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 /* Forward declarations for rcutree_plugin.h */
85 static inline void rcu_bootup_announce(void);
86 long rcu_batches_completed(void);
87 static void rcu_preempt_note_context_switch(int cpu);
88 static int rcu_preempted_readers(struct rcu_node *rnp);
89 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
90 static void rcu_print_task_stall(struct rcu_node *rnp);
91 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
92 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
93 #ifdef CONFIG_HOTPLUG_CPU
94 static void rcu_preempt_offline_tasks(struct rcu_state *rsp,
96 struct rcu_data *rdp);
97 static void rcu_preempt_offline_cpu(int cpu);
98 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
99 static void rcu_preempt_check_callbacks(int cpu);
100 static void rcu_preempt_process_callbacks(void);
101 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
102 static int rcu_preempt_pending(int cpu);
103 static int rcu_preempt_needs_cpu(int cpu);
104 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
105 static void __init __rcu_init_preempt(void);
109 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
110 * permit this function to be invoked without holding the root rcu_node
111 * structure's ->lock, but of course results can be subject to change.
113 static int rcu_gp_in_progress(struct rcu_state *rsp)
115 return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
119 * Note a quiescent state. Because we do not need to know
120 * how many quiescent states passed, just if there was at least
121 * one since the start of the grace period, this just sets a flag.
123 void rcu_sched_qs(int cpu)
125 struct rcu_data *rdp;
127 rdp = &per_cpu(rcu_sched_data, cpu);
128 rdp->passed_quiesc_completed = rdp->completed;
130 rdp->passed_quiesc = 1;
131 rcu_preempt_note_context_switch(cpu);
134 void rcu_bh_qs(int cpu)
136 struct rcu_data *rdp;
138 rdp = &per_cpu(rcu_bh_data, cpu);
139 rdp->passed_quiesc_completed = rdp->completed;
141 rdp->passed_quiesc = 1;
145 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
146 .dynticks_nesting = 1,
149 #endif /* #ifdef CONFIG_NO_HZ */
151 static int blimit = 10; /* Maximum callbacks per softirq. */
152 static int qhimark = 10000; /* If this many pending, ignore blimit. */
153 static int qlowmark = 100; /* Once only this many pending, use blimit. */
155 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
156 static int rcu_pending(int cpu);
159 * Return the number of RCU-sched batches processed thus far for debug & stats.
161 long rcu_batches_completed_sched(void)
163 return rcu_sched_state.completed;
165 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
168 * Return the number of RCU BH batches processed thus far for debug & stats.
170 long rcu_batches_completed_bh(void)
172 return rcu_bh_state.completed;
174 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
177 * Does the CPU have callbacks ready to be invoked?
180 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
182 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
186 * Does the current CPU require a yet-as-unscheduled grace period?
189 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
191 return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
195 * Return the root node of the specified rcu_state structure.
197 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
199 return &rsp->node[0];
205 * If the specified CPU is offline, tell the caller that it is in
206 * a quiescent state. Otherwise, whack it with a reschedule IPI.
207 * Grace periods can end up waiting on an offline CPU when that
208 * CPU is in the process of coming online -- it will be added to the
209 * rcu_node bitmasks before it actually makes it online. The same thing
210 * can happen while a CPU is in the process of coming online. Because this
211 * race is quite rare, we check for it after detecting that the grace
212 * period has been delayed rather than checking each and every CPU
213 * each and every time we start a new grace period.
215 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
218 * If the CPU is offline, it is in a quiescent state. We can
219 * trust its state not to change because interrupts are disabled.
221 if (cpu_is_offline(rdp->cpu)) {
226 /* If preemptable RCU, no point in sending reschedule IPI. */
227 if (rdp->preemptable)
230 /* The CPU is online, so send it a reschedule IPI. */
231 if (rdp->cpu != smp_processor_id())
232 smp_send_reschedule(rdp->cpu);
239 #endif /* #ifdef CONFIG_SMP */
244 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
246 * Enter nohz mode, in other words, -leave- the mode in which RCU
247 * read-side critical sections can occur. (Though RCU read-side
248 * critical sections can occur in irq handlers in nohz mode, a possibility
249 * handled by rcu_irq_enter() and rcu_irq_exit()).
251 void rcu_enter_nohz(void)
254 struct rcu_dynticks *rdtp;
256 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
257 local_irq_save(flags);
258 rdtp = &__get_cpu_var(rcu_dynticks);
260 rdtp->dynticks_nesting--;
261 WARN_ON_ONCE(rdtp->dynticks & 0x1);
262 local_irq_restore(flags);
266 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
268 * Exit nohz mode, in other words, -enter- the mode in which RCU
269 * read-side critical sections normally occur.
