From: Cho KyongHo <pullip.cho@samsung.com>
Date: Wed, 21 Dec 2016 02:17:20 +0000 (+0900)
Subject: mm: do not alloc cma page for shmem
X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=7f6c82e570cf79f779e9b9ddb9e8f2d62bcec18e;p=GitHub%2FLineageOS%2Fandroid_kernel_motorola_exynos9610.git

mm: do not alloc cma page for shmem

Some page caches of shmem tends to be pinned for a long time.
Therefore shmem should not allocate page caches from CMA because cma
pages should be always ready to be reclaimed promptly.

Change-Id: I024dad765feec07b0028851b0b98b7db7f3c85a5
Signed-off-by: Cho KyongHo <pullip.cho@samsung.com>
---

diff --git a/mm/shmem.c b/mm/shmem.c
index 6a5973d727f8..024e5c217f96 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -2170,6 +2170,13 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 		inode->i_generation = get_seconds();
+		/*
+		 * removal of __GFP_HIGHMEM breaks GFP_HIGHUSER_MOVABLE. It is
+		 * required not to allocate CMA pages to the page caches of
+		 * shmem.
+		 */
+		mapping_set_gfp_mask(inode->i_mapping,
+			mapping_gfp_mask(inode->i_mapping) & ~__GFP_HIGHMEM);
 		info = SHMEM_I(inode);
 		memset(info, 0, (char *)inode - (char *)info);
 		spin_lock_init(&info->lock);