rcu: Throttle callback invocation based on number of ready callbacks

Currently, rcu_do_batch() sizes its batches based on the total number
of callbacks in the callback list.  This can result in some strange
choices, for example, if there was 12,800 callbacks in the list, but
only 200 were ready to invoke, RCU would invoke 100 at a time (12,800
shifted down by seven bits).

A more measured approach would use the number that were actually ready
to invoke, an approach that has become feasible only recently given the
per-segment ->seglen counts in ->cblist.

This commit therefore bases the batch limit on the number of callbacks
ready to invoke instead of on the total number of callbacks.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>

kernel/rcu/rcu_segcblist.c

@@ -89,7 +89,7 @@ static void rcu_segcblist_set_len(struct rcu_segcblist *rsclp, long v)
 }
 
 /* Get the length of a segment of the rcu_segcblist structure. */
-static long rcu_segcblist_get_seglen(struct rcu_segcblist *rsclp, int seg)
+long rcu_segcblist_get_seglen(struct rcu_segcblist *rsclp, int seg)
 {
 	return READ_ONCE(rsclp->seglen[seg]);
 }

kernel/rcu/rcu_segcblist.h

@@ -15,6 +15,8 @@ static inline long rcu_cblist_n_cbs(struct rcu_cblist *rclp)
 	return READ_ONCE(rclp->len);
 }
 
+long rcu_segcblist_get_seglen(struct rcu_segcblist *rsclp, int seg);
+
 /* Return number of callbacks in segmented callback list by summing seglen. */
 long rcu_segcblist_n_segment_cbs(struct rcu_segcblist *rsclp);
 

kernel/rcu/tree.c

@@ -2057,7 +2057,7 @@ static void rcu_do_batch(struct rcu_data *rdp)
 	 */
 	rcu_nocb_lock_irqsave(rdp, flags);
 	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
-	pending = rcu_segcblist_n_cbs(&rdp->cblist);
+	pending = rcu_segcblist_get_seglen(&rdp->cblist, RCU_DONE_TAIL);
 	div = READ_ONCE(rcu_divisor);
 	div = div < 0 ? 7 : div > sizeof(long) * 8 - 2 ? sizeof(long) * 8 - 2 : div;
 	bl = max(rdp->blimit, pending >> div);