linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Mel Gorman <mel@csn.ul.ie>
To: Mel Gorman <mel@csn.ul.ie>,
	Linux Memory Management List <linux-mm@kvack.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>,
	Rik van Riel <riel@redhat.com>,
	KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>,
	Christoph Lameter <cl@linux-foundation.org>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Nick Piggin <npiggin@suse.de>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Lin Ming <ming.m.lin@intel.com>,
	Zhang Yanmin <yanmin_zhang@linux.intel.com>,
	Peter Zijlstra <peterz@infradead.org>
Subject: [PATCH 28/35] Batch free pages from migratetype per-cpu lists
Date: Mon, 16 Mar 2009 09:46:23 +0000	[thread overview]
Message-ID: <1237196790-7268-29-git-send-email-mel@csn.ul.ie> (raw)
In-Reply-To: <1237196790-7268-1-git-send-email-mel@csn.ul.ie>

When the PCP lists are too large, a number of pages are freed in bulk.
Currently the free lists are examined in a round-robin fashion but this
touches more cache lines than necessary. This patch frees pages from one
list at a time and uses the migratetype most recently used as the
starting point.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
---
 mm/page_alloc.c |   36 +++++++++++++++++++++++-------------
 1 files changed, 23 insertions(+), 13 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 3516b87..edadab1 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -547,32 +547,42 @@ static inline void bulk_add_pcp_page(struct per_cpu_pages *pcp,
  * pinned" detection logic.
  */
 static void free_pcppages_bulk(struct zone *zone, int count,
-					struct per_cpu_pages *pcp)
+					struct per_cpu_pages *pcp,
+					int migratetype)
 {
-	int migratetype = 0;
 	unsigned int freed = 0;
+	unsigned int bulkcount;
+	struct list_head *list;
 
 	spin_lock(&zone->lock);
 	zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
 	zone->pages_scanned = 0;
 
+	list = &pcp->lists[migratetype];
+	bulkcount = 1 + (count / (MIGRATE_PCPTYPES * 2));
 	while (freed < count) {
 		struct page *page;
-		struct list_head *list;
+		int thisfreed;
 
-		/* Remove pages from lists in a round-robin fashion */
-		do {
-			if (migratetype == MIGRATE_PCPTYPES)
+		/*
+		 * Move to another migratetype if this list is depleted or
+		 * we've freed enough in this batch
+		 */
+		while (list_empty(list) || bulkcount < 0) {
+			bulkcount = 1 + (count / (MIGRATE_PCPTYPES * 2));
+			if (++migratetype == MIGRATE_PCPTYPES)
 				migratetype = 0;
 			list = &pcp->lists[migratetype];
-			migratetype++;
-		} while (list_empty(list));
+		}
 
+		/* Remove from list and update counters */
 		page = list_entry(list->prev, struct page, lru);
 		rmv_pcp_page(pcp, page);
+		thisfreed = 1 << page->index;
+		freed += thisfreed;
+		bulkcount -= thisfreed;
 
-		freed += 1 << page->index;
-		__free_one_page(page, zone, page->index, page_private(page));
+		__free_one_page(page, zone, page->index, migratetype);
 	}
 	spin_unlock(&zone->lock);
 
@@ -969,7 +979,7 @@ void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
 		to_drain = pcp->batch;
 	else
 		to_drain = pcp->count;
-	free_pcppages_bulk(zone, to_drain, pcp);
+	free_pcppages_bulk(zone, to_drain, pcp, 0);
 	local_irq_restore(flags);
 }
 #endif
@@ -997,7 +1007,7 @@ static void drain_pages(unsigned int cpu)
 
 		pcp = &pset->pcp;
 		local_irq_save(flags);
-		free_pcppages_bulk(zone, pcp->count, pcp);
+		free_pcppages_bulk(zone, pcp->count, pcp, 0);
 		BUG_ON(pcp->count);
 		local_irq_restore(flags);
 	}
@@ -1110,7 +1120,7 @@ static void free_hot_cold_page(struct page *page, int order, int cold)
 	page->index = order;
 	add_pcp_page(pcp, page, cold);
 	if (pcp->count >= pcp->high)
-		free_pcppages_bulk(zone, pcp->batch, pcp);
+		free_pcppages_bulk(zone, pcp->batch, pcp, migratetype);
 
 out:
 	local_irq_restore(flags);
-- 
1.5.6.5

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

  parent reply	other threads:[~2009-03-16  9:44 UTC|newest]

Thread overview: 94+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2009-03-16  9:45 [PATCH 00/35] Cleanup and optimise the page allocator V3 Mel Gorman
2009-03-16  9:45 ` [PATCH 01/35] Replace __alloc_pages_internal() with __alloc_pages_nodemask() Mel Gorman
2009-03-16 15:49   ` Christoph Lameter
2009-03-16  9:45 ` [PATCH 02/35] Do not sanity check order in the fast path Mel Gorman
2009-03-16 15:52   ` Christoph Lameter
2009-03-16  9:45 ` [PATCH 03/35] Do not check NUMA node ID when the caller knows the node is valid Mel Gorman
2009-03-16  9:45 ` [PATCH 04/35] Check only once if the zonelist is suitable for the allocation Mel Gorman
2009-03-16  9:46 ` [PATCH 05/35] Break up the allocator entry point into fast and slow paths Mel Gorman
2009-03-16  9:46 ` [PATCH 06/35] Move check for disabled anti-fragmentation out of fastpath Mel Gorman
2009-03-16 15:54   ` Christoph Lameter
2009-03-16  9:46 ` [PATCH 07/35] Check in advance if the zonelist needs additional filtering Mel Gorman
2009-03-16  9:46 ` [PATCH 08/35] Calculate the preferred zone for allocation only once Mel Gorman
2009-03-16  9:46 ` [PATCH 09/35] Calculate the migratetype " Mel Gorman
2009-03-16  9:46 ` [PATCH 10/35] Calculate the alloc_flags " Mel Gorman
2009-03-16  9:46 ` [PATCH 11/35] Calculate the cold parameter " Mel Gorman
2009-03-16  9:46 ` [PATCH 12/35] Remove a branch by assuming __GFP_HIGH == ALLOC_HIGH Mel Gorman
2009-03-16  9:46 ` [PATCH 13/35] Inline __rmqueue_smallest() Mel Gorman
2009-03-16  9:46 ` [PATCH 14/35] Inline buffered_rmqueue() Mel Gorman
2009-03-16  9:46 ` [PATCH 15/35] Inline __rmqueue_fallback() Mel Gorman
2009-03-16 15:57   ` Christoph Lameter
2009-03-16 16:25     ` Mel Gorman
2009-03-16  9:46 ` [PATCH 16/35] Save text by reducing call sites of __rmqueue() Mel Gorman
2009-03-16  9:46 ` [PATCH 17/35] Do not call get_pageblock_migratetype() more than necessary Mel Gorman
2009-03-16 16:00   ` Christoph Lameter
2009-03-16  9:46 ` [PATCH 18/35] Do not disable interrupts in free_page_mlock() Mel Gorman
2009-03-16 16:05   ` Christoph Lameter
2009-03-16 16:29     ` Mel Gorman
2009-03-16  9:46 ` [PATCH 19/35] Do not setup zonelist cache when there is only one node Mel Gorman
2009-03-16 16:06   ` Christoph Lameter
2009-03-16  9:46 ` [PATCH 20/35] Use a pre-calculated value for num_online_nodes() Mel Gorman
2009-03-16 11:42   ` Nick Piggin
2009-03-16 11:46     ` Nick Piggin
2009-03-16 16:08   ` Christoph Lameter
2009-03-16 16:36     ` Mel Gorman
2009-03-16 16:47       ` Christoph Lameter
2009-03-18 15:08         ` Mel Gorman
2009-03-18 16:58           ` Christoph Lameter
2009-03-18 18:01             ` Mel Gorman
2009-03-18 19:10               ` Christoph Lameter
2009-03-19 20:43                 ` Christoph Lameter
2009-03-19 21:29                   ` Mel Gorman
2009-03-19 22:22                     ` Christoph Lameter
2009-03-19 22:33                       ` Mel Gorman
2009-03-19 22:42                         ` Christoph Lameter
2009-03-19 22:52                           ` Mel Gorman
2009-03-19 22:06                   ` Mel Gorman
2009-03-19 22:39                     ` Christoph Lameter
2009-03-19 22:21                   ` Mel Gorman
2009-03-19 22:24                     ` Christoph Lameter
2009-03-19 23:04                       ` Mel Gorman
2009-03-16  9:46 ` [PATCH 21/35] Do not check for compound pages during the page allocator sanity checks Mel Gorman
2009-03-16 16:09   ` Christoph Lameter
2009-03-16  9:46 ` [PATCH 22/35] Use allocation flags as an index to the zone watermark Mel Gorman
2009-03-16 16:11   ` Christoph Lameter
2009-03-16  9:46 ` [PATCH 23/35] Update NR_FREE_PAGES only as necessary Mel Gorman
2009-03-16 16:17   ` Christoph Lameter
2009-03-16 16:42     ` Mel Gorman
2009-03-16 16:48       ` Christoph Lameter
2009-03-16 16:58         ` Mel Gorman
2009-03-16  9:46 ` [PATCH 24/35] Convert gfp_zone() to use a table of precalculated values Mel Gorman
2009-03-16 16:19   ` Christoph Lameter
2009-03-16 16:45     ` Mel Gorman
2009-03-16  9:46 ` [PATCH 25/35] Re-sort GFP flags and fix whitespace alignment for easier reading Mel Gorman
2009-03-16  9:46 ` [PATCH 26/35] Use the per-cpu allocator for orders up to PAGE_ALLOC_COSTLY_ORDER Mel Gorman
2009-03-16 16:26   ` Christoph Lameter
2009-03-16 16:47     ` Mel Gorman
2009-03-16  9:46 ` [PATCH 27/35] Split per-cpu list into one-list-per-migrate-type Mel Gorman
2009-03-16  9:46 ` Mel Gorman [this message]
2009-03-16  9:46 ` [PATCH 29/35] Do not store the PCP high and batch watermarks in the per-cpu structure Mel Gorman
2009-03-16 16:30   ` Christoph Lameter
2009-03-16  9:46 ` [PATCH 30/35] Skip the PCP list search by counting the order and type of pages on list Mel Gorman
2009-03-16 16:31   ` Christoph Lameter
2009-03-16 16:51     ` Mel Gorman
2009-03-16  9:46 ` [PATCH 31/35] Optimistically check the first page on the PCP free list is suitable Mel Gorman
2009-03-16 16:33   ` Christoph Lameter
2009-03-16 16:52     ` Mel Gorman
2009-03-16  9:46 ` [PATCH 32/35] Inline next_zones_zonelist() of the zonelist scan in the fastpath Mel Gorman
2009-03-16  9:46 ` [PATCH 33/35] Do not merge buddies until they are needed by a high-order allocation or anti-fragmentation Mel Gorman
2009-03-16  9:46 ` [PATCH 34/35] Allow compound pages to be stored on the PCP lists Mel Gorman
2009-03-16 16:47   ` Christoph Lameter
2009-03-16  9:46 ` [PATCH 35/35] Allow up to 4MB PCP lists due to compound pages Mel Gorman
2009-03-16 10:40 ` [PATCH 00/35] Cleanup and optimise the page allocator V3 Nick Piggin
2009-03-16 11:19   ` Mel Gorman
2009-03-16 11:33     ` Nick Piggin
2009-03-16 12:02       ` Mel Gorman
2009-03-16 12:25         ` Nick Piggin
2009-03-16 13:32           ` Mel Gorman
2009-03-16 15:53             ` Nick Piggin
2009-03-16 16:56               ` Mel Gorman
2009-03-16 17:05                 ` Nick Piggin
2009-03-18 15:07                   ` Mel Gorman
2009-03-16 11:45 ` Nick Piggin
2009-03-16 12:11   ` Mel Gorman
2009-03-16 12:28     ` Nick Piggin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1237196790-7268-29-git-send-email-mel@csn.ul.ie \
    --to=mel@csn.ul.ie \
    --cc=cl@linux-foundation.org \
    --cc=hannes@cmpxchg.org \
    --cc=kosaki.motohiro@jp.fujitsu.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=ming.m.lin@intel.com \
    --cc=npiggin@suse.de \
    --cc=penberg@cs.helsinki.fi \
    --cc=peterz@infradead.org \
    --cc=riel@redhat.com \
    --cc=yanmin_zhang@linux.intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).