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>
50 #ifdef CONFIG_DEBUG_LOCK_ALLOC
51 static struct lock_class_key rcu_lock_key
;
52 struct lockdep_map rcu_lock_map
=
53 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key
);
54 EXPORT_SYMBOL_GPL(rcu_lock_map
);
57 /* Data structures. */
59 #define RCU_STATE_INITIALIZER(name) { \
60 .level = { &name.node[0] }, \
62 NUM_RCU_LVL_0, /* root of hierarchy. */ \
65 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
67 .signaled = RCU_SIGNAL_INIT, \
70 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
71 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
73 .n_force_qs_ngp = 0, \
76 struct rcu_state rcu_state
= RCU_STATE_INITIALIZER(rcu_state
);
77 DEFINE_PER_CPU(struct rcu_data
, rcu_data
);
79 struct rcu_state rcu_bh_state
= RCU_STATE_INITIALIZER(rcu_bh_state
);
80 DEFINE_PER_CPU(struct rcu_data
, rcu_bh_data
);
83 * Increment the quiescent state counter.
84 * The counter is a bit degenerated: 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. Thus just a flag.
88 void rcu_qsctr_inc(int cpu
)
90 struct rcu_data
*rdp
= &per_cpu(rcu_data
, cpu
);
91 rdp
->passed_quiesc
= 1;
92 rdp
->passed_quiesc_completed
= rdp
->completed
;
95 void rcu_bh_qsctr_inc(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 batches processed thus far for debug & stats.
118 long rcu_batches_completed(void)
120 return rcu_state
.completed
;
122 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
125 * Return the number of RCU BH batches processed thus far for debug & stats.
127 long rcu_batches_completed_bh(void)
129 return rcu_bh_state
.completed
;
131 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh
);
134 * Does the CPU have callbacks ready to be invoked?
137 cpu_has_callbacks_ready_to_invoke(struct rcu_data
*rdp
)
139 return &rdp
->nxtlist
!= rdp
->nxttail
[RCU_DONE_TAIL
];
143 * Does the current CPU require a yet-as-unscheduled grace period?
146 cpu_needs_another_gp(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
148 /* ACCESS_ONCE() because we are accessing outside of lock. */
149 return *rdp
->nxttail
[RCU_DONE_TAIL
] &&
150 ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
);
154 * Return the root node of the specified rcu_state structure.
156 static struct rcu_node
*rcu_get_root(struct rcu_state
*rsp
)
158 return &rsp
->node
[0];
164 * If the specified CPU is offline, tell the caller that it is in
165 * a quiescent state. Otherwise, whack it with a reschedule IPI.
166 * Grace periods can end up waiting on an offline CPU when that
167 * CPU is in the process of coming online -- it will be added to the
168 * rcu_node bitmasks before it actually makes it online. The same thing
169 * can happen while a CPU is in the process of coming online. Because this
170 * race is quite rare, we check for it after detecting that the grace
171 * period has been delayed rather than checking each and every CPU
172 * each and every time we start a new grace period.
174 static int rcu_implicit_offline_qs(struct rcu_data
*rdp
)
177 * If the CPU is offline, it is in a quiescent state. We can
178 * trust its state not to change because interrupts are disabled.
180 if (cpu_is_offline(rdp
->cpu
)) {
185 /* The CPU is online, so send it a reschedule IPI. */
186 if (rdp
->cpu
!= smp_processor_id())
187 smp_send_reschedule(rdp
->cpu
);
194 #endif /* #ifdef CONFIG_SMP */
197 static DEFINE_RATELIMIT_STATE(rcu_rs
, 10 * HZ
, 5);
200 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
202 * Enter nohz mode, in other words, -leave- the mode in which RCU
203 * read-side critical sections can occur. (Though RCU read-side
204 * critical sections can occur in irq handlers in nohz mode, a possibility
205 * handled by rcu_irq_enter() and rcu_irq_exit()).
207 void rcu_enter_nohz(void)
210 struct rcu_dynticks
*rdtp
;
212 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
213 local_irq_save(flags
);
214 rdtp
= &__get_cpu_var(rcu_dynticks
);
216 rdtp
->dynticks_nesting
--;
217 WARN_ON_RATELIMIT(rdtp
->dynticks
& 0x1, &rcu_rs
);
218 local_irq_restore(flags
);
222 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
224 * Exit nohz mode, in other words, -enter- the mode in which RCU
225 * read-side critical sections normally occur.
227 void rcu_exit_nohz(void)
230 struct rcu_dynticks
*rdtp
;
232 local_irq_save(flags
);
233 rdtp
= &__get_cpu_var(rcu_dynticks
);
235 rdtp
->dynticks_nesting
++;
236 WARN_ON_RATELIMIT(!(rdtp
->dynticks
& 0x1), &rcu_rs
);
237 local_irq_restore(flags
);
238 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
242 * rcu_nmi_enter - inform RCU of entry to NMI context
244 * If the CPU was idle with dynamic ticks active, and there is no
245 * irq handler running, this updates rdtp->dynticks_nmi to let the
246 * RCU grace-period handling know that the CPU is active.
248 void rcu_nmi_enter(void)
250 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
252 if (rdtp
->dynticks
& 0x1)
254 rdtp
->dynticks_nmi
++;
255 WARN_ON_RATELIMIT(!(rdtp
->dynticks_nmi
& 0x1), &rcu_rs
);
256 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
260 * rcu_nmi_exit - inform RCU of exit from NMI context
262 * If the CPU was idle with dynamic ticks active, and there is no
263 * irq handler running, this updates rdtp->dynticks_nmi to let the
264 * RCU grace-period handling know that the CPU is no longer active.
