1 /* SPDX-License-Identifier: GPL-2.0+ */
3 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
4 * Internal non-public definitions that provide either classic
5 * or preemptible semantics.
7 * Copyright Red Hat, 2009
8 * Copyright IBM Corporation, 2009
10 * Author: Ingo Molnar <mingo@elte.hu>
11 * Paul E. McKenney <paulmck@linux.ibm.com>
14 #include "../locking/rtmutex_common.h"
16 static bool rcu_rdp_is_offloaded(struct rcu_data *rdp)
19 * In order to read the offloaded state of an rdp in a safe
20 * and stable way and prevent from its value to be changed
21 * under us, we must either hold the barrier mutex, the cpu
22 * hotplug lock (read or write) or the nocb lock. Local
23 * non-preemptible reads are also safe. NOCB kthreads and
24 * timers have their own means of synchronization against the
25 * offloaded state updaters.
28 !(lockdep_is_held(&rcu_state.barrier_mutex) ||
29 (IS_ENABLED(CONFIG_HOTPLUG_CPU) && lockdep_is_cpus_held()) ||
30 rcu_lockdep_is_held_nocb(rdp) ||
31 (rdp == this_cpu_ptr(&rcu_data) &&
32 !(IS_ENABLED(CONFIG_PREEMPT_COUNT) && preemptible())) ||
33 rcu_current_is_nocb_kthread(rdp)),
34 "Unsafe read of RCU_NOCB offloaded state"
37 return rcu_segcblist_is_offloaded(&rdp->cblist);
41 * Check the RCU kernel configuration parameters and print informative
42 * messages about anything out of the ordinary.
44 static void __init rcu_bootup_announce_oddness(void)
46 if (IS_ENABLED(CONFIG_RCU_TRACE))
47 pr_info("\tRCU event tracing is enabled.\n");
48 if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) ||
49 (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32))
50 pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d.\n",
53 pr_info("\tHierarchical RCU autobalancing is disabled.\n");
54 if (IS_ENABLED(CONFIG_PROVE_RCU))
55 pr_info("\tRCU lockdep checking is enabled.\n");
56 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
57 pr_info("\tRCU strict (and thus non-scalable) grace periods are enabled.\n");
58 if (RCU_NUM_LVLS >= 4)
59 pr_info("\tFour(or more)-level hierarchy is enabled.\n");
60 if (RCU_FANOUT_LEAF != 16)
61 pr_info("\tBuild-time adjustment of leaf fanout to %d.\n",
63 if (rcu_fanout_leaf != RCU_FANOUT_LEAF)
64 pr_info("\tBoot-time adjustment of leaf fanout to %d.\n",
66 if (nr_cpu_ids != NR_CPUS)
67 pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids);
68 #ifdef CONFIG_RCU_BOOST
69 pr_info("\tRCU priority boosting: priority %d delay %d ms.\n",
70 kthread_prio, CONFIG_RCU_BOOST_DELAY);
72 if (blimit != DEFAULT_RCU_BLIMIT)
73 pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit);
74 if (qhimark != DEFAULT_RCU_QHIMARK)
75 pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark);
76 if (qlowmark != DEFAULT_RCU_QLOMARK)
77 pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark);
78 if (qovld != DEFAULT_RCU_QOVLD)
79 pr_info("\tBoot-time adjustment of callback overload level to %ld.\n", qovld);
80 if (jiffies_till_first_fqs != ULONG_MAX)
81 pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs);
82 if (jiffies_till_next_fqs != ULONG_MAX)
83 pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs);
84 if (jiffies_till_sched_qs != ULONG_MAX)
85 pr_info("\tBoot-time adjustment of scheduler-enlistment delay to %ld jiffies.\n", jiffies_till_sched_qs);
86 if (rcu_kick_kthreads)
87 pr_info("\tKick kthreads if too-long grace period.\n");
88 if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD))
89 pr_info("\tRCU callback double-/use-after-free debug is enabled.\n");
91 pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay);
93 pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay);
95 pr_info("\tRCU debug GP cleanup slowdown %d jiffies.\n", gp_cleanup_delay);
97 pr_info("\tRCU_SOFTIRQ processing moved to rcuc kthreads.\n");
98 if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG))
99 pr_info("\tRCU debug extended QS entry/exit.\n");
100 rcupdate_announce_bootup_oddness();
103 #ifdef CONFIG_PREEMPT_RCU
105 static void rcu_report_exp_rnp(struct rcu_node *rnp, bool wake);
106 static void rcu_read_unlock_special(struct task_struct *t);
109 * Tell them what RCU they are running.
111 static void __init rcu_bootup_announce(void)
113 pr_info("Preemptible hierarchical RCU implementation.\n");
114 rcu_bootup_announce_oddness();
117 /* Flags for rcu_preempt_ctxt_queue() decision table. */
118 #define RCU_GP_TASKS 0x8
119 #define RCU_EXP_TASKS 0x4
120 #define RCU_GP_BLKD 0x2
121 #define RCU_EXP_BLKD 0x1
124 * Queues a task preempted within an RCU-preempt read-side critical
125 * section into the appropriate location within the ->blkd_tasks list,
126 * depending on the states of any ongoing normal and expedited grace
127 * periods. The ->gp_tasks pointer indicates which element the normal
128 * grace period is waiting on (NULL if none), and the ->exp_tasks pointer
129 * indicates which element the expedited grace period is waiting on (again,
130 * NULL if none). If a grace period is waiting on a given element in the
131 * ->blkd_tasks list, it also waits on all subsequent elements. Thus,
132 * adding a task to the tail of the list blocks any grace period that is
133 * already waiting on one of the elements. In contrast, adding a task
134 * to the head of the list won't block any grace period that is already
135 * waiting on one of the elements.