271 void rcu_exit_nohz(void)
274 struct rcu_dynticks *rdtp;
276 local_irq_save(flags);
277 rdtp = &__get_cpu_var(rcu_dynticks);
279 rdtp->dynticks_nesting++;
280 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
281 local_irq_restore(flags);
282 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
286 * rcu_nmi_enter - inform RCU of entry to NMI context
288 * If the CPU was idle with dynamic ticks active, and there is no
289 * irq handler running, this updates rdtp->dynticks_nmi to let the
290 * RCU grace-period handling know that the CPU is active.
292 void rcu_nmi_enter(void)
294 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
296 if (rdtp->dynticks & 0x1)
298 rdtp->dynticks_nmi++;
299 WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1));
300 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
304 * rcu_nmi_exit - inform RCU of exit from NMI context
306 * If the CPU was idle with dynamic ticks active, and there is no
307 * irq handler running, this updates rdtp->dynticks_nmi to let the
308 * RCU grace-period handling know that the CPU is no longer active.
310 void rcu_nmi_exit(void)
312 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
314 if (rdtp->dynticks & 0x1)
316 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
317 rdtp->dynticks_nmi++;
318 WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1);
322 * rcu_irq_enter - inform RCU of entry to hard irq context
324 * If the CPU was idle with dynamic ticks active, this updates the
325 * rdtp->dynticks to let the RCU handling know that the CPU is active.
327 void rcu_irq_enter(void)
329 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
331 if (rdtp->dynticks_nesting++)
334 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
335 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
339 * rcu_irq_exit - inform RCU of exit from hard irq context
341 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
342 * to put let the RCU handling be aware that the CPU is going back to idle
345 void rcu_irq_exit(void)
347 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
349 if (--rdtp->dynticks_nesting)
351 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
353 WARN_ON_ONCE(rdtp->dynticks & 0x1);
355 /* If the interrupt queued a callback, get out of dyntick mode. */
356 if (__get_cpu_var(rcu_sched_data).nxtlist ||
357 __get_cpu_var(rcu_bh_data).nxtlist)
362 * Record the specified "completed" value, which is later used to validate
363 * dynticks counter manipulations. Specify "rsp->completed - 1" to
364 * unconditionally invalidate any future dynticks manipulations (which is
365 * useful at the beginning of a grace period).
367 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
369 rsp->dynticks_completed = comp;
375 * Recall the previously recorded value of the completion for dynticks.
377 static long dyntick_recall_completed(struct rcu_state *rsp)
379 return rsp->dynticks_completed;
383 * Snapshot the specified CPU's dynticks counter so that we can later
384 * credit them with an implicit quiescent state. Return 1 if this CPU
385 * is in dynticks idle mode, which is an extended quiescent state.
387 static int dyntick_save_progress_counter(struct rcu_data *rdp)
393 snap = rdp->dynticks->dynticks;
394 snap_nmi = rdp->dynticks->dynticks_nmi;
395 smp_mb(); /* Order sampling of snap with end of grace period. */
396 rdp->dynticks_snap = snap;
397 rdp->dynticks_nmi_snap = snap_nmi;
398 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
405 * Return true if the specified CPU has passed through a quiescent
406 * state by virtue of being in or having passed through an dynticks
407 * idle state since the last call to dyntick_save_progress_counter()
410 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
417 curr = rdp->dynticks->dynticks;
418 snap = rdp->dynticks_snap;
419 curr_nmi = rdp->dynticks->dynticks_nmi;
420 snap_nmi = rdp->dynticks_nmi_snap;
421 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
424 * If the CPU passed through or entered a dynticks idle phase with
425 * no active irq/NMI handlers, then we can safely pretend that the CPU
426 * already acknowledged the request to pass through a quiescent
427 * state. Either way, that CPU cannot possibly be in an RCU
428 * read-side critical section that started before the beginning
429 * of the current RCU grace period.
431 if ((curr != snap || (curr & 0x1) == 0) &&
432 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
437 /* Go check for the CPU being offline. */
438 return rcu_implicit_offline_qs(rdp);
441 #endif /* #ifdef CONFIG_SMP */
443 #else /* #ifdef CONFIG_NO_HZ */
445 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
452 * If there are no dynticks, then the only way that a CPU can passively
453 * be in a quiescent state is to be offline. Unlike dynticks idle, which
454 * is a point in time during the prior (already finished) grace period,
455 * an offline CPU is always in a quiescent state, and thus can be
456 * unconditionally applied. So just return the current value of completed.
