[2.6 patch] slub.c:early_kmem_cache_node_alloc() shouldn't be __init

[2.6 patch] slub.c:early_kmem_cache_node_alloc() shouldn't be __init

[Adrian Bunk]

This patch fixes the following section mismatch:

<--  snip  -->

...
  MODPOST vmlinux
WARNING: mm/built-in.o(.text+0x24bd3): Section mismatch: reference to .init.text:early_kmem_cache_node_alloc (between 'init_kmem_cache_nodes' and 'calculate_sizes')
...

<--  snip  -->

Signed-off-by: Adrian Bunk <bunk@stusta.de>

---
--- linux-2.6.22-rc6-mm1/mm/slub.c.old	2007-07-10 04:31:32.000000000 +0200
+++ linux-2.6.22-rc6-mm1/mm/slub.c	2007-07-10 04:31:52.000000000 +0200
@@ -1882,8 +1882,8 @@
  * Note that this function only works on the kmalloc_node_cache
  * when allocating for the kmalloc_node_cache.
  */
-static struct kmem_cache_node * __init early_kmem_cache_node_alloc(gfp_t gfpflags,
-								int node)
+static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags,
+							   int node)
 {
 	struct page *page;
 	struct kmem_cache_node *n;

Re: [2.6 patch] slub.c:early_kmem_cache_node_alloc() shouldn't be __init

[Christoph Lameter]

On Tue, 10 Jul 2007, Adrian Bunk wrote:

> This patch fixes the following section mismatch:

Nack.

Can you remove the warning? early_kmem_cache_node_alloc is only used 
during bootstrap. But it is called from a function that is also used 
later.

Re: [2.6 patch] slub.c:early_kmem_cache_node_alloc() shouldn't be __init

[Adrian Bunk]

On Tue, Jul 10, 2007 at 11:33:15AM -0700, Christoph Lameter wrote:
> On Tue, 10 Jul 2007, Adrian Bunk wrote:
> 
> > This patch fixes the following section mismatch:
> 
> Nack.
> 
> Can you remove the warning? early_kmem_cache_node_alloc is only used 
> during bootstrap. But it is called from a function that is also used 
> later.

The alternative is "noinline __init_refok" - the noinline is required 
because modern gcc versions already make it non-__init by inlining it 
into init_kmem_cache_nodes() without.

But I'm not convinced it's really worth manually forcing noinline only 
for moving a few bytes to __init in the CONFIG_NUMA=y case - in the  
long term this will only bit rot.

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

Re: [2.6 patch] slub.c:early_kmem_cache_node_alloc() shouldn't be __init

[Christoph Lameter]

On Tue, 10 Jul 2007, Adrian Bunk wrote:

> But I'm not convinced it's really worth manually forcing noinline only 
> for moving a few bytes to __init in the CONFIG_NUMA=y case - in the  
> long term this will only bit rot.

Allright then

Acked-by: Christoph Lameter <clameter@sgi.com>