ring-buffer: Check if ring buffer iterator has stale data
[linux-2.6.git] / kernel / rcutree.c
index 1b32cdd..53ae959 100644 (file)
@@ -25,7 +25,7 @@
  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
  *
  * For detailed explanation of Read-Copy Update mechanism see -
- *     Documentation/RCU
+ *     Documentation/RCU
  */
 #include <linux/types.h>
 #include <linux/kernel.h>
 #include <linux/cpu.h>
 #include <linux/mutex.h>
 #include <linux/time.h>
+#include <linux/kernel_stat.h>
 
 #include "rcutree.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. */
 
+static struct lock_class_key rcu_node_class[NUM_RCU_LVLS];
+
 #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 */ \
+               NUM_RCU_LVL_3, \
+               NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \
        }, \
-       .signaled = RCU_SIGNAL_INIT, \
+       .signaled = RCU_GP_IDLE, \
        .gpnum = -300, \
        .completed = -300, \
        .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
+       .orphan_cbs_list = NULL, \
+       .orphan_cbs_tail = &name.orphan_cbs_list, \
+       .orphan_qlen = 0, \
        .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
        .n_force_qs = 0, \
        .n_force_qs_ngp = 0, \
@@ -81,24 +81,18 @@ DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
 struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
 
-extern long rcu_batches_completed_sched(void);
-static struct rcu_node *rcu_get_root(struct rcu_state *rsp);
-static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp,
-                         struct rcu_node *rnp, unsigned long flags);
-static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags);
-#ifdef CONFIG_HOTPLUG_CPU
-static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp);
-#endif /* #ifdef CONFIG_HOTPLUG_CPU */
-static void __rcu_process_callbacks(struct rcu_state *rsp,
-                                   struct rcu_data *rdp);
-static void __call_rcu(struct rcu_head *head,
-                      void (*func)(struct rcu_head *rcu),
-                      struct rcu_state *rsp);
-static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp);
-static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp,
-                                          int preemptable);
+static int rcu_scheduler_active __read_mostly;
 
-#include "rcutree_plugin.h"
+
+/*
+ * Return true if an RCU grace period is in progress.  The ACCESS_ONCE()s
+ * permit this function to be invoked without holding the root rcu_node
+ * structure's ->lock, but of course results can be subject to change.
+ */
+static int rcu_gp_in_progress(struct rcu_state *rsp)
+{
+       return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
+}
 
 /*
  * Note a quiescent state.  Because we do not need to know
@@ -110,7 +104,7 @@ void rcu_sched_qs(int cpu)
        struct rcu_data *rdp;
 
        rdp = &per_cpu(rcu_sched_data, cpu);
-       rdp->passed_quiesc_completed = rdp->completed;
+       rdp->passed_quiesc_completed = rdp->gpnum - 1;
        barrier();
        rdp->passed_quiesc = 1;
        rcu_preempt_note_context_switch(cpu);
@@ -121,7 +115,7 @@ void rcu_bh_qs(int cpu)
        struct rcu_data *rdp;
 
        rdp = &per_cpu(rcu_bh_data, cpu);
-       rdp->passed_quiesc_completed = rdp->completed;
+       rdp->passed_quiesc_completed = rdp->gpnum - 1;
        barrier();
        rdp->passed_quiesc = 1;
 }
@@ -137,6 +131,10 @@ 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. */
 
+module_param(blimit, int, 0);
+module_param(qhimark, int, 0);
+module_param(qlowmark, int, 0);
+
 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
 static int rcu_pending(int cpu);
 
@@ -173,9 +171,7 @@ cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
 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 *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
 }
 
 /*
@@ -345,31 +341,12 @@ void rcu_irq_exit(void)
                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.
+ * is in dynticks idle mode, which is an extended quiescent state.
  */
 static int dyntick_save_progress_counter(struct rcu_data *rdp)
 {
@@ -429,24 +406,8 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
 
 #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;
@@ -475,30 +436,34 @@ static void print_other_cpu_stall(struct rcu_state *rsp)
        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) {
+       if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
                spin_unlock_irqrestore(&rnp->lock, flags);
                return;
        }
        rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
+
+       /*
+        * Now rat on any tasks that got kicked up to the root rcu_node
+        * due to CPU offlining.
+        */
+       rcu_print_task_stall(rnp);
        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++) {
+       rcu_for_each_leaf_node(rsp, rnp) {
                rcu_print_task_stall(rnp);
-               if (rnp_cur->qsmask == 0)
+               if (rnp->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);
+               for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
+                       if (rnp->qsmask & (1UL << cpu))
+                               printk(" %d", rnp->grplo + cpu);
        }
        printk(" (detected by %d, t=%ld jiffies)\n",
               smp_processor_id(), (long)(jiffies - rsp->gp_start));
@@ -537,8 +502,7 @@ static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
                /* We haven't checked in, so go dump stack. */
                print_cpu_stall(rsp);
 
