[04/11] rcu/nocb: Introduce nocb mutex

Commit Message

Neeraj Upadhyay Aug. 16, 2024, 4:56 a.m. UTC

From: Frederic Weisbecker <[e-mail]>

The barrier_mutex is used currently to protect (de-)offloading
operations and prevent from nocb_lock locking imbalance in rcu_barrier()
and shrinker, and also from misordered RCU barrier invocation.

Now since RCU (de-)offloading is going to happen on offline CPUs, an RCU
barrier will have to be executed while transitionning from offloaded to
de-offloaded state. And this can't happen while holding the
barrier_mutex.

Introduce a NOCB mutex to protect (de-)offloading transitions. The
barrier_mutex is still held for now when necessary to avoid barrier
callbacks reordering and nocb_lock imbalance.

Signed-off-by: Frederic Weisbecker <[e-mail]>
Signed-off-by: Paul E. McKenney <[e-mail]>
Reviewed-by: Paul E. McKenney <[e-mail]>
Signed-off-by: Neeraj Upadhyay <[e-mail]>
---
 kernel/rcu/tree.c        |  3 +++
 kernel/rcu/tree.h        |  1 +
 kernel/rcu/tree_nocb.h   | 20 ++++++++++++--------
 kernel/rcu/tree_plugin.h |  1 +
 4 files changed, 17 insertions(+), 8 deletions(-)

Patch

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index e641cc681901..2b9e713854b0 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -97,6 +97,9 @@  static struct rcu_state rcu_state = {
 	.srs_cleanup_work = __WORK_INITIALIZER(rcu_state.srs_cleanup_work,
 		rcu_sr_normal_gp_cleanup_work),
 	.srs_cleanups_pending = ATOMIC_INIT(0),
+#ifdef CONFIG_RCU_NOCB_CPU
+	.nocb_mutex = __MUTEX_INITIALIZER(rcu_state.nocb_mutex),
+#endif
 };
 
 /* Dump rcu_node combining tree at boot to verify correct setup. */
diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
index a297dc89a09c..16e6fe63d93c 100644
--- a/kernel/rcu/tree.h
+++ b/kernel/rcu/tree.h
@@ -421,6 +421,7 @@  struct rcu_state {
 	atomic_t srs_cleanups_pending; /* srs inflight worker cleanups. */
 
 #ifdef CONFIG_RCU_NOCB_CPU
+	struct mutex nocb_mutex;		/* Guards (de-)offloading */
 	int nocb_is_setup;			/* nocb is setup from boot */
 #endif
 };
diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
index fdd0616f2fd1..16bcb8b13a5e 100644
--- a/kernel/rcu/tree_nocb.h
+++ b/kernel/rcu/tree_nocb.h
@@ -1141,6 +1141,7 @@  int rcu_nocb_cpu_deoffload(int cpu)
 	int ret = 0;
 
 	cpus_read_lock();
+	mutex_lock(&rcu_state.nocb_mutex);
 	mutex_lock(&rcu_state.barrier_mutex);
 	if (rcu_rdp_is_offloaded(rdp)) {
 		if (cpu_online(cpu)) {
@@ -1153,6 +1154,7 @@  int rcu_nocb_cpu_deoffload(int cpu)
 		}
 	}
 	mutex_unlock(&rcu_state.barrier_mutex);
+	mutex_unlock(&rcu_state.nocb_mutex);
 	cpus_read_unlock();
 
 	return ret;
@@ -1228,6 +1230,7 @@  int rcu_nocb_cpu_offload(int cpu)
 	int ret = 0;
 
 	cpus_read_lock();
+	mutex_lock(&rcu_state.nocb_mutex);
 	mutex_lock(&rcu_state.barrier_mutex);
 	if (!rcu_rdp_is_offloaded(rdp)) {
 		if (cpu_online(cpu)) {
@@ -1240,6 +1243,7 @@  int rcu_nocb_cpu_offload(int cpu)
 		}
 	}
 	mutex_unlock(&rcu_state.barrier_mutex);
+	mutex_unlock(&rcu_state.nocb_mutex);
 	cpus_read_unlock();
 
 	return ret;
@@ -1257,7 +1261,7 @@  lazy_rcu_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
 		return 0;
 
