2 * Read-Copy Update mechanism for mutual exclusion
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright IBM Corporation, 2008
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
27 * For detailed explanation of Read-Copy Update mechanism see -
30 #include <linux/types.h>
31 #include <linux/kernel.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/smp.h>
35 #include <linux/rcupdate.h>
36 #include <linux/interrupt.h>
37 #include <linux/sched.h>
38 #include <linux/nmi.h>
39 #include <asm/atomic.h>
40 #include <linux/bitops.h>
41 #include <linux/module.h>
42 #include <linux/completion.h>
43 #include <linux/moduleparam.h>
44 #include <linux/percpu.h>
45 #include <linux/notifier.h>
46 #include <linux/cpu.h>
47 #include <linux/mutex.h>
48 #include <linux/time.h>
52 #ifdef CONFIG_DEBUG_LOCK_ALLOC
53 static struct lock_class_key rcu_lock_key;
54 struct lockdep_map rcu_lock_map =
55 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
56 EXPORT_SYMBOL_GPL(rcu_lock_map);
59 /* Data structures. */
61 #define RCU_STATE_INITIALIZER(name) { \
62 .level = { &name.node[0] }, \
64 NUM_RCU_LVL_0, /* root of hierarchy. */ \
67 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
69 .signaled = RCU_SIGNAL_INIT, \
72 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
73 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
75 .n_force_qs_ngp = 0, \
78 struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
79 DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
81 struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
82 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
86 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
87 * permit this function to be invoked without holding the root rcu_node
88 * structure's ->lock, but of course results can be subject to change.
90 static int rcu_gp_in_progress(struct rcu_state *rsp)
92 return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
96 * Note a quiescent state. Because we do not need to know
97 * how many quiescent states passed, just if there was at least
98 * one since the start of the grace period, this just sets a flag.
100 void rcu_sched_qs(int cpu)
102 struct rcu_data *rdp;
104 rdp = &per_cpu(rcu_sched_data, cpu);
105 rdp->passed_quiesc_completed = rdp->completed;
107 rdp->passed_quiesc = 1;
108 rcu_preempt_note_context_switch(cpu);
111 void rcu_bh_qs(int cpu)
113 struct rcu_data *rdp;
115 rdp = &per_cpu(rcu_bh_data, cpu);
116 rdp->passed_quiesc_completed = rdp->completed;
118 rdp->passed_quiesc = 1;
122 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
123 .dynticks_nesting = 1,
126 #endif /* #ifdef CONFIG_NO_HZ */
128 static int blimit = 10; /* Maximum callbacks per softirq. */
129 static int qhimark = 10000; /* If this many pending, ignore blimit. */
130 static int qlowmark = 100; /* Once only this many pending, use blimit. */
132 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
133 static int rcu_pending(int cpu);
136 * Return the number of RCU-sched batches processed thus far for debug & stats.
138 long rcu_batches_completed_sched(void)
140 return rcu_sched_state.completed;
142 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
145 * Return the number of RCU BH batches processed thus far for debug & stats.
147 long rcu_batches_completed_bh(void)
149 return rcu_bh_state.completed;
151 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
154 * Does the CPU have callbacks ready to be invoked?
157 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
159 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
163 * Does the current CPU require a yet-as-unscheduled grace period?
166 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
168 return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
172 * Return the root node of the specified rcu_state structure.
174 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
176 return &rsp->node[0];
182 * If the specified CPU is offline, tell the caller that it is in
183 * a quiescent state. Otherwise, whack it with a reschedule IPI.
184 * Grace periods can end up waiting on an offline CPU when that
185 * CPU is in the process of coming online -- it will be added to the
186 * rcu_node bitmasks before it actually makes it online. The same thing
187 * can happen while a CPU is in the process of coming online. Because this
188 * race is quite rare, we check for it after detecting that the grace
189 * period has been delayed rather than checking each and every CPU
190 * each and every time we start a new grace period.
192 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
195 * If the CPU is offline, it is in a quiescent state. We can
196 * trust its state not to change because interrupts are disabled.
198 if (cpu_is_offline(rdp->cpu)) {
203 /* If preemptable RCU, no point in sending reschedule IPI. */
204 if (rdp->preemptable)
207 /* The CPU is online, so send it a reschedule IPI. */
208 if (rdp->cpu != smp_processor_id())
209 smp_send_reschedule(rdp->cpu);
216 #endif /* #ifdef CONFIG_SMP */
221 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
223 * Enter nohz mode, in other words, -leave- the mode in which RCU
224 * read-side critical sections can occur. (Though RCU read-side
225 * critical sections can occur in irq handlers in nohz mode, a possibility
226 * handled by rcu_irq_enter() and rcu_irq_exit()).