-       } else if (rsp->gpnum != rsp->completed &&
-                  delta >= RCU_STALL_RAT_DELAY) {
+       } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
 
                /* They had two time units to dump stack, so complain. */
                print_other_cpu_stall(rsp);
@@ -560,13 +524,33 @@ static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
 /*
  * 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.
+ * that someone else started the grace period.  The caller must hold the
+ * ->lock of the leaf rcu_node structure corresponding to the current CPU,
+ *  and must have irqs disabled.
  */
+static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
+{
+       if (rdp->gpnum != rnp->gpnum) {
+               rdp->qs_pending = 1;
+               rdp->passed_quiesc = 0;
+               rdp->gpnum = rnp->gpnum;
+       }
+}
+
 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;
+       unsigned long flags;
+       struct rcu_node *rnp;
+
+       local_irq_save(flags);
+       rnp = rdp->mynode;
+       if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */
+           !spin_trylock(&rnp->lock)) { /* irqs already off, retry later. */
+               local_irq_restore(flags);
+               return;
+       }
+       __note_new_gpnum(rsp, rnp, rdp);
+       spin_unlock_irqrestore(&rnp->lock, flags);
 }
 
 /*
@@ -590,6 +574,79 @@ check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
 }
 
 /*
+ * 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.  In addition, the corresponding leaf rcu_node structure's
+ * ->lock must be held by the caller, with irqs disabled.
+ */
+static void
+__rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
+{
+       /* Did another grace period end? */
+       if (rdp->completed != rnp->completed) {
+
+               /* 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 = rnp->completed;
+       }
+}
+
+/*
+ * 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)
+{
+       unsigned long flags;
+       struct rcu_node *rnp;
+
+       local_irq_save(flags);
+       rnp = rdp->mynode;
+       if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */
+           !spin_trylock(&rnp->lock)) { /* irqs already off, retry later. */
+               local_irq_restore(flags);
+               return;
+       }
+       __rcu_process_gp_end(rsp, rnp, rdp);
+       spin_unlock_irqrestore(&rnp->lock, flags);
+}
+
+/*
+ * Do per-CPU grace-period initialization for running CPU.  The caller
+ * must hold the lock of the leaf rcu_node structure corresponding to
+ * this CPU.
+ */
+static void
+rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
+{
+       /* Prior grace period ended, so advance callbacks for current CPU. */
+       __rcu_process_gp_end(rsp, rnp, rdp);
+
+       /*
+        * Because this CPU just now started the new grace period, we know
+        * that all of its callbacks will be covered by this upcoming grace
+        * period, even the ones that were registered arbitrarily recently.
+        * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
+        *
+        * Other CPUs cannot be sure exactly when the grace period started.
+        * Therefore, their recently registered callbacks must pass through
+        * an additional RCU_NEXT_READY stage, so that they will be handled
+        * by the next RCU grace period.
+        */
+       rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
+       rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
+
+       /* Set state so that this CPU will detect the next quiescent state. */
+       __note_new_gpnum(rsp, rnp, rdp);
+}
+
+/*
  * 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
@@ -603,7 +660,23 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
        struct rcu_node *rnp = rcu_get_root(rsp);
 
        if (!cpu_needs_another_gp(rsp, rdp)) {
-               spin_unlock_irqrestore(&rnp->lock, flags);
+               if (rnp->completed == rsp->completed) {
+                       spin_unlock_irqrestore(&rnp->lock, flags);
+                       return;
+               }
+               spin_unlock(&rnp->lock);         /* irqs remain disabled. */
+
+               /*
+                * Propagate new ->completed value to rcu_node structures
+                * so that other CPUs don't have to wait until the start
+                * of the next grace period to process their callbacks.
+                */
+               rcu_for_each_node_breadth_first(rsp, rnp) {
+                       spin_lock(&rnp->lock);   /* irqs already disabled. */
+                       rnp->completed = rsp->completed;
+                       spin_unlock(&rnp->lock); /* irqs remain disabled. */
+               }
+               local_irq_restore(flags);
                return;
        }
 
