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 void rcu_read_unlock_special(struct task_struct *t);
87 static int rcu_preempted_readers_exp(struct rcu_node *rnp);
90 * Tell them what RCU they are running.
92 static void __init rcu_bootup_announce(void)
94 printk(KERN_INFO "Preemptible hierarchical RCU implementation.\n");
95 rcu_bootup_announce_oddness();
99 * Return the number of RCU-preempt batches processed thus far
100 * for debug and statistics.
102 long rcu_batches_completed_preempt(void)
104 return rcu_preempt_state.completed;
106 EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt);
109 * Return the number of RCU batches processed thus far for debug & stats.
111 long rcu_batches_completed(void)
113 return rcu_batches_completed_preempt();
115 EXPORT_SYMBOL_GPL(rcu_batches_completed);
118 * Force a quiescent state for preemptible RCU.
120 void rcu_force_quiescent_state(void)
122 force_quiescent_state(&rcu_preempt_state, 0);
124 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
127 * Record a preemptible-RCU quiescent state for the specified CPU. Note
128 * that this just means that the task currently running on the CPU is
129 * not in a quiescent state. There might be any number of tasks blocked
130 * while in an RCU read-side critical section.
132 * Unlike the other rcu_*_qs() functions, callers to this function
133 * must disable irqs in order to protect the assignment to
134 * ->rcu_read_unlock_special.
136 static void rcu_preempt_qs(int cpu)
138 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
140 rdp->passed_quiesce_gpnum = rdp->gpnum;
142 if (rdp->passed_quiesce == 0)
143 trace_rcu_grace_period("rcu_preempt", rdp->gpnum, "cpuqs");
144 rdp->passed_quiesce = 1;
145 current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
149 * We have entered the scheduler, and the current task might soon be
150 * context-switched away from. If this task is in an RCU read-side
151 * critical section, we will no longer be able to rely on the CPU to
152 * record that fact, so we enqueue the task on the blkd_tasks list.
153 * The task will dequeue itself when it exits the outermost enclosing
154 * RCU read-side critical section. Therefore, the current grace period
155 * cannot be permitted to complete until the blkd_tasks list entries
156 * predating the current grace period drain, in other words, until
157 * rnp->gp_tasks becomes NULL.
159 * Caller must disable preemption.
161 static void rcu_preempt_note_context_switch(int cpu)
163 struct task_struct *t = current;
165 struct rcu_data *rdp;
166 struct rcu_node *rnp;
168 if (t->rcu_read_lock_nesting > 0 &&
169 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
171 /* Possibly blocking in an RCU read-side critical section. */
172 rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu);
174 raw_spin_lock_irqsave(&rnp->lock, flags);
175 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
176 t->rcu_blocked_node = rnp;
179 * If this CPU has already checked in, then this task
180 * will hold up the next grace period rather than the
181 * current grace period. Queue the task accordingly.
182 * If the task is queued for the current grace period
183 * (i.e., this CPU has not yet passed through a quiescent
184 * state for the current grace period), then as long
185 * as that task remains queued, the current grace period
186 * cannot end. Note that there is some uncertainty as
187 * to exactly when the current grace period started.
188 * We take a conservative approach, which can result
189 * in unnecessarily waiting on tasks that started very
190 * slightly after the current grace period began. C'est
193 * But first, note that the current CPU must still be
196 WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0);
197 WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
198 if ((rnp->qsmask & rdp->grpmask) && rnp->gp_tasks != NULL) {
199 list_add(&t->rcu_node_entry, rnp->gp_tasks->prev);
200 rnp->gp_tasks = &t->rcu_node_entry;
201 #ifdef CONFIG_RCU_BOOST
202 if (rnp->boost_tasks != NULL)
203 rnp->boost_tasks = rnp->gp_tasks;
204 #endif /* #ifdef CONFIG_RCU_BOOST */
206 list_add(&t->rcu_node_entry, &rnp->blkd_tasks);
207 if (rnp->qsmask & rdp->grpmask)
208 rnp->gp_tasks = &t->rcu_node_entry;
210 trace_rcu_preempt_task(rdp->rsp->name,
212 (rnp->qsmask & rdp->grpmask)
215 raw_spin_unlock_irqrestore(&rnp->lock, flags);
216 } else if (t->rcu_read_lock_nesting < 0 &&
217 t->rcu_read_unlock_special) {
220 * Complete exit from RCU read-side critical section on
221 * behalf of preempted instance of __rcu_read_unlock().
223 rcu_read_unlock_special(t);
227 * Either we were not in an RCU read-side critical section to
228 * begin with, or we have now recorded that critical section
229 * globally. Either way, we can now note a quiescent state
230 * for this CPU. Again, if we were in an RCU read-side critical
231 * section, and if that critical section was blocking the current
232 * grace period, then the fact that the task has been enqueued
233 * means that we continue to block the current grace period.
235 local_irq_save(flags);
237 local_irq_restore(flags);
241 * Tree-preemptible RCU implementation for rcu_read_lock().
242 * Just increment ->rcu_read_lock_nesting, shared state will be updated
245 void __rcu_read_lock(void)
247 current->rcu_read_lock_nesting++;
248 barrier(); /* needed if we ever invoke rcu_read_lock in rcutree.c */
250 EXPORT_SYMBOL_GPL(__rcu_read_lock);
253 * Check for preempted RCU readers blocking the current grace period
254 * for the specified rcu_node structure. If the caller needs a reliable
255 * answer, it must hold the rcu_node's ->lock.
257 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
259 return rnp->gp_tasks != NULL;
263 * Record a quiescent state for all tasks that were previously queued
264 * on the specified rcu_node structure and that were blocking the current
265 * RCU grace period. The caller must hold the specified rnp->lock with
266 * irqs disabled, and this lock is released upon return, but irqs remain
269 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
270 __releases(rnp->lock)
273 struct rcu_node *rnp_p;
275 if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
276 raw_spin_unlock_irqrestore(&rnp->lock, flags);
277 return; /* Still need more quiescent states! */
283 * Either there is only one rcu_node in the tree,
284 * or tasks were kicked up to root rcu_node due to
285 * CPUs going offline.
287 rcu_report_qs_rsp(&rcu_preempt_state, flags);
291 /* Report up the rest of the hierarchy. */
293 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
294 raw_spin_lock(&rnp_p->lock); /* irqs already disabled. */
295 rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags);
299 * Advance a ->blkd_tasks-list pointer to the next entry, instead
300 * returning NULL if at the end of the list.
302 static struct list_head *rcu_next_node_entry(struct task_struct *t,
303 struct rcu_node *rnp)
305 struct list_head *np;
307 np = t->rcu_node_entry.next;
308 if (np == &rnp->blkd_tasks)
314 * Handle special cases during rcu_read_unlock(), such as needing to
315 * notify RCU core processing or task having blocked during the RCU
316 * read-side critical section.
318 static noinline void rcu_read_unlock_special(struct task_struct *t)
324 struct list_head *np;
325 #ifdef CONFIG_RCU_BOOST
326 struct rt_mutex *rbmp = NULL;
327 #endif /* #ifdef CONFIG_RCU_BOOST */
328 struct rcu_node *rnp;
331 /* NMI handlers cannot block and cannot safely manipulate state. */
335 local_irq_save(flags);
338 * If RCU core is waiting for this CPU to exit critical section,
339 * let it know that we have done so.
341 special = t->rcu_read_unlock_special;
342 if (special & RCU_READ_UNLOCK_NEED_QS) {
343 rcu_preempt_qs(smp_processor_id());
346 /* Hardware IRQ handlers cannot block. */
347 if (in_irq() || in_serving_softirq()) {
348 local_irq_restore(flags);
352 /* Clean up if blocked during RCU read-side critical section. */
353 if (special & RCU_READ_UNLOCK_BLOCKED) {
354 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
357 * Remove this task from the list it blocked on. The
358 * task can migrate while we acquire the lock, but at
359 * most one time. So at most two passes through loop.