137 * This queuing is imprecise, and can sometimes make an ongoing grace
138 * period wait for a task that is not strictly speaking blocking it.
139 * Given the choice, we needlessly block a normal grace period rather than
140 * blocking an expedited grace period.
142 * Note that an endless sequence of expedited grace periods still cannot
143 * indefinitely postpone a normal grace period. Eventually, all of the
144 * fixed number of preempted tasks blocking the normal grace period that are
145 * not also blocking the expedited grace period will resume and complete
146 * their RCU read-side critical sections. At that point, the ->gp_tasks
147 * pointer will equal the ->exp_tasks pointer, at which point the end of
148 * the corresponding expedited grace period will also be the end of the
149 * normal grace period.
151 static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp)
152 __releases(rnp->lock) /* But leaves rrupts disabled. */
154 int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) +
155 (rnp->exp_tasks ? RCU_EXP_TASKS : 0) +
156 (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) +
157 (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0);
158 struct task_struct *t = current;
160 raw_lockdep_assert_held_rcu_node(rnp);
161 WARN_ON_ONCE(rdp->mynode != rnp);
162 WARN_ON_ONCE(!rcu_is_leaf_node(rnp));
163 /* RCU better not be waiting on newly onlined CPUs! */
164 WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask &
168 * Decide where to queue the newly blocked task. In theory,
169 * this could be an if-statement. In practice, when I tried
170 * that, it was quite messy.
172 switch (blkd_state) {
175 case RCU_EXP_TASKS + RCU_GP_BLKD:
177 case RCU_GP_TASKS + RCU_EXP_TASKS:
180 * Blocking neither GP, or first task blocking the normal
181 * GP but not blocking the already-waiting expedited GP.
182 * Queue at the head of the list to avoid unnecessarily
183 * blocking the already-waiting GPs.
185 list_add(&t->rcu_node_entry, &rnp->blkd_tasks);
190 case RCU_GP_BLKD + RCU_EXP_BLKD:
191 case RCU_GP_TASKS + RCU_EXP_BLKD:
192 case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
193 case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
196 * First task arriving that blocks either GP, or first task
197 * arriving that blocks the expedited GP (with the normal
198 * GP already waiting), or a task arriving that blocks
199 * both GPs with both GPs already waiting. Queue at the
200 * tail of the list to avoid any GP waiting on any of the
201 * already queued tasks that are not blocking it.
203 list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks);
206 case RCU_EXP_TASKS + RCU_EXP_BLKD:
207 case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD:
208 case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD:
211 * Second or subsequent task blocking the expedited GP.
212 * The task either does not block the normal GP, or is the
213 * first task blocking the normal GP. Queue just after
214 * the first task blocking the expedited GP.
216 list_add(&t->rcu_node_entry, rnp->exp_tasks);
219 case RCU_GP_TASKS + RCU_GP_BLKD:
220 case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD:
223 * Second or subsequent task blocking the normal GP.
224 * The task does not block the expedited GP. Queue just
225 * after the first task blocking the normal GP.
227 list_add(&t->rcu_node_entry, rnp->gp_tasks);
232 /* Yet another exercise in excessive paranoia. */
238 * We have now queued the task. If it was the first one to
239 * block either grace period, update the ->gp_tasks and/or
240 * ->exp_tasks pointers, respectively, to reference the newly
243 if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) {
244 WRITE_ONCE(rnp->gp_tasks, &t->rcu_node_entry);
245 WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq);
247 if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD))
248 WRITE_ONCE(rnp->exp_tasks, &t->rcu_node_entry);
249 WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) !=
250 !(rnp->qsmask & rdp->grpmask));
251 WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) !=
252 !(rnp->expmask & rdp->grpmask));
253 raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */
256 * Report the quiescent state for the expedited GP. This expedited
257 * GP should not be able to end until we report, so there should be
258 * no need to check for a subsequent expedited GP. (Though we are
259 * still in a quiescent state in any case.)
261 if (blkd_state & RCU_EXP_BLKD && rdp->cpu_no_qs.b.exp)
262 rcu_report_exp_rdp(rdp);
264 WARN_ON_ONCE(rdp->cpu_no_qs.b.exp);
268 * Record a preemptible-RCU quiescent state for the specified CPU.
269 * Note that this does not necessarily mean that the task currently running
270 * on the CPU is in a quiescent state: Instead, it means that the current
271 * grace period need not wait on any RCU read-side critical section that
272 * starts later on this CPU. It also means that if the current task is
273 * in an RCU read-side critical section, it has already added itself to
274 * some leaf rcu_node structure's ->blkd_tasks list. In addition to the
275 * current task, there might be any number of other tasks blocked while
276 * in an RCU read-side critical section.
278 * Unlike non-preemptible-RCU, quiescent state reports for expedited
279 * grace periods are handled separately via deferred quiescent states
280 * and context switch events.
282 * Callers to this function must disable preemption.
284 static void rcu_qs(void)
286 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!\n");
287 if (__this_cpu_read(rcu_data.cpu_no_qs.b.norm)) {
288 trace_rcu_grace_period(TPS("rcu_preempt"),
289 __this_cpu_read(rcu_data.gp_seq),
291 __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
292 barrier(); /* Coordinate with rcu_flavor_sched_clock_irq(). */
293 WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, false);
298 * We have entered the scheduler, and the current task might soon be
299 * context-switched away from. If this task is in an RCU read-side
300 * critical section, we will no longer be able to rely on the CPU to
301 * record that fact, so we enqueue the task on the blkd_tasks list.
302 * The task will dequeue itself when it exits the outermost enclosing
303 * RCU read-side critical section. Therefore, the current grace period
304 * cannot be permitted to complete until the blkd_tasks list entries
305 * predating the current grace period drain, in other words, until
306 * rnp->gp_tasks becomes NULL.
308 * Caller must disable interrupts.
310 void rcu_note_context_switch(bool preempt)
312 struct task_struct *t = current;
313 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
314 struct rcu_node *rnp;
316 trace_rcu_utilization(TPS("Start context switch"));
317 lockdep_assert_irqs_disabled();
318 WARN_ONCE(!preempt && rcu_preempt_depth() > 0, "Voluntary context switch within RCU read-side critical section!");
319 if (rcu_preempt_depth() > 0 &&
320 !t->rcu_read_unlock_special.b.blocked) {
322 /* Possibly blocking in an RCU read-side critical section. */
324 raw_spin_lock_rcu_node(rnp);
325 t->rcu_read_unlock_special.b.blocked = true;
326 t->rcu_blocked_node = rnp;
329 * Verify the CPU's sanity, trace the preemption, and
330 * then queue the task as required based on the states
331 * of any ongoing and expedited grace periods.
333 WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0);
334 WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
335 trace_rcu_preempt_task(rcu_state.name,
337 (rnp->qsmask & rdp->grpmask)
339 : rcu_seq_snap(&rnp->gp_seq));
340 rcu_preempt_ctxt_queue(rnp, rdp);
342 rcu_preempt_deferred_qs(t);
346 * Either we were not in an RCU read-side critical section to
347 * begin with, or we have now recorded that critical section
348 * globally. Either way, we can now note a quiescent state
349 * for this CPU. Again, if we were in an RCU read-side critical
350 * section, and if that critical section was blocking the current
351 * grace period, then the fact that the task has been enqueued
352 * means that we continue to block the current grace period.
355 if (rdp->cpu_no_qs.b.exp)
356 rcu_report_exp_rdp(rdp);
357 rcu_tasks_qs(current, preempt);
358 trace_rcu_utilization(TPS("End context switch"));
360 EXPORT_SYMBOL_GPL(rcu_note_context_switch);
363 * Check for preempted RCU readers blocking the current grace period
364 * for the specified rcu_node structure. If the caller needs a reliable
365 * answer, it must hold the rcu_node's ->lock.
367 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
369 return READ_ONCE(rnp->gp_tasks) != NULL;
372 /* limit value for ->rcu_read_lock_nesting. */
373 #define RCU_NEST_PMAX (INT_MAX / 2)
375 static void rcu_preempt_read_enter(void)
377 WRITE_ONCE(current->rcu_read_lock_nesting, READ_ONCE(current->rcu_read_lock_nesting) + 1);
380 static int rcu_preempt_read_exit(void)
382 int ret = READ_ONCE(current->rcu_read_lock_nesting) - 1;
384 WRITE_ONCE(current->rcu_read_lock_nesting, ret);
388 static void rcu_preempt_depth_set(int val)
390 WRITE_ONCE(current->rcu_read_lock_nesting, val);
394 * Preemptible RCU implementation for rcu_read_lock().
395 * Just increment ->rcu_read_lock_nesting, shared state will be updated
398 void __rcu_read_lock(void)
400 rcu_preempt_read_enter();
401 if (IS_ENABLED(CONFIG_PROVE_LOCKING))
402 WARN_ON_ONCE(rcu_preempt_depth() > RCU_NEST_PMAX);
403 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) && rcu_state.gp_kthread)
404 WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, true);
405 barrier(); /* critical section after entry code. */
407 EXPORT_SYMBOL_GPL(__rcu_read_lock);
410 * Preemptible RCU implementation for rcu_read_unlock().
411 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
412 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
413 * invoke rcu_read_unlock_special() to clean up after a context switch
414 * in an RCU read-side critical section and other special cases.
416 void __rcu_read_unlock(void)
418 struct task_struct *t = current;
420 barrier(); // critical section before exit code.
421 if (rcu_preempt_read_exit() == 0) {
422 barrier(); // critical-section exit before .s check.
423 if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s)))
424 rcu_read_unlock_special(t);
426 if (IS_ENABLED(CONFIG_PROVE_LOCKING)) {
427 int rrln = rcu_preempt_depth();
429 WARN_ON_ONCE(rrln < 0 || rrln > RCU_NEST_PMAX);
432 EXPORT_SYMBOL_GPL(__rcu_read_unlock);
435 * Advance a ->blkd_tasks-list pointer to the next entry, instead
436 * returning NULL if at the end of the list.
438 static struct list_head *rcu_next_node_entry(struct task_struct *t,
439 struct rcu_node *rnp)
441 struct list_head *np;
443 np = t->rcu_node_entry.next;
444 if (np == &rnp->blkd_tasks)
450 * Return true if the specified rcu_node structure has tasks that were
451 * preempted within an RCU read-side critical section.
453 static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
455 return !list_empty(&rnp->blkd_tasks);
459 * Report deferred quiescent states. The deferral time can
460 * be quite short, for example, in the case of the call from
461 * rcu_read_unlock_special().
464 rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags)
469 struct list_head *np;
470 bool drop_boost_mutex = false;
471 struct rcu_data *rdp;
472 struct rcu_node *rnp;
473 union rcu_special special;
476 * If RCU core is waiting for this CPU to exit its critical section,
477 * report the fact that it has exited. Because irqs are disabled,
478 * t->rcu_read_unlock_special cannot change.
480 special = t->rcu_read_unlock_special;
481 rdp = this_cpu_ptr(&rcu_data);
482 if (!special.s && !rdp->cpu_no_qs.b.exp) {
483 local_irq_restore(flags);
486 t->rcu_read_unlock_special.s = 0;
487 if (special.b.need_qs) {
488 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) {
489 rcu_report_qs_rdp(rdp);
490 udelay(rcu_unlock_delay);
497 * Respond to a request by an expedited grace period for a
498 * quiescent state from this CPU. Note that requests from
499 * tasks are handled when removing the task from the
500 * blocked-tasks list below.
502 if (rdp->cpu_no_qs.b.exp)
503 rcu_report_exp_rdp(rdp);
505 /* Clean up if blocked during RCU read-side critical section. */
506 if (special.b.blocked) {
509 * Remove this task from the list it blocked on. The task
510 * now remains queued on the rcu_node corresponding to the
511 * CPU it first blocked on, so there is no longer any need
512 * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia.
514 rnp = t->rcu_blocked_node;
515 raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */
516 WARN_ON_ONCE(rnp != t->rcu_blocked_node);
517 WARN_ON_ONCE(!rcu_is_leaf_node(rnp));
518 empty_norm = !rcu_preempt_blocked_readers_cgp(rnp);
519 WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq &&
520 (!empty_norm || rnp->qsmask));
521 empty_exp = sync_rcu_exp_done(rnp);
522 smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
523 np = rcu_next_node_entry(t, rnp);
524 list_del_init(&t->rcu_node_entry);
525 t->rcu_blocked_node = NULL;
526 trace_rcu_unlock_preempted_task(TPS("rcu_preempt"),
527 rnp->gp_seq, t->pid);
528 if (&t->rcu_node_entry == rnp->gp_tasks)
529 WRITE_ONCE(rnp->gp_tasks, np);
530 if (&t->rcu_node_entry == rnp->exp_tasks)
531 WRITE_ONCE(rnp->exp_tasks, np);
532 if (IS_ENABLED(CONFIG_RCU_BOOST)) {
533 /* Snapshot ->boost_mtx ownership w/rnp->lock held. */
534 drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx.rtmutex) == t;
535 if (&t->rcu_node_entry == rnp->boost_tasks)
536 WRITE_ONCE(rnp->boost_tasks, np);
540 * If this was the last task on the current list, and if
541 * we aren't waiting on any CPUs, report the quiescent state.
542 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
543 * so we must take a snapshot of the expedited state.
545 empty_exp_now = sync_rcu_exp_done(rnp);
546 if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) {
547 trace_rcu_quiescent_state_report(TPS("preempt_rcu"),
554 rcu_report_unblock_qs_rnp(rnp, flags);
556 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
559 /* Unboost if we were boosted. */
560 if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex)
561 rt_mutex_futex_unlock(&rnp->boost_mtx.rtmutex);
564 * If this was the last task on the expedited lists,
565 * then we need to report up the rcu_node hierarchy.
567 if (!empty_exp && empty_exp_now)
568 rcu_report_exp_rnp(rnp, true);
570 local_irq_restore(flags);
575 * Is a deferred quiescent-state pending, and are we also not in
576 * an RCU read-side critical section? It is the caller's responsibility
577 * to ensure it is otherwise safe to report any deferred quiescent
578 * states. The reason for this is that it is safe to report a
579 * quiescent state during context switch even though preemption
580 * is disabled. This function cannot be expected to understand these
581 * nuances, so the caller must handle them.
583 static bool rcu_preempt_need_deferred_qs(struct task_struct *t)
585 return (__this_cpu_read(rcu_data.cpu_no_qs.b.exp) ||
586 READ_ONCE(t->rcu_read_unlock_special.s)) &&
587 rcu_preempt_depth() == 0;
591 * Report a deferred quiescent state if needed and safe to do so.
592 * As with rcu_preempt_need_deferred_qs(), "safe" involves only
593 * not being in an RCU read-side critical section. The caller must
594 * evaluate safety in terms of interrupt, softirq, and preemption
597 static void rcu_preempt_deferred_qs(struct task_struct *t)
601 if (!rcu_preempt_need_deferred_qs(t))
603 local_irq_save(flags);
604 rcu_preempt_deferred_qs_irqrestore(t, flags);
608 * Minimal handler to give the scheduler a chance to re-evaluate.
610 static void rcu_preempt_deferred_qs_handler(struct irq_work *iwp)
612 struct rcu_data *rdp;
614 rdp = container_of(iwp, struct rcu_data, defer_qs_iw);
615 rdp->defer_qs_iw_pending = false;
619 * Handle special cases during rcu_read_unlock(), such as needing to
620 * notify RCU core processing or task having blocked during the RCU
621 * read-side critical section.
623 static void rcu_read_unlock_special(struct task_struct *t)
626 bool irqs_were_disabled;
627 bool preempt_bh_were_disabled =
628 !!(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK));
630 /* NMI handlers cannot block and cannot safely manipulate state. */
634 local_irq_save(flags);
635 irqs_were_disabled = irqs_disabled_flags(flags);
636 if (preempt_bh_were_disabled || irqs_were_disabled) {
637 bool expboost; // Expedited GP in flight or possible boosting.
638 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
639 struct rcu_node *rnp = rdp->mynode;
641 expboost = (t->rcu_blocked_node && READ_ONCE(t->rcu_blocked_node->exp_tasks)) ||
642 (rdp->grpmask & READ_ONCE(rnp->expmask)) ||
643 IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) ||
644 (IS_ENABLED(CONFIG_RCU_BOOST) && irqs_were_disabled &&
645 t->rcu_blocked_node);
646 // Need to defer quiescent state until everything is enabled.
647 if (use_softirq && (in_hardirq() || (expboost && !irqs_were_disabled))) {
648 // Using softirq, safe to awaken, and either the
649 // wakeup is free or there is either an expedited
650 // GP in flight or a potential need to deboost.
651 raise_softirq_irqoff(RCU_SOFTIRQ);
653 // Enabling BH or preempt does reschedule, so...
654 // Also if no expediting and no possible deboosting,
655 // slow is OK. Plus nohz_full CPUs eventually get
657 set_tsk_need_resched(current);
658 set_preempt_need_resched();
659 if (IS_ENABLED(CONFIG_IRQ_WORK) && irqs_were_disabled &&
660 expboost && !rdp->defer_qs_iw_pending && cpu_online(rdp->cpu)) {
661 // Get scheduler to re-evaluate and call hooks.
662 // If !IRQ_WORK, FQS scan will eventually IPI.
663 init_irq_work(&rdp->defer_qs_iw, rcu_preempt_deferred_qs_handler);
664 rdp->defer_qs_iw_pending = true;
665 irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu);
668 local_irq_restore(flags);
671 rcu_preempt_deferred_qs_irqrestore(t, flags);
675 * Check that the list of blocked tasks for the newly completed grace
676 * period is in fact empty. It is a serious bug to complete a grace
677 * period that still has RCU readers blocked! This function must be
678 * invoked -before- updating this rnp's ->gp_seq.
680 * Also, if there are blocked tasks on the list, they automatically
681 * block the newly created grace period, so set up ->gp_tasks accordingly.
683 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
685 struct task_struct *t;
687 RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n");
688 raw_lockdep_assert_held_rcu_node(rnp);
689 if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)))
690 dump_blkd_tasks(rnp, 10);
691 if (rcu_preempt_has_tasks(rnp) &&
692 (rnp->qsmaskinit || rnp->wait_blkd_tasks)) {
693 WRITE_ONCE(rnp->gp_tasks, rnp->blkd_tasks.next);
694 t = container_of(rnp->gp_tasks, struct task_struct,
696 trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"),
697 rnp->gp_seq, t->pid);
699 WARN_ON_ONCE(rnp->qsmask);
703 * Check for a quiescent state from the current CPU, including voluntary
704 * context switches for Tasks RCU. When a task blocks, the task is
705 * recorded in the corresponding CPU's rcu_node structure, which is checked
706 * elsewhere, hence this function need only check for quiescent states
707 * related to the current CPU, not to those related to tasks.
709 static void rcu_flavor_sched_clock_irq(int user)
711 struct task_struct *t = current;
713 lockdep_assert_irqs_disabled();
714 if (user || rcu_is_cpu_rrupt_from_idle()) {
715 rcu_note_voluntary_context_switch(current);
717 if (rcu_preempt_depth() > 0 ||
718 (preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK))) {
719 /* No QS, force context switch if deferred. */
720 if (rcu_preempt_need_deferred_qs(t)) {
721 set_tsk_need_resched(t);
722 set_preempt_need_resched();
724 } else if (rcu_preempt_need_deferred_qs(t)) {
725 rcu_preempt_deferred_qs(t); /* Report deferred QS. */
727 } else if (!WARN_ON_ONCE(rcu_preempt_depth())) {
728 rcu_qs(); /* Report immediate QS. */
732 /* If GP is oldish, ask for help from rcu_read_unlock_special(). */
733 if (rcu_preempt_depth() > 0 &&
734 __this_cpu_read(rcu_data.core_needs_qs) &&
735 __this_cpu_read(rcu_data.cpu_no_qs.b.norm) &&
736 !t->rcu_read_unlock_special.b.need_qs &&
737 time_after(jiffies, rcu_state.gp_start + HZ))
738 t->rcu_read_unlock_special.b.need_qs = true;
742 * Check for a task exiting while in a preemptible-RCU read-side
743 * critical section, clean up if so. No need to issue warnings, as
744 * debug_check_no_locks_held() already does this if lockdep is enabled.
745 * Besides, if this function does anything other than just immediately
746 * return, there was a bug of some sort. Spewing warnings from this
747 * function is like as not to simply obscure important prior warnings.
751 struct task_struct *t = current;
753 if (unlikely(!list_empty(¤t->rcu_node_entry))) {
754 rcu_preempt_depth_set(1);
756 WRITE_ONCE(t->rcu_read_unlock_special.b.blocked, true);
757 } else if (unlikely(rcu_preempt_depth())) {
758 rcu_preempt_depth_set(1);
763 rcu_preempt_deferred_qs(current);
767 * Dump the blocked-tasks state, but limit the list dump to the
768 * specified number of elements.
771 dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
775 struct list_head *lhp;
777 struct rcu_data *rdp;
778 struct rcu_node *rnp1;
780 raw_lockdep_assert_held_rcu_node(rnp);
781 pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n",
782 __func__, rnp->grplo, rnp->grphi, rnp->level,
783 (long)READ_ONCE(rnp->gp_seq), (long)rnp->completedqs);
784 for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent)
785 pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n",
786 __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext);
787 pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n",
788 __func__, READ_ONCE(rnp->gp_tasks), data_race(rnp->boost_tasks),
789 READ_ONCE(rnp->exp_tasks));
790 pr_info("%s: ->blkd_tasks", __func__);
792 list_for_each(lhp, &rnp->blkd_tasks) {
798 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) {
799 rdp = per_cpu_ptr(&rcu_data, cpu);
800 onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp));
801 pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n",
803 (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags,
804 (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags);
808 #else /* #ifdef CONFIG_PREEMPT_RCU */
811 * If strict grace periods are enabled, and if the calling
812 * __rcu_read_unlock() marks the beginning of a quiescent state, immediately
813 * report that quiescent state and, if requested, spin for a bit.
815 void rcu_read_unlock_strict(void)
817 struct rcu_data *rdp;
819 if (irqs_disabled() || preempt_count() || !rcu_state.gp_kthread)
821 rdp = this_cpu_ptr(&rcu_data);
822 rcu_report_qs_rdp(rdp);
823 udelay(rcu_unlock_delay);
825 EXPORT_SYMBOL_GPL(rcu_read_unlock_strict);
828 * Tell them what RCU they are running.
830 static void __init rcu_bootup_announce(void)
832 pr_info("Hierarchical RCU implementation.\n");
833 rcu_bootup_announce_oddness();
837 * Note a quiescent state for PREEMPTION=n. Because we do not need to know
838 * how many quiescent states passed, just if there was at least one since
839 * the start of the grace period, this just sets a flag. The caller must
840 * have disabled preemption.
842 static void rcu_qs(void)
844 RCU_LOCKDEP_WARN(preemptible(), "rcu_qs() invoked with preemption enabled!!!");
845 if (!__this_cpu_read(rcu_data.cpu_no_qs.s))
847 trace_rcu_grace_period(TPS("rcu_sched"),
848 __this_cpu_read(rcu_data.gp_seq), TPS("cpuqs"));
849 __this_cpu_write(rcu_data.cpu_no_qs.b.norm, false);
850 if (__this_cpu_read(rcu_data.cpu_no_qs.b.exp))
851 rcu_report_exp_rdp(this_cpu_ptr(&rcu_data));
855 * Register an urgently needed quiescent state. If there is an
856 * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight
857 * dyntick-idle quiescent state visible to other CPUs, which will in
858 * some cases serve for expedited as well as normal grace periods.
859 * Either way, register a lightweight quiescent state.
861 void rcu_all_qs(void)
865 if (!raw_cpu_read(rcu_data.rcu_urgent_qs))
868 /* Load rcu_urgent_qs before other flags. */
869 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs))) {
873 this_cpu_write(rcu_data.rcu_urgent_qs, false);
874 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs))) {
875 local_irq_save(flags);
876 rcu_momentary_dyntick_idle();
877 local_irq_restore(flags);
882 EXPORT_SYMBOL_GPL(rcu_all_qs);
885 * Note a PREEMPTION=n context switch. The caller must have disabled interrupts.
887 void rcu_note_context_switch(bool preempt)
889 trace_rcu_utilization(TPS("Start context switch"));
891 /* Load rcu_urgent_qs before other flags. */
892 if (!smp_load_acquire(this_cpu_ptr(&rcu_data.rcu_urgent_qs)))
894 this_cpu_write(rcu_data.rcu_urgent_qs, false);
895 if (unlikely(raw_cpu_read(rcu_data.rcu_need_heavy_qs)))
896 rcu_momentary_dyntick_idle();
897 rcu_tasks_qs(current, preempt);
899 trace_rcu_utilization(TPS("End context switch"));
901 EXPORT_SYMBOL_GPL(rcu_note_context_switch);
904 * Because preemptible RCU does not exist, there are never any preempted
907 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
913 * Because there is no preemptible RCU, there can be no readers blocked.
915 static bool rcu_preempt_has_tasks(struct rcu_node *rnp)
921 * Because there is no preemptible RCU, there can be no deferred quiescent
924 static bool rcu_preempt_need_deferred_qs(struct task_struct *t)
929 // Except that we do need to respond to a request by an expedited grace
930 // period for a quiescent state from this CPU. Note that requests from
931 // tasks are handled when removing the task from the blocked-tasks list
933 static void rcu_preempt_deferred_qs(struct task_struct *t)
935 struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
937 if (rdp->cpu_no_qs.b.exp)
938 rcu_report_exp_rdp(rdp);
942 * Because there is no preemptible RCU, there can be no readers blocked,
943 * so there is no need to check for blocked tasks. So check only for
944 * bogus qsmask values.
946 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
948 WARN_ON_ONCE(rnp->qsmask);
952 * Check to see if this CPU is in a non-context-switch quiescent state,
953 * namely user mode and idle loop.
955 static void rcu_flavor_sched_clock_irq(int user)
957 if (user || rcu_is_cpu_rrupt_from_idle()) {
960 * Get here if this CPU took its interrupt from user
961 * mode or from the idle loop, and if this is not a
962 * nested interrupt. In this case, the CPU is in
963 * a quiescent state, so note it.
965 * No memory barrier is required here because rcu_qs()
966 * references only CPU-local variables that other CPUs
967 * neither access nor modify, at least not while the
968 * corresponding CPU is online.
976 * Because preemptible RCU does not exist, tasks cannot possibly exit
977 * while in preemptible RCU read-side critical sections.
984 * Dump the guaranteed-empty blocked-tasks state. Trust but verify.
987 dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
989 WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks));
992 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
995 * If boosting, set rcuc kthreads to realtime priority.
997 static void rcu_cpu_kthread_setup(unsigned int cpu)
999 #ifdef CONFIG_RCU_BOOST
1000 struct sched_param sp;
1002 sp.sched_priority = kthread_prio;
1003 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1004 #endif /* #ifdef CONFIG_RCU_BOOST */
1007 #ifdef CONFIG_RCU_BOOST
1010 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1011 * or ->boost_tasks, advancing the pointer to the next task in the
1012 * ->blkd_tasks list.
1014 * Note that irqs must be enabled: boosting the task can block.
1015 * Returns 1 if there are more tasks needing to be boosted.
1017 static int rcu_boost(struct rcu_node *rnp)
1019 unsigned long flags;
1020 struct task_struct *t;
1021 struct list_head *tb;
1023 if (READ_ONCE(rnp->exp_tasks) == NULL &&
1024 READ_ONCE(rnp->boost_tasks) == NULL)
1025 return 0; /* Nothing left to boost. */
1027 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1030 * Recheck under the lock: all tasks in need of boosting
1031 * might exit their RCU read-side critical sections on their own.
1033 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1034 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1039 * Preferentially boost tasks blocking expedited grace periods.
1040 * This cannot starve the normal grace periods because a second
1041 * expedited grace period must boost all blocked tasks, including
1042 * those blocking the pre-existing normal grace period.
1044 if (rnp->exp_tasks != NULL)
1045 tb = rnp->exp_tasks;
1047 tb = rnp->boost_tasks;
1050 * We boost task t by manufacturing an rt_mutex that appears to
1051 * be held by task t. We leave a pointer to that rt_mutex where
1052 * task t can find it, and task t will release the mutex when it
1053 * exits its outermost RCU read-side critical section. Then
1054 * simply acquiring this artificial rt_mutex will boost task
1055 * t's priority. (Thanks to tglx for suggesting this approach!)
1057 * Note that task t must acquire rnp->lock to remove itself from
1058 * the ->blkd_tasks list, which it will do from exit() if from
1059 * nowhere else. We therefore are guaranteed that task t will
1060 * stay around at least until we drop rnp->lock. Note that
1061 * rnp->lock also resolves races between our priority boosting
1062 * and task t's exiting its outermost RCU read-side critical
1065 t = container_of(tb, struct task_struct, rcu_node_entry);
1066 rt_mutex_init_proxy_locked(&rnp->boost_mtx.rtmutex, t);
1067 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1068 /* Lock only for side effect: boosts task t's priority. */
1069 rt_mutex_lock(&rnp->boost_mtx);
1070 rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */
1073 return READ_ONCE(rnp->exp_tasks) != NULL ||
1074 READ_ONCE(rnp->boost_tasks) != NULL;
1078 * Priority-boosting kthread, one per leaf rcu_node.
1080 static int rcu_boost_kthread(void *arg)
1082 struct rcu_node *rnp = (struct rcu_node *)arg;
1086 trace_rcu_utilization(TPS("Start boost kthread@init"));
1088 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_WAITING);
1089 trace_rcu_utilization(TPS("End boost kthread@rcu_wait"));
1090 rcu_wait(READ_ONCE(rnp->boost_tasks) ||
1091 READ_ONCE(rnp->exp_tasks));
1092 trace_rcu_utilization(TPS("Start boost kthread@rcu_wait"));
1093 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_RUNNING);
1094 more2boost = rcu_boost(rnp);
1100 WRITE_ONCE(rnp->boost_kthread_status, RCU_KTHREAD_YIELDING);
1101 trace_rcu_utilization(TPS("End boost kthread@rcu_yield"));
1102 schedule_timeout_idle(2);
1103 trace_rcu_utilization(TPS("Start boost kthread@rcu_yield"));
1108 trace_rcu_utilization(TPS("End boost kthread@notreached"));
1113 * Check to see if it is time to start boosting RCU readers that are
1114 * blocking the current grace period, and, if so, tell the per-rcu_node
1115 * kthread to start boosting them. If there is an expedited grace
1116 * period in progress, it is always time to boost.
1118 * The caller must hold rnp->lock, which this function releases.
1119 * The ->boost_kthread_task is immortal, so we don't need to worry
1120 * about it going away.
1122 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1123 __releases(rnp->lock)
1125 raw_lockdep_assert_held_rcu_node(rnp);
1126 if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
1127 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1130 if (rnp->exp_tasks != NULL ||
1131 (rnp->gp_tasks != NULL &&
1132 rnp->boost_tasks == NULL &&
1134 (!time_after(rnp->boost_time, jiffies) || rcu_state.cbovld))) {
1135 if (rnp->exp_tasks == NULL)
1136 WRITE_ONCE(rnp->boost_tasks, rnp->gp_tasks);
1137 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1138 rcu_wake_cond(rnp->boost_kthread_task,
1139 READ_ONCE(rnp->boost_kthread_status));
1141 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1146 * Is the current CPU running the RCU-callbacks kthread?
1147 * Caller must have preemption disabled.
1149 static bool rcu_is_callbacks_kthread(void)
1151 return __this_cpu_read(rcu_data.rcu_cpu_kthread_task) == current;
1154 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1157 * Do priority-boost accounting for the start of a new grace period.
1159 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1161 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1165 * Create an RCU-boost kthread for the specified node if one does not
1166 * already exist. We only create this kthread for preemptible RCU.
1168 static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1170 unsigned long flags;
1171 int rnp_index = rnp - rcu_get_root();
1172 struct sched_param sp;
1173 struct task_struct *t;
1175 if (rnp->boost_kthread_task || !rcu_scheduler_fully_active)
1178 rcu_state.boost = 1;
1180 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1181 "rcub/%d", rnp_index);
1182 if (WARN_ON_ONCE(IS_ERR(t)))
1185 raw_spin_lock_irqsave_rcu_node(rnp, flags);
1186 rnp->boost_kthread_task = t;
1187 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1188 sp.sched_priority = kthread_prio;
1189 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1190 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1194 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1195 * served by the rcu_node in question. The CPU hotplug lock is still
1196 * held, so the value of rnp->qsmaskinit will be stable.
1198 * We don't include outgoingcpu in the affinity set, use -1 if there is
1199 * no outgoing CPU. If there are no CPUs left in the affinity set,
1200 * this function allows the kthread to execute on any CPU.
1202 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1204 struct task_struct *t = rnp->boost_kthread_task;
1205 unsigned long mask = rcu_rnp_online_cpus(rnp);
1211 if (!zalloc_cpumask_var(&cm, GFP_KERNEL))
1213 for_each_leaf_node_possible_cpu(rnp, cpu)
1214 if ((mask & leaf_node_cpu_bit(rnp, cpu)) &&
1216 cpumask_set_cpu(cpu, cm);
1217 cpumask_and(cm, cm, housekeeping_cpumask(HK_FLAG_RCU));
1218 if (cpumask_weight(cm) == 0)
1219 cpumask_copy(cm, housekeeping_cpumask(HK_FLAG_RCU));
1220 set_cpus_allowed_ptr(t, cm);
1221 free_cpumask_var(cm);
1225 * Spawn boost kthreads -- called as soon as the scheduler is running.
1227 static void __init rcu_spawn_boost_kthreads(void)
1229 struct rcu_node *rnp;
1231 rcu_for_each_leaf_node(rnp)
1232 if (rcu_rnp_online_cpus(rnp))
1233 rcu_spawn_one_boost_kthread(rnp);
1236 #else /* #ifdef CONFIG_RCU_BOOST */
1238 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1239 __releases(rnp->lock)
1241 raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
1244 static bool rcu_is_callbacks_kthread(void)
1249 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1253 static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
1257 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1261 static void __init rcu_spawn_boost_kthreads(void)
1265 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1268 * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
1269 * grace-period kthread will do force_quiescent_state() processing?
1270 * The idea is to avoid waking up RCU core processing on such a
1271 * CPU unless the grace period has extended for too long.
1273 * This code relies on the fact that all NO_HZ_FULL CPUs are also
1274 * RCU_NOCB_CPU CPUs.
1276 static bool rcu_nohz_full_cpu(void)
1278 #ifdef CONFIG_NO_HZ_FULL
1279 if (tick_nohz_full_cpu(smp_processor_id()) &&
1280 (!rcu_gp_in_progress() ||
1281 time_before(jiffies, READ_ONCE(rcu_state.gp_start) + HZ)))
1283 #endif /* #ifdef CONFIG_NO_HZ_FULL */
1288 * Bind the RCU grace-period kthreads to the housekeeping CPU.
1290 static void rcu_bind_gp_kthread(void)
1292 if (!tick_nohz_full_enabled())
1294 housekeeping_affine(current, HK_FLAG_RCU);
1297 /* Record the current task on dyntick-idle entry. */
1298 static __always_inline void rcu_dynticks_task_enter(void)
1300 #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
1301 WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id());
1302 #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */
1305 /* Record no current task on dyntick-idle exit. */
1306 static __always_inline void rcu_dynticks_task_exit(void)
1308 #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL)
1309 WRITE_ONCE(current->rcu_tasks_idle_cpu, -1);
1310 #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */
1313 /* Turn on heavyweight RCU tasks trace readers on idle/user entry. */
1314 static __always_inline void rcu_dynticks_task_trace_enter(void)
1316 #ifdef CONFIG_TASKS_TRACE_RCU
1317 if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB))
1318 current->trc_reader_special.b.need_mb = true;
1319 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1322 /* Turn off heavyweight RCU tasks trace readers on idle/user exit. */
1323 static __always_inline void rcu_dynticks_task_trace_exit(void)
1325 #ifdef CONFIG_TASKS_TRACE_RCU
1326 if (IS_ENABLED(CONFIG_TASKS_TRACE_RCU_READ_MB))
1327 current->trc_reader_special.b.need_mb = false;
1328 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */