Re: [PATCH 1/2] NFS: call block plug around direct write

[Boaz Harrosh <bharrosh@panasas.com>]

On 05/16/2012 03:35 AM, Peng Tao wrote:

> On Wed, May 16, 2012 at 12:23 AM, Boaz Harrosh <bharrosh@panasas.com> wrote:
>> On 05/15/2012 07:08 PM, Christoph Hellwig wrote:
>>
>>> On Tue, May 15, 2012 at 11:38:22PM +0800, Peng Tao wrote:
>>>> We bypass generic_file_aio_write() but would want to call block plug.
>>>
>>> We in this case is the pnfs block driver at most.  Thus these should
>>> be pnfs block code.
> Agreed. Just that for buffer IO case all call into block plug, so I
> thought it might be OK in DIO case as well... Anyway. I will make it
> block specific and call it LD()->dio_begin and LD()->dio_end. I am
> cooking some dio alignment patches for block layout driver and would
> love to put them there as well.
> 
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>>
>> I thought so too.
>>
>> But reading the code more closely it might be hard for the blocklayout
>> to figure out the right state to call these two. Specially the call to
>> blk_finish_plug(). So you might need to add a new LD API such as
>> LD()->finish_plug() which is empty for others.
>>
>> But again inspecting the code it looks like blk_start_plug() is a no-op
>> and blk_finish_plug() is specially optimized for the empty case.
>>
> They are empty only when !CONFIG_BLOCK. I will make it block layout
> specific as said above.
> 


You miss understood. At the very top of blk_finish_plug() there
is an if (list_empty()) return. So if there was no activity between
plug/unplug the function is very fast.

But your above plan for alignment+plug sounds very good.

Thanks
Boaz

> Thanks,
> Tao
>> So is it worth it, the extra effort? I do understand the temptation
>> to get lazy here.
>>
>> Just my $0.017
>> Boaz