[deliverable/linux.git] / kernel / rcupdate.c

/*
 * Read-Copy Update mechanism for mutual exclusion
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
 *
 * Copyright (C) IBM Corporation, 2001
 *
 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
 *	    Manfred Spraul <manfred@colorfullife.com>
 * 
 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
 * Papers:
 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
 *
 * For detailed explanation of Read-Copy Update mechanism see -
 * 		http://lse.sourceforge.net/locking/rcupdate.html
 *
 */
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/smp.h>
#include <linux/rcupdate.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
#include <asm/atomic.h>
#include <linux/bitops.h>
#include <linux/module.h>
#include <linux/completion.h>
#include <linux/moduleparam.h>
#include <linux/percpu.h>
#include <linux/notifier.h>
#include <linux/rcupdate.h>
#include <linux/cpu.h>

/* Definition for rcupdate control block. */
struct rcu_ctrlblk rcu_ctrlblk = {
	.cur = -300,
	.completed = -300,
	.lock = SPIN_LOCK_UNLOCKED,
	.cpumask = CPU_MASK_NONE,
};
struct rcu_ctrlblk rcu_bh_ctrlblk = {
	.cur = -300,
	.completed = -300,
	.lock = SPIN_LOCK_UNLOCKED,
	.cpumask = CPU_MASK_NONE,
};

DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L };
DEFINE_PER_CPU(struct rcu_data, rcu_bh_data) = { 0L };

/* Fake initialization required by compiler */
static DEFINE_PER_CPU(struct tasklet_struct, rcu_tasklet) = {NULL};
static int blimit = 10;
static int qhimark = 10000;
static int qlowmark = 100;
#ifdef CONFIG_SMP
static int rsinterval = 1000;
#endif

static atomic_t rcu_barrier_cpu_count;
static struct semaphore rcu_barrier_sema;
static struct completion rcu_barrier_completion;

#ifdef CONFIG_SMP
static void force_quiescent_state(struct rcu_data *rdp,
			struct rcu_ctrlblk *rcp)
{
	int cpu;
	cpumask_t cpumask;
	set_need_resched();
	if (unlikely(rdp->qlen - rdp->last_rs_qlen > rsinterval)) {
		rdp->last_rs_qlen = rdp->qlen;
		/*
		 * Don't send IPI to itself. With irqs disabled,
		 * rdp->cpu is the current cpu.
		 */
		cpumask = rcp->cpumask;
		cpu_clear(rdp->cpu, cpumask);
		for_each_cpu_mask(cpu, cpumask)
			smp_send_reschedule(cpu);
	}
}
#else
static inline void force_quiescent_state(struct rcu_data *rdp,
			struct rcu_ctrlblk *rcp)
{
	set_need_resched();
}
#endif

/**
 * call_rcu - Queue an RCU callback for invocation after a grace period.
 * @head: structure to be used for queueing the RCU updates.
 * @func: actual update function to be invoked after the grace period
 *
 * The update function will be invoked some time after a full grace
 * period elapses, in other words after all currently executing RCU
 * read-side critical sections have completed.  RCU read-side critical
 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
 * and may be nested.
 */
void fastcall call_rcu(struct rcu_head *head,
				void (*func)(struct rcu_head *rcu))
{
	unsigned long flags;
	struct rcu_data *rdp;

	head->func = func;
	head->next = NULL;
	local_irq_save(flags);
	rdp = &__get_cpu_var(rcu_data);
	*rdp->nxttail = head;
	rdp->nxttail = &head->next;
	if (unlikely(++rdp->qlen > qhimark)) {
		rdp->blimit = INT_MAX;
		force_quiescent_state(rdp, &rcu_ctrlblk);
	}
	local_irq_restore(flags);
}

/**
 * call_rcu_bh - Queue an RCU for invocation after a quicker grace period.
 * @head: structure to be used for queueing the RCU updates.
 * @func: actual update function to be invoked after the grace period
 *
 * The update function will be invoked some time after a full grace
 * period elapses, in other words after all currently executing RCU
 * read-side critical sections have completed. call_rcu_bh() assumes
 * that the read-side critical sections end on completion of a softirq
 * handler. This means that read-side critical sections in process
 * context must not be interrupted by softirqs. This interface is to be
 * used when most of the read-side critical sections are in softirq context.
 * RCU read-side critical sections are delimited by rcu_read_lock() and
 * rcu_read_unlock(), * if in interrupt context or rcu_read_lock_bh()
 * and rcu_read_unlock_bh(), if in process context. These may be nested.
 */
void fastcall call_rcu_bh(struct rcu_head *head,
				void (*func)(struct rcu_head *rcu))
{
	unsigned long flags;
	struct rcu_data *rdp;

	head->func = func;
	head->next = NULL;
	local_irq_save(flags);
	rdp = &__get_cpu_var(rcu_bh_data);
	*rdp->nxttail = head;
	rdp->nxttail = &head->next;

	if (unlikely(++rdp->qlen > qhimark)) {
		rdp->blimit = INT_MAX;
		force_quiescent_state(rdp, &rcu_bh_ctrlblk);
	}

	local_irq_restore(flags);
}

/*
 * Return the number of RCU batches processed thus far.  Useful
 * for debug and statistics.
 */
long rcu_batches_completed(void)
{
	return rcu_ctrlblk.completed;
}

static void rcu_barrier_callback(struct rcu_head *notused)
{
	if (atomic_dec_and_test(&rcu_barrier_cpu_count))
		complete(&rcu_barrier_completion);
}

