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_state = RCU_STATE_INITIALIZER(rcu_state);
78 DEFINE_PER_CPU(struct rcu_data, rcu_data);
80 struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
81 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
84 * Increment the quiescent state counter.
85 * The counter is a bit degenerated: We do not need to know
86 * how many quiescent states passed, just if there was at least
87 * one since the start of the grace period. Thus just a flag.
89 void rcu_qsctr_inc(int cpu)
91 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
92 rdp->passed_quiesc = 1;
93 rdp->passed_quiesc_completed = rdp->completed;
96 void rcu_bh_qsctr_inc(int cpu)
98 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
99 rdp->passed_quiesc = 1;
100 rdp->passed_quiesc_completed = rdp->completed;
104 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
105 .dynticks_nesting = 1,
108 #endif /* #ifdef CONFIG_NO_HZ */
110 static int blimit = 10; /* Maximum callbacks per softirq. */
111 static int qhimark = 10000; /* If this many pending, ignore blimit. */
112 static int qlowmark = 100; /* Once only this many pending, use blimit. */
114 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
117 * Return the number of RCU batches processed thus far for debug & stats.
119 long rcu_batches_completed(void)
121 return rcu_state.completed;
123 EXPORT_SYMBOL_GPL(rcu_batches_completed);
126 * Return the number of RCU BH batches processed thus far for debug & stats.
128 long rcu_batches_completed_bh(void)
130 return rcu_bh_state.completed;
132 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
135 * Does the CPU have callbacks ready to be invoked?
138 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
140 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
144 * Does the current CPU require a yet-as-unscheduled grace period?
147 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
149 /* ACCESS_ONCE() because we are accessing outside of lock. */
150 return *rdp->nxttail[RCU_DONE_TAIL] &&
151 ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
155 * Return the root node of the specified rcu_state structure.
157 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
159 return &rsp->node[0];
165 * If the specified CPU is offline, tell the caller that it is in
166 * a quiescent state. Otherwise, whack it with a reschedule IPI.
167 * Grace periods can end up waiting on an offline CPU when that
168 * CPU is in the process of coming online -- it will be added to the
169 * rcu_node bitmasks before it actually makes it online. The same thing
170 * can happen while a CPU is in the process of coming online. Because this
171 * race is quite rare, we check for it after detecting that the grace
172 * period has been delayed rather than checking each and every CPU
173 * each and every time we start a new grace period.
175 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
178 * If the CPU is offline, it is in a quiescent state. We can
179 * trust its state not to change because interrupts are disabled.
181 if (cpu_is_offline(rdp->cpu)) {
186 /* The CPU is online, so send it a reschedule IPI. */
187 if (rdp->cpu != smp_processor_id())
188 smp_send_reschedule(rdp->cpu);
195 #endif /* #ifdef CONFIG_SMP */
198 static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5);
201 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
203 * Enter nohz mode, in other words, -leave- the mode in which RCU
204 * read-side critical sections can occur. (Though RCU read-side
205 * critical sections can occur in irq handlers in nohz mode, a possibility
206 * handled by rcu_irq_enter() and rcu_irq_exit()).
208 void rcu_enter_nohz(void)
211 struct rcu_dynticks *rdtp;
213 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
214 local_irq_save(flags);
215 rdtp = &__get_cpu_var(rcu_dynticks);
217 rdtp->dynticks_nesting--;
218 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
219 local_irq_restore(flags);
223 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
225 * Exit nohz mode, in other words, -enter- the mode in which RCU
226 * read-side critical sections normally occur.
228 void rcu_exit_nohz(void)
231 struct rcu_dynticks *rdtp;
233 local_irq_save(flags);
234 rdtp = &__get_cpu_var(rcu_dynticks);
236 rdtp->dynticks_nesting++;
237 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
238 local_irq_restore(flags);
239 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
243 * rcu_nmi_enter - inform RCU of entry to NMI context
245 * If the CPU was idle with dynamic ticks active, and there is no
246 * irq handler running, this updates rdtp->dynticks_nmi to let the
247 * RCU grace-period handling know that the CPU is active.
249 void rcu_nmi_enter(void)
251 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
253 if (rdtp->dynticks & 0x1)
255 rdtp->dynticks_nmi++;
256 WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs);
257 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
261 * rcu_nmi_exit - inform RCU of exit from NMI context
263 * If the CPU was idle with dynamic ticks active, and there is no
264 * irq handler running, this updates rdtp->dynticks_nmi to let the
265 * RCU grace-period handling know that the CPU is no longer active.