266 void rcu_nmi_exit(void)
268 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
270 if (rdtp
->dynticks
& 0x1)
272 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
273 rdtp
->dynticks_nmi
++;
274 WARN_ON_RATELIMIT(rdtp
->dynticks_nmi
& 0x1, &rcu_rs
);
278 * rcu_irq_enter - inform RCU of entry to hard irq context
280 * If the CPU was idle with dynamic ticks active, this updates the
281 * rdtp->dynticks to let the RCU handling know that the CPU is active.
283 void rcu_irq_enter(void)
285 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
287 if (rdtp
->dynticks_nesting
++)
290 WARN_ON_RATELIMIT(!(rdtp
->dynticks
& 0x1), &rcu_rs
);
291 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
295 * rcu_irq_exit - inform RCU of exit from hard irq context
297 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
298 * to put let the RCU handling be aware that the CPU is going back to idle
301 void rcu_irq_exit(void)
303 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
305 if (--rdtp
->dynticks_nesting
)
307 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
309 WARN_ON_RATELIMIT(rdtp
->dynticks
& 0x1, &rcu_rs
);
311 /* If the interrupt queued a callback, get out of dyntick mode. */
312 if (__get_cpu_var(rcu_data
).nxtlist
||
313 __get_cpu_var(rcu_bh_data
).nxtlist
)
318 * Record the specified "completed" value, which is later used to validate
319 * dynticks counter manipulations. Specify "rsp->completed - 1" to
320 * unconditionally invalidate any future dynticks manipulations (which is
321 * useful at the beginning of a grace period).
323 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
325 rsp
->dynticks_completed
= comp
;
331 * Recall the previously recorded value of the completion for dynticks.
333 static long dyntick_recall_completed(struct rcu_state
*rsp
)
335 return rsp
->dynticks_completed
;
339 * Snapshot the specified CPU's dynticks counter so that we can later
340 * credit them with an implicit quiescent state. Return 1 if this CPU
341 * is already in a quiescent state courtesy of dynticks idle mode.
343 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
349 snap
= rdp
->dynticks
->dynticks
;
350 snap_nmi
= rdp
->dynticks
->dynticks_nmi
;
351 smp_mb(); /* Order sampling of snap with end of grace period. */
352 rdp
->dynticks_snap
= snap
;
353 rdp
->dynticks_nmi_snap
= snap_nmi
;
354 ret
= ((snap
& 0x1) == 0) && ((snap_nmi
& 0x1) == 0);
361 * Return true if the specified CPU has passed through a quiescent
362 * state by virtue of being in or having passed through an dynticks
363 * idle state since the last call to dyntick_save_progress_counter()
366 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
373 curr
= rdp
->dynticks
->dynticks
;
374 snap
= rdp
->dynticks_snap
;
375 curr_nmi
= rdp
->dynticks
->dynticks_nmi
;
376 snap_nmi
= rdp
->dynticks_nmi_snap
;
377 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
380 * If the CPU passed through or entered a dynticks idle phase with
381 * no active irq/NMI handlers, then we can safely pretend that the CPU
382 * already acknowledged the request to pass through a quiescent
383 * state. Either way, that CPU cannot possibly be in an RCU
384 * read-side critical section that started before the beginning
385 * of the current RCU grace period.
387 if ((curr
!= snap
|| (curr
& 0x1) == 0) &&
388 (curr_nmi
!= snap_nmi
|| (curr_nmi
& 0x1) == 0)) {
393 /* Go check for the CPU being offline. */
394 return rcu_implicit_offline_qs(rdp
);
397 #endif /* #ifdef CONFIG_SMP */
399 #else /* #ifdef CONFIG_NO_HZ */
401 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
408 * If there are no dynticks, then the only way that a CPU can passively
409 * be in a quiescent state is to be offline. Unlike dynticks idle, which
410 * is a point in time during the prior (already finished) grace period,
411 * an offline CPU is always in a quiescent state, and thus can be
412 * unconditionally applied. So just return the current value of completed.