@@ -613,23 +686,15 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
        rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
        rsp->jiffies_force_qs = jiffies + 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) {
                rcu_preempt_check_blocked_tasks(rnp);
                rnp->qsmask = rnp->qsmaskinit;
                rnp->gpnum = rsp->gpnum;
+               rnp->completed = rsp->completed;
                rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
+               rcu_start_gp_per_cpu(rsp, rnp, rdp);
                spin_unlock_irqrestore(&rnp->lock, flags);
                return;
        }
@@ -657,70 +722,51 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
         * one corresponding to this CPU, due to the fact that we have
         * irqs disabled.
         */
-       for (rnp = &rsp->node[0]; rnp < &rsp->node[NUM_RCU_NODES]; rnp++) {
-               spin_lock(&rnp->lock);  /* irqs already disabled. */
+       rcu_for_each_node_breadth_first(rsp, rnp) {
+               spin_lock(&rnp->lock);          /* irqs already disabled. */
                rcu_preempt_check_blocked_tasks(rnp);
                rnp->qsmask = rnp->qsmaskinit;
                rnp->gpnum = rsp->gpnum;
-               spin_unlock(&rnp->lock);        /* irqs already disabled. */
+               rnp->completed = rsp->completed;
+               if (rnp == rdp->mynode)
+                       rcu_start_gp_per_cpu(rsp, rnp, rdp);
+               spin_unlock(&rnp->lock);        /* irqs remain disabled. */
        }
 
+       rnp = rcu_get_root(rsp);
+       spin_lock(&rnp->lock);                  /* irqs already disabled. */
        rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
+       spin_unlock(&rnp->lock);                /* irqs remain disabled. */
        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);
-}
-
-/*
- * Clean up after the prior grace period and let rcu_start_gp() start up
- * the next grace period if one is needed.  Note that the caller must
- * hold rnp->lock, as required by rcu_start_gp(), which will release it.
+ * Report a full set of quiescent states to the specified rcu_state
+ * data structure.  This involves cleaning up after the prior grace
+ * period and letting rcu_start_gp() start up the next grace period
+ * if one is needed.  Note that the caller must hold rnp->lock, as
+ * required by rcu_start_gp(), which will release it.
  */
-static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
-       __releases(rnp->lock)
+static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
+       __releases(rcu_get_root(rsp)->lock)
 {
-       WARN_ON_ONCE(rsp->completed == rsp->gpnum);
+       WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
        rsp->completed = rsp->gpnum;
-       rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
+       rsp->signaled = RCU_GP_IDLE;
        rcu_start_gp(rsp, flags);  /* releases root node's rnp->lock. */
 }
 
 /*
- * 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.
+ * Similar to rcu_report_qs_rdp(), for which it is a helper function.
+ * Allows quiescent states for a group of CPUs to be reported at one go
+ * to the specified rcu_node structure, though all the CPUs in the group
+ * must be represented by the same rcu_node structure (which need not be
+ * a leaf rcu_node structure, though it often will be).  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)
+rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
+                 struct rcu_node *rnp, unsigned long flags)
        __releases(rnp->lock)
 {
        struct rcu_node *rnp_c;
@@ -756,21 +802,23 @@ cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
 
        /*
         * Get here if we are the last CPU to pass through a quiescent
-        * state for this grace period.  Invoke cpu_quiet_msk_finish()
+        * state for this grace period.  Invoke rcu_report_qs_rsp()
         * to clean up and start the next grace period if one is needed.
         */
-       cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
+       rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
 }
 
 /*
- * Record a quiescent state for the specified CPU, which must either be
- * the current 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!
+ * Record a quiescent state for the specified CPU to that CPU's rcu_data
+ * structure.  This must be either called from the specified CPU, or
+ * called when the specified CPU is known to be offline (and when it is
+ * also known that no other CPU is concurrently trying to help the 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)
+rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
 {
        unsigned long flags;
        unsigned long mask;
@@ -778,15 +826,15 @@ cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
 
        rnp = rdp->mynode;
        spin_lock_irqsave(&rnp->lock, flags);
-       if (lastcomp != ACCESS_ONCE(rsp->completed)) {
+       if (lastcomp != rnp->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.
+                * CPU's bit already cleared in rcu_report_qs_rnp() if this
+                * race occurred.
                 */
                rdp->passed_quiesc = 0; /* try again later! */
                spin_unlock_irqrestore(&rnp->lock, flags);
@@ -804,7 +852,7 @@ cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
                 */
                rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
 
-               cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
+               rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
        }
 }
 
@@ -835,24 +883,73 @@ rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
        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);
+       /*
+        * Tell RCU we are done (but rcu_report_qs_rdp() will be the
+        * judge of that).
+        */
+       rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
 
 /*
+ * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the
+ * specified flavor of RCU.  The callbacks will be adopted by the next
+ * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever
+ * comes first.  Because this is invoked from the CPU_DYING notifier,
+ * irqs are already disabled.
+ */
+static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
+{
+       int i;
+       struct rcu_data *rdp = rsp->rda[smp_processor_id()];
+
+       if (rdp->nxtlist == NULL)
+               return;  /* irqs disabled, so comparison is stable. */
+       spin_lock(&rsp->onofflock);  /* irqs already disabled. */
+       *rsp->orphan_cbs_tail = rdp->nxtlist;
+       rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL];
+       rdp->nxtlist = NULL;
+       for (i = 0; i < RCU_NEXT_SIZE; i++)
+               rdp->nxttail[i] = &rdp->nxtlist;
+       rsp->orphan_qlen += rdp->qlen;
+       rdp->qlen = 0;
+       spin_unlock(&rsp->onofflock);  /* irqs remain disabled. */
+}
+
+/*
+ * Adopt previously orphaned RCU callbacks.
+ */
+static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
+{
+       unsigned long flags;
+       struct rcu_data *rdp;
+
+       spin_lock_irqsave(&rsp->onofflock, flags);
+       rdp = rsp->rda[smp_processor_id()];
+       if (rsp->orphan_cbs_list == NULL) {
+               spin_unlock_irqrestore(&rsp->onofflock, flags);
+               return;
+       }
+       *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list;
+       rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail;
+       rdp->qlen += rsp->orphan_qlen;
+       rsp->orphan_cbs_list = NULL;
+       rsp->orphan_cbs_tail = &rsp->orphan_cbs_list;
+       rsp->orphan_qlen = 0;
+       spin_unlock_irqrestore(&rsp->onofflock, flags);
+}
+
+/*
  * 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;
+       int need_report = 0;
        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. */
@@ -865,42 +962,34 @@ static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
                spin_lock(&rnp->lock);          /* irqs already disabled. */
                rnp->qsmaskinit &= ~mask;
                if (rnp->qsmaskinit != 0) {
-                       spin_unlock(&rnp->lock); /* irqs remain disabled. */
+                       if (rnp != rdp->mynode)
+                               spin_unlock(&rnp->lock); /* irqs remain disabled. */
                        break;
                }
-               rcu_preempt_offline_tasks(rsp, rnp, rdp);
+               if (rnp == rdp->mynode)
+                       need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp);
+               else
+                       spin_unlock(&rnp->lock); /* irqs remain disabled. */
                mask = rnp->grpmask;
-               spin_unlock(&rnp->lock);        /* irqs remain disabled. */
                rnp = rnp->parent;
        } while (rnp != NULL);
-       lastcomp = rsp->completed;
-
-       spin_unlock(&rsp->onofflock);           /* irqs remain disabled. */
 
        /*
-        * 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 its 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.
+        * We still hold the leaf rcu_node structure lock here, and
+        * irqs are still disabled.  The reason for this subterfuge is
+        * because invoking rcu_report_unblock_qs_rnp() with ->onofflock
+        * held leads to deadlock.
         */
-       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);
+       spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
+       rnp = rdp->mynode;
+       if (need_report & RCU_OFL_TASKS_NORM_GP)
+               rcu_report_unblock_qs_rnp(rnp, flags);
+       else
+               spin_unlock_irqrestore(&rnp->lock, flags);
+       if (need_report & RCU_OFL_TASKS_EXP_GP)
+               rcu_report_exp_rnp(rsp, rnp);
+
+       rcu_adopt_orphan_cbs(rsp);
 }
 
 /*
@@ -918,6 +1007,14 @@ static void rcu_offline_cpu(int cpu)
 
 #else /* #ifdef CONFIG_HOTPLUG_CPU */
 
+static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
+{
+}
+
+static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
+{
+}
+
 static void rcu_offline_cpu(int cpu)
 {
 }
@@ -928,7 +1025,7 @@ static void rcu_offline_cpu(int 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)
+static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
 {
        unsigned long flags;
        struct rcu_head *next, *list, **tail;
@@ -981,6 +1078,13 @@ static void rcu_do_batch(struct rcu_data *rdp)
        if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
                rdp->blimit = blimit;
 
+       /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
+       if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
+               rdp->qlen_last_fqs_check = 0;
+               rdp->n_force_qs_snap = rsp->n_force_qs;
+       } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
+               rdp->qlen_last_fqs_check = rdp->qlen;
+
        local_irq_restore(flags);
 
        /* Re-raise the RCU softirq if there are callbacks remaining. */
@@ -1050,33 +1154,32 @@ static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
        int cpu;
        unsigned long flags;
        unsigned long mask;
-       struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
-       struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
+       struct rcu_node *rnp;
 
-       for (; rnp_cur < rnp_end; rnp_cur++) {
+       rcu_for_each_leaf_node(rsp, rnp) {
                mask = 0;
-               spin_lock_irqsave(&rnp_cur->lock, flags);
-               if (rsp->completed != lastcomp) {
-                       spin_unlock_irqrestore(&rnp_cur->lock, flags);
+               spin_lock_irqsave(&rnp->lock, flags);
+               if (rnp->completed != lastcomp) {
+                       spin_unlock_irqrestore(&rnp->lock, flags);
                        return 1;
                }
-               if (rnp_cur->qsmask == 0) {
-                       spin_unlock_irqrestore(&rnp_cur->lock, flags);
+               if (rnp->qsmask == 0) {
+                       spin_unlock_irqrestore(&rnp->lock, flags);
                        continue;
                }
-               cpu = rnp_cur->grplo;
+               cpu = rnp->grplo;
                bit = 1;
-               for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
-                       if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
+               for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
+                       if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu]))
                                mask |= bit;
                }
-               if (mask != 0 && rsp->completed == lastcomp) {
+               if (mask != 0 && rnp->completed == lastcomp) {
 
-                       /* cpu_quiet_msk() releases rnp_cur->lock. */
-                       cpu_quiet_msk(mask, rsp, rnp_cur, flags);
+                       /* rcu_report_qs_rnp() releases rnp->lock. */
+                       rcu_report_qs_rnp(mask, rsp, rnp, flags);
                        continue;
                }
-               spin_unlock_irqrestore(&rnp_cur->lock, flags);
+               spin_unlock_irqrestore(&rnp->lock, flags);
        }
        return 0;
 }
@@ -1091,8 +1194,9 @@ static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
        long lastcomp;
        struct rcu_node *rnp = rcu_get_root(rsp);
        u8 signaled;
+       u8 forcenow;
 
-       if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
+       if (!rcu_gp_in_progress(rsp))
                return;  /* No grace period in progress, nothing to force. */
        if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
                rsp->n_force_qs_lh++; /* Inexact, can lose counts.  Tough! */
@@ -1103,19 +1207,20 @@ static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
                goto unlock_ret; /* no emergency and done recently. */
        rsp->n_force_qs++;
        spin_lock(&rnp->lock);
-       lastcomp = rsp->completed;
+       lastcomp = rsp->gpnum - 1;
        signaled = rsp->signaled;
        rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
-       if (lastcomp == rsp->gpnum) {
+       if(!rcu_gp_in_progress(rsp)) {
                rsp->n_force_qs_ngp++;
                spin_unlock(&rnp->lock);
                goto unlock_ret;  /* no GP in progress, time updated. */
        }
        spin_unlock(&rnp->lock);
        switch (signaled) {
+       case RCU_GP_IDLE:
        case RCU_GP_INIT:
 
-               break; /* grace period still initializing, ignore. */
+               break; /* grace period idle or initializing, ignore. */
 
        case RCU_SAVE_DYNTICK:
 
@@ -1126,20 +1231,29 @@ static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
                if (rcu_process_dyntick(rsp, lastcomp,
                                        dyntick_save_progress_counter))
                        goto unlock_ret;
+               /* fall into next case. */
+
+       case RCU_SAVE_COMPLETED:
 
                /* Update state, record completion counter. */
+               forcenow = 0;
                spin_lock(&rnp->lock);
-               if (lastcomp == rsp->completed) {
+               if (lastcomp + 1 == rsp->gpnum &&
+                   lastcomp == rsp->completed &&
+                   rsp->signaled == signaled) {
                        rsp->signaled = RCU_FORCE_QS;
-                       dyntick_record_completed(rsp, lastcomp);
+                       rsp->completed_fqs = lastcomp;
+                       forcenow = signaled == RCU_SAVE_COMPLETED;
                }
                spin_unlock(&rnp->lock);
-               break;
+               if (!forcenow)
+                       break;
+               /* fall into next case. */
 
        case RCU_FORCE_QS:
 
                /* Check dyntick-idle state, send IPI to laggarts. */
-               if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
+               if (rcu_process_dyntick(rsp, rsp->completed_fqs,
                                        rcu_implicit_dynticks_qs))
                        goto unlock_ret;
 
@@ -1195,7 +1309,7 @@ __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
        }
 
        /* If there are callbacks ready, invoke them. */
-       rcu_do_batch(rdp);
+       rcu_do_batch(rsp, rdp);
 }
 
 /*
@@ -1251,7 +1365,7 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
        rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
 
        /* Start a new grace period if one not already started. */
-       if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
+       if (!rcu_gp_in_progress(rsp)) {
                unsigned long nestflag;
                struct rcu_node *rnp_root = rcu_get_root(rsp);
 
@@ -1259,10 +1373,20 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
                rcu_start_gp(rsp, nestflag);  /* releases rnp_root->lock. */
        }
 
-       /* Force the grace period if too many callbacks or too long waiting. */
-       if (unlikely(++rdp->qlen > qhimark)) {
+       /*
+        * Force the grace period if too many callbacks or too long waiting.
+        * Enforce hysteresis, and don't invoke force_quiescent_state()
+        * if some other CPU has recently done so.  Also, don't bother
+        * invoking force_quiescent_state() if the newly enqueued callback
+        * is the only one waiting for a grace period to complete.
+        */
+       if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
                rdp->blimit = LONG_MAX;
-               force_quiescent_state(rsp, 0);
+               if (rsp->n_force_qs == rdp->n_force_qs_snap &&
+                   *rdp->nxttail[RCU_DONE_TAIL] != head)
+                       force_quiescent_state(rsp, 0);
+               rdp->n_force_qs_snap = rsp->n_force_qs;
+               rdp->qlen_last_fqs_check = rdp->qlen;
        } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
                force_quiescent_state(rsp, 1);
        local_irq_restore(flags);
@@ -1286,6 +1410,68 @@ void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
 }
 EXPORT_SYMBOL_GPL(call_rcu_bh);
 
+/**
+ * synchronize_sched - wait until an rcu-sched grace period has elapsed.
+ *
+ * Control will return to the caller some time after a full rcu-sched
+ * grace period has elapsed, in other words after all currently executing
+ * rcu-sched read-side critical sections have completed.   These read-side
+ * critical sections are delimited by rcu_read_lock_sched() and
+ * rcu_read_unlock_sched(), and may be nested.  Note that preempt_disable(),
+ * local_irq_disable(), and so on may be used in place of
+ * rcu_read_lock_sched().
+ *
+ * This means that all preempt_disable code sequences, including NMI and
+ * hardware-interrupt handlers, in progress on entry will have completed
+ * before this primitive returns.  However, this does not guarantee that
+ * softirq handlers will have completed, since in some kernels, these
+ * handlers can run in process context, and can block.
+ *
+ * This primitive provides the guarantees made by the (now removed)
+ * synchronize_kernel() API.  In contrast, synchronize_rcu() only
+ * guarantees that rcu_read_lock() sections will have completed.
+ * In "classic RCU", these two guarantees happen to be one and
+ * the same, but can differ in realtime RCU implementations.
+ */
+void synchronize_sched(void)
+{
+       struct rcu_synchronize rcu;
+
+       if (rcu_blocking_is_gp())
+               return;
+
+       init_completion(&rcu.completion);
+       /* Will wake me after RCU finished. */
+       call_rcu_sched(&rcu.head, wakeme_after_rcu);
+       /* Wait for it. */
+       wait_for_completion(&rcu.completion);
+}
+EXPORT_SYMBOL_GPL(synchronize_sched);
+
+/**
+ * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
+ *
+ * Control will return to the caller some time after a full rcu_bh grace
+ * period has elapsed, in other words after all currently executing rcu_bh
+ * read-side critical sections have completed.  RCU read-side critical
+ * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
+ * and may be nested.
+ */
+void synchronize_rcu_bh(void)
+{
+       struct rcu_synchronize rcu;
+
+       if (rcu_blocking_is_gp())
+               return;
+
+       init_completion(&rcu.completion);
+       /* Will wake me after RCU finished. */
+       call_rcu_bh(&rcu.head, wakeme_after_rcu);
+       /* Wait for it. */
+       wait_for_completion(&rcu.completion);
+}
+EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
+
 /*
  * Check to see if there is any immediate RCU-related work to be done
  * by the current CPU, for the specified type of RCU, returning 1 if so.
@@ -1295,6 +1481,8 @@ EXPORT_SYMBOL_GPL(call_rcu_bh);
  */
 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
 {
+       struct rcu_node *rnp = rdp->mynode;
+
        rdp->n_rcu_pending++;
 
        /* Check for CPU stalls, if enabled. */
@@ -1319,19 +1507,19 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
        }
 
        /* Has another RCU grace period completed?  */
-       if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
+       if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
                rdp->n_rp_gp_completed++;
                return 1;
        }
 
        /* Has a new RCU grace period started? */
-       if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
+       if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
                rdp->n_rp_gp_started++;
                return 1;
        }
 
        /* Has an RCU GP gone long enough to send resched IPIs &c? */
-       if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
+       if (rcu_gp_in_progress(rsp) &&
            ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
                rdp->n_rp_need_fqs++;
                return 1;
@@ -1369,6 +1557,97 @@ int rcu_needs_cpu(int cpu)
 }
 
 /*
+ * This function is invoked towards the end of the scheduler's initialization
+ * process.  Before this is called, the idle task might contain
+ * RCU read-side critical sections (during which time, this idle
+ * task is booting the system).  After this function is called, the
+ * idle tasks are prohibited from containing RCU read-side critical
+ * sections.
+ */
+void rcu_scheduler_starting(void)
+{
+       WARN_ON(num_online_cpus() != 1);
+       WARN_ON(nr_context_switches() > 0);
+       rcu_scheduler_active = 1;
+}
+
+static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
+static atomic_t rcu_barrier_cpu_count;
+static DEFINE_MUTEX(rcu_barrier_mutex);
+static struct completion rcu_barrier_completion;
+
+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 *type)
+{
+       int cpu = smp_processor_id();
+       struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
+       void (*call_rcu_func)(struct rcu_head *head,
+                             void (*func)(struct rcu_head *head));
+
+       atomic_inc(&rcu_barrier_cpu_count);
+       call_rcu_func = type;
+       call_rcu_func(head, rcu_barrier_callback);
+}
+
+/*
+ * Orchestrate the specified type of RCU barrier, waiting for all
+ * RCU callbacks of the specified type to complete.
+ */
+static void _rcu_barrier(struct rcu_state *rsp,
+                        void (*call_rcu_func)(struct rcu_head *head,
+                                              void (*func)(struct rcu_head *head)))
+{
+       BUG_ON(in_interrupt());
+       /* Take mutex to serialize concurrent rcu_barrier() requests. */
+       mutex_lock(&rcu_barrier_mutex);
+       init_completion(&rcu_barrier_completion);
+       /*
+        * Initialize rcu_barrier_cpu_count to 1, then invoke
+        * rcu_barrier_func() on each CPU, so that each CPU also has
+        * incremented rcu_barrier_cpu_count.  Only then is it safe to
+        * decrement rcu_barrier_cpu_count -- otherwise the first CPU
+        * might complete its grace period before all of the other CPUs
+        * did their increment, causing this function to return too
+        * early.
+        */
+       atomic_set(&rcu_barrier_cpu_count, 1);
+       preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */
+       rcu_adopt_orphan_cbs(rsp);
+       on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
+       preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */
+       if (atomic_dec_and_test(&rcu_barrier_cpu_count))
+               complete(&rcu_barrier_completion);
+       wait_for_completion(&rcu_barrier_completion);
+       mutex_unlock(&rcu_barrier_mutex);
+}
+
+/**
+ * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
+ */
+void rcu_barrier_bh(void)
+{
+       _rcu_barrier(&rcu_bh_state, call_rcu_bh);
+}
+EXPORT_SYMBOL_GPL(rcu_barrier_bh);
+
+/**
+ * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
+ */
+void rcu_barrier_sched(void)
+{
+       _rcu_barrier(&rcu_sched_state, call_rcu_sched);
+}
+EXPORT_SYMBOL_GPL(rcu_barrier_sched);
+
+/*
  * Do boot-time initialization of a CPU's per-CPU RCU data.
  */
 static void __init
@@ -1403,21 +1682,18 @@ static void __cpuinit
 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
 {
        unsigned long flags;
-       long lastcomp;
        unsigned long mask;
        struct rcu_data *rdp = rsp->rda[cpu];
        struct rcu_node *rnp = rcu_get_root(rsp);
 
        /* Set up local state, ensuring consistent view of global state. */
        spin_lock_irqsave(&rnp->lock, flags);
-       lastcomp = rsp->completed;
-       rdp->completed = lastcomp;
-       rdp->gpnum = lastcomp;
        rdp->passed_quiesc = 0;  /* We could be racing with new GP, */
        rdp->qs_pending = 1;     /*  so set up to respond to current GP. */
        rdp->beenonline = 1;     /* We have now been online. */
        rdp->preemptable = preemptable;
-       rdp->passed_quiesc_completed = lastcomp - 1;
+       rdp->qlen_last_fqs_check = 0;
+       rdp->n_force_qs_snap = rsp->n_force_qs;
        rdp->blimit = blimit;
        spin_unlock(&rnp->lock);                /* irqs remain disabled. */
 
