On Sun, Mar 31, 2024 at 09:19:32AM +0000, Alan Huang wrote:
> The two READ_ONCEs here imply that there might be updater concurrently
> modifies rcu_read_lock_nesting, although it's true there could be
> interrupt handlers enter rcu RSCS, but rcu_read_lock_nesting should
> remain unchanged after they finished. That is, READ_ONCE and then
> WRITE_ONCE the same variable is not a valid pattern, this patch thus
> removes the two READ_ONCEs.
This is true, but I will take the READ_ONCE() calls over the KCSAN
complaints. So I must avoid taking this one.
Please note that RCU uses strict KCSAN configuration [1].
Thanx, Paul
[1]
https://docs.google.com/document/d/1FwZaXSg3A55ivVoWffA9iMuhJ3_Gmj_E494dLYjjyLQ/edit?usp=sharing
> Signed-off-by: Alan Huang <[email protected]>
> ---
> kernel/rcu/tree_plugin.h | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> index 340bbefe5f65..c90edbb7fa40 100644
> --- a/kernel/rcu/tree_plugin.h
> +++ b/kernel/rcu/tree_plugin.h
> @@ -376,12 +376,12 @@ static int rcu_preempt_blocked_readers_cgp(struct
> rcu_node *rnp)
>
> static void rcu_preempt_read_enter(void)
> {
> - WRITE_ONCE(current->rcu_read_lock_nesting,
> READ_ONCE(current->rcu_read_lock_nesting) + 1);
> + WRITE_ONCE(current->rcu_read_lock_nesting,
> current->rcu_read_lock_nesting + 1);
> }
>
> static int rcu_preempt_read_exit(void)
> {
> - int ret = READ_ONCE(current->rcu_read_lock_nesting) - 1;
> + int ret = current->rcu_read_lock_nesting - 1;
>
> WRITE_ONCE(current->rcu_read_lock_nesting, ret);
> return ret;
> --
> 2.34.1
>
>
