Re: [PATCH 1/2] ftrace: Update ftrace_ops->next pointer with rcu_assign_pointer()

[Steven Rostedt <rostedt@goodmis.org>]

On Wed, 17 Mar 2021 18:25:28 +0800
Li Huafei <lihuafei1@huawei.com> wrote:

> The unregistered ftrace_ops may be freed by the caller, so we should use
> rcu_assign_pointer() in remove_ftrace_ops() to remove the ftrace_ops,
> which ensures that no more users will reference the ftrace_ops after
> synchronize_rcu() is called.

The patch is fine, but we don't use synchronize_rcu() to protect the
list. We use synchronize_rcu_tasks().

-- Steve


> 
> Signed-off-by: Li Huafei <lihuafei1@huawei.com>
> ---
>  kernel/trace/ftrace.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> index 4d8e35575549..2e315a145d20 100644
> --- a/kernel/trace/ftrace.c
> +++ b/kernel/trace/ftrace.c
> @@ -282,7 +282,7 @@ static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
>  			lockdep_is_held(&ftrace_lock)) == ops &&
>  	    rcu_dereference_protected(ops->next,
>  			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
> -		*list = &ftrace_list_end;
> +		rcu_assign_pointer(*list, &ftrace_list_end);
>  		return 0;
>  	}
>  
> @@ -293,7 +293,7 @@ static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
>  	if (*p != ops)
>  		return -1;
>  
> -	*p = (*p)->next;
> +	rcu_assign_pointer(*p, (*p)->next);
>  	return 0;
>  }
>