rcu: Employ jiffies-based backstop to callback time limit

Currently, if there are more than 100 ready-to-invoke RCU callbacks queued
on a given CPU, the rcu_do_batch() function sets a timeout for invocation
of the series.  This timeout defaulting to three milliseconds, and may
be adjusted using the rcutree.rcu_resched_ns kernel boot parameter.
This timeout is checked using local_clock(), but the overhead of this
function combined with the common-case very small callback-invocation
overhead means that local_clock() is checked every 32nd invocation.

This works well except for longer-than average callbacks.  For example,
a series of 500-microsecond-duration callbacks means that local_clock()
is checked only once every 16 milliseconds, which makes it difficult to
enforce a three-millisecond timeout.

This commit therefore adds a Kconfig option RCU_DOUBLE_CHECK_CB_TIME
that enables backup timeout checking using the coarser grained but
lighter weight jiffies.  If the jiffies counter detects a timeout,
then local_clock() is consulted even if this is not the 32nd callback.
This prevents the aforementioned 16-millisecond latency blow.

Reported-by: Domas Mituzas <dmituzas@meta.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>

2 files changed, 38 insertions, 8 deletions

diff --git a/kernel/rcu/Kconfig b/kernel/rcu/Kconfig
index 9071182b1284..bdd7eadb33d8 100644
--- a/kernel/rcu/Kconfig
+++ b/kernel/rcu/Kconfig
@@ -314,4 +314,22 @@ config RCU_LAZY
 	  To save power, batch RCU callbacks and flush after delay, memory
 	  pressure, or callback list growing too big.
 
+config RCU_DOUBLE_CHECK_CB_TIME
+	bool "RCU callback-batch backup time check"
+	depends on RCU_EXPERT
+	default n
+	help
+	  Use this option to provide more precise enforcement of the
+	  rcutree.rcu_resched_ns module parameter in situations where
+	  a single RCU callback might run for hundreds of microseconds,
+	  thus defeating the 32-callback batching used to amortize the
+	  cost of the fine-grained but expensive local_clock() function.
+
+	  This option rounds rcutree.rcu_resched_ns up to the next
+	  jiffy, and overrides the 32-callback batching if this limit
+	  is exceeded.
+
+	  Say Y here if you need tighter callback-limit enforcement.
+	  Say N here if you are unsure.
+
 endmenu # "RCU Subsystem"
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 9a5c160186d1..e2dbea6cee4b 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -2047,10 +2047,15 @@ rcu_check_quiescent_state(struct rcu_data *rdp)
 }
 
 /* Return true if callback-invocation time limit exceeded. */
-static bool rcu_do_batch_check_time(long count, long tlimit)
+static bool rcu_do_batch_check_time(long count, long tlimit,
+				    bool jlimit_check, unsigned long jlimit)
 {
 	// Invoke local_clock() only once per 32 consecutive callbacks.
-	return unlikely(tlimit) && !likely(count & 31) && local_clock() >= tlimit;
+	return unlikely(tlimit) &&
+	       (!likely(count & 31) ||
+		(IS_ENABLED(CONFIG_RCU_DOUBLE_CHECK_CB_TIME) &&
+		 jlimit_check && time_after(jiffies, jlimit))) &&
+	       local_clock() >= tlimit;
 }
 
 /*
@@ -2059,13 +2064,17 @@ static bool rcu_do_batch_check_time(long count, long tlimit)
  */
 static void rcu_do_batch(struct rcu_data *rdp)
 {
+	long bl;
+	long count = 0;
 	int div;
 	bool __maybe_unused empty;
 	unsigned long flags;
-	struct rcu_head *rhp;
+	unsigned long jlimit;
+	bool jlimit_check = false;
+	long pending;
 	struct rcu_cblist rcl = RCU_CBLIST_INITIALIZER(rcl);
-	long bl, count = 0;
-	long pending, tlimit = 0;
+	struct rcu_head *rhp;
+	long tlimit = 0;
 
 	/* If no callbacks are ready, just return. */
 	if (!rcu_segcblist_ready_cbs(&rdp->cblist)) {
@@ -2090,11 +2099,14 @@ static void rcu_do_batch(struct rcu_data *rdp)
 	div = div < 0 ? 7 : div > sizeof(long) * 8 - 2 ? sizeof(long) * 8 - 2 : div;
 	bl = max(rdp->blimit, pending >> div);
 	if ((in_serving_softirq() || rdp->rcu_cpu_kthread_status == RCU_KTHREAD_RUNNING) &&
-	    unlikely(bl > 100)) {
+	    (IS_ENABLED(CONFIG_RCU_DOUBLE_CHECK_CB_TIME) || unlikely(bl > 100))) {
+		const long npj = NSEC_PER_SEC / HZ;
 		long rrn = READ_ONCE(rcu_resched_ns);
 
 		rrn = rrn < NSEC_PER_MSEC ? NSEC_PER_MSEC : rrn > NSEC_PER_SEC ? NSEC_PER_SEC : rrn;
 		tlimit = local_clock() + rrn;
+		jlimit = jiffies + (rrn + npj + 1) / npj;
+		jlimit_check = true;
 	}
 	trace_rcu_batch_start(rcu_state.name,
 			      rcu_segcblist_n_cbs(&rdp->cblist), bl);
@@ -2134,7 +2146,7 @@ static void rcu_do_batch(struct rcu_data *rdp)
 			 * Make sure we don't spend too much time here and deprive other
 			 * softirq vectors of CPU cycles.
 			 */
-			if (rcu_do_batch_check_time(count, tlimit))
+			if (rcu_do_batch_check_time(count, tlimit, jlimit_check, jlimit))
 				break;
 		} else {
 			// In rcuc/rcuoc context, so no worries about
@@ -2147,7 +2159,7 @@ static void rcu_do_batch(struct rcu_data *rdp)
 			// But rcuc kthreads can delay quiescent-state
 			// reporting, so check time limits for them.
 			if (rdp->rcu_cpu_kthread_status == RCU_KTHREAD_RUNNING &&
-			    rcu_do_batch_check_time(count, tlimit)) {
+			    rcu_do_batch_check_time(count, tlimit, jlimit_check, jlimit)) {
 				rdp->rcu_cpu_has_work = 1;
 				break;
 			}