diff mbox series

[v2] mm: kmem: add lockdep assertion to obj_cgroup_memcg

Message ID 20240724095307.81264-1-songmuchun@bytedance.com (mailing list archive)
State New
Headers show
Series [v2] mm: kmem: add lockdep assertion to obj_cgroup_memcg | expand

Commit Message

Muchun Song July 24, 2024, 9:53 a.m. UTC
The obj_cgroup_memcg() is supposed to safe to prevent the returned
memory cgroup from being freed only when the caller is holding the
rcu read lock or objcg_lock or cgroup_mutex. It is very easy to
ignore thoes conditions when users call some upper APIs which call
obj_cgroup_memcg() internally like mem_cgroup_from_slab_obj() (See
the link below). So it is better to add lockdep assertion to
obj_cgroup_memcg() to find those issues ASAP.

Because there is no user of obj_cgroup_memcg() holding objcg_lock
to make the returned memory cgroup safe, do not add objcg_lock
assertion (We should export objcg_lock if we really want to do).
Additionally, this is some internal implementation detail of memcg
and should not be accessible outside memcg code.

Some users like __mem_cgroup_uncharge() do not care the lifetime
of the returned memory cgroup, which just want to know if the
folio is charged to a memory cgroup, therefore, they do not need
to hold the needed locks. In which case, introduce a new helper
folio_memcg_charged() to do this. Compare it to folio_memcg(), it
could eliminate a memory access of objcg->memcg for kmem, actually,
a really small gain.

Link: https://lore.kernel.org/all/20240718083607.42068-1-songmuchun@bytedance.com/
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
---
v2:
 - Remove mention of objcg_lock in obj_cgroup_memcg()(Shakeel Butt).

 include/linux/memcontrol.h | 20 +++++++++++++++++---
 mm/memcontrol.c            |  6 +++---
 2 files changed, 20 insertions(+), 6 deletions(-)

Comments

Vlastimil Babka (SUSE) July 24, 2024, 3:20 p.m. UTC | #1
On 7/24/24 11:53 AM, Muchun Song wrote:
> The obj_cgroup_memcg() is supposed to safe to prevent the returned
> memory cgroup from being freed only when the caller is holding the
> rcu read lock or objcg_lock or cgroup_mutex. It is very easy to
> ignore thoes conditions when users call some upper APIs which call
> obj_cgroup_memcg() internally like mem_cgroup_from_slab_obj() (See
> the link below). So it is better to add lockdep assertion to
> obj_cgroup_memcg() to find those issues ASAP.
> 
> Because there is no user of obj_cgroup_memcg() holding objcg_lock
> to make the returned memory cgroup safe, do not add objcg_lock
> assertion (We should export objcg_lock if we really want to do).
> Additionally, this is some internal implementation detail of memcg
> and should not be accessible outside memcg code.
> 
> Some users like __mem_cgroup_uncharge() do not care the lifetime
> of the returned memory cgroup, which just want to know if the
> folio is charged to a memory cgroup, therefore, they do not need
> to hold the needed locks. In which case, introduce a new helper
> folio_memcg_charged() to do this. Compare it to folio_memcg(), it
> could eliminate a memory access of objcg->memcg for kmem, actually,
> a really small gain.
> 
> Link: https://lore.kernel.org/all/20240718083607.42068-1-songmuchun@bytedance.com/
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> ---
> v2:
>  - Remove mention of objcg_lock in obj_cgroup_memcg()(Shakeel Butt).
> 
>  include/linux/memcontrol.h | 20 +++++++++++++++++---
>  mm/memcontrol.c            |  6 +++---
>  2 files changed, 20 insertions(+), 6 deletions(-)
> 
> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> index fc94879db4dff..742351945f683 100644
> --- a/include/linux/memcontrol.h
> +++ b/include/linux/memcontrol.h
> @@ -360,11 +360,11 @@ static inline bool folio_memcg_kmem(struct folio *folio);
>   * After the initialization objcg->memcg is always pointing at
>   * a valid memcg, but can be atomically swapped to the parent memcg.
>   *
> - * The caller must ensure that the returned memcg won't be released:
> - * e.g. acquire the rcu_read_lock or css_set_lock.
> + * The caller must ensure that the returned memcg won't be released.
>   */
>  static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
>  {
> +	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_is_held(&cgroup_mutex));

Maybe lockdep_assert_once() would be a better fit?

>  	return READ_ONCE(objcg->memcg);
>  }
>  
> @@ -438,6 +438,19 @@ static inline struct mem_cgroup *folio_memcg(struct folio *folio)
>  	return __folio_memcg(folio);
>  }
>  
> +/*
> + * folio_memcg_charged - If a folio is charged to a memory cgroup.
> + * @folio: Pointer to the folio.
> + *
> + * Returns true if folio is charged to a memory cgroup, otherwise returns false.
> + */
> +static inline bool folio_memcg_charged(struct folio *folio)
> +{
> +	if (folio_memcg_kmem(folio))
> +		return __folio_objcg(folio) != NULL;
> +	return __folio_memcg(folio) != NULL;
> +}
> +
>  /**
>   * folio_memcg_rcu - Locklessly get the memory cgroup associated with a folio.
>   * @folio: Pointer to the folio.
> @@ -454,7 +467,6 @@ static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio)
>  	unsigned long memcg_data = READ_ONCE(folio->memcg_data);
>  
>  	VM_BUG_ON_FOLIO(folio_test_slab(folio), folio);
> -	WARN_ON_ONCE(!rcu_read_lock_held());
>  
>  	if (memcg_data & MEMCG_DATA_KMEM) {
>  		struct obj_cgroup *objcg;
> @@ -463,6 +475,8 @@ static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio)
>  		return obj_cgroup_memcg(objcg);
>  	}
>  
> +	WARN_ON_ONCE(!rcu_read_lock_held());
> +
>  	return (struct mem_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK);
>  }
>  
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 622d4544edd24..3da0284573857 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -2366,7 +2366,7 @@ void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
>  
>  static void commit_charge(struct folio *folio, struct mem_cgroup *memcg)
>  {
> -	VM_BUG_ON_FOLIO(folio_memcg(folio), folio);
> +	VM_BUG_ON_FOLIO(folio_memcg_charged(folio), folio);
>  	/*
>  	 * Any of the following ensures page's memcg stability:
>  	 *
> @@ -4617,7 +4617,7 @@ void __mem_cgroup_uncharge(struct folio *folio)
>  	struct uncharge_gather ug;
>  
>  	/* Don't touch folio->lru of any random page, pre-check: */
> -	if (!folio_memcg(folio))
> +	if (!folio_memcg_charged(folio))
>  		return;
>  
>  	uncharge_gather_clear(&ug);
> @@ -4662,7 +4662,7 @@ void mem_cgroup_replace_folio(struct folio *old, struct folio *new)
>  		return;
>  
>  	/* Page cache replacement: new folio already charged? */
> -	if (folio_memcg(new))
> +	if (folio_memcg_charged(new))
>  		return;
>  
>  	memcg = folio_memcg(old);
Shakeel Butt July 24, 2024, 10:12 p.m. UTC | #2
On Wed, Jul 24, 2024 at 05:20:09PM GMT, Vlastimil Babka (SUSE) wrote:
> On 7/24/24 11:53 AM, Muchun Song wrote:
> > The obj_cgroup_memcg() is supposed to safe to prevent the returned
> > memory cgroup from being freed only when the caller is holding the
> > rcu read lock or objcg_lock or cgroup_mutex. It is very easy to
> > ignore thoes conditions when users call some upper APIs which call
> > obj_cgroup_memcg() internally like mem_cgroup_from_slab_obj() (See
> > the link below). So it is better to add lockdep assertion to
> > obj_cgroup_memcg() to find those issues ASAP.
> > 
> > Because there is no user of obj_cgroup_memcg() holding objcg_lock
> > to make the returned memory cgroup safe, do not add objcg_lock
> > assertion (We should export objcg_lock if we really want to do).
> > Additionally, this is some internal implementation detail of memcg
> > and should not be accessible outside memcg code.
> > 
> > Some users like __mem_cgroup_uncharge() do not care the lifetime
> > of the returned memory cgroup, which just want to know if the
> > folio is charged to a memory cgroup, therefore, they do not need
> > to hold the needed locks. In which case, introduce a new helper
> > folio_memcg_charged() to do this. Compare it to folio_memcg(), it
> > could eliminate a memory access of objcg->memcg for kmem, actually,
> > a really small gain.
> > 
> > Link: https://lore.kernel.org/all/20240718083607.42068-1-songmuchun@bytedance.com/
> > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > ---
> > v2:
> >  - Remove mention of objcg_lock in obj_cgroup_memcg()(Shakeel Butt).
> > 
> >  include/linux/memcontrol.h | 20 +++++++++++++++++---
> >  mm/memcontrol.c            |  6 +++---
> >  2 files changed, 20 insertions(+), 6 deletions(-)
> > 
> > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > index fc94879db4dff..742351945f683 100644
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -360,11 +360,11 @@ static inline bool folio_memcg_kmem(struct folio *folio);
> >   * After the initialization objcg->memcg is always pointing at
> >   * a valid memcg, but can be atomically swapped to the parent memcg.
> >   *
> > - * The caller must ensure that the returned memcg won't be released:
> > - * e.g. acquire the rcu_read_lock or css_set_lock.
> > + * The caller must ensure that the returned memcg won't be released.
> >   */
> >  static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
> >  {
> > +	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_is_held(&cgroup_mutex));
> 
> Maybe lockdep_assert_once() would be a better fit?
> 

