Btrfs devel - Feb 2013 - [PATCH v2] btrfs: use kmalloc for lzo de/compress buffer

The size of de/compress buffer is small enough to use kmalloc.
Allocating it with kmalloc rather than vmalloc is preferred.

This patch depends on my previous patch, “btrfs: fix decompress buffer size”.

v2: Using vmalloc for "workspace->mem" due to the size limit.

Signed-off-by: Kyungsik Lee <kyungsik.lee@lge.com>
Cc: David Sterba <dsterba@suse.cz>
---
 fs/btrfs/lzo.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/fs/btrfs/lzo.c b/fs/btrfs/lzo.c
index 223893a..aaaab2e 100644
--- a/fs/btrfs/lzo.c
+++ b/fs/btrfs/lzo.c
@@ -40,8 +40,8 @@ static void lzo_free_workspace(struct list_head *ws)
 {
 	struct workspace *workspace = list_entry(ws, struct workspace, list);
 
-	vfree(workspace->buf);
-	vfree(workspace->cbuf);
+	kfree(workspace->buf);
+	kfree(workspace->cbuf);
 	vfree(workspace->mem);
 	kfree(workspace);
 }
@@ -55,8 +55,9 @@ static struct list_head *lzo_alloc_workspace(void)
 		return ERR_PTR(-ENOMEM);
 
 	workspace->mem = vmalloc(LZO1X_MEM_COMPRESS);
-	workspace->buf = vmalloc(PAGE_CACHE_SIZE);
-	workspace->cbuf = vmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE));
+	workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
+	workspace->cbuf = kmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE),
+			GFP_NOFS);
 	if (!workspace->mem || !workspace->buf || !workspace->cbuf)
 		goto fail;
 
-- 
1.8.0.3

On Mon, Feb 18, 2013 at 04:56:04PM +0900, Kyungsik Lee
wrote:
> @@ -55,8 +55,9 @@ static struct list_head *lzo_alloc_workspace(void)
>  		return ERR_PTR(-ENOMEM);
>  
>  	workspace->mem = vmalloc(LZO1X_MEM_COMPRESS);
> -	workspace->buf = vmalloc(PAGE_CACHE_SIZE);
> -	workspace->cbuf = vmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE));
> +	workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
> +	workspace->cbuf = kmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE),
This is still larger than usual page size and allocator may issue a
warning, so if we want to use kmalloc (it''s ~4.4k in size and likely to
be available, but not always due to possible page fragmentation), the
__GFP_NOWARN should be supplied and if the allocation fails fall back to
vmalloc.

david

On Thu, Feb 21, 2013 at 02:41:53PM +0100, David Sterba
wrote:
> On Mon, Feb 18, 2013 at 04:56:04PM +0900, Kyungsik Lee wrote:
> > @@ -55,8 +55,9 @@ static struct list_head *lzo_alloc_workspace(void)
> >  		return ERR_PTR(-ENOMEM);
> >  
> >  	workspace->mem = vmalloc(LZO1X_MEM_COMPRESS);
> > -	workspace->buf = vmalloc(PAGE_CACHE_SIZE);
> > -	workspace->cbuf = vmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE));
> > +	workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
> > +	workspace->cbuf = kmalloc(lzo1x_worst_compress(PAGE_CACHE_SIZE),
> 
> This is still larger than usual page size and allocator may issue a
> warning, so if we want to use kmalloc (it''s ~4.4k in size and
likely to
> be available, but not always due to possible page fragmentation), the
> __GFP_NOWARN should be supplied and if the allocation fails fall back to
> vmalloc.
Right, there shoud be another fixed verson for this.

Thanks,
Kyungsik