From: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
To: linux-kernel@vger.kernel.org
Cc: mingo@kernel.org, jiangshanlai@gmail.com, dipankar@in.ibm.com,
        akpm@linux-foundation.org, mathieu.desnoyers@efficios.com,
        josh@joshtriplett.org, tglx@linutronix.de, peterz@infradead.org,
        rostedt@goodmis.org, dhowells@redhat.com, edumazet@google.com,
        fweisbec@gmail.com, oleg@redhat.com, bobby.prani@gmail.com,
        "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Subject: [PATCH tip/core/rcu 36/40] srcu: Crude control of expedited grace periods
Date: Wed, 12 Apr 2017 10:40:21 -0700
In-Reply-To: <20170412174003.GA23207@linux.vnet.ibm.com>
References: <20170412174003.GA23207@linux.vnet.ibm.com>
Message-Id: <1492018825-25634-36-git-send-email-paulmck@linux.vnet.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 7806
Lines: 215

SRCU's implementation of expedited grace periods has always assumed
that the SRCU instance is idle when the expedited request arrives.
This commit improves this a bit by maintaining a count of the number
of outstanding expedited requests, thus allowing prior non-expedited
grace periods accommodate these requests by shifting to expedited mode.
However, any non-expedited wait already in progress will still wait for
the full duration.

Improved control of expedited grace periods is planned, but one step
at a time.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 include/linux/srcu.h |  1 +
 kernel/rcu/srcu.c    | 84 ++++++++++++++++++++++++++++------------------------
 2 files changed, 47 insertions(+), 38 deletions(-)

diff --git a/include/linux/srcu.h b/include/linux/srcu.h
index e7dbc01b61a1..73a1b6296224 100644
--- a/include/linux/srcu.h
+++ b/include/linux/srcu.h
@@ -42,6 +42,7 @@ struct srcu_array {
 struct srcu_struct {
 	unsigned long completed;
 	unsigned long srcu_gp_seq;
+	atomic_t srcu_exp_cnt;
 	struct srcu_array __percpu *per_cpu_ref;
 	spinlock_t queue_lock; /* protect ->srcu_cblist */
 	struct rcu_segcblist srcu_cblist;
diff --git a/kernel/rcu/srcu.c b/kernel/rcu/srcu.c
index 97aec5d7b316..b62919be99e7 100644
--- a/kernel/rcu/srcu.c
+++ b/kernel/rcu/srcu.c
@@ -43,6 +43,7 @@ static int init_srcu_struct_fields(struct srcu_struct *sp)
 {
 	sp->completed = 0;
 	sp->srcu_gp_seq = 0;
+	atomic_set(&sp->srcu_exp_cnt, 0);
 	spin_lock_init(&sp->queue_lock);
 	rcu_segcblist_init(&sp->srcu_cblist);
 	INIT_DELAYED_WORK(&sp->work, process_srcu);
@@ -179,7 +180,6 @@ static bool srcu_readers_active(struct srcu_struct *sp)
 	return sum;
 }
 
-#define SRCU_CALLBACK_BATCH	10
 #define SRCU_INTERVAL		1
 
 /**
@@ -191,6 +191,7 @@ static bool srcu_readers_active(struct srcu_struct *sp)
  */
 void cleanup_srcu_struct(struct srcu_struct *sp)
 {
+	WARN_ON_ONCE(atomic_read(&sp->srcu_exp_cnt));
 	if (WARN_ON(srcu_readers_active(sp)))
 		return; /* Leakage unless caller handles error. */
 	if (WARN_ON(!rcu_segcblist_empty(&sp->srcu_cblist)))
@@ -238,13 +239,10 @@ EXPORT_SYMBOL_GPL(__srcu_read_unlock);
  * We use an adaptive strategy for synchronize_srcu() and especially for
  * synchronize_srcu_expedited().  We spin for a fixed time period
  * (defined below) to allow SRCU readers to exit their read-side critical
- * sections.  If there are still some readers after 10 microseconds,
- * we repeatedly block for 1-millisecond time periods.  This approach
- * has done well in testing, so there is no need for a config parameter.
+ * sections.  If there are still some readers after a few microseconds,
+ * we repeatedly block for 1-millisecond time periods.
  */
 #define SRCU_RETRY_CHECK_DELAY		5
-#define SYNCHRONIZE_SRCU_TRYCOUNT	2
-#define SYNCHRONIZE_SRCU_EXP_TRYCOUNT	12
 
 /*
  * Start an SRCU grace period.
@@ -261,16 +259,16 @@ static void srcu_gp_start(struct srcu_struct *sp)
 }
 
 /*
- * Wait until all readers counted by array index idx complete, but loop
- * a maximum of trycount times.  The caller must ensure that ->completed
- * is not changed while checking.
+ * Wait until all readers counted by array index idx complete, but
+ * loop an additional time if there is an expedited grace period pending.
+ * The caller must ensure that ->completed is not changed while checking.
  */
 static bool try_check_zero(struct srcu_struct *sp, int idx, int trycount)
 {
 	for (;;) {
 		if (srcu_readers_active_idx_check(sp, idx))
 			return true;
-		if (--trycount <= 0)
+		if (--trycount + !!atomic_read(&sp->srcu_exp_cnt) <= 0)
 			return false;
 		udelay(SRCU_RETRY_CHECK_DELAY);
 	}
@@ -358,7 +356,7 @@ static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay);
 /*
  * Helper function for synchronize_srcu() and synchronize_srcu_expedited().
  */
-static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
+static void __synchronize_srcu(struct srcu_struct *sp)
 {
 	struct rcu_synchronize rcu;
 	struct rcu_head *head = &rcu.head;
@@ -395,6 +393,32 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
 }
 
 /**
+ * synchronize_srcu_expedited - Brute-force SRCU grace period
+ * @sp: srcu_struct with which to synchronize.
+ *
+ * Wait for an SRCU grace period to elapse, but be more aggressive about
+ * spinning rather than blocking when waiting.
+ *
+ * Note that synchronize_srcu_expedited() has the same deadlock and
+ * memory-ordering properties as does synchronize_srcu().
+ */
+void synchronize_srcu_expedited(struct srcu_struct *sp)
+{
+	bool do_norm = rcu_gp_is_normal();
+
+	if (!do_norm) {
+		atomic_inc(&sp->srcu_exp_cnt);
+		smp_mb__after_atomic(); /* increment before GP. */
+	}
+	__synchronize_srcu(sp);
+	if (!do_norm) {
+		smp_mb__before_atomic(); /* GP before decrement. */
+		atomic_dec(&sp->srcu_exp_cnt);
+	}
+}
+EXPORT_SYMBOL_GPL(synchronize_srcu_expedited);
+
+/**
  * synchronize_srcu - wait for prior SRCU read-side critical-section completion
  * @sp: srcu_struct with which to synchronize.
  *
@@ -435,29 +459,14 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
  */
 void synchronize_srcu(struct srcu_struct *sp)
 {
-	__synchronize_srcu(sp, (rcu_gp_is_expedited() && !rcu_gp_is_normal())
-			   ? SYNCHRONIZE_SRCU_EXP_TRYCOUNT
-			   : SYNCHRONIZE_SRCU_TRYCOUNT);
+	if (rcu_gp_is_expedited())
+		synchronize_srcu_expedited(sp);
+	else
+		__synchronize_srcu(sp);
 }
 EXPORT_SYMBOL_GPL(synchronize_srcu);
 
 /**
- * synchronize_srcu_expedited - Brute-force SRCU grace period
- * @sp: srcu_struct with which to synchronize.
- *
- * Wait for an SRCU grace period to elapse, but be more aggressive about
- * spinning rather than blocking when waiting.
- *
- * Note that synchronize_srcu_expedited() has the same deadlock and
- * memory-ordering properties as does synchronize_srcu().
- */
-void synchronize_srcu_expedited(struct srcu_struct *sp)
-{
-	__synchronize_srcu(sp, SYNCHRONIZE_SRCU_EXP_TRYCOUNT);
-}
-EXPORT_SYMBOL_GPL(synchronize_srcu_expedited);
-
-/**
  * srcu_barrier - Wait until all in-flight call_srcu() callbacks complete.
  * @sp: srcu_struct on which to wait for in-flight callbacks.
  */
@@ -484,7 +493,7 @@ EXPORT_SYMBOL_GPL(srcu_batches_completed);
  * Core SRCU state machine.  Advance callbacks from ->batch_check0 to
  * ->batch_check1 and then to ->batch_done as readers drain.
  */
-static void srcu_advance_batches(struct srcu_struct *sp, int trycount)
+static void srcu_advance_batches(struct srcu_struct *sp)
 {
 	int idx;
 
@@ -515,8 +524,8 @@ static void srcu_advance_batches(struct srcu_struct *sp, int trycount)
 
 	if (rcu_seq_state(READ_ONCE(sp->srcu_gp_seq)) == SRCU_STATE_SCAN1) {
 		idx = 1 ^ (sp->completed & 1);
-		if (!try_check_zero(sp, idx, trycount))
-			return; /* readers present, retry after SRCU_INTERVAL */
+		if (!try_check_zero(sp, idx, 1))
+			return; /* readers present, retry later. */
 		srcu_flip(sp);
 		rcu_seq_set_state(&sp->srcu_gp_seq, SRCU_STATE_SCAN2);
 	}
@@ -528,9 +537,8 @@ static void srcu_advance_batches(struct srcu_struct *sp, int trycount)
 		 * so check at least twice in quick succession after a flip.
 		 */
 		idx = 1 ^ (sp->completed & 1);
-		trycount = trycount < 2 ? 2 : trycount;
-		if (!try_check_zero(sp, idx, trycount))
-			return; /* readers present, retry after SRCU_INTERVAL */
+		if (!try_check_zero(sp, idx, 2))
+			return; /* readers present, retry after later. */
 		srcu_gp_end(sp);
 	}
 }
@@ -596,8 +604,8 @@ void process_srcu(struct work_struct *work)
 
 	sp = container_of(work, struct srcu_struct, work.work);
 
-	srcu_advance_batches(sp, 1);
+	srcu_advance_batches(sp);
 	srcu_invoke_callbacks(sp);
-	srcu_reschedule(sp, SRCU_INTERVAL);
+	srcu_reschedule(sp, atomic_read(&sp->srcu_exp_cnt) ? 0 : SRCU_INTERVAL);
 }
 EXPORT_SYMBOL_GPL(process_srcu);
-- 
2.5.2