On Fri, Feb 28, 2014 at 03:15:00PM +0100, Vlastimil Babka wrote:
> In order to prevent race with set_pageblock_migratetype, most of calls to
> get_pageblock_migratetype have been moved under zone->lock. For the remaining
> call sites, the extra locking is undesirable, notably in free_hot_cold_page().
>
> This patch introduces a _nolock version to be used on these call sites, where
> a wrong value does not affect correctness. The function makes sure that the
> value does not exceed valid migratetype numbers. Such too-high values are
> assumed to be a result of race and caller-supplied fallback value is returned
> instead.
>
> Signed-off-by: Vlastimil Babka <vba...@suse.cz>
> ---
> include/linux/mmzone.h | 24 ++++++++++++++++++++++++
> mm/compaction.c | 14 +++++++++++---
> mm/memory-failure.c | 3 ++-
> mm/page_alloc.c | 22 +++++++++++++++++-----
> mm/vmstat.c | 2 +-
> 5 files changed, 55 insertions(+), 10 deletions(-)
>
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index fac5509..7c3f678 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -75,6 +75,30 @@ enum {
>
> extern int page_group_by_mobility_disabled;
>
> +/*
> + * When called without zone->lock held, a race with set_pageblock_migratetype
> + * may result in bogus values. Use this variant only when this does not
> affect
> + * correctness, and taking zone->lock would be costly. Values >=
> MIGRATE_TYPES
> + * are considered to be a result of this race and the value of race_fallback
> + * argument is returned instead.
> + */
> +static inline int get_pageblock_migratetype_nolock(struct page *page,
> + int race_fallback)
> +{
> + int ret = get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
> +
> + if (unlikely(ret >= MIGRATE_TYPES))
> + ret = race_fallback;
> +
> + return ret;
> +}
How about below forms?
get_pageblock_migratetype_locked(struct page *page)
get_pageblock_migratetype(struct page *page, int race_fallback)
get_pageblock_migratetype() and _nolock looks error-prone because developer
who try to use get_pageblock_migratetype() may not know that it needs lock.
Thanks.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
