Re: [S+Q Cleanup 3/6] slub: Remove static kmem_cache_cpu array for boot

[David Rientjes <rientjes@google.com>]

On Tue, 17 Aug 2010, Christoph Lameter wrote:

> Index: linux-2.6/mm/slub.c
> ===================================================================
> --- linux-2.6.orig/mm/slub.c	2010-08-13 10:32:45.000000000 -0500
> +++ linux-2.6/mm/slub.c	2010-08-13 10:32:50.000000000 -0500
> @@ -2062,23 +2062,14 @@ init_kmem_cache_node(struct kmem_cache_n
>  #endif
>  }
>  
> -static DEFINE_PER_CPU(struct kmem_cache_cpu, kmalloc_percpu[KMALLOC_CACHES]);
> -
>  static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
>  {
> -	if (s < kmalloc_caches + KMALLOC_CACHES && s >= kmalloc_caches)
> -		/*
> -		 * Boot time creation of the kmalloc array. Use static per cpu data
> -		 * since the per cpu allocator is not available yet.
> -		 */
> -		s->cpu_slab = kmalloc_percpu + (s - kmalloc_caches);
> -	else
> -		s->cpu_slab =  alloc_percpu(struct kmem_cache_cpu);
> +	BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
> +			SLUB_PAGE_SHIFT * sizeof(struct kmem_cache));

This fails with CONFIG_NODES_SHIFT=10 on x86_64, which means it will fail 
the ia64 defconfig as well.  struct kmem_cache stores nodemask pointers up 
to MAX_NUMNODES, which makes the conditional fail.

struct kmem_cache is 8376 bytes with that config (and CONFIG_SLUB_DEBUG), 
so it looks like PERCPU_DYNAMIC_EARLY_SIZE will need to be at least 117264 
for this not to fail (four orders larger than it currently is, or
12 << 14).  Tejun?

>  
> -	if (!s->cpu_slab)
> -		return 0;
> +	s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
>  
> -	return 1;
> +	return s->cpu_slab != NULL;
>  }
>  
>  #ifdef CONFIG_NUMA
> 
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>