Re: [PATCH 1/5 v2] memcg: replace unsigned long by u64 to avoid overflow

[Glauber Costa <glommer-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org>]

On 07/04/2012 05:16 PM, Johannes Weiner wrote:
> On Wed, Jul 04, 2012 at 04:29:48PM +0400, Glauber Costa wrote:
>> On 07/04/2012 04:24 PM, Wanpeng Li wrote:
>>> On Mon, Jun 25, 2012 at 05:30:11PM +0900, Kamezawa Hiroyuki wrote:
>>>> (2012/06/25 16:52), Johannes Weiner wrote:
>>>>> On Mon, Jun 25, 2012 at 02:04:20PM +0800, Wanpeng Li wrote:
>>>>>> Changlog:
>>>>>>
>>>>>> V2 -> V1:
>>>>>> * fix zone_page_state()::/include/linux/vmstat.h returns 'unsigned long'
>>>>>>
>>>>>> From: Wanpeng Li <liwp-23VcF4HTsmIX0ybBhKVfKdBPR1lH4CV8@public.gmane.org>
>>>>>>
>>>>>> Since the return value variable in mem_cgroup_zone_nr_lru_pages and
>>>>>> mem_cgroup_node_nr_lru_pages functions are u64, so replace the return
>>>>>> value of funtions by u64 to avoid overflow.
>>>>>>
>>>>>> Signed-off-by: Wanpeng Li <liwp.linux-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>>>> ---
>>>>>>  include/linux/vmstat.h |    2 +-
>>>>>>  mm/memcontrol.c        |    5 ++---
>>>>>>  2 files changed, 3 insertions(+), 4 deletions(-)
>>>>>>
>>>>>> diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
>>>>>> index 65efb92..6a14291 100644
>>>>>> --- a/include/linux/vmstat.h
>>>>>> +++ b/include/linux/vmstat.h
>>>>>> @@ -106,7 +106,7 @@ static inline unsigned long global_page_state(enum zone_stat_item item)
>>>>>>  	return x;
>>>>>>  }
>>>>>>
>>>>>> -static inline unsigned long zone_page_state(struct zone *zone,
>>>>>> +static inline u64 zone_page_state(struct zone *zone,
>>>>>>  					enum zone_stat_item item)
>>>>>>  {
>>>>>>  	long x = atomic_long_read(&zone->vm_stat[item]);
>>>>>
>>>>> We established that there is no known reason to use ulong for page
>>>>> counters and that IF YOU HAD ONE, you should obviously say so and then
>>>>> do a wholesale conversion.  But I don't think you have one.
>>>>>
>>>>> This patch makes absolutely no sense.
>>>>>
>>>> I agree. Then, Nack from me.
>>>>
>>>> Thanks,
>>>> -Kame
>>>
>>> static unsigned long
>>> mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
>>> 			int nid, unsigned int lru_mask)
>>> {
>>> 	u64 total = 0;
>>> 	int zid;
>>>
>>> 	for (zid = 0; zid < MAX_NR_ZONES; zid++)
>>> 		total += mem_cgroup_zone_nr_lru_pages(memcg,
>>> 					    nid, zid, lru_mask);
>>>
>>> 	return total;
>>> }
>>>
>>> Since you use unsigned long to caculate nr_pages and unsigned long long
>>> to caculate bytes, so u64 in function mem_cgroup_node_nr_lru_pages should 
>>> replace by unsigned long to save kernel stack, right?
>>
>> How many bytes do you intend to save by replacing "u64" with "unsigned
>> long" ? Have you asked yourself this question ?
> 
> I think fixing these types is a good thing.  Not for their memory
> savings but to clarify the code.  Otherwise, people will scratch their
> heads and waste time over whether there is something that requires the
> counter to be 64-bit on a 32-bit system when really there isn't.
> 
I don't see a reason not to be type consistent, it's just not a stack
issue at all. Being just a type fix, its benefits need to be offset
against any problems it may bring, such as conflicts with other people
working on that, risking doing a mid-term conversion that break
something, and all the likes.