From: zhouxianrong <zhouxianrong@xxxxxxxxxx> the reason for why to do this is based on below factors. 1. larg file read/write operations with order 0 can fragmentize memory rapidly. 2. when a special filesystem does not supply migratepage callback, kernel would fallback to default function fallback_migrate_page. but fallback_migrate_page could not migrate diry page nicely; specially kcompactd with MIGRATE_SYNC_LIGHT could not migrate diry pages due to this until clear_page_dirty_for_io in some procedure. i think it is not suitable here in this scenario. for dirty pages we should migrate it rather than skip or writeout it in kcomapctd with MIGRATE_SYNC_LIGHT. i think this problem is for all filesystem without migratepage not only for block device fs. so for compaction under large file writing supply migratepage for def_blk_aops. Signed-off-by: zhouxianrong <zhouxianrong@xxxxxxxxxx> --- fs/block_dev.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/fs/block_dev.c b/fs/block_dev.c index 1c62845..9343b60 100644 --- a/fs/block_dev.c +++ b/fs/block_dev.c @@ -2062,6 +2062,9 @@ static int blkdev_writepages(struct address_space *mapping, .releasepage = blkdev_releasepage, .direct_IO = blkdev_direct_IO, .is_dirty_writeback = buffer_check_dirty_writeback, +#ifdef CONFIG_MIGRATION + .migratepage = buffer_migrate_page, +#endif }; #define BLKDEV_FALLOC_FL_SUPPORTED \ -- 1.7.9.5
