Re: [PATCH] memory.min_usage (seqlock for res_counter)

[Pavel Emelyanov <xemul-GEFAQzZX7r8dnm+yROfE0A@public.gmane.org>]

KAMEZAWA Hiroyuki wrote:
> On Tue, 04 Dec 2007 14:10:42 +0300
> Pavel Emelyanov <xemul-GEFAQzZX7r8dnm+yROfE0A@public.gmane.org> wrote:
> 
>> KAMEZAWA Hiroyuki wrote:
>>> This is seqlock version res_counter.
>>> Maybe this this will reduce # of spin_lock.
>>>
>>> Pavel-san, How about this ?
>> AFAIS the readlock is used only in the check_under_limit(),
>> but I think, that even if we read usage and limit values
>> in this case non-atomically, this won't result in any 
>> dramatic sequence at all. No?
>>
> Reader can detect *any* changes in res_counter member which happens
> while they access res_counter between seq_begin/seq_retry.
> Memory barrier and "sequence" of seq_lock guarantees this.
> So..there is no dramatical situation.
> (it's used for accesing xtime.)
> 
> I'm sorry if I miss your point.

Sorry, let me explain it in other words.

I think, that protection in reader, that guarantees that it
will see the valid result, is not very important - even if
we compare usage and limit not atomically nothing serious
will happen (in this particular case)

> Thanks,
> -Kame
> 
>