rcutorture: ASSERT_EXCLUSIVE_WRITER() for ->rtort_pipe_count updates
authorPaul E. McKenney <paulmck@kernel.org>
Thu, 7 Mar 2024 01:00:39 +0000 (17:00 -0800)
committerUladzislau Rezki (Sony) <urezki@gmail.com>
Tue, 9 Apr 2024 13:10:13 +0000 (15:10 +0200)
It turns out that only one CPU at a time will ever invoke
rcu_torture_pipe_update_one() on a given rcu_torture structure.
This commit therefore adds three ASSERT_EXCLUSIVE_WRITER() calls
to enlist KCSAN's aid in checking this.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
kernel/rcu/rcutorture.c

index eff51a26216f300f51bb1164634bfe262ed63475..d8c12eba35b7fa6aad1454c7a7098cc2a8f557dd 100644 (file)
@@ -466,6 +466,7 @@ rcu_torture_pipe_update_one(struct rcu_torture *rp)
                i = RCU_TORTURE_PIPE_LEN;
        atomic_inc(&rcu_torture_wcount[i]);
        WRITE_ONCE(rp->rtort_pipe_count, i + 1);
+       ASSERT_EXCLUSIVE_WRITER(rp->rtort_pipe_count);
        if (rp->rtort_pipe_count >= RCU_TORTURE_PIPE_LEN) {
                rp->rtort_mbtest = 0;
                return true;
@@ -1399,6 +1400,7 @@ rcu_torture_writer(void *arg)
                if (rp == NULL)
                        continue;
                rp->rtort_pipe_count = 0;
+               ASSERT_EXCLUSIVE_WRITER(rp->rtort_pipe_count);
                rcu_torture_writer_state = RTWS_DELAY;
                udelay(torture_random(&rand) & 0x3ff);
                rcu_torture_writer_state = RTWS_REPLACE;
@@ -1414,6 +1416,7 @@ rcu_torture_writer(void *arg)
                        atomic_inc(&rcu_torture_wcount[i]);
                        WRITE_ONCE(old_rp->rtort_pipe_count,
                                   old_rp->rtort_pipe_count + 1);
+                       ASSERT_EXCLUSIVE_WRITER(old_rp->rtort_pipe_count);
 
                        // Make sure readers block polled grace periods.
                        if (cur_ops->get_gp_state && cur_ops->poll_gp_state) {