Date:   Mon, 23 Jul 2018 21:22:14 +0800
From:   Chao Yu <chao@...nel.org>
To:     Jaegeuk Kim <jaegeuk@...nel.org>
Cc:     linux-kernel@...r.kernel.org,
        linux-f2fs-devel@...ts.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH v2] f2fs: split discard command in prior to
 block layer

On 2018/7/23 20:47, Jaegeuk Kim wrote:
> On 07/15, Chao Yu wrote:
>> On 2018/7/15 11:13, Jaegeuk Kim wrote:
>>> On 07/15, Chao Yu wrote:
>>>> Hi Jaegeuk,
>>>>
>>>> On 2018/7/15 9:27, Jaegeuk Kim wrote:
>>>>> On 07/08, Chao Yu wrote:
>>>>>> From: Chao Yu <yuchao0@...wei.com>
>>>>>>
>>>>>> Some devices has small max_{hw,}discard_sectors, so that in
>>>>>> __blkdev_issue_discard(), one big size discard bio can be split
>>>>>> into multiple small size discard bios, result in heavy load in IO
>>>>>> scheduler and device, which can hang other sync IO for long time.
>>>>>>
>>>>>> Now, f2fs is trying to control discard commands more elaboratively,
>>>>>> in order to make less conflict in between discard IO and user IO
>>>>>> to enhance application's performance, so in this patch, we will
>>>>>> split discard bio in f2fs in prior to in block layer to reduce
>>>>>> issuing multiple discard bios in a short time.
>>>>>
>>>>> Can we just change __submit_discard_cmd() to submit the large candidate
>>>>> partially and adjust lstart in the original candidate?
>>>>
>>>> Yep, could you check __submit_discard_cmd() implementation below? Does that
>>>> match the thoughts from you?
>>>>
>>>> Thanks,
>>>>
>>>>>
>>>>>>
>>>>>> Signed-off-by: Chao Yu <yuchao0@...wei.com>
>>>>>> ---
>>>>>> v2:
>>>>>> - change to split discard command entry before submission.
>>>>>>  fs/f2fs/f2fs.h    |  13 +++---
>>>>>>  fs/f2fs/segment.c | 117 +++++++++++++++++++++++++++++++++++++-----------------
>>>>>>  2 files changed, 86 insertions(+), 44 deletions(-)
>>>>>>
>>>>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
>>>>>> index c48b655d5d8d..359526d88d3f 100644
>>>>>> --- a/fs/f2fs/f2fs.h
>>>>>> +++ b/fs/f2fs/f2fs.h
>>>>>> @@ -178,7 +178,6 @@ enum {
>>>>>>  
>>>>>>  #define MAX_DISCARD_BLOCKS(sbi)		BLKS_PER_SEC(sbi)
>>>>>>  #define DEF_MAX_DISCARD_REQUEST		8	/* issue 8 discards per round */
>>>>>> -#define DEF_MAX_DISCARD_LEN		512	/* Max. 2MB per discard */
>>>>>>  #define DEF_MIN_DISCARD_ISSUE_TIME	50	/* 50 ms, if exists */
>>>>>>  #define DEF_MID_DISCARD_ISSUE_TIME	500	/* 500 ms, if device busy */
>>>>>>  #define DEF_MAX_DISCARD_ISSUE_TIME	60000	/* 60 s, if no candidates */
>>>>>> @@ -709,22 +708,22 @@ static inline void set_extent_info(struct extent_info *ei, unsigned int fofs,
>>>>>>  }
>>>>>>  
>>>>>>  static inline bool __is_discard_mergeable(struct discard_info *back,
>>>>>> -						struct discard_info *front)
>>>>>> +			struct discard_info *front, unsigned int max_len)
>>>>>>  {
>>>>>>  	return (back->lstart + back->len == front->lstart) &&
>>>>>> -		(back->len + front->len < DEF_MAX_DISCARD_LEN);
>>>>>> +		(back->len + front->len <= max_len);
>>>>>>  }
>>>>>>  
>>>>>>  static inline bool __is_discard_back_mergeable(struct discard_info *cur,
>>>>>> -						struct discard_info *back)
>>>>>> +			struct discard_info *back, unsigned int max_len)
>>>>>>  {
>>>>>> -	return __is_discard_mergeable(back, cur);
>>>>>> +	return __is_discard_mergeable(back, cur, max_len);
>>>>>>  }
>>>>>>  
>>>>>>  static inline bool __is_discard_front_mergeable(struct discard_info *cur,
>>>>>> -						struct discard_info *front)
>>>>>> +			struct discard_info *front, unsigned int max_len)
>>>>>>  {
>>>>>> -	return __is_discard_mergeable(cur, front);
>>>>>> +	return __is_discard_mergeable(cur, front, max_len);
>>>>>>  }
>>>>>>  
>>>>>>  static inline bool __is_extent_mergeable(struct extent_info *back,
>>>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>>>>> index f7f56dd091b4..cbf8f3f9a8e7 100644
>>>>>> --- a/fs/f2fs/segment.c
>>>>>> +++ b/fs/f2fs/segment.c
>>>>>> @@ -966,17 +966,24 @@ static void __init_discard_policy(struct f2fs_sb_info *sbi,
>>>>>>  	}
>>>>>>  }
>>>>>>  
>>>>>> -
>>>
>>> ===================== BEGIN
>>>
>>>>>> +static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
>>>>>> +				struct block_device *bdev, block_t lstart,
>>>>>> +				block_t start, block_t len);
>>>>>>  /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
>>>>>>  static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
>>>>>>  						struct discard_policy *dpolicy,
>>>>>> -						struct discard_cmd *dc)
>>>>>> +						struct discard_cmd *dc,
>>>>>> +						unsigned int *issued)
>>>>>>  {
>>>>>> +	struct block_device *bdev = dc->bdev;
>>>>>> +	struct request_queue *q = bdev_get_queue(bdev);
>>>>>> +	unsigned int max_discard_blocks =
>>>>>> +			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
>>>>>>  	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
>>>>>>  	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
>>>>>>  					&(dcc->fstrim_list) : &(dcc->wait_list);
>>>>>> -	struct bio *bio = NULL;
>>>>>>  	int flag = dpolicy->sync ? REQ_SYNC : 0;
>>>>>> +	block_t lstart, start, len, total_len;
>>>>>>  
>>>>>>  	if (dc->state != D_PREP)
>>>>>>  		return;
>>>>>> @@ -984,30 +991,63 @@ static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
>>>>>>  	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
>>>>>>  		return;
>>>>>>  
>>>>>> -	trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
>>>>>> -
>>>>>> -	dc->error = __blkdev_issue_discard(dc->bdev,
>>>>>> -				SECTOR_FROM_BLOCK(dc->start),
>>>>>> -				SECTOR_FROM_BLOCK(dc->len),
>>>>>> -				GFP_NOFS, 0, &bio);
>>>>>> -	if (!dc->error) {
>>>>>> -		/* should keep before submission to avoid D_DONE right away */
>>>>>> -		dc->state = D_SUBMIT;
>>>>>> -		atomic_inc(&dcc->issued_discard);
>>>>>> -		atomic_inc(&dcc->issing_discard);
>>>>>> -		if (bio) {
>>>>>> -			bio->bi_private = dc;
>>>>>> -			bio->bi_end_io = f2fs_submit_discard_endio;
>>>>>> -			bio->bi_opf |= flag;
>>>>>> -			submit_bio(bio);
>>>>>> -			list_move_tail(&dc->list, wait_list);
>>>>>> -			__check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
>>>>>> -
>>>>>> -			f2fs_update_iostat(sbi, FS_DISCARD, 1);
>>>>>> +	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
>>>>>> +
>>>>>> +	lstart = dc->lstart;
>>>>>> +	start = dc->start;
>>>>>> +	len = dc->len;
>>>>>> +	total_len = len;
>>>>>> +
>>>>>> +	while (total_len && *issued < dpolicy->max_requests) {
>>>>>> +		struct bio *bio = NULL;
>>>>>> +
>>>>>> +		if (len > max_discard_blocks)
>>>>>> +			len = max_discard_blocks;
>>>>>> +
>>>>>> +		dc->error = __blkdev_issue_discard(bdev,
>>>>>> +					SECTOR_FROM_BLOCK(start),
>>>>>> +					SECTOR_FROM_BLOCK(len),
>>>>>> +					GFP_NOFS, 0, &bio);
>>>>>> +		if (!dc->error) {
>>>>>> +			/*
>>>>>> +			 * should keep before submission to avoid D_DONE
>>>>>> +			 * right away
>>>>>> +			 */
>>>>>> +			dc->state = D_SUBMIT;
>>>>>> +			dc->len = len;
>>>>>> +			atomic_inc(&dcc->issued_discard);
>>>>>> +			atomic_inc(&dcc->issing_discard);
>>>>>> +			if (bio) {
>>>>>> +				bio->bi_private = dc;
>>>>>> +				bio->bi_end_io = f2fs_submit_discard_endio;
>>>>>> +				bio->bi_opf |= flag;
>>>>>> +				submit_bio(bio);
>>>>>> +				list_move_tail(&dc->list, wait_list);
>>>>>> +				__check_sit_bitmap(sbi, dc->start,
>>>>>> +							dc->start + dc->len);
>>>>>> +
>>>>>> +				f2fs_update_iostat(sbi, FS_DISCARD, 1);
>>>>>> +			}
>>>>>> +		} else {
>>>>>> +			__remove_discard_cmd(sbi, dc);
>>>
>>> Keep the original one, since it's partially failed?
>>
>> During __submit_discard_cmd(), we are trying to split original dc into multiple
>> dc, and once we failed to submit some of them, we will remove corresponding dc,
>> the removal logic is the same as before, not being changed...
>>
>> If we want to keep failed dc and do retry, we'd better add another patch for
>> that, right?
> 
> I thought removing failed portion of the dc.
> 
>>
>>>
>>>>>> +			return;
>>>>>> +		}
>>>>>> +		lstart += len;
>>>>>> +		start += len;
>>>>>> +		total_len -= len;
>>>>>> +		len = total_len;
>>>>>> +		(*issued)++;
>>>>>> +
>>>>>> +		if (len) {
>>>>>> +			__update_discard_tree_range(sbi, bdev, lstart,
>>>>>> +								start, len);
>>>
>>> Do we really need this? We already grabbed one candidate which was not merged
>>> well.
>>
>> You mean submitting multiple bios based on one large size dc?
> 
> Yes.
> 
>>
>> Previously, one bio references to one dc as below, in end_io, it will change
>> dc->state which can be detected by its own waiter.
>>
>> 	bio->bi_private = dc;
>> 	bio->bi_end_io = f2fs_submit_discard_endio;
>> 	bio->bi_opf |= flag;
>>
>> If multiple bios reference to one dc, it will be more complicated to handle
>> that. How do you think?
> 
> I was concerned about too many entries in the lists. How about adding a refcnt
> to count how many bios were issued on the dc?

Got you, let me try to refactor a bit for that.

Thanks,

> 
> Thanks,
> 
>>
>> Thanks,
>>
>>>
>>>>>> +			dc = (struct discard_cmd *)f2fs_lookup_rb_tree(
>>>>>> +						&dcc->root, NULL, lstart);
>>>>>> +			f2fs_bug_on(sbi, !dc);
>>>>>>  		}
>>>>>> -	} else {
>>>>>> -		__remove_discard_cmd(sbi, dc);
>>>>>>  	}
>>>>>> +
>>>>>> +	return;
>>>>>>  }
>>>
>>> ======================= END
>>>
>>>
>>>>>>  
>>>>>>  static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
>>>>>> @@ -1088,10 +1128,11 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
>>>>>>  	struct discard_cmd *dc;
>>>>>>  	struct discard_info di = {0};
>>>>>>  	struct rb_node **insert_p = NULL, *insert_parent = NULL;
>>>>>> +	struct request_queue *q = bdev_get_queue(bdev);
>>>>>> +	unsigned int max_discard_blocks =
>>>>>> +			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
>>>>>>  	block_t end = lstart + len;
>>>>>>  
>>>>>> -	mutex_lock(&dcc->cmd_lock);
>>>>>> -
>>>>>>  	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
>>>>>>  					NULL, lstart,
>>>>>>  					(struct rb_entry **)&prev_dc,
>>>>>> @@ -1131,7 +1172,8 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
>>>>>>  
>>>>>>  		if (prev_dc && prev_dc->state == D_PREP &&
>>>>>>  			prev_dc->bdev == bdev &&
>>>>>> -			__is_discard_back_mergeable(&di, &prev_dc->di)) {
>>>>>> +			__is_discard_back_mergeable(&di, &prev_dc->di,
>>>>>> +							max_discard_blocks)) {
>>>>>>  			prev_dc->di.len += di.len;
>>>>>>  			dcc->undiscard_blks += di.len;
>>>>>>  			__relocate_discard_cmd(dcc, prev_dc);
>>>>>> @@ -1142,7 +1184,8 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
>>>>>>  
>>>>>>  		if (next_dc && next_dc->state == D_PREP &&
>>>>>>  			next_dc->bdev == bdev &&
>>>>>> -			__is_discard_front_mergeable(&di, &next_dc->di)) {
>>>>>> +			__is_discard_front_mergeable(&di, &next_dc->di,
>>>>>> +							max_discard_blocks)) {
>>>>>>  			next_dc->di.lstart = di.lstart;
>>>>>>  			next_dc->di.len += di.len;
>>>>>>  			next_dc->di.start = di.start;
>>>>>> @@ -1165,8 +1208,6 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
>>>>>>  		node = rb_next(&prev_dc->rb_node);
>>>>>>  		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
>>>>>>  	}
>>>>>> -
>>>>>> -	mutex_unlock(&dcc->cmd_lock);
>>>>>>  }
>>>>>>  
>>>>>>  static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
>>>>>> @@ -1181,7 +1222,9 @@ static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
>>>>>>  
>>>>>>  		blkstart -= FDEV(devi).start_blk;
>>>>>>  	}
>>>>>> +	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
>>>>>>  	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
>>>>>> +	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
>>>>>>  	return 0;
>>>>>>  }
>>>>>>  
>>>>>> @@ -1220,9 +1263,9 @@ static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
>>>>>>  		}
>>>>>>  
>>>>>>  		dcc->next_pos = dc->lstart + dc->len;
>>>>>> -		__submit_discard_cmd(sbi, dpolicy, dc);
>>>>>> +		__submit_discard_cmd(sbi, dpolicy, dc, &issued);
>>>>>>  
>>>>>> -		if (++issued >= dpolicy->max_requests)
>>>>>> +		if (issued >= dpolicy->max_requests)
>>>>>>  			break;
>>>>>>  next:
>>>>>>  		node = rb_next(&dc->rb_node);
>>>>>> @@ -1277,9 +1320,9 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
>>>>>>  				break;
>>>>>>  			}
>>>>>>  
>>>>>> -			__submit_discard_cmd(sbi, dpolicy, dc);
>>>>>> +			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
>>>>>>  
>>>>>> -			if (++issued >= dpolicy->max_requests)
>>>>>> +			if (issued >= dpolicy->max_requests)
>>>>>>  				break;
>>>>>>  		}
>>>>>>  		blk_finish_plug(&plug);
>>>>>> @@ -2475,9 +2518,9 @@ static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
>>>>>>  			goto skip;
>>>>>>  		}
>>>>>>  
>>>>>> -		__submit_discard_cmd(sbi, dpolicy, dc);
>>>>>> +		__submit_discard_cmd(sbi, dpolicy, dc, &issued);
>>>>>>  
>>>>>> -		if (++issued >= dpolicy->max_requests) {
>>>>>> +		if (issued >= dpolicy->max_requests) {
>>>>>>  			start = dc->lstart + dc->len;
>>>>>>  
>>>>>>  			blk_finish_plug(&plug);
>>>>>> -- 
>>>>>> 2.16.2.17.g38e79b1fd
> 
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@...ts.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> 

Powered by blists - more mailing lists