2 * Read-Copy Update mechanism for mutual exclusion
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright IBM Corporation, 2008
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
27 * For detailed explanation of Read-Copy Update mechanism see -
30 #include <linux/types.h>
31 #include <linux/kernel.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/smp.h>
35 #include <linux/rcupdate.h>
36 #include <linux/interrupt.h>
37 #include <linux/sched.h>
38 #include <asm/atomic.h>
39 #include <linux/bitops.h>
40 #include <linux/module.h>
41 #include <linux/completion.h>
42 #include <linux/moduleparam.h>
43 #include <linux/percpu.h>
44 #include <linux/notifier.h>
45 #include <linux/cpu.h>
46 #include <linux/mutex.h>
47 #include <linux/time.h>
51 #ifdef CONFIG_DEBUG_LOCK_ALLOC
52 static struct lock_class_key rcu_lock_key
;
53 struct lockdep_map rcu_lock_map
=
54 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key
);
55 EXPORT_SYMBOL_GPL(rcu_lock_map
);
58 /* Data structures. */
60 #define RCU_STATE_INITIALIZER(name) { \
61 .level = { &name.node[0] }, \
63 NUM_RCU_LVL_0, /* root of hierarchy. */ \
66 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
68 .signaled = RCU_SIGNAL_INIT, \
71 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
72 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
74 .n_force_qs_ngp = 0, \
77 struct rcu_state rcu_sched_state
= RCU_STATE_INITIALIZER(rcu_sched_state
);
78 DEFINE_PER_CPU(struct rcu_data
, rcu_sched_data
);
80 struct rcu_state rcu_bh_state
= RCU_STATE_INITIALIZER(rcu_bh_state
);
81 DEFINE_PER_CPU(struct rcu_data
, rcu_bh_data
);
84 * Note a quiescent state. Because we do not need to know
85 * how many quiescent states passed, just if there was at least
86 * one since the start of the grace period, this just sets a flag.
88 void rcu_sched_qs(int cpu
)
90 struct rcu_data
*rdp
= &per_cpu(rcu_sched_data
, cpu
);
91 rdp
->passed_quiesc
= 1;
92 rdp
->passed_quiesc_completed
= rdp
->completed
;
95 void rcu_bh_qs(int cpu
)
97 struct rcu_data
*rdp
= &per_cpu(rcu_bh_data
, cpu
);
98 rdp
->passed_quiesc
= 1;
99 rdp
->passed_quiesc_completed
= rdp
->completed
;
103 DEFINE_PER_CPU(struct rcu_dynticks
, rcu_dynticks
) = {
104 .dynticks_nesting
= 1,
107 #endif /* #ifdef CONFIG_NO_HZ */
109 static int blimit
= 10; /* Maximum callbacks per softirq. */
110 static int qhimark
= 10000; /* If this many pending, ignore blimit. */
111 static int qlowmark
= 100; /* Once only this many pending, use blimit. */
113 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
);
116 * Return the number of RCU-sched batches processed thus far for debug & stats.
118 long rcu_batches_completed_sched(void)
120 return rcu_sched_state
.completed
;
122 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched
);
125 * Return the number of RCU batches processed thus far for debug & stats.
126 * @@@ placeholder, maps to rcu_batches_completed_sched().
128 long rcu_batches_completed(void)
130 return rcu_batches_completed_sched();
132 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
135 * Return the number of RCU BH batches processed thus far for debug & stats.
137 long rcu_batches_completed_bh(void)
139 return rcu_bh_state
.completed
;
141 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh
);
144 * Does the CPU have callbacks ready to be invoked?
147 cpu_has_callbacks_ready_to_invoke(struct rcu_data
*rdp
)
149 return &rdp
->nxtlist
!= rdp
->nxttail
[RCU_DONE_TAIL
];
153 * Does the current CPU require a yet-as-unscheduled grace period?
156 cpu_needs_another_gp(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
158 /* ACCESS_ONCE() because we are accessing outside of lock. */
159 return *rdp
->nxttail
[RCU_DONE_TAIL
] &&
160 ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
);
164 * Return the root node of the specified rcu_state structure.
166 static struct rcu_node
*rcu_get_root(struct rcu_state
*rsp
)
168 return &rsp
->node
[0];
174 * If the specified CPU is offline, tell the caller that it is in
175 * a quiescent state. Otherwise, whack it with a reschedule IPI.
176 * Grace periods can end up waiting on an offline CPU when that
177 * CPU is in the process of coming online -- it will be added to the
178 * rcu_node bitmasks before it actually makes it online. The same thing
179 * can happen while a CPU is in the process of coming online. Because this
180 * race is quite rare, we check for it after detecting that the grace
181 * period has been delayed rather than checking each and every CPU
182 * each and every time we start a new grace period.
184 static int rcu_implicit_offline_qs(struct rcu_data
*rdp
)
187 * If the CPU is offline, it is in a quiescent state. We can
188 * trust its state not to change because interrupts are disabled.
190 if (cpu_is_offline(rdp
->cpu
)) {
195 /* The CPU is online, so send it a reschedule IPI. */
196 if (rdp
->cpu
!= smp_processor_id())
197 smp_send_reschedule(rdp
->cpu
);
204 #endif /* #ifdef CONFIG_SMP */
207 static DEFINE_RATELIMIT_STATE(rcu_rs
, 10 * HZ
, 5);
210 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
212 * Enter nohz mode, in other words, -leave- the mode in which RCU
213 * read-side critical sections can occur. (Though RCU read-side
214 * critical sections can occur in irq handlers in nohz mode, a possibility
215 * handled by rcu_irq_enter() and rcu_irq_exit()).
217 void rcu_enter_nohz(void)
220 struct rcu_dynticks
*rdtp
;
222 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
223 local_irq_save(flags
);
224 rdtp
= &__get_cpu_var(rcu_dynticks
);
226 rdtp
->dynticks_nesting
--;
227 WARN_ON_RATELIMIT(rdtp
->dynticks
& 0x1, &rcu_rs
);
228 local_irq_restore(flags
);
232 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
234 * Exit nohz mode, in other words, -enter- the mode in which RCU
235 * read-side critical sections normally occur.
237 void rcu_exit_nohz(void)
240 struct rcu_dynticks
*rdtp
;
242 local_irq_save(flags
);
243 rdtp
= &__get_cpu_var(rcu_dynticks
);
245 rdtp
->dynticks_nesting
++;
246 WARN_ON_RATELIMIT(!(rdtp
->dynticks
& 0x1), &rcu_rs
);
247 local_irq_restore(flags
);
248 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
252 * rcu_nmi_enter - inform RCU of entry to NMI context
254 * If the CPU was idle with dynamic ticks active, and there is no
255 * irq handler running, this updates rdtp->dynticks_nmi to let the
256 * RCU grace-period handling know that the CPU is active.
258 void rcu_nmi_enter(void)
260 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
262 if (rdtp
->dynticks
& 0x1)
264 rdtp
->dynticks_nmi
++;
265 WARN_ON_RATELIMIT(!(rdtp
->dynticks_nmi
& 0x1), &rcu_rs
);
266 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
270 * rcu_nmi_exit - inform RCU of exit from NMI context
272 * If the CPU was idle with dynamic ticks active, and there is no
273 * irq handler running, this updates rdtp->dynticks_nmi to let the
274 * RCU grace-period handling know that the CPU is no longer active.
276 void rcu_nmi_exit(void)
278 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
280 if (rdtp
->dynticks
& 0x1)
282 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
283 rdtp
->dynticks_nmi
++;
284 WARN_ON_RATELIMIT(rdtp
->dynticks_nmi
& 0x1, &rcu_rs
);
288 * rcu_irq_enter - inform RCU of entry to hard irq context
290 * If the CPU was idle with dynamic ticks active, this updates the
291 * rdtp->dynticks to let the RCU handling know that the CPU is active.
293 void rcu_irq_enter(void)
295 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
297 if (rdtp
->dynticks_nesting
++)
300 WARN_ON_RATELIMIT(!(rdtp
->dynticks
& 0x1), &rcu_rs
);
301 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
305 * rcu_irq_exit - inform RCU of exit from hard irq context
307 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
308 * to put let the RCU handling be aware that the CPU is going back to idle
311 void rcu_irq_exit(void)
313 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
315 if (--rdtp
->dynticks_nesting
)
317 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
319 WARN_ON_RATELIMIT(rdtp
->dynticks
& 0x1, &rcu_rs
);
321 /* If the interrupt queued a callback, get out of dyntick mode. */
322 if (__get_cpu_var(rcu_sched_data
).nxtlist
||
323 __get_cpu_var(rcu_bh_data
).nxtlist
)
328 * Record the specified "completed" value, which is later used to validate
329 * dynticks counter manipulations. Specify "rsp->completed - 1" to
330 * unconditionally invalidate any future dynticks manipulations (which is
331 * useful at the beginning of a grace period).
333 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
335 rsp
->dynticks_completed
= comp
;
341 * Recall the previously recorded value of the completion for dynticks.
343 static long dyntick_recall_completed(struct rcu_state
*rsp
)
345 return rsp
->dynticks_completed
;
349 * Snapshot the specified CPU's dynticks counter so that we can later
350 * credit them with an implicit quiescent state. Return 1 if this CPU
351 * is already in a quiescent state courtesy of dynticks idle mode.
353 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
359 snap
= rdp
->dynticks
->dynticks
;
360 snap_nmi
= rdp
->dynticks
->dynticks_nmi
;
361 smp_mb(); /* Order sampling of snap with end of grace period. */
362 rdp
->dynticks_snap
= snap
;
363 rdp
->dynticks_nmi_snap
= snap_nmi
;
364 ret
= ((snap
& 0x1) == 0) && ((snap_nmi
& 0x1) == 0);
371 * Return true if the specified CPU has passed through a quiescent
372 * state by virtue of being in or having passed through an dynticks
373 * idle state since the last call to dyntick_save_progress_counter()
376 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
383 curr
= rdp
->dynticks
->dynticks
;
384 snap
= rdp
->dynticks_snap
;
385 curr_nmi
= rdp
->dynticks
->dynticks_nmi
;
386 snap_nmi
= rdp
->dynticks_nmi_snap
;
387 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
390 * If the CPU passed through or entered a dynticks idle phase with
391 * no active irq/NMI handlers, then we can safely pretend that the CPU
392 * already acknowledged the request to pass through a quiescent
393 * state. Either way, that CPU cannot possibly be in an RCU
394 * read-side critical section that started before the beginning
395 * of the current RCU grace period.
397 if ((curr
!= snap
|| (curr
& 0x1) == 0) &&
398 (curr_nmi
!= snap_nmi
|| (curr_nmi
& 0x1) == 0)) {
403 /* Go check for the CPU being offline. */
404 return rcu_implicit_offline_qs(rdp
);
407 #endif /* #ifdef CONFIG_SMP */
409 #else /* #ifdef CONFIG_NO_HZ */
411 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
418 * If there are no dynticks, then the only way that a CPU can passively
419 * be in a quiescent state is to be offline. Unlike dynticks idle, which
420 * is a point in time during the prior (already finished) grace period,
421 * an offline CPU is always in a quiescent state, and thus can be
422 * unconditionally applied. So just return the current value of completed.
