"Tree RCU": scalable classic RCU implementation

This patch fixes a long-standing performance bug in classic RCU that
results in massive internal-to-RCU lock contention on systems with
more than a few hundred CPUs.  Although this patch creates a separate
flavor of RCU for ease of review and patch maintenance, it is intended
to replace classic RCU.

This patch still handles stress better than does mainline, so I am still
calling it ready for inclusion.  This patch is against the -tip tree.
Nevertheless, experience on an actual 1000+ CPU machine would still be
most welcome.

Most of the changes noted below were found while creating an rcutiny
(which should permit ejecting the current rcuclassic) and while doing
detailed line-by-line documentation.

Updates from v9 (http://lkml.org/lkml/2008/12/2/334):

o	Fixes from remainder of line-by-line code walkthrough,
	including comment spelling, initialization, undesirable
	narrowing due to type conversion, removing redundant memory
	barriers, removing redundant local-variable initialization,
	and removing redundant local variables.

	I do not believe that any of these fixes address the CPU-hotplug
	issues that Andi Kleen was seeing, but please do give it a whirl
	in case the machine is smarter than I am.

	A writeup from the walkthrough may be found at the following
	URL, in case you are suffering from terminal insomnia or
	masochism:

	http://www.kernel.org/pub/linux/kernel/people/paulmck/tmp/rcutree-walkthrough.2008.12.16a.pdf

o	Made rcutree tracing use seq_file, as suggested some time
	ago by Lai Jiangshan.

o	Added a .csv variant of the rcudata debugfs trace file, to allow
	people having thousands of CPUs to drop the data into
	a spreadsheet.	Tested with oocalc and gnumeric.  Updated
	documentation to suit.

Updates from v8 (http://lkml.org/lkml/2008/11/15/139):

o	Fix a theoretical race between grace-period initialization and
	force_quiescent_state() that could occur if more than three
	jiffies were required to carry out the grace-period
	initialization.  Which it might, if you had enough CPUs.

o	Apply Ingo's printk-standardization patch.

o	Substitute local variables for repeated accesses to global
	variables.

o	Fix comment misspellings and redundant (but harmless) increments
	of ->n_rcu_pending (this latter after having explicitly added it).

o	Apply checkpatch fixes.

Updates from v7 (http://lkml.org/lkml/2008/10/10/291):

o	Fixed a number of problems noted by Gautham Shenoy, including
	the cpu-stall-detection bug that he was having difficulty
	convincing me was real.  ;-)

o	Changed cpu-stall detection to wait for ten seconds rather than
	three in order to reduce false positive, as suggested by Ingo
	Molnar.

o	Produced a design document (http://lwn.net/Articles/305782/).
	The act of writing this document uncovered a number of both
	theoretical and "here and now" bugs as noted below.

o	Fix dynticks_nesting accounting confusion, simplify WARN_ON()
	condition, fix kerneldoc comments, and add memory barriers
	in dynticks interface functions.

o	Add more data to tracing.

o	Remove unused "rcu_barrier" field from rcu_data structure.

o	Count calls to rcu_pending() from scheduling-clock interrupt
	to use as a surrogate timebase should jiffies stop counting.

o	Fix a theoretical race between force_quiescent_state() and
	grace-period initialization.  Yes, initialization does have to
	go on for some jiffies for this race to occur, but given enough
	CPUs...

Updates from v6 (http://lkml.org/lkml/2008/9/23/448):

o	Fix a number of checkpatch.pl complaints.

o	Apply review comments from Ingo Molnar and Lai Jiangshan
	on the stall-detection code.

o	Fix several bugs in !CONFIG_SMP builds.

o	Fix a misspelled config-parameter name so that RCU now announces
	at boot time if stall detection is configured.

o	Run tests on numerous combinations of configurations parameters,
	which after the fixes above, now build and run correctly.

Updates from v5 (http://lkml.org/lkml/2008/9/15/92, bad subject line):

o	Fix a compiler error in the !CONFIG_FANOUT_EXACT case (blew a
	changeset some time ago, and finally got around to retesting
	this option).

o	Fix some tracing bugs in rcupreempt that caused incorrect
	totals to be printed.

o	I now test with a more brutal random-selection online/offline
	script (attached).  Probably more brutal than it needs to be
	on the people reading it as well, but so it goes.

o	A number of optimizations and usability improvements:

	o	Make rcu_pending() ignore the grace-period timeout when
		there is no grace period in progress.

	o	Make force_quiescent_state() avoid going for a global
		lock in the case where there is no grace period in
		progress.

	o	Rearrange struct fields to improve struct layout.

	o	Make call_rcu() initiate a grace period if RCU was
		idle, rather than waiting for the next scheduling
		clock interrupt.

	o	Invoke rcu_irq_enter() and rcu_irq_exit() only when
		idle, as suggested by Andi Kleen.  I still don't
		completely trust this change, and might back it out.

	o	Make CONFIG_RCU_TRACE be the single config variable
		manipulated for all forms of RCU, instead of the prior
		confusion.

	o	Document tracing files and formats for both rcupreempt
		and rcutree.

