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 * Suppress preemptible RCU's CPU stall warnings by pushing the
549 * time of the next stall-warning message comfortably far into the
552 static void rcu_preempt_stall_reset(void)
554 rcu_preempt_state
.jiffies_stall
= jiffies
+ ULONG_MAX
/ 2;
558 * Check that the list of blocked tasks for the newly completed grace
559 * period is in fact empty. It is a serious bug to complete a grace
560 * period that still has RCU readers blocked! This function must be
561 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
562 * must be held by the caller.
564 * Also, if there are blocked tasks on the list, they automatically
565 * block the newly created grace period, so set up ->gp_tasks accordingly.
567 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
569 WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp
));
570 if (!list_empty(&rnp
->blkd_tasks
))
571 rnp
->gp_tasks
= rnp
->blkd_tasks
.next
;
572 WARN_ON_ONCE(rnp
->qsmask
);
575 #ifdef CONFIG_HOTPLUG_CPU
578 * Handle tasklist migration for case in which all CPUs covered by the
579 * specified rcu_node have gone offline. Move them up to the root
580 * rcu_node. The reason for not just moving them to the immediate
581 * parent is to remove the need for rcu_read_unlock_special() to
582 * make more than two attempts to acquire the target rcu_node's lock.
583 * Returns true if there were tasks blocking the current RCU grace
586 * Returns 1 if there was previously a task blocking the current grace
587 * period on the specified rcu_node structure.
589 * The caller must hold rnp->lock with irqs disabled.
591 static int rcu_preempt_offline_tasks(struct rcu_state
*rsp
,
592 struct rcu_node
*rnp
,
593 struct rcu_data
*rdp
)
595 struct list_head
*lp
;
596 struct list_head
*lp_root
;
598 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
599 struct task_struct
*t
;
601 if (rnp
== rnp_root
) {
602 WARN_ONCE(1, "Last CPU thought to be offlined?");
603 return 0; /* Shouldn't happen: at least one CPU online. */
606 /* If we are on an internal node, complain bitterly. */
607 WARN_ON_ONCE(rnp
!= rdp
->mynode
);
610 * Move tasks up to root rcu_node. Don't try to get fancy for
611 * this corner-case operation -- just put this node's tasks
612 * at the head of the root node's list, and update the root node's
613 * ->gp_tasks and ->exp_tasks pointers to those of this node's,
614 * if non-NULL. This might result in waiting for more tasks than
615 * absolutely necessary, but this is a good performance/complexity
618 if (rcu_preempt_blocked_readers_cgp(rnp
) && rnp
->qsmask
== 0)
619 retval
|= RCU_OFL_TASKS_NORM_GP
;
620 if (rcu_preempted_readers_exp(rnp
))
621 retval
|= RCU_OFL_TASKS_EXP_GP
;
622 lp
= &rnp
->blkd_tasks
;
623 lp_root
= &rnp_root
->blkd_tasks
;
624 while (!list_empty(lp
)) {
625 t
= list_entry(lp
->next
, typeof(*t
), rcu_node_entry
);
626 raw_spin_lock(&rnp_root
->lock
); /* irqs already disabled */
627 list_del(&t
->rcu_node_entry
);
628 t
->rcu_blocked_node
= rnp_root
;
629 list_add(&t
->rcu_node_entry
, lp_root
);
630 if (&t
->rcu_node_entry
== rnp
->gp_tasks
)
631 rnp_root
->gp_tasks
= rnp
->gp_tasks
;
632 if (&t
->rcu_node_entry
== rnp
->exp_tasks
)
633 rnp_root
->exp_tasks
= rnp
->exp_tasks
;
634 #ifdef CONFIG_RCU_BOOST
635 if (&t
->rcu_node_entry
== rnp
->boost_tasks
)
636 rnp_root
->boost_tasks
= rnp
->boost_tasks
;
637 #endif /* #ifdef CONFIG_RCU_BOOST */
638 raw_spin_unlock(&rnp_root
->lock
); /* irqs still disabled */
641 #ifdef CONFIG_RCU_BOOST
642 /* In case root is being boosted and leaf is not. */
643 raw_spin_lock(&rnp_root
->lock
); /* irqs already disabled */
644 if (rnp_root
->boost_tasks
!= NULL
&&
645 rnp_root
->boost_tasks
!= rnp_root
->gp_tasks
)
646 rnp_root
->boost_tasks
= rnp_root
->gp_tasks
;
647 raw_spin_unlock(&rnp_root
->lock
); /* irqs still disabled */
648 #endif /* #ifdef CONFIG_RCU_BOOST */
650 rnp
->gp_tasks
= NULL
;
651 rnp
->exp_tasks
= NULL
;
655 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
658 * Do CPU-offline processing for preemptible RCU.
660 static void rcu_preempt_cleanup_dead_cpu(int cpu
)
662 rcu_cleanup_dead_cpu(cpu
, &rcu_preempt_state
);
666 * Check for a quiescent state from the current CPU. When a task blocks,
667 * the task is recorded in the corresponding CPU's rcu_node structure,
668 * which is checked elsewhere.
670 * Caller must disable hard irqs.
672 static void rcu_preempt_check_callbacks(int cpu
)
674 struct task_struct
*t
= current
;
676 if (t
->rcu_read_lock_nesting
== 0) {
680 if (t
->rcu_read_lock_nesting
> 0 &&
681 per_cpu(rcu_preempt_data
, cpu
).qs_pending
)
682 t
->rcu_read_unlock_special
|= RCU_READ_UNLOCK_NEED_QS
;
686 * Process callbacks for preemptible RCU.
688 static void rcu_preempt_process_callbacks(void)
690 __rcu_process_callbacks(&rcu_preempt_state
,
691 &__get_cpu_var(rcu_preempt_data
));
694 #ifdef CONFIG_RCU_BOOST
696 static void rcu_preempt_do_callbacks(void)
698 rcu_do_batch(&rcu_preempt_state
, &__get_cpu_var(rcu_preempt_data
));
701 #endif /* #ifdef CONFIG_RCU_BOOST */
704 * Queue a preemptible-RCU callback for invocation after a grace period.
706 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
708 __call_rcu(head
, func
, &rcu_preempt_state
, 0);
710 EXPORT_SYMBOL_GPL(call_rcu
);
713 * Queue an RCU callback for lazy invocation after a grace period.
714 * This will likely be later named something like "call_rcu_lazy()",
715 * but this change will require some way of tagging the lazy RCU
716 * callbacks in the list of pending callbacks. Until then, this
717 * function may only be called from __kfree_rcu().
719 void kfree_call_rcu(struct rcu_head
*head
,
720 void (*func
)(struct rcu_head
*rcu
))
722 __call_rcu(head
, func
, &rcu_preempt_state
, 1);
724 EXPORT_SYMBOL_GPL(kfree_call_rcu
);
727 * synchronize_rcu - wait until a grace period has elapsed.
729 * Control will return to the caller some time after a full grace
730 * period has elapsed, in other words after all currently executing RCU
731 * read-side critical sections have completed. Note, however, that
732 * upon return from synchronize_rcu(), the caller might well be executing
733 * concurrently with new RCU read-side critical sections that began while
734 * synchronize_rcu() was waiting. RCU read-side critical sections are
735 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
737 void synchronize_rcu(void)
739 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map
) &&
740 !lock_is_held(&rcu_lock_map
) &&
741 !lock_is_held(&rcu_sched_lock_map
),
742 "Illegal synchronize_rcu() in RCU read-side critical section");
743 if (!rcu_scheduler_active
)
745 wait_rcu_gp(call_rcu
);
747 EXPORT_SYMBOL_GPL(synchronize_rcu
);
749 static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq
);
750 static long sync_rcu_preempt_exp_count
;
751 static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex
);
754 * Return non-zero if there are any tasks in RCU read-side critical
755 * sections blocking the current preemptible-RCU expedited grace period.
