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 void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp,
85 struct rcu_node *rnp, unsigned long flags);
86 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags);
87 static void __rcu_process_callbacks(struct rcu_state *rsp,
88 struct rcu_data *rdp);
89 static void __call_rcu(struct rcu_head *head,
90 void (*func)(struct rcu_head *rcu),
91 struct rcu_state *rsp);
92 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp);
93 static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp,
96 #include "rcutree_plugin.h"
99 * Note a quiescent state. Because we do not need to know
100 * how many quiescent states passed, just if there was at least
101 * one since the start of the grace period, this just sets a flag.
103 void rcu_sched_qs(int cpu)
106 struct rcu_data *rdp;
108 local_irq_save(flags);
109 rdp = &per_cpu(rcu_sched_data, cpu);
110 rdp->passed_quiesc = 1;
111 rdp->passed_quiesc_completed = rdp->completed;
113 local_irq_restore(flags);
116 void rcu_bh_qs(int cpu)
119 struct rcu_data *rdp;
121 local_irq_save(flags);
122 rdp = &per_cpu(rcu_bh_data, cpu);
123 rdp->passed_quiesc = 1;
124 rdp->passed_quiesc_completed = rdp->completed;
125 local_irq_restore(flags);
129 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
130 .dynticks_nesting = 1,
133 #endif /* #ifdef CONFIG_NO_HZ */
135 static int blimit = 10; /* Maximum callbacks per softirq. */
136 static int qhimark = 10000; /* If this many pending, ignore blimit. */
137 static int qlowmark = 100; /* Once only this many pending, use blimit. */
139 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
140 static int rcu_pending(int cpu);
143 * Return the number of RCU-sched batches processed thus far for debug & stats.
145 long rcu_batches_completed_sched(void)
147 return rcu_sched_state.completed;
149 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
152 * Return the number of RCU BH batches processed thus far for debug & stats.
154 long rcu_batches_completed_bh(void)
156 return rcu_bh_state.completed;
158 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
161 * Does the CPU have callbacks ready to be invoked?
164 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
166 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
170 * Does the current CPU require a yet-as-unscheduled grace period?
173 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
175 /* ACCESS_ONCE() because we are accessing outside of lock. */
176 return *rdp->nxttail[RCU_DONE_TAIL] &&
177 ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
181 * Return the root node of the specified rcu_state structure.
183 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
185 return &rsp->node[0];
191 * If the specified CPU is offline, tell the caller that it is in
192 * a quiescent state. Otherwise, whack it with a reschedule IPI.
193 * Grace periods can end up waiting on an offline CPU when that
194 * CPU is in the process of coming online -- it will be added to the
195 * rcu_node bitmasks before it actually makes it online. The same thing
196 * can happen while a CPU is in the process of coming online. Because this
197 * race is quite rare, we check for it after detecting that the grace
198 * period has been delayed rather than checking each and every CPU
199 * each and every time we start a new grace period.
201 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
204 * If the CPU is offline, it is in a quiescent state. We can
205 * trust its state not to change because interrupts are disabled.
207 if (cpu_is_offline(rdp->cpu)) {
212 /* If preemptable RCU, no point in sending reschedule IPI. */
213 if (rdp->preemptable)
216 /* The CPU is online, so send it a reschedule IPI. */
217 if (rdp->cpu != smp_processor_id())
218 smp_send_reschedule(rdp->cpu);
225 #endif /* #ifdef CONFIG_SMP */
228 static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5);
231 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
233 * Enter nohz mode, in other words, -leave- the mode in which RCU
234 * read-side critical sections can occur. (Though RCU read-side
235 * critical sections can occur in irq handlers in nohz mode, a possibility
236 * handled by rcu_irq_enter() and rcu_irq_exit()).
238 void rcu_enter_nohz(void)
241 struct rcu_dynticks *rdtp;
243 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
244 local_irq_save(flags);
245 rdtp = &__get_cpu_var(rcu_dynticks);
247 rdtp->dynticks_nesting--;
248 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
249 local_irq_restore(flags);
253 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
255 * Exit nohz mode, in other words, -enter- the mode in which RCU
256 * read-side critical sections normally occur.
258 void rcu_exit_nohz(void)
261 struct rcu_dynticks *rdtp;
263 local_irq_save(flags);
264 rdtp = &__get_cpu_var(rcu_dynticks);
266 rdtp->dynticks_nesting++;
267 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
268 local_irq_restore(flags);
269 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
273 * rcu_nmi_enter - inform RCU of entry to NMI context
275 * If the CPU was idle with dynamic ticks active, and there is no
276 * irq handler running, this updates rdtp->dynticks_nmi to let the
277 * RCU grace-period handling know that the CPU is active.
279 void rcu_nmi_enter(void)
281 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
283 if (rdtp->dynticks & 0x1)
285 rdtp->dynticks_nmi++;
286 WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs);
287 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
291 * rcu_nmi_exit - inform RCU of exit from NMI context
293 * If the CPU was idle with dynamic ticks active, and there is no
294 * irq handler running, this updates rdtp->dynticks_nmi to let the
295 * RCU grace-period handling know that the CPU is no longer active.
297 void rcu_nmi_exit(void)
299 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
301 if (rdtp->dynticks & 0x1)
303 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
304 rdtp->dynticks_nmi++;
305 WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs);
309 * rcu_irq_enter - inform RCU of entry to hard irq context
311 * If the CPU was idle with dynamic ticks active, this updates the
312 * rdtp->dynticks to let the RCU handling know that the CPU is active.
314 void rcu_irq_enter(void)
316 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
318 if (rdtp->dynticks_nesting++)
321 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
322 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
326 * rcu_irq_exit - inform RCU of exit from hard irq context
328 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
329 * to put let the RCU handling be aware that the CPU is going back to idle
332 void rcu_irq_exit(void)
334 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
336 if (--rdtp->dynticks_nesting)
338 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
340 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
342 /* If the interrupt queued a callback, get out of dyntick mode. */
343 if (__get_cpu_var(rcu_sched_data).nxtlist ||
344 __get_cpu_var(rcu_bh_data).nxtlist)
349 * Record the specified "completed" value, which is later used to validate
350 * dynticks counter manipulations. Specify "rsp->completed - 1" to
351 * unconditionally invalidate any future dynticks manipulations (which is
352 * useful at the beginning of a grace period).
354 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
356 rsp->dynticks_completed = comp;
362 * Recall the previously recorded value of the completion for dynticks.
364 static long dyntick_recall_completed(struct rcu_state *rsp)
366 return rsp->dynticks_completed;
370 * Snapshot the specified CPU's dynticks counter so that we can later
371 * credit them with an implicit quiescent state. Return 1 if this CPU
372 * is already in a quiescent state courtesy of dynticks idle mode.
374 static int dyntick_save_progress_counter(struct rcu_data *rdp)
380 snap = rdp->dynticks->dynticks;
381 snap_nmi = rdp->dynticks->dynticks_nmi;
382 smp_mb(); /* Order sampling of snap with end of grace period. */
383 rdp->dynticks_snap = snap;
384 rdp->dynticks_nmi_snap = snap_nmi;
385 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
392 * Return true if the specified CPU has passed through a quiescent
393 * state by virtue of being in or having passed through an dynticks
394 * idle state since the last call to dyntick_save_progress_counter()
397 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
404 curr = rdp->dynticks->dynticks;
405 snap = rdp->dynticks_snap;
406 curr_nmi = rdp->dynticks->dynticks_nmi;
407 snap_nmi = rdp->dynticks_nmi_snap;
408 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
411 * If the CPU passed through or entered a dynticks idle phase with
412 * no active irq/NMI handlers, then we can safely pretend that the CPU
413 * already acknowledged the request to pass through a quiescent
414 * state. Either way, that CPU cannot possibly be in an RCU
415 * read-side critical section that started before the beginning
416 * of the current RCU grace period.
418 if ((curr != snap || (curr & 0x1) == 0) &&
419 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
424 /* Go check for the CPU being offline. */
425 return rcu_implicit_offline_qs(rdp);
428 #endif /* #ifdef CONFIG_SMP */
430 #else /* #ifdef CONFIG_NO_HZ */
432 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
439 * If there are no dynticks, then the only way that a CPU can passively
440 * be in a quiescent state is to be offline. Unlike dynticks idle, which
441 * is a point in time during the prior (already finished) grace period,
442 * an offline CPU is always in a quiescent state, and thus can be
443 * unconditionally applied. So just return the current value of completed.
445 static long dyntick_recall_completed(struct rcu_state *rsp)
447 return rsp->completed;
450 static int dyntick_save_progress_counter(struct rcu_data *rdp)
455 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
457 return rcu_implicit_offline_qs(rdp);
460 #endif /* #ifdef CONFIG_SMP */
462 #endif /* #else #ifdef CONFIG_NO_HZ */
464 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
466 static void record_gp_stall_check_time(struct rcu_state *rsp)
468 rsp->gp_start = jiffies;
469 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
472 static void print_other_cpu_stall(struct rcu_state *rsp)
477 struct rcu_node *rnp = rcu_get_root(rsp);
478 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
479 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
481 /* Only let one CPU complain about others per time interval. */
483 spin_lock_irqsave(&rnp->lock, flags);
484 delta = jiffies - rsp->jiffies_stall;
485 if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) {
486 spin_unlock_irqrestore(&rnp->lock, flags);
489 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
490 spin_unlock_irqrestore(&rnp->lock, flags);
492 /* OK, time to rat on our buddy... */
494 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
495 for (; rnp_cur < rnp_end; rnp_cur++) {
496 rcu_print_task_stall(rnp);
497 if (rnp_cur->qsmask == 0)
499 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
500 if (rnp_cur->qsmask & (1UL << cpu))
501 printk(" %d", rnp_cur->grplo + cpu);
503 printk(" (detected by %d, t=%ld jiffies)\n",
504 smp_processor_id(), (long)(jiffies - rsp->gp_start));
505 force_quiescent_state(rsp, 0); /* Kick them all. */
508 static void print_cpu_stall(struct rcu_state *rsp)
511 struct rcu_node *rnp = rcu_get_root(rsp);
513 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
514 smp_processor_id(), jiffies - rsp->gp_start);
516 spin_lock_irqsave(&rnp->lock, flags);
517 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
519 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
520 spin_unlock_irqrestore(&rnp->lock, flags);
521 set_need_resched(); /* kick ourselves to get things going. */
524 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
527 struct rcu_node *rnp;
529 delta = jiffies - rsp->jiffies_stall;
531 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
533 /* We haven't checked in, so go dump stack. */
534 print_cpu_stall(rsp);
536 } else if (rsp->gpnum != rsp->completed &&
537 delta >= RCU_STALL_RAT_DELAY) {
539 /* They had two time units to dump stack, so complain. */
540 print_other_cpu_stall(rsp);
544 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
546 static void record_gp_stall_check_time(struct rcu_state *rsp)
550 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
554 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
557 * Update CPU-local rcu_data state to record the newly noticed grace period.
558 * This is used both when we started the grace period and when we notice
559 * that someone else started the grace period.
561 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
564 rdp->passed_quiesc = 0;
565 rdp->gpnum = rsp->gpnum;
569 * Did someone else start a new RCU grace period start since we last
570 * checked? Update local state appropriately if so. Must be called
571 * on the CPU corresponding to rdp.
574 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
579 local_irq_save(flags);
580 if (rdp->gpnum != rsp->gpnum) {
581 note_new_gpnum(rsp, rdp);
584 local_irq_restore(flags);
589 * Start a new RCU grace period if warranted, re-initializing the hierarchy
590 * in preparation for detecting the next grace period. The caller must hold
591 * the root node's ->lock, which is released before return. Hard irqs must
595 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
596 __releases(rcu_get_root(rsp)->lock)
598 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
599 struct rcu_node *rnp = rcu_get_root(rsp);
600 struct rcu_node *rnp_cur;
601 struct rcu_node *rnp_end;
603 if (!cpu_needs_another_gp(rsp, rdp)) {
604 spin_unlock_irqrestore(&rnp->lock, flags);
608 /* Advance to a new grace period and initialize state. */
610 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
611 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
612 record_gp_stall_check_time(rsp);
613 dyntick_record_completed(rsp, rsp->completed - 1);
614 note_new_gpnum(rsp, rdp);
617 * Because we are first, we know that all our callbacks will
618 * be covered by this upcoming grace period, even the ones
619 * that were registered arbitrarily recently.
621 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
622 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
624 /* Special-case the common single-level case. */
625 if (NUM_RCU_NODES == 1) {
626 rnp->qsmask = rnp->qsmaskinit;
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 non-leaf RCU
640 * nodes for all currently online CPUs. This operation relies
641 * on the layout of the hierarchy within the rsp->node[] array.
642 * Note that other CPUs will access only the leaves of the
643 * hierarchy, which still indicate that no grace period is in
644 * progress. In addition, we have excluded CPU-hotplug operations.
646 * We therefore do not need to hold any locks. Any required
647 * memory barriers will be supplied by the locks guarding the
648 * leaf rcu_nodes in the hierarchy.
651 rnp_end = rsp->level[NUM_RCU_LVLS - 1];
652 for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++)
653 rnp_cur->qsmask = rnp_cur->qsmaskinit;
656 * Now set up the leaf nodes. Here we must be careful. First,
657 * we need to hold the lock in order to exclude other CPUs, which
658 * might be contending for the leaf nodes' locks. Second, as
659 * soon as we initialize a given leaf node, its CPUs might run
660 * up the rest of the hierarchy. We must therefore acquire locks
661 * for each node that we touch during this stage. (But we still
662 * are excluding CPU-hotplug operations.)
664 * Note that the grace period cannot complete until we finish
665 * the initialization process, as there will be at least one
666 * qsmask bit set in the root node until that time, namely the
667 * one corresponding to this CPU.
669 rnp_end = &rsp->node[NUM_RCU_NODES];
670 rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
671 for (; rnp_cur < rnp_end; rnp_cur++) {
672 spin_lock(&rnp_cur->lock); /* irqs already disabled. */
673 rnp_cur->qsmask = rnp_cur->qsmaskinit;
674 spin_unlock(&rnp_cur->lock); /* irqs already disabled. */
677 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
678 spin_unlock_irqrestore(&rsp->onofflock, flags);
682 * Advance this CPU's callbacks, but only if the current grace period
683 * has ended. This may be called only from the CPU to whom the rdp
687 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
692 local_irq_save(flags);
693 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
695 /* Did another grace period end? */
696 if (rdp->completed != completed_snap) {
698 /* Advance callbacks. No harm if list empty. */
699 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
700 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
701 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
703 /* Remember that we saw this grace-period completion. */
704 rdp->completed = completed_snap;
706 local_irq_restore(flags);
710 * Clean up after the prior grace period and let rcu_start_gp() start up
711 * the next grace period if one is needed. Note that the caller must
712 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
714 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
715 __releases(rnp->lock)
717 rsp->completed = rsp->gpnum;
718 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
719 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
723 * Similar to cpu_quiet(), for which it is a helper function. Allows
724 * a group of CPUs to be quieted at one go, though all the CPUs in the
725 * group must be represented by the same leaf rcu_node structure.
726 * That structure's lock must be held upon entry, and it is released
730 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
732 __releases(rnp->lock)
734 /* Walk up the rcu_node hierarchy. */
736 if (!(rnp->qsmask & mask)) {
738 /* Our bit has already been cleared, so done. */
739 spin_unlock_irqrestore(&rnp->lock, flags);
742 rnp->qsmask &= ~mask;
743 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
745 /* Other bits still set at this level, so done. */
746 spin_unlock_irqrestore(&rnp->lock, flags);
750 if (rnp->parent == NULL) {
752 /* No more levels. Exit loop holding root lock. */
756 spin_unlock_irqrestore(&rnp->lock, flags);
758 spin_lock_irqsave(&rnp->lock, flags);
762 * Get here if we are the last CPU to pass through a quiescent
763 * state for this grace period. Invoke cpu_quiet_msk_finish()
764 * to clean up and start the next grace period if one is needed.
766 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
770 * Record a quiescent state for the specified CPU, which must either be
771 * the current CPU or an offline CPU. The lastcomp argument is used to
772 * make sure we are still in the grace period of interest. We don't want
773 * to end the current grace period based on quiescent states detected in
774 * an earlier grace period!
777 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
781 struct rcu_node *rnp;
784 spin_lock_irqsave(&rnp->lock, flags);
785 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
788 * Someone beat us to it for this grace period, so leave.
789 * The race with GP start is resolved by the fact that we
790 * hold the leaf rcu_node lock, so that the per-CPU bits
791 * cannot yet be initialized -- so we would simply find our
792 * CPU's bit already cleared in cpu_quiet_msk() if this race
795 rdp->passed_quiesc = 0; /* try again later! */
796 spin_unlock_irqrestore(&rnp->lock, flags);
800 if ((rnp->qsmask & mask) == 0) {
801 spin_unlock_irqrestore(&rnp->lock, flags);
806 * This GP can't end until cpu checks in, so all of our
807 * callbacks can be processed during the next GP.
809 rdp = rsp->rda[smp_processor_id()];
810 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
812 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
817 * Check to see if there is a new grace period of which this CPU
818 * is not yet aware, and if so, set up local rcu_data state for it.
819 * Otherwise, see if this CPU has just passed through its first
820 * quiescent state for this grace period, and record that fact if so.
823 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
825 /* If there is now a new grace period, record and return. */
826 if (check_for_new_grace_period(rsp, rdp))
830 * Does this CPU still need to do its part for current grace period?
831 * If no, return and let the other CPUs do their part as well.
833 if (!rdp->qs_pending)
837 * Was there a quiescent state since the beginning of the grace
838 * period? If no, then exit and wait for the next call.
840 if (!rdp->passed_quiesc)
843 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
844 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
847 #ifdef CONFIG_HOTPLUG_CPU
850 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
851 * and move all callbacks from the outgoing CPU to the current one.
853 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
859 struct rcu_data *rdp = rsp->rda[cpu];
860 struct rcu_data *rdp_me;
861 struct rcu_node *rnp;
863 /* Exclude any attempts to start a new grace period. */
864 spin_lock_irqsave(&rsp->onofflock, flags);
866 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
868 mask = rdp->grpmask; /* rnp->grplo is constant. */
870 spin_lock(&rnp->lock); /* irqs already disabled. */
871 rnp->qsmaskinit &= ~mask;
872 if (rnp->qsmaskinit != 0) {
873 spin_unlock(&rnp->lock); /* irqs remain disabled. */
877 spin_unlock(&rnp->lock); /* irqs remain disabled. */
879 } while (rnp != NULL);
880 lastcomp = rsp->completed;
882 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
884 /* Being offline is a quiescent state, so go record it. */
885 cpu_quiet(cpu, rsp, rdp, lastcomp);
888 * Move callbacks from the outgoing CPU to the running CPU.
889 * Note that the outgoing CPU is now quiscent, so it is now
890 * (uncharacteristically) safe to access its rcu_data structure.
891 * Note also that we must carefully retain the order of the
892 * outgoing CPU's callbacks in order for rcu_barrier() to work
893 * correctly. Finally, note that we start all the callbacks
894 * afresh, even those that have passed through a grace period
895 * and are therefore ready to invoke. The theory is that hotplug
896 * events are rare, and that if they are frequent enough to
897 * indefinitely delay callbacks, you have far worse things to
900 rdp_me = rsp->rda[smp_processor_id()];
901 if (rdp->nxtlist != NULL) {
902 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
903 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
905 for (i = 0; i < RCU_NEXT_SIZE; i++)
906 rdp->nxttail[i] = &rdp->nxtlist;
907 rdp_me->qlen += rdp->qlen;
910 local_irq_restore(flags);
914 * Remove the specified CPU from the RCU hierarchy and move any pending
915 * callbacks that it might have to the current CPU. This code assumes
916 * that at least one CPU in the system will remain running at all times.
917 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
919 static void rcu_offline_cpu(int cpu)
921 __rcu_offline_cpu(cpu, &rcu_sched_state);
922 __rcu_offline_cpu(cpu, &rcu_bh_state);
925 #else /* #ifdef CONFIG_HOTPLUG_CPU */
927 static void rcu_offline_cpu(int cpu)
931 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
934 * Invoke any RCU callbacks that have made it to the end of their grace
935 * period. Thottle as specified by rdp->blimit.
937 static void rcu_do_batch(struct rcu_data *rdp)
940 struct rcu_head *next, *list, **tail;
943 /* If no callbacks are ready, just return.*/
944 if (!cpu_has_callbacks_ready_to_invoke(rdp))
948 * Extract the list of ready callbacks, disabling to prevent
949 * races with call_rcu() from interrupt handlers.
951 local_irq_save(flags);
953 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
954 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
955 tail = rdp->nxttail[RCU_DONE_TAIL];
956 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
957 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
958 rdp->nxttail[count] = &rdp->nxtlist;
959 local_irq_restore(flags);
961 /* Invoke callbacks. */
968 if (++count >= rdp->blimit)
972 local_irq_save(flags);
974 /* Update count, and requeue any remaining callbacks. */
977 *tail = rdp->nxtlist;
979 for (count = 0; count < RCU_NEXT_SIZE; count++)
980 if (&rdp->nxtlist == rdp->nxttail[count])
981 rdp->nxttail[count] = tail;
986 /* Reinstate batch limit if we have worked down the excess. */
987 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
988 rdp->blimit = blimit;
990 local_irq_restore(flags);
992 /* Re-raise the RCU softirq if there are callbacks remaining. */
993 if (cpu_has_callbacks_ready_to_invoke(rdp))
994 raise_softirq(RCU_SOFTIRQ);
998 * Check to see if this CPU is in a non-context-switch quiescent state
999 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1000 * Also schedule the RCU softirq handler.
1002 * This function must be called with hardirqs disabled. It is normally
1003 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1004 * false, there is no point in invoking rcu_check_callbacks().
1006 void rcu_check_callbacks(int cpu, int user)
1008 if (!rcu_pending(cpu))
1009 return; /* if nothing for RCU to do. */
1011 (idle_cpu(cpu) && rcu_scheduler_active &&
1012 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1015 * Get here if this CPU took its interrupt from user
1016 * mode or from the idle loop, and if this is not a
1017 * nested interrupt. In this case, the CPU is in
1018 * a quiescent state, so note it.
1020 * No memory barrier is required here because both
1021 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1022 * variables that other CPUs neither access nor modify,
1023 * at least not while the corresponding CPU is online.
1029 } else if (!in_softirq()) {
1032 * Get here if this CPU did not take its interrupt from
1033 * softirq, in other words, if it is not interrupting
1034 * a rcu_bh read-side critical section. This is an _bh
1035 * critical section, so note it.
1040 rcu_preempt_check_callbacks(cpu);
1041 raise_softirq(RCU_SOFTIRQ);
1047 * Scan the leaf rcu_node structures, processing dyntick state for any that
1048 * have not yet encountered a quiescent state, using the function specified.
1049 * Returns 1 if the current grace period ends while scanning (possibly
1050 * because we made it end).
1052 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1053 int (*f)(struct rcu_data *))
1057 unsigned long flags;
1059 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
1060 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1062 for (; rnp_cur < rnp_end; rnp_cur++) {
1064 spin_lock_irqsave(&rnp_cur->lock, flags);
1065 if (rsp->completed != lastcomp) {
1066 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1069 if (rnp_cur->qsmask == 0) {
1070 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1073 cpu = rnp_cur->grplo;
1075 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1076 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1079 if (mask != 0 && rsp->completed == lastcomp) {
1081 /* cpu_quiet_msk() releases rnp_cur->lock. */
1082 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1085 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1091 * Force quiescent states on reluctant CPUs, and also detect which
1092 * CPUs are in dyntick-idle mode.
1094 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1096 unsigned long flags;
1098 struct rcu_node *rnp = rcu_get_root(rsp);
1101 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
1102 return; /* No grace period in progress, nothing to force. */
1103 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1104 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1105 return; /* Someone else is already on the job. */
1108 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1109 goto unlock_ret; /* no emergency and done recently. */
1111 spin_lock(&rnp->lock);
1112 lastcomp = rsp->completed;
1113 signaled = rsp->signaled;
1114 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1115 if (lastcomp == rsp->gpnum) {
1116 rsp->n_force_qs_ngp++;
1117 spin_unlock(&rnp->lock);
1118 goto unlock_ret; /* no GP in progress, time updated. */
1120 spin_unlock(&rnp->lock);
1124 break; /* grace period still initializing, ignore. */
1126 case RCU_SAVE_DYNTICK:
1128 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1129 break; /* So gcc recognizes the dead code. */
1131 /* Record dyntick-idle state. */
1132 if (rcu_process_dyntick(rsp, lastcomp,
1133 dyntick_save_progress_counter))
1136 /* Update state, record completion counter. */
1137 spin_lock(&rnp->lock);
1138 if (lastcomp == rsp->completed) {
1139 rsp->signaled = RCU_FORCE_QS;
1140 dyntick_record_completed(rsp, lastcomp);
1142 spin_unlock(&rnp->lock);
1147 /* Check dyntick-idle state, send IPI to laggarts. */
1148 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1149 rcu_implicit_dynticks_qs))
1152 /* Leave state in case more forcing is required. */
1157 spin_unlock_irqrestore(&rsp->fqslock, flags);
1160 #else /* #ifdef CONFIG_SMP */
1162 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1167 #endif /* #else #ifdef CONFIG_SMP */
1170 * This does the RCU processing work from softirq context for the
1171 * specified rcu_state and rcu_data structures. This may be called
1172 * only from the CPU to whom the rdp belongs.
1175 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1177 unsigned long flags;
1179 WARN_ON_ONCE(rdp->beenonline == 0);
1182 * If an RCU GP has gone long enough, go check for dyntick
1183 * idle CPUs and, if needed, send resched IPIs.
1185 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1186 force_quiescent_state(rsp, 1);
1189 * Advance callbacks in response to end of earlier grace
1190 * period that some other CPU ended.
1192 rcu_process_gp_end(rsp, rdp);
1194 /* Update RCU state based on any recent quiescent states. */
1195 rcu_check_quiescent_state(rsp, rdp);
1197 /* Does this CPU require a not-yet-started grace period? */
1198 if (cpu_needs_another_gp(rsp, rdp)) {
1199 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1200 rcu_start_gp(rsp, flags); /* releases above lock */
1203 /* If there are callbacks ready, invoke them. */
1208 * Do softirq processing for the current CPU.
1210 static void rcu_process_callbacks(struct softirq_action *unused)
1213 * Memory references from any prior RCU read-side critical sections
1214 * executed by the interrupted code must be seen before any RCU
1215 * grace-period manipulations below.
1217 smp_mb(); /* See above block comment. */
1219 __rcu_process_callbacks(&rcu_sched_state,
1220 &__get_cpu_var(rcu_sched_data));
1221 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1222 rcu_preempt_process_callbacks();
1225 * Memory references from any later RCU read-side critical sections
1226 * executed by the interrupted code must be seen after any RCU
1227 * grace-period manipulations above.
1229 smp_mb(); /* See above block comment. */
1233 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1234 struct rcu_state *rsp)
1236 unsigned long flags;
1237 struct rcu_data *rdp;
1242 smp_mb(); /* Ensure RCU update seen before callback registry. */
1245 * Opportunistically note grace-period endings and beginnings.
1246 * Note that we might see a beginning right after we see an
1247 * end, but never vice versa, since this CPU has to pass through
1248 * a quiescent state betweentimes.
1250 local_irq_save(flags);
1251 rdp = rsp->rda[smp_processor_id()];
1252 rcu_process_gp_end(rsp, rdp);
1253 check_for_new_grace_period(rsp, rdp);
1255 /* Add the callback to our list. */
1256 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1257 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1259 /* Start a new grace period if one not already started. */
1260 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
1261 unsigned long nestflag;
1262 struct rcu_node *rnp_root = rcu_get_root(rsp);
1264 spin_lock_irqsave(&rnp_root->lock, nestflag);
1265 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1268 /* Force the grace period if too many callbacks or too long waiting. */
1269 if (unlikely(++rdp->qlen > qhimark)) {
1270 rdp->blimit = LONG_MAX;
1271 force_quiescent_state(rsp, 0);
1272 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1273 force_quiescent_state(rsp, 1);
1274 local_irq_restore(flags);
1278 * Queue an RCU-sched callback for invocation after a grace period.
1280 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1282 __call_rcu(head, func, &rcu_sched_state);
1284 EXPORT_SYMBOL_GPL(call_rcu_sched);
1287 * Queue an RCU for invocation after a quicker grace period.
1289 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1291 __call_rcu(head, func, &rcu_bh_state);
1293 EXPORT_SYMBOL_GPL(call_rcu_bh);
1296 * Check to see if there is any immediate RCU-related work to be done
1297 * by the current CPU, for the specified type of RCU, returning 1 if so.
1298 * The checks are in order of increasing expense: checks that can be
1299 * carried out against CPU-local state are performed first. However,
1300 * we must check for CPU stalls first, else we might not get a chance.
1302 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1304 rdp->n_rcu_pending++;
1306 /* Check for CPU stalls, if enabled. */
1307 check_cpu_stall(rsp, rdp);
1309 /* Is the RCU core waiting for a quiescent state from this CPU? */
1310 if (rdp->qs_pending) {
1311 rdp->n_rp_qs_pending++;
1315 /* Does this CPU have callbacks ready to invoke? */
1316 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1317 rdp->n_rp_cb_ready++;
1321 /* Has RCU gone idle with this CPU needing another grace period? */
1322 if (cpu_needs_another_gp(rsp, rdp)) {
1323 rdp->n_rp_cpu_needs_gp++;
1327 /* Has another RCU grace period completed? */
1328 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1329 rdp->n_rp_gp_completed++;
1333 /* Has a new RCU grace period started? */
1334 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1335 rdp->n_rp_gp_started++;
1339 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1340 if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
1341 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1342 rdp->n_rp_need_fqs++;
1347 rdp->n_rp_need_nothing++;
1352 * Check to see if there is any immediate RCU-related work to be done
1353 * by the current CPU, returning 1 if so. This function is part of the
1354 * RCU implementation; it is -not- an exported member of the RCU API.
1356 static int rcu_pending(int cpu)
1358 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1359 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1360 rcu_preempt_pending(cpu);
1364 * Check to see if any future RCU-related work will need to be done
1365 * by the current CPU, even if none need be done immediately, returning
1366 * 1 if so. This function is part of the RCU implementation; it is -not-
1367 * an exported member of the RCU API.
1369 int rcu_needs_cpu(int cpu)
1371 /* RCU callbacks either ready or pending? */
1372 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1373 per_cpu(rcu_bh_data, cpu).nxtlist ||
1374 rcu_preempt_needs_cpu(cpu);
1378 * Do boot-time initialization of a CPU's per-CPU RCU data.
1381 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1383 unsigned long flags;
1385 struct rcu_data *rdp = rsp->rda[cpu];
1386 struct rcu_node *rnp = rcu_get_root(rsp);
1388 /* Set up local state, ensuring consistent view of global state. */
1389 spin_lock_irqsave(&rnp->lock, flags);
1390 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1391 rdp->nxtlist = NULL;
1392 for (i = 0; i < RCU_NEXT_SIZE; i++)
1393 rdp->nxttail[i] = &rdp->nxtlist;
1396 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1397 #endif /* #ifdef CONFIG_NO_HZ */
1399 spin_unlock_irqrestore(&rnp->lock, flags);
1403 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1404 * offline event can be happening at a given time. Note also that we
1405 * can accept some slop in the rsp->completed access due to the fact
1406 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1408 static void __cpuinit
1409 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1411 unsigned long flags;
1414 struct rcu_data *rdp = rsp->rda[cpu];
1415 struct rcu_node *rnp = rcu_get_root(rsp);
1417 /* Set up local state, ensuring consistent view of global state. */
1418 spin_lock_irqsave(&rnp->lock, flags);
1419 lastcomp = rsp->completed;
1420 rdp->completed = lastcomp;
1421 rdp->gpnum = lastcomp;
1422 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1423 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1424 rdp->beenonline = 1; /* We have now been online. */
1425 rdp->preemptable = preemptable;
1426 rdp->passed_quiesc_completed = lastcomp - 1;
1427 rdp->blimit = blimit;
1428 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1431 * A new grace period might start here. If so, we won't be part
1432 * of it, but that is OK, as we are currently in a quiescent state.
1435 /* Exclude any attempts to start a new GP on large systems. */
1436 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1438 /* Add CPU to rcu_node bitmasks. */
1440 mask = rdp->grpmask;
1442 /* Exclude any attempts to start a new GP on small systems. */
1443 spin_lock(&rnp->lock); /* irqs already disabled. */
1444 rnp->qsmaskinit |= mask;
1445 mask = rnp->grpmask;
1446 spin_unlock(&rnp->lock); /* irqs already disabled. */
1448 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1450 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1453 * A new grace period might start here. If so, we will be part of
1454 * it, and its gpnum will be greater than ours, so we will
1455 * participate. It is also possible for the gpnum to have been
1456 * incremented before this function was called, and the bitmasks
1457 * to not be filled out until now, in which case we will also
1458 * participate due to our gpnum being behind.
1461 /* Since it is coming online, the CPU is in a quiescent state. */
1462 cpu_quiet(cpu, rsp, rdp, lastcomp);
1463 local_irq_restore(flags);
1466 static void __cpuinit rcu_online_cpu(int cpu)
1468 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1469 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1470 rcu_preempt_init_percpu_data(cpu);
1474 * Handle CPU online/offline notification events.
1476 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1477 unsigned long action, void *hcpu)
1479 long cpu = (long)hcpu;
1482 case CPU_UP_PREPARE:
1483 case CPU_UP_PREPARE_FROZEN:
1484 rcu_online_cpu(cpu);
1487 case CPU_DEAD_FROZEN:
1488 case CPU_UP_CANCELED:
1489 case CPU_UP_CANCELED_FROZEN:
1490 rcu_offline_cpu(cpu);
1499 * Compute the per-level fanout, either using the exact fanout specified
1500 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1502 #ifdef CONFIG_RCU_FANOUT_EXACT
1503 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1507 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1508 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1510 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1511 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1518 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1519 ccur = rsp->levelcnt[i];
1520 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1524 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1527 * Helper function for rcu_init() that initializes one rcu_state structure.
1529 static void __init rcu_init_one(struct rcu_state *rsp)
1534 struct rcu_node *rnp;
1536 /* Initialize the level-tracking arrays. */
1538 for (i = 1; i < NUM_RCU_LVLS; i++)
1539 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1540 rcu_init_levelspread(rsp);
1542 /* Initialize the elements themselves, starting from the leaves. */
1544 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1545 cpustride *= rsp->levelspread[i];
1546 rnp = rsp->level[i];
1547 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1548 spin_lock_init(&rnp->lock);
1551 rnp->qsmaskinit = 0;
1552 rnp->grplo = j * cpustride;
1553 rnp->grphi = (j + 1) * cpustride - 1;
1554 if (rnp->grphi >= NR_CPUS)
1555 rnp->grphi = NR_CPUS - 1;
1561 rnp->grpnum = j % rsp->levelspread[i - 1];
1562 rnp->grpmask = 1UL << rnp->grpnum;
1563 rnp->parent = rsp->level[i - 1] +
1564 j / rsp->levelspread[i - 1];
1567 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1568 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1574 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1575 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1578 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1580 rcu_init_one(rsp); \
1581 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1583 for_each_possible_cpu(i) { \
1584 if (i > rnp[j].grphi) \
1586 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1587 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1588 rcu_boot_init_percpu_data(i, rsp); \
1592 #ifdef CONFIG_TREE_PREEMPT_RCU
1594 void __init __rcu_init_preempt(void)
1596 int i; /* All used by RCU_INIT_FLAVOR(). */
1598 struct rcu_node *rnp;
1600 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
1603 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1605 void __init __rcu_init_preempt(void)
1609 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1611 void __init __rcu_init(void)
1613 int i; /* All used by RCU_INIT_FLAVOR(). */
1615 struct rcu_node *rnp;
1617 rcu_bootup_announce();
1618 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1619 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1620 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1621 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1622 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1623 __rcu_init_preempt();
1624 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1627 module_param(blimit, int, 0);
1628 module_param(qhimark, int, 0);
1629 module_param(qlowmark, int, 0);