414 static long dyntick_recall_completed(struct rcu_state
*rsp
)
416 return rsp
->completed
;
419 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
424 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
426 return rcu_implicit_offline_qs(rdp
);
429 #endif /* #ifdef CONFIG_SMP */
431 #endif /* #else #ifdef CONFIG_NO_HZ */
433 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
435 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
437 rsp
->gp_start
= jiffies
;
438 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_CHECK
;
441 static void print_other_cpu_stall(struct rcu_state
*rsp
)
446 struct rcu_node
*rnp
= rcu_get_root(rsp
);
447 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
448 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
450 /* Only let one CPU complain about others per time interval. */
452 spin_lock_irqsave(&rnp
->lock
, flags
);
453 delta
= jiffies
- rsp
->jiffies_stall
;
454 if (delta
< RCU_STALL_RAT_DELAY
|| rsp
->gpnum
== rsp
->completed
) {
455 spin_unlock_irqrestore(&rnp
->lock
, flags
);
458 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
459 spin_unlock_irqrestore(&rnp
->lock
, flags
);
461 /* OK, time to rat on our buddy... */
463 printk(KERN_ERR
"INFO: RCU detected CPU stalls:");
464 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
465 if (rnp_cur
->qsmask
== 0)
467 for (cpu
= 0; cpu
<= rnp_cur
->grphi
- rnp_cur
->grplo
; cpu
++)
468 if (rnp_cur
->qsmask
& (1UL << cpu
))
469 printk(" %d", rnp_cur
->grplo
+ cpu
);
471 printk(" (detected by %d, t=%ld jiffies)\n",
472 smp_processor_id(), (long)(jiffies
- rsp
->gp_start
));
473 trigger_all_cpu_backtrace();
475 force_quiescent_state(rsp
, 0); /* Kick them all. */
478 static void print_cpu_stall(struct rcu_state
*rsp
)
481 struct rcu_node
*rnp
= rcu_get_root(rsp
);
483 printk(KERN_ERR
"INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
484 smp_processor_id(), jiffies
- rsp
->gp_start
);
485 trigger_all_cpu_backtrace();
487 spin_lock_irqsave(&rnp
->lock
, flags
);
488 if ((long)(jiffies
- rsp
->jiffies_stall
) >= 0)
490 jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
491 spin_unlock_irqrestore(&rnp
->lock
, flags
);
493 set_need_resched(); /* kick ourselves to get things going. */
496 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
499 struct rcu_node
*rnp
;
501 delta
= jiffies
- rsp
->jiffies_stall
;
503 if ((rnp
->qsmask
& rdp
->grpmask
) && delta
>= 0) {
505 /* We haven't checked in, so go dump stack. */
506 print_cpu_stall(rsp
);
508 } else if (rsp
->gpnum
!= rsp
->completed
&&
509 delta
>= RCU_STALL_RAT_DELAY
) {
511 /* They had two time units to dump stack, so complain. */
512 print_other_cpu_stall(rsp
);
516 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
518 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
522 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
526 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
529 * Update CPU-local rcu_data state to record the newly noticed grace period.
530 * This is used both when we started the grace period and when we notice
531 * that someone else started the grace period.
533 static void note_new_gpnum(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
536 rdp
->passed_quiesc
= 0;
537 rdp
->gpnum
= rsp
->gpnum
;
541 * Did someone else start a new RCU grace period start since we last
542 * checked? Update local state appropriately if so. Must be called
543 * on the CPU corresponding to rdp.
546 check_for_new_grace_period(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
551 local_irq_save(flags
);
552 if (rdp
->gpnum
!= rsp
->gpnum
) {
553 note_new_gpnum(rsp
, rdp
);
556 local_irq_restore(flags
);
561 * Start a new RCU grace period if warranted, re-initializing the hierarchy
562 * in preparation for detecting the next grace period. The caller must hold
563 * the root node's ->lock, which is released before return. Hard irqs must
567 rcu_start_gp(struct rcu_state
*rsp
, unsigned long flags
)
568 __releases(rcu_get_root(rsp
)->lock
)
570 struct rcu_data
*rdp
= rsp
->rda
[smp_processor_id()];
571 struct rcu_node
*rnp
= rcu_get_root(rsp
);
572 struct rcu_node
*rnp_cur
;
573 struct rcu_node
*rnp_end
;
575 if (!cpu_needs_another_gp(rsp
, rdp
)) {
576 spin_unlock_irqrestore(&rnp
->lock
, flags
);
580 /* Advance to a new grace period and initialize state. */
582 rsp
->signaled
= RCU_GP_INIT
; /* Hold off force_quiescent_state. */
583 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
584 record_gp_stall_check_time(rsp
);
585 dyntick_record_completed(rsp
, rsp
->completed
- 1);
586 note_new_gpnum(rsp
, rdp
);
589 * Because we are first, we know that all our callbacks will
590 * be covered by this upcoming grace period, even the ones
591 * that were registered arbitrarily recently.
593 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
594 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
596 /* Special-case the common single-level case. */
597 if (NUM_RCU_NODES
== 1) {
598 rnp
->qsmask
= rnp
->qsmaskinit
;
599 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state OK. */
600 spin_unlock_irqrestore(&rnp
->lock
, flags
);
604 spin_unlock(&rnp
->lock
); /* leave irqs disabled. */
607 /* Exclude any concurrent CPU-hotplug operations. */
608 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
611 * Set the quiescent-state-needed bits in all the non-leaf RCU
612 * nodes for all currently online CPUs. This operation relies
613 * on the layout of the hierarchy within the rsp->node[] array.
614 * Note that other CPUs will access only the leaves of the
615 * hierarchy, which still indicate that no grace period is in
616 * progress. In addition, we have excluded CPU-hotplug operations.
618 * We therefore do not need to hold any locks. Any required
619 * memory barriers will be supplied by the locks guarding the
620 * leaf rcu_nodes in the hierarchy.
623 rnp_end
= rsp
->level
[NUM_RCU_LVLS
- 1];
624 for (rnp_cur
= &rsp
->node
[0]; rnp_cur
< rnp_end
; rnp_cur
++)
625 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
628 * Now set up the leaf nodes. Here we must be careful. First,
629 * we need to hold the lock in order to exclude other CPUs, which
630 * might be contending for the leaf nodes' locks. Second, as
631 * soon as we initialize a given leaf node, its CPUs might run
632 * up the rest of the hierarchy. We must therefore acquire locks
633 * for each node that we touch during this stage. (But we still
634 * are excluding CPU-hotplug operations.)
636 * Note that the grace period cannot complete until we finish
637 * the initialization process, as there will be at least one
638 * qsmask bit set in the root node until that time, namely the
639 * one corresponding to this CPU.
641 rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
642 rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
643 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
644 spin_lock(&rnp_cur
->lock
); /* irqs already disabled. */
645 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
646 spin_unlock(&rnp_cur
->lock
); /* irqs already disabled. */
649 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state now OK. */
650 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
654 * Advance this CPU's callbacks, but only if the current grace period
655 * has ended. This may be called only from the CPU to whom the rdp
659 rcu_process_gp_end(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
664 local_irq_save(flags
);
665 completed_snap
= ACCESS_ONCE(rsp
->completed
); /* outside of lock. */
667 /* Did another grace period end? */
668 if (rdp
->completed
!= completed_snap
) {
670 /* Advance callbacks. No harm if list empty. */
671 rdp
->nxttail
[RCU_DONE_TAIL
] = rdp
->nxttail
[RCU_WAIT_TAIL
];
672 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_READY_TAIL
];
673 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
675 /* Remember that we saw this grace-period completion. */
676 rdp
->completed
= completed_snap
;
678 local_irq_restore(flags
);
682 * Similar to cpu_quiet(), for which it is a helper function. Allows
683 * a group of CPUs to be quieted at one go, though all the CPUs in the
684 * group must be represented by the same leaf rcu_node structure.
685 * That structure's lock must be held upon entry, and it is released
689 cpu_quiet_msk(unsigned long mask
, struct rcu_state
*rsp
, struct rcu_node
*rnp
,
691 __releases(rnp
->lock
)
693 /* Walk up the rcu_node hierarchy. */
695 if (!(rnp
->qsmask
& mask
)) {
697 /* Our bit has already been cleared, so done. */
698 spin_unlock_irqrestore(&rnp
->lock
, flags
);
701 rnp
->qsmask
&= ~mask
;
702 if (rnp
->qsmask
!= 0) {
704 /* Other bits still set at this level, so done. */
705 spin_unlock_irqrestore(&rnp
->lock
, flags
);
709 if (rnp
->parent
== NULL
) {
711 /* No more levels. Exit loop holding root lock. */
715 spin_unlock_irqrestore(&rnp
->lock
, flags
);
717 spin_lock_irqsave(&rnp
->lock
, flags
);
721 * Get here if we are the last CPU to pass through a quiescent
722 * state for this grace period. Clean up and let rcu_start_gp()
723 * start up the next grace period if one is needed. Note that
724 * we still hold rnp->lock, as required by rcu_start_gp(), which
727 rsp
->completed
= rsp
->gpnum
;
728 rcu_process_gp_end(rsp
, rsp
->rda
[smp_processor_id()]);
729 rcu_start_gp(rsp
, flags
); /* releases rnp->lock. */
733 * Record a quiescent state for the specified CPU, which must either be
734 * the current CPU or an offline CPU. The lastcomp argument is used to
735 * make sure we are still in the grace period of interest. We don't want
736 * to end the current grace period based on quiescent states detected in
737 * an earlier grace period!
740 cpu_quiet(int cpu
, struct rcu_state
*rsp
, struct rcu_data
*rdp
, long lastcomp
)
744 struct rcu_node
*rnp
;
747 spin_lock_irqsave(&rnp
->lock
, flags
);
748 if (lastcomp
!= ACCESS_ONCE(rsp
->completed
)) {
751 * Someone beat us to it for this grace period, so leave.
752 * The race with GP start is resolved by the fact that we
753 * hold the leaf rcu_node lock, so that the per-CPU bits
754 * cannot yet be initialized -- so we would simply find our
755 * CPU's bit already cleared in cpu_quiet_msk() if this race
758 rdp
->passed_quiesc
= 0; /* try again later! */
759 spin_unlock_irqrestore(&rnp
->lock
, flags
);
763 if ((rnp
->qsmask
& mask
) == 0) {
764 spin_unlock_irqrestore(&rnp
->lock
, flags
);
769 * This GP can't end until cpu checks in, so all of our
770 * callbacks can be processed during the next GP.
772 rdp
= rsp
->rda
[smp_processor_id()];
773 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
775 cpu_quiet_msk(mask
, rsp
, rnp
, flags
); /* releases rnp->lock */
780 * Check to see if there is a new grace period of which this CPU
781 * is not yet aware, and if so, set up local rcu_data state for it.
782 * Otherwise, see if this CPU has just passed through its first
783 * quiescent state for this grace period, and record that fact if so.
786 rcu_check_quiescent_state(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
788 /* If there is now a new grace period, record and return. */
789 if (check_for_new_grace_period(rsp
, rdp
))
793 * Does this CPU still need to do its part for current grace period?
794 * If no, return and let the other CPUs do their part as well.
796 if (!rdp
->qs_pending
)
800 * Was there a quiescent state since the beginning of the grace
801 * period? If no, then exit and wait for the next call.
803 if (!rdp
->passed_quiesc
)
806 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
807 cpu_quiet(rdp
->cpu
, rsp
, rdp
, rdp
->passed_quiesc_completed
);
810 #ifdef CONFIG_HOTPLUG_CPU
813 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
814 * and move all callbacks from the outgoing CPU to the current one.
816 static void __rcu_offline_cpu(int cpu
, struct rcu_state
*rsp
)
822 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
823 struct rcu_data
*rdp_me
;
824 struct rcu_node
*rnp
;
826 /* Exclude any attempts to start a new grace period. */
827 spin_lock_irqsave(&rsp
->onofflock
, flags
);
829 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
831 mask
= rdp
->grpmask
; /* rnp->grplo is constant. */
833 spin_lock(&rnp
->lock
); /* irqs already disabled. */
834 rnp
->qsmaskinit
&= ~mask
;
835 if (rnp
->qsmaskinit
!= 0) {
836 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
840 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
842 } while (rnp
!= NULL
);
843 lastcomp
= rsp
->completed
;
845 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
847 /* Being offline is a quiescent state, so go record it. */
848 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
851 * Move callbacks from the outgoing CPU to the running CPU.