267 void rcu_nmi_exit(void)
269 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
271 if (rdtp->dynticks & 0x1)
273 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
274 rdtp->dynticks_nmi++;
275 WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs);
279 * rcu_irq_enter - inform RCU of entry to hard irq context
281 * If the CPU was idle with dynamic ticks active, this updates the
282 * rdtp->dynticks to let the RCU handling know that the CPU is active.
284 void rcu_irq_enter(void)
286 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
288 if (rdtp->dynticks_nesting++)
291 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
292 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
296 * rcu_irq_exit - inform RCU of exit from hard irq context
298 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
299 * to put let the RCU handling be aware that the CPU is going back to idle
302 void rcu_irq_exit(void)
304 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
306 if (--rdtp->dynticks_nesting)
308 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
310 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
312 /* If the interrupt queued a callback, get out of dyntick mode. */
313 if (__get_cpu_var(rcu_data).nxtlist ||
314 __get_cpu_var(rcu_bh_data).nxtlist)
319 * Record the specified "completed" value, which is later used to validate
320 * dynticks counter manipulations. Specify "rsp->completed - 1" to
321 * unconditionally invalidate any future dynticks manipulations (which is
322 * useful at the beginning of a grace period).
324 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
326 rsp->dynticks_completed = comp;
332 * Recall the previously recorded value of the completion for dynticks.
334 static long dyntick_recall_completed(struct rcu_state *rsp)
336 return rsp->dynticks_completed;
340 * Snapshot the specified CPU's dynticks counter so that we can later
341 * credit them with an implicit quiescent state. Return 1 if this CPU
342 * is already in a quiescent state courtesy of dynticks idle mode.
344 static int dyntick_save_progress_counter(struct rcu_data *rdp)
350 snap = rdp->dynticks->dynticks;
351 snap_nmi = rdp->dynticks->dynticks_nmi;
352 smp_mb(); /* Order sampling of snap with end of grace period. */
353 rdp->dynticks_snap = snap;
354 rdp->dynticks_nmi_snap = snap_nmi;
355 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
362 * Return true if the specified CPU has passed through a quiescent
363 * state by virtue of being in or having passed through an dynticks
364 * idle state since the last call to dyntick_save_progress_counter()
367 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
374 curr = rdp->dynticks->dynticks;
375 snap = rdp->dynticks_snap;
376 curr_nmi = rdp->dynticks->dynticks_nmi;
377 snap_nmi = rdp->dynticks_nmi_snap;
378 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
381 * If the CPU passed through or entered a dynticks idle phase with
382 * no active irq/NMI handlers, then we can safely pretend that the CPU
383 * already acknowledged the request to pass through a quiescent
384 * state. Either way, that CPU cannot possibly be in an RCU
385 * read-side critical section that started before the beginning
386 * of the current RCU grace period.
388 if ((curr != snap || (curr & 0x1) == 0) &&
389 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
394 /* Go check for the CPU being offline. */
395 return rcu_implicit_offline_qs(rdp);
398 #endif /* #ifdef CONFIG_SMP */
400 #else /* #ifdef CONFIG_NO_HZ */
402 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
409 * If there are no dynticks, then the only way that a CPU can passively
410 * be in a quiescent state is to be offline. Unlike dynticks idle, which
411 * is a point in time during the prior (already finished) grace period,
412 * an offline CPU is always in a quiescent state, and thus can be
413 * unconditionally applied. So just return the current value of completed.
415 static long dyntick_recall_completed(struct rcu_state *rsp)
417 return rsp->completed;
420 static int dyntick_save_progress_counter(struct rcu_data *rdp)
425 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
427 return rcu_implicit_offline_qs(rdp);
430 #endif /* #ifdef CONFIG_SMP */
432 #endif /* #else #ifdef CONFIG_NO_HZ */
434 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
436 static void record_gp_stall_check_time(struct rcu_state *rsp)
438 rsp->gp_start = jiffies;
439 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
442 static void print_other_cpu_stall(struct rcu_state *rsp)
447 struct rcu_node *rnp = rcu_get_root(rsp);
448 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
449 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
451 /* Only let one CPU complain about others per time interval. */
453 spin_lock_irqsave(&rnp->lock, flags);
454 delta = jiffies - rsp->jiffies_stall;
455 if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) {
456 spin_unlock_irqrestore(&rnp->lock, flags);
459 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
460 spin_unlock_irqrestore(&rnp->lock, flags);
462 /* OK, time to rat on our buddy... */
464 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
465 for (; rnp_cur < rnp_end; rnp_cur++) {
466 if (rnp_cur->qsmask == 0)
468 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
469 if (rnp_cur->qsmask & (1UL << cpu))
470 printk(" %d", rnp_cur->grplo + cpu);
472 printk(" (detected by %d, t=%ld jiffies)\n",
473 smp_processor_id(), (long)(jiffies - rsp->gp_start));
474 force_quiescent_state(rsp, 0); /* Kick them all. */
477 static void print_cpu_stall(struct rcu_state *rsp)
480 struct rcu_node *rnp = rcu_get_root(rsp);
482 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
483 smp_processor_id(), jiffies - rsp->gp_start);
485 spin_lock_irqsave(&rnp->lock, flags);
486 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
488 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
489 spin_unlock_irqrestore(&rnp->lock, flags);
490 set_need_resched(); /* kick ourselves to get things going. */
493 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
496 struct rcu_node *rnp;
498 delta = jiffies - rsp->jiffies_stall;
500 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
502 /* We haven't checked in, so go dump stack. */
503 print_cpu_stall(rsp);
505 } else if (rsp->gpnum != rsp->completed &&
506 delta >= RCU_STALL_RAT_DELAY) {
508 /* They had two time units to dump stack, so complain. */
509 print_other_cpu_stall(rsp);
513 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
515 static void record_gp_stall_check_time(struct rcu_state *rsp)
519 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
523 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
526 * Update CPU-local rcu_data state to record the newly noticed grace period.
