Re: [PATCH v2] mm: memcontrol: remove obsolete comment of mem_cgroup_unmark_under_oom()
On Wed 30-09-20 01:34:25, linmiaohe wrote:
> Michal Hocko wrote:
> > On Thu 17-09-20 06:59:00, Miaohe Lin wrote:
> >> Since commit 79dfdaccd1d5 ("memcg: make oom_lock 0 and 1 based rather
> >> than counter"), the mem_cgroup_unmark_under_oom() is added and the
> >> comment of the mem_cgroup_oom_unlock() is moved here. But this comment
> >> make no sense here because mem_cgroup_oom_lock() does not operate on
> >> under_oom field.
> >
> >OK, so I've looked into this more deeply and I finally remember why we have
> >this comment here. The point is that under_oom shouldn't underflow and that
> >we have to explicitly check for > 0 because a new child memcg could have
> >been added between mem_cgroup_mark_under_oom and mem_cgroup_unmark_under_oom.
> >
> >So the comment makes sense although it is not as helpful as it could be.
> >I think that changing it to the following will be more usefule
> >
> > /*
> > * Be careful about under_oom underflows becase a child memcg
> > * could have neem added after mem_cgroup_mark_under_oom
>
> Should it be s/neem/been/ ?
yep, fat fingers...
>
> > */
>
> Many thanks for detailed explanation. Will fix it in v2. Thanks again.
--
Michal Hocko
SUSE Labs
Re: [PATCH v2] mm: memcontrol: remove obsolete comment of mem_cgroup_unmark_under_oom()
Michal Hocko wrote:
> On Thu 17-09-20 06:59:00, Miaohe Lin wrote:
>> Since commit 79dfdaccd1d5 ("memcg: make oom_lock 0 and 1 based rather
>> than counter"), the mem_cgroup_unmark_under_oom() is added and the
>> comment of the mem_cgroup_oom_unlock() is moved here. But this comment
>> make no sense here because mem_cgroup_oom_lock() does not operate on
>> under_oom field.
>
>OK, so I've looked into this more deeply and I finally remember why we have
>this comment here. The point is that under_oom shouldn't underflow and that we
>have to explicitly check for > 0 because a new child memcg could have been
>added between mem_cgroup_mark_under_oom and mem_cgroup_unmark_under_oom.
>
>So the comment makes sense although it is not as helpful as it could be.
>I think that changing it to the following will be more usefule
>
> /*
>* Be careful about under_oom underflows becase a child memcg
>* could have neem added after mem_cgroup_mark_under_oom
Should it be s/neem/been/ ?
>*/
Many thanks for detailed explanation. Will fix it in v2. Thanks again.
Re: [PATCH v2] mm: memcontrol: remove obsolete comment of mem_cgroup_unmark_under_oom()
On Thu 17-09-20 06:59:00, Miaohe Lin wrote:
> Since commit 79dfdaccd1d5 ("memcg: make oom_lock 0 and 1 based rather than
> counter"), the mem_cgroup_unmark_under_oom() is added and the comment of
> the mem_cgroup_oom_unlock() is moved here. But this comment make no sense
> here because mem_cgroup_oom_lock() does not operate on under_oom field.
OK, so I've looked into this more deeply and I finally remember why we
have this comment here. The point is that under_oom shouldn't underflow
and that we have to explicitly check for > 0 because a new child memcg
could have been added between mem_cgroup_mark_under_oom and
mem_cgroup_unmark_under_oom.
So the comment makes sense although it is not as helpful as it could be.
I think that changing it to the following will be more usefule
/*
* Be careful about under_oom underflows becase a child memcg
* could have neem added after mem_cgroup_mark_under_oom
*/
>
> Signed-off-by: Miaohe Lin
> ---
> mm/memcontrol.c | 4
> 1 file changed, 4 deletions(-)
>
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index cd5f83de9a6f..e44f5afaf78b 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -1848,10 +1848,6 @@ static void mem_cgroup_unmark_under_oom(struct
> mem_cgroup *memcg)
> {
> struct mem_cgroup *iter;
>
> - /*
> - * When a new child is created while the hierarchy is under oom,
> - * mem_cgroup_oom_lock() may not be called. Watch for underflow.
> - */
> spin_lock(_oom_lock);
> for_each_mem_cgroup_tree(iter, memcg)
> if (iter->under_oom > 0)
> --
> 2.19.1
--
Michal Hocko
SUSE Labs
[PATCH v2] mm: memcontrol: remove obsolete comment of mem_cgroup_unmark_under_oom()
Since commit 79dfdaccd1d5 ("memcg: make oom_lock 0 and 1 based rather than
counter"), the mem_cgroup_unmark_under_oom() is added and the comment of
the mem_cgroup_oom_unlock() is moved here. But this comment make no sense
here because mem_cgroup_oom_lock() does not operate on under_oom field.
Signed-off-by: Miaohe Lin
---
mm/memcontrol.c | 4
1 file changed, 4 deletions(-)
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index cd5f83de9a6f..e44f5afaf78b 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -1848,10 +1848,6 @@ static void mem_cgroup_unmark_under_oom(struct
mem_cgroup *memcg)
{
struct mem_cgroup *iter;
- /*
-* When a new child is created while the hierarchy is under oom,
-* mem_cgroup_oom_lock() may not be called. Watch for underflow.
-*/
spin_lock(_oom_lock);
for_each_mem_cgroup_tree(iter, memcg)
if (iter->under_oom > 0)
--
2.19.1