 	/*  Protect rcu_nocb_mask against concurrent (de-)offloading. */
-	if (!mutex_trylock(&rcu_state.barrier_mutex))
+	if (!mutex_trylock(&rcu_state.nocb_mutex))
 		return 0;
 
 	/* Snapshot count of all CPUs */
@@ -1267,7 +1271,7 @@  lazy_rcu_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
 		count +=  READ_ONCE(rdp->lazy_len);
 	}
 
-	mutex_unlock(&rcu_state.barrier_mutex);
+	mutex_unlock(&rcu_state.nocb_mutex);
 
 	return count ? count : SHRINK_EMPTY;
 }
@@ -1285,9 +1289,9 @@  lazy_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 	 * Protect against concurrent (de-)offloading. Otherwise nocb locking
 	 * may be ignored or imbalanced.
 	 */
-	if (!mutex_trylock(&rcu_state.barrier_mutex)) {
+	if (!mutex_trylock(&rcu_state.nocb_mutex)) {
 		/*
-		 * But really don't insist if barrier_mutex is contended since we
+		 * But really don't insist if nocb_mutex is contended since we
 		 * can't guarantee that it will never engage in a dependency
 		 * chain involving memory allocation. The lock is seldom contended
 		 * anyway.
@@ -1326,7 +1330,7 @@  lazy_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 			break;
 	}
 
-	mutex_unlock(&rcu_state.barrier_mutex);
+	mutex_unlock(&rcu_state.nocb_mutex);
 
 	return count ? count : SHRINK_STOP;
 }
@@ -1473,15 +1477,15 @@  static void rcu_spawn_cpu_nocb_kthread(int cpu)
 	 * No need to protect against concurrent rcu_barrier()
 	 * because the number of callbacks should be 0 for a non-boot CPU,
 	 * therefore rcu_barrier() shouldn't even try to grab the nocb_lock.
-	 * But hold barrier_mutex to avoid nocb_lock imbalance from shrinker.
+	 * But hold nocb_mutex to avoid nocb_lock imbalance from shrinker.
 	 */
 	WARN_ON_ONCE(system_state > SYSTEM_BOOTING && rcu_segcblist_n_cbs(&rdp->cblist));
-	mutex_lock(&rcu_state.barrier_mutex);
+	mutex_lock(&rcu_state.nocb_mutex);
 	if (rcu_rdp_is_offloaded(rdp)) {
 		rcu_nocb_rdp_deoffload(rdp);
 		cpumask_clear_cpu(cpu, rcu_nocb_mask);
 	}
-	mutex_unlock(&rcu_state.barrier_mutex);
+	mutex_unlock(&rcu_state.nocb_mutex);
 }
 
 /* How many CB CPU IDs per GP kthread?  Default of -1 for sqrt(nr_cpu_ids). */
diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index f752b2a1d887..c662376c8af0 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -28,6 +28,7 @@  static bool rcu_rdp_is_offloaded(struct rcu_data *rdp)
 		!(lockdep_is_held(&rcu_state.barrier_mutex) ||
 		  (IS_ENABLED(CONFIG_HOTPLUG_CPU) && lockdep_is_cpus_held()) ||
 		  lockdep_is_held(&rdp->nocb_lock) ||
+		  lockdep_is_held(&rcu_state.nocb_mutex) ||
 		  (!(IS_ENABLED(CONFIG_PREEMPT_COUNT) && preemptible()) &&
 		   rdp == this_cpu_ptr(&rcu_data)) ||
 		  rcu_current_is_nocb_kthread(rdp)),