756 * If there is no preemptible-RCU expedited grace period currently in
757 * progress, returns zero unconditionally.
759 static int rcu_preempted_readers_exp(struct rcu_node
*rnp
)
761 return rnp
->exp_tasks
!= NULL
;
765 * return non-zero if there is no RCU expedited grace period in progress
766 * for the specified rcu_node structure, in other words, if all CPUs and
767 * tasks covered by the specified rcu_node structure have done their bit
768 * for the current expedited grace period. Works only for preemptible
769 * RCU -- other RCU implementation use other means.
771 * Caller must hold sync_rcu_preempt_exp_mutex.
773 static int sync_rcu_preempt_exp_done(struct rcu_node
*rnp
)
775 return !rcu_preempted_readers_exp(rnp
) &&
776 ACCESS_ONCE(rnp
->expmask
) == 0;
780 * Report the exit from RCU read-side critical section for the last task
781 * that queued itself during or before the current expedited preemptible-RCU
782 * grace period. This event is reported either to the rcu_node structure on
783 * which the task was queued or to one of that rcu_node structure's ancestors,
784 * recursively up the tree. (Calm down, calm down, we do the recursion
787 * Most callers will set the "wake" flag, but the task initiating the
788 * expedited grace period need not wake itself.
790 * Caller must hold sync_rcu_preempt_exp_mutex.
792 static void rcu_report_exp_rnp(struct rcu_state
*rsp
, struct rcu_node
*rnp
,
798 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
800 if (!sync_rcu_preempt_exp_done(rnp
)) {
801 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
804 if (rnp
->parent
== NULL
) {
805 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
807 wake_up(&sync_rcu_preempt_exp_wq
);
811 raw_spin_unlock(&rnp
->lock
); /* irqs remain disabled */
813 raw_spin_lock(&rnp
->lock
); /* irqs already disabled */
814 rnp
->expmask
&= ~mask
;
819 * Snapshot the tasks blocking the newly started preemptible-RCU expedited
820 * grace period for the specified rcu_node structure. If there are no such
821 * tasks, report it up the rcu_node hierarchy.
823 * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock.
826 sync_rcu_preempt_exp_init(struct rcu_state
*rsp
, struct rcu_node
*rnp
)
831 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
832 if (list_empty(&rnp
->blkd_tasks
))
833 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
835 rnp
->exp_tasks
= rnp
->blkd_tasks
.next
;
836 rcu_initiate_boost(rnp
, flags
); /* releases rnp->lock */
840 rcu_report_exp_rnp(rsp
, rnp
, false); /* Don't wake self. */
844 * synchronize_rcu_expedited - Brute-force RCU grace period
846 * Wait for an RCU-preempt grace period, but expedite it. The basic
847 * idea is to invoke synchronize_sched_expedited() to push all the tasks to
848 * the ->blkd_tasks lists and wait for this list to drain. This consumes
849 * significant time on all CPUs and is unfriendly to real-time workloads,
850 * so is thus not recommended for any sort of common-case code.
851 * In fact, if you are using synchronize_rcu_expedited() in a loop,
852 * please restructure your code to batch your updates, and then Use a
853 * single synchronize_rcu() instead.
855 * Note that it is illegal to call this function while holding any lock
856 * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal
857 * to call this function from a CPU-hotplug notifier. Failing to observe
858 * these restriction will result in deadlock.
860 void synchronize_rcu_expedited(void)
863 struct rcu_node
*rnp
;
864 struct rcu_state
*rsp
= &rcu_preempt_state
;
868 smp_mb(); /* Caller's modifications seen first by other CPUs. */
869 snap
= ACCESS_ONCE(sync_rcu_preempt_exp_count
) + 1;
870 smp_mb(); /* Above access cannot bleed into critical section. */
873 * Acquire lock, falling back to synchronize_rcu() if too many
874 * lock-acquisition failures. Of course, if someone does the
875 * expedited grace period for us, just leave.
877 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex
)) {
879 udelay(trycount
* num_online_cpus());
884 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count
) - snap
) > 0)
885 goto mb_ret
; /* Others did our work for us. */
887 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count
) - snap
) > 0)
888 goto unlock_mb_ret
; /* Others did our work for us. */
890 /* force all RCU readers onto ->blkd_tasks lists. */
891 synchronize_sched_expedited();
893 raw_spin_lock_irqsave(&rsp
->onofflock
, flags
);
895 /* Initialize ->expmask for all non-leaf rcu_node structures. */
896 rcu_for_each_nonleaf_node_breadth_first(rsp
, rnp
) {
897 raw_spin_lock(&rnp
->lock
); /* irqs already disabled. */
898 rnp
->expmask
= rnp
->qsmaskinit
;
899 raw_spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
902 /* Snapshot current state of ->blkd_tasks lists. */
903 rcu_for_each_leaf_node(rsp
, rnp
)
904 sync_rcu_preempt_exp_init(rsp
, rnp
);
905 if (NUM_RCU_NODES
> 1)
906 sync_rcu_preempt_exp_init(rsp
, rcu_get_root(rsp
));
908 raw_spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
910 /* Wait for snapshotted ->blkd_tasks lists to drain. */
911 rnp
= rcu_get_root(rsp
);
912 wait_event(sync_rcu_preempt_exp_wq
,
913 sync_rcu_preempt_exp_done(rnp
));
915 /* Clean up and exit. */
916 smp_mb(); /* ensure expedited GP seen before counter increment. */
917 ACCESS_ONCE(sync_rcu_preempt_exp_count
)++;
919 mutex_unlock(&sync_rcu_preempt_exp_mutex
);
921 smp_mb(); /* ensure subsequent action seen after grace period. */
923 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited
);
926 * Check to see if there is any immediate preemptible-RCU-related work
929 static int rcu_preempt_pending(int cpu
)
931 return __rcu_pending(&rcu_preempt_state
,
932 &per_cpu(rcu_preempt_data
, cpu
));
936 * Does preemptible RCU have callbacks on this CPU?
938 static int rcu_preempt_cpu_has_callbacks(int cpu
)
940 return !!per_cpu(rcu_preempt_data
, cpu
).nxtlist
;
944 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
946 void rcu_barrier(void)
948 _rcu_barrier(&rcu_preempt_state
);
950 EXPORT_SYMBOL_GPL(rcu_barrier
);
953 * Initialize preemptible RCU's per-CPU data.
955 static void __cpuinit
rcu_preempt_init_percpu_data(int cpu
)
957 rcu_init_percpu_data(cpu
, &rcu_preempt_state
, 1);
961 * Move preemptible RCU's callbacks from dying CPU to other online CPU
962 * and record a quiescent state.
964 static void rcu_preempt_cleanup_dying_cpu(void)
966 rcu_cleanup_dying_cpu(&rcu_preempt_state
);
970 * Initialize preemptible RCU's state structures.
972 static void __init
__rcu_init_preempt(void)
974 rcu_init_one(&rcu_preempt_state
, &rcu_preempt_data
);
977 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
979 static struct rcu_state
*rcu_state
= &rcu_sched_state
;
982 * Tell them what RCU they are running.
984 static void __init
rcu_bootup_announce(void)
986 printk(KERN_INFO
"Hierarchical RCU implementation.\n");
987 rcu_bootup_announce_oddness();
991 * Return the number of RCU batches processed thus far for debug & stats.
993 long rcu_batches_completed(void)
995 return rcu_batches_completed_sched();
997 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
1000 * Force a quiescent state for RCU, which, because there is no preemptible
1001 * RCU, becomes the same as rcu-sched.
