mm: accurately calculate zone->managed_pages for highmem zones

Commit "mm: introduce new field 'managed_pages' to struct zone" assumes
that all highmem pages will be freed into the buddy system by function
mem_init().  But that's not always true, some architectures may reserve
some highmem pages during boot.  For example PPC may allocate highmem
pages for giagant HugeTLB pages, and several architectures have code to
check PageReserved flag to exclude highmem pages allocated during boot
when freeing highmem pages into the buddy system.

So treat highmem pages in the same way as normal pages, that is to:
1) reset zone->managed_pages to zero in mem_init().
2) recalculate managed_pages when freeing pages into the buddy system.

Signed-off-by: Jiang Liu <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: Tejun Heo <[email protected]>
Cc: Joonsoo Kim <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Minchan Kim <[email protected]>
Cc: Kamezawa Hiroyuki <[email protected]>
Cc: Marek Szyprowski <[email protected]>
Cc: "Michael S. Tsirkin" <[email protected]>
Cc: <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: Catalin Marinas <[email protected]>
Cc: Chris Metcalf <[email protected]>
Cc: David Howells <[email protected]>
Cc: Geert Uytterhoeven <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Jeremy Fitzhardinge <[email protected]>
Cc: Jianguo Wu <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Michel Lespinasse <[email protected]>
Cc: Rik van Riel <[email protected]>
Cc: Rusty Russell <[email protected]>
Cc: Tang Chen <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Wen Congyang <[email protected]>
Cc: Will Deacon <[email protected]>
Cc: Yasuaki Ishimatsu <[email protected]>
Cc: Russell King <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

6 files changed, 48 insertions, 28 deletions

diff --git a/arch/metag/mm/init.c b/arch/metag/mm/init.c
index 5e2238dd72e0..d7595f58fad5 100644
--- a/arch/metag/mm/init.c
+++ b/arch/metag/mm/init.c
@@ -380,6 +380,12 @@ void __init mem_init(void)
 
 #ifdef CONFIG_HIGHMEM
 	unsigned long tmp;
+
+	/*
+	 * Explicitly reset zone->managed_pages because highmem pages are
+	 * freed before calling free_all_bootmem_node();
+	 */
+	reset_all_zones_managed_pages();
 	for (tmp = highstart_pfn; tmp < highend_pfn; tmp++)
 		free_highmem_page(pfn_to_page(tmp));
 	num_physpages += totalhigh_pages;
diff --git a/arch/x86/mm/highmem_32.c b/arch/x86/mm/highmem_32.c
index 252b8f5489ba..4500142bc4aa 100644
--- a/arch/x86/mm/highmem_32.c
+++ b/arch/x86/mm/highmem_32.c
@@ -1,6 +1,7 @@
 #include <linux/highmem.h>
 #include <linux/module.h>
 #include <linux/swap.h> /* for totalram_pages */
+#include <linux/bootmem.h>
 
 void *kmap(struct page *page)
 {
@@ -121,6 +122,11 @@ void __init set_highmem_pages_init(void)
 	struct zone *zone;
 	int nid;
 
+	/*
+	 * Explicitly reset zone->managed_pages because set_highmem_pages_init()
+	 * is invoked before free_all_bootmem()
+	 */
+	reset_all_zones_managed_pages();
 	for_each_zone(zone) {
 		unsigned long zone_start_pfn, zone_end_pfn;
 
diff --git a/include/linux/bootmem.h b/include/linux/bootmem.h
index 5f0b0e1f7c08..0e48c3221d82 100644
--- a/include/linux/bootmem.h
+++ b/include/linux/bootmem.h
@@ -46,6 +46,7 @@ extern unsigned long init_bootmem(unsigned long addr, unsigned long memend);
 
 extern unsigned long free_all_bootmem_node(pg_data_t *pgdat);
 extern unsigned long free_all_bootmem(void);
+extern void reset_all_zones_managed_pages(void);
 
 extern void free_bootmem_node(pg_data_t *pgdat,
 			      unsigned long addr,
diff --git a/mm/bootmem.c b/mm/bootmem.c
index 2b0bcb019ec2..eb792323187b 100644
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -241,20 +241,26 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 	return count;
 }
 
-static void reset_node_lowmem_managed_pages(pg_data_t *pgdat)
+static int reset_managed_pages_done __initdata;
+
+static inline void __init reset_node_managed_pages(pg_data_t *pgdat)
 {
 	struct zone *z;
 
-	/*
-	 * In free_area_init_core(), highmem zone's managed_pages is set to
-	 * present_pages, and bootmem allocator doesn't allocate from highmem
-	 * zones. So there's no need to recalculate managed_pages because all
-	 * highmem pages will be managed by the buddy system. Here highmem
-	 * zone also includes highmem movable zone.
-	 */
+	if (reset_managed_pages_done)
+		return;
+
 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
-		if (!is_highmem(z))
-			z->managed_pages = 0;
+		z->managed_pages = 0;
+}
+
+void __init reset_all_zones_managed_pages(void)
+{
+	struct pglist_data *pgdat;
+
+	for_each_online_pgdat(pgdat)
+		reset_node_managed_pages(pgdat);
+	reset_managed_pages_done = 1;
 }
 
 /**
@@ -266,7 +272,7 @@ static void reset_node_lowmem_managed_pages(pg_data_t *pgdat)
 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 {
 	register_page_bootmem_info_node(pgdat);
-	reset_node_lowmem_managed_pages(pgdat);
+	reset_node_managed_pages(pgdat);
 	return free_all_bootmem_core(pgdat->bdata);
 }
 
@@ -279,10 +285,8 @@ unsigned long __init free_all_bootmem(void)
 {
 	unsigned long total_pages = 0;
 	bootmem_data_t *bdata;
-	struct pglist_data *pgdat;
 
-	for_each_online_pgdat(pgdat)
-		reset_node_lowmem_managed_pages(pgdat);
+	reset_all_zones_managed_pages();
 
 	list_for_each_entry(bdata, &bdata_list, list)
 		total_pages += free_all_bootmem_core(bdata);
diff --git a/mm/nobootmem.c b/mm/nobootmem.c
index bdd3fa2fc73b..0ae8d91365af 100644
--- a/mm/nobootmem.c
+++ b/mm/nobootmem.c
@@ -137,20 +137,25 @@ static unsigned long __init free_low_memory_core_early(void)
 	return count;
 }
 
-static void reset_node_lowmem_managed_pages(pg_data_t *pgdat)
+static int reset_managed_pages_done __initdata;
+
+static inline void __init reset_node_managed_pages(pg_data_t *pgdat)
 {
 	struct zone *z;
 
-	/*
-	 * In free_area_init_core(), highmem zone's managed_pages is set to
-	 * present_pages, and bootmem allocator doesn't allocate from highmem
-	 * zones. So there's no need to recalculate managed_pages because all
-	 * highmem pages will be managed by the buddy system. Here highmem
-	 * zone also includes highmem movable zone.
-	 */
+	if (reset_managed_pages_done)
+		return;
 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
-		if (!is_highmem(z))
-			z->managed_pages = 0;
+		z->managed_pages = 0;
+}
+
+void __init reset_all_zones_managed_pages(void)
+{
+	struct pglist_data *pgdat;
+
+	for_each_online_pgdat(pgdat)
+		reset_node_managed_pages(pgdat);
+	reset_managed_pages_done = 1;
 }
 
 /**
@@ -160,10 +165,7 @@ static void reset_node_lowmem_managed_pages(pg_data_t *pgdat)
  */
 unsigned long __init free_all_bootmem(void)
 {
-	struct pglist_data *pgdat;
-
-	for_each_online_pgdat(pgdat)
-		reset_node_lowmem_managed_pages(pgdat);
+	reset_all_zones_managed_pages();
 
 	/*
 	 * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index f22542f6dc12..22438eba00b6 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -5232,6 +5232,7 @@ void free_highmem_page(struct page *page)
 {
 	__free_reserved_page(page);
 	totalram_pages++;
+	page_zone(page)->managed_pages++;
 	totalhigh_pages++;
 }
 #endif