Re: [PATCH 02/13] mm: use page_off_lru()

From: Michal Hocko
Date: Fri Sep 18 2020 - 03:37:06 EST


On Thu 17-09-20 21:00:40, Yu Zhao wrote:
> This patch replaces the only open-coded __ClearPageActive() with
> page_off_lru(). There is no open-coded __ClearPageUnevictable()s.
>
> Before this patch, we have:
> __ClearPageActive()
> add_page_to_lru_list()
>
> After this patch, we have:
> page_off_lru()
> if PageUnevictable()
> __ClearPageUnevictable()
> else if PageActive()
> __ClearPageActive()
> add_page_to_lru_list()
>
> Checking PageUnevictable() shouldn't be a problem because these two
> flags are mutually exclusive. Leaking either will trigger bad_page().

I am sorry but the changelog is really hard to grasp. What are you
trying to achieve, why and why it is safe. This should be a general
outline for any patch. I have already commented on the previous patch
and asked you for the explanation why removing __ClearPageActive from
this path is desirable and safe. I have specifically asked to clarify
the compound page situation as that is using its oen destructor in the
freeing path and that might result in page_off_lru to be not called.

> Signed-off-by: Yu Zhao <yuzhao@xxxxxxxxxx>
> ---
> mm/vmscan.c | 6 +-----
> 1 file changed, 1 insertion(+), 5 deletions(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 503fc5e1fe32..f257d2f61574 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -1845,7 +1845,6 @@ static unsigned noinline_for_stack move_pages_to_lru(struct lruvec *lruvec,
> int nr_pages, nr_moved = 0;
> LIST_HEAD(pages_to_free);
> struct page *page;
> - enum lru_list lru;
>
> while (!list_empty(list)) {
> page = lru_to_page(list);
> @@ -1860,14 +1859,11 @@ static unsigned noinline_for_stack move_pages_to_lru(struct lruvec *lruvec,
> lruvec = mem_cgroup_page_lruvec(page, pgdat);
>
> SetPageLRU(page);
> - lru = page_lru(page);
> -
> add_page_to_lru_list(page, lruvec, lru);
>
> if (put_page_testzero(page)) {
> __ClearPageLRU(page);
> - __ClearPageActive(page);
> - del_page_from_lru_list(page, lruvec, lru);
> + del_page_from_lru_list(page, lruvec, page_off_lru(page));
>
> if (unlikely(PageCompound(page))) {
> spin_unlock_irq(&pgdat->lru_lock);
> --
> 2.28.0.681.g6f77f65b4e-goog

--
Michal Hocko
SUSE Labs