2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions that provide either classic
4 * or preemptible semantics.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
20 * Copyright Red Hat, 2009
21 * Copyright IBM Corporation, 2009
23 * Author: Ingo Molnar <mingo@elte.hu>
24 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
27 #include <linux/delay.h>
29 #define RCU_KTHREAD_PRIO 1
31 #ifdef CONFIG_RCU_BOOST
32 #define RCU_BOOST_PRIO CONFIG_RCU_BOOST_PRIO
34 #define RCU_BOOST_PRIO RCU_KTHREAD_PRIO
38 * Check the RCU kernel configuration parameters and print informative
39 * messages about anything out of the ordinary. If you like #ifdef, you
40 * will love this function.
42 static void __init
rcu_bootup_announce_oddness(void)
44 #ifdef CONFIG_RCU_TRACE
45 printk(KERN_INFO
"\tRCU debugfs-based tracing is enabled.\n");
47 #if (defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 64) || (!defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 32)
48 printk(KERN_INFO
"\tCONFIG_RCU_FANOUT set to non-default value of %d\n",
51 #ifdef CONFIG_RCU_FANOUT_EXACT
52 printk(KERN_INFO
"\tHierarchical RCU autobalancing is disabled.\n");
54 #ifdef CONFIG_RCU_FAST_NO_HZ
56 "\tRCU dyntick-idle grace-period acceleration is enabled.\n");
58 #ifdef CONFIG_PROVE_RCU
59 printk(KERN_INFO
"\tRCU lockdep checking is enabled.\n");
61 #ifdef CONFIG_RCU_TORTURE_TEST_RUNNABLE
62 printk(KERN_INFO
"\tRCU torture testing starts during boot.\n");
64 #if defined(CONFIG_TREE_PREEMPT_RCU) && !defined(CONFIG_RCU_CPU_STALL_VERBOSE)
65 printk(KERN_INFO
"\tDump stacks of tasks blocking RCU-preempt GP.\n");
67 #if defined(CONFIG_RCU_CPU_STALL_INFO)
68 printk(KERN_INFO
"\tAdditional per-CPU info printed with stalls.\n");
70 #if NUM_RCU_LVL_4 != 0
71 printk(KERN_INFO
"\tFour-level hierarchy is enabled.\n");
73 if (rcu_fanout_leaf
!= CONFIG_RCU_FANOUT_LEAF
)
74 printk(KERN_INFO
"\tExperimental boot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf
);
75 if (nr_cpu_ids
!= NR_CPUS
)
76 printk(KERN_INFO
"\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%d.\n", NR_CPUS
, nr_cpu_ids
);
79 #ifdef CONFIG_TREE_PREEMPT_RCU
81 struct rcu_state rcu_preempt_state
=
82 RCU_STATE_INITIALIZER(rcu_preempt
, call_rcu
);
83 DEFINE_PER_CPU(struct rcu_data
, rcu_preempt_data
);
84 static struct rcu_state
*rcu_state
= &rcu_preempt_state
;
86 static int rcu_preempted_readers_exp(struct rcu_node
*rnp
);
89 * Tell them what RCU they are running.
91 static void __init
rcu_bootup_announce(void)
93 printk(KERN_INFO
"Preemptible hierarchical RCU implementation.\n");
94 rcu_bootup_announce_oddness();
98 * Return the number of RCU-preempt batches processed thus far
99 * for debug and statistics.
101 long rcu_batches_completed_preempt(void)
103 return rcu_preempt_state
.completed
;
105 EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt
);
108 * Return the number of RCU batches processed thus far for debug & stats.
110 long rcu_batches_completed(void)
112 return rcu_batches_completed_preempt();
114 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
117 * Force a quiescent state for preemptible RCU.
119 void rcu_force_quiescent_state(void)
121 force_quiescent_state(&rcu_preempt_state
, 0);
123 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state
);
126 * Record a preemptible-RCU quiescent state for the specified CPU. Note
127 * that this just means that the task currently running on the CPU is
128 * not in a quiescent state. There might be any number of tasks blocked
129 * while in an RCU read-side critical section.
131 * Unlike the other rcu_*_qs() functions, callers to this function
132 * must disable irqs in order to protect the assignment to
133 * ->rcu_read_unlock_special.
135 static void rcu_preempt_qs(int cpu
)
137 struct rcu_data
*rdp
= &per_cpu(rcu_preempt_data
, cpu
);
139 rdp
->passed_quiesce_gpnum
= rdp
->gpnum
;
141 if (rdp
->passed_quiesce
== 0)
142 trace_rcu_grace_period("rcu_preempt", rdp
->gpnum
, "cpuqs");
143 rdp
->passed_quiesce
= 1;
144 current
->rcu_read_unlock_special
&= ~RCU_READ_UNLOCK_NEED_QS
;
148 * We have entered the scheduler, and the current task might soon be
149 * context-switched away from. If this task is in an RCU read-side
150 * critical section, we will no longer be able to rely on the CPU to
151 * record that fact, so we enqueue the task on the blkd_tasks list.
152 * The task will dequeue itself when it exits the outermost enclosing
153 * RCU read-side critical section. Therefore, the current grace period
154 * cannot be permitted to complete until the blkd_tasks list entries
155 * predating the current grace period drain, in other words, until
156 * rnp->gp_tasks becomes NULL.
158 * Caller must disable preemption.
160 static void rcu_preempt_note_context_switch(int cpu
)
162 struct task_struct
*t
= current
;
164 struct rcu_data
*rdp
;
165 struct rcu_node
*rnp
;
167 if (t
->rcu_read_lock_nesting
> 0 &&
168 (t
->rcu_read_unlock_special
& RCU_READ_UNLOCK_BLOCKED
) == 0) {
170 /* Possibly blocking in an RCU read-side critical section. */
171 rdp
= per_cpu_ptr(rcu_preempt_state
.rda
, cpu
);
173 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
174 t
->rcu_read_unlock_special
|= RCU_READ_UNLOCK_BLOCKED
;
175 t
->rcu_blocked_node
= rnp
;
178 * If this CPU has already checked in, then this task
179 * will hold up the next grace period rather than the
180 * current grace period. Queue the task accordingly.
181 * If the task is queued for the current grace period
182 * (i.e., this CPU has not yet passed through a quiescent
183 * state for the current grace period), then as long
184 * as that task remains queued, the current grace period
185 * cannot end. Note that there is some uncertainty as
186 * to exactly when the current grace period started.
187 * We take a conservative approach, which can result
188 * in unnecessarily waiting on tasks that started very
189 * slightly after the current grace period began. C'est
192 * But first, note that the current CPU must still be
195 WARN_ON_ONCE((rdp
->grpmask
& rnp
->qsmaskinit
) == 0);
196 WARN_ON_ONCE(!list_empty(&t
->rcu_node_entry
));
197 if ((rnp
->qsmask
& rdp
->grpmask
) && rnp
->gp_tasks
!= NULL
) {
198 list_add(&t
->rcu_node_entry
, rnp
->gp_tasks
->prev
);
199 rnp
->gp_tasks
= &t
->rcu_node_entry
;
200 #ifdef CONFIG_RCU_BOOST
201 if (rnp
->boost_tasks
!= NULL
)
202 rnp
->boost_tasks
= rnp
->gp_tasks
;
203 #endif /* #ifdef CONFIG_RCU_BOOST */
205 list_add(&t
->rcu_node_entry
, &rnp
->blkd_tasks
);
206 if (rnp
->qsmask
& rdp
->grpmask
)
207 rnp
->gp_tasks
= &t
->rcu_node_entry
;
209 trace_rcu_preempt_task(rdp
->rsp
->name
,
211 (rnp
->qsmask
& rdp
->grpmask
)
214 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
215 } else if (t
->rcu_read_lock_nesting
< 0 &&
216 t
->rcu_read_unlock_special
) {
219 * Complete exit from RCU read-side critical section on
220 * behalf of preempted instance of __rcu_read_unlock().
222 rcu_read_unlock_special(t
);
226 * Either we were not in an RCU read-side critical section to
227 * begin with, or we have now recorded that critical section
228 * globally. Either way, we can now note a quiescent state
229 * for this CPU. Again, if we were in an RCU read-side critical
230 * section, and if that critical section was blocking the current
231 * grace period, then the fact that the task has been enqueued
232 * means that we continue to block the current grace period.
234 local_irq_save(flags
);
236 local_irq_restore(flags
);
240 * Check for preempted RCU readers blocking the current grace period
241 * for the specified rcu_node structure. If the caller needs a reliable
242 * answer, it must hold the rcu_node's ->lock.
244 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
)
246 return rnp
->gp_tasks
!= NULL
;
250 * Record a quiescent state for all tasks that were previously queued
251 * on the specified rcu_node structure and that were blocking the current
252 * RCU grace period. The caller must hold the specified rnp->lock with
253 * irqs disabled, and this lock is released upon return, but irqs remain
256 static void rcu_report_unblock_qs_rnp(struct rcu_node
*rnp
, unsigned long flags
)
257 __releases(rnp
->lock
)
260 struct rcu_node
*rnp_p
;
262 if (rnp
->qsmask
!= 0 || rcu_preempt_blocked_readers_cgp(rnp
)) {
263 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
264 return; /* Still need more quiescent states! */
270 * Either there is only one rcu_node in the tree,
271 * or tasks were kicked up to root rcu_node due to
272 * CPUs going offline.
274 rcu_report_qs_rsp(&rcu_preempt_state
, flags
);
278 /* Report up the rest of the hierarchy. */
280 raw_spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
281 raw_spin_lock(&rnp_p
->lock
); /* irqs already disabled. */
282 rcu_report_qs_rnp(mask
, &rcu_preempt_state
, rnp_p
, flags
);
286 * Advance a ->blkd_tasks-list pointer to the next entry, instead
287 * returning NULL if at the end of the list.
289 static struct list_head
*rcu_next_node_entry(struct task_struct
*t
,
290 struct rcu_node
*rnp
)
292 struct list_head
*np
;
294 np
= t
->rcu_node_entry
.next
;
295 if (np
== &rnp
->blkd_tasks
)
301 * Handle special cases during rcu_read_unlock(), such as needing to
302 * notify RCU core processing or task having blocked during the RCU
303 * read-side critical section.
305 void rcu_read_unlock_special(struct task_struct
*t
)
311 struct list_head
*np
;
312 #ifdef CONFIG_RCU_BOOST
313 struct rt_mutex
*rbmp
= NULL
;
314 #endif /* #ifdef CONFIG_RCU_BOOST */
315 struct rcu_node
*rnp
;
318 /* NMI handlers cannot block and cannot safely manipulate state. */
322 local_irq_save(flags
);
325 * If RCU core is waiting for this CPU to exit critical section,
326 * let it know that we have done so.
328 special
= t
->rcu_read_unlock_special
;
329 if (special
& RCU_READ_UNLOCK_NEED_QS
) {
330 rcu_preempt_qs(smp_processor_id());
333 /* Hardware IRQ handlers cannot block. */
334 if (in_irq() || in_serving_softirq()) {
335 local_irq_restore(flags
);
339 /* Clean up if blocked during RCU read-side critical section. */
340 if (special
& RCU_READ_UNLOCK_BLOCKED
) {
341 t
->rcu_read_unlock_special
&= ~RCU_READ_UNLOCK_BLOCKED
;
344 * Remove this task from the list it blocked on. The
345 * task can migrate while we acquire the lock, but at
346 * most one time. So at most two passes through loop.
349 rnp
= t
->rcu_blocked_node
;
350 raw_spin_lock(&rnp
->lock
); /* irqs already disabled. */
351 if (rnp
== t
->rcu_blocked_node
)
353 raw_spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
355 empty
= !rcu_preempt_blocked_readers_cgp(rnp
);
356 empty_exp
= !rcu_preempted_readers_exp(rnp
);
357 smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */
358 np
= rcu_next_node_entry(t
, rnp
);
359 list_del_init(&t
->rcu_node_entry
);
360 t
->rcu_blocked_node
= NULL
;
361 trace_rcu_unlock_preempted_task("rcu_preempt",
363 if (&t
->rcu_node_entry
== rnp
->gp_tasks
)
365 if (&t
->rcu_node_entry
== rnp
->exp_tasks
)
367 #ifdef CONFIG_RCU_BOOST
368 if (&t
->rcu_node_entry
== rnp
->boost_tasks
)
369 rnp
->boost_tasks
= np
;
370 /* Snapshot/clear ->rcu_boost_mutex with rcu_node lock held. */
371 if (t
->rcu_boost_mutex
) {
372 rbmp
= t
->rcu_boost_mutex
;
373 t
->rcu_boost_mutex
= NULL
;
375 #endif /* #ifdef CONFIG_RCU_BOOST */
378 * If this was the last task on the current list, and if
379 * we aren't waiting on any CPUs, report the quiescent state.
380 * Note that rcu_report_unblock_qs_rnp() releases rnp->lock,
381 * so we must take a snapshot of the expedited state.
383 empty_exp_now
= !rcu_preempted_readers_exp(rnp
);
384 if (!empty
&& !rcu_preempt_blocked_readers_cgp(rnp
)) {
385 trace_rcu_quiescent_state_report("preempt_rcu",
392 rcu_report_unblock_qs_rnp(rnp
, flags
);
394 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
396 #ifdef CONFIG_RCU_BOOST
397 /* Unboost if we were boosted. */
399 rt_mutex_unlock(rbmp
);
400 #endif /* #ifdef CONFIG_RCU_BOOST */
403 * If this was the last task on the expedited lists,
404 * then we need to report up the rcu_node hierarchy.
406 if (!empty_exp
&& empty_exp_now
)
407 rcu_report_exp_rnp(&rcu_preempt_state
, rnp
, true);
409 local_irq_restore(flags
);
413 #ifdef CONFIG_RCU_CPU_STALL_VERBOSE
416 * Dump detailed information for all tasks blocking the current RCU
417 * grace period on the specified rcu_node structure.
419 static void rcu_print_detail_task_stall_rnp(struct rcu_node
*rnp
)
422 struct task_struct
*t
;
424 if (!rcu_preempt_blocked_readers_cgp(rnp
))
426 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
427 t
= list_entry(rnp
->gp_tasks
,
428 struct task_struct
, rcu_node_entry
);
429 list_for_each_entry_continue(t
, &rnp
->blkd_tasks
, rcu_node_entry
)
431 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
435 * Dump detailed information for all tasks blocking the current RCU
438 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
)
440 struct rcu_node
*rnp
= rcu_get_root(rsp
);
442 rcu_print_detail_task_stall_rnp(rnp
);
443 rcu_for_each_leaf_node(rsp
, rnp
)
444 rcu_print_detail_task_stall_rnp(rnp
);
447 #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
449 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
)
453 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */
455 #ifdef CONFIG_RCU_CPU_STALL_INFO
457 static void rcu_print_task_stall_begin(struct rcu_node
*rnp
)
459 printk(KERN_ERR
"\tTasks blocked on level-%d rcu_node (CPUs %d-%d):",
460 rnp
->level
, rnp
->grplo
, rnp
->grphi
);
463 static void rcu_print_task_stall_end(void)
465 printk(KERN_CONT
"\n");
468 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
470 static void rcu_print_task_stall_begin(struct rcu_node
*rnp
)
474 static void rcu_print_task_stall_end(void)
478 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */
481 * Scan the current list of tasks blocked within RCU read-side critical
482 * sections, printing out the tid of each.
484 static int rcu_print_task_stall(struct rcu_node
*rnp
)
486 struct task_struct
*t
;
489 if (!rcu_preempt_blocked_readers_cgp(rnp
))
491 rcu_print_task_stall_begin(rnp
);
492 t
= list_entry(rnp
->gp_tasks
,
493 struct task_struct
, rcu_node_entry
);
494 list_for_each_entry_continue(t
, &rnp
->blkd_tasks
, rcu_node_entry
) {
495 printk(KERN_CONT
" P%d", t
->pid
);
498 rcu_print_task_stall_end();
503 * Check that the list of blocked tasks for the newly completed grace
504 * period is in fact empty. It is a serious bug to complete a grace
505 * period that still has RCU readers blocked! This function must be
506 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
507 * must be held by the caller.
509 * Also, if there are blocked tasks on the list, they automatically
510 * block the newly created grace period, so set up ->gp_tasks accordingly.
512 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
514 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp
));
515 if (!list_empty(&rnp
->blkd_tasks
))
516 rnp
->gp_tasks
= rnp
->blkd_tasks
.next
;
517 WARN_ON_ONCE(rnp
->qsmask
);
520 #ifdef CONFIG_HOTPLUG_CPU
523 * Handle tasklist migration for case in which all CPUs covered by the
524 * specified rcu_node have gone offline. Move them up to the root
525 * rcu_node. The reason for not just moving them to the immediate
526 * parent is to remove the need for rcu_read_unlock_special() to
527 * make more than two attempts to acquire the target rcu_node's lock.
528 * Returns true if there were tasks blocking the current RCU grace
531 * Returns 1 if there was previously a task blocking the current grace
532 * period on the specified rcu_node structure.
534 * The caller must hold rnp->lock with irqs disabled.
536 static int rcu_preempt_offline_tasks(struct rcu_state
*rsp
,
537 struct rcu_node
*rnp
,
538 struct rcu_data
*rdp
)
540 struct list_head
*lp
;
541 struct list_head
*lp_root
;
543 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
544 struct task_struct
*t
;
546 if (rnp
== rnp_root
) {
547 WARN_ONCE(1, "Last CPU thought to be offlined?");
548 return 0; /* Shouldn't happen: at least one CPU online. */
551 /* If we are on an internal node, complain bitterly. */
552 WARN_ON_ONCE(rnp
!= rdp
->mynode
);
555 * Move tasks up to root rcu_node. Don't try to get fancy for
556 * this corner-case operation -- just put this node's tasks
557 * at the head of the root node's list, and update the root node's
558 * ->gp_tasks and ->exp_tasks pointers to those of this node's,
559 * if non-NULL. This might result in waiting for more tasks than
560 * absolutely necessary, but this is a good performance/complexity
563 if (rcu_preempt_blocked_readers_cgp(rnp
) && rnp
->qsmask
== 0)
564 retval
|= RCU_OFL_TASKS_NORM_GP
;
565 if (rcu_preempted_readers_exp(rnp
))
566 retval
|= RCU_OFL_TASKS_EXP_GP
;
567 lp
= &rnp
->blkd_tasks
;
568 lp_root
= &rnp_root
->blkd_tasks
;
569 while (!list_empty(lp
)) {
570 t
= list_entry(lp
->next
, typeof(*t
), rcu_node_entry
);
571 raw_spin_lock(&rnp_root
->lock
); /* irqs already disabled */
572 list_del(&t
->rcu_node_entry
);
573 t
->rcu_blocked_node
= rnp_root
;
574 list_add(&t
->rcu_node_entry
, lp_root
);
575 if (&t
->rcu_node_entry
== rnp
->gp_tasks
)
576 rnp_root
->gp_tasks
= rnp
->gp_tasks
;
577 if (&t
->rcu_node_entry
== rnp
->exp_tasks
)
578 rnp_root
->exp_tasks
= rnp
->exp_tasks
;
579 #ifdef CONFIG_RCU_BOOST
580 if (&t
->rcu_node_entry
== rnp
->boost_tasks
)
581 rnp_root
->boost_tasks
= rnp
->boost_tasks
;
582 #endif /* #ifdef CONFIG_RCU_BOOST */
583 raw_spin_unlock(&rnp_root
->lock
); /* irqs still disabled */
586 #ifdef CONFIG_RCU_BOOST
587 /* In case root is being boosted and leaf is not. */
588 raw_spin_lock(&rnp_root
->lock
); /* irqs already disabled */
589 if (rnp_root
->boost_tasks
!= NULL
&&
590 rnp_root
->boost_tasks
!= rnp_root
->gp_tasks
)
591 rnp_root
->boost_tasks
= rnp_root
->gp_tasks
;
592 raw_spin_unlock(&rnp_root
->lock
); /* irqs still disabled */
593 #endif /* #ifdef CONFIG_RCU_BOOST */
595 rnp
->gp_tasks
= NULL
;
596 rnp
->exp_tasks
= NULL
;
600 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
603 * Check for a quiescent state from the current CPU. When a task blocks,
604 * the task is recorded in the corresponding CPU's rcu_node structure,
605 * which is checked elsewhere.
607 * Caller must disable hard irqs.
609 static void rcu_preempt_check_callbacks(int cpu
)
611 struct task_struct
*t
= current
;
613 if (t
->rcu_read_lock_nesting
== 0) {
617 if (t
->rcu_read_lock_nesting
> 0 &&
618 per_cpu(rcu_preempt_data
, cpu
).qs_pending
)
619 t
->rcu_read_unlock_special
|= RCU_READ_UNLOCK_NEED_QS
;
622 #ifdef CONFIG_RCU_BOOST
624 static void rcu_preempt_do_callbacks(void)
626 rcu_do_batch(&rcu_preempt_state
, &__get_cpu_var(rcu_preempt_data
));
629 #endif /* #ifdef CONFIG_RCU_BOOST */
632 * Queue a preemptible-RCU callback for invocation after a grace period.
634 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
636 __call_rcu(head
, func
, &rcu_preempt_state
, 0);
638 EXPORT_SYMBOL_GPL(call_rcu
);
641 * Queue an RCU callback for lazy invocation after a grace period.
642 * This will likely be later named something like "call_rcu_lazy()",
643 * but this change will require some way of tagging the lazy RCU
644 * callbacks in the list of pending callbacks. Until then, this
645 * function may only be called from __kfree_rcu().
647 void kfree_call_rcu(struct rcu_head
*head
,
648 void (*func
)(struct rcu_head
*rcu
))
650 __call_rcu(head
, func
, &rcu_preempt_state
, 1);
652 EXPORT_SYMBOL_GPL(kfree_call_rcu
);
655 * synchronize_rcu - wait until a grace period has elapsed.
657 * Control will return to the caller some time after a full grace
658 * period has elapsed, in other words after all currently executing RCU
659 * read-side critical sections have completed. Note, however, that
660 * upon return from synchronize_rcu(), the caller might well be executing
661 * concurrently with new RCU read-side critical sections that began while
662 * synchronize_rcu() was waiting. RCU read-side critical sections are
663 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
665 void synchronize_rcu(void)
667 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map
) &&
668 !lock_is_held(&rcu_lock_map
) &&
669 !lock_is_held(&rcu_sched_lock_map
),
670 "Illegal synchronize_rcu() in RCU read-side critical section");
671 if (!rcu_scheduler_active
)
673 wait_rcu_gp(call_rcu
);
675 EXPORT_SYMBOL_GPL(synchronize_rcu
);
677 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq
);
678 static long sync_rcu_preempt_exp_count
;
679 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex
);
682 * Return non-zero if there are any tasks in RCU read-side critical
683 * sections blocking the current preemptible-RCU expedited grace period.
684 * If there is no preemptible-RCU expedited grace period currently in
685 * progress, returns zero unconditionally.
687 static int rcu_preempted_readers_exp(struct rcu_node
*rnp
)
689 return rnp
->exp_tasks
!= NULL
;
693 * return non-zero if there is no RCU expedited grace period in progress
694 * for the specified rcu_node structure, in other words, if all CPUs and
695 * tasks covered by the specified rcu_node structure have done their bit
696 * for the current expedited grace period. Works only for preemptible
697 * RCU -- other RCU implementation use other means.
699 * Caller must hold sync_rcu_preempt_exp_mutex.
701 static int sync_rcu_preempt_exp_done(struct rcu_node
*rnp
)
703 return !rcu_preempted_readers_exp(rnp
) &&
704 ACCESS_ONCE(rnp
->expmask
) == 0;
708 * Report the exit from RCU read-side critical section for the last task
709 * that queued itself during or before the current expedited preemptible-RCU
710 * grace period. This event is reported either to the rcu_node structure on
711 * which the task was queued or to one of that rcu_node structure's ancestors,
712 * recursively up the tree. (Calm down, calm down, we do the recursion
715 * Most callers will set the "wake" flag, but the task initiating the
716 * expedited grace period need not wake itself.
718 * Caller must hold sync_rcu_preempt_exp_mutex.
720 static void rcu_report_exp_rnp(struct rcu_state
*rsp
, struct rcu_node
*rnp
,
726 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
728 if (!sync_rcu_preempt_exp_done(rnp
)) {
729 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
732 if (rnp
->parent
== NULL
) {
733 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
735 wake_up(&sync_rcu_preempt_exp_wq
);
739 raw_spin_unlock(&rnp
->lock
); /* irqs remain disabled */
741 raw_spin_lock(&rnp
->lock
); /* irqs already disabled */
742 rnp
->expmask
&= ~mask
;
747 * Snapshot the tasks blocking the newly started preemptible-RCU expedited
748 * grace period for the specified rcu_node structure. If there are no such
749 * tasks, report it up the rcu_node hierarchy.
751 * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock.
754 sync_rcu_preempt_exp_init(struct rcu_state
*rsp
, struct rcu_node
*rnp
)
759 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
760 if (list_empty(&rnp
->blkd_tasks
))
761 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
763 rnp
->exp_tasks
= rnp
->blkd_tasks
.next
;
764 rcu_initiate_boost(rnp
, flags
); /* releases rnp->lock */
768 rcu_report_exp_rnp(rsp
, rnp
, false); /* Don't wake self. */
772 * synchronize_rcu_expedited - Brute-force RCU grace period
774 * Wait for an RCU-preempt grace period, but expedite it. The basic
775 * idea is to invoke synchronize_sched_expedited() to push all the tasks to
776 * the ->blkd_tasks lists and wait for this list to drain. This consumes
777 * significant time on all CPUs and is unfriendly to real-time workloads,
778 * so is thus not recommended for any sort of common-case code.
779 * In fact, if you are using synchronize_rcu_expedited() in a loop,
780 * please restructure your code to batch your updates, and then Use a
781 * single synchronize_rcu() instead.
783 * Note that it is illegal to call this function while holding any lock
784 * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal
785 * to call this function from a CPU-hotplug notifier. Failing to observe
786 * these restriction will result in deadlock.
788 void synchronize_rcu_expedited(void)
791 struct rcu_node
*rnp
;
792 struct rcu_state
*rsp
= &rcu_preempt_state
;
796 smp_mb(); /* Caller's modifications seen first by other CPUs. */
797 snap
= ACCESS_ONCE(sync_rcu_preempt_exp_count
) + 1;
798 smp_mb(); /* Above access cannot bleed into critical section. */
801 * Acquire lock, falling back to synchronize_rcu() if too many
802 * lock-acquisition failures. Of course, if someone does the
803 * expedited grace period for us, just leave.
805 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex
)) {
807 udelay(trycount
* num_online_cpus());
812 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count
) - snap
) > 0)
813 goto mb_ret
; /* Others did our work for us. */
815 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count
) - snap
) > 0)
816 goto unlock_mb_ret
; /* Others did our work for us. */
818 /* force all RCU readers onto ->blkd_tasks lists. */
819 synchronize_sched_expedited();
821 raw_spin_lock_irqsave(&rsp
->onofflock
, flags
);
823 /* Initialize ->expmask for all non-leaf rcu_node structures. */
824 rcu_for_each_nonleaf_node_breadth_first(rsp
, rnp
) {
825 raw_spin_lock(&rnp
->lock
); /* irqs already disabled. */
826 rnp
->expmask
= rnp
->qsmaskinit
;
827 raw_spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
830 /* Snapshot current state of ->blkd_tasks lists. */
831 rcu_for_each_leaf_node(rsp
, rnp
)
832 sync_rcu_preempt_exp_init(rsp
, rnp
);
833 if (NUM_RCU_NODES
> 1)
834 sync_rcu_preempt_exp_init(rsp
, rcu_get_root(rsp
));
836 raw_spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
838 /* Wait for snapshotted ->blkd_tasks lists to drain. */
839 rnp
= rcu_get_root(rsp
);
840 wait_event(sync_rcu_preempt_exp_wq
,
841 sync_rcu_preempt_exp_done(rnp
));
843 /* Clean up and exit. */
844 smp_mb(); /* ensure expedited GP seen before counter increment. */
845 ACCESS_ONCE(sync_rcu_preempt_exp_count
)++;
847 mutex_unlock(&sync_rcu_preempt_exp_mutex
);
849 smp_mb(); /* ensure subsequent action seen after grace period. */
851 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited
);
854 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
856 void rcu_barrier(void)
858 _rcu_barrier(&rcu_preempt_state
);
860 EXPORT_SYMBOL_GPL(rcu_barrier
);
863 * Initialize preemptible RCU's state structures.
865 static void __init
__rcu_init_preempt(void)
867 rcu_init_one(&rcu_preempt_state
, &rcu_preempt_data
);
870 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
872 static struct rcu_state
*rcu_state
= &rcu_sched_state
;
875 * Tell them what RCU they are running.
877 static void __init
rcu_bootup_announce(void)
879 printk(KERN_INFO
"Hierarchical RCU implementation.\n");
880 rcu_bootup_announce_oddness();
884 * Return the number of RCU batches processed thus far for debug & stats.
886 long rcu_batches_completed(void)
888 return rcu_batches_completed_sched();
890 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
893 * Force a quiescent state for RCU, which, because there is no preemptible
894 * RCU, becomes the same as rcu-sched.
896 void rcu_force_quiescent_state(void)
898 rcu_sched_force_quiescent_state();
900 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state
);
903 * Because preemptible RCU does not exist, we never have to check for
904 * CPUs being in quiescent states.
906 static void rcu_preempt_note_context_switch(int cpu
)
911 * Because preemptible RCU does not exist, there are never any preempted
914 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
)
919 #ifdef CONFIG_HOTPLUG_CPU
921 /* Because preemptible RCU does not exist, no quieting of tasks. */
922 static void rcu_report_unblock_qs_rnp(struct rcu_node
*rnp
, unsigned long flags
)
924 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
927 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
930 * Because preemptible RCU does not exist, we never have to check for
931 * tasks blocked within RCU read-side critical sections.
933 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
)
938 * Because preemptible RCU does not exist, we never have to check for
939 * tasks blocked within RCU read-side critical sections.
941 static int rcu_print_task_stall(struct rcu_node
*rnp
)
947 * Because there is no preemptible RCU, there can be no readers blocked,
948 * so there is no need to check for blocked tasks. So check only for
949 * bogus qsmask values.
951 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
953 WARN_ON_ONCE(rnp
->qsmask
);
956 #ifdef CONFIG_HOTPLUG_CPU
959 * Because preemptible RCU does not exist, it never needs to migrate
960 * tasks that were blocked within RCU read-side critical sections, and
961 * such non-existent tasks cannot possibly have been blocking the current
964 static int rcu_preempt_offline_tasks(struct rcu_state
*rsp
,
965 struct rcu_node
*rnp
,
966 struct rcu_data
*rdp
)
971 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
974 * Because preemptible RCU does not exist, it never has any callbacks
977 static void rcu_preempt_check_callbacks(int cpu
)
982 * Queue an RCU callback for lazy invocation after a grace period.
983 * This will likely be later named something like "call_rcu_lazy()",
984 * but this change will require some way of tagging the lazy RCU
985 * callbacks in the list of pending callbacks. Until then, this
986 * function may only be called from __kfree_rcu().
988 * Because there is no preemptible RCU, we use RCU-sched instead.
990 void kfree_call_rcu(struct rcu_head
*head
,
991 void (*func
)(struct rcu_head
*rcu
))
993 __call_rcu(head
, func
, &rcu_sched_state
, 1);
995 EXPORT_SYMBOL_GPL(kfree_call_rcu
);
998 * Wait for an rcu-preempt grace period, but make it happen quickly.
999 * But because preemptible RCU does not exist, map to rcu-sched.
1001 void synchronize_rcu_expedited(void)
1003 synchronize_sched_expedited();
1005 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited
);
1007 #ifdef CONFIG_HOTPLUG_CPU
1010 * Because preemptible RCU does not exist, there is never any need to
1011 * report on tasks preempted in RCU read-side critical sections during
1012 * expedited RCU grace periods.
1014 static void rcu_report_exp_rnp(struct rcu_state
*rsp
, struct rcu_node
*rnp
,
1019 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1022 * Because preemptible RCU does not exist, rcu_barrier() is just
1023 * another name for rcu_barrier_sched().
1025 void rcu_barrier(void)
1027 rcu_barrier_sched();
1029 EXPORT_SYMBOL_GPL(rcu_barrier
);
1032 * Because preemptible RCU does not exist, it need not be initialized.
1034 static void __init
__rcu_init_preempt(void)
1038 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1040 #ifdef CONFIG_RCU_BOOST
1042 #include "rtmutex_common.h"
1044 #ifdef CONFIG_RCU_TRACE
1046 static void rcu_initiate_boost_trace(struct rcu_node
*rnp
)
1048 if (list_empty(&rnp
->blkd_tasks
))
1049 rnp
->n_balk_blkd_tasks
++;
1050 else if (rnp
->exp_tasks
== NULL
&& rnp
->gp_tasks
== NULL
)
1051 rnp
->n_balk_exp_gp_tasks
++;
1052 else if (rnp
->gp_tasks
!= NULL
&& rnp
->boost_tasks
!= NULL
)
1053 rnp
->n_balk_boost_tasks
++;
1054 else if (rnp
->gp_tasks
!= NULL
&& rnp
->qsmask
!= 0)
1055 rnp
->n_balk_notblocked
++;
1056 else if (rnp
->gp_tasks
!= NULL
&&
1057 ULONG_CMP_LT(jiffies
, rnp
->boost_time
))
1058 rnp
->n_balk_notyet
++;
1063 #else /* #ifdef CONFIG_RCU_TRACE */
1065 static void rcu_initiate_boost_trace(struct rcu_node
*rnp
)
1069 #endif /* #else #ifdef CONFIG_RCU_TRACE */
1072 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1073 * or ->boost_tasks, advancing the pointer to the next task in the
1074 * ->blkd_tasks list.
1076 * Note that irqs must be enabled: boosting the task can block.
1077 * Returns 1 if there are more tasks needing to be boosted.
1079 static int rcu_boost(struct rcu_node
*rnp
)
1081 unsigned long flags
;
1082 struct rt_mutex mtx
;
1083 struct task_struct
*t
;
1084 struct list_head
*tb
;
1086 if (rnp
->exp_tasks
== NULL
&& rnp
->boost_tasks
== NULL
)
1087 return 0; /* Nothing left to boost. */
1089 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1092 * Recheck under the lock: all tasks in need of boosting
1093 * might exit their RCU read-side critical sections on their own.
1095 if (rnp
->exp_tasks
== NULL
&& rnp
->boost_tasks
== NULL
) {
1096 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1101 * Preferentially boost tasks blocking expedited grace periods.
1102 * This cannot starve the normal grace periods because a second
1103 * expedited grace period must boost all blocked tasks, including
1104 * those blocking the pre-existing normal grace period.
1106 if (rnp
->exp_tasks
!= NULL
) {
1107 tb
= rnp
->exp_tasks
;
1108 rnp
->n_exp_boosts
++;
1110 tb
= rnp
->boost_tasks
;
1111 rnp
->n_normal_boosts
++;
1113 rnp
->n_tasks_boosted
++;
1116 * We boost task t by manufacturing an rt_mutex that appears to
1117 * be held by task t. We leave a pointer to that rt_mutex where
1118 * task t can find it, and task t will release the mutex when it
1119 * exits its outermost RCU read-side critical section. Then
1120 * simply acquiring this artificial rt_mutex will boost task
1121 * t's priority. (Thanks to tglx for suggesting this approach!)
1123 * Note that task t must acquire rnp->lock to remove itself from
1124 * the ->blkd_tasks list, which it will do from exit() if from
1125 * nowhere else. We therefore are guaranteed that task t will
1126 * stay around at least until we drop rnp->lock. Note that
1127 * rnp->lock also resolves races between our priority boosting
1128 * and task t's exiting its outermost RCU read-side critical
1131 t
= container_of(tb
, struct task_struct
, rcu_node_entry
);
1132 rt_mutex_init_proxy_locked(&mtx
, t
);
1133 t
->rcu_boost_mutex
= &mtx
;
1134 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1135 rt_mutex_lock(&mtx
); /* Side effect: boosts task t's priority. */
1136 rt_mutex_unlock(&mtx
); /* Keep lockdep happy. */
1138 return ACCESS_ONCE(rnp
->exp_tasks
) != NULL
||
1139 ACCESS_ONCE(rnp
->boost_tasks
) != NULL
;
1143 * Timer handler to initiate waking up of boost kthreads that
1144 * have yielded the CPU due to excessive numbers of tasks to
1145 * boost. We wake up the per-rcu_node kthread, which in turn
1146 * will wake up the booster kthread.
1148 static void rcu_boost_kthread_timer(unsigned long arg
)
1150 invoke_rcu_node_kthread((struct rcu_node
*)arg
);
1154 * Priority-boosting kthread. One per leaf rcu_node and one for the
1157 static int rcu_boost_kthread(void *arg
)
1159 struct rcu_node
*rnp
= (struct rcu_node
*)arg
;
1163 trace_rcu_utilization("Start boost kthread@init");
1165 rnp
->boost_kthread_status
= RCU_KTHREAD_WAITING
;
1166 trace_rcu_utilization("End boost kthread@rcu_wait");
1167 rcu_wait(rnp
->boost_tasks
|| rnp
->exp_tasks
);
1168 trace_rcu_utilization("Start boost kthread@rcu_wait");
1169 rnp
->boost_kthread_status
= RCU_KTHREAD_RUNNING
;
1170 more2boost
= rcu_boost(rnp
);
1176 trace_rcu_utilization("End boost kthread@rcu_yield");
1177 rcu_yield(rcu_boost_kthread_timer
, (unsigned long)rnp
);
1178 trace_rcu_utilization("Start boost kthread@rcu_yield");
1183 trace_rcu_utilization("End boost kthread@notreached");
1188 * Check to see if it is time to start boosting RCU readers that are
1189 * blocking the current grace period, and, if so, tell the per-rcu_node
1190 * kthread to start boosting them. If there is an expedited grace
1191 * period in progress, it is always time to boost.
1193 * The caller must hold rnp->lock, which this function releases,
1194 * but irqs remain disabled. The ->boost_kthread_task is immortal,
1195 * so we don't need to worry about it going away.
1197 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
)
1199 struct task_struct
*t
;
1201 if (!rcu_preempt_blocked_readers_cgp(rnp
) && rnp
->exp_tasks
== NULL
) {
1202 rnp
->n_balk_exp_gp_tasks
++;
1203 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1206 if (rnp
->exp_tasks
!= NULL
||
1207 (rnp
->gp_tasks
!= NULL
&&
1208 rnp
->boost_tasks
== NULL
&&
1210 ULONG_CMP_GE(jiffies
, rnp
->boost_time
))) {
1211 if (rnp
->exp_tasks
== NULL
)
1212 rnp
->boost_tasks
= rnp
->gp_tasks
;
1213 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1214 t
= rnp
->boost_kthread_task
;
1218 rcu_initiate_boost_trace(rnp
);
1219 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1224 * Wake up the per-CPU kthread to invoke RCU callbacks.
1226 static void invoke_rcu_callbacks_kthread(void)
1228 unsigned long flags
;
1230 local_irq_save(flags
);
1231 __this_cpu_write(rcu_cpu_has_work
, 1);
1232 if (__this_cpu_read(rcu_cpu_kthread_task
) != NULL
&&
1233 current
!= __this_cpu_read(rcu_cpu_kthread_task
))
1234 wake_up_process(__this_cpu_read(rcu_cpu_kthread_task
));
1235 local_irq_restore(flags
);
1239 * Is the current CPU running the RCU-callbacks kthread?
1240 * Caller must have preemption disabled.
1242 static bool rcu_is_callbacks_kthread(void)
1244 return __get_cpu_var(rcu_cpu_kthread_task
) == current
;
1248 * Set the affinity of the boost kthread. The CPU-hotplug locks are
1249 * held, so no one should be messing with the existence of the boost
1252 static void rcu_boost_kthread_setaffinity(struct rcu_node
*rnp
,
1255 struct task_struct
*t
;
1257 t
= rnp
->boost_kthread_task
;
1259 set_cpus_allowed_ptr(rnp
->boost_kthread_task
, cm
);
1262 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1265 * Do priority-boost accounting for the start of a new grace period.
1267 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
)
1269 rnp
->boost_time
= jiffies
+ RCU_BOOST_DELAY_JIFFIES
;
1273 * Create an RCU-boost kthread for the specified node if one does not
1274 * already exist. We only create this kthread for preemptible RCU.
1275 * Returns zero if all is well, a negated errno otherwise.
1277 static int __cpuinit
rcu_spawn_one_boost_kthread(struct rcu_state
*rsp
,
1278 struct rcu_node
*rnp
,
1281 unsigned long flags
;
1282 struct sched_param sp
;
1283 struct task_struct
*t
;
1285 if (&rcu_preempt_state
!= rsp
)
1288 if (rnp
->boost_kthread_task
!= NULL
)
1290 t
= kthread_create(rcu_boost_kthread
, (void *)rnp
,
1291 "rcub/%d", rnp_index
);
1294 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1295 rnp
->boost_kthread_task
= t
;
1296 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1297 sp
.sched_priority
= RCU_BOOST_PRIO
;
1298 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1299 wake_up_process(t
); /* get to TASK_INTERRUPTIBLE quickly. */
1303 #ifdef CONFIG_HOTPLUG_CPU
1306 * Stop the RCU's per-CPU kthread when its CPU goes offline,.
1308 static void rcu_stop_cpu_kthread(int cpu
)
1310 struct task_struct
*t
;
1312 /* Stop the CPU's kthread. */
1313 t
= per_cpu(rcu_cpu_kthread_task
, cpu
);
1315 per_cpu(rcu_cpu_kthread_task
, cpu
) = NULL
;
1320 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1322 static void rcu_kthread_do_work(void)
1324 rcu_do_batch(&rcu_sched_state
, &__get_cpu_var(rcu_sched_data
));
1325 rcu_do_batch(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1326 rcu_preempt_do_callbacks();
1330 * Wake up the specified per-rcu_node-structure kthread.
1331 * Because the per-rcu_node kthreads are immortal, we don't need
1332 * to do anything to keep them alive.
1334 static void invoke_rcu_node_kthread(struct rcu_node
*rnp
)
1336 struct task_struct
*t
;
1338 t
= rnp
->node_kthread_task
;
1344 * Set the specified CPU's kthread to run RT or not, as specified by
1345 * the to_rt argument. The CPU-hotplug locks are held, so the task
1346 * is not going away.
1348 static void rcu_cpu_kthread_setrt(int cpu
, int to_rt
)
1351 struct sched_param sp
;
1352 struct task_struct
*t
;
1354 t
= per_cpu(rcu_cpu_kthread_task
, cpu
);
1358 policy
= SCHED_FIFO
;
1359 sp
.sched_priority
= RCU_KTHREAD_PRIO
;
1361 policy
= SCHED_NORMAL
;
1362 sp
.sched_priority
= 0;
1364 sched_setscheduler_nocheck(t
, policy
, &sp
);
1368 * Timer handler to initiate the waking up of per-CPU kthreads that
1369 * have yielded the CPU due to excess numbers of RCU callbacks.
1370 * We wake up the per-rcu_node kthread, which in turn will wake up
1371 * the booster kthread.
1373 static void rcu_cpu_kthread_timer(unsigned long arg
)
1375 struct rcu_data
*rdp
= per_cpu_ptr(rcu_state
->rda
, arg
);
1376 struct rcu_node
*rnp
= rdp
->mynode
;
1378 atomic_or(rdp
->grpmask
, &rnp
->wakemask
);
1379 invoke_rcu_node_kthread(rnp
);
1383 * Drop to non-real-time priority and yield, but only after posting a
1384 * timer that will cause us to regain our real-time priority if we
1385 * remain preempted. Either way, we restore our real-time priority
1388 static void rcu_yield(void (*f
)(unsigned long), unsigned long arg
)
1390 struct sched_param sp
;
1391 struct timer_list yield_timer
;
1392 int prio
= current
->rt_priority
;
1394 setup_timer_on_stack(&yield_timer
, f
, arg
);
1395 mod_timer(&yield_timer
, jiffies
+ 2);
1396 sp
.sched_priority
= 0;
1397 sched_setscheduler_nocheck(current
, SCHED_NORMAL
, &sp
);
1398 set_user_nice(current
, 19);
1400 set_user_nice(current
, 0);
1401 sp
.sched_priority
= prio
;
1402 sched_setscheduler_nocheck(current
, SCHED_FIFO
, &sp
);
1403 del_timer(&yield_timer
);
1407 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1408 * This can happen while the corresponding CPU is either coming online
1409 * or going offline. We cannot wait until the CPU is fully online
1410 * before starting the kthread, because the various notifier functions
1411 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1412 * the corresponding CPU is online.
1414 * Return 1 if the kthread needs to stop, 0 otherwise.
1416 * Caller must disable bh. This function can momentarily enable it.
1418 static int rcu_cpu_kthread_should_stop(int cpu
)
1420 while (cpu_is_offline(cpu
) ||
1421 !cpumask_equal(¤t
->cpus_allowed
, cpumask_of(cpu
)) ||
1422 smp_processor_id() != cpu
) {
1423 if (kthread_should_stop())
1425 per_cpu(rcu_cpu_kthread_status
, cpu
) = RCU_KTHREAD_OFFCPU
;
1426 per_cpu(rcu_cpu_kthread_cpu
, cpu
) = raw_smp_processor_id();
1428 schedule_timeout_uninterruptible(1);
1429 if (!cpumask_equal(¤t
->cpus_allowed
, cpumask_of(cpu
)))
1430 set_cpus_allowed_ptr(current
, cpumask_of(cpu
));
1433 per_cpu(rcu_cpu_kthread_cpu
, cpu
) = cpu
;
1438 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1439 * RCU softirq used in flavors and configurations of RCU that do not
1440 * support RCU priority boosting.
1442 static int rcu_cpu_kthread(void *arg
)
1444 int cpu
= (int)(long)arg
;
1445 unsigned long flags
;
1447 unsigned int *statusp
= &per_cpu(rcu_cpu_kthread_status
, cpu
);
1449 char *workp
= &per_cpu(rcu_cpu_has_work
, cpu
);
1451 trace_rcu_utilization("Start CPU kthread@init");
1453 *statusp
= RCU_KTHREAD_WAITING
;
1454 trace_rcu_utilization("End CPU kthread@rcu_wait");
1455 rcu_wait(*workp
!= 0 || kthread_should_stop());
1456 trace_rcu_utilization("Start CPU kthread@rcu_wait");
1458 if (rcu_cpu_kthread_should_stop(cpu
)) {
1462 *statusp
= RCU_KTHREAD_RUNNING
;
1463 per_cpu(rcu_cpu_kthread_loops
, cpu
)++;
1464 local_irq_save(flags
);
1467 local_irq_restore(flags
);
1469 rcu_kthread_do_work();
1476 *statusp
= RCU_KTHREAD_YIELDING
;
1477 trace_rcu_utilization("End CPU kthread@rcu_yield");
1478 rcu_yield(rcu_cpu_kthread_timer
, (unsigned long)cpu
);
1479 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1483 *statusp
= RCU_KTHREAD_STOPPED
;
1484 trace_rcu_utilization("End CPU kthread@term");
1489 * Spawn a per-CPU kthread, setting up affinity and priority.
1490 * Because the CPU hotplug lock is held, no other CPU will be attempting
1491 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1492 * attempting to access it during boot, but the locking in kthread_bind()
1493 * will enforce sufficient ordering.
1495 * Please note that we cannot simply refuse to wake up the per-CPU
1496 * kthread because kthreads are created in TASK_UNINTERRUPTIBLE state,
1497 * which can result in softlockup complaints if the task ends up being
1498 * idle for more than a couple of minutes.
1500 * However, please note also that we cannot bind the per-CPU kthread to its
1501 * CPU until that CPU is fully online. We also cannot wait until the
1502 * CPU is fully online before we create its per-CPU kthread, as this would
1503 * deadlock the system when CPU notifiers tried waiting for grace
1504 * periods. So we bind the per-CPU kthread to its CPU only if the CPU
1505 * is online. If its CPU is not yet fully online, then the code in
1506 * rcu_cpu_kthread() will wait until it is fully online, and then do
1509 static int __cpuinit
rcu_spawn_one_cpu_kthread(int cpu
)
1511 struct sched_param sp
;
1512 struct task_struct
*t
;
1514 if (!rcu_scheduler_fully_active
||
1515 per_cpu(rcu_cpu_kthread_task
, cpu
) != NULL
)
1517 t
= kthread_create_on_node(rcu_cpu_kthread
,
1523 if (cpu_online(cpu
))
1524 kthread_bind(t
, cpu
);
1525 per_cpu(rcu_cpu_kthread_cpu
, cpu
) = cpu
;
1526 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task
, cpu
) != NULL
);
1527 sp
.sched_priority
= RCU_KTHREAD_PRIO
;
1528 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1529 per_cpu(rcu_cpu_kthread_task
, cpu
) = t
;
1530 wake_up_process(t
); /* Get to TASK_INTERRUPTIBLE quickly. */
1535 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1536 * kthreads when needed. We ignore requests to wake up kthreads
1537 * for offline CPUs, which is OK because force_quiescent_state()
1538 * takes care of this case.
1540 static int rcu_node_kthread(void *arg
)
1543 unsigned long flags
;
1545 struct rcu_node
*rnp
= (struct rcu_node
*)arg
;
1546 struct sched_param sp
;
1547 struct task_struct
*t
;
1550 rnp
->node_kthread_status
= RCU_KTHREAD_WAITING
;
1551 rcu_wait(atomic_read(&rnp
->wakemask
) != 0);
1552 rnp
->node_kthread_status
= RCU_KTHREAD_RUNNING
;
1553 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1554 mask
= atomic_xchg(&rnp
->wakemask
, 0);
1555 rcu_initiate_boost(rnp
, flags
); /* releases rnp->lock. */
1556 for (cpu
= rnp
->grplo
; cpu
<= rnp
->grphi
; cpu
++, mask
>>= 1) {
1557 if ((mask
& 0x1) == 0)
1560 t
= per_cpu(rcu_cpu_kthread_task
, cpu
);
1561 if (!cpu_online(cpu
) || t
== NULL
) {
1565 per_cpu(rcu_cpu_has_work
, cpu
) = 1;
1566 sp
.sched_priority
= RCU_KTHREAD_PRIO
;
1567 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1572 rnp
->node_kthread_status
= RCU_KTHREAD_STOPPED
;
1577 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1578 * served by the rcu_node in question. The CPU hotplug lock is still
1579 * held, so the value of rnp->qsmaskinit will be stable.
1581 * We don't include outgoingcpu in the affinity set, use -1 if there is
1582 * no outgoing CPU. If there are no CPUs left in the affinity set,
1583 * this function allows the kthread to execute on any CPU.
1585 static void rcu_node_kthread_setaffinity(struct rcu_node
*rnp
, int outgoingcpu
)
1589 unsigned long mask
= rnp
->qsmaskinit
;
1591 if (rnp
->node_kthread_task
== NULL
)
1593 if (!alloc_cpumask_var(&cm
, GFP_KERNEL
))
1596 for (cpu
= rnp
->grplo
; cpu
<= rnp
->grphi
; cpu
++, mask
>>= 1)
1597 if ((mask
& 0x1) && cpu
!= outgoingcpu
)
1598 cpumask_set_cpu(cpu
, cm
);
1599 if (cpumask_weight(cm
) == 0) {
1601 for (cpu
= rnp
->grplo
; cpu
<= rnp
->grphi
; cpu
++)
1602 cpumask_clear_cpu(cpu
, cm
);
1603 WARN_ON_ONCE(cpumask_weight(cm
) == 0);
1605 set_cpus_allowed_ptr(rnp
->node_kthread_task
, cm
);
1606 rcu_boost_kthread_setaffinity(rnp
, cm
);
1607 free_cpumask_var(cm
);
1611 * Spawn a per-rcu_node kthread, setting priority and affinity.
1612 * Called during boot before online/offline can happen, or, if
1613 * during runtime, with the main CPU-hotplug locks held. So only
1614 * one of these can be executing at a time.
1616 static int __cpuinit
rcu_spawn_one_node_kthread(struct rcu_state
*rsp
,
1617 struct rcu_node
*rnp
)
1619 unsigned long flags
;
1620 int rnp_index
= rnp
- &rsp
->node
[0];
1621 struct sched_param sp
;
1622 struct task_struct
*t
;
1624 if (!rcu_scheduler_fully_active
||
1625 rnp
->qsmaskinit
== 0)
1627 if (rnp
->node_kthread_task
== NULL
) {
1628 t
= kthread_create(rcu_node_kthread
, (void *)rnp
,
1629 "rcun/%d", rnp_index
);
1632 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1633 rnp
->node_kthread_task
= t
;
1634 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1635 sp
.sched_priority
= 99;
1636 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1637 wake_up_process(t
); /* get to TASK_INTERRUPTIBLE quickly. */
1639 return rcu_spawn_one_boost_kthread(rsp
, rnp
, rnp_index
);
1643 * Spawn all kthreads -- called as soon as the scheduler is running.
1645 static int __init
rcu_spawn_kthreads(void)
1648 struct rcu_node
*rnp
;
1650 rcu_scheduler_fully_active
= 1;
1651 for_each_possible_cpu(cpu
) {
1652 per_cpu(rcu_cpu_has_work
, cpu
) = 0;
1653 if (cpu_online(cpu
))
1654 (void)rcu_spawn_one_cpu_kthread(cpu
);
1656 rnp
= rcu_get_root(rcu_state
);
1657 (void)rcu_spawn_one_node_kthread(rcu_state
, rnp
);
1658 if (NUM_RCU_NODES
> 1) {
1659 rcu_for_each_leaf_node(rcu_state
, rnp
)
1660 (void)rcu_spawn_one_node_kthread(rcu_state
, rnp
);
1664 early_initcall(rcu_spawn_kthreads
);
1666 static void __cpuinit
rcu_prepare_kthreads(int cpu
)
1668 struct rcu_data
*rdp
= per_cpu_ptr(rcu_state
->rda
, cpu
);
1669 struct rcu_node
*rnp
= rdp
->mynode
;
1671 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1672 if (rcu_scheduler_fully_active
) {
1673 (void)rcu_spawn_one_cpu_kthread(cpu
);
1674 if (rnp
->node_kthread_task
== NULL
)
1675 (void)rcu_spawn_one_node_kthread(rcu_state
, rnp
);
1679 #else /* #ifdef CONFIG_RCU_BOOST */
1681 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
)
1683 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1686 static void invoke_rcu_callbacks_kthread(void)
1691 static bool rcu_is_callbacks_kthread(void)
1696 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
)
1700 #ifdef CONFIG_HOTPLUG_CPU
1702 static void rcu_stop_cpu_kthread(int cpu
)
1706 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1708 static void rcu_node_kthread_setaffinity(struct rcu_node
*rnp
, int outgoingcpu
)
1712 static void rcu_cpu_kthread_setrt(int cpu
, int to_rt
)
1716 static int __init
rcu_scheduler_really_started(void)
1718 rcu_scheduler_fully_active
= 1;
1721 early_initcall(rcu_scheduler_really_started
);
1723 static void __cpuinit
rcu_prepare_kthreads(int cpu
)
1727 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1729 #if !defined(CONFIG_RCU_FAST_NO_HZ)
1732 * Check to see if any future RCU-related work will need to be done
1733 * by the current CPU, even if none need be done immediately, returning
1734 * 1 if so. This function is part of the RCU implementation; it is -not-
1735 * an exported member of the RCU API.
1737 * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
1738 * any flavor of RCU.
1740 int rcu_needs_cpu(int cpu
, unsigned long *delta_jiffies
)
1742 *delta_jiffies
= ULONG_MAX
;
1743 return rcu_cpu_has_callbacks(cpu
);
1747 * Because we do not have RCU_FAST_NO_HZ, don't bother initializing for it.
1749 static void rcu_prepare_for_idle_init(int cpu
)
1754 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1757 static void rcu_cleanup_after_idle(int cpu
)
1762 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
1765 static void rcu_prepare_for_idle(int cpu
)
1770 * Don't bother keeping a running count of the number of RCU callbacks
1771 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1773 static void rcu_idle_count_callbacks_posted(void)
1777 #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1780 * This code is invoked when a CPU goes idle, at which point we want
1781 * to have the CPU do everything required for RCU so that it can enter
1782 * the energy-efficient dyntick-idle mode. This is handled by a
1783 * state machine implemented by rcu_prepare_for_idle() below.
1785 * The following three proprocessor symbols control this state machine:
1787 * RCU_IDLE_FLUSHES gives the maximum number of times that we will attempt
1788 * to satisfy RCU. Beyond this point, it is better to incur a periodic
1789 * scheduling-clock interrupt than to loop through the state machine
1791 * RCU_IDLE_OPT_FLUSHES gives the number of RCU_IDLE_FLUSHES that are
1792 * optional if RCU does not need anything immediately from this
1793 * CPU, even if this CPU still has RCU callbacks queued. The first
1794 * times through the state machine are mandatory: we need to give
1795 * the state machine a chance to communicate a quiescent state
1797 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1798 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1799 * is sized to be roughly one RCU grace period. Those energy-efficiency
1800 * benchmarkers who might otherwise be tempted to set this to a large
1801 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1802 * system. And if you are -that- concerned about energy efficiency,
1803 * just power the system down and be done with it!
1804 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1805 * permitted to sleep in dyntick-idle mode with only lazy RCU
1806 * callbacks pending. Setting this too high can OOM your system.
1808 * The values below work well in practice. If future workloads require
1809 * adjustment, they can be converted into kernel config parameters, though
1810 * making the state machine smarter might be a better option.
1812 #define RCU_IDLE_FLUSHES 5 /* Number of dyntick-idle tries. */
1813 #define RCU_IDLE_OPT_FLUSHES 3 /* Optional dyntick-idle tries. */
1814 #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */
1815 #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
1817 extern int tick_nohz_enabled
;
1820 * Does the specified flavor of RCU have non-lazy callbacks pending on
1821 * the specified CPU? Both RCU flavor and CPU are specified by the
1822 * rcu_data structure.
1824 static bool __rcu_cpu_has_nonlazy_callbacks(struct rcu_data
*rdp
)
1826 return rdp
->qlen
!= rdp
->qlen_lazy
;
1829 #ifdef CONFIG_TREE_PREEMPT_RCU
1832 * Are there non-lazy RCU-preempt callbacks? (There cannot be if there
1833 * is no RCU-preempt in the kernel.)
1835 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu
)
1837 struct rcu_data
*rdp
= &per_cpu(rcu_preempt_data
, cpu
);
1839 return __rcu_cpu_has_nonlazy_callbacks(rdp
);
1842 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1844 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu
)
1849 #endif /* else #ifdef CONFIG_TREE_PREEMPT_RCU */
1852 * Does any flavor of RCU have non-lazy callbacks on the specified CPU?
1854 static bool rcu_cpu_has_nonlazy_callbacks(int cpu
)
1856 return __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_sched_data
, cpu
)) ||
1857 __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_bh_data
, cpu
)) ||
1858 rcu_preempt_cpu_has_nonlazy_callbacks(cpu
);
1862 * Allow the CPU to enter dyntick-idle mode if either: (1) There are no
1863 * callbacks on this CPU, (2) this CPU has not yet attempted to enter
1864 * dyntick-idle mode, or (3) this CPU is in the process of attempting to
1865 * enter dyntick-idle mode. Otherwise, if we have recently tried and failed
1866 * to enter dyntick-idle mode, we refuse to try to enter it. After all,
1867 * it is better to incur scheduling-clock interrupts than to spin
1868 * continuously for the same time duration!
1870 * The delta_jiffies argument is used to store the time when RCU is
1871 * going to need the CPU again if it still has callbacks. The reason
1872 * for this is that rcu_prepare_for_idle() might need to post a timer,
1873 * but if so, it will do so after tick_nohz_stop_sched_tick() has set
1874 * the wakeup time for this CPU. This means that RCU's timer can be
1875 * delayed until the wakeup time, which defeats the purpose of posting
1878 int rcu_needs_cpu(int cpu
, unsigned long *delta_jiffies
)
1880 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
1882 /* Flag a new idle sojourn to the idle-entry state machine. */
1883 rdtp
->idle_first_pass
= 1;
1884 /* If no callbacks, RCU doesn't need the CPU. */
1885 if (!rcu_cpu_has_callbacks(cpu
)) {
1886 *delta_jiffies
= ULONG_MAX
;
1889 if (rdtp
->dyntick_holdoff
== jiffies
) {
1890 /* RCU recently tried and failed, so don't try again. */
1894 /* Set up for the possibility that RCU will post a timer. */
1895 if (rcu_cpu_has_nonlazy_callbacks(cpu
)) {
1896 *delta_jiffies
= round_up(RCU_IDLE_GP_DELAY
+ jiffies
,
1897 RCU_IDLE_GP_DELAY
) - jiffies
;
1899 *delta_jiffies
= jiffies
+ RCU_IDLE_LAZY_GP_DELAY
;
1900 *delta_jiffies
= round_jiffies(*delta_jiffies
) - jiffies
;
1906 * Handler for smp_call_function_single(). The only point of this
1907 * handler is to wake the CPU up, so the handler does only tracing.
1909 void rcu_idle_demigrate(void *unused
)
1911 trace_rcu_prep_idle("Demigrate");
1915 * Timer handler used to force CPU to start pushing its remaining RCU
1916 * callbacks in the case where it entered dyntick-idle mode with callbacks
1917 * pending. The hander doesn't really need to do anything because the
1918 * real work is done upon re-entry to idle, or by the next scheduling-clock
1919 * interrupt should idle not be re-entered.
1921 * One special case: the timer gets migrated without awakening the CPU
1922 * on which the timer was scheduled on. In this case, we must wake up
1923 * that CPU. We do so with smp_call_function_single().
1925 static void rcu_idle_gp_timer_func(unsigned long cpu_in
)
1927 int cpu
= (int)cpu_in
;
1929 trace_rcu_prep_idle("Timer");
1930 if (cpu
!= smp_processor_id())
1931 smp_call_function_single(cpu
, rcu_idle_demigrate
, NULL
, 0);
1933 WARN_ON_ONCE(1); /* Getting here can hang the system... */
1937 * Initialize the timer used to pull CPUs out of dyntick-idle mode.
1939 static void rcu_prepare_for_idle_init(int cpu
)
1941 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
1943 rdtp
->dyntick_holdoff
= jiffies
- 1;
1944 setup_timer(&rdtp
->idle_gp_timer
, rcu_idle_gp_timer_func
, cpu
);
1945 rdtp
->idle_gp_timer_expires
= jiffies
- 1;
1946 rdtp
->idle_first_pass
= 1;
1950 * Clean up for exit from idle. Because we are exiting from idle, there
1951 * is no longer any point to ->idle_gp_timer, so cancel it. This will
1952 * do nothing if this timer is not active, so just cancel it unconditionally.
1954 static void rcu_cleanup_after_idle(int cpu
)
1956 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
1958 del_timer(&rdtp
->idle_gp_timer
);
1959 trace_rcu_prep_idle("Cleanup after idle");
1960 rdtp
->tick_nohz_enabled_snap
= ACCESS_ONCE(tick_nohz_enabled
);
1964 * Check to see if any RCU-related work can be done by the current CPU,
1965 * and if so, schedule a softirq to get it done. This function is part
1966 * of the RCU implementation; it is -not- an exported member of the RCU API.
1968 * The idea is for the current CPU to clear out all work required by the
1969 * RCU core for the current grace period, so that this CPU can be permitted
1970 * to enter dyntick-idle mode. In some cases, it will need to be awakened
1971 * at the end of the grace period by whatever CPU ends the grace period.
1972 * This allows CPUs to go dyntick-idle more quickly, and to reduce the
1973 * number of wakeups by a modest integer factor.
1975 * Because it is not legal to invoke rcu_process_callbacks() with irqs
1976 * disabled, we do one pass of force_quiescent_state(), then do a
1977 * invoke_rcu_core() to cause rcu_process_callbacks() to be invoked
1978 * later. The ->dyntick_drain field controls the sequencing.
