mm/page_alloc: fix min_free_kbytes calculation regarding ZONE_MOVABLE
authorliuq <liuq131@chinatelecom.cn>
Sun, 25 Jun 2023 03:16:56 +0000 (11:16 +0800)
committerAndrew Morton <akpm@linux-foundation.org>
Fri, 18 Aug 2023 17:12:07 +0000 (10:12 -0700)
commit416ef04fe00c5f2f6fb8e13d8dbe1b5a0a274f83
tree1f03ef95311aa32095524943caa3d2251541107f
parenta524fcfe190da16bbf1311b6636f51d81f35d59a
mm/page_alloc: fix min_free_kbytes calculation regarding ZONE_MOVABLE

The current calculation of min_free_kbytes only uses ZONE_DMA and
ZONE_NORMAL pages,but the ZONE_MOVABLE zone->_watermark[WMARK_MIN] will
also divide part of min_free_kbytes.This will cause the min watermark of
ZONE_NORMAL to be too small in the presence of ZONE_MOVEABLE.

__GFP_HIGH and PF_MEMALLOC allocations usually don't need movable zone
pages, so just like ZONE_HIGHMEM, cap pages_min to a small value in
__setup_per_zone_wmarks().

On my testing machine with 16GB of memory (transparent hugepage is turned
off by default, and movablecore=12G is configured) The following is a
comparative test data of watermark_min

no patch add patch
ZONE_DMA 1 8
ZONE_DMA32 151 709
ZONE_NORMAL 233 1113
ZONE_MOVABLE 1434 128
min_free_kbytes 7288 7326

Link: https://lkml.kernel.org/r/20230625031656.23941-1-liuq131@chinatelecom.cn
Signed-off-by: liuq <liuq131@chinatelecom.cn>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/page_alloc.c