On Mon 03-08-15 18:02:27, Vladimir Davydov wrote:
> If transparent huge pages are enabled, we can isolate many more pages
> than we actually need to scan, because we count both single and huge
> pages equally in isolate_lru_pages().
>
> Since commit 5bc7b8aca942d ("mm: thp: add split tail pages to shrink
> page list in page reclaim"), we scan all the tail pages immediately
> after a huge page split (see shrink_page_list()). As a result, we can
> reclaim up to SWAP_CLUSTER_MAX * HPAGE_PMD_NR (512 MB) in one run!
512MB is really unexpected. But yeah, you are right. Mel has increased
SWAP_CLUSTER_MAX to 256 recently (mm: increase SWAP_CLUSTER_MAX to
batch TLB flushes) which I have missed. That has made the situation
potentially much worse. I guess this is worth mentioning in the
changelog because the original SWAP_CLUSTER_MAX (32) hasn't looked that
scary.
> This is easy to catch on memcg reclaim with zswap enabled. The latter
> makes swapout instant so that if we happen to scan an unreferenced huge
> page we will evict both its head and tail pages immediately, which is
> likely to result in excessive reclaim.
>
> Signed-off-by: Vladimir Davydov <[email protected]>
Reviewed-by: Michal Hocko <[email protected]>
> ---
> mm/vmscan.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 5221e19e98f4..94092fd3b96b 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -1387,7 +1387,8 @@ static unsigned long isolate_lru_pages(unsigned long
> nr_to_scan,
> unsigned long nr_taken = 0;
> unsigned long scan;
>
> - for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
> + for (scan = 0; scan < nr_to_scan && nr_taken < nr_to_scan &&
> + !list_empty(src); scan++) {
> struct page *page;
> int nr_pages;
>
> --
> 2.1.4
>
--
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
