Preempt-RCU: implementation

This patch implements a new version of RCU which allows its read-side
critical sections to be preempted. It uses a set of counter pairs
to keep track of the read-side critical sections and flips them
when all tasks exit read-side critical section. The details
of this implementation can be found in this paper -

	http://www.rdrop.com/users/paulmck/RCU/OLSrtRCU.2006.08.11a.pdf

and the article-

	http://lwn.net/Articles/253651/

This patch was developed as a part of the -rt kernel development and
meant to provide better latencies when read-side critical sections of
RCU don't disable preemption.  As a consequence of keeping track of RCU
readers, the readers have a slight overhead (optimizations in the paper).
This implementation co-exists with the "classic" RCU implementations
and can be switched to at compiler.

Also includes RCU tracing summarized in debugfs.

[ akpm@linux-foundation.org: build fixes on non-preempt architectures ]

Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Reviewed-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

6 files changed, 1166 insertions, 2 deletions

diff --git a/kernel/Kconfig.preempt b/kernel/Kconfig.preempt
index c64ce9c14207..61fa116efcde 100644
--- a/kernel/Kconfig.preempt
+++ b/kernel/Kconfig.preempt
@@ -63,3 +63,13 @@ config PREEMPT_BKL
 	  Say Y here if you are building a kernel for a desktop system.
 	  Say N if you are unsure.
 
+config RCU_TRACE
+	bool "Enable tracing for RCU - currently stats in debugfs"
+	select DEBUG_FS
+	default y
+	help
+	  This option provides tracing in RCU which presents stats
+	  in debugfs for debugging RCU implementation.
+
+	  Say Y here if you want to enable RCU tracing
+	  Say N if you are unsure.
diff --git a/kernel/Makefile b/kernel/Makefile
index def5dd6097a0..68755cd9a7e4 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -6,7 +6,7 @@ obj-y     = sched.o fork.o exec_domain.o panic.o printk.o profile.o \
 	    exit.o itimer.o time.o softirq.o resource.o \
 	    sysctl.o capability.o ptrace.o timer.o user.o user_namespace.o \
 	    signal.o sys.o kmod.o workqueue.o pid.o \
-	    rcupdate.o rcuclassic.o extable.o params.o posix-timers.o \
+	    rcupdate.o extable.o params.o posix-timers.o \
 	    kthread.o wait.o kfifo.o sys_ni.o posix-cpu-timers.o mutex.o \
 	    hrtimer.o rwsem.o latency.o nsproxy.o srcu.o \
 	    utsname.o notifier.o
@@ -52,6 +52,11 @@ obj-$(CONFIG_DETECT_SOFTLOCKUP) += softlockup.o
 obj-$(CONFIG_GENERIC_HARDIRQS) += irq/
 obj-$(CONFIG_SECCOMP) += seccomp.o
 obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o
+obj-$(CONFIG_CLASSIC_RCU) += rcuclassic.o
+obj-$(CONFIG_PREEMPT_RCU) += rcupreempt.o
+ifeq ($(CONFIG_PREEMPT_RCU),y)
+obj-$(CONFIG_RCU_TRACE) += rcupreempt_trace.o
+endif
 obj-$(CONFIG_RELAY) += relay.o
 obj-$(CONFIG_SYSCTL) += utsname_sysctl.o
 obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o
diff --git a/kernel/fork.c b/kernel/fork.c
index 930c51865ab4..9f8ef32cbc7a 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1045,6 +1045,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	copy_flags(clone_flags, p);
 	INIT_LIST_HEAD(&p->children);
 	INIT_LIST_HEAD(&p->sibling);
+#ifdef CONFIG_PREEMPT_RCU
+	p->rcu_read_lock_nesting = 0;
+	p->rcu_flipctr_idx = 0;
+#endif /* #ifdef CONFIG_PREEMPT_RCU */
 	p->vfork_done = NULL;
 	spin_lock_init(&p->alloc_lock);
 
diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c
index ce0cf16cab67..f4ffbd0f306f 100644
--- a/kernel/rcuclassic.c
+++ b/kernel/rcuclassic.c
@@ -45,7 +45,6 @@
 #include <linux/moduleparam.h>
 #include <linux/percpu.h>
 #include <linux/notifier.h>
-/* #include <linux/rcupdate.h> @@@ */
 #include <linux/cpu.h>
 #include <linux/mutex.h>
 
diff --git a/kernel/rcupreempt.c b/kernel/rcupreempt.c
new file mode 100644
index 000000000000..a5aabb1677f8
--- /dev/null
+++ b/kernel/rcupreempt.c
@@ -0,0 +1,816 @@
+/*
+ * Read-Copy Update mechanism for mutual exclusion, realtime implementation
+ *
+ * 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, 2006
+ *
+ * Authors: Paul E. McKenney <paulmck@us.ibm.com>
+ *		With thanks to Esben Nielsen, Bill Huey, and Ingo Molnar
+ *		for pushing me away from locks and towards counters, and
+ *		to Suparna Bhattacharya for pushing me completely away
+ *		from atomic instructions on the read side.
+ *
+ * Papers:  http://www.rdrop.com/users/paulmck/RCU
+ *
+ * Design Document: http://lwn.net/Articles/253651/
+ *
+ * For detailed explanation of Read-Copy Update mechanism see -
+ * 		Documentation/RCU/ *.txt
+ *
+ */
+#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>
+#include <linux/random.h>
+#include <linux/delay.h>
+#include <linux/byteorder/swabb.h>
+#include <linux/cpumask.h>
+#include <linux/rcupreempt_trace.h>
+
+/*
+ * Macro that prevents the compiler from reordering accesses, but does
+ * absolutely -nothing- to prevent CPUs from reordering.  This is used
+ * only to mediate communication between mainline code and hardware
+ * interrupt and NMI handlers.
