The Coverity checker spotted the following inconsequent NULL checking
introduced by commit 8ff12cfc009a2a38d87fa7058226fe197bb2696f:
<-- snip -->
...
static inline int is_end(void *addr)
{
return (unsigned long)addr & PAGE_MAPPING_ANON;
}
...
static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
{
...
if (c->freelist) <----------------------------------------
stat(c, DEACTIVATE_REMOTE_FREES);
/*
* Merge cpu freelist into freelist. Typically we get here
* because both freelists are empty. So this is unlikely
* to occur.
*
* We need to use _is_end here because deactivate slab may
* be called for a debug slab. Then c->freelist may contain
* a dummy pointer.
*/
while (unlikely(!is_end(c->freelist))) {
void **object;
tail = 0; /* Hot objects. Put the slab first */
/* Retrieve object from cpu_freelist */
object = c->freelist;
c->freelist = c->freelist[c->offset];
... ^^^^^^^^^^^^^^^^^^^^^^
<-- snip -->
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
Hi Adrian,
On 2/20/2008, "Adrian Bunk" <[email protected]> wrote:
> The Coverity checker spotted the following inconsequent NULL checking
> introduced by commit 8ff12cfc009a2a38d87fa7058226fe197bb2696f:
>
> <-- snip -->
>
> ...
> static inline int is_end(void *addr)
> {
> return (unsigned long)addr & PAGE_MAPPING_ANON;
> }
> ...
> static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
> {
> ...
> if (c->freelist) <----------------------------------------
> stat(c, DEACTIVATE_REMOTE_FREES);
I spotted this too. c->freelist should never be NULL so why not send a
patch to Christoph?
On Wed, Feb 20, 2008 at 03:52:44PM +0200, Pekka Enberg wrote:
>
> Hi Adrian,
>
> On 2/20/2008, "Adrian Bunk" <[email protected]> wrote:
> > The Coverity checker spotted the following inconsequent NULL checking
> > introduced by commit 8ff12cfc009a2a38d87fa7058226fe197bb2696f:
> >
> > <-- snip -->
> >
> > ...
> > static inline int is_end(void *addr)
> > {
> > return (unsigned long)addr & PAGE_MAPPING_ANON;
> > }
> > ...
> > static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
> > {
> > ...
> > if (c->freelist) <----------------------------------------
> > stat(c, DEACTIVATE_REMOTE_FREES);
>
> I spotted this too.
I missed that.
> c->freelist should never be NULL so why not send a
> patch to Christoph?
Patch below.
cu
Adrian
<-- snip -->
There's no reason for checking c->freelist for being NULL here (and we'd
anyway Oops below if it was).
Signed-off-by: Adrian Bunk <[email protected]>
---
dae2a3c60f258f3ad2522b85d79b735a89d702f0 diff --git a/mm/slub.c b/mm/slub.c
index 74c65af..072e0a6 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1404,8 +1404,7 @@ static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
struct page *page = c->page;
int tail = 1;
- if (c->freelist)
- stat(c, DEACTIVATE_REMOTE_FREES);
+ stat(c, DEACTIVATE_REMOTE_FREES);
/*
* Merge cpu freelist into freelist. Typically we get here
* because both freelists are empty. So this is unlikely
Adrian Bunk wrote:
> There's no reason for checking c->freelist for being NULL here (and we'd
> anyway Oops below if it was).
>
> Signed-off-by: Adrian Bunk <[email protected]>
>
> ---
> dae2a3c60f258f3ad2522b85d79b735a89d702f0 diff --git a/mm/slub.c b/mm/slub.c
> index 74c65af..072e0a6 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -1404,8 +1404,7 @@ static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
> struct page *page = c->page;
> int tail = 1;
>
> - if (c->freelist)
> - stat(c, DEACTIVATE_REMOTE_FREES);
> + stat(c, DEACTIVATE_REMOTE_FREES);
> /*
> * Merge cpu freelist into freelist. Typically we get here
> * because both freelists are empty. So this is unlikely
Christoph, please apply.
Reviewed-by: Pekka Enberg <[email protected]>
On Fri, 22 Feb 2008, Adrian Bunk wrote:
> There's no reason for checking c->freelist for being NULL here (and we'd
> anyway Oops below if it was).
Well we still need to check for the freelist being empty otherwise the
counter for remote frees does not work as intended. The check was
introduced at the time when page->end did not yet exist. At that time the
NULL check made sense.
From: Christoph Lameter <[email protected]>
Subject: Fix check for remote frees
The check for remote frees must check is_end() instead of != NULL.
We execute the !is_end() section rarely so move the check in there. Just do it
once by relying on tail being 1 only the first time we enter the loop.
Signed-off-by: Christoph Lameter <[email protected]>
---
dae2a3c60f258f3ad2522b85d79b735a89d702f0 diff --git a/mm/slub.c b/mm/slub.c
index 74c65af..072e0a6 100644
---
mm/slub.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c 2008-02-27 11:48:11.000000000 -0800
+++ linux-2.6/mm/slub.c 2008-02-27 11:51:07.000000000 -0800
@@ -1404,8 +1404,6 @@ static void deactivate_slab(struct kmem_
struct page *page = c->page;
int tail = 1;
- if (c->freelist)
- stat(c, DEACTIVATE_REMOTE_FREES);
/*
* Merge cpu freelist into freelist. Typically we get here
* because both freelists are empty. So this is unlikely
@@ -1418,6 +1416,8 @@ static void deactivate_slab(struct kmem_
while (unlikely(!is_end(c->freelist))) {
void **object;
+ if (unlikely(tail))
+ stat(c, DEACTIVATE_REMOTE_FREES);
tail = 0; /* Hot objects. Put the slab first */
/* Retrieve object from cpu_freelist */