Re: [PATCH 2/3] mm/kmemleak: fix partially freeing unknown object warning

[Catalin Marinas]

On Wed, Sep 27, 2023 at 11:59:22AM +0800, Liu Shixin wrote:
> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> index 54c2c90d3abc..5a2bbd85df57 100644
> --- a/mm/kmemleak.c
> +++ b/mm/kmemleak.c
> @@ -208,6 +208,8 @@ static struct rb_root object_tree_root = RB_ROOT;
>  static struct rb_root object_phys_tree_root = RB_ROOT;
>  /* protecting the access to object_list, object_tree_root (or object_phys_tree_root) */
>  static DEFINE_RAW_SPINLOCK(kmemleak_lock);
> +/* Serial delete_object_part() to ensure all objects is deleted correctly */
> +static DEFINE_RAW_SPINLOCK(delete_object_part_mutex);

Don't call this mutex, it implies sleeping.

>  
>  /* allocation caches for kmemleak internal data */
>  static struct kmem_cache *object_cache;
> @@ -784,13 +786,16 @@ static void delete_object_part(unsigned long ptr, size_t size, bool is_phys)
>  {
>  	struct kmemleak_object *object;
>  	unsigned long start, end;
> +	unsigned long flags;
>  
> +	raw_spin_lock_irqsave(&delete_object_part_mutex, flags);
>  	object = find_and_remove_object(ptr, 1, is_phys);
>  	if (!object) {
>  #ifdef DEBUG
>  		kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
>  			      ptr, size);
>  #endif
> +		raw_spin_unlock_irqrestore(&delete_object_part_mutex, flags);

I prefer a goto out and a single place for unlocking.

However, we already take the kmemleak_lock in find_and_remove_object().
So better to open-code that function here and avoid introducing a new
lock. __create_object() may need a new bool argument, no_lock or
something. Or just split it into separate functions for allocating the
kmemleak structure and adding it to the corresponding trees/lists under
a lock.

-- 
Catalin