852 * Note that the outgoing CPU is now quiscent, so it is now
853 * (uncharacteristically) safe to access it rcu_data structure.
854 * Note also that we must carefully retain the order of the
855 * outgoing CPU's callbacks in order for rcu_barrier() to work
856 * correctly. Finally, note that we start all the callbacks
857 * afresh, even those that have passed through a grace period
858 * and are therefore ready to invoke. The theory is that hotplug
859 * events are rare, and that if they are frequent enough to
860 * indefinitely delay callbacks, you have far worse things to
863 rdp_me
= rsp
->rda
[smp_processor_id()];
864 if (rdp
->nxtlist
!= NULL
) {
865 *rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxtlist
;
866 rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
868 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
869 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
870 rdp_me
->qlen
+= rdp
->qlen
;
873 local_irq_restore(flags
);
877 * Remove the specified CPU from the RCU hierarchy and move any pending
878 * callbacks that it might have to the current CPU. This code assumes
879 * that at least one CPU in the system will remain running at all times.
880 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
882 static void rcu_offline_cpu(int cpu
)
884 __rcu_offline_cpu(cpu
, &rcu_state
);
885 __rcu_offline_cpu(cpu
, &rcu_bh_state
);
888 #else /* #ifdef CONFIG_HOTPLUG_CPU */
890 static void rcu_offline_cpu(int cpu
)
894 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
897 * Invoke any RCU callbacks that have made it to the end of their grace
898 * period. Thottle as specified by rdp->blimit.
900 static void rcu_do_batch(struct rcu_data
*rdp
)
903 struct rcu_head
*next
, *list
, **tail
;
906 /* If no callbacks are ready, just return.*/
907 if (!cpu_has_callbacks_ready_to_invoke(rdp
))
911 * Extract the list of ready callbacks, disabling to prevent
912 * races with call_rcu() from interrupt handlers.
914 local_irq_save(flags
);
916 rdp
->nxtlist
= *rdp
->nxttail
[RCU_DONE_TAIL
];
917 *rdp
->nxttail
[RCU_DONE_TAIL
] = NULL
;
918 tail
= rdp
->nxttail
[RCU_DONE_TAIL
];
919 for (count
= RCU_NEXT_SIZE
- 1; count
>= 0; count
--)
920 if (rdp
->nxttail
[count
] == rdp
->nxttail
[RCU_DONE_TAIL
])
921 rdp
->nxttail
[count
] = &rdp
->nxtlist
;
922 local_irq_restore(flags
);
924 /* Invoke callbacks. */
931 if (++count
>= rdp
->blimit
)
935 local_irq_save(flags
);
937 /* Update count, and requeue any remaining callbacks. */
940 *tail
= rdp
->nxtlist
;
942 for (count
= 0; count
< RCU_NEXT_SIZE
; count
++)
943 if (&rdp
->nxtlist
== rdp
->nxttail
[count
])
944 rdp
->nxttail
[count
] = tail
;
949 /* Reinstate batch limit if we have worked down the excess. */
950 if (rdp
->blimit
== LONG_MAX
&& rdp
->qlen
<= qlowmark
)
951 rdp
->blimit
= blimit
;
953 local_irq_restore(flags
);
955 /* Re-raise the RCU softirq if there are callbacks remaining. */
956 if (cpu_has_callbacks_ready_to_invoke(rdp
))
957 raise_softirq(RCU_SOFTIRQ
);
961 * Check to see if this CPU is in a non-context-switch quiescent state
962 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
963 * Also schedule the RCU softirq handler.
965 * This function must be called with hardirqs disabled. It is normally
966 * invoked from the scheduling-clock interrupt. If rcu_pending returns
967 * false, there is no point in invoking rcu_check_callbacks().
969 void rcu_check_callbacks(int cpu
, int user
)
972 (idle_cpu(cpu
) && rcu_scheduler_active
&&
973 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT
))) {
976 * Get here if this CPU took its interrupt from user
977 * mode or from the idle loop, and if this is not a
978 * nested interrupt. In this case, the CPU is in
979 * a quiescent state, so count it.
981 * No memory barrier is required here because both
982 * rcu_qsctr_inc() and rcu_bh_qsctr_inc() reference
983 * only CPU-local variables that other CPUs neither
984 * access nor modify, at least not while the corresponding
989 rcu_bh_qsctr_inc(cpu
);
991 } else if (!in_softirq()) {
994 * Get here if this CPU did not take its interrupt from
995 * softirq, in other words, if it is not interrupting
996 * a rcu_bh read-side critical section. This is an _bh
997 * critical section, so count it.
1000 rcu_bh_qsctr_inc(cpu
);
1002 raise_softirq(RCU_SOFTIRQ
);
1008 * Scan the leaf rcu_node structures, processing dyntick state for any that
1009 * have not yet encountered a quiescent state, using the function specified.
1010 * Returns 1 if the current grace period ends while scanning (possibly
1011 * because we made it end).
1013 static int rcu_process_dyntick(struct rcu_state
*rsp
, long lastcomp
,
1014 int (*f
)(struct rcu_data
*))
1018 unsigned long flags
;
1020 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
1021 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
1023 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
1025 spin_lock_irqsave(&rnp_cur
->lock
, flags
);
1026 if (rsp
->completed
!= lastcomp
) {
1027 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1030 if (rnp_cur
->qsmask
== 0) {
1031 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1034 cpu
= rnp_cur
->grplo
;
1036 for (; cpu
<= rnp_cur
->grphi
; cpu
++, bit
<<= 1) {
1037 if ((rnp_cur
->qsmask
& bit
) != 0 && f(rsp
->rda
[cpu
]))
1040 if (mask
!= 0 && rsp
->completed
== lastcomp
) {
1042 /* cpu_quiet_msk() releases rnp_cur->lock. */
1043 cpu_quiet_msk(mask
, rsp
, rnp_cur
, flags
);
1046 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1052 * Force quiescent states on reluctant CPUs, and also detect which
1053 * CPUs are in dyntick-idle mode.
1055 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1057 unsigned long flags
;
1059 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1062 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
))
1063 return; /* No grace period in progress, nothing to force. */
1064 if (!spin_trylock_irqsave(&rsp
->fqslock
, flags
)) {
1065 rsp
->n_force_qs_lh
++; /* Inexact, can lose counts. Tough! */
1066 return; /* Someone else is already on the job. */
1069 (long)(rsp
->jiffies_force_qs
- jiffies
) >= 0)
1070 goto unlock_ret
; /* no emergency and done recently. */
1072 spin_lock(&rnp
->lock
);
1073 lastcomp
= rsp
->completed
;
1074 signaled
= rsp
->signaled
;
1075 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
1076 if (lastcomp
== rsp
->gpnum
) {
1077 rsp
->n_force_qs_ngp
++;
1078 spin_unlock(&rnp
->lock
);
1079 goto unlock_ret
; /* no GP in progress, time updated. */
1081 spin_unlock(&rnp
->lock
);
1085 break; /* grace period still initializing, ignore. */
1087 case RCU_SAVE_DYNTICK
:
1089 if (RCU_SIGNAL_INIT
!= RCU_SAVE_DYNTICK
)
1090 break; /* So gcc recognizes the dead code. */
1092 /* Record dyntick-idle state. */
1093 if (rcu_process_dyntick(rsp
, lastcomp
,
1094 dyntick_save_progress_counter
))
1097 /* Update state, record completion counter. */
1098 spin_lock(&rnp
->lock
);
1099 if (lastcomp
== rsp
->completed
) {
1100 rsp
->signaled
= RCU_FORCE_QS
;
1101 dyntick_record_completed(rsp
, lastcomp
);
1103 spin_unlock(&rnp
->lock
);
1108 /* Check dyntick-idle state, send IPI to laggarts. */
1109 if (rcu_process_dyntick(rsp
, dyntick_recall_completed(rsp
),
1110 rcu_implicit_dynticks_qs
))
1113 /* Leave state in case more forcing is required. */
1118 spin_unlock_irqrestore(&rsp
->fqslock
, flags
);
1121 #else /* #ifdef CONFIG_SMP */
1123 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1128 #endif /* #else #ifdef CONFIG_SMP */
1131 * This does the RCU processing work from softirq context for the
1132 * specified rcu_state and rcu_data structures. This may be called
1133 * only from the CPU to whom the rdp belongs.
1136 __rcu_process_callbacks(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1138 unsigned long flags
;
1141 * If an RCU GP has gone long enough, go check for dyntick
1142 * idle CPUs and, if needed, send resched IPIs.
1144 if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1145 force_quiescent_state(rsp
, 1);
1148 * Advance callbacks in response to end of earlier grace
1149 * period that some other CPU ended.
1151 rcu_process_gp_end(rsp
, rdp
);
1153 /* Update RCU state based on any recent quiescent states. */
1154 rcu_check_quiescent_state(rsp
, rdp
);
1156 /* Does this CPU require a not-yet-started grace period? */
1157 if (cpu_needs_another_gp(rsp
, rdp
)) {
1158 spin_lock_irqsave(&rcu_get_root(rsp
)->lock
, flags
);
1159 rcu_start_gp(rsp
, flags
); /* releases above lock */
1162 /* If there are callbacks ready, invoke them. */
1167 * Do softirq processing for the current CPU.
1169 static void rcu_process_callbacks(struct softirq_action
*unused
)
1172 * Memory references from any prior RCU read-side critical sections
1173 * executed by the interrupted code must be seen before any RCU
1174 * grace-period manipulations below.
1176 smp_mb(); /* See above block comment. */
1178 __rcu_process_callbacks(&rcu_state
, &__get_cpu_var(rcu_data
));
1179 __rcu_process_callbacks(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1182 * Memory references from any later RCU read-side critical sections
1183 * executed by the interrupted code must be seen after any RCU
1184 * grace-period manipulations above.
1186 smp_mb(); /* See above block comment. */
1190 __call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
),
1191 struct rcu_state
*rsp
)
1193 unsigned long flags
;
1194 struct rcu_data
*rdp
;
1199 smp_mb(); /* Ensure RCU update seen before callback registry. */
1202 * Opportunistically note grace-period endings and beginnings.
1203 * Note that we might see a beginning right after we see an
1204 * end, but never vice versa, since this CPU has to pass through
1205 * a quiescent state betweentimes.
1207 local_irq_save(flags
);
1208 rdp
= rsp
->rda
[smp_processor_id()];
1209 rcu_process_gp_end(rsp
, rdp
);
1210 check_for_new_grace_period(rsp
, rdp
);
1212 /* Add the callback to our list. */
1213 *rdp
->nxttail
[RCU_NEXT_TAIL
] = head
;
1214 rdp
->nxttail
[RCU_NEXT_TAIL
] = &head
->next
;
1216 /* Start a new grace period if one not already started. */
1217 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
)) {
1218 unsigned long nestflag
;
1219 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
1221 spin_lock_irqsave(&rnp_root
->lock
, nestflag
);
1222 rcu_start_gp(rsp
, nestflag
); /* releases rnp_root->lock. */
1225 /* Force the grace period if too many callbacks or too long waiting. */
1226 if (unlikely(++rdp
->qlen
> qhimark
)) {
1227 rdp
->blimit
= LONG_MAX
;
1228 force_quiescent_state(rsp
, 0);
1229 } else if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1230 force_quiescent_state(rsp
, 1);
1231 local_irq_restore(flags
);
1235 * Queue an RCU callback for invocation after a grace period.
1237 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1239 __call_rcu(head
, func
, &rcu_state
);
1241 EXPORT_SYMBOL_GPL(call_rcu
);
1244 * Queue an RCU for invocation after a quicker grace period.
1246 void call_rcu_bh(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1248 __call_rcu(head
, func
, &rcu_bh_state
);
1250 EXPORT_SYMBOL_GPL(call_rcu_bh
);
1253 * Check to see if there is any immediate RCU-related work to be done
1254 * by the current CPU, for the specified type of RCU, returning 1 if so.
1255 * The checks are in order of increasing expense: checks that can be
1256 * carried out against CPU-local state are performed first. However,
1257 * we must check for CPU stalls first, else we might not get a chance.
1259 static int __rcu_pending(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1261 rdp
->n_rcu_pending
++;
1263 /* Check for CPU stalls, if enabled. */
1264 check_cpu_stall(rsp
, rdp
);
1266 /* Is the RCU core waiting for a quiescent state from this CPU? */
1267 if (rdp
->qs_pending
) {
1268 rdp
->n_rp_qs_pending
++;
1272 /* Does this CPU have callbacks ready to invoke? */
1273 if (cpu_has_callbacks_ready_to_invoke(rdp
)) {
1274 rdp
->n_rp_cb_ready
++;
1278 /* Has RCU gone idle with this CPU needing another grace period? */
1279 if (cpu_needs_another_gp(rsp
, rdp
)) {
1280 rdp
->n_rp_cpu_needs_gp
++;
1284 /* Has another RCU grace period completed? */
1285 if (ACCESS_ONCE(rsp
->completed
) != rdp
->completed
) { /* outside lock */
1286 rdp
->n_rp_gp_completed
++;
1290 /* Has a new RCU grace period started? */
1291 if (ACCESS_ONCE(rsp
->gpnum
) != rdp
->gpnum
) { /* outside lock */
1292 rdp
->n_rp_gp_started
++;
1296 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1297 if (ACCESS_ONCE(rsp
->completed
) != ACCESS_ONCE(rsp
->gpnum
) &&
1298 ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)) {
1299 rdp
->n_rp_need_fqs
++;
1304 rdp
->n_rp_need_nothing
++;
1309 * Check to see if there is any immediate RCU-related work to be done
1310 * by the current CPU, returning 1 if so. This function is part of the
1311 * RCU implementation; it is -not- an exported member of the RCU API.
1313 int rcu_pending(int cpu
)
1315 return __rcu_pending(&rcu_state
, &per_cpu(rcu_data
, cpu
)) ||
1316 __rcu_pending(&rcu_bh_state
, &per_cpu(rcu_bh_data
, cpu
));
1320 * Check to see if any future RCU-related work will need to be done
1321 * by the current CPU, even if none need be done immediately, returning
1322 * 1 if so. This function is part of the RCU implementation; it is -not-
1323 * an exported member of the RCU API.
1325 int rcu_needs_cpu(int cpu
)
1327 /* RCU callbacks either ready or pending? */
1328 return per_cpu(rcu_data
, cpu
).nxtlist
||
1329 per_cpu(rcu_bh_data
, cpu
).nxtlist
;
1333 * Initialize a CPU's per-CPU RCU data. We take this "scorched earth"
1334 * approach so that we don't have to worry about how long the CPU has
1335 * been gone, or whether it ever was online previously. We do trust the
1336 * ->mynode field, as it is constant for a given struct rcu_data and
1337 * initialized during early boot.
1339 * Note that only one online or offline event can be happening at a given
1340 * time. Note also that we can accept some slop in the rsp->completed
1341 * access due to the fact that this CPU cannot possibly have any RCU
1342 * callbacks in flight yet.
1344 static void __cpuinit
1345 rcu_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1347 unsigned long flags
;
1351 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1352 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1354 /* Set up local state, ensuring consistent view of global state. */
1355 spin_lock_irqsave(&rnp
->lock
, flags
);
1356 lastcomp
= rsp
->completed
;
1357 rdp
->completed
= lastcomp
;
1358 rdp
->gpnum
= lastcomp
;
1359 rdp
->passed_quiesc
= 0; /* We could be racing with new GP, */
1360 rdp
->qs_pending
= 1; /* so set up to respond to current GP. */
1361 rdp
->beenonline
= 1; /* We have now been online. */
1362 rdp
->passed_quiesc_completed
= lastcomp
- 1;
1363 rdp
->grpmask
= 1UL << (cpu
- rdp
->mynode
->grplo
);
1364 rdp
->nxtlist
= NULL
;
1365 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
1366 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
1368 rdp
->blimit
= blimit
;
1370 rdp
->dynticks
= &per_cpu(rcu_dynticks
, cpu
);
1371 #endif /* #ifdef CONFIG_NO_HZ */
1373 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
1376 * A new grace period might start here. If so, we won't be part
1377 * of it, but that is OK, as we are currently in a quiescent state.
1380 /* Exclude any attempts to start a new GP on large systems. */
1381 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
1383 /* Add CPU to rcu_node bitmasks. */
1385 mask
= rdp
->grpmask
;
1387 /* Exclude any attempts to start a new GP on small systems. */
1388 spin_lock(&rnp
->lock
); /* irqs already disabled. */
1389 rnp
->qsmaskinit
|= mask
;
1390 mask
= rnp
->grpmask
;
1391 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
1393 } while (rnp
!= NULL
&& !(rnp
->qsmaskinit
& mask
));
1395 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
1398 * A new grace period might start here. If so, we will be part of
1399 * it, and its gpnum will be greater than ours, so we will
1400 * participate. It is also possible for the gpnum to have been
1401 * incremented before this function was called, and the bitmasks
1402 * to not be filled out until now, in which case we will also
1403 * participate due to our gpnum being behind.
1406 /* Since it is coming online, the CPU is in a quiescent state. */
1407 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
1408 local_irq_restore(flags
);
1411 static void __cpuinit
rcu_online_cpu(int cpu
)
1413 rcu_init_percpu_data(cpu
, &rcu_state
);
1414 rcu_init_percpu_data(cpu
, &rcu_bh_state
);
1415 open_softirq(RCU_SOFTIRQ
, rcu_process_callbacks
);
1419 * Handle CPU online/offline notifcation events.
1421 static int __cpuinit
rcu_cpu_notify(struct notifier_block
*self
,
1422 unsigned long action
, void *hcpu
)
1424 long cpu
= (long)hcpu
;
1427 case CPU_UP_PREPARE
:
1428 case CPU_UP_PREPARE_FROZEN
:
1429 rcu_online_cpu(cpu
);
1432 case CPU_DEAD_FROZEN
:
1433 case CPU_UP_CANCELED
:
1434 case CPU_UP_CANCELED_FROZEN
:
1435 rcu_offline_cpu(cpu
);
1444 * Compute the per-level fanout, either using the exact fanout specified
1445 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1447 #ifdef CONFIG_RCU_FANOUT_EXACT
1448 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1452 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--)
1453 rsp
->levelspread
[i
] = CONFIG_RCU_FANOUT
;
1455 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1456 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1463 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1464 ccur
= rsp
->levelcnt
[i
];
1465 rsp
->levelspread
[i
] = (cprv
+ ccur
- 1) / ccur
;
1469 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1472 * Helper function for rcu_init() that initializes one rcu_state structure.
1474 static void __init
rcu_init_one(struct rcu_state
*rsp
)
1479 struct rcu_node
*rnp
;
1481 /* Initialize the level-tracking arrays. */
1483 for (i
= 1; i
< NUM_RCU_LVLS
; i
++)
1484 rsp
->level
[i
] = rsp
->level
[i
- 1] + rsp
->levelcnt
[i
- 1];
1485 rcu_init_levelspread(rsp
);
1487 /* Initialize the elements themselves, starting from the leaves. */
1489 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1490 cpustride
*= rsp
->levelspread
[i
];
1491 rnp
= rsp
->level
[i
];
1492 for (j
= 0; j
< rsp
->levelcnt
[i
]; j
++, rnp
++) {
1493 spin_lock_init(&rnp
->lock
);
1495 rnp
->qsmaskinit
= 0;
1496 rnp
->grplo
= j
* cpustride
;
1497 rnp
->grphi
= (j
+ 1) * cpustride
- 1;
1498 if (rnp
->grphi
>= NR_CPUS
)
1499 rnp
->grphi
= NR_CPUS
- 1;
1505 rnp
->grpnum
= j
% rsp
->levelspread
[i
- 1];
1506 rnp
->grpmask
= 1UL << rnp
->grpnum
;
1507 rnp
->parent
= rsp
->level
[i
- 1] +
1508 j
/ rsp
->levelspread
[i
- 1];
1516 * Helper macro for __rcu_init(). To be used nowhere else!
1517 * Assigns leaf node pointers into each CPU's rcu_data structure.
1519 #define RCU_DATA_PTR_INIT(rsp, rcu_data) \
1521 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1523 for_each_possible_cpu(i) { \
1524 if (i > rnp[j].grphi) \
1526 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1527 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1531 static struct notifier_block __cpuinitdata rcu_nb
= {
1532 .notifier_call
= rcu_cpu_notify
,
1535 void __init
__rcu_init(void)
1537 int i
; /* All used by RCU_DATA_PTR_INIT(). */
1539 struct rcu_node
*rnp
;
1541 printk(KERN_INFO
"Hierarchical RCU implementation.\n");
1542 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1543 printk(KERN_INFO
"RCU-based detection of stalled CPUs is enabled.\n");
1544 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1545 rcu_init_one(&rcu_state
);
1546 RCU_DATA_PTR_INIT(&rcu_state
, rcu_data
);
1547 rcu_init_one(&rcu_bh_state
);
1548 RCU_DATA_PTR_INIT(&rcu_bh_state
, rcu_bh_data
);
1550 for_each_online_cpu(i
)
1551 rcu_cpu_notify(&rcu_nb
, CPU_UP_PREPARE
, (void *)(long)i
);
1552 /* Register notifier for non-boot CPUs */
1553 register_cpu_notifier(&rcu_nb
);
1556 module_param(blimit
, int, 0);
1557 module_param(qhimark
, int, 0);
1558 module_param(qlowmark
, int, 0);