Re: [PATCH 1/4] sparc64: Kill page table quicklists.

[Andrew Morton]

On Tue, 12 Jul 2011 14:26:09 +0200
Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> wrote:

> --- linux-2.6.orig/arch/sparc/mm/tsb.c
> +++ linux-2.6/arch/sparc/mm/tsb.c
> @@ -236,6 +236,8 @@ static void setup_tsb_params(struct mm_s
>  	}
>  }
>  
> +struct kmem_cache *pgtable_cache __read_mostly;
> +
>  static struct kmem_cache *tsb_caches[8] __read_mostly;
>  
>  static const char *tsb_cache_names[8] = {
> @@ -253,6 +255,15 @@ void __init pgtable_cache_init(void)
>  {
>  	unsigned long i;
>  
> +	pgtable_cache = kmem_cache_create("pgtable_cache",
> +					  PAGE_SIZE, PAGE_SIZE,
> +					  0,
> +					  _clear_page);

The use of slab constructors is often dubious from a cache usage POV. 
But the lifecycle of a page-table page might well be that it slowly
gets non-zeroes written into it and then slowly gets zeroes written
into it until it is all-zeroes and then we free it up.  And often only
a subset of the page will ever be written.

So it could be that the slab constructor behaviour is a good match
here.  And not just for sparc!

Did such thinking and/or any testing go into this decision?


> +	if (!pgtable_cache) {
> +		prom_printf("pgtable_cache_init(): Could not create!\n");
> +		prom_halt();
> +	}
> +
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/