2 * Read-Copy Update mechanism for mutual exclusion
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright IBM Corporation, 2008
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
27 * For detailed explanation of Read-Copy Update mechanism see -
30 #include <linux/types.h>
31 #include <linux/kernel.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/smp.h>
35 #include <linux/rcupdate.h>
36 #include <linux/interrupt.h>
37 #include <linux/sched.h>
38 #include <linux/nmi.h>
39 #include <asm/atomic.h>
40 #include <linux/bitops.h>
41 #include <linux/module.h>
42 #include <linux/completion.h>
43 #include <linux/moduleparam.h>
44 #include <linux/percpu.h>
45 #include <linux/notifier.h>
46 #include <linux/cpu.h>
47 #include <linux/mutex.h>
48 #include <linux/time.h>
52 #ifdef CONFIG_DEBUG_LOCK_ALLOC
53 static struct lock_class_key rcu_lock_key;
54 struct lockdep_map rcu_lock_map =
55 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
56 EXPORT_SYMBOL_GPL(rcu_lock_map);
59 /* Data structures. */
61 #define RCU_STATE_INITIALIZER(name) { \
62 .level = { &name.node[0] }, \
64 NUM_RCU_LVL_0, /* root of hierarchy. */ \
67 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
69 .signaled = RCU_SIGNAL_INIT, \
72 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
73 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
75 .n_force_qs_ngp = 0, \
78 struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
79 DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
81 struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
82 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
84 extern long rcu_batches_completed_sched(void);
85 static struct rcu_node *rcu_get_root(struct rcu_state *rsp);
86 static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp,
87 struct rcu_node *rnp, unsigned long flags);
88 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags);
89 #ifdef CONFIG_HOTPLUG_CPU
90 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp);
91 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
92 static void __rcu_process_callbacks(struct rcu_state *rsp,
93 struct rcu_data *rdp);
94 static void __call_rcu(struct rcu_head *head,
95 void (*func)(struct rcu_head *rcu),
96 struct rcu_state *rsp);
97 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp);
98 static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp,
101 #include "rcutree_plugin.h"
104 * Note a quiescent state. Because we do not need to know
105 * how many quiescent states passed, just if there was at least
106 * one since the start of the grace period, this just sets a flag.
108 void rcu_sched_qs(int cpu)
110 struct rcu_data *rdp;
112 rdp = &per_cpu(rcu_sched_data, cpu);
113 rdp->passed_quiesc_completed = rdp->completed;
115 rdp->passed_quiesc = 1;
116 rcu_preempt_note_context_switch(cpu);
119 void rcu_bh_qs(int cpu)
121 struct rcu_data *rdp;
123 rdp = &per_cpu(rcu_bh_data, cpu);
124 rdp->passed_quiesc_completed = rdp->completed;
126 rdp->passed_quiesc = 1;
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 */
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_ONCE(rdtp->dynticks & 0x1);
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_ONCE(!(rdtp->dynticks & 0x1));
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_ONCE(!(rdtp->dynticks_nmi & 0x1));
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_ONCE(rdtp->dynticks_nmi & 0x1);
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_ONCE(!(rdtp->dynticks & 0x1));
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_ONCE(rdtp->dynticks & 0x1);
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 trigger_all_cpu_backtrace();
507 force_quiescent_state(rsp, 0); /* Kick them all. */
510 static void print_cpu_stall(struct rcu_state *rsp)
513 struct rcu_node *rnp = rcu_get_root(rsp);
515 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
516 smp_processor_id(), jiffies - rsp->gp_start);
517 trigger_all_cpu_backtrace();
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);
525 set_need_resched(); /* kick ourselves to get things going. */
528 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
531 struct rcu_node *rnp;
533 delta = jiffies - rsp->jiffies_stall;
535 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
537 /* We haven't checked in, so go dump stack. */
538 print_cpu_stall(rsp);
540 } else if (rsp->gpnum != rsp->completed &&
541 delta >= RCU_STALL_RAT_DELAY) {
543 /* They had two time units to dump stack, so complain. */
544 print_other_cpu_stall(rsp);
548 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
550 static void record_gp_stall_check_time(struct rcu_state *rsp)
554 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
558 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
561 * Update CPU-local rcu_data state to record the newly noticed grace period.
562 * This is used both when we started the grace period and when we notice
563 * that someone else started the grace period.
565 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
568 rdp->passed_quiesc = 0;
569 rdp->gpnum = rsp->gpnum;
573 * Did someone else start a new RCU grace period start since we last
574 * checked? Update local state appropriately if so. Must be called
575 * on the CPU corresponding to rdp.
578 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
583 local_irq_save(flags);
584 if (rdp->gpnum != rsp->gpnum) {
585 note_new_gpnum(rsp, rdp);
588 local_irq_restore(flags);
593 * Start a new RCU grace period if warranted, re-initializing the hierarchy
594 * in preparation for detecting the next grace period. The caller must hold
595 * the root node's ->lock, which is released before return. Hard irqs must
599 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
600 __releases(rcu_get_root(rsp)->lock)
602 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
603 struct rcu_node *rnp = rcu_get_root(rsp);
604 struct rcu_node *rnp_cur;
605 struct rcu_node *rnp_end;
607 if (!cpu_needs_another_gp(rsp, rdp)) {
608 spin_unlock_irqrestore(&rnp->lock, flags);
612 /* Advance to a new grace period and initialize state. */
614 WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
615 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
616 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
617 record_gp_stall_check_time(rsp);
618 dyntick_record_completed(rsp, rsp->completed - 1);
619 note_new_gpnum(rsp, rdp);
622 * Because we are first, we know that all our callbacks will
623 * be covered by this upcoming grace period, even the ones
624 * that were registered arbitrarily recently.
626 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
627 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
629 /* Special-case the common single-level case. */
630 if (NUM_RCU_NODES == 1) {
631 rnp->qsmask = rnp->qsmaskinit;
632 rcu_preempt_check_blocked_tasks(rnp);
633 rnp->gpnum = rsp->gpnum;
634 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
635 spin_unlock_irqrestore(&rnp->lock, flags);
639 spin_unlock(&rnp->lock); /* leave irqs disabled. */
642 /* Exclude any concurrent CPU-hotplug operations. */
643 spin_lock(&rsp->onofflock); /* irqs already disabled. */
646 * Set the quiescent-state-needed bits in all the rcu_node
647 * structures for all currently online CPUs in breadth-first
648 * order, starting from the root rcu_node structure. This
649 * operation relies on the layout of the hierarchy within the
650 * rsp->node[] array. Note that other CPUs will access only
651 * the leaves of the hierarchy, which still indicate that no
652 * grace period is in progress, at least until the corresponding
653 * leaf node has been initialized. In addition, we have excluded
654 * CPU-hotplug operations.
656 * Note that the grace period cannot complete until we finish
657 * the initialization process, as there will be at least one
658 * qsmask bit set in the root node until that time, namely the
659 * one corresponding to this CPU, due to the fact that we have
662 rnp_end = &rsp->node[NUM_RCU_NODES];
663 for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++) {
664 spin_lock(&rnp_cur->lock); /* irqs already disabled. */
665 rnp_cur->qsmask = rnp_cur->qsmaskinit;
666 rcu_preempt_check_blocked_tasks(rnp);
667 rnp->gpnum = rsp->gpnum;
668 spin_unlock(&rnp_cur->lock); /* irqs already disabled. */
671 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
672 spin_unlock_irqrestore(&rsp->onofflock, flags);
676 * Advance this CPU's callbacks, but only if the current grace period
677 * has ended. This may be called only from the CPU to whom the rdp
681 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
686 local_irq_save(flags);
687 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
689 /* Did another grace period end? */
690 if (rdp->completed != completed_snap) {
692 /* Advance callbacks. No harm if list empty. */
693 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
694 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
695 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
697 /* Remember that we saw this grace-period completion. */
698 rdp->completed = completed_snap;
700 local_irq_restore(flags);
704 * Clean up after the prior grace period and let rcu_start_gp() start up
705 * the next grace period if one is needed. Note that the caller must
706 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
708 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
709 __releases(rnp->lock)
711 rsp->completed = rsp->gpnum;
712 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
713 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
717 * Similar to cpu_quiet(), for which it is a helper function. Allows
718 * a group of CPUs to be quieted at one go, though all the CPUs in the
719 * group must be represented by the same leaf rcu_node structure.
