Skip to contentSkip to conversation

mm: page_alloc: avoid marking zones full prematurely after zone_reclaim()

DISCUSSION Dan Carpenter 3 participants 6 messages Mar 27, 2013 – Apr 2, 2013 Memory management

lore
Share & tools
lore
#0 lore raw

Hello Mel Gorman,

The patch 290d1a3ce0ec: "mm: page_alloc: avoid marking zones full prematurely after zone_reclaim()" from Mar 23, 2013, leads to the following warning:
"mm/page_alloc.c:1957 get_page_from_freelist()
warn: bitwise AND condition is false here"

mm/page_alloc.c
1948 /*

  1949                                   * Failed to reclaim enough to meet watermark.
  1950                                   * Only mark the zone full if checking the min
  1951                                   * watermark or if we failed to reclaim just
  1952                                   * 1<<order pages or else the page allocator
  1953                                   * fastpath will prematurely mark zones full
  1954                                   * when the watermark is between the low and
  1955                                   * min watermarks.
  1956                                   */
  1957                                  if ((alloc_flags & ALLOC_WMARK_MIN) ||
                                                           ^^^^^^^^^^^^^^^

This is zero.

1958 ret == ZONE_RECLAIM_SOME)
1959 goto this_zone_full;

[snip]

  2333  static inline int
  2334  gfp_to_alloc_flags(gfp_t gfp_mask)
  2335  {
  2336          int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
                                  ^^^^^^^^^^^^^^^

2337 const gfp_t wait = gfp_mask & __GFP_WAIT;

regards,
dan carpenter

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#1 lore raw

[Adding Andrew into CC]

Hi Dan,

Quoted 7 lines · Dan Carpenter

Hello Mel Gorman,

The patch 290d1a3ce0ec: "mm: page_alloc: avoid marking zones full prematurely after zone_reclaim()" from Mar 23, 2013, leads to the following warning:
"mm/page_alloc.c:1957 get_page_from_freelist()
warn: bitwise AND condition is false here"

Dohh, I have totally missed this during review and I managed to burn myself on the similar issue in the past (gfp & GFP_NOWAIT). The follow up fix is bellow

Quoted 17 lines

mm/page_alloc.c
1948 /*

  1949                                   * Failed to reclaim enough to meet watermark.
  1950                                   * Only mark the zone full if checking the min
  1951                                   * watermark or if we failed to reclaim just
  1952                                   * 1<<order pages or else the page allocator
  1953                                   * fastpath will prematurely mark zones full
  1954                                   * when the watermark is between the low and
  1955                                   * min watermarks.
  1956                                   */
  1957                                  if ((alloc_flags & ALLOC_WMARK_MIN) ||
                                                           ^^^^^^^^^^^^^^^

This is zero.

1958 ret == ZONE_RECLAIM_SOME)
1959 goto this_zone_full;

Quoted 8 lines

[snip]

  2333  static inline int
  2334  gfp_to_alloc_flags(gfp_t gfp_mask)
  2335  {
  2336          int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
                                  ^^^^^^^^^^^^^^^

2337 const gfp_t wait = gfp_mask & __GFP_WAIT;

---

#2 lore raw

And just for record. I wanted to give credit to Dan for reporting this but I have no idea how when this gets merged into the original patch which is still sitting in the -mm queue. I will leave that to Andrew ;)

Quoted 40 lines · Michal Hocko

From b60e75c65b855a0df827a28a509e6761b4cf45dd Mon Sep 17 00:00:00 2001
From: Michal Hocko <[email protected]>
Date: Wed, 27 Mar 2013 17:53:09 +0100
Subject: [PATCH] mm:
mm-page_alloc-avoid-marking-zones-full-prematurely-after-zone_reclaim-fix

Dan Carpenter has reported that (alloc_flags & ALLOC_WMARK_MIN) test doesn't make much sense as the flag is 0 and it is in fact intended for wmark indexing rather than being used as a flag.

---
 mm/page_alloc.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/page_alloc.c +1 -1
index aa4b5c2..071e66a 100644
@@ -1953,7 +1953,7 @@ zonelist_scan:
 				 * when the watermark is between the low and
 				 * min watermarks.
 				 */
-				if ((alloc_flags & ALLOC_WMARK_MIN) ||
+				if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
 				    ret == ZONE_RECLAIM_SOME)
 					goto this_zone_full;
 
Signature · 11 lines
1.7.10.4

-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Signature · 8 lines
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#4 lore raw
Quoted 5 lines · Dan Carpenter

I still don't understand the code in gfp_to_alloc_flags().

	int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;

ORing with zero is odd.

It is a named zero ;) Something similar like GFP_NOWAIT AFAIU.

Signature · 8 lines
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#5 lore raw
In reply to Dan Carpenter
Quoted 6 lines · Dan Carpenter

I still don't understand the code in gfp_to_alloc_flags().

	int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;

ORing with zero is odd.

Thanks Dan for the report and thanks Michal for fixing it. I was offline for last week which lead to my tardy response.

The odditiy is that ALLOC_WMARK_MIN is not treated as a flag but as an offset within the zone->wmark so it starts as 0. It could have been written as

	int alloc_flags = ALLOC_CPUSET;

but then it would be easy to forget that in this path we are using the MIN watermark.

The "flag" is used as an offset because it eliminated a number of branches in the page allocator and was a micro-optimisation at the time.

Signature · 8 lines
Mel Gorman
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>