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_offline_cpu(int cpu, struct rcu_state *rsp);
88 static void __rcu_process_callbacks(struct rcu_state *rsp,
89 struct rcu_data *rdp);
90 static void __call_rcu(struct rcu_head *head,
91 void (*func)(struct rcu_head *rcu),
92 struct rcu_state *rsp);
93 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp);
94 static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp,
97 #include "rcutree_plugin.h"
100 * Note a quiescent state. Because we do not need to know
101 * how many quiescent states passed, just if there was at least
102 * one since the start of the grace period, this just sets a flag.
104 void rcu_sched_qs(int cpu)
107 struct rcu_data *rdp;
109 local_irq_save(flags);
110 rdp = &per_cpu(rcu_sched_data, cpu);
111 rdp->passed_quiesc = 1;
112 rdp->passed_quiesc_completed = rdp->completed;
114 local_irq_restore(flags);
117 void rcu_bh_qs(int cpu)
120 struct rcu_data *rdp;
122 local_irq_save(flags);
123 rdp = &per_cpu(rcu_bh_data, cpu);
124 rdp->passed_quiesc = 1;
125 rdp->passed_quiesc_completed = rdp->completed;
126 local_irq_restore(flags);
130 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
131 .dynticks_nesting = 1,
134 #endif /* #ifdef CONFIG_NO_HZ */
136 static int blimit = 10; /* Maximum callbacks per softirq. */
137 static int qhimark = 10000; /* If this many pending, ignore blimit. */
138 static int qlowmark = 100; /* Once only this many pending, use blimit. */
140 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
141 static int rcu_pending(int cpu);
144 * Return the number of RCU-sched batches processed thus far for debug & stats.
146 long rcu_batches_completed_sched(void)
148 return rcu_sched_state.completed;
150 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
153 * Return the number of RCU BH batches processed thus far for debug & stats.
155 long rcu_batches_completed_bh(void)
157 return rcu_bh_state.completed;
159 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
162 * Does the CPU have callbacks ready to be invoked?
165 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
167 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
171 * Does the current CPU require a yet-as-unscheduled grace period?
174 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
176 /* ACCESS_ONCE() because we are accessing outside of lock. */
177 return *rdp->nxttail[RCU_DONE_TAIL] &&
178 ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
182 * Return the root node of the specified rcu_state structure.
184 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
186 return &rsp->node[0];
192 * If the specified CPU is offline, tell the caller that it is in
193 * a quiescent state. Otherwise, whack it with a reschedule IPI.
194 * Grace periods can end up waiting on an offline CPU when that
195 * CPU is in the process of coming online -- it will be added to the
196 * rcu_node bitmasks before it actually makes it online. The same thing
197 * can happen while a CPU is in the process of coming online. Because this
198 * race is quite rare, we check for it after detecting that the grace
199 * period has been delayed rather than checking each and every CPU
200 * each and every time we start a new grace period.
202 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
205 * If the CPU is offline, it is in a quiescent state. We can
206 * trust its state not to change because interrupts are disabled.
208 if (cpu_is_offline(rdp->cpu)) {
213 /* If preemptable RCU, no point in sending reschedule IPI. */
214 if (rdp->preemptable)
217 /* The CPU is online, so send it a reschedule IPI. */
218 if (rdp->cpu != smp_processor_id())
219 smp_send_reschedule(rdp->cpu);
226 #endif /* #ifdef CONFIG_SMP */
229 static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5);
232 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
234 * Enter nohz mode, in other words, -leave- the mode in which RCU
235 * read-side critical sections can occur. (Though RCU read-side
236 * critical sections can occur in irq handlers in nohz mode, a possibility
237 * handled by rcu_irq_enter() and rcu_irq_exit()).
239 void rcu_enter_nohz(void)
242 struct rcu_dynticks *rdtp;
244 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
245 local_irq_save(flags);
246 rdtp = &__get_cpu_var(rcu_dynticks);
248 rdtp->dynticks_nesting--;
249 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
250 local_irq_restore(flags);
254 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
256 * Exit nohz mode, in other words, -enter- the mode in which RCU
257 * read-side critical sections normally occur.
259 void rcu_exit_nohz(void)
262 struct rcu_dynticks *rdtp;
264 local_irq_save(flags);
265 rdtp = &__get_cpu_var(rcu_dynticks);
267 rdtp->dynticks_nesting++;
268 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
269 local_irq_restore(flags);
270 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
274 * rcu_nmi_enter - inform RCU of entry to NMI context
276 * If the CPU was idle with dynamic ticks active, and there is no
277 * irq handler running, this updates rdtp->dynticks_nmi to let the
278 * RCU grace-period handling know that the CPU is active.