+ */
+#define ACCESS_ONCE(x) (*(volatile typeof(x) *)&(x))
+
+/*
+ * PREEMPT_RCU data structures.
+ */
+
+/*
+ * GP_STAGES specifies the number of times the state machine has
+ * to go through the all the rcu_try_flip_states (see below)
+ * in a single Grace Period.
+ *
+ * GP in GP_STAGES stands for Grace Period ;)
+ */
+#define GP_STAGES    2
+struct rcu_data {
+	spinlock_t	lock;		/* Protect rcu_data fields. */
+	long		completed;	/* Number of last completed batch. */
+	int		waitlistcount;
+	struct tasklet_struct rcu_tasklet;
+	struct rcu_head *nextlist;
+	struct rcu_head **nexttail;
+	struct rcu_head *waitlist[GP_STAGES];
+	struct rcu_head **waittail[GP_STAGES];
+	struct rcu_head *donelist;
+	struct rcu_head **donetail;
+	long rcu_flipctr[2];
+#ifdef CONFIG_RCU_TRACE
+	struct rcupreempt_trace trace;
+#endif /* #ifdef CONFIG_RCU_TRACE */
+};
+
+/*
+ * States for rcu_try_flip() and friends.
+ */
+
+enum rcu_try_flip_states {
+
+	/*
+	 * Stay here if nothing is happening. Flip the counter if somthing
+	 * starts happening. Denoted by "I"
+	 */
+	rcu_try_flip_idle_state,
+
+	/*
+	 * Wait here for all CPUs to notice that the counter has flipped. This
+	 * prevents the old set of counters from ever being incremented once
+	 * we leave this state, which in turn is necessary because we cannot
+	 * test any individual counter for zero -- we can only check the sum.
+	 * Denoted by "A".
+	 */
+	rcu_try_flip_waitack_state,
+
+	/*
+	 * Wait here for the sum of the old per-CPU counters to reach zero.
+	 * Denoted by "Z".
+	 */
+	rcu_try_flip_waitzero_state,
+
+	/*
+	 * Wait here for each of the other CPUs to execute a memory barrier.
+	 * This is necessary to ensure that these other CPUs really have
+	 * completed executing their RCU read-side critical sections, despite
+	 * their CPUs wildly reordering memory. Denoted by "M".
+	 */
+	rcu_try_flip_waitmb_state,
+};
+
+struct rcu_ctrlblk {
+	spinlock_t	fliplock;	/* Protect state-machine transitions. */
+	long		completed;	/* Number of last completed batch. */
+	enum rcu_try_flip_states rcu_try_flip_state; /* The current state of
+							the rcu state machine */
+};
+
+static DEFINE_PER_CPU(struct rcu_data, rcu_data);
+static struct rcu_ctrlblk rcu_ctrlblk = {
+	.fliplock = __SPIN_LOCK_UNLOCKED(rcu_ctrlblk.fliplock),
+	.completed = 0,
+	.rcu_try_flip_state = rcu_try_flip_idle_state,
+};
+
+
+#ifdef CONFIG_RCU_TRACE
+static char *rcu_try_flip_state_names[] =
+	{ "idle", "waitack", "waitzero", "waitmb" };
+#endif /* #ifdef CONFIG_RCU_TRACE */
+
+/*
+ * Enum and per-CPU flag to determine when each CPU has seen
+ * the most recent counter flip.
+ */
+
+enum rcu_flip_flag_values {
+	rcu_flip_seen,		/* Steady/initial state, last flip seen. */
+				/* Only GP detector can update. */
+	rcu_flipped		/* Flip just completed, need confirmation. */
+				/* Only corresponding CPU can update. */
+};
+static DEFINE_PER_CPU_SHARED_ALIGNED(enum rcu_flip_flag_values, rcu_flip_flag)
+								= rcu_flip_seen;
+
+/*
+ * Enum and per-CPU flag to determine when each CPU has executed the
+ * needed memory barrier to fence in memory references from its last RCU
+ * read-side critical section in the just-completed grace period.
+ */
+
+enum rcu_mb_flag_values {
+	rcu_mb_done,		/* Steady/initial state, no mb()s required. */
+				/* Only GP detector can update. */
+	rcu_mb_needed		/* Flip just completed, need an mb(). */
+				/* Only corresponding CPU can update. */
+};
+static DEFINE_PER_CPU_SHARED_ALIGNED(enum rcu_mb_flag_values, rcu_mb_flag)
+								= rcu_mb_done;
+
+/*
+ * RCU_DATA_ME: find the current CPU's rcu_data structure.
+ * RCU_DATA_CPU: find the specified CPU's rcu_data structure.
+ */
+#define RCU_DATA_ME()		(&__get_cpu_var(rcu_data))
+#define RCU_DATA_CPU(cpu)	(&per_cpu(rcu_data, cpu))
+
+/*
+ * Helper macro for tracing when the appropriate rcu_data is not
+ * cached in a local variable, but where the CPU number is so cached.
+ */
+#define RCU_TRACE_CPU(f, cpu) RCU_TRACE(f, &(RCU_DATA_CPU(cpu)->trace));
+
+/*
+ * Helper macro for tracing when the appropriate rcu_data is not
+ * cached in a local variable.
+ */
+#define RCU_TRACE_ME(f) RCU_TRACE(f, &(RCU_DATA_ME()->trace));
+
+/*
+ * Helper macro for tracing when the appropriate rcu_data is pointed
+ * to by a local variable.
