[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Tue, 10 Nov 2020 23:27:54 -0800 (PST)
From: Hugh Dickins <hughd@...gle.com>
To: Johannes Weiner <hannes@...xchg.org>
cc: Alex Shi <alex.shi@...ux.alibaba.com>, akpm@...ux-foundation.org,
mgorman@...hsingularity.net, tj@...nel.org, hughd@...gle.com,
khlebnikov@...dex-team.ru, daniel.m.jordan@...cle.com,
willy@...radead.org, lkp@...el.com, linux-mm@...ck.org,
linux-kernel@...r.kernel.org, cgroups@...r.kernel.org,
shakeelb@...gle.com, iamjoonsoo.kim@....com,
richard.weiyang@...il.com, kirill@...temov.name,
alexander.duyck@...il.com, rong.a.chen@...el.com, mhocko@...e.com,
vdavydov.dev@...il.com, shy828301@...il.com,
Vlastimil Babka <vbabka@...e.cz>,
Minchan Kim <minchan@...nel.org>
Subject: Re: [PATCH v20 08/20] mm: page_idle_get_page() does not need
lru_lock
On Mon, 2 Nov 2020, Johannes Weiner wrote:
> On Thu, Oct 29, 2020 at 06:44:53PM +0800, Alex Shi wrote:
> > From: Hugh Dickins <hughd@...gle.com>
> >
> > It is necessary for page_idle_get_page() to recheck PageLRU() after
> > get_page_unless_zero(), but holding lru_lock around that serves no
> > useful purpose, and adds to lru_lock contention: delete it.
> >
> > See https://lore.kernel.org/lkml/20150504031722.GA2768@blaptop for the
> > discussion that led to lru_lock there; but __page_set_anon_rmap() now
> > uses WRITE_ONCE(),
>
> That doesn't seem to be the case in Linus's or Andrew's tree. Am I
> missing a dependent patch series?
Sorry, I was out of action, then slower than ever, for a while.
Many thanks for calling out my falsehood there, Johannes.
What led me to write that? It has baffled me, but at last I see:
this patch to page_idle_get_page() was 0002 in my lru_lock patchset
against v5.3 last year, and 0001 was the patch which made it true.
Then when I checked against mainline, I must have got confused by
the similar WRITE_ONCE in page_move_anon_rmap().
Appended below, but not rediffed, and let's not hold up Alex's set
for the rest of it: it is all theoretical until the kernel gets to
be built with a suitably malicious compiler; but I'll follow up
with a fresh version of the below after his set is safely in.
>From a1abcbc2aac70c6ba47b8991992bb85b86b4a160 Mon Sep 17 00:00:00 2001
From: Hugh Dickins <hughd@...gle.com>
Date: Thu, 22 Aug 2019 15:49:44 -0700
Subject: [PATCH 1/9] mm: more WRITE_ONCE and READ_ONCE on page->mapping
v4.2 commit 414e2fb8ce5a ("rmap: fix theoretical race between do_wp_page
and shrink_active_list") added a WRITE_ONCE() where page_move_anon_rmap()
composes page->mapping from anon_vma pointer and PAGE_MAPPING_ANON.
Now do the same where __page_set_anon_rmap() does the same, and where
compaction.c applies PAGE_MAPPING_MOVABLE, and ksm.c PAGE_MAPPING_KSM.
rmap.c already uses READ_ONCE(page->mapping), but util.c should too:
add READ_ONCE() in page_rmapping(), page_anon_vma() and page_mapping().
Delete the then unused helper __page_rmapping().
I doubt that this commit fixes anything, but it's harmless and
unintrusive, and makes reasoning about page mapping flags easier.
What if a compiler implements "page->mapping = mapping" in other places
by, say, first assigning the odd bits of mapping, then adding in the
even bits? Then we shall not build the kernel with such a compiler.