228 void rcu_enter_nohz(void)
231 struct rcu_dynticks *rdtp;
233 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
234 local_irq_save(flags);
235 rdtp = &__get_cpu_var(rcu_dynticks);
237 rdtp->dynticks_nesting--;
238 WARN_ON_ONCE(rdtp->dynticks & 0x1);
239 local_irq_restore(flags);
243 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
245 * Exit nohz mode, in other words, -enter- the mode in which RCU
246 * read-side critical sections normally occur.
248 void rcu_exit_nohz(void)
251 struct rcu_dynticks *rdtp;
253 local_irq_save(flags);
254 rdtp = &__get_cpu_var(rcu_dynticks);
256 rdtp->dynticks_nesting++;
257 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
258 local_irq_restore(flags);
259 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
263 * rcu_nmi_enter - inform RCU of entry to NMI context
265 * If the CPU was idle with dynamic ticks active, and there is no
266 * irq handler running, this updates rdtp->dynticks_nmi to let the
267 * RCU grace-period handling know that the CPU is active.
269 void rcu_nmi_enter(void)
271 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
273 if (rdtp->dynticks & 0x1)
275 rdtp->dynticks_nmi++;
276 WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1));
277 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
281 * rcu_nmi_exit - inform RCU of exit from NMI context
283 * If the CPU was idle with dynamic ticks active, and there is no
284 * irq handler running, this updates rdtp->dynticks_nmi to let the
285 * RCU grace-period handling know that the CPU is no longer active.
287 void rcu_nmi_exit(void)
289 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
291 if (rdtp->dynticks & 0x1)
293 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
294 rdtp->dynticks_nmi++;
295 WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1);
299 * rcu_irq_enter - inform RCU of entry to hard irq context
301 * If the CPU was idle with dynamic ticks active, this updates the
302 * rdtp->dynticks to let the RCU handling know that the CPU is active.
304 void rcu_irq_enter(void)
306 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
308 if (rdtp->dynticks_nesting++)
311 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
312 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
316 * rcu_irq_exit - inform RCU of exit from hard irq context
318 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
319 * to put let the RCU handling be aware that the CPU is going back to idle
322 void rcu_irq_exit(void)
324 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
326 if (--rdtp->dynticks_nesting)
328 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
330 WARN_ON_ONCE(rdtp->dynticks & 0x1);
332 /* If the interrupt queued a callback, get out of dyntick mode. */
333 if (__get_cpu_var(rcu_sched_data).nxtlist ||
334 __get_cpu_var(rcu_bh_data).nxtlist)
339 * Record the specified "completed" value, which is later used to validate
340 * dynticks counter manipulations. Specify "rsp->completed - 1" to
341 * unconditionally invalidate any future dynticks manipulations (which is
342 * useful at the beginning of a grace period).
344 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
346 rsp->dynticks_completed = comp;
352 * Recall the previously recorded value of the completion for dynticks.
354 static long dyntick_recall_completed(struct rcu_state *rsp)
356 return rsp->dynticks_completed;
360 * Snapshot the specified CPU's dynticks counter so that we can later
361 * credit them with an implicit quiescent state. Return 1 if this CPU
362 * is in dynticks idle mode, which is an extended quiescent state.
364 static int dyntick_save_progress_counter(struct rcu_data *rdp)
370 snap = rdp->dynticks->dynticks;
371 snap_nmi = rdp->dynticks->dynticks_nmi;
372 smp_mb(); /* Order sampling of snap with end of grace period. */
373 rdp->dynticks_snap = snap;
374 rdp->dynticks_nmi_snap = snap_nmi;
375 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
382 * Return true if the specified CPU has passed through a quiescent
383 * state by virtue of being in or having passed through an dynticks
384 * idle state since the last call to dyntick_save_progress_counter()
387 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
394 curr = rdp->dynticks->dynticks;
395 snap = rdp->dynticks_snap;
396 curr_nmi = rdp->dynticks->dynticks_nmi;
397 snap_nmi = rdp->dynticks_nmi_snap;
398 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
401 * If the CPU passed through or entered a dynticks idle phase with
402 * no active irq/NMI handlers, then we can safely pretend that the CPU
403 * already acknowledged the request to pass through a quiescent
404 * state. Either way, that CPU cannot possibly be in an RCU
405 * read-side critical section that started before the beginning
406 * of the current RCU grace period.
408 if ((curr != snap || (curr & 0x1) == 0) &&
409 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
414 /* Go check for the CPU being offline. */
415 return rcu_implicit_offline_qs(rdp);
418 #endif /* #ifdef CONFIG_SMP */
420 #else /* #ifdef CONFIG_NO_HZ */
422 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
429 * If there are no dynticks, then the only way that a CPU can passively
430 * be in a quiescent state is to be offline. Unlike dynticks idle, which
431 * is a point in time during the prior (already finished) grace period,
432 * an offline CPU is always in a quiescent state, and thus can be
433 * unconditionally applied. So just return the current value of completed.
435 static long dyntick_recall_completed(struct rcu_state *rsp)
437 return rsp->completed;
440 static int dyntick_save_progress_counter(struct rcu_data *rdp)
445 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
447 return rcu_implicit_offline_qs(rdp);
450 #endif /* #ifdef CONFIG_SMP */
452 #endif /* #else #ifdef CONFIG_NO_HZ */
454 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
456 static void record_gp_stall_check_time(struct rcu_state *rsp)
458 rsp->gp_start = jiffies;
459 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
462 static void print_other_cpu_stall(struct rcu_state *rsp)
467 struct rcu_node *rnp = rcu_get_root(rsp);
468 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
469 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
471 /* Only let one CPU complain about others per time interval. */
473 spin_lock_irqsave(&rnp->lock, flags);
474 delta = jiffies - rsp->jiffies_stall;
475 if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
476 spin_unlock_irqrestore(&rnp->lock, flags);
479 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
480 spin_unlock_irqrestore(&rnp->lock, flags);
482 /* OK, time to rat on our buddy... */
484 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
485 for (; rnp_cur < rnp_end; rnp_cur++) {
486 rcu_print_task_stall(rnp);
487 if (rnp_cur->qsmask == 0)
489 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
490 if (rnp_cur->qsmask & (1UL << cpu))
491 printk(" %d", rnp_cur->grplo + cpu);
493 printk(" (detected by %d, t=%ld jiffies)\n",
494 smp_processor_id(), (long)(jiffies - rsp->gp_start));
495 trigger_all_cpu_backtrace();
497 force_quiescent_state(rsp, 0); /* Kick them all. */
500 static void print_cpu_stall(struct rcu_state *rsp)
503 struct rcu_node *rnp = rcu_get_root(rsp);
505 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
506 smp_processor_id(), jiffies - rsp->gp_start);
507 trigger_all_cpu_backtrace();
509 spin_lock_irqsave(&rnp->lock, flags);
510 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
512 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
513 spin_unlock_irqrestore(&rnp->lock, flags);
515 set_need_resched(); /* kick ourselves to get things going. */
518 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
521 struct rcu_node *rnp;
523 delta = jiffies - rsp->jiffies_stall;
525 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
527 /* We haven't checked in, so go dump stack. */
528 print_cpu_stall(rsp);
530 } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
532 /* They had two time units to dump stack, so complain. */
533 print_other_cpu_stall(rsp);
537 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
539 static void record_gp_stall_check_time(struct rcu_state *rsp)
543 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
547 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
550 * Update CPU-local rcu_data state to record the newly noticed grace period.
551 * This is used both when we started the grace period and when we notice
552 * that someone else started the grace period.
554 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
557 rdp->passed_quiesc = 0;
558 rdp->gpnum = rsp->gpnum;
562 * Did someone else start a new RCU grace period start since we last
563 * checked? Update local state appropriately if so. Must be called
564 * on the CPU corresponding to rdp.
567 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
572 local_irq_save(flags);
573 if (rdp->gpnum != rsp->gpnum) {
574 note_new_gpnum(rsp, rdp);
577 local_irq_restore(flags);
582 * Start a new RCU grace period if warranted, re-initializing the hierarchy
583 * in preparation for detecting the next grace period. The caller must hold
584 * the root node's ->lock, which is released before return. Hard irqs must
588 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
589 __releases(rcu_get_root(rsp)->lock)
591 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
592 struct rcu_node *rnp = rcu_get_root(rsp);
594 if (!cpu_needs_another_gp(rsp, rdp)) {
595 spin_unlock_irqrestore(&rnp->lock, flags);
599 /* Advance to a new grace period and initialize state. */
601 WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
602 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
603 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
604 record_gp_stall_check_time(rsp);
605 dyntick_record_completed(rsp, rsp->completed - 1);
606 note_new_gpnum(rsp, rdp);
609 * Because this CPU just now started the new grace period, we know
610 * that all of its callbacks will be covered by this upcoming grace
611 * period, even the ones that were registered arbitrarily recently.
612 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
614 * Other CPUs cannot be sure exactly when the grace period started.
615 * Therefore, their recently registered callbacks must pass through
616 * an additional RCU_NEXT_READY stage, so that they will be handled
617 * by the next RCU grace period.
