linux-fsdevel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Dmitry Osipenko <digetx@gmail.com>
To: Ming Lei <ming.lei@redhat.com>
Cc: Jens Axboe <axboe@fb.com>, Christoph Hellwig <hch@infradead.org>,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	Kent Overstreet <kent.overstreet@gmail.com>,
	Huang Ying <ying.huang@intel.com>,
	linux-kernel@vger.kernel.org, linux-block@vger.kernel.org,
	linux-fsdevel@vger.kernel.org, linux-mm@kvack.org,
	Theodore Ts'o <tytso@mit.edu>,
	"Darrick J . Wong" <darrick.wong@oracle.com>,
	Coly Li <colyli@suse.de>, Filipe Manana <fdmanana@gmail.com>,
	Ulf Hansson <ulf.hansson@linaro.org>,
	linux-mmc@vger.kernel.org
Subject: Re: [PATCH V4 13/45] block: blk-merge: try to make front segments in full size
Date: Wed, 10 Jan 2018 18:41:14 +0300	[thread overview]
Message-ID: <d8a13047-3b20-0166-947f-dc1af01ef0bd@gmail.com> (raw)
In-Reply-To: <20180110024046.GA18922@ming.t460p>

On 10.01.2018 05:40, Ming Lei wrote:
> On Tue, Jan 09, 2018 at 08:02:53PM +0300, Dmitry Osipenko wrote:
>> On 09.01.2018 17:33, Ming Lei wrote:
>>> On Tue, Jan 09, 2018 at 04:18:39PM +0300, Dmitry Osipenko wrote:
>>>> On 09.01.2018 05:34, Ming Lei wrote:
>>>>> On Tue, Jan 09, 2018 at 12:09:27AM +0300, Dmitry Osipenko wrote:
>>>>>> On 18.12.2017 15:22, Ming Lei wrote:
>>>>>>> When merging one bvec into segment, if the bvec is too big
>>>>>>> to merge, current policy is to move the whole bvec into another
>>>>>>> new segment.
>>>>>>>
>>>>>>> This patchset changes the policy into trying to maximize size of
>>>>>>> front segments, that means in above situation, part of bvec
>>>>>>> is merged into current segment, and the remainder is put
>>>>>>> into next segment.
>>>>>>>
>>>>>>> This patch prepares for support multipage bvec because
>>>>>>> it can be quite common to see this case and we should try
>>>>>>> to make front segments in full size.
>>>>>>>
>>>>>>> Signed-off-by: Ming Lei <ming.lei@redhat.com>
>>>>>>> ---
>>>>>>>  block/blk-merge.c | 54 +++++++++++++++++++++++++++++++++++++++++++++++++-----
>>>>>>>  1 file changed, 49 insertions(+), 5 deletions(-)
>>>>>>>
>>>>>>> diff --git a/block/blk-merge.c b/block/blk-merge.c
>>>>>>> index a476337a8ff4..42ceb89bc566 100644
>>>>>>> --- a/block/blk-merge.c
>>>>>>> +++ b/block/blk-merge.c
>>>>>>> @@ -109,6 +109,7 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
>>>>>>>  	bool do_split = true;
>>>>>>>  	struct bio *new = NULL;
>>>>>>>  	const unsigned max_sectors = get_max_io_size(q, bio);
>>>>>>> +	unsigned advance = 0;
>>>>>>>  
>>>>>>>  	bio_for_each_segment(bv, bio, iter) {
>>>>>>>  		/*
>>>>>>> @@ -134,12 +135,32 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
>>>>>>>  		}
>>>>>>>  
>>>>>>>  		if (bvprvp && blk_queue_cluster(q)) {
>>>>>>> -			if (seg_size + bv.bv_len > queue_max_segment_size(q))
>>>>>>> -				goto new_segment;
>>>>>>>  			if (!BIOVEC_PHYS_MERGEABLE(bvprvp, &bv))
>>>>>>>  				goto new_segment;
>>>>>>>  			if (!BIOVEC_SEG_BOUNDARY(q, bvprvp, &bv))
>>>>>>>  				goto new_segment;
>>>>>>> +			if (seg_size + bv.bv_len > queue_max_segment_size(q)) {
>>>>>>> +				/*
>>>>>>> +				 * On assumption is that initial value of
>>>>>>> +				 * @seg_size(equals to bv.bv_len) won't be
>>>>>>> +				 * bigger than max segment size, but will
>>>>>>> +				 * becomes false after multipage bvec comes.