720 * That structure's lock must be held upon entry, and it is released
724 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
726 __releases(rnp->lock)
728 /* Walk up the rcu_node hierarchy. */
730 if (!(rnp->qsmask & mask)) {
732 /* Our bit has already been cleared, so done. */
733 spin_unlock_irqrestore(&rnp->lock, flags);
736 rnp->qsmask &= ~mask;
737 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
739 /* Other bits still set at this level, so done. */
740 spin_unlock_irqrestore(&rnp->lock, flags);
744 if (rnp->parent == NULL) {
746 /* No more levels. Exit loop holding root lock. */
750 spin_unlock_irqrestore(&rnp->lock, flags);
752 spin_lock_irqsave(&rnp->lock, flags);
756 * Get here if we are the last CPU to pass through a quiescent
757 * state for this grace period. Invoke cpu_quiet_msk_finish()
758 * to clean up and start the next grace period if one is needed.
760 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
764 * Record a quiescent state for the specified CPU, which must either be
765 * the current CPU or an offline CPU. The lastcomp argument is used to
766 * make sure we are still in the grace period of interest. We don't want
767 * to end the current grace period based on quiescent states detected in
768 * an earlier grace period!
771 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
775 struct rcu_node *rnp;
778 spin_lock_irqsave(&rnp->lock, flags);
779 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
782 * Someone beat us to it for this grace period, so leave.
783 * The race with GP start is resolved by the fact that we
784 * hold the leaf rcu_node lock, so that the per-CPU bits
785 * cannot yet be initialized -- so we would simply find our
786 * CPU's bit already cleared in cpu_quiet_msk() if this race
789 rdp->passed_quiesc = 0; /* try again later! */
790 spin_unlock_irqrestore(&rnp->lock, flags);
794 if ((rnp->qsmask & mask) == 0) {
795 spin_unlock_irqrestore(&rnp->lock, flags);
800 * This GP can't end until cpu checks in, so all of our
801 * callbacks can be processed during the next GP.
803 rdp = rsp->rda[smp_processor_id()];
804 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
806 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
811 * Check to see if there is a new grace period of which this CPU
812 * is not yet aware, and if so, set up local rcu_data state for it.
813 * Otherwise, see if this CPU has just passed through its first
814 * quiescent state for this grace period, and record that fact if so.
817 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
819 /* If there is now a new grace period, record and return. */
820 if (check_for_new_grace_period(rsp, rdp))
824 * Does this CPU still need to do its part for current grace period?
825 * If no, return and let the other CPUs do their part as well.
827 if (!rdp->qs_pending)
831 * Was there a quiescent state since the beginning of the grace
832 * period? If no, then exit and wait for the next call.
834 if (!rdp->passed_quiesc)
837 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
838 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
841 #ifdef CONFIG_HOTPLUG_CPU
844 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
845 * and move all callbacks from the outgoing CPU to the current one.
847 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
853 struct rcu_data *rdp = rsp->rda[cpu];
854 struct rcu_data *rdp_me;
855 struct rcu_node *rnp;
857 /* Exclude any attempts to start a new grace period. */
858 spin_lock_irqsave(&rsp->onofflock, flags);
860 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
862 mask = rdp->grpmask; /* rnp->grplo is constant. */
864 spin_lock(&rnp->lock); /* irqs already disabled. */
865 rnp->qsmaskinit &= ~mask;
866 if (rnp->qsmaskinit != 0) {
867 spin_unlock(&rnp->lock); /* irqs remain disabled. */
870 rcu_preempt_offline_tasks(rsp, rnp);
872 spin_unlock(&rnp->lock); /* irqs remain disabled. */
874 } while (rnp != NULL);
875 lastcomp = rsp->completed;
877 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
879 /* Being offline is a quiescent state, so go record it. */
880 cpu_quiet(cpu, rsp, rdp, lastcomp);
883 * Move callbacks from the outgoing CPU to the running CPU.
884 * Note that the outgoing CPU is now quiscent, so it is now
885 * (uncharacteristically) safe to access its rcu_data structure.
886 * Note also that we must carefully retain the order of the
887 * outgoing CPU's callbacks in order for rcu_barrier() to work
888 * correctly. Finally, note that we start all the callbacks
889 * afresh, even those that have passed through a grace period
890 * and are therefore ready to invoke. The theory is that hotplug
891 * events are rare, and that if they are frequent enough to
892 * indefinitely delay callbacks, you have far worse things to
895 rdp_me = rsp->rda[smp_processor_id()];
896 if (rdp->nxtlist != NULL) {
897 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
898 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
900 for (i = 0; i < RCU_NEXT_SIZE; i++)
901 rdp->nxttail[i] = &rdp->nxtlist;
902 rdp_me->qlen += rdp->qlen;
905 local_irq_restore(flags);
909 * Remove the specified CPU from the RCU hierarchy and move any pending
910 * callbacks that it might have to the current CPU. This code assumes
911 * that at least one CPU in the system will remain running at all times.
912 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
914 static void rcu_offline_cpu(int cpu)
916 __rcu_offline_cpu(cpu, &rcu_sched_state);
917 __rcu_offline_cpu(cpu, &rcu_bh_state);
918 rcu_preempt_offline_cpu(cpu);
921 #else /* #ifdef CONFIG_HOTPLUG_CPU */
923 static void rcu_offline_cpu(int cpu)
927 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
930 * Invoke any RCU callbacks that have made it to the end of their grace
931 * period. Thottle as specified by rdp->blimit.
933 static void rcu_do_batch(struct rcu_data *rdp)
936 struct rcu_head *next, *list, **tail;
939 /* If no callbacks are ready, just return.*/
940 if (!cpu_has_callbacks_ready_to_invoke(rdp))
944 * Extract the list of ready callbacks, disabling to prevent
945 * races with call_rcu() from interrupt handlers.
947 local_irq_save(flags);
949 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
950 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
951 tail = rdp->nxttail[RCU_DONE_TAIL];
952 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
953 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
954 rdp->nxttail[count] = &rdp->nxtlist;
955 local_irq_restore(flags);
957 /* Invoke callbacks. */
964 if (++count >= rdp->blimit)
968 local_irq_save(flags);
970 /* Update count, and requeue any remaining callbacks. */
973 *tail = rdp->nxtlist;
975 for (count = 0; count < RCU_NEXT_SIZE; count++)
976 if (&rdp->nxtlist == rdp->nxttail[count])
977 rdp->nxttail[count] = tail;
982 /* Reinstate batch limit if we have worked down the excess. */
983 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
984 rdp->blimit = blimit;
986 local_irq_restore(flags);
988 /* Re-raise the RCU softirq if there are callbacks remaining. */
989 if (cpu_has_callbacks_ready_to_invoke(rdp))
990 raise_softirq(RCU_SOFTIRQ);
994 * Check to see if this CPU is in a non-context-switch quiescent state
995 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
996 * Also schedule the RCU softirq handler.
998 * This function must be called with hardirqs disabled. It is normally
999 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1000 * false, there is no point in invoking rcu_check_callbacks().
1002 void rcu_check_callbacks(int cpu, int user)
1004 if (!rcu_pending(cpu))
1005 return; /* if nothing for RCU to do. */
1007 (idle_cpu(cpu) && rcu_scheduler_active &&
1008 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1011 * Get here if this CPU took its interrupt from user
1012 * mode or from the idle loop, and if this is not a
1013 * nested interrupt. In this case, the CPU is in
1014 * a quiescent state, so note it.
1016 * No memory barrier is required here because both
1017 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1018 * variables that other CPUs neither access nor modify,
1019 * at least not while the corresponding CPU is online.
1025 } else if (!in_softirq()) {
1028 * Get here if this CPU did not take its interrupt from
1029 * softirq, in other words, if it is not interrupting
1030 * a rcu_bh read-side critical section. This is an _bh
1031 * critical section, so note it.
1036 rcu_preempt_check_callbacks(cpu);
1037 raise_softirq(RCU_SOFTIRQ);
1043 * Scan the leaf rcu_node structures, processing dyntick state for any that
1044 * have not yet encountered a quiescent state, using the function specified.
1045 * Returns 1 if the current grace period ends while scanning (possibly
1046 * because we made it end).
1048 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1049 int (*f)(struct rcu_data *))
1053 unsigned long flags;
1055 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
1056 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1058 for (; rnp_cur < rnp_end; rnp_cur++) {
1060 spin_lock_irqsave(&rnp_cur->lock, flags);
1061 if (rsp->completed != lastcomp) {
1062 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1065 if (rnp_cur->qsmask == 0) {
1066 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1069 cpu = rnp_cur->grplo;
1071 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1072 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1075 if (mask != 0 && rsp->completed == lastcomp) {
1077 /* cpu_quiet_msk() releases rnp_cur->lock. */
1078 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1081 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1087 * Force quiescent states on reluctant CPUs, and also detect which
1088 * CPUs are in dyntick-idle mode.
1090 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1092 unsigned long flags;
1094 struct rcu_node *rnp = rcu_get_root(rsp);
1097 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
1098 return; /* No grace period in progress, nothing to force. */
1099 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1100 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1101 return; /* Someone else is already on the job. */
1104 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1105 goto unlock_ret; /* no emergency and done recently. */
1107 spin_lock(&rnp->lock);
1108 lastcomp = rsp->completed;
1109 signaled = rsp->signaled;
1110 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1111 if (lastcomp == rsp->gpnum) {
1112 rsp->n_force_qs_ngp++;
1113 spin_unlock(&rnp->lock);
1114 goto unlock_ret; /* no GP in progress, time updated. */
1116 spin_unlock(&rnp->lock);
1120 break; /* grace period still initializing, ignore. */
1122 case RCU_SAVE_DYNTICK:
1124 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1125 break; /* So gcc recognizes the dead code. */
1127 /* Record dyntick-idle state. */
1128 if (rcu_process_dyntick(rsp, lastcomp,
1129 dyntick_save_progress_counter))
1132 /* Update state, record completion counter. */
1133 spin_lock(&rnp->lock);
1134 if (lastcomp == rsp->completed) {
1135 rsp->signaled = RCU_FORCE_QS;
1136 dyntick_record_completed(rsp, lastcomp);
1138 spin_unlock(&rnp->lock);
1143 /* Check dyntick-idle state, send IPI to laggarts. */
1144 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1145 rcu_implicit_dynticks_qs))
1148 /* Leave state in case more forcing is required. */
1153 spin_unlock_irqrestore(&rsp->fqslock, flags);
1156 #else /* #ifdef CONFIG_SMP */
1158 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1163 #endif /* #else #ifdef CONFIG_SMP */
1166 * This does the RCU processing work from softirq context for the
1167 * specified rcu_state and rcu_data structures. This may be called
1168 * only from the CPU to whom the rdp belongs.
1171 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1173 unsigned long flags;
1175 WARN_ON_ONCE(rdp->beenonline == 0);
1178 * If an RCU GP has gone long enough, go check for dyntick
1179 * idle CPUs and, if needed, send resched IPIs.
1181 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1182 force_quiescent_state(rsp, 1);
1185 * Advance callbacks in response to end of earlier grace
1186 * period that some other CPU ended.
1188 rcu_process_gp_end(rsp, rdp);
1190 /* Update RCU state based on any recent quiescent states. */
1191 rcu_check_quiescent_state(rsp, rdp);
1193 /* Does this CPU require a not-yet-started grace period? */
1194 if (cpu_needs_another_gp(rsp, rdp)) {
1195 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1196 rcu_start_gp(rsp, flags); /* releases above lock */
1199 /* If there are callbacks ready, invoke them. */
1204 * Do softirq processing for the current CPU.
1206 static void rcu_process_callbacks(struct softirq_action *unused)
1209 * Memory references from any prior RCU read-side critical sections
1210 * executed by the interrupted code must be seen before any RCU
1211 * grace-period manipulations below.
1213 smp_mb(); /* See above block comment. */
1215 __rcu_process_callbacks(&rcu_sched_state,
1216 &__get_cpu_var(rcu_sched_data));
1217 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1218 rcu_preempt_process_callbacks();
1221 * Memory references from any later RCU read-side critical sections
1222 * executed by the interrupted code must be seen after any RCU
1223 * grace-period manipulations above.
1225 smp_mb(); /* See above block comment. */
1229 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1230 struct rcu_state *rsp)
1232 unsigned long flags;
1233 struct rcu_data *rdp;
1238 smp_mb(); /* Ensure RCU update seen before callback registry. */
1241 * Opportunistically note grace-period endings and beginnings.
1242 * Note that we might see a beginning right after we see an
1243 * end, but never vice versa, since this CPU has to pass through
1244 * a quiescent state betweentimes.
1246 local_irq_save(flags);
1247 rdp = rsp->rda[smp_processor_id()];
1248 rcu_process_gp_end(rsp, rdp);
1249 check_for_new_grace_period(rsp, rdp);
1251 /* Add the callback to our list. */
1252 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1253 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1255 /* Start a new grace period if one not already started. */
1256 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
1257 unsigned long nestflag;
1258 struct rcu_node *rnp_root = rcu_get_root(rsp);
1260 spin_lock_irqsave(&rnp_root->lock, nestflag);
1261 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1264 /* Force the grace period if too many callbacks or too long waiting. */
1265 if (unlikely(++rdp->qlen > qhimark)) {
1266 rdp->blimit = LONG_MAX;
1267 force_quiescent_state(rsp, 0);
1268 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1269 force_quiescent_state(rsp, 1);
1270 local_irq_restore(flags);
1274 * Queue an RCU-sched callback for invocation after a grace period.
1276 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1278 __call_rcu(head, func, &rcu_sched_state);
1280 EXPORT_SYMBOL_GPL(call_rcu_sched);
1283 * Queue an RCU for invocation after a quicker grace period.
1285 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1287 __call_rcu(head, func, &rcu_bh_state);
1289 EXPORT_SYMBOL_GPL(call_rcu_bh);
1292 * Check to see if there is any immediate RCU-related work to be done
1293 * by the current CPU, for the specified type of RCU, returning 1 if so.
1294 * The checks are in order of increasing expense: checks that can be
1295 * carried out against CPU-local state are performed first. However,
1296 * we must check for CPU stalls first, else we might not get a chance.
1298 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1300 rdp->n_rcu_pending++;
1302 /* Check for CPU stalls, if enabled. */
1303 check_cpu_stall(rsp, rdp);
1305 /* Is the RCU core waiting for a quiescent state from this CPU? */
1306 if (rdp->qs_pending) {
1307 rdp->n_rp_qs_pending++;
1311 /* Does this CPU have callbacks ready to invoke? */
1312 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1313 rdp->n_rp_cb_ready++;
1317 /* Has RCU gone idle with this CPU needing another grace period? */
1318 if (cpu_needs_another_gp(rsp, rdp)) {
1319 rdp->n_rp_cpu_needs_gp++;
1323 /* Has another RCU grace period completed? */
1324 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1325 rdp->n_rp_gp_completed++;
1329 /* Has a new RCU grace period started? */
1330 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1331 rdp->n_rp_gp_started++;
1335 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1336 if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
1337 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1338 rdp->n_rp_need_fqs++;
1343 rdp->n_rp_need_nothing++;
1348 * Check to see if there is any immediate RCU-related work to be done
1349 * by the current CPU, returning 1 if so. This function is part of the
1350 * RCU implementation; it is -not- an exported member of the RCU API.
1352 static int rcu_pending(int cpu)
1354 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1355 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1356 rcu_preempt_pending(cpu);
1360 * Check to see if any future RCU-related work will need to be done
1361 * by the current CPU, even if none need be done immediately, returning
1362 * 1 if so. This function is part of the RCU implementation; it is -not-
1363 * an exported member of the RCU API.
1365 int rcu_needs_cpu(int cpu)
1367 /* RCU callbacks either ready or pending? */
1368 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1369 per_cpu(rcu_bh_data, cpu).nxtlist ||
1370 rcu_preempt_needs_cpu(cpu);
1374 * Do boot-time initialization of a CPU's per-CPU RCU data.
1377 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1379 unsigned long flags;
1381 struct rcu_data *rdp = rsp->rda[cpu];
1382 struct rcu_node *rnp = rcu_get_root(rsp);
1384 /* Set up local state, ensuring consistent view of global state. */
1385 spin_lock_irqsave(&rnp->lock, flags);
1386 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1387 rdp->nxtlist = NULL;
1388 for (i = 0; i < RCU_NEXT_SIZE; i++)
1389 rdp->nxttail[i] = &rdp->nxtlist;
1392 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1393 #endif /* #ifdef CONFIG_NO_HZ */
1395 spin_unlock_irqrestore(&rnp->lock, flags);
1399 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1400 * offline event can be happening at a given time. Note also that we
1401 * can accept some slop in the rsp->completed access due to the fact
1402 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1404 static void __cpuinit
1405 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1407 unsigned long flags;
1410 struct rcu_data *rdp = rsp->rda[cpu];
1411 struct rcu_node *rnp = rcu_get_root(rsp);
1413 /* Set up local state, ensuring consistent view of global state. */
1414 spin_lock_irqsave(&rnp->lock, flags);
1415 lastcomp = rsp->completed;
1416 rdp->completed = lastcomp;
1417 rdp->gpnum = lastcomp;
1418 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1419 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1420 rdp->beenonline = 1; /* We have now been online. */
1421 rdp->preemptable = preemptable;
1422 rdp->passed_quiesc_completed = lastcomp - 1;
1423 rdp->blimit = blimit;
1424 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1427 * A new grace period might start here. If so, we won't be part
1428 * of it, but that is OK, as we are currently in a quiescent state.
1431 /* Exclude any attempts to start a new GP on large systems. */
1432 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1434 /* Add CPU to rcu_node bitmasks. */
1436 mask = rdp->grpmask;
1438 /* Exclude any attempts to start a new GP on small systems. */
1439 spin_lock(&rnp->lock); /* irqs already disabled. */
1440 rnp->qsmaskinit |= mask;
1441 mask = rnp->grpmask;
1442 spin_unlock(&rnp->lock); /* irqs already disabled. */
1444 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1446 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1449 * A new grace period might start here. If so, we will be part of
1450 * it, and its gpnum will be greater than ours, so we will
1451 * participate. It is also possible for the gpnum to have been
1452 * incremented before this function was called, and the bitmasks
1453 * to not be filled out until now, in which case we will also
1454 * participate due to our gpnum being behind.
1457 /* Since it is coming online, the CPU is in a quiescent state. */
1458 cpu_quiet(cpu, rsp, rdp, lastcomp);
1459 local_irq_restore(flags);
1462 static void __cpuinit rcu_online_cpu(int cpu)
1464 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1465 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1466 rcu_preempt_init_percpu_data(cpu);
1470 * Handle CPU online/offline notification events.
1472 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1473 unsigned long action, void *hcpu)
1475 long cpu = (long)hcpu;
1478 case CPU_UP_PREPARE:
1479 case CPU_UP_PREPARE_FROZEN:
1480 rcu_online_cpu(cpu);
1483 case CPU_DEAD_FROZEN:
1484 case CPU_UP_CANCELED:
1485 case CPU_UP_CANCELED_FROZEN:
1486 rcu_offline_cpu(cpu);
1495 * Compute the per-level fanout, either using the exact fanout specified
1496 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1498 #ifdef CONFIG_RCU_FANOUT_EXACT
1499 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1503 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1504 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1506 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1507 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1514 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1515 ccur = rsp->levelcnt[i];
1516 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1520 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1523 * Helper function for rcu_init() that initializes one rcu_state structure.
1525 static void __init rcu_init_one(struct rcu_state *rsp)
1530 struct rcu_node *rnp;
1532 /* Initialize the level-tracking arrays. */
1534 for (i = 1; i < NUM_RCU_LVLS; i++)
1535 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1536 rcu_init_levelspread(rsp);
1538 /* Initialize the elements themselves, starting from the leaves. */
1540 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1541 cpustride *= rsp->levelspread[i];
1542 rnp = rsp->level[i];
1543 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1544 spin_lock_init(&rnp->lock);
1547 rnp->qsmaskinit = 0;
1548 rnp->grplo = j * cpustride;
1549 rnp->grphi = (j + 1) * cpustride - 1;
1550 if (rnp->grphi >= NR_CPUS)
1551 rnp->grphi = NR_CPUS - 1;
1557 rnp->grpnum = j % rsp->levelspread[i - 1];
1558 rnp->grpmask = 1UL << rnp->grpnum;
1559 rnp->parent = rsp->level[i - 1] +
1560 j / rsp->levelspread[i - 1];
1563 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1564 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1570 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1571 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1574 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1576 rcu_init_one(rsp); \
1577 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1579 for_each_possible_cpu(i) { \
1580 if (i > rnp[j].grphi) \
1582 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1583 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1584 rcu_boot_init_percpu_data(i, rsp); \
1588 #ifdef CONFIG_TREE_PREEMPT_RCU
1590 void __init __rcu_init_preempt(void)
1592 int i; /* All used by RCU_INIT_FLAVOR(). */
1594 struct rcu_node *rnp;
1596 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
1599 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1601 void __init __rcu_init_preempt(void)
1605 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1607 void __init __rcu_init(void)
1609 int i; /* All used by RCU_INIT_FLAVOR(). */
1611 struct rcu_node *rnp;
1613 rcu_bootup_announce();
1614 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1615 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1616 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1617 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1618 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1619 __rcu_init_preempt();
1620 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1623 module_param(blimit, int, 0);
1624 module_param(qhimark, int, 0);
1625 module_param(qlowmark, int, 0);