2013-07-17 14:30:15

by Heesub Shin

Subject: [PATCH] mm: zbud: fix condition check on allocation size

zbud_alloc() incorrectly verifies the size of allocation limit. It
should deny the allocation request greater than (PAGE_SIZE -
ZHDR_SIZE_ALIGNED - CHUNK_SIZE), not (PAGE_SIZE - ZHDR_SIZE_ALIGNED)
which has no remaining spaces for its buddy. There is no point in
spending the entire zbud page storing only a single page, since we don't
have any benefits.

Signed-off-by: Heesub Shin <[email protected]>
---
mm/zbud.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/zbud.c b/mm/zbud.c
index 9bb4710..ad1e781 100644
--- a/mm/zbud.c
+++ b/mm/zbud.c
@@ -257,7 +257,7 @@ int zbud_alloc(struct zbud_pool *pool, int size, gfp_t gfp,

if (size <= 0 || gfp & __GFP_HIGHMEM)
return -EINVAL;
- if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED)
+ if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
return -ENOSPC;
chunks = size_to_chunks(size);
spin_lock(&pool->lock);
--
1.8.3.2

2013-07-18 12:48:17

by Bob Liu

Subject: Re: [PATCH] mm: zbud: fix condition check on allocation size

On 07/17/2013 10:30 PM, Heesub Shin wrote:
> zbud_alloc() incorrectly verifies the size of allocation limit. It
> should deny the allocation request greater than (PAGE_SIZE -
> ZHDR_SIZE_ALIGNED - CHUNK_SIZE), not (PAGE_SIZE - ZHDR_SIZE_ALIGNED)
> which has no remaining spaces for its buddy. There is no point in
> spending the entire zbud page storing only a single page, since we don't
> have any benefits.
>
> Signed-off-by: Heesub Shin <[email protected]>

Looks good to me, although I'm thinking to make it more aggressive.
eg. minus two or three times of CHUNK_SIZE.

> ---
> mm/zbud.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/zbud.c b/mm/zbud.c
> index 9bb4710..ad1e781 100644
> --- a/mm/zbud.c
> +++ b/mm/zbud.c
> @@ -257,7 +257,7 @@ int zbud_alloc(struct zbud_pool *pool, int size, gfp_t gfp,
>
> if (size <= 0 || gfp & __GFP_HIGHMEM)
> return -EINVAL;
> - if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED)
> + if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
> return -ENOSPC;
> chunks = size_to_chunks(size);
> spin_lock(&pool->lock);
>

--
Regards,
-Bob