>>>>>>> +				 */
>>>>>>> +				advance = queue_max_segment_size(q) - seg_size;
>>>>>>> +
>>>>>>> +				if (advance > 0) {
>>>>>>> +					seg_size += advance;
>>>>>>> +					sectors += advance >> 9;
>>>>>>> +					bv.bv_len -= advance;
>>>>>>> +					bv.bv_offset += advance;
>>>>>>> +				}
>>>>>>> +
>>>>>>> +				/*
>>>>>>> +				 * Still need to put remainder of current
>>>>>>> +				 * bvec into a new segment.
>>>>>>> +				 */
>>>>>>> +				goto new_segment;
>>>>>>> +			}
>>>>>>>  
>>>>>>>  			seg_size += bv.bv_len;
>>>>>>>  			bvprv = bv;
>>>>>>> @@ -161,6 +182,12 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
>>>>>>>  		seg_size = bv.bv_len;
>>>>>>>  		sectors += bv.bv_len >> 9;
>>>>>>>  
>>>>>>> +		/* restore the bvec for iterator */
>>>>>>> +		if (advance) {
>>>>>>> +			bv.bv_len += advance;
>>>>>>> +			bv.bv_offset -= advance;
>>>>>>> +			advance = 0;
>>>>>>> +		}
>>>>>>>  	}
>>>>>>>  
>>>>>>>  	do_split = false;
>>>>>>> @@ -361,16 +388,29 @@ __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
>>>>>>>  {
>>>>>>>  
>>>>>>>  	int nbytes = bvec->bv_len;
>>>>>>> +	unsigned advance = 0;
>>>>>>>  
>>>>>>>  	if (*sg && *cluster) {
>>>>>>> -		if ((*sg)->length + nbytes > queue_max_segment_size(q))
>>>>>>> -			goto new_segment;
>>>>>>> -
>>>>>>>  		if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
>>>>>>>  			goto new_segment;
>>>>>>>  		if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
>>>>>>>  			goto new_segment;
>>>>>>>  
>>>>>>> +		/*
>>>>>>> +		 * try best to merge part of the bvec into previous
>>>>>>> +		 * segment and follow same policy with
>>>>>>> +		 * blk_bio_segment_split()
>>>>>>> +		 */
>>>>>>> +		if ((*sg)->length + nbytes > queue_max_segment_size(q)) {
>>>>>>> +			advance = queue_max_segment_size(q) - (*sg)->length;
>>>>>>> +			if (advance) {
>>>>>>> +				(*sg)->length += advance;
>>>>>>> +				bvec->bv_offset += advance;
>>>>>>> +				bvec->bv_len -= advance;
>>>>>>> +			}
>>>>>>> +			goto new_segment;
>>>>>>> +		}
>>>>>>> +
>>>>>>>  		(*sg)->length += nbytes;
>>>>>>>  	} else {
>>>>>>>  new_segment:
>>>>>>> @@ -393,6 +433,10 @@ __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
>>>>>>>  
>>>>>>>  		sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
>>>>>>>  		(*nsegs)++;
>>>>>>> +
>>>>>>> +		/* for making iterator happy */
>>>>>>> +		bvec->bv_offset -= advance;
>>>>>>> +		bvec->bv_len += advance;
>>>>>>>  	}
>>>>>>>  	*bvprv = *bvec;
>>>>>>>  }
>>>>>>>
>>>>>>
>>>>>> Hello,
>>>>>>
>>>>>> This patch breaks MMC on next-20180108, in particular MMC doesn't work anymore
>>>>>> with this patch on NVIDIA Tegra20:
>>>>>>
>>>>>> <3>[   36.622253] print_req_error: I/O error, dev mmcblk1, sector 512
>>>>>> <3>[   36.671233] print_req_error: I/O error, dev mmcblk2, sector 128
>>>>>> <3>[   36.711308] print_req_error: I/O error, dev mmcblk1, sector 31325304
>>>>>> <3>[   36.749232] print_req_error: I/O error, dev mmcblk2, sector 512
>>>>>> <3>[   36.761235] print_req_error: I/O error, dev mmcblk1, sector 31325816
>>>>>> <3>[   36.832039] print_req_error: I/O error, dev mmcblk2, sector 31259768
>>>>>> <3>[   99.793248] print_req_error: I/O error, dev mmcblk1, sector 31323136
>>>>>> <3>[   99.982043] print_req_error: I/O error, dev mmcblk1, sector 929792
>>>>>> <3>[   99.986301] print_req_error: I/O error, dev mmcblk1, sector 930816
>>>>>> <3>[  100.293624] print_req_error: I/O error, dev mmcblk1, sector 932864
>>>>>> <3>[  100.466839] print_req_error: I/O error, dev mmcblk1, sector 947200
>>>>>> <3>[  100.642955] print_req_error: I/O error, dev mmcblk1, sector 949248
>>>>>> <3>[  100.818838] print_req_error: I/O error, dev mmcblk1, sector 230400
>>>>>>
>>>>>> Any attempt of mounting MMC block dev ends with a kernel crash. Reverting this
>>>>>> patch fixes the issue.
>>>>>
>>>>> Hi Dmitry,
>>>>>
>>>>> Thanks for your report!
>>>>>
>>>>> Could you share us what the segment limits are on your MMC?
>>>>>
>>>>> 	cat /sys/block/mmcN/queue/max_segment_size
>>>>> 	cat /sys/block/mmcN/queue/max_segments
>>>>>
>>>>> Please test the following patch to see if your issue can be fixed?
>>>>>
>>>>> ---
>>>>> diff --git a/block/blk-merge.c b/block/blk-merge.c
>>>>> index 446f63e076aa..cfab36c26608 100644
>>>>> --- a/block/blk-merge.c
>>>>> +++ b/block/blk-merge.c
>>>>> @@ -431,12 +431,14 @@ __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
>>>>>  
>>>>>  		sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
>>>>>  		(*nsegs)++;
>>>>> +	}
>>>>>  
>>>>> +	*bvprv = *bvec;
>>>>> +	if (advance) {
>>>>>  		/* for making iterator happy */
>>>>>  		bvec->bv_offset -= advance;
>>>>>  		bvec->bv_len += advance;
>>>>>  	}
>>>>> -	*bvprv = *bvec;
>>>>>  }
>>>>>  
>>>>>  static inline int __blk_bvec_map_sg(struct request_queue *q, struct bio_vec bv,
>>>>
>>>> Hi Ming,
>>>>
>>>> I've tried your patch and unfortunately it doesn't help with the issue.
>>>>
>>>> Here are the segment limits:
>>>>
>>>> # cat /sys/block/mmc*/queue/max_segment_size
>>>> 65535
>>>
>>> Hi Dmitry,
>>>
>>> The 'max_segment_size' of 65535 should be the reason, could you test the
>>> following patch?
>>>
>>> ---
>>> diff --git a/block/blk-merge.c b/block/blk-merge.c
>>> index 446f63e076aa..38a66e3e678e 100644
>>> --- a/block/blk-merge.c
>>> +++ b/block/blk-merge.c
>>> @@ -12,6 +12,8 @@
>>>  
>>>  #include "blk.h"
>>>  
>>> +#define sector_align(x)   ALIGN_DOWN(x, 512)
>>> +
>>>  static struct bio *blk_bio_discard_split(struct request_queue *q,
>>>  					 struct bio *bio,
>>>  					 struct bio_set *bs,
>>> @@ -109,7 +111,7 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
>>>  	bool do_split = true;
>>>  	struct bio *new = NULL;
>>>  	const unsigned max_sectors = get_max_io_size(q, bio);
>>> -	unsigned advance = 0;
>>> +	int advance = 0;
>>>  
>>>  	bio_for_each_segment(bv, bio, iter) {
>>>  		/*
>>> @@ -144,8 +146,9 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
>>>  				 * bigger than max segment size, but this
>>>  				 * becomes false after multipage bvecs.