424 static long dyntick_recall_completed(struct rcu_state
*rsp
)
426 return rsp
->completed
;
429 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
434 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
436 return rcu_implicit_offline_qs(rdp
);
439 #endif /* #ifdef CONFIG_SMP */
441 #endif /* #else #ifdef CONFIG_NO_HZ */
443 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
445 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
447 rsp
->gp_start
= jiffies
;
448 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_CHECK
;
451 static void print_other_cpu_stall(struct rcu_state
*rsp
)
456 struct rcu_node
*rnp
= rcu_get_root(rsp
);
457 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
458 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
460 /* Only let one CPU complain about others per time interval. */
462 spin_lock_irqsave(&rnp
->lock
, flags
);
463 delta
= jiffies
- rsp
->jiffies_stall
;
464 if (delta
< RCU_STALL_RAT_DELAY
|| rsp
->gpnum
== rsp
->completed
) {
465 spin_unlock_irqrestore(&rnp
->lock
, flags
);
468 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
469 spin_unlock_irqrestore(&rnp
->lock
, flags
);
471 /* OK, time to rat on our buddy... */
473 printk(KERN_ERR
"INFO: RCU detected CPU stalls:");
474 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
475 if (rnp_cur
->qsmask
== 0)
477 for (cpu
= 0; cpu
<= rnp_cur
->grphi
- rnp_cur
->grplo
; cpu
++)
478 if (rnp_cur
->qsmask
& (1UL << cpu
))
479 printk(" %d", rnp_cur
->grplo
+ cpu
);
481 printk(" (detected by %d, t=%ld jiffies)\n",
482 smp_processor_id(), (long)(jiffies
- rsp
->gp_start
));
483 force_quiescent_state(rsp
, 0); /* Kick them all. */
486 static void print_cpu_stall(struct rcu_state
*rsp
)
489 struct rcu_node
*rnp
= rcu_get_root(rsp
);
491 printk(KERN_ERR
"INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
492 smp_processor_id(), jiffies
- rsp
->gp_start
);
494 spin_lock_irqsave(&rnp
->lock
, flags
);
495 if ((long)(jiffies
- rsp
->jiffies_stall
) >= 0)
497 jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
498 spin_unlock_irqrestore(&rnp
->lock
, flags
);
499 set_need_resched(); /* kick ourselves to get things going. */
502 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
505 struct rcu_node
*rnp
;
507 delta
= jiffies
- rsp
->jiffies_stall
;
509 if ((rnp
->qsmask
& rdp
->grpmask
) && delta
>= 0) {
511 /* We haven't checked in, so go dump stack. */
512 print_cpu_stall(rsp
);
514 } else if (rsp
->gpnum
!= rsp
->completed
&&
515 delta
>= RCU_STALL_RAT_DELAY
) {
517 /* They had two time units to dump stack, so complain. */
518 print_other_cpu_stall(rsp
);
522 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
524 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
528 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
532 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
535 * Update CPU-local rcu_data state to record the newly noticed grace period.
536 * This is used both when we started the grace period and when we notice
537 * that someone else started the grace period.
539 static void note_new_gpnum(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
542 rdp
->passed_quiesc
= 0;
543 rdp
->gpnum
= rsp
->gpnum
;
547 * Did someone else start a new RCU grace period start since we last
548 * checked? Update local state appropriately if so. Must be called
549 * on the CPU corresponding to rdp.
552 check_for_new_grace_period(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
557 local_irq_save(flags
);
558 if (rdp
->gpnum
!= rsp
->gpnum
) {
559 note_new_gpnum(rsp
, rdp
);
562 local_irq_restore(flags
);
567 * Start a new RCU grace period if warranted, re-initializing the hierarchy
568 * in preparation for detecting the next grace period. The caller must hold
569 * the root node's ->lock, which is released before return. Hard irqs must
573 rcu_start_gp(struct rcu_state
*rsp
, unsigned long flags
)
574 __releases(rcu_get_root(rsp
)->lock
)
576 struct rcu_data
*rdp
= rsp
->rda
[smp_processor_id()];
577 struct rcu_node
*rnp
= rcu_get_root(rsp
);
578 struct rcu_node
*rnp_cur
;
579 struct rcu_node
*rnp_end
;
581 if (!cpu_needs_another_gp(rsp
, rdp
)) {
582 spin_unlock_irqrestore(&rnp
->lock
, flags
);
586 /* Advance to a new grace period and initialize state. */
588 rsp
->signaled
= RCU_GP_INIT
; /* Hold off force_quiescent_state. */
589 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
590 record_gp_stall_check_time(rsp
);
591 dyntick_record_completed(rsp
, rsp
->completed
- 1);
592 note_new_gpnum(rsp
, rdp
);
595 * Because we are first, we know that all our callbacks will
596 * be covered by this upcoming grace period, even the ones
597 * that were registered arbitrarily recently.
599 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
600 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
602 /* Special-case the common single-level case. */
603 if (NUM_RCU_NODES
== 1) {
604 rnp
->qsmask
= rnp
->qsmaskinit
;
605 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state OK. */
606 spin_unlock_irqrestore(&rnp
->lock
, flags
);
610 spin_unlock(&rnp
->lock
); /* leave irqs disabled. */
613 /* Exclude any concurrent CPU-hotplug operations. */
614 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
617 * Set the quiescent-state-needed bits in all the non-leaf RCU
618 * nodes for all currently online CPUs. This operation relies
619 * on the layout of the hierarchy within the rsp->node[] array.
620 * Note that other CPUs will access only the leaves of the
621 * hierarchy, which still indicate that no grace period is in
622 * progress. In addition, we have excluded CPU-hotplug operations.
624 * We therefore do not need to hold any locks. Any required
625 * memory barriers will be supplied by the locks guarding the
626 * leaf rcu_nodes in the hierarchy.
