Re: [PATCH 1/2] block: move the padding adjustment to blk_rq_map_sg

[Tejun Heo <htejun@gmail.com>]

Hello, FUJITA.

FUJITA Tomonori wrote:
>>>>> The current code does padding only for requests that we call
>>>>> __blk_rq_map_user (bio_copy_user) for. You meant that if we create a
>>>>> new flag like REQ_NEED_PADDING and set it in blk_rq_map_user,
>>>>> blk_rq_map_sg can do padding only when it was set.
>>>> Yeap, that's what I meant.
>>> Thanks, I see. Then, I have one question: blk_rq_map_user_iov doesn't
>>> padding, thus libata still needs to adjust scatter list, right?
>> Yes but it doesn't, so blk_rq_map_user_iov() path is essentially broken 
>> regarding padding at the moment.  :-(
> 
> I see. So we need to use bounce buffers for blk_rq_map_user_iov like
> bio_copy_user does for blk_rq_map_user?

Yeap.  I think so.  It would be great if we can do it where all the PC 
commands pass through so that blk_rq_map_useR_iov() and 
blk_rq_map_user() don't have to do it separately and as in your earlier 
patch it's probably best to move sg manipulation close to drain handling.

Thanks.

-- 
tejun