Skip to contentSkip to conversation

Do not pass the SLAB flags as GFP in slob kmem_cache_create

PATCH Applied Catalin Marinas 5 participants 8 messages Dec 10, 2008 – Dec 11, 2008 Memory management

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Do not pass the SLAB flags as GFP in slob kmem_cache_create

It looks like the kmem_cache_create() function in the slob allocator passes the SLAB flags as GFP flags to the slob_alloc() function. The patch changes this call to pass GFP_KERNEL as the other allocators seem to do.

---
 mm/slob.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
mm/slob.c +1 -1
index ff5a98d..dce9258 100644
@@ -538,7 +538,7 @@ struct kmem_cache *kmem_cache_create(const char *name, size_t size,
 	struct kmem_cache *c;
 
 	c = slob_alloc(sizeof(struct kmem_cache),
-		flags, ARCH_KMALLOC_MINALIGN, -1);
+		GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
 
 	if (c) {
 		c->name = name;
#1 lore raw

[Catalin Marinas - Wed, Dec 10, 2008 at 05:36:33PM +0000]

| It looks like the kmem_cache_create() function in the slob allocator
| passes the SLAB flags as GFP flags to the slob_alloc() function. The
| patch changes this call to pass GFP_KERNEL as the other allocators seem
| to do.
| 
| Signed-off-by: Catalin Marinas <[email protected]>
| Cc: Matt Mackall <[email protected]>
| ---
|  mm/slob.c |    2 +-
|  1 files changed, 1 insertions(+), 1 deletions(-)
| 
| diff --git a/mm/slob.c b/mm/slob.c
| index ff5a98d..dce9258 100644
| --- a/mm/slob.c
| +++ b/mm/slob.c
| @@ -538,7 +538,7 @@ struct kmem_cache *kmem_cache_create(const char *name, size_t size,
|  	struct kmem_cache *c;
|  
|  	c = slob_alloc(sizeof(struct kmem_cache),
| -		flags, ARCH_KMALLOC_MINALIGN, -1);
| +		GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
|  
|  	if (c) {
|  		c->name = name;
|

Hi Catalin,

this would make the following line in slob_alloc

	...
	if (unlikely((gfp & __GFP_ZERO) && b))
		memset(b, 0, size);

...

useless. Not sure if it will be good :)

---
Some CC: added

- Cyrill -

#2 lore raw
Quoted 29 lines · Cyrill Gorcunov

[Catalin Marinas - Wed, Dec 10, 2008 at 05:36:33PM +0000]

| It looks like the kmem_cache_create() function in the slob allocator
| passes the SLAB flags as GFP flags to the slob_alloc() function. The
| patch changes this call to pass GFP_KERNEL as the other allocators seem
| to do.
| 
| Signed-off-by: Catalin Marinas <[email protected]>
| Cc: Matt Mackall <[email protected]>
| ---
|  mm/slob.c |    2 +-
|  1 files changed, 1 insertions(+), 1 deletions(-)
| 
| diff --git a/mm/slob.c b/mm/slob.c
| index ff5a98d..dce9258 100644
| --- a/mm/slob.c
| +++ b/mm/slob.c
| @@ -538,7 +538,7 @@ struct kmem_cache *kmem_cache_create(const char *name, size_t size,
|  	struct kmem_cache *c;
|  
|  	c = slob_alloc(sizeof(struct kmem_cache),
| -		flags, ARCH_KMALLOC_MINALIGN, -1);
| +		GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
|  
|  	if (c) {
|  		c->name = name;
|

Hi Catalin,

Quoted 8 lines

this would make the following line in slob_alloc

	...
	if (unlikely((gfp & __GFP_ZERO) && b))
		memset(b, 0, size);

...

useless. Not sure if it will be good :)

I think that's fine. Notice that this is not the allocation of an object, but the allocation of a cache descriptor. We save the flags inside the descriptor so that when the user calls kmem_cache_alloc, we can pass them along to slob_alloc, where GFP_ZERO can take effect.

So, if we're comfortable with the idea that we can only create caches in 'normal' contexts, then I think the patch is fine.

Signature · 1 line
Mathematics is the supreme nostalgia of our time.
#3 lore raw

[Matt Mackall - Wed, Dec 10, 2008 at 01:57:50PM -0600]
...

| I think that's fine. Notice that this is not the allocation of an
| object, but the allocation of a cache descriptor. We save the flags
| inside the descriptor so that when the user calls kmem_cache_alloc, we
| can pass them along to slob_alloc, where GFP_ZERO can take effect.
| 
| So, if we're comfortable with the idea that we can only create caches in
| 'normal' contexts, then I think the patch is fine.
| 
| -- 
| Mathematics is the supreme nostalgia of our time.
| 

Yes Matt, somehow missed that.

- Cyrill -

#4 lore raw
In reply to Cyrill Gorcunov
Quoted 8 lines · Cyrill Gorcunov

this would make the following line in slob_alloc

	...
	if (unlikely((gfp & __GFP_ZERO) && b))
		memset(b, 0, size);

...

useless. Not sure if it will be good :)

No. GFP_ZERO is set when kmalloc etc is called. Its not set on kmem_cache_create.

#5 lore raw

[Christoph Lameter - Wed, Dec 10, 2008 at 02:11:23PM -0600]

| On Wed, 10 Dec 2008, Cyrill Gorcunov wrote:
| 
| > this would make the following line in slob_alloc
| >
| > 	...
| > 	if (unlikely((gfp & __GFP_ZERO) && b))
| > 		memset(b, 0, size);
| >  	...
| >
| > useless. Not sure if it will be good :)
| 
| No. GFP_ZERO is set when kmalloc etc is called. Its not set on
| kmem_cache_create.
| 

yep, Matt already pointed, I just missed first time.

- Cyrill -

Acked-by on v1 1/1
#6 lore raw
In reply to Catalin Marinas
Quoted 7 lines · Catalin Marinas

It looks like the kmem_cache_create() function in the slob allocator passes the SLAB flags as GFP flags to the slob_alloc() function. The patch changes this call to pass GFP_KERNEL as the other allocators seem to do.

Good catch.

Signature · 1 line
Mathematics is the supreme nostalgia of our time.