rcu: Apply smp_mb__after_unlock_lock() to preserve grace periods

RCU must ensure that there is the equivalent of a full memory
barrier between any memory access preceding grace period and any
memory access following that same grace period, regardless of
which CPU(s) happen to execute the two memory accesses.
Therefore, downgrading UNLOCK+LOCK to no longer imply a full
memory barrier requires some adjustments to RCU.

This commit therefore adds smp_mb__after_unlock_lock()
invocations as needed after the RCU lock acquisitions that need
to be part of a full-memory-barrier UNLOCK+LOCK.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <linux-arch@vger.kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/1386799151-2219-7-git-send-email-paulmck@linux.vnet.ibm.comSigned-off-by: Ingo Molnar <mingo@kernel.org>

kernel/rcu/tree.c

@@ -1133,8 +1133,10 @@ rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp)
 	 * hold it, acquire the root rcu_node structure's lock in order to
 	 * start one (if needed).
 	 */
-	if (rnp != rnp_root)
+	if (rnp != rnp_root) {
 		raw_spin_lock(&rnp_root->lock);
+		smp_mb__after_unlock_lock();
+	}
 
 	/*
 	 * Get a new grace-period number.  If there really is no grace
@@ -1354,6 +1356,7 @@ static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp)
 		local_irq_restore(flags);
 		return;
 	}
+	smp_mb__after_unlock_lock();
 	__note_gp_changes(rsp, rnp, rdp);
 	raw_spin_unlock_irqrestore(&rnp->lock, flags);
 }
@@ -1368,6 +1371,7 @@ static int rcu_gp_init(struct rcu_state *rsp)
 
 	rcu_bind_gp_kthread();
 	raw_spin_lock_irq(&rnp->lock);
+	smp_mb__after_unlock_lock();
 	if (rsp->gp_flags == 0) {
 		/* Spurious wakeup, tell caller to go back to sleep.  */
 		raw_spin_unlock_irq(&rnp->lock);
@@ -1409,6 +1413,7 @@ static int rcu_gp_init(struct rcu_state *rsp)
 	 */
 	rcu_for_each_node_breadth_first(rsp, rnp) {
 		raw_spin_lock_irq(&rnp->lock);
+		smp_mb__after_unlock_lock();
 		rdp = this_cpu_ptr(rsp->rda);
 		rcu_preempt_check_blocked_tasks(rnp);
 		rnp->qsmask = rnp->qsmaskinit;
@@ -1463,6 +1468,7 @@ static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
 	/* Clear flag to prevent immediate re-entry. */
 	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
 		raw_spin_lock_irq(&rnp->lock);
+		smp_mb__after_unlock_lock();
 		rsp->gp_flags &= ~RCU_GP_FLAG_FQS;
 		raw_spin_unlock_irq(&rnp->lock);
 	}
@@ -1480,6 +1486,7 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
 	struct rcu_node *rnp = rcu_get_root(rsp);
 
 	raw_spin_lock_irq(&rnp->lock);
+	smp_mb__after_unlock_lock();
 	gp_duration = jiffies - rsp->gp_start;
 	if (gp_duration > rsp->gp_max)
 		rsp->gp_max = gp_duration;
@@ -1505,6 +1512,7 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
 	 */
 	rcu_for_each_node_breadth_first(rsp, rnp) {
 		raw_spin_lock_irq(&rnp->lock);
+		smp_mb__after_unlock_lock();
 		ACCESS_ONCE(rnp->completed) = rsp->gpnum;
 		rdp = this_cpu_ptr(rsp->rda);
 		if (rnp == rdp->mynode)
@@ -1515,6 +1523,7 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
 	}
 	rnp = rcu_get_root(rsp);
 	raw_spin_lock_irq(&rnp->lock);
+	smp_mb__after_unlock_lock();
 	rcu_nocb_gp_set(rnp, nocb);
 
 	rsp->completed = rsp->gpnum; /* Declare grace period done. */
@@ -1749,6 +1758,7 @@ rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
 		rnp_c = rnp;
 		rnp = rnp->parent;
 		raw_spin_lock_irqsave(&rnp->lock, flags);
+		smp_mb__after_unlock_lock();
 		WARN_ON_ONCE(rnp_c->qsmask);
 	}
 
@@ -1778,6 +1788,7 @@ rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
 
 	rnp = rdp->mynode;
 	raw_spin_lock_irqsave(&rnp->lock, flags);
+	smp_mb__after_unlock_lock();
 	if (rdp->passed_quiesce == 0 || rdp->gpnum != rnp->gpnum ||
 	    rnp->completed == rnp->gpnum) {
 
@@ -1992,6 +2003,7 @@ static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
 	mask = rdp->grpmask;	/* rnp->grplo is constant. */
 	do {
 		raw_spin_lock(&rnp->lock);	/* irqs already disabled. */
+		smp_mb__after_unlock_lock();
 		rnp->qsmaskinit &= ~mask;
 		if (rnp->qsmaskinit != 0) {
 			if (rnp != rdp->mynode)
@@ -2202,6 +2214,7 @@ static void force_qs_rnp(struct rcu_state *rsp,
 		cond_resched();
 		mask = 0;
 		raw_spin_lock_irqsave(&rnp->lock, flags);
+		smp_mb__after_unlock_lock();
 		if (!rcu_gp_in_progress(rsp)) {
 			raw_spin_unlock_irqrestore(&rnp->lock, flags);
 			return;
@@ -2231,6 +2244,7 @@ static void force_qs_rnp(struct rcu_state *rsp,
 	rnp = rcu_get_root(rsp);
 	if (rnp->qsmask == 0) {
 		raw_spin_lock_irqsave(&rnp->lock, flags);
+		smp_mb__after_unlock_lock();
 		rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
 	}
 }
@@ -2263,6 +2277,7 @@ static void force_quiescent_state(struct rcu_state *rsp)
 
 	/* Reached the root of the rcu_node tree, acquire lock. */
 	raw_spin_lock_irqsave(&rnp_old->lock, flags);
+	smp_mb__after_unlock_lock();
 	raw_spin_unlock(&rnp_old->fqslock);
 	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
 		rsp->n_force_qs_lh++;
@@ -2378,6 +2393,7 @@ static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp,
 			struct rcu_node *rnp_root = rcu_get_root(rsp);
 
 			raw_spin_lock(&rnp_root->lock);
+			smp_mb__after_unlock_lock();
 			rcu_start_gp(rsp);
 			raw_spin_unlock(&rnp_root->lock);
 		} else {

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);
 }
 
@@ -381,6 +383,7 @@ 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 +608,7 @@ 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 +633,7 @@ 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 +777,7 @@ 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);
@@ -787,6 +793,7 @@ static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
 		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 +813,7 @@ 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 +894,7 @@ 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 +1200,7 @@ 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 +1387,7 @@ 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 +1780,7 @@ 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. */
 	}
@@ -2209,6 +2221,7 @@ 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);