diff --git a/Documentation/core-api/rbtree.rst b/Documentation/core-api/rbtree.rst
index cce80e19087b..210045f128d6 100644
--- a/Documentation/core-api/rbtree.rst
+++ b/Documentation/core-api/rbtree.rst
@@ -257,8 +257,14 @@ When erasing a node, the user must call rb_erase_augmented() instead of
 rb_erase(). rb_erase_augmented() calls back into user provided functions
 to update the augmented information on affected subtrees.
 
-In both cases, the callbacks are provided through struct rb_augment_callbacks.
-3 callbacks must be defined:
+Alternatively, rb_add_augmented_cached() performs the whole insertion into
+a cached tree: it walks down to the insertion point, merges the new node's
+augmented information into every node on that path, links the node and
+rebalances.  The new node's augmented information must already describe
+the node alone when it is called.
+
+In all cases, the callbacks are provided through struct rb_augment_callbacks.
+3 callbacks must be defined, plus a fourth one for rb_add_augmented_cached():
 
 - A propagation callback, which updates the augmented value for a given
   node and its ancestors, up to a given stop point (or NULL to update
@@ -271,6 +277,10 @@ In both cases, the callbacks are provided through struct rb_augment_callbacks.
   subtree to a newly assigned subtree root AND recomputes the augmented
   information for the former subtree root.
 
+- A merge callback, which folds the augmented value of a node being inserted
+  into the augmented value of one of its future ancestors.  It is only used
+  by rb_add_augmented_cached().
+
 The compiled code for rb_erase_augmented() may inline the propagation and
 copy callbacks, which results in a large function, so each augmented rbtree
 user should have a single rb_erase_augmented() call site in order to limit
@@ -395,8 +405,19 @@ Insertion/removal are defined using the following augmented callbacks::
 	old->__subtree_last = compute_subtree_last(old);
   }
 
+  static void augment_merge(struct rb_node *rb, struct rb_node *rb_new)
+  {
+	struct interval_tree_node *node =
+		rb_entry(rb, struct interval_tree_node, rb);
+	struct interval_tree_node *new =
+		rb_entry(rb_new, struct interval_tree_node, rb);
+
+	if (node->__subtree_last < new->__subtree_last)
+		node->__subtree_last = new->__subtree_last;
+  }
+
   static const struct rb_augment_callbacks augment_callbacks = {
-	augment_propagate, augment_copy, augment_rotate
+	augment_propagate, augment_copy, augment_rotate, augment_merge
   };
 
   void interval_tree_insert(struct interval_tree_node *node,
diff --git a/include/linux/rbtree_augmented.h b/include/linux/rbtree_augmented.h
index 5a336ac203c1..b9fc1be86ae3 100644
--- a/include/linux/rbtree_augmented.h
+++ b/include/linux/rbtree_augmented.h
@@ -30,6 +30,7 @@ struct rb_augment_callbacks {
 	void (*propagate)(struct rb_node *node, struct rb_node *stop);
 	void (*copy)(struct rb_node *old, struct rb_node *new);
 	void (*rotate)(struct rb_node *old, struct rb_node *new);
+	void (*merge)(struct rb_node *node, struct rb_node *new);
 };
 
 extern void __rb_insert_augmented(struct rb_node *node, struct rb_root *root,
@@ -62,6 +63,12 @@ rb_insert_augmented_cached(struct rb_node *node,
 	rb_insert_augmented(node, &root->rb_root, augment);
 }
 
+/*
+ * Insert @node into the leftmost cached augmented tree @tree.
+ *
+ * The augmented data of @node must already describe @node alone; it is
+ * merged into every ancestor on the way down through augment->merge().
+ */
 static __always_inline struct rb_node *
 rb_add_augmented_cached(struct rb_node *node, struct rb_root_cached *tree,
 			bool (*less)(struct rb_node *, const struct rb_node *),
@@ -73,6 +80,7 @@ rb_add_augmented_cached(struct rb_node *node, struct rb_root_cached *tree,
 
 	while (*link) {
 		parent = *link;
+		augment->merge(parent, node);
 		if (less(node, parent)) {
 			link = &parent->rb_left;
 		} else {
@@ -82,7 +90,6 @@ rb_add_augmented_cached(struct rb_node *node, struct rb_root_cached *tree,
 	}
 
 	rb_link_node(node, parent, link);
-	augment->propagate(parent, NULL); /* suboptimal */
 	rb_insert_augmented_cached(node, tree, leftmost, augment);
 
 	return leftmost ? node : NULL;
@@ -177,6 +184,14 @@ RBNAME ## _compute_ ## n(RBSTRUCT *s, bool exit)			\
 #define RB_COMPUTE(n, RBNAME, RBSTRUCT, RBFIELD, x)			\
 	_RB_COMPUTE(n, RBNAME, RBSTRUCT, RBFIELD, RB_UNPACK x)
 
+/* fold @new, about to become a descendant of @node, into @node */
+#define __RB_MERGE(n, RBNAME, RBSTRUCT, RBFIELD, val, aug, fold)	\
+	node->aug = fold(node->aug, new->aug);
+#define _RB_MERGE(n, RBNAME, RBSTRUCT, RBFIELD, args)			\
+	__RB_MERGE(n, RBNAME, RBSTRUCT, RBFIELD, args)
+#define RB_MERGE(n, RBNAME, RBSTRUCT, RBFIELD, x)			\
+	_RB_MERGE(n, RBNAME, RBSTRUCT, RBFIELD, RB_UNPACK x)
+
 /*
  * Template for declaring augmented rbtree callbacks (generic multi fields)
  *
@@ -226,10 +241,18 @@ RBNAME ## _rotate(struct rb_node *rb_old, struct rb_node *rb_new)	\
 	RBNAME ## __copy(old, new);					\
 	RBNAME ## __compute(old, false);				\
 }									\
+static inline void							\
+RBNAME ## _merge(struct rb_node *rb, struct rb_node *rb_new)		\
+{									\
+	RBSTRUCT *node = rb_entry(rb, RBSTRUCT, RBFIELD);		\
+	RBSTRUCT *new = rb_entry(rb_new, RBSTRUCT, RBFIELD);		\
+	RB_FOR_EACH(RB_MERGE, RBNAME, RBSTRUCT, RBFIELD, RBAUG);	\
+}									\
 RBSTATIC const struct rb_augment_callbacks RBNAME = {			\
 	.propagate = RBNAME ## _propagate,				\
 	.copy = RBNAME ## _copy,					\
-	.rotate = RBNAME ## _rotate					\
+	.rotate = RBNAME ## _rotate,					\
+	.merge = RBNAME ## _merge					\
 };
 
 /*