Skip to contentSkip to conversation

ext4: replace kmem_cache_create with KMEM_CACHE

PATCH In review JunChao Sun 3 participants 7 messages Nov 9, 2022 – Nov 11, 2022 ext4 file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ext4: replace kmem_cache_create with KMEM_CACHE

Replace kmem_cache_create with KMEM_CACHE macro that guaranteed struct alignment

---
 fs/ext4/extents_status.c | 8 ++------
 fs/ext4/readpage.c       | 5 ++---
 2 files changed, 4 insertions(+), 9 deletions(-)
fs/ext4/extents_status.c +2 -6
index cd0a861853e3..97eccc0028a1 100644
@@ -155,9 +155,7 @@ static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
 
 int __init ext4_init_es(void)
 {
-	ext4_es_cachep = kmem_cache_create("ext4_extent_status",
-					   sizeof(struct extent_status),
-					   0, (SLAB_RECLAIM_ACCOUNT), NULL);
+	ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
 	if (ext4_es_cachep == NULL)
 		return -ENOMEM;
 	return 0;
@@ -1807,9 +1805,7 @@ static void ext4_print_pending_tree(struct inode *inode)
 
 int __init ext4_init_pending(void)
 {
-	ext4_pending_cachep = kmem_cache_create("ext4_pending_reservation",
-					   sizeof(struct pending_reservation),
-					   0, (SLAB_RECLAIM_ACCOUNT), NULL);
+	ext4_pending_cachep = KMEM_CACHE(pending_reservation, SLAB_RECLAIM_ACCOUNT);
 	if (ext4_pending_cachep == NULL)
 		return -ENOMEM;
 	return 0;
fs/ext4/readpage.c +2 -3
index 3d21eae267fc..773176e7f9f5 100644
@@ -410,9 +410,8 @@ int ext4_mpage_readpages(struct inode *inode,
 
 int __init ext4_init_post_read_processing(void)
 {
-	bio_post_read_ctx_cache =
-		kmem_cache_create("ext4_bio_post_read_ctx",
-				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
+	bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT);
+
 	if (!bio_post_read_ctx_cache)
 		goto fail;
 	bio_post_read_ctx_pool =
Signature · 1 line
2.17.1
reviewer Reviewed-by on v1 1/1
#1 lore raw
Quoted 4 lines · JunChao Sun

Replace kmem_cache_create with KMEM_CACHE macro that guaranteed struct alignment

Yeah, nice cleanups. Feel free to add:

Honza

Quoted 50 lines
---
 fs/ext4/extents_status.c | 8 ++------
 fs/ext4/readpage.c       | 5 ++---
 2 files changed, 4 insertions(+), 9 deletions(-)
fs/ext4/extents_status.c +2 -6
index cd0a861853e3..97eccc0028a1 100644
@@ -155,9 +155,7 @@ static void __revise_pending(struct inode *inode, ext4_lblk_t lblk,
 
 int __init ext4_init_es(void)
 {
-	ext4_es_cachep = kmem_cache_create("ext4_extent_status",
-					   sizeof(struct extent_status),
-					   0, (SLAB_RECLAIM_ACCOUNT), NULL);
+	ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
 	if (ext4_es_cachep == NULL)
 		return -ENOMEM;
 	return 0;
@@ -1807,9 +1805,7 @@ static void ext4_print_pending_tree(struct inode *inode)
 
 int __init ext4_init_pending(void)
 {
-	ext4_pending_cachep = kmem_cache_create("ext4_pending_reservation",
-					   sizeof(struct pending_reservation),
-					   0, (SLAB_RECLAIM_ACCOUNT), NULL);
+	ext4_pending_cachep = KMEM_CACHE(pending_reservation, SLAB_RECLAIM_ACCOUNT);
 	if (ext4_pending_cachep == NULL)
 		return -ENOMEM;
 	return 0;
fs/ext4/readpage.c +2 -3
index 3d21eae267fc..773176e7f9f5 100644
@@ -410,9 +410,8 @@ int ext4_mpage_readpages(struct inode *inode,
 
 int __init ext4_init_post_read_processing(void)
 {
-	bio_post_read_ctx_cache =
-		kmem_cache_create("ext4_bio_post_read_ctx",
-				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
+	bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT);
+
 	if (!bio_post_read_ctx_cache)
 		goto fail;
 	bio_post_read_ctx_pool =
Signature · 1 line
2.17.1
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#2 lore raw
In reply to JunChao Sun

fs/ext4/readpage.c int ext4_mpage_readpages(struct inode *inode, · line 414

Quoted 13 lines · JunChao Sun
fs/ext4/readpage.c +2 -3
index 3d21eae267fc..773176e7f9f5 100644
@@ -410,9 +410,8 @@ int ext4_mpage_readpages(struct inode *inode,
 
 int __init ext4_init_post_read_processing(void)
 {
-	bio_post_read_ctx_cache =
-		kmem_cache_create("ext4_bio_post_read_ctx",
-				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
+	bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT);
+

Why use SLAB_RECLAIM_ACCOUNT here?

- Eric

#3 lore raw

Yeah, maybe we should remove the SLAB_RECLAIM_ACCOUNT flag for static slab, and 16828088f9e51815 ("ext4: use KMEM_CACHE instead of kmem_cache_create") have done so. But should we remove SLAB_RECLAIM_ACCOUNT in this patch or belong to a separate patch?

Quoted 19 lines · Eric Biggers <[email protected]> 于2022年11月10日周四 02:00写道:

fs/ext4/readpage.c int ext4_mpage_readpages(struct inode *inode, · line 414

Quoted 13 lines · JunChao Sun
fs/ext4/readpage.c +2 -3
index 3d21eae267fc..773176e7f9f5 100644
@@ -410,9 +410,8 @@ int ext4_mpage_readpages(struct inode *inode,
 
 int __init ext4_init_post_read_processing(void)
 {
-     bio_post_read_ctx_cache =
-             kmem_cache_create("ext4_bio_post_read_ctx",
-                               sizeof(struct bio_post_read_ctx), 0, 0, NULL);
+     bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT);
+

Why use SLAB_RECLAIM_ACCOUNT here?

- Eric

#4 lore raw
Quoted 4 lines · JunChao Sun

Yeah, maybe we should remove the SLAB_RECLAIM_ACCOUNT flag for static slab, and 16828088f9e51815 ("ext4: use KMEM_CACHE instead of kmem_cache_create") have done so. But should we remove SLAB_RECLAIM_ACCOUNT in this patch or belong to a separate patch?

I'd just keep the slab flags the same in this patch. If any flags do need to be changed, that should be a separate patch.

I think SLAB_RECLAIM_ACCOUNT is meant for for things that are directly reclaimable, such as struct ext4_inode_info. Inodes are evictable, and when that happens, the corresponding struct ext4_inode_info gets freed.

bio_post_read_ctx_cache probably should use SLAB_TEMPORARY instead, since it is only used for temporary structures during I/O.

That being said, SLAB_TEMPORARY is currently #define'd to SLAB_RECLAIM_ACCOUNT, so currently it makes no difference in practice...

- Eric

#5 lore raw
Quoted 15 lines · Eric Biggers <[email protected]> 于2022年11月10日周四 10:47写道:
Quoted 13 lines · JunChao Sun

Yeah, maybe we should remove the SLAB_RECLAIM_ACCOUNT flag for static slab, and 16828088f9e51815 ("ext4: use KMEM_CACHE instead of kmem_cache_create") have done so. But should we remove SLAB_RECLAIM_ACCOUNT in this patch or belong to a separate patch?

I'd just keep the slab flags the same in this patch. If any flags do need to be changed, that should be a separate patch.

I think SLAB_RECLAIM_ACCOUNT is meant for for things that are directly reclaimable, such as struct ext4_inode_info. Inodes are evictable, and when that happens, the corresponding struct ext4_inode_info gets freed.

Quoted 5 lines
Quoted 5 lines

bio_post_read_ctx_cache probably should use SLAB_TEMPORARY instead, since it is only used for temporary structures during I/O.

That being said, SLAB_TEMPORARY is currently #define'd to SLAB_RECLAIM_ACCOUNT, so currently it makes no difference in practice...

Thanks for clarifying. I will send a separate patch to remove SLAB_RECLAIM_ACCOUNT.

Quoted 2 lines

- Eric

#6 lore raw
In reply to Eric Biggers
Quoted 17 lines · Eric Biggers <[email protected]> 于2022年11月10日周四 10:47写道:
Quoted 4 lines · JunChao Sun

Yeah, maybe we should remove the SLAB_RECLAIM_ACCOUNT flag for static slab, and 16828088f9e51815 ("ext4: use KMEM_CACHE instead of kmem_cache_create") have done so. But should we remove SLAB_RECLAIM_ACCOUNT in this patch or belong to a separate patch?

I'd just keep the slab flags the same in this patch. If any flags do need to be changed, that should be a separate patch.

Quoted 6 lines

I think SLAB_RECLAIM_ACCOUNT is meant for for things that are directly reclaimable, such as struct ext4_inode_info. Inodes are evictable, and when that happens, the corresponding struct ext4_inode_info gets freed.

bio_post_read_ctx_cache probably should use SLAB_TEMPORARY instead, since it is only used for temporary structures during I/O.

How to decide whether to use SLAB_RECLAIM_ACCOUNT or not when creating a slab cache? Is it based on whether the object is reclaimable or evictable, or the amount of memory the object may use? If the first, how to know whether an object is reclaimable or evictable? Btw, I saw that ext4 called ext4_es_register_shrinker() to register shrinker for struct extent_status, so SLAB_RECLAIM_ACCOUNT should be used when creating ext4_es_cachep, right?

Quoted 5 lines

That being said, SLAB_TEMPORARY is currently #define'd to SLAB_RECLAIM_ACCOUNT, so currently it makes no difference in practice...

- Eric