362 rnp = t->rcu_blocked_node;
363 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
364 if (rnp == t->rcu_blocked_node)
366 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
368 empty = !rcu_preempt_blocked_readers_cgp(rnp);
369 empty_exp = !rcu_preempted_readers_exp(rnp);
370 smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
371 np = rcu_next_node_entry(t, rnp);
372 list_del_init(&t->rcu_node_entry);
373 t->rcu_blocked_node = NULL;
374 trace_rcu_unlock_preempted_task("rcu_preempt",
376 if (&t->rcu_node_entry == rnp->gp_tasks)
378 if (&t->rcu_node_entry == rnp->exp_tasks)
380 #ifdef CONFIG_RCU_BOOST
381 if (&t->rcu_node_entry == rnp->boost_tasks)
382 rnp->boost_tasks = np;
383 /* Snapshot/clear ->rcu_boost_mutex with rcu_node lock held. */
384 if (t->rcu_boost_mutex) {
385 rbmp = t->rcu_boost_mutex;
386 t->rcu_boost_mutex = NULL;
388 #endif /* #ifdef CONFIG_RCU_BOOST */
391 * If this was the last task on the current list, and if
392 * we aren't waiting on any CPUs, report the quiescent state.
393 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
394 * so we must take a snapshot of the expedited state.
396 empty_exp_now = !rcu_preempted_readers_exp(rnp);
397 if (!empty && !rcu_preempt_blocked_readers_cgp(rnp)) {
398 trace_rcu_quiescent_state_report("preempt_rcu",
405 rcu_report_unblock_qs_rnp(rnp, flags);
407 raw_spin_unlock_irqrestore(&rnp->lock, flags);
409 #ifdef CONFIG_RCU_BOOST
410 /* Unboost if we were boosted. */
412 rt_mutex_unlock(rbmp);
413 #endif /* #ifdef CONFIG_RCU_BOOST */
416 * If this was the last task on the expedited lists,
417 * then we need to report up the rcu_node hierarchy.
419 if (!empty_exp && empty_exp_now)
420 rcu_report_exp_rnp(&rcu_preempt_state, rnp, true);
422 local_irq_restore(flags);
427 * Tree-preemptible RCU implementation for rcu_read_unlock().
428 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
429 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
430 * invoke rcu_read_unlock_special() to clean up after a context switch
431 * in an RCU read-side critical section and other special cases.
433 void __rcu_read_unlock(void)
435 struct task_struct *t = current;
437 if (t->rcu_read_lock_nesting != 1)
438 --t->rcu_read_lock_nesting;
440 barrier(); /* critical section before exit code. */
441 t->rcu_read_lock_nesting = INT_MIN;
442 barrier(); /* assign before ->rcu_read_unlock_special load */
443 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
444 rcu_read_unlock_special(t);
445 barrier(); /* ->rcu_read_unlock_special load before assign */
446 t->rcu_read_lock_nesting = 0;
448 #ifdef CONFIG_PROVE_LOCKING
450 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
452 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
454 #endif /* #ifdef CONFIG_PROVE_LOCKING */
456 EXPORT_SYMBOL_GPL(__rcu_read_unlock);
458 #ifdef CONFIG_RCU_CPU_STALL_VERBOSE
461 * Dump detailed information for all tasks blocking the current RCU
462 * grace period on the specified rcu_node structure.
464 static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
467 struct task_struct *t;
469 if (!rcu_preempt_blocked_readers_cgp(rnp))
471 raw_spin_lock_irqsave(&rnp->lock, flags);
472 t = list_entry(rnp->gp_tasks,
473 struct task_struct, rcu_node_entry);
474 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry)
476 raw_spin_unlock_irqrestore(&rnp->lock, flags);
480 * Dump detailed information for all tasks blocking the current RCU
483 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
485 struct rcu_node *rnp = rcu_get_root(rsp);
487 rcu_print_detail_task_stall_rnp(rnp);
488 rcu_for_each_leaf_node(rsp, rnp)
489 rcu_print_detail_task_stall_rnp(rnp);
492 #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
494 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
498 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
500 #ifdef CONFIG_RCU_CPU_STALL_INFO
502 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
504 printk(KERN_ERR "\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
505 rnp->level, rnp->grplo, rnp->grphi);
508 static void rcu_print_task_stall_end(void)
510 printk(KERN_CONT "\n");
513 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
515 static void rcu_print_task_stall_begin(struct rcu_node *rnp)
519 static void rcu_print_task_stall_end(void)
523 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */
526 * Scan the current list of tasks blocked within RCU read-side critical
527 * sections, printing out the tid of each.
529 static int rcu_print_task_stall(struct rcu_node *rnp)
531 struct task_struct *t;
534 if (!rcu_preempt_blocked_readers_cgp(rnp))
536 rcu_print_task_stall_begin(rnp);
537 t = list_entry(rnp->gp_tasks,
538 struct task_struct, rcu_node_entry);
539 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
540 printk(KERN_CONT " P%d", t->pid);
543 rcu_print_task_stall_end();
548 * Check that the list of blocked tasks for the newly completed grace
549 * period is in fact empty. It is a serious bug to complete a grace
550 * period that still has RCU readers blocked! This function must be
551 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
552 * must be held by the caller.
554 * Also, if there are blocked tasks on the list, they automatically
555 * block the newly created grace period, so set up ->gp_tasks accordingly.
557 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
559 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
560 if (!list_empty(&rnp->blkd_tasks))
561 rnp->gp_tasks = rnp->blkd_tasks.next;
562 WARN_ON_ONCE(rnp->qsmask);
565 #ifdef CONFIG_HOTPLUG_CPU
568 * Handle tasklist migration for case in which all CPUs covered by the
569 * specified rcu_node have gone offline. Move them up to the root
570 * rcu_node. The reason for not just moving them to the immediate
571 * parent is to remove the need for rcu_read_unlock_special() to
572 * make more than two attempts to acquire the target rcu_node's lock.
573 * Returns true if there were tasks blocking the current RCU grace
576 * Returns 1 if there was previously a task blocking the current grace
577 * period on the specified rcu_node structure.
579 * The caller must hold rnp->lock with irqs disabled.
581 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
582 struct rcu_node *rnp,
583 struct rcu_data *rdp)
585 struct list_head *lp;
586 struct list_head *lp_root;
588 struct rcu_node *rnp_root = rcu_get_root(rsp);
589 struct task_struct *t;
591 if (rnp == rnp_root) {
592 WARN_ONCE(1, "Last CPU thought to be offlined?");
593 return 0; /* Shouldn't happen: at least one CPU online. */
596 /* If we are on an internal node, complain bitterly. */
597 WARN_ON_ONCE(rnp != rdp->mynode);
600 * Move tasks up to root rcu_node. Don't try to get fancy for
601 * this corner-case operation -- just put this node's tasks
602 * at the head of the root node's list, and update the root node's
603 * ->gp_tasks and ->exp_tasks pointers to those of this node's,
604 * if non-NULL. This might result in waiting for more tasks than
605 * absolutely necessary, but this is a good performance/complexity
608 if (rcu_preempt_blocked_readers_cgp(rnp) && rnp->qsmask == 0)
609 retval |= RCU_OFL_TASKS_NORM_GP;
610 if (rcu_preempted_readers_exp(rnp))
611 retval |= RCU_OFL_TASKS_EXP_GP;
612 lp = &rnp->blkd_tasks;
613 lp_root = &rnp_root->blkd_tasks;
614 while (!list_empty(lp)) {
615 t = list_entry(lp->next, typeof(*t), rcu_node_entry);
616 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
617 list_del(&t->rcu_node_entry);
618 t->rcu_blocked_node = rnp_root;
619 list_add(&t->rcu_node_entry, lp_root);
620 if (&t->rcu_node_entry == rnp->gp_tasks)
621 rnp_root->gp_tasks = rnp->gp_tasks;
622 if (&t->rcu_node_entry == rnp->exp_tasks)
623 rnp_root->exp_tasks = rnp->exp_tasks;
624 #ifdef CONFIG_RCU_BOOST
625 if (&t->rcu_node_entry == rnp->boost_tasks)
626 rnp_root->boost_tasks = rnp->boost_tasks;
627 #endif /* #ifdef CONFIG_RCU_BOOST */
628 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
631 #ifdef CONFIG_RCU_BOOST
632 /* In case root is being boosted and leaf is not. */
633 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
634 if (rnp_root->boost_tasks != NULL &&
635 rnp_root->boost_tasks != rnp_root->gp_tasks)
636 rnp_root->boost_tasks = rnp_root->gp_tasks;
637 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
638 #endif /* #ifdef CONFIG_RCU_BOOST */
640 rnp->gp_tasks = NULL;
641 rnp->exp_tasks = NULL;
645 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
648 * Check for a quiescent state from the current CPU. When a task blocks,
649 * the task is recorded in the corresponding CPU's rcu_node structure,
650 * which is checked elsewhere.
652 * Caller must disable hard irqs.
654 static void rcu_preempt_check_callbacks(int cpu)
656 struct task_struct *t = current;
658 if (t->rcu_read_lock_nesting == 0) {
662 if (t->rcu_read_lock_nesting > 0 &&
663 per_cpu(rcu_preempt_data, cpu).qs_pending)
664 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
667 #ifdef CONFIG_RCU_BOOST
669 static void rcu_preempt_do_callbacks(void)
671 rcu_do_batch(&rcu_preempt_state, &__get_cpu_var(rcu_preempt_data));
674 #endif /* #ifdef CONFIG_RCU_BOOST */
677 * Queue a preemptible-RCU callback for invocation after a grace period.
679 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
681 __call_rcu(head, func, &rcu_preempt_state, 0);
683 EXPORT_SYMBOL_GPL(call_rcu);
686 * Queue an RCU callback for lazy invocation after a grace period.
687 * This will likely be later named something like "call_rcu_lazy()",
688 * but this change will require some way of tagging the lazy RCU
689 * callbacks in the list of pending callbacks. Until then, this
690 * function may only be called from __kfree_rcu().
692 void kfree_call_rcu(struct rcu_head *head,
693 void (*func)(struct rcu_head *rcu))
695 __call_rcu(head, func, &rcu_preempt_state, 1);
697 EXPORT_SYMBOL_GPL(kfree_call_rcu);
700 * synchronize_rcu - wait until a grace period has elapsed.
702 * Control will return to the caller some time after a full grace
703 * period has elapsed, in other words after all currently executing RCU
704 * read-side critical sections have completed. Note, however, that
705 * upon return from synchronize_rcu(), the caller might well be executing
706 * concurrently with new RCU read-side critical sections that began while
707 * synchronize_rcu() was waiting. RCU read-side critical sections are
708 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
710 void synchronize_rcu(void)
712 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
713 !lock_is_held(&rcu_lock_map) &&
714 !lock_is_held(&rcu_sched_lock_map),
715 "Illegal synchronize_rcu() in RCU read-side critical section");
716 if (!rcu_scheduler_active)
718 wait_rcu_gp(call_rcu);
720 EXPORT_SYMBOL_GPL(synchronize_rcu);
722 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
723 static long sync_rcu_preempt_exp_count;
724 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
727 * Return non-zero if there are any tasks in RCU read-side critical
728 * sections blocking the current preemptible-RCU expedited grace period.
729 * If there is no preemptible-RCU expedited grace period currently in
730 * progress, returns zero unconditionally.
732 static int rcu_preempted_readers_exp(struct rcu_node *rnp)
734 return rnp->exp_tasks != NULL;
738 * return non-zero if there is no RCU expedited grace period in progress
739 * for the specified rcu_node structure, in other words, if all CPUs and
740 * tasks covered by the specified rcu_node structure have done their bit
741 * for the current expedited grace period. Works only for preemptible
742 * RCU -- other RCU implementation use other means.
744 * Caller must hold sync_rcu_preempt_exp_mutex.
746 static int sync_rcu_preempt_exp_done(struct rcu_node *rnp)
748 return !rcu_preempted_readers_exp(rnp) &&
749 ACCESS_ONCE(rnp->expmask) == 0;
753 * Report the exit from RCU read-side critical section for the last task
754 * that queued itself during or before the current expedited preemptible-RCU
755 * grace period. This event is reported either to the rcu_node structure on
756 * which the task was queued or to one of that rcu_node structure's ancestors,
757 * recursively up the tree. (Calm down, calm down, we do the recursion
760 * Most callers will set the "wake" flag, but the task initiating the
761 * expedited grace period need not wake itself.
763 * Caller must hold sync_rcu_preempt_exp_mutex.
765 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
771 raw_spin_lock_irqsave(&rnp->lock, flags);
773 if (!sync_rcu_preempt_exp_done(rnp)) {
774 raw_spin_unlock_irqrestore(&rnp->lock, flags);
777 if (rnp->parent == NULL) {
778 raw_spin_unlock_irqrestore(&rnp->lock, flags);
780 wake_up(&sync_rcu_preempt_exp_wq);
784 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
786 raw_spin_lock(&rnp->lock); /* irqs already disabled */
787 rnp->expmask &= ~mask;
792 * Snapshot the tasks blocking the newly started preemptible-RCU expedited
793 * grace period for the specified rcu_node structure. If there are no such
794 * tasks, report it up the rcu_node hierarchy.
796 * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock.
799 sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp)
804 raw_spin_lock_irqsave(&rnp->lock, flags);
805 if (list_empty(&rnp->blkd_tasks))
806 raw_spin_unlock_irqrestore(&rnp->lock, flags);
808 rnp->exp_tasks = rnp->blkd_tasks.next;
809 rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
813 rcu_report_exp_rnp(rsp, rnp, false); /* Don't wake self. */
817 * synchronize_rcu_expedited - Brute-force RCU grace period
819 * Wait for an RCU-preempt grace period, but expedite it. The basic
820 * idea is to invoke synchronize_sched_expedited() to push all the tasks to
821 * the ->blkd_tasks lists and wait for this list to drain. This consumes
822 * significant time on all CPUs and is unfriendly to real-time workloads,
823 * so is thus not recommended for any sort of common-case code.
824 * In fact, if you are using synchronize_rcu_expedited() in a loop,
825 * please restructure your code to batch your updates, and then Use a
826 * single synchronize_rcu() instead.
828 * Note that it is illegal to call this function while holding any lock
829 * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal
830 * to call this function from a CPU-hotplug notifier. Failing to observe
831 * these restriction will result in deadlock.
833 void synchronize_rcu_expedited(void)
836 struct rcu_node *rnp;
837 struct rcu_state *rsp = &rcu_preempt_state;
841 smp_mb(); /* Caller's modifications seen first by other CPUs. */
842 snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1;
843 smp_mb(); /* Above access cannot bleed into critical section. */
846 * Acquire lock, falling back to synchronize_rcu() if too many
847 * lock-acquisition failures. Of course, if someone does the
848 * expedited grace period for us, just leave.
850 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) {
852 udelay(trycount * num_online_cpus());
857 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
858 goto mb_ret; /* Others did our work for us. */
860 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
861 goto unlock_mb_ret; /* Others did our work for us. */
863 /* force all RCU readers onto ->blkd_tasks lists. */
864 synchronize_sched_expedited();
866 raw_spin_lock_irqsave(&rsp->onofflock, flags);
868 /* Initialize ->expmask for all non-leaf rcu_node structures. */
869 rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) {
870 raw_spin_lock(&rnp->lock); /* irqs already disabled. */
871 rnp->expmask = rnp->qsmaskinit;
872 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
875 /* Snapshot current state of ->blkd_tasks lists. */
876 rcu_for_each_leaf_node(rsp, rnp)
877 sync_rcu_preempt_exp_init(rsp, rnp);
878 if (NUM_RCU_NODES > 1)
879 sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp));
881 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
883 /* Wait for snapshotted ->blkd_tasks lists to drain. */
884 rnp = rcu_get_root(rsp);
885 wait_event(sync_rcu_preempt_exp_wq,
886 sync_rcu_preempt_exp_done(rnp));
888 /* Clean up and exit. */
889 smp_mb(); /* ensure expedited GP seen before counter increment. */
890 ACCESS_ONCE(sync_rcu_preempt_exp_count)++;
892 mutex_unlock(&sync_rcu_preempt_exp_mutex);
894 smp_mb(); /* ensure subsequent action seen after grace period. */
896 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
899 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
901 void rcu_barrier(void)
903 _rcu_barrier(&rcu_preempt_state);
905 EXPORT_SYMBOL_GPL(rcu_barrier);
908 * Initialize preemptible RCU's state structures.
910 static void __init __rcu_init_preempt(void)
912 rcu_init_one(&rcu_preempt_state, &rcu_preempt_data);
915 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
917 static struct rcu_state *rcu_state = &rcu_sched_state;
920 * Tell them what RCU they are running.
922 static void __init rcu_bootup_announce(void)
924 printk(KERN_INFO "Hierarchical RCU implementation.\n");
925 rcu_bootup_announce_oddness();
929 * Return the number of RCU batches processed thus far for debug & stats.
931 long rcu_batches_completed(void)
933 return rcu_batches_completed_sched();
935 EXPORT_SYMBOL_GPL(rcu_batches_completed);
938 * Force a quiescent state for RCU, which, because there is no preemptible
939 * RCU, becomes the same as rcu-sched.
941 void rcu_force_quiescent_state(void)
943 rcu_sched_force_quiescent_state();
945 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
948 * Because preemptible RCU does not exist, we never have to check for
949 * CPUs being in quiescent states.
951 static void rcu_preempt_note_context_switch(int cpu)
956 * Because preemptible RCU does not exist, there are never any preempted
959 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp)
964 #ifdef CONFIG_HOTPLUG_CPU
966 /* Because preemptible RCU does not exist, no quieting of tasks. */
967 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
969 raw_spin_unlock_irqrestore(&rnp->lock, flags);
972 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
975 * Because preemptible RCU does not exist, we never have to check for
976 * tasks blocked within RCU read-side critical sections.
978 static void rcu_print_detail_task_stall(struct rcu_state *rsp)
983 * Because preemptible RCU does not exist, we never have to check for
984 * tasks blocked within RCU read-side critical sections.
986 static int rcu_print_task_stall(struct rcu_node *rnp)
992 * Because there is no preemptible RCU, there can be no readers blocked,
993 * so there is no need to check for blocked tasks. So check only for
994 * bogus qsmask values.
996 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
998 WARN_ON_ONCE(rnp->qsmask);
1001 #ifdef CONFIG_HOTPLUG_CPU
1004 * Because preemptible RCU does not exist, it never needs to migrate
1005 * tasks that were blocked within RCU read-side critical sections, and
1006 * such non-existent tasks cannot possibly have been blocking the current
1009 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
1010 struct rcu_node *rnp,
1011 struct rcu_data *rdp)
1016 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1019 * Because preemptible RCU does not exist, it never has any callbacks
1022 static void rcu_preempt_check_callbacks(int cpu)
1027 * Queue an RCU callback for lazy invocation after a grace period.
1028 * This will likely be later named something like "call_rcu_lazy()",
1029 * but this change will require some way of tagging the lazy RCU
1030 * callbacks in the list of pending callbacks. Until then, this
1031 * function may only be called from __kfree_rcu().
1033 * Because there is no preemptible RCU, we use RCU-sched instead.
1035 void kfree_call_rcu(struct rcu_head *head,
1036 void (*func)(struct rcu_head *rcu))
1038 __call_rcu(head, func, &rcu_sched_state, 1);
1040 EXPORT_SYMBOL_GPL(kfree_call_rcu);
1043 * Wait for an rcu-preempt grace period, but make it happen quickly.
1044 * But because preemptible RCU does not exist, map to rcu-sched.
1046 void synchronize_rcu_expedited(void)
1048 synchronize_sched_expedited();
1050 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
1052 #ifdef CONFIG_HOTPLUG_CPU
1055 * Because preemptible RCU does not exist, there is never any need to
1056 * report on tasks preempted in RCU read-side critical sections during
1057 * expedited RCU grace periods.
1059 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
1064 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1067 * Because preemptible RCU does not exist, rcu_barrier() is just
1068 * another name for rcu_barrier_sched().
1070 void rcu_barrier(void)
1072 rcu_barrier_sched();
1074 EXPORT_SYMBOL_GPL(rcu_barrier);
1077 * Because preemptible RCU does not exist, it need not be initialized.
1079 static void __init __rcu_init_preempt(void)
1083 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1085 #ifdef CONFIG_RCU_BOOST
1087 #include "rtmutex_common.h"
1089 #ifdef CONFIG_RCU_TRACE
1091 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1093 if (list_empty(&rnp->blkd_tasks))
1094 rnp->n_balk_blkd_tasks++;
1095 else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL)
1096 rnp->n_balk_exp_gp_tasks++;
1097 else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL)
1098 rnp->n_balk_boost_tasks++;
1099 else if (rnp->gp_tasks != NULL && rnp->qsmask != 0)
1100 rnp->n_balk_notblocked++;
1101 else if (rnp->gp_tasks != NULL &&
1102 ULONG_CMP_LT(jiffies, rnp->boost_time))
1103 rnp->n_balk_notyet++;
1108 #else /* #ifdef CONFIG_RCU_TRACE */
1110 static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1114 #endif /* #else #ifdef CONFIG_RCU_TRACE */
1117 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1118 * or ->boost_tasks, advancing the pointer to the next task in the
1119 * ->blkd_tasks list.
1121 * Note that irqs must be enabled: boosting the task can block.
1122 * Returns 1 if there are more tasks needing to be boosted.
1124 static int rcu_boost(struct rcu_node *rnp)
1126 unsigned long flags;
1127 struct rt_mutex mtx;
1128 struct task_struct *t;
1129 struct list_head *tb;
1131 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL)
1132 return 0; /* Nothing left to boost. */
1134 raw_spin_lock_irqsave(&rnp->lock, flags);
1137 * Recheck under the lock: all tasks in need of boosting
1138 * might exit their RCU read-side critical sections on their own.
1140 if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) {
1141 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1146 * Preferentially boost tasks blocking expedited grace periods.
1147 * This cannot starve the normal grace periods because a second
1148 * expedited grace period must boost all blocked tasks, including
1149 * those blocking the pre-existing normal grace period.
1151 if (rnp->exp_tasks != NULL) {
1152 tb = rnp->exp_tasks;
1153 rnp->n_exp_boosts++;
1155 tb = rnp->boost_tasks;
1156 rnp->n_normal_boosts++;
1158 rnp->n_tasks_boosted++;
1161 * We boost task t by manufacturing an rt_mutex that appears to
1162 * be held by task t. We leave a pointer to that rt_mutex where
1163 * task t can find it, and task t will release the mutex when it
1164 * exits its outermost RCU read-side critical section. Then
1165 * simply acquiring this artificial rt_mutex will boost task
1166 * t's priority. (Thanks to tglx for suggesting this approach!)
1168 * Note that task t must acquire rnp->lock to remove itself from
1169 * the ->blkd_tasks list, which it will do from exit() if from
1170 * nowhere else. We therefore are guaranteed that task t will
1171 * stay around at least until we drop rnp->lock. Note that
1172 * rnp->lock also resolves races between our priority boosting
1173 * and task t's exiting its outermost RCU read-side critical
1176 t = container_of(tb, struct task_struct, rcu_node_entry);
1177 rt_mutex_init_proxy_locked(&mtx, t);
1178 t->rcu_boost_mutex = &mtx;
1179 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1180 rt_mutex_lock(&mtx); /* Side effect: boosts task t's priority. */
1181 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */
1183 return ACCESS_ONCE(rnp->exp_tasks) != NULL ||
1184 ACCESS_ONCE(rnp->boost_tasks) != NULL;
1188 * Timer handler to initiate waking up of boost kthreads that
1189 * have yielded the CPU due to excessive numbers of tasks to
1190 * boost. We wake up the per-rcu_node kthread, which in turn
1191 * will wake up the booster kthread.
1193 static void rcu_boost_kthread_timer(unsigned long arg)
1195 invoke_rcu_node_kthread((struct rcu_node *)arg);
1199 * Priority-boosting kthread. One per leaf rcu_node and one for the
1202 static int rcu_boost_kthread(void *arg)
1204 struct rcu_node *rnp = (struct rcu_node *)arg;
1208 trace_rcu_utilization("Start boost kthread@init");
1210 rnp->boost_kthread_status = RCU_KTHREAD_WAITING;
1211 trace_rcu_utilization("End boost kthread@rcu_wait");
1212 rcu_wait(rnp->boost_tasks || rnp->exp_tasks);
1213 trace_rcu_utilization("Start boost kthread@rcu_wait");
1214 rnp->boost_kthread_status = RCU_KTHREAD_RUNNING;
1215 more2boost = rcu_boost(rnp);
1221 trace_rcu_utilization("End boost kthread@rcu_yield");
1222 rcu_yield(rcu_boost_kthread_timer, (unsigned long)rnp);
1223 trace_rcu_utilization("Start boost kthread@rcu_yield");
1228 trace_rcu_utilization("End boost kthread@notreached");
1233 * Check to see if it is time to start boosting RCU readers that are
1234 * blocking the current grace period, and, if so, tell the per-rcu_node
1235 * kthread to start boosting them. If there is an expedited grace
1236 * period in progress, it is always time to boost.
1238 * The caller must hold rnp->lock, which this function releases,
1239 * but irqs remain disabled. The ->boost_kthread_task is immortal,
1240 * so we don't need to worry about it going away.
1242 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1244 struct task_struct *t;
1246 if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
1247 rnp->n_balk_exp_gp_tasks++;
1248 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1251 if (rnp->exp_tasks != NULL ||
1252 (rnp->gp_tasks != NULL &&
1253 rnp->boost_tasks == NULL &&
1255 ULONG_CMP_GE(jiffies, rnp->boost_time))) {
1256 if (rnp->exp_tasks == NULL)
1257 rnp->boost_tasks = rnp->gp_tasks;
1258 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1259 t = rnp->boost_kthread_task;
1263 rcu_initiate_boost_trace(rnp);
1264 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1269 * Wake up the per-CPU kthread to invoke RCU callbacks.
1271 static void invoke_rcu_callbacks_kthread(void)
1273 unsigned long flags;
1275 local_irq_save(flags);
1276 __this_cpu_write(rcu_cpu_has_work, 1);
1277 if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
1278 current != __this_cpu_read(rcu_cpu_kthread_task))
1279 wake_up_process(__this_cpu_read(rcu_cpu_kthread_task));
1280 local_irq_restore(flags);
1284 * Is the current CPU running the RCU-callbacks kthread?
1285 * Caller must have preemption disabled.
1287 static bool rcu_is_callbacks_kthread(void)
1289 return __get_cpu_var(rcu_cpu_kthread_task) == current;
1293 * Set the affinity of the boost kthread. The CPU-hotplug locks are
1294 * held, so no one should be messing with the existence of the boost
1297 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
1300 struct task_struct *t;
1302 t = rnp->boost_kthread_task;
1304 set_cpus_allowed_ptr(rnp->boost_kthread_task, cm);
1307 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1310 * Do priority-boost accounting for the start of a new grace period.
1312 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1314 rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
1318 * Create an RCU-boost kthread for the specified node if one does not
1319 * already exist. We only create this kthread for preemptible RCU.
1320 * Returns zero if all is well, a negated errno otherwise.
1322 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
1323 struct rcu_node *rnp,
1326 unsigned long flags;
1327 struct sched_param sp;
1328 struct task_struct *t;
1330 if (&rcu_preempt_state != rsp)
1333 if (rnp->boost_kthread_task != NULL)
1335 t = kthread_create(rcu_boost_kthread, (void *)rnp,
1336 "rcub/%d", rnp_index);
1339 raw_spin_lock_irqsave(&rnp->lock, flags);
1340 rnp->boost_kthread_task = t;
1341 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1342 sp.sched_priority = RCU_BOOST_PRIO;
1343 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1344 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1348 #ifdef CONFIG_HOTPLUG_CPU
1351 * Stop the RCU's per-CPU kthread when its CPU goes offline,.
1353 static void rcu_stop_cpu_kthread(int cpu)
1355 struct task_struct *t;
1357 /* Stop the CPU's kthread. */
1358 t = per_cpu(rcu_cpu_kthread_task, cpu);
1360 per_cpu(rcu_cpu_kthread_task, cpu) = NULL;
1365 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1367 static void rcu_kthread_do_work(void)
1369 rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data));
1370 rcu_do_batch(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1371 rcu_preempt_do_callbacks();
1375 * Wake up the specified per-rcu_node-structure kthread.
1376 * Because the per-rcu_node kthreads are immortal, we don't need
1377 * to do anything to keep them alive.
1379 static void invoke_rcu_node_kthread(struct rcu_node *rnp)
1381 struct task_struct *t;
1383 t = rnp->node_kthread_task;
1389 * Set the specified CPU's kthread to run RT or not, as specified by
1390 * the to_rt argument. The CPU-hotplug locks are held, so the task
1391 * is not going away.
1393 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1396 struct sched_param sp;
1397 struct task_struct *t;
1399 t = per_cpu(rcu_cpu_kthread_task, cpu);
1403 policy = SCHED_FIFO;
1404 sp.sched_priority = RCU_KTHREAD_PRIO;
1406 policy = SCHED_NORMAL;
1407 sp.sched_priority = 0;
1409 sched_setscheduler_nocheck(t, policy, &sp);
1413 * Timer handler to initiate the waking up of per-CPU kthreads that
1414 * have yielded the CPU due to excess numbers of RCU callbacks.
1415 * We wake up the per-rcu_node kthread, which in turn will wake up
1416 * the booster kthread.
1418 static void rcu_cpu_kthread_timer(unsigned long arg)
1420 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg);
1421 struct rcu_node *rnp = rdp->mynode;
1423 atomic_or(rdp->grpmask, &rnp->wakemask);
1424 invoke_rcu_node_kthread(rnp);
1428 * Drop to non-real-time priority and yield, but only after posting a
1429 * timer that will cause us to regain our real-time priority if we
1430 * remain preempted. Either way, we restore our real-time priority
1433 static void rcu_yield(void (*f)(unsigned long), unsigned long arg)
1435 struct sched_param sp;
1436 struct timer_list yield_timer;
1437 int prio = current->rt_priority;
1439 setup_timer_on_stack(&yield_timer, f, arg);
1440 mod_timer(&yield_timer, jiffies + 2);
1441 sp.sched_priority = 0;
1442 sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
1443 set_user_nice(current, 19);
1445 set_user_nice(current, 0);
1446 sp.sched_priority = prio;
1447 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1448 del_timer(&yield_timer);
1452 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1453 * This can happen while the corresponding CPU is either coming online
1454 * or going offline. We cannot wait until the CPU is fully online
1455 * before starting the kthread, because the various notifier functions
1456 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1457 * the corresponding CPU is online.
1459 * Return 1 if the kthread needs to stop, 0 otherwise.
1461 * Caller must disable bh. This function can momentarily enable it.
1463 static int rcu_cpu_kthread_should_stop(int cpu)
1465 while (cpu_is_offline(cpu) ||
1466 !cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)) ||
1467 smp_processor_id() != cpu) {
1468 if (kthread_should_stop())
1470 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
1471 per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id();
1473 schedule_timeout_uninterruptible(1);
1474 if (!cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)))
1475 set_cpus_allowed_ptr(current, cpumask_of(cpu));
1478 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1483 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1484 * RCU softirq used in flavors and configurations of RCU that do not
1485 * support RCU priority boosting.
1487 static int rcu_cpu_kthread(void *arg)
1489 int cpu = (int)(long)arg;
1490 unsigned long flags;
1492 unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu);
1494 char *workp = &per_cpu(rcu_cpu_has_work, cpu);
1496 trace_rcu_utilization("Start CPU kthread@init");
1498 *statusp = RCU_KTHREAD_WAITING;
1499 trace_rcu_utilization("End CPU kthread@rcu_wait");
1500 rcu_wait(*workp != 0 || kthread_should_stop());
1501 trace_rcu_utilization("Start CPU kthread@rcu_wait");
1503 if (rcu_cpu_kthread_should_stop(cpu)) {
1507 *statusp = RCU_KTHREAD_RUNNING;
1508 per_cpu(rcu_cpu_kthread_loops, cpu)++;
1509 local_irq_save(flags);
1512 local_irq_restore(flags);
1514 rcu_kthread_do_work();
1521 *statusp = RCU_KTHREAD_YIELDING;
1522 trace_rcu_utilization("End CPU kthread@rcu_yield");
1523 rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu);
1524 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1528 *statusp = RCU_KTHREAD_STOPPED;
1529 trace_rcu_utilization("End CPU kthread@term");
1534 * Spawn a per-CPU kthread, setting up affinity and priority.
1535 * Because the CPU hotplug lock is held, no other CPU will be attempting
1536 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1537 * attempting to access it during boot, but the locking in kthread_bind()
1538 * will enforce sufficient ordering.
1540 * Please note that we cannot simply refuse to wake up the per-CPU
1541 * kthread because kthreads are created in TASK_UNINTERRUPTIBLE state,
1542 * which can result in softlockup complaints if the task ends up being
1543 * idle for more than a couple of minutes.
1545 * However, please note also that we cannot bind the per-CPU kthread to its
1546 * CPU until that CPU is fully online. We also cannot wait until the
1547 * CPU is fully online before we create its per-CPU kthread, as this would
1548 * deadlock the system when CPU notifiers tried waiting for grace
1549 * periods. So we bind the per-CPU kthread to its CPU only if the CPU
1550 * is online. If its CPU is not yet fully online, then the code in
1551 * rcu_cpu_kthread() will wait until it is fully online, and then do
1554 static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu)
1556 struct sched_param sp;
1557 struct task_struct *t;
1559 if (!rcu_scheduler_fully_active ||
1560 per_cpu(rcu_cpu_kthread_task, cpu) != NULL)
1562 t = kthread_create_on_node(rcu_cpu_kthread,
1568 if (cpu_online(cpu))
1569 kthread_bind(t, cpu);
1570 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1571 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL);
1572 sp.sched_priority = RCU_KTHREAD_PRIO;
1573 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1574 per_cpu(rcu_cpu_kthread_task, cpu) = t;
1575 wake_up_process(t); /* Get to TASK_INTERRUPTIBLE quickly. */
1580 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1581 * kthreads when needed. We ignore requests to wake up kthreads
1582 * for offline CPUs, which is OK because force_quiescent_state()
1583 * takes care of this case.
1585 static int rcu_node_kthread(void *arg)
1588 unsigned long flags;
1590 struct rcu_node *rnp = (struct rcu_node *)arg;
1591 struct sched_param sp;
1592 struct task_struct *t;
1595 rnp->node_kthread_status = RCU_KTHREAD_WAITING;
1596 rcu_wait(atomic_read(&rnp->wakemask) != 0);
1597 rnp->node_kthread_status = RCU_KTHREAD_RUNNING;
1598 raw_spin_lock_irqsave(&rnp->lock, flags);
1599 mask = atomic_xchg(&rnp->wakemask, 0);
1600 rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
1601 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) {
1602 if ((mask & 0x1) == 0)
1605 t = per_cpu(rcu_cpu_kthread_task, cpu);
1606 if (!cpu_online(cpu) || t == NULL) {
1610 per_cpu(rcu_cpu_has_work, cpu) = 1;
1611 sp.sched_priority = RCU_KTHREAD_PRIO;
1612 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1617 rnp->node_kthread_status = RCU_KTHREAD_STOPPED;
1622 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1623 * served by the rcu_node in question. The CPU hotplug lock is still
1624 * held, so the value of rnp->qsmaskinit will be stable.
1626 * We don't include outgoingcpu in the affinity set, use -1 if there is
1627 * no outgoing CPU. If there are no CPUs left in the affinity set,
1628 * this function allows the kthread to execute on any CPU.
1630 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1634 unsigned long mask = rnp->qsmaskinit;
1636 if (rnp->node_kthread_task == NULL)
1638 if (!alloc_cpumask_var(&cm, GFP_KERNEL))
1641 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1)
1642 if ((mask & 0x1) && cpu != outgoingcpu)
1643 cpumask_set_cpu(cpu, cm);
1644 if (cpumask_weight(cm) == 0) {
1646 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++)
1647 cpumask_clear_cpu(cpu, cm);
1648 WARN_ON_ONCE(cpumask_weight(cm) == 0);
1650 set_cpus_allowed_ptr(rnp->node_kthread_task, cm);
1651 rcu_boost_kthread_setaffinity(rnp, cm);
1652 free_cpumask_var(cm);
1656 * Spawn a per-rcu_node kthread, setting priority and affinity.
1657 * Called during boot before online/offline can happen, or, if
1658 * during runtime, with the main CPU-hotplug locks held. So only
1659 * one of these can be executing at a time.
1661 static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp,
1662 struct rcu_node *rnp)
1664 unsigned long flags;
1665 int rnp_index = rnp - &rsp->node[0];
1666 struct sched_param sp;
1667 struct task_struct *t;
1669 if (!rcu_scheduler_fully_active ||
1670 rnp->qsmaskinit == 0)
1672 if (rnp->node_kthread_task == NULL) {
1673 t = kthread_create(rcu_node_kthread, (void *)rnp,
1674 "rcun/%d", rnp_index);
1677 raw_spin_lock_irqsave(&rnp->lock, flags);
1678 rnp->node_kthread_task = t;
1679 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1680 sp.sched_priority = 99;
1681 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1682 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1684 return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index);
1688 * Spawn all kthreads -- called as soon as the scheduler is running.
1690 static int __init rcu_spawn_kthreads(void)
1693 struct rcu_node *rnp;
1695 rcu_scheduler_fully_active = 1;
1696 for_each_possible_cpu(cpu) {
1697 per_cpu(rcu_cpu_has_work, cpu) = 0;
1698 if (cpu_online(cpu))
1699 (void)rcu_spawn_one_cpu_kthread(cpu);
1701 rnp = rcu_get_root(rcu_state);
1702 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1703 if (NUM_RCU_NODES > 1) {
1704 rcu_for_each_leaf_node(rcu_state, rnp)
1705 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1709 early_initcall(rcu_spawn_kthreads);
1711 static void __cpuinit rcu_prepare_kthreads(int cpu)
1713 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu);
1714 struct rcu_node *rnp = rdp->mynode;
1716 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1717 if (rcu_scheduler_fully_active) {
1718 (void)rcu_spawn_one_cpu_kthread(cpu);
1719 if (rnp->node_kthread_task == NULL)
1720 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1724 #else /* #ifdef CONFIG_RCU_BOOST */
1726 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1728 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1731 static void invoke_rcu_callbacks_kthread(void)
1736 static bool rcu_is_callbacks_kthread(void)
1741 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1745 #ifdef CONFIG_HOTPLUG_CPU
1747 static void rcu_stop_cpu_kthread(int cpu)
1751 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1753 static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1757 static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1761 static int __init rcu_scheduler_really_started(void)
1763 rcu_scheduler_fully_active = 1;
1766 early_initcall(rcu_scheduler_really_started);
1768 static void __cpuinit rcu_prepare_kthreads(int cpu)
1772 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1774 #if !defined(CONFIG_RCU_FAST_NO_HZ)
1777 * Check to see if any future RCU-related work will need to be done
1778 * by the current CPU, even if none need be done immediately, returning
1779 * 1 if so. This function is part of the RCU implementation; it is -not-
1780 * an exported member of the RCU API.