527 * This is used both when we started the grace period and when we notice
528 * that someone else started the grace period.
530 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
533 rdp->passed_quiesc = 0;
534 rdp->gpnum = rsp->gpnum;
538 * Did someone else start a new RCU grace period start since we last
539 * checked? Update local state appropriately if so. Must be called
540 * on the CPU corresponding to rdp.
543 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
548 local_irq_save(flags);
549 if (rdp->gpnum != rsp->gpnum) {
550 note_new_gpnum(rsp, rdp);
553 local_irq_restore(flags);
558 * Start a new RCU grace period if warranted, re-initializing the hierarchy
559 * in preparation for detecting the next grace period. The caller must hold
560 * the root node's ->lock, which is released before return. Hard irqs must
564 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
565 __releases(rcu_get_root(rsp)->lock)
567 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
568 struct rcu_node *rnp = rcu_get_root(rsp);
569 struct rcu_node *rnp_cur;
570 struct rcu_node *rnp_end;
572 if (!cpu_needs_another_gp(rsp, rdp)) {
573 spin_unlock_irqrestore(&rnp->lock, flags);
577 /* Advance to a new grace period and initialize state. */
579 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
580 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
581 record_gp_stall_check_time(rsp);
582 dyntick_record_completed(rsp, rsp->completed - 1);
583 note_new_gpnum(rsp, rdp);
586 * Because we are first, we know that all our callbacks will
587 * be covered by this upcoming grace period, even the ones
588 * that were registered arbitrarily recently.
590 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
591 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
593 /* Special-case the common single-level case. */
594 if (NUM_RCU_NODES == 1) {
595 rnp->qsmask = rnp->qsmaskinit;
596 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
597 spin_unlock_irqrestore(&rnp->lock, flags);
601 spin_unlock(&rnp->lock); /* leave irqs disabled. */
604 /* Exclude any concurrent CPU-hotplug operations. */
605 spin_lock(&rsp->onofflock); /* irqs already disabled. */
608 * Set the quiescent-state-needed bits in all the non-leaf RCU
609 * nodes for all currently online CPUs. This operation relies
610 * on the layout of the hierarchy within the rsp->node[] array.
611 * Note that other CPUs will access only the leaves of the
612 * hierarchy, which still indicate that no grace period is in
613 * progress. In addition, we have excluded CPU-hotplug operations.
615 * We therefore do not need to hold any locks. Any required
616 * memory barriers will be supplied by the locks guarding the
617 * leaf rcu_nodes in the hierarchy.
620 rnp_end = rsp->level[NUM_RCU_LVLS - 1];
621 for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++)
622 rnp_cur->qsmask = rnp_cur->qsmaskinit;
625 * Now set up the leaf nodes. Here we must be careful. First,
626 * we need to hold the lock in order to exclude other CPUs, which
627 * might be contending for the leaf nodes' locks. Second, as
628 * soon as we initialize a given leaf node, its CPUs might run
629 * up the rest of the hierarchy. We must therefore acquire locks
630 * for each node that we touch during this stage. (But we still
631 * are excluding CPU-hotplug operations.)
633 * Note that the grace period cannot complete until we finish
634 * the initialization process, as there will be at least one
635 * qsmask bit set in the root node until that time, namely the
636 * one corresponding to this CPU.
638 rnp_end = &rsp->node[NUM_RCU_NODES];
639 rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
640 for (; rnp_cur < rnp_end; rnp_cur++) {
641 spin_lock(&rnp_cur->lock); /* irqs already disabled. */
642 rnp_cur->qsmask = rnp_cur->qsmaskinit;
643 spin_unlock(&rnp_cur->lock); /* irqs already disabled. */
646 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
647 spin_unlock_irqrestore(&rsp->onofflock, flags);
651 * Advance this CPU's callbacks, but only if the current grace period
652 * has ended. This may be called only from the CPU to whom the rdp
656 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
661 local_irq_save(flags);
662 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
664 /* Did another grace period end? */
665 if (rdp->completed != completed_snap) {
667 /* Advance callbacks. No harm if list empty. */
668 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
669 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
670 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
672 /* Remember that we saw this grace-period completion. */
673 rdp->completed = completed_snap;
675 local_irq_restore(flags);
679 * Similar to cpu_quiet(), for which it is a helper function. Allows
680 * a group of CPUs to be quieted at one go, though all the CPUs in the
681 * group must be represented by the same leaf rcu_node structure.
682 * That structure's lock must be held upon entry, and it is released
686 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
688 __releases(rnp->lock)
690 /* Walk up the rcu_node hierarchy. */
692 if (!(rnp->qsmask & mask)) {
694 /* Our bit has already been cleared, so done. */
695 spin_unlock_irqrestore(&rnp->lock, flags);
698 rnp->qsmask &= ~mask;
699 if (rnp->qsmask != 0) {
701 /* Other bits still set at this level, so done. */
702 spin_unlock_irqrestore(&rnp->lock, flags);
706 if (rnp->parent == NULL) {
708 /* No more levels. Exit loop holding root lock. */
712 spin_unlock_irqrestore(&rnp->lock, flags);
714 spin_lock_irqsave(&rnp->lock, flags);
718 * Get here if we are the last CPU to pass through a quiescent
719 * state for this grace period. Clean up and let rcu_start_gp()
720 * start up the next grace period if one is needed. Note that
721 * we still hold rnp->lock, as required by rcu_start_gp(), which
724 rsp->completed = rsp->gpnum;
725 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
726 rcu_start_gp(rsp, flags); /* releases rnp->lock. */
730 * Record a quiescent state for the specified CPU, which must either be
731 * the current CPU or an offline CPU. The lastcomp argument is used to
732 * make sure we are still in the grace period of interest. We don't want
733 * to end the current grace period based on quiescent states detected in
734 * an earlier grace period!
737 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
741 struct rcu_node *rnp;
744 spin_lock_irqsave(&rnp->lock, flags);
745 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
748 * Someone beat us to it for this grace period, so leave.
749 * The race with GP start is resolved by the fact that we
750 * hold the leaf rcu_node lock, so that the per-CPU bits
751 * cannot yet be initialized -- so we would simply find our
752 * CPU's bit already cleared in cpu_quiet_msk() if this race
755 rdp->passed_quiesc = 0; /* try again later! */
756 spin_unlock_irqrestore(&rnp->lock, flags);
760 if ((rnp->qsmask & mask) == 0) {
761 spin_unlock_irqrestore(&rnp->lock, flags);
766 * This GP can't end until cpu checks in, so all of our
767 * callbacks can be processed during the next GP.
769 rdp = rsp->rda[smp_processor_id()];
770 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
772 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
777 * Check to see if there is a new grace period of which this CPU
778 * is not yet aware, and if so, set up local rcu_data state for it.
779 * Otherwise, see if this CPU has just passed through its first
780 * quiescent state for this grace period, and record that fact if so.
783 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
785 /* If there is now a new grace period, record and return. */
786 if (check_for_new_grace_period(rsp, rdp))
790 * Does this CPU still need to do its part for current grace period?
791 * If no, return and let the other CPUs do their part as well.
793 if (!rdp->qs_pending)
797 * Was there a quiescent state since the beginning of the grace
798 * period? If no, then exit and wait for the next call.
800 if (!rdp->passed_quiesc)
803 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
804 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
807 #ifdef CONFIG_HOTPLUG_CPU
810 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
811 * and move all callbacks from the outgoing CPU to the current one.
813 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
819 struct rcu_data *rdp = rsp->rda[cpu];
820 struct rcu_data *rdp_me;
821 struct rcu_node *rnp;
823 /* Exclude any attempts to start a new grace period. */
824 spin_lock_irqsave(&rsp->onofflock, flags);
826 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
828 mask = rdp->grpmask; /* rnp->grplo is constant. */
830 spin_lock(&rnp->lock); /* irqs already disabled. */
831 rnp->qsmaskinit &= ~mask;
832 if (rnp->qsmaskinit != 0) {
833 spin_unlock(&rnp->lock); /* irqs already disabled. */
837 spin_unlock(&rnp->lock); /* irqs already disabled. */
839 } while (rnp != NULL);
840 lastcomp = rsp->completed;
842 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
844 /* Being offline is a quiescent state, so go record it. */
845 cpu_quiet(cpu, rsp, rdp, lastcomp);
848 * Move callbacks from the outgoing CPU to the running CPU.
849 * Note that the outgoing CPU is now quiscent, so it is now
850 * (uncharacteristically) safe to access it rcu_data structure.
851 * Note also that we must carefully retain the order of the
852 * outgoing CPU's callbacks in order for rcu_barrier() to work
853 * correctly. Finally, note that we start all the callbacks
854 * afresh, even those that have passed through a grace period
855 * and are therefore ready to invoke. The theory is that hotplug
856 * events are rare, and that if they are frequent enough to
857 * indefinitely delay callbacks, you have far worse things to
860 rdp_me = rsp->rda[smp_processor_id()];
861 if (rdp->nxtlist != NULL) {
862 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
863 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
865 for (i = 0; i < RCU_NEXT_SIZE; i++)
866 rdp->nxttail[i] = &rdp->nxtlist;
867 rdp_me->qlen += rdp->qlen;
870 local_irq_restore(flags);
874 * Remove the specified CPU from the RCU hierarchy and move any pending
875 * callbacks that it might have to the current CPU. This code assumes
876 * that at least one CPU in the system will remain running at all times.
877 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
879 static void rcu_offline_cpu(int cpu)
881 __rcu_offline_cpu(cpu, &rcu_state);
882 __rcu_offline_cpu(cpu, &rcu_bh_state);
885 #else /* #ifdef CONFIG_HOTPLUG_CPU */
887 static void rcu_offline_cpu(int cpu)
891 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
894 * Invoke any RCU callbacks that have made it to the end of their grace
895 * period. Thottle as specified by rdp->blimit.
897 static void rcu_do_batch(struct rcu_data *rdp)
900 struct rcu_head *next, *list, **tail;
903 /* If no callbacks are ready, just return.*/
904 if (!cpu_has_callbacks_ready_to_invoke(rdp))
908 * Extract the list of ready callbacks, disabling to prevent
909 * races with call_rcu() from interrupt handlers.
911 local_irq_save(flags);
913 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
914 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
915 tail = rdp->nxttail[RCU_DONE_TAIL];
916 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
917 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
918 rdp->nxttail[count] = &rdp->nxtlist;
919 local_irq_restore(flags);
921 /* Invoke callbacks. */
928 if (++count >= rdp->blimit)
932 local_irq_save(flags);
934 /* Update count, and requeue any remaining callbacks. */
937 *tail = rdp->nxtlist;
939 for (count = 0; count < RCU_NEXT_SIZE; count++)
940 if (&rdp->nxtlist == rdp->nxttail[count])
941 rdp->nxttail[count] = tail;
946 /* Reinstate batch limit if we have worked down the excess. */
947 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
948 rdp->blimit = blimit;
950 local_irq_restore(flags);
952 /* Re-raise the RCU softirq if there are callbacks remaining. */
953 if (cpu_has_callbacks_ready_to_invoke(rdp))
954 raise_softirq(RCU_SOFTIRQ);
958 * Check to see if this CPU is in a non-context-switch quiescent state
959 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
960 * Also schedule the RCU softirq handler.
962 * This function must be called with hardirqs disabled. It is normally
963 * invoked from the scheduling-clock interrupt. If rcu_pending returns
964 * false, there is no point in invoking rcu_check_callbacks().
966 void rcu_check_callbacks(int cpu, int user)
969 (idle_cpu(cpu) && rcu_scheduler_active &&
970 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
973 * Get here if this CPU took its interrupt from user
974 * mode or from the idle loop, and if this is not a
975 * nested interrupt. In this case, the CPU is in
976 * a quiescent state, so count it.
978 * No memory barrier is required here because both
979 * rcu_qsctr_inc() and rcu_bh_qsctr_inc() reference
980 * only CPU-local variables that other CPUs neither
981 * access nor modify, at least not while the corresponding
986 rcu_bh_qsctr_inc(cpu);
988 } else if (!in_softirq()) {
991 * Get here if this CPU did not take its interrupt from
992 * softirq, in other words, if it is not interrupting
993 * a rcu_bh read-side critical section. This is an _bh
994 * critical section, so count it.
997 rcu_bh_qsctr_inc(cpu);
999 raise_softirq(RCU_SOFTIRQ);
1005 * Scan the leaf rcu_node structures, processing dyntick state for any that
1006 * have not yet encountered a quiescent state, using the function specified.
1007 * Returns 1 if the current grace period ends while scanning (possibly
1008 * because we made it end).
1010 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1011 int (*f)(struct rcu_data *))
1015 unsigned long flags;
1017 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
1018 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1020 for (; rnp_cur < rnp_end; rnp_cur++) {
1022 spin_lock_irqsave(&rnp_cur->lock, flags);
1023 if (rsp->completed != lastcomp) {
1024 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1027 if (rnp_cur->qsmask == 0) {
1028 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1031 cpu = rnp_cur->grplo;
1033 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1034 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1037 if (mask != 0 && rsp->completed == lastcomp) {
1039 /* cpu_quiet_msk() releases rnp_cur->lock. */
1040 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1043 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1049 * Force quiescent states on reluctant CPUs, and also detect which
1050 * CPUs are in dyntick-idle mode.
1052 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1054 unsigned long flags;
1056 struct rcu_node *rnp = rcu_get_root(rsp);
1059 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
1060 return; /* No grace period in progress, nothing to force. */
1061 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1062 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1063 return; /* Someone else is already on the job. */
1066 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1067 goto unlock_ret; /* no emergency and done recently. */
1069 spin_lock(&rnp->lock);
1070 lastcomp = rsp->completed;
1071 signaled = rsp->signaled;
1072 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1073 if (lastcomp == rsp->gpnum) {
1074 rsp->n_force_qs_ngp++;
1075 spin_unlock(&rnp->lock);
1076 goto unlock_ret; /* no GP in progress, time updated. */
1078 spin_unlock(&rnp->lock);
1082 break; /* grace period still initializing, ignore. */
1084 case RCU_SAVE_DYNTICK:
1086 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1087 break; /* So gcc recognizes the dead code. */
1089 /* Record dyntick-idle state. */
1090 if (rcu_process_dyntick(rsp, lastcomp,
1091 dyntick_save_progress_counter))
1094 /* Update state, record completion counter. */
1095 spin_lock(&rnp->lock);
1096 if (lastcomp == rsp->completed) {
1097 rsp->signaled = RCU_FORCE_QS;
1098 dyntick_record_completed(rsp, lastcomp);
1100 spin_unlock(&rnp->lock);
1105 /* Check dyntick-idle state, send IPI to laggarts. */
1106 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1107 rcu_implicit_dynticks_qs))
1110 /* Leave state in case more forcing is required. */
1115 spin_unlock_irqrestore(&rsp->fqslock, flags);
1118 #else /* #ifdef CONFIG_SMP */
1120 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1125 #endif /* #else #ifdef CONFIG_SMP */
1128 * This does the RCU processing work from softirq context for the
1129 * specified rcu_state and rcu_data structures. This may be called
1130 * only from the CPU to whom the rdp belongs.
1133 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1135 unsigned long flags;
1137 WARN_ON_ONCE(rdp->beenonline == 0);
1140 * If an RCU GP has gone long enough, go check for dyntick
1141 * idle CPUs and, if needed, send resched IPIs.
1143 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1144 force_quiescent_state(rsp, 1);
1147 * Advance callbacks in response to end of earlier grace
1148 * period that some other CPU ended.
1150 rcu_process_gp_end(rsp, rdp);
1152 /* Update RCU state based on any recent quiescent states. */
1153 rcu_check_quiescent_state(rsp, rdp);
1155 /* Does this CPU require a not-yet-started grace period? */
1156 if (cpu_needs_another_gp(rsp, rdp)) {
1157 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1158 rcu_start_gp(rsp, flags); /* releases above lock */
1161 /* If there are callbacks ready, invoke them. */
1166 * Do softirq processing for the current CPU.
1168 static void rcu_process_callbacks(struct softirq_action *unused)
1171 * Memory references from any prior RCU read-side critical sections
1172 * executed by the interrupted code must be seen before any RCU
1173 * grace-period manipulations below.
1175 smp_mb(); /* See above block comment. */
1177 __rcu_process_callbacks(&rcu_state, &__get_cpu_var(rcu_data));
1178 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1181 * Memory references from any later RCU read-side critical sections
1182 * executed by the interrupted code must be seen after any RCU
1183 * grace-period manipulations above.
1185 smp_mb(); /* See above block comment. */
1189 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1190 struct rcu_state *rsp)
1192 unsigned long flags;
1193 struct rcu_data *rdp;
1198 smp_mb(); /* Ensure RCU update seen before callback registry. */
1201 * Opportunistically note grace-period endings and beginnings.
1202 * Note that we might see a beginning right after we see an
1203 * end, but never vice versa, since this CPU has to pass through
1204 * a quiescent state betweentimes.
1206 local_irq_save(flags);
1207 rdp = rsp->rda[smp_processor_id()];
1208 rcu_process_gp_end(rsp, rdp);
1209 check_for_new_grace_period(rsp, rdp);
1211 /* Add the callback to our list. */
1212 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1213 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1215 /* Start a new grace period if one not already started. */
1216 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
1217 unsigned long nestflag;
1218 struct rcu_node *rnp_root = rcu_get_root(rsp);
1220 spin_lock_irqsave(&rnp_root->lock, nestflag);
1221 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1224 /* Force the grace period if too many callbacks or too long waiting. */
1225 if (unlikely(++rdp->qlen > qhimark)) {
1226 rdp->blimit = LONG_MAX;
1227 force_quiescent_state(rsp, 0);
1228 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1229 force_quiescent_state(rsp, 1);
1230 local_irq_restore(flags);
1234 * Queue an RCU callback for invocation after a grace period.
1236 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1238 __call_rcu(head, func, &rcu_state);
1240 EXPORT_SYMBOL_GPL(call_rcu);
1243 * Queue an RCU for invocation after a quicker grace period.
1245 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1247 __call_rcu(head, func, &rcu_bh_state);
1249 EXPORT_SYMBOL_GPL(call_rcu_bh);
1252 * Check to see if there is any immediate RCU-related work to be done
1253 * by the current CPU, for the specified type of RCU, returning 1 if so.
1254 * The checks are in order of increasing expense: checks that can be
1255 * carried out against CPU-local state are performed first. However,
1256 * we must check for CPU stalls first, else we might not get a chance.
1258 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1260 rdp->n_rcu_pending++;
1262 /* Check for CPU stalls, if enabled. */
1263 check_cpu_stall(rsp, rdp);
1265 /* Is the RCU core waiting for a quiescent state from this CPU? */
1266 if (rdp->qs_pending) {
1267 rdp->n_rp_qs_pending++;
1271 /* Does this CPU have callbacks ready to invoke? */
1272 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1273 rdp->n_rp_cb_ready++;
1277 /* Has RCU gone idle with this CPU needing another grace period? */
1278 if (cpu_needs_another_gp(rsp, rdp)) {
1279 rdp->n_rp_cpu_needs_gp++;
1283 /* Has another RCU grace period completed? */
1284 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1285 rdp->n_rp_gp_completed++;
1289 /* Has a new RCU grace period started? */
1290 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1291 rdp->n_rp_gp_started++;
1295 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1296 if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
1297 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1298 rdp->n_rp_need_fqs++;
1303 rdp->n_rp_need_nothing++;
1308 * Check to see if there is any immediate RCU-related work to be done
1309 * by the current CPU, returning 1 if so. This function is part of the
1310 * RCU implementation; it is -not- an exported member of the RCU API.
1312 int rcu_pending(int cpu)
1314 return __rcu_pending(&rcu_state, &per_cpu(rcu_data, cpu)) ||
1315 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu));
1319 * Check to see if any future RCU-related work will need to be done
1320 * by the current CPU, even if none need be done immediately, returning
1321 * 1 if so. This function is part of the RCU implementation; it is -not-
1322 * an exported member of the RCU API.
1324 int rcu_needs_cpu(int cpu)
1326 /* RCU callbacks either ready or pending? */
1327 return per_cpu(rcu_data, cpu).nxtlist ||
1328 per_cpu(rcu_bh_data, cpu).nxtlist;
1332 * Do boot-time initialization of a CPU's per-CPU RCU data.
1335 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1337 unsigned long flags;
1339 struct rcu_data *rdp = rsp->rda[cpu];
1340 struct rcu_node *rnp = rcu_get_root(rsp);
1342 /* Set up local state, ensuring consistent view of global state. */
1343 spin_lock_irqsave(&rnp->lock, flags);
1344 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1345 rdp->nxtlist = NULL;
1346 for (i = 0; i < RCU_NEXT_SIZE; i++)
1347 rdp->nxttail[i] = &rdp->nxtlist;
1350 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1351 #endif /* #ifdef CONFIG_NO_HZ */
1353 spin_unlock_irqrestore(&rnp->lock, flags);
1357 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1358 * offline event can be happening at a given time. Note also that we
1359 * can accept some slop in the rsp->completed access due to the fact
1360 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1362 static void __cpuinit
1363 rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
1365 unsigned long flags;
1368 struct rcu_data *rdp = rsp->rda[cpu];
1369 struct rcu_node *rnp = rcu_get_root(rsp);
1371 /* Set up local state, ensuring consistent view of global state. */
1372 spin_lock_irqsave(&rnp->lock, flags);
1373 lastcomp = rsp->completed;
1374 rdp->completed = lastcomp;
1375 rdp->gpnum = lastcomp;
1376 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1377 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1378 rdp->beenonline = 1; /* We have now been online. */
1379 rdp->passed_quiesc_completed = lastcomp - 1;
1380 rdp->blimit = blimit;
1381 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1384 * A new grace period might start here. If so, we won't be part
1385 * of it, but that is OK, as we are currently in a quiescent state.
1388 /* Exclude any attempts to start a new GP on large systems. */
1389 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1391 /* Add CPU to rcu_node bitmasks. */
1393 mask = rdp->grpmask;
1395 /* Exclude any attempts to start a new GP on small systems. */
1396 spin_lock(&rnp->lock); /* irqs already disabled. */
1397 rnp->qsmaskinit |= mask;
1398 mask = rnp->grpmask;
1399 spin_unlock(&rnp->lock); /* irqs already disabled. */
1401 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1403 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1406 * A new grace period might start here. If so, we will be part of
1407 * it, and its gpnum will be greater than ours, so we will
1408 * participate. It is also possible for the gpnum to have been
1409 * incremented before this function was called, and the bitmasks
1410 * to not be filled out until now, in which case we will also
1411 * participate due to our gpnum being behind.
1414 /* Since it is coming online, the CPU is in a quiescent state. */
1415 cpu_quiet(cpu, rsp, rdp, lastcomp);
1416 local_irq_restore(flags);
1419 static void __cpuinit rcu_online_cpu(int cpu)
1421 rcu_init_percpu_data(cpu, &rcu_state);
1422 rcu_init_percpu_data(cpu, &rcu_bh_state);
1426 * Handle CPU online/offline notifcation events.
1428 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1429 unsigned long action, void *hcpu)
1431 long cpu = (long)hcpu;
1434 case CPU_UP_PREPARE:
1435 case CPU_UP_PREPARE_FROZEN:
1436 rcu_online_cpu(cpu);
1439 case CPU_DEAD_FROZEN:
1440 case CPU_UP_CANCELED:
1441 case CPU_UP_CANCELED_FROZEN:
1442 rcu_offline_cpu(cpu);
1451 * Compute the per-level fanout, either using the exact fanout specified
1452 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1454 #ifdef CONFIG_RCU_FANOUT_EXACT
1455 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1459 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1460 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1462 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1463 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1470 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1471 ccur = rsp->levelcnt[i];
1472 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1476 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1479 * Helper function for rcu_init() that initializes one rcu_state structure.
1481 static void __init rcu_init_one(struct rcu_state *rsp)
1486 struct rcu_node *rnp;
1488 /* Initialize the level-tracking arrays. */
1490 for (i = 1; i < NUM_RCU_LVLS; i++)
1491 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1492 rcu_init_levelspread(rsp);
1494 /* Initialize the elements themselves, starting from the leaves. */
1496 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1497 cpustride *= rsp->levelspread[i];
1498 rnp = rsp->level[i];
1499 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1500 spin_lock_init(&rnp->lock);
1502 rnp->qsmaskinit = 0;
1503 rnp->grplo = j * cpustride;
1504 rnp->grphi = (j + 1) * cpustride - 1;
1505 if (rnp->grphi >= NR_CPUS)
1506 rnp->grphi = NR_CPUS - 1;
1512 rnp->grpnum = j % rsp->levelspread[i - 1];
1513 rnp->grpmask = 1UL << rnp->grpnum;
1514 rnp->parent = rsp->level[i - 1] +
1515 j / rsp->levelspread[i - 1];
1523 * Helper macro for __rcu_init(). To be used nowhere else!
1524 * Assigns leaf node pointers into each CPU's rcu_data structure.
1526 #define RCU_DATA_PTR_INIT(rsp, rcu_data) \
1528 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1530 for_each_possible_cpu(i) { \
1531 if (i > rnp[j].grphi) \
1533 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1534 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1538 void __init __rcu_init(void)
1540 int i; /* All used by RCU_DATA_PTR_INIT(). */
1542 struct rcu_node *rnp;
1544 printk(KERN_INFO "Hierarchical RCU implementation.\n");
1545 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1546 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1547 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1548 rcu_init_one(&rcu_state);
1549 RCU_DATA_PTR_INIT(&rcu_state, rcu_data);
1550 for_each_possible_cpu(i)
1551 rcu_boot_init_percpu_data(i, &rcu_state);
1552 rcu_init_one(&rcu_bh_state);
1553 RCU_DATA_PTR_INIT(&rcu_bh_state, rcu_bh_data);
1554 for_each_possible_cpu(i)
1555 rcu_boot_init_percpu_data(i, &rcu_bh_state);
1556 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1559 module_param(blimit, int, 0);
1560 module_param(qhimark, int, 0);
1561 module_param(qlowmark, int, 0);