[PATCH] lib/sbitmap: allocate sb->map via kvzalloc_node

[PATCH] lib/sbitmap: allocate sb->map via kvzalloc_node

[Ming Lei]

sbitmap has been used in scsi for replacing atomic operations on
sdev->device_busy, so IOPS on some fast scsi storage can be improved.

However, sdev->device_busy can be changed in fast path, so we have to
allocate the sb->map statically. sdev->device_busy has been capped to 1024,
but some drivers may configure the default depth as < 8, then
cause each sbitmap word to hold only one bit. Finally 1024 * 128(
sizeof(sbitmap_word)) bytes is needed for sb->map, given it is order 5
allocation, sometimes it may fail.

Avoid the issue by using kvzalloc_node() for allocating sb->map.

Cc: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 include/linux/sbitmap.h | 2 +-
 lib/sbitmap.c           | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/sbitmap.h b/include/linux/sbitmap.h
index dffeb8281c2d..8f5a86e210b9 100644
--- a/include/linux/sbitmap.h
+++ b/include/linux/sbitmap.h
@@ -174,7 +174,7 @@ static inline unsigned int __map_depth(const struct sbitmap *sb, int index)
 static inline void sbitmap_free(struct sbitmap *sb)
 {
 	free_percpu(sb->alloc_hint);
-	kfree(sb->map);
+	kvfree(sb->map);
 	sb->map = NULL;
 }
 
diff --git a/lib/sbitmap.c b/lib/sbitmap.c
index 2eb3de18ded3..ae4fd4de9ebe 100644
--- a/lib/sbitmap.c
+++ b/lib/sbitmap.c
@@ -110,7 +110,7 @@ int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
 		sb->alloc_hint = NULL;
 	}
 
-	sb->map = kcalloc_node(sb->map_nr, sizeof(*sb->map), flags, node);
+	sb->map = kvzalloc_node(sb->map_nr * sizeof(*sb->map), flags, node);
 	if (!sb->map) {
 		free_percpu(sb->alloc_hint);
 		return -ENOMEM;
-- 
2.31.1

Re: [PATCH] lib/sbitmap: allocate sb->map via kvzalloc_node

[Chaitanya Kulkarni]

On 3/15/22 18:27, Ming Lei wrote:
> sbitmap has been used in scsi for replacing atomic operations on
> sdev->device_busy, so IOPS on some fast scsi storage can be improved.
> 
> However, sdev->device_busy can be changed in fast path, so we have to
> allocate the sb->map statically. sdev->device_busy has been capped to 1024,
> but some drivers may configure the default depth as < 8, then
> cause each sbitmap word to hold only one bit. Finally 1024 * 128(
> sizeof(sbitmap_word)) bytes is needed for sb->map, given it is order 5
> allocation, sometimes it may fail.
> 
> Avoid the issue by using kvzalloc_node() for allocating sb->map.
> 
> Cc: Ewan D. Milne <emilne@redhat.com>
> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> ---
>

maybe considering the total size before actual allocation
that will not buy us anything ?

Looks good.

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

-ck

Re: [PATCH] lib/sbitmap: allocate sb->map via kvzalloc_node

[Ming Lei]

Hello,

On Wed, Mar 16, 2022 at 09:27:08AM +0800, Ming Lei wrote:
> sbitmap has been used in scsi for replacing atomic operations on
> sdev->device_busy, so IOPS on some fast scsi storage can be improved.
> 
> However, sdev->device_busy can be changed in fast path, so we have to
> allocate the sb->map statically. sdev->device_busy has been capped to 1024,
> but some drivers may configure the default depth as < 8, then
> cause each sbitmap word to hold only one bit. Finally 1024 * 128(
> sizeof(sbitmap_word)) bytes is needed for sb->map, given it is order 5
> allocation, sometimes it may fail.
> 
> Avoid the issue by using kvzalloc_node() for allocating sb->map.
> 
> Cc: Ewan D. Milne <emilne@redhat.com>
> Signed-off-by: Ming Lei <ming.lei@redhat.com>

Ping...


thanks,
Ming

Re: [PATCH] lib/sbitmap: allocate sb->map via kvzalloc_node

[Jens Axboe]

On 3/21/22 6:25 PM, Ming Lei wrote:
> Hello,
> 
> On Wed, Mar 16, 2022 at 09:27:08AM +0800, Ming Lei wrote:
>> sbitmap has been used in scsi for replacing atomic operations on
>> sdev->device_busy, so IOPS on some fast scsi storage can be improved.
>>
>> However, sdev->device_busy can be changed in fast path, so we have to
>> allocate the sb->map statically. sdev->device_busy has been capped to 1024,
>> but some drivers may configure the default depth as < 8, then
>> cause each sbitmap word to hold only one bit. Finally 1024 * 128(
>> sizeof(sbitmap_word)) bytes is needed for sb->map, given it is order 5
>> allocation, sometimes it may fail.
>>
>> Avoid the issue by using kvzalloc_node() for allocating sb->map.
>>
>> Cc: Ewan D. Milne <emilne@redhat.com>
>> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> 
> Ping...

Was waiting on a response to the question on why we don't make this
dependent on size, but upon look at kvzalloc_node(), this is already
what it does. So I guess this looks fine to me too.

-- 
Jens Axboe

Re: [PATCH] lib/sbitmap: allocate sb->map via kvzalloc_node

[Jens Axboe]

On Wed, 16 Mar 2022 09:27:08 +0800, Ming Lei wrote:
> sbitmap has been used in scsi for replacing atomic operations on
> sdev->device_busy, so IOPS on some fast scsi storage can be improved.
> 
> However, sdev->device_busy can be changed in fast path, so we have to
> allocate the sb->map statically. sdev->device_busy has been capped to 1024,
> but some drivers may configure the default depth as < 8, then
> cause each sbitmap word to hold only one bit. Finally 1024 * 128(
> sizeof(sbitmap_word)) bytes is needed for sb->map, given it is order 5
> allocation, sometimes it may fail.
> 
> [...]

Applied, thanks!

[1/1] lib/sbitmap: allocate sb->map via kvzalloc_node
      commit: 863a66cdb4df25fd146d9851c3289072298566d5

Best regards,
-- 
Jens Axboe