458 static long dyntick_recall_completed(struct rcu_state *rsp)
460 return rsp->completed;
463 static int dyntick_save_progress_counter(struct rcu_data *rdp)
468 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
470 return rcu_implicit_offline_qs(rdp);
473 #endif /* #ifdef CONFIG_SMP */
475 #endif /* #else #ifdef CONFIG_NO_HZ */
477 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
479 static void record_gp_stall_check_time(struct rcu_state *rsp)
481 rsp->gp_start = jiffies;
482 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
485 static void print_other_cpu_stall(struct rcu_state *rsp)
490 struct rcu_node *rnp = rcu_get_root(rsp);
491 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
492 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
494 /* Only let one CPU complain about others per time interval. */
496 spin_lock_irqsave(&rnp->lock, flags);
497 delta = jiffies - rsp->jiffies_stall;
498 if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
499 spin_unlock_irqrestore(&rnp->lock, flags);
502 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
503 spin_unlock_irqrestore(&rnp->lock, flags);
505 /* OK, time to rat on our buddy... */
507 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
508 for (; rnp_cur < rnp_end; rnp_cur++) {
509 rcu_print_task_stall(rnp);
510 if (rnp_cur->qsmask == 0)
512 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
513 if (rnp_cur->qsmask & (1UL << cpu))
514 printk(" %d", rnp_cur->grplo + cpu);
516 printk(" (detected by %d, t=%ld jiffies)\n",
517 smp_processor_id(), (long)(jiffies - rsp->gp_start));
518 trigger_all_cpu_backtrace();
520 force_quiescent_state(rsp, 0); /* Kick them all. */
523 static void print_cpu_stall(struct rcu_state *rsp)
526 struct rcu_node *rnp = rcu_get_root(rsp);
528 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
529 smp_processor_id(), jiffies - rsp->gp_start);
530 trigger_all_cpu_backtrace();
532 spin_lock_irqsave(&rnp->lock, flags);
533 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
535 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
536 spin_unlock_irqrestore(&rnp->lock, flags);
538 set_need_resched(); /* kick ourselves to get things going. */
541 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
544 struct rcu_node *rnp;
546 delta = jiffies - rsp->jiffies_stall;
548 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
550 /* We haven't checked in, so go dump stack. */
551 print_cpu_stall(rsp);
553 } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
555 /* They had two time units to dump stack, so complain. */
556 print_other_cpu_stall(rsp);
560 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
562 static void record_gp_stall_check_time(struct rcu_state *rsp)
566 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
570 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
573 * Update CPU-local rcu_data state to record the newly noticed grace period.
574 * This is used both when we started the grace period and when we notice
575 * that someone else started the grace period.
577 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
580 rdp->passed_quiesc = 0;
581 rdp->gpnum = rsp->gpnum;
585 * Did someone else start a new RCU grace period start since we last
586 * checked? Update local state appropriately if so. Must be called
587 * on the CPU corresponding to rdp.
590 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
595 local_irq_save(flags);
596 if (rdp->gpnum != rsp->gpnum) {
597 note_new_gpnum(rsp, rdp);
600 local_irq_restore(flags);
605 * Start a new RCU grace period if warranted, re-initializing the hierarchy
606 * in preparation for detecting the next grace period. The caller must hold
607 * the root node's ->lock, which is released before return. Hard irqs must
611 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
612 __releases(rcu_get_root(rsp)->lock)
614 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
615 struct rcu_node *rnp = rcu_get_root(rsp);
617 if (!cpu_needs_another_gp(rsp, rdp)) {
618 spin_unlock_irqrestore(&rnp->lock, flags);
622 /* Advance to a new grace period and initialize state. */
624 WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
625 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
626 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
627 record_gp_stall_check_time(rsp);
628 dyntick_record_completed(rsp, rsp->completed - 1);
629 note_new_gpnum(rsp, rdp);
632 * Because this CPU just now started the new grace period, we know
633 * that all of its callbacks will be covered by this upcoming grace
634 * period, even the ones that were registered arbitrarily recently.
635 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
637 * Other CPUs cannot be sure exactly when the grace period started.
638 * Therefore, their recently registered callbacks must pass through
639 * an additional RCU_NEXT_READY stage, so that they will be handled
640 * by the next RCU grace period.
642 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
643 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
645 /* Special-case the common single-level case. */
646 if (NUM_RCU_NODES == 1) {
647 rcu_preempt_check_blocked_tasks(rnp);
648 rnp->qsmask = rnp->qsmaskinit;
649 rnp->gpnum = rsp->gpnum;
650 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
651 spin_unlock_irqrestore(&rnp->lock, flags);
655 spin_unlock(&rnp->lock); /* leave irqs disabled. */
658 /* Exclude any concurrent CPU-hotplug operations. */
659 spin_lock(&rsp->onofflock); /* irqs already disabled. */
662 * Set the quiescent-state-needed bits in all the rcu_node
663 * structures for all currently online CPUs in breadth-first
664 * order, starting from the root rcu_node structure. This
665 * operation relies on the layout of the hierarchy within the
666 * rsp->node[] array. Note that other CPUs will access only
667 * the leaves of the hierarchy, which still indicate that no
668 * grace period is in progress, at least until the corresponding
669 * leaf node has been initialized. In addition, we have excluded
670 * CPU-hotplug operations.
672 * Note that the grace period cannot complete until we finish
673 * the initialization process, as there will be at least one
674 * qsmask bit set in the root node until that time, namely the
675 * one corresponding to this CPU, due to the fact that we have
678 for (rnp = &rsp->node[0]; rnp < &rsp->node[NUM_RCU_NODES]; rnp++) {
679 spin_lock(&rnp->lock); /* irqs already disabled. */
680 rcu_preempt_check_blocked_tasks(rnp);
681 rnp->qsmask = rnp->qsmaskinit;
682 rnp->gpnum = rsp->gpnum;
683 spin_unlock(&rnp->lock); /* irqs already disabled. */
686 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
687 spin_unlock_irqrestore(&rsp->onofflock, flags);
691 * Advance this CPU's callbacks, but only if the current grace period
692 * has ended. This may be called only from the CPU to whom the rdp
696 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
701 local_irq_save(flags);
702 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
704 /* Did another grace period end? */
705 if (rdp->completed != completed_snap) {
707 /* Advance callbacks. No harm if list empty. */
708 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
709 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
710 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
712 /* Remember that we saw this grace-period completion. */
713 rdp->completed = completed_snap;
715 local_irq_restore(flags);
719 * Clean up after the prior grace period and let rcu_start_gp() start up
720 * the next grace period if one is needed. Note that the caller must
721 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
723 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
724 __releases(rcu_get_root(rsp)->lock)
726 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
727 rsp->completed = rsp->gpnum;
728 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
729 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
733 * Similar to cpu_quiet(), for which it is a helper function. Allows
734 * a group of CPUs to be quieted at one go, though all the CPUs in the
735 * group must be represented by the same leaf rcu_node structure.
736 * That structure's lock must be held upon entry, and it is released
740 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
742 __releases(rnp->lock)
744 struct rcu_node *rnp_c;
746 /* Walk up the rcu_node hierarchy. */
748 if (!(rnp->qsmask & mask)) {
750 /* Our bit has already been cleared, so done. */
751 spin_unlock_irqrestore(&rnp->lock, flags);
754 rnp->qsmask &= ~mask;
755 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
757 /* Other bits still set at this level, so done. */
758 spin_unlock_irqrestore(&rnp->lock, flags);
762 if (rnp->parent == NULL) {
764 /* No more levels. Exit loop holding root lock. */
768 spin_unlock_irqrestore(&rnp->lock, flags);
771 spin_lock_irqsave(&rnp->lock, flags);
772 WARN_ON_ONCE(rnp_c->qsmask);
776 * Get here if we are the last CPU to pass through a quiescent
777 * state for this grace period. Invoke cpu_quiet_msk_finish()
778 * to clean up and start the next grace period if one is needed.
780 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
784 * Record a quiescent state for the specified CPU, which must either be
785 * the current CPU. The lastcomp argument is used to make sure we are
786 * still in the grace period of interest. We don't want to end the current
787 * grace period based on quiescent states detected in an earlier grace
791 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
795 struct rcu_node *rnp;
798 spin_lock_irqsave(&rnp->lock, flags);
799 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
802 * Someone beat us to it for this grace period, so leave.
803 * The race with GP start is resolved by the fact that we
804 * hold the leaf rcu_node lock, so that the per-CPU bits
805 * cannot yet be initialized -- so we would simply find our
806 * CPU's bit already cleared in cpu_quiet_msk() if this race
809 rdp->passed_quiesc = 0; /* try again later! */
810 spin_unlock_irqrestore(&rnp->lock, flags);
814 if ((rnp->qsmask & mask) == 0) {
815 spin_unlock_irqrestore(&rnp->lock, flags);
820 * This GP can't end until cpu checks in, so all of our
821 * callbacks can be processed during the next GP.
823 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
825 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
830 * Check to see if there is a new grace period of which this CPU
831 * is not yet aware, and if so, set up local rcu_data state for it.
832 * Otherwise, see if this CPU has just passed through its first
833 * quiescent state for this grace period, and record that fact if so.
836 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
838 /* If there is now a new grace period, record and return. */
839 if (check_for_new_grace_period(rsp, rdp))
843 * Does this CPU still need to do its part for current grace period?
844 * If no, return and let the other CPUs do their part as well.
846 if (!rdp->qs_pending)
850 * Was there a quiescent state since the beginning of the grace
851 * period? If no, then exit and wait for the next call.
853 if (!rdp->passed_quiesc)
856 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
857 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
860 #ifdef CONFIG_HOTPLUG_CPU
863 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
864 * and move all callbacks from the outgoing CPU to the current one.
866 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
872 struct rcu_data *rdp = rsp->rda[cpu];
873 struct rcu_data *rdp_me;
874 struct rcu_node *rnp;
876 /* Exclude any attempts to start a new grace period. */
877 spin_lock_irqsave(&rsp->onofflock, flags);
879 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
880 rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */
881 mask = rdp->grpmask; /* rnp->grplo is constant. */
883 spin_lock(&rnp->lock); /* irqs already disabled. */
884 rnp->qsmaskinit &= ~mask;
885 if (rnp->qsmaskinit != 0) {
886 spin_unlock(&rnp->lock); /* irqs remain disabled. */
889 rcu_preempt_offline_tasks(rsp, rnp, rdp);
891 spin_unlock(&rnp->lock); /* irqs remain disabled. */
893 } while (rnp != NULL);
894 lastcomp = rsp->completed;
896 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
899 * Move callbacks from the outgoing CPU to the running CPU.
900 * Note that the outgoing CPU is now quiescent, so it is now
901 * (uncharacteristically) safe to access its rcu_data structure.
902 * Note also that we must carefully retain the order of the
903 * outgoing CPU's callbacks in order for rcu_barrier() to work
904 * correctly. Finally, note that we start all the callbacks
905 * afresh, even those that have passed through a grace period
906 * and are therefore ready to invoke. The theory is that hotplug
907 * events are rare, and that if they are frequent enough to
908 * indefinitely delay callbacks, you have far worse things to
911 rdp_me = rsp->rda[smp_processor_id()];
912 if (rdp->nxtlist != NULL) {
913 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
914 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
916 for (i = 0; i < RCU_NEXT_SIZE; i++)
917 rdp->nxttail[i] = &rdp->nxtlist;
918 rdp_me->qlen += rdp->qlen;
921 local_irq_restore(flags);
925 * Remove the specified CPU from the RCU hierarchy and move any pending
926 * callbacks that it might have to the current CPU. This code assumes
927 * that at least one CPU in the system will remain running at all times.
928 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
930 static void rcu_offline_cpu(int cpu)
932 __rcu_offline_cpu(cpu, &rcu_sched_state);
933 __rcu_offline_cpu(cpu, &rcu_bh_state);
934 rcu_preempt_offline_cpu(cpu);
937 #else /* #ifdef CONFIG_HOTPLUG_CPU */
939 static void rcu_offline_cpu(int cpu)
943 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
946 * Invoke any RCU callbacks that have made it to the end of their grace
947 * period. Thottle as specified by rdp->blimit.
949 static void rcu_do_batch(struct rcu_data *rdp)
952 struct rcu_head *next, *list, **tail;
955 /* If no callbacks are ready, just return.*/
956 if (!cpu_has_callbacks_ready_to_invoke(rdp))
960 * Extract the list of ready callbacks, disabling to prevent
961 * races with call_rcu() from interrupt handlers.
963 local_irq_save(flags);
965 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
966 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
967 tail = rdp->nxttail[RCU_DONE_TAIL];
968 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
969 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
970 rdp->nxttail[count] = &rdp->nxtlist;
971 local_irq_restore(flags);
973 /* Invoke callbacks. */
980 if (++count >= rdp->blimit)
984 local_irq_save(flags);
986 /* Update count, and requeue any remaining callbacks. */
989 *tail = rdp->nxtlist;
991 for (count = 0; count < RCU_NEXT_SIZE; count++)
992 if (&rdp->nxtlist == rdp->nxttail[count])
993 rdp->nxttail[count] = tail;
998 /* Reinstate batch limit if we have worked down the excess. */
999 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
1000 rdp->blimit = blimit;
1002 local_irq_restore(flags);
1004 /* Re-raise the RCU softirq if there are callbacks remaining. */
1005 if (cpu_has_callbacks_ready_to_invoke(rdp))
1006 raise_softirq(RCU_SOFTIRQ);
1010 * Check to see if this CPU is in a non-context-switch quiescent state
1011 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1012 * Also schedule the RCU softirq handler.
1014 * This function must be called with hardirqs disabled. It is normally
1015 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1016 * false, there is no point in invoking rcu_check_callbacks().
1018 void rcu_check_callbacks(int cpu, int user)
1020 if (!rcu_pending(cpu))
1021 return; /* if nothing for RCU to do. */
1023 (idle_cpu(cpu) && rcu_scheduler_active &&
1024 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1027 * Get here if this CPU took its interrupt from user
1028 * mode or from the idle loop, and if this is not a
1029 * nested interrupt. In this case, the CPU is in
1030 * a quiescent state, so note it.
1032 * No memory barrier is required here because both
1033 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1034 * variables that other CPUs neither access nor modify,
1035 * at least not while the corresponding CPU is online.
1041 } else if (!in_softirq()) {
1044 * Get here if this CPU did not take its interrupt from
1045 * softirq, in other words, if it is not interrupting
1046 * a rcu_bh read-side critical section. This is an _bh
1047 * critical section, so note it.
1052 rcu_preempt_check_callbacks(cpu);
1053 raise_softirq(RCU_SOFTIRQ);
1059 * Scan the leaf rcu_node structures, processing dyntick state for any that
1060 * have not yet encountered a quiescent state, using the function specified.
1061 * Returns 1 if the current grace period ends while scanning (possibly
1062 * because we made it end).
1064 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1065 int (*f)(struct rcu_data *))
1069 unsigned long flags;
1071 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
1072 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1074 for (; rnp_cur < rnp_end; rnp_cur++) {
1076 spin_lock_irqsave(&rnp_cur->lock, flags);
1077 if (rsp->completed != lastcomp) {
1078 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1081 if (rnp_cur->qsmask == 0) {
1082 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1085 cpu = rnp_cur->grplo;
1087 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1088 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1091 if (mask != 0 && rsp->completed == lastcomp) {
1093 /* cpu_quiet_msk() releases rnp_cur->lock. */
1094 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1097 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1103 * Force quiescent states on reluctant CPUs, and also detect which
1104 * CPUs are in dyntick-idle mode.
1106 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1108 unsigned long flags;
1110 struct rcu_node *rnp = rcu_get_root(rsp);
1113 if (!rcu_gp_in_progress(rsp))
1114 return; /* No grace period in progress, nothing to force. */
1115 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1116 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1117 return; /* Someone else is already on the job. */
1120 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1121 goto unlock_ret; /* no emergency and done recently. */
1123 spin_lock(&rnp->lock);
1124 lastcomp = rsp->completed;
1125 signaled = rsp->signaled;
1126 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1127 if (lastcomp == rsp->gpnum) {
1128 rsp->n_force_qs_ngp++;
1129 spin_unlock(&rnp->lock);
1130 goto unlock_ret; /* no GP in progress, time updated. */
1132 spin_unlock(&rnp->lock);
1136 break; /* grace period still initializing, ignore. */
1138 case RCU_SAVE_DYNTICK:
1140 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1141 break; /* So gcc recognizes the dead code. */
1143 /* Record dyntick-idle state. */
1144 if (rcu_process_dyntick(rsp, lastcomp,
1145 dyntick_save_progress_counter))
1148 /* Update state, record completion counter. */
1149 spin_lock(&rnp->lock);
1150 if (lastcomp == rsp->completed) {
1151 rsp->signaled = RCU_FORCE_QS;
1152 dyntick_record_completed(rsp, lastcomp);
1154 spin_unlock(&rnp->lock);
1159 /* Check dyntick-idle state, send IPI to laggarts. */
1160 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1161 rcu_implicit_dynticks_qs))
1164 /* Leave state in case more forcing is required. */
1169 spin_unlock_irqrestore(&rsp->fqslock, flags);
1172 #else /* #ifdef CONFIG_SMP */
1174 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1179 #endif /* #else #ifdef CONFIG_SMP */
1182 * This does the RCU processing work from softirq context for the
1183 * specified rcu_state and rcu_data structures. This may be called
1184 * only from the CPU to whom the rdp belongs.
1187 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1189 unsigned long flags;
1191 WARN_ON_ONCE(rdp->beenonline == 0);
1194 * If an RCU GP has gone long enough, go check for dyntick
1195 * idle CPUs and, if needed, send resched IPIs.
1197 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1198 force_quiescent_state(rsp, 1);
1201 * Advance callbacks in response to end of earlier grace
1202 * period that some other CPU ended.
1204 rcu_process_gp_end(rsp, rdp);
1206 /* Update RCU state based on any recent quiescent states. */
1207 rcu_check_quiescent_state(rsp, rdp);
1209 /* Does this CPU require a not-yet-started grace period? */
1210 if (cpu_needs_another_gp(rsp, rdp)) {
1211 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1212 rcu_start_gp(rsp, flags); /* releases above lock */
1215 /* If there are callbacks ready, invoke them. */
1220 * Do softirq processing for the current CPU.
1222 static void rcu_process_callbacks(struct softirq_action *unused)
1225 * Memory references from any prior RCU read-side critical sections
1226 * executed by the interrupted code must be seen before any RCU
1227 * grace-period manipulations below.
1229 smp_mb(); /* See above block comment. */
1231 __rcu_process_callbacks(&rcu_sched_state,
1232 &__get_cpu_var(rcu_sched_data));
1233 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1234 rcu_preempt_process_callbacks();
1237 * Memory references from any later RCU read-side critical sections
1238 * executed by the interrupted code must be seen after any RCU
1239 * grace-period manipulations above.
1241 smp_mb(); /* See above block comment. */
1245 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1246 struct rcu_state *rsp)
1248 unsigned long flags;
1249 struct rcu_data *rdp;
1254 smp_mb(); /* Ensure RCU update seen before callback registry. */
1257 * Opportunistically note grace-period endings and beginnings.
1258 * Note that we might see a beginning right after we see an
1259 * end, but never vice versa, since this CPU has to pass through
1260 * a quiescent state betweentimes.
1262 local_irq_save(flags);
1263 rdp = rsp->rda[smp_processor_id()];
1264 rcu_process_gp_end(rsp, rdp);
1265 check_for_new_grace_period(rsp, rdp);
1267 /* Add the callback to our list. */
1268 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1269 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1271 /* Start a new grace period if one not already started. */
1272 if (!rcu_gp_in_progress(rsp)) {
1273 unsigned long nestflag;
1274 struct rcu_node *rnp_root = rcu_get_root(rsp);
1276 spin_lock_irqsave(&rnp_root->lock, nestflag);
1277 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1280 /* Force the grace period if too many callbacks or too long waiting. */
1281 if (unlikely(++rdp->qlen > qhimark)) {
1282 rdp->blimit = LONG_MAX;
1283 force_quiescent_state(rsp, 0);
1284 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1285 force_quiescent_state(rsp, 1);
1286 local_irq_restore(flags);
1290 * Queue an RCU-sched callback for invocation after a grace period.
1292 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1294 __call_rcu(head, func, &rcu_sched_state);
1296 EXPORT_SYMBOL_GPL(call_rcu_sched);
1299 * Queue an RCU for invocation after a quicker grace period.
1301 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1303 __call_rcu(head, func, &rcu_bh_state);
1305 EXPORT_SYMBOL_GPL(call_rcu_bh);
1308 * Check to see if there is any immediate RCU-related work to be done
1309 * by the current CPU, for the specified type of RCU, returning 1 if so.
1310 * The checks are in order of increasing expense: checks that can be
1311 * carried out against CPU-local state are performed first. However,
1312 * we must check for CPU stalls first, else we might not get a chance.
1314 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1316 rdp->n_rcu_pending++;
1318 /* Check for CPU stalls, if enabled. */
1319 check_cpu_stall(rsp, rdp);
1321 /* Is the RCU core waiting for a quiescent state from this CPU? */
1322 if (rdp->qs_pending) {
1323 rdp->n_rp_qs_pending++;
1327 /* Does this CPU have callbacks ready to invoke? */
1328 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1329 rdp->n_rp_cb_ready++;
1333 /* Has RCU gone idle with this CPU needing another grace period? */
1334 if (cpu_needs_another_gp(rsp, rdp)) {
1335 rdp->n_rp_cpu_needs_gp++;
1339 /* Has another RCU grace period completed? */
1340 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1341 rdp->n_rp_gp_completed++;
1345 /* Has a new RCU grace period started? */
1346 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1347 rdp->n_rp_gp_started++;
1351 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1352 if (rcu_gp_in_progress(rsp) &&
1353 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1354 rdp->n_rp_need_fqs++;
1359 rdp->n_rp_need_nothing++;
1364 * Check to see if there is any immediate RCU-related work to be done
1365 * by the current CPU, returning 1 if so. This function is part of the
1366 * RCU implementation; it is -not- an exported member of the RCU API.
1368 static int rcu_pending(int cpu)
1370 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1371 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1372 rcu_preempt_pending(cpu);
1376 * Check to see if any future RCU-related work will need to be done
1377 * by the current CPU, even if none need be done immediately, returning
1378 * 1 if so. This function is part of the RCU implementation; it is -not-
1379 * an exported member of the RCU API.
1381 int rcu_needs_cpu(int cpu)
1383 /* RCU callbacks either ready or pending? */
1384 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1385 per_cpu(rcu_bh_data, cpu).nxtlist ||
1386 rcu_preempt_needs_cpu(cpu);
1390 * Do boot-time initialization of a CPU's per-CPU RCU data.
