lkml.org 
[lkml]   [2009]   [Dec]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH 5/7] Don't deactivate the page if trylock_page() is failed.
Date
From 7635eaa033cfcce7f351b5023952f23f0daffefe Mon Sep 17 00:00:00 2001
From: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Date: Fri, 4 Dec 2009 12:03:07 +0900
Subject: [PATCH 5/7] Don't deactivate the page if trylock_page() is failed.

Currently, wipe_page_reference() increment refctx->referenced variable
if trylock_page() is failed. but it is meaningless at all.
shrink_active_list() deactivate the page although the page was
referenced. The page shouldn't be deactivated with young bit. it
break reclaim basic theory and decrease reclaim throughput.

This patch introduce new SWAP_AGAIN return value to
wipe_page_reference().

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
---
mm/rmap.c | 5 ++++-
mm/vmscan.c | 15 +++++++++++++--
2 files changed, 17 insertions(+), 3 deletions(-)

diff --git a/mm/rmap.c b/mm/rmap.c
index 2f4451b..b84f350 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -539,6 +539,9 @@ static int wipe_page_reference_file(struct page *page,
*
* Quick test_and_clear_referenced for all mappings to a page,
* returns the number of ptes which referenced the page.
+ *
+ * SWAP_SUCCESS - success to wipe all ptes
+ * SWAP_AGAIN - temporary busy, try again later
*/
int wipe_page_reference(struct page *page,
struct mem_cgroup *memcg,
@@ -555,7 +558,7 @@ int wipe_page_reference(struct page *page,
(!PageAnon(page) || PageKsm(page))) {
we_locked = trylock_page(page);
if (!we_locked) {
- refctx->referenced++;
+ ret = SWAP_AGAIN;
goto out;
}
}
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 0db9c06..9684e40 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -577,6 +577,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
struct address_space *mapping;
struct page *page;
int may_enter_fs;
+ int ret;
struct page_reference_context refctx = {
.is_page_locked = 1,
};
@@ -621,7 +622,11 @@ static unsigned long shrink_page_list(struct list_head *page_list,
goto keep_locked;
}

- wipe_page_reference(page, sc->mem_cgroup, &refctx);
+ ret = wipe_page_reference(page, sc->mem_cgroup, &refctx);
+ if (ret == SWAP_AGAIN)
+ goto keep_locked;
+ VM_BUG_ON(ret != SWAP_SUCCESS);
+
/*
* In active use or really unfreeable? Activate it.
* If page which have PG_mlocked lost isoltation race,
@@ -1326,6 +1331,7 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
spin_unlock_irq(&zone->lru_lock);

while (!list_empty(&l_hold)) {
+ int ret;
struct page_reference_context refctx = {
.is_page_locked = 0,
};
@@ -1345,7 +1351,12 @@ static void shrink_active_list(unsigned long nr_pages, struct zone *zone,
continue;
}

- wipe_page_reference(page, sc->mem_cgroup, &refctx);
+ ret = wipe_page_reference(page, sc->mem_cgroup, &refctx);
+ if (ret == SWAP_AGAIN) {
+ list_add(&page->lru, &l_active);
+ continue;
+ }
+ VM_BUG_ON(ret != SWAP_SUCCESS);

if (refctx.referenced)
nr_rotated++;
--
1.6.5.2




\
 
 \ /
  Last update: 2009-12-04 09:47    [W:0.113 / U:5.500 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site