From 1c77e862b81e8fe84d3a908271c72da9711ea801 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Tue, 22 Jul 2025 14:57:33 -0700 Subject: [PATCH] srcu: Document srcu_flip() memory-barrier D relation to SRCU-fast The smp_mb() memory barrier at the end of srcu_flip() has a comment, but that comment does not make it clear that this memory barrier is an optimization, as opposed to being needed for correctness. This commit therefore adds this information and points out that it is omitted for SRCU-fast, where a much heavier weight synchronize_srcu() would be required. Signed-off-by: Paul E. McKenney Cc: Mathieu Desnoyers Cc: Steven Rostedt Cc: Sebastian Andrzej Siewior Cc: --- kernel/rcu/srcutree.c | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c index c5e8ebc493d5e..1ff94b76d91f1 100644 --- a/kernel/rcu/srcutree.c +++ b/kernel/rcu/srcutree.c @@ -1168,6 +1168,16 @@ static void srcu_flip(struct srcu_struct *ssp) * counter update. Note that both this memory barrier and the * one in srcu_readers_active_idx_check() provide the guarantee * for __srcu_read_lock(). + * + * Note that this is a performance optimization, in which we spend + * an otherwise unnecessary smp_mb() in order to reduce the number + * of full per-CPU-variable scans in srcu_readers_lock_idx() and + * srcu_readers_unlock_idx(). But this performance optimization + * is not so optimal for SRCU-fast, where we would be spending + * not smp_mb(), but rather synchronize_rcu(). At the same time, + * the overhead of the smp_mb() is in the noise, so there is no + * point in omitting it in the SRCU-fast case. So the same code + * is executed either way. */ smp_mb(); /* D */ /* Pairs with C. */ } -- 2.47.3