Re: [PATCH] memcg: clean up migration

[Minchan Kim <minchan.kim@gmail.com>]

Hi, Daisuke

On Mon, Feb 28, 2011 at 11:18 AM, Daisuke Nishimura
<nishimura@mxp.nes.nec.co.jp> wrote:
> On Mon, 28 Feb 2011 00:49:25 +0900
> Minchan Kim <minchan.kim@gmail.com> wrote:
>
>> This patch cleans up unncessary BUG_ON check and confusing
>> charge variable.
>>
>> That's because memcg charge/uncharge could be handled by
>> mem_cgroup_[prepare/end] migration itself so charge local variable
>> in unmap_and_move lost the role since we introduced 01b1ae63c2.
>>
>> And mem_cgroup_prepare_migratio return 0 if only it is successful.
>> Otherwise, it jumps to unlock label to clean up so BUG_ON(charge)
>> isn;t meaningless.
>>
>> Cc: Johannes Weiner <hannes@cmpxchg.org>
>> Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
>> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
>> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
>> Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
>
> It looks good to me, but I have one minor comment.
>
>> ---
>>  mm/memcontrol.c |    1 +
>>  mm/migrate.c    |   14 ++++----------
>>  2 files changed, 5 insertions(+), 10 deletions(-)
>>
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index 2fc97fc..6832926 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -2872,6 +2872,7 @@ static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
>>  /*
>>   * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
>>   * page belongs to.
>> + * Return 0 if charge is successful. Otherwise return -errno.
>>   */
>>  int mem_cgroup_prepare_migration(struct page *page,
>>       struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask)
>> diff --git a/mm/migrate.c b/mm/migrate.c
>> index eb083a6..737c2e5 100644
>> --- a/mm/migrate.c
>> +++ b/mm/migrate.c
>> @@ -622,7 +622,6 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
>>       int *result = NULL;
>>       struct page *newpage = get_new_page(page, private, &result);
>>       int remap_swapcache = 1;
>> -     int charge = 0;
>>       struct mem_cgroup *mem;
>>       struct anon_vma *anon_vma = NULL;
>>
>> @@ -637,9 +636,7 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
>>               if (unlikely(split_huge_page(page)))
>>                       goto move_newpage;
>>
>> -     /* prepare cgroup just returns 0 or -ENOMEM */
>>       rc = -EAGAIN;
>> -
>>       if (!trylock_page(page)) {
>>               if (!force)
>>                       goto move_newpage;
>> @@ -678,13 +675,11 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
>>       }
>>
>>       /* charge against new page */
>> -     charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
>> -     if (charge == -ENOMEM) {
>> -             rc = -ENOMEM;
>> +     rc = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
>> +     if (rc)
>>               goto unlock;
>> -     }
>> -     BUG_ON(charge);
>>
>> +     rc = -EAGAIN;
>>       if (PageWriteback(page)) {
>>               if (!force || !sync)
>>                       goto uncharge;
> How about
>
>        if (mem_cgroup_prepare_migration(..)) {
>                rc = -ENOMEM;
>                goto unlock;
>        }
>
> ?
>
> Re-setting "rc" to -EAGAIN is not necessary in this case.
> "if (mem_cgroup_...)" is commonly used in many places.
>
It works now but Johannes doesn't like it and me, either.
It makes unnecessary dependency which mem_cgroup_preparre_migration
can't propagate error to migrate_pages.
Although we don't need it, I want to remove such unnecessary dependency.

> Anyway,
>
>        Acked-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
>
>

Thanks!.

-- 
Kind regards,
Minchan Kim

--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>