/*
 * Called with preemption disabled, and from cross-cpu IRQ context.
 */
static void rcu_barrier_func(void *notused)
{
	int cpu = smp_processor_id();
	struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
	struct rcu_head *head;

	head = &rdp->barrier;
	atomic_inc(&rcu_barrier_cpu_count);
	call_rcu(head, rcu_barrier_callback);
}

/**
 * rcu_barrier - Wait until all the in-flight RCUs are complete.
 */
void rcu_barrier(void)
{
	BUG_ON(in_interrupt());
	/* Take cpucontrol semaphore to protect against CPU hotplug */
	down(&rcu_barrier_sema);
	init_completion(&rcu_barrier_completion);
	atomic_set(&rcu_barrier_cpu_count, 0);
	on_each_cpu(rcu_barrier_func, NULL, 0, 1);
	wait_for_completion(&rcu_barrier_completion);
	up(&rcu_barrier_sema);
}
EXPORT_SYMBOL_GPL(rcu_barrier);

/*
 * Invoke the completed RCU callbacks. They are expected to be in
 * a per-cpu list.
 */
static void rcu_do_batch(struct rcu_data *rdp)
{
	struct rcu_head *next, *list;
	int count = 0;

	list = rdp->donelist;
	while (list) {
		next = rdp->donelist = list->next;
		list->func(list);
		list = next;
		rdp->qlen--;
		if (++count >= rdp->blimit)
			break;
	}
	if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark)
		rdp->blimit = blimit;
	if (!rdp->donelist)
		rdp->donetail = &rdp->donelist;
	else
		tasklet_schedule(&per_cpu(rcu_tasklet, rdp->cpu));
}

/*
 * Grace period handling:
 * The grace period handling consists out of two steps:
 * - A new grace period is started.
 *   This is done by rcu_start_batch. The start is not broadcasted to
 *   all cpus, they must pick this up by comparing rcp->cur with
 *   rdp->quiescbatch. All cpus are recorded  in the
 *   rcu_ctrlblk.cpumask bitmap.
 * - All cpus must go through a quiescent state.
 *   Since the start of the grace period is not broadcasted, at least two
 *   calls to rcu_check_quiescent_state are required:
 *   The first call just notices that a new grace period is running. The
 *   following calls check if there was a quiescent state since the beginning
 *   of the grace period. If so, it updates rcu_ctrlblk.cpumask. If
 *   the bitmap is empty, then the grace period is completed.
 *   rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace
 *   period (if necessary).
 */
/*
 * Register a new batch of callbacks, and start it up if there is currently no
 * active batch and the batch to be registered has not already occurred.
 * Caller must hold rcu_ctrlblk.lock.
 */
static void rcu_start_batch(struct rcu_ctrlblk *rcp)
{
	if (rcp->next_pending &&
			rcp->completed == rcp->cur) {
		rcp->next_pending = 0;
		/*
		 * next_pending == 0 must be visible in
		 * __rcu_process_callbacks() before it can see new value of cur.
		 */
		smp_wmb();
		rcp->cur++;

		/*
		 * Accessing nohz_cpu_mask before incrementing rcp->cur needs a
		 * Barrier  Otherwise it can cause tickless idle CPUs to be
		 * included in rcp->cpumask, which will extend graceperiods
		 * unnecessarily.
		 */
		smp_mb();
		cpus_andnot(rcp->cpumask, cpu_online_map, nohz_cpu_mask);

	}
}

/*
 * cpu went through a quiescent state since the beginning of the grace period.
 * Clear it from the cpu mask and complete the grace period if it was the last
 * cpu. Start another grace period if someone has further entries pending
 */
static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp)
{
	cpu_clear(cpu, rcp->cpumask);
	if (cpus_empty(rcp->cpumask)) {
		/* batch completed ! */
		rcp->completed = rcp->cur;
		rcu_start_batch(rcp);
	}
}

/*
 * Check if the cpu has gone through a quiescent state (say context
 * switch). If so and if it already hasn't done so in this RCU
 * quiescent cycle, then indicate that it has done so.
 */
static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp,
					struct rcu_data *rdp)
{
	if (rdp->quiescbatch != rcp->cur) {
		/* start new grace period: */
		rdp->qs_pending = 1;
		rdp->passed_quiesc = 0;
		rdp->quiescbatch = rcp->cur;
		return;
	}

	/* Grace period already completed for this cpu?
	 * qs_pending is checked instead of the actual bitmap to avoid
	 * cacheline trashing.
	 */
	if (!rdp->qs_pending)
		return;

