | |
3848 | * CPU0 CPU1 CPU2 | |
3849 | * | |
3850 | * LOCK rq(0)->lock | |
3851 | * sched-out X | |
3852 | * sched-in Y | |
3853 | * UNLOCK rq(0)->lock | |
3854 | * | |
3855 | * LOCK rq(0)->lock // orders against CPU0 | |
3856 | * dequeue X | |
3857 | * UNLOCK rq(0)->lock | |
3858 | * | |
3859 | * LOCK rq(1)->lock | |
3860 | * enqueue X | |
3861 | * UNLOCK rq(1)->lock | |
3862 | * | |
3863 | * LOCK rq(1)->lock // orders against CPU2 | |
3864 | * sched-out Z | |
3865 | * sched-in X | |
3866 | * UNLOCK rq(1)->lock | |
3867 | * | |
3868 | * | |
3869 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3870 | * | |
3871 | * For blocking we (obviously) need to provide the same guarantee as for | |
3872 | * migration. However the means are completely different as there is no lock | |
3873 | * chain to provide order. Instead we do: | |
3874 | * | |
58877d34 PZ |
3875 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3876 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3877 | * |
3878 | * Example: | |
3879 | * | |
3880 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
3881 | * | |
3882 | * LOCK rq(0)->lock LOCK X->pi_lock | |
3883 | * dequeue X | |
3884 | * sched-out X | |
3885 | * smp_store_release(X->on_cpu, 0); | |
3886 | * | |
1f03e8d2 | 3887 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
3888 | * X->state = WAKING |
3889 | * set_task_cpu(X,2) | |
3890 | * | |
3891 | * LOCK rq(2)->lock | |
3892 | * enqueue X | |
3893 | * X->state = RUNNING | |
3894 | * UNLOCK rq(2)->lock | |
3895 | * | |
3896 | * LOCK rq(2)->lock // orders against CPU1 | |
3897 | * sched-out Z | |
3898 | * sched-in X | |
3899 | * UNLOCK rq(2)->lock | |
3900 | * | |
3901 | * UNLOCK X->pi_lock | |
3902 | * UNLOCK rq(0)->lock | |
3903 | * | |
3904 | * | |
7696f991 AP |
3905 | * However, for wakeups there is a second guarantee we must provide, namely we |
3906 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
3907 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
3908 | */ |
3909 | ||
9ed3811a | 3910 | /** |
1da177e4 | 3911 | * try_to_wake_up - wake up a thread |
9ed3811a | 3912 | * @p: the thread to be awakened |
1da177e4 | 3913 | * @state: the mask of task states that can be woken |
9ed3811a | 3914 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 3915 | * |
58877d34 PZ |
3916 | * Conceptually does: |
3917 | * | |
3918 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 3919 | * |
a2250238 PZ |
3920 | * If the task was not queued/runnable, also place it back on a runqueue. |
3921 | * | |
58877d34 PZ |
3922 | * This function is atomic against schedule() which would dequeue the task. |
3923 | * | |
3924 | * It issues a full memory barrier before accessing @p->state, see the comment | |
3925 | * with set_current_state(). | |
a2250238 | 3926 | * |
58877d34 | 3927 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 3928 | * |
58877d34 PZ |
3929 | * Relies on p->pi_lock stabilizing: |
3930 | * - p->sched_class | |
3931 | * - p->cpus_ptr | |
3932 | * - p->sched_task_group | |
3933 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
3934 | * | |
3935 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
3936 | * Takes rq->lock in: | |
3937 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
3938 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
3939 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
3940 | * | |
3941 | * As a consequence we race really badly with just about everything. See the | |
3942 | * many memory barriers and their comments for details. | |
7696f991 | 3943 | * |
a2250238 PZ |
3944 | * Return: %true if @p->state changes (an actual wakeup was done), |
3945 | * %false otherwise. | |
1da177e4 | 3946 | */ |
e4a52bcb PZ |
3947 | static int |
3948 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 3949 | { |
1da177e4 | 3950 | unsigned long flags; |
c05fbafb | 3951 | int cpu, success = 0; |
2398f2c6 | 3952 | |
e3d85487 | 3953 | preempt_disable(); |
aacedf26 PZ |
3954 | if (p == current) { |
3955 | /* | |
3956 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
3957 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 3958 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
3959 | * without taking any locks. |
3960 | * | |
3961 | * In particular: | |
3962 | * - we rely on Program-Order guarantees for all the ordering, | |
3963 | * - we're serialized against set_special_state() by virtue of | |
3964 | * it disabling IRQs (this allows not taking ->pi_lock). | |
3965 | */ | |
43295d73 | 3966 | if (!ttwu_state_match(p, state, &success)) |
e3d85487 | 3967 | goto out; |
aacedf26 | 3968 | |
aacedf26 | 3969 | trace_sched_waking(p); |
2f064a59 | 3970 | WRITE_ONCE(p->__state, TASK_RUNNING); |
aacedf26 PZ |
3971 | trace_sched_wakeup(p); |
3972 | goto out; | |
3973 | } | |
3974 | ||
e0acd0a6 ON |
3975 | /* |
3976 | * If we are going to wake up a thread waiting for CONDITION we | |
3977 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
3978 | * reordered with p->state check below. This pairs with smp_store_mb() |
3979 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 3980 | */ |
013fdb80 | 3981 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 3982 | smp_mb__after_spinlock(); |
43295d73 | 3983 | if (!ttwu_state_match(p, state, &success)) |
aacedf26 | 3984 | goto unlock; |
1da177e4 | 3985 | |
fbd705a0 PZ |
3986 | trace_sched_waking(p); |
3987 | ||
135e8c92 BS |
3988 | /* |
3989 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
3990 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
3991 | * in smp_cond_load_acquire() below. | |
3992 | * | |
3d85b270 AP |
3993 | * sched_ttwu_pending() try_to_wake_up() |
3994 | * STORE p->on_rq = 1 LOAD p->state | |
3995 | * UNLOCK rq->lock | |
3996 | * | |
3997 | * __schedule() (switch to task 'p') | |
3998 | * LOCK rq->lock smp_rmb(); | |
3999 | * smp_mb__after_spinlock(); | |
4000 | * UNLOCK rq->lock | |
135e8c92 BS |
4001 | * |
4002 | * [task p] | |
3d85b270 | 4003 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 4004 | * |
3d85b270 AP |
4005 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4006 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
4007 | * |
4008 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
4009 | */ |
4010 | smp_rmb(); | |
58877d34 | 4011 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 4012 | goto unlock; |
1da177e4 | 4013 | |
1da177e4 | 4014 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
4015 | /* |
4016 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
4017 | * possible to, falsely, observe p->on_cpu == 0. | |
4018 | * | |
4019 | * One must be running (->on_cpu == 1) in order to remove oneself | |
4020 | * from the runqueue. | |
4021 | * | |
3d85b270 AP |
4022 | * __schedule() (switch to task 'p') try_to_wake_up() |
4023 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
4024 | * UNLOCK rq->lock | |
4025 | * | |
4026 | * __schedule() (put 'p' to sleep) | |
4027 | * LOCK rq->lock smp_rmb(); | |
4028 | * smp_mb__after_spinlock(); | |
4029 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 4030 | * |
3d85b270 AP |
4031 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4032 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
4033 | * |
4034 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
4035 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
4036 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 4037 | */ |
dbfb089d PZ |
4038 | smp_acquire__after_ctrl_dep(); |
4039 | ||
4040 | /* | |
4041 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
4042 | * == 0), which means we need to do an enqueue, change p->state to | |
4043 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
4044 | * enqueue, such as ttwu_queue_wakelist(). | |
4045 | */ | |
2f064a59 | 4046 | WRITE_ONCE(p->__state, TASK_WAKING); |
ecf7d01c | 4047 | |
c6e7bd7a PZ |
4048 | /* |
4049 | * If the owning (remote) CPU is still in the middle of schedule() with | |
4050 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
4051 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
4052 | * let the waker make forward progress. This is safe because IRQs are | |
4053 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
4054 | * |
4055 | * Ensure we load task_cpu(p) after p->on_cpu: | |
4056 | * | |
4057 | * set_task_cpu(p, cpu); | |
4058 | * STORE p->cpu = @cpu | |
4059 | * __schedule() (switch to task 'p') | |
4060 | * LOCK rq->lock | |
4061 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
4062 | * STORE p->on_cpu = 1 LOAD p->cpu | |
4063 | * | |
4064 | * to ensure we observe the correct CPU on which the task is currently | |
4065 | * scheduling. | |
c6e7bd7a | 4066 | */ |
b6e13e85 | 4067 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 4068 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
4069 | goto unlock; |
4070 | ||
e9c84311 | 4071 | /* |
d1ccc66d | 4072 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 4073 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 4074 | * |
31cb1bc0 | 4075 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
4076 | * |
4077 | * This ensures that tasks getting woken will be fully ordered against | |
4078 | * their previous state and preserve Program Order. | |
0970d299 | 4079 | */ |
1f03e8d2 | 4080 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 4081 | |
3aef1551 | 4082 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc | 4083 | if (task_cpu(p) != cpu) { |
ec618b84 PZ |
4084 | if (p->in_iowait) { |
4085 | delayacct_blkio_end(p); | |
4086 | atomic_dec(&task_rq(p)->nr_iowait); | |
4087 | } | |
4088 | ||
f339b9dc | 4089 | wake_flags |= WF_MIGRATED; |
eb414681 | 4090 | psi_ttwu_dequeue(p); |
e4a52bcb | 4091 | set_task_cpu(p, cpu); |
f339b9dc | 4092 | } |
b6e13e85 PZ |
4093 | #else |
4094 | cpu = task_cpu(p); | |
1da177e4 | 4095 | #endif /* CONFIG_SMP */ |
1da177e4 | 4096 | |
b5179ac7 | 4097 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 4098 | unlock: |
013fdb80 | 4099 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
4100 | out: |
4101 | if (success) | |
b6e13e85 | 4102 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 4103 | preempt_enable(); |
1da177e4 LT |
4104 | |
4105 | return success; | |
4106 | } | |
4107 | ||
2beaf328 PM |
4108 | /** |
4109 | * try_invoke_on_locked_down_task - Invoke a function on task in fixed state | |
1b7af295 | 4110 | * @p: Process for which the function is to be invoked, can be @current. |
2beaf328 PM |
4111 | * @func: Function to invoke. |
4112 | * @arg: Argument to function. | |
4113 | * | |
4114 | * If the specified task can be quickly locked into a definite state | |
4115 | * (either sleeping or on a given runqueue), arrange to keep it in that | |
4116 | * state while invoking @func(@arg). This function can use ->on_rq and | |
4117 | * task_curr() to work out what the state is, if required. Given that | |
4118 | * @func can be invoked with a runqueue lock held, it had better be quite | |
4119 | * lightweight. | |
4120 | * | |
4121 | * Returns: | |
4122 | * @false if the task slipped out from under the locks. | |
4123 | * @true if the task was locked onto a runqueue or is sleeping. | |
4124 | * However, @func can override this by returning @false. | |
4125 | */ | |
4126 | bool try_invoke_on_locked_down_task(struct task_struct *p, bool (*func)(struct task_struct *t, void *arg), void *arg) | |
4127 | { | |
2beaf328 | 4128 | struct rq_flags rf; |
1b7af295 | 4129 | bool ret = false; |
2beaf328 PM |
4130 | struct rq *rq; |
4131 | ||
1b7af295 | 4132 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
2beaf328 PM |
4133 | if (p->on_rq) { |
4134 | rq = __task_rq_lock(p, &rf); | |
4135 | if (task_rq(p) == rq) | |
4136 | ret = func(p, arg); | |
4137 | rq_unlock(rq, &rf); | |
4138 | } else { | |
2f064a59 | 4139 | switch (READ_ONCE(p->__state)) { |
2beaf328 PM |
4140 | case TASK_RUNNING: |
4141 | case TASK_WAKING: | |
4142 | break; | |
4143 | default: | |
4144 | smp_rmb(); // See smp_rmb() comment in try_to_wake_up(). | |
4145 | if (!p->on_rq) | |
4146 | ret = func(p, arg); | |
4147 | } | |
4148 | } | |
1b7af295 | 4149 | raw_spin_unlock_irqrestore(&p->pi_lock, rf.flags); |
2beaf328 PM |
4150 | return ret; |
4151 | } | |
4152 | ||
50fa610a DH |
4153 | /** |
4154 | * wake_up_process - Wake up a specific process | |
4155 | * @p: The process to be woken up. | |
4156 | * | |
4157 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
4158 | * processes. |
4159 | * | |
4160 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 4161 | * |
7696f991 | 4162 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 4163 | */ |
7ad5b3a5 | 4164 | int wake_up_process(struct task_struct *p) |
1da177e4 | 4165 | { |
9067ac85 | 4166 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 4167 | } |
1da177e4 LT |
4168 | EXPORT_SYMBOL(wake_up_process); |
4169 | ||
7ad5b3a5 | 4170 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
4171 | { |
4172 | return try_to_wake_up(p, state, 0); | |
4173 | } | |
4174 | ||
1da177e4 LT |
4175 | /* |
4176 | * Perform scheduler related setup for a newly forked process p. | |
4177 | * p is forked by current. | |
dd41f596 IM |
4178 | * |
4179 | * __sched_fork() is basic setup used by init_idle() too: | |
4180 | */ | |
5e1576ed | 4181 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4182 | { |
fd2f4419 PZ |
4183 | p->on_rq = 0; |
4184 | ||
4185 | p->se.on_rq = 0; | |
dd41f596 IM |
4186 | p->se.exec_start = 0; |
4187 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 4188 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 4189 | p->se.nr_migrations = 0; |
da7a735e | 4190 | p->se.vruntime = 0; |
fd2f4419 | 4191 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 4192 | |
ad936d86 BP |
4193 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4194 | p->se.cfs_rq = NULL; | |
4195 | #endif | |
4196 | ||
6cfb0d5d | 4197 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 4198 | /* Even if schedstat is disabled, there should not be garbage */ |
41acab88 | 4199 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 4200 | #endif |
476d139c | 4201 | |
aab03e05 | 4202 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 4203 | init_dl_task_timer(&p->dl); |
209a0cbd | 4204 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 4205 | __dl_clear_params(p); |
aab03e05 | 4206 | |
fa717060 | 4207 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
4208 | p->rt.timeout = 0; |
4209 | p->rt.time_slice = sched_rr_timeslice; | |
4210 | p->rt.on_rq = 0; | |
4211 | p->rt.on_list = 0; | |
476d139c | 4212 | |
e107be36 AK |
4213 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4214 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
4215 | #endif | |
cbee9f88 | 4216 | |
5e1f0f09 MG |
4217 | #ifdef CONFIG_COMPACTION |
4218 | p->capture_control = NULL; | |
4219 | #endif | |
13784475 | 4220 | init_numa_balancing(clone_flags, p); |
a1488664 | 4221 | #ifdef CONFIG_SMP |
8c4890d1 | 4222 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 4223 | p->migration_pending = NULL; |
a1488664 | 4224 | #endif |
dd41f596 IM |
4225 | } |
4226 | ||
2a595721 SD |
4227 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
4228 | ||
1a687c2e | 4229 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 4230 | |
1a687c2e MG |
4231 | void set_numabalancing_state(bool enabled) |
4232 | { | |
4233 | if (enabled) | |
2a595721 | 4234 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 4235 | else |
2a595721 | 4236 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 4237 | } |
54a43d54 AK |
4238 | |
4239 | #ifdef CONFIG_PROC_SYSCTL | |
4240 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 4241 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
4242 | { |
4243 | struct ctl_table t; | |
4244 | int err; | |
2a595721 | 4245 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
4246 | |
4247 | if (write && !capable(CAP_SYS_ADMIN)) | |
4248 | return -EPERM; | |
4249 | ||
4250 | t = *table; | |
4251 | t.data = &state; | |
4252 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4253 | if (err < 0) | |
4254 | return err; | |
4255 | if (write) | |
4256 | set_numabalancing_state(state); | |
4257 | return err; | |
4258 | } | |
4259 | #endif | |
4260 | #endif | |
dd41f596 | 4261 | |
4698f88c JP |
4262 | #ifdef CONFIG_SCHEDSTATS |
4263 | ||
cb251765 MG |
4264 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4265 | ||
cb251765 MG |
4266 | static void set_schedstats(bool enabled) |
4267 | { | |
4268 | if (enabled) | |
4269 | static_branch_enable(&sched_schedstats); | |
4270 | else | |
4271 | static_branch_disable(&sched_schedstats); | |
4272 | } | |
4273 | ||
4274 | void force_schedstat_enabled(void) | |
4275 | { | |
4276 | if (!schedstat_enabled()) { | |
4277 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
4278 | static_branch_enable(&sched_schedstats); | |
4279 | } | |
4280 | } | |
4281 | ||
4282 | static int __init setup_schedstats(char *str) | |
4283 | { | |
4284 | int ret = 0; | |
4285 | if (!str) | |
4286 | goto out; | |
4287 | ||
4288 | if (!strcmp(str, "enable")) { | |
1faa491a | 4289 | set_schedstats(true); |
cb251765 MG |
4290 | ret = 1; |
4291 | } else if (!strcmp(str, "disable")) { | |
1faa491a | 4292 | set_schedstats(false); |
cb251765 MG |
4293 | ret = 1; |
4294 | } | |
4295 | out: | |
4296 | if (!ret) | |
4297 | pr_warn("Unable to parse schedstats=\n"); | |
4298 | ||
4299 | return ret; | |
4300 | } | |
4301 | __setup("schedstats=", setup_schedstats); | |
4302 | ||
4303 | #ifdef CONFIG_PROC_SYSCTL | |
32927393 CH |
4304 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
4305 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
4306 | { |
4307 | struct ctl_table t; | |
4308 | int err; | |
4309 | int state = static_branch_likely(&sched_schedstats); | |
4310 | ||
4311 | if (write && !capable(CAP_SYS_ADMIN)) | |
4312 | return -EPERM; | |
4313 | ||
4314 | t = *table; | |
4315 | t.data = &state; | |
4316 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4317 | if (err < 0) | |
4318 | return err; | |
4319 | if (write) | |
4320 | set_schedstats(state); | |
4321 | return err; | |
4322 | } | |
4698f88c | 4323 | #endif /* CONFIG_PROC_SYSCTL */ |
4698f88c | 4324 | #endif /* CONFIG_SCHEDSTATS */ |
dd41f596 IM |
4325 | |
4326 | /* | |
4327 | * fork()/clone()-time setup: | |
4328 | */ | |
aab03e05 | 4329 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4330 | { |
0122ec5b | 4331 | unsigned long flags; |
dd41f596 | 4332 | |
5e1576ed | 4333 | __sched_fork(clone_flags, p); |
06b83b5f | 4334 | /* |
7dc603c9 | 4335 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
4336 | * nobody will actually run it, and a signal or other external |
4337 | * event cannot wake it up and insert it on the runqueue either. | |
4338 | */ | |
2f064a59 | 4339 | p->__state = TASK_NEW; |
dd41f596 | 4340 | |
c350a04e MG |
4341 | /* |
4342 | * Make sure we do not leak PI boosting priority to the child. | |
4343 | */ | |
4344 | p->prio = current->normal_prio; | |
4345 | ||
e8f14172 PB |
4346 | uclamp_fork(p); |
4347 | ||
b9dc29e7 MG |
4348 | /* |
4349 | * Revert to default priority/policy on fork if requested. | |
4350 | */ | |
4351 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 4352 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 4353 | p->policy = SCHED_NORMAL; |
6c697bdf | 4354 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
4355 | p->rt_priority = 0; |
4356 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
4357 | p->static_prio = NICE_TO_PRIO(0); | |
4358 | ||
f558c2b8 | 4359 | p->prio = p->normal_prio = p->static_prio; |
9059393e | 4360 | set_load_weight(p, false); |
6c697bdf | 4361 | |
b9dc29e7 MG |
4362 | /* |
4363 | * We don't need the reset flag anymore after the fork. It has | |
4364 | * fulfilled its duty: | |
4365 | */ | |
4366 | p->sched_reset_on_fork = 0; | |
4367 | } | |
ca94c442 | 4368 | |
af0fffd9 | 4369 | if (dl_prio(p->prio)) |
aab03e05 | 4370 | return -EAGAIN; |
af0fffd9 | 4371 | else if (rt_prio(p->prio)) |
aab03e05 | 4372 | p->sched_class = &rt_sched_class; |
af0fffd9 | 4373 | else |
2ddbf952 | 4374 | p->sched_class = &fair_sched_class; |
b29739f9 | 4375 | |
7dc603c9 | 4376 | init_entity_runnable_average(&p->se); |
cd29fe6f | 4377 | |
86951599 PZ |
4378 | /* |
4379 | * The child is not yet in the pid-hash so no cgroup attach races, | |
4380 | * and the cgroup is pinned to this child due to cgroup_fork() | |
4381 | * is ran before sched_fork(). | |
4382 | * | |
4383 | * Silence PROVE_RCU. | |
4384 | */ | |
0122ec5b | 4385 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
ce3614da | 4386 | rseq_migrate(p); |
e210bffd | 4387 | /* |
d1ccc66d | 4388 | * We're setting the CPU for the first time, we don't migrate, |
e210bffd PZ |
4389 | * so use __set_task_cpu(). |
4390 | */ | |
af0fffd9 | 4391 | __set_task_cpu(p, smp_processor_id()); |
e210bffd PZ |
4392 | if (p->sched_class->task_fork) |
4393 | p->sched_class->task_fork(p); | |
0122ec5b | 4394 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 4395 | |
f6db8347 | 4396 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 4397 | if (likely(sched_info_on())) |
52f17b6c | 4398 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 4399 | #endif |
3ca7a440 PZ |
4400 | #if defined(CONFIG_SMP) |
4401 | p->on_cpu = 0; | |
4866cde0 | 4402 | #endif |
01028747 | 4403 | init_task_preempt_count(p); |
806c09a7 | 4404 | #ifdef CONFIG_SMP |
917b627d | 4405 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 4406 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 4407 | #endif |
aab03e05 | 4408 | return 0; |
1da177e4 LT |
4409 | } |
4410 | ||
13685c4a QY |
4411 | void sched_post_fork(struct task_struct *p) |
4412 | { | |
4413 | uclamp_post_fork(p); | |
4414 | } | |
4415 | ||
332ac17e DF |
4416 | unsigned long to_ratio(u64 period, u64 runtime) |
4417 | { | |
4418 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 4419 | return BW_UNIT; |
332ac17e DF |
4420 | |
4421 | /* | |
4422 | * Doing this here saves a lot of checks in all | |
4423 | * the calling paths, and returning zero seems | |
4424 | * safe for them anyway. | |
4425 | */ | |
4426 | if (period == 0) | |
4427 | return 0; | |
4428 | ||
c52f14d3 | 4429 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
4430 | } |
4431 | ||
1da177e4 LT |
4432 | /* |
4433 | * wake_up_new_task - wake up a newly created task for the first time. | |
4434 | * | |
4435 | * This function will do some initial scheduler statistics housekeeping | |
4436 | * that must be done for every newly created context, then puts the task | |
4437 | * on the runqueue and wakes it. | |
4438 | */ | |
3e51e3ed | 4439 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 4440 | { |
eb580751 | 4441 | struct rq_flags rf; |
dd41f596 | 4442 | struct rq *rq; |
fabf318e | 4443 | |
eb580751 | 4444 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
2f064a59 | 4445 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fabf318e PZ |
4446 | #ifdef CONFIG_SMP |
4447 | /* | |
4448 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 4449 | * - cpus_ptr can change in the fork path |
d1ccc66d | 4450 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
4451 | * |
4452 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
4453 | * as we're not fully set-up yet. | |
fabf318e | 4454 | */ |
32e839dd | 4455 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 4456 | rseq_migrate(p); |
3aef1551 | 4457 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 4458 | #endif |
b7fa30c9 | 4459 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 4460 | update_rq_clock(rq); |
d0fe0b9c | 4461 | post_init_entity_util_avg(p); |
0017d735 | 4462 | |
7a57f32a | 4463 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 4464 | trace_sched_wakeup_new(p); |
a7558e01 | 4465 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 4466 | #ifdef CONFIG_SMP |
0aaafaab PZ |
4467 | if (p->sched_class->task_woken) { |
4468 | /* | |
b19a888c | 4469 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
4470 | * drop it. |
4471 | */ | |
d8ac8971 | 4472 | rq_unpin_lock(rq, &rf); |
efbbd05a | 4473 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 4474 | rq_repin_lock(rq, &rf); |
0aaafaab | 4475 | } |
9a897c5a | 4476 | #endif |
eb580751 | 4477 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4478 | } |
4479 | ||
e107be36 AK |
4480 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4481 | ||
b7203428 | 4482 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 4483 | |
2ecd9d29 PZ |
4484 | void preempt_notifier_inc(void) |
4485 | { | |
b7203428 | 4486 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
4487 | } |
4488 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
4489 | ||
4490 | void preempt_notifier_dec(void) | |
4491 | { | |
b7203428 | 4492 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
4493 | } |
4494 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
4495 | ||
e107be36 | 4496 | /** |
80dd99b3 | 4497 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 4498 | * @notifier: notifier struct to register |
e107be36 AK |
4499 | */ |
4500 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
4501 | { | |
b7203428 | 4502 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
4503 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
4504 | ||
e107be36 AK |
4505 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
4506 | } | |
4507 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
4508 | ||
4509 | /** | |
4510 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 4511 | * @notifier: notifier struct to unregister |
e107be36 | 4512 | * |
d84525a8 | 4513 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
4514 | */ |
4515 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
4516 | { | |
4517 | hlist_del(¬ifier->link); | |
4518 | } | |
4519 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
4520 | ||
1cde2930 | 4521 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4522 | { |
4523 | struct preempt_notifier *notifier; | |
e107be36 | 4524 | |
b67bfe0d | 4525 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4526 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
4527 | } | |
4528 | ||
1cde2930 PZ |
4529 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
4530 | { | |
b7203428 | 4531 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4532 | __fire_sched_in_preempt_notifiers(curr); |
4533 | } | |
4534 | ||
e107be36 | 4535 | static void |
1cde2930 PZ |
4536 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4537 | struct task_struct *next) | |
e107be36 AK |
4538 | { |
4539 | struct preempt_notifier *notifier; | |
e107be36 | 4540 | |
b67bfe0d | 4541 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4542 | notifier->ops->sched_out(notifier, next); |
4543 | } | |
4544 | ||
1cde2930 PZ |
4545 | static __always_inline void |
4546 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
4547 | struct task_struct *next) | |
4548 | { | |
b7203428 | 4549 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4550 | __fire_sched_out_preempt_notifiers(curr, next); |
4551 | } | |
4552 | ||
6d6bc0ad | 4553 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4554 | |
1cde2930 | 4555 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4556 | { |
4557 | } | |
4558 | ||
1cde2930 | 4559 | static inline void |
e107be36 AK |
4560 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4561 | struct task_struct *next) | |
4562 | { | |
4563 | } | |
4564 | ||
6d6bc0ad | 4565 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4566 | |
31cb1bc0 | 4567 | static inline void prepare_task(struct task_struct *next) |
4568 | { | |
4569 | #ifdef CONFIG_SMP | |
4570 | /* | |
4571 | * Claim the task as running, we do this before switching to it | |
4572 | * such that any running task will have this set. | |
58877d34 PZ |
4573 | * |
4574 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 4575 | */ |
58877d34 | 4576 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 4577 | #endif |
4578 | } | |
4579 | ||
4580 | static inline void finish_task(struct task_struct *prev) | |
4581 | { | |
4582 | #ifdef CONFIG_SMP | |
4583 | /* | |
58877d34 PZ |
4584 | * This must be the very last reference to @prev from this CPU. After |
4585 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
4586 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 4587 | * finished. |
4588 | * | |
4589 | * In particular, the load of prev->state in finish_task_switch() must | |
4590 | * happen before this. | |
4591 | * | |
4592 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
4593 | */ | |
4594 | smp_store_release(&prev->on_cpu, 0); | |
4595 | #endif | |
4596 | } | |
4597 | ||
565790d2 PZ |
4598 | #ifdef CONFIG_SMP |
4599 | ||
4600 | static void do_balance_callbacks(struct rq *rq, struct callback_head *head) | |
4601 | { | |
4602 | void (*func)(struct rq *rq); | |
4603 | struct callback_head *next; | |
4604 | ||
5cb9eaa3 | 4605 | lockdep_assert_rq_held(rq); |
565790d2 PZ |
4606 | |
4607 | while (head) { | |
4608 | func = (void (*)(struct rq *))head->func; | |
4609 | next = head->next; | |
4610 | head->next = NULL; | |
4611 | head = next; | |
4612 | ||
4613 | func(rq); | |
4614 | } | |
4615 | } | |
4616 | ||
ae792702 PZ |
4617 | static void balance_push(struct rq *rq); |
4618 | ||
4619 | struct callback_head balance_push_callback = { | |
4620 | .next = NULL, | |
4621 | .func = (void (*)(struct callback_head *))balance_push, | |
4622 | }; | |
4623 | ||
565790d2 PZ |
4624 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) |
4625 | { | |
4626 | struct callback_head *head = rq->balance_callback; | |
4627 | ||
5cb9eaa3 | 4628 | lockdep_assert_rq_held(rq); |
ae792702 | 4629 | if (head) |
565790d2 PZ |
4630 | rq->balance_callback = NULL; |
4631 | ||
4632 | return head; | |
4633 | } | |
4634 | ||
4635 | static void __balance_callbacks(struct rq *rq) | |
4636 | { | |
4637 | do_balance_callbacks(rq, splice_balance_callbacks(rq)); | |
4638 | } | |
4639 | ||
4640 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4641 | { | |
4642 | unsigned long flags; | |
4643 | ||
4644 | if (unlikely(head)) { | |
5cb9eaa3 | 4645 | raw_spin_rq_lock_irqsave(rq, flags); |
565790d2 | 4646 | do_balance_callbacks(rq, head); |
5cb9eaa3 | 4647 | raw_spin_rq_unlock_irqrestore(rq, flags); |
565790d2 PZ |
4648 | } |
4649 | } | |
4650 | ||
4651 | #else | |
4652 | ||
4653 | static inline void __balance_callbacks(struct rq *rq) | |
4654 | { | |
4655 | } | |
4656 | ||
4657 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
4658 | { | |
4659 | return NULL; | |
4660 | } | |
4661 | ||
4662 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4663 | { | |
4664 | } | |
4665 | ||
4666 | #endif | |
4667 | ||
269d5992 PZ |
4668 | static inline void |
4669 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4670 | { |
269d5992 PZ |
4671 | /* |
4672 | * Since the runqueue lock will be released by the next | |
4673 | * task (which is an invalid locking op but in the case | |
4674 | * of the scheduler it's an obvious special-case), so we | |
4675 | * do an early lockdep release here: | |
4676 | */ | |
4677 | rq_unpin_lock(rq, rf); | |
9ef7e7e3 | 4678 | spin_release(&__rq_lockp(rq)->dep_map, _THIS_IP_); |
31cb1bc0 | 4679 | #ifdef CONFIG_DEBUG_SPINLOCK |
4680 | /* this is a valid case when another task releases the spinlock */ | |
5cb9eaa3 | 4681 | rq_lockp(rq)->owner = next; |
31cb1bc0 | 4682 | #endif |
269d5992 PZ |
4683 | } |
4684 | ||
4685 | static inline void finish_lock_switch(struct rq *rq) | |
4686 | { | |
31cb1bc0 | 4687 | /* |
4688 | * If we are tracking spinlock dependencies then we have to | |
4689 | * fix up the runqueue lock - which gets 'carried over' from | |
4690 | * prev into current: | |
4691 | */ | |
9ef7e7e3 | 4692 | spin_acquire(&__rq_lockp(rq)->dep_map, 0, 0, _THIS_IP_); |
ae792702 | 4693 | __balance_callbacks(rq); |
5cb9eaa3 | 4694 | raw_spin_rq_unlock_irq(rq); |
31cb1bc0 | 4695 | } |
4696 | ||
325ea10c IM |
4697 | /* |
4698 | * NOP if the arch has not defined these: | |
4699 | */ | |
4700 | ||
4701 | #ifndef prepare_arch_switch | |
4702 | # define prepare_arch_switch(next) do { } while (0) | |
4703 | #endif | |
4704 | ||
4705 | #ifndef finish_arch_post_lock_switch | |
4706 | # define finish_arch_post_lock_switch() do { } while (0) | |
4707 | #endif | |
4708 | ||
5fbda3ec TG |
4709 | static inline void kmap_local_sched_out(void) |
4710 | { | |
4711 | #ifdef CONFIG_KMAP_LOCAL | |
4712 | if (unlikely(current->kmap_ctrl.idx)) | |
4713 | __kmap_local_sched_out(); | |
4714 | #endif | |
4715 | } | |
4716 | ||
4717 | static inline void kmap_local_sched_in(void) | |
4718 | { | |
4719 | #ifdef CONFIG_KMAP_LOCAL | |
4720 | if (unlikely(current->kmap_ctrl.idx)) | |
4721 | __kmap_local_sched_in(); | |
4722 | #endif | |
4723 | } | |
4724 | ||
4866cde0 NP |
4725 | /** |
4726 | * prepare_task_switch - prepare to switch tasks | |
4727 | * @rq: the runqueue preparing to switch | |
421cee29 | 4728 | * @prev: the current task that is being switched out |
4866cde0 NP |
4729 | * @next: the task we are going to switch to. |
4730 | * | |
4731 | * This is called with the rq lock held and interrupts off. It must | |
4732 | * be paired with a subsequent finish_task_switch after the context | |
4733 | * switch. | |
4734 | * | |
4735 | * prepare_task_switch sets up locking and calls architecture specific | |
4736 | * hooks. | |
4737 | */ | |
e107be36 AK |
4738 | static inline void |
4739 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
4740 | struct task_struct *next) | |
4866cde0 | 4741 | { |
0ed557aa | 4742 | kcov_prepare_switch(prev); |
43148951 | 4743 | sched_info_switch(rq, prev, next); |
fe4b04fa | 4744 | perf_event_task_sched_out(prev, next); |
d7822b1e | 4745 | rseq_preempt(prev); |
e107be36 | 4746 | fire_sched_out_preempt_notifiers(prev, next); |
5fbda3ec | 4747 | kmap_local_sched_out(); |
31cb1bc0 | 4748 | prepare_task(next); |
4866cde0 NP |
4749 | prepare_arch_switch(next); |
4750 | } | |
4751 | ||
1da177e4 LT |
4752 | /** |
4753 | * finish_task_switch - clean up after a task-switch | |
4754 | * @prev: the thread we just switched away from. | |
4755 | * | |
4866cde0 NP |
4756 | * finish_task_switch must be called after the context switch, paired |
4757 | * with a prepare_task_switch call before the context switch. | |
4758 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
4759 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
4760 | * |
4761 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 4762 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
4763 | * with the lock held can cause deadlocks; see schedule() for |
4764 | * details.) | |
dfa50b60 ON |
4765 | * |
4766 | * The context switch have flipped the stack from under us and restored the | |
4767 | * local variables which were saved when this task called schedule() in the | |
4768 | * past. prev == current is still correct but we need to recalculate this_rq | |
4769 | * because prev may have moved to another CPU. | |
1da177e4 | 4770 | */ |
dfa50b60 | 4771 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
4772 | __releases(rq->lock) |
4773 | { | |
dfa50b60 | 4774 | struct rq *rq = this_rq(); |
1da177e4 | 4775 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 4776 | long prev_state; |
1da177e4 | 4777 | |
609ca066 PZ |
4778 | /* |
4779 | * The previous task will have left us with a preempt_count of 2 | |
4780 | * because it left us after: | |
4781 | * | |
4782 | * schedule() | |
4783 | * preempt_disable(); // 1 | |
4784 | * __schedule() | |
4785 | * raw_spin_lock_irq(&rq->lock) // 2 | |
4786 | * | |
4787 | * Also, see FORK_PREEMPT_COUNT. | |
4788 | */ | |
e2bf1c4b PZ |
4789 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
4790 | "corrupted preempt_count: %s/%d/0x%x\n", | |
4791 | current->comm, current->pid, preempt_count())) | |
4792 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 4793 | |
1da177e4 LT |
4794 | rq->prev_mm = NULL; |
4795 | ||
4796 | /* | |
4797 | * A task struct has one reference for the use as "current". | |
c394cc9f | 4798 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
4799 | * schedule one last time. The schedule call will never return, and |
4800 | * the scheduled task must drop that reference. | |
95913d97 PZ |
4801 | * |
4802 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 4803 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
4804 | * running on another CPU and we could rave with its RUNNING -> DEAD |
4805 | * transition, resulting in a double drop. | |
1da177e4 | 4806 | */ |
2f064a59 | 4807 | prev_state = READ_ONCE(prev->__state); |
bf9fae9f | 4808 | vtime_task_switch(prev); |
a8d757ef | 4809 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 4810 | finish_task(prev); |
0fdcccfa | 4811 | tick_nohz_task_switch(); |
31cb1bc0 | 4812 | finish_lock_switch(rq); |
01f23e16 | 4813 | finish_arch_post_lock_switch(); |
0ed557aa | 4814 | kcov_finish_switch(current); |
5fbda3ec TG |
4815 | /* |
4816 | * kmap_local_sched_out() is invoked with rq::lock held and | |
4817 | * interrupts disabled. There is no requirement for that, but the | |
4818 | * sched out code does not have an interrupt enabled section. | |
4819 | * Restoring the maps on sched in does not require interrupts being | |
4820 | * disabled either. | |
4821 | */ | |
4822 | kmap_local_sched_in(); | |
e8fa1362 | 4823 | |
e107be36 | 4824 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 4825 | /* |
70216e18 MD |
4826 | * When switching through a kernel thread, the loop in |
4827 | * membarrier_{private,global}_expedited() may have observed that | |
4828 | * kernel thread and not issued an IPI. It is therefore possible to | |
4829 | * schedule between user->kernel->user threads without passing though | |
4830 | * switch_mm(). Membarrier requires a barrier after storing to | |
4831 | * rq->curr, before returning to userspace, so provide them here: | |
4832 | * | |
4833 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
4834 | * provided by mmdrop(), | |
4835 | * - a sync_core for SYNC_CORE. | |
306e0604 | 4836 | */ |
70216e18 MD |
4837 | if (mm) { |
4838 | membarrier_mm_sync_core_before_usermode(mm); | |
1da177e4 | 4839 | mmdrop(mm); |
70216e18 | 4840 | } |
1cef1150 PZ |
4841 | if (unlikely(prev_state == TASK_DEAD)) { |
4842 | if (prev->sched_class->task_dead) | |
4843 | prev->sched_class->task_dead(prev); | |
68f24b08 | 4844 | |
1cef1150 PZ |
4845 | /* |
4846 | * Remove function-return probe instances associated with this | |
4847 | * task and put them back on the free list. | |
4848 | */ | |
4849 | kprobe_flush_task(prev); | |
4850 | ||
4851 | /* Task is done with its stack. */ | |
4852 | put_task_stack(prev); | |
4853 | ||
0ff7b2cf | 4854 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 4855 | } |
99e5ada9 | 4856 | |
dfa50b60 | 4857 | return rq; |
1da177e4 LT |
4858 | } |
4859 | ||
4860 | /** | |
4861 | * schedule_tail - first thing a freshly forked thread must call. | |
4862 | * @prev: the thread we just switched away from. | |
4863 | */ | |
722a9f92 | 4864 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
4865 | __releases(rq->lock) |
4866 | { | |
609ca066 PZ |
4867 | /* |
4868 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
4869 | * finish_task_switch() for details. | |
4870 | * | |
4871 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
4872 | * and the preempt_enable() will end up enabling preemption (on | |
4873 | * PREEMPT_COUNT kernels). | |
4874 | */ | |
4875 | ||
13c2235b | 4876 | finish_task_switch(prev); |
1a43a14a | 4877 | preempt_enable(); |
70b97a7f | 4878 | |
1da177e4 | 4879 | if (current->set_child_tid) |
b488893a | 4880 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
4881 | |
4882 | calculate_sigpending(); | |
1da177e4 LT |
4883 | } |
4884 | ||
4885 | /* | |
dfa50b60 | 4886 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 4887 | */ |
04936948 | 4888 | static __always_inline struct rq * |
70b97a7f | 4889 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 4890 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 4891 | { |
e107be36 | 4892 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 4893 | |
9226d125 ZA |
4894 | /* |
4895 | * For paravirt, this is coupled with an exit in switch_to to | |
4896 | * combine the page table reload and the switch backend into | |
4897 | * one hypercall. | |
4898 | */ | |
224101ed | 4899 | arch_start_context_switch(prev); |
9226d125 | 4900 | |
306e0604 | 4901 | /* |
139d025c PZ |
4902 | * kernel -> kernel lazy + transfer active |
4903 | * user -> kernel lazy + mmgrab() active | |
4904 | * | |
4905 | * kernel -> user switch + mmdrop() active | |
4906 | * user -> user switch | |
306e0604 | 4907 | */ |
139d025c PZ |
4908 | if (!next->mm) { // to kernel |
4909 | enter_lazy_tlb(prev->active_mm, next); | |
4910 | ||
4911 | next->active_mm = prev->active_mm; | |
4912 | if (prev->mm) // from user | |
4913 | mmgrab(prev->active_mm); | |
4914 | else | |
4915 | prev->active_mm = NULL; | |
4916 | } else { // to user | |
227a4aad | 4917 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
4918 | /* |
4919 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 4920 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
4921 | * |
4922 | * The below provides this either through switch_mm(), or in | |
4923 | * case 'prev->active_mm == next->mm' through | |
4924 | * finish_task_switch()'s mmdrop(). | |
4925 | */ | |
139d025c | 4926 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 4927 | |
139d025c PZ |
4928 | if (!prev->mm) { // from kernel |
4929 | /* will mmdrop() in finish_task_switch(). */ | |
4930 | rq->prev_mm = prev->active_mm; | |
4931 | prev->active_mm = NULL; | |
4932 | } | |
1da177e4 | 4933 | } |
92509b73 | 4934 | |
cb42c9a3 | 4935 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 4936 | |
269d5992 | 4937 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
4938 | |
4939 | /* Here we just switch the register state and the stack. */ | |
4940 | switch_to(prev, next, prev); | |
dd41f596 | 4941 | barrier(); |
dfa50b60 ON |
4942 | |
4943 | return finish_task_switch(prev); | |
1da177e4 LT |
4944 | } |
4945 | ||
4946 | /* | |
1c3e8264 | 4947 | * nr_running and nr_context_switches: |
1da177e4 LT |
4948 | * |
4949 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 4950 | * threads, total number of context switches performed since bootup. |
1da177e4 | 4951 | */ |
01aee8fd | 4952 | unsigned int nr_running(void) |
1da177e4 | 4953 | { |
01aee8fd | 4954 | unsigned int i, sum = 0; |
1da177e4 LT |
4955 | |
4956 | for_each_online_cpu(i) | |
4957 | sum += cpu_rq(i)->nr_running; | |
4958 | ||
4959 | return sum; | |
f711f609 | 4960 | } |
1da177e4 | 4961 | |
2ee507c4 | 4962 | /* |
d1ccc66d | 4963 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
4964 | * |
4965 | * Caution: this function does not check that the caller has disabled | |
4966 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
4967 | * race. The caller is responsible to use it correctly, for example: | |
4968 | * | |
dfcb245e | 4969 | * - from a non-preemptible section (of course) |
00cc1633 DD |
4970 | * |
4971 | * - from a thread that is bound to a single CPU | |
4972 | * | |
4973 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
4974 | */ |
4975 | bool single_task_running(void) | |
4976 | { | |
00cc1633 | 4977 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
4978 | } |
4979 | EXPORT_SYMBOL(single_task_running); | |
4980 | ||
1da177e4 | 4981 | unsigned long long nr_context_switches(void) |
46cb4b7c | 4982 | { |
cc94abfc SR |
4983 | int i; |
4984 | unsigned long long sum = 0; | |
46cb4b7c | 4985 | |
0a945022 | 4986 | for_each_possible_cpu(i) |
1da177e4 | 4987 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 4988 | |
1da177e4 LT |
4989 | return sum; |
4990 | } | |
483b4ee6 | 4991 | |
145d952a DL |
4992 | /* |
4993 | * Consumers of these two interfaces, like for example the cpuidle menu | |
4994 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
4995 | * for a CPU that has IO-wait which might not even end up running the task when | |
4996 | * it does become runnable. | |
4997 | */ | |
4998 | ||
8fc2858e | 4999 | unsigned int nr_iowait_cpu(int cpu) |
145d952a DL |
5000 | { |
5001 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
5002 | } | |
5003 | ||
e33a9bba | 5004 | /* |
b19a888c | 5005 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
5006 | * |
5007 | * The idea behind IO-wait account is to account the idle time that we could | |
5008 | * have spend running if it were not for IO. That is, if we were to improve the | |
5009 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
5010 | * | |
5011 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
5012 | * idle time as IO-wait, because if the storage were faster, it could've been | |
5013 | * running and we'd not be idle. | |
5014 | * | |
5015 | * This has been extended to SMP, by doing the same for each CPU. This however | |
5016 | * is broken. | |
5017 | * | |
5018 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
5019 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
5020 | * though, if the storage were faster, both could've ran at the same time, | |
5021 | * utilising both CPUs. | |
5022 | * | |
5023 | * This means, that when looking globally, the current IO-wait accounting on | |
5024 | * SMP is a lower bound, by reason of under accounting. | |
5025 | * | |
5026 | * Worse, since the numbers are provided per CPU, they are sometimes | |
5027 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
5028 | * associated with any one particular CPU, it can wake to another CPU than it | |
5029 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
5030 | * | |
5031 | * Task CPU affinities can make all that even more 'interesting'. | |
5032 | */ | |
5033 | ||
97455168 | 5034 | unsigned int nr_iowait(void) |
1da177e4 | 5035 | { |
97455168 | 5036 | unsigned int i, sum = 0; |
483b4ee6 | 5037 | |
0a945022 | 5038 | for_each_possible_cpu(i) |
145d952a | 5039 | sum += nr_iowait_cpu(i); |
46cb4b7c | 5040 | |
1da177e4 LT |
5041 | return sum; |
5042 | } | |
483b4ee6 | 5043 | |
dd41f596 | 5044 | #ifdef CONFIG_SMP |
8a0be9ef | 5045 | |
46cb4b7c | 5046 | /* |
38022906 PZ |
5047 | * sched_exec - execve() is a valuable balancing opportunity, because at |
5048 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 5049 | */ |
38022906 | 5050 | void sched_exec(void) |
46cb4b7c | 5051 | { |
38022906 | 5052 | struct task_struct *p = current; |
1da177e4 | 5053 | unsigned long flags; |
0017d735 | 5054 | int dest_cpu; |
46cb4b7c | 5055 | |
8f42ced9 | 5056 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 5057 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
5058 | if (dest_cpu == smp_processor_id()) |
5059 | goto unlock; | |
38022906 | 5060 | |
8f42ced9 | 5061 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 5062 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 5063 | |
8f42ced9 PZ |
5064 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5065 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
5066 | return; |
5067 | } | |
0017d735 | 5068 | unlock: |
8f42ced9 | 5069 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 5070 | } |
dd41f596 | 5071 | |
1da177e4 LT |
5072 | #endif |
5073 | ||
1da177e4 | 5074 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 5075 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
5076 | |
5077 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 5078 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 5079 | |
6075620b GG |
5080 | /* |
5081 | * The function fair_sched_class.update_curr accesses the struct curr | |
5082 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
5083 | * we observe a high rate of cache misses in practice. | |
5084 | * Prefetching this data results in improved performance. | |
5085 | */ | |
5086 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
5087 | { | |
5088 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
5089 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
5090 | #else | |
5091 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
5092 | #endif | |
5093 | prefetch(curr); | |
5094 | prefetch(&curr->exec_start); | |
5095 | } | |
5096 | ||
c5f8d995 HS |
5097 | /* |
5098 | * Return accounted runtime for the task. | |
5099 | * In case the task is currently running, return the runtime plus current's | |
5100 | * pending runtime that have not been accounted yet. | |
5101 | */ | |
5102 | unsigned long long task_sched_runtime(struct task_struct *p) | |
5103 | { | |
eb580751 | 5104 | struct rq_flags rf; |
c5f8d995 | 5105 | struct rq *rq; |
6e998916 | 5106 | u64 ns; |
c5f8d995 | 5107 | |
911b2898 PZ |
5108 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
5109 | /* | |
97fb7a0a | 5110 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
5111 | * So we have a optimization chance when the task's delta_exec is 0. |
5112 | * Reading ->on_cpu is racy, but this is ok. | |
5113 | * | |
d1ccc66d IM |
5114 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
5115 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 5116 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
5117 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
5118 | * been accounted, so we're correct here as well. | |
911b2898 | 5119 | */ |
da0c1e65 | 5120 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
5121 | return p->se.sum_exec_runtime; |
5122 | #endif | |
5123 | ||
eb580751 | 5124 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
5125 | /* |
5126 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
5127 | * project cycles that may never be accounted to this | |
5128 | * thread, breaking clock_gettime(). | |
5129 | */ | |
5130 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 5131 | prefetch_curr_exec_start(p); |
6e998916 SG |
5132 | update_rq_clock(rq); |
5133 | p->sched_class->update_curr(rq); | |
5134 | } | |
5135 | ns = p->se.sum_exec_runtime; | |
eb580751 | 5136 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
5137 | |
5138 | return ns; | |
5139 | } | |
48f24c4d | 5140 | |
c006fac5 PT |
5141 | #ifdef CONFIG_SCHED_DEBUG |
5142 | static u64 cpu_resched_latency(struct rq *rq) | |
5143 | { | |
5144 | int latency_warn_ms = READ_ONCE(sysctl_resched_latency_warn_ms); | |
5145 | u64 resched_latency, now = rq_clock(rq); | |
5146 | static bool warned_once; | |
5147 | ||
5148 | if (sysctl_resched_latency_warn_once && warned_once) | |
5149 | return 0; | |
5150 | ||
5151 | if (!need_resched() || !latency_warn_ms) | |
5152 | return 0; | |
5153 | ||
5154 | if (system_state == SYSTEM_BOOTING) | |
5155 | return 0; | |
5156 | ||
5157 | if (!rq->last_seen_need_resched_ns) { | |
5158 | rq->last_seen_need_resched_ns = now; | |
5159 | rq->ticks_without_resched = 0; | |
5160 | return 0; | |
5161 | } | |
5162 | ||
5163 | rq->ticks_without_resched++; | |
5164 | resched_latency = now - rq->last_seen_need_resched_ns; | |
5165 | if (resched_latency <= latency_warn_ms * NSEC_PER_MSEC) | |
5166 | return 0; | |
5167 | ||
5168 | warned_once = true; | |
5169 | ||
5170 | return resched_latency; | |
5171 | } | |
5172 | ||
5173 | static int __init setup_resched_latency_warn_ms(char *str) | |
5174 | { | |
5175 | long val; | |
5176 | ||
5177 | if ((kstrtol(str, 0, &val))) { | |
5178 | pr_warn("Unable to set resched_latency_warn_ms\n"); | |
5179 | return 1; | |
5180 | } | |
5181 | ||
5182 | sysctl_resched_latency_warn_ms = val; | |
5183 | return 1; | |
5184 | } | |
5185 | __setup("resched_latency_warn_ms=", setup_resched_latency_warn_ms); | |
5186 | #else | |
5187 | static inline u64 cpu_resched_latency(struct rq *rq) { return 0; } | |
5188 | #endif /* CONFIG_SCHED_DEBUG */ | |
5189 | ||
7835b98b CL |
5190 | /* |
5191 | * This function gets called by the timer code, with HZ frequency. | |
5192 | * We call it with interrupts disabled. | |
7835b98b CL |
5193 | */ |
5194 | void scheduler_tick(void) | |
5195 | { | |
7835b98b CL |
5196 | int cpu = smp_processor_id(); |
5197 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5198 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 5199 | struct rq_flags rf; |
b4eccf5f | 5200 | unsigned long thermal_pressure; |
c006fac5 | 5201 | u64 resched_latency; |
3e51f33f | 5202 | |
1567c3e3 | 5203 | arch_scale_freq_tick(); |
3e51f33f | 5204 | sched_clock_tick(); |
dd41f596 | 5205 | |
8a8c69c3 PZ |
5206 | rq_lock(rq, &rf); |
5207 | ||
3e51f33f | 5208 | update_rq_clock(rq); |
b4eccf5f | 5209 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 5210 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 5211 | curr->sched_class->task_tick(rq, curr, 0); |
c006fac5 PT |
5212 | if (sched_feat(LATENCY_WARN)) |
5213 | resched_latency = cpu_resched_latency(rq); | |
3289bdb4 | 5214 | calc_global_load_tick(rq); |
8a8c69c3 PZ |
5215 | |
5216 | rq_unlock(rq, &rf); | |
7835b98b | 5217 | |
c006fac5 PT |
5218 | if (sched_feat(LATENCY_WARN) && resched_latency) |
5219 | resched_latency_warn(cpu, resched_latency); | |
5220 | ||
e9d2b064 | 5221 | perf_event_task_tick(); |
e220d2dc | 5222 | |
e418e1c2 | 5223 | #ifdef CONFIG_SMP |
6eb57e0d | 5224 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 5225 | trigger_load_balance(rq); |
e418e1c2 | 5226 | #endif |
1da177e4 LT |
5227 | } |
5228 | ||
265f22a9 | 5229 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
5230 | |
5231 | struct tick_work { | |
5232 | int cpu; | |
b55bd585 | 5233 | atomic_t state; |
d84b3131 FW |
5234 | struct delayed_work work; |
5235 | }; | |
b55bd585 PM |
5236 | /* Values for ->state, see diagram below. */ |
5237 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
5238 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
5239 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
5240 | ||
5241 | /* | |
5242 | * State diagram for ->state: | |
5243 | * | |
5244 | * | |
5245 | * TICK_SCHED_REMOTE_OFFLINE | |
5246 | * | ^ | |
5247 | * | | | |
5248 | * | | sched_tick_remote() | |
5249 | * | | | |
5250 | * | | | |
5251 | * +--TICK_SCHED_REMOTE_OFFLINING | |
5252 | * | ^ | |
5253 | * | | | |
5254 | * sched_tick_start() | | sched_tick_stop() | |
5255 | * | | | |
5256 | * V | | |
5257 | * TICK_SCHED_REMOTE_RUNNING | |
5258 | * | |
5259 | * | |
5260 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
5261 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
5262 | */ | |
d84b3131 FW |
5263 | |
5264 | static struct tick_work __percpu *tick_work_cpu; | |
5265 | ||
5266 | static void sched_tick_remote(struct work_struct *work) | |
5267 | { | |
5268 | struct delayed_work *dwork = to_delayed_work(work); | |
5269 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
5270 | int cpu = twork->cpu; | |
5271 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 5272 | struct task_struct *curr; |
d84b3131 | 5273 | struct rq_flags rf; |
d9c0ffca | 5274 | u64 delta; |
b55bd585 | 5275 | int os; |
d84b3131 FW |
5276 | |
5277 | /* | |
5278 | * Handle the tick only if it appears the remote CPU is running in full | |
5279 | * dynticks mode. The check is racy by nature, but missing a tick or | |
5280 | * having one too much is no big deal because the scheduler tick updates | |
5281 | * statistics and checks timeslices in a time-independent way, regardless | |
5282 | * of when exactly it is running. | |
5283 | */ | |
488603b8 | 5284 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 5285 | goto out_requeue; |
d84b3131 | 5286 | |
d9c0ffca FW |
5287 | rq_lock_irq(rq, &rf); |
5288 | curr = rq->curr; | |
488603b8 | 5289 | if (cpu_is_offline(cpu)) |
d9c0ffca | 5290 | goto out_unlock; |
d84b3131 | 5291 | |
d9c0ffca | 5292 | update_rq_clock(rq); |
d9c0ffca | 5293 | |
488603b8 SW |
5294 | if (!is_idle_task(curr)) { |
5295 | /* | |
5296 | * Make sure the next tick runs within a reasonable | |
5297 | * amount of time. | |
5298 | */ | |
5299 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
5300 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
5301 | } | |
d9c0ffca FW |
5302 | curr->sched_class->task_tick(rq, curr, 0); |
5303 | ||
ebc0f83c | 5304 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
5305 | out_unlock: |
5306 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 5307 | out_requeue: |
ebc0f83c | 5308 | |
d84b3131 FW |
5309 | /* |
5310 | * Run the remote tick once per second (1Hz). This arbitrary | |
5311 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
5312 | * to keep scheduler internal stats reasonably up to date. But |
5313 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 5314 | */ |
b55bd585 PM |
5315 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
5316 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
5317 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
5318 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
5319 | } |
5320 | ||
5321 | static void sched_tick_start(int cpu) | |
5322 | { | |
b55bd585 | 5323 | int os; |
d84b3131 FW |
5324 | struct tick_work *twork; |
5325 | ||
5326 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
5327 | return; | |
5328 | ||
5329 | WARN_ON_ONCE(!tick_work_cpu); | |
5330 | ||
5331 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5332 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
5333 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
5334 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
5335 | twork->cpu = cpu; | |
5336 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
5337 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
5338 | } | |
d84b3131 FW |
5339 | } |
5340 | ||
5341 | #ifdef CONFIG_HOTPLUG_CPU | |
5342 | static void sched_tick_stop(int cpu) | |
5343 | { | |
5344 | struct tick_work *twork; | |
b55bd585 | 5345 | int os; |
d84b3131 FW |
5346 | |
5347 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
5348 | return; | |
5349 | ||
5350 | WARN_ON_ONCE(!tick_work_cpu); | |
5351 | ||
5352 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5353 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
5354 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
5355 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
5356 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
5357 | } |
5358 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5359 | ||
5360 | int __init sched_tick_offload_init(void) | |
5361 | { | |
5362 | tick_work_cpu = alloc_percpu(struct tick_work); | |
5363 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
5364 | return 0; |
5365 | } | |
5366 | ||
5367 | #else /* !CONFIG_NO_HZ_FULL */ | |
5368 | static inline void sched_tick_start(int cpu) { } | |
5369 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 5370 | #endif |
1da177e4 | 5371 | |
c1a280b6 | 5372 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 5373 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
5374 | /* |
5375 | * If the value passed in is equal to the current preempt count | |
5376 | * then we just disabled preemption. Start timing the latency. | |
5377 | */ | |
5378 | static inline void preempt_latency_start(int val) | |
5379 | { | |
5380 | if (preempt_count() == val) { | |
5381 | unsigned long ip = get_lock_parent_ip(); | |
5382 | #ifdef CONFIG_DEBUG_PREEMPT | |
5383 | current->preempt_disable_ip = ip; | |
5384 | #endif | |
5385 | trace_preempt_off(CALLER_ADDR0, ip); | |
5386 | } | |
5387 | } | |
7e49fcce | 5388 | |
edafe3a5 | 5389 | void preempt_count_add(int val) |
1da177e4 | 5390 | { |
6cd8a4bb | 5391 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5392 | /* |
5393 | * Underflow? | |
5394 | */ | |
9a11b49a IM |
5395 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5396 | return; | |
6cd8a4bb | 5397 | #endif |
bdb43806 | 5398 | __preempt_count_add(val); |
6cd8a4bb | 5399 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5400 | /* |
5401 | * Spinlock count overflowing soon? | |
5402 | */ | |
33859f7f MOS |
5403 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5404 | PREEMPT_MASK - 10); | |
6cd8a4bb | 5405 | #endif |
47252cfb | 5406 | preempt_latency_start(val); |
1da177e4 | 5407 | } |
bdb43806 | 5408 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 5409 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 5410 | |
47252cfb SR |
5411 | /* |
5412 | * If the value passed in equals to the current preempt count | |
5413 | * then we just enabled preemption. Stop timing the latency. | |
5414 | */ | |
5415 | static inline void preempt_latency_stop(int val) | |
5416 | { | |
5417 | if (preempt_count() == val) | |
5418 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
5419 | } | |
5420 | ||
edafe3a5 | 5421 | void preempt_count_sub(int val) |
1da177e4 | 5422 | { |
6cd8a4bb | 5423 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5424 | /* |
5425 | * Underflow? | |
5426 | */ | |
01e3eb82 | 5427 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5428 | return; |
1da177e4 LT |
5429 | /* |
5430 | * Is the spinlock portion underflowing? | |
5431 | */ | |
9a11b49a IM |
5432 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5433 | !(preempt_count() & PREEMPT_MASK))) | |
5434 | return; | |
6cd8a4bb | 5435 | #endif |
9a11b49a | 5436 | |
47252cfb | 5437 | preempt_latency_stop(val); |
bdb43806 | 5438 | __preempt_count_sub(val); |
1da177e4 | 5439 | } |
bdb43806 | 5440 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 5441 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 5442 | |
47252cfb SR |
5443 | #else |
5444 | static inline void preempt_latency_start(int val) { } | |
5445 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
5446 | #endif |
5447 | ||
59ddbcb2 IM |
5448 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
5449 | { | |
5450 | #ifdef CONFIG_DEBUG_PREEMPT | |
5451 | return p->preempt_disable_ip; | |
5452 | #else | |
5453 | return 0; | |
5454 | #endif | |
5455 | } | |
5456 | ||
1da177e4 | 5457 | /* |
dd41f596 | 5458 | * Print scheduling while atomic bug: |
1da177e4 | 5459 | */ |
dd41f596 | 5460 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5461 | { |
d1c6d149 VN |
5462 | /* Save this before calling printk(), since that will clobber it */ |
5463 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
5464 | ||
664dfa65 DJ |
5465 | if (oops_in_progress) |
5466 | return; | |
5467 | ||
3df0fc5b PZ |
5468 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
5469 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 5470 | |
dd41f596 | 5471 | debug_show_held_locks(prev); |
e21f5b15 | 5472 | print_modules(); |
dd41f596 IM |
5473 | if (irqs_disabled()) |
5474 | print_irqtrace_events(prev); | |
d1c6d149 VN |
5475 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
5476 | && in_atomic_preempt_off()) { | |
8f47b187 | 5477 | pr_err("Preemption disabled at:"); |
2062a4e8 | 5478 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 5479 | } |
748c7201 DBO |
5480 | if (panic_on_warn) |
5481 | panic("scheduling while atomic\n"); | |
5482 | ||
6135fc1e | 5483 | dump_stack(); |
373d4d09 | 5484 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 5485 | } |
1da177e4 | 5486 | |
dd41f596 IM |
5487 | /* |
5488 | * Various schedule()-time debugging checks and statistics: | |
5489 | */ | |
312364f3 | 5490 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 5491 | { |
0d9e2632 | 5492 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
5493 | if (task_stack_end_corrupted(prev)) |
5494 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
5495 | |
5496 | if (task_scs_end_corrupted(prev)) | |
5497 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 5498 | #endif |
b99def8b | 5499 | |
312364f3 | 5500 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
2f064a59 | 5501 | if (!preempt && READ_ONCE(prev->__state) && prev->non_block_count) { |
312364f3 DV |
5502 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", |
5503 | prev->comm, prev->pid, prev->non_block_count); | |
5504 | dump_stack(); | |
5505 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
5506 | } | |
5507 | #endif | |
5508 | ||
1dc0fffc | 5509 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 5510 | __schedule_bug(prev); |
1dc0fffc PZ |
5511 | preempt_count_set(PREEMPT_DISABLED); |
5512 | } | |
b3fbab05 | 5513 | rcu_sleep_check(); |
9f68b5b7 | 5514 | SCHED_WARN_ON(ct_state() == CONTEXT_USER); |
dd41f596 | 5515 | |
1da177e4 LT |
5516 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5517 | ||
ae92882e | 5518 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
5519 | } |
5520 | ||
457d1f46 CY |
5521 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
5522 | struct rq_flags *rf) | |
5523 | { | |
5524 | #ifdef CONFIG_SMP | |
5525 | const struct sched_class *class; | |
5526 | /* | |
5527 | * We must do the balancing pass before put_prev_task(), such | |
5528 | * that when we release the rq->lock the task is in the same | |
5529 | * state as before we took rq->lock. | |
5530 | * | |
5531 | * We can terminate the balance pass as soon as we know there is | |
5532 | * a runnable task of @class priority or higher. | |
5533 | */ | |
5534 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
5535 | if (class->balance(rq, prev, rf)) | |
5536 | break; | |
5537 | } | |
5538 | #endif | |
5539 | ||
5540 | put_prev_task(rq, prev); | |
5541 | } | |
5542 | ||
dd41f596 IM |
5543 | /* |
5544 | * Pick up the highest-prio task: | |
5545 | */ | |
5546 | static inline struct task_struct * | |
539f6512 | 5547 | __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 5548 | { |
49ee5768 | 5549 | const struct sched_class *class; |
dd41f596 | 5550 | struct task_struct *p; |
1da177e4 LT |
5551 | |
5552 | /* | |
0ba87bb2 PZ |
5553 | * Optimization: we know that if all tasks are in the fair class we can |
5554 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 5555 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 5556 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 5557 | */ |
aa93cd53 | 5558 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
5559 | rq->nr_running == rq->cfs.h_nr_running)) { |
5560 | ||
5d7d6056 | 5561 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 5562 | if (unlikely(p == RETRY_TASK)) |
67692435 | 5563 | goto restart; |
6ccdc84b | 5564 | |
1699949d | 5565 | /* Assume the next prioritized class is idle_sched_class */ |
5d7d6056 | 5566 | if (!p) { |
f488e105 | 5567 | put_prev_task(rq, prev); |
98c2f700 | 5568 | p = pick_next_task_idle(rq); |
f488e105 | 5569 | } |
6ccdc84b PZ |
5570 | |
5571 | return p; | |
1da177e4 LT |
5572 | } |
5573 | ||
67692435 | 5574 | restart: |
457d1f46 | 5575 | put_prev_task_balance(rq, prev, rf); |
67692435 | 5576 | |
34f971f6 | 5577 | for_each_class(class) { |
98c2f700 | 5578 | p = class->pick_next_task(rq); |
67692435 | 5579 | if (p) |
dd41f596 | 5580 | return p; |
dd41f596 | 5581 | } |
34f971f6 | 5582 | |
d1ccc66d IM |
5583 | /* The idle class should always have a runnable task: */ |
5584 | BUG(); | |
dd41f596 | 5585 | } |
1da177e4 | 5586 | |
9edeaea1 | 5587 | #ifdef CONFIG_SCHED_CORE |
539f6512 PZ |
5588 | static inline bool is_task_rq_idle(struct task_struct *t) |
5589 | { | |
5590 | return (task_rq(t)->idle == t); | |
5591 | } | |
5592 | ||
5593 | static inline bool cookie_equals(struct task_struct *a, unsigned long cookie) | |
5594 | { | |
5595 | return is_task_rq_idle(a) || (a->core_cookie == cookie); | |
5596 | } | |
5597 | ||
5598 | static inline bool cookie_match(struct task_struct *a, struct task_struct *b) | |
5599 | { | |
5600 | if (is_task_rq_idle(a) || is_task_rq_idle(b)) | |
5601 | return true; | |
5602 | ||
5603 | return a->core_cookie == b->core_cookie; | |
5604 | } | |
5605 | ||
5606 | // XXX fairness/fwd progress conditions | |
5607 | /* | |
5608 | * Returns | |
5609 | * - NULL if there is no runnable task for this class. | |
5610 | * - the highest priority task for this runqueue if it matches | |
5611 | * rq->core->core_cookie or its priority is greater than max. | |
5612 | * - Else returns idle_task. | |
5613 | */ | |
5614 | static struct task_struct * | |
c6047c2e | 5615 | pick_task(struct rq *rq, const struct sched_class *class, struct task_struct *max, bool in_fi) |
539f6512 PZ |
5616 | { |
5617 | struct task_struct *class_pick, *cookie_pick; | |
5618 | unsigned long cookie = rq->core->core_cookie; | |
5619 | ||
5620 | class_pick = class->pick_task(rq); | |
5621 | if (!class_pick) | |
5622 | return NULL; | |
5623 | ||
5624 | if (!cookie) { | |
5625 | /* | |
5626 | * If class_pick is tagged, return it only if it has | |
5627 | * higher priority than max. | |
5628 | */ | |
5629 | if (max && class_pick->core_cookie && | |
c6047c2e | 5630 | prio_less(class_pick, max, in_fi)) |
539f6512 PZ |
5631 | return idle_sched_class.pick_task(rq); |
5632 | ||
5633 | return class_pick; | |
5634 | } | |
5635 | ||
5636 | /* | |
5637 | * If class_pick is idle or matches cookie, return early. | |
5638 | */ | |
5639 | if (cookie_equals(class_pick, cookie)) | |
5640 | return class_pick; | |
5641 | ||
5642 | cookie_pick = sched_core_find(rq, cookie); | |
5643 | ||
5644 | /* | |
5645 | * If class > max && class > cookie, it is the highest priority task on | |
5646 | * the core (so far) and it must be selected, otherwise we must go with | |
5647 | * the cookie pick in order to satisfy the constraint. | |
5648 | */ | |
c6047c2e JFG |
5649 | if (prio_less(cookie_pick, class_pick, in_fi) && |
5650 | (!max || prio_less(max, class_pick, in_fi))) | |
539f6512 PZ |
5651 | return class_pick; |
5652 | ||
5653 | return cookie_pick; | |
5654 | } | |
5655 | ||
c6047c2e JFG |
5656 | extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi); |
5657 | ||
539f6512 PZ |
5658 | static struct task_struct * |
5659 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
5660 | { | |
5661 | struct task_struct *next, *max = NULL; | |
5662 | const struct sched_class *class; | |
5663 | const struct cpumask *smt_mask; | |
c6047c2e | 5664 | bool fi_before = false; |
d2dfa17b | 5665 | int i, j, cpu, occ = 0; |
539f6512 | 5666 | bool need_sync; |
539f6512 PZ |
5667 | |
5668 | if (!sched_core_enabled(rq)) | |
5669 | return __pick_next_task(rq, prev, rf); | |
5670 | ||
5671 | cpu = cpu_of(rq); | |
5672 | ||
5673 | /* Stopper task is switching into idle, no need core-wide selection. */ | |
5674 | if (cpu_is_offline(cpu)) { | |
5675 | /* | |
5676 | * Reset core_pick so that we don't enter the fastpath when | |
5677 | * coming online. core_pick would already be migrated to | |
5678 | * another cpu during offline. | |
5679 | */ | |
5680 | rq->core_pick = NULL; | |
5681 | return __pick_next_task(rq, prev, rf); | |
5682 | } | |
5683 | ||
5684 | /* | |
5685 | * If there were no {en,de}queues since we picked (IOW, the task | |
5686 | * pointers are all still valid), and we haven't scheduled the last | |
5687 | * pick yet, do so now. | |
5688 | * | |
5689 | * rq->core_pick can be NULL if no selection was made for a CPU because | |
5690 | * it was either offline or went offline during a sibling's core-wide | |
5691 | * selection. In this case, do a core-wide selection. | |
5692 | */ | |
5693 | if (rq->core->core_pick_seq == rq->core->core_task_seq && | |
5694 | rq->core->core_pick_seq != rq->core_sched_seq && | |
5695 | rq->core_pick) { | |
5696 | WRITE_ONCE(rq->core_sched_seq, rq->core->core_pick_seq); | |
5697 | ||
5698 | next = rq->core_pick; | |
5699 | if (next != prev) { | |
5700 | put_prev_task(rq, prev); | |
5701 | set_next_task(rq, next); | |
5702 | } | |
5703 | ||
5704 | rq->core_pick = NULL; | |
5705 | return next; | |
5706 | } | |
5707 | ||
5708 | put_prev_task_balance(rq, prev, rf); | |
5709 | ||
5710 | smt_mask = cpu_smt_mask(cpu); | |
7afbba11 JFG |
5711 | need_sync = !!rq->core->core_cookie; |
5712 | ||
5713 | /* reset state */ | |
5714 | rq->core->core_cookie = 0UL; | |
5715 | if (rq->core->core_forceidle) { | |
5716 | need_sync = true; | |
5717 | fi_before = true; | |
5718 | rq->core->core_forceidle = false; | |
5719 | } | |
539f6512 PZ |
5720 | |
5721 | /* | |
5722 | * core->core_task_seq, core->core_pick_seq, rq->core_sched_seq | |
5723 | * | |
5724 | * @task_seq guards the task state ({en,de}queues) | |
5725 | * @pick_seq is the @task_seq we did a selection on | |
5726 | * @sched_seq is the @pick_seq we scheduled | |
5727 | * | |
5728 | * However, preemptions can cause multiple picks on the same task set. | |
5729 | * 'Fix' this by also increasing @task_seq for every pick. | |
5730 | */ | |
5731 | rq->core->core_task_seq++; | |
539f6512 | 5732 | |
7afbba11 JFG |
5733 | /* |
5734 | * Optimize for common case where this CPU has no cookies | |
5735 | * and there are no cookied tasks running on siblings. | |
5736 | */ | |
5737 | if (!need_sync) { | |
5738 | for_each_class(class) { | |
5739 | next = class->pick_task(rq); | |
5740 | if (next) | |
5741 | break; | |
5742 | } | |
5743 | ||
5744 | if (!next->core_cookie) { | |
5745 | rq->core_pick = NULL; | |
c6047c2e JFG |
5746 | /* |
5747 | * For robustness, update the min_vruntime_fi for | |
5748 | * unconstrained picks as well. | |
5749 | */ | |
5750 | WARN_ON_ONCE(fi_before); | |
5751 | task_vruntime_update(rq, next, false); | |
7afbba11 JFG |
5752 | goto done; |
5753 | } | |
8039e96f | 5754 | } |
7afbba11 | 5755 | |
539f6512 PZ |
5756 | for_each_cpu(i, smt_mask) { |
5757 | struct rq *rq_i = cpu_rq(i); | |
5758 | ||
5759 | rq_i->core_pick = NULL; | |
5760 | ||
539f6512 PZ |
5761 | if (i != cpu) |
5762 | update_rq_clock(rq_i); | |
5763 | } | |
5764 | ||
5765 | /* | |
cc00c198 | 5766 | * Try and select tasks for each sibling in descending sched_class |
539f6512 PZ |
5767 | * order. |
5768 | */ | |
5769 | for_each_class(class) { | |
5770 | again: | |
5771 | for_each_cpu_wrap(i, smt_mask, cpu) { | |
5772 | struct rq *rq_i = cpu_rq(i); | |
5773 | struct task_struct *p; | |
5774 | ||
5775 | if (rq_i->core_pick) | |
5776 | continue; | |
5777 | ||
5778 | /* | |
5779 | * If this sibling doesn't yet have a suitable task to | |
cc00c198 | 5780 | * run; ask for the most eligible task, given the |
539f6512 PZ |
5781 | * highest priority task already selected for this |
5782 | * core. | |
5783 | */ | |
c6047c2e | 5784 | p = pick_task(rq_i, class, max, fi_before); |
7afbba11 | 5785 | if (!p) |
539f6512 | 5786 | continue; |
539f6512 | 5787 | |
d2dfa17b PZ |
5788 | if (!is_task_rq_idle(p)) |
5789 | occ++; | |
5790 | ||
539f6512 | 5791 | rq_i->core_pick = p; |
c6047c2e JFG |
5792 | if (rq_i->idle == p && rq_i->nr_running) { |
5793 | rq->core->core_forceidle = true; | |
5794 | if (!fi_before) | |
5795 | rq->core->core_forceidle_seq++; | |
5796 | } | |
539f6512 PZ |
5797 | |
5798 | /* | |
5799 | * If this new candidate is of higher priority than the | |
5800 | * previous; and they're incompatible; we need to wipe | |
5801 | * the slate and start over. pick_task makes sure that | |
5802 | * p's priority is more than max if it doesn't match | |
5803 | * max's cookie. | |
5804 | * | |
5805 | * NOTE: this is a linear max-filter and is thus bounded | |
5806 | * in execution time. | |
5807 | */ | |
5808 | if (!max || !cookie_match(max, p)) { | |
5809 | struct task_struct *old_max = max; | |
5810 | ||
5811 | rq->core->core_cookie = p->core_cookie; | |
5812 | max = p; | |
5813 | ||
5814 | if (old_max) { | |
c6047c2e | 5815 | rq->core->core_forceidle = false; |
539f6512 PZ |
5816 | for_each_cpu(j, smt_mask) { |
5817 | if (j == i) | |
5818 | continue; | |
5819 | ||
5820 | cpu_rq(j)->core_pick = NULL; | |
5821 | } | |
d2dfa17b | 5822 | occ = 1; |
539f6512 | 5823 | goto again; |
539f6512 | 5824 | } |
539f6512 PZ |
5825 | } |
5826 | } | |
539f6512 PZ |
5827 | } |
5828 | ||
5829 | rq->core->core_pick_seq = rq->core->core_task_seq; | |
5830 | next = rq->core_pick; | |
5831 | rq->core_sched_seq = rq->core->core_pick_seq; | |
5832 | ||
5833 | /* Something should have been selected for current CPU */ | |
5834 | WARN_ON_ONCE(!next); | |
5835 | ||
5836 | /* | |
5837 | * Reschedule siblings | |
5838 | * | |
5839 | * NOTE: L1TF -- at this point we're no longer running the old task and | |
5840 | * sending an IPI (below) ensures the sibling will no longer be running | |
5841 | * their task. This ensures there is no inter-sibling overlap between | |
5842 | * non-matching user state. | |
5843 | */ | |
5844 | for_each_cpu(i, smt_mask) { | |
5845 | struct rq *rq_i = cpu_rq(i); | |
5846 | ||
5847 | /* | |
5848 | * An online sibling might have gone offline before a task | |
5849 | * could be picked for it, or it might be offline but later | |
5850 | * happen to come online, but its too late and nothing was | |
5851 | * picked for it. That's Ok - it will pick tasks for itself, | |
5852 | * so ignore it. | |
5853 | */ | |
5854 | if (!rq_i->core_pick) | |
5855 | continue; | |
5856 | ||
c6047c2e JFG |
5857 | /* |
5858 | * Update for new !FI->FI transitions, or if continuing to be in !FI: | |
5859 | * fi_before fi update? | |
5860 | * 0 0 1 | |
5861 | * 0 1 1 | |
5862 | * 1 0 1 | |
5863 | * 1 1 0 | |
5864 | */ | |
5865 | if (!(fi_before && rq->core->core_forceidle)) | |
5866 | task_vruntime_update(rq_i, rq_i->core_pick, rq->core->core_forceidle); | |
539f6512 | 5867 | |
d2dfa17b PZ |
5868 | rq_i->core_pick->core_occupation = occ; |
5869 | ||
539f6512 PZ |
5870 | if (i == cpu) { |
5871 | rq_i->core_pick = NULL; | |
5872 | continue; | |
5873 | } | |
5874 | ||
5875 | /* Did we break L1TF mitigation requirements? */ | |
5876 | WARN_ON_ONCE(!cookie_match(next, rq_i->core_pick)); | |
5877 | ||
5878 | if (rq_i->curr == rq_i->core_pick) { | |
5879 | rq_i->core_pick = NULL; | |
5880 | continue; | |
5881 | } | |
5882 | ||
5883 | resched_curr(rq_i); | |
5884 | } | |
5885 | ||
5886 | done: | |
5887 | set_next_task(rq, next); | |
5888 | return next; | |
5889 | } | |
9edeaea1 | 5890 | |
d2dfa17b PZ |
5891 | static bool try_steal_cookie(int this, int that) |
5892 | { | |
5893 | struct rq *dst = cpu_rq(this), *src = cpu_rq(that); | |
5894 | struct task_struct *p; | |
5895 | unsigned long cookie; | |
5896 | bool success = false; | |
5897 | ||
5898 | local_irq_disable(); | |
5899 | double_rq_lock(dst, src); | |
5900 | ||
5901 | cookie = dst->core->core_cookie; | |
5902 | if (!cookie) | |
5903 | goto unlock; | |
5904 | ||
5905 | if (dst->curr != dst->idle) | |
5906 | goto unlock; | |
5907 | ||
5908 | p = sched_core_find(src, cookie); | |
5909 | if (p == src->idle) | |
5910 | goto unlock; | |
5911 | ||
5912 | do { | |
5913 | if (p == src->core_pick || p == src->curr) | |
5914 | goto next; | |
5915 | ||
5916 | if (!cpumask_test_cpu(this, &p->cpus_mask)) | |
5917 | goto next; | |
5918 | ||
5919 | if (p->core_occupation > dst->idle->core_occupation) | |
5920 | goto next; | |
5921 | ||
d2dfa17b PZ |
5922 | deactivate_task(src, p, 0); |
5923 | set_task_cpu(p, this); | |
5924 | activate_task(dst, p, 0); | |
d2dfa17b PZ |
5925 | |
5926 | resched_curr(dst); | |
5927 | ||
5928 | success = true; | |
5929 | break; | |
5930 | ||
5931 | next: | |
5932 | p = sched_core_next(p, cookie); | |
5933 | } while (p); | |
5934 | ||
5935 | unlock: | |
5936 | double_rq_unlock(dst, src); | |
5937 | local_irq_enable(); | |
5938 | ||
5939 | return success; | |
5940 | } | |
5941 | ||
5942 | static bool steal_cookie_task(int cpu, struct sched_domain *sd) | |
5943 | { | |
5944 | int i; | |
5945 | ||
5946 | for_each_cpu_wrap(i, sched_domain_span(sd), cpu) { | |
5947 | if (i == cpu) | |
5948 | continue; | |
5949 | ||
5950 | if (need_resched()) | |
5951 | break; | |
5952 | ||
5953 | if (try_steal_cookie(cpu, i)) | |
5954 | return true; | |
5955 | } | |
5956 | ||
5957 | return false; | |
5958 | } | |
5959 | ||
5960 | static void sched_core_balance(struct rq *rq) | |
5961 | { | |
5962 | struct sched_domain *sd; | |
5963 | int cpu = cpu_of(rq); | |
5964 | ||
5965 | preempt_disable(); | |
5966 | rcu_read_lock(); | |
5967 | raw_spin_rq_unlock_irq(rq); | |
5968 | for_each_domain(cpu, sd) { | |
5969 | if (need_resched()) | |
5970 | break; | |
5971 | ||
5972 | if (steal_cookie_task(cpu, sd)) | |
5973 | break; | |
5974 | } | |
5975 | raw_spin_rq_lock_irq(rq); | |
5976 | rcu_read_unlock(); | |
5977 | preempt_enable(); | |
5978 | } | |
5979 | ||
5980 | static DEFINE_PER_CPU(struct callback_head, core_balance_head); | |
5981 | ||
5982 | void queue_core_balance(struct rq *rq) | |
5983 | { | |
5984 | if (!sched_core_enabled(rq)) | |
5985 | return; | |
5986 | ||
5987 | if (!rq->core->core_cookie) | |
5988 | return; | |
5989 | ||
5990 | if (!rq->nr_running) /* not forced idle */ | |
5991 | return; | |
5992 | ||
5993 | queue_balance_callback(rq, &per_cpu(core_balance_head, rq->cpu), sched_core_balance); | |
5994 | } | |
5995 | ||
3c474b32 | 5996 | static void sched_core_cpu_starting(unsigned int cpu) |
9edeaea1 PZ |
5997 | { |
5998 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
3c474b32 PZ |
5999 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; |
6000 | unsigned long flags; | |
6001 | int t; | |
9edeaea1 | 6002 | |
3c474b32 | 6003 | sched_core_lock(cpu, &flags); |
9edeaea1 | 6004 | |
3c474b32 PZ |
6005 | WARN_ON_ONCE(rq->core != rq); |
6006 | ||
6007 | /* if we're the first, we'll be our own leader */ | |
6008 | if (cpumask_weight(smt_mask) == 1) | |
6009 | goto unlock; | |
6010 | ||
6011 | /* find the leader */ | |
6012 | for_each_cpu(t, smt_mask) { | |
6013 | if (t == cpu) | |
6014 | continue; | |
6015 | rq = cpu_rq(t); | |
6016 | if (rq->core == rq) { | |
6017 | core_rq = rq; | |
6018 | break; | |
9edeaea1 | 6019 | } |
3c474b32 | 6020 | } |
9edeaea1 | 6021 | |
3c474b32 PZ |
6022 | if (WARN_ON_ONCE(!core_rq)) /* whoopsie */ |
6023 | goto unlock; | |
9edeaea1 | 6024 | |
3c474b32 PZ |
6025 | /* install and validate core_rq */ |
6026 | for_each_cpu(t, smt_mask) { | |
6027 | rq = cpu_rq(t); | |
9edeaea1 | 6028 | |
3c474b32 | 6029 | if (t == cpu) |
9edeaea1 | 6030 | rq->core = core_rq; |
3c474b32 PZ |
6031 | |
6032 | WARN_ON_ONCE(rq->core != core_rq); | |
9edeaea1 | 6033 | } |
3c474b32 PZ |
6034 | |
6035 | unlock: | |
6036 | sched_core_unlock(cpu, &flags); | |
9edeaea1 | 6037 | } |
3c474b32 PZ |
6038 | |
6039 | static void sched_core_cpu_deactivate(unsigned int cpu) | |
6040 | { | |
6041 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
6042 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; | |
6043 | unsigned long flags; | |
6044 | int t; | |
6045 | ||
6046 | sched_core_lock(cpu, &flags); | |
6047 | ||
6048 | /* if we're the last man standing, nothing to do */ | |
6049 | if (cpumask_weight(smt_mask) == 1) { | |
6050 | WARN_ON_ONCE(rq->core != rq); | |
6051 | goto unlock; | |
6052 | } | |
6053 | ||
6054 | /* if we're not the leader, nothing to do */ | |
6055 | if (rq->core != rq) | |
6056 | goto unlock; | |
6057 | ||
6058 | /* find a new leader */ | |
6059 | for_each_cpu(t, smt_mask) { | |
6060 | if (t == cpu) | |
6061 | continue; | |
6062 | core_rq = cpu_rq(t); | |
6063 | break; | |
6064 | } | |
6065 | ||
6066 | if (WARN_ON_ONCE(!core_rq)) /* impossible */ | |
6067 | goto unlock; | |
6068 | ||
6069 | /* copy the shared state to the new leader */ | |
6070 | core_rq->core_task_seq = rq->core_task_seq; | |
6071 | core_rq->core_pick_seq = rq->core_pick_seq; | |
6072 | core_rq->core_cookie = rq->core_cookie; | |
6073 | core_rq->core_forceidle = rq->core_forceidle; | |
6074 | core_rq->core_forceidle_seq = rq->core_forceidle_seq; | |
6075 | ||
6076 | /* install new leader */ | |
6077 | for_each_cpu(t, smt_mask) { | |
6078 | rq = cpu_rq(t); | |
6079 | rq->core = core_rq; | |
6080 | } | |
6081 | ||
6082 | unlock: | |
6083 | sched_core_unlock(cpu, &flags); | |
6084 | } | |
6085 | ||
6086 | static inline void sched_core_cpu_dying(unsigned int cpu) | |
6087 | { | |
6088 | struct rq *rq = cpu_rq(cpu); | |
6089 | ||
6090 | if (rq->core != rq) | |
6091 | rq->core = rq; | |
6092 | } | |
6093 | ||
9edeaea1 PZ |
6094 | #else /* !CONFIG_SCHED_CORE */ |
6095 | ||
6096 | static inline void sched_core_cpu_starting(unsigned int cpu) {} | |
3c474b32 PZ |
6097 | static inline void sched_core_cpu_deactivate(unsigned int cpu) {} |
6098 | static inline void sched_core_cpu_dying(unsigned int cpu) {} | |
9edeaea1 | 6099 | |
539f6512 PZ |
6100 | static struct task_struct * |
6101 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
6102 | { | |
6103 | return __pick_next_task(rq, prev, rf); | |
6104 | } | |
6105 | ||
9edeaea1 PZ |
6106 | #endif /* CONFIG_SCHED_CORE */ |
6107 | ||
b4bfa3fc TG |
6108 | /* |
6109 | * Constants for the sched_mode argument of __schedule(). | |
6110 | * | |
6111 | * The mode argument allows RT enabled kernels to differentiate a | |
6112 | * preemption from blocking on an 'sleeping' spin/rwlock. Note that | |
6113 | * SM_MASK_PREEMPT for !RT has all bits set, which allows the compiler to | |
6114 | * optimize the AND operation out and just check for zero. | |
6115 | */ | |
6116 | #define SM_NONE 0x0 | |
6117 | #define SM_PREEMPT 0x1 | |
6991436c TG |
6118 | #define SM_RTLOCK_WAIT 0x2 |
6119 | ||
6120 | #ifndef CONFIG_PREEMPT_RT | |
6121 | # define SM_MASK_PREEMPT (~0U) | |
6122 | #else | |
6123 | # define SM_MASK_PREEMPT SM_PREEMPT | |
6124 | #endif | |
b4bfa3fc | 6125 | |
dd41f596 | 6126 | /* |
c259e01a | 6127 | * __schedule() is the main scheduler function. |
edde96ea PE |
6128 | * |
6129 | * The main means of driving the scheduler and thus entering this function are: | |
6130 | * | |
6131 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
6132 | * | |
6133 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
6134 | * paths. For example, see arch/x86/entry_64.S. | |
6135 | * | |
6136 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
6137 | * interrupt handler scheduler_tick(). | |
6138 | * | |
6139 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
6140 | * task to the run-queue and that's it. | |
6141 | * | |
6142 | * Now, if the new task added to the run-queue preempts the current | |
6143 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
6144 | * called on the nearest possible occasion: | |
6145 | * | |
c1a280b6 | 6146 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
6147 | * |
6148 | * - in syscall or exception context, at the next outmost | |
6149 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
6150 | * spin_unlock()!) | |
6151 | * | |
6152 | * - in IRQ context, return from interrupt-handler to | |
6153 | * preemptible context | |
6154 | * | |
c1a280b6 | 6155 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
6156 | * then at the next: |
6157 | * | |
6158 | * - cond_resched() call | |
6159 | * - explicit schedule() call | |
6160 | * - return from syscall or exception to user-space | |
6161 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 6162 | * |
b30f0e3f | 6163 | * WARNING: must be called with preemption disabled! |
dd41f596 | 6164 | */ |
b4bfa3fc | 6165 | static void __sched notrace __schedule(unsigned int sched_mode) |
dd41f596 IM |
6166 | { |
6167 | struct task_struct *prev, *next; | |
67ca7bde | 6168 | unsigned long *switch_count; |
dbfb089d | 6169 | unsigned long prev_state; |
d8ac8971 | 6170 | struct rq_flags rf; |
dd41f596 | 6171 | struct rq *rq; |
31656519 | 6172 | int cpu; |
dd41f596 | 6173 | |
dd41f596 IM |
6174 | cpu = smp_processor_id(); |
6175 | rq = cpu_rq(cpu); | |
dd41f596 | 6176 | prev = rq->curr; |
dd41f596 | 6177 | |
b4bfa3fc | 6178 | schedule_debug(prev, !!sched_mode); |
1da177e4 | 6179 | |
e0ee463c | 6180 | if (sched_feat(HRTICK) || sched_feat(HRTICK_DL)) |
f333fdc9 | 6181 | hrtick_clear(rq); |
8f4d37ec | 6182 | |
46a5d164 | 6183 | local_irq_disable(); |
b4bfa3fc | 6184 | rcu_note_context_switch(!!sched_mode); |
46a5d164 | 6185 | |
e0acd0a6 ON |
6186 | /* |
6187 | * Make sure that signal_pending_state()->signal_pending() below | |
6188 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
6189 | * done by the caller to avoid the race with signal_wake_up(): |
6190 | * | |
6191 | * __set_current_state(@state) signal_wake_up() | |
6192 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
6193 | * wake_up_state(p, state) | |
6194 | * LOCK rq->lock LOCK p->pi_state | |
6195 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
6196 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 6197 | * |
dbfb089d | 6198 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 6199 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 6200 | */ |
8a8c69c3 | 6201 | rq_lock(rq, &rf); |
d89e588c | 6202 | smp_mb__after_spinlock(); |
1da177e4 | 6203 | |
d1ccc66d IM |
6204 | /* Promote REQ to ACT */ |
6205 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 6206 | update_rq_clock(rq); |
9edfbfed | 6207 | |
246d86b5 | 6208 | switch_count = &prev->nivcsw; |
d136122f | 6209 | |
dbfb089d | 6210 | /* |
d136122f PZ |
6211 | * We must load prev->state once (task_struct::state is volatile), such |
6212 | * that: | |
6213 | * | |
6214 | * - we form a control dependency vs deactivate_task() below. | |
6215 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 6216 | */ |
2f064a59 | 6217 | prev_state = READ_ONCE(prev->__state); |
b4bfa3fc | 6218 | if (!(sched_mode & SM_MASK_PREEMPT) && prev_state) { |
dbfb089d | 6219 | if (signal_pending_state(prev_state, prev)) { |
2f064a59 | 6220 | WRITE_ONCE(prev->__state, TASK_RUNNING); |
21aa9af0 | 6221 | } else { |
dbfb089d PZ |
6222 | prev->sched_contributes_to_load = |
6223 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
6224 | !(prev_state & TASK_NOLOAD) && | |
6225 | !(prev->flags & PF_FROZEN); | |
6226 | ||
6227 | if (prev->sched_contributes_to_load) | |
6228 | rq->nr_uninterruptible++; | |
6229 | ||
6230 | /* | |
6231 | * __schedule() ttwu() | |
d136122f PZ |
6232 | * prev_state = prev->state; if (p->on_rq && ...) |
6233 | * if (prev_state) goto out; | |
6234 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
6235 | * p->state = TASK_WAKING | |
6236 | * | |
6237 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
6238 | * |
6239 | * After this, schedule() must not care about p->state any more. | |
6240 | */ | |
bce4dc80 | 6241 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 6242 | |
e33a9bba TH |
6243 | if (prev->in_iowait) { |
6244 | atomic_inc(&rq->nr_iowait); | |
6245 | delayacct_blkio_start(); | |
6246 | } | |
21aa9af0 | 6247 | } |
dd41f596 | 6248 | switch_count = &prev->nvcsw; |
1da177e4 LT |
6249 | } |
6250 | ||
d8ac8971 | 6251 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 6252 | clear_tsk_need_resched(prev); |
f27dde8d | 6253 | clear_preempt_need_resched(); |
c006fac5 PT |
6254 | #ifdef CONFIG_SCHED_DEBUG |
6255 | rq->last_seen_need_resched_ns = 0; | |
6256 | #endif | |
1da177e4 | 6257 | |
1da177e4 | 6258 | if (likely(prev != next)) { |
1da177e4 | 6259 | rq->nr_switches++; |
5311a98f EB |
6260 | /* |
6261 | * RCU users of rcu_dereference(rq->curr) may not see | |
6262 | * changes to task_struct made by pick_next_task(). | |
6263 | */ | |
6264 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
6265 | /* |
6266 | * The membarrier system call requires each architecture | |
6267 | * to have a full memory barrier after updating | |
306e0604 MD |
6268 | * rq->curr, before returning to user-space. |
6269 | * | |
6270 | * Here are the schemes providing that barrier on the | |
6271 | * various architectures: | |
6272 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
6273 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
6274 | * - finish_lock_switch() for weakly-ordered | |
6275 | * architectures where spin_unlock is a full barrier, | |
6276 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
6277 | * is a RELEASE barrier), | |
22e4ebb9 | 6278 | */ |
1da177e4 LT |
6279 | ++*switch_count; |
6280 | ||
af449901 | 6281 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
6282 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
6283 | ||
b4bfa3fc | 6284 | trace_sched_switch(sched_mode & SM_MASK_PREEMPT, prev, next); |
d1ccc66d IM |
6285 | |
6286 | /* Also unlocks the rq: */ | |
6287 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 6288 | } else { |
cb42c9a3 | 6289 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 6290 | |
565790d2 PZ |
6291 | rq_unpin_lock(rq, &rf); |
6292 | __balance_callbacks(rq); | |
5cb9eaa3 | 6293 | raw_spin_rq_unlock_irq(rq); |
565790d2 | 6294 | } |
1da177e4 | 6295 | } |
c259e01a | 6296 | |
9af6528e PZ |
6297 | void __noreturn do_task_dead(void) |
6298 | { | |
d1ccc66d | 6299 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 6300 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
6301 | |
6302 | /* Tell freezer to ignore us: */ | |
6303 | current->flags |= PF_NOFREEZE; | |
6304 | ||
b4bfa3fc | 6305 | __schedule(SM_NONE); |
9af6528e | 6306 | BUG(); |
d1ccc66d IM |
6307 | |
6308 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 6309 | for (;;) |
d1ccc66d | 6310 | cpu_relax(); |
9af6528e PZ |
6311 | } |
6312 | ||
9c40cef2 TG |
6313 | static inline void sched_submit_work(struct task_struct *tsk) |
6314 | { | |
c1cecf88 SAS |
6315 | unsigned int task_flags; |
6316 | ||
b03fbd4f | 6317 | if (task_is_running(tsk)) |
9c40cef2 | 6318 | return; |
6d25be57 | 6319 | |
c1cecf88 | 6320 | task_flags = tsk->flags; |
6d25be57 TG |
6321 | /* |
6322 | * If a worker went to sleep, notify and ask workqueue whether | |
6323 | * it wants to wake up a task to maintain concurrency. | |
6324 | * As this function is called inside the schedule() context, | |
6325 | * we disable preemption to avoid it calling schedule() again | |
62849a96 SAS |
6326 | * in the possible wakeup of a kworker and because wq_worker_sleeping() |
6327 | * requires it. | |
6d25be57 | 6328 | */ |
c1cecf88 | 6329 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6d25be57 | 6330 | preempt_disable(); |
c1cecf88 | 6331 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
6332 | wq_worker_sleeping(tsk); |
6333 | else | |
6334 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
6335 | preempt_enable_no_resched(); |
6336 | } | |
6337 | ||
b0fdc013 SAS |
6338 | if (tsk_is_pi_blocked(tsk)) |
6339 | return; | |
6340 | ||
9c40cef2 TG |
6341 | /* |
6342 | * If we are going to sleep and we have plugged IO queued, | |
6343 | * make sure to submit it to avoid deadlocks. | |
6344 | */ | |
6345 | if (blk_needs_flush_plug(tsk)) | |
6346 | blk_schedule_flush_plug(tsk); | |
6347 | } | |
6348 | ||
6d25be57 TG |
6349 | static void sched_update_worker(struct task_struct *tsk) |
6350 | { | |
771b53d0 JA |
6351 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6352 | if (tsk->flags & PF_WQ_WORKER) | |
6353 | wq_worker_running(tsk); | |
6354 | else | |
6355 | io_wq_worker_running(tsk); | |
6356 | } | |
6d25be57 TG |
6357 | } |
6358 | ||
722a9f92 | 6359 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 6360 | { |
9c40cef2 TG |
6361 | struct task_struct *tsk = current; |
6362 | ||
6363 | sched_submit_work(tsk); | |
bfd9b2b5 | 6364 | do { |
b30f0e3f | 6365 | preempt_disable(); |
b4bfa3fc | 6366 | __schedule(SM_NONE); |
b30f0e3f | 6367 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 6368 | } while (need_resched()); |
6d25be57 | 6369 | sched_update_worker(tsk); |
c259e01a | 6370 | } |
1da177e4 LT |
6371 | EXPORT_SYMBOL(schedule); |
6372 | ||
8663effb SRV |
6373 | /* |
6374 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
6375 | * state (have scheduled out non-voluntarily) by making sure that all | |
6376 | * tasks have either left the run queue or have gone into user space. | |
6377 | * As idle tasks do not do either, they must not ever be preempted | |
6378 | * (schedule out non-voluntarily). | |
6379 | * | |
6380 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
6381 | * never enables preemption because it does not call sched_submit_work(). | |
6382 | */ | |
6383 | void __sched schedule_idle(void) | |
6384 | { | |
6385 | /* | |
6386 | * As this skips calling sched_submit_work(), which the idle task does | |
6387 | * regardless because that function is a nop when the task is in a | |
6388 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
6389 | * current task can be in any other state. Note, idle is always in the | |
6390 | * TASK_RUNNING state. | |
6391 | */ | |
2f064a59 | 6392 | WARN_ON_ONCE(current->__state); |
8663effb | 6393 | do { |
b4bfa3fc | 6394 | __schedule(SM_NONE); |
8663effb SRV |
6395 | } while (need_resched()); |
6396 | } | |
6397 | ||
6775de49 | 6398 | #if defined(CONFIG_CONTEXT_TRACKING) && !defined(CONFIG_HAVE_CONTEXT_TRACKING_OFFSTACK) |
722a9f92 | 6399 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
6400 | { |
6401 | /* | |
6402 | * If we come here after a random call to set_need_resched(), | |
6403 | * or we have been woken up remotely but the IPI has not yet arrived, | |
6404 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
6405 | * we find a better solution. | |
7cc78f8f AL |
6406 | * |
6407 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 6408 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 6409 | * too frequently to make sense yet. |
20ab65e3 | 6410 | */ |
7cc78f8f | 6411 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 6412 | schedule(); |
7cc78f8f | 6413 | exception_exit(prev_state); |
20ab65e3 FW |
6414 | } |
6415 | #endif | |
6416 | ||
c5491ea7 TG |
6417 | /** |
6418 | * schedule_preempt_disabled - called with preemption disabled | |
6419 | * | |
6420 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
6421 | */ | |
6422 | void __sched schedule_preempt_disabled(void) | |
6423 | { | |
ba74c144 | 6424 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
6425 | schedule(); |
6426 | preempt_disable(); | |
6427 | } | |
6428 | ||
6991436c TG |
6429 | #ifdef CONFIG_PREEMPT_RT |
6430 | void __sched notrace schedule_rtlock(void) | |
6431 | { | |
6432 | do { | |
6433 | preempt_disable(); | |
6434 | __schedule(SM_RTLOCK_WAIT); | |
6435 | sched_preempt_enable_no_resched(); | |
6436 | } while (need_resched()); | |
6437 | } | |
6438 | NOKPROBE_SYMBOL(schedule_rtlock); | |
6439 | #endif | |
6440 | ||
06b1f808 | 6441 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
6442 | { |
6443 | do { | |
47252cfb SR |
6444 | /* |
6445 | * Because the function tracer can trace preempt_count_sub() | |
6446 | * and it also uses preempt_enable/disable_notrace(), if | |
6447 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6448 | * by the function tracer will call this function again and | |
6449 | * cause infinite recursion. | |
6450 | * | |
6451 | * Preemption must be disabled here before the function | |
6452 | * tracer can trace. Break up preempt_disable() into two | |
6453 | * calls. One to disable preemption without fear of being | |
6454 | * traced. The other to still record the preemption latency, | |
6455 | * which can also be traced by the function tracer. | |
6456 | */ | |
499d7955 | 6457 | preempt_disable_notrace(); |
47252cfb | 6458 | preempt_latency_start(1); |
b4bfa3fc | 6459 | __schedule(SM_PREEMPT); |
47252cfb | 6460 | preempt_latency_stop(1); |
499d7955 | 6461 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
6462 | |
6463 | /* | |
6464 | * Check again in case we missed a preemption opportunity | |
6465 | * between schedule and now. | |
6466 | */ | |
a18b5d01 FW |
6467 | } while (need_resched()); |
6468 | } | |
6469 | ||
c1a280b6 | 6470 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 6471 | /* |
a49b4f40 VS |
6472 | * This is the entry point to schedule() from in-kernel preemption |
6473 | * off of preempt_enable. | |
1da177e4 | 6474 | */ |
722a9f92 | 6475 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 6476 | { |