1782 * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
1783 * any flavor of RCU.
1785 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1787 *delta_jiffies = ULONG_MAX;
1788 return rcu_cpu_has_callbacks(cpu);
1792 * Because we do not have RCU_FAST_NO_HZ, don't bother initializing for it.
1794 static void rcu_prepare_for_idle_init(int cpu)
1799 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1802 static void rcu_cleanup_after_idle(int cpu)
1807 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
1810 static void rcu_prepare_for_idle(int cpu)
1815 * Don't bother keeping a running count of the number of RCU callbacks
1816 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1818 static void rcu_idle_count_callbacks_posted(void)
1822 #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1825 * This code is invoked when a CPU goes idle, at which point we want
1826 * to have the CPU do everything required for RCU so that it can enter
1827 * the energy-efficient dyntick-idle mode. This is handled by a
1828 * state machine implemented by rcu_prepare_for_idle() below.
1830 * The following three proprocessor symbols control this state machine:
1832 * RCU_IDLE_FLUSHES gives the maximum number of times that we will attempt
1833 * to satisfy RCU. Beyond this point, it is better to incur a periodic
1834 * scheduling-clock interrupt than to loop through the state machine
1836 * RCU_IDLE_OPT_FLUSHES gives the number of RCU_IDLE_FLUSHES that are
1837 * optional if RCU does not need anything immediately from this
1838 * CPU, even if this CPU still has RCU callbacks queued. The first
1839 * times through the state machine are mandatory: we need to give
1840 * the state machine a chance to communicate a quiescent state
1842 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1843 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1844 * is sized to be roughly one RCU grace period. Those energy-efficiency
1845 * benchmarkers who might otherwise be tempted to set this to a large
1846 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1847 * system. And if you are -that- concerned about energy efficiency,
1848 * just power the system down and be done with it!
1849 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1850 * permitted to sleep in dyntick-idle mode with only lazy RCU
1851 * callbacks pending. Setting this too high can OOM your system.
1853 * The values below work well in practice. If future workloads require
1854 * adjustment, they can be converted into kernel config parameters, though
1855 * making the state machine smarter might be a better option.
1857 #define RCU_IDLE_FLUSHES 5 /* Number of dyntick-idle tries. */
1858 #define RCU_IDLE_OPT_FLUSHES 3 /* Optional dyntick-idle tries. */
1859 #define RCU_IDLE_GP_DELAY 6 /* Roughly one grace period. */
1860 #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
1863 * Does the specified flavor of RCU have non-lazy callbacks pending on
1864 * the specified CPU? Both RCU flavor and CPU are specified by the
1865 * rcu_data structure.
1867 static bool __rcu_cpu_has_nonlazy_callbacks(struct rcu_data *rdp)
1869 return rdp->qlen != rdp->qlen_lazy;
1872 #ifdef CONFIG_TREE_PREEMPT_RCU
1875 * Are there non-lazy RCU-preempt callbacks? (There cannot be if there
1876 * is no RCU-preempt in the kernel.)
1878 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1880 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
1882 return __rcu_cpu_has_nonlazy_callbacks(rdp);
1885 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1887 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu)
1892 #endif /* else #ifdef CONFIG_TREE_PREEMPT_RCU */
1895 * Does any flavor of RCU have non-lazy callbacks on the specified CPU?
1897 static bool rcu_cpu_has_nonlazy_callbacks(int cpu)
1899 return __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_sched_data, cpu)) ||
1900 __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_bh_data, cpu)) ||
1901 rcu_preempt_cpu_has_nonlazy_callbacks(cpu);
1905 * Allow the CPU to enter dyntick-idle mode if either: (1) There are no
1906 * callbacks on this CPU, (2) this CPU has not yet attempted to enter
1907 * dyntick-idle mode, or (3) this CPU is in the process of attempting to
1908 * enter dyntick-idle mode. Otherwise, if we have recently tried and failed
1909 * to enter dyntick-idle mode, we refuse to try to enter it. After all,
1910 * it is better to incur scheduling-clock interrupts than to spin
1911 * continuously for the same time duration!
1913 * The delta_jiffies argument is used to store the time when RCU is
1914 * going to need the CPU again if it still has callbacks. The reason
1915 * for this is that rcu_prepare_for_idle() might need to post a timer,
1916 * but if so, it will do so after tick_nohz_stop_sched_tick() has set
1917 * the wakeup time for this CPU. This means that RCU's timer can be
1918 * delayed until the wakeup time, which defeats the purpose of posting
1921 int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1923 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1925 /* Flag a new idle sojourn to the idle-entry state machine. */
1926 rdtp->idle_first_pass = 1;
1927 /* If no callbacks, RCU doesn't need the CPU. */
1928 if (!rcu_cpu_has_callbacks(cpu)) {
1929 *delta_jiffies = ULONG_MAX;
1932 if (rdtp->dyntick_holdoff == jiffies) {
1933 /* RCU recently tried and failed, so don't try again. */
1937 /* Set up for the possibility that RCU will post a timer. */
1938 if (rcu_cpu_has_nonlazy_callbacks(cpu))
1939 *delta_jiffies = RCU_IDLE_GP_DELAY;
1941 *delta_jiffies = RCU_IDLE_LAZY_GP_DELAY;
1946 * Handler for smp_call_function_single(). The only point of this
1947 * handler is to wake the CPU up, so the handler does only tracing.
1949 void rcu_idle_demigrate(void *unused)
1951 trace_rcu_prep_idle("Demigrate");
1955 * Timer handler used to force CPU to start pushing its remaining RCU
1956 * callbacks in the case where it entered dyntick-idle mode with callbacks
1957 * pending. The hander doesn't really need to do anything because the
1958 * real work is done upon re-entry to idle, or by the next scheduling-clock
1959 * interrupt should idle not be re-entered.
1961 * One special case: the timer gets migrated without awakening the CPU
1962 * on which the timer was scheduled on. In this case, we must wake up
1963 * that CPU. We do so with smp_call_function_single().
1965 static void rcu_idle_gp_timer_func(unsigned long cpu_in)
1967 int cpu = (int)cpu_in;
1969 trace_rcu_prep_idle("Timer");
1970 if (cpu != smp_processor_id())
1971 smp_call_function_single(cpu, rcu_idle_demigrate, NULL, 0);
1973 WARN_ON_ONCE(1); /* Getting here can hang the system... */
1977 * Initialize the timer used to pull CPUs out of dyntick-idle mode.
1979 static void rcu_prepare_for_idle_init(int cpu)
1981 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1983 rdtp->dyntick_holdoff = jiffies - 1;
1984 setup_timer(&rdtp->idle_gp_timer, rcu_idle_gp_timer_func, cpu);
1985 rdtp->idle_gp_timer_expires = jiffies - 1;
1986 rdtp->idle_first_pass = 1;
1990 * Clean up for exit from idle. Because we are exiting from idle, there
1991 * is no longer any point to ->idle_gp_timer, so cancel it. This will
1992 * do nothing if this timer is not active, so just cancel it unconditionally.
1994 static void rcu_cleanup_after_idle(int cpu)
1996 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1998 del_timer(&rdtp->idle_gp_timer);
1999 trace_rcu_prep_idle("Cleanup after idle");
2003 * Check to see if any RCU-related work can be done by the current CPU,
2004 * and if so, schedule a softirq to get it done. This function is part
2005 * of the RCU implementation; it is -not- an exported member of the RCU API.
2007 * The idea is for the current CPU to clear out all work required by the
2008 * RCU core for the current grace period, so that this CPU can be permitted
2009 * to enter dyntick-idle mode. In some cases, it will need to be awakened
2010 * at the end of the grace period by whatever CPU ends the grace period.