Updates from v4 for those missing v5 given its bad subject line:

o	Separated dynticks interface so that NMIs and irqs call separate
	functions, greatly simplifying it.  In particular, this code
	no longer requires a proof of correctness.  ;-)

o	Separated dynticks state out into its own per-CPU structure,
	avoiding the duplicated accounting.

o	The case where a dynticks-idle CPU runs an irq handler that
	invokes call_rcu() is now correctly handled, forcing that CPU
	out of dynticks-idle mode.

o	Review comments have been applied (thank you all!!!).
	For but one example, fixed the dynticks-ordering issue that
	Manfred pointed out, saving me much debugging.  ;-)

o	Adjusted rcuclassic and rcupreempt to handle dynticks changes.

Attached is an updated patch to Classic RCU that applies a hierarchy,
greatly reducing the contention on the top-level lock for large machines.
This passes 10-hour concurrent rcutorture and online-offline testing on
128-CPU ppc64 without dynticks enabled, and exposes some timekeeping
bugs in presence of dynticks (exciting working on a system where
"sleep 1" hangs until interrupted...), which were fixed in the
2.6.27 kernel.  It is getting more reliable than mainline by some
measures, so the next version will be against -tip for inclusion.
See also Manfred Spraul's recent patches (or his earlier work from
2004 at http://marc.info/?l=linux-kernel&m=108546384711797&w=2).
We will converge onto a common patch in the fullness of time, but are
currently exploring different regions of the design space.  That said,
I have already gratefully stolen quite a few of Manfred's ideas.

This patch provides CONFIG_RCU_FANOUT, which controls the bushiness
of the RCU hierarchy.  Defaults to 32 on 32-bit machines and 64 on
64-bit machines.  If CONFIG_NR_CPUS is less than CONFIG_RCU_FANOUT,
there is no hierarchy.  By default, the RCU initialization code will
adjust CONFIG_RCU_FANOUT to balance the hierarchy, so strongly NUMA
architectures may choose to set CONFIG_RCU_FANOUT_EXACT to disable
this balancing, allowing the hierarchy to be exactly aligned to the
underlying hardware.  Up to two levels of hierarchy are permitted
(in addition to the root node), allowing up to 16,384 CPUs on 32-bit
systems and up to 262,144 CPUs on 64-bit systems.  I just know that I
am going to regret saying this, but this seems more than sufficient
for the foreseeable future.  (Some architectures might wish to set
CONFIG_RCU_FANOUT=4, which would limit such architectures to 64 CPUs.
If this becomes a real problem, additional levels can be added, but I
doubt that it will make a significant difference on real hardware.)

In the common case, a given CPU will manipulate its private rcu_data
structure and the rcu_node structure that it shares with its immediate
neighbors.  This can reduce both lock and memory contention by multiple
orders of magnitude, which should eliminate the need for the strange
manipulations that are reported to be required when running Linux on
very large systems.

Some shortcomings:

o	More bugs will probably surface as a result of an ongoing
	line-by-line code inspection.

	Patches will be provided as required.

o	There are probably hangs, rcutorture failures, &c.  Seems
	quite stable on a 128-CPU machine, but that is kind of small
	compared to 4096 CPUs.  However, seems to do better than
	mainline.

	Patches will be provided as required.

o	The memory footprint of this version is several KB larger
	than rcuclassic.

	A separate UP-only rcutiny patch will be provided, which will
	reduce the memory footprint significantly, even compared
	to the old rcuclassic.  One such patch passes light testing,
	and has a memory footprint smaller even than rcuclassic.
	Initial reaction from various embedded guys was "it is not
	worth it", so am putting it aside.

Credits:

o	Manfred Spraul for ideas, review comments, and bugs spotted,
	as well as some good friendly competition.  ;-)

o	Josh Triplett, Ingo Molnar, Peter Zijlstra, Mathieu Desnoyers,
	Lai Jiangshan, Andi Kleen, Andy Whitcroft, and Andrew Morton
	for reviews and comments.

o	Thomas Gleixner for much-needed help with some timer issues
	(see patches below).

o	Jon M. Tollefson, Tim Pepper, Andrew Theurer, Jose R. Santos,
	Andy Whitcroft, Darrick Wong, Nishanth Aravamudan, Anton
	Blanchard, Dave Kleikamp, and Nathan Lynch for keeping machines
	alive despite my heavy abuse^Wtesting.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

1 files changed, 1535 insertions, 0 deletions

diff --git a/kernel/rcutree.c b/kernel/rcutree.c
new file mode 100644
index 000000000000..a342b032112c
--- /dev/null
+++ b/kernel/rcutree.c
@@ -0,0 +1,1535 @@
+/*
+ * 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 IBM Corporation, 2008
+ *
+ * Authors: Dipankar Sarma <dipankar@in.ibm.com>
+ *	    Manfred Spraul <manfred@colorfullife.com>
+ *	    Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
+ *
+ * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
+ * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
+ *
+ * For detailed explanation of Read-Copy Update mechanism see -
+ * 	Documentation/RCU
+ */
+#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/cpu.h>
+#include <linux/mutex.h>
+#include <linux/time.h>
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+static struct lock_class_key rcu_lock_key;
+struct lockdep_map rcu_lock_map =
+	STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
+EXPORT_SYMBOL_GPL(rcu_lock_map);
+#endif
+
+/* Data structures. */
+
+#define RCU_STATE_INITIALIZER(name) { \
+	.level = { &name.node[0] }, \
+	.levelcnt = { \
+		NUM_RCU_LVL_0,  /* root of hierarchy. */ \
+		NUM_RCU_LVL_1, \
+		NUM_RCU_LVL_2, \
+		NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
+	}, \
+	.signaled = RCU_SIGNAL_INIT, \
+	.gpnum = -300, \
+	.completed = -300, \
+	.onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
+	.fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
+	.n_force_qs = 0, \
+	.n_force_qs_ngp = 0, \
+}
+
+struct rcu_state rcu_state = RCU_STATE_INITIALIZER(rcu_state);
+DEFINE_PER_CPU(struct rcu_data, rcu_data);
+
+struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
+DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
+
+#ifdef CONFIG_NO_HZ
+DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks);
+#endif /* #ifdef CONFIG_NO_HZ */
+
+static int blimit = 10;		/* Maximum callbacks per softirq. */
+static int qhimark = 10000;	/* If this many pending, ignore blimit. */
+static int qlowmark = 100;	/* Once only this many pending, use blimit. */
+
+static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
+
+/*
+ * Return the number of RCU batches processed thus far for debug & stats.
+ */
+long rcu_batches_completed(void)
+{
+	return rcu_state.completed;
+}
+EXPORT_SYMBOL_GPL(rcu_batches_completed);
+
+/*
+ * Return the number of RCU BH batches processed thus far for debug & stats.
+ */
+long rcu_batches_completed_bh(void)
+{
+	return rcu_bh_state.completed;
+}
+EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
+
+/*
+ * Does the CPU have callbacks ready to be invoked?
+ */
+static int
+cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
+{
+	return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
+}
+
+/*
+ * Does the current CPU require a yet-as-unscheduled grace period?
+ */
+static int
+cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
+{
+	/* ACCESS_ONCE() because we are accessing outside of lock. */
+	return *rdp->nxttail[RCU_DONE_TAIL] &&
+	       ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
+}
+
+/*
+ * Return the root node of the specified rcu_state structure.
+ */
+static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
+{
+	return &rsp->node[0];
+}
+
+#ifdef CONFIG_SMP
+
+/*
+ * If the specified CPU is offline, tell the caller that it is in
+ * a quiescent state.  Otherwise, whack it with a reschedule IPI.
+ * Grace periods can end up waiting on an offline CPU when that
+ * CPU is in the process of coming online -- it will be added to the
+ * rcu_node bitmasks before it actually makes it online.  The same thing
+ * can happen while a CPU is in the process of coming online.  Because this
+ * race is quite rare, we check for it after detecting that the grace
+ * period has been delayed rather than checking each and every CPU
+ * each and every time we start a new grace period.
+ */
+static int rcu_implicit_offline_qs(struct rcu_data *rdp)
+{
+	/*
+	 * If the CPU is offline, it is in a quiescent state.  We can
+	 * trust its state not to change because interrupts are disabled.
+	 */
+	if (cpu_is_offline(rdp->cpu)) {
+		rdp->offline_fqs++;
+		return 1;
+	}
+
+	/* The CPU is online, so send it a reschedule IPI. */
+	if (rdp->cpu != smp_processor_id())
+		smp_send_reschedule(rdp->cpu);
+	else
+		set_need_resched();
+	rdp->resched_ipi++;
+	return 0;
+}
+
+#endif /* #ifdef CONFIG_SMP */
+
+#ifdef CONFIG_NO_HZ
+static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5);
+
+/**
+ * rcu_enter_nohz - inform RCU that current CPU is entering nohz
+ *
+ * Enter nohz mode, in other words, -leave- the mode in which RCU
+ * read-side critical sections can occur.  (Though RCU read-side
+ * critical sections can occur in irq handlers in nohz mode, a possibility
+ * handled by rcu_irq_enter() and rcu_irq_exit()).
+ */
+void rcu_enter_nohz(void)
+{
+	unsigned long flags;
+	struct rcu_dynticks *rdtp;
+
+	smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
+	local_irq_save(flags);
+	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp->dynticks++;
+	rdtp->dynticks_nesting--;
+	WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
+	local_irq_restore(flags);
+}
+
+/*
+ * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
+ *
+ * Exit nohz mode, in other words, -enter- the mode in which RCU
+ * read-side critical sections normally occur.
+ */
+void rcu_exit_nohz(void)
+{
+	unsigned long flags;
+	struct rcu_dynticks *rdtp;
+
+	local_irq_save(flags);
+	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp->dynticks++;
+	rdtp->dynticks_nesting++;
+	WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
+	local_irq_restore(flags);
+	smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
+}
+
+/**
+ * rcu_nmi_enter - inform RCU of entry to NMI context
+ *
+ * If the CPU was idle with dynamic ticks active, and there is no
+ * irq handler running, this updates rdtp->dynticks_nmi to let the
+ * RCU grace-period handling know that the CPU is active.
+ */
+void rcu_nmi_enter(void)
+{
+	struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
+
+	if (rdtp->dynticks & 0x1)
+		return;
+	rdtp->dynticks_nmi++;
+	WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs);
+	smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
+}
+
+/**
+ * rcu_nmi_exit - inform RCU of exit from NMI context
+ *
+ * If the CPU was idle with dynamic ticks active, and there is no
+ * irq handler running, this updates rdtp->dynticks_nmi to let the
+ * RCU grace-period handling know that the CPU is no longer active.
+ */
+void rcu_nmi_exit(void)
+{
+	struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
+
+	if (rdtp->dynticks & 0x1)
+		return;
+	smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
+	rdtp->dynticks_nmi++;
+	WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs);
+}
+
+/**
+ * rcu_irq_enter - inform RCU of entry to hard irq context
+ *
+ * If the CPU was idle with dynamic ticks active, this updates the
+ * rdtp->dynticks to let the RCU handling know that the CPU is active.
+ */
+void rcu_irq_enter(void)
+{
+	struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
+
+	if (rdtp->dynticks_nesting++)
+		return;
+	rdtp->dynticks++;
+	WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
+	smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
+}
+
+/**
+ * rcu_irq_exit - inform RCU of exit from hard irq context
+ *
+ * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
+ * to put let the RCU handling be aware that the CPU is going back to idle
+ * with no ticks.
+ */
+void rcu_irq_exit(void)
+{
+	struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
+
+	if (--rdtp->dynticks_nesting)
+		return;
+	smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
+	rdtp->dynticks++;
+	WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
+
+	/* If the interrupt queued a callback, get out of dyntick mode. */
+	if (__get_cpu_var(rcu_data).nxtlist ||
+	    __get_cpu_var(rcu_bh_data).nxtlist)
+		set_need_resched();
+}
+
+/*
+ * Record the specified "completed" value, which is later used to validate
+ * dynticks counter manipulations.  Specify "rsp->completed - 1" to
+ * unconditionally invalidate any future dynticks manipulations (which is
+ * useful at the beginning of a grace period).
+ */
+static void dyntick_record_completed(struct rcu_state *rsp, long comp)
+{
+	rsp->dynticks_completed = comp;
+}
+
+#ifdef CONFIG_SMP
+
+/*
+ * Recall the previously recorded value of the completion for dynticks.
+ */