1003 void rcu_force_quiescent_state(void)
1005 rcu_sched_force_quiescent_state();
1007 EXPORT_SYMBOL_GPL(rcu_force_quiescent_state
);
1010 * Because preemptible RCU does not exist, we never have to check for
1011 * CPUs being in quiescent states.
1013 static void rcu_preempt_note_context_switch(int cpu
)
1018 * Because preemptible RCU does not exist, there are never any preempted
1021 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
)
1026 #ifdef CONFIG_HOTPLUG_CPU
1028 /* Because preemptible RCU does not exist, no quieting of tasks. */
1029 static void rcu_report_unblock_qs_rnp(struct rcu_node
*rnp
, unsigned long flags
)
1031 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1034 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1037 * Because preemptible RCU does not exist, we never have to check for
1038 * tasks blocked within RCU read-side critical sections.
1040 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
)
1045 * Because preemptible RCU does not exist, we never have to check for
1046 * tasks blocked within RCU read-side critical sections.
1048 static int rcu_print_task_stall(struct rcu_node
*rnp
)
1054 * Because preemptible RCU does not exist, there is no need to suppress
1055 * its CPU stall warnings.
1057 static void rcu_preempt_stall_reset(void)
1062 * Because there is no preemptible RCU, there can be no readers blocked,
1063 * so there is no need to check for blocked tasks. So check only for
1064 * bogus qsmask values.
1066 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
1068 WARN_ON_ONCE(rnp
->qsmask
);
1071 #ifdef CONFIG_HOTPLUG_CPU
1074 * Because preemptible RCU does not exist, it never needs to migrate
1075 * tasks that were blocked within RCU read-side critical sections, and
1076 * such non-existent tasks cannot possibly have been blocking the current
1079 static int rcu_preempt_offline_tasks(struct rcu_state
*rsp
,
1080 struct rcu_node
*rnp
,
1081 struct rcu_data
*rdp
)
1086 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1089 * Because preemptible RCU does not exist, it never needs CPU-offline
1092 static void rcu_preempt_cleanup_dead_cpu(int cpu
)
1097 * Because preemptible RCU does not exist, it never has any callbacks
1100 static void rcu_preempt_check_callbacks(int cpu
)
1105 * Because preemptible RCU does not exist, it never has any callbacks
1108 static void rcu_preempt_process_callbacks(void)
1113 * Queue an RCU callback for lazy invocation after a grace period.
1114 * This will likely be later named something like "call_rcu_lazy()",
1115 * but this change will require some way of tagging the lazy RCU
1116 * callbacks in the list of pending callbacks. Until then, this
1117 * function may only be called from __kfree_rcu().
1119 * Because there is no preemptible RCU, we use RCU-sched instead.
1121 void kfree_call_rcu(struct rcu_head
*head
,
1122 void (*func
)(struct rcu_head
*rcu
))
1124 __call_rcu(head
, func
, &rcu_sched_state
, 1);
1126 EXPORT_SYMBOL_GPL(kfree_call_rcu
);
1129 * Wait for an rcu-preempt grace period, but make it happen quickly.
1130 * But because preemptible RCU does not exist, map to rcu-sched.
1132 void synchronize_rcu_expedited(void)
1134 synchronize_sched_expedited();
1136 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited
);
1138 #ifdef CONFIG_HOTPLUG_CPU
1141 * Because preemptible RCU does not exist, there is never any need to
1142 * report on tasks preempted in RCU read-side critical sections during
1143 * expedited RCU grace periods.
1145 static void rcu_report_exp_rnp(struct rcu_state
*rsp
, struct rcu_node
*rnp
,
1150 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1153 * Because preemptible RCU does not exist, it never has any work to do.
1155 static int rcu_preempt_pending(int cpu
)
1161 * Because preemptible RCU does not exist, it never has callbacks
1163 static int rcu_preempt_cpu_has_callbacks(int cpu
)
1169 * Because preemptible RCU does not exist, rcu_barrier() is just
1170 * another name for rcu_barrier_sched().
1172 void rcu_barrier(void)
1174 rcu_barrier_sched();
1176 EXPORT_SYMBOL_GPL(rcu_barrier
);
1179 * Because preemptible RCU does not exist, there is no per-CPU
1180 * data to initialize.
1182 static void __cpuinit
rcu_preempt_init_percpu_data(int cpu
)
1187 * Because there is no preemptible RCU, there is no cleanup to do.
1189 static void rcu_preempt_cleanup_dying_cpu(void)
1194 * Because preemptible RCU does not exist, it need not be initialized.
1196 static void __init
__rcu_init_preempt(void)
1200 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1202 #ifdef CONFIG_RCU_BOOST
1204 #include "rtmutex_common.h"
1206 #ifdef CONFIG_RCU_TRACE
1208 static void rcu_initiate_boost_trace(struct rcu_node
*rnp
)
1210 if (list_empty(&rnp
->blkd_tasks
))
1211 rnp
->n_balk_blkd_tasks
++;
1212 else if (rnp
->exp_tasks
== NULL
&& rnp
->gp_tasks
== NULL
)
1213 rnp
->n_balk_exp_gp_tasks
++;
1214 else if (rnp
->gp_tasks
!= NULL
&& rnp
->boost_tasks
!= NULL
)
1215 rnp
->n_balk_boost_tasks
++;
1216 else if (rnp
->gp_tasks
!= NULL
&& rnp
->qsmask
!= 0)
1217 rnp
->n_balk_notblocked
++;
1218 else if (rnp
->gp_tasks
!= NULL
&&
1219 ULONG_CMP_LT(jiffies
, rnp
->boost_time
))
1220 rnp
->n_balk_notyet
++;
1225 #else /* #ifdef CONFIG_RCU_TRACE */
1227 static void rcu_initiate_boost_trace(struct rcu_node
*rnp
)
1231 #endif /* #else #ifdef CONFIG_RCU_TRACE */
1234 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1235 * or ->boost_tasks, advancing the pointer to the next task in the
1236 * ->blkd_tasks list.
1238 * Note that irqs must be enabled: boosting the task can block.
1239 * Returns 1 if there are more tasks needing to be boosted.
1241 static int rcu_boost(struct rcu_node
*rnp
)
1243 unsigned long flags
;
1244 struct rt_mutex mtx
;
1245 struct task_struct
*t
;
1246 struct list_head
*tb
;
1248 if (rnp
->exp_tasks
== NULL
&& rnp
->boost_tasks
== NULL
)
1249 return 0; /* Nothing left to boost. */
1251 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1254 * Recheck under the lock: all tasks in need of boosting
1255 * might exit their RCU read-side critical sections on their own.
1257 if (rnp
->exp_tasks
== NULL
&& rnp
->boost_tasks
== NULL
) {
1258 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1263 * Preferentially boost tasks blocking expedited grace periods.
1264 * This cannot starve the normal grace periods because a second
1265 * expedited grace period must boost all blocked tasks, including
1266 * those blocking the pre-existing normal grace period.
1268 if (rnp
->exp_tasks
!= NULL
) {
1269 tb
= rnp
->exp_tasks
;
1270 rnp
->n_exp_boosts
++;
1272 tb
= rnp
->boost_tasks
;
1273 rnp
->n_normal_boosts
++;
1275 rnp
->n_tasks_boosted
++;
1278 * We boost task t by manufacturing an rt_mutex that appears to
1279 * be held by task t. We leave a pointer to that rt_mutex where
1280 * task t can find it, and task t will release the mutex when it
1281 * exits its outermost RCU read-side critical section. Then
1282 * simply acquiring this artificial rt_mutex will boost task
1283 * t's priority. (Thanks to tglx for suggesting this approach!)
