Re: [syzbot] KCSAN: data-race in call_rcu / rcu_gp_kthread

[Marco Elver]

On Tue, 20 Jul 2021 at 15:18, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
[...]
> Good catch!  And yes, this would be hard to reproduce.
>
> How about as shown below?

Acked-by: Marco Elver <elver@xxxxxxxxxx>

I was merely a little surprised syzbot was able to exercise RCU in a
way that resulted in a data race your torture runs hadn't found yet
(or perhaps it did and missed?).

Thanks,
-- Marco

>                                                         Thanx, Paul
>
> ------------------------------------------------------------------------
>
> commit 43e0f01f3b6f510dbe31d02a8f4c909c45deff04
> Author: Paul E. McKenney <paulmck@xxxxxxxxxx>
> Date:   Tue Jul 20 06:16:27 2021 -0700
>
>     rcu: Mark accesses to rcu_state.n_force_qs
>
>     This commit marks accesses to the rcu_state.n_force_qs.  These data
>     races are hard to make happen, but syzkaller was equal to the task.
>
>     Reported-by: syzbot+e08a83a1940ec3846cd5@xxxxxxxxxxxxxxxxxxxxxxxxx
>     Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
>
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index a7379c44a2366..245bca7cdf6ee 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -1913,7 +1913,7 @@ static void rcu_gp_fqs(bool first_time)
>         struct rcu_node *rnp = rcu_get_root();
>
>         WRITE_ONCE(rcu_state.gp_activity, jiffies);
> -       rcu_state.n_force_qs++;
> +       WRITE_ONCE(rcu_state.n_force_qs, rcu_state.n_force_qs + 1);
>         if (first_time) {
>                 /* Collect dyntick-idle snapshots. */
>                 force_qs_rnp(dyntick_save_progress_counter);
> @@ -2556,7 +2556,7 @@ static void rcu_do_batch(struct rcu_data *rdp)
>         /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
>         if (count == 0 && rdp->qlen_last_fqs_check != 0) {
>                 rdp->qlen_last_fqs_check = 0;
> -               rdp->n_force_qs_snap = rcu_state.n_force_qs;
> +               rdp->n_force_qs_snap = READ_ONCE(rcu_state.n_force_qs);
>         } else if (count < rdp->qlen_last_fqs_check - qhimark)
>                 rdp->qlen_last_fqs_check = count;
>
> @@ -2904,10 +2904,10 @@ static void __call_rcu_core(struct rcu_data *rdp, struct rcu_head *head,
>                 } else {
>                         /* Give the grace period a kick. */
>                         rdp->blimit = DEFAULT_MAX_RCU_BLIMIT;
> -                       if (rcu_state.n_force_qs == rdp->n_force_qs_snap &&
> +                       if (READ_ONCE(rcu_state.n_force_qs) == rdp->n_force_qs_snap &&
>                             rcu_segcblist_first_pend_cb(&rdp->cblist) != head)
>                                 rcu_force_quiescent_state();
> -                       rdp->n_force_qs_snap = rcu_state.n_force_qs;
> +                       rdp->n_force_qs_snap = READ_ONCE(rcu_state.n_force_qs);
>                         rdp->qlen_last_fqs_check = rcu_segcblist_n_cbs(&rdp->cblist);
>                 }
>         }
> @@ -4134,7 +4134,7 @@ int rcutree_prepare_cpu(unsigned int cpu)
>         /* Set up local state, ensuring consistent view of global state. */
>         raw_spin_lock_irqsave_rcu_node(rnp, flags);
>         rdp->qlen_last_fqs_check = 0;
> -       rdp->n_force_qs_snap = rcu_state.n_force_qs;
> +       rdp->n_force_qs_snap = READ_ONCE(rcu_state.n_force_qs);
>         rdp->blimit = blimit;
>         rdp->dynticks_nesting = 1;      /* CPU not up, no tearing. */
>         rcu_dynticks_eqs_online();