Re: [PATCH -V4 04/10] memcg: Add HugeTLB extension

[KAMEZAWA Hiroyuki]

(2012/03/21 13:48), Aneesh Kumar K.V wrote:

> Glauber Costa <glommer@xxxxxxxxxxxxx> writes:
> 
>> On 03/19/2012 11:00 AM, KAMEZAWA Hiroyuki wrote:
>>> (2012/03/19 15:52), Aneesh Kumar K.V wrote:
>>>
>>>>
>>>>>> +#ifdef CONFIG_MEM_RES_CTLR_HUGETLB
>>>>>> +static bool mem_cgroup_have_hugetlb_usage(struct mem_cgroup *memcg)
>>>>>> +{
>>>>>> +	int idx;
>>>>>> +	for (idx = 0; idx<  hugetlb_max_hstate; idx++) {
>>>>>> +		if (memcg->hugepage[idx].usage>  0)
>>>>>> +			return 1;
>>>>>> +	}
>>>>>> +	return 0;
>>>>>> +}
>>>>>
>>>>>
>>>>> Please use res_counter_read_u64() rather than reading the value directly.
>>>>>
>>>>
>>>> The open-coded variant is mostly derived from mem_cgroup_force_empty. I
>>>> have updated the patch to use res_counter_read_u64.
>>>>
>>>
>>> Ah, ok. it's(maybe) my bad. I'll schedule a fix.
>>>
>> Kame,
>>
>> I actually have it ready here. I can submit it if you want.
>>
>> This one has bitten me as well when I was trying to experiment with the 
>> res_counter performance...
> 
> Do we really need memcg.res.usage to be accurate in that while loop ? If
> we miss a zero update because we encountered a partial update; in the
> next loop we will find it zero right ?
> 

At rmdir(), I assume there is no task in memcg. It means res->usage never
increase and no other thread than force_empty will touch res->counter.
So, I think memcg->res.usage > 0 never be wrong and we'll find correct comparison
by continuing the loop.

But recent kmem accounting at el may break the assumption (I'm not fully sure..)
So, I think it will be good to use res_counter_u64(). This part is not important for
performance, anyway.

Thanks,
-Kame



--
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/