Signed-off-by: Hugh Dickins <hughd@...gle.com>
Cc: Vladimir Davydov <vdavydov.dev@...il.com>
Cc: Vlastimil Babka <vbabka@...e.cz>
Cc: Minchan Kim <minchan@...nel.org>
Cc: Yu Zhao <yuzhao@...gle.com>
Cc: Alex Shi <alex.shi@...ux.alibaba.com>
---
mm/compaction.c | 7 ++++---
mm/ksm.c | 2 +-
mm/rmap.c | 7 ++++++-
mm/util.c | 24 ++++++++++--------------
4 files changed, 21 insertions(+), 19 deletions(-)
diff --git a/mm/compaction.c b/mm/compaction.c
index 952dc2fb24e5..c405f4362624 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -113,7 +113,8 @@ void __SetPageMovable(struct page *page, struct address_space *mapping)
{
VM_BUG_ON_PAGE(!PageLocked(page), page);
VM_BUG_ON_PAGE((unsigned long)mapping & PAGE_MAPPING_MOVABLE, page);
- page->mapping = (void *)((unsigned long)mapping | PAGE_MAPPING_MOVABLE);
+ WRITE_ONCE(page->mapping,
+ (unsigned long)mapping | PAGE_MAPPING_MOVABLE);
}
EXPORT_SYMBOL(__SetPageMovable);
@@ -126,8 +127,8 @@ void __ClearPageMovable(struct page *page)
* flag so that VM can catch up released page by driver after isolation.
* With it, VM migration doesn't try to put it back.
*/
- page->mapping = (void *)((unsigned long)page->mapping &
- PAGE_MAPPING_MOVABLE);
+ WRITE_ONCE(page->mapping,
+ (unsigned long)page->mapping & PAGE_MAPPING_MOVABLE);
}
EXPORT_SYMBOL(__ClearPageMovable);
diff --git a/mm/ksm.c b/mm/ksm.c
index 3dc4346411e4..426b6a40ea41 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -865,7 +865,7 @@ static inline struct stable_node *page_stable_node(struct page *page)
static inline void set_page_stable_node(struct page *page,
struct stable_node *stable_node)
{
- page->mapping = (void *)((unsigned long)stable_node | PAGE_MAPPING_KSM);
+ WRITE_ONCE(page->mapping, (unsigned long)stable_node | PAGE_MAPPING_KSM);
}
#ifdef CONFIG_SYSFS
diff --git a/mm/rmap.c b/mm/rmap.c
index 003377e24232..9480df437edc 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1044,7 +1044,12 @@ static void __page_set_anon_rmap(struct page *page,
anon_vma = anon_vma->root;
anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
- page->mapping = (struct address_space *) anon_vma;
+ /*
+ * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
+ * simultaneously, so a concurrent reader (eg page_referenced()'s
+ * PageAnon()) will not see one without the other.
+ */
+ WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
page->index = linear_page_index(vma, address);
}
diff --git a/mm/util.c b/mm/util.c
index e6351a80f248..09b9fcbedac3 100644
--- a/mm/util.c
+++ b/mm/util.c
@@ -489,21 +489,14 @@ void kvfree(const void *addr)
}
EXPORT_SYMBOL(kvfree);
-static inline void *__page_rmapping(struct page *page)
-{
- unsigned long mapping;
-
- mapping = (unsigned long)page->mapping;
- mapping &= ~PAGE_MAPPING_FLAGS;
-
- return (void *)mapping;
-}
-
/* Neutral page->mapping pointer to address_space or anon_vma or other */
void *page_rmapping(struct page *page)
{
+ unsigned long mapping;
+
page = compound_head(page);
- return __page_rmapping(page);
+ mapping = (unsigned long)READ_ONCE(page->mapping);
+ return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
}
/*
@@ -534,10 +527,11 @@ struct anon_vma *page_anon_vma(struct page *page)
unsigned long mapping;
page = compound_head(page);
- mapping = (unsigned long)page->mapping;
+ mapping = (unsigned long)READ_ONCE(page->mapping);
+ /* Return NULL if file or PageMovable or PageKsm */
if ((mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
return NULL;
- return __page_rmapping(page);
+ return (struct anon_vma *)(mapping & ~PAGE_MAPPING_FLAGS);
}
struct address_space *page_mapping(struct page *page)
@@ -557,10 +551,12 @@ struct address_space *page_mapping(struct page *page)
return swap_address_space(entry);
}
- mapping = page->mapping;
+ mapping = READ_ONCE(page->mapping);
+ /* Return NULL if PageAnon (including PageKsm) */
if ((unsigned long)mapping & PAGE_MAPPING_ANON)
return NULL;
+ /* Return struct address_space pointer if file or PageMovable */
return (void *)((unsigned long)mapping & ~PAGE_MAPPING_FLAGS);
}
EXPORT_SYMBOL(page_mapping);
--
2.23.0.187.g17f5b7556c-goog
Powered by blists - more mailing lists