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 <asm/atomic.h>
39 #include <linux/bitops.h>
40 #include <linux/module.h>
41 #include <linux/completion.h>
42 #include <linux/moduleparam.h>
43 #include <linux/percpu.h>
44 #include <linux/notifier.h>
45 #include <linux/cpu.h>
46 #include <linux/mutex.h>
47 #include <linux/time.h>
51 #ifdef CONFIG_DEBUG_LOCK_ALLOC
52 static struct lock_class_key rcu_lock_key;
53 struct lockdep_map rcu_lock_map =
54 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
55 EXPORT_SYMBOL_GPL(rcu_lock_map);
58 /* Data structures. */
60 #define RCU_STATE_INITIALIZER(name) { \
61 .level = { &name.node[0] }, \
63 NUM_RCU_LVL_0, /* root of hierarchy. */ \
66 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
68 .signaled = RCU_SIGNAL_INIT, \
71 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
72 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
74 .n_force_qs_ngp = 0, \
77 struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
78 DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
80 struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
81 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
83 extern long rcu_batches_completed_sched(void);
84 static struct rcu_node *rcu_get_root(struct rcu_state *rsp);
85 static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp,
86 struct rcu_node *rnp, unsigned long flags);
87 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags);
88 #ifdef CONFIG_HOTPLUG_CPU
89 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp);
90 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
91 static void __rcu_process_callbacks(struct rcu_state *rsp,
92 struct rcu_data *rdp);
93 static void __call_rcu(struct rcu_head *head,
94 void (*func)(struct rcu_head *rcu),
95 struct rcu_state *rsp);
96 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp);
97 static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp,
100 #include "rcutree_plugin.h"
103 * Note a quiescent state. Because we do not need to know
104 * how many quiescent states passed, just if there was at least
105 * one since the start of the grace period, this just sets a flag.
107 void rcu_sched_qs(int cpu)
110 struct rcu_data *rdp;
112 local_irq_save(flags);
113 rdp = &per_cpu(rcu_sched_data, cpu);
114 rdp->passed_quiesc = 1;
115 rdp->passed_quiesc_completed = rdp->completed;
117 local_irq_restore(flags);
120 void rcu_bh_qs(int cpu)
123 struct rcu_data *rdp;
125 local_irq_save(flags);
126 rdp = &per_cpu(rcu_bh_data, cpu);
127 rdp->passed_quiesc = 1;
128 rdp->passed_quiesc_completed = rdp->completed;
129 local_irq_restore(flags);
133 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
134 .dynticks_nesting = 1,
137 #endif /* #ifdef CONFIG_NO_HZ */
139 static int blimit = 10; /* Maximum callbacks per softirq. */
140 static int qhimark = 10000; /* If this many pending, ignore blimit. */
141 static int qlowmark = 100; /* Once only this many pending, use blimit. */
143 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
144 static int rcu_pending(int cpu);
147 * Return the number of RCU-sched batches processed thus far for debug & stats.
149 long rcu_batches_completed_sched(void)
151 return rcu_sched_state.completed;
153 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
156 * Return the number of RCU BH batches processed thus far for debug & stats.
158 long rcu_batches_completed_bh(void)
160 return rcu_bh_state.completed;
162 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
165 * Does the CPU have callbacks ready to be invoked?
168 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
170 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
174 * Does the current CPU require a yet-as-unscheduled grace period?
177 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
179 /* ACCESS_ONCE() because we are accessing outside of lock. */
180 return *rdp->nxttail[RCU_DONE_TAIL] &&
181 ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
185 * Return the root node of the specified rcu_state structure.
187 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
189 return &rsp->node[0];
195 * If the specified CPU is offline, tell the caller that it is in
196 * a quiescent state. Otherwise, whack it with a reschedule IPI.
197 * Grace periods can end up waiting on an offline CPU when that
198 * CPU is in the process of coming online -- it will be added to the
199 * rcu_node bitmasks before it actually makes it online. The same thing
200 * can happen while a CPU is in the process of coming online. Because this
201 * race is quite rare, we check for it after detecting that the grace
202 * period has been delayed rather than checking each and every CPU
203 * each and every time we start a new grace period.
205 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
208 * If the CPU is offline, it is in a quiescent state. We can
209 * trust its state not to change because interrupts are disabled.
211 if (cpu_is_offline(rdp->cpu)) {
216 /* If preemptable RCU, no point in sending reschedule IPI. */
217 if (rdp->preemptable)
220 /* The CPU is online, so send it a reschedule IPI. */
221 if (rdp->cpu != smp_processor_id())
222 smp_send_reschedule(rdp->cpu);
229 #endif /* #ifdef CONFIG_SMP */
234 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
236 * Enter nohz mode, in other words, -leave- the mode in which RCU
237 * read-side critical sections can occur. (Though RCU read-side
238 * critical sections can occur in irq handlers in nohz mode, a possibility
239 * handled by rcu_irq_enter() and rcu_irq_exit()).