@@ -1437,6 +1713,11 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
                spin_lock(&rnp->lock);  /* irqs already disabled. */
                rnp->qsmaskinit |= mask;
                mask = rnp->grpmask;
+               if (rnp == rdp->mynode) {
+                       rdp->gpnum = rnp->completed; /* if GP in progress... */
+                       rdp->completed = rnp->completed;
+                       rdp->passed_quiesc_completed = rnp->completed - 1;
+               }
                spin_unlock(&rnp->lock); /* irqs already disabled. */
                rnp = rnp->parent;
        } while (rnp != NULL && !(rnp->qsmaskinit & mask));
@@ -1454,8 +1735,8 @@ static void __cpuinit rcu_online_cpu(int cpu)
 /*
  * Handle CPU online/offline notification events.
  */
-int __cpuinit rcu_cpu_notify(struct notifier_block *self,
-                            unsigned long action, void *hcpu)
+static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
+                                   unsigned long action, void *hcpu)
 {
        long cpu = (long)hcpu;
 
@@ -1464,6 +1745,22 @@ int __cpuinit rcu_cpu_notify(struct notifier_block *self,
        case CPU_UP_PREPARE_FROZEN:
                rcu_online_cpu(cpu);
                break;
+       case CPU_DYING:
+       case CPU_DYING_FROZEN:
+               /*
+                * preempt_disable() in _rcu_barrier() prevents stop_machine(),
+                * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
+                * returns, all online cpus have queued rcu_barrier_func().
+                * The dying CPU clears its cpu_online_mask bit and
+                * moves all of its RCU callbacks to ->orphan_cbs_list
+                * in the context of stop_machine(), so subsequent calls
+                * to _rcu_barrier() will adopt these callbacks and only
+                * then queue rcu_barrier_func() on all remaining CPUs.
+                */
+               rcu_send_cbs_to_orphanage(&rcu_bh_state);
+               rcu_send_cbs_to_orphanage(&rcu_sched_state);
+               rcu_preempt_send_cbs_to_orphanage();
+               break;
        case CPU_DEAD:
        case CPU_DEAD_FROZEN:
        case CPU_UP_CANCELED:
@@ -1527,6 +1824,7 @@ static void __init rcu_init_one(struct rcu_state *rsp)
                rnp = rsp->level[i];
                for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
                        spin_lock_init(&rnp->lock);
+                       lockdep_set_class(&rnp->lock, &rcu_node_class[i]);
                        rnp->gpnum = 0;
                        rnp->qsmask = 0;
                        rnp->qsmaskinit = 0;
@@ -1547,6 +1845,8 @@ static void __init rcu_init_one(struct rcu_state *rsp)
                        rnp->level = i;
                        INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
                        INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
+                       INIT_LIST_HEAD(&rnp->blocked_tasks[2]);
+                       INIT_LIST_HEAD(&rnp->blocked_tasks[3]);
                }
        }
 }
@@ -1558,6 +1858,10 @@ static void __init rcu_init_one(struct rcu_state *rsp)
  */
 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
 do { \
+       int i; \
+       int j; \
+       struct rcu_node *rnp; \
+       \
        rcu_init_one(rsp); \
        rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
        j = 0; \
@@ -1570,41 +1874,30 @@ do { \
        } \
 } while (0)
 
-#ifdef CONFIG_TREE_PREEMPT_RCU
-
-void __init __rcu_init_preempt(void)
-{
-       int i;                  /* All used by RCU_INIT_FLAVOR(). */
-       int j;
-       struct rcu_node *rnp;
-
-       RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
-}
-
-#else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
-
-void __init __rcu_init_preempt(void)
+void __init rcu_init(void)
 {
-}
-
-#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
-
-void __init __rcu_init(void)
-{
-       int i;                  /* All used by RCU_INIT_FLAVOR(). */
-       int j;
-       struct rcu_node *rnp;
+       int i;
 
        rcu_bootup_announce();
 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
        printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
+#if NUM_RCU_LVL_4 != 0
+       printk(KERN_INFO "Experimental four-level hierarchy is enabled.\n");
+#endif /* #if NUM_RCU_LVL_4 != 0 */
        RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
        RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
        __rcu_init_preempt();
        open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
+
+       /*
+        * We don't need protection against CPU-hotplug here because
+        * this is called early in boot, before either interrupts
+        * or the scheduler are operational.
+        */
+       cpu_notifier(rcu_cpu_notify, 0);
+       for_each_online_cpu(i)
+               rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)i);
 }
 
-module_param(blimit, int, 0);
-module_param(qhimark, int, 0);
-module_param(qlowmark, int, 0);
+#include "rcutree_plugin.h"