	/* 
	 * Was there a quiescent state since the beginning of the grace
	 * period? If no, then exit and wait for the next call.
	 */
	if (!rdp->passed_quiesc)
		return;
	rdp->qs_pending = 0;

	spin_lock(&rcp->lock);
	/*
	 * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync
	 * during cpu startup. Ignore the quiescent state.
	 */
	if (likely(rdp->quiescbatch == rcp->cur))
		cpu_quiet(rdp->cpu, rcp);

	spin_unlock(&rcp->lock);
}


#ifdef CONFIG_HOTPLUG_CPU

/* warning! helper for rcu_offline_cpu. do not use elsewhere without reviewing
 * locking requirements, the list it's pulling from has to belong to a cpu
 * which is dead and hence not processing interrupts.
 */
static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list,
				struct rcu_head **tail)
{
	local_irq_disable();
	*this_rdp->nxttail = list;
	if (list)
		this_rdp->nxttail = tail;
	local_irq_enable();
}

static void __rcu_offline_cpu(struct rcu_data *this_rdp,
				struct rcu_ctrlblk *rcp, struct rcu_data *rdp)
{
	/* if the cpu going offline owns the grace period
	 * we can block indefinitely waiting for it, so flush
	 * it here
	 */
	spin_lock_bh(&rcp->lock);
	if (rcp->cur != rcp->completed)
		cpu_quiet(rdp->cpu, rcp);
	spin_unlock_bh(&rcp->lock);
	rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail);
	rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail);
	rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail);
}

static void rcu_offline_cpu(int cpu)
{
	struct rcu_data *this_rdp = &get_cpu_var(rcu_data);
	struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data);

	__rcu_offline_cpu(this_rdp, &rcu_ctrlblk,
					&per_cpu(rcu_data, cpu));
	__rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk,
					&per_cpu(rcu_bh_data, cpu));
	put_cpu_var(rcu_data);
	put_cpu_var(rcu_bh_data);
	tasklet_kill_immediate(&per_cpu(rcu_tasklet, cpu), cpu);
}

#else

static void rcu_offline_cpu(int cpu)
{
}

#endif

/*
 * This does the RCU processing work from tasklet context. 
 */
static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp,
					struct rcu_data *rdp)
{
	if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) {
		*rdp->donetail = rdp->curlist;
		rdp->donetail = rdp->curtail;
		rdp->curlist = NULL;
		rdp->curtail = &rdp->curlist;
	}

	local_irq_disable();
	if (rdp->nxtlist && !rdp->curlist) {
		rdp->curlist = rdp->nxtlist;
		rdp->curtail = rdp->nxttail;
		rdp->nxtlist = NULL;
		rdp->nxttail = &rdp->nxtlist;
		local_irq_enable();

		/*
		 * start the next batch of callbacks
		 */

		/* determine batch number */
		rdp->batch = rcp->cur + 1;
		/* see the comment and corresponding wmb() in
		 * the rcu_start_batch()
		 */
		smp_rmb();

		if (!rcp->next_pending) {
			/* and start it/schedule start if it's a new batch */
			spin_lock(&rcp->lock);
			rcp->next_pending = 1;
			rcu_start_batch(rcp);
			spin_unlock(&rcp->lock);
		}
	} else {
		local_irq_enable();
	}
	rcu_check_quiescent_state(rcp, rdp);
	if (rdp->donelist)
		rcu_do_batch(rdp);
}

static void rcu_process_callbacks(unsigned long unused)
{
	__rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data));
	__rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data));
}

static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp)
{
	/* This cpu has pending rcu entries and the grace period
	 * for them has completed.
	 */
	if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch))
		return 1;

	/* This cpu has no pending entries, but there are new entries */
	if (!rdp->curlist && rdp->nxtlist)
		return 1;

	/* This cpu has finished callbacks to invoke */
	if (rdp->donelist)
		return 1;

	/* The rcu core waits for a quiescent state from the cpu */
	if (rdp->quiescbatch != rcp->cur || rdp->qs_pending)
		return 1;

	/* nothing to do */
	return 0;
}

int rcu_pending(int cpu)
{
	return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) ||
		__rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu));
}

void rcu_check_callbacks(int cpu, int user)
{
	if (user || 
	    (idle_cpu(cpu) && !in_softirq() && 
				hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
		rcu_qsctr_inc(cpu);
		rcu_bh_qsctr_inc(cpu);
	} else if (!in_softirq())
		rcu_bh_qsctr_inc(cpu);
	tasklet_schedule(&per_cpu(rcu_tasklet, cpu));
}

static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp,
						struct rcu_data *rdp)
{
	memset(rdp, 0, sizeof(*rdp));
	rdp->curtail = &rdp->curlist;
	rdp->nxttail = &rdp->nxtlist;
	rdp->donetail = &rdp->donelist;
	rdp->quiescbatch = rcp->completed;
	rdp->qs_pending = 0;
	rdp->cpu = cpu;
	rdp->blimit = blimit;
}

static void __devinit rcu_online_cpu(int cpu)
{
	struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
	struct rcu_data *bh_rdp = &per_cpu(rcu_bh_data, cpu);

	rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp);
	rcu_init_percpu_data(cpu, &rcu_bh_ctrlblk, bh_rdp);
	tasklet_init(&per_cpu(rcu_tasklet, cpu), rcu_process_callbacks, 0UL);
}

static int __devinit rcu_cpu_notify(struct notifier_block *self, 
				unsigned long action, void *hcpu)
{
	long cpu = (long)hcpu;
	switch (action) {
	case CPU_UP_PREPARE:
		rcu_online_cpu(cpu);
		break;
	case CPU_DEAD:
		rcu_offline_cpu(cpu);
		break;
	default:
		break;
	}
	return NOTIFY_OK;
}

static struct notifier_block __devinitdata rcu_nb = {
	.notifier_call	= rcu_cpu_notify,
};

