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 | */ |
e66f6481 IM |
9 | #include <linux/highmem.h> |
10 | #include <linux/hrtimer_api.h> | |
11 | #include <linux/ktime_api.h> | |
12 | #include <linux/sched/signal.h> | |
13 | #include <linux/syscalls_api.h> | |
14 | #include <linux/debug_locks.h> | |
15 | #include <linux/prefetch.h> | |
16 | #include <linux/capability.h> | |
17 | #include <linux/pgtable_api.h> | |
18 | #include <linux/wait_bit.h> | |
19 | #include <linux/jiffies.h> | |
20 | #include <linux/spinlock_api.h> | |
21 | #include <linux/cpumask_api.h> | |
22 | #include <linux/lockdep_api.h> | |
23 | #include <linux/hardirq.h> | |
24 | #include <linux/softirq.h> | |
25 | #include <linux/refcount_api.h> | |
26 | #include <linux/topology.h> | |
27 | #include <linux/sched/clock.h> | |
28 | #include <linux/sched/cond_resched.h> | |
d664e399 | 29 | #include <linux/sched/cputime.h> |
e66f6481 | 30 | #include <linux/sched/debug.h> |
d664e399 TG |
31 | #include <linux/sched/hotplug.h> |
32 | #include <linux/sched/init.h> | |
e66f6481 IM |
33 | #include <linux/sched/isolation.h> |
34 | #include <linux/sched/loadavg.h> | |
35 | #include <linux/sched/mm.h> | |
36 | #include <linux/sched/nohz.h> | |
37 | #include <linux/sched/rseq_api.h> | |
38 | #include <linux/sched/rt.h> | |
1da177e4 | 39 | |
6a5850d1 | 40 | #include <linux/blkdev.h> |
e66f6481 IM |
41 | #include <linux/context_tracking.h> |
42 | #include <linux/cpuset.h> | |
43 | #include <linux/delayacct.h> | |
44 | #include <linux/init_task.h> | |
45 | #include <linux/interrupt.h> | |
46 | #include <linux/ioprio.h> | |
47 | #include <linux/kallsyms.h> | |
0ed557aa | 48 | #include <linux/kcov.h> |
e66f6481 IM |
49 | #include <linux/kprobes.h> |
50 | #include <linux/llist_api.h> | |
51 | #include <linux/mmu_context.h> | |
52 | #include <linux/mmzone.h> | |
53 | #include <linux/mutex_api.h> | |
54 | #include <linux/nmi.h> | |
55 | #include <linux/nospec.h> | |
56 | #include <linux/perf_event_api.h> | |
57 | #include <linux/profile.h> | |
58 | #include <linux/psi.h> | |
59 | #include <linux/rcuwait_api.h> | |
60 | #include <linux/sched/wake_q.h> | |
d08b9f0c | 61 | #include <linux/scs.h> |
e66f6481 IM |
62 | #include <linux/slab.h> |
63 | #include <linux/syscalls.h> | |
64 | #include <linux/vtime.h> | |
65 | #include <linux/wait_api.h> | |
66 | #include <linux/workqueue_api.h> | |
67 | ||
68 | #ifdef CONFIG_PREEMPT_DYNAMIC | |
a7b2553b IM |
69 | # ifdef CONFIG_GENERIC_ENTRY |
70 | # include <linux/entry-common.h> | |
71 | # endif | |
e66f6481 IM |
72 | #endif |
73 | ||
74 | #include <uapi/linux/sched/types.h> | |
0ed557aa | 75 | |
bc1cca97 | 76 | #include <asm/irq_regs.h> |
96f951ed | 77 | #include <asm/switch_to.h> |
5517d86b | 78 | #include <asm/tlb.h> |
1da177e4 | 79 | |
9d246053 | 80 | #define CREATE_TRACE_POINTS |
e66f6481 | 81 | #include <linux/sched/rseq_api.h> |
9d246053 PA |
82 | #include <trace/events/sched.h> |
83 | #undef CREATE_TRACE_POINTS | |
84 | ||
325ea10c | 85 | #include "sched.h" |
b9e9c6ca IM |
86 | #include "stats.h" |
87 | #include "autogroup.h" | |
6e0534f2 | 88 | |
e66f6481 | 89 | #include "autogroup.h" |
91c27493 | 90 | #include "pelt.h" |
1f8db415 | 91 | #include "smp.h" |
e66f6481 | 92 | #include "stats.h" |
1da177e4 | 93 | |
ea138446 | 94 | #include "../workqueue_internal.h" |
ed29b0b4 | 95 | #include "../../io_uring/io-wq.h" |
29d5e047 | 96 | #include "../smpboot.h" |
91c27493 | 97 | |
a056a5be QY |
98 | /* |
99 | * Export tracepoints that act as a bare tracehook (ie: have no trace event | |
100 | * associated with them) to allow external modules to probe them. | |
101 | */ | |
102 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_cfs_tp); | |
103 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp); | |
104 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp); | |
105 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp); | |
106 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp); | |
77cf151b | 107 | EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_thermal_tp); |
51cf18c9 | 108 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp); |
a056a5be | 109 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp); |
4581bea8 VD |
110 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp); |
111 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); | |
9d246053 | 112 | EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); |
a056a5be | 113 | |
029632fb | 114 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
dc61b1d6 | 115 | |
a73f863a | 116 | #ifdef CONFIG_SCHED_DEBUG |
bf5c91ba IM |
117 | /* |
118 | * Debugging: various feature bits | |
765cc3a4 PB |
119 | * |
120 | * If SCHED_DEBUG is disabled, each compilation unit has its own copy of | |
121 | * sysctl_sched_features, defined in sched.h, to allow constants propagation | |
122 | * at compile time and compiler optimization based on features default. | |
bf5c91ba | 123 | */ |
f00b45c1 PZ |
124 | #define SCHED_FEAT(name, enabled) \ |
125 | (1UL << __SCHED_FEAT_##name) * enabled | | |
bf5c91ba | 126 | const_debug unsigned int sysctl_sched_features = |
391e43da | 127 | #include "features.h" |
f00b45c1 | 128 | 0; |
f00b45c1 | 129 | #undef SCHED_FEAT |
c006fac5 PT |
130 | |
131 | /* | |
132 | * Print a warning if need_resched is set for the given duration (if | |
133 | * LATENCY_WARN is enabled). | |
134 | * | |
135 | * If sysctl_resched_latency_warn_once is set, only one warning will be shown | |
136 | * per boot. | |
137 | */ | |
138 | __read_mostly int sysctl_resched_latency_warn_ms = 100; | |
139 | __read_mostly int sysctl_resched_latency_warn_once = 1; | |
140 | #endif /* CONFIG_SCHED_DEBUG */ | |
f00b45c1 | 141 | |
b82d9fdd PZ |
142 | /* |
143 | * Number of tasks to iterate in a single balance run. | |
144 | * Limited because this is done with IRQs disabled. | |
145 | */ | |
c59862f8 | 146 | const_debug unsigned int sysctl_sched_nr_migrate = SCHED_NR_MIGRATE_BREAK; |
b82d9fdd | 147 | |
029632fb | 148 | __read_mostly int scheduler_running; |
6892b75e | 149 | |
9edeaea1 PZ |
150 | #ifdef CONFIG_SCHED_CORE |
151 | ||
152 | DEFINE_STATIC_KEY_FALSE(__sched_core_enabled); | |
153 | ||
8a311c74 PZ |
154 | /* kernel prio, less is more */ |
155 | static inline int __task_prio(struct task_struct *p) | |
156 | { | |
157 | if (p->sched_class == &stop_sched_class) /* trumps deadline */ | |
158 | return -2; | |
159 | ||
160 | if (rt_prio(p->prio)) /* includes deadline */ | |
161 | return p->prio; /* [-1, 99] */ | |
162 | ||
163 | if (p->sched_class == &idle_sched_class) | |
164 | return MAX_RT_PRIO + NICE_WIDTH; /* 140 */ | |
165 | ||
166 | return MAX_RT_PRIO + MAX_NICE; /* 120, squash fair */ | |
167 | } | |
168 | ||
169 | /* | |
170 | * l(a,b) | |
171 | * le(a,b) := !l(b,a) | |
172 | * g(a,b) := l(b,a) | |
173 | * ge(a,b) := !l(a,b) | |
174 | */ | |
175 | ||
176 | /* real prio, less is less */ | |
c6047c2e | 177 | static inline bool prio_less(struct task_struct *a, struct task_struct *b, bool in_fi) |
8a311c74 PZ |
178 | { |
179 | ||
180 | int pa = __task_prio(a), pb = __task_prio(b); | |
181 | ||
182 | if (-pa < -pb) | |
183 | return true; | |
184 | ||
185 | if (-pb < -pa) | |
186 | return false; | |
187 | ||
188 | if (pa == -1) /* dl_prio() doesn't work because of stop_class above */ | |
189 | return !dl_time_before(a->dl.deadline, b->dl.deadline); | |
190 | ||
c6047c2e JFG |
191 | if (pa == MAX_RT_PRIO + MAX_NICE) /* fair */ |
192 | return cfs_prio_less(a, b, in_fi); | |
8a311c74 PZ |
193 | |
194 | return false; | |
195 | } | |
196 | ||
197 | static inline bool __sched_core_less(struct task_struct *a, struct task_struct *b) | |
198 | { | |
199 | if (a->core_cookie < b->core_cookie) | |
200 | return true; | |
201 | ||
202 | if (a->core_cookie > b->core_cookie) | |
203 | return false; | |
204 | ||
205 | /* flip prio, so high prio is leftmost */ | |
4feee7d1 | 206 | if (prio_less(b, a, !!task_rq(a)->core->core_forceidle_count)) |
8a311c74 PZ |
207 | return true; |
208 | ||
209 | return false; | |
210 | } | |
211 | ||
212 | #define __node_2_sc(node) rb_entry((node), struct task_struct, core_node) | |
213 | ||
214 | static inline bool rb_sched_core_less(struct rb_node *a, const struct rb_node *b) | |
215 | { | |
216 | return __sched_core_less(__node_2_sc(a), __node_2_sc(b)); | |
217 | } | |
218 | ||
219 | static inline int rb_sched_core_cmp(const void *key, const struct rb_node *node) | |
220 | { | |
221 | const struct task_struct *p = __node_2_sc(node); | |
222 | unsigned long cookie = (unsigned long)key; | |
223 | ||
224 | if (cookie < p->core_cookie) | |
225 | return -1; | |
226 | ||
227 | if (cookie > p->core_cookie) | |
228 | return 1; | |
229 | ||
230 | return 0; | |
231 | } | |
232 | ||
6e33cad0 | 233 | void sched_core_enqueue(struct rq *rq, struct task_struct *p) |
8a311c74 PZ |
234 | { |
235 | rq->core->core_task_seq++; | |
236 | ||
237 | if (!p->core_cookie) | |
238 | return; | |
239 | ||
240 | rb_add(&p->core_node, &rq->core_tree, rb_sched_core_less); | |
241 | } | |
242 | ||
4feee7d1 | 243 | void sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags) |
8a311c74 PZ |
244 | { |
245 | rq->core->core_task_seq++; | |
246 | ||
4feee7d1 JD |
247 | if (sched_core_enqueued(p)) { |
248 | rb_erase(&p->core_node, &rq->core_tree); | |
249 | RB_CLEAR_NODE(&p->core_node); | |
250 | } | |
8a311c74 | 251 | |
4feee7d1 JD |
252 | /* |
253 | * Migrating the last task off the cpu, with the cpu in forced idle | |
254 | * state. Reschedule to create an accounting edge for forced idle, | |
255 | * and re-examine whether the core is still in forced idle state. | |
256 | */ | |
257 | if (!(flags & DEQUEUE_SAVE) && rq->nr_running == 1 && | |
258 | rq->core->core_forceidle_count && rq->curr == rq->idle) | |
259 | resched_curr(rq); | |
8a311c74 PZ |
260 | } |
261 | ||
262 | /* | |
263 | * Find left-most (aka, highest priority) task matching @cookie. | |
264 | */ | |
265 | static struct task_struct *sched_core_find(struct rq *rq, unsigned long cookie) | |
266 | { | |
267 | struct rb_node *node; | |
268 | ||
269 | node = rb_find_first((void *)cookie, &rq->core_tree, rb_sched_core_cmp); | |
270 | /* | |
271 | * The idle task always matches any cookie! | |
272 | */ | |
273 | if (!node) | |
274 | return idle_sched_class.pick_task(rq); | |
275 | ||
276 | return __node_2_sc(node); | |
277 | } | |
278 | ||
d2dfa17b PZ |
279 | static struct task_struct *sched_core_next(struct task_struct *p, unsigned long cookie) |
280 | { | |
281 | struct rb_node *node = &p->core_node; | |
282 | ||
283 | node = rb_next(node); | |
284 | if (!node) | |
285 | return NULL; | |
286 | ||
287 | p = container_of(node, struct task_struct, core_node); | |
288 | if (p->core_cookie != cookie) | |
289 | return NULL; | |
290 | ||
291 | return p; | |
292 | } | |
293 | ||
9edeaea1 PZ |
294 | /* |
295 | * Magic required such that: | |
296 | * | |
297 | * raw_spin_rq_lock(rq); | |
298 | * ... | |
299 | * raw_spin_rq_unlock(rq); | |
300 | * | |
301 | * ends up locking and unlocking the _same_ lock, and all CPUs | |
302 | * always agree on what rq has what lock. | |
303 | * | |
304 | * XXX entirely possible to selectively enable cores, don't bother for now. | |
305 | */ | |
306 | ||
307 | static DEFINE_MUTEX(sched_core_mutex); | |
875feb41 | 308 | static atomic_t sched_core_count; |
9edeaea1 PZ |
309 | static struct cpumask sched_core_mask; |
310 | ||
3c474b32 PZ |
311 | static void sched_core_lock(int cpu, unsigned long *flags) |
312 | { | |
313 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
314 | int t, i = 0; | |
315 | ||
316 | local_irq_save(*flags); | |
317 | for_each_cpu(t, smt_mask) | |
318 | raw_spin_lock_nested(&cpu_rq(t)->__lock, i++); | |
319 | } | |
320 | ||
321 | static void sched_core_unlock(int cpu, unsigned long *flags) | |
322 | { | |
323 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
324 | int t; | |
325 | ||
326 | for_each_cpu(t, smt_mask) | |
327 | raw_spin_unlock(&cpu_rq(t)->__lock); | |
328 | local_irq_restore(*flags); | |
329 | } | |
330 | ||
9edeaea1 PZ |
331 | static void __sched_core_flip(bool enabled) |
332 | { | |
3c474b32 PZ |
333 | unsigned long flags; |
334 | int cpu, t; | |
9edeaea1 PZ |
335 | |
336 | cpus_read_lock(); | |
337 | ||
338 | /* | |
339 | * Toggle the online cores, one by one. | |
340 | */ | |
341 | cpumask_copy(&sched_core_mask, cpu_online_mask); | |
342 | for_each_cpu(cpu, &sched_core_mask) { | |
343 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
344 | ||
3c474b32 | 345 | sched_core_lock(cpu, &flags); |
9edeaea1 PZ |
346 | |
347 | for_each_cpu(t, smt_mask) | |
348 | cpu_rq(t)->core_enabled = enabled; | |
349 | ||
4feee7d1 JD |
350 | cpu_rq(cpu)->core->core_forceidle_start = 0; |
351 | ||
3c474b32 | 352 | sched_core_unlock(cpu, &flags); |
9edeaea1 PZ |
353 | |
354 | cpumask_andnot(&sched_core_mask, &sched_core_mask, smt_mask); | |
355 | } | |
356 | ||
357 | /* | |
358 | * Toggle the offline CPUs. | |
359 | */ | |
585463f0 | 360 | for_each_cpu_andnot(cpu, cpu_possible_mask, cpu_online_mask) |
9edeaea1 PZ |
361 | cpu_rq(cpu)->core_enabled = enabled; |
362 | ||
363 | cpus_read_unlock(); | |
364 | } | |
365 | ||
8a311c74 | 366 | static void sched_core_assert_empty(void) |
9edeaea1 | 367 | { |
8a311c74 | 368 | int cpu; |
9edeaea1 | 369 | |
8a311c74 PZ |
370 | for_each_possible_cpu(cpu) |
371 | WARN_ON_ONCE(!RB_EMPTY_ROOT(&cpu_rq(cpu)->core_tree)); | |
372 | } | |
373 | ||
374 | static void __sched_core_enable(void) | |
375 | { | |
9edeaea1 PZ |
376 | static_branch_enable(&__sched_core_enabled); |
377 | /* | |
378 | * Ensure all previous instances of raw_spin_rq_*lock() have finished | |
379 | * and future ones will observe !sched_core_disabled(). | |
380 | */ | |
381 | synchronize_rcu(); | |
382 | __sched_core_flip(true); | |
8a311c74 | 383 | sched_core_assert_empty(); |
9edeaea1 PZ |
384 | } |
385 | ||
386 | static void __sched_core_disable(void) | |
387 | { | |
8a311c74 | 388 | sched_core_assert_empty(); |
9edeaea1 PZ |
389 | __sched_core_flip(false); |
390 | static_branch_disable(&__sched_core_enabled); | |
391 | } | |
392 | ||
393 | void sched_core_get(void) | |
394 | { | |
875feb41 PZ |
395 | if (atomic_inc_not_zero(&sched_core_count)) |
396 | return; | |
397 | ||
9edeaea1 | 398 | mutex_lock(&sched_core_mutex); |
875feb41 | 399 | if (!atomic_read(&sched_core_count)) |
9edeaea1 | 400 | __sched_core_enable(); |
875feb41 PZ |
401 | |
402 | smp_mb__before_atomic(); | |
403 | atomic_inc(&sched_core_count); | |
9edeaea1 PZ |
404 | mutex_unlock(&sched_core_mutex); |
405 | } | |
406 | ||
875feb41 | 407 | static void __sched_core_put(struct work_struct *work) |
9edeaea1 | 408 | { |
875feb41 | 409 | if (atomic_dec_and_mutex_lock(&sched_core_count, &sched_core_mutex)) { |
9edeaea1 | 410 | __sched_core_disable(); |
875feb41 PZ |
411 | mutex_unlock(&sched_core_mutex); |
412 | } | |
413 | } | |
414 | ||
415 | void sched_core_put(void) | |
416 | { | |
417 | static DECLARE_WORK(_work, __sched_core_put); | |
418 | ||
419 | /* | |
420 | * "There can be only one" | |
421 | * | |
422 | * Either this is the last one, or we don't actually need to do any | |
423 | * 'work'. If it is the last *again*, we rely on | |
424 | * WORK_STRUCT_PENDING_BIT. | |
425 | */ | |
426 | if (!atomic_add_unless(&sched_core_count, -1, 1)) | |
427 | schedule_work(&_work); | |
9edeaea1 PZ |
428 | } |
429 | ||
8a311c74 PZ |
430 | #else /* !CONFIG_SCHED_CORE */ |
431 | ||
432 | static inline void sched_core_enqueue(struct rq *rq, struct task_struct *p) { } | |
4feee7d1 JD |
433 | static inline void |
434 | sched_core_dequeue(struct rq *rq, struct task_struct *p, int flags) { } | |
8a311c74 | 435 | |
9edeaea1 PZ |
436 | #endif /* CONFIG_SCHED_CORE */ |
437 | ||
58877d34 PZ |
438 | /* |
439 | * Serialization rules: | |
440 | * | |
441 | * Lock order: | |
442 | * | |
443 | * p->pi_lock | |
444 | * rq->lock | |
445 | * hrtimer_cpu_base->lock (hrtimer_start() for bandwidth controls) | |
446 | * | |
447 | * rq1->lock | |
448 | * rq2->lock where: rq1 < rq2 | |
449 | * | |
450 | * Regular state: | |
451 | * | |
452 | * Normal scheduling state is serialized by rq->lock. __schedule() takes the | |
453 | * local CPU's rq->lock, it optionally removes the task from the runqueue and | |
b19a888c | 454 | * always looks at the local rq data structures to find the most eligible task |
58877d34 PZ |
455 | * to run next. |
456 | * | |
457 | * Task enqueue is also under rq->lock, possibly taken from another CPU. | |
458 | * Wakeups from another LLC domain might use an IPI to transfer the enqueue to | |
459 | * the local CPU to avoid bouncing the runqueue state around [ see | |
460 | * ttwu_queue_wakelist() ] | |
461 | * | |
462 | * Task wakeup, specifically wakeups that involve migration, are horribly | |
463 | * complicated to avoid having to take two rq->locks. | |
464 | * | |
465 | * Special state: | |
466 | * | |
467 | * System-calls and anything external will use task_rq_lock() which acquires | |
468 | * both p->pi_lock and rq->lock. As a consequence the state they change is | |
469 | * stable while holding either lock: | |
470 | * | |
471 | * - sched_setaffinity()/ | |
472 | * set_cpus_allowed_ptr(): p->cpus_ptr, p->nr_cpus_allowed | |
473 | * - set_user_nice(): p->se.load, p->*prio | |
474 | * - __sched_setscheduler(): p->sched_class, p->policy, p->*prio, | |
475 | * p->se.load, p->rt_priority, | |
476 | * p->dl.dl_{runtime, deadline, period, flags, bw, density} | |
477 | * - sched_setnuma(): p->numa_preferred_nid | |
39c42611 | 478 | * - sched_move_task(): p->sched_task_group |
58877d34 PZ |
479 | * - uclamp_update_active() p->uclamp* |
480 | * | |
481 | * p->state <- TASK_*: | |
482 | * | |
483 | * is changed locklessly using set_current_state(), __set_current_state() or | |
484 | * set_special_state(), see their respective comments, or by | |
485 | * try_to_wake_up(). This latter uses p->pi_lock to serialize against | |
486 | * concurrent self. | |
487 | * | |
488 | * p->on_rq <- { 0, 1 = TASK_ON_RQ_QUEUED, 2 = TASK_ON_RQ_MIGRATING }: | |
489 | * | |
490 | * is set by activate_task() and cleared by deactivate_task(), under | |
491 | * rq->lock. Non-zero indicates the task is runnable, the special | |
492 | * ON_RQ_MIGRATING state is used for migration without holding both | |
493 | * rq->locks. It indicates task_cpu() is not stable, see task_rq_lock(). | |
494 | * | |
495 | * p->on_cpu <- { 0, 1 }: | |
496 | * | |
497 | * is set by prepare_task() and cleared by finish_task() such that it will be | |
498 | * set before p is scheduled-in and cleared after p is scheduled-out, both | |
499 | * under rq->lock. Non-zero indicates the task is running on its CPU. | |
500 | * | |
501 | * [ The astute reader will observe that it is possible for two tasks on one | |
502 | * CPU to have ->on_cpu = 1 at the same time. ] | |
503 | * | |
504 | * task_cpu(p): is changed by set_task_cpu(), the rules are: | |
505 | * | |
506 | * - Don't call set_task_cpu() on a blocked task: | |
507 | * | |
508 | * We don't care what CPU we're not running on, this simplifies hotplug, | |
509 | * the CPU assignment of blocked tasks isn't required to be valid. | |
510 | * | |
511 | * - for try_to_wake_up(), called under p->pi_lock: | |
512 | * | |
513 | * This allows try_to_wake_up() to only take one rq->lock, see its comment. | |
514 | * | |
515 | * - for migration called under rq->lock: | |
516 | * [ see task_on_rq_migrating() in task_rq_lock() ] | |
517 | * | |
518 | * o move_queued_task() | |
519 | * o detach_task() | |
520 | * | |
521 | * - for migration called under double_rq_lock(): | |
522 | * | |
523 | * o __migrate_swap_task() | |
524 | * o push_rt_task() / pull_rt_task() | |
525 | * o push_dl_task() / pull_dl_task() | |
526 | * o dl_task_offline_migration() | |
527 | * | |
528 | */ | |
529 | ||
39d371b7 PZ |
530 | void raw_spin_rq_lock_nested(struct rq *rq, int subclass) |
531 | { | |
d66f1b06 PZ |
532 | raw_spinlock_t *lock; |
533 | ||
9edeaea1 PZ |
534 | /* Matches synchronize_rcu() in __sched_core_enable() */ |
535 | preempt_disable(); | |
d66f1b06 PZ |
536 | if (sched_core_disabled()) { |
537 | raw_spin_lock_nested(&rq->__lock, subclass); | |
9edeaea1 PZ |
538 | /* preempt_count *MUST* be > 1 */ |
539 | preempt_enable_no_resched(); | |
d66f1b06 PZ |
540 | return; |
541 | } | |
542 | ||
543 | for (;;) { | |
9ef7e7e3 | 544 | lock = __rq_lockp(rq); |
d66f1b06 | 545 | raw_spin_lock_nested(lock, subclass); |
9ef7e7e3 | 546 | if (likely(lock == __rq_lockp(rq))) { |
9edeaea1 PZ |
547 | /* preempt_count *MUST* be > 1 */ |
548 | preempt_enable_no_resched(); | |
d66f1b06 | 549 | return; |
9edeaea1 | 550 | } |
d66f1b06 PZ |
551 | raw_spin_unlock(lock); |
552 | } | |
39d371b7 PZ |
553 | } |
554 | ||
555 | bool raw_spin_rq_trylock(struct rq *rq) | |
556 | { | |
d66f1b06 PZ |
557 | raw_spinlock_t *lock; |
558 | bool ret; | |
559 | ||
9edeaea1 PZ |
560 | /* Matches synchronize_rcu() in __sched_core_enable() */ |
561 | preempt_disable(); | |
562 | if (sched_core_disabled()) { | |
563 | ret = raw_spin_trylock(&rq->__lock); | |
564 | preempt_enable(); | |
565 | return ret; | |
566 | } | |
d66f1b06 PZ |
567 | |
568 | for (;;) { | |
9ef7e7e3 | 569 | lock = __rq_lockp(rq); |
d66f1b06 | 570 | ret = raw_spin_trylock(lock); |
9ef7e7e3 | 571 | if (!ret || (likely(lock == __rq_lockp(rq)))) { |
9edeaea1 | 572 | preempt_enable(); |
d66f1b06 | 573 | return ret; |
9edeaea1 | 574 | } |
d66f1b06 PZ |
575 | raw_spin_unlock(lock); |
576 | } | |
39d371b7 PZ |
577 | } |
578 | ||
579 | void raw_spin_rq_unlock(struct rq *rq) | |
580 | { | |
581 | raw_spin_unlock(rq_lockp(rq)); | |
582 | } | |
583 | ||
d66f1b06 PZ |
584 | #ifdef CONFIG_SMP |
585 | /* | |
586 | * double_rq_lock - safely lock two runqueues | |
587 | */ | |
588 | void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
589 | { | |
590 | lockdep_assert_irqs_disabled(); | |
591 | ||
592 | if (rq_order_less(rq2, rq1)) | |
593 | swap(rq1, rq2); | |
594 | ||
595 | raw_spin_rq_lock(rq1); | |
2679a837 HJ |
596 | if (__rq_lockp(rq1) != __rq_lockp(rq2)) |
597 | raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING); | |
d66f1b06 | 598 | |
2679a837 | 599 | double_rq_clock_clear_update(rq1, rq2); |
d66f1b06 PZ |
600 | } |
601 | #endif | |
602 | ||
3e71a462 PZ |
603 | /* |
604 | * __task_rq_lock - lock the rq @p resides on. | |
605 | */ | |
eb580751 | 606 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
607 | __acquires(rq->lock) |
608 | { | |
609 | struct rq *rq; | |
610 | ||
611 | lockdep_assert_held(&p->pi_lock); | |
612 | ||
613 | for (;;) { | |
614 | rq = task_rq(p); | |
5cb9eaa3 | 615 | raw_spin_rq_lock(rq); |
3e71a462 | 616 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
d8ac8971 | 617 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
618 | return rq; |
619 | } | |
5cb9eaa3 | 620 | raw_spin_rq_unlock(rq); |
3e71a462 PZ |
621 | |
622 | while (unlikely(task_on_rq_migrating(p))) | |
623 | cpu_relax(); | |
624 | } | |
625 | } | |
626 | ||
627 | /* | |
628 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. | |
629 | */ | |
eb580751 | 630 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
3e71a462 PZ |
631 | __acquires(p->pi_lock) |
632 | __acquires(rq->lock) | |
633 | { | |
634 | struct rq *rq; | |
635 | ||
636 | for (;;) { | |
eb580751 | 637 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
3e71a462 | 638 | rq = task_rq(p); |
5cb9eaa3 | 639 | raw_spin_rq_lock(rq); |
3e71a462 PZ |
640 | /* |
641 | * move_queued_task() task_rq_lock() | |
642 | * | |
643 | * ACQUIRE (rq->lock) | |
644 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() | |
645 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); | |
646 | * [S] ->cpu = new_cpu [L] task_rq() | |
647 | * [L] ->on_rq | |
648 | * RELEASE (rq->lock) | |
649 | * | |
c546951d | 650 | * If we observe the old CPU in task_rq_lock(), the acquire of |
3e71a462 PZ |
651 | * the old rq->lock will fully serialize against the stores. |
652 | * | |
c546951d AP |
653 | * If we observe the new CPU in task_rq_lock(), the address |
654 | * dependency headed by '[L] rq = task_rq()' and the acquire | |
655 | * will pair with the WMB to ensure we then also see migrating. | |
3e71a462 PZ |
656 | */ |
657 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { | |
d8ac8971 | 658 | rq_pin_lock(rq, rf); |
3e71a462 PZ |
659 | return rq; |
660 | } | |
5cb9eaa3 | 661 | raw_spin_rq_unlock(rq); |
eb580751 | 662 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
3e71a462 PZ |
663 | |
664 | while (unlikely(task_on_rq_migrating(p))) | |
665 | cpu_relax(); | |
666 | } | |
667 | } | |
668 | ||
535b9552 IM |
669 | /* |
670 | * RQ-clock updating methods: | |
671 | */ | |
672 | ||
673 | static void update_rq_clock_task(struct rq *rq, s64 delta) | |
674 | { | |
675 | /* | |
676 | * In theory, the compile should just see 0 here, and optimize out the call | |
677 | * to sched_rt_avg_update. But I don't trust it... | |
678 | */ | |
11d4afd4 VG |
679 | s64 __maybe_unused steal = 0, irq_delta = 0; |
680 | ||
535b9552 IM |
681 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
682 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; | |
683 | ||
684 | /* | |
685 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
686 | * this case when a previous update_rq_clock() happened inside a | |
687 | * {soft,}irq region. | |
688 | * | |
689 | * When this happens, we stop ->clock_task and only update the | |
690 | * prev_irq_time stamp to account for the part that fit, so that a next | |
691 | * update will consume the rest. This ensures ->clock_task is | |
692 | * monotonic. | |
693 | * | |
694 | * It does however cause some slight miss-attribution of {soft,}irq | |
695 | * time, a more accurate solution would be to update the irq_time using | |
696 | * the current rq->clock timestamp, except that would require using | |
697 | * atomic ops. | |
698 | */ | |
699 | if (irq_delta > delta) | |
700 | irq_delta = delta; | |
701 | ||
702 | rq->prev_irq_time += irq_delta; | |
703 | delta -= irq_delta; | |
52b1364b | 704 | psi_account_irqtime(rq->curr, irq_delta); |
535b9552 IM |
705 | #endif |
706 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING | |
707 | if (static_key_false((¶virt_steal_rq_enabled))) { | |
708 | steal = paravirt_steal_clock(cpu_of(rq)); | |
709 | steal -= rq->prev_steal_time_rq; | |
710 | ||
711 | if (unlikely(steal > delta)) | |
712 | steal = delta; | |
713 | ||
714 | rq->prev_steal_time_rq += steal; | |
715 | delta -= steal; | |
716 | } | |
717 | #endif | |
718 | ||
719 | rq->clock_task += delta; | |
720 | ||
11d4afd4 | 721 | #ifdef CONFIG_HAVE_SCHED_AVG_IRQ |
535b9552 | 722 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
91c27493 | 723 | update_irq_load_avg(rq, irq_delta + steal); |
535b9552 | 724 | #endif |
23127296 | 725 | update_rq_clock_pelt(rq, delta); |
535b9552 IM |
726 | } |
727 | ||
728 | void update_rq_clock(struct rq *rq) | |
729 | { | |
730 | s64 delta; | |
731 | ||
5cb9eaa3 | 732 | lockdep_assert_rq_held(rq); |
535b9552 IM |
733 | |
734 | if (rq->clock_update_flags & RQCF_ACT_SKIP) | |
735 | return; | |
736 | ||
737 | #ifdef CONFIG_SCHED_DEBUG | |
26ae58d2 PZ |
738 | if (sched_feat(WARN_DOUBLE_CLOCK)) |
739 | SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED); | |
535b9552 IM |
740 | rq->clock_update_flags |= RQCF_UPDATED; |
741 | #endif | |
26ae58d2 | 742 | |
535b9552 IM |
743 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
744 | if (delta < 0) | |
745 | return; | |
746 | rq->clock += delta; | |
747 | update_rq_clock_task(rq, delta); | |
748 | } | |
749 | ||
8f4d37ec PZ |
750 | #ifdef CONFIG_SCHED_HRTICK |
751 | /* | |
752 | * Use HR-timers to deliver accurate preemption points. | |
8f4d37ec | 753 | */ |
8f4d37ec | 754 | |
8f4d37ec PZ |
755 | static void hrtick_clear(struct rq *rq) |
756 | { | |
757 | if (hrtimer_active(&rq->hrtick_timer)) | |
758 | hrtimer_cancel(&rq->hrtick_timer); | |
759 | } | |
760 | ||
8f4d37ec PZ |
761 | /* |
762 | * High-resolution timer tick. | |
763 | * Runs from hardirq context with interrupts disabled. | |
764 | */ | |
765 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
766 | { | |
767 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
8a8c69c3 | 768 | struct rq_flags rf; |
8f4d37ec PZ |
769 | |
770 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
771 | ||
8a8c69c3 | 772 | rq_lock(rq, &rf); |
3e51f33f | 773 | update_rq_clock(rq); |
8f4d37ec | 774 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
8a8c69c3 | 775 | rq_unlock(rq, &rf); |
8f4d37ec PZ |
776 | |
777 | return HRTIMER_NORESTART; | |
778 | } | |
779 | ||
95e904c7 | 780 | #ifdef CONFIG_SMP |
971ee28c | 781 | |
4961b6e1 | 782 | static void __hrtick_restart(struct rq *rq) |
971ee28c PZ |
783 | { |
784 | struct hrtimer *timer = &rq->hrtick_timer; | |
156ec6f4 | 785 | ktime_t time = rq->hrtick_time; |
971ee28c | 786 | |
156ec6f4 | 787 | hrtimer_start(timer, time, HRTIMER_MODE_ABS_PINNED_HARD); |
971ee28c PZ |
788 | } |
789 | ||
31656519 PZ |
790 | /* |
791 | * called from hardirq (IPI) context | |
792 | */ | |
793 | static void __hrtick_start(void *arg) | |
b328ca18 | 794 | { |
31656519 | 795 | struct rq *rq = arg; |
8a8c69c3 | 796 | struct rq_flags rf; |
b328ca18 | 797 | |
8a8c69c3 | 798 | rq_lock(rq, &rf); |
971ee28c | 799 | __hrtick_restart(rq); |
8a8c69c3 | 800 | rq_unlock(rq, &rf); |
b328ca18 PZ |
801 | } |
802 | ||
31656519 PZ |
803 | /* |
804 | * Called to set the hrtick timer state. | |
805 | * | |
806 | * called with rq->lock held and irqs disabled | |
807 | */ | |
029632fb | 808 | void hrtick_start(struct rq *rq, u64 delay) |
b328ca18 | 809 | { |
31656519 | 810 | struct hrtimer *timer = &rq->hrtick_timer; |
177ef2a6 | 811 | s64 delta; |
812 | ||
813 | /* | |
814 | * Don't schedule slices shorter than 10000ns, that just | |
815 | * doesn't make sense and can cause timer DoS. | |
816 | */ | |
817 | delta = max_t(s64, delay, 10000LL); | |
156ec6f4 | 818 | rq->hrtick_time = ktime_add_ns(timer->base->get_time(), delta); |
31656519 | 819 | |
fd3eafda | 820 | if (rq == this_rq()) |
971ee28c | 821 | __hrtick_restart(rq); |
fd3eafda | 822 | else |
c46fff2a | 823 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
b328ca18 PZ |
824 | } |
825 | ||
31656519 PZ |
826 | #else |
827 | /* | |
828 | * Called to set the hrtick timer state. | |
829 | * | |
830 | * called with rq->lock held and irqs disabled | |
831 | */ | |
029632fb | 832 | void hrtick_start(struct rq *rq, u64 delay) |
31656519 | 833 | { |
86893335 WL |
834 | /* |
835 | * Don't schedule slices shorter than 10000ns, that just | |
836 | * doesn't make sense. Rely on vruntime for fairness. | |
837 | */ | |
838 | delay = max_t(u64, delay, 10000LL); | |
4961b6e1 | 839 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
d5096aa6 | 840 | HRTIMER_MODE_REL_PINNED_HARD); |
31656519 | 841 | } |
90b5363a | 842 | |
31656519 | 843 | #endif /* CONFIG_SMP */ |
8f4d37ec | 844 | |
77a021be | 845 | static void hrtick_rq_init(struct rq *rq) |
8f4d37ec | 846 | { |
31656519 | 847 | #ifdef CONFIG_SMP |
545b8c8d | 848 | INIT_CSD(&rq->hrtick_csd, __hrtick_start, rq); |
31656519 | 849 | #endif |
d5096aa6 | 850 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD); |
31656519 | 851 | rq->hrtick_timer.function = hrtick; |
8f4d37ec | 852 | } |
006c75f1 | 853 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
854 | static inline void hrtick_clear(struct rq *rq) |
855 | { | |
856 | } | |
857 | ||
77a021be | 858 | static inline void hrtick_rq_init(struct rq *rq) |
8f4d37ec PZ |
859 | { |
860 | } | |
006c75f1 | 861 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 862 | |
5529578a FW |
863 | /* |
864 | * cmpxchg based fetch_or, macro so it works for different integer types | |
865 | */ | |
866 | #define fetch_or(ptr, mask) \ | |
867 | ({ \ | |
868 | typeof(ptr) _ptr = (ptr); \ | |
869 | typeof(mask) _mask = (mask); \ | |
c02d5546 | 870 | typeof(*_ptr) _val = *_ptr; \ |
5529578a | 871 | \ |
c02d5546 UB |
872 | do { \ |
873 | } while (!try_cmpxchg(_ptr, &_val, _val | _mask)); \ | |
874 | _val; \ | |
5529578a FW |
875 | }) |
876 | ||
e3baac47 | 877 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
fd99f91a PZ |
878 | /* |
879 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, | |
880 | * this avoids any races wrt polling state changes and thereby avoids | |
881 | * spurious IPIs. | |
882 | */ | |
c02d5546 | 883 | static inline bool set_nr_and_not_polling(struct task_struct *p) |
fd99f91a PZ |
884 | { |
885 | struct thread_info *ti = task_thread_info(p); | |
886 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); | |
887 | } | |
e3baac47 PZ |
888 | |
889 | /* | |
890 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. | |
891 | * | |
892 | * If this returns true, then the idle task promises to call | |
893 | * sched_ttwu_pending() and reschedule soon. | |
894 | */ | |
895 | static bool set_nr_if_polling(struct task_struct *p) | |
896 | { | |
897 | struct thread_info *ti = task_thread_info(p); | |
c02d5546 | 898 | typeof(ti->flags) val = READ_ONCE(ti->flags); |
e3baac47 PZ |
899 | |
900 | for (;;) { | |
901 | if (!(val & _TIF_POLLING_NRFLAG)) | |
902 | return false; | |
903 | if (val & _TIF_NEED_RESCHED) | |
904 | return true; | |
c02d5546 | 905 | if (try_cmpxchg(&ti->flags, &val, val | _TIF_NEED_RESCHED)) |
e3baac47 | 906 | break; |
e3baac47 PZ |
907 | } |
908 | return true; | |
909 | } | |
910 | ||
fd99f91a | 911 | #else |
c02d5546 | 912 | static inline bool set_nr_and_not_polling(struct task_struct *p) |
fd99f91a PZ |
913 | { |
914 | set_tsk_need_resched(p); | |
915 | return true; | |
916 | } | |
e3baac47 PZ |
917 | |
918 | #ifdef CONFIG_SMP | |
c02d5546 | 919 | static inline bool set_nr_if_polling(struct task_struct *p) |
e3baac47 PZ |
920 | { |
921 | return false; | |
922 | } | |
923 | #endif | |
fd99f91a PZ |
924 | #endif |
925 | ||
07879c6a | 926 | static bool __wake_q_add(struct wake_q_head *head, struct task_struct *task) |
76751049 PZ |
927 | { |
928 | struct wake_q_node *node = &task->wake_q; | |
929 | ||
930 | /* | |
931 | * Atomically grab the task, if ->wake_q is !nil already it means | |
b19a888c | 932 | * it's already queued (either by us or someone else) and will get the |
76751049 PZ |
933 | * wakeup due to that. |
934 | * | |
4c4e3731 PZ |
935 | * In order to ensure that a pending wakeup will observe our pending |
936 | * state, even in the failed case, an explicit smp_mb() must be used. | |
76751049 | 937 | */ |
4c4e3731 | 938 | smp_mb__before_atomic(); |
87ff19cb | 939 | if (unlikely(cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))) |
07879c6a | 940 | return false; |
76751049 PZ |
941 | |
942 | /* | |
943 | * The head is context local, there can be no concurrency. | |
944 | */ | |
945 | *head->lastp = node; | |
946 | head->lastp = &node->next; | |
07879c6a DB |
947 | return true; |
948 | } | |
949 | ||
950 | /** | |
951 | * wake_q_add() - queue a wakeup for 'later' waking. | |
952 | * @head: the wake_q_head to add @task to | |
953 | * @task: the task to queue for 'later' wakeup | |
954 | * | |
955 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
956 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
957 | * instantly. | |
958 | * | |
959 | * This function must be used as-if it were wake_up_process(); IOW the task | |
960 | * must be ready to be woken at this location. | |
961 | */ | |
962 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) | |
963 | { | |
964 | if (__wake_q_add(head, task)) | |
965 | get_task_struct(task); | |
966 | } | |
967 | ||
968 | /** | |
969 | * wake_q_add_safe() - safely queue a wakeup for 'later' waking. | |
970 | * @head: the wake_q_head to add @task to | |
971 | * @task: the task to queue for 'later' wakeup | |
972 | * | |
973 | * Queue a task for later wakeup, most likely by the wake_up_q() call in the | |
974 | * same context, _HOWEVER_ this is not guaranteed, the wakeup can come | |
975 | * instantly. | |
976 | * | |
977 | * This function must be used as-if it were wake_up_process(); IOW the task | |
978 | * must be ready to be woken at this location. | |
979 | * | |
980 | * This function is essentially a task-safe equivalent to wake_q_add(). Callers | |
981 | * that already hold reference to @task can call the 'safe' version and trust | |
982 | * wake_q to do the right thing depending whether or not the @task is already | |
983 | * queued for wakeup. | |
984 | */ | |
985 | void wake_q_add_safe(struct wake_q_head *head, struct task_struct *task) | |
986 | { | |
987 | if (!__wake_q_add(head, task)) | |
988 | put_task_struct(task); | |
76751049 PZ |
989 | } |
990 | ||
991 | void wake_up_q(struct wake_q_head *head) | |
992 | { | |
993 | struct wake_q_node *node = head->first; | |
994 | ||
995 | while (node != WAKE_Q_TAIL) { | |
996 | struct task_struct *task; | |
997 | ||
998 | task = container_of(node, struct task_struct, wake_q); | |
d1ccc66d | 999 | /* Task can safely be re-inserted now: */ |
76751049 PZ |
1000 | node = node->next; |
1001 | task->wake_q.next = NULL; | |
1002 | ||
1003 | /* | |
7696f991 AP |
1004 | * wake_up_process() executes a full barrier, which pairs with |
1005 | * the queueing in wake_q_add() so as not to miss wakeups. | |
76751049 PZ |
1006 | */ |
1007 | wake_up_process(task); | |
1008 | put_task_struct(task); | |
1009 | } | |
1010 | } | |
1011 | ||
c24d20db | 1012 | /* |
8875125e | 1013 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
c24d20db IM |
1014 | * |
1015 | * On UP this means the setting of the need_resched flag, on SMP it | |
1016 | * might also involve a cross-CPU call to trigger the scheduler on | |
1017 | * the target CPU. | |
1018 | */ | |
8875125e | 1019 | void resched_curr(struct rq *rq) |
c24d20db | 1020 | { |
8875125e | 1021 | struct task_struct *curr = rq->curr; |
c24d20db IM |
1022 | int cpu; |
1023 | ||
5cb9eaa3 | 1024 | lockdep_assert_rq_held(rq); |
c24d20db | 1025 | |
8875125e | 1026 | if (test_tsk_need_resched(curr)) |
c24d20db IM |
1027 | return; |
1028 | ||
8875125e | 1029 | cpu = cpu_of(rq); |
fd99f91a | 1030 | |
f27dde8d | 1031 | if (cpu == smp_processor_id()) { |
8875125e | 1032 | set_tsk_need_resched(curr); |
f27dde8d | 1033 | set_preempt_need_resched(); |
c24d20db | 1034 | return; |
f27dde8d | 1035 | } |
c24d20db | 1036 | |
8875125e | 1037 | if (set_nr_and_not_polling(curr)) |
c24d20db | 1038 | smp_send_reschedule(cpu); |
dfc68f29 AL |
1039 | else |
1040 | trace_sched_wake_idle_without_ipi(cpu); | |
c24d20db IM |
1041 | } |
1042 | ||
029632fb | 1043 | void resched_cpu(int cpu) |
c24d20db IM |
1044 | { |
1045 | struct rq *rq = cpu_rq(cpu); | |
1046 | unsigned long flags; | |
1047 | ||
5cb9eaa3 | 1048 | raw_spin_rq_lock_irqsave(rq, flags); |
a0982dfa PM |
1049 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
1050 | resched_curr(rq); | |
5cb9eaa3 | 1051 | raw_spin_rq_unlock_irqrestore(rq, flags); |
c24d20db | 1052 | } |
06d8308c | 1053 | |
b021fe3e | 1054 | #ifdef CONFIG_SMP |
3451d024 | 1055 | #ifdef CONFIG_NO_HZ_COMMON |
83cd4fe2 | 1056 | /* |
d1ccc66d IM |
1057 | * In the semi idle case, use the nearest busy CPU for migrating timers |
1058 | * from an idle CPU. This is good for power-savings. | |
83cd4fe2 VP |
1059 | * |
1060 | * We don't do similar optimization for completely idle system, as | |
d1ccc66d IM |
1061 | * selecting an idle CPU will add more delays to the timers than intended |
1062 | * (as that CPU's timer base may not be uptodate wrt jiffies etc). | |
83cd4fe2 | 1063 | */ |
bc7a34b8 | 1064 | int get_nohz_timer_target(void) |
83cd4fe2 | 1065 | { |
e938b9c9 | 1066 | int i, cpu = smp_processor_id(), default_cpu = -1; |
83cd4fe2 | 1067 | struct sched_domain *sd; |
031e3bd8 | 1068 | const struct cpumask *hk_mask; |
83cd4fe2 | 1069 | |
04d4e665 | 1070 | if (housekeeping_cpu(cpu, HK_TYPE_TIMER)) { |
e938b9c9 WL |
1071 | if (!idle_cpu(cpu)) |
1072 | return cpu; | |
1073 | default_cpu = cpu; | |
1074 | } | |
6201b4d6 | 1075 | |
04d4e665 | 1076 | hk_mask = housekeeping_cpumask(HK_TYPE_TIMER); |
031e3bd8 | 1077 | |
057f3fad | 1078 | rcu_read_lock(); |
83cd4fe2 | 1079 | for_each_domain(cpu, sd) { |
031e3bd8 | 1080 | for_each_cpu_and(i, sched_domain_span(sd), hk_mask) { |
44496922 WL |
1081 | if (cpu == i) |
1082 | continue; | |
1083 | ||
e938b9c9 | 1084 | if (!idle_cpu(i)) { |
057f3fad PZ |
1085 | cpu = i; |
1086 | goto unlock; | |
1087 | } | |
1088 | } | |
83cd4fe2 | 1089 | } |
9642d18e | 1090 | |
e938b9c9 | 1091 | if (default_cpu == -1) |
04d4e665 | 1092 | default_cpu = housekeeping_any_cpu(HK_TYPE_TIMER); |
e938b9c9 | 1093 | cpu = default_cpu; |
057f3fad PZ |
1094 | unlock: |
1095 | rcu_read_unlock(); | |
83cd4fe2 VP |
1096 | return cpu; |
1097 | } | |
d1ccc66d | 1098 | |
06d8308c TG |
1099 | /* |
1100 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1101 | * idle CPU then this timer might expire before the next timer event | |
1102 | * which is scheduled to wake up that CPU. In case of a completely | |
1103 | * idle system the next event might even be infinite time into the | |
1104 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1105 | * leaves the inner idle loop so the newly added timer is taken into | |
1106 | * account when the CPU goes back to idle and evaluates the timer | |
1107 | * wheel for the next timer event. | |
1108 | */ | |
1c20091e | 1109 | static void wake_up_idle_cpu(int cpu) |
06d8308c TG |
1110 | { |
1111 | struct rq *rq = cpu_rq(cpu); | |
1112 | ||
1113 | if (cpu == smp_processor_id()) | |
1114 | return; | |
1115 | ||
67b9ca70 | 1116 | if (set_nr_and_not_polling(rq->idle)) |
06d8308c | 1117 | smp_send_reschedule(cpu); |
dfc68f29 AL |
1118 | else |
1119 | trace_sched_wake_idle_without_ipi(cpu); | |
45bf76df IM |
1120 | } |
1121 | ||
c5bfece2 | 1122 | static bool wake_up_full_nohz_cpu(int cpu) |
1c20091e | 1123 | { |
53c5fa16 FW |
1124 | /* |
1125 | * We just need the target to call irq_exit() and re-evaluate | |
1126 | * the next tick. The nohz full kick at least implies that. | |
1127 | * If needed we can still optimize that later with an | |
1128 | * empty IRQ. | |
1129 | */ | |
379d9ecb PM |
1130 | if (cpu_is_offline(cpu)) |
1131 | return true; /* Don't try to wake offline CPUs. */ | |
c5bfece2 | 1132 | if (tick_nohz_full_cpu(cpu)) { |
1c20091e FW |
1133 | if (cpu != smp_processor_id() || |
1134 | tick_nohz_tick_stopped()) | |
53c5fa16 | 1135 | tick_nohz_full_kick_cpu(cpu); |
1c20091e FW |
1136 | return true; |
1137 | } | |
1138 | ||
1139 | return false; | |
1140 | } | |
1141 | ||
379d9ecb PM |
1142 | /* |
1143 | * Wake up the specified CPU. If the CPU is going offline, it is the | |
1144 | * caller's responsibility to deal with the lost wakeup, for example, | |
1145 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. | |
1146 | */ | |
1c20091e FW |
1147 | void wake_up_nohz_cpu(int cpu) |
1148 | { | |
c5bfece2 | 1149 | if (!wake_up_full_nohz_cpu(cpu)) |
1c20091e FW |
1150 | wake_up_idle_cpu(cpu); |
1151 | } | |
1152 | ||
19a1f5ec | 1153 | static void nohz_csd_func(void *info) |
45bf76df | 1154 | { |
19a1f5ec PZ |
1155 | struct rq *rq = info; |
1156 | int cpu = cpu_of(rq); | |
1157 | unsigned int flags; | |
873b4c65 VG |
1158 | |
1159 | /* | |
19a1f5ec | 1160 | * Release the rq::nohz_csd. |
873b4c65 | 1161 | */ |
c6f88654 | 1162 | flags = atomic_fetch_andnot(NOHZ_KICK_MASK | NOHZ_NEWILB_KICK, nohz_flags(cpu)); |
19a1f5ec | 1163 | WARN_ON(!(flags & NOHZ_KICK_MASK)); |
45bf76df | 1164 | |
19a1f5ec PZ |
1165 | rq->idle_balance = idle_cpu(cpu); |
1166 | if (rq->idle_balance && !need_resched()) { | |
1167 | rq->nohz_idle_balance = flags; | |
90b5363a PZI |
1168 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
1169 | } | |
2069dd75 PZ |
1170 | } |
1171 | ||
3451d024 | 1172 | #endif /* CONFIG_NO_HZ_COMMON */ |
d842de87 | 1173 | |
ce831b38 | 1174 | #ifdef CONFIG_NO_HZ_FULL |
76d92ac3 | 1175 | bool sched_can_stop_tick(struct rq *rq) |
ce831b38 | 1176 | { |
76d92ac3 FW |
1177 | int fifo_nr_running; |
1178 | ||
1179 | /* Deadline tasks, even if single, need the tick */ | |
1180 | if (rq->dl.dl_nr_running) | |
1181 | return false; | |
1182 | ||
1e78cdbd | 1183 | /* |
b19a888c | 1184 | * If there are more than one RR tasks, we need the tick to affect the |
2548d546 | 1185 | * actual RR behaviour. |
1e78cdbd | 1186 | */ |
76d92ac3 FW |
1187 | if (rq->rt.rr_nr_running) { |
1188 | if (rq->rt.rr_nr_running == 1) | |
1189 | return true; | |
1190 | else | |
1191 | return false; | |
1e78cdbd RR |
1192 | } |
1193 | ||
2548d546 PZ |
1194 | /* |
1195 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no | |
1196 | * forced preemption between FIFO tasks. | |
1197 | */ | |
1198 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; | |
1199 | if (fifo_nr_running) | |
1200 | return true; | |
1201 | ||
1202 | /* | |
1203 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; | |
1204 | * if there's more than one we need the tick for involuntary | |
1205 | * preemption. | |
1206 | */ | |
1207 | if (rq->nr_running > 1) | |
541b8264 | 1208 | return false; |
ce831b38 | 1209 | |
541b8264 | 1210 | return true; |
ce831b38 FW |
1211 | } |
1212 | #endif /* CONFIG_NO_HZ_FULL */ | |
6d6bc0ad | 1213 | #endif /* CONFIG_SMP */ |
18d95a28 | 1214 | |
a790de99 PT |
1215 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
1216 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) | |
c09595f6 | 1217 | /* |
8277434e PT |
1218 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
1219 | * node and @up when leaving it for the final time. | |
1220 | * | |
1221 | * Caller must hold rcu_lock or sufficient equivalent. | |
c09595f6 | 1222 | */ |
029632fb | 1223 | int walk_tg_tree_from(struct task_group *from, |
8277434e | 1224 | tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1225 | { |
1226 | struct task_group *parent, *child; | |
eb755805 | 1227 | int ret; |
c09595f6 | 1228 | |
8277434e PT |
1229 | parent = from; |
1230 | ||
c09595f6 | 1231 | down: |
eb755805 PZ |
1232 | ret = (*down)(parent, data); |
1233 | if (ret) | |
8277434e | 1234 | goto out; |
c09595f6 PZ |
1235 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1236 | parent = child; | |
1237 | goto down; | |
1238 | ||
1239 | up: | |
1240 | continue; | |
1241 | } | |
eb755805 | 1242 | ret = (*up)(parent, data); |
8277434e PT |
1243 | if (ret || parent == from) |
1244 | goto out; | |
c09595f6 PZ |
1245 | |
1246 | child = parent; | |
1247 | parent = parent->parent; | |
1248 | if (parent) | |
1249 | goto up; | |
8277434e | 1250 | out: |
eb755805 | 1251 | return ret; |
c09595f6 PZ |
1252 | } |
1253 | ||
029632fb | 1254 | int tg_nop(struct task_group *tg, void *data) |
eb755805 | 1255 | { |
e2b245f8 | 1256 | return 0; |
eb755805 | 1257 | } |
18d95a28 PZ |
1258 | #endif |
1259 | ||
b1e82065 | 1260 | static void set_load_weight(struct task_struct *p, bool update_load) |
45bf76df | 1261 | { |
f05998d4 NR |
1262 | int prio = p->static_prio - MAX_RT_PRIO; |
1263 | struct load_weight *load = &p->se.load; | |
1264 | ||
dd41f596 IM |
1265 | /* |
1266 | * SCHED_IDLE tasks get minimal weight: | |
1267 | */ | |
1da1843f | 1268 | if (task_has_idle_policy(p)) { |
c8b28116 | 1269 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 1270 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
1271 | return; |
1272 | } | |
71f8bd46 | 1273 | |
9059393e VG |
1274 | /* |
1275 | * SCHED_OTHER tasks have to update their load when changing their | |
1276 | * weight | |
1277 | */ | |
1278 | if (update_load && p->sched_class == &fair_sched_class) { | |
1279 | reweight_task(p, prio); | |
1280 | } else { | |
1281 | load->weight = scale_load(sched_prio_to_weight[prio]); | |
1282 | load->inv_weight = sched_prio_to_wmult[prio]; | |
1283 | } | |
71f8bd46 IM |
1284 | } |
1285 | ||
69842cba | 1286 | #ifdef CONFIG_UCLAMP_TASK |
2480c093 PB |
1287 | /* |
1288 | * Serializes updates of utilization clamp values | |
1289 | * | |
1290 | * The (slow-path) user-space triggers utilization clamp value updates which | |
1291 | * can require updates on (fast-path) scheduler's data structures used to | |
1292 | * support enqueue/dequeue operations. | |
1293 | * While the per-CPU rq lock protects fast-path update operations, user-space | |
1294 | * requests are serialized using a mutex to reduce the risk of conflicting | |
1295 | * updates or API abuses. | |
1296 | */ | |
1297 | static DEFINE_MUTEX(uclamp_mutex); | |
1298 | ||
e8f14172 | 1299 | /* Max allowed minimum utilization */ |
494dcdf4 | 1300 | static unsigned int __maybe_unused sysctl_sched_uclamp_util_min = SCHED_CAPACITY_SCALE; |
e8f14172 PB |
1301 | |
1302 | /* Max allowed maximum utilization */ | |
494dcdf4 | 1303 | static unsigned int __maybe_unused sysctl_sched_uclamp_util_max = SCHED_CAPACITY_SCALE; |
e8f14172 | 1304 | |
13685c4a QY |
1305 | /* |
1306 | * By default RT tasks run at the maximum performance point/capacity of the | |
1307 | * system. Uclamp enforces this by always setting UCLAMP_MIN of RT tasks to | |
1308 | * SCHED_CAPACITY_SCALE. | |
1309 | * | |
1310 | * This knob allows admins to change the default behavior when uclamp is being | |
1311 | * used. In battery powered devices, particularly, running at the maximum | |
1312 | * capacity and frequency will increase energy consumption and shorten the | |
1313 | * battery life. | |
1314 | * | |
1315 | * This knob only affects RT tasks that their uclamp_se->user_defined == false. | |
1316 | * | |
1317 | * This knob will not override the system default sched_util_clamp_min defined | |
1318 | * above. | |
1319 | */ | |
3267e015 | 1320 | static unsigned int sysctl_sched_uclamp_util_min_rt_default = SCHED_CAPACITY_SCALE; |
13685c4a | 1321 | |
e8f14172 PB |
1322 | /* All clamps are required to be less or equal than these values */ |
1323 | static struct uclamp_se uclamp_default[UCLAMP_CNT]; | |
69842cba | 1324 | |
46609ce2 QY |
1325 | /* |
1326 | * This static key is used to reduce the uclamp overhead in the fast path. It | |
1327 | * primarily disables the call to uclamp_rq_{inc, dec}() in | |
1328 | * enqueue/dequeue_task(). | |
1329 | * | |
1330 | * This allows users to continue to enable uclamp in their kernel config with | |
1331 | * minimum uclamp overhead in the fast path. | |
1332 | * | |
1333 | * As soon as userspace modifies any of the uclamp knobs, the static key is | |
1334 | * enabled, since we have an actual users that make use of uclamp | |
1335 | * functionality. | |
1336 | * | |
1337 | * The knobs that would enable this static key are: | |
1338 | * | |
1339 | * * A task modifying its uclamp value with sched_setattr(). | |
1340 | * * An admin modifying the sysctl_sched_uclamp_{min, max} via procfs. | |
1341 | * * An admin modifying the cgroup cpu.uclamp.{min, max} | |
1342 | */ | |
1343 | DEFINE_STATIC_KEY_FALSE(sched_uclamp_used); | |
1344 | ||
69842cba PB |
1345 | /* Integer rounded range for each bucket */ |
1346 | #define UCLAMP_BUCKET_DELTA DIV_ROUND_CLOSEST(SCHED_CAPACITY_SCALE, UCLAMP_BUCKETS) | |
1347 | ||
1348 | #define for_each_clamp_id(clamp_id) \ | |
1349 | for ((clamp_id) = 0; (clamp_id) < UCLAMP_CNT; (clamp_id)++) | |
1350 | ||
1351 | static inline unsigned int uclamp_bucket_id(unsigned int clamp_value) | |
1352 | { | |
6d2f8909 | 1353 | return min_t(unsigned int, clamp_value / UCLAMP_BUCKET_DELTA, UCLAMP_BUCKETS - 1); |
69842cba PB |
1354 | } |
1355 | ||
7763baac | 1356 | static inline unsigned int uclamp_none(enum uclamp_id clamp_id) |
69842cba PB |
1357 | { |
1358 | if (clamp_id == UCLAMP_MIN) | |
1359 | return 0; | |
1360 | return SCHED_CAPACITY_SCALE; | |
1361 | } | |
1362 | ||
a509a7cd PB |
1363 | static inline void uclamp_se_set(struct uclamp_se *uc_se, |
1364 | unsigned int value, bool user_defined) | |
69842cba PB |
1365 | { |
1366 | uc_se->value = value; | |
1367 | uc_se->bucket_id = uclamp_bucket_id(value); | |
a509a7cd | 1368 | uc_se->user_defined = user_defined; |
69842cba PB |
1369 | } |
1370 | ||
e496187d | 1371 | static inline unsigned int |
0413d7f3 | 1372 | uclamp_idle_value(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
1373 | unsigned int clamp_value) |
1374 | { | |
1375 | /* | |
1376 | * Avoid blocked utilization pushing up the frequency when we go | |
1377 | * idle (which drops the max-clamp) by retaining the last known | |
1378 | * max-clamp. | |
1379 | */ | |
1380 | if (clamp_id == UCLAMP_MAX) { | |
1381 | rq->uclamp_flags |= UCLAMP_FLAG_IDLE; | |
1382 | return clamp_value; | |
1383 | } | |
1384 | ||
1385 | return uclamp_none(UCLAMP_MIN); | |
1386 | } | |
1387 | ||
0413d7f3 | 1388 | static inline void uclamp_idle_reset(struct rq *rq, enum uclamp_id clamp_id, |
e496187d PB |
1389 | unsigned int clamp_value) |
1390 | { | |
1391 | /* Reset max-clamp retention only on idle exit */ | |
1392 | if (!(rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
1393 | return; | |
1394 | ||
1395 | WRITE_ONCE(rq->uclamp[clamp_id].value, clamp_value); | |
1396 | } | |
1397 | ||
69842cba | 1398 | static inline |
7763baac | 1399 | unsigned int uclamp_rq_max_value(struct rq *rq, enum uclamp_id clamp_id, |
0413d7f3 | 1400 | unsigned int clamp_value) |
69842cba PB |
1401 | { |
1402 | struct uclamp_bucket *bucket = rq->uclamp[clamp_id].bucket; | |
1403 | int bucket_id = UCLAMP_BUCKETS - 1; | |
1404 | ||
1405 | /* | |
1406 | * Since both min and max clamps are max aggregated, find the | |
1407 | * top most bucket with tasks in. | |
1408 | */ | |
1409 | for ( ; bucket_id >= 0; bucket_id--) { | |
1410 | if (!bucket[bucket_id].tasks) | |
1411 | continue; | |
1412 | return bucket[bucket_id].value; | |
1413 | } | |
1414 | ||
1415 | /* No tasks -- default clamp values */ | |
e496187d | 1416 | return uclamp_idle_value(rq, clamp_id, clamp_value); |
69842cba PB |
1417 | } |
1418 | ||
13685c4a QY |
1419 | static void __uclamp_update_util_min_rt_default(struct task_struct *p) |
1420 | { | |
1421 | unsigned int default_util_min; | |
1422 | struct uclamp_se *uc_se; | |
1423 | ||
1424 | lockdep_assert_held(&p->pi_lock); | |
1425 | ||
1426 | uc_se = &p->uclamp_req[UCLAMP_MIN]; | |
1427 | ||
1428 | /* Only sync if user didn't override the default */ | |
1429 | if (uc_se->user_defined) | |
1430 | return; | |
1431 | ||
1432 | default_util_min = sysctl_sched_uclamp_util_min_rt_default; | |
1433 | uclamp_se_set(uc_se, default_util_min, false); | |
1434 | } | |
1435 | ||
1436 | static void uclamp_update_util_min_rt_default(struct task_struct *p) | |
1437 | { | |
1438 | struct rq_flags rf; | |
1439 | struct rq *rq; | |
1440 | ||
1441 | if (!rt_task(p)) | |
1442 | return; | |
1443 | ||
1444 | /* Protect updates to p->uclamp_* */ | |
1445 | rq = task_rq_lock(p, &rf); | |
1446 | __uclamp_update_util_min_rt_default(p); | |
1447 | task_rq_unlock(rq, p, &rf); | |
1448 | } | |
1449 | ||
3eac870a | 1450 | static inline struct uclamp_se |
0413d7f3 | 1451 | uclamp_tg_restrict(struct task_struct *p, enum uclamp_id clamp_id) |
3eac870a | 1452 | { |
0213b708 | 1453 | /* Copy by value as we could modify it */ |
3eac870a PB |
1454 | struct uclamp_se uc_req = p->uclamp_req[clamp_id]; |
1455 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0213b708 | 1456 | unsigned int tg_min, tg_max, value; |
3eac870a PB |
1457 | |
1458 | /* | |
1459 | * Tasks in autogroups or root task group will be | |
1460 | * restricted by system defaults. | |
1461 | */ | |
1462 | if (task_group_is_autogroup(task_group(p))) | |
1463 | return uc_req; | |
1464 | if (task_group(p) == &root_task_group) | |
1465 | return uc_req; | |
1466 | ||
0213b708 QY |
1467 | tg_min = task_group(p)->uclamp[UCLAMP_MIN].value; |
1468 | tg_max = task_group(p)->uclamp[UCLAMP_MAX].value; | |
1469 | value = uc_req.value; | |
1470 | value = clamp(value, tg_min, tg_max); | |
1471 | uclamp_se_set(&uc_req, value, false); | |
3eac870a PB |
1472 | #endif |
1473 | ||
1474 | return uc_req; | |
1475 | } | |
1476 | ||
e8f14172 PB |
1477 | /* |
1478 | * The effective clamp bucket index of a task depends on, by increasing | |
1479 | * priority: | |
1480 | * - the task specific clamp value, when explicitly requested from userspace | |
3eac870a PB |
1481 | * - the task group effective clamp value, for tasks not either in the root |
1482 | * group or in an autogroup | |
e8f14172 PB |
1483 | * - the system default clamp value, defined by the sysadmin |
1484 | */ | |
1485 | static inline struct uclamp_se | |
0413d7f3 | 1486 | uclamp_eff_get(struct task_struct *p, enum uclamp_id clamp_id) |
e8f14172 | 1487 | { |
3eac870a | 1488 | struct uclamp_se uc_req = uclamp_tg_restrict(p, clamp_id); |
e8f14172 PB |
1489 | struct uclamp_se uc_max = uclamp_default[clamp_id]; |
1490 | ||
1491 | /* System default restrictions always apply */ | |
1492 | if (unlikely(uc_req.value > uc_max.value)) | |
1493 | return uc_max; | |
1494 | ||
1495 | return uc_req; | |
1496 | } | |
1497 | ||
686516b5 | 1498 | unsigned long uclamp_eff_value(struct task_struct *p, enum uclamp_id clamp_id) |
9d20ad7d PB |
1499 | { |
1500 | struct uclamp_se uc_eff; | |
1501 | ||
1502 | /* Task currently refcounted: use back-annotated (effective) value */ | |
1503 | if (p->uclamp[clamp_id].active) | |
686516b5 | 1504 | return (unsigned long)p->uclamp[clamp_id].value; |
9d20ad7d PB |
1505 | |
1506 | uc_eff = uclamp_eff_get(p, clamp_id); | |
1507 | ||
686516b5 | 1508 | return (unsigned long)uc_eff.value; |
9d20ad7d PB |
1509 | } |
1510 | ||
69842cba PB |
1511 | /* |
1512 | * When a task is enqueued on a rq, the clamp bucket currently defined by the | |
1513 | * task's uclamp::bucket_id is refcounted on that rq. This also immediately | |
1514 | * updates the rq's clamp value if required. | |
60daf9c1 PB |
1515 | * |
1516 | * Tasks can have a task-specific value requested from user-space, track | |
1517 | * within each bucket the maximum value for tasks refcounted in it. | |
1518 | * This "local max aggregation" allows to track the exact "requested" value | |
1519 | * for each bucket when all its RUNNABLE tasks require the same clamp. | |
69842cba PB |
1520 | */ |
1521 | static inline void uclamp_rq_inc_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1522 | enum uclamp_id clamp_id) |
69842cba PB |
1523 | { |
1524 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1525 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1526 | struct uclamp_bucket *bucket; | |
1527 | ||
5cb9eaa3 | 1528 | lockdep_assert_rq_held(rq); |
69842cba | 1529 | |
e8f14172 PB |
1530 | /* Update task effective clamp */ |
1531 | p->uclamp[clamp_id] = uclamp_eff_get(p, clamp_id); | |
1532 | ||
69842cba PB |
1533 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
1534 | bucket->tasks++; | |
e8f14172 | 1535 | uc_se->active = true; |
69842cba | 1536 | |
e496187d PB |
1537 | uclamp_idle_reset(rq, clamp_id, uc_se->value); |
1538 | ||
60daf9c1 PB |
1539 | /* |
1540 | * Local max aggregation: rq buckets always track the max | |
1541 | * "requested" clamp value of its RUNNABLE tasks. | |
1542 | */ | |
1543 | if (bucket->tasks == 1 || uc_se->value > bucket->value) | |
1544 | bucket->value = uc_se->value; | |
1545 | ||
69842cba | 1546 | if (uc_se->value > READ_ONCE(uc_rq->value)) |
60daf9c1 | 1547 | WRITE_ONCE(uc_rq->value, uc_se->value); |
69842cba PB |
1548 | } |
1549 | ||
1550 | /* | |
1551 | * When a task is dequeued from a rq, the clamp bucket refcounted by the task | |
1552 | * is released. If this is the last task reference counting the rq's max | |
1553 | * active clamp value, then the rq's clamp value is updated. | |
1554 | * | |
1555 | * Both refcounted tasks and rq's cached clamp values are expected to be | |
1556 | * always valid. If it's detected they are not, as defensive programming, | |
1557 | * enforce the expected state and warn. | |
1558 | */ | |
1559 | static inline void uclamp_rq_dec_id(struct rq *rq, struct task_struct *p, | |
0413d7f3 | 1560 | enum uclamp_id clamp_id) |
69842cba PB |
1561 | { |
1562 | struct uclamp_rq *uc_rq = &rq->uclamp[clamp_id]; | |
1563 | struct uclamp_se *uc_se = &p->uclamp[clamp_id]; | |
1564 | struct uclamp_bucket *bucket; | |
e496187d | 1565 | unsigned int bkt_clamp; |
69842cba PB |
1566 | unsigned int rq_clamp; |
1567 | ||
5cb9eaa3 | 1568 | lockdep_assert_rq_held(rq); |
69842cba | 1569 | |
46609ce2 QY |
1570 | /* |
1571 | * If sched_uclamp_used was enabled after task @p was enqueued, | |
1572 | * we could end up with unbalanced call to uclamp_rq_dec_id(). | |
1573 | * | |
1574 | * In this case the uc_se->active flag should be false since no uclamp | |
1575 | * accounting was performed at enqueue time and we can just return | |
1576 | * here. | |
1577 | * | |
b19a888c | 1578 | * Need to be careful of the following enqueue/dequeue ordering |
46609ce2 QY |
1579 | * problem too |
1580 | * | |
1581 | * enqueue(taskA) | |
1582 | * // sched_uclamp_used gets enabled | |
1583 | * enqueue(taskB) | |
1584 | * dequeue(taskA) | |
b19a888c | 1585 | * // Must not decrement bucket->tasks here |
46609ce2 QY |
1586 | * dequeue(taskB) |
1587 | * | |
1588 | * where we could end up with stale data in uc_se and | |
1589 | * bucket[uc_se->bucket_id]. | |
1590 | * | |
1591 | * The following check here eliminates the possibility of such race. | |
1592 | */ | |
1593 | if (unlikely(!uc_se->active)) | |
1594 | return; | |
1595 | ||
69842cba | 1596 | bucket = &uc_rq->bucket[uc_se->bucket_id]; |
46609ce2 | 1597 | |
69842cba PB |
1598 | SCHED_WARN_ON(!bucket->tasks); |
1599 | if (likely(bucket->tasks)) | |
1600 | bucket->tasks--; | |
46609ce2 | 1601 | |
e8f14172 | 1602 | uc_se->active = false; |
69842cba | 1603 | |
60daf9c1 PB |
1604 | /* |
1605 | * Keep "local max aggregation" simple and accept to (possibly) | |
1606 | * overboost some RUNNABLE tasks in the same bucket. | |
1607 | * The rq clamp bucket value is reset to its base value whenever | |
1608 | * there are no more RUNNABLE tasks refcounting it. | |
1609 | */ | |
69842cba PB |
1610 | if (likely(bucket->tasks)) |
1611 | return; | |
1612 | ||
1613 | rq_clamp = READ_ONCE(uc_rq->value); | |
1614 | /* | |
1615 | * Defensive programming: this should never happen. If it happens, | |
1616 | * e.g. due to future modification, warn and fixup the expected value. | |
1617 | */ | |
1618 | SCHED_WARN_ON(bucket->value > rq_clamp); | |
e496187d PB |
1619 | if (bucket->value >= rq_clamp) { |
1620 | bkt_clamp = uclamp_rq_max_value(rq, clamp_id, uc_se->value); | |
1621 | WRITE_ONCE(uc_rq->value, bkt_clamp); | |
1622 | } | |
69842cba PB |
1623 | } |
1624 | ||
1625 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) | |
1626 | { | |
0413d7f3 | 1627 | enum uclamp_id clamp_id; |
69842cba | 1628 | |
46609ce2 QY |
1629 | /* |
1630 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1631 | * | |
1632 | * The condition is constructed such that a NOP is generated when | |
1633 | * sched_uclamp_used is disabled. | |
1634 | */ | |
1635 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1636 | return; | |
1637 | ||
69842cba PB |
1638 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1639 | return; | |
1640 | ||
1641 | for_each_clamp_id(clamp_id) | |
1642 | uclamp_rq_inc_id(rq, p, clamp_id); | |
e496187d PB |
1643 | |
1644 | /* Reset clamp idle holding when there is one RUNNABLE task */ | |
1645 | if (rq->uclamp_flags & UCLAMP_FLAG_IDLE) | |
1646 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
69842cba PB |
1647 | } |
1648 | ||
1649 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) | |
1650 | { | |
0413d7f3 | 1651 | enum uclamp_id clamp_id; |
69842cba | 1652 | |
46609ce2 QY |
1653 | /* |
1654 | * Avoid any overhead until uclamp is actually used by the userspace. | |
1655 | * | |
1656 | * The condition is constructed such that a NOP is generated when | |
1657 | * sched_uclamp_used is disabled. | |
1658 | */ | |
1659 | if (!static_branch_unlikely(&sched_uclamp_used)) | |
1660 | return; | |
1661 | ||
69842cba PB |
1662 | if (unlikely(!p->sched_class->uclamp_enabled)) |
1663 | return; | |
1664 | ||
1665 | for_each_clamp_id(clamp_id) | |
1666 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1667 | } | |
1668 | ||
ca4984a7 QP |
1669 | static inline void uclamp_rq_reinc_id(struct rq *rq, struct task_struct *p, |
1670 | enum uclamp_id clamp_id) | |
1671 | { | |
1672 | if (!p->uclamp[clamp_id].active) | |
1673 | return; | |
1674 | ||
1675 | uclamp_rq_dec_id(rq, p, clamp_id); | |
1676 | uclamp_rq_inc_id(rq, p, clamp_id); | |
1677 | ||
1678 | /* | |
1679 | * Make sure to clear the idle flag if we've transiently reached 0 | |
1680 | * active tasks on rq. | |
1681 | */ | |
1682 | if (clamp_id == UCLAMP_MAX && (rq->uclamp_flags & UCLAMP_FLAG_IDLE)) | |
1683 | rq->uclamp_flags &= ~UCLAMP_FLAG_IDLE; | |
1684 | } | |
1685 | ||
babbe170 | 1686 | static inline void |
0213b708 | 1687 | uclamp_update_active(struct task_struct *p) |
babbe170 | 1688 | { |
0213b708 | 1689 | enum uclamp_id clamp_id; |
babbe170 PB |
1690 | struct rq_flags rf; |
1691 | struct rq *rq; | |
1692 | ||
1693 | /* | |
1694 | * Lock the task and the rq where the task is (or was) queued. | |
1695 | * | |
1696 | * We might lock the (previous) rq of a !RUNNABLE task, but that's the | |
1697 | * price to pay to safely serialize util_{min,max} updates with | |
1698 | * enqueues, dequeues and migration operations. | |
1699 | * This is the same locking schema used by __set_cpus_allowed_ptr(). | |
1700 | */ | |
1701 | rq = task_rq_lock(p, &rf); | |
1702 | ||
1703 | /* | |
1704 | * Setting the clamp bucket is serialized by task_rq_lock(). | |
1705 | * If the task is not yet RUNNABLE and its task_struct is not | |
1706 | * affecting a valid clamp bucket, the next time it's enqueued, | |
1707 | * it will already see the updated clamp bucket value. | |
1708 | */ | |
ca4984a7 QP |
1709 | for_each_clamp_id(clamp_id) |
1710 | uclamp_rq_reinc_id(rq, p, clamp_id); | |
babbe170 PB |
1711 | |
1712 | task_rq_unlock(rq, p, &rf); | |
1713 | } | |
1714 | ||
e3b8b6a0 | 1715 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
babbe170 | 1716 | static inline void |
0213b708 | 1717 | uclamp_update_active_tasks(struct cgroup_subsys_state *css) |
babbe170 PB |
1718 | { |
1719 | struct css_task_iter it; | |
1720 | struct task_struct *p; | |
babbe170 PB |
1721 | |
1722 | css_task_iter_start(css, 0, &it); | |
0213b708 QY |
1723 | while ((p = css_task_iter_next(&it))) |
1724 | uclamp_update_active(p); | |
babbe170 PB |
1725 | css_task_iter_end(&it); |
1726 | } | |
1727 | ||
7274a5c1 | 1728 | static void cpu_util_update_eff(struct cgroup_subsys_state *css); |
494dcdf4 Y |
1729 | #endif |
1730 | ||
1731 | #ifdef CONFIG_SYSCTL | |
1732 | #ifdef CONFIG_UCLAMP_TASK | |
1733 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
7274a5c1 PB |
1734 | static void uclamp_update_root_tg(void) |
1735 | { | |
1736 | struct task_group *tg = &root_task_group; | |
1737 | ||
1738 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MIN], | |
1739 | sysctl_sched_uclamp_util_min, false); | |
1740 | uclamp_se_set(&tg->uclamp_req[UCLAMP_MAX], | |
1741 | sysctl_sched_uclamp_util_max, false); | |
1742 | ||
1743 | rcu_read_lock(); | |
1744 | cpu_util_update_eff(&root_task_group.css); | |
1745 | rcu_read_unlock(); | |
1746 | } | |
1747 | #else | |
1748 | static void uclamp_update_root_tg(void) { } | |
1749 | #endif | |
1750 | ||
494dcdf4 Y |
1751 | static void uclamp_sync_util_min_rt_default(void) |
1752 | { | |
1753 | struct task_struct *g, *p; | |
1754 | ||
1755 | /* | |
1756 | * copy_process() sysctl_uclamp | |
1757 | * uclamp_min_rt = X; | |
1758 | * write_lock(&tasklist_lock) read_lock(&tasklist_lock) | |
1759 | * // link thread smp_mb__after_spinlock() | |
1760 | * write_unlock(&tasklist_lock) read_unlock(&tasklist_lock); | |
1761 | * sched_post_fork() for_each_process_thread() | |
1762 | * __uclamp_sync_rt() __uclamp_sync_rt() | |
1763 | * | |
1764 | * Ensures that either sched_post_fork() will observe the new | |
1765 | * uclamp_min_rt or for_each_process_thread() will observe the new | |
1766 | * task. | |
1767 | */ | |
1768 | read_lock(&tasklist_lock); | |
1769 | smp_mb__after_spinlock(); | |
1770 | read_unlock(&tasklist_lock); | |
1771 | ||
1772 | rcu_read_lock(); | |
1773 | for_each_process_thread(g, p) | |
1774 | uclamp_update_util_min_rt_default(p); | |
1775 | rcu_read_unlock(); | |
1776 | } | |
1777 | ||
3267e015 | 1778 | static int sysctl_sched_uclamp_handler(struct ctl_table *table, int write, |
32927393 | 1779 | void *buffer, size_t *lenp, loff_t *ppos) |
e8f14172 | 1780 | { |
7274a5c1 | 1781 | bool update_root_tg = false; |
13685c4a | 1782 | int old_min, old_max, old_min_rt; |
e8f14172 PB |
1783 | int result; |
1784 | ||
2480c093 | 1785 | mutex_lock(&uclamp_mutex); |
e8f14172 PB |
1786 | old_min = sysctl_sched_uclamp_util_min; |
1787 | old_max = sysctl_sched_uclamp_util_max; | |
13685c4a | 1788 | old_min_rt = sysctl_sched_uclamp_util_min_rt_default; |
e8f14172 PB |
1789 | |
1790 | result = proc_dointvec(table, write, buffer, lenp, ppos); | |
1791 | if (result) | |
1792 | goto undo; | |
1793 | if (!write) | |
1794 | goto done; | |
1795 | ||
1796 | if (sysctl_sched_uclamp_util_min > sysctl_sched_uclamp_util_max || | |
13685c4a QY |
1797 | sysctl_sched_uclamp_util_max > SCHED_CAPACITY_SCALE || |
1798 | sysctl_sched_uclamp_util_min_rt_default > SCHED_CAPACITY_SCALE) { | |
1799 | ||
e8f14172 PB |
1800 | result = -EINVAL; |
1801 | goto undo; | |
1802 | } | |
1803 | ||
1804 | if (old_min != sysctl_sched_uclamp_util_min) { | |
1805 | uclamp_se_set(&uclamp_default[UCLAMP_MIN], | |
a509a7cd | 1806 | sysctl_sched_uclamp_util_min, false); |
7274a5c1 | 1807 | update_root_tg = true; |
e8f14172 PB |
1808 | } |
1809 | if (old_max != sysctl_sched_uclamp_util_max) { | |
1810 | uclamp_se_set(&uclamp_default[UCLAMP_MAX], | |
a509a7cd | 1811 | sysctl_sched_uclamp_util_max, false); |
7274a5c1 | 1812 | update_root_tg = true; |
e8f14172 PB |
1813 | } |
1814 | ||
46609ce2 QY |
1815 | if (update_root_tg) { |
1816 | static_branch_enable(&sched_uclamp_used); | |
7274a5c1 | 1817 | uclamp_update_root_tg(); |
46609ce2 | 1818 | } |
7274a5c1 | 1819 | |
13685c4a QY |
1820 | if (old_min_rt != sysctl_sched_uclamp_util_min_rt_default) { |
1821 | static_branch_enable(&sched_uclamp_used); | |
1822 | uclamp_sync_util_min_rt_default(); | |
1823 | } | |
7274a5c1 | 1824 | |
e8f14172 | 1825 | /* |
7274a5c1 PB |
1826 | * We update all RUNNABLE tasks only when task groups are in use. |
1827 | * Otherwise, keep it simple and do just a lazy update at each next | |
1828 | * task enqueue time. | |
e8f14172 | 1829 | */ |
7274a5c1 | 1830 | |
e8f14172 PB |
1831 | goto done; |
1832 | ||
1833 | undo: | |
1834 | sysctl_sched_uclamp_util_min = old_min; | |
1835 | sysctl_sched_uclamp_util_max = old_max; | |
13685c4a | 1836 | sysctl_sched_uclamp_util_min_rt_default = old_min_rt; |
e8f14172 | 1837 | done: |
2480c093 | 1838 | mutex_unlock(&uclamp_mutex); |
e8f14172 PB |
1839 | |
1840 | return result; | |
1841 | } | |
494dcdf4 Y |
1842 | #endif |
1843 | #endif | |
e8f14172 | 1844 | |
a509a7cd PB |
1845 | static int uclamp_validate(struct task_struct *p, |
1846 | const struct sched_attr *attr) | |
1847 | { | |
480a6ca2 DE |
1848 | int util_min = p->uclamp_req[UCLAMP_MIN].value; |
1849 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd | 1850 | |
480a6ca2 DE |
1851 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { |
1852 | util_min = attr->sched_util_min; | |
a509a7cd | 1853 | |
480a6ca2 DE |
1854 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) |
1855 | return -EINVAL; | |
1856 | } | |
1857 | ||
1858 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | |
1859 | util_max = attr->sched_util_max; | |
1860 | ||
1861 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | |
1862 | return -EINVAL; | |
1863 | } | |
1864 | ||
1865 | if (util_min != -1 && util_max != -1 && util_min > util_max) | |
a509a7cd PB |
1866 | return -EINVAL; |
1867 | ||
e65855a5 QY |
1868 | /* |
1869 | * We have valid uclamp attributes; make sure uclamp is enabled. | |
1870 | * | |
1871 | * We need to do that here, because enabling static branches is a | |
1872 | * blocking operation which obviously cannot be done while holding | |
1873 | * scheduler locks. | |
1874 | */ | |
1875 | static_branch_enable(&sched_uclamp_used); | |
1876 | ||
a509a7cd PB |
1877 | return 0; |
1878 | } | |
1879 | ||
480a6ca2 DE |
1880 | static bool uclamp_reset(const struct sched_attr *attr, |
1881 | enum uclamp_id clamp_id, | |
1882 | struct uclamp_se *uc_se) | |
1883 | { | |
1884 | /* Reset on sched class change for a non user-defined clamp value. */ | |
1885 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | |
1886 | !uc_se->user_defined) | |
1887 | return true; | |
1888 | ||
1889 | /* Reset on sched_util_{min,max} == -1. */ | |
1890 | if (clamp_id == UCLAMP_MIN && | |
1891 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | |
1892 | attr->sched_util_min == -1) { | |
1893 | return true; | |
1894 | } | |
1895 | ||
1896 | if (clamp_id == UCLAMP_MAX && | |
1897 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | |
1898 | attr->sched_util_max == -1) { | |
1899 | return true; | |
1900 | } | |
1901 | ||
1902 | return false; | |
1903 | } | |
1904 | ||
a509a7cd PB |
1905 | static void __setscheduler_uclamp(struct task_struct *p, |
1906 | const struct sched_attr *attr) | |
1907 | { | |
0413d7f3 | 1908 | enum uclamp_id clamp_id; |
1a00d999 | 1909 | |
1a00d999 PB |
1910 | for_each_clamp_id(clamp_id) { |
1911 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | |
480a6ca2 | 1912 | unsigned int value; |
1a00d999 | 1913 | |
480a6ca2 | 1914 | if (!uclamp_reset(attr, clamp_id, uc_se)) |
1a00d999 PB |
1915 | continue; |
1916 | ||
13685c4a QY |
1917 | /* |
1918 | * RT by default have a 100% boost value that could be modified | |
1919 | * at runtime. | |
1920 | */ | |
1a00d999 | 1921 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) |
480a6ca2 | 1922 | value = sysctl_sched_uclamp_util_min_rt_default; |
13685c4a | 1923 | else |
480a6ca2 DE |
1924 | value = uclamp_none(clamp_id); |
1925 | ||
1926 | uclamp_se_set(uc_se, value, false); | |
1a00d999 | 1927 | |
1a00d999 PB |
1928 | } |
1929 | ||
a509a7cd PB |
1930 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) |
1931 | return; | |
1932 | ||
480a6ca2 DE |
1933 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && |
1934 | attr->sched_util_min != -1) { | |
a509a7cd PB |
1935 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], |
1936 | attr->sched_util_min, true); | |
1937 | } | |
1938 | ||
480a6ca2 DE |
1939 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && |
1940 | attr->sched_util_max != -1) { | |
a509a7cd PB |
1941 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], |
1942 | attr->sched_util_max, true); | |
1943 | } | |
1944 | } | |
1945 | ||
e8f14172 PB |
1946 | static void uclamp_fork(struct task_struct *p) |
1947 | { | |
0413d7f3 | 1948 | enum uclamp_id clamp_id; |
e8f14172 | 1949 | |
13685c4a QY |
1950 | /* |
1951 | * We don't need to hold task_rq_lock() when updating p->uclamp_* here | |
1952 | * as the task is still at its early fork stages. | |
1953 | */ | |
e8f14172 PB |
1954 | for_each_clamp_id(clamp_id) |
1955 | p->uclamp[clamp_id].active = false; | |
a87498ac PB |
1956 | |
1957 | if (likely(!p->sched_reset_on_fork)) | |
1958 | return; | |
1959 | ||
1960 | for_each_clamp_id(clamp_id) { | |
eaf5a92e QP |
1961 | uclamp_se_set(&p->uclamp_req[clamp_id], |
1962 | uclamp_none(clamp_id), false); | |
a87498ac | 1963 | } |
e8f14172 PB |
1964 | } |
1965 | ||
13685c4a QY |
1966 | static void uclamp_post_fork(struct task_struct *p) |
1967 | { | |
1968 | uclamp_update_util_min_rt_default(p); | |
1969 | } | |
1970 | ||
d81ae8aa QY |
1971 | static void __init init_uclamp_rq(struct rq *rq) |
1972 | { | |
1973 | enum uclamp_id clamp_id; | |
1974 | struct uclamp_rq *uc_rq = rq->uclamp; | |
1975 | ||
1976 | for_each_clamp_id(clamp_id) { | |
1977 | uc_rq[clamp_id] = (struct uclamp_rq) { | |
1978 | .value = uclamp_none(clamp_id) | |
1979 | }; | |
1980 | } | |
1981 | ||
315c4f88 | 1982 | rq->uclamp_flags = UCLAMP_FLAG_IDLE; |
d81ae8aa QY |
1983 | } |
1984 | ||
69842cba PB |
1985 | static void __init init_uclamp(void) |
1986 | { | |
e8f14172 | 1987 | struct uclamp_se uc_max = {}; |
0413d7f3 | 1988 | enum uclamp_id clamp_id; |
69842cba PB |
1989 | int cpu; |
1990 | ||
d81ae8aa QY |
1991 | for_each_possible_cpu(cpu) |
1992 | init_uclamp_rq(cpu_rq(cpu)); | |
69842cba | 1993 | |
69842cba | 1994 | for_each_clamp_id(clamp_id) { |
e8f14172 | 1995 | uclamp_se_set(&init_task.uclamp_req[clamp_id], |
a509a7cd | 1996 | uclamp_none(clamp_id), false); |
69842cba | 1997 | } |
e8f14172 PB |
1998 | |
1999 | /* System defaults allow max clamp values for both indexes */ | |
a509a7cd | 2000 | uclamp_se_set(&uc_max, uclamp_none(UCLAMP_MAX), false); |
2480c093 | 2001 | for_each_clamp_id(clamp_id) { |
e8f14172 | 2002 | uclamp_default[clamp_id] = uc_max; |
2480c093 PB |
2003 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
2004 | root_task_group.uclamp_req[clamp_id] = uc_max; | |
0b60ba2d | 2005 | root_task_group.uclamp[clamp_id] = uc_max; |
2480c093 PB |
2006 | #endif |
2007 | } | |
69842cba PB |
2008 | } |
2009 | ||
2010 | #else /* CONFIG_UCLAMP_TASK */ | |
2011 | static inline void uclamp_rq_inc(struct rq *rq, struct task_struct *p) { } | |
2012 | static inline void uclamp_rq_dec(struct rq *rq, struct task_struct *p) { } | |
a509a7cd PB |
2013 | static inline int uclamp_validate(struct task_struct *p, |
2014 | const struct sched_attr *attr) | |
2015 | { | |
2016 | return -EOPNOTSUPP; | |
2017 | } | |
2018 | static void __setscheduler_uclamp(struct task_struct *p, | |
2019 | const struct sched_attr *attr) { } | |
e8f14172 | 2020 | static inline void uclamp_fork(struct task_struct *p) { } |
13685c4a | 2021 | static inline void uclamp_post_fork(struct task_struct *p) { } |
69842cba PB |
2022 | static inline void init_uclamp(void) { } |
2023 | #endif /* CONFIG_UCLAMP_TASK */ | |
2024 | ||
a1dfb631 MT |
2025 | bool sched_task_on_rq(struct task_struct *p) |
2026 | { | |
2027 | return task_on_rq_queued(p); | |
2028 | } | |
2029 | ||
42a20f86 KC |
2030 | unsigned long get_wchan(struct task_struct *p) |
2031 | { | |
2032 | unsigned long ip = 0; | |
2033 | unsigned int state; | |
2034 | ||
2035 | if (!p || p == current) | |
2036 | return 0; | |
2037 | ||
2038 | /* Only get wchan if task is blocked and we can keep it that way. */ | |
2039 | raw_spin_lock_irq(&p->pi_lock); | |
2040 | state = READ_ONCE(p->__state); | |
2041 | smp_rmb(); /* see try_to_wake_up() */ | |
2042 | if (state != TASK_RUNNING && state != TASK_WAKING && !p->on_rq) | |
2043 | ip = __get_wchan(p); | |
2044 | raw_spin_unlock_irq(&p->pi_lock); | |
2045 | ||
2046 | return ip; | |
2047 | } | |
2048 | ||
1de64443 | 2049 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 2050 | { |
0a67d1ee PZ |
2051 | if (!(flags & ENQUEUE_NOCLOCK)) |
2052 | update_rq_clock(rq); | |
2053 | ||
eb414681 | 2054 | if (!(flags & ENQUEUE_RESTORE)) { |
4e29fb70 | 2055 | sched_info_enqueue(rq, p); |
eb414681 JW |
2056 | psi_enqueue(p, flags & ENQUEUE_WAKEUP); |
2057 | } | |
0a67d1ee | 2058 | |
69842cba | 2059 | uclamp_rq_inc(rq, p); |
371fd7e7 | 2060 | p->sched_class->enqueue_task(rq, p, flags); |
8a311c74 PZ |
2061 | |
2062 | if (sched_core_enabled(rq)) | |
2063 | sched_core_enqueue(rq, p); | |
71f8bd46 IM |
2064 | } |
2065 | ||
1de64443 | 2066 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 2067 | { |
8a311c74 | 2068 | if (sched_core_enabled(rq)) |
4feee7d1 | 2069 | sched_core_dequeue(rq, p, flags); |
8a311c74 | 2070 | |
0a67d1ee PZ |
2071 | if (!(flags & DEQUEUE_NOCLOCK)) |
2072 | update_rq_clock(rq); | |
2073 | ||
eb414681 | 2074 | if (!(flags & DEQUEUE_SAVE)) { |
4e29fb70 | 2075 | sched_info_dequeue(rq, p); |
eb414681 JW |
2076 | psi_dequeue(p, flags & DEQUEUE_SLEEP); |
2077 | } | |
0a67d1ee | 2078 | |
69842cba | 2079 | uclamp_rq_dec(rq, p); |
371fd7e7 | 2080 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
2081 | } |
2082 | ||
029632fb | 2083 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 2084 | { |
371fd7e7 | 2085 | enqueue_task(rq, p, flags); |
7dd77884 PZ |
2086 | |
2087 | p->on_rq = TASK_ON_RQ_QUEUED; | |
1e3c88bd PZ |
2088 | } |
2089 | ||
029632fb | 2090 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd | 2091 | { |
7dd77884 PZ |
2092 | p->on_rq = (flags & DEQUEUE_SLEEP) ? 0 : TASK_ON_RQ_MIGRATING; |
2093 | ||
371fd7e7 | 2094 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
2095 | } |
2096 | ||
f558c2b8 | 2097 | static inline int __normal_prio(int policy, int rt_prio, int nice) |
14531189 | 2098 | { |
f558c2b8 PZ |
2099 | int prio; |
2100 | ||
2101 | if (dl_policy(policy)) | |
2102 | prio = MAX_DL_PRIO - 1; | |
2103 | else if (rt_policy(policy)) | |
2104 | prio = MAX_RT_PRIO - 1 - rt_prio; | |
2105 | else | |
2106 | prio = NICE_TO_PRIO(nice); | |
2107 | ||
2108 | return prio; | |
14531189 IM |
2109 | } |
2110 | ||
b29739f9 IM |
2111 | /* |
2112 | * Calculate the expected normal priority: i.e. priority | |
2113 | * without taking RT-inheritance into account. Might be | |
2114 | * boosted by interactivity modifiers. Changes upon fork, | |
2115 | * setprio syscalls, and whenever the interactivity | |
2116 | * estimator recalculates. | |
2117 | */ | |
36c8b586 | 2118 | static inline int normal_prio(struct task_struct *p) |
b29739f9 | 2119 | { |
f558c2b8 | 2120 | return __normal_prio(p->policy, p->rt_priority, PRIO_TO_NICE(p->static_prio)); |
b29739f9 IM |
2121 | } |
2122 | ||
2123 | /* | |
2124 | * Calculate the current priority, i.e. the priority | |
2125 | * taken into account by the scheduler. This value might | |
2126 | * be boosted by RT tasks, or might be boosted by | |
2127 | * interactivity modifiers. Will be RT if the task got | |
2128 | * RT-boosted. If not then it returns p->normal_prio. | |
2129 | */ | |
36c8b586 | 2130 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2131 | { |
2132 | p->normal_prio = normal_prio(p); | |
2133 | /* | |
2134 | * If we are RT tasks or we were boosted to RT priority, | |
2135 | * keep the priority unchanged. Otherwise, update priority | |
2136 | * to the normal priority: | |
2137 | */ | |
2138 | if (!rt_prio(p->prio)) | |
2139 | return p->normal_prio; | |
2140 | return p->prio; | |
2141 | } | |
2142 | ||
1da177e4 LT |
2143 | /** |
2144 | * task_curr - is this task currently executing on a CPU? | |
2145 | * @p: the task in question. | |
e69f6186 YB |
2146 | * |
2147 | * Return: 1 if the task is currently executing. 0 otherwise. | |
1da177e4 | 2148 | */ |
36c8b586 | 2149 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2150 | { |
2151 | return cpu_curr(task_cpu(p)) == p; | |
2152 | } | |
2153 | ||
67dfa1b7 | 2154 | /* |
4c9a4bc8 PZ |
2155 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
2156 | * use the balance_callback list if you want balancing. | |
2157 | * | |
2158 | * this means any call to check_class_changed() must be followed by a call to | |
2159 | * balance_callback(). | |
67dfa1b7 | 2160 | */ |
cb469845 SR |
2161 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2162 | const struct sched_class *prev_class, | |
da7a735e | 2163 | int oldprio) |
cb469845 SR |
2164 | { |
2165 | if (prev_class != p->sched_class) { | |
2166 | if (prev_class->switched_from) | |
da7a735e | 2167 | prev_class->switched_from(rq, p); |
4c9a4bc8 | 2168 | |
da7a735e | 2169 | p->sched_class->switched_to(rq, p); |
2d3d891d | 2170 | } else if (oldprio != p->prio || dl_task(p)) |
da7a735e | 2171 | p->sched_class->prio_changed(rq, p, oldprio); |
cb469845 SR |
2172 | } |
2173 | ||
029632fb | 2174 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1e5a7405 | 2175 | { |
aa93cd53 | 2176 | if (p->sched_class == rq->curr->sched_class) |
1e5a7405 | 2177 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
546a3fee | 2178 | else if (sched_class_above(p->sched_class, rq->curr->sched_class)) |
aa93cd53 | 2179 | resched_curr(rq); |
1e5a7405 PZ |
2180 | |
2181 | /* | |
2182 | * A queue event has occurred, and we're going to schedule. In | |
2183 | * this case, we can save a useless back to back clock update. | |
2184 | */ | |
da0c1e65 | 2185 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
adcc8da8 | 2186 | rq_clock_skip_update(rq); |
1e5a7405 PZ |
2187 | } |
2188 | ||
1da177e4 | 2189 | #ifdef CONFIG_SMP |
175f0e25 | 2190 | |
af449901 PZ |
2191 | static void |
2192 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags); | |
2193 | ||
2194 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
2195 | const struct cpumask *new_mask, | |
2196 | u32 flags); | |
2197 | ||
2198 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
2199 | { | |
2200 | if (likely(!p->migration_disabled)) | |
2201 | return; | |
2202 | ||
2203 | if (p->cpus_ptr != &p->cpus_mask) | |
2204 | return; | |
2205 | ||
2206 | /* | |
2207 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
2208 | */ | |
2209 | __do_set_cpus_allowed(p, cpumask_of(rq->cpu), SCA_MIGRATE_DISABLE); | |
2210 | } | |
2211 | ||
2212 | void migrate_disable(void) | |
2213 | { | |
3015ef4b TG |
2214 | struct task_struct *p = current; |
2215 | ||
2216 | if (p->migration_disabled) { | |
2217 | p->migration_disabled++; | |
af449901 | 2218 | return; |
3015ef4b | 2219 | } |
af449901 | 2220 | |
3015ef4b TG |
2221 | preempt_disable(); |
2222 | this_rq()->nr_pinned++; | |
2223 | p->migration_disabled = 1; | |
2224 | preempt_enable(); | |
af449901 PZ |
2225 | } |
2226 | EXPORT_SYMBOL_GPL(migrate_disable); | |
2227 | ||
2228 | void migrate_enable(void) | |
2229 | { | |
2230 | struct task_struct *p = current; | |
2231 | ||
6d337eab PZ |
2232 | if (p->migration_disabled > 1) { |
2233 | p->migration_disabled--; | |
af449901 | 2234 | return; |
6d337eab | 2235 | } |
af449901 | 2236 | |
9d0df377 SAS |
2237 | if (WARN_ON_ONCE(!p->migration_disabled)) |
2238 | return; | |
2239 | ||
6d337eab PZ |
2240 | /* |
2241 | * Ensure stop_task runs either before or after this, and that | |
2242 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
2243 | */ | |
2244 | preempt_disable(); | |
2245 | if (p->cpus_ptr != &p->cpus_mask) | |
2246 | __set_cpus_allowed_ptr(p, &p->cpus_mask, SCA_MIGRATE_ENABLE); | |
2247 | /* | |
2248 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
2249 | * regular cpus_mask, otherwise things that race (eg. | |
2250 | * select_fallback_rq) get confused. | |
2251 | */ | |
af449901 | 2252 | barrier(); |
6d337eab | 2253 | p->migration_disabled = 0; |
3015ef4b | 2254 | this_rq()->nr_pinned--; |
6d337eab | 2255 | preempt_enable(); |
af449901 PZ |
2256 | } |
2257 | EXPORT_SYMBOL_GPL(migrate_enable); | |
2258 | ||
3015ef4b TG |
2259 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2260 | { | |
2261 | return rq->nr_pinned; | |
2262 | } | |
2263 | ||
175f0e25 | 2264 | /* |
bee98539 | 2265 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
2266 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
2267 | */ | |
2268 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
2269 | { | |
5ba2ffba | 2270 | /* When not in the task's cpumask, no point in looking further. */ |
3bd37062 | 2271 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
2272 | return false; |
2273 | ||
5ba2ffba PZ |
2274 | /* migrate_disabled() must be allowed to finish. */ |
2275 | if (is_migration_disabled(p)) | |
175f0e25 PZ |
2276 | return cpu_online(cpu); |
2277 | ||
5ba2ffba PZ |
2278 | /* Non kernel threads are not allowed during either online or offline. */ |
2279 | if (!(p->flags & PF_KTHREAD)) | |
9ae606bc | 2280 | return cpu_active(cpu) && task_cpu_possible(cpu, p); |
5ba2ffba PZ |
2281 | |
2282 | /* KTHREAD_IS_PER_CPU is always allowed. */ | |
2283 | if (kthread_is_per_cpu(p)) | |
2284 | return cpu_online(cpu); | |
2285 | ||
2286 | /* Regular kernel threads don't get to stay during offline. */ | |
b5c44773 | 2287 | if (cpu_dying(cpu)) |
5ba2ffba PZ |
2288 | return false; |
2289 | ||
2290 | /* But are allowed during online. */ | |
2291 | return cpu_online(cpu); | |
175f0e25 PZ |
2292 | } |
2293 | ||
5cc389bc PZ |
2294 | /* |
2295 | * This is how migration works: | |
2296 | * | |
2297 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
2298 | * stop_one_cpu(). | |
2299 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
2300 | * off the CPU) | |
2301 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
2302 | * 4) if it's in the wrong runqueue then the migration thread removes | |
2303 | * it and puts it into the right queue. | |
2304 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
2305 | * is done. | |
2306 | */ | |
2307 | ||
2308 | /* | |
2309 | * move_queued_task - move a queued task to new rq. | |
2310 | * | |
2311 | * Returns (locked) new rq. Old rq's lock is released. | |
2312 | */ | |
8a8c69c3 PZ |
2313 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
2314 | struct task_struct *p, int new_cpu) | |
5cc389bc | 2315 | { |
5cb9eaa3 | 2316 | lockdep_assert_rq_held(rq); |
5cc389bc | 2317 | |
58877d34 | 2318 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 2319 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 2320 | rq_unlock(rq, rf); |
5cc389bc PZ |
2321 | |
2322 | rq = cpu_rq(new_cpu); | |
2323 | ||
8a8c69c3 | 2324 | rq_lock(rq, rf); |
09348d75 | 2325 | WARN_ON_ONCE(task_cpu(p) != new_cpu); |
58877d34 | 2326 | activate_task(rq, p, 0); |
5cc389bc PZ |
2327 | check_preempt_curr(rq, p, 0); |
2328 | ||
2329 | return rq; | |
2330 | } | |
2331 | ||
2332 | struct migration_arg { | |
6d337eab PZ |
2333 | struct task_struct *task; |
2334 | int dest_cpu; | |
2335 | struct set_affinity_pending *pending; | |
2336 | }; | |
2337 | ||
50caf9c1 PZ |
2338 | /* |
2339 | * @refs: number of wait_for_completion() | |
2340 | * @stop_pending: is @stop_work in use | |
2341 | */ | |
6d337eab PZ |
2342 | struct set_affinity_pending { |
2343 | refcount_t refs; | |
9e81889c | 2344 | unsigned int stop_pending; |
6d337eab PZ |
2345 | struct completion done; |
2346 | struct cpu_stop_work stop_work; | |
2347 | struct migration_arg arg; | |
5cc389bc PZ |
2348 | }; |
2349 | ||
2350 | /* | |
d1ccc66d | 2351 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
2352 | * this because either it can't run here any more (set_cpus_allowed() |
2353 | * away from this CPU, or CPU going down), or because we're | |
2354 | * attempting to rebalance this task on exec (sched_exec). | |
2355 | * | |
2356 | * So we race with normal scheduler movements, but that's OK, as long | |
2357 | * as the task is no longer on this CPU. | |
5cc389bc | 2358 | */ |
8a8c69c3 PZ |
2359 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
2360 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 2361 | { |
5cc389bc | 2362 | /* Affinity changed (again). */ |
175f0e25 | 2363 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 2364 | return rq; |
5cc389bc | 2365 | |
15ff991e | 2366 | update_rq_clock(rq); |
8a8c69c3 | 2367 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
2368 | |
2369 | return rq; | |
5cc389bc PZ |
2370 | } |
2371 | ||
2372 | /* | |
2373 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
2374 | * and performs thread migration by bumping thread off CPU then | |
2375 | * 'pushing' onto another runqueue. | |
2376 | */ | |
2377 | static int migration_cpu_stop(void *data) | |
2378 | { | |
2379 | struct migration_arg *arg = data; | |
c20cf065 | 2380 | struct set_affinity_pending *pending = arg->pending; |
5e16bbc2 PZ |
2381 | struct task_struct *p = arg->task; |
2382 | struct rq *rq = this_rq(); | |
6d337eab | 2383 | bool complete = false; |
8a8c69c3 | 2384 | struct rq_flags rf; |
5cc389bc PZ |
2385 | |
2386 | /* | |
d1ccc66d IM |
2387 | * The original target CPU might have gone down and we might |
2388 | * be on another CPU but it doesn't matter. | |
5cc389bc | 2389 | */ |
6d337eab | 2390 | local_irq_save(rf.flags); |
5cc389bc PZ |
2391 | /* |
2392 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 2393 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
2394 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
2395 | */ | |
16bf5a5e | 2396 | flush_smp_call_function_queue(); |
5e16bbc2 PZ |
2397 | |
2398 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 2399 | rq_lock(rq, &rf); |
6d337eab | 2400 | |
e140749c VS |
2401 | /* |
2402 | * If we were passed a pending, then ->stop_pending was set, thus | |
2403 | * p->migration_pending must have remained stable. | |
2404 | */ | |
2405 | WARN_ON_ONCE(pending && pending != p->migration_pending); | |
2406 | ||
5e16bbc2 PZ |
2407 | /* |
2408 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
2409 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
2410 | * we're holding p->pi_lock. | |
2411 | */ | |
bf89a304 | 2412 | if (task_rq(p) == rq) { |
6d337eab PZ |
2413 | if (is_migration_disabled(p)) |
2414 | goto out; | |
2415 | ||
2416 | if (pending) { | |
e140749c | 2417 | p->migration_pending = NULL; |
6d337eab | 2418 | complete = true; |
6d337eab | 2419 | |
3f1bc119 PZ |
2420 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) |
2421 | goto out; | |
3f1bc119 | 2422 | } |
6d337eab | 2423 | |
bf89a304 | 2424 | if (task_on_rq_queued(p)) |
475ea6c6 | 2425 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 | 2426 | else |
475ea6c6 | 2427 | p->wake_cpu = arg->dest_cpu; |
6d337eab | 2428 | |
3f1bc119 PZ |
2429 | /* |
2430 | * XXX __migrate_task() can fail, at which point we might end | |
2431 | * up running on a dodgy CPU, AFAICT this can only happen | |
2432 | * during CPU hotplug, at which point we'll get pushed out | |
2433 | * anyway, so it's probably not a big deal. | |
2434 | */ | |
2435 | ||
c20cf065 | 2436 | } else if (pending) { |
6d337eab PZ |
2437 | /* |
2438 | * This happens when we get migrated between migrate_enable()'s | |
2439 | * preempt_enable() and scheduling the stopper task. At that | |
2440 | * point we're a regular task again and not current anymore. | |
2441 | * | |
2442 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
2443 | * more likely. | |
2444 | */ | |
2445 | ||
d707faa6 VS |
2446 | /* |
2447 | * The task moved before the stopper got to run. We're holding | |
2448 | * ->pi_lock, so the allowed mask is stable - if it got | |
2449 | * somewhere allowed, we're done. | |
2450 | */ | |
c20cf065 | 2451 | if (cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { |
e140749c | 2452 | p->migration_pending = NULL; |
d707faa6 VS |
2453 | complete = true; |
2454 | goto out; | |
2455 | } | |
2456 | ||
6d337eab PZ |
2457 | /* |
2458 | * When migrate_enable() hits a rq mis-match we can't reliably | |
2459 | * determine is_migration_disabled() and so have to chase after | |
2460 | * it. | |
2461 | */ | |
9e81889c | 2462 | WARN_ON_ONCE(!pending->stop_pending); |
6d337eab PZ |
2463 | task_rq_unlock(rq, p, &rf); |
2464 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
2465 | &pending->arg, &pending->stop_work); | |
2466 | return 0; | |
bf89a304 | 2467 | } |
6d337eab | 2468 | out: |
9e81889c PZ |
2469 | if (pending) |
2470 | pending->stop_pending = false; | |
6d337eab PZ |
2471 | task_rq_unlock(rq, p, &rf); |
2472 | ||
2473 | if (complete) | |
2474 | complete_all(&pending->done); | |
2475 | ||
5cc389bc PZ |
2476 | return 0; |
2477 | } | |
2478 | ||
a7c81556 PZ |
2479 | int push_cpu_stop(void *arg) |
2480 | { | |
2481 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2482 | struct task_struct *p = arg; | |
2483 | ||
2484 | raw_spin_lock_irq(&p->pi_lock); | |
5cb9eaa3 | 2485 | raw_spin_rq_lock(rq); |
a7c81556 PZ |
2486 | |
2487 | if (task_rq(p) != rq) | |
2488 | goto out_unlock; | |
2489 | ||
2490 | if (is_migration_disabled(p)) { | |
2491 | p->migration_flags |= MDF_PUSH; | |
2492 | goto out_unlock; | |
2493 | } | |
2494 | ||
2495 | p->migration_flags &= ~MDF_PUSH; | |
2496 | ||
2497 | if (p->sched_class->find_lock_rq) | |
2498 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2499 | |
a7c81556 PZ |
2500 | if (!lowest_rq) |
2501 | goto out_unlock; | |
2502 | ||
2503 | // XXX validate p is still the highest prio task | |
2504 | if (task_rq(p) == rq) { | |
2505 | deactivate_task(rq, p, 0); | |
2506 | set_task_cpu(p, lowest_rq->cpu); | |
2507 | activate_task(lowest_rq, p, 0); | |
2508 | resched_curr(lowest_rq); | |
2509 | } | |
2510 | ||
2511 | double_unlock_balance(rq, lowest_rq); | |
2512 | ||
2513 | out_unlock: | |
2514 | rq->push_busy = false; | |
5cb9eaa3 | 2515 | raw_spin_rq_unlock(rq); |
a7c81556 PZ |
2516 | raw_spin_unlock_irq(&p->pi_lock); |
2517 | ||
2518 | put_task_struct(p); | |
5cc389bc PZ |
2519 | return 0; |
2520 | } | |
2521 | ||
c5b28038 PZ |
2522 | /* |
2523 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2524 | * actually call this function. | |
2525 | */ | |
9cfc3e18 | 2526 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask, u32 flags) |
5cc389bc | 2527 | { |
af449901 PZ |
2528 | if (flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2529 | p->cpus_ptr = new_mask; | |
2530 | return; | |
2531 | } | |
2532 | ||
3bd37062 | 2533 | cpumask_copy(&p->cpus_mask, new_mask); |
5cc389bc PZ |
2534 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
2535 | } | |
2536 | ||
9cfc3e18 PZ |
2537 | static void |
2538 | __do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask, u32 flags) | |
c5b28038 | 2539 | { |
6c37067e PZ |
2540 | struct rq *rq = task_rq(p); |
2541 | bool queued, running; | |
2542 | ||
af449901 PZ |
2543 | /* |
2544 | * This here violates the locking rules for affinity, since we're only | |
2545 | * supposed to change these variables while holding both rq->lock and | |
2546 | * p->pi_lock. | |
2547 | * | |
2548 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2549 | * accesses these variables under p->pi_lock and only does so after | |
2550 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2551 | * before finish_task(). | |
2552 | * | |
2553 | * XXX do further audits, this smells like something putrid. | |
2554 | */ | |
2555 | if (flags & SCA_MIGRATE_DISABLE) | |
2556 | SCHED_WARN_ON(!p->on_cpu); | |
2557 | else | |
2558 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2559 | |
2560 | queued = task_on_rq_queued(p); | |
2561 | running = task_current(rq, p); | |
2562 | ||
2563 | if (queued) { | |
2564 | /* | |
2565 | * Because __kthread_bind() calls this on blocked tasks without | |
2566 | * holding rq->lock. | |
2567 | */ | |
5cb9eaa3 | 2568 | lockdep_assert_rq_held(rq); |
7a57f32a | 2569 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2570 | } |
2571 | if (running) | |
2572 | put_prev_task(rq, p); | |
2573 | ||
9cfc3e18 | 2574 | p->sched_class->set_cpus_allowed(p, new_mask, flags); |
6c37067e | 2575 | |
6c37067e | 2576 | if (queued) |
7134b3e9 | 2577 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2578 | if (running) |
03b7fad1 | 2579 | set_next_task(rq, p); |
c5b28038 PZ |
2580 | } |
2581 | ||
9cfc3e18 PZ |
2582 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2583 | { | |
2584 | __do_set_cpus_allowed(p, new_mask, 0); | |
2585 | } | |
2586 | ||
b90ca8ba WD |
2587 | int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, |
2588 | int node) | |
2589 | { | |
2590 | if (!src->user_cpus_ptr) | |
2591 | return 0; | |
2592 | ||
2593 | dst->user_cpus_ptr = kmalloc_node(cpumask_size(), GFP_KERNEL, node); | |
2594 | if (!dst->user_cpus_ptr) | |
2595 | return -ENOMEM; | |
2596 | ||
2597 | cpumask_copy(dst->user_cpus_ptr, src->user_cpus_ptr); | |
2598 | return 0; | |
2599 | } | |
2600 | ||
07ec77a1 WD |
2601 | static inline struct cpumask *clear_user_cpus_ptr(struct task_struct *p) |
2602 | { | |
2603 | struct cpumask *user_mask = NULL; | |
2604 | ||
2605 | swap(p->user_cpus_ptr, user_mask); | |
2606 | ||
2607 | return user_mask; | |
2608 | } | |
2609 | ||
b90ca8ba WD |
2610 | void release_user_cpus_ptr(struct task_struct *p) |
2611 | { | |
07ec77a1 | 2612 | kfree(clear_user_cpus_ptr(p)); |
b90ca8ba WD |
2613 | } |
2614 | ||
6d337eab | 2615 | /* |
c777d847 VS |
2616 | * This function is wildly self concurrent; here be dragons. |
2617 | * | |
2618 | * | |
2619 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2620 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2621 | * running, we have to kick it out using the CPU stopper. | |
2622 | * | |
2623 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2624 | * Consider: | |
2625 | * | |
2626 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2627 | * | |
2628 | * P0@CPU0 P1 | |
2629 | * | |
2630 | * migrate_disable(); | |
2631 | * <preempted> | |
2632 | * set_cpus_allowed_ptr(P0, [1]); | |
2633 | * | |
2634 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2635 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2636 | * This means we need the following scheme: | |
2637 | * | |
2638 | * P0@CPU0 P1 | |
2639 | * | |
2640 | * migrate_disable(); | |
2641 | * <preempted> | |
2642 | * set_cpus_allowed_ptr(P0, [1]); | |
2643 | * <blocks> | |
2644 | * <resumes> | |
2645 | * migrate_enable(); | |
2646 | * __set_cpus_allowed_ptr(); | |
2647 | * <wakes local stopper> | |
2648 | * `--> <woken on migration completion> | |
2649 | * | |
2650 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2651 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2652 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2653 | * should come into effect at the end of the Migrate-Disable region is the last | |
2654 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2655 | * but we still need to properly signal those waiting tasks at the appropriate | |
2656 | * moment. | |
2657 | * | |
2658 | * This is implemented using struct set_affinity_pending. The first | |
2659 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2660 | * setup an instance of that struct and install it on the targeted task_struct. | |
2661 | * Any and all further callers will reuse that instance. Those then wait for | |
2662 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2663 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2664 | * | |
2665 | * | |
2666 | * (1) In the cases covered above. There is one more where the completion is | |
2667 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
e140749c VS |
2668 | * occurs after the stopper bailed out due to the targeted task still being |
2669 | * Migrate-Disable. Consider: | |
c777d847 VS |
2670 | * |
2671 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2672 | * | |
e140749c VS |
2673 | * CPU0 P1 P2 |
2674 | * <P0> | |
2675 | * migrate_disable(); | |
2676 | * <preempted> | |
c777d847 VS |
2677 | * set_cpus_allowed_ptr(P0, [1]); |
2678 | * <blocks> | |
e140749c VS |
2679 | * <migration/0> |
2680 | * migration_cpu_stop() | |
2681 | * is_migration_disabled() | |
2682 | * <bails> | |
c777d847 VS |
2683 | * set_cpus_allowed_ptr(P0, [0, 1]); |
2684 | * <signal completion> | |
2685 | * <awakes> | |
2686 | * | |
2687 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2688 | * pending affinity completion is preceded by an uninstallation of | |
2689 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2690 | */ |
2691 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2692 | int dest_cpu, unsigned int flags) | |
2693 | { | |
2694 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
9e81889c | 2695 | bool stop_pending, complete = false; |
6d337eab PZ |
2696 | |
2697 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2698 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2699 | struct task_struct *push_task = NULL; |
2700 | ||
2701 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2702 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2703 | rq->push_busy = true; | |
2704 | push_task = get_task_struct(p); | |
2705 | } | |
2706 | ||
50caf9c1 PZ |
2707 | /* |
2708 | * If there are pending waiters, but no pending stop_work, | |
2709 | * then complete now. | |
2710 | */ | |
6d337eab | 2711 | pending = p->migration_pending; |
50caf9c1 | 2712 | if (pending && !pending->stop_pending) { |
6d337eab PZ |
2713 | p->migration_pending = NULL; |
2714 | complete = true; | |
2715 | } | |
50caf9c1 | 2716 | |
6d337eab PZ |
2717 | task_rq_unlock(rq, p, rf); |
2718 | ||
a7c81556 PZ |
2719 | if (push_task) { |
2720 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2721 | p, &rq->push_work); | |
2722 | } | |
2723 | ||
6d337eab | 2724 | if (complete) |
50caf9c1 | 2725 | complete_all(&pending->done); |
6d337eab PZ |
2726 | |
2727 | return 0; | |
2728 | } | |
2729 | ||
2730 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2731 | /* serialized by p->pi_lock */ | |
2732 | if (!p->migration_pending) { | |
c777d847 | 2733 | /* Install the request */ |
6d337eab PZ |
2734 | refcount_set(&my_pending.refs, 1); |
2735 | init_completion(&my_pending.done); | |
8a6edb52 PZ |
2736 | my_pending.arg = (struct migration_arg) { |
2737 | .task = p, | |
475ea6c6 | 2738 | .dest_cpu = dest_cpu, |
8a6edb52 PZ |
2739 | .pending = &my_pending, |
2740 | }; | |
2741 | ||
6d337eab PZ |
2742 | p->migration_pending = &my_pending; |
2743 | } else { | |
2744 | pending = p->migration_pending; | |
2745 | refcount_inc(&pending->refs); | |
475ea6c6 VS |
2746 | /* |
2747 | * Affinity has changed, but we've already installed a | |
2748 | * pending. migration_cpu_stop() *must* see this, else | |
2749 | * we risk a completion of the pending despite having a | |
2750 | * task on a disallowed CPU. | |
2751 | * | |
2752 | * Serialized by p->pi_lock, so this is safe. | |
2753 | */ | |
2754 | pending->arg.dest_cpu = dest_cpu; | |
6d337eab PZ |
2755 | } |
2756 | } | |
2757 | pending = p->migration_pending; | |
2758 | /* | |
2759 | * - !MIGRATE_ENABLE: | |
2760 | * we'll have installed a pending if there wasn't one already. | |
2761 | * | |
2762 | * - MIGRATE_ENABLE: | |
2763 | * we're here because the current CPU isn't matching anymore, | |
2764 | * the only way that can happen is because of a concurrent | |
2765 | * set_cpus_allowed_ptr() call, which should then still be | |
2766 | * pending completion. | |
2767 | * | |
2768 | * Either way, we really should have a @pending here. | |
2769 | */ | |
2770 | if (WARN_ON_ONCE(!pending)) { | |
2771 | task_rq_unlock(rq, p, rf); | |
2772 | return -EINVAL; | |
2773 | } | |
2774 | ||
0b9d46fc | 2775 | if (task_on_cpu(rq, p) || READ_ONCE(p->__state) == TASK_WAKING) { |
c777d847 | 2776 | /* |
58b1a450 PZ |
2777 | * MIGRATE_ENABLE gets here because 'p == current', but for |
2778 | * anything else we cannot do is_migration_disabled(), punt | |
2779 | * and have the stopper function handle it all race-free. | |
c777d847 | 2780 | */ |
9e81889c PZ |
2781 | stop_pending = pending->stop_pending; |
2782 | if (!stop_pending) | |
2783 | pending->stop_pending = true; | |
58b1a450 | 2784 | |
58b1a450 PZ |
2785 | if (flags & SCA_MIGRATE_ENABLE) |
2786 | p->migration_flags &= ~MDF_PUSH; | |
50caf9c1 | 2787 | |
6d337eab | 2788 | task_rq_unlock(rq, p, rf); |
8a6edb52 | 2789 | |
9e81889c PZ |
2790 | if (!stop_pending) { |
2791 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2792 | &pending->arg, &pending->stop_work); | |
2793 | } | |
6d337eab | 2794 | |
58b1a450 PZ |
2795 | if (flags & SCA_MIGRATE_ENABLE) |
2796 | return 0; | |
6d337eab PZ |
2797 | } else { |
2798 | ||
2799 | if (!is_migration_disabled(p)) { | |
2800 | if (task_on_rq_queued(p)) | |
2801 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2802 | ||
50caf9c1 PZ |
2803 | if (!pending->stop_pending) { |
2804 | p->migration_pending = NULL; | |
2805 | complete = true; | |
2806 | } | |
6d337eab PZ |
2807 | } |
2808 | task_rq_unlock(rq, p, rf); | |
2809 | ||
6d337eab PZ |
2810 | if (complete) |
2811 | complete_all(&pending->done); | |
2812 | } | |
2813 | ||
2814 | wait_for_completion(&pending->done); | |
2815 | ||
2816 | if (refcount_dec_and_test(&pending->refs)) | |
50caf9c1 | 2817 | wake_up_var(&pending->refs); /* No UaF, just an address */ |
6d337eab | 2818 | |
c777d847 VS |
2819 | /* |
2820 | * Block the original owner of &pending until all subsequent callers | |
2821 | * have seen the completion and decremented the refcount | |
2822 | */ | |
6d337eab PZ |
2823 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2824 | ||
50caf9c1 PZ |
2825 | /* ARGH */ |
2826 | WARN_ON_ONCE(my_pending.stop_pending); | |
2827 | ||
6d337eab PZ |
2828 | return 0; |
2829 | } | |
2830 | ||
5cc389bc | 2831 | /* |
07ec77a1 | 2832 | * Called with both p->pi_lock and rq->lock held; drops both before returning. |
5cc389bc | 2833 | */ |
07ec77a1 WD |
2834 | static int __set_cpus_allowed_ptr_locked(struct task_struct *p, |
2835 | const struct cpumask *new_mask, | |
2836 | u32 flags, | |
2837 | struct rq *rq, | |
2838 | struct rq_flags *rf) | |
2839 | __releases(rq->lock) | |
2840 | __releases(p->pi_lock) | |
5cc389bc | 2841 | { |
234a503e | 2842 | const struct cpumask *cpu_allowed_mask = task_cpu_possible_mask(p); |
e9d867a6 | 2843 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
234a503e | 2844 | bool kthread = p->flags & PF_KTHREAD; |
07ec77a1 | 2845 | struct cpumask *user_mask = NULL; |
5cc389bc PZ |
2846 | unsigned int dest_cpu; |
2847 | int ret = 0; | |
2848 | ||
a499c3ea | 2849 | update_rq_clock(rq); |
5cc389bc | 2850 | |
234a503e | 2851 | if (kthread || is_migration_disabled(p)) { |
e9d867a6 | 2852 | /* |
741ba80f PZ |
2853 | * Kernel threads are allowed on online && !active CPUs, |
2854 | * however, during cpu-hot-unplug, even these might get pushed | |
2855 | * away if not KTHREAD_IS_PER_CPU. | |
af449901 PZ |
2856 | * |
2857 | * Specifically, migration_disabled() tasks must not fail the | |
2858 | * cpumask_any_and_distribute() pick below, esp. so on | |
2859 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2860 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2861 | */ |
2862 | cpu_valid_mask = cpu_online_mask; | |
2863 | } | |
2864 | ||
234a503e WD |
2865 | if (!kthread && !cpumask_subset(new_mask, cpu_allowed_mask)) { |
2866 | ret = -EINVAL; | |
2867 | goto out; | |
2868 | } | |
2869 | ||
25834c73 PZ |
2870 | /* |
2871 | * Must re-check here, to close a race against __kthread_bind(), | |
2872 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2873 | */ | |
9cfc3e18 | 2874 | if ((flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2875 | ret = -EINVAL; |
2876 | goto out; | |
2877 | } | |
2878 | ||
885b3ba4 VS |
2879 | if (!(flags & SCA_MIGRATE_ENABLE)) { |
2880 | if (cpumask_equal(&p->cpus_mask, new_mask)) | |
2881 | goto out; | |
2882 | ||
2883 | if (WARN_ON_ONCE(p == current && | |
2884 | is_migration_disabled(p) && | |
2885 | !cpumask_test_cpu(task_cpu(p), new_mask))) { | |
2886 | ret = -EBUSY; | |
2887 | goto out; | |
2888 | } | |
2889 | } | |
5cc389bc | 2890 | |
46a87b38 PT |
2891 | /* |
2892 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2893 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2894 | * immediately required to distribute the tasks within their new mask. | |
2895 | */ | |
2896 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, new_mask); | |
714e501e | 2897 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2898 | ret = -EINVAL; |
2899 | goto out; | |
2900 | } | |
2901 | ||
9cfc3e18 | 2902 | __do_set_cpus_allowed(p, new_mask, flags); |
5cc389bc | 2903 | |
07ec77a1 WD |
2904 | if (flags & SCA_USER) |
2905 | user_mask = clear_user_cpus_ptr(p); | |
2906 | ||
2907 | ret = affine_move_task(rq, p, rf, dest_cpu, flags); | |
2908 | ||
2909 | kfree(user_mask); | |
2910 | ||
2911 | return ret; | |
5cc389bc | 2912 | |
5cc389bc | 2913 | out: |
07ec77a1 | 2914 | task_rq_unlock(rq, p, rf); |
5cc389bc PZ |
2915 | |
2916 | return ret; | |
2917 | } | |
25834c73 | 2918 | |
07ec77a1 WD |
2919 | /* |
2920 | * Change a given task's CPU affinity. Migrate the thread to a | |
2921 | * proper CPU and schedule it away if the CPU it's executing on | |
2922 | * is removed from the allowed bitmask. | |
2923 | * | |
2924 | * NOTE: the caller must have a valid reference to the task, the | |
2925 | * task must not exit() & deallocate itself prematurely. The | |
2926 | * call is not atomic; no spinlocks may be held. | |
2927 | */ | |
2928 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
2929 | const struct cpumask *new_mask, u32 flags) | |
2930 | { | |
2931 | struct rq_flags rf; | |
2932 | struct rq *rq; | |
2933 | ||
2934 | rq = task_rq_lock(p, &rf); | |
2935 | return __set_cpus_allowed_ptr_locked(p, new_mask, flags, rq, &rf); | |
2936 | } | |
2937 | ||
25834c73 PZ |
2938 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
2939 | { | |
9cfc3e18 | 2940 | return __set_cpus_allowed_ptr(p, new_mask, 0); |
25834c73 | 2941 | } |
5cc389bc PZ |
2942 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2943 | ||
07ec77a1 WD |
2944 | /* |
2945 | * Change a given task's CPU affinity to the intersection of its current | |
2946 | * affinity mask and @subset_mask, writing the resulting mask to @new_mask | |
2947 | * and pointing @p->user_cpus_ptr to a copy of the old mask. | |
2948 | * If the resulting mask is empty, leave the affinity unchanged and return | |
2949 | * -EINVAL. | |
2950 | */ | |
2951 | static int restrict_cpus_allowed_ptr(struct task_struct *p, | |
2952 | struct cpumask *new_mask, | |
2953 | const struct cpumask *subset_mask) | |
2954 | { | |
2955 | struct cpumask *user_mask = NULL; | |
2956 | struct rq_flags rf; | |
2957 | struct rq *rq; | |
2958 | int err; | |
2959 | ||
2960 | if (!p->user_cpus_ptr) { | |
2961 | user_mask = kmalloc(cpumask_size(), GFP_KERNEL); | |
2962 | if (!user_mask) | |
2963 | return -ENOMEM; | |
2964 | } | |
2965 | ||
2966 | rq = task_rq_lock(p, &rf); | |
2967 | ||
2968 | /* | |
2969 | * Forcefully restricting the affinity of a deadline task is | |
2970 | * likely to cause problems, so fail and noisily override the | |
2971 | * mask entirely. | |
2972 | */ | |
2973 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { | |
2974 | err = -EPERM; | |
2975 | goto err_unlock; | |
2976 | } | |
2977 | ||
2978 | if (!cpumask_and(new_mask, &p->cpus_mask, subset_mask)) { | |
2979 | err = -EINVAL; | |
2980 | goto err_unlock; | |
2981 | } | |
2982 | ||
2983 | /* | |
2984 | * We're about to butcher the task affinity, so keep track of what | |
2985 | * the user asked for in case we're able to restore it later on. | |
2986 | */ | |
2987 | if (user_mask) { | |
2988 | cpumask_copy(user_mask, p->cpus_ptr); | |
2989 | p->user_cpus_ptr = user_mask; | |
2990 | } | |
2991 | ||
2992 | return __set_cpus_allowed_ptr_locked(p, new_mask, 0, rq, &rf); | |
2993 | ||
2994 | err_unlock: | |
2995 | task_rq_unlock(rq, p, &rf); | |
2996 | kfree(user_mask); | |
2997 | return err; | |
2998 | } | |
2999 | ||
3000 | /* | |
3001 | * Restrict the CPU affinity of task @p so that it is a subset of | |
3002 | * task_cpu_possible_mask() and point @p->user_cpu_ptr to a copy of the | |
3003 | * old affinity mask. If the resulting mask is empty, we warn and walk | |
3004 | * up the cpuset hierarchy until we find a suitable mask. | |
3005 | */ | |
3006 | void force_compatible_cpus_allowed_ptr(struct task_struct *p) | |
3007 | { | |
3008 | cpumask_var_t new_mask; | |
3009 | const struct cpumask *override_mask = task_cpu_possible_mask(p); | |
3010 | ||
3011 | alloc_cpumask_var(&new_mask, GFP_KERNEL); | |
3012 | ||
3013 | /* | |
3014 | * __migrate_task() can fail silently in the face of concurrent | |
3015 | * offlining of the chosen destination CPU, so take the hotplug | |
3016 | * lock to ensure that the migration succeeds. | |
3017 | */ | |
3018 | cpus_read_lock(); | |
3019 | if (!cpumask_available(new_mask)) | |
3020 | goto out_set_mask; | |
3021 | ||
3022 | if (!restrict_cpus_allowed_ptr(p, new_mask, override_mask)) | |
3023 | goto out_free_mask; | |
3024 | ||
3025 | /* | |
3026 | * We failed to find a valid subset of the affinity mask for the | |
3027 | * task, so override it based on its cpuset hierarchy. | |
3028 | */ | |
3029 | cpuset_cpus_allowed(p, new_mask); | |
3030 | override_mask = new_mask; | |
3031 | ||
3032 | out_set_mask: | |
3033 | if (printk_ratelimit()) { | |
3034 | printk_deferred("Overriding affinity for process %d (%s) to CPUs %*pbl\n", | |
3035 | task_pid_nr(p), p->comm, | |
3036 | cpumask_pr_args(override_mask)); | |
3037 | } | |
3038 | ||
3039 | WARN_ON(set_cpus_allowed_ptr(p, override_mask)); | |
3040 | out_free_mask: | |
3041 | cpus_read_unlock(); | |
3042 | free_cpumask_var(new_mask); | |
3043 | } | |
3044 | ||
3045 | static int | |
3046 | __sched_setaffinity(struct task_struct *p, const struct cpumask *mask); | |
3047 | ||
3048 | /* | |
3049 | * Restore the affinity of a task @p which was previously restricted by a | |
3050 | * call to force_compatible_cpus_allowed_ptr(). This will clear (and free) | |
3051 | * @p->user_cpus_ptr. | |
3052 | * | |
3053 | * It is the caller's responsibility to serialise this with any calls to | |
3054 | * force_compatible_cpus_allowed_ptr(@p). | |
3055 | */ | |
3056 | void relax_compatible_cpus_allowed_ptr(struct task_struct *p) | |
3057 | { | |
3058 | struct cpumask *user_mask = p->user_cpus_ptr; | |
3059 | unsigned long flags; | |
3060 | ||
3061 | /* | |
3062 | * Try to restore the old affinity mask. If this fails, then | |
3063 | * we free the mask explicitly to avoid it being inherited across | |
3064 | * a subsequent fork(). | |
3065 | */ | |
3066 | if (!user_mask || !__sched_setaffinity(p, user_mask)) | |
3067 | return; | |
3068 | ||
3069 | raw_spin_lock_irqsave(&p->pi_lock, flags); | |
3070 | user_mask = clear_user_cpus_ptr(p); | |
3071 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
3072 | ||
3073 | kfree(user_mask); | |
3074 | } | |
3075 | ||
dd41f596 | 3076 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 3077 | { |
e2912009 | 3078 | #ifdef CONFIG_SCHED_DEBUG |
2f064a59 PZ |
3079 | unsigned int state = READ_ONCE(p->__state); |
3080 | ||
e2912009 PZ |
3081 | /* |
3082 | * We should never call set_task_cpu() on a blocked task, | |
3083 | * ttwu() will sort out the placement. | |
3084 | */ | |
2f064a59 | 3085 | WARN_ON_ONCE(state != TASK_RUNNING && state != TASK_WAKING && !p->on_rq); |
0122ec5b | 3086 | |
3ea94de1 JP |
3087 | /* |
3088 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
3089 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
3090 | * time relying on p->on_rq. | |
3091 | */ | |
2f064a59 | 3092 | WARN_ON_ONCE(state == TASK_RUNNING && |
3ea94de1 JP |
3093 | p->sched_class == &fair_sched_class && |
3094 | (p->on_rq && !task_on_rq_migrating(p))); | |
3095 | ||
0122ec5b | 3096 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
3097 | /* |
3098 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
3099 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
3100 | * | |
3101 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 3102 | * see task_group(). |
6c6c54e1 PZ |
3103 | * |
3104 | * Furthermore, all task_rq users should acquire both locks, see | |
3105 | * task_rq_lock(). | |
3106 | */ | |
0122ec5b | 3107 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
9ef7e7e3 | 3108 | lockdep_is_held(__rq_lockp(task_rq(p))))); |
0122ec5b | 3109 | #endif |
4ff9083b PZ |
3110 | /* |
3111 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
3112 | */ | |
3113 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
3114 | |
3115 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
3116 | #endif |
3117 | ||
de1d7286 | 3118 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 3119 | |
0c69774e | 3120 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 3121 | if (p->sched_class->migrate_task_rq) |
1327237a | 3122 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 3123 | p->se.nr_migrations++; |
d7822b1e | 3124 | rseq_migrate(p); |
ff303e66 | 3125 | perf_event_task_migrate(p); |
0c69774e | 3126 | } |
dd41f596 IM |
3127 | |
3128 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
3129 | } |
3130 | ||
0ad4e3df | 3131 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
3132 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
3133 | { | |
da0c1e65 | 3134 | if (task_on_rq_queued(p)) { |
ac66f547 | 3135 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 3136 | struct rq_flags srf, drf; |
ac66f547 PZ |
3137 | |
3138 | src_rq = task_rq(p); | |
3139 | dst_rq = cpu_rq(cpu); | |
3140 | ||
8a8c69c3 PZ |
3141 | rq_pin_lock(src_rq, &srf); |
3142 | rq_pin_lock(dst_rq, &drf); | |
3143 | ||
ac66f547 PZ |
3144 | deactivate_task(src_rq, p, 0); |
3145 | set_task_cpu(p, cpu); | |
3146 | activate_task(dst_rq, p, 0); | |
3147 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
3148 | |
3149 | rq_unpin_lock(dst_rq, &drf); | |
3150 | rq_unpin_lock(src_rq, &srf); | |
3151 | ||
ac66f547 PZ |
3152 | } else { |
3153 | /* | |
3154 | * Task isn't running anymore; make it appear like we migrated | |
3155 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 3156 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
3157 | */ |
3158 | p->wake_cpu = cpu; | |
3159 | } | |
3160 | } | |
3161 | ||
3162 | struct migration_swap_arg { | |
3163 | struct task_struct *src_task, *dst_task; | |
3164 | int src_cpu, dst_cpu; | |
3165 | }; | |
3166 | ||
3167 | static int migrate_swap_stop(void *data) | |
3168 | { | |
3169 | struct migration_swap_arg *arg = data; | |
3170 | struct rq *src_rq, *dst_rq; | |
3171 | int ret = -EAGAIN; | |
3172 | ||
62694cd5 PZ |
3173 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
3174 | return -EAGAIN; | |
3175 | ||
ac66f547 PZ |
3176 | src_rq = cpu_rq(arg->src_cpu); |
3177 | dst_rq = cpu_rq(arg->dst_cpu); | |
3178 | ||
74602315 PZ |
3179 | double_raw_lock(&arg->src_task->pi_lock, |
3180 | &arg->dst_task->pi_lock); | |
ac66f547 | 3181 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 3182 | |
ac66f547 PZ |
3183 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
3184 | goto unlock; | |
3185 | ||
3186 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
3187 | goto unlock; | |
3188 | ||
3bd37062 | 3189 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
3190 | goto unlock; |
3191 | ||
3bd37062 | 3192 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
3193 | goto unlock; |
3194 | ||
3195 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
3196 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
3197 | ||
3198 | ret = 0; | |
3199 | ||
3200 | unlock: | |
3201 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
3202 | raw_spin_unlock(&arg->dst_task->pi_lock); |
3203 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
3204 | |
3205 | return ret; | |
3206 | } | |
3207 | ||
3208 | /* | |
3209 | * Cross migrate two tasks | |
3210 | */ | |
0ad4e3df SD |
3211 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
3212 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
3213 | { |
3214 | struct migration_swap_arg arg; | |
3215 | int ret = -EINVAL; | |
3216 | ||
ac66f547 PZ |
3217 | arg = (struct migration_swap_arg){ |
3218 | .src_task = cur, | |
0ad4e3df | 3219 | .src_cpu = curr_cpu, |
ac66f547 | 3220 | .dst_task = p, |
0ad4e3df | 3221 | .dst_cpu = target_cpu, |
ac66f547 PZ |
3222 | }; |
3223 | ||
3224 | if (arg.src_cpu == arg.dst_cpu) | |
3225 | goto out; | |
3226 | ||
6acce3ef PZ |
3227 | /* |
3228 | * These three tests are all lockless; this is OK since all of them | |
3229 | * will be re-checked with proper locks held further down the line. | |
3230 | */ | |
ac66f547 PZ |
3231 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
3232 | goto out; | |
3233 | ||
3bd37062 | 3234 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
3235 | goto out; |
3236 | ||
3bd37062 | 3237 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
3238 | goto out; |
3239 | ||
286549dc | 3240 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
3241 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
3242 | ||
3243 | out: | |
ac66f547 PZ |
3244 | return ret; |
3245 | } | |
0ad4e3df | 3246 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 3247 | |
1da177e4 LT |
3248 | /* |
3249 | * wait_task_inactive - wait for a thread to unschedule. | |
3250 | * | |
f9fc8cad PZ |
3251 | * Wait for the thread to block in any of the states set in @match_state. |
3252 | * If it changes, i.e. @p might have woken up, then return zero. When we | |
3253 | * succeed in waiting for @p to be off its CPU, we return a positive number | |
3254 | * (its total switch count). If a second call a short while later returns the | |
3255 | * same number, the caller can be sure that @p has remained unscheduled the | |
3256 | * whole time. | |
85ba2d86 | 3257 | * |
1da177e4 LT |
3258 | * The caller must ensure that the task *will* unschedule sometime soon, |
3259 | * else this function might spin for a *long* time. This function can't | |
3260 | * be called with interrupts off, or it may introduce deadlock with | |
3261 | * smp_call_function() if an IPI is sent by the same process we are | |
3262 | * waiting to become inactive. | |
3263 | */ | |
2f064a59 | 3264 | unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state) |
1da177e4 | 3265 | { |
da0c1e65 | 3266 | int running, queued; |
eb580751 | 3267 | struct rq_flags rf; |
85ba2d86 | 3268 | unsigned long ncsw; |
70b97a7f | 3269 | struct rq *rq; |
1da177e4 | 3270 | |
3a5c359a AK |
3271 | for (;;) { |
3272 | /* | |
3273 | * We do the initial early heuristics without holding | |
3274 | * any task-queue locks at all. We'll only try to get | |
3275 | * the runqueue lock when things look like they will | |
3276 | * work out! | |
3277 | */ | |
3278 | rq = task_rq(p); | |
fa490cfd | 3279 | |
3a5c359a AK |
3280 | /* |
3281 | * If the task is actively running on another CPU | |
3282 | * still, just relax and busy-wait without holding | |
3283 | * any locks. | |
3284 | * | |
3285 | * NOTE! Since we don't hold any locks, it's not | |
3286 | * even sure that "rq" stays as the right runqueue! | |
0b9d46fc | 3287 | * But we don't care, since "task_on_cpu()" will |
3a5c359a AK |
3288 | * return false if the runqueue has changed and p |
3289 | * is actually now running somewhere else! | |
3290 | */ | |
0b9d46fc | 3291 | while (task_on_cpu(rq, p)) { |
f9fc8cad | 3292 | if (!(READ_ONCE(p->__state) & match_state)) |
85ba2d86 | 3293 | return 0; |
3a5c359a | 3294 | cpu_relax(); |
85ba2d86 | 3295 | } |
fa490cfd | 3296 | |
3a5c359a AK |
3297 | /* |
3298 | * Ok, time to look more closely! We need the rq | |
3299 | * lock now, to be *sure*. If we're wrong, we'll | |
3300 | * just go back and repeat. | |
3301 | */ | |
eb580751 | 3302 | rq = task_rq_lock(p, &rf); |
27a9da65 | 3303 | trace_sched_wait_task(p); |
0b9d46fc | 3304 | running = task_on_cpu(rq, p); |
da0c1e65 | 3305 | queued = task_on_rq_queued(p); |
85ba2d86 | 3306 | ncsw = 0; |
f9fc8cad | 3307 | if (READ_ONCE(p->__state) & match_state) |
93dcf55f | 3308 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 3309 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 3310 | |
85ba2d86 RM |
3311 | /* |
3312 | * If it changed from the expected state, bail out now. | |
3313 | */ | |
3314 | if (unlikely(!ncsw)) | |
3315 | break; | |
3316 | ||
3a5c359a AK |
3317 | /* |
3318 | * Was it really running after all now that we | |
3319 | * checked with the proper locks actually held? | |
3320 | * | |
3321 | * Oops. Go back and try again.. | |
3322 | */ | |
3323 | if (unlikely(running)) { | |
3324 | cpu_relax(); | |
3325 | continue; | |
3326 | } | |
fa490cfd | 3327 | |
3a5c359a AK |
3328 | /* |
3329 | * It's not enough that it's not actively running, | |
3330 | * it must be off the runqueue _entirely_, and not | |
3331 | * preempted! | |
3332 | * | |
80dd99b3 | 3333 | * So if it was still runnable (but just not actively |
3a5c359a AK |
3334 | * running right now), it's preempted, and we should |
3335 | * yield - it could be a while. | |
3336 | */ | |
da0c1e65 | 3337 | if (unlikely(queued)) { |
8b0e1953 | 3338 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
3339 | |
3340 | set_current_state(TASK_UNINTERRUPTIBLE); | |
c33627e9 | 3341 | schedule_hrtimeout(&to, HRTIMER_MODE_REL_HARD); |
3a5c359a AK |
3342 | continue; |
3343 | } | |
fa490cfd | 3344 | |
3a5c359a AK |
3345 | /* |
3346 | * Ahh, all good. It wasn't running, and it wasn't | |
3347 | * runnable, which means that it will never become | |
3348 | * running in the future either. We're all done! | |
3349 | */ | |
3350 | break; | |
3351 | } | |
85ba2d86 RM |
3352 | |
3353 | return ncsw; | |
1da177e4 LT |
3354 | } |
3355 | ||
3356 | /*** | |
3357 | * kick_process - kick a running thread to enter/exit the kernel | |
3358 | * @p: the to-be-kicked thread | |
3359 | * | |
3360 | * Cause a process which is running on another CPU to enter | |
3361 | * kernel-mode, without any delay. (to get signals handled.) | |
3362 | * | |
25985edc | 3363 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
3364 | * because all it wants to ensure is that the remote task enters |
3365 | * the kernel. If the IPI races and the task has been migrated | |
3366 | * to another CPU then no harm is done and the purpose has been | |
3367 | * achieved as well. | |
3368 | */ | |
36c8b586 | 3369 | void kick_process(struct task_struct *p) |
1da177e4 LT |
3370 | { |
3371 | int cpu; | |
3372 | ||
3373 | preempt_disable(); | |
3374 | cpu = task_cpu(p); | |
3375 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
3376 | smp_send_reschedule(cpu); | |
3377 | preempt_enable(); | |
3378 | } | |
b43e3521 | 3379 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 3380 | |
30da688e | 3381 | /* |
3bd37062 | 3382 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
3383 | * |
3384 | * A few notes on cpu_active vs cpu_online: | |
3385 | * | |
3386 | * - cpu_active must be a subset of cpu_online | |
3387 | * | |
97fb7a0a | 3388 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 3389 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 3390 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
3391 | * see it. |
3392 | * | |
d1ccc66d | 3393 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 3394 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 3395 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
3396 | * off. |
3397 | * | |
3398 | * This means that fallback selection must not select !active CPUs. | |
3399 | * And can assume that any active CPU must be online. Conversely | |
3400 | * select_task_rq() below may allow selection of !active CPUs in order | |
3401 | * to satisfy the above rules. | |
30da688e | 3402 | */ |
5da9a0fb PZ |
3403 | static int select_fallback_rq(int cpu, struct task_struct *p) |
3404 | { | |
aa00d89c TC |
3405 | int nid = cpu_to_node(cpu); |
3406 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
3407 | enum { cpuset, possible, fail } state = cpuset; |
3408 | int dest_cpu; | |
5da9a0fb | 3409 | |
aa00d89c | 3410 | /* |
d1ccc66d IM |
3411 | * If the node that the CPU is on has been offlined, cpu_to_node() |
3412 | * will return -1. There is no CPU on the node, and we should | |
3413 | * select the CPU on the other node. | |
aa00d89c TC |
3414 | */ |
3415 | if (nid != -1) { | |
3416 | nodemask = cpumask_of_node(nid); | |
3417 | ||
3418 | /* Look for allowed, online CPU in same node. */ | |
3419 | for_each_cpu(dest_cpu, nodemask) { | |
9ae606bc | 3420 | if (is_cpu_allowed(p, dest_cpu)) |
aa00d89c TC |
3421 | return dest_cpu; |
3422 | } | |
2baab4e9 | 3423 | } |
5da9a0fb | 3424 | |
2baab4e9 PZ |
3425 | for (;;) { |
3426 | /* Any allowed, online CPU? */ | |
3bd37062 | 3427 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 3428 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 3429 | continue; |
175f0e25 | 3430 | |
2baab4e9 PZ |
3431 | goto out; |
3432 | } | |
5da9a0fb | 3433 | |
e73e85f0 | 3434 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
3435 | switch (state) { |
3436 | case cpuset: | |
97c0054d | 3437 | if (cpuset_cpus_allowed_fallback(p)) { |
e73e85f0 ON |
3438 | state = possible; |
3439 | break; | |
3440 | } | |
df561f66 | 3441 | fallthrough; |
2baab4e9 | 3442 | case possible: |
af449901 PZ |
3443 | /* |
3444 | * XXX When called from select_task_rq() we only | |
3445 | * hold p->pi_lock and again violate locking order. | |
3446 | * | |
3447 | * More yuck to audit. | |
3448 | */ | |
9ae606bc | 3449 | do_set_cpus_allowed(p, task_cpu_possible_mask(p)); |
2baab4e9 PZ |
3450 | state = fail; |
3451 | break; | |
2baab4e9 PZ |
3452 | case fail: |
3453 | BUG(); | |
3454 | break; | |
3455 | } | |
3456 | } | |
3457 | ||
3458 | out: | |
3459 | if (state != cpuset) { | |
3460 | /* | |
3461 | * Don't tell them about moving exiting tasks or | |
3462 | * kernel threads (both mm NULL), since they never | |
3463 | * leave kernel. | |
3464 | */ | |
3465 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 3466 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
3467 | task_pid_nr(p), p->comm, cpu); |
3468 | } | |
5da9a0fb PZ |
3469 | } |
3470 | ||
3471 | return dest_cpu; | |
3472 | } | |
3473 | ||
e2912009 | 3474 | /* |
3bd37062 | 3475 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 3476 | */ |
970b13ba | 3477 | static inline |
3aef1551 | 3478 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 3479 | { |
cbce1a68 PZ |
3480 | lockdep_assert_held(&p->pi_lock); |
3481 | ||
af449901 | 3482 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 3483 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 3484 | else |
3bd37062 | 3485 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
3486 | |
3487 | /* | |
3488 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 3489 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 3490 | * CPU. |
e2912009 PZ |
3491 | * |
3492 | * Since this is common to all placement strategies, this lives here. | |
3493 | * | |
3494 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
3495 | * not worry about this generic constraint ] | |
3496 | */ | |
7af443ee | 3497 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 3498 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
3499 | |
3500 | return cpu; | |
970b13ba | 3501 | } |
09a40af5 | 3502 | |
f5832c19 NP |
3503 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
3504 | { | |
ded467dc | 3505 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
3506 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
3507 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
3508 | ||
3509 | if (stop) { | |
3510 | /* | |
3511 | * Make it appear like a SCHED_FIFO task, its something | |
3512 | * userspace knows about and won't get confused about. | |
3513 | * | |
3514 | * Also, it will make PI more or less work without too | |
3515 | * much confusion -- but then, stop work should not | |
3516 | * rely on PI working anyway. | |
3517 | */ | |
3518 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
3519 | ||
3520 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
3521 | |
3522 | /* | |
3523 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
3524 | * adjust the effective priority of a task. As a result, | |
3525 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
3526 | * which can then trigger wakeups of the stop thread to push | |
3527 | * around the current task. | |
3528 | * | |
3529 | * The stop task itself will never be part of the PI-chain, it | |
3530 | * never blocks, therefore that ->pi_lock recursion is safe. | |
3531 | * Tell lockdep about this by placing the stop->pi_lock in its | |
3532 | * own class. | |
3533 | */ | |
3534 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
3535 | } |
3536 | ||
3537 | cpu_rq(cpu)->stop = stop; | |
3538 | ||
3539 | if (old_stop) { | |
3540 | /* | |
3541 | * Reset it back to a normal scheduling class so that | |
3542 | * it can die in pieces. | |
3543 | */ | |
3544 | old_stop->sched_class = &rt_sched_class; | |
3545 | } | |
3546 | } | |
3547 | ||
74d862b6 | 3548 | #else /* CONFIG_SMP */ |
25834c73 PZ |
3549 | |
3550 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
9cfc3e18 PZ |
3551 | const struct cpumask *new_mask, |
3552 | u32 flags) | |
25834c73 PZ |
3553 | { |
3554 | return set_cpus_allowed_ptr(p, new_mask); | |
3555 | } | |
3556 | ||
af449901 PZ |
3557 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
3558 | ||
3015ef4b TG |
3559 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
3560 | { | |
3561 | return false; | |
3562 | } | |
3563 | ||
74d862b6 | 3564 | #endif /* !CONFIG_SMP */ |
970b13ba | 3565 | |
d7c01d27 | 3566 | static void |
b84cb5df | 3567 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 3568 | { |
4fa8d299 | 3569 | struct rq *rq; |
b84cb5df | 3570 | |
4fa8d299 JP |
3571 | if (!schedstat_enabled()) |
3572 | return; | |
3573 | ||
3574 | rq = this_rq(); | |
d7c01d27 | 3575 | |
4fa8d299 JP |
3576 | #ifdef CONFIG_SMP |
3577 | if (cpu == rq->cpu) { | |
b85c8b71 | 3578 | __schedstat_inc(rq->ttwu_local); |
ceeadb83 | 3579 | __schedstat_inc(p->stats.nr_wakeups_local); |
d7c01d27 PZ |
3580 | } else { |
3581 | struct sched_domain *sd; | |
3582 | ||
ceeadb83 | 3583 | __schedstat_inc(p->stats.nr_wakeups_remote); |
057f3fad | 3584 | rcu_read_lock(); |
4fa8d299 | 3585 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 3586 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 3587 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
3588 | break; |
3589 | } | |
3590 | } | |
057f3fad | 3591 | rcu_read_unlock(); |
d7c01d27 | 3592 | } |
f339b9dc PZ |
3593 | |
3594 | if (wake_flags & WF_MIGRATED) | |
ceeadb83 | 3595 | __schedstat_inc(p->stats.nr_wakeups_migrate); |
d7c01d27 PZ |
3596 | #endif /* CONFIG_SMP */ |
3597 | ||
b85c8b71 | 3598 | __schedstat_inc(rq->ttwu_count); |
ceeadb83 | 3599 | __schedstat_inc(p->stats.nr_wakeups); |
d7c01d27 PZ |
3600 | |
3601 | if (wake_flags & WF_SYNC) | |
ceeadb83 | 3602 | __schedstat_inc(p->stats.nr_wakeups_sync); |
d7c01d27 PZ |
3603 | } |
3604 | ||
23f41eeb PZ |
3605 | /* |
3606 | * Mark the task runnable and perform wakeup-preemption. | |
3607 | */ | |
e7904a28 | 3608 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3609 | struct rq_flags *rf) |
9ed3811a | 3610 | { |
9ed3811a | 3611 | check_preempt_curr(rq, p, wake_flags); |
2f064a59 | 3612 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fbd705a0 PZ |
3613 | trace_sched_wakeup(p); |
3614 | ||
9ed3811a | 3615 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
3616 | if (p->sched_class->task_woken) { |
3617 | /* | |
b19a888c | 3618 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 3619 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 3620 | */ |
d8ac8971 | 3621 | rq_unpin_lock(rq, rf); |
9ed3811a | 3622 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3623 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 3624 | } |
9ed3811a | 3625 | |
e69c6341 | 3626 | if (rq->idle_stamp) { |
78becc27 | 3627 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 3628 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 3629 | |
abfafa54 JL |
3630 | update_avg(&rq->avg_idle, delta); |
3631 | ||
3632 | if (rq->avg_idle > max) | |
9ed3811a | 3633 | rq->avg_idle = max; |
abfafa54 | 3634 | |
94aafc3e PZ |
3635 | rq->wake_stamp = jiffies; |
3636 | rq->wake_avg_idle = rq->avg_idle / 2; | |
3637 | ||
9ed3811a TH |
3638 | rq->idle_stamp = 0; |
3639 | } | |
3640 | #endif | |
3641 | } | |
3642 | ||
c05fbafb | 3643 | static void |
e7904a28 | 3644 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3645 | struct rq_flags *rf) |
c05fbafb | 3646 | { |
77558e4d | 3647 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 3648 | |
5cb9eaa3 | 3649 | lockdep_assert_rq_held(rq); |
cbce1a68 | 3650 | |
c05fbafb PZ |
3651 | if (p->sched_contributes_to_load) |
3652 | rq->nr_uninterruptible--; | |
b5179ac7 | 3653 | |
dbfb089d | 3654 | #ifdef CONFIG_SMP |
b5179ac7 | 3655 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 3656 | en_flags |= ENQUEUE_MIGRATED; |
ec618b84 | 3657 | else |
c05fbafb | 3658 | #endif |
ec618b84 PZ |
3659 | if (p->in_iowait) { |
3660 | delayacct_blkio_end(p); | |
3661 | atomic_dec(&task_rq(p)->nr_iowait); | |
3662 | } | |
c05fbafb | 3663 | |
1b174a2c | 3664 | activate_task(rq, p, en_flags); |
d8ac8971 | 3665 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
3666 | } |
3667 | ||
3668 | /* | |
58877d34 PZ |
3669 | * Consider @p being inside a wait loop: |
3670 | * | |
3671 | * for (;;) { | |
3672 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
3673 | * | |
3674 | * if (CONDITION) | |
3675 | * break; | |
3676 | * | |
3677 | * schedule(); | |
3678 | * } | |
3679 | * __set_current_state(TASK_RUNNING); | |
3680 | * | |
3681 | * between set_current_state() and schedule(). In this case @p is still | |
3682 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3683 | * an atomic manner. | |
3684 | * | |
3685 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3686 | * then schedule() must still happen and p->state can be changed to | |
3687 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3688 | * need to do a full wakeup with enqueue. | |
3689 | * | |
3690 | * Returns: %true when the wakeup is done, | |
3691 | * %false otherwise. | |
c05fbafb | 3692 | */ |
58877d34 | 3693 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3694 | { |
eb580751 | 3695 | struct rq_flags rf; |
c05fbafb PZ |
3696 | struct rq *rq; |
3697 | int ret = 0; | |
3698 | ||
eb580751 | 3699 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3700 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3701 | /* check_preempt_curr() may use rq clock */ |
3702 | update_rq_clock(rq); | |
d8ac8971 | 3703 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3704 | ret = 1; |
3705 | } | |
eb580751 | 3706 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3707 | |
3708 | return ret; | |
3709 | } | |
3710 | ||
317f3941 | 3711 | #ifdef CONFIG_SMP |
a1488664 | 3712 | void sched_ttwu_pending(void *arg) |
317f3941 | 3713 | { |
a1488664 | 3714 | struct llist_node *llist = arg; |
317f3941 | 3715 | struct rq *rq = this_rq(); |
73215849 | 3716 | struct task_struct *p, *t; |
d8ac8971 | 3717 | struct rq_flags rf; |
317f3941 | 3718 | |
e3baac47 PZ |
3719 | if (!llist) |
3720 | return; | |
3721 | ||
126c2092 PZ |
3722 | /* |
3723 | * rq::ttwu_pending racy indication of out-standing wakeups. | |
3724 | * Races such that false-negatives are possible, since they | |
3725 | * are shorter lived that false-positives would be. | |
3726 | */ | |
3727 | WRITE_ONCE(rq->ttwu_pending, 0); | |
3728 | ||
8a8c69c3 | 3729 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3730 | update_rq_clock(rq); |
317f3941 | 3731 | |
8c4890d1 | 3732 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3733 | if (WARN_ON_ONCE(p->on_cpu)) |
3734 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3735 | ||
3736 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3737 | set_task_cpu(p, cpu_of(rq)); | |
3738 | ||
73215849 | 3739 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3740 | } |
317f3941 | 3741 | |
8a8c69c3 | 3742 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3743 | } |
3744 | ||
b2a02fc4 | 3745 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3746 | { |
b2a02fc4 | 3747 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3748 | |
b2a02fc4 PZ |
3749 | if (!set_nr_if_polling(rq->idle)) |
3750 | arch_send_call_function_single_ipi(cpu); | |
3751 | else | |
3752 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3753 | } |
3754 | ||
2ebb1771 MG |
3755 | /* |
3756 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3757 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3758 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3759 | * of the wakeup instead of the waker. | |
3760 | */ | |
3761 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3762 | { |
e3baac47 PZ |
3763 | struct rq *rq = cpu_rq(cpu); |
3764 | ||
b7e7ade3 PZ |
3765 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3766 | ||
126c2092 | 3767 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3768 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3769 | } |
d6aa8f85 | 3770 | |
f6be8af1 CL |
3771 | void wake_up_if_idle(int cpu) |
3772 | { | |
3773 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3774 | struct rq_flags rf; |
f6be8af1 | 3775 | |
fd7de1e8 AL |
3776 | rcu_read_lock(); |
3777 | ||
3778 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3779 | goto out; | |
f6be8af1 | 3780 | |
8850cb66 PZ |
3781 | rq_lock_irqsave(rq, &rf); |
3782 | if (is_idle_task(rq->curr)) | |
3783 | resched_curr(rq); | |
3784 | /* Else CPU is not idle, do nothing here: */ | |
3785 | rq_unlock_irqrestore(rq, &rf); | |
fd7de1e8 AL |
3786 | |
3787 | out: | |
3788 | rcu_read_unlock(); | |
f6be8af1 CL |
3789 | } |
3790 | ||
39be3501 | 3791 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 | 3792 | { |
42dc938a VD |
3793 | if (this_cpu == that_cpu) |
3794 | return true; | |
3795 | ||
518cd623 PZ |
3796 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); |
3797 | } | |
c6e7bd7a | 3798 | |
751d4cbc | 3799 | static inline bool ttwu_queue_cond(struct task_struct *p, int cpu) |
2ebb1771 | 3800 | { |
5ba2ffba PZ |
3801 | /* |
3802 | * Do not complicate things with the async wake_list while the CPU is | |
3803 | * in hotplug state. | |
3804 | */ | |
3805 | if (!cpu_active(cpu)) | |
3806 | return false; | |
3807 | ||
751d4cbc MG |
3808 | /* Ensure the task will still be allowed to run on the CPU. */ |
3809 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) | |
3810 | return false; | |
3811 | ||
2ebb1771 MG |
3812 | /* |
3813 | * If the CPU does not share cache, then queue the task on the | |
3814 | * remote rqs wakelist to avoid accessing remote data. | |
3815 | */ | |
3816 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3817 | return true; | |
3818 | ||
f3dd3f67 TD |
3819 | if (cpu == smp_processor_id()) |
3820 | return false; | |
3821 | ||
2ebb1771 | 3822 | /* |
f3dd3f67 TD |
3823 | * If the wakee cpu is idle, or the task is descheduling and the |
3824 | * only running task on the CPU, then use the wakelist to offload | |
3825 | * the task activation to the idle (or soon-to-be-idle) CPU as | |
3826 | * the current CPU is likely busy. nr_running is checked to | |
3827 | * avoid unnecessary task stacking. | |
28156108 TD |
3828 | * |
3829 | * Note that we can only get here with (wakee) p->on_rq=0, | |
3830 | * p->on_cpu can be whatever, we've done the dequeue, so | |
3831 | * the wakee has been accounted out of ->nr_running. | |
2ebb1771 | 3832 | */ |
f3dd3f67 | 3833 | if (!cpu_rq(cpu)->nr_running) |
2ebb1771 MG |
3834 | return true; |
3835 | ||
3836 | return false; | |
3837 | } | |
3838 | ||
3839 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3840 | { |
751d4cbc | 3841 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(p, cpu)) { |
c6e7bd7a | 3842 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3843 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3844 | return true; |
3845 | } | |
3846 | ||
3847 | return false; | |
3848 | } | |
58877d34 PZ |
3849 | |
3850 | #else /* !CONFIG_SMP */ | |
3851 | ||
3852 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3853 | { | |
3854 | return false; | |
3855 | } | |
3856 | ||
d6aa8f85 | 3857 | #endif /* CONFIG_SMP */ |
317f3941 | 3858 | |
b5179ac7 | 3859 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3860 | { |
3861 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3862 | struct rq_flags rf; |
c05fbafb | 3863 | |
2ebb1771 | 3864 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3865 | return; |
317f3941 | 3866 | |
8a8c69c3 | 3867 | rq_lock(rq, &rf); |
77558e4d | 3868 | update_rq_clock(rq); |
d8ac8971 | 3869 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3870 | rq_unlock(rq, &rf); |
9ed3811a TH |
3871 | } |
3872 | ||
43295d73 TG |
3873 | /* |
3874 | * Invoked from try_to_wake_up() to check whether the task can be woken up. | |
3875 | * | |
3876 | * The caller holds p::pi_lock if p != current or has preemption | |
3877 | * disabled when p == current. | |
5f220be2 TG |
3878 | * |
3879 | * The rules of PREEMPT_RT saved_state: | |
3880 | * | |
3881 | * The related locking code always holds p::pi_lock when updating | |
3882 | * p::saved_state, which means the code is fully serialized in both cases. | |
3883 | * | |
3884 | * The lock wait and lock wakeups happen via TASK_RTLOCK_WAIT. No other | |
3885 | * bits set. This allows to distinguish all wakeup scenarios. | |
43295d73 TG |
3886 | */ |
3887 | static __always_inline | |
3888 | bool ttwu_state_match(struct task_struct *p, unsigned int state, int *success) | |
3889 | { | |
5f220be2 TG |
3890 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)) { |
3891 | WARN_ON_ONCE((state & TASK_RTLOCK_WAIT) && | |
3892 | state != TASK_RTLOCK_WAIT); | |
3893 | } | |
3894 | ||
43295d73 TG |
3895 | if (READ_ONCE(p->__state) & state) { |
3896 | *success = 1; | |
3897 | return true; | |
3898 | } | |
5f220be2 TG |
3899 | |
3900 | #ifdef CONFIG_PREEMPT_RT | |
3901 | /* | |
3902 | * Saved state preserves the task state across blocking on | |
3903 | * an RT lock. If the state matches, set p::saved_state to | |
3904 | * TASK_RUNNING, but do not wake the task because it waits | |
3905 | * for a lock wakeup. Also indicate success because from | |
3906 | * the regular waker's point of view this has succeeded. | |
3907 | * | |
3908 | * After acquiring the lock the task will restore p::__state | |
3909 | * from p::saved_state which ensures that the regular | |
3910 | * wakeup is not lost. The restore will also set | |
3911 | * p::saved_state to TASK_RUNNING so any further tests will | |
3912 | * not result in false positives vs. @success | |
3913 | */ | |
3914 | if (p->saved_state & state) { | |
3915 | p->saved_state = TASK_RUNNING; | |
3916 | *success = 1; | |
3917 | } | |
3918 | #endif | |
43295d73 TG |
3919 | return false; |
3920 | } | |
3921 | ||
8643cda5 PZ |
3922 | /* |
3923 | * Notes on Program-Order guarantees on SMP systems. | |
3924 | * | |
3925 | * MIGRATION | |
3926 | * | |
3927 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3928 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3929 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3930 | * |
3931 | * For migration (of runnable tasks) this is provided by the following means: | |
3932 | * | |
3933 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3934 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3935 | * rq(c1)->lock (if not at the same time, then in that order). | |
3936 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3937 | * | |
7696f991 | 3938 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3939 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3940 | * |
3941 | * Example: | |
3942 | * | |
3943 | * CPU0 CPU1 CPU2 | |
3944 | * | |
3945 | * LOCK rq(0)->lock | |
3946 | * sched-out X | |
3947 | * sched-in Y | |
3948 | * UNLOCK rq(0)->lock | |
3949 | * | |
3950 | * LOCK rq(0)->lock // orders against CPU0 | |
3951 | * dequeue X | |
3952 | * UNLOCK rq(0)->lock | |
3953 | * | |
3954 | * LOCK rq(1)->lock | |
3955 | * enqueue X | |
3956 | * UNLOCK rq(1)->lock | |
3957 | * | |
3958 | * LOCK rq(1)->lock // orders against CPU2 | |
3959 | * sched-out Z | |
3960 | * sched-in X | |
3961 | * UNLOCK rq(1)->lock | |
3962 | * | |
3963 | * | |
3964 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3965 | * | |
3966 | * For blocking we (obviously) need to provide the same guarantee as for | |
3967 | * migration. However the means are completely different as there is no lock | |
3968 | * chain to provide order. Instead we do: | |
3969 | * | |
58877d34 PZ |
3970 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3971 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3972 | * |
3973 | * Example: | |
3974 | * | |
3975 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
3976 | * | |
3977 | * LOCK rq(0)->lock LOCK X->pi_lock | |
3978 | * dequeue X | |
3979 | * sched-out X | |
3980 | * smp_store_release(X->on_cpu, 0); | |
3981 | * | |
1f03e8d2 | 3982 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
3983 | * X->state = WAKING |
3984 | * set_task_cpu(X,2) | |
3985 | * | |
3986 | * LOCK rq(2)->lock | |
3987 | * enqueue X | |
3988 | * X->state = RUNNING | |
3989 | * UNLOCK rq(2)->lock | |
3990 | * | |
3991 | * LOCK rq(2)->lock // orders against CPU1 | |
3992 | * sched-out Z | |
3993 | * sched-in X | |
3994 | * UNLOCK rq(2)->lock | |
3995 | * | |
3996 | * UNLOCK X->pi_lock | |
3997 | * UNLOCK rq(0)->lock | |
3998 | * | |
3999 | * | |
7696f991 AP |
4000 | * However, for wakeups there is a second guarantee we must provide, namely we |
4001 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
4002 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
4003 | */ |
4004 | ||
9ed3811a | 4005 | /** |
1da177e4 | 4006 | * try_to_wake_up - wake up a thread |
9ed3811a | 4007 | * @p: the thread to be awakened |
1da177e4 | 4008 | * @state: the mask of task states that can be woken |
9ed3811a | 4009 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 4010 | * |
58877d34 PZ |
4011 | * Conceptually does: |
4012 | * | |
4013 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 4014 | * |
a2250238 PZ |
4015 | * If the task was not queued/runnable, also place it back on a runqueue. |
4016 | * | |
58877d34 PZ |
4017 | * This function is atomic against schedule() which would dequeue the task. |
4018 | * | |
4019 | * It issues a full memory barrier before accessing @p->state, see the comment | |
4020 | * with set_current_state(). | |
a2250238 | 4021 | * |
58877d34 | 4022 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 4023 | * |
58877d34 PZ |
4024 | * Relies on p->pi_lock stabilizing: |
4025 | * - p->sched_class | |
4026 | * - p->cpus_ptr | |
4027 | * - p->sched_task_group | |
4028 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
4029 | * | |
4030 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
4031 | * Takes rq->lock in: | |
4032 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
4033 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
4034 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
4035 | * | |
4036 | * As a consequence we race really badly with just about everything. See the | |
4037 | * many memory barriers and their comments for details. | |
7696f991 | 4038 | * |
a2250238 PZ |
4039 | * Return: %true if @p->state changes (an actual wakeup was done), |
4040 | * %false otherwise. | |
1da177e4 | 4041 | */ |
e4a52bcb PZ |
4042 | static int |
4043 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 4044 | { |
1da177e4 | 4045 | unsigned long flags; |
c05fbafb | 4046 | int cpu, success = 0; |
2398f2c6 | 4047 | |
e3d85487 | 4048 | preempt_disable(); |
aacedf26 PZ |
4049 | if (p == current) { |
4050 | /* | |
4051 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
4052 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 4053 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
4054 | * without taking any locks. |
4055 | * | |
4056 | * In particular: | |
4057 | * - we rely on Program-Order guarantees for all the ordering, | |
4058 | * - we're serialized against set_special_state() by virtue of | |
4059 | * it disabling IRQs (this allows not taking ->pi_lock). | |
4060 | */ | |
43295d73 | 4061 | if (!ttwu_state_match(p, state, &success)) |
e3d85487 | 4062 | goto out; |
aacedf26 | 4063 | |
aacedf26 | 4064 | trace_sched_waking(p); |
2f064a59 | 4065 | WRITE_ONCE(p->__state, TASK_RUNNING); |
aacedf26 PZ |
4066 | trace_sched_wakeup(p); |
4067 | goto out; | |
4068 | } | |
4069 | ||
e0acd0a6 ON |
4070 | /* |
4071 | * If we are going to wake up a thread waiting for CONDITION we | |
4072 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
4073 | * reordered with p->state check below. This pairs with smp_store_mb() |
4074 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 4075 | */ |
013fdb80 | 4076 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 4077 | smp_mb__after_spinlock(); |
43295d73 | 4078 | if (!ttwu_state_match(p, state, &success)) |
aacedf26 | 4079 | goto unlock; |
1da177e4 | 4080 | |
fbd705a0 PZ |
4081 | trace_sched_waking(p); |
4082 | ||
135e8c92 BS |
4083 | /* |
4084 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
4085 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
4086 | * in smp_cond_load_acquire() below. | |
4087 | * | |
3d85b270 AP |
4088 | * sched_ttwu_pending() try_to_wake_up() |
4089 | * STORE p->on_rq = 1 LOAD p->state | |
4090 | * UNLOCK rq->lock | |
4091 | * | |
4092 | * __schedule() (switch to task 'p') | |
4093 | * LOCK rq->lock smp_rmb(); | |
4094 | * smp_mb__after_spinlock(); | |
4095 | * UNLOCK rq->lock | |
135e8c92 BS |
4096 | * |
4097 | * [task p] | |
3d85b270 | 4098 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 4099 | * |
3d85b270 AP |
4100 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4101 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
4102 | * |
4103 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
4104 | */ |
4105 | smp_rmb(); | |
58877d34 | 4106 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 4107 | goto unlock; |
1da177e4 | 4108 | |
1da177e4 | 4109 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
4110 | /* |
4111 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
4112 | * possible to, falsely, observe p->on_cpu == 0. | |
4113 | * | |
4114 | * One must be running (->on_cpu == 1) in order to remove oneself | |
4115 | * from the runqueue. | |
4116 | * | |
3d85b270 AP |
4117 | * __schedule() (switch to task 'p') try_to_wake_up() |
4118 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
4119 | * UNLOCK rq->lock | |
4120 | * | |
4121 | * __schedule() (put 'p' to sleep) | |
4122 | * LOCK rq->lock smp_rmb(); | |
4123 | * smp_mb__after_spinlock(); | |
4124 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 4125 | * |
3d85b270 AP |
4126 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4127 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
4128 | * |
4129 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
4130 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
4131 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 4132 | */ |
dbfb089d PZ |
4133 | smp_acquire__after_ctrl_dep(); |
4134 | ||
4135 | /* | |
4136 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
4137 | * == 0), which means we need to do an enqueue, change p->state to | |
4138 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
4139 | * enqueue, such as ttwu_queue_wakelist(). | |
4140 | */ | |
2f064a59 | 4141 | WRITE_ONCE(p->__state, TASK_WAKING); |
ecf7d01c | 4142 | |
c6e7bd7a PZ |
4143 | /* |
4144 | * If the owning (remote) CPU is still in the middle of schedule() with | |
4145 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
4146 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
4147 | * let the waker make forward progress. This is safe because IRQs are | |
4148 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
4149 | * |
4150 | * Ensure we load task_cpu(p) after p->on_cpu: | |
4151 | * | |
4152 | * set_task_cpu(p, cpu); | |
4153 | * STORE p->cpu = @cpu | |
4154 | * __schedule() (switch to task 'p') | |
4155 | * LOCK rq->lock | |
4156 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
4157 | * STORE p->on_cpu = 1 LOAD p->cpu | |
4158 | * | |
4159 | * to ensure we observe the correct CPU on which the task is currently | |
4160 | * scheduling. | |
c6e7bd7a | 4161 | */ |
b6e13e85 | 4162 | if (smp_load_acquire(&p->on_cpu) && |
f3dd3f67 | 4163 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags)) |
c6e7bd7a PZ |
4164 | goto unlock; |
4165 | ||
e9c84311 | 4166 | /* |
d1ccc66d | 4167 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 4168 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 4169 | * |
31cb1bc0 | 4170 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
4171 | * |
4172 | * This ensures that tasks getting woken will be fully ordered against | |
4173 | * their previous state and preserve Program Order. | |
0970d299 | 4174 | */ |
1f03e8d2 | 4175 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 4176 | |
3aef1551 | 4177 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc | 4178 | if (task_cpu(p) != cpu) { |
ec618b84 PZ |
4179 | if (p->in_iowait) { |
4180 | delayacct_blkio_end(p); | |
4181 | atomic_dec(&task_rq(p)->nr_iowait); | |
4182 | } | |
4183 | ||
f339b9dc | 4184 | wake_flags |= WF_MIGRATED; |
eb414681 | 4185 | psi_ttwu_dequeue(p); |
e4a52bcb | 4186 | set_task_cpu(p, cpu); |
f339b9dc | 4187 | } |
b6e13e85 PZ |
4188 | #else |
4189 | cpu = task_cpu(p); | |
1da177e4 | 4190 | #endif /* CONFIG_SMP */ |
1da177e4 | 4191 | |
b5179ac7 | 4192 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 4193 | unlock: |
013fdb80 | 4194 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
4195 | out: |
4196 | if (success) | |
b6e13e85 | 4197 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 4198 | preempt_enable(); |
1da177e4 LT |
4199 | |
4200 | return success; | |
4201 | } | |
4202 | ||
2beaf328 | 4203 | /** |
9b3c4ab3 | 4204 | * task_call_func - Invoke a function on task in fixed state |
1b7af295 | 4205 | * @p: Process for which the function is to be invoked, can be @current. |
2beaf328 PM |
4206 | * @func: Function to invoke. |
4207 | * @arg: Argument to function. | |
4208 | * | |
f6ac18fa PZ |
4209 | * Fix the task in it's current state by avoiding wakeups and or rq operations |
4210 | * and call @func(@arg) on it. This function can use ->on_rq and task_curr() | |
4211 | * to work out what the state is, if required. Given that @func can be invoked | |
4212 | * with a runqueue lock held, it had better be quite lightweight. | |
2beaf328 PM |
4213 | * |
4214 | * Returns: | |
f6ac18fa | 4215 | * Whatever @func returns |
2beaf328 | 4216 | */ |
9b3c4ab3 | 4217 | int task_call_func(struct task_struct *p, task_call_f func, void *arg) |
2beaf328 | 4218 | { |
f6ac18fa PZ |
4219 | struct rq *rq = NULL; |
4220 | unsigned int state; | |
2beaf328 | 4221 | struct rq_flags rf; |
9b3c4ab3 | 4222 | int ret; |
2beaf328 | 4223 | |
1b7af295 | 4224 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
f6ac18fa PZ |
4225 | |
4226 | state = READ_ONCE(p->__state); | |
4227 | ||
4228 | /* | |
4229 | * Ensure we load p->on_rq after p->__state, otherwise it would be | |
4230 | * possible to, falsely, observe p->on_rq == 0. | |
4231 | * | |
4232 | * See try_to_wake_up() for a longer comment. | |
4233 | */ | |
4234 | smp_rmb(); | |
4235 | ||
4236 | /* | |
4237 | * Since pi->lock blocks try_to_wake_up(), we don't need rq->lock when | |
4238 | * the task is blocked. Make sure to check @state since ttwu() can drop | |
4239 | * locks at the end, see ttwu_queue_wakelist(). | |
4240 | */ | |
4241 | if (state == TASK_RUNNING || state == TASK_WAKING || p->on_rq) | |
2beaf328 | 4242 | rq = __task_rq_lock(p, &rf); |
f6ac18fa PZ |
4243 | |
4244 | /* | |
4245 | * At this point the task is pinned; either: | |
4246 | * - blocked and we're holding off wakeups (pi->lock) | |
4247 | * - woken, and we're holding off enqueue (rq->lock) | |
4248 | * - queued, and we're holding off schedule (rq->lock) | |
4249 | * - running, and we're holding off de-schedule (rq->lock) | |
4250 | * | |
4251 | * The called function (@func) can use: task_curr(), p->on_rq and | |
4252 | * p->__state to differentiate between these states. | |
4253 | */ | |
4254 | ret = func(p, arg); | |
4255 | ||
4256 | if (rq) | |
2beaf328 | 4257 | rq_unlock(rq, &rf); |
f6ac18fa | 4258 | |
1b7af295 | 4259 | raw_spin_unlock_irqrestore(&p->pi_lock, rf.flags); |
2beaf328 PM |
4260 | return ret; |
4261 | } | |
4262 | ||
e386b672 PM |
4263 | /** |
4264 | * cpu_curr_snapshot - Return a snapshot of the currently running task | |
4265 | * @cpu: The CPU on which to snapshot the task. | |
4266 | * | |
4267 | * Returns the task_struct pointer of the task "currently" running on | |
4268 | * the specified CPU. If the same task is running on that CPU throughout, | |
4269 | * the return value will be a pointer to that task's task_struct structure. | |
4270 | * If the CPU did any context switches even vaguely concurrently with the | |
4271 | * execution of this function, the return value will be a pointer to the | |
4272 | * task_struct structure of a randomly chosen task that was running on | |
4273 | * that CPU somewhere around the time that this function was executing. | |
4274 | * | |
4275 | * If the specified CPU was offline, the return value is whatever it | |
4276 | * is, perhaps a pointer to the task_struct structure of that CPU's idle | |
4277 | * task, but there is no guarantee. Callers wishing a useful return | |
4278 | * value must take some action to ensure that the specified CPU remains | |
4279 | * online throughout. | |
4280 | * | |
4281 | * This function executes full memory barriers before and after fetching | |
4282 | * the pointer, which permits the caller to confine this function's fetch | |
4283 | * with respect to the caller's accesses to other shared variables. | |
4284 | */ | |
4285 | struct task_struct *cpu_curr_snapshot(int cpu) | |
4286 | { | |
4287 | struct task_struct *t; | |
4288 | ||
4289 | smp_mb(); /* Pairing determined by caller's synchronization design. */ | |
4290 | t = rcu_dereference(cpu_curr(cpu)); | |
4291 | smp_mb(); /* Pairing determined by caller's synchronization design. */ | |
4292 | return t; | |
4293 | } | |
4294 | ||
50fa610a DH |
4295 | /** |
4296 | * wake_up_process - Wake up a specific process | |
4297 | * @p: The process to be woken up. | |
4298 | * | |
4299 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
4300 | * processes. |
4301 | * | |
4302 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 4303 | * |
7696f991 | 4304 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 4305 | */ |
7ad5b3a5 | 4306 | int wake_up_process(struct task_struct *p) |
1da177e4 | 4307 | { |
9067ac85 | 4308 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 4309 | } |
1da177e4 LT |
4310 | EXPORT_SYMBOL(wake_up_process); |
4311 | ||
7ad5b3a5 | 4312 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
4313 | { |
4314 | return try_to_wake_up(p, state, 0); | |
4315 | } | |
4316 | ||
1da177e4 LT |
4317 | /* |
4318 | * Perform scheduler related setup for a newly forked process p. | |
4319 | * p is forked by current. | |
dd41f596 IM |
4320 | * |
4321 | * __sched_fork() is basic setup used by init_idle() too: | |
4322 | */ | |
5e1576ed | 4323 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4324 | { |
fd2f4419 PZ |
4325 | p->on_rq = 0; |
4326 | ||
4327 | p->se.on_rq = 0; | |
dd41f596 IM |
4328 | p->se.exec_start = 0; |
4329 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 4330 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 4331 | p->se.nr_migrations = 0; |
da7a735e | 4332 | p->se.vruntime = 0; |
fd2f4419 | 4333 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 4334 | |
ad936d86 BP |
4335 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4336 | p->se.cfs_rq = NULL; | |
4337 | #endif | |
4338 | ||
6cfb0d5d | 4339 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 4340 | /* Even if schedstat is disabled, there should not be garbage */ |
ceeadb83 | 4341 | memset(&p->stats, 0, sizeof(p->stats)); |
6cfb0d5d | 4342 | #endif |
476d139c | 4343 | |
aab03e05 | 4344 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 4345 | init_dl_task_timer(&p->dl); |
209a0cbd | 4346 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 4347 | __dl_clear_params(p); |
aab03e05 | 4348 | |
fa717060 | 4349 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
4350 | p->rt.timeout = 0; |
4351 | p->rt.time_slice = sched_rr_timeslice; | |
4352 | p->rt.on_rq = 0; | |
4353 | p->rt.on_list = 0; | |
476d139c | 4354 | |
e107be36 AK |
4355 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4356 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
4357 | #endif | |
cbee9f88 | 4358 | |
5e1f0f09 MG |
4359 | #ifdef CONFIG_COMPACTION |
4360 | p->capture_control = NULL; | |
4361 | #endif | |
13784475 | 4362 | init_numa_balancing(clone_flags, p); |
a1488664 | 4363 | #ifdef CONFIG_SMP |
8c4890d1 | 4364 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 4365 | p->migration_pending = NULL; |
a1488664 | 4366 | #endif |
dd41f596 IM |
4367 | } |
4368 | ||
2a595721 SD |
4369 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
4370 | ||
1a687c2e | 4371 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 4372 | |
c574bbe9 HY |
4373 | int sysctl_numa_balancing_mode; |
4374 | ||
4375 | static void __set_numabalancing_state(bool enabled) | |
1a687c2e MG |
4376 | { |
4377 | if (enabled) | |
2a595721 | 4378 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 4379 | else |
2a595721 | 4380 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 4381 | } |
54a43d54 | 4382 | |
c574bbe9 HY |
4383 | void set_numabalancing_state(bool enabled) |
4384 | { | |
4385 | if (enabled) | |
4386 | sysctl_numa_balancing_mode = NUMA_BALANCING_NORMAL; | |
4387 | else | |
4388 | sysctl_numa_balancing_mode = NUMA_BALANCING_DISABLED; | |
4389 | __set_numabalancing_state(enabled); | |
4390 | } | |
4391 | ||
54a43d54 | 4392 | #ifdef CONFIG_PROC_SYSCTL |
c959924b HY |
4393 | static void reset_memory_tiering(void) |
4394 | { | |
4395 | struct pglist_data *pgdat; | |
4396 | ||
4397 | for_each_online_pgdat(pgdat) { | |
4398 | pgdat->nbp_threshold = 0; | |
4399 | pgdat->nbp_th_nr_cand = node_page_state(pgdat, PGPROMOTE_CANDIDATE); | |
4400 | pgdat->nbp_th_start = jiffies_to_msecs(jiffies); | |
4401 | } | |
4402 | } | |
4403 | ||
0dff89c4 | 4404 | static int sysctl_numa_balancing(struct ctl_table *table, int write, |
32927393 | 4405 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
4406 | { |
4407 | struct ctl_table t; | |
4408 | int err; | |
c574bbe9 | 4409 | int state = sysctl_numa_balancing_mode; |
54a43d54 AK |
4410 | |
4411 | if (write && !capable(CAP_SYS_ADMIN)) | |
4412 | return -EPERM; | |
4413 | ||
4414 | t = *table; | |
4415 | t.data = &state; | |
4416 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4417 | if (err < 0) | |
4418 | return err; | |
c574bbe9 | 4419 | if (write) { |
c959924b HY |
4420 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && |
4421 | (state & NUMA_BALANCING_MEMORY_TIERING)) | |
4422 | reset_memory_tiering(); | |
c574bbe9 HY |
4423 | sysctl_numa_balancing_mode = state; |
4424 | __set_numabalancing_state(state); | |
4425 | } | |
54a43d54 AK |
4426 | return err; |
4427 | } | |
4428 | #endif | |
4429 | #endif | |
dd41f596 | 4430 | |
4698f88c JP |
4431 | #ifdef CONFIG_SCHEDSTATS |
4432 | ||
cb251765 MG |
4433 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4434 | ||
cb251765 MG |
4435 | static void set_schedstats(bool enabled) |
4436 | { | |
4437 | if (enabled) | |
4438 | static_branch_enable(&sched_schedstats); | |
4439 | else | |
4440 | static_branch_disable(&sched_schedstats); | |
4441 | } | |
4442 | ||
4443 | void force_schedstat_enabled(void) | |
4444 | { | |
4445 | if (!schedstat_enabled()) { | |
4446 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
4447 | static_branch_enable(&sched_schedstats); | |
4448 | } | |
4449 | } | |
4450 | ||
4451 | static int __init setup_schedstats(char *str) | |
4452 | { | |
4453 | int ret = 0; | |
4454 | if (!str) | |
4455 | goto out; | |
4456 | ||
4457 | if (!strcmp(str, "enable")) { | |
1faa491a | 4458 | set_schedstats(true); |
cb251765 MG |
4459 | ret = 1; |
4460 | } else if (!strcmp(str, "disable")) { | |
1faa491a | 4461 | set_schedstats(false); |
cb251765 MG |
4462 | ret = 1; |
4463 | } | |
4464 | out: | |
4465 | if (!ret) | |
4466 | pr_warn("Unable to parse schedstats=\n"); | |
4467 | ||
4468 | return ret; | |
4469 | } | |
4470 | __setup("schedstats=", setup_schedstats); | |
4471 | ||
4472 | #ifdef CONFIG_PROC_SYSCTL | |
f5ef06d5 | 4473 | static int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
32927393 | 4474 | size_t *lenp, loff_t *ppos) |
cb251765 MG |
4475 | { |
4476 | struct ctl_table t; | |
4477 | int err; | |
4478 | int state = static_branch_likely(&sched_schedstats); | |
4479 | ||
4480 | if (write && !capable(CAP_SYS_ADMIN)) | |
4481 | return -EPERM; | |
4482 | ||
4483 | t = *table; | |
4484 | t.data = &state; | |
4485 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4486 | if (err < 0) | |
4487 | return err; | |
4488 | if (write) | |
4489 | set_schedstats(state); | |
4490 | return err; | |
4491 | } | |
4698f88c | 4492 | #endif /* CONFIG_PROC_SYSCTL */ |
4698f88c | 4493 | #endif /* CONFIG_SCHEDSTATS */ |
dd41f596 | 4494 | |
3267e015 ZN |
4495 | #ifdef CONFIG_SYSCTL |
4496 | static struct ctl_table sched_core_sysctls[] = { | |
4497 | #ifdef CONFIG_SCHEDSTATS | |
f5ef06d5 ZN |
4498 | { |
4499 | .procname = "sched_schedstats", | |
4500 | .data = NULL, | |
4501 | .maxlen = sizeof(unsigned int), | |
4502 | .mode = 0644, | |
4503 | .proc_handler = sysctl_schedstats, | |
4504 | .extra1 = SYSCTL_ZERO, | |
4505 | .extra2 = SYSCTL_ONE, | |
4506 | }, | |
3267e015 ZN |
4507 | #endif /* CONFIG_SCHEDSTATS */ |
4508 | #ifdef CONFIG_UCLAMP_TASK | |
4509 | { | |
4510 | .procname = "sched_util_clamp_min", | |
4511 | .data = &sysctl_sched_uclamp_util_min, | |
4512 | .maxlen = sizeof(unsigned int), | |
4513 | .mode = 0644, | |
4514 | .proc_handler = sysctl_sched_uclamp_handler, | |
4515 | }, | |
4516 | { | |
4517 | .procname = "sched_util_clamp_max", | |
4518 | .data = &sysctl_sched_uclamp_util_max, | |
4519 | .maxlen = sizeof(unsigned int), | |
4520 | .mode = 0644, | |
4521 | .proc_handler = sysctl_sched_uclamp_handler, | |
4522 | }, | |
4523 | { | |
4524 | .procname = "sched_util_clamp_min_rt_default", | |
4525 | .data = &sysctl_sched_uclamp_util_min_rt_default, | |
4526 | .maxlen = sizeof(unsigned int), | |
4527 | .mode = 0644, | |
4528 | .proc_handler = sysctl_sched_uclamp_handler, | |
4529 | }, | |
4530 | #endif /* CONFIG_UCLAMP_TASK */ | |
0dff89c4 KW |
4531 | #ifdef CONFIG_NUMA_BALANCING |
4532 | { | |
4533 | .procname = "numa_balancing", | |
4534 | .data = NULL, /* filled in by handler */ | |
4535 | .maxlen = sizeof(unsigned int), | |
4536 | .mode = 0644, | |
4537 | .proc_handler = sysctl_numa_balancing, | |
4538 | .extra1 = SYSCTL_ZERO, | |
4539 | .extra2 = SYSCTL_FOUR, | |
4540 | }, | |
4541 | #endif /* CONFIG_NUMA_BALANCING */ | |
f5ef06d5 ZN |
4542 | {} |
4543 | }; | |
3267e015 | 4544 | static int __init sched_core_sysctl_init(void) |
f5ef06d5 | 4545 | { |
3267e015 | 4546 | register_sysctl_init("kernel", sched_core_sysctls); |
f5ef06d5 ZN |
4547 | return 0; |
4548 | } | |
3267e015 ZN |
4549 | late_initcall(sched_core_sysctl_init); |
4550 | #endif /* CONFIG_SYSCTL */ | |
dd41f596 IM |
4551 | |
4552 | /* | |
4553 | * fork()/clone()-time setup: | |
4554 | */ | |
aab03e05 | 4555 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4556 | { |
5e1576ed | 4557 | __sched_fork(clone_flags, p); |
06b83b5f | 4558 | /* |
7dc603c9 | 4559 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
4560 | * nobody will actually run it, and a signal or other external |
4561 | * event cannot wake it up and insert it on the runqueue either. | |
4562 | */ | |
2f064a59 | 4563 | p->__state = TASK_NEW; |
dd41f596 | 4564 | |
c350a04e MG |
4565 | /* |
4566 | * Make sure we do not leak PI boosting priority to the child. | |
4567 | */ | |
4568 | p->prio = current->normal_prio; | |
4569 | ||
e8f14172 PB |
4570 | uclamp_fork(p); |
4571 | ||
b9dc29e7 MG |
4572 | /* |
4573 | * Revert to default priority/policy on fork if requested. | |
4574 | */ | |
4575 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 4576 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 4577 | p->policy = SCHED_NORMAL; |
6c697bdf | 4578 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
4579 | p->rt_priority = 0; |
4580 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
4581 | p->static_prio = NICE_TO_PRIO(0); | |
4582 | ||
f558c2b8 | 4583 | p->prio = p->normal_prio = p->static_prio; |
b1e82065 | 4584 | set_load_weight(p, false); |
6c697bdf | 4585 | |
b9dc29e7 MG |
4586 | /* |
4587 | * We don't need the reset flag anymore after the fork. It has | |
4588 | * fulfilled its duty: | |
4589 | */ | |
4590 | p->sched_reset_on_fork = 0; | |
4591 | } | |
ca94c442 | 4592 | |
af0fffd9 | 4593 | if (dl_prio(p->prio)) |
aab03e05 | 4594 | return -EAGAIN; |
af0fffd9 | 4595 | else if (rt_prio(p->prio)) |
aab03e05 | 4596 | p->sched_class = &rt_sched_class; |
af0fffd9 | 4597 | else |
2ddbf952 | 4598 | p->sched_class = &fair_sched_class; |
b29739f9 | 4599 | |
7dc603c9 | 4600 | init_entity_runnable_average(&p->se); |
cd29fe6f | 4601 | |
b1e82065 | 4602 | |
f6db8347 | 4603 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 4604 | if (likely(sched_info_on())) |
52f17b6c | 4605 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 4606 | #endif |
3ca7a440 PZ |
4607 | #if defined(CONFIG_SMP) |
4608 | p->on_cpu = 0; | |
4866cde0 | 4609 | #endif |
01028747 | 4610 | init_task_preempt_count(p); |
806c09a7 | 4611 | #ifdef CONFIG_SMP |
917b627d | 4612 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 4613 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 4614 | #endif |
aab03e05 | 4615 | return 0; |
1da177e4 LT |
4616 | } |
4617 | ||
b1e82065 | 4618 | void sched_cgroup_fork(struct task_struct *p, struct kernel_clone_args *kargs) |
13685c4a | 4619 | { |
4ef0c5c6 | 4620 | unsigned long flags; |
4ef0c5c6 | 4621 | |
b1e82065 PZ |
4622 | /* |
4623 | * Because we're not yet on the pid-hash, p->pi_lock isn't strictly | |
4624 | * required yet, but lockdep gets upset if rules are violated. | |
4625 | */ | |
4ef0c5c6 ZQ |
4626 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
4627 | #ifdef CONFIG_CGROUP_SCHED | |
b1e82065 PZ |
4628 | if (1) { |
4629 | struct task_group *tg; | |
4630 | tg = container_of(kargs->cset->subsys[cpu_cgrp_id], | |
4631 | struct task_group, css); | |
4632 | tg = autogroup_task_group(p, tg); | |
4633 | p->sched_task_group = tg; | |
4634 | } | |
4ef0c5c6 ZQ |
4635 | #endif |
4636 | rseq_migrate(p); | |
4637 | /* | |
4638 | * We're setting the CPU for the first time, we don't migrate, | |
4639 | * so use __set_task_cpu(). | |
4640 | */ | |
4641 | __set_task_cpu(p, smp_processor_id()); | |
4642 | if (p->sched_class->task_fork) | |
4643 | p->sched_class->task_fork(p); | |
4644 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
b1e82065 | 4645 | } |
4ef0c5c6 | 4646 | |
b1e82065 PZ |
4647 | void sched_post_fork(struct task_struct *p) |
4648 | { | |
13685c4a QY |
4649 | uclamp_post_fork(p); |
4650 | } | |
4651 | ||
332ac17e DF |
4652 | unsigned long to_ratio(u64 period, u64 runtime) |
4653 | { | |
4654 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 4655 | return BW_UNIT; |
332ac17e DF |
4656 | |
4657 | /* | |
4658 | * Doing this here saves a lot of checks in all | |
4659 | * the calling paths, and returning zero seems | |
4660 | * safe for them anyway. | |
4661 | */ | |
4662 | if (period == 0) | |
4663 | return 0; | |
4664 | ||
c52f14d3 | 4665 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
4666 | } |
4667 | ||
1da177e4 LT |
4668 | /* |
4669 | * wake_up_new_task - wake up a newly created task for the first time. | |
4670 | * | |
4671 | * This function will do some initial scheduler statistics housekeeping | |
4672 | * that must be done for every newly created context, then puts the task | |
4673 | * on the runqueue and wakes it. | |
4674 | */ | |
3e51e3ed | 4675 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 4676 | { |
eb580751 | 4677 | struct rq_flags rf; |
dd41f596 | 4678 | struct rq *rq; |
fabf318e | 4679 | |
eb580751 | 4680 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
2f064a59 | 4681 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fabf318e PZ |
4682 | #ifdef CONFIG_SMP |
4683 | /* | |
4684 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 4685 | * - cpus_ptr can change in the fork path |
d1ccc66d | 4686 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
4687 | * |
4688 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
4689 | * as we're not fully set-up yet. | |
fabf318e | 4690 | */ |
32e839dd | 4691 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 4692 | rseq_migrate(p); |
3aef1551 | 4693 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 4694 | #endif |
b7fa30c9 | 4695 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 4696 | update_rq_clock(rq); |
d0fe0b9c | 4697 | post_init_entity_util_avg(p); |
0017d735 | 4698 | |
7a57f32a | 4699 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 4700 | trace_sched_wakeup_new(p); |
a7558e01 | 4701 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 4702 | #ifdef CONFIG_SMP |
0aaafaab PZ |
4703 | if (p->sched_class->task_woken) { |
4704 | /* | |
b19a888c | 4705 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
4706 | * drop it. |
4707 | */ | |
d8ac8971 | 4708 | rq_unpin_lock(rq, &rf); |
efbbd05a | 4709 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 4710 | rq_repin_lock(rq, &rf); |
0aaafaab | 4711 | } |
9a897c5a | 4712 | #endif |
eb580751 | 4713 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4714 | } |
4715 | ||
e107be36 AK |
4716 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4717 | ||
b7203428 | 4718 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 4719 | |
2ecd9d29 PZ |
4720 | void preempt_notifier_inc(void) |
4721 | { | |
b7203428 | 4722 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
4723 | } |
4724 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
4725 | ||
4726 | void preempt_notifier_dec(void) | |
4727 | { | |
b7203428 | 4728 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
4729 | } |
4730 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
4731 | ||
e107be36 | 4732 | /** |
80dd99b3 | 4733 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 4734 | * @notifier: notifier struct to register |
e107be36 AK |
4735 | */ |
4736 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
4737 | { | |
b7203428 | 4738 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
4739 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
4740 | ||
e107be36 AK |
4741 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
4742 | } | |
4743 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
4744 | ||
4745 | /** | |
4746 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 4747 | * @notifier: notifier struct to unregister |
e107be36 | 4748 | * |
d84525a8 | 4749 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
4750 | */ |
4751 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
4752 | { | |
4753 | hlist_del(¬ifier->link); | |
4754 | } | |
4755 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
4756 | ||
1cde2930 | 4757 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4758 | { |
4759 | struct preempt_notifier *notifier; | |
e107be36 | 4760 | |
b67bfe0d | 4761 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4762 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
4763 | } | |
4764 | ||
1cde2930 PZ |
4765 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
4766 | { | |
b7203428 | 4767 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4768 | __fire_sched_in_preempt_notifiers(curr); |
4769 | } | |
4770 | ||
e107be36 | 4771 | static void |
1cde2930 PZ |
4772 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4773 | struct task_struct *next) | |
e107be36 AK |
4774 | { |
4775 | struct preempt_notifier *notifier; | |
e107be36 | 4776 | |
b67bfe0d | 4777 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4778 | notifier->ops->sched_out(notifier, next); |
4779 | } | |
4780 | ||
1cde2930 PZ |
4781 | static __always_inline void |
4782 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
4783 | struct task_struct *next) | |
4784 | { | |
b7203428 | 4785 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4786 | __fire_sched_out_preempt_notifiers(curr, next); |
4787 | } | |
4788 | ||
6d6bc0ad | 4789 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4790 | |
1cde2930 | 4791 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4792 | { |
4793 | } | |
4794 | ||
1cde2930 | 4795 | static inline void |
e107be36 AK |
4796 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4797 | struct task_struct *next) | |
4798 | { | |
4799 | } | |
4800 | ||
6d6bc0ad | 4801 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4802 | |
31cb1bc0 | 4803 | static inline void prepare_task(struct task_struct *next) |
4804 | { | |
4805 | #ifdef CONFIG_SMP | |
4806 | /* | |
4807 | * Claim the task as running, we do this before switching to it | |
4808 | * such that any running task will have this set. | |
58877d34 | 4809 | * |
f3dd3f67 TD |
4810 | * See the smp_load_acquire(&p->on_cpu) case in ttwu() and |
4811 | * its ordering comment. | |
31cb1bc0 | 4812 | */ |
58877d34 | 4813 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 4814 | #endif |
4815 | } | |
4816 | ||
4817 | static inline void finish_task(struct task_struct *prev) | |
4818 | { | |
4819 | #ifdef CONFIG_SMP | |
4820 | /* | |
58877d34 PZ |
4821 | * This must be the very last reference to @prev from this CPU. After |
4822 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
4823 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 4824 | * finished. |
4825 | * | |
4826 | * In particular, the load of prev->state in finish_task_switch() must | |
4827 | * happen before this. | |
4828 | * | |
4829 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
4830 | */ | |
4831 | smp_store_release(&prev->on_cpu, 0); | |
4832 | #endif | |
4833 | } | |
4834 | ||
565790d2 PZ |
4835 | #ifdef CONFIG_SMP |
4836 | ||
8e5bad7d | 4837 | static void do_balance_callbacks(struct rq *rq, struct balance_callback *head) |
565790d2 PZ |
4838 | { |
4839 | void (*func)(struct rq *rq); | |
8e5bad7d | 4840 | struct balance_callback *next; |
565790d2 | 4841 | |
5cb9eaa3 | 4842 | lockdep_assert_rq_held(rq); |
565790d2 PZ |
4843 | |
4844 | while (head) { | |
4845 | func = (void (*)(struct rq *))head->func; | |
4846 | next = head->next; | |
4847 | head->next = NULL; | |
4848 | head = next; | |
4849 | ||
4850 | func(rq); | |
4851 | } | |
4852 | } | |
4853 | ||
ae792702 PZ |
4854 | static void balance_push(struct rq *rq); |
4855 | ||
04193d59 PZ |
4856 | /* |
4857 | * balance_push_callback is a right abuse of the callback interface and plays | |
4858 | * by significantly different rules. | |
4859 | * | |
4860 | * Where the normal balance_callback's purpose is to be ran in the same context | |
4861 | * that queued it (only later, when it's safe to drop rq->lock again), | |
4862 | * balance_push_callback is specifically targeted at __schedule(). | |
4863 | * | |
4864 | * This abuse is tolerated because it places all the unlikely/odd cases behind | |
4865 | * a single test, namely: rq->balance_callback == NULL. | |
4866 | */ | |
8e5bad7d | 4867 | struct balance_callback balance_push_callback = { |
ae792702 | 4868 | .next = NULL, |
8e5bad7d | 4869 | .func = balance_push, |
ae792702 PZ |
4870 | }; |
4871 | ||
8e5bad7d | 4872 | static inline struct balance_callback * |
04193d59 | 4873 | __splice_balance_callbacks(struct rq *rq, bool split) |
565790d2 | 4874 | { |
8e5bad7d | 4875 | struct balance_callback *head = rq->balance_callback; |
565790d2 | 4876 | |
04193d59 PZ |
4877 | if (likely(!head)) |
4878 | return NULL; | |
4879 | ||
5cb9eaa3 | 4880 | lockdep_assert_rq_held(rq); |
04193d59 PZ |
4881 | /* |
4882 | * Must not take balance_push_callback off the list when | |
4883 | * splice_balance_callbacks() and balance_callbacks() are not | |
4884 | * in the same rq->lock section. | |
4885 | * | |
4886 | * In that case it would be possible for __schedule() to interleave | |
4887 | * and observe the list empty. | |
4888 | */ | |
4889 | if (split && head == &balance_push_callback) | |
4890 | head = NULL; | |
4891 | else | |
565790d2 PZ |
4892 | rq->balance_callback = NULL; |
4893 | ||
4894 | return head; | |
4895 | } | |
4896 | ||
8e5bad7d | 4897 | static inline struct balance_callback *splice_balance_callbacks(struct rq *rq) |
04193d59 PZ |
4898 | { |
4899 | return __splice_balance_callbacks(rq, true); | |
4900 | } | |
4901 | ||
565790d2 PZ |
4902 | static void __balance_callbacks(struct rq *rq) |
4903 | { | |
04193d59 | 4904 | do_balance_callbacks(rq, __splice_balance_callbacks(rq, false)); |
565790d2 PZ |
4905 | } |
4906 | ||
8e5bad7d | 4907 | static inline void balance_callbacks(struct rq *rq, struct balance_callback *head) |
565790d2 PZ |
4908 | { |
4909 | unsigned long flags; | |
4910 | ||
4911 | if (unlikely(head)) { | |
5cb9eaa3 | 4912 | raw_spin_rq_lock_irqsave(rq, flags); |
565790d2 | 4913 | do_balance_callbacks(rq, head); |
5cb9eaa3 | 4914 | raw_spin_rq_unlock_irqrestore(rq, flags); |
565790d2 PZ |
4915 | } |
4916 | } | |
4917 | ||
4918 | #else | |
4919 | ||
4920 | static inline void __balance_callbacks(struct rq *rq) | |
4921 | { | |
4922 | } | |
4923 | ||
8e5bad7d | 4924 | static inline struct balance_callback *splice_balance_callbacks(struct rq *rq) |
565790d2 PZ |
4925 | { |
4926 | return NULL; | |
4927 | } | |
4928 | ||
8e5bad7d | 4929 | static inline void balance_callbacks(struct rq *rq, struct balance_callback *head) |
565790d2 PZ |
4930 | { |
4931 | } | |
4932 | ||
4933 | #endif | |
4934 | ||
269d5992 PZ |
4935 | static inline void |
4936 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4937 | { |
269d5992 PZ |
4938 | /* |
4939 | * Since the runqueue lock will be released by the next | |
4940 | * task (which is an invalid locking op but in the case | |
4941 | * of the scheduler it's an obvious special-case), so we | |
4942 | * do an early lockdep release here: | |
4943 | */ | |
4944 | rq_unpin_lock(rq, rf); | |
9ef7e7e3 | 4945 | spin_release(&__rq_lockp(rq)->dep_map, _THIS_IP_); |
31cb1bc0 | 4946 | #ifdef CONFIG_DEBUG_SPINLOCK |
4947 | /* this is a valid case when another task releases the spinlock */ | |
5cb9eaa3 | 4948 | rq_lockp(rq)->owner = next; |
31cb1bc0 | 4949 | #endif |
269d5992 PZ |
4950 | } |
4951 | ||
4952 | static inline void finish_lock_switch(struct rq *rq) | |
4953 | { | |
31cb1bc0 | 4954 | /* |
4955 | * If we are tracking spinlock dependencies then we have to | |
4956 | * fix up the runqueue lock - which gets 'carried over' from | |
4957 | * prev into current: | |
4958 | */ | |
9ef7e7e3 | 4959 | spin_acquire(&__rq_lockp(rq)->dep_map, 0, 0, _THIS_IP_); |
ae792702 | 4960 | __balance_callbacks(rq); |
5cb9eaa3 | 4961 | raw_spin_rq_unlock_irq(rq); |
31cb1bc0 | 4962 | } |
4963 | ||
325ea10c IM |
4964 | /* |
4965 | * NOP if the arch has not defined these: | |
4966 | */ | |
4967 | ||
4968 | #ifndef prepare_arch_switch | |
4969 | # define prepare_arch_switch(next) do { } while (0) | |
4970 | #endif | |
4971 | ||
4972 | #ifndef finish_arch_post_lock_switch | |
4973 | # define finish_arch_post_lock_switch() do { } while (0) | |
4974 | #endif | |
4975 | ||
5fbda3ec TG |
4976 | static inline void kmap_local_sched_out(void) |
4977 | { | |
4978 | #ifdef CONFIG_KMAP_LOCAL | |
4979 | if (unlikely(current->kmap_ctrl.idx)) | |
4980 | __kmap_local_sched_out(); | |
4981 | #endif | |
4982 | } | |
4983 | ||
4984 | static inline void kmap_local_sched_in(void) | |
4985 | { | |
4986 | #ifdef CONFIG_KMAP_LOCAL | |
4987 | if (unlikely(current->kmap_ctrl.idx)) | |
4988 | __kmap_local_sched_in(); | |
4989 | #endif | |
4990 | } | |
4991 | ||
4866cde0 NP |
4992 | /** |
4993 | * prepare_task_switch - prepare to switch tasks | |
4994 | * @rq: the runqueue preparing to switch | |
421cee29 | 4995 | * @prev: the current task that is being switched out |
4866cde0 NP |
4996 | * @next: the task we are going to switch to. |
4997 | * | |
4998 | * This is called with the rq lock held and interrupts off. It must | |
4999 | * be paired with a subsequent finish_task_switch after the context | |
5000 | * switch. | |
5001 | * | |
5002 | * prepare_task_switch sets up locking and calls architecture specific | |
5003 | * hooks. | |
5004 | */ | |
e107be36 AK |
5005 | static inline void |
5006 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
5007 | struct task_struct *next) | |
4866cde0 | 5008 | { |
0ed557aa | 5009 | kcov_prepare_switch(prev); |
43148951 | 5010 | sched_info_switch(rq, prev, next); |
fe4b04fa | 5011 | perf_event_task_sched_out(prev, next); |
d7822b1e | 5012 | rseq_preempt(prev); |
e107be36 | 5013 | fire_sched_out_preempt_notifiers(prev, next); |
5fbda3ec | 5014 | kmap_local_sched_out(); |
31cb1bc0 | 5015 | prepare_task(next); |
4866cde0 NP |
5016 | prepare_arch_switch(next); |
5017 | } | |
5018 | ||
1da177e4 LT |
5019 | /** |
5020 | * finish_task_switch - clean up after a task-switch | |
5021 | * @prev: the thread we just switched away from. | |
5022 | * | |
4866cde0 NP |
5023 | * finish_task_switch must be called after the context switch, paired |
5024 | * with a prepare_task_switch call before the context switch. | |
5025 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
5026 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
5027 | * |
5028 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 5029 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
5030 | * with the lock held can cause deadlocks; see schedule() for |
5031 | * details.) | |
dfa50b60 ON |
5032 | * |
5033 | * The context switch have flipped the stack from under us and restored the | |
5034 | * local variables which were saved when this task called schedule() in the | |
5035 | * past. prev == current is still correct but we need to recalculate this_rq | |
5036 | * because prev may have moved to another CPU. | |
1da177e4 | 5037 | */ |
dfa50b60 | 5038 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
5039 | __releases(rq->lock) |
5040 | { | |
dfa50b60 | 5041 | struct rq *rq = this_rq(); |
1da177e4 | 5042 | struct mm_struct *mm = rq->prev_mm; |
fa2c3254 | 5043 | unsigned int prev_state; |
1da177e4 | 5044 | |
609ca066 PZ |
5045 | /* |
5046 | * The previous task will have left us with a preempt_count of 2 | |
5047 | * because it left us after: | |
5048 | * | |
5049 | * schedule() | |
5050 | * preempt_disable(); // 1 | |
5051 | * __schedule() | |
5052 | * raw_spin_lock_irq(&rq->lock) // 2 | |
5053 | * | |
5054 | * Also, see FORK_PREEMPT_COUNT. | |
5055 | */ | |
e2bf1c4b PZ |
5056 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
5057 | "corrupted preempt_count: %s/%d/0x%x\n", | |
5058 | current->comm, current->pid, preempt_count())) | |
5059 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 5060 | |
1da177e4 LT |
5061 | rq->prev_mm = NULL; |
5062 | ||
5063 | /* | |
5064 | * A task struct has one reference for the use as "current". | |
c394cc9f | 5065 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
5066 | * schedule one last time. The schedule call will never return, and |
5067 | * the scheduled task must drop that reference. | |
95913d97 PZ |
5068 | * |
5069 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 5070 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
5071 | * running on another CPU and we could rave with its RUNNING -> DEAD |
5072 | * transition, resulting in a double drop. | |
1da177e4 | 5073 | */ |
2f064a59 | 5074 | prev_state = READ_ONCE(prev->__state); |
bf9fae9f | 5075 | vtime_task_switch(prev); |
a8d757ef | 5076 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 5077 | finish_task(prev); |
0fdcccfa | 5078 | tick_nohz_task_switch(); |
31cb1bc0 | 5079 | finish_lock_switch(rq); |
01f23e16 | 5080 | finish_arch_post_lock_switch(); |
0ed557aa | 5081 | kcov_finish_switch(current); |
5fbda3ec TG |
5082 | /* |
5083 | * kmap_local_sched_out() is invoked with rq::lock held and | |
5084 | * interrupts disabled. There is no requirement for that, but the | |
5085 | * sched out code does not have an interrupt enabled section. | |
5086 | * Restoring the maps on sched in does not require interrupts being | |
5087 | * disabled either. | |
5088 | */ | |
5089 | kmap_local_sched_in(); | |
e8fa1362 | 5090 | |
e107be36 | 5091 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 5092 | /* |
70216e18 MD |
5093 | * When switching through a kernel thread, the loop in |
5094 | * membarrier_{private,global}_expedited() may have observed that | |
5095 | * kernel thread and not issued an IPI. It is therefore possible to | |
5096 | * schedule between user->kernel->user threads without passing though | |
5097 | * switch_mm(). Membarrier requires a barrier after storing to | |
5098 | * rq->curr, before returning to userspace, so provide them here: | |
5099 | * | |
5100 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
5101 | * provided by mmdrop(), | |
5102 | * - a sync_core for SYNC_CORE. | |
306e0604 | 5103 | */ |
70216e18 MD |
5104 | if (mm) { |
5105 | membarrier_mm_sync_core_before_usermode(mm); | |
8d491de6 | 5106 | mmdrop_sched(mm); |
70216e18 | 5107 | } |
1cef1150 PZ |
5108 | if (unlikely(prev_state == TASK_DEAD)) { |
5109 | if (prev->sched_class->task_dead) | |
5110 | prev->sched_class->task_dead(prev); | |
68f24b08 | 5111 | |
1cef1150 PZ |
5112 | /* Task is done with its stack. */ |
5113 | put_task_stack(prev); | |
5114 | ||
0ff7b2cf | 5115 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 5116 | } |
99e5ada9 | 5117 | |
dfa50b60 | 5118 | return rq; |
1da177e4 LT |
5119 | } |
5120 | ||
5121 | /** | |
5122 | * schedule_tail - first thing a freshly forked thread must call. | |
5123 | * @prev: the thread we just switched away from. | |
5124 | */ | |
722a9f92 | 5125 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
5126 | __releases(rq->lock) |
5127 | { | |
609ca066 PZ |
5128 | /* |
5129 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
5130 | * finish_task_switch() for details. | |
5131 | * | |
5132 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
5133 | * and the preempt_enable() will end up enabling preemption (on | |
5134 | * PREEMPT_COUNT kernels). | |
5135 | */ | |
5136 | ||
13c2235b | 5137 | finish_task_switch(prev); |
1a43a14a | 5138 | preempt_enable(); |
70b97a7f | 5139 | |
1da177e4 | 5140 | if (current->set_child_tid) |
b488893a | 5141 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
5142 | |
5143 | calculate_sigpending(); | |
1da177e4 LT |
5144 | } |
5145 | ||
5146 | /* | |
dfa50b60 | 5147 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 5148 | */ |
04936948 | 5149 | static __always_inline struct rq * |
70b97a7f | 5150 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 5151 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 5152 | { |
e107be36 | 5153 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 5154 | |
9226d125 ZA |
5155 | /* |
5156 | * For paravirt, this is coupled with an exit in switch_to to | |
5157 | * combine the page table reload and the switch backend into | |
5158 | * one hypercall. | |
5159 | */ | |
224101ed | 5160 | arch_start_context_switch(prev); |
9226d125 | 5161 | |
306e0604 | 5162 | /* |
139d025c PZ |
5163 | * kernel -> kernel lazy + transfer active |
5164 | * user -> kernel lazy + mmgrab() active | |
5165 | * | |
5166 | * kernel -> user switch + mmdrop() active | |
5167 | * user -> user switch | |
306e0604 | 5168 | */ |
139d025c PZ |
5169 | if (!next->mm) { // to kernel |
5170 | enter_lazy_tlb(prev->active_mm, next); | |
5171 | ||
5172 | next->active_mm = prev->active_mm; | |
5173 | if (prev->mm) // from user | |
5174 | mmgrab(prev->active_mm); | |
5175 | else | |
5176 | prev->active_mm = NULL; | |
5177 | } else { // to user | |
227a4aad | 5178 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
5179 | /* |
5180 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 5181 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
5182 | * |
5183 | * The below provides this either through switch_mm(), or in | |
5184 | * case 'prev->active_mm == next->mm' through | |
5185 | * finish_task_switch()'s mmdrop(). | |
5186 | */ | |
139d025c | 5187 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
bd74fdae | 5188 | lru_gen_use_mm(next->mm); |
1da177e4 | 5189 | |
139d025c PZ |
5190 | if (!prev->mm) { // from kernel |
5191 | /* will mmdrop() in finish_task_switch(). */ | |
5192 | rq->prev_mm = prev->active_mm; | |
5193 | prev->active_mm = NULL; | |
5194 | } | |
1da177e4 | 5195 | } |
92509b73 | 5196 | |
cb42c9a3 | 5197 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 5198 | |
269d5992 | 5199 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
5200 | |
5201 | /* Here we just switch the register state and the stack. */ | |
5202 | switch_to(prev, next, prev); | |
dd41f596 | 5203 | barrier(); |
dfa50b60 ON |
5204 | |
5205 | return finish_task_switch(prev); | |
1da177e4 LT |
5206 | } |
5207 | ||
5208 | /* | |
1c3e8264 | 5209 | * nr_running and nr_context_switches: |
1da177e4 LT |
5210 | * |
5211 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 5212 | * threads, total number of context switches performed since bootup. |
1da177e4 | 5213 | */ |
01aee8fd | 5214 | unsigned int nr_running(void) |
1da177e4 | 5215 | { |
01aee8fd | 5216 | unsigned int i, sum = 0; |
1da177e4 LT |
5217 | |
5218 | for_each_online_cpu(i) | |
5219 | sum += cpu_rq(i)->nr_running; | |
5220 | ||
5221 | return sum; | |
f711f609 | 5222 | } |
1da177e4 | 5223 | |
2ee507c4 | 5224 | /* |
d1ccc66d | 5225 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
5226 | * |
5227 | * Caution: this function does not check that the caller has disabled | |
5228 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
5229 | * race. The caller is responsible to use it correctly, for example: | |
5230 | * | |
dfcb245e | 5231 | * - from a non-preemptible section (of course) |
00cc1633 DD |
5232 | * |
5233 | * - from a thread that is bound to a single CPU | |
5234 | * | |
5235 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
5236 | */ |
5237 | bool single_task_running(void) | |
5238 | { | |
00cc1633 | 5239 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
5240 | } |
5241 | EXPORT_SYMBOL(single_task_running); | |
5242 | ||
1da177e4 | 5243 | unsigned long long nr_context_switches(void) |
46cb4b7c | 5244 | { |
cc94abfc SR |
5245 | int i; |
5246 | unsigned long long sum = 0; | |
46cb4b7c | 5247 | |
0a945022 | 5248 | for_each_possible_cpu(i) |
1da177e4 | 5249 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 5250 | |
1da177e4 LT |
5251 | return sum; |
5252 | } | |
483b4ee6 | 5253 | |
145d952a DL |
5254 | /* |
5255 | * Consumers of these two interfaces, like for example the cpuidle menu | |
5256 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
5257 | * for a CPU that has IO-wait which might not even end up running the task when | |
5258 | * it does become runnable. | |
5259 | */ | |
5260 | ||
8fc2858e | 5261 | unsigned int nr_iowait_cpu(int cpu) |
145d952a DL |
5262 | { |
5263 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
5264 | } | |
5265 | ||
e33a9bba | 5266 | /* |
b19a888c | 5267 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
5268 | * |
5269 | * The idea behind IO-wait account is to account the idle time that we could | |
5270 | * have spend running if it were not for IO. That is, if we were to improve the | |
5271 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
5272 | * | |
5273 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
5274 | * idle time as IO-wait, because if the storage were faster, it could've been | |
5275 | * running and we'd not be idle. | |
5276 | * | |
5277 | * This has been extended to SMP, by doing the same for each CPU. This however | |
5278 | * is broken. | |
5279 | * | |
5280 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
5281 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
5282 | * though, if the storage were faster, both could've ran at the same time, | |
5283 | * utilising both CPUs. | |
5284 | * | |
5285 | * This means, that when looking globally, the current IO-wait accounting on | |
5286 | * SMP is a lower bound, by reason of under accounting. | |
5287 | * | |
5288 | * Worse, since the numbers are provided per CPU, they are sometimes | |
5289 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
5290 | * associated with any one particular CPU, it can wake to another CPU than it | |
5291 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
5292 | * | |
5293 | * Task CPU affinities can make all that even more 'interesting'. | |
5294 | */ | |
5295 | ||
97455168 | 5296 | unsigned int nr_iowait(void) |
1da177e4 | 5297 | { |
97455168 | 5298 | unsigned int i, sum = 0; |
483b4ee6 | 5299 | |
0a945022 | 5300 | for_each_possible_cpu(i) |
145d952a | 5301 | sum += nr_iowait_cpu(i); |
46cb4b7c | 5302 | |
1da177e4 LT |
5303 | return sum; |
5304 | } | |
483b4ee6 | 5305 | |
dd41f596 | 5306 | #ifdef CONFIG_SMP |
8a0be9ef | 5307 | |
46cb4b7c | 5308 | /* |
38022906 PZ |
5309 | * sched_exec - execve() is a valuable balancing opportunity, because at |
5310 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 5311 | */ |
38022906 | 5312 | void sched_exec(void) |
46cb4b7c | 5313 | { |
38022906 | 5314 | struct task_struct *p = current; |
1da177e4 | 5315 | unsigned long flags; |
0017d735 | 5316 | int dest_cpu; |
46cb4b7c | 5317 | |
8f42ced9 | 5318 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 5319 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
5320 | if (dest_cpu == smp_processor_id()) |
5321 | goto unlock; | |
38022906 | 5322 | |
8f42ced9 | 5323 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 5324 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 5325 | |
8f42ced9 PZ |
5326 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5327 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
5328 | return; |
5329 | } | |
0017d735 | 5330 | unlock: |
8f42ced9 | 5331 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 5332 | } |
dd41f596 | 5333 | |
1da177e4 LT |
5334 | #endif |
5335 | ||
1da177e4 | 5336 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 5337 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
5338 | |
5339 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 5340 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 5341 | |
6075620b GG |
5342 | /* |
5343 | * The function fair_sched_class.update_curr accesses the struct curr | |
5344 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
5345 | * we observe a high rate of cache misses in practice. | |
5346 | * Prefetching this data results in improved performance. | |
5347 | */ | |
5348 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
5349 | { | |
5350 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
5351 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
5352 | #else | |
5353 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
5354 | #endif | |
5355 | prefetch(curr); | |
5356 | prefetch(&curr->exec_start); | |
5357 | } | |
5358 | ||
c5f8d995 HS |
5359 | /* |
5360 | * Return accounted runtime for the task. | |
5361 | * In case the task is currently running, return the runtime plus current's | |
5362 | * pending runtime that have not been accounted yet. | |
5363 | */ | |
5364 | unsigned long long task_sched_runtime(struct task_struct *p) | |
5365 | { | |
eb580751 | 5366 | struct rq_flags rf; |
c5f8d995 | 5367 | struct rq *rq; |
6e998916 | 5368 | u64 ns; |
c5f8d995 | 5369 | |
911b2898 PZ |
5370 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
5371 | /* | |
97fb7a0a | 5372 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
5373 | * So we have a optimization chance when the task's delta_exec is 0. |
5374 | * Reading ->on_cpu is racy, but this is ok. | |
5375 | * | |
d1ccc66d IM |
5376 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
5377 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 5378 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
5379 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
5380 | * been accounted, so we're correct here as well. | |
911b2898 | 5381 | */ |
da0c1e65 | 5382 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
5383 | return p->se.sum_exec_runtime; |
5384 | #endif | |
5385 | ||
eb580751 | 5386 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
5387 | /* |
5388 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
5389 | * project cycles that may never be accounted to this | |
5390 | * thread, breaking clock_gettime(). | |
5391 | */ | |
5392 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 5393 | prefetch_curr_exec_start(p); |
6e998916 SG |
5394 | update_rq_clock(rq); |
5395 | p->sched_class->update_curr(rq); | |
5396 | } | |
5397 | ns = p->se.sum_exec_runtime; | |
eb580751 | 5398 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
5399 | |
5400 | return ns; | |
5401 | } | |
48f24c4d | 5402 | |
c006fac5 PT |
5403 | #ifdef CONFIG_SCHED_DEBUG |
5404 | static u64 cpu_resched_latency(struct rq *rq) | |
5405 | { | |
5406 | int latency_warn_ms = READ_ONCE(sysctl_resched_latency_warn_ms); | |
5407 | u64 resched_latency, now = rq_clock(rq); | |
5408 | static bool warned_once; | |
5409 | ||
5410 | if (sysctl_resched_latency_warn_once && warned_once) | |
5411 | return 0; | |
5412 | ||
5413 | if (!need_resched() || !latency_warn_ms) | |
5414 | return 0; | |
5415 | ||
5416 | if (system_state == SYSTEM_BOOTING) | |
5417 | return 0; | |
5418 | ||
5419 | if (!rq->last_seen_need_resched_ns) { | |
5420 | rq->last_seen_need_resched_ns = now; | |
5421 | rq->ticks_without_resched = 0; | |
5422 | return 0; | |
5423 | } | |
5424 | ||
5425 | rq->ticks_without_resched++; | |
5426 | resched_latency = now - rq->last_seen_need_resched_ns; | |
5427 | if (resched_latency <= latency_warn_ms * NSEC_PER_MSEC) | |
5428 | return 0; | |
5429 | ||
5430 | warned_once = true; | |
5431 | ||
5432 | return resched_latency; | |
5433 | } | |
5434 | ||
5435 | static int __init setup_resched_latency_warn_ms(char *str) | |
5436 | { | |
5437 | long val; | |
5438 | ||
5439 | if ((kstrtol(str, 0, &val))) { | |
5440 | pr_warn("Unable to set resched_latency_warn_ms\n"); | |
5441 | return 1; | |
5442 | } | |
5443 | ||
5444 | sysctl_resched_latency_warn_ms = val; | |
5445 | return 1; | |
5446 | } | |
5447 | __setup("resched_latency_warn_ms=", setup_resched_latency_warn_ms); | |
5448 | #else | |
5449 | static inline u64 cpu_resched_latency(struct rq *rq) { return 0; } | |
5450 | #endif /* CONFIG_SCHED_DEBUG */ | |
5451 | ||
7835b98b CL |
5452 | /* |
5453 | * This function gets called by the timer code, with HZ frequency. | |
5454 | * We call it with interrupts disabled. | |
7835b98b CL |
5455 | */ |
5456 | void scheduler_tick(void) | |
5457 | { | |
7835b98b CL |
5458 | int cpu = smp_processor_id(); |
5459 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5460 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 5461 | struct rq_flags rf; |
b4eccf5f | 5462 | unsigned long thermal_pressure; |
c006fac5 | 5463 | u64 resched_latency; |
3e51f33f | 5464 | |
1567c3e3 | 5465 | arch_scale_freq_tick(); |
3e51f33f | 5466 | sched_clock_tick(); |
dd41f596 | 5467 | |
8a8c69c3 PZ |
5468 | rq_lock(rq, &rf); |
5469 | ||
3e51f33f | 5470 | update_rq_clock(rq); |
b4eccf5f | 5471 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 5472 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 5473 | curr->sched_class->task_tick(rq, curr, 0); |
c006fac5 PT |
5474 | if (sched_feat(LATENCY_WARN)) |
5475 | resched_latency = cpu_resched_latency(rq); | |
3289bdb4 | 5476 | calc_global_load_tick(rq); |
4feee7d1 | 5477 | sched_core_tick(rq); |
8a8c69c3 PZ |
5478 | |
5479 | rq_unlock(rq, &rf); | |
7835b98b | 5480 | |
c006fac5 PT |
5481 | if (sched_feat(LATENCY_WARN) && resched_latency) |
5482 | resched_latency_warn(cpu, resched_latency); | |
5483 | ||
e9d2b064 | 5484 | perf_event_task_tick(); |
e220d2dc | 5485 | |
e418e1c2 | 5486 | #ifdef CONFIG_SMP |
6eb57e0d | 5487 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 5488 | trigger_load_balance(rq); |
e418e1c2 | 5489 | #endif |
1da177e4 LT |
5490 | } |
5491 | ||
265f22a9 | 5492 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
5493 | |
5494 | struct tick_work { | |
5495 | int cpu; | |
b55bd585 | 5496 | atomic_t state; |
d84b3131 FW |
5497 | struct delayed_work work; |
5498 | }; | |
b55bd585 PM |
5499 | /* Values for ->state, see diagram below. */ |
5500 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
5501 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
5502 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
5503 | ||
5504 | /* | |
5505 | * State diagram for ->state: | |
5506 | * | |
5507 | * | |
5508 | * TICK_SCHED_REMOTE_OFFLINE | |
5509 | * | ^ | |
5510 | * | | | |
5511 | * | | sched_tick_remote() | |
5512 | * | | | |
5513 | * | | | |
5514 | * +--TICK_SCHED_REMOTE_OFFLINING | |
5515 | * | ^ | |
5516 | * | | | |
5517 | * sched_tick_start() | | sched_tick_stop() | |
5518 | * | | | |
5519 | * V | | |
5520 | * TICK_SCHED_REMOTE_RUNNING | |
5521 | * | |
5522 | * | |
5523 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
5524 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
5525 | */ | |
d84b3131 FW |
5526 | |
5527 | static struct tick_work __percpu *tick_work_cpu; | |
5528 | ||
5529 | static void sched_tick_remote(struct work_struct *work) | |
5530 | { | |
5531 | struct delayed_work *dwork = to_delayed_work(work); | |
5532 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
5533 | int cpu = twork->cpu; | |
5534 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 5535 | struct task_struct *curr; |
d84b3131 | 5536 | struct rq_flags rf; |
d9c0ffca | 5537 | u64 delta; |
b55bd585 | 5538 | int os; |
d84b3131 FW |
5539 | |
5540 | /* | |
5541 | * Handle the tick only if it appears the remote CPU is running in full | |
5542 | * dynticks mode. The check is racy by nature, but missing a tick or | |
5543 | * having one too much is no big deal because the scheduler tick updates | |
5544 | * statistics and checks timeslices in a time-independent way, regardless | |
5545 | * of when exactly it is running. | |
5546 | */ | |
488603b8 | 5547 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 5548 | goto out_requeue; |
d84b3131 | 5549 | |
d9c0ffca FW |
5550 | rq_lock_irq(rq, &rf); |
5551 | curr = rq->curr; | |
488603b8 | 5552 | if (cpu_is_offline(cpu)) |
d9c0ffca | 5553 | goto out_unlock; |
d84b3131 | 5554 | |
d9c0ffca | 5555 | update_rq_clock(rq); |
d9c0ffca | 5556 | |
488603b8 SW |
5557 | if (!is_idle_task(curr)) { |
5558 | /* | |
5559 | * Make sure the next tick runs within a reasonable | |
5560 | * amount of time. | |
5561 | */ | |
5562 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
5563 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
5564 | } | |
d9c0ffca FW |
5565 | curr->sched_class->task_tick(rq, curr, 0); |
5566 | ||
ebc0f83c | 5567 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
5568 | out_unlock: |
5569 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 5570 | out_requeue: |
ebc0f83c | 5571 | |
d84b3131 FW |
5572 | /* |
5573 | * Run the remote tick once per second (1Hz). This arbitrary | |
5574 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
5575 | * to keep scheduler internal stats reasonably up to date. But |
5576 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 5577 | */ |
b55bd585 PM |
5578 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
5579 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
5580 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
5581 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
5582 | } |
5583 | ||
5584 | static void sched_tick_start(int cpu) | |
5585 | { | |
b55bd585 | 5586 | int os; |
d84b3131 FW |
5587 | struct tick_work *twork; |
5588 | ||
04d4e665 | 5589 | if (housekeeping_cpu(cpu, HK_TYPE_TICK)) |
d84b3131 FW |
5590 | return; |
5591 | ||
5592 | WARN_ON_ONCE(!tick_work_cpu); | |
5593 | ||
5594 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5595 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
5596 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
5597 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
5598 | twork->cpu = cpu; | |
5599 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
5600 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
5601 | } | |
d84b3131 FW |
5602 | } |
5603 | ||
5604 | #ifdef CONFIG_HOTPLUG_CPU | |
5605 | static void sched_tick_stop(int cpu) | |
5606 | { | |
5607 | struct tick_work *twork; | |
b55bd585 | 5608 | int os; |
d84b3131 | 5609 | |
04d4e665 | 5610 | if (housekeeping_cpu(cpu, HK_TYPE_TICK)) |
d84b3131 FW |
5611 | return; |
5612 | ||
5613 | WARN_ON_ONCE(!tick_work_cpu); | |
5614 | ||
5615 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5616 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
5617 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
5618 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
5619 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
5620 | } |
5621 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5622 | ||
5623 | int __init sched_tick_offload_init(void) | |
5624 | { | |
5625 | tick_work_cpu = alloc_percpu(struct tick_work); | |
5626 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
5627 | return 0; |
5628 | } | |
5629 | ||
5630 | #else /* !CONFIG_NO_HZ_FULL */ | |
5631 | static inline void sched_tick_start(int cpu) { } | |
5632 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 5633 | #endif |
1da177e4 | 5634 | |
c1a280b6 | 5635 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 5636 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
5637 | /* |
5638 | * If the value passed in is equal to the current preempt count | |
5639 | * then we just disabled preemption. Start timing the latency. | |
5640 | */ | |
5641 | static inline void preempt_latency_start(int val) | |
5642 | { | |
5643 | if (preempt_count() == val) { | |
5644 | unsigned long ip = get_lock_parent_ip(); | |
5645 | #ifdef CONFIG_DEBUG_PREEMPT | |
5646 | current->preempt_disable_ip = ip; | |
5647 | #endif | |
5648 | trace_preempt_off(CALLER_ADDR0, ip); | |
5649 | } | |
5650 | } | |
7e49fcce | 5651 | |
edafe3a5 | 5652 | void preempt_count_add(int val) |
1da177e4 | 5653 | { |
6cd8a4bb | 5654 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5655 | /* |
5656 | * Underflow? | |
5657 | */ | |
9a11b49a IM |
5658 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5659 | return; | |
6cd8a4bb | 5660 | #endif |
bdb43806 | 5661 | __preempt_count_add(val); |
6cd8a4bb | 5662 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5663 | /* |
5664 | * Spinlock count overflowing soon? | |
5665 | */ | |
33859f7f MOS |
5666 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5667 | PREEMPT_MASK - 10); | |
6cd8a4bb | 5668 | #endif |
47252cfb | 5669 | preempt_latency_start(val); |
1da177e4 | 5670 | } |
bdb43806 | 5671 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 5672 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 5673 | |
47252cfb SR |
5674 | /* |
5675 | * If the value passed in equals to the current preempt count | |
5676 | * then we just enabled preemption. Stop timing the latency. | |
5677 | */ | |
5678 | static inline void preempt_latency_stop(int val) | |
5679 | { | |
5680 | if (preempt_count() == val) | |
5681 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
5682 | } | |
5683 | ||
edafe3a5 | 5684 | void preempt_count_sub(int val) |
1da177e4 | 5685 | { |
6cd8a4bb | 5686 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5687 | /* |
5688 | * Underflow? | |
5689 | */ | |
01e3eb82 | 5690 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5691 | return; |
1da177e4 LT |
5692 | /* |
5693 | * Is the spinlock portion underflowing? | |
5694 | */ | |
9a11b49a IM |
5695 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5696 | !(preempt_count() & PREEMPT_MASK))) | |
5697 | return; | |
6cd8a4bb | 5698 | #endif |
9a11b49a | 5699 | |
47252cfb | 5700 | preempt_latency_stop(val); |
bdb43806 | 5701 | __preempt_count_sub(val); |
1da177e4 | 5702 | } |
bdb43806 | 5703 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 5704 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 5705 | |
47252cfb SR |
5706 | #else |
5707 | static inline void preempt_latency_start(int val) { } | |
5708 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
5709 | #endif |
5710 | ||
59ddbcb2 IM |
5711 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
5712 | { | |
5713 | #ifdef CONFIG_DEBUG_PREEMPT | |
5714 | return p->preempt_disable_ip; | |
5715 | #else | |
5716 | return 0; | |
5717 | #endif | |
5718 | } | |
5719 | ||
1da177e4 | 5720 | /* |
dd41f596 | 5721 | * Print scheduling while atomic bug: |
1da177e4 | 5722 | */ |
dd41f596 | 5723 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5724 | { |
d1c6d149 VN |
5725 | /* Save this before calling printk(), since that will clobber it */ |
5726 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
5727 | ||
664dfa65 DJ |
5728 | if (oops_in_progress) |
5729 | return; | |
5730 | ||
3df0fc5b PZ |
5731 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
5732 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 5733 | |
dd41f596 | 5734 | debug_show_held_locks(prev); |
e21f5b15 | 5735 | print_modules(); |
dd41f596 IM |
5736 | if (irqs_disabled()) |
5737 | print_irqtrace_events(prev); | |
d1c6d149 VN |
5738 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
5739 | && in_atomic_preempt_off()) { | |
8f47b187 | 5740 | pr_err("Preemption disabled at:"); |
2062a4e8 | 5741 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 5742 | } |
748c7201 DBO |
5743 | if (panic_on_warn) |
5744 | panic("scheduling while atomic\n"); | |
5745 | ||
6135fc1e | 5746 | dump_stack(); |
373d4d09 | 5747 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 5748 | } |
1da177e4 | 5749 | |
dd41f596 IM |
5750 | /* |
5751 | * Various schedule()-time debugging checks and statistics: | |
5752 | */ | |
312364f3 | 5753 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 5754 | { |
0d9e2632 | 5755 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
5756 | if (task_stack_end_corrupted(prev)) |
5757 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
5758 | |
5759 | if (task_scs_end_corrupted(prev)) | |
5760 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 5761 | #endif |
b99def8b | 5762 | |
312364f3 | 5763 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
2f064a59 | 5764 | if (!preempt && READ_ONCE(prev->__state) && prev->non_block_count) { |
312364f3 DV |
5765 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", |
5766 | prev->comm, prev->pid, prev->non_block_count); | |
5767 | dump_stack(); | |
5768 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
5769 | } | |
5770 | #endif | |
5771 | ||
1dc0fffc | 5772 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 5773 | __schedule_bug(prev); |
1dc0fffc PZ |
5774 | preempt_count_set(PREEMPT_DISABLED); |
5775 | } | |
b3fbab05 | 5776 | rcu_sleep_check(); |
9f68b5b7 | 5777 | SCHED_WARN_ON(ct_state() == CONTEXT_USER); |
dd41f596 | 5778 | |
1da177e4 LT |
5779 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5780 | ||
ae92882e | 5781 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
5782 | } |
5783 | ||
457d1f46 CY |
5784 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
5785 | struct rq_flags *rf) | |
5786 | { | |
5787 | #ifdef CONFIG_SMP | |
5788 | const struct sched_class *class; | |
5789 | /* | |
5790 | * We must do the balancing pass before put_prev_task(), such | |
5791 | * that when we release the rq->lock the task is in the same | |
5792 | * state as before we took rq->lock. | |
5793 | * | |
5794 | * We can terminate the balance pass as soon as we know there is | |
5795 | * a runnable task of @class priority or higher. | |
5796 | */ | |
5797 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
5798 | if (class->balance(rq, prev, rf)) | |
5799 | break; | |
5800 | } | |
5801 | #endif | |
5802 | ||
5803 | put_prev_task(rq, prev); | |
5804 | } | |
5805 | ||
dd41f596 IM |
5806 | /* |
5807 | * Pick up the highest-prio task: | |
5808 | */ | |
5809 | static inline struct task_struct * | |
539f6512 | 5810 | __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 5811 | { |
49ee5768 | 5812 | const struct sched_class *class; |
dd41f596 | 5813 | struct task_struct *p; |
1da177e4 LT |
5814 | |
5815 | /* | |
0ba87bb2 PZ |
5816 | * Optimization: we know that if all tasks are in the fair class we can |
5817 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 5818 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 5819 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 5820 | */ |
546a3fee | 5821 | if (likely(!sched_class_above(prev->sched_class, &fair_sched_class) && |
0ba87bb2 PZ |
5822 | rq->nr_running == rq->cfs.h_nr_running)) { |
5823 | ||
5d7d6056 | 5824 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 5825 | if (unlikely(p == RETRY_TASK)) |
67692435 | 5826 | goto restart; |
6ccdc84b | 5827 | |
1699949d | 5828 | /* Assume the next prioritized class is idle_sched_class */ |
5d7d6056 | 5829 | if (!p) { |
f488e105 | 5830 | put_prev_task(rq, prev); |
98c2f700 | 5831 | p = pick_next_task_idle(rq); |
f488e105 | 5832 | } |
6ccdc84b PZ |
5833 | |
5834 | return p; | |
1da177e4 LT |
5835 | } |
5836 | ||
67692435 | 5837 | restart: |
457d1f46 | 5838 | put_prev_task_balance(rq, prev, rf); |
67692435 | 5839 | |
34f971f6 | 5840 | for_each_class(class) { |
98c2f700 | 5841 | p = class->pick_next_task(rq); |
67692435 | 5842 | if (p) |
dd41f596 | 5843 | return p; |
dd41f596 | 5844 | } |
34f971f6 | 5845 | |
bc9ffef3 | 5846 | BUG(); /* The idle class should always have a runnable task. */ |
dd41f596 | 5847 | } |
1da177e4 | 5848 | |
9edeaea1 | 5849 | #ifdef CONFIG_SCHED_CORE |
539f6512 PZ |
5850 | static inline bool is_task_rq_idle(struct task_struct *t) |
5851 | { | |
5852 | return (task_rq(t)->idle == t); | |
5853 | } | |
5854 | ||
5855 | static inline bool cookie_equals(struct task_struct *a, unsigned long cookie) | |
5856 | { | |
5857 | return is_task_rq_idle(a) || (a->core_cookie == cookie); | |
5858 | } | |
5859 | ||
5860 | static inline bool cookie_match(struct task_struct *a, struct task_struct *b) | |
5861 | { | |
5862 | if (is_task_rq_idle(a) || is_task_rq_idle(b)) | |
5863 | return true; | |
5864 | ||
5865 | return a->core_cookie == b->core_cookie; | |
5866 | } | |
5867 | ||
bc9ffef3 | 5868 | static inline struct task_struct *pick_task(struct rq *rq) |
539f6512 | 5869 | { |
bc9ffef3 PZ |
5870 | const struct sched_class *class; |
5871 | struct task_struct *p; | |
539f6512 | 5872 | |
bc9ffef3 PZ |
5873 | for_each_class(class) { |
5874 | p = class->pick_task(rq); | |
5875 | if (p) | |
5876 | return p; | |
539f6512 PZ |
5877 | } |
5878 | ||
bc9ffef3 | 5879 | BUG(); /* The idle class should always have a runnable task. */ |
539f6512 PZ |
5880 | } |
5881 | ||
c6047c2e JFG |
5882 | extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi); |
5883 | ||
5b6547ed PZ |
5884 | static void queue_core_balance(struct rq *rq); |
5885 | ||
539f6512 PZ |
5886 | static struct task_struct * |
5887 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
5888 | { | |
bc9ffef3 | 5889 | struct task_struct *next, *p, *max = NULL; |
539f6512 | 5890 | const struct cpumask *smt_mask; |
c6047c2e | 5891 | bool fi_before = false; |
4feee7d1 | 5892 | bool core_clock_updated = (rq == rq->core); |
bc9ffef3 PZ |
5893 | unsigned long cookie; |
5894 | int i, cpu, occ = 0; | |
5895 | struct rq *rq_i; | |
539f6512 | 5896 | bool need_sync; |
539f6512 PZ |
5897 | |
5898 | if (!sched_core_enabled(rq)) | |
5899 | return __pick_next_task(rq, prev, rf); | |
5900 | ||
5901 | cpu = cpu_of(rq); | |
5902 | ||
5903 | /* Stopper task is switching into idle, no need core-wide selection. */ | |
5904 | if (cpu_is_offline(cpu)) { | |
5905 | /* | |
5906 | * Reset core_pick so that we don't enter the fastpath when | |
5907 | * coming online. core_pick would already be migrated to | |
5908 | * another cpu during offline. | |
5909 | */ | |
5910 | rq->core_pick = NULL; | |
5911 | return __pick_next_task(rq, prev, rf); | |
5912 | } | |
5913 | ||
5914 | /* | |
5915 | * If there were no {en,de}queues since we picked (IOW, the task | |
5916 | * pointers are all still valid), and we haven't scheduled the last | |
5917 | * pick yet, do so now. | |
5918 | * | |
5919 | * rq->core_pick can be NULL if no selection was made for a CPU because | |
5920 | * it was either offline or went offline during a sibling's core-wide | |
5921 | * selection. In this case, do a core-wide selection. | |
5922 | */ | |
5923 | if (rq->core->core_pick_seq == rq->core->core_task_seq && | |
5924 | rq->core->core_pick_seq != rq->core_sched_seq && | |
5925 | rq->core_pick) { | |
5926 | WRITE_ONCE(rq->core_sched_seq, rq->core->core_pick_seq); | |
5927 | ||
5928 | next = rq->core_pick; | |
5929 | if (next != prev) { | |
5930 | put_prev_task(rq, prev); | |
5931 | set_next_task(rq, next); | |
5932 | } | |
5933 | ||
5934 | rq->core_pick = NULL; | |
5b6547ed | 5935 | goto out; |
539f6512 PZ |
5936 | } |
5937 | ||
5938 | put_prev_task_balance(rq, prev, rf); | |
5939 | ||
5940 | smt_mask = cpu_smt_mask(cpu); | |
7afbba11 JFG |
5941 | need_sync = !!rq->core->core_cookie; |
5942 | ||
5943 | /* reset state */ | |
5944 | rq->core->core_cookie = 0UL; | |
4feee7d1 JD |
5945 | if (rq->core->core_forceidle_count) { |
5946 | if (!core_clock_updated) { | |
5947 | update_rq_clock(rq->core); | |
5948 | core_clock_updated = true; | |
5949 | } | |
5950 | sched_core_account_forceidle(rq); | |
5951 | /* reset after accounting force idle */ | |
5952 | rq->core->core_forceidle_start = 0; | |
5953 | rq->core->core_forceidle_count = 0; | |
5954 | rq->core->core_forceidle_occupation = 0; | |
7afbba11 JFG |
5955 | need_sync = true; |
5956 | fi_before = true; | |
7afbba11 | 5957 | } |
539f6512 PZ |
5958 | |
5959 | /* | |
5960 | * core->core_task_seq, core->core_pick_seq, rq->core_sched_seq | |
5961 | * | |
5962 | * @task_seq guards the task state ({en,de}queues) | |
5963 | * @pick_seq is the @task_seq we did a selection on | |
5964 | * @sched_seq is the @pick_seq we scheduled | |
5965 | * | |
5966 | * However, preemptions can cause multiple picks on the same task set. | |
5967 | * 'Fix' this by also increasing @task_seq for every pick. | |
5968 | */ | |
5969 | rq->core->core_task_seq++; | |
539f6512 | 5970 | |
7afbba11 JFG |
5971 | /* |
5972 | * Optimize for common case where this CPU has no cookies | |
5973 | * and there are no cookied tasks running on siblings. | |
5974 | */ | |
5975 | if (!need_sync) { | |
bc9ffef3 | 5976 | next = pick_task(rq); |
7afbba11 JFG |
5977 | if (!next->core_cookie) { |
5978 | rq->core_pick = NULL; | |
c6047c2e JFG |
5979 | /* |
5980 | * For robustness, update the min_vruntime_fi for | |
5981 | * unconstrained picks as well. | |
5982 | */ | |
5983 | WARN_ON_ONCE(fi_before); | |
5984 | task_vruntime_update(rq, next, false); | |
5b6547ed | 5985 | goto out_set_next; |
7afbba11 | 5986 | } |
8039e96f | 5987 | } |
7afbba11 | 5988 | |
bc9ffef3 PZ |
5989 | /* |
5990 | * For each thread: do the regular task pick and find the max prio task | |
5991 | * amongst them. | |
5992 | * | |
5993 | * Tie-break prio towards the current CPU | |
5994 | */ | |
5995 | for_each_cpu_wrap(i, smt_mask, cpu) { | |
5996 | rq_i = cpu_rq(i); | |
539f6512 | 5997 | |
4feee7d1 JD |
5998 | /* |
5999 | * Current cpu always has its clock updated on entrance to | |
6000 | * pick_next_task(). If the current cpu is not the core, | |
6001 | * the core may also have been updated above. | |
6002 | */ | |
6003 | if (i != cpu && (rq_i != rq->core || !core_clock_updated)) | |
539f6512 | 6004 | update_rq_clock(rq_i); |
bc9ffef3 PZ |
6005 | |
6006 | p = rq_i->core_pick = pick_task(rq_i); | |
6007 | if (!max || prio_less(max, p, fi_before)) | |
6008 | max = p; | |
539f6512 PZ |
6009 | } |
6010 | ||
bc9ffef3 PZ |
6011 | cookie = rq->core->core_cookie = max->core_cookie; |
6012 | ||
539f6512 | 6013 | /* |
bc9ffef3 PZ |
6014 | * For each thread: try and find a runnable task that matches @max or |
6015 | * force idle. | |
539f6512 | 6016 | */ |
bc9ffef3 PZ |
6017 | for_each_cpu(i, smt_mask) { |
6018 | rq_i = cpu_rq(i); | |
6019 | p = rq_i->core_pick; | |
539f6512 | 6020 | |
bc9ffef3 PZ |
6021 | if (!cookie_equals(p, cookie)) { |
6022 | p = NULL; | |
6023 | if (cookie) | |
6024 | p = sched_core_find(rq_i, cookie); | |
7afbba11 | 6025 | if (!p) |
bc9ffef3 PZ |
6026 | p = idle_sched_class.pick_task(rq_i); |
6027 | } | |
539f6512 | 6028 | |
bc9ffef3 | 6029 | rq_i->core_pick = p; |
d2dfa17b | 6030 | |
bc9ffef3 PZ |
6031 | if (p == rq_i->idle) { |
6032 | if (rq_i->nr_running) { | |
4feee7d1 | 6033 | rq->core->core_forceidle_count++; |
c6047c2e JFG |
6034 | if (!fi_before) |
6035 | rq->core->core_forceidle_seq++; | |
6036 | } | |
bc9ffef3 PZ |
6037 | } else { |
6038 | occ++; | |
539f6512 | 6039 | } |
539f6512 PZ |
6040 | } |
6041 | ||
4feee7d1 | 6042 | if (schedstat_enabled() && rq->core->core_forceidle_count) { |
b171501f | 6043 | rq->core->core_forceidle_start = rq_clock(rq->core); |
4feee7d1 JD |
6044 | rq->core->core_forceidle_occupation = occ; |
6045 | } | |
6046 | ||
539f6512 PZ |
6047 | rq->core->core_pick_seq = rq->core->core_task_seq; |
6048 | next = rq->core_pick; | |
6049 | rq->core_sched_seq = rq->core->core_pick_seq; | |
6050 | ||
6051 | /* Something should have been selected for current CPU */ | |
6052 | WARN_ON_ONCE(!next); | |
6053 | ||
6054 | /* | |
6055 | * Reschedule siblings | |
6056 | * | |
6057 | * NOTE: L1TF -- at this point we're no longer running the old task and | |
6058 | * sending an IPI (below) ensures the sibling will no longer be running | |
6059 | * their task. This ensures there is no inter-sibling overlap between | |
6060 | * non-matching user state. | |
6061 | */ | |
6062 | for_each_cpu(i, smt_mask) { | |
bc9ffef3 | 6063 | rq_i = cpu_rq(i); |
539f6512 PZ |
6064 | |
6065 | /* | |
6066 | * An online sibling might have gone offline before a task | |
6067 | * could be picked for it, or it might be offline but later | |
6068 | * happen to come online, but its too late and nothing was | |
6069 | * picked for it. That's Ok - it will pick tasks for itself, | |
6070 | * so ignore it. | |
6071 | */ | |
6072 | if (!rq_i->core_pick) | |
6073 | continue; | |
6074 | ||
c6047c2e JFG |
6075 | /* |
6076 | * Update for new !FI->FI transitions, or if continuing to be in !FI: | |
6077 | * fi_before fi update? | |
6078 | * 0 0 1 | |
6079 | * 0 1 1 | |
6080 | * 1 0 1 | |
6081 | * 1 1 0 | |
6082 | */ | |
4feee7d1 JD |
6083 | if (!(fi_before && rq->core->core_forceidle_count)) |
6084 | task_vruntime_update(rq_i, rq_i->core_pick, !!rq->core->core_forceidle_count); | |
539f6512 | 6085 | |
d2dfa17b PZ |
6086 | rq_i->core_pick->core_occupation = occ; |
6087 | ||
539f6512 PZ |
6088 | if (i == cpu) { |
6089 | rq_i->core_pick = NULL; | |
6090 | continue; | |
6091 | } | |
6092 | ||
6093 | /* Did we break L1TF mitigation requirements? */ | |
6094 | WARN_ON_ONCE(!cookie_match(next, rq_i->core_pick)); | |
6095 | ||
6096 | if (rq_i->curr == rq_i->core_pick) { | |
6097 | rq_i->core_pick = NULL; | |
6098 | continue; | |
6099 | } | |
6100 | ||
6101 | resched_curr(rq_i); | |
6102 | } | |
6103 | ||
5b6547ed | 6104 | out_set_next: |
539f6512 | 6105 | set_next_task(rq, next); |
5b6547ed PZ |
6106 | out: |
6107 | if (rq->core->core_forceidle_count && next == rq->idle) | |
6108 | queue_core_balance(rq); | |
6109 | ||
539f6512 PZ |
6110 | return next; |
6111 | } | |
9edeaea1 | 6112 | |
d2dfa17b PZ |
6113 | static bool try_steal_cookie(int this, int that) |
6114 | { | |
6115 | struct rq *dst = cpu_rq(this), *src = cpu_rq(that); | |
6116 | struct task_struct *p; | |
6117 | unsigned long cookie; | |
6118 | bool success = false; | |
6119 | ||
6120 | local_irq_disable(); | |
6121 | double_rq_lock(dst, src); | |
6122 | ||
6123 | cookie = dst->core->core_cookie; | |
6124 | if (!cookie) | |
6125 | goto unlock; | |
6126 | ||
6127 | if (dst->curr != dst->idle) | |
6128 | goto unlock; | |
6129 | ||
6130 | p = sched_core_find(src, cookie); | |
6131 | if (p == src->idle) | |
6132 | goto unlock; | |
6133 | ||
6134 | do { | |
6135 | if (p == src->core_pick || p == src->curr) | |
6136 | goto next; | |
6137 | ||
386ef214 | 6138 | if (!is_cpu_allowed(p, this)) |
d2dfa17b PZ |
6139 | goto next; |
6140 | ||
6141 | if (p->core_occupation > dst->idle->core_occupation) | |
6142 | goto next; | |
6143 | ||
d2dfa17b PZ |
6144 | deactivate_task(src, p, 0); |
6145 | set_task_cpu(p, this); | |
6146 | activate_task(dst, p, 0); | |
d2dfa17b PZ |
6147 | |
6148 | resched_curr(dst); | |
6149 | ||
6150 | success = true; | |
6151 | break; | |
6152 | ||
6153 | next: | |
6154 | p = sched_core_next(p, cookie); | |
6155 | } while (p); | |
6156 | ||
6157 | unlock: | |
6158 | double_rq_unlock(dst, src); | |
6159 | local_irq_enable(); | |
6160 | ||
6161 | return success; | |
6162 | } | |
6163 | ||
6164 | static bool steal_cookie_task(int cpu, struct sched_domain *sd) | |
6165 | { | |
6166 | int i; | |
6167 | ||
6168 | for_each_cpu_wrap(i, sched_domain_span(sd), cpu) { | |
6169 | if (i == cpu) | |
6170 | continue; | |
6171 | ||
6172 | if (need_resched()) | |
6173 | break; | |
6174 | ||
6175 | if (try_steal_cookie(cpu, i)) | |
6176 | return true; | |
6177 | } | |
6178 | ||
6179 | return false; | |
6180 | } | |
6181 | ||
6182 | static void sched_core_balance(struct rq *rq) | |
6183 | { | |
6184 | struct sched_domain *sd; | |
6185 | int cpu = cpu_of(rq); | |
6186 | ||
6187 | preempt_disable(); | |
6188 | rcu_read_lock(); | |
6189 | raw_spin_rq_unlock_irq(rq); | |
6190 | for_each_domain(cpu, sd) { | |
6191 | if (need_resched()) | |
6192 | break; | |
6193 | ||
6194 | if (steal_cookie_task(cpu, sd)) | |
6195 | break; | |
6196 | } | |
6197 | raw_spin_rq_lock_irq(rq); | |
6198 | rcu_read_unlock(); | |
6199 | preempt_enable(); | |
6200 | } | |
6201 | ||
8e5bad7d | 6202 | static DEFINE_PER_CPU(struct balance_callback, core_balance_head); |
d2dfa17b | 6203 | |
5b6547ed | 6204 | static void queue_core_balance(struct rq *rq) |
d2dfa17b PZ |
6205 | { |
6206 | if (!sched_core_enabled(rq)) | |
6207 | return; | |
6208 | ||
6209 | if (!rq->core->core_cookie) | |
6210 | return; | |
6211 | ||
6212 | if (!rq->nr_running) /* not forced idle */ | |
6213 | return; | |
6214 | ||
6215 | queue_balance_callback(rq, &per_cpu(core_balance_head, rq->cpu), sched_core_balance); | |
6216 | } | |
6217 | ||
3c474b32 | 6218 | static void sched_core_cpu_starting(unsigned int cpu) |
9edeaea1 PZ |
6219 | { |
6220 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
3c474b32 PZ |
6221 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; |
6222 | unsigned long flags; | |
6223 | int t; | |
9edeaea1 | 6224 | |
3c474b32 | 6225 | sched_core_lock(cpu, &flags); |
9edeaea1 | 6226 | |
3c474b32 PZ |
6227 | WARN_ON_ONCE(rq->core != rq); |
6228 | ||
6229 | /* if we're the first, we'll be our own leader */ | |
6230 | if (cpumask_weight(smt_mask) == 1) | |
6231 | goto unlock; | |
6232 | ||
6233 | /* find the leader */ | |
6234 | for_each_cpu(t, smt_mask) { | |
6235 | if (t == cpu) | |
6236 | continue; | |
6237 | rq = cpu_rq(t); | |
6238 | if (rq->core == rq) { | |
6239 | core_rq = rq; | |
6240 | break; | |
9edeaea1 | 6241 | } |
3c474b32 | 6242 | } |
9edeaea1 | 6243 | |
3c474b32 PZ |
6244 | if (WARN_ON_ONCE(!core_rq)) /* whoopsie */ |
6245 | goto unlock; | |
9edeaea1 | 6246 | |
3c474b32 PZ |
6247 | /* install and validate core_rq */ |
6248 | for_each_cpu(t, smt_mask) { | |
6249 | rq = cpu_rq(t); | |
9edeaea1 | 6250 | |
3c474b32 | 6251 | if (t == cpu) |
9edeaea1 | 6252 | rq->core = core_rq; |
3c474b32 PZ |
6253 | |
6254 | WARN_ON_ONCE(rq->core != core_rq); | |
9edeaea1 | 6255 | } |
3c474b32 PZ |
6256 | |
6257 | unlock: | |
6258 | sched_core_unlock(cpu, &flags); | |
9edeaea1 | 6259 | } |
3c474b32 PZ |
6260 | |
6261 | static void sched_core_cpu_deactivate(unsigned int cpu) | |
6262 | { | |
6263 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
6264 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; | |
6265 | unsigned long flags; | |
6266 | int t; | |
6267 | ||
6268 | sched_core_lock(cpu, &flags); | |
6269 | ||
6270 | /* if we're the last man standing, nothing to do */ | |
6271 | if (cpumask_weight(smt_mask) == 1) { | |
6272 | WARN_ON_ONCE(rq->core != rq); | |
6273 | goto unlock; | |
6274 | } | |
6275 | ||
6276 | /* if we're not the leader, nothing to do */ | |
6277 | if (rq->core != rq) | |
6278 | goto unlock; | |
6279 | ||
6280 | /* find a new leader */ | |
6281 | for_each_cpu(t, smt_mask) { | |
6282 | if (t == cpu) | |
6283 | continue; | |
6284 | core_rq = cpu_rq(t); | |
6285 | break; | |
6286 | } | |
6287 | ||
6288 | if (WARN_ON_ONCE(!core_rq)) /* impossible */ | |
6289 | goto unlock; | |
6290 | ||
6291 | /* copy the shared state to the new leader */ | |
4feee7d1 JD |
6292 | core_rq->core_task_seq = rq->core_task_seq; |
6293 | core_rq->core_pick_seq = rq->core_pick_seq; | |
6294 | core_rq->core_cookie = rq->core_cookie; | |
6295 | core_rq->core_forceidle_count = rq->core_forceidle_count; | |
6296 | core_rq->core_forceidle_seq = rq->core_forceidle_seq; | |
6297 | core_rq->core_forceidle_occupation = rq->core_forceidle_occupation; | |
6298 | ||
6299 | /* | |
6300 | * Accounting edge for forced idle is handled in pick_next_task(). | |
6301 | * Don't need another one here, since the hotplug thread shouldn't | |
6302 | * have a cookie. | |
6303 | */ | |
6304 | core_rq->core_forceidle_start = 0; | |
3c474b32 PZ |
6305 | |
6306 | /* install new leader */ | |
6307 | for_each_cpu(t, smt_mask) { | |
6308 | rq = cpu_rq(t); | |
6309 | rq->core = core_rq; | |
6310 | } | |
6311 | ||
6312 | unlock: | |
6313 | sched_core_unlock(cpu, &flags); | |
6314 | } | |
6315 | ||
6316 | static inline void sched_core_cpu_dying(unsigned int cpu) | |
6317 | { | |
6318 | struct rq *rq = cpu_rq(cpu); | |
6319 | ||
6320 | if (rq->core != rq) | |
6321 | rq->core = rq; | |
6322 | } | |
6323 | ||
9edeaea1 PZ |
6324 | #else /* !CONFIG_SCHED_CORE */ |
6325 | ||
6326 | static inline void sched_core_cpu_starting(unsigned int cpu) {} | |
3c474b32 PZ |
6327 | static inline void sched_core_cpu_deactivate(unsigned int cpu) {} |
6328 | static inline void sched_core_cpu_dying(unsigned int cpu) {} | |
9edeaea1 | 6329 | |
539f6512 PZ |
6330 | static struct task_struct * |
6331 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
6332 | { | |
6333 | return __pick_next_task(rq, prev, rf); | |
6334 | } | |
6335 | ||
9edeaea1 PZ |
6336 | #endif /* CONFIG_SCHED_CORE */ |
6337 | ||
b4bfa3fc TG |
6338 | /* |
6339 | * Constants for the sched_mode argument of __schedule(). | |
6340 | * | |
6341 | * The mode argument allows RT enabled kernels to differentiate a | |
6342 | * preemption from blocking on an 'sleeping' spin/rwlock. Note that | |
6343 | * SM_MASK_PREEMPT for !RT has all bits set, which allows the compiler to | |
6344 | * optimize the AND operation out and just check for zero. | |
6345 | */ | |
6346 | #define SM_NONE 0x0 | |
6347 | #define SM_PREEMPT 0x1 | |
6991436c TG |
6348 | #define SM_RTLOCK_WAIT 0x2 |
6349 | ||
6350 | #ifndef CONFIG_PREEMPT_RT | |
6351 | # define SM_MASK_PREEMPT (~0U) | |
6352 | #else | |
6353 | # define SM_MASK_PREEMPT SM_PREEMPT | |
6354 | #endif | |
b4bfa3fc | 6355 | |
dd41f596 | 6356 | /* |
c259e01a | 6357 | * __schedule() is the main scheduler function. |
edde96ea PE |
6358 | * |
6359 | * The main means of driving the scheduler and thus entering this function are: | |
6360 | * | |
6361 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
6362 | * | |
6363 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
6364 | * paths. For example, see arch/x86/entry_64.S. | |
6365 | * | |
6366 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
6367 | * interrupt handler scheduler_tick(). | |
6368 | * | |
6369 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
6370 | * task to the run-queue and that's it. | |
6371 | * | |
6372 | * Now, if the new task added to the run-queue preempts the current | |
6373 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
6374 | * called on the nearest possible occasion: | |
6375 | * | |
c1a280b6 | 6376 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
6377 | * |
6378 | * - in syscall or exception context, at the next outmost | |
6379 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
6380 | * spin_unlock()!) | |
6381 | * | |
6382 | * - in IRQ context, return from interrupt-handler to | |
6383 | * preemptible context | |
6384 | * | |
c1a280b6 | 6385 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
6386 | * then at the next: |
6387 | * | |
6388 | * - cond_resched() call | |
6389 | * - explicit schedule() call | |
6390 | * - return from syscall or exception to user-space | |
6391 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 6392 | * |
b30f0e3f | 6393 | * WARNING: must be called with preemption disabled! |
dd41f596 | 6394 | */ |
b4bfa3fc | 6395 | static void __sched notrace __schedule(unsigned int sched_mode) |
dd41f596 IM |
6396 | { |
6397 | struct task_struct *prev, *next; | |
67ca7bde | 6398 | unsigned long *switch_count; |
dbfb089d | 6399 | unsigned long prev_state; |
d8ac8971 | 6400 | struct rq_flags rf; |
dd41f596 | 6401 | struct rq *rq; |
31656519 | 6402 | int cpu; |
dd41f596 | 6403 | |
dd41f596 IM |
6404 | cpu = smp_processor_id(); |
6405 | rq = cpu_rq(cpu); | |
dd41f596 | 6406 | prev = rq->curr; |
dd41f596 | 6407 | |
b4bfa3fc | 6408 | schedule_debug(prev, !!sched_mode); |
1da177e4 | 6409 | |
e0ee463c | 6410 | if (sched_feat(HRTICK) || sched_feat(HRTICK_DL)) |
f333fdc9 | 6411 | hrtick_clear(rq); |
8f4d37ec | 6412 | |
46a5d164 | 6413 | local_irq_disable(); |
b4bfa3fc | 6414 | rcu_note_context_switch(!!sched_mode); |
46a5d164 | 6415 | |
e0acd0a6 ON |
6416 | /* |
6417 | * Make sure that signal_pending_state()->signal_pending() below | |
6418 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
6419 | * done by the caller to avoid the race with signal_wake_up(): |
6420 | * | |
6421 | * __set_current_state(@state) signal_wake_up() | |
6422 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
6423 | * wake_up_state(p, state) | |
6424 | * LOCK rq->lock LOCK p->pi_state | |
6425 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
6426 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 6427 | * |
dbfb089d | 6428 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 6429 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 6430 | */ |
8a8c69c3 | 6431 | rq_lock(rq, &rf); |
d89e588c | 6432 | smp_mb__after_spinlock(); |
1da177e4 | 6433 | |
d1ccc66d IM |
6434 | /* Promote REQ to ACT */ |
6435 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 6436 | update_rq_clock(rq); |
9edfbfed | 6437 | |
246d86b5 | 6438 | switch_count = &prev->nivcsw; |
d136122f | 6439 | |
dbfb089d | 6440 | /* |
d136122f | 6441 | * We must load prev->state once (task_struct::state is volatile), such |
2500ad1c | 6442 | * that we form a control dependency vs deactivate_task() below. |
dbfb089d | 6443 | */ |
2f064a59 | 6444 | prev_state = READ_ONCE(prev->__state); |
b4bfa3fc | 6445 | if (!(sched_mode & SM_MASK_PREEMPT) && prev_state) { |
dbfb089d | 6446 | if (signal_pending_state(prev_state, prev)) { |
2f064a59 | 6447 | WRITE_ONCE(prev->__state, TASK_RUNNING); |
21aa9af0 | 6448 | } else { |
dbfb089d PZ |
6449 | prev->sched_contributes_to_load = |
6450 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
6451 | !(prev_state & TASK_NOLOAD) && | |
f5d39b02 | 6452 | !(prev_state & TASK_FROZEN); |
dbfb089d PZ |
6453 | |
6454 | if (prev->sched_contributes_to_load) | |
6455 | rq->nr_uninterruptible++; | |
6456 | ||
6457 | /* | |
6458 | * __schedule() ttwu() | |
d136122f PZ |
6459 | * prev_state = prev->state; if (p->on_rq && ...) |
6460 | * if (prev_state) goto out; | |
6461 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
6462 | * p->state = TASK_WAKING | |
6463 | * | |
6464 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
6465 | * |
6466 | * After this, schedule() must not care about p->state any more. | |
6467 | */ | |
bce4dc80 | 6468 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 6469 | |
e33a9bba TH |
6470 | if (prev->in_iowait) { |
6471 | atomic_inc(&rq->nr_iowait); | |
6472 | delayacct_blkio_start(); | |
6473 | } | |
21aa9af0 | 6474 | } |
dd41f596 | 6475 | switch_count = &prev->nvcsw; |
1da177e4 LT |
6476 | } |
6477 | ||
d8ac8971 | 6478 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 6479 | clear_tsk_need_resched(prev); |
f27dde8d | 6480 | clear_preempt_need_resched(); |
c006fac5 PT |
6481 | #ifdef CONFIG_SCHED_DEBUG |
6482 | rq->last_seen_need_resched_ns = 0; | |
6483 | #endif | |
1da177e4 | 6484 | |
1da177e4 | 6485 | if (likely(prev != next)) { |
1da177e4 | 6486 | rq->nr_switches++; |
5311a98f EB |
6487 | /* |
6488 | * RCU users of rcu_dereference(rq->curr) may not see | |
6489 | * changes to task_struct made by pick_next_task(). | |
6490 | */ | |
6491 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
6492 | /* |
6493 | * The membarrier system call requires each architecture | |
6494 | * to have a full memory barrier after updating | |
306e0604 MD |
6495 | * rq->curr, before returning to user-space. |
6496 | * | |
6497 | * Here are the schemes providing that barrier on the | |
6498 | * various architectures: | |
6499 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
6500 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
6501 | * - finish_lock_switch() for weakly-ordered | |
6502 | * architectures where spin_unlock is a full barrier, | |
6503 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
6504 | * is a RELEASE barrier), | |
22e4ebb9 | 6505 | */ |
1da177e4 LT |
6506 | ++*switch_count; |
6507 | ||
af449901 | 6508 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
6509 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
6510 | ||
9c2136be | 6511 | trace_sched_switch(sched_mode & SM_MASK_PREEMPT, prev, next, prev_state); |
d1ccc66d IM |
6512 | |
6513 | /* Also unlocks the rq: */ | |
6514 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 6515 | } else { |
cb42c9a3 | 6516 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 6517 | |
565790d2 PZ |
6518 | rq_unpin_lock(rq, &rf); |
6519 | __balance_callbacks(rq); | |
5cb9eaa3 | 6520 | raw_spin_rq_unlock_irq(rq); |
565790d2 | 6521 | } |
1da177e4 | 6522 | } |
c259e01a | 6523 | |
9af6528e PZ |
6524 | void __noreturn do_task_dead(void) |
6525 | { | |
d1ccc66d | 6526 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 6527 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
6528 | |
6529 | /* Tell freezer to ignore us: */ | |
6530 | current->flags |= PF_NOFREEZE; | |
6531 | ||
b4bfa3fc | 6532 | __schedule(SM_NONE); |
9af6528e | 6533 | BUG(); |
d1ccc66d IM |
6534 | |
6535 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 6536 | for (;;) |
d1ccc66d | 6537 | cpu_relax(); |
9af6528e PZ |
6538 | } |
6539 | ||
9c40cef2 TG |
6540 | static inline void sched_submit_work(struct task_struct *tsk) |
6541 | { | |
c1cecf88 SAS |
6542 | unsigned int task_flags; |
6543 | ||
b03fbd4f | 6544 | if (task_is_running(tsk)) |
9c40cef2 | 6545 | return; |
6d25be57 | 6546 | |
c1cecf88 | 6547 | task_flags = tsk->flags; |
6d25be57 | 6548 | /* |
b945efcd TG |
6549 | * If a worker goes to sleep, notify and ask workqueue whether it |
6550 | * wants to wake up a task to maintain concurrency. | |
6d25be57 | 6551 | */ |
c1cecf88 | 6552 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
c1cecf88 | 6553 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
6554 | wq_worker_sleeping(tsk); |
6555 | else | |
6556 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
6557 | } |
6558 | ||
401e4963 JK |
6559 | /* |
6560 | * spinlock and rwlock must not flush block requests. This will | |
6561 | * deadlock if the callback attempts to acquire a lock which is | |
6562 | * already acquired. | |
6563 | */ | |
6564 | SCHED_WARN_ON(current->__state & TASK_RTLOCK_WAIT); | |
b0fdc013 | 6565 | |
9c40cef2 TG |
6566 | /* |
6567 | * If we are going to sleep and we have plugged IO queued, | |
6568 | * make sure to submit it to avoid deadlocks. | |
6569 | */ | |
aa8dccca | 6570 | blk_flush_plug(tsk->plug, true); |
9c40cef2 TG |
6571 | } |
6572 | ||
6d25be57 TG |
6573 | static void sched_update_worker(struct task_struct *tsk) |
6574 | { | |
771b53d0 JA |
6575 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6576 | if (tsk->flags & PF_WQ_WORKER) | |
6577 | wq_worker_running(tsk); | |
6578 | else | |
6579 | io_wq_worker_running(tsk); | |
6580 | } | |
6d25be57 TG |
6581 | } |
6582 | ||
722a9f92 | 6583 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 6584 | { |
9c40cef2 TG |
6585 | struct task_struct *tsk = current; |
6586 | ||
6587 | sched_submit_work(tsk); | |
bfd9b2b5 | 6588 | do { |
b30f0e3f | 6589 | preempt_disable(); |
b4bfa3fc | 6590 | __schedule(SM_NONE); |
b30f0e3f | 6591 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 6592 | } while (need_resched()); |
6d25be57 | 6593 | sched_update_worker(tsk); |
c259e01a | 6594 | } |
1da177e4 LT |
6595 | EXPORT_SYMBOL(schedule); |
6596 | ||
8663effb SRV |
6597 | /* |
6598 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
6599 | * state (have scheduled out non-voluntarily) by making sure that all | |
6600 | * tasks have either left the run queue or have gone into user space. | |
6601 | * As idle tasks do not do either, they must not ever be preempted | |
6602 | * (schedule out non-voluntarily). | |
6603 | * | |
6604 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
6605 | * never enables preemption because it does not call sched_submit_work(). | |
6606 | */ | |
6607 | void __sched schedule_idle(void) | |
6608 | { | |
6609 | /* | |
6610 | * As this skips calling sched_submit_work(), which the idle task does | |
6611 | * regardless because that function is a nop when the task is in a | |
6612 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
6613 | * current task can be in any other state. Note, idle is always in the | |
6614 | * TASK_RUNNING state. | |
6615 | */ | |
2f064a59 | 6616 | WARN_ON_ONCE(current->__state); |
8663effb | 6617 | do { |
b4bfa3fc | 6618 | __schedule(SM_NONE); |
8663effb SRV |
6619 | } while (need_resched()); |
6620 | } | |
6621 | ||
24a9c541 | 6622 | #if defined(CONFIG_CONTEXT_TRACKING_USER) && !defined(CONFIG_HAVE_CONTEXT_TRACKING_USER_OFFSTACK) |
722a9f92 | 6623 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
6624 | { |
6625 | /* | |
6626 | * If we come here after a random call to set_need_resched(), | |
6627 | * or we have been woken up remotely but the IPI has not yet arrived, | |
6628 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
6629 | * we find a better solution. | |
7cc78f8f AL |
6630 | * |
6631 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 6632 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 6633 | * too frequently to make sense yet. |
20ab65e3 | 6634 | */ |
7cc78f8f | 6635 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 6636 | schedule(); |
7cc78f8f | 6637 | exception_exit(prev_state); |
20ab65e3 FW |
6638 | } |
6639 | #endif | |
6640 | ||
c5491ea7 TG |
6641 | /** |
6642 | * schedule_preempt_disabled - called with preemption disabled | |
6643 | * | |
6644 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
6645 | */ | |
6646 | void __sched schedule_preempt_disabled(void) | |
6647 | { | |
ba74c144 | 6648 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
6649 | schedule(); |
6650 | preempt_disable(); | |
6651 | } | |
6652 | ||
6991436c TG |
6653 | #ifdef CONFIG_PREEMPT_RT |
6654 | void __sched notrace schedule_rtlock(void) | |
6655 | { | |
6656 | do { | |
6657 | preempt_disable(); | |
6658 | __schedule(SM_RTLOCK_WAIT); | |
6659 | sched_preempt_enable_no_resched(); | |
6660 | } while (need_resched()); | |
6661 | } | |
6662 | NOKPROBE_SYMBOL(schedule_rtlock); | |
6663 | #endif | |
6664 | ||
06b1f808 | 6665 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
6666 | { |
6667 | do { | |
47252cfb SR |
6668 | /* |
6669 | * Because the function tracer can trace preempt_count_sub() | |
6670 | * and it also uses preempt_enable/disable_notrace(), if | |
6671 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6672 | * by the function tracer will call this function again and | |
6673 | * cause infinite recursion. | |
6674 | * | |
6675 | * Preemption must be disabled here before the function | |
6676 | * tracer can trace. Break up preempt_disable() into two | |
6677 | * calls. One to disable preemption without fear of being | |
6678 | * traced. The other to still record the preemption latency, | |
6679 | * which can also be traced by the function tracer. | |
6680 | */ | |
499d7955 | 6681 | preempt_disable_notrace(); |
47252cfb | 6682 | preempt_latency_start(1); |
b4bfa3fc | 6683 | __schedule(SM_PREEMPT); |
47252cfb | 6684 | preempt_latency_stop(1); |
499d7955 | 6685 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
6686 | |
6687 | /* | |
6688 | * Check again in case we missed a preemption opportunity | |
6689 | * between schedule and now. | |
6690 | */ | |
a18b5d01 FW |
6691 | } while (need_resched()); |
6692 | } | |
6693 | ||
c1a280b6 | 6694 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 6695 | /* |
a49b4f40 VS |
6696 | * This is the entry point to schedule() from in-kernel preemption |
6697 | * off of preempt_enable. | |
1da177e4 | 6698 | */ |
722a9f92 | 6699 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 6700 | { |
1da177e4 LT |
6701 | /* |
6702 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 6703 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 6704 | */ |
fbb00b56 | 6705 | if (likely(!preemptible())) |
1da177e4 | 6706 | return; |
a18b5d01 | 6707 | preempt_schedule_common(); |
1da177e4 | 6708 | } |
376e2424 | 6709 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 6710 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 6711 | |
2c9a98d3 | 6712 | #ifdef CONFIG_PREEMPT_DYNAMIC |
99cf983c | 6713 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
6714 | #ifndef preempt_schedule_dynamic_enabled |
6715 | #define preempt_schedule_dynamic_enabled preempt_schedule | |
6716 | #define preempt_schedule_dynamic_disabled NULL | |
6717 | #endif | |
6718 | DEFINE_STATIC_CALL(preempt_schedule, preempt_schedule_dynamic_enabled); | |
ef72661e | 6719 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule); |
99cf983c MR |
6720 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
6721 | static DEFINE_STATIC_KEY_TRUE(sk_dynamic_preempt_schedule); | |
6722 | void __sched notrace dynamic_preempt_schedule(void) | |
6723 | { | |
6724 | if (!static_branch_unlikely(&sk_dynamic_preempt_schedule)) | |
6725 | return; | |
6726 | preempt_schedule(); | |
6727 | } | |
6728 | NOKPROBE_SYMBOL(dynamic_preempt_schedule); | |
6729 | EXPORT_SYMBOL(dynamic_preempt_schedule); | |
6730 | #endif | |
2c9a98d3 | 6731 | #endif |
2c9a98d3 | 6732 | |
009f60e2 | 6733 | /** |
4eaca0a8 | 6734 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
6735 | * |
6736 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
6737 | * recursion and tracing preempt enabling caused by the tracing | |
6738 | * infrastructure itself. But as tracing can happen in areas coming | |
6739 | * from userspace or just about to enter userspace, a preempt enable | |
6740 | * can occur before user_exit() is called. This will cause the scheduler | |
6741 | * to be called when the system is still in usermode. | |
6742 | * | |
6743 | * To prevent this, the preempt_enable_notrace will use this function | |
6744 | * instead of preempt_schedule() to exit user context if needed before | |
6745 | * calling the scheduler. | |
6746 | */ | |
4eaca0a8 | 6747 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
6748 | { |
6749 | enum ctx_state prev_ctx; | |
6750 | ||
6751 | if (likely(!preemptible())) | |
6752 | return; | |
6753 | ||
6754 | do { | |
47252cfb SR |
6755 | /* |
6756 | * Because the function tracer can trace preempt_count_sub() | |
6757 | * and it also uses preempt_enable/disable_notrace(), if | |
6758 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6759 | * by the function tracer will call this function again and | |
6760 | * cause infinite recursion. | |
6761 | * | |
6762 | * Preemption must be disabled here before the function | |
6763 | * tracer can trace. Break up preempt_disable() into two | |
6764 | * calls. One to disable preemption without fear of being | |
6765 | * traced. The other to still record the preemption latency, | |
6766 | * which can also be traced by the function tracer. | |
6767 | */ | |
3d8f74dd | 6768 | preempt_disable_notrace(); |
47252cfb | 6769 | preempt_latency_start(1); |
009f60e2 ON |
6770 | /* |
6771 | * Needs preempt disabled in case user_exit() is traced | |
6772 | * and the tracer calls preempt_enable_notrace() causing | |
6773 | * an infinite recursion. | |
6774 | */ | |
6775 | prev_ctx = exception_enter(); | |
b4bfa3fc | 6776 | __schedule(SM_PREEMPT); |
009f60e2 ON |
6777 | exception_exit(prev_ctx); |
6778 | ||
47252cfb | 6779 | preempt_latency_stop(1); |
3d8f74dd | 6780 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
6781 | } while (need_resched()); |
6782 | } | |
4eaca0a8 | 6783 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 6784 | |
2c9a98d3 | 6785 | #ifdef CONFIG_PREEMPT_DYNAMIC |
99cf983c | 6786 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
6787 | #ifndef preempt_schedule_notrace_dynamic_enabled |
6788 | #define preempt_schedule_notrace_dynamic_enabled preempt_schedule_notrace | |
6789 | #define preempt_schedule_notrace_dynamic_disabled NULL | |
2c9a98d3 | 6790 | #endif |
8a69fe0b | 6791 | DEFINE_STATIC_CALL(preempt_schedule_notrace, preempt_schedule_notrace_dynamic_enabled); |
ef72661e | 6792 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule_notrace); |
99cf983c MR |
6793 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
6794 | static DEFINE_STATIC_KEY_TRUE(sk_dynamic_preempt_schedule_notrace); | |
6795 | void __sched notrace dynamic_preempt_schedule_notrace(void) | |
c597bfdd | 6796 | { |
99cf983c MR |
6797 | if (!static_branch_unlikely(&sk_dynamic_preempt_schedule_notrace)) |
6798 | return; | |
6799 | preempt_schedule_notrace(); | |
c597bfdd | 6800 | } |
99cf983c MR |
6801 | NOKPROBE_SYMBOL(dynamic_preempt_schedule_notrace); |
6802 | EXPORT_SYMBOL(dynamic_preempt_schedule_notrace); | |
6803 | #endif | |
2c9a98d3 | 6804 | #endif |
c597bfdd | 6805 | |
c1a280b6 | 6806 | #endif /* CONFIG_PREEMPTION */ |
826bfeb3 | 6807 | |
1da177e4 | 6808 | /* |
a49b4f40 | 6809 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
6810 | * off of irq context. |
6811 | * Note, that this is called and return with irqs disabled. This will | |
6812 | * protect us against recursive calling from irq. | |
6813 | */ | |
722a9f92 | 6814 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 6815 | { |
b22366cd | 6816 | enum ctx_state prev_state; |
6478d880 | 6817 | |
2ed6e34f | 6818 | /* Catch callers which need to be fixed */ |
f27dde8d | 6819 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 6820 | |
b22366cd FW |
6821 | prev_state = exception_enter(); |
6822 | ||
3a5c359a | 6823 | do { |
3d8f74dd | 6824 | preempt_disable(); |
3a5c359a | 6825 | local_irq_enable(); |
b4bfa3fc | 6826 | __schedule(SM_PREEMPT); |
3a5c359a | 6827 | local_irq_disable(); |
3d8f74dd | 6828 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 6829 | } while (need_resched()); |
b22366cd FW |
6830 | |
6831 | exception_exit(prev_state); | |
1da177e4 LT |
6832 | } |
6833 | ||
ac6424b9 | 6834 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 6835 | void *key) |
1da177e4 | 6836 | { |
062d3f95 | 6837 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 6838 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 6839 | } |
1da177e4 LT |
6840 | EXPORT_SYMBOL(default_wake_function); |
6841 | ||
f558c2b8 PZ |
6842 | static void __setscheduler_prio(struct task_struct *p, int prio) |
6843 | { | |
6844 | if (dl_prio(prio)) | |
6845 | p->sched_class = &dl_sched_class; | |
6846 | else if (rt_prio(prio)) | |
6847 | p->sched_class = &rt_sched_class; | |
6848 | else | |
6849 | p->sched_class = &fair_sched_class; | |
6850 | ||
6851 | p->prio = prio; | |
6852 | } | |
6853 | ||
b29739f9 IM |
6854 | #ifdef CONFIG_RT_MUTEXES |
6855 | ||
acd58620 PZ |
6856 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
6857 | { | |
6858 | if (pi_task) | |
6859 | prio = min(prio, pi_task->prio); | |
6860 | ||
6861 | return prio; | |
6862 | } | |
6863 | ||
6864 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
6865 | { | |
6866 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
6867 | ||
6868 | return __rt_effective_prio(pi_task, prio); | |
6869 | } | |
6870 | ||
b29739f9 IM |
6871 | /* |
6872 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
6873 | * @p: task to boost |
6874 | * @pi_task: donor task | |
b29739f9 IM |
6875 | * |
6876 | * This function changes the 'effective' priority of a task. It does | |
6877 | * not touch ->normal_prio like __setscheduler(). | |
6878 | * | |
c365c292 TG |
6879 | * Used by the rt_mutex code to implement priority inheritance |
6880 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 6881 | */ |
acd58620 | 6882 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 6883 | { |
acd58620 | 6884 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 6885 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 6886 | const struct sched_class *prev_class; |
eb580751 PZ |
6887 | struct rq_flags rf; |
6888 | struct rq *rq; | |
b29739f9 | 6889 | |
acd58620 PZ |
6890 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
6891 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
6892 | ||
6893 | /* | |
6894 | * If nothing changed; bail early. | |
6895 | */ | |
6896 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
6897 | return; | |
b29739f9 | 6898 | |
eb580751 | 6899 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 6900 | update_rq_clock(rq); |
acd58620 PZ |
6901 | /* |
6902 | * Set under pi_lock && rq->lock, such that the value can be used under | |
6903 | * either lock. | |
6904 | * | |
6905 | * Note that there is loads of tricky to make this pointer cache work | |
6906 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
6907 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
6908 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 6909 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
6910 | */ |
6911 | p->pi_top_task = pi_task; | |
6912 | ||
6913 | /* | |
6914 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
6915 | */ | |
6916 | if (prio == p->prio && !dl_prio(prio)) | |
6917 | goto out_unlock; | |
b29739f9 | 6918 | |
1c4dd99b TG |
6919 | /* |
6920 | * Idle task boosting is a nono in general. There is one | |
6921 | * exception, when PREEMPT_RT and NOHZ is active: | |
6922 | * | |
6923 | * The idle task calls get_next_timer_interrupt() and holds | |
6924 | * the timer wheel base->lock on the CPU and another CPU wants | |
6925 | * to access the timer (probably to cancel it). We can safely | |
6926 | * ignore the boosting request, as the idle CPU runs this code | |
6927 | * with interrupts disabled and will complete the lock | |
6928 | * protected section without being interrupted. So there is no | |
6929 | * real need to boost. | |
6930 | */ | |
6931 | if (unlikely(p == rq->idle)) { | |
6932 | WARN_ON(p != rq->curr); | |
6933 | WARN_ON(p->pi_blocked_on); | |
6934 | goto out_unlock; | |
6935 | } | |
6936 | ||
b91473ff | 6937 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 6938 | oldprio = p->prio; |
ff77e468 PZ |
6939 | |
6940 | if (oldprio == prio) | |
6941 | queue_flag &= ~DEQUEUE_MOVE; | |
6942 | ||
83ab0aa0 | 6943 | prev_class = p->sched_class; |
da0c1e65 | 6944 | queued = task_on_rq_queued(p); |
051a1d1a | 6945 | running = task_current(rq, p); |
da0c1e65 | 6946 | if (queued) |
ff77e468 | 6947 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 6948 | if (running) |
f3cd1c4e | 6949 | put_prev_task(rq, p); |
dd41f596 | 6950 | |
2d3d891d DF |
6951 | /* |
6952 | * Boosting condition are: | |
6953 | * 1. -rt task is running and holds mutex A | |
6954 | * --> -dl task blocks on mutex A | |
6955 | * | |
6956 | * 2. -dl task is running and holds mutex A | |
6957 | * --> -dl task blocks on mutex A and could preempt the | |
6958 | * running task | |
6959 | */ | |
6960 | if (dl_prio(prio)) { | |
466af29b | 6961 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
6962 | (pi_task && dl_prio(pi_task->prio) && |
6963 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2279f540 | 6964 | p->dl.pi_se = pi_task->dl.pi_se; |
ff77e468 | 6965 | queue_flag |= ENQUEUE_REPLENISH; |
2279f540 JL |
6966 | } else { |
6967 | p->dl.pi_se = &p->dl; | |
6968 | } | |
2d3d891d DF |
6969 | } else if (rt_prio(prio)) { |
6970 | if (dl_prio(oldprio)) | |
2279f540 | 6971 | p->dl.pi_se = &p->dl; |
2d3d891d | 6972 | if (oldprio < prio) |
ff77e468 | 6973 | queue_flag |= ENQUEUE_HEAD; |
2d3d891d DF |
6974 | } else { |
6975 | if (dl_prio(oldprio)) | |
2279f540 | 6976 | p->dl.pi_se = &p->dl; |
746db944 BS |
6977 | if (rt_prio(oldprio)) |
6978 | p->rt.timeout = 0; | |
2d3d891d | 6979 | } |
dd41f596 | 6980 | |
f558c2b8 | 6981 | __setscheduler_prio(p, prio); |
b29739f9 | 6982 | |
da0c1e65 | 6983 | if (queued) |
ff77e468 | 6984 | enqueue_task(rq, p, queue_flag); |
a399d233 | 6985 | if (running) |
03b7fad1 | 6986 | set_next_task(rq, p); |
cb469845 | 6987 | |
da7a735e | 6988 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 6989 | out_unlock: |
d1ccc66d IM |
6990 | /* Avoid rq from going away on us: */ |
6991 | preempt_disable(); | |
4c9a4bc8 | 6992 | |
565790d2 PZ |
6993 | rq_unpin_lock(rq, &rf); |
6994 | __balance_callbacks(rq); | |
5cb9eaa3 | 6995 | raw_spin_rq_unlock(rq); |
565790d2 | 6996 | |
4c9a4bc8 | 6997 | preempt_enable(); |
b29739f9 | 6998 | } |
acd58620 PZ |
6999 | #else |
7000 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
7001 | { | |
7002 | return prio; | |
7003 | } | |
b29739f9 | 7004 | #endif |
d50dde5a | 7005 | |
36c8b586 | 7006 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 7007 | { |
49bd21ef | 7008 | bool queued, running; |
53a23364 | 7009 | int old_prio; |
eb580751 | 7010 | struct rq_flags rf; |
70b97a7f | 7011 | struct rq *rq; |
1da177e4 | 7012 | |
75e45d51 | 7013 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
7014 | return; |
7015 | /* | |
7016 | * We have to be careful, if called from sys_setpriority(), | |
7017 | * the task might be in the middle of scheduling on another CPU. | |
7018 | */ | |
eb580751 | 7019 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
7020 | update_rq_clock(rq); |
7021 | ||
1da177e4 LT |
7022 | /* |
7023 | * The RT priorities are set via sched_setscheduler(), but we still | |
7024 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 7025 | * it won't have any effect on scheduling until the task is |
aab03e05 | 7026 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 7027 | */ |
aab03e05 | 7028 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
7029 | p->static_prio = NICE_TO_PRIO(nice); |
7030 | goto out_unlock; | |
7031 | } | |
da0c1e65 | 7032 | queued = task_on_rq_queued(p); |
49bd21ef | 7033 | running = task_current(rq, p); |
da0c1e65 | 7034 | if (queued) |
7a57f32a | 7035 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
7036 | if (running) |
7037 | put_prev_task(rq, p); | |
1da177e4 | 7038 | |
1da177e4 | 7039 | p->static_prio = NICE_TO_PRIO(nice); |
b1e82065 | 7040 | set_load_weight(p, true); |
b29739f9 IM |
7041 | old_prio = p->prio; |
7042 | p->prio = effective_prio(p); | |
1da177e4 | 7043 | |
5443a0be | 7044 | if (queued) |
7134b3e9 | 7045 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 7046 | if (running) |
03b7fad1 | 7047 | set_next_task(rq, p); |
5443a0be FW |
7048 | |
7049 | /* | |
7050 | * If the task increased its priority or is running and | |
7051 | * lowered its priority, then reschedule its CPU: | |
7052 | */ | |
7053 | p->sched_class->prio_changed(rq, p, old_prio); | |
7054 | ||
1da177e4 | 7055 | out_unlock: |
eb580751 | 7056 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 7057 | } |
1da177e4 LT |
7058 | EXPORT_SYMBOL(set_user_nice); |
7059 | ||
e43379f1 | 7060 | /* |
700a7833 CG |
7061 | * is_nice_reduction - check if nice value is an actual reduction |
7062 | * | |
7063 | * Similar to can_nice() but does not perform a capability check. | |
7064 | * | |
e43379f1 MM |
7065 | * @p: task |
7066 | * @nice: nice value | |
7067 | */ | |
700a7833 | 7068 | static bool is_nice_reduction(const struct task_struct *p, const int nice) |
e43379f1 | 7069 | { |
d1ccc66d | 7070 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 7071 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 7072 | |
700a7833 CG |
7073 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE)); |
7074 | } | |
7075 | ||
7076 | /* | |
7077 | * can_nice - check if a task can reduce its nice value | |
7078 | * @p: task | |
7079 | * @nice: nice value | |
7080 | */ | |
7081 | int can_nice(const struct task_struct *p, const int nice) | |
7082 | { | |
7083 | return is_nice_reduction(p, nice) || capable(CAP_SYS_NICE); | |
e43379f1 MM |
7084 | } |
7085 | ||
1da177e4 LT |
7086 | #ifdef __ARCH_WANT_SYS_NICE |
7087 | ||
7088 | /* | |
7089 | * sys_nice - change the priority of the current process. | |
7090 | * @increment: priority increment | |
7091 | * | |
7092 | * sys_setpriority is a more generic, but much slower function that | |
7093 | * does similar things. | |
7094 | */ | |
5add95d4 | 7095 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 7096 | { |
48f24c4d | 7097 | long nice, retval; |
1da177e4 LT |
7098 | |
7099 | /* | |
7100 | * Setpriority might change our priority at the same moment. | |
7101 | * We don't have to worry. Conceptually one call occurs first | |
7102 | * and we have a single winner. | |
7103 | */ | |
a9467fa3 | 7104 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 7105 | nice = task_nice(current) + increment; |
1da177e4 | 7106 | |
a9467fa3 | 7107 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
7108 | if (increment < 0 && !can_nice(current, nice)) |
7109 | return -EPERM; | |
7110 | ||
1da177e4 LT |
7111 | retval = security_task_setnice(current, nice); |
7112 | if (retval) | |
7113 | return retval; | |
7114 | ||
7115 | set_user_nice(current, nice); | |
7116 | return 0; | |
7117 | } | |
7118 | ||
7119 | #endif | |
7120 | ||
7121 | /** | |
7122 | * task_prio - return the priority value of a given task. | |
7123 | * @p: the task in question. | |
7124 | * | |
e69f6186 | 7125 | * Return: The priority value as seen by users in /proc. |
c541bb78 DE |
7126 | * |
7127 | * sched policy return value kernel prio user prio/nice | |
7128 | * | |
7129 | * normal, batch, idle [0 ... 39] [100 ... 139] 0/[-20 ... 19] | |
7130 | * fifo, rr [-2 ... -100] [98 ... 0] [1 ... 99] | |
7131 | * deadline -101 -1 0 | |
1da177e4 | 7132 | */ |
36c8b586 | 7133 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
7134 | { |
7135 | return p->prio - MAX_RT_PRIO; | |
7136 | } | |
7137 | ||
1da177e4 | 7138 | /** |
d1ccc66d | 7139 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 7140 | * @cpu: the processor in question. |
e69f6186 YB |
7141 | * |
7142 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
7143 | */ |
7144 | int idle_cpu(int cpu) | |
7145 | { | |
908a3283 TG |
7146 | struct rq *rq = cpu_rq(cpu); |
7147 | ||
7148 | if (rq->curr != rq->idle) | |
7149 | return 0; | |
7150 | ||
7151 | if (rq->nr_running) | |
7152 | return 0; | |
7153 | ||
7154 | #ifdef CONFIG_SMP | |
126c2092 | 7155 | if (rq->ttwu_pending) |
908a3283 TG |
7156 | return 0; |
7157 | #endif | |
7158 | ||
7159 | return 1; | |
1da177e4 LT |
7160 | } |
7161 | ||
943d355d RJ |
7162 | /** |
7163 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
7164 | * @cpu: the CPU in question. | |
7165 | * | |
7166 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
7167 | */ | |
7168 | int available_idle_cpu(int cpu) | |
7169 | { | |
7170 | if (!idle_cpu(cpu)) | |
7171 | return 0; | |
7172 | ||
247f2f6f RJ |
7173 | if (vcpu_is_preempted(cpu)) |
7174 | return 0; | |
7175 | ||
908a3283 | 7176 | return 1; |
1da177e4 LT |
7177 | } |
7178 | ||
1da177e4 | 7179 | /** |
d1ccc66d | 7180 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 7181 | * @cpu: the processor in question. |
e69f6186 | 7182 | * |
d1ccc66d | 7183 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 7184 | */ |
36c8b586 | 7185 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
7186 | { |
7187 | return cpu_rq(cpu)->idle; | |
7188 | } | |
7189 | ||
7d6a905f VK |
7190 | #ifdef CONFIG_SMP |
7191 | /* | |
7192 | * This function computes an effective utilization for the given CPU, to be | |
7193 | * used for frequency selection given the linear relation: f = u * f_max. | |
7194 | * | |
7195 | * The scheduler tracks the following metrics: | |
7196 | * | |
7197 | * cpu_util_{cfs,rt,dl,irq}() | |
7198 | * cpu_bw_dl() | |
7199 | * | |
7200 | * Where the cfs,rt and dl util numbers are tracked with the same metric and | |
7201 | * synchronized windows and are thus directly comparable. | |
7202 | * | |
7203 | * The cfs,rt,dl utilization are the running times measured with rq->clock_task | |
7204 | * which excludes things like IRQ and steal-time. These latter are then accrued | |
7205 | * in the irq utilization. | |
7206 | * | |
7207 | * The DL bandwidth number otoh is not a measured metric but a value computed | |
7208 | * based on the task model parameters and gives the minimal utilization | |
7209 | * required to meet deadlines. | |
7210 | */ | |
a5418be9 | 7211 | unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, |
bb447999 | 7212 | enum cpu_util_type type, |
7d6a905f VK |
7213 | struct task_struct *p) |
7214 | { | |
bb447999 | 7215 | unsigned long dl_util, util, irq, max; |
7d6a905f VK |
7216 | struct rq *rq = cpu_rq(cpu); |
7217 | ||
bb447999 DE |
7218 | max = arch_scale_cpu_capacity(cpu); |
7219 | ||
7d6a905f VK |
7220 | if (!uclamp_is_used() && |
7221 | type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) { | |
7222 | return max; | |
7223 | } | |
7224 | ||
7225 | /* | |
7226 | * Early check to see if IRQ/steal time saturates the CPU, can be | |
7227 | * because of inaccuracies in how we track these -- see | |
7228 | * update_irq_load_avg(). | |
7229 | */ | |
7230 | irq = cpu_util_irq(rq); | |
7231 | if (unlikely(irq >= max)) | |
7232 | return max; | |
7233 | ||
7234 | /* | |
7235 | * Because the time spend on RT/DL tasks is visible as 'lost' time to | |
7236 | * CFS tasks and we use the same metric to track the effective | |
7237 | * utilization (PELT windows are synchronized) we can directly add them | |
7238 | * to obtain the CPU's actual utilization. | |
7239 | * | |
7240 | * CFS and RT utilization can be boosted or capped, depending on | |
7241 | * utilization clamp constraints requested by currently RUNNABLE | |
7242 | * tasks. | |
7243 | * When there are no CFS RUNNABLE tasks, clamps are released and | |
7244 | * frequency will be gracefully reduced with the utilization decay. | |
7245 | */ | |
7246 | util = util_cfs + cpu_util_rt(rq); | |
7247 | if (type == FREQUENCY_UTIL) | |
7248 | util = uclamp_rq_util_with(rq, util, p); | |
7249 | ||
7250 | dl_util = cpu_util_dl(rq); | |
7251 | ||
7252 | /* | |
7253 | * For frequency selection we do not make cpu_util_dl() a permanent part | |
7254 | * of this sum because we want to use cpu_bw_dl() later on, but we need | |
7255 | * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such | |
7256 | * that we select f_max when there is no idle time. | |
7257 | * | |
7258 | * NOTE: numerical errors or stop class might cause us to not quite hit | |
7259 | * saturation when we should -- something for later. | |
7260 | */ | |
7261 | if (util + dl_util >= max) | |
7262 | return max; | |
7263 | ||
7264 | /* | |
7265 | * OTOH, for energy computation we need the estimated running time, so | |
7266 | * include util_dl and ignore dl_bw. | |
7267 | */ | |
7268 | if (type == ENERGY_UTIL) | |
7269 | util += dl_util; | |
7270 | ||
7271 | /* | |
7272 | * There is still idle time; further improve the number by using the | |
7273 | * irq metric. Because IRQ/steal time is hidden from the task clock we | |
7274 | * need to scale the task numbers: | |
7275 | * | |
7276 | * max - irq | |
7277 | * U' = irq + --------- * U | |
7278 | * max | |
7279 | */ | |
7280 | util = scale_irq_capacity(util, irq, max); | |
7281 | util += irq; | |
7282 | ||
7283 | /* | |
7284 | * Bandwidth required by DEADLINE must always be granted while, for | |
7285 | * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism | |
7286 | * to gracefully reduce the frequency when no tasks show up for longer | |
7287 | * periods of time. | |
7288 | * | |
7289 | * Ideally we would like to set bw_dl as min/guaranteed freq and util + | |
7290 | * bw_dl as requested freq. However, cpufreq is not yet ready for such | |
7291 | * an interface. So, we only do the latter for now. | |
7292 | */ | |
7293 | if (type == FREQUENCY_UTIL) | |
7294 | util += cpu_bw_dl(rq); | |
7295 | ||
7296 | return min(max, util); | |
7297 | } | |
a5418be9 | 7298 | |
bb447999 | 7299 | unsigned long sched_cpu_util(int cpu) |
a5418be9 | 7300 | { |
bb447999 | 7301 | return effective_cpu_util(cpu, cpu_util_cfs(cpu), ENERGY_UTIL, NULL); |
a5418be9 | 7302 | } |
7d6a905f VK |
7303 | #endif /* CONFIG_SMP */ |
7304 | ||
1da177e4 LT |
7305 | /** |
7306 | * find_process_by_pid - find a process with a matching PID value. | |
7307 | * @pid: the pid in question. | |
e69f6186 YB |
7308 | * |
7309 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 7310 | */ |
a9957449 | 7311 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 7312 | { |
228ebcbe | 7313 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
7314 | } |
7315 | ||
c13db6b1 SR |
7316 | /* |
7317 | * sched_setparam() passes in -1 for its policy, to let the functions | |
7318 | * it calls know not to change it. | |
7319 | */ | |
7320 | #define SETPARAM_POLICY -1 | |
7321 | ||
c365c292 TG |
7322 | static void __setscheduler_params(struct task_struct *p, |
7323 | const struct sched_attr *attr) | |
1da177e4 | 7324 | { |
d50dde5a DF |
7325 | int policy = attr->sched_policy; |
7326 | ||
c13db6b1 | 7327 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
7328 | policy = p->policy; |
7329 | ||
1da177e4 | 7330 | p->policy = policy; |
d50dde5a | 7331 | |
aab03e05 DF |
7332 | if (dl_policy(policy)) |
7333 | __setparam_dl(p, attr); | |
39fd8fd2 | 7334 | else if (fair_policy(policy)) |
d50dde5a DF |
7335 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
7336 | ||
39fd8fd2 PZ |
7337 | /* |
7338 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
7339 | * !rt_policy. Always setting this ensures that things like | |
7340 | * getparam()/getattr() don't report silly values for !rt tasks. | |
7341 | */ | |
7342 | p->rt_priority = attr->sched_priority; | |
383afd09 | 7343 | p->normal_prio = normal_prio(p); |
b1e82065 | 7344 | set_load_weight(p, true); |
c365c292 | 7345 | } |
39fd8fd2 | 7346 | |
c69e8d9c | 7347 | /* |
d1ccc66d | 7348 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
7349 | */ |
7350 | static bool check_same_owner(struct task_struct *p) | |
7351 | { | |
7352 | const struct cred *cred = current_cred(), *pcred; | |
7353 | bool match; | |
7354 | ||
7355 | rcu_read_lock(); | |
7356 | pcred = __task_cred(p); | |
9c806aa0 EB |
7357 | match = (uid_eq(cred->euid, pcred->euid) || |
7358 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
7359 | rcu_read_unlock(); |
7360 | return match; | |
7361 | } | |
7362 | ||
700a7833 CG |
7363 | /* |
7364 | * Allow unprivileged RT tasks to decrease priority. | |
7365 | * Only issue a capable test if needed and only once to avoid an audit | |
7366 | * event on permitted non-privileged operations: | |
7367 | */ | |
7368 | static int user_check_sched_setscheduler(struct task_struct *p, | |
7369 | const struct sched_attr *attr, | |
7370 | int policy, int reset_on_fork) | |
7371 | { | |
7372 | if (fair_policy(policy)) { | |
7373 | if (attr->sched_nice < task_nice(p) && | |
7374 | !is_nice_reduction(p, attr->sched_nice)) | |
7375 | goto req_priv; | |
7376 | } | |
7377 | ||
7378 | if (rt_policy(policy)) { | |
7379 | unsigned long rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); | |
7380 | ||
7381 | /* Can't set/change the rt policy: */ | |
7382 | if (policy != p->policy && !rlim_rtprio) | |
7383 | goto req_priv; | |
7384 | ||
7385 | /* Can't increase priority: */ | |
7386 | if (attr->sched_priority > p->rt_priority && | |
7387 | attr->sched_priority > rlim_rtprio) | |
7388 | goto req_priv; | |
7389 | } | |
7390 | ||
7391 | /* | |
7392 | * Can't set/change SCHED_DEADLINE policy at all for now | |
7393 | * (safest behavior); in the future we would like to allow | |
7394 | * unprivileged DL tasks to increase their relative deadline | |
7395 | * or reduce their runtime (both ways reducing utilization) | |
7396 | */ | |
7397 | if (dl_policy(policy)) | |
7398 | goto req_priv; | |
7399 | ||
7400 | /* | |
7401 | * Treat SCHED_IDLE as nice 20. Only allow a switch to | |
7402 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
7403 | */ | |
7404 | if (task_has_idle_policy(p) && !idle_policy(policy)) { | |
7405 | if (!is_nice_reduction(p, task_nice(p))) | |
7406 | goto req_priv; | |
7407 | } | |
7408 | ||
7409 | /* Can't change other user's priorities: */ | |
7410 | if (!check_same_owner(p)) | |
7411 | goto req_priv; | |
7412 | ||
7413 | /* Normal users shall not reset the sched_reset_on_fork flag: */ | |
7414 | if (p->sched_reset_on_fork && !reset_on_fork) | |
7415 | goto req_priv; | |
7416 | ||
7417 | return 0; | |
7418 | ||
7419 | req_priv: | |
7420 | if (!capable(CAP_SYS_NICE)) | |
7421 | return -EPERM; | |
7422 | ||
7423 | return 0; | |
7424 | } | |
7425 | ||
d50dde5a DF |
7426 | static int __sched_setscheduler(struct task_struct *p, |
7427 | const struct sched_attr *attr, | |
dbc7f069 | 7428 | bool user, bool pi) |
1da177e4 | 7429 | { |
f558c2b8 PZ |
7430 | int oldpolicy = -1, policy = attr->sched_policy; |
7431 | int retval, oldprio, newprio, queued, running; | |
83ab0aa0 | 7432 | const struct sched_class *prev_class; |
8e5bad7d | 7433 | struct balance_callback *head; |
eb580751 | 7434 | struct rq_flags rf; |
ca94c442 | 7435 | int reset_on_fork; |
7a57f32a | 7436 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 7437 | struct rq *rq; |
1da177e4 | 7438 | |
896bbb25 SRV |
7439 | /* The pi code expects interrupts enabled */ |
7440 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 7441 | recheck: |
d1ccc66d | 7442 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
7443 | if (policy < 0) { |
7444 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 7445 | policy = oldpolicy = p->policy; |
ca94c442 | 7446 | } else { |
7479f3c9 | 7447 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 7448 | |
20f9cd2a | 7449 | if (!valid_policy(policy)) |
ca94c442 LP |
7450 | return -EINVAL; |
7451 | } | |
7452 | ||
794a56eb | 7453 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
7454 | return -EINVAL; |
7455 | ||
1da177e4 LT |
7456 | /* |
7457 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
ae18ad28 | 7458 | * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL, |
dd41f596 | 7459 | * SCHED_BATCH and SCHED_IDLE is 0. |
1da177e4 | 7460 | */ |
ae18ad28 | 7461 | if (attr->sched_priority > MAX_RT_PRIO-1) |
1da177e4 | 7462 | return -EINVAL; |
aab03e05 DF |
7463 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
7464 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
7465 | return -EINVAL; |
7466 | ||
725aad24 | 7467 | if (user) { |
700a7833 CG |
7468 | retval = user_check_sched_setscheduler(p, attr, policy, reset_on_fork); |
7469 | if (retval) | |
7470 | return retval; | |
7471 | ||
794a56eb JL |
7472 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
7473 | return -EINVAL; | |
7474 | ||
b0ae1981 | 7475 | retval = security_task_setscheduler(p); |
725aad24 JF |
7476 | if (retval) |
7477 | return retval; | |
7478 | } | |
7479 | ||
a509a7cd PB |
7480 | /* Update task specific "requested" clamps */ |
7481 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
7482 | retval = uclamp_validate(p, attr); | |
7483 | if (retval) | |
7484 | return retval; | |
7485 | } | |
7486 | ||
710da3c8 JL |
7487 | if (pi) |
7488 | cpuset_read_lock(); | |
7489 | ||
b29739f9 | 7490 | /* |
d1ccc66d | 7491 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 7492 | * changing the priority of the task: |
0122ec5b | 7493 | * |
25985edc | 7494 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
7495 | * runqueue lock must be held. |
7496 | */ | |
eb580751 | 7497 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 7498 | update_rq_clock(rq); |
dc61b1d6 | 7499 | |
34f971f6 | 7500 | /* |
d1ccc66d | 7501 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
7502 | */ |
7503 | if (p == rq->stop) { | |
4b211f2b MP |
7504 | retval = -EINVAL; |
7505 | goto unlock; | |
34f971f6 PZ |
7506 | } |
7507 | ||
a51e9198 | 7508 | /* |
d6b1e911 TG |
7509 | * If not changing anything there's no need to proceed further, |
7510 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 7511 | */ |
d50dde5a | 7512 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 7513 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
7514 | goto change; |
7515 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
7516 | goto change; | |
75381608 | 7517 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 7518 | goto change; |
a509a7cd PB |
7519 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
7520 | goto change; | |
d50dde5a | 7521 | |
d6b1e911 | 7522 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
7523 | retval = 0; |
7524 | goto unlock; | |
a51e9198 | 7525 | } |
d50dde5a | 7526 | change: |
a51e9198 | 7527 | |
dc61b1d6 | 7528 | if (user) { |
332ac17e | 7529 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
7530 | /* |
7531 | * Do not allow realtime tasks into groups that have no runtime | |
7532 | * assigned. | |
7533 | */ | |
7534 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
7535 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
7536 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
7537 | retval = -EPERM; |
7538 | goto unlock; | |
dc61b1d6 | 7539 | } |
dc61b1d6 | 7540 | #endif |
332ac17e | 7541 | #ifdef CONFIG_SMP |
794a56eb JL |
7542 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
7543 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 7544 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
7545 | |
7546 | /* | |
7547 | * Don't allow tasks with an affinity mask smaller than | |
7548 | * the entire root_domain to become SCHED_DEADLINE. We | |
7549 | * will also fail if there's no bandwidth available. | |
7550 | */ | |
3bd37062 | 7551 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 7552 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
7553 | retval = -EPERM; |
7554 | goto unlock; | |
332ac17e DF |
7555 | } |
7556 | } | |
7557 | #endif | |
7558 | } | |
dc61b1d6 | 7559 | |
d1ccc66d | 7560 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
7561 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
7562 | policy = oldpolicy = -1; | |
eb580751 | 7563 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
7564 | if (pi) |
7565 | cpuset_read_unlock(); | |
1da177e4 LT |
7566 | goto recheck; |
7567 | } | |
332ac17e DF |
7568 | |
7569 | /* | |
7570 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
7571 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
7572 | * is available. | |
7573 | */ | |
06a76fe0 | 7574 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
7575 | retval = -EBUSY; |
7576 | goto unlock; | |
332ac17e DF |
7577 | } |
7578 | ||
c365c292 TG |
7579 | p->sched_reset_on_fork = reset_on_fork; |
7580 | oldprio = p->prio; | |
7581 | ||
f558c2b8 | 7582 | newprio = __normal_prio(policy, attr->sched_priority, attr->sched_nice); |
dbc7f069 PZ |
7583 | if (pi) { |
7584 | /* | |
7585 | * Take priority boosted tasks into account. If the new | |
7586 | * effective priority is unchanged, we just store the new | |
7587 | * normal parameters and do not touch the scheduler class and | |
7588 | * the runqueue. This will be done when the task deboost | |
7589 | * itself. | |
7590 | */ | |
f558c2b8 PZ |
7591 | newprio = rt_effective_prio(p, newprio); |
7592 | if (newprio == oldprio) | |
ff77e468 | 7593 | queue_flags &= ~DEQUEUE_MOVE; |
c365c292 TG |
7594 | } |
7595 | ||
da0c1e65 | 7596 | queued = task_on_rq_queued(p); |
051a1d1a | 7597 | running = task_current(rq, p); |
da0c1e65 | 7598 | if (queued) |
ff77e468 | 7599 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 7600 | if (running) |
f3cd1c4e | 7601 | put_prev_task(rq, p); |
f6b53205 | 7602 | |
83ab0aa0 | 7603 | prev_class = p->sched_class; |
a509a7cd | 7604 | |
f558c2b8 PZ |
7605 | if (!(attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)) { |
7606 | __setscheduler_params(p, attr); | |
7607 | __setscheduler_prio(p, newprio); | |
7608 | } | |
a509a7cd | 7609 | __setscheduler_uclamp(p, attr); |
f6b53205 | 7610 | |
da0c1e65 | 7611 | if (queued) { |
81a44c54 TG |
7612 | /* |
7613 | * We enqueue to tail when the priority of a task is | |
7614 | * increased (user space view). | |
7615 | */ | |
ff77e468 PZ |
7616 | if (oldprio < p->prio) |
7617 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 7618 | |
ff77e468 | 7619 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 7620 | } |
a399d233 | 7621 | if (running) |
03b7fad1 | 7622 | set_next_task(rq, p); |
cb469845 | 7623 | |
da7a735e | 7624 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
7625 | |
7626 | /* Avoid rq from going away on us: */ | |
7627 | preempt_disable(); | |
565790d2 | 7628 | head = splice_balance_callbacks(rq); |
eb580751 | 7629 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 7630 | |
710da3c8 JL |
7631 | if (pi) { |
7632 | cpuset_read_unlock(); | |
dbc7f069 | 7633 | rt_mutex_adjust_pi(p); |
710da3c8 | 7634 | } |
95e02ca9 | 7635 | |
d1ccc66d | 7636 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 7637 | balance_callbacks(rq, head); |
4c9a4bc8 | 7638 | preempt_enable(); |
95e02ca9 | 7639 | |
1da177e4 | 7640 | return 0; |
4b211f2b MP |
7641 | |
7642 | unlock: | |
7643 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
7644 | if (pi) |
7645 | cpuset_read_unlock(); | |
4b211f2b | 7646 | return retval; |
1da177e4 | 7647 | } |
961ccddd | 7648 | |
7479f3c9 PZ |
7649 | static int _sched_setscheduler(struct task_struct *p, int policy, |
7650 | const struct sched_param *param, bool check) | |
7651 | { | |
7652 | struct sched_attr attr = { | |
7653 | .sched_policy = policy, | |
7654 | .sched_priority = param->sched_priority, | |
7655 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
7656 | }; | |
7657 | ||
c13db6b1 SR |
7658 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
7659 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
7660 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
7661 | policy &= ~SCHED_RESET_ON_FORK; | |
7662 | attr.sched_policy = policy; | |
7663 | } | |
7664 | ||
dbc7f069 | 7665 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 7666 | } |
961ccddd RR |
7667 | /** |
7668 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
7669 | * @p: the task in question. | |
7670 | * @policy: new policy. | |
7671 | * @param: structure containing the new RT priority. | |
7672 | * | |
7318d4cc PZ |
7673 | * Use sched_set_fifo(), read its comment. |
7674 | * | |
e69f6186 YB |
7675 | * Return: 0 on success. An error code otherwise. |
7676 | * | |
961ccddd RR |
7677 | * NOTE that the task may be already dead. |
7678 | */ | |
7679 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 7680 | const struct sched_param *param) |
961ccddd | 7681 | { |
7479f3c9 | 7682 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 7683 | } |
1da177e4 | 7684 | |
d50dde5a DF |
7685 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
7686 | { | |
dbc7f069 | 7687 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 7688 | } |
d50dde5a | 7689 | |
794a56eb JL |
7690 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
7691 | { | |
7692 | return __sched_setscheduler(p, attr, false, true); | |
7693 | } | |
1eb5dde6 | 7694 | EXPORT_SYMBOL_GPL(sched_setattr_nocheck); |
794a56eb | 7695 | |
961ccddd RR |
7696 | /** |
7697 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
7698 | * @p: the task in question. | |
7699 | * @policy: new policy. | |
7700 | * @param: structure containing the new RT priority. | |
7701 | * | |
7702 | * Just like sched_setscheduler, only don't bother checking if the | |
7703 | * current context has permission. For example, this is needed in | |
7704 | * stop_machine(): we create temporary high priority worker threads, | |
7705 | * but our caller might not have that capability. | |
e69f6186 YB |
7706 | * |
7707 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
7708 | */ |
7709 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 7710 | const struct sched_param *param) |
961ccddd | 7711 | { |
7479f3c9 | 7712 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
7713 | } |
7714 | ||
7318d4cc PZ |
7715 | /* |
7716 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
7717 | * incapable of resource management, which is the one thing an OS really should | |
7718 | * be doing. | |
7719 | * | |
7720 | * This is of course the reason it is limited to privileged users only. | |
7721 | * | |
7722 | * Worse still; it is fundamentally impossible to compose static priority | |
7723 | * workloads. You cannot take two correctly working static prio workloads | |
7724 | * and smash them together and still expect them to work. | |
7725 | * | |
7726 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
7727 | * | |
7728 | * MAX_RT_PRIO / 2 | |
7729 | * | |
7730 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
7731 | * know enough information to make a sensible choice. | |
7732 | */ | |
8b700983 | 7733 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
7734 | { |
7735 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 7736 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7737 | } |
7738 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
7739 | ||
7740 | /* | |
7741 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
7742 | */ | |
8b700983 | 7743 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
7744 | { |
7745 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 7746 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7747 | } |
7748 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
7749 | ||
8b700983 | 7750 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
7751 | { |
7752 | struct sched_attr attr = { | |
7753 | .sched_policy = SCHED_NORMAL, | |
7754 | .sched_nice = nice, | |
7755 | }; | |
8b700983 | 7756 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
7757 | } |
7758 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 7759 | |
95cdf3b7 IM |
7760 | static int |
7761 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 7762 | { |
1da177e4 LT |
7763 | struct sched_param lparam; |
7764 | struct task_struct *p; | |
36c8b586 | 7765 | int retval; |
1da177e4 LT |
7766 | |
7767 | if (!param || pid < 0) | |
7768 | return -EINVAL; | |
7769 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
7770 | return -EFAULT; | |
5fe1d75f ON |
7771 | |
7772 | rcu_read_lock(); | |
7773 | retval = -ESRCH; | |
1da177e4 | 7774 | p = find_process_by_pid(pid); |
710da3c8 JL |
7775 | if (likely(p)) |
7776 | get_task_struct(p); | |
5fe1d75f | 7777 | rcu_read_unlock(); |
36c8b586 | 7778 | |
710da3c8 JL |
7779 | if (likely(p)) { |
7780 | retval = sched_setscheduler(p, policy, &lparam); | |
7781 | put_task_struct(p); | |
7782 | } | |
7783 | ||
1da177e4 LT |
7784 | return retval; |
7785 | } | |
7786 | ||
d50dde5a DF |
7787 | /* |
7788 | * Mimics kernel/events/core.c perf_copy_attr(). | |
7789 | */ | |
d1ccc66d | 7790 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
7791 | { |
7792 | u32 size; | |
7793 | int ret; | |
7794 | ||
d1ccc66d | 7795 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
7796 | memset(attr, 0, sizeof(*attr)); |
7797 | ||
7798 | ret = get_user(size, &uattr->size); | |
7799 | if (ret) | |
7800 | return ret; | |
7801 | ||
d1ccc66d IM |
7802 | /* ABI compatibility quirk: */ |
7803 | if (!size) | |
d50dde5a | 7804 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 7805 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
7806 | goto err_size; |
7807 | ||
dff3a85f AS |
7808 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
7809 | if (ret) { | |
7810 | if (ret == -E2BIG) | |
7811 | goto err_size; | |
7812 | return ret; | |
d50dde5a DF |
7813 | } |
7814 | ||
a509a7cd PB |
7815 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
7816 | size < SCHED_ATTR_SIZE_VER1) | |
7817 | return -EINVAL; | |
7818 | ||
d50dde5a | 7819 | /* |
d1ccc66d | 7820 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
7821 | * to be strict and return an error on out-of-bounds values? |
7822 | */ | |
75e45d51 | 7823 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 7824 | |
e78c7bca | 7825 | return 0; |
d50dde5a DF |
7826 | |
7827 | err_size: | |
7828 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 7829 | return -E2BIG; |
d50dde5a DF |
7830 | } |
7831 | ||
f4dddf90 QP |
7832 | static void get_params(struct task_struct *p, struct sched_attr *attr) |
7833 | { | |
7834 | if (task_has_dl_policy(p)) | |
7835 | __getparam_dl(p, attr); | |
7836 | else if (task_has_rt_policy(p)) | |
7837 | attr->sched_priority = p->rt_priority; | |
7838 | else | |
7839 | attr->sched_nice = task_nice(p); | |
7840 | } | |
7841 | ||
1da177e4 LT |
7842 | /** |
7843 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
7844 | * @pid: the pid in question. | |
7845 | * @policy: new policy. | |
7846 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7847 | * |
7848 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7849 | */ |
d1ccc66d | 7850 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 7851 | { |
c21761f1 JB |
7852 | if (policy < 0) |
7853 | return -EINVAL; | |
7854 | ||
1da177e4 LT |
7855 | return do_sched_setscheduler(pid, policy, param); |
7856 | } | |
7857 | ||
7858 | /** | |
7859 | * sys_sched_setparam - set/change the RT priority of a thread | |
7860 | * @pid: the pid in question. | |
7861 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7862 | * |
7863 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7864 | */ |
5add95d4 | 7865 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7866 | { |
c13db6b1 | 7867 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
7868 | } |
7869 | ||
d50dde5a DF |
7870 | /** |
7871 | * sys_sched_setattr - same as above, but with extended sched_attr | |
7872 | * @pid: the pid in question. | |
5778fccf | 7873 | * @uattr: structure containing the extended parameters. |
db66d756 | 7874 | * @flags: for future extension. |
d50dde5a | 7875 | */ |
6d35ab48 PZ |
7876 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
7877 | unsigned int, flags) | |
d50dde5a DF |
7878 | { |
7879 | struct sched_attr attr; | |
7880 | struct task_struct *p; | |
7881 | int retval; | |
7882 | ||
6d35ab48 | 7883 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
7884 | return -EINVAL; |
7885 | ||
143cf23d MK |
7886 | retval = sched_copy_attr(uattr, &attr); |
7887 | if (retval) | |
7888 | return retval; | |
d50dde5a | 7889 | |
b14ed2c2 | 7890 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 7891 | return -EINVAL; |
1d6362fa PB |
7892 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
7893 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
7894 | |
7895 | rcu_read_lock(); | |
7896 | retval = -ESRCH; | |
7897 | p = find_process_by_pid(pid); | |
a509a7cd PB |
7898 | if (likely(p)) |
7899 | get_task_struct(p); | |
d50dde5a DF |
7900 | rcu_read_unlock(); |
7901 | ||
a509a7cd | 7902 | if (likely(p)) { |
f4dddf90 QP |
7903 | if (attr.sched_flags & SCHED_FLAG_KEEP_PARAMS) |
7904 | get_params(p, &attr); | |
a509a7cd PB |
7905 | retval = sched_setattr(p, &attr); |
7906 | put_task_struct(p); | |
7907 | } | |
7908 | ||
d50dde5a DF |
7909 | return retval; |
7910 | } | |
7911 | ||
1da177e4 LT |
7912 | /** |
7913 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
7914 | * @pid: the pid in question. | |
e69f6186 YB |
7915 | * |
7916 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
7917 | * code. | |
1da177e4 | 7918 | */ |
5add95d4 | 7919 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 7920 | { |
36c8b586 | 7921 | struct task_struct *p; |
3a5c359a | 7922 | int retval; |
1da177e4 LT |
7923 | |
7924 | if (pid < 0) | |
3a5c359a | 7925 | return -EINVAL; |
1da177e4 LT |
7926 | |
7927 | retval = -ESRCH; | |
5fe85be0 | 7928 | rcu_read_lock(); |
1da177e4 LT |
7929 | p = find_process_by_pid(pid); |
7930 | if (p) { | |
7931 | retval = security_task_getscheduler(p); | |
7932 | if (!retval) | |
ca94c442 LP |
7933 | retval = p->policy |
7934 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 7935 | } |
5fe85be0 | 7936 | rcu_read_unlock(); |
1da177e4 LT |
7937 | return retval; |
7938 | } | |
7939 | ||
7940 | /** | |
ca94c442 | 7941 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
7942 | * @pid: the pid in question. |
7943 | * @param: structure containing the RT priority. | |
e69f6186 YB |
7944 | * |
7945 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
7946 | * code. | |
1da177e4 | 7947 | */ |
5add95d4 | 7948 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7949 | { |
ce5f7f82 | 7950 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 7951 | struct task_struct *p; |
3a5c359a | 7952 | int retval; |
1da177e4 LT |
7953 | |
7954 | if (!param || pid < 0) | |
3a5c359a | 7955 | return -EINVAL; |
1da177e4 | 7956 | |
5fe85be0 | 7957 | rcu_read_lock(); |
1da177e4 LT |
7958 | p = find_process_by_pid(pid); |
7959 | retval = -ESRCH; | |
7960 | if (!p) | |
7961 | goto out_unlock; | |
7962 | ||
7963 | retval = security_task_getscheduler(p); | |
7964 | if (retval) | |
7965 | goto out_unlock; | |
7966 | ||
ce5f7f82 PZ |
7967 | if (task_has_rt_policy(p)) |
7968 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 7969 | rcu_read_unlock(); |
1da177e4 LT |
7970 | |
7971 | /* | |
7972 | * This one might sleep, we cannot do it with a spinlock held ... | |
7973 | */ | |
7974 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
7975 | ||
1da177e4 LT |
7976 | return retval; |
7977 | ||
7978 | out_unlock: | |
5fe85be0 | 7979 | rcu_read_unlock(); |
1da177e4 LT |
7980 | return retval; |
7981 | } | |
7982 | ||
1251201c IM |
7983 | /* |
7984 | * Copy the kernel size attribute structure (which might be larger | |
7985 | * than what user-space knows about) to user-space. | |
7986 | * | |
7987 | * Note that all cases are valid: user-space buffer can be larger or | |
7988 | * smaller than the kernel-space buffer. The usual case is that both | |
7989 | * have the same size. | |
7990 | */ | |
7991 | static int | |
7992 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
7993 | struct sched_attr *kattr, | |
7994 | unsigned int usize) | |
d50dde5a | 7995 | { |
1251201c | 7996 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 7997 | |
96d4f267 | 7998 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
7999 | return -EFAULT; |
8000 | ||
8001 | /* | |
1251201c IM |
8002 | * sched_getattr() ABI forwards and backwards compatibility: |
8003 | * | |
8004 | * If usize == ksize then we just copy everything to user-space and all is good. | |
8005 | * | |
8006 | * If usize < ksize then we only copy as much as user-space has space for, | |
8007 | * this keeps ABI compatibility as well. We skip the rest. | |
8008 | * | |
8009 | * If usize > ksize then user-space is using a newer version of the ABI, | |
8010 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
8011 | * detect the kernel's knowledge of attributes from the attr->size value | |
8012 | * which is set to ksize in this case. | |
d50dde5a | 8013 | */ |
1251201c | 8014 | kattr->size = min(usize, ksize); |
d50dde5a | 8015 | |
1251201c | 8016 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
8017 | return -EFAULT; |
8018 | ||
22400674 | 8019 | return 0; |
d50dde5a DF |
8020 | } |
8021 | ||
8022 | /** | |
aab03e05 | 8023 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 8024 | * @pid: the pid in question. |
5778fccf | 8025 | * @uattr: structure containing the extended parameters. |
dff3a85f | 8026 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 8027 | * @flags: for future extension. |
d50dde5a | 8028 | */ |
6d35ab48 | 8029 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 8030 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 8031 | { |
1251201c | 8032 | struct sched_attr kattr = { }; |
d50dde5a DF |
8033 | struct task_struct *p; |
8034 | int retval; | |
8035 | ||
1251201c IM |
8036 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
8037 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
8038 | return -EINVAL; |
8039 | ||
8040 | rcu_read_lock(); | |
8041 | p = find_process_by_pid(pid); | |
8042 | retval = -ESRCH; | |
8043 | if (!p) | |
8044 | goto out_unlock; | |
8045 | ||
8046 | retval = security_task_getscheduler(p); | |
8047 | if (retval) | |
8048 | goto out_unlock; | |
8049 | ||
1251201c | 8050 | kattr.sched_policy = p->policy; |
7479f3c9 | 8051 | if (p->sched_reset_on_fork) |
1251201c | 8052 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
f4dddf90 | 8053 | get_params(p, &kattr); |
7ad721bf | 8054 | kattr.sched_flags &= SCHED_FLAG_ALL; |
d50dde5a | 8055 | |
a509a7cd | 8056 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
8057 | /* |
8058 | * This could race with another potential updater, but this is fine | |
8059 | * because it'll correctly read the old or the new value. We don't need | |
8060 | * to guarantee who wins the race as long as it doesn't return garbage. | |
8061 | */ | |
1251201c IM |
8062 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
8063 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
8064 | #endif |
8065 | ||
d50dde5a DF |
8066 | rcu_read_unlock(); |
8067 | ||
1251201c | 8068 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
8069 | |
8070 | out_unlock: | |
8071 | rcu_read_unlock(); | |
8072 | return retval; | |
8073 | } | |
8074 | ||
234b8ab6 WD |
8075 | #ifdef CONFIG_SMP |
8076 | int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask) | |
1da177e4 | 8077 | { |
234b8ab6 WD |
8078 | int ret = 0; |
8079 | ||
8080 | /* | |
8081 | * If the task isn't a deadline task or admission control is | |
8082 | * disabled then we don't care about affinity changes. | |
8083 | */ | |
8084 | if (!task_has_dl_policy(p) || !dl_bandwidth_enabled()) | |
8085 | return 0; | |
8086 | ||
8087 | /* | |
8088 | * Since bandwidth control happens on root_domain basis, | |
8089 | * if admission test is enabled, we only admit -deadline | |
8090 | * tasks allowed to run on all the CPUs in the task's | |
8091 | * root_domain. | |
8092 | */ | |
8093 | rcu_read_lock(); | |
8094 | if (!cpumask_subset(task_rq(p)->rd->span, mask)) | |
8095 | ret = -EBUSY; | |
8096 | rcu_read_unlock(); | |
8097 | return ret; | |
8098 | } | |
8099 | #endif | |
8100 | ||
db3b02ae WD |
8101 | static int |
8102 | __sched_setaffinity(struct task_struct *p, const struct cpumask *mask) | |
1da177e4 | 8103 | { |
36c8b586 | 8104 | int retval; |
5a16f3d3 | 8105 | cpumask_var_t cpus_allowed, new_mask; |
1da177e4 | 8106 | |
db3b02ae WD |
8107 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) |
8108 | return -ENOMEM; | |
1da177e4 | 8109 | |
5a16f3d3 RR |
8110 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { |
8111 | retval = -ENOMEM; | |
8112 | goto out_free_cpus_allowed; | |
8113 | } | |
e4099a5e PZ |
8114 | |
8115 | cpuset_cpus_allowed(p, cpus_allowed); | |
db3b02ae | 8116 | cpumask_and(new_mask, mask, cpus_allowed); |
e4099a5e | 8117 | |
234b8ab6 WD |
8118 | retval = dl_task_check_affinity(p, new_mask); |
8119 | if (retval) | |
8120 | goto out_free_new_mask; | |
49246274 | 8121 | again: |
07ec77a1 | 8122 | retval = __set_cpus_allowed_ptr(p, new_mask, SCA_CHECK | SCA_USER); |
db3b02ae WD |
8123 | if (retval) |
8124 | goto out_free_new_mask; | |
1da177e4 | 8125 | |
db3b02ae WD |
8126 | cpuset_cpus_allowed(p, cpus_allowed); |
8127 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8128 | /* | |
8129 | * We must have raced with a concurrent cpuset update. | |
8130 | * Just reset the cpumask to the cpuset's cpus_allowed. | |
8131 | */ | |
8132 | cpumask_copy(new_mask, cpus_allowed); | |
8133 | goto again; | |
8707d8b8 | 8134 | } |
db3b02ae | 8135 | |
16303ab2 | 8136 | out_free_new_mask: |
5a16f3d3 RR |
8137 | free_cpumask_var(new_mask); |
8138 | out_free_cpus_allowed: | |
8139 | free_cpumask_var(cpus_allowed); | |
db3b02ae WD |
8140 | return retval; |
8141 | } | |
8142 | ||
8143 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) | |
8144 | { | |
36c8b586 IM |
8145 | struct task_struct *p; |
8146 | int retval; | |
1da177e4 | 8147 | |
23f5d142 | 8148 | rcu_read_lock(); |
1da177e4 LT |
8149 | |
8150 | p = find_process_by_pid(pid); | |
8151 | if (!p) { | |
23f5d142 | 8152 | rcu_read_unlock(); |
1da177e4 LT |
8153 | return -ESRCH; |
8154 | } | |
8155 | ||
23f5d142 | 8156 | /* Prevent p going away */ |
1da177e4 | 8157 | get_task_struct(p); |
23f5d142 | 8158 | rcu_read_unlock(); |
1da177e4 | 8159 | |
14a40ffc TH |
8160 | if (p->flags & PF_NO_SETAFFINITY) { |
8161 | retval = -EINVAL; | |
8162 | goto out_put_task; | |
8163 | } | |
db3b02ae | 8164 | |
4c44aaaf EB |
8165 | if (!check_same_owner(p)) { |
8166 | rcu_read_lock(); | |
8167 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
8168 | rcu_read_unlock(); | |
db3b02ae WD |
8169 | retval = -EPERM; |
8170 | goto out_put_task; | |
4c44aaaf EB |
8171 | } |
8172 | rcu_read_unlock(); | |
8173 | } | |
1da177e4 | 8174 | |
b0ae1981 | 8175 | retval = security_task_setscheduler(p); |
e7834f8f | 8176 | if (retval) |
db3b02ae | 8177 | goto out_put_task; |
1da177e4 | 8178 | |
db3b02ae | 8179 | retval = __sched_setaffinity(p, in_mask); |
5a16f3d3 | 8180 | out_put_task: |
1da177e4 | 8181 | put_task_struct(p); |
1da177e4 LT |
8182 | return retval; |
8183 | } | |
8184 | ||
8185 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 8186 | struct cpumask *new_mask) |
1da177e4 | 8187 | { |
96f874e2 RR |
8188 | if (len < cpumask_size()) |
8189 | cpumask_clear(new_mask); | |
8190 | else if (len > cpumask_size()) | |
8191 | len = cpumask_size(); | |
8192 | ||
1da177e4 LT |
8193 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
8194 | } | |
8195 | ||
8196 | /** | |
d1ccc66d | 8197 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
8198 | * @pid: pid of the process |
8199 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8200 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
8201 | * |
8202 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 8203 | */ |
5add95d4 HC |
8204 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
8205 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 8206 | { |
5a16f3d3 | 8207 | cpumask_var_t new_mask; |
1da177e4 LT |
8208 | int retval; |
8209 | ||
5a16f3d3 RR |
8210 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
8211 | return -ENOMEM; | |
1da177e4 | 8212 | |
5a16f3d3 RR |
8213 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
8214 | if (retval == 0) | |
8215 | retval = sched_setaffinity(pid, new_mask); | |
8216 | free_cpumask_var(new_mask); | |
8217 | return retval; | |
1da177e4 LT |
8218 | } |
8219 | ||
96f874e2 | 8220 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 8221 | { |
36c8b586 | 8222 | struct task_struct *p; |
31605683 | 8223 | unsigned long flags; |
1da177e4 | 8224 | int retval; |
1da177e4 | 8225 | |
23f5d142 | 8226 | rcu_read_lock(); |
1da177e4 LT |
8227 | |
8228 | retval = -ESRCH; | |
8229 | p = find_process_by_pid(pid); | |
8230 | if (!p) | |
8231 | goto out_unlock; | |
8232 | ||
e7834f8f DQ |
8233 | retval = security_task_getscheduler(p); |
8234 | if (retval) | |
8235 | goto out_unlock; | |
8236 | ||
013fdb80 | 8237 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 8238 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 8239 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
8240 | |
8241 | out_unlock: | |
23f5d142 | 8242 | rcu_read_unlock(); |
1da177e4 | 8243 | |
9531b62f | 8244 | return retval; |
1da177e4 LT |
8245 | } |
8246 | ||
8247 | /** | |
d1ccc66d | 8248 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
8249 | * @pid: pid of the process |
8250 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8251 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 8252 | * |
599b4840 ZW |
8253 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
8254 | * error code otherwise. | |
1da177e4 | 8255 | */ |
5add95d4 HC |
8256 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
8257 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
8258 | { |
8259 | int ret; | |
f17c8607 | 8260 | cpumask_var_t mask; |
1da177e4 | 8261 | |
84fba5ec | 8262 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
8263 | return -EINVAL; |
8264 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
8265 | return -EINVAL; |
8266 | ||
f17c8607 RR |
8267 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
8268 | return -ENOMEM; | |
1da177e4 | 8269 | |
f17c8607 RR |
8270 | ret = sched_getaffinity(pid, mask); |
8271 | if (ret == 0) { | |
4de373a1 | 8272 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
8273 | |
8274 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
8275 | ret = -EFAULT; |
8276 | else | |
cd3d8031 | 8277 | ret = retlen; |
f17c8607 RR |
8278 | } |
8279 | free_cpumask_var(mask); | |
1da177e4 | 8280 | |
f17c8607 | 8281 | return ret; |
1da177e4 LT |
8282 | } |
8283 | ||
7d4dd4f1 | 8284 | static void do_sched_yield(void) |
1da177e4 | 8285 | { |
8a8c69c3 PZ |
8286 | struct rq_flags rf; |
8287 | struct rq *rq; | |
8288 | ||
246b3b33 | 8289 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 8290 | |
ae92882e | 8291 | schedstat_inc(rq->yld_count); |
4530d7ab | 8292 | current->sched_class->yield_task(rq); |
1da177e4 | 8293 | |
8a8c69c3 | 8294 | preempt_disable(); |
345a957f | 8295 | rq_unlock_irq(rq, &rf); |
ba74c144 | 8296 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
8297 | |
8298 | schedule(); | |
7d4dd4f1 | 8299 | } |
1da177e4 | 8300 | |
59a74b15 MCC |
8301 | /** |
8302 | * sys_sched_yield - yield the current processor to other threads. | |
8303 | * | |
8304 | * This function yields the current CPU to other tasks. If there are no | |
8305 | * other threads running on this CPU then this function will return. | |
8306 | * | |
8307 | * Return: 0. | |
8308 | */ | |
7d4dd4f1 DB |
8309 | SYSCALL_DEFINE0(sched_yield) |
8310 | { | |
8311 | do_sched_yield(); | |
1da177e4 LT |
8312 | return 0; |
8313 | } | |
8314 | ||
b965f1dd PZI |
8315 | #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) |
8316 | int __sched __cond_resched(void) | |
1da177e4 | 8317 | { |
fe32d3cd | 8318 | if (should_resched(0)) { |
a18b5d01 | 8319 | preempt_schedule_common(); |
1da177e4 LT |
8320 | return 1; |
8321 | } | |
50895825 FW |
8322 | /* |
8323 | * In preemptible kernels, ->rcu_read_lock_nesting tells the tick | |
8324 | * whether the current CPU is in an RCU read-side critical section, | |
8325 | * so the tick can report quiescent states even for CPUs looping | |
8326 | * in kernel context. In contrast, in non-preemptible kernels, | |
8327 | * RCU readers leave no in-memory hints, which means that CPU-bound | |
8328 | * processes executing in kernel context might never report an | |
8329 | * RCU quiescent state. Therefore, the following code causes | |
8330 | * cond_resched() to report a quiescent state, but only when RCU | |
8331 | * is in urgent need of one. | |
8332 | */ | |
b965f1dd | 8333 | #ifndef CONFIG_PREEMPT_RCU |
f79c3ad6 | 8334 | rcu_all_qs(); |
b965f1dd | 8335 | #endif |
1da177e4 LT |
8336 | return 0; |
8337 | } | |
b965f1dd PZI |
8338 | EXPORT_SYMBOL(__cond_resched); |
8339 | #endif | |
8340 | ||
8341 | #ifdef CONFIG_PREEMPT_DYNAMIC | |
99cf983c | 8342 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
8343 | #define cond_resched_dynamic_enabled __cond_resched |
8344 | #define cond_resched_dynamic_disabled ((void *)&__static_call_return0) | |
b965f1dd | 8345 | DEFINE_STATIC_CALL_RET0(cond_resched, __cond_resched); |
ef72661e | 8346 | EXPORT_STATIC_CALL_TRAMP(cond_resched); |
b965f1dd | 8347 | |
8a69fe0b MR |
8348 | #define might_resched_dynamic_enabled __cond_resched |
8349 | #define might_resched_dynamic_disabled ((void *)&__static_call_return0) | |
b965f1dd | 8350 | DEFINE_STATIC_CALL_RET0(might_resched, __cond_resched); |
ef72661e | 8351 | EXPORT_STATIC_CALL_TRAMP(might_resched); |
99cf983c MR |
8352 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
8353 | static DEFINE_STATIC_KEY_FALSE(sk_dynamic_cond_resched); | |
8354 | int __sched dynamic_cond_resched(void) | |
8355 | { | |
8356 | if (!static_branch_unlikely(&sk_dynamic_cond_resched)) | |
8357 | return 0; | |
8358 | return __cond_resched(); | |
8359 | } | |
8360 | EXPORT_SYMBOL(dynamic_cond_resched); | |
8361 | ||
8362 | static DEFINE_STATIC_KEY_FALSE(sk_dynamic_might_resched); | |
8363 | int __sched dynamic_might_resched(void) | |
8364 | { | |
8365 | if (!static_branch_unlikely(&sk_dynamic_might_resched)) | |
8366 | return 0; | |
8367 | return __cond_resched(); | |
8368 | } | |
8369 | EXPORT_SYMBOL(dynamic_might_resched); | |
8370 | #endif | |
35a773a0 | 8371 | #endif |
1da177e4 LT |
8372 | |
8373 | /* | |
613afbf8 | 8374 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
8375 | * call schedule, and on return reacquire the lock. |
8376 | * | |
c1a280b6 | 8377 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
8378 | * operations here to prevent schedule() from being called twice (once via |
8379 | * spin_unlock(), once by hand). | |
8380 | */ | |
613afbf8 | 8381 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 8382 | { |
fe32d3cd | 8383 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
8384 | int ret = 0; |
8385 | ||
f607c668 PZ |
8386 | lockdep_assert_held(lock); |
8387 | ||
4a81e832 | 8388 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 8389 | spin_unlock(lock); |
7e406d1f | 8390 | if (!_cond_resched()) |
95c354fe | 8391 | cpu_relax(); |
6df3cecb | 8392 | ret = 1; |
1da177e4 | 8393 | spin_lock(lock); |
1da177e4 | 8394 | } |
6df3cecb | 8395 | return ret; |
1da177e4 | 8396 | } |
613afbf8 | 8397 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 8398 | |
f3d4b4b1 BG |
8399 | int __cond_resched_rwlock_read(rwlock_t *lock) |
8400 | { | |
8401 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8402 | int ret = 0; | |
8403 | ||
8404 | lockdep_assert_held_read(lock); | |
8405 | ||
8406 | if (rwlock_needbreak(lock) || resched) { | |
8407 | read_unlock(lock); | |
7e406d1f | 8408 | if (!_cond_resched()) |
f3d4b4b1 BG |
8409 | cpu_relax(); |
8410 | ret = 1; | |
8411 | read_lock(lock); | |
8412 | } | |
8413 | return ret; | |
8414 | } | |
8415 | EXPORT_SYMBOL(__cond_resched_rwlock_read); | |
8416 | ||
8417 | int __cond_resched_rwlock_write(rwlock_t *lock) | |
8418 | { | |
8419 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8420 | int ret = 0; | |
8421 | ||
8422 | lockdep_assert_held_write(lock); | |
8423 | ||
8424 | if (rwlock_needbreak(lock) || resched) { | |
8425 | write_unlock(lock); | |
7e406d1f | 8426 | if (!_cond_resched()) |
f3d4b4b1 BG |
8427 | cpu_relax(); |
8428 | ret = 1; | |
8429 | write_lock(lock); | |
8430 | } | |
8431 | return ret; | |
8432 | } | |
8433 | EXPORT_SYMBOL(__cond_resched_rwlock_write); | |
8434 | ||
4c748558 MR |
8435 | #ifdef CONFIG_PREEMPT_DYNAMIC |
8436 | ||
33c64734 | 8437 | #ifdef CONFIG_GENERIC_ENTRY |
4c748558 | 8438 | #include <linux/entry-common.h> |
33c64734 | 8439 | #endif |
4c748558 MR |
8440 | |
8441 | /* | |
8442 | * SC:cond_resched | |
8443 | * SC:might_resched | |
8444 | * SC:preempt_schedule | |
8445 | * SC:preempt_schedule_notrace | |
8446 | * SC:irqentry_exit_cond_resched | |
8447 | * | |
8448 | * | |
8449 | * NONE: | |
8450 | * cond_resched <- __cond_resched | |
8451 | * might_resched <- RET0 | |
8452 | * preempt_schedule <- NOP | |
8453 | * preempt_schedule_notrace <- NOP | |
8454 | * irqentry_exit_cond_resched <- NOP | |
8455 | * | |
8456 | * VOLUNTARY: | |
8457 | * cond_resched <- __cond_resched | |
8458 | * might_resched <- __cond_resched | |
8459 | * preempt_schedule <- NOP | |
8460 | * preempt_schedule_notrace <- NOP | |
8461 | * irqentry_exit_cond_resched <- NOP | |
8462 | * | |
8463 | * FULL: | |
8464 | * cond_resched <- RET0 | |
8465 | * might_resched <- RET0 | |
8466 | * preempt_schedule <- preempt_schedule | |
8467 | * preempt_schedule_notrace <- preempt_schedule_notrace | |
8468 | * irqentry_exit_cond_resched <- irqentry_exit_cond_resched | |
8469 | */ | |
8470 | ||
8471 | enum { | |
8472 | preempt_dynamic_undefined = -1, | |
8473 | preempt_dynamic_none, | |
8474 | preempt_dynamic_voluntary, | |
8475 | preempt_dynamic_full, | |
8476 | }; | |
8477 | ||
8478 | int preempt_dynamic_mode = preempt_dynamic_undefined; | |
8479 | ||
8480 | int sched_dynamic_mode(const char *str) | |
8481 | { | |
8482 | if (!strcmp(str, "none")) | |
8483 | return preempt_dynamic_none; | |
8484 | ||
8485 | if (!strcmp(str, "voluntary")) | |
8486 | return preempt_dynamic_voluntary; | |
8487 | ||
8488 | if (!strcmp(str, "full")) | |
8489 | return preempt_dynamic_full; | |
8490 | ||
8491 | return -EINVAL; | |
8492 | } | |
8493 | ||
99cf983c | 8494 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
8495 | #define preempt_dynamic_enable(f) static_call_update(f, f##_dynamic_enabled) |
8496 | #define preempt_dynamic_disable(f) static_call_update(f, f##_dynamic_disabled) | |
99cf983c MR |
8497 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
8498 | #define preempt_dynamic_enable(f) static_key_enable(&sk_dynamic_##f.key) | |
8499 | #define preempt_dynamic_disable(f) static_key_disable(&sk_dynamic_##f.key) | |
8500 | #else | |
8501 | #error "Unsupported PREEMPT_DYNAMIC mechanism" | |
8502 | #endif | |
8a69fe0b | 8503 | |
4c748558 MR |
8504 | void sched_dynamic_update(int mode) |
8505 | { | |
8506 | /* | |
8507 | * Avoid {NONE,VOLUNTARY} -> FULL transitions from ever ending up in | |
8508 | * the ZERO state, which is invalid. | |
8509 | */ | |
8a69fe0b MR |
8510 | preempt_dynamic_enable(cond_resched); |
8511 | preempt_dynamic_enable(might_resched); | |
8512 | preempt_dynamic_enable(preempt_schedule); | |
8513 | preempt_dynamic_enable(preempt_schedule_notrace); | |
8514 | preempt_dynamic_enable(irqentry_exit_cond_resched); | |
4c748558 MR |
8515 | |
8516 | switch (mode) { | |
8517 | case preempt_dynamic_none: | |
8a69fe0b MR |
8518 | preempt_dynamic_enable(cond_resched); |
8519 | preempt_dynamic_disable(might_resched); | |
8520 | preempt_dynamic_disable(preempt_schedule); | |
8521 | preempt_dynamic_disable(preempt_schedule_notrace); | |
8522 | preempt_dynamic_disable(irqentry_exit_cond_resched); | |
4c748558 MR |
8523 | pr_info("Dynamic Preempt: none\n"); |
8524 | break; | |
8525 | ||
8526 | case preempt_dynamic_voluntary: | |
8a69fe0b MR |
8527 | preempt_dynamic_enable(cond_resched); |
8528 | preempt_dynamic_enable(might_resched); | |
8529 | preempt_dynamic_disable(preempt_schedule); | |
8530 | preempt_dynamic_disable(preempt_schedule_notrace); | |
8531 | preempt_dynamic_disable(irqentry_exit_cond_resched); | |
4c748558 MR |
8532 | pr_info("Dynamic Preempt: voluntary\n"); |
8533 | break; | |
8534 | ||
8535 | case preempt_dynamic_full: | |
8a69fe0b MR |
8536 | preempt_dynamic_disable(cond_resched); |
8537 | preempt_dynamic_disable(might_resched); | |
8538 | preempt_dynamic_enable(preempt_schedule); | |
8539 | preempt_dynamic_enable(preempt_schedule_notrace); | |
8540 | preempt_dynamic_enable(irqentry_exit_cond_resched); | |
4c748558 MR |
8541 | pr_info("Dynamic Preempt: full\n"); |
8542 | break; | |
8543 | } | |
8544 | ||
8545 | preempt_dynamic_mode = mode; | |
8546 | } | |
8547 | ||
8548 | static int __init setup_preempt_mode(char *str) | |
8549 | { | |
8550 | int mode = sched_dynamic_mode(str); | |
8551 | if (mode < 0) { | |
8552 | pr_warn("Dynamic Preempt: unsupported mode: %s\n", str); | |
8553 | return 0; | |
8554 | } | |
8555 | ||
8556 | sched_dynamic_update(mode); | |
8557 | return 1; | |
8558 | } | |
8559 | __setup("preempt=", setup_preempt_mode); | |
8560 | ||
8561 | static void __init preempt_dynamic_init(void) | |
8562 | { | |
8563 | if (preempt_dynamic_mode == preempt_dynamic_undefined) { | |
8564 | if (IS_ENABLED(CONFIG_PREEMPT_NONE)) { | |
8565 | sched_dynamic_update(preempt_dynamic_none); | |
8566 | } else if (IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY)) { | |
8567 | sched_dynamic_update(preempt_dynamic_voluntary); | |
8568 | } else { | |
8569 | /* Default static call setting, nothing to do */ | |
8570 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT)); | |
8571 | preempt_dynamic_mode = preempt_dynamic_full; | |
8572 | pr_info("Dynamic Preempt: full\n"); | |
8573 | } | |
8574 | } | |
8575 | } | |
8576 | ||
cfe43f47 VS |
8577 | #define PREEMPT_MODEL_ACCESSOR(mode) \ |
8578 | bool preempt_model_##mode(void) \ | |
8579 | { \ | |
8580 | WARN_ON_ONCE(preempt_dynamic_mode == preempt_dynamic_undefined); \ | |
8581 | return preempt_dynamic_mode == preempt_dynamic_##mode; \ | |
8582 | } \ | |
8583 | EXPORT_SYMBOL_GPL(preempt_model_##mode) | |
8584 | ||
8585 | PREEMPT_MODEL_ACCESSOR(none); | |
8586 | PREEMPT_MODEL_ACCESSOR(voluntary); | |
8587 | PREEMPT_MODEL_ACCESSOR(full); | |
8588 | ||
4c748558 MR |
8589 | #else /* !CONFIG_PREEMPT_DYNAMIC */ |
8590 | ||
8591 | static inline void preempt_dynamic_init(void) { } | |
8592 | ||
8593 | #endif /* #ifdef CONFIG_PREEMPT_DYNAMIC */ | |
8594 | ||
1da177e4 LT |
8595 | /** |
8596 | * yield - yield the current processor to other threads. | |
8597 | * | |
8e3fabfd PZ |
8598 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
8599 | * | |
8600 | * The scheduler is at all times free to pick the calling task as the most | |
8601 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 8602 | * it, it's already broken. |
8e3fabfd PZ |
8603 | * |
8604 | * Typical broken usage is: | |
8605 | * | |
8606 | * while (!event) | |
d1ccc66d | 8607 | * yield(); |
8e3fabfd PZ |
8608 | * |
8609 | * where one assumes that yield() will let 'the other' process run that will | |
8610 | * make event true. If the current task is a SCHED_FIFO task that will never | |
8611 | * happen. Never use yield() as a progress guarantee!! | |
8612 | * | |
8613 | * If you want to use yield() to wait for something, use wait_event(). | |
8614 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
8615 | * If you still want to use yield(), do not! | |
1da177e4 LT |
8616 | */ |
8617 | void __sched yield(void) | |
8618 | { | |
8619 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 8620 | do_sched_yield(); |
1da177e4 | 8621 | } |
1da177e4 LT |
8622 | EXPORT_SYMBOL(yield); |
8623 | ||
d95f4122 MG |
8624 | /** |
8625 | * yield_to - yield the current processor to another thread in | |
8626 | * your thread group, or accelerate that thread toward the | |
8627 | * processor it's on. | |
16addf95 RD |
8628 | * @p: target task |
8629 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
8630 | * |
8631 | * It's the caller's job to ensure that the target task struct | |
8632 | * can't go away on us before we can do any checks. | |
8633 | * | |
e69f6186 | 8634 | * Return: |
7b270f60 PZ |
8635 | * true (>0) if we indeed boosted the target task. |
8636 | * false (0) if we failed to boost the target. | |
8637 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 8638 | */ |
fa93384f | 8639 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
8640 | { |
8641 | struct task_struct *curr = current; | |
8642 | struct rq *rq, *p_rq; | |
8643 | unsigned long flags; | |
c3c18640 | 8644 | int yielded = 0; |
d95f4122 MG |
8645 | |
8646 | local_irq_save(flags); | |
8647 | rq = this_rq(); | |
8648 | ||
8649 | again: | |
8650 | p_rq = task_rq(p); | |
7b270f60 PZ |
8651 | /* |
8652 | * If we're the only runnable task on the rq and target rq also | |
8653 | * has only one task, there's absolutely no point in yielding. | |
8654 | */ | |
8655 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
8656 | yielded = -ESRCH; | |
8657 | goto out_irq; | |
8658 | } | |
8659 | ||
d95f4122 | 8660 | double_rq_lock(rq, p_rq); |
39e24d8f | 8661 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
8662 | double_rq_unlock(rq, p_rq); |
8663 | goto again; | |
8664 | } | |
8665 | ||
8666 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 8667 | goto out_unlock; |
d95f4122 MG |
8668 | |
8669 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 8670 | goto out_unlock; |
d95f4122 | 8671 | |
0b9d46fc | 8672 | if (task_on_cpu(p_rq, p) || !task_is_running(p)) |
7b270f60 | 8673 | goto out_unlock; |
d95f4122 | 8674 | |
0900acf2 | 8675 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 8676 | if (yielded) { |
ae92882e | 8677 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
8678 | /* |
8679 | * Make p's CPU reschedule; pick_next_entity takes care of | |
8680 | * fairness. | |
8681 | */ | |
8682 | if (preempt && rq != p_rq) | |
8875125e | 8683 | resched_curr(p_rq); |
6d1cafd8 | 8684 | } |
d95f4122 | 8685 | |
7b270f60 | 8686 | out_unlock: |
d95f4122 | 8687 | double_rq_unlock(rq, p_rq); |
7b270f60 | 8688 | out_irq: |
d95f4122 MG |
8689 | local_irq_restore(flags); |
8690 | ||
7b270f60 | 8691 | if (yielded > 0) |
d95f4122 MG |
8692 | schedule(); |
8693 | ||
8694 | return yielded; | |
8695 | } | |
8696 | EXPORT_SYMBOL_GPL(yield_to); | |
8697 | ||
10ab5643 TH |
8698 | int io_schedule_prepare(void) |
8699 | { | |
8700 | int old_iowait = current->in_iowait; | |
8701 | ||
8702 | current->in_iowait = 1; | |
aa8dccca | 8703 | blk_flush_plug(current->plug, true); |
10ab5643 TH |
8704 | return old_iowait; |
8705 | } | |
8706 | ||
8707 | void io_schedule_finish(int token) | |
8708 | { | |
8709 | current->in_iowait = token; | |
8710 | } | |
8711 | ||
1da177e4 | 8712 | /* |
41a2d6cf | 8713 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 8714 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 8715 | */ |
1da177e4 LT |
8716 | long __sched io_schedule_timeout(long timeout) |
8717 | { | |
10ab5643 | 8718 | int token; |
1da177e4 LT |
8719 | long ret; |
8720 | ||
10ab5643 | 8721 | token = io_schedule_prepare(); |
1da177e4 | 8722 | ret = schedule_timeout(timeout); |
10ab5643 | 8723 | io_schedule_finish(token); |
9cff8ade | 8724 | |
1da177e4 LT |
8725 | return ret; |
8726 | } | |
9cff8ade | 8727 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 8728 | |
e3b929b0 | 8729 | void __sched io_schedule(void) |
10ab5643 TH |
8730 | { |
8731 | int token; | |
8732 | ||
8733 | token = io_schedule_prepare(); | |
8734 | schedule(); | |
8735 | io_schedule_finish(token); | |
8736 | } | |
8737 | EXPORT_SYMBOL(io_schedule); | |
8738 | ||
1da177e4 LT |
8739 | /** |
8740 | * sys_sched_get_priority_max - return maximum RT priority. | |
8741 | * @policy: scheduling class. | |
8742 | * | |
e69f6186 YB |
8743 | * Return: On success, this syscall returns the maximum |
8744 | * rt_priority that can be used by a given scheduling class. | |
8745 | * On failure, a negative error code is returned. | |
1da177e4 | 8746 | */ |
5add95d4 | 8747 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
8748 | { |
8749 | int ret = -EINVAL; | |
8750 | ||
8751 | switch (policy) { | |
8752 | case SCHED_FIFO: | |
8753 | case SCHED_RR: | |
ae18ad28 | 8754 | ret = MAX_RT_PRIO-1; |
1da177e4 | 8755 | break; |
aab03e05 | 8756 | case SCHED_DEADLINE: |
1da177e4 | 8757 | case SCHED_NORMAL: |
b0a9499c | 8758 | case SCHED_BATCH: |
dd41f596 | 8759 | case SCHED_IDLE: |
1da177e4 LT |
8760 | ret = 0; |
8761 | break; | |
8762 | } | |
8763 | return ret; | |
8764 | } | |
8765 | ||
8766 | /** | |
8767 | * sys_sched_get_priority_min - return minimum RT priority. | |
8768 | * @policy: scheduling class. | |
8769 | * | |
e69f6186 YB |
8770 | * Return: On success, this syscall returns the minimum |
8771 | * rt_priority that can be used by a given scheduling class. | |
8772 | * On failure, a negative error code is returned. | |
1da177e4 | 8773 | */ |
5add95d4 | 8774 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
8775 | { |
8776 | int ret = -EINVAL; | |
8777 | ||
8778 | switch (policy) { | |
8779 | case SCHED_FIFO: | |
8780 | case SCHED_RR: | |
8781 | ret = 1; | |
8782 | break; | |
aab03e05 | 8783 | case SCHED_DEADLINE: |
1da177e4 | 8784 | case SCHED_NORMAL: |
b0a9499c | 8785 | case SCHED_BATCH: |
dd41f596 | 8786 | case SCHED_IDLE: |
1da177e4 LT |
8787 | ret = 0; |
8788 | } | |
8789 | return ret; | |
8790 | } | |
8791 | ||
abca5fc5 | 8792 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 8793 | { |
36c8b586 | 8794 | struct task_struct *p; |
a4ec24b4 | 8795 | unsigned int time_slice; |
eb580751 | 8796 | struct rq_flags rf; |
dba091b9 | 8797 | struct rq *rq; |
3a5c359a | 8798 | int retval; |
1da177e4 LT |
8799 | |
8800 | if (pid < 0) | |
3a5c359a | 8801 | return -EINVAL; |
1da177e4 LT |
8802 | |
8803 | retval = -ESRCH; | |
1a551ae7 | 8804 | rcu_read_lock(); |
1da177e4 LT |
8805 | p = find_process_by_pid(pid); |
8806 | if (!p) | |
8807 | goto out_unlock; | |
8808 | ||
8809 | retval = security_task_getscheduler(p); | |
8810 | if (retval) | |
8811 | goto out_unlock; | |
8812 | ||
eb580751 | 8813 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
8814 | time_slice = 0; |
8815 | if (p->sched_class->get_rr_interval) | |
8816 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 8817 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 8818 | |
1a551ae7 | 8819 | rcu_read_unlock(); |
abca5fc5 AV |
8820 | jiffies_to_timespec64(time_slice, t); |
8821 | return 0; | |
3a5c359a | 8822 | |
1da177e4 | 8823 | out_unlock: |
1a551ae7 | 8824 | rcu_read_unlock(); |
1da177e4 LT |
8825 | return retval; |
8826 | } | |
8827 | ||
2064a5ab RD |
8828 | /** |
8829 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
8830 | * @pid: pid of the process. | |
8831 | * @interval: userspace pointer to the timeslice value. | |
8832 | * | |
8833 | * this syscall writes the default timeslice value of a given process | |
8834 | * into the user-space timespec buffer. A value of '0' means infinity. | |
8835 | * | |
8836 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
8837 | * an error code. | |
8838 | */ | |
abca5fc5 | 8839 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 8840 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
8841 | { |
8842 | struct timespec64 t; | |
8843 | int retval = sched_rr_get_interval(pid, &t); | |
8844 | ||
8845 | if (retval == 0) | |
8846 | retval = put_timespec64(&t, interval); | |
8847 | ||
8848 | return retval; | |
8849 | } | |
8850 | ||
474b9c77 | 8851 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
8852 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
8853 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
8854 | { |
8855 | struct timespec64 t; | |
8856 | int retval = sched_rr_get_interval(pid, &t); | |
8857 | ||
8858 | if (retval == 0) | |
9afc5eee | 8859 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
8860 | return retval; |
8861 | } | |
8862 | #endif | |
8863 | ||
82a1fcb9 | 8864 | void sched_show_task(struct task_struct *p) |
1da177e4 | 8865 | { |
1da177e4 | 8866 | unsigned long free = 0; |
4e79752c | 8867 | int ppid; |
c930b2c0 | 8868 | |
38200502 TH |
8869 | if (!try_get_task_stack(p)) |
8870 | return; | |
20435d84 | 8871 | |
cc172ff3 | 8872 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 | 8873 | |
b03fbd4f | 8874 | if (task_is_running(p)) |
cc172ff3 | 8875 | pr_cont(" running task "); |
1da177e4 | 8876 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 8877 | free = stack_not_used(p); |
1da177e4 | 8878 | #endif |
a90e984c | 8879 | ppid = 0; |
4e79752c | 8880 | rcu_read_lock(); |
a90e984c ON |
8881 | if (pid_alive(p)) |
8882 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 8883 | rcu_read_unlock(); |
0f03d680 | 8884 | pr_cont(" stack:%-5lu pid:%-5d ppid:%-6d flags:0x%08lx\n", |
cc172ff3 | 8885 | free, task_pid_nr(p), ppid, |
0569b245 | 8886 | read_task_thread_flags(p)); |
1da177e4 | 8887 | |
3d1cb205 | 8888 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 8889 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 8890 | show_stack(p, NULL, KERN_INFO); |
38200502 | 8891 | put_task_stack(p); |
1da177e4 | 8892 | } |
0032f4e8 | 8893 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 8894 | |
5d68cc95 PZ |
8895 | static inline bool |
8896 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
8897 | { | |
2f064a59 PZ |
8898 | unsigned int state = READ_ONCE(p->__state); |
8899 | ||
5d68cc95 PZ |
8900 | /* no filter, everything matches */ |
8901 | if (!state_filter) | |
8902 | return true; | |
8903 | ||
8904 | /* filter, but doesn't match */ | |
2f064a59 | 8905 | if (!(state & state_filter)) |
5d68cc95 PZ |
8906 | return false; |
8907 | ||
8908 | /* | |
8909 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
8910 | * TASK_KILLABLE). | |
8911 | */ | |
5aec788a | 8912 | if (state_filter == TASK_UNINTERRUPTIBLE && (state & TASK_NOLOAD)) |
5d68cc95 PZ |
8913 | return false; |
8914 | ||
8915 | return true; | |
8916 | } | |
8917 | ||
8918 | ||
2f064a59 | 8919 | void show_state_filter(unsigned int state_filter) |
1da177e4 | 8920 | { |
36c8b586 | 8921 | struct task_struct *g, *p; |
1da177e4 | 8922 | |
510f5acc | 8923 | rcu_read_lock(); |
5d07f420 | 8924 | for_each_process_thread(g, p) { |
1da177e4 LT |
8925 | /* |
8926 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 8927 | * console might take a lot of time: |
57675cb9 AR |
8928 | * Also, reset softlockup watchdogs on all CPUs, because |
8929 | * another CPU might be blocked waiting for us to process | |
8930 | * an IPI. | |
1da177e4 LT |
8931 | */ |
8932 | touch_nmi_watchdog(); | |
57675cb9 | 8933 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 8934 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 8935 | sched_show_task(p); |
5d07f420 | 8936 | } |
1da177e4 | 8937 | |
dd41f596 | 8938 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
8939 | if (!state_filter) |
8940 | sysrq_sched_debug_show(); | |
dd41f596 | 8941 | #endif |
510f5acc | 8942 | rcu_read_unlock(); |
e59e2ae2 IM |
8943 | /* |
8944 | * Only show locks if all tasks are dumped: | |
8945 | */ | |
93335a21 | 8946 | if (!state_filter) |
e59e2ae2 | 8947 | debug_show_all_locks(); |
1da177e4 LT |
8948 | } |
8949 | ||
f340c0d1 IM |
8950 | /** |
8951 | * init_idle - set up an idle thread for a given CPU | |
8952 | * @idle: task in question | |
d1ccc66d | 8953 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
8954 | * |
8955 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
8956 | * flag, to make booting more robust. | |
8957 | */ | |
f1a0a376 | 8958 | void __init init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 8959 | { |
70b97a7f | 8960 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
8961 | unsigned long flags; |
8962 | ||
ff51ff84 PZ |
8963 | __sched_fork(0, idle); |
8964 | ||
25834c73 | 8965 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5cb9eaa3 | 8966 | raw_spin_rq_lock(rq); |
5cbd54ef | 8967 | |
2f064a59 | 8968 | idle->__state = TASK_RUNNING; |
dd41f596 | 8969 | idle->se.exec_start = sched_clock(); |
00b89fe0 VS |
8970 | /* |
8971 | * PF_KTHREAD should already be set at this point; regardless, make it | |
8972 | * look like a proper per-CPU kthread. | |
8973 | */ | |
8974 | idle->flags |= PF_IDLE | PF_KTHREAD | PF_NO_SETAFFINITY; | |
8975 | kthread_set_per_cpu(idle, cpu); | |
dd41f596 | 8976 | |
de9b8f5d PZ |
8977 | #ifdef CONFIG_SMP |
8978 | /* | |
b19a888c | 8979 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
8980 | * in that case do_set_cpus_allowed() will not do the right thing. |
8981 | * | |
8982 | * And since this is boot we can forgo the serialization. | |
8983 | */ | |
9cfc3e18 | 8984 | set_cpus_allowed_common(idle, cpumask_of(cpu), 0); |
de9b8f5d | 8985 | #endif |
6506cf6c PZ |
8986 | /* |
8987 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 8988 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
8989 | * lockdep check in task_group() will fail. |
8990 | * | |
8991 | * Similar case to sched_fork(). / Alternatively we could | |
8992 | * use task_rq_lock() here and obtain the other rq->lock. | |
8993 | * | |
8994 | * Silence PROVE_RCU | |
8995 | */ | |
8996 | rcu_read_lock(); | |
dd41f596 | 8997 | __set_task_cpu(idle, cpu); |
6506cf6c | 8998 | rcu_read_unlock(); |
1da177e4 | 8999 | |
5311a98f EB |
9000 | rq->idle = idle; |
9001 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 9002 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 9003 | #ifdef CONFIG_SMP |
3ca7a440 | 9004 | idle->on_cpu = 1; |
4866cde0 | 9005 | #endif |
5cb9eaa3 | 9006 | raw_spin_rq_unlock(rq); |
25834c73 | 9007 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); |
1da177e4 LT |
9008 | |
9009 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 9010 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 9011 | |
dd41f596 IM |
9012 | /* |
9013 | * The idle tasks have their own, simple scheduling class: | |
9014 | */ | |
9015 | idle->sched_class = &idle_sched_class; | |
868baf07 | 9016 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 9017 | vtime_init_idle(idle, cpu); |
de9b8f5d | 9018 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
9019 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
9020 | #endif | |
19978ca6 IM |
9021 | } |
9022 | ||
e1d4eeec NP |
9023 | #ifdef CONFIG_SMP |
9024 | ||
f82f8042 JL |
9025 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
9026 | const struct cpumask *trial) | |
9027 | { | |
06a76fe0 | 9028 | int ret = 1; |
f82f8042 | 9029 | |
1087ad4e | 9030 | if (cpumask_empty(cur)) |
bb2bc55a MG |
9031 | return ret; |
9032 | ||
06a76fe0 | 9033 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
9034 | |
9035 | return ret; | |
9036 | } | |
9037 | ||
7f51412a | 9038 | int task_can_attach(struct task_struct *p, |
b6e8d40d | 9039 | const struct cpumask *cs_effective_cpus) |
7f51412a JL |
9040 | { |
9041 | int ret = 0; | |
9042 | ||
9043 | /* | |
9044 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 9045 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
9046 | * affinity and isolating such threads by their set of |
9047 | * allowed nodes is unnecessary. Thus, cpusets are not | |
9048 | * applicable for such threads. This prevents checking for | |
9049 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 9050 | * before cpus_mask may be changed. |
7f51412a JL |
9051 | */ |
9052 | if (p->flags & PF_NO_SETAFFINITY) { | |
9053 | ret = -EINVAL; | |
9054 | goto out; | |
9055 | } | |
9056 | ||
7f51412a | 9057 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
b6e8d40d WL |
9058 | cs_effective_cpus)) { |
9059 | int cpu = cpumask_any_and(cpu_active_mask, cs_effective_cpus); | |
772b6539 | 9060 | |
b6e8d40d WL |
9061 | if (unlikely(cpu >= nr_cpu_ids)) |
9062 | return -EINVAL; | |
772b6539 DE |
9063 | ret = dl_cpu_busy(cpu, p); |
9064 | } | |
7f51412a | 9065 | |
7f51412a JL |
9066 | out: |
9067 | return ret; | |
9068 | } | |
9069 | ||
f2cb1360 | 9070 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 9071 | |
e6628d5b MG |
9072 | #ifdef CONFIG_NUMA_BALANCING |
9073 | /* Migrate current task p to target_cpu */ | |
9074 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
9075 | { | |
9076 | struct migration_arg arg = { p, target_cpu }; | |
9077 | int curr_cpu = task_cpu(p); | |
9078 | ||
9079 | if (curr_cpu == target_cpu) | |
9080 | return 0; | |
9081 | ||
3bd37062 | 9082 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
9083 | return -EINVAL; |
9084 | ||
9085 | /* TODO: This is not properly updating schedstats */ | |
9086 | ||
286549dc | 9087 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
9088 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
9089 | } | |
0ec8aa00 PZ |
9090 | |
9091 | /* | |
9092 | * Requeue a task on a given node and accurately track the number of NUMA | |
9093 | * tasks on the runqueues | |
9094 | */ | |
9095 | void sched_setnuma(struct task_struct *p, int nid) | |
9096 | { | |
da0c1e65 | 9097 | bool queued, running; |
eb580751 PZ |
9098 | struct rq_flags rf; |
9099 | struct rq *rq; | |
0ec8aa00 | 9100 | |
eb580751 | 9101 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 9102 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
9103 | running = task_current(rq, p); |
9104 | ||
da0c1e65 | 9105 | if (queued) |
1de64443 | 9106 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 9107 | if (running) |
f3cd1c4e | 9108 | put_prev_task(rq, p); |
0ec8aa00 PZ |
9109 | |
9110 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 9111 | |
da0c1e65 | 9112 | if (queued) |
7134b3e9 | 9113 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 9114 | if (running) |
03b7fad1 | 9115 | set_next_task(rq, p); |
eb580751 | 9116 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 9117 | } |
5cc389bc | 9118 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 9119 | |
1da177e4 | 9120 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 9121 | /* |
d1ccc66d | 9122 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 9123 | * offline. |
054b9108 | 9124 | */ |
48c5ccae | 9125 | void idle_task_exit(void) |
1da177e4 | 9126 | { |
48c5ccae | 9127 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 9128 | |
48c5ccae | 9129 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 9130 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 9131 | |
a53efe5f | 9132 | if (mm != &init_mm) { |
252d2a41 | 9133 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
9134 | finish_arch_post_lock_switch(); |
9135 | } | |
bf2c59fc PZ |
9136 | |
9137 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
9138 | } |
9139 | ||
2558aacf | 9140 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 9141 | { |
2558aacf PZ |
9142 | struct task_struct *p = arg; |
9143 | struct rq *rq = this_rq(); | |
9144 | struct rq_flags rf; | |
9145 | int cpu; | |
1da177e4 | 9146 | |
2558aacf PZ |
9147 | raw_spin_lock_irq(&p->pi_lock); |
9148 | rq_lock(rq, &rf); | |
3f1d2a31 | 9149 | |
2558aacf PZ |
9150 | update_rq_clock(rq); |
9151 | ||
9152 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
9153 | cpu = select_fallback_rq(rq->cpu, p); | |
9154 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 9155 | } |
3f1d2a31 | 9156 | |
2558aacf PZ |
9157 | rq_unlock(rq, &rf); |
9158 | raw_spin_unlock_irq(&p->pi_lock); | |
9159 | ||
9160 | put_task_struct(p); | |
9161 | ||
9162 | return 0; | |
10e7071b | 9163 | } |
3f1d2a31 | 9164 | |
2558aacf PZ |
9165 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
9166 | ||
48f24c4d | 9167 | /* |
2558aacf | 9168 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
b5c44773 PZ |
9169 | * |
9170 | * This is enabled below SCHED_AP_ACTIVE; when !cpu_active(), but only | |
9171 | * effective when the hotplug motion is down. | |
1da177e4 | 9172 | */ |
2558aacf | 9173 | static void balance_push(struct rq *rq) |
1da177e4 | 9174 | { |
2558aacf PZ |
9175 | struct task_struct *push_task = rq->curr; |
9176 | ||
5cb9eaa3 | 9177 | lockdep_assert_rq_held(rq); |
b5c44773 | 9178 | |
ae792702 PZ |
9179 | /* |
9180 | * Ensure the thing is persistent until balance_push_set(.on = false); | |
9181 | */ | |
9182 | rq->balance_callback = &balance_push_callback; | |
1da177e4 | 9183 | |
b5c44773 | 9184 | /* |
868ad33b TG |
9185 | * Only active while going offline and when invoked on the outgoing |
9186 | * CPU. | |
b5c44773 | 9187 | */ |
868ad33b | 9188 | if (!cpu_dying(rq->cpu) || rq != this_rq()) |
b5c44773 PZ |
9189 | return; |
9190 | ||
1da177e4 | 9191 | /* |
2558aacf PZ |
9192 | * Both the cpu-hotplug and stop task are in this case and are |
9193 | * required to complete the hotplug process. | |
1da177e4 | 9194 | */ |
00b89fe0 | 9195 | if (kthread_is_per_cpu(push_task) || |
5ba2ffba PZ |
9196 | is_migration_disabled(push_task)) { |
9197 | ||
f2469a1f TG |
9198 | /* |
9199 | * If this is the idle task on the outgoing CPU try to wake | |
9200 | * up the hotplug control thread which might wait for the | |
9201 | * last task to vanish. The rcuwait_active() check is | |
9202 | * accurate here because the waiter is pinned on this CPU | |
9203 | * and can't obviously be running in parallel. | |
3015ef4b TG |
9204 | * |
9205 | * On RT kernels this also has to check whether there are | |
9206 | * pinned and scheduled out tasks on the runqueue. They | |
9207 | * need to leave the migrate disabled section first. | |
f2469a1f | 9208 | */ |
3015ef4b TG |
9209 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
9210 | rcuwait_active(&rq->hotplug_wait)) { | |
5cb9eaa3 | 9211 | raw_spin_rq_unlock(rq); |
f2469a1f | 9212 | rcuwait_wake_up(&rq->hotplug_wait); |
5cb9eaa3 | 9213 | raw_spin_rq_lock(rq); |
f2469a1f | 9214 | } |
2558aacf | 9215 | return; |
f2469a1f | 9216 | } |
48f24c4d | 9217 | |
2558aacf | 9218 | get_task_struct(push_task); |
77bd3970 | 9219 | /* |
2558aacf PZ |
9220 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
9221 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 9222 | */ |
5cb9eaa3 | 9223 | raw_spin_rq_unlock(rq); |
2558aacf PZ |
9224 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, |
9225 | this_cpu_ptr(&push_work)); | |
9226 | /* | |
9227 | * At this point need_resched() is true and we'll take the loop in | |
9228 | * schedule(). The next pick is obviously going to be the stop task | |
5ba2ffba | 9229 | * which kthread_is_per_cpu() and will push this task away. |
2558aacf | 9230 | */ |
5cb9eaa3 | 9231 | raw_spin_rq_lock(rq); |
2558aacf | 9232 | } |
77bd3970 | 9233 | |
2558aacf PZ |
9234 | static void balance_push_set(int cpu, bool on) |
9235 | { | |
9236 | struct rq *rq = cpu_rq(cpu); | |
9237 | struct rq_flags rf; | |
48c5ccae | 9238 | |
2558aacf | 9239 | rq_lock_irqsave(rq, &rf); |
22f667c9 PZ |
9240 | if (on) { |
9241 | WARN_ON_ONCE(rq->balance_callback); | |
ae792702 | 9242 | rq->balance_callback = &balance_push_callback; |
22f667c9 | 9243 | } else if (rq->balance_callback == &balance_push_callback) { |
ae792702 | 9244 | rq->balance_callback = NULL; |
22f667c9 | 9245 | } |
2558aacf PZ |
9246 | rq_unlock_irqrestore(rq, &rf); |
9247 | } | |
e692ab53 | 9248 | |
f2469a1f TG |
9249 | /* |
9250 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
9251 | * inactive. All tasks which are not per CPU kernel threads are either | |
9252 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
9253 | * during wakeup. Wait until the CPU is quiescent. | |
9254 | */ | |
9255 | static void balance_hotplug_wait(void) | |
9256 | { | |
9257 | struct rq *rq = this_rq(); | |
5473e0cc | 9258 | |
3015ef4b TG |
9259 | rcuwait_wait_event(&rq->hotplug_wait, |
9260 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
9261 | TASK_UNINTERRUPTIBLE); |
9262 | } | |
5473e0cc | 9263 | |
2558aacf | 9264 | #else |
dce48a84 | 9265 | |
2558aacf PZ |
9266 | static inline void balance_push(struct rq *rq) |
9267 | { | |
dce48a84 | 9268 | } |
dce48a84 | 9269 | |
2558aacf PZ |
9270 | static inline void balance_push_set(int cpu, bool on) |
9271 | { | |
9272 | } | |
9273 | ||
f2469a1f TG |
9274 | static inline void balance_hotplug_wait(void) |
9275 | { | |
dce48a84 | 9276 | } |
f2469a1f | 9277 | |
1da177e4 LT |
9278 | #endif /* CONFIG_HOTPLUG_CPU */ |
9279 | ||
f2cb1360 | 9280 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
9281 | { |
9282 | if (!rq->online) { | |
9283 | const struct sched_class *class; | |
9284 | ||
c6c4927b | 9285 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
9286 | rq->online = 1; |
9287 | ||
9288 | for_each_class(class) { | |
9289 | if (class->rq_online) | |
9290 | class->rq_online(rq); | |
9291 | } | |
9292 | } | |
9293 | } | |
9294 | ||
f2cb1360 | 9295 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
9296 | { |
9297 | if (rq->online) { | |
9298 | const struct sched_class *class; | |
9299 | ||
9300 | for_each_class(class) { | |
9301 | if (class->rq_offline) | |
9302 | class->rq_offline(rq); | |
9303 | } | |
9304 | ||
c6c4927b | 9305 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
9306 | rq->online = 0; |
9307 | } | |
9308 | } | |
9309 | ||
d1ccc66d IM |
9310 | /* |
9311 | * used to mark begin/end of suspend/resume: | |
9312 | */ | |
9313 | static int num_cpus_frozen; | |
d35be8ba | 9314 | |
1da177e4 | 9315 | /* |
3a101d05 TH |
9316 | * Update cpusets according to cpu_active mask. If cpusets are |
9317 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
9318 | * around partition_sched_domains(). | |
d35be8ba SB |
9319 | * |
9320 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
9321 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 9322 | */ |
40190a78 | 9323 | static void cpuset_cpu_active(void) |
e761b772 | 9324 | { |
40190a78 | 9325 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
9326 | /* |
9327 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
9328 | * resume sequence. As long as this is not the last online | |
9329 | * operation in the resume sequence, just build a single sched | |
9330 | * domain, ignoring cpusets. | |
9331 | */ | |
50e76632 PZ |
9332 | partition_sched_domains(1, NULL, NULL); |
9333 | if (--num_cpus_frozen) | |
135fb3e1 | 9334 | return; |
d35be8ba SB |
9335 | /* |
9336 | * This is the last CPU online operation. So fall through and | |
9337 | * restore the original sched domains by considering the | |
9338 | * cpuset configurations. | |
9339 | */ | |
50e76632 | 9340 | cpuset_force_rebuild(); |
3a101d05 | 9341 | } |
30e03acd | 9342 | cpuset_update_active_cpus(); |
3a101d05 | 9343 | } |
e761b772 | 9344 | |
40190a78 | 9345 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 9346 | { |
40190a78 | 9347 | if (!cpuhp_tasks_frozen) { |
772b6539 DE |
9348 | int ret = dl_cpu_busy(cpu, NULL); |
9349 | ||
9350 | if (ret) | |
9351 | return ret; | |
30e03acd | 9352 | cpuset_update_active_cpus(); |
135fb3e1 | 9353 | } else { |
d35be8ba SB |
9354 | num_cpus_frozen++; |
9355 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 9356 | } |
135fb3e1 | 9357 | return 0; |
e761b772 | 9358 | } |
e761b772 | 9359 | |
40190a78 | 9360 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 9361 | { |
7d976699 | 9362 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 9363 | struct rq_flags rf; |
7d976699 | 9364 | |
22f667c9 | 9365 | /* |
b5c44773 PZ |
9366 | * Clear the balance_push callback and prepare to schedule |
9367 | * regular tasks. | |
22f667c9 | 9368 | */ |
2558aacf PZ |
9369 | balance_push_set(cpu, false); |
9370 | ||
ba2591a5 PZ |
9371 | #ifdef CONFIG_SCHED_SMT |
9372 | /* | |
c5511d03 | 9373 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 9374 | */ |
c5511d03 PZI |
9375 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
9376 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 9377 | #endif |
40190a78 | 9378 | set_cpu_active(cpu, true); |
135fb3e1 | 9379 | |
40190a78 | 9380 | if (sched_smp_initialized) { |
0fb3978b | 9381 | sched_update_numa(cpu, true); |
135fb3e1 | 9382 | sched_domains_numa_masks_set(cpu); |
40190a78 | 9383 | cpuset_cpu_active(); |
e761b772 | 9384 | } |
7d976699 TG |
9385 | |
9386 | /* | |
9387 | * Put the rq online, if not already. This happens: | |
9388 | * | |
9389 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 9390 | * after all CPUs have been brought up. |
7d976699 TG |
9391 | * |
9392 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
9393 | * domains. | |
9394 | */ | |
8a8c69c3 | 9395 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
9396 | if (rq->rd) { |
9397 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9398 | set_rq_online(rq); | |
9399 | } | |
8a8c69c3 | 9400 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 9401 | |
40190a78 | 9402 | return 0; |
135fb3e1 TG |
9403 | } |
9404 | ||
40190a78 | 9405 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 9406 | { |
120455c5 PZ |
9407 | struct rq *rq = cpu_rq(cpu); |
9408 | struct rq_flags rf; | |
135fb3e1 TG |
9409 | int ret; |
9410 | ||
e0b257c3 AMB |
9411 | /* |
9412 | * Remove CPU from nohz.idle_cpus_mask to prevent participating in | |
9413 | * load balancing when not active | |
9414 | */ | |
9415 | nohz_balance_exit_idle(rq); | |
9416 | ||
40190a78 | 9417 | set_cpu_active(cpu, false); |
741ba80f PZ |
9418 | |
9419 | /* | |
9420 | * From this point forward, this CPU will refuse to run any task that | |
9421 | * is not: migrate_disable() or KTHREAD_IS_PER_CPU, and will actively | |
9422 | * push those tasks away until this gets cleared, see | |
9423 | * sched_cpu_dying(). | |
9424 | */ | |
975707f2 PZ |
9425 | balance_push_set(cpu, true); |
9426 | ||
b2454caa | 9427 | /* |
975707f2 PZ |
9428 | * We've cleared cpu_active_mask / set balance_push, wait for all |
9429 | * preempt-disabled and RCU users of this state to go away such that | |
9430 | * all new such users will observe it. | |
b2454caa | 9431 | * |
5ba2ffba PZ |
9432 | * Specifically, we rely on ttwu to no longer target this CPU, see |
9433 | * ttwu_queue_cond() and is_cpu_allowed(). | |
9434 | * | |
b2454caa PZ |
9435 | * Do sync before park smpboot threads to take care the rcu boost case. |
9436 | */ | |
309ba859 | 9437 | synchronize_rcu(); |
40190a78 | 9438 | |
120455c5 PZ |
9439 | rq_lock_irqsave(rq, &rf); |
9440 | if (rq->rd) { | |
9441 | update_rq_clock(rq); | |
9442 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9443 | set_rq_offline(rq); | |
9444 | } | |
9445 | rq_unlock_irqrestore(rq, &rf); | |
9446 | ||
c5511d03 PZI |
9447 | #ifdef CONFIG_SCHED_SMT |
9448 | /* | |
9449 | * When going down, decrement the number of cores with SMT present. | |
9450 | */ | |
9451 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
9452 | static_branch_dec_cpuslocked(&sched_smt_present); | |
3c474b32 PZ |
9453 | |
9454 | sched_core_cpu_deactivate(cpu); | |
c5511d03 PZI |
9455 | #endif |
9456 | ||
40190a78 TG |
9457 | if (!sched_smp_initialized) |
9458 | return 0; | |
9459 | ||
0fb3978b | 9460 | sched_update_numa(cpu, false); |
40190a78 TG |
9461 | ret = cpuset_cpu_inactive(cpu); |
9462 | if (ret) { | |
2558aacf | 9463 | balance_push_set(cpu, false); |
40190a78 | 9464 | set_cpu_active(cpu, true); |
0fb3978b | 9465 | sched_update_numa(cpu, true); |
40190a78 | 9466 | return ret; |
135fb3e1 | 9467 | } |
40190a78 TG |
9468 | sched_domains_numa_masks_clear(cpu); |
9469 | return 0; | |
135fb3e1 TG |
9470 | } |
9471 | ||
94baf7a5 TG |
9472 | static void sched_rq_cpu_starting(unsigned int cpu) |
9473 | { | |
9474 | struct rq *rq = cpu_rq(cpu); | |
9475 | ||
9476 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
9477 | update_max_interval(); |
9478 | } | |
9479 | ||
135fb3e1 TG |
9480 | int sched_cpu_starting(unsigned int cpu) |
9481 | { | |
9edeaea1 | 9482 | sched_core_cpu_starting(cpu); |
94baf7a5 | 9483 | sched_rq_cpu_starting(cpu); |
d84b3131 | 9484 | sched_tick_start(cpu); |
135fb3e1 | 9485 | return 0; |
e761b772 | 9486 | } |
e761b772 | 9487 | |
f2785ddb | 9488 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
9489 | |
9490 | /* | |
9491 | * Invoked immediately before the stopper thread is invoked to bring the | |
9492 | * CPU down completely. At this point all per CPU kthreads except the | |
9493 | * hotplug thread (current) and the stopper thread (inactive) have been | |
9494 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
9495 | * any of those which might be on the way out are gone. | |
9496 | * | |
9497 | * If after this point a bound task is being woken on this CPU then the | |
9498 | * responsible hotplug callback has failed to do it's job. | |
9499 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
9500 | */ | |
9501 | int sched_cpu_wait_empty(unsigned int cpu) | |
9502 | { | |
9503 | balance_hotplug_wait(); | |
9504 | return 0; | |
9505 | } | |
9506 | ||
9507 | /* | |
9508 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
9509 | * might have. Called from the CPU stopper task after ensuring that the | |
9510 | * stopper is the last running task on the CPU, so nr_active count is | |
9511 | * stable. We need to take the teardown thread which is calling this into | |
9512 | * account, so we hand in adjust = 1 to the load calculation. | |
9513 | * | |
9514 | * Also see the comment "Global load-average calculations". | |
9515 | */ | |
9516 | static void calc_load_migrate(struct rq *rq) | |
9517 | { | |
9518 | long delta = calc_load_fold_active(rq, 1); | |
9519 | ||
9520 | if (delta) | |
9521 | atomic_long_add(delta, &calc_load_tasks); | |
9522 | } | |
9523 | ||
36c6e17b VS |
9524 | static void dump_rq_tasks(struct rq *rq, const char *loglvl) |
9525 | { | |
9526 | struct task_struct *g, *p; | |
9527 | int cpu = cpu_of(rq); | |
9528 | ||
5cb9eaa3 | 9529 | lockdep_assert_rq_held(rq); |
36c6e17b VS |
9530 | |
9531 | printk("%sCPU%d enqueued tasks (%u total):\n", loglvl, cpu, rq->nr_running); | |
9532 | for_each_process_thread(g, p) { | |
9533 | if (task_cpu(p) != cpu) | |
9534 | continue; | |
9535 | ||
9536 | if (!task_on_rq_queued(p)) | |
9537 | continue; | |
9538 | ||
9539 | printk("%s\tpid: %d, name: %s\n", loglvl, p->pid, p->comm); | |
9540 | } | |
9541 | } | |
9542 | ||
f2785ddb TG |
9543 | int sched_cpu_dying(unsigned int cpu) |
9544 | { | |
9545 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 9546 | struct rq_flags rf; |
f2785ddb TG |
9547 | |
9548 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 9549 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
9550 | |
9551 | rq_lock_irqsave(rq, &rf); | |
36c6e17b VS |
9552 | if (rq->nr_running != 1 || rq_has_pinned_tasks(rq)) { |
9553 | WARN(true, "Dying CPU not properly vacated!"); | |
9554 | dump_rq_tasks(rq, KERN_WARNING); | |
9555 | } | |
8a8c69c3 PZ |
9556 | rq_unlock_irqrestore(rq, &rf); |
9557 | ||
f2785ddb TG |
9558 | calc_load_migrate(rq); |
9559 | update_max_interval(); | |
e5ef27d0 | 9560 | hrtick_clear(rq); |
3c474b32 | 9561 | sched_core_cpu_dying(cpu); |
f2785ddb TG |
9562 | return 0; |
9563 | } | |
9564 | #endif | |
9565 | ||
1da177e4 LT |
9566 | void __init sched_init_smp(void) |
9567 | { | |
0fb3978b | 9568 | sched_init_numa(NUMA_NO_NODE); |
cb83b629 | 9569 | |
6acce3ef PZ |
9570 | /* |
9571 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 9572 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 9573 | * happen. |
6acce3ef | 9574 | */ |
712555ee | 9575 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 9576 | sched_init_domains(cpu_active_mask); |
712555ee | 9577 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 9578 | |
5c1e1767 | 9579 | /* Move init over to a non-isolated CPU */ |
04d4e665 | 9580 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_TYPE_DOMAIN)) < 0) |
5c1e1767 | 9581 | BUG(); |
15faafc6 | 9582 | current->flags &= ~PF_NO_SETAFFINITY; |
19978ca6 | 9583 | sched_init_granularity(); |
4212823f | 9584 | |
0e3900e6 | 9585 | init_sched_rt_class(); |
1baca4ce | 9586 | init_sched_dl_class(); |
1b568f0a | 9587 | |
e26fbffd | 9588 | sched_smp_initialized = true; |
1da177e4 | 9589 | } |
e26fbffd TG |
9590 | |
9591 | static int __init migration_init(void) | |
9592 | { | |
77a5352b | 9593 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 9594 | return 0; |
1da177e4 | 9595 | } |
e26fbffd TG |
9596 | early_initcall(migration_init); |
9597 | ||
1da177e4 LT |
9598 | #else |
9599 | void __init sched_init_smp(void) | |
9600 | { | |
19978ca6 | 9601 | sched_init_granularity(); |
1da177e4 LT |
9602 | } |
9603 | #endif /* CONFIG_SMP */ | |
9604 | ||
9605 | int in_sched_functions(unsigned long addr) | |
9606 | { | |
1da177e4 LT |
9607 | return in_lock_functions(addr) || |
9608 | (addr >= (unsigned long)__sched_text_start | |
9609 | && addr < (unsigned long)__sched_text_end); | |
9610 | } | |
9611 | ||
029632fb | 9612 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
9613 | /* |
9614 | * Default task group. | |
9615 | * Every task in system belongs to this group at bootup. | |
9616 | */ | |
029632fb | 9617 | struct task_group root_task_group; |
35cf4e50 | 9618 | LIST_HEAD(task_groups); |
b0367629 WL |
9619 | |
9620 | /* Cacheline aligned slab cache for task_group */ | |
9621 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 9622 | #endif |
6f505b16 | 9623 | |
1da177e4 LT |
9624 | void __init sched_init(void) |
9625 | { | |
a1dc0446 | 9626 | unsigned long ptr = 0; |
55627e3c | 9627 | int i; |
434d53b0 | 9628 | |
c3a340f7 | 9629 | /* Make sure the linker didn't screw up */ |
546a3fee PZ |
9630 | BUG_ON(&idle_sched_class != &fair_sched_class + 1 || |
9631 | &fair_sched_class != &rt_sched_class + 1 || | |
9632 | &rt_sched_class != &dl_sched_class + 1); | |
c3a340f7 | 9633 | #ifdef CONFIG_SMP |
546a3fee | 9634 | BUG_ON(&dl_sched_class != &stop_sched_class + 1); |
c3a340f7 SRV |
9635 | #endif |
9636 | ||
5822a454 | 9637 | wait_bit_init(); |
9dcb8b68 | 9638 | |
434d53b0 | 9639 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 9640 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
9641 | #endif |
9642 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 9643 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 9644 | #endif |
a1dc0446 QC |
9645 | if (ptr) { |
9646 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
9647 | |
9648 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 9649 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
9650 | ptr += nr_cpu_ids * sizeof(void **); |
9651 | ||
07e06b01 | 9652 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 9653 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 9654 | |
b1d1779e WY |
9655 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
9656 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 9657 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 9658 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9659 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
9660 | ptr += nr_cpu_ids * sizeof(void **); |
9661 | ||
07e06b01 | 9662 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
9663 | ptr += nr_cpu_ids * sizeof(void **); |
9664 | ||
6d6bc0ad | 9665 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 9666 | } |
dd41f596 | 9667 | |
d1ccc66d | 9668 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
332ac17e | 9669 | |
57d885fe GH |
9670 | #ifdef CONFIG_SMP |
9671 | init_defrootdomain(); | |
9672 | #endif | |
9673 | ||
d0b27fa7 | 9674 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9675 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 9676 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 9677 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 9678 | |
7c941438 | 9679 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
9680 | task_group_cache = KMEM_CACHE(task_group, 0); |
9681 | ||
07e06b01 YZ |
9682 | list_add(&root_task_group.list, &task_groups); |
9683 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 9684 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 9685 | autogroup_init(&init_task); |
7c941438 | 9686 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 9687 | |
0a945022 | 9688 | for_each_possible_cpu(i) { |
70b97a7f | 9689 | struct rq *rq; |
1da177e4 LT |
9690 | |
9691 | rq = cpu_rq(i); | |
5cb9eaa3 | 9692 | raw_spin_lock_init(&rq->__lock); |
7897986b | 9693 | rq->nr_running = 0; |
dce48a84 TG |
9694 | rq->calc_load_active = 0; |
9695 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 9696 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
9697 | init_rt_rq(&rq->rt); |
9698 | init_dl_rq(&rq->dl); | |
dd41f596 | 9699 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 9700 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 9701 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 9702 | /* |
d1ccc66d | 9703 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
9704 | * |
9705 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
9706 | * gets 100% of the CPU resources in the system. This overall |
9707 | * system CPU resource is divided among the tasks of | |
07e06b01 | 9708 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
9709 | * based on each entity's (task or task-group's) weight |
9710 | * (se->load.weight). | |
9711 | * | |
07e06b01 | 9712 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 9713 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 9714 | * then A0's share of the CPU resource is: |
354d60c2 | 9715 | * |
0d905bca | 9716 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 9717 | * |
07e06b01 YZ |
9718 | * We achieve this by letting root_task_group's tasks sit |
9719 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 9720 | */ |
07e06b01 | 9721 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
9722 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9723 | ||
9724 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9725 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9726 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 9727 | #endif |
1da177e4 | 9728 | #ifdef CONFIG_SMP |
41c7ce9a | 9729 | rq->sd = NULL; |
57d885fe | 9730 | rq->rd = NULL; |
ca6d75e6 | 9731 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
b5c44773 | 9732 | rq->balance_callback = &balance_push_callback; |
1da177e4 | 9733 | rq->active_balance = 0; |
dd41f596 | 9734 | rq->next_balance = jiffies; |
1da177e4 | 9735 | rq->push_cpu = 0; |
0a2966b4 | 9736 | rq->cpu = i; |
1f11eb6a | 9737 | rq->online = 0; |
eae0c9df MG |
9738 | rq->idle_stamp = 0; |
9739 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
94aafc3e PZ |
9740 | rq->wake_stamp = jiffies; |
9741 | rq->wake_avg_idle = rq->avg_idle; | |
9bd721c5 | 9742 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
9743 | |
9744 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
9745 | ||
dc938520 | 9746 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 9747 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 9748 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 9749 | atomic_set(&rq->nohz_flags, 0); |
90b5363a | 9750 | |
545b8c8d | 9751 | INIT_CSD(&rq->nohz_csd, nohz_csd_func, rq); |
83cd4fe2 | 9752 | #endif |
f2469a1f TG |
9753 | #ifdef CONFIG_HOTPLUG_CPU |
9754 | rcuwait_init(&rq->hotplug_wait); | |
83cd4fe2 | 9755 | #endif |
9fd81dd5 | 9756 | #endif /* CONFIG_SMP */ |
77a021be | 9757 | hrtick_rq_init(rq); |
1da177e4 | 9758 | atomic_set(&rq->nr_iowait, 0); |
9edeaea1 PZ |
9759 | |
9760 | #ifdef CONFIG_SCHED_CORE | |
3c474b32 | 9761 | rq->core = rq; |
539f6512 | 9762 | rq->core_pick = NULL; |
9edeaea1 | 9763 | rq->core_enabled = 0; |
539f6512 | 9764 | rq->core_tree = RB_ROOT; |
4feee7d1 JD |
9765 | rq->core_forceidle_count = 0; |
9766 | rq->core_forceidle_occupation = 0; | |
9767 | rq->core_forceidle_start = 0; | |
539f6512 PZ |
9768 | |
9769 | rq->core_cookie = 0UL; | |
9edeaea1 | 9770 | #endif |
1da177e4 LT |
9771 | } |
9772 | ||
b1e82065 | 9773 | set_load_weight(&init_task, false); |
b50f60ce | 9774 | |
1da177e4 LT |
9775 | /* |
9776 | * The boot idle thread does lazy MMU switching as well: | |
9777 | */ | |
f1f10076 | 9778 | mmgrab(&init_mm); |
1da177e4 LT |
9779 | enter_lazy_tlb(&init_mm, current); |
9780 | ||
40966e31 EB |
9781 | /* |
9782 | * The idle task doesn't need the kthread struct to function, but it | |
9783 | * is dressed up as a per-CPU kthread and thus needs to play the part | |
9784 | * if we want to avoid special-casing it in code that deals with per-CPU | |
9785 | * kthreads. | |
9786 | */ | |
dd621ee0 | 9787 | WARN_ON(!set_kthread_struct(current)); |
40966e31 | 9788 | |
1da177e4 LT |
9789 | /* |
9790 | * Make us the idle thread. Technically, schedule() should not be | |
9791 | * called from this thread, however somewhere below it might be, | |
9792 | * but because we are the idle thread, we just pick up running again | |
9793 | * when this runqueue becomes "idle". | |
9794 | */ | |
9795 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9796 | |
9797 | calc_load_update = jiffies + LOAD_FREQ; | |
9798 | ||
bf4d83f6 | 9799 | #ifdef CONFIG_SMP |
29d5e047 | 9800 | idle_thread_set_boot_cpu(); |
b5c44773 | 9801 | balance_push_set(smp_processor_id(), false); |
029632fb PZ |
9802 | #endif |
9803 | init_sched_fair_class(); | |
6a7b3dc3 | 9804 | |
eb414681 JW |
9805 | psi_init(); |
9806 | ||
69842cba PB |
9807 | init_uclamp(); |
9808 | ||
c597bfdd FW |
9809 | preempt_dynamic_init(); |
9810 | ||
6892b75e | 9811 | scheduler_running = 1; |
1da177e4 LT |
9812 | } |
9813 | ||
d902db1e | 9814 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 | 9815 | |
42a38756 | 9816 | void __might_sleep(const char *file, int line) |
1da177e4 | 9817 | { |
d6c23bb3 | 9818 | unsigned int state = get_current_state(); |
8eb23b9f PZ |
9819 | /* |
9820 | * Blocking primitives will set (and therefore destroy) current->state, | |
9821 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
9822 | * otherwise we will destroy state. | |
9823 | */ | |
d6c23bb3 | 9824 | WARN_ONCE(state != TASK_RUNNING && current->task_state_change, |
8eb23b9f | 9825 | "do not call blocking ops when !TASK_RUNNING; " |
d6c23bb3 | 9826 | "state=%x set at [<%p>] %pS\n", state, |
8eb23b9f | 9827 | (void *)current->task_state_change, |
00845eb9 | 9828 | (void *)current->task_state_change); |
8eb23b9f | 9829 | |
42a38756 | 9830 | __might_resched(file, line, 0); |
3427445a PZ |
9831 | } |
9832 | EXPORT_SYMBOL(__might_sleep); | |
9833 | ||
8d713b69 TG |
9834 | static void print_preempt_disable_ip(int preempt_offset, unsigned long ip) |
9835 | { | |
9836 | if (!IS_ENABLED(CONFIG_DEBUG_PREEMPT)) | |
9837 | return; | |
9838 | ||
9839 | if (preempt_count() == preempt_offset) | |
9840 | return; | |
9841 | ||
9842 | pr_err("Preemption disabled at:"); | |
9843 | print_ip_sym(KERN_ERR, ip); | |
9844 | } | |
9845 | ||
50e081b9 TG |
9846 | static inline bool resched_offsets_ok(unsigned int offsets) |
9847 | { | |
9848 | unsigned int nested = preempt_count(); | |
9849 | ||
9850 | nested += rcu_preempt_depth() << MIGHT_RESCHED_RCU_SHIFT; | |
9851 | ||
9852 | return nested == offsets; | |
9853 | } | |
9854 | ||
9855 | void __might_resched(const char *file, int line, unsigned int offsets) | |
1da177e4 | 9856 | { |
d1ccc66d IM |
9857 | /* Ratelimiting timestamp: */ |
9858 | static unsigned long prev_jiffy; | |
9859 | ||
d1c6d149 | 9860 | unsigned long preempt_disable_ip; |
1da177e4 | 9861 | |
d1ccc66d IM |
9862 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
9863 | rcu_sleep_check(); | |
9864 | ||
50e081b9 | 9865 | if ((resched_offsets_ok(offsets) && !irqs_disabled() && |
312364f3 | 9866 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
9867 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
9868 | oops_in_progress) | |
aef745fc | 9869 | return; |
1c3c5eab | 9870 | |
aef745fc IM |
9871 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
9872 | return; | |
9873 | prev_jiffy = jiffies; | |
9874 | ||
d1ccc66d | 9875 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
9876 | preempt_disable_ip = get_preempt_disable_ip(current); |
9877 | ||
a45ed302 TG |
9878 | pr_err("BUG: sleeping function called from invalid context at %s:%d\n", |
9879 | file, line); | |
9880 | pr_err("in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", | |
9881 | in_atomic(), irqs_disabled(), current->non_block_count, | |
9882 | current->pid, current->comm); | |
8d713b69 | 9883 | pr_err("preempt_count: %x, expected: %x\n", preempt_count(), |
50e081b9 | 9884 | offsets & MIGHT_RESCHED_PREEMPT_MASK); |
8d713b69 TG |
9885 | |
9886 | if (IS_ENABLED(CONFIG_PREEMPT_RCU)) { | |
50e081b9 TG |
9887 | pr_err("RCU nest depth: %d, expected: %u\n", |
9888 | rcu_preempt_depth(), offsets >> MIGHT_RESCHED_RCU_SHIFT); | |
8d713b69 | 9889 | } |
aef745fc | 9890 | |
a8b686b3 | 9891 | if (task_stack_end_corrupted(current)) |
a45ed302 | 9892 | pr_emerg("Thread overran stack, or stack corrupted\n"); |
a8b686b3 | 9893 | |
aef745fc IM |
9894 | debug_show_held_locks(current); |
9895 | if (irqs_disabled()) | |
9896 | print_irqtrace_events(current); | |
8d713b69 | 9897 | |
50e081b9 TG |
9898 | print_preempt_disable_ip(offsets & MIGHT_RESCHED_PREEMPT_MASK, |
9899 | preempt_disable_ip); | |
8d713b69 | 9900 | |
aef745fc | 9901 | dump_stack(); |
f0b22e39 | 9902 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 9903 | } |
874f670e | 9904 | EXPORT_SYMBOL(__might_resched); |
568f1967 PZ |
9905 | |
9906 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
9907 | { | |
9908 | static unsigned long prev_jiffy; | |
9909 | ||
9910 | if (irqs_disabled()) | |
9911 | return; | |
9912 | ||
9913 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
9914 | return; | |
9915 | ||
9916 | if (preempt_count() > preempt_offset) | |
9917 | return; | |
9918 | ||
9919 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9920 | return; | |
9921 | prev_jiffy = jiffies; | |
9922 | ||
9923 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
9924 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9925 | in_atomic(), irqs_disabled(), | |
9926 | current->pid, current->comm); | |
9927 | ||
9928 | debug_show_held_locks(current); | |
9929 | dump_stack(); | |
9930 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
9931 | } | |
9932 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
9933 | |
9934 | #ifdef CONFIG_SMP | |
9935 | void __cant_migrate(const char *file, int line) | |
9936 | { | |
9937 | static unsigned long prev_jiffy; | |
9938 | ||
9939 | if (irqs_disabled()) | |
9940 | return; | |
9941 | ||
9942 | if (is_migration_disabled(current)) | |
9943 | return; | |
9944 | ||
9945 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
9946 | return; | |
9947 | ||
9948 | if (preempt_count() > 0) | |
9949 | return; | |
9950 | ||
9951 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9952 | return; | |
9953 | prev_jiffy = jiffies; | |
9954 | ||
9955 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
9956 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
9957 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
9958 | current->pid, current->comm); | |
9959 | ||
9960 | debug_show_held_locks(current); | |
9961 | dump_stack(); | |
9962 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
9963 | } | |
9964 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
9965 | #endif | |
1da177e4 LT |
9966 | #endif |
9967 | ||
9968 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 9969 | void normalize_rt_tasks(void) |
3a5e4dc1 | 9970 | { |
dbc7f069 | 9971 | struct task_struct *g, *p; |
d50dde5a DF |
9972 | struct sched_attr attr = { |
9973 | .sched_policy = SCHED_NORMAL, | |
9974 | }; | |
1da177e4 | 9975 | |
3472eaa1 | 9976 | read_lock(&tasklist_lock); |
5d07f420 | 9977 | for_each_process_thread(g, p) { |
178be793 IM |
9978 | /* |
9979 | * Only normalize user tasks: | |
9980 | */ | |
3472eaa1 | 9981 | if (p->flags & PF_KTHREAD) |
178be793 IM |
9982 | continue; |
9983 | ||
4fa8d299 | 9984 | p->se.exec_start = 0; |
ceeadb83 YS |
9985 | schedstat_set(p->stats.wait_start, 0); |
9986 | schedstat_set(p->stats.sleep_start, 0); | |
9987 | schedstat_set(p->stats.block_start, 0); | |
dd41f596 | 9988 | |
aab03e05 | 9989 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
9990 | /* |
9991 | * Renice negative nice level userspace | |
9992 | * tasks back to 0: | |
9993 | */ | |
3472eaa1 | 9994 | if (task_nice(p) < 0) |
dd41f596 | 9995 | set_user_nice(p, 0); |
1da177e4 | 9996 | continue; |
dd41f596 | 9997 | } |
1da177e4 | 9998 | |
dbc7f069 | 9999 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 10000 | } |
3472eaa1 | 10001 | read_unlock(&tasklist_lock); |
1da177e4 LT |
10002 | } |
10003 | ||
10004 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 10005 | |
67fc4e0c | 10006 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 10007 | /* |
67fc4e0c | 10008 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
10009 | * |
10010 | * They can only be called when the whole system has been | |
10011 | * stopped - every CPU needs to be quiescent, and no scheduling | |
10012 | * activity can take place. Using them for anything else would | |
10013 | * be a serious bug, and as a result, they aren't even visible | |
10014 | * under any other configuration. | |
10015 | */ | |
10016 | ||
10017 | /** | |
d1ccc66d | 10018 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
10019 | * @cpu: the processor in question. |
10020 | * | |
10021 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
10022 | * |
10023 | * Return: The current task for @cpu. | |
1df5c10a | 10024 | */ |
36c8b586 | 10025 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
10026 | { |
10027 | return cpu_curr(cpu); | |
10028 | } | |
10029 | ||
67fc4e0c JW |
10030 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
10031 | ||
10032 | #ifdef CONFIG_IA64 | |
1df5c10a | 10033 | /** |
5feeb783 | 10034 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
10035 | * @cpu: the processor in question. |
10036 | * @p: the task pointer to set. | |
10037 | * | |
10038 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 10039 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 10040 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
10041 | * must be called with all CPU's synchronized, and interrupts disabled, the |
10042 | * and caller must save the original value of the current task (see | |
10043 | * curr_task() above) and restore that value before reenabling interrupts and | |
10044 | * re-starting the system. | |
10045 | * | |
10046 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
10047 | */ | |
a458ae2e | 10048 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
10049 | { |
10050 | cpu_curr(cpu) = p; | |
10051 | } | |
10052 | ||
10053 | #endif | |
29f59db3 | 10054 | |
7c941438 | 10055 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
10056 | /* task_group_lock serializes the addition/removal of task groups */ |
10057 | static DEFINE_SPINLOCK(task_group_lock); | |
10058 | ||
2480c093 PB |
10059 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
10060 | struct task_group *parent) | |
10061 | { | |
10062 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 10063 | enum uclamp_id clamp_id; |
2480c093 PB |
10064 | |
10065 | for_each_clamp_id(clamp_id) { | |
10066 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
10067 | uclamp_none(clamp_id), false); | |
0b60ba2d | 10068 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
10069 | } |
10070 | #endif | |
10071 | } | |
10072 | ||
2f5177f0 | 10073 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
10074 | { |
10075 | free_fair_sched_group(tg); | |
10076 | free_rt_sched_group(tg); | |
e9aa1dd1 | 10077 | autogroup_free(tg); |
b0367629 | 10078 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
10079 | } |
10080 | ||
b027789e MK |
10081 | static void sched_free_group_rcu(struct rcu_head *rcu) |
10082 | { | |
10083 | sched_free_group(container_of(rcu, struct task_group, rcu)); | |
10084 | } | |
10085 | ||
10086 | static void sched_unregister_group(struct task_group *tg) | |
10087 | { | |
10088 | unregister_fair_sched_group(tg); | |
10089 | unregister_rt_sched_group(tg); | |
10090 | /* | |
10091 | * We have to wait for yet another RCU grace period to expire, as | |
10092 | * print_cfs_stats() might run concurrently. | |
10093 | */ | |
10094 | call_rcu(&tg->rcu, sched_free_group_rcu); | |
10095 | } | |
10096 | ||
bccbe08a | 10097 | /* allocate runqueue etc for a new task group */ |
ec7dc8ac | 10098 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
10099 | { |
10100 | struct task_group *tg; | |
bccbe08a | 10101 | |
b0367629 | 10102 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
10103 | if (!tg) |
10104 | return ERR_PTR(-ENOMEM); | |
10105 | ||
ec7dc8ac | 10106 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
10107 | goto err; |
10108 | ||
ec7dc8ac | 10109 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
10110 | goto err; |
10111 | ||
2480c093 PB |
10112 | alloc_uclamp_sched_group(tg, parent); |
10113 | ||
ace783b9 LZ |
10114 | return tg; |
10115 | ||
10116 | err: | |
2f5177f0 | 10117 | sched_free_group(tg); |
ace783b9 LZ |
10118 | return ERR_PTR(-ENOMEM); |
10119 | } | |
10120 | ||
10121 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
10122 | { | |
10123 | unsigned long flags; | |
10124 | ||
8ed36996 | 10125 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 10126 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 10127 | |
d1ccc66d IM |
10128 | /* Root should already exist: */ |
10129 | WARN_ON(!parent); | |
f473aa5e PZ |
10130 | |
10131 | tg->parent = parent; | |
f473aa5e | 10132 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 10133 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 10134 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
10135 | |
10136 | online_fair_sched_group(tg); | |
29f59db3 SV |
10137 | } |
10138 | ||
9b5b7751 | 10139 | /* rcu callback to free various structures associated with a task group */ |
b027789e | 10140 | static void sched_unregister_group_rcu(struct rcu_head *rhp) |
29f59db3 | 10141 | { |
d1ccc66d | 10142 | /* Now it should be safe to free those cfs_rqs: */ |
b027789e | 10143 | sched_unregister_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
10144 | } |
10145 | ||
4cf86d77 | 10146 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 10147 | { |
d1ccc66d | 10148 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
b027789e | 10149 | call_rcu(&tg->rcu, sched_unregister_group_rcu); |
ace783b9 LZ |
10150 | } |
10151 | ||
b027789e | 10152 | void sched_release_group(struct task_group *tg) |
29f59db3 | 10153 | { |
8ed36996 | 10154 | unsigned long flags; |
29f59db3 | 10155 | |
b027789e MK |
10156 | /* |
10157 | * Unlink first, to avoid walk_tg_tree_from() from finding us (via | |
10158 | * sched_cfs_period_timer()). | |
10159 | * | |
10160 | * For this to be effective, we have to wait for all pending users of | |
10161 | * this task group to leave their RCU critical section to ensure no new | |
10162 | * user will see our dying task group any more. Specifically ensure | |
10163 | * that tg_unthrottle_up() won't add decayed cfs_rq's to it. | |
10164 | * | |
10165 | * We therefore defer calling unregister_fair_sched_group() to | |
10166 | * sched_unregister_group() which is guarantied to get called only after the | |
10167 | * current RCU grace period has expired. | |
10168 | */ | |
3d4b47b4 | 10169 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 10170 | list_del_rcu(&tg->list); |
f473aa5e | 10171 | list_del_rcu(&tg->siblings); |
8ed36996 | 10172 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
10173 | } |
10174 | ||
39c42611 | 10175 | static void sched_change_group(struct task_struct *tsk) |
29f59db3 | 10176 | { |
8323f26c | 10177 | struct task_group *tg; |
29f59db3 | 10178 | |
f7b8a47d KT |
10179 | /* |
10180 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
10181 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
10182 | * to prevent lockdep warnings. | |
10183 | */ | |
10184 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
10185 | struct task_group, css); |
10186 | tg = autogroup_task_group(tsk, tg); | |
10187 | tsk->sched_task_group = tg; | |
10188 | ||
810b3817 | 10189 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b | 10190 | if (tsk->sched_class->task_change_group) |
39c42611 | 10191 | tsk->sched_class->task_change_group(tsk); |
b2b5ce02 | 10192 | else |
810b3817 | 10193 | #endif |
b2b5ce02 | 10194 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
10195 | } |
10196 | ||
10197 | /* | |
10198 | * Change task's runqueue when it moves between groups. | |
10199 | * | |
10200 | * The caller of this function should have put the task in its new group by | |
10201 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
10202 | * its new group. | |
10203 | */ | |
10204 | void sched_move_task(struct task_struct *tsk) | |
10205 | { | |
7a57f32a PZ |
10206 | int queued, running, queue_flags = |
10207 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
10208 | struct rq_flags rf; |
10209 | struct rq *rq; | |
10210 | ||
10211 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 10212 | update_rq_clock(rq); |
ea86cb4b VG |
10213 | |
10214 | running = task_current(rq, tsk); | |
10215 | queued = task_on_rq_queued(tsk); | |
10216 | ||
10217 | if (queued) | |
7a57f32a | 10218 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 10219 | if (running) |
ea86cb4b VG |
10220 | put_prev_task(rq, tsk); |
10221 | ||
39c42611 | 10222 | sched_change_group(tsk); |
810b3817 | 10223 | |
da0c1e65 | 10224 | if (queued) |
7a57f32a | 10225 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 10226 | if (running) { |
03b7fad1 | 10227 | set_next_task(rq, tsk); |
2a4b03ff VG |
10228 | /* |
10229 | * After changing group, the running task may have joined a | |
10230 | * throttled one but it's still the running task. Trigger a | |
10231 | * resched to make sure that task can still run. | |
10232 | */ | |
10233 | resched_curr(rq); | |
10234 | } | |
29f59db3 | 10235 | |
eb580751 | 10236 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 10237 | } |
68318b8e | 10238 | |
a7c6d554 | 10239 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 10240 | { |
a7c6d554 | 10241 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
10242 | } |
10243 | ||
eb95419b TH |
10244 | static struct cgroup_subsys_state * |
10245 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 10246 | { |
eb95419b TH |
10247 | struct task_group *parent = css_tg(parent_css); |
10248 | struct task_group *tg; | |
68318b8e | 10249 | |
eb95419b | 10250 | if (!parent) { |
68318b8e | 10251 | /* This is early initialization for the top cgroup */ |
07e06b01 | 10252 | return &root_task_group.css; |
68318b8e SV |
10253 | } |
10254 | ||
ec7dc8ac | 10255 | tg = sched_create_group(parent); |
68318b8e SV |
10256 | if (IS_ERR(tg)) |
10257 | return ERR_PTR(-ENOMEM); | |
10258 | ||
68318b8e SV |
10259 | return &tg->css; |
10260 | } | |
10261 | ||
96b77745 KK |
10262 | /* Expose task group only after completing cgroup initialization */ |
10263 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
10264 | { | |
10265 | struct task_group *tg = css_tg(css); | |
10266 | struct task_group *parent = css_tg(css->parent); | |
10267 | ||
10268 | if (parent) | |
10269 | sched_online_group(tg, parent); | |
7226017a QY |
10270 | |
10271 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
10272 | /* Propagate the effective uclamp value for the new group */ | |
93b73858 QY |
10273 | mutex_lock(&uclamp_mutex); |
10274 | rcu_read_lock(); | |
7226017a | 10275 | cpu_util_update_eff(css); |
93b73858 QY |
10276 | rcu_read_unlock(); |
10277 | mutex_unlock(&uclamp_mutex); | |
7226017a QY |
10278 | #endif |
10279 | ||
96b77745 KK |
10280 | return 0; |
10281 | } | |
10282 | ||
2f5177f0 | 10283 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 10284 | { |
eb95419b | 10285 | struct task_group *tg = css_tg(css); |
ace783b9 | 10286 | |
b027789e | 10287 | sched_release_group(tg); |
ace783b9 LZ |
10288 | } |
10289 | ||
eb95419b | 10290 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 10291 | { |
eb95419b | 10292 | struct task_group *tg = css_tg(css); |
68318b8e | 10293 | |
2f5177f0 PZ |
10294 | /* |
10295 | * Relies on the RCU grace period between css_released() and this. | |
10296 | */ | |
b027789e | 10297 | sched_unregister_group(tg); |
ace783b9 LZ |
10298 | } |
10299 | ||
df16b71c | 10300 | #ifdef CONFIG_RT_GROUP_SCHED |
1f7dd3e5 | 10301 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 10302 | { |
bb9d97b6 | 10303 | struct task_struct *task; |
1f7dd3e5 | 10304 | struct cgroup_subsys_state *css; |
bb9d97b6 | 10305 | |
1f7dd3e5 | 10306 | cgroup_taskset_for_each(task, css, tset) { |
eb95419b | 10307 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 10308 | return -EINVAL; |
bb9d97b6 | 10309 | } |
df16b71c | 10310 | return 0; |
be367d09 | 10311 | } |
df16b71c | 10312 | #endif |
68318b8e | 10313 | |
1f7dd3e5 | 10314 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 10315 | { |
bb9d97b6 | 10316 | struct task_struct *task; |
1f7dd3e5 | 10317 | struct cgroup_subsys_state *css; |
bb9d97b6 | 10318 | |
1f7dd3e5 | 10319 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 10320 | sched_move_task(task); |
68318b8e SV |
10321 | } |
10322 | ||
2480c093 | 10323 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
10324 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
10325 | { | |
10326 | struct cgroup_subsys_state *top_css = css; | |
10327 | struct uclamp_se *uc_parent = NULL; | |
10328 | struct uclamp_se *uc_se = NULL; | |
10329 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 10330 | enum uclamp_id clamp_id; |
0b60ba2d PB |
10331 | unsigned int clamps; |
10332 | ||
93b73858 QY |
10333 | lockdep_assert_held(&uclamp_mutex); |
10334 | SCHED_WARN_ON(!rcu_read_lock_held()); | |
10335 | ||
0b60ba2d PB |
10336 | css_for_each_descendant_pre(css, top_css) { |
10337 | uc_parent = css_tg(css)->parent | |
10338 | ? css_tg(css)->parent->uclamp : NULL; | |
10339 | ||
10340 | for_each_clamp_id(clamp_id) { | |
10341 | /* Assume effective clamps matches requested clamps */ | |
10342 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
10343 | /* Cap effective clamps with parent's effective clamps */ | |
10344 | if (uc_parent && | |
10345 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
10346 | eff[clamp_id] = uc_parent[clamp_id].value; | |
10347 | } | |
10348 | } | |
10349 | /* Ensure protection is always capped by limit */ | |
10350 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
10351 | ||
10352 | /* Propagate most restrictive effective clamps */ | |
10353 | clamps = 0x0; | |
10354 | uc_se = css_tg(css)->uclamp; | |
10355 | for_each_clamp_id(clamp_id) { | |
10356 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
10357 | continue; | |
10358 | uc_se[clamp_id].value = eff[clamp_id]; | |
10359 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
10360 | clamps |= (0x1 << clamp_id); | |
10361 | } | |
babbe170 | 10362 | if (!clamps) { |
0b60ba2d | 10363 | css = css_rightmost_descendant(css); |
babbe170 PB |
10364 | continue; |
10365 | } | |
10366 | ||
10367 | /* Immediately update descendants RUNNABLE tasks */ | |
0213b708 | 10368 | uclamp_update_active_tasks(css); |
0b60ba2d PB |
10369 | } |
10370 | } | |
2480c093 PB |
10371 | |
10372 | /* | |
10373 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
10374 | * C expression. Since there is no way to convert a macro argument (N) into a | |
10375 | * character constant, use two levels of macros. | |
10376 | */ | |
10377 | #define _POW10(exp) ((unsigned int)1e##exp) | |
10378 | #define POW10(exp) _POW10(exp) | |
10379 | ||
10380 | struct uclamp_request { | |
10381 | #define UCLAMP_PERCENT_SHIFT 2 | |
10382 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
10383 | s64 percent; | |
10384 | u64 util; | |
10385 | int ret; | |
10386 | }; | |
10387 | ||
10388 | static inline struct uclamp_request | |
10389 | capacity_from_percent(char *buf) | |
10390 | { | |
10391 | struct uclamp_request req = { | |
10392 | .percent = UCLAMP_PERCENT_SCALE, | |
10393 | .util = SCHED_CAPACITY_SCALE, | |
10394 | .ret = 0, | |
10395 | }; | |
10396 | ||
10397 | buf = strim(buf); | |
10398 | if (strcmp(buf, "max")) { | |
10399 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
10400 | &req.percent); | |
10401 | if (req.ret) | |
10402 | return req; | |
b562d140 | 10403 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
10404 | req.ret = -ERANGE; |
10405 | return req; | |
10406 | } | |
10407 | ||
10408 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
10409 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
10410 | } | |
10411 | ||
10412 | return req; | |
10413 | } | |
10414 | ||
10415 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
10416 | size_t nbytes, loff_t off, | |
10417 | enum uclamp_id clamp_id) | |
10418 | { | |
10419 | struct uclamp_request req; | |
10420 | struct task_group *tg; | |
10421 | ||
10422 | req = capacity_from_percent(buf); | |
10423 | if (req.ret) | |
10424 | return req.ret; | |
10425 | ||
46609ce2 QY |
10426 | static_branch_enable(&sched_uclamp_used); |
10427 | ||
2480c093 PB |
10428 | mutex_lock(&uclamp_mutex); |
10429 | rcu_read_lock(); | |
10430 | ||
10431 | tg = css_tg(of_css(of)); | |
10432 | if (tg->uclamp_req[clamp_id].value != req.util) | |
10433 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
10434 | ||
10435 | /* | |
10436 | * Because of not recoverable conversion rounding we keep track of the | |
10437 | * exact requested value | |
10438 | */ | |
10439 | tg->uclamp_pct[clamp_id] = req.percent; | |
10440 | ||
0b60ba2d PB |
10441 | /* Update effective clamps to track the most restrictive value */ |
10442 | cpu_util_update_eff(of_css(of)); | |
10443 | ||
2480c093 PB |
10444 | rcu_read_unlock(); |
10445 | mutex_unlock(&uclamp_mutex); | |
10446 | ||
10447 | return nbytes; | |
10448 | } | |
10449 | ||
10450 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
10451 | char *buf, size_t nbytes, | |
10452 | loff_t off) | |
10453 | { | |
10454 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
10455 | } | |
10456 | ||
10457 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
10458 | char *buf, size_t nbytes, | |
10459 | loff_t off) | |
10460 | { | |
10461 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
10462 | } | |
10463 | ||
10464 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
10465 | enum uclamp_id clamp_id) | |
10466 | { | |
10467 | struct task_group *tg; | |
10468 | u64 util_clamp; | |
10469 | u64 percent; | |
10470 | u32 rem; | |
10471 | ||
10472 | rcu_read_lock(); | |
10473 | tg = css_tg(seq_css(sf)); | |
10474 | util_clamp = tg->uclamp_req[clamp_id].value; | |
10475 | rcu_read_unlock(); | |
10476 | ||
10477 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
10478 | seq_puts(sf, "max\n"); | |
10479 | return; | |
10480 | } | |
10481 | ||
10482 | percent = tg->uclamp_pct[clamp_id]; | |
10483 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
10484 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
10485 | } | |
10486 | ||
10487 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
10488 | { | |
10489 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
10490 | return 0; | |
10491 | } | |
10492 | ||
10493 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
10494 | { | |
10495 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
10496 | return 0; | |
10497 | } | |
10498 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
10499 | ||
052f1dc7 | 10500 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
10501 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
10502 | struct cftype *cftype, u64 shareval) | |
68318b8e | 10503 | { |
5b61d50a KK |
10504 | if (shareval > scale_load_down(ULONG_MAX)) |
10505 | shareval = MAX_SHARES; | |
182446d0 | 10506 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
10507 | } |
10508 | ||
182446d0 TH |
10509 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
10510 | struct cftype *cft) | |
68318b8e | 10511 | { |
182446d0 | 10512 | struct task_group *tg = css_tg(css); |
68318b8e | 10513 | |
c8b28116 | 10514 | return (u64) scale_load_down(tg->shares); |
68318b8e | 10515 | } |
ab84d31e PT |
10516 | |
10517 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
10518 | static DEFINE_MUTEX(cfs_constraints_mutex); |
10519 | ||
ab84d31e | 10520 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 10521 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
10522 | /* More than 203 days if BW_SHIFT equals 20. */ |
10523 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 10524 | |
a790de99 PT |
10525 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
10526 | ||
f4183717 HC |
10527 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota, |
10528 | u64 burst) | |
ab84d31e | 10529 | { |
56f570e5 | 10530 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 10531 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
10532 | |
10533 | if (tg == &root_task_group) | |
10534 | return -EINVAL; | |
10535 | ||
10536 | /* | |
10537 | * Ensure we have at some amount of bandwidth every period. This is | |
10538 | * to prevent reaching a state of large arrears when throttled via | |
10539 | * entity_tick() resulting in prolonged exit starvation. | |
10540 | */ | |
10541 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
10542 | return -EINVAL; | |
10543 | ||
10544 | /* | |
3b03706f | 10545 | * Likewise, bound things on the other side by preventing insane quota |
ab84d31e PT |
10546 | * periods. This also allows us to normalize in computing quota |
10547 | * feasibility. | |
10548 | */ | |
10549 | if (period > max_cfs_quota_period) | |
10550 | return -EINVAL; | |
10551 | ||
d505b8af HC |
10552 | /* |
10553 | * Bound quota to defend quota against overflow during bandwidth shift. | |
10554 | */ | |
10555 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
10556 | return -EINVAL; | |
10557 | ||
f4183717 HC |
10558 | if (quota != RUNTIME_INF && (burst > quota || |
10559 | burst + quota > max_cfs_runtime)) | |
10560 | return -EINVAL; | |
10561 | ||
0e59bdae KT |
10562 | /* |
10563 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
10564 | * unthrottle_offline_cfs_rqs(). | |
10565 | */ | |
746f5ea9 | 10566 | cpus_read_lock(); |
a790de99 PT |
10567 | mutex_lock(&cfs_constraints_mutex); |
10568 | ret = __cfs_schedulable(tg, period, quota); | |
10569 | if (ret) | |
10570 | goto out_unlock; | |
10571 | ||
58088ad0 | 10572 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 10573 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
10574 | /* |
10575 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
10576 | * before making related changes, and on->off must occur afterwards | |
10577 | */ | |
10578 | if (runtime_enabled && !runtime_was_enabled) | |
10579 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
10580 | raw_spin_lock_irq(&cfs_b->lock); |
10581 | cfs_b->period = ns_to_ktime(period); | |
10582 | cfs_b->quota = quota; | |
f4183717 | 10583 | cfs_b->burst = burst; |
58088ad0 | 10584 | |
a9cf55b2 | 10585 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
10586 | |
10587 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
10588 | if (runtime_enabled) |
10589 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 10590 | |
ab84d31e PT |
10591 | raw_spin_unlock_irq(&cfs_b->lock); |
10592 | ||
0e59bdae | 10593 | for_each_online_cpu(i) { |
ab84d31e | 10594 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 10595 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 10596 | struct rq_flags rf; |
ab84d31e | 10597 | |
8a8c69c3 | 10598 | rq_lock_irq(rq, &rf); |
58088ad0 | 10599 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 10600 | cfs_rq->runtime_remaining = 0; |
671fd9da | 10601 | |
029632fb | 10602 | if (cfs_rq->throttled) |
671fd9da | 10603 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 10604 | rq_unlock_irq(rq, &rf); |
ab84d31e | 10605 | } |
1ee14e6c BS |
10606 | if (runtime_was_enabled && !runtime_enabled) |
10607 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
10608 | out_unlock: |
10609 | mutex_unlock(&cfs_constraints_mutex); | |
746f5ea9 | 10610 | cpus_read_unlock(); |
ab84d31e | 10611 | |
a790de99 | 10612 | return ret; |
ab84d31e PT |
10613 | } |
10614 | ||
b1546edc | 10615 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e | 10616 | { |
f4183717 | 10617 | u64 quota, period, burst; |
ab84d31e | 10618 | |
029632fb | 10619 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
f4183717 | 10620 | burst = tg->cfs_bandwidth.burst; |
ab84d31e PT |
10621 | if (cfs_quota_us < 0) |
10622 | quota = RUNTIME_INF; | |
1a8b4540 | 10623 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 10624 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
10625 | else |
10626 | return -EINVAL; | |
ab84d31e | 10627 | |
f4183717 | 10628 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10629 | } |
10630 | ||
b1546edc | 10631 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
10632 | { |
10633 | u64 quota_us; | |
10634 | ||
029632fb | 10635 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
10636 | return -1; |
10637 | ||
029632fb | 10638 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
10639 | do_div(quota_us, NSEC_PER_USEC); |
10640 | ||
10641 | return quota_us; | |
10642 | } | |
10643 | ||
b1546edc | 10644 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e | 10645 | { |
f4183717 | 10646 | u64 quota, period, burst; |
ab84d31e | 10647 | |
1a8b4540 KK |
10648 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
10649 | return -EINVAL; | |
10650 | ||
ab84d31e | 10651 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 10652 | quota = tg->cfs_bandwidth.quota; |
f4183717 | 10653 | burst = tg->cfs_bandwidth.burst; |
ab84d31e | 10654 | |
f4183717 | 10655 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10656 | } |
10657 | ||
b1546edc | 10658 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
10659 | { |
10660 | u64 cfs_period_us; | |
10661 | ||
029632fb | 10662 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
10663 | do_div(cfs_period_us, NSEC_PER_USEC); |
10664 | ||
10665 | return cfs_period_us; | |
10666 | } | |
10667 | ||
f4183717 HC |
10668 | static int tg_set_cfs_burst(struct task_group *tg, long cfs_burst_us) |
10669 | { | |
10670 | u64 quota, period, burst; | |
10671 | ||
10672 | if ((u64)cfs_burst_us > U64_MAX / NSEC_PER_USEC) | |
10673 | return -EINVAL; | |
10674 | ||
10675 | burst = (u64)cfs_burst_us * NSEC_PER_USEC; | |
10676 | period = ktime_to_ns(tg->cfs_bandwidth.period); | |
10677 | quota = tg->cfs_bandwidth.quota; | |
10678 | ||
10679 | return tg_set_cfs_bandwidth(tg, period, quota, burst); | |
10680 | } | |
10681 | ||
10682 | static long tg_get_cfs_burst(struct task_group *tg) | |
10683 | { | |
10684 | u64 burst_us; | |
10685 | ||
10686 | burst_us = tg->cfs_bandwidth.burst; | |
10687 | do_div(burst_us, NSEC_PER_USEC); | |
10688 | ||
10689 | return burst_us; | |
10690 | } | |
10691 | ||
182446d0 TH |
10692 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
10693 | struct cftype *cft) | |
ab84d31e | 10694 | { |
182446d0 | 10695 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
10696 | } |
10697 | ||
182446d0 TH |
10698 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
10699 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 10700 | { |
182446d0 | 10701 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
10702 | } |
10703 | ||
182446d0 TH |
10704 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
10705 | struct cftype *cft) | |
ab84d31e | 10706 | { |
182446d0 | 10707 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
10708 | } |
10709 | ||
182446d0 TH |
10710 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
10711 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 10712 | { |
182446d0 | 10713 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
10714 | } |
10715 | ||
f4183717 HC |
10716 | static u64 cpu_cfs_burst_read_u64(struct cgroup_subsys_state *css, |
10717 | struct cftype *cft) | |
10718 | { | |
10719 | return tg_get_cfs_burst(css_tg(css)); | |
10720 | } | |
10721 | ||
10722 | static int cpu_cfs_burst_write_u64(struct cgroup_subsys_state *css, | |
10723 | struct cftype *cftype, u64 cfs_burst_us) | |
10724 | { | |
10725 | return tg_set_cfs_burst(css_tg(css), cfs_burst_us); | |
10726 | } | |
10727 | ||
a790de99 PT |
10728 | struct cfs_schedulable_data { |
10729 | struct task_group *tg; | |
10730 | u64 period, quota; | |
10731 | }; | |
10732 | ||
10733 | /* | |
10734 | * normalize group quota/period to be quota/max_period | |
10735 | * note: units are usecs | |
10736 | */ | |
10737 | static u64 normalize_cfs_quota(struct task_group *tg, | |
10738 | struct cfs_schedulable_data *d) | |
10739 | { | |
10740 | u64 quota, period; | |
10741 | ||
10742 | if (tg == d->tg) { | |
10743 | period = d->period; | |
10744 | quota = d->quota; | |
10745 | } else { | |
10746 | period = tg_get_cfs_period(tg); | |
10747 | quota = tg_get_cfs_quota(tg); | |
10748 | } | |
10749 | ||
10750 | /* note: these should typically be equivalent */ | |
10751 | if (quota == RUNTIME_INF || quota == -1) | |
10752 | return RUNTIME_INF; | |
10753 | ||
10754 | return to_ratio(period, quota); | |
10755 | } | |
10756 | ||
10757 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
10758 | { | |
10759 | struct cfs_schedulable_data *d = data; | |
029632fb | 10760 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
10761 | s64 quota = 0, parent_quota = -1; |
10762 | ||
10763 | if (!tg->parent) { | |
10764 | quota = RUNTIME_INF; | |
10765 | } else { | |
029632fb | 10766 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
10767 | |
10768 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 10769 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
10770 | |
10771 | /* | |
c53593e5 TH |
10772 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
10773 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 10774 | * limit is set: |
a790de99 | 10775 | */ |
c53593e5 TH |
10776 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
10777 | quota = min(quota, parent_quota); | |
10778 | } else { | |
10779 | if (quota == RUNTIME_INF) | |
10780 | quota = parent_quota; | |
10781 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
10782 | return -EINVAL; | |
10783 | } | |
a790de99 | 10784 | } |
9c58c79a | 10785 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
10786 | |
10787 | return 0; | |
10788 | } | |
10789 | ||
10790 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
10791 | { | |
8277434e | 10792 | int ret; |
a790de99 PT |
10793 | struct cfs_schedulable_data data = { |
10794 | .tg = tg, | |
10795 | .period = period, | |
10796 | .quota = quota, | |
10797 | }; | |
10798 | ||
10799 | if (quota != RUNTIME_INF) { | |
10800 | do_div(data.period, NSEC_PER_USEC); | |
10801 | do_div(data.quota, NSEC_PER_USEC); | |
10802 | } | |
10803 | ||
8277434e PT |
10804 | rcu_read_lock(); |
10805 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
10806 | rcu_read_unlock(); | |
10807 | ||
10808 | return ret; | |
a790de99 | 10809 | } |
e8da1b18 | 10810 | |
a1f7164c | 10811 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 10812 | { |
2da8ca82 | 10813 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 10814 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 10815 | |
44ffc75b TH |
10816 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
10817 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
10818 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 10819 | |
3d6c50c2 | 10820 | if (schedstat_enabled() && tg != &root_task_group) { |
ceeadb83 | 10821 | struct sched_statistics *stats; |
3d6c50c2 YW |
10822 | u64 ws = 0; |
10823 | int i; | |
10824 | ||
ceeadb83 YS |
10825 | for_each_possible_cpu(i) { |
10826 | stats = __schedstats_from_se(tg->se[i]); | |
10827 | ws += schedstat_val(stats->wait_sum); | |
10828 | } | |
3d6c50c2 YW |
10829 | |
10830 | seq_printf(sf, "wait_sum %llu\n", ws); | |
10831 | } | |
10832 | ||
bcb1704a HC |
10833 | seq_printf(sf, "nr_bursts %d\n", cfs_b->nr_burst); |
10834 | seq_printf(sf, "burst_time %llu\n", cfs_b->burst_time); | |
10835 | ||
e8da1b18 NR |
10836 | return 0; |
10837 | } | |
ab84d31e | 10838 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 10839 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10840 | |
052f1dc7 | 10841 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
10842 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
10843 | struct cftype *cft, s64 val) | |
6f505b16 | 10844 | { |
182446d0 | 10845 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
10846 | } |
10847 | ||
182446d0 TH |
10848 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
10849 | struct cftype *cft) | |
6f505b16 | 10850 | { |
182446d0 | 10851 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 10852 | } |
d0b27fa7 | 10853 | |
182446d0 TH |
10854 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
10855 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 10856 | { |
182446d0 | 10857 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
10858 | } |
10859 | ||
182446d0 TH |
10860 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
10861 | struct cftype *cft) | |
d0b27fa7 | 10862 | { |
182446d0 | 10863 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 10864 | } |
6d6bc0ad | 10865 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10866 | |
30400039 JD |
10867 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10868 | static s64 cpu_idle_read_s64(struct cgroup_subsys_state *css, | |
10869 | struct cftype *cft) | |
10870 | { | |
10871 | return css_tg(css)->idle; | |
10872 | } | |
10873 | ||
10874 | static int cpu_idle_write_s64(struct cgroup_subsys_state *css, | |
10875 | struct cftype *cft, s64 idle) | |
10876 | { | |
10877 | return sched_group_set_idle(css_tg(css), idle); | |
10878 | } | |
10879 | #endif | |
10880 | ||
a1f7164c | 10881 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 10882 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10883 | { |
10884 | .name = "shares", | |
f4c753b7 PM |
10885 | .read_u64 = cpu_shares_read_u64, |
10886 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10887 | }, |
30400039 JD |
10888 | { |
10889 | .name = "idle", | |
10890 | .read_s64 = cpu_idle_read_s64, | |
10891 | .write_s64 = cpu_idle_write_s64, | |
10892 | }, | |
052f1dc7 | 10893 | #endif |
ab84d31e PT |
10894 | #ifdef CONFIG_CFS_BANDWIDTH |
10895 | { | |
10896 | .name = "cfs_quota_us", | |
10897 | .read_s64 = cpu_cfs_quota_read_s64, | |
10898 | .write_s64 = cpu_cfs_quota_write_s64, | |
10899 | }, | |
10900 | { | |
10901 | .name = "cfs_period_us", | |
10902 | .read_u64 = cpu_cfs_period_read_u64, | |
10903 | .write_u64 = cpu_cfs_period_write_u64, | |
10904 | }, | |
f4183717 HC |
10905 | { |
10906 | .name = "cfs_burst_us", | |
10907 | .read_u64 = cpu_cfs_burst_read_u64, | |
10908 | .write_u64 = cpu_cfs_burst_write_u64, | |
10909 | }, | |
e8da1b18 NR |
10910 | { |
10911 | .name = "stat", | |
a1f7164c | 10912 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 10913 | }, |
ab84d31e | 10914 | #endif |
052f1dc7 | 10915 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10916 | { |
9f0c1e56 | 10917 | .name = "rt_runtime_us", |
06ecb27c PM |
10918 | .read_s64 = cpu_rt_runtime_read, |
10919 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10920 | }, |
d0b27fa7 PZ |
10921 | { |
10922 | .name = "rt_period_us", | |
f4c753b7 PM |
10923 | .read_u64 = cpu_rt_period_read_uint, |
10924 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10925 | }, |
2480c093 PB |
10926 | #endif |
10927 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
10928 | { | |
10929 | .name = "uclamp.min", | |
10930 | .flags = CFTYPE_NOT_ON_ROOT, | |
10931 | .seq_show = cpu_uclamp_min_show, | |
10932 | .write = cpu_uclamp_min_write, | |
10933 | }, | |
10934 | { | |
10935 | .name = "uclamp.max", | |
10936 | .flags = CFTYPE_NOT_ON_ROOT, | |
10937 | .seq_show = cpu_uclamp_max_show, | |
10938 | .write = cpu_uclamp_max_write, | |
10939 | }, | |
052f1dc7 | 10940 | #endif |
d1ccc66d | 10941 | { } /* Terminate */ |
68318b8e SV |
10942 | }; |
10943 | ||
d41bf8c9 TH |
10944 | static int cpu_extra_stat_show(struct seq_file *sf, |
10945 | struct cgroup_subsys_state *css) | |
0d593634 | 10946 | { |
0d593634 TH |
10947 | #ifdef CONFIG_CFS_BANDWIDTH |
10948 | { | |
d41bf8c9 | 10949 | struct task_group *tg = css_tg(css); |
0d593634 | 10950 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
bcb1704a | 10951 | u64 throttled_usec, burst_usec; |
0d593634 TH |
10952 | |
10953 | throttled_usec = cfs_b->throttled_time; | |
10954 | do_div(throttled_usec, NSEC_PER_USEC); | |
bcb1704a HC |
10955 | burst_usec = cfs_b->burst_time; |
10956 | do_div(burst_usec, NSEC_PER_USEC); | |
0d593634 TH |
10957 | |
10958 | seq_printf(sf, "nr_periods %d\n" | |
10959 | "nr_throttled %d\n" | |
bcb1704a HC |
10960 | "throttled_usec %llu\n" |
10961 | "nr_bursts %d\n" | |
10962 | "burst_usec %llu\n", | |
0d593634 | 10963 | cfs_b->nr_periods, cfs_b->nr_throttled, |
bcb1704a | 10964 | throttled_usec, cfs_b->nr_burst, burst_usec); |
0d593634 TH |
10965 | } |
10966 | #endif | |
10967 | return 0; | |
10968 | } | |
10969 | ||
10970 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
10971 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
10972 | struct cftype *cft) | |
10973 | { | |
10974 | struct task_group *tg = css_tg(css); | |
10975 | u64 weight = scale_load_down(tg->shares); | |
10976 | ||
10977 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
10978 | } | |
10979 | ||
10980 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
10981 | struct cftype *cft, u64 weight) | |
10982 | { | |
10983 | /* | |
10984 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
10985 | * values which are 1, 100 and 10000 respectively. While it loses | |
10986 | * a bit of range on both ends, it maps pretty well onto the shares | |
10987 | * value used by scheduler and the round-trip conversions preserve | |
10988 | * the original value over the entire range. | |
10989 | */ | |
10990 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
10991 | return -ERANGE; | |
10992 | ||
10993 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
10994 | ||
10995 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
10996 | } | |
10997 | ||
10998 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
10999 | struct cftype *cft) | |
11000 | { | |
11001 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
11002 | int last_delta = INT_MAX; | |
11003 | int prio, delta; | |
11004 | ||
11005 | /* find the closest nice value to the current weight */ | |
11006 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
11007 | delta = abs(sched_prio_to_weight[prio] - weight); | |
11008 | if (delta >= last_delta) | |
11009 | break; | |
11010 | last_delta = delta; | |
11011 | } | |
11012 | ||
11013 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
11014 | } | |
11015 | ||
11016 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
11017 | struct cftype *cft, s64 nice) | |
11018 | { | |
11019 | unsigned long weight; | |
7281c8de | 11020 | int idx; |
0d593634 TH |
11021 | |
11022 | if (nice < MIN_NICE || nice > MAX_NICE) | |
11023 | return -ERANGE; | |
11024 | ||
7281c8de PZ |
11025 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
11026 | idx = array_index_nospec(idx, 40); | |
11027 | weight = sched_prio_to_weight[idx]; | |
11028 | ||
0d593634 TH |
11029 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
11030 | } | |
11031 | #endif | |
11032 | ||
11033 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
11034 | long period, long quota) | |
11035 | { | |
11036 | if (quota < 0) | |
11037 | seq_puts(sf, "max"); | |
11038 | else | |
11039 | seq_printf(sf, "%ld", quota); | |
11040 | ||
11041 | seq_printf(sf, " %ld\n", period); | |
11042 | } | |
11043 | ||
11044 | /* caller should put the current value in *@periodp before calling */ | |
11045 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
11046 | u64 *periodp, u64 *quotap) | |
11047 | { | |
11048 | char tok[21]; /* U64_MAX */ | |
11049 | ||
4c47acd8 | 11050 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
11051 | return -EINVAL; |
11052 | ||
11053 | *periodp *= NSEC_PER_USEC; | |
11054 | ||
11055 | if (sscanf(tok, "%llu", quotap)) | |
11056 | *quotap *= NSEC_PER_USEC; | |
11057 | else if (!strcmp(tok, "max")) | |
11058 | *quotap = RUNTIME_INF; | |
11059 | else | |
11060 | return -EINVAL; | |
11061 | ||
11062 | return 0; | |
11063 | } | |
11064 | ||
11065 | #ifdef CONFIG_CFS_BANDWIDTH | |
11066 | static int cpu_max_show(struct seq_file *sf, void *v) | |
11067 | { | |
11068 | struct task_group *tg = css_tg(seq_css(sf)); | |
11069 | ||
11070 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
11071 | return 0; | |
11072 | } | |
11073 | ||
11074 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
11075 | char *buf, size_t nbytes, loff_t off) | |
11076 | { | |
11077 | struct task_group *tg = css_tg(of_css(of)); | |
11078 | u64 period = tg_get_cfs_period(tg); | |
f4183717 | 11079 | u64 burst = tg_get_cfs_burst(tg); |
0d593634 TH |
11080 | u64 quota; |
11081 | int ret; | |
11082 | ||
11083 | ret = cpu_period_quota_parse(buf, &period, "a); | |
11084 | if (!ret) | |
f4183717 | 11085 | ret = tg_set_cfs_bandwidth(tg, period, quota, burst); |
0d593634 TH |
11086 | return ret ?: nbytes; |
11087 | } | |
11088 | #endif | |
11089 | ||
11090 | static struct cftype cpu_files[] = { | |
0d593634 TH |
11091 | #ifdef CONFIG_FAIR_GROUP_SCHED |
11092 | { | |
11093 | .name = "weight", | |
11094 | .flags = CFTYPE_NOT_ON_ROOT, | |
11095 | .read_u64 = cpu_weight_read_u64, | |
11096 | .write_u64 = cpu_weight_write_u64, | |
11097 | }, | |
11098 | { | |
11099 | .name = "weight.nice", | |
11100 | .flags = CFTYPE_NOT_ON_ROOT, | |
11101 | .read_s64 = cpu_weight_nice_read_s64, | |
11102 | .write_s64 = cpu_weight_nice_write_s64, | |
11103 | }, | |
30400039 JD |
11104 | { |
11105 | .name = "idle", | |
11106 | .flags = CFTYPE_NOT_ON_ROOT, | |
11107 | .read_s64 = cpu_idle_read_s64, | |
11108 | .write_s64 = cpu_idle_write_s64, | |
11109 | }, | |
0d593634 TH |
11110 | #endif |
11111 | #ifdef CONFIG_CFS_BANDWIDTH | |
11112 | { | |
11113 | .name = "max", | |
11114 | .flags = CFTYPE_NOT_ON_ROOT, | |
11115 | .seq_show = cpu_max_show, | |
11116 | .write = cpu_max_write, | |
11117 | }, | |
f4183717 HC |
11118 | { |
11119 | .name = "max.burst", | |
11120 | .flags = CFTYPE_NOT_ON_ROOT, | |
11121 | .read_u64 = cpu_cfs_burst_read_u64, | |
11122 | .write_u64 = cpu_cfs_burst_write_u64, | |
11123 | }, | |
2480c093 PB |
11124 | #endif |
11125 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
11126 | { | |
11127 | .name = "uclamp.min", | |
11128 | .flags = CFTYPE_NOT_ON_ROOT, | |
11129 | .seq_show = cpu_uclamp_min_show, | |
11130 | .write = cpu_uclamp_min_write, | |
11131 | }, | |
11132 | { | |
11133 | .name = "uclamp.max", | |
11134 | .flags = CFTYPE_NOT_ON_ROOT, | |
11135 | .seq_show = cpu_uclamp_max_show, | |
11136 | .write = cpu_uclamp_max_write, | |
11137 | }, | |
0d593634 TH |
11138 | #endif |
11139 | { } /* terminate */ | |
11140 | }; | |
11141 | ||
073219e9 | 11142 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 11143 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 11144 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 11145 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 11146 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 11147 | .css_extra_stat_show = cpu_extra_stat_show, |
df16b71c | 11148 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 | 11149 | .can_attach = cpu_cgroup_can_attach, |
df16b71c | 11150 | #endif |
bb9d97b6 | 11151 | .attach = cpu_cgroup_attach, |
a1f7164c | 11152 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 11153 | .dfl_cftypes = cpu_files, |
b38e42e9 | 11154 | .early_init = true, |
0d593634 | 11155 | .threaded = true, |
68318b8e SV |
11156 | }; |
11157 | ||
052f1dc7 | 11158 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 11159 | |
b637a328 PM |
11160 | void dump_cpu_task(int cpu) |
11161 | { | |
bc1cca97 ZL |
11162 | if (cpu == smp_processor_id() && in_hardirq()) { |
11163 | struct pt_regs *regs; | |
11164 | ||
11165 | regs = get_irq_regs(); | |
11166 | if (regs) { | |
11167 | show_regs(regs); | |
11168 | return; | |
11169 | } | |
11170 | } | |
11171 | ||
e73dfe30 ZL |
11172 | if (trigger_single_cpu_backtrace(cpu)) |
11173 | return; | |
11174 | ||
b637a328 PM |
11175 | pr_info("Task dump for CPU %d:\n", cpu); |
11176 | sched_show_task(cpu_curr(cpu)); | |
11177 | } | |
ed82b8a1 AK |
11178 | |
11179 | /* | |
11180 | * Nice levels are multiplicative, with a gentle 10% change for every | |
11181 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
11182 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
11183 | * that remained on nice 0. | |
11184 | * | |
11185 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
11186 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
11187 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
11188 | * If a task goes up by ~10% and another task goes down by ~10% then | |
11189 | * the relative distance between them is ~25%.) | |
11190 | */ | |
11191 | const int sched_prio_to_weight[40] = { | |
11192 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
11193 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
11194 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
11195 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
11196 | /* 0 */ 1024, 820, 655, 526, 423, | |
11197 | /* 5 */ 335, 272, 215, 172, 137, | |
11198 | /* 10 */ 110, 87, 70, 56, 45, | |
11199 | /* 15 */ 36, 29, 23, 18, 15, | |
11200 | }; | |
11201 | ||
11202 | /* | |
11203 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
11204 | * | |
11205 | * In cases where the weight does not change often, we can use the | |
11206 | * precalculated inverse to speed up arithmetics by turning divisions | |
11207 | * into multiplications: | |
11208 | */ | |
11209 | const u32 sched_prio_to_wmult[40] = { | |
11210 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
11211 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
11212 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
11213 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
11214 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
11215 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
11216 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
11217 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
11218 | }; | |
14a7405b | 11219 | |
9d246053 PA |
11220 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
11221 | { | |
11222 | trace_sched_update_nr_running_tp(rq, count); | |
11223 | } |