1393 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1395 unsigned long flags;
1397 struct rcu_data *rdp = rsp->rda[cpu];
1398 struct rcu_node *rnp = rcu_get_root(rsp);
1400 /* Set up local state, ensuring consistent view of global state. */
1401 spin_lock_irqsave(&rnp->lock, flags);
1402 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1403 rdp->nxtlist = NULL;
1404 for (i = 0; i < RCU_NEXT_SIZE; i++)
1405 rdp->nxttail[i] = &rdp->nxtlist;
1408 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1409 #endif /* #ifdef CONFIG_NO_HZ */
1411 spin_unlock_irqrestore(&rnp->lock, flags);
1415 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1416 * offline event can be happening at a given time. Note also that we
1417 * can accept some slop in the rsp->completed access due to the fact
1418 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1420 static void __cpuinit
1421 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1423 unsigned long flags;
1426 struct rcu_data *rdp = rsp->rda[cpu];
1427 struct rcu_node *rnp = rcu_get_root(rsp);
1429 /* Set up local state, ensuring consistent view of global state. */
1430 spin_lock_irqsave(&rnp->lock, flags);
1431 lastcomp = rsp->completed;
1432 rdp->completed = lastcomp;
1433 rdp->gpnum = lastcomp;
1434 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1435 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1436 rdp->beenonline = 1; /* We have now been online. */
1437 rdp->preemptable = preemptable;
1438 rdp->passed_quiesc_completed = lastcomp - 1;
1439 rdp->blimit = blimit;
1440 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1443 * A new grace period might start here. If so, we won't be part
1444 * of it, but that is OK, as we are currently in a quiescent state.
1447 /* Exclude any attempts to start a new GP on large systems. */
1448 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1450 /* Add CPU to rcu_node bitmasks. */
1452 mask = rdp->grpmask;
1454 /* Exclude any attempts to start a new GP on small systems. */
1455 spin_lock(&rnp->lock); /* irqs already disabled. */
1456 rnp->qsmaskinit |= mask;
1457 mask = rnp->grpmask;
1458 spin_unlock(&rnp->lock); /* irqs already disabled. */
1460 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1462 spin_unlock_irqrestore(&rsp->onofflock, flags);
1465 static void __cpuinit rcu_online_cpu(int cpu)
1467 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1468 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1469 rcu_preempt_init_percpu_data(cpu);
1473 * Handle CPU online/offline notification events.
1475 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1476 unsigned long action, void *hcpu)
1478 long cpu = (long)hcpu;
1481 case CPU_UP_PREPARE:
1482 case CPU_UP_PREPARE_FROZEN:
1483 rcu_online_cpu(cpu);
1486 case CPU_DEAD_FROZEN:
1487 case CPU_UP_CANCELED:
1488 case CPU_UP_CANCELED_FROZEN:
1489 rcu_offline_cpu(cpu);
1498 * Compute the per-level fanout, either using the exact fanout specified
1499 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1501 #ifdef CONFIG_RCU_FANOUT_EXACT
1502 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1506 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1507 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1509 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1510 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1517 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1518 ccur = rsp->levelcnt[i];
1519 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1523 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1526 * Helper function for rcu_init() that initializes one rcu_state structure.
1528 static void __init rcu_init_one(struct rcu_state *rsp)
1533 struct rcu_node *rnp;
1535 /* Initialize the level-tracking arrays. */
1537 for (i = 1; i < NUM_RCU_LVLS; i++)
1538 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1539 rcu_init_levelspread(rsp);
1541 /* Initialize the elements themselves, starting from the leaves. */
1543 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1544 cpustride *= rsp->levelspread[i];
1545 rnp = rsp->level[i];
1546 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1547 spin_lock_init(&rnp->lock);
1550 rnp->qsmaskinit = 0;
1551 rnp->grplo = j * cpustride;
1552 rnp->grphi = (j + 1) * cpustride - 1;
1553 if (rnp->grphi >= NR_CPUS)
1554 rnp->grphi = NR_CPUS - 1;
1560 rnp->grpnum = j % rsp->levelspread[i - 1];
1561 rnp->grpmask = 1UL << rnp->grpnum;
1562 rnp->parent = rsp->level[i - 1] +
1563 j / rsp->levelspread[i - 1];
1566 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1567 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1573 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1574 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1577 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1579 rcu_init_one(rsp); \
1580 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1582 for_each_possible_cpu(i) { \
1583 if (i > rnp[j].grphi) \
1585 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1586 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1587 rcu_boot_init_percpu_data(i, rsp); \
1591 void __init __rcu_init(void)
1593 int i; /* All used by RCU_INIT_FLAVOR(). */
1595 struct rcu_node *rnp;
1597 rcu_bootup_announce();
1598 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1599 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1600 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1601 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1602 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1603 __rcu_init_preempt();
1604 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1607 #include "rcutree_plugin.h"
1609 module_param(blimit, int, 0);
1610 module_param(qhimark, int, 0);
1611 module_param(qlowmark, int, 0);