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
);
114 static int rcu_pending(int cpu
);
117 * Return the number of RCU-sched batches processed thus far for debug & stats.
119 long rcu_batches_completed_sched(void)
121 return rcu_sched_state
.completed
;
123 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched
);
126 * Return the number of RCU batches processed thus far for debug & stats.
127 * @@@ placeholder, maps to rcu_batches_completed_sched().
129 long rcu_batches_completed(void)
131 return rcu_batches_completed_sched();
133 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
136 * Return the number of RCU BH batches processed thus far for debug & stats.
138 long rcu_batches_completed_bh(void)
140 return rcu_bh_state
.completed
;
142 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh
);
145 * Does the CPU have callbacks ready to be invoked?
148 cpu_has_callbacks_ready_to_invoke(struct rcu_data
*rdp
)
150 return &rdp
->nxtlist
!= rdp
->nxttail
[RCU_DONE_TAIL
];
154 * Does the current CPU require a yet-as-unscheduled grace period?
157 cpu_needs_another_gp(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
159 /* ACCESS_ONCE() because we are accessing outside of lock. */
160 return *rdp
->nxttail
[RCU_DONE_TAIL
] &&
161 ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
);
165 * Return the root node of the specified rcu_state structure.
167 static struct rcu_node
*rcu_get_root(struct rcu_state
*rsp
)
169 return &rsp
->node
[0];
175 * If the specified CPU is offline, tell the caller that it is in
176 * a quiescent state. Otherwise, whack it with a reschedule IPI.
177 * Grace periods can end up waiting on an offline CPU when that
178 * CPU is in the process of coming online -- it will be added to the
179 * rcu_node bitmasks before it actually makes it online. The same thing
180 * can happen while a CPU is in the process of coming online. Because this
181 * race is quite rare, we check for it after detecting that the grace
182 * period has been delayed rather than checking each and every CPU
183 * each and every time we start a new grace period.
185 static int rcu_implicit_offline_qs(struct rcu_data
*rdp
)
188 * If the CPU is offline, it is in a quiescent state. We can
189 * trust its state not to change because interrupts are disabled.
191 if (cpu_is_offline(rdp
->cpu
)) {
196 /* The CPU is online, so send it a reschedule IPI. */
197 if (rdp
->cpu
!= smp_processor_id())
198 smp_send_reschedule(rdp
->cpu
);
205 #endif /* #ifdef CONFIG_SMP */
208 static DEFINE_RATELIMIT_STATE(rcu_rs
, 10 * HZ
, 5);
211 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
213 * Enter nohz mode, in other words, -leave- the mode in which RCU
214 * read-side critical sections can occur. (Though RCU read-side
215 * critical sections can occur in irq handlers in nohz mode, a possibility
216 * handled by rcu_irq_enter() and rcu_irq_exit()).
218 void rcu_enter_nohz(void)
221 struct rcu_dynticks
*rdtp
;
223 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
224 local_irq_save(flags
);
225 rdtp
= &__get_cpu_var(rcu_dynticks
);
227 rdtp
->dynticks_nesting
--;
228 WARN_ON_RATELIMIT(rdtp
->dynticks
& 0x1, &rcu_rs
);
229 local_irq_restore(flags
);
233 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
235 * Exit nohz mode, in other words, -enter- the mode in which RCU
236 * read-side critical sections normally occur.
238 void rcu_exit_nohz(void)
241 struct rcu_dynticks
*rdtp
;
243 local_irq_save(flags
);
244 rdtp
= &__get_cpu_var(rcu_dynticks
);
246 rdtp
->dynticks_nesting
++;
247 WARN_ON_RATELIMIT(!(rdtp
->dynticks
& 0x1), &rcu_rs
);
248 local_irq_restore(flags
);
249 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
253 * rcu_nmi_enter - inform RCU of entry to NMI context
255 * If the CPU was idle with dynamic ticks active, and there is no
256 * irq handler running, this updates rdtp->dynticks_nmi to let the
257 * RCU grace-period handling know that the CPU is active.
259 void rcu_nmi_enter(void)
261 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
263 if (rdtp
->dynticks
& 0x1)
265 rdtp
->dynticks_nmi
++;
266 WARN_ON_RATELIMIT(!(rdtp
->dynticks_nmi
& 0x1), &rcu_rs
);
267 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
271 * rcu_nmi_exit - inform RCU of exit from NMI context
273 * If the CPU was idle with dynamic ticks active, and there is no
274 * irq handler running, this updates rdtp->dynticks_nmi to let the
275 * RCU grace-period handling know that the CPU is no longer active.
277 void rcu_nmi_exit(void)
279 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
281 if (rdtp
->dynticks
& 0x1)
283 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
284 rdtp
->dynticks_nmi
++;
285 WARN_ON_RATELIMIT(rdtp
->dynticks_nmi
& 0x1, &rcu_rs
);
289 * rcu_irq_enter - inform RCU of entry to hard irq context
291 * If the CPU was idle with dynamic ticks active, this updates the
292 * rdtp->dynticks to let the RCU handling know that the CPU is active.
294 void rcu_irq_enter(void)
296 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
298 if (rdtp
->dynticks_nesting
++)
301 WARN_ON_RATELIMIT(!(rdtp
->dynticks
& 0x1), &rcu_rs
);
302 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
306 * rcu_irq_exit - inform RCU of exit from hard irq context
308 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
309 * to put let the RCU handling be aware that the CPU is going back to idle
312 void rcu_irq_exit(void)
314 struct rcu_dynticks
*rdtp
= &__get_cpu_var(rcu_dynticks
);
316 if (--rdtp
->dynticks_nesting
)
318 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
320 WARN_ON_RATELIMIT(rdtp
->dynticks
& 0x1, &rcu_rs
);
322 /* If the interrupt queued a callback, get out of dyntick mode. */
323 if (__get_cpu_var(rcu_sched_data
).nxtlist
||
324 __get_cpu_var(rcu_bh_data
).nxtlist
)
329 * Record the specified "completed" value, which is later used to validate
330 * dynticks counter manipulations. Specify "rsp->completed - 1" to
331 * unconditionally invalidate any future dynticks manipulations (which is
332 * useful at the beginning of a grace period).