619 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
620 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
622 /* Special-case the common single-level case. */
623 if (NUM_RCU_NODES == 1) {
624 rcu_preempt_check_blocked_tasks(rnp);
625 rnp->qsmask = rnp->qsmaskinit;
626 rnp->gpnum = rsp->gpnum;
627 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
628 spin_unlock_irqrestore(&rnp->lock, flags);
632 spin_unlock(&rnp->lock); /* leave irqs disabled. */
635 /* Exclude any concurrent CPU-hotplug operations. */
636 spin_lock(&rsp->onofflock); /* irqs already disabled. */
639 * Set the quiescent-state-needed bits in all the rcu_node
640 * structures for all currently online CPUs in breadth-first
641 * order, starting from the root rcu_node structure. This
642 * operation relies on the layout of the hierarchy within the
643 * rsp->node[] array. Note that other CPUs will access only
644 * the leaves of the hierarchy, which still indicate that no
645 * grace period is in progress, at least until the corresponding
646 * leaf node has been initialized. In addition, we have excluded
647 * CPU-hotplug operations.
649 * Note that the grace period cannot complete until we finish
650 * the initialization process, as there will be at least one
651 * qsmask bit set in the root node until that time, namely the
652 * one corresponding to this CPU, due to the fact that we have
655 for (rnp = &rsp->node[0]; rnp < &rsp->node[NUM_RCU_NODES]; rnp++) {
656 spin_lock(&rnp->lock); /* irqs already disabled. */
657 rcu_preempt_check_blocked_tasks(rnp);
658 rnp->qsmask = rnp->qsmaskinit;
659 rnp->gpnum = rsp->gpnum;
660 spin_unlock(&rnp->lock); /* irqs already disabled. */
663 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
664 spin_unlock_irqrestore(&rsp->onofflock, flags);
668 * Advance this CPU's callbacks, but only if the current grace period
669 * has ended. This may be called only from the CPU to whom the rdp
673 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
678 local_irq_save(flags);
679 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
681 /* Did another grace period end? */
682 if (rdp->completed != completed_snap) {
684 /* Advance callbacks. No harm if list empty. */
685 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
686 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
687 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
689 /* Remember that we saw this grace-period completion. */
690 rdp->completed = completed_snap;
692 local_irq_restore(flags);
696 * Clean up after the prior grace period and let rcu_start_gp() start up
697 * the next grace period if one is needed. Note that the caller must
698 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
700 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
701 __releases(rcu_get_root(rsp)->lock)
703 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
704 rsp->completed = rsp->gpnum;
705 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
706 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
710 * Similar to cpu_quiet(), for which it is a helper function. Allows
711 * a group of CPUs to be quieted at one go, though all the CPUs in the
712 * group must be represented by the same leaf rcu_node structure.
713 * That structure's lock must be held upon entry, and it is released
717 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
719 __releases(rnp->lock)
721 struct rcu_node *rnp_c;
723 /* Walk up the rcu_node hierarchy. */
725 if (!(rnp->qsmask & mask)) {
727 /* Our bit has already been cleared, so done. */
728 spin_unlock_irqrestore(&rnp->lock, flags);
731 rnp->qsmask &= ~mask;
732 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
734 /* Other bits still set at this level, so done. */
735 spin_unlock_irqrestore(&rnp->lock, flags);
739 if (rnp->parent == NULL) {
741 /* No more levels. Exit loop holding root lock. */
745 spin_unlock_irqrestore(&rnp->lock, flags);
748 spin_lock_irqsave(&rnp->lock, flags);
749 WARN_ON_ONCE(rnp_c->qsmask);
753 * Get here if we are the last CPU to pass through a quiescent
754 * state for this grace period. Invoke cpu_quiet_msk_finish()
755 * to clean up and start the next grace period if one is needed.
757 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
761 * Record a quiescent state for the specified CPU, which must either be
762 * the current CPU. The lastcomp argument is used to make sure we are
763 * still in the grace period of interest. We don't want to end the current
764 * grace period based on quiescent states detected in an earlier grace
768 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
772 struct rcu_node *rnp;
775 spin_lock_irqsave(&rnp->lock, flags);
776 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
779 * Someone beat us to it for this grace period, so leave.
780 * The race with GP start is resolved by the fact that we
781 * hold the leaf rcu_node lock, so that the per-CPU bits
782 * cannot yet be initialized -- so we would simply find our
783 * CPU's bit already cleared in cpu_quiet_msk() if this race
786 rdp->passed_quiesc = 0; /* try again later! */
787 spin_unlock_irqrestore(&rnp->lock, flags);
791 if ((rnp->qsmask & mask) == 0) {
792 spin_unlock_irqrestore(&rnp->lock, flags);
797 * This GP can't end until cpu checks in, so all of our
798 * callbacks can be processed during the next GP.
800 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
802 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
807 * Check to see if there is a new grace period of which this CPU
808 * is not yet aware, and if so, set up local rcu_data state for it.
809 * Otherwise, see if this CPU has just passed through its first
810 * quiescent state for this grace period, and record that fact if so.
813 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
815 /* If there is now a new grace period, record and return. */
816 if (check_for_new_grace_period(rsp, rdp))
820 * Does this CPU still need to do its part for current grace period?
821 * If no, return and let the other CPUs do their part as well.
823 if (!rdp->qs_pending)
827 * Was there a quiescent state since the beginning of the grace
828 * period? If no, then exit and wait for the next call.
830 if (!rdp->passed_quiesc)
833 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
834 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
837 #ifdef CONFIG_HOTPLUG_CPU
840 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
841 * and move all callbacks from the outgoing CPU to the current one.
843 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
849 struct rcu_data *rdp = rsp->rda[cpu];
850 struct rcu_data *rdp_me;
851 struct rcu_node *rnp;
853 /* Exclude any attempts to start a new grace period. */
854 spin_lock_irqsave(&rsp->onofflock, flags);
856 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
857 rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */
858 mask = rdp->grpmask; /* rnp->grplo is constant. */
860 spin_lock(&rnp->lock); /* irqs already disabled. */
861 rnp->qsmaskinit &= ~mask;
862 if (rnp->qsmaskinit != 0) {
863 spin_unlock(&rnp->lock); /* irqs remain disabled. */
866 rcu_preempt_offline_tasks(rsp, rnp, rdp);
868 spin_unlock(&rnp->lock); /* irqs remain disabled. */
870 } while (rnp != NULL);
871 lastcomp = rsp->completed;
873 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
876 * Move callbacks from the outgoing CPU to the running CPU.
877 * Note that the outgoing CPU is now quiescent, so it is now
878 * (uncharacteristically) safe to access its rcu_data structure.
879 * Note also that we must carefully retain the order of the
880 * outgoing CPU's callbacks in order for rcu_barrier() to work
881 * correctly. Finally, note that we start all the callbacks
882 * afresh, even those that have passed through a grace period
883 * and are therefore ready to invoke. The theory is that hotplug
884 * events are rare, and that if they are frequent enough to
885 * indefinitely delay callbacks, you have far worse things to
888 rdp_me = rsp->rda[smp_processor_id()];
889 if (rdp->nxtlist != NULL) {
890 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
891 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
893 for (i = 0; i < RCU_NEXT_SIZE; i++)
894 rdp->nxttail[i] = &rdp->nxtlist;
895 rdp_me->qlen += rdp->qlen;
898 local_irq_restore(flags);
902 * Remove the specified CPU from the RCU hierarchy and move any pending
903 * callbacks that it might have to the current CPU. This code assumes
904 * that at least one CPU in the system will remain running at all times.
905 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
907 static void rcu_offline_cpu(int cpu)
909 __rcu_offline_cpu(cpu, &rcu_sched_state);
910 __rcu_offline_cpu(cpu, &rcu_bh_state);
911 rcu_preempt_offline_cpu(cpu);
914 #else /* #ifdef CONFIG_HOTPLUG_CPU */
916 static void rcu_offline_cpu(int cpu)
920 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
923 * Invoke any RCU callbacks that have made it to the end of their grace
924 * period. Thottle as specified by rdp->blimit.
926 static void rcu_do_batch(struct rcu_data *rdp)
929 struct rcu_head *next, *list, **tail;
932 /* If no callbacks are ready, just return.*/
933 if (!cpu_has_callbacks_ready_to_invoke(rdp))
937 * Extract the list of ready callbacks, disabling to prevent
938 * races with call_rcu() from interrupt handlers.
940 local_irq_save(flags);
942 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
943 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
944 tail = rdp->nxttail[RCU_DONE_TAIL];
945 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
946 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
947 rdp->nxttail[count] = &rdp->nxtlist;
948 local_irq_restore(flags);
950 /* Invoke callbacks. */
957 if (++count >= rdp->blimit)
961 local_irq_save(flags);
963 /* Update count, and requeue any remaining callbacks. */
966 *tail = rdp->nxtlist;
968 for (count = 0; count < RCU_NEXT_SIZE; count++)
969 if (&rdp->nxtlist == rdp->nxttail[count])
970 rdp->nxttail[count] = tail;
975 /* Reinstate batch limit if we have worked down the excess. */
976 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
977 rdp->blimit = blimit;
979 local_irq_restore(flags);
981 /* Re-raise the RCU softirq if there are callbacks remaining. */
982 if (cpu_has_callbacks_ready_to_invoke(rdp))
983 raise_softirq(RCU_SOFTIRQ);
987 * Check to see if this CPU is in a non-context-switch quiescent state
988 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
989 * Also schedule the RCU softirq handler.
991 * This function must be called with hardirqs disabled. It is normally
992 * invoked from the scheduling-clock interrupt. If rcu_pending returns
993 * false, there is no point in invoking rcu_check_callbacks().
995 void rcu_check_callbacks(int cpu, int user)
997 if (!rcu_pending(cpu))
998 return; /* if nothing for RCU to do. */
1000 (idle_cpu(cpu) && rcu_scheduler_active &&
1001 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1004 * Get here if this CPU took its interrupt from user
1005 * mode or from the idle loop, and if this is not a
1006 * nested interrupt. In this case, the CPU is in
1007 * a quiescent state, so note it.
1009 * No memory barrier is required here because both
1010 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1011 * variables that other CPUs neither access nor modify,
1012 * at least not while the corresponding CPU is online.
1018 } else if (!in_softirq()) {
1021 * Get here if this CPU did not take its interrupt from
1022 * softirq, in other words, if it is not interrupting
1023 * a rcu_bh read-side critical section. This is an _bh
1024 * critical section, so note it.
1029 rcu_preempt_check_callbacks(cpu);
1030 raise_softirq(RCU_SOFTIRQ);
1036 * Scan the leaf rcu_node structures, processing dyntick state for any that
1037 * have not yet encountered a quiescent state, using the function specified.
1038 * Returns 1 if the current grace period ends while scanning (possibly
1039 * because we made it end).
1041 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1042 int (*f)(struct rcu_data *))
1046 unsigned long flags;
1048 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
1049 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1051 for (; rnp_cur < rnp_end; rnp_cur++) {
1053 spin_lock_irqsave(&rnp_cur->lock, flags);
1054 if (rsp->completed != lastcomp) {
1055 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1058 if (rnp_cur->qsmask == 0) {
1059 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1062 cpu = rnp_cur->grplo;
1064 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1065 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1068 if (mask != 0 && rsp->completed == lastcomp) {
1070 /* cpu_quiet_msk() releases rnp_cur->lock. */
1071 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1074 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1080 * Force quiescent states on reluctant CPUs, and also detect which
1081 * CPUs are in dyntick-idle mode.
1083 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1085 unsigned long flags;
1087 struct rcu_node *rnp = rcu_get_root(rsp);
1090 if (!rcu_gp_in_progress(rsp))
1091 return; /* No grace period in progress, nothing to force. */
1092 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1093 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1094 return; /* Someone else is already on the job. */
1097 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1098 goto unlock_ret; /* no emergency and done recently. */
1100 spin_lock(&rnp->lock);
1101 lastcomp = rsp->completed;
1102 signaled = rsp->signaled;
1103 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1104 if (lastcomp == rsp->gpnum) {
1105 rsp->n_force_qs_ngp++;
1106 spin_unlock(&rnp->lock);
1107 goto unlock_ret; /* no GP in progress, time updated. */
1109 spin_unlock(&rnp->lock);
1113 break; /* grace period still initializing, ignore. */
1115 case RCU_SAVE_DYNTICK:
1117 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1118 break; /* So gcc recognizes the dead code. */
1120 /* Record dyntick-idle state. */
1121 if (rcu_process_dyntick(rsp, lastcomp,
1122 dyntick_save_progress_counter))
1125 /* Update state, record completion counter. */
1126 spin_lock(&rnp->lock);
1127 if (lastcomp == rsp->completed) {
1128 rsp->signaled = RCU_FORCE_QS;
1129 dyntick_record_completed(rsp, lastcomp);
1131 spin_unlock(&rnp->lock);
1136 /* Check dyntick-idle state, send IPI to laggarts. */
1137 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1138 rcu_implicit_dynticks_qs))
1141 /* Leave state in case more forcing is required. */
1146 spin_unlock_irqrestore(&rsp->fqslock, flags);
1149 #else /* #ifdef CONFIG_SMP */
1151 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1156 #endif /* #else #ifdef CONFIG_SMP */
1159 * This does the RCU processing work from softirq context for the
1160 * specified rcu_state and rcu_data structures. This may be called
1161 * only from the CPU to whom the rdp belongs.
1164 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1166 unsigned long flags;
1168 WARN_ON_ONCE(rdp->beenonline == 0);
1171 * If an RCU GP has gone long enough, go check for dyntick
1172 * idle CPUs and, if needed, send resched IPIs.
1174 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1175 force_quiescent_state(rsp, 1);
1178 * Advance callbacks in response to end of earlier grace
1179 * period that some other CPU ended.
1181 rcu_process_gp_end(rsp, rdp);
1183 /* Update RCU state based on any recent quiescent states. */
1184 rcu_check_quiescent_state(rsp, rdp);
1186 /* Does this CPU require a not-yet-started grace period? */
1187 if (cpu_needs_another_gp(rsp, rdp)) {
1188 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1189 rcu_start_gp(rsp, flags); /* releases above lock */
1192 /* If there are callbacks ready, invoke them. */
1197 * Do softirq processing for the current CPU.
1199 static void rcu_process_callbacks(struct softirq_action *unused)
1202 * Memory references from any prior RCU read-side critical sections
1203 * executed by the interrupted code must be seen before any RCU
1204 * grace-period manipulations below.
1206 smp_mb(); /* See above block comment. */
1208 __rcu_process_callbacks(&rcu_sched_state,
1209 &__get_cpu_var(rcu_sched_data));
1210 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1211 rcu_preempt_process_callbacks();
1214 * Memory references from any later RCU read-side critical sections
1215 * executed by the interrupted code must be seen after any RCU
1216 * grace-period manipulations above.
1218 smp_mb(); /* See above block comment. */
1222 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1223 struct rcu_state *rsp)
1225 unsigned long flags;
1226 struct rcu_data *rdp;
1231 smp_mb(); /* Ensure RCU update seen before callback registry. */
1234 * Opportunistically note grace-period endings and beginnings.
1235 * Note that we might see a beginning right after we see an
1236 * end, but never vice versa, since this CPU has to pass through
1237 * a quiescent state betweentimes.
1239 local_irq_save(flags);
1240 rdp = rsp->rda[smp_processor_id()];
1241 rcu_process_gp_end(rsp, rdp);
1242 check_for_new_grace_period(rsp, rdp);
1244 /* Add the callback to our list. */
1245 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1246 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1248 /* Start a new grace period if one not already started. */
1249 if (!rcu_gp_in_progress(rsp)) {
1250 unsigned long nestflag;
1251 struct rcu_node *rnp_root = rcu_get_root(rsp);
1253 spin_lock_irqsave(&rnp_root->lock, nestflag);
1254 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1257 /* Force the grace period if too many callbacks or too long waiting. */
1258 if (unlikely(++rdp->qlen > qhimark)) {
1259 rdp->blimit = LONG_MAX;
1260 force_quiescent_state(rsp, 0);
1261 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1262 force_quiescent_state(rsp, 1);
1263 local_irq_restore(flags);
1267 * Queue an RCU-sched callback for invocation after a grace period.
1269 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1271 __call_rcu(head, func, &rcu_sched_state);
1273 EXPORT_SYMBOL_GPL(call_rcu_sched);
1276 * Queue an RCU for invocation after a quicker grace period.
1278 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1280 __call_rcu(head, func, &rcu_bh_state);
1282 EXPORT_SYMBOL_GPL(call_rcu_bh);
1285 * Check to see if there is any immediate RCU-related work to be done
1286 * by the current CPU, for the specified type of RCU, returning 1 if so.
1287 * The checks are in order of increasing expense: checks that can be
1288 * carried out against CPU-local state are performed first. However,
1289 * we must check for CPU stalls first, else we might not get a chance.
1291 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1293 rdp->n_rcu_pending++;
1295 /* Check for CPU stalls, if enabled. */
1296 check_cpu_stall(rsp, rdp);
1298 /* Is the RCU core waiting for a quiescent state from this CPU? */
1299 if (rdp->qs_pending) {
1300 rdp->n_rp_qs_pending++;
1304 /* Does this CPU have callbacks ready to invoke? */
1305 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1306 rdp->n_rp_cb_ready++;
1310 /* Has RCU gone idle with this CPU needing another grace period? */
1311 if (cpu_needs_another_gp(rsp, rdp)) {
1312 rdp->n_rp_cpu_needs_gp++;
1316 /* Has another RCU grace period completed? */
1317 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1318 rdp->n_rp_gp_completed++;
1322 /* Has a new RCU grace period started? */
1323 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1324 rdp->n_rp_gp_started++;
1328 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1329 if (rcu_gp_in_progress(rsp) &&
1330 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1331 rdp->n_rp_need_fqs++;
1336 rdp->n_rp_need_nothing++;
1341 * Check to see if there is any immediate RCU-related work to be done
1342 * by the current CPU, returning 1 if so. This function is part of the
1343 * RCU implementation; it is -not- an exported member of the RCU API.
1345 static int rcu_pending(int cpu)
1347 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1348 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1349 rcu_preempt_pending(cpu);
1353 * Check to see if any future RCU-related work will need to be done
1354 * by the current CPU, even if none need be done immediately, returning
1355 * 1 if so. This function is part of the RCU implementation; it is -not-
1356 * an exported member of the RCU API.
1358 int rcu_needs_cpu(int cpu)
1360 /* RCU callbacks either ready or pending? */
1361 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1362 per_cpu(rcu_bh_data, cpu).nxtlist ||
1363 rcu_preempt_needs_cpu(cpu);
1367 * Do boot-time initialization of a CPU's per-CPU RCU data.
1370 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1372 unsigned long flags;
1374 struct rcu_data *rdp = rsp->rda[cpu];
1375 struct rcu_node *rnp = rcu_get_root(rsp);
1377 /* Set up local state, ensuring consistent view of global state. */
1378 spin_lock_irqsave(&rnp->lock, flags);
1379 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1380 rdp->nxtlist = NULL;
1381 for (i = 0; i < RCU_NEXT_SIZE; i++)
1382 rdp->nxttail[i] = &rdp->nxtlist;
1385 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1386 #endif /* #ifdef CONFIG_NO_HZ */
1388 spin_unlock_irqrestore(&rnp->lock, flags);
1392 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1393 * offline event can be happening at a given time. Note also that we
1394 * can accept some slop in the rsp->completed access due to the fact
1395 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1397 static void __cpuinit
1398 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1400 unsigned long flags;
1403 struct rcu_data *rdp = rsp->rda[cpu];
1404 struct rcu_node *rnp = rcu_get_root(rsp);
1406 /* Set up local state, ensuring consistent view of global state. */
1407 spin_lock_irqsave(&rnp->lock, flags);
1408 lastcomp = rsp->completed;
1409 rdp->completed = lastcomp;
1410 rdp->gpnum = lastcomp;
1411 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1412 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1413 rdp->beenonline = 1; /* We have now been online. */
1414 rdp->preemptable = preemptable;
1415 rdp->passed_quiesc_completed = lastcomp - 1;
1416 rdp->blimit = blimit;
1417 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1420 * A new grace period might start here. If so, we won't be part
1421 * of it, but that is OK, as we are currently in a quiescent state.
1424 /* Exclude any attempts to start a new GP on large systems. */
1425 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1427 /* Add CPU to rcu_node bitmasks. */
1429 mask = rdp->grpmask;
1431 /* Exclude any attempts to start a new GP on small systems. */
1432 spin_lock(&rnp->lock); /* irqs already disabled. */
1433 rnp->qsmaskinit |= mask;
1434 mask = rnp->grpmask;
1435 spin_unlock(&rnp->lock); /* irqs already disabled. */
1437 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1439 spin_unlock_irqrestore(&rsp->onofflock, flags);
1442 static void __cpuinit rcu_online_cpu(int cpu)
1444 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1445 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1446 rcu_preempt_init_percpu_data(cpu);
1450 * Handle CPU online/offline notification events.
1452 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1453 unsigned long action, void *hcpu)
1455 long cpu = (long)hcpu;
1458 case CPU_UP_PREPARE:
1459 case CPU_UP_PREPARE_FROZEN:
1460 rcu_online_cpu(cpu);
1463 case CPU_DEAD_FROZEN:
1464 case CPU_UP_CANCELED:
1465 case CPU_UP_CANCELED_FROZEN:
1466 rcu_offline_cpu(cpu);
1475 * Compute the per-level fanout, either using the exact fanout specified
1476 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1478 #ifdef CONFIG_RCU_FANOUT_EXACT
1479 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1483 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1484 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1486 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1487 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1494 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1495 ccur = rsp->levelcnt[i];
1496 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1500 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1503 * Helper function for rcu_init() that initializes one rcu_state structure.
1505 static void __init rcu_init_one(struct rcu_state *rsp)
1510 struct rcu_node *rnp;
1512 /* Initialize the level-tracking arrays. */
1514 for (i = 1; i < NUM_RCU_LVLS; i++)
1515 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1516 rcu_init_levelspread(rsp);
1518 /* Initialize the elements themselves, starting from the leaves. */
1520 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1521 cpustride *= rsp->levelspread[i];
1522 rnp = rsp->level[i];
1523 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1524 spin_lock_init(&rnp->lock);
1527 rnp->qsmaskinit = 0;
1528 rnp->grplo = j * cpustride;
1529 rnp->grphi = (j + 1) * cpustride - 1;
1530 if (rnp->grphi >= NR_CPUS)
1531 rnp->grphi = NR_CPUS - 1;
1537 rnp->grpnum = j % rsp->levelspread[i - 1];
1538 rnp->grpmask = 1UL << rnp->grpnum;
1539 rnp->parent = rsp->level[i - 1] +
1540 j / rsp->levelspread[i - 1];
1543 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1544 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1550 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1551 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1554 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1556 rcu_init_one(rsp); \
1557 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1559 for_each_possible_cpu(i) { \
1560 if (i > rnp[j].grphi) \
1562 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1563 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1564 rcu_boot_init_percpu_data(i, rsp); \
1568 void __init __rcu_init(void)
1570 int i; /* All used by RCU_INIT_FLAVOR(). */
1572 struct rcu_node *rnp;
1574 rcu_bootup_announce();
1575 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1576 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1577 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1578 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1579 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1580 __rcu_init_preempt();
1581 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1584 #include "rcutree_plugin.h"
1586 module_param(blimit, int, 0);
1587 module_param(qhimark, int, 0);
1588 module_param(qlowmark, int, 0);