2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions that provide either classic
4 * or preemptible semantics.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
20 * Copyright Red Hat, 2009
21 * Copyright IBM Corporation, 2009
23 * Author: Ingo Molnar <mingo@elte.hu>
24 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
27 #include <linux/delay.h>
29 #define RCU_KTHREAD_PRIO 1
31 #ifdef CONFIG_RCU_BOOST
32 #define RCU_BOOST_PRIO CONFIG_RCU_BOOST_PRIO
34 #define RCU_BOOST_PRIO RCU_KTHREAD_PRIO
38 * Check the RCU kernel configuration parameters and print informative
39 * messages about anything out of the ordinary. If you like #ifdef, you
40 * will love this function.
42 static void __init rcu_bootup_announce_oddness(void)
44 #ifdef CONFIG_RCU_TRACE
45 printk(KERN_INFO "\tRCU debugfs-based tracing is enabled.\n");
47 #if (defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 64) || (!defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 32)
48 printk(KERN_INFO "\tCONFIG_RCU_FANOUT set to non-default value of %d\n",
51 #ifdef CONFIG_RCU_FANOUT_EXACT
52 printk(KERN_INFO "\tHierarchical RCU autobalancing is disabled.\n");
54 #ifdef CONFIG_RCU_FAST_NO_HZ
56 "\tRCU dyntick-idle grace-period acceleration is enabled.\n");
58 #ifdef CONFIG_PROVE_RCU
59 printk(KERN_INFO "\tRCU lockdep checking is enabled.\n");
61 #ifdef CONFIG_RCU_TORTURE_TEST_RUNNABLE
62 printk(KERN_INFO "\tRCU torture testing starts during boot.\n");
64 #if defined(CONFIG_TREE_PREEMPT_RCU) && !defined(CONFIG_RCU_CPU_STALL_VERBOSE)
65 printk(KERN_INFO "\tDump stacks of tasks blocking RCU-preempt GP.\n");
67 #if defined(CONFIG_RCU_CPU_STALL_INFO)
68 printk(KERN_INFO "\tAdditional per-CPU info printed with stalls.\n");
70 #if NUM_RCU_LVL_4 != 0
71 printk(KERN_INFO "\tFour-level hierarchy is enabled.\n");
73 if (rcu_fanout_leaf != CONFIG_RCU_FANOUT_LEAF)
74 printk(KERN_INFO "\tExperimental boot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf);
75 if (nr_cpu_ids != NR_CPUS)
76 printk(KERN_INFO "\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%d.\n", NR_CPUS, nr_cpu_ids);
79 #ifdef CONFIG_TREE_PREEMPT_RCU
81 struct rcu_state rcu_preempt_state =
82 RCU_STATE_INITIALIZER(rcu_preempt, call_rcu);
83 DEFINE_PER_CPU(struct rcu_data, rcu_preempt_data);
84 static struct rcu_state *rcu_state = &rcu_preempt_state;
86 static int rcu_preempted_readers_exp(struct rcu_node *rnp);
89 * Tell them what RCU they are running.
91 static void __init rcu_bootup_announce(void)
93 printk(KERN_INFO "Preemptible hierarchical RCU implementation.\n");
94 rcu_bootup_announce_oddness();
98 * Return the number of RCU-preempt batches processed thus far
99 * for debug and statistics.
101 long rcu_batches_completed_preempt(void)
103 return rcu_preempt_state.completed;
105 EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt);
108 * Return the number of RCU batches processed thus far for debug & stats.
110 long rcu_batches_completed(void)
112 return rcu_batches_completed_preempt();
114 EXPORT_SYMBOL_GPL(rcu_batches_completed);
117 * Force a quiescent state for preemptible RCU.
119 void rcu_force_quiescent_state(void)
121 force_quiescent_state(&rcu_preempt_state, 0);
123 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
126 * Record a preemptible-RCU quiescent state for the specified CPU. Note
127 * that this just means that the task currently running on the CPU is
128 * not in a quiescent state. There might be any number of tasks blocked
129 * while in an RCU read-side critical section.
131 * Unlike the other rcu_*_qs() functions, callers to this function
132 * must disable irqs in order to protect the assignment to
133 * ->rcu_read_unlock_special.
135 static void rcu_preempt_qs(int cpu)
137 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
139 rdp->passed_quiesce_gpnum = rdp->gpnum;
141 if (rdp->passed_quiesce == 0)
142 trace_rcu_grace_period("rcu_preempt", rdp->gpnum, "cpuqs");
143 rdp->passed_quiesce = 1;
144 current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
148 * We have entered the scheduler, and the current task might soon be
149 * context-switched away from. If this task is in an RCU read-side
150 * critical section, we will no longer be able to rely on the CPU to
151 * record that fact, so we enqueue the task on the blkd_tasks list.
152 * The task will dequeue itself when it exits the outermost enclosing
153 * RCU read-side critical section. Therefore, the current grace period
154 * cannot be permitted to complete until the blkd_tasks list entries
155 * predating the current grace period drain, in other words, until
156 * rnp->gp_tasks becomes NULL.
158 * Caller must disable preemption.
160 static void rcu_preempt_note_context_switch(int cpu)
162 struct task_struct *t = current;
164 struct rcu_data *rdp;
165 struct rcu_node *rnp;
167 if (t->rcu_read_lock_nesting > 0 &&
168 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
170 /* Possibly blocking in an RCU read-side critical section. */
171 rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu);
173 raw_spin_lock_irqsave(&rnp->lock, flags);
174 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
175 t->rcu_blocked_node = rnp;
178 * If this CPU has already checked in, then this task
179 * will hold up the next grace period rather than the
180 * current grace period. Queue the task accordingly.
181 * If the task is queued for the current grace period
182 * (i.e., this CPU has not yet passed through a quiescent
183 * state for the current grace period), then as long
184 * as that task remains queued, the current grace period
185 * cannot end. Note that there is some uncertainty as
186 * to exactly when the current grace period started.
187 * We take a conservative approach, which can result
188 * in unnecessarily waiting on tasks that started very
189 * slightly after the current grace period began. C'est
192 * But first, note that the current CPU must still be
195 WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0);
196 WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
197 if ((rnp->qsmask & rdp->grpmask) && rnp->gp_tasks != NULL) {
198 list_add(&t->rcu_node_entry, rnp->gp_tasks->prev);
199 rnp->gp_tasks = &t->rcu_node_entry;
200 #ifdef CONFIG_RCU_BOOST
201 if (rnp->boost_tasks != NULL)
202 rnp->boost_tasks = rnp->gp_tasks;
203 #endif /* #ifdef CONFIG_RCU_BOOST */
205 list_add(&t->rcu_node_entry, &rnp->blkd_tasks);
206 if (rnp->qsmask & rdp->grpmask)
207 rnp->gp_tasks = &t->rcu_node_entry;
209 trace_rcu_preempt_task(rdp->rsp->name,
211 (rnp->qsmask & rdp->grpmask)
214 raw_spin_unlock_irqrestore(&rnp->lock, flags);
215 } else if (t->rcu_read_lock_nesting < 0 &&
216 t->rcu_read_unlock_special) {
219 * Complete exit from RCU read-side critical section on
220 * behalf of preempted instance of __rcu_read_unlock().
222 rcu_read_unlock_special(t);
226 * Either we were not in an RCU read-side critical section to
227 * begin with, or we have now recorded that critical section
228 * globally. Either way, we can now note a quiescent state
229 * for this CPU. Again, if we were in an RCU read-side critical
230 * section, and if that critical section was blocking the current
231 * grace period, then the fact that the task has been enqueued
232 * means that we continue to block the current grace period.
234 local_irq_save(flags);
236 local_irq_restore(flags);
240 * Check for preempted RCU readers blocking the current grace period
241 * for the specified rcu_node structure. If the caller needs a reliable
242 * answer, it must hold the rcu_node's ->lock.
244 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
246 return rnp->gp_tasks != NULL;
250 * Record a quiescent state for all tasks that were previously queued
251 * on the specified rcu_node structure and that were blocking the current
252 * RCU grace period. The caller must hold the specified rnp->lock with
253 * irqs disabled, and this lock is released upon return, but irqs remain
256 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
257 __releases(rnp->lock)
260 struct rcu_node *rnp_p;
262 if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
263 raw_spin_unlock_irqrestore(&rnp->lock, flags);
264 return; /* Still need more quiescent states! */
270 * Either there is only one rcu_node in the tree,
271 * or tasks were kicked up to root rcu_node due to
272 * CPUs going offline.
274 rcu_report_qs_rsp(&rcu_preempt_state, flags);
278 /* Report up the rest of the hierarchy. */
280 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
281 raw_spin_lock(&rnp_p->lock); /* irqs already disabled. */
282 rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags);
286 * Advance a ->blkd_tasks-list pointer to the next entry, instead
287 * returning NULL if at the end of the list.
289 static struct list_head *rcu_next_node_entry(struct task_struct *t,
290 struct rcu_node *rnp)
292 struct list_head *np;
294 np = t->rcu_node_entry.next;
295 if (np == &rnp->blkd_tasks)
301 * Handle special cases during rcu_read_unlock(), such as needing to
302 * notify RCU core processing or task having blocked during the RCU
303 * read-side critical section.
305 void rcu_read_unlock_special(struct task_struct *t)
311 struct list_head *np;
312 #ifdef CONFIG_RCU_BOOST
313 struct rt_mutex *rbmp = NULL;
314 #endif /* #ifdef CONFIG_RCU_BOOST */
315 struct rcu_node *rnp;
318 /* NMI handlers cannot block and cannot safely manipulate state. */
322 local_irq_save(flags);
325 * If RCU core is waiting for this CPU to exit critical section,
326 * let it know that we have done so.
328 special = t->rcu_read_unlock_special;
329 if (special & RCU_READ_UNLOCK_NEED_QS) {
330 rcu_preempt_qs(smp_processor_id());
333 /* Hardware IRQ handlers cannot block. */
334 if (in_irq() || in_serving_softirq()) {
335 local_irq_restore(flags);
339 /* Clean up if blocked during RCU read-side critical section. */
340 if (special & RCU_READ_UNLOCK_BLOCKED) {
341 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
344 * Remove this task from the list it blocked on. The
345 * task can migrate while we acquire the lock, but at
346 * most one time. So at most two passes through loop.
349 rnp = t->rcu_blocked_node;
350 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
351 if (rnp == t->rcu_blocked_node)
353 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
355 empty = !rcu_preempt_blocked_readers_cgp(rnp);
356 empty_exp = !rcu_preempted_readers_exp(rnp);
357 smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
358 np = rcu_next_node_entry(t, rnp);
359 list_del_init(&t->rcu_node_entry);
360 t->rcu_blocked_node = NULL;
361 trace_rcu_unlock_preempted_task("rcu_preempt",
363 if (&t->rcu_node_entry == rnp->gp_tasks)
365 if (&t->rcu_node_entry == rnp->exp_tasks)
367 #ifdef CONFIG_RCU_BOOST
368 if (&t->rcu_node_entry == rnp->boost_tasks)
369 rnp->boost_tasks = np;
370 /* Snapshot/clear ->rcu_boost_mutex with rcu_node lock held. */
371 if (t->rcu_boost_mutex) {
372 rbmp = t->rcu_boost_mutex;
373 t->rcu_boost_mutex = NULL;
375 #endif /* #ifdef CONFIG_RCU_BOOST */
378 * If this was the last task on the current list, and if
379 * we aren't waiting on any CPUs, report the quiescent state.
380 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
381 * so we must take a snapshot of the expedited state.
383 empty_exp_now = !rcu_preempted_readers_exp(rnp);
384 if (!empty && !rcu_preempt_blocked_readers_cgp(rnp)) {
385 trace_rcu_quiescent_state_report("preempt_rcu",
392 rcu_report_unblock_qs_rnp(rnp, flags);
394 raw_spin_unlock_irqrestore(&rnp->lock, flags);
397 #ifdef CONFIG_RCU_BOOST
398 /* Unboost if we were boosted. */
400 rt_mutex_unlock(rbmp);
401 #endif /* #ifdef CONFIG_RCU_BOOST */
404 * If this was the last task on the expedited lists,
405 * then we need to report up the rcu_node hierarchy.
407 if (!empty_exp && empty_exp_now)
408 rcu_report_exp_rnp(&rcu_preempt_state, rnp, true);
410 local_irq_restore(flags);
414 #ifdef CONFIG_RCU_CPU_STALL_VERBOSE
417 * Dump detailed information for all tasks blocking the current RCU
418 * grace period on the specified rcu_node structure.
420 static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
423 struct task_struct *t;
425 raw_spin_lock_irqsave(&rnp->lock, flags);
426 if (!rcu_preempt_blocked_readers_cgp(rnp)) {
427 raw_spin_unlock_irqrestore(&rnp->lock, flags);
430 t = list_entry(rnp->gp_tasks,
431 struct task_struct, rcu_node_entry);
432 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry)
434 raw_spin_unlock_irqrestore(&rnp->lock, flags);
438 * Dump detailed information for all tasks blocking the current RCU
441 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
443 struct rcu_node *rnp = rcu_get_root(rsp);
445 rcu_print_detail_task_stall_rnp(rnp);
446 rcu_for_each_leaf_node(rsp, rnp)
447 rcu_print_detail_task_stall_rnp(rnp);
450 #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
452 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
456 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
458 #ifdef CONFIG_RCU_CPU_STALL_INFO
460 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
462 printk(KERN_ERR "\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
463 rnp->level, rnp->grplo, rnp->grphi);
466 static void rcu_print_task_stall_end(void)
468 printk(KERN_CONT "\n");
471 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
473 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
477 static void rcu_print_task_stall_end(void)
481 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */
484 * Scan the current list of tasks blocked within RCU read-side critical
485 * sections, printing out the tid of each.
487 static int rcu_print_task_stall(struct rcu_node *rnp)
489 struct task_struct *t;
492 if (!rcu_preempt_blocked_readers_cgp(rnp))
494 rcu_print_task_stall_begin(rnp);
495 t = list_entry(rnp->gp_tasks,
496 struct task_struct, rcu_node_entry);
497 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
498 printk(KERN_CONT " P%d", t->pid);
501 rcu_print_task_stall_end();
506 * Check that the list of blocked tasks for the newly completed grace
507 * period is in fact empty. It is a serious bug to complete a grace
508 * period that still has RCU readers blocked! This function must be
509 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
510 * must be held by the caller.
512 * Also, if there are blocked tasks on the list, they automatically
513 * block the newly created grace period, so set up ->gp_tasks accordingly.
515 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
517 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
518 if (!list_empty(&rnp->blkd_tasks))
519 rnp->gp_tasks = rnp->blkd_tasks.next;
520 WARN_ON_ONCE(rnp->qsmask);
523 #ifdef CONFIG_HOTPLUG_CPU
526 * Handle tasklist migration for case in which all CPUs covered by the
527 * specified rcu_node have gone offline. Move them up to the root
528 * rcu_node. The reason for not just moving them to the immediate
529 * parent is to remove the need for rcu_read_unlock_special() to
530 * make more than two attempts to acquire the target rcu_node's lock.
531 * Returns true if there were tasks blocking the current RCU grace
534 * Returns 1 if there was previously a task blocking the current grace
535 * period on the specified rcu_node structure.
537 * The caller must hold rnp->lock with irqs disabled.
539 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
540 struct rcu_node *rnp,
541 struct rcu_data *rdp)
543 struct list_head *lp;
544 struct list_head *lp_root;
546 struct rcu_node *rnp_root = rcu_get_root(rsp);
547 struct task_struct *t;
549 if (rnp == rnp_root) {
550 WARN_ONCE(1, "Last CPU thought to be offlined?");
551 return 0; /* Shouldn't happen: at least one CPU online. */
554 /* If we are on an internal node, complain bitterly. */
555 WARN_ON_ONCE(rnp != rdp->mynode);
558 * Move tasks up to root rcu_node. Don't try to get fancy for
559 * this corner-case operation -- just put this node's tasks
560 * at the head of the root node's list, and update the root node's
561 * ->gp_tasks and ->exp_tasks pointers to those of this node's,
562 * if non-NULL. This might result in waiting for more tasks than
563 * absolutely necessary, but this is a good performance/complexity
566 if (rcu_preempt_blocked_readers_cgp(rnp) && rnp->qsmask == 0)
567 retval |= RCU_OFL_TASKS_NORM_GP;
568 if (rcu_preempted_readers_exp(rnp))
569 retval |= RCU_OFL_TASKS_EXP_GP;
570 lp = &rnp->blkd_tasks;
571 lp_root = &rnp_root->blkd_tasks;
572 while (!list_empty(lp)) {
573 t = list_entry(lp->next, typeof(*t), rcu_node_entry);
574 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
575 list_del(&t->rcu_node_entry);
576 t->rcu_blocked_node = rnp_root;
577 list_add(&t->rcu_node_entry, lp_root);
578 if (&t->rcu_node_entry == rnp->gp_tasks)
579 rnp_root->gp_tasks = rnp->gp_tasks;
580 if (&t->rcu_node_entry == rnp->exp_tasks)
581 rnp_root->exp_tasks = rnp->exp_tasks;
582 #ifdef CONFIG_RCU_BOOST
583 if (&t->rcu_node_entry == rnp->boost_tasks)
584 rnp_root->boost_tasks = rnp->boost_tasks;
585 #endif /* #ifdef CONFIG_RCU_BOOST */
586 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
589 rnp->gp_tasks = NULL;
590 rnp->exp_tasks = NULL;
591 #ifdef CONFIG_RCU_BOOST
592 rnp->boost_tasks = NULL;
594 * In case root is being boosted and leaf was not. Make sure
595 * that we boost the tasks blocking the current grace period
598 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
599 if (rnp_root->boost_tasks != NULL &&
600 rnp_root->boost_tasks != rnp_root->gp_tasks &&
601 rnp_root->boost_tasks != rnp_root->exp_tasks)
602 rnp_root->boost_tasks = rnp_root->gp_tasks;
603 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
604 #endif /* #ifdef CONFIG_RCU_BOOST */
609 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
612 * Check for a quiescent state from the current CPU. When a task blocks,
613 * the task is recorded in the corresponding CPU's rcu_node structure,
614 * which is checked elsewhere.
616 * Caller must disable hard irqs.
618 static void rcu_preempt_check_callbacks(int cpu)
620 struct task_struct *t = current;
622 if (t->rcu_read_lock_nesting == 0) {
626 if (t->rcu_read_lock_nesting > 0 &&
627 per_cpu(rcu_preempt_data, cpu).qs_pending)
628 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
631 #ifdef CONFIG_RCU_BOOST
633 static void rcu_preempt_do_callbacks(void)
635 rcu_do_batch(&rcu_preempt_state, &__get_cpu_var(rcu_preempt_data));
638 #endif /* #ifdef CONFIG_RCU_BOOST */
641 * Queue a preemptible-RCU callback for invocation after a grace period.
643 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
645 __call_rcu(head, func, &rcu_preempt_state, 0);
647 EXPORT_SYMBOL_GPL(call_rcu);
650 * Queue an RCU callback for lazy invocation after a grace period.
651 * This will likely be later named something like "call_rcu_lazy()",
652 * but this change will require some way of tagging the lazy RCU
653 * callbacks in the list of pending callbacks. Until then, this
654 * function may only be called from __kfree_rcu().
656 void kfree_call_rcu(struct rcu_head *head,
657 void (*func)(struct rcu_head *rcu))
659 __call_rcu(head, func, &rcu_preempt_state, 1);
661 EXPORT_SYMBOL_GPL(kfree_call_rcu);
664 * synchronize_rcu - wait until a grace period has elapsed.
666 * Control will return to the caller some time after a full grace
667 * period has elapsed, in other words after all currently executing RCU
668 * read-side critical sections have completed. Note, however, that
669 * upon return from synchronize_rcu(), the caller might well be executing
670 * concurrently with new RCU read-side critical sections that began while
671 * synchronize_rcu() was waiting. RCU read-side critical sections are
672 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
674 void synchronize_rcu(void)
676 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
677 !lock_is_held(&rcu_lock_map) &&
678 !lock_is_held(&rcu_sched_lock_map),
679 "Illegal synchronize_rcu() in RCU read-side critical section");
680 if (!rcu_scheduler_active)
682 wait_rcu_gp(call_rcu);
684 EXPORT_SYMBOL_GPL(synchronize_rcu);
686 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
687 static long sync_rcu_preempt_exp_count;
688 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
691 * Return non-zero if there are any tasks in RCU read-side critical
692 * sections blocking the current preemptible-RCU expedited grace period.
693 * If there is no preemptible-RCU expedited grace period currently in
694 * progress, returns zero unconditionally.
696 static int rcu_preempted_readers_exp(struct rcu_node *rnp)
698 return rnp->exp_tasks != NULL;
702 * return non-zero if there is no RCU expedited grace period in progress
703 * for the specified rcu_node structure, in other words, if all CPUs and
704 * tasks covered by the specified rcu_node structure have done their bit
705 * for the current expedited grace period. Works only for preemptible
706 * RCU -- other RCU implementation use other means.
708 * Caller must hold sync_rcu_preempt_exp_mutex.
710 static int sync_rcu_preempt_exp_done(struct rcu_node *rnp)
712 return !rcu_preempted_readers_exp(rnp) &&
713 ACCESS_ONCE(rnp->expmask) == 0;
717 * Report the exit from RCU read-side critical section for the last task
718 * that queued itself during or before the current expedited preemptible-RCU
719 * grace period. This event is reported either to the rcu_node structure on
720 * which the task was queued or to one of that rcu_node structure's ancestors,
721 * recursively up the tree. (Calm down, calm down, we do the recursion
724 * Most callers will set the "wake" flag, but the task initiating the
725 * expedited grace period need not wake itself.
727 * Caller must hold sync_rcu_preempt_exp_mutex.
729 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
735 raw_spin_lock_irqsave(&rnp->lock, flags);
737 if (!sync_rcu_preempt_exp_done(rnp)) {
738 raw_spin_unlock_irqrestore(&rnp->lock, flags);
741 if (rnp->parent == NULL) {
742 raw_spin_unlock_irqrestore(&rnp->lock, flags);
744 wake_up(&sync_rcu_preempt_exp_wq);
748 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
750 raw_spin_lock(&rnp->lock); /* irqs already disabled */
751 rnp->expmask &= ~mask;
756 * Snapshot the tasks blocking the newly started preemptible-RCU expedited
757 * grace period for the specified rcu_node structure. If there are no such
758 * tasks, report it up the rcu_node hierarchy.
760 * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock.
763 sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp)
768 raw_spin_lock_irqsave(&rnp->lock, flags);
769 if (list_empty(&rnp->blkd_tasks)) {
770 raw_spin_unlock_irqrestore(&rnp->lock, flags);
772 rnp->exp_tasks = rnp->blkd_tasks.next;
773 rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
777 rcu_report_exp_rnp(rsp, rnp, false); /* Don't wake self. */
781 * synchronize_rcu_expedited - Brute-force RCU grace period
783 * Wait for an RCU-preempt grace period, but expedite it. The basic
784 * idea is to invoke synchronize_sched_expedited() to push all the tasks to
785 * the ->blkd_tasks lists and wait for this list to drain. This consumes
786 * significant time on all CPUs and is unfriendly to real-time workloads,
787 * so is thus not recommended for any sort of common-case code.
788 * In fact, if you are using synchronize_rcu_expedited() in a loop,
789 * please restructure your code to batch your updates, and then Use a
790 * single synchronize_rcu() instead.
792 * Note that it is illegal to call this function while holding any lock
793 * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal
794 * to call this function from a CPU-hotplug notifier. Failing to observe
795 * these restriction will result in deadlock.
797 void synchronize_rcu_expedited(void)
800 struct rcu_node *rnp;
801 struct rcu_state *rsp = &rcu_preempt_state;
805 smp_mb(); /* Caller's modifications seen first by other CPUs. */
806 snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1;
807 smp_mb(); /* Above access cannot bleed into critical section. */
810 * Acquire lock, falling back to synchronize_rcu() if too many
811 * lock-acquisition failures. Of course, if someone does the
812 * expedited grace period for us, just leave.
814 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) {
815 if (trycount++ < 10) {
816 udelay(trycount * num_online_cpus());
821 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
822 goto mb_ret; /* Others did our work for us. */
824 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
825 goto unlock_mb_ret; /* Others did our work for us. */
827 /* force all RCU readers onto ->blkd_tasks lists. */
828 synchronize_sched_expedited();
830 raw_spin_lock_irqsave(&rsp->onofflock, flags);
832 /* Initialize ->expmask for all non-leaf rcu_node structures. */
833 rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) {
834 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
835 rnp->expmask = rnp->qsmaskinit;
836 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
839 /* Snapshot current state of ->blkd_tasks lists. */
840 rcu_for_each_leaf_node(rsp, rnp)
841 sync_rcu_preempt_exp_init(rsp, rnp);
842 if (NUM_RCU_NODES > 1)
843 sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp));
845 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
847 /* Wait for snapshotted ->blkd_tasks lists to drain. */
848 rnp = rcu_get_root(rsp);
849 wait_event(sync_rcu_preempt_exp_wq,
850 sync_rcu_preempt_exp_done(rnp));
852 /* Clean up and exit. */
853 smp_mb(); /* ensure expedited GP seen before counter increment. */
854 ACCESS_ONCE(sync_rcu_preempt_exp_count)++;
856 mutex_unlock(&sync_rcu_preempt_exp_mutex);
858 smp_mb(); /* ensure subsequent action seen after grace period. */
860 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
863 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
865 void rcu_barrier(void)
867 _rcu_barrier(&rcu_preempt_state);
869 EXPORT_SYMBOL_GPL(rcu_barrier);
872 * Initialize preemptible RCU's state structures.
874 static void __init __rcu_init_preempt(void)
876 rcu_init_one(&rcu_preempt_state, &rcu_preempt_data);
879 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
881 static struct rcu_state *rcu_state = &rcu_sched_state;
884 * Tell them what RCU they are running.
886 static void __init rcu_bootup_announce(void)
888 printk(KERN_INFO "Hierarchical RCU implementation.\n");
889 rcu_bootup_announce_oddness();
893 * Return the number of RCU batches processed thus far for debug & stats.
895 long rcu_batches_completed(void)
897 return rcu_batches_completed_sched();
899 EXPORT_SYMBOL_GPL(rcu_batches_completed);
902 * Force a quiescent state for RCU, which, because there is no preemptible
903 * RCU, becomes the same as rcu-sched.
905 void rcu_force_quiescent_state(void)
907 rcu_sched_force_quiescent_state();
909 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
912 * Because preemptible RCU does not exist, we never have to check for
913 * CPUs being in quiescent states.
915 static void rcu_preempt_note_context_switch(int cpu)
920 * Because preemptible RCU does not exist, there are never any preempted
923 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
928 #ifdef CONFIG_HOTPLUG_CPU
930 /* Because preemptible RCU does not exist, no quieting of tasks. */
931 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
933 raw_spin_unlock_irqrestore(&rnp->lock, flags);
936 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
939 * Because preemptible RCU does not exist, we never have to check for
940 * tasks blocked within RCU read-side critical sections.
942 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
947 * Because preemptible RCU does not exist, we never have to check for
948 * tasks blocked within RCU read-side critical sections.
950 static int rcu_print_task_stall(struct rcu_node *rnp)
956 * Because there is no preemptible RCU, there can be no readers blocked,
957 * so there is no need to check for blocked tasks. So check only for
958 * bogus qsmask values.
960 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
962 WARN_ON_ONCE(rnp->qsmask);
965 #ifdef CONFIG_HOTPLUG_CPU
968 * Because preemptible RCU does not exist, it never needs to migrate
969 * tasks that were blocked within RCU read-side critical sections, and
970 * such non-existent tasks cannot possibly have been blocking the current
973 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
974 struct rcu_node *rnp,
975 struct rcu_data *rdp)
980 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
983 * Because preemptible RCU does not exist, it never has any callbacks
986 static void rcu_preempt_check_callbacks(int cpu)
991 * Queue an RCU callback for lazy invocation after a grace period.
992 * This will likely be later named something like "call_rcu_lazy()",
993 * but this change will require some way of tagging the lazy RCU
994 * callbacks in the list of pending callbacks. Until then, this
995 * function may only be called from __kfree_rcu().
997 * Because there is no preemptible RCU, we use RCU-sched instead.
999 void kfree_call_rcu(struct rcu_head *head,
1000 void (*func)(struct rcu_head *rcu))
1002 __call_rcu(head, func, &rcu_sched_state, 1);
1004 EXPORT_SYMBOL_GPL(kfree_call_rcu);
1007 * Wait for an rcu-preempt grace period, but make it happen quickly.
1008 * But because preemptible RCU does not exist, map to rcu-sched.
1010 void synchronize_rcu_expedited(void)
1012 synchronize_sched_expedited();
1014 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
1016 #ifdef CONFIG_HOTPLUG_CPU
1019 * Because preemptible RCU does not exist, there is never any need to
1020 * report on tasks preempted in RCU read-side critical sections during
1021 * expedited RCU grace periods.
1023 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
1028 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1031 * Because preemptible RCU does not exist, rcu_barrier() is just
1032 * another name for rcu_barrier_sched().
1034 void rcu_barrier(void)
1036 rcu_barrier_sched();
1038 EXPORT_SYMBOL_GPL(rcu_barrier);
1041 * Because preemptible RCU does not exist, it need not be initialized.
1043 static void __init __rcu_init_preempt(void)
1047 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1049 #ifdef CONFIG_RCU_BOOST
1051 #include "rtmutex_common.h"
1053 #ifdef CONFIG_RCU_TRACE
1055 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1057 if (list_empty(&rnp->blkd_tasks))
1058 rnp->n_balk_blkd_tasks++;
1059 else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL)
1060 rnp->n_balk_exp_gp_tasks++;
1061 else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL)
1062 rnp->n_balk_boost_tasks++;
1063 else if (rnp->gp_tasks != NULL && rnp->qsmask != 0)
1064 rnp->n_balk_notblocked++;
1065 else if (rnp->gp_tasks != NULL &&
1066 ULONG_CMP_LT(jiffies, rnp->boost_time))
1067 rnp->n_balk_notyet++;
1072 #else /* #ifdef CONFIG_RCU_TRACE */
1074 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1078 #endif /* #else #ifdef CONFIG_RCU_TRACE */
1081 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1082 * or ->boost_tasks, advancing the pointer to the next task in the
1083 * ->blkd_tasks list.
1085 * Note that irqs must be enabled: boosting the task can block.
1086 * Returns 1 if there are more tasks needing to be boosted.
1088 static int rcu_boost(struct rcu_node *rnp)
1090 unsigned long flags;
1091 struct rt_mutex mtx;
1092 struct task_struct *t;
1093 struct list_head *tb;
1095 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL)
1096 return 0; /* Nothing left to boost. */
1098 raw_spin_lock_irqsave(&rnp->lock, flags);
1101 * Recheck under the lock: all tasks in need of boosting
1102 * might exit their RCU read-side critical sections on their own.
1104 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1105 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1110 * Preferentially boost tasks blocking expedited grace periods.
1111 * This cannot starve the normal grace periods because a second
1112 * expedited grace period must boost all blocked tasks, including
1113 * those blocking the pre-existing normal grace period.
1115 if (rnp->exp_tasks != NULL) {
1116 tb = rnp->exp_tasks;
1117 rnp->n_exp_boosts++;
1119 tb = rnp->boost_tasks;
1120 rnp->n_normal_boosts++;
1122 rnp->n_tasks_boosted++;
1125 * We boost task t by manufacturing an rt_mutex that appears to
1126 * be held by task t. We leave a pointer to that rt_mutex where
1127 * task t can find it, and task t will release the mutex when it
1128 * exits its outermost RCU read-side critical section. Then
1129 * simply acquiring this artificial rt_mutex will boost task
1130 * t's priority. (Thanks to tglx for suggesting this approach!)
1132 * Note that task t must acquire rnp->lock to remove itself from
1133 * the ->blkd_tasks list, which it will do from exit() if from
1134 * nowhere else. We therefore are guaranteed that task t will
1135 * stay around at least until we drop rnp->lock. Note that
1136 * rnp->lock also resolves races between our priority boosting
1137 * and task t's exiting its outermost RCU read-side critical
1140 t = container_of(tb, struct task_struct, rcu_node_entry);
1141 rt_mutex_init_proxy_locked(&mtx, t);
1142 t->rcu_boost_mutex = &mtx;
1143 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1144 rt_mutex_lock(&mtx); /* Side effect: boosts task t's priority. */
1145 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */
1147 return ACCESS_ONCE(rnp->exp_tasks) != NULL ||
1148 ACCESS_ONCE(rnp->boost_tasks) != NULL;
1152 * Timer handler to initiate waking up of boost kthreads that
1153 * have yielded the CPU due to excessive numbers of tasks to
1154 * boost. We wake up the per-rcu_node kthread, which in turn
1155 * will wake up the booster kthread.
1157 static void rcu_boost_kthread_timer(unsigned long arg)
1159 invoke_rcu_node_kthread((struct rcu_node *)arg);
1163 * Priority-boosting kthread. One per leaf rcu_node and one for the
1166 static int rcu_boost_kthread(void *arg)
1168 struct rcu_node *rnp = (struct rcu_node *)arg;
1172 trace_rcu_utilization("Start boost kthread@init");
1174 rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
1175 trace_rcu_utilization("End boost kthread@rcu_wait");
1176 rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
1177 trace_rcu_utilization("Start boost kthread@rcu_wait");
1178 rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
1179 more2boost = rcu_boost(rnp);
1185 trace_rcu_utilization("End boost kthread@rcu_yield");
1186 rcu_yield(rcu_boost_kthread_timer, (unsigned long)rnp);
1187 trace_rcu_utilization("Start boost kthread@rcu_yield");
1192 trace_rcu_utilization("End boost kthread@notreached");
1197 * Check to see if it is time to start boosting RCU readers that are
1198 * blocking the current grace period, and, if so, tell the per-rcu_node
1199 * kthread to start boosting them. If there is an expedited grace
1200 * period in progress, it is always time to boost.
1202 * The caller must hold rnp->lock, which this function releases.
1203 * The ->boost_kthread_task is immortal, so we don't need to worry
1204 * about it going away.
1206 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1208 struct task_struct *t;
1210 if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
1211 rnp->n_balk_exp_gp_tasks++;
1212 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1215 if (rnp->exp_tasks != NULL ||
1216 (rnp->gp_tasks != NULL &&
1217 rnp->boost_tasks == NULL &&
1219 ULONG_CMP_GE(jiffies, rnp->boost_time))) {
1220 if (rnp->exp_tasks == NULL)
1221 rnp->boost_tasks = rnp->gp_tasks;
1222 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1223 t = rnp->boost_kthread_task;
1227 rcu_initiate_boost_trace(rnp);
1228 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1233 * Wake up the per-CPU kthread to invoke RCU callbacks.
1235 static void invoke_rcu_callbacks_kthread(void)
1237 unsigned long flags;
1239 local_irq_save(flags);
1240 __this_cpu_write(rcu_cpu_has_work, 1);
1241 if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
1242 current != __this_cpu_read(rcu_cpu_kthread_task))
1243 wake_up_process(__this_cpu_read(rcu_cpu_kthread_task));
1244 local_irq_restore(flags);
1248 * Is the current CPU running the RCU-callbacks kthread?
1249 * Caller must have preemption disabled.
1251 static bool rcu_is_callbacks_kthread(void)
1253 return __get_cpu_var(rcu_cpu_kthread_task) == current;
1257 * Set the affinity of the boost kthread. The CPU-hotplug locks are
1258 * held, so no one should be messing with the existence of the boost
1261 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
1264 struct task_struct *t;
1266 t = rnp->boost_kthread_task;
1268 set_cpus_allowed_ptr(rnp->boost_kthread_task, cm);
1271 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1274 * Do priority-boost accounting for the start of a new grace period.
1276 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1278 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1282 * Create an RCU-boost kthread for the specified node if one does not
1283 * already exist. We only create this kthread for preemptible RCU.
1284 * Returns zero if all is well, a negated errno otherwise.
1286 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
1287 struct rcu_node *rnp,
1290 unsigned long flags;
1291 struct sched_param sp;
1292 struct task_struct *t;
1294 if (&rcu_preempt_state != rsp)
1297 if (rnp->boost_kthread_task != NULL)
1299 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1300 "rcub/%d", rnp_index);
1303 raw_spin_lock_irqsave(&rnp->lock, flags);
1304 rnp->boost_kthread_task = t;
1305 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1306 sp.sched_priority = RCU_BOOST_PRIO;
1307 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1308 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1312 #ifdef CONFIG_HOTPLUG_CPU
1315 * Stop the RCU's per-CPU kthread when its CPU goes offline,.
1317 static void rcu_stop_cpu_kthread(int cpu)
1319 struct task_struct *t;
1321 /* Stop the CPU's kthread. */
1322 t = per_cpu(rcu_cpu_kthread_task, cpu);
1324 per_cpu(rcu_cpu_kthread_task, cpu) = NULL;
1329 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1331 static void rcu_kthread_do_work(void)
1333 rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data));
1334 rcu_do_batch(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1335 rcu_preempt_do_callbacks();
1339 * Wake up the specified per-rcu_node-structure kthread.
1340 * Because the per-rcu_node kthreads are immortal, we don't need
1341 * to do anything to keep them alive.
1343 static void invoke_rcu_node_kthread(struct rcu_node *rnp)
1345 struct task_struct *t;
1347 t = rnp->node_kthread_task;
1353 * Set the specified CPU's kthread to run RT or not, as specified by
1354 * the to_rt argument. The CPU-hotplug locks are held, so the task
1355 * is not going away.
1357 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1360 struct sched_param sp;
1361 struct task_struct *t;
1363 t = per_cpu(rcu_cpu_kthread_task, cpu);
1367 policy = SCHED_FIFO;
1368 sp.sched_priority = RCU_KTHREAD_PRIO;
1370 policy = SCHED_NORMAL;
1371 sp.sched_priority = 0;
1373 sched_setscheduler_nocheck(t, policy, &sp);
1377 * Timer handler to initiate the waking up of per-CPU kthreads that
1378 * have yielded the CPU due to excess numbers of RCU callbacks.
1379 * We wake up the per-rcu_node kthread, which in turn will wake up
1380 * the booster kthread.
1382 static void rcu_cpu_kthread_timer(unsigned long arg)
1384 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg);
1385 struct rcu_node *rnp = rdp->mynode;
1387 atomic_or(rdp->grpmask, &rnp->wakemask);
1388 invoke_rcu_node_kthread(rnp);
1392 * Drop to non-real-time priority and yield, but only after posting a
1393 * timer that will cause us to regain our real-time priority if we
1394 * remain preempted. Either way, we restore our real-time priority
1397 static void rcu_yield(void (*f)(unsigned long), unsigned long arg)
1399 struct sched_param sp;
1400 struct timer_list yield_timer;
1401 int prio = current->rt_priority;
1403 setup_timer_on_stack(&yield_timer, f, arg);
1404 mod_timer(&yield_timer, jiffies + 2);
1405 sp.sched_priority = 0;
1406 sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
1407 set_user_nice(current, 19);
1409 set_user_nice(current, 0);
1410 sp.sched_priority = prio;
1411 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1412 del_timer(&yield_timer);
1416 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1417 * This can happen while the corresponding CPU is either coming online
1418 * or going offline. We cannot wait until the CPU is fully online
1419 * before starting the kthread, because the various notifier functions
1420 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1421 * the corresponding CPU is online.
1423 * Return 1 if the kthread needs to stop, 0 otherwise.
1425 * Caller must disable bh. This function can momentarily enable it.
1427 static int rcu_cpu_kthread_should_stop(int cpu)
1429 while (cpu_is_offline(cpu) ||
1430 !cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)) ||
1431 smp_processor_id() != cpu) {
1432 if (kthread_should_stop())
1434 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
1435 per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id();
1437 schedule_timeout_uninterruptible(1);
1438 if (!cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)))
1439 set_cpus_allowed_ptr(current, cpumask_of(cpu));
1442 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1447 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1448 * RCU softirq used in flavors and configurations of RCU that do not
1449 * support RCU priority boosting.
1451 static int rcu_cpu_kthread(void *arg)
1453 int cpu = (int)(long)arg;
1454 unsigned long flags;
1456 unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu);
1458 char *workp = &per_cpu(rcu_cpu_has_work, cpu);
1460 trace_rcu_utilization("Start CPU kthread@init");
1462 *statusp = RCU_KTHREAD_WAITING;
1463 trace_rcu_utilization("End CPU kthread@rcu_wait");
1464 rcu_wait(*workp != 0 || kthread_should_stop());
1465 trace_rcu_utilization("Start CPU kthread@rcu_wait");
1467 if (rcu_cpu_kthread_should_stop(cpu)) {
1471 *statusp = RCU_KTHREAD_RUNNING;
1472 per_cpu(rcu_cpu_kthread_loops, cpu)++;
1473 local_irq_save(flags);
1476 local_irq_restore(flags);
1478 rcu_kthread_do_work();
1485 *statusp = RCU_KTHREAD_YIELDING;
1486 trace_rcu_utilization("End CPU kthread@rcu_yield");
1487 rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu);
1488 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1492 *statusp = RCU_KTHREAD_STOPPED;
1493 trace_rcu_utilization("End CPU kthread@term");
1498 * Spawn a per-CPU kthread, setting up affinity and priority.
1499 * Because the CPU hotplug lock is held, no other CPU will be attempting
1500 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1501 * attempting to access it during boot, but the locking in kthread_bind()
1502 * will enforce sufficient ordering.
1504 * Please note that we cannot simply refuse to wake up the per-CPU
1505 * kthread because kthreads are created in TASK_UNINTERRUPTIBLE state,
1506 * which can result in softlockup complaints if the task ends up being
1507 * idle for more than a couple of minutes.
1509 * However, please note also that we cannot bind the per-CPU kthread to its
1510 * CPU until that CPU is fully online. We also cannot wait until the
1511 * CPU is fully online before we create its per-CPU kthread, as this would
1512 * deadlock the system when CPU notifiers tried waiting for grace
1513 * periods. So we bind the per-CPU kthread to its CPU only if the CPU
1514 * is online. If its CPU is not yet fully online, then the code in
1515 * rcu_cpu_kthread() will wait until it is fully online, and then do
1518 static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu)
1520 struct sched_param sp;
1521 struct task_struct *t;
1523 if (!rcu_scheduler_fully_active ||
1524 per_cpu(rcu_cpu_kthread_task, cpu) != NULL)
1526 t = kthread_create_on_node(rcu_cpu_kthread,
1532 if (cpu_online(cpu))
1533 kthread_bind(t, cpu);
1534 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1535 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL);
1536 sp.sched_priority = RCU_KTHREAD_PRIO;
1537 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1538 per_cpu(rcu_cpu_kthread_task, cpu) = t;
1539 wake_up_process(t); /* Get to TASK_INTERRUPTIBLE quickly. */
1544 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1545 * kthreads when needed. We ignore requests to wake up kthreads
1546 * for offline CPUs, which is OK because force_quiescent_state()
1547 * takes care of this case.
1549 static int rcu_node_kthread(void *arg)
1552 unsigned long flags;
1554 struct rcu_node *rnp = (struct rcu_node *)arg;
1555 struct sched_param sp;
1556 struct task_struct *t;
1559 rnp->node_kthread_status = RCU_KTHREAD_WAITING;
1560 rcu_wait(atomic_read(&rnp->wakemask) != 0);
1561 rnp->node_kthread_status = RCU_KTHREAD_RUNNING;
1562 raw_spin_lock_irqsave(&rnp->lock, flags);
1563 mask = atomic_xchg(&rnp->wakemask, 0);
1564 rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
1565 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) {
1566 if ((mask & 0x1) == 0)
1569 t = per_cpu(rcu_cpu_kthread_task, cpu);
1570 if (!cpu_online(cpu) || t == NULL) {
1574 per_cpu(rcu_cpu_has_work, cpu) = 1;
1575 sp.sched_priority = RCU_KTHREAD_PRIO;
1576 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1581 rnp->node_kthread_status = RCU_KTHREAD_STOPPED;
1586 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1587 * served by the rcu_node in question. The CPU hotplug lock is still
1588 * held, so the value of rnp->qsmaskinit will be stable.
1590 * We don't include outgoingcpu in the affinity set, use -1 if there is
1591 * no outgoing CPU. If there are no CPUs left in the affinity set,
1592 * this function allows the kthread to execute on any CPU.
1594 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1598 unsigned long mask = rnp->qsmaskinit;
1600 if (rnp->node_kthread_task == NULL)
1602 if (!alloc_cpumask_var(&cm, GFP_KERNEL))
1605 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1)
1606 if ((mask & 0x1) && cpu != outgoingcpu)
1607 cpumask_set_cpu(cpu, cm);
1608 if (cpumask_weight(cm) == 0) {
1610 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++)
1611 cpumask_clear_cpu(cpu, cm);
1612 WARN_ON_ONCE(cpumask_weight(cm) == 0);
1614 set_cpus_allowed_ptr(rnp->node_kthread_task, cm);
1615 rcu_boost_kthread_setaffinity(rnp, cm);
1616 free_cpumask_var(cm);
1620 * Spawn a per-rcu_node kthread, setting priority and affinity.
1621 * Called during boot before online/offline can happen, or, if
1622 * during runtime, with the main CPU-hotplug locks held. So only
1623 * one of these can be executing at a time.
1625 static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp,
1626 struct rcu_node *rnp)
1628 unsigned long flags;
1629 int rnp_index = rnp - &rsp->node[0];
1630 struct sched_param sp;
1631 struct task_struct *t;
1633 if (!rcu_scheduler_fully_active ||
1634 rnp->qsmaskinit == 0)
1636 if (rnp->node_kthread_task == NULL) {
1637 t = kthread_create(rcu_node_kthread, (void *)rnp,
1638 "rcun/%d", rnp_index);
1641 raw_spin_lock_irqsave(&rnp->lock, flags);
1642 rnp->node_kthread_task = t;
1643 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1644 sp.sched_priority = 99;
1645 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1646 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1648 return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index);
1652 * Spawn all kthreads -- called as soon as the scheduler is running.
1654 static int __init rcu_spawn_kthreads(void)
1657 struct rcu_node *rnp;
1659 rcu_scheduler_fully_active = 1;
1660 for_each_possible_cpu(cpu) {
1661 per_cpu(rcu_cpu_has_work, cpu) = 0;
1662 if (cpu_online(cpu))
1663 (void)rcu_spawn_one_cpu_kthread(cpu);
1665 rnp = rcu_get_root(rcu_state);
1666 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1667 if (NUM_RCU_NODES > 1) {
1668 rcu_for_each_leaf_node(rcu_state, rnp)
1669 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1673 early_initcall(rcu_spawn_kthreads);
1675 static void __cpuinit rcu_prepare_kthreads(int cpu)
1677 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
1678 struct rcu_node *rnp = rdp->mynode;
1680 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1681 if (rcu_scheduler_fully_active) {
1682 (void)rcu_spawn_one_cpu_kthread(cpu);
1683 if (rnp->node_kthread_task == NULL)
1684 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1688 #else /* #ifdef CONFIG_RCU_BOOST */
1690 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1692 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1695 static void invoke_rcu_callbacks_kthread(void)
1700 static bool rcu_is_callbacks_kthread(void)
1705 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1709 #ifdef CONFIG_HOTPLUG_CPU
1711 static void rcu_stop_cpu_kthread(int cpu)
1715 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1717 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1721 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1725 static int __init rcu_scheduler_really_started(void)
1727 rcu_scheduler_fully_active = 1;
1730 early_initcall(rcu_scheduler_really_started);
1732 static void __cpuinit rcu_prepare_kthreads(int cpu)
1736 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1738 #if !defined(CONFIG_RCU_FAST_NO_HZ)
1741 * Check to see if any future RCU-related work will need to be done
1742 * by the current CPU, even if none need be done immediately, returning
1743 * 1 if so. This function is part of the RCU implementation; it is -not-
1744 * an exported member of the RCU API.
1746 * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
1747 * any flavor of RCU.
1749 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1751 *delta_jiffies = ULONG_MAX;
1752 return rcu_cpu_has_callbacks(cpu);
1756 * Because we do not have RCU_FAST_NO_HZ, don't bother initializing for it.
1758 static void rcu_prepare_for_idle_init(int cpu)
1763 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1766 static void rcu_cleanup_after_idle(int cpu)
1771 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
1774 static void rcu_prepare_for_idle(int cpu)
1779 * Don't bother keeping a running count of the number of RCU callbacks
1780 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1782 static void rcu_idle_count_callbacks_posted(void)
1786 #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1789 * This code is invoked when a CPU goes idle, at which point we want
1790 * to have the CPU do everything required for RCU so that it can enter
1791 * the energy-efficient dyntick-idle mode. This is handled by a
1792 * state machine implemented by rcu_prepare_for_idle() below.
1794 * The following three proprocessor symbols control this state machine:
1796 * RCU_IDLE_FLUSHES gives the maximum number of times that we will attempt
1797 * to satisfy RCU. Beyond this point, it is better to incur a periodic
1798 * scheduling-clock interrupt than to loop through the state machine
1800 * RCU_IDLE_OPT_FLUSHES gives the number of RCU_IDLE_FLUSHES that are
1801 * optional if RCU does not need anything immediately from this
1802 * CPU, even if this CPU still has RCU callbacks queued. The first
1803 * times through the state machine are mandatory: we need to give
1804 * the state machine a chance to communicate a quiescent state
1806 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1807 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1808 * is sized to be roughly one RCU grace period. Those energy-efficiency
1809 * benchmarkers who might otherwise be tempted to set this to a large
1810 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1811 * system. And if you are -that- concerned about energy efficiency,
1812 * just power the system down and be done with it!
1813 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1814 * permitted to sleep in dyntick-idle mode with only lazy RCU
1815 * callbacks pending. Setting this too high can OOM your system.
1817 * The values below work well in practice. If future workloads require
1818 * adjustment, they can be converted into kernel config parameters, though
1819 * making the state machine smarter might be a better option.
1821 #define RCU_IDLE_FLUSHES 5 /* Number of dyntick-idle tries. */
1822 #define RCU_IDLE_OPT_FLUSHES 3 /* Optional dyntick-idle tries. */
1823 #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */
1824 #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
1826 extern int tick_nohz_enabled;
1829 * Does the specified flavor of RCU have non-lazy callbacks pending on
1830 * the specified CPU? Both RCU flavor and CPU are specified by the
1831 * rcu_data structure.
1833 static bool __rcu_cpu_has_nonlazy_callbacks(struct rcu_data *rdp)
1835 return rdp->qlen != rdp->qlen_lazy;
1838 #ifdef CONFIG_TREE_PREEMPT_RCU
1841 * Are there non-lazy RCU-preempt callbacks? (There cannot be if there
1842 * is no RCU-preempt in the kernel.)
1844 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1846 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
1848 return __rcu_cpu_has_nonlazy_callbacks(rdp);
1851 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1853 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1858 #endif /* else #ifdef CONFIG_TREE_PREEMPT_RCU */
1861 * Does any flavor of RCU have non-lazy callbacks on the specified CPU?
1863 static bool rcu_cpu_has_nonlazy_callbacks(int cpu)
1865 return __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_sched_data, cpu)) ||
1866 __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_bh_data, cpu)) ||
1867 rcu_preempt_cpu_has_nonlazy_callbacks(cpu);
1871 * Allow the CPU to enter dyntick-idle mode if either: (1) There are no
1872 * callbacks on this CPU, (2) this CPU has not yet attempted to enter
1873 * dyntick-idle mode, or (3) this CPU is in the process of attempting to
1874 * enter dyntick-idle mode. Otherwise, if we have recently tried and failed
1875 * to enter dyntick-idle mode, we refuse to try to enter it. After all,
1876 * it is better to incur scheduling-clock interrupts than to spin
1877 * continuously for the same time duration!
1879 * The delta_jiffies argument is used to store the time when RCU is
1880 * going to need the CPU again if it still has callbacks. The reason
1881 * for this is that rcu_prepare_for_idle() might need to post a timer,
1882 * but if so, it will do so after tick_nohz_stop_sched_tick() has set
1883 * the wakeup time for this CPU. This means that RCU's timer can be
1884 * delayed until the wakeup time, which defeats the purpose of posting
1887 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1889 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1891 /* Flag a new idle sojourn to the idle-entry state machine. */
1892 rdtp->idle_first_pass = 1;
1893 /* If no callbacks, RCU doesn't need the CPU. */
1894 if (!rcu_cpu_has_callbacks(cpu)) {
1895 *delta_jiffies = ULONG_MAX;
1898 if (rdtp->dyntick_holdoff == jiffies) {
1899 /* RCU recently tried and failed, so don't try again. */
1903 /* Set up for the possibility that RCU will post a timer. */
1904 if (rcu_cpu_has_nonlazy_callbacks(cpu)) {
1905 *delta_jiffies = round_up(RCU_IDLE_GP_DELAY + jiffies,
1906 RCU_IDLE_GP_DELAY) - jiffies;
1908 *delta_jiffies = jiffies + RCU_IDLE_LAZY_GP_DELAY;
1909 *delta_jiffies = round_jiffies(*delta_jiffies) - jiffies;
1915 * Handler for smp_call_function_single(). The only point of this
1916 * handler is to wake the CPU up, so the handler does only tracing.
1918 void rcu_idle_demigrate(void *unused)
1920 trace_rcu_prep_idle("Demigrate");
1924 * Timer handler used to force CPU to start pushing its remaining RCU
1925 * callbacks in the case where it entered dyntick-idle mode with callbacks
1926 * pending. The hander doesn't really need to do anything because the
1927 * real work is done upon re-entry to idle, or by the next scheduling-clock
1928 * interrupt should idle not be re-entered.
1930 * One special case: the timer gets migrated without awakening the CPU
1931 * on which the timer was scheduled on. In this case, we must wake up
1932 * that CPU. We do so with smp_call_function_single().
1934 static void rcu_idle_gp_timer_func(unsigned long cpu_in)
1936 int cpu = (int)cpu_in;
1938 trace_rcu_prep_idle("Timer");
1939 if (cpu != smp_processor_id())
1940 smp_call_function_single(cpu, rcu_idle_demigrate, NULL, 0);
1942 WARN_ON_ONCE(1); /* Getting here can hang the system... */
1946 * Initialize the timer used to pull CPUs out of dyntick-idle mode.
1948 static void rcu_prepare_for_idle_init(int cpu)
1950 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1952 rdtp->dyntick_holdoff = jiffies - 1;
1953 setup_timer(&rdtp->idle_gp_timer, rcu_idle_gp_timer_func, cpu);
1954 rdtp->idle_gp_timer_expires = jiffies - 1;
1955 rdtp->idle_first_pass = 1;
1959 * Clean up for exit from idle. Because we are exiting from idle, there
1960 * is no longer any point to ->idle_gp_timer, so cancel it. This will
1961 * do nothing if this timer is not active, so just cancel it unconditionally.
1963 static void rcu_cleanup_after_idle(int cpu)
1965 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1967 del_timer(&rdtp->idle_gp_timer);
1968 trace_rcu_prep_idle("Cleanup after idle");
1969 rdtp->tick_nohz_enabled_snap = ACCESS_ONCE(tick_nohz_enabled);
1973 * Check to see if any RCU-related work can be done by the current CPU,
1974 * and if so, schedule a softirq to get it done. This function is part
1975 * of the RCU implementation; it is -not- an exported member of the RCU API.
1977 * The idea is for the current CPU to clear out all work required by the
1978 * RCU core for the current grace period, so that this CPU can be permitted
1979 * to enter dyntick-idle mode. In some cases, it will need to be awakened
1980 * at the end of the grace period by whatever CPU ends the grace period.
1981 * This allows CPUs to go dyntick-idle more quickly, and to reduce the
1982 * number of wakeups by a modest integer factor.
1984 * Because it is not legal to invoke rcu_process_callbacks() with irqs
1985 * disabled, we do one pass of force_quiescent_state(), then do a
1986 * invoke_rcu_core() to cause rcu_process_callbacks() to be invoked
1987 * later. The ->dyntick_drain field controls the sequencing.
1989 * The caller must have disabled interrupts.
1991 static void rcu_prepare_for_idle(int cpu)
1993 struct timer_list *tp;
1994 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1997 /* Handle nohz enablement switches conservatively. */
1998 tne = ACCESS_ONCE(tick_nohz_enabled);
1999 if (tne != rdtp->tick_nohz_enabled_snap) {
2000 if (rcu_cpu_has_callbacks(cpu))
2001 invoke_rcu_core(); /* force nohz to see update. */
2002 rdtp->tick_nohz_enabled_snap = tne;
2009 * If this is an idle re-entry, for example, due to use of
2010 * RCU_NONIDLE() or the new idle-loop tracing API within the idle
2011 * loop, then don't take any state-machine actions, unless the
2012 * momentary exit from idle queued additional non-lazy callbacks.
2013 * Instead, repost the ->idle_gp_timer if this CPU has callbacks
2016 if (!rdtp->idle_first_pass &&
2017 (rdtp->nonlazy_posted == rdtp->nonlazy_posted_snap)) {
2018 if (rcu_cpu_has_callbacks(cpu)) {
2019 tp = &rdtp->idle_gp_timer;
2020 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2024 rdtp->idle_first_pass = 0;
2025 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted - 1;
2028 * If there are no callbacks on this CPU, enter dyntick-idle mode.
2029 * Also reset state to avoid prejudicing later attempts.
2031 if (!rcu_cpu_has_callbacks(cpu)) {
2032 rdtp->dyntick_holdoff = jiffies - 1;
2033 rdtp->dyntick_drain = 0;
2034 trace_rcu_prep_idle("No callbacks");
2039 * If in holdoff mode, just return. We will presumably have
2040 * refrained from disabling the scheduling-clock tick.
2042 if (rdtp->dyntick_holdoff == jiffies) {
2043 trace_rcu_prep_idle("In holdoff");
2047 /* Check and update the ->dyntick_drain sequencing. */
2048 if (rdtp->dyntick_drain <= 0) {
2049 /* First time through, initialize the counter. */
2050 rdtp->dyntick_drain = RCU_IDLE_FLUSHES;
2051 } else if (rdtp->dyntick_drain <= RCU_IDLE_OPT_FLUSHES &&
2052 !rcu_pending(cpu) &&
2053 !local_softirq_pending()) {
2054 /* Can we go dyntick-idle despite still having callbacks? */
2055 rdtp->dyntick_drain = 0;
2056 rdtp->dyntick_holdoff = jiffies;
2057 if (rcu_cpu_has_nonlazy_callbacks(cpu)) {
2058 trace_rcu_prep_idle("Dyntick with callbacks");
2059 rdtp->idle_gp_timer_expires =
2060 round_up(jiffies + RCU_IDLE_GP_DELAY,
2063 rdtp->idle_gp_timer_expires =
2064 round_jiffies(jiffies + RCU_IDLE_LAZY_GP_DELAY);
2065 trace_rcu_prep_idle("Dyntick with lazy callbacks");
2067 tp = &rdtp->idle_gp_timer;
2068 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2069 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted;
2070 return; /* Nothing more to do immediately. */
2071 } else if (--(rdtp->dyntick_drain) <= 0) {
2072 /* We have hit the limit, so time to give up. */
2073 rdtp->dyntick_holdoff = jiffies;
2074 trace_rcu_prep_idle("Begin holdoff");
2075 invoke_rcu_core(); /* Force the CPU out of dyntick-idle. */
2080 * Do one step of pushing the remaining RCU callbacks through
2081 * the RCU core state machine.
2083 #ifdef CONFIG_TREE_PREEMPT_RCU
2084 if (per_cpu(rcu_preempt_data, cpu).nxtlist) {
2085 rcu_preempt_qs(cpu);
2086 force_quiescent_state(&rcu_preempt_state, 0);
2088 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2089 if (per_cpu(rcu_sched_data, cpu).nxtlist) {
2091 force_quiescent_state(&rcu_sched_state, 0);
2093 if (per_cpu(rcu_bh_data, cpu).nxtlist) {
2095 force_quiescent_state(&rcu_bh_state, 0);
2099 * If RCU callbacks are still pending, RCU still needs this CPU.
2100 * So try forcing the callbacks through the grace period.
2102 if (rcu_cpu_has_callbacks(cpu)) {
2103 trace_rcu_prep_idle("More callbacks");
2106 trace_rcu_prep_idle("Callbacks drained");
2111 * Keep a running count of the number of non-lazy callbacks posted
2112 * on this CPU. This running counter (which is never decremented) allows
2113 * rcu_prepare_for_idle() to detect when something out of the idle loop
2114 * posts a callback, even if an equal number of callbacks are invoked.
2115 * Of course, callbacks should only be posted from within a trace event
2116 * designed to be called from idle or from within RCU_NONIDLE().
2118 static void rcu_idle_count_callbacks_posted(void)
2120 __this_cpu_add(rcu_dynticks.nonlazy_posted, 1);
2123 #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
2125 #ifdef CONFIG_RCU_CPU_STALL_INFO
2127 #ifdef CONFIG_RCU_FAST_NO_HZ
2129 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2131 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2132 struct timer_list *tltp = &rdtp->idle_gp_timer;
2134 sprintf(cp, "drain=%d %c timer=%lu",
2135 rdtp->dyntick_drain,
2136 rdtp->dyntick_holdoff == jiffies ? 'H' : '.',
2137 timer_pending(tltp) ? tltp->expires - jiffies : -1);
2140 #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
2142 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2147 #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
2149 /* Initiate the stall-info list. */
2150 static void print_cpu_stall_info_begin(void)
2152 printk(KERN_CONT "\n");
2156 * Print out diagnostic information for the specified stalled CPU.
2158 * If the specified CPU is aware of the current RCU grace period
2159 * (flavor specified by rsp), then print the number of scheduling
2160 * clock interrupts the CPU has taken during the time that it has
2161 * been aware. Otherwise, print the number of RCU grace periods
2162 * that this CPU is ignorant of, for example, "1" if the CPU was
2163 * aware of the previous grace period.
2165 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
2167 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2169 char fast_no_hz[72];
2170 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2171 struct rcu_dynticks *rdtp = rdp->dynticks;
2173 unsigned long ticks_value;
2175 if (rsp->gpnum == rdp->gpnum) {
2176 ticks_title = "ticks this GP";
2177 ticks_value = rdp->ticks_this_gp;
2179 ticks_title = "GPs behind";
2180 ticks_value = rsp->gpnum - rdp->gpnum;
2182 print_cpu_stall_fast_no_hz(fast_no_hz, cpu);
2183 printk(KERN_ERR "\t%d: (%lu %s) idle=%03x/%llx/%d %s\n",
2184 cpu, ticks_value, ticks_title,
2185 atomic_read(&rdtp->dynticks) & 0xfff,
2186 rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting,
2190 /* Terminate the stall-info list. */
2191 static void print_cpu_stall_info_end(void)
2193 printk(KERN_ERR "\t");
2196 /* Zero ->ticks_this_gp for all flavors of RCU. */
2197 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2199 rdp->ticks_this_gp = 0;
2202 /* Increment ->ticks_this_gp for all flavors of RCU. */
2203 static void increment_cpu_stall_ticks(void)
2205 struct rcu_state *rsp;
2207 for_each_rcu_flavor(rsp)
2208 __this_cpu_ptr(rsp->rda)->ticks_this_gp++;
2211 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
2213 static void print_cpu_stall_info_begin(void)
2215 printk(KERN_CONT " {");
2218 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2220 printk(KERN_CONT " %d", cpu);
2223 static void print_cpu_stall_info_end(void)
2225 printk(KERN_CONT "} ");
2228 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2232 static void increment_cpu_stall_ticks(void)
2236 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */