[HMM 02/15] mm/put_page: move ZONE_DEVICE page reference decrement v2

[JÃrÃme Glisse]

Move page reference decrement of ZONE_DEVICE from put_page()
to put_zone_device_page() this does not affect non ZONE_DEVICE
page.

Doing this allow to catch when a ZONE_DEVICE page refcount reach
1 which means the device is no longer reference by any one (unlike
page from other zone, ZONE_DEVICE page refcount never reach 0).

This patch is just a preparatory patch for HMM.

Changes since v1:
  - commit message

Signed-off-by: JÃrÃme Glisse <jglisse@xxxxxxxxxx>
Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
Cc: Ross Zwisler <ross.zwisler@xxxxxxxxxxxxxxx>
---
 include/linux/mm.h | 14 +++++++++++---
 kernel/memremap.c  |  6 ++++++
 2 files changed, 17 insertions(+), 3 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index c82e8db..022423c 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -821,11 +821,19 @@ static inline void put_page(struct page *page)
 {
 	page = compound_head(page);
 
+	/*
+	 * ZONE_DEVICE pages should never have their refcount reach 0 (this
+	 * would be a bug), so call page_ref_dec() in put_zone_device_page()
+	 * to decrement page refcount and skip __put_page() here, as this
+	 * would worsen things if a ZONE_DEVICE had a refcount bug.
+	 */
+	if (unlikely(is_zone_device_page(page))) {
+		put_zone_device_page(page);
+		return;
+	}
+
 	if (put_page_testzero(page))
 		__put_page(page);
-
-	if (unlikely(is_zone_device_page(page)))
-		put_zone_device_page(page);
 }
 
 #if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
diff --git a/kernel/memremap.c b/kernel/memremap.c
index ea714ee..97ef676 100644
--- a/kernel/memremap.c
+++ b/kernel/memremap.c
@@ -190,6 +190,12 @@ EXPORT_SYMBOL(get_zone_device_page);
 
 void put_zone_device_page(struct page *page)
 {
+	/*
+	 * ZONE_DEVICE page refcount should never reach 0 and never be freed
+	 * to kernel memory allocator.
+	 */
+	page_ref_dec(page);
+
 	put_dev_pagemap(page->pgmap);
 }
 EXPORT_SYMBOL(put_zone_device_page);
-- 
2.9.3