[PATCH v2] mm: check_new_page_bad() directly returns in __PG_HWPOISON case

[Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>]

On Wed, May 18, 2016 at 11:31:07AM +0200, Vlastimil Babka wrote:
> On 05/18/2016 11:21 AM, Mel Gorman wrote:
> > On Tue, May 17, 2016 at 04:42:55PM +0900, Naoya Horiguchi wrote:
> > > There's a race window between checking page->flags and unpoisoning, which
> > > taints kernel with "BUG: Bad page state". That's overkill. It's safer to
> > > use bad_flags to detect hwpoisoned page.
> > > 
> > 
> > I'm not quite getting this one. Minimally, instead of = __PG_HWPOISON, it
> > should have been (bad_flags & __PG_POISON). As Vlastimil already pointed
> > out, __PG_HWPOISON can be 0. What I'm not getting is why this fixes the
> > race. The current race is
> > 
> > 1. Check poison, set bad_flags
> > 2. poison clears in parallel
> > 3. Check page->flag state in bad_page and trigger warning
> > 
> > The code changes it to
> > 
> > 1. Check poison, set bad_flags
> > 2. poison clears in parallel
> > 3. Check bad_flags and trigger warning
> 
> I think you got step 3 here wrong. It's "skip the warning since we have set
> bad_flags to hwpoison and bad_flags didn't change due to parallel unpoison".
> 
> Perhaps the question is why do we need to split the handling between
> check_new_page_bad() and bad_page() like this? It might have been different
> in the past, but seems like at this point we only look for hwpoison from
> check_new_page_bad(). But a cleanup can come later.

Thanks for clarification. check_new_page_bad() is the only function interested
in hwpoison flag, so we had better move the hwpoison related code in bad_page()
to check_new_page_bad().

Thanks,
Naoya Horiguchi
---
>From c600b1ee6c36b3df6973f5365b4179c92f3c08e3 Mon Sep 17 00:00:00 2001
From: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Date: Wed, 18 May 2016 18:42:57 +0900
Subject: [PATCH v2] mm: check_new_page_bad() directly returns in __PG_HWPOISON
 case

Currently we check page->flags twice for "HWPoisoned" case of
check_new_page_bad(), which can cause a race with unpoisoning.
This race unnecessarily taints kernel with "BUG: Bad page state".
check_new_page_bad() is the only caller of bad_page() which is interested
in __PG_HWPOISON, so let's move the hwpoison related code in bad_page()
to it.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
---
 mm/page_alloc.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 5b269bc3eca7..59b938ddfb2d 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -522,12 +522,6 @@ static void bad_page(struct page *page, const char *reason,
 	static unsigned long nr_shown;
 	static unsigned long nr_unshown;
 
-	/* Don't complain about poisoned pages */
-	if (PageHWPoison(page)) {
-		page_mapcount_reset(page); /* remove PageBuddy */
-		return;
-	}
-
 	/*
 	 * Allow a burst of 60 reports, then keep quiet for that minute;
 	 * or allow a steady drip of one report per second.
@@ -1654,6 +1648,9 @@ static void check_new_page_bad(struct page *page)
 	if (unlikely(page->flags & __PG_HWPOISON)) {
 		bad_reason = "HWPoisoned (hardware-corrupted)";
 		bad_flags = __PG_HWPOISON;
+		/* Don't complain about hwpoisoned pages */
+		page_mapcount_reset(page); /* remove PageBuddy */
+		return;
 	}
 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
 		bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
-- 
2.5.5