f2fs: split discard command in prior to block layer

From: Chao Yu <[email protected]>

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.

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/f2fs.h    | 13 ++++++-------
 fs/f2fs/segment.c | 25 ++++++++++++++++++++++---
 2 files changed, 28 insertions(+), 10 deletions(-)

fs/f2fs/f2fs.h +6 -7

index a9da5a089cb4..a09d2b2d9520 100644
@@ -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 */

@@ -701,22 +700,22 @@ 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,

fs/f2fs/segment.c +22 -3

index 4648561e2bfd..8e417a12684d 100644
@@ -1086,6 +1086,9 @@ 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);

@@ -1129,7 +1132,8 @@ 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);

@@ -1140,7 +1144,8 @@ 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;

@@ -1170,7 +1175,11 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 		struct block_device *bdev, block_t blkstart, block_t blklen)
 {
+	struct request_queue *q = bdev_get_queue(bdev);
+	unsigned int max_discard_blocks =
+			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
 	block_t lblkstart = blkstart;
+	block_t total_len = blklen;
 
 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
 

@@ -1179,7 +1188,17 @@ static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 
 		blkstart -= FDEV(devi).start_blk;
 	}
-	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
+
+	while (total_len) {
+		if (blklen > max_discard_blocks)
+			blklen = max_discard_blocks;
+		__update_discard_tree_range(sbi, bdev, lblkstart,
+						blkstart, blklen);
+		lblkstart += blklen;
+		blkstart += blklen;
+		total_len -= blklen;
+		blklen = total_len;
+	}
 	return 0;
 }
 

Signature · 1 line

2.16.2.17.g38e79b1fd

Quoted 7 lines · Chao Yu

From: Chao Yu <[email protected]>

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.

Quoted 5 lines

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.

Hi Chao,

In terms of # of candidates, can we control this when actually issuing the discard commands?

Thanks,

Quoted 115 lines

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/f2fs.h    | 13 ++++++-------
 fs/f2fs/segment.c | 25 ++++++++++++++++++++++---
 2 files changed, 28 insertions(+), 10 deletions(-)

fs/f2fs/f2fs.h +6 -7

index a9da5a089cb4..a09d2b2d9520 100644
@@ -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 */

@@ -701,22 +700,22 @@ 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,

fs/f2fs/segment.c +22 -3

index 4648561e2bfd..8e417a12684d 100644
@@ -1086,6 +1086,9 @@ 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);

@@ -1129,7 +1132,8 @@ 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);

@@ -1140,7 +1144,8 @@ 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;

@@ -1170,7 +1175,11 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 		struct block_device *bdev, block_t blkstart, block_t blklen)
 {
+	struct request_queue *q = bdev_get_queue(bdev);
+	unsigned int max_discard_blocks =
+			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
 	block_t lblkstart = blkstart;
+	block_t total_len = blklen;
 
 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
 

@@ -1179,7 +1188,17 @@ static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 
 		blkstart -= FDEV(devi).start_blk;
 	}
-	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
+
+	while (total_len) {
+		if (blklen > max_discard_blocks)
+			blklen = max_discard_blocks;
+		__update_discard_tree_range(sbi, bdev, lblkstart,
+						blkstart, blklen);
+		lblkstart += blklen;
+		blkstart += blklen;
+		total_len -= blklen;
+		blklen = total_len;
+	}
 	return 0;
 }
 

Signature · 1 line

2.16.2.17.g38e79b1fd

Hi Jaegeuk,

Quoted 14 lines · Jaegeuk Kim

Quoted 12 lines · Chao Yu

From: Chao Yu <[email protected]>

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.

Quoted 4 lines

Hi Chao,

In terms of # of candidates, can we control this when actually issuing the discard commands?

IIUC, you mean once we pick one discard entry in rbtree, if max_{hw,}discard_sectors is smaller than size of this discard, then we can split it into smaller ones by discard_sectors, and just issue one or partials of them?

Thanks,

Quoted 116 lines

Thanks,

Quoted 113 lines

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/f2fs.h    | 13 ++++++-------
 fs/f2fs/segment.c | 25 ++++++++++++++++++++++---
 2 files changed, 28 insertions(+), 10 deletions(-)

fs/f2fs/f2fs.h +6 -7

index a9da5a089cb4..a09d2b2d9520 100644
@@ -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 */

@@ -701,22 +700,22 @@ 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,

fs/f2fs/segment.c +22 -3

index 4648561e2bfd..8e417a12684d 100644
@@ -1086,6 +1086,9 @@ 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);

@@ -1129,7 +1132,8 @@ 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);

@@ -1140,7 +1144,8 @@ 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;

@@ -1170,7 +1175,11 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 		struct block_device *bdev, block_t blkstart, block_t blklen)
 {
+	struct request_queue *q = bdev_get_queue(bdev);
+	unsigned int max_discard_blocks =
+			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
 	block_t lblkstart = blkstart;
+	block_t total_len = blklen;
 
 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
 

@@ -1179,7 +1188,17 @@ static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 
 		blkstart -= FDEV(devi).start_blk;
 	}
-	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
+
+	while (total_len) {
+		if (blklen > max_discard_blocks)
+			blklen = max_discard_blocks;
+		__update_discard_tree_range(sbi, bdev, lblkstart,
+						blkstart, blklen);
+		lblkstart += blklen;
+		blkstart += blklen;
+		total_len -= blklen;
+		blklen = total_len;
+	}
 	return 0;
 }
 

Quoted 2 lines

Quoted 2 lines

Signature · 1 line

2.16.2.17.g38e79b1fd

Quoted 19 lines · Chao Yu

Hi Jaegeuk,

Quoted 16 lines · Jaegeuk Kim

Quoted 7 lines · Chao Yu

From: Chao Yu <[email protected]>

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.

Quoted 5 lines

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.

Hi Chao,

Quoted 6 lines

Quoted 2 lines

In terms of # of candidates, can we control this when actually issuing the discard commands?

IIUC, you mean once we pick one discard entry in rbtree, if max_{hw,}discard_sectors is smaller than size of this discard, then we can split it into smaller ones by discard_sectors, and just issue one or partials of them?

Yes, sort of.

Quoted 120 lines

Thanks,

Quoted 116 lines

Thanks,

Quoted 113 lines

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/f2fs.h    | 13 ++++++-------
 fs/f2fs/segment.c | 25 ++++++++++++++++++++++---
 2 files changed, 28 insertions(+), 10 deletions(-)

fs/f2fs/f2fs.h +6 -7

index a9da5a089cb4..a09d2b2d9520 100644
@@ -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 */

@@ -701,22 +700,22 @@ 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,

fs/f2fs/segment.c +22 -3

index 4648561e2bfd..8e417a12684d 100644
@@ -1086,6 +1086,9 @@ 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);

@@ -1129,7 +1132,8 @@ 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);

@@ -1140,7 +1144,8 @@ 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;

@@ -1170,7 +1175,11 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 		struct block_device *bdev, block_t blkstart, block_t blklen)
 {
+	struct request_queue *q = bdev_get_queue(bdev);
+	unsigned int max_discard_blocks =
+			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
 	block_t lblkstart = blkstart;
+	block_t total_len = blklen;
 
 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
 

@@ -1179,7 +1188,17 @@ static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
 
 		blkstart -= FDEV(devi).start_blk;
 	}
-	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
+
+	while (total_len) {
+		if (blklen > max_discard_blocks)
+			blklen = max_discard_blocks;
+		__update_discard_tree_range(sbi, bdev, lblkstart,
+						blkstart, blklen);
+		lblkstart += blklen;
+		blkstart += blklen;
+		total_len -= blklen;
+		blklen = total_len;
+	}
 	return 0;
 }
 

Quoted 2 lines

Quoted 2 lines

Quoted 2 lines

Signature · 1 line

2.16.2.17.g38e79b1fd