[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20240524-b4-ksm-scan-optimize-v1-4-053b31bd7ab4@linux.dev>
Date: Fri, 24 May 2024 16:56:53 +0800
From: Chengming Zhou <chengming.zhou@...ux.dev>
To: Andrew Morton <akpm@...ux-foundation.org>, david@...hat.com,
aarcange@...hat.com, hughd@...gle.com, shr@...kernel.io
Cc: linux-mm@...ck.org, linux-kernel@...r.kernel.org,
zhouchengming@...edance.com, Chengming Zhou <chengming.zhou@...ux.dev>
Subject: [PATCH 4/4] mm/ksm: use ksm page itself if no another ksm page is
found on stable tree
It's interesting that a mapped ksm page also need to stable_tree_search(),
instead of using stable_tree_insert() directly. The reason is that we have
a minor optimization for migrated ksm page that has only one mapcount, in
which case we can find another ksm page that already on the stable tree
to replace it.
But what if we can't find another shareable candidate on the stable tree?
Obviously, we should just return the ksm page itself if it has been
inserted on the tree. And we shouldn't return NULL if no another ksm page
is found on the tree, since we will still map on this ksm page but the
rmap_item will be removed out to insert on the unstable tree if we return
NULL in this case.
We can ignore the is_page_sharing_candidate() check in this case, since
max_page_sharing_bypass is set to true in cmp_and_merge_page().
Signed-off-by: Chengming Zhou <chengming.zhou@...ux.dev>
---
mm/ksm.c | 19 +++++--------------
1 file changed, 5 insertions(+), 14 deletions(-)
diff --git a/mm/ksm.c b/mm/ksm.c
index f923699452ed..6dea83998258 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -1940,11 +1940,8 @@ static struct page *stable_tree_search(struct page *page)
rb_link_node(&page_node->node, parent, new);
rb_insert_color(&page_node->node, root);
out:
- if (is_page_sharing_candidate(page_node)) {
- folio_get(folio);
- return &folio->page;
- } else
- return NULL;
+ folio_get(folio);
+ return &folio->page;
replace:
/*
@@ -1966,10 +1963,7 @@ static struct page *stable_tree_search(struct page *page)
rb_replace_node(&stable_node_dup->node,
&page_node->node,
root);
- if (is_page_sharing_candidate(page_node))
- folio_get(folio);
- else
- folio = NULL;
+ folio_get(folio);
} else {
rb_erase(&stable_node_dup->node, root);
folio = NULL;
@@ -1982,10 +1976,7 @@ static struct page *stable_tree_search(struct page *page)
list_del(&page_node->list);
DO_NUMA(page_node->nid = nid);
stable_node_chain_add_dup(page_node, stable_node);
- if (is_page_sharing_candidate(page_node))
- folio_get(folio);
- else
- folio = NULL;
+ folio_get(folio);
} else {
folio = NULL;
}
@@ -2009,7 +2000,7 @@ static struct page *stable_tree_search(struct page *page)
stable_node = alloc_stable_node_chain(stable_node_dup,
root);
if (!stable_node)
- return NULL;
+ goto out;
}
/*
* Add this stable_node dup that was
--
2.45.1
Powered by blists - more mailing lists