241 void rcu_enter_nohz(void)
244 struct rcu_dynticks *rdtp;
246 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
247 local_irq_save(flags);
248 rdtp = &__get_cpu_var(rcu_dynticks);
250 rdtp->dynticks_nesting--;
251 WARN_ON_ONCE(rdtp->dynticks & 0x1);
252 local_irq_restore(flags);
256 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
258 * Exit nohz mode, in other words, -enter- the mode in which RCU
259 * read-side critical sections normally occur.
261 void rcu_exit_nohz(void)
264 struct rcu_dynticks *rdtp;
266 local_irq_save(flags);
267 rdtp = &__get_cpu_var(rcu_dynticks);
269 rdtp->dynticks_nesting++;
270 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
271 local_irq_restore(flags);
272 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
276 * rcu_nmi_enter - inform RCU of entry to NMI context
278 * If the CPU was idle with dynamic ticks active, and there is no
279 * irq handler running, this updates rdtp->dynticks_nmi to let the
280 * RCU grace-period handling know that the CPU is active.
282 void rcu_nmi_enter(void)
284 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
286 if (rdtp->dynticks & 0x1)
288 rdtp->dynticks_nmi++;
289 WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1));
290 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
294 * rcu_nmi_exit - inform RCU of exit from NMI context
296 * If the CPU was idle with dynamic ticks active, and there is no
297 * irq handler running, this updates rdtp->dynticks_nmi to let the
298 * RCU grace-period handling know that the CPU is no longer active.
300 void rcu_nmi_exit(void)
302 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
304 if (rdtp->dynticks & 0x1)
306 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
307 rdtp->dynticks_nmi++;
308 WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1);
312 * rcu_irq_enter - inform RCU of entry to hard irq context
314 * If the CPU was idle with dynamic ticks active, this updates the
315 * rdtp->dynticks to let the RCU handling know that the CPU is active.
317 void rcu_irq_enter(void)
319 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
321 if (rdtp->dynticks_nesting++)
324 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
325 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
329 * rcu_irq_exit - inform RCU of exit from hard irq context
331 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
332 * to put let the RCU handling be aware that the CPU is going back to idle
335 void rcu_irq_exit(void)
337 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
339 if (--rdtp->dynticks_nesting)
341 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
343 WARN_ON_ONCE(rdtp->dynticks & 0x1);
345 /* If the interrupt queued a callback, get out of dyntick mode. */
346 if (__get_cpu_var(rcu_sched_data).nxtlist ||
347 __get_cpu_var(rcu_bh_data).nxtlist)
352 * Record the specified "completed" value, which is later used to validate
353 * dynticks counter manipulations. Specify "rsp->completed - 1" to
354 * unconditionally invalidate any future dynticks manipulations (which is
355 * useful at the beginning of a grace period).
357 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
359 rsp->dynticks_completed = comp;
365 * Recall the previously recorded value of the completion for dynticks.
367 static long dyntick_recall_completed(struct rcu_state *rsp)
369 return rsp->dynticks_completed;
373 * Snapshot the specified CPU's dynticks counter so that we can later
374 * credit them with an implicit quiescent state. Return 1 if this CPU
375 * is already in a quiescent state courtesy of dynticks idle mode.
377 static int dyntick_save_progress_counter(struct rcu_data *rdp)
383 snap = rdp->dynticks->dynticks;
384 snap_nmi = rdp->dynticks->dynticks_nmi;
385 smp_mb(); /* Order sampling of snap with end of grace period. */
386 rdp->dynticks_snap = snap;
387 rdp->dynticks_nmi_snap = snap_nmi;
388 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
395 * Return true if the specified CPU has passed through a quiescent
396 * state by virtue of being in or having passed through an dynticks
397 * idle state since the last call to dyntick_save_progress_counter()
400 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
407 curr = rdp->dynticks->dynticks;
408 snap = rdp->dynticks_snap;
409 curr_nmi = rdp->dynticks->dynticks_nmi;
410 snap_nmi = rdp->dynticks_nmi_snap;
411 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
414 * If the CPU passed through or entered a dynticks idle phase with
415 * no active irq/NMI handlers, then we can safely pretend that the CPU
416 * already acknowledged the request to pass through a quiescent
417 * state. Either way, that CPU cannot possibly be in an RCU
418 * read-side critical section that started before the beginning
419 * of the current RCU grace period.
421 if ((curr != snap || (curr & 0x1) == 0) &&
422 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
427 /* Go check for the CPU being offline. */
428 return rcu_implicit_offline_qs(rdp);
431 #endif /* #ifdef CONFIG_SMP */
433 #else /* #ifdef CONFIG_NO_HZ */
435 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
442 * If there are no dynticks, then the only way that a CPU can passively
443 * be in a quiescent state is to be offline. Unlike dynticks idle, which
444 * is a point in time during the prior (already finished) grace period,
445 * an offline CPU is always in a quiescent state, and thus can be
446 * unconditionally applied. So just return the current value of completed.
448 static long dyntick_recall_completed(struct rcu_state *rsp)
450 return rsp->completed;
453 static int dyntick_save_progress_counter(struct rcu_data *rdp)
458 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
460 return rcu_implicit_offline_qs(rdp);
463 #endif /* #ifdef CONFIG_SMP */
465 #endif /* #else #ifdef CONFIG_NO_HZ */
467 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
469 static void record_gp_stall_check_time(struct rcu_state *rsp)
471 rsp->gp_start = jiffies;
472 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
475 static void print_other_cpu_stall(struct rcu_state *rsp)
480 struct rcu_node *rnp = rcu_get_root(rsp);
481 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
482 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
484 /* Only let one CPU complain about others per time interval. */
486 spin_lock_irqsave(&rnp->lock, flags);
487 delta = jiffies - rsp->jiffies_stall;
488 if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) {
489 spin_unlock_irqrestore(&rnp->lock, flags);
492 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
493 spin_unlock_irqrestore(&rnp->lock, flags);
495 /* OK, time to rat on our buddy... */
497 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
498 for (; rnp_cur < rnp_end; rnp_cur++) {
499 rcu_print_task_stall(rnp);
500 if (rnp_cur->qsmask == 0)
502 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
503 if (rnp_cur->qsmask & (1UL << cpu))
504 printk(" %d", rnp_cur->grplo + cpu);
506 printk(" (detected by %d, t=%ld jiffies)\n",
507 smp_processor_id(), (long)(jiffies - rsp->gp_start));
508 force_quiescent_state(rsp, 0); /* Kick them all. */
511 static void print_cpu_stall(struct rcu_state *rsp)
514 struct rcu_node *rnp = rcu_get_root(rsp);
516 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
517 smp_processor_id(), jiffies - rsp->gp_start);
519 spin_lock_irqsave(&rnp->lock, flags);
520 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
522 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
523 spin_unlock_irqrestore(&rnp->lock, flags);
524 set_need_resched(); /* kick ourselves to get things going. */
527 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
530 struct rcu_node *rnp;
532 delta = jiffies - rsp->jiffies_stall;
534 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
536 /* We haven't checked in, so go dump stack. */
537 print_cpu_stall(rsp);
539 } else if (rsp->gpnum != rsp->completed &&
540 delta >= RCU_STALL_RAT_DELAY) {
542 /* They had two time units to dump stack, so complain. */
543 print_other_cpu_stall(rsp);
547 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
549 static void record_gp_stall_check_time(struct rcu_state *rsp)
553 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
557 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
560 * Update CPU-local rcu_data state to record the newly noticed grace period.
561 * This is used both when we started the grace period and when we notice
562 * that someone else started the grace period.
564 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
567 rdp->passed_quiesc = 0;
568 rdp->gpnum = rsp->gpnum;
572 * Did someone else start a new RCU grace period start since we last
573 * checked? Update local state appropriately if so. Must be called
574 * on the CPU corresponding to rdp.
577 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
582 local_irq_save(flags);
583 if (rdp->gpnum != rsp->gpnum) {
584 note_new_gpnum(rsp, rdp);
587 local_irq_restore(flags);
592 * Start a new RCU grace period if warranted, re-initializing the hierarchy
593 * in preparation for detecting the next grace period. The caller must hold
594 * the root node's ->lock, which is released before return. Hard irqs must
598 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
599 __releases(rcu_get_root(rsp)->lock)
601 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
602 struct rcu_node *rnp = rcu_get_root(rsp);
603 struct rcu_node *rnp_cur;
604 struct rcu_node *rnp_end;
606 if (!cpu_needs_another_gp(rsp, rdp)) {
607 spin_unlock_irqrestore(&rnp->lock, flags);
611 /* Advance to a new grace period and initialize state. */
613 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
614 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
615 record_gp_stall_check_time(rsp);
616 dyntick_record_completed(rsp, rsp->completed - 1);
617 note_new_gpnum(rsp, rdp);
620 * Because we are first, we know that all our callbacks will
621 * be covered by this upcoming grace period, even the ones
622 * that were registered arbitrarily recently.
624 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
625 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
627 /* Special-case the common single-level case. */
628 if (NUM_RCU_NODES == 1) {
629 rnp->qsmask = rnp->qsmaskinit;
630 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
631 spin_unlock_irqrestore(&rnp->lock, flags);
635 spin_unlock(&rnp->lock); /* leave irqs disabled. */
638 /* Exclude any concurrent CPU-hotplug operations. */
639 spin_lock(&rsp->onofflock); /* irqs already disabled. */
642 * Set the quiescent-state-needed bits in all the non-leaf RCU
643 * nodes for all currently online CPUs. This operation relies
644 * on the layout of the hierarchy within the rsp->node[] array.
645 * Note that other CPUs will access only the leaves of the
646 * hierarchy, which still indicate that no grace period is in
647 * progress. In addition, we have excluded CPU-hotplug operations.
649 * We therefore do not need to hold any locks. Any required
650 * memory barriers will be supplied by the locks guarding the
651 * leaf rcu_nodes in the hierarchy.
654 rnp_end = rsp->level[NUM_RCU_LVLS - 1];
655 for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++)
656 rnp_cur->qsmask = rnp_cur->qsmaskinit;
659 * Now set up the leaf nodes. Here we must be careful. First,
660 * we need to hold the lock in order to exclude other CPUs, which
661 * might be contending for the leaf nodes' locks. Second, as
662 * soon as we initialize a given leaf node, its CPUs might run
663 * up the rest of the hierarchy. We must therefore acquire locks
664 * for each node that we touch during this stage. (But we still
665 * are excluding CPU-hotplug operations.)
667 * Note that the grace period cannot complete until we finish
668 * the initialization process, as there will be at least one
669 * qsmask bit set in the root node until that time, namely the
670 * one corresponding to this CPU.
672 rnp_end = &rsp->node[NUM_RCU_NODES];
673 rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
674 for (; rnp_cur < rnp_end; rnp_cur++) {
675 spin_lock(&rnp_cur->lock); /* irqs already disabled. */
676 rnp_cur->qsmask = rnp_cur->qsmaskinit;
677 spin_unlock(&rnp_cur->lock); /* irqs already disabled. */
680 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
681 spin_unlock_irqrestore(&rsp->onofflock, flags);
685 * Advance this CPU's callbacks, but only if the current grace period
686 * has ended. This may be called only from the CPU to whom the rdp
690 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
695 local_irq_save(flags);
696 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
698 /* Did another grace period end? */
699 if (rdp->completed != completed_snap) {
701 /* Advance callbacks. No harm if list empty. */
702 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
703 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
704 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
706 /* Remember that we saw this grace-period completion. */
707 rdp->completed = completed_snap;
709 local_irq_restore(flags);
713 * Clean up after the prior grace period and let rcu_start_gp() start up
714 * the next grace period if one is needed. Note that the caller must
715 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
717 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
718 __releases(rnp->lock)
720 rsp->completed = rsp->gpnum;
721 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
722 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
726 * Similar to cpu_quiet(), for which it is a helper function. Allows
727 * a group of CPUs to be quieted at one go, though all the CPUs in the
728 * group must be represented by the same leaf rcu_node structure.
729 * That structure's lock must be held upon entry, and it is released
733 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
735 __releases(rnp->lock)
737 /* Walk up the rcu_node hierarchy. */
739 if (!(rnp->qsmask & mask)) {
741 /* Our bit has already been cleared, so done. */
742 spin_unlock_irqrestore(&rnp->lock, flags);
745 rnp->qsmask &= ~mask;
746 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
748 /* Other bits still set at this level, so done. */
749 spin_unlock_irqrestore(&rnp->lock, flags);
753 if (rnp->parent == NULL) {
755 /* No more levels. Exit loop holding root lock. */
759 spin_unlock_irqrestore(&rnp->lock, flags);
761 spin_lock_irqsave(&rnp->lock, flags);
765 * Get here if we are the last CPU to pass through a quiescent
766 * state for this grace period. Invoke cpu_quiet_msk_finish()
767 * to clean up and start the next grace period if one is needed.
769 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
773 * Record a quiescent state for the specified CPU, which must either be
774 * the current CPU or an offline CPU. The lastcomp argument is used to
775 * make sure we are still in the grace period of interest. We don't want
776 * to end the current grace period based on quiescent states detected in
777 * an earlier grace period!
780 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
784 struct rcu_node *rnp;
787 spin_lock_irqsave(&rnp->lock, flags);
788 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
791 * Someone beat us to it for this grace period, so leave.
792 * The race with GP start is resolved by the fact that we
793 * hold the leaf rcu_node lock, so that the per-CPU bits
794 * cannot yet be initialized -- so we would simply find our
795 * CPU's bit already cleared in cpu_quiet_msk() if this race
798 rdp->passed_quiesc = 0; /* try again later! */
799 spin_unlock_irqrestore(&rnp->lock, flags);
803 if ((rnp->qsmask & mask) == 0) {
804 spin_unlock_irqrestore(&rnp->lock, flags);
809 * This GP can't end until cpu checks in, so all of our
810 * callbacks can be processed during the next GP.
812 rdp = rsp->rda[smp_processor_id()];
813 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
815 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
820 * Check to see if there is a new grace period of which this CPU
821 * is not yet aware, and if so, set up local rcu_data state for it.
822 * Otherwise, see if this CPU has just passed through its first
823 * quiescent state for this grace period, and record that fact if so.
826 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
828 /* If there is now a new grace period, record and return. */
829 if (check_for_new_grace_period(rsp, rdp))
833 * Does this CPU still need to do its part for current grace period?
834 * If no, return and let the other CPUs do their part as well.
836 if (!rdp->qs_pending)
840 * Was there a quiescent state since the beginning of the grace
841 * period? If no, then exit and wait for the next call.
843 if (!rdp->passed_quiesc)
846 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
847 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
850 #ifdef CONFIG_HOTPLUG_CPU
853 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
854 * and move all callbacks from the outgoing CPU to the current one.
856 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
862 struct rcu_data *rdp = rsp->rda[cpu];
863 struct rcu_data *rdp_me;
864 struct rcu_node *rnp;
866 /* Exclude any attempts to start a new grace period. */
867 spin_lock_irqsave(&rsp->onofflock, flags);
869 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
871 mask = rdp->grpmask; /* rnp->grplo is constant. */
873 spin_lock(&rnp->lock); /* irqs already disabled. */
874 rnp->qsmaskinit &= ~mask;
875 if (rnp->qsmaskinit != 0) {
876 spin_unlock(&rnp->lock); /* irqs remain disabled. */
879 rcu_preempt_offline_tasks(rsp, rnp);
881 spin_unlock(&rnp->lock); /* irqs remain disabled. */
883 } while (rnp != NULL);
884 lastcomp = rsp->completed;
886 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
888 /* Being offline is a quiescent state, so go record it. */
889 cpu_quiet(cpu, rsp, rdp, lastcomp);
892 * Move callbacks from the outgoing CPU to the running CPU.
893 * Note that the outgoing CPU is now quiscent, so it is now
894 * (uncharacteristically) safe to access its rcu_data structure.
895 * Note also that we must carefully retain the order of the
896 * outgoing CPU's callbacks in order for rcu_barrier() to work
897 * correctly. Finally, note that we start all the callbacks
898 * afresh, even those that have passed through a grace period
899 * and are therefore ready to invoke. The theory is that hotplug
900 * events are rare, and that if they are frequent enough to
901 * indefinitely delay callbacks, you have far worse things to
904 rdp_me = rsp->rda[smp_processor_id()];
905 if (rdp->nxtlist != NULL) {
906 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
907 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
909 for (i = 0; i < RCU_NEXT_SIZE; i++)
910 rdp->nxttail[i] = &rdp->nxtlist;
911 rdp_me->qlen += rdp->qlen;
914 local_irq_restore(flags);
918 * Remove the specified CPU from the RCU hierarchy and move any pending
919 * callbacks that it might have to the current CPU. This code assumes
920 * that at least one CPU in the system will remain running at all times.
921 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
923 static void rcu_offline_cpu(int cpu)
925 __rcu_offline_cpu(cpu, &rcu_sched_state);
926 __rcu_offline_cpu(cpu, &rcu_bh_state);
927 rcu_preempt_offline_cpu(cpu);
930 #else /* #ifdef CONFIG_HOTPLUG_CPU */
932 static void rcu_offline_cpu(int cpu)
936 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
939 * Invoke any RCU callbacks that have made it to the end of their grace
940 * period. Thottle as specified by rdp->blimit.
942 static void rcu_do_batch(struct rcu_data *rdp)
945 struct rcu_head *next, *list, **tail;
948 /* If no callbacks are ready, just return.*/
949 if (!cpu_has_callbacks_ready_to_invoke(rdp))
953 * Extract the list of ready callbacks, disabling to prevent
954 * races with call_rcu() from interrupt handlers.
956 local_irq_save(flags);
958 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
959 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
960 tail = rdp->nxttail[RCU_DONE_TAIL];
961 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
962 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
963 rdp->nxttail[count] = &rdp->nxtlist;
964 local_irq_restore(flags);
966 /* Invoke callbacks. */
973 if (++count >= rdp->blimit)
977 local_irq_save(flags);
979 /* Update count, and requeue any remaining callbacks. */
982 *tail = rdp->nxtlist;
984 for (count = 0; count < RCU_NEXT_SIZE; count++)
985 if (&rdp->nxtlist == rdp->nxttail[count])
986 rdp->nxttail[count] = tail;
991 /* Reinstate batch limit if we have worked down the excess. */
992 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
993 rdp->blimit = blimit;
995 local_irq_restore(flags);
997 /* Re-raise the RCU softirq if there are callbacks remaining. */
998 if (cpu_has_callbacks_ready_to_invoke(rdp))
999 raise_softirq(RCU_SOFTIRQ);
1003 * Check to see if this CPU is in a non-context-switch quiescent state
1004 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1005 * Also schedule the RCU softirq handler.
1007 * This function must be called with hardirqs disabled. It is normally
1008 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1009 * false, there is no point in invoking rcu_check_callbacks().
1011 void rcu_check_callbacks(int cpu, int user)
1013 if (!rcu_pending(cpu))
1014 return; /* if nothing for RCU to do. */
1016 (idle_cpu(cpu) && rcu_scheduler_active &&
1017 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1020 * Get here if this CPU took its interrupt from user
1021 * mode or from the idle loop, and if this is not a
1022 * nested interrupt. In this case, the CPU is in
1023 * a quiescent state, so note it.
1025 * No memory barrier is required here because both
1026 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1027 * variables that other CPUs neither access nor modify,
1028 * at least not while the corresponding CPU is online.
1034 } else if (!in_softirq()) {
1037 * Get here if this CPU did not take its interrupt from
1038 * softirq, in other words, if it is not interrupting
1039 * a rcu_bh read-side critical section. This is an _bh
1040 * critical section, so note it.
1045 rcu_preempt_check_callbacks(cpu);
1046 raise_softirq(RCU_SOFTIRQ);
1052 * Scan the leaf rcu_node structures, processing dyntick state for any that
1053 * have not yet encountered a quiescent state, using the function specified.
1054 * Returns 1 if the current grace period ends while scanning (possibly
1055 * because we made it end).
1057 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1058 int (*f)(struct rcu_data *))
1062 unsigned long flags;
1064 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
1065 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1067 for (; rnp_cur < rnp_end; rnp_cur++) {
1069 spin_lock_irqsave(&rnp_cur->lock, flags);
1070 if (rsp->completed != lastcomp) {
1071 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1074 if (rnp_cur->qsmask == 0) {
1075 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1078 cpu = rnp_cur->grplo;
1080 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1081 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1084 if (mask != 0 && rsp->completed == lastcomp) {
1086 /* cpu_quiet_msk() releases rnp_cur->lock. */
1087 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1090 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1096 * Force quiescent states on reluctant CPUs, and also detect which
1097 * CPUs are in dyntick-idle mode.
1099 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1101 unsigned long flags;
1103 struct rcu_node *rnp = rcu_get_root(rsp);
1106 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
1107 return; /* No grace period in progress, nothing to force. */
1108 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1109 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1110 return; /* Someone else is already on the job. */
1113 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1114 goto unlock_ret; /* no emergency and done recently. */
1116 spin_lock(&rnp->lock);
1117 lastcomp = rsp->completed;
1118 signaled = rsp->signaled;
1119 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1120 if (lastcomp == rsp->gpnum) {
1121 rsp->n_force_qs_ngp++;
1122 spin_unlock(&rnp->lock);
1123 goto unlock_ret; /* no GP in progress, time updated. */
1125 spin_unlock(&rnp->lock);
1129 break; /* grace period still initializing, ignore. */
1131 case RCU_SAVE_DYNTICK:
1133 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1134 break; /* So gcc recognizes the dead code. */
1136 /* Record dyntick-idle state. */
1137 if (rcu_process_dyntick(rsp, lastcomp,
1138 dyntick_save_progress_counter))
1141 /* Update state, record completion counter. */
1142 spin_lock(&rnp->lock);
1143 if (lastcomp == rsp->completed) {
1144 rsp->signaled = RCU_FORCE_QS;
1145 dyntick_record_completed(rsp, lastcomp);
1147 spin_unlock(&rnp->lock);
1152 /* Check dyntick-idle state, send IPI to laggarts. */
1153 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1154 rcu_implicit_dynticks_qs))
1157 /* Leave state in case more forcing is required. */
1162 spin_unlock_irqrestore(&rsp->fqslock, flags);
1165 #else /* #ifdef CONFIG_SMP */
1167 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1172 #endif /* #else #ifdef CONFIG_SMP */
1175 * This does the RCU processing work from softirq context for the
1176 * specified rcu_state and rcu_data structures. This may be called
1177 * only from the CPU to whom the rdp belongs.
1180 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1182 unsigned long flags;
1184 WARN_ON_ONCE(rdp->beenonline == 0);
1187 * If an RCU GP has gone long enough, go check for dyntick
1188 * idle CPUs and, if needed, send resched IPIs.
1190 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1191 force_quiescent_state(rsp, 1);
1194 * Advance callbacks in response to end of earlier grace
1195 * period that some other CPU ended.
1197 rcu_process_gp_end(rsp, rdp);
1199 /* Update RCU state based on any recent quiescent states. */
1200 rcu_check_quiescent_state(rsp, rdp);
1202 /* Does this CPU require a not-yet-started grace period? */
1203 if (cpu_needs_another_gp(rsp, rdp)) {
1204 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1205 rcu_start_gp(rsp, flags); /* releases above lock */
1208 /* If there are callbacks ready, invoke them. */
1213 * Do softirq processing for the current CPU.
1215 static void rcu_process_callbacks(struct softirq_action *unused)
1218 * Memory references from any prior RCU read-side critical sections
1219 * executed by the interrupted code must be seen before any RCU
1220 * grace-period manipulations below.
1222 smp_mb(); /* See above block comment. */
1224 __rcu_process_callbacks(&rcu_sched_state,
1225 &__get_cpu_var(rcu_sched_data));
1226 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1227 rcu_preempt_process_callbacks();
1230 * Memory references from any later RCU read-side critical sections
1231 * executed by the interrupted code must be seen after any RCU
1232 * grace-period manipulations above.
1234 smp_mb(); /* See above block comment. */
1238 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1239 struct rcu_state *rsp)
1241 unsigned long flags;
1242 struct rcu_data *rdp;
1247 smp_mb(); /* Ensure RCU update seen before callback registry. */
1250 * Opportunistically note grace-period endings and beginnings.
1251 * Note that we might see a beginning right after we see an
1252 * end, but never vice versa, since this CPU has to pass through
1253 * a quiescent state betweentimes.
1255 local_irq_save(flags);
1256 rdp = rsp->rda[smp_processor_id()];
1257 rcu_process_gp_end(rsp, rdp);
1258 check_for_new_grace_period(rsp, rdp);
1260 /* Add the callback to our list. */
1261 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1262 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1264 /* Start a new grace period if one not already started. */
1265 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
1266 unsigned long nestflag;
1267 struct rcu_node *rnp_root = rcu_get_root(rsp);
1269 spin_lock_irqsave(&rnp_root->lock, nestflag);
1270 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1273 /* Force the grace period if too many callbacks or too long waiting. */
1274 if (unlikely(++rdp->qlen > qhimark)) {
1275 rdp->blimit = LONG_MAX;
1276 force_quiescent_state(rsp, 0);
1277 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1278 force_quiescent_state(rsp, 1);
1279 local_irq_restore(flags);
1283 * Queue an RCU-sched callback for invocation after a grace period.
1285 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1287 __call_rcu(head, func, &rcu_sched_state);
1289 EXPORT_SYMBOL_GPL(call_rcu_sched);
1292 * Queue an RCU for invocation after a quicker grace period.
1294 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1296 __call_rcu(head, func, &rcu_bh_state);
1298 EXPORT_SYMBOL_GPL(call_rcu_bh);
1301 * Check to see if there is any immediate RCU-related work to be done
1302 * by the current CPU, for the specified type of RCU, returning 1 if so.
1303 * The checks are in order of increasing expense: checks that can be
1304 * carried out against CPU-local state are performed first. However,
1305 * we must check for CPU stalls first, else we might not get a chance.
1307 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1309 rdp->n_rcu_pending++;
1311 /* Check for CPU stalls, if enabled. */
1312 check_cpu_stall(rsp, rdp);
1314 /* Is the RCU core waiting for a quiescent state from this CPU? */
1315 if (rdp->qs_pending) {
1316 rdp->n_rp_qs_pending++;
1320 /* Does this CPU have callbacks ready to invoke? */
1321 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1322 rdp->n_rp_cb_ready++;
1326 /* Has RCU gone idle with this CPU needing another grace period? */
1327 if (cpu_needs_another_gp(rsp, rdp)) {
1328 rdp->n_rp_cpu_needs_gp++;
1332 /* Has another RCU grace period completed? */
1333 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1334 rdp->n_rp_gp_completed++;
1338 /* Has a new RCU grace period started? */
1339 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1340 rdp->n_rp_gp_started++;
1344 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1345 if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
1346 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1347 rdp->n_rp_need_fqs++;
1352 rdp->n_rp_need_nothing++;
1357 * Check to see if there is any immediate RCU-related work to be done
1358 * by the current CPU, returning 1 if so. This function is part of the
1359 * RCU implementation; it is -not- an exported member of the RCU API.
1361 static int rcu_pending(int cpu)
1363 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1364 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1365 rcu_preempt_pending(cpu);
1369 * Check to see if any future RCU-related work will need to be done
1370 * by the current CPU, even if none need be done immediately, returning
1371 * 1 if so. This function is part of the RCU implementation; it is -not-
1372 * an exported member of the RCU API.
1374 int rcu_needs_cpu(int cpu)
1376 /* RCU callbacks either ready or pending? */
1377 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1378 per_cpu(rcu_bh_data, cpu).nxtlist ||
1379 rcu_preempt_needs_cpu(cpu);
1383 * Do boot-time initialization of a CPU's per-CPU RCU data.
1386 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1388 unsigned long flags;
1390 struct rcu_data *rdp = rsp->rda[cpu];
1391 struct rcu_node *rnp = rcu_get_root(rsp);
1393 /* Set up local state, ensuring consistent view of global state. */
1394 spin_lock_irqsave(&rnp->lock, flags);
1395 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1396 rdp->nxtlist = NULL;
1397 for (i = 0; i < RCU_NEXT_SIZE; i++)
1398 rdp->nxttail[i] = &rdp->nxtlist;
1401 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1402 #endif /* #ifdef CONFIG_NO_HZ */
1404 spin_unlock_irqrestore(&rnp->lock, flags);
1408 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1409 * offline event can be happening at a given time. Note also that we
1410 * can accept some slop in the rsp->completed access due to the fact
1411 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1413 static void __cpuinit
1414 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1416 unsigned long flags;
1419 struct rcu_data *rdp = rsp->rda[cpu];
1420 struct rcu_node *rnp = rcu_get_root(rsp);
1422 /* Set up local state, ensuring consistent view of global state. */
1423 spin_lock_irqsave(&rnp->lock, flags);
1424 lastcomp = rsp->completed;
1425 rdp->completed = lastcomp;
1426 rdp->gpnum = lastcomp;
1427 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1428 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1429 rdp->beenonline = 1; /* We have now been online. */
1430 rdp->preemptable = preemptable;
1431 rdp->passed_quiesc_completed = lastcomp - 1;
1432 rdp->blimit = blimit;
1433 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1436 * A new grace period might start here. If so, we won't be part
1437 * of it, but that is OK, as we are currently in a quiescent state.
1440 /* Exclude any attempts to start a new GP on large systems. */
1441 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1443 /* Add CPU to rcu_node bitmasks. */
1445 mask = rdp->grpmask;
1447 /* Exclude any attempts to start a new GP on small systems. */
1448 spin_lock(&rnp->lock); /* irqs already disabled. */
1449 rnp->qsmaskinit |= mask;
1450 mask = rnp->grpmask;
1451 spin_unlock(&rnp->lock); /* irqs already disabled. */
1453 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1455 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1458 * A new grace period might start here. If so, we will be part of
1459 * it, and its gpnum will be greater than ours, so we will
1460 * participate. It is also possible for the gpnum to have been
1461 * incremented before this function was called, and the bitmasks
1462 * to not be filled out until now, in which case we will also
1463 * participate due to our gpnum being behind.
1466 /* Since it is coming online, the CPU is in a quiescent state. */
1467 cpu_quiet(cpu, rsp, rdp, lastcomp);
1468 local_irq_restore(flags);
1471 static void __cpuinit rcu_online_cpu(int cpu)
1473 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1474 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1475 rcu_preempt_init_percpu_data(cpu);
1479 * Handle CPU online/offline notification events.
1481 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1482 unsigned long action, void *hcpu)
1484 long cpu = (long)hcpu;
1487 case CPU_UP_PREPARE:
1488 case CPU_UP_PREPARE_FROZEN:
1489 rcu_online_cpu(cpu);
1492 case CPU_DEAD_FROZEN:
1493 case CPU_UP_CANCELED:
1494 case CPU_UP_CANCELED_FROZEN:
1495 rcu_offline_cpu(cpu);
1504 * Compute the per-level fanout, either using the exact fanout specified
1505 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1507 #ifdef CONFIG_RCU_FANOUT_EXACT
1508 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1512 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1513 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1515 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1516 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1523 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1524 ccur = rsp->levelcnt[i];
1525 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1529 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1532 * Helper function for rcu_init() that initializes one rcu_state structure.
1534 static void __init rcu_init_one(struct rcu_state *rsp)
1539 struct rcu_node *rnp;
1541 /* Initialize the level-tracking arrays. */
1543 for (i = 1; i < NUM_RCU_LVLS; i++)
1544 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1545 rcu_init_levelspread(rsp);
1547 /* Initialize the elements themselves, starting from the leaves. */
1549 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1550 cpustride *= rsp->levelspread[i];
1551 rnp = rsp->level[i];
1552 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1553 spin_lock_init(&rnp->lock);
1556 rnp->qsmaskinit = 0;
1557 rnp->grplo = j * cpustride;
1558 rnp->grphi = (j + 1) * cpustride - 1;
1559 if (rnp->grphi >= NR_CPUS)
1560 rnp->grphi = NR_CPUS - 1;
1566 rnp->grpnum = j % rsp->levelspread[i - 1];
1567 rnp->grpmask = 1UL << rnp->grpnum;
1568 rnp->parent = rsp->level[i - 1] +
1569 j / rsp->levelspread[i - 1];
1572 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1573 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1579 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1580 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1583 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1585 rcu_init_one(rsp); \
1586 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1588 for_each_possible_cpu(i) { \
1589 if (i > rnp[j].grphi) \
1591 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1592 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1593 rcu_boot_init_percpu_data(i, rsp); \
1597 #ifdef CONFIG_TREE_PREEMPT_RCU
1599 void __init __rcu_init_preempt(void)
1601 int i; /* All used by RCU_INIT_FLAVOR(). */
1603 struct rcu_node *rnp;
1605 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
1608 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1610 void __init __rcu_init_preempt(void)
1614 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1616 void __init __rcu_init(void)
1618 int i; /* All used by RCU_INIT_FLAVOR(). */
1620 struct rcu_node *rnp;
1622 rcu_bootup_announce();
1623 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1624 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1625 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1626 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1627 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1628 __rcu_init_preempt();
1629 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1632 module_param(blimit, int, 0);
1633 module_param(qhimark, int, 0);
1634 module_param(qlowmark, int, 0);