buddy allocator without bitmap [3/4]

diff -puN mm/page_alloc.c~eliminate-bitmap-free mm/page_alloc.c
@@ -157,6 +157,27 @@ static void destroy_compound_page(struct
 #endif		/* CONFIG_HUGETLB_PAGE */
 
 /*
+ * This function checks whether a page is free && is the buddy
+ * we can do coalesce if
+ * (a) the buddy is free and
+ * (b) the buddy is on the buddy system
+ * (c) the buddy has the same order.
+ * for recording page's order, we use private field and PG_private.
+ *
+ * Because page_count(page) == 0, and zone->lock is aquired.
+ * Atomic page->flags operation is needless here.
+ */
+static inline int page_is_buddy(struct page *page, int order)
+{
+	if ((page->flags & (1 << PG_private)) &&
+	    (page_order(page) == order) &&
+	    !(page->flags & (1 << PG_reserved)) &&
+            page_count(page) == 0)
+		return 1;
+	return 0;
+}
+
+/*
  * Freeing function for a buddy system allocator.
  *
  * The concept of a buddy system is to maintain direct-mapped table

@@ -168,9 +189,12 @@ static void destroy_compound_page(struct
  * at the bottom level available, and propagating the changes upward
  * as necessary, plus some accounting needed to play nicely with other
  * parts of the VM system.
- * At each level, we keep one bit for each pair of blocks, which
- * is set to 1 iff only one of the pair is allocated.  So when we
- * are allocating or freeing one, we can derive the state of the
+ *
+ * At each level, we keep a list of pages, which are head of chunk of
+ * pages at the level. A page, which is a head of chunks, has its order
+ * in page structure itself and PG_private flag is set. we can get an
+ * order of a page by calling  page_order().
+ * So we are allocating or freeing one, we can derive the state of the
  * other.  That is, if we allocate a small block, and both were
  * free, the remainder of the region must be split into blocks.
  * If a block is freed, and its buddy is also free, then this

@@ -180,42 +204,68 @@ static void destroy_compound_page(struct
  */
 
 static inline void __free_pages_bulk (struct page *page, struct page *base,
-		struct zone *zone, struct free_area *area, unsigned int order)
+		struct zone *zone, unsigned int order)
 {
-	unsigned long page_idx, index, mask;
-
+	unsigned long page_idx, mask;
+	struct page *coalesced_page;
+	
 	if (order)
 		destroy_compound_page(page, order);
+
 	mask = (~0UL) << order;
 	page_idx = page - base;
 	if (page_idx & ~mask)
 		BUG();
-	index = page_idx >> (1 + order);
-
 	zone->free_pages += 1 << order;
+	BUG_ON(bad_range(zone,page));
+
 	while (order < MAX_ORDER-1) {
-		struct page *buddy1, *buddy2;
+		struct page *buddy;
+		
+		buddy = base + (page_idx ^ (1 << order));
 
-		BUG_ON(area >= zone->free_area + MAX_ORDER);
-		if (!__test_and_change_bit(index, area->map))
+		/*
+		 * quick check, if order < zone->aligned_order
+		 * every page has its buddy in this order.
+		 * buddy is guaranteed to be valid page.
+		 */
+		if (order >= zone->aligned_order) {
+			if (zone->nr_mem_map > 1) {
+				/*
+				 * there may be hole in zone's memmap &&
+				 * hole is not aligned in this order.
+				 * currently, I think CONFIG_VIRTUAL_MEM_MAP
+				 * case is only case to reach here.
+				 * Is there any other case ?
+				 */
+				/*
+				 * Is there better call than pfn_valid ?
+				 */
+				if (!pfn_valid(zone->zone_start_pfn
+					       + (page_idx ^ (1 << order))))
+					break;
+			}
+			/* we need range check */
+			if (bad_range(zone, buddy))
+				break;
+		}
+		if (!page_is_buddy(buddy, order))
 			/*
 			 * the buddy page is still allocated.
 			 */
 			break;
-
-		/* Move the buddy up one level. */
-		buddy1 = base + (page_idx ^ (1 << order));
-		buddy2 = base + page_idx;
-		BUG_ON(bad_range(zone, buddy1));
-		BUG_ON(bad_range(zone, buddy2));
-		list_del(&buddy1->lru);
-		mask <<= 1;
 		order++;
-		area++;
-		index >>= 1;
+		mask <<= 1;
 		page_idx &= mask;
-	}
-	list_add(&(base + page_idx)->lru, &area->free_list);
+		list_del(&buddy->lru);
+		/* for propriety of PG_private bit, we clear it */
+		buddy->flags &= ~(1 << PG_private);
+	}
+	/* record the final order of the page */
+	coalesced_page = base + page_idx;
+	coalesced_page->flags |= (1 << PG_private);
+	coalesced_page->private = order;
+	list_add(&coalesced_page->lru, &zone->free_area[order].free_list);
 }
 
 static inline void free_pages_check(const char *function, struct page *page)

@@ -253,12 +303,10 @@ free_pages_bulk(struct zone *zone, int c
 		struct list_head *list, unsigned int order)
 {
 	unsigned long flags;
-	struct free_area *area;
 	struct page *base, *page = NULL;
 	int ret = 0;
 
 	base = zone->zone_mem_map;
-	area = zone->free_area + order;
 	spin_lock_irqsave(&zone->lock, flags);
 	zone->all_unreclaimable = 0;
 	zone->pages_scanned = 0;

@@ -266,7 +314,7 @@ free_pages_bulk(struct zone *zone, int c
 		page = list_entry(list->prev, struct page, lru);
 		/* have to delete it as __free_pages_bulk list manipulates */
 		list_del(&page->lru);
-		__free_pages_bulk(page, base, zone, area, order);
+		__free_pages_bulk(page, base, zone, order);
 		ret++;
 	}
 	spin_unlock_irqrestore(&zone->lock, flags);