Skip to contentSkip to conversation

ext4: add cond_resched() to ext4_mb_init_backend()

PATCH In review Khazhismel Kumykov 4 participants 6 messages Apr 16, 2019 – Apr 25, 2019 ext4 file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ext4: add cond_resched() to ext4_mb_init_backend()

on non-preempt kernels for filesystems with large number of groups we may take a long time (>50 ticks) initializing all the groups.

---
 fs/ext4/mballoc.c | 1 +
 1 file changed, 1 insertion(+)
fs/ext4/mballoc.c +1 -0
index 8ef5f12bbee2..c89f497ccf50 100644
@@ -2490,6 +2490,7 @@ static int ext4_mb_init_backend(struct super_block *sb)
 	sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
 	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
 	for (i = 0; i < ngroups; i++) {
+		cond_resched();
 		desc = ext4_get_group_desc(sb, i, NULL);
 		if (desc == NULL) {
 			ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
reviewer
#1 lore raw
Quoted 4 lines · Khazhismel Kumykov

on non-preempt kernels for filesystems with large number of groups we may take a long time (>50 ticks) initializing all the groups.

Thanks for the patch. I'm not opposed to this but I'm just wondering: Has this caused any real issues to you? Or how have you come across this particular loop? Because I'd think we have more similar loops in ext4 codebase...

Honza

Quoted 19 lines
---
 fs/ext4/mballoc.c | 1 +
 1 file changed, 1 insertion(+)
fs/ext4/mballoc.c +1 -0
index 8ef5f12bbee2..c89f497ccf50 100644
@@ -2490,6 +2490,7 @@ static int ext4_mb_init_backend(struct super_block *sb)
 	sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
 	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
 	for (i = 0; i < ngroups; i++) {
+		cond_resched();
 		desc = ext4_get_group_desc(sb, i, NULL);
 		if (desc == NULL) {
 			ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
Signature · 1 line
2.21.0.392.gf8f6787159e-goog
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#2 lore raw
Quoted 5 lines · Jan Kara
Quoted 3 lines · Khazhismel Kumykov

on non-preempt kernels for filesystems with large number of groups we may take a long time (>50 ticks) initializing all the groups.

Quoted 9 lines
Quoted 1 line

Thanks for the patch. I'm not opposed to this but I'm just wondering: Has this caused any real issues to you? Or how have you come across this particular loop? Because I'd think we have more similar loops in ext4 codebase...

Honza

We have some instrumentation to warn for these longer periods without scheduling, which admittedly does trigger relatively frequently from various places. This particular loop just so happens to be during system startup where it caught my eye.

#3 lore raw

PATCH v2 1/1 ext4: cond_resched in work-heavy group loops

These functions may take a long time looping over many groups, which may cause issues for non-preempt kernels. ext4_mb_init_backend()
ext4_setup_system_zone()
ext4_mb_release()

---
v2:
- a few other places that in testing showed to be slow

 fs/ext4/block_validity.c | 1 +
 fs/ext4/mballoc.c        | 2 ++
 2 files changed, 3 insertions(+)
fs/ext4/block_validity.c +1 -0
index 913061c0de1b..16134469ea3c 100644
@@ -155,6 +155,7 @@ int ext4_setup_system_zone(struct super_block *sb)
 		return 0;
 
 	for (i=0; i < ngroups; i++) {
+		cond_resched();
 		if (ext4_bg_has_super(sb, i) &&
 		    ((i < 5) || ((i % flex_size) == 0)))
 			add_system_zone(sbi, ext4_group_first_block_no(sb, i),
fs/ext4/mballoc.c +2 -0
index 8ef5f12bbee2..99ba720dbb7a 100644
@@ -2490,6 +2490,7 @@ static int ext4_mb_init_backend(struct super_block *sb)
 	sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
 	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
 	for (i = 0; i < ngroups; i++) {
+		cond_resched();
 		desc = ext4_get_group_desc(sb, i, NULL);
 		if (desc == NULL) {
 			ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
@@ -2705,6 +2706,7 @@ int ext4_mb_release(struct super_block *sb)
 
 	if (sbi->s_group_info) {
 		for (i = 0; i < ngroups; i++) {
+			cond_resched();
 			grinfo = ext4_get_group_info(sb, i);
 #ifdef DOUBLE_CHECK
 			kfree(grinfo->bb_bitmap);
Signature · 1 line
2.21.0.593.g511ec345e18-goog
Reviewed-by on v2 1/1
#4 lore raw

Re: [PATCH v2] ext4: cond_resched in work-heavy group loops

Quoted 8 lines · Khazhismel Kumykov

These functions may take a long time looping over many groups, which may cause issues for non-preempt kernels. ext4_mb_init_backend()
ext4_setup_system_zone()
ext4_mb_release()

Quoted 41 lines

---
v2:
- a few other places that in testing showed to be slow

fs/ext4/block_validity.c | 1 +
fs/ext4/mballoc.c | 2 ++

2 files changed, 3 insertions(+)
fs/ext4/block_validity.c +1 -0
index 913061c0de1b..16134469ea3c 100644
@@ -155,6 +155,7 @@ int ext4_setup_system_zone(struct super_block *sb)
 		return 0;
 
 	for (i=0; i < ngroups; i++) {
+		cond_resched();
 		if (ext4_bg_has_super(sb, i) &&
 		    ((i < 5) || ((i % flex_size) == 0)))
 			add_system_zone(sbi, ext4_group_first_block_no(sb, i),
fs/ext4/mballoc.c +2 -1
index 8ef5f12bbee2..99ba720dbb7a 100644
@@ -2490,6 +2490,7 @@ static int ext4_mb_init_backend(struct super_block *sb)
 	sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
 	EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
 	for (i = 0; i < ngroups; i++) {
+		cond_resched();
 		desc = ext4_get_group_desc(sb, i, NULL);
 		if (desc == NULL) {
 			ext4_msg(sb, KERN_ERR, "can't read descriptor %u", i);
@@ -2705,6 +2706,7 @@ int ext4_mb_release(struct super_block *sb)
 
 	if (sbi->s_group_info) {
 		for (i = 0; i < ngroups; i++) {
+			cond_resched();
 			grinfo = ext4_get_group_info(sb, i);
 #ifdef DOUBLE_CHECK
 			kfree(grinfo->bb_bitmap);
--
Quoted 2 lines

2.21.0.593.g511ec345e18-goog

Cheers, Andreas