So something like:
	lockdep_assert_once(rcu_read_lock_held() || lockdep_is_held(&cgroup_mutex));
Roman Gushchin July 24, 2024, 10:38 p.m. UTC | #3
On Wed, Jul 24, 2024 at 05:20:09PM +0200, Vlastimil Babka (SUSE) wrote:
> On 7/24/24 11:53 AM, Muchun Song wrote:
> > The obj_cgroup_memcg() is supposed to safe to prevent the returned
> > memory cgroup from being freed only when the caller is holding the
> > rcu read lock or objcg_lock or cgroup_mutex. It is very easy to
> > ignore thoes conditions when users call some upper APIs which call
> > obj_cgroup_memcg() internally like mem_cgroup_from_slab_obj() (See
> > the link below). So it is better to add lockdep assertion to
> > obj_cgroup_memcg() to find those issues ASAP.
> > 
> > Because there is no user of obj_cgroup_memcg() holding objcg_lock
> > to make the returned memory cgroup safe, do not add objcg_lock
> > assertion (We should export objcg_lock if we really want to do).
> > Additionally, this is some internal implementation detail of memcg
> > and should not be accessible outside memcg code.
> > 
> > Some users like __mem_cgroup_uncharge() do not care the lifetime
> > of the returned memory cgroup, which just want to know if the
> > folio is charged to a memory cgroup, therefore, they do not need
> > to hold the needed locks. In which case, introduce a new helper
> > folio_memcg_charged() to do this. Compare it to folio_memcg(), it
> > could eliminate a memory access of objcg->memcg for kmem, actually,
> > a really small gain.
> > 
> > Link: https://lore.kernel.org/all/20240718083607.42068-1-songmuchun@bytedance.com/
> > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > ---
> > v2:
> >  - Remove mention of objcg_lock in obj_cgroup_memcg()(Shakeel Butt).
> > 
> >  include/linux/memcontrol.h | 20 +++++++++++++++++---
> >  mm/memcontrol.c            |  6 +++---
> >  2 files changed, 20 insertions(+), 6 deletions(-)
> > 
> > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > index fc94879db4dff..742351945f683 100644
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -360,11 +360,11 @@ static inline bool folio_memcg_kmem(struct folio *folio);
> >   * After the initialization objcg->memcg is always pointing at
> >   * a valid memcg, but can be atomically swapped to the parent memcg.
> >   *
> > - * The caller must ensure that the returned memcg won't be released:
> > - * e.g. acquire the rcu_read_lock or css_set_lock.
> > + * The caller must ensure that the returned memcg won't be released.
> >   */
> >  static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
> >  {
> > +	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_is_held(&cgroup_mutex));
> 
> Maybe lockdep_assert_once() would be a better fit?

100%.
diff mbox series

Patch

diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index fc94879db4dff..742351945f683 100644
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -360,11 +360,11 @@  static inline bool folio_memcg_kmem(struct folio *folio);
  * After the initialization objcg->memcg is always pointing at
  * a valid memcg, but can be atomically swapped to the parent memcg.
  *
- * The caller must ensure that the returned memcg won't be released:
- * e.g. acquire the rcu_read_lock or css_set_lock.
+ * The caller must ensure that the returned memcg won't be released.
  */
 static inline struct mem_cgroup *obj_cgroup_memcg(struct obj_cgroup *objcg)
 {
+	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_is_held(&cgroup_mutex));
 	return READ_ONCE(objcg->memcg);
 }
 
@@ -438,6 +438,19 @@  static inline struct mem_cgroup *folio_memcg(struct folio *folio)
 	return __folio_memcg(folio);
 }
 
+/*
+ * folio_memcg_charged - If a folio is charged to a memory cgroup.
+ * @folio: Pointer to the folio.
+ *
+ * Returns true if folio is charged to a memory cgroup, otherwise returns false.
+ */
+static inline bool folio_memcg_charged(struct folio *folio)
+{
+	if (folio_memcg_kmem(folio))
+		return __folio_objcg(folio) != NULL;
+	return __folio_memcg(folio) != NULL;
+}
+
 /**
  * folio_memcg_rcu - Locklessly get the memory cgroup associated with a folio.
  * @folio: Pointer to the folio.
@@ -454,7 +467,6 @@  static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio)
 	unsigned long memcg_data = READ_ONCE(folio->memcg_data);
 
 	VM_BUG_ON_FOLIO(folio_test_slab(folio), folio);
-	WARN_ON_ONCE(!rcu_read_lock_held());
 
 	if (memcg_data & MEMCG_DATA_KMEM) {
 		struct obj_cgroup *objcg;
@@ -463,6 +475,8 @@  static inline struct mem_cgroup *folio_memcg_rcu(struct folio *folio)
 		return obj_cgroup_memcg(objcg);
 	}
 
+	WARN_ON_ONCE(!rcu_read_lock_held());
+
 	return (struct mem_cgroup *)(memcg_data & ~OBJEXTS_FLAGS_MASK);
 }
 
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 622d4544edd24..3da0284573857 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2366,7 +2366,7 @@  void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
 
 static void commit_charge(struct folio *folio, struct mem_cgroup *memcg)
 {
-	VM_BUG_ON_FOLIO(folio_memcg(folio), folio);
+	VM_BUG_ON_FOLIO(folio_memcg_charged(folio), folio);
 	/*
 	 * Any of the following ensures page's memcg stability:
 	 *
@@ -4617,7 +4617,7 @@  void __mem_cgroup_uncharge(struct folio *folio)
 	struct uncharge_gather ug;
 
 	/* Don't touch folio->lru of any random page, pre-check: */
-	if (!folio_memcg(folio))
+	if (!folio_memcg_charged(folio))
 		return;
 
 	uncharge_gather_clear(&ug);
@@ -4662,7 +4662,7 @@  void mem_cgroup_replace_folio(struct folio *old, struct folio *new)
 		return;
 
 	/* Page cache replacement: new folio already charged? */
-	if (folio_memcg(new))
+	if (folio_memcg_charged(new))
 		return;
 
 	memcg = folio_memcg(old);