1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * kernel/sched/syscalls.c
4 *
5 * Core kernel scheduler syscalls related code
6 *
7 * Copyright (C) 1991-2002 Linus Torvalds
8 * Copyright (C) 1998-2024 Ingo Molnar, Red Hat
9 */
10 #include <linux/sched.h>
11 #include <linux/cpuset.h>
12 #include <linux/sched/debug.h>
13
14 #include <uapi/linux/sched/types.h>
15
16 #include "sched.h"
17 #include "autogroup.h"
18
__normal_prio(int policy,int rt_prio,int nice)19 static inline int __normal_prio(int policy, int rt_prio, int nice)
20 {
21 int prio;
22
23 if (dl_policy(policy))
24 prio = MAX_DL_PRIO - 1;
25 else if (rt_policy(policy))
26 prio = MAX_RT_PRIO - 1 - rt_prio;
27 else
28 prio = NICE_TO_PRIO(nice);
29
30 return prio;
31 }
32
33 /*
34 * Calculate the expected normal priority: i.e. priority
35 * without taking RT-inheritance into account. Might be
36 * boosted by interactivity modifiers. Changes upon fork,
37 * setprio syscalls, and whenever the interactivity
38 * estimator recalculates.
39 */
normal_prio(struct task_struct * p)40 static inline int normal_prio(struct task_struct *p)
41 {
42 return __normal_prio(p->policy, p->rt_priority, PRIO_TO_NICE(p->static_prio));
43 }
44
45 /*
46 * Calculate the current priority, i.e. the priority
47 * taken into account by the scheduler. This value might
48 * be boosted by RT tasks, or might be boosted by
49 * interactivity modifiers. Will be RT if the task got
50 * RT-boosted. If not then it returns p->normal_prio.
51 */
effective_prio(struct task_struct * p)52 static int effective_prio(struct task_struct *p)
53 {
54 p->normal_prio = normal_prio(p);
55 /*
56 * If we are RT tasks or we were boosted to RT priority,
57 * keep the priority unchanged. Otherwise, update priority
58 * to the normal priority:
59 */
60 if (!rt_or_dl_prio(p->prio))
61 return p->normal_prio;
62 return p->prio;
63 }
64
set_user_nice(struct task_struct * p,long nice)65 void set_user_nice(struct task_struct *p, long nice)
66 {
67 bool queued, running;
68 struct rq *rq;
69 int old_prio;
70
71 if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
72 return;
73 /*
74 * We have to be careful, if called from sys_setpriority(),
75 * the task might be in the middle of scheduling on another CPU.
76 */
77 CLASS(task_rq_lock, rq_guard)(p);
78 rq = rq_guard.rq;
79
80 update_rq_clock(rq);
81
82 /*
83 * The RT priorities are set via sched_setscheduler(), but we still
84 * allow the 'normal' nice value to be set - but as expected
85 * it won't have any effect on scheduling until the task is
86 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
87 */
88 if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
89 p->static_prio = NICE_TO_PRIO(nice);
90 return;
91 }
92
93 queued = task_on_rq_queued(p);
94 running = task_current_donor(rq, p);
95 if (queued)
96 dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
97 if (running)
98 put_prev_task(rq, p);
99
100 p->static_prio = NICE_TO_PRIO(nice);
101 set_load_weight(p, true);
102 old_prio = p->prio;
103 p->prio = effective_prio(p);
104
105 if (queued)
106 enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
107 if (running)
108 set_next_task(rq, p);
109
110 /*
111 * If the task increased its priority or is running and
112 * lowered its priority, then reschedule its CPU:
113 */
114 p->sched_class->prio_changed(rq, p, old_prio);
115 }
116 EXPORT_SYMBOL(set_user_nice);
117
118 /*
119 * is_nice_reduction - check if nice value is an actual reduction
120 *
121 * Similar to can_nice() but does not perform a capability check.
122 *
123 * @p: task
124 * @nice: nice value
125 */
is_nice_reduction(const struct task_struct * p,const int nice)126 static bool is_nice_reduction(const struct task_struct *p, const int nice)
127 {
128 /* Convert nice value [19,-20] to rlimit style value [1,40]: */
129 int nice_rlim = nice_to_rlimit(nice);
130
131 return (nice_rlim <= task_rlimit(p, RLIMIT_NICE));
132 }
133
134 /*
135 * can_nice - check if a task can reduce its nice value
136 * @p: task
137 * @nice: nice value
138 */
can_nice(const struct task_struct * p,const int nice)139 int can_nice(const struct task_struct *p, const int nice)
140 {
141 return is_nice_reduction(p, nice) || capable(CAP_SYS_NICE);
142 }
143
144 #ifdef __ARCH_WANT_SYS_NICE
145
146 /*
147 * sys_nice - change the priority of the current process.
148 * @increment: priority increment
149 *
150 * sys_setpriority is a more generic, but much slower function that
151 * does similar things.
152 */
SYSCALL_DEFINE1(nice,int,increment)153 SYSCALL_DEFINE1(nice, int, increment)
154 {
155 long nice, retval;
156
157 /*
158 * Setpriority might change our priority at the same moment.
159 * We don't have to worry. Conceptually one call occurs first
160 * and we have a single winner.
161 */
162 increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
163 nice = task_nice(current) + increment;
164
165 nice = clamp_val(nice, MIN_NICE, MAX_NICE);
166 if (increment < 0 && !can_nice(current, nice))
167 return -EPERM;
168
169 retval = security_task_setnice(current, nice);
170 if (retval)
171 return retval;
172
173 set_user_nice(current, nice);
174 return 0;
175 }
176
177 #endif
178
179 /**
180 * task_prio - return the priority value of a given task.
181 * @p: the task in question.
182 *
183 * Return: The priority value as seen by users in /proc.
184 *
185 * sched policy return value kernel prio user prio/nice
186 *
187 * normal, batch, idle [0 ... 39] [100 ... 139] 0/[-20 ... 19]
188 * fifo, rr [-2 ... -100] [98 ... 0] [1 ... 99]
189 * deadline -101 -1 0
190 */
task_prio(const struct task_struct * p)191 int task_prio(const struct task_struct *p)
192 {
193 return p->prio - MAX_RT_PRIO;
194 }
195
196 /**
197 * idle_cpu - is a given CPU idle currently?
198 * @cpu: the processor in question.
199 *
200 * Return: 1 if the CPU is currently idle. 0 otherwise.
201 */
idle_cpu(int cpu)202 int idle_cpu(int cpu)
203 {
204 struct rq *rq = cpu_rq(cpu);
205
206 if (rq->curr != rq->idle)
207 return 0;
208
209 if (rq->nr_running)
210 return 0;
211
212 #ifdef CONFIG_SMP
213 if (rq->ttwu_pending)
214 return 0;
215 #endif
216
217 return 1;
218 }
219
220 /**
221 * available_idle_cpu - is a given CPU idle for enqueuing work.
222 * @cpu: the CPU in question.
223 *
224 * Return: 1 if the CPU is currently idle. 0 otherwise.
225 */
available_idle_cpu(int cpu)226 int available_idle_cpu(int cpu)
227 {
228 if (!idle_cpu(cpu))
229 return 0;
230
231 if (vcpu_is_preempted(cpu))
232 return 0;
233
234 return 1;
235 }
236
237 /**
238 * idle_task - return the idle task for a given CPU.
239 * @cpu: the processor in question.
240 *
241 * Return: The idle task for the CPU @cpu.
242 */
idle_task(int cpu)243 struct task_struct *idle_task(int cpu)
244 {
245 return cpu_rq(cpu)->idle;
246 }
247
248 #ifdef CONFIG_SCHED_CORE
sched_core_idle_cpu(int cpu)249 int sched_core_idle_cpu(int cpu)
250 {
251 struct rq *rq = cpu_rq(cpu);
252
253 if (sched_core_enabled(rq) && rq->curr == rq->idle)
254 return 1;
255
256 return idle_cpu(cpu);
257 }
258
259 #endif
260
261 /**
262 * find_process_by_pid - find a process with a matching PID value.
263 * @pid: the pid in question.
264 *
265 * The task of @pid, if found. %NULL otherwise.
266 */
find_process_by_pid(pid_t pid)267 static struct task_struct *find_process_by_pid(pid_t pid)
268 {
269 return pid ? find_task_by_vpid(pid) : current;
270 }
271
find_get_task(pid_t pid)272 static struct task_struct *find_get_task(pid_t pid)
273 {
274 struct task_struct *p;
275 guard(rcu)();
276
277 p = find_process_by_pid(pid);
278 if (likely(p))
279 get_task_struct(p);
280
281 return p;
282 }
283
DEFINE_CLASS(find_get_task,struct task_struct *,if (_T)put_task_struct (_T),find_get_task (pid),pid_t pid)284 DEFINE_CLASS(find_get_task, struct task_struct *, if (_T) put_task_struct(_T),
285 find_get_task(pid), pid_t pid)
286
287 /*
288 * sched_setparam() passes in -1 for its policy, to let the functions
289 * it calls know not to change it.
290 */
291 #define SETPARAM_POLICY -1
292
293 static void __setscheduler_params(struct task_struct *p,
294 const struct sched_attr *attr)
295 {
296 int policy = attr->sched_policy;
297
298 if (policy == SETPARAM_POLICY)
299 policy = p->policy;
300
301 p->policy = policy;
302
303 if (dl_policy(policy)) {
304 __setparam_dl(p, attr);
305 } else if (fair_policy(policy)) {
306 p->static_prio = NICE_TO_PRIO(attr->sched_nice);
307 if (attr->sched_runtime) {
308 p->se.custom_slice = 1;
309 p->se.slice = clamp_t(u64, attr->sched_runtime,
310 NSEC_PER_MSEC/10, /* HZ=1000 * 10 */
311 NSEC_PER_MSEC*100); /* HZ=100 / 10 */
312 } else {
313 p->se.custom_slice = 0;
314 p->se.slice = sysctl_sched_base_slice;
315 }
316 }
317
318 /* rt-policy tasks do not have a timerslack */
319 if (rt_or_dl_task_policy(p)) {
320 p->timer_slack_ns = 0;
321 } else if (p->timer_slack_ns == 0) {
322 /* when switching back to non-rt policy, restore timerslack */
323 p->timer_slack_ns = p->default_timer_slack_ns;
324 }
325
326 /*
327 * __sched_setscheduler() ensures attr->sched_priority == 0 when
328 * !rt_policy. Always setting this ensures that things like
329 * getparam()/getattr() don't report silly values for !rt tasks.
330 */
331 p->rt_priority = attr->sched_priority;
332 p->normal_prio = normal_prio(p);
333 set_load_weight(p, true);
334 }
335
336 /*
337 * Check the target process has a UID that matches the current process's:
338 */
check_same_owner(struct task_struct * p)339 static bool check_same_owner(struct task_struct *p)
340 {
341 const struct cred *cred = current_cred(), *pcred;
342 guard(rcu)();
343
344 pcred = __task_cred(p);
345 return (uid_eq(cred->euid, pcred->euid) ||
346 uid_eq(cred->euid, pcred->uid));
347 }
348
349 #ifdef CONFIG_UCLAMP_TASK
350
uclamp_validate(struct task_struct * p,const struct sched_attr * attr)351 static int uclamp_validate(struct task_struct *p,
352 const struct sched_attr *attr)
353 {
354 int util_min = p->uclamp_req[UCLAMP_MIN].value;
355 int util_max = p->uclamp_req[UCLAMP_MAX].value;
356
357 if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) {
358 util_min = attr->sched_util_min;
359
360 if (util_min + 1 > SCHED_CAPACITY_SCALE + 1)
361 return -EINVAL;
362 }
363
364 if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) {
365 util_max = attr->sched_util_max;
366
367 if (util_max + 1 > SCHED_CAPACITY_SCALE + 1)
368 return -EINVAL;
369 }
370
371 if (util_min != -1 && util_max != -1 && util_min > util_max)
372 return -EINVAL;
373
374 /*
375 * We have valid uclamp attributes; make sure uclamp is enabled.
376 *
377 * We need to do that here, because enabling static branches is a
378 * blocking operation which obviously cannot be done while holding
379 * scheduler locks.
380 */
381 static_branch_enable(&sched_uclamp_used);
382
383 return 0;
384 }
385
uclamp_reset(const struct sched_attr * attr,enum uclamp_id clamp_id,struct uclamp_se * uc_se)386 static bool uclamp_reset(const struct sched_attr *attr,
387 enum uclamp_id clamp_id,
388 struct uclamp_se *uc_se)
389 {
390 /* Reset on sched class change for a non user-defined clamp value. */
391 if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) &&
392 !uc_se->user_defined)
393 return true;
394
395 /* Reset on sched_util_{min,max} == -1. */
396 if (clamp_id == UCLAMP_MIN &&
397 attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN &&
398 attr->sched_util_min == -1) {
399 return true;
400 }
401
402 if (clamp_id == UCLAMP_MAX &&
403 attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX &&
404 attr->sched_util_max == -1) {
405 return true;
406 }
407
408 return false;
409 }
410
__setscheduler_uclamp(struct task_struct * p,const struct sched_attr * attr)411 static void __setscheduler_uclamp(struct task_struct *p,
412 const struct sched_attr *attr)
413 {
414 enum uclamp_id clamp_id;
415
416 for_each_clamp_id(clamp_id) {
417 struct uclamp_se *uc_se = &p->uclamp_req[clamp_id];
418 unsigned int value;
419
420 if (!uclamp_reset(attr, clamp_id, uc_se))
421 continue;
422
423 /*
424 * RT by default have a 100% boost value that could be modified
425 * at runtime.
426 */
427 if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
428 value = sysctl_sched_uclamp_util_min_rt_default;
429 else
430 value = uclamp_none(clamp_id);
431
432 uclamp_se_set(uc_se, value, false);
433
434 }
435
436 if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)))
437 return;
438
439 if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN &&
440 attr->sched_util_min != -1) {
441 uclamp_se_set(&p->uclamp_req[UCLAMP_MIN],
442 attr->sched_util_min, true);
443 }
444
445 if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX &&
446 attr->sched_util_max != -1) {
447 uclamp_se_set(&p->uclamp_req[UCLAMP_MAX],
448 attr->sched_util_max, true);
449 }
450 }
451
452 #else /* !CONFIG_UCLAMP_TASK: */
453
uclamp_validate(struct task_struct * p,const struct sched_attr * attr)454 static inline int uclamp_validate(struct task_struct *p,
455 const struct sched_attr *attr)
456 {
457 return -EOPNOTSUPP;
458 }
__setscheduler_uclamp(struct task_struct * p,const struct sched_attr * attr)459 static void __setscheduler_uclamp(struct task_struct *p,
460 const struct sched_attr *attr) { }
461 #endif
462
463 /*
464 * Allow unprivileged RT tasks to decrease priority.
465 * Only issue a capable test if needed and only once to avoid an audit
466 * event on permitted non-privileged operations:
467 */
user_check_sched_setscheduler(struct task_struct * p,const struct sched_attr * attr,int policy,int reset_on_fork)468 static int user_check_sched_setscheduler(struct task_struct *p,
469 const struct sched_attr *attr,
470 int policy, int reset_on_fork)
471 {
472 if (fair_policy(policy)) {
473 if (attr->sched_nice < task_nice(p) &&
474 !is_nice_reduction(p, attr->sched_nice))
475 goto req_priv;
476 }
477
478 if (rt_policy(policy)) {
479 unsigned long rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO);
480
481 /* Can't set/change the rt policy: */
482 if (policy != p->policy && !rlim_rtprio)
483 goto req_priv;
484
485 /* Can't increase priority: */
486 if (attr->sched_priority > p->rt_priority &&
487 attr->sched_priority > rlim_rtprio)
488 goto req_priv;
489 }
490
491 /*
492 * Can't set/change SCHED_DEADLINE policy at all for now
493 * (safest behavior); in the future we would like to allow
494 * unprivileged DL tasks to increase their relative deadline
495 * or reduce their runtime (both ways reducing utilization)
496 */
497 if (dl_policy(policy))
498 goto req_priv;
499
500 /*
501 * Treat SCHED_IDLE as nice 20. Only allow a switch to
502 * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
503 */
504 if (task_has_idle_policy(p) && !idle_policy(policy)) {
505 if (!is_nice_reduction(p, task_nice(p)))
506 goto req_priv;
507 }
508
509 /* Can't change other user's priorities: */
510 if (!check_same_owner(p))
511 goto req_priv;
512
513 /* Normal users shall not reset the sched_reset_on_fork flag: */
514 if (p->sched_reset_on_fork && !reset_on_fork)
515 goto req_priv;
516
517 return 0;
518
519 req_priv:
520 if (!capable(CAP_SYS_NICE))
521 return -EPERM;
522
523 return 0;
524 }
525
__sched_setscheduler(struct task_struct * p,const struct sched_attr * attr,bool user,bool pi)526 int __sched_setscheduler(struct task_struct *p,
527 const struct sched_attr *attr,
528 bool user, bool pi)
529 {
530 int oldpolicy = -1, policy = attr->sched_policy;
531 int retval, oldprio, newprio, queued, running;
532 const struct sched_class *prev_class, *next_class;
533 struct balance_callback *head;
534 struct rq_flags rf;
535 int reset_on_fork;
536 int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
537 struct rq *rq;
538 bool cpuset_locked = false;
539
540 /* The pi code expects interrupts enabled */
541 BUG_ON(pi && in_interrupt());
542 recheck:
543 /* Double check policy once rq lock held: */
544 if (policy < 0) {
545 reset_on_fork = p->sched_reset_on_fork;
546 policy = oldpolicy = p->policy;
547 } else {
548 reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
549
550 if (!valid_policy(policy))
551 return -EINVAL;
552 }
553
554 if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
555 return -EINVAL;
556
557 /*
558 * Valid priorities for SCHED_FIFO and SCHED_RR are
559 * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL,
560 * SCHED_BATCH and SCHED_IDLE is 0.
561 */
562 if (attr->sched_priority > MAX_RT_PRIO-1)
563 return -EINVAL;
564 if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
565 (rt_policy(policy) != (attr->sched_priority != 0)))
566 return -EINVAL;
567
568 if (user) {
569 retval = user_check_sched_setscheduler(p, attr, policy, reset_on_fork);
570 if (retval)
571 return retval;
572
573 if (attr->sched_flags & SCHED_FLAG_SUGOV)
574 return -EINVAL;
575
576 retval = security_task_setscheduler(p);
577 if (retval)
578 return retval;
579 }
580
581 /* Update task specific "requested" clamps */
582 if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) {
583 retval = uclamp_validate(p, attr);
584 if (retval)
585 return retval;
586 }
587
588 /*
589 * SCHED_DEADLINE bandwidth accounting relies on stable cpusets
590 * information.
591 */
592 if (dl_policy(policy) || dl_policy(p->policy)) {
593 cpuset_locked = true;
594 cpuset_lock();
595 }
596
597 /*
598 * Make sure no PI-waiters arrive (or leave) while we are
599 * changing the priority of the task:
600 *
601 * To be able to change p->policy safely, the appropriate
602 * runqueue lock must be held.
603 */
604 rq = task_rq_lock(p, &rf);
605 update_rq_clock(rq);
606
607 /*
608 * Changing the policy of the stop threads its a very bad idea:
609 */
610 if (p == rq->stop) {
611 retval = -EINVAL;
612 goto unlock;
613 }
614
615 retval = scx_check_setscheduler(p, policy);
616 if (retval)
617 goto unlock;
618
619 /*
620 * If not changing anything there's no need to proceed further,
621 * but store a possible modification of reset_on_fork.
622 */
623 if (unlikely(policy == p->policy)) {
624 if (fair_policy(policy) &&
625 (attr->sched_nice != task_nice(p) ||
626 (attr->sched_runtime != p->se.slice)))
627 goto change;
628 if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
629 goto change;
630 if (dl_policy(policy) && dl_param_changed(p, attr))
631 goto change;
632 if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)
633 goto change;
634
635 p->sched_reset_on_fork = reset_on_fork;
636 retval = 0;
637 goto unlock;
638 }
639 change:
640
641 if (user) {
642 #ifdef CONFIG_RT_GROUP_SCHED
643 /*
644 * Do not allow real-time tasks into groups that have no runtime
645 * assigned.
646 */
647 if (rt_bandwidth_enabled() && rt_policy(policy) &&
648 task_group(p)->rt_bandwidth.rt_runtime == 0 &&
649 !task_group_is_autogroup(task_group(p))) {
650 retval = -EPERM;
651 goto unlock;
652 }
653 #endif
654 #ifdef CONFIG_SMP
655 if (dl_bandwidth_enabled() && dl_policy(policy) &&
656 !(attr->sched_flags & SCHED_FLAG_SUGOV)) {
657 cpumask_t *span = rq->rd->span;
658
659 /*
660 * Don't allow tasks with an affinity mask smaller than
661 * the entire root_domain to become SCHED_DEADLINE. We
662 * will also fail if there's no bandwidth available.
663 */
664 if (!cpumask_subset(span, p->cpus_ptr) ||
665 rq->rd->dl_bw.bw == 0) {
666 retval = -EPERM;
667 goto unlock;
668 }
669 }
670 #endif
671 }
672
673 /* Re-check policy now with rq lock held: */
674 if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
675 policy = oldpolicy = -1;
676 task_rq_unlock(rq, p, &rf);
677 if (cpuset_locked)
678 cpuset_unlock();
679 goto recheck;
680 }
681
682 /*
683 * If setscheduling to SCHED_DEADLINE (or changing the parameters
684 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
685 * is available.
686 */
687 if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
688 retval = -EBUSY;
689 goto unlock;
690 }
691
692 p->sched_reset_on_fork = reset_on_fork;
693 oldprio = p->prio;
694
695 newprio = __normal_prio(policy, attr->sched_priority, attr->sched_nice);
696 if (pi) {
697 /*
698 * Take priority boosted tasks into account. If the new
699 * effective priority is unchanged, we just store the new
700 * normal parameters and do not touch the scheduler class and
701 * the runqueue. This will be done when the task deboost
702 * itself.
703 */
704 newprio = rt_effective_prio(p, newprio);
705 if (newprio == oldprio)
706 queue_flags &= ~DEQUEUE_MOVE;
707 }
708
709 prev_class = p->sched_class;
710 next_class = __setscheduler_class(policy, newprio);
711
712 if (prev_class != next_class && p->se.sched_delayed)
713 dequeue_task(rq, p, DEQUEUE_SLEEP | DEQUEUE_DELAYED | DEQUEUE_NOCLOCK);
714
715 queued = task_on_rq_queued(p);
716 running = task_current_donor(rq, p);
717 if (queued)
718 dequeue_task(rq, p, queue_flags);
719 if (running)
720 put_prev_task(rq, p);
721
722 if (!(attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)) {
723 __setscheduler_params(p, attr);
724 p->sched_class = next_class;
725 p->prio = newprio;
726 }
727 __setscheduler_uclamp(p, attr);
728 check_class_changing(rq, p, prev_class);
729
730 if (queued) {
731 /*
732 * We enqueue to tail when the priority of a task is
733 * increased (user space view).
734 */
735 if (oldprio < p->prio)
736 queue_flags |= ENQUEUE_HEAD;
737
738 enqueue_task(rq, p, queue_flags);
739 }
740 if (running)
741 set_next_task(rq, p);
742
743 check_class_changed(rq, p, prev_class, oldprio);
744
745 /* Avoid rq from going away on us: */
746 preempt_disable();
747 head = splice_balance_callbacks(rq);
748 task_rq_unlock(rq, p, &rf);
749
750 if (pi) {
751 if (cpuset_locked)
752 cpuset_unlock();
753 rt_mutex_adjust_pi(p);
754 }
755
756 /* Run balance callbacks after we've adjusted the PI chain: */
757 balance_callbacks(rq, head);
758 preempt_enable();
759
760 return 0;
761
762 unlock:
763 task_rq_unlock(rq, p, &rf);
764 if (cpuset_locked)
765 cpuset_unlock();
766 return retval;
767 }
768
_sched_setscheduler(struct task_struct * p,int policy,const struct sched_param * param,bool check)769 static int _sched_setscheduler(struct task_struct *p, int policy,
770 const struct sched_param *param, bool check)
771 {
772 struct sched_attr attr = {
773 .sched_policy = policy,
774 .sched_priority = param->sched_priority,
775 .sched_nice = PRIO_TO_NICE(p->static_prio),
776 };
777
778 if (p->se.custom_slice)
779 attr.sched_runtime = p->se.slice;
780
781 /* Fixup the legacy SCHED_RESET_ON_FORK hack. */
782 if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
783 attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
784 policy &= ~SCHED_RESET_ON_FORK;
785 attr.sched_policy = policy;
786 }
787
788 return __sched_setscheduler(p, &attr, check, true);
789 }
790 /**
791 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
792 * @p: the task in question.
793 * @policy: new policy.
794 * @param: structure containing the new RT priority.
795 *
796 * Use sched_set_fifo(), read its comment.
797 *
798 * Return: 0 on success. An error code otherwise.
799 *
800 * NOTE that the task may be already dead.
801 */
sched_setscheduler(struct task_struct * p,int policy,const struct sched_param * param)802 int sched_setscheduler(struct task_struct *p, int policy,
803 const struct sched_param *param)
804 {
805 return _sched_setscheduler(p, policy, param, true);
806 }
807
sched_setattr(struct task_struct * p,const struct sched_attr * attr)808 int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
809 {
810 return __sched_setscheduler(p, attr, true, true);
811 }
812
sched_setattr_nocheck(struct task_struct * p,const struct sched_attr * attr)813 int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
814 {
815 return __sched_setscheduler(p, attr, false, true);
816 }
817 EXPORT_SYMBOL_GPL(sched_setattr_nocheck);
818
819 /**
820 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernel-space.
821 * @p: the task in question.
822 * @policy: new policy.
823 * @param: structure containing the new RT priority.
824 *
825 * Just like sched_setscheduler, only don't bother checking if the
826 * current context has permission. For example, this is needed in
827 * stop_machine(): we create temporary high priority worker threads,
828 * but our caller might not have that capability.
829 *
830 * Return: 0 on success. An error code otherwise.
831 */
sched_setscheduler_nocheck(struct task_struct * p,int policy,const struct sched_param * param)832 int sched_setscheduler_nocheck(struct task_struct *p, int policy,
833 const struct sched_param *param)
834 {
835 return _sched_setscheduler(p, policy, param, false);
836 }
837
838 /*
839 * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally
840 * incapable of resource management, which is the one thing an OS really should
841 * be doing.
842 *
843 * This is of course the reason it is limited to privileged users only.
844 *
845 * Worse still; it is fundamentally impossible to compose static priority
846 * workloads. You cannot take two correctly working static prio workloads
847 * and smash them together and still expect them to work.
848 *
849 * For this reason 'all' FIFO tasks the kernel creates are basically at:
850 *
851 * MAX_RT_PRIO / 2
852 *
853 * The administrator _MUST_ configure the system, the kernel simply doesn't
854 * know enough information to make a sensible choice.
855 */
sched_set_fifo(struct task_struct * p)856 void sched_set_fifo(struct task_struct *p)
857 {
858 struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 };
859 WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0);
860 }
861 EXPORT_SYMBOL_GPL(sched_set_fifo);
862
863 /*
864 * For when you don't much care about FIFO, but want to be above SCHED_NORMAL.
865 */
sched_set_fifo_low(struct task_struct * p)866 void sched_set_fifo_low(struct task_struct *p)
867 {
868 struct sched_param sp = { .sched_priority = 1 };
869 WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0);
870 }
871 EXPORT_SYMBOL_GPL(sched_set_fifo_low);
872
sched_set_normal(struct task_struct * p,int nice)873 void sched_set_normal(struct task_struct *p, int nice)
874 {
875 struct sched_attr attr = {
876 .sched_policy = SCHED_NORMAL,
877 .sched_nice = nice,
878 };
879 WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0);
880 }
881 EXPORT_SYMBOL_GPL(sched_set_normal);
882
883 static int
do_sched_setscheduler(pid_t pid,int policy,struct sched_param __user * param)884 do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
885 {
886 struct sched_param lparam;
887
888 if (!param || pid < 0)
889 return -EINVAL;
890 if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
891 return -EFAULT;
892
893 CLASS(find_get_task, p)(pid);
894 if (!p)
895 return -ESRCH;
896
897 return sched_setscheduler(p, policy, &lparam);
898 }
899
900 /*
901 * Mimics kernel/events/core.c perf_copy_attr().
902 */
sched_copy_attr(struct sched_attr __user * uattr,struct sched_attr * attr)903 static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
904 {
905 u32 size;
906 int ret;
907
908 /* Zero the full structure, so that a short copy will be nice: */
909 memset(attr, 0, sizeof(*attr));
910
911 ret = get_user(size, &uattr->size);
912 if (ret)
913 return ret;
914
915 /* ABI compatibility quirk: */
916 if (!size)
917 size = SCHED_ATTR_SIZE_VER0;
918 if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE)
919 goto err_size;
920
921 ret = copy_struct_from_user(attr, sizeof(*attr), uattr, size);
922 if (ret) {
923 if (ret == -E2BIG)
924 goto err_size;
925 return ret;
926 }
927
928 if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) &&
929 size < SCHED_ATTR_SIZE_VER1)
930 return -EINVAL;
931
932 /*
933 * XXX: Do we want to be lenient like existing syscalls; or do we want
934 * to be strict and return an error on out-of-bounds values?
935 */
936 attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
937
938 return 0;
939
940 err_size:
941 put_user(sizeof(*attr), &uattr->size);
942 return -E2BIG;
943 }
944
get_params(struct task_struct * p,struct sched_attr * attr)945 static void get_params(struct task_struct *p, struct sched_attr *attr)
946 {
947 if (task_has_dl_policy(p)) {
948 __getparam_dl(p, attr);
949 } else if (task_has_rt_policy(p)) {
950 attr->sched_priority = p->rt_priority;
951 } else {
952 attr->sched_nice = task_nice(p);
953 attr->sched_runtime = p->se.slice;
954 }
955 }
956
957 /**
958 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
959 * @pid: the pid in question.
960 * @policy: new policy.
961 * @param: structure containing the new RT priority.
962 *
963 * Return: 0 on success. An error code otherwise.
964 */
SYSCALL_DEFINE3(sched_setscheduler,pid_t,pid,int,policy,struct sched_param __user *,param)965 SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
966 {
967 if (policy < 0)
968 return -EINVAL;
969
970 return do_sched_setscheduler(pid, policy, param);
971 }
972
973 /**
974 * sys_sched_setparam - set/change the RT priority of a thread
975 * @pid: the pid in question.
976 * @param: structure containing the new RT priority.
977 *
978 * Return: 0 on success. An error code otherwise.
979 */
SYSCALL_DEFINE2(sched_setparam,pid_t,pid,struct sched_param __user *,param)980 SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
981 {
982 return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
983 }
984
985 /**
986 * sys_sched_setattr - same as above, but with extended sched_attr
987 * @pid: the pid in question.
988 * @uattr: structure containing the extended parameters.
989 * @flags: for future extension.
990 */
SYSCALL_DEFINE3(sched_setattr,pid_t,pid,struct sched_attr __user *,uattr,unsigned int,flags)991 SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
992 unsigned int, flags)
993 {
994 struct sched_attr attr;
995 int retval;
996
997 if (!uattr || pid < 0 || flags)
998 return -EINVAL;
999
1000 retval = sched_copy_attr(uattr, &attr);
1001 if (retval)
1002 return retval;
1003
1004 if ((int)attr.sched_policy < 0)
1005 return -EINVAL;
1006 if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY)
1007 attr.sched_policy = SETPARAM_POLICY;
1008
1009 CLASS(find_get_task, p)(pid);
1010 if (!p)
1011 return -ESRCH;
1012
1013 if (attr.sched_flags & SCHED_FLAG_KEEP_PARAMS)
1014 get_params(p, &attr);
1015
1016 return sched_setattr(p, &attr);
1017 }
1018
1019 /**
1020 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
1021 * @pid: the pid in question.
1022 *
1023 * Return: On success, the policy of the thread. Otherwise, a negative error
1024 * code.
1025 */
SYSCALL_DEFINE1(sched_getscheduler,pid_t,pid)1026 SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
1027 {
1028 struct task_struct *p;
1029 int retval;
1030
1031 if (pid < 0)
1032 return -EINVAL;
1033
1034 guard(rcu)();
1035 p = find_process_by_pid(pid);
1036 if (!p)
1037 return -ESRCH;
1038
1039 retval = security_task_getscheduler(p);
1040 if (!retval) {
1041 retval = p->policy;
1042 if (p->sched_reset_on_fork)
1043 retval |= SCHED_RESET_ON_FORK;
1044 }
1045 return retval;
1046 }
1047
1048 /**
1049 * sys_sched_getparam - get the RT priority of a thread
1050 * @pid: the pid in question.
1051 * @param: structure containing the RT priority.
1052 *
1053 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
1054 * code.
1055 */
SYSCALL_DEFINE2(sched_getparam,pid_t,pid,struct sched_param __user *,param)1056 SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
1057 {
1058 struct sched_param lp = { .sched_priority = 0 };
1059 struct task_struct *p;
1060 int retval;
1061
1062 if (!param || pid < 0)
1063 return -EINVAL;
1064
1065 scoped_guard (rcu) {
1066 p = find_process_by_pid(pid);
1067 if (!p)
1068 return -ESRCH;
1069
1070 retval = security_task_getscheduler(p);
1071 if (retval)
1072 return retval;
1073
1074 if (task_has_rt_policy(p))
1075 lp.sched_priority = p->rt_priority;
1076 }
1077
1078 /*
1079 * This one might sleep, we cannot do it with a spinlock held ...
1080 */
1081 return copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;
1082 }
1083
1084 /**
1085 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
1086 * @pid: the pid in question.
1087 * @uattr: structure containing the extended parameters.
1088 * @usize: sizeof(attr) for fwd/bwd comp.
1089 * @flags: for future extension.
1090 */
SYSCALL_DEFINE4(sched_getattr,pid_t,pid,struct sched_attr __user *,uattr,unsigned int,usize,unsigned int,flags)1091 SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
1092 unsigned int, usize, unsigned int, flags)
1093 {
1094 struct sched_attr kattr = { };
1095 struct task_struct *p;
1096 int retval;
1097
1098 if (!uattr || pid < 0 || usize > PAGE_SIZE ||
1099 usize < SCHED_ATTR_SIZE_VER0 || flags)
1100 return -EINVAL;
1101
1102 scoped_guard (rcu) {
1103 p = find_process_by_pid(pid);
1104 if (!p)
1105 return -ESRCH;
1106
1107 retval = security_task_getscheduler(p);
1108 if (retval)
1109 return retval;
1110
1111 kattr.sched_policy = p->policy;
1112 if (p->sched_reset_on_fork)
1113 kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
1114 get_params(p, &kattr);
1115 kattr.sched_flags &= SCHED_FLAG_ALL;
1116
1117 #ifdef CONFIG_UCLAMP_TASK
1118 /*
1119 * This could race with another potential updater, but this is fine
1120 * because it'll correctly read the old or the new value. We don't need
1121 * to guarantee who wins the race as long as it doesn't return garbage.
1122 */
1123 kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value;
1124 kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value;
1125 #endif
1126 }
1127
1128 kattr.size = min(usize, sizeof(kattr));
1129 return copy_struct_to_user(uattr, usize, &kattr, sizeof(kattr), NULL);
1130 }
1131
1132 #ifdef CONFIG_SMP
dl_task_check_affinity(struct task_struct * p,const struct cpumask * mask)1133 int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1134 {
1135 /*
1136 * If the task isn't a deadline task or admission control is
1137 * disabled then we don't care about affinity changes.
1138 */
1139 if (!task_has_dl_policy(p) || !dl_bandwidth_enabled())
1140 return 0;
1141
1142 /*
1143 * Since bandwidth control happens on root_domain basis,
1144 * if admission test is enabled, we only admit -deadline
1145 * tasks allowed to run on all the CPUs in the task's
1146 * root_domain.
1147 */
1148 guard(rcu)();
1149 if (!cpumask_subset(task_rq(p)->rd->span, mask))
1150 return -EBUSY;
1151
1152 return 0;
1153 }
1154 #endif /* CONFIG_SMP */
1155
__sched_setaffinity(struct task_struct * p,struct affinity_context * ctx)1156 int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx)
1157 {
1158 int retval;
1159 cpumask_var_t cpus_allowed, new_mask;
1160
1161 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL))
1162 return -ENOMEM;
1163
1164 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
1165 retval = -ENOMEM;
1166 goto out_free_cpus_allowed;
1167 }
1168
1169 cpuset_cpus_allowed(p, cpus_allowed);
1170 cpumask_and(new_mask, ctx->new_mask, cpus_allowed);
1171
1172 ctx->new_mask = new_mask;
1173 ctx->flags |= SCA_CHECK;
1174
1175 retval = dl_task_check_affinity(p, new_mask);
1176 if (retval)
1177 goto out_free_new_mask;
1178
1179 retval = __set_cpus_allowed_ptr(p, ctx);
1180 if (retval)
1181 goto out_free_new_mask;
1182
1183 cpuset_cpus_allowed(p, cpus_allowed);
1184 if (!cpumask_subset(new_mask, cpus_allowed)) {
1185 /*
1186 * We must have raced with a concurrent cpuset update.
1187 * Just reset the cpumask to the cpuset's cpus_allowed.
1188 */
1189 cpumask_copy(new_mask, cpus_allowed);
1190
1191 /*
1192 * If SCA_USER is set, a 2nd call to __set_cpus_allowed_ptr()
1193 * will restore the previous user_cpus_ptr value.
1194 *
1195 * In the unlikely event a previous user_cpus_ptr exists,
1196 * we need to further restrict the mask to what is allowed
1197 * by that old user_cpus_ptr.
1198 */
1199 if (unlikely((ctx->flags & SCA_USER) && ctx->user_mask)) {
1200 bool empty = !cpumask_and(new_mask, new_mask,
1201 ctx->user_mask);
1202
1203 if (empty)
1204 cpumask_copy(new_mask, cpus_allowed);
1205 }
1206 __set_cpus_allowed_ptr(p, ctx);
1207 retval = -EINVAL;
1208 }
1209
1210 out_free_new_mask:
1211 free_cpumask_var(new_mask);
1212 out_free_cpus_allowed:
1213 free_cpumask_var(cpus_allowed);
1214 return retval;
1215 }
1216
sched_setaffinity(pid_t pid,const struct cpumask * in_mask)1217 long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
1218 {
1219 struct affinity_context ac;
1220 struct cpumask *user_mask;
1221 int retval;
1222
1223 CLASS(find_get_task, p)(pid);
1224 if (!p)
1225 return -ESRCH;
1226
1227 if (p->flags & PF_NO_SETAFFINITY)
1228 return -EINVAL;
1229
1230 if (!check_same_owner(p)) {
1231 guard(rcu)();
1232 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1233 return -EPERM;
1234 }
1235
1236 retval = security_task_setscheduler(p);
1237 if (retval)
1238 return retval;
1239
1240 /*
1241 * With non-SMP configs, user_cpus_ptr/user_mask isn't used and
1242 * alloc_user_cpus_ptr() returns NULL.
1243 */
1244 user_mask = alloc_user_cpus_ptr(NUMA_NO_NODE);
1245 if (user_mask) {
1246 cpumask_copy(user_mask, in_mask);
1247 } else if (IS_ENABLED(CONFIG_SMP)) {
1248 return -ENOMEM;
1249 }
1250
1251 ac = (struct affinity_context){
1252 .new_mask = in_mask,
1253 .user_mask = user_mask,
1254 .flags = SCA_USER,
1255 };
1256
1257 retval = __sched_setaffinity(p, &ac);
1258 kfree(ac.user_mask);
1259
1260 return retval;
1261 }
1262
get_user_cpu_mask(unsigned long __user * user_mask_ptr,unsigned len,struct cpumask * new_mask)1263 static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
1264 struct cpumask *new_mask)
1265 {
1266 if (len < cpumask_size())
1267 cpumask_clear(new_mask);
1268 else if (len > cpumask_size())
1269 len = cpumask_size();
1270
1271 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
1272 }
1273
1274 /**
1275 * sys_sched_setaffinity - set the CPU affinity of a process
1276 * @pid: pid of the process
1277 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
1278 * @user_mask_ptr: user-space pointer to the new CPU mask
1279 *
1280 * Return: 0 on success. An error code otherwise.
1281 */
SYSCALL_DEFINE3(sched_setaffinity,pid_t,pid,unsigned int,len,unsigned long __user *,user_mask_ptr)1282 SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
1283 unsigned long __user *, user_mask_ptr)
1284 {
1285 cpumask_var_t new_mask;
1286 int retval;
1287
1288 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
1289 return -ENOMEM;
1290
1291 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
1292 if (retval == 0)
1293 retval = sched_setaffinity(pid, new_mask);
1294 free_cpumask_var(new_mask);
1295 return retval;
1296 }
1297
sched_getaffinity(pid_t pid,struct cpumask * mask)1298 long sched_getaffinity(pid_t pid, struct cpumask *mask)
1299 {
1300 struct task_struct *p;
1301 int retval;
1302
1303 guard(rcu)();
1304 p = find_process_by_pid(pid);
1305 if (!p)
1306 return -ESRCH;
1307
1308 retval = security_task_getscheduler(p);
1309 if (retval)
1310 return retval;
1311
1312 guard(raw_spinlock_irqsave)(&p->pi_lock);
1313 cpumask_and(mask, &p->cpus_mask, cpu_active_mask);
1314
1315 return 0;
1316 }
1317
1318 /**
1319 * sys_sched_getaffinity - get the CPU affinity of a process
1320 * @pid: pid of the process
1321 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
1322 * @user_mask_ptr: user-space pointer to hold the current CPU mask
1323 *
1324 * Return: size of CPU mask copied to user_mask_ptr on success. An
1325 * error code otherwise.
1326 */
SYSCALL_DEFINE3(sched_getaffinity,pid_t,pid,unsigned int,len,unsigned long __user *,user_mask_ptr)1327 SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
1328 unsigned long __user *, user_mask_ptr)
1329 {
1330 int ret;
1331 cpumask_var_t mask;
1332
1333 if ((len * BITS_PER_BYTE) < nr_cpu_ids)
1334 return -EINVAL;
1335 if (len & (sizeof(unsigned long)-1))
1336 return -EINVAL;
1337
1338 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1339 return -ENOMEM;
1340
1341 ret = sched_getaffinity(pid, mask);
1342 if (ret == 0) {
1343 unsigned int retlen = min(len, cpumask_size());
1344
1345 if (copy_to_user(user_mask_ptr, cpumask_bits(mask), retlen))
1346 ret = -EFAULT;
1347 else
1348 ret = retlen;
1349 }
1350 free_cpumask_var(mask);
1351
1352 return ret;
1353 }
1354
do_sched_yield(void)1355 static void do_sched_yield(void)
1356 {
1357 struct rq_flags rf;
1358 struct rq *rq;
1359
1360 rq = this_rq_lock_irq(&rf);
1361
1362 schedstat_inc(rq->yld_count);
1363 current->sched_class->yield_task(rq);
1364
1365 preempt_disable();
1366 rq_unlock_irq(rq, &rf);
1367 sched_preempt_enable_no_resched();
1368
1369 schedule();
1370 }
1371
1372 /**
1373 * sys_sched_yield - yield the current processor to other threads.
1374 *
1375 * This function yields the current CPU to other tasks. If there are no
1376 * other threads running on this CPU then this function will return.
1377 *
1378 * Return: 0.
1379 */
SYSCALL_DEFINE0(sched_yield)1380 SYSCALL_DEFINE0(sched_yield)
1381 {
1382 do_sched_yield();
1383 return 0;
1384 }
1385
1386 /**
1387 * yield - yield the current processor to other threads.
1388 *
1389 * Do not ever use this function, there's a 99% chance you're doing it wrong.
1390 *
1391 * The scheduler is at all times free to pick the calling task as the most
1392 * eligible task to run, if removing the yield() call from your code breaks
1393 * it, it's already broken.
1394 *
1395 * Typical broken usage is:
1396 *
1397 * while (!event)
1398 * yield();
1399 *
1400 * where one assumes that yield() will let 'the other' process run that will
1401 * make event true. If the current task is a SCHED_FIFO task that will never
1402 * happen. Never use yield() as a progress guarantee!!
1403 *
1404 * If you want to use yield() to wait for something, use wait_event().
1405 * If you want to use yield() to be 'nice' for others, use cond_resched().
1406 * If you still want to use yield(), do not!
1407 */
yield(void)1408 void __sched yield(void)
1409 {
1410 set_current_state(TASK_RUNNING);
1411 do_sched_yield();
1412 }
1413 EXPORT_SYMBOL(yield);
1414
1415 /**
1416 * yield_to - yield the current processor to another thread in
1417 * your thread group, or accelerate that thread toward the
1418 * processor it's on.
1419 * @p: target task
1420 * @preempt: whether task preemption is allowed or not
1421 *
1422 * It's the caller's job to ensure that the target task struct
1423 * can't go away on us before we can do any checks.
1424 *
1425 * Return:
1426 * true (>0) if we indeed boosted the target task.
1427 * false (0) if we failed to boost the target.
1428 * -ESRCH if there's no task to yield to.
1429 */
yield_to(struct task_struct * p,bool preempt)1430 int __sched yield_to(struct task_struct *p, bool preempt)
1431 {
1432 struct task_struct *curr = current;
1433 struct rq *rq, *p_rq;
1434 int yielded = 0;
1435
1436 scoped_guard (irqsave) {
1437 rq = this_rq();
1438
1439 again:
1440 p_rq = task_rq(p);
1441 /*
1442 * If we're the only runnable task on the rq and target rq also
1443 * has only one task, there's absolutely no point in yielding.
1444 */
1445 if (rq->nr_running == 1 && p_rq->nr_running == 1)
1446 return -ESRCH;
1447
1448 guard(double_rq_lock)(rq, p_rq);
1449 if (task_rq(p) != p_rq)
1450 goto again;
1451
1452 if (!curr->sched_class->yield_to_task)
1453 return 0;
1454
1455 if (curr->sched_class != p->sched_class)
1456 return 0;
1457
1458 if (task_on_cpu(p_rq, p) || !task_is_running(p))
1459 return 0;
1460
1461 yielded = curr->sched_class->yield_to_task(rq, p);
1462 if (yielded) {
1463 schedstat_inc(rq->yld_count);
1464 /*
1465 * Make p's CPU reschedule; pick_next_entity
1466 * takes care of fairness.
1467 */
1468 if (preempt && rq != p_rq)
1469 resched_curr(p_rq);
1470 }
1471 }
1472
1473 if (yielded)
1474 schedule();
1475
1476 return yielded;
1477 }
1478 EXPORT_SYMBOL_GPL(yield_to);
1479
1480 /**
1481 * sys_sched_get_priority_max - return maximum RT priority.
1482 * @policy: scheduling class.
1483 *
1484 * Return: On success, this syscall returns the maximum
1485 * rt_priority that can be used by a given scheduling class.
1486 * On failure, a negative error code is returned.
1487 */
SYSCALL_DEFINE1(sched_get_priority_max,int,policy)1488 SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
1489 {
1490 int ret = -EINVAL;
1491
1492 switch (policy) {
1493 case SCHED_FIFO:
1494 case SCHED_RR:
1495 ret = MAX_RT_PRIO-1;
1496 break;
1497 case SCHED_DEADLINE:
1498 case SCHED_NORMAL:
1499 case SCHED_BATCH:
1500 case SCHED_IDLE:
1501 case SCHED_EXT:
1502 ret = 0;
1503 break;
1504 }
1505 return ret;
1506 }
1507
1508 /**
1509 * sys_sched_get_priority_min - return minimum RT priority.
1510 * @policy: scheduling class.
1511 *
1512 * Return: On success, this syscall returns the minimum
1513 * rt_priority that can be used by a given scheduling class.
1514 * On failure, a negative error code is returned.
1515 */
SYSCALL_DEFINE1(sched_get_priority_min,int,policy)1516 SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
1517 {
1518 int ret = -EINVAL;
1519
1520 switch (policy) {
1521 case SCHED_FIFO:
1522 case SCHED_RR:
1523 ret = 1;
1524 break;
1525 case SCHED_DEADLINE:
1526 case SCHED_NORMAL:
1527 case SCHED_BATCH:
1528 case SCHED_IDLE:
1529 case SCHED_EXT:
1530 ret = 0;
1531 }
1532 return ret;
1533 }
1534
sched_rr_get_interval(pid_t pid,struct timespec64 * t)1535 static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
1536 {
1537 unsigned int time_slice = 0;
1538 int retval;
1539
1540 if (pid < 0)
1541 return -EINVAL;
1542
1543 scoped_guard (rcu) {
1544 struct task_struct *p = find_process_by_pid(pid);
1545 if (!p)
1546 return -ESRCH;
1547
1548 retval = security_task_getscheduler(p);
1549 if (retval)
1550 return retval;
1551
1552 scoped_guard (task_rq_lock, p) {
1553 struct rq *rq = scope.rq;
1554 if (p->sched_class->get_rr_interval)
1555 time_slice = p->sched_class->get_rr_interval(rq, p);
1556 }
1557 }
1558
1559 jiffies_to_timespec64(time_slice, t);
1560 return 0;
1561 }
1562
1563 /**
1564 * sys_sched_rr_get_interval - return the default time-slice of a process.
1565 * @pid: pid of the process.
1566 * @interval: userspace pointer to the time-slice value.
1567 *
1568 * this syscall writes the default time-slice value of a given process
1569 * into the user-space timespec buffer. A value of '0' means infinity.
1570 *
1571 * Return: On success, 0 and the time-slice is in @interval. Otherwise,
1572 * an error code.
1573 */
SYSCALL_DEFINE2(sched_rr_get_interval,pid_t,pid,struct __kernel_timespec __user *,interval)1574 SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
1575 struct __kernel_timespec __user *, interval)
1576 {
1577 struct timespec64 t;
1578 int retval = sched_rr_get_interval(pid, &t);
1579
1580 if (retval == 0)
1581 retval = put_timespec64(&t, interval);
1582
1583 return retval;
1584 }
1585
1586 #ifdef CONFIG_COMPAT_32BIT_TIME
SYSCALL_DEFINE2(sched_rr_get_interval_time32,pid_t,pid,struct old_timespec32 __user *,interval)1587 SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid,
1588 struct old_timespec32 __user *, interval)
1589 {
1590 struct timespec64 t;
1591 int retval = sched_rr_get_interval(pid, &t);
1592
1593 if (retval == 0)
1594 retval = put_old_timespec32(&t, interval);
1595 return retval;
1596 }
1597 #endif
1598