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

[Sedat Dilek <sedat.dilek@gmail.com>]

On Wed, Jul 3, 2013 at 6:34 PM, Michal Hocko <mhocko@suse.cz> wrote:
> On Wed 03-07-13 18:11:28, Sedat Dilek wrote:
>> On Wed, Jul 3, 2013 at 5:59 PM, Michal Hocko <mhocko@suse.cz> wrote:
>> > On Wed 03-07-13 17:53:21, Sedat Dilek wrote:
>> >> On Wed, Jul 3, 2013 at 5:20 PM, Michal Hocko <mhocko@suse.cz> wrote:
>> >> > On Wed 03-07-13 20:51:00, Li Zefan wrote:
>> >> > [...]
>> >> >> [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 <fengguang.wu@intel.com>
>> >> >> Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
>> >> >> Signed-off-by: Li Zefan <lizefan@huawei.com>
>> >> >
>> >> > 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 :-).

>> 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 -

>> 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
>>
>> > [...]
>> > --
>> > 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>