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 "\tExperimental four-level hierarchy is enabled.\n");
75 #ifdef CONFIG_TREE_PREEMPT_RCU
77 struct rcu_state rcu_preempt_state = RCU_STATE_INITIALIZER(rcu_preempt);
78 DEFINE_PER_CPU(struct rcu_data, rcu_preempt_data);
79 static struct rcu_state *rcu_state = &rcu_preempt_state;
81 static int rcu_preempted_readers_exp(struct rcu_node *rnp);
84 * Tell them what RCU they are running.
86 static void __init rcu_bootup_announce(void)
88 printk(KERN_INFO "Preemptible hierarchical RCU implementation.\n");
89 rcu_bootup_announce_oddness();
93 * Return the number of RCU-preempt batches processed thus far
94 * for debug and statistics.
96 long rcu_batches_completed_preempt(void)
98 return rcu_preempt_state.completed;
100 EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt);
103 * Return the number of RCU batches processed thus far for debug & stats.
105 long rcu_batches_completed(void)
107 return rcu_batches_completed_preempt();
109 EXPORT_SYMBOL_GPL(rcu_batches_completed);
112 * Force a quiescent state for preemptible RCU.
114 void rcu_force_quiescent_state(void)
116 force_quiescent_state(&rcu_preempt_state, 0);
118 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
121 * Record a preemptible-RCU quiescent state for the specified CPU. Note
122 * that this just means that the task currently running on the CPU is
123 * not in a quiescent state. There might be any number of tasks blocked
124 * while in an RCU read-side critical section.
126 * Unlike the other rcu_*_qs() functions, callers to this function
127 * must disable irqs in order to protect the assignment to
128 * ->rcu_read_unlock_special.
130 static void rcu_preempt_qs(int cpu)
132 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
134 rdp->passed_quiesce_gpnum = rdp->gpnum;
136 if (rdp->passed_quiesce == 0)
137 trace_rcu_grace_period("rcu_preempt", rdp->gpnum, "cpuqs");
138 rdp->passed_quiesce = 1;
139 current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
143 * We have entered the scheduler, and the current task might soon be
144 * context-switched away from. If this task is in an RCU read-side
145 * critical section, we will no longer be able to rely on the CPU to
146 * record that fact, so we enqueue the task on the blkd_tasks list.
147 * The task will dequeue itself when it exits the outermost enclosing
148 * RCU read-side critical section. Therefore, the current grace period
149 * cannot be permitted to complete until the blkd_tasks list entries
150 * predating the current grace period drain, in other words, until
151 * rnp->gp_tasks becomes NULL.
153 * Caller must disable preemption.
155 static void rcu_preempt_note_context_switch(int cpu)
157 struct task_struct *t = current;
159 struct rcu_data *rdp;
160 struct rcu_node *rnp;
162 if (t->rcu_read_lock_nesting > 0 &&
163 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
165 /* Possibly blocking in an RCU read-side critical section. */
166 rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu);
168 raw_spin_lock_irqsave(&rnp->lock, flags);
169 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
170 t->rcu_blocked_node = rnp;
173 * If this CPU has already checked in, then this task
174 * will hold up the next grace period rather than the
175 * current grace period. Queue the task accordingly.
176 * If the task is queued for the current grace period
177 * (i.e., this CPU has not yet passed through a quiescent
178 * state for the current grace period), then as long
179 * as that task remains queued, the current grace period
180 * cannot end. Note that there is some uncertainty as
181 * to exactly when the current grace period started.
182 * We take a conservative approach, which can result
183 * in unnecessarily waiting on tasks that started very
184 * slightly after the current grace period began. C'est
187 * But first, note that the current CPU must still be
190 WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0);
191 WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
192 if ((rnp->qsmask & rdp->grpmask) && rnp->gp_tasks != NULL) {
193 list_add(&t->rcu_node_entry, rnp->gp_tasks->prev);
194 rnp->gp_tasks = &t->rcu_node_entry;
195 #ifdef CONFIG_RCU_BOOST
196 if (rnp->boost_tasks != NULL)
197 rnp->boost_tasks = rnp->gp_tasks;
198 #endif /* #ifdef CONFIG_RCU_BOOST */
200 list_add(&t->rcu_node_entry, &rnp->blkd_tasks);
201 if (rnp->qsmask & rdp->grpmask)
202 rnp->gp_tasks = &t->rcu_node_entry;
204 trace_rcu_preempt_task(rdp->rsp->name,
206 (rnp->qsmask & rdp->grpmask)
209 raw_spin_unlock_irqrestore(&rnp->lock, flags);
210 } else if (t->rcu_read_lock_nesting < 0 &&
211 t->rcu_read_unlock_special) {
214 * Complete exit from RCU read-side critical section on
215 * behalf of preempted instance of __rcu_read_unlock().
217 rcu_read_unlock_special(t);
221 * Either we were not in an RCU read-side critical section to
222 * begin with, or we have now recorded that critical section
223 * globally. Either way, we can now note a quiescent state
224 * for this CPU. Again, if we were in an RCU read-side critical
225 * section, and if that critical section was blocking the current
226 * grace period, then the fact that the task has been enqueued
227 * means that we continue to block the current grace period.
229 local_irq_save(flags);
231 local_irq_restore(flags);
235 * Check for preempted RCU readers blocking the current grace period
236 * for the specified rcu_node structure. If the caller needs a reliable
237 * answer, it must hold the rcu_node's ->lock.
239 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
241 return rnp->gp_tasks != NULL;
245 * Record a quiescent state for all tasks that were previously queued
246 * on the specified rcu_node structure and that were blocking the current
247 * RCU grace period. The caller must hold the specified rnp->lock with
248 * irqs disabled, and this lock is released upon return, but irqs remain
251 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
252 __releases(rnp->lock)
255 struct rcu_node *rnp_p;
257 if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
258 raw_spin_unlock_irqrestore(&rnp->lock, flags);
259 return; /* Still need more quiescent states! */
265 * Either there is only one rcu_node in the tree,
266 * or tasks were kicked up to root rcu_node due to
267 * CPUs going offline.
269 rcu_report_qs_rsp(&rcu_preempt_state, flags);
273 /* Report up the rest of the hierarchy. */
275 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
276 raw_spin_lock(&rnp_p->lock); /* irqs already disabled. */
277 rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags);
281 * Advance a ->blkd_tasks-list pointer to the next entry, instead
282 * returning NULL if at the end of the list.
284 static struct list_head *rcu_next_node_entry(struct task_struct *t,
285 struct rcu_node *rnp)
287 struct list_head *np;
289 np = t->rcu_node_entry.next;
290 if (np == &rnp->blkd_tasks)
296 * Handle special cases during rcu_read_unlock(), such as needing to
297 * notify RCU core processing or task having blocked during the RCU
298 * read-side critical section.
300 void rcu_read_unlock_special(struct task_struct *t)
306 struct list_head *np;
307 #ifdef CONFIG_RCU_BOOST
308 struct rt_mutex *rbmp = NULL;
309 #endif /* #ifdef CONFIG_RCU_BOOST */
310 struct rcu_node *rnp;
313 /* NMI handlers cannot block and cannot safely manipulate state. */
317 local_irq_save(flags);
320 * If RCU core is waiting for this CPU to exit critical section,
321 * let it know that we have done so.
323 special = t->rcu_read_unlock_special;
324 if (special & RCU_READ_UNLOCK_NEED_QS) {
325 rcu_preempt_qs(smp_processor_id());
328 /* Hardware IRQ handlers cannot block. */
329 if (in_irq() || in_serving_softirq()) {
330 local_irq_restore(flags);
334 /* Clean up if blocked during RCU read-side critical section. */
335 if (special & RCU_READ_UNLOCK_BLOCKED) {
336 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
339 * Remove this task from the list it blocked on. The
340 * task can migrate while we acquire the lock, but at
341 * most one time. So at most two passes through loop.
344 rnp = t->rcu_blocked_node;
345 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
346 if (rnp == t->rcu_blocked_node)
348 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
350 empty = !rcu_preempt_blocked_readers_cgp(rnp);
351 empty_exp = !rcu_preempted_readers_exp(rnp);
352 smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
353 np = rcu_next_node_entry(t, rnp);
354 list_del_init(&t->rcu_node_entry);
355 t->rcu_blocked_node = NULL;
356 trace_rcu_unlock_preempted_task("rcu_preempt",
358 if (&t->rcu_node_entry == rnp->gp_tasks)
360 if (&t->rcu_node_entry == rnp->exp_tasks)
362 #ifdef CONFIG_RCU_BOOST
363 if (&t->rcu_node_entry == rnp->boost_tasks)
364 rnp->boost_tasks = np;
365 /* Snapshot/clear ->rcu_boost_mutex with rcu_node lock held. */
366 if (t->rcu_boost_mutex) {
367 rbmp = t->rcu_boost_mutex;
368 t->rcu_boost_mutex = NULL;
370 #endif /* #ifdef CONFIG_RCU_BOOST */
373 * If this was the last task on the current list, and if
374 * we aren't waiting on any CPUs, report the quiescent state.
375 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
376 * so we must take a snapshot of the expedited state.
378 empty_exp_now = !rcu_preempted_readers_exp(rnp);
379 if (!empty && !rcu_preempt_blocked_readers_cgp(rnp)) {
380 trace_rcu_quiescent_state_report("preempt_rcu",
387 rcu_report_unblock_qs_rnp(rnp, flags);
389 raw_spin_unlock_irqrestore(&rnp->lock, flags);
391 #ifdef CONFIG_RCU_BOOST
392 /* Unboost if we were boosted. */
394 rt_mutex_unlock(rbmp);
395 #endif /* #ifdef CONFIG_RCU_BOOST */
398 * If this was the last task on the expedited lists,
399 * then we need to report up the rcu_node hierarchy.
401 if (!empty_exp && empty_exp_now)
402 rcu_report_exp_rnp(&rcu_preempt_state, rnp, true);
404 local_irq_restore(flags);
408 #ifdef CONFIG_RCU_CPU_STALL_VERBOSE
411 * Dump detailed information for all tasks blocking the current RCU
412 * grace period on the specified rcu_node structure.
414 static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
417 struct task_struct *t;
419 if (!rcu_preempt_blocked_readers_cgp(rnp))
421 raw_spin_lock_irqsave(&rnp->lock, flags);
422 t = list_entry(rnp->gp_tasks,
423 struct task_struct, rcu_node_entry);
424 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry)
426 raw_spin_unlock_irqrestore(&rnp->lock, flags);
430 * Dump detailed information for all tasks blocking the current RCU
433 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
435 struct rcu_node *rnp = rcu_get_root(rsp);
437 rcu_print_detail_task_stall_rnp(rnp);
438 rcu_for_each_leaf_node(rsp, rnp)
439 rcu_print_detail_task_stall_rnp(rnp);
442 #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
444 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
448 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
450 #ifdef CONFIG_RCU_CPU_STALL_INFO
452 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
454 printk(KERN_ERR "\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
455 rnp->level, rnp->grplo, rnp->grphi);
458 static void rcu_print_task_stall_end(void)
460 printk(KERN_CONT "\n");
463 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
465 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
469 static void rcu_print_task_stall_end(void)
473 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */
476 * Scan the current list of tasks blocked within RCU read-side critical
477 * sections, printing out the tid of each.
479 static int rcu_print_task_stall(struct rcu_node *rnp)
481 struct task_struct *t;
484 if (!rcu_preempt_blocked_readers_cgp(rnp))
486 rcu_print_task_stall_begin(rnp);
487 t = list_entry(rnp->gp_tasks,
488 struct task_struct, rcu_node_entry);
489 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
490 printk(KERN_CONT " P%d", t->pid);
493 rcu_print_task_stall_end();
498 * Suppress preemptible RCU's CPU stall warnings by pushing the
499 * time of the next stall-warning message comfortably far into the
502 static void rcu_preempt_stall_reset(void)
504 rcu_preempt_state.jiffies_stall = jiffies + ULONG_MAX / 2;
508 * Check that the list of blocked tasks for the newly completed grace
509 * period is in fact empty. It is a serious bug to complete a grace
510 * period that still has RCU readers blocked! This function must be
511 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
512 * must be held by the caller.
514 * Also, if there are blocked tasks on the list, they automatically
515 * block the newly created grace period, so set up ->gp_tasks accordingly.
517 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
519 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
520 if (!list_empty(&rnp->blkd_tasks))
521 rnp->gp_tasks = rnp->blkd_tasks.next;
522 WARN_ON_ONCE(rnp->qsmask);
525 #ifdef CONFIG_HOTPLUG_CPU
528 * Handle tasklist migration for case in which all CPUs covered by the
529 * specified rcu_node have gone offline. Move them up to the root
530 * rcu_node. The reason for not just moving them to the immediate
531 * parent is to remove the need for rcu_read_unlock_special() to
532 * make more than two attempts to acquire the target rcu_node's lock.
533 * Returns true if there were tasks blocking the current RCU grace
536 * Returns 1 if there was previously a task blocking the current grace
537 * period on the specified rcu_node structure.
539 * The caller must hold rnp->lock with irqs disabled.
541 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
542 struct rcu_node *rnp,
543 struct rcu_data *rdp)
545 struct list_head *lp;
546 struct list_head *lp_root;
548 struct rcu_node *rnp_root = rcu_get_root(rsp);
549 struct task_struct *t;
551 if (rnp == rnp_root) {
552 WARN_ONCE(1, "Last CPU thought to be offlined?");
553 return 0; /* Shouldn't happen: at least one CPU online. */
556 /* If we are on an internal node, complain bitterly. */
557 WARN_ON_ONCE(rnp != rdp->mynode);
560 * Move tasks up to root rcu_node. Don't try to get fancy for
561 * this corner-case operation -- just put this node's tasks
562 * at the head of the root node's list, and update the root node's
563 * ->gp_tasks and ->exp_tasks pointers to those of this node's,
564 * if non-NULL. This might result in waiting for more tasks than
565 * absolutely necessary, but this is a good performance/complexity
568 if (rcu_preempt_blocked_readers_cgp(rnp) && rnp->qsmask == 0)
569 retval |= RCU_OFL_TASKS_NORM_GP;
570 if (rcu_preempted_readers_exp(rnp))
571 retval |= RCU_OFL_TASKS_EXP_GP;
572 lp = &rnp->blkd_tasks;
573 lp_root = &rnp_root->blkd_tasks;
574 while (!list_empty(lp)) {
575 t = list_entry(lp->next, typeof(*t), rcu_node_entry);
576 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
577 list_del(&t->rcu_node_entry);
578 t->rcu_blocked_node = rnp_root;
579 list_add(&t->rcu_node_entry, lp_root);
580 if (&t->rcu_node_entry == rnp->gp_tasks)
581 rnp_root->gp_tasks = rnp->gp_tasks;
582 if (&t->rcu_node_entry == rnp->exp_tasks)
583 rnp_root->exp_tasks = rnp->exp_tasks;
584 #ifdef CONFIG_RCU_BOOST
585 if (&t->rcu_node_entry == rnp->boost_tasks)
586 rnp_root->boost_tasks = rnp->boost_tasks;
587 #endif /* #ifdef CONFIG_RCU_BOOST */
588 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
591 #ifdef CONFIG_RCU_BOOST
592 /* In case root is being boosted and leaf is not. */
593 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
594 if (rnp_root->boost_tasks != NULL &&
595 rnp_root->boost_tasks != rnp_root->gp_tasks)
596 rnp_root->boost_tasks = rnp_root->gp_tasks;
597 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
598 #endif /* #ifdef CONFIG_RCU_BOOST */
600 rnp->gp_tasks = NULL;
601 rnp->exp_tasks = NULL;
605 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
608 * Do CPU-offline processing for preemptible RCU.
610 static void rcu_preempt_cleanup_dead_cpu(int cpu)
612 rcu_cleanup_dead_cpu(cpu, &rcu_preempt_state);
616 * Check for a quiescent state from the current CPU. When a task blocks,
617 * the task is recorded in the corresponding CPU's rcu_node structure,
618 * which is checked elsewhere.
