1 files changed, 89 insertions, 13 deletions

diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index 08a765232432..6e2ef4b2b920 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -204,6 +204,7 @@ static void rcu_preempt_note_context_switch(int cpu)
 		rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu);
 		rnp = rdp->mynode;
 		raw_spin_lock_irqsave(&rnp->lock, flags);
+		smp_mb__after_unlock_lock();
 		t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
 		t->rcu_blocked_node = rnp;
 
@@ -312,6 +313,7 @@ static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags)
 	mask = rnp->grpmask;
 	raw_spin_unlock(&rnp->lock);	/* irqs remain disabled. */
 	raw_spin_lock(&rnp_p->lock);	/* irqs already disabled. */
+	smp_mb__after_unlock_lock();
 	rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags);
 }
 
@@ -361,10 +363,14 @@ void rcu_read_unlock_special(struct task_struct *t)
 	special = t->rcu_read_unlock_special;
 	if (special & RCU_READ_UNLOCK_NEED_QS) {
 		rcu_preempt_qs(smp_processor_id());
+		if (!t->rcu_read_unlock_special) {
+			local_irq_restore(flags);
+			return;
+		}
 	}
 
-	/* Hardware IRQ handlers cannot block. */
-	if (in_irq() || in_serving_softirq()) {
+	/* Hardware IRQ handlers cannot block, complain if they get here. */
+	if (WARN_ON_ONCE(in_irq() || in_serving_softirq())) {
 		local_irq_restore(flags);
 		return;
 	}
@@ -381,6 +387,7 @@ void rcu_read_unlock_special(struct task_struct *t)
 		for (;;) {
 			rnp = t->rcu_blocked_node;
 			raw_spin_lock(&rnp->lock);  /* irqs already disabled. */
+			smp_mb__after_unlock_lock();
 			if (rnp == t->rcu_blocked_node)
 				break;
 			raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
@@ -605,6 +612,7 @@ static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
 	while (!list_empty(lp)) {
 		t = list_entry(lp->next, typeof(*t), rcu_node_entry);
 		raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
+		smp_mb__after_unlock_lock();
 		list_del(&t->rcu_node_entry);
 		t->rcu_blocked_node = rnp_root;
 		list_add(&t->rcu_node_entry, lp_root);
@@ -629,6 +637,7 @@ static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
 	 * in this case.
 	 */
 	raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
+	smp_mb__after_unlock_lock();
 	if (rnp_root->boost_tasks != NULL &&
 	    rnp_root->boost_tasks != rnp_root->gp_tasks &&
 	    rnp_root->boost_tasks != rnp_root->exp_tasks)
@@ -772,6 +781,7 @@ static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
 	unsigned long mask;
 
 	raw_spin_lock_irqsave(&rnp->lock, flags);
+	smp_mb__after_unlock_lock();
 	for (;;) {
 		if (!sync_rcu_preempt_exp_done(rnp)) {
 			raw_spin_unlock_irqrestore(&rnp->lock, flags);
@@ -779,14 +789,17 @@ static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
 		}
 		if (rnp->parent == NULL) {
 			raw_spin_unlock_irqrestore(&rnp->lock, flags);
-			if (wake)
+			if (wake) {
+				smp_mb(); /* EGP done before wake_up(). */
 				wake_up(&sync_rcu_preempt_exp_wq);
+			}
 			break;
 		}
 		mask = rnp->grpmask;
 		raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
 		rnp = rnp->parent;
 		raw_spin_lock(&rnp->lock); /* irqs already disabled */
+		smp_mb__after_unlock_lock();
 		rnp->expmask &= ~mask;
 	}
 }
@@ -806,6 +819,7 @@ sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp)
 	int must_wait = 0;
 
 	raw_spin_lock_irqsave(&rnp->lock, flags);
+	smp_mb__after_unlock_lock();
 	if (list_empty(&rnp->blkd_tasks)) {
 		raw_spin_unlock_irqrestore(&rnp->lock, flags);
 	} else {
@@ -886,6 +900,7 @@ void synchronize_rcu_expedited(void)
 	/* Initialize ->expmask for all non-leaf rcu_node structures. */
 	rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) {
 		raw_spin_lock_irqsave(&rnp->lock, flags);
+		smp_mb__after_unlock_lock();
 		rnp->expmask = rnp->qsmaskinit;
 		raw_spin_unlock_irqrestore(&rnp->lock, flags);
 	}
@@ -1191,6 +1206,7 @@ static int rcu_boost(struct rcu_node *rnp)
 		return 0;  /* Nothing left to boost. */
 
 	raw_spin_lock_irqsave(&rnp->lock, flags);
+	smp_mb__after_unlock_lock();
 
 	/*
 	 * Recheck under the lock: all tasks in need of boosting
@@ -1377,6 +1393,7 @@ static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
 	if (IS_ERR(t))
 		return PTR_ERR(t);
 	raw_spin_lock_irqsave(&rnp->lock, flags);
+	smp_mb__after_unlock_lock();
 	rnp->boost_kthread_task = t;
 	raw_spin_unlock_irqrestore(&rnp->lock, flags);
 	sp.sched_priority = RCU_BOOST_PRIO;
@@ -1769,6 +1786,7 @@ static void rcu_prepare_for_idle(int cpu)
 			continue;
 		rnp = rdp->mynode;
 		raw_spin_lock(&rnp->lock); /* irqs already disabled. */
+		smp_mb__after_unlock_lock();
 		rcu_accelerate_cbs(rsp, rnp, rdp);
 		raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
 	}
@@ -1852,6 +1870,7 @@ static int rcu_oom_notify(struct notifier_block *self,
 
 	/* Wait for callbacks from earlier instance to complete. */
 	wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0);
+	smp_mb(); /* Ensure callback reuse happens after callback invocation. */
 
 	/*
 	 * Prevent premature wakeup: ensure that all increments happen
@@ -2101,7 +2120,8 @@ bool rcu_is_nocb_cpu(int cpu)
 static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
 				    struct rcu_head *rhp,
 				    struct rcu_head **rhtp,
-				    int rhcount, int rhcount_lazy)
+				    int rhcount, int rhcount_lazy,
+				    unsigned long flags)
 {
 	int len;
 	struct rcu_head **old_rhpp;
@@ -2122,9 +2142,16 @@ static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
 	}
 	len = atomic_long_read(&rdp->nocb_q_count);
 	if (old_rhpp == &rdp->nocb_head) {
-		wake_up(&rdp->nocb_wq); /* ... only if queue was empty ... */
+		if (!irqs_disabled_flags(flags)) {
+			wake_up(&rdp->nocb_wq); /* ... if queue was empty ... */
+			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
+					    TPS("WakeEmpty"));
+		} else {
+			rdp->nocb_defer_wakeup = true;
+			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
+					    TPS("WakeEmptyIsDeferred"));
+		}
 		rdp->qlen_last_fqs_check = 0;
-		trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WakeEmpty"));
 	} else if (len > rdp->qlen_last_fqs_check + qhimark) {
 		wake_up_process(t); /* ... or if many callbacks queued. */
 		rdp->qlen_last_fqs_check = LONG_MAX / 2;
@@ -2145,12 +2172,12 @@ static void __call_rcu_nocb_enqueue(struct rcu_data *rdp,
  * "rcuo" kthread can find it.
  */
 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
-			    bool lazy)
+			    bool lazy, unsigned long flags)
 {
 
 	if (!rcu_is_nocb_cpu(rdp->cpu))
 		return 0;
-	__call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy);
+	__call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags);
 	if (__is_kfree_rcu_offset((unsigned long)rhp->func))
 		trace_rcu_kfree_callback(rdp->rsp->name, rhp,
 					 (unsigned long)rhp->func,
@@ -2168,7 +2195,8 @@ static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
  * not a no-CBs CPU.
  */
 static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
-						     struct rcu_data *rdp)
+						     struct rcu_data *rdp,
+						     unsigned long flags)
 {
 	long ql = rsp->qlen;
 	long qll = rsp->qlen_lazy;
@@ -2182,14 +2210,14 @@ static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
 	/* First, enqueue the donelist, if any.  This preserves CB ordering. */
 	if (rsp->orphan_donelist != NULL) {
 		__call_rcu_nocb_enqueue(rdp, rsp->orphan_donelist,
-					rsp->orphan_donetail, ql, qll);
+					rsp->orphan_donetail, ql, qll, flags);
 		ql = qll = 0;
 		rsp->orphan_donelist = NULL;
 		rsp->orphan_donetail = &rsp->orphan_donelist;
 	}
 	if (rsp->orphan_nxtlist != NULL) {
 		__call_rcu_nocb_enqueue(rdp, rsp->orphan_nxtlist,
-					rsp->orphan_nxttail, ql, qll);
+					rsp->orphan_nxttail, ql, qll, flags);
 		ql = qll = 0;
 		rsp->orphan_nxtlist = NULL;
 		rsp->orphan_nxttail = &rsp->orphan_nxtlist;
@@ -2209,6 +2237,7 @@ static void rcu_nocb_wait_gp(struct rcu_data *rdp)
 	struct rcu_node *rnp = rdp->mynode;
 
 	raw_spin_lock_irqsave(&rnp->lock, flags);
+	smp_mb__after_unlock_lock();
 	c = rcu_start_future_gp(rnp, rdp);
 	raw_spin_unlock_irqrestore(&rnp->lock, flags);
 
@@ -2250,6 +2279,7 @@ static int rcu_nocb_kthread(void *arg)
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
 					    TPS("Sleep"));
 			wait_event_interruptible(rdp->nocb_wq, rdp->nocb_head);
+			/* Memory barrier provide by xchg() below. */
 		} else if (firsttime) {
 			firsttime = 0;
 			trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu,
@@ -2310,6 +2340,22 @@ static int rcu_nocb_kthread(void *arg)
 	return 0;
 }
 
+/* Is a deferred wakeup of rcu_nocb_kthread() required? */
+static bool rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp)
+{
+	return ACCESS_ONCE(rdp->nocb_defer_wakeup);
+}
+
+/* Do a deferred wakeup of rcu_nocb_kthread(). */
+static void do_nocb_deferred_wakeup(struct rcu_data *rdp)
+{
+	if (!rcu_nocb_need_deferred_wakeup(rdp))
+		return;
+	ACCESS_ONCE(rdp->nocb_defer_wakeup) = false;
+	wake_up(&rdp->nocb_wq);
+	trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("DeferredWakeEmpty"));
+}
+
 /* Initialize per-rcu_data variables for no-CBs CPUs. */
 static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp)
 {
@@ -2365,13 +2411,14 @@ static void rcu_init_one_nocb(struct rcu_node *rnp)
 }
 
 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
-			    bool lazy)
+			    bool lazy, unsigned long flags)
 {
 	return 0;
 }
 
 static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
-						     struct rcu_data *rdp)
+						     struct rcu_data *rdp,
+						     unsigned long flags)
 {
 	return 0;
 }
@@ -2380,6 +2427,15 @@ static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp)
 {
 }
 
+static bool rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp)
+{
+	return false;
+}
+
+static void do_nocb_deferred_wakeup(struct rcu_data *rdp)
+{
+}
+
 static void __init rcu_spawn_nocb_kthreads(struct rcu_state *rsp)
 {
 }
@@ -2829,3 +2885,23 @@ static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp)
 }
 
 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
+
+/*
+ * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the
+ * grace-period kthread will do force_quiescent_state() processing?
+ * The idea is to avoid waking up RCU core processing on such a
+ * CPU unless the grace period has extended for too long.
+ *
+ * This code relies on the fact that all NO_HZ_FULL CPUs are also
+ * CONFIG_RCU_NOCB_CPUs.
+ */
+static bool rcu_nohz_full_cpu(struct rcu_state *rsp)
+{
+#ifdef CONFIG_NO_HZ_FULL
+	if (tick_nohz_full_cpu(smp_processor_id()) &&
+	    (!rcu_gp_in_progress(rsp) ||
+	     ULONG_CMP_LT(jiffies, ACCESS_ONCE(rsp->gp_start) + HZ)))
+		return 1;
+#endif /* #ifdef CONFIG_NO_HZ_FULL */
+	return 0;
+}