On Mon, May 13, 2013 at 8:25 PM, Michal Hocko <mhocko@xxxxxxx> wrote:
> On Mon 13-05-13 13:05:24, Sha Zhengju wrote:
>> From: Sha Zhengju <handai.szj@xxxxxxxxxx>
>>
>> Change the first argument of mem_cgroup_{update,inc,dec}_page_stat() from
>> 'struct page *' to 'struct mem_cgroup *', and so move PageCgroupUsed(pc)
>> checking out of mem_cgroup_update_page_stat(). This is a prepare patch for
>> the following memcg page stat lock simplifying.
>
> No, please do not do this because it just spreads memcg specific code
> out of memcontrol.c. Besides that the patch is not correct.
> [...]
>> --- a/mm/rmap.c
>> +++ b/mm/rmap.c
>> @@ -1109,12 +1109,24 @@ void page_add_file_rmap(struct page *page)
>> {
>> bool locked;
>> unsigned long flags;
>> + struct page_cgroup *pc;
>> + struct mem_cgroup *memcg = NULL;
>>
>> mem_cgroup_begin_update_page_stat(page, &locked, &flags);
>> + pc = lookup_page_cgroup(page);
>> +
>> + rcu_read_lock();
>> + memcg = pc->mem_cgroup;
>
> a) unnecessary RCU take for memcg disabled and b) worse KABOOM in that case
> as page_cgroup is NULL. We really do not want to put
> mem_cgroup_disabled() tests all over the place. The idea behind
> mem_cgroup_begin_update_page_stat was to be almost a noop for !memcg
> (and the real noop for !CONFIG_MEMCG).
It's indeed an unwise behavior. How about also wrapping it in
mm/memcontrol.c or memcontrol.h?
>
> so Nak to this approach
--
Thanks,
Sha
--
To unsubscribe from this list: send the line "unsubscribe cgroups" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
