srcu: Apply trivial callback lists to shrink Tiny SRCU

The rcu_segcblist structure provides quite a bit of functionality, and
Tiny SRCU needs almost none of it.  So this commit replaces Tiny SRCU's
uses of rcu_segcblist with a simple singly linked list with tail pointer.
This change significantly reduces Tiny SRCU's memory footprint, more
than making up for the growth caused by the creation of rcu_segcblist.c

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

include/linux/srcutiny.h

@@ -33,9 +33,8 @@ struct srcu_struct {
 	u8 srcu_gp_waiting;		/* GP waiting for readers? */
 	struct swait_queue_head srcu_wq;
 					/* Last srcu_read_unlock() wakes GP. */
-	unsigned long srcu_gp_seq;	/* GP seq # for callback tagging. */
+	struct rcu_head *srcu_cb_head;	/* Pending callbacks: Head. */
-	struct rcu_segcblist srcu_cblist;
+	struct rcu_head **srcu_cb_tail;	/* Pending callbacks: Tail. */
-					/* Pending SRCU callbacks. */
 	struct work_struct srcu_work;	/* For driving grace periods. */
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 	struct lockdep_map dep_map;
@@ -47,7 +46,7 @@ void srcu_drive_gp(struct work_struct *wp);
 #define __SRCU_STRUCT_INIT(name)					\
 {									\
 	.srcu_wq = __SWAIT_QUEUE_HEAD_INITIALIZER(name.srcu_wq),	\
-	.srcu_cblist = RCU_SEGCBLIST_INITIALIZER(name.srcu_cblist),	\
+	.srcu_cb_tail = &name.srcu_cb_head,				\
 	.srcu_work = __WORK_INITIALIZER(name.srcu_work, srcu_drive_gp),	\
 	__SRCU_DEP_MAP_INIT(name)					\
 }

init/Kconfig

@@ -573,7 +573,7 @@ config RCU_STALL_COMMON
 	  making these warnings mandatory for the tree variants.
 
 config RCU_NEED_SEGCBLIST
-	def_bool ( TREE_RCU || PREEMPT_RCU || TINY_SRCU || TREE_SRCU )
+	def_bool ( TREE_RCU || PREEMPT_RCU || TREE_SRCU )
 
 config CONTEXT_TRACKING
        bool

kernel/rcu/rcu.h

@@ -398,7 +398,7 @@ static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
 	if (test_type != SRCU_FLAVOR)
 		return;
 	*flags = 0;
-	*completed = sp->srcu_gp_seq;
+	*completed = sp->srcu_idx;
 	*gpnum = *completed;
 }
 

kernel/rcu/srcutiny.c

@@ -38,8 +38,8 @@ static int init_srcu_struct_fields(struct srcu_struct *sp)
 	sp->srcu_lock_nesting[0] = 0;
 	sp->srcu_lock_nesting[1] = 0;
 	init_swait_queue_head(&sp->srcu_wq);
-	sp->srcu_gp_seq = 0;
+	sp->srcu_cb_head = NULL;
-	rcu_segcblist_init(&sp->srcu_cblist);
+	sp->srcu_cb_tail = &sp->srcu_cb_head;
 	sp->srcu_gp_running = false;
 	sp->srcu_gp_waiting = false;
 	sp->srcu_idx = 0;
@@ -88,10 +88,10 @@ void cleanup_srcu_struct(struct srcu_struct *sp)
 {
 	WARN_ON(sp->srcu_lock_nesting[0] || sp->srcu_lock_nesting[1]);
 	flush_work(&sp->srcu_work);
-	WARN_ON(rcu_seq_state(sp->srcu_gp_seq));
 	WARN_ON(sp->srcu_gp_running);
 	WARN_ON(sp->srcu_gp_waiting);
-	WARN_ON(!rcu_segcblist_empty(&sp->srcu_cblist));
+	WARN_ON(sp->srcu_cb_head);
+	WARN_ON(&sp->srcu_cb_head != sp->srcu_cb_tail);
 }
 EXPORT_SYMBOL_GPL(cleanup_srcu_struct);
 
@@ -117,52 +117,44 @@ EXPORT_SYMBOL_GPL(__srcu_read_unlock);
 void srcu_drive_gp(struct work_struct *wp)
 {
 	int idx;
-	struct rcu_cblist ready_cbs;
+	struct rcu_head *lh;
-	struct srcu_struct *sp;
 	struct rcu_head *rhp;
+	struct srcu_struct *sp;
 
 	sp = container_of(wp, struct srcu_struct, srcu_work);
-	if (sp->srcu_gp_running || rcu_segcblist_empty(&sp->srcu_cblist))
+	if (sp->srcu_gp_running || !READ_ONCE(sp->srcu_cb_head))
 		return; /* Already running or nothing to do. */
 
-	/* Tag recently arrived callbacks and wait for readers. */
+	/* Remove recently arrived callbacks and wait for readers. */
 	WRITE_ONCE(sp->srcu_gp_running, true);
-	rcu_segcblist_accelerate(&sp->srcu_cblist,
+	local_irq_disable();
-				 rcu_seq_snap(&sp->srcu_gp_seq));
+	lh = sp->srcu_cb_head;
-	rcu_seq_start(&sp->srcu_gp_seq);
+	sp->srcu_cb_head = NULL;
+	sp->srcu_cb_tail = &sp->srcu_cb_head;
+	local_irq_enable();
 	idx = sp->srcu_idx;
 	WRITE_ONCE(sp->srcu_idx, !sp->srcu_idx);
 	WRITE_ONCE(sp->srcu_gp_waiting, true);  /* srcu_read_unlock() wakes! */
 	swait_event(sp->srcu_wq, !READ_ONCE(sp->srcu_lock_nesting[idx]));
 	WRITE_ONCE(sp->srcu_gp_waiting, false); /* srcu_read_unlock() cheap. */
-	rcu_seq_end(&sp->srcu_gp_seq);
 
-	/* Update callback list based on GP, and invoke ready callbacks. */
+	/* Invoke the callbacks we removed above. */
-	rcu_segcblist_advance(&sp->srcu_cblist,
+	while (lh) {
-			      rcu_seq_current(&sp->srcu_gp_seq));
+		rhp = lh;
-	if (rcu_segcblist_ready_cbs(&sp->srcu_cblist)) {
+		lh = lh->next;
-		rcu_cblist_init(&ready_cbs);
+		local_bh_disable();
-		local_irq_disable();
+		rhp->func(rhp);
-		rcu_segcblist_extract_done_cbs(&sp->srcu_cblist, &ready_cbs);
+		local_bh_enable();
-		local_irq_enable();
-		rhp = rcu_cblist_dequeue(&ready_cbs);
-		for (; rhp != NULL; rhp = rcu_cblist_dequeue(&ready_cbs)) {
-			local_bh_disable();
-			rhp->func(rhp);
-			local_bh_enable();
-		}
-		local_irq_disable();
-		rcu_segcblist_insert_count(&sp->srcu_cblist, &ready_cbs);
-		local_irq_enable();
 	}
-	WRITE_ONCE(sp->srcu_gp_running, false);
 
 	/*
-	 * If more callbacks, reschedule ourselves.  This can race with
+	 * Enable rescheduling, and if there are more callbacks,
-	 * a call_srcu() at interrupt level, but the ->srcu_gp_running
+	 * reschedule ourselves.  This can race with a call_srcu()
-	 * checks will straighten that out.
+	 * at interrupt level, but the ->srcu_gp_running checks will
+	 * straighten that out.
 	 */
-	if (!rcu_segcblist_empty(&sp->srcu_cblist))
+	WRITE_ONCE(sp->srcu_gp_running, false);
+	if (READ_ONCE(sp->srcu_cb_head))
 		schedule_work(&sp->srcu_work);
 }
 EXPORT_SYMBOL_GPL(srcu_drive_gp);
@@ -171,14 +163,16 @@ EXPORT_SYMBOL_GPL(srcu_drive_gp);
  * Enqueue an SRCU callback on the specified srcu_struct structure,
  * initiating grace-period processing if it is not already running.
  */
-void call_srcu(struct srcu_struct *sp, struct rcu_head *head,
+void call_srcu(struct srcu_struct *sp, struct rcu_head *rhp,
 	       rcu_callback_t func)
 {
 	unsigned long flags;
 
-	head->func = func;
+	rhp->func = func;
+	rhp->next = NULL;
 	local_irq_save(flags);
-	rcu_segcblist_enqueue(&sp->srcu_cblist, head, false);
+	*sp->srcu_cb_tail = rhp;
+	sp->srcu_cb_tail = &rhp->next;
 	local_irq_restore(flags);
 	if (!READ_ONCE(sp->srcu_gp_running))
 		schedule_work(&sp->srcu_work);