280 void rcu_nmi_enter(void)
282 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
284 if (rdtp->dynticks & 0x1)
286 rdtp->dynticks_nmi++;
287 WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs);
288 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
292 * rcu_nmi_exit - inform RCU of exit from NMI context
294 * If the CPU was idle with dynamic ticks active, and there is no
295 * irq handler running, this updates rdtp->dynticks_nmi to let the
296 * RCU grace-period handling know that the CPU is no longer active.
298 void rcu_nmi_exit(void)
300 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
302 if (rdtp->dynticks & 0x1)
304 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
305 rdtp->dynticks_nmi++;
306 WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs);
310 * rcu_irq_enter - inform RCU of entry to hard irq context
312 * If the CPU was idle with dynamic ticks active, this updates the
313 * rdtp->dynticks to let the RCU handling know that the CPU is active.
315 void rcu_irq_enter(void)
317 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
319 if (rdtp->dynticks_nesting++)
322 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
323 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
327 * rcu_irq_exit - inform RCU of exit from hard irq context
329 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
330 * to put let the RCU handling be aware that the CPU is going back to idle
333 void rcu_irq_exit(void)
335 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
337 if (--rdtp->dynticks_nesting)
339 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
341 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
343 /* If the interrupt queued a callback, get out of dyntick mode. */
344 if (__get_cpu_var(rcu_sched_data).nxtlist ||
345 __get_cpu_var(rcu_bh_data).nxtlist)
350 * Record the specified "completed" value, which is later used to validate
351 * dynticks counter manipulations. Specify "rsp->completed - 1" to
352 * unconditionally invalidate any future dynticks manipulations (which is
353 * useful at the beginning of a grace period).
355 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
357 rsp->dynticks_completed = comp;
363 * Recall the previously recorded value of the completion for dynticks.
365 static long dyntick_recall_completed(struct rcu_state *rsp)
367 return rsp->dynticks_completed;
371 * Snapshot the specified CPU's dynticks counter so that we can later
372 * credit them with an implicit quiescent state. Return 1 if this CPU
373 * is already in a quiescent state courtesy of dynticks idle mode.
375 static int dyntick_save_progress_counter(struct rcu_data *rdp)
381 snap = rdp->dynticks->dynticks;
382 snap_nmi = rdp->dynticks->dynticks_nmi;
383 smp_mb(); /* Order sampling of snap with end of grace period. */
384 rdp->dynticks_snap = snap;
385 rdp->dynticks_nmi_snap = snap_nmi;
386 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
393 * Return true if the specified CPU has passed through a quiescent
394 * state by virtue of being in or having passed through an dynticks
395 * idle state since the last call to dyntick_save_progress_counter()
398 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
405 curr = rdp->dynticks->dynticks;
406 snap = rdp->dynticks_snap;
407 curr_nmi = rdp->dynticks->dynticks_nmi;
408 snap_nmi = rdp->dynticks_nmi_snap;
409 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
412 * If the CPU passed through or entered a dynticks idle phase with
413 * no active irq/NMI handlers, then we can safely pretend that the CPU
414 * already acknowledged the request to pass through a quiescent
415 * state. Either way, that CPU cannot possibly be in an RCU
416 * read-side critical section that started before the beginning
417 * of the current RCU grace period.
419 if ((curr != snap || (curr & 0x1) == 0) &&
420 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
425 /* Go check for the CPU being offline. */
426 return rcu_implicit_offline_qs(rdp);
429 #endif /* #ifdef CONFIG_SMP */
431 #else /* #ifdef CONFIG_NO_HZ */
433 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
440 * If there are no dynticks, then the only way that a CPU can passively
441 * be in a quiescent state is to be offline. Unlike dynticks idle, which
442 * is a point in time during the prior (already finished) grace period,
443 * an offline CPU is always in a quiescent state, and thus can be
444 * unconditionally applied. So just return the current value of completed.