334 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
336 rsp
->dynticks_completed
= comp
;
342 * Recall the previously recorded value of the completion for dynticks.
344 static long dyntick_recall_completed(struct rcu_state
*rsp
)
346 return rsp
->dynticks_completed
;
350 * Snapshot the specified CPU's dynticks counter so that we can later
351 * credit them with an implicit quiescent state. Return 1 if this CPU
352 * is already in a quiescent state courtesy of dynticks idle mode.
354 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
360 snap
= rdp
->dynticks
->dynticks
;
361 snap_nmi
= rdp
->dynticks
->dynticks_nmi
;
362 smp_mb(); /* Order sampling of snap with end of grace period. */
363 rdp
->dynticks_snap
= snap
;
364 rdp
->dynticks_nmi_snap
= snap_nmi
;
365 ret
= ((snap
& 0x1) == 0) && ((snap_nmi
& 0x1) == 0);
372 * Return true if the specified CPU has passed through a quiescent
373 * state by virtue of being in or having passed through an dynticks
374 * idle state since the last call to dyntick_save_progress_counter()
377 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
384 curr
= rdp
->dynticks
->dynticks
;
385 snap
= rdp
->dynticks_snap
;
386 curr_nmi
= rdp
->dynticks
->dynticks_nmi
;
387 snap_nmi
= rdp
->dynticks_nmi_snap
;
388 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
391 * If the CPU passed through or entered a dynticks idle phase with
392 * no active irq/NMI handlers, then we can safely pretend that the CPU
393 * already acknowledged the request to pass through a quiescent
394 * state. Either way, that CPU cannot possibly be in an RCU
395 * read-side critical section that started before the beginning
396 * of the current RCU grace period.
398 if ((curr
!= snap
|| (curr
& 0x1) == 0) &&
399 (curr_nmi
!= snap_nmi
|| (curr_nmi
& 0x1) == 0)) {
404 /* Go check for the CPU being offline. */
405 return rcu_implicit_offline_qs(rdp
);
408 #endif /* #ifdef CONFIG_SMP */
410 #else /* #ifdef CONFIG_NO_HZ */
412 static void dyntick_record_completed(struct rcu_state
*rsp
, long comp
)
419 * If there are no dynticks, then the only way that a CPU can passively
420 * be in a quiescent state is to be offline. Unlike dynticks idle, which
421 * is a point in time during the prior (already finished) grace period,
422 * an offline CPU is always in a quiescent state, and thus can be
423 * unconditionally applied. So just return the current value of completed.
425 static long dyntick_recall_completed(struct rcu_state
*rsp
)
427 return rsp
->completed
;
430 static int dyntick_save_progress_counter(struct rcu_data
*rdp
)
435 static int rcu_implicit_dynticks_qs(struct rcu_data
*rdp
)
437 return rcu_implicit_offline_qs(rdp
);
440 #endif /* #ifdef CONFIG_SMP */
442 #endif /* #else #ifdef CONFIG_NO_HZ */
444 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
446 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
448 rsp
->gp_start
= jiffies
;
449 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_CHECK
;
452 static void print_other_cpu_stall(struct rcu_state
*rsp
)
457 struct rcu_node
*rnp
= rcu_get_root(rsp
);
458 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
459 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
461 /* Only let one CPU complain about others per time interval. */
463 spin_lock_irqsave(&rnp
->lock
, flags
);
464 delta
= jiffies
- rsp
->jiffies_stall
;
465 if (delta
< RCU_STALL_RAT_DELAY
|| rsp
->gpnum
== rsp
->completed
) {
466 spin_unlock_irqrestore(&rnp
->lock
, flags
);
469 rsp
->jiffies_stall
= jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
470 spin_unlock_irqrestore(&rnp
->lock
, flags
);
472 /* OK, time to rat on our buddy... */
474 printk(KERN_ERR
"INFO: RCU detected CPU stalls:");
475 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
476 if (rnp_cur
->qsmask
== 0)
478 for (cpu
= 0; cpu
<= rnp_cur
->grphi
- rnp_cur
->grplo
; cpu
++)
479 if (rnp_cur
->qsmask
& (1UL << cpu
))
480 printk(" %d", rnp_cur
->grplo
+ cpu
);
482 printk(" (detected by %d, t=%ld jiffies)\n",
483 smp_processor_id(), (long)(jiffies
- rsp
->gp_start
));
484 force_quiescent_state(rsp
, 0); /* Kick them all. */
487 static void print_cpu_stall(struct rcu_state
*rsp
)
490 struct rcu_node
*rnp
= rcu_get_root(rsp
);
492 printk(KERN_ERR
"INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
493 smp_processor_id(), jiffies
- rsp
->gp_start
);
495 spin_lock_irqsave(&rnp
->lock
, flags
);
496 if ((long)(jiffies
- rsp
->jiffies_stall
) >= 0)
498 jiffies
+ RCU_SECONDS_TILL_STALL_RECHECK
;
499 spin_unlock_irqrestore(&rnp
->lock
, flags
);
500 set_need_resched(); /* kick ourselves to get things going. */
503 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
506 struct rcu_node
*rnp
;
508 delta
= jiffies
- rsp
->jiffies_stall
;
510 if ((rnp
->qsmask
& rdp
->grpmask
) && delta
>= 0) {
512 /* We haven't checked in, so go dump stack. */
513 print_cpu_stall(rsp
);
515 } else if (rsp
->gpnum
!= rsp
->completed
&&
516 delta
>= RCU_STALL_RAT_DELAY
) {
518 /* They had two time units to dump stack, so complain. */
519 print_other_cpu_stall(rsp
);
523 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
525 static void record_gp_stall_check_time(struct rcu_state
*rsp
)
529 static void check_cpu_stall(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
533 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
536 * Update CPU-local rcu_data state to record the newly noticed grace period.
537 * This is used both when we started the grace period and when we notice
538 * that someone else started the grace period.
540 static void note_new_gpnum(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
543 rdp
->passed_quiesc
= 0;
544 rdp
->gpnum
= rsp
->gpnum
;
548 * Did someone else start a new RCU grace period start since we last
549 * checked? Update local state appropriately if so. Must be called
550 * on the CPU corresponding to rdp.
553 check_for_new_grace_period(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
558 local_irq_save(flags
);
559 if (rdp
->gpnum
!= rsp
->gpnum
) {
560 note_new_gpnum(rsp
, rdp
);
563 local_irq_restore(flags
);
568 * Start a new RCU grace period if warranted, re-initializing the hierarchy
569 * in preparation for detecting the next grace period. The caller must hold
570 * the root node's ->lock, which is released before return. Hard irqs must
574 rcu_start_gp(struct rcu_state
*rsp
, unsigned long flags
)
575 __releases(rcu_get_root(rsp
)->lock
)
577 struct rcu_data
*rdp
= rsp
->rda
[smp_processor_id()];
578 struct rcu_node
*rnp
= rcu_get_root(rsp
);
579 struct rcu_node
*rnp_cur
;
580 struct rcu_node
*rnp_end
;
582 if (!cpu_needs_another_gp(rsp
, rdp
)) {
583 spin_unlock_irqrestore(&rnp
->lock
, flags
);
587 /* Advance to a new grace period and initialize state. */
589 rsp
->signaled
= RCU_GP_INIT
; /* Hold off force_quiescent_state. */
590 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
591 record_gp_stall_check_time(rsp
);
592 dyntick_record_completed(rsp
, rsp
->completed
- 1);
593 note_new_gpnum(rsp
, rdp
);
596 * Because we are first, we know that all our callbacks will
597 * be covered by this upcoming grace period, even the ones
598 * that were registered arbitrarily recently.
600 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
601 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
603 /* Special-case the common single-level case. */
604 if (NUM_RCU_NODES
== 1) {
605 rnp
->qsmask
= rnp
->qsmaskinit
;
606 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state OK. */
607 spin_unlock_irqrestore(&rnp
->lock
, flags
);
611 spin_unlock(&rnp
->lock
); /* leave irqs disabled. */
614 /* Exclude any concurrent CPU-hotplug operations. */
615 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
618 * Set the quiescent-state-needed bits in all the non-leaf RCU
619 * nodes for all currently online CPUs. This operation relies
620 * on the layout of the hierarchy within the rsp->node[] array.
621 * Note that other CPUs will access only the leaves of the
622 * hierarchy, which still indicate that no grace period is in
623 * progress. In addition, we have excluded CPU-hotplug operations.
625 * We therefore do not need to hold any locks. Any required
626 * memory barriers will be supplied by the locks guarding the
627 * leaf rcu_nodes in the hierarchy.
630 rnp_end
= rsp
->level
[NUM_RCU_LVLS
- 1];
631 for (rnp_cur
= &rsp
->node
[0]; rnp_cur
< rnp_end
; rnp_cur
++)
632 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
635 * Now set up the leaf nodes. Here we must be careful. First,
636 * we need to hold the lock in order to exclude other CPUs, which
637 * might be contending for the leaf nodes' locks. Second, as
638 * soon as we initialize a given leaf node, its CPUs might run
639 * up the rest of the hierarchy. We must therefore acquire locks
640 * for each node that we touch during this stage. (But we still
641 * are excluding CPU-hotplug operations.)
643 * Note that the grace period cannot complete until we finish
644 * the initialization process, as there will be at least one
645 * qsmask bit set in the root node until that time, namely the
646 * one corresponding to this CPU.
648 rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
649 rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
650 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
651 spin_lock(&rnp_cur
->lock
); /* irqs already disabled. */
652 rnp_cur
->qsmask
= rnp_cur
->qsmaskinit
;
653 spin_unlock(&rnp_cur
->lock
); /* irqs already disabled. */
656 rsp
->signaled
= RCU_SIGNAL_INIT
; /* force_quiescent_state now OK. */
657 spin_unlock_irqrestore(&rsp
->onofflock
, flags
);
661 * Advance this CPU's callbacks, but only if the current grace period
662 * has ended. This may be called only from the CPU to whom the rdp
666 rcu_process_gp_end(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
671 local_irq_save(flags
);
672 completed_snap
= ACCESS_ONCE(rsp
->completed
); /* outside of lock. */
674 /* Did another grace period end? */
675 if (rdp
->completed
!= completed_snap
) {
677 /* Advance callbacks. No harm if list empty. */
678 rdp
->nxttail
[RCU_DONE_TAIL
] = rdp
->nxttail
[RCU_WAIT_TAIL
];
679 rdp
->nxttail
[RCU_WAIT_TAIL
] = rdp
->nxttail
[RCU_NEXT_READY_TAIL
];
680 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
682 /* Remember that we saw this grace-period completion. */
683 rdp
->completed
= completed_snap
;
685 local_irq_restore(flags
);
689 * Similar to cpu_quiet(), for which it is a helper function. Allows
690 * a group of CPUs to be quieted at one go, though all the CPUs in the
691 * group must be represented by the same leaf rcu_node structure.
692 * That structure's lock must be held upon entry, and it is released
696 cpu_quiet_msk(unsigned long mask
, struct rcu_state
*rsp
, struct rcu_node
*rnp
,
698 __releases(rnp
->lock
)
700 /* Walk up the rcu_node hierarchy. */
702 if (!(rnp
->qsmask
& mask
)) {
704 /* Our bit has already been cleared, so done. */
705 spin_unlock_irqrestore(&rnp
->lock
, flags
);
708 rnp
->qsmask
&= ~mask
;
709 if (rnp
->qsmask
!= 0) {
711 /* Other bits still set at this level, so done. */
712 spin_unlock_irqrestore(&rnp
->lock
, flags
);
716 if (rnp
->parent
== NULL
) {
718 /* No more levels. Exit loop holding root lock. */
722 spin_unlock_irqrestore(&rnp
->lock
, flags
);
724 spin_lock_irqsave(&rnp
->lock
, flags
);
728 * Get here if we are the last CPU to pass through a quiescent
729 * state for this grace period. Clean up and let rcu_start_gp()
730 * start up the next grace period if one is needed. Note that
731 * we still hold rnp->lock, as required by rcu_start_gp(), which
734 rsp
->completed
= rsp
->gpnum
;
735 rcu_process_gp_end(rsp
, rsp
->rda
[smp_processor_id()]);
736 rcu_start_gp(rsp
, flags
); /* releases rnp->lock. */
740 * Record a quiescent state for the specified CPU, which must either be
741 * the current CPU or an offline CPU. The lastcomp argument is used to
742 * make sure we are still in the grace period of interest. We don't want
743 * to end the current grace period based on quiescent states detected in
744 * an earlier grace period!
747 cpu_quiet(int cpu
, struct rcu_state
*rsp
, struct rcu_data
*rdp
, long lastcomp
)
751 struct rcu_node
*rnp
;
754 spin_lock_irqsave(&rnp
->lock
, flags
);
755 if (lastcomp
!= ACCESS_ONCE(rsp
->completed
)) {
758 * Someone beat us to it for this grace period, so leave.
759 * The race with GP start is resolved by the fact that we
760 * hold the leaf rcu_node lock, so that the per-CPU bits
761 * cannot yet be initialized -- so we would simply find our
762 * CPU's bit already cleared in cpu_quiet_msk() if this race
765 rdp
->passed_quiesc
= 0; /* try again later! */
766 spin_unlock_irqrestore(&rnp
->lock
, flags
);
770 if ((rnp
->qsmask
& mask
) == 0) {
771 spin_unlock_irqrestore(&rnp
->lock
, flags
);
776 * This GP can't end until cpu checks in, so all of our
777 * callbacks can be processed during the next GP.
779 rdp
= rsp
->rda
[smp_processor_id()];
780 rdp
->nxttail
[RCU_NEXT_READY_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
782 cpu_quiet_msk(mask
, rsp
, rnp
, flags
); /* releases rnp->lock */
787 * Check to see if there is a new grace period of which this CPU
788 * is not yet aware, and if so, set up local rcu_data state for it.
789 * Otherwise, see if this CPU has just passed through its first
790 * quiescent state for this grace period, and record that fact if so.
793 rcu_check_quiescent_state(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
795 /* If there is now a new grace period, record and return. */
796 if (check_for_new_grace_period(rsp
, rdp
))
800 * Does this CPU still need to do its part for current grace period?
801 * If no, return and let the other CPUs do their part as well.
803 if (!rdp
->qs_pending
)
807 * Was there a quiescent state since the beginning of the grace
808 * period? If no, then exit and wait for the next call.
810 if (!rdp
->passed_quiesc
)
813 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
814 cpu_quiet(rdp
->cpu
, rsp
, rdp
, rdp
->passed_quiesc_completed
);
817 #ifdef CONFIG_HOTPLUG_CPU
820 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
821 * and move all callbacks from the outgoing CPU to the current one.
823 static void __rcu_offline_cpu(int cpu
, struct rcu_state
*rsp
)
829 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
830 struct rcu_data
*rdp_me
;
831 struct rcu_node
*rnp
;
833 /* Exclude any attempts to start a new grace period. */
834 spin_lock_irqsave(&rsp
->onofflock
, flags
);
836 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
838 mask
= rdp
->grpmask
; /* rnp->grplo is constant. */
840 spin_lock(&rnp
->lock
); /* irqs already disabled. */
841 rnp
->qsmaskinit
&= ~mask
;
842 if (rnp
->qsmaskinit
!= 0) {
843 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
847 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
849 } while (rnp
!= NULL
);
850 lastcomp
= rsp
->completed
;
852 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
854 /* Being offline is a quiescent state, so go record it. */
855 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
858 * Move callbacks from the outgoing CPU to the running CPU.
859 * Note that the outgoing CPU is now quiscent, so it is now
860 * (uncharacteristically) safe to access its rcu_data structure.
861 * Note also that we must carefully retain the order of the
862 * outgoing CPU's callbacks in order for rcu_barrier() to work
863 * correctly. Finally, note that we start all the callbacks
864 * afresh, even those that have passed through a grace period
865 * and are therefore ready to invoke. The theory is that hotplug
866 * events are rare, and that if they are frequent enough to
867 * indefinitely delay callbacks, you have far worse things to
870 rdp_me
= rsp
->rda
[smp_processor_id()];
871 if (rdp
->nxtlist
!= NULL
) {
872 *rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxtlist
;
873 rdp_me
->nxttail
[RCU_NEXT_TAIL
] = rdp
->nxttail
[RCU_NEXT_TAIL
];
875 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
876 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
877 rdp_me
->qlen
+= rdp
->qlen
;
880 local_irq_restore(flags
);
884 * Remove the specified CPU from the RCU hierarchy and move any pending
885 * callbacks that it might have to the current CPU. This code assumes
886 * that at least one CPU in the system will remain running at all times.
887 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
889 static void rcu_offline_cpu(int cpu
)
891 __rcu_offline_cpu(cpu
, &rcu_sched_state
);
892 __rcu_offline_cpu(cpu
, &rcu_bh_state
);
895 #else /* #ifdef CONFIG_HOTPLUG_CPU */
897 static void rcu_offline_cpu(int cpu
)
901 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
904 * Invoke any RCU callbacks that have made it to the end of their grace
905 * period. Thottle as specified by rdp->blimit.
907 static void rcu_do_batch(struct rcu_data
*rdp
)
910 struct rcu_head
*next
, *list
, **tail
;
913 /* If no callbacks are ready, just return.*/
914 if (!cpu_has_callbacks_ready_to_invoke(rdp
))
918 * Extract the list of ready callbacks, disabling to prevent
919 * races with call_rcu() from interrupt handlers.
921 local_irq_save(flags
);
923 rdp
->nxtlist
= *rdp
->nxttail
[RCU_DONE_TAIL
];
924 *rdp
->nxttail
[RCU_DONE_TAIL
] = NULL
;
925 tail
= rdp
->nxttail
[RCU_DONE_TAIL
];
926 for (count
= RCU_NEXT_SIZE
- 1; count
>= 0; count
--)
927 if (rdp
->nxttail
[count
] == rdp
->nxttail
[RCU_DONE_TAIL
])
928 rdp
->nxttail
[count
] = &rdp
->nxtlist
;
929 local_irq_restore(flags
);
931 /* Invoke callbacks. */
938 if (++count
>= rdp
->blimit
)
942 local_irq_save(flags
);
944 /* Update count, and requeue any remaining callbacks. */
947 *tail
= rdp
->nxtlist
;
949 for (count
= 0; count
< RCU_NEXT_SIZE
; count
++)
950 if (&rdp
->nxtlist
== rdp
->nxttail
[count
])
951 rdp
->nxttail
[count
] = tail
;
956 /* Reinstate batch limit if we have worked down the excess. */
957 if (rdp
->blimit
== LONG_MAX
&& rdp
->qlen
<= qlowmark
)
958 rdp
->blimit
= blimit
;
960 local_irq_restore(flags
);
962 /* Re-raise the RCU softirq if there are callbacks remaining. */
963 if (cpu_has_callbacks_ready_to_invoke(rdp
))
964 raise_softirq(RCU_SOFTIRQ
);
968 * Check to see if this CPU is in a non-context-switch quiescent state
969 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
970 * Also schedule the RCU softirq handler.
972 * This function must be called with hardirqs disabled. It is normally
973 * invoked from the scheduling-clock interrupt. If rcu_pending returns
974 * false, there is no point in invoking rcu_check_callbacks().
976 void rcu_check_callbacks(int cpu
, int user
)
978 if (!rcu_pending(cpu
))
979 return; /* if nothing for RCU to do. */
981 (idle_cpu(cpu
) && rcu_scheduler_active
&&
982 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT
))) {
985 * Get here if this CPU took its interrupt from user
986 * mode or from the idle loop, and if this is not a
987 * nested interrupt. In this case, the CPU is in
988 * a quiescent state, so note it.
990 * No memory barrier is required here because both
991 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
992 * variables that other CPUs neither access nor modify,
993 * at least not while the corresponding CPU is online.
999 } else if (!in_softirq()) {
1002 * Get here if this CPU did not take its interrupt from
1003 * softirq, in other words, if it is not interrupting
1004 * a rcu_bh read-side critical section. This is an _bh
1005 * critical section, so note it.
1010 raise_softirq(RCU_SOFTIRQ
);
1016 * Scan the leaf rcu_node structures, processing dyntick state for any that
1017 * have not yet encountered a quiescent state, using the function specified.
1018 * Returns 1 if the current grace period ends while scanning (possibly
1019 * because we made it end).
1021 static int rcu_process_dyntick(struct rcu_state
*rsp
, long lastcomp
,
1022 int (*f
)(struct rcu_data
*))
1026 unsigned long flags
;
1028 struct rcu_node
*rnp_cur
= rsp
->level
[NUM_RCU_LVLS
- 1];
1029 struct rcu_node
*rnp_end
= &rsp
->node
[NUM_RCU_NODES
];
1031 for (; rnp_cur
< rnp_end
; rnp_cur
++) {
1033 spin_lock_irqsave(&rnp_cur
->lock
, flags
);
1034 if (rsp
->completed
!= lastcomp
) {
1035 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1038 if (rnp_cur
->qsmask
== 0) {
1039 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1042 cpu
= rnp_cur
->grplo
;
1044 for (; cpu
<= rnp_cur
->grphi
; cpu
++, bit
<<= 1) {
1045 if ((rnp_cur
->qsmask
& bit
) != 0 && f(rsp
->rda
[cpu
]))
1048 if (mask
!= 0 && rsp
->completed
== lastcomp
) {
1050 /* cpu_quiet_msk() releases rnp_cur->lock. */
1051 cpu_quiet_msk(mask
, rsp
, rnp_cur
, flags
);
1054 spin_unlock_irqrestore(&rnp_cur
->lock
, flags
);
1060 * Force quiescent states on reluctant CPUs, and also detect which
1061 * CPUs are in dyntick-idle mode.
1063 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1065 unsigned long flags
;
1067 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1070 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
))
1071 return; /* No grace period in progress, nothing to force. */
1072 if (!spin_trylock_irqsave(&rsp
->fqslock
, flags
)) {
1073 rsp
->n_force_qs_lh
++; /* Inexact, can lose counts. Tough! */
1074 return; /* Someone else is already on the job. */
1077 (long)(rsp
->jiffies_force_qs
- jiffies
) >= 0)
1078 goto unlock_ret
; /* no emergency and done recently. */
1080 spin_lock(&rnp
->lock
);
1081 lastcomp
= rsp
->completed
;
1082 signaled
= rsp
->signaled
;
1083 rsp
->jiffies_force_qs
= jiffies
+ RCU_JIFFIES_TILL_FORCE_QS
;
1084 if (lastcomp
== rsp
->gpnum
) {
1085 rsp
->n_force_qs_ngp
++;
1086 spin_unlock(&rnp
->lock
);
1087 goto unlock_ret
; /* no GP in progress, time updated. */
1089 spin_unlock(&rnp
->lock
);
1093 break; /* grace period still initializing, ignore. */
1095 case RCU_SAVE_DYNTICK
:
1097 if (RCU_SIGNAL_INIT
!= RCU_SAVE_DYNTICK
)
1098 break; /* So gcc recognizes the dead code. */
1100 /* Record dyntick-idle state. */
1101 if (rcu_process_dyntick(rsp
, lastcomp
,
1102 dyntick_save_progress_counter
))
1105 /* Update state, record completion counter. */
1106 spin_lock(&rnp
->lock
);
1107 if (lastcomp
== rsp
->completed
) {
1108 rsp
->signaled
= RCU_FORCE_QS
;
1109 dyntick_record_completed(rsp
, lastcomp
);
1111 spin_unlock(&rnp
->lock
);
1116 /* Check dyntick-idle state, send IPI to laggarts. */
1117 if (rcu_process_dyntick(rsp
, dyntick_recall_completed(rsp
),
1118 rcu_implicit_dynticks_qs
))
1121 /* Leave state in case more forcing is required. */
1126 spin_unlock_irqrestore(&rsp
->fqslock
, flags
);
1129 #else /* #ifdef CONFIG_SMP */
1131 static void force_quiescent_state(struct rcu_state
*rsp
, int relaxed
)
1136 #endif /* #else #ifdef CONFIG_SMP */
1139 * This does the RCU processing work from softirq context for the
1140 * specified rcu_state and rcu_data structures. This may be called
1141 * only from the CPU to whom the rdp belongs.
1144 __rcu_process_callbacks(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1146 unsigned long flags
;
1148 WARN_ON_ONCE(rdp
->beenonline
== 0);
1151 * If an RCU GP has gone long enough, go check for dyntick
1152 * idle CPUs and, if needed, send resched IPIs.
1154 if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1155 force_quiescent_state(rsp
, 1);
1158 * Advance callbacks in response to end of earlier grace
1159 * period that some other CPU ended.
1161 rcu_process_gp_end(rsp
, rdp
);
1163 /* Update RCU state based on any recent quiescent states. */
1164 rcu_check_quiescent_state(rsp
, rdp
);
1166 /* Does this CPU require a not-yet-started grace period? */
1167 if (cpu_needs_another_gp(rsp
, rdp
)) {
1168 spin_lock_irqsave(&rcu_get_root(rsp
)->lock
, flags
);
1169 rcu_start_gp(rsp
, flags
); /* releases above lock */
1172 /* If there are callbacks ready, invoke them. */
1177 * Do softirq processing for the current CPU.
1179 static void rcu_process_callbacks(struct softirq_action
*unused
)
1182 * Memory references from any prior RCU read-side critical sections
1183 * executed by the interrupted code must be seen before any RCU
1184 * grace-period manipulations below.
1186 smp_mb(); /* See above block comment. */
1188 __rcu_process_callbacks(&rcu_sched_state
,
1189 &__get_cpu_var(rcu_sched_data
));
1190 __rcu_process_callbacks(&rcu_bh_state
, &__get_cpu_var(rcu_bh_data
));
1193 * Memory references from any later RCU read-side critical sections
1194 * executed by the interrupted code must be seen after any RCU
1195 * grace-period manipulations above.
1197 smp_mb(); /* See above block comment. */
1201 __call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
),
1202 struct rcu_state
*rsp
)
1204 unsigned long flags
;
1205 struct rcu_data
*rdp
;
1210 smp_mb(); /* Ensure RCU update seen before callback registry. */
1213 * Opportunistically note grace-period endings and beginnings.
1214 * Note that we might see a beginning right after we see an
1215 * end, but never vice versa, since this CPU has to pass through
1216 * a quiescent state betweentimes.
1218 local_irq_save(flags
);
1219 rdp
= rsp
->rda
[smp_processor_id()];
1220 rcu_process_gp_end(rsp
, rdp
);
1221 check_for_new_grace_period(rsp
, rdp
);
1223 /* Add the callback to our list. */
1224 *rdp
->nxttail
[RCU_NEXT_TAIL
] = head
;
1225 rdp
->nxttail
[RCU_NEXT_TAIL
] = &head
->next
;
1227 /* Start a new grace period if one not already started. */
1228 if (ACCESS_ONCE(rsp
->completed
) == ACCESS_ONCE(rsp
->gpnum
)) {
1229 unsigned long nestflag
;
1230 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
1232 spin_lock_irqsave(&rnp_root
->lock
, nestflag
);
1233 rcu_start_gp(rsp
, nestflag
); /* releases rnp_root->lock. */
1236 /* Force the grace period if too many callbacks or too long waiting. */
1237 if (unlikely(++rdp
->qlen
> qhimark
)) {
1238 rdp
->blimit
= LONG_MAX
;
1239 force_quiescent_state(rsp
, 0);
1240 } else if ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)
1241 force_quiescent_state(rsp
, 1);
1242 local_irq_restore(flags
);
1246 * Queue an RCU-sched callback for invocation after a grace period.
1248 void call_rcu_sched(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1250 __call_rcu(head
, func
, &rcu_sched_state
);
1252 EXPORT_SYMBOL_GPL(call_rcu_sched
);
1255 * @@@ Queue an RCU callback for invocation after a grace period.
1256 * @@@ Placeholder pending rcutree_plugin.h.
1258 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1260 call_rcu_sched(head
, func
);
1262 EXPORT_SYMBOL_GPL(call_rcu
);
1266 * Queue an RCU for invocation after a quicker grace period.
1268 void call_rcu_bh(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
1270 __call_rcu(head
, func
, &rcu_bh_state
);
1272 EXPORT_SYMBOL_GPL(call_rcu_bh
);
1275 * Check to see if there is any immediate RCU-related work to be done
1276 * by the current CPU, for the specified type of RCU, returning 1 if so.
1277 * The checks are in order of increasing expense: checks that can be
1278 * carried out against CPU-local state are performed first. However,
1279 * we must check for CPU stalls first, else we might not get a chance.
1281 static int __rcu_pending(struct rcu_state
*rsp
, struct rcu_data
*rdp
)
1283 rdp
->n_rcu_pending
++;
1285 /* Check for CPU stalls, if enabled. */
1286 check_cpu_stall(rsp
, rdp
);
1288 /* Is the RCU core waiting for a quiescent state from this CPU? */
1289 if (rdp
->qs_pending
) {
1290 rdp
->n_rp_qs_pending
++;
1294 /* Does this CPU have callbacks ready to invoke? */
1295 if (cpu_has_callbacks_ready_to_invoke(rdp
)) {
1296 rdp
->n_rp_cb_ready
++;
1300 /* Has RCU gone idle with this CPU needing another grace period? */
1301 if (cpu_needs_another_gp(rsp
, rdp
)) {
1302 rdp
->n_rp_cpu_needs_gp
++;
1306 /* Has another RCU grace period completed? */
1307 if (ACCESS_ONCE(rsp
->completed
) != rdp
->completed
) { /* outside lock */
1308 rdp
->n_rp_gp_completed
++;
1312 /* Has a new RCU grace period started? */
1313 if (ACCESS_ONCE(rsp
->gpnum
) != rdp
->gpnum
) { /* outside lock */
1314 rdp
->n_rp_gp_started
++;
1318 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1319 if (ACCESS_ONCE(rsp
->completed
) != ACCESS_ONCE(rsp
->gpnum
) &&
1320 ((long)(ACCESS_ONCE(rsp
->jiffies_force_qs
) - jiffies
) < 0)) {
1321 rdp
->n_rp_need_fqs
++;
1326 rdp
->n_rp_need_nothing
++;
1331 * Check to see if there is any immediate RCU-related work to be done
1332 * by the current CPU, returning 1 if so. This function is part of the
1333 * RCU implementation; it is -not- an exported member of the RCU API.
1335 static int rcu_pending(int cpu
)
1337 return __rcu_pending(&rcu_sched_state
, &per_cpu(rcu_sched_data
, cpu
)) ||
1338 __rcu_pending(&rcu_bh_state
, &per_cpu(rcu_bh_data
, cpu
));
1342 * Check to see if any future RCU-related work will need to be done
1343 * by the current CPU, even if none need be done immediately, returning
1344 * 1 if so. This function is part of the RCU implementation; it is -not-
1345 * an exported member of the RCU API.
1347 int rcu_needs_cpu(int cpu
)
1349 /* RCU callbacks either ready or pending? */
1350 return per_cpu(rcu_sched_data
, cpu
).nxtlist
||
1351 per_cpu(rcu_bh_data
, cpu
).nxtlist
;
1355 * Do boot-time initialization of a CPU's per-CPU RCU data.
1358 rcu_boot_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1360 unsigned long flags
;
1362 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1363 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1365 /* Set up local state, ensuring consistent view of global state. */
1366 spin_lock_irqsave(&rnp
->lock
, flags
);
1367 rdp
->grpmask
= 1UL << (cpu
- rdp
->mynode
->grplo
);
1368 rdp
->nxtlist
= NULL
;
1369 for (i
= 0; i
< RCU_NEXT_SIZE
; i
++)
1370 rdp
->nxttail
[i
] = &rdp
->nxtlist
;
1373 rdp
->dynticks
= &per_cpu(rcu_dynticks
, cpu
);
1374 #endif /* #ifdef CONFIG_NO_HZ */
1376 spin_unlock_irqrestore(&rnp
->lock
, flags
);
1380 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1381 * offline event can be happening at a given time. Note also that we
1382 * can accept some slop in the rsp->completed access due to the fact
1383 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1385 static void __cpuinit
1386 rcu_init_percpu_data(int cpu
, struct rcu_state
*rsp
)
1388 unsigned long flags
;
1391 struct rcu_data
*rdp
= rsp
->rda
[cpu
];
1392 struct rcu_node
*rnp
= rcu_get_root(rsp
);
1394 /* Set up local state, ensuring consistent view of global state. */
1395 spin_lock_irqsave(&rnp
->lock
, flags
);
1396 lastcomp
= rsp
->completed
;
1397 rdp
->completed
= lastcomp
;
1398 rdp
->gpnum
= lastcomp
;
1399 rdp
->passed_quiesc
= 0; /* We could be racing with new GP, */
1400 rdp
->qs_pending
= 1; /* so set up to respond to current GP. */
1401 rdp
->beenonline
= 1; /* We have now been online. */
1402 rdp
->passed_quiesc_completed
= lastcomp
- 1;
1403 rdp
->blimit
= blimit
;
1404 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
1407 * A new grace period might start here. If so, we won't be part
1408 * of it, but that is OK, as we are currently in a quiescent state.
1411 /* Exclude any attempts to start a new GP on large systems. */
1412 spin_lock(&rsp
->onofflock
); /* irqs already disabled. */
1414 /* Add CPU to rcu_node bitmasks. */
1416 mask
= rdp
->grpmask
;
1418 /* Exclude any attempts to start a new GP on small systems. */
1419 spin_lock(&rnp
->lock
); /* irqs already disabled. */
1420 rnp
->qsmaskinit
|= mask
;
1421 mask
= rnp
->grpmask
;
1422 spin_unlock(&rnp
->lock
); /* irqs already disabled. */
1424 } while (rnp
!= NULL
&& !(rnp
->qsmaskinit
& mask
));
1426 spin_unlock(&rsp
->onofflock
); /* irqs remain disabled. */
1429 * A new grace period might start here. If so, we will be part of
1430 * it, and its gpnum will be greater than ours, so we will
1431 * participate. It is also possible for the gpnum to have been
1432 * incremented before this function was called, and the bitmasks
1433 * to not be filled out until now, in which case we will also
1434 * participate due to our gpnum being behind.
1437 /* Since it is coming online, the CPU is in a quiescent state. */
1438 cpu_quiet(cpu
, rsp
, rdp
, lastcomp
);
1439 local_irq_restore(flags
);
1442 static void __cpuinit
rcu_online_cpu(int cpu
)
1444 rcu_init_percpu_data(cpu
, &rcu_sched_state
);
1445 rcu_init_percpu_data(cpu
, &rcu_bh_state
);
1449 * Handle CPU online/offline notifcation events.
1451 int __cpuinit
rcu_cpu_notify(struct notifier_block
*self
,
1452 unsigned long action
, void *hcpu
)
1454 long cpu
= (long)hcpu
;
1457 case CPU_UP_PREPARE
:
1458 case CPU_UP_PREPARE_FROZEN
:
1459 rcu_online_cpu(cpu
);
1462 case CPU_DEAD_FROZEN
:
1463 case CPU_UP_CANCELED
:
1464 case CPU_UP_CANCELED_FROZEN
:
1465 rcu_offline_cpu(cpu
);
1474 * Compute the per-level fanout, either using the exact fanout specified
1475 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1477 #ifdef CONFIG_RCU_FANOUT_EXACT
1478 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1482 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--)
1483 rsp
->levelspread
[i
] = CONFIG_RCU_FANOUT
;
1485 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1486 static void __init
rcu_init_levelspread(struct rcu_state
*rsp
)
1493 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1494 ccur
= rsp
->levelcnt
[i
];
1495 rsp
->levelspread
[i
] = (cprv
+ ccur
- 1) / ccur
;
1499 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1502 * Helper function for rcu_init() that initializes one rcu_state structure.
1504 static void __init
rcu_init_one(struct rcu_state
*rsp
)
1509 struct rcu_node
*rnp
;
1511 /* Initialize the level-tracking arrays. */
1513 for (i
= 1; i
< NUM_RCU_LVLS
; i
++)
1514 rsp
->level
[i
] = rsp
->level
[i
- 1] + rsp
->levelcnt
[i
- 1];
1515 rcu_init_levelspread(rsp
);
1517 /* Initialize the elements themselves, starting from the leaves. */
1519 for (i
= NUM_RCU_LVLS
- 1; i
>= 0; i
--) {
1520 cpustride
*= rsp
->levelspread
[i
];
1521 rnp
= rsp
->level
[i
];
1522 for (j
= 0; j
< rsp
->levelcnt
[i
]; j
++, rnp
++) {
1523 spin_lock_init(&rnp
->lock
);
1525 rnp
->qsmaskinit
= 0;
1526 rnp
->grplo
= j
* cpustride
;
1527 rnp
->grphi
= (j
+ 1) * cpustride
- 1;
1528 if (rnp
->grphi
>= NR_CPUS
)
1529 rnp
->grphi
= NR_CPUS
- 1;
1535 rnp
->grpnum
= j
% rsp
->levelspread
[i
- 1];
1536 rnp
->grpmask
= 1UL << rnp
->grpnum
;
1537 rnp
->parent
= rsp
->level
[i
- 1] +
1538 j
/ rsp
->levelspread
[i
- 1];
1546 * Helper macro for __rcu_init(). To be used nowhere else!
1547 * Assigns leaf node pointers into each CPU's rcu_data structure.
1549 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1551 rcu_init_one(rsp); \
1552 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1554 for_each_possible_cpu(i) { \
1555 if (i > rnp[j].grphi) \
1557 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1558 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1559 rcu_boot_init_percpu_data(i, rsp); \
1563 void __init
__rcu_init(void)
1565 int i
; /* All used by RCU_DATA_PTR_INIT(). */
1567 struct rcu_node
*rnp
;
1569 printk(KERN_INFO
"Hierarchical RCU implementation.\n");
1570 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1571 printk(KERN_INFO
"RCU-based detection of stalled CPUs is enabled.\n");
1572 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1573 RCU_INIT_FLAVOR(&rcu_sched_state
, rcu_sched_data
);
1574 RCU_INIT_FLAVOR(&rcu_bh_state
, rcu_bh_data
);
1575 open_softirq(RCU_SOFTIRQ
, rcu_process_callbacks
);
1578 module_param(blimit
, int, 0);
1579 module_param(qhimark
, int, 0);
1580 module_param(qlowmark
, int, 0);