Re: [PATCH 3/4] rhashtable: use bit_spin_locks to protect hash bucket.

[Herbert Xu]

Hi Neil:

On Tue, Apr 02, 2019 at 10:07:45AM +1100, NeilBrown wrote:
>
> @@ -263,13 +311,13 @@ void rhashtable_free_and_destroy(struct rhashtable *ht,
>  				 void *arg);
>  void rhashtable_destroy(struct rhashtable *ht);
>  
> -struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
> -					    unsigned int hash);
> -struct rhash_head __rcu **__rht_bucket_nested(const struct bucket_table *tbl,
> -					      unsigned int hash);
> -struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
> -						   struct bucket_table *tbl,
> +struct rhash_lock_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
> +						 unsigned int hash);

I don't think this opaque type should be marked as __rcu.  Because
you can't directly dereference it and once you put it through rht_ptr
then that's the pointer that should carry the __rcu marker.

If you add the __rcu here then you generate a lot of extra noise
in the code that isn't needed.

Cheers,
-- 
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt