Re: [PATCH v5] mm: Fix possible deadlock in kmemleak

[Waiman Long <llong@redhat.com>]

On 8/22/25 3:35 AM, Gu Bowen wrote:
> There are some AA deadlock issues in kmemleak, similar to the situation
> reported by Breno [1]. The deadlock path is as follows:
>
> mem_pool_alloc()
>    -> raw_spin_lock_irqsave(&kmemleak_lock, flags);
>        -> pr_warn()
>            -> netconsole subsystem
> 	     -> netpoll
> 	         -> __alloc_skb
> 		   -> __create_object
> 		     -> raw_spin_lock_irqsave(&kmemleak_lock, flags);
>
> To solve this problem, switch to printk_safe mode before printing warning
> message, this will redirect all printk()-s to a special per-CPU buffer,
> which will be flushed later from a safe context (irq work), and this
> deadlock problem can be avoided. The proper API to use should be
> printk_deferred_enter()/printk_deferred_exit() [2]. Another way is to
> place the warn print after kmemleak is released.
>
> [1]
> https://lore.kernel.org/all/20250731-kmemleak_lock-v1-1-728fd470198f@debian.org/#t
> [2]
> https://lore.kernel.org/all/5ca375cd-4a20-4807-b897-68b289626550@redhat.com/
> ====================
>
> Signed-off-by: Gu Bowen <gubowen5@huawei.com>
> ---
>   mm/kmemleak.c | 27 ++++++++++++++++++++-------
>   1 file changed, 20 insertions(+), 7 deletions(-)
>
> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> index 84265983f239..1ac56ceb29b6 100644
> --- a/mm/kmemleak.c
> +++ b/mm/kmemleak.c
> @@ -437,9 +437,15 @@ static struct kmemleak_object *__lookup_object(unsigned long ptr, int alias,
>   		else if (untagged_objp == untagged_ptr || alias)
>   			return object;
>   		else {
> +			/*
> +			 * Printk deferring due to the kmemleak_lock held.
> +			 * This is done to avoid deadlock.
> +			 */
> +			printk_deferred_enter();
>   			kmemleak_warn("Found object by alias at 0x%08lx\n",
>   				      ptr);
>   			dump_object_info(object);
> +			printk_deferred_exit();
>   			break;
>   		}
>   	}
> @@ -736,6 +742,11 @@ static int __link_object(struct kmemleak_object *object, unsigned long ptr,
>   		else if (untagged_objp + parent->size <= untagged_ptr)
>   			link = &parent->rb_node.rb_right;
>   		else {
> +			/*
> +			 * Printk deferring due to the kmemleak_lock held.
> +			 * This is done to avoid deadlock.
> +			 */
> +			printk_deferred_enter();
>   			kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
>   				      ptr);
>   			/*
> @@ -743,6 +754,7 @@ static int __link_object(struct kmemleak_object *object, unsigned long ptr,
>   			 * be freed while the kmemleak_lock is held.
>   			 */
>   			dump_object_info(parent);
> +			printk_deferred_exit();
>   			return -EEXIST;
>   		}
>   	}
> @@ -856,13 +868,8 @@ static void delete_object_part(unsigned long ptr, size_t size,
>   
>   	raw_spin_lock_irqsave(&kmemleak_lock, flags);
>   	object = __find_and_remove_object(ptr, 1, objflags);
> -	if (!object) {
> -#ifdef DEBUG
> -		kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
> -			      ptr, size);
> -#endif
> +	if (!object)
>   		goto unlock;
> -	}
>   
>   	/*
>   	 * Create one or two objects that may result from the memory block
> @@ -882,8 +889,14 @@ static void delete_object_part(unsigned long ptr, size_t size,
>   
>   unlock:
>   	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
> -	if (object)
> +	if (object) {
>   		__delete_object(object);
> +	} else {
> +#ifdef DEBUG
> +		kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
> +			      ptr, size);
> +#endif
> +	}
>   
>   out:
>   	if (object_l)
Reviewed-by: Waiman Long <longman@redhat.com>