620 * Caller must disable hard irqs.
622 static void rcu_preempt_check_callbacks(int cpu)
624 struct task_struct *t = current;
626 if (t->rcu_read_lock_nesting == 0) {
630 if (t->rcu_read_lock_nesting > 0 &&
631 per_cpu(rcu_preempt_data, cpu).qs_pending)
632 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
636 * Process callbacks for preemptible RCU.
638 static void rcu_preempt_process_callbacks(void)
640 __rcu_process_callbacks(&rcu_preempt_state,
641 &__get_cpu_var(rcu_preempt_data));
644 #ifdef CONFIG_RCU_BOOST
646 static void rcu_preempt_do_callbacks(void)
648 rcu_do_batch(&rcu_preempt_state, &__get_cpu_var(rcu_preempt_data));
651 #endif /* #ifdef CONFIG_RCU_BOOST */
654 * Queue a preemptible-RCU callback for invocation after a grace period.
656 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
658 __call_rcu(head, func, &rcu_preempt_state, 0);
660 EXPORT_SYMBOL_GPL(call_rcu);
663 * Queue an RCU callback for lazy invocation after a grace period.
664 * This will likely be later named something like "call_rcu_lazy()",
665 * but this change will require some way of tagging the lazy RCU
666 * callbacks in the list of pending callbacks. Until then, this
667 * function may only be called from __kfree_rcu().
669 void kfree_call_rcu(struct rcu_head *head,
670 void (*func)(struct rcu_head *rcu))
672 __call_rcu(head, func, &rcu_preempt_state, 1);
674 EXPORT_SYMBOL_GPL(kfree_call_rcu);
677 * synchronize_rcu - wait until a grace period has elapsed.
679 * Control will return to the caller some time after a full grace
680 * period has elapsed, in other words after all currently executing RCU
681 * read-side critical sections have completed. Note, however, that
682 * upon return from synchronize_rcu(), the caller might well be executing
683 * concurrently with new RCU read-side critical sections that began while
684 * synchronize_rcu() was waiting. RCU read-side critical sections are
685 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
687 void synchronize_rcu(void)
689 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
690 !lock_is_held(&rcu_lock_map) &&
691 !lock_is_held(&rcu_sched_lock_map),
692 "Illegal synchronize_rcu() in RCU read-side critical section");
693 if (!rcu_scheduler_active)
695 wait_rcu_gp(call_rcu);
697 EXPORT_SYMBOL_GPL(synchronize_rcu);
699 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
700 static long sync_rcu_preempt_exp_count;
701 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
704 * Return non-zero if there are any tasks in RCU read-side critical
705 * sections blocking the current preemptible-RCU expedited grace period.
706 * If there is no preemptible-RCU expedited grace period currently in
707 * progress, returns zero unconditionally.
709 static int rcu_preempted_readers_exp(struct rcu_node *rnp)
711 return rnp->exp_tasks != NULL;
715 * return non-zero if there is no RCU expedited grace period in progress
716 * for the specified rcu_node structure, in other words, if all CPUs and
717 * tasks covered by the specified rcu_node structure have done their bit
718 * for the current expedited grace period. Works only for preemptible
719 * RCU -- other RCU implementation use other means.
721 * Caller must hold sync_rcu_preempt_exp_mutex.
723 static int sync_rcu_preempt_exp_done(struct rcu_node *rnp)
725 return !rcu_preempted_readers_exp(rnp) &&
726 ACCESS_ONCE(rnp->expmask) == 0;
730 * Report the exit from RCU read-side critical section for the last task
731 * that queued itself during or before the current expedited preemptible-RCU
732 * grace period. This event is reported either to the rcu_node structure on
733 * which the task was queued or to one of that rcu_node structure's ancestors,
734 * recursively up the tree. (Calm down, calm down, we do the recursion
737 * Most callers will set the "wake" flag, but the task initiating the
738 * expedited grace period need not wake itself.
740 * Caller must hold sync_rcu_preempt_exp_mutex.
742 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
748 raw_spin_lock_irqsave(&rnp->lock, flags);
750 if (!sync_rcu_preempt_exp_done(rnp)) {
751 raw_spin_unlock_irqrestore(&rnp->lock, flags);
754 if (rnp->parent == NULL) {
755 raw_spin_unlock_irqrestore(&rnp->lock, flags);
757 wake_up(&sync_rcu_preempt_exp_wq);
761 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
763 raw_spin_lock(&rnp->lock); /* irqs already disabled */
764 rnp->expmask &= ~mask;
769 * Snapshot the tasks blocking the newly started preemptible-RCU expedited
770 * grace period for the specified rcu_node structure. If there are no such
771 * tasks, report it up the rcu_node hierarchy.
773 * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock.
776 sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp)
781 raw_spin_lock_irqsave(&rnp->lock, flags);
782 if (list_empty(&rnp->blkd_tasks))
783 raw_spin_unlock_irqrestore(&rnp->lock, flags);
785 rnp->exp_tasks = rnp->blkd_tasks.next;
786 rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
790 rcu_report_exp_rnp(rsp, rnp, false); /* Don't wake self. */
794 * synchronize_rcu_expedited - Brute-force RCU grace period
796 * Wait for an RCU-preempt grace period, but expedite it. The basic
797 * idea is to invoke synchronize_sched_expedited() to push all the tasks to
798 * the ->blkd_tasks lists and wait for this list to drain. This consumes
799 * significant time on all CPUs and is unfriendly to real-time workloads,
800 * so is thus not recommended for any sort of common-case code.
801 * In fact, if you are using synchronize_rcu_expedited() in a loop,
802 * please restructure your code to batch your updates, and then Use a
803 * single synchronize_rcu() instead.
805 * Note that it is illegal to call this function while holding any lock
806 * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal
807 * to call this function from a CPU-hotplug notifier. Failing to observe
808 * these restriction will result in deadlock.
810 void synchronize_rcu_expedited(void)
813 struct rcu_node *rnp;
814 struct rcu_state *rsp = &rcu_preempt_state;
818 smp_mb(); /* Caller's modifications seen first by other CPUs. */
819 snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1;
820 smp_mb(); /* Above access cannot bleed into critical section. */
823 * Acquire lock, falling back to synchronize_rcu() if too many
824 * lock-acquisition failures. Of course, if someone does the
825 * expedited grace period for us, just leave.
827 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) {
829 udelay(trycount * num_online_cpus());
834 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
835 goto mb_ret; /* Others did our work for us. */
837 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
838 goto unlock_mb_ret; /* Others did our work for us. */
840 /* force all RCU readers onto ->blkd_tasks lists. */
841 synchronize_sched_expedited();
843 raw_spin_lock_irqsave(&rsp->onofflock, flags);
845 /* Initialize ->expmask for all non-leaf rcu_node structures. */
846 rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) {
847 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
848 rnp->expmask = rnp->qsmaskinit;
849 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
852 /* Snapshot current state of ->blkd_tasks lists. */
853 rcu_for_each_leaf_node(rsp, rnp)
854 sync_rcu_preempt_exp_init(rsp, rnp);
855 if (NUM_RCU_NODES > 1)
856 sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp));
858 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
860 /* Wait for snapshotted ->blkd_tasks lists to drain. */
861 rnp = rcu_get_root(rsp);
862 wait_event(sync_rcu_preempt_exp_wq,
863 sync_rcu_preempt_exp_done(rnp));
865 /* Clean up and exit. */
866 smp_mb(); /* ensure expedited GP seen before counter increment. */
867 ACCESS_ONCE(sync_rcu_preempt_exp_count)++;
869 mutex_unlock(&sync_rcu_preempt_exp_mutex);
871 smp_mb(); /* ensure subsequent action seen after grace period. */
873 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
876 * Check to see if there is any immediate preemptible-RCU-related work
879 static int rcu_preempt_pending(int cpu)
881 return __rcu_pending(&rcu_preempt_state,
882 &per_cpu(rcu_preempt_data, cpu));
886 * Does preemptible RCU have callbacks on this CPU?
888 static int rcu_preempt_cpu_has_callbacks(int cpu)
890 return !!per_cpu(rcu_preempt_data, cpu).nxtlist;
894 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
896 void rcu_barrier(void)
898 _rcu_barrier(&rcu_preempt_state, call_rcu);
900 EXPORT_SYMBOL_GPL(rcu_barrier);
903 * Initialize preemptible RCU's per-CPU data.
905 static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
907 rcu_init_percpu_data(cpu, &rcu_preempt_state, 1);
911 * Move preemptible RCU's callbacks from dying CPU to other online CPU
912 * and record a quiescent state.
914 static void rcu_preempt_cleanup_dying_cpu(void)
916 rcu_cleanup_dying_cpu(&rcu_preempt_state);
920 * Initialize preemptible RCU's state structures.
922 static void __init __rcu_init_preempt(void)
924 rcu_init_one(&rcu_preempt_state, &rcu_preempt_data);
927 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
929 static struct rcu_state *rcu_state = &rcu_sched_state;
932 * Tell them what RCU they are running.
934 static void __init rcu_bootup_announce(void)
936 printk(KERN_INFO "Hierarchical RCU implementation.\n");
937 rcu_bootup_announce_oddness();
941 * Return the number of RCU batches processed thus far for debug & stats.
943 long rcu_batches_completed(void)
945 return rcu_batches_completed_sched();
947 EXPORT_SYMBOL_GPL(rcu_batches_completed);
950 * Force a quiescent state for RCU, which, because there is no preemptible
951 * RCU, becomes the same as rcu-sched.
953 void rcu_force_quiescent_state(void)
955 rcu_sched_force_quiescent_state();
957 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
960 * Because preemptible RCU does not exist, we never have to check for
961 * CPUs being in quiescent states.
963 static void rcu_preempt_note_context_switch(int cpu)
968 * Because preemptible RCU does not exist, there are never any preempted
971 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
976 #ifdef CONFIG_HOTPLUG_CPU
978 /* Because preemptible RCU does not exist, no quieting of tasks. */
979 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
981 raw_spin_unlock_irqrestore(&rnp->lock, flags);
984 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
987 * Because preemptible RCU does not exist, we never have to check for
988 * tasks blocked within RCU read-side critical sections.
990 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
995 * Because preemptible RCU does not exist, we never have to check for
996 * tasks blocked within RCU read-side critical sections.
998 static int rcu_print_task_stall(struct rcu_node *rnp)
1004 * Because preemptible RCU does not exist, there is no need to suppress
1005 * its CPU stall warnings.
1007 static void rcu_preempt_stall_reset(void)
1012 * Because there is no preemptible RCU, there can be no readers blocked,
1013 * so there is no need to check for blocked tasks. So check only for
1014 * bogus qsmask values.
1016 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
1018 WARN_ON_ONCE(rnp->qsmask);
1021 #ifdef CONFIG_HOTPLUG_CPU
1024 * Because preemptible RCU does not exist, it never needs to migrate
1025 * tasks that were blocked within RCU read-side critical sections, and
1026 * such non-existent tasks cannot possibly have been blocking the current
1029 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
1030 struct rcu_node *rnp,
1031 struct rcu_data *rdp)
1036 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1039 * Because preemptible RCU does not exist, it never needs CPU-offline
1042 static void rcu_preempt_cleanup_dead_cpu(int cpu)
1047 * Because preemptible RCU does not exist, it never has any callbacks
1050 static void rcu_preempt_check_callbacks(int cpu)
1055 * Because preemptible RCU does not exist, it never has any callbacks
1058 static void rcu_preempt_process_callbacks(void)
1063 * Queue an RCU callback for lazy invocation after a grace period.
1064 * This will likely be later named something like "call_rcu_lazy()",
1065 * but this change will require some way of tagging the lazy RCU
1066 * callbacks in the list of pending callbacks. Until then, this
1067 * function may only be called from __kfree_rcu().
1069 * Because there is no preemptible RCU, we use RCU-sched instead.
1071 void kfree_call_rcu(struct rcu_head *head,
1072 void (*func)(struct rcu_head *rcu))
1074 __call_rcu(head, func, &rcu_sched_state, 1);
1076 EXPORT_SYMBOL_GPL(kfree_call_rcu);
1079 * Wait for an rcu-preempt grace period, but make it happen quickly.
1080 * But because preemptible RCU does not exist, map to rcu-sched.
1082 void synchronize_rcu_expedited(void)
1084 synchronize_sched_expedited();
1086 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
1088 #ifdef CONFIG_HOTPLUG_CPU
1091 * Because preemptible RCU does not exist, there is never any need to
1092 * report on tasks preempted in RCU read-side critical sections during
1093 * expedited RCU grace periods.
1095 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
1100 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1103 * Because preemptible RCU does not exist, it never has any work to do.
1105 static int rcu_preempt_pending(int cpu)
1111 * Because preemptible RCU does not exist, it never has callbacks
1113 static int rcu_preempt_cpu_has_callbacks(int cpu)
1119 * Because preemptible RCU does not exist, rcu_barrier() is just
1120 * another name for rcu_barrier_sched().
1122 void rcu_barrier(void)
1124 rcu_barrier_sched();
1126 EXPORT_SYMBOL_GPL(rcu_barrier);
1129 * Because preemptible RCU does not exist, there is no per-CPU
1130 * data to initialize.
1132 static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
1137 * Because there is no preemptible RCU, there is no cleanup to do.
1139 static void rcu_preempt_cleanup_dying_cpu(void)
1144 * Because preemptible RCU does not exist, it need not be initialized.
1146 static void __init __rcu_init_preempt(void)
1150 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1152 #ifdef CONFIG_RCU_BOOST
1154 #include "rtmutex_common.h"
1156 #ifdef CONFIG_RCU_TRACE
1158 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1160 if (list_empty(&rnp->blkd_tasks))
1161 rnp->n_balk_blkd_tasks++;
1162 else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL)
1163 rnp->n_balk_exp_gp_tasks++;
1164 else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL)
1165 rnp->n_balk_boost_tasks++;
1166 else if (rnp->gp_tasks != NULL && rnp->qsmask != 0)
1167 rnp->n_balk_notblocked++;
1168 else if (rnp->gp_tasks != NULL &&
1169 ULONG_CMP_LT(jiffies, rnp->boost_time))
1170 rnp->n_balk_notyet++;
1175 #else /* #ifdef CONFIG_RCU_TRACE */
1177 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1181 #endif /* #else #ifdef CONFIG_RCU_TRACE */
1184 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1185 * or ->boost_tasks, advancing the pointer to the next task in the
1186 * ->blkd_tasks list.
1188 * Note that irqs must be enabled: boosting the task can block.
1189 * Returns 1 if there are more tasks needing to be boosted.
1191 static int rcu_boost(struct rcu_node *rnp)
1193 unsigned long flags;
1194 struct rt_mutex mtx;
1195 struct task_struct *t;
1196 struct list_head *tb;
1198 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL)
1199 return 0; /* Nothing left to boost. */
1201 raw_spin_lock_irqsave(&rnp->lock, flags);
1204 * Recheck under the lock: all tasks in need of boosting
1205 * might exit their RCU read-side critical sections on their own.
1207 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1208 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1213 * Preferentially boost tasks blocking expedited grace periods.
1214 * This cannot starve the normal grace periods because a second
1215 * expedited grace period must boost all blocked tasks, including
1216 * those blocking the pre-existing normal grace period.
1218 if (rnp->exp_tasks != NULL) {
1219 tb = rnp->exp_tasks;
1220 rnp->n_exp_boosts++;
1222 tb = rnp->boost_tasks;
1223 rnp->n_normal_boosts++;
1225 rnp->n_tasks_boosted++;
1228 * We boost task t by manufacturing an rt_mutex that appears to
1229 * be held by task t. We leave a pointer to that rt_mutex where
1230 * task t can find it, and task t will release the mutex when it
1231 * exits its outermost RCU read-side critical section. Then
1232 * simply acquiring this artificial rt_mutex will boost task
1233 * t's priority. (Thanks to tglx for suggesting this approach!)
1235 * Note that task t must acquire rnp->lock to remove itself from
1236 * the ->blkd_tasks list, which it will do from exit() if from
1237 * nowhere else. We therefore are guaranteed that task t will
1238 * stay around at least until we drop rnp->lock. Note that
1239 * rnp->lock also resolves races between our priority boosting
1240 * and task t's exiting its outermost RCU read-side critical
1243 t = container_of(tb, struct task_struct, rcu_node_entry);
1244 rt_mutex_init_proxy_locked(&mtx, t);
1245 t->rcu_boost_mutex = &mtx;
1246 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1247 rt_mutex_lock(&mtx); /* Side effect: boosts task t's priority. */
1248 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */
1250 return ACCESS_ONCE(rnp->exp_tasks) != NULL ||
1251 ACCESS_ONCE(rnp->boost_tasks) != NULL;
1255 * Timer handler to initiate waking up of boost kthreads that
1256 * have yielded the CPU due to excessive numbers of tasks to
1257 * boost. We wake up the per-rcu_node kthread, which in turn
1258 * will wake up the booster kthread.
1260 static void rcu_boost_kthread_timer(unsigned long arg)
1262 invoke_rcu_node_kthread((struct rcu_node *)arg);
1266 * Priority-boosting kthread. One per leaf rcu_node and one for the
1269 static int rcu_boost_kthread(void *arg)
1271 struct rcu_node *rnp = (struct rcu_node *)arg;
1275 trace_rcu_utilization("Start boost kthread@init");
1277 rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
1278 trace_rcu_utilization("End boost kthread@rcu_wait");
1279 rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
1280 trace_rcu_utilization("Start boost kthread@rcu_wait");
1281 rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
1282 more2boost = rcu_boost(rnp);
1288 trace_rcu_utilization("End boost kthread@rcu_yield");
1289 rcu_yield(rcu_boost_kthread_timer, (unsigned long)rnp);
1290 trace_rcu_utilization("Start boost kthread@rcu_yield");
1295 trace_rcu_utilization("End boost kthread@notreached");
1300 * Check to see if it is time to start boosting RCU readers that are
1301 * blocking the current grace period, and, if so, tell the per-rcu_node
1302 * kthread to start boosting them. If there is an expedited grace
1303 * period in progress, it is always time to boost.
1305 * The caller must hold rnp->lock, which this function releases,
1306 * but irqs remain disabled. The ->boost_kthread_task is immortal,
1307 * so we don't need to worry about it going away.
1309 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1311 struct task_struct *t;
1313 if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
1314 rnp->n_balk_exp_gp_tasks++;
1315 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1318 if (rnp->exp_tasks != NULL ||
1319 (rnp->gp_tasks != NULL &&
1320 rnp->boost_tasks == NULL &&
1322 ULONG_CMP_GE(jiffies, rnp->boost_time))) {
1323 if (rnp->exp_tasks == NULL)
1324 rnp->boost_tasks = rnp->gp_tasks;
1325 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1326 t = rnp->boost_kthread_task;
1330 rcu_initiate_boost_trace(rnp);
1331 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1336 * Wake up the per-CPU kthread to invoke RCU callbacks.
1338 static void invoke_rcu_callbacks_kthread(void)
1340 unsigned long flags;
1342 local_irq_save(flags);
1343 __this_cpu_write(rcu_cpu_has_work, 1);
1344 if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
1345 current != __this_cpu_read(rcu_cpu_kthread_task))
1346 wake_up_process(__this_cpu_read(rcu_cpu_kthread_task));
1347 local_irq_restore(flags);
1351 * Is the current CPU running the RCU-callbacks kthread?
1352 * Caller must have preemption disabled.
1354 static bool rcu_is_callbacks_kthread(void)
1356 return __get_cpu_var(rcu_cpu_kthread_task) == current;
1360 * Set the affinity of the boost kthread. The CPU-hotplug locks are
1361 * held, so no one should be messing with the existence of the boost
1364 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
1367 struct task_struct *t;
1369 t = rnp->boost_kthread_task;
1371 set_cpus_allowed_ptr(rnp->boost_kthread_task, cm);
1374 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1377 * Do priority-boost accounting for the start of a new grace period.
1379 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1381 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1385 * Create an RCU-boost kthread for the specified node if one does not
1386 * already exist. We only create this kthread for preemptible RCU.
1387 * Returns zero if all is well, a negated errno otherwise.
1389 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
1390 struct rcu_node *rnp,
1393 unsigned long flags;
1394 struct sched_param sp;
1395 struct task_struct *t;
1397 if (&rcu_preempt_state != rsp)
1400 if (rnp->boost_kthread_task != NULL)
1402 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1403 "rcub/%d", rnp_index);
1406 raw_spin_lock_irqsave(&rnp->lock, flags);
1407 rnp->boost_kthread_task = t;
1408 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1409 sp.sched_priority = RCU_BOOST_PRIO;
1410 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1411 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1415 #ifdef CONFIG_HOTPLUG_CPU
1418 * Stop the RCU's per-CPU kthread when its CPU goes offline,.
1420 static void rcu_stop_cpu_kthread(int cpu)
1422 struct task_struct *t;
1424 /* Stop the CPU's kthread. */
1425 t = per_cpu(rcu_cpu_kthread_task, cpu);
1427 per_cpu(rcu_cpu_kthread_task, cpu) = NULL;
1432 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1434 static void rcu_kthread_do_work(void)
1436 rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data));
1437 rcu_do_batch(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1438 rcu_preempt_do_callbacks();
1442 * Wake up the specified per-rcu_node-structure kthread.
1443 * Because the per-rcu_node kthreads are immortal, we don't need
1444 * to do anything to keep them alive.
1446 static void invoke_rcu_node_kthread(struct rcu_node *rnp)
1448 struct task_struct *t;
1450 t = rnp->node_kthread_task;
1456 * Set the specified CPU's kthread to run RT or not, as specified by
1457 * the to_rt argument. The CPU-hotplug locks are held, so the task
1458 * is not going away.
1460 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1463 struct sched_param sp;
1464 struct task_struct *t;
1466 t = per_cpu(rcu_cpu_kthread_task, cpu);
1470 policy = SCHED_FIFO;
1471 sp.sched_priority = RCU_KTHREAD_PRIO;
1473 policy = SCHED_NORMAL;
1474 sp.sched_priority = 0;
1476 sched_setscheduler_nocheck(t, policy, &sp);
1480 * Timer handler to initiate the waking up of per-CPU kthreads that
1481 * have yielded the CPU due to excess numbers of RCU callbacks.
1482 * We wake up the per-rcu_node kthread, which in turn will wake up
1483 * the booster kthread.
1485 static void rcu_cpu_kthread_timer(unsigned long arg)
1487 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg);
1488 struct rcu_node *rnp = rdp->mynode;
1490 atomic_or(rdp->grpmask, &rnp->wakemask);
1491 invoke_rcu_node_kthread(rnp);
1495 * Drop to non-real-time priority and yield, but only after posting a
1496 * timer that will cause us to regain our real-time priority if we
1497 * remain preempted. Either way, we restore our real-time priority
1500 static void rcu_yield(void (*f)(unsigned long), unsigned long arg)
1502 struct sched_param sp;
1503 struct timer_list yield_timer;
1504 int prio = current->rt_priority;
1506 setup_timer_on_stack(&yield_timer, f, arg);
1507 mod_timer(&yield_timer, jiffies + 2);
1508 sp.sched_priority = 0;
1509 sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
1510 set_user_nice(current, 19);
1512 set_user_nice(current, 0);
1513 sp.sched_priority = prio;
1514 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1515 del_timer(&yield_timer);
1519 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1520 * This can happen while the corresponding CPU is either coming online
1521 * or going offline. We cannot wait until the CPU is fully online
1522 * before starting the kthread, because the various notifier functions
1523 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1524 * the corresponding CPU is online.
1526 * Return 1 if the kthread needs to stop, 0 otherwise.
1528 * Caller must disable bh. This function can momentarily enable it.
1530 static int rcu_cpu_kthread_should_stop(int cpu)
1532 while (cpu_is_offline(cpu) ||
1533 !cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)) ||
1534 smp_processor_id() != cpu) {
1535 if (kthread_should_stop())
1537 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
1538 per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id();
1540 schedule_timeout_uninterruptible(1);
1541 if (!cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)))
1542 set_cpus_allowed_ptr(current, cpumask_of(cpu));
1545 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1550 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1551 * RCU softirq used in flavors and configurations of RCU that do not
1552 * support RCU priority boosting.
1554 static int rcu_cpu_kthread(void *arg)
1556 int cpu = (int)(long)arg;
1557 unsigned long flags;
1559 unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu);
1561 char *workp = &per_cpu(rcu_cpu_has_work, cpu);
1563 trace_rcu_utilization("Start CPU kthread@init");
1565 *statusp = RCU_KTHREAD_WAITING;
1566 trace_rcu_utilization("End CPU kthread@rcu_wait");
1567 rcu_wait(*workp != 0 || kthread_should_stop());
1568 trace_rcu_utilization("Start CPU kthread@rcu_wait");
1570 if (rcu_cpu_kthread_should_stop(cpu)) {
1574 *statusp = RCU_KTHREAD_RUNNING;
1575 per_cpu(rcu_cpu_kthread_loops, cpu)++;
1576 local_irq_save(flags);
1579 local_irq_restore(flags);
1581 rcu_kthread_do_work();
1588 *statusp = RCU_KTHREAD_YIELDING;
1589 trace_rcu_utilization("End CPU kthread@rcu_yield");
1590 rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu);
1591 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1595 *statusp = RCU_KTHREAD_STOPPED;
1596 trace_rcu_utilization("End CPU kthread@term");
1601 * Spawn a per-CPU kthread, setting up affinity and priority.
1602 * Because the CPU hotplug lock is held, no other CPU will be attempting
1603 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1604 * attempting to access it during boot, but the locking in kthread_bind()
1605 * will enforce sufficient ordering.
1607 * Please note that we cannot simply refuse to wake up the per-CPU
1608 * kthread because kthreads are created in TASK_UNINTERRUPTIBLE state,
1609 * which can result in softlockup complaints if the task ends up being
1610 * idle for more than a couple of minutes.
1612 * However, please note also that we cannot bind the per-CPU kthread to its
1613 * CPU until that CPU is fully online. We also cannot wait until the
1614 * CPU is fully online before we create its per-CPU kthread, as this would
1615 * deadlock the system when CPU notifiers tried waiting for grace
1616 * periods. So we bind the per-CPU kthread to its CPU only if the CPU
1617 * is online. If its CPU is not yet fully online, then the code in
1618 * rcu_cpu_kthread() will wait until it is fully online, and then do
1621 static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu)
1623 struct sched_param sp;
1624 struct task_struct *t;
1626 if (!rcu_scheduler_fully_active ||
1627 per_cpu(rcu_cpu_kthread_task, cpu) != NULL)
1629 t = kthread_create_on_node(rcu_cpu_kthread,
1635 if (cpu_online(cpu))
1636 kthread_bind(t, cpu);
1637 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1638 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL);
1639 sp.sched_priority = RCU_KTHREAD_PRIO;
1640 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1641 per_cpu(rcu_cpu_kthread_task, cpu) = t;
1642 wake_up_process(t); /* Get to TASK_INTERRUPTIBLE quickly. */
1647 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1648 * kthreads when needed. We ignore requests to wake up kthreads
1649 * for offline CPUs, which is OK because force_quiescent_state()
1650 * takes care of this case.
1652 static int rcu_node_kthread(void *arg)
1655 unsigned long flags;
1657 struct rcu_node *rnp = (struct rcu_node *)arg;
1658 struct sched_param sp;
1659 struct task_struct *t;
1662 rnp->node_kthread_status = RCU_KTHREAD_WAITING;
1663 rcu_wait(atomic_read(&rnp->wakemask) != 0);
1664 rnp->node_kthread_status = RCU_KTHREAD_RUNNING;
1665 raw_spin_lock_irqsave(&rnp->lock, flags);
1666 mask = atomic_xchg(&rnp->wakemask, 0);
1667 rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
1668 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) {
1669 if ((mask & 0x1) == 0)
1672 t = per_cpu(rcu_cpu_kthread_task, cpu);
1673 if (!cpu_online(cpu) || t == NULL) {
1677 per_cpu(rcu_cpu_has_work, cpu) = 1;
1678 sp.sched_priority = RCU_KTHREAD_PRIO;
1679 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1684 rnp->node_kthread_status = RCU_KTHREAD_STOPPED;
1689 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1690 * served by the rcu_node in question. The CPU hotplug lock is still
1691 * held, so the value of rnp->qsmaskinit will be stable.
1693 * We don't include outgoingcpu in the affinity set, use -1 if there is
1694 * no outgoing CPU. If there are no CPUs left in the affinity set,
1695 * this function allows the kthread to execute on any CPU.
1697 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1701 unsigned long mask = rnp->qsmaskinit;
1703 if (rnp->node_kthread_task == NULL)
1705 if (!alloc_cpumask_var(&cm, GFP_KERNEL))
1708 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1)
1709 if ((mask & 0x1) && cpu != outgoingcpu)
1710 cpumask_set_cpu(cpu, cm);
1711 if (cpumask_weight(cm) == 0) {
1713 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++)
1714 cpumask_clear_cpu(cpu, cm);
1715 WARN_ON_ONCE(cpumask_weight(cm) == 0);
1717 set_cpus_allowed_ptr(rnp->node_kthread_task, cm);
1718 rcu_boost_kthread_setaffinity(rnp, cm);
1719 free_cpumask_var(cm);
1723 * Spawn a per-rcu_node kthread, setting priority and affinity.
1724 * Called during boot before online/offline can happen, or, if
1725 * during runtime, with the main CPU-hotplug locks held. So only
1726 * one of these can be executing at a time.
1728 static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp,
1729 struct rcu_node *rnp)
1731 unsigned long flags;
1732 int rnp_index = rnp - &rsp->node[0];
1733 struct sched_param sp;
1734 struct task_struct *t;
1736 if (!rcu_scheduler_fully_active ||
1737 rnp->qsmaskinit == 0)
1739 if (rnp->node_kthread_task == NULL) {
1740 t = kthread_create(rcu_node_kthread, (void *)rnp,
1741 "rcun/%d", rnp_index);
1744 raw_spin_lock_irqsave(&rnp->lock, flags);
1745 rnp->node_kthread_task = t;
1746 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1747 sp.sched_priority = 99;
1748 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1749 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1751 return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index);
1755 * Spawn all kthreads -- called as soon as the scheduler is running.
1757 static int __init rcu_spawn_kthreads(void)
1760 struct rcu_node *rnp;
1762 rcu_scheduler_fully_active = 1;
1763 for_each_possible_cpu(cpu) {
1764 per_cpu(rcu_cpu_has_work, cpu) = 0;
1765 if (cpu_online(cpu))
1766 (void)rcu_spawn_one_cpu_kthread(cpu);
1768 rnp = rcu_get_root(rcu_state);
1769 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1770 if (NUM_RCU_NODES > 1) {
1771 rcu_for_each_leaf_node(rcu_state, rnp)
1772 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1776 early_initcall(rcu_spawn_kthreads);
1778 static void __cpuinit rcu_prepare_kthreads(int cpu)
1780 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
1781 struct rcu_node *rnp = rdp->mynode;
1783 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1784 if (rcu_scheduler_fully_active) {
1785 (void)rcu_spawn_one_cpu_kthread(cpu);
1786 if (rnp->node_kthread_task == NULL)
1787 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1791 #else /* #ifdef CONFIG_RCU_BOOST */
1793 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1795 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1798 static void invoke_rcu_callbacks_kthread(void)
1803 static bool rcu_is_callbacks_kthread(void)
1808 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1812 #ifdef CONFIG_HOTPLUG_CPU
1814 static void rcu_stop_cpu_kthread(int cpu)
1818 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1820 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1824 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1828 static int __init rcu_scheduler_really_started(void)
1830 rcu_scheduler_fully_active = 1;
1833 early_initcall(rcu_scheduler_really_started);
1835 static void __cpuinit rcu_prepare_kthreads(int cpu)
1839 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1841 #if !defined(CONFIG_RCU_FAST_NO_HZ)
1844 * Check to see if any future RCU-related work will need to be done
1845 * by the current CPU, even if none need be done immediately, returning
1846 * 1 if so. This function is part of the RCU implementation; it is -not-
1847 * an exported member of the RCU API.
1849 * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
1850 * any flavor of RCU.
1852 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1854 *delta_jiffies = ULONG_MAX;
1855 return rcu_cpu_has_callbacks(cpu);
1859 * Because we do not have RCU_FAST_NO_HZ, don't bother initializing for it.
1861 static void rcu_prepare_for_idle_init(int cpu)
1866 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1869 static void rcu_cleanup_after_idle(int cpu)
1874 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
1877 static void rcu_prepare_for_idle(int cpu)
1882 * Don't bother keeping a running count of the number of RCU callbacks
1883 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1885 static void rcu_idle_count_callbacks_posted(void)
1889 #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1892 * This code is invoked when a CPU goes idle, at which point we want
1893 * to have the CPU do everything required for RCU so that it can enter
1894 * the energy-efficient dyntick-idle mode. This is handled by a
1895 * state machine implemented by rcu_prepare_for_idle() below.
1897 * The following three proprocessor symbols control this state machine:
1899 * RCU_IDLE_FLUSHES gives the maximum number of times that we will attempt
1900 * to satisfy RCU. Beyond this point, it is better to incur a periodic
1901 * scheduling-clock interrupt than to loop through the state machine
1903 * RCU_IDLE_OPT_FLUSHES gives the number of RCU_IDLE_FLUSHES that are
1904 * optional if RCU does not need anything immediately from this
1905 * CPU, even if this CPU still has RCU callbacks queued. The first
1906 * times through the state machine are mandatory: we need to give
1907 * the state machine a chance to communicate a quiescent state
1909 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1910 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1911 * is sized to be roughly one RCU grace period. Those energy-efficiency
1912 * benchmarkers who might otherwise be tempted to set this to a large
1913 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1914 * system. And if you are -that- concerned about energy efficiency,
1915 * just power the system down and be done with it!
1916 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1917 * permitted to sleep in dyntick-idle mode with only lazy RCU
1918 * callbacks pending. Setting this too high can OOM your system.
1920 * The values below work well in practice. If future workloads require
1921 * adjustment, they can be converted into kernel config parameters, though
1922 * making the state machine smarter might be a better option.
1924 #define RCU_IDLE_FLUSHES 5 /* Number of dyntick-idle tries. */
1925 #define RCU_IDLE_OPT_FLUSHES 3 /* Optional dyntick-idle tries. */
1926 #define RCU_IDLE_GP_DELAY 6 /* Roughly one grace period. */
1927 #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
1930 * Does the specified flavor of RCU have non-lazy callbacks pending on
1931 * the specified CPU? Both RCU flavor and CPU are specified by the
1932 * rcu_data structure.
1934 static bool __rcu_cpu_has_nonlazy_callbacks(struct rcu_data *rdp)
1936 return rdp->qlen != rdp->qlen_lazy;
1939 #ifdef CONFIG_TREE_PREEMPT_RCU
1942 * Are there non-lazy RCU-preempt callbacks? (There cannot be if there
1943 * is no RCU-preempt in the kernel.)
1945 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1947 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
1949 return __rcu_cpu_has_nonlazy_callbacks(rdp);
1952 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1954 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1959 #endif /* else #ifdef CONFIG_TREE_PREEMPT_RCU */
1962 * Does any flavor of RCU have non-lazy callbacks on the specified CPU?
1964 static bool rcu_cpu_has_nonlazy_callbacks(int cpu)
1966 return __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_sched_data, cpu)) ||
1967 __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_bh_data, cpu)) ||
1968 rcu_preempt_cpu_has_nonlazy_callbacks(cpu);
1972 * Allow the CPU to enter dyntick-idle mode if either: (1) There are no
1973 * callbacks on this CPU, (2) this CPU has not yet attempted to enter
1974 * dyntick-idle mode, or (3) this CPU is in the process of attempting to
1975 * enter dyntick-idle mode. Otherwise, if we have recently tried and failed
1976 * to enter dyntick-idle mode, we refuse to try to enter it. After all,
1977 * it is better to incur scheduling-clock interrupts than to spin
1978 * continuously for the same time duration!
1980 * The delta_jiffies argument is used to store the time when RCU is
1981 * going to need the CPU again if it still has callbacks. The reason
1982 * for this is that rcu_prepare_for_idle() might need to post a timer,
1983 * but if so, it will do so after tick_nohz_stop_sched_tick() has set
1984 * the wakeup time for this CPU. This means that RCU's timer can be
1985 * delayed until the wakeup time, which defeats the purpose of posting
1988 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1990 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1992 /* Flag a new idle sojourn to the idle-entry state machine. */
1993 rdtp->idle_first_pass = 1;
1994 /* If no callbacks, RCU doesn't need the CPU. */
1995 if (!rcu_cpu_has_callbacks(cpu)) {
1996 *delta_jiffies = ULONG_MAX;
1999 if (rdtp->dyntick_holdoff == jiffies) {
2000 /* RCU recently tried and failed, so don't try again. */
2004 /* Set up for the possibility that RCU will post a timer. */
2005 if (rcu_cpu_has_nonlazy_callbacks(cpu))
2006 *delta_jiffies = RCU_IDLE_GP_DELAY;
2008 *delta_jiffies = RCU_IDLE_LAZY_GP_DELAY;
2013 * Handler for smp_call_function_single(). The only point of this
2014 * handler is to wake the CPU up, so the handler does only tracing.
2016 void rcu_idle_demigrate(void *unused)
2018 trace_rcu_prep_idle("Demigrate");
2022 * Timer handler used to force CPU to start pushing its remaining RCU
2023 * callbacks in the case where it entered dyntick-idle mode with callbacks
2024 * pending. The hander doesn't really need to do anything because the
2025 * real work is done upon re-entry to idle, or by the next scheduling-clock
2026 * interrupt should idle not be re-entered.
2028 * One special case: the timer gets migrated without awakening the CPU
2029 * on which the timer was scheduled on. In this case, we must wake up
2030 * that CPU. We do so with smp_call_function_single().
2032 static void rcu_idle_gp_timer_func(unsigned long cpu_in)
2034 int cpu = (int)cpu_in;
2036 trace_rcu_prep_idle("Timer");
2037 if (cpu != smp_processor_id())
2038 smp_call_function_single(cpu, rcu_idle_demigrate, NULL, 0);
2040 WARN_ON_ONCE(1); /* Getting here can hang the system... */
2044 * Initialize the timer used to pull CPUs out of dyntick-idle mode.
2046 static void rcu_prepare_for_idle_init(int cpu)
2048 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2050 rdtp->dyntick_holdoff = jiffies - 1;
2051 setup_timer(&rdtp->idle_gp_timer, rcu_idle_gp_timer_func, cpu);
2052 rdtp->idle_gp_timer_expires = jiffies - 1;
2053 rdtp->idle_first_pass = 1;
2057 * Clean up for exit from idle. Because we are exiting from idle, there
2058 * is no longer any point to ->idle_gp_timer, so cancel it. This will
2059 * do nothing if this timer is not active, so just cancel it unconditionally.
2061 static void rcu_cleanup_after_idle(int cpu)
2063 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2065 del_timer(&rdtp->idle_gp_timer);
2066 trace_rcu_prep_idle("Cleanup after idle");
2070 * Check to see if any RCU-related work can be done by the current CPU,
2071 * and if so, schedule a softirq to get it done. This function is part
2072 * of the RCU implementation; it is -not- an exported member of the RCU API.
2074 * The idea is for the current CPU to clear out all work required by the
2075 * RCU core for the current grace period, so that this CPU can be permitted
2076 * to enter dyntick-idle mode. In some cases, it will need to be awakened
2077 * at the end of the grace period by whatever CPU ends the grace period.
2078 * This allows CPUs to go dyntick-idle more quickly, and to reduce the
2079 * number of wakeups by a modest integer factor.
2081 * Because it is not legal to invoke rcu_process_callbacks() with irqs
2082 * disabled, we do one pass of force_quiescent_state(), then do a
2083 * invoke_rcu_core() to cause rcu_process_callbacks() to be invoked
2084 * later. The ->dyntick_drain field controls the sequencing.
2086 * The caller must have disabled interrupts.
2088 static void rcu_prepare_for_idle(int cpu)
2090 struct timer_list *tp;
2091 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2094 * If this is an idle re-entry, for example, due to use of
2095 * RCU_NONIDLE() or the new idle-loop tracing API within the idle
2096 * loop, then don't take any state-machine actions, unless the
2097 * momentary exit from idle queued additional non-lazy callbacks.
2098 * Instead, repost the ->idle_gp_timer if this CPU has callbacks
2101 if (!rdtp->idle_first_pass &&
2102 (rdtp->nonlazy_posted == rdtp->nonlazy_posted_snap)) {
2103 if (rcu_cpu_has_callbacks(cpu)) {
2104 tp = &rdtp->idle_gp_timer;
2105 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2109 rdtp->idle_first_pass = 0;
2110 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted - 1;
2113 * If there are no callbacks on this CPU, enter dyntick-idle mode.
2114 * Also reset state to avoid prejudicing later attempts.
2116 if (!rcu_cpu_has_callbacks(cpu)) {
2117 rdtp->dyntick_holdoff = jiffies - 1;
2118 rdtp->dyntick_drain = 0;
2119 trace_rcu_prep_idle("No callbacks");
2124 * If in holdoff mode, just return. We will presumably have
2125 * refrained from disabling the scheduling-clock tick.
2127 if (rdtp->dyntick_holdoff == jiffies) {
2128 trace_rcu_prep_idle("In holdoff");
2132 /* Check and update the ->dyntick_drain sequencing. */
2133 if (rdtp->dyntick_drain <= 0) {
2134 /* First time through, initialize the counter. */
2135 rdtp->dyntick_drain = RCU_IDLE_FLUSHES;
2136 } else if (rdtp->dyntick_drain <= RCU_IDLE_OPT_FLUSHES &&
2137 !rcu_pending(cpu) &&
2138 !local_softirq_pending()) {
2139 /* Can we go dyntick-idle despite still having callbacks? */
2140 rdtp->dyntick_drain = 0;
2141 rdtp->dyntick_holdoff = jiffies;
2142 if (rcu_cpu_has_nonlazy_callbacks(cpu)) {
2143 trace_rcu_prep_idle("Dyntick with callbacks");
2144 rdtp->idle_gp_timer_expires =
2145 jiffies + RCU_IDLE_GP_DELAY;
2147 rdtp->idle_gp_timer_expires =
2148 jiffies + RCU_IDLE_LAZY_GP_DELAY;
2149 trace_rcu_prep_idle("Dyntick with lazy callbacks");
2151 tp = &rdtp->idle_gp_timer;
2152 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2153 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted;
2154 return; /* Nothing more to do immediately. */
2155 } else if (--(rdtp->dyntick_drain) <= 0) {
2156 /* We have hit the limit, so time to give up. */
2157 rdtp->dyntick_holdoff = jiffies;
2158 trace_rcu_prep_idle("Begin holdoff");
2159 invoke_rcu_core(); /* Force the CPU out of dyntick-idle. */
2164 * Do one step of pushing the remaining RCU callbacks through
2165 * the RCU core state machine.
2167 #ifdef CONFIG_TREE_PREEMPT_RCU
2168 if (per_cpu(rcu_preempt_data, cpu).nxtlist) {
2169 rcu_preempt_qs(cpu);
2170 force_quiescent_state(&rcu_preempt_state, 0);
2172 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2173 if (per_cpu(rcu_sched_data, cpu).nxtlist) {
2175 force_quiescent_state(&rcu_sched_state, 0);
2177 if (per_cpu(rcu_bh_data, cpu).nxtlist) {
2179 force_quiescent_state(&rcu_bh_state, 0);
2183 * If RCU callbacks are still pending, RCU still needs this CPU.
2184 * So try forcing the callbacks through the grace period.
2186 if (rcu_cpu_has_callbacks(cpu)) {
2187 trace_rcu_prep_idle("More callbacks");
2190 trace_rcu_prep_idle("Callbacks drained");
2194 * Keep a running count of the number of non-lazy callbacks posted
2195 * on this CPU. This running counter (which is never decremented) allows
2196 * rcu_prepare_for_idle() to detect when something out of the idle loop
2197 * posts a callback, even if an equal number of callbacks are invoked.
2198 * Of course, callbacks should only be posted from within a trace event
2199 * designed to be called from idle or from within RCU_NONIDLE().
2201 static void rcu_idle_count_callbacks_posted(void)
2203 __this_cpu_add(rcu_dynticks.nonlazy_posted, 1);
2206 #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
2208 #ifdef CONFIG_RCU_CPU_STALL_INFO
2210 #ifdef CONFIG_RCU_FAST_NO_HZ
2212 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2214 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2215 struct timer_list *tltp = &rdtp->idle_gp_timer;
2217 sprintf(cp, "drain=%d %c timer=%lu",
2218 rdtp->dyntick_drain,
2219 rdtp->dyntick_holdoff == jiffies ? 'H' : '.',
2220 timer_pending(tltp) ? tltp->expires - jiffies : -1);
2223 #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
2225 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2229 #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
2231 /* Initiate the stall-info list. */
2232 static void print_cpu_stall_info_begin(void)
2234 printk(KERN_CONT "\n");
2238 * Print out diagnostic information for the specified stalled CPU.
2240 * If the specified CPU is aware of the current RCU grace period
2241 * (flavor specified by rsp), then print the number of scheduling
2242 * clock interrupts the CPU has taken during the time that it has
2243 * been aware. Otherwise, print the number of RCU grace periods
2244 * that this CPU is ignorant of, for example, "1" if the CPU was
2245 * aware of the previous grace period.
2247 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
2249 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2251 char fast_no_hz[72];
2252 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2253 struct rcu_dynticks *rdtp = rdp->dynticks;
2255 unsigned long ticks_value;
2257 if (rsp->gpnum == rdp->gpnum) {
2258 ticks_title = "ticks this GP";
2259 ticks_value = rdp->ticks_this_gp;
2261 ticks_title = "GPs behind";
2262 ticks_value = rsp->gpnum - rdp->gpnum;
2264 print_cpu_stall_fast_no_hz(fast_no_hz, cpu);
2265 printk(KERN_ERR "\t%d: (%lu %s) idle=%03x/%llx/%d %s\n",
2266 cpu, ticks_value, ticks_title,
2267 atomic_read(&rdtp->dynticks) & 0xfff,
2268 rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting,
2272 /* Terminate the stall-info list. */
2273 static void print_cpu_stall_info_end(void)
2275 printk(KERN_ERR "\t");
2278 /* Zero ->ticks_this_gp for all flavors of RCU. */
2279 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2281 rdp->ticks_this_gp = 0;
2284 /* Increment ->ticks_this_gp for all flavors of RCU. */
2285 static void increment_cpu_stall_ticks(void)
2287 __get_cpu_var(rcu_sched_data).ticks_this_gp++;
2288 __get_cpu_var(rcu_bh_data).ticks_this_gp++;
2289 #ifdef CONFIG_TREE_PREEMPT_RCU
2290 __get_cpu_var(rcu_preempt_data).ticks_this_gp++;
2291 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2294 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
2296 static void print_cpu_stall_info_begin(void)
2298 printk(KERN_CONT " {");
2301 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2303 printk(KERN_CONT " %d", cpu);
2306 static void print_cpu_stall_info_end(void)
2308 printk(KERN_CONT "} ");
2311 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2315 static void increment_cpu_stall_ticks(void)
2319 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */