2018-11-26 16:56:56

by Qian Cai

Subject: [PATCH] debugobjects: avoid recursive calls with kmemleak

CONFIG_DEBUG_OBJECTS_RCU_HEAD does not play well with kmemleak due to
recursive calls.

fill_pool
kmemleak_ignore
make_black_object
put_object
__call_rcu (kernel/rcu/tree.c)
debug_rcu_head_queue
debug_object_activate
debug_object_init
fill_pool
kmemleak_ignore
make_black_object
...

Hence, adding SLAB_NOLEAKTRACE to kmem_cache_create() to not register a
newly allocated debug objects at all.

Suggested-by: Catalin Marinas <[email protected]>
Signed-off-by: Qian Cai <[email protected]>
---
lib/debugobjects.c | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/lib/debugobjects.c b/lib/debugobjects.c
index eab7727b46ed..55437fd5128b 100644
--- a/lib/debugobjects.c
+++ b/lib/debugobjects.c
@@ -135,7 +135,6 @@ static void fill_pool(void)
if (!new)
return;

- kmemleak_ignore(new);
raw_spin_lock_irqsave(&pool_lock, flags);
hlist_add_head(&new->node, &obj_pool);
debug_objects_allocated++;
@@ -1128,7 +1127,6 @@ static int __init debug_objects_replace_static_objects(void)
obj = kmem_cache_zalloc(obj_cache, GFP_KERNEL);
if (!obj)
goto free;
- kmemleak_ignore(obj);
hlist_add_head(&obj->node, &objects);
}

@@ -1182,7 +1180,8 @@ void __init debug_objects_mem_init(void)

obj_cache = kmem_cache_create("debug_objects_cache",
sizeof (struct debug_obj), 0,
- SLAB_DEBUG_OBJECTS, NULL);
+ SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE,
+ NULL);

if (!obj_cache || debug_objects_replace_static_objects()) {
debug_objects_enabled = 0;
--
2.17.2 (Apple Git-113)

2018-11-26 17:14:30

by Waiman Long

Subject: Re: [PATCH] debugobjects: avoid recursive calls with kmemleak

On 11/26/2018 11:53 AM, Qian Cai wrote:
> CONFIG_DEBUG_OBJECTS_RCU_HEAD does not play well with kmemleak due to
> recursive calls.
>
> fill_pool
> kmemleak_ignore
> make_black_object
> put_object
> __call_rcu (kernel/rcu/tree.c)
> debug_rcu_head_queue
> debug_object_activate
> debug_object_init
> fill_pool
> kmemleak_ignore
> make_black_object
> ...
>
> Hence, adding SLAB_NOLEAKTRACE to kmem_cache_create() to not register a
> newly allocated debug objects at all.
>
> Suggested-by: Catalin Marinas <[email protected]>
> Signed-off-by: Qian Cai <[email protected]>
> ---
> lib/debugobjects.c | 5 ++---
> 1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/lib/debugobjects.c b/lib/debugobjects.c
> index eab7727b46ed..55437fd5128b 100644
> --- a/lib/debugobjects.c
> +++ b/lib/debugobjects.c
> @@ -135,7 +135,6 @@ static void fill_pool(void)
> if (!new)
> return;
>
> - kmemleak_ignore(new);
> raw_spin_lock_irqsave(&pool_lock, flags);
> hlist_add_head(&new->node, &obj_pool);
> debug_objects_allocated++;
> @@ -1128,7 +1127,6 @@ static int __init debug_objects_replace_static_objects(void)
> obj = kmem_cache_zalloc(obj_cache, GFP_KERNEL);
> if (!obj)
> goto free;
> - kmemleak_ignore(obj);
> hlist_add_head(&obj->node, &objects);
> }
>
> @@ -1182,7 +1180,8 @@ void __init debug_objects_mem_init(void)
>
> obj_cache = kmem_cache_create("debug_objects_cache",
> sizeof (struct debug_obj), 0,
> - SLAB_DEBUG_OBJECTS, NULL);
> + SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE,
> + NULL);
>
> if (!obj_cache || debug_objects_replace_static_objects()) {
> debug_objects_enabled = 0;

Acked-by: Waiman Long <[email protected]>

2018-11-26 17:25:15

by Catalin Marinas

Subject: Re: [PATCH] debugobjects: avoid recursive calls with kmemleak

On Mon, Nov 26, 2018 at 11:53:43AM -0500, Qian Cai wrote:
> CONFIG_DEBUG_OBJECTS_RCU_HEAD does not play well with kmemleak due to
> recursive calls.
>
> fill_pool
> kmemleak_ignore
> make_black_object
> put_object
> __call_rcu (kernel/rcu/tree.c)
> debug_rcu_head_queue
> debug_object_activate
> debug_object_init
> fill_pool
> kmemleak_ignore
> make_black_object
> ...
>
> Hence, adding SLAB_NOLEAKTRACE to kmem_cache_create() to not register a
> newly allocated debug objects at all.
>
> Suggested-by: Catalin Marinas <[email protected]>
> Signed-off-by: Qian Cai <[email protected]>

Acked-by: Catalin Marinas <[email protected]>