mm-page_alloc-reduce-lock-sections-further.patch

From: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Fri, 3 Jul 2009 08:44:37 -0500
Subject: mm: page_alloc reduce lock sections further

Split out the pages which are to be freed into a separate list and
call free_pages_bulk() outside of the percpu page allocator locks.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 mm/page_alloc.c |   81 +++++++++++++++++++++++++++++++++++++++++---------------
 1 file changed, 60 insertions(+), 21 deletions(-)

Index: linux-stable/mm/page_alloc.c
===================================================================
--- linux-stable.orig/mm/page_alloc.c
+++ linux-stable/mm/page_alloc.c
@@ -637,7 +637,7 @@ static inline int free_pages_check(struc
 }
 
 /*
- * Frees a number of pages from the PCP lists
+ * Frees a number of pages which have been collected from the pcp lists.
  * Assumes all pages on list are in same zone, and of same order.
  * count is the number of pages to free.
  *
@@ -648,16 +648,42 @@ static inline int free_pages_check(struc
  * pinned" detection logic.
  */
 static void free_pcppages_bulk(struct zone *zone, int count,
-					struct per_cpu_pages *pcp)
+			       struct list_head *list)
 {
-	int migratetype = 0;
-	int batch_free = 0;
 	int to_free = count;
+	unsigned long flags;
 
-	spin_lock(&zone->lock);
+	spin_lock_irqsave(&zone->lock, flags);
 	zone->all_unreclaimable = 0;
 	zone->pages_scanned = 0;
 
+	while (!list_empty(list)) {
+		struct page *page = list_first_entry(list, struct page, lru);
+
+		/* must delete as __free_one_page list manipulates */
+		list_del(&page->lru);
+		/* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
+		__free_one_page(page, zone, 0, page_private(page));
+		trace_mm_page_pcpu_drain(page, 0, page_private(page));
+		to_free--;
+	}
+	WARN_ON(to_free != 0);
+	__mod_zone_page_state(zone, NR_FREE_PAGES, count);
+	spin_unlock_irqrestore(&zone->lock, flags);
+}
+
+/*
+ * Moves a number of pages from the PCP lists to free list which
+ * is freed outside of the locked region.
+ *
+ * Assumes all pages on list are in same zone, and of same order.
+ * count is the number of pages to free.
+ */
+static void isolate_pcp_pages(int to_free, struct per_cpu_pages *src,
+			      struct list_head *dst)
+{
+	int migratetype = 0, batch_free = 0;
+
 	while (to_free) {
 		struct page *page;
 		struct list_head *list;
@@ -673,7 +699,7 @@ static void free_pcppages_bulk(struct zo
 			batch_free++;
 			if (++migratetype == MIGRATE_PCPTYPES)
 				migratetype = 0;
-			list = &pcp->lists[migratetype];
+			list = &src->lists[migratetype];
 		} while (list_empty(list));
 
 		/* This is the only non-empty list. Free them all. */
@@ -682,27 +708,24 @@ static void free_pcppages_bulk(struct zo
 
 		do {
 			page = list_last_entry(list, struct page, lru);
-			/* must delete as __free_one_page list manipulates */
 			list_del(&page->lru);
-			/* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
-			__free_one_page(page, zone, 0, page_private(page));
-			trace_mm_page_pcpu_drain(page, 0, page_private(page));
+			list_add(&page->lru, dst);
 		} while (--to_free && --batch_free && !list_empty(list));
 	}
-	__mod_zone_page_state(zone, NR_FREE_PAGES, count);
-	spin_unlock(&zone->lock);
 }
 
 static void free_one_page(struct zone *zone, struct page *page, int order,
 				int migratetype)
 {
-	spin_lock(&zone->lock);
+	unsigned long flags;
+
+	spin_lock_irqsave(&zone->lock, flags);
 	zone->all_unreclaimable = 0;
 	zone->pages_scanned = 0;
 
 	__free_one_page(page, zone, order, migratetype);
 	__mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
-	spin_unlock(&zone->lock);
+	spin_unlock_irqrestore(&zone->lock, flags);
 }
 
 static bool free_pages_prepare(struct page *page, unsigned int order)
@@ -1168,6 +1191,7 @@ static int rmqueue_bulk(struct zone *zon
 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
 {
 	unsigned long flags;
+	LIST_HEAD(dst);
 	int to_drain;
 
 	local_lock_irqsave(pa_lock, flags);
@@ -1176,10 +1200,11 @@ void drain_zone_pages(struct zone *zone,
 	else
 		to_drain = pcp->count;
 	if (to_drain > 0) {
-		free_pcppages_bulk(zone, to_drain, pcp);
+		isolate_pcp_pages(to_drain, pcp, &dst);
 		pcp->count -= to_drain;
 	}
 	local_unlock_irqrestore(pa_lock, flags);
+	free_pcppages_bulk(zone, to_drain, &dst);
 }
 #endif
 
@@ -1198,16 +1223,21 @@ static void drain_pages(unsigned int cpu
 	for_each_populated_zone(zone) {
 		struct per_cpu_pageset *pset;
 		struct per_cpu_pages *pcp;
+		LIST_HEAD(dst);
+		int count;
 
 		cpu_lock_irqsave(cpu, flags);
 		pset = per_cpu_ptr(zone->pageset, cpu);
 
 		pcp = &pset->pcp;
-		if (pcp->count) {
-			free_pcppages_bulk(zone, pcp->count, pcp);
+		count = pcp->count;
+		if (count) {
+			isolate_pcp_pages(count, pcp, &dst);
 			pcp->count = 0;
 		}
 		cpu_unlock_irqrestore(cpu, flags);
+		if (count)
+			free_pcppages_bulk(zone, count, &dst);
 	}
 }
 
@@ -1348,8 +1378,15 @@ void free_hot_cold_page(struct page *pag
 		list_add(&page->lru, &pcp->lists[migratetype]);
 	pcp->count++;
 	if (pcp->count >= pcp->high) {
-		free_pcppages_bulk(zone, pcp->batch, pcp);
+		LIST_HEAD(dst);
+		int count;
+
+		isolate_pcp_pages(pcp->batch, pcp, &dst);
 		pcp->count -= pcp->batch;
+		count = pcp->batch;
+		local_unlock_irqrestore(pa_lock, flags);
+		free_pcppages_bulk(zone, count, &dst);
+		return;
 	}
 
 out:
@@ -5888,14 +5925,16 @@ static int __meminit __zone_pcp_update(v
 	for_each_possible_cpu(cpu) {
 		struct per_cpu_pageset *pset;
 		struct per_cpu_pages *pcp;
+		LIST_HEAD(dst);
 
 		pset = per_cpu_ptr(zone->pageset, cpu);
 		pcp = &pset->pcp;
 
 		cpu_lock_irqsave(cpu, flags);
-		if (pcp->count > 0)
-			free_pcppages_bulk(zone, pcp->count, pcp);
-		setup_pageset(pset, batch);
+		if (pcp->count > 0) {
+			isolate_pcp_pages(pcp->count, pcp, &dst);
+			free_pcppages_bulk(zone, pcp->count, &dst);
+		}
 		cpu_unlock_irqrestore(cpu, flags);
 	}
 	return 0;