2011 * This allows CPUs to go dyntick-idle more quickly, and to reduce the
2012 * number of wakeups by a modest integer factor.
2014 * Because it is not legal to invoke rcu_process_callbacks() with irqs
2015 * disabled, we do one pass of force_quiescent_state(), then do a
2016 * invoke_rcu_core() to cause rcu_process_callbacks() to be invoked
2017 * later. The ->dyntick_drain field controls the sequencing.
2019 * The caller must have disabled interrupts.
2021 static void rcu_prepare_for_idle(int cpu)
2023 struct timer_list *tp;
2024 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2027 * If this is an idle re-entry, for example, due to use of
2028 * RCU_NONIDLE() or the new idle-loop tracing API within the idle
2029 * loop, then don't take any state-machine actions, unless the
2030 * momentary exit from idle queued additional non-lazy callbacks.
2031 * Instead, repost the ->idle_gp_timer if this CPU has callbacks
2034 if (!rdtp->idle_first_pass &&
2035 (rdtp->nonlazy_posted == rdtp->nonlazy_posted_snap)) {
2036 if (rcu_cpu_has_callbacks(cpu)) {
2037 tp = &rdtp->idle_gp_timer;
2038 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2042 rdtp->idle_first_pass = 0;
2043 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted - 1;
2046 * If there are no callbacks on this CPU, enter dyntick-idle mode.
2047 * Also reset state to avoid prejudicing later attempts.
2049 if (!rcu_cpu_has_callbacks(cpu)) {
2050 rdtp->dyntick_holdoff = jiffies - 1;
2051 rdtp->dyntick_drain = 0;
2052 trace_rcu_prep_idle("No callbacks");
2057 * If in holdoff mode, just return. We will presumably have
2058 * refrained from disabling the scheduling-clock tick.
2060 if (rdtp->dyntick_holdoff == jiffies) {
2061 trace_rcu_prep_idle("In holdoff");
2065 /* Check and update the ->dyntick_drain sequencing. */
2066 if (rdtp->dyntick_drain <= 0) {
2067 /* First time through, initialize the counter. */
2068 rdtp->dyntick_drain = RCU_IDLE_FLUSHES;
2069 } else if (rdtp->dyntick_drain <= RCU_IDLE_OPT_FLUSHES &&
2070 !rcu_pending(cpu) &&
2071 !local_softirq_pending()) {
2072 /* Can we go dyntick-idle despite still having callbacks? */
2073 rdtp->dyntick_drain = 0;
2074 rdtp->dyntick_holdoff = jiffies;
2075 if (rcu_cpu_has_nonlazy_callbacks(cpu)) {
2076 trace_rcu_prep_idle("Dyntick with callbacks");
2077 rdtp->idle_gp_timer_expires =
2078 jiffies + RCU_IDLE_GP_DELAY;
2080 rdtp->idle_gp_timer_expires =
2081 jiffies + RCU_IDLE_LAZY_GP_DELAY;
2082 trace_rcu_prep_idle("Dyntick with lazy callbacks");
2084 tp = &rdtp->idle_gp_timer;
2085 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
2086 rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted;
2087 return; /* Nothing more to do immediately. */
2088 } else if (--(rdtp->dyntick_drain) <= 0) {
2089 /* We have hit the limit, so time to give up. */
2090 rdtp->dyntick_holdoff = jiffies;
2091 trace_rcu_prep_idle("Begin holdoff");
2092 invoke_rcu_core(); /* Force the CPU out of dyntick-idle. */
2097 * Do one step of pushing the remaining RCU callbacks through
2098 * the RCU core state machine.
2100 #ifdef CONFIG_TREE_PREEMPT_RCU
2101 if (per_cpu(rcu_preempt_data, cpu).nxtlist) {
2102 rcu_preempt_qs(cpu);
2103 force_quiescent_state(&rcu_preempt_state, 0);
2105 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2106 if (per_cpu(rcu_sched_data, cpu).nxtlist) {
2108 force_quiescent_state(&rcu_sched_state, 0);
2110 if (per_cpu(rcu_bh_data, cpu).nxtlist) {
2112 force_quiescent_state(&rcu_bh_state, 0);
2116 * If RCU callbacks are still pending, RCU still needs this CPU.
2117 * So try forcing the callbacks through the grace period.
2119 if (rcu_cpu_has_callbacks(cpu)) {
2120 trace_rcu_prep_idle("More callbacks");
2123 trace_rcu_prep_idle("Callbacks drained");
2127 * Keep a running count of the number of non-lazy callbacks posted
2128 * on this CPU. This running counter (which is never decremented) allows
2129 * rcu_prepare_for_idle() to detect when something out of the idle loop
2130 * posts a callback, even if an equal number of callbacks are invoked.
2131 * Of course, callbacks should only be posted from within a trace event
2132 * designed to be called from idle or from within RCU_NONIDLE().
2134 static void rcu_idle_count_callbacks_posted(void)
2136 __this_cpu_add(rcu_dynticks.nonlazy_posted, 1);
2139 #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
2141 #ifdef CONFIG_RCU_CPU_STALL_INFO
2143 #ifdef CONFIG_RCU_FAST_NO_HZ
2145 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2147 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2148 struct timer_list *tltp = &rdtp->idle_gp_timer;
2150 sprintf(cp, "drain=%d %c timer=%lu",
2151 rdtp->dyntick_drain,
2152 rdtp->dyntick_holdoff == jiffies ? 'H' : '.',
2153 timer_pending(tltp) ? tltp->expires - jiffies : -1);
2156 #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
2158 static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2162 #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
2164 /* Initiate the stall-info list. */
2165 static void print_cpu_stall_info_begin(void)
2167 printk(KERN_CONT "\n");
2171 * Print out diagnostic information for the specified stalled CPU.
2173 * If the specified CPU is aware of the current RCU grace period
2174 * (flavor specified by rsp), then print the number of scheduling
2175 * clock interrupts the CPU has taken during the time that it has
2176 * been aware. Otherwise, print the number of RCU grace periods
2177 * that this CPU is ignorant of, for example, "1" if the CPU was
2178 * aware of the previous grace period.
2180 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
2182 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2184 char fast_no_hz[72];
2185 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2186 struct rcu_dynticks *rdtp = rdp->dynticks;
2188 unsigned long ticks_value;
2190 if (rsp->gpnum == rdp->gpnum) {
2191 ticks_title = "ticks this GP";
2192 ticks_value = rdp->ticks_this_gp;
2194 ticks_title = "GPs behind";
2195 ticks_value = rsp->gpnum - rdp->gpnum;
2197 print_cpu_stall_fast_no_hz(fast_no_hz, cpu);
2198 printk(KERN_ERR "\t%d: (%lu %s) idle=%03x/%llx/%d %s\n",
2199 cpu, ticks_value, ticks_title,
2200 atomic_read(&rdtp->dynticks) & 0xfff,
2201 rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting,
2205 /* Terminate the stall-info list. */
2206 static void print_cpu_stall_info_end(void)
2208 printk(KERN_ERR "\t");
2211 /* Zero ->ticks_this_gp for all flavors of RCU. */
2212 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2214 rdp->ticks_this_gp = 0;
2217 /* Increment ->ticks_this_gp for all flavors of RCU. */
2218 static void increment_cpu_stall_ticks(void)
2220 __get_cpu_var(rcu_sched_data).ticks_this_gp++;
2221 __get_cpu_var(rcu_bh_data).ticks_this_gp++;
2222 #ifdef CONFIG_TREE_PREEMPT_RCU
2223 __get_cpu_var(rcu_preempt_data).ticks_this_gp++;
2224 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2227 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
2229 static void print_cpu_stall_info_begin(void)
2231 printk(KERN_CONT " {");
2234 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu)
2236 printk(KERN_CONT " %d", cpu);
2239 static void print_cpu_stall_info_end(void)
2241 printk(KERN_CONT "} ");
2244 static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2248 static void increment_cpu_stall_ticks(void)
2252 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */