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> |
6a5850d1 | 16 | #include <linux/blkdev.h> |
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 | */ | |
691925f3 TG |
77 | #ifdef CONFIG_PREEMPT_RT |
78 | const_debug unsigned int sysctl_sched_nr_migrate = 8; | |
79 | #else | |
b82d9fdd | 80 | const_debug unsigned int sysctl_sched_nr_migrate = 32; |
691925f3 | 81 | #endif |
b82d9fdd | 82 | |
fa85ae24 | 83 | /* |
d1ccc66d | 84 | * period over which we measure -rt task CPU usage in us. |
fa85ae24 PZ |
85 | * default: 1s |
86 | */ | |
9f0c1e56 | 87 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 88 | |
029632fb | 89 | __read_mostly int scheduler_running; |
6892b75e | 90 | |
9edeaea1 PZ |
91 | #ifdef CONFIG_SCHED_CORE |
92 | ||
93 | DEFINE_STATIC_KEY_FALSE(__sched_core_enabled); | |
94 | ||
8a311c74 PZ |
95 | /* kernel prio, less is more */ |
96 | static inline int __task_prio(struct task_struct *p) | |
97 | { | |
98 | if (p->sched_class == &stop_sched_class) /* trumps deadline */ | |
99 | return -2; | |
100 | ||
101 | if (rt_prio(p->prio)) /* includes deadline */ | |
102 | return p->prio; /* [-1, 99] */ | |
103 | ||
104 | if (p->sched_class == &idle_sched_class) | |
105 | return MAX_RT_PRIO + NICE_WIDTH; /* 140 */ | |
106 | ||
107 | return MAX_RT_PRIO + MAX_NICE; /* 120, squash fair */ | |
108 | } | |
109 | ||
110 | /* | |
111 | * l(a,b) | |
112 | * le(a,b) := !l(b,a) | |
113 | * g(a,b) := l(b,a) | |
114 | * ge(a,b) := !l(a,b) | |
115 | */ | |
116 | ||
117 | /* real prio, less is less */ | |
c6047c2e | 118 | static inline bool prio_less(struct task_struct *a, struct task_struct *b, bool in_fi) |
8a311c74 PZ |
119 | { |
120 | ||
121 | int pa = __task_prio(a), pb = __task_prio(b); | |
122 | ||
123 | if (-pa < -pb) | |
124 | return true; | |
125 | ||
126 | if (-pb < -pa) | |
127 | return false; | |
128 | ||
129 | if (pa == -1) /* dl_prio() doesn't work because of stop_class above */ | |
130 | return !dl_time_before(a->dl.deadline, b->dl.deadline); | |
131 | ||
c6047c2e JFG |
132 | if (pa == MAX_RT_PRIO + MAX_NICE) /* fair */ |
133 | return cfs_prio_less(a, b, in_fi); | |
8a311c74 PZ |
134 | |
135 | return false; | |
136 | } | |
137 | ||
138 | static inline bool __sched_core_less(struct task_struct *a, struct task_struct *b) | |
139 | { | |
140 | if (a->core_cookie < b->core_cookie) | |
141 | return true; | |
142 | ||
143 | if (a->core_cookie > b->core_cookie) | |
144 | return false; | |
145 | ||
146 | /* flip prio, so high prio is leftmost */ | |
c6047c2e | 147 | if (prio_less(b, a, task_rq(a)->core->core_forceidle)) |
8a311c74 PZ |
148 | return true; |
149 | ||
150 | return false; | |
151 | } | |
152 | ||
153 | #define __node_2_sc(node) rb_entry((node), struct task_struct, core_node) | |
154 | ||
155 | static inline bool rb_sched_core_less(struct rb_node *a, const struct rb_node *b) | |
156 | { | |
157 | return __sched_core_less(__node_2_sc(a), __node_2_sc(b)); | |
158 | } | |
159 | ||
160 | static inline int rb_sched_core_cmp(const void *key, const struct rb_node *node) | |
161 | { | |
162 | const struct task_struct *p = __node_2_sc(node); | |
163 | unsigned long cookie = (unsigned long)key; | |
164 | ||
165 | if (cookie < p->core_cookie) | |
166 | return -1; | |
167 | ||
168 | if (cookie > p->core_cookie) | |
169 | return 1; | |
170 | ||
171 | return 0; | |
172 | } | |
173 | ||
6e33cad0 | 174 | void sched_core_enqueue(struct rq *rq, struct task_struct *p) |
8a311c74 PZ |
175 | { |
176 | rq->core->core_task_seq++; | |
177 | ||
178 | if (!p->core_cookie) | |
179 | return; | |
180 | ||
181 | rb_add(&p->core_node, &rq->core_tree, rb_sched_core_less); | |
182 | } | |
183 | ||
6e33cad0 | 184 | void sched_core_dequeue(struct rq *rq, struct task_struct *p) |
8a311c74 PZ |
185 | { |
186 | rq->core->core_task_seq++; | |
187 | ||
6e33cad0 | 188 | if (!sched_core_enqueued(p)) |
8a311c74 PZ |
189 | return; |
190 | ||
191 | rb_erase(&p->core_node, &rq->core_tree); | |
6e33cad0 | 192 | RB_CLEAR_NODE(&p->core_node); |
8a311c74 PZ |
193 | } |
194 | ||
195 | /* | |
196 | * Find left-most (aka, highest priority) task matching @cookie. | |
197 | */ | |
198 | static struct task_struct *sched_core_find(struct rq *rq, unsigned long cookie) | |
199 | { | |
200 | struct rb_node *node; | |
201 | ||
202 | node = rb_find_first((void *)cookie, &rq->core_tree, rb_sched_core_cmp); | |
203 | /* | |
204 | * The idle task always matches any cookie! | |
205 | */ | |
206 | if (!node) | |
207 | return idle_sched_class.pick_task(rq); | |
208 | ||
209 | return __node_2_sc(node); | |
210 | } | |
211 | ||
d2dfa17b PZ |
212 | static struct task_struct *sched_core_next(struct task_struct *p, unsigned long cookie) |
213 | { | |
214 | struct rb_node *node = &p->core_node; | |
215 | ||
216 | node = rb_next(node); | |
217 | if (!node) | |
218 | return NULL; | |
219 | ||
220 | p = container_of(node, struct task_struct, core_node); | |
221 | if (p->core_cookie != cookie) | |
222 | return NULL; | |
223 | ||
224 | return p; | |
225 | } | |
226 | ||
9edeaea1 PZ |
227 | /* |
228 | * Magic required such that: | |
229 | * | |
230 | * raw_spin_rq_lock(rq); | |
231 | * ... | |
232 | * raw_spin_rq_unlock(rq); | |
233 | * | |
234 | * ends up locking and unlocking the _same_ lock, and all CPUs | |
235 | * always agree on what rq has what lock. | |
236 | * | |
237 | * XXX entirely possible to selectively enable cores, don't bother for now. | |
238 | */ | |
239 | ||
240 | static DEFINE_MUTEX(sched_core_mutex); | |
875feb41 | 241 | static atomic_t sched_core_count; |
9edeaea1 PZ |
242 | static struct cpumask sched_core_mask; |
243 | ||
3c474b32 PZ |
244 | static void sched_core_lock(int cpu, unsigned long *flags) |
245 | { | |
246 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
247 | int t, i = 0; | |
248 | ||
249 | local_irq_save(*flags); | |
250 | for_each_cpu(t, smt_mask) | |
251 | raw_spin_lock_nested(&cpu_rq(t)->__lock, i++); | |
252 | } | |
253 | ||
254 | static void sched_core_unlock(int cpu, unsigned long *flags) | |
255 | { | |
256 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
257 | int t; | |
258 | ||
259 | for_each_cpu(t, smt_mask) | |
260 | raw_spin_unlock(&cpu_rq(t)->__lock); | |
261 | local_irq_restore(*flags); | |
262 | } | |
263 | ||
9edeaea1 PZ |
264 | static void __sched_core_flip(bool enabled) |
265 | { | |
3c474b32 PZ |
266 | unsigned long flags; |
267 | int cpu, t; | |
9edeaea1 PZ |
268 | |
269 | cpus_read_lock(); | |
270 | ||
271 | /* | |
272 | * Toggle the online cores, one by one. | |
273 | */ | |
274 | cpumask_copy(&sched_core_mask, cpu_online_mask); | |
275 | for_each_cpu(cpu, &sched_core_mask) { | |
276 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
277 | ||
3c474b32 | 278 | sched_core_lock(cpu, &flags); |
9edeaea1 PZ |
279 | |
280 | for_each_cpu(t, smt_mask) | |
281 | cpu_rq(t)->core_enabled = enabled; | |
282 | ||
3c474b32 | 283 | sched_core_unlock(cpu, &flags); |
9edeaea1 PZ |
284 | |
285 | cpumask_andnot(&sched_core_mask, &sched_core_mask, smt_mask); | |
286 | } | |
287 | ||
288 | /* | |
289 | * Toggle the offline CPUs. | |
290 | */ | |
291 | cpumask_copy(&sched_core_mask, cpu_possible_mask); | |
292 | cpumask_andnot(&sched_core_mask, &sched_core_mask, cpu_online_mask); | |
293 | ||
294 | for_each_cpu(cpu, &sched_core_mask) | |
295 | cpu_rq(cpu)->core_enabled = enabled; | |
296 | ||
297 | cpus_read_unlock(); | |
298 | } | |
299 | ||
8a311c74 | 300 | static void sched_core_assert_empty(void) |
9edeaea1 | 301 | { |
8a311c74 | 302 | int cpu; |
9edeaea1 | 303 | |
8a311c74 PZ |
304 | for_each_possible_cpu(cpu) |
305 | WARN_ON_ONCE(!RB_EMPTY_ROOT(&cpu_rq(cpu)->core_tree)); | |
306 | } | |
307 | ||
308 | static void __sched_core_enable(void) | |
309 | { | |
9edeaea1 PZ |
310 | static_branch_enable(&__sched_core_enabled); |
311 | /* | |
312 | * Ensure all previous instances of raw_spin_rq_*lock() have finished | |
313 | * and future ones will observe !sched_core_disabled(). | |
314 | */ | |
315 | synchronize_rcu(); | |
316 | __sched_core_flip(true); | |
8a311c74 | 317 | sched_core_assert_empty(); |
9edeaea1 PZ |
318 | } |
319 | ||
320 | static void __sched_core_disable(void) | |
321 | { | |
8a311c74 | 322 | sched_core_assert_empty(); |
9edeaea1 PZ |
323 | __sched_core_flip(false); |
324 | static_branch_disable(&__sched_core_enabled); | |
325 | } | |
326 | ||
327 | void sched_core_get(void) | |
328 | { | |
875feb41 PZ |
329 | if (atomic_inc_not_zero(&sched_core_count)) |
330 | return; | |
331 | ||
9edeaea1 | 332 | mutex_lock(&sched_core_mutex); |
875feb41 | 333 | if (!atomic_read(&sched_core_count)) |
9edeaea1 | 334 | __sched_core_enable(); |
875feb41 PZ |
335 | |
336 | smp_mb__before_atomic(); | |
337 | atomic_inc(&sched_core_count); | |
9edeaea1 PZ |
338 | mutex_unlock(&sched_core_mutex); |
339 | } | |
340 | ||
875feb41 | 341 | static void __sched_core_put(struct work_struct *work) |
9edeaea1 | 342 | { |
875feb41 | 343 | if (atomic_dec_and_mutex_lock(&sched_core_count, &sched_core_mutex)) { |
9edeaea1 | 344 | __sched_core_disable(); |
875feb41 PZ |
345 | mutex_unlock(&sched_core_mutex); |
346 | } | |
347 | } | |
348 | ||
349 | void sched_core_put(void) | |
350 | { | |
351 | static DECLARE_WORK(_work, __sched_core_put); | |
352 | ||
353 | /* | |
354 | * "There can be only one" | |
355 | * | |
356 | * Either this is the last one, or we don't actually need to do any | |
357 | * 'work'. If it is the last *again*, we rely on | |
358 | * WORK_STRUCT_PENDING_BIT. | |
359 | */ | |
360 | if (!atomic_add_unless(&sched_core_count, -1, 1)) | |
361 | schedule_work(&_work); | |
9edeaea1 PZ |
362 | } |
363 | ||
8a311c74 PZ |
364 | #else /* !CONFIG_SCHED_CORE */ |
365 | ||
366 | static inline void sched_core_enqueue(struct rq *rq, struct task_struct *p) { } | |
367 | static inline void sched_core_dequeue(struct rq *rq, struct task_struct *p) { } | |
368 | ||
9edeaea1 PZ |
369 | #endif /* CONFIG_SCHED_CORE */ |
370 | ||
9f0c1e56 PZ |
371 | /* |
372 | * part of the period that we allow rt tasks to run in us. | |
373 | * default: 0.95s | |
374 | */ | |
375 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 376 | |
58877d34 PZ |
377 | |
378 | /* | |
379 | * Serialization rules: | |
380 | * | |
381 | * Lock order: | |
382 | * | |
383 | * p->pi_lock | |
384 | * rq->lock | |
385 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
386 | * | |
387 | * rq1->lock | |
388 | * rq2->lock where: rq1 < rq2 | |
389 | * | |
390 | * Regular state: | |
391 | * | |
392 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
393 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
b19a888c | 394 | * always looks at the local rq data structures to find the most eligible task |
58877d34 PZ |
395 | * to run next. |
396 | * | |
397 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
398 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
399 | * the local CPU to avoid bouncing the runqueue state around [ see | |
400 | * ttwu_queue_wakelist() ] | |
401 | * | |
402 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
403 | * complicated to avoid having to take two rq->locks. | |
404 | * | |
405 | * Special state: | |
406 | * | |
407 | * System-calls and anything external will use task_rq_lock() which acquires | |
408 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
409 | * stable while holding either lock: | |
410 | * | |
411 | * - sched_setaffinity()/ | |
412 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
413 | * - set_user_nice(): p->se.load, p->*prio | |
414 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
415 | * p->se.load, p->rt_priority, | |
416 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
417 | * - sched_setnuma(): p->numa_preferred_nid | |
418 | * - sched_move_task()/ | |
419 | * cpu_cgroup_fork(): p->sched_task_group | |
420 | * - uclamp_update_active() p->uclamp* | |
421 | * | |
422 | * p->state <- TASK_*: | |
423 | * | |
424 | * is changed locklessly using set_current_state(), __set_current_state() or | |
425 | * set_special_state(), see their respective comments, or by | |
426 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
427 | * concurrent self. | |
428 | * | |
429 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
430 | * | |
431 | * is set by activate_task() and cleared by deactivate_task(), under | |
432 | * rq->lock. Non-zero indicates the task is runnable, the special | |
433 | * ON_RQ_MIGRATING state is used for migration without holding both | |
434 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
435 | * | |
436 | * p->on_cpu <- { 0, 1 }: | |
437 | * | |
438 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
439 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
440 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
441 | * | |
442 | * [ The astute reader will observe that it is possible for two tasks on one | |
443 | * CPU to have ->on_cpu = 1 at the same time. ] | |
444 | * | |
445 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
446 | * | |
447 | * - Don't call set_task_cpu() on a blocked task: | |
448 | * | |
449 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
450 | * the CPU assignment of blocked tasks isn't required to be valid. | |
451 | * | |
452 | * - for try_to_wake_up(), called under p->pi_lock: | |
453 | * | |
454 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
455 | * | |
456 | * - for migration called under rq->lock: | |
457 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
458 | * | |
459 | * o move_queued_task() | |
460 | * o detach_task() | |
461 | * | |
462 | * - for migration called under double_rq_lock(): | |
463 | * | |
464 | * o __migrate_swap_task() | |
465 | * o push_rt_task() / pull_rt_task() | |
466 | * o push_dl_task() / pull_dl_task() | |
467 | * o dl_task_offline_migration() | |
468 | * | |
469 | */ | |
470 | ||
39d371b7 PZ |
471 | void raw_spin_rq_lock_nested(struct rq *rq, int subclass) |
472 | { | |
d66f1b06 PZ |
473 | raw_spinlock_t *lock; |
474 | ||
9edeaea1 PZ |
475 | /* Matches synchronize_rcu() in __sched_core_enable() */ |
476 | preempt_disable(); | |
d66f1b06 PZ |
477 | if (sched_core_disabled()) { |
478 | raw_spin_lock_nested(&rq->__lock, subclass); | |
9edeaea1 PZ |
479 | /* preempt_count *MUST* be > 1 */ |
480 | preempt_enable_no_resched(); | |
d66f1b06 PZ |
481 | return; |
482 | } | |
483 | ||
484 | for (;;) { | |
9ef7e7e3 | 485 | lock = __rq_lockp(rq); |
d66f1b06 | 486 | raw_spin_lock_nested(lock, subclass); |
9ef7e7e3 | 487 | if (likely(lock == __rq_lockp(rq))) { |
9edeaea1 PZ |
488 | /* preempt_count *MUST* be > 1 */ |
489 | preempt_enable_no_resched(); | |
d66f1b06 | 490 | return; |
9edeaea1 | 491 | } |
d66f1b06 PZ |
492 | raw_spin_unlock(lock); |
493 | } | |
39d371b7 PZ |
494 | } |
495 | ||
496 | bool raw_spin_rq_trylock(struct rq *rq) | |
497 | { | |
d66f1b06 PZ |
498 | raw_spinlock_t *lock; |
499 | bool ret; | |
500 | ||
9edeaea1 PZ |
501 | /* Matches synchronize_rcu() in __sched_core_enable() */ |
502 | preempt_disable(); | |
503 | if (sched_core_disabled()) { | |
504 | ret = raw_spin_trylock(&rq->__lock); | |
505 | preempt_enable(); | |
506 | return ret; | |
507 | } | |
d66f1b06 PZ |
508 | |
509 | for (;;) { | |
9ef7e7e3 | 510 | lock = __rq_lockp(rq); |
d66f1b06 | 511 | ret = raw_spin_trylock(lock); |
9ef7e7e3 | 512 | if (!ret || (likely(lock == __rq_lockp(rq)))) { |
9edeaea1 | 513 | preempt_enable(); |
d66f1b06 | 514 | return ret; |
9edeaea1 | 515 | } |
d66f1b06 PZ |
516 | raw_spin_unlock(lock); |
517 | } | |
39d371b7 PZ |
518 | } |
519 | ||
520 | void raw_spin_rq_unlock(struct rq *rq) | |
521 | { | |
522 | raw_spin_unlock(rq_lockp(rq)); | |
523 | } | |
524 | ||
d66f1b06 PZ |
525 | #ifdef CONFIG_SMP |
526 | /* | |
527 | * double_rq_lock - safely lock two runqueues | |
528 | */ | |
529 | void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
530 | { | |
531 | lockdep_assert_irqs_disabled(); | |
532 | ||
533 | if (rq_order_less(rq2, rq1)) | |
534 | swap(rq1, rq2); | |
535 | ||
536 | raw_spin_rq_lock(rq1); | |
9ef7e7e3 | 537 | if (__rq_lockp(rq1) == __rq_lockp(rq2)) |
d66f1b06 PZ |
538 | return; |
539 | ||
540 | raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING); | |
541 | } | |
542 | #endif | |
543 | ||
3e71a462 PZ |
544 | /* |
545 | * __task_rq_lock - lock the rq @p resides on. | |
546 | */ | |
eb580751 | 547 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
548 | __acquires(rq->lock) |
549 | { | |
550 | struct rq *rq; | |
551 | ||
552 | lockdep_assert_held(&p->pi_lock); | |
553 | ||
554 | for (;;) { | |
555 | rq = task_rq(p); | |
5cb9eaa3 | 556 | raw_spin_rq_lock(rq); |
3e71a462 | 557 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
d8ac8971 | 558 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
559 | return rq; |
560 | } | |
5cb9eaa3 | 561 | raw_spin_rq_unlock(rq); |
3e71a462 PZ |
562 | |
563 | while (unlikely(task_on_rq_migrating(p))) | |
564 | cpu_relax(); | |
565 | } | |
566 | } | |
567 | ||
568 | /* | |
569 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
570 | */ | |
eb580751 | 571 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
572 | __acquires(p->pi_lock) |
573 | __acquires(rq->lock) | |
574 | { | |
575 | struct rq *rq; | |
576 | ||
577 | for (;;) { | |
eb580751 | 578 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 | 579 | rq = task_rq(p); |
5cb9eaa3 | 580 | raw_spin_rq_lock(rq); |
3e71a462 PZ |
581 | /* |
582 | * move_queued_task() task_rq_lock() | |
583 | * | |
584 | * ACQUIRE (rq->lock) | |
585 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
586 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
587 | * [S] ->cpu = new_cpu [L] task_rq() | |
588 | * [L] ->on_rq | |
589 | * RELEASE (rq->lock) | |
590 | * | |
c546951d | 591 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
592 | * the old rq->lock will fully serialize against the stores. |
593 | * | |
c546951d AP |
594 | * If we observe the new CPU in task_rq_lock(), the address |
595 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
596 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
597 | */ |
598 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 599 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
600 | return rq; |
601 | } | |
5cb9eaa3 | 602 | raw_spin_rq_unlock(rq); |
eb580751 | 603 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
604 | |
605 | while (unlikely(task_on_rq_migrating(p))) | |
606 | cpu_relax(); | |
607 | } | |
608 | } | |
609 | ||
535b9552 IM |
610 | /* |
611 | * RQ-clock updating methods: | |
612 | */ | |
613 | ||
614 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
615 | { | |
616 | /* | |
617 | * In theory, the compile should just see 0 here, and optimize out the call | |
618 | * to sched_rt_avg_update. But I don't trust it... | |
619 | */ | |
11d4afd4 VG |
620 | s64 __maybe_unused steal = 0, irq_delta = 0; |
621 | ||
535b9552 IM |
622 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
623 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
624 | ||
625 | /* | |
626 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
627 | * this case when a previous update_rq_clock() happened inside a | |
628 | * {soft,}irq region. | |
629 | * | |
630 | * When this happens, we stop ->clock_task and only update the | |
631 | * prev_irq_time stamp to account for the part that fit, so that a next | |
632 | * update will consume the rest. This ensures ->clock_task is | |
633 | * monotonic. | |
634 | * | |
635 | * It does however cause some slight miss-attribution of {soft,}irq | |
636 | * time, a more accurate solution would be to update the irq_time using | |
637 | * the current rq->clock timestamp, except that would require using | |
638 | * atomic ops. | |
639 | */ | |
640 | if (irq_delta > delta) | |
641 | irq_delta = delta; | |
642 | ||
643 | rq->prev_irq_time += irq_delta; | |
644 | delta -= irq_delta; | |
645 | #endif | |
646 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
647 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
648 | steal = paravirt_steal_clock(cpu_of(rq)); | |
649 | steal -= rq->prev_steal_time_rq; | |
650 | ||
651 | if (unlikely(steal > delta)) | |
652 | steal = delta; | |
653 | ||
654 | rq->prev_steal_time_rq += steal; | |
655 | delta -= steal; | |
656 | } | |
657 | #endif | |
658 | ||
659 | rq->clock_task += delta; | |
660 | ||
11d4afd4 | 661 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 662 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 663 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 664 | #endif |
23127296 | 665 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
666 | } |
667 | ||
668 | void update_rq_clock(struct rq *rq) | |
669 | { | |
670 | s64 delta; | |
671 | ||
5cb9eaa3 | 672 | lockdep_assert_rq_held(rq); |
535b9552 IM |
673 | |
674 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
675 | return; | |
676 | ||
677 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
678 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
679 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
680 | rq->clock_update_flags |= RQCF_UPDATED; |
681 | #endif | |
26ae58d2 | 682 | |
535b9552 IM |
683 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
684 | if (delta < 0) | |
685 | return; | |
686 | rq->clock += delta; | |
687 | update_rq_clock_task(rq, delta); | |
688 | } | |
689 | ||
8f4d37ec PZ |
690 | #ifdef CONFIG_SCHED_HRTICK |
691 | /* | |
692 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 693 | */ |
8f4d37ec | 694 | |
8f4d37ec PZ |
695 | static void hrtick_clear(struct rq *rq) |
696 | { | |
697 | if (hrtimer_active(&rq->hrtick_timer)) | |
698 | hrtimer_cancel(&rq->hrtick_timer); | |
699 | } | |
700 | ||
8f4d37ec PZ |
701 | /* |
702 | * High-resolution timer tick. | |
703 | * Runs from hardirq context with interrupts disabled. | |
704 | */ | |
705 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
706 | { | |
707 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 708 | struct rq_flags rf; |
8f4d37ec PZ |
709 | |
710 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
711 | ||
8a8c69c3 | 712 | rq_lock(rq, &rf); |
3e51f33f | 713 | update_rq_clock(rq); |
8f4d37ec | 714 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 715 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
716 | |
717 | return HRTIMER_NORESTART; | |
718 | } | |
719 | ||
95e904c7 | 720 | #ifdef CONFIG_SMP |
971ee28c | 721 | |
4961b6e1 | 722 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
723 | { |
724 | struct hrtimer *timer = &rq->hrtick_timer; | |
156ec6f4 | 725 | ktime_t time = rq->hrtick_time; |
971ee28c | 726 | |
156ec6f4 | 727 | hrtimer_start(timer, time, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
728 | } |
729 | ||
31656519 PZ |
730 | /* |
731 | * called from hardirq (IPI) context | |
732 | */ | |
733 | static void __hrtick_start(void *arg) | |
b328ca18 | 734 | { |
31656519 | 735 | struct rq *rq = arg; |
8a8c69c3 | 736 | struct rq_flags rf; |
b328ca18 | 737 | |
8a8c69c3 | 738 | rq_lock(rq, &rf); |
971ee28c | 739 | __hrtick_restart(rq); |
8a8c69c3 | 740 | rq_unlock(rq, &rf); |
b328ca18 PZ |
741 | } |
742 | ||
31656519 PZ |
743 | /* |
744 | * Called to set the hrtick timer state. | |
745 | * | |
746 | * called with rq->lock held and irqs disabled | |
747 | */ | |
029632fb | 748 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 749 | { |
31656519 | 750 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 751 | s64 delta; |
752 | ||
753 | /* | |
754 | * Don't schedule slices shorter than 10000ns, that just | |
755 | * doesn't make sense and can cause timer DoS. | |
756 | */ | |
757 | delta = max_t(s64, delay, 10000LL); | |
156ec6f4 | 758 | rq->hrtick_time = ktime_add_ns(timer->base->get_time(), delta); |
31656519 | 759 | |
fd3eafda | 760 | if (rq == this_rq()) |
971ee28c | 761 | __hrtick_restart(rq); |
fd3eafda | 762 | else |
c46fff2a | 763 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
764 | } |
765 | ||
31656519 PZ |
766 | #else |
767 | /* | |
768 | * Called to set the hrtick timer state. | |
769 | * | |
770 | * called with rq->lock held and irqs disabled | |
771 | */ | |
029632fb | 772 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 773 | { |
86893335 WL |
774 | /* |
775 | * Don't schedule slices shorter than 10000ns, that just | |
776 | * doesn't make sense. Rely on vruntime for fairness. | |
777 | */ | |
778 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 779 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 780 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 781 | } |
90b5363a | 782 | |
31656519 | 783 | #endif /* CONFIG_SMP */ |
8f4d37ec | 784 | |
77a021be | 785 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 786 | { |
31656519 | 787 | #ifdef CONFIG_SMP |
545b8c8d | 788 | INIT_CSD(&rq->hrtick_csd, __hrtick_start, rq); |
31656519 | 789 | #endif |
d5096aa6 | 790 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 791 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 792 | } |
006c75f1 | 793 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
794 | static inline void hrtick_clear(struct rq *rq) |
795 | { | |
796 | } | |
797 | ||
77a021be | 798 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
799 | { |
800 | } | |
006c75f1 | 801 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 802 | |
5529578a FW |
803 | /* |
804 | * cmpxchg based fetch_or, macro so it works for different integer types | |
805 | */ | |
806 | #define fetch_or(ptr, mask) \ | |
807 | ({ \ | |
808 | typeof(ptr) _ptr = (ptr); \ | |
809 | typeof(mask) _mask = (mask); \ | |
810 | typeof(*_ptr) _old, _val = *_ptr; \ | |
811 | \ | |
812 | for (;;) { \ | |
813 | _old = cmpxchg(_ptr, _val, _val | _mask); \ | |
814 | if (_old == _val) \ | |
815 | break; \ | |
816 | _val = _old; \ | |
817 | } \ | |
818 | _old; \ | |
819 | }) | |
820 | ||
e3baac47 | 821 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
822 | /* |
823 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
824 | * this avoids any races wrt polling state changes and thereby avoids | |
825 | * spurious IPIs. | |
826 | */ | |
827 | static bool set_nr_and_not_polling(struct task_struct *p) | |
828 | { | |
829 | struct thread_info *ti = task_thread_info(p); | |
830 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
831 | } | |
e3baac47 PZ |
832 | |
833 | /* | |
834 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
835 | * | |
836 | * If this returns true, then the idle task promises to call | |
837 | * sched_ttwu_pending() and reschedule soon. | |
838 | */ | |
839 | static bool set_nr_if_polling(struct task_struct *p) | |
840 | { | |
841 | struct thread_info *ti = task_thread_info(p); | |
316c1608 | 842 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
e3baac47 PZ |
843 | |
844 | for (;;) { | |
845 | if (!(val & _TIF_POLLING_NRFLAG)) | |
846 | return false; | |
847 | if (val & _TIF_NEED_RESCHED) | |
848 | return true; | |
849 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); | |
850 | if (old == val) | |
851 | break; | |
852 | val = old; | |
853 | } | |
854 | return true; | |
855 | } | |
856 | ||
fd99f91a PZ |
857 | #else |
858 | static bool set_nr_and_not_polling(struct task_struct *p) | |
859 | { | |
860 | set_tsk_need_resched(p); | |
861 | return true; | |
862 | } | |
e3baac47 PZ |
863 | |
864 | #ifdef CONFIG_SMP | |
865 | static bool set_nr_if_polling(struct task_struct *p) | |
866 | { | |
867 | return false; | |
868 | } | |
869 | #endif | |
fd99f91a PZ |
870 | #endif |
871 | ||
07879c6a | 872 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
873 | { |
874 | struct wake_q_node *node = &task->wake_q; | |
875 | ||
876 | /* | |
877 | * Atomically grab the task, if ->wake_q is !nil already it means | |
b19a888c | 878 | * it's already queued (either by us or someone else) and will get the |
76751049 PZ |
879 | * wakeup due to that. |
880 | * | |
4c4e3731 PZ |
881 | * In order to ensure that a pending wakeup will observe our pending |
882 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 883 | */ |
4c4e3731 | 884 | smp_mb__before_atomic(); |
87ff19cb | 885 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 886 | return false; |
76751049 PZ |
887 | |
888 | /* | |
889 | * The head is context local, there can be no concurrency. | |
890 | */ | |
891 | *head->lastp = node; | |
892 | head->lastp = &node->next; | |
07879c6a DB |
893 | return true; |
894 | } | |
895 | ||
896 | /** | |
897 | * wake_q_add() - queue a wakeup for 'later' waking. | |
898 | * @head: the wake_q_head to add @task to | |
899 | * @task: the task to queue for 'later' wakeup | |
900 | * | |
901 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
902 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
903 | * instantly. | |
904 | * | |
905 | * This function must be used as-if it were wake_up_process(); IOW the task | |
906 | * must be ready to be woken at this location. | |
907 | */ | |
908 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
909 | { | |
910 | if (__wake_q_add(head, task)) | |
911 | get_task_struct(task); | |
912 | } | |
913 | ||
914 | /** | |
915 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
916 | * @head: the wake_q_head to add @task to | |
917 | * @task: the task to queue for 'later' wakeup | |
918 | * | |
919 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
920 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
921 | * instantly. | |
922 | * | |
923 | * This function must be used as-if it were wake_up_process(); IOW the task | |
924 | * must be ready to be woken at this location. | |
925 | * | |
926 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
927 | * that already hold reference to @task can call the 'safe' version and trust | |
928 | * wake_q to do the right thing depending whether or not the @task is already | |
929 | * queued for wakeup. | |
930 | */ | |
931 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
932 | { | |
933 | if (!__wake_q_add(head, task)) | |
934 | put_task_struct(task); | |
76751049 PZ |
935 | } |
936 | ||
937 | void wake_up_q(struct wake_q_head *head) | |
938 | { | |
939 | struct wake_q_node *node = head->first; | |
940 | ||
941 | while (node != WAKE_Q_TAIL) { | |
942 | struct task_struct *task; | |
943 | ||
944 | task = container_of(node, struct task_struct, wake_q); | |
d1ccc66d | 945 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
946 | node = node->next; |
947 | task->wake_q.next = NULL; | |
948 | ||
949 | /* | |
7696f991 AP |
950 | * wake_up_process() executes a full barrier, which pairs with |
951 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
952 | */ |
953 | wake_up_process(task); | |
954 | put_task_struct(task); | |
955 | } | |
956 | } | |
957 | ||
c24d20db | 958 | /* |
8875125e | 959 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
960 | * |
961 | * On UP this means the setting of the need_resched flag, on SMP it | |
962 | * might also involve a cross-CPU call to trigger the scheduler on | |
963 | * the target CPU. | |
964 | */ | |
8875125e | 965 | void resched_curr(struct rq *rq) |
c24d20db | 966 | { |
8875125e | 967 | struct task_struct *curr = rq->curr; |
c24d20db IM |
968 | int cpu; |
969 | ||
5cb9eaa3 | 970 | lockdep_assert_rq_held(rq); |
c24d20db | 971 | |
8875125e | 972 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
973 | return; |
974 | ||
8875125e | 975 | cpu = cpu_of(rq); |
fd99f91a | 976 | |
f27dde8d | 977 | if (cpu == smp_processor_id()) { |
8875125e | 978 | set_tsk_need_resched(curr); |
f27dde8d | 979 | set_preempt_need_resched(); |
c24d20db | 980 | return; |
f27dde8d | 981 | } |
c24d20db | 982 | |
8875125e | 983 | if (set_nr_and_not_polling(curr)) |
c24d20db | 984 | smp_send_reschedule(cpu); |
dfc68f29 AL |
985 | else |
986 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
987 | } |
988 | ||
029632fb | 989 | void resched_cpu(int cpu) |
c24d20db IM |
990 | { |
991 | struct rq *rq = cpu_rq(cpu); | |
992 | unsigned long flags; | |
993 | ||
5cb9eaa3 | 994 | raw_spin_rq_lock_irqsave(rq, flags); |
a0982dfa PM |
995 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
996 | resched_curr(rq); | |
5cb9eaa3 | 997 | raw_spin_rq_unlock_irqrestore(rq, flags); |
c24d20db | 998 | } |
06d8308c | 999 | |
b021fe3e | 1000 | #ifdef CONFIG_SMP |
3451d024 | 1001 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 1002 | /* |
d1ccc66d IM |
1003 | * In the semi idle case, use the nearest busy CPU for migrating timers |
1004 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
1005 | * |
1006 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
1007 | * selecting an idle CPU will add more delays to the timers than intended |
1008 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 1009 | */ |
bc7a34b8 | 1010 | int get_nohz_timer_target(void) |
83cd4fe2 | 1011 | { |
e938b9c9 | 1012 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 | 1013 | struct sched_domain *sd; |
031e3bd8 | 1014 | const struct cpumask *hk_mask; |
83cd4fe2 | 1015 | |
e938b9c9 WL |
1016 | if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) { |
1017 | if (!idle_cpu(cpu)) | |
1018 | return cpu; | |
1019 | default_cpu = cpu; | |
1020 | } | |
6201b4d6 | 1021 | |
031e3bd8 YZ |
1022 | hk_mask = housekeeping_cpumask(HK_FLAG_TIMER); |
1023 | ||
057f3fad | 1024 | rcu_read_lock(); |
83cd4fe2 | 1025 | for_each_domain(cpu, sd) { |
031e3bd8 | 1026 | for_each_cpu_and(i, sched_domain_span(sd), hk_mask) { |
44496922 WL |
1027 | if (cpu == i) |
1028 | continue; | |
1029 | ||
e938b9c9 | 1030 | if (!idle_cpu(i)) { |
057f3fad PZ |
1031 | cpu = i; |
1032 | goto unlock; | |
1033 | } | |
1034 | } | |
83cd4fe2 | 1035 | } |
9642d18e | 1036 | |
e938b9c9 WL |
1037 | if (default_cpu == -1) |
1038 | default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER); | |
1039 | cpu = default_cpu; | |
057f3fad PZ |
1040 | unlock: |
1041 | rcu_read_unlock(); | |
83cd4fe2 VP |
1042 | return cpu; |
1043 | } | |
d1ccc66d | 1044 | |
06d8308c TG |
1045 | /* |
1046 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1047 | * idle CPU then this timer might expire before the next timer event | |
1048 | * which is scheduled to wake up that CPU. In case of a completely | |
1049 | * idle system the next event might even be infinite time into the | |
1050 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1051 | * leaves the inner idle loop so the newly added timer is taken into | |
1052 | * account when the CPU goes back to idle and evaluates the timer | |
1053 | * wheel for the next timer event. | |
1054 | */ | |
1c20091e | 1055 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
1056 | { |
1057 | struct rq *rq = cpu_rq(cpu); | |
1058 | ||
1059 | if (cpu == smp_processor_id()) | |
1060 | return; | |
1061 | ||
67b9ca70 | 1062 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 1063 | smp_send_reschedule(cpu); |
dfc68f29 AL |
1064 | else |
1065 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
1066 | } |
1067 | ||
c5bfece2 | 1068 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 1069 | { |
53c5fa16 FW |
1070 | /* |
1071 | * We just need the target to call irq_exit() and re-evaluate | |
1072 | * the next tick. The nohz full kick at least implies that. | |
1073 | * If needed we can still optimize that later with an | |
1074 | * empty IRQ. | |
1075 | */ | |
379d9ecb PM |
1076 | if (cpu_is_offline(cpu)) |
1077 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 1078 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
1079 | if (cpu != smp_processor_id() || |
1080 | tick_nohz_tick_stopped()) | |
53c5fa16 | 1081 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
1082 | return true; |
1083 | } | |
1084 | ||
1085 | return false; | |
1086 | } | |
1087 | ||
379d9ecb PM |
1088 | /* |
1089 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
1090 | * caller's responsibility to deal with the lost wakeup, for example, | |
1091 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
1092 | */ | |
1c20091e FW |
1093 | void wake_up_nohz_cpu(int cpu) |
1094 | { | |
c5bfece2 | 1095 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
1096 | wake_up_idle_cpu(cpu); |
1097 | } | |
1098 | ||
19a1f5ec | 1099 | static void nohz_csd_func(void *info) |
45bf76df | 1100 | { |
19a1f5ec PZ |
1101 | struct rq *rq = info; |
1102 | int cpu = cpu_of(rq); | |
1103 | unsigned int flags; | |
873b4c65 VG |
1104 | |
1105 | /* | |
19a1f5ec | 1106 | * Release the rq::nohz_csd. |
873b4c65 | 1107 | */ |
c6f88654 | 1108 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK | NOHZ_NEWILB_KICK, nohz_flags(cpu)); |
19a1f5ec | 1109 | WARN_ON(!(flags & NOHZ_KICK_MASK)); |
45bf76df | 1110 | |
19a1f5ec PZ |
1111 | rq->idle_balance = idle_cpu(cpu); |
1112 | if (rq->idle_balance && !need_resched()) { | |
1113 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
1114 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
1115 | } | |
2069dd75 PZ |
1116 | } |
1117 | ||
3451d024 | 1118 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 1119 | |
ce831b38 | 1120 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 1121 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 1122 | { |
76d92ac3 FW |
1123 | int fifo_nr_running; |
1124 | ||
1125 | /* Deadline tasks, even if single, need the tick */ | |
1126 | if (rq->dl.dl_nr_running) | |
1127 | return false; | |
1128 | ||
1e78cdbd | 1129 | /* |
b19a888c | 1130 | * If there are more than one RR tasks, we need the tick to affect the |
2548d546 | 1131 | * actual RR behaviour. |
1e78cdbd | 1132 | */ |
76d92ac3 FW |
1133 | if (rq->rt.rr_nr_running) { |
1134 | if (rq->rt.rr_nr_running == 1) | |
1135 | return true; | |
1136 | else | |
1137 | return false; | |
1e78cdbd RR |
1138 | } |
1139 | ||
2548d546 PZ |
1140 | /* |
1141 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
1142 | * forced preemption between FIFO tasks. | |
1143 | */ | |
1144 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
1145 | if (fifo_nr_running) | |
1146 | return true; | |
1147 | ||
1148 | /* | |
1149 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
1150 | * if there's more than one we need the tick for involuntary | |
1151 | * preemption. | |
1152 | */ | |
1153 | if (rq->nr_running > 1) | |
541b8264 | 1154 | return false; |
ce831b38 | 1155 | |
541b8264 | 1156 | return true; |
ce831b38 FW |
1157 | } |
1158 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 1159 | #endif /* CONFIG_SMP */ |
18d95a28 | 1160 | |
a790de99 PT |
1161 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
1162 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 1163 | /* |
8277434e PT |
1164 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
1165 | * node and @up when leaving it for the final time. | |
1166 | * | |
1167 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 1168 | */ |
029632fb | 1169 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 1170 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1171 | { |
1172 | struct task_group *parent, *child; | |
eb755805 | 1173 | int ret; |
c09595f6 | 1174 | |
8277434e PT |
1175 | parent = from; |
1176 | ||
c09595f6 | 1177 | down: |
eb755805 PZ |
1178 | ret = (*down)(parent, data); |
1179 | if (ret) | |
8277434e | 1180 | goto out; |
c09595f6 PZ |
1181 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1182 | parent = child; | |
1183 | goto down; | |
1184 | ||
1185 | up: | |
1186 | continue; | |
1187 | } | |
eb755805 | 1188 | ret = (*up)(parent, data); |
8277434e PT |
1189 | if (ret || parent == from) |
1190 | goto out; | |
c09595f6 PZ |
1191 | |
1192 | child = parent; | |
1193 | parent = parent->parent; | |
1194 | if (parent) | |
1195 | goto up; | |
8277434e | 1196 | out: |
eb755805 | 1197 | return ret; |
c09595f6 PZ |
1198 | } |
1199 | ||
029632fb | 1200 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 1201 | { |
e2b245f8 | 1202 | return 0; |
eb755805 | 1203 | } |
18d95a28 PZ |
1204 | #endif |
1205 | ||
9059393e | 1206 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 1207 | { |
f05998d4 NR |
1208 | int prio = p->static_prio - MAX_RT_PRIO; |
1209 | struct load_weight *load = &p->se.load; | |
1210 | ||
dd41f596 IM |
1211 | /* |
1212 | * SCHED_IDLE tasks get minimal weight: | |
1213 | */ | |
1da1843f | 1214 | if (task_has_idle_policy(p)) { |
c8b28116 | 1215 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 1216 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
1217 | return; |
1218 | } | |
71f8bd46 | 1219 | |
9059393e VG |
1220 | /* |
1221 | * SCHED_OTHER tasks have to update their load when changing their | |
1222 | * weight | |
1223 | */ | |
1224 | if (update_load && p->sched_class == &fair_sched_class) { | |
1225 | reweight_task(p, prio); | |
1226 | } else { | |
1227 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
1228 | load->inv_weight = sched_prio_to_wmult[prio]; | |
1229 | } | |
71f8bd46 IM |
1230 | } |
1231 | ||
69842cba | 1232 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
1233 | /* |
1234 | * Serializes updates of utilization clamp values | |
1235 | * | |
1236 | * The (slow-path) user-space triggers utilization clamp value updates which | |
1237 | * can require updates on (fast-path) scheduler's data structures used to | |
1238 | * support enqueue/dequeue operations. | |
1239 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
1240 | * requests are serialized using a mutex to reduce the risk of conflicting | |
1241 | * updates or API abuses. | |
1242 | */ | |
1243 | static DEFINE_MUTEX(uclamp_mutex); | |
1244 | ||
e8f14172 PB |
1245 | /* Max allowed minimum utilization */ |
1246 | unsigned int sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; | |
1247 | ||
1248 | /* Max allowed maximum utilization */ | |
1249 | unsigned int sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; | |
1250 | ||
13685c4a QY |
1251 | /* |
1252 | * By default RT tasks run at the maximum performance point/capacity of the | |
1253 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
1254 | * SCHED_CAPACITY_SCALE. | |
1255 | * | |
1256 | * This knob allows admins to change the default behavior when uclamp is being | |
1257 | * used. In battery powered devices, particularly, running at the maximum | |
1258 | * capacity and frequency will increase energy consumption and shorten the | |
1259 | * battery life. | |
1260 | * | |
1261 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
1262 | * | |
1263 | * This knob will not override the system default sched_util_clamp_min defined | |
1264 | * above. | |
1265 | */ | |
1266 | unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; | |
1267 | ||
e8f14172 PB |
1268 | /* All clamps are required to be less or equal than these values */ |
1269 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 1270 | |
46609ce2 QY |
1271 | /* |
1272 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
1273 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
1274 | * enqueue/dequeue_task(). | |
1275 | * | |
1276 | * This allows users to continue to enable uclamp in their kernel config with | |
1277 | * minimum uclamp overhead in the fast path. | |
1278 | * | |
1279 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
1280 | * enabled, since we have an actual users that make use of uclamp | |
1281 | * functionality. | |
1282 | * | |
1283 | * The knobs that would enable this static key are: | |
1284 | * | |
1285 | * * A task modifying its uclamp value with sched_setattr(). | |
1286 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
1287 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
1288 | */ | |
1289 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
1290 | ||
69842cba PB |
1291 | /* Integer rounded range for each bucket */ |
1292 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
1293 | ||
1294 | #define for_each_clamp_id(clamp_id) \ | |
1295 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
1296 | ||
1297 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
1298 | { | |
6d2f8909 | 1299 | return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1); |
69842cba PB |
1300 | } |
1301 | ||
7763baac | 1302 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
1303 | { |
1304 | if (clamp_id == UCLAMP_MIN) | |
1305 | return 0; | |
1306 | return SCHED_CAPACITY_SCALE; | |
1307 | } | |
1308 | ||
a509a7cd PB |
1309 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
1310 | unsigned int value, bool user_defined) | |
69842cba PB |
1311 | { |
1312 | uc_se->value = value; | |
1313 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 1314 | uc_se->user_defined = user_defined; |
69842cba PB |
1315 | } |
1316 | ||
e496187d | 1317 | static inline unsigned int |
0413d7f3 | 1318 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
1319 | unsigned int clamp_value) |
1320 | { | |
1321 | /* | |
1322 | * Avoid blocked utilization pushing up the frequency when we go | |
1323 | * idle (which drops the max-clamp) by retaining the last known | |
1324 | * max-clamp. | |
1325 | */ | |
1326 | if (clamp_id == UCLAMP_MAX) { | |
1327 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
1328 | return clamp_value; | |
1329 | } | |
1330 | ||
1331 | return uclamp_none(UCLAMP_MIN); | |
1332 | } | |
1333 | ||
0413d7f3 | 1334 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
1335 | unsigned int clamp_value) |
1336 | { | |
1337 | /* Reset max-clamp retention only on idle exit */ | |
1338 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
1339 | return; | |
1340 | ||
1341 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
1342 | } | |
1343 | ||
69842cba | 1344 | static inline |
7763baac | 1345 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 1346 | unsigned int clamp_value) |
69842cba PB |
1347 | { |
1348 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
1349 | int bucket_id = UCLAMP_BUCKETS - 1; | |
1350 | ||
1351 | /* | |
1352 | * Since both min and max clamps are max aggregated, find the | |
1353 | * top most bucket with tasks in. | |
1354 | */ | |
1355 | for ( ; bucket_id >= 0; bucket_id--) { | |
1356 | if (!bucket[bucket_id].tasks) | |
1357 | continue; | |
1358 | return bucket[bucket_id].value; | |
1359 | } | |
1360 | ||
1361 | /* No tasks -- default clamp values */ | |
e496187d | 1362 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
1363 | } |
1364 | ||
13685c4a QY |
1365 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
1366 | { | |
1367 | unsigned int default_util_min; | |
1368 | struct uclamp_se *uc_se; | |
1369 | ||
1370 | lockdep_assert_held(&p->pi_lock); | |
1371 | ||
1372 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1373 | ||
1374 | /* Only sync if user didn't override the default */ | |
1375 | if (uc_se->user_defined) | |
1376 | return; | |
1377 | ||
1378 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1379 | uclamp_se_set(uc_se, default_util_min, false); | |
1380 | } | |
1381 | ||
1382 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1383 | { | |
1384 | struct rq_flags rf; | |
1385 | struct rq *rq; | |
1386 | ||
1387 | if (!rt_task(p)) | |
1388 | return; | |
1389 | ||
1390 | /* Protect updates to p->uclamp_* */ | |
1391 | rq = task_rq_lock(p, &rf); | |
1392 | __uclamp_update_util_min_rt_default(p); | |
1393 | task_rq_unlock(rq, p, &rf); | |
1394 | } | |
1395 | ||
1396 | static void uclamp_sync_util_min_rt_default(void) | |
1397 | { | |
1398 | struct task_struct *g, *p; | |
1399 | ||
1400 | /* | |
1401 | * copy_process() sysctl_uclamp | |
1402 | * uclamp_min_rt = X; | |
1403 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1404 | * // link thread smp_mb__after_spinlock() | |
1405 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1406 | * sched_post_fork() for_each_process_thread() | |
1407 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1408 | * | |
1409 | * Ensures that either sched_post_fork() will observe the new | |
1410 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1411 | * task. | |
1412 | */ | |
1413 | read_lock(&tasklist_lock); | |
1414 | smp_mb__after_spinlock(); | |
1415 | read_unlock(&tasklist_lock); | |
1416 | ||
1417 | rcu_read_lock(); | |
1418 | for_each_process_thread(g, p) | |
1419 | uclamp_update_util_min_rt_default(p); | |
1420 | rcu_read_unlock(); | |
1421 | } | |
1422 | ||
3eac870a | 1423 | static inline struct uclamp_se |
0413d7f3 | 1424 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a | 1425 | { |
0213b708 | 1426 | /* Copy by value as we could modify it */ |
3eac870a PB |
1427 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; |
1428 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0213b708 | 1429 | unsigned int tg_min, tg_max, value; |
3eac870a PB |
1430 | |
1431 | /* | |
1432 | * Tasks in autogroups or root task group will be | |
1433 | * restricted by system defaults. | |
1434 | */ | |
1435 | if (task_group_is_autogroup(task_group(p))) | |
1436 | return uc_req; | |
1437 | if (task_group(p) == &root_task_group) | |
1438 | return uc_req; | |
1439 | ||
0213b708 QY |
1440 | tg_min = task_group(p)->uclamp[UCLAMP_MIN].value; |
1441 | tg_max = task_group(p)->uclamp[UCLAMP_MAX].value; | |
1442 | value = uc_req.value; | |
1443 | value = clamp(value, tg_min, tg_max); | |
1444 | uclamp_se_set(&uc_req, value, false); | |
3eac870a PB |
1445 | #endif |
1446 | ||
1447 | return uc_req; | |
1448 | } | |
1449 | ||
e8f14172 PB |
1450 | /* |
1451 | * The effective clamp bucket index of a task depends on, by increasing | |
1452 | * priority: | |
1453 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1454 | * - the task group effective clamp value, for tasks not either in the root |
1455 | * group or in an autogroup | |
e8f14172 PB |
1456 | * - the system default clamp value, defined by the sysadmin |
1457 | */ | |
1458 | static inline struct uclamp_se | |
0413d7f3 | 1459 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1460 | { |
3eac870a | 1461 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1462 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1463 | ||
1464 | /* System default restrictions always apply */ | |
1465 | if (unlikely(uc_req.value > uc_max.value)) | |
1466 | return uc_max; | |
1467 | ||
1468 | return uc_req; | |
1469 | } | |
1470 | ||
686516b5 | 1471 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1472 | { |
1473 | struct uclamp_se uc_eff; | |
1474 | ||
1475 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1476 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1477 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1478 | |
1479 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1480 | ||
686516b5 | 1481 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1482 | } |
1483 | ||
69842cba PB |
1484 | /* |
1485 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1486 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1487 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1488 | * |
1489 | * Tasks can have a task-specific value requested from user-space, track | |
1490 | * within each bucket the maximum value for tasks refcounted in it. | |
1491 | * This "local max aggregation" allows to track the exact "requested" value | |
1492 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1493 | */ |
1494 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1495 | enum uclamp_id clamp_id) |
69842cba PB |
1496 | { |
1497 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1498 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1499 | struct uclamp_bucket *bucket; | |
1500 | ||
5cb9eaa3 | 1501 | lockdep_assert_rq_held(rq); |
69842cba | 1502 | |
e8f14172 PB |
1503 | /* Update task effective clamp */ |
1504 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1505 | ||
69842cba PB |
1506 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1507 | bucket->tasks++; | |
e8f14172 | 1508 | uc_se->active = true; |
69842cba | 1509 | |
e496187d PB |
1510 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1511 | ||
60daf9c1 PB |
1512 | /* |
1513 | * Local max aggregation: rq buckets always track the max | |
1514 | * "requested" clamp value of its RUNNABLE tasks. | |
1515 | */ | |
1516 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1517 | bucket->value = uc_se->value; | |
1518 | ||
69842cba | 1519 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1520 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1521 | } |
1522 | ||
1523 | /* | |
1524 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1525 | * is released. If this is the last task reference counting the rq's max | |
1526 | * active clamp value, then the rq's clamp value is updated. | |
1527 | * | |
1528 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1529 | * always valid. If it's detected they are not, as defensive programming, | |
1530 | * enforce the expected state and warn. | |
1531 | */ | |
1532 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1533 | enum uclamp_id clamp_id) |
69842cba PB |
1534 | { |
1535 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1536 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1537 | struct uclamp_bucket *bucket; | |
e496187d | 1538 | unsigned int bkt_clamp; |
69842cba PB |
1539 | unsigned int rq_clamp; |
1540 | ||
5cb9eaa3 | 1541 | lockdep_assert_rq_held(rq); |
69842cba | 1542 | |
46609ce2 QY |
1543 | /* |
1544 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1545 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1546 | * | |
1547 | * In this case the uc_se->active flag should be false since no uclamp | |
1548 | * accounting was performed at enqueue time and we can just return | |
1549 | * here. | |
1550 | * | |
b19a888c | 1551 | * Need to be careful of the following enqueue/dequeue ordering |
46609ce2 QY |
1552 | * problem too |
1553 | * | |
1554 | * enqueue(taskA) | |
1555 | * // sched_uclamp_used gets enabled | |
1556 | * enqueue(taskB) | |
1557 | * dequeue(taskA) | |
b19a888c | 1558 | * // Must not decrement bucket->tasks here |
46609ce2 QY |
1559 | * dequeue(taskB) |
1560 | * | |
1561 | * where we could end up with stale data in uc_se and | |
1562 | * bucket[uc_se->bucket_id]. | |
1563 | * | |
1564 | * The following check here eliminates the possibility of such race. | |
1565 | */ | |
1566 | if (unlikely(!uc_se->active)) | |
1567 | return; | |
1568 | ||
69842cba | 1569 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1570 | |
69842cba PB |
1571 | SCHED_WARN_ON(!bucket->tasks); |
1572 | if (likely(bucket->tasks)) | |
1573 | bucket->tasks--; | |
46609ce2 | 1574 | |
e8f14172 | 1575 | uc_se->active = false; |
69842cba | 1576 | |
60daf9c1 PB |
1577 | /* |
1578 | * Keep "local max aggregation" simple and accept to (possibly) | |
1579 | * overboost some RUNNABLE tasks in the same bucket. | |
1580 | * The rq clamp bucket value is reset to its base value whenever | |
1581 | * there are no more RUNNABLE tasks refcounting it. | |
1582 | */ | |
69842cba PB |
1583 | if (likely(bucket->tasks)) |
1584 | return; | |
1585 | ||
1586 | rq_clamp = READ_ONCE(uc_rq->value); | |
1587 | /* | |
1588 | * Defensive programming: this should never happen. If it happens, | |
1589 | * e.g. due to future modification, warn and fixup the expected value. | |
1590 | */ | |
1591 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1592 | if (bucket->value >= rq_clamp) { |
1593 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1594 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1595 | } | |
69842cba PB |
1596 | } |
1597 | ||
1598 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1599 | { | |
0413d7f3 | 1600 | enum uclamp_id clamp_id; |
69842cba | 1601 | |
46609ce2 QY |
1602 | /* |
1603 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1604 | * | |
1605 | * The condition is constructed such that a NOP is generated when | |
1606 | * sched_uclamp_used is disabled. | |
1607 | */ | |
1608 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1609 | return; | |
1610 | ||
69842cba PB |
1611 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1612 | return; | |
1613 | ||
1614 | for_each_clamp_id(clamp_id) | |
1615 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1616 | |
1617 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1618 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1619 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1620 | } |
1621 | ||
1622 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1623 | { | |
0413d7f3 | 1624 | enum uclamp_id clamp_id; |
69842cba | 1625 | |
46609ce2 QY |
1626 | /* |
1627 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1628 | * | |
1629 | * The condition is constructed such that a NOP is generated when | |
1630 | * sched_uclamp_used is disabled. | |
1631 | */ | |
1632 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1633 | return; | |
1634 | ||
69842cba PB |
1635 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1636 | return; | |
1637 | ||
1638 | for_each_clamp_id(clamp_id) | |
1639 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1640 | } | |
1641 | ||
ca4984a7 QP |
1642 | static inline void uclamp_rq_reinc_id(struct rq *rq, struct task_struct *p, |
1643 | enum uclamp_id clamp_id) | |
1644 | { | |
1645 | if (!p->uclamp[clamp_id].active) | |
1646 | return; | |
1647 | ||
1648 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1649 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1650 | ||
1651 | /* | |
1652 | * Make sure to clear the idle flag if we've transiently reached 0 | |
1653 | * active tasks on rq. | |
1654 | */ | |
1655 | if (clamp_id == UCLAMP_MAX && (rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
1656 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
1657 | } | |
1658 | ||
babbe170 | 1659 | static inline void |
0213b708 | 1660 | uclamp_update_active(struct task_struct *p) |
babbe170 | 1661 | { |
0213b708 | 1662 | enum uclamp_id clamp_id; |
babbe170 PB |
1663 | struct rq_flags rf; |
1664 | struct rq *rq; | |
1665 | ||
1666 | /* | |
1667 | * Lock the task and the rq where the task is (or was) queued. | |
1668 | * | |
1669 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1670 | * price to pay to safely serialize util_{min,max} updates with | |
1671 | * enqueues, dequeues and migration operations. | |
1672 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1673 | */ | |
1674 | rq = task_rq_lock(p, &rf); | |
1675 | ||
1676 | /* | |
1677 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1678 | * If the task is not yet RUNNABLE and its task_struct is not | |
1679 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1680 | * it will already see the updated clamp bucket value. | |
1681 | */ | |
ca4984a7 QP |
1682 | for_each_clamp_id(clamp_id) |
1683 | uclamp_rq_reinc_id(rq, p, clamp_id); | |
babbe170 PB |
1684 | |
1685 | task_rq_unlock(rq, p, &rf); | |
1686 | } | |
1687 | ||
e3b8b6a0 | 1688 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 | 1689 | static inline void |
0213b708 | 1690 | uclamp_update_active_tasks(struct cgroup_subsys_state *css) |
babbe170 PB |
1691 | { |
1692 | struct css_task_iter it; | |
1693 | struct task_struct *p; | |
babbe170 PB |
1694 | |
1695 | css_task_iter_start(css, 0, &it); | |
0213b708 QY |
1696 | while ((p = css_task_iter_next(&it))) |
1697 | uclamp_update_active(p); | |
babbe170 PB |
1698 | css_task_iter_end(&it); |
1699 | } | |
1700 | ||
7274a5c1 PB |
1701 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
1702 | static void uclamp_update_root_tg(void) | |
1703 | { | |
1704 | struct task_group *tg = &root_task_group; | |
1705 | ||
1706 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1707 | sysctl_sched_uclamp_util_min, false); | |
1708 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1709 | sysctl_sched_uclamp_util_max, false); | |
1710 | ||
1711 | rcu_read_lock(); | |
1712 | cpu_util_update_eff(&root_task_group.css); | |
1713 | rcu_read_unlock(); | |
1714 | } | |
1715 | #else | |
1716 | static void uclamp_update_root_tg(void) { } | |
1717 | #endif | |
1718 | ||
e8f14172 | 1719 | int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1720 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1721 | { |
7274a5c1 | 1722 | bool update_root_tg = false; |
13685c4a | 1723 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1724 | int result; |
1725 | ||
2480c093 | 1726 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1727 | old_min = sysctl_sched_uclamp_util_min; |
1728 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1729 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1730 | |
1731 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1732 | if (result) | |
1733 | goto undo; | |
1734 | if (!write) | |
1735 | goto done; | |
1736 | ||
1737 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1738 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1739 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1740 | ||
e8f14172 PB |
1741 | result = -EINVAL; |
1742 | goto undo; | |
1743 | } | |
1744 | ||
1745 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1746 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1747 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1748 | update_root_tg = true; |
e8f14172 PB |
1749 | } |
1750 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1751 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1752 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1753 | update_root_tg = true; |
e8f14172 PB |
1754 | } |
1755 | ||
46609ce2 QY |
1756 | if (update_root_tg) { |
1757 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1758 | uclamp_update_root_tg(); |
46609ce2 | 1759 | } |
7274a5c1 | 1760 | |
13685c4a QY |
1761 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1762 | static_branch_enable(&sched_uclamp_used); | |
1763 | uclamp_sync_util_min_rt_default(); | |
1764 | } | |
7274a5c1 | 1765 | |
e8f14172 | 1766 | /* |
7274a5c1 PB |
1767 | * We update all RUNNABLE tasks only when task groups are in use. |
1768 | * Otherwise, keep it simple and do just a lazy update at each next | |
1769 | * task enqueue time. | |
e8f14172 | 1770 | */ |
7274a5c1 | 1771 | |
e8f14172 PB |
1772 | goto done; |
1773 | ||
1774 | undo: | |
1775 | sysctl_sched_uclamp_util_min = old_min; | |
1776 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1777 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1778 | done: |
2480c093 | 1779 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1780 | |
1781 | return result; | |
1782 | } | |
1783 | ||
a509a7cd PB |
1784 | static int uclamp_validate(struct task_struct *p, |
1785 | const struct sched_attr *attr) | |
1786 | { | |
480a6ca2 DE |
1787 | int util_min = p->uclamp_req[UCLAMP_MIN].value; |
1788 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd | 1789 | |
480a6ca2 DE |
1790 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { |
1791 | util_min = attr->sched_util_min; | |
a509a7cd | 1792 | |
480a6ca2 DE |
1793 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) |
1794 | return -EINVAL; | |
1795 | } | |
1796 | ||
1797 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1798 | util_max = attr->sched_util_max; | |
1799 | ||
1800 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | |
1801 | return -EINVAL; | |
1802 | } | |
1803 | ||
1804 | if (util_min != -1 && util_max != -1 && util_min > util_max) | |
a509a7cd PB |
1805 | return -EINVAL; |
1806 | ||
e65855a5 QY |
1807 | /* |
1808 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1809 | * | |
1810 | * We need to do that here, because enabling static branches is a | |
1811 | * blocking operation which obviously cannot be done while holding | |
1812 | * scheduler locks. | |
1813 | */ | |
1814 | static_branch_enable(&sched_uclamp_used); | |
1815 | ||
a509a7cd PB |
1816 | return 0; |
1817 | } | |
1818 | ||
480a6ca2 DE |
1819 | static bool uclamp_reset(const struct sched_attr *attr, |
1820 | enum uclamp_id clamp_id, | |
1821 | struct uclamp_se *uc_se) | |
1822 | { | |
1823 | /* Reset on sched class change for a non user-defined clamp value. */ | |
1824 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | |
1825 | !uc_se->user_defined) | |
1826 | return true; | |
1827 | ||
1828 | /* Reset on sched_util_{min,max} == -1. */ | |
1829 | if (clamp_id == UCLAMP_MIN && | |
1830 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | |
1831 | attr->sched_util_min == -1) { | |
1832 | return true; | |
1833 | } | |
1834 | ||
1835 | if (clamp_id == UCLAMP_MAX && | |
1836 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | |
1837 | attr->sched_util_max == -1) { | |
1838 | return true; | |
1839 | } | |
1840 | ||
1841 | return false; | |
1842 | } | |
1843 | ||
a509a7cd PB |
1844 | static void __setscheduler_uclamp(struct task_struct *p, |
1845 | const struct sched_attr *attr) | |
1846 | { | |
0413d7f3 | 1847 | enum uclamp_id clamp_id; |
1a00d999 | 1848 | |
1a00d999 PB |
1849 | for_each_clamp_id(clamp_id) { |
1850 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
480a6ca2 | 1851 | unsigned int value; |
1a00d999 | 1852 | |
480a6ca2 | 1853 | if (!uclamp_reset(attr, clamp_id, uc_se)) |
1a00d999 PB |
1854 | continue; |
1855 | ||
13685c4a QY |
1856 | /* |
1857 | * RT by default have a 100% boost value that could be modified | |
1858 | * at runtime. | |
1859 | */ | |
1a00d999 | 1860 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
480a6ca2 | 1861 | value = sysctl_sched_uclamp_util_min_rt_default; |
13685c4a | 1862 | else |
480a6ca2 DE |
1863 | value = uclamp_none(clamp_id); |
1864 | ||
1865 | uclamp_se_set(uc_se, value, false); | |
1a00d999 | 1866 | |
1a00d999 PB |
1867 | } |
1868 | ||
a509a7cd PB |
1869 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1870 | return; | |
1871 | ||
480a6ca2 DE |
1872 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && |
1873 | attr->sched_util_min != -1) { | |
a509a7cd PB |
1874 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], |
1875 | attr->sched_util_min, true); | |
1876 | } | |
1877 | ||
480a6ca2 DE |
1878 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && |
1879 | attr->sched_util_max != -1) { | |
a509a7cd PB |
1880 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], |
1881 | attr->sched_util_max, true); | |
1882 | } | |
1883 | } | |
1884 | ||
e8f14172 PB |
1885 | static void uclamp_fork(struct task_struct *p) |
1886 | { | |
0413d7f3 | 1887 | enum uclamp_id clamp_id; |
e8f14172 | 1888 | |
13685c4a QY |
1889 | /* |
1890 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1891 | * as the task is still at its early fork stages. | |
1892 | */ | |
e8f14172 PB |
1893 | for_each_clamp_id(clamp_id) |
1894 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1895 | |
1896 | if (likely(!p->sched_reset_on_fork)) | |
1897 | return; | |
1898 | ||
1899 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1900 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1901 | uclamp_none(clamp_id), false); | |
a87498ac | 1902 | } |
e8f14172 PB |
1903 | } |
1904 | ||
13685c4a QY |
1905 | static void uclamp_post_fork(struct task_struct *p) |
1906 | { | |
1907 | uclamp_update_util_min_rt_default(p); | |
1908 | } | |
1909 | ||
d81ae8aa QY |
1910 | static void __init init_uclamp_rq(struct rq *rq) |
1911 | { | |
1912 | enum uclamp_id clamp_id; | |
1913 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1914 | ||
1915 | for_each_clamp_id(clamp_id) { | |
1916 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1917 | .value = uclamp_none(clamp_id) | |
1918 | }; | |
1919 | } | |
1920 | ||
1921 | rq->uclamp_flags = 0; | |
1922 | } | |
1923 | ||
69842cba PB |
1924 | static void __init init_uclamp(void) |
1925 | { | |
e8f14172 | 1926 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1927 | enum uclamp_id clamp_id; |
69842cba PB |
1928 | int cpu; |
1929 | ||
d81ae8aa QY |
1930 | for_each_possible_cpu(cpu) |
1931 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1932 | |
69842cba | 1933 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1934 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1935 | uclamp_none(clamp_id), false); |
69842cba | 1936 | } |
e8f14172 PB |
1937 | |
1938 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 1939 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 1940 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1941 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
1942 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
1943 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 1944 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
1945 | #endif |
1946 | } | |
69842cba PB |
1947 | } |
1948 | ||
1949 | #else /* CONFIG_UCLAMP_TASK */ | |
1950 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
1951 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
1952 | static inline int uclamp_validate(struct task_struct *p, |
1953 | const struct sched_attr *attr) | |
1954 | { | |
1955 | return -EOPNOTSUPP; | |
1956 | } | |
1957 | static void __setscheduler_uclamp(struct task_struct *p, | |
1958 | const struct sched_attr *attr) { } | |
e8f14172 | 1959 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 1960 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
1961 | static inline void init_uclamp(void) { } |
1962 | #endif /* CONFIG_UCLAMP_TASK */ | |
1963 | ||
a1dfb631 MT |
1964 | bool sched_task_on_rq(struct task_struct *p) |
1965 | { | |
1966 | return task_on_rq_queued(p); | |
1967 | } | |
1968 | ||
42a20f86 KC |
1969 | unsigned long get_wchan(struct task_struct *p) |
1970 | { | |
1971 | unsigned long ip = 0; | |
1972 | unsigned int state; | |
1973 | ||
1974 | if (!p || p == current) | |
1975 | return 0; | |
1976 | ||
1977 | /* Only get wchan if task is blocked and we can keep it that way. */ | |
1978 | raw_spin_lock_irq(&p->pi_lock); | |
1979 | state = READ_ONCE(p->__state); | |
1980 | smp_rmb(); /* see try_to_wake_up() */ | |
1981 | if (state != TASK_RUNNING && state != TASK_WAKING && !p->on_rq) | |
1982 | ip = __get_wchan(p); | |
1983 | raw_spin_unlock_irq(&p->pi_lock); | |
1984 | ||
1985 | return ip; | |
1986 | } | |
1987 | ||
1de64443 | 1988 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1989 | { |
0a67d1ee PZ |
1990 | if (!(flags & ENQUEUE_NOCLOCK)) |
1991 | update_rq_clock(rq); | |
1992 | ||
eb414681 | 1993 | if (!(flags & ENQUEUE_RESTORE)) { |
4e29fb70 | 1994 | sched_info_enqueue(rq, p); |
eb414681 JW |
1995 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
1996 | } | |
0a67d1ee | 1997 | |
69842cba | 1998 | uclamp_rq_inc(rq, p); |
371fd7e7 | 1999 | p->sched_class->enqueue_task(rq, p, flags); |
8a311c74 PZ |
2000 | |
2001 | if (sched_core_enabled(rq)) | |
2002 | sched_core_enqueue(rq, p); | |
71f8bd46 IM |
2003 | } |
2004 | ||
1de64443 | 2005 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 2006 | { |
8a311c74 PZ |
2007 | if (sched_core_enabled(rq)) |
2008 | sched_core_dequeue(rq, p); | |
2009 | ||
0a67d1ee PZ |
2010 | if (!(flags & DEQUEUE_NOCLOCK)) |
2011 | update_rq_clock(rq); | |
2012 | ||
eb414681 | 2013 | if (!(flags & DEQUEUE_SAVE)) { |
4e29fb70 | 2014 | sched_info_dequeue(rq, p); |
eb414681 JW |
2015 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
2016 | } | |
0a67d1ee | 2017 | |
69842cba | 2018 | uclamp_rq_dec(rq, p); |
371fd7e7 | 2019 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
2020 | } |
2021 | ||
029632fb | 2022 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 2023 | { |
371fd7e7 | 2024 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
2025 | |
2026 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
2027 | } |
2028 | ||
029632fb | 2029 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 2030 | { |
7dd77884 PZ |
2031 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
2032 | ||
371fd7e7 | 2033 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
2034 | } |
2035 | ||
f558c2b8 | 2036 | static inline int __normal_prio(int policy, int rt_prio, int nice) |
14531189 | 2037 | { |
f558c2b8 PZ |
2038 | int prio; |
2039 | ||
2040 | if (dl_policy(policy)) | |
2041 | prio = MAX_DL_PRIO - 1; | |
2042 | else if (rt_policy(policy)) | |
2043 | prio = MAX_RT_PRIO - 1 - rt_prio; | |
2044 | else | |
2045 | prio = NICE_TO_PRIO(nice); | |
2046 | ||
2047 | return prio; | |
14531189 IM |
2048 | } |
2049 | ||
b29739f9 IM |
2050 | /* |
2051 | * Calculate the expected normal priority: i.e. priority | |
2052 | * without taking RT-inheritance into account. Might be | |
2053 | * boosted by interactivity modifiers. Changes upon fork, | |
2054 | * setprio syscalls, and whenever the interactivity | |
2055 | * estimator recalculates. | |
2056 | */ | |
36c8b586 | 2057 | static inline int normal_prio(struct task_struct *p) |
b29739f9 | 2058 | { |
f558c2b8 | 2059 | return __normal_prio(p->policy, p->rt_priority, PRIO_TO_NICE(p->static_prio)); |
b29739f9 IM |
2060 | } |
2061 | ||
2062 | /* | |
2063 | * Calculate the current priority, i.e. the priority | |
2064 | * taken into account by the scheduler. This value might | |
2065 | * be boosted by RT tasks, or might be boosted by | |
2066 | * interactivity modifiers. Will be RT if the task got | |
2067 | * RT-boosted. If not then it returns p->normal_prio. | |
2068 | */ | |
36c8b586 | 2069 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2070 | { |
2071 | p->normal_prio = normal_prio(p); | |
2072 | /* | |
2073 | * If we are RT tasks or we were boosted to RT priority, | |
2074 | * keep the priority unchanged. Otherwise, update priority | |
2075 | * to the normal priority: | |
2076 | */ | |
2077 | if (!rt_prio(p->prio)) | |
2078 | return p->normal_prio; | |
2079 | return p->prio; | |
2080 | } | |
2081 | ||
1da177e4 LT |
2082 | /** |
2083 | * task_curr - is this task currently executing on a CPU? | |
2084 | * @p: the task in question. | |
e69f6186 YB |
2085 | * |
2086 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 2087 | */ |
36c8b586 | 2088 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2089 | { |
2090 | return cpu_curr(task_cpu(p)) == p; | |
2091 | } | |
2092 | ||
67dfa1b7 | 2093 | /* |
4c9a4bc8 PZ |
2094 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
2095 | * use the balance_callback list if you want balancing. | |
2096 | * | |
2097 | * this means any call to check_class_changed() must be followed by a call to | |
2098 | * balance_callback(). | |
67dfa1b7 | 2099 | */ |
cb469845 SR |
2100 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2101 | const struct sched_class *prev_class, | |
da7a735e | 2102 | int oldprio) |
cb469845 SR |
2103 | { |
2104 | if (prev_class != p->sched_class) { | |
2105 | if (prev_class->switched_from) | |
da7a735e | 2106 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 2107 | |
da7a735e | 2108 | p->sched_class->switched_to(rq, p); |
2d3d891d | 2109 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 2110 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
2111 | } |
2112 | ||
029632fb | 2113 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 2114 | { |
aa93cd53 | 2115 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 2116 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
aa93cd53 KT |
2117 | else if (p->sched_class > rq->curr->sched_class) |
2118 | resched_curr(rq); | |
1e5a7405 PZ |
2119 | |
2120 | /* | |
2121 | * A queue event has occurred, and we're going to schedule. In | |
2122 | * this case, we can save a useless back to back clock update. | |
2123 | */ | |
da0c1e65 | 2124 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 2125 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
2126 | } |
2127 | ||
1da177e4 | 2128 | #ifdef CONFIG_SMP |
175f0e25 | 2129 | |
af449901 PZ |
2130 | static void |
2131 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags); | |
2132 | ||
2133 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
2134 | const struct cpumask *new_mask, | |
2135 | u32 flags); | |
2136 | ||
2137 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
2138 | { | |
2139 | if (likely(!p->migration_disabled)) | |
2140 | return; | |
2141 | ||
2142 | if (p->cpus_ptr != &p->cpus_mask) | |
2143 | return; | |
2144 | ||
2145 | /* | |
2146 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
2147 | */ | |
2148 | __do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE); | |
2149 | } | |
2150 | ||
2151 | void migrate_disable(void) | |
2152 | { | |
3015ef4b TG |
2153 | struct task_struct *p = current; |
2154 | ||
2155 | if (p->migration_disabled) { | |
2156 | p->migration_disabled++; | |
af449901 | 2157 | return; |
3015ef4b | 2158 | } |
af449901 | 2159 | |
3015ef4b TG |
2160 | preempt_disable(); |
2161 | this_rq()->nr_pinned++; | |
2162 | p->migration_disabled = 1; | |
2163 | preempt_enable(); | |
af449901 PZ |
2164 | } |
2165 | EXPORT_SYMBOL_GPL(migrate_disable); | |
2166 | ||
2167 | void migrate_enable(void) | |
2168 | { | |
2169 | struct task_struct *p = current; | |
2170 | ||
6d337eab PZ |
2171 | if (p->migration_disabled > 1) { |
2172 | p->migration_disabled--; | |
af449901 | 2173 | return; |
6d337eab | 2174 | } |
af449901 | 2175 | |
6d337eab PZ |
2176 | /* |
2177 | * Ensure stop_task runs either before or after this, and that | |
2178 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
2179 | */ | |
2180 | preempt_disable(); | |
2181 | if (p->cpus_ptr != &p->cpus_mask) | |
2182 | __set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE); | |
2183 | /* | |
2184 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
2185 | * regular cpus_mask, otherwise things that race (eg. | |
2186 | * select_fallback_rq) get confused. | |
2187 | */ | |
af449901 | 2188 | barrier(); |
6d337eab | 2189 | p->migration_disabled = 0; |
3015ef4b | 2190 | this_rq()->nr_pinned--; |
6d337eab | 2191 | preempt_enable(); |
af449901 PZ |
2192 | } |
2193 | EXPORT_SYMBOL_GPL(migrate_enable); | |
2194 | ||
3015ef4b TG |
2195 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2196 | { | |
2197 | return rq->nr_pinned; | |
2198 | } | |
2199 | ||
175f0e25 | 2200 | /* |
bee98539 | 2201 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
2202 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
2203 | */ | |
2204 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
2205 | { | |
5ba2ffba | 2206 | /* When not in the task's cpumask, no point in looking further. */ |
3bd37062 | 2207 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
2208 | return false; |
2209 | ||
5ba2ffba PZ |
2210 | /* migrate_disabled() must be allowed to finish. */ |
2211 | if (is_migration_disabled(p)) | |
175f0e25 PZ |
2212 | return cpu_online(cpu); |
2213 | ||
5ba2ffba PZ |
2214 | /* Non kernel threads are not allowed during either online or offline. */ |
2215 | if (!(p->flags & PF_KTHREAD)) | |
9ae606bc | 2216 | return cpu_active(cpu) && task_cpu_possible(cpu, p); |
5ba2ffba PZ |
2217 | |
2218 | /* KTHREAD_IS_PER_CPU is always allowed. */ | |
2219 | if (kthread_is_per_cpu(p)) | |
2220 | return cpu_online(cpu); | |
2221 | ||
2222 | /* Regular kernel threads don't get to stay during offline. */ | |
b5c44773 | 2223 | if (cpu_dying(cpu)) |
5ba2ffba PZ |
2224 | return false; |
2225 | ||
2226 | /* But are allowed during online. */ | |
2227 | return cpu_online(cpu); | |
175f0e25 PZ |
2228 | } |
2229 | ||
5cc389bc PZ |
2230 | /* |
2231 | * This is how migration works: | |
2232 | * | |
2233 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
2234 | * stop_one_cpu(). | |
2235 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
2236 | * off the CPU) | |
2237 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
2238 | * 4) if it's in the wrong runqueue then the migration thread removes | |
2239 | * it and puts it into the right queue. | |
2240 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
2241 | * is done. | |
2242 | */ | |
2243 | ||
2244 | /* | |
2245 | * move_queued_task - move a queued task to new rq. | |
2246 | * | |
2247 | * Returns (locked) new rq. Old rq's lock is released. | |
2248 | */ | |
8a8c69c3 PZ |
2249 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
2250 | struct task_struct *p, int new_cpu) | |
5cc389bc | 2251 | { |
5cb9eaa3 | 2252 | lockdep_assert_rq_held(rq); |
5cc389bc | 2253 | |
58877d34 | 2254 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 2255 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 2256 | rq_unlock(rq, rf); |
5cc389bc PZ |
2257 | |
2258 | rq = cpu_rq(new_cpu); | |
2259 | ||
8a8c69c3 | 2260 | rq_lock(rq, rf); |
5cc389bc | 2261 | BUG_ON(task_cpu(p) != new_cpu); |
58877d34 | 2262 | activate_task(rq, p, 0); |
5cc389bc PZ |
2263 | check_preempt_curr(rq, p, 0); |
2264 | ||
2265 | return rq; | |
2266 | } | |
2267 | ||
2268 | struct migration_arg { | |
6d337eab PZ |
2269 | struct task_struct *task; |
2270 | int dest_cpu; | |
2271 | struct set_affinity_pending *pending; | |
2272 | }; | |
2273 | ||
50caf9c1 PZ |
2274 | /* |
2275 | * @refs: number of wait_for_completion() | |
2276 | * @stop_pending: is @stop_work in use | |
2277 | */ | |
6d337eab PZ |
2278 | struct set_affinity_pending { |
2279 | refcount_t refs; | |
9e81889c | 2280 | unsigned int stop_pending; |
6d337eab PZ |
2281 | struct completion done; |
2282 | struct cpu_stop_work stop_work; | |
2283 | struct migration_arg arg; | |
5cc389bc PZ |
2284 | }; |
2285 | ||
2286 | /* | |
d1ccc66d | 2287 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
2288 | * this because either it can't run here any more (set_cpus_allowed() |
2289 | * away from this CPU, or CPU going down), or because we're | |
2290 | * attempting to rebalance this task on exec (sched_exec). | |
2291 | * | |
2292 | * So we race with normal scheduler movements, but that's OK, as long | |
2293 | * as the task is no longer on this CPU. | |
5cc389bc | 2294 | */ |
8a8c69c3 PZ |
2295 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
2296 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 2297 | { |
5cc389bc | 2298 | /* Affinity changed (again). */ |
175f0e25 | 2299 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 2300 | return rq; |
5cc389bc | 2301 | |
15ff991e | 2302 | update_rq_clock(rq); |
8a8c69c3 | 2303 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
2304 | |
2305 | return rq; | |
5cc389bc PZ |
2306 | } |
2307 | ||
2308 | /* | |
2309 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
2310 | * and performs thread migration by bumping thread off CPU then | |
2311 | * 'pushing' onto another runqueue. | |
2312 | */ | |
2313 | static int migration_cpu_stop(void *data) | |
2314 | { | |
2315 | struct migration_arg *arg = data; | |
c20cf065 | 2316 | struct set_affinity_pending *pending = arg->pending; |
5e16bbc2 PZ |
2317 | struct task_struct *p = arg->task; |
2318 | struct rq *rq = this_rq(); | |
6d337eab | 2319 | bool complete = false; |
8a8c69c3 | 2320 | struct rq_flags rf; |
5cc389bc PZ |
2321 | |
2322 | /* | |
d1ccc66d IM |
2323 | * The original target CPU might have gone down and we might |
2324 | * be on another CPU but it doesn't matter. | |
5cc389bc | 2325 | */ |
6d337eab | 2326 | local_irq_save(rf.flags); |
5cc389bc PZ |
2327 | /* |
2328 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 2329 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
2330 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
2331 | */ | |
a1488664 | 2332 | flush_smp_call_function_from_idle(); |
5e16bbc2 PZ |
2333 | |
2334 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 2335 | rq_lock(rq, &rf); |
6d337eab | 2336 | |
e140749c VS |
2337 | /* |
2338 | * If we were passed a pending, then ->stop_pending was set, thus | |
2339 | * p->migration_pending must have remained stable. | |
2340 | */ | |
2341 | WARN_ON_ONCE(pending && pending != p->migration_pending); | |
2342 | ||
5e16bbc2 PZ |
2343 | /* |
2344 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
2345 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
2346 | * we're holding p->pi_lock. | |
2347 | */ | |
bf89a304 | 2348 | if (task_rq(p) == rq) { |
6d337eab PZ |
2349 | if (is_migration_disabled(p)) |
2350 | goto out; | |
2351 | ||
2352 | if (pending) { | |
e140749c | 2353 | p->migration_pending = NULL; |
6d337eab | 2354 | complete = true; |
6d337eab | 2355 | |
3f1bc119 PZ |
2356 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) |
2357 | goto out; | |
3f1bc119 | 2358 | } |
6d337eab | 2359 | |
bf89a304 | 2360 | if (task_on_rq_queued(p)) |
475ea6c6 | 2361 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 | 2362 | else |
475ea6c6 | 2363 | p->wake_cpu = arg->dest_cpu; |
6d337eab | 2364 | |
3f1bc119 PZ |
2365 | /* |
2366 | * XXX __migrate_task() can fail, at which point we might end | |
2367 | * up running on a dodgy CPU, AFAICT this can only happen | |
2368 | * during CPU hotplug, at which point we'll get pushed out | |
2369 | * anyway, so it's probably not a big deal. | |
2370 | */ | |
2371 | ||
c20cf065 | 2372 | } else if (pending) { |
6d337eab PZ |
2373 | /* |
2374 | * This happens when we get migrated between migrate_enable()'s | |
2375 | * preempt_enable() and scheduling the stopper task. At that | |
2376 | * point we're a regular task again and not current anymore. | |
2377 | * | |
2378 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
2379 | * more likely. | |
2380 | */ | |
2381 | ||
d707faa6 VS |
2382 | /* |
2383 | * The task moved before the stopper got to run. We're holding | |
2384 | * ->pi_lock, so the allowed mask is stable - if it got | |
2385 | * somewhere allowed, we're done. | |
2386 | */ | |
c20cf065 | 2387 | if (cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { |
e140749c | 2388 | p->migration_pending = NULL; |
d707faa6 VS |
2389 | complete = true; |
2390 | goto out; | |
2391 | } | |
2392 | ||
6d337eab PZ |
2393 | /* |
2394 | * When migrate_enable() hits a rq mis-match we can't reliably | |
2395 | * determine is_migration_disabled() and so have to chase after | |
2396 | * it. | |
2397 | */ | |
9e81889c | 2398 | WARN_ON_ONCE(!pending->stop_pending); |
6d337eab PZ |
2399 | task_rq_unlock(rq, p, &rf); |
2400 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
2401 | &pending->arg, &pending->stop_work); | |
2402 | return 0; | |
bf89a304 | 2403 | } |
6d337eab | 2404 | out: |
9e81889c PZ |
2405 | if (pending) |
2406 | pending->stop_pending = false; | |
6d337eab PZ |
2407 | task_rq_unlock(rq, p, &rf); |
2408 | ||
2409 | if (complete) | |
2410 | complete_all(&pending->done); | |
2411 | ||
5cc389bc PZ |
2412 | return 0; |
2413 | } | |
2414 | ||
a7c81556 PZ |
2415 | int push_cpu_stop(void *arg) |
2416 | { | |
2417 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2418 | struct task_struct *p = arg; | |
2419 | ||
2420 | raw_spin_lock_irq(&p->pi_lock); | |
5cb9eaa3 | 2421 | raw_spin_rq_lock(rq); |
a7c81556 PZ |
2422 | |
2423 | if (task_rq(p) != rq) | |
2424 | goto out_unlock; | |
2425 | ||
2426 | if (is_migration_disabled(p)) { | |
2427 | p->migration_flags |= MDF_PUSH; | |
2428 | goto out_unlock; | |
2429 | } | |
2430 | ||
2431 | p->migration_flags &= ~MDF_PUSH; | |
2432 | ||
2433 | if (p->sched_class->find_lock_rq) | |
2434 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2435 | |
a7c81556 PZ |
2436 | if (!lowest_rq) |
2437 | goto out_unlock; | |
2438 | ||
2439 | // XXX validate p is still the highest prio task | |
2440 | if (task_rq(p) == rq) { | |
2441 | deactivate_task(rq, p, 0); | |
2442 | set_task_cpu(p, lowest_rq->cpu); | |
2443 | activate_task(lowest_rq, p, 0); | |
2444 | resched_curr(lowest_rq); | |
2445 | } | |
2446 | ||
2447 | double_unlock_balance(rq, lowest_rq); | |
2448 | ||
2449 | out_unlock: | |
2450 | rq->push_busy = false; | |
5cb9eaa3 | 2451 | raw_spin_rq_unlock(rq); |
a7c81556 PZ |
2452 | raw_spin_unlock_irq(&p->pi_lock); |
2453 | ||
2454 | put_task_struct(p); | |
5cc389bc PZ |
2455 | return 0; |
2456 | } | |
2457 | ||
c5b28038 PZ |
2458 | /* |
2459 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2460 | * actually call this function. | |
2461 | */ | |
9cfc3e18 | 2462 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 2463 | { |
af449901 PZ |
2464 | if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2465 | p->cpus_ptr = new_mask; | |
2466 | return; | |
2467 | } | |
2468 | ||
3bd37062 | 2469 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
2470 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
2471 | } | |
2472 | ||
9cfc3e18 PZ |
2473 | static void |
2474 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 2475 | { |
6c37067e PZ |
2476 | struct rq *rq = task_rq(p); |
2477 | bool queued, running; | |
2478 | ||
af449901 PZ |
2479 | /* |
2480 | * This here violates the locking rules for affinity, since we're only | |
2481 | * supposed to change these variables while holding both rq->lock and | |
2482 | * p->pi_lock. | |
2483 | * | |
2484 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2485 | * accesses these variables under p->pi_lock and only does so after | |
2486 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2487 | * before finish_task(). | |
2488 | * | |
2489 | * XXX do further audits, this smells like something putrid. | |
2490 | */ | |
2491 | if (flags & SCA_MIGRATE_DISABLE) | |
2492 | SCHED_WARN_ON(!p->on_cpu); | |
2493 | else | |
2494 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2495 | |
2496 | queued = task_on_rq_queued(p); | |
2497 | running = task_current(rq, p); | |
2498 | ||
2499 | if (queued) { | |
2500 | /* | |
2501 | * Because __kthread_bind() calls this on blocked tasks without | |
2502 | * holding rq->lock. | |
2503 | */ | |
5cb9eaa3 | 2504 | lockdep_assert_rq_held(rq); |
7a57f32a | 2505 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2506 | } |
2507 | if (running) | |
2508 | put_prev_task(rq, p); | |
2509 | ||
9cfc3e18 | 2510 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 2511 | |
6c37067e | 2512 | if (queued) |
7134b3e9 | 2513 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2514 | if (running) |
03b7fad1 | 2515 | set_next_task(rq, p); |
c5b28038 PZ |
2516 | } |
2517 | ||
9cfc3e18 PZ |
2518 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2519 | { | |
2520 | __do_set_cpus_allowed(p, new_mask, 0); | |
2521 | } | |
2522 | ||
b90ca8ba WD |
2523 | int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, |
2524 | int node) | |
2525 | { | |
2526 | if (!src->user_cpus_ptr) | |
2527 | return 0; | |
2528 | ||
2529 | dst->user_cpus_ptr = kmalloc_node(cpumask_size(), GFP_KERNEL, node); | |
2530 | if (!dst->user_cpus_ptr) | |
2531 | return -ENOMEM; | |
2532 | ||
2533 | cpumask_copy(dst->user_cpus_ptr, src->user_cpus_ptr); | |
2534 | return 0; | |
2535 | } | |
2536 | ||
07ec77a1 WD |
2537 | static inline struct cpumask *clear_user_cpus_ptr(struct task_struct *p) |
2538 | { | |
2539 | struct cpumask *user_mask = NULL; | |
2540 | ||
2541 | swap(p->user_cpus_ptr, user_mask); | |
2542 | ||
2543 | return user_mask; | |
2544 | } | |
2545 | ||
b90ca8ba WD |
2546 | void release_user_cpus_ptr(struct task_struct *p) |
2547 | { | |
07ec77a1 | 2548 | kfree(clear_user_cpus_ptr(p)); |
b90ca8ba WD |
2549 | } |
2550 | ||
6d337eab | 2551 | /* |
c777d847 VS |
2552 | * This function is wildly self concurrent; here be dragons. |
2553 | * | |
2554 | * | |
2555 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2556 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2557 | * running, we have to kick it out using the CPU stopper. | |
2558 | * | |
2559 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2560 | * Consider: | |
2561 | * | |
2562 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2563 | * | |
2564 | * P0@CPU0 P1 | |
2565 | * | |
2566 | * migrate_disable(); | |
2567 | * <preempted> | |
2568 | * set_cpus_allowed_ptr(P0, [1]); | |
2569 | * | |
2570 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2571 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2572 | * This means we need the following scheme: | |
2573 | * | |
2574 | * P0@CPU0 P1 | |
2575 | * | |
2576 | * migrate_disable(); | |
2577 | * <preempted> | |
2578 | * set_cpus_allowed_ptr(P0, [1]); | |
2579 | * <blocks> | |
2580 | * <resumes> | |
2581 | * migrate_enable(); | |
2582 | * __set_cpus_allowed_ptr(); | |
2583 | * <wakes local stopper> | |
2584 | * `--> <woken on migration completion> | |
2585 | * | |
2586 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2587 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2588 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2589 | * should come into effect at the end of the Migrate-Disable region is the last | |
2590 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2591 | * but we still need to properly signal those waiting tasks at the appropriate | |
2592 | * moment. | |
2593 | * | |
2594 | * This is implemented using struct set_affinity_pending. The first | |
2595 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2596 | * setup an instance of that struct and install it on the targeted task_struct. | |
2597 | * Any and all further callers will reuse that instance. Those then wait for | |
2598 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2599 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2600 | * | |
2601 | * | |
2602 | * (1) In the cases covered above. There is one more where the completion is | |
2603 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
e140749c VS |
2604 | * occurs after the stopper bailed out due to the targeted task still being |
2605 | * Migrate-Disable. Consider: | |
c777d847 VS |
2606 | * |
2607 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2608 | * | |
e140749c VS |
2609 | * CPU0 P1 P2 |
2610 | * <P0> | |
2611 | * migrate_disable(); | |
2612 | * <preempted> | |
c777d847 VS |
2613 | * set_cpus_allowed_ptr(P0, [1]); |
2614 | * <blocks> | |
e140749c VS |
2615 | * <migration/0> |
2616 | * migration_cpu_stop() | |
2617 | * is_migration_disabled() | |
2618 | * <bails> | |
c777d847 VS |
2619 | * set_cpus_allowed_ptr(P0, [0, 1]); |
2620 | * <signal completion> | |
2621 | * <awakes> | |
2622 | * | |
2623 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2624 | * pending affinity completion is preceded by an uninstallation of | |
2625 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2626 | */ |
2627 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2628 | int dest_cpu, unsigned int flags) | |
2629 | { | |
2630 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
9e81889c | 2631 | bool stop_pending, complete = false; |
6d337eab PZ |
2632 | |
2633 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2634 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2635 | struct task_struct *push_task = NULL; |
2636 | ||
2637 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2638 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2639 | rq->push_busy = true; | |
2640 | push_task = get_task_struct(p); | |
2641 | } | |
2642 | ||
50caf9c1 PZ |
2643 | /* |
2644 | * If there are pending waiters, but no pending stop_work, | |
2645 | * then complete now. | |
2646 | */ | |
6d337eab | 2647 | pending = p->migration_pending; |
50caf9c1 | 2648 | if (pending && !pending->stop_pending) { |
6d337eab PZ |
2649 | p->migration_pending = NULL; |
2650 | complete = true; | |
2651 | } | |
50caf9c1 | 2652 | |
6d337eab PZ |
2653 | task_rq_unlock(rq, p, rf); |
2654 | ||
a7c81556 PZ |
2655 | if (push_task) { |
2656 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2657 | p, &rq->push_work); | |
2658 | } | |
2659 | ||
6d337eab | 2660 | if (complete) |
50caf9c1 | 2661 | complete_all(&pending->done); |
6d337eab PZ |
2662 | |
2663 | return 0; | |
2664 | } | |
2665 | ||
2666 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2667 | /* serialized by p->pi_lock */ | |
2668 | if (!p->migration_pending) { | |
c777d847 | 2669 | /* Install the request */ |
6d337eab PZ |
2670 | refcount_set(&my_pending.refs, 1); |
2671 | init_completion(&my_pending.done); | |
8a6edb52 PZ |
2672 | my_pending.arg = (struct migration_arg) { |
2673 | .task = p, | |
475ea6c6 | 2674 | .dest_cpu = dest_cpu, |
8a6edb52 PZ |
2675 | .pending = &my_pending, |
2676 | }; | |
2677 | ||
6d337eab PZ |
2678 | p->migration_pending = &my_pending; |
2679 | } else { | |
2680 | pending = p->migration_pending; | |
2681 | refcount_inc(&pending->refs); | |
475ea6c6 VS |
2682 | /* |
2683 | * Affinity has changed, but we've already installed a | |
2684 | * pending. migration_cpu_stop() *must* see this, else | |
2685 | * we risk a completion of the pending despite having a | |
2686 | * task on a disallowed CPU. | |
2687 | * | |
2688 | * Serialized by p->pi_lock, so this is safe. | |
2689 | */ | |
2690 | pending->arg.dest_cpu = dest_cpu; | |
6d337eab PZ |
2691 | } |
2692 | } | |
2693 | pending = p->migration_pending; | |
2694 | /* | |
2695 | * - !MIGRATE_ENABLE: | |
2696 | * we'll have installed a pending if there wasn't one already. | |
2697 | * | |
2698 | * - MIGRATE_ENABLE: | |
2699 | * we're here because the current CPU isn't matching anymore, | |
2700 | * the only way that can happen is because of a concurrent | |
2701 | * set_cpus_allowed_ptr() call, which should then still be | |
2702 | * pending completion. | |
2703 | * | |
2704 | * Either way, we really should have a @pending here. | |
2705 | */ | |
2706 | if (WARN_ON_ONCE(!pending)) { | |
2707 | task_rq_unlock(rq, p, rf); | |
2708 | return -EINVAL; | |
2709 | } | |
2710 | ||
2f064a59 | 2711 | if (task_running(rq, p) || READ_ONCE(p->__state) == TASK_WAKING) { |
c777d847 | 2712 | /* |
58b1a450 PZ |
2713 | * MIGRATE_ENABLE gets here because 'p == current', but for |
2714 | * anything else we cannot do is_migration_disabled(), punt | |
2715 | * and have the stopper function handle it all race-free. | |
c777d847 | 2716 | */ |
9e81889c PZ |
2717 | stop_pending = pending->stop_pending; |
2718 | if (!stop_pending) | |
2719 | pending->stop_pending = true; | |
58b1a450 | 2720 | |
58b1a450 PZ |
2721 | if (flags & SCA_MIGRATE_ENABLE) |
2722 | p->migration_flags &= ~MDF_PUSH; | |
50caf9c1 | 2723 | |
6d337eab | 2724 | task_rq_unlock(rq, p, rf); |
8a6edb52 | 2725 | |
9e81889c PZ |
2726 | if (!stop_pending) { |
2727 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2728 | &pending->arg, &pending->stop_work); | |
2729 | } | |
6d337eab | 2730 | |
58b1a450 PZ |
2731 | if (flags & SCA_MIGRATE_ENABLE) |
2732 | return 0; | |
6d337eab PZ |
2733 | } else { |
2734 | ||
2735 | if (!is_migration_disabled(p)) { | |
2736 | if (task_on_rq_queued(p)) | |
2737 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2738 | ||
50caf9c1 PZ |
2739 | if (!pending->stop_pending) { |
2740 | p->migration_pending = NULL; | |
2741 | complete = true; | |
2742 | } | |
6d337eab PZ |
2743 | } |
2744 | task_rq_unlock(rq, p, rf); | |
2745 | ||
6d337eab PZ |
2746 | if (complete) |
2747 | complete_all(&pending->done); | |
2748 | } | |
2749 | ||
2750 | wait_for_completion(&pending->done); | |
2751 | ||
2752 | if (refcount_dec_and_test(&pending->refs)) | |
50caf9c1 | 2753 | wake_up_var(&pending->refs); /* No UaF, just an address */ |
6d337eab | 2754 | |
c777d847 VS |
2755 | /* |
2756 | * Block the original owner of &pending until all subsequent callers | |
2757 | * have seen the completion and decremented the refcount | |
2758 | */ | |
6d337eab PZ |
2759 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2760 | ||
50caf9c1 PZ |
2761 | /* ARGH */ |
2762 | WARN_ON_ONCE(my_pending.stop_pending); | |
2763 | ||
6d337eab PZ |
2764 | return 0; |
2765 | } | |
2766 | ||
5cc389bc | 2767 | /* |
07ec77a1 | 2768 | * Called with both p->pi_lock and rq->lock held; drops both before returning. |
5cc389bc | 2769 | */ |
07ec77a1 WD |
2770 | static int __set_cpus_allowed_ptr_locked(struct task_struct *p, |
2771 | const struct cpumask *new_mask, | |
2772 | u32 flags, | |
2773 | struct rq *rq, | |
2774 | struct rq_flags *rf) | |
2775 | __releases(rq->lock) | |
2776 | __releases(p->pi_lock) | |
5cc389bc | 2777 | { |
234a503e | 2778 | const struct cpumask *cpu_allowed_mask = task_cpu_possible_mask(p); |
e9d867a6 | 2779 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
234a503e | 2780 | bool kthread = p->flags & PF_KTHREAD; |
07ec77a1 | 2781 | struct cpumask *user_mask = NULL; |
5cc389bc PZ |
2782 | unsigned int dest_cpu; |
2783 | int ret = 0; | |
2784 | ||
a499c3ea | 2785 | update_rq_clock(rq); |
5cc389bc | 2786 | |
234a503e | 2787 | if (kthread || is_migration_disabled(p)) { |
e9d867a6 | 2788 | /* |
741ba80f PZ |
2789 | * Kernel threads are allowed on online && !active CPUs, |
2790 | * however, during cpu-hot-unplug, even these might get pushed | |
2791 | * away if not KTHREAD_IS_PER_CPU. | |
af449901 PZ |
2792 | * |
2793 | * Specifically, migration_disabled() tasks must not fail the | |
2794 | * cpumask_any_and_distribute() pick below, esp. so on | |
2795 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2796 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2797 | */ |
2798 | cpu_valid_mask = cpu_online_mask; | |
2799 | } | |
2800 | ||
234a503e WD |
2801 | if (!kthread && !cpumask_subset(new_mask, cpu_allowed_mask)) { |
2802 | ret = -EINVAL; | |
2803 | goto out; | |
2804 | } | |
2805 | ||
25834c73 PZ |
2806 | /* |
2807 | * Must re-check here, to close a race against __kthread_bind(), | |
2808 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2809 | */ | |
9cfc3e18 | 2810 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2811 | ret = -EINVAL; |
2812 | goto out; | |
2813 | } | |
2814 | ||
885b3ba4 VS |
2815 | if (!(flags & SCA_MIGRATE_ENABLE)) { |
2816 | if (cpumask_equal(&p->cpus_mask, new_mask)) | |
2817 | goto out; | |
2818 | ||
2819 | if (WARN_ON_ONCE(p == current && | |
2820 | is_migration_disabled(p) && | |
2821 | !cpumask_test_cpu(task_cpu(p), new_mask))) { | |
2822 | ret = -EBUSY; | |
2823 | goto out; | |
2824 | } | |
2825 | } | |
5cc389bc | 2826 | |
46a87b38 PT |
2827 | /* |
2828 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2829 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2830 | * immediately required to distribute the tasks within their new mask. | |
2831 | */ | |
2832 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 2833 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2834 | ret = -EINVAL; |
2835 | goto out; | |
2836 | } | |
2837 | ||
9cfc3e18 | 2838 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 2839 | |
07ec77a1 WD |
2840 | if (flags & SCA_USER) |
2841 | user_mask = clear_user_cpus_ptr(p); | |
2842 | ||
2843 | ret = affine_move_task(rq, p, rf, dest_cpu, flags); | |
2844 | ||
2845 | kfree(user_mask); | |
2846 | ||
2847 | return ret; | |
5cc389bc | 2848 | |
5cc389bc | 2849 | out: |
07ec77a1 | 2850 | task_rq_unlock(rq, p, rf); |
5cc389bc PZ |
2851 | |
2852 | return ret; | |
2853 | } | |
25834c73 | 2854 | |
07ec77a1 WD |
2855 | /* |
2856 | * Change a given task's CPU affinity. Migrate the thread to a | |
2857 | * proper CPU and schedule it away if the CPU it's executing on | |
2858 | * is removed from the allowed bitmask. | |
2859 | * | |
2860 | * NOTE: the caller must have a valid reference to the task, the | |
2861 | * task must not exit() & deallocate itself prematurely. The | |
2862 | * call is not atomic; no spinlocks may be held. | |
2863 | */ | |
2864 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
2865 | const struct cpumask *new_mask, u32 flags) | |
2866 | { | |
2867 | struct rq_flags rf; | |
2868 | struct rq *rq; | |
2869 | ||
2870 | rq = task_rq_lock(p, &rf); | |
2871 | return __set_cpus_allowed_ptr_locked(p, new_mask, flags, rq, &rf); | |
2872 | } | |
2873 | ||
25834c73 PZ |
2874 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
2875 | { | |
9cfc3e18 | 2876 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 2877 | } |
5cc389bc PZ |
2878 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2879 | ||
07ec77a1 WD |
2880 | /* |
2881 | * Change a given task's CPU affinity to the intersection of its current | |
2882 | * affinity mask and @subset_mask, writing the resulting mask to @new_mask | |
2883 | * and pointing @p->user_cpus_ptr to a copy of the old mask. | |
2884 | * If the resulting mask is empty, leave the affinity unchanged and return | |
2885 | * -EINVAL. | |
2886 | */ | |
2887 | static int restrict_cpus_allowed_ptr(struct task_struct *p, | |
2888 | struct cpumask *new_mask, | |
2889 | const struct cpumask *subset_mask) | |
2890 | { | |
2891 | struct cpumask *user_mask = NULL; | |
2892 | struct rq_flags rf; | |
2893 | struct rq *rq; | |
2894 | int err; | |
2895 | ||
2896 | if (!p->user_cpus_ptr) { | |
2897 | user_mask = kmalloc(cpumask_size(), GFP_KERNEL); | |
2898 | if (!user_mask) | |
2899 | return -ENOMEM; | |
2900 | } | |
2901 | ||
2902 | rq = task_rq_lock(p, &rf); | |
2903 | ||
2904 | /* | |
2905 | * Forcefully restricting the affinity of a deadline task is | |
2906 | * likely to cause problems, so fail and noisily override the | |
2907 | * mask entirely. | |
2908 | */ | |
2909 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { | |
2910 | err = -EPERM; | |
2911 | goto err_unlock; | |
2912 | } | |
2913 | ||
2914 | if (!cpumask_and(new_mask, &p->cpus_mask, subset_mask)) { | |
2915 | err = -EINVAL; | |
2916 | goto err_unlock; | |
2917 | } | |
2918 | ||
2919 | /* | |
2920 | * We're about to butcher the task affinity, so keep track of what | |
2921 | * the user asked for in case we're able to restore it later on. | |
2922 | */ | |
2923 | if (user_mask) { | |
2924 | cpumask_copy(user_mask, p->cpus_ptr); | |
2925 | p->user_cpus_ptr = user_mask; | |
2926 | } | |
2927 | ||
2928 | return __set_cpus_allowed_ptr_locked(p, new_mask, 0, rq, &rf); | |
2929 | ||
2930 | err_unlock: | |
2931 | task_rq_unlock(rq, p, &rf); | |
2932 | kfree(user_mask); | |
2933 | return err; | |
2934 | } | |
2935 | ||
2936 | /* | |
2937 | * Restrict the CPU affinity of task @p so that it is a subset of | |
2938 | * task_cpu_possible_mask() and point @p->user_cpu_ptr to a copy of the | |
2939 | * old affinity mask. If the resulting mask is empty, we warn and walk | |
2940 | * up the cpuset hierarchy until we find a suitable mask. | |
2941 | */ | |
2942 | void force_compatible_cpus_allowed_ptr(struct task_struct *p) | |
2943 | { | |
2944 | cpumask_var_t new_mask; | |
2945 | const struct cpumask *override_mask = task_cpu_possible_mask(p); | |
2946 | ||
2947 | alloc_cpumask_var(&new_mask, GFP_KERNEL); | |
2948 | ||
2949 | /* | |
2950 | * __migrate_task() can fail silently in the face of concurrent | |
2951 | * offlining of the chosen destination CPU, so take the hotplug | |
2952 | * lock to ensure that the migration succeeds. | |
2953 | */ | |
2954 | cpus_read_lock(); | |
2955 | if (!cpumask_available(new_mask)) | |
2956 | goto out_set_mask; | |
2957 | ||
2958 | if (!restrict_cpus_allowed_ptr(p, new_mask, override_mask)) | |
2959 | goto out_free_mask; | |
2960 | ||
2961 | /* | |
2962 | * We failed to find a valid subset of the affinity mask for the | |
2963 | * task, so override it based on its cpuset hierarchy. | |
2964 | */ | |
2965 | cpuset_cpus_allowed(p, new_mask); | |
2966 | override_mask = new_mask; | |
2967 | ||
2968 | out_set_mask: | |
2969 | if (printk_ratelimit()) { | |
2970 | printk_deferred("Overriding affinity for process %d (%s) to CPUs %*pbl\n", | |
2971 | task_pid_nr(p), p->comm, | |
2972 | cpumask_pr_args(override_mask)); | |
2973 | } | |
2974 | ||
2975 | WARN_ON(set_cpus_allowed_ptr(p, override_mask)); | |
2976 | out_free_mask: | |
2977 | cpus_read_unlock(); | |
2978 | free_cpumask_var(new_mask); | |
2979 | } | |
2980 | ||
2981 | static int | |
2982 | __sched_setaffinity(struct task_struct *p, const struct cpumask *mask); | |
2983 | ||
2984 | /* | |
2985 | * Restore the affinity of a task @p which was previously restricted by a | |
2986 | * call to force_compatible_cpus_allowed_ptr(). This will clear (and free) | |
2987 | * @p->user_cpus_ptr. | |
2988 | * | |
2989 | * It is the caller's responsibility to serialise this with any calls to | |
2990 | * force_compatible_cpus_allowed_ptr(@p). | |
2991 | */ | |
2992 | void relax_compatible_cpus_allowed_ptr(struct task_struct *p) | |
2993 | { | |
2994 | struct cpumask *user_mask = p->user_cpus_ptr; | |
2995 | unsigned long flags; | |
2996 | ||
2997 | /* | |
2998 | * Try to restore the old affinity mask. If this fails, then | |
2999 | * we free the mask explicitly to avoid it being inherited across | |
3000 | * a subsequent fork(). | |
3001 | */ | |
3002 | if (!user_mask || !__sched_setaffinity(p, user_mask)) | |
3003 | return; | |
3004 | ||
3005 | raw_spin_lock_irqsave(&p->pi_lock, flags); | |
3006 | user_mask = clear_user_cpus_ptr(p); | |
3007 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
3008 | ||
3009 | kfree(user_mask); | |
3010 | } | |
3011 | ||
dd41f596 | 3012 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 3013 | { |
e2912009 | 3014 | #ifdef CONFIG_SCHED_DEBUG |
2f064a59 PZ |
3015 | unsigned int state = READ_ONCE(p->__state); |
3016 | ||
e2912009 PZ |
3017 | /* |
3018 | * We should never call set_task_cpu() on a blocked task, | |
3019 | * ttwu() will sort out the placement. | |
3020 | */ | |
2f064a59 | 3021 | WARN_ON_ONCE(state != TASK_RUNNING && state != TASK_WAKING && !p->on_rq); |
0122ec5b | 3022 | |
3ea94de1 JP |
3023 | /* |
3024 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
3025 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
3026 | * time relying on p->on_rq. | |
3027 | */ | |
2f064a59 | 3028 | WARN_ON_ONCE(state == TASK_RUNNING && |
3ea94de1 JP |
3029 | p->sched_class == &fair_sched_class && |
3030 | (p->on_rq && !task_on_rq_migrating(p))); | |
3031 | ||
0122ec5b | 3032 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
3033 | /* |
3034 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
3035 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
3036 | * | |
3037 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 3038 | * see task_group(). |
6c6c54e1 PZ |
3039 | * |
3040 | * Furthermore, all task_rq users should acquire both locks, see | |
3041 | * task_rq_lock(). | |
3042 | */ | |
0122ec5b | 3043 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
9ef7e7e3 | 3044 | lockdep_is_held(__rq_lockp(task_rq(p))))); |
0122ec5b | 3045 | #endif |
4ff9083b PZ |
3046 | /* |
3047 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
3048 | */ | |
3049 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
3050 | |
3051 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
3052 | #endif |
3053 | ||
de1d7286 | 3054 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 3055 | |
0c69774e | 3056 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 3057 | if (p->sched_class->migrate_task_rq) |
1327237a | 3058 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 3059 | p->se.nr_migrations++; |
d7822b1e | 3060 | rseq_migrate(p); |
ff303e66 | 3061 | perf_event_task_migrate(p); |
0c69774e | 3062 | } |
dd41f596 IM |
3063 | |
3064 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
3065 | } |
3066 | ||
0ad4e3df | 3067 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
3068 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
3069 | { | |
da0c1e65 | 3070 | if (task_on_rq_queued(p)) { |
ac66f547 | 3071 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 3072 | struct rq_flags srf, drf; |
ac66f547 PZ |
3073 | |
3074 | src_rq = task_rq(p); | |
3075 | dst_rq = cpu_rq(cpu); | |
3076 | ||
8a8c69c3 PZ |
3077 | rq_pin_lock(src_rq, &srf); |
3078 | rq_pin_lock(dst_rq, &drf); | |
3079 | ||
ac66f547 PZ |
3080 | deactivate_task(src_rq, p, 0); |
3081 | set_task_cpu(p, cpu); | |
3082 | activate_task(dst_rq, p, 0); | |
3083 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
3084 | |
3085 | rq_unpin_lock(dst_rq, &drf); | |
3086 | rq_unpin_lock(src_rq, &srf); | |
3087 | ||
ac66f547 PZ |
3088 | } else { |
3089 | /* | |
3090 | * Task isn't running anymore; make it appear like we migrated | |
3091 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 3092 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
3093 | */ |
3094 | p->wake_cpu = cpu; | |
3095 | } | |
3096 | } | |
3097 | ||
3098 | struct migration_swap_arg { | |
3099 | struct task_struct *src_task, *dst_task; | |
3100 | int src_cpu, dst_cpu; | |
3101 | }; | |
3102 | ||
3103 | static int migrate_swap_stop(void *data) | |
3104 | { | |
3105 | struct migration_swap_arg *arg = data; | |
3106 | struct rq *src_rq, *dst_rq; | |
3107 | int ret = -EAGAIN; | |
3108 | ||
62694cd5 PZ |
3109 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
3110 | return -EAGAIN; | |
3111 | ||
ac66f547 PZ |
3112 | src_rq = cpu_rq(arg->src_cpu); |
3113 | dst_rq = cpu_rq(arg->dst_cpu); | |
3114 | ||
74602315 PZ |
3115 | double_raw_lock(&arg->src_task->pi_lock, |
3116 | &arg->dst_task->pi_lock); | |
ac66f547 | 3117 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 3118 | |
ac66f547 PZ |
3119 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
3120 | goto unlock; | |
3121 | ||
3122 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
3123 | goto unlock; | |
3124 | ||
3bd37062 | 3125 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
3126 | goto unlock; |
3127 | ||
3bd37062 | 3128 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
3129 | goto unlock; |
3130 | ||
3131 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
3132 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
3133 | ||
3134 | ret = 0; | |
3135 | ||
3136 | unlock: | |
3137 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
3138 | raw_spin_unlock(&arg->dst_task->pi_lock); |
3139 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
3140 | |
3141 | return ret; | |
3142 | } | |
3143 | ||
3144 | /* | |
3145 | * Cross migrate two tasks | |
3146 | */ | |
0ad4e3df SD |
3147 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
3148 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
3149 | { |
3150 | struct migration_swap_arg arg; | |
3151 | int ret = -EINVAL; | |
3152 | ||
ac66f547 PZ |
3153 | arg = (struct migration_swap_arg){ |
3154 | .src_task = cur, | |
0ad4e3df | 3155 | .src_cpu = curr_cpu, |
ac66f547 | 3156 | .dst_task = p, |
0ad4e3df | 3157 | .dst_cpu = target_cpu, |
ac66f547 PZ |
3158 | }; |
3159 | ||
3160 | if (arg.src_cpu == arg.dst_cpu) | |
3161 | goto out; | |
3162 | ||
6acce3ef PZ |
3163 | /* |
3164 | * These three tests are all lockless; this is OK since all of them | |
3165 | * will be re-checked with proper locks held further down the line. | |
3166 | */ | |
ac66f547 PZ |
3167 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
3168 | goto out; | |
3169 | ||
3bd37062 | 3170 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
3171 | goto out; |
3172 | ||
3bd37062 | 3173 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
3174 | goto out; |
3175 | ||
286549dc | 3176 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
3177 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
3178 | ||
3179 | out: | |
ac66f547 PZ |
3180 | return ret; |
3181 | } | |
0ad4e3df | 3182 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 3183 | |
1da177e4 LT |
3184 | /* |
3185 | * wait_task_inactive - wait for a thread to unschedule. | |
3186 | * | |
85ba2d86 RM |
3187 | * If @match_state is nonzero, it's the @p->state value just checked and |
3188 | * not expected to change. If it changes, i.e. @p might have woken up, | |
3189 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
3190 | * we return a positive number (its total switch count). If a second call | |
3191 | * a short while later returns the same number, the caller can be sure that | |
3192 | * @p has remained unscheduled the whole time. | |
3193 | * | |
1da177e4 LT |
3194 | * The caller must ensure that the task *will* unschedule sometime soon, |
3195 | * else this function might spin for a *long* time. This function can't | |
3196 | * be called with interrupts off, or it may introduce deadlock with | |
3197 | * smp_call_function() if an IPI is sent by the same process we are | |
3198 | * waiting to become inactive. | |
3199 | */ | |
2f064a59 | 3200 | unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state) |
1da177e4 | 3201 | { |
da0c1e65 | 3202 | int running, queued; |
eb580751 | 3203 | struct rq_flags rf; |
85ba2d86 | 3204 | unsigned long ncsw; |
70b97a7f | 3205 | struct rq *rq; |
1da177e4 | 3206 | |
3a5c359a AK |
3207 | for (;;) { |
3208 | /* | |
3209 | * We do the initial early heuristics without holding | |
3210 | * any task-queue locks at all. We'll only try to get | |
3211 | * the runqueue lock when things look like they will | |
3212 | * work out! | |
3213 | */ | |
3214 | rq = task_rq(p); | |
fa490cfd | 3215 | |
3a5c359a AK |
3216 | /* |
3217 | * If the task is actively running on another CPU | |
3218 | * still, just relax and busy-wait without holding | |
3219 | * any locks. | |
3220 | * | |
3221 | * NOTE! Since we don't hold any locks, it's not | |
3222 | * even sure that "rq" stays as the right runqueue! | |
3223 | * But we don't care, since "task_running()" will | |
3224 | * return false if the runqueue has changed and p | |
3225 | * is actually now running somewhere else! | |
3226 | */ | |
85ba2d86 | 3227 | while (task_running(rq, p)) { |
2f064a59 | 3228 | if (match_state && unlikely(READ_ONCE(p->__state) != match_state)) |
85ba2d86 | 3229 | return 0; |
3a5c359a | 3230 | cpu_relax(); |
85ba2d86 | 3231 | } |
fa490cfd | 3232 | |
3a5c359a AK |
3233 | /* |
3234 | * Ok, time to look more closely! We need the rq | |
3235 | * lock now, to be *sure*. If we're wrong, we'll | |
3236 | * just go back and repeat. | |
3237 | */ | |
eb580751 | 3238 | rq = task_rq_lock(p, &rf); |
27a9da65 | 3239 | trace_sched_wait_task(p); |
3a5c359a | 3240 | running = task_running(rq, p); |
da0c1e65 | 3241 | queued = task_on_rq_queued(p); |
85ba2d86 | 3242 | ncsw = 0; |
2f064a59 | 3243 | if (!match_state || READ_ONCE(p->__state) == match_state) |
93dcf55f | 3244 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 3245 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 3246 | |
85ba2d86 RM |
3247 | /* |
3248 | * If it changed from the expected state, bail out now. | |
3249 | */ | |
3250 | if (unlikely(!ncsw)) | |
3251 | break; | |
3252 | ||
3a5c359a AK |
3253 | /* |
3254 | * Was it really running after all now that we | |
3255 | * checked with the proper locks actually held? | |
3256 | * | |
3257 | * Oops. Go back and try again.. | |
3258 | */ | |
3259 | if (unlikely(running)) { | |
3260 | cpu_relax(); | |
3261 | continue; | |
3262 | } | |
fa490cfd | 3263 | |
3a5c359a AK |
3264 | /* |
3265 | * It's not enough that it's not actively running, | |
3266 | * it must be off the runqueue _entirely_, and not | |
3267 | * preempted! | |
3268 | * | |
80dd99b3 | 3269 | * So if it was still runnable (but just not actively |
3a5c359a AK |
3270 | * running right now), it's preempted, and we should |
3271 | * yield - it could be a while. | |
3272 | */ | |
da0c1e65 | 3273 | if (unlikely(queued)) { |
8b0e1953 | 3274 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
3275 | |
3276 | set_current_state(TASK_UNINTERRUPTIBLE); | |
c33627e9 | 3277 | schedule_hrtimeout(&to, HRTIMER_MODE_REL_HARD); |
3a5c359a AK |
3278 | continue; |
3279 | } | |
fa490cfd | 3280 | |
3a5c359a AK |
3281 | /* |
3282 | * Ahh, all good. It wasn't running, and it wasn't | |
3283 | * runnable, which means that it will never become | |
3284 | * running in the future either. We're all done! | |
3285 | */ | |
3286 | break; | |
3287 | } | |
85ba2d86 RM |
3288 | |
3289 | return ncsw; | |
1da177e4 LT |
3290 | } |
3291 | ||
3292 | /*** | |
3293 | * kick_process - kick a running thread to enter/exit the kernel | |
3294 | * @p: the to-be-kicked thread | |
3295 | * | |
3296 | * Cause a process which is running on another CPU to enter | |
3297 | * kernel-mode, without any delay. (to get signals handled.) | |
3298 | * | |
25985edc | 3299 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
3300 | * because all it wants to ensure is that the remote task enters |
3301 | * the kernel. If the IPI races and the task has been migrated | |
3302 | * to another CPU then no harm is done and the purpose has been | |
3303 | * achieved as well. | |
3304 | */ | |
36c8b586 | 3305 | void kick_process(struct task_struct *p) |
1da177e4 LT |
3306 | { |
3307 | int cpu; | |
3308 | ||
3309 | preempt_disable(); | |
3310 | cpu = task_cpu(p); | |
3311 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
3312 | smp_send_reschedule(cpu); | |
3313 | preempt_enable(); | |
3314 | } | |
b43e3521 | 3315 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 3316 | |
30da688e | 3317 | /* |
3bd37062 | 3318 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
3319 | * |
3320 | * A few notes on cpu_active vs cpu_online: | |
3321 | * | |
3322 | * - cpu_active must be a subset of cpu_online | |
3323 | * | |
97fb7a0a | 3324 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 3325 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 3326 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
3327 | * see it. |
3328 | * | |
d1ccc66d | 3329 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 3330 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 3331 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
3332 | * off. |
3333 | * | |
3334 | * This means that fallback selection must not select !active CPUs. | |
3335 | * And can assume that any active CPU must be online. Conversely | |
3336 | * select_task_rq() below may allow selection of !active CPUs in order | |
3337 | * to satisfy the above rules. | |
30da688e | 3338 | */ |
5da9a0fb PZ |
3339 | static int select_fallback_rq(int cpu, struct task_struct *p) |
3340 | { | |
aa00d89c TC |
3341 | int nid = cpu_to_node(cpu); |
3342 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
3343 | enum { cpuset, possible, fail } state = cpuset; |
3344 | int dest_cpu; | |
5da9a0fb | 3345 | |
aa00d89c | 3346 | /* |
d1ccc66d IM |
3347 | * If the node that the CPU is on has been offlined, cpu_to_node() |
3348 | * will return -1. There is no CPU on the node, and we should | |
3349 | * select the CPU on the other node. | |
aa00d89c TC |
3350 | */ |
3351 | if (nid != -1) { | |
3352 | nodemask = cpumask_of_node(nid); | |
3353 | ||
3354 | /* Look for allowed, online CPU in same node. */ | |
3355 | for_each_cpu(dest_cpu, nodemask) { | |
9ae606bc | 3356 | if (is_cpu_allowed(p, dest_cpu)) |
aa00d89c TC |
3357 | return dest_cpu; |
3358 | } | |
2baab4e9 | 3359 | } |
5da9a0fb | 3360 | |
2baab4e9 PZ |
3361 | for (;;) { |
3362 | /* Any allowed, online CPU? */ | |
3bd37062 | 3363 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 3364 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 3365 | continue; |
175f0e25 | 3366 | |
2baab4e9 PZ |
3367 | goto out; |
3368 | } | |
5da9a0fb | 3369 | |
e73e85f0 | 3370 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
3371 | switch (state) { |
3372 | case cpuset: | |
97c0054d | 3373 | if (cpuset_cpus_allowed_fallback(p)) { |
e73e85f0 ON |
3374 | state = possible; |
3375 | break; | |
3376 | } | |
df561f66 | 3377 | fallthrough; |
2baab4e9 | 3378 | case possible: |
af449901 PZ |
3379 | /* |
3380 | * XXX When called from select_task_rq() we only | |
3381 | * hold p->pi_lock and again violate locking order. | |
3382 | * | |
3383 | * More yuck to audit. | |
3384 | */ | |
9ae606bc | 3385 | do_set_cpus_allowed(p, task_cpu_possible_mask(p)); |
2baab4e9 PZ |
3386 | state = fail; |
3387 | break; | |
2baab4e9 PZ |
3388 | case fail: |
3389 | BUG(); | |
3390 | break; | |
3391 | } | |
3392 | } | |
3393 | ||
3394 | out: | |
3395 | if (state != cpuset) { | |
3396 | /* | |
3397 | * Don't tell them about moving exiting tasks or | |
3398 | * kernel threads (both mm NULL), since they never | |
3399 | * leave kernel. | |
3400 | */ | |
3401 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 3402 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
3403 | task_pid_nr(p), p->comm, cpu); |
3404 | } | |
5da9a0fb PZ |
3405 | } |
3406 | ||
3407 | return dest_cpu; | |
3408 | } | |
3409 | ||
e2912009 | 3410 | /* |
3bd37062 | 3411 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 3412 | */ |
970b13ba | 3413 | static inline |
3aef1551 | 3414 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 3415 | { |
cbce1a68 PZ |
3416 | lockdep_assert_held(&p->pi_lock); |
3417 | ||
af449901 | 3418 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 3419 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 3420 | else |
3bd37062 | 3421 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
3422 | |
3423 | /* | |
3424 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 3425 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 3426 | * CPU. |
e2912009 PZ |
3427 | * |
3428 | * Since this is common to all placement strategies, this lives here. | |
3429 | * | |
3430 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
3431 | * not worry about this generic constraint ] | |
3432 | */ | |
7af443ee | 3433 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 3434 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
3435 | |
3436 | return cpu; | |
970b13ba | 3437 | } |
09a40af5 | 3438 | |
f5832c19 NP |
3439 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
3440 | { | |
ded467dc | 3441 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
3442 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
3443 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
3444 | ||
3445 | if (stop) { | |
3446 | /* | |
3447 | * Make it appear like a SCHED_FIFO task, its something | |
3448 | * userspace knows about and won't get confused about. | |
3449 | * | |
3450 | * Also, it will make PI more or less work without too | |
3451 | * much confusion -- but then, stop work should not | |
3452 | * rely on PI working anyway. | |
3453 | */ | |
3454 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
3455 | ||
3456 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
3457 | |
3458 | /* | |
3459 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
3460 | * adjust the effective priority of a task. As a result, | |
3461 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
3462 | * which can then trigger wakeups of the stop thread to push | |
3463 | * around the current task. | |
3464 | * | |
3465 | * The stop task itself will never be part of the PI-chain, it | |
3466 | * never blocks, therefore that ->pi_lock recursion is safe. | |
3467 | * Tell lockdep about this by placing the stop->pi_lock in its | |
3468 | * own class. | |
3469 | */ | |
3470 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
3471 | } |
3472 | ||
3473 | cpu_rq(cpu)->stop = stop; | |
3474 | ||
3475 | if (old_stop) { | |
3476 | /* | |
3477 | * Reset it back to a normal scheduling class so that | |
3478 | * it can die in pieces. | |
3479 | */ | |
3480 | old_stop->sched_class = &rt_sched_class; | |
3481 | } | |
3482 | } | |
3483 | ||
74d862b6 | 3484 | #else /* CONFIG_SMP */ |
25834c73 PZ |
3485 | |
3486 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
3487 | const struct cpumask *new_mask, |
3488 | u32 flags) | |
25834c73 PZ |
3489 | { |
3490 | return set_cpus_allowed_ptr(p, new_mask); | |
3491 | } | |
3492 | ||
af449901 PZ |
3493 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
3494 | ||
3015ef4b TG |
3495 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
3496 | { | |
3497 | return false; | |
3498 | } | |
3499 | ||
74d862b6 | 3500 | #endif /* !CONFIG_SMP */ |
970b13ba | 3501 | |
d7c01d27 | 3502 | static void |
b84cb5df | 3503 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 3504 | { |
4fa8d299 | 3505 | struct rq *rq; |
b84cb5df | 3506 | |
4fa8d299 JP |
3507 | if (!schedstat_enabled()) |
3508 | return; | |
3509 | ||
3510 | rq = this_rq(); | |
d7c01d27 | 3511 | |
4fa8d299 JP |
3512 | #ifdef CONFIG_SMP |
3513 | if (cpu == rq->cpu) { | |
b85c8b71 | 3514 | __schedstat_inc(rq->ttwu_local); |
ceeadb83 | 3515 | __schedstat_inc(p->stats.nr_wakeups_local); |
d7c01d27 PZ |
3516 | } else { |
3517 | struct sched_domain *sd; | |
3518 | ||
ceeadb83 | 3519 | __schedstat_inc(p->stats.nr_wakeups_remote); |
057f3fad | 3520 | rcu_read_lock(); |
4fa8d299 | 3521 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 3522 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 3523 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
3524 | break; |
3525 | } | |
3526 | } | |
057f3fad | 3527 | rcu_read_unlock(); |
d7c01d27 | 3528 | } |
f339b9dc PZ |
3529 | |
3530 | if (wake_flags & WF_MIGRATED) | |
ceeadb83 | 3531 | __schedstat_inc(p->stats.nr_wakeups_migrate); |
d7c01d27 PZ |
3532 | #endif /* CONFIG_SMP */ |
3533 | ||
b85c8b71 | 3534 | __schedstat_inc(rq->ttwu_count); |
ceeadb83 | 3535 | __schedstat_inc(p->stats.nr_wakeups); |
d7c01d27 PZ |
3536 | |
3537 | if (wake_flags & WF_SYNC) | |
ceeadb83 | 3538 | __schedstat_inc(p->stats.nr_wakeups_sync); |
d7c01d27 PZ |
3539 | } |
3540 | ||
23f41eeb PZ |
3541 | /* |
3542 | * Mark the task runnable and perform wakeup-preemption. | |
3543 | */ | |
e7904a28 | 3544 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3545 | struct rq_flags *rf) |
9ed3811a | 3546 | { |
9ed3811a | 3547 | check_preempt_curr(rq, p, wake_flags); |
2f064a59 | 3548 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fbd705a0 PZ |
3549 | trace_sched_wakeup(p); |
3550 | ||
9ed3811a | 3551 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
3552 | if (p->sched_class->task_woken) { |
3553 | /* | |
b19a888c | 3554 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 3555 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 3556 | */ |
d8ac8971 | 3557 | rq_unpin_lock(rq, rf); |
9ed3811a | 3558 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3559 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 3560 | } |
9ed3811a | 3561 | |
e69c6341 | 3562 | if (rq->idle_stamp) { |
78becc27 | 3563 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 3564 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 3565 | |
abfafa54 JL |
3566 | update_avg(&rq->avg_idle, delta); |
3567 | ||
3568 | if (rq->avg_idle > max) | |
9ed3811a | 3569 | rq->avg_idle = max; |
abfafa54 | 3570 | |
94aafc3e PZ |
3571 | rq->wake_stamp = jiffies; |
3572 | rq->wake_avg_idle = rq->avg_idle / 2; | |
3573 | ||
9ed3811a TH |
3574 | rq->idle_stamp = 0; |
3575 | } | |
3576 | #endif | |
3577 | } | |
3578 | ||
c05fbafb | 3579 | static void |
e7904a28 | 3580 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3581 | struct rq_flags *rf) |
c05fbafb | 3582 | { |
77558e4d | 3583 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 3584 | |
5cb9eaa3 | 3585 | lockdep_assert_rq_held(rq); |
cbce1a68 | 3586 | |
c05fbafb PZ |
3587 | if (p->sched_contributes_to_load) |
3588 | rq->nr_uninterruptible--; | |
b5179ac7 | 3589 | |
dbfb089d | 3590 | #ifdef CONFIG_SMP |
b5179ac7 | 3591 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 3592 | en_flags |= ENQUEUE_MIGRATED; |
ec618b84 | 3593 | else |
c05fbafb | 3594 | #endif |
ec618b84 PZ |
3595 | if (p->in_iowait) { |
3596 | delayacct_blkio_end(p); | |
3597 | atomic_dec(&task_rq(p)->nr_iowait); | |
3598 | } | |
c05fbafb | 3599 | |
1b174a2c | 3600 | activate_task(rq, p, en_flags); |
d8ac8971 | 3601 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
3602 | } |
3603 | ||
3604 | /* | |
58877d34 PZ |
3605 | * Consider @p being inside a wait loop: |
3606 | * | |
3607 | * for (;;) { | |
3608 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
3609 | * | |
3610 | * if (CONDITION) | |
3611 | * break; | |
3612 | * | |
3613 | * schedule(); | |
3614 | * } | |
3615 | * __set_current_state(TASK_RUNNING); | |
3616 | * | |
3617 | * between set_current_state() and schedule(). In this case @p is still | |
3618 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3619 | * an atomic manner. | |
3620 | * | |
3621 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3622 | * then schedule() must still happen and p->state can be changed to | |
3623 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3624 | * need to do a full wakeup with enqueue. | |
3625 | * | |
3626 | * Returns: %true when the wakeup is done, | |
3627 | * %false otherwise. | |
c05fbafb | 3628 | */ |
58877d34 | 3629 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3630 | { |
eb580751 | 3631 | struct rq_flags rf; |
c05fbafb PZ |
3632 | struct rq *rq; |
3633 | int ret = 0; | |
3634 | ||
eb580751 | 3635 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3636 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3637 | /* check_preempt_curr() may use rq clock */ |
3638 | update_rq_clock(rq); | |
d8ac8971 | 3639 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3640 | ret = 1; |
3641 | } | |
eb580751 | 3642 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3643 | |
3644 | return ret; | |
3645 | } | |
3646 | ||
317f3941 | 3647 | #ifdef CONFIG_SMP |
a1488664 | 3648 | void sched_ttwu_pending(void *arg) |
317f3941 | 3649 | { |
a1488664 | 3650 | struct llist_node *llist = arg; |
317f3941 | 3651 | struct rq *rq = this_rq(); |
73215849 | 3652 | struct task_struct *p, *t; |
d8ac8971 | 3653 | struct rq_flags rf; |
317f3941 | 3654 | |
e3baac47 PZ |
3655 | if (!llist) |
3656 | return; | |
3657 | ||
126c2092 PZ |
3658 | /* |
3659 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
3660 | * Races such that false-negatives are possible, since they | |
3661 | * are shorter lived that false-positives would be. | |
3662 | */ | |
3663 | WRITE_ONCE(rq->ttwu_pending, 0); | |
3664 | ||
8a8c69c3 | 3665 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3666 | update_rq_clock(rq); |
317f3941 | 3667 | |
8c4890d1 | 3668 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3669 | if (WARN_ON_ONCE(p->on_cpu)) |
3670 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3671 | ||
3672 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3673 | set_task_cpu(p, cpu_of(rq)); | |
3674 | ||
73215849 | 3675 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3676 | } |
317f3941 | 3677 | |
8a8c69c3 | 3678 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3679 | } |
3680 | ||
b2a02fc4 | 3681 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3682 | { |
b2a02fc4 | 3683 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3684 | |
b2a02fc4 PZ |
3685 | if (!set_nr_if_polling(rq->idle)) |
3686 | arch_send_call_function_single_ipi(cpu); | |
3687 | else | |
3688 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3689 | } |
3690 | ||
2ebb1771 MG |
3691 | /* |
3692 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3693 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3694 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3695 | * of the wakeup instead of the waker. | |
3696 | */ | |
3697 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3698 | { |
e3baac47 PZ |
3699 | struct rq *rq = cpu_rq(cpu); |
3700 | ||
b7e7ade3 PZ |
3701 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3702 | ||
126c2092 | 3703 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3704 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3705 | } |
d6aa8f85 | 3706 | |
f6be8af1 CL |
3707 | void wake_up_if_idle(int cpu) |
3708 | { | |
3709 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3710 | struct rq_flags rf; |
f6be8af1 | 3711 | |
fd7de1e8 AL |
3712 | rcu_read_lock(); |
3713 | ||
3714 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3715 | goto out; | |
f6be8af1 | 3716 | |
8850cb66 PZ |
3717 | rq_lock_irqsave(rq, &rf); |
3718 | if (is_idle_task(rq->curr)) | |
3719 | resched_curr(rq); | |
3720 | /* Else CPU is not idle, do nothing here: */ | |
3721 | rq_unlock_irqrestore(rq, &rf); | |
fd7de1e8 AL |
3722 | |
3723 | out: | |
3724 | rcu_read_unlock(); | |
f6be8af1 CL |
3725 | } |
3726 | ||
39be3501 | 3727 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 | 3728 | { |
42dc938a VD |
3729 | if (this_cpu == that_cpu) |
3730 | return true; | |
3731 | ||
518cd623 PZ |
3732 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); |
3733 | } | |
c6e7bd7a | 3734 | |
2ebb1771 MG |
3735 | static inline bool ttwu_queue_cond(int cpu, int wake_flags) |
3736 | { | |
5ba2ffba PZ |
3737 | /* |
3738 | * Do not complicate things with the async wake_list while the CPU is | |
3739 | * in hotplug state. | |
3740 | */ | |
3741 | if (!cpu_active(cpu)) | |
3742 | return false; | |
3743 | ||
2ebb1771 MG |
3744 | /* |
3745 | * If the CPU does not share cache, then queue the task on the | |
3746 | * remote rqs wakelist to avoid accessing remote data. | |
3747 | */ | |
3748 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3749 | return true; | |
3750 | ||
3751 | /* | |
3752 | * If the task is descheduling and the only running task on the | |
3753 | * CPU then use the wakelist to offload the task activation to | |
3754 | * the soon-to-be-idle CPU as the current CPU is likely busy. | |
3755 | * nr_running is checked to avoid unnecessary task stacking. | |
3756 | */ | |
739f70b4 | 3757 | if ((wake_flags & WF_ON_CPU) && cpu_rq(cpu)->nr_running <= 1) |
2ebb1771 MG |
3758 | return true; |
3759 | ||
3760 | return false; | |
3761 | } | |
3762 | ||
3763 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3764 | { |
2ebb1771 | 3765 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(cpu, wake_flags)) { |
b6e13e85 PZ |
3766 | if (WARN_ON_ONCE(cpu == smp_processor_id())) |
3767 | return false; | |
3768 | ||
c6e7bd7a | 3769 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3770 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3771 | return true; |
3772 | } | |
3773 | ||
3774 | return false; | |
3775 | } | |
58877d34 PZ |
3776 | |
3777 | #else /* !CONFIG_SMP */ | |
3778 | ||
3779 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3780 | { | |
3781 | return false; | |
3782 | } | |
3783 | ||
d6aa8f85 | 3784 | #endif /* CONFIG_SMP */ |
317f3941 | 3785 | |
b5179ac7 | 3786 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3787 | { |
3788 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3789 | struct rq_flags rf; |
c05fbafb | 3790 | |
2ebb1771 | 3791 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3792 | return; |
317f3941 | 3793 | |
8a8c69c3 | 3794 | rq_lock(rq, &rf); |
77558e4d | 3795 | update_rq_clock(rq); |
d8ac8971 | 3796 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3797 | rq_unlock(rq, &rf); |
9ed3811a TH |
3798 | } |
3799 | ||
43295d73 TG |
3800 | /* |
3801 | * Invoked from try_to_wake_up() to check whether the task can be woken up. | |
3802 | * | |
3803 | * The caller holds p::pi_lock if p != current or has preemption | |
3804 | * disabled when p == current. | |
5f220be2 TG |
3805 | * |
3806 | * The rules of PREEMPT_RT saved_state: | |
3807 | * | |
3808 | * The related locking code always holds p::pi_lock when updating | |
3809 | * p::saved_state, which means the code is fully serialized in both cases. | |
3810 | * | |
3811 | * The lock wait and lock wakeups happen via TASK_RTLOCK_WAIT. No other | |
3812 | * bits set. This allows to distinguish all wakeup scenarios. | |
43295d73 TG |
3813 | */ |
3814 | static __always_inline | |
3815 | bool ttwu_state_match(struct task_struct *p, unsigned int state, int *success) | |
3816 | { | |
5f220be2 TG |
3817 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)) { |
3818 | WARN_ON_ONCE((state & TASK_RTLOCK_WAIT) && | |
3819 | state != TASK_RTLOCK_WAIT); | |
3820 | } | |
3821 | ||
43295d73 TG |
3822 | if (READ_ONCE(p->__state) & state) { |
3823 | *success = 1; | |
3824 | return true; | |
3825 | } | |
5f220be2 TG |
3826 | |
3827 | #ifdef CONFIG_PREEMPT_RT | |
3828 | /* | |
3829 | * Saved state preserves the task state across blocking on | |
3830 | * an RT lock. If the state matches, set p::saved_state to | |
3831 | * TASK_RUNNING, but do not wake the task because it waits | |
3832 | * for a lock wakeup. Also indicate success because from | |
3833 | * the regular waker's point of view this has succeeded. | |
3834 | * | |
3835 | * After acquiring the lock the task will restore p::__state | |
3836 | * from p::saved_state which ensures that the regular | |
3837 | * wakeup is not lost. The restore will also set | |
3838 | * p::saved_state to TASK_RUNNING so any further tests will | |
3839 | * not result in false positives vs. @success | |
3840 | */ | |
3841 | if (p->saved_state & state) { | |
3842 | p->saved_state = TASK_RUNNING; | |
3843 | *success = 1; | |
3844 | } | |
3845 | #endif | |
43295d73 TG |
3846 | return false; |
3847 | } | |
3848 | ||
8643cda5 PZ |
3849 | /* |
3850 | * Notes on Program-Order guarantees on SMP systems. | |
3851 | * | |
3852 | * MIGRATION | |
3853 | * | |
3854 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3855 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3856 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3857 | * |
3858 | * For migration (of runnable tasks) this is provided by the following means: | |
3859 | * | |
3860 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3861 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3862 | * rq(c1)->lock (if not at the same time, then in that order). | |
3863 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3864 | * | |
7696f991 | 3865 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3866 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3867 | * |
3868 | * Example: | |
3869 | * | |
3870 | * CPU0 CPU1 CPU2 | |
3871 | * | |
3872 | * LOCK rq(0)->lock | |
3873 | * sched-out X | |
3874 | * sched-in Y | |
3875 | * UNLOCK rq(0)->lock | |
3876 | * | |
3877 | * LOCK rq(0)->lock // orders against CPU0 | |
3878 | * dequeue X | |
3879 | * UNLOCK rq(0)->lock | |
3880 | * | |
3881 | * LOCK rq(1)->lock | |
3882 | * enqueue X | |
3883 | * UNLOCK rq(1)->lock | |
3884 | * | |
3885 | * LOCK rq(1)->lock // orders against CPU2 | |
3886 | * sched-out Z | |
3887 | * sched-in X | |
3888 | * UNLOCK rq(1)->lock | |
3889 | * | |
3890 | * | |
3891 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3892 | * | |
3893 | * For blocking we (obviously) need to provide the same guarantee as for | |
3894 | * migration. However the means are completely different as there is no lock | |
3895 | * chain to provide order. Instead we do: | |
3896 | * | |
58877d34 PZ |
3897 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3898 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3899 | * |
3900 | * Example: | |
3901 | * | |
3902 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
3903 | * | |
3904 | * LOCK rq(0)->lock LOCK X->pi_lock | |
3905 | * dequeue X | |
3906 | * sched-out X | |
3907 | * smp_store_release(X->on_cpu, 0); | |
3908 | * | |
1f03e8d2 | 3909 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
3910 | * X->state = WAKING |
3911 | * set_task_cpu(X,2) | |
3912 | * | |
3913 | * LOCK rq(2)->lock | |
3914 | * enqueue X | |
3915 | * X->state = RUNNING | |
3916 | * UNLOCK rq(2)->lock | |
3917 | * | |
3918 | * LOCK rq(2)->lock // orders against CPU1 | |
3919 | * sched-out Z | |
3920 | * sched-in X | |
3921 | * UNLOCK rq(2)->lock | |
3922 | * | |
3923 | * UNLOCK X->pi_lock | |
3924 | * UNLOCK rq(0)->lock | |
3925 | * | |
3926 | * | |
7696f991 AP |
3927 | * However, for wakeups there is a second guarantee we must provide, namely we |
3928 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
3929 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
3930 | */ |
3931 | ||
9ed3811a | 3932 | /** |
1da177e4 | 3933 | * try_to_wake_up - wake up a thread |
9ed3811a | 3934 | * @p: the thread to be awakened |
1da177e4 | 3935 | * @state: the mask of task states that can be woken |
9ed3811a | 3936 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 3937 | * |
58877d34 PZ |
3938 | * Conceptually does: |
3939 | * | |
3940 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 3941 | * |
a2250238 PZ |
3942 | * If the task was not queued/runnable, also place it back on a runqueue. |
3943 | * | |
58877d34 PZ |
3944 | * This function is atomic against schedule() which would dequeue the task. |
3945 | * | |
3946 | * It issues a full memory barrier before accessing @p->state, see the comment | |
3947 | * with set_current_state(). | |
a2250238 | 3948 | * |
58877d34 | 3949 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 3950 | * |
58877d34 PZ |
3951 | * Relies on p->pi_lock stabilizing: |
3952 | * - p->sched_class | |
3953 | * - p->cpus_ptr | |
3954 | * - p->sched_task_group | |
3955 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
3956 | * | |
3957 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
3958 | * Takes rq->lock in: | |
3959 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
3960 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
3961 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
3962 | * | |
3963 | * As a consequence we race really badly with just about everything. See the | |
3964 | * many memory barriers and their comments for details. | |
7696f991 | 3965 | * |
a2250238 PZ |
3966 | * Return: %true if @p->state changes (an actual wakeup was done), |
3967 | * %false otherwise. | |
1da177e4 | 3968 | */ |
e4a52bcb PZ |
3969 | static int |
3970 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 3971 | { |
1da177e4 | 3972 | unsigned long flags; |
c05fbafb | 3973 | int cpu, success = 0; |
2398f2c6 | 3974 | |
e3d85487 | 3975 | preempt_disable(); |
aacedf26 PZ |
3976 | if (p == current) { |
3977 | /* | |
3978 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
3979 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 3980 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
3981 | * without taking any locks. |
3982 | * | |
3983 | * In particular: | |
3984 | * - we rely on Program-Order guarantees for all the ordering, | |
3985 | * - we're serialized against set_special_state() by virtue of | |
3986 | * it disabling IRQs (this allows not taking ->pi_lock). | |
3987 | */ | |
43295d73 | 3988 | if (!ttwu_state_match(p, state, &success)) |
e3d85487 | 3989 | goto out; |
aacedf26 | 3990 | |
aacedf26 | 3991 | trace_sched_waking(p); |
2f064a59 | 3992 | WRITE_ONCE(p->__state, TASK_RUNNING); |
aacedf26 PZ |
3993 | trace_sched_wakeup(p); |
3994 | goto out; | |
3995 | } | |
3996 | ||
e0acd0a6 ON |
3997 | /* |
3998 | * If we are going to wake up a thread waiting for CONDITION we | |
3999 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
4000 | * reordered with p->state check below. This pairs with smp_store_mb() |
4001 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 4002 | */ |
013fdb80 | 4003 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 4004 | smp_mb__after_spinlock(); |
43295d73 | 4005 | if (!ttwu_state_match(p, state, &success)) |
aacedf26 | 4006 | goto unlock; |
1da177e4 | 4007 | |
fbd705a0 PZ |
4008 | trace_sched_waking(p); |
4009 | ||
135e8c92 BS |
4010 | /* |
4011 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
4012 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
4013 | * in smp_cond_load_acquire() below. | |
4014 | * | |
3d85b270 AP |
4015 | * sched_ttwu_pending() try_to_wake_up() |
4016 | * STORE p->on_rq = 1 LOAD p->state | |
4017 | * UNLOCK rq->lock | |
4018 | * | |
4019 | * __schedule() (switch to task 'p') | |
4020 | * LOCK rq->lock smp_rmb(); | |
4021 | * smp_mb__after_spinlock(); | |
4022 | * UNLOCK rq->lock | |
135e8c92 BS |
4023 | * |
4024 | * [task p] | |
3d85b270 | 4025 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 4026 | * |
3d85b270 AP |
4027 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4028 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
4029 | * |
4030 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
4031 | */ |
4032 | smp_rmb(); | |
58877d34 | 4033 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 4034 | goto unlock; |
1da177e4 | 4035 | |
1da177e4 | 4036 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
4037 | /* |
4038 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
4039 | * possible to, falsely, observe p->on_cpu == 0. | |
4040 | * | |
4041 | * One must be running (->on_cpu == 1) in order to remove oneself | |
4042 | * from the runqueue. | |
4043 | * | |
3d85b270 AP |
4044 | * __schedule() (switch to task 'p') try_to_wake_up() |
4045 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
4046 | * UNLOCK rq->lock | |
4047 | * | |
4048 | * __schedule() (put 'p' to sleep) | |
4049 | * LOCK rq->lock smp_rmb(); | |
4050 | * smp_mb__after_spinlock(); | |
4051 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 4052 | * |
3d85b270 AP |
4053 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4054 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
4055 | * |
4056 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
4057 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
4058 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 4059 | */ |
dbfb089d PZ |
4060 | smp_acquire__after_ctrl_dep(); |
4061 | ||
4062 | /* | |
4063 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
4064 | * == 0), which means we need to do an enqueue, change p->state to | |
4065 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
4066 | * enqueue, such as ttwu_queue_wakelist(). | |
4067 | */ | |
2f064a59 | 4068 | WRITE_ONCE(p->__state, TASK_WAKING); |
ecf7d01c | 4069 | |
c6e7bd7a PZ |
4070 | /* |
4071 | * If the owning (remote) CPU is still in the middle of schedule() with | |
4072 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
4073 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
4074 | * let the waker make forward progress. This is safe because IRQs are | |
4075 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
4076 | * |
4077 | * Ensure we load task_cpu(p) after p->on_cpu: | |
4078 | * | |
4079 | * set_task_cpu(p, cpu); | |
4080 | * STORE p->cpu = @cpu | |
4081 | * __schedule() (switch to task 'p') | |
4082 | * LOCK rq->lock | |
4083 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
4084 | * STORE p->on_cpu = 1 LOAD p->cpu | |
4085 | * | |
4086 | * to ensure we observe the correct CPU on which the task is currently | |
4087 | * scheduling. | |
c6e7bd7a | 4088 | */ |
b6e13e85 | 4089 | if (smp_load_acquire(&p->on_cpu) && |
739f70b4 | 4090 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags | WF_ON_CPU)) |
c6e7bd7a PZ |
4091 | goto unlock; |
4092 | ||
e9c84311 | 4093 | /* |
d1ccc66d | 4094 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 4095 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 4096 | * |
31cb1bc0 | 4097 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
4098 | * |
4099 | * This ensures that tasks getting woken will be fully ordered against | |
4100 | * their previous state and preserve Program Order. | |
0970d299 | 4101 | */ |
1f03e8d2 | 4102 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 4103 | |
3aef1551 | 4104 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc | 4105 | if (task_cpu(p) != cpu) { |
ec618b84 PZ |
4106 | if (p->in_iowait) { |
4107 | delayacct_blkio_end(p); | |
4108 | atomic_dec(&task_rq(p)->nr_iowait); | |
4109 | } | |
4110 | ||
f339b9dc | 4111 | wake_flags |= WF_MIGRATED; |
eb414681 | 4112 | psi_ttwu_dequeue(p); |
e4a52bcb | 4113 | set_task_cpu(p, cpu); |
f339b9dc | 4114 | } |
b6e13e85 PZ |
4115 | #else |
4116 | cpu = task_cpu(p); | |
1da177e4 | 4117 | #endif /* CONFIG_SMP */ |
1da177e4 | 4118 | |
b5179ac7 | 4119 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 4120 | unlock: |
013fdb80 | 4121 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
4122 | out: |
4123 | if (success) | |
b6e13e85 | 4124 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 4125 | preempt_enable(); |
1da177e4 LT |
4126 | |
4127 | return success; | |
4128 | } | |
4129 | ||
2beaf328 | 4130 | /** |
9b3c4ab3 | 4131 | * task_call_func - Invoke a function on task in fixed state |
1b7af295 | 4132 | * @p: Process for which the function is to be invoked, can be @current. |
2beaf328 PM |
4133 | * @func: Function to invoke. |
4134 | * @arg: Argument to function. | |
4135 | * | |
f6ac18fa PZ |
4136 | * Fix the task in it's current state by avoiding wakeups and or rq operations |
4137 | * and call @func(@arg) on it. This function can use ->on_rq and task_curr() | |
4138 | * to work out what the state is, if required. Given that @func can be invoked | |
4139 | * with a runqueue lock held, it had better be quite lightweight. | |
2beaf328 PM |
4140 | * |
4141 | * Returns: | |
f6ac18fa | 4142 | * Whatever @func returns |
2beaf328 | 4143 | */ |
9b3c4ab3 | 4144 | int task_call_func(struct task_struct *p, task_call_f func, void *arg) |
2beaf328 | 4145 | { |
f6ac18fa PZ |
4146 | struct rq *rq = NULL; |
4147 | unsigned int state; | |
2beaf328 | 4148 | struct rq_flags rf; |
9b3c4ab3 | 4149 | int ret; |
2beaf328 | 4150 | |
1b7af295 | 4151 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
f6ac18fa PZ |
4152 | |
4153 | state = READ_ONCE(p->__state); | |
4154 | ||
4155 | /* | |
4156 | * Ensure we load p->on_rq after p->__state, otherwise it would be | |
4157 | * possible to, falsely, observe p->on_rq == 0. | |
4158 | * | |
4159 | * See try_to_wake_up() for a longer comment. | |
4160 | */ | |
4161 | smp_rmb(); | |
4162 | ||
4163 | /* | |
4164 | * Since pi->lock blocks try_to_wake_up(), we don't need rq->lock when | |
4165 | * the task is blocked. Make sure to check @state since ttwu() can drop | |
4166 | * locks at the end, see ttwu_queue_wakelist(). | |
4167 | */ | |
4168 | if (state == TASK_RUNNING || state == TASK_WAKING || p->on_rq) | |
2beaf328 | 4169 | rq = __task_rq_lock(p, &rf); |
f6ac18fa PZ |
4170 | |
4171 | /* | |
4172 | * At this point the task is pinned; either: | |
4173 | * - blocked and we're holding off wakeups (pi->lock) | |
4174 | * - woken, and we're holding off enqueue (rq->lock) | |
4175 | * - queued, and we're holding off schedule (rq->lock) | |
4176 | * - running, and we're holding off de-schedule (rq->lock) | |
4177 | * | |
4178 | * The called function (@func) can use: task_curr(), p->on_rq and | |
4179 | * p->__state to differentiate between these states. | |
4180 | */ | |
4181 | ret = func(p, arg); | |
4182 | ||
4183 | if (rq) | |
2beaf328 | 4184 | rq_unlock(rq, &rf); |
f6ac18fa | 4185 | |
1b7af295 | 4186 | raw_spin_unlock_irqrestore(&p->pi_lock, rf.flags); |
2beaf328 PM |
4187 | return ret; |
4188 | } | |
4189 | ||
50fa610a DH |
4190 | /** |
4191 | * wake_up_process - Wake up a specific process | |
4192 | * @p: The process to be woken up. | |
4193 | * | |
4194 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
4195 | * processes. |
4196 | * | |
4197 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 4198 | * |
7696f991 | 4199 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 4200 | */ |
7ad5b3a5 | 4201 | int wake_up_process(struct task_struct *p) |
1da177e4 | 4202 | { |
9067ac85 | 4203 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 4204 | } |
1da177e4 LT |
4205 | EXPORT_SYMBOL(wake_up_process); |
4206 | ||
7ad5b3a5 | 4207 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
4208 | { |
4209 | return try_to_wake_up(p, state, 0); | |
4210 | } | |
4211 | ||
1da177e4 LT |
4212 | /* |
4213 | * Perform scheduler related setup for a newly forked process p. | |
4214 | * p is forked by current. | |
dd41f596 IM |
4215 | * |
4216 | * __sched_fork() is basic setup used by init_idle() too: | |
4217 | */ | |
5e1576ed | 4218 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4219 | { |
fd2f4419 PZ |
4220 | p->on_rq = 0; |
4221 | ||
4222 | p->se.on_rq = 0; | |
dd41f596 IM |
4223 | p->se.exec_start = 0; |
4224 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 4225 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 4226 | p->se.nr_migrations = 0; |
da7a735e | 4227 | p->se.vruntime = 0; |
fd2f4419 | 4228 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 4229 | |
ad936d86 BP |
4230 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4231 | p->se.cfs_rq = NULL; | |
4232 | #endif | |
4233 | ||
6cfb0d5d | 4234 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 4235 | /* Even if schedstat is disabled, there should not be garbage */ |
ceeadb83 | 4236 | memset(&p->stats, 0, sizeof(p->stats)); |
6cfb0d5d | 4237 | #endif |
476d139c | 4238 | |
aab03e05 | 4239 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 4240 | init_dl_task_timer(&p->dl); |
209a0cbd | 4241 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 4242 | __dl_clear_params(p); |
aab03e05 | 4243 | |
fa717060 | 4244 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
4245 | p->rt.timeout = 0; |
4246 | p->rt.time_slice = sched_rr_timeslice; | |
4247 | p->rt.on_rq = 0; | |
4248 | p->rt.on_list = 0; | |
476d139c | 4249 | |
e107be36 AK |
4250 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4251 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
4252 | #endif | |
cbee9f88 | 4253 | |
5e1f0f09 MG |
4254 | #ifdef CONFIG_COMPACTION |
4255 | p->capture_control = NULL; | |
4256 | #endif | |
13784475 | 4257 | init_numa_balancing(clone_flags, p); |
a1488664 | 4258 | #ifdef CONFIG_SMP |
8c4890d1 | 4259 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 4260 | p->migration_pending = NULL; |
a1488664 | 4261 | #endif |
dd41f596 IM |
4262 | } |
4263 | ||
2a595721 SD |
4264 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
4265 | ||
1a687c2e | 4266 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 4267 | |
1a687c2e MG |
4268 | void set_numabalancing_state(bool enabled) |
4269 | { | |
4270 | if (enabled) | |
2a595721 | 4271 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 4272 | else |
2a595721 | 4273 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 4274 | } |
54a43d54 AK |
4275 | |
4276 | #ifdef CONFIG_PROC_SYSCTL | |
4277 | int sysctl_numa_balancing(struct ctl_table *table, int write, | |
32927393 | 4278 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
4279 | { |
4280 | struct ctl_table t; | |
4281 | int err; | |
2a595721 | 4282 | int state = static_branch_likely(&sched_numa_balancing); |
54a43d54 AK |
4283 | |
4284 | if (write && !capable(CAP_SYS_ADMIN)) | |
4285 | return -EPERM; | |
4286 | ||
4287 | t = *table; | |
4288 | t.data = &state; | |
4289 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4290 | if (err < 0) | |
4291 | return err; | |
4292 | if (write) | |
4293 | set_numabalancing_state(state); | |
4294 | return err; | |
4295 | } | |
4296 | #endif | |
4297 | #endif | |
dd41f596 | 4298 | |
4698f88c JP |
4299 | #ifdef CONFIG_SCHEDSTATS |
4300 | ||
cb251765 MG |
4301 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4302 | ||
cb251765 MG |
4303 | static void set_schedstats(bool enabled) |
4304 | { | |
4305 | if (enabled) | |
4306 | static_branch_enable(&sched_schedstats); | |
4307 | else | |
4308 | static_branch_disable(&sched_schedstats); | |
4309 | } | |
4310 | ||
4311 | void force_schedstat_enabled(void) | |
4312 | { | |
4313 | if (!schedstat_enabled()) { | |
4314 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
4315 | static_branch_enable(&sched_schedstats); | |
4316 | } | |
4317 | } | |
4318 | ||
4319 | static int __init setup_schedstats(char *str) | |
4320 | { | |
4321 | int ret = 0; | |
4322 | if (!str) | |
4323 | goto out; | |
4324 | ||
4325 | if (!strcmp(str, "enable")) { | |
1faa491a | 4326 | set_schedstats(true); |
cb251765 MG |
4327 | ret = 1; |
4328 | } else if (!strcmp(str, "disable")) { | |
1faa491a | 4329 | set_schedstats(false); |
cb251765 MG |
4330 | ret = 1; |
4331 | } | |
4332 | out: | |
4333 | if (!ret) | |
4334 | pr_warn("Unable to parse schedstats=\n"); | |
4335 | ||
4336 | return ret; | |
4337 | } | |
4338 | __setup("schedstats=", setup_schedstats); | |
4339 | ||
4340 | #ifdef CONFIG_PROC_SYSCTL | |
32927393 CH |
4341 | int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
4342 | size_t *lenp, loff_t *ppos) | |
cb251765 MG |
4343 | { |
4344 | struct ctl_table t; | |
4345 | int err; | |
4346 | int state = static_branch_likely(&sched_schedstats); | |
4347 | ||
4348 | if (write && !capable(CAP_SYS_ADMIN)) | |
4349 | return -EPERM; | |
4350 | ||
4351 | t = *table; | |
4352 | t.data = &state; | |
4353 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4354 | if (err < 0) | |
4355 | return err; | |
4356 | if (write) | |
4357 | set_schedstats(state); | |
4358 | return err; | |
4359 | } | |
4698f88c | 4360 | #endif /* CONFIG_PROC_SYSCTL */ |
4698f88c | 4361 | #endif /* CONFIG_SCHEDSTATS */ |
dd41f596 IM |
4362 | |
4363 | /* | |
4364 | * fork()/clone()-time setup: | |
4365 | */ | |
aab03e05 | 4366 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4367 | { |
5e1576ed | 4368 | __sched_fork(clone_flags, p); |
06b83b5f | 4369 | /* |
7dc603c9 | 4370 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
4371 | * nobody will actually run it, and a signal or other external |
4372 | * event cannot wake it up and insert it on the runqueue either. | |
4373 | */ | |
2f064a59 | 4374 | p->__state = TASK_NEW; |
dd41f596 | 4375 | |
c350a04e MG |
4376 | /* |
4377 | * Make sure we do not leak PI boosting priority to the child. | |
4378 | */ | |
4379 | p->prio = current->normal_prio; | |
4380 | ||
e8f14172 PB |
4381 | uclamp_fork(p); |
4382 | ||
b9dc29e7 MG |
4383 | /* |
4384 | * Revert to default priority/policy on fork if requested. | |
4385 | */ | |
4386 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 4387 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 4388 | p->policy = SCHED_NORMAL; |
6c697bdf | 4389 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
4390 | p->rt_priority = 0; |
4391 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
4392 | p->static_prio = NICE_TO_PRIO(0); | |
4393 | ||
f558c2b8 | 4394 | p->prio = p->normal_prio = p->static_prio; |
9059393e | 4395 | set_load_weight(p, false); |
6c697bdf | 4396 | |
b9dc29e7 MG |
4397 | /* |
4398 | * We don't need the reset flag anymore after the fork. It has | |
4399 | * fulfilled its duty: | |
4400 | */ | |
4401 | p->sched_reset_on_fork = 0; | |
4402 | } | |
ca94c442 | 4403 | |
af0fffd9 | 4404 | if (dl_prio(p->prio)) |
aab03e05 | 4405 | return -EAGAIN; |
af0fffd9 | 4406 | else if (rt_prio(p->prio)) |
aab03e05 | 4407 | p->sched_class = &rt_sched_class; |
af0fffd9 | 4408 | else |
2ddbf952 | 4409 | p->sched_class = &fair_sched_class; |
b29739f9 | 4410 | |
7dc603c9 | 4411 | init_entity_runnable_average(&p->se); |
cd29fe6f | 4412 | |
f6db8347 | 4413 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 4414 | if (likely(sched_info_on())) |
52f17b6c | 4415 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 4416 | #endif |
3ca7a440 PZ |
4417 | #if defined(CONFIG_SMP) |
4418 | p->on_cpu = 0; | |
4866cde0 | 4419 | #endif |
01028747 | 4420 | init_task_preempt_count(p); |
806c09a7 | 4421 | #ifdef CONFIG_SMP |
917b627d | 4422 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 4423 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 4424 | #endif |
aab03e05 | 4425 | return 0; |
1da177e4 LT |
4426 | } |
4427 | ||
4ef0c5c6 | 4428 | void sched_post_fork(struct task_struct *p, struct kernel_clone_args *kargs) |
13685c4a | 4429 | { |
4ef0c5c6 ZQ |
4430 | unsigned long flags; |
4431 | #ifdef CONFIG_CGROUP_SCHED | |
4432 | struct task_group *tg; | |
4433 | #endif | |
4434 | ||
4435 | raw_spin_lock_irqsave(&p->pi_lock, flags); | |
4436 | #ifdef CONFIG_CGROUP_SCHED | |
4437 | tg = container_of(kargs->cset->subsys[cpu_cgrp_id], | |
4438 | struct task_group, css); | |
4439 | p->sched_task_group = autogroup_task_group(p, tg); | |
4440 | #endif | |
4441 | rseq_migrate(p); | |
4442 | /* | |
4443 | * We're setting the CPU for the first time, we don't migrate, | |
4444 | * so use __set_task_cpu(). | |
4445 | */ | |
4446 | __set_task_cpu(p, smp_processor_id()); | |
4447 | if (p->sched_class->task_fork) | |
4448 | p->sched_class->task_fork(p); | |
4449 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4450 | ||
13685c4a QY |
4451 | uclamp_post_fork(p); |
4452 | } | |
4453 | ||
332ac17e DF |
4454 | unsigned long to_ratio(u64 period, u64 runtime) |
4455 | { | |
4456 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 4457 | return BW_UNIT; |
332ac17e DF |
4458 | |
4459 | /* | |
4460 | * Doing this here saves a lot of checks in all | |
4461 | * the calling paths, and returning zero seems | |
4462 | * safe for them anyway. | |
4463 | */ | |
4464 | if (period == 0) | |
4465 | return 0; | |
4466 | ||
c52f14d3 | 4467 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
4468 | } |
4469 | ||
1da177e4 LT |
4470 | /* |
4471 | * wake_up_new_task - wake up a newly created task for the first time. | |
4472 | * | |
4473 | * This function will do some initial scheduler statistics housekeeping | |
4474 | * that must be done for every newly created context, then puts the task | |
4475 | * on the runqueue and wakes it. | |
4476 | */ | |
3e51e3ed | 4477 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 4478 | { |
eb580751 | 4479 | struct rq_flags rf; |
dd41f596 | 4480 | struct rq *rq; |
fabf318e | 4481 | |
eb580751 | 4482 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
2f064a59 | 4483 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fabf318e PZ |
4484 | #ifdef CONFIG_SMP |
4485 | /* | |
4486 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 4487 | * - cpus_ptr can change in the fork path |
d1ccc66d | 4488 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
4489 | * |
4490 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
4491 | * as we're not fully set-up yet. | |
fabf318e | 4492 | */ |
32e839dd | 4493 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 4494 | rseq_migrate(p); |
3aef1551 | 4495 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 4496 | #endif |
b7fa30c9 | 4497 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 4498 | update_rq_clock(rq); |
d0fe0b9c | 4499 | post_init_entity_util_avg(p); |
0017d735 | 4500 | |
7a57f32a | 4501 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 4502 | trace_sched_wakeup_new(p); |
a7558e01 | 4503 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 4504 | #ifdef CONFIG_SMP |
0aaafaab PZ |
4505 | if (p->sched_class->task_woken) { |
4506 | /* | |
b19a888c | 4507 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
4508 | * drop it. |
4509 | */ | |
d8ac8971 | 4510 | rq_unpin_lock(rq, &rf); |
efbbd05a | 4511 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 4512 | rq_repin_lock(rq, &rf); |
0aaafaab | 4513 | } |
9a897c5a | 4514 | #endif |
eb580751 | 4515 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4516 | } |
4517 | ||
e107be36 AK |
4518 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4519 | ||
b7203428 | 4520 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 4521 | |
2ecd9d29 PZ |
4522 | void preempt_notifier_inc(void) |
4523 | { | |
b7203428 | 4524 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
4525 | } |
4526 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
4527 | ||
4528 | void preempt_notifier_dec(void) | |
4529 | { | |
b7203428 | 4530 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
4531 | } |
4532 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
4533 | ||
e107be36 | 4534 | /** |
80dd99b3 | 4535 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 4536 | * @notifier: notifier struct to register |
e107be36 AK |
4537 | */ |
4538 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
4539 | { | |
b7203428 | 4540 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
4541 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
4542 | ||
e107be36 AK |
4543 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
4544 | } | |
4545 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
4546 | ||
4547 | /** | |
4548 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 4549 | * @notifier: notifier struct to unregister |
e107be36 | 4550 | * |
d84525a8 | 4551 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
4552 | */ |
4553 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
4554 | { | |
4555 | hlist_del(¬ifier->link); | |
4556 | } | |
4557 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
4558 | ||
1cde2930 | 4559 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4560 | { |
4561 | struct preempt_notifier *notifier; | |
e107be36 | 4562 | |
b67bfe0d | 4563 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4564 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
4565 | } | |
4566 | ||
1cde2930 PZ |
4567 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
4568 | { | |
b7203428 | 4569 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4570 | __fire_sched_in_preempt_notifiers(curr); |
4571 | } | |
4572 | ||
e107be36 | 4573 | static void |
1cde2930 PZ |
4574 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4575 | struct task_struct *next) | |
e107be36 AK |
4576 | { |
4577 | struct preempt_notifier *notifier; | |
e107be36 | 4578 | |
b67bfe0d | 4579 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4580 | notifier->ops->sched_out(notifier, next); |
4581 | } | |
4582 | ||
1cde2930 PZ |
4583 | static __always_inline void |
4584 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
4585 | struct task_struct *next) | |
4586 | { | |
b7203428 | 4587 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4588 | __fire_sched_out_preempt_notifiers(curr, next); |
4589 | } | |
4590 | ||
6d6bc0ad | 4591 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4592 | |
1cde2930 | 4593 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4594 | { |
4595 | } | |
4596 | ||
1cde2930 | 4597 | static inline void |
e107be36 AK |
4598 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4599 | struct task_struct *next) | |
4600 | { | |
4601 | } | |
4602 | ||
6d6bc0ad | 4603 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4604 | |
31cb1bc0 | 4605 | static inline void prepare_task(struct task_struct *next) |
4606 | { | |
4607 | #ifdef CONFIG_SMP | |
4608 | /* | |
4609 | * Claim the task as running, we do this before switching to it | |
4610 | * such that any running task will have this set. | |
58877d34 PZ |
4611 | * |
4612 | * See the ttwu() WF_ON_CPU case and its ordering comment. | |
31cb1bc0 | 4613 | */ |
58877d34 | 4614 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 4615 | #endif |
4616 | } | |
4617 | ||
4618 | static inline void finish_task(struct task_struct *prev) | |
4619 | { | |
4620 | #ifdef CONFIG_SMP | |
4621 | /* | |
58877d34 PZ |
4622 | * This must be the very last reference to @prev from this CPU. After |
4623 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
4624 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 4625 | * finished. |
4626 | * | |
4627 | * In particular, the load of prev->state in finish_task_switch() must | |
4628 | * happen before this. | |
4629 | * | |
4630 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
4631 | */ | |
4632 | smp_store_release(&prev->on_cpu, 0); | |
4633 | #endif | |
4634 | } | |
4635 | ||
565790d2 PZ |
4636 | #ifdef CONFIG_SMP |
4637 | ||
4638 | static void do_balance_callbacks(struct rq *rq, struct callback_head *head) | |
4639 | { | |
4640 | void (*func)(struct rq *rq); | |
4641 | struct callback_head *next; | |
4642 | ||
5cb9eaa3 | 4643 | lockdep_assert_rq_held(rq); |
565790d2 PZ |
4644 | |
4645 | while (head) { | |
4646 | func = (void (*)(struct rq *))head->func; | |
4647 | next = head->next; | |
4648 | head->next = NULL; | |
4649 | head = next; | |
4650 | ||
4651 | func(rq); | |
4652 | } | |
4653 | } | |
4654 | ||
ae792702 PZ |
4655 | static void balance_push(struct rq *rq); |
4656 | ||
4657 | struct callback_head balance_push_callback = { | |
4658 | .next = NULL, | |
4659 | .func = (void (*)(struct callback_head *))balance_push, | |
4660 | }; | |
4661 | ||
565790d2 PZ |
4662 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) |
4663 | { | |
4664 | struct callback_head *head = rq->balance_callback; | |
4665 | ||
5cb9eaa3 | 4666 | lockdep_assert_rq_held(rq); |
ae792702 | 4667 | if (head) |
565790d2 PZ |
4668 | rq->balance_callback = NULL; |
4669 | ||
4670 | return head; | |
4671 | } | |
4672 | ||
4673 | static void __balance_callbacks(struct rq *rq) | |
4674 | { | |
4675 | do_balance_callbacks(rq, splice_balance_callbacks(rq)); | |
4676 | } | |
4677 | ||
4678 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4679 | { | |
4680 | unsigned long flags; | |
4681 | ||
4682 | if (unlikely(head)) { | |
5cb9eaa3 | 4683 | raw_spin_rq_lock_irqsave(rq, flags); |
565790d2 | 4684 | do_balance_callbacks(rq, head); |
5cb9eaa3 | 4685 | raw_spin_rq_unlock_irqrestore(rq, flags); |
565790d2 PZ |
4686 | } |
4687 | } | |
4688 | ||
4689 | #else | |
4690 | ||
4691 | static inline void __balance_callbacks(struct rq *rq) | |
4692 | { | |
4693 | } | |
4694 | ||
4695 | static inline struct callback_head *splice_balance_callbacks(struct rq *rq) | |
4696 | { | |
4697 | return NULL; | |
4698 | } | |
4699 | ||
4700 | static inline void balance_callbacks(struct rq *rq, struct callback_head *head) | |
4701 | { | |
4702 | } | |
4703 | ||
4704 | #endif | |
4705 | ||
269d5992 PZ |
4706 | static inline void |
4707 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4708 | { |
269d5992 PZ |
4709 | /* |
4710 | * Since the runqueue lock will be released by the next | |
4711 | * task (which is an invalid locking op but in the case | |
4712 | * of the scheduler it's an obvious special-case), so we | |
4713 | * do an early lockdep release here: | |
4714 | */ | |
4715 | rq_unpin_lock(rq, rf); | |
9ef7e7e3 | 4716 | spin_release(&__rq_lockp(rq)->dep_map, _THIS_IP_); |
31cb1bc0 | 4717 | #ifdef CONFIG_DEBUG_SPINLOCK |
4718 | /* this is a valid case when another task releases the spinlock */ | |
5cb9eaa3 | 4719 | rq_lockp(rq)->owner = next; |
31cb1bc0 | 4720 | #endif |
269d5992 PZ |
4721 | } |
4722 | ||
4723 | static inline void finish_lock_switch(struct rq *rq) | |
4724 | { | |
31cb1bc0 | 4725 | /* |
4726 | * If we are tracking spinlock dependencies then we have to | |
4727 | * fix up the runqueue lock - which gets 'carried over' from | |
4728 | * prev into current: | |
4729 | */ | |
9ef7e7e3 | 4730 | spin_acquire(&__rq_lockp(rq)->dep_map, 0, 0, _THIS_IP_); |
ae792702 | 4731 | __balance_callbacks(rq); |
5cb9eaa3 | 4732 | raw_spin_rq_unlock_irq(rq); |
31cb1bc0 | 4733 | } |
4734 | ||
325ea10c IM |
4735 | /* |
4736 | * NOP if the arch has not defined these: | |
4737 | */ | |
4738 | ||
4739 | #ifndef prepare_arch_switch | |
4740 | # define prepare_arch_switch(next) do { } while (0) | |
4741 | #endif | |
4742 | ||
4743 | #ifndef finish_arch_post_lock_switch | |
4744 | # define finish_arch_post_lock_switch() do { } while (0) | |
4745 | #endif | |
4746 | ||
5fbda3ec TG |
4747 | static inline void kmap_local_sched_out(void) |
4748 | { | |
4749 | #ifdef CONFIG_KMAP_LOCAL | |
4750 | if (unlikely(current->kmap_ctrl.idx)) | |
4751 | __kmap_local_sched_out(); | |
4752 | #endif | |
4753 | } | |
4754 | ||
4755 | static inline void kmap_local_sched_in(void) | |
4756 | { | |
4757 | #ifdef CONFIG_KMAP_LOCAL | |
4758 | if (unlikely(current->kmap_ctrl.idx)) | |
4759 | __kmap_local_sched_in(); | |
4760 | #endif | |
4761 | } | |
4762 | ||
4866cde0 NP |
4763 | /** |
4764 | * prepare_task_switch - prepare to switch tasks | |
4765 | * @rq: the runqueue preparing to switch | |
421cee29 | 4766 | * @prev: the current task that is being switched out |
4866cde0 NP |
4767 | * @next: the task we are going to switch to. |
4768 | * | |
4769 | * This is called with the rq lock held and interrupts off. It must | |
4770 | * be paired with a subsequent finish_task_switch after the context | |
4771 | * switch. | |
4772 | * | |
4773 | * prepare_task_switch sets up locking and calls architecture specific | |
4774 | * hooks. | |
4775 | */ | |
e107be36 AK |
4776 | static inline void |
4777 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
4778 | struct task_struct *next) | |
4866cde0 | 4779 | { |
0ed557aa | 4780 | kcov_prepare_switch(prev); |
43148951 | 4781 | sched_info_switch(rq, prev, next); |
fe4b04fa | 4782 | perf_event_task_sched_out(prev, next); |
d7822b1e | 4783 | rseq_preempt(prev); |
e107be36 | 4784 | fire_sched_out_preempt_notifiers(prev, next); |
5fbda3ec | 4785 | kmap_local_sched_out(); |
31cb1bc0 | 4786 | prepare_task(next); |
4866cde0 NP |
4787 | prepare_arch_switch(next); |
4788 | } | |
4789 | ||
1da177e4 LT |
4790 | /** |
4791 | * finish_task_switch - clean up after a task-switch | |
4792 | * @prev: the thread we just switched away from. | |
4793 | * | |
4866cde0 NP |
4794 | * finish_task_switch must be called after the context switch, paired |
4795 | * with a prepare_task_switch call before the context switch. | |
4796 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
4797 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
4798 | * |
4799 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 4800 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
4801 | * with the lock held can cause deadlocks; see schedule() for |
4802 | * details.) | |
dfa50b60 ON |
4803 | * |
4804 | * The context switch have flipped the stack from under us and restored the | |
4805 | * local variables which were saved when this task called schedule() in the | |
4806 | * past. prev == current is still correct but we need to recalculate this_rq | |
4807 | * because prev may have moved to another CPU. | |
1da177e4 | 4808 | */ |
dfa50b60 | 4809 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
4810 | __releases(rq->lock) |
4811 | { | |
dfa50b60 | 4812 | struct rq *rq = this_rq(); |
1da177e4 | 4813 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 4814 | long prev_state; |
1da177e4 | 4815 | |
609ca066 PZ |
4816 | /* |
4817 | * The previous task will have left us with a preempt_count of 2 | |
4818 | * because it left us after: | |
4819 | * | |
4820 | * schedule() | |
4821 | * preempt_disable(); // 1 | |
4822 | * __schedule() | |
4823 | * raw_spin_lock_irq(&rq->lock) // 2 | |
4824 | * | |
4825 | * Also, see FORK_PREEMPT_COUNT. | |
4826 | */ | |
e2bf1c4b PZ |
4827 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
4828 | "corrupted preempt_count: %s/%d/0x%x\n", | |
4829 | current->comm, current->pid, preempt_count())) | |
4830 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 4831 | |
1da177e4 LT |
4832 | rq->prev_mm = NULL; |
4833 | ||
4834 | /* | |
4835 | * A task struct has one reference for the use as "current". | |
c394cc9f | 4836 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
4837 | * schedule one last time. The schedule call will never return, and |
4838 | * the scheduled task must drop that reference. | |
95913d97 PZ |
4839 | * |
4840 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 4841 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
4842 | * running on another CPU and we could rave with its RUNNING -> DEAD |
4843 | * transition, resulting in a double drop. | |
1da177e4 | 4844 | */ |
2f064a59 | 4845 | prev_state = READ_ONCE(prev->__state); |
bf9fae9f | 4846 | vtime_task_switch(prev); |
a8d757ef | 4847 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 4848 | finish_task(prev); |
0fdcccfa | 4849 | tick_nohz_task_switch(); |
31cb1bc0 | 4850 | finish_lock_switch(rq); |
01f23e16 | 4851 | finish_arch_post_lock_switch(); |
0ed557aa | 4852 | kcov_finish_switch(current); |
5fbda3ec TG |
4853 | /* |
4854 | * kmap_local_sched_out() is invoked with rq::lock held and | |
4855 | * interrupts disabled. There is no requirement for that, but the | |
4856 | * sched out code does not have an interrupt enabled section. | |
4857 | * Restoring the maps on sched in does not require interrupts being | |
4858 | * disabled either. | |
4859 | */ | |
4860 | kmap_local_sched_in(); | |
e8fa1362 | 4861 | |
e107be36 | 4862 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 4863 | /* |
70216e18 MD |
4864 | * When switching through a kernel thread, the loop in |
4865 | * membarrier_{private,global}_expedited() may have observed that | |
4866 | * kernel thread and not issued an IPI. It is therefore possible to | |
4867 | * schedule between user->kernel->user threads without passing though | |
4868 | * switch_mm(). Membarrier requires a barrier after storing to | |
4869 | * rq->curr, before returning to userspace, so provide them here: | |
4870 | * | |
4871 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
4872 | * provided by mmdrop(), | |
4873 | * - a sync_core for SYNC_CORE. | |
306e0604 | 4874 | */ |
70216e18 MD |
4875 | if (mm) { |
4876 | membarrier_mm_sync_core_before_usermode(mm); | |
8d491de6 | 4877 | mmdrop_sched(mm); |
70216e18 | 4878 | } |
1cef1150 PZ |
4879 | if (unlikely(prev_state == TASK_DEAD)) { |
4880 | if (prev->sched_class->task_dead) | |
4881 | prev->sched_class->task_dead(prev); | |
68f24b08 | 4882 | |
1cef1150 PZ |
4883 | /* Task is done with its stack. */ |
4884 | put_task_stack(prev); | |
4885 | ||
0ff7b2cf | 4886 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 4887 | } |
99e5ada9 | 4888 | |
dfa50b60 | 4889 | return rq; |
1da177e4 LT |
4890 | } |
4891 | ||
4892 | /** | |
4893 | * schedule_tail - first thing a freshly forked thread must call. | |
4894 | * @prev: the thread we just switched away from. | |
4895 | */ | |
722a9f92 | 4896 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
4897 | __releases(rq->lock) |
4898 | { | |
609ca066 PZ |
4899 | /* |
4900 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
4901 | * finish_task_switch() for details. | |
4902 | * | |
4903 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
4904 | * and the preempt_enable() will end up enabling preemption (on | |
4905 | * PREEMPT_COUNT kernels). | |
4906 | */ | |
4907 | ||
13c2235b | 4908 | finish_task_switch(prev); |
1a43a14a | 4909 | preempt_enable(); |
70b97a7f | 4910 | |
1da177e4 | 4911 | if (current->set_child_tid) |
b488893a | 4912 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
4913 | |
4914 | calculate_sigpending(); | |
1da177e4 LT |
4915 | } |
4916 | ||
4917 | /* | |
dfa50b60 | 4918 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 4919 | */ |
04936948 | 4920 | static __always_inline struct rq * |
70b97a7f | 4921 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 4922 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 4923 | { |
e107be36 | 4924 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 4925 | |
9226d125 ZA |
4926 | /* |
4927 | * For paravirt, this is coupled with an exit in switch_to to | |
4928 | * combine the page table reload and the switch backend into | |
4929 | * one hypercall. | |
4930 | */ | |
224101ed | 4931 | arch_start_context_switch(prev); |
9226d125 | 4932 | |
306e0604 | 4933 | /* |
139d025c PZ |
4934 | * kernel -> kernel lazy + transfer active |
4935 | * user -> kernel lazy + mmgrab() active | |
4936 | * | |
4937 | * kernel -> user switch + mmdrop() active | |
4938 | * user -> user switch | |
306e0604 | 4939 | */ |
139d025c PZ |
4940 | if (!next->mm) { // to kernel |
4941 | enter_lazy_tlb(prev->active_mm, next); | |
4942 | ||
4943 | next->active_mm = prev->active_mm; | |
4944 | if (prev->mm) // from user | |
4945 | mmgrab(prev->active_mm); | |
4946 | else | |
4947 | prev->active_mm = NULL; | |
4948 | } else { // to user | |
227a4aad | 4949 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
4950 | /* |
4951 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 4952 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
4953 | * |
4954 | * The below provides this either through switch_mm(), or in | |
4955 | * case 'prev->active_mm == next->mm' through | |
4956 | * finish_task_switch()'s mmdrop(). | |
4957 | */ | |
139d025c | 4958 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
1da177e4 | 4959 | |
139d025c PZ |
4960 | if (!prev->mm) { // from kernel |
4961 | /* will mmdrop() in finish_task_switch(). */ | |
4962 | rq->prev_mm = prev->active_mm; | |
4963 | prev->active_mm = NULL; | |
4964 | } | |
1da177e4 | 4965 | } |
92509b73 | 4966 | |
cb42c9a3 | 4967 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 4968 | |
269d5992 | 4969 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
4970 | |
4971 | /* Here we just switch the register state and the stack. */ | |
4972 | switch_to(prev, next, prev); | |
dd41f596 | 4973 | barrier(); |
dfa50b60 ON |
4974 | |
4975 | return finish_task_switch(prev); | |
1da177e4 LT |
4976 | } |
4977 | ||
4978 | /* | |
1c3e8264 | 4979 | * nr_running and nr_context_switches: |
1da177e4 LT |
4980 | * |
4981 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 4982 | * threads, total number of context switches performed since bootup. |
1da177e4 | 4983 | */ |
01aee8fd | 4984 | unsigned int nr_running(void) |
1da177e4 | 4985 | { |
01aee8fd | 4986 | unsigned int i, sum = 0; |
1da177e4 LT |
4987 | |
4988 | for_each_online_cpu(i) | |
4989 | sum += cpu_rq(i)->nr_running; | |
4990 | ||
4991 | return sum; | |
f711f609 | 4992 | } |
1da177e4 | 4993 | |
2ee507c4 | 4994 | /* |
d1ccc66d | 4995 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
4996 | * |
4997 | * Caution: this function does not check that the caller has disabled | |
4998 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
4999 | * race. The caller is responsible to use it correctly, for example: | |
5000 | * | |
dfcb245e | 5001 | * - from a non-preemptible section (of course) |
00cc1633 DD |
5002 | * |
5003 | * - from a thread that is bound to a single CPU | |
5004 | * | |
5005 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
5006 | */ |
5007 | bool single_task_running(void) | |
5008 | { | |
00cc1633 | 5009 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
5010 | } |
5011 | EXPORT_SYMBOL(single_task_running); | |
5012 | ||
1da177e4 | 5013 | unsigned long long nr_context_switches(void) |
46cb4b7c | 5014 | { |
cc94abfc SR |
5015 | int i; |
5016 | unsigned long long sum = 0; | |
46cb4b7c | 5017 | |
0a945022 | 5018 | for_each_possible_cpu(i) |
1da177e4 | 5019 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 5020 | |
1da177e4 LT |
5021 | return sum; |
5022 | } | |
483b4ee6 | 5023 | |
145d952a DL |
5024 | /* |
5025 | * Consumers of these two interfaces, like for example the cpuidle menu | |
5026 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
5027 | * for a CPU that has IO-wait which might not even end up running the task when | |
5028 | * it does become runnable. | |
5029 | */ | |
5030 | ||
8fc2858e | 5031 | unsigned int nr_iowait_cpu(int cpu) |
145d952a DL |
5032 | { |
5033 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
5034 | } | |
5035 | ||
e33a9bba | 5036 | /* |
b19a888c | 5037 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
5038 | * |
5039 | * The idea behind IO-wait account is to account the idle time that we could | |
5040 | * have spend running if it were not for IO. That is, if we were to improve the | |
5041 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
5042 | * | |
5043 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
5044 | * idle time as IO-wait, because if the storage were faster, it could've been | |
5045 | * running and we'd not be idle. | |
5046 | * | |
5047 | * This has been extended to SMP, by doing the same for each CPU. This however | |
5048 | * is broken. | |
5049 | * | |
5050 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
5051 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
5052 | * though, if the storage were faster, both could've ran at the same time, | |
5053 | * utilising both CPUs. | |
5054 | * | |
5055 | * This means, that when looking globally, the current IO-wait accounting on | |
5056 | * SMP is a lower bound, by reason of under accounting. | |
5057 | * | |
5058 | * Worse, since the numbers are provided per CPU, they are sometimes | |
5059 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
5060 | * associated with any one particular CPU, it can wake to another CPU than it | |
5061 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
5062 | * | |
5063 | * Task CPU affinities can make all that even more 'interesting'. | |
5064 | */ | |
5065 | ||
97455168 | 5066 | unsigned int nr_iowait(void) |
1da177e4 | 5067 | { |
97455168 | 5068 | unsigned int i, sum = 0; |
483b4ee6 | 5069 | |
0a945022 | 5070 | for_each_possible_cpu(i) |
145d952a | 5071 | sum += nr_iowait_cpu(i); |
46cb4b7c | 5072 | |
1da177e4 LT |
5073 | return sum; |
5074 | } | |
483b4ee6 | 5075 | |
dd41f596 | 5076 | #ifdef CONFIG_SMP |
8a0be9ef | 5077 | |
46cb4b7c | 5078 | /* |
38022906 PZ |
5079 | * sched_exec - execve() is a valuable balancing opportunity, because at |
5080 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 5081 | */ |
38022906 | 5082 | void sched_exec(void) |
46cb4b7c | 5083 | { |
38022906 | 5084 | struct task_struct *p = current; |
1da177e4 | 5085 | unsigned long flags; |
0017d735 | 5086 | int dest_cpu; |
46cb4b7c | 5087 | |
8f42ced9 | 5088 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 5089 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
5090 | if (dest_cpu == smp_processor_id()) |
5091 | goto unlock; | |
38022906 | 5092 | |
8f42ced9 | 5093 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 5094 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 5095 | |
8f42ced9 PZ |
5096 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5097 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
5098 | return; |
5099 | } | |
0017d735 | 5100 | unlock: |
8f42ced9 | 5101 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 5102 | } |
dd41f596 | 5103 | |
1da177e4 LT |
5104 | #endif |
5105 | ||
1da177e4 | 5106 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 5107 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
5108 | |
5109 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 5110 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 5111 | |
6075620b GG |
5112 | /* |
5113 | * The function fair_sched_class.update_curr accesses the struct curr | |
5114 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
5115 | * we observe a high rate of cache misses in practice. | |
5116 | * Prefetching this data results in improved performance. | |
5117 | */ | |
5118 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
5119 | { | |
5120 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
5121 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
5122 | #else | |
5123 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
5124 | #endif | |
5125 | prefetch(curr); | |
5126 | prefetch(&curr->exec_start); | |
5127 | } | |
5128 | ||
c5f8d995 HS |
5129 | /* |
5130 | * Return accounted runtime for the task. | |
5131 | * In case the task is currently running, return the runtime plus current's | |
5132 | * pending runtime that have not been accounted yet. | |
5133 | */ | |
5134 | unsigned long long task_sched_runtime(struct task_struct *p) | |
5135 | { | |
eb580751 | 5136 | struct rq_flags rf; |
c5f8d995 | 5137 | struct rq *rq; |
6e998916 | 5138 | u64 ns; |
c5f8d995 | 5139 | |
911b2898 PZ |
5140 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
5141 | /* | |
97fb7a0a | 5142 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
5143 | * So we have a optimization chance when the task's delta_exec is 0. |
5144 | * Reading ->on_cpu is racy, but this is ok. | |
5145 | * | |
d1ccc66d IM |
5146 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
5147 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 5148 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
5149 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
5150 | * been accounted, so we're correct here as well. | |
911b2898 | 5151 | */ |
da0c1e65 | 5152 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
5153 | return p->se.sum_exec_runtime; |
5154 | #endif | |
5155 | ||
eb580751 | 5156 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
5157 | /* |
5158 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
5159 | * project cycles that may never be accounted to this | |
5160 | * thread, breaking clock_gettime(). | |
5161 | */ | |
5162 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 5163 | prefetch_curr_exec_start(p); |
6e998916 SG |
5164 | update_rq_clock(rq); |
5165 | p->sched_class->update_curr(rq); | |
5166 | } | |
5167 | ns = p->se.sum_exec_runtime; | |
eb580751 | 5168 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
5169 | |
5170 | return ns; | |
5171 | } | |
48f24c4d | 5172 | |
c006fac5 PT |
5173 | #ifdef CONFIG_SCHED_DEBUG |
5174 | static u64 cpu_resched_latency(struct rq *rq) | |
5175 | { | |
5176 | int latency_warn_ms = READ_ONCE(sysctl_resched_latency_warn_ms); | |
5177 | u64 resched_latency, now = rq_clock(rq); | |
5178 | static bool warned_once; | |
5179 | ||
5180 | if (sysctl_resched_latency_warn_once && warned_once) | |
5181 | return 0; | |
5182 | ||
5183 | if (!need_resched() || !latency_warn_ms) | |
5184 | return 0; | |
5185 | ||
5186 | if (system_state == SYSTEM_BOOTING) | |
5187 | return 0; | |
5188 | ||
5189 | if (!rq->last_seen_need_resched_ns) { | |
5190 | rq->last_seen_need_resched_ns = now; | |
5191 | rq->ticks_without_resched = 0; | |
5192 | return 0; | |
5193 | } | |
5194 | ||
5195 | rq->ticks_without_resched++; | |
5196 | resched_latency = now - rq->last_seen_need_resched_ns; | |
5197 | if (resched_latency <= latency_warn_ms * NSEC_PER_MSEC) | |
5198 | return 0; | |
5199 | ||
5200 | warned_once = true; | |
5201 | ||
5202 | return resched_latency; | |
5203 | } | |
5204 | ||
5205 | static int __init setup_resched_latency_warn_ms(char *str) | |
5206 | { | |
5207 | long val; | |
5208 | ||
5209 | if ((kstrtol(str, 0, &val))) { | |
5210 | pr_warn("Unable to set resched_latency_warn_ms\n"); | |
5211 | return 1; | |
5212 | } | |
5213 | ||
5214 | sysctl_resched_latency_warn_ms = val; | |
5215 | return 1; | |
5216 | } | |
5217 | __setup("resched_latency_warn_ms=", setup_resched_latency_warn_ms); | |
5218 | #else | |
5219 | static inline u64 cpu_resched_latency(struct rq *rq) { return 0; } | |
5220 | #endif /* CONFIG_SCHED_DEBUG */ | |
5221 | ||
7835b98b CL |
5222 | /* |
5223 | * This function gets called by the timer code, with HZ frequency. | |
5224 | * We call it with interrupts disabled. | |
7835b98b CL |
5225 | */ |
5226 | void scheduler_tick(void) | |
5227 | { | |
7835b98b CL |
5228 | int cpu = smp_processor_id(); |
5229 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5230 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 5231 | struct rq_flags rf; |
b4eccf5f | 5232 | unsigned long thermal_pressure; |
c006fac5 | 5233 | u64 resched_latency; |
3e51f33f | 5234 | |
1567c3e3 | 5235 | arch_scale_freq_tick(); |
3e51f33f | 5236 | sched_clock_tick(); |
dd41f596 | 5237 | |
8a8c69c3 PZ |
5238 | rq_lock(rq, &rf); |
5239 | ||
3e51f33f | 5240 | update_rq_clock(rq); |
b4eccf5f | 5241 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 5242 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 5243 | curr->sched_class->task_tick(rq, curr, 0); |
c006fac5 PT |
5244 | if (sched_feat(LATENCY_WARN)) |
5245 | resched_latency = cpu_resched_latency(rq); | |
3289bdb4 | 5246 | calc_global_load_tick(rq); |
8a8c69c3 PZ |
5247 | |
5248 | rq_unlock(rq, &rf); | |
7835b98b | 5249 | |
c006fac5 PT |
5250 | if (sched_feat(LATENCY_WARN) && resched_latency) |
5251 | resched_latency_warn(cpu, resched_latency); | |
5252 | ||
e9d2b064 | 5253 | perf_event_task_tick(); |
e220d2dc | 5254 | |
e418e1c2 | 5255 | #ifdef CONFIG_SMP |
6eb57e0d | 5256 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 5257 | trigger_load_balance(rq); |
e418e1c2 | 5258 | #endif |
1da177e4 LT |
5259 | } |
5260 | ||
265f22a9 | 5261 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
5262 | |
5263 | struct tick_work { | |
5264 | int cpu; | |
b55bd585 | 5265 | atomic_t state; |
d84b3131 FW |
5266 | struct delayed_work work; |
5267 | }; | |
b55bd585 PM |
5268 | /* Values for ->state, see diagram below. */ |
5269 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
5270 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
5271 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
5272 | ||
5273 | /* | |
5274 | * State diagram for ->state: | |
5275 | * | |
5276 | * | |
5277 | * TICK_SCHED_REMOTE_OFFLINE | |
5278 | * | ^ | |
5279 | * | | | |
5280 | * | | sched_tick_remote() | |
5281 | * | | | |
5282 | * | | | |
5283 | * +--TICK_SCHED_REMOTE_OFFLINING | |
5284 | * | ^ | |
5285 | * | | | |
5286 | * sched_tick_start() | | sched_tick_stop() | |
5287 | * | | | |
5288 | * V | | |
5289 | * TICK_SCHED_REMOTE_RUNNING | |
5290 | * | |
5291 | * | |
5292 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
5293 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
5294 | */ | |
d84b3131 FW |
5295 | |
5296 | static struct tick_work __percpu *tick_work_cpu; | |
5297 | ||
5298 | static void sched_tick_remote(struct work_struct *work) | |
5299 | { | |
5300 | struct delayed_work *dwork = to_delayed_work(work); | |
5301 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
5302 | int cpu = twork->cpu; | |
5303 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 5304 | struct task_struct *curr; |
d84b3131 | 5305 | struct rq_flags rf; |
d9c0ffca | 5306 | u64 delta; |
b55bd585 | 5307 | int os; |
d84b3131 FW |
5308 | |
5309 | /* | |
5310 | * Handle the tick only if it appears the remote CPU is running in full | |
5311 | * dynticks mode. The check is racy by nature, but missing a tick or | |
5312 | * having one too much is no big deal because the scheduler tick updates | |
5313 | * statistics and checks timeslices in a time-independent way, regardless | |
5314 | * of when exactly it is running. | |
5315 | */ | |
488603b8 | 5316 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 5317 | goto out_requeue; |
d84b3131 | 5318 | |
d9c0ffca FW |
5319 | rq_lock_irq(rq, &rf); |
5320 | curr = rq->curr; | |
488603b8 | 5321 | if (cpu_is_offline(cpu)) |
d9c0ffca | 5322 | goto out_unlock; |
d84b3131 | 5323 | |
d9c0ffca | 5324 | update_rq_clock(rq); |
d9c0ffca | 5325 | |
488603b8 SW |
5326 | if (!is_idle_task(curr)) { |
5327 | /* | |
5328 | * Make sure the next tick runs within a reasonable | |
5329 | * amount of time. | |
5330 | */ | |
5331 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
5332 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
5333 | } | |
d9c0ffca FW |
5334 | curr->sched_class->task_tick(rq, curr, 0); |
5335 | ||
ebc0f83c | 5336 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
5337 | out_unlock: |
5338 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 5339 | out_requeue: |
ebc0f83c | 5340 | |
d84b3131 FW |
5341 | /* |
5342 | * Run the remote tick once per second (1Hz). This arbitrary | |
5343 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
5344 | * to keep scheduler internal stats reasonably up to date. But |
5345 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 5346 | */ |
b55bd585 PM |
5347 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
5348 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
5349 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
5350 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
5351 | } |
5352 | ||
5353 | static void sched_tick_start(int cpu) | |
5354 | { | |
b55bd585 | 5355 | int os; |
d84b3131 FW |
5356 | struct tick_work *twork; |
5357 | ||
5358 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
5359 | return; | |
5360 | ||
5361 | WARN_ON_ONCE(!tick_work_cpu); | |
5362 | ||
5363 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5364 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
5365 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
5366 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
5367 | twork->cpu = cpu; | |
5368 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
5369 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
5370 | } | |
d84b3131 FW |
5371 | } |
5372 | ||
5373 | #ifdef CONFIG_HOTPLUG_CPU | |
5374 | static void sched_tick_stop(int cpu) | |
5375 | { | |
5376 | struct tick_work *twork; | |
b55bd585 | 5377 | int os; |
d84b3131 FW |
5378 | |
5379 | if (housekeeping_cpu(cpu, HK_FLAG_TICK)) | |
5380 | return; | |
5381 | ||
5382 | WARN_ON_ONCE(!tick_work_cpu); | |
5383 | ||
5384 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5385 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
5386 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
5387 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
5388 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
5389 | } |
5390 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5391 | ||
5392 | int __init sched_tick_offload_init(void) | |
5393 | { | |
5394 | tick_work_cpu = alloc_percpu(struct tick_work); | |
5395 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
5396 | return 0; |
5397 | } | |
5398 | ||
5399 | #else /* !CONFIG_NO_HZ_FULL */ | |
5400 | static inline void sched_tick_start(int cpu) { } | |
5401 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 5402 | #endif |
1da177e4 | 5403 | |
c1a280b6 | 5404 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 5405 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
5406 | /* |
5407 | * If the value passed in is equal to the current preempt count | |
5408 | * then we just disabled preemption. Start timing the latency. | |
5409 | */ | |
5410 | static inline void preempt_latency_start(int val) | |
5411 | { | |
5412 | if (preempt_count() == val) { | |
5413 | unsigned long ip = get_lock_parent_ip(); | |
5414 | #ifdef CONFIG_DEBUG_PREEMPT | |
5415 | current->preempt_disable_ip = ip; | |
5416 | #endif | |
5417 | trace_preempt_off(CALLER_ADDR0, ip); | |
5418 | } | |
5419 | } | |
7e49fcce | 5420 | |
edafe3a5 | 5421 | void preempt_count_add(int val) |
1da177e4 | 5422 | { |
6cd8a4bb | 5423 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5424 | /* |
5425 | * Underflow? | |
5426 | */ | |
9a11b49a IM |
5427 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5428 | return; | |
6cd8a4bb | 5429 | #endif |
bdb43806 | 5430 | __preempt_count_add(val); |
6cd8a4bb | 5431 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5432 | /* |
5433 | * Spinlock count overflowing soon? | |
5434 | */ | |
33859f7f MOS |
5435 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5436 | PREEMPT_MASK - 10); | |
6cd8a4bb | 5437 | #endif |
47252cfb | 5438 | preempt_latency_start(val); |
1da177e4 | 5439 | } |
bdb43806 | 5440 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 5441 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 5442 | |
47252cfb SR |
5443 | /* |
5444 | * If the value passed in equals to the current preempt count | |
5445 | * then we just enabled preemption. Stop timing the latency. | |
5446 | */ | |
5447 | static inline void preempt_latency_stop(int val) | |
5448 | { | |
5449 | if (preempt_count() == val) | |
5450 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
5451 | } | |
5452 | ||
edafe3a5 | 5453 | void preempt_count_sub(int val) |
1da177e4 | 5454 | { |
6cd8a4bb | 5455 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5456 | /* |
5457 | * Underflow? | |
5458 | */ | |
01e3eb82 | 5459 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5460 | return; |
1da177e4 LT |
5461 | /* |
5462 | * Is the spinlock portion underflowing? | |
5463 | */ | |
9a11b49a IM |
5464 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5465 | !(preempt_count() & PREEMPT_MASK))) | |
5466 | return; | |
6cd8a4bb | 5467 | #endif |
9a11b49a | 5468 | |
47252cfb | 5469 | preempt_latency_stop(val); |
bdb43806 | 5470 | __preempt_count_sub(val); |
1da177e4 | 5471 | } |
bdb43806 | 5472 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 5473 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 5474 | |
47252cfb SR |
5475 | #else |
5476 | static inline void preempt_latency_start(int val) { } | |
5477 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
5478 | #endif |
5479 | ||
59ddbcb2 IM |
5480 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
5481 | { | |
5482 | #ifdef CONFIG_DEBUG_PREEMPT | |
5483 | return p->preempt_disable_ip; | |
5484 | #else | |
5485 | return 0; | |
5486 | #endif | |
5487 | } | |
5488 | ||
1da177e4 | 5489 | /* |
dd41f596 | 5490 | * Print scheduling while atomic bug: |
1da177e4 | 5491 | */ |
dd41f596 | 5492 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5493 | { |
d1c6d149 VN |
5494 | /* Save this before calling printk(), since that will clobber it */ |
5495 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
5496 | ||
664dfa65 DJ |
5497 | if (oops_in_progress) |
5498 | return; | |
5499 | ||
3df0fc5b PZ |
5500 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
5501 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 5502 | |
dd41f596 | 5503 | debug_show_held_locks(prev); |
e21f5b15 | 5504 | print_modules(); |
dd41f596 IM |
5505 | if (irqs_disabled()) |
5506 | print_irqtrace_events(prev); | |
d1c6d149 VN |
5507 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
5508 | && in_atomic_preempt_off()) { | |
8f47b187 | 5509 | pr_err("Preemption disabled at:"); |
2062a4e8 | 5510 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 5511 | } |
748c7201 DBO |
5512 | if (panic_on_warn) |
5513 | panic("scheduling while atomic\n"); | |
5514 | ||
6135fc1e | 5515 | dump_stack(); |
373d4d09 | 5516 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 5517 | } |
1da177e4 | 5518 | |
dd41f596 IM |
5519 | /* |
5520 | * Various schedule()-time debugging checks and statistics: | |
5521 | */ | |
312364f3 | 5522 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 5523 | { |
0d9e2632 | 5524 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
5525 | if (task_stack_end_corrupted(prev)) |
5526 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
5527 | |
5528 | if (task_scs_end_corrupted(prev)) | |
5529 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 5530 | #endif |
b99def8b | 5531 | |
312364f3 | 5532 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
2f064a59 | 5533 | if (!preempt && READ_ONCE(prev->__state) && prev->non_block_count) { |
312364f3 DV |
5534 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", |
5535 | prev->comm, prev->pid, prev->non_block_count); | |
5536 | dump_stack(); | |
5537 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
5538 | } | |
5539 | #endif | |
5540 | ||
1dc0fffc | 5541 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 5542 | __schedule_bug(prev); |
1dc0fffc PZ |
5543 | preempt_count_set(PREEMPT_DISABLED); |
5544 | } | |
b3fbab05 | 5545 | rcu_sleep_check(); |
9f68b5b7 | 5546 | SCHED_WARN_ON(ct_state() == CONTEXT_USER); |
dd41f596 | 5547 | |
1da177e4 LT |
5548 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5549 | ||
ae92882e | 5550 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
5551 | } |
5552 | ||
457d1f46 CY |
5553 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
5554 | struct rq_flags *rf) | |
5555 | { | |
5556 | #ifdef CONFIG_SMP | |
5557 | const struct sched_class *class; | |
5558 | /* | |
5559 | * We must do the balancing pass before put_prev_task(), such | |
5560 | * that when we release the rq->lock the task is in the same | |
5561 | * state as before we took rq->lock. | |
5562 | * | |
5563 | * We can terminate the balance pass as soon as we know there is | |
5564 | * a runnable task of @class priority or higher. | |
5565 | */ | |
5566 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
5567 | if (class->balance(rq, prev, rf)) | |
5568 | break; | |
5569 | } | |
5570 | #endif | |
5571 | ||
5572 | put_prev_task(rq, prev); | |
5573 | } | |
5574 | ||
dd41f596 IM |
5575 | /* |
5576 | * Pick up the highest-prio task: | |
5577 | */ | |
5578 | static inline struct task_struct * | |
539f6512 | 5579 | __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 5580 | { |
49ee5768 | 5581 | const struct sched_class *class; |
dd41f596 | 5582 | struct task_struct *p; |
1da177e4 LT |
5583 | |
5584 | /* | |
0ba87bb2 PZ |
5585 | * Optimization: we know that if all tasks are in the fair class we can |
5586 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 5587 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 5588 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 5589 | */ |
aa93cd53 | 5590 | if (likely(prev->sched_class <= &fair_sched_class && |
0ba87bb2 PZ |
5591 | rq->nr_running == rq->cfs.h_nr_running)) { |
5592 | ||
5d7d6056 | 5593 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 5594 | if (unlikely(p == RETRY_TASK)) |
67692435 | 5595 | goto restart; |
6ccdc84b | 5596 | |
1699949d | 5597 | /* Assume the next prioritized class is idle_sched_class */ |
5d7d6056 | 5598 | if (!p) { |
f488e105 | 5599 | put_prev_task(rq, prev); |
98c2f700 | 5600 | p = pick_next_task_idle(rq); |
f488e105 | 5601 | } |
6ccdc84b PZ |
5602 | |
5603 | return p; | |
1da177e4 LT |
5604 | } |
5605 | ||
67692435 | 5606 | restart: |
457d1f46 | 5607 | put_prev_task_balance(rq, prev, rf); |
67692435 | 5608 | |
34f971f6 | 5609 | for_each_class(class) { |
98c2f700 | 5610 | p = class->pick_next_task(rq); |
67692435 | 5611 | if (p) |
dd41f596 | 5612 | return p; |
dd41f596 | 5613 | } |
34f971f6 | 5614 | |
bc9ffef3 | 5615 | BUG(); /* The idle class should always have a runnable task. */ |
dd41f596 | 5616 | } |
1da177e4 | 5617 | |
9edeaea1 | 5618 | #ifdef CONFIG_SCHED_CORE |
539f6512 PZ |
5619 | static inline bool is_task_rq_idle(struct task_struct *t) |
5620 | { | |
5621 | return (task_rq(t)->idle == t); | |
5622 | } | |
5623 | ||
5624 | static inline bool cookie_equals(struct task_struct *a, unsigned long cookie) | |
5625 | { | |
5626 | return is_task_rq_idle(a) || (a->core_cookie == cookie); | |
5627 | } | |
5628 | ||
5629 | static inline bool cookie_match(struct task_struct *a, struct task_struct *b) | |
5630 | { | |
5631 | if (is_task_rq_idle(a) || is_task_rq_idle(b)) | |
5632 | return true; | |
5633 | ||
5634 | return a->core_cookie == b->core_cookie; | |
5635 | } | |
5636 | ||
bc9ffef3 | 5637 | static inline struct task_struct *pick_task(struct rq *rq) |
539f6512 | 5638 | { |
bc9ffef3 PZ |
5639 | const struct sched_class *class; |
5640 | struct task_struct *p; | |
539f6512 | 5641 | |
bc9ffef3 PZ |
5642 | for_each_class(class) { |
5643 | p = class->pick_task(rq); | |
5644 | if (p) | |
5645 | return p; | |
539f6512 PZ |
5646 | } |
5647 | ||
bc9ffef3 | 5648 | BUG(); /* The idle class should always have a runnable task. */ |
539f6512 PZ |
5649 | } |
5650 | ||
c6047c2e JFG |
5651 | extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi); |
5652 | ||
539f6512 PZ |
5653 | static struct task_struct * |
5654 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
5655 | { | |
bc9ffef3 | 5656 | struct task_struct *next, *p, *max = NULL; |
539f6512 | 5657 | const struct cpumask *smt_mask; |
c6047c2e | 5658 | bool fi_before = false; |
bc9ffef3 PZ |
5659 | unsigned long cookie; |
5660 | int i, cpu, occ = 0; | |
5661 | struct rq *rq_i; | |
539f6512 | 5662 | bool need_sync; |
539f6512 PZ |
5663 | |
5664 | if (!sched_core_enabled(rq)) | |
5665 | return __pick_next_task(rq, prev, rf); | |
5666 | ||
5667 | cpu = cpu_of(rq); | |
5668 | ||
5669 | /* Stopper task is switching into idle, no need core-wide selection. */ | |
5670 | if (cpu_is_offline(cpu)) { | |
5671 | /* | |
5672 | * Reset core_pick so that we don't enter the fastpath when | |
5673 | * coming online. core_pick would already be migrated to | |
5674 | * another cpu during offline. | |
5675 | */ | |
5676 | rq->core_pick = NULL; | |
5677 | return __pick_next_task(rq, prev, rf); | |
5678 | } | |
5679 | ||
5680 | /* | |
5681 | * If there were no {en,de}queues since we picked (IOW, the task | |
5682 | * pointers are all still valid), and we haven't scheduled the last | |
5683 | * pick yet, do so now. | |
5684 | * | |
5685 | * rq->core_pick can be NULL if no selection was made for a CPU because | |
5686 | * it was either offline or went offline during a sibling's core-wide | |
5687 | * selection. In this case, do a core-wide selection. | |
5688 | */ | |
5689 | if (rq->core->core_pick_seq == rq->core->core_task_seq && | |
5690 | rq->core->core_pick_seq != rq->core_sched_seq && | |
5691 | rq->core_pick) { | |
5692 | WRITE_ONCE(rq->core_sched_seq, rq->core->core_pick_seq); | |
5693 | ||
5694 | next = rq->core_pick; | |
5695 | if (next != prev) { | |
5696 | put_prev_task(rq, prev); | |
5697 | set_next_task(rq, next); | |
5698 | } | |
5699 | ||
5700 | rq->core_pick = NULL; | |
5701 | return next; | |
5702 | } | |
5703 | ||
5704 | put_prev_task_balance(rq, prev, rf); | |
5705 | ||
5706 | smt_mask = cpu_smt_mask(cpu); | |
7afbba11 JFG |
5707 | need_sync = !!rq->core->core_cookie; |
5708 | ||
5709 | /* reset state */ | |
5710 | rq->core->core_cookie = 0UL; | |
5711 | if (rq->core->core_forceidle) { | |
5712 | need_sync = true; | |
5713 | fi_before = true; | |
5714 | rq->core->core_forceidle = false; | |
5715 | } | |
539f6512 PZ |
5716 | |
5717 | /* | |
5718 | * core->core_task_seq, core->core_pick_seq, rq->core_sched_seq | |
5719 | * | |
5720 | * @task_seq guards the task state ({en,de}queues) | |
5721 | * @pick_seq is the @task_seq we did a selection on | |
5722 | * @sched_seq is the @pick_seq we scheduled | |
5723 | * | |
5724 | * However, preemptions can cause multiple picks on the same task set. | |
5725 | * 'Fix' this by also increasing @task_seq for every pick. | |
5726 | */ | |
5727 | rq->core->core_task_seq++; | |
539f6512 | 5728 | |
7afbba11 JFG |
5729 | /* |
5730 | * Optimize for common case where this CPU has no cookies | |
5731 | * and there are no cookied tasks running on siblings. | |
5732 | */ | |
5733 | if (!need_sync) { | |
bc9ffef3 | 5734 | next = pick_task(rq); |
7afbba11 JFG |
5735 | if (!next->core_cookie) { |
5736 | rq->core_pick = NULL; | |
c6047c2e JFG |
5737 | /* |
5738 | * For robustness, update the min_vruntime_fi for | |
5739 | * unconstrained picks as well. | |
5740 | */ | |
5741 | WARN_ON_ONCE(fi_before); | |
5742 | task_vruntime_update(rq, next, false); | |
7afbba11 JFG |
5743 | goto done; |
5744 | } | |
8039e96f | 5745 | } |
7afbba11 | 5746 | |
bc9ffef3 PZ |
5747 | /* |
5748 | * For each thread: do the regular task pick and find the max prio task | |
5749 | * amongst them. | |
5750 | * | |
5751 | * Tie-break prio towards the current CPU | |
5752 | */ | |
5753 | for_each_cpu_wrap(i, smt_mask, cpu) { | |
5754 | rq_i = cpu_rq(i); | |
539f6512 | 5755 | |
539f6512 PZ |
5756 | if (i != cpu) |
5757 | update_rq_clock(rq_i); | |
bc9ffef3 PZ |
5758 | |
5759 | p = rq_i->core_pick = pick_task(rq_i); | |
5760 | if (!max || prio_less(max, p, fi_before)) | |
5761 | max = p; | |
539f6512 PZ |
5762 | } |
5763 | ||
bc9ffef3 PZ |
5764 | cookie = rq->core->core_cookie = max->core_cookie; |
5765 | ||
539f6512 | 5766 | /* |
bc9ffef3 PZ |
5767 | * For each thread: try and find a runnable task that matches @max or |
5768 | * force idle. | |
539f6512 | 5769 | */ |
bc9ffef3 PZ |
5770 | for_each_cpu(i, smt_mask) { |
5771 | rq_i = cpu_rq(i); | |
5772 | p = rq_i->core_pick; | |
539f6512 | 5773 | |
bc9ffef3 PZ |
5774 | if (!cookie_equals(p, cookie)) { |
5775 | p = NULL; | |
5776 | if (cookie) | |
5777 | p = sched_core_find(rq_i, cookie); | |
7afbba11 | 5778 | if (!p) |
bc9ffef3 PZ |
5779 | p = idle_sched_class.pick_task(rq_i); |
5780 | } | |
539f6512 | 5781 | |
bc9ffef3 | 5782 | rq_i->core_pick = p; |
d2dfa17b | 5783 | |
bc9ffef3 PZ |
5784 | if (p == rq_i->idle) { |
5785 | if (rq_i->nr_running) { | |
c6047c2e JFG |
5786 | rq->core->core_forceidle = true; |
5787 | if (!fi_before) | |
5788 | rq->core->core_forceidle_seq++; | |
5789 | } | |
bc9ffef3 PZ |
5790 | } else { |
5791 | occ++; | |
539f6512 | 5792 | } |
539f6512 PZ |
5793 | } |
5794 | ||
5795 | rq->core->core_pick_seq = rq->core->core_task_seq; | |
5796 | next = rq->core_pick; | |
5797 | rq->core_sched_seq = rq->core->core_pick_seq; | |
5798 | ||
5799 | /* Something should have been selected for current CPU */ | |
5800 | WARN_ON_ONCE(!next); | |
5801 | ||
5802 | /* | |
5803 | * Reschedule siblings | |
5804 | * | |
5805 | * NOTE: L1TF -- at this point we're no longer running the old task and | |
5806 | * sending an IPI (below) ensures the sibling will no longer be running | |
5807 | * their task. This ensures there is no inter-sibling overlap between | |
5808 | * non-matching user state. | |
5809 | */ | |
5810 | for_each_cpu(i, smt_mask) { | |
bc9ffef3 | 5811 | rq_i = cpu_rq(i); |
539f6512 PZ |
5812 | |
5813 | /* | |
5814 | * An online sibling might have gone offline before a task | |
5815 | * could be picked for it, or it might be offline but later | |
5816 | * happen to come online, but its too late and nothing was | |
5817 | * picked for it. That's Ok - it will pick tasks for itself, | |
5818 | * so ignore it. | |
5819 | */ | |
5820 | if (!rq_i->core_pick) | |
5821 | continue; | |
5822 | ||
c6047c2e JFG |
5823 | /* |
5824 | * Update for new !FI->FI transitions, or if continuing to be in !FI: | |
5825 | * fi_before fi update? | |
5826 | * 0 0 1 | |
5827 | * 0 1 1 | |
5828 | * 1 0 1 | |
5829 | * 1 1 0 | |
5830 | */ | |
5831 | if (!(fi_before && rq->core->core_forceidle)) | |
5832 | task_vruntime_update(rq_i, rq_i->core_pick, rq->core->core_forceidle); | |
539f6512 | 5833 | |
d2dfa17b PZ |
5834 | rq_i->core_pick->core_occupation = occ; |
5835 | ||
539f6512 PZ |
5836 | if (i == cpu) { |
5837 | rq_i->core_pick = NULL; | |
5838 | continue; | |
5839 | } | |
5840 | ||
5841 | /* Did we break L1TF mitigation requirements? */ | |
5842 | WARN_ON_ONCE(!cookie_match(next, rq_i->core_pick)); | |
5843 | ||
5844 | if (rq_i->curr == rq_i->core_pick) { | |
5845 | rq_i->core_pick = NULL; | |
5846 | continue; | |
5847 | } | |
5848 | ||
5849 | resched_curr(rq_i); | |
5850 | } | |
5851 | ||
5852 | done: | |
5853 | set_next_task(rq, next); | |
5854 | return next; | |
5855 | } | |
9edeaea1 | 5856 | |
d2dfa17b PZ |
5857 | static bool try_steal_cookie(int this, int that) |
5858 | { | |
5859 | struct rq *dst = cpu_rq(this), *src = cpu_rq(that); | |
5860 | struct task_struct *p; | |
5861 | unsigned long cookie; | |
5862 | bool success = false; | |
5863 | ||
5864 | local_irq_disable(); | |
5865 | double_rq_lock(dst, src); | |
5866 | ||
5867 | cookie = dst->core->core_cookie; | |
5868 | if (!cookie) | |
5869 | goto unlock; | |
5870 | ||
5871 | if (dst->curr != dst->idle) | |
5872 | goto unlock; | |
5873 | ||
5874 | p = sched_core_find(src, cookie); | |
5875 | if (p == src->idle) | |
5876 | goto unlock; | |
5877 | ||
5878 | do { | |
5879 | if (p == src->core_pick || p == src->curr) | |
5880 | goto next; | |
5881 | ||
5882 | if (!cpumask_test_cpu(this, &p->cpus_mask)) | |
5883 | goto next; | |
5884 | ||
5885 | if (p->core_occupation > dst->idle->core_occupation) | |
5886 | goto next; | |
5887 | ||
d2dfa17b PZ |
5888 | deactivate_task(src, p, 0); |
5889 | set_task_cpu(p, this); | |
5890 | activate_task(dst, p, 0); | |
d2dfa17b PZ |
5891 | |
5892 | resched_curr(dst); | |
5893 | ||
5894 | success = true; | |
5895 | break; | |
5896 | ||
5897 | next: | |
5898 | p = sched_core_next(p, cookie); | |
5899 | } while (p); | |
5900 | ||
5901 | unlock: | |
5902 | double_rq_unlock(dst, src); | |
5903 | local_irq_enable(); | |
5904 | ||
5905 | return success; | |
5906 | } | |
5907 | ||
5908 | static bool steal_cookie_task(int cpu, struct sched_domain *sd) | |
5909 | { | |
5910 | int i; | |
5911 | ||
5912 | for_each_cpu_wrap(i, sched_domain_span(sd), cpu) { | |
5913 | if (i == cpu) | |
5914 | continue; | |
5915 | ||
5916 | if (need_resched()) | |
5917 | break; | |
5918 | ||
5919 | if (try_steal_cookie(cpu, i)) | |
5920 | return true; | |
5921 | } | |
5922 | ||
5923 | return false; | |
5924 | } | |
5925 | ||
5926 | static void sched_core_balance(struct rq *rq) | |
5927 | { | |
5928 | struct sched_domain *sd; | |
5929 | int cpu = cpu_of(rq); | |
5930 | ||
5931 | preempt_disable(); | |
5932 | rcu_read_lock(); | |
5933 | raw_spin_rq_unlock_irq(rq); | |
5934 | for_each_domain(cpu, sd) { | |
5935 | if (need_resched()) | |
5936 | break; | |
5937 | ||
5938 | if (steal_cookie_task(cpu, sd)) | |
5939 | break; | |
5940 | } | |
5941 | raw_spin_rq_lock_irq(rq); | |
5942 | rcu_read_unlock(); | |
5943 | preempt_enable(); | |
5944 | } | |
5945 | ||
5946 | static DEFINE_PER_CPU(struct callback_head, core_balance_head); | |
5947 | ||
5948 | void queue_core_balance(struct rq *rq) | |
5949 | { | |
5950 | if (!sched_core_enabled(rq)) | |
5951 | return; | |
5952 | ||
5953 | if (!rq->core->core_cookie) | |
5954 | return; | |
5955 | ||
5956 | if (!rq->nr_running) /* not forced idle */ | |
5957 | return; | |
5958 | ||
5959 | queue_balance_callback(rq, &per_cpu(core_balance_head, rq->cpu), sched_core_balance); | |
5960 | } | |
5961 | ||
3c474b32 | 5962 | static void sched_core_cpu_starting(unsigned int cpu) |
9edeaea1 PZ |
5963 | { |
5964 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
3c474b32 PZ |
5965 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; |
5966 | unsigned long flags; | |
5967 | int t; | |
9edeaea1 | 5968 | |
3c474b32 | 5969 | sched_core_lock(cpu, &flags); |
9edeaea1 | 5970 | |
3c474b32 PZ |
5971 | WARN_ON_ONCE(rq->core != rq); |
5972 | ||
5973 | /* if we're the first, we'll be our own leader */ | |
5974 | if (cpumask_weight(smt_mask) == 1) | |
5975 | goto unlock; | |
5976 | ||
5977 | /* find the leader */ | |
5978 | for_each_cpu(t, smt_mask) { | |
5979 | if (t == cpu) | |
5980 | continue; | |
5981 | rq = cpu_rq(t); | |
5982 | if (rq->core == rq) { | |
5983 | core_rq = rq; | |
5984 | break; | |
9edeaea1 | 5985 | } |
3c474b32 | 5986 | } |
9edeaea1 | 5987 | |
3c474b32 PZ |
5988 | if (WARN_ON_ONCE(!core_rq)) /* whoopsie */ |
5989 | goto unlock; | |
9edeaea1 | 5990 | |
3c474b32 PZ |
5991 | /* install and validate core_rq */ |
5992 | for_each_cpu(t, smt_mask) { | |
5993 | rq = cpu_rq(t); | |
9edeaea1 | 5994 | |
3c474b32 | 5995 | if (t == cpu) |
9edeaea1 | 5996 | rq->core = core_rq; |
3c474b32 PZ |
5997 | |
5998 | WARN_ON_ONCE(rq->core != core_rq); | |
9edeaea1 | 5999 | } |
3c474b32 PZ |
6000 | |
6001 | unlock: | |
6002 | sched_core_unlock(cpu, &flags); | |
9edeaea1 | 6003 | } |
3c474b32 PZ |
6004 | |
6005 | static void sched_core_cpu_deactivate(unsigned int cpu) | |
6006 | { | |
6007 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
6008 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; | |
6009 | unsigned long flags; | |
6010 | int t; | |
6011 | ||
6012 | sched_core_lock(cpu, &flags); | |
6013 | ||
6014 | /* if we're the last man standing, nothing to do */ | |
6015 | if (cpumask_weight(smt_mask) == 1) { | |
6016 | WARN_ON_ONCE(rq->core != rq); | |
6017 | goto unlock; | |
6018 | } | |
6019 | ||
6020 | /* if we're not the leader, nothing to do */ | |
6021 | if (rq->core != rq) | |
6022 | goto unlock; | |
6023 | ||
6024 | /* find a new leader */ | |
6025 | for_each_cpu(t, smt_mask) { | |
6026 | if (t == cpu) | |
6027 | continue; | |
6028 | core_rq = cpu_rq(t); | |
6029 | break; | |
6030 | } | |
6031 | ||
6032 | if (WARN_ON_ONCE(!core_rq)) /* impossible */ | |
6033 | goto unlock; | |
6034 | ||
6035 | /* copy the shared state to the new leader */ | |
6036 | core_rq->core_task_seq = rq->core_task_seq; | |
6037 | core_rq->core_pick_seq = rq->core_pick_seq; | |
6038 | core_rq->core_cookie = rq->core_cookie; | |
6039 | core_rq->core_forceidle = rq->core_forceidle; | |
6040 | core_rq->core_forceidle_seq = rq->core_forceidle_seq; | |
6041 | ||
6042 | /* install new leader */ | |
6043 | for_each_cpu(t, smt_mask) { | |
6044 | rq = cpu_rq(t); | |
6045 | rq->core = core_rq; | |
6046 | } | |
6047 | ||
6048 | unlock: | |
6049 | sched_core_unlock(cpu, &flags); | |
6050 | } | |
6051 | ||
6052 | static inline void sched_core_cpu_dying(unsigned int cpu) | |
6053 | { | |
6054 | struct rq *rq = cpu_rq(cpu); | |
6055 | ||
6056 | if (rq->core != rq) | |
6057 | rq->core = rq; | |
6058 | } | |
6059 | ||
9edeaea1 PZ |
6060 | #else /* !CONFIG_SCHED_CORE */ |
6061 | ||
6062 | static inline void sched_core_cpu_starting(unsigned int cpu) {} | |
3c474b32 PZ |
6063 | static inline void sched_core_cpu_deactivate(unsigned int cpu) {} |
6064 | static inline void sched_core_cpu_dying(unsigned int cpu) {} | |
9edeaea1 | 6065 | |
539f6512 PZ |
6066 | static struct task_struct * |
6067 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
6068 | { | |
6069 | return __pick_next_task(rq, prev, rf); | |
6070 | } | |
6071 | ||
9edeaea1 PZ |
6072 | #endif /* CONFIG_SCHED_CORE */ |
6073 | ||
b4bfa3fc TG |
6074 | /* |
6075 | * Constants for the sched_mode argument of __schedule(). | |
6076 | * | |
6077 | * The mode argument allows RT enabled kernels to differentiate a | |
6078 | * preemption from blocking on an 'sleeping' spin/rwlock. Note that | |
6079 | * SM_MASK_PREEMPT for !RT has all bits set, which allows the compiler to | |
6080 | * optimize the AND operation out and just check for zero. | |
6081 | */ | |
6082 | #define SM_NONE 0x0 | |
6083 | #define SM_PREEMPT 0x1 | |
6991436c TG |
6084 | #define SM_RTLOCK_WAIT 0x2 |
6085 | ||
6086 | #ifndef CONFIG_PREEMPT_RT | |
6087 | # define SM_MASK_PREEMPT (~0U) | |
6088 | #else | |
6089 | # define SM_MASK_PREEMPT SM_PREEMPT | |
6090 | #endif | |
b4bfa3fc | 6091 | |
dd41f596 | 6092 | /* |
c259e01a | 6093 | * __schedule() is the main scheduler function. |
edde96ea PE |
6094 | * |
6095 | * The main means of driving the scheduler and thus entering this function are: | |
6096 | * | |
6097 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
6098 | * | |
6099 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
6100 | * paths. For example, see arch/x86/entry_64.S. | |
6101 | * | |
6102 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
6103 | * interrupt handler scheduler_tick(). | |
6104 | * | |
6105 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
6106 | * task to the run-queue and that's it. | |
6107 | * | |
6108 | * Now, if the new task added to the run-queue preempts the current | |
6109 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
6110 | * called on the nearest possible occasion: | |
6111 | * | |
c1a280b6 | 6112 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
6113 | * |
6114 | * - in syscall or exception context, at the next outmost | |
6115 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
6116 | * spin_unlock()!) | |
6117 | * | |
6118 | * - in IRQ context, return from interrupt-handler to | |
6119 | * preemptible context | |
6120 | * | |
c1a280b6 | 6121 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
6122 | * then at the next: |
6123 | * | |
6124 | * - cond_resched() call | |
6125 | * - explicit schedule() call | |
6126 | * - return from syscall or exception to user-space | |
6127 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 6128 | * |
b30f0e3f | 6129 | * WARNING: must be called with preemption disabled! |
dd41f596 | 6130 | */ |
b4bfa3fc | 6131 | static void __sched notrace __schedule(unsigned int sched_mode) |
dd41f596 IM |
6132 | { |
6133 | struct task_struct *prev, *next; | |
67ca7bde | 6134 | unsigned long *switch_count; |
dbfb089d | 6135 | unsigned long prev_state; |
d8ac8971 | 6136 | struct rq_flags rf; |
dd41f596 | 6137 | struct rq *rq; |
31656519 | 6138 | int cpu; |
dd41f596 | 6139 | |
dd41f596 IM |
6140 | cpu = smp_processor_id(); |
6141 | rq = cpu_rq(cpu); | |
dd41f596 | 6142 | prev = rq->curr; |
dd41f596 | 6143 | |
b4bfa3fc | 6144 | schedule_debug(prev, !!sched_mode); |
1da177e4 | 6145 | |
e0ee463c | 6146 | if (sched_feat(HRTICK) || sched_feat(HRTICK_DL)) |
f333fdc9 | 6147 | hrtick_clear(rq); |
8f4d37ec | 6148 | |
46a5d164 | 6149 | local_irq_disable(); |
b4bfa3fc | 6150 | rcu_note_context_switch(!!sched_mode); |
46a5d164 | 6151 | |
e0acd0a6 ON |
6152 | /* |
6153 | * Make sure that signal_pending_state()->signal_pending() below | |
6154 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
6155 | * done by the caller to avoid the race with signal_wake_up(): |
6156 | * | |
6157 | * __set_current_state(@state) signal_wake_up() | |
6158 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
6159 | * wake_up_state(p, state) | |
6160 | * LOCK rq->lock LOCK p->pi_state | |
6161 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
6162 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 6163 | * |
dbfb089d | 6164 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 6165 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 6166 | */ |
8a8c69c3 | 6167 | rq_lock(rq, &rf); |
d89e588c | 6168 | smp_mb__after_spinlock(); |
1da177e4 | 6169 | |
d1ccc66d IM |
6170 | /* Promote REQ to ACT */ |
6171 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 6172 | update_rq_clock(rq); |
9edfbfed | 6173 | |
246d86b5 | 6174 | switch_count = &prev->nivcsw; |
d136122f | 6175 | |
dbfb089d | 6176 | /* |
d136122f PZ |
6177 | * We must load prev->state once (task_struct::state is volatile), such |
6178 | * that: | |
6179 | * | |
6180 | * - we form a control dependency vs deactivate_task() below. | |
6181 | * - ptrace_{,un}freeze_traced() can change ->state underneath us. | |
dbfb089d | 6182 | */ |
2f064a59 | 6183 | prev_state = READ_ONCE(prev->__state); |
b4bfa3fc | 6184 | if (!(sched_mode & SM_MASK_PREEMPT) && prev_state) { |
dbfb089d | 6185 | if (signal_pending_state(prev_state, prev)) { |
2f064a59 | 6186 | WRITE_ONCE(prev->__state, TASK_RUNNING); |
21aa9af0 | 6187 | } else { |
dbfb089d PZ |
6188 | prev->sched_contributes_to_load = |
6189 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
6190 | !(prev_state & TASK_NOLOAD) && | |
6191 | !(prev->flags & PF_FROZEN); | |
6192 | ||
6193 | if (prev->sched_contributes_to_load) | |
6194 | rq->nr_uninterruptible++; | |
6195 | ||
6196 | /* | |
6197 | * __schedule() ttwu() | |
d136122f PZ |
6198 | * prev_state = prev->state; if (p->on_rq && ...) |
6199 | * if (prev_state) goto out; | |
6200 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
6201 | * p->state = TASK_WAKING | |
6202 | * | |
6203 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
6204 | * |
6205 | * After this, schedule() must not care about p->state any more. | |
6206 | */ | |
bce4dc80 | 6207 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 6208 | |
e33a9bba TH |
6209 | if (prev->in_iowait) { |
6210 | atomic_inc(&rq->nr_iowait); | |
6211 | delayacct_blkio_start(); | |
6212 | } | |
21aa9af0 | 6213 | } |
dd41f596 | 6214 | switch_count = &prev->nvcsw; |
1da177e4 LT |
6215 | } |
6216 | ||
d8ac8971 | 6217 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 6218 | clear_tsk_need_resched(prev); |
f27dde8d | 6219 | clear_preempt_need_resched(); |
c006fac5 PT |
6220 | #ifdef CONFIG_SCHED_DEBUG |
6221 | rq->last_seen_need_resched_ns = 0; | |
6222 | #endif | |
1da177e4 | 6223 | |
1da177e4 | 6224 | if (likely(prev != next)) { |
1da177e4 | 6225 | rq->nr_switches++; |
5311a98f EB |
6226 | /* |
6227 | * RCU users of rcu_dereference(rq->curr) may not see | |
6228 | * changes to task_struct made by pick_next_task(). | |
6229 | */ | |
6230 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
6231 | /* |
6232 | * The membarrier system call requires each architecture | |
6233 | * to have a full memory barrier after updating | |
306e0604 MD |
6234 | * rq->curr, before returning to user-space. |
6235 | * | |
6236 | * Here are the schemes providing that barrier on the | |
6237 | * various architectures: | |
6238 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
6239 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
6240 | * - finish_lock_switch() for weakly-ordered | |
6241 | * architectures where spin_unlock is a full barrier, | |
6242 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
6243 | * is a RELEASE barrier), | |
22e4ebb9 | 6244 | */ |
1da177e4 LT |
6245 | ++*switch_count; |
6246 | ||
af449901 | 6247 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
6248 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
6249 | ||
b4bfa3fc | 6250 | trace_sched_switch(sched_mode & SM_MASK_PREEMPT, prev, next); |
d1ccc66d IM |
6251 | |
6252 | /* Also unlocks the rq: */ | |
6253 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 6254 | } else { |
cb42c9a3 | 6255 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 6256 | |
565790d2 PZ |
6257 | rq_unpin_lock(rq, &rf); |
6258 | __balance_callbacks(rq); | |
5cb9eaa3 | 6259 | raw_spin_rq_unlock_irq(rq); |
565790d2 | 6260 | } |
1da177e4 | 6261 | } |
c259e01a | 6262 | |
9af6528e PZ |
6263 | void __noreturn do_task_dead(void) |
6264 | { | |
d1ccc66d | 6265 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 6266 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
6267 | |
6268 | /* Tell freezer to ignore us: */ | |
6269 | current->flags |= PF_NOFREEZE; | |
6270 | ||
b4bfa3fc | 6271 | __schedule(SM_NONE); |
9af6528e | 6272 | BUG(); |
d1ccc66d IM |
6273 | |
6274 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 6275 | for (;;) |
d1ccc66d | 6276 | cpu_relax(); |
9af6528e PZ |
6277 | } |
6278 | ||
9c40cef2 TG |
6279 | static inline void sched_submit_work(struct task_struct *tsk) |
6280 | { | |
c1cecf88 SAS |
6281 | unsigned int task_flags; |
6282 | ||
b03fbd4f | 6283 | if (task_is_running(tsk)) |
9c40cef2 | 6284 | return; |
6d25be57 | 6285 | |
c1cecf88 | 6286 | task_flags = tsk->flags; |
6d25be57 | 6287 | /* |
b945efcd TG |
6288 | * If a worker goes to sleep, notify and ask workqueue whether it |
6289 | * wants to wake up a task to maintain concurrency. | |
6d25be57 | 6290 | */ |
c1cecf88 | 6291 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
c1cecf88 | 6292 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
6293 | wq_worker_sleeping(tsk); |
6294 | else | |
6295 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
6296 | } |
6297 | ||
b0fdc013 SAS |
6298 | if (tsk_is_pi_blocked(tsk)) |
6299 | return; | |
6300 | ||
9c40cef2 TG |
6301 | /* |
6302 | * If we are going to sleep and we have plugged IO queued, | |
6303 | * make sure to submit it to avoid deadlocks. | |
6304 | */ | |
6305 | if (blk_needs_flush_plug(tsk)) | |
008f75a2 | 6306 | blk_flush_plug(tsk->plug, true); |
9c40cef2 TG |
6307 | } |
6308 | ||
6d25be57 TG |
6309 | static void sched_update_worker(struct task_struct *tsk) |
6310 | { | |
771b53d0 JA |
6311 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6312 | if (tsk->flags & PF_WQ_WORKER) | |
6313 | wq_worker_running(tsk); | |
6314 | else | |
6315 | io_wq_worker_running(tsk); | |
6316 | } | |
6d25be57 TG |
6317 | } |
6318 | ||
722a9f92 | 6319 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 6320 | { |
9c40cef2 TG |
6321 | struct task_struct *tsk = current; |
6322 | ||
6323 | sched_submit_work(tsk); | |
bfd9b2b5 | 6324 | do { |
b30f0e3f | 6325 | preempt_disable(); |
b4bfa3fc | 6326 | __schedule(SM_NONE); |
b30f0e3f | 6327 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 6328 | } while (need_resched()); |
6d25be57 | 6329 | sched_update_worker(tsk); |
c259e01a | 6330 | } |
1da177e4 LT |
6331 | EXPORT_SYMBOL(schedule); |
6332 | ||
8663effb SRV |
6333 | /* |
6334 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
6335 | * state (have scheduled out non-voluntarily) by making sure that all | |
6336 | * tasks have either left the run queue or have gone into user space. | |
6337 | * As idle tasks do not do either, they must not ever be preempted | |
6338 | * (schedule out non-voluntarily). | |
6339 | * | |
6340 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
6341 | * never enables preemption because it does not call sched_submit_work(). | |
6342 | */ | |
6343 | void __sched schedule_idle(void) | |
6344 | { | |
6345 | /* | |
6346 | * As this skips calling sched_submit_work(), which the idle task does | |
6347 | * regardless because that function is a nop when the task is in a | |
6348 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
6349 | * current task can be in any other state. Note, idle is always in the | |
6350 | * TASK_RUNNING state. | |
6351 | */ | |
2f064a59 | 6352 | WARN_ON_ONCE(current->__state); |
8663effb | 6353 | do { |
b4bfa3fc | 6354 | __schedule(SM_NONE); |
8663effb SRV |
6355 | } while (need_resched()); |
6356 | } | |
6357 | ||
6775de49 | 6358 | #if defined(CONFIG_CONTEXT_TRACKING) && !defined(CONFIG_HAVE_CONTEXT_TRACKING_OFFSTACK) |
722a9f92 | 6359 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
6360 | { |
6361 | /* | |
6362 | * If we come here after a random call to set_need_resched(), | |
6363 | * or we have been woken up remotely but the IPI has not yet arrived, | |
6364 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
6365 | * we find a better solution. | |
7cc78f8f AL |
6366 | * |
6367 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 6368 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 6369 | * too frequently to make sense yet. |
20ab65e3 | 6370 | */ |
7cc78f8f | 6371 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 6372 | schedule(); |
7cc78f8f | 6373 | exception_exit(prev_state); |
20ab65e3 FW |
6374 | } |
6375 | #endif | |
6376 | ||
c5491ea7 TG |
6377 | /** |
6378 | * schedule_preempt_disabled - called with preemption disabled | |
6379 | * | |
6380 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
6381 | */ | |
6382 | void __sched schedule_preempt_disabled(void) | |
6383 | { | |
ba74c144 | 6384 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
6385 | schedule(); |
6386 | preempt_disable(); | |
6387 | } | |
6388 | ||
6991436c TG |
6389 | #ifdef CONFIG_PREEMPT_RT |
6390 | void __sched notrace schedule_rtlock(void) | |
6391 | { | |
6392 | do { | |
6393 | preempt_disable(); | |
6394 | __schedule(SM_RTLOCK_WAIT); | |
6395 | sched_preempt_enable_no_resched(); | |
6396 | } while (need_resched()); | |
6397 | } | |
6398 | NOKPROBE_SYMBOL(schedule_rtlock); | |
6399 | #endif | |
6400 | ||
06b1f808 | 6401 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
6402 | { |
6403 | do { | |
47252cfb SR |
6404 | /* |
6405 | * Because the function tracer can trace preempt_count_sub() | |
6406 | * and it also uses preempt_enable/disable_notrace(), if | |
6407 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6408 | * by the function tracer will call this function again and | |
6409 | * cause infinite recursion. | |
6410 | * | |
6411 | * Preemption must be disabled here before the function | |
6412 | * tracer can trace. Break up preempt_disable() into two | |
6413 | * calls. One to disable preemption without fear of being | |
6414 | * traced. The other to still record the preemption latency, | |
6415 | * which can also be traced by the function tracer. | |
6416 | */ | |
499d7955 | 6417 | preempt_disable_notrace(); |
47252cfb | 6418 | preempt_latency_start(1); |
b4bfa3fc | 6419 | __schedule(SM_PREEMPT); |
47252cfb | 6420 | preempt_latency_stop(1); |
499d7955 | 6421 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
6422 | |
6423 | /* | |
6424 | * Check again in case we missed a preemption opportunity | |
6425 | * between schedule and now. | |
6426 | */ | |
a18b5d01 FW |
6427 | } while (need_resched()); |
6428 | } | |
6429 | ||
c1a280b6 | 6430 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 6431 | /* |
a49b4f40 VS |
6432 | * This is the entry point to schedule() from in-kernel preemption |
6433 | * off of preempt_enable. | |
1da177e4 | 6434 | */ |
722a9f92 | 6435 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 6436 | { |
1da177e4 LT |
6437 | /* |
6438 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 6439 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 6440 | */ |
fbb00b56 | 6441 | if (likely(!preemptible())) |
1da177e4 LT |
6442 | return; |
6443 | ||
a18b5d01 | 6444 | preempt_schedule_common(); |
1da177e4 | 6445 | } |
376e2424 | 6446 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 6447 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 6448 | |
2c9a98d3 PZI |
6449 | #ifdef CONFIG_PREEMPT_DYNAMIC |
6450 | DEFINE_STATIC_CALL(preempt_schedule, __preempt_schedule_func); | |
ef72661e | 6451 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule); |
2c9a98d3 PZI |
6452 | #endif |
6453 | ||
6454 | ||
009f60e2 | 6455 | /** |
4eaca0a8 | 6456 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
6457 | * |
6458 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
6459 | * recursion and tracing preempt enabling caused by the tracing | |
6460 | * infrastructure itself. But as tracing can happen in areas coming | |
6461 | * from userspace or just about to enter userspace, a preempt enable | |
6462 | * can occur before user_exit() is called. This will cause the scheduler | |
6463 | * to be called when the system is still in usermode. | |
6464 | * | |
6465 | * To prevent this, the preempt_enable_notrace will use this function | |
6466 | * instead of preempt_schedule() to exit user context if needed before | |
6467 | * calling the scheduler. | |
6468 | */ | |
4eaca0a8 | 6469 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
6470 | { |
6471 | enum ctx_state prev_ctx; | |
6472 | ||
6473 | if (likely(!preemptible())) | |
6474 | return; | |
6475 | ||
6476 | do { | |
47252cfb SR |
6477 | /* |
6478 | * Because the function tracer can trace preempt_count_sub() | |
6479 | * and it also uses preempt_enable/disable_notrace(), if | |
6480 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6481 | * by the function tracer will call this function again and | |
6482 | * cause infinite recursion. | |
6483 | * | |
6484 | * Preemption must be disabled here before the function | |
6485 | * tracer can trace. Break up preempt_disable() into two | |
6486 | * calls. One to disable preemption without fear of being | |
6487 | * traced. The other to still record the preemption latency, | |
6488 | * which can also be traced by the function tracer. | |
6489 | */ | |
3d8f74dd | 6490 | preempt_disable_notrace(); |
47252cfb | 6491 | preempt_latency_start(1); |
009f60e2 ON |
6492 | /* |
6493 | * Needs preempt disabled in case user_exit() is traced | |
6494 | * and the tracer calls preempt_enable_notrace() causing | |
6495 | * an infinite recursion. | |
6496 | */ | |
6497 | prev_ctx = exception_enter(); | |
b4bfa3fc | 6498 | __schedule(SM_PREEMPT); |
009f60e2 ON |
6499 | exception_exit(prev_ctx); |
6500 | ||
47252cfb | 6501 | preempt_latency_stop(1); |
3d8f74dd | 6502 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
6503 | } while (need_resched()); |
6504 | } | |
4eaca0a8 | 6505 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 6506 | |
2c9a98d3 PZI |
6507 | #ifdef CONFIG_PREEMPT_DYNAMIC |
6508 | DEFINE_STATIC_CALL(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
ef72661e | 6509 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule_notrace); |
2c9a98d3 PZI |
6510 | #endif |
6511 | ||
c1a280b6 | 6512 | #endif /* CONFIG_PREEMPTION */ |
1da177e4 | 6513 | |
826bfeb3 PZI |
6514 | #ifdef CONFIG_PREEMPT_DYNAMIC |
6515 | ||
6516 | #include <linux/entry-common.h> | |
6517 | ||
6518 | /* | |
6519 | * SC:cond_resched | |
6520 | * SC:might_resched | |
6521 | * SC:preempt_schedule | |
6522 | * SC:preempt_schedule_notrace | |
6523 | * SC:irqentry_exit_cond_resched | |
6524 | * | |
6525 | * | |
6526 | * NONE: | |
6527 | * cond_resched <- __cond_resched | |
6528 | * might_resched <- RET0 | |
6529 | * preempt_schedule <- NOP | |
6530 | * preempt_schedule_notrace <- NOP | |
6531 | * irqentry_exit_cond_resched <- NOP | |
6532 | * | |
6533 | * VOLUNTARY: | |
6534 | * cond_resched <- __cond_resched | |
6535 | * might_resched <- __cond_resched | |
6536 | * preempt_schedule <- NOP | |
6537 | * preempt_schedule_notrace <- NOP | |
6538 | * irqentry_exit_cond_resched <- NOP | |
6539 | * | |
6540 | * FULL: | |
6541 | * cond_resched <- RET0 | |
6542 | * might_resched <- RET0 | |
6543 | * preempt_schedule <- preempt_schedule | |
6544 | * preempt_schedule_notrace <- preempt_schedule_notrace | |
6545 | * irqentry_exit_cond_resched <- irqentry_exit_cond_resched | |
6546 | */ | |
e59e10f8 PZ |
6547 | |
6548 | enum { | |
c597bfdd FW |
6549 | preempt_dynamic_undefined = -1, |
6550 | preempt_dynamic_none, | |
e59e10f8 PZ |
6551 | preempt_dynamic_voluntary, |
6552 | preempt_dynamic_full, | |
6553 | }; | |
6554 | ||
c597bfdd | 6555 | int preempt_dynamic_mode = preempt_dynamic_undefined; |
e59e10f8 | 6556 | |
1011dcce | 6557 | int sched_dynamic_mode(const char *str) |
826bfeb3 | 6558 | { |
e59e10f8 | 6559 | if (!strcmp(str, "none")) |
7e1b2eb7 | 6560 | return preempt_dynamic_none; |
e59e10f8 PZ |
6561 | |
6562 | if (!strcmp(str, "voluntary")) | |
7e1b2eb7 | 6563 | return preempt_dynamic_voluntary; |
e59e10f8 PZ |
6564 | |
6565 | if (!strcmp(str, "full")) | |
7e1b2eb7 | 6566 | return preempt_dynamic_full; |
e59e10f8 | 6567 | |
c4681f3f | 6568 | return -EINVAL; |
e59e10f8 PZ |
6569 | } |
6570 | ||
1011dcce | 6571 | void sched_dynamic_update(int mode) |
e59e10f8 PZ |
6572 | { |
6573 | /* | |
6574 | * Avoid {NONE,VOLUNTARY} -> FULL transitions from ever ending up in | |
6575 | * the ZERO state, which is invalid. | |
6576 | */ | |
6577 | static_call_update(cond_resched, __cond_resched); | |
6578 | static_call_update(might_resched, __cond_resched); | |
6579 | static_call_update(preempt_schedule, __preempt_schedule_func); | |
6580 | static_call_update(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
6581 | static_call_update(irqentry_exit_cond_resched, irqentry_exit_cond_resched); | |
6582 | ||
6583 | switch (mode) { | |
6584 | case preempt_dynamic_none: | |
826bfeb3 | 6585 | static_call_update(cond_resched, __cond_resched); |
9432bbd9 PZ |
6586 | static_call_update(might_resched, (void *)&__static_call_return0); |
6587 | static_call_update(preempt_schedule, NULL); | |
6588 | static_call_update(preempt_schedule_notrace, NULL); | |
6589 | static_call_update(irqentry_exit_cond_resched, NULL); | |
e59e10f8 PZ |
6590 | pr_info("Dynamic Preempt: none\n"); |
6591 | break; | |
6592 | ||
6593 | case preempt_dynamic_voluntary: | |
826bfeb3 PZI |
6594 | static_call_update(cond_resched, __cond_resched); |
6595 | static_call_update(might_resched, __cond_resched); | |
9432bbd9 PZ |
6596 | static_call_update(preempt_schedule, NULL); |
6597 | static_call_update(preempt_schedule_notrace, NULL); | |
6598 | static_call_update(irqentry_exit_cond_resched, NULL); | |
e59e10f8 PZ |
6599 | pr_info("Dynamic Preempt: voluntary\n"); |
6600 | break; | |
6601 | ||
6602 | case preempt_dynamic_full: | |
9432bbd9 PZ |
6603 | static_call_update(cond_resched, (void *)&__static_call_return0); |
6604 | static_call_update(might_resched, (void *)&__static_call_return0); | |
826bfeb3 PZI |
6605 | static_call_update(preempt_schedule, __preempt_schedule_func); |
6606 | static_call_update(preempt_schedule_notrace, __preempt_schedule_notrace_func); | |
6607 | static_call_update(irqentry_exit_cond_resched, irqentry_exit_cond_resched); | |
e59e10f8 PZ |
6608 | pr_info("Dynamic Preempt: full\n"); |
6609 | break; | |
6610 | } | |
6611 | ||
6612 | preempt_dynamic_mode = mode; | |
6613 | } | |
6614 | ||
6615 | static int __init setup_preempt_mode(char *str) | |
6616 | { | |
6617 | int mode = sched_dynamic_mode(str); | |
6618 | if (mode < 0) { | |
6619 | pr_warn("Dynamic Preempt: unsupported mode: %s\n", str); | |
826bfeb3 PZI |
6620 | return 1; |
6621 | } | |
e59e10f8 PZ |
6622 | |
6623 | sched_dynamic_update(mode); | |
826bfeb3 PZI |
6624 | return 0; |
6625 | } | |
6626 | __setup("preempt=", setup_preempt_mode); | |
6627 | ||
c597bfdd FW |
6628 | static void __init preempt_dynamic_init(void) |
6629 | { | |
6630 | if (preempt_dynamic_mode == preempt_dynamic_undefined) { | |
a8b76910 | 6631 | if (IS_ENABLED(CONFIG_PREEMPT_NONE)) { |
c597bfdd | 6632 | sched_dynamic_update(preempt_dynamic_none); |
a8b76910 | 6633 | } else if (IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY)) { |
c597bfdd FW |
6634 | sched_dynamic_update(preempt_dynamic_voluntary); |
6635 | } else { | |
6636 | /* Default static call setting, nothing to do */ | |
a8b76910 | 6637 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT)); |
c597bfdd FW |
6638 | preempt_dynamic_mode = preempt_dynamic_full; |
6639 | pr_info("Dynamic Preempt: full\n"); | |
6640 | } | |
6641 | } | |
6642 | } | |
6643 | ||
6644 | #else /* !CONFIG_PREEMPT_DYNAMIC */ | |
6645 | ||
6646 | static inline void preempt_dynamic_init(void) { } | |
6647 | ||
6648 | #endif /* #ifdef CONFIG_PREEMPT_DYNAMIC */ | |
826bfeb3 | 6649 | |
1da177e4 | 6650 | /* |
a49b4f40 | 6651 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
6652 | * off of irq context. |
6653 | * Note, that this is called and return with irqs disabled. This will | |
6654 | * protect us against recursive calling from irq. | |
6655 | */ | |
722a9f92 | 6656 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 6657 | { |
b22366cd | 6658 | enum ctx_state prev_state; |
6478d880 | 6659 | |
2ed6e34f | 6660 | /* Catch callers which need to be fixed */ |
f27dde8d | 6661 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 6662 | |
b22366cd FW |
6663 | prev_state = exception_enter(); |
6664 | ||
3a5c359a | 6665 | do { |
3d8f74dd | 6666 | preempt_disable(); |
3a5c359a | 6667 | local_irq_enable(); |
b4bfa3fc | 6668 | __schedule(SM_PREEMPT); |
3a5c359a | 6669 | local_irq_disable(); |
3d8f74dd | 6670 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 6671 | } while (need_resched()); |
b22366cd FW |
6672 | |
6673 | exception_exit(prev_state); | |
1da177e4 LT |
6674 | } |
6675 | ||
ac6424b9 | 6676 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 6677 | void *key) |
1da177e4 | 6678 | { |
062d3f95 | 6679 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 6680 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 6681 | } |
1da177e4 LT |
6682 | EXPORT_SYMBOL(default_wake_function); |
6683 | ||
f558c2b8 PZ |
6684 | static void __setscheduler_prio(struct task_struct *p, int prio) |
6685 | { | |
6686 | if (dl_prio(prio)) | |
6687 | p->sched_class = &dl_sched_class; | |
6688 | else if (rt_prio(prio)) | |
6689 | p->sched_class = &rt_sched_class; | |
6690 | else | |
6691 | p->sched_class = &fair_sched_class; | |
6692 | ||
6693 | p->prio = prio; | |
6694 | } | |
6695 | ||
b29739f9 IM |
6696 | #ifdef CONFIG_RT_MUTEXES |
6697 | ||
acd58620 PZ |
6698 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
6699 | { | |
6700 | if (pi_task) | |
6701 | prio = min(prio, pi_task->prio); | |
6702 | ||
6703 | return prio; | |
6704 | } | |
6705 | ||
6706 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
6707 | { | |
6708 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
6709 | ||
6710 | return __rt_effective_prio(pi_task, prio); | |
6711 | } | |
6712 | ||
b29739f9 IM |
6713 | /* |
6714 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
6715 | * @p: task to boost |
6716 | * @pi_task: donor task | |
b29739f9 IM |
6717 | * |
6718 | * This function changes the 'effective' priority of a task. It does | |
6719 | * not touch ->normal_prio like __setscheduler(). | |
6720 | * | |
c365c292 TG |
6721 | * Used by the rt_mutex code to implement priority inheritance |
6722 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 6723 | */ |
acd58620 | 6724 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 6725 | { |
acd58620 | 6726 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 6727 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 6728 | const struct sched_class *prev_class; |
eb580751 PZ |
6729 | struct rq_flags rf; |
6730 | struct rq *rq; | |
b29739f9 | 6731 | |
acd58620 PZ |
6732 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
6733 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
6734 | ||
6735 | /* | |
6736 | * If nothing changed; bail early. | |
6737 | */ | |
6738 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
6739 | return; | |
b29739f9 | 6740 | |
eb580751 | 6741 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 6742 | update_rq_clock(rq); |
acd58620 PZ |
6743 | /* |
6744 | * Set under pi_lock && rq->lock, such that the value can be used under | |
6745 | * either lock. | |
6746 | * | |
6747 | * Note that there is loads of tricky to make this pointer cache work | |
6748 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
6749 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
6750 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 6751 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
6752 | */ |
6753 | p->pi_top_task = pi_task; | |
6754 | ||
6755 | /* | |
6756 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
6757 | */ | |
6758 | if (prio == p->prio && !dl_prio(prio)) | |
6759 | goto out_unlock; | |
b29739f9 | 6760 | |
1c4dd99b TG |
6761 | /* |
6762 | * Idle task boosting is a nono in general. There is one | |
6763 | * exception, when PREEMPT_RT and NOHZ is active: | |
6764 | * | |
6765 | * The idle task calls get_next_timer_interrupt() and holds | |
6766 | * the timer wheel base->lock on the CPU and another CPU wants | |
6767 | * to access the timer (probably to cancel it). We can safely | |
6768 | * ignore the boosting request, as the idle CPU runs this code | |
6769 | * with interrupts disabled and will complete the lock | |
6770 | * protected section without being interrupted. So there is no | |
6771 | * real need to boost. | |
6772 | */ | |
6773 | if (unlikely(p == rq->idle)) { | |
6774 | WARN_ON(p != rq->curr); | |
6775 | WARN_ON(p->pi_blocked_on); | |
6776 | goto out_unlock; | |
6777 | } | |
6778 | ||
b91473ff | 6779 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 6780 | oldprio = p->prio; |
ff77e468 PZ |
6781 | |
6782 | if (oldprio == prio) | |
6783 | queue_flag &= ~DEQUEUE_MOVE; | |
6784 | ||
83ab0aa0 | 6785 | prev_class = p->sched_class; |
da0c1e65 | 6786 | queued = task_on_rq_queued(p); |
051a1d1a | 6787 | running = task_current(rq, p); |
da0c1e65 | 6788 | if (queued) |
ff77e468 | 6789 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 6790 | if (running) |
f3cd1c4e | 6791 | put_prev_task(rq, p); |
dd41f596 | 6792 | |
2d3d891d DF |
6793 | /* |
6794 | * Boosting condition are: | |
6795 | * 1. -rt task is running and holds mutex A | |
6796 | * --> -dl task blocks on mutex A | |
6797 | * | |
6798 | * 2. -dl task is running and holds mutex A | |
6799 | * --> -dl task blocks on mutex A and could preempt the | |
6800 | * running task | |
6801 | */ | |
6802 | if (dl_prio(prio)) { | |
466af29b | 6803 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
6804 | (pi_task && dl_prio(pi_task->prio) && |
6805 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2279f540 | 6806 | p->dl.pi_se = pi_task->dl.pi_se; |
ff77e468 | 6807 | queue_flag |= ENQUEUE_REPLENISH; |
2279f540 JL |
6808 | } else { |
6809 | p->dl.pi_se = &p->dl; | |
6810 | } | |
2d3d891d DF |
6811 | } else if (rt_prio(prio)) { |
6812 | if (dl_prio(oldprio)) | |
2279f540 | 6813 | p->dl.pi_se = &p->dl; |
2d3d891d | 6814 | if (oldprio < prio) |
ff77e468 | 6815 | queue_flag |= ENQUEUE_HEAD; |
2d3d891d DF |
6816 | } else { |
6817 | if (dl_prio(oldprio)) | |
2279f540 | 6818 | p->dl.pi_se = &p->dl; |
746db944 BS |
6819 | if (rt_prio(oldprio)) |
6820 | p->rt.timeout = 0; | |
2d3d891d | 6821 | } |
dd41f596 | 6822 | |
f558c2b8 | 6823 | __setscheduler_prio(p, prio); |
b29739f9 | 6824 | |
da0c1e65 | 6825 | if (queued) |
ff77e468 | 6826 | enqueue_task(rq, p, queue_flag); |
a399d233 | 6827 | if (running) |
03b7fad1 | 6828 | set_next_task(rq, p); |
cb469845 | 6829 | |
da7a735e | 6830 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 6831 | out_unlock: |
d1ccc66d IM |
6832 | /* Avoid rq from going away on us: */ |
6833 | preempt_disable(); | |
4c9a4bc8 | 6834 | |
565790d2 PZ |
6835 | rq_unpin_lock(rq, &rf); |
6836 | __balance_callbacks(rq); | |
5cb9eaa3 | 6837 | raw_spin_rq_unlock(rq); |
565790d2 | 6838 | |
4c9a4bc8 | 6839 | preempt_enable(); |
b29739f9 | 6840 | } |
acd58620 PZ |
6841 | #else |
6842 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
6843 | { | |
6844 | return prio; | |
6845 | } | |
b29739f9 | 6846 | #endif |
d50dde5a | 6847 | |
36c8b586 | 6848 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6849 | { |
49bd21ef | 6850 | bool queued, running; |
53a23364 | 6851 | int old_prio; |
eb580751 | 6852 | struct rq_flags rf; |
70b97a7f | 6853 | struct rq *rq; |
1da177e4 | 6854 | |
75e45d51 | 6855 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
6856 | return; |
6857 | /* | |
6858 | * We have to be careful, if called from sys_setpriority(), | |
6859 | * the task might be in the middle of scheduling on another CPU. | |
6860 | */ | |
eb580751 | 6861 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
6862 | update_rq_clock(rq); |
6863 | ||
1da177e4 LT |
6864 | /* |
6865 | * The RT priorities are set via sched_setscheduler(), but we still | |
6866 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 6867 | * it won't have any effect on scheduling until the task is |
aab03e05 | 6868 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 6869 | */ |
aab03e05 | 6870 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
6871 | p->static_prio = NICE_TO_PRIO(nice); |
6872 | goto out_unlock; | |
6873 | } | |
da0c1e65 | 6874 | queued = task_on_rq_queued(p); |
49bd21ef | 6875 | running = task_current(rq, p); |
da0c1e65 | 6876 | if (queued) |
7a57f32a | 6877 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
6878 | if (running) |
6879 | put_prev_task(rq, p); | |
1da177e4 | 6880 | |
1da177e4 | 6881 | p->static_prio = NICE_TO_PRIO(nice); |
9059393e | 6882 | set_load_weight(p, true); |
b29739f9 IM |
6883 | old_prio = p->prio; |
6884 | p->prio = effective_prio(p); | |
1da177e4 | 6885 | |
5443a0be | 6886 | if (queued) |
7134b3e9 | 6887 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 6888 | if (running) |
03b7fad1 | 6889 | set_next_task(rq, p); |
5443a0be FW |
6890 | |
6891 | /* | |
6892 | * If the task increased its priority or is running and | |
6893 | * lowered its priority, then reschedule its CPU: | |
6894 | */ | |
6895 | p->sched_class->prio_changed(rq, p, old_prio); | |
6896 | ||
1da177e4 | 6897 | out_unlock: |
eb580751 | 6898 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 6899 | } |
1da177e4 LT |
6900 | EXPORT_SYMBOL(set_user_nice); |
6901 | ||
e43379f1 MM |
6902 | /* |
6903 | * can_nice - check if a task can reduce its nice value | |
6904 | * @p: task | |
6905 | * @nice: nice value | |
6906 | */ | |
36c8b586 | 6907 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6908 | { |
d1ccc66d | 6909 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 6910 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 6911 | |
78d7d407 | 6912 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
6913 | capable(CAP_SYS_NICE)); |
6914 | } | |
6915 | ||
1da177e4 LT |
6916 | #ifdef __ARCH_WANT_SYS_NICE |
6917 | ||
6918 | /* | |
6919 | * sys_nice - change the priority of the current process. | |
6920 | * @increment: priority increment | |
6921 | * | |
6922 | * sys_setpriority is a more generic, but much slower function that | |
6923 | * does similar things. | |
6924 | */ | |
5add95d4 | 6925 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6926 | { |
48f24c4d | 6927 | long nice, retval; |
1da177e4 LT |
6928 | |
6929 | /* | |
6930 | * Setpriority might change our priority at the same moment. | |
6931 | * We don't have to worry. Conceptually one call occurs first | |
6932 | * and we have a single winner. | |
6933 | */ | |
a9467fa3 | 6934 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 6935 | nice = task_nice(current) + increment; |
1da177e4 | 6936 | |
a9467fa3 | 6937 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
6938 | if (increment < 0 && !can_nice(current, nice)) |
6939 | return -EPERM; | |
6940 | ||
1da177e4 LT |
6941 | retval = security_task_setnice(current, nice); |
6942 | if (retval) | |
6943 | return retval; | |
6944 | ||
6945 | set_user_nice(current, nice); | |
6946 | return 0; | |
6947 | } | |
6948 | ||
6949 | #endif | |
6950 | ||
6951 | /** | |
6952 | * task_prio - return the priority value of a given task. | |
6953 | * @p: the task in question. | |
6954 | * | |
e69f6186 | 6955 | * Return: The priority value as seen by users in /proc. |
c541bb78 DE |
6956 | * |
6957 | * sched policy return value kernel prio user prio/nice | |
6958 | * | |
6959 | * normal, batch, idle [0 ... 39] [100 ... 139] 0/[-20 ... 19] | |
6960 | * fifo, rr [-2 ... -100] [98 ... 0] [1 ... 99] | |
6961 | * deadline -101 -1 0 | |
1da177e4 | 6962 | */ |
36c8b586 | 6963 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6964 | { |
6965 | return p->prio - MAX_RT_PRIO; | |
6966 | } | |
6967 | ||
1da177e4 | 6968 | /** |
d1ccc66d | 6969 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 6970 | * @cpu: the processor in question. |
e69f6186 YB |
6971 | * |
6972 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
6973 | */ |
6974 | int idle_cpu(int cpu) | |
6975 | { | |
908a3283 TG |
6976 | struct rq *rq = cpu_rq(cpu); |
6977 | ||
6978 | if (rq->curr != rq->idle) | |
6979 | return 0; | |
6980 | ||
6981 | if (rq->nr_running) | |
6982 | return 0; | |
6983 | ||
6984 | #ifdef CONFIG_SMP | |
126c2092 | 6985 | if (rq->ttwu_pending) |
908a3283 TG |
6986 | return 0; |
6987 | #endif | |
6988 | ||
6989 | return 1; | |
1da177e4 LT |
6990 | } |
6991 | ||
943d355d RJ |
6992 | /** |
6993 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
6994 | * @cpu: the CPU in question. | |
6995 | * | |
6996 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
6997 | */ | |
6998 | int available_idle_cpu(int cpu) | |
6999 | { | |
7000 | if (!idle_cpu(cpu)) | |
7001 | return 0; | |
7002 | ||
247f2f6f RJ |
7003 | if (vcpu_is_preempted(cpu)) |
7004 | return 0; | |
7005 | ||
908a3283 | 7006 | return 1; |
1da177e4 LT |
7007 | } |
7008 | ||
1da177e4 | 7009 | /** |
d1ccc66d | 7010 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 7011 | * @cpu: the processor in question. |
e69f6186 | 7012 | * |
d1ccc66d | 7013 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 7014 | */ |
36c8b586 | 7015 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
7016 | { |
7017 | return cpu_rq(cpu)->idle; | |
7018 | } | |
7019 | ||
7d6a905f VK |
7020 | #ifdef CONFIG_SMP |
7021 | /* | |
7022 | * This function computes an effective utilization for the given CPU, to be | |
7023 | * used for frequency selection given the linear relation: f = u * f_max. | |
7024 | * | |
7025 | * The scheduler tracks the following metrics: | |
7026 | * | |
7027 | * cpu_util_{cfs,rt,dl,irq}() | |
7028 | * cpu_bw_dl() | |
7029 | * | |
7030 | * Where the cfs,rt and dl util numbers are tracked with the same metric and | |
7031 | * synchronized windows and are thus directly comparable. | |
7032 | * | |
7033 | * The cfs,rt,dl utilization are the running times measured with rq->clock_task | |
7034 | * which excludes things like IRQ and steal-time. These latter are then accrued | |
7035 | * in the irq utilization. | |
7036 | * | |
7037 | * The DL bandwidth number otoh is not a measured metric but a value computed | |
7038 | * based on the task model parameters and gives the minimal utilization | |
7039 | * required to meet deadlines. | |
7040 | */ | |
a5418be9 VK |
7041 | unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, |
7042 | unsigned long max, enum cpu_util_type type, | |
7d6a905f VK |
7043 | struct task_struct *p) |
7044 | { | |
7045 | unsigned long dl_util, util, irq; | |
7046 | struct rq *rq = cpu_rq(cpu); | |
7047 | ||
7048 | if (!uclamp_is_used() && | |
7049 | type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) { | |
7050 | return max; | |
7051 | } | |
7052 | ||
7053 | /* | |
7054 | * Early check to see if IRQ/steal time saturates the CPU, can be | |
7055 | * because of inaccuracies in how we track these -- see | |
7056 | * update_irq_load_avg(). | |
7057 | */ | |
7058 | irq = cpu_util_irq(rq); | |
7059 | if (unlikely(irq >= max)) | |
7060 | return max; | |
7061 | ||
7062 | /* | |
7063 | * Because the time spend on RT/DL tasks is visible as 'lost' time to | |
7064 | * CFS tasks and we use the same metric to track the effective | |
7065 | * utilization (PELT windows are synchronized) we can directly add them | |
7066 | * to obtain the CPU's actual utilization. | |
7067 | * | |
7068 | * CFS and RT utilization can be boosted or capped, depending on | |
7069 | * utilization clamp constraints requested by currently RUNNABLE | |
7070 | * tasks. | |
7071 | * When there are no CFS RUNNABLE tasks, clamps are released and | |
7072 | * frequency will be gracefully reduced with the utilization decay. | |
7073 | */ | |
7074 | util = util_cfs + cpu_util_rt(rq); | |
7075 | if (type == FREQUENCY_UTIL) | |
7076 | util = uclamp_rq_util_with(rq, util, p); | |
7077 | ||
7078 | dl_util = cpu_util_dl(rq); | |
7079 | ||
7080 | /* | |
7081 | * For frequency selection we do not make cpu_util_dl() a permanent part | |
7082 | * of this sum because we want to use cpu_bw_dl() later on, but we need | |
7083 | * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such | |
7084 | * that we select f_max when there is no idle time. | |
7085 | * | |
7086 | * NOTE: numerical errors or stop class might cause us to not quite hit | |
7087 | * saturation when we should -- something for later. | |
7088 | */ | |
7089 | if (util + dl_util >= max) | |
7090 | return max; | |
7091 | ||
7092 | /* | |
7093 | * OTOH, for energy computation we need the estimated running time, so | |
7094 | * include util_dl and ignore dl_bw. | |
7095 | */ | |
7096 | if (type == ENERGY_UTIL) | |
7097 | util += dl_util; | |
7098 | ||
7099 | /* | |
7100 | * There is still idle time; further improve the number by using the | |
7101 | * irq metric. Because IRQ/steal time is hidden from the task clock we | |
7102 | * need to scale the task numbers: | |
7103 | * | |
7104 | * max - irq | |
7105 | * U' = irq + --------- * U | |
7106 | * max | |
7107 | */ | |
7108 | util = scale_irq_capacity(util, irq, max); | |
7109 | util += irq; | |
7110 | ||
7111 | /* | |
7112 | * Bandwidth required by DEADLINE must always be granted while, for | |
7113 | * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism | |
7114 | * to gracefully reduce the frequency when no tasks show up for longer | |
7115 | * periods of time. | |
7116 | * | |
7117 | * Ideally we would like to set bw_dl as min/guaranteed freq and util + | |
7118 | * bw_dl as requested freq. However, cpufreq is not yet ready for such | |
7119 | * an interface. So, we only do the latter for now. | |
7120 | */ | |
7121 | if (type == FREQUENCY_UTIL) | |
7122 | util += cpu_bw_dl(rq); | |
7123 | ||
7124 | return min(max, util); | |
7125 | } | |
a5418be9 VK |
7126 | |
7127 | unsigned long sched_cpu_util(int cpu, unsigned long max) | |
7128 | { | |
7129 | return effective_cpu_util(cpu, cpu_util_cfs(cpu_rq(cpu)), max, | |
7130 | ENERGY_UTIL, NULL); | |
7131 | } | |
7d6a905f VK |
7132 | #endif /* CONFIG_SMP */ |
7133 | ||
1da177e4 LT |
7134 | /** |
7135 | * find_process_by_pid - find a process with a matching PID value. | |
7136 | * @pid: the pid in question. | |
e69f6186 YB |
7137 | * |
7138 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 7139 | */ |
a9957449 | 7140 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 7141 | { |
228ebcbe | 7142 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
7143 | } |
7144 | ||
c13db6b1 SR |
7145 | /* |
7146 | * sched_setparam() passes in -1 for its policy, to let the functions | |
7147 | * it calls know not to change it. | |
7148 | */ | |
7149 | #define SETPARAM_POLICY -1 | |
7150 | ||
c365c292 TG |
7151 | static void __setscheduler_params(struct task_struct *p, |
7152 | const struct sched_attr *attr) | |
1da177e4 | 7153 | { |
d50dde5a DF |
7154 | int policy = attr->sched_policy; |
7155 | ||
c13db6b1 | 7156 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
7157 | policy = p->policy; |
7158 | ||
1da177e4 | 7159 | p->policy = policy; |
d50dde5a | 7160 | |
aab03e05 DF |
7161 | if (dl_policy(policy)) |
7162 | __setparam_dl(p, attr); | |
39fd8fd2 | 7163 | else if (fair_policy(policy)) |
d50dde5a DF |
7164 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
7165 | ||
39fd8fd2 PZ |
7166 | /* |
7167 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
7168 | * !rt_policy. Always setting this ensures that things like | |
7169 | * getparam()/getattr() don't report silly values for !rt tasks. | |
7170 | */ | |
7171 | p->rt_priority = attr->sched_priority; | |
383afd09 | 7172 | p->normal_prio = normal_prio(p); |
9059393e | 7173 | set_load_weight(p, true); |
c365c292 | 7174 | } |
39fd8fd2 | 7175 | |
c69e8d9c | 7176 | /* |
d1ccc66d | 7177 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
7178 | */ |
7179 | static bool check_same_owner(struct task_struct *p) | |
7180 | { | |
7181 | const struct cred *cred = current_cred(), *pcred; | |
7182 | bool match; | |
7183 | ||
7184 | rcu_read_lock(); | |
7185 | pcred = __task_cred(p); | |
9c806aa0 EB |
7186 | match = (uid_eq(cred->euid, pcred->euid) || |
7187 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
7188 | rcu_read_unlock(); |
7189 | return match; | |
7190 | } | |
7191 | ||
d50dde5a DF |
7192 | static int __sched_setscheduler(struct task_struct *p, |
7193 | const struct sched_attr *attr, | |
dbc7f069 | 7194 | bool user, bool pi) |
1da177e4 | 7195 | { |
f558c2b8 PZ |
7196 | int oldpolicy = -1, policy = attr->sched_policy; |
7197 | int retval, oldprio, newprio, queued, running; | |
83ab0aa0 | 7198 | const struct sched_class *prev_class; |
565790d2 | 7199 | struct callback_head *head; |
eb580751 | 7200 | struct rq_flags rf; |
ca94c442 | 7201 | int reset_on_fork; |
7a57f32a | 7202 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 7203 | struct rq *rq; |
1da177e4 | 7204 | |
896bbb25 SRV |
7205 | /* The pi code expects interrupts enabled */ |
7206 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 7207 | recheck: |
d1ccc66d | 7208 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
7209 | if (policy < 0) { |
7210 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 7211 | policy = oldpolicy = p->policy; |
ca94c442 | 7212 | } else { |
7479f3c9 | 7213 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 7214 | |
20f9cd2a | 7215 | if (!valid_policy(policy)) |
ca94c442 LP |
7216 | return -EINVAL; |
7217 | } | |
7218 | ||
794a56eb | 7219 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
7220 | return -EINVAL; |
7221 | ||
1da177e4 LT |
7222 | /* |
7223 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
ae18ad28 | 7224 | * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL, |
dd41f596 | 7225 | * SCHED_BATCH and SCHED_IDLE is 0. |
1da177e4 | 7226 | */ |
ae18ad28 | 7227 | if (attr->sched_priority > MAX_RT_PRIO-1) |
1da177e4 | 7228 | return -EINVAL; |
aab03e05 DF |
7229 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
7230 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
7231 | return -EINVAL; |
7232 | ||
37e4ab3f OC |
7233 | /* |
7234 | * Allow unprivileged RT tasks to decrease priority: | |
7235 | */ | |
961ccddd | 7236 | if (user && !capable(CAP_SYS_NICE)) { |
d50dde5a | 7237 | if (fair_policy(policy)) { |
d0ea0268 | 7238 | if (attr->sched_nice < task_nice(p) && |
eaad4513 | 7239 | !can_nice(p, attr->sched_nice)) |
d50dde5a DF |
7240 | return -EPERM; |
7241 | } | |
7242 | ||
e05606d3 | 7243 | if (rt_policy(policy)) { |
a44702e8 ON |
7244 | unsigned long rlim_rtprio = |
7245 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 | 7246 | |
d1ccc66d | 7247 | /* Can't set/change the rt policy: */ |
8dc3e909 ON |
7248 | if (policy != p->policy && !rlim_rtprio) |
7249 | return -EPERM; | |
7250 | ||
d1ccc66d | 7251 | /* Can't increase priority: */ |
d50dde5a DF |
7252 | if (attr->sched_priority > p->rt_priority && |
7253 | attr->sched_priority > rlim_rtprio) | |
8dc3e909 ON |
7254 | return -EPERM; |
7255 | } | |
c02aa73b | 7256 | |
d44753b8 JL |
7257 | /* |
7258 | * Can't set/change SCHED_DEADLINE policy at all for now | |
7259 | * (safest behavior); in the future we would like to allow | |
7260 | * unprivileged DL tasks to increase their relative deadline | |
7261 | * or reduce their runtime (both ways reducing utilization) | |
7262 | */ | |
7263 | if (dl_policy(policy)) | |
7264 | return -EPERM; | |
7265 | ||
dd41f596 | 7266 | /* |
c02aa73b DH |
7267 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
7268 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 7269 | */ |
1da1843f | 7270 | if (task_has_idle_policy(p) && !idle_policy(policy)) { |
d0ea0268 | 7271 | if (!can_nice(p, task_nice(p))) |
c02aa73b DH |
7272 | return -EPERM; |
7273 | } | |
5fe1d75f | 7274 | |
d1ccc66d | 7275 | /* Can't change other user's priorities: */ |
c69e8d9c | 7276 | if (!check_same_owner(p)) |
37e4ab3f | 7277 | return -EPERM; |
ca94c442 | 7278 | |
d1ccc66d | 7279 | /* Normal users shall not reset the sched_reset_on_fork flag: */ |
ca94c442 LP |
7280 | if (p->sched_reset_on_fork && !reset_on_fork) |
7281 | return -EPERM; | |
37e4ab3f | 7282 | } |
1da177e4 | 7283 | |
725aad24 | 7284 | if (user) { |
794a56eb JL |
7285 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
7286 | return -EINVAL; | |
7287 | ||
b0ae1981 | 7288 | retval = security_task_setscheduler(p); |
725aad24 JF |
7289 | if (retval) |
7290 | return retval; | |
7291 | } | |
7292 | ||
a509a7cd PB |
7293 | /* Update task specific "requested" clamps */ |
7294 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
7295 | retval = uclamp_validate(p, attr); | |
7296 | if (retval) | |
7297 | return retval; | |
7298 | } | |
7299 | ||
710da3c8 JL |
7300 | if (pi) |
7301 | cpuset_read_lock(); | |
7302 | ||
b29739f9 | 7303 | /* |
d1ccc66d | 7304 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 7305 | * changing the priority of the task: |
0122ec5b | 7306 | * |
25985edc | 7307 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
7308 | * runqueue lock must be held. |
7309 | */ | |
eb580751 | 7310 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 7311 | update_rq_clock(rq); |
dc61b1d6 | 7312 | |
34f971f6 | 7313 | /* |
d1ccc66d | 7314 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
7315 | */ |
7316 | if (p == rq->stop) { | |
4b211f2b MP |
7317 | retval = -EINVAL; |
7318 | goto unlock; | |
34f971f6 PZ |
7319 | } |
7320 | ||
a51e9198 | 7321 | /* |
d6b1e911 TG |
7322 | * If not changing anything there's no need to proceed further, |
7323 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 7324 | */ |
d50dde5a | 7325 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 7326 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
7327 | goto change; |
7328 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
7329 | goto change; | |
75381608 | 7330 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 7331 | goto change; |
a509a7cd PB |
7332 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
7333 | goto change; | |
d50dde5a | 7334 | |
d6b1e911 | 7335 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
7336 | retval = 0; |
7337 | goto unlock; | |
a51e9198 | 7338 | } |
d50dde5a | 7339 | change: |
a51e9198 | 7340 | |
dc61b1d6 | 7341 | if (user) { |
332ac17e | 7342 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
7343 | /* |
7344 | * Do not allow realtime tasks into groups that have no runtime | |
7345 | * assigned. | |
7346 | */ | |
7347 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
7348 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
7349 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
7350 | retval = -EPERM; |
7351 | goto unlock; | |
dc61b1d6 | 7352 | } |
dc61b1d6 | 7353 | #endif |
332ac17e | 7354 | #ifdef CONFIG_SMP |
794a56eb JL |
7355 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
7356 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 7357 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
7358 | |
7359 | /* | |
7360 | * Don't allow tasks with an affinity mask smaller than | |
7361 | * the entire root_domain to become SCHED_DEADLINE. We | |
7362 | * will also fail if there's no bandwidth available. | |
7363 | */ | |
3bd37062 | 7364 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 7365 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
7366 | retval = -EPERM; |
7367 | goto unlock; | |
332ac17e DF |
7368 | } |
7369 | } | |
7370 | #endif | |
7371 | } | |
dc61b1d6 | 7372 | |
d1ccc66d | 7373 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
7374 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
7375 | policy = oldpolicy = -1; | |
eb580751 | 7376 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
7377 | if (pi) |
7378 | cpuset_read_unlock(); | |
1da177e4 LT |
7379 | goto recheck; |
7380 | } | |
332ac17e DF |
7381 | |
7382 | /* | |
7383 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
7384 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
7385 | * is available. | |
7386 | */ | |
06a76fe0 | 7387 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
7388 | retval = -EBUSY; |
7389 | goto unlock; | |
332ac17e DF |
7390 | } |
7391 | ||
c365c292 TG |
7392 | p->sched_reset_on_fork = reset_on_fork; |
7393 | oldprio = p->prio; | |
7394 | ||
f558c2b8 | 7395 | newprio = __normal_prio(policy, attr->sched_priority, attr->sched_nice); |
dbc7f069 PZ |
7396 | if (pi) { |
7397 | /* | |
7398 | * Take priority boosted tasks into account. If the new | |
7399 | * effective priority is unchanged, we just store the new | |
7400 | * normal parameters and do not touch the scheduler class and | |
7401 | * the runqueue. This will be done when the task deboost | |
7402 | * itself. | |
7403 | */ | |
f558c2b8 PZ |
7404 | newprio = rt_effective_prio(p, newprio); |
7405 | if (newprio == oldprio) | |
ff77e468 | 7406 | queue_flags &= ~DEQUEUE_MOVE; |
c365c292 TG |
7407 | } |
7408 | ||
da0c1e65 | 7409 | queued = task_on_rq_queued(p); |
051a1d1a | 7410 | running = task_current(rq, p); |
da0c1e65 | 7411 | if (queued) |
ff77e468 | 7412 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 7413 | if (running) |
f3cd1c4e | 7414 | put_prev_task(rq, p); |
f6b53205 | 7415 | |
83ab0aa0 | 7416 | prev_class = p->sched_class; |
a509a7cd | 7417 | |
f558c2b8 PZ |
7418 | if (!(attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)) { |
7419 | __setscheduler_params(p, attr); | |
7420 | __setscheduler_prio(p, newprio); | |
7421 | } | |
a509a7cd | 7422 | __setscheduler_uclamp(p, attr); |
f6b53205 | 7423 | |
da0c1e65 | 7424 | if (queued) { |
81a44c54 TG |
7425 | /* |
7426 | * We enqueue to tail when the priority of a task is | |
7427 | * increased (user space view). | |
7428 | */ | |
ff77e468 PZ |
7429 | if (oldprio < p->prio) |
7430 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 7431 | |
ff77e468 | 7432 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 7433 | } |
a399d233 | 7434 | if (running) |
03b7fad1 | 7435 | set_next_task(rq, p); |
cb469845 | 7436 | |
da7a735e | 7437 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
7438 | |
7439 | /* Avoid rq from going away on us: */ | |
7440 | preempt_disable(); | |
565790d2 | 7441 | head = splice_balance_callbacks(rq); |
eb580751 | 7442 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 7443 | |
710da3c8 JL |
7444 | if (pi) { |
7445 | cpuset_read_unlock(); | |
dbc7f069 | 7446 | rt_mutex_adjust_pi(p); |
710da3c8 | 7447 | } |
95e02ca9 | 7448 | |
d1ccc66d | 7449 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 7450 | balance_callbacks(rq, head); |
4c9a4bc8 | 7451 | preempt_enable(); |
95e02ca9 | 7452 | |
1da177e4 | 7453 | return 0; |
4b211f2b MP |
7454 | |
7455 | unlock: | |
7456 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
7457 | if (pi) |
7458 | cpuset_read_unlock(); | |
4b211f2b | 7459 | return retval; |
1da177e4 | 7460 | } |
961ccddd | 7461 | |
7479f3c9 PZ |
7462 | static int _sched_setscheduler(struct task_struct *p, int policy, |
7463 | const struct sched_param *param, bool check) | |
7464 | { | |
7465 | struct sched_attr attr = { | |
7466 | .sched_policy = policy, | |
7467 | .sched_priority = param->sched_priority, | |
7468 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
7469 | }; | |
7470 | ||
c13db6b1 SR |
7471 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
7472 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
7473 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
7474 | policy &= ~SCHED_RESET_ON_FORK; | |
7475 | attr.sched_policy = policy; | |
7476 | } | |
7477 | ||
dbc7f069 | 7478 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 7479 | } |
961ccddd RR |
7480 | /** |
7481 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
7482 | * @p: the task in question. | |
7483 | * @policy: new policy. | |
7484 | * @param: structure containing the new RT priority. | |
7485 | * | |
7318d4cc PZ |
7486 | * Use sched_set_fifo(), read its comment. |
7487 | * | |
e69f6186 YB |
7488 | * Return: 0 on success. An error code otherwise. |
7489 | * | |
961ccddd RR |
7490 | * NOTE that the task may be already dead. |
7491 | */ | |
7492 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 7493 | const struct sched_param *param) |
961ccddd | 7494 | { |
7479f3c9 | 7495 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 7496 | } |
1da177e4 | 7497 | |
d50dde5a DF |
7498 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
7499 | { | |
dbc7f069 | 7500 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 7501 | } |
d50dde5a | 7502 | |
794a56eb JL |
7503 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
7504 | { | |
7505 | return __sched_setscheduler(p, attr, false, true); | |
7506 | } | |
1eb5dde6 | 7507 | EXPORT_SYMBOL_GPL(sched_setattr_nocheck); |
794a56eb | 7508 | |
961ccddd RR |
7509 | /** |
7510 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
7511 | * @p: the task in question. | |
7512 | * @policy: new policy. | |
7513 | * @param: structure containing the new RT priority. | |
7514 | * | |
7515 | * Just like sched_setscheduler, only don't bother checking if the | |
7516 | * current context has permission. For example, this is needed in | |
7517 | * stop_machine(): we create temporary high priority worker threads, | |
7518 | * but our caller might not have that capability. | |
e69f6186 YB |
7519 | * |
7520 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
7521 | */ |
7522 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 7523 | const struct sched_param *param) |
961ccddd | 7524 | { |
7479f3c9 | 7525 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
7526 | } |
7527 | ||
7318d4cc PZ |
7528 | /* |
7529 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
7530 | * incapable of resource management, which is the one thing an OS really should | |
7531 | * be doing. | |
7532 | * | |
7533 | * This is of course the reason it is limited to privileged users only. | |
7534 | * | |
7535 | * Worse still; it is fundamentally impossible to compose static priority | |
7536 | * workloads. You cannot take two correctly working static prio workloads | |
7537 | * and smash them together and still expect them to work. | |
7538 | * | |
7539 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
7540 | * | |
7541 | * MAX_RT_PRIO / 2 | |
7542 | * | |
7543 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
7544 | * know enough information to make a sensible choice. | |
7545 | */ | |
8b700983 | 7546 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
7547 | { |
7548 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 7549 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7550 | } |
7551 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
7552 | ||
7553 | /* | |
7554 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
7555 | */ | |
8b700983 | 7556 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
7557 | { |
7558 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 7559 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7560 | } |
7561 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
7562 | ||
8b700983 | 7563 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
7564 | { |
7565 | struct sched_attr attr = { | |
7566 | .sched_policy = SCHED_NORMAL, | |
7567 | .sched_nice = nice, | |
7568 | }; | |
8b700983 | 7569 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
7570 | } |
7571 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 7572 | |
95cdf3b7 IM |
7573 | static int |
7574 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 7575 | { |
1da177e4 LT |
7576 | struct sched_param lparam; |
7577 | struct task_struct *p; | |
36c8b586 | 7578 | int retval; |
1da177e4 LT |
7579 | |
7580 | if (!param || pid < 0) | |
7581 | return -EINVAL; | |
7582 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
7583 | return -EFAULT; | |
5fe1d75f ON |
7584 | |
7585 | rcu_read_lock(); | |
7586 | retval = -ESRCH; | |
1da177e4 | 7587 | p = find_process_by_pid(pid); |
710da3c8 JL |
7588 | if (likely(p)) |
7589 | get_task_struct(p); | |
5fe1d75f | 7590 | rcu_read_unlock(); |
36c8b586 | 7591 | |
710da3c8 JL |
7592 | if (likely(p)) { |
7593 | retval = sched_setscheduler(p, policy, &lparam); | |
7594 | put_task_struct(p); | |
7595 | } | |
7596 | ||
1da177e4 LT |
7597 | return retval; |
7598 | } | |
7599 | ||
d50dde5a DF |
7600 | /* |
7601 | * Mimics kernel/events/core.c perf_copy_attr(). | |
7602 | */ | |
d1ccc66d | 7603 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
7604 | { |
7605 | u32 size; | |
7606 | int ret; | |
7607 | ||
d1ccc66d | 7608 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
7609 | memset(attr, 0, sizeof(*attr)); |
7610 | ||
7611 | ret = get_user(size, &uattr->size); | |
7612 | if (ret) | |
7613 | return ret; | |
7614 | ||
d1ccc66d IM |
7615 | /* ABI compatibility quirk: */ |
7616 | if (!size) | |
d50dde5a | 7617 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 7618 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
7619 | goto err_size; |
7620 | ||
dff3a85f AS |
7621 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
7622 | if (ret) { | |
7623 | if (ret == -E2BIG) | |
7624 | goto err_size; | |
7625 | return ret; | |
d50dde5a DF |
7626 | } |
7627 | ||
a509a7cd PB |
7628 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
7629 | size < SCHED_ATTR_SIZE_VER1) | |
7630 | return -EINVAL; | |
7631 | ||
d50dde5a | 7632 | /* |
d1ccc66d | 7633 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
7634 | * to be strict and return an error on out-of-bounds values? |
7635 | */ | |
75e45d51 | 7636 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 7637 | |
e78c7bca | 7638 | return 0; |
d50dde5a DF |
7639 | |
7640 | err_size: | |
7641 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 7642 | return -E2BIG; |
d50dde5a DF |
7643 | } |
7644 | ||
f4dddf90 QP |
7645 | static void get_params(struct task_struct *p, struct sched_attr *attr) |
7646 | { | |
7647 | if (task_has_dl_policy(p)) | |
7648 | __getparam_dl(p, attr); | |
7649 | else if (task_has_rt_policy(p)) | |
7650 | attr->sched_priority = p->rt_priority; | |
7651 | else | |
7652 | attr->sched_nice = task_nice(p); | |
7653 | } | |
7654 | ||
1da177e4 LT |
7655 | /** |
7656 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
7657 | * @pid: the pid in question. | |
7658 | * @policy: new policy. | |
7659 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7660 | * |
7661 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7662 | */ |
d1ccc66d | 7663 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 7664 | { |
c21761f1 JB |
7665 | if (policy < 0) |
7666 | return -EINVAL; | |
7667 | ||
1da177e4 LT |
7668 | return do_sched_setscheduler(pid, policy, param); |
7669 | } | |
7670 | ||
7671 | /** | |
7672 | * sys_sched_setparam - set/change the RT priority of a thread | |
7673 | * @pid: the pid in question. | |
7674 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7675 | * |
7676 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7677 | */ |
5add95d4 | 7678 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7679 | { |
c13db6b1 | 7680 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
7681 | } |
7682 | ||
d50dde5a DF |
7683 | /** |
7684 | * sys_sched_setattr - same as above, but with extended sched_attr | |
7685 | * @pid: the pid in question. | |
5778fccf | 7686 | * @uattr: structure containing the extended parameters. |
db66d756 | 7687 | * @flags: for future extension. |
d50dde5a | 7688 | */ |
6d35ab48 PZ |
7689 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
7690 | unsigned int, flags) | |
d50dde5a DF |
7691 | { |
7692 | struct sched_attr attr; | |
7693 | struct task_struct *p; | |
7694 | int retval; | |
7695 | ||
6d35ab48 | 7696 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
7697 | return -EINVAL; |
7698 | ||
143cf23d MK |
7699 | retval = sched_copy_attr(uattr, &attr); |
7700 | if (retval) | |
7701 | return retval; | |
d50dde5a | 7702 | |
b14ed2c2 | 7703 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 7704 | return -EINVAL; |
1d6362fa PB |
7705 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
7706 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
7707 | |
7708 | rcu_read_lock(); | |
7709 | retval = -ESRCH; | |
7710 | p = find_process_by_pid(pid); | |
a509a7cd PB |
7711 | if (likely(p)) |
7712 | get_task_struct(p); | |
d50dde5a DF |
7713 | rcu_read_unlock(); |
7714 | ||
a509a7cd | 7715 | if (likely(p)) { |
f4dddf90 QP |
7716 | if (attr.sched_flags & SCHED_FLAG_KEEP_PARAMS) |
7717 | get_params(p, &attr); | |
a509a7cd PB |
7718 | retval = sched_setattr(p, &attr); |
7719 | put_task_struct(p); | |
7720 | } | |
7721 | ||
d50dde5a DF |
7722 | return retval; |
7723 | } | |
7724 | ||
1da177e4 LT |
7725 | /** |
7726 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
7727 | * @pid: the pid in question. | |
e69f6186 YB |
7728 | * |
7729 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
7730 | * code. | |
1da177e4 | 7731 | */ |
5add95d4 | 7732 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 7733 | { |
36c8b586 | 7734 | struct task_struct *p; |
3a5c359a | 7735 | int retval; |
1da177e4 LT |
7736 | |
7737 | if (pid < 0) | |
3a5c359a | 7738 | return -EINVAL; |
1da177e4 LT |
7739 | |
7740 | retval = -ESRCH; | |
5fe85be0 | 7741 | rcu_read_lock(); |
1da177e4 LT |
7742 | p = find_process_by_pid(pid); |
7743 | if (p) { | |
7744 | retval = security_task_getscheduler(p); | |
7745 | if (!retval) | |
ca94c442 LP |
7746 | retval = p->policy |
7747 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 7748 | } |
5fe85be0 | 7749 | rcu_read_unlock(); |
1da177e4 LT |
7750 | return retval; |
7751 | } | |
7752 | ||
7753 | /** | |
ca94c442 | 7754 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
7755 | * @pid: the pid in question. |
7756 | * @param: structure containing the RT priority. | |
e69f6186 YB |
7757 | * |
7758 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
7759 | * code. | |
1da177e4 | 7760 | */ |
5add95d4 | 7761 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7762 | { |
ce5f7f82 | 7763 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 7764 | struct task_struct *p; |
3a5c359a | 7765 | int retval; |
1da177e4 LT |
7766 | |
7767 | if (!param || pid < 0) | |
3a5c359a | 7768 | return -EINVAL; |
1da177e4 | 7769 | |
5fe85be0 | 7770 | rcu_read_lock(); |
1da177e4 LT |
7771 | p = find_process_by_pid(pid); |
7772 | retval = -ESRCH; | |
7773 | if (!p) | |
7774 | goto out_unlock; | |
7775 | ||
7776 | retval = security_task_getscheduler(p); | |
7777 | if (retval) | |
7778 | goto out_unlock; | |
7779 | ||
ce5f7f82 PZ |
7780 | if (task_has_rt_policy(p)) |
7781 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 7782 | rcu_read_unlock(); |
1da177e4 LT |
7783 | |
7784 | /* | |
7785 | * This one might sleep, we cannot do it with a spinlock held ... | |
7786 | */ | |
7787 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
7788 | ||
1da177e4 LT |
7789 | return retval; |
7790 | ||
7791 | out_unlock: | |
5fe85be0 | 7792 | rcu_read_unlock(); |
1da177e4 LT |
7793 | return retval; |
7794 | } | |
7795 | ||
1251201c IM |
7796 | /* |
7797 | * Copy the kernel size attribute structure (which might be larger | |
7798 | * than what user-space knows about) to user-space. | |
7799 | * | |
7800 | * Note that all cases are valid: user-space buffer can be larger or | |
7801 | * smaller than the kernel-space buffer. The usual case is that both | |
7802 | * have the same size. | |
7803 | */ | |
7804 | static int | |
7805 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
7806 | struct sched_attr *kattr, | |
7807 | unsigned int usize) | |
d50dde5a | 7808 | { |
1251201c | 7809 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 7810 | |
96d4f267 | 7811 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
7812 | return -EFAULT; |
7813 | ||
7814 | /* | |
1251201c IM |
7815 | * sched_getattr() ABI forwards and backwards compatibility: |
7816 | * | |
7817 | * If usize == ksize then we just copy everything to user-space and all is good. | |
7818 | * | |
7819 | * If usize < ksize then we only copy as much as user-space has space for, | |
7820 | * this keeps ABI compatibility as well. We skip the rest. | |
7821 | * | |
7822 | * If usize > ksize then user-space is using a newer version of the ABI, | |
7823 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
7824 | * detect the kernel's knowledge of attributes from the attr->size value | |
7825 | * which is set to ksize in this case. | |
d50dde5a | 7826 | */ |
1251201c | 7827 | kattr->size = min(usize, ksize); |
d50dde5a | 7828 | |
1251201c | 7829 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
7830 | return -EFAULT; |
7831 | ||
22400674 | 7832 | return 0; |
d50dde5a DF |
7833 | } |
7834 | ||
7835 | /** | |
aab03e05 | 7836 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 7837 | * @pid: the pid in question. |
5778fccf | 7838 | * @uattr: structure containing the extended parameters. |
dff3a85f | 7839 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 7840 | * @flags: for future extension. |
d50dde5a | 7841 | */ |
6d35ab48 | 7842 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 7843 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 7844 | { |
1251201c | 7845 | struct sched_attr kattr = { }; |
d50dde5a DF |
7846 | struct task_struct *p; |
7847 | int retval; | |
7848 | ||
1251201c IM |
7849 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
7850 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
7851 | return -EINVAL; |
7852 | ||
7853 | rcu_read_lock(); | |
7854 | p = find_process_by_pid(pid); | |
7855 | retval = -ESRCH; | |
7856 | if (!p) | |
7857 | goto out_unlock; | |
7858 | ||
7859 | retval = security_task_getscheduler(p); | |
7860 | if (retval) | |
7861 | goto out_unlock; | |
7862 | ||
1251201c | 7863 | kattr.sched_policy = p->policy; |
7479f3c9 | 7864 | if (p->sched_reset_on_fork) |
1251201c | 7865 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
f4dddf90 | 7866 | get_params(p, &kattr); |
7ad721bf | 7867 | kattr.sched_flags &= SCHED_FLAG_ALL; |
d50dde5a | 7868 | |
a509a7cd | 7869 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
7870 | /* |
7871 | * This could race with another potential updater, but this is fine | |
7872 | * because it'll correctly read the old or the new value. We don't need | |
7873 | * to guarantee who wins the race as long as it doesn't return garbage. | |
7874 | */ | |
1251201c IM |
7875 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
7876 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
7877 | #endif |
7878 | ||
d50dde5a DF |
7879 | rcu_read_unlock(); |
7880 | ||
1251201c | 7881 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
7882 | |
7883 | out_unlock: | |
7884 | rcu_read_unlock(); | |
7885 | return retval; | |
7886 | } | |
7887 | ||
234b8ab6 WD |
7888 | #ifdef CONFIG_SMP |
7889 | int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask) | |
1da177e4 | 7890 | { |
234b8ab6 WD |
7891 | int ret = 0; |
7892 | ||
7893 | /* | |
7894 | * If the task isn't a deadline task or admission control is | |
7895 | * disabled then we don't care about affinity changes. | |
7896 | */ | |
7897 | if (!task_has_dl_policy(p) || !dl_bandwidth_enabled()) | |
7898 | return 0; | |
7899 | ||
7900 | /* | |
7901 | * Since bandwidth control happens on root_domain basis, | |
7902 | * if admission test is enabled, we only admit -deadline | |
7903 | * tasks allowed to run on all the CPUs in the task's | |
7904 | * root_domain. | |
7905 | */ | |
7906 | rcu_read_lock(); | |
7907 | if (!cpumask_subset(task_rq(p)->rd->span, mask)) | |
7908 | ret = -EBUSY; | |
7909 | rcu_read_unlock(); | |
7910 | return ret; | |
7911 | } | |
7912 | #endif | |
7913 | ||
db3b02ae WD |
7914 | static int |
7915 | __sched_setaffinity(struct task_struct *p, const struct cpumask *mask) | |
1da177e4 | 7916 | { |
36c8b586 | 7917 | int retval; |
5a16f3d3 | 7918 | cpumask_var_t cpus_allowed, new_mask; |
1da177e4 | 7919 | |
db3b02ae WD |
7920 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) |
7921 | return -ENOMEM; | |
1da177e4 | 7922 | |
5a16f3d3 RR |
7923 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { |
7924 | retval = -ENOMEM; | |
7925 | goto out_free_cpus_allowed; | |
7926 | } | |
e4099a5e PZ |
7927 | |
7928 | cpuset_cpus_allowed(p, cpus_allowed); | |
db3b02ae | 7929 | cpumask_and(new_mask, mask, cpus_allowed); |
e4099a5e | 7930 | |
234b8ab6 WD |
7931 | retval = dl_task_check_affinity(p, new_mask); |
7932 | if (retval) | |
7933 | goto out_free_new_mask; | |
49246274 | 7934 | again: |
07ec77a1 | 7935 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK | SCA_USER); |
db3b02ae WD |
7936 | if (retval) |
7937 | goto out_free_new_mask; | |
1da177e4 | 7938 | |
db3b02ae WD |
7939 | cpuset_cpus_allowed(p, cpus_allowed); |
7940 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
7941 | /* | |
7942 | * We must have raced with a concurrent cpuset update. | |
7943 | * Just reset the cpumask to the cpuset's cpus_allowed. | |
7944 | */ | |
7945 | cpumask_copy(new_mask, cpus_allowed); | |
7946 | goto again; | |
8707d8b8 | 7947 | } |
db3b02ae | 7948 | |
16303ab2 | 7949 | out_free_new_mask: |
5a16f3d3 RR |
7950 | free_cpumask_var(new_mask); |
7951 | out_free_cpus_allowed: | |
7952 | free_cpumask_var(cpus_allowed); | |
db3b02ae WD |
7953 | return retval; |
7954 | } | |
7955 | ||
7956 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) | |
7957 | { | |
36c8b586 IM |
7958 | struct task_struct *p; |
7959 | int retval; | |
1da177e4 | 7960 | |
23f5d142 | 7961 | rcu_read_lock(); |
1da177e4 LT |
7962 | |
7963 | p = find_process_by_pid(pid); | |
7964 | if (!p) { | |
23f5d142 | 7965 | rcu_read_unlock(); |
1da177e4 LT |
7966 | return -ESRCH; |
7967 | } | |
7968 | ||
23f5d142 | 7969 | /* Prevent p going away */ |
1da177e4 | 7970 | get_task_struct(p); |
23f5d142 | 7971 | rcu_read_unlock(); |
1da177e4 | 7972 | |
14a40ffc TH |
7973 | if (p->flags & PF_NO_SETAFFINITY) { |
7974 | retval = -EINVAL; | |
7975 | goto out_put_task; | |
7976 | } | |
db3b02ae | 7977 | |
4c44aaaf EB |
7978 | if (!check_same_owner(p)) { |
7979 | rcu_read_lock(); | |
7980 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
7981 | rcu_read_unlock(); | |
db3b02ae WD |
7982 | retval = -EPERM; |
7983 | goto out_put_task; | |
4c44aaaf EB |
7984 | } |
7985 | rcu_read_unlock(); | |
7986 | } | |
1da177e4 | 7987 | |
b0ae1981 | 7988 | retval = security_task_setscheduler(p); |
e7834f8f | 7989 | if (retval) |
db3b02ae | 7990 | goto out_put_task; |
1da177e4 | 7991 | |
db3b02ae | 7992 | retval = __sched_setaffinity(p, in_mask); |
5a16f3d3 | 7993 | out_put_task: |
1da177e4 | 7994 | put_task_struct(p); |
1da177e4 LT |
7995 | return retval; |
7996 | } | |
7997 | ||
7998 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 7999 | struct cpumask *new_mask) |
1da177e4 | 8000 | { |
96f874e2 RR |
8001 | if (len < cpumask_size()) |
8002 | cpumask_clear(new_mask); | |
8003 | else if (len > cpumask_size()) | |
8004 | len = cpumask_size(); | |
8005 | ||
1da177e4 LT |
8006 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
8007 | } | |
8008 | ||
8009 | /** | |
d1ccc66d | 8010 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
8011 | * @pid: pid of the process |
8012 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8013 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
8014 | * |
8015 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 8016 | */ |
5add95d4 HC |
8017 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
8018 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 8019 | { |
5a16f3d3 | 8020 | cpumask_var_t new_mask; |
1da177e4 LT |
8021 | int retval; |
8022 | ||
5a16f3d3 RR |
8023 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
8024 | return -ENOMEM; | |
1da177e4 | 8025 | |
5a16f3d3 RR |
8026 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
8027 | if (retval == 0) | |
8028 | retval = sched_setaffinity(pid, new_mask); | |
8029 | free_cpumask_var(new_mask); | |
8030 | return retval; | |
1da177e4 LT |
8031 | } |
8032 | ||
96f874e2 | 8033 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 8034 | { |
36c8b586 | 8035 | struct task_struct *p; |
31605683 | 8036 | unsigned long flags; |
1da177e4 | 8037 | int retval; |
1da177e4 | 8038 | |
23f5d142 | 8039 | rcu_read_lock(); |
1da177e4 LT |
8040 | |
8041 | retval = -ESRCH; | |
8042 | p = find_process_by_pid(pid); | |
8043 | if (!p) | |
8044 | goto out_unlock; | |
8045 | ||
e7834f8f DQ |
8046 | retval = security_task_getscheduler(p); |
8047 | if (retval) | |
8048 | goto out_unlock; | |
8049 | ||
013fdb80 | 8050 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 8051 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 8052 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
8053 | |
8054 | out_unlock: | |
23f5d142 | 8055 | rcu_read_unlock(); |
1da177e4 | 8056 | |
9531b62f | 8057 | return retval; |
1da177e4 LT |
8058 | } |
8059 | ||
8060 | /** | |
d1ccc66d | 8061 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
8062 | * @pid: pid of the process |
8063 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8064 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 8065 | * |
599b4840 ZW |
8066 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
8067 | * error code otherwise. | |
1da177e4 | 8068 | */ |
5add95d4 HC |
8069 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
8070 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
8071 | { |
8072 | int ret; | |
f17c8607 | 8073 | cpumask_var_t mask; |
1da177e4 | 8074 | |
84fba5ec | 8075 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
8076 | return -EINVAL; |
8077 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
8078 | return -EINVAL; |
8079 | ||
f17c8607 RR |
8080 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
8081 | return -ENOMEM; | |
1da177e4 | 8082 | |
f17c8607 RR |
8083 | ret = sched_getaffinity(pid, mask); |
8084 | if (ret == 0) { | |
4de373a1 | 8085 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
8086 | |
8087 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
8088 | ret = -EFAULT; |
8089 | else | |
cd3d8031 | 8090 | ret = retlen; |
f17c8607 RR |
8091 | } |
8092 | free_cpumask_var(mask); | |
1da177e4 | 8093 | |
f17c8607 | 8094 | return ret; |
1da177e4 LT |
8095 | } |
8096 | ||
7d4dd4f1 | 8097 | static void do_sched_yield(void) |
1da177e4 | 8098 | { |
8a8c69c3 PZ |
8099 | struct rq_flags rf; |
8100 | struct rq *rq; | |
8101 | ||
246b3b33 | 8102 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 8103 | |
ae92882e | 8104 | schedstat_inc(rq->yld_count); |
4530d7ab | 8105 | current->sched_class->yield_task(rq); |
1da177e4 | 8106 | |
8a8c69c3 | 8107 | preempt_disable(); |
345a957f | 8108 | rq_unlock_irq(rq, &rf); |
ba74c144 | 8109 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
8110 | |
8111 | schedule(); | |
7d4dd4f1 | 8112 | } |
1da177e4 | 8113 | |
59a74b15 MCC |
8114 | /** |
8115 | * sys_sched_yield - yield the current processor to other threads. | |
8116 | * | |
8117 | * This function yields the current CPU to other tasks. If there are no | |
8118 | * other threads running on this CPU then this function will return. | |
8119 | * | |
8120 | * Return: 0. | |
8121 | */ | |
7d4dd4f1 DB |
8122 | SYSCALL_DEFINE0(sched_yield) |
8123 | { | |
8124 | do_sched_yield(); | |
1da177e4 LT |
8125 | return 0; |
8126 | } | |
8127 | ||
b965f1dd PZI |
8128 | #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) |
8129 | int __sched __cond_resched(void) | |
1da177e4 | 8130 | { |
fe32d3cd | 8131 | if (should_resched(0)) { |
a18b5d01 | 8132 | preempt_schedule_common(); |
1da177e4 LT |
8133 | return 1; |
8134 | } | |
50895825 FW |
8135 | /* |
8136 | * In preemptible kernels, ->rcu_read_lock_nesting tells the tick | |
8137 | * whether the current CPU is in an RCU read-side critical section, | |
8138 | * so the tick can report quiescent states even for CPUs looping | |
8139 | * in kernel context. In contrast, in non-preemptible kernels, | |
8140 | * RCU readers leave no in-memory hints, which means that CPU-bound | |
8141 | * processes executing in kernel context might never report an | |
8142 | * RCU quiescent state. Therefore, the following code causes | |
8143 | * cond_resched() to report a quiescent state, but only when RCU | |
8144 | * is in urgent need of one. | |
8145 | */ | |
b965f1dd | 8146 | #ifndef CONFIG_PREEMPT_RCU |
f79c3ad6 | 8147 | rcu_all_qs(); |
b965f1dd | 8148 | #endif |
1da177e4 LT |
8149 | return 0; |
8150 | } | |
b965f1dd PZI |
8151 | EXPORT_SYMBOL(__cond_resched); |
8152 | #endif | |
8153 | ||
8154 | #ifdef CONFIG_PREEMPT_DYNAMIC | |
8155 | DEFINE_STATIC_CALL_RET0(cond_resched, __cond_resched); | |
ef72661e | 8156 | EXPORT_STATIC_CALL_TRAMP(cond_resched); |
b965f1dd PZI |
8157 | |
8158 | DEFINE_STATIC_CALL_RET0(might_resched, __cond_resched); | |
ef72661e | 8159 | EXPORT_STATIC_CALL_TRAMP(might_resched); |
35a773a0 | 8160 | #endif |
1da177e4 LT |
8161 | |
8162 | /* | |
613afbf8 | 8163 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
8164 | * call schedule, and on return reacquire the lock. |
8165 | * | |
c1a280b6 | 8166 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
8167 | * operations here to prevent schedule() from being called twice (once via |
8168 | * spin_unlock(), once by hand). | |
8169 | */ | |
613afbf8 | 8170 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 8171 | { |
fe32d3cd | 8172 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
8173 | int ret = 0; |
8174 | ||
f607c668 PZ |
8175 | lockdep_assert_held(lock); |
8176 | ||
4a81e832 | 8177 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 8178 | spin_unlock(lock); |
d86ee480 | 8179 | if (resched) |
a18b5d01 | 8180 | preempt_schedule_common(); |
95c354fe NP |
8181 | else |
8182 | cpu_relax(); | |
6df3cecb | 8183 | ret = 1; |
1da177e4 | 8184 | spin_lock(lock); |
1da177e4 | 8185 | } |
6df3cecb | 8186 | return ret; |
1da177e4 | 8187 | } |
613afbf8 | 8188 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 8189 | |
f3d4b4b1 BG |
8190 | int __cond_resched_rwlock_read(rwlock_t *lock) |
8191 | { | |
8192 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8193 | int ret = 0; | |
8194 | ||
8195 | lockdep_assert_held_read(lock); | |
8196 | ||
8197 | if (rwlock_needbreak(lock) || resched) { | |
8198 | read_unlock(lock); | |
8199 | if (resched) | |
8200 | preempt_schedule_common(); | |
8201 | else | |
8202 | cpu_relax(); | |
8203 | ret = 1; | |
8204 | read_lock(lock); | |
8205 | } | |
8206 | return ret; | |
8207 | } | |
8208 | EXPORT_SYMBOL(__cond_resched_rwlock_read); | |
8209 | ||
8210 | int __cond_resched_rwlock_write(rwlock_t *lock) | |
8211 | { | |
8212 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8213 | int ret = 0; | |
8214 | ||
8215 | lockdep_assert_held_write(lock); | |
8216 | ||
8217 | if (rwlock_needbreak(lock) || resched) { | |
8218 | write_unlock(lock); | |
8219 | if (resched) | |
8220 | preempt_schedule_common(); | |
8221 | else | |
8222 | cpu_relax(); | |
8223 | ret = 1; | |
8224 | write_lock(lock); | |
8225 | } | |
8226 | return ret; | |
8227 | } | |
8228 | EXPORT_SYMBOL(__cond_resched_rwlock_write); | |
8229 | ||
1da177e4 LT |
8230 | /** |
8231 | * yield - yield the current processor to other threads. | |
8232 | * | |
8e3fabfd PZ |
8233 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
8234 | * | |
8235 | * The scheduler is at all times free to pick the calling task as the most | |
8236 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 8237 | * it, it's already broken. |
8e3fabfd PZ |
8238 | * |
8239 | * Typical broken usage is: | |
8240 | * | |
8241 | * while (!event) | |
d1ccc66d | 8242 | * yield(); |
8e3fabfd PZ |
8243 | * |
8244 | * where one assumes that yield() will let 'the other' process run that will | |
8245 | * make event true. If the current task is a SCHED_FIFO task that will never | |
8246 | * happen. Never use yield() as a progress guarantee!! | |
8247 | * | |
8248 | * If you want to use yield() to wait for something, use wait_event(). | |
8249 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
8250 | * If you still want to use yield(), do not! | |
1da177e4 LT |
8251 | */ |
8252 | void __sched yield(void) | |
8253 | { | |
8254 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 8255 | do_sched_yield(); |
1da177e4 | 8256 | } |
1da177e4 LT |
8257 | EXPORT_SYMBOL(yield); |
8258 | ||
d95f4122 MG |
8259 | /** |
8260 | * yield_to - yield the current processor to another thread in | |
8261 | * your thread group, or accelerate that thread toward the | |
8262 | * processor it's on. | |
16addf95 RD |
8263 | * @p: target task |
8264 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
8265 | * |
8266 | * It's the caller's job to ensure that the target task struct | |
8267 | * can't go away on us before we can do any checks. | |
8268 | * | |
e69f6186 | 8269 | * Return: |
7b270f60 PZ |
8270 | * true (>0) if we indeed boosted the target task. |
8271 | * false (0) if we failed to boost the target. | |
8272 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 8273 | */ |
fa93384f | 8274 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
8275 | { |
8276 | struct task_struct *curr = current; | |
8277 | struct rq *rq, *p_rq; | |
8278 | unsigned long flags; | |
c3c18640 | 8279 | int yielded = 0; |
d95f4122 MG |
8280 | |
8281 | local_irq_save(flags); | |
8282 | rq = this_rq(); | |
8283 | ||
8284 | again: | |
8285 | p_rq = task_rq(p); | |
7b270f60 PZ |
8286 | /* |
8287 | * If we're the only runnable task on the rq and target rq also | |
8288 | * has only one task, there's absolutely no point in yielding. | |
8289 | */ | |
8290 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
8291 | yielded = -ESRCH; | |
8292 | goto out_irq; | |
8293 | } | |
8294 | ||
d95f4122 | 8295 | double_rq_lock(rq, p_rq); |
39e24d8f | 8296 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
8297 | double_rq_unlock(rq, p_rq); |
8298 | goto again; | |
8299 | } | |
8300 | ||
8301 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 8302 | goto out_unlock; |
d95f4122 MG |
8303 | |
8304 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 8305 | goto out_unlock; |
d95f4122 | 8306 | |
b03fbd4f | 8307 | if (task_running(p_rq, p) || !task_is_running(p)) |
7b270f60 | 8308 | goto out_unlock; |
d95f4122 | 8309 | |
0900acf2 | 8310 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 8311 | if (yielded) { |
ae92882e | 8312 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
8313 | /* |
8314 | * Make p's CPU reschedule; pick_next_entity takes care of | |
8315 | * fairness. | |
8316 | */ | |
8317 | if (preempt && rq != p_rq) | |
8875125e | 8318 | resched_curr(p_rq); |
6d1cafd8 | 8319 | } |
d95f4122 | 8320 | |
7b270f60 | 8321 | out_unlock: |
d95f4122 | 8322 | double_rq_unlock(rq, p_rq); |
7b270f60 | 8323 | out_irq: |
d95f4122 MG |
8324 | local_irq_restore(flags); |
8325 | ||
7b270f60 | 8326 | if (yielded > 0) |
d95f4122 MG |
8327 | schedule(); |
8328 | ||
8329 | return yielded; | |
8330 | } | |
8331 | EXPORT_SYMBOL_GPL(yield_to); | |
8332 | ||
10ab5643 TH |
8333 | int io_schedule_prepare(void) |
8334 | { | |
8335 | int old_iowait = current->in_iowait; | |
8336 | ||
8337 | current->in_iowait = 1; | |
008f75a2 CH |
8338 | if (current->plug) |
8339 | blk_flush_plug(current->plug, true); | |
10ab5643 TH |
8340 | |
8341 | return old_iowait; | |
8342 | } | |
8343 | ||
8344 | void io_schedule_finish(int token) | |
8345 | { | |
8346 | current->in_iowait = token; | |
8347 | } | |
8348 | ||
1da177e4 | 8349 | /* |
41a2d6cf | 8350 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 8351 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 8352 | */ |
1da177e4 LT |
8353 | long __sched io_schedule_timeout(long timeout) |
8354 | { | |
10ab5643 | 8355 | int token; |
1da177e4 LT |
8356 | long ret; |
8357 | ||
10ab5643 | 8358 | token = io_schedule_prepare(); |
1da177e4 | 8359 | ret = schedule_timeout(timeout); |
10ab5643 | 8360 | io_schedule_finish(token); |
9cff8ade | 8361 | |
1da177e4 LT |
8362 | return ret; |
8363 | } | |
9cff8ade | 8364 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 8365 | |
e3b929b0 | 8366 | void __sched io_schedule(void) |
10ab5643 TH |
8367 | { |
8368 | int token; | |
8369 | ||
8370 | token = io_schedule_prepare(); | |
8371 | schedule(); | |
8372 | io_schedule_finish(token); | |
8373 | } | |
8374 | EXPORT_SYMBOL(io_schedule); | |
8375 | ||
1da177e4 LT |
8376 | /** |
8377 | * sys_sched_get_priority_max - return maximum RT priority. | |
8378 | * @policy: scheduling class. | |
8379 | * | |
e69f6186 YB |
8380 | * Return: On success, this syscall returns the maximum |
8381 | * rt_priority that can be used by a given scheduling class. | |
8382 | * On failure, a negative error code is returned. | |
1da177e4 | 8383 | */ |
5add95d4 | 8384 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
8385 | { |
8386 | int ret = -EINVAL; | |
8387 | ||
8388 | switch (policy) { | |
8389 | case SCHED_FIFO: | |
8390 | case SCHED_RR: | |
ae18ad28 | 8391 | ret = MAX_RT_PRIO-1; |
1da177e4 | 8392 | break; |
aab03e05 | 8393 | case SCHED_DEADLINE: |
1da177e4 | 8394 | case SCHED_NORMAL: |
b0a9499c | 8395 | case SCHED_BATCH: |
dd41f596 | 8396 | case SCHED_IDLE: |
1da177e4 LT |
8397 | ret = 0; |
8398 | break; | |
8399 | } | |
8400 | return ret; | |
8401 | } | |
8402 | ||
8403 | /** | |
8404 | * sys_sched_get_priority_min - return minimum RT priority. | |
8405 | * @policy: scheduling class. | |
8406 | * | |
e69f6186 YB |
8407 | * Return: On success, this syscall returns the minimum |
8408 | * rt_priority that can be used by a given scheduling class. | |
8409 | * On failure, a negative error code is returned. | |
1da177e4 | 8410 | */ |
5add95d4 | 8411 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
8412 | { |
8413 | int ret = -EINVAL; | |
8414 | ||
8415 | switch (policy) { | |
8416 | case SCHED_FIFO: | |
8417 | case SCHED_RR: | |
8418 | ret = 1; | |
8419 | break; | |
aab03e05 | 8420 | case SCHED_DEADLINE: |
1da177e4 | 8421 | case SCHED_NORMAL: |
b0a9499c | 8422 | case SCHED_BATCH: |
dd41f596 | 8423 | case SCHED_IDLE: |
1da177e4 LT |
8424 | ret = 0; |
8425 | } | |
8426 | return ret; | |
8427 | } | |
8428 | ||
abca5fc5 | 8429 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 8430 | { |
36c8b586 | 8431 | struct task_struct *p; |
a4ec24b4 | 8432 | unsigned int time_slice; |
eb580751 | 8433 | struct rq_flags rf; |
dba091b9 | 8434 | struct rq *rq; |
3a5c359a | 8435 | int retval; |
1da177e4 LT |
8436 | |
8437 | if (pid < 0) | |
3a5c359a | 8438 | return -EINVAL; |
1da177e4 LT |
8439 | |
8440 | retval = -ESRCH; | |
1a551ae7 | 8441 | rcu_read_lock(); |
1da177e4 LT |
8442 | p = find_process_by_pid(pid); |
8443 | if (!p) | |
8444 | goto out_unlock; | |
8445 | ||
8446 | retval = security_task_getscheduler(p); | |
8447 | if (retval) | |
8448 | goto out_unlock; | |
8449 | ||
eb580751 | 8450 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
8451 | time_slice = 0; |
8452 | if (p->sched_class->get_rr_interval) | |
8453 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 8454 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 8455 | |
1a551ae7 | 8456 | rcu_read_unlock(); |
abca5fc5 AV |
8457 | jiffies_to_timespec64(time_slice, t); |
8458 | return 0; | |
3a5c359a | 8459 | |
1da177e4 | 8460 | out_unlock: |
1a551ae7 | 8461 | rcu_read_unlock(); |
1da177e4 LT |
8462 | return retval; |
8463 | } | |
8464 | ||
2064a5ab RD |
8465 | /** |
8466 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
8467 | * @pid: pid of the process. | |
8468 | * @interval: userspace pointer to the timeslice value. | |
8469 | * | |
8470 | * this syscall writes the default timeslice value of a given process | |
8471 | * into the user-space timespec buffer. A value of '0' means infinity. | |
8472 | * | |
8473 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
8474 | * an error code. | |
8475 | */ | |
abca5fc5 | 8476 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 8477 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
8478 | { |
8479 | struct timespec64 t; | |
8480 | int retval = sched_rr_get_interval(pid, &t); | |
8481 | ||
8482 | if (retval == 0) | |
8483 | retval = put_timespec64(&t, interval); | |
8484 | ||
8485 | return retval; | |
8486 | } | |
8487 | ||
474b9c77 | 8488 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
8489 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
8490 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
8491 | { |
8492 | struct timespec64 t; | |
8493 | int retval = sched_rr_get_interval(pid, &t); | |
8494 | ||
8495 | if (retval == 0) | |
9afc5eee | 8496 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
8497 | return retval; |
8498 | } | |
8499 | #endif | |
8500 | ||
82a1fcb9 | 8501 | void sched_show_task(struct task_struct *p) |
1da177e4 | 8502 | { |
1da177e4 | 8503 | unsigned long free = 0; |
4e79752c | 8504 | int ppid; |
c930b2c0 | 8505 | |
38200502 TH |
8506 | if (!try_get_task_stack(p)) |
8507 | return; | |
20435d84 | 8508 | |
cc172ff3 | 8509 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 | 8510 | |
b03fbd4f | 8511 | if (task_is_running(p)) |
cc172ff3 | 8512 | pr_cont(" running task "); |
1da177e4 | 8513 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 8514 | free = stack_not_used(p); |
1da177e4 | 8515 | #endif |
a90e984c | 8516 | ppid = 0; |
4e79752c | 8517 | rcu_read_lock(); |
a90e984c ON |
8518 | if (pid_alive(p)) |
8519 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 8520 | rcu_read_unlock(); |
cc172ff3 LZ |
8521 | pr_cont(" stack:%5lu pid:%5d ppid:%6d flags:0x%08lx\n", |
8522 | free, task_pid_nr(p), ppid, | |
aa47b7e0 | 8523 | (unsigned long)task_thread_info(p)->flags); |
1da177e4 | 8524 | |
3d1cb205 | 8525 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 8526 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 8527 | show_stack(p, NULL, KERN_INFO); |
38200502 | 8528 | put_task_stack(p); |
1da177e4 | 8529 | } |
0032f4e8 | 8530 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 8531 | |
5d68cc95 PZ |
8532 | static inline bool |
8533 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
8534 | { | |
2f064a59 PZ |
8535 | unsigned int state = READ_ONCE(p->__state); |
8536 | ||
5d68cc95 PZ |
8537 | /* no filter, everything matches */ |
8538 | if (!state_filter) | |
8539 | return true; | |
8540 | ||
8541 | /* filter, but doesn't match */ | |
2f064a59 | 8542 | if (!(state & state_filter)) |
5d68cc95 PZ |
8543 | return false; |
8544 | ||
8545 | /* | |
8546 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
8547 | * TASK_KILLABLE). | |
8548 | */ | |
2f064a59 | 8549 | if (state_filter == TASK_UNINTERRUPTIBLE && state == TASK_IDLE) |
5d68cc95 PZ |
8550 | return false; |
8551 | ||
8552 | return true; | |
8553 | } | |
8554 | ||
8555 | ||
2f064a59 | 8556 | void show_state_filter(unsigned int state_filter) |
1da177e4 | 8557 | { |
36c8b586 | 8558 | struct task_struct *g, *p; |
1da177e4 | 8559 | |
510f5acc | 8560 | rcu_read_lock(); |
5d07f420 | 8561 | for_each_process_thread(g, p) { |
1da177e4 LT |
8562 | /* |
8563 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 8564 | * console might take a lot of time: |
57675cb9 AR |
8565 | * Also, reset softlockup watchdogs on all CPUs, because |
8566 | * another CPU might be blocked waiting for us to process | |
8567 | * an IPI. | |
1da177e4 LT |
8568 | */ |
8569 | touch_nmi_watchdog(); | |
57675cb9 | 8570 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 8571 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 8572 | sched_show_task(p); |
5d07f420 | 8573 | } |
1da177e4 | 8574 | |
dd41f596 | 8575 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
8576 | if (!state_filter) |
8577 | sysrq_sched_debug_show(); | |
dd41f596 | 8578 | #endif |
510f5acc | 8579 | rcu_read_unlock(); |
e59e2ae2 IM |
8580 | /* |
8581 | * Only show locks if all tasks are dumped: | |
8582 | */ | |
93335a21 | 8583 | if (!state_filter) |
e59e2ae2 | 8584 | debug_show_all_locks(); |
1da177e4 LT |
8585 | } |
8586 | ||
f340c0d1 IM |
8587 | /** |
8588 | * init_idle - set up an idle thread for a given CPU | |
8589 | * @idle: task in question | |
d1ccc66d | 8590 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
8591 | * |
8592 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
8593 | * flag, to make booting more robust. | |
8594 | */ | |
f1a0a376 | 8595 | void __init init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 8596 | { |
70b97a7f | 8597 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
8598 | unsigned long flags; |
8599 | ||
ff51ff84 PZ |
8600 | __sched_fork(0, idle); |
8601 | ||
00b89fe0 VS |
8602 | /* |
8603 | * The idle task doesn't need the kthread struct to function, but it | |
8604 | * is dressed up as a per-CPU kthread and thus needs to play the part | |
8605 | * if we want to avoid special-casing it in code that deals with per-CPU | |
8606 | * kthreads. | |
8607 | */ | |
8608 | set_kthread_struct(idle); | |
8609 | ||
25834c73 | 8610 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5cb9eaa3 | 8611 | raw_spin_rq_lock(rq); |
5cbd54ef | 8612 | |
2f064a59 | 8613 | idle->__state = TASK_RUNNING; |
dd41f596 | 8614 | idle->se.exec_start = sched_clock(); |
00b89fe0 VS |
8615 | /* |
8616 | * PF_KTHREAD should already be set at this point; regardless, make it | |
8617 | * look like a proper per-CPU kthread. | |
8618 | */ | |
8619 | idle->flags |= PF_IDLE | PF_KTHREAD | PF_NO_SETAFFINITY; | |
8620 | kthread_set_per_cpu(idle, cpu); | |
dd41f596 | 8621 | |
d08b9f0c | 8622 | scs_task_reset(idle); |
e1b77c92 MR |
8623 | kasan_unpoison_task_stack(idle); |
8624 | ||
de9b8f5d PZ |
8625 | #ifdef CONFIG_SMP |
8626 | /* | |
b19a888c | 8627 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
8628 | * in that case do_set_cpus_allowed() will not do the right thing. |
8629 | * | |
8630 | * And since this is boot we can forgo the serialization. | |
8631 | */ | |
9cfc3e18 | 8632 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 8633 | #endif |
6506cf6c PZ |
8634 | /* |
8635 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 8636 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
8637 | * lockdep check in task_group() will fail. |
8638 | * | |
8639 | * Similar case to sched_fork(). / Alternatively we could | |
8640 | * use task_rq_lock() here and obtain the other rq->lock. | |
8641 | * | |
8642 | * Silence PROVE_RCU | |
8643 | */ | |
8644 | rcu_read_lock(); | |
dd41f596 | 8645 | __set_task_cpu(idle, cpu); |
6506cf6c | 8646 | rcu_read_unlock(); |
1da177e4 | 8647 | |
5311a98f EB |
8648 | rq->idle = idle; |
8649 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 8650 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 8651 | #ifdef CONFIG_SMP |
3ca7a440 | 8652 | idle->on_cpu = 1; |
4866cde0 | 8653 | #endif |
5cb9eaa3 | 8654 | raw_spin_rq_unlock(rq); |
25834c73 | 8655 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); |
1da177e4 LT |
8656 | |
8657 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 8658 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 8659 | |
dd41f596 IM |
8660 | /* |
8661 | * The idle tasks have their own, simple scheduling class: | |
8662 | */ | |
8663 | idle->sched_class = &idle_sched_class; | |
868baf07 | 8664 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 8665 | vtime_init_idle(idle, cpu); |
de9b8f5d | 8666 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
8667 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
8668 | #endif | |
19978ca6 IM |
8669 | } |
8670 | ||
e1d4eeec NP |
8671 | #ifdef CONFIG_SMP |
8672 | ||
f82f8042 JL |
8673 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
8674 | const struct cpumask *trial) | |
8675 | { | |
06a76fe0 | 8676 | int ret = 1; |
f82f8042 | 8677 | |
bb2bc55a MG |
8678 | if (!cpumask_weight(cur)) |
8679 | return ret; | |
8680 | ||
06a76fe0 | 8681 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
8682 | |
8683 | return ret; | |
8684 | } | |
8685 | ||
7f51412a JL |
8686 | int task_can_attach(struct task_struct *p, |
8687 | const struct cpumask *cs_cpus_allowed) | |
8688 | { | |
8689 | int ret = 0; | |
8690 | ||
8691 | /* | |
8692 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 8693 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
8694 | * affinity and isolating such threads by their set of |
8695 | * allowed nodes is unnecessary. Thus, cpusets are not | |
8696 | * applicable for such threads. This prevents checking for | |
8697 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 8698 | * before cpus_mask may be changed. |
7f51412a JL |
8699 | */ |
8700 | if (p->flags & PF_NO_SETAFFINITY) { | |
8701 | ret = -EINVAL; | |
8702 | goto out; | |
8703 | } | |
8704 | ||
7f51412a | 8705 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
06a76fe0 NP |
8706 | cs_cpus_allowed)) |
8707 | ret = dl_task_can_attach(p, cs_cpus_allowed); | |
7f51412a | 8708 | |
7f51412a JL |
8709 | out: |
8710 | return ret; | |
8711 | } | |
8712 | ||
f2cb1360 | 8713 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 8714 | |
e6628d5b MG |
8715 | #ifdef CONFIG_NUMA_BALANCING |
8716 | /* Migrate current task p to target_cpu */ | |
8717 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
8718 | { | |
8719 | struct migration_arg arg = { p, target_cpu }; | |
8720 | int curr_cpu = task_cpu(p); | |
8721 | ||
8722 | if (curr_cpu == target_cpu) | |
8723 | return 0; | |
8724 | ||
3bd37062 | 8725 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
8726 | return -EINVAL; |
8727 | ||
8728 | /* TODO: This is not properly updating schedstats */ | |
8729 | ||
286549dc | 8730 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
8731 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
8732 | } | |
0ec8aa00 PZ |
8733 | |
8734 | /* | |
8735 | * Requeue a task on a given node and accurately track the number of NUMA | |
8736 | * tasks on the runqueues | |
8737 | */ | |
8738 | void sched_setnuma(struct task_struct *p, int nid) | |
8739 | { | |
da0c1e65 | 8740 | bool queued, running; |
eb580751 PZ |
8741 | struct rq_flags rf; |
8742 | struct rq *rq; | |
0ec8aa00 | 8743 | |
eb580751 | 8744 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 8745 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
8746 | running = task_current(rq, p); |
8747 | ||
da0c1e65 | 8748 | if (queued) |
1de64443 | 8749 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 8750 | if (running) |
f3cd1c4e | 8751 | put_prev_task(rq, p); |
0ec8aa00 PZ |
8752 | |
8753 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 8754 | |
da0c1e65 | 8755 | if (queued) |
7134b3e9 | 8756 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 8757 | if (running) |
03b7fad1 | 8758 | set_next_task(rq, p); |
eb580751 | 8759 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 8760 | } |
5cc389bc | 8761 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 8762 | |
1da177e4 | 8763 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 8764 | /* |
d1ccc66d | 8765 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 8766 | * offline. |
054b9108 | 8767 | */ |
48c5ccae | 8768 | void idle_task_exit(void) |
1da177e4 | 8769 | { |
48c5ccae | 8770 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 8771 | |
48c5ccae | 8772 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 8773 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 8774 | |
a53efe5f | 8775 | if (mm != &init_mm) { |
252d2a41 | 8776 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
8777 | finish_arch_post_lock_switch(); |
8778 | } | |
bf2c59fc | 8779 | |
63acd42c | 8780 | scs_task_reset(current); |
bf2c59fc | 8781 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ |
1da177e4 LT |
8782 | } |
8783 | ||
2558aacf | 8784 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 8785 | { |
2558aacf PZ |
8786 | struct task_struct *p = arg; |
8787 | struct rq *rq = this_rq(); | |
8788 | struct rq_flags rf; | |
8789 | int cpu; | |
1da177e4 | 8790 | |
2558aacf PZ |
8791 | raw_spin_lock_irq(&p->pi_lock); |
8792 | rq_lock(rq, &rf); | |
3f1d2a31 | 8793 | |
2558aacf PZ |
8794 | update_rq_clock(rq); |
8795 | ||
8796 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
8797 | cpu = select_fallback_rq(rq->cpu, p); | |
8798 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 8799 | } |
3f1d2a31 | 8800 | |
2558aacf PZ |
8801 | rq_unlock(rq, &rf); |
8802 | raw_spin_unlock_irq(&p->pi_lock); | |
8803 | ||
8804 | put_task_struct(p); | |
8805 | ||
8806 | return 0; | |
10e7071b | 8807 | } |
3f1d2a31 | 8808 | |
2558aacf PZ |
8809 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
8810 | ||
48f24c4d | 8811 | /* |
2558aacf | 8812 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
b5c44773 PZ |
8813 | * |
8814 | * This is enabled below SCHED_AP_ACTIVE; when !cpu_active(), but only | |
8815 | * effective when the hotplug motion is down. | |
1da177e4 | 8816 | */ |
2558aacf | 8817 | static void balance_push(struct rq *rq) |
1da177e4 | 8818 | { |
2558aacf PZ |
8819 | struct task_struct *push_task = rq->curr; |
8820 | ||
5cb9eaa3 | 8821 | lockdep_assert_rq_held(rq); |
b5c44773 | 8822 | |
ae792702 PZ |
8823 | /* |
8824 | * Ensure the thing is persistent until balance_push_set(.on = false); | |
8825 | */ | |
8826 | rq->balance_callback = &balance_push_callback; | |
1da177e4 | 8827 | |
b5c44773 | 8828 | /* |
868ad33b TG |
8829 | * Only active while going offline and when invoked on the outgoing |
8830 | * CPU. | |
b5c44773 | 8831 | */ |
868ad33b | 8832 | if (!cpu_dying(rq->cpu) || rq != this_rq()) |
b5c44773 PZ |
8833 | return; |
8834 | ||
1da177e4 | 8835 | /* |
2558aacf PZ |
8836 | * Both the cpu-hotplug and stop task are in this case and are |
8837 | * required to complete the hotplug process. | |
1da177e4 | 8838 | */ |
00b89fe0 | 8839 | if (kthread_is_per_cpu(push_task) || |
5ba2ffba PZ |
8840 | is_migration_disabled(push_task)) { |
8841 | ||
f2469a1f TG |
8842 | /* |
8843 | * If this is the idle task on the outgoing CPU try to wake | |
8844 | * up the hotplug control thread which might wait for the | |
8845 | * last task to vanish. The rcuwait_active() check is | |
8846 | * accurate here because the waiter is pinned on this CPU | |
8847 | * and can't obviously be running in parallel. | |
3015ef4b TG |
8848 | * |
8849 | * On RT kernels this also has to check whether there are | |
8850 | * pinned and scheduled out tasks on the runqueue. They | |
8851 | * need to leave the migrate disabled section first. | |
f2469a1f | 8852 | */ |
3015ef4b TG |
8853 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
8854 | rcuwait_active(&rq->hotplug_wait)) { | |
5cb9eaa3 | 8855 | raw_spin_rq_unlock(rq); |
f2469a1f | 8856 | rcuwait_wake_up(&rq->hotplug_wait); |
5cb9eaa3 | 8857 | raw_spin_rq_lock(rq); |
f2469a1f | 8858 | } |
2558aacf | 8859 | return; |
f2469a1f | 8860 | } |
48f24c4d | 8861 | |
2558aacf | 8862 | get_task_struct(push_task); |
77bd3970 | 8863 | /* |
2558aacf PZ |
8864 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
8865 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 8866 | */ |
5cb9eaa3 | 8867 | raw_spin_rq_unlock(rq); |
2558aacf PZ |
8868 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, |
8869 | this_cpu_ptr(&push_work)); | |
8870 | /* | |
8871 | * At this point need_resched() is true and we'll take the loop in | |
8872 | * schedule(). The next pick is obviously going to be the stop task | |
5ba2ffba | 8873 | * which kthread_is_per_cpu() and will push this task away. |
2558aacf | 8874 | */ |
5cb9eaa3 | 8875 | raw_spin_rq_lock(rq); |
2558aacf | 8876 | } |
77bd3970 | 8877 | |
2558aacf PZ |
8878 | static void balance_push_set(int cpu, bool on) |
8879 | { | |
8880 | struct rq *rq = cpu_rq(cpu); | |
8881 | struct rq_flags rf; | |
48c5ccae | 8882 | |
2558aacf | 8883 | rq_lock_irqsave(rq, &rf); |
22f667c9 PZ |
8884 | if (on) { |
8885 | WARN_ON_ONCE(rq->balance_callback); | |
ae792702 | 8886 | rq->balance_callback = &balance_push_callback; |
22f667c9 | 8887 | } else if (rq->balance_callback == &balance_push_callback) { |
ae792702 | 8888 | rq->balance_callback = NULL; |
22f667c9 | 8889 | } |
2558aacf PZ |
8890 | rq_unlock_irqrestore(rq, &rf); |
8891 | } | |
e692ab53 | 8892 | |
f2469a1f TG |
8893 | /* |
8894 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
8895 | * inactive. All tasks which are not per CPU kernel threads are either | |
8896 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
8897 | * during wakeup. Wait until the CPU is quiescent. | |
8898 | */ | |
8899 | static void balance_hotplug_wait(void) | |
8900 | { | |
8901 | struct rq *rq = this_rq(); | |
5473e0cc | 8902 | |
3015ef4b TG |
8903 | rcuwait_wait_event(&rq->hotplug_wait, |
8904 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
8905 | TASK_UNINTERRUPTIBLE); |
8906 | } | |
5473e0cc | 8907 | |
2558aacf | 8908 | #else |
dce48a84 | 8909 | |
2558aacf PZ |
8910 | static inline void balance_push(struct rq *rq) |
8911 | { | |
dce48a84 | 8912 | } |
dce48a84 | 8913 | |
2558aacf PZ |
8914 | static inline void balance_push_set(int cpu, bool on) |
8915 | { | |
8916 | } | |
8917 | ||
f2469a1f TG |
8918 | static inline void balance_hotplug_wait(void) |
8919 | { | |
dce48a84 | 8920 | } |
f2469a1f | 8921 | |
1da177e4 LT |
8922 | #endif /* CONFIG_HOTPLUG_CPU */ |
8923 | ||
f2cb1360 | 8924 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
8925 | { |
8926 | if (!rq->online) { | |
8927 | const struct sched_class *class; | |
8928 | ||
c6c4927b | 8929 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
8930 | rq->online = 1; |
8931 | ||
8932 | for_each_class(class) { | |
8933 | if (class->rq_online) | |
8934 | class->rq_online(rq); | |
8935 | } | |
8936 | } | |
8937 | } | |
8938 | ||
f2cb1360 | 8939 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
8940 | { |
8941 | if (rq->online) { | |
8942 | const struct sched_class *class; | |
8943 | ||
8944 | for_each_class(class) { | |
8945 | if (class->rq_offline) | |
8946 | class->rq_offline(rq); | |
8947 | } | |
8948 | ||
c6c4927b | 8949 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
8950 | rq->online = 0; |
8951 | } | |
8952 | } | |
8953 | ||
d1ccc66d IM |
8954 | /* |
8955 | * used to mark begin/end of suspend/resume: | |
8956 | */ | |
8957 | static int num_cpus_frozen; | |
d35be8ba | 8958 | |
1da177e4 | 8959 | /* |
3a101d05 TH |
8960 | * Update cpusets according to cpu_active mask. If cpusets are |
8961 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
8962 | * around partition_sched_domains(). | |
d35be8ba SB |
8963 | * |
8964 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
8965 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 8966 | */ |
40190a78 | 8967 | static void cpuset_cpu_active(void) |
e761b772 | 8968 | { |
40190a78 | 8969 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
8970 | /* |
8971 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
8972 | * resume sequence. As long as this is not the last online | |
8973 | * operation in the resume sequence, just build a single sched | |
8974 | * domain, ignoring cpusets. | |
8975 | */ | |
50e76632 PZ |
8976 | partition_sched_domains(1, NULL, NULL); |
8977 | if (--num_cpus_frozen) | |
135fb3e1 | 8978 | return; |
d35be8ba SB |
8979 | /* |
8980 | * This is the last CPU online operation. So fall through and | |
8981 | * restore the original sched domains by considering the | |
8982 | * cpuset configurations. | |
8983 | */ | |
50e76632 | 8984 | cpuset_force_rebuild(); |
3a101d05 | 8985 | } |
30e03acd | 8986 | cpuset_update_active_cpus(); |
3a101d05 | 8987 | } |
e761b772 | 8988 | |
40190a78 | 8989 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 8990 | { |
40190a78 | 8991 | if (!cpuhp_tasks_frozen) { |
06a76fe0 | 8992 | if (dl_cpu_busy(cpu)) |
135fb3e1 | 8993 | return -EBUSY; |
30e03acd | 8994 | cpuset_update_active_cpus(); |
135fb3e1 | 8995 | } else { |
d35be8ba SB |
8996 | num_cpus_frozen++; |
8997 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 8998 | } |
135fb3e1 | 8999 | return 0; |
e761b772 | 9000 | } |
e761b772 | 9001 | |
40190a78 | 9002 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 9003 | { |
7d976699 | 9004 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 9005 | struct rq_flags rf; |
7d976699 | 9006 | |
22f667c9 | 9007 | /* |
b5c44773 PZ |
9008 | * Clear the balance_push callback and prepare to schedule |
9009 | * regular tasks. | |
22f667c9 | 9010 | */ |
2558aacf PZ |
9011 | balance_push_set(cpu, false); |
9012 | ||
ba2591a5 PZ |
9013 | #ifdef CONFIG_SCHED_SMT |
9014 | /* | |
c5511d03 | 9015 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 9016 | */ |
c5511d03 PZI |
9017 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
9018 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 9019 | #endif |
40190a78 | 9020 | set_cpu_active(cpu, true); |
135fb3e1 | 9021 | |
40190a78 | 9022 | if (sched_smp_initialized) { |
135fb3e1 | 9023 | sched_domains_numa_masks_set(cpu); |
40190a78 | 9024 | cpuset_cpu_active(); |
e761b772 | 9025 | } |
7d976699 TG |
9026 | |
9027 | /* | |
9028 | * Put the rq online, if not already. This happens: | |
9029 | * | |
9030 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 9031 | * after all CPUs have been brought up. |
7d976699 TG |
9032 | * |
9033 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
9034 | * domains. | |
9035 | */ | |
8a8c69c3 | 9036 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
9037 | if (rq->rd) { |
9038 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9039 | set_rq_online(rq); | |
9040 | } | |
8a8c69c3 | 9041 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 9042 | |
40190a78 | 9043 | return 0; |
135fb3e1 TG |
9044 | } |
9045 | ||
40190a78 | 9046 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 9047 | { |
120455c5 PZ |
9048 | struct rq *rq = cpu_rq(cpu); |
9049 | struct rq_flags rf; | |
135fb3e1 TG |
9050 | int ret; |
9051 | ||
e0b257c3 AMB |
9052 | /* |
9053 | * Remove CPU from nohz.idle_cpus_mask to prevent participating in | |
9054 | * load balancing when not active | |
9055 | */ | |
9056 | nohz_balance_exit_idle(rq); | |
9057 | ||
40190a78 | 9058 | set_cpu_active(cpu, false); |
741ba80f PZ |
9059 | |
9060 | /* | |
9061 | * From this point forward, this CPU will refuse to run any task that | |
9062 | * is not: migrate_disable() or KTHREAD_IS_PER_CPU, and will actively | |
9063 | * push those tasks away until this gets cleared, see | |
9064 | * sched_cpu_dying(). | |
9065 | */ | |
975707f2 PZ |
9066 | balance_push_set(cpu, true); |
9067 | ||
b2454caa | 9068 | /* |
975707f2 PZ |
9069 | * We've cleared cpu_active_mask / set balance_push, wait for all |
9070 | * preempt-disabled and RCU users of this state to go away such that | |
9071 | * all new such users will observe it. | |
b2454caa | 9072 | * |
5ba2ffba PZ |
9073 | * Specifically, we rely on ttwu to no longer target this CPU, see |
9074 | * ttwu_queue_cond() and is_cpu_allowed(). | |
9075 | * | |
b2454caa PZ |
9076 | * Do sync before park smpboot threads to take care the rcu boost case. |
9077 | */ | |
309ba859 | 9078 | synchronize_rcu(); |
40190a78 | 9079 | |
120455c5 PZ |
9080 | rq_lock_irqsave(rq, &rf); |
9081 | if (rq->rd) { | |
9082 | update_rq_clock(rq); | |
9083 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9084 | set_rq_offline(rq); | |
9085 | } | |
9086 | rq_unlock_irqrestore(rq, &rf); | |
9087 | ||
c5511d03 PZI |
9088 | #ifdef CONFIG_SCHED_SMT |
9089 | /* | |
9090 | * When going down, decrement the number of cores with SMT present. | |
9091 | */ | |
9092 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
9093 | static_branch_dec_cpuslocked(&sched_smt_present); | |
3c474b32 PZ |
9094 | |
9095 | sched_core_cpu_deactivate(cpu); | |
c5511d03 PZI |
9096 | #endif |
9097 | ||
40190a78 TG |
9098 | if (!sched_smp_initialized) |
9099 | return 0; | |
9100 | ||
9101 | ret = cpuset_cpu_inactive(cpu); | |
9102 | if (ret) { | |
2558aacf | 9103 | balance_push_set(cpu, false); |
40190a78 TG |
9104 | set_cpu_active(cpu, true); |
9105 | return ret; | |
135fb3e1 | 9106 | } |
40190a78 TG |
9107 | sched_domains_numa_masks_clear(cpu); |
9108 | return 0; | |
135fb3e1 TG |
9109 | } |
9110 | ||
94baf7a5 TG |
9111 | static void sched_rq_cpu_starting(unsigned int cpu) |
9112 | { | |
9113 | struct rq *rq = cpu_rq(cpu); | |
9114 | ||
9115 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
9116 | update_max_interval(); |
9117 | } | |
9118 | ||
135fb3e1 TG |
9119 | int sched_cpu_starting(unsigned int cpu) |
9120 | { | |
9edeaea1 | 9121 | sched_core_cpu_starting(cpu); |
94baf7a5 | 9122 | sched_rq_cpu_starting(cpu); |
d84b3131 | 9123 | sched_tick_start(cpu); |
135fb3e1 | 9124 | return 0; |
e761b772 | 9125 | } |
e761b772 | 9126 | |
f2785ddb | 9127 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
9128 | |
9129 | /* | |
9130 | * Invoked immediately before the stopper thread is invoked to bring the | |
9131 | * CPU down completely. At this point all per CPU kthreads except the | |
9132 | * hotplug thread (current) and the stopper thread (inactive) have been | |
9133 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
9134 | * any of those which might be on the way out are gone. | |
9135 | * | |
9136 | * If after this point a bound task is being woken on this CPU then the | |
9137 | * responsible hotplug callback has failed to do it's job. | |
9138 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
9139 | */ | |
9140 | int sched_cpu_wait_empty(unsigned int cpu) | |
9141 | { | |
9142 | balance_hotplug_wait(); | |
9143 | return 0; | |
9144 | } | |
9145 | ||
9146 | /* | |
9147 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
9148 | * might have. Called from the CPU stopper task after ensuring that the | |
9149 | * stopper is the last running task on the CPU, so nr_active count is | |
9150 | * stable. We need to take the teardown thread which is calling this into | |
9151 | * account, so we hand in adjust = 1 to the load calculation. | |
9152 | * | |
9153 | * Also see the comment "Global load-average calculations". | |
9154 | */ | |
9155 | static void calc_load_migrate(struct rq *rq) | |
9156 | { | |
9157 | long delta = calc_load_fold_active(rq, 1); | |
9158 | ||
9159 | if (delta) | |
9160 | atomic_long_add(delta, &calc_load_tasks); | |
9161 | } | |
9162 | ||
36c6e17b VS |
9163 | static void dump_rq_tasks(struct rq *rq, const char *loglvl) |
9164 | { | |
9165 | struct task_struct *g, *p; | |
9166 | int cpu = cpu_of(rq); | |
9167 | ||
5cb9eaa3 | 9168 | lockdep_assert_rq_held(rq); |
36c6e17b VS |
9169 | |
9170 | printk("%sCPU%d enqueued tasks (%u total):\n", loglvl, cpu, rq->nr_running); | |
9171 | for_each_process_thread(g, p) { | |
9172 | if (task_cpu(p) != cpu) | |
9173 | continue; | |
9174 | ||
9175 | if (!task_on_rq_queued(p)) | |
9176 | continue; | |
9177 | ||
9178 | printk("%s\tpid: %d, name: %s\n", loglvl, p->pid, p->comm); | |
9179 | } | |
9180 | } | |
9181 | ||
f2785ddb TG |
9182 | int sched_cpu_dying(unsigned int cpu) |
9183 | { | |
9184 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 9185 | struct rq_flags rf; |
f2785ddb TG |
9186 | |
9187 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 9188 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
9189 | |
9190 | rq_lock_irqsave(rq, &rf); | |
36c6e17b VS |
9191 | if (rq->nr_running != 1 || rq_has_pinned_tasks(rq)) { |
9192 | WARN(true, "Dying CPU not properly vacated!"); | |
9193 | dump_rq_tasks(rq, KERN_WARNING); | |
9194 | } | |
8a8c69c3 PZ |
9195 | rq_unlock_irqrestore(rq, &rf); |
9196 | ||
f2785ddb TG |
9197 | calc_load_migrate(rq); |
9198 | update_max_interval(); | |
e5ef27d0 | 9199 | hrtick_clear(rq); |
3c474b32 | 9200 | sched_core_cpu_dying(cpu); |
f2785ddb TG |
9201 | return 0; |
9202 | } | |
9203 | #endif | |
9204 | ||
1da177e4 LT |
9205 | void __init sched_init_smp(void) |
9206 | { | |
cb83b629 PZ |
9207 | sched_init_numa(); |
9208 | ||
6acce3ef PZ |
9209 | /* |
9210 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 9211 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 9212 | * happen. |
6acce3ef | 9213 | */ |
712555ee | 9214 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 9215 | sched_init_domains(cpu_active_mask); |
712555ee | 9216 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 9217 | |
5c1e1767 | 9218 | /* Move init over to a non-isolated CPU */ |
edb93821 | 9219 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0) |
5c1e1767 | 9220 | BUG(); |
15faafc6 | 9221 | current->flags &= ~PF_NO_SETAFFINITY; |
19978ca6 | 9222 | sched_init_granularity(); |
4212823f | 9223 | |
0e3900e6 | 9224 | init_sched_rt_class(); |
1baca4ce | 9225 | init_sched_dl_class(); |
1b568f0a | 9226 | |
e26fbffd | 9227 | sched_smp_initialized = true; |
1da177e4 | 9228 | } |
e26fbffd TG |
9229 | |
9230 | static int __init migration_init(void) | |
9231 | { | |
77a5352b | 9232 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 9233 | return 0; |
1da177e4 | 9234 | } |
e26fbffd TG |
9235 | early_initcall(migration_init); |
9236 | ||
1da177e4 LT |
9237 | #else |
9238 | void __init sched_init_smp(void) | |
9239 | { | |
19978ca6 | 9240 | sched_init_granularity(); |
1da177e4 LT |
9241 | } |
9242 | #endif /* CONFIG_SMP */ | |
9243 | ||
9244 | int in_sched_functions(unsigned long addr) | |
9245 | { | |
1da177e4 LT |
9246 | return in_lock_functions(addr) || |
9247 | (addr >= (unsigned long)__sched_text_start | |
9248 | && addr < (unsigned long)__sched_text_end); | |
9249 | } | |
9250 | ||
029632fb | 9251 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
9252 | /* |
9253 | * Default task group. | |
9254 | * Every task in system belongs to this group at bootup. | |
9255 | */ | |
029632fb | 9256 | struct task_group root_task_group; |
35cf4e50 | 9257 | LIST_HEAD(task_groups); |
b0367629 WL |
9258 | |
9259 | /* Cacheline aligned slab cache for task_group */ | |
9260 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 9261 | #endif |
6f505b16 | 9262 | |
e6252c3e | 9263 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
10e2f1ac | 9264 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
6f505b16 | 9265 | |
1da177e4 LT |
9266 | void __init sched_init(void) |
9267 | { | |
a1dc0446 | 9268 | unsigned long ptr = 0; |
55627e3c | 9269 | int i; |
434d53b0 | 9270 | |
c3a340f7 SRV |
9271 | /* Make sure the linker didn't screw up */ |
9272 | BUG_ON(&idle_sched_class + 1 != &fair_sched_class || | |
9273 | &fair_sched_class + 1 != &rt_sched_class || | |
9274 | &rt_sched_class + 1 != &dl_sched_class); | |
9275 | #ifdef CONFIG_SMP | |
9276 | BUG_ON(&dl_sched_class + 1 != &stop_sched_class); | |
9277 | #endif | |
9278 | ||
5822a454 | 9279 | wait_bit_init(); |
9dcb8b68 | 9280 | |
434d53b0 | 9281 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 9282 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
9283 | #endif |
9284 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 9285 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 9286 | #endif |
a1dc0446 QC |
9287 | if (ptr) { |
9288 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
9289 | |
9290 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 9291 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
9292 | ptr += nr_cpu_ids * sizeof(void **); |
9293 | ||
07e06b01 | 9294 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 9295 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 9296 | |
b1d1779e WY |
9297 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
9298 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 9299 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 9300 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9301 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
9302 | ptr += nr_cpu_ids * sizeof(void **); |
9303 | ||
07e06b01 | 9304 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
9305 | ptr += nr_cpu_ids * sizeof(void **); |
9306 | ||
6d6bc0ad | 9307 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 9308 | } |
df7c8e84 | 9309 | #ifdef CONFIG_CPUMASK_OFFSTACK |
b74e6278 AT |
9310 | for_each_possible_cpu(i) { |
9311 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( | |
9312 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
10e2f1ac PZ |
9313 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
9314 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); | |
434d53b0 | 9315 | } |
b74e6278 | 9316 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
dd41f596 | 9317 | |
d1ccc66d IM |
9318 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
9319 | init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime()); | |
332ac17e | 9320 | |
57d885fe GH |
9321 | #ifdef CONFIG_SMP |
9322 | init_defrootdomain(); | |
9323 | #endif | |
9324 | ||
d0b27fa7 | 9325 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9326 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 9327 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 9328 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 9329 | |
7c941438 | 9330 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
9331 | task_group_cache = KMEM_CACHE(task_group, 0); |
9332 | ||
07e06b01 YZ |
9333 | list_add(&root_task_group.list, &task_groups); |
9334 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 9335 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 9336 | autogroup_init(&init_task); |
7c941438 | 9337 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 9338 | |
0a945022 | 9339 | for_each_possible_cpu(i) { |
70b97a7f | 9340 | struct rq *rq; |
1da177e4 LT |
9341 | |
9342 | rq = cpu_rq(i); | |
5cb9eaa3 | 9343 | raw_spin_lock_init(&rq->__lock); |
7897986b | 9344 | rq->nr_running = 0; |
dce48a84 TG |
9345 | rq->calc_load_active = 0; |
9346 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 9347 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
9348 | init_rt_rq(&rq->rt); |
9349 | init_dl_rq(&rq->dl); | |
dd41f596 | 9350 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 9351 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 9352 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 9353 | /* |
d1ccc66d | 9354 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
9355 | * |
9356 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
9357 | * gets 100% of the CPU resources in the system. This overall |
9358 | * system CPU resource is divided among the tasks of | |
07e06b01 | 9359 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
9360 | * based on each entity's (task or task-group's) weight |
9361 | * (se->load.weight). | |
9362 | * | |
07e06b01 | 9363 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 9364 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 9365 | * then A0's share of the CPU resource is: |
354d60c2 | 9366 | * |
0d905bca | 9367 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 9368 | * |
07e06b01 YZ |
9369 | * We achieve this by letting root_task_group's tasks sit |
9370 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 9371 | */ |
07e06b01 | 9372 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
9373 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9374 | ||
9375 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9376 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9377 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 9378 | #endif |
1da177e4 | 9379 | #ifdef CONFIG_SMP |
41c7ce9a | 9380 | rq->sd = NULL; |
57d885fe | 9381 | rq->rd = NULL; |
ca6d75e6 | 9382 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
b5c44773 | 9383 | rq->balance_callback = &balance_push_callback; |
1da177e4 | 9384 | rq->active_balance = 0; |
dd41f596 | 9385 | rq->next_balance = jiffies; |
1da177e4 | 9386 | rq->push_cpu = 0; |
0a2966b4 | 9387 | rq->cpu = i; |
1f11eb6a | 9388 | rq->online = 0; |
eae0c9df MG |
9389 | rq->idle_stamp = 0; |
9390 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
94aafc3e PZ |
9391 | rq->wake_stamp = jiffies; |
9392 | rq->wake_avg_idle = rq->avg_idle; | |
9bd721c5 | 9393 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
9394 | |
9395 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
9396 | ||
dc938520 | 9397 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 9398 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 9399 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 9400 | atomic_set(&rq->nohz_flags, 0); |
90b5363a | 9401 | |
545b8c8d | 9402 | INIT_CSD(&rq->nohz_csd, nohz_csd_func, rq); |
83cd4fe2 | 9403 | #endif |
f2469a1f TG |
9404 | #ifdef CONFIG_HOTPLUG_CPU |
9405 | rcuwait_init(&rq->hotplug_wait); | |
83cd4fe2 | 9406 | #endif |
9fd81dd5 | 9407 | #endif /* CONFIG_SMP */ |
77a021be | 9408 | hrtick_rq_init(rq); |
1da177e4 | 9409 | atomic_set(&rq->nr_iowait, 0); |
9edeaea1 PZ |
9410 | |
9411 | #ifdef CONFIG_SCHED_CORE | |
3c474b32 | 9412 | rq->core = rq; |
539f6512 | 9413 | rq->core_pick = NULL; |
9edeaea1 | 9414 | rq->core_enabled = 0; |
539f6512 PZ |
9415 | rq->core_tree = RB_ROOT; |
9416 | rq->core_forceidle = false; | |
9417 | ||
9418 | rq->core_cookie = 0UL; | |
9edeaea1 | 9419 | #endif |
1da177e4 LT |
9420 | } |
9421 | ||
9059393e | 9422 | set_load_weight(&init_task, false); |
b50f60ce | 9423 | |
1da177e4 LT |
9424 | /* |
9425 | * The boot idle thread does lazy MMU switching as well: | |
9426 | */ | |
f1f10076 | 9427 | mmgrab(&init_mm); |
1da177e4 LT |
9428 | enter_lazy_tlb(&init_mm, current); |
9429 | ||
9430 | /* | |
9431 | * Make us the idle thread. Technically, schedule() should not be | |
9432 | * called from this thread, however somewhere below it might be, | |
9433 | * but because we are the idle thread, we just pick up running again | |
9434 | * when this runqueue becomes "idle". | |
9435 | */ | |
9436 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9437 | |
9438 | calc_load_update = jiffies + LOAD_FREQ; | |
9439 | ||
bf4d83f6 | 9440 | #ifdef CONFIG_SMP |
29d5e047 | 9441 | idle_thread_set_boot_cpu(); |
b5c44773 | 9442 | balance_push_set(smp_processor_id(), false); |
029632fb PZ |
9443 | #endif |
9444 | init_sched_fair_class(); | |
6a7b3dc3 | 9445 | |
eb414681 JW |
9446 | psi_init(); |
9447 | ||
69842cba PB |
9448 | init_uclamp(); |
9449 | ||
c597bfdd FW |
9450 | preempt_dynamic_init(); |
9451 | ||
6892b75e | 9452 | scheduler_running = 1; |
1da177e4 LT |
9453 | } |
9454 | ||
d902db1e | 9455 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 | 9456 | |
42a38756 | 9457 | void __might_sleep(const char *file, int line) |
1da177e4 | 9458 | { |
d6c23bb3 | 9459 | unsigned int state = get_current_state(); |
8eb23b9f PZ |
9460 | /* |
9461 | * Blocking primitives will set (and therefore destroy) current->state, | |
9462 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
9463 | * otherwise we will destroy state. | |
9464 | */ | |
d6c23bb3 | 9465 | WARN_ONCE(state != TASK_RUNNING && current->task_state_change, |
8eb23b9f | 9466 | "do not call blocking ops when !TASK_RUNNING; " |
d6c23bb3 | 9467 | "state=%x set at [<%p>] %pS\n", state, |
8eb23b9f | 9468 | (void *)current->task_state_change, |
00845eb9 | 9469 | (void *)current->task_state_change); |
8eb23b9f | 9470 | |
42a38756 | 9471 | __might_resched(file, line, 0); |
3427445a PZ |
9472 | } |
9473 | EXPORT_SYMBOL(__might_sleep); | |
9474 | ||
8d713b69 TG |
9475 | static void print_preempt_disable_ip(int preempt_offset, unsigned long ip) |
9476 | { | |
9477 | if (!IS_ENABLED(CONFIG_DEBUG_PREEMPT)) | |
9478 | return; | |
9479 | ||
9480 | if (preempt_count() == preempt_offset) | |
9481 | return; | |
9482 | ||
9483 | pr_err("Preemption disabled at:"); | |
9484 | print_ip_sym(KERN_ERR, ip); | |
9485 | } | |
9486 | ||
50e081b9 TG |
9487 | static inline bool resched_offsets_ok(unsigned int offsets) |
9488 | { | |
9489 | unsigned int nested = preempt_count(); | |
9490 | ||
9491 | nested += rcu_preempt_depth() << MIGHT_RESCHED_RCU_SHIFT; | |
9492 | ||
9493 | return nested == offsets; | |
9494 | } | |
9495 | ||
9496 | void __might_resched(const char *file, int line, unsigned int offsets) | |
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 | ||
50e081b9 | 9506 | if ((resched_offsets_ok(offsets) && !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 | ||
a45ed302 TG |
9519 | pr_err("BUG: sleeping function called from invalid context at %s:%d\n", |
9520 | file, line); | |
9521 | pr_err("in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", | |
9522 | in_atomic(), irqs_disabled(), current->non_block_count, | |
9523 | current->pid, current->comm); | |
8d713b69 | 9524 | pr_err("preempt_count: %x, expected: %x\n", preempt_count(), |
50e081b9 | 9525 | offsets & MIGHT_RESCHED_PREEMPT_MASK); |
8d713b69 TG |
9526 | |
9527 | if (IS_ENABLED(CONFIG_PREEMPT_RCU)) { | |
50e081b9 TG |
9528 | pr_err("RCU nest depth: %d, expected: %u\n", |
9529 | rcu_preempt_depth(), offsets >> MIGHT_RESCHED_RCU_SHIFT); | |
8d713b69 | 9530 | } |
aef745fc | 9531 | |
a8b686b3 | 9532 | if (task_stack_end_corrupted(current)) |
a45ed302 | 9533 | pr_emerg("Thread overran stack, or stack corrupted\n"); |
a8b686b3 | 9534 | |
aef745fc IM |
9535 | debug_show_held_locks(current); |
9536 | if (irqs_disabled()) | |
9537 | print_irqtrace_events(current); | |
8d713b69 | 9538 | |
50e081b9 TG |
9539 | print_preempt_disable_ip(offsets & MIGHT_RESCHED_PREEMPT_MASK, |
9540 | preempt_disable_ip); | |
8d713b69 | 9541 | |
aef745fc | 9542 | dump_stack(); |
f0b22e39 | 9543 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 9544 | } |
874f670e | 9545 | EXPORT_SYMBOL(__might_resched); |
568f1967 PZ |
9546 | |
9547 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
9548 | { | |
9549 | static unsigned long prev_jiffy; | |
9550 | ||
9551 | if (irqs_disabled()) | |
9552 | return; | |
9553 | ||
9554 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
9555 | return; | |
9556 | ||
9557 | if (preempt_count() > preempt_offset) | |
9558 | return; | |
9559 | ||
9560 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9561 | return; | |
9562 | prev_jiffy = jiffies; | |
9563 | ||
9564 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
9565 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9566 | in_atomic(), irqs_disabled(), | |
9567 | current->pid, current->comm); | |
9568 | ||
9569 | debug_show_held_locks(current); | |
9570 | dump_stack(); | |
9571 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
9572 | } | |
9573 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
9574 | |
9575 | #ifdef CONFIG_SMP | |
9576 | void __cant_migrate(const char *file, int line) | |
9577 | { | |
9578 | static unsigned long prev_jiffy; | |
9579 | ||
9580 | if (irqs_disabled()) | |
9581 | return; | |
9582 | ||
9583 | if (is_migration_disabled(current)) | |
9584 | return; | |
9585 | ||
9586 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
9587 | return; | |
9588 | ||
9589 | if (preempt_count() > 0) | |
9590 | return; | |
9591 | ||
9592 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9593 | return; | |
9594 | prev_jiffy = jiffies; | |
9595 | ||
9596 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
9597 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
9598 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
9599 | current->pid, current->comm); | |
9600 | ||
9601 | debug_show_held_locks(current); | |
9602 | dump_stack(); | |
9603 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
9604 | } | |
9605 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
9606 | #endif | |
1da177e4 LT |
9607 | #endif |
9608 | ||
9609 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 9610 | void normalize_rt_tasks(void) |
3a5e4dc1 | 9611 | { |
dbc7f069 | 9612 | struct task_struct *g, *p; |
d50dde5a DF |
9613 | struct sched_attr attr = { |
9614 | .sched_policy = SCHED_NORMAL, | |
9615 | }; | |
1da177e4 | 9616 | |
3472eaa1 | 9617 | read_lock(&tasklist_lock); |
5d07f420 | 9618 | for_each_process_thread(g, p) { |
178be793 IM |
9619 | /* |
9620 | * Only normalize user tasks: | |
9621 | */ | |
3472eaa1 | 9622 | if (p->flags & PF_KTHREAD) |
178be793 IM |
9623 | continue; |
9624 | ||
4fa8d299 | 9625 | p->se.exec_start = 0; |
ceeadb83 YS |
9626 | schedstat_set(p->stats.wait_start, 0); |
9627 | schedstat_set(p->stats.sleep_start, 0); | |
9628 | schedstat_set(p->stats.block_start, 0); | |
dd41f596 | 9629 | |
aab03e05 | 9630 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
9631 | /* |
9632 | * Renice negative nice level userspace | |
9633 | * tasks back to 0: | |
9634 | */ | |
3472eaa1 | 9635 | if (task_nice(p) < 0) |
dd41f596 | 9636 | set_user_nice(p, 0); |
1da177e4 | 9637 | continue; |
dd41f596 | 9638 | } |
1da177e4 | 9639 | |
dbc7f069 | 9640 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 9641 | } |
3472eaa1 | 9642 | read_unlock(&tasklist_lock); |
1da177e4 LT |
9643 | } |
9644 | ||
9645 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 9646 | |
67fc4e0c | 9647 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 9648 | /* |
67fc4e0c | 9649 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
9650 | * |
9651 | * They can only be called when the whole system has been | |
9652 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9653 | * activity can take place. Using them for anything else would | |
9654 | * be a serious bug, and as a result, they aren't even visible | |
9655 | * under any other configuration. | |
9656 | */ | |
9657 | ||
9658 | /** | |
d1ccc66d | 9659 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
9660 | * @cpu: the processor in question. |
9661 | * | |
9662 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
9663 | * |
9664 | * Return: The current task for @cpu. | |
1df5c10a | 9665 | */ |
36c8b586 | 9666 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9667 | { |
9668 | return cpu_curr(cpu); | |
9669 | } | |
9670 | ||
67fc4e0c JW |
9671 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
9672 | ||
9673 | #ifdef CONFIG_IA64 | |
1df5c10a | 9674 | /** |
5feeb783 | 9675 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
9676 | * @cpu: the processor in question. |
9677 | * @p: the task pointer to set. | |
9678 | * | |
9679 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 9680 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 9681 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
9682 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9683 | * and caller must save the original value of the current task (see | |
9684 | * curr_task() above) and restore that value before reenabling interrupts and | |
9685 | * re-starting the system. | |
9686 | * | |
9687 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9688 | */ | |
a458ae2e | 9689 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9690 | { |
9691 | cpu_curr(cpu) = p; | |
9692 | } | |
9693 | ||
9694 | #endif | |
29f59db3 | 9695 | |
7c941438 | 9696 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
9697 | /* task_group_lock serializes the addition/removal of task groups */ |
9698 | static DEFINE_SPINLOCK(task_group_lock); | |
9699 | ||
2480c093 PB |
9700 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
9701 | struct task_group *parent) | |
9702 | { | |
9703 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 9704 | enum uclamp_id clamp_id; |
2480c093 PB |
9705 | |
9706 | for_each_clamp_id(clamp_id) { | |
9707 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
9708 | uclamp_none(clamp_id), false); | |
0b60ba2d | 9709 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
9710 | } |
9711 | #endif | |
9712 | } | |
9713 | ||
2f5177f0 | 9714 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
9715 | { |
9716 | free_fair_sched_group(tg); | |
9717 | free_rt_sched_group(tg); | |
e9aa1dd1 | 9718 | autogroup_free(tg); |
b0367629 | 9719 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
9720 | } |
9721 | ||
b027789e MK |
9722 | static void sched_free_group_rcu(struct rcu_head *rcu) |
9723 | { | |
9724 | sched_free_group(container_of(rcu, struct task_group, rcu)); | |
9725 | } | |
9726 | ||
9727 | static void sched_unregister_group(struct task_group *tg) | |
9728 | { | |
9729 | unregister_fair_sched_group(tg); | |
9730 | unregister_rt_sched_group(tg); | |
9731 | /* | |
9732 | * We have to wait for yet another RCU grace period to expire, as | |
9733 | * print_cfs_stats() might run concurrently. | |
9734 | */ | |
9735 | call_rcu(&tg->rcu, sched_free_group_rcu); | |
9736 | } | |
9737 | ||
bccbe08a | 9738 | /* allocate runqueue etc for a new task group */ |
ec7dc8ac | 9739 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9740 | { |
9741 | struct task_group *tg; | |
bccbe08a | 9742 | |
b0367629 | 9743 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
9744 | if (!tg) |
9745 | return ERR_PTR(-ENOMEM); | |
9746 | ||
ec7dc8ac | 9747 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9748 | goto err; |
9749 | ||
ec7dc8ac | 9750 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9751 | goto err; |
9752 | ||
2480c093 PB |
9753 | alloc_uclamp_sched_group(tg, parent); |
9754 | ||
ace783b9 LZ |
9755 | return tg; |
9756 | ||
9757 | err: | |
2f5177f0 | 9758 | sched_free_group(tg); |
ace783b9 LZ |
9759 | return ERR_PTR(-ENOMEM); |
9760 | } | |
9761 | ||
9762 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
9763 | { | |
9764 | unsigned long flags; | |
9765 | ||
8ed36996 | 9766 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 9767 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 9768 | |
d1ccc66d IM |
9769 | /* Root should already exist: */ |
9770 | WARN_ON(!parent); | |
f473aa5e PZ |
9771 | |
9772 | tg->parent = parent; | |
f473aa5e | 9773 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9774 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9775 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
9776 | |
9777 | online_fair_sched_group(tg); | |
29f59db3 SV |
9778 | } |
9779 | ||
9b5b7751 | 9780 | /* rcu callback to free various structures associated with a task group */ |
b027789e | 9781 | static void sched_unregister_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9782 | { |
d1ccc66d | 9783 | /* Now it should be safe to free those cfs_rqs: */ |
b027789e | 9784 | sched_unregister_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9785 | } |
9786 | ||
4cf86d77 | 9787 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 9788 | { |
d1ccc66d | 9789 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
b027789e | 9790 | call_rcu(&tg->rcu, sched_unregister_group_rcu); |
ace783b9 LZ |
9791 | } |
9792 | ||
b027789e | 9793 | void sched_release_group(struct task_group *tg) |
29f59db3 | 9794 | { |
8ed36996 | 9795 | unsigned long flags; |
29f59db3 | 9796 | |
b027789e MK |
9797 | /* |
9798 | * Unlink first, to avoid walk_tg_tree_from() from finding us (via | |
9799 | * sched_cfs_period_timer()). | |
9800 | * | |
9801 | * For this to be effective, we have to wait for all pending users of | |
9802 | * this task group to leave their RCU critical section to ensure no new | |
9803 | * user will see our dying task group any more. Specifically ensure | |
9804 | * that tg_unthrottle_up() won't add decayed cfs_rq's to it. | |
9805 | * | |
9806 | * We therefore defer calling unregister_fair_sched_group() to | |
9807 | * sched_unregister_group() which is guarantied to get called only after the | |
9808 | * current RCU grace period has expired. | |
9809 | */ | |
3d4b47b4 | 9810 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 9811 | list_del_rcu(&tg->list); |
f473aa5e | 9812 | list_del_rcu(&tg->siblings); |
8ed36996 | 9813 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
9814 | } |
9815 | ||
ea86cb4b | 9816 | static void sched_change_group(struct task_struct *tsk, int type) |
29f59db3 | 9817 | { |
8323f26c | 9818 | struct task_group *tg; |
29f59db3 | 9819 | |
f7b8a47d KT |
9820 | /* |
9821 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
9822 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
9823 | * to prevent lockdep warnings. | |
9824 | */ | |
9825 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
9826 | struct task_group, css); |
9827 | tg = autogroup_task_group(tsk, tg); | |
9828 | tsk->sched_task_group = tg; | |
9829 | ||
810b3817 | 9830 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b VG |
9831 | if (tsk->sched_class->task_change_group) |
9832 | tsk->sched_class->task_change_group(tsk, type); | |
b2b5ce02 | 9833 | else |
810b3817 | 9834 | #endif |
b2b5ce02 | 9835 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
9836 | } |
9837 | ||
9838 | /* | |
9839 | * Change task's runqueue when it moves between groups. | |
9840 | * | |
9841 | * The caller of this function should have put the task in its new group by | |
9842 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
9843 | * its new group. | |
9844 | */ | |
9845 | void sched_move_task(struct task_struct *tsk) | |
9846 | { | |
7a57f32a PZ |
9847 | int queued, running, queue_flags = |
9848 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
9849 | struct rq_flags rf; |
9850 | struct rq *rq; | |
9851 | ||
9852 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 9853 | update_rq_clock(rq); |
ea86cb4b VG |
9854 | |
9855 | running = task_current(rq, tsk); | |
9856 | queued = task_on_rq_queued(tsk); | |
9857 | ||
9858 | if (queued) | |
7a57f32a | 9859 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 9860 | if (running) |
ea86cb4b VG |
9861 | put_prev_task(rq, tsk); |
9862 | ||
9863 | sched_change_group(tsk, TASK_MOVE_GROUP); | |
810b3817 | 9864 | |
da0c1e65 | 9865 | if (queued) |
7a57f32a | 9866 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 9867 | if (running) { |
03b7fad1 | 9868 | set_next_task(rq, tsk); |
2a4b03ff VG |
9869 | /* |
9870 | * After changing group, the running task may have joined a | |
9871 | * throttled one but it's still the running task. Trigger a | |
9872 | * resched to make sure that task can still run. | |
9873 | */ | |
9874 | resched_curr(rq); | |
9875 | } | |
29f59db3 | 9876 | |
eb580751 | 9877 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 9878 | } |
68318b8e | 9879 | |
a7c6d554 | 9880 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 9881 | { |
a7c6d554 | 9882 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
9883 | } |
9884 | ||
eb95419b TH |
9885 | static struct cgroup_subsys_state * |
9886 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 9887 | { |
eb95419b TH |
9888 | struct task_group *parent = css_tg(parent_css); |
9889 | struct task_group *tg; | |
68318b8e | 9890 | |
eb95419b | 9891 | if (!parent) { |
68318b8e | 9892 | /* This is early initialization for the top cgroup */ |
07e06b01 | 9893 | return &root_task_group.css; |
68318b8e SV |
9894 | } |
9895 | ||
ec7dc8ac | 9896 | tg = sched_create_group(parent); |
68318b8e SV |
9897 | if (IS_ERR(tg)) |
9898 | return ERR_PTR(-ENOMEM); | |
9899 | ||
68318b8e SV |
9900 | return &tg->css; |
9901 | } | |
9902 | ||
96b77745 KK |
9903 | /* Expose task group only after completing cgroup initialization */ |
9904 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
9905 | { | |
9906 | struct task_group *tg = css_tg(css); | |
9907 | struct task_group *parent = css_tg(css->parent); | |
9908 | ||
9909 | if (parent) | |
9910 | sched_online_group(tg, parent); | |
7226017a QY |
9911 | |
9912 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
9913 | /* Propagate the effective uclamp value for the new group */ | |
93b73858 QY |
9914 | mutex_lock(&uclamp_mutex); |
9915 | rcu_read_lock(); | |
7226017a | 9916 | cpu_util_update_eff(css); |
93b73858 QY |
9917 | rcu_read_unlock(); |
9918 | mutex_unlock(&uclamp_mutex); | |
7226017a QY |
9919 | #endif |
9920 | ||
96b77745 KK |
9921 | return 0; |
9922 | } | |
9923 | ||
2f5177f0 | 9924 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 9925 | { |
eb95419b | 9926 | struct task_group *tg = css_tg(css); |
ace783b9 | 9927 | |
b027789e | 9928 | sched_release_group(tg); |
ace783b9 LZ |
9929 | } |
9930 | ||
eb95419b | 9931 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 9932 | { |
eb95419b | 9933 | struct task_group *tg = css_tg(css); |
68318b8e | 9934 | |
2f5177f0 PZ |
9935 | /* |
9936 | * Relies on the RCU grace period between css_released() and this. | |
9937 | */ | |
b027789e | 9938 | sched_unregister_group(tg); |
ace783b9 LZ |
9939 | } |
9940 | ||
ea86cb4b VG |
9941 | /* |
9942 | * This is called before wake_up_new_task(), therefore we really only | |
9943 | * have to set its group bits, all the other stuff does not apply. | |
9944 | */ | |
b53202e6 | 9945 | static void cpu_cgroup_fork(struct task_struct *task) |
eeb61e53 | 9946 | { |
ea86cb4b VG |
9947 | struct rq_flags rf; |
9948 | struct rq *rq; | |
9949 | ||
9950 | rq = task_rq_lock(task, &rf); | |
9951 | ||
80f5c1b8 | 9952 | update_rq_clock(rq); |
ea86cb4b VG |
9953 | sched_change_group(task, TASK_SET_GROUP); |
9954 | ||
9955 | task_rq_unlock(rq, task, &rf); | |
eeb61e53 KT |
9956 | } |
9957 | ||
1f7dd3e5 | 9958 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 9959 | { |
bb9d97b6 | 9960 | struct task_struct *task; |
1f7dd3e5 | 9961 | struct cgroup_subsys_state *css; |
7dc603c9 | 9962 | int ret = 0; |
bb9d97b6 | 9963 | |
1f7dd3e5 | 9964 | cgroup_taskset_for_each(task, css, tset) { |
b68aa230 | 9965 | #ifdef CONFIG_RT_GROUP_SCHED |
eb95419b | 9966 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 9967 | return -EINVAL; |
b68aa230 | 9968 | #endif |
7dc603c9 | 9969 | /* |
b19a888c | 9970 | * Serialize against wake_up_new_task() such that if it's |
7dc603c9 PZ |
9971 | * running, we're sure to observe its full state. |
9972 | */ | |
9973 | raw_spin_lock_irq(&task->pi_lock); | |
9974 | /* | |
9975 | * Avoid calling sched_move_task() before wake_up_new_task() | |
9976 | * has happened. This would lead to problems with PELT, due to | |
9977 | * move wanting to detach+attach while we're not attached yet. | |
9978 | */ | |
2f064a59 | 9979 | if (READ_ONCE(task->__state) == TASK_NEW) |
7dc603c9 PZ |
9980 | ret = -EINVAL; |
9981 | raw_spin_unlock_irq(&task->pi_lock); | |
9982 | ||
9983 | if (ret) | |
9984 | break; | |
bb9d97b6 | 9985 | } |
7dc603c9 | 9986 | return ret; |
be367d09 | 9987 | } |
68318b8e | 9988 | |
1f7dd3e5 | 9989 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 9990 | { |
bb9d97b6 | 9991 | struct task_struct *task; |
1f7dd3e5 | 9992 | struct cgroup_subsys_state *css; |
bb9d97b6 | 9993 | |
1f7dd3e5 | 9994 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 9995 | sched_move_task(task); |
68318b8e SV |
9996 | } |
9997 | ||
2480c093 | 9998 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
9999 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
10000 | { | |
10001 | struct cgroup_subsys_state *top_css = css; | |
10002 | struct uclamp_se *uc_parent = NULL; | |
10003 | struct uclamp_se *uc_se = NULL; | |
10004 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 10005 | enum uclamp_id clamp_id; |
0b60ba2d PB |
10006 | unsigned int clamps; |
10007 | ||
93b73858 QY |
10008 | lockdep_assert_held(&uclamp_mutex); |
10009 | SCHED_WARN_ON(!rcu_read_lock_held()); | |
10010 | ||
0b60ba2d PB |
10011 | css_for_each_descendant_pre(css, top_css) { |
10012 | uc_parent = css_tg(css)->parent | |
10013 | ? css_tg(css)->parent->uclamp : NULL; | |
10014 | ||
10015 | for_each_clamp_id(clamp_id) { | |
10016 | /* Assume effective clamps matches requested clamps */ | |
10017 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
10018 | /* Cap effective clamps with parent's effective clamps */ | |
10019 | if (uc_parent && | |
10020 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
10021 | eff[clamp_id] = uc_parent[clamp_id].value; | |
10022 | } | |
10023 | } | |
10024 | /* Ensure protection is always capped by limit */ | |
10025 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
10026 | ||
10027 | /* Propagate most restrictive effective clamps */ | |
10028 | clamps = 0x0; | |
10029 | uc_se = css_tg(css)->uclamp; | |
10030 | for_each_clamp_id(clamp_id) { | |
10031 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
10032 | continue; | |
10033 | uc_se[clamp_id].value = eff[clamp_id]; | |
10034 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
10035 | clamps |= (0x1 << clamp_id); | |
10036 | } | |
babbe170 | 10037 | if (!clamps) { |
0b60ba2d | 10038 | css = css_rightmost_descendant(css); |
babbe170 PB |
10039 | continue; |
10040 | } | |
10041 | ||
10042 | /* Immediately update descendants RUNNABLE tasks */ | |
0213b708 | 10043 | uclamp_update_active_tasks(css); |
0b60ba2d PB |
10044 | } |
10045 | } | |
2480c093 PB |
10046 | |
10047 | /* | |
10048 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
10049 | * C expression. Since there is no way to convert a macro argument (N) into a | |
10050 | * character constant, use two levels of macros. | |
10051 | */ | |
10052 | #define _POW10(exp) ((unsigned int)1e##exp) | |
10053 | #define POW10(exp) _POW10(exp) | |
10054 | ||
10055 | struct uclamp_request { | |
10056 | #define UCLAMP_PERCENT_SHIFT 2 | |
10057 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
10058 | s64 percent; | |
10059 | u64 util; | |
10060 | int ret; | |
10061 | }; | |
10062 | ||
10063 | static inline struct uclamp_request | |
10064 | capacity_from_percent(char *buf) | |
10065 | { | |
10066 | struct uclamp_request req = { | |
10067 | .percent = UCLAMP_PERCENT_SCALE, | |
10068 | .util = SCHED_CAPACITY_SCALE, | |
10069 | .ret = 0, | |
10070 | }; | |
10071 | ||
10072 | buf = strim(buf); | |
10073 | if (strcmp(buf, "max")) { | |
10074 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
10075 | &req.percent); | |
10076 | if (req.ret) | |
10077 | return req; | |
b562d140 | 10078 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
10079 | req.ret = -ERANGE; |
10080 | return req; | |
10081 | } | |
10082 | ||
10083 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
10084 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
10085 | } | |
10086 | ||
10087 | return req; | |
10088 | } | |
10089 | ||
10090 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
10091 | size_t nbytes, loff_t off, | |
10092 | enum uclamp_id clamp_id) | |
10093 | { | |
10094 | struct uclamp_request req; | |
10095 | struct task_group *tg; | |
10096 | ||
10097 | req = capacity_from_percent(buf); | |
10098 | if (req.ret) | |
10099 | return req.ret; | |
10100 | ||
46609ce2 QY |
10101 | static_branch_enable(&sched_uclamp_used); |
10102 | ||
2480c093 PB |
10103 | mutex_lock(&uclamp_mutex); |
10104 | rcu_read_lock(); | |
10105 | ||
10106 | tg = css_tg(of_css(of)); | |
10107 | if (tg->uclamp_req[clamp_id].value != req.util) | |
10108 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
10109 | ||
10110 | /* | |
10111 | * Because of not recoverable conversion rounding we keep track of the | |
10112 | * exact requested value | |
10113 | */ | |
10114 | tg->uclamp_pct[clamp_id] = req.percent; | |
10115 | ||
0b60ba2d PB |
10116 | /* Update effective clamps to track the most restrictive value */ |
10117 | cpu_util_update_eff(of_css(of)); | |
10118 | ||
2480c093 PB |
10119 | rcu_read_unlock(); |
10120 | mutex_unlock(&uclamp_mutex); | |
10121 | ||
10122 | return nbytes; | |
10123 | } | |
10124 | ||
10125 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
10126 | char *buf, size_t nbytes, | |
10127 | loff_t off) | |
10128 | { | |
10129 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
10130 | } | |
10131 | ||
10132 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
10133 | char *buf, size_t nbytes, | |
10134 | loff_t off) | |
10135 | { | |
10136 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
10137 | } | |
10138 | ||
10139 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
10140 | enum uclamp_id clamp_id) | |
10141 | { | |
10142 | struct task_group *tg; | |
10143 | u64 util_clamp; | |
10144 | u64 percent; | |
10145 | u32 rem; | |
10146 | ||
10147 | rcu_read_lock(); | |
10148 | tg = css_tg(seq_css(sf)); | |
10149 | util_clamp = tg->uclamp_req[clamp_id].value; | |
10150 | rcu_read_unlock(); | |
10151 | ||
10152 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
10153 | seq_puts(sf, "max\n"); | |
10154 | return; | |
10155 | } | |
10156 | ||
10157 | percent = tg->uclamp_pct[clamp_id]; | |
10158 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
10159 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
10160 | } | |
10161 | ||
10162 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
10163 | { | |
10164 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
10165 | return 0; | |
10166 | } | |
10167 | ||
10168 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
10169 | { | |
10170 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
10171 | return 0; | |
10172 | } | |
10173 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
10174 | ||
052f1dc7 | 10175 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
10176 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
10177 | struct cftype *cftype, u64 shareval) | |
68318b8e | 10178 | { |
5b61d50a KK |
10179 | if (shareval > scale_load_down(ULONG_MAX)) |
10180 | shareval = MAX_SHARES; | |
182446d0 | 10181 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
10182 | } |
10183 | ||
182446d0 TH |
10184 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
10185 | struct cftype *cft) | |
68318b8e | 10186 | { |
182446d0 | 10187 | struct task_group *tg = css_tg(css); |
68318b8e | 10188 | |
c8b28116 | 10189 | return (u64) scale_load_down(tg->shares); |
68318b8e | 10190 | } |
ab84d31e PT |
10191 | |
10192 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
10193 | static DEFINE_MUTEX(cfs_constraints_mutex); |
10194 | ||
ab84d31e | 10195 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 10196 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
10197 | /* More than 203 days if BW_SHIFT equals 20. */ |
10198 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 10199 | |
a790de99 PT |
10200 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
10201 | ||
f4183717 HC |
10202 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota, |
10203 | u64 burst) | |
ab84d31e | 10204 | { |
56f570e5 | 10205 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 10206 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
10207 | |
10208 | if (tg == &root_task_group) | |
10209 | return -EINVAL; | |
10210 | ||
10211 | /* | |
10212 | * Ensure we have at some amount of bandwidth every period. This is | |
10213 | * to prevent reaching a state of large arrears when throttled via | |
10214 | * entity_tick() resulting in prolonged exit starvation. | |
10215 | */ | |
10216 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
10217 | return -EINVAL; | |
10218 | ||
10219 | /* | |
3b03706f | 10220 | * Likewise, bound things on the other side by preventing insane quota |
ab84d31e PT |
10221 | * periods. This also allows us to normalize in computing quota |
10222 | * feasibility. | |
10223 | */ | |
10224 | if (period > max_cfs_quota_period) | |
10225 | return -EINVAL; | |
10226 | ||
d505b8af HC |
10227 | /* |
10228 | * Bound quota to defend quota against overflow during bandwidth shift. | |
10229 | */ | |
10230 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
10231 | return -EINVAL; | |
10232 | ||
f4183717 HC |
10233 | if (quota != RUNTIME_INF && (burst > quota || |
10234 | burst + quota > max_cfs_runtime)) | |
10235 | return -EINVAL; | |
10236 | ||
0e59bdae KT |
10237 | /* |
10238 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
10239 | * unthrottle_offline_cfs_rqs(). | |
10240 | */ | |
746f5ea9 | 10241 | cpus_read_lock(); |
a790de99 PT |
10242 | mutex_lock(&cfs_constraints_mutex); |
10243 | ret = __cfs_schedulable(tg, period, quota); | |
10244 | if (ret) | |
10245 | goto out_unlock; | |
10246 | ||
58088ad0 | 10247 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 10248 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
10249 | /* |
10250 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
10251 | * before making related changes, and on->off must occur afterwards | |
10252 | */ | |
10253 | if (runtime_enabled && !runtime_was_enabled) | |
10254 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
10255 | raw_spin_lock_irq(&cfs_b->lock); |
10256 | cfs_b->period = ns_to_ktime(period); | |
10257 | cfs_b->quota = quota; | |
f4183717 | 10258 | cfs_b->burst = burst; |
58088ad0 | 10259 | |
a9cf55b2 | 10260 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
10261 | |
10262 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
10263 | if (runtime_enabled) |
10264 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 10265 | |
ab84d31e PT |
10266 | raw_spin_unlock_irq(&cfs_b->lock); |
10267 | ||
0e59bdae | 10268 | for_each_online_cpu(i) { |
ab84d31e | 10269 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 10270 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 10271 | struct rq_flags rf; |
ab84d31e | 10272 | |
8a8c69c3 | 10273 | rq_lock_irq(rq, &rf); |
58088ad0 | 10274 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 10275 | cfs_rq->runtime_remaining = 0; |
671fd9da | 10276 | |
029632fb | 10277 | if (cfs_rq->throttled) |
671fd9da | 10278 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 10279 | rq_unlock_irq(rq, &rf); |
ab84d31e | 10280 | } |
1ee14e6c BS |
10281 | if (runtime_was_enabled && !runtime_enabled) |
10282 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
10283 | out_unlock: |
10284 | mutex_unlock(&cfs_constraints_mutex); | |
746f5ea9 | 10285 | cpus_read_unlock(); |
ab84d31e | 10286 | |
a790de99 | 10287 | return ret; |
ab84d31e PT |
10288 | } |
10289 | ||
b1546edc | 10290 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e | 10291 | { |
f4183717 | 10292 | u64 quota, period, burst; |
ab84d31e | 10293 | |
029632fb | 10294 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
f4183717 | 10295 | burst = tg->cfs_bandwidth.burst; |
ab84d31e PT |
10296 | if (cfs_quota_us < 0) |
10297 | quota = RUNTIME_INF; | |
1a8b4540 | 10298 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 10299 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
10300 | else |
10301 | return -EINVAL; | |
ab84d31e | 10302 | |
f4183717 | 10303 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10304 | } |
10305 | ||
b1546edc | 10306 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
10307 | { |
10308 | u64 quota_us; | |
10309 | ||
029632fb | 10310 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
10311 | return -1; |
10312 | ||
029632fb | 10313 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
10314 | do_div(quota_us, NSEC_PER_USEC); |
10315 | ||
10316 | return quota_us; | |
10317 | } | |
10318 | ||
b1546edc | 10319 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e | 10320 | { |
f4183717 | 10321 | u64 quota, period, burst; |
ab84d31e | 10322 | |
1a8b4540 KK |
10323 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
10324 | return -EINVAL; | |
10325 | ||
ab84d31e | 10326 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 10327 | quota = tg->cfs_bandwidth.quota; |
f4183717 | 10328 | burst = tg->cfs_bandwidth.burst; |
ab84d31e | 10329 | |
f4183717 | 10330 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10331 | } |
10332 | ||
b1546edc | 10333 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
10334 | { |
10335 | u64 cfs_period_us; | |
10336 | ||
029632fb | 10337 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
10338 | do_div(cfs_period_us, NSEC_PER_USEC); |
10339 | ||
10340 | return cfs_period_us; | |
10341 | } | |
10342 | ||
f4183717 HC |
10343 | static int tg_set_cfs_burst(struct task_group *tg, long cfs_burst_us) |
10344 | { | |
10345 | u64 quota, period, burst; | |
10346 | ||
10347 | if ((u64)cfs_burst_us > U64_MAX / NSEC_PER_USEC) | |
10348 | return -EINVAL; | |
10349 | ||
10350 | burst = (u64)cfs_burst_us * NSEC_PER_USEC; | |
10351 | period = ktime_to_ns(tg->cfs_bandwidth.period); | |
10352 | quota = tg->cfs_bandwidth.quota; | |
10353 | ||
10354 | return tg_set_cfs_bandwidth(tg, period, quota, burst); | |
10355 | } | |
10356 | ||
10357 | static long tg_get_cfs_burst(struct task_group *tg) | |
10358 | { | |
10359 | u64 burst_us; | |
10360 | ||
10361 | burst_us = tg->cfs_bandwidth.burst; | |
10362 | do_div(burst_us, NSEC_PER_USEC); | |
10363 | ||
10364 | return burst_us; | |
10365 | } | |
10366 | ||
182446d0 TH |
10367 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
10368 | struct cftype *cft) | |
ab84d31e | 10369 | { |
182446d0 | 10370 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
10371 | } |
10372 | ||
182446d0 TH |
10373 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
10374 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 10375 | { |
182446d0 | 10376 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
10377 | } |
10378 | ||
182446d0 TH |
10379 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
10380 | struct cftype *cft) | |
ab84d31e | 10381 | { |
182446d0 | 10382 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
10383 | } |
10384 | ||
182446d0 TH |
10385 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
10386 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 10387 | { |
182446d0 | 10388 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
10389 | } |
10390 | ||
f4183717 HC |
10391 | static u64 cpu_cfs_burst_read_u64(struct cgroup_subsys_state *css, |
10392 | struct cftype *cft) | |
10393 | { | |
10394 | return tg_get_cfs_burst(css_tg(css)); | |
10395 | } | |
10396 | ||
10397 | static int cpu_cfs_burst_write_u64(struct cgroup_subsys_state *css, | |
10398 | struct cftype *cftype, u64 cfs_burst_us) | |
10399 | { | |
10400 | return tg_set_cfs_burst(css_tg(css), cfs_burst_us); | |
10401 | } | |
10402 | ||
a790de99 PT |
10403 | struct cfs_schedulable_data { |
10404 | struct task_group *tg; | |
10405 | u64 period, quota; | |
10406 | }; | |
10407 | ||
10408 | /* | |
10409 | * normalize group quota/period to be quota/max_period | |
10410 | * note: units are usecs | |
10411 | */ | |
10412 | static u64 normalize_cfs_quota(struct task_group *tg, | |
10413 | struct cfs_schedulable_data *d) | |
10414 | { | |
10415 | u64 quota, period; | |
10416 | ||
10417 | if (tg == d->tg) { | |
10418 | period = d->period; | |
10419 | quota = d->quota; | |
10420 | } else { | |
10421 | period = tg_get_cfs_period(tg); | |
10422 | quota = tg_get_cfs_quota(tg); | |
10423 | } | |
10424 | ||
10425 | /* note: these should typically be equivalent */ | |
10426 | if (quota == RUNTIME_INF || quota == -1) | |
10427 | return RUNTIME_INF; | |
10428 | ||
10429 | return to_ratio(period, quota); | |
10430 | } | |
10431 | ||
10432 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
10433 | { | |
10434 | struct cfs_schedulable_data *d = data; | |
029632fb | 10435 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
10436 | s64 quota = 0, parent_quota = -1; |
10437 | ||
10438 | if (!tg->parent) { | |
10439 | quota = RUNTIME_INF; | |
10440 | } else { | |
029632fb | 10441 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
10442 | |
10443 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 10444 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
10445 | |
10446 | /* | |
c53593e5 TH |
10447 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
10448 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 10449 | * limit is set: |
a790de99 | 10450 | */ |
c53593e5 TH |
10451 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
10452 | quota = min(quota, parent_quota); | |
10453 | } else { | |
10454 | if (quota == RUNTIME_INF) | |
10455 | quota = parent_quota; | |
10456 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
10457 | return -EINVAL; | |
10458 | } | |
a790de99 | 10459 | } |
9c58c79a | 10460 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
10461 | |
10462 | return 0; | |
10463 | } | |
10464 | ||
10465 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
10466 | { | |
8277434e | 10467 | int ret; |
a790de99 PT |
10468 | struct cfs_schedulable_data data = { |
10469 | .tg = tg, | |
10470 | .period = period, | |
10471 | .quota = quota, | |
10472 | }; | |
10473 | ||
10474 | if (quota != RUNTIME_INF) { | |
10475 | do_div(data.period, NSEC_PER_USEC); | |
10476 | do_div(data.quota, NSEC_PER_USEC); | |
10477 | } | |
10478 | ||
8277434e PT |
10479 | rcu_read_lock(); |
10480 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
10481 | rcu_read_unlock(); | |
10482 | ||
10483 | return ret; | |
a790de99 | 10484 | } |
e8da1b18 | 10485 | |
a1f7164c | 10486 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 10487 | { |
2da8ca82 | 10488 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 10489 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 10490 | |
44ffc75b TH |
10491 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
10492 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
10493 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 10494 | |
3d6c50c2 | 10495 | if (schedstat_enabled() && tg != &root_task_group) { |
ceeadb83 | 10496 | struct sched_statistics *stats; |
3d6c50c2 YW |
10497 | u64 ws = 0; |
10498 | int i; | |
10499 | ||
ceeadb83 YS |
10500 | for_each_possible_cpu(i) { |
10501 | stats = __schedstats_from_se(tg->se[i]); | |
10502 | ws += schedstat_val(stats->wait_sum); | |
10503 | } | |
3d6c50c2 YW |
10504 | |
10505 | seq_printf(sf, "wait_sum %llu\n", ws); | |
10506 | } | |
10507 | ||
bcb1704a HC |
10508 | seq_printf(sf, "nr_bursts %d\n", cfs_b->nr_burst); |
10509 | seq_printf(sf, "burst_time %llu\n", cfs_b->burst_time); | |
10510 | ||
e8da1b18 NR |
10511 | return 0; |
10512 | } | |
ab84d31e | 10513 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 10514 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10515 | |
052f1dc7 | 10516 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
10517 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
10518 | struct cftype *cft, s64 val) | |
6f505b16 | 10519 | { |
182446d0 | 10520 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
10521 | } |
10522 | ||
182446d0 TH |
10523 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
10524 | struct cftype *cft) | |
6f505b16 | 10525 | { |
182446d0 | 10526 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 10527 | } |
d0b27fa7 | 10528 | |
182446d0 TH |
10529 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
10530 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 10531 | { |
182446d0 | 10532 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
10533 | } |
10534 | ||
182446d0 TH |
10535 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
10536 | struct cftype *cft) | |
d0b27fa7 | 10537 | { |
182446d0 | 10538 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 10539 | } |
6d6bc0ad | 10540 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10541 | |
30400039 JD |
10542 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10543 | static s64 cpu_idle_read_s64(struct cgroup_subsys_state *css, | |
10544 | struct cftype *cft) | |
10545 | { | |
10546 | return css_tg(css)->idle; | |
10547 | } | |
10548 | ||
10549 | static int cpu_idle_write_s64(struct cgroup_subsys_state *css, | |
10550 | struct cftype *cft, s64 idle) | |
10551 | { | |
10552 | return sched_group_set_idle(css_tg(css), idle); | |
10553 | } | |
10554 | #endif | |
10555 | ||
a1f7164c | 10556 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 10557 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10558 | { |
10559 | .name = "shares", | |
f4c753b7 PM |
10560 | .read_u64 = cpu_shares_read_u64, |
10561 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10562 | }, |
30400039 JD |
10563 | { |
10564 | .name = "idle", | |
10565 | .read_s64 = cpu_idle_read_s64, | |
10566 | .write_s64 = cpu_idle_write_s64, | |
10567 | }, | |
052f1dc7 | 10568 | #endif |
ab84d31e PT |
10569 | #ifdef CONFIG_CFS_BANDWIDTH |
10570 | { | |
10571 | .name = "cfs_quota_us", | |
10572 | .read_s64 = cpu_cfs_quota_read_s64, | |
10573 | .write_s64 = cpu_cfs_quota_write_s64, | |
10574 | }, | |
10575 | { | |
10576 | .name = "cfs_period_us", | |
10577 | .read_u64 = cpu_cfs_period_read_u64, | |
10578 | .write_u64 = cpu_cfs_period_write_u64, | |
10579 | }, | |
f4183717 HC |
10580 | { |
10581 | .name = "cfs_burst_us", | |
10582 | .read_u64 = cpu_cfs_burst_read_u64, | |
10583 | .write_u64 = cpu_cfs_burst_write_u64, | |
10584 | }, | |
e8da1b18 NR |
10585 | { |
10586 | .name = "stat", | |
a1f7164c | 10587 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 10588 | }, |
ab84d31e | 10589 | #endif |
052f1dc7 | 10590 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10591 | { |
9f0c1e56 | 10592 | .name = "rt_runtime_us", |
06ecb27c PM |
10593 | .read_s64 = cpu_rt_runtime_read, |
10594 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10595 | }, |
d0b27fa7 PZ |
10596 | { |
10597 | .name = "rt_period_us", | |
f4c753b7 PM |
10598 | .read_u64 = cpu_rt_period_read_uint, |
10599 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10600 | }, |
2480c093 PB |
10601 | #endif |
10602 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
10603 | { | |
10604 | .name = "uclamp.min", | |
10605 | .flags = CFTYPE_NOT_ON_ROOT, | |
10606 | .seq_show = cpu_uclamp_min_show, | |
10607 | .write = cpu_uclamp_min_write, | |
10608 | }, | |
10609 | { | |
10610 | .name = "uclamp.max", | |
10611 | .flags = CFTYPE_NOT_ON_ROOT, | |
10612 | .seq_show = cpu_uclamp_max_show, | |
10613 | .write = cpu_uclamp_max_write, | |
10614 | }, | |
052f1dc7 | 10615 | #endif |
d1ccc66d | 10616 | { } /* Terminate */ |
68318b8e SV |
10617 | }; |
10618 | ||
d41bf8c9 TH |
10619 | static int cpu_extra_stat_show(struct seq_file *sf, |
10620 | struct cgroup_subsys_state *css) | |
0d593634 | 10621 | { |
0d593634 TH |
10622 | #ifdef CONFIG_CFS_BANDWIDTH |
10623 | { | |
d41bf8c9 | 10624 | struct task_group *tg = css_tg(css); |
0d593634 | 10625 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
bcb1704a | 10626 | u64 throttled_usec, burst_usec; |
0d593634 TH |
10627 | |
10628 | throttled_usec = cfs_b->throttled_time; | |
10629 | do_div(throttled_usec, NSEC_PER_USEC); | |
bcb1704a HC |
10630 | burst_usec = cfs_b->burst_time; |
10631 | do_div(burst_usec, NSEC_PER_USEC); | |
0d593634 TH |
10632 | |
10633 | seq_printf(sf, "nr_periods %d\n" | |
10634 | "nr_throttled %d\n" | |
bcb1704a HC |
10635 | "throttled_usec %llu\n" |
10636 | "nr_bursts %d\n" | |
10637 | "burst_usec %llu\n", | |
0d593634 | 10638 | cfs_b->nr_periods, cfs_b->nr_throttled, |
bcb1704a | 10639 | throttled_usec, cfs_b->nr_burst, burst_usec); |
0d593634 TH |
10640 | } |
10641 | #endif | |
10642 | return 0; | |
10643 | } | |
10644 | ||
10645 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
10646 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
10647 | struct cftype *cft) | |
10648 | { | |
10649 | struct task_group *tg = css_tg(css); | |
10650 | u64 weight = scale_load_down(tg->shares); | |
10651 | ||
10652 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
10653 | } | |
10654 | ||
10655 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
10656 | struct cftype *cft, u64 weight) | |
10657 | { | |
10658 | /* | |
10659 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
10660 | * values which are 1, 100 and 10000 respectively. While it loses | |
10661 | * a bit of range on both ends, it maps pretty well onto the shares | |
10662 | * value used by scheduler and the round-trip conversions preserve | |
10663 | * the original value over the entire range. | |
10664 | */ | |
10665 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
10666 | return -ERANGE; | |
10667 | ||
10668 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
10669 | ||
10670 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
10671 | } | |
10672 | ||
10673 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
10674 | struct cftype *cft) | |
10675 | { | |
10676 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
10677 | int last_delta = INT_MAX; | |
10678 | int prio, delta; | |
10679 | ||
10680 | /* find the closest nice value to the current weight */ | |
10681 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
10682 | delta = abs(sched_prio_to_weight[prio] - weight); | |
10683 | if (delta >= last_delta) | |
10684 | break; | |
10685 | last_delta = delta; | |
10686 | } | |
10687 | ||
10688 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
10689 | } | |
10690 | ||
10691 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
10692 | struct cftype *cft, s64 nice) | |
10693 | { | |
10694 | unsigned long weight; | |
7281c8de | 10695 | int idx; |
0d593634 TH |
10696 | |
10697 | if (nice < MIN_NICE || nice > MAX_NICE) | |
10698 | return -ERANGE; | |
10699 | ||
7281c8de PZ |
10700 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
10701 | idx = array_index_nospec(idx, 40); | |
10702 | weight = sched_prio_to_weight[idx]; | |
10703 | ||
0d593634 TH |
10704 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
10705 | } | |
10706 | #endif | |
10707 | ||
10708 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
10709 | long period, long quota) | |
10710 | { | |
10711 | if (quota < 0) | |
10712 | seq_puts(sf, "max"); | |
10713 | else | |
10714 | seq_printf(sf, "%ld", quota); | |
10715 | ||
10716 | seq_printf(sf, " %ld\n", period); | |
10717 | } | |
10718 | ||
10719 | /* caller should put the current value in *@periodp before calling */ | |
10720 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
10721 | u64 *periodp, u64 *quotap) | |
10722 | { | |
10723 | char tok[21]; /* U64_MAX */ | |
10724 | ||
4c47acd8 | 10725 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
10726 | return -EINVAL; |
10727 | ||
10728 | *periodp *= NSEC_PER_USEC; | |
10729 | ||
10730 | if (sscanf(tok, "%llu", quotap)) | |
10731 | *quotap *= NSEC_PER_USEC; | |
10732 | else if (!strcmp(tok, "max")) | |
10733 | *quotap = RUNTIME_INF; | |
10734 | else | |
10735 | return -EINVAL; | |
10736 | ||
10737 | return 0; | |
10738 | } | |
10739 | ||
10740 | #ifdef CONFIG_CFS_BANDWIDTH | |
10741 | static int cpu_max_show(struct seq_file *sf, void *v) | |
10742 | { | |
10743 | struct task_group *tg = css_tg(seq_css(sf)); | |
10744 | ||
10745 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
10746 | return 0; | |
10747 | } | |
10748 | ||
10749 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
10750 | char *buf, size_t nbytes, loff_t off) | |
10751 | { | |
10752 | struct task_group *tg = css_tg(of_css(of)); | |
10753 | u64 period = tg_get_cfs_period(tg); | |
f4183717 | 10754 | u64 burst = tg_get_cfs_burst(tg); |
0d593634 TH |
10755 | u64 quota; |
10756 | int ret; | |
10757 | ||
10758 | ret = cpu_period_quota_parse(buf, &period, "a); | |
10759 | if (!ret) | |
f4183717 | 10760 | ret = tg_set_cfs_bandwidth(tg, period, quota, burst); |
0d593634 TH |
10761 | return ret ?: nbytes; |
10762 | } | |
10763 | #endif | |
10764 | ||
10765 | static struct cftype cpu_files[] = { | |
0d593634 TH |
10766 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10767 | { | |
10768 | .name = "weight", | |
10769 | .flags = CFTYPE_NOT_ON_ROOT, | |
10770 | .read_u64 = cpu_weight_read_u64, | |
10771 | .write_u64 = cpu_weight_write_u64, | |
10772 | }, | |
10773 | { | |
10774 | .name = "weight.nice", | |
10775 | .flags = CFTYPE_NOT_ON_ROOT, | |
10776 | .read_s64 = cpu_weight_nice_read_s64, | |
10777 | .write_s64 = cpu_weight_nice_write_s64, | |
10778 | }, | |
30400039 JD |
10779 | { |
10780 | .name = "idle", | |
10781 | .flags = CFTYPE_NOT_ON_ROOT, | |
10782 | .read_s64 = cpu_idle_read_s64, | |
10783 | .write_s64 = cpu_idle_write_s64, | |
10784 | }, | |
0d593634 TH |
10785 | #endif |
10786 | #ifdef CONFIG_CFS_BANDWIDTH | |
10787 | { | |
10788 | .name = "max", | |
10789 | .flags = CFTYPE_NOT_ON_ROOT, | |
10790 | .seq_show = cpu_max_show, | |
10791 | .write = cpu_max_write, | |
10792 | }, | |
f4183717 HC |
10793 | { |
10794 | .name = "max.burst", | |
10795 | .flags = CFTYPE_NOT_ON_ROOT, | |
10796 | .read_u64 = cpu_cfs_burst_read_u64, | |
10797 | .write_u64 = cpu_cfs_burst_write_u64, | |
10798 | }, | |
2480c093 PB |
10799 | #endif |
10800 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
10801 | { | |
10802 | .name = "uclamp.min", | |
10803 | .flags = CFTYPE_NOT_ON_ROOT, | |
10804 | .seq_show = cpu_uclamp_min_show, | |
10805 | .write = cpu_uclamp_min_write, | |
10806 | }, | |
10807 | { | |
10808 | .name = "uclamp.max", | |
10809 | .flags = CFTYPE_NOT_ON_ROOT, | |
10810 | .seq_show = cpu_uclamp_max_show, | |
10811 | .write = cpu_uclamp_max_write, | |
10812 | }, | |
0d593634 TH |
10813 | #endif |
10814 | { } /* terminate */ | |
10815 | }; | |
10816 | ||
073219e9 | 10817 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 10818 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 10819 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 10820 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 10821 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 10822 | .css_extra_stat_show = cpu_extra_stat_show, |
eeb61e53 | 10823 | .fork = cpu_cgroup_fork, |
bb9d97b6 TH |
10824 | .can_attach = cpu_cgroup_can_attach, |
10825 | .attach = cpu_cgroup_attach, | |
a1f7164c | 10826 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 10827 | .dfl_cftypes = cpu_files, |
b38e42e9 | 10828 | .early_init = true, |
0d593634 | 10829 | .threaded = true, |
68318b8e SV |
10830 | }; |
10831 | ||
052f1dc7 | 10832 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 10833 | |
b637a328 PM |
10834 | void dump_cpu_task(int cpu) |
10835 | { | |
10836 | pr_info("Task dump for CPU %d:\n", cpu); | |
10837 | sched_show_task(cpu_curr(cpu)); | |
10838 | } | |
ed82b8a1 AK |
10839 | |
10840 | /* | |
10841 | * Nice levels are multiplicative, with a gentle 10% change for every | |
10842 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
10843 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
10844 | * that remained on nice 0. | |
10845 | * | |
10846 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
10847 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
10848 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
10849 | * If a task goes up by ~10% and another task goes down by ~10% then | |
10850 | * the relative distance between them is ~25%.) | |
10851 | */ | |
10852 | const int sched_prio_to_weight[40] = { | |
10853 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
10854 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
10855 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
10856 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
10857 | /* 0 */ 1024, 820, 655, 526, 423, | |
10858 | /* 5 */ 335, 272, 215, 172, 137, | |
10859 | /* 10 */ 110, 87, 70, 56, 45, | |
10860 | /* 15 */ 36, 29, 23, 18, 15, | |
10861 | }; | |
10862 | ||
10863 | /* | |
10864 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
10865 | * | |
10866 | * In cases where the weight does not change often, we can use the | |
10867 | * precalculated inverse to speed up arithmetics by turning divisions | |
10868 | * into multiplications: | |
10869 | */ | |
10870 | const u32 sched_prio_to_wmult[40] = { | |
10871 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
10872 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
10873 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
10874 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
10875 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
10876 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
10877 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
10878 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
10879 | }; | |
14a7405b | 10880 | |
9d246053 PA |
10881 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
10882 | { | |
10883 | trace_sched_update_nr_running_tp(rq, count); | |
10884 | } |