Re: [PATCH] memcg: clean up migration

[Minchan Kim]

Hi, Daisuke

On Mon, Feb 28, 2011 at 11:18 AM, Daisuke Nishimura
<nishimura@xxxxxxxxxxxxxxxxx> wrote:
> On Mon, 28 Feb 2011 00:49:25 +0900
> Minchan Kim <minchan.kim@xxxxxxxxx> 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@xxxxxxxxxxx>
>> Cc: Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx>
>> Cc: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>
>> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
>> Signed-off-by: Minchan Kim <minchan.kim@xxxxxxxxx>
>
> It looks good to me, but I have one minor comment.
>
>> ---
>> Âmm/memcontrol.c | Â Â1 +
>> Âmm/migrate.c  Â|  14 ++++----------
>> Â2 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,
>> Â/*
>> Â * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
>> Â * page belongs to.
>> + * Return 0 if charge is successful. Otherwise return -errno.
>> Â */
>> Âint mem_cgroup_prepare_migration(struct page *page,
>> Â Â Â 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,
>> Â Â Â int *result = NULL;
>> Â Â Â struct page *newpage = get_new_page(page, private, &result);
>> Â Â Â int remap_swapcache = 1;
>> - Â Â int charge = 0;
>> Â Â Â struct mem_cgroup *mem;
>> Â Â Â struct anon_vma *anon_vma = NULL;
>>
>> @@ -637,9 +636,7 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
>> Â Â Â Â Â Â Â if (unlikely(split_huge_page(page)))
>> Â Â Â Â Â Â Â Â Â Â Â goto move_newpage;
>>
>> - Â Â /* prepare cgroup just returns 0 or -ENOMEM */
>> Â Â Â rc = -EAGAIN;
>> -
>> Â Â Â if (!trylock_page(page)) {
>> Â Â Â Â Â Â Â if (!force)
>> Â Â Â Â Â Â Â Â Â Â Â goto move_newpage;
>> @@ -678,13 +675,11 @@ static int unmap_and_move(new_page_t get_new_page, unsigned long private,
>> Â Â Â }
>>
>> Â Â Â /* charge against new page */
>> - Â Â charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
>> - Â Â if (charge == -ENOMEM) {
>> - Â Â Â Â Â Â rc = -ENOMEM;
>> + Â Â rc = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
>> + Â Â if (rc)
>> Â Â Â Â Â Â Â goto unlock;
>> - Â Â }
>> - Â Â BUG_ON(charge);
>>
>> + Â Â rc = -EAGAIN;
>> Â Â Â if (PageWriteback(page)) {
>> Â Â Â Â Â Â Â if (!force || !sync)
>> Â Â Â Â Â Â Â Â Â Â Â goto uncharge;
> How about
>
> Â Â Â Âif (mem_cgroup_prepare_migration(..)) {
> Â Â Â Â Â Â Â Ârc = -ENOMEM;
> Â Â Â Â Â Â Â Âgoto unlock;
> Â Â Â Â}
>
> ?
>
> 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,
>
> Â Â Â ÂAcked-by: Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx>
>
>

Thanks!.

-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/