446 static long dyntick_recall_completed(struct rcu_state *rsp)
448 return rsp->completed;
451 static int dyntick_save_progress_counter(struct rcu_data *rdp)
456 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
458 return rcu_implicit_offline_qs(rdp);
461 #endif /* #ifdef CONFIG_SMP */
463 #endif /* #else #ifdef CONFIG_NO_HZ */
465 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
467 static void record_gp_stall_check_time(struct rcu_state *rsp)
469 rsp->gp_start = jiffies;
470 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
473 static void print_other_cpu_stall(struct rcu_state *rsp)
478 struct rcu_node *rnp = rcu_get_root(rsp);
479 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
480 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
482 /* Only let one CPU complain about others per time interval. */
484 spin_lock_irqsave(&rnp->lock, flags);
485 delta = jiffies - rsp->jiffies_stall;
486 if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) {
487 spin_unlock_irqrestore(&rnp->lock, flags);
490 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
491 spin_unlock_irqrestore(&rnp->lock, flags);
493 /* OK, time to rat on our buddy... */
495 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
496 for (; rnp_cur < rnp_end; rnp_cur++) {
497 rcu_print_task_stall(rnp);
498 if (rnp_cur->qsmask == 0)
500 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
501 if (rnp_cur->qsmask & (1UL << cpu))
502 printk(" %d", rnp_cur->grplo + cpu);
504 printk(" (detected by %d, t=%ld jiffies)\n",
505 smp_processor_id(), (long)(jiffies - rsp->gp_start));
506 force_quiescent_state(rsp, 0); /* Kick them all. */
509 static void print_cpu_stall(struct rcu_state *rsp)
512 struct rcu_node *rnp = rcu_get_root(rsp);
514 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
515 smp_processor_id(), jiffies - rsp->gp_start);
517 spin_lock_irqsave(&rnp->lock, flags);
518 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
520 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
521 spin_unlock_irqrestore(&rnp->lock, flags);
522 set_need_resched(); /* kick ourselves to get things going. */
525 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
528 struct rcu_node *rnp;
530 delta = jiffies - rsp->jiffies_stall;
532 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
534 /* We haven't checked in, so go dump stack. */
535 print_cpu_stall(rsp);
537 } else if (rsp->gpnum != rsp->completed &&
538 delta >= RCU_STALL_RAT_DELAY) {
540 /* They had two time units to dump stack, so complain. */
541 print_other_cpu_stall(rsp);
545 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
547 static void record_gp_stall_check_time(struct rcu_state *rsp)
551 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
555 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
558 * Update CPU-local rcu_data state to record the newly noticed grace period.
559 * This is used both when we started the grace period and when we notice
560 * that someone else started the grace period.
562 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
565 rdp->passed_quiesc = 0;
566 rdp->gpnum = rsp->gpnum;
570 * Did someone else start a new RCU grace period start since we last
571 * checked? Update local state appropriately if so. Must be called
572 * on the CPU corresponding to rdp.
575 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
580 local_irq_save(flags);
581 if (rdp->gpnum != rsp->gpnum) {
582 note_new_gpnum(rsp, rdp);
585 local_irq_restore(flags);
590 * Start a new RCU grace period if warranted, re-initializing the hierarchy
591 * in preparation for detecting the next grace period. The caller must hold
592 * the root node's ->lock, which is released before return. Hard irqs must
596 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
597 __releases(rcu_get_root(rsp)->lock)
599 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
600 struct rcu_node *rnp = rcu_get_root(rsp);
601 struct rcu_node *rnp_cur;
602 struct rcu_node *rnp_end;
604 if (!cpu_needs_another_gp(rsp, rdp)) {
605 spin_unlock_irqrestore(&rnp->lock, flags);
609 /* Advance to a new grace period and initialize state. */
611 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
612 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
613 record_gp_stall_check_time(rsp);
614 dyntick_record_completed(rsp, rsp->completed - 1);
615 note_new_gpnum(rsp, rdp);
618 * Because we are first, we know that all our callbacks will
619 * be covered by this upcoming grace period, even the ones
620 * that were registered arbitrarily recently.
622 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
623 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
625 /* Special-case the common single-level case. */
626 if (NUM_RCU_NODES == 1) {
627 rnp->qsmask = rnp->qsmaskinit;
628 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
629 spin_unlock_irqrestore(&rnp->lock, flags);
633 spin_unlock(&rnp->lock); /* leave irqs disabled. */
636 /* Exclude any concurrent CPU-hotplug operations. */
637 spin_lock(&rsp->onofflock); /* irqs already disabled. */
640 * Set the quiescent-state-needed bits in all the non-leaf RCU
641 * nodes for all currently online CPUs. This operation relies
642 * on the layout of the hierarchy within the rsp->node[] array.
643 * Note that other CPUs will access only the leaves of the
644 * hierarchy, which still indicate that no grace period is in
645 * progress. In addition, we have excluded CPU-hotplug operations.
647 * We therefore do not need to hold any locks. Any required
648 * memory barriers will be supplied by the locks guarding the
649 * leaf rcu_nodes in the hierarchy.
652 rnp_end = rsp->level[NUM_RCU_LVLS - 1];
653 for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++)
654 rnp_cur->qsmask = rnp_cur->qsmaskinit;
657 * Now set up the leaf nodes. Here we must be careful. First,
658 * we need to hold the lock in order to exclude other CPUs, which
659 * might be contending for the leaf nodes' locks. Second, as
660 * soon as we initialize a given leaf node, its CPUs might run
661 * up the rest of the hierarchy. We must therefore acquire locks
662 * for each node that we touch during this stage. (But we still
663 * are excluding CPU-hotplug operations.)
665 * Note that the grace period cannot complete until we finish
666 * the initialization process, as there will be at least one
667 * qsmask bit set in the root node until that time, namely the
668 * one corresponding to this CPU.
670 rnp_end = &rsp->node[NUM_RCU_NODES];
671 rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
672 for (; rnp_cur < rnp_end; rnp_cur++) {
673 spin_lock(&rnp_cur->lock); /* irqs already disabled. */
674 rnp_cur->qsmask = rnp_cur->qsmaskinit;
675 spin_unlock(&rnp_cur->lock); /* irqs already disabled. */
678 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
679 spin_unlock_irqrestore(&rsp->onofflock, flags);
683 * Advance this CPU's callbacks, but only if the current grace period
684 * has ended. This may be called only from the CPU to whom the rdp
688 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
693 local_irq_save(flags);
694 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
696 /* Did another grace period end? */
697 if (rdp->completed != completed_snap) {
699 /* Advance callbacks. No harm if list empty. */
700 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
701 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
702 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
704 /* Remember that we saw this grace-period completion. */
705 rdp->completed = completed_snap;
707 local_irq_restore(flags);
711 * Clean up after the prior grace period and let rcu_start_gp() start up
712 * the next grace period if one is needed. Note that the caller must
713 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
715 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
716 __releases(rnp->lock)
718 rsp->completed = rsp->gpnum;
719 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
720 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
724 * Similar to cpu_quiet(), for which it is a helper function. Allows
725 * a group of CPUs to be quieted at one go, though all the CPUs in the
726 * group must be represented by the same leaf rcu_node structure.
727 * That structure's lock must be held upon entry, and it is released
731 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
733 __releases(rnp->lock)
735 /* Walk up the rcu_node hierarchy. */
737 if (!(rnp->qsmask & mask)) {
739 /* Our bit has already been cleared, so done. */
740 spin_unlock_irqrestore(&rnp->lock, flags);
743 rnp->qsmask &= ~mask;
744 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
746 /* Other bits still set at this level, so done. */
747 spin_unlock_irqrestore(&rnp->lock, flags);
751 if (rnp->parent == NULL) {
753 /* No more levels. Exit loop holding root lock. */
757 spin_unlock_irqrestore(&rnp->lock, flags);
759 spin_lock_irqsave(&rnp->lock, flags);
763 * Get here if we are the last CPU to pass through a quiescent
764 * state for this grace period. Invoke cpu_quiet_msk_finish()
765 * to clean up and start the next grace period if one is needed.
767 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
771 * Record a quiescent state for the specified CPU, which must either be
772 * the current CPU or an offline CPU. The lastcomp argument is used to
773 * make sure we are still in the grace period of interest. We don't want
774 * to end the current grace period based on quiescent states detected in
775 * an earlier grace period!
778 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
782 struct rcu_node *rnp;
785 spin_lock_irqsave(&rnp->lock, flags);
786 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
789 * Someone beat us to it for this grace period, so leave.
790 * The race with GP start is resolved by the fact that we
791 * hold the leaf rcu_node lock, so that the per-CPU bits
792 * cannot yet be initialized -- so we would simply find our
793 * CPU's bit already cleared in cpu_quiet_msk() if this race
796 rdp->passed_quiesc = 0; /* try again later! */
797 spin_unlock_irqrestore(&rnp->lock, flags);
801 if ((rnp->qsmask & mask) == 0) {
802 spin_unlock_irqrestore(&rnp->lock, flags);
807 * This GP can't end until cpu checks in, so all of our
808 * callbacks can be processed during the next GP.
810 rdp = rsp->rda[smp_processor_id()];
811 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
813 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
818 * Check to see if there is a new grace period of which this CPU
819 * is not yet aware, and if so, set up local rcu_data state for it.
820 * Otherwise, see if this CPU has just passed through its first
821 * quiescent state for this grace period, and record that fact if so.
824 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
826 /* If there is now a new grace period, record and return. */
827 if (check_for_new_grace_period(rsp, rdp))
831 * Does this CPU still need to do its part for current grace period?
832 * If no, return and let the other CPUs do their part as well.
834 if (!rdp->qs_pending)
838 * Was there a quiescent state since the beginning of the grace
839 * period? If no, then exit and wait for the next call.
841 if (!rdp->passed_quiesc)
844 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
845 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
848 #ifdef CONFIG_HOTPLUG_CPU
851 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
852 * and move all callbacks from the outgoing CPU to the current one.
854 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
860 struct rcu_data *rdp = rsp->rda[cpu];
861 struct rcu_data *rdp_me;
862 struct rcu_node *rnp;
864 /* Exclude any attempts to start a new grace period. */
865 spin_lock_irqsave(&rsp->onofflock, flags);
867 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
869 mask = rdp->grpmask; /* rnp->grplo is constant. */
871 spin_lock(&rnp->lock); /* irqs already disabled. */
872 rnp->qsmaskinit &= ~mask;
873 if (rnp->qsmaskinit != 0) {
874 spin_unlock(&rnp->lock); /* irqs remain disabled. */
878 spin_unlock(&rnp->lock); /* irqs remain disabled. */
880 } while (rnp != NULL);
881 lastcomp = rsp->completed;
883 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
885 /* Being offline is a quiescent state, so go record it. */
886 cpu_quiet(cpu, rsp, rdp, lastcomp);
889 * Move callbacks from the outgoing CPU to the running CPU.
890 * Note that the outgoing CPU is now quiscent, so it is now
891 * (uncharacteristically) safe to access its rcu_data structure.
892 * Note also that we must carefully retain the order of the
893 * outgoing CPU's callbacks in order for rcu_barrier() to work
894 * correctly. Finally, note that we start all the callbacks
895 * afresh, even those that have passed through a grace period
896 * and are therefore ready to invoke. The theory is that hotplug
897 * events are rare, and that if they are frequent enough to
898 * indefinitely delay callbacks, you have far worse things to
901 rdp_me = rsp->rda[smp_processor_id()];
902 if (rdp->nxtlist != NULL) {
903 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
904 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
906 for (i = 0; i < RCU_NEXT_SIZE; i++)
907 rdp->nxttail[i] = &rdp->nxtlist;
908 rdp_me->qlen += rdp->qlen;
911 local_irq_restore(flags);
915 * Remove the specified CPU from the RCU hierarchy and move any pending
916 * callbacks that it might have to the current CPU. This code assumes
917 * that at least one CPU in the system will remain running at all times.
918 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
920 static void rcu_offline_cpu(int cpu)
922 __rcu_offline_cpu(cpu, &rcu_sched_state);
923 __rcu_offline_cpu(cpu, &rcu_bh_state);
924 rcu_preempt_offline_cpu(cpu);
927 #else /* #ifdef CONFIG_HOTPLUG_CPU */
929 static void rcu_offline_cpu(int cpu)
933 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
936 * Invoke any RCU callbacks that have made it to the end of their grace
937 * period. Thottle as specified by rdp->blimit.
939 static void rcu_do_batch(struct rcu_data *rdp)
942 struct rcu_head *next, *list, **tail;
945 /* If no callbacks are ready, just return.*/
946 if (!cpu_has_callbacks_ready_to_invoke(rdp))
950 * Extract the list of ready callbacks, disabling to prevent
951 * races with call_rcu() from interrupt handlers.
953 local_irq_save(flags);
955 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
956 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
957 tail = rdp->nxttail[RCU_DONE_TAIL];
958 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
959 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
960 rdp->nxttail[count] = &rdp->nxtlist;
961 local_irq_restore(flags);
963 /* Invoke callbacks. */
970 if (++count >= rdp->blimit)
974 local_irq_save(flags);
976 /* Update count, and requeue any remaining callbacks. */
979 *tail = rdp->nxtlist;
981 for (count = 0; count < RCU_NEXT_SIZE; count++)
982 if (&rdp->nxtlist == rdp->nxttail[count])
983 rdp->nxttail[count] = tail;
988 /* Reinstate batch limit if we have worked down the excess. */
989 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
990 rdp->blimit = blimit;
992 local_irq_restore(flags);
994 /* Re-raise the RCU softirq if there are callbacks remaining. */
995 if (cpu_has_callbacks_ready_to_invoke(rdp))
996 raise_softirq(RCU_SOFTIRQ);
1000 * Check to see if this CPU is in a non-context-switch quiescent state
1001 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1002 * Also schedule the RCU softirq handler.
1004 * This function must be called with hardirqs disabled. It is normally
1005 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1006 * false, there is no point in invoking rcu_check_callbacks().
1008 void rcu_check_callbacks(int cpu, int user)
1010 if (!rcu_pending(cpu))
1011 return; /* if nothing for RCU to do. */
1013 (idle_cpu(cpu) && rcu_scheduler_active &&
1014 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1017 * Get here if this CPU took its interrupt from user
1018 * mode or from the idle loop, and if this is not a
1019 * nested interrupt. In this case, the CPU is in
1020 * a quiescent state, so note it.
1022 * No memory barrier is required here because both
1023 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1024 * variables that other CPUs neither access nor modify,
1025 * at least not while the corresponding CPU is online.
1031 } else if (!in_softirq()) {
1034 * Get here if this CPU did not take its interrupt from
1035 * softirq, in other words, if it is not interrupting
1036 * a rcu_bh read-side critical section. This is an _bh
1037 * critical section, so note it.
1042 rcu_preempt_check_callbacks(cpu);
1043 raise_softirq(RCU_SOFTIRQ);
1049 * Scan the leaf rcu_node structures, processing dyntick state for any that
1050 * have not yet encountered a quiescent state, using the function specified.
1051 * Returns 1 if the current grace period ends while scanning (possibly
1052 * because we made it end).
1054 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1055 int (*f)(struct rcu_data *))
1059 unsigned long flags;
1061 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
1062 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1064 for (; rnp_cur < rnp_end; rnp_cur++) {
1066 spin_lock_irqsave(&rnp_cur->lock, flags);
1067 if (rsp->completed != lastcomp) {
1068 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1071 if (rnp_cur->qsmask == 0) {
1072 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1075 cpu = rnp_cur->grplo;
1077 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1078 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1081 if (mask != 0 && rsp->completed == lastcomp) {
1083 /* cpu_quiet_msk() releases rnp_cur->lock. */
1084 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1087 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1093 * Force quiescent states on reluctant CPUs, and also detect which
1094 * CPUs are in dyntick-idle mode.
1096 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1098 unsigned long flags;
1100 struct rcu_node *rnp = rcu_get_root(rsp);
1103 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
1104 return; /* No grace period in progress, nothing to force. */
1105 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1106 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1107 return; /* Someone else is already on the job. */
1110 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1111 goto unlock_ret; /* no emergency and done recently. */
1113 spin_lock(&rnp->lock);
1114 lastcomp = rsp->completed;
1115 signaled = rsp->signaled;
1116 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1117 if (lastcomp == rsp->gpnum) {
1118 rsp->n_force_qs_ngp++;
1119 spin_unlock(&rnp->lock);
1120 goto unlock_ret; /* no GP in progress, time updated. */
1122 spin_unlock(&rnp->lock);
1126 break; /* grace period still initializing, ignore. */
1128 case RCU_SAVE_DYNTICK:
1130 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1131 break; /* So gcc recognizes the dead code. */
1133 /* Record dyntick-idle state. */
1134 if (rcu_process_dyntick(rsp, lastcomp,
1135 dyntick_save_progress_counter))
1138 /* Update state, record completion counter. */
1139 spin_lock(&rnp->lock);
1140 if (lastcomp == rsp->completed) {
1141 rsp->signaled = RCU_FORCE_QS;
1142 dyntick_record_completed(rsp, lastcomp);
1144 spin_unlock(&rnp->lock);
1149 /* Check dyntick-idle state, send IPI to laggarts. */
1150 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1151 rcu_implicit_dynticks_qs))
1154 /* Leave state in case more forcing is required. */
1159 spin_unlock_irqrestore(&rsp->fqslock, flags);
1162 #else /* #ifdef CONFIG_SMP */
1164 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1169 #endif /* #else #ifdef CONFIG_SMP */
1172 * This does the RCU processing work from softirq context for the
1173 * specified rcu_state and rcu_data structures. This may be called
1174 * only from the CPU to whom the rdp belongs.
1177 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1179 unsigned long flags;
1181 WARN_ON_ONCE(rdp->beenonline == 0);
1184 * If an RCU GP has gone long enough, go check for dyntick
1185 * idle CPUs and, if needed, send resched IPIs.
1187 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1188 force_quiescent_state(rsp, 1);
1191 * Advance callbacks in response to end of earlier grace
1192 * period that some other CPU ended.
1194 rcu_process_gp_end(rsp, rdp);
1196 /* Update RCU state based on any recent quiescent states. */
1197 rcu_check_quiescent_state(rsp, rdp);
1199 /* Does this CPU require a not-yet-started grace period? */
1200 if (cpu_needs_another_gp(rsp, rdp)) {
1201 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1202 rcu_start_gp(rsp, flags); /* releases above lock */
1205 /* If there are callbacks ready, invoke them. */
1210 * Do softirq processing for the current CPU.
1212 static void rcu_process_callbacks(struct softirq_action *unused)
1215 * Memory references from any prior RCU read-side critical sections
1216 * executed by the interrupted code must be seen before any RCU
1217 * grace-period manipulations below.
1219 smp_mb(); /* See above block comment. */
1221 __rcu_process_callbacks(&rcu_sched_state,
1222 &__get_cpu_var(rcu_sched_data));
1223 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1224 rcu_preempt_process_callbacks();
1227 * Memory references from any later RCU read-side critical sections
1228 * executed by the interrupted code must be seen after any RCU
1229 * grace-period manipulations above.
1231 smp_mb(); /* See above block comment. */
1235 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1236 struct rcu_state *rsp)
1238 unsigned long flags;
1239 struct rcu_data *rdp;
1244 smp_mb(); /* Ensure RCU update seen before callback registry. */
1247 * Opportunistically note grace-period endings and beginnings.
1248 * Note that we might see a beginning right after we see an
1249 * end, but never vice versa, since this CPU has to pass through
1250 * a quiescent state betweentimes.
1252 local_irq_save(flags);
1253 rdp = rsp->rda[smp_processor_id()];
1254 rcu_process_gp_end(rsp, rdp);
1255 check_for_new_grace_period(rsp, rdp);
1257 /* Add the callback to our list. */
1258 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1259 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1261 /* Start a new grace period if one not already started. */
1262 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
1263 unsigned long nestflag;
1264 struct rcu_node *rnp_root = rcu_get_root(rsp);
1266 spin_lock_irqsave(&rnp_root->lock, nestflag);
1267 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1270 /* Force the grace period if too many callbacks or too long waiting. */
1271 if (unlikely(++rdp->qlen > qhimark)) {
1272 rdp->blimit = LONG_MAX;
1273 force_quiescent_state(rsp, 0);
1274 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1275 force_quiescent_state(rsp, 1);
1276 local_irq_restore(flags);
1280 * Queue an RCU-sched callback for invocation after a grace period.
1282 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1284 __call_rcu(head, func, &rcu_sched_state);
1286 EXPORT_SYMBOL_GPL(call_rcu_sched);
1289 * Queue an RCU for invocation after a quicker grace period.
1291 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1293 __call_rcu(head, func, &rcu_bh_state);
1295 EXPORT_SYMBOL_GPL(call_rcu_bh);
1298 * Check to see if there is any immediate RCU-related work to be done
1299 * by the current CPU, for the specified type of RCU, returning 1 if so.
1300 * The checks are in order of increasing expense: checks that can be
1301 * carried out against CPU-local state are performed first. However,
1302 * we must check for CPU stalls first, else we might not get a chance.
1304 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1306 rdp->n_rcu_pending++;
1308 /* Check for CPU stalls, if enabled. */
1309 check_cpu_stall(rsp, rdp);
1311 /* Is the RCU core waiting for a quiescent state from this CPU? */
1312 if (rdp->qs_pending) {
1313 rdp->n_rp_qs_pending++;
1317 /* Does this CPU have callbacks ready to invoke? */
1318 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1319 rdp->n_rp_cb_ready++;
1323 /* Has RCU gone idle with this CPU needing another grace period? */
1324 if (cpu_needs_another_gp(rsp, rdp)) {
1325 rdp->n_rp_cpu_needs_gp++;
1329 /* Has another RCU grace period completed? */
1330 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1331 rdp->n_rp_gp_completed++;
1335 /* Has a new RCU grace period started? */
1336 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1337 rdp->n_rp_gp_started++;
1341 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1342 if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
1343 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1344 rdp->n_rp_need_fqs++;
1349 rdp->n_rp_need_nothing++;
1354 * Check to see if there is any immediate RCU-related work to be done
1355 * by the current CPU, returning 1 if so. This function is part of the
1356 * RCU implementation; it is -not- an exported member of the RCU API.
1358 static int rcu_pending(int cpu)
1360 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1361 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1362 rcu_preempt_pending(cpu);
1366 * Check to see if any future RCU-related work will need to be done
1367 * by the current CPU, even if none need be done immediately, returning
1368 * 1 if so. This function is part of the RCU implementation; it is -not-
1369 * an exported member of the RCU API.
1371 int rcu_needs_cpu(int cpu)
1373 /* RCU callbacks either ready or pending? */
1374 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1375 per_cpu(rcu_bh_data, cpu).nxtlist ||
1376 rcu_preempt_needs_cpu(cpu);
1380 * Do boot-time initialization of a CPU's per-CPU RCU data.
1383 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1385 unsigned long flags;
1387 struct rcu_data *rdp = rsp->rda[cpu];
1388 struct rcu_node *rnp = rcu_get_root(rsp);
1390 /* Set up local state, ensuring consistent view of global state. */
1391 spin_lock_irqsave(&rnp->lock, flags);
1392 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1393 rdp->nxtlist = NULL;
1394 for (i = 0; i < RCU_NEXT_SIZE; i++)
1395 rdp->nxttail[i] = &rdp->nxtlist;
1398 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1399 #endif /* #ifdef CONFIG_NO_HZ */
1401 spin_unlock_irqrestore(&rnp->lock, flags);
1405 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1406 * offline event can be happening at a given time. Note also that we
1407 * can accept some slop in the rsp->completed access due to the fact
1408 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1410 static void __cpuinit
1411 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1413 unsigned long flags;
1416 struct rcu_data *rdp = rsp->rda[cpu];
1417 struct rcu_node *rnp = rcu_get_root(rsp);
1419 /* Set up local state, ensuring consistent view of global state. */
1420 spin_lock_irqsave(&rnp->lock, flags);
1421 lastcomp = rsp->completed;
1422 rdp->completed = lastcomp;
1423 rdp->gpnum = lastcomp;
1424 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1425 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1426 rdp->beenonline = 1; /* We have now been online. */
1427 rdp->preemptable = preemptable;
1428 rdp->passed_quiesc_completed = lastcomp - 1;
1429 rdp->blimit = blimit;
1430 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1433 * A new grace period might start here. If so, we won't be part
1434 * of it, but that is OK, as we are currently in a quiescent state.
1437 /* Exclude any attempts to start a new GP on large systems. */
1438 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1440 /* Add CPU to rcu_node bitmasks. */
1442 mask = rdp->grpmask;
1444 /* Exclude any attempts to start a new GP on small systems. */
1445 spin_lock(&rnp->lock); /* irqs already disabled. */
1446 rnp->qsmaskinit |= mask;
1447 mask = rnp->grpmask;
1448 spin_unlock(&rnp->lock); /* irqs already disabled. */
1450 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1452 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1455 * A new grace period might start here. If so, we will be part of
1456 * it, and its gpnum will be greater than ours, so we will
1457 * participate. It is also possible for the gpnum to have been
1458 * incremented before this function was called, and the bitmasks
1459 * to not be filled out until now, in which case we will also
1460 * participate due to our gpnum being behind.
1463 /* Since it is coming online, the CPU is in a quiescent state. */
1464 cpu_quiet(cpu, rsp, rdp, lastcomp);
1465 local_irq_restore(flags);
1468 static void __cpuinit rcu_online_cpu(int cpu)
1470 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1471 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1472 rcu_preempt_init_percpu_data(cpu);
1476 * Handle CPU online/offline notification events.
1478 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1479 unsigned long action, void *hcpu)
1481 long cpu = (long)hcpu;
1484 case CPU_UP_PREPARE:
1485 case CPU_UP_PREPARE_FROZEN:
1486 rcu_online_cpu(cpu);
1489 case CPU_DEAD_FROZEN:
1490 case CPU_UP_CANCELED:
1491 case CPU_UP_CANCELED_FROZEN:
1492 rcu_offline_cpu(cpu);
1501 * Compute the per-level fanout, either using the exact fanout specified
1502 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1504 #ifdef CONFIG_RCU_FANOUT_EXACT
1505 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1509 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1510 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1512 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1513 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1520 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1521 ccur = rsp->levelcnt[i];
1522 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1526 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1529 * Helper function for rcu_init() that initializes one rcu_state structure.
1531 static void __init rcu_init_one(struct rcu_state *rsp)
1536 struct rcu_node *rnp;
1538 /* Initialize the level-tracking arrays. */
1540 for (i = 1; i < NUM_RCU_LVLS; i++)
1541 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1542 rcu_init_levelspread(rsp);
1544 /* Initialize the elements themselves, starting from the leaves. */
1546 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1547 cpustride *= rsp->levelspread[i];
1548 rnp = rsp->level[i];
1549 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1550 spin_lock_init(&rnp->lock);
1553 rnp->qsmaskinit = 0;
1554 rnp->grplo = j * cpustride;
1555 rnp->grphi = (j + 1) * cpustride - 1;
1556 if (rnp->grphi >= NR_CPUS)
1557 rnp->grphi = NR_CPUS - 1;
1563 rnp->grpnum = j % rsp->levelspread[i - 1];
1564 rnp->grpmask = 1UL << rnp->grpnum;
1565 rnp->parent = rsp->level[i - 1] +
1566 j / rsp->levelspread[i - 1];
1569 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1570 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1576 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1577 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1580 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1582 rcu_init_one(rsp); \
1583 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1585 for_each_possible_cpu(i) { \
1586 if (i > rnp[j].grphi) \
1588 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1589 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1590 rcu_boot_init_percpu_data(i, rsp); \
1594 #ifdef CONFIG_TREE_PREEMPT_RCU
1596 void __init __rcu_init_preempt(void)
1598 int i; /* All used by RCU_INIT_FLAVOR(). */
1600 struct rcu_node *rnp;
1602 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
1605 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1607 void __init __rcu_init_preempt(void)
1611 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1613 void __init __rcu_init(void)
1615 int i; /* All used by RCU_INIT_FLAVOR(). */
1617 struct rcu_node *rnp;
1619 rcu_bootup_announce();
1620 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1621 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1622 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1623 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1624 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1625 __rcu_init_preempt();
1626 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1629 module_param(blimit, int, 0);
1630 module_param(qhimark, int, 0);
1631 module_param(qlowmark, int, 0);