Skip to contentSkip to conversation

f2fs: remove a waiter for checkpoint completion

PATCH In review Jaegeuk Kim 2 participants 6 messages Aug 3, 2020 – Aug 4, 2020 F2FS file system

lore
Share & tools
lore

Jaegeuk Kim

author maintainer
#0 lore raw

PATCH 1/1 f2fs: remove a waiter for checkpoint completion

It doesn't need to wait for checkpoint being completed triggered by end_io.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
---
 fs/f2fs/checkpoint.c | 6 +-----
 fs/f2fs/data.c       | 4 ----
 fs/f2fs/f2fs.h       | 1 -
 fs/f2fs/super.c      | 1 -
 4 files changed, 1 insertion(+), 11 deletions(-)
fs/f2fs/checkpoint.c +1 -5
index 99c8061da55b9..2bdddc725e677 100644
@@ -1255,11 +1255,7 @@ static void unblock_operations(struct f2fs_sb_info *sbi)
 
 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
 {
-	DEFINE_WAIT(wait);
-
 	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
-
 		if (!get_pages(sbi, type))
 			break;
 
@@ -1271,9 +1267,9 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
 							FS_CP_META_IO);
 		else if (type == F2FS_WB_CP_DATA)
 			f2fs_submit_merged_write(sbi, DATA);
+
 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 	}
-	finish_wait(&sbi->cp_wait, &wait);
 }
 
 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
fs/f2fs/data.c +0 -4
index c1b676be67b9a..588d2871bbe69 100644
@@ -383,10 +383,6 @@ static void f2fs_write_end_io(struct bio *bio)
 		clear_cold_data(page);
 		end_page_writeback(page);
 	}
-	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
-				wq_has_sleeper(&sbi->cp_wait))
-		wake_up(&sbi->cp_wait);
-
 	bio_put(bio);
 }
 
fs/f2fs/f2fs.h +0 -1
index 02811ce15059b..a036539363030 100644
@@ -1437,7 +1437,6 @@ struct f2fs_sb_info {
 	struct rw_semaphore cp_rwsem;		/* blocking FS operations */
 	struct rw_semaphore node_write;		/* locking node writes */
 	struct rw_semaphore node_change;	/* locking node change */
-	wait_queue_head_t cp_wait;
 	unsigned long last_time[MAX_TIME];	/* to store time in jiffies */
 	long interval_time[MAX_TIME];		/* to store thresholds */
 
fs/f2fs/super.c +0 -1
index 83bf9a02f83f2..5f4d9abc079d6 100644
@@ -3540,7 +3540,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
 
 	init_rwsem(&sbi->cp_rwsem);
 	init_rwsem(&sbi->quota_sem);
-	init_waitqueue_head(&sbi->cp_wait);
 	init_sb_info(sbi);
 
 	err = init_percpu_info(sbi);
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
#1 lore raw
Quoted 31 lines · Jaegeuk Kim

It doesn't need to wait for checkpoint being completed triggered by end_io.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
---
  fs/f2fs/checkpoint.c | 6 +-----
  fs/f2fs/data.c       | 4 ----
  fs/f2fs/f2fs.h       | 1 -
  fs/f2fs/super.c      | 1 -
  4 files changed, 1 insertion(+), 11 deletions(-)
fs/f2fs/checkpoint.c +0 -0
index 99c8061da55b9..2bdddc725e677 100644

fs/f2fs/checkpoint.c static void unblock_operations(struct f2fs_sb_info *sbi) · line 1261

Quoted 7 lines
fs/f2fs/checkpoint.c +0 -3
@@ -1255,11 +1255,7 @@ static void unblock_operations(struct f2fs_sb_info *sbi)
 
  void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
  {
-	DEFINE_WAIT(wait);
-
  	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);

Wouldn't that case high cpu usage before io end?

Quoted 55 lines
fs/f2fs/checkpoint.c +1 -2
-
  		if (!get_pages(sbi, type))
  			break;
 
@@ -1271,9 +1267,9 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
  							FS_CP_META_IO);
  		else if (type == F2FS_WB_CP_DATA)
  			f2fs_submit_merged_write(sbi, DATA);
+
  		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
  	}
-	finish_wait(&sbi->cp_wait, &wait);
  }
 
  static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
fs/f2fs/data.c +0 -4
index c1b676be67b9a..588d2871bbe69 100644
@@ -383,10 +383,6 @@ static void f2fs_write_end_io(struct bio *bio)
  		clear_cold_data(page);
  		end_page_writeback(page);
  	}
-	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
-				wq_has_sleeper(&sbi->cp_wait))
-		wake_up(&sbi->cp_wait);
-
  	bio_put(bio);
  }
 
fs/f2fs/f2fs.h +0 -1
index 02811ce15059b..a036539363030 100644
@@ -1437,7 +1437,6 @@ struct f2fs_sb_info {
  	struct rw_semaphore cp_rwsem;		/* blocking FS operations */
  	struct rw_semaphore node_write;		/* locking node writes */
  	struct rw_semaphore node_change;	/* locking node change */
-	wait_queue_head_t cp_wait;
  	unsigned long last_time[MAX_TIME];	/* to store time in jiffies */
  	long interval_time[MAX_TIME];		/* to store thresholds */
 
fs/f2fs/super.c +0 -1
index 83bf9a02f83f2..5f4d9abc079d6 100644
@@ -3540,7 +3540,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
 
  	init_rwsem(&sbi->cp_rwsem);
  	init_rwsem(&sbi->quota_sem);
-	init_waitqueue_head(&sbi->cp_wait);
  	init_sb_info(sbi);
 
  	err = init_percpu_info(sbi);
author maintainer
#2 lore raw
Quoted 27 lines · Chao Yu
Quoted 26 lines · Jaegeuk Kim

It doesn't need to wait for checkpoint being completed triggered by end_io.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
---
  fs/f2fs/checkpoint.c | 6 +-----
  fs/f2fs/data.c       | 4 ----
  fs/f2fs/f2fs.h       | 1 -
  fs/f2fs/super.c      | 1 -
  4 files changed, 1 insertion(+), 11 deletions(-)
Quoted 13 lines

fs/f2fs/checkpoint.c static void unblock_operations(struct f2fs_sb_info *sbi) · line 1260

Quoted 11 lines
fs/f2fs/checkpoint.c +0 -3
index 99c8061da55b9..2bdddc725e677 100644
@@ -1255,11 +1255,7 @@ static void unblock_operations(struct f2fs_sb_info *sbi)
  void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
  {
-	DEFINE_WAIT(wait);
-
  	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);

Wouldn't that case high cpu usage before io end?

This is a critical path to wait for IO completion in checkpoint, which would be better to wait for it to avoid long latency to continue filesystem operations. Moreover, I expect io_schedule_timeout() can mitigate such the CPU overhead and actually we don't need to make there-in context switches as well.

Quoted 50 lines
Quoted 50 lines
fs/f2fs/checkpoint.c +1 -2
-
  		if (!get_pages(sbi, type))
  			break;
@@ -1271,9 +1267,9 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
  							FS_CP_META_IO);
  		else if (type == F2FS_WB_CP_DATA)
  			f2fs_submit_merged_write(sbi, DATA);
+
  		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
  	}
-	finish_wait(&sbi->cp_wait, &wait);
  }
  static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
fs/f2fs/data.c +0 -4
index c1b676be67b9a..588d2871bbe69 100644
@@ -383,10 +383,6 @@ static void f2fs_write_end_io(struct bio *bio)
  		clear_cold_data(page);
  		end_page_writeback(page);
  	}
-	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
-				wq_has_sleeper(&sbi->cp_wait))
-		wake_up(&sbi->cp_wait);
-
  	bio_put(bio);
  }
fs/f2fs/f2fs.h +0 -1
index 02811ce15059b..a036539363030 100644
@@ -1437,7 +1437,6 @@ struct f2fs_sb_info {
  	struct rw_semaphore cp_rwsem;		/* blocking FS operations */
  	struct rw_semaphore node_write;		/* locking node writes */
  	struct rw_semaphore node_change;	/* locking node change */
-	wait_queue_head_t cp_wait;
  	unsigned long last_time[MAX_TIME];	/* to store time in jiffies */
  	long interval_time[MAX_TIME];		/* to store thresholds */
fs/f2fs/super.c +0 -1
index 83bf9a02f83f2..5f4d9abc079d6 100644
@@ -3540,7 +3540,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
  	init_rwsem(&sbi->cp_rwsem);
  	init_rwsem(&sbi->quota_sem);
-	init_waitqueue_head(&sbi->cp_wait);
  	init_sb_info(sbi);
  	err = init_percpu_info(sbi);
 
#3 lore raw
Quoted 42 lines · Jaegeuk Kim
Quoted 27 lines · Chao Yu
Quoted 26 lines · Jaegeuk Kim

It doesn't need to wait for checkpoint being completed triggered by end_io.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
---
   fs/f2fs/checkpoint.c | 6 +-----
   fs/f2fs/data.c       | 4 ----
   fs/f2fs/f2fs.h       | 1 -
   fs/f2fs/super.c      | 1 -
   4 files changed, 1 insertion(+), 11 deletions(-)
Quoted 13 lines

fs/f2fs/checkpoint.c static void unblock_operations(struct f2fs_sb_info *sbi) · line 1260

Quoted 11 lines
fs/f2fs/checkpoint.c +0 -3
index 99c8061da55b9..2bdddc725e677 100644
@@ -1255,11 +1255,7 @@ static void unblock_operations(struct f2fs_sb_info *sbi)
   void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
   {
-	DEFINE_WAIT(wait);
-
   	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);

Wouldn't that case high cpu usage before io end?

Quoted 2 lines

This is a critical path to wait for IO completion in checkpoint, which would be better to wait for it to avoid long latency to continue filesystem operations.

Yup, in previous implementation, last end_io wakes up checkpoint() waiter, we didn't waste any time there.

Quoted 2 lines

Moreover, I expect io_schedule_timeout() can mitigate such the CPU overhead and actually we don't need to make there-in context switches as well.

Then io_schedule_timeout() in this loop may give CPU time slice to other thread until scheduler reselect checkpoint(), that would cause longer latency?

Quoted 53 lines
Quoted 43 lines
Quoted 43 lines
fs/f2fs/checkpoint.c +1 -2
-
   		if (!get_pages(sbi, type))
   			break;
@@ -1271,9 +1267,9 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
   							FS_CP_META_IO);
   		else if (type == F2FS_WB_CP_DATA)
   			f2fs_submit_merged_write(sbi, DATA);
+
   		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
   	}
-	finish_wait(&sbi->cp_wait, &wait);
   }
   static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
fs/f2fs/data.c +0 -4
index c1b676be67b9a..588d2871bbe69 100644
@@ -383,10 +383,6 @@ static void f2fs_write_end_io(struct bio *bio)
   		clear_cold_data(page);
   		end_page_writeback(page);
   	}
-	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
-				wq_has_sleeper(&sbi->cp_wait))
-		wake_up(&sbi->cp_wait);
-
   	bio_put(bio);
   }
fs/f2fs/f2fs.h +0 -1
index 02811ce15059b..a036539363030 100644
@@ -1437,7 +1437,6 @@ struct f2fs_sb_info {
   	struct rw_semaphore cp_rwsem;		/* blocking FS operations */
   	struct rw_semaphore node_write;		/* locking node writes */
   	struct rw_semaphore node_change;	/* locking node change */
-	wait_queue_head_t cp_wait;
   	unsigned long last_time[MAX_TIME];	/* to store time in jiffies */
   	long interval_time[MAX_TIME];		/* to store thresholds */
fs/f2fs/super.c +0 -1
index 83bf9a02f83f2..5f4d9abc079d6 100644
--- a/fs/f2fs/super.c
Quoted 8 lines

fs/f2fs/super.c static int f2fs_fill_super(struct super_block *sb, void *data, int silent) · line 3544

Quoted 8 lines

+++ b/fs/f2fs/super.c

fs/f2fs/super.c +0 -1
@@ -3540,7 +3540,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
   	init_rwsem(&sbi->cp_rwsem);
   	init_rwsem(&sbi->quota_sem);
-	init_waitqueue_head(&sbi->cp_wait);
   	init_sb_info(sbi);
   	err = init_percpu_info(sbi);
 

.

author maintainer
#4 lore raw
Quoted 46 lines · Chao Yu
Quoted 42 lines · Jaegeuk Kim
Quoted 27 lines · Chao Yu
Quoted 26 lines · Jaegeuk Kim

It doesn't need to wait for checkpoint being completed triggered by end_io.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
---
   fs/f2fs/checkpoint.c | 6 +-----
   fs/f2fs/data.c       | 4 ----
   fs/f2fs/f2fs.h       | 1 -
   fs/f2fs/super.c      | 1 -
   4 files changed, 1 insertion(+), 11 deletions(-)
Quoted 13 lines

fs/f2fs/checkpoint.c static void unblock_operations(struct f2fs_sb_info *sbi) · line 1260

Quoted 11 lines
fs/f2fs/checkpoint.c +0 -3
index 99c8061da55b9..2bdddc725e677 100644
@@ -1255,11 +1255,7 @@ static void unblock_operations(struct f2fs_sb_info *sbi)
   void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
   {
-	DEFINE_WAIT(wait);
-
   	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);

Wouldn't that case high cpu usage before io end?

Quoted 2 lines

This is a critical path to wait for IO completion in checkpoint, which would be better to wait for it to avoid long latency to continue filesystem operations.

Quoted 8 lines

Yup, in previous implementation, last end_io wakes up checkpoint() waiter, we didn't waste any time there.

Quoted 2 lines

Moreover, I expect io_schedule_timeout() can mitigate such the CPU overhead and actually we don't need to make there-in context switches as well.

Then io_schedule_timeout() in this loop may give CPU time slice to other thread until scheduler reselect checkpoint(), that would cause longer latency?

Hmm, how about this then?

From 4956afa1cedc019cabf6e8bff7bc48d3bcf7a3f5 Mon Sep 17 00:00:00 2001
From: Jaegeuk Kim <[email protected]>
Date: Mon, 3 Aug 2020 19:37:12 -0700
Subject: [PATCH] f2fs: prepare a waiter before entering io_schedule

This is to avoid sleep() in the waiter thread.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
---
 fs/f2fs/checkpoint.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/f2fs/checkpoint.c +2 -2
index 99c8061da55b9..ff807e14c8911 100644
@@ -1258,8 +1258,6 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
 	DEFINE_WAIT(wait);
 
 	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
-
 		if (!get_pages(sbi, type))
 			break;
 
@@ -1271,6 +1269,8 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
 							FS_CP_META_IO);
 		else if (type == F2FS_WB_CP_DATA)
 			f2fs_submit_merged_write(sbi, DATA);
+
+		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
 	}
 	finish_wait(&sbi->cp_wait, &wait);
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Reviewed-by
#5 lore raw
Quoted 80 lines · Jaegeuk Kim
Quoted 46 lines · Chao Yu
Quoted 42 lines · Jaegeuk Kim
Quoted 27 lines · Chao Yu
Quoted 26 lines · Jaegeuk Kim

It doesn't need to wait for checkpoint being completed triggered by end_io.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
---
    fs/f2fs/checkpoint.c | 6 +-----
    fs/f2fs/data.c       | 4 ----
    fs/f2fs/f2fs.h       | 1 -
    fs/f2fs/super.c      | 1 -
    4 files changed, 1 insertion(+), 11 deletions(-)
Quoted 13 lines

fs/f2fs/checkpoint.c static void unblock_operations(struct f2fs_sb_info *sbi) · line 1260

Quoted 11 lines
fs/f2fs/checkpoint.c +0 -3
index 99c8061da55b9..2bdddc725e677 100644
@@ -1255,11 +1255,7 @@ static void unblock_operations(struct f2fs_sb_info *sbi)
    void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
    {
-	DEFINE_WAIT(wait);
-
    	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);

Wouldn't that case high cpu usage before io end?

Quoted 2 lines

This is a critical path to wait for IO completion in checkpoint, which would be better to wait for it to avoid long latency to continue filesystem operations.

Quoted 8 lines

Yup, in previous implementation, last end_io wakes up checkpoint() waiter, we didn't waste any time there.

Quoted 2 lines

Moreover, I expect io_schedule_timeout() can mitigate such the CPU overhead and actually we don't need to make there-in context switches as well.

Then io_schedule_timeout() in this loop may give CPU time slice to other thread until scheduler reselect checkpoint(), that would cause longer latency?

Hmm, how about this then?

Quoted 1 line

From 4956afa1cedc019cabf6e8bff7bc48d3bcf7a3f5 Mon Sep 17 00:00:00 2001

From: Jaegeuk Kim <[email protected]>
Date: Mon, 3 Aug 2020 19:37:12 -0700
Subject: [PATCH] f2fs: prepare a waiter before entering io_schedule

This is to avoid sleep() in the waiter thread.

[   20.157753] ------------[ cut here ]------------
[   20.158393] do not call blocking ops when !TASK_RUNNING; state=2 set at [<0000000096354225>] prepare_to_wait+0xcd/0x430
[   20.159858] WARNING: CPU: 1 PID: 1152 at kernel/sched/core.c:7142 __might_sleep+0x149/0x1a0
...
[   20.176110]  __submit_merged_write_cond+0x191/0x310
[   20.176739]  f2fs_submit_merged_write+0x18/0x20
[   20.177323]  f2fs_wait_on_all_pages+0x269/0x2d0
[   20.177899]  ? block_operations+0x980/0x980
[   20.178441]  ? __kasan_check_read+0x11/0x20
[   20.178975]  ? finish_wait+0x260/0x260
[   20.179488]  ? percpu_counter_set+0x147/0x230
[   20.180049]  do_checkpoint+0x1757/0x2a50
[   20.180558]  f2fs_write_checkpoint+0x840/0xaf0
[   20.181126]  f2fs_sync_fs+0x287/0x4a0
Quoted 2 lines

Better,

Thanks,

Quoted 27 lines
---
  fs/f2fs/checkpoint.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
fs/f2fs/checkpoint.c +2 -2
index 99c8061da55b9..ff807e14c8911 100644
@@ -1258,8 +1258,6 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
  	DEFINE_WAIT(wait);
 
  	for (;;) {
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
-
  		if (!get_pages(sbi, type))
  			break;
 
@@ -1271,6 +1269,8 @@ void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
  							FS_CP_META_IO);
  		else if (type == F2FS_WB_CP_DATA)
  			f2fs_submit_merged_write(sbi, DATA);
+
+		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
  		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
  	}
  	finish_wait(&sbi->cp_wait, &wait);