From 2b494fe0b8811f60e2ac80901235425b51237ac1 Mon Sep 17 00:00:00 2001
From: Paul Gortmaker <paul.gortmaker@windriver.com>
Date: Sun, 23 Jan 2011 19:17:40 -0500
Subject: [PATCH] page_alloc: augment percpu pages support for newer kernels

This is a follow-on to:

	mm-page_alloc-rt-friendly-per-cpu-pages.patch
	(tip ff3fd6afd788760c846a2f4449487debb6c4b0ac)

and:

	mm-page_alloc-reduce-lock-sections-further.patch
	(tip 46167aec68f48cbbeff23cae9173bc4d19a7bcda)

which is embedded in the big merge up to 33rt (5f854cfc024)

At some point it probably makes sense to munge all three of
these together -- whee.

See the constituent bits spread around in a git tip repo with:

   git diff 5f854cfc024622e4aae14d7cf422f6ff86278688^2 \
	5f854cfc024622e4aae14d7cf422f6ff86278688 mm/page_alloc.c

(the above merge changeset is a superset of this delta).

You can find the origin of this change in the tip merge commit:

     commit 5f854cfc024622e4aae14d7cf422f6ff86278688
     Merge: cc24da0 4ec62b2
     Author: Thomas Gleixner <tglx@linutronix.de>
     Date:   Sun Feb 21 20:17:22 2010 +0100

         Forward to 2.6.33-rc8

         Merge branch 'linus' into rt/head with a pile of conflicts.

         Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Normally there are not significant changes/additions in a merge commit that
are not from any other "normal" commit.  But in this case there are, so
break them out into separate explicit commits.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 75d0e57..bde9ea1 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -170,7 +170,7 @@ static inline void __lock_cpu_pcp(unsigned long *flags, int cpu)
 {
 #ifdef CONFIG_PREEMPT_RT
 	spin_lock(&__get_cpu_lock(pcp_locks, cpu));
-	flags = 0;
+	*flags = 0;
 #else
 	local_irq_save(*flags);
 #endif
@@ -571,18 +571,48 @@ static inline int free_pages_check(struct page *page)
  * 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 = 0;
-	int batch_free = 0;
 	unsigned long flags;
 
-
 	spin_lock_irqsave(&zone->lock, flags);
 	zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
 	zone->pages_scanned = 0;
 
 	__mod_zone_page_state(zone, NR_FREE_PAGES, count);
+
+	for (migratetype =0; migratetype < MIGRATE_PCPTYPES; migratetype++) {
+		struct list_head *list = &pcp->lists[migratetype];
+
+		while (!list_empty(list)) {
+			struct page *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));
+#ifdef CONFIG_PREEMPT_RT
+			cond_resched_lock(&zone->lock);
+#endif
+			count--;
+		}
+	}
+	WARN_ON(count != 0);
+	spin_unlock_irqrestore(&zone->lock, flags);
+}
+
+static void isolate_pcp_pages(int count, struct per_cpu_pages *src,
+			      struct per_cpu_pages *dst)
+{
+	int migratetype, batch_free = 0;
+
+	for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
+		INIT_LIST_HEAD(&dst->lists[migratetype]);
+	migratetype = 0;
+
 	while (count) {
 		struct page *page;
 		struct list_head *list;
@@ -598,22 +628,16 @@ static void free_pcppages_bulk(struct zone *zone, int count,
 			batch_free++;
 			if (++migratetype == MIGRATE_PCPTYPES)
 				migratetype = 0;
-			list = &pcp->lists[migratetype];
+			list = &src->lists[migratetype];
 		} while (list_empty(list));
 
 		do {
-			page = list_entry(list->prev, struct page, lru);
+			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));
-#ifdef CONFIG_PREEMPT_RT
-			cond_resched_lock(&zone->lock);
-#endif
+			list_add(&page->lru, &dst->lists[migratetype]);
 		} while (--count && --batch_free && !list_empty(list));
 	}
-	spin_unlock_irqrestore(&zone->lock, flags);
 }
 
 static void free_one_page(struct zone *zone, struct page *page, int order,
@@ -1021,16 +1045,6 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
 	return i;
 }
 
