Re: [PATCH] improve speed of freeing bootmem

[Andrew Morton <akpm@osdl.org>]

Josh Aas <josha@sgi.com> wrote:
>
> Attached is a patch that greatly improves the speed of freeing boot 
> memory.

hm, OK.  I have a vague feeling that Bill Irwin had patches to fix this up
ages ago.


A few nits:

> --- a/mm/bootmem.c	2004-08-05 15:33:39.000000000 -0500
> +++ b/mm/bootmem.c	2004-08-06 13:42:33.000000000 -0500
> @@ -259,6 +259,7 @@ static unsigned long __init free_all_boo
>  	unsigned long i, count, total = 0;
>  	unsigned long idx;
>  	unsigned long *map; 
> +	int gofast = 0;
>  
>  	BUG_ON(!bdata->node_bootmem_map);
>  
> @@ -267,14 +268,32 @@ static unsigned long __init free_all_boo
>  	page = virt_to_page(phys_to_virt(bdata->node_boot_start));
>  	idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
>  	map = bdata->node_bootmem_map;
> +	if (bdata->node_boot_start = 0 ||
> +	    ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
> +		gofast = 1;

A comment describing the above reasoning would be nice.

>  	for (i = 0; i < idx; ) {
>  		unsigned long v = ~map[i / BITS_PER_LONG];
> -		if (v) {
> +		if (gofast && v = ~0UL) {
> +			int j;
> +
> +			count += BITS_PER_LONG;
> +			ClearPageReservedNoAtomic(page);
> +			set_page_count(page, 1);
> +			for (j = 1; j < BITS_PER_LONG; j++) {
> +				if (j + 16 < BITS_PER_LONG) {
> +                      			prefetchw(page + j + 16);
> +                                }

The whitespace/tabbing has gone funny here.

> +#define ClearPageReservedNoAtomic(page)	(page)->flags &= ~(1UL << PG_reserved)

The naming convention we used in 2.4 for the nonatomic operation was
__ClearPageReserved(), so can we please stick with that?

And this macro can use __clear_bit() rather than open-coding it.