f2fs: add ckpt_valid_blocks to the section entry

when performing buffered writes in a large section,
overhead is incurred due to the iteration through
ckpt_valid_blocks within the section.
when SEGS_PER_SEC is 128, this overhead accounts for 20% within
the f2fs_write_single_data_page routine.
as the size of the section increases, the overhead also grows.
to handle this problem ckpt_valid_blocks is
added within the section entries.

Test
insmod null_blk.ko nr_devices=1 completion_nsec=1 submit_queues=8
hw_queue_depth=64 max_sectors=512 bs=4096 memory_backed=1
make_f2fs /dev/block/nullb0
make_f2fs -s 128 /dev/block/nullb0
fio --bs=512k --size=1536M --rw=write --name=1
--filename=/mnt/test_dir/seq_write
--ioengine=io_uring --iodepth=64 --end_fsync=1

before
SEGS_PER_SEC 1
2556MiB/s
SEGS_PER_SEC 128
2145MiB/s

after
SEGS_PER_SEC 1
2556MiB/s
SEGS_PER_SEC 128
2556MiB/s

Signed-off-by: yohan.joung <[e-mail]>
Reviewed-by: Chao Yu <[e-mail]>
Signed-off-by: Jaegeuk Kim <[e-mail]>

fs/f2fs/segment.c

···
  * that the consecutive input blocks belong to the same segment.
  */
 static int update_sit_entry_for_release(struct f2fs_sb_info *sbi, struct seg_entry *se,
-				block_t blkaddr, unsigned int offset, int del)
+				unsigned int segno, block_t blkaddr, unsigned int offset, int del)
 {
 	bool exist;
 #ifdef CONFIG_F2FS_CHECK_FS
···
 				f2fs_test_and_clear_bit(offset + i, se->discard_map))
 			sbi->discard_blks++;
 
-		if (!f2fs_test_bit(offset + i, se->ckpt_valid_map))
+		if (!f2fs_test_bit(offset + i, se->ckpt_valid_map)) {
 			se->ckpt_valid_blocks -= 1;
+			if (__is_large_section(sbi))
+				get_sec_entry(sbi, segno)->ckpt_valid_blocks -= 1;
+		}
 	}
+
+	if (__is_large_section(sbi))
+		sanity_check_valid_blocks(sbi, segno);
 
 	return del;
 }
 
 static int update_sit_entry_for_alloc(struct f2fs_sb_info *sbi, struct seg_entry *se,
-				block_t blkaddr, unsigned int offset, int del)
+				unsigned int segno, block_t blkaddr, unsigned int offset, int del)
 {
 	bool exist;
 #ifdef CONFIG_F2FS_CHECK_FS
···
 	 * or newly invalidated.
 	 */
 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
-		if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
+		if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) {
 			se->ckpt_valid_blocks++;
+			if (__is_large_section(sbi))
+				get_sec_entry(sbi, segno)->ckpt_valid_blocks++;
+		}
 	}
 
-	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
+	if (!f2fs_test_bit(offset, se->ckpt_valid_map)) {
 		se->ckpt_valid_blocks += del;
+		if (__is_large_section(sbi))
+			get_sec_entry(sbi, segno)->ckpt_valid_blocks += del;
+	}
+
+	if (__is_large_section(sbi))
+		sanity_check_valid_blocks(sbi, segno);
 
 	return del;
 }
···
 
 	/* Update valid block bitmap */
 	if (del > 0) {
-		del = update_sit_entry_for_alloc(sbi, se, blkaddr, offset, del);
+		del = update_sit_entry_for_alloc(sbi, se, segno, blkaddr, offset, del);
 	} else {
-		del = update_sit_entry_for_release(sbi, se, blkaddr, offset, del);
+		del = update_sit_entry_for_release(sbi, se, segno, blkaddr, offset, del);
 	}
 
 	__mark_sit_entry_dirty(sbi, segno);
···
 						&raw_sit->entries[sit_offset]);
 			}
 
+			/* update ckpt_valid_block */
+			if (__is_large_section(sbi)) {
+				set_ckpt_valid_blocks(sbi, segno);
+				sanity_check_valid_blocks(sbi, segno);
+			}
+
 			__clear_bit(segno, bitmap);
 			sit_i->dirty_sentries--;
 			ses->entry_cnt--;
···
 		}
 	}
 	up_read(&curseg->journal_rwsem);
+
+	/* update ckpt_valid_block */
+	if (__is_large_section(sbi)) {
+		unsigned int segno;
+
+		for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
+			set_ckpt_valid_blocks(sbi, segno);
+			sanity_check_valid_blocks(sbi, segno);
+		}
+	}
 
 	if (err)
 		return err;

fs/f2fs/segment.h

···
 
 struct sec_entry {
 	unsigned int valid_blocks;	/* # of valid blocks in a section */
+	unsigned int ckpt_valid_blocks; /* # of valid blocks last cp in a section */
 };
 
 #define MAX_SKIP_GC_COUNT			16
···
 static inline unsigned int get_ckpt_valid_blocks(struct f2fs_sb_info *sbi,
 				unsigned int segno, bool use_section)
 {
-	if (use_section && __is_large_section(sbi)) {
-		unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
-		unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno);
-		unsigned int blocks = 0;
-		int i;
-
-		for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) {
-			struct seg_entry *se = get_seg_entry(sbi, start_segno);
-
-			blocks += se->ckpt_valid_blocks;
-		}
-		return blocks;
-	}
-	return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
+	if (use_section && __is_large_section(sbi))
+		return get_sec_entry(sbi, segno)->ckpt_valid_blocks;
+	else
+		return get_seg_entry(sbi, segno)->ckpt_valid_blocks;
 }
 
+static inline void set_ckpt_valid_blocks(struct f2fs_sb_info *sbi,
+		unsigned int segno)
+{
+	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
+	unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno);
+	unsigned int blocks = 0;
+	int i;
+
+	for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) {
+		struct seg_entry *se = get_seg_entry(sbi, start_segno);
+
+		blocks += se->ckpt_valid_blocks;
+	}
+	get_sec_entry(sbi, segno)->ckpt_valid_blocks = blocks;
+}
+
+#ifdef CONFIG_F2FS_CHECK_FS
+static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi,
+		unsigned int segno)
+{
+	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
+	unsigned int start_segno = GET_SEG_FROM_SEC(sbi, secno);
+	unsigned int blocks = 0;
+	int i;
+
+	for (i = 0; i < SEGS_PER_SEC(sbi); i++, start_segno++) {
+		struct seg_entry *se = get_seg_entry(sbi, start_segno);
+
+		blocks += se->ckpt_valid_blocks;
+	}
+
+	if (blocks != get_sec_entry(sbi, segno)->ckpt_valid_blocks) {
+		f2fs_err(sbi,
+			"Inconsistent ckpt valid blocks: "
+			"seg entry(%d) vs sec entry(%d) at secno %d",
+			blocks, get_sec_entry(sbi, segno)->ckpt_valid_blocks, secno);
+		f2fs_bug_on(sbi, 1);
+	}
+}
+#else
+static inline void sanity_check_valid_blocks(struct f2fs_sb_info *sbi,
+			unsigned int segno)
+{
+}
+#endif
 static inline void seg_info_from_raw_sit(struct seg_entry *se,
 					struct f2fs_sit_entry *rs)
 {