1da177e4 LT |
6477 | /* |
6478 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 6479 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 6480 | */ |
fbb00b56 | 6481 | if (likely(!preemptible())) |
1da177e4 LT |
6482 | return; |
6483 | ||
a18b5d01 | 6484 | preempt_schedule_common(); |
1da177e4 | 6485 | } |
376e2424 | 6486 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 6487 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 6488 | |
2c9a98d3 PZI |
6489 | #ifdef CONFIG_PREEMPT_DYNAMIC |
6490 | DEFINE_STATIC_CALL(preempt_schedule, __preempt_schedule_func); | |
ef72661e | 6491 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule); |
2c9a98d3 PZI |
6492 | #endif |
6493 | ||
6494 | ||
009f60e2 | 6495 | /** |
4eaca0a8 | 6496 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
6497 | * |
6498 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
6499 | * recursion and tracing preempt enabling caused by the tracing | |
6500 | * infrastructure itself. But as tracing can happen in areas coming | |
6501 | * from userspace or just about to enter userspace, a preempt enable | |
6502 | * can occur before user_exit() is called. This will cause the scheduler | |
6503 | * to be called when the system is still in usermode. | |
6504 | * | |
6505 | * To prevent this, the preempt_enable_notrace will use this function | |
6506 | * instead of preempt_schedule() to exit user context if needed before | |
6507 | * calling the scheduler. | |
6508 | */ | |
4eaca0a8 | 6509 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
6510 | { |
6511 | enum ctx_state prev_ctx; | |
6512 | ||
6513 | if (likely(!preemptible())) | |
6514 | return; | |
6515 | ||
6516 | do { | |
47252cfb SR |
6517 | /* |
6518 | * Because the function tracer can trace preempt_count_sub() | |
6519 | * and it also uses preempt_enable/disable_notrace(), if | |
6520 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6521 | * by the function tracer will call this function again and | |
6522 | * cause infinite recursion. | |
6523 | * | |
6524 | * Preemption must be disabled here before the function | |
6525 | * tracer can trace. Break up preempt_disable() into two | |
6526 | * calls. One to disable preemption without fear of being | |
6527 | * traced. The other to still record the preemption latency, | |
6528 | * which can also be traced by the function tracer. | |
6529 | */ | |
3d8f74dd | 6530 | preempt_disable_notrace(); |
47252cfb | 6531 | preempt_latency_start(1); |
009f60e2 ON |
6532 | /* |
6533 | * Needs preempt disabled in case user_exit() is traced | |
6534 | * and the tracer calls preempt_enable_notrace() causing | |
6535 | * an infinite recursion. | |
6536 | */ | |
6537 | prev_ctx = exception_enter(); | |
b4bfa3fc | 6538 | __schedule(SM_PREEMPT); |
009f60e2 ON |
6539 | exception_exit(prev_ctx); |
6540 | ||
47252cfb | 6541 | preempt_latency_stop(1); |
3d8f74dd | 6542 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
6543 | } while (need_resched()); |
6544 | } | |
4eaca0a8 | 6545 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 6546 | |
2c9a98d3 PZI |
6547 | #ifdef CONFIG_PREEMPT_DYNAMIC |
6548 | DEFINE_STATIC_CALL(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
ef72661e | 6549 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule_notrace); |
2c9a98d3 PZI |
6550 | #endif |
6551 | ||
c1a280b6 | 6552 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 | 6553 | |
826bfeb3 PZI |
6554 | #ifdef CONFIG_PREEMPT_DYNAMIC |
6555 | ||
6556 | #include <linux/entry-common.h> | |
6557 | ||
6558 | /* | |
6559 | * SC:cond_resched | |
6560 | * SC:might_resched | |
6561 | * SC:preempt_schedule | |
6562 | * SC:preempt_schedule_notrace | |
6563 | * SC:irqentry_exit_cond_resched | |
6564 | * | |
6565 | * | |
6566 | * NONE: | |
6567 | * cond_resched <- __cond_resched | |
6568 | * might_resched <- RET0 | |
6569 | * preempt_schedule <- NOP | |
6570 | * preempt_schedule_notrace <- NOP | |
6571 | * irqentry_exit_cond_resched <- NOP | |
6572 | * | |
6573 | * VOLUNTARY: | |
6574 | * cond_resched <- __cond_resched | |
6575 | * might_resched <- __cond_resched | |
6576 | * preempt_schedule <- NOP | |
6577 | * preempt_schedule_notrace <- NOP | |
6578 | * irqentry_exit_cond_resched <- NOP | |
6579 | * | |
6580 | * FULL: | |
6581 | * cond_resched <- RET0 | |
6582 | * might_resched <- RET0 | |
6583 | * preempt_schedule <- preempt_schedule | |
6584 | * preempt_schedule_notrace <- preempt_schedule_notrace | |
6585 | * irqentry_exit_cond_resched <- irqentry_exit_cond_resched | |
6586 | */ | |
e59e10f8 PZ |
6587 | |
6588 | enum { | |
6589 | preempt_dynamic_none = 0, | |
6590 | preempt_dynamic_voluntary, | |
6591 | preempt_dynamic_full, | |
6592 | }; | |
6593 | ||
1011dcce | 6594 | int preempt_dynamic_mode = preempt_dynamic_full; |
e59e10f8 | 6595 | |
1011dcce | 6596 | int sched_dynamic_mode(const char *str) |
826bfeb3 | 6597 | { |
e59e10f8 | 6598 | if (!strcmp(str, "none")) |
7e1b2eb7 | 6599 | return preempt_dynamic_none; |
e59e10f8 PZ |
6600 | |
6601 | if (!strcmp(str, "voluntary")) | |
7e1b2eb7 | 6602 | return preempt_dynamic_voluntary; |
e59e10f8 PZ |
6603 | |
6604 | if (!strcmp(str, "full")) | |
7e1b2eb7 | 6605 | return preempt_dynamic_full; |
e59e10f8 | 6606 | |
c4681f3f | 6607 | return -EINVAL; |
e59e10f8 PZ |
6608 | } |
6609 | ||
1011dcce | 6610 | void sched_dynamic_update(int mode) |
e59e10f8 PZ |
6611 | { |
6612 | /* | |
6613 | * Avoid {NONE,VOLUNTARY} -> FULL transitions from ever ending up in | |
6614 | * the ZERO state, which is invalid. | |
6615 | */ | |
6616 | static_call_update(cond_resched, __cond_resched); | |
6617 | static_call_update(might_resched, __cond_resched); | |
6618 | static_call_update(preempt_schedule, __preempt_schedule_func); | |
6619 | static_call_update(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
6620 | static_call_update(irqentry_exit_cond_resched, irqentry_exit_cond_resched); | |
6621 | ||
6622 | switch (mode) { | |
6623 | case preempt_dynamic_none: | |
826bfeb3 | 6624 | static_call_update(cond_resched, __cond_resched); |
9432bbd9 PZ |
6625 | static_call_update(might_resched, (void *)&__static_call_return0); |
6626 | static_call_update(preempt_schedule, NULL); | |
6627 | static_call_update(preempt_schedule_notrace, NULL); | |
6628 | static_call_update(irqentry_exit_cond_resched, NULL); | |
e59e10f8 PZ |
6629 | pr_info("Dynamic Preempt: none\n"); |
6630 | break; | |
6631 | ||
6632 | case preempt_dynamic_voluntary: | |
826bfeb3 PZI |
6633 | static_call_update(cond_resched, __cond_resched); |
6634 | static_call_update(might_resched, __cond_resched); | |
9432bbd9 PZ |
6635 | static_call_update(preempt_schedule, NULL); |
6636 | static_call_update(preempt_schedule_notrace, NULL); | |
6637 | static_call_update(irqentry_exit_cond_resched, NULL); | |
e59e10f8 PZ |
6638 | pr_info("Dynamic Preempt: voluntary\n"); |
6639 | break; | |
6640 | ||
6641 | case preempt_dynamic_full: | |
9432bbd9 PZ |
6642 | static_call_update(cond_resched, (void *)&__static_call_return0); |
6643 | static_call_update(might_resched, (void *)&__static_call_return0); | |
826bfeb3 PZI |
6644 | static_call_update(preempt_schedule, __preempt_schedule_func); |
6645 | static_call_update(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
6646 | static_call_update(irqentry_exit_cond_resched, irqentry_exit_cond_resched); | |
e59e10f8 PZ |
6647 | pr_info("Dynamic Preempt: full\n"); |
6648 | break; | |
6649 | } | |
6650 | ||
6651 | preempt_dynamic_mode = mode; | |
6652 | } | |
6653 | ||
6654 | static int __init setup_preempt_mode(char *str) | |
6655 | { | |
6656 | int mode = sched_dynamic_mode(str); | |
6657 | if (mode < 0) { | |
6658 | pr_warn("Dynamic Preempt: unsupported mode: %s\n", str); | |
826bfeb3 PZI |
6659 | return 1; |
6660 | } | |
e59e10f8 PZ |
6661 | |
6662 | sched_dynamic_update(mode); | |
826bfeb3 PZI |
6663 | return 0; |
6664 | } | |
6665 | __setup("preempt=", setup_preempt_mode); | |
6666 | ||
6667 | #endif /* CONFIG_PREEMPT_DYNAMIC */ | |
6668 | ||
1da177e4 | 6669 | /* |
a49b4f40 | 6670 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
6671 | * off of irq context. |
6672 | * Note, that this is called and return with irqs disabled. This will | |
6673 | * protect us against recursive calling from irq. | |
6674 | */ | |
722a9f92 | 6675 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 6676 | { |
b22366cd | 6677 | enum ctx_state prev_state; |
6478d880 | 6678 | |
2ed6e34f | 6679 | /* Catch callers which need to be fixed */ |
f27dde8d | 6680 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 6681 | |
b22366cd FW |
6682 | prev_state = exception_enter(); |
6683 | ||
3a5c359a | 6684 | do { |
3d8f74dd | 6685 | preempt_disable(); |
3a5c359a | 6686 | local_irq_enable(); |
b4bfa3fc | 6687 | __schedule(SM_PREEMPT); |
3a5c359a | 6688 | local_irq_disable(); |
3d8f74dd | 6689 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 6690 | } while (need_resched()); |
b22366cd FW |
6691 | |
6692 | exception_exit(prev_state); | |
1da177e4 LT |
6693 | } |
6694 | ||
ac6424b9 | 6695 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 6696 | void *key) |
1da177e4 | 6697 | { |
062d3f95 | 6698 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 6699 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 6700 | } |
1da177e4 LT |
6701 | EXPORT_SYMBOL(default_wake_function); |
6702 | ||
f558c2b8 PZ |
6703 | static void __setscheduler_prio(struct task_struct *p, int prio) |
6704 | { | |
6705 | if (dl_prio(prio)) | |
6706 | p->sched_class = &dl_sched_class; | |
6707 | else if (rt_prio(prio)) | |
6708 | p->sched_class = &rt_sched_class; | |
6709 | else | |
6710 | p->sched_class = &fair_sched_class; | |
6711 | ||
6712 | p->prio = prio; | |
6713 | } | |
6714 | ||
b29739f9 IM |
6715 | #ifdef CONFIG_RT_MUTEXES |
6716 | ||
acd58620 PZ |
6717 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
6718 | { | |
6719 | if (pi_task) | |
6720 | prio = min(prio, pi_task->prio); | |
6721 | ||
6722 | return prio; | |
6723 | } | |
6724 | ||
6725 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
6726 | { | |
6727 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
6728 | ||
6729 | return __rt_effective_prio(pi_task, prio); | |
6730 | } | |
6731 | ||
b29739f9 IM |
6732 | /* |
6733 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
6734 | * @p: task to boost |
6735 | * @pi_task: donor task | |
b29739f9 IM |
6736 | * |
6737 | * This function changes the 'effective' priority of a task. It does | |
6738 | * not touch ->normal_prio like __setscheduler(). | |
6739 | * | |
c365c292 TG |
6740 | * Used by the rt_mutex code to implement priority inheritance |
6741 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 6742 | */ |
acd58620 | 6743 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 6744 | { |
acd58620 | 6745 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 6746 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 6747 | const struct sched_class *prev_class; |
eb580751 PZ |
6748 | struct rq_flags rf; |
6749 | struct rq *rq; | |
b29739f9 | 6750 | |
acd58620 PZ |
6751 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
6752 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
6753 | ||
6754 | /* | |
6755 | * If nothing changed; bail early. | |
6756 | */ | |
6757 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
6758 | return; | |
b29739f9 | 6759 | |
eb580751 | 6760 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 6761 | update_rq_clock(rq); |
acd58620 PZ |
6762 | /* |
6763 | * Set under pi_lock && rq->lock, such that the value can be used under | |
6764 | * either lock. | |
6765 | * | |
6766 | * Note that there is loads of tricky to make this pointer cache work | |
6767 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
6768 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
6769 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 6770 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
6771 | */ |
6772 | p->pi_top_task = pi_task; | |
6773 | ||
6774 | /* | |
6775 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
6776 | */ | |
6777 | if (prio == p->prio && !dl_prio(prio)) | |
6778 | goto out_unlock; | |
b29739f9 | 6779 | |
1c4dd99b TG |
6780 | /* |
6781 | * Idle task boosting is a nono in general. There is one | |
6782 | * exception, when PREEMPT_RT and NOHZ is active: | |
6783 | * | |
6784 | * The idle task calls get_next_timer_interrupt() and holds | |
6785 | * the timer wheel base->lock on the CPU and another CPU wants | |
6786 | * to access the timer (probably to cancel it). We can safely | |
6787 | * ignore the boosting request, as the idle CPU runs this code | |
6788 | * with interrupts disabled and will complete the lock | |
6789 | * protected section without being interrupted. So there is no | |
6790 | * real need to boost. | |
6791 | */ | |
6792 | if (unlikely(p == rq->idle)) { | |
6793 | WARN_ON(p != rq->curr); | |
6794 | WARN_ON(p->pi_blocked_on); | |
6795 | goto out_unlock; | |
6796 | } | |
6797 | ||
b91473ff | 6798 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 6799 | oldprio = p->prio; |
ff77e468 PZ |
6800 | |
6801 | if (oldprio == prio) | |
6802 | queue_flag &= ~DEQUEUE_MOVE; | |
6803 | ||
83ab0aa0 | 6804 | prev_class = p->sched_class; |
da0c1e65 | 6805 | queued = task_on_rq_queued(p); |
051a1d1a | 6806 | running = task_current(rq, p); |
da0c1e65 | 6807 | if (queued) |
ff77e468 | 6808 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 6809 | if (running) |
f3cd1c4e | 6810 | put_prev_task(rq, p); |
dd41f596 | 6811 | |
2d3d891d DF |
6812 | /* |
6813 | * Boosting condition are: | |
6814 | * 1. -rt task is running and holds mutex A | |
6815 | * --> -dl task blocks on mutex A | |
6816 | * | |
6817 | * 2. -dl task is running and holds mutex A | |
6818 | * --> -dl task blocks on mutex A and could preempt the | |
6819 | * running task | |
6820 | */ | |
6821 | if (dl_prio(prio)) { | |
466af29b | 6822 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
6823 | (pi_task && dl_prio(pi_task->prio) && |
6824 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2279f540 | 6825 | p->dl.pi_se = pi_task->dl.pi_se; |
ff77e468 | 6826 | queue_flag |= ENQUEUE_REPLENISH; |
2279f540 JL |
6827 | } else { |
6828 | p->dl.pi_se = &p->dl; | |
6829 | } | |
2d3d891d DF |
6830 | } else if (rt_prio(prio)) { |
6831 | if (dl_prio(oldprio)) | |
2279f540 | 6832 | p->dl.pi_se = &p->dl; |
2d3d891d | 6833 | if (oldprio < prio) |
ff77e468 | 6834 | queue_flag |= ENQUEUE_HEAD; |
2d3d891d DF |
6835 | } else { |
6836 | if (dl_prio(oldprio)) | |
2279f540 | 6837 | p->dl.pi_se = &p->dl; |
746db944 BS |
6838 | if (rt_prio(oldprio)) |
6839 | p->rt.timeout = 0; | |
2d3d891d | 6840 | } |
dd41f596 | 6841 | |
f558c2b8 | 6842 | __setscheduler_prio(p, prio); |
b29739f9 | 6843 | |
da0c1e65 | 6844 | if (queued) |
ff77e468 | 6845 | enqueue_task(rq, p, queue_flag); |
a399d233 | 6846 | if (running) |
03b7fad1 | 6847 | set_next_task(rq, p); |
cb469845 | 6848 | |
da7a735e | 6849 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 6850 | out_unlock: |
d1ccc66d IM |
6851 | /* Avoid rq from going away on us: */ |
6852 | preempt_disable(); | |
4c9a4bc8 | 6853 | |
565790d2 PZ |
6854 | rq_unpin_lock(rq, &rf); |
6855 | __balance_callbacks(rq); | |
5cb9eaa3 | 6856 | raw_spin_rq_unlock(rq); |
565790d2 | 6857 | |
4c9a4bc8 | 6858 | preempt_enable(); |
b29739f9 | 6859 | } |
acd58620 PZ |
6860 | #else |
6861 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
6862 | { | |
6863 | return prio; | |
6864 | } | |
b29739f9 | 6865 | #endif |
d50dde5a | 6866 | |
36c8b586 | 6867 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6868 | { |
49bd21ef | 6869 | bool queued, running; |
53a23364 | 6870 | int old_prio; |
eb580751 | 6871 | struct rq_flags rf; |
70b97a7f | 6872 | struct rq *rq; |
1da177e4 | 6873 | |
75e45d51 | 6874 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
6875 | return; |
6876 | /* | |
6877 | * We have to be careful, if called from sys_setpriority(), | |
6878 | * the task might be in the middle of scheduling on another CPU. | |
6879 | */ | |
eb580751 | 6880 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
6881 | update_rq_clock(rq); |
6882 | ||
1da177e4 LT |
6883 | /* |
6884 | * The RT priorities are set via sched_setscheduler(), but we still | |
6885 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 6886 | * it won't have any effect on scheduling until the task is |
aab03e05 | 6887 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 6888 | */ |
aab03e05 | 6889 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
6890 | p->static_prio = NICE_TO_PRIO(nice); |
6891 | goto out_unlock; | |
6892 | } | |
da0c1e65 | 6893 | queued = task_on_rq_queued(p); |
49bd21ef | 6894 | running = task_current(rq, p); |
da0c1e65 | 6895 | if (queued) |
7a57f32a | 6896 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
6897 | if (running) |
6898 | put_prev_task(rq, p); | |
1da177e4 | 6899 | |
1da177e4 | 6900 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 6901 | set_load_weight(p, true); |
b29739f9 IM |
6902 | old_prio = p->prio; |
6903 | p->prio = effective_prio(p); | |
1da177e4 | 6904 | |
5443a0be | 6905 | if (queued) |
7134b3e9 | 6906 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 6907 | if (running) |
03b7fad1 | 6908 | set_next_task(rq, p); |
5443a0be FW |
6909 | |
6910 | /* | |
6911 | * If the task increased its priority or is running and | |
6912 | * lowered its priority, then reschedule its CPU: | |
6913 | */ | |
6914 | p->sched_class->prio_changed(rq, p, old_prio); | |
6915 | ||
1da177e4 | 6916 | out_unlock: |
eb580751 | 6917 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 6918 | } |
1da177e4 LT |
6919 | EXPORT_SYMBOL(set_user_nice); |
6920 | ||
e43379f1 MM |
6921 | /* |
6922 | * can_nice - check if a task can reduce its nice value | |
6923 | * @p: task | |
6924 | * @nice: nice value | |
6925 | */ | |
36c8b586 | 6926 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6927 | { |
d1ccc66d | 6928 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 6929 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 6930 | |
78d7d407 | 6931 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
6932 | capable(CAP_SYS_NICE)); |
6933 | } | |
6934 | ||
1da177e4 LT |
6935 | #ifdef __ARCH_WANT_SYS_NICE |
6936 | ||
6937 | /* | |
6938 | * sys_nice - change the priority of the current process. | |
6939 | * @increment: priority increment | |
6940 | * | |
6941 | * sys_setpriority is a more generic, but much slower function that | |
6942 | * does similar things. | |
6943 | */ | |
5add95d4 | 6944 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6945 | { |
48f24c4d | 6946 | long nice, retval; |
1da177e4 LT |
6947 | |
6948 | /* | |
6949 | * Setpriority might change our priority at the same moment. | |
6950 | * We don't have to worry. Conceptually one call occurs first | |
6951 | * and we have a single winner. | |
6952 | */ | |
a9467fa3 | 6953 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 6954 | nice = task_nice(current) + increment; |
1da177e4 | 6955 | |
a9467fa3 | 6956 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
6957 | if (increment < 0 && !can_nice(current, nice)) |
6958 | return -EPERM; | |
6959 | ||
1da177e4 LT |
6960 | retval = security_task_setnice(current, nice); |
6961 | if (retval) | |
6962 | return retval; | |
6963 | ||
6964 | set_user_nice(current, nice); | |
6965 | return 0; | |
6966 | } | |
6967 | ||
6968 | #endif | |
6969 | ||
6970 | /** | |
6971 | * task_prio - return the priority value of a given task. | |
6972 | * @p: the task in question. | |
6973 | * | |
e69f6186 | 6974 | * Return: The priority value as seen by users in /proc. |
c541bb78 DE |
6975 | * |
6976 | * sched policy return value kernel prio user prio/nice | |
6977 | * | |
6978 | * normal, batch, idle [0 ... 39] [100 ... 139] 0/[-20 ... 19] | |
6979 | * fifo, rr [-2 ... -100] [98 ... 0] [1 ... 99] | |
6980 | * deadline -101 -1 0 | |
1da177e4 | 6981 | */ |
36c8b586 | 6982 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6983 | { |
6984 | return p->prio - MAX_RT_PRIO; | |
6985 | } | |
6986 | ||
1da177e4 | 6987 | /** |
d1ccc66d | 6988 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 6989 | * @cpu: the processor in question. |
e69f6186 YB |
6990 | * |
6991 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
6992 | */ |
6993 | int idle_cpu(int cpu) | |
6994 | { | |
908a3283 TG |
6995 | struct rq *rq = cpu_rq(cpu); |
6996 | ||
6997 | if (rq->curr != rq->idle) | |
6998 | return 0; | |
6999 | ||
7000 | if (rq->nr_running) | |
7001 | return 0; | |
7002 | ||
7003 | #ifdef CONFIG_SMP | |
126c2092 | 7004 | if (rq->ttwu_pending) |
908a3283 TG |
7005 | return 0; |
7006 | #endif | |
7007 | ||
7008 | return 1; | |
1da177e4 LT |
7009 | } |
7010 | ||
943d355d RJ |
7011 | /** |
7012 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
7013 | * @cpu: the CPU in question. | |
7014 | * | |
7015 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
7016 | */ | |
7017 | int available_idle_cpu(int cpu) | |
7018 | { | |
7019 | if (!idle_cpu(cpu)) | |
7020 | return 0; | |
7021 | ||
247f2f6f RJ |
7022 | if (vcpu_is_preempted(cpu)) |
7023 | return 0; | |
7024 | ||
908a3283 | 7025 | return 1; |
1da177e4 LT |
7026 | } |
7027 | ||
1da177e4 | 7028 | /** |
d1ccc66d | 7029 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 7030 | * @cpu: the processor in question. |
e69f6186 | 7031 | * |
d1ccc66d | 7032 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 7033 | */ |
36c8b586 | 7034 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
7035 | { |
7036 | return cpu_rq(cpu)->idle; | |
7037 | } | |
7038 | ||
7d6a905f VK |
7039 | #ifdef CONFIG_SMP |
7040 | /* | |
7041 | * This function computes an effective utilization for the given CPU, to be | |
7042 | * used for frequency selection given the linear relation: f = u * f_max. | |
7043 | * | |
7044 | * The scheduler tracks the following metrics: | |
7045 | * | |
7046 | * cpu_util_{cfs,rt,dl,irq}() | |
7047 | * cpu_bw_dl() | |
7048 | * | |
7049 | * Where the cfs,rt and dl util numbers are tracked with the same metric and | |
7050 | * synchronized windows and are thus directly comparable. | |
7051 | * | |
7052 | * The cfs,rt,dl utilization are the running times measured with rq->clock_task | |
7053 | * which excludes things like IRQ and steal-time. These latter are then accrued | |
7054 | * in the irq utilization. | |
7055 | * | |
7056 | * The DL bandwidth number otoh is not a measured metric but a value computed | |
7057 | * based on the task model parameters and gives the minimal utilization | |
7058 | * required to meet deadlines. | |
7059 | */ | |
a5418be9 VK |
7060 | unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, |
7061 | unsigned long max, enum cpu_util_type type, | |
7d6a905f VK |
7062 | struct task_struct *p) |
7063 | { | |
7064 | unsigned long dl_util, util, irq; | |
7065 | struct rq *rq = cpu_rq(cpu); | |
7066 | ||
7067 | if (!uclamp_is_used() && | |
7068 | type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) { | |
7069 | return max; | |
7070 | } | |
7071 | ||
7072 | /* | |
7073 | * Early check to see if IRQ/steal time saturates the CPU, can be | |
7074 | * because of inaccuracies in how we track these -- see | |
7075 | * update_irq_load_avg(). | |
7076 | */ | |
7077 | irq = cpu_util_irq(rq); | |
7078 | if (unlikely(irq >= max)) | |
7079 | return max; | |
7080 | ||
7081 | /* | |
7082 | * Because the time spend on RT/DL tasks is visible as 'lost' time to | |
7083 | * CFS tasks and we use the same metric to track the effective | |
7084 | * utilization (PELT windows are synchronized) we can directly add them | |
7085 | * to obtain the CPU's actual utilization. | |
7086 | * | |
7087 | * CFS and RT utilization can be boosted or capped, depending on | |
7088 | * utilization clamp constraints requested by currently RUNNABLE | |
7089 | * tasks. | |
7090 | * When there are no CFS RUNNABLE tasks, clamps are released and | |
7091 | * frequency will be gracefully reduced with the utilization decay. | |
7092 | */ | |
7093 | util = util_cfs + cpu_util_rt(rq); | |
7094 | if (type == FREQUENCY_UTIL) | |
7095 | util = uclamp_rq_util_with(rq, util, p); | |
7096 | ||
7097 | dl_util = cpu_util_dl(rq); | |
7098 | ||
7099 | /* | |
7100 | * For frequency selection we do not make cpu_util_dl() a permanent part | |
7101 | * of this sum because we want to use cpu_bw_dl() later on, but we need | |
7102 | * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such | |
7103 | * that we select f_max when there is no idle time. | |
7104 | * | |
7105 | * NOTE: numerical errors or stop class might cause us to not quite hit | |
7106 | * saturation when we should -- something for later. | |
7107 | */ | |
7108 | if (util + dl_util >= max) | |
7109 | return max; | |
7110 | ||
7111 | /* | |
7112 | * OTOH, for energy computation we need the estimated running time, so | |
7113 | * include util_dl and ignore dl_bw. | |
7114 | */ | |
7115 | if (type == ENERGY_UTIL) | |
7116 | util += dl_util; | |
7117 | ||
7118 | /* | |
7119 | * There is still idle time; further improve the number by using the | |
7120 | * irq metric. Because IRQ/steal time is hidden from the task clock we | |
7121 | * need to scale the task numbers: | |
7122 | * | |
7123 | * max - irq | |
7124 | * U' = irq + --------- * U | |
7125 | * max | |
7126 | */ | |
7127 | util = scale_irq_capacity(util, irq, max); | |
7128 | util += irq; | |
7129 | ||
7130 | /* | |
7131 | * Bandwidth required by DEADLINE must always be granted while, for | |
7132 | * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism | |
7133 | * to gracefully reduce the frequency when no tasks show up for longer | |
7134 | * periods of time. | |
7135 | * | |
7136 | * Ideally we would like to set bw_dl as min/guaranteed freq and util + | |
7137 | * bw_dl as requested freq. However, cpufreq is not yet ready for such | |
7138 | * an interface. So, we only do the latter for now. | |
7139 | */ | |
7140 | if (type == FREQUENCY_UTIL) | |
7141 | util += cpu_bw_dl(rq); | |
7142 | ||
7143 | return min(max, util); | |
7144 | } | |
a5418be9 VK |
7145 | |
7146 | unsigned long sched_cpu_util(int cpu, unsigned long max) | |
7147 | { | |
7148 | return effective_cpu_util(cpu, cpu_util_cfs(cpu_rq(cpu)), max, | |
7149 | ENERGY_UTIL, NULL); | |
7150 | } | |
7d6a905f VK |
7151 | #endif /* CONFIG_SMP */ |
7152 | ||
1da177e4 LT |
7153 | /** |
7154 | * find_process_by_pid - find a process with a matching PID value. | |
7155 | * @pid: the pid in question. | |
e69f6186 YB |
7156 | * |
7157 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 7158 | */ |
a9957449 | 7159 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 7160 | { |
228ebcbe | 7161 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
7162 | } |
7163 | ||
c13db6b1 SR |
7164 | /* |
7165 | * sched_setparam() passes in -1 for its policy, to let the functions | |
7166 | * it calls know not to change it. | |
7167 | */ | |
7168 | #define SETPARAM_POLICY -1 | |
7169 | ||
c365c292 TG |
7170 | static void __setscheduler_params(struct task_struct *p, |
7171 | const struct sched_attr *attr) | |
1da177e4 | 7172 | { |
d50dde5a DF |
7173 | int policy = attr->sched_policy; |
7174 | ||
c13db6b1 | 7175 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
7176 | policy = p->policy; |
7177 | ||
1da177e4 | 7178 | p->policy = policy; |
d50dde5a | 7179 | |
aab03e05 DF |
7180 | if (dl_policy(policy)) |
7181 | __setparam_dl(p, attr); | |
39fd8fd2 | 7182 | else if (fair_policy(policy)) |
d50dde5a DF |
7183 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
7184 | ||
39fd8fd2 PZ |
7185 | /* |
7186 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
7187 | * !rt_policy. Always setting this ensures that things like | |
7188 | * getparam()/getattr() don't report silly values for !rt tasks. | |
7189 | */ | |
7190 | p->rt_priority = attr->sched_priority; | |
383afd09 | 7191 | p->normal_prio = normal_prio(p); |
9059393e | 7192 | set_load_weight(p, true); |
c365c292 | 7193 | } |
39fd8fd2 | 7194 | |
c69e8d9c | 7195 | /* |
d1ccc66d | 7196 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
7197 | */ |
7198 | static bool check_same_owner(struct task_struct *p) | |
7199 | { | |
7200 | const struct cred *cred = current_cred(), *pcred; | |
7201 | bool match; | |
7202 | ||
7203 | rcu_read_lock(); | |
7204 | pcred = __task_cred(p); | |
9c806aa0 EB |
7205 | match = (uid_eq(cred->euid, pcred->euid) || |
7206 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
7207 | rcu_read_unlock(); |
7208 | return match; | |
7209 | } | |
7210 | ||
d50dde5a DF |
7211 | static int __sched_setscheduler(struct task_struct *p, |
7212 | const struct sched_attr *attr, | |
dbc7f069 | 7213 | bool user, bool pi) |
1da177e4 | 7214 | { |
f558c2b8 PZ |
7215 | int oldpolicy = -1, policy = attr->sched_policy; |
7216 | int retval, oldprio, newprio, queued, running; | |
83ab0aa0 | 7217 | const struct sched_class *prev_class; |
565790d2 | 7218 | struct callback_head *head; |
eb580751 | 7219 | struct rq_flags rf; |
ca94c442 | 7220 | int reset_on_fork; |
7a57f32a | 7221 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 7222 | struct rq *rq; |
1da177e4 | 7223 | |
896bbb25 SRV |
7224 | /* The pi code expects interrupts enabled */ |
7225 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 7226 | recheck: |
d1ccc66d | 7227 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
7228 | if (policy < 0) { |
7229 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 7230 | policy = oldpolicy = p->policy; |
ca94c442 | 7231 | } else { |
7479f3c9 | 7232 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 7233 | |
20f9cd2a | 7234 | if (!valid_policy(policy)) |
ca94c442 LP |
7235 | return -EINVAL; |
7236 | } | |
7237 | ||
794a56eb | 7238 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
7239 | return -EINVAL; |
7240 | ||
1da177e4 LT |
7241 | /* |
7242 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
ae18ad28 | 7243 | * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL, |
dd41f596 | 7244 | * SCHED_BATCH and SCHED_IDLE is 0. |
1da177e4 | 7245 | */ |
ae18ad28 | 7246 | if (attr->sched_priority > MAX_RT_PRIO-1) |
1da177e4 | 7247 | return -EINVAL; |
aab03e05 DF |
7248 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
7249 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
7250 | return -EINVAL; |
7251 | ||
37e4ab3f OC |
7252 | /* |
7253 | * Allow unprivileged RT tasks to decrease priority: | |
7254 | */ | |
961ccddd | 7255 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 7256 | if (fair_policy(policy)) { |
d0ea0268 | 7257 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 7258 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
7259 | return -EPERM; |
7260 | } | |
7261 | ||
e05606d3 | 7262 | if (rt_policy(policy)) { |
a44702e8 ON |
7263 | unsigned long rlim_rtprio = |
7264 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 7265 | |
d1ccc66d | 7266 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
7267 | if (policy != p->policy && !rlim_rtprio) |
7268 | return -EPERM; | |
7269 | ||
d1ccc66d | 7270 | /* Can't increase priority: */ |
d50dde5a DF |
7271 | if (attr->sched_priority > p->rt_priority && |
7272 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
7273 | return -EPERM; |
7274 | } | |
c02aa73b | 7275 | |
d44753b8 JL |
7276 | /* |
7277 | * Can't set/change SCHED_DEADLINE policy at all for now | |
7278 | * (safest behavior); in the future we would like to allow | |
7279 | * unprivileged DL tasks to increase their relative deadline | |
7280 | * or reduce their runtime (both ways reducing utilization) | |
7281 | */ | |
7282 | if (dl_policy(policy)) | |
7283 | return -EPERM; | |
7284 | ||
dd41f596 | 7285 | /* |
c02aa73b DH |
7286 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
7287 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 7288 | */ |
1da1843f | 7289 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 7290 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
7291 | return -EPERM; |
7292 | } | |
5fe1d75f | 7293 | |
d1ccc66d | 7294 | /* Can't change other user's priorities: */ |
c69e8d9c | 7295 | if (!check_same_owner(p)) |
37e4ab3f | 7296 | return -EPERM; |
ca94c442 | 7297 | |
d1ccc66d | 7298 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
7299 | if (p->sched_reset_on_fork && !reset_on_fork) |
7300 | return -EPERM; | |
37e4ab3f | 7301 | } |
1da177e4 | 7302 | |
725aad24 | 7303 | if (user) { |
794a56eb JL |
7304 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
7305 | return -EINVAL; | |
7306 | ||
b0ae1981 | 7307 | retval = security_task_setscheduler(p); |
725aad24 JF |
7308 | if (retval) |
7309 | return retval; | |
7310 | } | |
7311 | ||
a509a7cd PB |
7312 | /* Update task specific "requested" clamps */ |
7313 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
7314 | retval = uclamp_validate(p, attr); | |
7315 | if (retval) | |
7316 | return retval; | |
7317 | } | |
7318 | ||
710da3c8 JL |
7319 | if (pi) |
7320 | cpuset_read_lock(); | |
7321 | ||
b29739f9 | 7322 | /* |
d1ccc66d | 7323 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 7324 | * changing the priority of the task: |
0122ec5b | 7325 | * |
25985edc | 7326 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
7327 | * runqueue lock must be held. |
7328 | */ | |
eb580751 | 7329 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 7330 | update_rq_clock(rq); |
dc61b1d6 | 7331 | |
34f971f6 | 7332 | /* |
d1ccc66d | 7333 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
7334 | */ |
7335 | if (p == rq->stop) { | |
4b211f2b MP |
7336 | retval = -EINVAL; |
7337 | goto unlock; | |
34f971f6 PZ |
7338 | } |
7339 | ||
a51e9198 | 7340 | /* |
d6b1e911 TG |
7341 | * If not changing anything there's no need to proceed further, |
7342 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 7343 | */ |
d50dde5a | 7344 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 7345 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
7346 | goto change; |
7347 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
7348 | goto change; | |
75381608 | 7349 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 7350 | goto change; |
a509a7cd PB |
7351 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
7352 | goto change; | |
d50dde5a | 7353 | |
d6b1e911 | 7354 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
7355 | retval = 0; |
7356 | goto unlock; | |
a51e9198 | 7357 | } |
d50dde5a | 7358 | change: |
a51e9198 | 7359 | |
dc61b1d6 | 7360 | if (user) { |
332ac17e | 7361 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
7362 | /* |
7363 | * Do not allow realtime tasks into groups that have no runtime | |
7364 | * assigned. | |
7365 | */ | |
7366 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
7367 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
7368 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
7369 | retval = -EPERM; |
7370 | goto unlock; | |
dc61b1d6 | 7371 | } |
dc61b1d6 | 7372 | #endif |
332ac17e | 7373 | #ifdef CONFIG_SMP |
794a56eb JL |
7374 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
7375 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 7376 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
7377 | |
7378 | /* | |
7379 | * Don't allow tasks with an affinity mask smaller than | |
7380 | * the entire root_domain to become SCHED_DEADLINE. We | |
7381 | * will also fail if there's no bandwidth available. | |
7382 | */ | |
3bd37062 | 7383 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 7384 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
7385 | retval = -EPERM; |
7386 | goto unlock; | |
332ac17e DF |
7387 | } |
7388 | } | |
7389 | #endif | |
7390 | } | |
dc61b1d6 | 7391 | |
d1ccc66d | 7392 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
7393 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
7394 | policy = oldpolicy = -1; | |
eb580751 | 7395 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
7396 | if (pi) |
7397 | cpuset_read_unlock(); | |
1da177e4 LT |
7398 | goto recheck; |
7399 | } | |
332ac17e DF |
7400 | |
7401 | /* | |
7402 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
7403 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
7404 | * is available. | |
7405 | */ | |
06a76fe0 | 7406 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
7407 | retval = -EBUSY; |
7408 | goto unlock; | |
332ac17e DF |
7409 | } |
7410 | ||
c365c292 TG |
7411 | p->sched_reset_on_fork = reset_on_fork; |
7412 | oldprio = p->prio; | |
7413 | ||
f558c2b8 | 7414 | newprio = __normal_prio(policy, attr->sched_priority, attr->sched_nice); |
dbc7f069 PZ |
7415 | if (pi) { |
7416 | /* | |
7417 | * Take priority boosted tasks into account. If the new | |
7418 | * effective priority is unchanged, we just store the new | |
7419 | * normal parameters and do not touch the scheduler class and | |
7420 | * the runqueue. This will be done when the task deboost | |
7421 | * itself. | |
7422 | */ | |
f558c2b8 PZ |
7423 | newprio = rt_effective_prio(p, newprio); |
7424 | if (newprio == oldprio) | |
ff77e468 | 7425 | queue_flags &= ~DEQUEUE_MOVE; |
c365c292 TG |
7426 | } |
7427 | ||
da0c1e65 | 7428 | queued = task_on_rq_queued(p); |
051a1d1a | 7429 | running = task_current(rq, p); |
da0c1e65 | 7430 | if (queued) |
ff77e468 | 7431 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 7432 | if (running) |
f3cd1c4e | 7433 | put_prev_task(rq, p); |
f6b53205 | 7434 | |
83ab0aa0 | 7435 | prev_class = p->sched_class; |
a509a7cd | 7436 | |
f558c2b8 PZ |
7437 | if (!(attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)) { |
7438 | __setscheduler_params(p, attr); | |
7439 | __setscheduler_prio(p, newprio); | |
7440 | } | |
a509a7cd | 7441 | __setscheduler_uclamp(p, attr); |
f6b53205 | 7442 | |
da0c1e65 | 7443 | if (queued) { |
81a44c54 TG |
7444 | /* |
7445 | * We enqueue to tail when the priority of a task is | |
7446 | * increased (user space view). | |
7447 | */ | |
ff77e468 PZ |
7448 | if (oldprio < p->prio) |
7449 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 7450 | |
ff77e468 | 7451 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 7452 | } |
a399d233 | 7453 | if (running) |
03b7fad1 | 7454 | set_next_task(rq, p); |
cb469845 | 7455 | |
da7a735e | 7456 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
7457 | |
7458 | /* Avoid rq from going away on us: */ | |
7459 | preempt_disable(); | |
565790d2 | 7460 | head = splice_balance_callbacks(rq); |
eb580751 | 7461 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 7462 | |
710da3c8 JL |
7463 | if (pi) { |
7464 | cpuset_read_unlock(); | |
dbc7f069 | 7465 | rt_mutex_adjust_pi(p); |
710da3c8 | 7466 | } |
95e02ca9 | 7467 | |
d1ccc66d | 7468 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 7469 | balance_callbacks(rq, head); |
4c9a4bc8 | 7470 | preempt_enable(); |
95e02ca9 | 7471 | |
1da177e4 | 7472 | return 0; |
4b211f2b MP |
7473 | |
7474 | unlock: | |
7475 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
7476 | if (pi) |
7477 | cpuset_read_unlock(); | |
4b211f2b | 7478 | return retval; |
1da177e4 | 7479 | } |
961ccddd | 7480 | |
7479f3c9 PZ |
7481 | static int _sched_setscheduler(struct task_struct *p, int policy, |
7482 | const struct sched_param *param, bool check) | |
7483 | { | |
7484 | struct sched_attr attr = { | |
7485 | .sched_policy = policy, | |
7486 | .sched_priority = param->sched_priority, | |
7487 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
7488 | }; | |
7489 | ||
c13db6b1 SR |
7490 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
7491 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
7492 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
7493 | policy &= ~SCHED_RESET_ON_FORK; | |
7494 | attr.sched_policy = policy; | |
7495 | } | |
7496 | ||
dbc7f069 | 7497 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 7498 | } |
961ccddd RR |
7499 | /** |
7500 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
7501 | * @p: the task in question. | |
7502 | * @policy: new policy. | |
7503 | * @param: structure containing the new RT priority. | |
7504 | * | |
7318d4cc PZ |
7505 | * Use sched_set_fifo(), read its comment. |
7506 | * | |
e69f6186 YB |
7507 | * Return: 0 on success. An error code otherwise. |
7508 | * | |
961ccddd RR |
7509 | * NOTE that the task may be already dead. |
7510 | */ | |
7511 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 7512 | const struct sched_param *param) |
961ccddd | 7513 | { |
7479f3c9 | 7514 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 7515 | } |
1da177e4 | 7516 | |
d50dde5a DF |
7517 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
7518 | { | |
dbc7f069 | 7519 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 7520 | } |
d50dde5a | 7521 | |
794a56eb JL |
7522 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
7523 | { | |
7524 | return __sched_setscheduler(p, attr, false, true); | |
7525 | } | |
1eb5dde6 | 7526 | EXPORT_SYMBOL_GPL(sched_setattr_nocheck); |
794a56eb | 7527 | |
961ccddd RR |
7528 | /** |
7529 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
7530 | * @p: the task in question. | |
7531 | * @policy: new policy. | |
7532 | * @param: structure containing the new RT priority. | |
7533 | * | |
7534 | * Just like sched_setscheduler, only don't bother checking if the | |
7535 | * current context has permission. For example, this is needed in | |
7536 | * stop_machine(): we create temporary high priority worker threads, | |
7537 | * but our caller might not have that capability. | |
e69f6186 YB |
7538 | * |
7539 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
7540 | */ |
7541 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 7542 | const struct sched_param *param) |
961ccddd | 7543 | { |
7479f3c9 | 7544 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
7545 | } |
7546 | ||
7318d4cc PZ |
7547 | /* |
7548 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
7549 | * incapable of resource management, which is the one thing an OS really should | |
7550 | * be doing. | |
7551 | * | |
7552 | * This is of course the reason it is limited to privileged users only. | |
7553 | * | |
7554 | * Worse still; it is fundamentally impossible to compose static priority | |
7555 | * workloads. You cannot take two correctly working static prio workloads | |
7556 | * and smash them together and still expect them to work. | |
7557 | * | |
7558 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
7559 | * | |
7560 | * MAX_RT_PRIO / 2 | |
7561 | * | |
7562 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
7563 | * know enough information to make a sensible choice. | |
7564 | */ | |
8b700983 | 7565 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
7566 | { |
7567 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 7568 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7569 | } |
7570 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
7571 | ||
7572 | /* | |
7573 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
7574 | */ | |
8b700983 | 7575 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
7576 | { |
7577 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 7578 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7579 | } |
7580 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
7581 | ||
8b700983 | 7582 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
7583 | { |
7584 | struct sched_attr attr = { | |
7585 | .sched_policy = SCHED_NORMAL, | |
7586 | .sched_nice = nice, | |
7587 | }; | |
8b700983 | 7588 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
7589 | } |
7590 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 7591 | |
95cdf3b7 IM |
7592 | static int |
7593 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 7594 | { |
1da177e4 LT |
7595 | struct sched_param lparam; |
7596 | struct task_struct *p; | |
36c8b586 | 7597 | int retval; |
1da177e4 LT |
7598 | |
7599 | if (!param || pid < 0) | |
7600 | return -EINVAL; | |
7601 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
7602 | return -EFAULT; | |
5fe1d75f ON |
7603 | |
7604 | rcu_read_lock(); | |
7605 | retval = -ESRCH; | |
1da177e4 | 7606 | p = find_process_by_pid(pid); |
710da3c8 JL |
7607 | if (likely(p)) |
7608 | get_task_struct(p); | |
5fe1d75f | 7609 | rcu_read_unlock(); |
36c8b586 | 7610 | |
710da3c8 JL |
7611 | if (likely(p)) { |
7612 | retval = sched_setscheduler(p, policy, &lparam); | |
7613 | put_task_struct(p); | |
7614 | } | |
7615 | ||
1da177e4 LT |
7616 | return retval; |
7617 | } | |
7618 | ||
d50dde5a DF |
7619 | /* |
7620 | * Mimics kernel/events/core.c perf_copy_attr(). | |
7621 | */ | |
d1ccc66d | 7622 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
7623 | { |
7624 | u32 size; | |
7625 | int ret; | |
7626 | ||
d1ccc66d | 7627 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
7628 | memset(attr, 0, sizeof(*attr)); |
7629 | ||
7630 | ret = get_user(size, &uattr->size); | |
7631 | if (ret) | |
7632 | return ret; | |
7633 | ||
d1ccc66d IM |
7634 | /* ABI compatibility quirk: */ |
7635 | if (!size) | |
d50dde5a | 7636 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 7637 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
7638 | goto err_size; |
7639 | ||
dff3a85f AS |
7640 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
7641 | if (ret) { | |
7642 | if (ret == -E2BIG) | |
7643 | goto err_size; | |
7644 | return ret; | |
d50dde5a DF |
7645 | } |
7646 | ||
a509a7cd PB |
7647 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
7648 | size < SCHED_ATTR_SIZE_VER1) | |
7649 | return -EINVAL; | |
7650 | ||
d50dde5a | 7651 | /* |
d1ccc66d | 7652 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
7653 | * to be strict and return an error on out-of-bounds values? |
7654 | */ | |
75e45d51 | 7655 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 7656 | |
e78c7bca | 7657 | return 0; |
d50dde5a DF |
7658 | |
7659 | err_size: | |
7660 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 7661 | return -E2BIG; |
d50dde5a DF |
7662 | } |
7663 | ||
f4dddf90 QP |
7664 | static void get_params(struct task_struct *p, struct sched_attr *attr) |
7665 | { | |
7666 | if (task_has_dl_policy(p)) | |
7667 | __getparam_dl(p, attr); | |
7668 | else if (task_has_rt_policy(p)) | |
7669 | attr->sched_priority = p->rt_priority; | |
7670 | else | |
7671 | attr->sched_nice = task_nice(p); | |
7672 | } | |
7673 | ||
1da177e4 LT |
7674 | /** |
7675 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
7676 | * @pid: the pid in question. | |
7677 | * @policy: new policy. | |
7678 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7679 | * |
7680 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7681 | */ |
d1ccc66d | 7682 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 7683 | { |
c21761f1 JB |
7684 | if (policy < 0) |
7685 | return -EINVAL; | |
7686 | ||
1da177e4 LT |
7687 | return do_sched_setscheduler(pid, policy, param); |
7688 | } | |
7689 | ||
7690 | /** | |
7691 | * sys_sched_setparam - set/change the RT priority of a thread | |
7692 | * @pid: the pid in question. | |
7693 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7694 | * |
7695 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7696 | */ |
5add95d4 | 7697 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7698 | { |
c13db6b1 | 7699 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
7700 | } |
7701 | ||
d50dde5a DF |
7702 | /** |
7703 | * sys_sched_setattr - same as above, but with extended sched_attr | |
7704 | * @pid: the pid in question. | |
5778fccf | 7705 | * @uattr: structure containing the extended parameters. |
db66d756 | 7706 | * @flags: for future extension. |
d50dde5a | 7707 | */ |
6d35ab48 PZ |
7708 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
7709 | unsigned int, flags) | |
d50dde5a DF |
7710 | { |
7711 | struct sched_attr attr; | |
7712 | struct task_struct *p; | |
7713 | int retval; | |
7714 | ||
6d35ab48 | 7715 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
7716 | return -EINVAL; |
7717 | ||
143cf23d MK |
7718 | retval = sched_copy_attr(uattr, &attr); |
7719 | if (retval) | |
7720 | return retval; | |
d50dde5a | 7721 | |
b14ed2c2 | 7722 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 7723 | return -EINVAL; |
1d6362fa PB |
7724 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
7725 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
7726 | |
7727 | rcu_read_lock(); | |
7728 | retval = -ESRCH; | |
7729 | p = find_process_by_pid(pid); | |
a509a7cd PB |
7730 | if (likely(p)) |
7731 | get_task_struct(p); | |
d50dde5a DF |
7732 | rcu_read_unlock(); |
7733 | ||
a509a7cd | 7734 | if (likely(p)) { |
f4dddf90 QP |
7735 | if (attr.sched_flags & SCHED_FLAG_KEEP_PARAMS) |
7736 | get_params(p, &attr); | |
a509a7cd PB |
7737 | retval = sched_setattr(p, &attr); |
7738 | put_task_struct(p); | |
7739 | } | |
7740 | ||
d50dde5a DF |
7741 | return retval; |
7742 | } | |
7743 | ||
1da177e4 LT |
7744 | /** |
7745 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
7746 | * @pid: the pid in question. | |
e69f6186 YB |
7747 | * |
7748 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
7749 | * code. | |
1da177e4 | 7750 | */ |
5add95d4 | 7751 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 7752 | { |
36c8b586 | 7753 | struct task_struct *p; |
3a5c359a | 7754 | int retval; |
1da177e4 LT |
7755 | |
7756 | if (pid < 0) | |
3a5c359a | 7757 | return -EINVAL; |
1da177e4 LT |
7758 | |
7759 | retval = -ESRCH; | |
5fe85be0 | 7760 | rcu_read_lock(); |
1da177e4 LT |
7761 | p = find_process_by_pid(pid); |
7762 | if (p) { | |
7763 | retval = security_task_getscheduler(p); | |
7764 | if (!retval) | |
ca94c442 LP |
7765 | retval = p->policy |
7766 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 7767 | } |
5fe85be0 | 7768 | rcu_read_unlock(); |
1da177e4 LT |
7769 | return retval; |
7770 | } | |
7771 | ||
7772 | /** | |
ca94c442 | 7773 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
7774 | * @pid: the pid in question. |
7775 | * @param: structure containing the RT priority. | |
e69f6186 YB |
7776 | * |
7777 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
7778 | * code. | |
1da177e4 | 7779 | */ |
5add95d4 | 7780 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7781 | { |
ce5f7f82 | 7782 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 7783 | struct task_struct *p; |
3a5c359a | 7784 | int retval; |
1da177e4 LT |
7785 | |
7786 | if (!param || pid < 0) | |
3a5c359a | 7787 | return -EINVAL; |
1da177e4 | 7788 | |
5fe85be0 | 7789 | rcu_read_lock(); |
1da177e4 LT |
7790 | p = find_process_by_pid(pid); |
7791 | retval = -ESRCH; | |
7792 | if (!p) | |
7793 | goto out_unlock; | |
7794 | ||
7795 | retval = security_task_getscheduler(p); | |
7796 | if (retval) | |
7797 | goto out_unlock; | |
7798 | ||
ce5f7f82 PZ |
7799 | if (task_has_rt_policy(p)) |
7800 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 7801 | rcu_read_unlock(); |
1da177e4 LT |
7802 | |
7803 | /* | |
7804 | * This one might sleep, we cannot do it with a spinlock held ... | |
7805 | */ | |
7806 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
7807 | ||
1da177e4 LT |
7808 | return retval; |
7809 | ||
7810 | out_unlock: | |
5fe85be0 | 7811 | rcu_read_unlock(); |
1da177e4 LT |
7812 | return retval; |
7813 | } | |
7814 | ||
1251201c IM |
7815 | /* |
7816 | * Copy the kernel size attribute structure (which might be larger | |
7817 | * than what user-space knows about) to user-space. | |
7818 | * | |
7819 | * Note that all cases are valid: user-space buffer can be larger or | |
7820 | * smaller than the kernel-space buffer. The usual case is that both | |
7821 | * have the same size. | |
7822 | */ | |
7823 | static int | |
7824 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
7825 | struct sched_attr *kattr, | |
7826 | unsigned int usize) | |
d50dde5a | 7827 | { |
1251201c | 7828 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 7829 | |
96d4f267 | 7830 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
7831 | return -EFAULT; |
7832 | ||
7833 | /* | |
1251201c IM |
7834 | * sched_getattr() ABI forwards and backwards compatibility: |
7835 | * | |
7836 | * If usize == ksize then we just copy everything to user-space and all is good. | |
7837 | * | |
7838 | * If usize < ksize then we only copy as much as user-space has space for, | |
7839 | * this keeps ABI compatibility as well. We skip the rest. | |
7840 | * | |
7841 | * If usize > ksize then user-space is using a newer version of the ABI, | |
7842 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
7843 | * detect the kernel's knowledge of attributes from the attr->size value | |
7844 | * which is set to ksize in this case. | |
d50dde5a | 7845 | */ |
1251201c | 7846 | kattr->size = min(usize, ksize); |
d50dde5a | 7847 | |
1251201c | 7848 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
7849 | return -EFAULT; |
7850 | ||
22400674 | 7851 | return 0; |
d50dde5a DF |
7852 | } |
7853 | ||
7854 | /** | |
aab03e05 | 7855 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 7856 | * @pid: the pid in question. |
5778fccf | 7857 | * @uattr: structure containing the extended parameters. |
dff3a85f | 7858 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 7859 | * @flags: for future extension. |
d50dde5a | 7860 | */ |
6d35ab48 | 7861 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 7862 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 7863 | { |
1251201c | 7864 | struct sched_attr kattr = { }; |
d50dde5a DF |
7865 | struct task_struct *p; |
7866 | int retval; | |
7867 | ||
1251201c IM |
7868 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
7869 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
7870 | return -EINVAL; |
7871 | ||
7872 | rcu_read_lock(); | |
7873 | p = find_process_by_pid(pid); | |
7874 | retval = -ESRCH; | |
7875 | if (!p) | |
7876 | goto out_unlock; | |
7877 | ||
7878 | retval = security_task_getscheduler(p); | |
7879 | if (retval) | |
7880 | goto out_unlock; | |
7881 | ||
1251201c | 7882 | kattr.sched_policy = p->policy; |
7479f3c9 | 7883 | if (p->sched_reset_on_fork) |
1251201c | 7884 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
f4dddf90 | 7885 | get_params(p, &kattr); |
7ad721bf | 7886 | kattr.sched_flags &= SCHED_FLAG_ALL; |
d50dde5a | 7887 | |
a509a7cd | 7888 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
7889 | /* |
7890 | * This could race with another potential updater, but this is fine | |
7891 | * because it'll correctly read the old or the new value. We don't need | |
7892 | * to guarantee who wins the race as long as it doesn't return garbage. | |
7893 | */ | |
1251201c IM |
7894 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
7895 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
7896 | #endif |
7897 | ||
d50dde5a DF |
7898 | rcu_read_unlock(); |
7899 | ||
1251201c | 7900 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
7901 | |
7902 | out_unlock: | |
7903 | rcu_read_unlock(); | |
7904 | return retval; | |
7905 | } | |
7906 | ||
234b8ab6 WD |
7907 | #ifdef CONFIG_SMP |
7908 | int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask) | |
1da177e4 | 7909 | { |
234b8ab6 WD |
7910 | int ret = 0; |
7911 | ||
7912 | /* | |
7913 | * If the task isn't a deadline task or admission control is | |
7914 | * disabled then we don't care about affinity changes. | |
7915 | */ | |
7916 | if (!task_has_dl_policy(p) || !dl_bandwidth_enabled()) | |
7917 | return 0; | |
7918 | ||
7919 | /* | |
7920 | * Since bandwidth control happens on root_domain basis, | |
7921 | * if admission test is enabled, we only admit -deadline | |
7922 | * tasks allowed to run on all the CPUs in the task's | |
7923 | * root_domain. | |
7924 | */ | |
7925 | rcu_read_lock(); | |
7926 | if (!cpumask_subset(task_rq(p)->rd->span, mask)) | |
7927 | ret = -EBUSY; | |
7928 | rcu_read_unlock(); | |
7929 | return ret; | |
7930 | } | |
7931 | #endif | |
7932 | ||
db3b02ae WD |
7933 | static int |
7934 | __sched_setaffinity(struct task_struct *p, const struct cpumask *mask) | |
1da177e4 | 7935 | { |
36c8b586 | 7936 | int retval; |
5a16f3d3 | 7937 | cpumask_var_t cpus_allowed, new_mask; |
1da177e4 | 7938 | |
db3b02ae WD |
7939 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) |
7940 | return -ENOMEM; | |
1da177e4 | 7941 | |
5a16f3d3 RR |
7942 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { |
7943 | retval = -ENOMEM; | |
7944 | goto out_free_cpus_allowed; | |
7945 | } | |
e4099a5e PZ |
7946 | |
7947 | cpuset_cpus_allowed(p, cpus_allowed); | |
db3b02ae | 7948 | cpumask_and(new_mask, mask, cpus_allowed); |
e4099a5e | 7949 | |
234b8ab6 WD |
7950 | retval = dl_task_check_affinity(p, new_mask); |
7951 | if (retval) | |
7952 | goto out_free_new_mask; | |
49246274 | 7953 | again: |
07ec77a1 | 7954 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK | SCA_USER); |
db3b02ae WD |
7955 | if (retval) |
7956 | goto out_free_new_mask; | |
1da177e4 | 7957 | |
db3b02ae WD |
7958 | cpuset_cpus_allowed(p, cpus_allowed); |
7959 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
7960 | /* | |
7961 | * We must have raced with a concurrent cpuset update. | |
7962 | * Just reset the cpumask to the cpuset's cpus_allowed. | |
7963 | */ | |
7964 | cpumask_copy(new_mask, cpus_allowed); | |
7965 | goto again; | |
8707d8b8 | 7966 | } |
db3b02ae | 7967 | |
16303ab2 | 7968 | out_free_new_mask: |
5a16f3d3 RR |
7969 | free_cpumask_var(new_mask); |
7970 | out_free_cpus_allowed: | |
7971 | free_cpumask_var(cpus_allowed); | |
db3b02ae WD |
7972 | return retval; |
7973 | } | |
7974 | ||
7975 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) | |
7976 | { | |
36c8b586 IM |
7977 | struct task_struct *p; |
7978 | int retval; | |
1da177e4 | 7979 | |
23f5d142 | 7980 | rcu_read_lock(); |
1da177e4 LT |
7981 | |
7982 | p = find_process_by_pid(pid); | |
7983 | if (!p) { | |
23f5d142 | 7984 | rcu_read_unlock(); |
1da177e4 LT |
7985 | return -ESRCH; |
7986 | } | |
7987 | ||
23f5d142 | 7988 | /* Prevent p going away */ |
1da177e4 | 7989 | get_task_struct(p); |
23f5d142 | 7990 | rcu_read_unlock(); |
1da177e4 | 7991 | |
14a40ffc TH |
7992 | if (p->flags & PF_NO_SETAFFINITY) { |
7993 | retval = -EINVAL; | |
7994 | goto out_put_task; | |
7995 | } | |
db3b02ae | 7996 | |
4c44aaaf EB |
7997 | if (!check_same_owner(p)) { |
7998 | rcu_read_lock(); | |
7999 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
8000 | rcu_read_unlock(); | |
db3b02ae WD |
8001 | retval = -EPERM; |
8002 | goto out_put_task; | |
4c44aaaf EB |
8003 | } |
8004 | rcu_read_unlock(); | |
8005 | } | |
1da177e4 | 8006 | |
b0ae1981 | 8007 | retval = security_task_setscheduler(p); |
e7834f8f | 8008 | if (retval) |
db3b02ae | 8009 | goto out_put_task; |
1da177e4 | 8010 | |
db3b02ae | 8011 | retval = __sched_setaffinity(p, in_mask); |
5a16f3d3 | 8012 | out_put_task: |
1da177e4 | 8013 | put_task_struct(p); |
1da177e4 LT |
8014 | return retval; |
8015 | } | |
8016 | ||
8017 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 8018 | struct cpumask *new_mask) |
1da177e4 | 8019 | { |
96f874e2 RR |
8020 | if (len < cpumask_size()) |
8021 | cpumask_clear(new_mask); | |
8022 | else if (len > cpumask_size()) | |
8023 | len = cpumask_size(); | |
8024 | ||
1da177e4 LT |
8025 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
8026 | } | |
8027 | ||
8028 | /** | |
d1ccc66d | 8029 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
8030 | * @pid: pid of the process |
8031 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8032 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
8033 | * |
8034 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 8035 | */ |
5add95d4 HC |
8036 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
8037 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 8038 | { |
5a16f3d3 | 8039 | cpumask_var_t new_mask; |
1da177e4 LT |
8040 | int retval; |
8041 | ||
5a16f3d3 RR |
8042 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
8043 | return -ENOMEM; | |
1da177e4 | 8044 | |
5a16f3d3 RR |
8045 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
8046 | if (retval == 0) | |
8047 | retval = sched_setaffinity(pid, new_mask); | |
8048 | free_cpumask_var(new_mask); | |
8049 | return retval; | |
1da177e4 LT |
8050 | } |
8051 | ||
96f874e2 | 8052 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 8053 | { |
36c8b586 | 8054 | struct task_struct *p; |
31605683 | 8055 | unsigned long flags; |
1da177e4 | 8056 | int retval; |
1da177e4 | 8057 | |
23f5d142 | 8058 | rcu_read_lock(); |
1da177e4 LT |
8059 | |
8060 | retval = -ESRCH; | |
8061 | p = find_process_by_pid(pid); | |
8062 | if (!p) | |
8063 | goto out_unlock; | |
8064 | ||
e7834f8f DQ |
8065 | retval = security_task_getscheduler(p); |
8066 | if (retval) | |
8067 | goto out_unlock; | |
8068 | ||
013fdb80 | 8069 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 8070 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 8071 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
8072 | |
8073 | out_unlock: | |
23f5d142 | 8074 | rcu_read_unlock(); |
1da177e4 | 8075 | |
9531b62f | 8076 | return retval; |
1da177e4 LT |
8077 | } |
8078 | ||
8079 | /** | |
d1ccc66d | 8080 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
8081 | * @pid: pid of the process |
8082 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8083 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 8084 | * |
599b4840 ZW |
8085 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
8086 | * error code otherwise. | |
1da177e4 | 8087 | */ |
5add95d4 HC |
8088 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
8089 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
8090 | { |
8091 | int ret; | |
f17c8607 | 8092 | cpumask_var_t mask; |
1da177e4 | 8093 | |
84fba5ec | 8094 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
8095 | return -EINVAL; |
8096 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
8097 | return -EINVAL; |
8098 | ||
f17c8607 RR |
8099 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
8100 | return -ENOMEM; | |
1da177e4 | 8101 | |
f17c8607 RR |
8102 | ret = sched_getaffinity(pid, mask); |
8103 | if (ret == 0) { | |
4de373a1 | 8104 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
8105 | |
8106 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
8107 | ret = -EFAULT; |
8108 | else | |
cd3d8031 | 8109 | ret = retlen; |
f17c8607 RR |
8110 | } |
8111 | free_cpumask_var(mask); | |
1da177e4 | 8112 | |
f17c8607 | 8113 | return ret; |
1da177e4 LT |
8114 | } |
8115 | ||
7d4dd4f1 | 8116 | static void do_sched_yield(void) |
1da177e4 | 8117 | { |
8a8c69c3 PZ |
8118 | struct rq_flags rf; |
8119 | struct rq *rq; | |
8120 | ||
246b3b33 | 8121 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 8122 | |
ae92882e | 8123 | schedstat_inc(rq->yld_count); |
4530d7ab | 8124 | current->sched_class->yield_task(rq); |
1da177e4 | 8125 | |
8a8c69c3 | 8126 | preempt_disable(); |
345a957f | 8127 | rq_unlock_irq(rq, &rf); |
ba74c144 | 8128 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
8129 | |
8130 | schedule(); | |
7d4dd4f1 | 8131 | } |
1da177e4 | 8132 | |
59a74b15 MCC |
8133 | /** |
8134 | * sys_sched_yield - yield the current processor to other threads. | |
8135 | * | |
8136 | * This function yields the current CPU to other tasks. If there are no | |
8137 | * other threads running on this CPU then this function will return. | |
8138 | * | |
8139 | * Return: 0. | |
8140 | */ | |
7d4dd4f1 DB |
8141 | SYSCALL_DEFINE0(sched_yield) |
8142 | { | |
8143 | do_sched_yield(); | |
1da177e4 LT |
8144 | return 0; |
8145 | } | |
8146 | ||
b965f1dd PZI |
8147 | #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) |
8148 | int __sched __cond_resched(void) | |
1da177e4 | 8149 | { |
fe32d3cd | 8150 | if (should_resched(0)) { |
a18b5d01 | 8151 | preempt_schedule_common(); |
1da177e4 LT |
8152 | return 1; |
8153 | } | |
50895825 FW |
8154 | /* |
8155 | * In preemptible kernels, ->rcu_read_lock_nesting tells the tick | |
8156 | * whether the current CPU is in an RCU read-side critical section, | |
8157 | * so the tick can report quiescent states even for CPUs looping | |
8158 | * in kernel context. In contrast, in non-preemptible kernels, | |
8159 | * RCU readers leave no in-memory hints, which means that CPU-bound | |
8160 | * processes executing in kernel context might never report an | |
8161 | * RCU quiescent state. Therefore, the following code causes | |
8162 | * cond_resched() to report a quiescent state, but only when RCU | |
8163 | * is in urgent need of one. | |
8164 | */ | |
b965f1dd | 8165 | #ifndef CONFIG_PREEMPT_RCU |
f79c3ad6 | 8166 | rcu_all_qs(); |
b965f1dd | 8167 | #endif |
1da177e4 LT |
8168 | return 0; |
8169 | } | |
b965f1dd PZI |
8170 | EXPORT_SYMBOL(__cond_resched); |
8171 | #endif | |
8172 | ||
8173 | #ifdef CONFIG_PREEMPT_DYNAMIC | |
8174 | DEFINE_STATIC_CALL_RET0(cond_resched, __cond_resched); | |
ef72661e | 8175 | EXPORT_STATIC_CALL_TRAMP(cond_resched); |
b965f1dd PZI |
8176 | |
8177 | DEFINE_STATIC_CALL_RET0(might_resched, __cond_resched); | |
ef72661e | 8178 | EXPORT_STATIC_CALL_TRAMP(might_resched); |
35a773a0 | 8179 | #endif |
1da177e4 LT |
8180 | |
8181 | /* | |
613afbf8 | 8182 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
8183 | * call schedule, and on return reacquire the lock. |
8184 | * | |
c1a280b6 | 8185 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
8186 | * operations here to prevent schedule() from being called twice (once via |
8187 | * spin_unlock(), once by hand). | |
8188 | */ | |
613afbf8 | 8189 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 8190 | { |
fe32d3cd | 8191 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
8192 | int ret = 0; |
8193 | ||
f607c668 PZ |
8194 | lockdep_assert_held(lock); |
8195 | ||
4a81e832 | 8196 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 8197 | spin_unlock(lock); |
d86ee480 | 8198 | if (resched) |
a18b5d01 | 8199 | preempt_schedule_common(); |
95c354fe NP |
8200 | else |
8201 | cpu_relax(); | |
6df3cecb | 8202 | ret = 1; |
1da177e4 | 8203 | spin_lock(lock); |
1da177e4 | 8204 | } |
6df3cecb | 8205 | return ret; |
1da177e4 | 8206 | } |
613afbf8 | 8207 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 8208 | |
f3d4b4b1 BG |
8209 | int __cond_resched_rwlock_read(rwlock_t *lock) |
8210 | { | |
8211 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8212 | int ret = 0; | |
8213 | ||
8214 | lockdep_assert_held_read(lock); | |
8215 | ||
8216 | if (rwlock_needbreak(lock) || resched) { | |
8217 | read_unlock(lock); | |
8218 | if (resched) | |
8219 | preempt_schedule_common(); | |
8220 | else | |
8221 | cpu_relax(); | |
8222 | ret = 1; | |
8223 | read_lock(lock); | |
8224 | } | |
8225 | return ret; | |
8226 | } | |
8227 | EXPORT_SYMBOL(__cond_resched_rwlock_read); | |
8228 | ||
8229 | int __cond_resched_rwlock_write(rwlock_t *lock) | |
8230 | { | |
8231 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8232 | int ret = 0; | |
8233 | ||
8234 | lockdep_assert_held_write(lock); | |
8235 | ||
8236 | if (rwlock_needbreak(lock) || resched) { | |
8237 | write_unlock(lock); | |
8238 | if (resched) | |
8239 | preempt_schedule_common(); | |
8240 | else | |
8241 | cpu_relax(); | |
8242 | ret = 1; | |
8243 | write_lock(lock); | |
8244 | } | |
8245 | return ret; | |
8246 | } | |
8247 | EXPORT_SYMBOL(__cond_resched_rwlock_write); | |
8248 | ||
1da177e4 LT |
8249 | /** |
8250 | * yield - yield the current processor to other threads. | |
8251 | * | |
8e3fabfd PZ |
8252 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
8253 | * | |
8254 | * The scheduler is at all times free to pick the calling task as the most | |
8255 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 8256 | * it, it's already broken. |
8e3fabfd PZ |
8257 | * |
8258 | * Typical broken usage is: | |
8259 | * | |
8260 | * while (!event) | |
d1ccc66d | 8261 | * yield(); |
8e3fabfd PZ |
8262 | * |
8263 | * where one assumes that yield() will let 'the other' process run that will | |
8264 | * make event true. If the current task is a SCHED_FIFO task that will never | |
8265 | * happen. Never use yield() as a progress guarantee!! | |
8266 | * | |
8267 | * If you want to use yield() to wait for something, use wait_event(). | |
8268 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
8269 | * If you still want to use yield(), do not! | |
1da177e4 LT |
8270 | */ |
8271 | void __sched yield(void) | |
8272 | { | |
8273 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 8274 | do_sched_yield(); |
1da177e4 | 8275 | } |
1da177e4 LT |
8276 | EXPORT_SYMBOL(yield); |
8277 | ||
d95f4122 MG |
8278 | /** |
8279 | * yield_to - yield the current processor to another thread in | |
8280 | * your thread group, or accelerate that thread toward the | |
8281 | * processor it's on. | |
16addf95 RD |
8282 | * @p: target task |
8283 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
8284 | * |
8285 | * It's the caller's job to ensure that the target task struct | |
8286 | * can't go away on us before we can do any checks. | |
8287 | * | |
e69f6186 | 8288 | * Return: |
7b270f60 PZ |
8289 | * true (>0) if we indeed boosted the target task. |
8290 | * false (0) if we failed to boost the target. | |
8291 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 8292 | */ |
fa93384f | 8293 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
8294 | { |
8295 | struct task_struct *curr = current; | |
8296 | struct rq *rq, *p_rq; | |
8297 | unsigned long flags; | |
c3c18640 | 8298 | int yielded = 0; |
d95f4122 MG |
8299 | |
8300 | local_irq_save(flags); | |
8301 | rq = this_rq(); | |
8302 | ||
8303 | again: | |
8304 | p_rq = task_rq(p); | |
7b270f60 PZ |
8305 | /* |
8306 | * If we're the only runnable task on the rq and target rq also | |
8307 | * has only one task, there's absolutely no point in yielding. | |
8308 | */ | |
8309 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
8310 | yielded = -ESRCH; | |
8311 | goto out_irq; | |
8312 | } | |
8313 | ||
d95f4122 | 8314 | double_rq_lock(rq, p_rq); |
39e24d8f | 8315 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
8316 | double_rq_unlock(rq, p_rq); |
8317 | goto again; | |
8318 | } | |
8319 | ||
8320 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 8321 | goto out_unlock; |
d95f4122 MG |
8322 | |
8323 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 8324 | goto out_unlock; |
d95f4122 | 8325 | |
b03fbd4f | 8326 | if (task_running(p_rq, p) || !task_is_running(p)) |
7b270f60 | 8327 | goto out_unlock; |
d95f4122 | 8328 | |
0900acf2 | 8329 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 8330 | if (yielded) { |
ae92882e | 8331 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
8332 | /* |
8333 | * Make p's CPU reschedule; pick_next_entity takes care of | |
8334 | * fairness. | |
8335 | */ | |
8336 | if (preempt && rq != p_rq) | |
8875125e | 8337 | resched_curr(p_rq); |
6d1cafd8 | 8338 | } |
d95f4122 | 8339 | |
7b270f60 | 8340 | out_unlock: |
d95f4122 | 8341 | double_rq_unlock(rq, p_rq); |
7b270f60 | 8342 | out_irq: |
d95f4122 MG |
8343 | local_irq_restore(flags); |
8344 | ||
7b270f60 | 8345 | if (yielded > 0) |
d95f4122 MG |
8346 | schedule(); |
8347 | ||
8348 | return yielded; | |
8349 | } | |
8350 | EXPORT_SYMBOL_GPL(yield_to); | |
8351 | ||
10ab5643 TH |
8352 | int io_schedule_prepare(void) |
8353 | { | |
8354 | int old_iowait = current->in_iowait; | |
8355 | ||
8356 | current->in_iowait = 1; | |
8357 | blk_schedule_flush_plug(current); | |
8358 | ||
8359 | return old_iowait; | |
8360 | } | |
8361 | ||
8362 | void io_schedule_finish(int token) | |
8363 | { | |
8364 | current->in_iowait = token; | |
8365 | } | |
8366 | ||
1da177e4 | 8367 | /* |
41a2d6cf | 8368 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 8369 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 8370 | */ |
1da177e4 LT |
8371 | long __sched io_schedule_timeout(long timeout) |
8372 | { | |
10ab5643 | 8373 | int token; |
1da177e4 LT |
8374 | long ret; |
8375 | ||
10ab5643 | 8376 | token = io_schedule_prepare(); |
1da177e4 | 8377 | ret = schedule_timeout(timeout); |
10ab5643 | 8378 | io_schedule_finish(token); |
9cff8ade | 8379 | |
1da177e4 LT |
8380 | return ret; |
8381 | } | |
9cff8ade | 8382 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 8383 | |
e3b929b0 | 8384 | void __sched io_schedule(void) |
10ab5643 TH |
8385 | { |
8386 | int token; | |
8387 | ||
8388 | token = io_schedule_prepare(); | |
8389 | schedule(); | |
8390 | io_schedule_finish(token); | |
8391 | } | |
8392 | EXPORT_SYMBOL(io_schedule); | |
8393 | ||
1da177e4 LT |
8394 | /** |
8395 | * sys_sched_get_priority_max - return maximum RT priority. | |
8396 | * @policy: scheduling class. | |
8397 | * | |
e69f6186 YB |
8398 | * Return: On success, this syscall returns the maximum |
8399 | * rt_priority that can be used by a given scheduling class. | |
8400 | * On failure, a negative error code is returned. | |
1da177e4 | 8401 | */ |
5add95d4 | 8402 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
8403 | { |
8404 | int ret = -EINVAL; | |
8405 | ||
8406 | switch (policy) { | |
8407 | case SCHED_FIFO: | |
8408 | case SCHED_RR: | |
ae18ad28 | 8409 | ret = MAX_RT_PRIO-1; |
1da177e4 | 8410 | break; |
aab03e05 | 8411 | case SCHED_DEADLINE: |
1da177e4 | 8412 | case SCHED_NORMAL: |
b0a9499c | 8413 | case SCHED_BATCH: |
dd41f596 | 8414 | case SCHED_IDLE: |
1da177e4 LT |
8415 | ret = 0; |
8416 | break; | |
8417 | } | |
8418 | return ret; | |
8419 | } | |
8420 | ||
8421 | /** | |
8422 | * sys_sched_get_priority_min - return minimum RT priority. | |
8423 | * @policy: scheduling class. | |
8424 | * | |
e69f6186 YB |
8425 | * Return: On success, this syscall returns the minimum |
8426 | * rt_priority that can be used by a given scheduling class. | |
8427 | * On failure, a negative error code is returned. | |
1da177e4 | 8428 | */ |
5add95d4 | 8429 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
8430 | { |
8431 | int ret = -EINVAL; | |
8432 | ||
8433 | switch (policy) { | |
8434 | case SCHED_FIFO: | |
8435 | case SCHED_RR: | |
8436 | ret = 1; | |
8437 | break; | |
aab03e05 | 8438 | case SCHED_DEADLINE: |
1da177e4 | 8439 | case SCHED_NORMAL: |
b0a9499c | 8440 | case SCHED_BATCH: |
dd41f596 | 8441 | case SCHED_IDLE: |
1da177e4 LT |
8442 | ret = 0; |
8443 | } | |
8444 | return ret; | |
8445 | } | |
8446 | ||
abca5fc5 | 8447 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 8448 | { |
36c8b586 | 8449 | struct task_struct *p; |
a4ec24b4 | 8450 | unsigned int time_slice; |
eb580751 | 8451 | struct rq_flags rf; |
dba091b9 | 8452 | struct rq *rq; |
3a5c359a | 8453 | int retval; |
1da177e4 LT |
8454 | |
8455 | if (pid < 0) | |
3a5c359a | 8456 | return -EINVAL; |
1da177e4 LT |
8457 | |
8458 | retval = -ESRCH; | |
1a551ae7 | 8459 | rcu_read_lock(); |
1da177e4 LT |
8460 | p = find_process_by_pid(pid); |
8461 | if (!p) | |
8462 | goto out_unlock; | |
8463 | ||
8464 | retval = security_task_getscheduler(p); | |
8465 | if (retval) | |
8466 | goto out_unlock; | |
8467 | ||
eb580751 | 8468 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
8469 | time_slice = 0; |
8470 | if (p->sched_class->get_rr_interval) | |
8471 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 8472 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 8473 | |
1a551ae7 | 8474 | rcu_read_unlock(); |
abca5fc5 AV |
8475 | jiffies_to_timespec64(time_slice, t); |
8476 | return 0; | |
3a5c359a | 8477 | |
1da177e4 | 8478 | out_unlock: |
1a551ae7 | 8479 | rcu_read_unlock(); |
1da177e4 LT |
8480 | return retval; |
8481 | } | |
8482 | ||
2064a5ab RD |
8483 | /** |
8484 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
8485 | * @pid: pid of the process. | |
8486 | * @interval: userspace pointer to the timeslice value. | |
8487 | * | |
8488 | * this syscall writes the default timeslice value of a given process | |
8489 | * into the user-space timespec buffer. A value of '0' means infinity. | |
8490 | * | |
8491 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
8492 | * an error code. | |
8493 | */ | |
abca5fc5 | 8494 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 8495 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
8496 | { |
8497 | struct timespec64 t; | |
8498 | int retval = sched_rr_get_interval(pid, &t); | |
8499 | ||
8500 | if (retval == 0) | |
8501 | retval = put_timespec64(&t, interval); | |
8502 | ||
8503 | return retval; | |
8504 | } | |
8505 | ||
474b9c77 | 8506 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
8507 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
8508 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
8509 | { |
8510 | struct timespec64 t; | |
8511 | int retval = sched_rr_get_interval(pid, &t); | |
8512 | ||
8513 | if (retval == 0) | |
9afc5eee | 8514 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
8515 | return retval; |
8516 | } | |
8517 | #endif | |
8518 | ||
82a1fcb9 | 8519 | void sched_show_task(struct task_struct *p) |
1da177e4 | 8520 | { |
1da177e4 | 8521 | unsigned long free = 0; |
4e79752c | 8522 | int ppid; |
c930b2c0 | 8523 | |
38200502 TH |
8524 | if (!try_get_task_stack(p)) |
8525 | return; | |
20435d84 | 8526 | |
cc172ff3 | 8527 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 | 8528 | |
b03fbd4f | 8529 | if (task_is_running(p)) |
cc172ff3 | 8530 | pr_cont(" running task "); |
1da177e4 | 8531 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 8532 | free = stack_not_used(p); |
1da177e4 | 8533 | #endif |
a90e984c | 8534 | ppid = 0; |
4e79752c | 8535 | rcu_read_lock(); |
a90e984c ON |
8536 | if (pid_alive(p)) |
8537 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 8538 | rcu_read_unlock(); |
cc172ff3 LZ |
8539 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
8540 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 8541 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 8542 | |
3d1cb205 | 8543 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 8544 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 8545 | show_stack(p, NULL, KERN_INFO); |
38200502 | 8546 | put_task_stack(p); |
1da177e4 | 8547 | } |
0032f4e8 | 8548 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 8549 | |
5d68cc95 PZ |
8550 | static inline bool |
8551 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
8552 | { | |
2f064a59 PZ |
8553 | unsigned int state = READ_ONCE(p->__state); |
8554 | ||
5d68cc95 PZ |
8555 | /* no filter, everything matches */ |
8556 | if (!state_filter) | |
8557 | return true; | |
8558 | ||
8559 | /* filter, but doesn't match */ | |
2f064a59 | 8560 | if (!(state & state_filter)) |
5d68cc95 PZ |
8561 | return false; |
8562 | ||
8563 | /* | |
8564 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
8565 | * TASK_KILLABLE). | |
8566 | */ | |
2f064a59 | 8567 | if (state_filter == TASK_UNINTERRUPTIBLE && state == TASK_IDLE) |
5d68cc95 PZ |
8568 | return false; |
8569 | ||
8570 | return true; | |
8571 | } | |
8572 | ||
8573 | ||
2f064a59 | 8574 | void show_state_filter(unsigned int state_filter) |
1da177e4 | 8575 | { |
36c8b586 | 8576 | struct task_struct *g, *p; |
1da177e4 | 8577 | |
510f5acc | 8578 | rcu_read_lock(); |
5d07f420 | 8579 | for_each_process_thread(g, p) { |
1da177e4 LT |
8580 | /* |
8581 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 8582 | * console might take a lot of time: |
57675cb9 AR |
8583 | * Also, reset softlockup watchdogs on all CPUs, because |
8584 | * another CPU might be blocked waiting for us to process | |
8585 | * an IPI. | |
1da177e4 LT |
8586 | */ |
8587 | touch_nmi_watchdog(); | |
57675cb9 | 8588 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 8589 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 8590 | sched_show_task(p); |
5d07f420 | 8591 | } |
1da177e4 | 8592 | |
dd41f596 | 8593 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
8594 | if (!state_filter) |
8595 | sysrq_sched_debug_show(); | |
dd41f596 | 8596 | #endif |
510f5acc | 8597 | rcu_read_unlock(); |
e59e2ae2 IM |
8598 | /* |
8599 | * Only show locks if all tasks are dumped: | |
8600 | */ | |
93335a21 | 8601 | if (!state_filter) |
e59e2ae2 | 8602 | debug_show_all_locks(); |
1da177e4 LT |
8603 | } |
8604 | ||
f340c0d1 IM |
8605 | /** |
8606 | * init_idle - set up an idle thread for a given CPU | |
8607 | * @idle: task in question | |
d1ccc66d | 8608 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
8609 | * |
8610 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
8611 | * flag, to make booting more robust. | |
8612 | */ | |
f1a0a376 | 8613 | void __init init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 8614 | { |
70b97a7f | 8615 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
8616 | unsigned long flags; |
8617 | ||
ff51ff84 PZ |
8618 | __sched_fork(0, idle); |
8619 | ||
00b89fe0 VS |
8620 | /* |
8621 | * The idle task doesn't need the kthread struct to function, but it | |
8622 | * is dressed up as a per-CPU kthread and thus needs to play the part | |
8623 | * if we want to avoid special-casing it in code that deals with per-CPU | |
8624 | * kthreads. | |
8625 | */ | |
8626 | set_kthread_struct(idle); | |
8627 | ||
25834c73 | 8628 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5cb9eaa3 | 8629 | raw_spin_rq_lock(rq); |
5cbd54ef | 8630 | |
2f064a59 | 8631 | idle->__state = TASK_RUNNING; |
dd41f596 | 8632 | idle->se.exec_start = sched_clock(); |
00b89fe0 VS |
8633 | /* |
8634 | * PF_KTHREAD should already be set at this point; regardless, make it | |
8635 | * look like a proper per-CPU kthread. | |
8636 | */ | |
8637 | idle->flags |= PF_IDLE | PF_KTHREAD | PF_NO_SETAFFINITY; | |
8638 | kthread_set_per_cpu(idle, cpu); | |
dd41f596 | 8639 | |
d08b9f0c | 8640 | scs_task_reset(idle); |
e1b77c92 MR |
8641 | kasan_unpoison_task_stack(idle); |
8642 | ||
de9b8f5d PZ |
8643 | #ifdef CONFIG_SMP |
8644 | /* | |
b19a888c | 8645 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
8646 | * in that case do_set_cpus_allowed() will not do the right thing. |
8647 | * | |
8648 | * And since this is boot we can forgo the serialization. | |
8649 | */ | |
9cfc3e18 | 8650 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 8651 | #endif |
6506cf6c PZ |
8652 | /* |
8653 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 8654 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
8655 | * lockdep check in task_group() will fail. |
8656 | * | |
8657 | * Similar case to sched_fork(). / Alternatively we could | |
8658 | * use task_rq_lock() here and obtain the other rq->lock. | |
8659 | * | |
8660 | * Silence PROVE_RCU | |
8661 | */ | |
8662 | rcu_read_lock(); | |
dd41f596 | 8663 | __set_task_cpu(idle, cpu); |
6506cf6c | 8664 | rcu_read_unlock(); |
1da177e4 | 8665 | |
5311a98f EB |
8666 | rq->idle = idle; |
8667 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 8668 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 8669 | #ifdef CONFIG_SMP |
3ca7a440 | 8670 | idle->on_cpu = 1; |
4866cde0 | 8671 | #endif |
5cb9eaa3 | 8672 | raw_spin_rq_unlock(rq); |
25834c73 | 8673 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); |
1da177e4 LT |
8674 | |
8675 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 8676 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 8677 | |
dd41f596 IM |
8678 | /* |
8679 | * The idle tasks have their own, simple scheduling class: | |
8680 | */ | |
8681 | idle->sched_class = &idle_sched_class; | |
868baf07 | 8682 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 8683 | vtime_init_idle(idle, cpu); |
de9b8f5d | 8684 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
8685 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
8686 | #endif | |
19978ca6 IM |
8687 | } |
8688 | ||
e1d4eeec NP |
8689 | #ifdef CONFIG_SMP |
8690 | ||
f82f8042 JL |
8691 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
8692 | const struct cpumask *trial) | |
8693 | { | |
06a76fe0 | 8694 | int ret = 1; |
f82f8042 | 8695 | |
bb2bc55a MG |
8696 | if (!cpumask_weight(cur)) |
8697 | return ret; | |
8698 | ||
06a76fe0 | 8699 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
8700 | |
8701 | return ret; | |
8702 | } | |
8703 | ||
7f51412a JL |
8704 | int task_can_attach(struct task_struct *p, |
8705 | const struct cpumask *cs_cpus_allowed) | |
8706 | { | |
8707 | int ret = 0; | |
8708 | ||
8709 | /* | |
8710 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 8711 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
8712 | * affinity and isolating such threads by their set of |
8713 | * allowed nodes is unnecessary. Thus, cpusets are not | |
8714 | * applicable for such threads. This prevents checking for | |
8715 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 8716 | * before cpus_mask may be changed. |
7f51412a JL |
8717 | */ |
8718 | if (p->flags & PF_NO_SETAFFINITY) { | |
8719 | ret = -EINVAL; | |
8720 | goto out; | |
8721 | } | |
8722 | ||
7f51412a | 8723 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
8724 | cs_cpus_allowed)) |
8725 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 8726 | |
7f51412a JL |
8727 | out: |
8728 | return ret; | |
8729 | } | |
8730 | ||
f2cb1360 | 8731 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 8732 | |
e6628d5b MG |
8733 | #ifdef CONFIG_NUMA_BALANCING |
8734 | /* Migrate current task p to target_cpu */ | |
8735 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
8736 | { | |
8737 | struct migration_arg arg = { p, target_cpu }; | |
8738 | int curr_cpu = task_cpu(p); | |
8739 | ||
8740 | if (curr_cpu == target_cpu) | |
8741 | return 0; | |
8742 | ||
3bd37062 | 8743 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
8744 | return -EINVAL; |
8745 | ||
8746 | /* TODO: This is not properly updating schedstats */ | |
8747 | ||
286549dc | 8748 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
8749 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
8750 | } | |
0ec8aa00 PZ |
8751 | |
8752 | /* | |
8753 | * Requeue a task on a given node and accurately track the number of NUMA | |
8754 | * tasks on the runqueues | |
8755 | */ | |
8756 | void sched_setnuma(struct task_struct *p, int nid) | |
8757 | { | |
da0c1e65 | 8758 | bool queued, running; |
eb580751 PZ |
8759 | struct rq_flags rf; |
8760 | struct rq *rq; | |
0ec8aa00 | 8761 | |
eb580751 | 8762 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 8763 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
8764 | running = task_current(rq, p); |
8765 | ||
da0c1e65 | 8766 | if (queued) |
1de64443 | 8767 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 8768 | if (running) |
f3cd1c4e | 8769 | put_prev_task(rq, p); |
0ec8aa00 PZ |
8770 | |
8771 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 8772 | |
da0c1e65 | 8773 | if (queued) |
7134b3e9 | 8774 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 8775 | if (running) |
03b7fad1 | 8776 | set_next_task(rq, p); |
eb580751 | 8777 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 8778 | } |
5cc389bc | 8779 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 8780 | |
1da177e4 | 8781 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 8782 | /* |
d1ccc66d | 8783 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 8784 | * offline. |
054b9108 | 8785 | */ |
48c5ccae | 8786 | void idle_task_exit(void) |
1da177e4 | 8787 | { |
48c5ccae | 8788 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 8789 | |
48c5ccae | 8790 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 8791 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 8792 | |
a53efe5f | 8793 | if (mm != &init_mm) { |
252d2a41 | 8794 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
8795 | finish_arch_post_lock_switch(); |
8796 | } | |
bf2c59fc PZ |
8797 | |
8798 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
8799 | } |
8800 | ||
2558aacf | 8801 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 8802 | { |
2558aacf PZ |
8803 | struct task_struct *p = arg; |
8804 | struct rq *rq = this_rq(); | |
8805 | struct rq_flags rf; | |
8806 | int cpu; | |
1da177e4 | 8807 | |
2558aacf PZ |
8808 | raw_spin_lock_irq(&p->pi_lock); |
8809 | rq_lock(rq, &rf); | |
3f1d2a31 | 8810 | |
2558aacf PZ |
8811 | update_rq_clock(rq); |
8812 | ||
8813 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
8814 | cpu = select_fallback_rq(rq->cpu, p); | |
8815 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 8816 | } |
3f1d2a31 | 8817 | |
2558aacf PZ |
8818 | rq_unlock(rq, &rf); |
8819 | raw_spin_unlock_irq(&p->pi_lock); | |
8820 | ||
8821 | put_task_struct(p); | |
8822 | ||
8823 | return 0; | |
10e7071b | 8824 | } |
3f1d2a31 | 8825 | |
2558aacf PZ |
8826 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
8827 | ||
48f24c4d | 8828 | /* |
2558aacf | 8829 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
b5c44773 PZ |
8830 | * |
8831 | * This is enabled below SCHED_AP_ACTIVE; when !cpu_active(), but only | |
8832 | * effective when the hotplug motion is down. | |
1da177e4 | 8833 | */ |
2558aacf | 8834 | static void balance_push(struct rq *rq) |
1da177e4 | 8835 | { |
2558aacf PZ |
8836 | struct task_struct *push_task = rq->curr; |
8837 | ||
5cb9eaa3 | 8838 | lockdep_assert_rq_held(rq); |
b5c44773 | 8839 | |
ae792702 PZ |
8840 | /* |
8841 | * Ensure the thing is persistent until balance_push_set(.on = false); | |
8842 | */ | |
8843 | rq->balance_callback = &balance_push_callback; | |
1da177e4 | 8844 | |
b5c44773 | 8845 | /* |
868ad33b TG |
8846 | * Only active while going offline and when invoked on the outgoing |
8847 | * CPU. | |
b5c44773 | 8848 | */ |
868ad33b | 8849 | if (!cpu_dying(rq->cpu) || rq != this_rq()) |
b5c44773 PZ |
8850 | return; |
8851 | ||
1da177e4 | 8852 | /* |
2558aacf PZ |
8853 | * Both the cpu-hotplug and stop task are in this case and are |
8854 | * required to complete the hotplug process. | |
1da177e4 | 8855 | */ |
00b89fe0 | 8856 | if (kthread_is_per_cpu(push_task) || |
5ba2ffba PZ |
8857 | is_migration_disabled(push_task)) { |
8858 | ||
f2469a1f TG |
8859 | /* |
8860 | * If this is the idle task on the outgoing CPU try to wake | |
8861 | * up the hotplug control thread which might wait for the | |
8862 | * last task to vanish. The rcuwait_active() check is | |
8863 | * accurate here because the waiter is pinned on this CPU | |
8864 | * and can't obviously be running in parallel. | |
3015ef4b TG |
8865 | * |
8866 | * On RT kernels this also has to check whether there are | |
8867 | * pinned and scheduled out tasks on the runqueue. They | |
8868 | * need to leave the migrate disabled section first. | |
f2469a1f | 8869 | */ |
3015ef4b TG |
8870 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
8871 | rcuwait_active(&rq->hotplug_wait)) { | |
5cb9eaa3 | 8872 | raw_spin_rq_unlock(rq); |
f2469a1f | 8873 | rcuwait_wake_up(&rq->hotplug_wait); |
5cb9eaa3 | 8874 | raw_spin_rq_lock(rq); |
f2469a1f | 8875 | } |
2558aacf | 8876 | return; |
f2469a1f | 8877 | } |
48f24c4d | 8878 | |
2558aacf | 8879 | get_task_struct(push_task); |
77bd3970 | 8880 | /* |
2558aacf PZ |
8881 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
8882 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 8883 | */ |
5cb9eaa3 | 8884 | raw_spin_rq_unlock(rq); |
2558aacf PZ |
8885 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, |
8886 | this_cpu_ptr(&push_work)); | |
8887 | /* | |
8888 | * At this point need_resched() is true and we'll take the loop in | |
8889 | * schedule(). The next pick is obviously going to be the stop task | |
5ba2ffba | 8890 | * which kthread_is_per_cpu() and will push this task away. |
2558aacf | 8891 | */ |
5cb9eaa3 | 8892 | raw_spin_rq_lock(rq); |
2558aacf | 8893 | } |
77bd3970 | 8894 | |
2558aacf PZ |
8895 | static void balance_push_set(int cpu, bool on) |
8896 | { | |
8897 | struct rq *rq = cpu_rq(cpu); | |
8898 | struct rq_flags rf; | |
48c5ccae | 8899 | |
2558aacf | 8900 | rq_lock_irqsave(rq, &rf); |
22f667c9 PZ |
8901 | if (on) { |
8902 | WARN_ON_ONCE(rq->balance_callback); | |
ae792702 | 8903 | rq->balance_callback = &balance_push_callback; |
22f667c9 | 8904 | } else if (rq->balance_callback == &balance_push_callback) { |
ae792702 | 8905 | rq->balance_callback = NULL; |
22f667c9 | 8906 | } |
2558aacf PZ |
8907 | rq_unlock_irqrestore(rq, &rf); |
8908 | } | |
e692ab53 | 8909 | |
f2469a1f TG |
8910 | /* |
8911 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
8912 | * inactive. All tasks which are not per CPU kernel threads are either | |
8913 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
8914 | * during wakeup. Wait until the CPU is quiescent. | |
8915 | */ | |
8916 | static void balance_hotplug_wait(void) | |
8917 | { | |
8918 | struct rq *rq = this_rq(); | |
5473e0cc | 8919 | |
3015ef4b TG |
8920 | rcuwait_wait_event(&rq->hotplug_wait, |
8921 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
8922 | TASK_UNINTERRUPTIBLE); |
8923 | } | |
5473e0cc | 8924 | |
2558aacf | 8925 | #else |
dce48a84 | 8926 | |
2558aacf PZ |
8927 | static inline void balance_push(struct rq *rq) |
8928 | { | |
dce48a84 | 8929 | } |
dce48a84 | 8930 | |
2558aacf PZ |
8931 | static inline void balance_push_set(int cpu, bool on) |
8932 | { | |
8933 | } | |
8934 | ||
f2469a1f TG |
8935 | static inline void balance_hotplug_wait(void) |
8936 | { | |
dce48a84 | 8937 | } |
f2469a1f | 8938 | |
1da177e4 LT |
8939 | #endif /* CONFIG_HOTPLUG_CPU */ |
8940 | ||
f2cb1360 | 8941 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
8942 | { |
8943 | if (!rq->online) { | |
8944 | const struct sched_class *class; | |
8945 | ||
c6c4927b | 8946 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
8947 | rq->online = 1; |
8948 | ||
8949 | for_each_class(class) { | |
8950 | if (class->rq_online) | |
8951 | class->rq_online(rq); | |
8952 | } | |
8953 | } | |
8954 | } | |
8955 | ||
f2cb1360 | 8956 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
8957 | { |
8958 | if (rq->online) { | |
8959 | const struct sched_class *class; | |
8960 | ||
8961 | for_each_class(class) { | |
8962 | if (class->rq_offline) | |
8963 | class->rq_offline(rq); | |
8964 | } | |
8965 | ||
c6c4927b | 8966 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
8967 | rq->online = 0; |
8968 | } | |
8969 | } | |
8970 | ||
d1ccc66d IM |
8971 | /* |
8972 | * used to mark begin/end of suspend/resume: | |
8973 | */ | |
8974 | static int num_cpus_frozen; | |
d35be8ba | 8975 | |
1da177e4 | 8976 | /* |
3a101d05 TH |
8977 | * Update cpusets according to cpu_active mask. If cpusets are |
8978 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
8979 | * around partition_sched_domains(). | |
d35be8ba SB |
8980 | * |
8981 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
8982 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 8983 | */ |
40190a78 | 8984 | static void cpuset_cpu_active(void) |
e761b772 | 8985 | { |
40190a78 | 8986 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
8987 | /* |
8988 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
8989 | * resume sequence. As long as this is not the last online | |
8990 | * operation in the resume sequence, just build a single sched | |
8991 | * domain, ignoring cpusets. | |
8992 | */ | |
50e76632 PZ |
8993 | partition_sched_domains(1, NULL, NULL); |
8994 | if (--num_cpus_frozen) | |
135fb3e1 | 8995 | return; |
d35be8ba SB |
8996 | /* |
8997 | * This is the last CPU online operation. So fall through and | |
8998 | * restore the original sched domains by considering the | |
8999 | * cpuset configurations. | |
9000 | */ | |
50e76632 | 9001 | cpuset_force_rebuild(); |
3a101d05 | 9002 | } |
30e03acd | 9003 | cpuset_update_active_cpus(); |
3a101d05 | 9004 | } |
e761b772 | 9005 | |
40190a78 | 9006 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 9007 | { |
40190a78 | 9008 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 9009 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 9010 | return -EBUSY; |
30e03acd | 9011 | cpuset_update_active_cpus(); |
135fb3e1 | 9012 | } else { |
d35be8ba SB |
9013 | num_cpus_frozen++; |
9014 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 9015 | } |
135fb3e1 | 9016 | return 0; |
e761b772 | 9017 | } |
e761b772 | 9018 | |
40190a78 | 9019 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 9020 | { |
7d976699 | 9021 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 9022 | struct rq_flags rf; |
7d976699 | 9023 | |
22f667c9 | 9024 | /* |
b5c44773 PZ |
9025 | * Clear the balance_push callback and prepare to schedule |
9026 | * regular tasks. | |
22f667c9 | 9027 | */ |
2558aacf PZ |
9028 | balance_push_set(cpu, false); |
9029 | ||
ba2591a5 PZ |
9030 | #ifdef CONFIG_SCHED_SMT |
9031 | /* | |
c5511d03 | 9032 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 9033 | */ |
c5511d03 PZI |
9034 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
9035 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 9036 | #endif |
40190a78 | 9037 | set_cpu_active(cpu, true); |
135fb3e1 | 9038 | |
40190a78 | 9039 | if (sched_smp_initialized) { |
135fb3e1 | 9040 | sched_domains_numa_masks_set(cpu); |
40190a78 | 9041 | cpuset_cpu_active(); |
e761b772 | 9042 | } |
7d976699 TG |
9043 | |
9044 | /* | |
9045 | * Put the rq online, if not already. This happens: | |
9046 | * | |
9047 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 9048 | * after all CPUs have been brought up. |
7d976699 TG |
9049 | * |
9050 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
9051 | * domains. | |
9052 | */ | |
8a8c69c3 | 9053 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
9054 | if (rq->rd) { |
9055 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9056 | set_rq_online(rq); | |
9057 | } | |
8a8c69c3 | 9058 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 9059 | |
40190a78 | 9060 | return 0; |
135fb3e1 TG |
9061 | } |
9062 | ||
40190a78 | 9063 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 9064 | { |
120455c5 PZ |
9065 | struct rq *rq = cpu_rq(cpu); |
9066 | struct rq_flags rf; | |
135fb3e1 TG |
9067 | int ret; |
9068 | ||
e0b257c3 AMB |
9069 | /* |
9070 | * Remove CPU from nohz.idle_cpus_mask to prevent participating in | |
9071 | * load balancing when not active | |
9072 | */ | |
9073 | nohz_balance_exit_idle(rq); | |
9074 | ||
40190a78 | 9075 | set_cpu_active(cpu, false); |
741ba80f PZ |
9076 | |
9077 | /* | |
9078 | * From this point forward, this CPU will refuse to run any task that | |
9079 | * is not: migrate_disable() or KTHREAD_IS_PER_CPU, and will actively | |
9080 | * push those tasks away until this gets cleared, see | |
9081 | * sched_cpu_dying(). | |
9082 | */ | |
975707f2 PZ |
9083 | balance_push_set(cpu, true); |
9084 | ||
b2454caa | 9085 | /* |
975707f2 PZ |
9086 | * We've cleared cpu_active_mask / set balance_push, wait for all |
9087 | * preempt-disabled and RCU users of this state to go away such that | |
9088 | * all new such users will observe it. | |
b2454caa | 9089 | * |
5ba2ffba PZ |
9090 | * Specifically, we rely on ttwu to no longer target this CPU, see |
9091 | * ttwu_queue_cond() and is_cpu_allowed(). | |
9092 | * | |
b2454caa PZ |
9093 | * Do sync before park smpboot threads to take care the rcu boost case. |
9094 | */ | |
309ba859 | 9095 | synchronize_rcu(); |
40190a78 | 9096 | |
120455c5 PZ |
9097 | rq_lock_irqsave(rq, &rf); |
9098 | if (rq->rd) { | |
9099 | update_rq_clock(rq); | |
9100 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9101 | set_rq_offline(rq); | |
9102 | } | |
9103 | rq_unlock_irqrestore(rq, &rf); | |
9104 | ||
c5511d03 PZI |
9105 | #ifdef CONFIG_SCHED_SMT |
9106 | /* | |
9107 | * When going down, decrement the number of cores with SMT present. | |
9108 | */ | |
9109 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
9110 | static_branch_dec_cpuslocked(&sched_smt_present); | |
3c474b32 PZ |
9111 | |
9112 | sched_core_cpu_deactivate(cpu); | |
c5511d03 PZI |
9113 | #endif |
9114 | ||
40190a78 TG |
9115 | if (!sched_smp_initialized) |
9116 | return 0; | |
9117 | ||
9118 | ret = cpuset_cpu_inactive(cpu); | |
9119 | if (ret) { | |
2558aacf | 9120 | balance_push_set(cpu, false); |
40190a78 TG |
9121 | set_cpu_active(cpu, true); |
9122 | return ret; | |
135fb3e1 | 9123 | } |
40190a78 TG |
9124 | sched_domains_numa_masks_clear(cpu); |
9125 | return 0; | |
135fb3e1 TG |
9126 | } |
9127 | ||
94baf7a5 TG |
9128 | static void sched_rq_cpu_starting(unsigned int cpu) |
9129 | { | |
9130 | struct rq *rq = cpu_rq(cpu); | |
9131 | ||
9132 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
9133 | update_max_interval(); |
9134 | } | |
9135 | ||
135fb3e1 TG |
9136 | int sched_cpu_starting(unsigned int cpu) |
9137 | { | |
9edeaea1 | 9138 | sched_core_cpu_starting(cpu); |
94baf7a5 | 9139 | sched_rq_cpu_starting(cpu); |
d84b3131 | 9140 | sched_tick_start(cpu); |
135fb3e1 | 9141 | return 0; |
e761b772 | 9142 | } |
e761b772 | 9143 | |
f2785ddb | 9144 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
9145 | |
9146 | /* | |
9147 | * Invoked immediately before the stopper thread is invoked to bring the | |
9148 | * CPU down completely. At this point all per CPU kthreads except the | |
9149 | * hotplug thread (current) and the stopper thread (inactive) have been | |
9150 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
9151 | * any of those which might be on the way out are gone. | |
9152 | * | |
9153 | * If after this point a bound task is being woken on this CPU then the | |
9154 | * responsible hotplug callback has failed to do it's job. | |
9155 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
9156 | */ | |
9157 | int sched_cpu_wait_empty(unsigned int cpu) | |
9158 | { | |
9159 | balance_hotplug_wait(); | |
9160 | return 0; | |
9161 | } | |
9162 | ||
9163 | /* | |
9164 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
9165 | * might have. Called from the CPU stopper task after ensuring that the | |
9166 | * stopper is the last running task on the CPU, so nr_active count is | |
9167 | * stable. We need to take the teardown thread which is calling this into | |
9168 | * account, so we hand in adjust = 1 to the load calculation. | |
9169 | * | |
9170 | * Also see the comment "Global load-average calculations". | |
9171 | */ | |
9172 | static void calc_load_migrate(struct rq *rq) | |
9173 | { | |
9174 | long delta = calc_load_fold_active(rq, 1); | |
9175 | ||
9176 | if (delta) | |
9177 | atomic_long_add(delta, &calc_load_tasks); | |
9178 | } | |
9179 | ||
36c6e17b VS |
9180 | static void dump_rq_tasks(struct rq *rq, const char *loglvl) |
9181 | { | |
9182 | struct task_struct *g, *p; | |
9183 | int cpu = cpu_of(rq); | |
9184 | ||
5cb9eaa3 | 9185 | lockdep_assert_rq_held(rq); |
36c6e17b VS |
9186 | |
9187 | printk("%sCPU%d enqueued tasks (%u total):\n", loglvl, cpu, rq->nr_running); | |
9188 | for_each_process_thread(g, p) { | |
9189 | if (task_cpu(p) != cpu) | |
9190 | continue; | |
9191 | ||
9192 | if (!task_on_rq_queued(p)) | |
9193 | continue; | |
9194 | ||
9195 | printk("%s\tpid: %d, name: %s\n", loglvl, p->pid, p->comm); | |
9196 | } | |
9197 | } | |
9198 | ||
f2785ddb TG |
9199 | int sched_cpu_dying(unsigned int cpu) |
9200 | { | |
9201 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 9202 | struct rq_flags rf; |
f2785ddb TG |
9203 | |
9204 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 9205 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
9206 | |
9207 | rq_lock_irqsave(rq, &rf); | |
36c6e17b VS |
9208 | if (rq->nr_running != 1 || rq_has_pinned_tasks(rq)) { |
9209 | WARN(true, "Dying CPU not properly vacated!"); | |
9210 | dump_rq_tasks(rq, KERN_WARNING); | |
9211 | } | |
8a8c69c3 PZ |
9212 | rq_unlock_irqrestore(rq, &rf); |
9213 | ||
f2785ddb TG |
9214 | calc_load_migrate(rq); |
9215 | update_max_interval(); | |
e5ef27d0 | 9216 | hrtick_clear(rq); |
3c474b32 | 9217 | sched_core_cpu_dying(cpu); |
f2785ddb TG |
9218 | return 0; |
9219 | } | |
9220 | #endif | |
9221 | ||
1da177e4 LT |
9222 | void __init sched_init_smp(void) |
9223 | { | |
cb83b629 PZ |
9224 | sched_init_numa(); |
9225 | ||
6acce3ef PZ |
9226 | /* |
9227 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 9228 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 9229 | * happen. |
6acce3ef | 9230 | */ |
712555ee | 9231 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 9232 | sched_init_domains(cpu_active_mask); |
712555ee | 9233 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 9234 | |
5c1e1767 | 9235 | /* Move init over to a non-isolated CPU */ |
edb93821 | 9236 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 9237 | BUG(); |
15faafc6 | 9238 | current->flags &= ~PF_NO_SETAFFINITY; |
19978ca6 | 9239 | sched_init_granularity(); |
4212823f | 9240 | |
0e3900e6 | 9241 | init_sched_rt_class(); |
1baca4ce | 9242 | init_sched_dl_class(); |
1b568f0a | 9243 | |
e26fbffd | 9244 | sched_smp_initialized = true; |
1da177e4 | 9245 | } |
e26fbffd TG |
9246 | |
9247 | static int __init migration_init(void) | |
9248 | { | |
77a5352b | 9249 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 9250 | return 0; |
1da177e4 | 9251 | } |
e26fbffd TG |
9252 | early_initcall(migration_init); |
9253 | ||
1da177e4 LT |
9254 | #else |
9255 | void __init sched_init_smp(void) | |
9256 | { | |
19978ca6 | 9257 | sched_init_granularity(); |
1da177e4 LT |
9258 | } |
9259 | #endif /* CONFIG_SMP */ | |
9260 | ||
9261 | int in_sched_functions(unsigned long addr) | |
9262 | { | |
1da177e4 LT |
9263 | return in_lock_functions(addr) || |
9264 | (addr >= (unsigned long)__sched_text_start | |
9265 | && addr < (unsigned long)__sched_text_end); | |
9266 | } | |
9267 | ||
029632fb | 9268 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
9269 | /* |
9270 | * Default task group. | |
9271 | * Every task in system belongs to this group at bootup. | |
9272 | */ | |
029632fb | 9273 | struct task_group root_task_group; |
35cf4e50 | 9274 | LIST_HEAD(task_groups); |
b0367629 WL |
9275 | |
9276 | /* Cacheline aligned slab cache for task_group */ | |
9277 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 9278 | #endif |
6f505b16 | 9279 | |
e6252c3e | 9280 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 9281 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 9282 | |
1da177e4 LT |
9283 | void __init sched_init(void) |
9284 | { | |
a1dc0446 | 9285 | unsigned long ptr = 0; |
55627e3c | 9286 | int i; |
434d53b0 | 9287 | |
c3a340f7 SRV |
9288 | /* Make sure the linker didn't screw up */ |
9289 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
9290 | &fair_sched_class + 1 != &rt_sched_class || | |
9291 | &rt_sched_class + 1 != &dl_sched_class); | |
9292 | #ifdef CONFIG_SMP | |
9293 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
9294 | #endif | |
9295 | ||
5822a454 | 9296 | wait_bit_init(); |
9dcb8b68 | 9297 | |
434d53b0 | 9298 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 9299 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
9300 | #endif |
9301 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 9302 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 9303 | #endif |
a1dc0446 QC |
9304 | if (ptr) { |
9305 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
9306 | |
9307 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 9308 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
9309 | ptr += nr_cpu_ids * sizeof(void **); |
9310 | ||
07e06b01 | 9311 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 9312 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 9313 | |
b1d1779e WY |
9314 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
9315 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 9316 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 9317 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9318 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
9319 | ptr += nr_cpu_ids * sizeof(void **); |
9320 | ||
07e06b01 | 9321 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
9322 | ptr += nr_cpu_ids * sizeof(void **); |
9323 | ||
6d6bc0ad | 9324 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 9325 | } |
df7c8e84 | 9326 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
9327 | for_each_possible_cpu(i) { |
9328 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
9329 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
9330 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
9331 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 9332 | } |
b74e6278 | 9333 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 9334 | |
d1ccc66d IM |
9335 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
9336 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 9337 | |
57d885fe GH |
9338 | #ifdef CONFIG_SMP |
9339 | init_defrootdomain(); | |
9340 | #endif | |
9341 | ||
d0b27fa7 | 9342 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9343 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 9344 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 9345 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 9346 | |
7c941438 | 9347 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
9348 | task_group_cache = KMEM_CACHE(task_group, 0); |
9349 | ||
07e06b01 YZ |
9350 | list_add(&root_task_group.list, &task_groups); |
9351 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 9352 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 9353 | autogroup_init(&init_task); |
7c941438 | 9354 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 9355 | |
0a945022 | 9356 | for_each_possible_cpu(i) { |
70b97a7f | 9357 | struct rq *rq; |
1da177e4 LT |
9358 | |
9359 | rq = cpu_rq(i); | |
5cb9eaa3 | 9360 | raw_spin_lock_init(&rq->__lock); |
7897986b | 9361 | rq->nr_running = 0; |
dce48a84 TG |
9362 | rq->calc_load_active = 0; |
9363 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 9364 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
9365 | init_rt_rq(&rq->rt); |
9366 | init_dl_rq(&rq->dl); | |
dd41f596 | 9367 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 9368 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 9369 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 9370 | /* |
d1ccc66d | 9371 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
9372 | * |
9373 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
9374 | * gets 100% of the CPU resources in the system. This overall |
9375 | * system CPU resource is divided among the tasks of | |
07e06b01 | 9376 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
9377 | * based on each entity's (task or task-group's) weight |
9378 | * (se->load.weight). | |
9379 | * | |
07e06b01 | 9380 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 9381 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 9382 | * then A0's share of the CPU resource is: |
354d60c2 | 9383 | * |
0d905bca | 9384 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 9385 | * |
07e06b01 YZ |
9386 | * We achieve this by letting root_task_group's tasks sit |
9387 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 9388 | */ |
07e06b01 | 9389 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
9390 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9391 | ||
9392 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9393 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9394 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 9395 | #endif |
1da177e4 | 9396 | #ifdef CONFIG_SMP |
41c7ce9a | 9397 | rq->sd = NULL; |
57d885fe | 9398 | rq->rd = NULL; |
ca6d75e6 | 9399 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
b5c44773 | 9400 | rq->balance_callback = &balance_push_callback; |
1da177e4 | 9401 | rq->active_balance = 0; |
dd41f596 | 9402 | rq->next_balance = jiffies; |
1da177e4 | 9403 | rq->push_cpu = 0; |
0a2966b4 | 9404 | rq->cpu = i; |
1f11eb6a | 9405 | rq->online = 0; |
eae0c9df MG |
9406 | rq->idle_stamp = 0; |
9407 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
94aafc3e PZ |
9408 | rq->wake_stamp = jiffies; |
9409 | rq->wake_avg_idle = rq->avg_idle; | |
9bd721c5 | 9410 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
9411 | |
9412 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
9413 | ||
dc938520 | 9414 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 9415 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 9416 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 9417 | atomic_set(&rq->nohz_flags, 0); |
90b5363a | 9418 | |
545b8c8d | 9419 | INIT_CSD(&rq->nohz_csd, nohz_csd_func, rq); |
83cd4fe2 | 9420 | #endif |
f2469a1f TG |
9421 | #ifdef CONFIG_HOTPLUG_CPU |
9422 | rcuwait_init(&rq->hotplug_wait); | |
83cd4fe2 | 9423 | #endif |
9fd81dd5 | 9424 | #endif /* CONFIG_SMP */ |
77a021be | 9425 | hrtick_rq_init(rq); |
1da177e4 | 9426 | atomic_set(&rq->nr_iowait, 0); |
9edeaea1 PZ |
9427 | |
9428 | #ifdef CONFIG_SCHED_CORE | |
3c474b32 | 9429 | rq->core = rq; |
539f6512 | 9430 | rq->core_pick = NULL; |
9edeaea1 | 9431 | rq->core_enabled = 0; |
539f6512 PZ |
9432 | rq->core_tree = RB_ROOT; |
9433 | rq->core_forceidle = false; | |
9434 | ||
9435 | rq->core_cookie = 0UL; | |
9edeaea1 | 9436 | #endif |
1da177e4 LT |
9437 | } |
9438 | ||
9059393e | 9439 | set_load_weight(&init_task, false); |
b50f60ce | 9440 | |
1da177e4 LT |
9441 | /* |
9442 | * The boot idle thread does lazy MMU switching as well: | |
9443 | */ | |
f1f10076 | 9444 | mmgrab(&init_mm); |
1da177e4 LT |
9445 | enter_lazy_tlb(&init_mm, current); |
9446 | ||
9447 | /* | |
9448 | * Make us the idle thread. Technically, schedule() should not be | |
9449 | * called from this thread, however somewhere below it might be, | |
9450 | * but because we are the idle thread, we just pick up running again | |
9451 | * when this runqueue becomes "idle". | |
9452 | */ | |
9453 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9454 | |
9455 | calc_load_update = jiffies + LOAD_FREQ; | |
9456 | ||
bf4d83f6 | 9457 | #ifdef CONFIG_SMP |
29d5e047 | 9458 | idle_thread_set_boot_cpu(); |
b5c44773 | 9459 | balance_push_set(smp_processor_id(), false); |
029632fb PZ |
9460 | #endif |
9461 | init_sched_fair_class(); | |
6a7b3dc3 | 9462 | |
eb414681 JW |
9463 | psi_init(); |
9464 | ||
69842cba PB |
9465 | init_uclamp(); |
9466 | ||
6892b75e | 9467 | scheduler_running = 1; |
1da177e4 LT |
9468 | } |
9469 | ||
d902db1e | 9470 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 FW |
9471 | static inline int preempt_count_equals(int preempt_offset) |
9472 | { | |
da7142e2 | 9473 | int nested = preempt_count() + rcu_preempt_depth(); |
e4aafea2 | 9474 | |
4ba8216c | 9475 | return (nested == preempt_offset); |
e4aafea2 FW |
9476 | } |
9477 | ||
d894837f | 9478 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 9479 | { |
d6c23bb3 | 9480 | unsigned int state = get_current_state(); |
8eb23b9f PZ |
9481 | /* |
9482 | * Blocking primitives will set (and therefore destroy) current->state, | |
9483 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
9484 | * otherwise we will destroy state. | |
9485 | */ | |
d6c23bb3 | 9486 | WARN_ONCE(state != TASK_RUNNING && current->task_state_change, |
8eb23b9f | 9487 | "do not call blocking ops when !TASK_RUNNING; " |
d6c23bb3 | 9488 | "state=%x set at [<%p>] %pS\n", state, |
8eb23b9f | 9489 | (void *)current->task_state_change, |
00845eb9 | 9490 | (void *)current->task_state_change); |
8eb23b9f | 9491 | |
3427445a PZ |
9492 | ___might_sleep(file, line, preempt_offset); |
9493 | } | |
9494 | EXPORT_SYMBOL(__might_sleep); | |
9495 | ||
9496 | void ___might_sleep(const char *file, int line, int preempt_offset) | |
1da177e4 | 9497 | { |
d1ccc66d IM |
9498 | /* Ratelimiting timestamp: */ |
9499 | static unsigned long prev_jiffy; | |
9500 | ||
d1c6d149 | 9501 | unsigned long preempt_disable_ip; |
1da177e4 | 9502 | |
d1ccc66d IM |
9503 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
9504 | rcu_sleep_check(); | |
9505 | ||
db273be2 | 9506 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
312364f3 | 9507 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
9508 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
9509 | oops_in_progress) | |
aef745fc | 9510 | return; |
1c3c5eab | 9511 | |
aef745fc IM |
9512 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
9513 | return; | |
9514 | prev_jiffy = jiffies; | |
9515 | ||
d1ccc66d | 9516 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
9517 | preempt_disable_ip = get_preempt_disable_ip(current); |
9518 | ||
3df0fc5b PZ |
9519 | printk(KERN_ERR |
9520 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9521 | file, line); | |
9522 | printk(KERN_ERR | |
312364f3 DV |
9523 | "in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", |
9524 | in_atomic(), irqs_disabled(), current->non_block_count, | |
3df0fc5b | 9525 | current->pid, current->comm); |
aef745fc | 9526 | |
a8b686b3 ES |
9527 | if (task_stack_end_corrupted(current)) |
9528 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); | |
9529 | ||
aef745fc IM |
9530 | debug_show_held_locks(current); |
9531 | if (irqs_disabled()) | |
9532 | print_irqtrace_events(current); | |
d1c6d149 VN |
9533 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
9534 | && !preempt_count_equals(preempt_offset)) { | |
8f47b187 | 9535 | pr_err("Preemption disabled at:"); |
2062a4e8 | 9536 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 9537 | } |
aef745fc | 9538 | dump_stack(); |
f0b22e39 | 9539 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 9540 | } |
3427445a | 9541 | EXPORT_SYMBOL(___might_sleep); |
568f1967 PZ |
9542 | |
9543 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
9544 | { | |
9545 | static unsigned long prev_jiffy; | |
9546 | ||
9547 | if (irqs_disabled()) | |
9548 | return; | |
9549 | ||
9550 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
9551 | return; | |
9552 | ||
9553 | if (preempt_count() > preempt_offset) | |
9554 | return; | |
9555 | ||
9556 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9557 | return; | |
9558 | prev_jiffy = jiffies; | |
9559 | ||
9560 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
9561 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9562 | in_atomic(), irqs_disabled(), | |
9563 | current->pid, current->comm); | |
9564 | ||
9565 | debug_show_held_locks(current); | |
9566 | dump_stack(); | |
9567 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
9568 | } | |
9569 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
9570 | |
9571 | #ifdef CONFIG_SMP | |
9572 | void __cant_migrate(const char *file, int line) | |
9573 | { | |
9574 | static unsigned long prev_jiffy; | |
9575 | ||
9576 | if (irqs_disabled()) | |
9577 | return; | |
9578 | ||
9579 | if (is_migration_disabled(current)) | |
9580 | return; | |
9581 | ||
9582 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
9583 | return; | |
9584 | ||
9585 | if (preempt_count() > 0) | |
9586 | return; | |
9587 | ||
9588 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9589 | return; | |
9590 | prev_jiffy = jiffies; | |
9591 | ||
9592 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
9593 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
9594 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
9595 | current->pid, current->comm); | |
9596 | ||
9597 | debug_show_held_locks(current); | |
9598 | dump_stack(); | |
9599 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
9600 | } | |
9601 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
9602 | #endif | |
1da177e4 LT |
9603 | #endif |
9604 | ||
9605 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 9606 | void normalize_rt_tasks(void) |
3a5e4dc1 | 9607 | { |
dbc7f069 | 9608 | struct task_struct *g, *p; |
d50dde5a DF |
9609 | struct sched_attr attr = { |
9610 | .sched_policy = SCHED_NORMAL, | |
9611 | }; | |
1da177e4 | 9612 | |
3472eaa1 | 9613 | read_lock(&tasklist_lock); |
5d07f420 | 9614 | for_each_process_thread(g, p) { |
178be793 IM |
9615 | /* |
9616 | * Only normalize user tasks: | |
9617 | */ | |
3472eaa1 | 9618 | if (p->flags & PF_KTHREAD) |
178be793 IM |
9619 | continue; |
9620 | ||
4fa8d299 JP |
9621 | p->se.exec_start = 0; |
9622 | schedstat_set(p->se.statistics.wait_start, 0); | |
9623 | schedstat_set(p->se.statistics.sleep_start, 0); | |
9624 | schedstat_set(p->se.statistics.block_start, 0); | |
dd41f596 | 9625 | |
aab03e05 | 9626 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
9627 | /* |
9628 | * Renice negative nice level userspace | |
9629 | * tasks back to 0: | |
9630 | */ | |
3472eaa1 | 9631 | if (task_nice(p) < 0) |
dd41f596 | 9632 | set_user_nice(p, 0); |
1da177e4 | 9633 | continue; |
dd41f596 | 9634 | } |
1da177e4 | 9635 | |
dbc7f069 | 9636 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 9637 | } |
3472eaa1 | 9638 | read_unlock(&tasklist_lock); |
1da177e4 LT |
9639 | } |
9640 | ||
9641 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 9642 | |
67fc4e0c | 9643 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 9644 | /* |
67fc4e0c | 9645 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
9646 | * |
9647 | * They can only be called when the whole system has been | |
9648 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9649 | * activity can take place. Using them for anything else would | |
9650 | * be a serious bug, and as a result, they aren't even visible | |
9651 | * under any other configuration. | |
9652 | */ | |
9653 | ||
9654 | /** | |
d1ccc66d | 9655 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
9656 | * @cpu: the processor in question. |
9657 | * | |
9658 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
9659 | * |
9660 | * Return: The current task for @cpu. | |
1df5c10a | 9661 | */ |
36c8b586 | 9662 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9663 | { |
9664 | return cpu_curr(cpu); | |
9665 | } | |
9666 | ||
67fc4e0c JW |
9667 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
9668 | ||
9669 | #ifdef CONFIG_IA64 | |
1df5c10a | 9670 | /** |
5feeb783 | 9671 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
9672 | * @cpu: the processor in question. |
9673 | * @p: the task pointer to set. | |
9674 | * | |
9675 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 9676 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 9677 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
9678 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9679 | * and caller must save the original value of the current task (see | |
9680 | * curr_task() above) and restore that value before reenabling interrupts and | |
9681 | * re-starting the system. | |
9682 | * | |
9683 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9684 | */ | |
a458ae2e | 9685 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9686 | { |
9687 | cpu_curr(cpu) = p; | |
9688 | } | |
9689 | ||
9690 | #endif | |
29f59db3 | 9691 | |
7c941438 | 9692 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
9693 | /* task_group_lock serializes the addition/removal of task groups */ |
9694 | static DEFINE_SPINLOCK(task_group_lock); | |
9695 | ||
2480c093 PB |
9696 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
9697 | struct task_group *parent) | |
9698 | { | |
9699 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 9700 | enum uclamp_id clamp_id; |
2480c093 PB |
9701 | |
9702 | for_each_clamp_id(clamp_id) { | |
9703 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
9704 | uclamp_none(clamp_id), false); | |
0b60ba2d | 9705 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
9706 | } |
9707 | #endif | |
9708 | } | |
9709 | ||
2f5177f0 | 9710 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
9711 | { |
9712 | free_fair_sched_group(tg); | |
9713 | free_rt_sched_group(tg); | |
e9aa1dd1 | 9714 | autogroup_free(tg); |
b0367629 | 9715 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
9716 | } |
9717 | ||
9718 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9719 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9720 | { |
9721 | struct task_group *tg; | |
bccbe08a | 9722 | |
b0367629 | 9723 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
9724 | if (!tg) |
9725 | return ERR_PTR(-ENOMEM); | |
9726 | ||
ec7dc8ac | 9727 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9728 | goto err; |
9729 | ||
ec7dc8ac | 9730 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9731 | goto err; |
9732 | ||
2480c093 PB |
9733 | alloc_uclamp_sched_group(tg, parent); |
9734 | ||
ace783b9 LZ |
9735 | return tg; |
9736 | ||
9737 | err: | |
2f5177f0 | 9738 | sched_free_group(tg); |
ace783b9 LZ |
9739 | return ERR_PTR(-ENOMEM); |
9740 | } | |
9741 | ||
9742 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
9743 | { | |
9744 | unsigned long flags; | |
9745 | ||
8ed36996 | 9746 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 9747 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 9748 | |
d1ccc66d IM |
9749 | /* Root should already exist: */ |
9750 | WARN_ON(!parent); | |
f473aa5e PZ |
9751 | |
9752 | tg->parent = parent; | |
f473aa5e | 9753 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9754 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9755 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
9756 | |
9757 | online_fair_sched_group(tg); | |
29f59db3 SV |
9758 | } |
9759 | ||
9b5b7751 | 9760 | /* rcu callback to free various structures associated with a task group */ |
2f5177f0 | 9761 | static void sched_free_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9762 | { |
d1ccc66d | 9763 | /* Now it should be safe to free those cfs_rqs: */ |
2f5177f0 | 9764 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9765 | } |
9766 | ||
4cf86d77 | 9767 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 9768 | { |
d1ccc66d | 9769 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
2f5177f0 | 9770 | call_rcu(&tg->rcu, sched_free_group_rcu); |
ace783b9 LZ |
9771 | } |
9772 | ||
9773 | void sched_offline_group(struct task_group *tg) | |
29f59db3 | 9774 | { |
8ed36996 | 9775 | unsigned long flags; |
29f59db3 | 9776 | |
d1ccc66d | 9777 | /* End participation in shares distribution: */ |
6fe1f348 | 9778 | unregister_fair_sched_group(tg); |
3d4b47b4 PZ |
9779 | |
9780 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 9781 | list_del_rcu(&tg->list); |
f473aa5e | 9782 | list_del_rcu(&tg->siblings); |
8ed36996 | 9783 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
9784 | } |
9785 | ||
ea86cb4b | 9786 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 9787 | { |
8323f26c | 9788 | struct task_group *tg; |
29f59db3 | 9789 | |
f7b8a47d KT |
9790 | /* |
9791 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
9792 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
9793 | * to prevent lockdep warnings. | |
9794 | */ | |
9795 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
9796 | struct task_group, css); |
9797 | tg = autogroup_task_group(tsk, tg); | |
9798 | tsk->sched_task_group = tg; | |
9799 | ||
810b3817 | 9800 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
9801 | if (tsk->sched_class->task_change_group) |
9802 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 9803 | else |
810b3817 | 9804 | #endif |
b2b5ce02 | 9805 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
9806 | } |
9807 | ||
9808 | /* | |
9809 | * Change task's runqueue when it moves between groups. | |
9810 | * | |
9811 | * The caller of this function should have put the task in its new group by | |
9812 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
9813 | * its new group. | |
9814 | */ | |
9815 | void sched_move_task(struct task_struct *tsk) | |
9816 | { | |
7a57f32a PZ |
9817 | int queued, running, queue_flags = |
9818 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
9819 | struct rq_flags rf; |
9820 | struct rq *rq; | |
9821 | ||
9822 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 9823 | update_rq_clock(rq); |
ea86cb4b VG |
9824 | |
9825 | running = task_current(rq, tsk); | |
9826 | queued = task_on_rq_queued(tsk); | |
9827 | ||
9828 | if (queued) | |
7a57f32a | 9829 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 9830 | if (running) |
ea86cb4b VG |
9831 | put_prev_task(rq, tsk); |
9832 | ||
9833 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 9834 | |
da0c1e65 | 9835 | if (queued) |
7a57f32a | 9836 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 9837 | if (running) { |
03b7fad1 | 9838 | set_next_task(rq, tsk); |
2a4b03ff VG |
9839 | /* |
9840 | * After changing group, the running task may have joined a | |
9841 | * throttled one but it's still the running task. Trigger a | |
9842 | * resched to make sure that task can still run. | |
9843 | */ | |
9844 | resched_curr(rq); | |
9845 | } | |
29f59db3 | 9846 | |
eb580751 | 9847 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 9848 | } |
68318b8e | 9849 | |
a7c6d554 | 9850 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 9851 | { |
a7c6d554 | 9852 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
9853 | } |
9854 | ||
eb95419b TH |
9855 | static struct cgroup_subsys_state * |
9856 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 9857 | { |
eb95419b TH |
9858 | struct task_group *parent = css_tg(parent_css); |
9859 | struct task_group *tg; | |
68318b8e | 9860 | |
eb95419b | 9861 | if (!parent) { |
68318b8e | 9862 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9863 | return &root_task_group.css; |
68318b8e SV |
9864 | } |
9865 | ||
ec7dc8ac | 9866 | tg = sched_create_group(parent); |
68318b8e SV |
9867 | if (IS_ERR(tg)) |
9868 | return ERR_PTR(-ENOMEM); | |
9869 | ||
68318b8e SV |
9870 | return &tg->css; |
9871 | } | |
9872 | ||
96b77745 KK |
9873 | /* Expose task group only after completing cgroup initialization */ |
9874 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
9875 | { | |
9876 | struct task_group *tg = css_tg(css); | |
9877 | struct task_group *parent = css_tg(css->parent); | |
9878 | ||
9879 | if (parent) | |
9880 | sched_online_group(tg, parent); | |
7226017a QY |
9881 | |
9882 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9883 | /* Propagate the effective uclamp value for the new group */ | |
93b73858 QY |
9884 | mutex_lock(&uclamp_mutex); |
9885 | rcu_read_lock(); | |
7226017a | 9886 | cpu_util_update_eff(css); |
93b73858 QY |
9887 | rcu_read_unlock(); |
9888 | mutex_unlock(&uclamp_mutex); | |
7226017a QY |
9889 | #endif |
9890 | ||
96b77745 KK |
9891 | return 0; |
9892 | } | |
9893 | ||
2f5177f0 | 9894 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 9895 | { |
eb95419b | 9896 | struct task_group *tg = css_tg(css); |
ace783b9 | 9897 | |
2f5177f0 | 9898 | sched_offline_group(tg); |
ace783b9 LZ |
9899 | } |
9900 | ||
eb95419b | 9901 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 9902 | { |
eb95419b | 9903 | struct task_group *tg = css_tg(css); |
68318b8e | 9904 | |
2f5177f0 PZ |
9905 | /* |
9906 | * Relies on the RCU grace period between css_released() and this. | |
9907 | */ | |
9908 | sched_free_group(tg); | |
ace783b9 LZ |
9909 | } |
9910 | ||
ea86cb4b VG |
9911 | /* |
9912 | * This is called before wake_up_new_task(), therefore we really only | |
9913 | * have to set its group bits, all the other stuff does not apply. | |
9914 | */ | |
b53202e6 | 9915 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 9916 | { |
ea86cb4b VG |
9917 | struct rq_flags rf; |
9918 | struct rq *rq; | |
9919 | ||
9920 | rq = task_rq_lock(task, &rf); | |
9921 | ||
80f5c1b8 | 9922 | update_rq_clock(rq); |
ea86cb4b VG |
9923 | sched_change_group(task, TASK_SET_GROUP); |
9924 | ||
9925 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
9926 | } |
9927 | ||
1f7dd3e5 | 9928 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 9929 | { |
bb9d97b6 | 9930 | struct task_struct *task; |
1f7dd3e5 | 9931 | struct cgroup_subsys_state *css; |
7dc603c9 | 9932 | int ret = 0; |
bb9d97b6 | 9933 | |
1f7dd3e5 | 9934 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 9935 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 9936 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 9937 | return -EINVAL; |
b68aa230 | 9938 | #endif |
7dc603c9 | 9939 | /* |
b19a888c | 9940 | * Serialize against wake_up_new_task() such that if it's |
7dc603c9 PZ |
9941 | * running, we're sure to observe its full state. |
9942 | */ | |
9943 | raw_spin_lock_irq(&task->pi_lock); | |
9944 | /* | |
9945 | * Avoid calling sched_move_task() before wake_up_new_task() | |
9946 | * has happened. This would lead to problems with PELT, due to | |
9947 | * move wanting to detach+attach while we're not attached yet. | |
9948 | */ | |
2f064a59 | 9949 | if (READ_ONCE(task->__state) == TASK_NEW) |
7dc603c9 PZ |