629 rnp_end
= rsp
->level
[NUM_RCU_LVLS
- 1];
630 for (rnp_cur
= &rsp
->node
[0]; rnp_cur
< rnp_end
; rnp_cur
++)
631 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
634 * Now set up the leaf nodes. Here we must be careful. First,
635 * we need to hold the lock in order to exclude other CPUs, which
636 * might be contending for the leaf nodes' locks. Second, as
637 * soon as we initialize a given leaf node, its CPUs might run
638 * up the rest of the hierarchy. We must therefore acquire locks
639 * for each node that we touch during this stage. (But we still
640 * are excluding CPU-hotplug operations.)
642 * Note that the grace period cannot complete until we finish
643 * the initialization process, as there will be at least one
644 * qsmask bit set in the root node until that time, namely the
645 * one corresponding to this CPU.
647 rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
648 rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
649 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
650 spin_lock(&rnp_cur
->lock
); /* irqs already disabled. */
651 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
652 spin_unlock(&rnp_cur
->lock
); /* irqs already disabled. */
655 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state now OK. */
656 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
660 * Advance this CPU's callbacks, but only if the current grace period
661 * has ended. This may be called only from the CPU to whom the rdp
665 rcu_process_gp_end(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
670 local_irq_save(flags
);
671 completed_snap
= ACCESS_ONCE(rsp
->completed
); /* outside of lock. */
673 /* Did another grace period end? */
674 if (rdp
->completed
!= completed_snap
) {
676 /* Advance callbacks. No harm if list empty. */
677 rdp
->nxttail
[RCU_DONE_TAIL
] = rdp
->nxttail
[RCU_WAIT_TAIL
];
678 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_READY_TAIL
];
679 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
681 /* Remember that we saw this grace-period completion. */
682 rdp
->completed
= completed_snap
;
684 local_irq_restore(flags
);
688 * Similar to cpu_quiet(), for which it is a helper function. Allows
689 * a group of CPUs to be quieted at one go, though all the CPUs in the
690 * group must be represented by the same leaf rcu_node structure.
691 * That structure's lock must be held upon entry, and it is released
695 cpu_quiet_msk(unsigned long mask
, struct rcu_state
*rsp
, struct rcu_node
*rnp
,
697 __releases(rnp
->lock
)
699 /* Walk up the rcu_node hierarchy. */
701 if (!(rnp
->qsmask
& mask
)) {
703 /* Our bit has already been cleared, so done. */
704 spin_unlock_irqrestore(&rnp
->lock
, flags
);
707 rnp
->qsmask
&= ~mask
;
708 if (rnp
->qsmask
!= 0) {
710 /* Other bits still set at this level, so done. */
711 spin_unlock_irqrestore(&rnp
->lock
, flags
);
715 if (rnp
->parent
== NULL
) {
717 /* No more levels. Exit loop holding root lock. */
721 spin_unlock_irqrestore(&rnp
->lock
, flags
);
723 spin_lock_irqsave(&rnp
->lock
, flags
);
727 * Get here if we are the last CPU to pass through a quiescent
728 * state for this grace period. Clean up and let rcu_start_gp()
729 * start up the next grace period if one is needed. Note that
730 * we still hold rnp->lock, as required by rcu_start_gp(), which
733 rsp
->completed
= rsp
->gpnum
;
734 rcu_process_gp_end(rsp
, rsp
->rda
[smp_processor_id()]);
735 rcu_start_gp(rsp
, flags
); /* releases rnp->lock. */
739 * Record a quiescent state for the specified CPU, which must either be
740 * the current CPU or an offline CPU. The lastcomp argument is used to
741 * make sure we are still in the grace period of interest. We don't want
742 * to end the current grace period based on quiescent states detected in
743 * an earlier grace period!
746 cpu_quiet(int cpu
, struct rcu_state
*rsp
, struct rcu_data
*rdp
, long lastcomp
)
750 struct rcu_node
*rnp
;
753 spin_lock_irqsave(&rnp
->lock
, flags
);
754 if (lastcomp
!= ACCESS_ONCE(rsp
->completed
)) {
757 * Someone beat us to it for this grace period, so leave.
758 * The race with GP start is resolved by the fact that we
759 * hold the leaf rcu_node lock, so that the per-CPU bits
760 * cannot yet be initialized -- so we would simply find our
761 * CPU's bit already cleared in cpu_quiet_msk() if this race
764 rdp
->passed_quiesc
= 0; /* try again later! */
765 spin_unlock_irqrestore(&rnp
->lock
, flags
);
769 if ((rnp
->qsmask
& mask
) == 0) {
770 spin_unlock_irqrestore(&rnp
->lock
, flags
);
775 * This GP can't end until cpu checks in, so all of our
776 * callbacks can be processed during the next GP.
778 rdp
= rsp
->rda
[smp_processor_id()];
779 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
781 cpu_quiet_msk(mask
, rsp
, rnp
, flags
); /* releases rnp->lock */
786 * Check to see if there is a new grace period of which this CPU
787 * is not yet aware, and if so, set up local rcu_data state for it.
788 * Otherwise, see if this CPU has just passed through its first
789 * quiescent state for this grace period, and record that fact if so.
792 rcu_check_quiescent_state(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
794 /* If there is now a new grace period, record and return. */
795 if (check_for_new_grace_period(rsp
, rdp
))
799 * Does this CPU still need to do its part for current grace period?
800 * If no, return and let the other CPUs do their part as well.
802 if (!rdp
->qs_pending
)
806 * Was there a quiescent state since the beginning of the grace
807 * period? If no, then exit and wait for the next call.
809 if (!rdp
->passed_quiesc
)
812 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
813 cpu_quiet(rdp
->cpu
, rsp
, rdp
, rdp
->passed_quiesc_completed
);
816 #ifdef CONFIG_HOTPLUG_CPU
819 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
820 * and move all callbacks from the outgoing CPU to the current one.
822 static void __rcu_offline_cpu(int cpu
, struct rcu_state
*rsp
)
828 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
829 struct rcu_data
*rdp_me
;
830 struct rcu_node
*rnp
;
832 /* Exclude any attempts to start a new grace period. */
833 spin_lock_irqsave(&rsp
->onofflock
, flags
);
835 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
837 mask
= rdp
->grpmask
; /* rnp->grplo is constant. */
839 spin_lock(&rnp
->lock
); /* irqs already disabled. */
840 rnp
->qsmaskinit
&= ~mask
;
841 if (rnp
->qsmaskinit
!= 0) {
842 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
846 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
848 } while (rnp
!= NULL
);
849 lastcomp
= rsp
->completed
;
851 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
853 /* Being offline is a quiescent state, so go record it. */
854 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
857 * Move callbacks from the outgoing CPU to the running CPU.
858 * Note that the outgoing CPU is now quiscent, so it is now
859 * (uncharacteristically) safe to access its rcu_data structure.
860 * Note also that we must carefully retain the order of the
861 * outgoing CPU's callbacks in order for rcu_barrier() to work
862 * correctly. Finally, note that we start all the callbacks
863 * afresh, even those that have passed through a grace period
864 * and are therefore ready to invoke. The theory is that hotplug
865 * events are rare, and that if they are frequent enough to
866 * indefinitely delay callbacks, you have far worse things to
869 rdp_me
= rsp
->rda
[smp_processor_id()];
870 if (rdp
->nxtlist
!= NULL
) {
871 *rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxtlist
;
872 rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
874 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
875 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
876 rdp_me
->qlen
+= rdp
->qlen
;
879 local_irq_restore(flags
);
883 * Remove the specified CPU from the RCU hierarchy and move any pending
884 * callbacks that it might have to the current CPU. This code assumes
885 * that at least one CPU in the system will remain running at all times.
886 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
888 static void rcu_offline_cpu(int cpu
)
890 __rcu_offline_cpu(cpu
, &rcu_sched_state
);
891 __rcu_offline_cpu(cpu
, &rcu_bh_state
);
894 #else /* #ifdef CONFIG_HOTPLUG_CPU */
896 static void rcu_offline_cpu(int cpu
)
900 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
903 * Invoke any RCU callbacks that have made it to the end of their grace
904 * period. Thottle as specified by rdp->blimit.
906 static void rcu_do_batch(struct rcu_data
*rdp
)
909 struct rcu_head
*next
, *list
, **tail
;
912 /* If no callbacks are ready, just return.*/
913 if (!cpu_has_callbacks_ready_to_invoke(rdp
))
917 * Extract the list of ready callbacks, disabling to prevent
918 * races with call_rcu() from interrupt handlers.
920 local_irq_save(flags
);
922 rdp
->nxtlist
= *rdp
->nxttail
[RCU_DONE_TAIL
];
923 *rdp
->nxttail
[RCU_DONE_TAIL
] = NULL
;
924 tail
= rdp
->nxttail
[RCU_DONE_TAIL
];
925 for (count
= RCU_NEXT_SIZE
- 1; count
>= 0; count
--)
926 if (rdp
->nxttail
[count
] == rdp
->nxttail
[RCU_DONE_TAIL
])
927 rdp
->nxttail
[count
] = &rdp
->nxtlist
;
928 local_irq_restore(flags
);
930 /* Invoke callbacks. */
937 if (++count
>= rdp
->blimit
)
941 local_irq_save(flags
);
943 /* Update count, and requeue any remaining callbacks. */
946 *tail
= rdp
->nxtlist
;
948 for (count
= 0; count
< RCU_NEXT_SIZE
; count
++)
949 if (&rdp
->nxtlist
== rdp
->nxttail
[count
])
950 rdp
->nxttail
[count
] = tail
;
955 /* Reinstate batch limit if we have worked down the excess. */
956 if (rdp
->blimit
== LONG_MAX
&& rdp
->qlen
<= qlowmark
)
957 rdp
->blimit
= blimit
;
959 local_irq_restore(flags
);
961 /* Re-raise the RCU softirq if there are callbacks remaining. */
962 if (cpu_has_callbacks_ready_to_invoke(rdp
))
963 raise_softirq(RCU_SOFTIRQ
);
967 * Check to see if this CPU is in a non-context-switch quiescent state
968 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
969 * Also schedule the RCU softirq handler.
971 * This function must be called with hardirqs disabled. It is normally
972 * invoked from the scheduling-clock interrupt. If rcu_pending returns
973 * false, there is no point in invoking rcu_check_callbacks().
975 void rcu_check_callbacks(int cpu
, int user
)
978 (idle_cpu(cpu
) && rcu_scheduler_active
&&
979 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT
))) {
982 * Get here if this CPU took its interrupt from user
983 * mode or from the idle loop, and if this is not a
984 * nested interrupt. In this case, the CPU is in
985 * a quiescent state, so note it.
987 * No memory barrier is required here because both
988 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
989 * variables that other CPUs neither access nor modify,
990 * at least not while the corresponding CPU is online.
996 } else if (!in_softirq()) {
999 * Get here if this CPU did not take its interrupt from
1000 * softirq, in other words, if it is not interrupting
1001 * a rcu_bh read-side critical section. This is an _bh
1002 * critical section, so note it.
1007 raise_softirq(RCU_SOFTIRQ
);
1013 * Scan the leaf rcu_node structures, processing dyntick state for any that
1014 * have not yet encountered a quiescent state, using the function specified.
1015 * Returns 1 if the current grace period ends while scanning (possibly
1016 * because we made it end).
1018 static int rcu_process_dyntick(struct rcu_state
*rsp
, long lastcomp
,
1019 int (*f
)(struct rcu_data
*))
1023 unsigned long flags
;
1025 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
1026 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
1028 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
1030 spin_lock_irqsave(&rnp_cur
->lock
, flags
);
1031 if (rsp
->completed
!= lastcomp
) {
1032 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1035 if (rnp_cur
->qsmask
== 0) {
1036 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1039 cpu
= rnp_cur
->grplo
;
1041 for (; cpu
<= rnp_cur
->grphi
; cpu
++, bit
<<= 1) {
1042 if ((rnp_cur
->qsmask
& bit
) != 0 && f(rsp
->rda
[cpu
]))
1045 if (mask
!= 0 && rsp
->completed
== lastcomp
) {
1047 /* cpu_quiet_msk() releases rnp_cur->lock. */
1048 cpu_quiet_msk(mask
, rsp
, rnp_cur
, flags
);
1051 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1057 * Force quiescent states on reluctant CPUs, and also detect which
1058 * CPUs are in dyntick-idle mode.
1060 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1062 unsigned long flags
;
1064 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1067 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
))
1068 return; /* No grace period in progress, nothing to force. */
1069 if (!spin_trylock_irqsave(&rsp
->fqslock
, flags
)) {
1070 rsp
->n_force_qs_lh
++; /* Inexact, can lose counts. Tough! */
1071 return; /* Someone else is already on the job. */
1074 (long)(rsp
->jiffies_force_qs
- jiffies
) >= 0)
1075 goto unlock_ret
; /* no emergency and done recently. */
1077 spin_lock(&rnp
->lock
);
1078 lastcomp
= rsp
->completed
;
1079 signaled
= rsp
->signaled
;
1080 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
1081 if (lastcomp
== rsp
->gpnum
) {
1082 rsp
->n_force_qs_ngp
++;
1083 spin_unlock(&rnp
->lock
);
1084 goto unlock_ret
; /* no GP in progress, time updated. */
1086 spin_unlock(&rnp
->lock
);
1090 break; /* grace period still initializing, ignore. */
1092 case RCU_SAVE_DYNTICK
:
1094 if (RCU_SIGNAL_INIT
!= RCU_SAVE_DYNTICK
)
1095 break; /* So gcc recognizes the dead code. */
1097 /* Record dyntick-idle state. */
1098 if (rcu_process_dyntick(rsp
, lastcomp
,
1099 dyntick_save_progress_counter
))
1102 /* Update state, record completion counter. */
1103 spin_lock(&rnp
->lock
);
1104 if (lastcomp
== rsp
->completed
) {
1105 rsp
->signaled
= RCU_FORCE_QS
;
1106 dyntick_record_completed(rsp
, lastcomp
);
1108 spin_unlock(&rnp
->lock
);
1113 /* Check dyntick-idle state, send IPI to laggarts. */
1114 if (rcu_process_dyntick(rsp
, dyntick_recall_completed(rsp
),
1115 rcu_implicit_dynticks_qs
))
1118 /* Leave state in case more forcing is required. */
1123 spin_unlock_irqrestore(&rsp
->fqslock
, flags
);
1126 #else /* #ifdef CONFIG_SMP */
1128 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1133 #endif /* #else #ifdef CONFIG_SMP */
1136 * This does the RCU processing work from softirq context for the
1137 * specified rcu_state and rcu_data structures. This may be called
1138 * only from the CPU to whom the rdp belongs.
1141 __rcu_process_callbacks(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1143 unsigned long flags
;
1145 WARN_ON_ONCE(rdp
->beenonline
== 0);
1148 * If an RCU GP has gone long enough, go check for dyntick
1149 * idle CPUs and, if needed, send resched IPIs.
1151 if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1152 force_quiescent_state(rsp
, 1);
1155 * Advance callbacks in response to end of earlier grace
1156 * period that some other CPU ended.
1158 rcu_process_gp_end(rsp
, rdp
);
1160 /* Update RCU state based on any recent quiescent states. */
1161 rcu_check_quiescent_state(rsp
, rdp
);
1163 /* Does this CPU require a not-yet-started grace period? */
1164 if (cpu_needs_another_gp(rsp
, rdp
)) {
1165 spin_lock_irqsave(&rcu_get_root(rsp
)->lock
, flags
);
1166 rcu_start_gp(rsp
, flags
); /* releases above lock */
1169 /* If there are callbacks ready, invoke them. */
1174 * Do softirq processing for the current CPU.
1176 static void rcu_process_callbacks(struct softirq_action
*unused
)
1179 * Memory references from any prior RCU read-side critical sections
1180 * executed by the interrupted code must be seen before any RCU
1181 * grace-period manipulations below.
1183 smp_mb(); /* See above block comment. */
1185 __rcu_process_callbacks(&rcu_sched_state
,
1186 &__get_cpu_var(rcu_sched_data
));
1187 __rcu_process_callbacks(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1190 * Memory references from any later RCU read-side critical sections
1191 * executed by the interrupted code must be seen after any RCU
1192 * grace-period manipulations above.
1194 smp_mb(); /* See above block comment. */
1198 __call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
),
1199 struct rcu_state
*rsp
)
1201 unsigned long flags
;
1202 struct rcu_data
*rdp
;
1207 smp_mb(); /* Ensure RCU update seen before callback registry. */
1210 * Opportunistically note grace-period endings and beginnings.
1211 * Note that we might see a beginning right after we see an
1212 * end, but never vice versa, since this CPU has to pass through
1213 * a quiescent state betweentimes.
1215 local_irq_save(flags
);
1216 rdp
= rsp
->rda
[smp_processor_id()];
1217 rcu_process_gp_end(rsp
, rdp
);
1218 check_for_new_grace_period(rsp
, rdp
);
1220 /* Add the callback to our list. */
1221 *rdp
->nxttail
[RCU_NEXT_TAIL
] = head
;
1222 rdp
->nxttail
[RCU_NEXT_TAIL
] = &head
->next
;
1224 /* Start a new grace period if one not already started. */
1225 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
)) {
1226 unsigned long nestflag
;
1227 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
1229 spin_lock_irqsave(&rnp_root
->lock
, nestflag
);
1230 rcu_start_gp(rsp
, nestflag
); /* releases rnp_root->lock. */
1233 /* Force the grace period if too many callbacks or too long waiting. */
1234 if (unlikely(++rdp
->qlen
> qhimark
)) {
1235 rdp
->blimit
= LONG_MAX
;
1236 force_quiescent_state(rsp
, 0);
1237 } else if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1238 force_quiescent_state(rsp
, 1);
1239 local_irq_restore(flags
);
1243 * Queue an RCU-sched callback for invocation after a grace period.
1245 void call_rcu_sched(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1247 __call_rcu(head
, func
, &rcu_sched_state
);
1249 EXPORT_SYMBOL_GPL(call_rcu_sched
);
1252 * @@@ Queue an RCU callback for invocation after a grace period.
1253 * @@@ Placeholder pending rcutree_plugin.h.
1255 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1257 call_rcu_sched(head
, func
);
1259 EXPORT_SYMBOL_GPL(call_rcu
);
1263 * Queue an RCU for invocation after a quicker grace period.
1265 void call_rcu_bh(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1267 __call_rcu(head
, func
, &rcu_bh_state
);
1269 EXPORT_SYMBOL_GPL(call_rcu_bh
);
1272 * Check to see if there is any immediate RCU-related work to be done
1273 * by the current CPU, for the specified type of RCU, returning 1 if so.
1274 * The checks are in order of increasing expense: checks that can be
1275 * carried out against CPU-local state are performed first. However,
1276 * we must check for CPU stalls first, else we might not get a chance.
1278 static int __rcu_pending(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1280 rdp
->n_rcu_pending
++;
1282 /* Check for CPU stalls, if enabled. */
1283 check_cpu_stall(rsp
, rdp
);
1285 /* Is the RCU core waiting for a quiescent state from this CPU? */
1286 if (rdp
->qs_pending
) {
1287 rdp
->n_rp_qs_pending
++;
1291 /* Does this CPU have callbacks ready to invoke? */
1292 if (cpu_has_callbacks_ready_to_invoke(rdp
)) {
1293 rdp
->n_rp_cb_ready
++;
1297 /* Has RCU gone idle with this CPU needing another grace period? */
1298 if (cpu_needs_another_gp(rsp
, rdp
)) {
1299 rdp
->n_rp_cpu_needs_gp
++;
1303 /* Has another RCU grace period completed? */
1304 if (ACCESS_ONCE(rsp
->completed
) != rdp
->completed
) { /* outside lock */
1305 rdp
->n_rp_gp_completed
++;
1309 /* Has a new RCU grace period started? */
1310 if (ACCESS_ONCE(rsp
->gpnum
) != rdp
->gpnum
) { /* outside lock */
1311 rdp
->n_rp_gp_started
++;
1315 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1316 if (ACCESS_ONCE(rsp
->completed
) != ACCESS_ONCE(rsp
->gpnum
) &&
1317 ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)) {
1318 rdp
->n_rp_need_fqs
++;
1323 rdp
->n_rp_need_nothing
++;
1328 * Check to see if there is any immediate RCU-related work to be done
1329 * by the current CPU, returning 1 if so. This function is part of the
1330 * RCU implementation; it is -not- an exported member of the RCU API.
1332 int rcu_pending(int cpu
)
1334 return __rcu_pending(&rcu_sched_state
, &per_cpu(rcu_sched_data
, cpu
)) ||
1335 __rcu_pending(&rcu_bh_state
, &per_cpu(rcu_bh_data
, cpu
));
1339 * Check to see if any future RCU-related work will need to be done
1340 * by the current CPU, even if none need be done immediately, returning
1341 * 1 if so. This function is part of the RCU implementation; it is -not-
1342 * an exported member of the RCU API.
1344 int rcu_needs_cpu(int cpu
)
1346 /* RCU callbacks either ready or pending? */
1347 return per_cpu(rcu_sched_data
, cpu
).nxtlist
||
1348 per_cpu(rcu_bh_data
, cpu
).nxtlist
;
1352 * Do boot-time initialization of a CPU's per-CPU RCU data.
1355 rcu_boot_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1357 unsigned long flags
;
1359 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1360 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1362 /* Set up local state, ensuring consistent view of global state. */
1363 spin_lock_irqsave(&rnp
->lock
, flags
);
1364 rdp
->grpmask
= 1UL << (cpu
- rdp
->mynode
->grplo
);
1365 rdp
->nxtlist
= NULL
;
1366 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
1367 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
1370 rdp
->dynticks
= &per_cpu(rcu_dynticks
, cpu
);
1371 #endif /* #ifdef CONFIG_NO_HZ */
1373 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1377 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1378 * offline event can be happening at a given time. Note also that we
1379 * can accept some slop in the rsp->completed access due to the fact
1380 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1382 static void __cpuinit
1383 rcu_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1385 unsigned long flags
;
1388 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1389 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1391 /* Set up local state, ensuring consistent view of global state. */
1392 spin_lock_irqsave(&rnp
->lock
, flags
);
1393 lastcomp
= rsp
->completed
;
1394 rdp
->completed
= lastcomp
;
1395 rdp
->gpnum
= lastcomp
;
1396 rdp
->passed_quiesc
= 0; /* We could be racing with new GP, */
1397 rdp
->qs_pending
= 1; /* so set up to respond to current GP. */
1398 rdp
->beenonline
= 1; /* We have now been online. */
1399 rdp
->passed_quiesc_completed
= lastcomp
- 1;
1400 rdp
->blimit
= blimit
;
1401 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
1404 * A new grace period might start here. If so, we won't be part
1405 * of it, but that is OK, as we are currently in a quiescent state.
1408 /* Exclude any attempts to start a new GP on large systems. */
1409 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
1411 /* Add CPU to rcu_node bitmasks. */
1413 mask
= rdp
->grpmask
;
1415 /* Exclude any attempts to start a new GP on small systems. */
1416 spin_lock(&rnp
->lock
); /* irqs already disabled. */
1417 rnp
->qsmaskinit
|= mask
;
1418 mask
= rnp
->grpmask
;
1419 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
1421 } while (rnp
!= NULL
&& !(rnp
->qsmaskinit
& mask
));
1423 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
1426 * A new grace period might start here. If so, we will be part of
1427 * it, and its gpnum will be greater than ours, so we will
1428 * participate. It is also possible for the gpnum to have been
1429 * incremented before this function was called, and the bitmasks
1430 * to not be filled out until now, in which case we will also
1431 * participate due to our gpnum being behind.
1434 /* Since it is coming online, the CPU is in a quiescent state. */
1435 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
1436 local_irq_restore(flags
);
1439 static void __cpuinit
rcu_online_cpu(int cpu
)
1441 rcu_init_percpu_data(cpu
, &rcu_sched_state
);
1442 rcu_init_percpu_data(cpu
, &rcu_bh_state
);
1446 * Handle CPU online/offline notifcation events.
1448 int __cpuinit
rcu_cpu_notify(struct notifier_block
*self
,
1449 unsigned long action
, void *hcpu
)
1451 long cpu
= (long)hcpu
;
1454 case CPU_UP_PREPARE
:
1455 case CPU_UP_PREPARE_FROZEN
:
1456 rcu_online_cpu(cpu
);
1459 case CPU_DEAD_FROZEN
:
1460 case CPU_UP_CANCELED
:
1461 case CPU_UP_CANCELED_FROZEN
:
1462 rcu_offline_cpu(cpu
);
1471 * Compute the per-level fanout, either using the exact fanout specified
1472 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1474 #ifdef CONFIG_RCU_FANOUT_EXACT
1475 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1479 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--)
1480 rsp
->levelspread
[i
] = CONFIG_RCU_FANOUT
;
1482 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1483 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1490 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1491 ccur
= rsp
->levelcnt
[i
];
1492 rsp
->levelspread
[i
] = (cprv
+ ccur
- 1) / ccur
;
1496 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1499 * Helper function for rcu_init() that initializes one rcu_state structure.
1501 static void __init
rcu_init_one(struct rcu_state
*rsp
)
1506 struct rcu_node
*rnp
;
1508 /* Initialize the level-tracking arrays. */
1510 for (i
= 1; i
< NUM_RCU_LVLS
; i
++)
1511 rsp
->level
[i
] = rsp
->level
[i
- 1] + rsp
->levelcnt
[i
- 1];
1512 rcu_init_levelspread(rsp
);
1514 /* Initialize the elements themselves, starting from the leaves. */
1516 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1517 cpustride
*= rsp
->levelspread
[i
];
1518 rnp
= rsp
->level
[i
];
1519 for (j
= 0; j
< rsp
->levelcnt
[i
]; j
++, rnp
++) {
1520 spin_lock_init(&rnp
->lock
);
1522 rnp
->qsmaskinit
= 0;
1523 rnp
->grplo
= j
* cpustride
;
1524 rnp
->grphi
= (j
+ 1) * cpustride
- 1;
1525 if (rnp
->grphi
>= NR_CPUS
)
1526 rnp
->grphi
= NR_CPUS
- 1;
1532 rnp
->grpnum
= j
% rsp
->levelspread
[i
- 1];
1533 rnp
->grpmask
= 1UL << rnp
->grpnum
;
1534 rnp
->parent
= rsp
->level
[i
- 1] +
1535 j
/ rsp
->levelspread
[i
- 1];
1543 * Helper macro for __rcu_init(). To be used nowhere else!
1544 * Assigns leaf node pointers into each CPU's rcu_data structure.
1546 #define RCU_DATA_PTR_INIT(rsp, rcu_data) \
1548 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1550 for_each_possible_cpu(i) { \
1551 if (i > rnp[j].grphi) \
1553 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1554 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1558 void __init
__rcu_init(void)
1560 int i
; /* All used by RCU_DATA_PTR_INIT(). */
1562 struct rcu_node
*rnp
;
1564 printk(KERN_INFO
"Hierarchical RCU implementation.\n");
1565 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1566 printk(KERN_INFO
"RCU-based detection of stalled CPUs is enabled.\n");
1567 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1568 rcu_init_one(&rcu_sched_state
);
1569 RCU_DATA_PTR_INIT(&rcu_sched_state
, rcu_sched_data
);
1570 for_each_possible_cpu(i
)
1571 rcu_boot_init_percpu_data(i
, &rcu_sched_state
);
1572 rcu_init_one(&rcu_bh_state
);
1573 RCU_DATA_PTR_INIT(&rcu_bh_state
, rcu_bh_data
);
1574 for_each_possible_cpu(i
)
1575 rcu_boot_init_percpu_data(i
, &rcu_bh_state
);
1576 open_softirq(RCU_SOFTIRQ
, rcu_process_callbacks
);
1579 module_param(blimit
, int, 0);
1580 module_param(qhimark
, int, 0);
1581 module_param(qlowmark
, int, 0);