+static long dyntick_recall_completed(struct rcu_state *rsp)
+{
+	return rsp->dynticks_completed;
+}
+
+/*
+ * Snapshot the specified CPU's dynticks counter so that we can later
+ * credit them with an implicit quiescent state.  Return 1 if this CPU
+ * is already in a quiescent state courtesy of dynticks idle mode.
+ */
+static int dyntick_save_progress_counter(struct rcu_data *rdp)
+{
+	int ret;
+	int snap;
+	int snap_nmi;
+
+	snap = rdp->dynticks->dynticks;
+	snap_nmi = rdp->dynticks->dynticks_nmi;
+	smp_mb();	/* Order sampling of snap with end of grace period. */
+	rdp->dynticks_snap = snap;
+	rdp->dynticks_nmi_snap = snap_nmi;
+	ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
+	if (ret)
+		rdp->dynticks_fqs++;
+	return ret;
+}
+
+/*
+ * Return true if the specified CPU has passed through a quiescent
+ * state by virtue of being in or having passed through an dynticks
+ * idle state since the last call to dyntick_save_progress_counter()
+ * for this same CPU.
+ */
+static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
+{
+	long curr;
+	long curr_nmi;
+	long snap;
+	long snap_nmi;
+
+	curr = rdp->dynticks->dynticks;
+	snap = rdp->dynticks_snap;
+	curr_nmi = rdp->dynticks->dynticks_nmi;
+	snap_nmi = rdp->dynticks_nmi_snap;
+	smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
+
+	/*
+	 * If the CPU passed through or entered a dynticks idle phase with
+	 * no active irq/NMI handlers, then we can safely pretend that the CPU
+	 * already acknowledged the request to pass through a quiescent
+	 * state.  Either way, that CPU cannot possibly be in an RCU
+	 * read-side critical section that started before the beginning
+	 * of the current RCU grace period.
+	 */
+	if ((curr != snap || (curr & 0x1) == 0) &&
+	    (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
+		rdp->dynticks_fqs++;
+		return 1;
+	}
+
+	/* Go check for the CPU being offline. */
+	return rcu_implicit_offline_qs(rdp);
+}
+
+#endif /* #ifdef CONFIG_SMP */
+
+#else /* #ifdef CONFIG_NO_HZ */
+
+static void dyntick_record_completed(struct rcu_state *rsp, long comp)
+{
+}
+
+#ifdef CONFIG_SMP
+
+/*
+ * If there are no dynticks, then the only way that a CPU can passively
+ * be in a quiescent state is to be offline.  Unlike dynticks idle, which
+ * is a point in time during the prior (already finished) grace period,
+ * an offline CPU is always in a quiescent state, and thus can be
+ * unconditionally applied.  So just return the current value of completed.
+ */
+static long dyntick_recall_completed(struct rcu_state *rsp)
+{
+	return rsp->completed;
+}
+
+static int dyntick_save_progress_counter(struct rcu_data *rdp)
+{
+	return 0;
+}
+
+static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
+{
+	return rcu_implicit_offline_qs(rdp);
+}
+
+#endif /* #ifdef CONFIG_SMP */
+
+#endif /* #else #ifdef CONFIG_NO_HZ */
+
+#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
+
+static void record_gp_stall_check_time(struct rcu_state *rsp)
+{
+	rsp->gp_start = jiffies;
+	rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
+}
+
+static void print_other_cpu_stall(struct rcu_state *rsp)
+{
+	int cpu;
+	long delta;
+	unsigned long flags;
+	struct rcu_node *rnp = rcu_get_root(rsp);
+	struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
+	struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
+
+	/* Only let one CPU complain about others per time interval. */
+
+	spin_lock_irqsave(&rnp->lock, flags);
+	delta = jiffies - rsp->jiffies_stall;
+	if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) {
+		spin_unlock_irqrestore(&rnp->lock, flags);
+		return;
+	}
+	rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
+	spin_unlock_irqrestore(&rnp->lock, flags);
+
+	/* OK, time to rat on our buddy... */
+
+	printk(KERN_ERR "INFO: RCU detected CPU stalls:");
+	for (; rnp_cur < rnp_end; rnp_cur++) {
+		if (rnp_cur->qsmask == 0)
+			continue;
+		for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
+			if (rnp_cur->qsmask & (1UL << cpu))
+				printk(" %d", rnp_cur->grplo + cpu);
+	}
+	printk(" (detected by %d, t=%ld jiffies)\n",
+	       smp_processor_id(), (long)(jiffies - rsp->gp_start));
+	force_quiescent_state(rsp, 0);  /* Kick them all. */
+}
+
+static void print_cpu_stall(struct rcu_state *rsp)
+{
+	unsigned long flags;
+	struct rcu_node *rnp = rcu_get_root(rsp);
+
+	printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
+			smp_processor_id(), jiffies - rsp->gp_start);
+	dump_stack();
+	spin_lock_irqsave(&rnp->lock, flags);
+	if ((long)(jiffies - rsp->jiffies_stall) >= 0)
+		rsp->jiffies_stall =
+			jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
+	spin_unlock_irqrestore(&rnp->lock, flags);
+	set_need_resched();  /* kick ourselves to get things going. */
+}
+
+static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
+{
+	long delta;
+	struct rcu_node *rnp;
+
+	delta = jiffies - rsp->jiffies_stall;
+	rnp = rdp->mynode;
+	if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
+
+		/* We haven't checked in, so go dump stack. */
+		print_cpu_stall(rsp);
+
+	} else if (rsp->gpnum != rsp->completed &&
+		   delta >= RCU_STALL_RAT_DELAY) {
+
+		/* They had two time units to dump stack, so complain. */
+		print_other_cpu_stall(rsp);
+	}
+}
+
+#else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
+
+static void record_gp_stall_check_time(struct rcu_state *rsp)
+{
+}
+
+static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
+{
+}
+
+#endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
+
+/*
+ * Update CPU-local rcu_data state to record the newly noticed grace period.
+ * This is used both when we started the grace period and when we notice
+ * that someone else started the grace period.
+ */
+static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
+{
+	rdp->qs_pending = 1;
+	rdp->passed_quiesc = 0;
+	rdp->gpnum = rsp->gpnum;
+	rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending +
+				      RCU_JIFFIES_TILL_FORCE_QS;
+}
+
+/*
+ * Did someone else start a new RCU grace period start since we last
+ * checked?  Update local state appropriately if so.  Must be called
+ * on the CPU corresponding to rdp.
+ */
+static int
+check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
+{
+	unsigned long flags;
+	int ret = 0;
+
+	local_irq_save(flags);
+	if (rdp->gpnum != rsp->gpnum) {
+		note_new_gpnum(rsp, rdp);
+		ret = 1;
+	}
+	local_irq_restore(flags);
+	return ret;
+}
+
+/*
+ * Start a new RCU grace period if warranted, re-initializing the hierarchy
+ * in preparation for detecting the next grace period.  The caller must hold
+ * the root node's ->lock, which is released before return.  Hard irqs must
+ * be disabled.
+ */
+static void
+rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
+	__releases(rcu_get_root(rsp)->lock)
+{
+	struct rcu_data *rdp = rsp->rda[smp_processor_id()];
+	struct rcu_node *rnp = rcu_get_root(rsp);
+	struct rcu_node *rnp_cur;
+	struct rcu_node *rnp_end;
+
+	if (!cpu_needs_another_gp(rsp, rdp)) {
+		spin_unlock_irqrestore(&rnp->lock, flags);
+		return;
+	}
+
+	/* Advance to a new grace period and initialize state. */
+	rsp->gpnum++;
+	rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
+	rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
+	rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending +
+				      RCU_JIFFIES_TILL_FORCE_QS;
+	record_gp_stall_check_time(rsp);
+	dyntick_record_completed(rsp, rsp->completed - 1);
+	note_new_gpnum(rsp, rdp);
+
+	/*
+	 * Because we are first, we know that all our callbacks will
+	 * be covered by this upcoming grace period, even the ones
+	 * that were registered arbitrarily recently.
+	 */
+	rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
+	rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
+
+	/* Special-case the common single-level case. */
+	if (NUM_RCU_NODES == 1) {
+		rnp->qsmask = rnp->qsmaskinit;
+		spin_unlock_irqrestore(&rnp->lock, flags);
+		return;
+	}
+
+	spin_unlock(&rnp->lock);  /* leave irqs disabled. */
+
+
+	/* Exclude any concurrent CPU-hotplug operations. */
+	spin_lock(&rsp->onofflock);  /* irqs already disabled. */
+
+	/*
+	 * Set the quiescent-state-needed bits in all the non-leaf RCU
+	 * nodes for all currently online CPUs.  This operation relies
+	 * on the layout of the hierarchy within the rsp->node[] array.
+	 * Note that other CPUs will access only the leaves of the
+	 * hierarchy, which still indicate that no grace period is in
+	 * progress.  In addition, we have excluded CPU-hotplug operations.
+	 *
+	 * We therefore do not need to hold any locks.  Any required
+	 * memory barriers will be supplied by the locks guarding the
+	 * leaf rcu_nodes in the hierarchy.
+	 */
+
+	rnp_end = rsp->level[NUM_RCU_LVLS - 1];
+	for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++)
+		rnp_cur->qsmask = rnp_cur->qsmaskinit;
+
+	/*
+	 * Now set up the leaf nodes.  Here we must be careful.  First,
+	 * we need to hold the lock in order to exclude other CPUs, which
+	 * might be contending for the leaf nodes' locks.  Second, as
+	 * soon as we initialize a given leaf node, its CPUs might run
+	 * up the rest of the hierarchy.  We must therefore acquire locks
+	 * for each node that we touch during this stage.  (But we still
+	 * are excluding CPU-hotplug operations.)
+	 *
+	 * Note that the grace period cannot complete until we finish
+	 * the initialization process, as there will be at least one
+	 * qsmask bit set in the root node until that time, namely the
+	 * one corresponding to this CPU.
+	 */
+	rnp_end = &rsp->node[NUM_RCU_NODES];
+	rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
+	for (; rnp_cur < rnp_end; rnp_cur++) {
+		spin_lock(&rnp_cur->lock);	/* irqs already disabled. */
+		rnp_cur->qsmask = rnp_cur->qsmaskinit;
+		spin_unlock(&rnp_cur->lock);	/* irqs already disabled. */
+	}
+
+	rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
+	spin_unlock_irqrestore(&rsp->onofflock, flags);
+}
+
+/*
+ * Advance this CPU's callbacks, but only if the current grace period
+ * has ended.  This may be called only from the CPU to whom the rdp
+ * belongs.
+ */
+static void
+rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
+{
+	long completed_snap;
+	unsigned long flags;
+
+	local_irq_save(flags);
+	completed_snap = ACCESS_ONCE(rsp->completed);  /* outside of lock. */
+
+	/* Did another grace period end? */
+	if (rdp->completed != completed_snap) {
+
+		/* Advance callbacks.  No harm if list empty. */
+		rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
+		rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
+		rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
+
+		/* Remember that we saw this grace-period completion. */
+		rdp->completed = completed_snap;
+	}
+	local_irq_restore(flags);
+}
+
+/*
+ * Similar to cpu_quiet(), for which it is a helper function.  Allows
+ * a group of CPUs to be quieted at one go, though all the CPUs in the
+ * group must be represented by the same leaf rcu_node structure.
+ * That structure's lock must be held upon entry, and it is released
+ * before return.
+ */
+static void
+cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
+	      unsigned long flags)
+	__releases(rnp->lock)
+{
+	/* Walk up the rcu_node hierarchy. */
+	for (;;) {
+		if (!(rnp->qsmask & mask)) {
+
+			/* Our bit has already been cleared, so done. */
+			spin_unlock_irqrestore(&rnp->lock, flags);
+			return;
+		}
+		rnp->qsmask &= ~mask;
+		if (rnp->qsmask != 0) {
+
+			/* Other bits still set at this level, so done. */
+			spin_unlock_irqrestore(&rnp->lock, flags);
+			return;
+		}
+		mask = rnp->grpmask;
+		if (rnp->parent == NULL) {
+
+			/* No more levels.  Exit loop holding root lock. */
+
+			break;
+		}
+		spin_unlock_irqrestore(&rnp->lock, flags);
+		rnp = rnp->parent;
+		spin_lock_irqsave(&rnp->lock, flags);
+	}
+
+	/*
+	 * Get here if we are the last CPU to pass through a quiescent
+	 * state for this grace period.  Clean up and let rcu_start_gp()
+	 * start up the next grace period if one is needed.  Note that
+	 * we still hold rnp->lock, as required by rcu_start_gp(), which
+	 * will release it.
+	 */
+	rsp->completed = rsp->gpnum;
+	rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
+	rcu_start_gp(rsp, flags);  /* releases rnp->lock. */
+}
+
+/*
+ * Record a quiescent state for the specified CPU, which must either be
+ * the current CPU or an offline CPU.  The lastcomp argument is used to
+ * make sure we are still in the grace period of interest.  We don't want
+ * to end the current grace period based on quiescent states detected in
+ * an earlier grace period!
+ */
+static void
+cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
+{
+	unsigned long flags;
+	unsigned long mask;
+	struct rcu_node *rnp;
+
+	rnp = rdp->mynode;
+	spin_lock_irqsave(&rnp->lock, flags);
+	if (lastcomp != ACCESS_ONCE(rsp->completed)) {
+
+		/*
+		 * Someone beat us to it for this grace period, so leave.
+		 * The race with GP start is resolved by the fact that we
+		 * hold the leaf rcu_node lock, so that the per-CPU bits
+		 * cannot yet be initialized -- so we would simply find our
+		 * CPU's bit already cleared in cpu_quiet_msk() if this race
+		 * occurred.
+		 */
+		rdp->passed_quiesc = 0;	/* try again later! */
+		spin_unlock_irqrestore(&rnp->lock, flags);
+		return;
+	}
+	mask = rdp->grpmask;
+	if ((rnp->qsmask & mask) == 0) {
+		spin_unlock_irqrestore(&rnp->lock, flags);
+	} else {
+		rdp->qs_pending = 0;
+
+		/*
+		 * This GP can't end until cpu checks in, so all of our
+		 * callbacks can be processed during the next GP.
+		 */
+		rdp = rsp->rda[smp_processor_id()];
+		rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
+
+		cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
+	}
+}
+
+/*
+ * Check to see if there is a new grace period of which this CPU
+ * is not yet aware, and if so, set up local rcu_data state for it.
+ * Otherwise, see if this CPU has just passed through its first
+ * quiescent state for this grace period, and record that fact if so.
+ */
+static void
+rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
+{
+	/* If there is now a new grace period, record and return. */
+	if (check_for_new_grace_period(rsp, rdp))
+		return;
+
+	/*
+	 * Does this CPU still need to do its part for current grace period?
+	 * If no, return and let the other CPUs do their part as well.
+	 */
+	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;
+
+	/* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
+	cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
+}
+
+#ifdef CONFIG_HOTPLUG_CPU
+
+/*
+ * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
+ * and move all callbacks from the outgoing CPU to the current one.
+ */
+static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
+{
+	int i;
+	unsigned long flags;
+	long lastcomp;
+	unsigned long mask;
+	struct rcu_data *rdp = rsp->rda[cpu];
+	struct rcu_data *rdp_me;
+	struct rcu_node *rnp;
+
+	/* Exclude any attempts to start a new grace period. */
+	spin_lock_irqsave(&rsp->onofflock, flags);
+
+	/* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
+	rnp = rdp->mynode;
+	mask = rdp->grpmask;	/* rnp->grplo is constant. */
+	do {
+		spin_lock(&rnp->lock);		/* irqs already disabled. */
+		rnp->qsmaskinit &= ~mask;
+		if (rnp->qsmaskinit != 0) {
+			spin_unlock(&rnp->lock); /* irqs already disabled. */
+			break;
+		}
+		mask = rnp->grpmask;
+		spin_unlock(&rnp->lock);	/* irqs already disabled. */
+		rnp = rnp->parent;
+	} while (rnp != NULL);
+	lastcomp = rsp->completed;
+
+	spin_unlock(&rsp->onofflock);		/* irqs remain disabled. */
+
+	/* Being offline is a quiescent state, so go record it. */
+	cpu_quiet(cpu, rsp, rdp, lastcomp);
+
+	/*
+	 * Move callbacks from the outgoing CPU to the running CPU.
+	 * Note that the outgoing CPU is now quiscent, so it is now
+	 * (uncharacteristically) safe to access it rcu_data structure.
+	 * Note also that we must carefully retain the order of the
+	 * outgoing CPU's callbacks in order for rcu_barrier() to work
+	 * correctly.  Finally, note that we start all the callbacks
+	 * afresh, even those that have passed through a grace period
+	 * and are therefore ready to invoke.  The theory is that hotplug
+	 * events are rare, and that if they are frequent enough to
+	 * indefinitely delay callbacks, you have far worse things to
+	 * be worrying about.
+	 */
+	rdp_me = rsp->rda[smp_processor_id()];
+	if (rdp->nxtlist != NULL) {
+		*rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
+		rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
+		rdp->nxtlist = NULL;
+		for (i = 0; i < RCU_NEXT_SIZE; i++)
+			rdp->nxttail[i] = &rdp->nxtlist;
+		rdp_me->qlen += rdp->qlen;
+		rdp->qlen = 0;
+	}
+	local_irq_restore(flags);
+}
+
+/*
+ * Remove the specified CPU from the RCU hierarchy and move any pending
+ * callbacks that it might have to the current CPU.  This code assumes
+ * that at least one CPU in the system will remain running at all times.
+ * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
+ */
+static void rcu_offline_cpu(int cpu)
+{
+	__rcu_offline_cpu(cpu, &rcu_state);
+	__rcu_offline_cpu(cpu, &rcu_bh_state);
+}
+
+#else /* #ifdef CONFIG_HOTPLUG_CPU */
+
+static void rcu_offline_cpu(int cpu)
+{
+}
+
+#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
+
+/*
+ * Invoke any RCU callbacks that have made it to the end of their grace
+ * period.  Thottle as specified by rdp->blimit.
+ */
+static void rcu_do_batch(struct rcu_data *rdp)
+{
+	unsigned long flags;
+	struct rcu_head *next, *list, **tail;
+	int count;
+
+	/* If no callbacks are ready, just return.*/
+	if (!cpu_has_callbacks_ready_to_invoke(rdp))
+		return;
+
+	/*
+	 * Extract the list of ready callbacks, disabling to prevent
+	 * races with call_rcu() from interrupt handlers.
+	 */
+	local_irq_save(flags);
+	list = rdp->nxtlist;
+	rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
+	*rdp->nxttail[RCU_DONE_TAIL] = NULL;
+	tail = rdp->nxttail[RCU_DONE_TAIL];
+	for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
+		if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
+			rdp->nxttail[count] = &rdp->nxtlist;
+	local_irq_restore(flags);
+
+	/* Invoke callbacks. */
+	count = 0;
+	while (list) {
+		next = list->next;
+		prefetch(next);
+		list->func(list);
+		list = next;
+		if (++count >= rdp->blimit)