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 if (!rcu_preempt_blocked_readers_cgp(rnp))
427 raw_spin_lock_irqsave(&rnp->lock, flags);
428 t = list_entry(rnp->gp_tasks,
429 struct task_struct, rcu_node_entry);
430 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry)
432 raw_spin_unlock_irqrestore(&rnp->lock, flags);
436 * Dump detailed information for all tasks blocking the current RCU
439 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
441 struct rcu_node *rnp = rcu_get_root(rsp);
443 rcu_print_detail_task_stall_rnp(rnp);
444 rcu_for_each_leaf_node(rsp, rnp)
445 rcu_print_detail_task_stall_rnp(rnp);
448 #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
450 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
454 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
456 #ifdef CONFIG_RCU_CPU_STALL_INFO
458 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
460 printk(KERN_ERR "\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
461 rnp->level, rnp->grplo, rnp->grphi);
464 static void rcu_print_task_stall_end(void)
466 printk(KERN_CONT "\n");
469 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
471 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
475 static void rcu_print_task_stall_end(void)
479 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */
482 * Scan the current list of tasks blocked within RCU read-side critical
483 * sections, printing out the tid of each.
485 static int rcu_print_task_stall(struct rcu_node *rnp)
487 struct task_struct *t;
490 if (!rcu_preempt_blocked_readers_cgp(rnp))
492 rcu_print_task_stall_begin(rnp);
493 t = list_entry(rnp->gp_tasks,
494 struct task_struct, rcu_node_entry);
495 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
496 printk(KERN_CONT " P%d", t->pid);
499 rcu_print_task_stall_end();
504 * Check that the list of blocked tasks for the newly completed grace
505 * period is in fact empty. It is a serious bug to complete a grace
506 * period that still has RCU readers blocked! This function must be
507 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
508 * must be held by the caller.
510 * Also, if there are blocked tasks on the list, they automatically
511 * block the newly created grace period, so set up ->gp_tasks accordingly.
513 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
515 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
516 if (!list_empty(&rnp->blkd_tasks))
517 rnp->gp_tasks = rnp->blkd_tasks.next;
518 WARN_ON_ONCE(rnp->qsmask);
521 #ifdef CONFIG_HOTPLUG_CPU
524 * Handle tasklist migration for case in which all CPUs covered by the
525 * specified rcu_node have gone offline. Move them up to the root
526 * rcu_node. The reason for not just moving them to the immediate
527 * parent is to remove the need for rcu_read_unlock_special() to
528 * make more than two attempts to acquire the target rcu_node's lock.
529 * Returns true if there were tasks blocking the current RCU grace
532 * Returns 1 if there was previously a task blocking the current grace
533 * period on the specified rcu_node structure.
535 * The caller must hold rnp->lock with irqs disabled.
537 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
538 struct rcu_node *rnp,
539 struct rcu_data *rdp)
541 struct list_head *lp;
542 struct list_head *lp_root;
544 struct rcu_node *rnp_root = rcu_get_root(rsp);
545 struct task_struct *t;
547 if (rnp == rnp_root) {
548 WARN_ONCE(1, "Last CPU thought to be offlined?");
549 return 0; /* Shouldn't happen: at least one CPU online. */
552 /* If we are on an internal node, complain bitterly. */
553 WARN_ON_ONCE(rnp != rdp->mynode);
556 * Move tasks up to root rcu_node. Don't try to get fancy for
557 * this corner-case operation -- just put this node's tasks
558 * at the head of the root node's list, and update the root node's
559 * ->gp_tasks and ->exp_tasks pointers to those of this node's,
560 * if non-NULL. This might result in waiting for more tasks than
561 * absolutely necessary, but this is a good performance/complexity
564 if (rcu_preempt_blocked_readers_cgp(rnp) && rnp->qsmask == 0)
565 retval |= RCU_OFL_TASKS_NORM_GP;
566 if (rcu_preempted_readers_exp(rnp))
567 retval |= RCU_OFL_TASKS_EXP_GP;
568 lp = &rnp->blkd_tasks;
569 lp_root = &rnp_root->blkd_tasks;
570 while (!list_empty(lp)) {
571 t = list_entry(lp->next, typeof(*t), rcu_node_entry);
572 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
573 list_del(&t->rcu_node_entry);
574 t->rcu_blocked_node = rnp_root;
575 list_add(&t->rcu_node_entry, lp_root);
576 if (&t->rcu_node_entry == rnp->gp_tasks)
577 rnp_root->gp_tasks = rnp->gp_tasks;
578 if (&t->rcu_node_entry == rnp->exp_tasks)
579 rnp_root->exp_tasks = rnp->exp_tasks;
580 #ifdef CONFIG_RCU_BOOST
581 if (&t->rcu_node_entry == rnp->boost_tasks)
582 rnp_root->boost_tasks = rnp->boost_tasks;
583 #endif /* #ifdef CONFIG_RCU_BOOST */
584 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
587 rnp->gp_tasks = NULL;
588 rnp->exp_tasks = NULL;
589 #ifdef CONFIG_RCU_BOOST
590 rnp->boost_tasks = NULL;
592 * In case root is being boosted and leaf was not. Make sure
593 * that we boost the tasks blocking the current grace period
596 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
597 if (rnp_root->boost_tasks != NULL &&
598 rnp_root->boost_tasks != rnp_root->gp_tasks &&
599 rnp_root->boost_tasks != rnp_root->exp_tasks)
600 rnp_root->boost_tasks = rnp_root->gp_tasks;
601 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
602 #endif /* #ifdef CONFIG_RCU_BOOST */
607 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
610 * Check for a quiescent state from the current CPU. When a task blocks,
611 * the task is recorded in the corresponding CPU's rcu_node structure,
612 * which is checked elsewhere.
614 * Caller must disable hard irqs.
616 static void rcu_preempt_check_callbacks(int cpu)
618 struct task_struct *t = current;
620 if (t->rcu_read_lock_nesting == 0) {
624 if (t->rcu_read_lock_nesting > 0 &&
625 per_cpu(rcu_preempt_data, cpu).qs_pending)
626 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
629 #ifdef CONFIG_RCU_BOOST
631 static void rcu_preempt_do_callbacks(void)
633 rcu_do_batch(&rcu_preempt_state, &__get_cpu_var(rcu_preempt_data));
636 #endif /* #ifdef CONFIG_RCU_BOOST */
639 * Queue a preemptible-RCU callback for invocation after a grace period.
641 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
643 __call_rcu(head, func, &rcu_preempt_state, 0);
645 EXPORT_SYMBOL_GPL(call_rcu);
648 * Queue an RCU callback for lazy invocation after a grace period.
649 * This will likely be later named something like "call_rcu_lazy()",
650 * but this change will require some way of tagging the lazy RCU
651 * callbacks in the list of pending callbacks. Until then, this
652 * function may only be called from __kfree_rcu().
654 void kfree_call_rcu(struct rcu_head *head,
655 void (*func)(struct rcu_head *rcu))
657 __call_rcu(head, func, &rcu_preempt_state, 1);
659 EXPORT_SYMBOL_GPL(kfree_call_rcu);
662 * synchronize_rcu - wait until a grace period has elapsed.
664 * Control will return to the caller some time after a full grace
665 * period has elapsed, in other words after all currently executing RCU
666 * read-side critical sections have completed. Note, however, that
667 * upon return from synchronize_rcu(), the caller might well be executing
668 * concurrently with new RCU read-side critical sections that began while
669 * synchronize_rcu() was waiting. RCU read-side critical sections are
670 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
672 void synchronize_rcu(void)
674 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
675 !lock_is_held(&rcu_lock_map) &&
676 !lock_is_held(&rcu_sched_lock_map),
677 "Illegal synchronize_rcu() in RCU read-side critical section");
678 if (!rcu_scheduler_active)
680 wait_rcu_gp(call_rcu);
682 EXPORT_SYMBOL_GPL(synchronize_rcu);
684 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
685 static long sync_rcu_preempt_exp_count;
686 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
689 * Return non-zero if there are any tasks in RCU read-side critical
690 * sections blocking the current preemptible-RCU expedited grace period.
691 * If there is no preemptible-RCU expedited grace period currently in
692 * progress, returns zero unconditionally.
694 static int rcu_preempted_readers_exp(struct rcu_node *rnp)
696 return rnp->exp_tasks != NULL;
700 * return non-zero if there is no RCU expedited grace period in progress
701 * for the specified rcu_node structure, in other words, if all CPUs and
702 * tasks covered by the specified rcu_node structure have done their bit
703 * for the current expedited grace period. Works only for preemptible
704 * RCU -- other RCU implementation use other means.
706 * Caller must hold sync_rcu_preempt_exp_mutex.
708 static int sync_rcu_preempt_exp_done(struct rcu_node *rnp)
710 return !rcu_preempted_readers_exp(rnp) &&
711 ACCESS_ONCE(rnp->expmask) == 0;
715 * Report the exit from RCU read-side critical section for the last task
716 * that queued itself during or before the current expedited preemptible-RCU
717 * grace period. This event is reported either to the rcu_node structure on
718 * which the task was queued or to one of that rcu_node structure's ancestors,
719 * recursively up the tree. (Calm down, calm down, we do the recursion
722 * Most callers will set the "wake" flag, but the task initiating the
723 * expedited grace period need not wake itself.
725 * Caller must hold sync_rcu_preempt_exp_mutex.
727 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
733 raw_spin_lock_irqsave(&rnp->lock, flags);
735 if (!sync_rcu_preempt_exp_done(rnp)) {
736 raw_spin_unlock_irqrestore(&rnp->lock, flags);
739 if (rnp->parent == NULL) {
740 raw_spin_unlock_irqrestore(&rnp->lock, flags);
742 wake_up(&sync_rcu_preempt_exp_wq);
746 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
748 raw_spin_lock(&rnp->lock); /* irqs already disabled */
749 rnp->expmask &= ~mask;
754 * Snapshot the tasks blocking the newly started preemptible-RCU expedited
755 * grace period for the specified rcu_node structure. If there are no such
756 * tasks, report it up the rcu_node hierarchy.
758 * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock.
761 sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp)
766 raw_spin_lock_irqsave(&rnp->lock, flags);
767 if (list_empty(&rnp->blkd_tasks)) {
768 raw_spin_unlock_irqrestore(&rnp->lock, flags);
770 rnp->exp_tasks = rnp->blkd_tasks.next;
771 rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
775 rcu_report_exp_rnp(rsp, rnp, false); /* Don't wake self. */
779 * synchronize_rcu_expedited - Brute-force RCU grace period
781 * Wait for an RCU-preempt grace period, but expedite it. The basic
782 * idea is to invoke synchronize_sched_expedited() to push all the tasks to
783 * the ->blkd_tasks lists and wait for this list to drain. This consumes
784 * significant time on all CPUs and is unfriendly to real-time workloads,
785 * so is thus not recommended for any sort of common-case code.
786 * In fact, if you are using synchronize_rcu_expedited() in a loop,
787 * please restructure your code to batch your updates, and then Use a
788 * single synchronize_rcu() instead.
790 * Note that it is illegal to call this function while holding any lock
791 * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal
792 * to call this function from a CPU-hotplug notifier. Failing to observe
793 * these restriction will result in deadlock.
795 void synchronize_rcu_expedited(void)
798 struct rcu_node *rnp;
799 struct rcu_state *rsp = &rcu_preempt_state;
803 smp_mb(); /* Caller's modifications seen first by other CPUs. */
804 snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1;
805 smp_mb(); /* Above access cannot bleed into critical section. */
808 * Acquire lock, falling back to synchronize_rcu() if too many
809 * lock-acquisition failures. Of course, if someone does the
810 * expedited grace period for us, just leave.
812 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) {
813 if (trycount++ < 10) {
814 udelay(trycount * num_online_cpus());
819 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
820 goto mb_ret; /* Others did our work for us. */
822 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
823 goto unlock_mb_ret; /* Others did our work for us. */
825 /* force all RCU readers onto ->blkd_tasks lists. */
826 synchronize_sched_expedited();
828 raw_spin_lock_irqsave(&rsp->onofflock, flags);
830 /* Initialize ->expmask for all non-leaf rcu_node structures. */
831 rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) {
832 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
833 rnp->expmask = rnp->qsmaskinit;
834 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
837 /* Snapshot current state of ->blkd_tasks lists. */
838 rcu_for_each_leaf_node(rsp, rnp)
839 sync_rcu_preempt_exp_init(rsp, rnp);
840 if (NUM_RCU_NODES > 1)
841 sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp));
843 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
845 /* Wait for snapshotted ->blkd_tasks lists to drain. */
846 rnp = rcu_get_root(rsp);
847 wait_event(sync_rcu_preempt_exp_wq,
848 sync_rcu_preempt_exp_done(rnp));
850 /* Clean up and exit. */
851 smp_mb(); /* ensure expedited GP seen before counter increment. */
852 ACCESS_ONCE(sync_rcu_preempt_exp_count)++;
854 mutex_unlock(&sync_rcu_preempt_exp_mutex);
856 smp_mb(); /* ensure subsequent action seen after grace period. */
858 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
861 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
863 void rcu_barrier(void)
865 _rcu_barrier(&rcu_preempt_state);
867 EXPORT_SYMBOL_GPL(rcu_barrier);
870 * Initialize preemptible RCU's state structures.
872 static void __init __rcu_init_preempt(void)
874 rcu_init_one(&rcu_preempt_state, &rcu_preempt_data);
877 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
879 static struct rcu_state *rcu_state = &rcu_sched_state;
882 * Tell them what RCU they are running.
884 static void __init rcu_bootup_announce(void)
886 printk(KERN_INFO "Hierarchical RCU implementation.\n");
887 rcu_bootup_announce_oddness();
891 * Return the number of RCU batches processed thus far for debug & stats.
893 long rcu_batches_completed(void)
895 return rcu_batches_completed_sched();
897 EXPORT_SYMBOL_GPL(rcu_batches_completed);
900 * Force a quiescent state for RCU, which, because there is no preemptible
901 * RCU, becomes the same as rcu-sched.
903 void rcu_force_quiescent_state(void)
905 rcu_sched_force_quiescent_state();
907 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
910 * Because preemptible RCU does not exist, we never have to check for
911 * CPUs being in quiescent states.
913 static void rcu_preempt_note_context_switch(int cpu)
918 * Because preemptible RCU does not exist, there are never any preempted
921 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
926 #ifdef CONFIG_HOTPLUG_CPU
928 /* Because preemptible RCU does not exist, no quieting of tasks. */
929 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
931 raw_spin_unlock_irqrestore(&rnp->lock, flags);
934 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
937 * Because preemptible RCU does not exist, we never have to check for
938 * tasks blocked within RCU read-side critical sections.
940 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
945 * Because preemptible RCU does not exist, we never have to check for
946 * tasks blocked within RCU read-side critical sections.
948 static int rcu_print_task_stall(struct rcu_node *rnp)
954 * Because there is no preemptible RCU, there can be no readers blocked,
955 * so there is no need to check for blocked tasks. So check only for
956 * bogus qsmask values.
958 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
960 WARN_ON_ONCE(rnp->qsmask);
963 #ifdef CONFIG_HOTPLUG_CPU
966 * Because preemptible RCU does not exist, it never needs to migrate
967 * tasks that were blocked within RCU read-side critical sections, and
968 * such non-existent tasks cannot possibly have been blocking the current
971 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
972 struct rcu_node *rnp,
973 struct rcu_data *rdp)
978 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
981 * Because preemptible RCU does not exist, it never has any callbacks
984 static void rcu_preempt_check_callbacks(int cpu)
989 * Queue an RCU callback for lazy invocation after a grace period.
990 * This will likely be later named something like "call_rcu_lazy()",
991 * but this change will require some way of tagging the lazy RCU
992 * callbacks in the list of pending callbacks. Until then, this
993 * function may only be called from __kfree_rcu().
995 * Because there is no preemptible RCU, we use RCU-sched instead.
997 void kfree_call_rcu(struct rcu_head *head,
998 void (*func)(struct rcu_head *rcu))
1000 __call_rcu(head, func, &rcu_sched_state, 1);
1002 EXPORT_SYMBOL_GPL(kfree_call_rcu);
1005 * Wait for an rcu-preempt grace period, but make it happen quickly.
1006 * But because preemptible RCU does not exist, map to rcu-sched.
1008 void synchronize_rcu_expedited(void)
1010 synchronize_sched_expedited();
1012 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
1014 #ifdef CONFIG_HOTPLUG_CPU
1017 * Because preemptible RCU does not exist, there is never any need to
1018 * report on tasks preempted in RCU read-side critical sections during
1019 * expedited RCU grace periods.
1021 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
1026 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1029 * Because preemptible RCU does not exist, rcu_barrier() is just
1030 * another name for rcu_barrier_sched().
1032 void rcu_barrier(void)
1034 rcu_barrier_sched();
1036 EXPORT_SYMBOL_GPL(rcu_barrier);
1039 * Because preemptible RCU does not exist, it need not be initialized.
1041 static void __init __rcu_init_preempt(void)
1045 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1047 #ifdef CONFIG_RCU_BOOST
1049 #include "rtmutex_common.h"
1051 #ifdef CONFIG_RCU_TRACE
1053 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1055 if (list_empty(&rnp->blkd_tasks))
1056 rnp->n_balk_blkd_tasks++;
1057 else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL)
1058 rnp->n_balk_exp_gp_tasks++;
1059 else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL)
1060 rnp->n_balk_boost_tasks++;
1061 else if (rnp->gp_tasks != NULL && rnp->qsmask != 0)
1062 rnp->n_balk_notblocked++;
1063 else if (rnp->gp_tasks != NULL &&
1064 ULONG_CMP_LT(jiffies, rnp->boost_time))
1065 rnp->n_balk_notyet++;
1070 #else /* #ifdef CONFIG_RCU_TRACE */
1072 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1076 #endif /* #else #ifdef CONFIG_RCU_TRACE */
1079 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1080 * or ->boost_tasks, advancing the pointer to the next task in the
1081 * ->blkd_tasks list.
1083 * Note that irqs must be enabled: boosting the task can block.
1084 * Returns 1 if there are more tasks needing to be boosted.
1086 static int rcu_boost(struct rcu_node *rnp)
1088 unsigned long flags;
1089 struct rt_mutex mtx;
1090 struct task_struct *t;
1091 struct list_head *tb;
1093 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL)
1094 return 0; /* Nothing left to boost. */
1096 raw_spin_lock_irqsave(&rnp->lock, flags);
1099 * Recheck under the lock: all tasks in need of boosting
1100 * might exit their RCU read-side critical sections on their own.
1102 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1103 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1108 * Preferentially boost tasks blocking expedited grace periods.
1109 * This cannot starve the normal grace periods because a second
1110 * expedited grace period must boost all blocked tasks, including
1111 * those blocking the pre-existing normal grace period.
1113 if (rnp->exp_tasks != NULL) {
1114 tb = rnp->exp_tasks;
1115 rnp->n_exp_boosts++;
1117 tb = rnp->boost_tasks;
1118 rnp->n_normal_boosts++;
1120 rnp->n_tasks_boosted++;
1123 * We boost task t by manufacturing an rt_mutex that appears to
1124 * be held by task t. We leave a pointer to that rt_mutex where
1125 * task t can find it, and task t will release the mutex when it
1126 * exits its outermost RCU read-side critical section. Then
1127 * simply acquiring this artificial rt_mutex will boost task
1128 * t's priority. (Thanks to tglx for suggesting this approach!)
1130 * Note that task t must acquire rnp->lock to remove itself from
1131 * the ->blkd_tasks list, which it will do from exit() if from
1132 * nowhere else. We therefore are guaranteed that task t will
1133 * stay around at least until we drop rnp->lock. Note that
1134 * rnp->lock also resolves races between our priority boosting
1135 * and task t's exiting its outermost RCU read-side critical
1138 t = container_of(tb, struct task_struct, rcu_node_entry);
1139 rt_mutex_init_proxy_locked(&mtx, t);
1140 t->rcu_boost_mutex = &mtx;
1141 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1142 rt_mutex_lock(&mtx); /* Side effect: boosts task t's priority. */
1143 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */
1145 return ACCESS_ONCE(rnp->exp_tasks) != NULL ||
1146 ACCESS_ONCE(rnp->boost_tasks) != NULL;
1150 * Timer handler to initiate waking up of boost kthreads that
1151 * have yielded the CPU due to excessive numbers of tasks to
1152 * boost. We wake up the per-rcu_node kthread, which in turn
1153 * will wake up the booster kthread.
1155 static void rcu_boost_kthread_timer(unsigned long arg)
1157 invoke_rcu_node_kthread((struct rcu_node *)arg);
1161 * Priority-boosting kthread. One per leaf rcu_node and one for the
1164 static int rcu_boost_kthread(void *arg)
1166 struct rcu_node *rnp = (struct rcu_node *)arg;
1170 trace_rcu_utilization("Start boost kthread@init");
1172 rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
1173 trace_rcu_utilization("End boost kthread@rcu_wait");
1174 rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
1175 trace_rcu_utilization("Start boost kthread@rcu_wait");
1176 rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
1177 more2boost = rcu_boost(rnp);
1183 trace_rcu_utilization("End boost kthread@rcu_yield");
1184 rcu_yield(rcu_boost_kthread_timer, (unsigned long)rnp);
1185 trace_rcu_utilization("Start boost kthread@rcu_yield");
1190 trace_rcu_utilization("End boost kthread@notreached");
1195 * Check to see if it is time to start boosting RCU readers that are
1196 * blocking the current grace period, and, if so, tell the per-rcu_node
1197 * kthread to start boosting them. If there is an expedited grace
1198 * period in progress, it is always time to boost.
1200 * The caller must hold rnp->lock, which this function releases,
1201 * but irqs remain disabled. The ->boost_kthread_task is immortal,
1202 * so we don't need to worry about it going away.
1204 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1206 struct task_struct *t;
1208 if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
1209 rnp->n_balk_exp_gp_tasks++;
1210 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1213 if (rnp->exp_tasks != NULL ||
1214 (rnp->gp_tasks != NULL &&
1215 rnp->boost_tasks == NULL &&
1217 ULONG_CMP_GE(jiffies, rnp->boost_time))) {
1218 if (rnp->exp_tasks == NULL)
1219 rnp->boost_tasks = rnp->gp_tasks;
1220 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1221 t = rnp->boost_kthread_task;
1225 rcu_initiate_boost_trace(rnp);
1226 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1231 * Wake up the per-CPU kthread to invoke RCU callbacks.
1233 static void invoke_rcu_callbacks_kthread(void)
1235 unsigned long flags;
1237 local_irq_save(flags);
1238 __this_cpu_write(rcu_cpu_has_work, 1);
1239 if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
1240 current != __this_cpu_read(rcu_cpu_kthread_task))
1241 wake_up_process(__this_cpu_read(rcu_cpu_kthread_task));
1242 local_irq_restore(flags);
1246 * Is the current CPU running the RCU-callbacks kthread?
1247 * Caller must have preemption disabled.
1249 static bool rcu_is_callbacks_kthread(void)
1251 return __get_cpu_var(rcu_cpu_kthread_task) == current;
1255 * Set the affinity of the boost kthread. The CPU-hotplug locks are
1256 * held, so no one should be messing with the existence of the boost
1259 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
1262 struct task_struct *t;
1264 t = rnp->boost_kthread_task;
1266 set_cpus_allowed_ptr(rnp->boost_kthread_task, cm);
1269 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1272 * Do priority-boost accounting for the start of a new grace period.
1274 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1276 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1280 * Create an RCU-boost kthread for the specified node if one does not
1281 * already exist. We only create this kthread for preemptible RCU.
1282 * Returns zero if all is well, a negated errno otherwise.
1284 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
1285 struct rcu_node *rnp,
1288 unsigned long flags;
1289 struct sched_param sp;
1290 struct task_struct *t;
1292 if (&rcu_preempt_state != rsp)
1295 if (rnp->boost_kthread_task != NULL)
1297 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1298 "rcub/%d", rnp_index);
1301 raw_spin_lock_irqsave(&rnp->lock, flags);
1302 rnp->boost_kthread_task = t;
1303 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1304 sp.sched_priority = RCU_BOOST_PRIO;
1305 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1306 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1310 #ifdef CONFIG_HOTPLUG_CPU
1313 * Stop the RCU's per-CPU kthread when its CPU goes offline,.
1315 static void rcu_stop_cpu_kthread(int cpu)
1317 struct task_struct *t;
1319 /* Stop the CPU's kthread. */
1320 t = per_cpu(rcu_cpu_kthread_task, cpu);
1322 per_cpu(rcu_cpu_kthread_task, cpu) = NULL;
1327 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1329 static void rcu_kthread_do_work(void)
1331 rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data));
1332 rcu_do_batch(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1333 rcu_preempt_do_callbacks();
1337 * Wake up the specified per-rcu_node-structure kthread.
1338 * Because the per-rcu_node kthreads are immortal, we don't need
1339 * to do anything to keep them alive.
1341 static void invoke_rcu_node_kthread(struct rcu_node *rnp)
1343 struct task_struct *t;
1345 t = rnp->node_kthread_task;
1351 * Set the specified CPU's kthread to run RT or not, as specified by
1352 * the to_rt argument. The CPU-hotplug locks are held, so the task
1353 * is not going away.
1355 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1358 struct sched_param sp;
1359 struct task_struct *t;
1361 t = per_cpu(rcu_cpu_kthread_task, cpu);
1365 policy = SCHED_FIFO;
1366 sp.sched_priority = RCU_KTHREAD_PRIO;
1368 policy = SCHED_NORMAL;
1369 sp.sched_priority = 0;
1371 sched_setscheduler_nocheck(t, policy, &sp);
1375 * Timer handler to initiate the waking up of per-CPU kthreads that
1376 * have yielded the CPU due to excess numbers of RCU callbacks.
1377 * We wake up the per-rcu_node kthread, which in turn will wake up
1378 * the booster kthread.
1380 static void rcu_cpu_kthread_timer(unsigned long arg)
1382 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg);
1383 struct rcu_node *rnp = rdp->mynode;
1385 atomic_or(rdp->grpmask, &rnp->wakemask);
1386 invoke_rcu_node_kthread(rnp);
1390 * Drop to non-real-time priority and yield, but only after posting a
1391 * timer that will cause us to regain our real-time priority if we
1392 * remain preempted. Either way, we restore our real-time priority
1395 static void rcu_yield(void (*f)(unsigned long), unsigned long arg)
1397 struct sched_param sp;
1398 struct timer_list yield_timer;
1399 int prio = current->rt_priority;
1401 setup_timer_on_stack(&yield_timer, f, arg);
1402 mod_timer(&yield_timer, jiffies + 2);
1403 sp.sched_priority = 0;
1404 sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
1405 set_user_nice(current, 19);
1407 set_user_nice(current, 0);
1408 sp.sched_priority = prio;
1409 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1410 del_timer(&yield_timer);
1414 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1415 * This can happen while the corresponding CPU is either coming online
1416 * or going offline. We cannot wait until the CPU is fully online
1417 * before starting the kthread, because the various notifier functions
1418 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1419 * the corresponding CPU is online.
1421 * Return 1 if the kthread needs to stop, 0 otherwise.
1423 * Caller must disable bh. This function can momentarily enable it.
1425 static int rcu_cpu_kthread_should_stop(int cpu)
1427 while (cpu_is_offline(cpu) ||
1428 !cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)) ||
1429 smp_processor_id() != cpu) {
1430 if (kthread_should_stop())
1432 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
1433 per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id();
1435 schedule_timeout_uninterruptible(1);
1436 if (!cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)))
1437 set_cpus_allowed_ptr(current, cpumask_of(cpu));
1440 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1445 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1446 * RCU softirq used in flavors and configurations of RCU that do not
1447 * support RCU priority boosting.
1449 static int rcu_cpu_kthread(void *arg)
1451 int cpu = (int)(long)arg;
1452 unsigned long flags;
1454 unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu);
1456 char *workp = &per_cpu(rcu_cpu_has_work, cpu);
1458 trace_rcu_utilization("Start CPU kthread@init");
1460 *statusp = RCU_KTHREAD_WAITING;
1461 trace_rcu_utilization("End CPU kthread@rcu_wait");
1462 rcu_wait(*workp != 0 || kthread_should_stop());
1463 trace_rcu_utilization("Start CPU kthread@rcu_wait");
1465 if (rcu_cpu_kthread_should_stop(cpu)) {
1469 *statusp = RCU_KTHREAD_RUNNING;
1470 per_cpu(rcu_cpu_kthread_loops, cpu)++;
1471 local_irq_save(flags);
1474 local_irq_restore(flags);
1476 rcu_kthread_do_work();
1483 *statusp = RCU_KTHREAD_YIELDING;
1484 trace_rcu_utilization("End CPU kthread@rcu_yield");
1485 rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu);
1486 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1490 *statusp = RCU_KTHREAD_STOPPED;
1491 trace_rcu_utilization("End CPU kthread@term");
1496 * Spawn a per-CPU kthread, setting up affinity and priority.
1497 * Because the CPU hotplug lock is held, no other CPU will be attempting
1498 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1499 * attempting to access it during boot, but the locking in kthread_bind()
1500 * will enforce sufficient ordering.
1502 * Please note that we cannot simply refuse to wake up the per-CPU
1503 * kthread because kthreads are created in TASK_UNINTERRUPTIBLE state,
1504 * which can result in softlockup complaints if the task ends up being
1505 * idle for more than a couple of minutes.
1507 * However, please note also that we cannot bind the per-CPU kthread to its
1508 * CPU until that CPU is fully online. We also cannot wait until the
1509 * CPU is fully online before we create its per-CPU kthread, as this would
1510 * deadlock the system when CPU notifiers tried waiting for grace
1511 * periods. So we bind the per-CPU kthread to its CPU only if the CPU
1512 * is online. If its CPU is not yet fully online, then the code in
1513 * rcu_cpu_kthread() will wait until it is fully online, and then do
1516 static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu)
1518 struct sched_param sp;
1519 struct task_struct *t;
1521 if (!rcu_scheduler_fully_active ||
1522 per_cpu(rcu_cpu_kthread_task, cpu) != NULL)
1524 t = kthread_create_on_node(rcu_cpu_kthread,
1530 if (cpu_online(cpu))
1531 kthread_bind(t, cpu);
1532 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1533 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL);
1534 sp.sched_priority = RCU_KTHREAD_PRIO;
1535 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1536 per_cpu(rcu_cpu_kthread_task, cpu) = t;
1537 wake_up_process(t); /* Get to TASK_INTERRUPTIBLE quickly. */
1542 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1543 * kthreads when needed. We ignore requests to wake up kthreads
1544 * for offline CPUs, which is OK because force_quiescent_state()
1545 * takes care of this case.
1547 static int rcu_node_kthread(void *arg)
1550 unsigned long flags;
1552 struct rcu_node *rnp = (struct rcu_node *)arg;
1553 struct sched_param sp;
1554 struct task_struct *t;
1557 rnp->node_kthread_status = RCU_KTHREAD_WAITING;
1558 rcu_wait(atomic_read(&rnp->wakemask) != 0);
1559 rnp->node_kthread_status = RCU_KTHREAD_RUNNING;
1560 raw_spin_lock_irqsave(&rnp->lock, flags);
1561 mask = atomic_xchg(&rnp->wakemask, 0);
1562 rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
1563 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) {
1564 if ((mask & 0x1) == 0)
1567 t = per_cpu(rcu_cpu_kthread_task, cpu);
1568 if (!cpu_online(cpu) || t == NULL) {
1572 per_cpu(rcu_cpu_has_work, cpu) = 1;
1573 sp.sched_priority = RCU_KTHREAD_PRIO;
1574 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1579 rnp->node_kthread_status = RCU_KTHREAD_STOPPED;
1584 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1585 * served by the rcu_node in question. The CPU hotplug lock is still
1586 * held, so the value of rnp->qsmaskinit will be stable.
1588 * We don't include outgoingcpu in the affinity set, use -1 if there is
1589 * no outgoing CPU. If there are no CPUs left in the affinity set,
1590 * this function allows the kthread to execute on any CPU.
1592 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1596 unsigned long mask = rnp->qsmaskinit;
1598 if (rnp->node_kthread_task == NULL)
1600 if (!alloc_cpumask_var(&cm, GFP_KERNEL))
1603 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1)
1604 if ((mask & 0x1) && cpu != outgoingcpu)
1605 cpumask_set_cpu(cpu, cm);
1606 if (cpumask_weight(cm) == 0) {
1608 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++)
1609 cpumask_clear_cpu(cpu, cm);
1610 WARN_ON_ONCE(cpumask_weight(cm) == 0);
1612 set_cpus_allowed_ptr(rnp->node_kthread_task, cm);
1613 rcu_boost_kthread_setaffinity(rnp, cm);
1614 free_cpumask_var(cm);
1618 * Spawn a per-rcu_node kthread, setting priority and affinity.
1619 * Called during boot before online/offline can happen, or, if
1620 * during runtime, with the main CPU-hotplug locks held. So only
1621 * one of these can be executing at a time.
1623 static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp,
1624 struct rcu_node *rnp)
1626 unsigned long flags;
1627 int rnp_index = rnp - &rsp->node[0];
1628 struct sched_param sp;
1629 struct task_struct *t;
1631 if (!rcu_scheduler_fully_active ||
1632 rnp->qsmaskinit == 0)
1634 if (rnp->node_kthread_task == NULL) {
1635 t = kthread_create(rcu_node_kthread, (void *)rnp,
1636 "rcun/%d", rnp_index);
1639 raw_spin_lock_irqsave(&rnp->lock, flags);
1640 rnp->node_kthread_task = t;
1641 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1642 sp.sched_priority = 99;
1643 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1644 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1646 return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index);
1650 * Spawn all kthreads -- called as soon as the scheduler is running.
1652 static int __init rcu_spawn_kthreads(void)
1655 struct rcu_node *rnp;
1657 rcu_scheduler_fully_active = 1;
1658 for_each_possible_cpu(cpu) {
1659 per_cpu(rcu_cpu_has_work, cpu) = 0;
1660 if (cpu_online(cpu))
1661 (void)rcu_spawn_one_cpu_kthread(cpu);
1663 rnp = rcu_get_root(rcu_state);
1664 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1665 if (NUM_RCU_NODES > 1) {
1666 rcu_for_each_leaf_node(rcu_state, rnp)
1667 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1671 early_initcall(rcu_spawn_kthreads);
1673 static void __cpuinit rcu_prepare_kthreads(int cpu)
1675 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
1676 struct rcu_node *rnp = rdp->mynode;
1678 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1679 if (rcu_scheduler_fully_active) {
1680 (void)rcu_spawn_one_cpu_kthread(cpu);
1681 if (rnp->node_kthread_task == NULL)
1682 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1686 #else /* #ifdef CONFIG_RCU_BOOST */
1688 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1690 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1693 static void invoke_rcu_callbacks_kthread(void)
1698 static bool rcu_is_callbacks_kthread(void)
1703 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1707 #ifdef CONFIG_HOTPLUG_CPU
1709 static void rcu_stop_cpu_kthread(int cpu)
1713 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1715 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1719 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1723 static int __init rcu_scheduler_really_started(void)
1725 rcu_scheduler_fully_active = 1;
1728 early_initcall(rcu_scheduler_really_started);
1730 static void __cpuinit rcu_prepare_kthreads(int cpu)
1734 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1736 #if !defined(CONFIG_RCU_FAST_NO_HZ)
1739 * Check to see if any future RCU-related work will need to be done
1740 * by the current CPU, even if none need be done immediately, returning
1741 * 1 if so. This function is part of the RCU implementation; it is -not-
1742 * an exported member of the RCU API.
1744 * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
1745 * any flavor of RCU.
1747 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1749 *delta_jiffies = ULONG_MAX;
1750 return rcu_cpu_has_callbacks(cpu);
1754 * Because we do not have RCU_FAST_NO_HZ, don't bother initializing for it.
1756 static void rcu_prepare_for_idle_init(int cpu)
1761 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1764 static void rcu_cleanup_after_idle(int cpu)
1769 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
1772 static void rcu_prepare_for_idle(int cpu)
1777 * Don't bother keeping a running count of the number of RCU callbacks
1778 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1780 static void rcu_idle_count_callbacks_posted(void)
1784 #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1787 * This code is invoked when a CPU goes idle, at which point we want
1788 * to have the CPU do everything required for RCU so that it can enter
1789 * the energy-efficient dyntick-idle mode. This is handled by a
1790 * state machine implemented by rcu_prepare_for_idle() below.
1792 * The following three proprocessor symbols control this state machine:
1794 * RCU_IDLE_FLUSHES gives the maximum number of times that we will attempt
1795 * to satisfy RCU. Beyond this point, it is better to incur a periodic
1796 * scheduling-clock interrupt than to loop through the state machine
1798 * RCU_IDLE_OPT_FLUSHES gives the number of RCU_IDLE_FLUSHES that are
1799 * optional if RCU does not need anything immediately from this
1800 * CPU, even if this CPU still has RCU callbacks queued. The first
1801 * times through the state machine are mandatory: we need to give
1802 * the state machine a chance to communicate a quiescent state
1804 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1805 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1806 * is sized to be roughly one RCU grace period. Those energy-efficiency
1807 * benchmarkers who might otherwise be tempted to set this to a large
1808 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1809 * system. And if you are -that- concerned about energy efficiency,
1810 * just power the system down and be done with it!
1811 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1812 * permitted to sleep in dyntick-idle mode with only lazy RCU
1813 * callbacks pending. Setting this too high can OOM your system.
1815 * The values below work well in practice. If future workloads require
1816 * adjustment, they can be converted into kernel config parameters, though
1817 * making the state machine smarter might be a better option.
1819 #define RCU_IDLE_FLUSHES 5 /* Number of dyntick-idle tries. */
1820 #define RCU_IDLE_OPT_FLUSHES 3 /* Optional dyntick-idle tries. */
1821 #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */
1822 #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
1824 extern int tick_nohz_enabled;
1827 * Does the specified flavor of RCU have non-lazy callbacks pending on
1828 * the specified CPU? Both RCU flavor and CPU are specified by the
1829 * rcu_data structure.
1831 static bool __rcu_cpu_has_nonlazy_callbacks(struct rcu_data *rdp)
1833 return rdp->qlen != rdp->qlen_lazy;
1836 #ifdef CONFIG_TREE_PREEMPT_RCU
1839 * Are there non-lazy RCU-preempt callbacks? (There cannot be if there
1840 * is no RCU-preempt in the kernel.)
1842 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1844 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
1846 return __rcu_cpu_has_nonlazy_callbacks(rdp);
1849 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1851 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1856 #endif /* else #ifdef CONFIG_TREE_PREEMPT_RCU */
1859 * Does any flavor of RCU have non-lazy callbacks on the specified CPU?
1861 static bool rcu_cpu_has_nonlazy_callbacks(int cpu)
1863 return __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_sched_data, cpu)) ||
1864 __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_bh_data, cpu)) ||
1865 rcu_preempt_cpu_has_nonlazy_callbacks(cpu);
1869 * Allow the CPU to enter dyntick-idle mode if either: (1) There are no
1870 * callbacks on this CPU, (2) this CPU has not yet attempted to enter
1871 * dyntick-idle mode, or (3) this CPU is in the process of attempting to
1872 * enter dyntick-idle mode. Otherwise, if we have recently tried and failed
1873 * to enter dyntick-idle mode, we refuse to try to enter it. After all,
1874 * it is better to incur scheduling-clock interrupts than to spin
1875 * continuously for the same time duration!
1877 * The delta_jiffies argument is used to store the time when RCU is
1878 * going to need the CPU again if it still has callbacks. The reason
1879 * for this is that rcu_prepare_for_idle() might need to post a timer,
1880 * but if so, it will do so after tick_nohz_stop_sched_tick() has set
1881 * the wakeup time for this CPU. This means that RCU's timer can be
1882 * delayed until the wakeup time, which defeats the purpose of posting
1885 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1887 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1889 /* Flag a new idle sojourn to the idle-entry state machine. */
1890 rdtp->idle_first_pass = 1;
1891 /* If no callbacks, RCU doesn't need the CPU. */
1892 if (!rcu_cpu_has_callbacks(cpu)) {
1893 *delta_jiffies = ULONG_MAX;
1896 if (rdtp->dyntick_holdoff == jiffies) {
1897 /* RCU recently tried and failed, so don't try again. */
1901 /* Set up for the possibility that RCU will post a timer. */
1902 if (rcu_cpu_has_nonlazy_callbacks(cpu)) {
1903 *delta_jiffies = round_up(RCU_IDLE_GP_DELAY + jiffies,
1904 RCU_IDLE_GP_DELAY) - jiffies;
1906 *delta_jiffies = jiffies + RCU_IDLE_LAZY_GP_DELAY;
1907 *delta_jiffies = round_jiffies(*delta_jiffies) - jiffies;
1913 * Handler for smp_call_function_single(). The only point of this
1914 * handler is to wake the CPU up, so the handler does only tracing.
1916 void rcu_idle_demigrate(void *unused)
1918 trace_rcu_prep_idle("Demigrate");
1922 * Timer handler used to force CPU to start pushing its remaining RCU
1923 * callbacks in the case where it entered dyntick-idle mode with callbacks
1924 * pending. The hander doesn't really need to do anything because the
1925 * real work is done upon re-entry to idle, or by the next scheduling-clock
1926 * interrupt should idle not be re-entered.
1928 * One special case: the timer gets migrated without awakening the CPU
1929 * on which the timer was scheduled on. In this case, we must wake up
1930 * that CPU. We do so with smp_call_function_single().
1932 static void rcu_idle_gp_timer_func(unsigned long cpu_in)
1934 int cpu = (int)cpu_in;
1936 trace_rcu_prep_idle("Timer");
1937 if (cpu != smp_processor_id())
1938 smp_call_function_single(cpu, rcu_idle_demigrate, NULL, 0);
1940 WARN_ON_ONCE(1); /* Getting here can hang the system... */
1944 * Initialize the timer used to pull CPUs out of dyntick-idle mode.
1946 static void rcu_prepare_for_idle_init(int cpu)
1948 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1950 rdtp->dyntick_holdoff = jiffies - 1;
1951 setup_timer(&rdtp->idle_gp_timer, rcu_idle_gp_timer_func, cpu);
1952 rdtp->idle_gp_timer_expires = jiffies - 1;
1953 rdtp->idle_first_pass = 1;
1957 * Clean up for exit from idle. Because we are exiting from idle, there
1958 * is no longer any point to ->idle_gp_timer, so cancel it. This will
1959 * do nothing if this timer is not active, so just cancel it unconditionally.
1961 static void rcu_cleanup_after_idle(int cpu)
1963 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1965 del_timer(&rdtp->idle_gp_timer);
1966 trace_rcu_prep_idle("Cleanup after idle");
1967 rdtp->tick_nohz_enabled_snap = ACCESS_ONCE(tick_nohz_enabled);
1971 * Check to see if any RCU-related work can be done by the current CPU,
1972 * and if so, schedule a softirq to get it done. This function is part
1973 * of the RCU implementation; it is -not- an exported member of the RCU API.
1975 * The idea is for the current CPU to clear out all work required by the
1976 * RCU core for the current grace period, so that this CPU can be permitted
1977 * to enter dyntick-idle mode. In some cases, it will need to be awakened
1978 * at the end of the grace period by whatever CPU ends the grace period.
1979 * This allows CPUs to go dyntick-idle more quickly, and to reduce the
1980 * number of wakeups by a modest integer factor.
1982 * Because it is not legal to invoke rcu_process_callbacks() with irqs
1983 * disabled, we do one pass of force_quiescent_state(), then do a
1984 * invoke_rcu_core() to cause rcu_process_callbacks() to be invoked
1985 * later. The ->dyntick_drain field controls the sequencing.
1987 * The caller must have disabled interrupts.
1989 static void rcu_prepare_for_idle(int cpu)
1991 struct timer_list *tp;
1992 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1995 /* Handle nohz enablement switches conservatively. */
1996 tne = ACCESS_ONCE(tick_nohz_enabled);
1997 if (tne != rdtp->tick_nohz_enabled_snap) {
1998 if (rcu_cpu_has_callbacks(cpu))
1999 invoke_rcu_core(); /* force nohz to see update. */
2000 rdtp->tick_nohz_enabled_snap = tne;
2007 * If this is an idle re-entry, for example, due to use of
2008 * RCU_NONIDLE() or the new idle-loop tracing API within the idle
2009 * loop, then don't take any state-machine actions, unless the
2010 * momentary exit from idle queued additional non-lazy callbacks.
2011 * Instead, repost the ->idle_gp_timer if this CPU has callbacks
2014 if (!rdtp->idle_first_pass &&
2015 (rdtp->nonlazy_posted == rdtp->nonlazy_posted_snap)) {
2016 if (rcu_cpu_has_callbacks(cpu)) {
2017 tp = &rdtp->idle_gp_timer;
2018 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2022 rdtp->idle_first_pass = 0;
2023 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted - 1;
2026 * If there are no callbacks on this CPU, enter dyntick-idle mode.
2027 * Also reset state to avoid prejudicing later attempts.
2029 if (!rcu_cpu_has_callbacks(cpu)) {
2030 rdtp->dyntick_holdoff = jiffies - 1;
2031 rdtp->dyntick_drain = 0;
2032 trace_rcu_prep_idle("No callbacks");
2037 * If in holdoff mode, just return. We will presumably have
2038 * refrained from disabling the scheduling-clock tick.
2040 if (rdtp->dyntick_holdoff == jiffies) {
2041 trace_rcu_prep_idle("In holdoff");
2045 /* Check and update the ->dyntick_drain sequencing. */
2046 if (rdtp->dyntick_drain <= 0) {
2047 /* First time through, initialize the counter. */
2048 rdtp->dyntick_drain = RCU_IDLE_FLUSHES;
2049 } else if (rdtp->dyntick_drain <= RCU_IDLE_OPT_FLUSHES &&
2050 !rcu_pending(cpu) &&
2051 !local_softirq_pending()) {
2052 /* Can we go dyntick-idle despite still having callbacks? */
2053 rdtp->dyntick_drain = 0;
2054 rdtp->dyntick_holdoff = jiffies;
2055 if (rcu_cpu_has_nonlazy_callbacks(cpu)) {
2056 trace_rcu_prep_idle("Dyntick with callbacks");
2057 rdtp->idle_gp_timer_expires =
2058 round_up(jiffies + RCU_IDLE_GP_DELAY,
2061 rdtp->idle_gp_timer_expires =
2062 round_jiffies(jiffies + RCU_IDLE_LAZY_GP_DELAY);
2063 trace_rcu_prep_idle("Dyntick with lazy callbacks");
2065 tp = &rdtp->idle_gp_timer;
2066 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2067 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted;
2068 return; /* Nothing more to do immediately. */
2069 } else if (--(rdtp->dyntick_drain) <= 0) {
2070 /* We have hit the limit, so time to give up. */
2071 rdtp->dyntick_holdoff = jiffies;
2072 trace_rcu_prep_idle("Begin holdoff");
2073 invoke_rcu_core(); /* Force the CPU out of dyntick-idle. */
2078 * Do one step of pushing the remaining RCU callbacks through
2079 * the RCU core state machine.
2081 #ifdef CONFIG_TREE_PREEMPT_RCU
2082 if (per_cpu(rcu_preempt_data, cpu).nxtlist) {
2083 rcu_preempt_qs(cpu);
2084 force_quiescent_state(&rcu_preempt_state, 0);
2086 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2087 if (per_cpu(rcu_sched_data, cpu).nxtlist) {
2089 force_quiescent_state(&rcu_sched_state, 0);
2091 if (per_cpu(rcu_bh_data, cpu).nxtlist) {
2093 force_quiescent_state(&rcu_bh_state, 0);
2097 * If RCU callbacks are still pending, RCU still needs this CPU.
2098 * So try forcing the callbacks through the grace period.
2100 if (rcu_cpu_has_callbacks(cpu)) {
2101 trace_rcu_prep_idle("More callbacks");
2104 trace_rcu_prep_idle("Callbacks drained");
2109 * Keep a running count of the number of non-lazy callbacks posted
2110 * on this CPU. This running counter (which is never decremented) allows
2111 * rcu_prepare_for_idle() to detect when something out of the idle loop
2112 * posts a callback, even if an equal number of callbacks are invoked.
2113 * Of course, callbacks should only be posted from within a trace event
2114 * designed to be called from idle or from within RCU_NONIDLE().
2116 static void rcu_idle_count_callbacks_posted(void)
2118 __this_cpu_add(rcu_dynticks.nonlazy_posted, 1);
2121 #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
2123 #ifdef CONFIG_RCU_CPU_STALL_INFO
2125 #ifdef CONFIG_RCU_FAST_NO_HZ
2127 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2129 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2130 struct timer_list *tltp = &rdtp->idle_gp_timer;
2132 sprintf(cp, "drain=%d %c timer=%lu",
2133 rdtp->dyntick_drain,
2134 rdtp->dyntick_holdoff == jiffies ? 'H' : '.',
2135 timer_pending(tltp) ? tltp->expires - jiffies : -1);
2138 #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
2140 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2145 #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
2147 /* Initiate the stall-info list. */
2148 static void print_cpu_stall_info_begin(void)
2150 printk(KERN_CONT "\n");
2154 * Print out diagnostic information for the specified stalled CPU.
2156 * If the specified CPU is aware of the current RCU grace period
2157 * (flavor specified by rsp), then print the number of scheduling
2158 * clock interrupts the CPU has taken during the time that it has
2159 * been aware. Otherwise, print the number of RCU grace periods
2160 * that this CPU is ignorant of, for example, "1" if the CPU was
2161 * aware of the previous grace period.
2163 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
2165 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2167 char fast_no_hz[72];
2168 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2169 struct rcu_dynticks *rdtp = rdp->dynticks;
2171 unsigned long ticks_value;
2173 if (rsp->gpnum == rdp->gpnum) {
2174 ticks_title = "ticks this GP";
2175 ticks_value = rdp->ticks_this_gp;
2177 ticks_title = "GPs behind";
2178 ticks_value = rsp->gpnum - rdp->gpnum;
2180 print_cpu_stall_fast_no_hz(fast_no_hz, cpu);
2181 printk(KERN_ERR "\t%d: (%lu %s) idle=%03x/%llx/%d %s\n",
2182 cpu, ticks_value, ticks_title,
2183 atomic_read(&rdtp->dynticks) & 0xfff,
2184 rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting,
2188 /* Terminate the stall-info list. */
2189 static void print_cpu_stall_info_end(void)
2191 printk(KERN_ERR "\t");
2194 /* Zero ->ticks_this_gp for all flavors of RCU. */
2195 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2197 rdp->ticks_this_gp = 0;
2200 /* Increment ->ticks_this_gp for all flavors of RCU. */
2201 static void increment_cpu_stall_ticks(void)
2203 __get_cpu_var(rcu_sched_data).ticks_this_gp++;
2204 __get_cpu_var(rcu_bh_data).ticks_this_gp++;
2205 #ifdef CONFIG_TREE_PREEMPT_RCU
2206 __get_cpu_var(rcu_preempt_data).ticks_this_gp++;
2207 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2210 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
2212 static void print_cpu_stall_info_begin(void)
2214 printk(KERN_CONT " {");
2217 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2219 printk(KERN_CONT " %d", cpu);
2222 static void print_cpu_stall_info_end(void)
2224 printk(KERN_CONT "} ");
2227 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2231 static void increment_cpu_stall_ticks(void)
2235 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */