lkml.org 
[lkml]   [2011]   [Dec]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH RFC] bootmem: micro optimize freeing pages in bulks
On Thu, Dec 01, 2011 at 11:10:55PM +0100, Uwe Kleine-König wrote:
> The first entry of bdata->node_bootmem_map holds the data for
> bdata->node_min_pfn up to bdata->node_min_pfn + BITS_PER_LONG - 1. So
> the test for freeing all pages of a single map entry can be slightly
> relaxed.

Agreed. The optimization is tiny - we may lose one bulk order-5/6
free per node and do it in 32/64 order-0 frees instead (we touch each
page anyway one way or another), but the code makes more sense with
your change.

[ Btw, what's worse is start being unaligned, because we won't do a
single batch free then. The single-page loop should probably just
move to the next BITS_PER_WORD boundary and then retry the aligned
batch frees. Oh, well... ]

> Moreover use DIV_ROUND_UP in another place instead of open coding it.

Agreed.

> Cc: Johannes Weiner <hannes@saeurebad.de>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> ---
> Hello,
>
> I'm not sure the current code is correct (and my patch doesn't fix it):
>
> If
>
> aligned && vec == ~0UL
>
> evalutates to true, but
>
> start + BITS_PER_LONG <= end
>
> does not (or "< end" resp.) the else branch still frees all BITS_PER_LONG
> pages. Is this intended? If yes, the last check can better be omitted
> resulting in the pages being freed in a bulk.
> If not, the loop in the else branch should only do something like:
>
> while (vec && off < min(BITS_PER_LONG, end - start)) {
> ...

I would think this is fine because node_bootmem_map, which is where
vec points to, is sized in multiples of pages, and zeroed word-wise.
So even if end is not aligned, we can rely on !vec.

> diff --git a/mm/bootmem.c b/mm/bootmem.c
> index fc22150..1e7d791 100644
> --- a/mm/bootmem.c
> +++ b/mm/bootmem.c
> @@ -56,7 +56,7 @@ early_param("bootmem_debug", bootmem_debug_setup);
>
> static unsigned long __init bootmap_bytes(unsigned long pages)
> {
> - unsigned long bytes = (pages + 7) / 8;
> + unsigned long bytes = DIV_ROUND_UP(pages, 8);
>
> return ALIGN(bytes, sizeof(long));
> }
> @@ -197,7 +197,7 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
> idx = start - bdata->node_min_pfn;
> vec = ~map[idx / BITS_PER_LONG];
>
> - if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
> + if (aligned && vec == ~0UL && start + BITS_PER_LONG <= end) {
> int order = ilog2(BITS_PER_LONG);
>
> __free_pages_bootmem(pfn_to_page(start), order);

Acked-by: Johannes Weiner <hannes@cmpxchg.org>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2011-12-13 11:09    [W:0.035 / U:0.172 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site