+ */
+#define RCU_TRACE_RDP(f, rdp) RCU_TRACE(f, &((rdp)->trace));
+
+/*
+ * Return the number of RCU batches processed thus far.  Useful
+ * for debug and statistics.
+ */
+long rcu_batches_completed(void)
+{
+	return rcu_ctrlblk.completed;
+}
+EXPORT_SYMBOL_GPL(rcu_batches_completed);
+
+EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
+
+void __rcu_read_lock(void)
+{
+	int idx;
+	struct task_struct *t = current;
+	int nesting;
+
+	nesting = ACCESS_ONCE(t->rcu_read_lock_nesting);
+	if (nesting != 0) {
+
+		/* An earlier rcu_read_lock() covers us, just count it. */
+
+		t->rcu_read_lock_nesting = nesting + 1;
+
+	} else {
+		unsigned long flags;
+
+		/*
+		 * We disable interrupts for the following reasons:
+		 * - If we get scheduling clock interrupt here, and we
+		 *   end up acking the counter flip, it's like a promise
+		 *   that we will never increment the old counter again.
+		 *   Thus we will break that promise if that
+		 *   scheduling clock interrupt happens between the time
+		 *   we pick the .completed field and the time that we
+		 *   increment our counter.
+		 *
+		 * - We don't want to be preempted out here.
+		 *
+		 * NMIs can still occur, of course, and might themselves
+		 * contain rcu_read_lock().
+		 */
+
+		local_irq_save(flags);
+
+		/*
+		 * Outermost nesting of rcu_read_lock(), so increment
+		 * the current counter for the current CPU.  Use volatile
+		 * casts to prevent the compiler from reordering.
+		 */
+
+		idx = ACCESS_ONCE(rcu_ctrlblk.completed) & 0x1;
+		ACCESS_ONCE(RCU_DATA_ME()->rcu_flipctr[idx])++;
+
+		/*
+		 * Now that the per-CPU counter has been incremented, we
+		 * are protected from races with rcu_read_lock() invoked
+		 * from NMI handlers on this CPU.  We can therefore safely
+		 * increment the nesting counter, relieving further NMIs
+		 * of the need to increment the per-CPU counter.
+		 */
+
+		ACCESS_ONCE(t->rcu_read_lock_nesting) = nesting + 1;
+
+		/*
+		 * Now that we have preventing any NMIs from storing
+		 * to the ->rcu_flipctr_idx, we can safely use it to
+		 * remember which counter to decrement in the matching
+		 * rcu_read_unlock().
+		 */
+
+		ACCESS_ONCE(t->rcu_flipctr_idx) = idx;
+		local_irq_restore(flags);
+	}
+}
+EXPORT_SYMBOL_GPL(__rcu_read_lock);
+
+void __rcu_read_unlock(void)
+{
+	int idx;
+	struct task_struct *t = current;
+	int nesting;
+
+	nesting = ACCESS_ONCE(t->rcu_read_lock_nesting);
+	if (nesting > 1) {
+
+		/*
+		 * We are still protected by the enclosing rcu_read_lock(),
+		 * so simply decrement the counter.
+		 */
+
+		t->rcu_read_lock_nesting = nesting - 1;
+
+	} else {
+		unsigned long flags;
+
+		/*
+		 * Disable local interrupts to prevent the grace-period
+		 * detection state machine from seeing us half-done.
+		 * NMIs can still occur, of course, and might themselves
+		 * contain rcu_read_lock() and rcu_read_unlock().
+		 */
+
+		local_irq_save(flags);
+
+		/*
+		 * Outermost nesting of rcu_read_unlock(), so we must
+		 * decrement the current counter for the current CPU.
+		 * This must be done carefully, because NMIs can
+		 * occur at any point in this code, and any rcu_read_lock()
+		 * and rcu_read_unlock() pairs in the NMI handlers
+		 * must interact non-destructively with this code.
+		 * Lots of volatile casts, and -very- careful ordering.
+		 *
+		 * Changes to this code, including this one, must be
+		 * inspected, validated, and tested extremely carefully!!!
+		 */
+
+		/*
+		 * First, pick up the index.
+		 */
+
+		idx = ACCESS_ONCE(t->rcu_flipctr_idx);
+
+		/*
+		 * Now that we have fetched the counter index, it is
+		 * safe to decrement the per-task RCU nesting counter.
+		 * After this, any interrupts or NMIs will increment and
+		 * decrement the per-CPU counters.
+		 */
+		ACCESS_ONCE(t->rcu_read_lock_nesting) = nesting - 1;
+
+		/*
+		 * It is now safe to decrement this task's nesting count.
+		 * NMIs that occur after this statement will route their
+		 * rcu_read_lock() calls through this "else" clause, and
+		 * will thus start incrementing the per-CPU counter on
+		 * their own.  They will also clobber ->rcu_flipctr_idx,
+		 * but that is OK, since we have already fetched it.
+		 */
+
+		ACCESS_ONCE(RCU_DATA_ME()->rcu_flipctr[idx])--;
+		local_irq_restore(flags);
+	}
+}
+EXPORT_SYMBOL_GPL(__rcu_read_unlock);
+
+/*
+ * If a global counter flip has occurred since the last time that we
+ * advanced callbacks, advance them.  Hardware interrupts must be
+ * disabled when calling this function.
+ */
+static void __rcu_advance_callbacks(struct rcu_data *rdp)
+{
+	int cpu;
+	int i;
+	int wlc = 0;
+
+	if (rdp->completed != rcu_ctrlblk.completed) {
+		if (rdp->waitlist[GP_STAGES - 1] != NULL) {
+			*rdp->donetail = rdp->waitlist[GP_STAGES - 1];
+			rdp->donetail = rdp->waittail[GP_STAGES - 1];
+			RCU_TRACE_RDP(rcupreempt_trace_move2done, rdp);
+		}
+		for (i = GP_STAGES - 2; i >= 0; i--) {
+			if (rdp->waitlist[i] != NULL) {
+				rdp->waitlist[i + 1] = rdp->waitlist[i];
+				rdp->waittail[i + 1] = rdp->waittail[i];
+				wlc++;
+			} else {
+				rdp->waitlist[i + 1] = NULL;
+				rdp->waittail[i + 1] =
+					&rdp->waitlist[i + 1];
+			}
+		}
+		if (rdp->nextlist != NULL) {
+			rdp->waitlist[0] = rdp->nextlist;
+			rdp->waittail[0] = rdp->nexttail;
+			wlc++;
+			rdp->nextlist = NULL;
+			rdp->nexttail = &rdp->nextlist;
+			RCU_TRACE_RDP(rcupreempt_trace_move2wait, rdp);
+		} else {
+			rdp->waitlist[0] = NULL;
+			rdp->waittail[0] = &rdp->waitlist[0];
+		}
+		rdp->waitlistcount = wlc;
+		rdp->completed = rcu_ctrlblk.completed;
+	}
+
+	/*
+	 * Check to see if this CPU needs to report that it has seen
+	 * the most recent counter flip, thereby declaring that all
+	 * subsequent rcu_read_lock() invocations will respect this flip.
+	 */
+
+	cpu = raw_smp_processor_id();
+	if (per_cpu(rcu_flip_flag, cpu) == rcu_flipped) {
+		smp_mb();  /* Subsequent counter accesses must see new value */
+		per_cpu(rcu_flip_flag, cpu) = rcu_flip_seen;
+		smp_mb();  /* Subsequent RCU read-side critical sections */
+			   /*  seen -after- acknowledgement. */
+	}
+}
+
+/*
+ * Get here when RCU is idle.  Decide whether we need to
+ * move out of idle state, and return non-zero if so.
+ * "Straightforward" approach for the moment, might later
+ * use callback-list lengths, grace-period duration, or
+ * some such to determine when to exit idle state.
+ * Might also need a pre-idle test that does not acquire
+ * the lock, but let's get the simple case working first...
+ */
+
+static int
+rcu_try_flip_idle(void)
+{
+	int cpu;
+
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_i1);
+	if (!rcu_pending(smp_processor_id())) {
+		RCU_TRACE_ME(rcupreempt_trace_try_flip_ie1);
+		return 0;
+	}
+
+	/*
+	 * Do the flip.
+	 */
+
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_g1);
+	rcu_ctrlblk.completed++;  /* stands in for rcu_try_flip_g2 */
+
+	/*
+	 * Need a memory barrier so that other CPUs see the new
+	 * counter value before they see the subsequent change of all
+	 * the rcu_flip_flag instances to rcu_flipped.
+	 */
+
+	smp_mb();	/* see above block comment. */
+
+	/* Now ask each CPU for acknowledgement of the flip. */
+
+	for_each_possible_cpu(cpu)
+		per_cpu(rcu_flip_flag, cpu) = rcu_flipped;
+
+	return 1;
+}
+
+/*
+ * Wait for CPUs to acknowledge the flip.
+ */
+
+static int
+rcu_try_flip_waitack(void)
+{
+	int cpu;
+
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_a1);
+	for_each_possible_cpu(cpu)
+		if (per_cpu(rcu_flip_flag, cpu) != rcu_flip_seen) {
+			RCU_TRACE_ME(rcupreempt_trace_try_flip_ae1);
+			return 0;
+		}
+
+	/*
+	 * Make sure our checks above don't bleed into subsequent
+	 * waiting for the sum of the counters to reach zero.
+	 */
+
+	smp_mb();	/* see above block comment. */
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_a2);
+	return 1;
+}
+
+/*
+ * Wait for collective ``last'' counter to reach zero,
+ * then tell all CPUs to do an end-of-grace-period memory barrier.
+ */
+
+static int
+rcu_try_flip_waitzero(void)
+{
+	int cpu;
+	int lastidx = !(rcu_ctrlblk.completed & 0x1);
+	int sum = 0;
+
+	/* Check to see if the sum of the "last" counters is zero. */
+
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_z1);
+	for_each_possible_cpu(cpu)
+		sum += RCU_DATA_CPU(cpu)->rcu_flipctr[lastidx];
+	if (sum != 0) {
+		RCU_TRACE_ME(rcupreempt_trace_try_flip_ze1);
+		return 0;
+	}
+
+	/*
+	 * This ensures that the other CPUs see the call for
+	 * memory barriers -after- the sum to zero has been
+	 * detected here
+	 */
+	smp_mb();  /*  ^^^^^^^^^^^^ */
+
+	/* Call for a memory barrier from each CPU. */
+	for_each_possible_cpu(cpu)
+		per_cpu(rcu_mb_flag, cpu) = rcu_mb_needed;
+
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_z2);
+	return 1;
+}
+
+/*
+ * Wait for all CPUs to do their end-of-grace-period memory barrier.
+ * Return 0 once all CPUs have done so.
+ */
+
+static int
+rcu_try_flip_waitmb(void)
+{
+	int cpu;
+
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_m1);
+	for_each_possible_cpu(cpu)
+		if (per_cpu(rcu_mb_flag, cpu) != rcu_mb_done) {
+			RCU_TRACE_ME(rcupreempt_trace_try_flip_me1);
+			return 0;
+		}
+
+	smp_mb(); /* Ensure that the above checks precede any following flip. */
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_m2);
+	return 1;
+}
+
+/*
+ * Attempt a single flip of the counters.  Remember, a single flip does
+ * -not- constitute a grace period.  Instead, the interval between
+ * at least GP_STAGES consecutive flips is a grace period.
+ *
+ * If anyone is nuts enough to run this CONFIG_PREEMPT_RCU implementation
+ * on a large SMP, they might want to use a hierarchical organization of
+ * the per-CPU-counter pairs.
+ */
+static void rcu_try_flip(void)
+{
+	unsigned long flags;
+
+	RCU_TRACE_ME(rcupreempt_trace_try_flip_1);
+	if (unlikely(!spin_trylock_irqsave(&rcu_ctrlblk.fliplock, flags))) {
+		RCU_TRACE_ME(rcupreempt_trace_try_flip_e1);
+		return;
+	}
+
+	/*
+	 * Take the next transition(s) through the RCU grace-period
+	 * flip-counter state machine.
+	 */
+
+	switch (rcu_ctrlblk.rcu_try_flip_state) {
+	case rcu_try_flip_idle_state:
+		if (rcu_try_flip_idle())
+			rcu_ctrlblk.rcu_try_flip_state =
+				rcu_try_flip_waitack_state;
+		break;
+	case rcu_try_flip_waitack_state:
+		if (rcu_try_flip_waitack())
+			rcu_ctrlblk.rcu_try_flip_state =
+				rcu_try_flip_waitzero_state;
+		break;
+	case rcu_try_flip_waitzero_state:
+		if (rcu_try_flip_waitzero())
+			rcu_ctrlblk.rcu_try_flip_state =
+				rcu_try_flip_waitmb_state;
+		break;
+	case rcu_try_flip_waitmb_state:
+		if (rcu_try_flip_waitmb())
+			rcu_ctrlblk.rcu_try_flip_state =
+				rcu_try_flip_idle_state;
+	}
+	spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);
+}
+
+/*
+ * Check to see if this CPU needs to do a memory barrier in order to
+ * ensure that any prior RCU read-side critical sections have committed
+ * their counter manipulations and critical-section memory references
+ * before declaring the grace period to be completed.
+ */
+static void rcu_check_mb(int cpu)
+{
+	if (per_cpu(rcu_mb_flag, cpu) == rcu_mb_needed) {
+		smp_mb();  /* Ensure RCU read-side accesses are visible. */
+		per_cpu(rcu_mb_flag, cpu) = rcu_mb_done;
+	}
+}
+
+void rcu_check_callbacks(int cpu, int user)
+{
+	unsigned long flags;
+	struct rcu_data *rdp = RCU_DATA_CPU(cpu);
+
+	rcu_check_mb(cpu);
+	if (rcu_ctrlblk.completed == rdp->completed)
+		rcu_try_flip();
+	spin_lock_irqsave(&rdp->lock, flags);
+	RCU_TRACE_RDP(rcupreempt_trace_check_callbacks, rdp);
+	__rcu_advance_callbacks(rdp);
+	if (rdp->donelist == NULL) {
+		spin_unlock_irqrestore(&rdp->lock, flags);
+	} else {
+		spin_unlock_irqrestore(&rdp->lock, flags);
+		raise_softirq(RCU_SOFTIRQ);
+	}
+}
+
+/*
+ * Needed by dynticks, to make sure all RCU processing has finished
+ * when we go idle:
+ */
+void rcu_advance_callbacks(int cpu, int user)
+{
+	unsigned long flags;
+	struct rcu_data *rdp = RCU_DATA_CPU(cpu);
+
+	if (rcu_ctrlblk.completed == rdp->completed) {
+		rcu_try_flip();
+		if (rcu_ctrlblk.completed == rdp->completed)
+			return;
+	}
+	spin_lock_irqsave(&rdp->lock, flags);
+	RCU_TRACE_RDP(rcupreempt_trace_check_callbacks, rdp);
+	__rcu_advance_callbacks(rdp);
+	spin_unlock_irqrestore(&rdp->lock, flags);
+}
+
+static void rcu_process_callbacks(struct softirq_action *unused)
+{
+	unsigned long flags;
+	struct rcu_head *next, *list;
+	struct rcu_data *rdp = RCU_DATA_ME();
+
+	spin_lock_irqsave(&rdp->lock, flags);
+	list = rdp->donelist;
+	if (list == NULL) {
+		spin_unlock_irqrestore(&rdp->lock, flags);
+		return;
+	}
+	rdp->donelist = NULL;
+	rdp->donetail = &rdp->donelist;
+	RCU_TRACE_RDP(rcupreempt_trace_done_remove, rdp);
+	spin_unlock_irqrestore(&rdp->lock, flags);
+	while (list) {
+		next = list->next;
+		list->func(list);
+		list = next;
+		RCU_TRACE_ME(rcupreempt_trace_invoke);
+	}
+}
+
+void 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 = RCU_DATA_ME();
+	spin_lock(&rdp->lock);
+	__rcu_advance_callbacks(rdp);
+	*rdp->nexttail = head;
+	rdp->nexttail = &head->next;
+	RCU_TRACE_RDP(rcupreempt_trace_next_add, rdp);
+	spin_unlock(&rdp->lock);
+	local_irq_restore(flags);
+}
+EXPORT_SYMBOL_GPL(call_rcu);
+
+/*
+ * Wait until all currently running preempt_disable() code segments
+ * (including hardware-irq-disable segments) complete.  Note that
+ * in -rt this does -not- necessarily result in all currently executing
+ * interrupt -handlers- having completed.
+ */
+void __synchronize_sched(void)
+{
+	cpumask_t oldmask;
+	int cpu;
+
+	if (sched_getaffinity(0, &oldmask) < 0)
+		oldmask = cpu_possible_map;
+	for_each_online_cpu(cpu) {
+		sched_setaffinity(0, cpumask_of_cpu(cpu));
+		schedule();
+	}
+	sched_setaffinity(0, oldmask);
+}
+EXPORT_SYMBOL_GPL(__synchronize_sched);
+
+/*
+ * Check to see if any future RCU-related work will need to be done
+ * by the current CPU, even if none need be done immediately, returning
+ * 1 if so.  Assumes that notifiers would take care of handling any
+ * outstanding requests from the RCU core.
+ *
+ * This function is part of the RCU implementation; it is -not-
+ * an exported member of the RCU API.
+ */
+int rcu_needs_cpu(int cpu)
+{
+	struct rcu_data *rdp = RCU_DATA_CPU(cpu);
+
+	return (rdp->donelist != NULL ||
+		!!rdp->waitlistcount ||
+		rdp->nextlist != NULL);
+}
+
+int rcu_pending(int cpu)
+{
+	struct rcu_data *rdp = RCU_DATA_CPU(cpu);
+
+	/* The CPU has at least one callback queued somewhere. */
+
+	if (rdp->donelist != NULL ||
+	    !!rdp->waitlistcount ||
+	    rdp->nextlist != NULL)
+		return 1;
+
+	/* The RCU core needs an acknowledgement from this CPU. */
+
+	if ((per_cpu(rcu_flip_flag, cpu) == rcu_flipped) ||
+	    (per_cpu(rcu_mb_flag, cpu) == rcu_mb_needed))
+		return 1;
+
+	/* This CPU has fallen behind the global grace-period number. */
+
+	if (rdp->completed != rcu_ctrlblk.completed)
+		return 1;
+
+	/* Nothing needed from this CPU. */
+
+	return 0;
+}
+
+void __init __rcu_init(void)
+{
+	int cpu;
+	int i;
+	struct rcu_data *rdp;
+
+	printk(KERN_NOTICE "Preemptible RCU implementation.\n");
+	for_each_possible_cpu(cpu) {
+		rdp = RCU_DATA_CPU(cpu);
+		spin_lock_init(&rdp->lock);
+		rdp->completed = 0;
+		rdp->waitlistcount = 0;
+		rdp->nextlist = NULL;
+		rdp->nexttail = &rdp->nextlist;
+		for (i = 0; i < GP_STAGES; i++) {
+			rdp->waitlist[i] = NULL;
+			rdp->waittail[i] = &rdp->waitlist[i];
+		}
+		rdp->donelist = NULL;
+		rdp->donetail = &rdp->donelist;
+		rdp->rcu_flipctr[0] = 0;
+		rdp->rcu_flipctr[1] = 0;
+	}
+	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks, NULL);
+}
+
+/*
+ * Deprecated, use synchronize_rcu() or synchronize_sched() instead.
+ */
+void synchronize_kernel(void)
+{
+	synchronize_rcu();
+}
+
+#ifdef CONFIG_RCU_TRACE
+long *rcupreempt_flipctr(int cpu)
+{
+	return &RCU_DATA_CPU(cpu)->rcu_flipctr[0];
+}
+EXPORT_SYMBOL_GPL(rcupreempt_flipctr);
+
+int rcupreempt_flip_flag(int cpu)
+{
+	return per_cpu(rcu_flip_flag, cpu);
+}
+EXPORT_SYMBOL_GPL(rcupreempt_flip_flag);
+
+int rcupreempt_mb_flag(int cpu)
+{
+	return per_cpu(rcu_mb_flag, cpu);
+}
+EXPORT_SYMBOL_GPL(rcupreempt_mb_flag);
+
+char *rcupreempt_try_flip_state_name(void)
+{
+	return rcu_try_flip_state_names[rcu_ctrlblk.rcu_try_flip_state];
+}
+EXPORT_SYMBOL_GPL(rcupreempt_try_flip_state_name);
+
+struct rcupreempt_trace *rcupreempt_trace_cpu(int cpu)
+{
+	struct rcu_data *rdp = RCU_DATA_CPU(cpu);
+
+	return &rdp->trace;
+}
+EXPORT_SYMBOL_GPL(rcupreempt_trace_cpu);
+
+#endif /* #ifdef RCU_TRACE */
diff --git a/kernel/rcupreempt_trace.c b/kernel/rcupreempt_trace.c
new file mode 100644
index 000000000000..49ac4947af24
--- /dev/null
+++ b/kernel/rcupreempt_trace.c
@@ -0,0 +1,330 @@
+/*
+ * Read-Copy Update tracing for realtime implementation
+ *
+ * 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, 2006
+ *
+ * Papers:  http://www.rdrop.com/users/paulmck/RCU
+ *
+ * For detailed explanation of Read-Copy Update mechanism see -
+ * 		Documentation/RCU/ *.txt
+ *
+ */
+#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>
+#include <linux/mutex.h>
+#include <linux/rcupreempt_trace.h>
+#include <linux/debugfs.h>
+
+static struct mutex rcupreempt_trace_mutex;
+static char *rcupreempt_trace_buf;
+#define RCUPREEMPT_TRACE_BUF_SIZE 4096
+
+void rcupreempt_trace_move2done(struct rcupreempt_trace *trace)
+{
+	trace->done_length += trace->wait_length;
+	trace->done_add += trace->wait_length;
+	trace->wait_length = 0;
+}
+void rcupreempt_trace_move2wait(struct rcupreempt_trace *trace)
+{
+	trace->wait_length += trace->next_length;
+	trace->wait_add += trace->next_length;
+	trace->next_length = 0;
+}
+void rcupreempt_trace_try_flip_1(struct rcupreempt_trace *trace)
+{
+	atomic_inc(&trace->rcu_try_flip_1);
+}
+void rcupreempt_trace_try_flip_e1(struct rcupreempt_trace *trace)
+{
+	atomic_inc(&trace->rcu_try_flip_e1);
+}
+void rcupreempt_trace_try_flip_i1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_i1++;
+}
+void rcupreempt_trace_try_flip_ie1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_ie1++;
+}
+void rcupreempt_trace_try_flip_g1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_g1++;
+}
+void rcupreempt_trace_try_flip_a1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_a1++;
+}
+void rcupreempt_trace_try_flip_ae1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_ae1++;
+}
+void rcupreempt_trace_try_flip_a2(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_a2++;
+}
+void rcupreempt_trace_try_flip_z1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_z1++;
+}
+void rcupreempt_trace_try_flip_ze1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_ze1++;
+}
+void rcupreempt_trace_try_flip_z2(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_z2++;
+}
+void rcupreempt_trace_try_flip_m1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_m1++;
+}
+void rcupreempt_trace_try_flip_me1(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_me1++;
+}
+void rcupreempt_trace_try_flip_m2(struct rcupreempt_trace *trace)
+{
+	trace->rcu_try_flip_m2++;
+}
+void rcupreempt_trace_check_callbacks(struct rcupreempt_trace *trace)
+{
+	trace->rcu_check_callbacks++;
+}
+void rcupreempt_trace_done_remove(struct rcupreempt_trace *trace)
+{
+	trace->done_remove += trace->done_length;
+	trace->done_length = 0;
+}
+void rcupreempt_trace_invoke(struct rcupreempt_trace *trace)
+{
+	atomic_inc(&trace->done_invoked);
+}
+void rcupreempt_trace_next_add(struct rcupreempt_trace *trace)
+{
+	trace->next_add++;
+	trace->next_length++;
+}
+
+static void rcupreempt_trace_sum(struct rcupreempt_trace *sp)
+{
+	struct rcupreempt_trace *cp;
+	int cpu;
+
+	memset(sp, 0, sizeof(*sp));
+	for_each_possible_cpu(cpu) {
+		cp = rcupreempt_trace_cpu(cpu);
+		sp->next_length += cp->next_length;
+		sp->next_add += cp->next_add;
+		sp->wait_length += cp->wait_length;
+		sp->wait_add += cp->wait_add;
+		sp->done_length += cp->done_length;
+		sp->done_add += cp->done_add;
+		sp->done_remove += cp->done_remove;
+		atomic_set(&sp->done_invoked, atomic_read(&cp->done_invoked));
+		sp->rcu_check_callbacks += cp->rcu_check_callbacks;
+		atomic_set(&sp->rcu_try_flip_1,
+			   atomic_read(&cp->rcu_try_flip_1));
+		atomic_set(&sp->rcu_try_flip_e1,
+			   atomic_read(&cp->rcu_try_flip_e1));
+		sp->rcu_try_flip_i1 += cp->rcu_try_flip_i1;
+		sp->rcu_try_flip_ie1 += cp->rcu_try_flip_ie1;
+		sp->rcu_try_flip_g1 += cp->rcu_try_flip_g1;
+		sp->rcu_try_flip_a1 += cp->rcu_try_flip_a1;
+		sp->rcu_try_flip_ae1 += cp->rcu_try_flip_ae1;
+		sp->rcu_try_flip_a2 += cp->rcu_try_flip_a2;
+		sp->rcu_try_flip_z1 += cp->rcu_try_flip_z1;
+		sp->rcu_try_flip_ze1 += cp->rcu_try_flip_ze1;
+		sp->rcu_try_flip_z2 += cp->rcu_try_flip_z2;
+		sp->rcu_try_flip_m1 += cp->rcu_try_flip_m1;
+		sp->rcu_try_flip_me1 += cp->rcu_try_flip_me1;
+		sp->rcu_try_flip_m2 += cp->rcu_try_flip_m2;
+	}
+}
+
+static ssize_t rcustats_read(struct file *filp, char __user *buffer,
+				size_t count, loff_t *ppos)
+{
+	struct rcupreempt_trace trace;
+	ssize_t bcount;
+	int cnt = 0;
+
+	rcupreempt_trace_sum(&trace);
+	mutex_lock(&rcupreempt_trace_mutex);
+	snprintf(&rcupreempt_trace_buf[cnt], RCUPREEMPT_TRACE_BUF_SIZE - cnt,
+		 "ggp=%ld rcc=%ld\n",
+		 rcu_batches_completed(),
+		 trace.rcu_check_callbacks);
+	snprintf(&rcupreempt_trace_buf[cnt], RCUPREEMPT_TRACE_BUF_SIZE - cnt,
+		 "na=%ld nl=%ld wa=%ld wl=%ld da=%ld dl=%ld dr=%ld di=%d\n"
+		 "1=%d e1=%d i1=%ld ie1=%ld g1=%ld a1=%ld ae1=%ld a2=%ld\n"
+		 "z1=%ld ze1=%ld z2=%ld m1=%ld me1=%ld m2=%ld\n",
+
+		 trace.next_add, trace.next_length,
+		 trace.wait_add, trace.wait_length,
+		 trace.done_add, trace.done_length,
+		 trace.done_remove, atomic_read(&trace.done_invoked),
+		 atomic_read(&trace.rcu_try_flip_1),
+		 atomic_read(&trace.rcu_try_flip_e1),
+		 trace.rcu_try_flip_i1, trace.rcu_try_flip_ie1,
+		 trace.rcu_try_flip_g1,
+		 trace.rcu_try_flip_a1, trace.rcu_try_flip_ae1,
+			 trace.rcu_try_flip_a2,
+		 trace.rcu_try_flip_z1, trace.rcu_try_flip_ze1,
+			 trace.rcu_try_flip_z2,
+		 trace.rcu_try_flip_m1, trace.rcu_try_flip_me1,
+			trace.rcu_try_flip_m2);
+	bcount = simple_read_from_buffer(buffer, count, ppos,
+			rcupreempt_trace_buf, strlen(rcupreempt_trace_buf));
+	mutex_unlock(&rcupreempt_trace_mutex);
+	return bcount;
+}
+
+static ssize_t rcugp_read(struct file *filp, char __user *buffer,
+				size_t count, loff_t *ppos)
+{
+	long oldgp = rcu_batches_completed();
+	ssize_t bcount;
+
+	mutex_lock(&rcupreempt_trace_mutex);
+	synchronize_rcu();
+	snprintf(rcupreempt_trace_buf, RCUPREEMPT_TRACE_BUF_SIZE,
+		"oldggp=%ld  newggp=%ld\n", oldgp, rcu_batches_completed());
+	bcount = simple_read_from_buffer(buffer, count, ppos,
+			rcupreempt_trace_buf, strlen(rcupreempt_trace_buf));
+	mutex_unlock(&rcupreempt_trace_mutex);
+	return bcount;
+}
+
+static ssize_t rcuctrs_read(struct file *filp, char __user *buffer,
+				size_t count, loff_t *ppos)
+{
+	int cnt = 0;
+	int cpu;
+	int f = rcu_batches_completed() & 0x1;
+	ssize_t bcount;
+
+	mutex_lock(&rcupreempt_trace_mutex);
+
+	cnt += snprintf(&rcupreempt_trace_buf[cnt], RCUPREEMPT_TRACE_BUF_SIZE,
+				"CPU last cur F M\n");
+	for_each_online_cpu(cpu) {
+		long *flipctr = rcupreempt_flipctr(cpu);
+		cnt += snprintf(&rcupreempt_trace_buf[cnt],
+				RCUPREEMPT_TRACE_BUF_SIZE - cnt,
+					"%3d %4ld %3ld %d %d\n",
+			       cpu,
+			       flipctr[!f],
+			       flipctr[f],
+			       rcupreempt_flip_flag(cpu),
+			       rcupreempt_mb_flag(cpu));
+	}
+	cnt += snprintf(&rcupreempt_trace_buf[cnt],
+			RCUPREEMPT_TRACE_BUF_SIZE - cnt,
+			"ggp = %ld, state = %s\n",
+			rcu_batches_completed(),
+			rcupreempt_try_flip_state_name());
+	cnt += snprintf(&rcupreempt_trace_buf[cnt],
+			RCUPREEMPT_TRACE_BUF_SIZE - cnt,
+			"\n");
+	bcount = simple_read_from_buffer(buffer, count, ppos,
+			rcupreempt_trace_buf, strlen(rcupreempt_trace_buf));
+	mutex_unlock(&rcupreempt_trace_mutex);
+	return bcount;
+}
+
+static struct file_operations rcustats_fops = {
+	.owner = THIS_MODULE,
+	.read = rcustats_read,
+};
+
+static struct file_operations rcugp_fops = {
+	.owner = THIS_MODULE,
+	.read = rcugp_read,
+};
+
+static struct file_operations rcuctrs_fops = {
+	.owner = THIS_MODULE,
+	.read = rcuctrs_read,
+};
+
+static struct dentry *rcudir, *statdir, *ctrsdir, *gpdir;
+static int rcupreempt_debugfs_init(void)
+{
+	rcudir = debugfs_create_dir("rcu", NULL);
+	if (!rcudir)
+		goto out;
+	statdir = debugfs_create_file("rcustats", 0444, rcudir,
+						NULL, &rcustats_fops);
+	if (!statdir)
+		goto free_out;
+
+	gpdir = debugfs_create_file("rcugp", 0444, rcudir, NULL, &rcugp_fops);
+	if (!gpdir)
+		goto free_out;
+
+	ctrsdir = debugfs_create_file("rcuctrs", 0444, rcudir,
+						NULL, &rcuctrs_fops);
+	if (!ctrsdir)
+		goto free_out;
+	return 0;
+free_out:
+	if (statdir)
+		debugfs_remove(statdir);
+	if (gpdir)
+		debugfs_remove(gpdir);
+	debugfs_remove(rcudir);
+out:
+	return 1;
+}
+
+static int __init rcupreempt_trace_init(void)
+{
+	mutex_init(&rcupreempt_trace_mutex);
+	rcupreempt_trace_buf = kmalloc(RCUPREEMPT_TRACE_BUF_SIZE, GFP_KERNEL);
+	if (!rcupreempt_trace_buf)
+		return 1;
+	return rcupreempt_debugfs_init();
+}
+
+static void __exit rcupreempt_trace_cleanup(void)
+{
+	debugfs_remove(statdir);
+	debugfs_remove(gpdir);
+	debugfs_remove(ctrsdir);
+	debugfs_remove(rcudir);
+	kfree(rcupreempt_trace_buf);
+}
+
+
+module_init(rcupreempt_trace_init);
+module_exit(rcupreempt_trace_cleanup);