diff mbox series

[RFC,bpf-next,03/10] mm, memcg: Add new helper obj_cgroup_from_current()

Message ID 20220619155032.32515-4-laoar.shao@gmail.com (mailing list archive)
State New
Headers show
Series bpf, mm: Recharge pages when reuse bpf map | expand

Commit Message

Yafang Shao June 19, 2022, 3:50 p.m. UTC
The difference between get_obj_cgroup_from_current() and obj_cgroup_from_current()
is that the later one doesn't add objcg's refcnt.

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
---
 include/linux/memcontrol.h |  1 +
 mm/memcontrol.c            | 24 ++++++++++++++++++++++++
 2 files changed, 25 insertions(+)

Comments

Roman Gushchin June 23, 2022, 3:01 a.m. UTC | #1
On Sun, Jun 19, 2022 at 03:50:25PM +0000, Yafang Shao wrote:
> The difference between get_obj_cgroup_from_current() and obj_cgroup_from_current()
> is that the later one doesn't add objcg's refcnt.
> 
> Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
> ---
>  include/linux/memcontrol.h |  1 +
>  mm/memcontrol.c            | 24 ++++++++++++++++++++++++
>  2 files changed, 25 insertions(+)
> 
> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> index cf074156c6ac..402b42670bcd 100644
> --- a/include/linux/memcontrol.h
> +++ b/include/linux/memcontrol.h
> @@ -1703,6 +1703,7 @@ bool mem_cgroup_kmem_disabled(void);
>  int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
>  void __memcg_kmem_uncharge_page(struct page *page, int order);
>  
> +struct obj_cgroup *obj_cgroup_from_current(void);
>  struct obj_cgroup *get_obj_cgroup_from_current(void);
>  struct obj_cgroup *get_obj_cgroup_from_page(struct page *page);
>  
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index abec50f31fe6..350a7849dac3 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -2950,6 +2950,30 @@ struct obj_cgroup *get_obj_cgroup_from_page(struct page *page)
>  	return objcg;
>  }
>  
> +__always_inline struct obj_cgroup *obj_cgroup_from_current(void)
> +{
> +	struct obj_cgroup *objcg = NULL;
> +	struct mem_cgroup *memcg;
> +
> +	if (memcg_kmem_bypass())
> +		return NULL;
> +
> +	rcu_read_lock();
> +	if (unlikely(active_memcg()))
> +		memcg = active_memcg();
> +	else
> +		memcg = mem_cgroup_from_task(current);
> +
> +	for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
> +		objcg = rcu_dereference(memcg->objcg);
> +		if (objcg)
> +			break;
> +	}
> +	rcu_read_unlock();

Hm, what prevents the objcg from being released here? Under which conditions
it's safe to call it?
Yafang Shao June 25, 2022, 1:54 p.m. UTC | #2
On Thu, Jun 23, 2022 at 11:01 AM Roman Gushchin
<roman.gushchin@linux.dev> wrote:
>
> On Sun, Jun 19, 2022 at 03:50:25PM +0000, Yafang Shao wrote:
> > The difference between get_obj_cgroup_from_current() and obj_cgroup_from_current()
> > is that the later one doesn't add objcg's refcnt.
> >
> > Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
> > ---
> >  include/linux/memcontrol.h |  1 +
> >  mm/memcontrol.c            | 24 ++++++++++++++++++++++++
> >  2 files changed, 25 insertions(+)
> >
> > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > index cf074156c6ac..402b42670bcd 100644
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -1703,6 +1703,7 @@ bool mem_cgroup_kmem_disabled(void);
> >  int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
> >  void __memcg_kmem_uncharge_page(struct page *page, int order);
> >
> > +struct obj_cgroup *obj_cgroup_from_current(void);
> >  struct obj_cgroup *get_obj_cgroup_from_current(void);
> >  struct obj_cgroup *get_obj_cgroup_from_page(struct page *page);
> >
> > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > index abec50f31fe6..350a7849dac3 100644
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -2950,6 +2950,30 @@ struct obj_cgroup *get_obj_cgroup_from_page(struct page *page)
> >       return objcg;
> >  }
> >
> > +__always_inline struct obj_cgroup *obj_cgroup_from_current(void)
> > +{
> > +     struct obj_cgroup *objcg = NULL;
> > +     struct mem_cgroup *memcg;
> > +
> > +     if (memcg_kmem_bypass())
> > +             return NULL;
> > +
> > +     rcu_read_lock();
> > +     if (unlikely(active_memcg()))
> > +             memcg = active_memcg();
> > +     else
> > +             memcg = mem_cgroup_from_task(current);
> > +
> > +     for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
> > +             objcg = rcu_dereference(memcg->objcg);
> > +             if (objcg)
> > +                     break;
> > +     }
> > +     rcu_read_unlock();
>
> Hm, what prevents the objcg from being released here? Under which conditions
> it's safe to call it?

obj_cgroup_from_current() is used when we know the objcg's refcnt has
already been incremented.
For example in my case, it is called after we have already call get_
parent_mem_cgroup().
I should add a comment or a WARN_ON() in this function.
Roman Gushchin June 26, 2022, 1:52 a.m. UTC | #3
On Sat, Jun 25, 2022 at 09:54:17PM +0800, Yafang Shao wrote:
> On Thu, Jun 23, 2022 at 11:01 AM Roman Gushchin
> <roman.gushchin@linux.dev> wrote:
> >
> > On Sun, Jun 19, 2022 at 03:50:25PM +0000, Yafang Shao wrote:
> > > The difference between get_obj_cgroup_from_current() and obj_cgroup_from_current()
> > > is that the later one doesn't add objcg's refcnt.
> > >
> > > Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
> > > ---
> > >  include/linux/memcontrol.h |  1 +
> > >  mm/memcontrol.c            | 24 ++++++++++++++++++++++++
> > >  2 files changed, 25 insertions(+)
> > >
> > > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > > index cf074156c6ac..402b42670bcd 100644
> > > --- a/include/linux/memcontrol.h
> > > +++ b/include/linux/memcontrol.h
> > > @@ -1703,6 +1703,7 @@ bool mem_cgroup_kmem_disabled(void);
> > >  int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
> > >  void __memcg_kmem_uncharge_page(struct page *page, int order);
> > >
> > > +struct obj_cgroup *obj_cgroup_from_current(void);
> > >  struct obj_cgroup *get_obj_cgroup_from_current(void);
> > >  struct obj_cgroup *get_obj_cgroup_from_page(struct page *page);
> > >
> > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > > index abec50f31fe6..350a7849dac3 100644
> > > --- a/mm/memcontrol.c
> > > +++ b/mm/memcontrol.c
> > > @@ -2950,6 +2950,30 @@ struct obj_cgroup *get_obj_cgroup_from_page(struct page *page)
> > >       return objcg;
> > >  }
> > >
> > > +__always_inline struct obj_cgroup *obj_cgroup_from_current(void)
> > > +{
> > > +     struct obj_cgroup *objcg = NULL;
> > > +     struct mem_cgroup *memcg;
> > > +
> > > +     if (memcg_kmem_bypass())
> > > +             return NULL;
> > > +
> > > +     rcu_read_lock();
> > > +     if (unlikely(active_memcg()))
> > > +             memcg = active_memcg();
> > > +     else
> > > +             memcg = mem_cgroup_from_task(current);
> > > +
> > > +     for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
> > > +             objcg = rcu_dereference(memcg->objcg);
> > > +             if (objcg)
> > > +                     break;
> > > +     }
> > > +     rcu_read_unlock();
> >
> > Hm, what prevents the objcg from being released here? Under which conditions
> > it's safe to call it?
> 
> obj_cgroup_from_current() is used when we know the objcg's refcnt has
> already been incremented.
> For example in my case, it is called after we have already call get_
> parent_mem_cgroup().
> I should add a comment or a WARN_ON() in this function.

Yes, it's very confusing, please add a big comment explaining under which
conditions it's safe to call it.

Thanks!
diff mbox series

Patch

diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index cf074156c6ac..402b42670bcd 100644
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -1703,6 +1703,7 @@  bool mem_cgroup_kmem_disabled(void);
 int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order);
 void __memcg_kmem_uncharge_page(struct page *page, int order);
 
+struct obj_cgroup *obj_cgroup_from_current(void);
 struct obj_cgroup *get_obj_cgroup_from_current(void);
 struct obj_cgroup *get_obj_cgroup_from_page(struct page *page);
 
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index abec50f31fe6..350a7849dac3 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2950,6 +2950,30 @@  struct obj_cgroup *get_obj_cgroup_from_page(struct page *page)
 	return objcg;
 }
 
+__always_inline struct obj_cgroup *obj_cgroup_from_current(void)
+{
+	struct obj_cgroup *objcg = NULL;
+	struct mem_cgroup *memcg;
+
+	if (memcg_kmem_bypass())
+		return NULL;
+
+	rcu_read_lock();
+	if (unlikely(active_memcg()))
+		memcg = active_memcg();
+	else
+		memcg = mem_cgroup_from_task(current);
+
+	for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
+		objcg = rcu_dereference(memcg->objcg);
+		if (objcg)
+			break;
+	}
+	rcu_read_unlock();
+
+	return objcg;
+}
+
 static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages)
 {
 	mod_memcg_state(memcg, MEMCG_KMEM, nr_pages);