rcu: Add WRITE_ONCE() to rcu_state ->gp_req_activity (2906d215) · Commits · 戴 / test

The rcu_state structure's ->gp_req_activity field is read locklessly, so this commit adds the WRITE_ONCE() to an update in order to provide proper documentation and READ_ONCE()/WRITE_ONCE() pairing. This data race was reported by KCSAN. Not appropriate for backporting due to failure being unlikely. Signed-off-by:

Paul E. McKenney <paulmck@kernel.org>

kernel/rcu/tree.c

+2 −2

Original line number	Diff line number	Diff line
		@@ -1200,7 +1200,7 @@ static bool rcu_start_this_gp(struct rcu_node rnp_start, struct rcu_data rdp,
		}
		trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startedroot"));
		WRITE_ONCE(rcu_state.gp_flags, rcu_state.gp_flags \| RCU_GP_FLAG_INIT);
		rcu_state.gp_req_activity = jiffies;
		WRITE_ONCE(rcu_state.gp_req_activity, jiffies);
		if (!rcu_state.gp_kthread) {
		trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("NoGPkthread"));
		goto unlock_out;
		@@ -1775,7 +1775,7 @@ static void rcu_gp_cleanup(void)
		rcu_segcblist_is_offloaded(&rdp->cblist);
		if ((offloaded \|\| !rcu_accelerate_cbs(rnp, rdp)) && needgp) {
		WRITE_ONCE(rcu_state.gp_flags, RCU_GP_FLAG_INIT);
		rcu_state.gp_req_activity = jiffies;
		WRITE_ONCE(rcu_state.gp_req_activity, jiffies);
		trace_rcu_grace_period(rcu_state.name,
		READ_ONCE(rcu_state.gp_seq),
		TPS("newreq"));

Admin message