Re: [PATCH 1/2] nfsd: cleanup setting of default max_block_size

[Mi Jinlong <kinglongmee@gmail.com>]

于 2012-2-4 10:16, J. Bruce Fields 写道:
> On Sat, Feb 04, 2012 at 12:24:31AM +0800, Mi Jinlong wrote:
>> 于 2012-2-4 4:49, J. Bruce Fields 写道:
>>> On Wed, Feb 01, 2012 at 04:46:56PM -0500, bfields wrote:
>>>> On Tue, Jan 31, 2012 at 09:11:48AM +0800, Mi Jinlong wrote:
>>>>>    Should use target = i.totalram<<   (PAGE_SHIFT - 12);
>>>>>
>>>>>    target = i.totalram<<   PAGE_SHIFT; and
>>>>>    target<<= 12;
>>>>>    means target = i.totalram<<   (PAGE_SHIFT + 12);
>>>>
>>>> Yes, thanks for catching that.
>>>>
>>>> Also, splitting up the calculation as I did above risks overflow at the
>>>> first step.
>>>>
>>>> I'll fix that....
>>>
>>> Here are the fixed patches.--b.
>>>
>>> > From 87b0fc7deb5feccf93b022f6a976e8441152dbb2 Mon Sep 17 00:00:00 2001
>>> From: "J. Bruce Fields"<bfields@redhat.com>
>>> Date: Mon, 30 Jan 2012 16:18:35 -0500
>>> Subject: [PATCH 1/2] nfsd: cleanup setting of default max_block_size
>>>
>>> Move calculation of the default into a helper function.
>>>
>>> Get rid of an unused variable "err" while we're there.
>>>
>>> Thanks to Mi Jinlong for catching an arithmetic error in a previous
>>> version.
>>>
>>> Cc: Mi Jinlong<mijinlong@cn.fujitsu.com>
>>> Signed-off-by: J. Bruce Fields<bfields@redhat.com>
>>> ---
>>>   fs/nfsd/nfssvc.c |   44 ++++++++++++++++++++++++--------------------
>>>   1 files changed, 24 insertions(+), 20 deletions(-)
>>>
>>> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
>>> index eda7d7e..e9eb408 100644
>>> --- a/fs/nfsd/nfssvc.c
>>> +++ b/fs/nfsd/nfssvc.c
>>> @@ -307,33 +307,37 @@ static void set_max_drc(void)
>>>   	dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
>>>   }
>>>
>>> -int nfsd_create_serv(void)
>>> +static int nfsd_get_default_max_blksize(void)
>>>   {
>>> -	int err = 0;
>>> +	struct sysinfo i;
>>> +	unsigned long long target;
>>> +	unsigned long ret;
>>> +
>>> +	si_meminfo(&i);
>>> +	target = i.totalram<<   PAGE_SHIFT;
>>> +	/*
>>> +	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
>>> +	 * machines, but only uses 32K on 128M machines.  Bottom out at
>>> +	 * 8K on 32M and smaller.  Of course, this is only a default.
>>> +	 */
>>> +	target>>= 12;
>>
>>     Why don't using target = i.totalram<<  (PAGE_SHIFT - 12) as before?
>
> Dividing the calculation into two steps (first convert from pages to
> bytes, then divide by 4096) makes it more obvious, and allows me to
> stick the comment before the part of the calculation it explains.
>
> So the result seems easier to read.

   Yes,

>
>>     The result of the two forms is more likely different.
>
> The result is the same as long as there's no overflow at the first step
> (which would require a machine with an exabyte of ram).
>
> Seem reasonable?

   Yes, that is my only concern.

thanks,
Mi Jinlong