The commit ad53f92e(fix incorrect isolation behavior by rechecking migratetype)
patch series describe the race between page isolation and alloc/free path, and
fix the race.
Now, after the pageblock has been isolated, free buddy pages are already in
the free_list[MIGRATE_ISOLATE] and will not be allocated for usage. So the
current freepage_migratetype check is unnecessary and it will cause redundant
page move. That is to say, even if the buddy page's migratetype is not
MIGRATE_ISOLATE, the page is in free_list[MIGRATE_ISOLATE], we just move it
from free_list[MIGRATE_ISOLATE] to free_list[MIGRATE_ISOLATE].
This patch removes the unnecessary freepage_migratetype check and the
redundant page moving.
Signed-off-by: Weijie Yang <[email protected]>
---
mm/page_isolation.c | 17 +----------------
1 file changed, 1 insertion(+), 16 deletions(-)
diff --git a/mm/page_isolation.c b/mm/page_isolation.c
index c8778f7..6e5174d 100644
--- a/mm/page_isolation.c
+++ b/mm/page_isolation.c
@@ -221,23 +221,8 @@ __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
continue;
}
page = pfn_to_page(pfn);
- if (PageBuddy(page)) {
- /*
- * If race between isolatation and allocation happens,
- * some free pages could be in MIGRATE_MOVABLE list
- * although pageblock's migratation type of the page
- * is MIGRATE_ISOLATE. Catch it and move the page into
- * MIGRATE_ISOLATE list.
- */
- if (get_freepage_migratetype(page) != MIGRATE_ISOLATE) {
- struct page *end_page;
-
- end_page = page + (1 << page_order(page)) - 1;
- move_freepages(page_zone(page), page, end_page,
- MIGRATE_ISOLATE);
- }
+ if (PageBuddy(page))
pfn += 1 << page_order(page);
- }
else if (page_count(page) == 0 &&
get_freepage_migratetype(page) == MIGRATE_ISOLATE)
pfn += 1;
--
1.7.10.4
On 12/09/2014 08:50 AM, Weijie Yang wrote:
> The commit ad53f92e(fix incorrect isolation behavior by rechecking migratetype)
> patch series describe the race between page isolation and alloc/free path, and
> fix the race.
>
> Now, after the pageblock has been isolated, free buddy pages are already in
> the free_list[MIGRATE_ISOLATE] and will not be allocated for usage. So the
> current freepage_migratetype check is unnecessary and it will cause redundant
> page move. That is to say, even if the buddy page's migratetype is not
> MIGRATE_ISOLATE, the page is in free_list[MIGRATE_ISOLATE], we just move it
> from free_list[MIGRATE_ISOLATE] to free_list[MIGRATE_ISOLATE].
>
> This patch removes the unnecessary freepage_migratetype check and the
> redundant page moving.
>
> Signed-off-by: Weijie Yang <[email protected]>
With great hope that Joonsoo won't find another corner case J
Acked-by: Vlastimil Babka <[email protected]>
> ---
> mm/page_isolation.c | 17 +----------------
> 1 file changed, 1 insertion(+), 16 deletions(-)
>
> diff --git a/mm/page_isolation.c b/mm/page_isolation.c
> index c8778f7..6e5174d 100644
> --- a/mm/page_isolation.c
> +++ b/mm/page_isolation.c
> @@ -221,23 +221,8 @@ __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
> continue;
> }
> page = pfn_to_page(pfn);
> - if (PageBuddy(page)) {
> - /*
> - * If race between isolatation and allocation happens,
> - * some free pages could be in MIGRATE_MOVABLE list
> - * although pageblock's migratation type of the page
> - * is MIGRATE_ISOLATE. Catch it and move the page into
> - * MIGRATE_ISOLATE list.
> - */
> - if (get_freepage_migratetype(page) != MIGRATE_ISOLATE) {
> - struct page *end_page;
> -
> - end_page = page + (1 << page_order(page)) - 1;
> - move_freepages(page_zone(page), page, end_page,
> - MIGRATE_ISOLATE);
> - }
> + if (PageBuddy(page))
> pfn += 1 << page_order(page);
> - }
> else if (page_count(page) == 0 &&
> get_freepage_migratetype(page) == MIGRATE_ISOLATE)
> pfn += 1;
>