1285 * Note that task t must acquire rnp->lock to remove itself from
1286 * the ->blkd_tasks list, which it will do from exit() if from
1287 * nowhere else. We therefore are guaranteed that task t will
1288 * stay around at least until we drop rnp->lock. Note that
1289 * rnp->lock also resolves races between our priority boosting
1290 * and task t's exiting its outermost RCU read-side critical
1293 t
= container_of(tb
, struct task_struct
, rcu_node_entry
);
1294 rt_mutex_init_proxy_locked(&mtx
, t
);
1295 t
->rcu_boost_mutex
= &mtx
;
1296 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1297 rt_mutex_lock(&mtx
); /* Side effect: boosts task t's priority. */
1298 rt_mutex_unlock(&mtx
); /* Keep lockdep happy. */
1300 return ACCESS_ONCE(rnp
->exp_tasks
) != NULL
||
1301 ACCESS_ONCE(rnp
->boost_tasks
) != NULL
;
1305 * Timer handler to initiate waking up of boost kthreads that
1306 * have yielded the CPU due to excessive numbers of tasks to
1307 * boost. We wake up the per-rcu_node kthread, which in turn
1308 * will wake up the booster kthread.
1310 static void rcu_boost_kthread_timer(unsigned long arg
)
1312 invoke_rcu_node_kthread((struct rcu_node
*)arg
);
1316 * Priority-boosting kthread. One per leaf rcu_node and one for the
1319 static int rcu_boost_kthread(void *arg
)
1321 struct rcu_node
*rnp
= (struct rcu_node
*)arg
;
1325 trace_rcu_utilization("Start boost kthread@init");
1327 rnp
->boost_kthread_status
= RCU_KTHREAD_WAITING
;
1328 trace_rcu_utilization("End boost kthread@rcu_wait");
1329 rcu_wait(rnp
->boost_tasks
|| rnp
->exp_tasks
);
1330 trace_rcu_utilization("Start boost kthread@rcu_wait");
1331 rnp
->boost_kthread_status
= RCU_KTHREAD_RUNNING
;
1332 more2boost
= rcu_boost(rnp
);
1338 trace_rcu_utilization("End boost kthread@rcu_yield");
1339 rcu_yield(rcu_boost_kthread_timer
, (unsigned long)rnp
);
1340 trace_rcu_utilization("Start boost kthread@rcu_yield");
1345 trace_rcu_utilization("End boost kthread@notreached");
1350 * Check to see if it is time to start boosting RCU readers that are
1351 * blocking the current grace period, and, if so, tell the per-rcu_node
1352 * kthread to start boosting them. If there is an expedited grace
1353 * period in progress, it is always time to boost.
1355 * The caller must hold rnp->lock, which this function releases,
1356 * but irqs remain disabled. The ->boost_kthread_task is immortal,
1357 * so we don't need to worry about it going away.
1359 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
)
1361 struct task_struct
*t
;
1363 if (!rcu_preempt_blocked_readers_cgp(rnp
) && rnp
->exp_tasks
== NULL
) {
1364 rnp
->n_balk_exp_gp_tasks
++;
1365 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1368 if (rnp
->exp_tasks
!= NULL
||
1369 (rnp
->gp_tasks
!= NULL
&&
1370 rnp
->boost_tasks
== NULL
&&
1372 ULONG_CMP_GE(jiffies
, rnp
->boost_time
))) {
1373 if (rnp
->exp_tasks
== NULL
)
1374 rnp
->boost_tasks
= rnp
->gp_tasks
;
1375 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1376 t
= rnp
->boost_kthread_task
;
1380 rcu_initiate_boost_trace(rnp
);
1381 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1386 * Wake up the per-CPU kthread to invoke RCU callbacks.
1388 static void invoke_rcu_callbacks_kthread(void)
1390 unsigned long flags
;
1392 local_irq_save(flags
);
1393 __this_cpu_write(rcu_cpu_has_work
, 1);
1394 if (__this_cpu_read(rcu_cpu_kthread_task
) != NULL
&&
1395 current
!= __this_cpu_read(rcu_cpu_kthread_task
))
1396 wake_up_process(__this_cpu_read(rcu_cpu_kthread_task
));
1397 local_irq_restore(flags
);
1401 * Is the current CPU running the RCU-callbacks kthread?
1402 * Caller must have preemption disabled.
1404 static bool rcu_is_callbacks_kthread(void)
1406 return __get_cpu_var(rcu_cpu_kthread_task
) == current
;
1410 * Set the affinity of the boost kthread. The CPU-hotplug locks are
1411 * held, so no one should be messing with the existence of the boost
1414 static void rcu_boost_kthread_setaffinity(struct rcu_node
*rnp
,
1417 struct task_struct
*t
;
1419 t
= rnp
->boost_kthread_task
;
1421 set_cpus_allowed_ptr(rnp
->boost_kthread_task
, cm
);
1424 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1427 * Do priority-boost accounting for the start of a new grace period.
1429 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
)
1431 rnp
->boost_time
= jiffies
+ RCU_BOOST_DELAY_JIFFIES
;
1435 * Create an RCU-boost kthread for the specified node if one does not
1436 * already exist. We only create this kthread for preemptible RCU.
1437 * Returns zero if all is well, a negated errno otherwise.
1439 static int __cpuinit
rcu_spawn_one_boost_kthread(struct rcu_state
*rsp
,
1440 struct rcu_node
*rnp
,
1443 unsigned long flags
;
1444 struct sched_param sp
;
1445 struct task_struct
*t
;
1447 if (&rcu_preempt_state
!= rsp
)
1450 if (rnp
->boost_kthread_task
!= NULL
)
1452 t
= kthread_create(rcu_boost_kthread
, (void *)rnp
,
1453 "rcub/%d", rnp_index
);
1456 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1457 rnp
->boost_kthread_task
= t
;
1458 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1459 sp
.sched_priority
= RCU_BOOST_PRIO
;
1460 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1461 wake_up_process(t
); /* get to TASK_INTERRUPTIBLE quickly. */
1465 #ifdef CONFIG_HOTPLUG_CPU
1468 * Stop the RCU's per-CPU kthread when its CPU goes offline,.
1470 static void rcu_stop_cpu_kthread(int cpu
)
1472 struct task_struct
*t
;
1474 /* Stop the CPU's kthread. */
1475 t
= per_cpu(rcu_cpu_kthread_task
, cpu
);
1477 per_cpu(rcu_cpu_kthread_task
, cpu
) = NULL
;
1482 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1484 static void rcu_kthread_do_work(void)
1486 rcu_do_batch(&rcu_sched_state
, &__get_cpu_var(rcu_sched_data
));
1487 rcu_do_batch(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1488 rcu_preempt_do_callbacks();
1492 * Wake up the specified per-rcu_node-structure kthread.
1493 * Because the per-rcu_node kthreads are immortal, we don't need
1494 * to do anything to keep them alive.
1496 static void invoke_rcu_node_kthread(struct rcu_node
*rnp
)
1498 struct task_struct
*t
;
1500 t
= rnp
->node_kthread_task
;
1506 * Set the specified CPU's kthread to run RT or not, as specified by
1507 * the to_rt argument. The CPU-hotplug locks are held, so the task
1508 * is not going away.
1510 static void rcu_cpu_kthread_setrt(int cpu
, int to_rt
)
1513 struct sched_param sp
;
1514 struct task_struct
*t
;
1516 t
= per_cpu(rcu_cpu_kthread_task
, cpu
);
1520 policy
= SCHED_FIFO
;
1521 sp
.sched_priority
= RCU_KTHREAD_PRIO
;
1523 policy
= SCHED_NORMAL
;
1524 sp
.sched_priority
= 0;
1526 sched_setscheduler_nocheck(t
, policy
, &sp
);
1530 * Timer handler to initiate the waking up of per-CPU kthreads that
1531 * have yielded the CPU due to excess numbers of RCU callbacks.
1532 * We wake up the per-rcu_node kthread, which in turn will wake up
1533 * the booster kthread.
1535 static void rcu_cpu_kthread_timer(unsigned long arg
)
1537 struct rcu_data
*rdp
= per_cpu_ptr(rcu_state
->rda
, arg
);
1538 struct rcu_node
*rnp
= rdp
->mynode
;
1540 atomic_or(rdp
->grpmask
, &rnp
->wakemask
);
1541 invoke_rcu_node_kthread(rnp
);
1545 * Drop to non-real-time priority and yield, but only after posting a
1546 * timer that will cause us to regain our real-time priority if we
1547 * remain preempted. Either way, we restore our real-time priority
1550 static void rcu_yield(void (*f
)(unsigned long), unsigned long arg
)
1552 struct sched_param sp
;
1553 struct timer_list yield_timer
;
1554 int prio
= current
->rt_priority
;
1556 setup_timer_on_stack(&yield_timer
, f
, arg
);
1557 mod_timer(&yield_timer
, jiffies
+ 2);
1558 sp
.sched_priority
= 0;
1559 sched_setscheduler_nocheck(current
, SCHED_NORMAL
, &sp
);
1560 set_user_nice(current
, 19);
1562 set_user_nice(current
, 0);
1563 sp
.sched_priority
= prio
;
1564 sched_setscheduler_nocheck(current
, SCHED_FIFO
, &sp
);
1565 del_timer(&yield_timer
);
1569 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1570 * This can happen while the corresponding CPU is either coming online
1571 * or going offline. We cannot wait until the CPU is fully online
1572 * before starting the kthread, because the various notifier functions
1573 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1574 * the corresponding CPU is online.
1576 * Return 1 if the kthread needs to stop, 0 otherwise.
1578 * Caller must disable bh. This function can momentarily enable it.
1580 static int rcu_cpu_kthread_should_stop(int cpu
)
1582 while (cpu_is_offline(cpu
) ||
1583 !cpumask_equal(¤t
->cpus_allowed
, cpumask_of(cpu
)) ||
1584 smp_processor_id() != cpu
) {
1585 if (kthread_should_stop())
1587 per_cpu(rcu_cpu_kthread_status
, cpu
) = RCU_KTHREAD_OFFCPU
;
1588 per_cpu(rcu_cpu_kthread_cpu
, cpu
) = raw_smp_processor_id();
1590 schedule_timeout_uninterruptible(1);
1591 if (!cpumask_equal(¤t
->cpus_allowed
, cpumask_of(cpu
)))
1592 set_cpus_allowed_ptr(current
, cpumask_of(cpu
));
1595 per_cpu(rcu_cpu_kthread_cpu
, cpu
) = cpu
;
1600 * Per-CPU kernel thread that invokes RCU callbacks. This replaces the
1601 * RCU softirq used in flavors and configurations of RCU that do not
1602 * support RCU priority boosting.
1604 static int rcu_cpu_kthread(void *arg
)
1606 int cpu
= (int)(long)arg
;
1607 unsigned long flags
;
1609 unsigned int *statusp
= &per_cpu(rcu_cpu_kthread_status
, cpu
);
1611 char *workp
= &per_cpu(rcu_cpu_has_work
, cpu
);
1613 trace_rcu_utilization("Start CPU kthread@init");
1615 *statusp
= RCU_KTHREAD_WAITING
;
1616 trace_rcu_utilization("End CPU kthread@rcu_wait");
1617 rcu_wait(*workp
!= 0 || kthread_should_stop());
1618 trace_rcu_utilization("Start CPU kthread@rcu_wait");
1620 if (rcu_cpu_kthread_should_stop(cpu
)) {
1624 *statusp
= RCU_KTHREAD_RUNNING
;
1625 per_cpu(rcu_cpu_kthread_loops
, cpu
)++;
1626 local_irq_save(flags
);
1629 local_irq_restore(flags
);
1631 rcu_kthread_do_work();
1638 *statusp
= RCU_KTHREAD_YIELDING
;
1639 trace_rcu_utilization("End CPU kthread@rcu_yield");
1640 rcu_yield(rcu_cpu_kthread_timer
, (unsigned long)cpu
);
1641 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1645 *statusp
= RCU_KTHREAD_STOPPED
;
1646 trace_rcu_utilization("End CPU kthread@term");
1651 * Spawn a per-CPU kthread, setting up affinity and priority.
1652 * Because the CPU hotplug lock is held, no other CPU will be attempting
1653 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1654 * attempting to access it during boot, but the locking in kthread_bind()
1655 * will enforce sufficient ordering.
1657 * Please note that we cannot simply refuse to wake up the per-CPU
1658 * kthread because kthreads are created in TASK_UNINTERRUPTIBLE state,
1659 * which can result in softlockup complaints if the task ends up being
1660 * idle for more than a couple of minutes.
1662 * However, please note also that we cannot bind the per-CPU kthread to its
1663 * CPU until that CPU is fully online. We also cannot wait until the
1664 * CPU is fully online before we create its per-CPU kthread, as this would
1665 * deadlock the system when CPU notifiers tried waiting for grace
1666 * periods. So we bind the per-CPU kthread to its CPU only if the CPU
1667 * is online. If its CPU is not yet fully online, then the code in
1668 * rcu_cpu_kthread() will wait until it is fully online, and then do
1671 static int __cpuinit
rcu_spawn_one_cpu_kthread(int cpu
)
1673 struct sched_param sp
;
1674 struct task_struct
*t
;
1676 if (!rcu_scheduler_fully_active
||
1677 per_cpu(rcu_cpu_kthread_task
, cpu
) != NULL
)
1679 t
= kthread_create_on_node(rcu_cpu_kthread
,
1685 if (cpu_online(cpu
))
1686 kthread_bind(t
, cpu
);
1687 per_cpu(rcu_cpu_kthread_cpu
, cpu
) = cpu
;
1688 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task
, cpu
) != NULL
);
1689 sp
.sched_priority
= RCU_KTHREAD_PRIO
;
1690 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1691 per_cpu(rcu_cpu_kthread_task
, cpu
) = t
;
1692 wake_up_process(t
); /* Get to TASK_INTERRUPTIBLE quickly. */
1697 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1698 * kthreads when needed. We ignore requests to wake up kthreads
1699 * for offline CPUs, which is OK because force_quiescent_state()
1700 * takes care of this case.
1702 static int rcu_node_kthread(void *arg
)
1705 unsigned long flags
;
1707 struct rcu_node
*rnp
= (struct rcu_node
*)arg
;
1708 struct sched_param sp
;
1709 struct task_struct
*t
;
1712 rnp
->node_kthread_status
= RCU_KTHREAD_WAITING
;
1713 rcu_wait(atomic_read(&rnp
->wakemask
) != 0);
1714 rnp
->node_kthread_status
= RCU_KTHREAD_RUNNING
;
1715 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1716 mask
= atomic_xchg(&rnp
->wakemask
, 0);
1717 rcu_initiate_boost(rnp
, flags
); /* releases rnp->lock. */
1718 for (cpu
= rnp
->grplo
; cpu
<= rnp
->grphi
; cpu
++, mask
>>= 1) {
1719 if ((mask
& 0x1) == 0)
1722 t
= per_cpu(rcu_cpu_kthread_task
, cpu
);
1723 if (!cpu_online(cpu
) || t
== NULL
) {
1727 per_cpu(rcu_cpu_has_work
, cpu
) = 1;
1728 sp
.sched_priority
= RCU_KTHREAD_PRIO
;
1729 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1734 rnp
->node_kthread_status
= RCU_KTHREAD_STOPPED
;
1739 * Set the per-rcu_node kthread's affinity to cover all CPUs that are
1740 * served by the rcu_node in question. The CPU hotplug lock is still
1741 * held, so the value of rnp->qsmaskinit will be stable.
1743 * We don't include outgoingcpu in the affinity set, use -1 if there is
1744 * no outgoing CPU. If there are no CPUs left in the affinity set,
1745 * this function allows the kthread to execute on any CPU.
1747 static void rcu_node_kthread_setaffinity(struct rcu_node
*rnp
, int outgoingcpu
)
1751 unsigned long mask
= rnp
->qsmaskinit
;
1753 if (rnp
->node_kthread_task
== NULL
)
1755 if (!alloc_cpumask_var(&cm
, GFP_KERNEL
))
1758 for (cpu
= rnp
->grplo
; cpu
<= rnp
->grphi
; cpu
++, mask
>>= 1)
1759 if ((mask
& 0x1) && cpu
!= outgoingcpu
)
1760 cpumask_set_cpu(cpu
, cm
);
1761 if (cpumask_weight(cm
) == 0) {
1763 for (cpu
= rnp
->grplo
; cpu
<= rnp
->grphi
; cpu
++)
1764 cpumask_clear_cpu(cpu
, cm
);
1765 WARN_ON_ONCE(cpumask_weight(cm
) == 0);
1767 set_cpus_allowed_ptr(rnp
->node_kthread_task
, cm
);
1768 rcu_boost_kthread_setaffinity(rnp
, cm
);
1769 free_cpumask_var(cm
);
1773 * Spawn a per-rcu_node kthread, setting priority and affinity.
1774 * Called during boot before online/offline can happen, or, if
1775 * during runtime, with the main CPU-hotplug locks held. So only
1776 * one of these can be executing at a time.
1778 static int __cpuinit
rcu_spawn_one_node_kthread(struct rcu_state
*rsp
,
1779 struct rcu_node
*rnp
)
1781 unsigned long flags
;
1782 int rnp_index
= rnp
- &rsp
->node
[0];
1783 struct sched_param sp
;
1784 struct task_struct
*t
;
1786 if (!rcu_scheduler_fully_active
||
1787 rnp
->qsmaskinit
== 0)
1789 if (rnp
->node_kthread_task
== NULL
) {
1790 t
= kthread_create(rcu_node_kthread
, (void *)rnp
,
1791 "rcun/%d", rnp_index
);
1794 raw_spin_lock_irqsave(&rnp
->lock
, flags
);
1795 rnp
->node_kthread_task
= t
;
1796 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1797 sp
.sched_priority
= 99;
1798 sched_setscheduler_nocheck(t
, SCHED_FIFO
, &sp
);
1799 wake_up_process(t
); /* get to TASK_INTERRUPTIBLE quickly. */
1801 return rcu_spawn_one_boost_kthread(rsp
, rnp
, rnp_index
);
1805 * Spawn all kthreads -- called as soon as the scheduler is running.
1807 static int __init
rcu_spawn_kthreads(void)
1810 struct rcu_node
*rnp
;
1812 rcu_scheduler_fully_active
= 1;
1813 for_each_possible_cpu(cpu
) {
1814 per_cpu(rcu_cpu_has_work
, cpu
) = 0;
1815 if (cpu_online(cpu
))
1816 (void)rcu_spawn_one_cpu_kthread(cpu
);
1818 rnp
= rcu_get_root(rcu_state
);
1819 (void)rcu_spawn_one_node_kthread(rcu_state
, rnp
);
1820 if (NUM_RCU_NODES
> 1) {
1821 rcu_for_each_leaf_node(rcu_state
, rnp
)
1822 (void)rcu_spawn_one_node_kthread(rcu_state
, rnp
);
1826 early_initcall(rcu_spawn_kthreads
);
1828 static void __cpuinit
rcu_prepare_kthreads(int cpu
)
1830 struct rcu_data
*rdp
= per_cpu_ptr(rcu_state
->rda
, cpu
);
1831 struct rcu_node
*rnp
= rdp
->mynode
;
1833 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1834 if (rcu_scheduler_fully_active
) {
1835 (void)rcu_spawn_one_cpu_kthread(cpu
);
1836 if (rnp
->node_kthread_task
== NULL
)
1837 (void)rcu_spawn_one_node_kthread(rcu_state
, rnp
);
1841 #else /* #ifdef CONFIG_RCU_BOOST */
1843 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
)
1845 raw_spin_unlock_irqrestore(&rnp
->lock
, flags
);
1848 static void invoke_rcu_callbacks_kthread(void)
1853 static bool rcu_is_callbacks_kthread(void)
1858 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
)
1862 #ifdef CONFIG_HOTPLUG_CPU
1864 static void rcu_stop_cpu_kthread(int cpu
)
1868 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
1870 static void rcu_node_kthread_setaffinity(struct rcu_node
*rnp
, int outgoingcpu
)
1874 static void rcu_cpu_kthread_setrt(int cpu
, int to_rt
)
1878 static int __init
rcu_scheduler_really_started(void)
1880 rcu_scheduler_fully_active
= 1;
1883 early_initcall(rcu_scheduler_really_started
);
1885 static void __cpuinit
rcu_prepare_kthreads(int cpu
)
1889 #endif /* #else #ifdef CONFIG_RCU_BOOST */
1891 #if !defined(CONFIG_RCU_FAST_NO_HZ)
1894 * Check to see if any future RCU-related work will need to be done
1895 * by the current CPU, even if none need be done immediately, returning
1896 * 1 if so. This function is part of the RCU implementation; it is -not-
1897 * an exported member of the RCU API.
1899 * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs
1900 * any flavor of RCU.
1902 int rcu_needs_cpu(int cpu
, unsigned long *delta_jiffies
)
1904 *delta_jiffies
= ULONG_MAX
;
1905 return rcu_cpu_has_callbacks(cpu
);
1909 * Because we do not have RCU_FAST_NO_HZ, don't bother initializing for it.
1911 static void rcu_prepare_for_idle_init(int cpu
)
1916 * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up
1919 static void rcu_cleanup_after_idle(int cpu
)
1924 * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n,
1927 static void rcu_prepare_for_idle(int cpu
)
1932 * Don't bother keeping a running count of the number of RCU callbacks
1933 * posted because CONFIG_RCU_FAST_NO_HZ=n.
1935 static void rcu_idle_count_callbacks_posted(void)
1939 #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */
1942 * This code is invoked when a CPU goes idle, at which point we want
1943 * to have the CPU do everything required for RCU so that it can enter
1944 * the energy-efficient dyntick-idle mode. This is handled by a
1945 * state machine implemented by rcu_prepare_for_idle() below.
1947 * The following three proprocessor symbols control this state machine:
1949 * RCU_IDLE_FLUSHES gives the maximum number of times that we will attempt
1950 * to satisfy RCU. Beyond this point, it is better to incur a periodic
1951 * scheduling-clock interrupt than to loop through the state machine
1953 * RCU_IDLE_OPT_FLUSHES gives the number of RCU_IDLE_FLUSHES that are
1954 * optional if RCU does not need anything immediately from this
1955 * CPU, even if this CPU still has RCU callbacks queued. The first
1956 * times through the state machine are mandatory: we need to give
1957 * the state machine a chance to communicate a quiescent state
1959 * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted
1960 * to sleep in dyntick-idle mode with RCU callbacks pending. This
1961 * is sized to be roughly one RCU grace period. Those energy-efficiency
1962 * benchmarkers who might otherwise be tempted to set this to a large
1963 * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your
1964 * system. And if you are -that- concerned about energy efficiency,
1965 * just power the system down and be done with it!
1966 * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is
1967 * permitted to sleep in dyntick-idle mode with only lazy RCU
1968 * callbacks pending. Setting this too high can OOM your system.
1970 * The values below work well in practice. If future workloads require
1971 * adjustment, they can be converted into kernel config parameters, though
1972 * making the state machine smarter might be a better option.
1974 #define RCU_IDLE_FLUSHES 5 /* Number of dyntick-idle tries. */
1975 #define RCU_IDLE_OPT_FLUSHES 3 /* Optional dyntick-idle tries. */
1976 #define RCU_IDLE_GP_DELAY 6 /* Roughly one grace period. */
1977 #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */
1980 * Does the specified flavor of RCU have non-lazy callbacks pending on
1981 * the specified CPU? Both RCU flavor and CPU are specified by the
1982 * rcu_data structure.
1984 static bool __rcu_cpu_has_nonlazy_callbacks(struct rcu_data
*rdp
)
1986 return rdp
->qlen
!= rdp
->qlen_lazy
;
1989 #ifdef CONFIG_TREE_PREEMPT_RCU
1992 * Are there non-lazy RCU-preempt callbacks? (There cannot be if there
1993 * is no RCU-preempt in the kernel.)
1995 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu
)
1997 struct rcu_data
*rdp
= &per_cpu(rcu_preempt_data
, cpu
);
1999 return __rcu_cpu_has_nonlazy_callbacks(rdp
);
2002 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2004 static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu
)
2009 #endif /* else #ifdef CONFIG_TREE_PREEMPT_RCU */
2012 * Does any flavor of RCU have non-lazy callbacks on the specified CPU?
2014 static bool rcu_cpu_has_nonlazy_callbacks(int cpu
)
2016 return __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_sched_data
, cpu
)) ||
2017 __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_bh_data
, cpu
)) ||
2018 rcu_preempt_cpu_has_nonlazy_callbacks(cpu
);
2022 * Allow the CPU to enter dyntick-idle mode if either: (1) There are no
2023 * callbacks on this CPU, (2) this CPU has not yet attempted to enter
2024 * dyntick-idle mode, or (3) this CPU is in the process of attempting to
2025 * enter dyntick-idle mode. Otherwise, if we have recently tried and failed
2026 * to enter dyntick-idle mode, we refuse to try to enter it. After all,
2027 * it is better to incur scheduling-clock interrupts than to spin
2028 * continuously for the same time duration!
2030 * The delta_jiffies argument is used to store the time when RCU is
2031 * going to need the CPU again if it still has callbacks. The reason
2032 * for this is that rcu_prepare_for_idle() might need to post a timer,
2033 * but if so, it will do so after tick_nohz_stop_sched_tick() has set
2034 * the wakeup time for this CPU. This means that RCU's timer can be
2035 * delayed until the wakeup time, which defeats the purpose of posting
2038 int rcu_needs_cpu(int cpu
, unsigned long *delta_jiffies
)
2040 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
2042 /* Flag a new idle sojourn to the idle-entry state machine. */
2043 rdtp
->idle_first_pass
= 1;
2044 /* If no callbacks, RCU doesn't need the CPU. */
2045 if (!rcu_cpu_has_callbacks(cpu
)) {
2046 *delta_jiffies
= ULONG_MAX
;
2049 if (rdtp
->dyntick_holdoff
== jiffies
) {
2050 /* RCU recently tried and failed, so don't try again. */
2054 /* Set up for the possibility that RCU will post a timer. */
2055 if (rcu_cpu_has_nonlazy_callbacks(cpu
))
2056 *delta_jiffies
= RCU_IDLE_GP_DELAY
;
2058 *delta_jiffies
= RCU_IDLE_LAZY_GP_DELAY
;
2063 * Handler for smp_call_function_single(). The only point of this
2064 * handler is to wake the CPU up, so the handler does only tracing.
2066 void rcu_idle_demigrate(void *unused
)
2068 trace_rcu_prep_idle("Demigrate");
2072 * Timer handler used to force CPU to start pushing its remaining RCU
2073 * callbacks in the case where it entered dyntick-idle mode with callbacks
2074 * pending. The hander doesn't really need to do anything because the
2075 * real work is done upon re-entry to idle, or by the next scheduling-clock
2076 * interrupt should idle not be re-entered.
2078 * One special case: the timer gets migrated without awakening the CPU
2079 * on which the timer was scheduled on. In this case, we must wake up
2080 * that CPU. We do so with smp_call_function_single().
2082 static void rcu_idle_gp_timer_func(unsigned long cpu_in
)
2084 int cpu
= (int)cpu_in
;
2086 trace_rcu_prep_idle("Timer");
2087 if (cpu
!= smp_processor_id())
2088 smp_call_function_single(cpu
, rcu_idle_demigrate
, NULL
, 0);
2090 WARN_ON_ONCE(1); /* Getting here can hang the system... */
2094 * Initialize the timer used to pull CPUs out of dyntick-idle mode.
2096 static void rcu_prepare_for_idle_init(int cpu
)
2098 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
2100 rdtp
->dyntick_holdoff
= jiffies
- 1;
2101 setup_timer(&rdtp
->idle_gp_timer
, rcu_idle_gp_timer_func
, cpu
);
2102 rdtp
->idle_gp_timer_expires
= jiffies
- 1;
2103 rdtp
->idle_first_pass
= 1;
2107 * Clean up for exit from idle. Because we are exiting from idle, there
2108 * is no longer any point to ->idle_gp_timer, so cancel it. This will
2109 * do nothing if this timer is not active, so just cancel it unconditionally.
2111 static void rcu_cleanup_after_idle(int cpu
)
2113 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
2115 del_timer(&rdtp
->idle_gp_timer
);
2116 trace_rcu_prep_idle("Cleanup after idle");
2120 * Check to see if any RCU-related work can be done by the current CPU,
2121 * and if so, schedule a softirq to get it done. This function is part
2122 * of the RCU implementation; it is -not- an exported member of the RCU API.
2124 * The idea is for the current CPU to clear out all work required by the
2125 * RCU core for the current grace period, so that this CPU can be permitted
2126 * to enter dyntick-idle mode. In some cases, it will need to be awakened
2127 * at the end of the grace period by whatever CPU ends the grace period.
2128 * This allows CPUs to go dyntick-idle more quickly, and to reduce the
2129 * number of wakeups by a modest integer factor.
2131 * Because it is not legal to invoke rcu_process_callbacks() with irqs
2132 * disabled, we do one pass of force_quiescent_state(), then do a
2133 * invoke_rcu_core() to cause rcu_process_callbacks() to be invoked
2134 * later. The ->dyntick_drain field controls the sequencing.
2136 * The caller must have disabled interrupts.
2138 static void rcu_prepare_for_idle(int cpu
)
2140 struct timer_list
*tp
;
2141 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
2144 * If this is an idle re-entry, for example, due to use of
2145 * RCU_NONIDLE() or the new idle-loop tracing API within the idle
2146 * loop, then don't take any state-machine actions, unless the
2147 * momentary exit from idle queued additional non-lazy callbacks.
2148 * Instead, repost the ->idle_gp_timer if this CPU has callbacks
2151 if (!rdtp
->idle_first_pass
&&
2152 (rdtp
->nonlazy_posted
== rdtp
->nonlazy_posted_snap
)) {
2153 if (rcu_cpu_has_callbacks(cpu
)) {
2154 tp
= &rdtp
->idle_gp_timer
;
2155 mod_timer_pinned(tp
, rdtp
->idle_gp_timer_expires
);
2159 rdtp
->idle_first_pass
= 0;
2160 rdtp
->nonlazy_posted_snap
= rdtp
->nonlazy_posted
- 1;
2163 * If there are no callbacks on this CPU, enter dyntick-idle mode.
2164 * Also reset state to avoid prejudicing later attempts.
2166 if (!rcu_cpu_has_callbacks(cpu
)) {
2167 rdtp
->dyntick_holdoff
= jiffies
- 1;
2168 rdtp
->dyntick_drain
= 0;
2169 trace_rcu_prep_idle("No callbacks");
2174 * If in holdoff mode, just return. We will presumably have
2175 * refrained from disabling the scheduling-clock tick.
2177 if (rdtp
->dyntick_holdoff
== jiffies
) {
2178 trace_rcu_prep_idle("In holdoff");
2182 /* Check and update the ->dyntick_drain sequencing. */
2183 if (rdtp
->dyntick_drain
<= 0) {
2184 /* First time through, initialize the counter. */
2185 rdtp
->dyntick_drain
= RCU_IDLE_FLUSHES
;
2186 } else if (rdtp
->dyntick_drain
<= RCU_IDLE_OPT_FLUSHES
&&
2187 !rcu_pending(cpu
) &&
2188 !local_softirq_pending()) {
2189 /* Can we go dyntick-idle despite still having callbacks? */
2190 rdtp
->dyntick_drain
= 0;
2191 rdtp
->dyntick_holdoff
= jiffies
;
2192 if (rcu_cpu_has_nonlazy_callbacks(cpu
)) {
2193 trace_rcu_prep_idle("Dyntick with callbacks");
2194 rdtp
->idle_gp_timer_expires
=
2195 jiffies
+ RCU_IDLE_GP_DELAY
;
2197 rdtp
->idle_gp_timer_expires
=
2198 jiffies
+ RCU_IDLE_LAZY_GP_DELAY
;
2199 trace_rcu_prep_idle("Dyntick with lazy callbacks");
2201 tp
= &rdtp
->idle_gp_timer
;
2202 mod_timer_pinned(tp
, rdtp
->idle_gp_timer_expires
);
2203 rdtp
->nonlazy_posted_snap
= rdtp
->nonlazy_posted
;
2204 return; /* Nothing more to do immediately. */
2205 } else if (--(rdtp
->dyntick_drain
) <= 0) {
2206 /* We have hit the limit, so time to give up. */
2207 rdtp
->dyntick_holdoff
= jiffies
;
2208 trace_rcu_prep_idle("Begin holdoff");
2209 invoke_rcu_core(); /* Force the CPU out of dyntick-idle. */
2214 * Do one step of pushing the remaining RCU callbacks through
2215 * the RCU core state machine.
2217 #ifdef CONFIG_TREE_PREEMPT_RCU
2218 if (per_cpu(rcu_preempt_data
, cpu
).nxtlist
) {
2219 rcu_preempt_qs(cpu
);
2220 force_quiescent_state(&rcu_preempt_state
, 0);
2222 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2223 if (per_cpu(rcu_sched_data
, cpu
).nxtlist
) {
2225 force_quiescent_state(&rcu_sched_state
, 0);
2227 if (per_cpu(rcu_bh_data
, cpu
).nxtlist
) {
2229 force_quiescent_state(&rcu_bh_state
, 0);
2233 * If RCU callbacks are still pending, RCU still needs this CPU.
2234 * So try forcing the callbacks through the grace period.
2236 if (rcu_cpu_has_callbacks(cpu
)) {
2237 trace_rcu_prep_idle("More callbacks");
2240 trace_rcu_prep_idle("Callbacks drained");
2244 * Keep a running count of the number of non-lazy callbacks posted
2245 * on this CPU. This running counter (which is never decremented) allows
2246 * rcu_prepare_for_idle() to detect when something out of the idle loop
2247 * posts a callback, even if an equal number of callbacks are invoked.
2248 * Of course, callbacks should only be posted from within a trace event
2249 * designed to be called from idle or from within RCU_NONIDLE().
2251 static void rcu_idle_count_callbacks_posted(void)
2253 __this_cpu_add(rcu_dynticks
.nonlazy_posted
, 1);
2256 #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
2258 #ifdef CONFIG_RCU_CPU_STALL_INFO
2260 #ifdef CONFIG_RCU_FAST_NO_HZ
2262 static void print_cpu_stall_fast_no_hz(char *cp
, int cpu
)
2264 struct rcu_dynticks
*rdtp
= &per_cpu(rcu_dynticks
, cpu
);
2265 struct timer_list
*tltp
= &rdtp
->idle_gp_timer
;
2267 sprintf(cp
, "drain=%d %c timer=%lu",
2268 rdtp
->dyntick_drain
,
2269 rdtp
->dyntick_holdoff
== jiffies
? 'H' : '.',
2270 timer_pending(tltp
) ? tltp
->expires
- jiffies
: -1);
2273 #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
2275 static void print_cpu_stall_fast_no_hz(char *cp
, int cpu
)
2279 #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */
2281 /* Initiate the stall-info list. */
2282 static void print_cpu_stall_info_begin(void)
2284 printk(KERN_CONT
"\n");
2288 * Print out diagnostic information for the specified stalled CPU.
2290 * If the specified CPU is aware of the current RCU grace period
2291 * (flavor specified by rsp), then print the number of scheduling
2292 * clock interrupts the CPU has taken during the time that it has
2293 * been aware. Otherwise, print the number of RCU grace periods
2294 * that this CPU is ignorant of, for example, "1" if the CPU was
2295 * aware of the previous grace period.
2297 * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info.
2299 static void print_cpu_stall_info(struct rcu_state
*rsp
, int cpu
)
2301 char fast_no_hz
[72];
2302 struct rcu_data
*rdp
= per_cpu_ptr(rsp
->rda
, cpu
);
2303 struct rcu_dynticks
*rdtp
= rdp
->dynticks
;
2305 unsigned long ticks_value
;
2307 if (rsp
->gpnum
== rdp
->gpnum
) {
2308 ticks_title
= "ticks this GP";
2309 ticks_value
= rdp
->ticks_this_gp
;
2311 ticks_title
= "GPs behind";
2312 ticks_value
= rsp
->gpnum
- rdp
->gpnum
;
2314 print_cpu_stall_fast_no_hz(fast_no_hz
, cpu
);
2315 printk(KERN_ERR
"\t%d: (%lu %s) idle=%03x/%llx/%d %s\n",
2316 cpu
, ticks_value
, ticks_title
,
2317 atomic_read(&rdtp
->dynticks
) & 0xfff,
2318 rdtp
->dynticks_nesting
, rdtp
->dynticks_nmi_nesting
,
2322 /* Terminate the stall-info list. */
2323 static void print_cpu_stall_info_end(void)
2325 printk(KERN_ERR
"\t");
2328 /* Zero ->ticks_this_gp for all flavors of RCU. */
2329 static void zero_cpu_stall_ticks(struct rcu_data
*rdp
)
2331 rdp
->ticks_this_gp
= 0;
2334 /* Increment ->ticks_this_gp for all flavors of RCU. */
2335 static void increment_cpu_stall_ticks(void)
2337 __get_cpu_var(rcu_sched_data
).ticks_this_gp
++;
2338 __get_cpu_var(rcu_bh_data
).ticks_this_gp
++;
2339 #ifdef CONFIG_TREE_PREEMPT_RCU
2340 __get_cpu_var(rcu_preempt_data
).ticks_this_gp
++;
2341 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2344 #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
2346 static void print_cpu_stall_info_begin(void)
2348 printk(KERN_CONT
" {");
2351 static void print_cpu_stall_info(struct rcu_state
*rsp
, int cpu
)
2353 printk(KERN_CONT
" %d", cpu
);
2356 static void print_cpu_stall_info_end(void)
2358 printk(KERN_CONT
"} ");
2361 static void zero_cpu_stall_ticks(struct rcu_data
*rdp
)
2365 static void increment_cpu_stall_ticks(void)
2369 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */