srcu: Specify auto-expedite holdoff time

On small systems, in the absence of readers, expedited SRCU grace
periods can complete in less than a microsecond.  This means that an
eight-CPU system can have all CPUs doing synchronize_srcu() in a tight
loop and almost always expedite.  This might actually be desirable in
some situations, but in general it is a good way to needlessly burn
CPU cycles.  And in those situations where it is desirable, your friend
is the function synchronize_srcu_expedited().

For other situations, this commit adds a kernel parameter that specifies
a holdoff between completing the last SRCU grace period and auto-expediting
the next.  If the next grace period starts before the holdoff expires,
auto-expediting is disabled.  The holdoff is 50 microseconds by default,
and can be tuned to the desired number of nanoseconds.  A value of zero
disables auto-expediting.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Mike Galbraith <efault@gmx.de>

3 files changed, 26 insertions, 1 deletions

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index facc20a3f962..4a4b9266c4de 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -3779,6 +3779,14 @@
 	spia_pedr=
 	spia_peddr=
 
+	srcutree.exp_holdoff [KNL]
+			Specifies how many nanoseconds must elapse
+			since the end of the last SRCU grace period for
+			a given srcu_struct until the next normal SRCU
+			grace period will be considered for automatic
+			expediting.  Set to zero to disable automatic
+			expediting.
+
 	stacktrace	[FTRACE]
 			Enabled the stack tracer on boot up.
 
diff --git a/include/linux/srcutree.h b/include/linux/srcutree.h
index 86df48d3e97b..32e86d85fd11 100644
--- a/include/linux/srcutree.h
+++ b/include/linux/srcutree.h
@@ -84,6 +84,7 @@ struct srcu_struct {
 	unsigned long srcu_gp_seq;		/* Grace-period seq #. */
 	unsigned long srcu_gp_seq_needed;	/* Latest gp_seq needed. */
 	unsigned long srcu_gp_seq_needed_exp;	/* Furthest future exp GP. */
+	unsigned long srcu_last_gp_end;		/* Last GP end timestamp (ns) */
 	struct srcu_data __percpu *sda;		/* Per-CPU srcu_data array. */
 	unsigned long srcu_barrier_seq;		/* srcu_barrier seq #. */
 	struct mutex srcu_barrier_mutex;	/* Serialize barrier ops. */
diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
index 2286e06fd159..74c283f9d15e 100644
--- a/kernel/rcu/srcutree.c
+++ b/kernel/rcu/srcutree.c
@@ -34,10 +34,14 @@
 #include <linux/sched.h>
 #include <linux/smp.h>
 #include <linux/delay.h>
+#include <linux/module.h>
 #include <linux/srcu.h>
 
 #include "rcu.h"
 
+ulong exp_holdoff = 50 * 1000; /* Holdoff (ns) for auto-expediting. */
+module_param(exp_holdoff, ulong, 0444);
+
 static void srcu_invoke_callbacks(struct work_struct *work);
 static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay);
 
@@ -145,6 +149,7 @@ static int init_srcu_struct_fields(struct srcu_struct *sp, bool is_static)
 		sp->sda = alloc_percpu(struct srcu_data);
 	init_srcu_struct_nodes(sp, is_static);
 	sp->srcu_gp_seq_needed_exp = 0;
+	sp->srcu_last_gp_end = ktime_get_mono_fast_ns();
 	smp_store_release(&sp->srcu_gp_seq_needed, 0); /* Init done. */
 	return sp->sda ? 0 : -ENOMEM;
 }
@@ -498,6 +503,7 @@ static void srcu_gp_end(struct srcu_struct *sp)
 	idx = rcu_seq_state(sp->srcu_gp_seq);
 	WARN_ON_ONCE(idx != SRCU_STATE_SCAN2);
 	cbdelay = srcu_get_delay(sp);
+	sp->srcu_last_gp_end = ktime_get_mono_fast_ns();
 	rcu_seq_end(&sp->srcu_gp_seq);
 	gpseq = rcu_seq_current(&sp->srcu_gp_seq);
 	if (ULONG_CMP_LT(sp->srcu_gp_seq_needed_exp, gpseq))
@@ -700,9 +706,10 @@ static void srcu_flip(struct srcu_struct *sp)
  */
 static bool srcu_might_be_idle(struct srcu_struct *sp)
 {
+	unsigned long curseq;
 	unsigned long flags;
 	struct srcu_data *sdp;
-	unsigned long curseq;
+	unsigned long t;
 
 	/* If the local srcu_data structure has callbacks, not idle.  */
 	local_irq_save(flags);
@@ -718,6 +725,15 @@ static bool srcu_might_be_idle(struct srcu_struct *sp)
 	 * Exact information would require acquiring locks, which would
 	 * kill scalability, hence the probabalistic nature of the probe.
 	 */
+
+	/* First, see if enough time has passed since the last GP. */
+	t = ktime_get_mono_fast_ns();
+	if (exp_holdoff == 0 ||
+	    time_in_range_open(t, sp->srcu_last_gp_end,
+			       sp->srcu_last_gp_end + exp_holdoff))
+		return false; /* Too soon after last GP. */
+
+	/* Next, check for probable idleness. */
 	curseq = rcu_seq_current(&sp->srcu_gp_seq);
 	smp_mb(); /* Order ->srcu_gp_seq with ->srcu_gp_seq_needed. */
 	if (ULONG_CMP_LT(curseq, READ_ONCE(sp->srcu_gp_seq_needed)))