]>
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 | ||
24422603 | 1395 | uclamp_rq_set(rq, clamp_id, clamp_value); |
e496187d PB |
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 | ||
24422603 QY |
1546 | if (uc_se->value > uclamp_rq_get(rq, clamp_id)) |
1547 | uclamp_rq_set(rq, clamp_id, 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 | ||
24422603 | 1613 | rq_clamp = uclamp_rq_get(rq, clamp_id); |
69842cba PB |
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); | |
24422603 | 1621 | uclamp_rq_set(rq, clamp_id, bkt_clamp); |
e496187d | 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); |
52b33d87 | 2056 | psi_enqueue(p, (flags & ENQUEUE_WAKEUP) && !(flags & ENQUEUE_MIGRATED)); |
eb414681 | 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 | 2191 | static void |
713a2e21 | 2192 | __do_set_cpus_allowed(struct task_struct *p, struct affinity_context *ctx); |
af449901 PZ |
2193 | |
2194 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
713a2e21 | 2195 | struct affinity_context *ctx); |
af449901 PZ |
2196 | |
2197 | static void migrate_disable_switch(struct rq *rq, struct task_struct *p) | |
2198 | { | |
713a2e21 WL |
2199 | struct affinity_context ac = { |
2200 | .new_mask = cpumask_of(rq->cpu), | |
2201 | .flags = SCA_MIGRATE_DISABLE, | |
2202 | }; | |
2203 | ||
af449901 PZ |
2204 | if (likely(!p->migration_disabled)) |
2205 | return; | |
2206 | ||
2207 | if (p->cpus_ptr != &p->cpus_mask) | |
2208 | return; | |
2209 | ||
2210 | /* | |
2211 | * Violates locking rules! see comment in __do_set_cpus_allowed(). | |
2212 | */ | |
713a2e21 | 2213 | __do_set_cpus_allowed(p, &ac); |
af449901 PZ |
2214 | } |
2215 | ||
2216 | void migrate_disable(void) | |
2217 | { | |
3015ef4b TG |
2218 | struct task_struct *p = current; |
2219 | ||
2220 | if (p->migration_disabled) { | |
2221 | p->migration_disabled++; | |
af449901 | 2222 | return; |
3015ef4b | 2223 | } |
af449901 | 2224 | |
3015ef4b TG |
2225 | preempt_disable(); |
2226 | this_rq()->nr_pinned++; | |
2227 | p->migration_disabled = 1; | |
2228 | preempt_enable(); | |
af449901 PZ |
2229 | } |
2230 | EXPORT_SYMBOL_GPL(migrate_disable); | |
2231 | ||
2232 | void migrate_enable(void) | |
2233 | { | |
2234 | struct task_struct *p = current; | |
713a2e21 WL |
2235 | struct affinity_context ac = { |
2236 | .new_mask = &p->cpus_mask, | |
2237 | .flags = SCA_MIGRATE_ENABLE, | |
2238 | }; | |
af449901 | 2239 | |
6d337eab PZ |
2240 | if (p->migration_disabled > 1) { |
2241 | p->migration_disabled--; | |
af449901 | 2242 | return; |
6d337eab | 2243 | } |
af449901 | 2244 | |
9d0df377 SAS |
2245 | if (WARN_ON_ONCE(!p->migration_disabled)) |
2246 | return; | |
2247 | ||
6d337eab PZ |
2248 | /* |
2249 | * Ensure stop_task runs either before or after this, and that | |
2250 | * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule(). | |
2251 | */ | |
2252 | preempt_disable(); | |
2253 | if (p->cpus_ptr != &p->cpus_mask) | |
713a2e21 | 2254 | __set_cpus_allowed_ptr(p, &ac); |
6d337eab PZ |
2255 | /* |
2256 | * Mustn't clear migration_disabled() until cpus_ptr points back at the | |
2257 | * regular cpus_mask, otherwise things that race (eg. | |
2258 | * select_fallback_rq) get confused. | |
2259 | */ | |
af449901 | 2260 | barrier(); |
6d337eab | 2261 | p->migration_disabled = 0; |
3015ef4b | 2262 | this_rq()->nr_pinned--; |
6d337eab | 2263 | preempt_enable(); |
af449901 PZ |
2264 | } |
2265 | EXPORT_SYMBOL_GPL(migrate_enable); | |
2266 | ||
3015ef4b TG |
2267 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
2268 | { | |
2269 | return rq->nr_pinned; | |
2270 | } | |
2271 | ||
175f0e25 | 2272 | /* |
bee98539 | 2273 | * Per-CPU kthreads are allowed to run on !active && online CPUs, see |
175f0e25 PZ |
2274 | * __set_cpus_allowed_ptr() and select_fallback_rq(). |
2275 | */ | |
2276 | static inline bool is_cpu_allowed(struct task_struct *p, int cpu) | |
2277 | { | |
5ba2ffba | 2278 | /* When not in the task's cpumask, no point in looking further. */ |
3bd37062 | 2279 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) |
175f0e25 PZ |
2280 | return false; |
2281 | ||
5ba2ffba PZ |
2282 | /* migrate_disabled() must be allowed to finish. */ |
2283 | if (is_migration_disabled(p)) | |
175f0e25 PZ |
2284 | return cpu_online(cpu); |
2285 | ||
5ba2ffba PZ |
2286 | /* Non kernel threads are not allowed during either online or offline. */ |
2287 | if (!(p->flags & PF_KTHREAD)) | |
9ae606bc | 2288 | return cpu_active(cpu) && task_cpu_possible(cpu, p); |
5ba2ffba PZ |
2289 | |
2290 | /* KTHREAD_IS_PER_CPU is always allowed. */ | |
2291 | if (kthread_is_per_cpu(p)) | |
2292 | return cpu_online(cpu); | |
2293 | ||
2294 | /* Regular kernel threads don't get to stay during offline. */ | |
b5c44773 | 2295 | if (cpu_dying(cpu)) |
5ba2ffba PZ |
2296 | return false; |
2297 | ||
2298 | /* But are allowed during online. */ | |
2299 | return cpu_online(cpu); | |
175f0e25 PZ |
2300 | } |
2301 | ||
5cc389bc PZ |
2302 | /* |
2303 | * This is how migration works: | |
2304 | * | |
2305 | * 1) we invoke migration_cpu_stop() on the target CPU using | |
2306 | * stop_one_cpu(). | |
2307 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
2308 | * off the CPU) | |
2309 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
2310 | * 4) if it's in the wrong runqueue then the migration thread removes | |
2311 | * it and puts it into the right queue. | |
2312 | * 5) stopper completes and stop_one_cpu() returns and the migration | |
2313 | * is done. | |
2314 | */ | |
2315 | ||
2316 | /* | |
2317 | * move_queued_task - move a queued task to new rq. | |
2318 | * | |
2319 | * Returns (locked) new rq. Old rq's lock is released. | |
2320 | */ | |
8a8c69c3 PZ |
2321 | static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf, |
2322 | struct task_struct *p, int new_cpu) | |
5cc389bc | 2323 | { |
5cb9eaa3 | 2324 | lockdep_assert_rq_held(rq); |
5cc389bc | 2325 | |
58877d34 | 2326 | deactivate_task(rq, p, DEQUEUE_NOCLOCK); |
5cc389bc | 2327 | set_task_cpu(p, new_cpu); |
8a8c69c3 | 2328 | rq_unlock(rq, rf); |
5cc389bc PZ |
2329 | |
2330 | rq = cpu_rq(new_cpu); | |
2331 | ||
8a8c69c3 | 2332 | rq_lock(rq, rf); |
09348d75 | 2333 | WARN_ON_ONCE(task_cpu(p) != new_cpu); |
58877d34 | 2334 | activate_task(rq, p, 0); |
5cc389bc PZ |
2335 | check_preempt_curr(rq, p, 0); |
2336 | ||
2337 | return rq; | |
2338 | } | |
2339 | ||
2340 | struct migration_arg { | |
6d337eab PZ |
2341 | struct task_struct *task; |
2342 | int dest_cpu; | |
2343 | struct set_affinity_pending *pending; | |
2344 | }; | |
2345 | ||
50caf9c1 PZ |
2346 | /* |
2347 | * @refs: number of wait_for_completion() | |
2348 | * @stop_pending: is @stop_work in use | |
2349 | */ | |
6d337eab PZ |
2350 | struct set_affinity_pending { |
2351 | refcount_t refs; | |
9e81889c | 2352 | unsigned int stop_pending; |
6d337eab PZ |
2353 | struct completion done; |
2354 | struct cpu_stop_work stop_work; | |
2355 | struct migration_arg arg; | |
5cc389bc PZ |
2356 | }; |
2357 | ||
2358 | /* | |
d1ccc66d | 2359 | * Move (not current) task off this CPU, onto the destination CPU. We're doing |
5cc389bc PZ |
2360 | * this because either it can't run here any more (set_cpus_allowed() |
2361 | * away from this CPU, or CPU going down), or because we're | |
2362 | * attempting to rebalance this task on exec (sched_exec). | |
2363 | * | |
2364 | * So we race with normal scheduler movements, but that's OK, as long | |
2365 | * as the task is no longer on this CPU. | |
5cc389bc | 2366 | */ |
8a8c69c3 PZ |
2367 | static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf, |
2368 | struct task_struct *p, int dest_cpu) | |
5cc389bc | 2369 | { |
5cc389bc | 2370 | /* Affinity changed (again). */ |
175f0e25 | 2371 | if (!is_cpu_allowed(p, dest_cpu)) |
5e16bbc2 | 2372 | return rq; |
5cc389bc | 2373 | |
15ff991e | 2374 | update_rq_clock(rq); |
8a8c69c3 | 2375 | rq = move_queued_task(rq, rf, p, dest_cpu); |
5e16bbc2 PZ |
2376 | |
2377 | return rq; | |
5cc389bc PZ |
2378 | } |
2379 | ||
2380 | /* | |
2381 | * migration_cpu_stop - this will be executed by a highprio stopper thread | |
2382 | * and performs thread migration by bumping thread off CPU then | |
2383 | * 'pushing' onto another runqueue. | |
2384 | */ | |
2385 | static int migration_cpu_stop(void *data) | |
2386 | { | |
2387 | struct migration_arg *arg = data; | |
c20cf065 | 2388 | struct set_affinity_pending *pending = arg->pending; |
5e16bbc2 PZ |
2389 | struct task_struct *p = arg->task; |
2390 | struct rq *rq = this_rq(); | |
6d337eab | 2391 | bool complete = false; |
8a8c69c3 | 2392 | struct rq_flags rf; |
5cc389bc PZ |
2393 | |
2394 | /* | |
d1ccc66d IM |
2395 | * The original target CPU might have gone down and we might |
2396 | * be on another CPU but it doesn't matter. | |
5cc389bc | 2397 | */ |
6d337eab | 2398 | local_irq_save(rf.flags); |
5cc389bc PZ |
2399 | /* |
2400 | * We need to explicitly wake pending tasks before running | |
3bd37062 | 2401 | * __migrate_task() such that we will not miss enforcing cpus_ptr |
5cc389bc PZ |
2402 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
2403 | */ | |
16bf5a5e | 2404 | flush_smp_call_function_queue(); |
5e16bbc2 PZ |
2405 | |
2406 | raw_spin_lock(&p->pi_lock); | |
8a8c69c3 | 2407 | rq_lock(rq, &rf); |
6d337eab | 2408 | |
e140749c VS |
2409 | /* |
2410 | * If we were passed a pending, then ->stop_pending was set, thus | |
2411 | * p->migration_pending must have remained stable. | |
2412 | */ | |
2413 | WARN_ON_ONCE(pending && pending != p->migration_pending); | |
2414 | ||
5e16bbc2 PZ |
2415 | /* |
2416 | * If task_rq(p) != rq, it cannot be migrated here, because we're | |
2417 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because | |
2418 | * we're holding p->pi_lock. | |
2419 | */ | |
bf89a304 | 2420 | if (task_rq(p) == rq) { |
6d337eab PZ |
2421 | if (is_migration_disabled(p)) |
2422 | goto out; | |
2423 | ||
2424 | if (pending) { | |
e140749c | 2425 | p->migration_pending = NULL; |
6d337eab | 2426 | complete = true; |
6d337eab | 2427 | |
3f1bc119 PZ |
2428 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) |
2429 | goto out; | |
3f1bc119 | 2430 | } |
6d337eab | 2431 | |
bf89a304 | 2432 | if (task_on_rq_queued(p)) |
475ea6c6 | 2433 | rq = __migrate_task(rq, &rf, p, arg->dest_cpu); |
bf89a304 | 2434 | else |
475ea6c6 | 2435 | p->wake_cpu = arg->dest_cpu; |
6d337eab | 2436 | |
3f1bc119 PZ |
2437 | /* |
2438 | * XXX __migrate_task() can fail, at which point we might end | |
2439 | * up running on a dodgy CPU, AFAICT this can only happen | |
2440 | * during CPU hotplug, at which point we'll get pushed out | |
2441 | * anyway, so it's probably not a big deal. | |
2442 | */ | |
2443 | ||
c20cf065 | 2444 | } else if (pending) { |
6d337eab PZ |
2445 | /* |
2446 | * This happens when we get migrated between migrate_enable()'s | |
2447 | * preempt_enable() and scheduling the stopper task. At that | |
2448 | * point we're a regular task again and not current anymore. | |
2449 | * | |
2450 | * A !PREEMPT kernel has a giant hole here, which makes it far | |
2451 | * more likely. | |
2452 | */ | |
2453 | ||
d707faa6 VS |
2454 | /* |
2455 | * The task moved before the stopper got to run. We're holding | |
2456 | * ->pi_lock, so the allowed mask is stable - if it got | |
2457 | * somewhere allowed, we're done. | |
2458 | */ | |
c20cf065 | 2459 | if (cpumask_test_cpu(task_cpu(p), p->cpus_ptr)) { |
e140749c | 2460 | p->migration_pending = NULL; |
d707faa6 VS |
2461 | complete = true; |
2462 | goto out; | |
2463 | } | |
2464 | ||
6d337eab PZ |
2465 | /* |
2466 | * When migrate_enable() hits a rq mis-match we can't reliably | |
2467 | * determine is_migration_disabled() and so have to chase after | |
2468 | * it. | |
2469 | */ | |
9e81889c | 2470 | WARN_ON_ONCE(!pending->stop_pending); |
6d337eab PZ |
2471 | task_rq_unlock(rq, p, &rf); |
2472 | stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop, | |
2473 | &pending->arg, &pending->stop_work); | |
2474 | return 0; | |
bf89a304 | 2475 | } |
6d337eab | 2476 | out: |
9e81889c PZ |
2477 | if (pending) |
2478 | pending->stop_pending = false; | |
6d337eab PZ |
2479 | task_rq_unlock(rq, p, &rf); |
2480 | ||
2481 | if (complete) | |
2482 | complete_all(&pending->done); | |
2483 | ||
5cc389bc PZ |
2484 | return 0; |
2485 | } | |
2486 | ||
a7c81556 PZ |
2487 | int push_cpu_stop(void *arg) |
2488 | { | |
2489 | struct rq *lowest_rq = NULL, *rq = this_rq(); | |
2490 | struct task_struct *p = arg; | |
2491 | ||
2492 | raw_spin_lock_irq(&p->pi_lock); | |
5cb9eaa3 | 2493 | raw_spin_rq_lock(rq); |
a7c81556 PZ |
2494 | |
2495 | if (task_rq(p) != rq) | |
2496 | goto out_unlock; | |
2497 | ||
2498 | if (is_migration_disabled(p)) { | |
2499 | p->migration_flags |= MDF_PUSH; | |
2500 | goto out_unlock; | |
2501 | } | |
2502 | ||
2503 | p->migration_flags &= ~MDF_PUSH; | |
2504 | ||
2505 | if (p->sched_class->find_lock_rq) | |
2506 | lowest_rq = p->sched_class->find_lock_rq(p, rq); | |
5e16bbc2 | 2507 | |
a7c81556 PZ |
2508 | if (!lowest_rq) |
2509 | goto out_unlock; | |
2510 | ||
2511 | // XXX validate p is still the highest prio task | |
2512 | if (task_rq(p) == rq) { | |
2513 | deactivate_task(rq, p, 0); | |
2514 | set_task_cpu(p, lowest_rq->cpu); | |
2515 | activate_task(lowest_rq, p, 0); | |
2516 | resched_curr(lowest_rq); | |
2517 | } | |
2518 | ||
2519 | double_unlock_balance(rq, lowest_rq); | |
2520 | ||
2521 | out_unlock: | |
2522 | rq->push_busy = false; | |
5cb9eaa3 | 2523 | raw_spin_rq_unlock(rq); |
a7c81556 PZ |
2524 | raw_spin_unlock_irq(&p->pi_lock); |
2525 | ||
2526 | put_task_struct(p); | |
5cc389bc PZ |
2527 | return 0; |
2528 | } | |
2529 | ||
c5b28038 PZ |
2530 | /* |
2531 | * sched_class::set_cpus_allowed must do the below, but is not required to | |
2532 | * actually call this function. | |
2533 | */ | |
713a2e21 | 2534 | void set_cpus_allowed_common(struct task_struct *p, struct affinity_context *ctx) |
5cc389bc | 2535 | { |
713a2e21 WL |
2536 | if (ctx->flags & (SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) { |
2537 | p->cpus_ptr = ctx->new_mask; | |
af449901 PZ |
2538 | return; |
2539 | } | |
2540 | ||
713a2e21 WL |
2541 | cpumask_copy(&p->cpus_mask, ctx->new_mask); |
2542 | p->nr_cpus_allowed = cpumask_weight(ctx->new_mask); | |
8f9ea86f WL |
2543 | |
2544 | /* | |
2545 | * Swap in a new user_cpus_ptr if SCA_USER flag set | |
2546 | */ | |
2547 | if (ctx->flags & SCA_USER) | |
2548 | swap(p->user_cpus_ptr, ctx->user_mask); | |
5cc389bc PZ |
2549 | } |
2550 | ||
9cfc3e18 | 2551 | static void |
713a2e21 | 2552 | __do_set_cpus_allowed(struct task_struct *p, struct affinity_context *ctx) |
c5b28038 | 2553 | { |
6c37067e PZ |
2554 | struct rq *rq = task_rq(p); |
2555 | bool queued, running; | |
2556 | ||
af449901 PZ |
2557 | /* |
2558 | * This here violates the locking rules for affinity, since we're only | |
2559 | * supposed to change these variables while holding both rq->lock and | |
2560 | * p->pi_lock. | |
2561 | * | |
2562 | * HOWEVER, it magically works, because ttwu() is the only code that | |
2563 | * accesses these variables under p->pi_lock and only does so after | |
2564 | * smp_cond_load_acquire(&p->on_cpu, !VAL), and we're in __schedule() | |
2565 | * before finish_task(). | |
2566 | * | |
2567 | * XXX do further audits, this smells like something putrid. | |
2568 | */ | |
713a2e21 | 2569 | if (ctx->flags & SCA_MIGRATE_DISABLE) |
af449901 PZ |
2570 | SCHED_WARN_ON(!p->on_cpu); |
2571 | else | |
2572 | lockdep_assert_held(&p->pi_lock); | |
6c37067e PZ |
2573 | |
2574 | queued = task_on_rq_queued(p); | |
2575 | running = task_current(rq, p); | |
2576 | ||
2577 | if (queued) { | |
2578 | /* | |
2579 | * Because __kthread_bind() calls this on blocked tasks without | |
2580 | * holding rq->lock. | |
2581 | */ | |
5cb9eaa3 | 2582 | lockdep_assert_rq_held(rq); |
7a57f32a | 2583 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
6c37067e PZ |
2584 | } |
2585 | if (running) | |
2586 | put_prev_task(rq, p); | |
2587 | ||
713a2e21 | 2588 | p->sched_class->set_cpus_allowed(p, ctx); |
6c37067e | 2589 | |
6c37067e | 2590 | if (queued) |
7134b3e9 | 2591 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 2592 | if (running) |
03b7fad1 | 2593 | set_next_task(rq, p); |
c5b28038 PZ |
2594 | } |
2595 | ||
851a723e WL |
2596 | /* |
2597 | * Used for kthread_bind() and select_fallback_rq(), in both cases the user | |
2598 | * affinity (if any) should be destroyed too. | |
2599 | */ | |
9cfc3e18 PZ |
2600 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
2601 | { | |
713a2e21 WL |
2602 | struct affinity_context ac = { |
2603 | .new_mask = new_mask, | |
851a723e WL |
2604 | .user_mask = NULL, |
2605 | .flags = SCA_USER, /* clear the user requested mask */ | |
713a2e21 WL |
2606 | }; |
2607 | ||
2608 | __do_set_cpus_allowed(p, &ac); | |
851a723e | 2609 | kfree(ac.user_mask); |
9cfc3e18 PZ |
2610 | } |
2611 | ||
b90ca8ba WD |
2612 | int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, |
2613 | int node) | |
2614 | { | |
8f9ea86f WL |
2615 | unsigned long flags; |
2616 | ||
b90ca8ba WD |
2617 | if (!src->user_cpus_ptr) |
2618 | return 0; | |
2619 | ||
2620 | dst->user_cpus_ptr = kmalloc_node(cpumask_size(), GFP_KERNEL, node); | |
2621 | if (!dst->user_cpus_ptr) | |
2622 | return -ENOMEM; | |
2623 | ||
8f9ea86f WL |
2624 | /* Use pi_lock to protect content of user_cpus_ptr */ |
2625 | raw_spin_lock_irqsave(&src->pi_lock, flags); | |
b90ca8ba | 2626 | cpumask_copy(dst->user_cpus_ptr, src->user_cpus_ptr); |
8f9ea86f | 2627 | raw_spin_unlock_irqrestore(&src->pi_lock, flags); |
b90ca8ba WD |
2628 | return 0; |
2629 | } | |
2630 | ||
07ec77a1 WD |
2631 | static inline struct cpumask *clear_user_cpus_ptr(struct task_struct *p) |
2632 | { | |
2633 | struct cpumask *user_mask = NULL; | |
2634 | ||
2635 | swap(p->user_cpus_ptr, user_mask); | |
2636 | ||
2637 | return user_mask; | |
2638 | } | |
2639 | ||
b90ca8ba WD |
2640 | void release_user_cpus_ptr(struct task_struct *p) |
2641 | { | |
07ec77a1 | 2642 | kfree(clear_user_cpus_ptr(p)); |
b90ca8ba WD |
2643 | } |
2644 | ||
6d337eab | 2645 | /* |
c777d847 VS |
2646 | * This function is wildly self concurrent; here be dragons. |
2647 | * | |
2648 | * | |
2649 | * When given a valid mask, __set_cpus_allowed_ptr() must block until the | |
2650 | * designated task is enqueued on an allowed CPU. If that task is currently | |
2651 | * running, we have to kick it out using the CPU stopper. | |
2652 | * | |
2653 | * Migrate-Disable comes along and tramples all over our nice sandcastle. | |
2654 | * Consider: | |
2655 | * | |
2656 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2657 | * | |
2658 | * P0@CPU0 P1 | |
2659 | * | |
2660 | * migrate_disable(); | |
2661 | * <preempted> | |
2662 | * set_cpus_allowed_ptr(P0, [1]); | |
2663 | * | |
2664 | * P1 *cannot* return from this set_cpus_allowed_ptr() call until P0 executes | |
2665 | * its outermost migrate_enable() (i.e. it exits its Migrate-Disable region). | |
2666 | * This means we need the following scheme: | |
2667 | * | |
2668 | * P0@CPU0 P1 | |
2669 | * | |
2670 | * migrate_disable(); | |
2671 | * <preempted> | |
2672 | * set_cpus_allowed_ptr(P0, [1]); | |
2673 | * <blocks> | |
2674 | * <resumes> | |
2675 | * migrate_enable(); | |
2676 | * __set_cpus_allowed_ptr(); | |
2677 | * <wakes local stopper> | |
2678 | * `--> <woken on migration completion> | |
2679 | * | |
2680 | * Now the fun stuff: there may be several P1-like tasks, i.e. multiple | |
2681 | * concurrent set_cpus_allowed_ptr(P0, [*]) calls. CPU affinity changes of any | |
2682 | * task p are serialized by p->pi_lock, which we can leverage: the one that | |
2683 | * should come into effect at the end of the Migrate-Disable region is the last | |
2684 | * one. This means we only need to track a single cpumask (i.e. p->cpus_mask), | |
2685 | * but we still need to properly signal those waiting tasks at the appropriate | |
2686 | * moment. | |
2687 | * | |
2688 | * This is implemented using struct set_affinity_pending. The first | |
2689 | * __set_cpus_allowed_ptr() caller within a given Migrate-Disable region will | |
2690 | * setup an instance of that struct and install it on the targeted task_struct. | |
2691 | * Any and all further callers will reuse that instance. Those then wait for | |
2692 | * a completion signaled at the tail of the CPU stopper callback (1), triggered | |
2693 | * on the end of the Migrate-Disable region (i.e. outermost migrate_enable()). | |
2694 | * | |
2695 | * | |
2696 | * (1) In the cases covered above. There is one more where the completion is | |
2697 | * signaled within affine_move_task() itself: when a subsequent affinity request | |
e140749c VS |
2698 | * occurs after the stopper bailed out due to the targeted task still being |
2699 | * Migrate-Disable. Consider: | |
c777d847 VS |
2700 | * |
2701 | * Initial conditions: P0->cpus_mask = [0, 1] | |
2702 | * | |
e140749c VS |
2703 | * CPU0 P1 P2 |
2704 | * <P0> | |
2705 | * migrate_disable(); | |
2706 | * <preempted> | |
c777d847 VS |
2707 | * set_cpus_allowed_ptr(P0, [1]); |
2708 | * <blocks> | |
e140749c VS |
2709 | * <migration/0> |
2710 | * migration_cpu_stop() | |
2711 | * is_migration_disabled() | |
2712 | * <bails> | |
c777d847 VS |
2713 | * set_cpus_allowed_ptr(P0, [0, 1]); |
2714 | * <signal completion> | |
2715 | * <awakes> | |
2716 | * | |
2717 | * Note that the above is safe vs a concurrent migrate_enable(), as any | |
2718 | * pending affinity completion is preceded by an uninstallation of | |
2719 | * p->migration_pending done with p->pi_lock held. | |
6d337eab PZ |
2720 | */ |
2721 | static int affine_move_task(struct rq *rq, struct task_struct *p, struct rq_flags *rf, | |
2722 | int dest_cpu, unsigned int flags) | |
5584e8ac WL |
2723 | __releases(rq->lock) |
2724 | __releases(p->pi_lock) | |
6d337eab PZ |
2725 | { |
2726 | struct set_affinity_pending my_pending = { }, *pending = NULL; | |
9e81889c | 2727 | bool stop_pending, complete = false; |
6d337eab PZ |
2728 | |
2729 | /* Can the task run on the task's current CPU? If so, we're done */ | |
2730 | if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) { | |
a7c81556 PZ |
2731 | struct task_struct *push_task = NULL; |
2732 | ||
2733 | if ((flags & SCA_MIGRATE_ENABLE) && | |
2734 | (p->migration_flags & MDF_PUSH) && !rq->push_busy) { | |
2735 | rq->push_busy = true; | |
2736 | push_task = get_task_struct(p); | |
2737 | } | |
2738 | ||
50caf9c1 PZ |
2739 | /* |
2740 | * If there are pending waiters, but no pending stop_work, | |
2741 | * then complete now. | |
2742 | */ | |
6d337eab | 2743 | pending = p->migration_pending; |
50caf9c1 | 2744 | if (pending && !pending->stop_pending) { |
6d337eab PZ |
2745 | p->migration_pending = NULL; |
2746 | complete = true; | |
2747 | } | |
50caf9c1 | 2748 | |
6d337eab PZ |
2749 | task_rq_unlock(rq, p, rf); |
2750 | ||
a7c81556 PZ |
2751 | if (push_task) { |
2752 | stop_one_cpu_nowait(rq->cpu, push_cpu_stop, | |
2753 | p, &rq->push_work); | |
2754 | } | |
2755 | ||
6d337eab | 2756 | if (complete) |
50caf9c1 | 2757 | complete_all(&pending->done); |
6d337eab PZ |
2758 | |
2759 | return 0; | |
2760 | } | |
2761 | ||
2762 | if (!(flags & SCA_MIGRATE_ENABLE)) { | |
2763 | /* serialized by p->pi_lock */ | |
2764 | if (!p->migration_pending) { | |
c777d847 | 2765 | /* Install the request */ |
6d337eab PZ |
2766 | refcount_set(&my_pending.refs, 1); |
2767 | init_completion(&my_pending.done); | |
8a6edb52 PZ |
2768 | my_pending.arg = (struct migration_arg) { |
2769 | .task = p, | |
475ea6c6 | 2770 | .dest_cpu = dest_cpu, |
8a6edb52 PZ |
2771 | .pending = &my_pending, |
2772 | }; | |
2773 | ||
6d337eab PZ |
2774 | p->migration_pending = &my_pending; |
2775 | } else { | |
2776 | pending = p->migration_pending; | |
2777 | refcount_inc(&pending->refs); | |
475ea6c6 VS |
2778 | /* |
2779 | * Affinity has changed, but we've already installed a | |
2780 | * pending. migration_cpu_stop() *must* see this, else | |
2781 | * we risk a completion of the pending despite having a | |
2782 | * task on a disallowed CPU. | |
2783 | * | |
2784 | * Serialized by p->pi_lock, so this is safe. | |
2785 | */ | |
2786 | pending->arg.dest_cpu = dest_cpu; | |
6d337eab PZ |
2787 | } |
2788 | } | |
2789 | pending = p->migration_pending; | |
2790 | /* | |
2791 | * - !MIGRATE_ENABLE: | |
2792 | * we'll have installed a pending if there wasn't one already. | |
2793 | * | |
2794 | * - MIGRATE_ENABLE: | |
2795 | * we're here because the current CPU isn't matching anymore, | |
2796 | * the only way that can happen is because of a concurrent | |
2797 | * set_cpus_allowed_ptr() call, which should then still be | |
2798 | * pending completion. | |
2799 | * | |
2800 | * Either way, we really should have a @pending here. | |
2801 | */ | |
2802 | if (WARN_ON_ONCE(!pending)) { | |
2803 | task_rq_unlock(rq, p, rf); | |
2804 | return -EINVAL; | |
2805 | } | |
2806 | ||
0b9d46fc | 2807 | if (task_on_cpu(rq, p) || READ_ONCE(p->__state) == TASK_WAKING) { |
c777d847 | 2808 | /* |
58b1a450 PZ |
2809 | * MIGRATE_ENABLE gets here because 'p == current', but for |
2810 | * anything else we cannot do is_migration_disabled(), punt | |
2811 | * and have the stopper function handle it all race-free. | |
c777d847 | 2812 | */ |
9e81889c PZ |
2813 | stop_pending = pending->stop_pending; |
2814 | if (!stop_pending) | |
2815 | pending->stop_pending = true; | |
58b1a450 | 2816 | |
58b1a450 PZ |
2817 | if (flags & SCA_MIGRATE_ENABLE) |
2818 | p->migration_flags &= ~MDF_PUSH; | |
50caf9c1 | 2819 | |
6d337eab | 2820 | task_rq_unlock(rq, p, rf); |
8a6edb52 | 2821 | |
9e81889c PZ |
2822 | if (!stop_pending) { |
2823 | stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop, | |
2824 | &pending->arg, &pending->stop_work); | |
2825 | } | |
6d337eab | 2826 | |
58b1a450 PZ |
2827 | if (flags & SCA_MIGRATE_ENABLE) |
2828 | return 0; | |
6d337eab PZ |
2829 | } else { |
2830 | ||
2831 | if (!is_migration_disabled(p)) { | |
2832 | if (task_on_rq_queued(p)) | |
2833 | rq = move_queued_task(rq, rf, p, dest_cpu); | |
2834 | ||
50caf9c1 PZ |
2835 | if (!pending->stop_pending) { |
2836 | p->migration_pending = NULL; | |
2837 | complete = true; | |
2838 | } | |
6d337eab PZ |
2839 | } |
2840 | task_rq_unlock(rq, p, rf); | |
2841 | ||
6d337eab PZ |
2842 | if (complete) |
2843 | complete_all(&pending->done); | |
2844 | } | |
2845 | ||
2846 | wait_for_completion(&pending->done); | |
2847 | ||
2848 | if (refcount_dec_and_test(&pending->refs)) | |
50caf9c1 | 2849 | wake_up_var(&pending->refs); /* No UaF, just an address */ |
6d337eab | 2850 | |
c777d847 VS |
2851 | /* |
2852 | * Block the original owner of &pending until all subsequent callers | |
2853 | * have seen the completion and decremented the refcount | |
2854 | */ | |
6d337eab PZ |
2855 | wait_var_event(&my_pending.refs, !refcount_read(&my_pending.refs)); |
2856 | ||
50caf9c1 PZ |
2857 | /* ARGH */ |
2858 | WARN_ON_ONCE(my_pending.stop_pending); | |
2859 | ||
6d337eab PZ |
2860 | return 0; |
2861 | } | |
2862 | ||
5cc389bc | 2863 | /* |
07ec77a1 | 2864 | * Called with both p->pi_lock and rq->lock held; drops both before returning. |
5cc389bc | 2865 | */ |
07ec77a1 | 2866 | static int __set_cpus_allowed_ptr_locked(struct task_struct *p, |
713a2e21 | 2867 | struct affinity_context *ctx, |
07ec77a1 WD |
2868 | struct rq *rq, |
2869 | struct rq_flags *rf) | |
2870 | __releases(rq->lock) | |
2871 | __releases(p->pi_lock) | |
5cc389bc | 2872 | { |
234a503e | 2873 | const struct cpumask *cpu_allowed_mask = task_cpu_possible_mask(p); |
e9d867a6 | 2874 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
234a503e | 2875 | bool kthread = p->flags & PF_KTHREAD; |
5cc389bc PZ |
2876 | unsigned int dest_cpu; |
2877 | int ret = 0; | |
2878 | ||
a499c3ea | 2879 | update_rq_clock(rq); |
5cc389bc | 2880 | |
234a503e | 2881 | if (kthread || is_migration_disabled(p)) { |
e9d867a6 | 2882 | /* |
741ba80f PZ |
2883 | * Kernel threads are allowed on online && !active CPUs, |
2884 | * however, during cpu-hot-unplug, even these might get pushed | |
2885 | * away if not KTHREAD_IS_PER_CPU. | |
af449901 PZ |
2886 | * |
2887 | * Specifically, migration_disabled() tasks must not fail the | |
2888 | * cpumask_any_and_distribute() pick below, esp. so on | |
2889 | * SCA_MIGRATE_ENABLE, otherwise we'll not call | |
2890 | * set_cpus_allowed_common() and actually reset p->cpus_ptr. | |
e9d867a6 PZI |
2891 | */ |
2892 | cpu_valid_mask = cpu_online_mask; | |
2893 | } | |
2894 | ||
713a2e21 | 2895 | if (!kthread && !cpumask_subset(ctx->new_mask, cpu_allowed_mask)) { |
234a503e WD |
2896 | ret = -EINVAL; |
2897 | goto out; | |
2898 | } | |
2899 | ||
25834c73 PZ |
2900 | /* |
2901 | * Must re-check here, to close a race against __kthread_bind(), | |
2902 | * sched_setaffinity() is not guaranteed to observe the flag. | |
2903 | */ | |
713a2e21 | 2904 | if ((ctx->flags & SCA_CHECK) && (p->flags & PF_NO_SETAFFINITY)) { |
25834c73 PZ |
2905 | ret = -EINVAL; |
2906 | goto out; | |
2907 | } | |
2908 | ||
713a2e21 WL |
2909 | if (!(ctx->flags & SCA_MIGRATE_ENABLE)) { |
2910 | if (cpumask_equal(&p->cpus_mask, ctx->new_mask)) | |
885b3ba4 VS |
2911 | goto out; |
2912 | ||
2913 | if (WARN_ON_ONCE(p == current && | |
2914 | is_migration_disabled(p) && | |
713a2e21 | 2915 | !cpumask_test_cpu(task_cpu(p), ctx->new_mask))) { |
885b3ba4 VS |
2916 | ret = -EBUSY; |
2917 | goto out; | |
2918 | } | |
2919 | } | |
5cc389bc | 2920 | |
46a87b38 PT |
2921 | /* |
2922 | * Picking a ~random cpu helps in cases where we are changing affinity | |
2923 | * for groups of tasks (ie. cpuset), so that load balancing is not | |
2924 | * immediately required to distribute the tasks within their new mask. | |
2925 | */ | |
713a2e21 | 2926 | dest_cpu = cpumask_any_and_distribute(cpu_valid_mask, ctx->new_mask); |
714e501e | 2927 | if (dest_cpu >= nr_cpu_ids) { |
5cc389bc PZ |
2928 | ret = -EINVAL; |
2929 | goto out; | |
2930 | } | |
2931 | ||
713a2e21 | 2932 | __do_set_cpus_allowed(p, ctx); |
07ec77a1 | 2933 | |
8f9ea86f | 2934 | return affine_move_task(rq, p, rf, dest_cpu, ctx->flags); |
5cc389bc | 2935 | |
5cc389bc | 2936 | out: |
07ec77a1 | 2937 | task_rq_unlock(rq, p, rf); |
5cc389bc PZ |
2938 | |
2939 | return ret; | |
2940 | } | |
25834c73 | 2941 | |
07ec77a1 WD |
2942 | /* |
2943 | * Change a given task's CPU affinity. Migrate the thread to a | |
2944 | * proper CPU and schedule it away if the CPU it's executing on | |
2945 | * is removed from the allowed bitmask. | |
2946 | * | |
2947 | * NOTE: the caller must have a valid reference to the task, the | |
2948 | * task must not exit() & deallocate itself prematurely. The | |
2949 | * call is not atomic; no spinlocks may be held. | |
2950 | */ | |
2951 | static int __set_cpus_allowed_ptr(struct task_struct *p, | |
713a2e21 | 2952 | struct affinity_context *ctx) |
07ec77a1 WD |
2953 | { |
2954 | struct rq_flags rf; | |
2955 | struct rq *rq; | |
2956 | ||
2957 | rq = task_rq_lock(p, &rf); | |
da019032 WL |
2958 | /* |
2959 | * Masking should be skipped if SCA_USER or any of the SCA_MIGRATE_* | |
2960 | * flags are set. | |
2961 | */ | |
2962 | if (p->user_cpus_ptr && | |
2963 | !(ctx->flags & (SCA_USER | SCA_MIGRATE_ENABLE | SCA_MIGRATE_DISABLE)) && | |
2964 | cpumask_and(rq->scratch_mask, ctx->new_mask, p->user_cpus_ptr)) | |
2965 | ctx->new_mask = rq->scratch_mask; | |
2966 | ||
713a2e21 | 2967 | return __set_cpus_allowed_ptr_locked(p, ctx, rq, &rf); |
07ec77a1 WD |
2968 | } |
2969 | ||
25834c73 PZ |
2970 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
2971 | { | |
713a2e21 WL |
2972 | struct affinity_context ac = { |
2973 | .new_mask = new_mask, | |
2974 | .flags = 0, | |
2975 | }; | |
2976 | ||
2977 | return __set_cpus_allowed_ptr(p, &ac); | |
25834c73 | 2978 | } |
5cc389bc PZ |
2979 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
2980 | ||
07ec77a1 WD |
2981 | /* |
2982 | * Change a given task's CPU affinity to the intersection of its current | |
8f9ea86f WL |
2983 | * affinity mask and @subset_mask, writing the resulting mask to @new_mask. |
2984 | * If user_cpus_ptr is defined, use it as the basis for restricting CPU | |
2985 | * affinity or use cpu_online_mask instead. | |
2986 | * | |
07ec77a1 WD |
2987 | * If the resulting mask is empty, leave the affinity unchanged and return |
2988 | * -EINVAL. | |
2989 | */ | |
2990 | static int restrict_cpus_allowed_ptr(struct task_struct *p, | |
2991 | struct cpumask *new_mask, | |
2992 | const struct cpumask *subset_mask) | |
2993 | { | |
8f9ea86f WL |
2994 | struct affinity_context ac = { |
2995 | .new_mask = new_mask, | |
2996 | .flags = 0, | |
2997 | }; | |
07ec77a1 WD |
2998 | struct rq_flags rf; |
2999 | struct rq *rq; | |
3000 | int err; | |
3001 | ||
07ec77a1 WD |
3002 | rq = task_rq_lock(p, &rf); |
3003 | ||
3004 | /* | |
3005 | * Forcefully restricting the affinity of a deadline task is | |
3006 | * likely to cause problems, so fail and noisily override the | |
3007 | * mask entirely. | |
3008 | */ | |
3009 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { | |
3010 | err = -EPERM; | |
3011 | goto err_unlock; | |
3012 | } | |
3013 | ||
8f9ea86f | 3014 | if (!cpumask_and(new_mask, task_user_cpus(p), subset_mask)) { |
07ec77a1 WD |
3015 | err = -EINVAL; |
3016 | goto err_unlock; | |
3017 | } | |
3018 | ||
713a2e21 | 3019 | return __set_cpus_allowed_ptr_locked(p, &ac, rq, &rf); |
07ec77a1 WD |
3020 | |
3021 | err_unlock: | |
3022 | task_rq_unlock(rq, p, &rf); | |
07ec77a1 WD |
3023 | return err; |
3024 | } | |
3025 | ||
3026 | /* | |
3027 | * Restrict the CPU affinity of task @p so that it is a subset of | |
5584e8ac | 3028 | * task_cpu_possible_mask() and point @p->user_cpus_ptr to a copy of the |
07ec77a1 WD |
3029 | * old affinity mask. If the resulting mask is empty, we warn and walk |
3030 | * up the cpuset hierarchy until we find a suitable mask. | |
3031 | */ | |
3032 | void force_compatible_cpus_allowed_ptr(struct task_struct *p) | |
3033 | { | |
3034 | cpumask_var_t new_mask; | |
3035 | const struct cpumask *override_mask = task_cpu_possible_mask(p); | |
3036 | ||
3037 | alloc_cpumask_var(&new_mask, GFP_KERNEL); | |
3038 | ||
3039 | /* | |
3040 | * __migrate_task() can fail silently in the face of concurrent | |
3041 | * offlining of the chosen destination CPU, so take the hotplug | |
3042 | * lock to ensure that the migration succeeds. | |
3043 | */ | |
3044 | cpus_read_lock(); | |
3045 | if (!cpumask_available(new_mask)) | |
3046 | goto out_set_mask; | |
3047 | ||
3048 | if (!restrict_cpus_allowed_ptr(p, new_mask, override_mask)) | |
3049 | goto out_free_mask; | |
3050 | ||
3051 | /* | |
3052 | * We failed to find a valid subset of the affinity mask for the | |
3053 | * task, so override it based on its cpuset hierarchy. | |
3054 | */ | |
3055 | cpuset_cpus_allowed(p, new_mask); | |
3056 | override_mask = new_mask; | |
3057 | ||
3058 | out_set_mask: | |
3059 | if (printk_ratelimit()) { | |
3060 | printk_deferred("Overriding affinity for process %d (%s) to CPUs %*pbl\n", | |
3061 | task_pid_nr(p), p->comm, | |
3062 | cpumask_pr_args(override_mask)); | |
3063 | } | |
3064 | ||
3065 | WARN_ON(set_cpus_allowed_ptr(p, override_mask)); | |
3066 | out_free_mask: | |
3067 | cpus_read_unlock(); | |
3068 | free_cpumask_var(new_mask); | |
3069 | } | |
3070 | ||
3071 | static int | |
713a2e21 | 3072 | __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx); |
07ec77a1 WD |
3073 | |
3074 | /* | |
3075 | * Restore the affinity of a task @p which was previously restricted by a | |
8f9ea86f | 3076 | * call to force_compatible_cpus_allowed_ptr(). |
07ec77a1 WD |
3077 | * |
3078 | * It is the caller's responsibility to serialise this with any calls to | |
3079 | * force_compatible_cpus_allowed_ptr(@p). | |
3080 | */ | |
3081 | void relax_compatible_cpus_allowed_ptr(struct task_struct *p) | |
3082 | { | |
713a2e21 | 3083 | struct affinity_context ac = { |
8f9ea86f WL |
3084 | .new_mask = task_user_cpus(p), |
3085 | .flags = 0, | |
713a2e21 | 3086 | }; |
8f9ea86f | 3087 | int ret; |
07ec77a1 WD |
3088 | |
3089 | /* | |
8f9ea86f WL |
3090 | * Try to restore the old affinity mask with __sched_setaffinity(). |
3091 | * Cpuset masking will be done there too. | |
07ec77a1 | 3092 | */ |
8f9ea86f WL |
3093 | ret = __sched_setaffinity(p, &ac); |
3094 | WARN_ON_ONCE(ret); | |
07ec77a1 WD |
3095 | } |
3096 | ||
dd41f596 | 3097 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 3098 | { |
e2912009 | 3099 | #ifdef CONFIG_SCHED_DEBUG |
2f064a59 PZ |
3100 | unsigned int state = READ_ONCE(p->__state); |
3101 | ||
e2912009 PZ |
3102 | /* |
3103 | * We should never call set_task_cpu() on a blocked task, | |
3104 | * ttwu() will sort out the placement. | |
3105 | */ | |
2f064a59 | 3106 | WARN_ON_ONCE(state != TASK_RUNNING && state != TASK_WAKING && !p->on_rq); |
0122ec5b | 3107 | |
3ea94de1 JP |
3108 | /* |
3109 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, | |
3110 | * because schedstat_wait_{start,end} rebase migrating task's wait_start | |
3111 | * time relying on p->on_rq. | |
3112 | */ | |
2f064a59 | 3113 | WARN_ON_ONCE(state == TASK_RUNNING && |
3ea94de1 JP |
3114 | p->sched_class == &fair_sched_class && |
3115 | (p->on_rq && !task_on_rq_migrating(p))); | |
3116 | ||
0122ec5b | 3117 | #ifdef CONFIG_LOCKDEP |
6c6c54e1 PZ |
3118 | /* |
3119 | * The caller should hold either p->pi_lock or rq->lock, when changing | |
3120 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. | |
3121 | * | |
3122 | * sched_move_task() holds both and thus holding either pins the cgroup, | |
8323f26c | 3123 | * see task_group(). |
6c6c54e1 PZ |
3124 | * |
3125 | * Furthermore, all task_rq users should acquire both locks, see | |
3126 | * task_rq_lock(). | |
3127 | */ | |
0122ec5b | 3128 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
9ef7e7e3 | 3129 | lockdep_is_held(__rq_lockp(task_rq(p))))); |
0122ec5b | 3130 | #endif |
4ff9083b PZ |
3131 | /* |
3132 | * Clearly, migrating tasks to offline CPUs is a fairly daft thing. | |
3133 | */ | |
3134 | WARN_ON_ONCE(!cpu_online(new_cpu)); | |
af449901 PZ |
3135 | |
3136 | WARN_ON_ONCE(is_migration_disabled(p)); | |
e2912009 PZ |
3137 | #endif |
3138 | ||
de1d7286 | 3139 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 3140 | |
0c69774e | 3141 | if (task_cpu(p) != new_cpu) { |
0a74bef8 | 3142 | if (p->sched_class->migrate_task_rq) |
1327237a | 3143 | p->sched_class->migrate_task_rq(p, new_cpu); |
0c69774e | 3144 | p->se.nr_migrations++; |
d7822b1e | 3145 | rseq_migrate(p); |
ff303e66 | 3146 | perf_event_task_migrate(p); |
0c69774e | 3147 | } |
dd41f596 IM |
3148 | |
3149 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
3150 | } |
3151 | ||
0ad4e3df | 3152 | #ifdef CONFIG_NUMA_BALANCING |
ac66f547 PZ |
3153 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
3154 | { | |
da0c1e65 | 3155 | if (task_on_rq_queued(p)) { |
ac66f547 | 3156 | struct rq *src_rq, *dst_rq; |
8a8c69c3 | 3157 | struct rq_flags srf, drf; |
ac66f547 PZ |
3158 | |
3159 | src_rq = task_rq(p); | |
3160 | dst_rq = cpu_rq(cpu); | |
3161 | ||
8a8c69c3 PZ |
3162 | rq_pin_lock(src_rq, &srf); |
3163 | rq_pin_lock(dst_rq, &drf); | |
3164 | ||
ac66f547 PZ |
3165 | deactivate_task(src_rq, p, 0); |
3166 | set_task_cpu(p, cpu); | |
3167 | activate_task(dst_rq, p, 0); | |
3168 | check_preempt_curr(dst_rq, p, 0); | |
8a8c69c3 PZ |
3169 | |
3170 | rq_unpin_lock(dst_rq, &drf); | |
3171 | rq_unpin_lock(src_rq, &srf); | |
3172 | ||
ac66f547 PZ |
3173 | } else { |
3174 | /* | |
3175 | * Task isn't running anymore; make it appear like we migrated | |
3176 | * it before it went to sleep. This means on wakeup we make the | |
d1ccc66d | 3177 | * previous CPU our target instead of where it really is. |
ac66f547 PZ |
3178 | */ |
3179 | p->wake_cpu = cpu; | |
3180 | } | |
3181 | } | |
3182 | ||
3183 | struct migration_swap_arg { | |
3184 | struct task_struct *src_task, *dst_task; | |
3185 | int src_cpu, dst_cpu; | |
3186 | }; | |
3187 | ||
3188 | static int migrate_swap_stop(void *data) | |
3189 | { | |
3190 | struct migration_swap_arg *arg = data; | |
3191 | struct rq *src_rq, *dst_rq; | |
3192 | int ret = -EAGAIN; | |
3193 | ||
62694cd5 PZ |
3194 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
3195 | return -EAGAIN; | |
3196 | ||
ac66f547 PZ |
3197 | src_rq = cpu_rq(arg->src_cpu); |
3198 | dst_rq = cpu_rq(arg->dst_cpu); | |
3199 | ||
74602315 PZ |
3200 | double_raw_lock(&arg->src_task->pi_lock, |
3201 | &arg->dst_task->pi_lock); | |
ac66f547 | 3202 | double_rq_lock(src_rq, dst_rq); |
62694cd5 | 3203 | |
ac66f547 PZ |
3204 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
3205 | goto unlock; | |
3206 | ||
3207 | if (task_cpu(arg->src_task) != arg->src_cpu) | |
3208 | goto unlock; | |
3209 | ||
3bd37062 | 3210 | if (!cpumask_test_cpu(arg->dst_cpu, arg->src_task->cpus_ptr)) |
ac66f547 PZ |
3211 | goto unlock; |
3212 | ||
3bd37062 | 3213 | if (!cpumask_test_cpu(arg->src_cpu, arg->dst_task->cpus_ptr)) |
ac66f547 PZ |
3214 | goto unlock; |
3215 | ||
3216 | __migrate_swap_task(arg->src_task, arg->dst_cpu); | |
3217 | __migrate_swap_task(arg->dst_task, arg->src_cpu); | |
3218 | ||
3219 | ret = 0; | |
3220 | ||
3221 | unlock: | |
3222 | double_rq_unlock(src_rq, dst_rq); | |
74602315 PZ |
3223 | raw_spin_unlock(&arg->dst_task->pi_lock); |
3224 | raw_spin_unlock(&arg->src_task->pi_lock); | |
ac66f547 PZ |
3225 | |
3226 | return ret; | |
3227 | } | |
3228 | ||
3229 | /* | |
3230 | * Cross migrate two tasks | |
3231 | */ | |
0ad4e3df SD |
3232 | int migrate_swap(struct task_struct *cur, struct task_struct *p, |
3233 | int target_cpu, int curr_cpu) | |
ac66f547 PZ |
3234 | { |
3235 | struct migration_swap_arg arg; | |
3236 | int ret = -EINVAL; | |
3237 | ||
ac66f547 PZ |
3238 | arg = (struct migration_swap_arg){ |
3239 | .src_task = cur, | |
0ad4e3df | 3240 | .src_cpu = curr_cpu, |
ac66f547 | 3241 | .dst_task = p, |
0ad4e3df | 3242 | .dst_cpu = target_cpu, |
ac66f547 PZ |
3243 | }; |
3244 | ||
3245 | if (arg.src_cpu == arg.dst_cpu) | |
3246 | goto out; | |
3247 | ||
6acce3ef PZ |
3248 | /* |
3249 | * These three tests are all lockless; this is OK since all of them | |
3250 | * will be re-checked with proper locks held further down the line. | |
3251 | */ | |
ac66f547 PZ |
3252 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
3253 | goto out; | |
3254 | ||
3bd37062 | 3255 | if (!cpumask_test_cpu(arg.dst_cpu, arg.src_task->cpus_ptr)) |
ac66f547 PZ |
3256 | goto out; |
3257 | ||
3bd37062 | 3258 | if (!cpumask_test_cpu(arg.src_cpu, arg.dst_task->cpus_ptr)) |
ac66f547 PZ |
3259 | goto out; |
3260 | ||
286549dc | 3261 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
ac66f547 PZ |
3262 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
3263 | ||
3264 | out: | |
ac66f547 PZ |
3265 | return ret; |
3266 | } | |
0ad4e3df | 3267 | #endif /* CONFIG_NUMA_BALANCING */ |
ac66f547 | 3268 | |
1da177e4 LT |
3269 | /* |
3270 | * wait_task_inactive - wait for a thread to unschedule. | |
3271 | * | |
f9fc8cad PZ |
3272 | * Wait for the thread to block in any of the states set in @match_state. |
3273 | * If it changes, i.e. @p might have woken up, then return zero. When we | |
3274 | * succeed in waiting for @p to be off its CPU, we return a positive number | |
3275 | * (its total switch count). If a second call a short while later returns the | |
3276 | * same number, the caller can be sure that @p has remained unscheduled the | |
3277 | * whole time. | |
85ba2d86 | 3278 | * |
1da177e4 LT |
3279 | * The caller must ensure that the task *will* unschedule sometime soon, |
3280 | * else this function might spin for a *long* time. This function can't | |
3281 | * be called with interrupts off, or it may introduce deadlock with | |
3282 | * smp_call_function() if an IPI is sent by the same process we are | |
3283 | * waiting to become inactive. | |
3284 | */ | |
2f064a59 | 3285 | unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state) |
1da177e4 | 3286 | { |
da0c1e65 | 3287 | int running, queued; |
eb580751 | 3288 | struct rq_flags rf; |
85ba2d86 | 3289 | unsigned long ncsw; |
70b97a7f | 3290 | struct rq *rq; |
1da177e4 | 3291 | |
3a5c359a AK |
3292 | for (;;) { |
3293 | /* | |
3294 | * We do the initial early heuristics without holding | |
3295 | * any task-queue locks at all. We'll only try to get | |
3296 | * the runqueue lock when things look like they will | |
3297 | * work out! | |
3298 | */ | |
3299 | rq = task_rq(p); | |
fa490cfd | 3300 | |
3a5c359a AK |
3301 | /* |
3302 | * If the task is actively running on another CPU | |
3303 | * still, just relax and busy-wait without holding | |
3304 | * any locks. | |
3305 | * | |
3306 | * NOTE! Since we don't hold any locks, it's not | |
3307 | * even sure that "rq" stays as the right runqueue! | |
0b9d46fc | 3308 | * But we don't care, since "task_on_cpu()" will |
3a5c359a AK |
3309 | * return false if the runqueue has changed and p |
3310 | * is actually now running somewhere else! | |
3311 | */ | |
0b9d46fc | 3312 | while (task_on_cpu(rq, p)) { |
f9fc8cad | 3313 | if (!(READ_ONCE(p->__state) & match_state)) |
85ba2d86 | 3314 | return 0; |
3a5c359a | 3315 | cpu_relax(); |
85ba2d86 | 3316 | } |
fa490cfd | 3317 | |
3a5c359a AK |
3318 | /* |
3319 | * Ok, time to look more closely! We need the rq | |
3320 | * lock now, to be *sure*. If we're wrong, we'll | |
3321 | * just go back and repeat. | |
3322 | */ | |
eb580751 | 3323 | rq = task_rq_lock(p, &rf); |
27a9da65 | 3324 | trace_sched_wait_task(p); |
0b9d46fc | 3325 | running = task_on_cpu(rq, p); |
da0c1e65 | 3326 | queued = task_on_rq_queued(p); |
85ba2d86 | 3327 | ncsw = 0; |
f9fc8cad | 3328 | if (READ_ONCE(p->__state) & match_state) |
93dcf55f | 3329 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
eb580751 | 3330 | task_rq_unlock(rq, p, &rf); |
fa490cfd | 3331 | |
85ba2d86 RM |
3332 | /* |
3333 | * If it changed from the expected state, bail out now. | |
3334 | */ | |
3335 | if (unlikely(!ncsw)) | |
3336 | break; | |
3337 | ||
3a5c359a AK |
3338 | /* |
3339 | * Was it really running after all now that we | |
3340 | * checked with the proper locks actually held? | |
3341 | * | |
3342 | * Oops. Go back and try again.. | |
3343 | */ | |
3344 | if (unlikely(running)) { | |
3345 | cpu_relax(); | |
3346 | continue; | |
3347 | } | |
fa490cfd | 3348 | |
3a5c359a AK |
3349 | /* |
3350 | * It's not enough that it's not actively running, | |
3351 | * it must be off the runqueue _entirely_, and not | |
3352 | * preempted! | |
3353 | * | |
80dd99b3 | 3354 | * So if it was still runnable (but just not actively |
3a5c359a AK |
3355 | * running right now), it's preempted, and we should |
3356 | * yield - it could be a while. | |
3357 | */ | |
da0c1e65 | 3358 | if (unlikely(queued)) { |
8b0e1953 | 3359 | ktime_t to = NSEC_PER_SEC / HZ; |
8eb90c30 TG |
3360 | |
3361 | set_current_state(TASK_UNINTERRUPTIBLE); | |
c33627e9 | 3362 | schedule_hrtimeout(&to, HRTIMER_MODE_REL_HARD); |
3a5c359a AK |
3363 | continue; |
3364 | } | |
fa490cfd | 3365 | |
3a5c359a AK |
3366 | /* |
3367 | * Ahh, all good. It wasn't running, and it wasn't | |
3368 | * runnable, which means that it will never become | |
3369 | * running in the future either. We're all done! | |
3370 | */ | |
3371 | break; | |
3372 | } | |
85ba2d86 RM |
3373 | |
3374 | return ncsw; | |
1da177e4 LT |
3375 | } |
3376 | ||
3377 | /*** | |
3378 | * kick_process - kick a running thread to enter/exit the kernel | |
3379 | * @p: the to-be-kicked thread | |
3380 | * | |
3381 | * Cause a process which is running on another CPU to enter | |
3382 | * kernel-mode, without any delay. (to get signals handled.) | |
3383 | * | |
25985edc | 3384 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
3385 | * because all it wants to ensure is that the remote task enters |
3386 | * the kernel. If the IPI races and the task has been migrated | |
3387 | * to another CPU then no harm is done and the purpose has been | |
3388 | * achieved as well. | |
3389 | */ | |
36c8b586 | 3390 | void kick_process(struct task_struct *p) |
1da177e4 LT |
3391 | { |
3392 | int cpu; | |
3393 | ||
3394 | preempt_disable(); | |
3395 | cpu = task_cpu(p); | |
3396 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
3397 | smp_send_reschedule(cpu); | |
3398 | preempt_enable(); | |
3399 | } | |
b43e3521 | 3400 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 | 3401 | |
30da688e | 3402 | /* |
3bd37062 | 3403 | * ->cpus_ptr is protected by both rq->lock and p->pi_lock |
e9d867a6 PZI |
3404 | * |
3405 | * A few notes on cpu_active vs cpu_online: | |
3406 | * | |
3407 | * - cpu_active must be a subset of cpu_online | |
3408 | * | |
97fb7a0a | 3409 | * - on CPU-up we allow per-CPU kthreads on the online && !active CPU, |
e9d867a6 | 3410 | * see __set_cpus_allowed_ptr(). At this point the newly online |
d1ccc66d | 3411 | * CPU isn't yet part of the sched domains, and balancing will not |
e9d867a6 PZI |
3412 | * see it. |
3413 | * | |
d1ccc66d | 3414 | * - on CPU-down we clear cpu_active() to mask the sched domains and |
e9d867a6 | 3415 | * avoid the load balancer to place new tasks on the to be removed |
d1ccc66d | 3416 | * CPU. Existing tasks will remain running there and will be taken |
e9d867a6 PZI |
3417 | * off. |
3418 | * | |
3419 | * This means that fallback selection must not select !active CPUs. | |
3420 | * And can assume that any active CPU must be online. Conversely | |
3421 | * select_task_rq() below may allow selection of !active CPUs in order | |
3422 | * to satisfy the above rules. | |
30da688e | 3423 | */ |
5da9a0fb PZ |
3424 | static int select_fallback_rq(int cpu, struct task_struct *p) |
3425 | { | |
aa00d89c TC |
3426 | int nid = cpu_to_node(cpu); |
3427 | const struct cpumask *nodemask = NULL; | |
2baab4e9 PZ |
3428 | enum { cpuset, possible, fail } state = cpuset; |
3429 | int dest_cpu; | |
5da9a0fb | 3430 | |
aa00d89c | 3431 | /* |
d1ccc66d IM |
3432 | * If the node that the CPU is on has been offlined, cpu_to_node() |
3433 | * will return -1. There is no CPU on the node, and we should | |
3434 | * select the CPU on the other node. | |
aa00d89c TC |
3435 | */ |
3436 | if (nid != -1) { | |
3437 | nodemask = cpumask_of_node(nid); | |
3438 | ||
3439 | /* Look for allowed, online CPU in same node. */ | |
3440 | for_each_cpu(dest_cpu, nodemask) { | |
9ae606bc | 3441 | if (is_cpu_allowed(p, dest_cpu)) |
aa00d89c TC |
3442 | return dest_cpu; |
3443 | } | |
2baab4e9 | 3444 | } |
5da9a0fb | 3445 | |
2baab4e9 PZ |
3446 | for (;;) { |
3447 | /* Any allowed, online CPU? */ | |
3bd37062 | 3448 | for_each_cpu(dest_cpu, p->cpus_ptr) { |
175f0e25 | 3449 | if (!is_cpu_allowed(p, dest_cpu)) |
2baab4e9 | 3450 | continue; |
175f0e25 | 3451 | |
2baab4e9 PZ |
3452 | goto out; |
3453 | } | |
5da9a0fb | 3454 | |
e73e85f0 | 3455 | /* No more Mr. Nice Guy. */ |
2baab4e9 PZ |
3456 | switch (state) { |
3457 | case cpuset: | |
97c0054d | 3458 | if (cpuset_cpus_allowed_fallback(p)) { |
e73e85f0 ON |
3459 | state = possible; |
3460 | break; | |
3461 | } | |
df561f66 | 3462 | fallthrough; |
2baab4e9 | 3463 | case possible: |
af449901 PZ |
3464 | /* |
3465 | * XXX When called from select_task_rq() we only | |
3466 | * hold p->pi_lock and again violate locking order. | |
3467 | * | |
3468 | * More yuck to audit. | |
3469 | */ | |
9ae606bc | 3470 | do_set_cpus_allowed(p, task_cpu_possible_mask(p)); |
2baab4e9 PZ |
3471 | state = fail; |
3472 | break; | |
2baab4e9 PZ |
3473 | case fail: |
3474 | BUG(); | |
3475 | break; | |
3476 | } | |
3477 | } | |
3478 | ||
3479 | out: | |
3480 | if (state != cpuset) { | |
3481 | /* | |
3482 | * Don't tell them about moving exiting tasks or | |
3483 | * kernel threads (both mm NULL), since they never | |
3484 | * leave kernel. | |
3485 | */ | |
3486 | if (p->mm && printk_ratelimit()) { | |
aac74dc4 | 3487 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
2baab4e9 PZ |
3488 | task_pid_nr(p), p->comm, cpu); |
3489 | } | |
5da9a0fb PZ |
3490 | } |
3491 | ||
3492 | return dest_cpu; | |
3493 | } | |
3494 | ||
e2912009 | 3495 | /* |
3bd37062 | 3496 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_ptr is stable. |
e2912009 | 3497 | */ |
970b13ba | 3498 | static inline |
3aef1551 | 3499 | int select_task_rq(struct task_struct *p, int cpu, int wake_flags) |
970b13ba | 3500 | { |
cbce1a68 PZ |
3501 | lockdep_assert_held(&p->pi_lock); |
3502 | ||
af449901 | 3503 | if (p->nr_cpus_allowed > 1 && !is_migration_disabled(p)) |
3aef1551 | 3504 | cpu = p->sched_class->select_task_rq(p, cpu, wake_flags); |
e9d867a6 | 3505 | else |
3bd37062 | 3506 | cpu = cpumask_any(p->cpus_ptr); |
e2912009 PZ |
3507 | |
3508 | /* | |
3509 | * In order not to call set_task_cpu() on a blocking task we need | |
3bd37062 | 3510 | * to rely on ttwu() to place the task on a valid ->cpus_ptr |
d1ccc66d | 3511 | * CPU. |
e2912009 PZ |
3512 | * |
3513 | * Since this is common to all placement strategies, this lives here. | |
3514 | * | |
3515 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
3516 | * not worry about this generic constraint ] | |
3517 | */ | |
7af443ee | 3518 | if (unlikely(!is_cpu_allowed(p, cpu))) |
5da9a0fb | 3519 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
3520 | |
3521 | return cpu; | |
970b13ba | 3522 | } |
09a40af5 | 3523 | |
f5832c19 NP |
3524 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
3525 | { | |
ded467dc | 3526 | static struct lock_class_key stop_pi_lock; |
f5832c19 NP |
3527 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
3528 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
3529 | ||
3530 | if (stop) { | |
3531 | /* | |
3532 | * Make it appear like a SCHED_FIFO task, its something | |
3533 | * userspace knows about and won't get confused about. | |
3534 | * | |
3535 | * Also, it will make PI more or less work without too | |
3536 | * much confusion -- but then, stop work should not | |
3537 | * rely on PI working anyway. | |
3538 | */ | |
3539 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
3540 | ||
3541 | stop->sched_class = &stop_sched_class; | |
ded467dc PZ |
3542 | |
3543 | /* | |
3544 | * The PI code calls rt_mutex_setprio() with ->pi_lock held to | |
3545 | * adjust the effective priority of a task. As a result, | |
3546 | * rt_mutex_setprio() can trigger (RT) balancing operations, | |
3547 | * which can then trigger wakeups of the stop thread to push | |
3548 | * around the current task. | |
3549 | * | |
3550 | * The stop task itself will never be part of the PI-chain, it | |
3551 | * never blocks, therefore that ->pi_lock recursion is safe. | |
3552 | * Tell lockdep about this by placing the stop->pi_lock in its | |
3553 | * own class. | |
3554 | */ | |
3555 | lockdep_set_class(&stop->pi_lock, &stop_pi_lock); | |
f5832c19 NP |
3556 | } |
3557 | ||
3558 | cpu_rq(cpu)->stop = stop; | |
3559 | ||
3560 | if (old_stop) { | |
3561 | /* | |
3562 | * Reset it back to a normal scheduling class so that | |
3563 | * it can die in pieces. | |
3564 | */ | |
3565 | old_stop->sched_class = &rt_sched_class; | |
3566 | } | |
3567 | } | |
3568 | ||
74d862b6 | 3569 | #else /* CONFIG_SMP */ |
25834c73 PZ |
3570 | |
3571 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, | |
713a2e21 | 3572 | struct affinity_context *ctx) |
25834c73 | 3573 | { |
713a2e21 | 3574 | return set_cpus_allowed_ptr(p, ctx->new_mask); |
25834c73 PZ |
3575 | } |
3576 | ||
af449901 PZ |
3577 | static inline void migrate_disable_switch(struct rq *rq, struct task_struct *p) { } |
3578 | ||
3015ef4b TG |
3579 | static inline bool rq_has_pinned_tasks(struct rq *rq) |
3580 | { | |
3581 | return false; | |
3582 | } | |
3583 | ||
74d862b6 | 3584 | #endif /* !CONFIG_SMP */ |
970b13ba | 3585 | |
d7c01d27 | 3586 | static void |
b84cb5df | 3587 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 3588 | { |
4fa8d299 | 3589 | struct rq *rq; |
b84cb5df | 3590 | |
4fa8d299 JP |
3591 | if (!schedstat_enabled()) |
3592 | return; | |
3593 | ||
3594 | rq = this_rq(); | |
d7c01d27 | 3595 | |
4fa8d299 JP |
3596 | #ifdef CONFIG_SMP |
3597 | if (cpu == rq->cpu) { | |
b85c8b71 | 3598 | __schedstat_inc(rq->ttwu_local); |
ceeadb83 | 3599 | __schedstat_inc(p->stats.nr_wakeups_local); |
d7c01d27 PZ |
3600 | } else { |
3601 | struct sched_domain *sd; | |
3602 | ||
ceeadb83 | 3603 | __schedstat_inc(p->stats.nr_wakeups_remote); |
057f3fad | 3604 | rcu_read_lock(); |
4fa8d299 | 3605 | for_each_domain(rq->cpu, sd) { |
d7c01d27 | 3606 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
b85c8b71 | 3607 | __schedstat_inc(sd->ttwu_wake_remote); |
d7c01d27 PZ |
3608 | break; |
3609 | } | |
3610 | } | |
057f3fad | 3611 | rcu_read_unlock(); |
d7c01d27 | 3612 | } |
f339b9dc PZ |
3613 | |
3614 | if (wake_flags & WF_MIGRATED) | |
ceeadb83 | 3615 | __schedstat_inc(p->stats.nr_wakeups_migrate); |
d7c01d27 PZ |
3616 | #endif /* CONFIG_SMP */ |
3617 | ||
b85c8b71 | 3618 | __schedstat_inc(rq->ttwu_count); |
ceeadb83 | 3619 | __schedstat_inc(p->stats.nr_wakeups); |
d7c01d27 PZ |
3620 | |
3621 | if (wake_flags & WF_SYNC) | |
ceeadb83 | 3622 | __schedstat_inc(p->stats.nr_wakeups_sync); |
d7c01d27 PZ |
3623 | } |
3624 | ||
23f41eeb PZ |
3625 | /* |
3626 | * Mark the task runnable and perform wakeup-preemption. | |
3627 | */ | |
e7904a28 | 3628 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3629 | struct rq_flags *rf) |
9ed3811a | 3630 | { |
9ed3811a | 3631 | check_preempt_curr(rq, p, wake_flags); |
2f064a59 | 3632 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fbd705a0 PZ |
3633 | trace_sched_wakeup(p); |
3634 | ||
9ed3811a | 3635 | #ifdef CONFIG_SMP |
4c9a4bc8 PZ |
3636 | if (p->sched_class->task_woken) { |
3637 | /* | |
b19a888c | 3638 | * Our task @p is fully woken up and running; so it's safe to |
cbce1a68 | 3639 | * drop the rq->lock, hereafter rq is only used for statistics. |
4c9a4bc8 | 3640 | */ |
d8ac8971 | 3641 | rq_unpin_lock(rq, rf); |
9ed3811a | 3642 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 3643 | rq_repin_lock(rq, rf); |
4c9a4bc8 | 3644 | } |
9ed3811a | 3645 | |
e69c6341 | 3646 | if (rq->idle_stamp) { |
78becc27 | 3647 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
9bd721c5 | 3648 | u64 max = 2*rq->max_idle_balance_cost; |
9ed3811a | 3649 | |
abfafa54 JL |
3650 | update_avg(&rq->avg_idle, delta); |
3651 | ||
3652 | if (rq->avg_idle > max) | |
9ed3811a | 3653 | rq->avg_idle = max; |
abfafa54 | 3654 | |
94aafc3e PZ |
3655 | rq->wake_stamp = jiffies; |
3656 | rq->wake_avg_idle = rq->avg_idle / 2; | |
3657 | ||
9ed3811a TH |
3658 | rq->idle_stamp = 0; |
3659 | } | |
3660 | #endif | |
3661 | } | |
3662 | ||
c05fbafb | 3663 | static void |
e7904a28 | 3664 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
d8ac8971 | 3665 | struct rq_flags *rf) |
c05fbafb | 3666 | { |
77558e4d | 3667 | int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK; |
b5179ac7 | 3668 | |
5cb9eaa3 | 3669 | lockdep_assert_rq_held(rq); |
cbce1a68 | 3670 | |
c05fbafb PZ |
3671 | if (p->sched_contributes_to_load) |
3672 | rq->nr_uninterruptible--; | |
b5179ac7 | 3673 | |
dbfb089d | 3674 | #ifdef CONFIG_SMP |
b5179ac7 | 3675 | if (wake_flags & WF_MIGRATED) |
59efa0ba | 3676 | en_flags |= ENQUEUE_MIGRATED; |
ec618b84 | 3677 | else |
c05fbafb | 3678 | #endif |
ec618b84 PZ |
3679 | if (p->in_iowait) { |
3680 | delayacct_blkio_end(p); | |
3681 | atomic_dec(&task_rq(p)->nr_iowait); | |
3682 | } | |
c05fbafb | 3683 | |
1b174a2c | 3684 | activate_task(rq, p, en_flags); |
d8ac8971 | 3685 | ttwu_do_wakeup(rq, p, wake_flags, rf); |
c05fbafb PZ |
3686 | } |
3687 | ||
3688 | /* | |
58877d34 PZ |
3689 | * Consider @p being inside a wait loop: |
3690 | * | |
3691 | * for (;;) { | |
3692 | * set_current_state(TASK_UNINTERRUPTIBLE); | |
3693 | * | |
3694 | * if (CONDITION) | |
3695 | * break; | |
3696 | * | |
3697 | * schedule(); | |
3698 | * } | |
3699 | * __set_current_state(TASK_RUNNING); | |
3700 | * | |
3701 | * between set_current_state() and schedule(). In this case @p is still | |
3702 | * runnable, so all that needs doing is change p->state back to TASK_RUNNING in | |
3703 | * an atomic manner. | |
3704 | * | |
3705 | * By taking task_rq(p)->lock we serialize against schedule(), if @p->on_rq | |
3706 | * then schedule() must still happen and p->state can be changed to | |
3707 | * TASK_RUNNING. Otherwise we lost the race, schedule() has happened, and we | |
3708 | * need to do a full wakeup with enqueue. | |
3709 | * | |
3710 | * Returns: %true when the wakeup is done, | |
3711 | * %false otherwise. | |
c05fbafb | 3712 | */ |
58877d34 | 3713 | static int ttwu_runnable(struct task_struct *p, int wake_flags) |
c05fbafb | 3714 | { |
eb580751 | 3715 | struct rq_flags rf; |
c05fbafb PZ |
3716 | struct rq *rq; |
3717 | int ret = 0; | |
3718 | ||
eb580751 | 3719 | rq = __task_rq_lock(p, &rf); |
da0c1e65 | 3720 | if (task_on_rq_queued(p)) { |
1ad4ec0d FW |
3721 | /* check_preempt_curr() may use rq clock */ |
3722 | update_rq_clock(rq); | |
d8ac8971 | 3723 | ttwu_do_wakeup(rq, p, wake_flags, &rf); |
c05fbafb PZ |
3724 | ret = 1; |
3725 | } | |
eb580751 | 3726 | __task_rq_unlock(rq, &rf); |
c05fbafb PZ |
3727 | |
3728 | return ret; | |
3729 | } | |
3730 | ||
317f3941 | 3731 | #ifdef CONFIG_SMP |
a1488664 | 3732 | void sched_ttwu_pending(void *arg) |
317f3941 | 3733 | { |
a1488664 | 3734 | struct llist_node *llist = arg; |
317f3941 | 3735 | struct rq *rq = this_rq(); |
73215849 | 3736 | struct task_struct *p, *t; |
d8ac8971 | 3737 | struct rq_flags rf; |
317f3941 | 3738 | |
e3baac47 PZ |
3739 | if (!llist) |
3740 | return; | |
3741 | ||
8a8c69c3 | 3742 | rq_lock_irqsave(rq, &rf); |
77558e4d | 3743 | update_rq_clock(rq); |
317f3941 | 3744 | |
8c4890d1 | 3745 | llist_for_each_entry_safe(p, t, llist, wake_entry.llist) { |
b6e13e85 PZ |
3746 | if (WARN_ON_ONCE(p->on_cpu)) |
3747 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
3748 | ||
3749 | if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq))) | |
3750 | set_task_cpu(p, cpu_of(rq)); | |
3751 | ||
73215849 | 3752 | ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf); |
b6e13e85 | 3753 | } |
317f3941 | 3754 | |
d6962c4f TD |
3755 | /* |
3756 | * Must be after enqueueing at least once task such that | |
3757 | * idle_cpu() does not observe a false-negative -- if it does, | |
3758 | * it is possible for select_idle_siblings() to stack a number | |
3759 | * of tasks on this CPU during that window. | |
3760 | * | |
3761 | * It is ok to clear ttwu_pending when another task pending. | |
3762 | * We will receive IPI after local irq enabled and then enqueue it. | |
3763 | * Since now nr_running > 0, idle_cpu() will always get correct result. | |
3764 | */ | |
3765 | WRITE_ONCE(rq->ttwu_pending, 0); | |
8a8c69c3 | 3766 | rq_unlock_irqrestore(rq, &rf); |
317f3941 PZ |
3767 | } |
3768 | ||
b2a02fc4 | 3769 | void send_call_function_single_ipi(int cpu) |
317f3941 | 3770 | { |
b2a02fc4 | 3771 | struct rq *rq = cpu_rq(cpu); |
ca38062e | 3772 | |
b2a02fc4 PZ |
3773 | if (!set_nr_if_polling(rq->idle)) |
3774 | arch_send_call_function_single_ipi(cpu); | |
3775 | else | |
3776 | trace_sched_wake_idle_without_ipi(cpu); | |
317f3941 PZ |
3777 | } |
3778 | ||
2ebb1771 MG |
3779 | /* |
3780 | * Queue a task on the target CPUs wake_list and wake the CPU via IPI if | |
3781 | * necessary. The wakee CPU on receipt of the IPI will queue the task | |
3782 | * via sched_ttwu_wakeup() for activation so the wakee incurs the cost | |
3783 | * of the wakeup instead of the waker. | |
3784 | */ | |
3785 | static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
317f3941 | 3786 | { |
e3baac47 PZ |
3787 | struct rq *rq = cpu_rq(cpu); |
3788 | ||
b7e7ade3 PZ |
3789 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
3790 | ||
126c2092 | 3791 | WRITE_ONCE(rq->ttwu_pending, 1); |
8c4890d1 | 3792 | __smp_call_single_queue(cpu, &p->wake_entry.llist); |
317f3941 | 3793 | } |
d6aa8f85 | 3794 | |
f6be8af1 CL |
3795 | void wake_up_if_idle(int cpu) |
3796 | { | |
3797 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 3798 | struct rq_flags rf; |
f6be8af1 | 3799 | |
fd7de1e8 AL |
3800 | rcu_read_lock(); |
3801 | ||
3802 | if (!is_idle_task(rcu_dereference(rq->curr))) | |
3803 | goto out; | |
f6be8af1 | 3804 | |
8850cb66 PZ |
3805 | rq_lock_irqsave(rq, &rf); |
3806 | if (is_idle_task(rq->curr)) | |
3807 | resched_curr(rq); | |
3808 | /* Else CPU is not idle, do nothing here: */ | |
3809 | rq_unlock_irqrestore(rq, &rf); | |
fd7de1e8 AL |
3810 | |
3811 | out: | |
3812 | rcu_read_unlock(); | |
f6be8af1 CL |
3813 | } |
3814 | ||
39be3501 | 3815 | bool cpus_share_cache(int this_cpu, int that_cpu) |
518cd623 | 3816 | { |
42dc938a VD |
3817 | if (this_cpu == that_cpu) |
3818 | return true; | |
3819 | ||
518cd623 PZ |
3820 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); |
3821 | } | |
c6e7bd7a | 3822 | |
751d4cbc | 3823 | static inline bool ttwu_queue_cond(struct task_struct *p, int cpu) |
2ebb1771 | 3824 | { |
5ba2ffba PZ |
3825 | /* |
3826 | * Do not complicate things with the async wake_list while the CPU is | |
3827 | * in hotplug state. | |
3828 | */ | |
3829 | if (!cpu_active(cpu)) | |
3830 | return false; | |
3831 | ||
751d4cbc MG |
3832 | /* Ensure the task will still be allowed to run on the CPU. */ |
3833 | if (!cpumask_test_cpu(cpu, p->cpus_ptr)) | |
3834 | return false; | |
3835 | ||
2ebb1771 MG |
3836 | /* |
3837 | * If the CPU does not share cache, then queue the task on the | |
3838 | * remote rqs wakelist to avoid accessing remote data. | |
3839 | */ | |
3840 | if (!cpus_share_cache(smp_processor_id(), cpu)) | |
3841 | return true; | |
3842 | ||
f3dd3f67 TD |
3843 | if (cpu == smp_processor_id()) |
3844 | return false; | |
3845 | ||
2ebb1771 | 3846 | /* |
f3dd3f67 TD |
3847 | * If the wakee cpu is idle, or the task is descheduling and the |
3848 | * only running task on the CPU, then use the wakelist to offload | |
3849 | * the task activation to the idle (or soon-to-be-idle) CPU as | |
3850 | * the current CPU is likely busy. nr_running is checked to | |
3851 | * avoid unnecessary task stacking. | |
28156108 TD |
3852 | * |
3853 | * Note that we can only get here with (wakee) p->on_rq=0, | |
3854 | * p->on_cpu can be whatever, we've done the dequeue, so | |
3855 | * the wakee has been accounted out of ->nr_running. | |
2ebb1771 | 3856 | */ |
f3dd3f67 | 3857 | if (!cpu_rq(cpu)->nr_running) |
2ebb1771 MG |
3858 | return true; |
3859 | ||
3860 | return false; | |
3861 | } | |
3862 | ||
3863 | static bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
c6e7bd7a | 3864 | { |
751d4cbc | 3865 | if (sched_feat(TTWU_QUEUE) && ttwu_queue_cond(p, cpu)) { |
c6e7bd7a | 3866 | sched_clock_cpu(cpu); /* Sync clocks across CPUs */ |
2ebb1771 | 3867 | __ttwu_queue_wakelist(p, cpu, wake_flags); |
c6e7bd7a PZ |
3868 | return true; |
3869 | } | |
3870 | ||
3871 | return false; | |
3872 | } | |
58877d34 PZ |
3873 | |
3874 | #else /* !CONFIG_SMP */ | |
3875 | ||
3876 | static inline bool ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags) | |
3877 | { | |
3878 | return false; | |
3879 | } | |
3880 | ||
d6aa8f85 | 3881 | #endif /* CONFIG_SMP */ |
317f3941 | 3882 | |
b5179ac7 | 3883 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
c05fbafb PZ |
3884 | { |
3885 | struct rq *rq = cpu_rq(cpu); | |
d8ac8971 | 3886 | struct rq_flags rf; |
c05fbafb | 3887 | |
2ebb1771 | 3888 | if (ttwu_queue_wakelist(p, cpu, wake_flags)) |
317f3941 | 3889 | return; |
317f3941 | 3890 | |
8a8c69c3 | 3891 | rq_lock(rq, &rf); |
77558e4d | 3892 | update_rq_clock(rq); |
d8ac8971 | 3893 | ttwu_do_activate(rq, p, wake_flags, &rf); |
8a8c69c3 | 3894 | rq_unlock(rq, &rf); |
9ed3811a TH |
3895 | } |
3896 | ||
43295d73 TG |
3897 | /* |
3898 | * Invoked from try_to_wake_up() to check whether the task can be woken up. | |
3899 | * | |
3900 | * The caller holds p::pi_lock if p != current or has preemption | |
3901 | * disabled when p == current. | |
5f220be2 TG |
3902 | * |
3903 | * The rules of PREEMPT_RT saved_state: | |
3904 | * | |
3905 | * The related locking code always holds p::pi_lock when updating | |
3906 | * p::saved_state, which means the code is fully serialized in both cases. | |
3907 | * | |
3908 | * The lock wait and lock wakeups happen via TASK_RTLOCK_WAIT. No other | |
3909 | * bits set. This allows to distinguish all wakeup scenarios. | |
43295d73 TG |
3910 | */ |
3911 | static __always_inline | |
3912 | bool ttwu_state_match(struct task_struct *p, unsigned int state, int *success) | |
3913 | { | |
5f220be2 TG |
3914 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)) { |
3915 | WARN_ON_ONCE((state & TASK_RTLOCK_WAIT) && | |
3916 | state != TASK_RTLOCK_WAIT); | |
3917 | } | |
3918 | ||
43295d73 TG |
3919 | if (READ_ONCE(p->__state) & state) { |
3920 | *success = 1; | |
3921 | return true; | |
3922 | } | |
5f220be2 TG |
3923 | |
3924 | #ifdef CONFIG_PREEMPT_RT | |
3925 | /* | |
3926 | * Saved state preserves the task state across blocking on | |
3927 | * an RT lock. If the state matches, set p::saved_state to | |
3928 | * TASK_RUNNING, but do not wake the task because it waits | |
3929 | * for a lock wakeup. Also indicate success because from | |
3930 | * the regular waker's point of view this has succeeded. | |
3931 | * | |
3932 | * After acquiring the lock the task will restore p::__state | |
3933 | * from p::saved_state which ensures that the regular | |
3934 | * wakeup is not lost. The restore will also set | |
3935 | * p::saved_state to TASK_RUNNING so any further tests will | |
3936 | * not result in false positives vs. @success | |
3937 | */ | |
3938 | if (p->saved_state & state) { | |
3939 | p->saved_state = TASK_RUNNING; | |
3940 | *success = 1; | |
3941 | } | |
3942 | #endif | |
43295d73 TG |
3943 | return false; |
3944 | } | |
3945 | ||
8643cda5 PZ |
3946 | /* |
3947 | * Notes on Program-Order guarantees on SMP systems. | |
3948 | * | |
3949 | * MIGRATION | |
3950 | * | |
3951 | * The basic program-order guarantee on SMP systems is that when a task [t] | |
d1ccc66d IM |
3952 | * migrates, all its activity on its old CPU [c0] happens-before any subsequent |
3953 | * execution on its new CPU [c1]. | |
8643cda5 PZ |
3954 | * |
3955 | * For migration (of runnable tasks) this is provided by the following means: | |
3956 | * | |
3957 | * A) UNLOCK of the rq(c0)->lock scheduling out task t | |
3958 | * B) migration for t is required to synchronize *both* rq(c0)->lock and | |
3959 | * rq(c1)->lock (if not at the same time, then in that order). | |
3960 | * C) LOCK of the rq(c1)->lock scheduling in task | |
3961 | * | |
7696f991 | 3962 | * Release/acquire chaining guarantees that B happens after A and C after B. |
d1ccc66d | 3963 | * Note: the CPU doing B need not be c0 or c1 |
8643cda5 PZ |
3964 | * |
3965 | * Example: | |
3966 | * | |
3967 | * CPU0 CPU1 CPU2 | |
3968 | * | |
3969 | * LOCK rq(0)->lock | |
3970 | * sched-out X | |
3971 | * sched-in Y | |
3972 | * UNLOCK rq(0)->lock | |
3973 | * | |
3974 | * LOCK rq(0)->lock // orders against CPU0 | |
3975 | * dequeue X | |
3976 | * UNLOCK rq(0)->lock | |
3977 | * | |
3978 | * LOCK rq(1)->lock | |
3979 | * enqueue X | |
3980 | * UNLOCK rq(1)->lock | |
3981 | * | |
3982 | * LOCK rq(1)->lock // orders against CPU2 | |
3983 | * sched-out Z | |
3984 | * sched-in X | |
3985 | * UNLOCK rq(1)->lock | |
3986 | * | |
3987 | * | |
3988 | * BLOCKING -- aka. SLEEP + WAKEUP | |
3989 | * | |
3990 | * For blocking we (obviously) need to provide the same guarantee as for | |
3991 | * migration. However the means are completely different as there is no lock | |
3992 | * chain to provide order. Instead we do: | |
3993 | * | |
58877d34 PZ |
3994 | * 1) smp_store_release(X->on_cpu, 0) -- finish_task() |
3995 | * 2) smp_cond_load_acquire(!X->on_cpu) -- try_to_wake_up() | |
8643cda5 PZ |
3996 | * |
3997 | * Example: | |
3998 | * | |
3999 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) | |
4000 | * | |
4001 | * LOCK rq(0)->lock LOCK X->pi_lock | |
4002 | * dequeue X | |
4003 | * sched-out X | |
4004 | * smp_store_release(X->on_cpu, 0); | |
4005 | * | |
1f03e8d2 | 4006 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
8643cda5 PZ |
4007 | * X->state = WAKING |
4008 | * set_task_cpu(X,2) | |
4009 | * | |
4010 | * LOCK rq(2)->lock | |
4011 | * enqueue X | |
4012 | * X->state = RUNNING | |
4013 | * UNLOCK rq(2)->lock | |
4014 | * | |
4015 | * LOCK rq(2)->lock // orders against CPU1 | |
4016 | * sched-out Z | |
4017 | * sched-in X | |
4018 | * UNLOCK rq(2)->lock | |
4019 | * | |
4020 | * UNLOCK X->pi_lock | |
4021 | * UNLOCK rq(0)->lock | |
4022 | * | |
4023 | * | |
7696f991 AP |
4024 | * However, for wakeups there is a second guarantee we must provide, namely we |
4025 | * must ensure that CONDITION=1 done by the caller can not be reordered with | |
4026 | * accesses to the task state; see try_to_wake_up() and set_current_state(). | |
8643cda5 PZ |
4027 | */ |
4028 | ||
9ed3811a | 4029 | /** |
1da177e4 | 4030 | * try_to_wake_up - wake up a thread |
9ed3811a | 4031 | * @p: the thread to be awakened |
1da177e4 | 4032 | * @state: the mask of task states that can be woken |
9ed3811a | 4033 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 | 4034 | * |
58877d34 PZ |
4035 | * Conceptually does: |
4036 | * | |
4037 | * If (@state & @p->state) @p->state = TASK_RUNNING. | |
1da177e4 | 4038 | * |
a2250238 PZ |
4039 | * If the task was not queued/runnable, also place it back on a runqueue. |
4040 | * | |
58877d34 PZ |
4041 | * This function is atomic against schedule() which would dequeue the task. |
4042 | * | |
4043 | * It issues a full memory barrier before accessing @p->state, see the comment | |
4044 | * with set_current_state(). | |
a2250238 | 4045 | * |
58877d34 | 4046 | * Uses p->pi_lock to serialize against concurrent wake-ups. |
a2250238 | 4047 | * |
58877d34 PZ |
4048 | * Relies on p->pi_lock stabilizing: |
4049 | * - p->sched_class | |
4050 | * - p->cpus_ptr | |
4051 | * - p->sched_task_group | |
4052 | * in order to do migration, see its use of select_task_rq()/set_task_cpu(). | |
4053 | * | |
4054 | * Tries really hard to only take one task_rq(p)->lock for performance. | |
4055 | * Takes rq->lock in: | |
4056 | * - ttwu_runnable() -- old rq, unavoidable, see comment there; | |
4057 | * - ttwu_queue() -- new rq, for enqueue of the task; | |
4058 | * - psi_ttwu_dequeue() -- much sadness :-( accounting will kill us. | |
4059 | * | |
4060 | * As a consequence we race really badly with just about everything. See the | |
4061 | * many memory barriers and their comments for details. | |
7696f991 | 4062 | * |
a2250238 PZ |
4063 | * Return: %true if @p->state changes (an actual wakeup was done), |
4064 | * %false otherwise. | |
1da177e4 | 4065 | */ |
e4a52bcb PZ |
4066 | static int |
4067 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 4068 | { |
1da177e4 | 4069 | unsigned long flags; |
c05fbafb | 4070 | int cpu, success = 0; |
2398f2c6 | 4071 | |
e3d85487 | 4072 | preempt_disable(); |
aacedf26 PZ |
4073 | if (p == current) { |
4074 | /* | |
4075 | * We're waking current, this means 'p->on_rq' and 'task_cpu(p) | |
4076 | * == smp_processor_id()'. Together this means we can special | |
58877d34 | 4077 | * case the whole 'p->on_rq && ttwu_runnable()' case below |
aacedf26 PZ |
4078 | * without taking any locks. |
4079 | * | |
4080 | * In particular: | |
4081 | * - we rely on Program-Order guarantees for all the ordering, | |
4082 | * - we're serialized against set_special_state() by virtue of | |
4083 | * it disabling IRQs (this allows not taking ->pi_lock). | |
4084 | */ | |
43295d73 | 4085 | if (!ttwu_state_match(p, state, &success)) |
e3d85487 | 4086 | goto out; |
aacedf26 | 4087 | |
aacedf26 | 4088 | trace_sched_waking(p); |
2f064a59 | 4089 | WRITE_ONCE(p->__state, TASK_RUNNING); |
aacedf26 PZ |
4090 | trace_sched_wakeup(p); |
4091 | goto out; | |
4092 | } | |
4093 | ||
e0acd0a6 ON |
4094 | /* |
4095 | * If we are going to wake up a thread waiting for CONDITION we | |
4096 | * need to ensure that CONDITION=1 done by the caller can not be | |
58877d34 PZ |
4097 | * reordered with p->state check below. This pairs with smp_store_mb() |
4098 | * in set_current_state() that the waiting thread does. | |
e0acd0a6 | 4099 | */ |
013fdb80 | 4100 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
d89e588c | 4101 | smp_mb__after_spinlock(); |
43295d73 | 4102 | if (!ttwu_state_match(p, state, &success)) |
aacedf26 | 4103 | goto unlock; |
1da177e4 | 4104 | |
fbd705a0 PZ |
4105 | trace_sched_waking(p); |
4106 | ||
135e8c92 BS |
4107 | /* |
4108 | * Ensure we load p->on_rq _after_ p->state, otherwise it would | |
4109 | * be possible to, falsely, observe p->on_rq == 0 and get stuck | |
4110 | * in smp_cond_load_acquire() below. | |
4111 | * | |
3d85b270 AP |
4112 | * sched_ttwu_pending() try_to_wake_up() |
4113 | * STORE p->on_rq = 1 LOAD p->state | |
4114 | * UNLOCK rq->lock | |
4115 | * | |
4116 | * __schedule() (switch to task 'p') | |
4117 | * LOCK rq->lock smp_rmb(); | |
4118 | * smp_mb__after_spinlock(); | |
4119 | * UNLOCK rq->lock | |
135e8c92 BS |
4120 | * |
4121 | * [task p] | |
3d85b270 | 4122 | * STORE p->state = UNINTERRUPTIBLE LOAD p->on_rq |
135e8c92 | 4123 | * |
3d85b270 AP |
4124 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4125 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
2beaf328 PM |
4126 | * |
4127 | * A similar smb_rmb() lives in try_invoke_on_locked_down_task(). | |
135e8c92 BS |
4128 | */ |
4129 | smp_rmb(); | |
58877d34 | 4130 | if (READ_ONCE(p->on_rq) && ttwu_runnable(p, wake_flags)) |
aacedf26 | 4131 | goto unlock; |
1da177e4 | 4132 | |
1da177e4 | 4133 | #ifdef CONFIG_SMP |
ecf7d01c PZ |
4134 | /* |
4135 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be | |
4136 | * possible to, falsely, observe p->on_cpu == 0. | |
4137 | * | |
4138 | * One must be running (->on_cpu == 1) in order to remove oneself | |
4139 | * from the runqueue. | |
4140 | * | |
3d85b270 AP |
4141 | * __schedule() (switch to task 'p') try_to_wake_up() |
4142 | * STORE p->on_cpu = 1 LOAD p->on_rq | |
4143 | * UNLOCK rq->lock | |
4144 | * | |
4145 | * __schedule() (put 'p' to sleep) | |
4146 | * LOCK rq->lock smp_rmb(); | |
4147 | * smp_mb__after_spinlock(); | |
4148 | * STORE p->on_rq = 0 LOAD p->on_cpu | |
ecf7d01c | 4149 | * |
3d85b270 AP |
4150 | * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in |
4151 | * __schedule(). See the comment for smp_mb__after_spinlock(). | |
dbfb089d PZ |
4152 | * |
4153 | * Form a control-dep-acquire with p->on_rq == 0 above, to ensure | |
4154 | * schedule()'s deactivate_task() has 'happened' and p will no longer | |
4155 | * care about it's own p->state. See the comment in __schedule(). | |
ecf7d01c | 4156 | */ |
dbfb089d PZ |
4157 | smp_acquire__after_ctrl_dep(); |
4158 | ||
4159 | /* | |
4160 | * We're doing the wakeup (@success == 1), they did a dequeue (p->on_rq | |
4161 | * == 0), which means we need to do an enqueue, change p->state to | |
4162 | * TASK_WAKING such that we can unlock p->pi_lock before doing the | |
4163 | * enqueue, such as ttwu_queue_wakelist(). | |
4164 | */ | |
2f064a59 | 4165 | WRITE_ONCE(p->__state, TASK_WAKING); |
ecf7d01c | 4166 | |
c6e7bd7a PZ |
4167 | /* |
4168 | * If the owning (remote) CPU is still in the middle of schedule() with | |
4169 | * this task as prev, considering queueing p on the remote CPUs wake_list | |
4170 | * which potentially sends an IPI instead of spinning on p->on_cpu to | |
4171 | * let the waker make forward progress. This is safe because IRQs are | |
4172 | * disabled and the IPI will deliver after on_cpu is cleared. | |
b6e13e85 PZ |
4173 | * |
4174 | * Ensure we load task_cpu(p) after p->on_cpu: | |
4175 | * | |
4176 | * set_task_cpu(p, cpu); | |
4177 | * STORE p->cpu = @cpu | |
4178 | * __schedule() (switch to task 'p') | |
4179 | * LOCK rq->lock | |
4180 | * smp_mb__after_spin_lock() smp_cond_load_acquire(&p->on_cpu) | |
4181 | * STORE p->on_cpu = 1 LOAD p->cpu | |
4182 | * | |
4183 | * to ensure we observe the correct CPU on which the task is currently | |
4184 | * scheduling. | |
c6e7bd7a | 4185 | */ |
b6e13e85 | 4186 | if (smp_load_acquire(&p->on_cpu) && |
f3dd3f67 | 4187 | ttwu_queue_wakelist(p, task_cpu(p), wake_flags)) |
c6e7bd7a PZ |
4188 | goto unlock; |
4189 | ||
e9c84311 | 4190 | /* |
d1ccc66d | 4191 | * If the owning (remote) CPU is still in the middle of schedule() with |
b19a888c | 4192 | * this task as prev, wait until it's done referencing the task. |
b75a2253 | 4193 | * |
31cb1bc0 | 4194 | * Pairs with the smp_store_release() in finish_task(). |
b75a2253 PZ |
4195 | * |
4196 | * This ensures that tasks getting woken will be fully ordered against | |
4197 | * their previous state and preserve Program Order. | |
0970d299 | 4198 | */ |
1f03e8d2 | 4199 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
1da177e4 | 4200 | |
3aef1551 | 4201 | cpu = select_task_rq(p, p->wake_cpu, wake_flags | WF_TTWU); |
f339b9dc | 4202 | if (task_cpu(p) != cpu) { |
ec618b84 PZ |
4203 | if (p->in_iowait) { |
4204 | delayacct_blkio_end(p); | |
4205 | atomic_dec(&task_rq(p)->nr_iowait); | |
4206 | } | |
4207 | ||
f339b9dc | 4208 | wake_flags |= WF_MIGRATED; |
eb414681 | 4209 | psi_ttwu_dequeue(p); |
e4a52bcb | 4210 | set_task_cpu(p, cpu); |
f339b9dc | 4211 | } |
b6e13e85 PZ |
4212 | #else |
4213 | cpu = task_cpu(p); | |
1da177e4 | 4214 | #endif /* CONFIG_SMP */ |
1da177e4 | 4215 | |
b5179ac7 | 4216 | ttwu_queue(p, cpu, wake_flags); |
aacedf26 | 4217 | unlock: |
013fdb80 | 4218 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
aacedf26 PZ |
4219 | out: |
4220 | if (success) | |
b6e13e85 | 4221 | ttwu_stat(p, task_cpu(p), wake_flags); |
e3d85487 | 4222 | preempt_enable(); |
1da177e4 LT |
4223 | |
4224 | return success; | |
4225 | } | |
4226 | ||
91dabf33 PZ |
4227 | static bool __task_needs_rq_lock(struct task_struct *p) |
4228 | { | |
4229 | unsigned int state = READ_ONCE(p->__state); | |
4230 | ||
4231 | /* | |
4232 | * Since pi->lock blocks try_to_wake_up(), we don't need rq->lock when | |
4233 | * the task is blocked. Make sure to check @state since ttwu() can drop | |
4234 | * locks at the end, see ttwu_queue_wakelist(). | |
4235 | */ | |
4236 | if (state == TASK_RUNNING || state == TASK_WAKING) | |
4237 | return true; | |
4238 | ||
4239 | /* | |
4240 | * Ensure we load p->on_rq after p->__state, otherwise it would be | |
4241 | * possible to, falsely, observe p->on_rq == 0. | |
4242 | * | |
4243 | * See try_to_wake_up() for a longer comment. | |
4244 | */ | |
4245 | smp_rmb(); | |
4246 | if (p->on_rq) | |
4247 | return true; | |
4248 | ||
4249 | #ifdef CONFIG_SMP | |
4250 | /* | |
4251 | * Ensure the task has finished __schedule() and will not be referenced | |
4252 | * anymore. Again, see try_to_wake_up() for a longer comment. | |
4253 | */ | |
4254 | smp_rmb(); | |
4255 | smp_cond_load_acquire(&p->on_cpu, !VAL); | |
4256 | #endif | |
4257 | ||
4258 | return false; | |
4259 | } | |
4260 | ||
2beaf328 | 4261 | /** |
9b3c4ab3 | 4262 | * task_call_func - Invoke a function on task in fixed state |
1b7af295 | 4263 | * @p: Process for which the function is to be invoked, can be @current. |
2beaf328 PM |
4264 | * @func: Function to invoke. |
4265 | * @arg: Argument to function. | |
4266 | * | |
f6ac18fa PZ |
4267 | * Fix the task in it's current state by avoiding wakeups and or rq operations |
4268 | * and call @func(@arg) on it. This function can use ->on_rq and task_curr() | |
4269 | * to work out what the state is, if required. Given that @func can be invoked | |
4270 | * with a runqueue lock held, it had better be quite lightweight. | |
2beaf328 PM |
4271 | * |
4272 | * Returns: | |
f6ac18fa | 4273 | * Whatever @func returns |
2beaf328 | 4274 | */ |
9b3c4ab3 | 4275 | int task_call_func(struct task_struct *p, task_call_f func, void *arg) |
2beaf328 | 4276 | { |
f6ac18fa | 4277 | struct rq *rq = NULL; |
2beaf328 | 4278 | struct rq_flags rf; |
9b3c4ab3 | 4279 | int ret; |
2beaf328 | 4280 | |
1b7af295 | 4281 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
f6ac18fa | 4282 | |
91dabf33 | 4283 | if (__task_needs_rq_lock(p)) |
2beaf328 | 4284 | rq = __task_rq_lock(p, &rf); |
f6ac18fa PZ |
4285 | |
4286 | /* | |
4287 | * At this point the task is pinned; either: | |
4288 | * - blocked and we're holding off wakeups (pi->lock) | |
4289 | * - woken, and we're holding off enqueue (rq->lock) | |
4290 | * - queued, and we're holding off schedule (rq->lock) | |
4291 | * - running, and we're holding off de-schedule (rq->lock) | |
4292 | * | |
4293 | * The called function (@func) can use: task_curr(), p->on_rq and | |
4294 | * p->__state to differentiate between these states. | |
4295 | */ | |
4296 | ret = func(p, arg); | |
4297 | ||
4298 | if (rq) | |
2beaf328 | 4299 | rq_unlock(rq, &rf); |
f6ac18fa | 4300 | |
1b7af295 | 4301 | raw_spin_unlock_irqrestore(&p->pi_lock, rf.flags); |
2beaf328 PM |
4302 | return ret; |
4303 | } | |
4304 | ||
e386b672 PM |
4305 | /** |
4306 | * cpu_curr_snapshot - Return a snapshot of the currently running task | |
4307 | * @cpu: The CPU on which to snapshot the task. | |
4308 | * | |
4309 | * Returns the task_struct pointer of the task "currently" running on | |
4310 | * the specified CPU. If the same task is running on that CPU throughout, | |
4311 | * the return value will be a pointer to that task's task_struct structure. | |
4312 | * If the CPU did any context switches even vaguely concurrently with the | |
4313 | * execution of this function, the return value will be a pointer to the | |
4314 | * task_struct structure of a randomly chosen task that was running on | |
4315 | * that CPU somewhere around the time that this function was executing. | |
4316 | * | |
4317 | * If the specified CPU was offline, the return value is whatever it | |
4318 | * is, perhaps a pointer to the task_struct structure of that CPU's idle | |
4319 | * task, but there is no guarantee. Callers wishing a useful return | |
4320 | * value must take some action to ensure that the specified CPU remains | |
4321 | * online throughout. | |
4322 | * | |
4323 | * This function executes full memory barriers before and after fetching | |
4324 | * the pointer, which permits the caller to confine this function's fetch | |
4325 | * with respect to the caller's accesses to other shared variables. | |
4326 | */ | |
4327 | struct task_struct *cpu_curr_snapshot(int cpu) | |
4328 | { | |
4329 | struct task_struct *t; | |
4330 | ||
4331 | smp_mb(); /* Pairing determined by caller's synchronization design. */ | |
4332 | t = rcu_dereference(cpu_curr(cpu)); | |
4333 | smp_mb(); /* Pairing determined by caller's synchronization design. */ | |
4334 | return t; | |
4335 | } | |
4336 | ||
50fa610a DH |
4337 | /** |
4338 | * wake_up_process - Wake up a specific process | |
4339 | * @p: The process to be woken up. | |
4340 | * | |
4341 | * Attempt to wake up the nominated process and move it to the set of runnable | |
e69f6186 YB |
4342 | * processes. |
4343 | * | |
4344 | * Return: 1 if the process was woken up, 0 if it was already running. | |
50fa610a | 4345 | * |
7696f991 | 4346 | * This function executes a full memory barrier before accessing the task state. |
50fa610a | 4347 | */ |
7ad5b3a5 | 4348 | int wake_up_process(struct task_struct *p) |
1da177e4 | 4349 | { |
9067ac85 | 4350 | return try_to_wake_up(p, TASK_NORMAL, 0); |
1da177e4 | 4351 | } |
1da177e4 LT |
4352 | EXPORT_SYMBOL(wake_up_process); |
4353 | ||
7ad5b3a5 | 4354 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
4355 | { |
4356 | return try_to_wake_up(p, state, 0); | |
4357 | } | |
4358 | ||
1da177e4 LT |
4359 | /* |
4360 | * Perform scheduler related setup for a newly forked process p. | |
4361 | * p is forked by current. | |
dd41f596 IM |
4362 | * |
4363 | * __sched_fork() is basic setup used by init_idle() too: | |
4364 | */ | |
5e1576ed | 4365 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4366 | { |
fd2f4419 PZ |
4367 | p->on_rq = 0; |
4368 | ||
4369 | p->se.on_rq = 0; | |
dd41f596 IM |
4370 | p->se.exec_start = 0; |
4371 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 4372 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 4373 | p->se.nr_migrations = 0; |
da7a735e | 4374 | p->se.vruntime = 0; |
fd2f4419 | 4375 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d | 4376 | |
ad936d86 BP |
4377 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4378 | p->se.cfs_rq = NULL; | |
4379 | #endif | |
4380 | ||
6cfb0d5d | 4381 | #ifdef CONFIG_SCHEDSTATS |
cb251765 | 4382 | /* Even if schedstat is disabled, there should not be garbage */ |
ceeadb83 | 4383 | memset(&p->stats, 0, sizeof(p->stats)); |
6cfb0d5d | 4384 | #endif |
476d139c | 4385 | |
aab03e05 | 4386 | RB_CLEAR_NODE(&p->dl.rb_node); |
40767b0d | 4387 | init_dl_task_timer(&p->dl); |
209a0cbd | 4388 | init_dl_inactive_task_timer(&p->dl); |
a5e7be3b | 4389 | __dl_clear_params(p); |
aab03e05 | 4390 | |
fa717060 | 4391 | INIT_LIST_HEAD(&p->rt.run_list); |
ff77e468 PZ |
4392 | p->rt.timeout = 0; |
4393 | p->rt.time_slice = sched_rr_timeslice; | |
4394 | p->rt.on_rq = 0; | |
4395 | p->rt.on_list = 0; | |
476d139c | 4396 | |
e107be36 AK |
4397 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4398 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
4399 | #endif | |
cbee9f88 | 4400 | |
5e1f0f09 MG |
4401 | #ifdef CONFIG_COMPACTION |
4402 | p->capture_control = NULL; | |
4403 | #endif | |
13784475 | 4404 | init_numa_balancing(clone_flags, p); |
a1488664 | 4405 | #ifdef CONFIG_SMP |
8c4890d1 | 4406 | p->wake_entry.u_flags = CSD_TYPE_TTWU; |
6d337eab | 4407 | p->migration_pending = NULL; |
a1488664 | 4408 | #endif |
dd41f596 IM |
4409 | } |
4410 | ||
2a595721 SD |
4411 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
4412 | ||
1a687c2e | 4413 | #ifdef CONFIG_NUMA_BALANCING |
c3b9bc5b | 4414 | |
c574bbe9 HY |
4415 | int sysctl_numa_balancing_mode; |
4416 | ||
4417 | static void __set_numabalancing_state(bool enabled) | |
1a687c2e MG |
4418 | { |
4419 | if (enabled) | |
2a595721 | 4420 | static_branch_enable(&sched_numa_balancing); |
1a687c2e | 4421 | else |
2a595721 | 4422 | static_branch_disable(&sched_numa_balancing); |
1a687c2e | 4423 | } |
54a43d54 | 4424 | |
c574bbe9 HY |
4425 | void set_numabalancing_state(bool enabled) |
4426 | { | |
4427 | if (enabled) | |
4428 | sysctl_numa_balancing_mode = NUMA_BALANCING_NORMAL; | |
4429 | else | |
4430 | sysctl_numa_balancing_mode = NUMA_BALANCING_DISABLED; | |
4431 | __set_numabalancing_state(enabled); | |
4432 | } | |
4433 | ||
54a43d54 | 4434 | #ifdef CONFIG_PROC_SYSCTL |
c959924b HY |
4435 | static void reset_memory_tiering(void) |
4436 | { | |
4437 | struct pglist_data *pgdat; | |
4438 | ||
4439 | for_each_online_pgdat(pgdat) { | |
4440 | pgdat->nbp_threshold = 0; | |
4441 | pgdat->nbp_th_nr_cand = node_page_state(pgdat, PGPROMOTE_CANDIDATE); | |
4442 | pgdat->nbp_th_start = jiffies_to_msecs(jiffies); | |
4443 | } | |
4444 | } | |
4445 | ||
0dff89c4 | 4446 | static int sysctl_numa_balancing(struct ctl_table *table, int write, |
32927393 | 4447 | void *buffer, size_t *lenp, loff_t *ppos) |
54a43d54 AK |
4448 | { |
4449 | struct ctl_table t; | |
4450 | int err; | |
c574bbe9 | 4451 | int state = sysctl_numa_balancing_mode; |
54a43d54 AK |
4452 | |
4453 | if (write && !capable(CAP_SYS_ADMIN)) | |
4454 | return -EPERM; | |
4455 | ||
4456 | t = *table; | |
4457 | t.data = &state; | |
4458 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4459 | if (err < 0) | |
4460 | return err; | |
c574bbe9 | 4461 | if (write) { |
c959924b HY |
4462 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) && |
4463 | (state & NUMA_BALANCING_MEMORY_TIERING)) | |
4464 | reset_memory_tiering(); | |
c574bbe9 HY |
4465 | sysctl_numa_balancing_mode = state; |
4466 | __set_numabalancing_state(state); | |
4467 | } | |
54a43d54 AK |
4468 | return err; |
4469 | } | |
4470 | #endif | |
4471 | #endif | |
dd41f596 | 4472 | |
4698f88c JP |
4473 | #ifdef CONFIG_SCHEDSTATS |
4474 | ||
cb251765 MG |
4475 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
4476 | ||
cb251765 MG |
4477 | static void set_schedstats(bool enabled) |
4478 | { | |
4479 | if (enabled) | |
4480 | static_branch_enable(&sched_schedstats); | |
4481 | else | |
4482 | static_branch_disable(&sched_schedstats); | |
4483 | } | |
4484 | ||
4485 | void force_schedstat_enabled(void) | |
4486 | { | |
4487 | if (!schedstat_enabled()) { | |
4488 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); | |
4489 | static_branch_enable(&sched_schedstats); | |
4490 | } | |
4491 | } | |
4492 | ||
4493 | static int __init setup_schedstats(char *str) | |
4494 | { | |
4495 | int ret = 0; | |
4496 | if (!str) | |
4497 | goto out; | |
4498 | ||
4499 | if (!strcmp(str, "enable")) { | |
1faa491a | 4500 | set_schedstats(true); |
cb251765 MG |
4501 | ret = 1; |
4502 | } else if (!strcmp(str, "disable")) { | |
1faa491a | 4503 | set_schedstats(false); |
cb251765 MG |
4504 | ret = 1; |
4505 | } | |
4506 | out: | |
4507 | if (!ret) | |
4508 | pr_warn("Unable to parse schedstats=\n"); | |
4509 | ||
4510 | return ret; | |
4511 | } | |
4512 | __setup("schedstats=", setup_schedstats); | |
4513 | ||
4514 | #ifdef CONFIG_PROC_SYSCTL | |
f5ef06d5 | 4515 | static int sysctl_schedstats(struct ctl_table *table, int write, void *buffer, |
32927393 | 4516 | size_t *lenp, loff_t *ppos) |
cb251765 MG |
4517 | { |
4518 | struct ctl_table t; | |
4519 | int err; | |
4520 | int state = static_branch_likely(&sched_schedstats); | |
4521 | ||
4522 | if (write && !capable(CAP_SYS_ADMIN)) | |
4523 | return -EPERM; | |
4524 | ||
4525 | t = *table; | |
4526 | t.data = &state; | |
4527 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); | |
4528 | if (err < 0) | |
4529 | return err; | |
4530 | if (write) | |
4531 | set_schedstats(state); | |
4532 | return err; | |
4533 | } | |
4698f88c | 4534 | #endif /* CONFIG_PROC_SYSCTL */ |
4698f88c | 4535 | #endif /* CONFIG_SCHEDSTATS */ |
dd41f596 | 4536 | |
3267e015 ZN |
4537 | #ifdef CONFIG_SYSCTL |
4538 | static struct ctl_table sched_core_sysctls[] = { | |
4539 | #ifdef CONFIG_SCHEDSTATS | |
f5ef06d5 ZN |
4540 | { |
4541 | .procname = "sched_schedstats", | |
4542 | .data = NULL, | |
4543 | .maxlen = sizeof(unsigned int), | |
4544 | .mode = 0644, | |
4545 | .proc_handler = sysctl_schedstats, | |
4546 | .extra1 = SYSCTL_ZERO, | |
4547 | .extra2 = SYSCTL_ONE, | |
4548 | }, | |
3267e015 ZN |
4549 | #endif /* CONFIG_SCHEDSTATS */ |
4550 | #ifdef CONFIG_UCLAMP_TASK | |
4551 | { | |
4552 | .procname = "sched_util_clamp_min", | |
4553 | .data = &sysctl_sched_uclamp_util_min, | |
4554 | .maxlen = sizeof(unsigned int), | |
4555 | .mode = 0644, | |
4556 | .proc_handler = sysctl_sched_uclamp_handler, | |
4557 | }, | |
4558 | { | |
4559 | .procname = "sched_util_clamp_max", | |
4560 | .data = &sysctl_sched_uclamp_util_max, | |
4561 | .maxlen = sizeof(unsigned int), | |
4562 | .mode = 0644, | |
4563 | .proc_handler = sysctl_sched_uclamp_handler, | |
4564 | }, | |
4565 | { | |
4566 | .procname = "sched_util_clamp_min_rt_default", | |
4567 | .data = &sysctl_sched_uclamp_util_min_rt_default, | |
4568 | .maxlen = sizeof(unsigned int), | |
4569 | .mode = 0644, | |
4570 | .proc_handler = sysctl_sched_uclamp_handler, | |
4571 | }, | |
4572 | #endif /* CONFIG_UCLAMP_TASK */ | |
0dff89c4 KW |
4573 | #ifdef CONFIG_NUMA_BALANCING |
4574 | { | |
4575 | .procname = "numa_balancing", | |
4576 | .data = NULL, /* filled in by handler */ | |
4577 | .maxlen = sizeof(unsigned int), | |
4578 | .mode = 0644, | |
4579 | .proc_handler = sysctl_numa_balancing, | |
4580 | .extra1 = SYSCTL_ZERO, | |
4581 | .extra2 = SYSCTL_FOUR, | |
4582 | }, | |
4583 | #endif /* CONFIG_NUMA_BALANCING */ | |
f5ef06d5 ZN |
4584 | {} |
4585 | }; | |
3267e015 | 4586 | static int __init sched_core_sysctl_init(void) |
f5ef06d5 | 4587 | { |
3267e015 | 4588 | register_sysctl_init("kernel", sched_core_sysctls); |
f5ef06d5 ZN |
4589 | return 0; |
4590 | } | |
3267e015 ZN |
4591 | late_initcall(sched_core_sysctl_init); |
4592 | #endif /* CONFIG_SYSCTL */ | |
dd41f596 IM |
4593 | |
4594 | /* | |
4595 | * fork()/clone()-time setup: | |
4596 | */ | |
aab03e05 | 4597 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
dd41f596 | 4598 | { |
5e1576ed | 4599 | __sched_fork(clone_flags, p); |
06b83b5f | 4600 | /* |
7dc603c9 | 4601 | * We mark the process as NEW here. This guarantees that |
06b83b5f PZ |
4602 | * nobody will actually run it, and a signal or other external |
4603 | * event cannot wake it up and insert it on the runqueue either. | |
4604 | */ | |
2f064a59 | 4605 | p->__state = TASK_NEW; |
dd41f596 | 4606 | |
c350a04e MG |
4607 | /* |
4608 | * Make sure we do not leak PI boosting priority to the child. | |
4609 | */ | |
4610 | p->prio = current->normal_prio; | |
4611 | ||
e8f14172 PB |
4612 | uclamp_fork(p); |
4613 | ||
b9dc29e7 MG |
4614 | /* |
4615 | * Revert to default priority/policy on fork if requested. | |
4616 | */ | |
4617 | if (unlikely(p->sched_reset_on_fork)) { | |
aab03e05 | 4618 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
b9dc29e7 | 4619 | p->policy = SCHED_NORMAL; |
6c697bdf | 4620 | p->static_prio = NICE_TO_PRIO(0); |
c350a04e MG |
4621 | p->rt_priority = 0; |
4622 | } else if (PRIO_TO_NICE(p->static_prio) < 0) | |
4623 | p->static_prio = NICE_TO_PRIO(0); | |
4624 | ||
f558c2b8 | 4625 | p->prio = p->normal_prio = p->static_prio; |
b1e82065 | 4626 | set_load_weight(p, false); |
6c697bdf | 4627 | |
b9dc29e7 MG |
4628 | /* |
4629 | * We don't need the reset flag anymore after the fork. It has | |
4630 | * fulfilled its duty: | |
4631 | */ | |
4632 | p->sched_reset_on_fork = 0; | |
4633 | } | |
ca94c442 | 4634 | |
af0fffd9 | 4635 | if (dl_prio(p->prio)) |
aab03e05 | 4636 | return -EAGAIN; |
af0fffd9 | 4637 | else if (rt_prio(p->prio)) |
aab03e05 | 4638 | p->sched_class = &rt_sched_class; |
af0fffd9 | 4639 | else |
2ddbf952 | 4640 | p->sched_class = &fair_sched_class; |
b29739f9 | 4641 | |
7dc603c9 | 4642 | init_entity_runnable_average(&p->se); |
cd29fe6f | 4643 | |
b1e82065 | 4644 | |
f6db8347 | 4645 | #ifdef CONFIG_SCHED_INFO |
dd41f596 | 4646 | if (likely(sched_info_on())) |
52f17b6c | 4647 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 4648 | #endif |
3ca7a440 PZ |
4649 | #if defined(CONFIG_SMP) |
4650 | p->on_cpu = 0; | |
4866cde0 | 4651 | #endif |
01028747 | 4652 | init_task_preempt_count(p); |
806c09a7 | 4653 | #ifdef CONFIG_SMP |
917b627d | 4654 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
1baca4ce | 4655 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
806c09a7 | 4656 | #endif |
aab03e05 | 4657 | return 0; |
1da177e4 LT |
4658 | } |
4659 | ||
b1e82065 | 4660 | void sched_cgroup_fork(struct task_struct *p, struct kernel_clone_args *kargs) |
13685c4a | 4661 | { |
4ef0c5c6 | 4662 | unsigned long flags; |
4ef0c5c6 | 4663 | |
b1e82065 PZ |
4664 | /* |
4665 | * Because we're not yet on the pid-hash, p->pi_lock isn't strictly | |
4666 | * required yet, but lockdep gets upset if rules are violated. | |
4667 | */ | |
4ef0c5c6 ZQ |
4668 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
4669 | #ifdef CONFIG_CGROUP_SCHED | |
b1e82065 PZ |
4670 | if (1) { |
4671 | struct task_group *tg; | |
4672 | tg = container_of(kargs->cset->subsys[cpu_cgrp_id], | |
4673 | struct task_group, css); | |
4674 | tg = autogroup_task_group(p, tg); | |
4675 | p->sched_task_group = tg; | |
4676 | } | |
4ef0c5c6 ZQ |
4677 | #endif |
4678 | rseq_migrate(p); | |
4679 | /* | |
4680 | * We're setting the CPU for the first time, we don't migrate, | |
4681 | * so use __set_task_cpu(). | |
4682 | */ | |
4683 | __set_task_cpu(p, smp_processor_id()); | |
4684 | if (p->sched_class->task_fork) | |
4685 | p->sched_class->task_fork(p); | |
4686 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
b1e82065 | 4687 | } |
4ef0c5c6 | 4688 | |
b1e82065 PZ |
4689 | void sched_post_fork(struct task_struct *p) |
4690 | { | |
13685c4a QY |
4691 | uclamp_post_fork(p); |
4692 | } | |
4693 | ||
332ac17e DF |
4694 | unsigned long to_ratio(u64 period, u64 runtime) |
4695 | { | |
4696 | if (runtime == RUNTIME_INF) | |
c52f14d3 | 4697 | return BW_UNIT; |
332ac17e DF |
4698 | |
4699 | /* | |
4700 | * Doing this here saves a lot of checks in all | |
4701 | * the calling paths, and returning zero seems | |
4702 | * safe for them anyway. | |
4703 | */ | |
4704 | if (period == 0) | |
4705 | return 0; | |
4706 | ||
c52f14d3 | 4707 | return div64_u64(runtime << BW_SHIFT, period); |
332ac17e DF |
4708 | } |
4709 | ||
1da177e4 LT |
4710 | /* |
4711 | * wake_up_new_task - wake up a newly created task for the first time. | |
4712 | * | |
4713 | * This function will do some initial scheduler statistics housekeeping | |
4714 | * that must be done for every newly created context, then puts the task | |
4715 | * on the runqueue and wakes it. | |
4716 | */ | |
3e51e3ed | 4717 | void wake_up_new_task(struct task_struct *p) |
1da177e4 | 4718 | { |
eb580751 | 4719 | struct rq_flags rf; |
dd41f596 | 4720 | struct rq *rq; |
fabf318e | 4721 | |
eb580751 | 4722 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
2f064a59 | 4723 | WRITE_ONCE(p->__state, TASK_RUNNING); |
fabf318e PZ |
4724 | #ifdef CONFIG_SMP |
4725 | /* | |
4726 | * Fork balancing, do it here and not earlier because: | |
3bd37062 | 4727 | * - cpus_ptr can change in the fork path |
d1ccc66d | 4728 | * - any previously selected CPU might disappear through hotplug |
e210bffd PZ |
4729 | * |
4730 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, | |
4731 | * as we're not fully set-up yet. | |
fabf318e | 4732 | */ |
32e839dd | 4733 | p->recent_used_cpu = task_cpu(p); |
ce3614da | 4734 | rseq_migrate(p); |
3aef1551 | 4735 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), WF_FORK)); |
0017d735 | 4736 | #endif |
b7fa30c9 | 4737 | rq = __task_rq_lock(p, &rf); |
4126bad6 | 4738 | update_rq_clock(rq); |
d0fe0b9c | 4739 | post_init_entity_util_avg(p); |
0017d735 | 4740 | |
7a57f32a | 4741 | activate_task(rq, p, ENQUEUE_NOCLOCK); |
fbd705a0 | 4742 | trace_sched_wakeup_new(p); |
a7558e01 | 4743 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 4744 | #ifdef CONFIG_SMP |
0aaafaab PZ |
4745 | if (p->sched_class->task_woken) { |
4746 | /* | |
b19a888c | 4747 | * Nothing relies on rq->lock after this, so it's fine to |
0aaafaab PZ |
4748 | * drop it. |
4749 | */ | |
d8ac8971 | 4750 | rq_unpin_lock(rq, &rf); |
efbbd05a | 4751 | p->sched_class->task_woken(rq, p); |
d8ac8971 | 4752 | rq_repin_lock(rq, &rf); |
0aaafaab | 4753 | } |
9a897c5a | 4754 | #endif |
eb580751 | 4755 | task_rq_unlock(rq, p, &rf); |
1da177e4 LT |
4756 | } |
4757 | ||
e107be36 AK |
4758 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
4759 | ||
b7203428 | 4760 | static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key); |
1cde2930 | 4761 | |
2ecd9d29 PZ |
4762 | void preempt_notifier_inc(void) |
4763 | { | |
b7203428 | 4764 | static_branch_inc(&preempt_notifier_key); |
2ecd9d29 PZ |
4765 | } |
4766 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); | |
4767 | ||
4768 | void preempt_notifier_dec(void) | |
4769 | { | |
b7203428 | 4770 | static_branch_dec(&preempt_notifier_key); |
2ecd9d29 PZ |
4771 | } |
4772 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); | |
4773 | ||
e107be36 | 4774 | /** |
80dd99b3 | 4775 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 4776 | * @notifier: notifier struct to register |
e107be36 AK |
4777 | */ |
4778 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
4779 | { | |
b7203428 | 4780 | if (!static_branch_unlikely(&preempt_notifier_key)) |
2ecd9d29 PZ |
4781 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
4782 | ||
e107be36 AK |
4783 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
4784 | } | |
4785 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
4786 | ||
4787 | /** | |
4788 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 4789 | * @notifier: notifier struct to unregister |
e107be36 | 4790 | * |
d84525a8 | 4791 | * This is *not* safe to call from within a preemption notifier. |
e107be36 AK |
4792 | */ |
4793 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
4794 | { | |
4795 | hlist_del(¬ifier->link); | |
4796 | } | |
4797 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
4798 | ||
1cde2930 | 4799 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4800 | { |
4801 | struct preempt_notifier *notifier; | |
e107be36 | 4802 | |
b67bfe0d | 4803 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4804 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
4805 | } | |
4806 | ||
1cde2930 PZ |
4807 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
4808 | { | |
b7203428 | 4809 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4810 | __fire_sched_in_preempt_notifiers(curr); |
4811 | } | |
4812 | ||
e107be36 | 4813 | static void |
1cde2930 PZ |
4814 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4815 | struct task_struct *next) | |
e107be36 AK |
4816 | { |
4817 | struct preempt_notifier *notifier; | |
e107be36 | 4818 | |
b67bfe0d | 4819 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
e107be36 AK |
4820 | notifier->ops->sched_out(notifier, next); |
4821 | } | |
4822 | ||
1cde2930 PZ |
4823 | static __always_inline void |
4824 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
4825 | struct task_struct *next) | |
4826 | { | |
b7203428 | 4827 | if (static_branch_unlikely(&preempt_notifier_key)) |
1cde2930 PZ |
4828 | __fire_sched_out_preempt_notifiers(curr, next); |
4829 | } | |
4830 | ||
6d6bc0ad | 4831 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4832 | |
1cde2930 | 4833 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
e107be36 AK |
4834 | { |
4835 | } | |
4836 | ||
1cde2930 | 4837 | static inline void |
e107be36 AK |
4838 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
4839 | struct task_struct *next) | |
4840 | { | |
4841 | } | |
4842 | ||
6d6bc0ad | 4843 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 4844 | |
31cb1bc0 | 4845 | static inline void prepare_task(struct task_struct *next) |
4846 | { | |
4847 | #ifdef CONFIG_SMP | |
4848 | /* | |
4849 | * Claim the task as running, we do this before switching to it | |
4850 | * such that any running task will have this set. | |
58877d34 | 4851 | * |
f3dd3f67 TD |
4852 | * See the smp_load_acquire(&p->on_cpu) case in ttwu() and |
4853 | * its ordering comment. | |
31cb1bc0 | 4854 | */ |
58877d34 | 4855 | WRITE_ONCE(next->on_cpu, 1); |
31cb1bc0 | 4856 | #endif |
4857 | } | |
4858 | ||
4859 | static inline void finish_task(struct task_struct *prev) | |
4860 | { | |
4861 | #ifdef CONFIG_SMP | |
4862 | /* | |
58877d34 PZ |
4863 | * This must be the very last reference to @prev from this CPU. After |
4864 | * p->on_cpu is cleared, the task can be moved to a different CPU. We | |
4865 | * must ensure this doesn't happen until the switch is completely | |
31cb1bc0 | 4866 | * finished. |
4867 | * | |
4868 | * In particular, the load of prev->state in finish_task_switch() must | |
4869 | * happen before this. | |
4870 | * | |
4871 | * Pairs with the smp_cond_load_acquire() in try_to_wake_up(). | |
4872 | */ | |
4873 | smp_store_release(&prev->on_cpu, 0); | |
4874 | #endif | |
4875 | } | |
4876 | ||
565790d2 PZ |
4877 | #ifdef CONFIG_SMP |
4878 | ||
8e5bad7d | 4879 | static void do_balance_callbacks(struct rq *rq, struct balance_callback *head) |
565790d2 PZ |
4880 | { |
4881 | void (*func)(struct rq *rq); | |
8e5bad7d | 4882 | struct balance_callback *next; |
565790d2 | 4883 | |
5cb9eaa3 | 4884 | lockdep_assert_rq_held(rq); |
565790d2 PZ |
4885 | |
4886 | while (head) { | |
4887 | func = (void (*)(struct rq *))head->func; | |
4888 | next = head->next; | |
4889 | head->next = NULL; | |
4890 | head = next; | |
4891 | ||
4892 | func(rq); | |
4893 | } | |
4894 | } | |
4895 | ||
ae792702 PZ |
4896 | static void balance_push(struct rq *rq); |
4897 | ||
04193d59 PZ |
4898 | /* |
4899 | * balance_push_callback is a right abuse of the callback interface and plays | |
4900 | * by significantly different rules. | |
4901 | * | |
4902 | * Where the normal balance_callback's purpose is to be ran in the same context | |
4903 | * that queued it (only later, when it's safe to drop rq->lock again), | |
4904 | * balance_push_callback is specifically targeted at __schedule(). | |
4905 | * | |
4906 | * This abuse is tolerated because it places all the unlikely/odd cases behind | |
4907 | * a single test, namely: rq->balance_callback == NULL. | |
4908 | */ | |
8e5bad7d | 4909 | struct balance_callback balance_push_callback = { |
ae792702 | 4910 | .next = NULL, |
8e5bad7d | 4911 | .func = balance_push, |
ae792702 PZ |
4912 | }; |
4913 | ||
8e5bad7d | 4914 | static inline struct balance_callback * |
04193d59 | 4915 | __splice_balance_callbacks(struct rq *rq, bool split) |
565790d2 | 4916 | { |
8e5bad7d | 4917 | struct balance_callback *head = rq->balance_callback; |
565790d2 | 4918 | |
04193d59 PZ |
4919 | if (likely(!head)) |
4920 | return NULL; | |
4921 | ||
5cb9eaa3 | 4922 | lockdep_assert_rq_held(rq); |
04193d59 PZ |
4923 | /* |
4924 | * Must not take balance_push_callback off the list when | |
4925 | * splice_balance_callbacks() and balance_callbacks() are not | |
4926 | * in the same rq->lock section. | |
4927 | * | |
4928 | * In that case it would be possible for __schedule() to interleave | |
4929 | * and observe the list empty. | |
4930 | */ | |
4931 | if (split && head == &balance_push_callback) | |
4932 | head = NULL; | |
4933 | else | |
565790d2 PZ |
4934 | rq->balance_callback = NULL; |
4935 | ||
4936 | return head; | |
4937 | } | |
4938 | ||
8e5bad7d | 4939 | static inline struct balance_callback *splice_balance_callbacks(struct rq *rq) |
04193d59 PZ |
4940 | { |
4941 | return __splice_balance_callbacks(rq, true); | |
4942 | } | |
4943 | ||
565790d2 PZ |
4944 | static void __balance_callbacks(struct rq *rq) |
4945 | { | |
04193d59 | 4946 | do_balance_callbacks(rq, __splice_balance_callbacks(rq, false)); |
565790d2 PZ |
4947 | } |
4948 | ||
8e5bad7d | 4949 | static inline void balance_callbacks(struct rq *rq, struct balance_callback *head) |
565790d2 PZ |
4950 | { |
4951 | unsigned long flags; | |
4952 | ||
4953 | if (unlikely(head)) { | |
5cb9eaa3 | 4954 | raw_spin_rq_lock_irqsave(rq, flags); |
565790d2 | 4955 | do_balance_callbacks(rq, head); |
5cb9eaa3 | 4956 | raw_spin_rq_unlock_irqrestore(rq, flags); |
565790d2 PZ |
4957 | } |
4958 | } | |
4959 | ||
4960 | #else | |
4961 | ||
4962 | static inline void __balance_callbacks(struct rq *rq) | |
4963 | { | |
4964 | } | |
4965 | ||
8e5bad7d | 4966 | static inline struct balance_callback *splice_balance_callbacks(struct rq *rq) |
565790d2 PZ |
4967 | { |
4968 | return NULL; | |
4969 | } | |
4970 | ||
8e5bad7d | 4971 | static inline void balance_callbacks(struct rq *rq, struct balance_callback *head) |
565790d2 PZ |
4972 | { |
4973 | } | |
4974 | ||
4975 | #endif | |
4976 | ||
269d5992 PZ |
4977 | static inline void |
4978 | prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf) | |
31cb1bc0 | 4979 | { |
269d5992 PZ |
4980 | /* |
4981 | * Since the runqueue lock will be released by the next | |
4982 | * task (which is an invalid locking op but in the case | |
4983 | * of the scheduler it's an obvious special-case), so we | |
4984 | * do an early lockdep release here: | |
4985 | */ | |
4986 | rq_unpin_lock(rq, rf); | |
9ef7e7e3 | 4987 | spin_release(&__rq_lockp(rq)->dep_map, _THIS_IP_); |
31cb1bc0 | 4988 | #ifdef CONFIG_DEBUG_SPINLOCK |
4989 | /* this is a valid case when another task releases the spinlock */ | |
5cb9eaa3 | 4990 | rq_lockp(rq)->owner = next; |
31cb1bc0 | 4991 | #endif |
269d5992 PZ |
4992 | } |
4993 | ||
4994 | static inline void finish_lock_switch(struct rq *rq) | |
4995 | { | |
31cb1bc0 | 4996 | /* |
4997 | * If we are tracking spinlock dependencies then we have to | |
4998 | * fix up the runqueue lock - which gets 'carried over' from | |
4999 | * prev into current: | |
5000 | */ | |
9ef7e7e3 | 5001 | spin_acquire(&__rq_lockp(rq)->dep_map, 0, 0, _THIS_IP_); |
ae792702 | 5002 | __balance_callbacks(rq); |
5cb9eaa3 | 5003 | raw_spin_rq_unlock_irq(rq); |
31cb1bc0 | 5004 | } |
5005 | ||
325ea10c IM |
5006 | /* |
5007 | * NOP if the arch has not defined these: | |
5008 | */ | |
5009 | ||
5010 | #ifndef prepare_arch_switch | |
5011 | # define prepare_arch_switch(next) do { } while (0) | |
5012 | #endif | |
5013 | ||
5014 | #ifndef finish_arch_post_lock_switch | |
5015 | # define finish_arch_post_lock_switch() do { } while (0) | |
5016 | #endif | |
5017 | ||
5fbda3ec TG |
5018 | static inline void kmap_local_sched_out(void) |
5019 | { | |
5020 | #ifdef CONFIG_KMAP_LOCAL | |
5021 | if (unlikely(current->kmap_ctrl.idx)) | |
5022 | __kmap_local_sched_out(); | |
5023 | #endif | |
5024 | } | |
5025 | ||
5026 | static inline void kmap_local_sched_in(void) | |
5027 | { | |
5028 | #ifdef CONFIG_KMAP_LOCAL | |
5029 | if (unlikely(current->kmap_ctrl.idx)) | |
5030 | __kmap_local_sched_in(); | |
5031 | #endif | |
5032 | } | |
5033 | ||
4866cde0 NP |
5034 | /** |
5035 | * prepare_task_switch - prepare to switch tasks | |
5036 | * @rq: the runqueue preparing to switch | |
421cee29 | 5037 | * @prev: the current task that is being switched out |
4866cde0 NP |
5038 | * @next: the task we are going to switch to. |
5039 | * | |
5040 | * This is called with the rq lock held and interrupts off. It must | |
5041 | * be paired with a subsequent finish_task_switch after the context | |
5042 | * switch. | |
5043 | * | |
5044 | * prepare_task_switch sets up locking and calls architecture specific | |
5045 | * hooks. | |
5046 | */ | |
e107be36 AK |
5047 | static inline void |
5048 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
5049 | struct task_struct *next) | |
4866cde0 | 5050 | { |
0ed557aa | 5051 | kcov_prepare_switch(prev); |
43148951 | 5052 | sched_info_switch(rq, prev, next); |
fe4b04fa | 5053 | perf_event_task_sched_out(prev, next); |
d7822b1e | 5054 | rseq_preempt(prev); |
e107be36 | 5055 | fire_sched_out_preempt_notifiers(prev, next); |
5fbda3ec | 5056 | kmap_local_sched_out(); |
31cb1bc0 | 5057 | prepare_task(next); |
4866cde0 NP |
5058 | prepare_arch_switch(next); |
5059 | } | |
5060 | ||
1da177e4 LT |
5061 | /** |
5062 | * finish_task_switch - clean up after a task-switch | |
5063 | * @prev: the thread we just switched away from. | |
5064 | * | |
4866cde0 NP |
5065 | * finish_task_switch must be called after the context switch, paired |
5066 | * with a prepare_task_switch call before the context switch. | |
5067 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
5068 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
5069 | * |
5070 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 5071 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
5072 | * with the lock held can cause deadlocks; see schedule() for |
5073 | * details.) | |
dfa50b60 ON |
5074 | * |
5075 | * The context switch have flipped the stack from under us and restored the | |
5076 | * local variables which were saved when this task called schedule() in the | |
5077 | * past. prev == current is still correct but we need to recalculate this_rq | |
5078 | * because prev may have moved to another CPU. | |
1da177e4 | 5079 | */ |
dfa50b60 | 5080 | static struct rq *finish_task_switch(struct task_struct *prev) |
1da177e4 LT |
5081 | __releases(rq->lock) |
5082 | { | |
dfa50b60 | 5083 | struct rq *rq = this_rq(); |
1da177e4 | 5084 | struct mm_struct *mm = rq->prev_mm; |
fa2c3254 | 5085 | unsigned int prev_state; |
1da177e4 | 5086 | |
609ca066 PZ |
5087 | /* |
5088 | * The previous task will have left us with a preempt_count of 2 | |
5089 | * because it left us after: | |
5090 | * | |
5091 | * schedule() | |
5092 | * preempt_disable(); // 1 | |
5093 | * __schedule() | |
5094 | * raw_spin_lock_irq(&rq->lock) // 2 | |
5095 | * | |
5096 | * Also, see FORK_PREEMPT_COUNT. | |
5097 | */ | |
e2bf1c4b PZ |
5098 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
5099 | "corrupted preempt_count: %s/%d/0x%x\n", | |
5100 | current->comm, current->pid, preempt_count())) | |
5101 | preempt_count_set(FORK_PREEMPT_COUNT); | |
609ca066 | 5102 | |
1da177e4 LT |
5103 | rq->prev_mm = NULL; |
5104 | ||
5105 | /* | |
5106 | * A task struct has one reference for the use as "current". | |
c394cc9f | 5107 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
5108 | * schedule one last time. The schedule call will never return, and |
5109 | * the scheduled task must drop that reference. | |
95913d97 PZ |
5110 | * |
5111 | * We must observe prev->state before clearing prev->on_cpu (in | |
31cb1bc0 | 5112 | * finish_task), otherwise a concurrent wakeup can get prev |
95913d97 PZ |
5113 | * running on another CPU and we could rave with its RUNNING -> DEAD |
5114 | * transition, resulting in a double drop. | |
1da177e4 | 5115 | */ |
2f064a59 | 5116 | prev_state = READ_ONCE(prev->__state); |
bf9fae9f | 5117 | vtime_task_switch(prev); |
a8d757ef | 5118 | perf_event_task_sched_in(prev, current); |
31cb1bc0 | 5119 | finish_task(prev); |
0fdcccfa | 5120 | tick_nohz_task_switch(); |
31cb1bc0 | 5121 | finish_lock_switch(rq); |
01f23e16 | 5122 | finish_arch_post_lock_switch(); |
0ed557aa | 5123 | kcov_finish_switch(current); |
5fbda3ec TG |
5124 | /* |
5125 | * kmap_local_sched_out() is invoked with rq::lock held and | |
5126 | * interrupts disabled. There is no requirement for that, but the | |
5127 | * sched out code does not have an interrupt enabled section. | |
5128 | * Restoring the maps on sched in does not require interrupts being | |
5129 | * disabled either. | |
5130 | */ | |
5131 | kmap_local_sched_in(); | |
e8fa1362 | 5132 | |
e107be36 | 5133 | fire_sched_in_preempt_notifiers(current); |
306e0604 | 5134 | /* |
70216e18 MD |
5135 | * When switching through a kernel thread, the loop in |
5136 | * membarrier_{private,global}_expedited() may have observed that | |
5137 | * kernel thread and not issued an IPI. It is therefore possible to | |
5138 | * schedule between user->kernel->user threads without passing though | |
5139 | * switch_mm(). Membarrier requires a barrier after storing to | |
5140 | * rq->curr, before returning to userspace, so provide them here: | |
5141 | * | |
5142 | * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly | |
5143 | * provided by mmdrop(), | |
5144 | * - a sync_core for SYNC_CORE. | |
306e0604 | 5145 | */ |
70216e18 MD |
5146 | if (mm) { |
5147 | membarrier_mm_sync_core_before_usermode(mm); | |
8d491de6 | 5148 | mmdrop_sched(mm); |
70216e18 | 5149 | } |
1cef1150 PZ |
5150 | if (unlikely(prev_state == TASK_DEAD)) { |
5151 | if (prev->sched_class->task_dead) | |
5152 | prev->sched_class->task_dead(prev); | |
68f24b08 | 5153 | |
1cef1150 PZ |
5154 | /* Task is done with its stack. */ |
5155 | put_task_stack(prev); | |
5156 | ||
0ff7b2cf | 5157 | put_task_struct_rcu_user(prev); |
c6fd91f0 | 5158 | } |
99e5ada9 | 5159 | |
dfa50b60 | 5160 | return rq; |
1da177e4 LT |
5161 | } |
5162 | ||
5163 | /** | |
5164 | * schedule_tail - first thing a freshly forked thread must call. | |
5165 | * @prev: the thread we just switched away from. | |
5166 | */ | |
722a9f92 | 5167 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
5168 | __releases(rq->lock) |
5169 | { | |
609ca066 PZ |
5170 | /* |
5171 | * New tasks start with FORK_PREEMPT_COUNT, see there and | |
5172 | * finish_task_switch() for details. | |
5173 | * | |
5174 | * finish_task_switch() will drop rq->lock() and lower preempt_count | |
5175 | * and the preempt_enable() will end up enabling preemption (on | |
5176 | * PREEMPT_COUNT kernels). | |
5177 | */ | |
5178 | ||
13c2235b | 5179 | finish_task_switch(prev); |
1a43a14a | 5180 | preempt_enable(); |
70b97a7f | 5181 | |
1da177e4 | 5182 | if (current->set_child_tid) |
b488893a | 5183 | put_user(task_pid_vnr(current), current->set_child_tid); |
088fe47c EB |
5184 | |
5185 | calculate_sigpending(); | |
1da177e4 LT |
5186 | } |
5187 | ||
5188 | /* | |
dfa50b60 | 5189 | * context_switch - switch to the new MM and the new thread's register state. |
1da177e4 | 5190 | */ |
04936948 | 5191 | static __always_inline struct rq * |
70b97a7f | 5192 | context_switch(struct rq *rq, struct task_struct *prev, |
d8ac8971 | 5193 | struct task_struct *next, struct rq_flags *rf) |
1da177e4 | 5194 | { |
e107be36 | 5195 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 5196 | |
9226d125 ZA |
5197 | /* |
5198 | * For paravirt, this is coupled with an exit in switch_to to | |
5199 | * combine the page table reload and the switch backend into | |
5200 | * one hypercall. | |
5201 | */ | |
224101ed | 5202 | arch_start_context_switch(prev); |
9226d125 | 5203 | |
306e0604 | 5204 | /* |
139d025c PZ |
5205 | * kernel -> kernel lazy + transfer active |
5206 | * user -> kernel lazy + mmgrab() active | |
5207 | * | |
5208 | * kernel -> user switch + mmdrop() active | |
5209 | * user -> user switch | |
306e0604 | 5210 | */ |
139d025c PZ |
5211 | if (!next->mm) { // to kernel |
5212 | enter_lazy_tlb(prev->active_mm, next); | |
5213 | ||
5214 | next->active_mm = prev->active_mm; | |
5215 | if (prev->mm) // from user | |
5216 | mmgrab(prev->active_mm); | |
5217 | else | |
5218 | prev->active_mm = NULL; | |
5219 | } else { // to user | |
227a4aad | 5220 | membarrier_switch_mm(rq, prev->active_mm, next->mm); |
139d025c PZ |
5221 | /* |
5222 | * sys_membarrier() requires an smp_mb() between setting | |
227a4aad | 5223 | * rq->curr / membarrier_switch_mm() and returning to userspace. |
139d025c PZ |
5224 | * |
5225 | * The below provides this either through switch_mm(), or in | |
5226 | * case 'prev->active_mm == next->mm' through | |
5227 | * finish_task_switch()'s mmdrop(). | |
5228 | */ | |
139d025c | 5229 | switch_mm_irqs_off(prev->active_mm, next->mm, next); |
bd74fdae | 5230 | lru_gen_use_mm(next->mm); |
1da177e4 | 5231 | |
139d025c PZ |
5232 | if (!prev->mm) { // from kernel |
5233 | /* will mmdrop() in finish_task_switch(). */ | |
5234 | rq->prev_mm = prev->active_mm; | |
5235 | prev->active_mm = NULL; | |
5236 | } | |
1da177e4 | 5237 | } |
92509b73 | 5238 | |
cb42c9a3 | 5239 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
92509b73 | 5240 | |
269d5992 | 5241 | prepare_lock_switch(rq, next, rf); |
1da177e4 LT |
5242 | |
5243 | /* Here we just switch the register state and the stack. */ | |
5244 | switch_to(prev, next, prev); | |
dd41f596 | 5245 | barrier(); |
dfa50b60 ON |
5246 | |
5247 | return finish_task_switch(prev); | |
1da177e4 LT |
5248 | } |
5249 | ||
5250 | /* | |
1c3e8264 | 5251 | * nr_running and nr_context_switches: |
1da177e4 LT |
5252 | * |
5253 | * externally visible scheduler statistics: current number of runnable | |
1c3e8264 | 5254 | * threads, total number of context switches performed since bootup. |
1da177e4 | 5255 | */ |
01aee8fd | 5256 | unsigned int nr_running(void) |
1da177e4 | 5257 | { |
01aee8fd | 5258 | unsigned int i, sum = 0; |
1da177e4 LT |
5259 | |
5260 | for_each_online_cpu(i) | |
5261 | sum += cpu_rq(i)->nr_running; | |
5262 | ||
5263 | return sum; | |
f711f609 | 5264 | } |
1da177e4 | 5265 | |
2ee507c4 | 5266 | /* |
d1ccc66d | 5267 | * Check if only the current task is running on the CPU. |
00cc1633 DD |
5268 | * |
5269 | * Caution: this function does not check that the caller has disabled | |
5270 | * preemption, thus the result might have a time-of-check-to-time-of-use | |
5271 | * race. The caller is responsible to use it correctly, for example: | |
5272 | * | |
dfcb245e | 5273 | * - from a non-preemptible section (of course) |
00cc1633 DD |
5274 | * |
5275 | * - from a thread that is bound to a single CPU | |
5276 | * | |
5277 | * - in a loop with very short iterations (e.g. a polling loop) | |
2ee507c4 TC |
5278 | */ |
5279 | bool single_task_running(void) | |
5280 | { | |
00cc1633 | 5281 | return raw_rq()->nr_running == 1; |
2ee507c4 TC |
5282 | } |
5283 | EXPORT_SYMBOL(single_task_running); | |
5284 | ||
1da177e4 | 5285 | unsigned long long nr_context_switches(void) |
46cb4b7c | 5286 | { |
cc94abfc SR |
5287 | int i; |
5288 | unsigned long long sum = 0; | |
46cb4b7c | 5289 | |
0a945022 | 5290 | for_each_possible_cpu(i) |
1da177e4 | 5291 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 5292 | |
1da177e4 LT |
5293 | return sum; |
5294 | } | |
483b4ee6 | 5295 | |
145d952a DL |
5296 | /* |
5297 | * Consumers of these two interfaces, like for example the cpuidle menu | |
5298 | * governor, are using nonsensical data. Preferring shallow idle state selection | |
5299 | * for a CPU that has IO-wait which might not even end up running the task when | |
5300 | * it does become runnable. | |
5301 | */ | |
5302 | ||
8fc2858e | 5303 | unsigned int nr_iowait_cpu(int cpu) |
145d952a DL |
5304 | { |
5305 | return atomic_read(&cpu_rq(cpu)->nr_iowait); | |
5306 | } | |
5307 | ||
e33a9bba | 5308 | /* |
b19a888c | 5309 | * IO-wait accounting, and how it's mostly bollocks (on SMP). |
e33a9bba TH |
5310 | * |
5311 | * The idea behind IO-wait account is to account the idle time that we could | |
5312 | * have spend running if it were not for IO. That is, if we were to improve the | |
5313 | * storage performance, we'd have a proportional reduction in IO-wait time. | |
5314 | * | |
5315 | * This all works nicely on UP, where, when a task blocks on IO, we account | |
5316 | * idle time as IO-wait, because if the storage were faster, it could've been | |
5317 | * running and we'd not be idle. | |
5318 | * | |
5319 | * This has been extended to SMP, by doing the same for each CPU. This however | |
5320 | * is broken. | |
5321 | * | |
5322 | * Imagine for instance the case where two tasks block on one CPU, only the one | |
5323 | * CPU will have IO-wait accounted, while the other has regular idle. Even | |
5324 | * though, if the storage were faster, both could've ran at the same time, | |
5325 | * utilising both CPUs. | |
5326 | * | |
5327 | * This means, that when looking globally, the current IO-wait accounting on | |
5328 | * SMP is a lower bound, by reason of under accounting. | |
5329 | * | |
5330 | * Worse, since the numbers are provided per CPU, they are sometimes | |
5331 | * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly | |
5332 | * associated with any one particular CPU, it can wake to another CPU than it | |
5333 | * blocked on. This means the per CPU IO-wait number is meaningless. | |
5334 | * | |
5335 | * Task CPU affinities can make all that even more 'interesting'. | |
5336 | */ | |
5337 | ||
97455168 | 5338 | unsigned int nr_iowait(void) |
1da177e4 | 5339 | { |
97455168 | 5340 | unsigned int i, sum = 0; |
483b4ee6 | 5341 | |
0a945022 | 5342 | for_each_possible_cpu(i) |
145d952a | 5343 | sum += nr_iowait_cpu(i); |
46cb4b7c | 5344 | |
1da177e4 LT |
5345 | return sum; |
5346 | } | |
483b4ee6 | 5347 | |
dd41f596 | 5348 | #ifdef CONFIG_SMP |
8a0be9ef | 5349 | |
46cb4b7c | 5350 | /* |
38022906 PZ |
5351 | * sched_exec - execve() is a valuable balancing opportunity, because at |
5352 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 5353 | */ |
38022906 | 5354 | void sched_exec(void) |
46cb4b7c | 5355 | { |
38022906 | 5356 | struct task_struct *p = current; |
1da177e4 | 5357 | unsigned long flags; |
0017d735 | 5358 | int dest_cpu; |
46cb4b7c | 5359 | |
8f42ced9 | 5360 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3aef1551 | 5361 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), WF_EXEC); |
0017d735 PZ |
5362 | if (dest_cpu == smp_processor_id()) |
5363 | goto unlock; | |
38022906 | 5364 | |
8f42ced9 | 5365 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 5366 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 5367 | |
8f42ced9 PZ |
5368 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5369 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
5370 | return; |
5371 | } | |
0017d735 | 5372 | unlock: |
8f42ced9 | 5373 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 5374 | } |
dd41f596 | 5375 | |
1da177e4 LT |
5376 | #endif |
5377 | ||
1da177e4 | 5378 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3292beb3 | 5379 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
1da177e4 LT |
5380 | |
5381 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3292beb3 | 5382 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
1da177e4 | 5383 | |
6075620b GG |
5384 | /* |
5385 | * The function fair_sched_class.update_curr accesses the struct curr | |
5386 | * and its field curr->exec_start; when called from task_sched_runtime(), | |
5387 | * we observe a high rate of cache misses in practice. | |
5388 | * Prefetching this data results in improved performance. | |
5389 | */ | |
5390 | static inline void prefetch_curr_exec_start(struct task_struct *p) | |
5391 | { | |
5392 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
5393 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; | |
5394 | #else | |
5395 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; | |
5396 | #endif | |
5397 | prefetch(curr); | |
5398 | prefetch(&curr->exec_start); | |
5399 | } | |
5400 | ||
c5f8d995 HS |
5401 | /* |
5402 | * Return accounted runtime for the task. | |
5403 | * In case the task is currently running, return the runtime plus current's | |
5404 | * pending runtime that have not been accounted yet. | |
5405 | */ | |
5406 | unsigned long long task_sched_runtime(struct task_struct *p) | |
5407 | { | |
eb580751 | 5408 | struct rq_flags rf; |
c5f8d995 | 5409 | struct rq *rq; |
6e998916 | 5410 | u64 ns; |
c5f8d995 | 5411 | |
911b2898 PZ |
5412 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
5413 | /* | |
97fb7a0a | 5414 | * 64-bit doesn't need locks to atomically read a 64-bit value. |
911b2898 PZ |
5415 | * So we have a optimization chance when the task's delta_exec is 0. |
5416 | * Reading ->on_cpu is racy, but this is ok. | |
5417 | * | |
d1ccc66d IM |
5418 | * If we race with it leaving CPU, we'll take a lock. So we're correct. |
5419 | * If we race with it entering CPU, unaccounted time is 0. This is | |
911b2898 | 5420 | * indistinguishable from the read occurring a few cycles earlier. |
4036ac15 MG |
5421 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
5422 | * been accounted, so we're correct here as well. | |
911b2898 | 5423 | */ |
da0c1e65 | 5424 | if (!p->on_cpu || !task_on_rq_queued(p)) |
911b2898 PZ |
5425 | return p->se.sum_exec_runtime; |
5426 | #endif | |
5427 | ||
eb580751 | 5428 | rq = task_rq_lock(p, &rf); |
6e998916 SG |
5429 | /* |
5430 | * Must be ->curr _and_ ->on_rq. If dequeued, we would | |
5431 | * project cycles that may never be accounted to this | |
5432 | * thread, breaking clock_gettime(). | |
5433 | */ | |
5434 | if (task_current(rq, p) && task_on_rq_queued(p)) { | |
6075620b | 5435 | prefetch_curr_exec_start(p); |
6e998916 SG |
5436 | update_rq_clock(rq); |
5437 | p->sched_class->update_curr(rq); | |
5438 | } | |
5439 | ns = p->se.sum_exec_runtime; | |
eb580751 | 5440 | task_rq_unlock(rq, p, &rf); |
c5f8d995 HS |
5441 | |
5442 | return ns; | |
5443 | } | |
48f24c4d | 5444 | |
c006fac5 PT |
5445 | #ifdef CONFIG_SCHED_DEBUG |
5446 | static u64 cpu_resched_latency(struct rq *rq) | |
5447 | { | |
5448 | int latency_warn_ms = READ_ONCE(sysctl_resched_latency_warn_ms); | |
5449 | u64 resched_latency, now = rq_clock(rq); | |
5450 | static bool warned_once; | |
5451 | ||
5452 | if (sysctl_resched_latency_warn_once && warned_once) | |
5453 | return 0; | |
5454 | ||
5455 | if (!need_resched() || !latency_warn_ms) | |
5456 | return 0; | |
5457 | ||
5458 | if (system_state == SYSTEM_BOOTING) | |
5459 | return 0; | |
5460 | ||
5461 | if (!rq->last_seen_need_resched_ns) { | |
5462 | rq->last_seen_need_resched_ns = now; | |
5463 | rq->ticks_without_resched = 0; | |
5464 | return 0; | |
5465 | } | |
5466 | ||
5467 | rq->ticks_without_resched++; | |
5468 | resched_latency = now - rq->last_seen_need_resched_ns; | |
5469 | if (resched_latency <= latency_warn_ms * NSEC_PER_MSEC) | |
5470 | return 0; | |
5471 | ||
5472 | warned_once = true; | |
5473 | ||
5474 | return resched_latency; | |
5475 | } | |
5476 | ||
5477 | static int __init setup_resched_latency_warn_ms(char *str) | |
5478 | { | |
5479 | long val; | |
5480 | ||
5481 | if ((kstrtol(str, 0, &val))) { | |
5482 | pr_warn("Unable to set resched_latency_warn_ms\n"); | |
5483 | return 1; | |
5484 | } | |
5485 | ||
5486 | sysctl_resched_latency_warn_ms = val; | |
5487 | return 1; | |
5488 | } | |
5489 | __setup("resched_latency_warn_ms=", setup_resched_latency_warn_ms); | |
5490 | #else | |
5491 | static inline u64 cpu_resched_latency(struct rq *rq) { return 0; } | |
5492 | #endif /* CONFIG_SCHED_DEBUG */ | |
5493 | ||
7835b98b CL |
5494 | /* |
5495 | * This function gets called by the timer code, with HZ frequency. | |
5496 | * We call it with interrupts disabled. | |
7835b98b CL |
5497 | */ |
5498 | void scheduler_tick(void) | |
5499 | { | |
7835b98b CL |
5500 | int cpu = smp_processor_id(); |
5501 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5502 | struct task_struct *curr = rq->curr; |
8a8c69c3 | 5503 | struct rq_flags rf; |
b4eccf5f | 5504 | unsigned long thermal_pressure; |
c006fac5 | 5505 | u64 resched_latency; |
3e51f33f | 5506 | |
1567c3e3 | 5507 | arch_scale_freq_tick(); |
3e51f33f | 5508 | sched_clock_tick(); |
dd41f596 | 5509 | |
8a8c69c3 PZ |
5510 | rq_lock(rq, &rf); |
5511 | ||
3e51f33f | 5512 | update_rq_clock(rq); |
b4eccf5f | 5513 | thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq)); |
05289b90 | 5514 | update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure); |
fa85ae24 | 5515 | curr->sched_class->task_tick(rq, curr, 0); |
c006fac5 PT |
5516 | if (sched_feat(LATENCY_WARN)) |
5517 | resched_latency = cpu_resched_latency(rq); | |
3289bdb4 | 5518 | calc_global_load_tick(rq); |
4feee7d1 | 5519 | sched_core_tick(rq); |
8a8c69c3 PZ |
5520 | |
5521 | rq_unlock(rq, &rf); | |
7835b98b | 5522 | |
c006fac5 PT |
5523 | if (sched_feat(LATENCY_WARN) && resched_latency) |
5524 | resched_latency_warn(cpu, resched_latency); | |
5525 | ||
e9d2b064 | 5526 | perf_event_task_tick(); |
e220d2dc | 5527 | |
e418e1c2 | 5528 | #ifdef CONFIG_SMP |
6eb57e0d | 5529 | rq->idle_balance = idle_cpu(cpu); |
7caff66f | 5530 | trigger_load_balance(rq); |
e418e1c2 | 5531 | #endif |
1da177e4 LT |
5532 | } |
5533 | ||
265f22a9 | 5534 | #ifdef CONFIG_NO_HZ_FULL |
d84b3131 FW |
5535 | |
5536 | struct tick_work { | |
5537 | int cpu; | |
b55bd585 | 5538 | atomic_t state; |
d84b3131 FW |
5539 | struct delayed_work work; |
5540 | }; | |
b55bd585 PM |
5541 | /* Values for ->state, see diagram below. */ |
5542 | #define TICK_SCHED_REMOTE_OFFLINE 0 | |
5543 | #define TICK_SCHED_REMOTE_OFFLINING 1 | |
5544 | #define TICK_SCHED_REMOTE_RUNNING 2 | |
5545 | ||
5546 | /* | |
5547 | * State diagram for ->state: | |
5548 | * | |
5549 | * | |
5550 | * TICK_SCHED_REMOTE_OFFLINE | |
5551 | * | ^ | |
5552 | * | | | |
5553 | * | | sched_tick_remote() | |
5554 | * | | | |
5555 | * | | | |
5556 | * +--TICK_SCHED_REMOTE_OFFLINING | |
5557 | * | ^ | |
5558 | * | | | |
5559 | * sched_tick_start() | | sched_tick_stop() | |
5560 | * | | | |
5561 | * V | | |
5562 | * TICK_SCHED_REMOTE_RUNNING | |
5563 | * | |
5564 | * | |
5565 | * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote() | |
5566 | * and sched_tick_start() are happy to leave the state in RUNNING. | |
5567 | */ | |
d84b3131 FW |
5568 | |
5569 | static struct tick_work __percpu *tick_work_cpu; | |
5570 | ||
5571 | static void sched_tick_remote(struct work_struct *work) | |
5572 | { | |
5573 | struct delayed_work *dwork = to_delayed_work(work); | |
5574 | struct tick_work *twork = container_of(dwork, struct tick_work, work); | |
5575 | int cpu = twork->cpu; | |
5576 | struct rq *rq = cpu_rq(cpu); | |
d9c0ffca | 5577 | struct task_struct *curr; |
d84b3131 | 5578 | struct rq_flags rf; |
d9c0ffca | 5579 | u64 delta; |
b55bd585 | 5580 | int os; |
d84b3131 FW |
5581 | |
5582 | /* | |
5583 | * Handle the tick only if it appears the remote CPU is running in full | |
5584 | * dynticks mode. The check is racy by nature, but missing a tick or | |
5585 | * having one too much is no big deal because the scheduler tick updates | |
5586 | * statistics and checks timeslices in a time-independent way, regardless | |
5587 | * of when exactly it is running. | |
5588 | */ | |
488603b8 | 5589 | if (!tick_nohz_tick_stopped_cpu(cpu)) |
d9c0ffca | 5590 | goto out_requeue; |
d84b3131 | 5591 | |
d9c0ffca FW |
5592 | rq_lock_irq(rq, &rf); |
5593 | curr = rq->curr; | |
488603b8 | 5594 | if (cpu_is_offline(cpu)) |
d9c0ffca | 5595 | goto out_unlock; |
d84b3131 | 5596 | |
d9c0ffca | 5597 | update_rq_clock(rq); |
d9c0ffca | 5598 | |
488603b8 SW |
5599 | if (!is_idle_task(curr)) { |
5600 | /* | |
5601 | * Make sure the next tick runs within a reasonable | |
5602 | * amount of time. | |
5603 | */ | |
5604 | delta = rq_clock_task(rq) - curr->se.exec_start; | |
5605 | WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3); | |
5606 | } | |
d9c0ffca FW |
5607 | curr->sched_class->task_tick(rq, curr, 0); |
5608 | ||
ebc0f83c | 5609 | calc_load_nohz_remote(rq); |
d9c0ffca FW |
5610 | out_unlock: |
5611 | rq_unlock_irq(rq, &rf); | |
d9c0ffca | 5612 | out_requeue: |
ebc0f83c | 5613 | |
d84b3131 FW |
5614 | /* |
5615 | * Run the remote tick once per second (1Hz). This arbitrary | |
5616 | * frequency is large enough to avoid overload but short enough | |
b55bd585 PM |
5617 | * to keep scheduler internal stats reasonably up to date. But |
5618 | * first update state to reflect hotplug activity if required. | |
d84b3131 | 5619 | */ |
b55bd585 PM |
5620 | os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING); |
5621 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE); | |
5622 | if (os == TICK_SCHED_REMOTE_RUNNING) | |
5623 | queue_delayed_work(system_unbound_wq, dwork, HZ); | |
d84b3131 FW |
5624 | } |
5625 | ||
5626 | static void sched_tick_start(int cpu) | |
5627 | { | |
b55bd585 | 5628 | int os; |
d84b3131 FW |
5629 | struct tick_work *twork; |
5630 | ||
04d4e665 | 5631 | if (housekeeping_cpu(cpu, HK_TYPE_TICK)) |
d84b3131 FW |
5632 | return; |
5633 | ||
5634 | WARN_ON_ONCE(!tick_work_cpu); | |
5635 | ||
5636 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5637 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING); |
5638 | WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING); | |
5639 | if (os == TICK_SCHED_REMOTE_OFFLINE) { | |
5640 | twork->cpu = cpu; | |
5641 | INIT_DELAYED_WORK(&twork->work, sched_tick_remote); | |
5642 | queue_delayed_work(system_unbound_wq, &twork->work, HZ); | |
5643 | } | |
d84b3131 FW |
5644 | } |
5645 | ||
5646 | #ifdef CONFIG_HOTPLUG_CPU | |
5647 | static void sched_tick_stop(int cpu) | |
5648 | { | |
5649 | struct tick_work *twork; | |
b55bd585 | 5650 | int os; |
d84b3131 | 5651 | |
04d4e665 | 5652 | if (housekeeping_cpu(cpu, HK_TYPE_TICK)) |
d84b3131 FW |
5653 | return; |
5654 | ||
5655 | WARN_ON_ONCE(!tick_work_cpu); | |
5656 | ||
5657 | twork = per_cpu_ptr(tick_work_cpu, cpu); | |
b55bd585 PM |
5658 | /* There cannot be competing actions, but don't rely on stop-machine. */ |
5659 | os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING); | |
5660 | WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING); | |
5661 | /* Don't cancel, as this would mess up the state machine. */ | |
d84b3131 FW |
5662 | } |
5663 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5664 | ||
5665 | int __init sched_tick_offload_init(void) | |
5666 | { | |
5667 | tick_work_cpu = alloc_percpu(struct tick_work); | |
5668 | BUG_ON(!tick_work_cpu); | |
d84b3131 FW |
5669 | return 0; |
5670 | } | |
5671 | ||
5672 | #else /* !CONFIG_NO_HZ_FULL */ | |
5673 | static inline void sched_tick_start(int cpu) { } | |
5674 | static inline void sched_tick_stop(int cpu) { } | |
265f22a9 | 5675 | #endif |
1da177e4 | 5676 | |
c1a280b6 | 5677 | #if defined(CONFIG_PREEMPTION) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
c3bc8fd6 | 5678 | defined(CONFIG_TRACE_PREEMPT_TOGGLE)) |
47252cfb SR |
5679 | /* |
5680 | * If the value passed in is equal to the current preempt count | |
5681 | * then we just disabled preemption. Start timing the latency. | |
5682 | */ | |
5683 | static inline void preempt_latency_start(int val) | |
5684 | { | |
5685 | if (preempt_count() == val) { | |
5686 | unsigned long ip = get_lock_parent_ip(); | |
5687 | #ifdef CONFIG_DEBUG_PREEMPT | |
5688 | current->preempt_disable_ip = ip; | |
5689 | #endif | |
5690 | trace_preempt_off(CALLER_ADDR0, ip); | |
5691 | } | |
5692 | } | |
7e49fcce | 5693 | |
edafe3a5 | 5694 | void preempt_count_add(int val) |
1da177e4 | 5695 | { |
6cd8a4bb | 5696 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5697 | /* |
5698 | * Underflow? | |
5699 | */ | |
9a11b49a IM |
5700 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5701 | return; | |
6cd8a4bb | 5702 | #endif |
bdb43806 | 5703 | __preempt_count_add(val); |
6cd8a4bb | 5704 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5705 | /* |
5706 | * Spinlock count overflowing soon? | |
5707 | */ | |
33859f7f MOS |
5708 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5709 | PREEMPT_MASK - 10); | |
6cd8a4bb | 5710 | #endif |
47252cfb | 5711 | preempt_latency_start(val); |
1da177e4 | 5712 | } |
bdb43806 | 5713 | EXPORT_SYMBOL(preempt_count_add); |
edafe3a5 | 5714 | NOKPROBE_SYMBOL(preempt_count_add); |
1da177e4 | 5715 | |
47252cfb SR |
5716 | /* |
5717 | * If the value passed in equals to the current preempt count | |
5718 | * then we just enabled preemption. Stop timing the latency. | |
5719 | */ | |
5720 | static inline void preempt_latency_stop(int val) | |
5721 | { | |
5722 | if (preempt_count() == val) | |
5723 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); | |
5724 | } | |
5725 | ||
edafe3a5 | 5726 | void preempt_count_sub(int val) |
1da177e4 | 5727 | { |
6cd8a4bb | 5728 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5729 | /* |
5730 | * Underflow? | |
5731 | */ | |
01e3eb82 | 5732 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5733 | return; |
1da177e4 LT |
5734 | /* |
5735 | * Is the spinlock portion underflowing? | |
5736 | */ | |
9a11b49a IM |
5737 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5738 | !(preempt_count() & PREEMPT_MASK))) | |
5739 | return; | |
6cd8a4bb | 5740 | #endif |
9a11b49a | 5741 | |
47252cfb | 5742 | preempt_latency_stop(val); |
bdb43806 | 5743 | __preempt_count_sub(val); |
1da177e4 | 5744 | } |
bdb43806 | 5745 | EXPORT_SYMBOL(preempt_count_sub); |
edafe3a5 | 5746 | NOKPROBE_SYMBOL(preempt_count_sub); |
1da177e4 | 5747 | |
47252cfb SR |
5748 | #else |
5749 | static inline void preempt_latency_start(int val) { } | |
5750 | static inline void preempt_latency_stop(int val) { } | |
1da177e4 LT |
5751 | #endif |
5752 | ||
59ddbcb2 IM |
5753 | static inline unsigned long get_preempt_disable_ip(struct task_struct *p) |
5754 | { | |
5755 | #ifdef CONFIG_DEBUG_PREEMPT | |
5756 | return p->preempt_disable_ip; | |
5757 | #else | |
5758 | return 0; | |
5759 | #endif | |
5760 | } | |
5761 | ||
1da177e4 | 5762 | /* |
dd41f596 | 5763 | * Print scheduling while atomic bug: |
1da177e4 | 5764 | */ |
dd41f596 | 5765 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5766 | { |
d1c6d149 VN |
5767 | /* Save this before calling printk(), since that will clobber it */ |
5768 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); | |
5769 | ||
664dfa65 DJ |
5770 | if (oops_in_progress) |
5771 | return; | |
5772 | ||
3df0fc5b PZ |
5773 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
5774 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 5775 | |
dd41f596 | 5776 | debug_show_held_locks(prev); |
e21f5b15 | 5777 | print_modules(); |
dd41f596 IM |
5778 | if (irqs_disabled()) |
5779 | print_irqtrace_events(prev); | |
d1c6d149 VN |
5780 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
5781 | && in_atomic_preempt_off()) { | |
8f47b187 | 5782 | pr_err("Preemption disabled at:"); |
2062a4e8 | 5783 | print_ip_sym(KERN_ERR, preempt_disable_ip); |
8f47b187 | 5784 | } |
79cc1ba7 | 5785 | check_panic_on_warn("scheduling while atomic"); |
748c7201 | 5786 | |
6135fc1e | 5787 | dump_stack(); |
373d4d09 | 5788 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
dd41f596 | 5789 | } |
1da177e4 | 5790 | |
dd41f596 IM |
5791 | /* |
5792 | * Various schedule()-time debugging checks and statistics: | |
5793 | */ | |
312364f3 | 5794 | static inline void schedule_debug(struct task_struct *prev, bool preempt) |
dd41f596 | 5795 | { |
0d9e2632 | 5796 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
29d64551 JH |
5797 | if (task_stack_end_corrupted(prev)) |
5798 | panic("corrupted stack end detected inside scheduler\n"); | |
88485be5 WD |
5799 | |
5800 | if (task_scs_end_corrupted(prev)) | |
5801 | panic("corrupted shadow stack detected inside scheduler\n"); | |
0d9e2632 | 5802 | #endif |
b99def8b | 5803 | |
312364f3 | 5804 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
2f064a59 | 5805 | if (!preempt && READ_ONCE(prev->__state) && prev->non_block_count) { |
312364f3 DV |
5806 | printk(KERN_ERR "BUG: scheduling in a non-blocking section: %s/%d/%i\n", |
5807 | prev->comm, prev->pid, prev->non_block_count); | |
5808 | dump_stack(); | |
5809 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
5810 | } | |
5811 | #endif | |
5812 | ||
1dc0fffc | 5813 | if (unlikely(in_atomic_preempt_off())) { |
dd41f596 | 5814 | __schedule_bug(prev); |
1dc0fffc PZ |
5815 | preempt_count_set(PREEMPT_DISABLED); |
5816 | } | |
b3fbab05 | 5817 | rcu_sleep_check(); |
9f68b5b7 | 5818 | SCHED_WARN_ON(ct_state() == CONTEXT_USER); |
dd41f596 | 5819 | |
1da177e4 LT |
5820 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5821 | ||
ae92882e | 5822 | schedstat_inc(this_rq()->sched_count); |
dd41f596 IM |
5823 | } |
5824 | ||
457d1f46 CY |
5825 | static void put_prev_task_balance(struct rq *rq, struct task_struct *prev, |
5826 | struct rq_flags *rf) | |
5827 | { | |
5828 | #ifdef CONFIG_SMP | |
5829 | const struct sched_class *class; | |
5830 | /* | |
5831 | * We must do the balancing pass before put_prev_task(), such | |
5832 | * that when we release the rq->lock the task is in the same | |
5833 | * state as before we took rq->lock. | |
5834 | * | |
5835 | * We can terminate the balance pass as soon as we know there is | |
5836 | * a runnable task of @class priority or higher. | |
5837 | */ | |
5838 | for_class_range(class, prev->sched_class, &idle_sched_class) { | |
5839 | if (class->balance(rq, prev, rf)) | |
5840 | break; | |
5841 | } | |
5842 | #endif | |
5843 | ||
5844 | put_prev_task(rq, prev); | |
5845 | } | |
5846 | ||
dd41f596 IM |
5847 | /* |
5848 | * Pick up the highest-prio task: | |
5849 | */ | |
5850 | static inline struct task_struct * | |
539f6512 | 5851 | __pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) |
dd41f596 | 5852 | { |
49ee5768 | 5853 | const struct sched_class *class; |
dd41f596 | 5854 | struct task_struct *p; |
1da177e4 LT |
5855 | |
5856 | /* | |
0ba87bb2 PZ |
5857 | * Optimization: we know that if all tasks are in the fair class we can |
5858 | * call that function directly, but only if the @prev task wasn't of a | |
b19a888c | 5859 | * higher scheduling class, because otherwise those lose the |
0ba87bb2 | 5860 | * opportunity to pull in more work from other CPUs. |
1da177e4 | 5861 | */ |
546a3fee | 5862 | if (likely(!sched_class_above(prev->sched_class, &fair_sched_class) && |
0ba87bb2 PZ |
5863 | rq->nr_running == rq->cfs.h_nr_running)) { |
5864 | ||
5d7d6056 | 5865 | p = pick_next_task_fair(rq, prev, rf); |
6ccdc84b | 5866 | if (unlikely(p == RETRY_TASK)) |
67692435 | 5867 | goto restart; |
6ccdc84b | 5868 | |
1699949d | 5869 | /* Assume the next prioritized class is idle_sched_class */ |
5d7d6056 | 5870 | if (!p) { |
f488e105 | 5871 | put_prev_task(rq, prev); |
98c2f700 | 5872 | p = pick_next_task_idle(rq); |
f488e105 | 5873 | } |
6ccdc84b PZ |
5874 | |
5875 | return p; | |
1da177e4 LT |
5876 | } |
5877 | ||
67692435 | 5878 | restart: |
457d1f46 | 5879 | put_prev_task_balance(rq, prev, rf); |
67692435 | 5880 | |
34f971f6 | 5881 | for_each_class(class) { |
98c2f700 | 5882 | p = class->pick_next_task(rq); |
67692435 | 5883 | if (p) |
dd41f596 | 5884 | return p; |
dd41f596 | 5885 | } |
34f971f6 | 5886 | |
bc9ffef3 | 5887 | BUG(); /* The idle class should always have a runnable task. */ |
dd41f596 | 5888 | } |
1da177e4 | 5889 | |
9edeaea1 | 5890 | #ifdef CONFIG_SCHED_CORE |
539f6512 PZ |
5891 | static inline bool is_task_rq_idle(struct task_struct *t) |
5892 | { | |
5893 | return (task_rq(t)->idle == t); | |
5894 | } | |
5895 | ||
5896 | static inline bool cookie_equals(struct task_struct *a, unsigned long cookie) | |
5897 | { | |
5898 | return is_task_rq_idle(a) || (a->core_cookie == cookie); | |
5899 | } | |
5900 | ||
5901 | static inline bool cookie_match(struct task_struct *a, struct task_struct *b) | |
5902 | { | |
5903 | if (is_task_rq_idle(a) || is_task_rq_idle(b)) | |
5904 | return true; | |
5905 | ||
5906 | return a->core_cookie == b->core_cookie; | |
5907 | } | |
5908 | ||
bc9ffef3 | 5909 | static inline struct task_struct *pick_task(struct rq *rq) |
539f6512 | 5910 | { |
bc9ffef3 PZ |
5911 | const struct sched_class *class; |
5912 | struct task_struct *p; | |
539f6512 | 5913 | |
bc9ffef3 PZ |
5914 | for_each_class(class) { |
5915 | p = class->pick_task(rq); | |
5916 | if (p) | |
5917 | return p; | |
539f6512 PZ |
5918 | } |
5919 | ||
bc9ffef3 | 5920 | BUG(); /* The idle class should always have a runnable task. */ |
539f6512 PZ |
5921 | } |
5922 | ||
c6047c2e JFG |
5923 | extern void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi); |
5924 | ||
5b6547ed PZ |
5925 | static void queue_core_balance(struct rq *rq); |
5926 | ||
539f6512 PZ |
5927 | static struct task_struct * |
5928 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
5929 | { | |
bc9ffef3 | 5930 | struct task_struct *next, *p, *max = NULL; |
539f6512 | 5931 | const struct cpumask *smt_mask; |
c6047c2e | 5932 | bool fi_before = false; |
4feee7d1 | 5933 | bool core_clock_updated = (rq == rq->core); |
bc9ffef3 PZ |
5934 | unsigned long cookie; |
5935 | int i, cpu, occ = 0; | |
5936 | struct rq *rq_i; | |
539f6512 | 5937 | bool need_sync; |
539f6512 PZ |
5938 | |
5939 | if (!sched_core_enabled(rq)) | |
5940 | return __pick_next_task(rq, prev, rf); | |
5941 | ||
5942 | cpu = cpu_of(rq); | |
5943 | ||
5944 | /* Stopper task is switching into idle, no need core-wide selection. */ | |
5945 | if (cpu_is_offline(cpu)) { | |
5946 | /* | |
5947 | * Reset core_pick so that we don't enter the fastpath when | |
5948 | * coming online. core_pick would already be migrated to | |
5949 | * another cpu during offline. | |
5950 | */ | |
5951 | rq->core_pick = NULL; | |
5952 | return __pick_next_task(rq, prev, rf); | |
5953 | } | |
5954 | ||
5955 | /* | |
5956 | * If there were no {en,de}queues since we picked (IOW, the task | |
5957 | * pointers are all still valid), and we haven't scheduled the last | |
5958 | * pick yet, do so now. | |
5959 | * | |
5960 | * rq->core_pick can be NULL if no selection was made for a CPU because | |
5961 | * it was either offline or went offline during a sibling's core-wide | |
5962 | * selection. In this case, do a core-wide selection. | |
5963 | */ | |
5964 | if (rq->core->core_pick_seq == rq->core->core_task_seq && | |
5965 | rq->core->core_pick_seq != rq->core_sched_seq && | |
5966 | rq->core_pick) { | |
5967 | WRITE_ONCE(rq->core_sched_seq, rq->core->core_pick_seq); | |
5968 | ||
5969 | next = rq->core_pick; | |
5970 | if (next != prev) { | |
5971 | put_prev_task(rq, prev); | |
5972 | set_next_task(rq, next); | |
5973 | } | |
5974 | ||
5975 | rq->core_pick = NULL; | |
5b6547ed | 5976 | goto out; |
539f6512 PZ |
5977 | } |
5978 | ||
5979 | put_prev_task_balance(rq, prev, rf); | |
5980 | ||
5981 | smt_mask = cpu_smt_mask(cpu); | |
7afbba11 JFG |
5982 | need_sync = !!rq->core->core_cookie; |
5983 | ||
5984 | /* reset state */ | |
5985 | rq->core->core_cookie = 0UL; | |
4feee7d1 JD |
5986 | if (rq->core->core_forceidle_count) { |
5987 | if (!core_clock_updated) { | |
5988 | update_rq_clock(rq->core); | |
5989 | core_clock_updated = true; | |
5990 | } | |
5991 | sched_core_account_forceidle(rq); | |
5992 | /* reset after accounting force idle */ | |
5993 | rq->core->core_forceidle_start = 0; | |
5994 | rq->core->core_forceidle_count = 0; | |
5995 | rq->core->core_forceidle_occupation = 0; | |
7afbba11 JFG |
5996 | need_sync = true; |
5997 | fi_before = true; | |
7afbba11 | 5998 | } |
539f6512 PZ |
5999 | |
6000 | /* | |
6001 | * core->core_task_seq, core->core_pick_seq, rq->core_sched_seq | |
6002 | * | |
6003 | * @task_seq guards the task state ({en,de}queues) | |
6004 | * @pick_seq is the @task_seq we did a selection on | |
6005 | * @sched_seq is the @pick_seq we scheduled | |
6006 | * | |
6007 | * However, preemptions can cause multiple picks on the same task set. | |
6008 | * 'Fix' this by also increasing @task_seq for every pick. | |
6009 | */ | |
6010 | rq->core->core_task_seq++; | |
539f6512 | 6011 | |
7afbba11 JFG |
6012 | /* |
6013 | * Optimize for common case where this CPU has no cookies | |
6014 | * and there are no cookied tasks running on siblings. | |
6015 | */ | |
6016 | if (!need_sync) { | |
bc9ffef3 | 6017 | next = pick_task(rq); |
7afbba11 JFG |
6018 | if (!next->core_cookie) { |
6019 | rq->core_pick = NULL; | |
c6047c2e JFG |
6020 | /* |
6021 | * For robustness, update the min_vruntime_fi for | |
6022 | * unconstrained picks as well. | |
6023 | */ | |
6024 | WARN_ON_ONCE(fi_before); | |
6025 | task_vruntime_update(rq, next, false); | |
5b6547ed | 6026 | goto out_set_next; |
7afbba11 | 6027 | } |
8039e96f | 6028 | } |
7afbba11 | 6029 | |
bc9ffef3 PZ |
6030 | /* |
6031 | * For each thread: do the regular task pick and find the max prio task | |
6032 | * amongst them. | |
6033 | * | |
6034 | * Tie-break prio towards the current CPU | |
6035 | */ | |
6036 | for_each_cpu_wrap(i, smt_mask, cpu) { | |
6037 | rq_i = cpu_rq(i); | |
539f6512 | 6038 | |
4feee7d1 JD |
6039 | /* |
6040 | * Current cpu always has its clock updated on entrance to | |
6041 | * pick_next_task(). If the current cpu is not the core, | |
6042 | * the core may also have been updated above. | |
6043 | */ | |
6044 | if (i != cpu && (rq_i != rq->core || !core_clock_updated)) | |
539f6512 | 6045 | update_rq_clock(rq_i); |
bc9ffef3 PZ |
6046 | |
6047 | p = rq_i->core_pick = pick_task(rq_i); | |
6048 | if (!max || prio_less(max, p, fi_before)) | |
6049 | max = p; | |
539f6512 PZ |
6050 | } |
6051 | ||
bc9ffef3 PZ |
6052 | cookie = rq->core->core_cookie = max->core_cookie; |
6053 | ||
539f6512 | 6054 | /* |
bc9ffef3 PZ |
6055 | * For each thread: try and find a runnable task that matches @max or |
6056 | * force idle. | |
539f6512 | 6057 | */ |
bc9ffef3 PZ |
6058 | for_each_cpu(i, smt_mask) { |
6059 | rq_i = cpu_rq(i); | |
6060 | p = rq_i->core_pick; | |
539f6512 | 6061 | |
bc9ffef3 PZ |
6062 | if (!cookie_equals(p, cookie)) { |
6063 | p = NULL; | |
6064 | if (cookie) | |
6065 | p = sched_core_find(rq_i, cookie); | |
7afbba11 | 6066 | if (!p) |
bc9ffef3 PZ |
6067 | p = idle_sched_class.pick_task(rq_i); |
6068 | } | |
539f6512 | 6069 | |
bc9ffef3 | 6070 | rq_i->core_pick = p; |
d2dfa17b | 6071 | |
bc9ffef3 PZ |
6072 | if (p == rq_i->idle) { |
6073 | if (rq_i->nr_running) { | |
4feee7d1 | 6074 | rq->core->core_forceidle_count++; |
c6047c2e JFG |
6075 | if (!fi_before) |
6076 | rq->core->core_forceidle_seq++; | |
6077 | } | |
bc9ffef3 PZ |
6078 | } else { |
6079 | occ++; | |
539f6512 | 6080 | } |
539f6512 PZ |
6081 | } |
6082 | ||
4feee7d1 | 6083 | if (schedstat_enabled() && rq->core->core_forceidle_count) { |
b171501f | 6084 | rq->core->core_forceidle_start = rq_clock(rq->core); |
4feee7d1 JD |
6085 | rq->core->core_forceidle_occupation = occ; |
6086 | } | |
6087 | ||
539f6512 PZ |
6088 | rq->core->core_pick_seq = rq->core->core_task_seq; |
6089 | next = rq->core_pick; | |
6090 | rq->core_sched_seq = rq->core->core_pick_seq; | |
6091 | ||
6092 | /* Something should have been selected for current CPU */ | |
6093 | WARN_ON_ONCE(!next); | |
6094 | ||
6095 | /* | |
6096 | * Reschedule siblings | |
6097 | * | |
6098 | * NOTE: L1TF -- at this point we're no longer running the old task and | |
6099 | * sending an IPI (below) ensures the sibling will no longer be running | |
6100 | * their task. This ensures there is no inter-sibling overlap between | |
6101 | * non-matching user state. | |
6102 | */ | |
6103 | for_each_cpu(i, smt_mask) { | |
bc9ffef3 | 6104 | rq_i = cpu_rq(i); |
539f6512 PZ |
6105 | |
6106 | /* | |
6107 | * An online sibling might have gone offline before a task | |
6108 | * could be picked for it, or it might be offline but later | |
6109 | * happen to come online, but its too late and nothing was | |
6110 | * picked for it. That's Ok - it will pick tasks for itself, | |
6111 | * so ignore it. | |
6112 | */ | |
6113 | if (!rq_i->core_pick) | |
6114 | continue; | |
6115 | ||
c6047c2e JFG |
6116 | /* |
6117 | * Update for new !FI->FI transitions, or if continuing to be in !FI: | |
6118 | * fi_before fi update? | |
6119 | * 0 0 1 | |
6120 | * 0 1 1 | |
6121 | * 1 0 1 | |
6122 | * 1 1 0 | |
6123 | */ | |
4feee7d1 JD |
6124 | if (!(fi_before && rq->core->core_forceidle_count)) |
6125 | task_vruntime_update(rq_i, rq_i->core_pick, !!rq->core->core_forceidle_count); | |
539f6512 | 6126 | |
d2dfa17b PZ |
6127 | rq_i->core_pick->core_occupation = occ; |
6128 | ||
539f6512 PZ |
6129 | if (i == cpu) { |
6130 | rq_i->core_pick = NULL; | |
6131 | continue; | |
6132 | } | |
6133 | ||
6134 | /* Did we break L1TF mitigation requirements? */ | |
6135 | WARN_ON_ONCE(!cookie_match(next, rq_i->core_pick)); | |
6136 | ||
6137 | if (rq_i->curr == rq_i->core_pick) { | |
6138 | rq_i->core_pick = NULL; | |
6139 | continue; | |
6140 | } | |
6141 | ||
6142 | resched_curr(rq_i); | |
6143 | } | |
6144 | ||
5b6547ed | 6145 | out_set_next: |
539f6512 | 6146 | set_next_task(rq, next); |
5b6547ed PZ |
6147 | out: |
6148 | if (rq->core->core_forceidle_count && next == rq->idle) | |
6149 | queue_core_balance(rq); | |
6150 | ||
539f6512 PZ |
6151 | return next; |
6152 | } | |
9edeaea1 | 6153 | |
d2dfa17b PZ |
6154 | static bool try_steal_cookie(int this, int that) |
6155 | { | |
6156 | struct rq *dst = cpu_rq(this), *src = cpu_rq(that); | |
6157 | struct task_struct *p; | |
6158 | unsigned long cookie; | |
6159 | bool success = false; | |
6160 | ||
6161 | local_irq_disable(); | |
6162 | double_rq_lock(dst, src); | |
6163 | ||
6164 | cookie = dst->core->core_cookie; | |
6165 | if (!cookie) | |
6166 | goto unlock; | |
6167 | ||
6168 | if (dst->curr != dst->idle) | |
6169 | goto unlock; | |
6170 | ||
6171 | p = sched_core_find(src, cookie); | |
6172 | if (p == src->idle) | |
6173 | goto unlock; | |
6174 | ||
6175 | do { | |
6176 | if (p == src->core_pick || p == src->curr) | |
6177 | goto next; | |
6178 | ||
386ef214 | 6179 | if (!is_cpu_allowed(p, this)) |
d2dfa17b PZ |
6180 | goto next; |
6181 | ||
6182 | if (p->core_occupation > dst->idle->core_occupation) | |
6183 | goto next; | |
6184 | ||
d2dfa17b PZ |
6185 | deactivate_task(src, p, 0); |
6186 | set_task_cpu(p, this); | |
6187 | activate_task(dst, p, 0); | |
d2dfa17b PZ |
6188 | |
6189 | resched_curr(dst); | |
6190 | ||
6191 | success = true; | |
6192 | break; | |
6193 | ||
6194 | next: | |
6195 | p = sched_core_next(p, cookie); | |
6196 | } while (p); | |
6197 | ||
6198 | unlock: | |
6199 | double_rq_unlock(dst, src); | |
6200 | local_irq_enable(); | |
6201 | ||
6202 | return success; | |
6203 | } | |
6204 | ||
6205 | static bool steal_cookie_task(int cpu, struct sched_domain *sd) | |
6206 | { | |
6207 | int i; | |
6208 | ||
6209 | for_each_cpu_wrap(i, sched_domain_span(sd), cpu) { | |
6210 | if (i == cpu) | |
6211 | continue; | |
6212 | ||
6213 | if (need_resched()) | |
6214 | break; | |
6215 | ||
6216 | if (try_steal_cookie(cpu, i)) | |
6217 | return true; | |
6218 | } | |
6219 | ||
6220 | return false; | |
6221 | } | |
6222 | ||
6223 | static void sched_core_balance(struct rq *rq) | |
6224 | { | |
6225 | struct sched_domain *sd; | |
6226 | int cpu = cpu_of(rq); | |
6227 | ||
6228 | preempt_disable(); | |
6229 | rcu_read_lock(); | |
6230 | raw_spin_rq_unlock_irq(rq); | |
6231 | for_each_domain(cpu, sd) { | |
6232 | if (need_resched()) | |
6233 | break; | |
6234 | ||
6235 | if (steal_cookie_task(cpu, sd)) | |
6236 | break; | |
6237 | } | |
6238 | raw_spin_rq_lock_irq(rq); | |
6239 | rcu_read_unlock(); | |
6240 | preempt_enable(); | |
6241 | } | |
6242 | ||
8e5bad7d | 6243 | static DEFINE_PER_CPU(struct balance_callback, core_balance_head); |
d2dfa17b | 6244 | |
5b6547ed | 6245 | static void queue_core_balance(struct rq *rq) |
d2dfa17b PZ |
6246 | { |
6247 | if (!sched_core_enabled(rq)) | |
6248 | return; | |
6249 | ||
6250 | if (!rq->core->core_cookie) | |
6251 | return; | |
6252 | ||
6253 | if (!rq->nr_running) /* not forced idle */ | |
6254 | return; | |
6255 | ||
6256 | queue_balance_callback(rq, &per_cpu(core_balance_head, rq->cpu), sched_core_balance); | |
6257 | } | |
6258 | ||
3c474b32 | 6259 | static void sched_core_cpu_starting(unsigned int cpu) |
9edeaea1 PZ |
6260 | { |
6261 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
3c474b32 PZ |
6262 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; |
6263 | unsigned long flags; | |
6264 | int t; | |
9edeaea1 | 6265 | |
3c474b32 | 6266 | sched_core_lock(cpu, &flags); |
9edeaea1 | 6267 | |
3c474b32 PZ |
6268 | WARN_ON_ONCE(rq->core != rq); |
6269 | ||
6270 | /* if we're the first, we'll be our own leader */ | |
6271 | if (cpumask_weight(smt_mask) == 1) | |
6272 | goto unlock; | |
6273 | ||
6274 | /* find the leader */ | |
6275 | for_each_cpu(t, smt_mask) { | |
6276 | if (t == cpu) | |
6277 | continue; | |
6278 | rq = cpu_rq(t); | |
6279 | if (rq->core == rq) { | |
6280 | core_rq = rq; | |
6281 | break; | |
9edeaea1 | 6282 | } |
3c474b32 | 6283 | } |
9edeaea1 | 6284 | |
3c474b32 PZ |
6285 | if (WARN_ON_ONCE(!core_rq)) /* whoopsie */ |
6286 | goto unlock; | |
9edeaea1 | 6287 | |
3c474b32 PZ |
6288 | /* install and validate core_rq */ |
6289 | for_each_cpu(t, smt_mask) { | |
6290 | rq = cpu_rq(t); | |
9edeaea1 | 6291 | |
3c474b32 | 6292 | if (t == cpu) |
9edeaea1 | 6293 | rq->core = core_rq; |
3c474b32 PZ |
6294 | |
6295 | WARN_ON_ONCE(rq->core != core_rq); | |
9edeaea1 | 6296 | } |
3c474b32 PZ |
6297 | |
6298 | unlock: | |
6299 | sched_core_unlock(cpu, &flags); | |
9edeaea1 | 6300 | } |
3c474b32 PZ |
6301 | |
6302 | static void sched_core_cpu_deactivate(unsigned int cpu) | |
6303 | { | |
6304 | const struct cpumask *smt_mask = cpu_smt_mask(cpu); | |
6305 | struct rq *rq = cpu_rq(cpu), *core_rq = NULL; | |
6306 | unsigned long flags; | |
6307 | int t; | |
6308 | ||
6309 | sched_core_lock(cpu, &flags); | |
6310 | ||
6311 | /* if we're the last man standing, nothing to do */ | |
6312 | if (cpumask_weight(smt_mask) == 1) { | |
6313 | WARN_ON_ONCE(rq->core != rq); | |
6314 | goto unlock; | |
6315 | } | |
6316 | ||
6317 | /* if we're not the leader, nothing to do */ | |
6318 | if (rq->core != rq) | |
6319 | goto unlock; | |
6320 | ||
6321 | /* find a new leader */ | |
6322 | for_each_cpu(t, smt_mask) { | |
6323 | if (t == cpu) | |
6324 | continue; | |
6325 | core_rq = cpu_rq(t); | |
6326 | break; | |
6327 | } | |
6328 | ||
6329 | if (WARN_ON_ONCE(!core_rq)) /* impossible */ | |
6330 | goto unlock; | |
6331 | ||
6332 | /* copy the shared state to the new leader */ | |
4feee7d1 JD |
6333 | core_rq->core_task_seq = rq->core_task_seq; |
6334 | core_rq->core_pick_seq = rq->core_pick_seq; | |
6335 | core_rq->core_cookie = rq->core_cookie; | |
6336 | core_rq->core_forceidle_count = rq->core_forceidle_count; | |
6337 | core_rq->core_forceidle_seq = rq->core_forceidle_seq; | |
6338 | core_rq->core_forceidle_occupation = rq->core_forceidle_occupation; | |
6339 | ||
6340 | /* | |
6341 | * Accounting edge for forced idle is handled in pick_next_task(). | |
6342 | * Don't need another one here, since the hotplug thread shouldn't | |
6343 | * have a cookie. | |
6344 | */ | |
6345 | core_rq->core_forceidle_start = 0; | |
3c474b32 PZ |
6346 | |
6347 | /* install new leader */ | |
6348 | for_each_cpu(t, smt_mask) { | |
6349 | rq = cpu_rq(t); | |
6350 | rq->core = core_rq; | |
6351 | } | |
6352 | ||
6353 | unlock: | |
6354 | sched_core_unlock(cpu, &flags); | |
6355 | } | |
6356 | ||
6357 | static inline void sched_core_cpu_dying(unsigned int cpu) | |
6358 | { | |
6359 | struct rq *rq = cpu_rq(cpu); | |
6360 | ||
6361 | if (rq->core != rq) | |
6362 | rq->core = rq; | |
6363 | } | |
6364 | ||
9edeaea1 PZ |
6365 | #else /* !CONFIG_SCHED_CORE */ |
6366 | ||
6367 | static inline void sched_core_cpu_starting(unsigned int cpu) {} | |
3c474b32 PZ |
6368 | static inline void sched_core_cpu_deactivate(unsigned int cpu) {} |
6369 | static inline void sched_core_cpu_dying(unsigned int cpu) {} | |
9edeaea1 | 6370 | |
539f6512 PZ |
6371 | static struct task_struct * |
6372 | pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf) | |
6373 | { | |
6374 | return __pick_next_task(rq, prev, rf); | |
6375 | } | |
6376 | ||
9edeaea1 PZ |
6377 | #endif /* CONFIG_SCHED_CORE */ |
6378 | ||
b4bfa3fc TG |
6379 | /* |
6380 | * Constants for the sched_mode argument of __schedule(). | |
6381 | * | |
6382 | * The mode argument allows RT enabled kernels to differentiate a | |
6383 | * preemption from blocking on an 'sleeping' spin/rwlock. Note that | |
6384 | * SM_MASK_PREEMPT for !RT has all bits set, which allows the compiler to | |
6385 | * optimize the AND operation out and just check for zero. | |
6386 | */ | |
6387 | #define SM_NONE 0x0 | |
6388 | #define SM_PREEMPT 0x1 | |
6991436c TG |
6389 | #define SM_RTLOCK_WAIT 0x2 |
6390 | ||
6391 | #ifndef CONFIG_PREEMPT_RT | |
6392 | # define SM_MASK_PREEMPT (~0U) | |
6393 | #else | |
6394 | # define SM_MASK_PREEMPT SM_PREEMPT | |
6395 | #endif | |
b4bfa3fc | 6396 | |
dd41f596 | 6397 | /* |
c259e01a | 6398 | * __schedule() is the main scheduler function. |
edde96ea PE |
6399 | * |
6400 | * The main means of driving the scheduler and thus entering this function are: | |
6401 | * | |
6402 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. | |
6403 | * | |
6404 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return | |
6405 | * paths. For example, see arch/x86/entry_64.S. | |
6406 | * | |
6407 | * To drive preemption between tasks, the scheduler sets the flag in timer | |
6408 | * interrupt handler scheduler_tick(). | |
6409 | * | |
6410 | * 3. Wakeups don't really cause entry into schedule(). They add a | |
6411 | * task to the run-queue and that's it. | |
6412 | * | |
6413 | * Now, if the new task added to the run-queue preempts the current | |
6414 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets | |
6415 | * called on the nearest possible occasion: | |
6416 | * | |
c1a280b6 | 6417 | * - If the kernel is preemptible (CONFIG_PREEMPTION=y): |
edde96ea PE |
6418 | * |
6419 | * - in syscall or exception context, at the next outmost | |
6420 | * preempt_enable(). (this might be as soon as the wake_up()'s | |
6421 | * spin_unlock()!) | |
6422 | * | |
6423 | * - in IRQ context, return from interrupt-handler to | |
6424 | * preemptible context | |
6425 | * | |
c1a280b6 | 6426 | * - If the kernel is not preemptible (CONFIG_PREEMPTION is not set) |
edde96ea PE |
6427 | * then at the next: |
6428 | * | |
6429 | * - cond_resched() call | |
6430 | * - explicit schedule() call | |
6431 | * - return from syscall or exception to user-space | |
6432 | * - return from interrupt-handler to user-space | |
bfd9b2b5 | 6433 | * |
b30f0e3f | 6434 | * WARNING: must be called with preemption disabled! |
dd41f596 | 6435 | */ |
b4bfa3fc | 6436 | static void __sched notrace __schedule(unsigned int sched_mode) |
dd41f596 IM |
6437 | { |
6438 | struct task_struct *prev, *next; | |
67ca7bde | 6439 | unsigned long *switch_count; |
dbfb089d | 6440 | unsigned long prev_state; |
d8ac8971 | 6441 | struct rq_flags rf; |
dd41f596 | 6442 | struct rq *rq; |
31656519 | 6443 | int cpu; |
dd41f596 | 6444 | |
dd41f596 IM |
6445 | cpu = smp_processor_id(); |
6446 | rq = cpu_rq(cpu); | |
dd41f596 | 6447 | prev = rq->curr; |
dd41f596 | 6448 | |
b4bfa3fc | 6449 | schedule_debug(prev, !!sched_mode); |
1da177e4 | 6450 | |
e0ee463c | 6451 | if (sched_feat(HRTICK) || sched_feat(HRTICK_DL)) |
f333fdc9 | 6452 | hrtick_clear(rq); |
8f4d37ec | 6453 | |
46a5d164 | 6454 | local_irq_disable(); |
b4bfa3fc | 6455 | rcu_note_context_switch(!!sched_mode); |
46a5d164 | 6456 | |
e0acd0a6 ON |
6457 | /* |
6458 | * Make sure that signal_pending_state()->signal_pending() below | |
6459 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) | |
dbfb089d PZ |
6460 | * done by the caller to avoid the race with signal_wake_up(): |
6461 | * | |
6462 | * __set_current_state(@state) signal_wake_up() | |
6463 | * schedule() set_tsk_thread_flag(p, TIF_SIGPENDING) | |
6464 | * wake_up_state(p, state) | |
6465 | * LOCK rq->lock LOCK p->pi_state | |
6466 | * smp_mb__after_spinlock() smp_mb__after_spinlock() | |
6467 | * if (signal_pending_state()) if (p->state & @state) | |
306e0604 | 6468 | * |
dbfb089d | 6469 | * Also, the membarrier system call requires a full memory barrier |
306e0604 | 6470 | * after coming from user-space, before storing to rq->curr. |
e0acd0a6 | 6471 | */ |
8a8c69c3 | 6472 | rq_lock(rq, &rf); |
d89e588c | 6473 | smp_mb__after_spinlock(); |
1da177e4 | 6474 | |
d1ccc66d IM |
6475 | /* Promote REQ to ACT */ |
6476 | rq->clock_update_flags <<= 1; | |
bce4dc80 | 6477 | update_rq_clock(rq); |
9edfbfed | 6478 | |
246d86b5 | 6479 | switch_count = &prev->nivcsw; |
d136122f | 6480 | |
dbfb089d | 6481 | /* |
d136122f | 6482 | * We must load prev->state once (task_struct::state is volatile), such |
2500ad1c | 6483 | * that we form a control dependency vs deactivate_task() below. |
dbfb089d | 6484 | */ |
2f064a59 | 6485 | prev_state = READ_ONCE(prev->__state); |
b4bfa3fc | 6486 | if (!(sched_mode & SM_MASK_PREEMPT) && prev_state) { |
dbfb089d | 6487 | if (signal_pending_state(prev_state, prev)) { |
2f064a59 | 6488 | WRITE_ONCE(prev->__state, TASK_RUNNING); |
21aa9af0 | 6489 | } else { |
dbfb089d PZ |
6490 | prev->sched_contributes_to_load = |
6491 | (prev_state & TASK_UNINTERRUPTIBLE) && | |
6492 | !(prev_state & TASK_NOLOAD) && | |
f5d39b02 | 6493 | !(prev_state & TASK_FROZEN); |
dbfb089d PZ |
6494 | |
6495 | if (prev->sched_contributes_to_load) | |
6496 | rq->nr_uninterruptible++; | |
6497 | ||
6498 | /* | |
6499 | * __schedule() ttwu() | |
d136122f PZ |
6500 | * prev_state = prev->state; if (p->on_rq && ...) |
6501 | * if (prev_state) goto out; | |
6502 | * p->on_rq = 0; smp_acquire__after_ctrl_dep(); | |
6503 | * p->state = TASK_WAKING | |
6504 | * | |
6505 | * Where __schedule() and ttwu() have matching control dependencies. | |
dbfb089d PZ |
6506 | * |
6507 | * After this, schedule() must not care about p->state any more. | |
6508 | */ | |
bce4dc80 | 6509 | deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK); |
2acca55e | 6510 | |
e33a9bba TH |
6511 | if (prev->in_iowait) { |
6512 | atomic_inc(&rq->nr_iowait); | |
6513 | delayacct_blkio_start(); | |
6514 | } | |
21aa9af0 | 6515 | } |
dd41f596 | 6516 | switch_count = &prev->nvcsw; |
1da177e4 LT |
6517 | } |
6518 | ||
d8ac8971 | 6519 | next = pick_next_task(rq, prev, &rf); |
f26f9aff | 6520 | clear_tsk_need_resched(prev); |
f27dde8d | 6521 | clear_preempt_need_resched(); |
c006fac5 PT |
6522 | #ifdef CONFIG_SCHED_DEBUG |
6523 | rq->last_seen_need_resched_ns = 0; | |
6524 | #endif | |
1da177e4 | 6525 | |
1da177e4 | 6526 | if (likely(prev != next)) { |
1da177e4 | 6527 | rq->nr_switches++; |
5311a98f EB |
6528 | /* |
6529 | * RCU users of rcu_dereference(rq->curr) may not see | |
6530 | * changes to task_struct made by pick_next_task(). | |
6531 | */ | |
6532 | RCU_INIT_POINTER(rq->curr, next); | |
22e4ebb9 MD |
6533 | /* |
6534 | * The membarrier system call requires each architecture | |
6535 | * to have a full memory barrier after updating | |
306e0604 MD |
6536 | * rq->curr, before returning to user-space. |
6537 | * | |
6538 | * Here are the schemes providing that barrier on the | |
6539 | * various architectures: | |
6540 | * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC. | |
6541 | * switch_mm() rely on membarrier_arch_switch_mm() on PowerPC. | |
6542 | * - finish_lock_switch() for weakly-ordered | |
6543 | * architectures where spin_unlock is a full barrier, | |
6544 | * - switch_to() for arm64 (weakly-ordered, spin_unlock | |
6545 | * is a RELEASE barrier), | |
22e4ebb9 | 6546 | */ |
1da177e4 LT |
6547 | ++*switch_count; |
6548 | ||
af449901 | 6549 | migrate_disable_switch(rq, prev); |
b05e75d6 JW |
6550 | psi_sched_switch(prev, next, !task_on_rq_queued(prev)); |
6551 | ||
9c2136be | 6552 | trace_sched_switch(sched_mode & SM_MASK_PREEMPT, prev, next, prev_state); |
d1ccc66d IM |
6553 | |
6554 | /* Also unlocks the rq: */ | |
6555 | rq = context_switch(rq, prev, next, &rf); | |
cbce1a68 | 6556 | } else { |
cb42c9a3 | 6557 | rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP); |
1da177e4 | 6558 | |
565790d2 PZ |
6559 | rq_unpin_lock(rq, &rf); |
6560 | __balance_callbacks(rq); | |
5cb9eaa3 | 6561 | raw_spin_rq_unlock_irq(rq); |
565790d2 | 6562 | } |
1da177e4 | 6563 | } |
c259e01a | 6564 | |
9af6528e PZ |
6565 | void __noreturn do_task_dead(void) |
6566 | { | |
d1ccc66d | 6567 | /* Causes final put_task_struct in finish_task_switch(): */ |
b5bf9a90 | 6568 | set_special_state(TASK_DEAD); |
d1ccc66d IM |
6569 | |
6570 | /* Tell freezer to ignore us: */ | |
6571 | current->flags |= PF_NOFREEZE; | |
6572 | ||
b4bfa3fc | 6573 | __schedule(SM_NONE); |
9af6528e | 6574 | BUG(); |
d1ccc66d IM |
6575 | |
6576 | /* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */ | |
9af6528e | 6577 | for (;;) |
d1ccc66d | 6578 | cpu_relax(); |
9af6528e PZ |
6579 | } |
6580 | ||
9c40cef2 TG |
6581 | static inline void sched_submit_work(struct task_struct *tsk) |
6582 | { | |
c1cecf88 SAS |
6583 | unsigned int task_flags; |
6584 | ||
b03fbd4f | 6585 | if (task_is_running(tsk)) |
9c40cef2 | 6586 | return; |
6d25be57 | 6587 | |
c1cecf88 | 6588 | task_flags = tsk->flags; |
6d25be57 | 6589 | /* |
b945efcd TG |
6590 | * If a worker goes to sleep, notify and ask workqueue whether it |
6591 | * wants to wake up a task to maintain concurrency. | |
6d25be57 | 6592 | */ |
c1cecf88 | 6593 | if (task_flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
c1cecf88 | 6594 | if (task_flags & PF_WQ_WORKER) |
771b53d0 JA |
6595 | wq_worker_sleeping(tsk); |
6596 | else | |
6597 | io_wq_worker_sleeping(tsk); | |
6d25be57 TG |
6598 | } |
6599 | ||
401e4963 JK |
6600 | /* |
6601 | * spinlock and rwlock must not flush block requests. This will | |
6602 | * deadlock if the callback attempts to acquire a lock which is | |
6603 | * already acquired. | |
6604 | */ | |
6605 | SCHED_WARN_ON(current->__state & TASK_RTLOCK_WAIT); | |
b0fdc013 | 6606 | |
9c40cef2 TG |
6607 | /* |
6608 | * If we are going to sleep and we have plugged IO queued, | |
6609 | * make sure to submit it to avoid deadlocks. | |
6610 | */ | |
aa8dccca | 6611 | blk_flush_plug(tsk->plug, true); |
9c40cef2 TG |
6612 | } |
6613 | ||
6d25be57 TG |
6614 | static void sched_update_worker(struct task_struct *tsk) |
6615 | { | |
771b53d0 JA |
6616 | if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) { |
6617 | if (tsk->flags & PF_WQ_WORKER) | |
6618 | wq_worker_running(tsk); | |
6619 | else | |
6620 | io_wq_worker_running(tsk); | |
6621 | } | |
6d25be57 TG |
6622 | } |
6623 | ||
722a9f92 | 6624 | asmlinkage __visible void __sched schedule(void) |
c259e01a | 6625 | { |
9c40cef2 TG |
6626 | struct task_struct *tsk = current; |
6627 | ||
6628 | sched_submit_work(tsk); | |
bfd9b2b5 | 6629 | do { |
b30f0e3f | 6630 | preempt_disable(); |
b4bfa3fc | 6631 | __schedule(SM_NONE); |
b30f0e3f | 6632 | sched_preempt_enable_no_resched(); |
bfd9b2b5 | 6633 | } while (need_resched()); |
6d25be57 | 6634 | sched_update_worker(tsk); |
c259e01a | 6635 | } |
1da177e4 LT |
6636 | EXPORT_SYMBOL(schedule); |
6637 | ||
8663effb SRV |
6638 | /* |
6639 | * synchronize_rcu_tasks() makes sure that no task is stuck in preempted | |
6640 | * state (have scheduled out non-voluntarily) by making sure that all | |
6641 | * tasks have either left the run queue or have gone into user space. | |
6642 | * As idle tasks do not do either, they must not ever be preempted | |
6643 | * (schedule out non-voluntarily). | |
6644 | * | |
6645 | * schedule_idle() is similar to schedule_preempt_disable() except that it | |
6646 | * never enables preemption because it does not call sched_submit_work(). | |
6647 | */ | |
6648 | void __sched schedule_idle(void) | |
6649 | { | |
6650 | /* | |
6651 | * As this skips calling sched_submit_work(), which the idle task does | |
6652 | * regardless because that function is a nop when the task is in a | |
6653 | * TASK_RUNNING state, make sure this isn't used someplace that the | |
6654 | * current task can be in any other state. Note, idle is always in the | |
6655 | * TASK_RUNNING state. | |
6656 | */ | |
2f064a59 | 6657 | WARN_ON_ONCE(current->__state); |
8663effb | 6658 | do { |
b4bfa3fc | 6659 | __schedule(SM_NONE); |
8663effb SRV |
6660 | } while (need_resched()); |
6661 | } | |
6662 | ||
24a9c541 | 6663 | #if defined(CONFIG_CONTEXT_TRACKING_USER) && !defined(CONFIG_HAVE_CONTEXT_TRACKING_USER_OFFSTACK) |
722a9f92 | 6664 | asmlinkage __visible void __sched schedule_user(void) |
20ab65e3 FW |
6665 | { |
6666 | /* | |
6667 | * If we come here after a random call to set_need_resched(), | |
6668 | * or we have been woken up remotely but the IPI has not yet arrived, | |
6669 | * we haven't yet exited the RCU idle mode. Do it here manually until | |
6670 | * we find a better solution. | |
7cc78f8f AL |
6671 | * |
6672 | * NB: There are buggy callers of this function. Ideally we | |
c467ea76 | 6673 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
7cc78f8f | 6674 | * too frequently to make sense yet. |
20ab65e3 | 6675 | */ |
7cc78f8f | 6676 | enum ctx_state prev_state = exception_enter(); |
20ab65e3 | 6677 | schedule(); |
7cc78f8f | 6678 | exception_exit(prev_state); |
20ab65e3 FW |
6679 | } |
6680 | #endif | |
6681 | ||
c5491ea7 TG |
6682 | /** |
6683 | * schedule_preempt_disabled - called with preemption disabled | |
6684 | * | |
6685 | * Returns with preemption disabled. Note: preempt_count must be 1 | |
6686 | */ | |
6687 | void __sched schedule_preempt_disabled(void) | |
6688 | { | |
ba74c144 | 6689 | sched_preempt_enable_no_resched(); |
c5491ea7 TG |
6690 | schedule(); |
6691 | preempt_disable(); | |
6692 | } | |
6693 | ||
6991436c TG |
6694 | #ifdef CONFIG_PREEMPT_RT |
6695 | void __sched notrace schedule_rtlock(void) | |
6696 | { | |
6697 | do { | |
6698 | preempt_disable(); | |
6699 | __schedule(SM_RTLOCK_WAIT); | |
6700 | sched_preempt_enable_no_resched(); | |
6701 | } while (need_resched()); | |
6702 | } | |
6703 | NOKPROBE_SYMBOL(schedule_rtlock); | |
6704 | #endif | |
6705 | ||
06b1f808 | 6706 | static void __sched notrace preempt_schedule_common(void) |
a18b5d01 FW |
6707 | { |
6708 | do { | |
47252cfb SR |
6709 | /* |
6710 | * Because the function tracer can trace preempt_count_sub() | |
6711 | * and it also uses preempt_enable/disable_notrace(), if | |
6712 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6713 | * by the function tracer will call this function again and | |
6714 | * cause infinite recursion. | |
6715 | * | |
6716 | * Preemption must be disabled here before the function | |
6717 | * tracer can trace. Break up preempt_disable() into two | |
6718 | * calls. One to disable preemption without fear of being | |
6719 | * traced. The other to still record the preemption latency, | |
6720 | * which can also be traced by the function tracer. | |
6721 | */ | |
499d7955 | 6722 | preempt_disable_notrace(); |
47252cfb | 6723 | preempt_latency_start(1); |
b4bfa3fc | 6724 | __schedule(SM_PREEMPT); |
47252cfb | 6725 | preempt_latency_stop(1); |
499d7955 | 6726 | preempt_enable_no_resched_notrace(); |
a18b5d01 FW |
6727 | |
6728 | /* | |
6729 | * Check again in case we missed a preemption opportunity | |
6730 | * between schedule and now. | |
6731 | */ | |
a18b5d01 FW |
6732 | } while (need_resched()); |
6733 | } | |
6734 | ||
c1a280b6 | 6735 | #ifdef CONFIG_PREEMPTION |
1da177e4 | 6736 | /* |
a49b4f40 VS |
6737 | * This is the entry point to schedule() from in-kernel preemption |
6738 | * off of preempt_enable. | |
1da177e4 | 6739 | */ |
722a9f92 | 6740 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
1da177e4 | 6741 | { |
1da177e4 LT |
6742 | /* |
6743 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 6744 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 6745 | */ |
fbb00b56 | 6746 | if (likely(!preemptible())) |
1da177e4 | 6747 | return; |
a18b5d01 | 6748 | preempt_schedule_common(); |
1da177e4 | 6749 | } |
376e2424 | 6750 | NOKPROBE_SYMBOL(preempt_schedule); |
1da177e4 | 6751 | EXPORT_SYMBOL(preempt_schedule); |
009f60e2 | 6752 | |
2c9a98d3 | 6753 | #ifdef CONFIG_PREEMPT_DYNAMIC |
99cf983c | 6754 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
6755 | #ifndef preempt_schedule_dynamic_enabled |
6756 | #define preempt_schedule_dynamic_enabled preempt_schedule | |
6757 | #define preempt_schedule_dynamic_disabled NULL | |
6758 | #endif | |
6759 | DEFINE_STATIC_CALL(preempt_schedule, preempt_schedule_dynamic_enabled); | |
ef72661e | 6760 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule); |
99cf983c MR |
6761 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
6762 | static DEFINE_STATIC_KEY_TRUE(sk_dynamic_preempt_schedule); | |
6763 | void __sched notrace dynamic_preempt_schedule(void) | |
6764 | { | |
6765 | if (!static_branch_unlikely(&sk_dynamic_preempt_schedule)) | |
6766 | return; | |
6767 | preempt_schedule(); | |
6768 | } | |
6769 | NOKPROBE_SYMBOL(dynamic_preempt_schedule); | |
6770 | EXPORT_SYMBOL(dynamic_preempt_schedule); | |
6771 | #endif | |
2c9a98d3 | 6772 | #endif |
2c9a98d3 | 6773 | |
009f60e2 | 6774 | /** |
4eaca0a8 | 6775 | * preempt_schedule_notrace - preempt_schedule called by tracing |
009f60e2 ON |
6776 | * |
6777 | * The tracing infrastructure uses preempt_enable_notrace to prevent | |
6778 | * recursion and tracing preempt enabling caused by the tracing | |
6779 | * infrastructure itself. But as tracing can happen in areas coming | |
6780 | * from userspace or just about to enter userspace, a preempt enable | |
6781 | * can occur before user_exit() is called. This will cause the scheduler | |
6782 | * to be called when the system is still in usermode. | |
6783 | * | |
6784 | * To prevent this, the preempt_enable_notrace will use this function | |
6785 | * instead of preempt_schedule() to exit user context if needed before | |
6786 | * calling the scheduler. | |
6787 | */ | |
4eaca0a8 | 6788 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
009f60e2 ON |
6789 | { |
6790 | enum ctx_state prev_ctx; | |
6791 | ||
6792 | if (likely(!preemptible())) | |
6793 | return; | |
6794 | ||
6795 | do { | |
47252cfb SR |
6796 | /* |
6797 | * Because the function tracer can trace preempt_count_sub() | |
6798 | * and it also uses preempt_enable/disable_notrace(), if | |
6799 | * NEED_RESCHED is set, the preempt_enable_notrace() called | |
6800 | * by the function tracer will call this function again and | |
6801 | * cause infinite recursion. | |
6802 | * | |
6803 | * Preemption must be disabled here before the function | |
6804 | * tracer can trace. Break up preempt_disable() into two | |
6805 | * calls. One to disable preemption without fear of being | |
6806 | * traced. The other to still record the preemption latency, | |
6807 | * which can also be traced by the function tracer. | |
6808 | */ | |
3d8f74dd | 6809 | preempt_disable_notrace(); |
47252cfb | 6810 | preempt_latency_start(1); |
009f60e2 ON |
6811 | /* |
6812 | * Needs preempt disabled in case user_exit() is traced | |
6813 | * and the tracer calls preempt_enable_notrace() causing | |
6814 | * an infinite recursion. | |
6815 | */ | |
6816 | prev_ctx = exception_enter(); | |
b4bfa3fc | 6817 | __schedule(SM_PREEMPT); |
009f60e2 ON |
6818 | exception_exit(prev_ctx); |
6819 | ||
47252cfb | 6820 | preempt_latency_stop(1); |
3d8f74dd | 6821 | preempt_enable_no_resched_notrace(); |
009f60e2 ON |
6822 | } while (need_resched()); |
6823 | } | |
4eaca0a8 | 6824 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
009f60e2 | 6825 | |
2c9a98d3 | 6826 | #ifdef CONFIG_PREEMPT_DYNAMIC |
99cf983c | 6827 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
6828 | #ifndef preempt_schedule_notrace_dynamic_enabled |
6829 | #define preempt_schedule_notrace_dynamic_enabled preempt_schedule_notrace | |
6830 | #define preempt_schedule_notrace_dynamic_disabled NULL | |
2c9a98d3 | 6831 | #endif |
8a69fe0b | 6832 | DEFINE_STATIC_CALL(preempt_schedule_notrace, preempt_schedule_notrace_dynamic_enabled); |
ef72661e | 6833 | EXPORT_STATIC_CALL_TRAMP(preempt_schedule_notrace); |
99cf983c MR |
6834 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
6835 | static DEFINE_STATIC_KEY_TRUE(sk_dynamic_preempt_schedule_notrace); | |
6836 | void __sched notrace dynamic_preempt_schedule_notrace(void) | |
c597bfdd | 6837 | { |
99cf983c MR |
6838 | if (!static_branch_unlikely(&sk_dynamic_preempt_schedule_notrace)) |
6839 | return; | |
6840 | preempt_schedule_notrace(); | |
c597bfdd | 6841 | } |
99cf983c MR |
6842 | NOKPROBE_SYMBOL(dynamic_preempt_schedule_notrace); |
6843 | EXPORT_SYMBOL(dynamic_preempt_schedule_notrace); | |
6844 | #endif | |
2c9a98d3 | 6845 | #endif |
c597bfdd | 6846 | |
c1a280b6 | 6847 | #endif /* CONFIG_PREEMPTION */ |
826bfeb3 | 6848 | |
1da177e4 | 6849 | /* |
a49b4f40 | 6850 | * This is the entry point to schedule() from kernel preemption |
1da177e4 LT |
6851 | * off of irq context. |
6852 | * Note, that this is called and return with irqs disabled. This will | |
6853 | * protect us against recursive calling from irq. | |
6854 | */ | |
722a9f92 | 6855 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
1da177e4 | 6856 | { |
b22366cd | 6857 | enum ctx_state prev_state; |
6478d880 | 6858 | |
2ed6e34f | 6859 | /* Catch callers which need to be fixed */ |
f27dde8d | 6860 | BUG_ON(preempt_count() || !irqs_disabled()); |
1da177e4 | 6861 | |
b22366cd FW |
6862 | prev_state = exception_enter(); |
6863 | ||
3a5c359a | 6864 | do { |
3d8f74dd | 6865 | preempt_disable(); |
3a5c359a | 6866 | local_irq_enable(); |
b4bfa3fc | 6867 | __schedule(SM_PREEMPT); |
3a5c359a | 6868 | local_irq_disable(); |
3d8f74dd | 6869 | sched_preempt_enable_no_resched(); |
5ed0cec0 | 6870 | } while (need_resched()); |
b22366cd FW |
6871 | |
6872 | exception_exit(prev_state); | |
1da177e4 LT |
6873 | } |
6874 | ||
ac6424b9 | 6875 | int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 6876 | void *key) |
1da177e4 | 6877 | { |
062d3f95 | 6878 | WARN_ON_ONCE(IS_ENABLED(CONFIG_SCHED_DEBUG) && wake_flags & ~WF_SYNC); |
63859d4f | 6879 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 6880 | } |
1da177e4 LT |
6881 | EXPORT_SYMBOL(default_wake_function); |
6882 | ||
f558c2b8 PZ |
6883 | static void __setscheduler_prio(struct task_struct *p, int prio) |
6884 | { | |
6885 | if (dl_prio(prio)) | |
6886 | p->sched_class = &dl_sched_class; | |
6887 | else if (rt_prio(prio)) | |
6888 | p->sched_class = &rt_sched_class; | |
6889 | else | |
6890 | p->sched_class = &fair_sched_class; | |
6891 | ||
6892 | p->prio = prio; | |
6893 | } | |
6894 | ||
b29739f9 IM |
6895 | #ifdef CONFIG_RT_MUTEXES |
6896 | ||
acd58620 PZ |
6897 | static inline int __rt_effective_prio(struct task_struct *pi_task, int prio) |
6898 | { | |
6899 | if (pi_task) | |
6900 | prio = min(prio, pi_task->prio); | |
6901 | ||
6902 | return prio; | |
6903 | } | |
6904 | ||
6905 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
6906 | { | |
6907 | struct task_struct *pi_task = rt_mutex_get_top_task(p); | |
6908 | ||
6909 | return __rt_effective_prio(pi_task, prio); | |
6910 | } | |
6911 | ||
b29739f9 IM |
6912 | /* |
6913 | * rt_mutex_setprio - set the current priority of a task | |
acd58620 PZ |
6914 | * @p: task to boost |
6915 | * @pi_task: donor task | |
b29739f9 IM |
6916 | * |
6917 | * This function changes the 'effective' priority of a task. It does | |
6918 | * not touch ->normal_prio like __setscheduler(). | |
6919 | * | |
c365c292 TG |
6920 | * Used by the rt_mutex code to implement priority inheritance |
6921 | * logic. Call site only calls if the priority of the task changed. | |
b29739f9 | 6922 | */ |
acd58620 | 6923 | void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task) |
b29739f9 | 6924 | { |
acd58620 | 6925 | int prio, oldprio, queued, running, queue_flag = |
7a57f32a | 6926 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
83ab0aa0 | 6927 | const struct sched_class *prev_class; |
eb580751 PZ |
6928 | struct rq_flags rf; |
6929 | struct rq *rq; | |
b29739f9 | 6930 | |
acd58620 PZ |
6931 | /* XXX used to be waiter->prio, not waiter->task->prio */ |
6932 | prio = __rt_effective_prio(pi_task, p->normal_prio); | |
6933 | ||
6934 | /* | |
6935 | * If nothing changed; bail early. | |
6936 | */ | |
6937 | if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio)) | |
6938 | return; | |
b29739f9 | 6939 | |
eb580751 | 6940 | rq = __task_rq_lock(p, &rf); |
80f5c1b8 | 6941 | update_rq_clock(rq); |
acd58620 PZ |
6942 | /* |
6943 | * Set under pi_lock && rq->lock, such that the value can be used under | |
6944 | * either lock. | |
6945 | * | |
6946 | * Note that there is loads of tricky to make this pointer cache work | |
6947 | * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to | |
6948 | * ensure a task is de-boosted (pi_task is set to NULL) before the | |
6949 | * task is allowed to run again (and can exit). This ensures the pointer | |
b19a888c | 6950 | * points to a blocked task -- which guarantees the task is present. |
acd58620 PZ |
6951 | */ |
6952 | p->pi_top_task = pi_task; | |
6953 | ||
6954 | /* | |
6955 | * For FIFO/RR we only need to set prio, if that matches we're done. | |
6956 | */ | |
6957 | if (prio == p->prio && !dl_prio(prio)) | |
6958 | goto out_unlock; | |
b29739f9 | 6959 | |
1c4dd99b TG |
6960 | /* |
6961 | * Idle task boosting is a nono in general. There is one | |
6962 | * exception, when PREEMPT_RT and NOHZ is active: | |
6963 | * | |
6964 | * The idle task calls get_next_timer_interrupt() and holds | |
6965 | * the timer wheel base->lock on the CPU and another CPU wants | |
6966 | * to access the timer (probably to cancel it). We can safely | |
6967 | * ignore the boosting request, as the idle CPU runs this code | |
6968 | * with interrupts disabled and will complete the lock | |
6969 | * protected section without being interrupted. So there is no | |
6970 | * real need to boost. | |
6971 | */ | |
6972 | if (unlikely(p == rq->idle)) { | |
6973 | WARN_ON(p != rq->curr); | |
6974 | WARN_ON(p->pi_blocked_on); | |
6975 | goto out_unlock; | |
6976 | } | |
6977 | ||
b91473ff | 6978 | trace_sched_pi_setprio(p, pi_task); |
d5f9f942 | 6979 | oldprio = p->prio; |
ff77e468 PZ |
6980 | |
6981 | if (oldprio == prio) | |
6982 | queue_flag &= ~DEQUEUE_MOVE; | |
6983 | ||
83ab0aa0 | 6984 | prev_class = p->sched_class; |
da0c1e65 | 6985 | queued = task_on_rq_queued(p); |
051a1d1a | 6986 | running = task_current(rq, p); |
da0c1e65 | 6987 | if (queued) |
ff77e468 | 6988 | dequeue_task(rq, p, queue_flag); |
0e1f3483 | 6989 | if (running) |
f3cd1c4e | 6990 | put_prev_task(rq, p); |
dd41f596 | 6991 | |
2d3d891d DF |
6992 | /* |
6993 | * Boosting condition are: | |
6994 | * 1. -rt task is running and holds mutex A | |
6995 | * --> -dl task blocks on mutex A | |
6996 | * | |
6997 | * 2. -dl task is running and holds mutex A | |
6998 | * --> -dl task blocks on mutex A and could preempt the | |
6999 | * running task | |
7000 | */ | |
7001 | if (dl_prio(prio)) { | |
466af29b | 7002 | if (!dl_prio(p->normal_prio) || |
740797ce JL |
7003 | (pi_task && dl_prio(pi_task->prio) && |
7004 | dl_entity_preempt(&pi_task->dl, &p->dl))) { | |
2279f540 | 7005 | p->dl.pi_se = pi_task->dl.pi_se; |
ff77e468 | 7006 | queue_flag |= ENQUEUE_REPLENISH; |
2279f540 JL |
7007 | } else { |
7008 | p->dl.pi_se = &p->dl; | |
7009 | } | |
2d3d891d DF |
7010 | } else if (rt_prio(prio)) { |
7011 | if (dl_prio(oldprio)) | |
2279f540 | 7012 | p->dl.pi_se = &p->dl; |
2d3d891d | 7013 | if (oldprio < prio) |
ff77e468 | 7014 | queue_flag |= ENQUEUE_HEAD; |
2d3d891d DF |
7015 | } else { |
7016 | if (dl_prio(oldprio)) | |
2279f540 | 7017 | p->dl.pi_se = &p->dl; |
746db944 BS |
7018 | if (rt_prio(oldprio)) |
7019 | p->rt.timeout = 0; | |
2d3d891d | 7020 | } |
dd41f596 | 7021 | |
f558c2b8 | 7022 | __setscheduler_prio(p, prio); |
b29739f9 | 7023 | |
da0c1e65 | 7024 | if (queued) |
ff77e468 | 7025 | enqueue_task(rq, p, queue_flag); |
a399d233 | 7026 | if (running) |
03b7fad1 | 7027 | set_next_task(rq, p); |
cb469845 | 7028 | |
da7a735e | 7029 | check_class_changed(rq, p, prev_class, oldprio); |
1c4dd99b | 7030 | out_unlock: |
d1ccc66d IM |
7031 | /* Avoid rq from going away on us: */ |
7032 | preempt_disable(); | |
4c9a4bc8 | 7033 | |
565790d2 PZ |
7034 | rq_unpin_lock(rq, &rf); |
7035 | __balance_callbacks(rq); | |
5cb9eaa3 | 7036 | raw_spin_rq_unlock(rq); |
565790d2 | 7037 | |
4c9a4bc8 | 7038 | preempt_enable(); |
b29739f9 | 7039 | } |
acd58620 PZ |
7040 | #else |
7041 | static inline int rt_effective_prio(struct task_struct *p, int prio) | |
7042 | { | |
7043 | return prio; | |
7044 | } | |
b29739f9 | 7045 | #endif |
d50dde5a | 7046 | |
36c8b586 | 7047 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 7048 | { |
49bd21ef | 7049 | bool queued, running; |
53a23364 | 7050 | int old_prio; |
eb580751 | 7051 | struct rq_flags rf; |
70b97a7f | 7052 | struct rq *rq; |
1da177e4 | 7053 | |
75e45d51 | 7054 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
1da177e4 LT |
7055 | return; |
7056 | /* | |
7057 | * We have to be careful, if called from sys_setpriority(), | |
7058 | * the task might be in the middle of scheduling on another CPU. | |
7059 | */ | |
eb580751 | 7060 | rq = task_rq_lock(p, &rf); |
2fb8d367 PZ |
7061 | update_rq_clock(rq); |
7062 | ||
1da177e4 LT |
7063 | /* |
7064 | * The RT priorities are set via sched_setscheduler(), but we still | |
7065 | * allow the 'normal' nice value to be set - but as expected | |
b19a888c | 7066 | * it won't have any effect on scheduling until the task is |
aab03e05 | 7067 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
1da177e4 | 7068 | */ |
aab03e05 | 7069 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
1da177e4 LT |
7070 | p->static_prio = NICE_TO_PRIO(nice); |
7071 | goto out_unlock; | |
7072 | } | |
da0c1e65 | 7073 | queued = task_on_rq_queued(p); |
49bd21ef | 7074 | running = task_current(rq, p); |
da0c1e65 | 7075 | if (queued) |
7a57f32a | 7076 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); |
49bd21ef PZ |
7077 | if (running) |
7078 | put_prev_task(rq, p); | |
1da177e4 | 7079 | |
1da177e4 | 7080 | p->static_prio = NICE_TO_PRIO(nice); |
b1e82065 | 7081 | set_load_weight(p, true); |
b29739f9 IM |
7082 | old_prio = p->prio; |
7083 | p->prio = effective_prio(p); | |
1da177e4 | 7084 | |
5443a0be | 7085 | if (queued) |
7134b3e9 | 7086 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
49bd21ef | 7087 | if (running) |
03b7fad1 | 7088 | set_next_task(rq, p); |
5443a0be FW |
7089 | |
7090 | /* | |
7091 | * If the task increased its priority or is running and | |
7092 | * lowered its priority, then reschedule its CPU: | |
7093 | */ | |
7094 | p->sched_class->prio_changed(rq, p, old_prio); | |
7095 | ||
1da177e4 | 7096 | out_unlock: |
eb580751 | 7097 | task_rq_unlock(rq, p, &rf); |
1da177e4 | 7098 | } |
1da177e4 LT |
7099 | EXPORT_SYMBOL(set_user_nice); |
7100 | ||
e43379f1 | 7101 | /* |
700a7833 CG |
7102 | * is_nice_reduction - check if nice value is an actual reduction |
7103 | * | |
7104 | * Similar to can_nice() but does not perform a capability check. | |
7105 | * | |
e43379f1 MM |
7106 | * @p: task |
7107 | * @nice: nice value | |
7108 | */ | |
700a7833 | 7109 | static bool is_nice_reduction(const struct task_struct *p, const int nice) |
e43379f1 | 7110 | { |
d1ccc66d | 7111 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ |
7aa2c016 | 7112 | int nice_rlim = nice_to_rlimit(nice); |
48f24c4d | 7113 | |
700a7833 CG |
7114 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE)); |
7115 | } | |
7116 | ||
7117 | /* | |
7118 | * can_nice - check if a task can reduce its nice value | |
7119 | * @p: task | |
7120 | * @nice: nice value | |
7121 | */ | |
7122 | int can_nice(const struct task_struct *p, const int nice) | |
7123 | { | |
7124 | return is_nice_reduction(p, nice) || capable(CAP_SYS_NICE); | |
e43379f1 MM |
7125 | } |
7126 | ||
1da177e4 LT |
7127 | #ifdef __ARCH_WANT_SYS_NICE |
7128 | ||
7129 | /* | |
7130 | * sys_nice - change the priority of the current process. | |
7131 | * @increment: priority increment | |
7132 | * | |
7133 | * sys_setpriority is a more generic, but much slower function that | |
7134 | * does similar things. | |
7135 | */ | |
5add95d4 | 7136 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 7137 | { |
48f24c4d | 7138 | long nice, retval; |
1da177e4 LT |
7139 | |
7140 | /* | |
7141 | * Setpriority might change our priority at the same moment. | |
7142 | * We don't have to worry. Conceptually one call occurs first | |
7143 | * and we have a single winner. | |
7144 | */ | |
a9467fa3 | 7145 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
d0ea0268 | 7146 | nice = task_nice(current) + increment; |
1da177e4 | 7147 | |
a9467fa3 | 7148 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
e43379f1 MM |
7149 | if (increment < 0 && !can_nice(current, nice)) |
7150 | return -EPERM; | |
7151 | ||
1da177e4 LT |
7152 | retval = security_task_setnice(current, nice); |
7153 | if (retval) | |
7154 | return retval; | |
7155 | ||
7156 | set_user_nice(current, nice); | |
7157 | return 0; | |
7158 | } | |
7159 | ||
7160 | #endif | |
7161 | ||
7162 | /** | |
7163 | * task_prio - return the priority value of a given task. | |
7164 | * @p: the task in question. | |
7165 | * | |
e69f6186 | 7166 | * Return: The priority value as seen by users in /proc. |
c541bb78 DE |
7167 | * |
7168 | * sched policy return value kernel prio user prio/nice | |
7169 | * | |
7170 | * normal, batch, idle [0 ... 39] [100 ... 139] 0/[-20 ... 19] | |
7171 | * fifo, rr [-2 ... -100] [98 ... 0] [1 ... 99] | |
7172 | * deadline -101 -1 0 | |
1da177e4 | 7173 | */ |
36c8b586 | 7174 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
7175 | { |
7176 | return p->prio - MAX_RT_PRIO; | |
7177 | } | |
7178 | ||
1da177e4 | 7179 | /** |
d1ccc66d | 7180 | * idle_cpu - is a given CPU idle currently? |
1da177e4 | 7181 | * @cpu: the processor in question. |
e69f6186 YB |
7182 | * |
7183 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
1da177e4 LT |
7184 | */ |
7185 | int idle_cpu(int cpu) | |
7186 | { | |
908a3283 TG |
7187 | struct rq *rq = cpu_rq(cpu); |
7188 | ||
7189 | if (rq->curr != rq->idle) | |
7190 | return 0; | |
7191 | ||
7192 | if (rq->nr_running) | |
7193 | return 0; | |
7194 | ||
7195 | #ifdef CONFIG_SMP | |
126c2092 | 7196 | if (rq->ttwu_pending) |
908a3283 TG |
7197 | return 0; |
7198 | #endif | |
7199 | ||
7200 | return 1; | |
1da177e4 LT |
7201 | } |
7202 | ||
943d355d RJ |
7203 | /** |
7204 | * available_idle_cpu - is a given CPU idle for enqueuing work. | |
7205 | * @cpu: the CPU in question. | |
7206 | * | |
7207 | * Return: 1 if the CPU is currently idle. 0 otherwise. | |
7208 | */ | |
7209 | int available_idle_cpu(int cpu) | |
7210 | { | |
7211 | if (!idle_cpu(cpu)) | |
7212 | return 0; | |
7213 | ||
247f2f6f RJ |
7214 | if (vcpu_is_preempted(cpu)) |
7215 | return 0; | |
7216 | ||
908a3283 | 7217 | return 1; |
1da177e4 LT |
7218 | } |
7219 | ||
1da177e4 | 7220 | /** |
d1ccc66d | 7221 | * idle_task - return the idle task for a given CPU. |
1da177e4 | 7222 | * @cpu: the processor in question. |
e69f6186 | 7223 | * |
d1ccc66d | 7224 | * Return: The idle task for the CPU @cpu. |
1da177e4 | 7225 | */ |
36c8b586 | 7226 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
7227 | { |
7228 | return cpu_rq(cpu)->idle; | |
7229 | } | |
7230 | ||
7d6a905f VK |
7231 | #ifdef CONFIG_SMP |
7232 | /* | |
7233 | * This function computes an effective utilization for the given CPU, to be | |
7234 | * used for frequency selection given the linear relation: f = u * f_max. | |
7235 | * | |
7236 | * The scheduler tracks the following metrics: | |
7237 | * | |
7238 | * cpu_util_{cfs,rt,dl,irq}() | |
7239 | * cpu_bw_dl() | |
7240 | * | |
7241 | * Where the cfs,rt and dl util numbers are tracked with the same metric and | |
7242 | * synchronized windows and are thus directly comparable. | |
7243 | * | |
7244 | * The cfs,rt,dl utilization are the running times measured with rq->clock_task | |
7245 | * which excludes things like IRQ and steal-time. These latter are then accrued | |
7246 | * in the irq utilization. | |
7247 | * | |
7248 | * The DL bandwidth number otoh is not a measured metric but a value computed | |
7249 | * based on the task model parameters and gives the minimal utilization | |
7250 | * required to meet deadlines. | |
7251 | */ | |
a5418be9 | 7252 | unsigned long effective_cpu_util(int cpu, unsigned long util_cfs, |
bb447999 | 7253 | enum cpu_util_type type, |
7d6a905f VK |
7254 | struct task_struct *p) |
7255 | { | |
bb447999 | 7256 | unsigned long dl_util, util, irq, max; |
7d6a905f VK |
7257 | struct rq *rq = cpu_rq(cpu); |
7258 | ||
bb447999 DE |
7259 | max = arch_scale_cpu_capacity(cpu); |
7260 | ||
7d6a905f VK |
7261 | if (!uclamp_is_used() && |
7262 | type == FREQUENCY_UTIL && rt_rq_is_runnable(&rq->rt)) { | |
7263 | return max; | |
7264 | } | |
7265 | ||
7266 | /* | |
7267 | * Early check to see if IRQ/steal time saturates the CPU, can be | |
7268 | * because of inaccuracies in how we track these -- see | |
7269 | * update_irq_load_avg(). | |
7270 | */ | |
7271 | irq = cpu_util_irq(rq); | |
7272 | if (unlikely(irq >= max)) | |
7273 | return max; | |
7274 | ||
7275 | /* | |
7276 | * Because the time spend on RT/DL tasks is visible as 'lost' time to | |
7277 | * CFS tasks and we use the same metric to track the effective | |
7278 | * utilization (PELT windows are synchronized) we can directly add them | |
7279 | * to obtain the CPU's actual utilization. | |
7280 | * | |
7281 | * CFS and RT utilization can be boosted or capped, depending on | |
7282 | * utilization clamp constraints requested by currently RUNNABLE | |
7283 | * tasks. | |
7284 | * When there are no CFS RUNNABLE tasks, clamps are released and | |
7285 | * frequency will be gracefully reduced with the utilization decay. | |
7286 | */ | |
7287 | util = util_cfs + cpu_util_rt(rq); | |
7288 | if (type == FREQUENCY_UTIL) | |
7289 | util = uclamp_rq_util_with(rq, util, p); | |
7290 | ||
7291 | dl_util = cpu_util_dl(rq); | |
7292 | ||
7293 | /* | |
7294 | * For frequency selection we do not make cpu_util_dl() a permanent part | |
7295 | * of this sum because we want to use cpu_bw_dl() later on, but we need | |
7296 | * to check if the CFS+RT+DL sum is saturated (ie. no idle time) such | |
7297 | * that we select f_max when there is no idle time. | |
7298 | * | |
7299 | * NOTE: numerical errors or stop class might cause us to not quite hit | |
7300 | * saturation when we should -- something for later. | |
7301 | */ | |
7302 | if (util + dl_util >= max) | |
7303 | return max; | |
7304 | ||
7305 | /* | |
7306 | * OTOH, for energy computation we need the estimated running time, so | |
7307 | * include util_dl and ignore dl_bw. | |
7308 | */ | |
7309 | if (type == ENERGY_UTIL) | |
7310 | util += dl_util; | |
7311 | ||
7312 | /* | |
7313 | * There is still idle time; further improve the number by using the | |
7314 | * irq metric. Because IRQ/steal time is hidden from the task clock we | |
7315 | * need to scale the task numbers: | |
7316 | * | |
7317 | * max - irq | |
7318 | * U' = irq + --------- * U | |
7319 | * max | |
7320 | */ | |
7321 | util = scale_irq_capacity(util, irq, max); | |
7322 | util += irq; | |
7323 | ||
7324 | /* | |
7325 | * Bandwidth required by DEADLINE must always be granted while, for | |
7326 | * FAIR and RT, we use blocked utilization of IDLE CPUs as a mechanism | |
7327 | * to gracefully reduce the frequency when no tasks show up for longer | |
7328 | * periods of time. | |
7329 | * | |
7330 | * Ideally we would like to set bw_dl as min/guaranteed freq and util + | |
7331 | * bw_dl as requested freq. However, cpufreq is not yet ready for such | |
7332 | * an interface. So, we only do the latter for now. | |
7333 | */ | |
7334 | if (type == FREQUENCY_UTIL) | |
7335 | util += cpu_bw_dl(rq); | |
7336 | ||
7337 | return min(max, util); | |
7338 | } | |
a5418be9 | 7339 | |
bb447999 | 7340 | unsigned long sched_cpu_util(int cpu) |
a5418be9 | 7341 | { |
bb447999 | 7342 | return effective_cpu_util(cpu, cpu_util_cfs(cpu), ENERGY_UTIL, NULL); |
a5418be9 | 7343 | } |
7d6a905f VK |
7344 | #endif /* CONFIG_SMP */ |
7345 | ||
1da177e4 LT |
7346 | /** |
7347 | * find_process_by_pid - find a process with a matching PID value. | |
7348 | * @pid: the pid in question. | |
e69f6186 YB |
7349 | * |
7350 | * The task of @pid, if found. %NULL otherwise. | |
1da177e4 | 7351 | */ |
a9957449 | 7352 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 7353 | { |
228ebcbe | 7354 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
7355 | } |
7356 | ||
c13db6b1 SR |
7357 | /* |
7358 | * sched_setparam() passes in -1 for its policy, to let the functions | |
7359 | * it calls know not to change it. | |
7360 | */ | |
7361 | #define SETPARAM_POLICY -1 | |
7362 | ||
c365c292 TG |
7363 | static void __setscheduler_params(struct task_struct *p, |
7364 | const struct sched_attr *attr) | |
1da177e4 | 7365 | { |
d50dde5a DF |
7366 | int policy = attr->sched_policy; |
7367 | ||
c13db6b1 | 7368 | if (policy == SETPARAM_POLICY) |
39fd8fd2 PZ |
7369 | policy = p->policy; |
7370 | ||
1da177e4 | 7371 | p->policy = policy; |
d50dde5a | 7372 | |
aab03e05 DF |
7373 | if (dl_policy(policy)) |
7374 | __setparam_dl(p, attr); | |
39fd8fd2 | 7375 | else if (fair_policy(policy)) |
d50dde5a DF |
7376 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
7377 | ||
39fd8fd2 PZ |
7378 | /* |
7379 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | |
7380 | * !rt_policy. Always setting this ensures that things like | |
7381 | * getparam()/getattr() don't report silly values for !rt tasks. | |
7382 | */ | |
7383 | p->rt_priority = attr->sched_priority; | |
383afd09 | 7384 | p->normal_prio = normal_prio(p); |
b1e82065 | 7385 | set_load_weight(p, true); |
c365c292 | 7386 | } |
39fd8fd2 | 7387 | |
c69e8d9c | 7388 | /* |
d1ccc66d | 7389 | * Check the target process has a UID that matches the current process's: |
c69e8d9c DH |
7390 | */ |
7391 | static bool check_same_owner(struct task_struct *p) | |
7392 | { | |
7393 | const struct cred *cred = current_cred(), *pcred; | |
7394 | bool match; | |
7395 | ||
7396 | rcu_read_lock(); | |
7397 | pcred = __task_cred(p); | |
9c806aa0 EB |
7398 | match = (uid_eq(cred->euid, pcred->euid) || |
7399 | uid_eq(cred->euid, pcred->uid)); | |
c69e8d9c DH |
7400 | rcu_read_unlock(); |
7401 | return match; | |
7402 | } | |
7403 | ||
700a7833 CG |
7404 | /* |
7405 | * Allow unprivileged RT tasks to decrease priority. | |
7406 | * Only issue a capable test if needed and only once to avoid an audit | |
7407 | * event on permitted non-privileged operations: | |
7408 | */ | |
7409 | static int user_check_sched_setscheduler(struct task_struct *p, | |
7410 | const struct sched_attr *attr, | |
7411 | int policy, int reset_on_fork) | |
7412 | { | |
7413 | if (fair_policy(policy)) { | |
7414 | if (attr->sched_nice < task_nice(p) && | |
7415 | !is_nice_reduction(p, attr->sched_nice)) | |
7416 | goto req_priv; | |
7417 | } | |
7418 | ||
7419 | if (rt_policy(policy)) { | |
7420 | unsigned long rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); | |
7421 | ||
7422 | /* Can't set/change the rt policy: */ | |
7423 | if (policy != p->policy && !rlim_rtprio) | |
7424 | goto req_priv; | |
7425 | ||
7426 | /* Can't increase priority: */ | |
7427 | if (attr->sched_priority > p->rt_priority && | |
7428 | attr->sched_priority > rlim_rtprio) | |
7429 | goto req_priv; | |
7430 | } | |
7431 | ||
7432 | /* | |
7433 | * Can't set/change SCHED_DEADLINE policy at all for now | |
7434 | * (safest behavior); in the future we would like to allow | |
7435 | * unprivileged DL tasks to increase their relative deadline | |
7436 | * or reduce their runtime (both ways reducing utilization) | |
7437 | */ | |
7438 | if (dl_policy(policy)) | |
7439 | goto req_priv; | |
7440 | ||
7441 | /* | |
7442 | * Treat SCHED_IDLE as nice 20. Only allow a switch to | |
7443 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
7444 | */ | |
7445 | if (task_has_idle_policy(p) && !idle_policy(policy)) { | |
7446 | if (!is_nice_reduction(p, task_nice(p))) | |
7447 | goto req_priv; | |
7448 | } | |
7449 | ||
7450 | /* Can't change other user's priorities: */ | |
7451 | if (!check_same_owner(p)) | |
7452 | goto req_priv; | |
7453 | ||
7454 | /* Normal users shall not reset the sched_reset_on_fork flag: */ | |
7455 | if (p->sched_reset_on_fork && !reset_on_fork) | |
7456 | goto req_priv; | |
7457 | ||
7458 | return 0; | |
7459 | ||
7460 | req_priv: | |
7461 | if (!capable(CAP_SYS_NICE)) | |
7462 | return -EPERM; | |
7463 | ||
7464 | return 0; | |
7465 | } | |
7466 | ||
d50dde5a DF |
7467 | static int __sched_setscheduler(struct task_struct *p, |
7468 | const struct sched_attr *attr, | |
dbc7f069 | 7469 | bool user, bool pi) |
1da177e4 | 7470 | { |
f558c2b8 PZ |
7471 | int oldpolicy = -1, policy = attr->sched_policy; |
7472 | int retval, oldprio, newprio, queued, running; | |
83ab0aa0 | 7473 | const struct sched_class *prev_class; |
8e5bad7d | 7474 | struct balance_callback *head; |
eb580751 | 7475 | struct rq_flags rf; |
ca94c442 | 7476 | int reset_on_fork; |
7a57f32a | 7477 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; |
eb580751 | 7478 | struct rq *rq; |
1da177e4 | 7479 | |
896bbb25 SRV |
7480 | /* The pi code expects interrupts enabled */ |
7481 | BUG_ON(pi && in_interrupt()); | |
1da177e4 | 7482 | recheck: |
d1ccc66d | 7483 | /* Double check policy once rq lock held: */ |
ca94c442 LP |
7484 | if (policy < 0) { |
7485 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 7486 | policy = oldpolicy = p->policy; |
ca94c442 | 7487 | } else { |
7479f3c9 | 7488 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
ca94c442 | 7489 | |
20f9cd2a | 7490 | if (!valid_policy(policy)) |
ca94c442 LP |
7491 | return -EINVAL; |
7492 | } | |
7493 | ||
794a56eb | 7494 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) |
7479f3c9 PZ |
7495 | return -EINVAL; |
7496 | ||
1da177e4 LT |
7497 | /* |
7498 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
ae18ad28 | 7499 | * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL, |
dd41f596 | 7500 | * SCHED_BATCH and SCHED_IDLE is 0. |
1da177e4 | 7501 | */ |
ae18ad28 | 7502 | if (attr->sched_priority > MAX_RT_PRIO-1) |
1da177e4 | 7503 | return -EINVAL; |
aab03e05 DF |
7504 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
7505 | (rt_policy(policy) != (attr->sched_priority != 0))) | |
1da177e4 LT |
7506 | return -EINVAL; |
7507 | ||
725aad24 | 7508 | if (user) { |
700a7833 CG |
7509 | retval = user_check_sched_setscheduler(p, attr, policy, reset_on_fork); |
7510 | if (retval) | |
7511 | return retval; | |
7512 | ||
794a56eb JL |
7513 | if (attr->sched_flags & SCHED_FLAG_SUGOV) |
7514 | return -EINVAL; | |
7515 | ||
b0ae1981 | 7516 | retval = security_task_setscheduler(p); |
725aad24 JF |
7517 | if (retval) |
7518 | return retval; | |
7519 | } | |
7520 | ||
a509a7cd PB |
7521 | /* Update task specific "requested" clamps */ |
7522 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | |
7523 | retval = uclamp_validate(p, attr); | |
7524 | if (retval) | |
7525 | return retval; | |
7526 | } | |
7527 | ||
710da3c8 JL |
7528 | if (pi) |
7529 | cpuset_read_lock(); | |
7530 | ||
b29739f9 | 7531 | /* |
d1ccc66d | 7532 | * Make sure no PI-waiters arrive (or leave) while we are |
b29739f9 | 7533 | * changing the priority of the task: |
0122ec5b | 7534 | * |
25985edc | 7535 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
7536 | * runqueue lock must be held. |
7537 | */ | |
eb580751 | 7538 | rq = task_rq_lock(p, &rf); |
80f5c1b8 | 7539 | update_rq_clock(rq); |
dc61b1d6 | 7540 | |
34f971f6 | 7541 | /* |
d1ccc66d | 7542 | * Changing the policy of the stop threads its a very bad idea: |
34f971f6 PZ |
7543 | */ |
7544 | if (p == rq->stop) { | |
4b211f2b MP |
7545 | retval = -EINVAL; |
7546 | goto unlock; | |
34f971f6 PZ |
7547 | } |
7548 | ||
a51e9198 | 7549 | /* |
d6b1e911 TG |
7550 | * If not changing anything there's no need to proceed further, |
7551 | * but store a possible modification of reset_on_fork. | |
a51e9198 | 7552 | */ |
d50dde5a | 7553 | if (unlikely(policy == p->policy)) { |
d0ea0268 | 7554 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
d50dde5a DF |
7555 | goto change; |
7556 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | |
7557 | goto change; | |
75381608 | 7558 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
aab03e05 | 7559 | goto change; |
a509a7cd PB |
7560 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) |
7561 | goto change; | |
d50dde5a | 7562 | |
d6b1e911 | 7563 | p->sched_reset_on_fork = reset_on_fork; |
4b211f2b MP |
7564 | retval = 0; |
7565 | goto unlock; | |
a51e9198 | 7566 | } |
d50dde5a | 7567 | change: |
a51e9198 | 7568 | |
dc61b1d6 | 7569 | if (user) { |
332ac17e | 7570 | #ifdef CONFIG_RT_GROUP_SCHED |
dc61b1d6 PZ |
7571 | /* |
7572 | * Do not allow realtime tasks into groups that have no runtime | |
7573 | * assigned. | |
7574 | */ | |
7575 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
7576 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
7577 | !task_group_is_autogroup(task_group(p))) { | |
4b211f2b MP |
7578 | retval = -EPERM; |
7579 | goto unlock; | |
dc61b1d6 | 7580 | } |
dc61b1d6 | 7581 | #endif |
332ac17e | 7582 | #ifdef CONFIG_SMP |
794a56eb JL |
7583 | if (dl_bandwidth_enabled() && dl_policy(policy) && |
7584 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | |
332ac17e | 7585 | cpumask_t *span = rq->rd->span; |
332ac17e DF |
7586 | |
7587 | /* | |
7588 | * Don't allow tasks with an affinity mask smaller than | |
7589 | * the entire root_domain to become SCHED_DEADLINE. We | |
7590 | * will also fail if there's no bandwidth available. | |
7591 | */ | |
3bd37062 | 7592 | if (!cpumask_subset(span, p->cpus_ptr) || |
e4099a5e | 7593 | rq->rd->dl_bw.bw == 0) { |
4b211f2b MP |
7594 | retval = -EPERM; |
7595 | goto unlock; | |
332ac17e DF |
7596 | } |
7597 | } | |
7598 | #endif | |
7599 | } | |
dc61b1d6 | 7600 | |
d1ccc66d | 7601 | /* Re-check policy now with rq lock held: */ |
1da177e4 LT |
7602 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
7603 | policy = oldpolicy = -1; | |
eb580751 | 7604 | task_rq_unlock(rq, p, &rf); |
710da3c8 JL |
7605 | if (pi) |
7606 | cpuset_read_unlock(); | |
1da177e4 LT |
7607 | goto recheck; |
7608 | } | |
332ac17e DF |
7609 | |
7610 | /* | |
7611 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | |
7612 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | |
7613 | * is available. | |
7614 | */ | |
06a76fe0 | 7615 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { |
4b211f2b MP |
7616 | retval = -EBUSY; |
7617 | goto unlock; | |
332ac17e DF |
7618 | } |
7619 | ||
c365c292 TG |
7620 | p->sched_reset_on_fork = reset_on_fork; |
7621 | oldprio = p->prio; | |
7622 | ||
f558c2b8 | 7623 | newprio = __normal_prio(policy, attr->sched_priority, attr->sched_nice); |
dbc7f069 PZ |
7624 | if (pi) { |
7625 | /* | |
7626 | * Take priority boosted tasks into account. If the new | |
7627 | * effective priority is unchanged, we just store the new | |
7628 | * normal parameters and do not touch the scheduler class and | |
7629 | * the runqueue. This will be done when the task deboost | |
7630 | * itself. | |
7631 | */ | |
f558c2b8 PZ |
7632 | newprio = rt_effective_prio(p, newprio); |
7633 | if (newprio == oldprio) | |
ff77e468 | 7634 | queue_flags &= ~DEQUEUE_MOVE; |
c365c292 TG |
7635 | } |
7636 | ||
da0c1e65 | 7637 | queued = task_on_rq_queued(p); |
051a1d1a | 7638 | running = task_current(rq, p); |
da0c1e65 | 7639 | if (queued) |
ff77e468 | 7640 | dequeue_task(rq, p, queue_flags); |
0e1f3483 | 7641 | if (running) |
f3cd1c4e | 7642 | put_prev_task(rq, p); |
f6b53205 | 7643 | |
83ab0aa0 | 7644 | prev_class = p->sched_class; |
a509a7cd | 7645 | |
f558c2b8 PZ |
7646 | if (!(attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)) { |
7647 | __setscheduler_params(p, attr); | |
7648 | __setscheduler_prio(p, newprio); | |
7649 | } | |
a509a7cd | 7650 | __setscheduler_uclamp(p, attr); |
f6b53205 | 7651 | |
da0c1e65 | 7652 | if (queued) { |
81a44c54 TG |
7653 | /* |
7654 | * We enqueue to tail when the priority of a task is | |
7655 | * increased (user space view). | |
7656 | */ | |
ff77e468 PZ |
7657 | if (oldprio < p->prio) |
7658 | queue_flags |= ENQUEUE_HEAD; | |
1de64443 | 7659 | |
ff77e468 | 7660 | enqueue_task(rq, p, queue_flags); |
81a44c54 | 7661 | } |
a399d233 | 7662 | if (running) |
03b7fad1 | 7663 | set_next_task(rq, p); |
cb469845 | 7664 | |
da7a735e | 7665 | check_class_changed(rq, p, prev_class, oldprio); |
d1ccc66d IM |
7666 | |
7667 | /* Avoid rq from going away on us: */ | |
7668 | preempt_disable(); | |
565790d2 | 7669 | head = splice_balance_callbacks(rq); |
eb580751 | 7670 | task_rq_unlock(rq, p, &rf); |
b29739f9 | 7671 | |
710da3c8 JL |
7672 | if (pi) { |
7673 | cpuset_read_unlock(); | |
dbc7f069 | 7674 | rt_mutex_adjust_pi(p); |
710da3c8 | 7675 | } |
95e02ca9 | 7676 | |
d1ccc66d | 7677 | /* Run balance callbacks after we've adjusted the PI chain: */ |
565790d2 | 7678 | balance_callbacks(rq, head); |
4c9a4bc8 | 7679 | preempt_enable(); |
95e02ca9 | 7680 | |
1da177e4 | 7681 | return 0; |
4b211f2b MP |
7682 | |
7683 | unlock: | |
7684 | task_rq_unlock(rq, p, &rf); | |
710da3c8 JL |
7685 | if (pi) |
7686 | cpuset_read_unlock(); | |
4b211f2b | 7687 | return retval; |
1da177e4 | 7688 | } |
961ccddd | 7689 | |
7479f3c9 PZ |
7690 | static int _sched_setscheduler(struct task_struct *p, int policy, |
7691 | const struct sched_param *param, bool check) | |
7692 | { | |
7693 | struct sched_attr attr = { | |
7694 | .sched_policy = policy, | |
7695 | .sched_priority = param->sched_priority, | |
7696 | .sched_nice = PRIO_TO_NICE(p->static_prio), | |
7697 | }; | |
7698 | ||
c13db6b1 SR |
7699 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
7700 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | |
7479f3c9 PZ |
7701 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
7702 | policy &= ~SCHED_RESET_ON_FORK; | |
7703 | attr.sched_policy = policy; | |
7704 | } | |
7705 | ||
dbc7f069 | 7706 | return __sched_setscheduler(p, &attr, check, true); |
7479f3c9 | 7707 | } |
961ccddd RR |
7708 | /** |
7709 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
7710 | * @p: the task in question. | |
7711 | * @policy: new policy. | |
7712 | * @param: structure containing the new RT priority. | |
7713 | * | |
7318d4cc PZ |
7714 | * Use sched_set_fifo(), read its comment. |
7715 | * | |
e69f6186 YB |
7716 | * Return: 0 on success. An error code otherwise. |
7717 | * | |
961ccddd RR |
7718 | * NOTE that the task may be already dead. |
7719 | */ | |
7720 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 7721 | const struct sched_param *param) |
961ccddd | 7722 | { |
7479f3c9 | 7723 | return _sched_setscheduler(p, policy, param, true); |
961ccddd | 7724 | } |
1da177e4 | 7725 | |
d50dde5a DF |
7726 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
7727 | { | |
dbc7f069 | 7728 | return __sched_setscheduler(p, attr, true, true); |
d50dde5a | 7729 | } |
d50dde5a | 7730 | |
794a56eb JL |
7731 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) |
7732 | { | |
7733 | return __sched_setscheduler(p, attr, false, true); | |
7734 | } | |
1eb5dde6 | 7735 | EXPORT_SYMBOL_GPL(sched_setattr_nocheck); |
794a56eb | 7736 | |
961ccddd RR |
7737 | /** |
7738 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
7739 | * @p: the task in question. | |
7740 | * @policy: new policy. | |
7741 | * @param: structure containing the new RT priority. | |
7742 | * | |
7743 | * Just like sched_setscheduler, only don't bother checking if the | |
7744 | * current context has permission. For example, this is needed in | |
7745 | * stop_machine(): we create temporary high priority worker threads, | |
7746 | * but our caller might not have that capability. | |
e69f6186 YB |
7747 | * |
7748 | * Return: 0 on success. An error code otherwise. | |
961ccddd RR |
7749 | */ |
7750 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 7751 | const struct sched_param *param) |
961ccddd | 7752 | { |
7479f3c9 | 7753 | return _sched_setscheduler(p, policy, param, false); |
961ccddd RR |
7754 | } |
7755 | ||
7318d4cc PZ |
7756 | /* |
7757 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | |
7758 | * incapable of resource management, which is the one thing an OS really should | |
7759 | * be doing. | |
7760 | * | |
7761 | * This is of course the reason it is limited to privileged users only. | |
7762 | * | |
7763 | * Worse still; it is fundamentally impossible to compose static priority | |
7764 | * workloads. You cannot take two correctly working static prio workloads | |
7765 | * and smash them together and still expect them to work. | |
7766 | * | |
7767 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | |
7768 | * | |
7769 | * MAX_RT_PRIO / 2 | |
7770 | * | |
7771 | * The administrator _MUST_ configure the system, the kernel simply doesn't | |
7772 | * know enough information to make a sensible choice. | |
7773 | */ | |
8b700983 | 7774 | void sched_set_fifo(struct task_struct *p) |
7318d4cc PZ |
7775 | { |
7776 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | |
8b700983 | 7777 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7778 | } |
7779 | EXPORT_SYMBOL_GPL(sched_set_fifo); | |
7780 | ||
7781 | /* | |
7782 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | |
7783 | */ | |
8b700983 | 7784 | void sched_set_fifo_low(struct task_struct *p) |
7318d4cc PZ |
7785 | { |
7786 | struct sched_param sp = { .sched_priority = 1 }; | |
8b700983 | 7787 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); |
7318d4cc PZ |
7788 | } |
7789 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | |
7790 | ||
8b700983 | 7791 | void sched_set_normal(struct task_struct *p, int nice) |
7318d4cc PZ |
7792 | { |
7793 | struct sched_attr attr = { | |
7794 | .sched_policy = SCHED_NORMAL, | |
7795 | .sched_nice = nice, | |
7796 | }; | |
8b700983 | 7797 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); |
7318d4cc PZ |
7798 | } |
7799 | EXPORT_SYMBOL_GPL(sched_set_normal); | |
961ccddd | 7800 | |
95cdf3b7 IM |
7801 | static int |
7802 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 7803 | { |
1da177e4 LT |
7804 | struct sched_param lparam; |
7805 | struct task_struct *p; | |
36c8b586 | 7806 | int retval; |
1da177e4 LT |
7807 | |
7808 | if (!param || pid < 0) | |
7809 | return -EINVAL; | |
7810 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
7811 | return -EFAULT; | |
5fe1d75f ON |
7812 | |
7813 | rcu_read_lock(); | |
7814 | retval = -ESRCH; | |
1da177e4 | 7815 | p = find_process_by_pid(pid); |
710da3c8 JL |
7816 | if (likely(p)) |
7817 | get_task_struct(p); | |
5fe1d75f | 7818 | rcu_read_unlock(); |
36c8b586 | 7819 | |
710da3c8 JL |
7820 | if (likely(p)) { |
7821 | retval = sched_setscheduler(p, policy, &lparam); | |
7822 | put_task_struct(p); | |
7823 | } | |
7824 | ||
1da177e4 LT |
7825 | return retval; |
7826 | } | |
7827 | ||
d50dde5a DF |
7828 | /* |
7829 | * Mimics kernel/events/core.c perf_copy_attr(). | |
7830 | */ | |
d1ccc66d | 7831 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) |
d50dde5a DF |
7832 | { |
7833 | u32 size; | |
7834 | int ret; | |
7835 | ||
d1ccc66d | 7836 | /* Zero the full structure, so that a short copy will be nice: */ |
d50dde5a DF |
7837 | memset(attr, 0, sizeof(*attr)); |
7838 | ||
7839 | ret = get_user(size, &uattr->size); | |
7840 | if (ret) | |
7841 | return ret; | |
7842 | ||
d1ccc66d IM |
7843 | /* ABI compatibility quirk: */ |
7844 | if (!size) | |
d50dde5a | 7845 | size = SCHED_ATTR_SIZE_VER0; |
dff3a85f | 7846 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) |
d50dde5a DF |
7847 | goto err_size; |
7848 | ||
dff3a85f AS |
7849 | ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size); |
7850 | if (ret) { | |
7851 | if (ret == -E2BIG) | |
7852 | goto err_size; | |
7853 | return ret; | |
d50dde5a DF |
7854 | } |
7855 | ||
a509a7cd PB |
7856 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && |
7857 | size < SCHED_ATTR_SIZE_VER1) | |
7858 | return -EINVAL; | |
7859 | ||
d50dde5a | 7860 | /* |
d1ccc66d | 7861 | * XXX: Do we want to be lenient like existing syscalls; or do we want |
d50dde5a DF |
7862 | * to be strict and return an error on out-of-bounds values? |
7863 | */ | |
75e45d51 | 7864 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
d50dde5a | 7865 | |
e78c7bca | 7866 | return 0; |
d50dde5a DF |
7867 | |
7868 | err_size: | |
7869 | put_user(sizeof(*attr), &uattr->size); | |
e78c7bca | 7870 | return -E2BIG; |
d50dde5a DF |
7871 | } |
7872 | ||
f4dddf90 QP |
7873 | static void get_params(struct task_struct *p, struct sched_attr *attr) |
7874 | { | |
7875 | if (task_has_dl_policy(p)) | |
7876 | __getparam_dl(p, attr); | |
7877 | else if (task_has_rt_policy(p)) | |
7878 | attr->sched_priority = p->rt_priority; | |
7879 | else | |
7880 | attr->sched_nice = task_nice(p); | |
7881 | } | |
7882 | ||
1da177e4 LT |
7883 | /** |
7884 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
7885 | * @pid: the pid in question. | |
7886 | * @policy: new policy. | |
7887 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7888 | * |
7889 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7890 | */ |
d1ccc66d | 7891 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) |
1da177e4 | 7892 | { |
c21761f1 JB |
7893 | if (policy < 0) |
7894 | return -EINVAL; | |
7895 | ||
1da177e4 LT |
7896 | return do_sched_setscheduler(pid, policy, param); |
7897 | } | |
7898 | ||
7899 | /** | |
7900 | * sys_sched_setparam - set/change the RT priority of a thread | |
7901 | * @pid: the pid in question. | |
7902 | * @param: structure containing the new RT priority. | |
e69f6186 YB |
7903 | * |
7904 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 7905 | */ |
5add95d4 | 7906 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7907 | { |
c13db6b1 | 7908 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
1da177e4 LT |
7909 | } |
7910 | ||
d50dde5a DF |
7911 | /** |
7912 | * sys_sched_setattr - same as above, but with extended sched_attr | |
7913 | * @pid: the pid in question. | |
5778fccf | 7914 | * @uattr: structure containing the extended parameters. |
db66d756 | 7915 | * @flags: for future extension. |
d50dde5a | 7916 | */ |
6d35ab48 PZ |
7917 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
7918 | unsigned int, flags) | |
d50dde5a DF |
7919 | { |
7920 | struct sched_attr attr; | |
7921 | struct task_struct *p; | |
7922 | int retval; | |
7923 | ||
6d35ab48 | 7924 | if (!uattr || pid < 0 || flags) |
d50dde5a DF |
7925 | return -EINVAL; |
7926 | ||
143cf23d MK |
7927 | retval = sched_copy_attr(uattr, &attr); |
7928 | if (retval) | |
7929 | return retval; | |
d50dde5a | 7930 | |
b14ed2c2 | 7931 | if ((int)attr.sched_policy < 0) |
dbdb2275 | 7932 | return -EINVAL; |
1d6362fa PB |
7933 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) |
7934 | attr.sched_policy = SETPARAM_POLICY; | |
d50dde5a DF |
7935 | |
7936 | rcu_read_lock(); | |
7937 | retval = -ESRCH; | |
7938 | p = find_process_by_pid(pid); | |
a509a7cd PB |
7939 | if (likely(p)) |
7940 | get_task_struct(p); | |
d50dde5a DF |
7941 | rcu_read_unlock(); |
7942 | ||
a509a7cd | 7943 | if (likely(p)) { |
f4dddf90 QP |
7944 | if (attr.sched_flags & SCHED_FLAG_KEEP_PARAMS) |
7945 | get_params(p, &attr); | |
a509a7cd PB |
7946 | retval = sched_setattr(p, &attr); |
7947 | put_task_struct(p); | |
7948 | } | |
7949 | ||
d50dde5a DF |
7950 | return retval; |
7951 | } | |
7952 | ||
1da177e4 LT |
7953 | /** |
7954 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
7955 | * @pid: the pid in question. | |
e69f6186 YB |
7956 | * |
7957 | * Return: On success, the policy of the thread. Otherwise, a negative error | |
7958 | * code. | |
1da177e4 | 7959 | */ |
5add95d4 | 7960 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 7961 | { |
36c8b586 | 7962 | struct task_struct *p; |
3a5c359a | 7963 | int retval; |
1da177e4 LT |
7964 | |
7965 | if (pid < 0) | |
3a5c359a | 7966 | return -EINVAL; |
1da177e4 LT |
7967 | |
7968 | retval = -ESRCH; | |
5fe85be0 | 7969 | rcu_read_lock(); |
1da177e4 LT |
7970 | p = find_process_by_pid(pid); |
7971 | if (p) { | |
7972 | retval = security_task_getscheduler(p); | |
7973 | if (!retval) | |
ca94c442 LP |
7974 | retval = p->policy |
7975 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 7976 | } |
5fe85be0 | 7977 | rcu_read_unlock(); |
1da177e4 LT |
7978 | return retval; |
7979 | } | |
7980 | ||
7981 | /** | |
ca94c442 | 7982 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
7983 | * @pid: the pid in question. |
7984 | * @param: structure containing the RT priority. | |
e69f6186 YB |
7985 | * |
7986 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | |
7987 | * code. | |
1da177e4 | 7988 | */ |
5add95d4 | 7989 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 | 7990 | { |
ce5f7f82 | 7991 | struct sched_param lp = { .sched_priority = 0 }; |
36c8b586 | 7992 | struct task_struct *p; |
3a5c359a | 7993 | int retval; |
1da177e4 LT |
7994 | |
7995 | if (!param || pid < 0) | |
3a5c359a | 7996 | return -EINVAL; |
1da177e4 | 7997 | |
5fe85be0 | 7998 | rcu_read_lock(); |
1da177e4 LT |
7999 | p = find_process_by_pid(pid); |
8000 | retval = -ESRCH; | |
8001 | if (!p) | |
8002 | goto out_unlock; | |
8003 | ||
8004 | retval = security_task_getscheduler(p); | |
8005 | if (retval) | |
8006 | goto out_unlock; | |
8007 | ||
ce5f7f82 PZ |
8008 | if (task_has_rt_policy(p)) |
8009 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 8010 | rcu_read_unlock(); |
1da177e4 LT |
8011 | |
8012 | /* | |
8013 | * This one might sleep, we cannot do it with a spinlock held ... | |
8014 | */ | |
8015 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
8016 | ||
1da177e4 LT |
8017 | return retval; |
8018 | ||
8019 | out_unlock: | |
5fe85be0 | 8020 | rcu_read_unlock(); |
1da177e4 LT |
8021 | return retval; |
8022 | } | |
8023 | ||
1251201c IM |
8024 | /* |
8025 | * Copy the kernel size attribute structure (which might be larger | |
8026 | * than what user-space knows about) to user-space. | |
8027 | * | |
8028 | * Note that all cases are valid: user-space buffer can be larger or | |
8029 | * smaller than the kernel-space buffer. The usual case is that both | |
8030 | * have the same size. | |
8031 | */ | |
8032 | static int | |
8033 | sched_attr_copy_to_user(struct sched_attr __user *uattr, | |
8034 | struct sched_attr *kattr, | |
8035 | unsigned int usize) | |
d50dde5a | 8036 | { |
1251201c | 8037 | unsigned int ksize = sizeof(*kattr); |
d50dde5a | 8038 | |
96d4f267 | 8039 | if (!access_ok(uattr, usize)) |
d50dde5a DF |
8040 | return -EFAULT; |
8041 | ||
8042 | /* | |
1251201c IM |
8043 | * sched_getattr() ABI forwards and backwards compatibility: |
8044 | * | |
8045 | * If usize == ksize then we just copy everything to user-space and all is good. | |
8046 | * | |
8047 | * If usize < ksize then we only copy as much as user-space has space for, | |
8048 | * this keeps ABI compatibility as well. We skip the rest. | |
8049 | * | |
8050 | * If usize > ksize then user-space is using a newer version of the ABI, | |
8051 | * which part the kernel doesn't know about. Just ignore it - tooling can | |
8052 | * detect the kernel's knowledge of attributes from the attr->size value | |
8053 | * which is set to ksize in this case. | |
d50dde5a | 8054 | */ |
1251201c | 8055 | kattr->size = min(usize, ksize); |
d50dde5a | 8056 | |
1251201c | 8057 | if (copy_to_user(uattr, kattr, kattr->size)) |
d50dde5a DF |
8058 | return -EFAULT; |
8059 | ||
22400674 | 8060 | return 0; |
d50dde5a DF |
8061 | } |
8062 | ||
8063 | /** | |
aab03e05 | 8064 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
d50dde5a | 8065 | * @pid: the pid in question. |
5778fccf | 8066 | * @uattr: structure containing the extended parameters. |
dff3a85f | 8067 | * @usize: sizeof(attr) for fwd/bwd comp. |
db66d756 | 8068 | * @flags: for future extension. |
d50dde5a | 8069 | */ |
6d35ab48 | 8070 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
1251201c | 8071 | unsigned int, usize, unsigned int, flags) |
d50dde5a | 8072 | { |
1251201c | 8073 | struct sched_attr kattr = { }; |
d50dde5a DF |
8074 | struct task_struct *p; |
8075 | int retval; | |
8076 | ||
1251201c IM |
8077 | if (!uattr || pid < 0 || usize > PAGE_SIZE || |
8078 | usize < SCHED_ATTR_SIZE_VER0 || flags) | |
d50dde5a DF |
8079 | return -EINVAL; |
8080 | ||
8081 | rcu_read_lock(); | |
8082 | p = find_process_by_pid(pid); | |
8083 | retval = -ESRCH; | |
8084 | if (!p) | |
8085 | goto out_unlock; | |
8086 | ||
8087 | retval = security_task_getscheduler(p); | |
8088 | if (retval) | |
8089 | goto out_unlock; | |
8090 | ||
1251201c | 8091 | kattr.sched_policy = p->policy; |
7479f3c9 | 8092 | if (p->sched_reset_on_fork) |
1251201c | 8093 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
f4dddf90 | 8094 | get_params(p, &kattr); |
7ad721bf | 8095 | kattr.sched_flags &= SCHED_FLAG_ALL; |
d50dde5a | 8096 | |
a509a7cd | 8097 | #ifdef CONFIG_UCLAMP_TASK |
13685c4a QY |
8098 | /* |
8099 | * This could race with another potential updater, but this is fine | |
8100 | * because it'll correctly read the old or the new value. We don't need | |
8101 | * to guarantee who wins the race as long as it doesn't return garbage. | |
8102 | */ | |
1251201c IM |
8103 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; |
8104 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | |
a509a7cd PB |
8105 | #endif |
8106 | ||
d50dde5a DF |
8107 | rcu_read_unlock(); |
8108 | ||
1251201c | 8109 | return sched_attr_copy_to_user(uattr, &kattr, usize); |
d50dde5a DF |
8110 | |
8111 | out_unlock: | |
8112 | rcu_read_unlock(); | |
8113 | return retval; | |
8114 | } | |
8115 | ||
234b8ab6 WD |
8116 | #ifdef CONFIG_SMP |
8117 | int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask) | |
1da177e4 | 8118 | { |
234b8ab6 WD |
8119 | int ret = 0; |
8120 | ||
8121 | /* | |
8122 | * If the task isn't a deadline task or admission control is | |
8123 | * disabled then we don't care about affinity changes. | |
8124 | */ | |
8125 | if (!task_has_dl_policy(p) || !dl_bandwidth_enabled()) | |
8126 | return 0; | |
8127 | ||
8128 | /* | |
8129 | * Since bandwidth control happens on root_domain basis, | |
8130 | * if admission test is enabled, we only admit -deadline | |
8131 | * tasks allowed to run on all the CPUs in the task's | |
8132 | * root_domain. | |
8133 | */ | |
8134 | rcu_read_lock(); | |
8135 | if (!cpumask_subset(task_rq(p)->rd->span, mask)) | |
8136 | ret = -EBUSY; | |
8137 | rcu_read_unlock(); | |
8138 | return ret; | |
8139 | } | |
8140 | #endif | |
8141 | ||
db3b02ae | 8142 | static int |
713a2e21 | 8143 | __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx) |
1da177e4 | 8144 | { |
36c8b586 | 8145 | int retval; |
5a16f3d3 | 8146 | cpumask_var_t cpus_allowed, new_mask; |
1da177e4 | 8147 | |
db3b02ae WD |
8148 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) |
8149 | return -ENOMEM; | |
1da177e4 | 8150 | |
5a16f3d3 RR |
8151 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { |
8152 | retval = -ENOMEM; | |
8153 | goto out_free_cpus_allowed; | |
8154 | } | |
e4099a5e PZ |
8155 | |
8156 | cpuset_cpus_allowed(p, cpus_allowed); | |
713a2e21 WL |
8157 | cpumask_and(new_mask, ctx->new_mask, cpus_allowed); |
8158 | ||
8159 | ctx->new_mask = new_mask; | |
8160 | ctx->flags |= SCA_CHECK; | |
e4099a5e | 8161 | |
234b8ab6 WD |
8162 | retval = dl_task_check_affinity(p, new_mask); |
8163 | if (retval) | |
8164 | goto out_free_new_mask; | |
8f9ea86f | 8165 | |
713a2e21 | 8166 | retval = __set_cpus_allowed_ptr(p, ctx); |
db3b02ae WD |
8167 | if (retval) |
8168 | goto out_free_new_mask; | |
1da177e4 | 8169 | |
db3b02ae WD |
8170 | cpuset_cpus_allowed(p, cpus_allowed); |
8171 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8172 | /* | |
8173 | * We must have raced with a concurrent cpuset update. | |
8174 | * Just reset the cpumask to the cpuset's cpus_allowed. | |
8175 | */ | |
8176 | cpumask_copy(new_mask, cpus_allowed); | |
8f9ea86f WL |
8177 | |
8178 | /* | |
8179 | * If SCA_USER is set, a 2nd call to __set_cpus_allowed_ptr() | |
8180 | * will restore the previous user_cpus_ptr value. | |
8181 | * | |
8182 | * In the unlikely event a previous user_cpus_ptr exists, | |
8183 | * we need to further restrict the mask to what is allowed | |
8184 | * by that old user_cpus_ptr. | |
8185 | */ | |
8186 | if (unlikely((ctx->flags & SCA_USER) && ctx->user_mask)) { | |
8187 | bool empty = !cpumask_and(new_mask, new_mask, | |
8188 | ctx->user_mask); | |
8189 | ||
8190 | if (WARN_ON_ONCE(empty)) | |
8191 | cpumask_copy(new_mask, cpus_allowed); | |
8192 | } | |
8193 | __set_cpus_allowed_ptr(p, ctx); | |
8194 | retval = -EINVAL; | |
8707d8b8 | 8195 | } |
db3b02ae | 8196 | |
16303ab2 | 8197 | out_free_new_mask: |
5a16f3d3 RR |
8198 | free_cpumask_var(new_mask); |
8199 | out_free_cpus_allowed: | |
8200 | free_cpumask_var(cpus_allowed); | |
db3b02ae WD |
8201 | return retval; |
8202 | } | |
8203 | ||
8204 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) | |
8205 | { | |
8f9ea86f WL |
8206 | struct affinity_context ac; |
8207 | struct cpumask *user_mask; | |
36c8b586 IM |
8208 | struct task_struct *p; |
8209 | int retval; | |
1da177e4 | 8210 | |
23f5d142 | 8211 | rcu_read_lock(); |
1da177e4 LT |
8212 | |
8213 | p = find_process_by_pid(pid); | |
8214 | if (!p) { | |
23f5d142 | 8215 | rcu_read_unlock(); |
1da177e4 LT |
8216 | return -ESRCH; |
8217 | } | |
8218 | ||
23f5d142 | 8219 | /* Prevent p going away */ |
1da177e4 | 8220 | get_task_struct(p); |
23f5d142 | 8221 | rcu_read_unlock(); |
1da177e4 | 8222 | |
14a40ffc TH |
8223 | if (p->flags & PF_NO_SETAFFINITY) { |
8224 | retval = -EINVAL; | |
8225 | goto out_put_task; | |
8226 | } | |
db3b02ae | 8227 | |
4c44aaaf EB |
8228 | if (!check_same_owner(p)) { |
8229 | rcu_read_lock(); | |
8230 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { | |
8231 | rcu_read_unlock(); | |
db3b02ae WD |
8232 | retval = -EPERM; |
8233 | goto out_put_task; | |
4c44aaaf EB |
8234 | } |
8235 | rcu_read_unlock(); | |
8236 | } | |
1da177e4 | 8237 | |
b0ae1981 | 8238 | retval = security_task_setscheduler(p); |
e7834f8f | 8239 | if (retval) |
db3b02ae | 8240 | goto out_put_task; |
1da177e4 | 8241 | |
8f9ea86f WL |
8242 | user_mask = kmalloc(cpumask_size(), GFP_KERNEL); |
8243 | if (!user_mask) { | |
8244 | retval = -ENOMEM; | |
8245 | goto out_put_task; | |
8246 | } | |
8247 | cpumask_copy(user_mask, in_mask); | |
8248 | ac = (struct affinity_context){ | |
8249 | .new_mask = in_mask, | |
8250 | .user_mask = user_mask, | |
8251 | .flags = SCA_USER, | |
8252 | }; | |
8253 | ||
713a2e21 | 8254 | retval = __sched_setaffinity(p, &ac); |
8f9ea86f WL |
8255 | kfree(ac.user_mask); |
8256 | ||
5a16f3d3 | 8257 | out_put_task: |
1da177e4 | 8258 | put_task_struct(p); |
1da177e4 LT |
8259 | return retval; |
8260 | } | |
8261 | ||
8262 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 8263 | struct cpumask *new_mask) |
1da177e4 | 8264 | { |
96f874e2 RR |
8265 | if (len < cpumask_size()) |
8266 | cpumask_clear(new_mask); | |
8267 | else if (len > cpumask_size()) | |
8268 | len = cpumask_size(); | |
8269 | ||
1da177e4 LT |
8270 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
8271 | } | |
8272 | ||
8273 | /** | |
d1ccc66d | 8274 | * sys_sched_setaffinity - set the CPU affinity of a process |
1da177e4 LT |
8275 | * @pid: pid of the process |
8276 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8277 | * @user_mask_ptr: user-space pointer to the new CPU mask |
e69f6186 YB |
8278 | * |
8279 | * Return: 0 on success. An error code otherwise. | |
1da177e4 | 8280 | */ |
5add95d4 HC |
8281 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
8282 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 8283 | { |
5a16f3d3 | 8284 | cpumask_var_t new_mask; |
1da177e4 LT |
8285 | int retval; |
8286 | ||
5a16f3d3 RR |
8287 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
8288 | return -ENOMEM; | |
1da177e4 | 8289 | |
5a16f3d3 RR |
8290 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
8291 | if (retval == 0) | |
8292 | retval = sched_setaffinity(pid, new_mask); | |
8293 | free_cpumask_var(new_mask); | |
8294 | return retval; | |
1da177e4 LT |
8295 | } |
8296 | ||
96f874e2 | 8297 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 8298 | { |
36c8b586 | 8299 | struct task_struct *p; |
31605683 | 8300 | unsigned long flags; |
1da177e4 | 8301 | int retval; |
1da177e4 | 8302 | |
23f5d142 | 8303 | rcu_read_lock(); |
1da177e4 LT |
8304 | |
8305 | retval = -ESRCH; | |
8306 | p = find_process_by_pid(pid); | |
8307 | if (!p) | |
8308 | goto out_unlock; | |
8309 | ||
e7834f8f DQ |
8310 | retval = security_task_getscheduler(p); |
8311 | if (retval) | |
8312 | goto out_unlock; | |
8313 | ||
013fdb80 | 8314 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
3bd37062 | 8315 | cpumask_and(mask, &p->cpus_mask, cpu_active_mask); |
013fdb80 | 8316 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
8317 | |
8318 | out_unlock: | |
23f5d142 | 8319 | rcu_read_unlock(); |
1da177e4 | 8320 | |
9531b62f | 8321 | return retval; |
1da177e4 LT |
8322 | } |
8323 | ||
8324 | /** | |
d1ccc66d | 8325 | * sys_sched_getaffinity - get the CPU affinity of a process |
1da177e4 LT |
8326 | * @pid: pid of the process |
8327 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
d1ccc66d | 8328 | * @user_mask_ptr: user-space pointer to hold the current CPU mask |
e69f6186 | 8329 | * |
599b4840 ZW |
8330 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
8331 | * error code otherwise. | |
1da177e4 | 8332 | */ |
5add95d4 HC |
8333 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
8334 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
8335 | { |
8336 | int ret; | |
f17c8607 | 8337 | cpumask_var_t mask; |
1da177e4 | 8338 | |
84fba5ec | 8339 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
8340 | return -EINVAL; |
8341 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
8342 | return -EINVAL; |
8343 | ||
f17c8607 RR |
8344 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
8345 | return -ENOMEM; | |
1da177e4 | 8346 | |
f17c8607 RR |
8347 | ret = sched_getaffinity(pid, mask); |
8348 | if (ret == 0) { | |
4de373a1 | 8349 | unsigned int retlen = min(len, cpumask_size()); |
cd3d8031 KM |
8350 | |
8351 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
8352 | ret = -EFAULT; |
8353 | else | |
cd3d8031 | 8354 | ret = retlen; |
f17c8607 RR |
8355 | } |
8356 | free_cpumask_var(mask); | |
1da177e4 | 8357 | |
f17c8607 | 8358 | return ret; |
1da177e4 LT |
8359 | } |
8360 | ||
7d4dd4f1 | 8361 | static void do_sched_yield(void) |
1da177e4 | 8362 | { |
8a8c69c3 PZ |
8363 | struct rq_flags rf; |
8364 | struct rq *rq; | |
8365 | ||
246b3b33 | 8366 | rq = this_rq_lock_irq(&rf); |
1da177e4 | 8367 | |
ae92882e | 8368 | schedstat_inc(rq->yld_count); |
4530d7ab | 8369 | current->sched_class->yield_task(rq); |
1da177e4 | 8370 | |
8a8c69c3 | 8371 | preempt_disable(); |
345a957f | 8372 | rq_unlock_irq(rq, &rf); |
ba74c144 | 8373 | sched_preempt_enable_no_resched(); |
1da177e4 LT |
8374 | |
8375 | schedule(); | |
7d4dd4f1 | 8376 | } |
1da177e4 | 8377 | |
59a74b15 MCC |
8378 | /** |
8379 | * sys_sched_yield - yield the current processor to other threads. | |
8380 | * | |
8381 | * This function yields the current CPU to other tasks. If there are no | |
8382 | * other threads running on this CPU then this function will return. | |
8383 | * | |
8384 | * Return: 0. | |
8385 | */ | |
7d4dd4f1 DB |
8386 | SYSCALL_DEFINE0(sched_yield) |
8387 | { | |
8388 | do_sched_yield(); | |
1da177e4 LT |
8389 | return 0; |
8390 | } | |
8391 | ||
b965f1dd PZI |
8392 | #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) |
8393 | int __sched __cond_resched(void) | |
1da177e4 | 8394 | { |
fe32d3cd | 8395 | if (should_resched(0)) { |
a18b5d01 | 8396 | preempt_schedule_common(); |
1da177e4 LT |
8397 | return 1; |
8398 | } | |
50895825 FW |
8399 | /* |
8400 | * In preemptible kernels, ->rcu_read_lock_nesting tells the tick | |
8401 | * whether the current CPU is in an RCU read-side critical section, | |
8402 | * so the tick can report quiescent states even for CPUs looping | |
8403 | * in kernel context. In contrast, in non-preemptible kernels, | |
8404 | * RCU readers leave no in-memory hints, which means that CPU-bound | |
8405 | * processes executing in kernel context might never report an | |
8406 | * RCU quiescent state. Therefore, the following code causes | |
8407 | * cond_resched() to report a quiescent state, but only when RCU | |
8408 | * is in urgent need of one. | |
8409 | */ | |
b965f1dd | 8410 | #ifndef CONFIG_PREEMPT_RCU |
f79c3ad6 | 8411 | rcu_all_qs(); |
b965f1dd | 8412 | #endif |
1da177e4 LT |
8413 | return 0; |
8414 | } | |
b965f1dd PZI |
8415 | EXPORT_SYMBOL(__cond_resched); |
8416 | #endif | |
8417 | ||
8418 | #ifdef CONFIG_PREEMPT_DYNAMIC | |
99cf983c | 8419 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
8420 | #define cond_resched_dynamic_enabled __cond_resched |
8421 | #define cond_resched_dynamic_disabled ((void *)&__static_call_return0) | |
b965f1dd | 8422 | DEFINE_STATIC_CALL_RET0(cond_resched, __cond_resched); |
ef72661e | 8423 | EXPORT_STATIC_CALL_TRAMP(cond_resched); |
b965f1dd | 8424 | |
8a69fe0b MR |
8425 | #define might_resched_dynamic_enabled __cond_resched |
8426 | #define might_resched_dynamic_disabled ((void *)&__static_call_return0) | |
b965f1dd | 8427 | DEFINE_STATIC_CALL_RET0(might_resched, __cond_resched); |
ef72661e | 8428 | EXPORT_STATIC_CALL_TRAMP(might_resched); |
99cf983c MR |
8429 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
8430 | static DEFINE_STATIC_KEY_FALSE(sk_dynamic_cond_resched); | |
8431 | int __sched dynamic_cond_resched(void) | |
8432 | { | |
8433 | if (!static_branch_unlikely(&sk_dynamic_cond_resched)) | |
8434 | return 0; | |
8435 | return __cond_resched(); | |
8436 | } | |
8437 | EXPORT_SYMBOL(dynamic_cond_resched); | |
8438 | ||
8439 | static DEFINE_STATIC_KEY_FALSE(sk_dynamic_might_resched); | |
8440 | int __sched dynamic_might_resched(void) | |
8441 | { | |
8442 | if (!static_branch_unlikely(&sk_dynamic_might_resched)) | |
8443 | return 0; | |
8444 | return __cond_resched(); | |
8445 | } | |
8446 | EXPORT_SYMBOL(dynamic_might_resched); | |
8447 | #endif | |
35a773a0 | 8448 | #endif |
1da177e4 LT |
8449 | |
8450 | /* | |
613afbf8 | 8451 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
8452 | * call schedule, and on return reacquire the lock. |
8453 | * | |
c1a280b6 | 8454 | * This works OK both with and without CONFIG_PREEMPTION. We do strange low-level |
1da177e4 LT |
8455 | * operations here to prevent schedule() from being called twice (once via |
8456 | * spin_unlock(), once by hand). | |
8457 | */ | |
613afbf8 | 8458 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 8459 | { |
fe32d3cd | 8460 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
6df3cecb JK |
8461 | int ret = 0; |
8462 | ||
f607c668 PZ |
8463 | lockdep_assert_held(lock); |
8464 | ||
4a81e832 | 8465 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 8466 | spin_unlock(lock); |
7e406d1f | 8467 | if (!_cond_resched()) |
95c354fe | 8468 | cpu_relax(); |
6df3cecb | 8469 | ret = 1; |
1da177e4 | 8470 | spin_lock(lock); |
1da177e4 | 8471 | } |
6df3cecb | 8472 | return ret; |
1da177e4 | 8473 | } |
613afbf8 | 8474 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 8475 | |
f3d4b4b1 BG |
8476 | int __cond_resched_rwlock_read(rwlock_t *lock) |
8477 | { | |
8478 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8479 | int ret = 0; | |
8480 | ||
8481 | lockdep_assert_held_read(lock); | |
8482 | ||
8483 | if (rwlock_needbreak(lock) || resched) { | |
8484 | read_unlock(lock); | |
7e406d1f | 8485 | if (!_cond_resched()) |
f3d4b4b1 BG |
8486 | cpu_relax(); |
8487 | ret = 1; | |
8488 | read_lock(lock); | |
8489 | } | |
8490 | return ret; | |
8491 | } | |
8492 | EXPORT_SYMBOL(__cond_resched_rwlock_read); | |
8493 | ||
8494 | int __cond_resched_rwlock_write(rwlock_t *lock) | |
8495 | { | |
8496 | int resched = should_resched(PREEMPT_LOCK_OFFSET); | |
8497 | int ret = 0; | |
8498 | ||
8499 | lockdep_assert_held_write(lock); | |
8500 | ||
8501 | if (rwlock_needbreak(lock) || resched) { | |
8502 | write_unlock(lock); | |
7e406d1f | 8503 | if (!_cond_resched()) |
f3d4b4b1 BG |
8504 | cpu_relax(); |
8505 | ret = 1; | |
8506 | write_lock(lock); | |
8507 | } | |
8508 | return ret; | |
8509 | } | |
8510 | EXPORT_SYMBOL(__cond_resched_rwlock_write); | |
8511 | ||
4c748558 MR |
8512 | #ifdef CONFIG_PREEMPT_DYNAMIC |
8513 | ||
33c64734 | 8514 | #ifdef CONFIG_GENERIC_ENTRY |
4c748558 | 8515 | #include <linux/entry-common.h> |
33c64734 | 8516 | #endif |
4c748558 MR |
8517 | |
8518 | /* | |
8519 | * SC:cond_resched | |
8520 | * SC:might_resched | |
8521 | * SC:preempt_schedule | |
8522 | * SC:preempt_schedule_notrace | |
8523 | * SC:irqentry_exit_cond_resched | |
8524 | * | |
8525 | * | |
8526 | * NONE: | |
8527 | * cond_resched <- __cond_resched | |
8528 | * might_resched <- RET0 | |
8529 | * preempt_schedule <- NOP | |
8530 | * preempt_schedule_notrace <- NOP | |
8531 | * irqentry_exit_cond_resched <- NOP | |
8532 | * | |
8533 | * VOLUNTARY: | |
8534 | * cond_resched <- __cond_resched | |
8535 | * might_resched <- __cond_resched | |
8536 | * preempt_schedule <- NOP | |
8537 | * preempt_schedule_notrace <- NOP | |
8538 | * irqentry_exit_cond_resched <- NOP | |
8539 | * | |
8540 | * FULL: | |
8541 | * cond_resched <- RET0 | |
8542 | * might_resched <- RET0 | |
8543 | * preempt_schedule <- preempt_schedule | |
8544 | * preempt_schedule_notrace <- preempt_schedule_notrace | |
8545 | * irqentry_exit_cond_resched <- irqentry_exit_cond_resched | |
8546 | */ | |
8547 | ||
8548 | enum { | |
8549 | preempt_dynamic_undefined = -1, | |
8550 | preempt_dynamic_none, | |
8551 | preempt_dynamic_voluntary, | |
8552 | preempt_dynamic_full, | |
8553 | }; | |
8554 | ||
8555 | int preempt_dynamic_mode = preempt_dynamic_undefined; | |
8556 | ||
8557 | int sched_dynamic_mode(const char *str) | |
8558 | { | |
8559 | if (!strcmp(str, "none")) | |
8560 | return preempt_dynamic_none; | |
8561 | ||
8562 | if (!strcmp(str, "voluntary")) | |
8563 | return preempt_dynamic_voluntary; | |
8564 | ||
8565 | if (!strcmp(str, "full")) | |
8566 | return preempt_dynamic_full; | |
8567 | ||
8568 | return -EINVAL; | |
8569 | } | |
8570 | ||
99cf983c | 8571 | #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) |
8a69fe0b MR |
8572 | #define preempt_dynamic_enable(f) static_call_update(f, f##_dynamic_enabled) |
8573 | #define preempt_dynamic_disable(f) static_call_update(f, f##_dynamic_disabled) | |
99cf983c MR |
8574 | #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) |
8575 | #define preempt_dynamic_enable(f) static_key_enable(&sk_dynamic_##f.key) | |
8576 | #define preempt_dynamic_disable(f) static_key_disable(&sk_dynamic_##f.key) | |
8577 | #else | |
8578 | #error "Unsupported PREEMPT_DYNAMIC mechanism" | |
8579 | #endif | |
8a69fe0b | 8580 | |
4c748558 MR |
8581 | void sched_dynamic_update(int mode) |
8582 | { | |
8583 | /* | |
8584 | * Avoid {NONE,VOLUNTARY} -> FULL transitions from ever ending up in | |
8585 | * the ZERO state, which is invalid. | |
8586 | */ | |
8a69fe0b MR |
8587 | preempt_dynamic_enable(cond_resched); |
8588 | preempt_dynamic_enable(might_resched); | |
8589 | preempt_dynamic_enable(preempt_schedule); | |
8590 | preempt_dynamic_enable(preempt_schedule_notrace); | |
8591 | preempt_dynamic_enable(irqentry_exit_cond_resched); | |
4c748558 MR |
8592 | |
8593 | switch (mode) { | |
8594 | case preempt_dynamic_none: | |
8a69fe0b MR |
8595 | preempt_dynamic_enable(cond_resched); |
8596 | preempt_dynamic_disable(might_resched); | |
8597 | preempt_dynamic_disable(preempt_schedule); | |
8598 | preempt_dynamic_disable(preempt_schedule_notrace); | |
8599 | preempt_dynamic_disable(irqentry_exit_cond_resched); | |
4c748558 MR |
8600 | pr_info("Dynamic Preempt: none\n"); |
8601 | break; | |
8602 | ||
8603 | case preempt_dynamic_voluntary: | |
8a69fe0b MR |
8604 | preempt_dynamic_enable(cond_resched); |
8605 | preempt_dynamic_enable(might_resched); | |
8606 | preempt_dynamic_disable(preempt_schedule); | |
8607 | preempt_dynamic_disable(preempt_schedule_notrace); | |
8608 | preempt_dynamic_disable(irqentry_exit_cond_resched); | |
4c748558 MR |
8609 | pr_info("Dynamic Preempt: voluntary\n"); |
8610 | break; | |
8611 | ||
8612 | case preempt_dynamic_full: | |
8a69fe0b MR |
8613 | preempt_dynamic_disable(cond_resched); |
8614 | preempt_dynamic_disable(might_resched); | |
8615 | preempt_dynamic_enable(preempt_schedule); | |
8616 | preempt_dynamic_enable(preempt_schedule_notrace); | |
8617 | preempt_dynamic_enable(irqentry_exit_cond_resched); | |
4c748558 MR |
8618 | pr_info("Dynamic Preempt: full\n"); |
8619 | break; | |
8620 | } | |
8621 | ||
8622 | preempt_dynamic_mode = mode; | |
8623 | } | |
8624 | ||
8625 | static int __init setup_preempt_mode(char *str) | |
8626 | { | |
8627 | int mode = sched_dynamic_mode(str); | |
8628 | if (mode < 0) { | |
8629 | pr_warn("Dynamic Preempt: unsupported mode: %s\n", str); | |
8630 | return 0; | |
8631 | } | |
8632 | ||
8633 | sched_dynamic_update(mode); | |
8634 | return 1; | |
8635 | } | |
8636 | __setup("preempt=", setup_preempt_mode); | |
8637 | ||
8638 | static void __init preempt_dynamic_init(void) | |
8639 | { | |
8640 | if (preempt_dynamic_mode == preempt_dynamic_undefined) { | |
8641 | if (IS_ENABLED(CONFIG_PREEMPT_NONE)) { | |
8642 | sched_dynamic_update(preempt_dynamic_none); | |
8643 | } else if (IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY)) { | |
8644 | sched_dynamic_update(preempt_dynamic_voluntary); | |
8645 | } else { | |
8646 | /* Default static call setting, nothing to do */ | |
8647 | WARN_ON_ONCE(!IS_ENABLED(CONFIG_PREEMPT)); | |
8648 | preempt_dynamic_mode = preempt_dynamic_full; | |
8649 | pr_info("Dynamic Preempt: full\n"); | |
8650 | } | |
8651 | } | |
8652 | } | |
8653 | ||
cfe43f47 VS |
8654 | #define PREEMPT_MODEL_ACCESSOR(mode) \ |
8655 | bool preempt_model_##mode(void) \ | |
8656 | { \ | |
8657 | WARN_ON_ONCE(preempt_dynamic_mode == preempt_dynamic_undefined); \ | |
8658 | return preempt_dynamic_mode == preempt_dynamic_##mode; \ | |
8659 | } \ | |
8660 | EXPORT_SYMBOL_GPL(preempt_model_##mode) | |
8661 | ||
8662 | PREEMPT_MODEL_ACCESSOR(none); | |
8663 | PREEMPT_MODEL_ACCESSOR(voluntary); | |
8664 | PREEMPT_MODEL_ACCESSOR(full); | |
8665 | ||
4c748558 MR |
8666 | #else /* !CONFIG_PREEMPT_DYNAMIC */ |
8667 | ||
8668 | static inline void preempt_dynamic_init(void) { } | |
8669 | ||
8670 | #endif /* #ifdef CONFIG_PREEMPT_DYNAMIC */ | |
8671 | ||
1da177e4 LT |
8672 | /** |
8673 | * yield - yield the current processor to other threads. | |
8674 | * | |
8e3fabfd PZ |
8675 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
8676 | * | |
8677 | * The scheduler is at all times free to pick the calling task as the most | |
8678 | * eligible task to run, if removing the yield() call from your code breaks | |
b19a888c | 8679 | * it, it's already broken. |
8e3fabfd PZ |
8680 | * |
8681 | * Typical broken usage is: | |
8682 | * | |
8683 | * while (!event) | |
d1ccc66d | 8684 | * yield(); |
8e3fabfd PZ |
8685 | * |
8686 | * where one assumes that yield() will let 'the other' process run that will | |
8687 | * make event true. If the current task is a SCHED_FIFO task that will never | |
8688 | * happen. Never use yield() as a progress guarantee!! | |
8689 | * | |
8690 | * If you want to use yield() to wait for something, use wait_event(). | |
8691 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | |
8692 | * If you still want to use yield(), do not! | |
1da177e4 LT |
8693 | */ |
8694 | void __sched yield(void) | |
8695 | { | |
8696 | set_current_state(TASK_RUNNING); | |
7d4dd4f1 | 8697 | do_sched_yield(); |
1da177e4 | 8698 | } |
1da177e4 LT |
8699 | EXPORT_SYMBOL(yield); |
8700 | ||
d95f4122 MG |
8701 | /** |
8702 | * yield_to - yield the current processor to another thread in | |
8703 | * your thread group, or accelerate that thread toward the | |
8704 | * processor it's on. | |
16addf95 RD |
8705 | * @p: target task |
8706 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
8707 | * |
8708 | * It's the caller's job to ensure that the target task struct | |
8709 | * can't go away on us before we can do any checks. | |
8710 | * | |
e69f6186 | 8711 | * Return: |
7b270f60 PZ |
8712 | * true (>0) if we indeed boosted the target task. |
8713 | * false (0) if we failed to boost the target. | |
8714 | * -ESRCH if there's no task to yield to. | |
d95f4122 | 8715 | */ |
fa93384f | 8716 | int __sched yield_to(struct task_struct *p, bool preempt) |
d95f4122 MG |
8717 | { |
8718 | struct task_struct *curr = current; | |
8719 | struct rq *rq, *p_rq; | |
8720 | unsigned long flags; | |
c3c18640 | 8721 | int yielded = 0; |
d95f4122 MG |
8722 | |
8723 | local_irq_save(flags); | |
8724 | rq = this_rq(); | |
8725 | ||
8726 | again: | |
8727 | p_rq = task_rq(p); | |
7b270f60 PZ |
8728 | /* |
8729 | * If we're the only runnable task on the rq and target rq also | |
8730 | * has only one task, there's absolutely no point in yielding. | |
8731 | */ | |
8732 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { | |
8733 | yielded = -ESRCH; | |
8734 | goto out_irq; | |
8735 | } | |
8736 | ||
d95f4122 | 8737 | double_rq_lock(rq, p_rq); |
39e24d8f | 8738 | if (task_rq(p) != p_rq) { |
d95f4122 MG |
8739 | double_rq_unlock(rq, p_rq); |
8740 | goto again; | |
8741 | } | |
8742 | ||
8743 | if (!curr->sched_class->yield_to_task) | |
7b270f60 | 8744 | goto out_unlock; |
d95f4122 MG |
8745 | |
8746 | if (curr->sched_class != p->sched_class) | |
7b270f60 | 8747 | goto out_unlock; |
d95f4122 | 8748 | |
0b9d46fc | 8749 | if (task_on_cpu(p_rq, p) || !task_is_running(p)) |
7b270f60 | 8750 | goto out_unlock; |
d95f4122 | 8751 | |
0900acf2 | 8752 | yielded = curr->sched_class->yield_to_task(rq, p); |
6d1cafd8 | 8753 | if (yielded) { |
ae92882e | 8754 | schedstat_inc(rq->yld_count); |
6d1cafd8 VP |
8755 | /* |
8756 | * Make p's CPU reschedule; pick_next_entity takes care of | |
8757 | * fairness. | |
8758 | */ | |
8759 | if (preempt && rq != p_rq) | |
8875125e | 8760 | resched_curr(p_rq); |
6d1cafd8 | 8761 | } |
d95f4122 | 8762 | |
7b270f60 | 8763 | out_unlock: |
d95f4122 | 8764 | double_rq_unlock(rq, p_rq); |
7b270f60 | 8765 | out_irq: |
d95f4122 MG |
8766 | local_irq_restore(flags); |
8767 | ||
7b270f60 | 8768 | if (yielded > 0) |
d95f4122 MG |
8769 | schedule(); |
8770 | ||
8771 | return yielded; | |
8772 | } | |
8773 | EXPORT_SYMBOL_GPL(yield_to); | |
8774 | ||
10ab5643 TH |
8775 | int io_schedule_prepare(void) |
8776 | { | |
8777 | int old_iowait = current->in_iowait; | |
8778 | ||
8779 | current->in_iowait = 1; | |
aa8dccca | 8780 | blk_flush_plug(current->plug, true); |
10ab5643 TH |
8781 | return old_iowait; |
8782 | } | |
8783 | ||
8784 | void io_schedule_finish(int token) | |
8785 | { | |
8786 | current->in_iowait = token; | |
8787 | } | |
8788 | ||
1da177e4 | 8789 | /* |
41a2d6cf | 8790 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 8791 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 | 8792 | */ |
1da177e4 LT |
8793 | long __sched io_schedule_timeout(long timeout) |
8794 | { | |
10ab5643 | 8795 | int token; |
1da177e4 LT |
8796 | long ret; |
8797 | ||
10ab5643 | 8798 | token = io_schedule_prepare(); |
1da177e4 | 8799 | ret = schedule_timeout(timeout); |
10ab5643 | 8800 | io_schedule_finish(token); |
9cff8ade | 8801 | |
1da177e4 LT |
8802 | return ret; |
8803 | } | |
9cff8ade | 8804 | EXPORT_SYMBOL(io_schedule_timeout); |
1da177e4 | 8805 | |
e3b929b0 | 8806 | void __sched io_schedule(void) |
10ab5643 TH |
8807 | { |
8808 | int token; | |
8809 | ||
8810 | token = io_schedule_prepare(); | |
8811 | schedule(); | |
8812 | io_schedule_finish(token); | |
8813 | } | |
8814 | EXPORT_SYMBOL(io_schedule); | |
8815 | ||
1da177e4 LT |
8816 | /** |
8817 | * sys_sched_get_priority_max - return maximum RT priority. | |
8818 | * @policy: scheduling class. | |
8819 | * | |
e69f6186 YB |
8820 | * Return: On success, this syscall returns the maximum |
8821 | * rt_priority that can be used by a given scheduling class. | |
8822 | * On failure, a negative error code is returned. | |
1da177e4 | 8823 | */ |
5add95d4 | 8824 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
8825 | { |
8826 | int ret = -EINVAL; | |
8827 | ||
8828 | switch (policy) { | |
8829 | case SCHED_FIFO: | |
8830 | case SCHED_RR: | |
ae18ad28 | 8831 | ret = MAX_RT_PRIO-1; |
1da177e4 | 8832 | break; |
aab03e05 | 8833 | case SCHED_DEADLINE: |
1da177e4 | 8834 | case SCHED_NORMAL: |
b0a9499c | 8835 | case SCHED_BATCH: |
dd41f596 | 8836 | case SCHED_IDLE: |
1da177e4 LT |
8837 | ret = 0; |
8838 | break; | |
8839 | } | |
8840 | return ret; | |
8841 | } | |
8842 | ||
8843 | /** | |
8844 | * sys_sched_get_priority_min - return minimum RT priority. | |
8845 | * @policy: scheduling class. | |
8846 | * | |
e69f6186 YB |
8847 | * Return: On success, this syscall returns the minimum |
8848 | * rt_priority that can be used by a given scheduling class. | |
8849 | * On failure, a negative error code is returned. | |
1da177e4 | 8850 | */ |
5add95d4 | 8851 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
8852 | { |
8853 | int ret = -EINVAL; | |
8854 | ||
8855 | switch (policy) { | |
8856 | case SCHED_FIFO: | |
8857 | case SCHED_RR: | |
8858 | ret = 1; | |
8859 | break; | |
aab03e05 | 8860 | case SCHED_DEADLINE: |
1da177e4 | 8861 | case SCHED_NORMAL: |
b0a9499c | 8862 | case SCHED_BATCH: |
dd41f596 | 8863 | case SCHED_IDLE: |
1da177e4 LT |
8864 | ret = 0; |
8865 | } | |
8866 | return ret; | |
8867 | } | |
8868 | ||
abca5fc5 | 8869 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) |
1da177e4 | 8870 | { |
36c8b586 | 8871 | struct task_struct *p; |
a4ec24b4 | 8872 | unsigned int time_slice; |
eb580751 | 8873 | struct rq_flags rf; |
dba091b9 | 8874 | struct rq *rq; |
3a5c359a | 8875 | int retval; |
1da177e4 LT |
8876 | |
8877 | if (pid < 0) | |
3a5c359a | 8878 | return -EINVAL; |
1da177e4 LT |
8879 | |
8880 | retval = -ESRCH; | |
1a551ae7 | 8881 | rcu_read_lock(); |
1da177e4 LT |
8882 | p = find_process_by_pid(pid); |
8883 | if (!p) | |
8884 | goto out_unlock; | |
8885 | ||
8886 | retval = security_task_getscheduler(p); | |
8887 | if (retval) | |
8888 | goto out_unlock; | |
8889 | ||
eb580751 | 8890 | rq = task_rq_lock(p, &rf); |
a57beec5 PZ |
8891 | time_slice = 0; |
8892 | if (p->sched_class->get_rr_interval) | |
8893 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
eb580751 | 8894 | task_rq_unlock(rq, p, &rf); |
a4ec24b4 | 8895 | |
1a551ae7 | 8896 | rcu_read_unlock(); |
abca5fc5 AV |
8897 | jiffies_to_timespec64(time_slice, t); |
8898 | return 0; | |
3a5c359a | 8899 | |
1da177e4 | 8900 | out_unlock: |
1a551ae7 | 8901 | rcu_read_unlock(); |
1da177e4 LT |
8902 | return retval; |
8903 | } | |
8904 | ||
2064a5ab RD |
8905 | /** |
8906 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
8907 | * @pid: pid of the process. | |
8908 | * @interval: userspace pointer to the timeslice value. | |
8909 | * | |
8910 | * this syscall writes the default timeslice value of a given process | |
8911 | * into the user-space timespec buffer. A value of '0' means infinity. | |
8912 | * | |
8913 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, | |
8914 | * an error code. | |
8915 | */ | |
abca5fc5 | 8916 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
474b9c77 | 8917 | struct __kernel_timespec __user *, interval) |
abca5fc5 AV |
8918 | { |
8919 | struct timespec64 t; | |
8920 | int retval = sched_rr_get_interval(pid, &t); | |
8921 | ||
8922 | if (retval == 0) | |
8923 | retval = put_timespec64(&t, interval); | |
8924 | ||
8925 | return retval; | |
8926 | } | |
8927 | ||
474b9c77 | 8928 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 AB |
8929 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, |
8930 | struct old_timespec32 __user *, interval) | |
abca5fc5 AV |
8931 | { |
8932 | struct timespec64 t; | |
8933 | int retval = sched_rr_get_interval(pid, &t); | |
8934 | ||
8935 | if (retval == 0) | |
9afc5eee | 8936 | retval = put_old_timespec32(&t, interval); |
abca5fc5 AV |
8937 | return retval; |
8938 | } | |
8939 | #endif | |
8940 | ||
82a1fcb9 | 8941 | void sched_show_task(struct task_struct *p) |
1da177e4 | 8942 | { |
1da177e4 | 8943 | unsigned long free = 0; |
4e79752c | 8944 | int ppid; |
c930b2c0 | 8945 | |
38200502 TH |
8946 | if (!try_get_task_stack(p)) |
8947 | return; | |
20435d84 | 8948 | |
cc172ff3 | 8949 | pr_info("task:%-15.15s state:%c", p->comm, task_state_to_char(p)); |
20435d84 | 8950 | |
b03fbd4f | 8951 | if (task_is_running(p)) |
cc172ff3 | 8952 | pr_cont(" running task "); |
1da177e4 | 8953 | #ifdef CONFIG_DEBUG_STACK_USAGE |
7c9f8861 | 8954 | free = stack_not_used(p); |
1da177e4 | 8955 | #endif |
a90e984c | 8956 | ppid = 0; |
4e79752c | 8957 | rcu_read_lock(); |
a90e984c ON |
8958 | if (pid_alive(p)) |
8959 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); | |
4e79752c | 8960 | rcu_read_unlock(); |
0f03d680 | 8961 | pr_cont(" stack:%-5lu pid:%-5d ppid:%-6d flags:0x%08lx\n", |
cc172ff3 | 8962 | free, task_pid_nr(p), ppid, |
0569b245 | 8963 | read_task_thread_flags(p)); |
1da177e4 | 8964 | |
3d1cb205 | 8965 | print_worker_info(KERN_INFO, p); |
a8b62fd0 | 8966 | print_stop_info(KERN_INFO, p); |
9cb8f069 | 8967 | show_stack(p, NULL, KERN_INFO); |
38200502 | 8968 | put_task_stack(p); |
1da177e4 | 8969 | } |
0032f4e8 | 8970 | EXPORT_SYMBOL_GPL(sched_show_task); |
1da177e4 | 8971 | |
5d68cc95 PZ |
8972 | static inline bool |
8973 | state_filter_match(unsigned long state_filter, struct task_struct *p) | |
8974 | { | |
2f064a59 PZ |
8975 | unsigned int state = READ_ONCE(p->__state); |
8976 | ||
5d68cc95 PZ |
8977 | /* no filter, everything matches */ |
8978 | if (!state_filter) | |
8979 | return true; | |
8980 | ||
8981 | /* filter, but doesn't match */ | |
2f064a59 | 8982 | if (!(state & state_filter)) |
5d68cc95 PZ |
8983 | return false; |
8984 | ||
8985 | /* | |
8986 | * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows | |
8987 | * TASK_KILLABLE). | |
8988 | */ | |
5aec788a | 8989 | if (state_filter == TASK_UNINTERRUPTIBLE && (state & TASK_NOLOAD)) |
5d68cc95 PZ |
8990 | return false; |
8991 | ||
8992 | return true; | |
8993 | } | |
8994 | ||
8995 | ||
2f064a59 | 8996 | void show_state_filter(unsigned int state_filter) |
1da177e4 | 8997 | { |
36c8b586 | 8998 | struct task_struct *g, *p; |
1da177e4 | 8999 | |
510f5acc | 9000 | rcu_read_lock(); |
5d07f420 | 9001 | for_each_process_thread(g, p) { |
1da177e4 LT |
9002 | /* |
9003 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 9004 | * console might take a lot of time: |
57675cb9 AR |
9005 | * Also, reset softlockup watchdogs on all CPUs, because |
9006 | * another CPU might be blocked waiting for us to process | |
9007 | * an IPI. | |
1da177e4 LT |
9008 | */ |
9009 | touch_nmi_watchdog(); | |
57675cb9 | 9010 | touch_all_softlockup_watchdogs(); |
5d68cc95 | 9011 | if (state_filter_match(state_filter, p)) |
82a1fcb9 | 9012 | sched_show_task(p); |
5d07f420 | 9013 | } |
1da177e4 | 9014 | |
dd41f596 | 9015 | #ifdef CONFIG_SCHED_DEBUG |
fb90a6e9 RV |
9016 | if (!state_filter) |
9017 | sysrq_sched_debug_show(); | |
dd41f596 | 9018 | #endif |
510f5acc | 9019 | rcu_read_unlock(); |
e59e2ae2 IM |
9020 | /* |
9021 | * Only show locks if all tasks are dumped: | |
9022 | */ | |
93335a21 | 9023 | if (!state_filter) |
e59e2ae2 | 9024 | debug_show_all_locks(); |
1da177e4 LT |
9025 | } |
9026 | ||
f340c0d1 IM |
9027 | /** |
9028 | * init_idle - set up an idle thread for a given CPU | |
9029 | * @idle: task in question | |
d1ccc66d | 9030 | * @cpu: CPU the idle task belongs to |
f340c0d1 IM |
9031 | * |
9032 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
9033 | * flag, to make booting more robust. | |
9034 | */ | |
f1a0a376 | 9035 | void __init init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 9036 | { |
713a2e21 WL |
9037 | #ifdef CONFIG_SMP |
9038 | struct affinity_context ac = (struct affinity_context) { | |
9039 | .new_mask = cpumask_of(cpu), | |
9040 | .flags = 0, | |
9041 | }; | |
9042 | #endif | |
70b97a7f | 9043 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
9044 | unsigned long flags; |
9045 | ||
ff51ff84 PZ |
9046 | __sched_fork(0, idle); |
9047 | ||
25834c73 | 9048 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
5cb9eaa3 | 9049 | raw_spin_rq_lock(rq); |
5cbd54ef | 9050 | |
2f064a59 | 9051 | idle->__state = TASK_RUNNING; |
dd41f596 | 9052 | idle->se.exec_start = sched_clock(); |
00b89fe0 VS |
9053 | /* |
9054 | * PF_KTHREAD should already be set at this point; regardless, make it | |
9055 | * look like a proper per-CPU kthread. | |
9056 | */ | |
9057 | idle->flags |= PF_IDLE | PF_KTHREAD | PF_NO_SETAFFINITY; | |
9058 | kthread_set_per_cpu(idle, cpu); | |
dd41f596 | 9059 | |
de9b8f5d PZ |
9060 | #ifdef CONFIG_SMP |
9061 | /* | |
b19a888c | 9062 | * It's possible that init_idle() gets called multiple times on a task, |
de9b8f5d PZ |
9063 | * in that case do_set_cpus_allowed() will not do the right thing. |
9064 | * | |
9065 | * And since this is boot we can forgo the serialization. | |
9066 | */ | |
713a2e21 | 9067 | set_cpus_allowed_common(idle, &ac); |
de9b8f5d | 9068 | #endif |
6506cf6c PZ |
9069 | /* |
9070 | * We're having a chicken and egg problem, even though we are | |
d1ccc66d | 9071 | * holding rq->lock, the CPU isn't yet set to this CPU so the |
6506cf6c PZ |
9072 | * lockdep check in task_group() will fail. |
9073 | * | |
9074 | * Similar case to sched_fork(). / Alternatively we could | |
9075 | * use task_rq_lock() here and obtain the other rq->lock. | |
9076 | * | |
9077 | * Silence PROVE_RCU | |
9078 | */ | |
9079 | rcu_read_lock(); | |
dd41f596 | 9080 | __set_task_cpu(idle, cpu); |
6506cf6c | 9081 | rcu_read_unlock(); |
1da177e4 | 9082 | |
5311a98f EB |
9083 | rq->idle = idle; |
9084 | rcu_assign_pointer(rq->curr, idle); | |
da0c1e65 | 9085 | idle->on_rq = TASK_ON_RQ_QUEUED; |
de9b8f5d | 9086 | #ifdef CONFIG_SMP |
3ca7a440 | 9087 | idle->on_cpu = 1; |
4866cde0 | 9088 | #endif |
5cb9eaa3 | 9089 | raw_spin_rq_unlock(rq); |
25834c73 | 9090 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); |
1da177e4 LT |
9091 | |
9092 | /* Set the preempt count _outside_ the spinlocks! */ | |
01028747 | 9093 | init_idle_preempt_count(idle, cpu); |
55cd5340 | 9094 | |
dd41f596 IM |
9095 | /* |
9096 | * The idle tasks have their own, simple scheduling class: | |
9097 | */ | |
9098 | idle->sched_class = &idle_sched_class; | |
868baf07 | 9099 | ftrace_graph_init_idle_task(idle, cpu); |
45eacc69 | 9100 | vtime_init_idle(idle, cpu); |
de9b8f5d | 9101 | #ifdef CONFIG_SMP |
f1c6f1a7 CE |
9102 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
9103 | #endif | |
19978ca6 IM |
9104 | } |
9105 | ||
e1d4eeec NP |
9106 | #ifdef CONFIG_SMP |
9107 | ||
f82f8042 JL |
9108 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
9109 | const struct cpumask *trial) | |
9110 | { | |
06a76fe0 | 9111 | int ret = 1; |
f82f8042 | 9112 | |
1087ad4e | 9113 | if (cpumask_empty(cur)) |
bb2bc55a MG |
9114 | return ret; |
9115 | ||
06a76fe0 | 9116 | ret = dl_cpuset_cpumask_can_shrink(cur, trial); |
f82f8042 JL |
9117 | |
9118 | return ret; | |
9119 | } | |
9120 | ||
7f51412a | 9121 | int task_can_attach(struct task_struct *p, |
b6e8d40d | 9122 | const struct cpumask *cs_effective_cpus) |
7f51412a JL |
9123 | { |
9124 | int ret = 0; | |
9125 | ||
9126 | /* | |
9127 | * Kthreads which disallow setaffinity shouldn't be moved | |
d1ccc66d | 9128 | * to a new cpuset; we don't want to change their CPU |
7f51412a JL |
9129 | * affinity and isolating such threads by their set of |
9130 | * allowed nodes is unnecessary. Thus, cpusets are not | |
9131 | * applicable for such threads. This prevents checking for | |
9132 | * success of set_cpus_allowed_ptr() on all attached tasks | |
3bd37062 | 9133 | * before cpus_mask may be changed. |
7f51412a JL |
9134 | */ |
9135 | if (p->flags & PF_NO_SETAFFINITY) { | |
9136 | ret = -EINVAL; | |
9137 | goto out; | |
9138 | } | |
9139 | ||
7f51412a | 9140 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
b6e8d40d WL |
9141 | cs_effective_cpus)) { |
9142 | int cpu = cpumask_any_and(cpu_active_mask, cs_effective_cpus); | |
772b6539 | 9143 | |
b6e8d40d WL |
9144 | if (unlikely(cpu >= nr_cpu_ids)) |
9145 | return -EINVAL; | |
772b6539 DE |
9146 | ret = dl_cpu_busy(cpu, p); |
9147 | } | |
7f51412a | 9148 | |
7f51412a JL |
9149 | out: |
9150 | return ret; | |
9151 | } | |
9152 | ||
f2cb1360 | 9153 | bool sched_smp_initialized __read_mostly; |
e26fbffd | 9154 | |
e6628d5b MG |
9155 | #ifdef CONFIG_NUMA_BALANCING |
9156 | /* Migrate current task p to target_cpu */ | |
9157 | int migrate_task_to(struct task_struct *p, int target_cpu) | |
9158 | { | |
9159 | struct migration_arg arg = { p, target_cpu }; | |
9160 | int curr_cpu = task_cpu(p); | |
9161 | ||
9162 | if (curr_cpu == target_cpu) | |
9163 | return 0; | |
9164 | ||
3bd37062 | 9165 | if (!cpumask_test_cpu(target_cpu, p->cpus_ptr)) |
e6628d5b MG |
9166 | return -EINVAL; |
9167 | ||
9168 | /* TODO: This is not properly updating schedstats */ | |
9169 | ||
286549dc | 9170 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
e6628d5b MG |
9171 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
9172 | } | |
0ec8aa00 PZ |
9173 | |
9174 | /* | |
9175 | * Requeue a task on a given node and accurately track the number of NUMA | |
9176 | * tasks on the runqueues | |
9177 | */ | |
9178 | void sched_setnuma(struct task_struct *p, int nid) | |
9179 | { | |
da0c1e65 | 9180 | bool queued, running; |
eb580751 PZ |
9181 | struct rq_flags rf; |
9182 | struct rq *rq; | |
0ec8aa00 | 9183 | |
eb580751 | 9184 | rq = task_rq_lock(p, &rf); |
da0c1e65 | 9185 | queued = task_on_rq_queued(p); |
0ec8aa00 PZ |
9186 | running = task_current(rq, p); |
9187 | ||
da0c1e65 | 9188 | if (queued) |
1de64443 | 9189 | dequeue_task(rq, p, DEQUEUE_SAVE); |
0ec8aa00 | 9190 | if (running) |
f3cd1c4e | 9191 | put_prev_task(rq, p); |
0ec8aa00 PZ |
9192 | |
9193 | p->numa_preferred_nid = nid; | |
0ec8aa00 | 9194 | |
da0c1e65 | 9195 | if (queued) |
7134b3e9 | 9196 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); |
a399d233 | 9197 | if (running) |
03b7fad1 | 9198 | set_next_task(rq, p); |
eb580751 | 9199 | task_rq_unlock(rq, p, &rf); |
0ec8aa00 | 9200 | } |
5cc389bc | 9201 | #endif /* CONFIG_NUMA_BALANCING */ |
f7b4cddc | 9202 | |
1da177e4 | 9203 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 9204 | /* |
d1ccc66d | 9205 | * Ensure that the idle task is using init_mm right before its CPU goes |
48c5ccae | 9206 | * offline. |
054b9108 | 9207 | */ |
48c5ccae | 9208 | void idle_task_exit(void) |
1da177e4 | 9209 | { |
48c5ccae | 9210 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 9211 | |
48c5ccae | 9212 | BUG_ON(cpu_online(smp_processor_id())); |
bf2c59fc | 9213 | BUG_ON(current != this_rq()->idle); |
e76bd8d9 | 9214 | |
a53efe5f | 9215 | if (mm != &init_mm) { |
252d2a41 | 9216 | switch_mm(mm, &init_mm, current); |
a53efe5f MS |
9217 | finish_arch_post_lock_switch(); |
9218 | } | |
bf2c59fc PZ |
9219 | |
9220 | /* finish_cpu(), as ran on the BP, will clean up the active_mm state */ | |
1da177e4 LT |
9221 | } |
9222 | ||
2558aacf | 9223 | static int __balance_push_cpu_stop(void *arg) |
1da177e4 | 9224 | { |
2558aacf PZ |
9225 | struct task_struct *p = arg; |
9226 | struct rq *rq = this_rq(); | |
9227 | struct rq_flags rf; | |
9228 | int cpu; | |
1da177e4 | 9229 | |
2558aacf PZ |
9230 | raw_spin_lock_irq(&p->pi_lock); |
9231 | rq_lock(rq, &rf); | |
3f1d2a31 | 9232 | |
2558aacf PZ |
9233 | update_rq_clock(rq); |
9234 | ||
9235 | if (task_rq(p) == rq && task_on_rq_queued(p)) { | |
9236 | cpu = select_fallback_rq(rq->cpu, p); | |
9237 | rq = __migrate_task(rq, &rf, p, cpu); | |
10e7071b | 9238 | } |
3f1d2a31 | 9239 | |
2558aacf PZ |
9240 | rq_unlock(rq, &rf); |
9241 | raw_spin_unlock_irq(&p->pi_lock); | |
9242 | ||
9243 | put_task_struct(p); | |
9244 | ||
9245 | return 0; | |
10e7071b | 9246 | } |
3f1d2a31 | 9247 | |
2558aacf PZ |
9248 | static DEFINE_PER_CPU(struct cpu_stop_work, push_work); |
9249 | ||
48f24c4d | 9250 | /* |
2558aacf | 9251 | * Ensure we only run per-cpu kthreads once the CPU goes !active. |
b5c44773 PZ |
9252 | * |
9253 | * This is enabled below SCHED_AP_ACTIVE; when !cpu_active(), but only | |
9254 | * effective when the hotplug motion is down. | |
1da177e4 | 9255 | */ |
2558aacf | 9256 | static void balance_push(struct rq *rq) |
1da177e4 | 9257 | { |
2558aacf PZ |
9258 | struct task_struct *push_task = rq->curr; |
9259 | ||
5cb9eaa3 | 9260 | lockdep_assert_rq_held(rq); |
b5c44773 | 9261 | |
ae792702 PZ |
9262 | /* |
9263 | * Ensure the thing is persistent until balance_push_set(.on = false); | |
9264 | */ | |
9265 | rq->balance_callback = &balance_push_callback; | |
1da177e4 | 9266 | |
b5c44773 | 9267 | /* |
868ad33b TG |
9268 | * Only active while going offline and when invoked on the outgoing |
9269 | * CPU. | |
b5c44773 | 9270 | */ |
868ad33b | 9271 | if (!cpu_dying(rq->cpu) || rq != this_rq()) |
b5c44773 PZ |
9272 | return; |
9273 | ||
1da177e4 | 9274 | /* |
2558aacf PZ |
9275 | * Both the cpu-hotplug and stop task are in this case and are |
9276 | * required to complete the hotplug process. | |
1da177e4 | 9277 | */ |
00b89fe0 | 9278 | if (kthread_is_per_cpu(push_task) || |
5ba2ffba PZ |
9279 | is_migration_disabled(push_task)) { |
9280 | ||
f2469a1f TG |
9281 | /* |
9282 | * If this is the idle task on the outgoing CPU try to wake | |
9283 | * up the hotplug control thread which might wait for the | |
9284 | * last task to vanish. The rcuwait_active() check is | |
9285 | * accurate here because the waiter is pinned on this CPU | |
9286 | * and can't obviously be running in parallel. | |
3015ef4b TG |
9287 | * |
9288 | * On RT kernels this also has to check whether there are | |
9289 | * pinned and scheduled out tasks on the runqueue. They | |
9290 | * need to leave the migrate disabled section first. | |
f2469a1f | 9291 | */ |
3015ef4b TG |
9292 | if (!rq->nr_running && !rq_has_pinned_tasks(rq) && |
9293 | rcuwait_active(&rq->hotplug_wait)) { | |
5cb9eaa3 | 9294 | raw_spin_rq_unlock(rq); |
f2469a1f | 9295 | rcuwait_wake_up(&rq->hotplug_wait); |
5cb9eaa3 | 9296 | raw_spin_rq_lock(rq); |
f2469a1f | 9297 | } |
2558aacf | 9298 | return; |
f2469a1f | 9299 | } |
48f24c4d | 9300 | |
2558aacf | 9301 | get_task_struct(push_task); |
77bd3970 | 9302 | /* |
2558aacf PZ |
9303 | * Temporarily drop rq->lock such that we can wake-up the stop task. |
9304 | * Both preemption and IRQs are still disabled. | |
77bd3970 | 9305 | */ |
5cb9eaa3 | 9306 | raw_spin_rq_unlock(rq); |
2558aacf PZ |
9307 | stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task, |
9308 | this_cpu_ptr(&push_work)); | |
9309 | /* | |
9310 | * At this point need_resched() is true and we'll take the loop in | |
9311 | * schedule(). The next pick is obviously going to be the stop task | |
5ba2ffba | 9312 | * which kthread_is_per_cpu() and will push this task away. |
2558aacf | 9313 | */ |
5cb9eaa3 | 9314 | raw_spin_rq_lock(rq); |
2558aacf | 9315 | } |
77bd3970 | 9316 | |
2558aacf PZ |
9317 | static void balance_push_set(int cpu, bool on) |
9318 | { | |
9319 | struct rq *rq = cpu_rq(cpu); | |
9320 | struct rq_flags rf; | |
48c5ccae | 9321 | |
2558aacf | 9322 | rq_lock_irqsave(rq, &rf); |
22f667c9 PZ |
9323 | if (on) { |
9324 | WARN_ON_ONCE(rq->balance_callback); | |
ae792702 | 9325 | rq->balance_callback = &balance_push_callback; |
22f667c9 | 9326 | } else if (rq->balance_callback == &balance_push_callback) { |
ae792702 | 9327 | rq->balance_callback = NULL; |
22f667c9 | 9328 | } |
2558aacf PZ |
9329 | rq_unlock_irqrestore(rq, &rf); |
9330 | } | |
e692ab53 | 9331 | |
f2469a1f TG |
9332 | /* |
9333 | * Invoked from a CPUs hotplug control thread after the CPU has been marked | |
9334 | * inactive. All tasks which are not per CPU kernel threads are either | |
9335 | * pushed off this CPU now via balance_push() or placed on a different CPU | |
9336 | * during wakeup. Wait until the CPU is quiescent. | |
9337 | */ | |
9338 | static void balance_hotplug_wait(void) | |
9339 | { | |
9340 | struct rq *rq = this_rq(); | |
5473e0cc | 9341 | |
3015ef4b TG |
9342 | rcuwait_wait_event(&rq->hotplug_wait, |
9343 | rq->nr_running == 1 && !rq_has_pinned_tasks(rq), | |
f2469a1f TG |
9344 | TASK_UNINTERRUPTIBLE); |
9345 | } | |
5473e0cc | 9346 | |
2558aacf | 9347 | #else |
dce48a84 | 9348 | |
2558aacf PZ |
9349 | static inline void balance_push(struct rq *rq) |
9350 | { | |
dce48a84 | 9351 | } |
dce48a84 | 9352 | |
2558aacf PZ |
9353 | static inline void balance_push_set(int cpu, bool on) |
9354 | { | |
9355 | } | |
9356 | ||
f2469a1f TG |
9357 | static inline void balance_hotplug_wait(void) |
9358 | { | |
dce48a84 | 9359 | } |
f2469a1f | 9360 | |
1da177e4 LT |
9361 | #endif /* CONFIG_HOTPLUG_CPU */ |
9362 | ||
f2cb1360 | 9363 | void set_rq_online(struct rq *rq) |
1f11eb6a GH |
9364 | { |
9365 | if (!rq->online) { | |
9366 | const struct sched_class *class; | |
9367 | ||
c6c4927b | 9368 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
9369 | rq->online = 1; |
9370 | ||
9371 | for_each_class(class) { | |
9372 | if (class->rq_online) | |
9373 | class->rq_online(rq); | |
9374 | } | |
9375 | } | |
9376 | } | |
9377 | ||
f2cb1360 | 9378 | void set_rq_offline(struct rq *rq) |
1f11eb6a GH |
9379 | { |
9380 | if (rq->online) { | |
9381 | const struct sched_class *class; | |
9382 | ||
9383 | for_each_class(class) { | |
9384 | if (class->rq_offline) | |
9385 | class->rq_offline(rq); | |
9386 | } | |
9387 | ||
c6c4927b | 9388 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
9389 | rq->online = 0; |
9390 | } | |
9391 | } | |
9392 | ||
d1ccc66d IM |
9393 | /* |
9394 | * used to mark begin/end of suspend/resume: | |
9395 | */ | |
9396 | static int num_cpus_frozen; | |
d35be8ba | 9397 | |
1da177e4 | 9398 | /* |
3a101d05 TH |
9399 | * Update cpusets according to cpu_active mask. If cpusets are |
9400 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
9401 | * around partition_sched_domains(). | |
d35be8ba SB |
9402 | * |
9403 | * If we come here as part of a suspend/resume, don't touch cpusets because we | |
9404 | * want to restore it back to its original state upon resume anyway. | |
1da177e4 | 9405 | */ |
40190a78 | 9406 | static void cpuset_cpu_active(void) |
e761b772 | 9407 | { |
40190a78 | 9408 | if (cpuhp_tasks_frozen) { |
d35be8ba SB |
9409 | /* |
9410 | * num_cpus_frozen tracks how many CPUs are involved in suspend | |
9411 | * resume sequence. As long as this is not the last online | |
9412 | * operation in the resume sequence, just build a single sched | |
9413 | * domain, ignoring cpusets. | |
9414 | */ | |
50e76632 PZ |
9415 | partition_sched_domains(1, NULL, NULL); |
9416 | if (--num_cpus_frozen) | |
135fb3e1 | 9417 | return; |
d35be8ba SB |
9418 | /* |
9419 | * This is the last CPU online operation. So fall through and | |
9420 | * restore the original sched domains by considering the | |
9421 | * cpuset configurations. | |
9422 | */ | |
50e76632 | 9423 | cpuset_force_rebuild(); |
3a101d05 | 9424 | } |
30e03acd | 9425 | cpuset_update_active_cpus(); |
3a101d05 | 9426 | } |
e761b772 | 9427 | |
40190a78 | 9428 | static int cpuset_cpu_inactive(unsigned int cpu) |
3a101d05 | 9429 | { |
40190a78 | 9430 | if (!cpuhp_tasks_frozen) { |
772b6539 DE |
9431 | int ret = dl_cpu_busy(cpu, NULL); |
9432 | ||
9433 | if (ret) | |
9434 | return ret; | |
30e03acd | 9435 | cpuset_update_active_cpus(); |
135fb3e1 | 9436 | } else { |
d35be8ba SB |
9437 | num_cpus_frozen++; |
9438 | partition_sched_domains(1, NULL, NULL); | |
e761b772 | 9439 | } |
135fb3e1 | 9440 | return 0; |
e761b772 | 9441 | } |
e761b772 | 9442 | |
40190a78 | 9443 | int sched_cpu_activate(unsigned int cpu) |
135fb3e1 | 9444 | { |
7d976699 | 9445 | struct rq *rq = cpu_rq(cpu); |
8a8c69c3 | 9446 | struct rq_flags rf; |
7d976699 | 9447 | |
22f667c9 | 9448 | /* |
b5c44773 PZ |
9449 | * Clear the balance_push callback and prepare to schedule |
9450 | * regular tasks. | |
22f667c9 | 9451 | */ |
2558aacf PZ |
9452 | balance_push_set(cpu, false); |
9453 | ||
ba2591a5 PZ |
9454 | #ifdef CONFIG_SCHED_SMT |
9455 | /* | |
c5511d03 | 9456 | * When going up, increment the number of cores with SMT present. |
ba2591a5 | 9457 | */ |
c5511d03 PZI |
9458 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) |
9459 | static_branch_inc_cpuslocked(&sched_smt_present); | |
ba2591a5 | 9460 | #endif |
40190a78 | 9461 | set_cpu_active(cpu, true); |
135fb3e1 | 9462 | |
40190a78 | 9463 | if (sched_smp_initialized) { |
0fb3978b | 9464 | sched_update_numa(cpu, true); |
135fb3e1 | 9465 | sched_domains_numa_masks_set(cpu); |
40190a78 | 9466 | cpuset_cpu_active(); |
e761b772 | 9467 | } |
7d976699 TG |
9468 | |
9469 | /* | |
9470 | * Put the rq online, if not already. This happens: | |
9471 | * | |
9472 | * 1) In the early boot process, because we build the real domains | |
d1ccc66d | 9473 | * after all CPUs have been brought up. |
7d976699 TG |
9474 | * |
9475 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the | |
9476 | * domains. | |
9477 | */ | |
8a8c69c3 | 9478 | rq_lock_irqsave(rq, &rf); |
7d976699 TG |
9479 | if (rq->rd) { |
9480 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9481 | set_rq_online(rq); | |
9482 | } | |
8a8c69c3 | 9483 | rq_unlock_irqrestore(rq, &rf); |
7d976699 | 9484 | |
40190a78 | 9485 | return 0; |
135fb3e1 TG |
9486 | } |
9487 | ||
40190a78 | 9488 | int sched_cpu_deactivate(unsigned int cpu) |
135fb3e1 | 9489 | { |
120455c5 PZ |
9490 | struct rq *rq = cpu_rq(cpu); |
9491 | struct rq_flags rf; | |
135fb3e1 TG |
9492 | int ret; |
9493 | ||
e0b257c3 AMB |
9494 | /* |
9495 | * Remove CPU from nohz.idle_cpus_mask to prevent participating in | |
9496 | * load balancing when not active | |
9497 | */ | |
9498 | nohz_balance_exit_idle(rq); | |
9499 | ||
40190a78 | 9500 | set_cpu_active(cpu, false); |
741ba80f PZ |
9501 | |
9502 | /* | |
9503 | * From this point forward, this CPU will refuse to run any task that | |
9504 | * is not: migrate_disable() or KTHREAD_IS_PER_CPU, and will actively | |
9505 | * push those tasks away until this gets cleared, see | |
9506 | * sched_cpu_dying(). | |
9507 | */ | |
975707f2 PZ |
9508 | balance_push_set(cpu, true); |
9509 | ||
b2454caa | 9510 | /* |
975707f2 PZ |
9511 | * We've cleared cpu_active_mask / set balance_push, wait for all |
9512 | * preempt-disabled and RCU users of this state to go away such that | |
9513 | * all new such users will observe it. | |
b2454caa | 9514 | * |
5ba2ffba PZ |
9515 | * Specifically, we rely on ttwu to no longer target this CPU, see |
9516 | * ttwu_queue_cond() and is_cpu_allowed(). | |
9517 | * | |
b2454caa PZ |
9518 | * Do sync before park smpboot threads to take care the rcu boost case. |
9519 | */ | |
309ba859 | 9520 | synchronize_rcu(); |
40190a78 | 9521 | |
120455c5 PZ |
9522 | rq_lock_irqsave(rq, &rf); |
9523 | if (rq->rd) { | |
9524 | update_rq_clock(rq); | |
9525 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); | |
9526 | set_rq_offline(rq); | |
9527 | } | |
9528 | rq_unlock_irqrestore(rq, &rf); | |
9529 | ||
c5511d03 PZI |
9530 | #ifdef CONFIG_SCHED_SMT |
9531 | /* | |
9532 | * When going down, decrement the number of cores with SMT present. | |
9533 | */ | |
9534 | if (cpumask_weight(cpu_smt_mask(cpu)) == 2) | |
9535 | static_branch_dec_cpuslocked(&sched_smt_present); | |
3c474b32 PZ |
9536 | |
9537 | sched_core_cpu_deactivate(cpu); | |
c5511d03 PZI |
9538 | #endif |
9539 | ||
40190a78 TG |
9540 | if (!sched_smp_initialized) |
9541 | return 0; | |
9542 | ||
0fb3978b | 9543 | sched_update_numa(cpu, false); |
40190a78 TG |
9544 | ret = cpuset_cpu_inactive(cpu); |
9545 | if (ret) { | |
2558aacf | 9546 | balance_push_set(cpu, false); |
40190a78 | 9547 | set_cpu_active(cpu, true); |
0fb3978b | 9548 | sched_update_numa(cpu, true); |
40190a78 | 9549 | return ret; |
135fb3e1 | 9550 | } |
40190a78 TG |
9551 | sched_domains_numa_masks_clear(cpu); |
9552 | return 0; | |
135fb3e1 TG |
9553 | } |
9554 | ||
94baf7a5 TG |
9555 | static void sched_rq_cpu_starting(unsigned int cpu) |
9556 | { | |
9557 | struct rq *rq = cpu_rq(cpu); | |
9558 | ||
9559 | rq->calc_load_update = calc_load_update; | |
94baf7a5 TG |
9560 | update_max_interval(); |
9561 | } | |
9562 | ||
135fb3e1 TG |
9563 | int sched_cpu_starting(unsigned int cpu) |
9564 | { | |
9edeaea1 | 9565 | sched_core_cpu_starting(cpu); |
94baf7a5 | 9566 | sched_rq_cpu_starting(cpu); |
d84b3131 | 9567 | sched_tick_start(cpu); |
135fb3e1 | 9568 | return 0; |
e761b772 | 9569 | } |
e761b772 | 9570 | |
f2785ddb | 9571 | #ifdef CONFIG_HOTPLUG_CPU |
1cf12e08 TG |
9572 | |
9573 | /* | |
9574 | * Invoked immediately before the stopper thread is invoked to bring the | |
9575 | * CPU down completely. At this point all per CPU kthreads except the | |
9576 | * hotplug thread (current) and the stopper thread (inactive) have been | |
9577 | * either parked or have been unbound from the outgoing CPU. Ensure that | |
9578 | * any of those which might be on the way out are gone. | |
9579 | * | |
9580 | * If after this point a bound task is being woken on this CPU then the | |
9581 | * responsible hotplug callback has failed to do it's job. | |
9582 | * sched_cpu_dying() will catch it with the appropriate fireworks. | |
9583 | */ | |
9584 | int sched_cpu_wait_empty(unsigned int cpu) | |
9585 | { | |
9586 | balance_hotplug_wait(); | |
9587 | return 0; | |
9588 | } | |
9589 | ||
9590 | /* | |
9591 | * Since this CPU is going 'away' for a while, fold any nr_active delta we | |
9592 | * might have. Called from the CPU stopper task after ensuring that the | |
9593 | * stopper is the last running task on the CPU, so nr_active count is | |
9594 | * stable. We need to take the teardown thread which is calling this into | |
9595 | * account, so we hand in adjust = 1 to the load calculation. | |
9596 | * | |
9597 | * Also see the comment "Global load-average calculations". | |
9598 | */ | |
9599 | static void calc_load_migrate(struct rq *rq) | |
9600 | { | |
9601 | long delta = calc_load_fold_active(rq, 1); | |
9602 | ||
9603 | if (delta) | |
9604 | atomic_long_add(delta, &calc_load_tasks); | |
9605 | } | |
9606 | ||
36c6e17b VS |
9607 | static void dump_rq_tasks(struct rq *rq, const char *loglvl) |
9608 | { | |
9609 | struct task_struct *g, *p; | |
9610 | int cpu = cpu_of(rq); | |
9611 | ||
5cb9eaa3 | 9612 | lockdep_assert_rq_held(rq); |
36c6e17b VS |
9613 | |
9614 | printk("%sCPU%d enqueued tasks (%u total):\n", loglvl, cpu, rq->nr_running); | |
9615 | for_each_process_thread(g, p) { | |
9616 | if (task_cpu(p) != cpu) | |
9617 | continue; | |
9618 | ||
9619 | if (!task_on_rq_queued(p)) | |
9620 | continue; | |
9621 | ||
9622 | printk("%s\tpid: %d, name: %s\n", loglvl, p->pid, p->comm); | |
9623 | } | |
9624 | } | |
9625 | ||
f2785ddb TG |
9626 | int sched_cpu_dying(unsigned int cpu) |
9627 | { | |
9628 | struct rq *rq = cpu_rq(cpu); | |
8a8c69c3 | 9629 | struct rq_flags rf; |
f2785ddb TG |
9630 | |
9631 | /* Handle pending wakeups and then migrate everything off */ | |
d84b3131 | 9632 | sched_tick_stop(cpu); |
8a8c69c3 PZ |
9633 | |
9634 | rq_lock_irqsave(rq, &rf); | |
36c6e17b VS |
9635 | if (rq->nr_running != 1 || rq_has_pinned_tasks(rq)) { |
9636 | WARN(true, "Dying CPU not properly vacated!"); | |
9637 | dump_rq_tasks(rq, KERN_WARNING); | |
9638 | } | |
8a8c69c3 PZ |
9639 | rq_unlock_irqrestore(rq, &rf); |
9640 | ||
f2785ddb TG |
9641 | calc_load_migrate(rq); |
9642 | update_max_interval(); | |
e5ef27d0 | 9643 | hrtick_clear(rq); |
3c474b32 | 9644 | sched_core_cpu_dying(cpu); |
f2785ddb TG |
9645 | return 0; |
9646 | } | |
9647 | #endif | |
9648 | ||
1da177e4 LT |
9649 | void __init sched_init_smp(void) |
9650 | { | |
0fb3978b | 9651 | sched_init_numa(NUMA_NO_NODE); |
cb83b629 | 9652 | |
6acce3ef PZ |
9653 | /* |
9654 | * There's no userspace yet to cause hotplug operations; hence all the | |
d1ccc66d | 9655 | * CPU masks are stable and all blatant races in the below code cannot |
b5a4e2bb | 9656 | * happen. |
6acce3ef | 9657 | */ |
712555ee | 9658 | mutex_lock(&sched_domains_mutex); |
8d5dc512 | 9659 | sched_init_domains(cpu_active_mask); |
712555ee | 9660 | mutex_unlock(&sched_domains_mutex); |
e761b772 | 9661 | |
5c1e1767 | 9662 | /* Move init over to a non-isolated CPU */ |
04d4e665 | 9663 | if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_TYPE_DOMAIN)) < 0) |
5c1e1767 | 9664 | BUG(); |
15faafc6 | 9665 | current->flags &= ~PF_NO_SETAFFINITY; |
19978ca6 | 9666 | sched_init_granularity(); |
4212823f | 9667 | |
0e3900e6 | 9668 | init_sched_rt_class(); |
1baca4ce | 9669 | init_sched_dl_class(); |
1b568f0a | 9670 | |
e26fbffd | 9671 | sched_smp_initialized = true; |
1da177e4 | 9672 | } |
e26fbffd TG |
9673 | |
9674 | static int __init migration_init(void) | |
9675 | { | |
77a5352b | 9676 | sched_cpu_starting(smp_processor_id()); |
e26fbffd | 9677 | return 0; |
1da177e4 | 9678 | } |
e26fbffd TG |
9679 | early_initcall(migration_init); |
9680 | ||
1da177e4 LT |
9681 | #else |
9682 | void __init sched_init_smp(void) | |
9683 | { | |
19978ca6 | 9684 | sched_init_granularity(); |
1da177e4 LT |
9685 | } |
9686 | #endif /* CONFIG_SMP */ | |
9687 | ||
9688 | int in_sched_functions(unsigned long addr) | |
9689 | { | |
1da177e4 LT |
9690 | return in_lock_functions(addr) || |
9691 | (addr >= (unsigned long)__sched_text_start | |
9692 | && addr < (unsigned long)__sched_text_end); | |
9693 | } | |
9694 | ||
029632fb | 9695 | #ifdef CONFIG_CGROUP_SCHED |
27b4b931 LZ |
9696 | /* |
9697 | * Default task group. | |
9698 | * Every task in system belongs to this group at bootup. | |
9699 | */ | |
029632fb | 9700 | struct task_group root_task_group; |
35cf4e50 | 9701 | LIST_HEAD(task_groups); |
b0367629 WL |
9702 | |
9703 | /* Cacheline aligned slab cache for task_group */ | |
9704 | static struct kmem_cache *task_group_cache __read_mostly; | |
052f1dc7 | 9705 | #endif |
6f505b16 | 9706 | |
1da177e4 LT |
9707 | void __init sched_init(void) |
9708 | { | |
a1dc0446 | 9709 | unsigned long ptr = 0; |
55627e3c | 9710 | int i; |
434d53b0 | 9711 | |
c3a340f7 | 9712 | /* Make sure the linker didn't screw up */ |
546a3fee PZ |
9713 | BUG_ON(&idle_sched_class != &fair_sched_class + 1 || |
9714 | &fair_sched_class != &rt_sched_class + 1 || | |
9715 | &rt_sched_class != &dl_sched_class + 1); | |
c3a340f7 | 9716 | #ifdef CONFIG_SMP |
546a3fee | 9717 | BUG_ON(&dl_sched_class != &stop_sched_class + 1); |
c3a340f7 SRV |
9718 | #endif |
9719 | ||
5822a454 | 9720 | wait_bit_init(); |
9dcb8b68 | 9721 | |
434d53b0 | 9722 | #ifdef CONFIG_FAIR_GROUP_SCHED |
a1dc0446 | 9723 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 MT |
9724 | #endif |
9725 | #ifdef CONFIG_RT_GROUP_SCHED | |
a1dc0446 | 9726 | ptr += 2 * nr_cpu_ids * sizeof(void **); |
434d53b0 | 9727 | #endif |
a1dc0446 QC |
9728 | if (ptr) { |
9729 | ptr = (unsigned long)kzalloc(ptr, GFP_NOWAIT); | |
434d53b0 MT |
9730 | |
9731 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 9732 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
9733 | ptr += nr_cpu_ids * sizeof(void **); |
9734 | ||
07e06b01 | 9735 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 9736 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 9737 | |
b1d1779e WY |
9738 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
9739 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); | |
6d6bc0ad | 9740 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 9741 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9742 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
9743 | ptr += nr_cpu_ids * sizeof(void **); |
9744 | ||
07e06b01 | 9745 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
9746 | ptr += nr_cpu_ids * sizeof(void **); |
9747 | ||
6d6bc0ad | 9748 | #endif /* CONFIG_RT_GROUP_SCHED */ |
b74e6278 | 9749 | } |
dd41f596 | 9750 | |
d1ccc66d | 9751 | init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime()); |
332ac17e | 9752 | |
57d885fe GH |
9753 | #ifdef CONFIG_SMP |
9754 | init_defrootdomain(); | |
9755 | #endif | |
9756 | ||
d0b27fa7 | 9757 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9758 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 9759 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 9760 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 9761 | |
7c941438 | 9762 | #ifdef CONFIG_CGROUP_SCHED |
b0367629 WL |
9763 | task_group_cache = KMEM_CACHE(task_group, 0); |
9764 | ||
07e06b01 YZ |
9765 | list_add(&root_task_group.list, &task_groups); |
9766 | INIT_LIST_HEAD(&root_task_group.children); | |
f4d6f6c2 | 9767 | INIT_LIST_HEAD(&root_task_group.siblings); |
5091faa4 | 9768 | autogroup_init(&init_task); |
7c941438 | 9769 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 9770 | |
0a945022 | 9771 | for_each_possible_cpu(i) { |
70b97a7f | 9772 | struct rq *rq; |
1da177e4 LT |
9773 | |
9774 | rq = cpu_rq(i); | |
5cb9eaa3 | 9775 | raw_spin_lock_init(&rq->__lock); |
7897986b | 9776 | rq->nr_running = 0; |
dce48a84 TG |
9777 | rq->calc_load_active = 0; |
9778 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
acb5a9ba | 9779 | init_cfs_rq(&rq->cfs); |
07c54f7a AV |
9780 | init_rt_rq(&rq->rt); |
9781 | init_dl_rq(&rq->dl); | |
dd41f596 | 9782 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6f505b16 | 9783 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
9c2791f9 | 9784 | rq->tmp_alone_branch = &rq->leaf_cfs_rq_list; |
354d60c2 | 9785 | /* |
d1ccc66d | 9786 | * How much CPU bandwidth does root_task_group get? |
354d60c2 DG |
9787 | * |
9788 | * In case of task-groups formed thr' the cgroup filesystem, it | |
d1ccc66d IM |
9789 | * gets 100% of the CPU resources in the system. This overall |
9790 | * system CPU resource is divided among the tasks of | |
07e06b01 | 9791 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
9792 | * based on each entity's (task or task-group's) weight |
9793 | * (se->load.weight). | |
9794 | * | |
07e06b01 | 9795 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 | 9796 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
d1ccc66d | 9797 | * then A0's share of the CPU resource is: |
354d60c2 | 9798 | * |
0d905bca | 9799 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 9800 | * |
07e06b01 YZ |
9801 | * We achieve this by letting root_task_group's tasks sit |
9802 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 9803 | */ |
07e06b01 | 9804 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
9805 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9806 | ||
9807 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9808 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 9809 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 9810 | #endif |
1da177e4 | 9811 | #ifdef CONFIG_SMP |
41c7ce9a | 9812 | rq->sd = NULL; |
57d885fe | 9813 | rq->rd = NULL; |
ca6d75e6 | 9814 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
b5c44773 | 9815 | rq->balance_callback = &balance_push_callback; |
1da177e4 | 9816 | rq->active_balance = 0; |
dd41f596 | 9817 | rq->next_balance = jiffies; |
1da177e4 | 9818 | rq->push_cpu = 0; |
0a2966b4 | 9819 | rq->cpu = i; |
1f11eb6a | 9820 | rq->online = 0; |
eae0c9df MG |
9821 | rq->idle_stamp = 0; |
9822 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
94aafc3e PZ |
9823 | rq->wake_stamp = jiffies; |
9824 | rq->wake_avg_idle = rq->avg_idle; | |
9bd721c5 | 9825 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
367456c7 PZ |
9826 | |
9827 | INIT_LIST_HEAD(&rq->cfs_tasks); | |
9828 | ||
dc938520 | 9829 | rq_attach_root(rq, &def_root_domain); |
3451d024 | 9830 | #ifdef CONFIG_NO_HZ_COMMON |
e022e0d3 | 9831 | rq->last_blocked_load_update_tick = jiffies; |
a22e47a4 | 9832 | atomic_set(&rq->nohz_flags, 0); |
90b5363a | 9833 | |
545b8c8d | 9834 | INIT_CSD(&rq->nohz_csd, nohz_csd_func, rq); |
83cd4fe2 | 9835 | #endif |
f2469a1f TG |
9836 | #ifdef CONFIG_HOTPLUG_CPU |
9837 | rcuwait_init(&rq->hotplug_wait); | |
83cd4fe2 | 9838 | #endif |
9fd81dd5 | 9839 | #endif /* CONFIG_SMP */ |
77a021be | 9840 | hrtick_rq_init(rq); |
1da177e4 | 9841 | atomic_set(&rq->nr_iowait, 0); |
9edeaea1 PZ |
9842 | |
9843 | #ifdef CONFIG_SCHED_CORE | |
3c474b32 | 9844 | rq->core = rq; |
539f6512 | 9845 | rq->core_pick = NULL; |
9edeaea1 | 9846 | rq->core_enabled = 0; |
539f6512 | 9847 | rq->core_tree = RB_ROOT; |
4feee7d1 JD |
9848 | rq->core_forceidle_count = 0; |
9849 | rq->core_forceidle_occupation = 0; | |
9850 | rq->core_forceidle_start = 0; | |
539f6512 PZ |
9851 | |
9852 | rq->core_cookie = 0UL; | |
9edeaea1 | 9853 | #endif |
da019032 | 9854 | zalloc_cpumask_var_node(&rq->scratch_mask, GFP_KERNEL, cpu_to_node(i)); |
1da177e4 LT |
9855 | } |
9856 | ||
b1e82065 | 9857 | set_load_weight(&init_task, false); |
b50f60ce | 9858 | |
1da177e4 LT |
9859 | /* |
9860 | * The boot idle thread does lazy MMU switching as well: | |
9861 | */ | |
f1f10076 | 9862 | mmgrab(&init_mm); |
1da177e4 LT |
9863 | enter_lazy_tlb(&init_mm, current); |
9864 | ||
40966e31 EB |
9865 | /* |
9866 | * The idle task doesn't need the kthread struct to function, but it | |
9867 | * is dressed up as a per-CPU kthread and thus needs to play the part | |
9868 | * if we want to avoid special-casing it in code that deals with per-CPU | |
9869 | * kthreads. | |
9870 | */ | |
dd621ee0 | 9871 | WARN_ON(!set_kthread_struct(current)); |
40966e31 | 9872 | |
1da177e4 LT |
9873 | /* |
9874 | * Make us the idle thread. Technically, schedule() should not be | |
9875 | * called from this thread, however somewhere below it might be, | |
9876 | * but because we are the idle thread, we just pick up running again | |
9877 | * when this runqueue becomes "idle". | |
9878 | */ | |
9879 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9880 | |
9881 | calc_load_update = jiffies + LOAD_FREQ; | |
9882 | ||
bf4d83f6 | 9883 | #ifdef CONFIG_SMP |
29d5e047 | 9884 | idle_thread_set_boot_cpu(); |
b5c44773 | 9885 | balance_push_set(smp_processor_id(), false); |
029632fb PZ |
9886 | #endif |
9887 | init_sched_fair_class(); | |
6a7b3dc3 | 9888 | |
eb414681 JW |
9889 | psi_init(); |
9890 | ||
69842cba PB |
9891 | init_uclamp(); |
9892 | ||
c597bfdd FW |
9893 | preempt_dynamic_init(); |
9894 | ||
6892b75e | 9895 | scheduler_running = 1; |
1da177e4 LT |
9896 | } |
9897 | ||
d902db1e | 9898 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
e4aafea2 | 9899 | |
42a38756 | 9900 | void __might_sleep(const char *file, int line) |
1da177e4 | 9901 | { |
d6c23bb3 | 9902 | unsigned int state = get_current_state(); |
8eb23b9f PZ |
9903 | /* |
9904 | * Blocking primitives will set (and therefore destroy) current->state, | |
9905 | * since we will exit with TASK_RUNNING make sure we enter with it, | |
9906 | * otherwise we will destroy state. | |
9907 | */ | |
d6c23bb3 | 9908 | WARN_ONCE(state != TASK_RUNNING && current->task_state_change, |
8eb23b9f | 9909 | "do not call blocking ops when !TASK_RUNNING; " |
d6c23bb3 | 9910 | "state=%x set at [<%p>] %pS\n", state, |
8eb23b9f | 9911 | (void *)current->task_state_change, |
00845eb9 | 9912 | (void *)current->task_state_change); |
8eb23b9f | 9913 | |
42a38756 | 9914 | __might_resched(file, line, 0); |
3427445a PZ |
9915 | } |
9916 | EXPORT_SYMBOL(__might_sleep); | |
9917 | ||
8d713b69 TG |
9918 | static void print_preempt_disable_ip(int preempt_offset, unsigned long ip) |
9919 | { | |
9920 | if (!IS_ENABLED(CONFIG_DEBUG_PREEMPT)) | |
9921 | return; | |
9922 | ||
9923 | if (preempt_count() == preempt_offset) | |
9924 | return; | |
9925 | ||
9926 | pr_err("Preemption disabled at:"); | |
9927 | print_ip_sym(KERN_ERR, ip); | |
9928 | } | |
9929 | ||
50e081b9 TG |
9930 | static inline bool resched_offsets_ok(unsigned int offsets) |
9931 | { | |
9932 | unsigned int nested = preempt_count(); | |
9933 | ||
9934 | nested += rcu_preempt_depth() << MIGHT_RESCHED_RCU_SHIFT; | |
9935 | ||
9936 | return nested == offsets; | |
9937 | } | |
9938 | ||
9939 | void __might_resched(const char *file, int line, unsigned int offsets) | |
1da177e4 | 9940 | { |
d1ccc66d IM |
9941 | /* Ratelimiting timestamp: */ |
9942 | static unsigned long prev_jiffy; | |
9943 | ||
d1c6d149 | 9944 | unsigned long preempt_disable_ip; |
1da177e4 | 9945 | |
d1ccc66d IM |
9946 | /* WARN_ON_ONCE() by default, no rate limit required: */ |
9947 | rcu_sleep_check(); | |
9948 | ||
50e081b9 | 9949 | if ((resched_offsets_ok(offsets) && !irqs_disabled() && |
312364f3 | 9950 | !is_idle_task(current) && !current->non_block_count) || |
1c3c5eab TG |
9951 | system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING || |
9952 | oops_in_progress) | |
aef745fc | 9953 | return; |
1c3c5eab | 9954 | |
aef745fc IM |
9955 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
9956 | return; | |
9957 | prev_jiffy = jiffies; | |
9958 | ||
d1ccc66d | 9959 | /* Save this before calling printk(), since that will clobber it: */ |
d1c6d149 VN |
9960 | preempt_disable_ip = get_preempt_disable_ip(current); |
9961 | ||
a45ed302 TG |
9962 | pr_err("BUG: sleeping function called from invalid context at %s:%d\n", |
9963 | file, line); | |
9964 | pr_err("in_atomic(): %d, irqs_disabled(): %d, non_block: %d, pid: %d, name: %s\n", | |
9965 | in_atomic(), irqs_disabled(), current->non_block_count, | |
9966 | current->pid, current->comm); | |
8d713b69 | 9967 | pr_err("preempt_count: %x, expected: %x\n", preempt_count(), |
50e081b9 | 9968 | offsets & MIGHT_RESCHED_PREEMPT_MASK); |
8d713b69 TG |
9969 | |
9970 | if (IS_ENABLED(CONFIG_PREEMPT_RCU)) { | |
50e081b9 TG |
9971 | pr_err("RCU nest depth: %d, expected: %u\n", |
9972 | rcu_preempt_depth(), offsets >> MIGHT_RESCHED_RCU_SHIFT); | |
8d713b69 | 9973 | } |
aef745fc | 9974 | |
a8b686b3 | 9975 | if (task_stack_end_corrupted(current)) |
a45ed302 | 9976 | pr_emerg("Thread overran stack, or stack corrupted\n"); |
a8b686b3 | 9977 | |
aef745fc IM |
9978 | debug_show_held_locks(current); |
9979 | if (irqs_disabled()) | |
9980 | print_irqtrace_events(current); | |
8d713b69 | 9981 | |
50e081b9 TG |
9982 | print_preempt_disable_ip(offsets & MIGHT_RESCHED_PREEMPT_MASK, |
9983 | preempt_disable_ip); | |
8d713b69 | 9984 | |
aef745fc | 9985 | dump_stack(); |
f0b22e39 | 9986 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
1da177e4 | 9987 | } |
874f670e | 9988 | EXPORT_SYMBOL(__might_resched); |
568f1967 PZ |
9989 | |
9990 | void __cant_sleep(const char *file, int line, int preempt_offset) | |
9991 | { | |
9992 | static unsigned long prev_jiffy; | |
9993 | ||
9994 | if (irqs_disabled()) | |
9995 | return; | |
9996 | ||
9997 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
9998 | return; | |
9999 | ||
10000 | if (preempt_count() > preempt_offset) | |
10001 | return; | |
10002 | ||
10003 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
10004 | return; | |
10005 | prev_jiffy = jiffies; | |
10006 | ||
10007 | printk(KERN_ERR "BUG: assuming atomic context at %s:%d\n", file, line); | |
10008 | printk(KERN_ERR "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
10009 | in_atomic(), irqs_disabled(), | |
10010 | current->pid, current->comm); | |
10011 | ||
10012 | debug_show_held_locks(current); | |
10013 | dump_stack(); | |
10014 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
10015 | } | |
10016 | EXPORT_SYMBOL_GPL(__cant_sleep); | |
74d862b6 TG |
10017 | |
10018 | #ifdef CONFIG_SMP | |
10019 | void __cant_migrate(const char *file, int line) | |
10020 | { | |
10021 | static unsigned long prev_jiffy; | |
10022 | ||
10023 | if (irqs_disabled()) | |
10024 | return; | |
10025 | ||
10026 | if (is_migration_disabled(current)) | |
10027 | return; | |
10028 | ||
10029 | if (!IS_ENABLED(CONFIG_PREEMPT_COUNT)) | |
10030 | return; | |
10031 | ||
10032 | if (preempt_count() > 0) | |
10033 | return; | |
10034 | ||
10035 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
10036 | return; | |
10037 | prev_jiffy = jiffies; | |
10038 | ||
10039 | pr_err("BUG: assuming non migratable context at %s:%d\n", file, line); | |
10040 | pr_err("in_atomic(): %d, irqs_disabled(): %d, migration_disabled() %u pid: %d, name: %s\n", | |
10041 | in_atomic(), irqs_disabled(), is_migration_disabled(current), | |
10042 | current->pid, current->comm); | |
10043 | ||
10044 | debug_show_held_locks(current); | |
10045 | dump_stack(); | |
10046 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); | |
10047 | } | |
10048 | EXPORT_SYMBOL_GPL(__cant_migrate); | |
10049 | #endif | |
1da177e4 LT |
10050 | #endif |
10051 | ||
10052 | #ifdef CONFIG_MAGIC_SYSRQ | |
dbc7f069 | 10053 | void normalize_rt_tasks(void) |
3a5e4dc1 | 10054 | { |
dbc7f069 | 10055 | struct task_struct *g, *p; |
d50dde5a DF |
10056 | struct sched_attr attr = { |
10057 | .sched_policy = SCHED_NORMAL, | |
10058 | }; | |
1da177e4 | 10059 | |
3472eaa1 | 10060 | read_lock(&tasklist_lock); |
5d07f420 | 10061 | for_each_process_thread(g, p) { |
178be793 IM |
10062 | /* |
10063 | * Only normalize user tasks: | |
10064 | */ | |
3472eaa1 | 10065 | if (p->flags & PF_KTHREAD) |
178be793 IM |
10066 | continue; |
10067 | ||
4fa8d299 | 10068 | p->se.exec_start = 0; |
ceeadb83 YS |
10069 | schedstat_set(p->stats.wait_start, 0); |
10070 | schedstat_set(p->stats.sleep_start, 0); | |
10071 | schedstat_set(p->stats.block_start, 0); | |
dd41f596 | 10072 | |
aab03e05 | 10073 | if (!dl_task(p) && !rt_task(p)) { |
dd41f596 IM |
10074 | /* |
10075 | * Renice negative nice level userspace | |
10076 | * tasks back to 0: | |
10077 | */ | |
3472eaa1 | 10078 | if (task_nice(p) < 0) |
dd41f596 | 10079 | set_user_nice(p, 0); |
1da177e4 | 10080 | continue; |
dd41f596 | 10081 | } |
1da177e4 | 10082 | |
dbc7f069 | 10083 | __sched_setscheduler(p, &attr, false, false); |
5d07f420 | 10084 | } |
3472eaa1 | 10085 | read_unlock(&tasklist_lock); |
1da177e4 LT |
10086 | } |
10087 | ||
10088 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 10089 | |
67fc4e0c | 10090 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 10091 | /* |
67fc4e0c | 10092 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
10093 | * |
10094 | * They can only be called when the whole system has been | |
10095 | * stopped - every CPU needs to be quiescent, and no scheduling | |
10096 | * activity can take place. Using them for anything else would | |
10097 | * be a serious bug, and as a result, they aren't even visible | |
10098 | * under any other configuration. | |
10099 | */ | |
10100 | ||
10101 | /** | |
d1ccc66d | 10102 | * curr_task - return the current task for a given CPU. |
1df5c10a LT |
10103 | * @cpu: the processor in question. |
10104 | * | |
10105 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
e69f6186 YB |
10106 | * |
10107 | * Return: The current task for @cpu. | |
1df5c10a | 10108 | */ |
36c8b586 | 10109 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
10110 | { |
10111 | return cpu_curr(cpu); | |
10112 | } | |
10113 | ||
67fc4e0c JW |
10114 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
10115 | ||
10116 | #ifdef CONFIG_IA64 | |
1df5c10a | 10117 | /** |
5feeb783 | 10118 | * ia64_set_curr_task - set the current task for a given CPU. |
1df5c10a LT |
10119 | * @cpu: the processor in question. |
10120 | * @p: the task pointer to set. | |
10121 | * | |
10122 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf | 10123 | * are serviced on a separate stack. It allows the architecture to switch the |
d1ccc66d | 10124 | * notion of the current task on a CPU in a non-blocking manner. This function |
1df5c10a LT |
10125 | * must be called with all CPU's synchronized, and interrupts disabled, the |
10126 | * and caller must save the original value of the current task (see | |
10127 | * curr_task() above) and restore that value before reenabling interrupts and | |
10128 | * re-starting the system. | |
10129 | * | |
10130 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
10131 | */ | |
a458ae2e | 10132 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
10133 | { |
10134 | cpu_curr(cpu) = p; | |
10135 | } | |
10136 | ||
10137 | #endif | |
29f59db3 | 10138 | |
7c941438 | 10139 | #ifdef CONFIG_CGROUP_SCHED |
029632fb PZ |
10140 | /* task_group_lock serializes the addition/removal of task groups */ |
10141 | static DEFINE_SPINLOCK(task_group_lock); | |
10142 | ||
2480c093 PB |
10143 | static inline void alloc_uclamp_sched_group(struct task_group *tg, |
10144 | struct task_group *parent) | |
10145 | { | |
10146 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
0413d7f3 | 10147 | enum uclamp_id clamp_id; |
2480c093 PB |
10148 | |
10149 | for_each_clamp_id(clamp_id) { | |
10150 | uclamp_se_set(&tg->uclamp_req[clamp_id], | |
10151 | uclamp_none(clamp_id), false); | |
0b60ba2d | 10152 | tg->uclamp[clamp_id] = parent->uclamp[clamp_id]; |
2480c093 PB |
10153 | } |
10154 | #endif | |
10155 | } | |
10156 | ||
2f5177f0 | 10157 | static void sched_free_group(struct task_group *tg) |
bccbe08a PZ |
10158 | { |
10159 | free_fair_sched_group(tg); | |
10160 | free_rt_sched_group(tg); | |
e9aa1dd1 | 10161 | autogroup_free(tg); |
b0367629 | 10162 | kmem_cache_free(task_group_cache, tg); |
bccbe08a PZ |
10163 | } |
10164 | ||
b027789e MK |
10165 | static void sched_free_group_rcu(struct rcu_head *rcu) |
10166 | { | |
10167 | sched_free_group(container_of(rcu, struct task_group, rcu)); | |
10168 | } | |
10169 | ||
10170 | static void sched_unregister_group(struct task_group *tg) | |
10171 | { | |
10172 | unregister_fair_sched_group(tg); | |
10173 | unregister_rt_sched_group(tg); | |
10174 | /* | |
10175 | * We have to wait for yet another RCU grace period to expire, as | |
10176 | * print_cfs_stats() might run concurrently. | |
10177 | */ | |
10178 | call_rcu(&tg->rcu, sched_free_group_rcu); | |
10179 | } | |
10180 | ||
bccbe08a | 10181 | /* allocate runqueue etc for a new task group */ |
ec7dc8ac | 10182 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
10183 | { |
10184 | struct task_group *tg; | |
bccbe08a | 10185 | |
b0367629 | 10186 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
bccbe08a PZ |
10187 | if (!tg) |
10188 | return ERR_PTR(-ENOMEM); | |
10189 | ||
ec7dc8ac | 10190 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
10191 | goto err; |
10192 | ||
ec7dc8ac | 10193 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
10194 | goto err; |
10195 | ||
2480c093 PB |
10196 | alloc_uclamp_sched_group(tg, parent); |
10197 | ||
ace783b9 LZ |
10198 | return tg; |
10199 | ||
10200 | err: | |
2f5177f0 | 10201 | sched_free_group(tg); |
ace783b9 LZ |
10202 | return ERR_PTR(-ENOMEM); |
10203 | } | |
10204 | ||
10205 | void sched_online_group(struct task_group *tg, struct task_group *parent) | |
10206 | { | |
10207 | unsigned long flags; | |
10208 | ||
8ed36996 | 10209 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 10210 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e | 10211 | |
d1ccc66d IM |
10212 | /* Root should already exist: */ |
10213 | WARN_ON(!parent); | |
f473aa5e PZ |
10214 | |
10215 | tg->parent = parent; | |
f473aa5e | 10216 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 10217 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 10218 | spin_unlock_irqrestore(&task_group_lock, flags); |
8663e24d PZ |
10219 | |
10220 | online_fair_sched_group(tg); | |
29f59db3 SV |
10221 | } |
10222 | ||
9b5b7751 | 10223 | /* rcu callback to free various structures associated with a task group */ |
b027789e | 10224 | static void sched_unregister_group_rcu(struct rcu_head *rhp) |
29f59db3 | 10225 | { |
d1ccc66d | 10226 | /* Now it should be safe to free those cfs_rqs: */ |
b027789e | 10227 | sched_unregister_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
10228 | } |
10229 | ||
4cf86d77 | 10230 | void sched_destroy_group(struct task_group *tg) |
ace783b9 | 10231 | { |
d1ccc66d | 10232 | /* Wait for possible concurrent references to cfs_rqs complete: */ |
b027789e | 10233 | call_rcu(&tg->rcu, sched_unregister_group_rcu); |
ace783b9 LZ |
10234 | } |
10235 | ||
b027789e | 10236 | void sched_release_group(struct task_group *tg) |
29f59db3 | 10237 | { |
8ed36996 | 10238 | unsigned long flags; |
29f59db3 | 10239 | |
b027789e MK |
10240 | /* |
10241 | * Unlink first, to avoid walk_tg_tree_from() from finding us (via | |
10242 | * sched_cfs_period_timer()). | |
10243 | * | |
10244 | * For this to be effective, we have to wait for all pending users of | |
10245 | * this task group to leave their RCU critical section to ensure no new | |
10246 | * user will see our dying task group any more. Specifically ensure | |
10247 | * that tg_unthrottle_up() won't add decayed cfs_rq's to it. | |
10248 | * | |
10249 | * We therefore defer calling unregister_fair_sched_group() to | |
10250 | * sched_unregister_group() which is guarantied to get called only after the | |
10251 | * current RCU grace period has expired. | |
10252 | */ | |
3d4b47b4 | 10253 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 10254 | list_del_rcu(&tg->list); |
f473aa5e | 10255 | list_del_rcu(&tg->siblings); |
8ed36996 | 10256 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 SV |
10257 | } |
10258 | ||
39c42611 | 10259 | static void sched_change_group(struct task_struct *tsk) |
29f59db3 | 10260 | { |
8323f26c | 10261 | struct task_group *tg; |
29f59db3 | 10262 | |
f7b8a47d KT |
10263 | /* |
10264 | * All callers are synchronized by task_rq_lock(); we do not use RCU | |
10265 | * which is pointless here. Thus, we pass "true" to task_css_check() | |
10266 | * to prevent lockdep warnings. | |
10267 | */ | |
10268 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), | |
8323f26c PZ |
10269 | struct task_group, css); |
10270 | tg = autogroup_task_group(tsk, tg); | |
10271 | tsk->sched_task_group = tg; | |
10272 | ||
810b3817 | 10273 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ea86cb4b | 10274 | if (tsk->sched_class->task_change_group) |
39c42611 | 10275 | tsk->sched_class->task_change_group(tsk); |
b2b5ce02 | 10276 | else |
810b3817 | 10277 | #endif |
b2b5ce02 | 10278 | set_task_rq(tsk, task_cpu(tsk)); |
ea86cb4b VG |
10279 | } |
10280 | ||
10281 | /* | |
10282 | * Change task's runqueue when it moves between groups. | |
10283 | * | |
10284 | * The caller of this function should have put the task in its new group by | |
10285 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect | |
10286 | * its new group. | |
10287 | */ | |
10288 | void sched_move_task(struct task_struct *tsk) | |
10289 | { | |
7a57f32a PZ |
10290 | int queued, running, queue_flags = |
10291 | DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | |
ea86cb4b VG |
10292 | struct rq_flags rf; |
10293 | struct rq *rq; | |
10294 | ||
10295 | rq = task_rq_lock(tsk, &rf); | |
1b1d6225 | 10296 | update_rq_clock(rq); |
ea86cb4b VG |
10297 | |
10298 | running = task_current(rq, tsk); | |
10299 | queued = task_on_rq_queued(tsk); | |
10300 | ||
10301 | if (queued) | |
7a57f32a | 10302 | dequeue_task(rq, tsk, queue_flags); |
bb3bac2c | 10303 | if (running) |
ea86cb4b VG |
10304 | put_prev_task(rq, tsk); |
10305 | ||
39c42611 | 10306 | sched_change_group(tsk); |
810b3817 | 10307 | |
da0c1e65 | 10308 | if (queued) |
7a57f32a | 10309 | enqueue_task(rq, tsk, queue_flags); |
2a4b03ff | 10310 | if (running) { |
03b7fad1 | 10311 | set_next_task(rq, tsk); |
2a4b03ff VG |
10312 | /* |
10313 | * After changing group, the running task may have joined a | |
10314 | * throttled one but it's still the running task. Trigger a | |
10315 | * resched to make sure that task can still run. | |
10316 | */ | |
10317 | resched_curr(rq); | |
10318 | } | |
29f59db3 | 10319 | |
eb580751 | 10320 | task_rq_unlock(rq, tsk, &rf); |
29f59db3 | 10321 | } |
68318b8e | 10322 | |
a7c6d554 | 10323 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
68318b8e | 10324 | { |
a7c6d554 | 10325 | return css ? container_of(css, struct task_group, css) : NULL; |
68318b8e SV |
10326 | } |
10327 | ||
eb95419b TH |
10328 | static struct cgroup_subsys_state * |
10329 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
68318b8e | 10330 | { |
eb95419b TH |
10331 | struct task_group *parent = css_tg(parent_css); |
10332 | struct task_group *tg; | |
68318b8e | 10333 | |
eb95419b | 10334 | if (!parent) { |
68318b8e | 10335 | /* This is early initialization for the top cgroup */ |
07e06b01 | 10336 | return &root_task_group.css; |
68318b8e SV |
10337 | } |
10338 | ||
ec7dc8ac | 10339 | tg = sched_create_group(parent); |
68318b8e SV |
10340 | if (IS_ERR(tg)) |
10341 | return ERR_PTR(-ENOMEM); | |
10342 | ||
68318b8e SV |
10343 | return &tg->css; |
10344 | } | |
10345 | ||
96b77745 KK |
10346 | /* Expose task group only after completing cgroup initialization */ |
10347 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) | |
10348 | { | |
10349 | struct task_group *tg = css_tg(css); | |
10350 | struct task_group *parent = css_tg(css->parent); | |
10351 | ||
10352 | if (parent) | |
10353 | sched_online_group(tg, parent); | |
7226017a QY |
10354 | |
10355 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
10356 | /* Propagate the effective uclamp value for the new group */ | |
93b73858 QY |
10357 | mutex_lock(&uclamp_mutex); |
10358 | rcu_read_lock(); | |
7226017a | 10359 | cpu_util_update_eff(css); |
93b73858 QY |
10360 | rcu_read_unlock(); |
10361 | mutex_unlock(&uclamp_mutex); | |
7226017a QY |
10362 | #endif |
10363 | ||
96b77745 KK |
10364 | return 0; |
10365 | } | |
10366 | ||
2f5177f0 | 10367 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
ace783b9 | 10368 | { |
eb95419b | 10369 | struct task_group *tg = css_tg(css); |
ace783b9 | 10370 | |
b027789e | 10371 | sched_release_group(tg); |
ace783b9 LZ |
10372 | } |
10373 | ||
eb95419b | 10374 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
68318b8e | 10375 | { |
eb95419b | 10376 | struct task_group *tg = css_tg(css); |
68318b8e | 10377 | |
2f5177f0 PZ |
10378 | /* |
10379 | * Relies on the RCU grace period between css_released() and this. | |
10380 | */ | |
b027789e | 10381 | sched_unregister_group(tg); |
ace783b9 LZ |
10382 | } |
10383 | ||
df16b71c | 10384 | #ifdef CONFIG_RT_GROUP_SCHED |
1f7dd3e5 | 10385 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
68318b8e | 10386 | { |
bb9d97b6 | 10387 | struct task_struct *task; |
1f7dd3e5 | 10388 | struct cgroup_subsys_state *css; |
bb9d97b6 | 10389 | |
1f7dd3e5 | 10390 | cgroup_taskset_for_each(task, css, tset) { |
eb95419b | 10391 | if (!sched_rt_can_attach(css_tg(css), task)) |
bb9d97b6 | 10392 | return -EINVAL; |
bb9d97b6 | 10393 | } |
df16b71c | 10394 | return 0; |
be367d09 | 10395 | } |
df16b71c | 10396 | #endif |
68318b8e | 10397 | |
1f7dd3e5 | 10398 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
68318b8e | 10399 | { |
bb9d97b6 | 10400 | struct task_struct *task; |
1f7dd3e5 | 10401 | struct cgroup_subsys_state *css; |
bb9d97b6 | 10402 | |
1f7dd3e5 | 10403 | cgroup_taskset_for_each(task, css, tset) |
bb9d97b6 | 10404 | sched_move_task(task); |
68318b8e SV |
10405 | } |
10406 | ||
2480c093 | 10407 | #ifdef CONFIG_UCLAMP_TASK_GROUP |
0b60ba2d PB |
10408 | static void cpu_util_update_eff(struct cgroup_subsys_state *css) |
10409 | { | |
10410 | struct cgroup_subsys_state *top_css = css; | |
10411 | struct uclamp_se *uc_parent = NULL; | |
10412 | struct uclamp_se *uc_se = NULL; | |
10413 | unsigned int eff[UCLAMP_CNT]; | |
0413d7f3 | 10414 | enum uclamp_id clamp_id; |
0b60ba2d PB |
10415 | unsigned int clamps; |
10416 | ||
93b73858 QY |
10417 | lockdep_assert_held(&uclamp_mutex); |
10418 | SCHED_WARN_ON(!rcu_read_lock_held()); | |
10419 | ||
0b60ba2d PB |
10420 | css_for_each_descendant_pre(css, top_css) { |
10421 | uc_parent = css_tg(css)->parent | |
10422 | ? css_tg(css)->parent->uclamp : NULL; | |
10423 | ||
10424 | for_each_clamp_id(clamp_id) { | |
10425 | /* Assume effective clamps matches requested clamps */ | |
10426 | eff[clamp_id] = css_tg(css)->uclamp_req[clamp_id].value; | |
10427 | /* Cap effective clamps with parent's effective clamps */ | |
10428 | if (uc_parent && | |
10429 | eff[clamp_id] > uc_parent[clamp_id].value) { | |
10430 | eff[clamp_id] = uc_parent[clamp_id].value; | |
10431 | } | |
10432 | } | |
10433 | /* Ensure protection is always capped by limit */ | |
10434 | eff[UCLAMP_MIN] = min(eff[UCLAMP_MIN], eff[UCLAMP_MAX]); | |
10435 | ||
10436 | /* Propagate most restrictive effective clamps */ | |
10437 | clamps = 0x0; | |
10438 | uc_se = css_tg(css)->uclamp; | |
10439 | for_each_clamp_id(clamp_id) { | |
10440 | if (eff[clamp_id] == uc_se[clamp_id].value) | |
10441 | continue; | |
10442 | uc_se[clamp_id].value = eff[clamp_id]; | |
10443 | uc_se[clamp_id].bucket_id = uclamp_bucket_id(eff[clamp_id]); | |
10444 | clamps |= (0x1 << clamp_id); | |
10445 | } | |
babbe170 | 10446 | if (!clamps) { |
0b60ba2d | 10447 | css = css_rightmost_descendant(css); |
babbe170 PB |
10448 | continue; |
10449 | } | |
10450 | ||
10451 | /* Immediately update descendants RUNNABLE tasks */ | |
0213b708 | 10452 | uclamp_update_active_tasks(css); |
0b60ba2d PB |
10453 | } |
10454 | } | |
2480c093 PB |
10455 | |
10456 | /* | |
10457 | * Integer 10^N with a given N exponent by casting to integer the literal "1eN" | |
10458 | * C expression. Since there is no way to convert a macro argument (N) into a | |
10459 | * character constant, use two levels of macros. | |
10460 | */ | |
10461 | #define _POW10(exp) ((unsigned int)1e##exp) | |
10462 | #define POW10(exp) _POW10(exp) | |
10463 | ||
10464 | struct uclamp_request { | |
10465 | #define UCLAMP_PERCENT_SHIFT 2 | |
10466 | #define UCLAMP_PERCENT_SCALE (100 * POW10(UCLAMP_PERCENT_SHIFT)) | |
10467 | s64 percent; | |
10468 | u64 util; | |
10469 | int ret; | |
10470 | }; | |
10471 | ||
10472 | static inline struct uclamp_request | |
10473 | capacity_from_percent(char *buf) | |
10474 | { | |
10475 | struct uclamp_request req = { | |
10476 | .percent = UCLAMP_PERCENT_SCALE, | |
10477 | .util = SCHED_CAPACITY_SCALE, | |
10478 | .ret = 0, | |
10479 | }; | |
10480 | ||
10481 | buf = strim(buf); | |
10482 | if (strcmp(buf, "max")) { | |
10483 | req.ret = cgroup_parse_float(buf, UCLAMP_PERCENT_SHIFT, | |
10484 | &req.percent); | |
10485 | if (req.ret) | |
10486 | return req; | |
b562d140 | 10487 | if ((u64)req.percent > UCLAMP_PERCENT_SCALE) { |
2480c093 PB |
10488 | req.ret = -ERANGE; |
10489 | return req; | |
10490 | } | |
10491 | ||
10492 | req.util = req.percent << SCHED_CAPACITY_SHIFT; | |
10493 | req.util = DIV_ROUND_CLOSEST_ULL(req.util, UCLAMP_PERCENT_SCALE); | |
10494 | } | |
10495 | ||
10496 | return req; | |
10497 | } | |
10498 | ||
10499 | static ssize_t cpu_uclamp_write(struct kernfs_open_file *of, char *buf, | |
10500 | size_t nbytes, loff_t off, | |
10501 | enum uclamp_id clamp_id) | |
10502 | { | |
10503 | struct uclamp_request req; | |
10504 | struct task_group *tg; | |
10505 | ||
10506 | req = capacity_from_percent(buf); | |
10507 | if (req.ret) | |
10508 | return req.ret; | |
10509 | ||
46609ce2 QY |
10510 | static_branch_enable(&sched_uclamp_used); |
10511 | ||
2480c093 PB |
10512 | mutex_lock(&uclamp_mutex); |
10513 | rcu_read_lock(); | |
10514 | ||
10515 | tg = css_tg(of_css(of)); | |
10516 | if (tg->uclamp_req[clamp_id].value != req.util) | |
10517 | uclamp_se_set(&tg->uclamp_req[clamp_id], req.util, false); | |
10518 | ||
10519 | /* | |
10520 | * Because of not recoverable conversion rounding we keep track of the | |
10521 | * exact requested value | |
10522 | */ | |
10523 | tg->uclamp_pct[clamp_id] = req.percent; | |
10524 | ||
0b60ba2d PB |
10525 | /* Update effective clamps to track the most restrictive value */ |
10526 | cpu_util_update_eff(of_css(of)); | |
10527 | ||
2480c093 PB |
10528 | rcu_read_unlock(); |
10529 | mutex_unlock(&uclamp_mutex); | |
10530 | ||
10531 | return nbytes; | |
10532 | } | |
10533 | ||
10534 | static ssize_t cpu_uclamp_min_write(struct kernfs_open_file *of, | |
10535 | char *buf, size_t nbytes, | |
10536 | loff_t off) | |
10537 | { | |
10538 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MIN); | |
10539 | } | |
10540 | ||
10541 | static ssize_t cpu_uclamp_max_write(struct kernfs_open_file *of, | |
10542 | char *buf, size_t nbytes, | |
10543 | loff_t off) | |
10544 | { | |
10545 | return cpu_uclamp_write(of, buf, nbytes, off, UCLAMP_MAX); | |
10546 | } | |
10547 | ||
10548 | static inline void cpu_uclamp_print(struct seq_file *sf, | |
10549 | enum uclamp_id clamp_id) | |
10550 | { | |
10551 | struct task_group *tg; | |
10552 | u64 util_clamp; | |
10553 | u64 percent; | |
10554 | u32 rem; | |
10555 | ||
10556 | rcu_read_lock(); | |
10557 | tg = css_tg(seq_css(sf)); | |
10558 | util_clamp = tg->uclamp_req[clamp_id].value; | |
10559 | rcu_read_unlock(); | |
10560 | ||
10561 | if (util_clamp == SCHED_CAPACITY_SCALE) { | |
10562 | seq_puts(sf, "max\n"); | |
10563 | return; | |
10564 | } | |
10565 | ||
10566 | percent = tg->uclamp_pct[clamp_id]; | |
10567 | percent = div_u64_rem(percent, POW10(UCLAMP_PERCENT_SHIFT), &rem); | |
10568 | seq_printf(sf, "%llu.%0*u\n", percent, UCLAMP_PERCENT_SHIFT, rem); | |
10569 | } | |
10570 | ||
10571 | static int cpu_uclamp_min_show(struct seq_file *sf, void *v) | |
10572 | { | |
10573 | cpu_uclamp_print(sf, UCLAMP_MIN); | |
10574 | return 0; | |
10575 | } | |
10576 | ||
10577 | static int cpu_uclamp_max_show(struct seq_file *sf, void *v) | |
10578 | { | |
10579 | cpu_uclamp_print(sf, UCLAMP_MAX); | |
10580 | return 0; | |
10581 | } | |
10582 | #endif /* CONFIG_UCLAMP_TASK_GROUP */ | |
10583 | ||
052f1dc7 | 10584 | #ifdef CONFIG_FAIR_GROUP_SCHED |
182446d0 TH |
10585 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
10586 | struct cftype *cftype, u64 shareval) | |
68318b8e | 10587 | { |
5b61d50a KK |
10588 | if (shareval > scale_load_down(ULONG_MAX)) |
10589 | shareval = MAX_SHARES; | |
182446d0 | 10590 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
68318b8e SV |
10591 | } |
10592 | ||
182446d0 TH |
10593 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
10594 | struct cftype *cft) | |
68318b8e | 10595 | { |
182446d0 | 10596 | struct task_group *tg = css_tg(css); |
68318b8e | 10597 | |
c8b28116 | 10598 | return (u64) scale_load_down(tg->shares); |
68318b8e | 10599 | } |
ab84d31e PT |
10600 | |
10601 | #ifdef CONFIG_CFS_BANDWIDTH | |
a790de99 PT |
10602 | static DEFINE_MUTEX(cfs_constraints_mutex); |
10603 | ||
ab84d31e | 10604 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
b1546edc | 10605 | static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
d505b8af HC |
10606 | /* More than 203 days if BW_SHIFT equals 20. */ |
10607 | static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC; | |
ab84d31e | 10608 | |
a790de99 PT |
10609 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
10610 | ||
f4183717 HC |
10611 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota, |
10612 | u64 burst) | |
ab84d31e | 10613 | { |
56f570e5 | 10614 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
029632fb | 10615 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
ab84d31e PT |
10616 | |
10617 | if (tg == &root_task_group) | |
10618 | return -EINVAL; | |
10619 | ||
10620 | /* | |
10621 | * Ensure we have at some amount of bandwidth every period. This is | |
10622 | * to prevent reaching a state of large arrears when throttled via | |
10623 | * entity_tick() resulting in prolonged exit starvation. | |
10624 | */ | |
10625 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) | |
10626 | return -EINVAL; | |
10627 | ||
10628 | /* | |
3b03706f | 10629 | * Likewise, bound things on the other side by preventing insane quota |
ab84d31e PT |
10630 | * periods. This also allows us to normalize in computing quota |
10631 | * feasibility. | |
10632 | */ | |
10633 | if (period > max_cfs_quota_period) | |
10634 | return -EINVAL; | |
10635 | ||
d505b8af HC |
10636 | /* |
10637 | * Bound quota to defend quota against overflow during bandwidth shift. | |
10638 | */ | |
10639 | if (quota != RUNTIME_INF && quota > max_cfs_runtime) | |
10640 | return -EINVAL; | |
10641 | ||
f4183717 HC |
10642 | if (quota != RUNTIME_INF && (burst > quota || |
10643 | burst + quota > max_cfs_runtime)) | |
10644 | return -EINVAL; | |
10645 | ||
0e59bdae KT |
10646 | /* |
10647 | * Prevent race between setting of cfs_rq->runtime_enabled and | |
10648 | * unthrottle_offline_cfs_rqs(). | |
10649 | */ | |
746f5ea9 | 10650 | cpus_read_lock(); |
a790de99 PT |
10651 | mutex_lock(&cfs_constraints_mutex); |
10652 | ret = __cfs_schedulable(tg, period, quota); | |
10653 | if (ret) | |
10654 | goto out_unlock; | |
10655 | ||
58088ad0 | 10656 | runtime_enabled = quota != RUNTIME_INF; |
56f570e5 | 10657 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
1ee14e6c BS |
10658 | /* |
10659 | * If we need to toggle cfs_bandwidth_used, off->on must occur | |
10660 | * before making related changes, and on->off must occur afterwards | |
10661 | */ | |
10662 | if (runtime_enabled && !runtime_was_enabled) | |
10663 | cfs_bandwidth_usage_inc(); | |
ab84d31e PT |
10664 | raw_spin_lock_irq(&cfs_b->lock); |
10665 | cfs_b->period = ns_to_ktime(period); | |
10666 | cfs_b->quota = quota; | |
f4183717 | 10667 | cfs_b->burst = burst; |
58088ad0 | 10668 | |
a9cf55b2 | 10669 | __refill_cfs_bandwidth_runtime(cfs_b); |
d1ccc66d IM |
10670 | |
10671 | /* Restart the period timer (if active) to handle new period expiry: */ | |
77a4d1a1 PZ |
10672 | if (runtime_enabled) |
10673 | start_cfs_bandwidth(cfs_b); | |
d1ccc66d | 10674 | |
ab84d31e PT |
10675 | raw_spin_unlock_irq(&cfs_b->lock); |
10676 | ||
0e59bdae | 10677 | for_each_online_cpu(i) { |
ab84d31e | 10678 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
029632fb | 10679 | struct rq *rq = cfs_rq->rq; |
8a8c69c3 | 10680 | struct rq_flags rf; |
ab84d31e | 10681 | |
8a8c69c3 | 10682 | rq_lock_irq(rq, &rf); |
58088ad0 | 10683 | cfs_rq->runtime_enabled = runtime_enabled; |
ab84d31e | 10684 | cfs_rq->runtime_remaining = 0; |
671fd9da | 10685 | |
029632fb | 10686 | if (cfs_rq->throttled) |
671fd9da | 10687 | unthrottle_cfs_rq(cfs_rq); |
8a8c69c3 | 10688 | rq_unlock_irq(rq, &rf); |
ab84d31e | 10689 | } |
1ee14e6c BS |
10690 | if (runtime_was_enabled && !runtime_enabled) |
10691 | cfs_bandwidth_usage_dec(); | |
a790de99 PT |
10692 | out_unlock: |
10693 | mutex_unlock(&cfs_constraints_mutex); | |
746f5ea9 | 10694 | cpus_read_unlock(); |
ab84d31e | 10695 | |
a790de99 | 10696 | return ret; |
ab84d31e PT |
10697 | } |
10698 | ||
b1546edc | 10699 | static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
ab84d31e | 10700 | { |
f4183717 | 10701 | u64 quota, period, burst; |
ab84d31e | 10702 | |
029632fb | 10703 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
f4183717 | 10704 | burst = tg->cfs_bandwidth.burst; |
ab84d31e PT |
10705 | if (cfs_quota_us < 0) |
10706 | quota = RUNTIME_INF; | |
1a8b4540 | 10707 | else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC) |
ab84d31e | 10708 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
1a8b4540 KK |
10709 | else |
10710 | return -EINVAL; | |
ab84d31e | 10711 | |
f4183717 | 10712 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10713 | } |
10714 | ||
b1546edc | 10715 | static long tg_get_cfs_quota(struct task_group *tg) |
ab84d31e PT |
10716 | { |
10717 | u64 quota_us; | |
10718 | ||
029632fb | 10719 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
ab84d31e PT |
10720 | return -1; |
10721 | ||
029632fb | 10722 | quota_us = tg->cfs_bandwidth.quota; |
ab84d31e PT |
10723 | do_div(quota_us, NSEC_PER_USEC); |
10724 | ||
10725 | return quota_us; | |
10726 | } | |
10727 | ||
b1546edc | 10728 | static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
ab84d31e | 10729 | { |
f4183717 | 10730 | u64 quota, period, burst; |
ab84d31e | 10731 | |
1a8b4540 KK |
10732 | if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC) |
10733 | return -EINVAL; | |
10734 | ||
ab84d31e | 10735 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
029632fb | 10736 | quota = tg->cfs_bandwidth.quota; |
f4183717 | 10737 | burst = tg->cfs_bandwidth.burst; |
ab84d31e | 10738 | |
f4183717 | 10739 | return tg_set_cfs_bandwidth(tg, period, quota, burst); |
ab84d31e PT |
10740 | } |
10741 | ||
b1546edc | 10742 | static long tg_get_cfs_period(struct task_group *tg) |
ab84d31e PT |
10743 | { |
10744 | u64 cfs_period_us; | |
10745 | ||
029632fb | 10746 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
ab84d31e PT |
10747 | do_div(cfs_period_us, NSEC_PER_USEC); |
10748 | ||
10749 | return cfs_period_us; | |
10750 | } | |
10751 | ||
f4183717 HC |
10752 | static int tg_set_cfs_burst(struct task_group *tg, long cfs_burst_us) |
10753 | { | |
10754 | u64 quota, period, burst; | |
10755 | ||
10756 | if ((u64)cfs_burst_us > U64_MAX / NSEC_PER_USEC) | |
10757 | return -EINVAL; | |
10758 | ||
10759 | burst = (u64)cfs_burst_us * NSEC_PER_USEC; | |
10760 | period = ktime_to_ns(tg->cfs_bandwidth.period); | |
10761 | quota = tg->cfs_bandwidth.quota; | |
10762 | ||
10763 | return tg_set_cfs_bandwidth(tg, period, quota, burst); | |
10764 | } | |
10765 | ||
10766 | static long tg_get_cfs_burst(struct task_group *tg) | |
10767 | { | |
10768 | u64 burst_us; | |
10769 | ||
10770 | burst_us = tg->cfs_bandwidth.burst; | |
10771 | do_div(burst_us, NSEC_PER_USEC); | |
10772 | ||
10773 | return burst_us; | |
10774 | } | |
10775 | ||
182446d0 TH |
10776 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
10777 | struct cftype *cft) | |
ab84d31e | 10778 | { |
182446d0 | 10779 | return tg_get_cfs_quota(css_tg(css)); |
ab84d31e PT |
10780 | } |
10781 | ||
182446d0 TH |
10782 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
10783 | struct cftype *cftype, s64 cfs_quota_us) | |
ab84d31e | 10784 | { |
182446d0 | 10785 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
ab84d31e PT |
10786 | } |
10787 | ||
182446d0 TH |
10788 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
10789 | struct cftype *cft) | |
ab84d31e | 10790 | { |
182446d0 | 10791 | return tg_get_cfs_period(css_tg(css)); |
ab84d31e PT |
10792 | } |
10793 | ||
182446d0 TH |
10794 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
10795 | struct cftype *cftype, u64 cfs_period_us) | |
ab84d31e | 10796 | { |
182446d0 | 10797 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
ab84d31e PT |
10798 | } |
10799 | ||
f4183717 HC |
10800 | static u64 cpu_cfs_burst_read_u64(struct cgroup_subsys_state *css, |
10801 | struct cftype *cft) | |
10802 | { | |
10803 | return tg_get_cfs_burst(css_tg(css)); | |
10804 | } | |
10805 | ||
10806 | static int cpu_cfs_burst_write_u64(struct cgroup_subsys_state *css, | |
10807 | struct cftype *cftype, u64 cfs_burst_us) | |
10808 | { | |
10809 | return tg_set_cfs_burst(css_tg(css), cfs_burst_us); | |
10810 | } | |
10811 | ||
a790de99 PT |
10812 | struct cfs_schedulable_data { |
10813 | struct task_group *tg; | |
10814 | u64 period, quota; | |
10815 | }; | |
10816 | ||
10817 | /* | |
10818 | * normalize group quota/period to be quota/max_period | |
10819 | * note: units are usecs | |
10820 | */ | |
10821 | static u64 normalize_cfs_quota(struct task_group *tg, | |
10822 | struct cfs_schedulable_data *d) | |
10823 | { | |
10824 | u64 quota, period; | |
10825 | ||
10826 | if (tg == d->tg) { | |
10827 | period = d->period; | |
10828 | quota = d->quota; | |
10829 | } else { | |
10830 | period = tg_get_cfs_period(tg); | |
10831 | quota = tg_get_cfs_quota(tg); | |
10832 | } | |
10833 | ||
10834 | /* note: these should typically be equivalent */ | |
10835 | if (quota == RUNTIME_INF || quota == -1) | |
10836 | return RUNTIME_INF; | |
10837 | ||
10838 | return to_ratio(period, quota); | |
10839 | } | |
10840 | ||
10841 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) | |
10842 | { | |
10843 | struct cfs_schedulable_data *d = data; | |
029632fb | 10844 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
a790de99 PT |
10845 | s64 quota = 0, parent_quota = -1; |
10846 | ||
10847 | if (!tg->parent) { | |
10848 | quota = RUNTIME_INF; | |
10849 | } else { | |
029632fb | 10850 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
a790de99 PT |
10851 | |
10852 | quota = normalize_cfs_quota(tg, d); | |
9c58c79a | 10853 | parent_quota = parent_b->hierarchical_quota; |
a790de99 PT |
10854 | |
10855 | /* | |
c53593e5 TH |
10856 | * Ensure max(child_quota) <= parent_quota. On cgroup2, |
10857 | * always take the min. On cgroup1, only inherit when no | |
d1ccc66d | 10858 | * limit is set: |
a790de99 | 10859 | */ |
c53593e5 TH |
10860 | if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) { |
10861 | quota = min(quota, parent_quota); | |
10862 | } else { | |
10863 | if (quota == RUNTIME_INF) | |
10864 | quota = parent_quota; | |
10865 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) | |
10866 | return -EINVAL; | |
10867 | } | |
a790de99 | 10868 | } |
9c58c79a | 10869 | cfs_b->hierarchical_quota = quota; |
a790de99 PT |
10870 | |
10871 | return 0; | |
10872 | } | |
10873 | ||
10874 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) | |
10875 | { | |
8277434e | 10876 | int ret; |
a790de99 PT |
10877 | struct cfs_schedulable_data data = { |
10878 | .tg = tg, | |
10879 | .period = period, | |
10880 | .quota = quota, | |
10881 | }; | |
10882 | ||
10883 | if (quota != RUNTIME_INF) { | |
10884 | do_div(data.period, NSEC_PER_USEC); | |
10885 | do_div(data.quota, NSEC_PER_USEC); | |
10886 | } | |
10887 | ||
8277434e PT |
10888 | rcu_read_lock(); |
10889 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); | |
10890 | rcu_read_unlock(); | |
10891 | ||
10892 | return ret; | |
a790de99 | 10893 | } |
e8da1b18 | 10894 | |
a1f7164c | 10895 | static int cpu_cfs_stat_show(struct seq_file *sf, void *v) |
e8da1b18 | 10896 | { |
2da8ca82 | 10897 | struct task_group *tg = css_tg(seq_css(sf)); |
029632fb | 10898 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
e8da1b18 | 10899 | |
44ffc75b TH |
10900 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
10901 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); | |
10902 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); | |
e8da1b18 | 10903 | |
3d6c50c2 | 10904 | if (schedstat_enabled() && tg != &root_task_group) { |
ceeadb83 | 10905 | struct sched_statistics *stats; |
3d6c50c2 YW |
10906 | u64 ws = 0; |
10907 | int i; | |
10908 | ||
ceeadb83 YS |
10909 | for_each_possible_cpu(i) { |
10910 | stats = __schedstats_from_se(tg->se[i]); | |
10911 | ws += schedstat_val(stats->wait_sum); | |
10912 | } | |
3d6c50c2 YW |
10913 | |
10914 | seq_printf(sf, "wait_sum %llu\n", ws); | |
10915 | } | |
10916 | ||
bcb1704a HC |
10917 | seq_printf(sf, "nr_bursts %d\n", cfs_b->nr_burst); |
10918 | seq_printf(sf, "burst_time %llu\n", cfs_b->burst_time); | |
10919 | ||
e8da1b18 NR |
10920 | return 0; |
10921 | } | |
ab84d31e | 10922 | #endif /* CONFIG_CFS_BANDWIDTH */ |
6d6bc0ad | 10923 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10924 | |
052f1dc7 | 10925 | #ifdef CONFIG_RT_GROUP_SCHED |
182446d0 TH |
10926 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
10927 | struct cftype *cft, s64 val) | |
6f505b16 | 10928 | { |
182446d0 | 10929 | return sched_group_set_rt_runtime(css_tg(css), val); |
6f505b16 PZ |
10930 | } |
10931 | ||
182446d0 TH |
10932 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
10933 | struct cftype *cft) | |
6f505b16 | 10934 | { |
182446d0 | 10935 | return sched_group_rt_runtime(css_tg(css)); |
6f505b16 | 10936 | } |
d0b27fa7 | 10937 | |
182446d0 TH |
10938 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
10939 | struct cftype *cftype, u64 rt_period_us) | |
d0b27fa7 | 10940 | { |
182446d0 | 10941 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
d0b27fa7 PZ |
10942 | } |
10943 | ||
182446d0 TH |
10944 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
10945 | struct cftype *cft) | |
d0b27fa7 | 10946 | { |
182446d0 | 10947 | return sched_group_rt_period(css_tg(css)); |
d0b27fa7 | 10948 | } |
6d6bc0ad | 10949 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10950 | |
30400039 JD |
10951 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10952 | static s64 cpu_idle_read_s64(struct cgroup_subsys_state *css, | |
10953 | struct cftype *cft) | |
10954 | { | |
10955 | return css_tg(css)->idle; | |
10956 | } | |
10957 | ||
10958 | static int cpu_idle_write_s64(struct cgroup_subsys_state *css, | |
10959 | struct cftype *cft, s64 idle) | |
10960 | { | |
10961 | return sched_group_set_idle(css_tg(css), idle); | |
10962 | } | |
10963 | #endif | |
10964 | ||
a1f7164c | 10965 | static struct cftype cpu_legacy_files[] = { |
052f1dc7 | 10966 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10967 | { |
10968 | .name = "shares", | |
f4c753b7 PM |
10969 | .read_u64 = cpu_shares_read_u64, |
10970 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10971 | }, |
30400039 JD |
10972 | { |
10973 | .name = "idle", | |
10974 | .read_s64 = cpu_idle_read_s64, | |
10975 | .write_s64 = cpu_idle_write_s64, | |
10976 | }, | |
052f1dc7 | 10977 | #endif |
ab84d31e PT |
10978 | #ifdef CONFIG_CFS_BANDWIDTH |
10979 | { | |
10980 | .name = "cfs_quota_us", | |
10981 | .read_s64 = cpu_cfs_quota_read_s64, | |
10982 | .write_s64 = cpu_cfs_quota_write_s64, | |
10983 | }, | |
10984 | { | |
10985 | .name = "cfs_period_us", | |
10986 | .read_u64 = cpu_cfs_period_read_u64, | |
10987 | .write_u64 = cpu_cfs_period_write_u64, | |
10988 | }, | |
f4183717 HC |
10989 | { |
10990 | .name = "cfs_burst_us", | |
10991 | .read_u64 = cpu_cfs_burst_read_u64, | |
10992 | .write_u64 = cpu_cfs_burst_write_u64, | |
10993 | }, | |
e8da1b18 NR |
10994 | { |
10995 | .name = "stat", | |
a1f7164c | 10996 | .seq_show = cpu_cfs_stat_show, |
e8da1b18 | 10997 | }, |
ab84d31e | 10998 | #endif |
052f1dc7 | 10999 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 11000 | { |
9f0c1e56 | 11001 | .name = "rt_runtime_us", |
06ecb27c PM |
11002 | .read_s64 = cpu_rt_runtime_read, |
11003 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 11004 | }, |
d0b27fa7 PZ |
11005 | { |
11006 | .name = "rt_period_us", | |
f4c753b7 PM |
11007 | .read_u64 = cpu_rt_period_read_uint, |
11008 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 11009 | }, |
2480c093 PB |
11010 | #endif |
11011 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
11012 | { | |
11013 | .name = "uclamp.min", | |
11014 | .flags = CFTYPE_NOT_ON_ROOT, | |
11015 | .seq_show = cpu_uclamp_min_show, | |
11016 | .write = cpu_uclamp_min_write, | |
11017 | }, | |
11018 | { | |
11019 | .name = "uclamp.max", | |
11020 | .flags = CFTYPE_NOT_ON_ROOT, | |
11021 | .seq_show = cpu_uclamp_max_show, | |
11022 | .write = cpu_uclamp_max_write, | |
11023 | }, | |
052f1dc7 | 11024 | #endif |
d1ccc66d | 11025 | { } /* Terminate */ |
68318b8e SV |
11026 | }; |
11027 | ||
d41bf8c9 TH |
11028 | static int cpu_extra_stat_show(struct seq_file *sf, |
11029 | struct cgroup_subsys_state *css) | |
0d593634 | 11030 | { |
0d593634 TH |
11031 | #ifdef CONFIG_CFS_BANDWIDTH |
11032 | { | |
d41bf8c9 | 11033 | struct task_group *tg = css_tg(css); |
0d593634 | 11034 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
bcb1704a | 11035 | u64 throttled_usec, burst_usec; |
0d593634 TH |
11036 | |
11037 | throttled_usec = cfs_b->throttled_time; | |
11038 | do_div(throttled_usec, NSEC_PER_USEC); | |
bcb1704a HC |
11039 | burst_usec = cfs_b->burst_time; |
11040 | do_div(burst_usec, NSEC_PER_USEC); | |
0d593634 TH |
11041 | |
11042 | seq_printf(sf, "nr_periods %d\n" | |
11043 | "nr_throttled %d\n" | |
bcb1704a HC |
11044 | "throttled_usec %llu\n" |
11045 | "nr_bursts %d\n" | |
11046 | "burst_usec %llu\n", | |
0d593634 | 11047 | cfs_b->nr_periods, cfs_b->nr_throttled, |
bcb1704a | 11048 | throttled_usec, cfs_b->nr_burst, burst_usec); |
0d593634 TH |
11049 | } |
11050 | #endif | |
11051 | return 0; | |
11052 | } | |
11053 | ||
11054 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
11055 | static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css, | |
11056 | struct cftype *cft) | |
11057 | { | |
11058 | struct task_group *tg = css_tg(css); | |
11059 | u64 weight = scale_load_down(tg->shares); | |
11060 | ||
11061 | return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024); | |
11062 | } | |
11063 | ||
11064 | static int cpu_weight_write_u64(struct cgroup_subsys_state *css, | |
11065 | struct cftype *cft, u64 weight) | |
11066 | { | |
11067 | /* | |
11068 | * cgroup weight knobs should use the common MIN, DFL and MAX | |
11069 | * values which are 1, 100 and 10000 respectively. While it loses | |
11070 | * a bit of range on both ends, it maps pretty well onto the shares | |
11071 | * value used by scheduler and the round-trip conversions preserve | |
11072 | * the original value over the entire range. | |
11073 | */ | |
11074 | if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX) | |
11075 | return -ERANGE; | |
11076 | ||
11077 | weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL); | |
11078 | ||
11079 | return sched_group_set_shares(css_tg(css), scale_load(weight)); | |
11080 | } | |
11081 | ||
11082 | static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css, | |
11083 | struct cftype *cft) | |
11084 | { | |
11085 | unsigned long weight = scale_load_down(css_tg(css)->shares); | |
11086 | int last_delta = INT_MAX; | |
11087 | int prio, delta; | |
11088 | ||
11089 | /* find the closest nice value to the current weight */ | |
11090 | for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) { | |
11091 | delta = abs(sched_prio_to_weight[prio] - weight); | |
11092 | if (delta >= last_delta) | |
11093 | break; | |
11094 | last_delta = delta; | |
11095 | } | |
11096 | ||
11097 | return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO); | |
11098 | } | |
11099 | ||
11100 | static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css, | |
11101 | struct cftype *cft, s64 nice) | |
11102 | { | |
11103 | unsigned long weight; | |
7281c8de | 11104 | int idx; |
0d593634 TH |
11105 | |
11106 | if (nice < MIN_NICE || nice > MAX_NICE) | |
11107 | return -ERANGE; | |
11108 | ||
7281c8de PZ |
11109 | idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO; |
11110 | idx = array_index_nospec(idx, 40); | |
11111 | weight = sched_prio_to_weight[idx]; | |
11112 | ||
0d593634 TH |
11113 | return sched_group_set_shares(css_tg(css), scale_load(weight)); |
11114 | } | |
11115 | #endif | |
11116 | ||
11117 | static void __maybe_unused cpu_period_quota_print(struct seq_file *sf, | |
11118 | long period, long quota) | |
11119 | { | |
11120 | if (quota < 0) | |
11121 | seq_puts(sf, "max"); | |
11122 | else | |
11123 | seq_printf(sf, "%ld", quota); | |
11124 | ||
11125 | seq_printf(sf, " %ld\n", period); | |
11126 | } | |
11127 | ||
11128 | /* caller should put the current value in *@periodp before calling */ | |
11129 | static int __maybe_unused cpu_period_quota_parse(char *buf, | |
11130 | u64 *periodp, u64 *quotap) | |
11131 | { | |
11132 | char tok[21]; /* U64_MAX */ | |
11133 | ||
4c47acd8 | 11134 | if (sscanf(buf, "%20s %llu", tok, periodp) < 1) |
0d593634 TH |
11135 | return -EINVAL; |
11136 | ||
11137 | *periodp *= NSEC_PER_USEC; | |
11138 | ||
11139 | if (sscanf(tok, "%llu", quotap)) | |
11140 | *quotap *= NSEC_PER_USEC; | |
11141 | else if (!strcmp(tok, "max")) | |
11142 | *quotap = RUNTIME_INF; | |
11143 | else | |
11144 | return -EINVAL; | |
11145 | ||
11146 | return 0; | |
11147 | } | |
11148 | ||
11149 | #ifdef CONFIG_CFS_BANDWIDTH | |
11150 | static int cpu_max_show(struct seq_file *sf, void *v) | |
11151 | { | |
11152 | struct task_group *tg = css_tg(seq_css(sf)); | |
11153 | ||
11154 | cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg)); | |
11155 | return 0; | |
11156 | } | |
11157 | ||
11158 | static ssize_t cpu_max_write(struct kernfs_open_file *of, | |
11159 | char *buf, size_t nbytes, loff_t off) | |
11160 | { | |
11161 | struct task_group *tg = css_tg(of_css(of)); | |
11162 | u64 period = tg_get_cfs_period(tg); | |
f4183717 | 11163 | u64 burst = tg_get_cfs_burst(tg); |
0d593634 TH |
11164 | u64 quota; |
11165 | int ret; | |
11166 | ||
11167 | ret = cpu_period_quota_parse(buf, &period, "a); | |
11168 | if (!ret) | |
f4183717 | 11169 | ret = tg_set_cfs_bandwidth(tg, period, quota, burst); |
0d593634 TH |
11170 | return ret ?: nbytes; |
11171 | } | |
11172 | #endif | |
11173 | ||
11174 | static struct cftype cpu_files[] = { | |
0d593634 TH |
11175 | #ifdef CONFIG_FAIR_GROUP_SCHED |
11176 | { | |
11177 | .name = "weight", | |
11178 | .flags = CFTYPE_NOT_ON_ROOT, | |
11179 | .read_u64 = cpu_weight_read_u64, | |
11180 | .write_u64 = cpu_weight_write_u64, | |
11181 | }, | |
11182 | { | |
11183 | .name = "weight.nice", | |
11184 | .flags = CFTYPE_NOT_ON_ROOT, | |
11185 | .read_s64 = cpu_weight_nice_read_s64, | |
11186 | .write_s64 = cpu_weight_nice_write_s64, | |
11187 | }, | |
30400039 JD |
11188 | { |
11189 | .name = "idle", | |
11190 | .flags = CFTYPE_NOT_ON_ROOT, | |
11191 | .read_s64 = cpu_idle_read_s64, | |
11192 | .write_s64 = cpu_idle_write_s64, | |
11193 | }, | |
0d593634 TH |
11194 | #endif |
11195 | #ifdef CONFIG_CFS_BANDWIDTH | |
11196 | { | |
11197 | .name = "max", | |
11198 | .flags = CFTYPE_NOT_ON_ROOT, | |
11199 | .seq_show = cpu_max_show, | |
11200 | .write = cpu_max_write, | |
11201 | }, | |
f4183717 HC |
11202 | { |
11203 | .name = "max.burst", | |
11204 | .flags = CFTYPE_NOT_ON_ROOT, | |
11205 | .read_u64 = cpu_cfs_burst_read_u64, | |
11206 | .write_u64 = cpu_cfs_burst_write_u64, | |
11207 | }, | |
2480c093 PB |
11208 | #endif |
11209 | #ifdef CONFIG_UCLAMP_TASK_GROUP | |
11210 | { | |
11211 | .name = "uclamp.min", | |
11212 | .flags = CFTYPE_NOT_ON_ROOT, | |
11213 | .seq_show = cpu_uclamp_min_show, | |
11214 | .write = cpu_uclamp_min_write, | |
11215 | }, | |
11216 | { | |
11217 | .name = "uclamp.max", | |
11218 | .flags = CFTYPE_NOT_ON_ROOT, | |
11219 | .seq_show = cpu_uclamp_max_show, | |
11220 | .write = cpu_uclamp_max_write, | |
11221 | }, | |
0d593634 TH |
11222 | #endif |
11223 | { } /* terminate */ | |
11224 | }; | |
11225 | ||
073219e9 | 11226 | struct cgroup_subsys cpu_cgrp_subsys = { |
92fb9748 | 11227 | .css_alloc = cpu_cgroup_css_alloc, |
96b77745 | 11228 | .css_online = cpu_cgroup_css_online, |
2f5177f0 | 11229 | .css_released = cpu_cgroup_css_released, |
92fb9748 | 11230 | .css_free = cpu_cgroup_css_free, |
d41bf8c9 | 11231 | .css_extra_stat_show = cpu_extra_stat_show, |
df16b71c | 11232 | #ifdef CONFIG_RT_GROUP_SCHED |
bb9d97b6 | 11233 | .can_attach = cpu_cgroup_can_attach, |
df16b71c | 11234 | #endif |
bb9d97b6 | 11235 | .attach = cpu_cgroup_attach, |
a1f7164c | 11236 | .legacy_cftypes = cpu_legacy_files, |
0d593634 | 11237 | .dfl_cftypes = cpu_files, |
b38e42e9 | 11238 | .early_init = true, |
0d593634 | 11239 | .threaded = true, |
68318b8e SV |
11240 | }; |
11241 | ||
052f1dc7 | 11242 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 | 11243 | |
b637a328 PM |
11244 | void dump_cpu_task(int cpu) |
11245 | { | |
bc1cca97 ZL |
11246 | if (cpu == smp_processor_id() && in_hardirq()) { |
11247 | struct pt_regs *regs; | |
11248 | ||
11249 | regs = get_irq_regs(); | |
11250 | if (regs) { | |
11251 | show_regs(regs); | |
11252 | return; | |
11253 | } | |
11254 | } | |
11255 | ||
e73dfe30 ZL |
11256 | if (trigger_single_cpu_backtrace(cpu)) |
11257 | return; | |
11258 | ||
b637a328 PM |
11259 | pr_info("Task dump for CPU %d:\n", cpu); |
11260 | sched_show_task(cpu_curr(cpu)); | |
11261 | } | |
ed82b8a1 AK |
11262 | |
11263 | /* | |
11264 | * Nice levels are multiplicative, with a gentle 10% change for every | |
11265 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
11266 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
11267 | * that remained on nice 0. | |
11268 | * | |
11269 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
11270 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
11271 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. | |
11272 | * If a task goes up by ~10% and another task goes down by ~10% then | |
11273 | * the relative distance between them is ~25%.) | |
11274 | */ | |
11275 | const int sched_prio_to_weight[40] = { | |
11276 | /* -20 */ 88761, 71755, 56483, 46273, 36291, | |
11277 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
11278 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
11279 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
11280 | /* 0 */ 1024, 820, 655, 526, 423, | |
11281 | /* 5 */ 335, 272, 215, 172, 137, | |
11282 | /* 10 */ 110, 87, 70, 56, 45, | |
11283 | /* 15 */ 36, 29, 23, 18, 15, | |
11284 | }; | |
11285 | ||
11286 | /* | |
11287 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. | |
11288 | * | |
11289 | * In cases where the weight does not change often, we can use the | |
11290 | * precalculated inverse to speed up arithmetics by turning divisions | |
11291 | * into multiplications: | |
11292 | */ | |
11293 | const u32 sched_prio_to_wmult[40] = { | |
11294 | /* -20 */ 48388, 59856, 76040, 92818, 118348, | |
11295 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
11296 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
11297 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
11298 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
11299 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
11300 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
11301 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
11302 | }; | |
14a7405b | 11303 | |
9d246053 PA |
11304 | void call_trace_sched_update_nr_running(struct rq *rq, int count) |
11305 | { | |
11306 | trace_sched_update_nr_running_tp(rq, count); | |
11307 | } |