>>>  				 */
>>> -				advance = queue_max_segment_size(q) - seg_size;
>>> -
>>> +				advance = sector_align(
>>> +						queue_max_segment_size(q) -
>>> +						seg_size);
>>>  				if (advance > 0) {
>>>  					seg_size += advance;
>>>  					sectors += advance >> 9;
>>> @@ -386,7 +389,7 @@ __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
>>>  {
>>>  
>>>  	int nbytes = bvec->bv_len;
>>> -	unsigned advance = 0;
>>> +	int advance = 0;
>>>  
>>>  	if (*sg && *cluster) {
>>>  		if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
>>> @@ -400,8 +403,9 @@ __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
>>>  		 * blk_bio_segment_split()
>>>  		 */
>>>  		if ((*sg)->length + nbytes > queue_max_segment_size(q)) {
>>> -			advance = queue_max_segment_size(q) - (*sg)->length;
>>> -			if (advance) {
>>> +			advance = sector_align(queue_max_segment_size(q) -
>>> +					(*sg)->length);
>>> +			if (advance > 0) {
>>>  				(*sg)->length += advance;
>>>  				bvec->bv_offset += advance;
>>>  				bvec->bv_len -= advance;
>>> @@ -431,12 +435,14 @@ __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
>>>  
>>>  		sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
>>>  		(*nsegs)++;
>>> +	}
>>>  
>>> +	*bvprv = *bvec;
>>> +	if (advance > 0) {
>>>  		/* for making iterator happy */
>>>  		bvec->bv_offset -= advance;
>>>  		bvec->bv_len += advance;
>>>  	}
>>> -	*bvprv = *bvec;
>>>  }
>>>  
>>>  static inline int __blk_bvec_map_sg(struct request_queue *q, struct bio_vec bv,
>>
>> This patch doesn't help either.
> 
> OK, I will send a revert later.
> 
> Thinking of the patch further, we don't need this kind of logic for
> multipage bvec at all since almost all bvecs won't be contiguous if
> bio_add_page() is used after multipage bvec is enabled.

Revert works for me, thanks.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

  reply	other threads:[~2018-01-10 15:41 UTC|newest]

Thread overview: 54+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-12-18 12:22 [PATCH V4 00/45] block: support multipage bvec Ming Lei
2017-12-18 12:22 ` [PATCH V4 01/45] block: introduce bio helpers for converting to " Ming Lei
2017-12-18 12:22 ` [PATCH V4 02/45] block: conver to bio_first_bvec_all & bio_first_page_all Ming Lei
2017-12-18 12:22 ` [PATCH V4 03/45] fs: convert to bio_last_bvec_all() Ming Lei
2017-12-18 12:22 ` [PATCH V4 04/45] block: bounce: avoid direct access to bvec table Ming Lei
2017-12-18 12:22 ` [PATCH V4 05/45] block: bounce: don't access bio->bi_io_vec in copy_to_high_bio_irq Ming Lei
2017-12-18 12:22 ` [PATCH V4 06/45] dm: limit the max bio size as BIO_MAX_PAGES * PAGE_SIZE Ming Lei
2017-12-18 12:22 ` [PATCH V4 07/45] bcache: comment on direct access to bvec table Ming Lei
2017-12-18 12:22 ` [PATCH V4 08/45] block: move bio_alloc_pages() to bcache Ming Lei
2017-12-18 12:22 ` [PATCH V4 09/45] btrfs: avoid access to .bi_vcnt directly Ming Lei
2017-12-18 12:22 ` [PATCH V4 10/45] btrfs: avoid to access bvec table directly for a cloned bio Ming Lei
2017-12-18 12:22 ` [PATCH V4 11/45] dm-crypt: don't clear bvec->bv_page in crypt_free_buffer_pages() Ming Lei
2017-12-18 12:22 ` [PATCH V4 12/45] blk-merge: compute bio->bi_seg_front_size efficiently Ming Lei
2017-12-18 12:22 ` [PATCH V4 13/45] block: blk-merge: try to make front segments in full size Ming Lei
2018-01-08 21:09   ` Dmitry Osipenko
2018-01-09  2:34     ` Ming Lei
2018-01-09 13:18       ` Dmitry Osipenko
2018-01-09 14:33         ` Ming Lei
2018-01-09 17:02           ` Dmitry Osipenko
2018-01-10  2:40             ` Ming Lei
2018-01-10 15:41               ` Dmitry Osipenko [this message]
2017-12-18 12:22 ` [PATCH V4 14/45] block: blk-merge: remove unnecessary check Ming Lei
2017-12-18 12:22 ` [PATCH V4 15/45] block: rename bio_for_each_segment* with bio_for_each_page* Ming Lei
2018-01-06 16:19   ` Jens Axboe
2017-12-18 12:22 ` [PATCH V4 16/45] block: rename rq_for_each_segment as rq_for_each_page Ming Lei
2017-12-18 12:22 ` [PATCH V4 17/45] block: rename bio_segments() with bio_pages() Ming Lei
2017-12-18 12:22 ` [PATCH V4 18/45] block: introduce multipage page bvec helpers Ming Lei
2017-12-18 12:22 ` [PATCH V4 19/45] block: introduce bio_for_each_segment() Ming Lei
2017-12-18 12:22 ` [PATCH V4 20/45] block: use bio_for_each_segment() to compute segments count Ming Lei
2017-12-18 12:22 ` [PATCH V4 21/45] block: use bio_for_each_segment() to map sg Ming Lei
2017-12-18 12:22 ` [PATCH V4 22/45] block: introduce segment_last_page() Ming Lei
2017-12-18 12:22 ` [PATCH V4 23/45] fs/buffer.c: use bvec iterator to truncate the bio Ming Lei
2017-12-18 12:22 ` [PATCH V4 24/45] btrfs: use segment_last_page to get bio's last page Ming Lei
2017-12-18 12:22 ` [PATCH V4 25/45] block: implement bio_pages_all() via bio_for_each_page_all() Ming Lei
2017-12-18 12:22 ` [PATCH V4 26/45] block: introduce bio_segments() Ming Lei
2017-12-18 12:22 ` [PATCH V4 27/45] block: introduce rq_for_each_segment() Ming Lei
2017-12-18 12:22 ` [PATCH V4 28/45] block: loop: pass segments to iov_iter Ming Lei
2017-12-18 12:22 ` [PATCH V4 29/45] block: bio: introduce bio_for_each_page_all2 and bio_for_each_segment_all Ming Lei
2017-12-18 12:22 ` [PATCH V4 30/45] block: deal with dirtying pages for multipage bvec Ming Lei
2017-12-18 12:22 ` [PATCH V4 31/45] block: convert to bio_for_each_page_all2() Ming Lei
2017-12-18 12:22 ` [PATCH V4 32/45] md/dm/bcache: conver to bio_for_each_page_all2 and bio_for_each_segment Ming Lei
2017-12-18 12:22 ` [PATCH V4 33/45] fs: conver to bio_for_each_page_all2 Ming Lei
2017-12-18 12:22 ` [PATCH V4 34/45] btrfs: " Ming Lei
2017-12-18 12:22 ` [PATCH V4 35/45] ext4: " Ming Lei
2017-12-18 12:22 ` [PATCH V4 36/45] f2fs: " Ming Lei
2017-12-18 12:22 ` [PATCH V4 37/45] xfs: " Ming Lei
2017-12-18 12:22 ` [PATCH V4 38/45] exofs: " Ming Lei
2017-12-18 12:22 ` [PATCH V4 39/45] gfs2: " Ming Lei
2017-12-18 12:22 ` [PATCH V4 40/45] block: kill bio_for_each_page_all() Ming Lei
2017-12-18 12:22 ` [PATCH V4 41/45] block: rename bio_for_each_page_all2 as bio_for_each_page_all Ming Lei
2017-12-18 12:22 ` [PATCH V4 42/45] block: enable multipage bvecs Ming Lei
2017-12-18 12:22 ` [PATCH V4 43/45] block: bio: pass segments to bio if bio_add_page() is bypassed Ming Lei
2017-12-18 12:22 ` [PATCH V4 44/45] block: always define BIO_MAX_PAGES as 256 Ming Lei
2017-12-18 12:22 ` [PATCH V4 45/45] block: document usage of bio iterator helpers Ming Lei

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=d8a13047-3b20-0166-947f-dc1af01ef0bd@gmail.com \
    --to=digetx@gmail.com \
    --cc=axboe@fb.com \
    --cc=colyli@suse.de \
    --cc=darrick.wong@oracle.com \
    --cc=fdmanana@gmail.com \
    --cc=hch@infradead.org \
    --cc=kent.overstreet@gmail.com \
    --cc=linux-block@vger.kernel.org \
    --cc=linux-fsdevel@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=linux-mmc@vger.kernel.org \
    --cc=ming.lei@redhat.com \
    --cc=tytso@mit.edu \
    --cc=ulf.hansson@linaro.org \
    --cc=viro@zeniv.linux.org.uk \
    --cc=ying.huang@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).