-static void
-isolate_pcp_pages(int count, struct list_head *src, struct list_head *dst)
-{
-	while (count--) {
-		struct page *page = list_last_entry(src, struct page, lru);
-		list_move(&page->lru, dst);
-	}
-}
-
-
 #ifdef CONFIG_NUMA
 /*
  * Called from the vmstat counter updater to drain pagesets of this
@@ -1042,20 +1056,19 @@ isolate_pcp_pages(int count, struct list_head *src, struct list_head *dst)
  */
 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
 {
-	LIST_HEAD(free_list);
+	struct per_cpu_pages dst;
 	unsigned long flags;
-	int to_drain;
-	int this_cpu;
+	int to_drain, this_cpu;
 
 	lock_cpu_pcp(&flags, &this_cpu);
 	if (pcp->count >= pcp->batch)
 		to_drain = pcp->batch;
 	else
 		to_drain = pcp->count;
-	isolate_pcp_pages(to_drain, &pcp->list, &free_list);
+	isolate_pcp_pages(to_drain, pcp, &dst);
 	pcp->count -= to_drain;
 	unlock_cpu_pcp(flags, this_cpu);
-	free_pages_bulk(zone, to_drain, &free_list, 0);
+	free_pcppages_bulk(zone, to_drain, &dst);
 }
 #endif
 
@@ -1073,8 +1086,7 @@ static void drain_pages(unsigned int cpu)
 
 	for_each_populated_zone(zone) {
 		struct per_cpu_pageset *pset;
-		struct per_cpu_pages *pcp;
-		LIST_HEAD(free_list);
+		struct per_cpu_pages *pcp, dst;
 		int count;
 
 		__lock_cpu_pcp(&flags, cpu);
@@ -1084,12 +1096,13 @@ static void drain_pages(unsigned int cpu)
 			WARN_ON(1);
 			continue;
 		}
+
 		pcp = &pset->pcp;
-		isolate_pcp_pages(pcp->count, &pcp->list, &free_list);
+		isolate_pcp_pages(pcp->count, pcp, &dst);
 		count = pcp->count;
 		pcp->count = 0;
 		unlock_cpu_pcp(flags, cpu);
-		free_pages_bulk(zone, count, &free_list, 0);
+		free_pcppages_bulk(zone, count, &dst);
 	}
 }
 
@@ -1194,8 +1207,8 @@ static void free_hot_cold_page(struct page *page, int cold)
 	struct per_cpu_pageset *pset;
 	struct per_cpu_pages *pcp;
 	unsigned long flags;
-	int migratetype;
-	int count, this_cpu, wasMlocked = __TestClearPageMlocked(page);
+	int migratetype, this_cpu, count;
+	int wasMlocked = __TestClearPageMlocked(page);
 
 	kmemcheck_free_shadow(page, 0);
 
@@ -1240,15 +1253,17 @@ static void free_hot_cold_page(struct page *page, int cold)
 		list_add(&page->lru, &pcp->lists[migratetype]);
 	pcp->count++;
 	if (pcp->count >= pcp->high) {
-		LIST_HEAD(free_list);
+		struct per_cpu_pages dst;
 
-		isolate_pcp_pages(pcp->batch, &pcp->list, &free_list);
+		isolate_pcp_pages(pcp->batch, pcp, &dst);
 		pcp->count -= pcp->batch;
 		count = pcp->batch;
 		put_zone_pcp(zone, flags, this_cpu);
-		free_pages_bulk(zone, count, &free_list, 0);
-	} else
-out:		put_zone_pcp(zone, flags, this_cpu);
+		free_pcppages_bulk(zone, count, &dst);
+		return;
+	}
+out:
+	put_zone_pcp(zone, flags, this_cpu);
 }
 
 void free_hot_page(struct page *page)
@@ -1305,11 +1320,10 @@ again:
 	pset = get_zone_pcp(zone, &flags, &this_cpu);
 
 	if (likely(order == 0)) {
-		struct list_head *list;
 		struct per_cpu_pages *pcp = &pset->pcp;
+		struct list_head *list;
 
 		list = &pcp->lists[migratetype];
-
 		if (list_empty(list)) {
 			pcp->count += rmqueue_bulk(zone, 0,
 					pcp->batch, list,
-- 
1.7.1.1