9950 | ret = -EINVAL; |
9951 | raw_spin_unlock_irq(&task->pi_lock); | |
9952 | ||
9953 | if (ret) | |
9954 | break; | |
bb9d97b6 | 9955 | } |
7dc603c9 | 9956 | return ret; |
be367d09 | 9957 | } |
68318b8e | 9958 | |
1f7dd3e5 | 9959 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 9960 | { |
bb9d97b6 | 9961 | struct task_struct *task; |
1f7dd3e5 | 9962 | struct cgroup_subsys_state *css; |
bb9d97b6 | 9963 | |
1f7dd3e5 | 9964 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 9965 | sched_move_task(task); |
68318b8e SV |
9966 | } |
9967 | ||
2480c093 | 9968 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
9969 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
9970 | { | |
9971 | struct cgroup_subsys_state *top_css = css; | |
9972 | struct uclamp_se *uc_parent = NULL; | |
9973 | struct uclamp_se *uc_se = NULL; | |
9974 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 9975 | enum uclamp_id clamp_id; |
0b60ba2d PB |
9976 | unsigned int clamps; |
9977 | ||
93b73858 QY |
9978 | lockdep_assert_held(&uclamp_mutex); |
9979 | SCHED_WARN_ON(!rcu_read_lock_held()); | |
9980 | ||
0b60ba2d PB |
9981 | css_for_each_descendant_pre(css, top_css) { |
9982 | uc_parent = css_tg(css)->parent | |
9983 | ? css_tg(css)->parent->uclamp : NULL; | |
9984 | ||
9985 | for_each_clamp_id(clamp_id) { | |
9986 | /* Assume effective clamps matches requested clamps */ | |
9987 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
9988 | /* Cap effective clamps with parent's effective clamps */ | |
9989 | if (uc_parent && | |
9990 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
9991 | eff[clamp_id] = uc_parent[clamp_id].value; | |
9992 | } | |
9993 | } | |
9994 | /* Ensure protection is always capped by limit */ | |
9995 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
9996 | ||
9997 | /* Propagate most restrictive effective clamps */ | |
9998 | clamps = 0x0; | |
9999 | uc_se = css_tg(css)->uclamp; | |
10000 | for_each_clamp_id(clamp_id) { | |
10001 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
10002 | continue; | |
10003 | uc_se[clamp_id].value = eff[clamp_id]; | |
10004 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
10005 | clamps |= (0x1 << clamp_id); | |
10006 | } | |
babbe170 | 10007 | if (!clamps) { |
0b60ba2d | 10008 | css = css_rightmost_descendant(css); |
babbe170 PB |
10009 | continue; |
10010 | } | |
10011 | ||
10012 | /* Immediately update descendants RUNNABLE tasks */ | |
0213b708 | 10013 | uclamp_update_active_tasks(css); |
0b60ba2d PB |
10014 | } |
10015 | } | |
2480c093 PB |
10016 | |
10017 | /* | |
10018 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
10019 | * C expression. Since there is no way to convert a macro argument (N) into a | |
10020 | * character constant, use two levels of macros. | |
10021 | */ | |
10022 | #define _POW10(exp) ((unsigned int)1e##exp) | |
10023 | #define POW10(exp) _POW10(exp) | |
10024 | ||
10025 | struct uclamp_request { | |
10026 | #define UCLAMP_PERCENT_SHIFT 2 | |
10027 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
10028 | s64 percent; | |
10029 | u64 util; | |
10030 | int ret; | |
10031 | }; | |
10032 | ||
10033 | static inline struct uclamp_request | |
10034 | capacity_from_percent(char *buf) | |
10035 | { | |
10036 | struct uclamp_request req = { | |
10037 | .percent = UCLAMP_PERCENT_SCALE, | |
10038 | .util = SCHED_CAPACITY_SCALE, | |
10039 | .ret = 0, | |
10040 | }; | |
10041 | ||
10042 | buf = strim(buf); | |
10043 | if (strcmp(buf, "max")) { | |
10044 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
10045 | &req.percent); | |
10046 | if (req.ret) | |
10047 | return req; | |
b562d140 | 10048 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
10049 | req.ret = -ERANGE; |
10050 | return req; | |
10051 | } | |
10052 | ||
10053 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
10054 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
10055 | } | |
10056 | ||
10057 | return req; | |
10058 | } | |
10059 | ||
10060 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
10061 | size_t nbytes, loff_t off, | |
10062 | enum uclamp_id clamp_id) | |
10063 | { | |
10064 | struct uclamp_request req; | |
10065 | struct task_group *tg; | |
10066 | ||
10067 | req = capacity_from_percent(buf); | |
10068 | if (req.ret) | |
10069 | return req.ret; | |
10070 | ||
46609ce2 QY |
10071 | static_branch_enable(&sched_uclamp_used); |
10072 | ||
2480c093 PB |
10073 | mutex_lock(&uclamp_mutex); |
10074 | rcu_read_lock(); | |
10075 | ||
10076 | tg = css_tg(of_css(of)); | |
10077 | if (tg->uclamp_req[clamp_id].value != req.util) | |
10078 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
10079 | ||
10080 | /* | |
10081 | * Because of not recoverable conversion rounding we keep track of the | |
10082 | * exact requested value | |
10083 | */ | |
10084 | tg->uclamp_pct[clamp_id] = req.percent; | |
10085 | ||
0b60ba2d PB |
10086 | /* Update effective clamps to track the most restrictive value */ |
10087 | cpu_util_update_eff(of_css(of)); | |
10088 | ||
2480c093 PB |
10089 | rcu_read_unlock(); |
10090 | mutex_unlock(&uclamp_mutex); | |
10091 | ||
10092 | return nbytes; | |
10093 | } | |
10094 | ||
10095 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
10096 | char *buf, size_t nbytes, | |
10097 | loff_t off) | |
10098 | { | |
10099 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
10100 | } | |
10101 | ||
10102 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
10103 | char *buf, size_t nbytes, | |
10104 | loff_t off) | |
10105 | { | |
10106 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
10107 | } | |
10108 | ||
10109 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
10110 | enum uclamp_id clamp_id) | |
10111 | { | |
10112 | struct task_group *tg; | |
10113 | u64 util_clamp; | |
10114 | u64 percent; | |
10115 | u32 rem; | |
10116 | ||
10117 | rcu_read_lock(); | |
10118 | tg = css_tg(seq_css(sf)); | |
10119 | util_clamp = tg->uclamp_req[clamp_id].value; | |
10120 | rcu_read_unlock(); | |
10121 | ||
10122 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
10123 | seq_puts(sf, "max\n"); | |
10124 | return; | |
10125 | } | |
10126 | ||
10127 | percent = tg->uclamp_pct[clamp_id]; | |
10128 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
10129 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
10130 | } | |
10131 | ||
10132 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
10133 | { | |
10134 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
10135 | return 0; | |
10136 | } | |
10137 | ||
10138 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
10139 | { | |
10140 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
10141 | return 0; | |
10142 | } | |
10143 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
10144 | ||
052f1dc7 | 10145 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
10146 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
10147 | struct cftype *cftype, u64 shareval) | |
68318b8e | 10148 | { |
5b61d50a KK |
10149 | if (shareval > scale_load_down(ULONG_MAX)) |
10150 | shareval = MAX_SHARES; | |
182446d0 | 10151 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
10152 | } |
10153 | ||
182446d0 TH |
10154 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
10155 | struct cftype *cft) | |
68318b8e | 10156 | { |
182446d0 | 10157 | struct task_group *tg = css_tg(css); |
68318b8e | 10158 | |
c8b28116 | 10159 | return (u64) scale_load_down(tg->shares); |
68318b8e | 10160 | } |
ab84d31e PT |
10161 | |
10162 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
10163 | static DEFINE_MUTEX(cfs_constraints_mutex); |
10164 | ||
ab84d31e | 10165 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 10166 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
10167 | /* More than 203 days if BW_SHIFT equals 20. */ |
10168 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 10169 | |
a790de99 PT |
10170 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
10171 | ||
f4183717 HC |
10172 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota, |
10173 | u64 burst) | |
ab84d31e | 10174 | { |
56f570e5 | 10175 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 10176 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
10177 | |
10178 | if (tg == &root_task_group) | |
10179 | return -EINVAL; | |
10180 | ||
10181 | /* | |
10182 | * Ensure we have at some amount of bandwidth every period. This is | |
10183 | * to prevent reaching a state of large arrears when throttled via | |
10184 | * entity_tick() resulting in prolonged exit starvation. | |
10185 | */ | |
10186 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
10187 | return -EINVAL; | |
10188 | ||
10189 | /* | |
3b03706f | 10190 | * Likewise, bound things on the other side by preventing insane quota |
ab84d31e PT |
10191 | * periods. This also allows us to normalize in computing quota |
10192 | * feasibility. | |
10193 | */ | |
10194 | if (period > max_cfs_quota_period) | |
10195 | return -EINVAL; | |
10196 | ||
d505b8af HC |
10197 | /* |
10198 | * Bound quota to defend quota against overflow during bandwidth shift. | |
10199 | */ | |
10200 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
10201 | return -EINVAL; | |
10202 | ||
f4183717 HC |
10203 | if (quota != RUNTIME_INF && (burst > quota || |
10204 | burst + quota > max_cfs_runtime)) | |
10205 | return -EINVAL; | |
10206 | ||
0e59bdae KT |
10207 | /* |
10208 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
10209 | * unthrottle_offline_cfs_rqs(). | |
10210 | */ | |
746f5ea9 | 10211 | cpus_read_lock(); |
a790de99 PT |
10212 | mutex_lock(&cfs_constraints_mutex); |
10213 | ret = __cfs_schedulable(tg, period, quota); | |
10214 | if (ret) | |
10215 | goto out_unlock; | |
10216 | ||
58088ad0 | 10217 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 10218 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
10219 | /* |
10220 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
10221 | * before making related changes, and on->off must occur afterwards | |
10222 | */ | |
10223 | if (runtime_enabled && !runtime_was_enabled) | |
10224 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
10225 | raw_spin_lock_irq(&cfs_b->lock); |
10226 | cfs_b->period = ns_to_ktime(period); | |
10227 | cfs_b->quota = quota; | |
f4183717 | 10228 | cfs_b->burst = burst; |
58088ad0 | 10229 | |
a9cf55b2 | 10230 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
10231 | |
10232 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
10233 | if (runtime_enabled) |
10234 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 10235 | |
ab84d31e PT |
10236 | raw_spin_unlock_irq(&cfs_b->lock); |
10237 | ||
0e59bdae | 10238 | for_each_online_cpu(i) { |
ab84d31e | 10239 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 10240 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 10241 | struct rq_flags rf; |
ab84d31e | 10242 | |
8a8c69c3 | 10243 | rq_lock_irq(rq, &rf); |
58088ad0 | 10244 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 10245 | cfs_rq->runtime_remaining = 0; |
671fd9da | 10246 | |
029632fb | 10247 | if (cfs_rq->throttled) |
671fd9da | 10248 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 10249 | rq_unlock_irq(rq, &rf); |
ab84d31e | 10250 | } |
1ee14e6c BS |
10251 | if (runtime_was_enabled && !runtime_enabled) |
10252 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
10253 | out_unlock: |
10254 | mutex_unlock(&cfs_constraints_mutex); | |
746f5ea9 | 10255 | cpus_read_unlock(); |
ab84d31e | 10256 | |
a790de99 | 10257 | return ret; |
ab84d31e PT |
10258 | } |
10259 | ||
b1546edc | 10260 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e | 10261 | { |
f4183717 | 10262 | u64 quota, period, burst; |
ab84d31e | 10263 | |
029632fb | 10264 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
f4183717 | 10265 | burst = tg->cfs_bandwidth.burst; |
ab84d31e PT |
10266 | if (cfs_quota_us < 0) |
10267 | quota = RUNTIME_INF; | |
1a8b4540 | 10268 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 10269 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
10270 | else |
10271 | return -EINVAL; | |
ab84d31e | 10272 | |
f4183717 | 10273 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10274 | } |
10275 | ||
b1546edc | 10276 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
10277 | { |
10278 | u64 quota_us; | |
10279 | ||
029632fb | 10280 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
10281 | return -1; |
10282 | ||
029632fb | 10283 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
10284 | do_div(quota_us, NSEC_PER_USEC); |
10285 | ||
10286 | return quota_us; | |
10287 | } | |
10288 | ||
b1546edc | 10289 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e | 10290 | { |
f4183717 | 10291 | u64 quota, period, burst; |
ab84d31e | 10292 | |
1a8b4540 KK |
10293 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
10294 | return -EINVAL; | |
10295 | ||
ab84d31e | 10296 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 10297 | quota = tg->cfs_bandwidth.quota; |
f4183717 | 10298 | burst = tg->cfs_bandwidth.burst; |
ab84d31e | 10299 | |
f4183717 | 10300 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10301 | } |
10302 | ||
b1546edc | 10303 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
10304 | { |
10305 | u64 cfs_period_us; | |
10306 | ||
029632fb | 10307 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
10308 | do_div(cfs_period_us, NSEC_PER_USEC); |
10309 | ||
10310 | return cfs_period_us; | |
10311 | } | |
10312 | ||
f4183717 HC |
10313 | static int tg_set_cfs_burst(struct task_group *tg, long cfs_burst_us) |
10314 | { | |
10315 | u64 quota, period, burst; | |
10316 | ||
10317 | if ((u64)cfs_burst_us > U64_MAX / NSEC_PER_USEC) | |
10318 | return -EINVAL; | |
10319 | ||
10320 | burst = (u64)cfs_burst_us * NSEC_PER_USEC; | |
10321 | period = ktime_to_ns(tg->cfs_bandwidth.period); | |
10322 | quota = tg->cfs_bandwidth.quota; | |
10323 | ||
10324 | return tg_set_cfs_bandwidth(tg, period, quota, burst); | |
10325 | } | |
10326 | ||
10327 | static long tg_get_cfs_burst(struct task_group *tg) | |
10328 | { | |
10329 | u64 burst_us; | |
10330 | ||
10331 | burst_us = tg->cfs_bandwidth.burst; | |
10332 | do_div(burst_us, NSEC_PER_USEC); | |
10333 | ||
10334 | return burst_us; | |
10335 | } | |
10336 | ||
182446d0 TH |
10337 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
10338 | struct cftype *cft) | |
ab84d31e | 10339 | { |
182446d0 | 10340 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
10341 | } |
10342 | ||
182446d0 TH |
10343 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
10344 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 10345 | { |
182446d0 | 10346 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
10347 | } |
10348 | ||
182446d0 TH |
10349 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
10350 | struct cftype *cft) | |
ab84d31e | 10351 | { |
182446d0 | 10352 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
10353 | } |
10354 | ||
182446d0 TH |
10355 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
10356 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 10357 | { |
182446d0 | 10358 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
10359 | } |
10360 | ||
f4183717 HC |
10361 | static u64 cpu_cfs_burst_read_u64(struct cgroup_subsys_state *css, |
10362 | struct cftype *cft) | |
10363 | { | |
10364 | return tg_get_cfs_burst(css_tg(css)); | |
10365 | } | |
10366 | ||
10367 | static int cpu_cfs_burst_write_u64(struct cgroup_subsys_state *css, | |
10368 | struct cftype *cftype, u64 cfs_burst_us) | |
10369 | { | |
10370 | return tg_set_cfs_burst(css_tg(css), cfs_burst_us); | |
10371 | } | |
10372 | ||
a790de99 PT |
10373 | struct cfs_schedulable_data { |
10374 | struct task_group *tg; | |
10375 | u64 period, quota; | |
10376 | }; | |
10377 | ||
10378 | /* | |
10379 | * normalize group quota/period to be quota/max_period | |
10380 | * note: units are usecs | |
10381 | */ | |
10382 | static u64 normalize_cfs_quota(struct task_group *tg, | |
10383 | struct cfs_schedulable_data *d) | |
10384 | { | |
10385 | u64 quota, period; | |
10386 | ||
10387 | if (tg == d->tg) { | |
10388 | period = d->period; | |
10389 | quota = d->quota; | |
10390 | } else { | |
10391 | period = tg_get_cfs_period(tg); | |
10392 | quota = tg_get_cfs_quota(tg); | |
10393 | } | |
10394 | ||
10395 | /* note: these should typically be equivalent */ | |
10396 | if (quota == RUNTIME_INF || quota == -1) | |
10397 | return RUNTIME_INF; | |
10398 | ||
10399 | return to_ratio(period, quota); | |
10400 | } | |
10401 | ||
10402 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
10403 | { | |
10404 | struct cfs_schedulable_data *d = data; | |
029632fb | 10405 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
10406 | s64 quota = 0, parent_quota = -1; |
10407 | ||
10408 | if (!tg->parent) { | |
10409 | quota = RUNTIME_INF; | |
10410 | } else { | |
029632fb | 10411 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
10412 | |
10413 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 10414 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
10415 | |
10416 | /* | |
c53593e5 TH |
10417 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
10418 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 10419 | * limit is set: |
a790de99 | 10420 | */ |
c53593e5 TH |
10421 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
10422 | quota = min(quota, parent_quota); | |
10423 | } else { | |
10424 | if (quota == RUNTIME_INF) | |
10425 | quota = parent_quota; | |
10426 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
10427 | return -EINVAL; | |
10428 | } | |
a790de99 | 10429 | } |
9c58c79a | 10430 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
10431 | |
10432 | return 0; | |
10433 | } | |
10434 | ||
10435 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
10436 | { | |
8277434e | 10437 | int ret; |
a790de99 PT |
10438 | struct cfs_schedulable_data data = { |
10439 | .tg = tg, | |
10440 | .period = period, | |
10441 | .quota = quota, | |
10442 | }; | |
10443 | ||
10444 | if (quota != RUNTIME_INF) { | |
10445 | do_div(data.period, NSEC_PER_USEC); | |
10446 | do_div(data.quota, NSEC_PER_USEC); | |
10447 | } | |
10448 | ||
8277434e PT |
10449 | rcu_read_lock(); |
10450 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
10451 | rcu_read_unlock(); | |
10452 | ||
10453 | return ret; | |
a790de99 | 10454 | } |
e8da1b18 | 10455 | |
a1f7164c | 10456 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 10457 | { |
2da8ca82 | 10458 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 10459 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 10460 | |
44ffc75b TH |
10461 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
10462 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
10463 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 10464 | |
3d6c50c2 YW |
10465 | if (schedstat_enabled() && tg != &root_task_group) { |
10466 | u64 ws = 0; | |
10467 | int i; | |
10468 | ||
10469 | for_each_possible_cpu(i) | |
10470 | ws += schedstat_val(tg->se[i]->statistics.wait_sum); | |
10471 | ||
10472 | seq_printf(sf, "wait_sum %llu\n", ws); | |
10473 | } | |
10474 | ||
e8da1b18 NR |
10475 | return 0; |
10476 | } | |
ab84d31e | 10477 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 10478 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10479 | |
052f1dc7 | 10480 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
10481 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
10482 | struct cftype *cft, s64 val) | |
6f505b16 | 10483 | { |
182446d0 | 10484 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
10485 | } |
10486 | ||
182446d0 TH |
10487 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
10488 | struct cftype *cft) | |
6f505b16 | 10489 | { |
182446d0 | 10490 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 10491 | } |
d0b27fa7 | 10492 | |
182446d0 TH |
10493 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
10494 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 10495 | { |
182446d0 | 10496 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
10497 | } |
10498 | ||
182446d0 TH |
10499 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
10500 | struct cftype *cft) | |
d0b27fa7 | 10501 | { |
182446d0 | 10502 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 10503 | } |
6d6bc0ad | 10504 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10505 | |
30400039 JD |
10506 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10507 | static s64 cpu_idle_read_s64(struct cgroup_subsys_state *css, | |
10508 | struct cftype *cft) | |
10509 | { | |
10510 | return css_tg(css)->idle; | |
10511 | } | |
10512 | ||
10513 | static int cpu_idle_write_s64(struct cgroup_subsys_state *css, | |
10514 | struct cftype *cft, s64 idle) | |
10515 | { | |
10516 | return sched_group_set_idle(css_tg(css), idle); | |
10517 | } | |
10518 | #endif | |
10519 | ||
a1f7164c | 10520 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 10521 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10522 | { |
10523 | .name = "shares", | |
f4c753b7 PM |
10524 | .read_u64 = cpu_shares_read_u64, |
10525 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10526 | }, |
30400039 JD |
10527 | { |
10528 | .name = "idle", | |
10529 | .read_s64 = cpu_idle_read_s64, | |
10530 | .write_s64 = cpu_idle_write_s64, | |
10531 | }, | |
052f1dc7 | 10532 | #endif |
ab84d31e PT |
10533 | #ifdef CONFIG_CFS_BANDWIDTH |
10534 | { | |
10535 | .name = "cfs_quota_us", | |
10536 | .read_s64 = cpu_cfs_quota_read_s64, | |
10537 | .write_s64 = cpu_cfs_quota_write_s64, | |
10538 | }, | |
10539 | { | |
10540 | .name = "cfs_period_us", | |
10541 | .read_u64 = cpu_cfs_period_read_u64, | |
10542 | .write_u64 = cpu_cfs_period_write_u64, | |
10543 | }, | |
f4183717 HC |
10544 | { |
10545 | .name = "cfs_burst_us", | |
10546 | .read_u64 = cpu_cfs_burst_read_u64, | |
10547 | .write_u64 = cpu_cfs_burst_write_u64, | |
10548 | }, | |
e8da1b18 NR |
10549 | { |
10550 | .name = "stat", | |
a1f7164c | 10551 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 10552 | }, |
ab84d31e | 10553 | #endif |
052f1dc7 | 10554 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10555 | { |
9f0c1e56 | 10556 | .name = "rt_runtime_us", |
06ecb27c PM |
10557 | .read_s64 = cpu_rt_runtime_read, |
10558 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10559 | }, |
d0b27fa7 PZ |
10560 | { |
10561 | .name = "rt_period_us", | |
f4c753b7 PM |
10562 | .read_u64 = cpu_rt_period_read_uint, |
10563 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10564 | }, |
2480c093 PB |
10565 | #endif |
10566 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
10567 | { | |
10568 | .name = "uclamp.min", | |
10569 | .flags = CFTYPE_NOT_ON_ROOT, | |
10570 | .seq_show = cpu_uclamp_min_show, | |
10571 | .write = cpu_uclamp_min_write, | |
10572 | }, | |
10573 | { | |
10574 | .name = "uclamp.max", | |
10575 | .flags = CFTYPE_NOT_ON_ROOT, | |
10576 | .seq_show = cpu_uclamp_max_show, | |
10577 | .write = cpu_uclamp_max_write, | |
10578 | }, | |
052f1dc7 | 10579 | #endif |
d1ccc66d | 10580 | { } /* Terminate */ |
68318b8e SV |
10581 | }; |
10582 | ||
d41bf8c9 TH |
10583 | static int cpu_extra_stat_show(struct seq_file *sf, |
10584 | struct cgroup_subsys_state *css) | |
0d593634 | 10585 | { |
0d593634 TH |
10586 | #ifdef CONFIG_CFS_BANDWIDTH |
10587 | { | |
d41bf8c9 | 10588 | struct task_group *tg = css_tg(css); |
0d593634 TH |
10589 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
10590 | u64 throttled_usec; | |
10591 | ||
10592 | throttled_usec = cfs_b->throttled_time; | |
10593 | do_div(throttled_usec, NSEC_PER_USEC); | |
10594 | ||
10595 | seq_printf(sf, "nr_periods %d\n" | |
10596 | "nr_throttled %d\n" | |
10597 | "throttled_usec %llu\n", | |
10598 | cfs_b->nr_periods, cfs_b->nr_throttled, | |
10599 | throttled_usec); | |
10600 | } | |
10601 | #endif | |
10602 | return 0; | |
10603 | } | |
10604 | ||
10605 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
10606 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
10607 | struct cftype *cft) | |
10608 | { | |
10609 | struct task_group *tg = css_tg(css); | |
10610 | u64 weight = scale_load_down(tg->shares); | |
10611 | ||
10612 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
10613 | } | |
10614 | ||
10615 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
10616 | struct cftype *cft, u64 weight) | |
10617 | { | |
10618 | /* | |
10619 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
10620 | * values which are 1, 100 and 10000 respectively. While it loses | |
10621 | * a bit of range on both ends, it maps pretty well onto the shares | |
10622 | * value used by scheduler and the round-trip conversions preserve | |
10623 | * the original value over the entire range. | |
10624 | */ | |
10625 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
10626 | return -ERANGE; | |
10627 | ||
10628 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
10629 | ||
10630 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
10631 | } | |
10632 | ||
10633 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
10634 | struct cftype *cft) | |
10635 | { | |
10636 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
10637 | int last_delta = INT_MAX; | |
10638 | int prio, delta; | |
10639 | ||
10640 | /* find the closest nice value to the current weight */ | |
10641 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
10642 | delta = abs(sched_prio_to_weight[prio] - weight); | |
10643 | if (delta >= last_delta) | |
10644 | break; | |
10645 | last_delta = delta; | |
10646 | } | |
10647 | ||
10648 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
10649 | } | |
10650 | ||
10651 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
10652 | struct cftype *cft, s64 nice) | |
10653 | { | |
10654 | unsigned long weight; | |
7281c8de | 10655 | int idx; |
0d593634 TH |
10656 | |
10657 | if (nice < MIN_NICE || nice > MAX_NICE) | |
10658 | return -ERANGE; | |
10659 | ||
7281c8de PZ |
10660 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
10661 | idx = array_index_nospec(idx, 40); | |
10662 | weight = sched_prio_to_weight[idx]; | |
10663 | ||
0d593634 TH |
10664 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
10665 | } | |
10666 | #endif | |
10667 | ||
10668 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
10669 | long period, long quota) | |
10670 | { | |
10671 | if (quota < 0) | |
10672 | seq_puts(sf, "max"); | |
10673 | else | |
10674 | seq_printf(sf, "%ld", quota); | |
10675 | ||
10676 | seq_printf(sf, " %ld\n", period); | |
10677 | } | |
10678 | ||
10679 | /* caller should put the current value in *@periodp before calling */ | |
10680 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
10681 | u64 *periodp, u64 *quotap) | |
10682 | { | |
10683 | char tok[21]; /* U64_MAX */ | |
10684 | ||
4c47acd8 | 10685 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
10686 | return -EINVAL; |
10687 | ||
10688 | *periodp *= NSEC_PER_USEC; | |
10689 | ||
10690 | if (sscanf(tok, "%llu", quotap)) | |
10691 | *quotap *= NSEC_PER_USEC; | |
10692 | else if (!strcmp(tok, "max")) | |
10693 | *quotap = RUNTIME_INF; | |
10694 | else | |
10695 | return -EINVAL; | |
10696 | ||
10697 | return 0; | |
10698 | } | |
10699 | ||
10700 | #ifdef CONFIG_CFS_BANDWIDTH | |
10701 | static int cpu_max_show(struct seq_file *sf, void *v) | |
10702 | { | |
10703 | struct task_group *tg = css_tg(seq_css(sf)); | |
10704 | ||
10705 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
10706 | return 0; | |
10707 | } | |
10708 | ||
10709 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
10710 | char *buf, size_t nbytes, loff_t off) | |
10711 | { | |
10712 | struct task_group *tg = css_tg(of_css(of)); | |
10713 | u64 period = tg_get_cfs_period(tg); | |
f4183717 | 10714 | u64 burst = tg_get_cfs_burst(tg); |
0d593634 TH |
10715 | u64 quota; |
10716 | int ret; | |
10717 | ||
10718 | ret = cpu_period_quota_parse(buf, &period, "a); | |
10719 | if (!ret) | |
f4183717 | 10720 | ret = tg_set_cfs_bandwidth(tg, period, quota, burst); |
0d593634 TH |
10721 | return ret ?: nbytes; |
10722 | } | |
10723 | #endif | |
10724 | ||
10725 | static struct cftype cpu_files[] = { | |
0d593634 TH |
10726 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10727 | { | |
10728 | .name = "weight", | |
10729 | .flags = CFTYPE_NOT_ON_ROOT, | |
10730 | .read_u64 = cpu_weight_read_u64, | |
10731 | .write_u64 = cpu_weight_write_u64, | |
10732 | }, | |
10733 | { | |
10734 | .name = "weight.nice", | |
10735 | .flags = CFTYPE_NOT_ON_ROOT, | |
10736 | .read_s64 = cpu_weight_nice_read_s64, | |
10737 | .write_s64 = cpu_weight_nice_write_s64, | |
10738 | }, | |
30400039 JD |
10739 | { |
10740 | .name = "idle", | |
10741 | .flags = CFTYPE_NOT_ON_ROOT, | |
10742 | .read_s64 = cpu_idle_read_s64, | |
10743 | .write_s64 = cpu_idle_write_s64, | |
10744 | }, | |
0d593634 TH |
10745 | #endif |
10746 | #ifdef CONFIG_CFS_BANDWIDTH | |
10747 | { | |
10748 | .name = "max", | |
10749 | .flags = CFTYPE_NOT_ON_ROOT, | |
10750 | .seq_show = cpu_max_show, | |
10751 | .write = cpu_max_write, | |
10752 | }, | |
f4183717 HC |
10753 | { |
10754 | .name = "max.burst", | |
10755 | .flags = CFTYPE_NOT_ON_ROOT, | |
10756 | .read_u64 = cpu_cfs_burst_read_u64, | |
10757 | .write_u64 = cpu_cfs_burst_write_u64, | |
10758 | }, | |
2480c093 PB |
10759 | #endif |
10760 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
10761 | { | |
10762 | .name = "uclamp.min", | |
10763 | .flags = CFTYPE_NOT_ON_ROOT, | |
10764 | .seq_show = cpu_uclamp_min_show, | |
10765 | .write = cpu_uclamp_min_write, | |
10766 | }, | |
10767 | { | |
10768 | .name = "uclamp.max", | |
10769 | .flags = CFTYPE_NOT_ON_ROOT, | |
10770 | .seq_show = cpu_uclamp_max_show, | |
10771 | .write = cpu_uclamp_max_write, | |
10772 | }, | |
0d593634 TH |
10773 | #endif |
10774 | { } /* terminate */ | |
10775 | }; | |
10776 | ||
073219e9 | 10777 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 10778 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 10779 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 10780 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 10781 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 10782 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 10783 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
10784 | .can_attach = cpu_cgroup_can_attach, |
10785 | .attach = cpu_cgroup_attach, | |
a1f7164c | 10786 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 10787 | .dfl_cftypes = cpu_files, |
b38e42e9 | 10788 | .early_init = true, |
0d593634 | 10789 | .threaded = true, |
68318b8e SV |
10790 | }; |
10791 | ||
052f1dc7 | 10792 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 10793 | |
b637a328 PM |
10794 | void dump_cpu_task(int cpu) |
10795 | { | |
10796 | pr_info("Task dump for CPU %d:\n", cpu); | |
10797 | sched_show_task(cpu_curr(cpu)); | |
10798 | } | |
ed82b8a1 AK |
10799 | |
10800 | /* | |
10801 | * Nice levels are multiplicative, with a gentle 10% change for every | |
10802 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
10803 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
10804 | * that remained on nice 0. | |
10805 | * | |
10806 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
10807 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
10808 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
10809 | * If a task goes up by ~10% and another task goes down by ~10% then | |
10810 | * the relative distance between them is ~25%.) | |
10811 | */ | |
10812 | const int sched_prio_to_weight[40] = { | |
10813 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
10814 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
10815 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
10816 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
10817 | /* 0 */ 1024, 820, 655, 526, 423, | |
10818 | /* 5 */ 335, 272, 215, 172, 137, | |
10819 | /* 10 */ 110, 87, 70, 56, 45, | |
10820 | /* 15 */ 36, 29, 23, 18, 15, | |
10821 | }; | |
10822 | ||
10823 | /* | |
10824 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
10825 | * | |
10826 | * In cases where the weight does not change often, we can use the | |
10827 | * precalculated inverse to speed up arithmetics by turning divisions | |
10828 | * into multiplications: | |
10829 | */ | |
10830 | const u32 sched_prio_to_wmult[40] = { | |
10831 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
10832 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
10833 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
10834 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
10835 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
10836 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
10837 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
10838 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
10839 | }; | |
14a7405b | 10840 | |
9d246053 PA |
10841 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
10842 | { | |
10843 | trace_sched_update_nr_running_tp(rq, count); | |
10844 | } |