Re: [PATCH next-20130703] net: sock: Add ifdef CONFIG_MEMCG_KMEM for mem_cgroup_sockets_{init,destroy}

On Wed, Jul 3, 2013 at 6:34 PM, Michal Hocko [off-list ref] wrote:

On Wed 03-07-13 18:11:28, Sedat Dilek wrote:

quoted

On Wed, Jul 3, 2013 at 5:59 PM, Michal Hocko [off-list ref] wrote:

quoted

On Wed 03-07-13 17:53:21, Sedat Dilek wrote:

quoted

On Wed, Jul 3, 2013 at 5:20 PM, Michal Hocko [off-list ref] wrote:

quoted

On Wed 03-07-13 20:51:00, Li Zefan wrote:
[...]

quoted

[PATCH] memcg: fix build error if CONFIG_MEMCG_KMEM=n

Fix this build error:

mm/built-in.o: In function `mem_cgroup_css_free':
memcontrol.c:(.text+0x5caa6): undefined reference to
'mem_cgroup_sockets_destroy'

Reported-by: Fengguang Wu <redacted>
Reported-by: Stephen Rothwell <redacted>
Signed-off-by: Li Zefan <redacted>

I am seeing the same thing I just didn't get to reporting it.
The other approach is not bad as well but I find this tiny better
because mem_cgroup_css_free should care only about a single cleanup
function for whole kmem. If that one needs to do tcp kmem specific
cleanup then it should be done inside kmem_cgroup_css_offline.

As said in my other mail, for me this makes sense as it is a followup.

But, still I don't know why sock.c has is own mem_cgroup_sockets_{init,destroy}.

That is the only definition AFAICS (except for !CONFIG_NET where it
expands to NOOP). Please note that memcg_init_kmem is a common kmem
initializator and it needs to be prepared for !CONFIG_NET.

The same applies to _destroy.
Makes more sense now?

So, that stuff comes originally from the net-tree.

No, it all came from tcp kmem accounting. It is a memcg thingy and I
guess it was placed into sock.c because it depends on some static
symbols there (e.g. proto_list_mutex).

memcg thingies should belong to mm-tree :-).

quoted

I understand the !CONFIG_NET case, but lack the understanding why
memcontrol.c needs _destroy.

Because it is memcg specific and it has to be called when a group is
destroyed.

I looked again into my local GIT tree where I applied Li Zefan's patch.
memcg_destroy_kmem() makes sense with existing memcg_init_kmem().

I have now a better picture, but it's balck/white not coloured :-).

Thanks for your patience and explanations.

- Sedat -

quoted

Can you explain that (sorry /me is no mm-geek)?

- Sedat -

[1] http://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/tree/net/core/sock.c?id=next-20130703#n147
[2] http://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/tree/include/net/sock.h?id=next-20130703#n73

quoted

[...]
--
Michal Hocko
SUSE Labs

--
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>