/*
 * Initializes rcu mechanism.  Assumed to be called early.
 * That is before local timer(SMP) or jiffie timer (uniproc) is setup.
 * Note that rcu_qsctr and friends are implicitly
 * initialized due to the choice of ``0'' for RCU_CTR_INVALID.
 */
void __init rcu_init(void)
{
	sema_init(&rcu_barrier_sema, 1);
	rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE,
			(void *)(long)smp_processor_id());
	/* Register notifier for non-boot CPUs */
	register_cpu_notifier(&rcu_nb);
}

struct rcu_synchronize {
	struct rcu_head head;
	struct completion completion;
};

/* Because of FASTCALL declaration of complete, we use this wrapper */
static void wakeme_after_rcu(struct rcu_head  *head)
{
	struct rcu_synchronize *rcu;

	rcu = container_of(head, struct rcu_synchronize, head);
	complete(&rcu->completion);
}

/**
 * synchronize_rcu - wait until a grace period has elapsed.
 *
 * Control will return to the caller some time after a full grace
 * period has elapsed, in other words after all currently executing RCU
 * read-side critical sections have completed.  RCU read-side critical
 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
 * and may be nested.
 *
 * If your read-side code is not protected by rcu_read_lock(), do -not-
 * use synchronize_rcu().
 */
void synchronize_rcu(void)
{
	struct rcu_synchronize rcu;

	init_completion(&rcu.completion);
	/* Will wake me after RCU finished */
	call_rcu(&rcu.head, wakeme_after_rcu);

	/* Wait for it */
	wait_for_completion(&rcu.completion);
}

/*
 * Deprecated, use synchronize_rcu() or synchronize_sched() instead.
 */
void synchronize_kernel(void)
{
	synchronize_rcu();
}

module_param(blimit, int, 0);
module_param(qhimark, int, 0);
module_param(qlowmark, int, 0);
#ifdef CONFIG_SMP
module_param(rsinterval, int, 0);
#endif
EXPORT_SYMBOL_GPL(rcu_batches_completed);
EXPORT_SYMBOL(call_rcu);  /* WARNING: GPL-only in April 2006. */
EXPORT_SYMBOL(call_rcu_bh);  /* WARNING: GPL-only in April 2006. */
EXPORT_SYMBOL_GPL(synchronize_rcu);
EXPORT_SYMBOL(synchronize_kernel);  /* WARNING: GPL-only in April 2006. */
Commit	Line	Data
1da177e4 LT	1	/*
	2	* Read-Copy Update mechanism for mutual exclusion
	3	*
	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.
	8	*
	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.
	13	*
	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.
	17	*
	18	* Copyright (C) IBM Corporation, 2001
	19	*
	20	* Authors: Dipankar Sarma <dipankar@in.ibm.com>
	21	* Manfred Spraul <manfred@colorfullife.com>
	22	*
	23	* Based on the original work by Paul McKenney <paulmck@us.ibm.com>
	24	* and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
	25	* Papers:
	26	* http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
	27	* http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
	28	*
	29	* For detailed explanation of Read-Copy Update mechanism see -
	30	* http://lse.sourceforge.net/locking/rcupdate.html
	31	*
	32	*/
	33	#include <linux/types.h>
	34	#include <linux/kernel.h>
	35	#include <linux/init.h>
	36	#include <linux/spinlock.h>
	37	#include <linux/smp.h>
e56d0903	38	#include <linux/rcupdate.h>
1da177e4 LT	39	#include <linux/interrupt.h>
	40	#include <linux/sched.h>
	41	#include <asm/atomic.h>
	42	#include <linux/bitops.h>
	43	#include <linux/module.h>
	44	#include <linux/completion.h>
	45	#include <linux/moduleparam.h>
	46	#include <linux/percpu.h>
	47	#include <linux/notifier.h>
	48	#include <linux/rcupdate.h>
	49	#include <linux/cpu.h>
	50
	51	/* Definition for rcupdate control block. */
69a0b315 ON	52	struct rcu_ctrlblk rcu_ctrlblk = {
	53	.cur = -300,
	54	.completed = -300,
	55	.lock = SPIN_LOCK_UNLOCKED,
	56	.cpumask = CPU_MASK_NONE,
	57	};
	58	struct rcu_ctrlblk rcu_bh_ctrlblk = {
	59	.cur = -300,
	60	.completed = -300,
	61	.lock = SPIN_LOCK_UNLOCKED,
	62	.cpumask = CPU_MASK_NONE,
1da177e4	63	};
1da177e4 LT	64
	65	DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L };
	66	DEFINE_PER_CPU(struct rcu_data, rcu_bh_data) = { 0L };
	67
	68	/* Fake initialization required by compiler */
	69	static DEFINE_PER_CPU(struct tasklet_struct, rcu_tasklet) = {NULL};
21a1ea9e DS	70	static int blimit = 10;
	71	static int qhimark = 10000;
	72	static int qlowmark = 100;
	73	#ifdef CONFIG_SMP
	74	static int rsinterval = 1000;
	75	#endif
	76
	77	static atomic_t rcu_barrier_cpu_count;
	78	static struct semaphore rcu_barrier_sema;
	79	static struct completion rcu_barrier_completion;
	80
	81	#ifdef CONFIG_SMP
	82	static void force_quiescent_state(struct rcu_data *rdp,
	83	struct rcu_ctrlblk *rcp)
	84	{
	85	int cpu;
	86	cpumask_t cpumask;
	87	set_need_resched();
	88	if (unlikely(rdp->qlen - rdp->last_rs_qlen > rsinterval)) {
	89	rdp->last_rs_qlen = rdp->qlen;
	90	/*
	91	* Don't send IPI to itself. With irqs disabled,
	92	* rdp->cpu is the current cpu.
	93	*/
	94	cpumask = rcp->cpumask;
	95	cpu_clear(rdp->cpu, cpumask);
	96	for_each_cpu_mask(cpu, cpumask)
	97	smp_send_reschedule(cpu);
	98	}
	99	}
	100	#else
	101	static inline void force_quiescent_state(struct rcu_data *rdp,
	102	struct rcu_ctrlblk *rcp)
	103	{
	104	set_need_resched();
	105	}
	106	#endif
1da177e4 LT	107
	108	/**
	109	* call_rcu - Queue an RCU callback for invocation after a grace period.
	110	* @head: structure to be used for queueing the RCU updates.
	111	* @func: actual update function to be invoked after the grace period
	112	*
	113	* The update function will be invoked some time after a full grace
	114	* period elapses, in other words after all currently executing RCU
	115	* read-side critical sections have completed. RCU read-side critical
	116	* sections are delimited by rcu_read_lock() and rcu_read_unlock(),
	117	* and may be nested.
	118	*/
	119	void fastcall call_rcu(struct rcu_head *head,
	120	void (func)(struct rcu_head rcu))
	121	{
	122	unsigned long flags;
	123	struct rcu_data *rdp;
	124
	125	head->func = func;
	126	head->next = NULL;
	127	local_irq_save(flags);
	128	rdp = &__get_cpu_var(rcu_data);
	129	*rdp->nxttail = head;
	130	rdp->nxttail = &head->next;
21a1ea9e DS	131	if (unlikely(++rdp->qlen > qhimark)) {
	132	rdp->blimit = INT_MAX;
	133	force_quiescent_state(rdp, &rcu_ctrlblk);
	134	}
1da177e4 LT	135	local_irq_restore(flags);
	136	}
	137
	138	/**
	139	* call_rcu_bh - Queue an RCU for invocation after a quicker grace period.
	140	* @head: structure to be used for queueing the RCU updates.
	141	* @func: actual update function to be invoked after the grace period
	142	*
	143	* The update function will be invoked some time after a full grace
	144	* period elapses, in other words after all currently executing RCU
	145	* read-side critical sections have completed. call_rcu_bh() assumes
	146	* that the read-side critical sections end on completion of a softirq
	147	* handler. This means that read-side critical sections in process
	148	* context must not be interrupted by softirqs. This interface is to be
	149	* used when most of the read-side critical sections are in softirq context.
	150	* RCU read-side critical sections are delimited by rcu_read_lock() and
	151	* rcu_read_unlock(), * if in interrupt context or rcu_read_lock_bh()
	152	* and rcu_read_unlock_bh(), if in process context. These may be nested.
	153	*/
	154	void fastcall call_rcu_bh(struct rcu_head *head,
	155	void (func)(struct rcu_head rcu))
	156	{
	157	unsigned long flags;
	158	struct rcu_data *rdp;
	159
	160	head->func = func;
	161	head->next = NULL;
	162	local_irq_save(flags);
	163	rdp = &__get_cpu_var(rcu_bh_data);
	164	*rdp->nxttail = head;
	165	rdp->nxttail = &head->next;
21a1ea9e DS	166
	167	if (unlikely(++rdp->qlen > qhimark)) {
	168	rdp->blimit = INT_MAX;
	169	force_quiescent_state(rdp, &rcu_bh_ctrlblk);
	170	}
	171
1da177e4 LT	172	local_irq_restore(flags);
	173	}
	174
a241ec65 PM	175	/*
	176	* Return the number of RCU batches processed thus far. Useful
	177	* for debug and statistics.
	178	*/
	179	long rcu_batches_completed(void)
	180	{
	181	return rcu_ctrlblk.completed;
	182	}
	183
ab4720ec DS	184	static void rcu_barrier_callback(struct rcu_head *notused)
	185	{
	186	if (atomic_dec_and_test(&rcu_barrier_cpu_count))
	187	complete(&rcu_barrier_completion);
	188	}
	189
	190	/*
	191	* Called with preemption disabled, and from cross-cpu IRQ context.
	192	*/
	193	static void rcu_barrier_func(void *notused)
	194	{
	195	int cpu = smp_processor_id();
	196	struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
	197	struct rcu_head *head;
	198
	199	head = &rdp->barrier;
	200	atomic_inc(&rcu_barrier_cpu_count);
	201	call_rcu(head, rcu_barrier_callback);
	202	}
	203
	204	/**
	205	* rcu_barrier - Wait until all the in-flight RCUs are complete.
	206	*/
	207	void rcu_barrier(void)
	208	{
	209	BUG_ON(in_interrupt());
	210	/* Take cpucontrol semaphore to protect against CPU hotplug */
	211	down(&rcu_barrier_sema);
	212	init_completion(&rcu_barrier_completion);
	213	atomic_set(&rcu_barrier_cpu_count, 0);
	214	on_each_cpu(rcu_barrier_func, NULL, 0, 1);
	215	wait_for_completion(&rcu_barrier_completion);
	216	up(&rcu_barrier_sema);
	217	}
	218	EXPORT_SYMBOL_GPL(rcu_barrier);
	219
1da177e4 LT	220	/*
	221	* Invoke the completed RCU callbacks. They are expected to be in
	222	* a per-cpu list.
	223	*/
	224	static void rcu_do_batch(struct rcu_data *rdp)
	225	{
	226	struct rcu_head next, list;
	227	int count = 0;
	228
	229	list = rdp->donelist;
	230	while (list) {
	231	next = rdp->donelist = list->next;
	232	list->func(list);
	233	list = next;
21a1ea9e DS	234	rdp->qlen--;
21a1ea9e DS	235	if (++count >= rdp->blimit)
1da177e4 LT	236	break;
1da177e4 LT	237	}
21a1ea9e DS	238	if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark)
21a1ea9e DS	239	rdp->blimit = blimit;
1da177e4 LT	240	if (!rdp->donelist)
	241	rdp->donetail = &rdp->donelist;
	242	else
	243	tasklet_schedule(&per_cpu(rcu_tasklet, rdp->cpu));
	244	}
	245
	246	/*
	247	* Grace period handling:
	248	* The grace period handling consists out of two steps:
	249	* - A new grace period is started.
	250	* This is done by rcu_start_batch. The start is not broadcasted to
	251	* all cpus, they must pick this up by comparing rcp->cur with
	252	* rdp->quiescbatch. All cpus are recorded in the
69a0b315	253	* rcu_ctrlblk.cpumask bitmap.
1da177e4 LT	254	* - All cpus must go through a quiescent state.
	255	* Since the start of the grace period is not broadcasted, at least two
	256	* calls to rcu_check_quiescent_state are required:
	257	* The first call just notices that a new grace period is running. The
	258	* following calls check if there was a quiescent state since the beginning
69a0b315	259	* of the grace period. If so, it updates rcu_ctrlblk.cpumask. If
1da177e4 LT	260	* the bitmap is empty, then the grace period is completed.
	261	* rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace
	262	* period (if necessary).
	263	*/
	264	/*
	265	* Register a new batch of callbacks, and start it up if there is currently no
	266	* active batch and the batch to be registered has not already occurred.
69a0b315	267	* Caller must hold rcu_ctrlblk.lock.
1da177e4	268	*/
69a0b315	269	static void rcu_start_batch(struct rcu_ctrlblk *rcp)
1da177e4	270	{
1da177e4 LT	271	if (rcp->next_pending &&
1da177e4 LT	272	rcp->completed == rcp->cur) {
1da177e4	273	rcp->next_pending = 0;
c3f59023 SV	274	/*
	275	* next_pending == 0 must be visible in
	276	* __rcu_process_callbacks() before it can see new value of cur.
1da177e4 LT	277	*/
	278	smp_wmb();
	279	rcp->cur++;
c3f59023 SV	280
	281	/*
	282	* Accessing nohz_cpu_mask before incrementing rcp->cur needs a
	283	* Barrier Otherwise it can cause tickless idle CPUs to be
69a0b315	284	* included in rcp->cpumask, which will extend graceperiods
c3f59023 SV	285	* unnecessarily.
	286	*/
	287	smp_mb();
69a0b315	288	cpus_andnot(rcp->cpumask, cpu_online_map, nohz_cpu_mask);
c3f59023	289
1da177e4 LT	290	}
	291	}
	292
	293	/*
	294	* cpu went through a quiescent state since the beginning of the grace period.
	295	* Clear it from the cpu mask and complete the grace period if it was the last
	296	* cpu. Start another grace period if someone has further entries pending
	297	*/
69a0b315	298	static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp)
1da177e4	299	{
69a0b315 ON	300	cpu_clear(cpu, rcp->cpumask);
69a0b315 ON	301	if (cpus_empty(rcp->cpumask)) {
1da177e4 LT	302	/* batch completed ! */
1da177e4 LT	303	rcp->completed = rcp->cur;
69a0b315	304	rcu_start_batch(rcp);
1da177e4 LT	305	}
	306	}
	307
	308	/*
	309	* Check if the cpu has gone through a quiescent state (say context
	310	* switch). If so and if it already hasn't done so in this RCU
	311	* quiescent cycle, then indicate that it has done so.
	312	*/
	313	static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp,
69a0b315	314	struct rcu_data *rdp)
1da177e4 LT	315	{
	316	if (rdp->quiescbatch != rcp->cur) {
	317	/* start new grace period: */
	318	rdp->qs_pending = 1;
	319	rdp->passed_quiesc = 0;
	320	rdp->quiescbatch = rcp->cur;
	321	return;
	322	}
	323
	324	/* Grace period already completed for this cpu?
	325	* qs_pending is checked instead of the actual bitmap to avoid
	326	* cacheline trashing.
	327	*/
	328	if (!rdp->qs_pending)
	329	return;
	330
	331	/*
	332	* Was there a quiescent state since the beginning of the grace
	333	* period? If no, then exit and wait for the next call.
	334	*/
	335	if (!rdp->passed_quiesc)
	336	return;
	337	rdp->qs_pending = 0;
	338
69a0b315	339	spin_lock(&rcp->lock);
1da177e4 LT	340	/*
	341	* rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync
	342	* during cpu startup. Ignore the quiescent state.
	343	*/
	344	if (likely(rdp->quiescbatch == rcp->cur))
69a0b315	345	cpu_quiet(rdp->cpu, rcp);
1da177e4	346
69a0b315	347	spin_unlock(&rcp->lock);
1da177e4 LT	348	}
	349
	350
	351	#ifdef CONFIG_HOTPLUG_CPU
	352
	353	/* warning! helper for rcu_offline_cpu. do not use elsewhere without reviewing
	354	* locking requirements, the list it's pulling from has to belong to a cpu
	355	* which is dead and hence not processing interrupts.
	356	*/
	357	static void rcu_move_batch(struct rcu_data this_rdp, struct rcu_head list,
	358	struct rcu_head **tail)
	359	{
	360	local_irq_disable();
	361	*this_rdp->nxttail = list;
	362	if (list)
	363	this_rdp->nxttail = tail;
	364	local_irq_enable();
	365	}
	366
	367	static void __rcu_offline_cpu(struct rcu_data *this_rdp,
69a0b315	368	struct rcu_ctrlblk rcp, struct rcu_data rdp)
1da177e4 LT	369	{
	370	/* if the cpu going offline owns the grace period
	371	* we can block indefinitely waiting for it, so flush
	372	* it here
	373	*/
69a0b315	374	spin_lock_bh(&rcp->lock);
1da177e4	375	if (rcp->cur != rcp->completed)
69a0b315 ON	376	cpu_quiet(rdp->cpu, rcp);
69a0b315 ON	377	spin_unlock_bh(&rcp->lock);
1da177e4 LT	378	rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail);
1da177e4 LT	379	rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail);
a9c82815	380	rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail);
1da177e4	381	}
a9c82815	382
1da177e4 LT	383	static void rcu_offline_cpu(int cpu)
	384	{
	385	struct rcu_data *this_rdp = &get_cpu_var(rcu_data);
	386	struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data);
	387
69a0b315	388	__rcu_offline_cpu(this_rdp, &rcu_ctrlblk,
1da177e4	389	&per_cpu(rcu_data, cpu));
69a0b315	390	__rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk,
1da177e4 LT	391	&per_cpu(rcu_bh_data, cpu));
	392	put_cpu_var(rcu_data);
	393	put_cpu_var(rcu_bh_data);
	394	tasklet_kill_immediate(&per_cpu(rcu_tasklet, cpu), cpu);
	395	}
	396
	397	#else
	398
	399	static void rcu_offline_cpu(int cpu)
	400	{
	401	}
	402
	403	#endif
	404
	405	/*
	406	* This does the RCU processing work from tasklet context.
	407	*/
	408	static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp,
69a0b315	409	struct rcu_data *rdp)
1da177e4 LT	410	{
	411	if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) {
	412	*rdp->donetail = rdp->curlist;
	413	rdp->donetail = rdp->curtail;
	414	rdp->curlist = NULL;
	415	rdp->curtail = &rdp->curlist;
	416	}
	417
	418	local_irq_disable();
	419	if (rdp->nxtlist && !rdp->curlist) {
	420	rdp->curlist = rdp->nxtlist;
	421	rdp->curtail = rdp->nxttail;
	422	rdp->nxtlist = NULL;
	423	rdp->nxttail = &rdp->nxtlist;
	424	local_irq_enable();
	425
	426	/*
	427	* start the next batch of callbacks
	428	*/
	429
	430	/* determine batch number */
	431	rdp->batch = rcp->cur + 1;
	432	/* see the comment and corresponding wmb() in
	433	* the rcu_start_batch()
	434	*/
	435	smp_rmb();
	436
	437	if (!rcp->next_pending) {
	438	/* and start it/schedule start if it's a new batch */
69a0b315	439	spin_lock(&rcp->lock);
dbc1651f	440	rcp->next_pending = 1;
69a0b315 ON	441	rcu_start_batch(rcp);
69a0b315 ON	442	spin_unlock(&rcp->lock);
1da177e4 LT	443	}
	444	} else {
	445	local_irq_enable();
	446	}
69a0b315	447	rcu_check_quiescent_state(rcp, rdp);
1da177e4 LT	448	if (rdp->donelist)
	449	rcu_do_batch(rdp);
	450	}
	451
	452	static void rcu_process_callbacks(unsigned long unused)
	453	{
69a0b315 ON	454	__rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data));
69a0b315 ON	455	__rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data));
1da177e4 LT	456	}
1da177e4 LT	457
67751777 ON	458	static int __rcu_pending(struct rcu_ctrlblk rcp, struct rcu_data rdp)
	459	{
	460	/* This cpu has pending rcu entries and the grace period
	461	* for them has completed.
	462	*/
	463	if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch))
	464	return 1;
	465
	466	/* This cpu has no pending entries, but there are new entries */
	467	if (!rdp->curlist && rdp->nxtlist)
	468	return 1;
	469
	470	/* This cpu has finished callbacks to invoke */
	471	if (rdp->donelist)
	472	return 1;
	473
	474	/* The rcu core waits for a quiescent state from the cpu */
	475	if (rdp->quiescbatch != rcp->cur \|\| rdp->qs_pending)
	476	return 1;
	477
	478	/* nothing to do */
	479	return 0;
	480	}
	481
	482	int rcu_pending(int cpu)
	483	{
	484	return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) \|\|
	485	__rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu));
	486	}
	487
1da177e4 LT	488	void rcu_check_callbacks(int cpu, int user)
	489	{
	490	if (user \|\|
	491	(idle_cpu(cpu) && !in_softirq() &&
	492	hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
	493	rcu_qsctr_inc(cpu);
	494	rcu_bh_qsctr_inc(cpu);
	495	} else if (!in_softirq())
	496	rcu_bh_qsctr_inc(cpu);
	497	tasklet_schedule(&per_cpu(rcu_tasklet, cpu));
	498	}
	499
	500	static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp,
	501	struct rcu_data *rdp)
	502	{
	503	memset(rdp, 0, sizeof(*rdp));
	504	rdp->curtail = &rdp->curlist;
	505	rdp->nxttail = &rdp->nxtlist;
	506	rdp->donetail = &rdp->donelist;
	507	rdp->quiescbatch = rcp->completed;
	508	rdp->qs_pending = 0;
	509	rdp->cpu = cpu;
21a1ea9e	510	rdp->blimit = blimit;
1da177e4 LT	511	}
	512
	513	static void __devinit rcu_online_cpu(int cpu)
	514	{
	515	struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
	516	struct rcu_data *bh_rdp = &per_cpu(rcu_bh_data, cpu);
	517
	518	rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp);
	519	rcu_init_percpu_data(cpu, &rcu_bh_ctrlblk, bh_rdp);
	520	tasklet_init(&per_cpu(rcu_tasklet, cpu), rcu_process_callbacks, 0UL);
	521	}
	522
	523	static int __devinit rcu_cpu_notify(struct notifier_block *self,
	524	unsigned long action, void *hcpu)
	525	{
	526	long cpu = (long)hcpu;
	527	switch (action) {
	528	case CPU_UP_PREPARE:
	529	rcu_online_cpu(cpu);
	530	break;
	531	case CPU_DEAD:
	532	rcu_offline_cpu(cpu);
	533	break;
	534	default:
	535	break;
	536	}
	537	return NOTIFY_OK;
	538	}
	539
	540	static struct notifier_block __devinitdata rcu_nb = {
	541	.notifier_call = rcu_cpu_notify,
	542	};
	543
	544	/*
	545	* Initializes rcu mechanism. Assumed to be called early.
	546	* That is before local timer(SMP) or jiffie timer (uniproc) is setup.
	547	* Note that rcu_qsctr and friends are implicitly
	548	* initialized due to the choice of ``0'' for RCU_CTR_INVALID.
	549	*/
	550	void __init rcu_init(void)
	551	{
ab4720ec	552	sema_init(&rcu_barrier_sema, 1);
1da177e4 LT	553	rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE,
	554	(void *)(long)smp_processor_id());
	555	/* Register notifier for non-boot CPUs */
	556	register_cpu_notifier(&rcu_nb);
	557	}
	558
	559	struct rcu_synchronize {
	560	struct rcu_head head;
	561	struct completion completion;
	562	};
	563
	564	/* Because of FASTCALL declaration of complete, we use this wrapper */
	565	static void wakeme_after_rcu(struct rcu_head *head)
	566	{
	567	struct rcu_synchronize *rcu;
	568
	569	rcu = container_of(head, struct rcu_synchronize, head);
	570	complete(&rcu->completion);
	571	}
	572
	573	/**
9b06e818	574	* synchronize_rcu - wait until a grace period has elapsed.
1da177e4 LT	575	*
	576	* Control will return to the caller some time after a full grace
	577	* period has elapsed, in other words after all currently executing RCU
	578	* read-side critical sections have completed. RCU read-side critical
	579	* sections are delimited by rcu_read_lock() and rcu_read_unlock(),
	580	* and may be nested.
9b06e818 PM	581	*
	582	* If your read-side code is not protected by rcu_read_lock(), do -not-
	583	* use synchronize_rcu().
1da177e4	584	*/
9b06e818	585	void synchronize_rcu(void)
1da177e4 LT	586	{
	587	struct rcu_synchronize rcu;
	588
	589	init_completion(&rcu.completion);
	590	/* Will wake me after RCU finished */
	591	call_rcu(&rcu.head, wakeme_after_rcu);
	592
	593	/* Wait for it */
	594	wait_for_completion(&rcu.completion);
	595	}
	596
9b06e818 PM	597	/*
	598	* Deprecated, use synchronize_rcu() or synchronize_sched() instead.
	599	*/
	600	void synchronize_kernel(void)
	601	{
	602	synchronize_rcu();
	603	}
	604
21a1ea9e DS	605	module_param(blimit, int, 0);
	606	module_param(qhimark, int, 0);
	607	module_param(qlowmark, int, 0);
	608	#ifdef CONFIG_SMP
	609	module_param(rsinterval, int, 0);
	610	#endif
a241ec65	611	EXPORT_SYMBOL_GPL(rcu_batches_completed);
66cf8f14 PM	612	EXPORT_SYMBOL(call_rcu); /* WARNING: GPL-only in April 2006. */
66cf8f14 PM	613	EXPORT_SYMBOL(call_rcu_bh); /* WARNING: GPL-only in April 2006. */
9b06e818	614	EXPORT_SYMBOL_GPL(synchronize_rcu);
66cf8f14	615	EXPORT_SYMBOL(synchronize_kernel); /* WARNING: GPL-only in April 2006. */