1980 * The caller must have disabled interrupts.
1982 static void rcu_prepare_for_idle(int cpu
)
1984 struct timer_list
*tp
;
1985 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
1988 /* Handle nohz enablement switches conservatively. */
1989 tne
= ACCESS_ONCE(tick_nohz_enabled
);
1990 if (tne
!= rdtp
->tick_nohz_enabled_snap
) {
1991 if (rcu_cpu_has_callbacks(cpu
))
1992 invoke_rcu_core(); /* force nohz to see update. */
1993 rdtp
->tick_nohz_enabled_snap
= tne
;
2000 * If this is an idle re-entry, for example, due to use of
2001 * RCU_NONIDLE() or the new idle-loop tracing API within the idle
2002 * loop, then don't take any state-machine actions, unless the
2003 * momentary exit from idle queued additional non-lazy callbacks.
2004 * Instead, repost the ->idle_gp_timer if this CPU has callbacks
2007 if (!rdtp
->idle_first_pass
&&
2008 (rdtp
->nonlazy_posted
== rdtp
->nonlazy_posted_snap
)) {
2009 if (rcu_cpu_has_callbacks(cpu
)) {
2010 tp
= &rdtp
->idle_gp_timer
;
2011 mod_timer_pinned(tp
, rdtp
->idle_gp_timer_expires
);
2015 rdtp
->idle_first_pass
= 0;
2016 rdtp
->nonlazy_posted_snap
= rdtp
->nonlazy_posted
- 1;
2019 * If there are no callbacks on this CPU, enter dyntick-idle mode.
2020 * Also reset state to avoid prejudicing later attempts.
2022 if (!rcu_cpu_has_callbacks(cpu
)) {
2023 rdtp
->dyntick_holdoff
= jiffies
- 1;
2024 rdtp
->dyntick_drain
= 0;
2025 trace_rcu_prep_idle("No callbacks");
2030 * If in holdoff mode, just return. We will presumably have
2031 * refrained from disabling the scheduling-clock tick.
2033 if (rdtp
->dyntick_holdoff
== jiffies
) {
2034 trace_rcu_prep_idle("In holdoff");
2038 /* Check and update the ->dyntick_drain sequencing. */
2039 if (rdtp
->dyntick_drain
<= 0) {
2040 /* First time through, initialize the counter. */
2041 rdtp
->dyntick_drain
= RCU_IDLE_FLUSHES
;
2042 } else if (rdtp
->dyntick_drain
<= RCU_IDLE_OPT_FLUSHES
&&
2043 !rcu_pending(cpu
) &&
2044 !local_softirq_pending()) {
2045 /* Can we go dyntick-idle despite still having callbacks? */
2046 rdtp
->dyntick_drain
= 0;
2047 rdtp
->dyntick_holdoff
= jiffies
;
2048 if (rcu_cpu_has_nonlazy_callbacks(cpu
)) {
2049 trace_rcu_prep_idle("Dyntick with callbacks");
2050 rdtp
->idle_gp_timer_expires
=
2051 round_up(jiffies
+ RCU_IDLE_GP_DELAY
,
2054 rdtp
->idle_gp_timer_expires
=
2055 round_jiffies(jiffies
+ RCU_IDLE_LAZY_GP_DELAY
);
2056 trace_rcu_prep_idle("Dyntick with lazy callbacks");
2058 tp
= &rdtp
->idle_gp_timer
;
2059 mod_timer_pinned(tp
, rdtp
->idle_gp_timer_expires
);
2060 rdtp
->nonlazy_posted_snap
= rdtp
->nonlazy_posted
;
2061 return; /* Nothing more to do immediately. */
2062 } else if (--(rdtp
->dyntick_drain
) <= 0) {
2063 /* We have hit the limit, so time to give up. */
2064 rdtp
->dyntick_holdoff
= jiffies
;
2065 trace_rcu_prep_idle("Begin holdoff");
2066 invoke_rcu_core(); /* Force the CPU out of dyntick-idle. */
2071 * Do one step of pushing the remaining RCU callbacks through
2072 * the RCU core state machine.
2074 #ifdef CONFIG_TREE_PREEMPT_RCU
2075 if (per_cpu(rcu_preempt_data
, cpu
).nxtlist
) {
2076 rcu_preempt_qs(cpu
);
2077 force_quiescent_state(&rcu_preempt_state
, 0);
2079 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2080 if (per_cpu(rcu_sched_data
, cpu
).nxtlist
) {
2082 force_quiescent_state(&rcu_sched_state
, 0);
2084 if (per_cpu(rcu_bh_data
, cpu
).nxtlist
) {
2086 force_quiescent_state(&rcu_bh_state
, 0);
2090 * If RCU callbacks are still pending, RCU still needs this CPU.
2091 * So try forcing the callbacks through the grace period.
2093 if (rcu_cpu_has_callbacks(cpu
)) {
2094 trace_rcu_prep_idle("More callbacks");
2097 trace_rcu_prep_idle("Callbacks drained");
2101 * Keep a running count of the number of non-lazy callbacks posted
2102 * on this CPU. This running counter (which is never decremented) allows
2103 * rcu_prepare_for_idle() to detect when something out of the idle loop
2104 * posts a callback, even if an equal number of callbacks are invoked.
2105 * Of course, callbacks should only be posted from within a trace event
2106 * designed to be called from idle or from within RCU_NONIDLE().
2108 static void rcu_idle_count_callbacks_posted(void)
2110 __this_cpu_add(rcu_dynticks
.nonlazy_posted
, 1);
2113 #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
2115 #ifdef CONFIG_RCU_CPU_STALL_INFO
2117 #ifdef CONFIG_RCU_FAST_NO_HZ
2119 static void print_cpu_stall_fast_no_hz(char *cp
, int cpu
)
2121 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
2122 struct timer_list
*tltp
= &rdtp
->idle_gp_timer
;
2124 sprintf(cp
, "drain=%d %c timer=%lu",
2125 rdtp
->dyntick_drain
,
2126 rdtp
->dyntick_holdoff
== jiffies
? 'H' : '.',
2127 timer_pending(tltp
) ? tltp
->expires
- jiffies
: -1);
2130 #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
2132 static void print_cpu_stall_fast_no_hz(char *cp
, int cpu
)
2137 #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
2139 /* Initiate the stall-info list. */
2140 static void print_cpu_stall_info_begin(void)
2142 printk(KERN_CONT
"\n");
2146 * Print out diagnostic information for the specified stalled CPU.
2148 * If the specified CPU is aware of the current RCU grace period
2149 * (flavor specified by rsp), then print the number of scheduling
2150 * clock interrupts the CPU has taken during the time that it has
2151 * been aware. Otherwise, print the number of RCU grace periods
2152 * that this CPU is ignorant of, for example, "1" if the CPU was
2153 * aware of the previous grace period.
2155 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
2157 static void print_cpu_stall_info(struct rcu_state
*rsp
, int cpu
)
2159 char fast_no_hz
[72];
2160 struct rcu_data
*rdp
= per_cpu_ptr(rsp
->rda
, cpu
);
2161 struct rcu_dynticks
*rdtp
= rdp
->dynticks
;
2163 unsigned long ticks_value
;
2165 if (rsp
->gpnum
== rdp
->gpnum
) {
2166 ticks_title
= "ticks this GP";
2167 ticks_value
= rdp
->ticks_this_gp
;
2169 ticks_title
= "GPs behind";
2170 ticks_value
= rsp
->gpnum
- rdp
->gpnum
;
2172 print_cpu_stall_fast_no_hz(fast_no_hz
, cpu
);
2173 printk(KERN_ERR
"\t%d: (%lu %s) idle=%03x/%llx/%d %s\n",
2174 cpu
, ticks_value
, ticks_title
,
2175 atomic_read(&rdtp
->dynticks
) & 0xfff,
2176 rdtp
->dynticks_nesting
, rdtp
->dynticks_nmi_nesting
,
2180 /* Terminate the stall-info list. */
2181 static void print_cpu_stall_info_end(void)
2183 printk(KERN_ERR
"\t");
2186 /* Zero ->ticks_this_gp for all flavors of RCU. */
2187 static void zero_cpu_stall_ticks(struct rcu_data
*rdp
)
2189 rdp
->ticks_this_gp
= 0;
2192 /* Increment ->ticks_this_gp for all flavors of RCU. */
2193 static void increment_cpu_stall_ticks(void)
2195 __get_cpu_var(rcu_sched_data
).ticks_this_gp
++;
2196 __get_cpu_var(rcu_bh_data
).ticks_this_gp
++;
2197 #ifdef CONFIG_TREE_PREEMPT_RCU
2198 __get_cpu_var(rcu_preempt_data
).ticks_this_gp
++;
2199 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2202 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
2204 static void print_cpu_stall_info_begin(void)
2206 printk(KERN_CONT
" {");
2209 static void print_cpu_stall_info(struct rcu_state
*rsp
, int cpu
)
2211 printk(KERN_CONT
" %d", cpu
);
2214 static void print_cpu_stall_info_end(void)
2216 printk(KERN_CONT
"} ");
2219 static void zero_cpu_stall_ticks(struct rcu_data
*rdp
)
2223 static void increment_cpu_stall_ticks(void)
2227 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */