Re: [PATCH rcu/dev -fixes 3/4] rcu/tree: Avoid using xchg() in kfree_call_rcu_add_ptr_to_bulk()

[Uladzislau Rezki]

On Mon, Apr 20, 2020 at 11:38:36AM -0400, Joel Fernandes (Google) wrote:
> From: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
> 
> There is no need to use xchg(), the access is serialized by krcp->lock.
> The xchg() function adds some atomic barriers which remain hidden in
> x86's disassembly but are visible on ARM for instance.
> 
> Replace xchg() with a load + store.
> 
> Acked-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
> ---
>  kernel/rcu/tree.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index cd61649e1b001..f6eb3aee0935e 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -3088,7 +3088,8 @@ kfree_call_rcu_add_ptr_to_bulk(struct kfree_rcu_cpu *krcp,
>  	/* Check if a new block is required. */
>  	if (!krcp->bhead ||
>  			krcp->bhead->nr_records == KFREE_BULK_MAX_ENTR) {
> -		bnode = xchg(&krcp->bcached, NULL);
> +		bnode = krcp->bcached;
> +		krcp->bcached = NULL;
>  		if (!bnode) {
>  			WARN_ON_ONCE(sizeof(struct kfree_rcu_bulk_data) > PAGE_SIZE);
>  
>
But we populate the cache without holding the krcp->lock. That is why it
is xchg here. See below:

<snip>
  if (cmpxchg(&krcp->bcached, NULL, bhead))
     free_page((unsigned long) bhead);
<snip>

we do not hold any krcp->lock here, we do not need it.

--
Vlad Rezki