Re: [PATCH 0/3] mm: replace atomic_t with percpu_ref in mempolicy.

[Michal Hocko <mhocko@suse.com>]

On Mon 05-12-22 00:14:29, Zhongkun He wrote:
[...]
> +/* Obtain a reference on the specified mpol */
>  static inline void mpol_get(struct mempolicy *pol)
>  {
>  	if (pol)

Shouldn't this be mpol_needs_cond_ref?

> -		atomic_inc(&pol->refcnt);
> +		percpu_ref_get(&pol->refcnt);
> +}
> +
> +static inline bool mpol_tryget(struct mempolicy *pol)
> +{
> +	return pol && percpu_ref_tryget(&pol->refcnt);
>  }
>  
> +/*
> + * This function initiates destruction of mempolicy.

This is not a useful comment. It would be much more helpful to say when
this should be called.

> + */
> +static inline void mpol_kill(struct mempolicy *pol)
> +{
> +        if (pol)
> +                percpu_ref_kill(&pol->refcnt);
> +}
> +
> +
>  extern bool __mpol_equal(struct mempolicy *a, struct mempolicy *b);
>  static inline bool mpol_equal(struct mempolicy *a, struct mempolicy *b)
>  {
> @@ -197,11 +210,15 @@ static inline void mpol_put(struct mempolicy *p)
>  {
>  }
>  
> -static inline void mpol_cond_put(struct mempolicy *pol)
> +static inline void mpol_get(struct mempolicy *pol)
>  {
>  }
>  
> -static inline void mpol_get(struct mempolicy *pol)
> +static inline bool mpol_tryget(struct mempolicy *pol)
> +{
> +}

This should return false, right?

[...]
> +/* Obtain a reference on the specified task mempolicy */

Again, this is pretty much clear from the name. It would be more useful
to explain how the pointer can be used - e.g. needs to call mpol_put
or mpol_kill depending on the calling context.

> +static mempolicy *get_task_mpol(struct task_struct *p)
> +{
> +	struct mempolicy *pol;
> +
> +	rcu_read_lock();
> +	pol = rcu_dereference(p->mempolicy);
> +
> +	if (!pol || mpol_tryget(pol))

Shouldn't be !mpol_tryget?

> +		pol = NULL;
> +	rcu_read_unlock();
> +
> +	return pol;
> +}
> +

I do not see any rcu_assign_pointer for the newly created policy so this
seems incomplete. Ditto no mpol_kill calls. I am unlikely to get into
follow up patches now. Please split up the work so that it is reviewable
more easily and then I can have a further look.

Thanks!
-- 
Michal Hocko
SUSE Labs