From b6c39a47e7436adf16f186d7b45758855f45f6c6 Mon Sep 17 00:00:00 2001 From: Kefeng Wang Date: Thu, 23 May 2024 09:19:10 +0800 Subject: [PATCH] mm: add thp anon pmd size mapping align control hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9R7N8 CVE: NA ------------------------------------------------- After committing df0ad0579ac9 ('mm: align larger anonymous mappings on THP boundaries'), we found a number of issues, eg, a significant increase in memory access latency on arm64, which was addressed by commit bb713c2a4a43 (arm64: mmap: disable align larger anonymous mappings on THP boundaries), but subsequently found on x86 that the patch also resulted in a significant degradation of the SPECCPU test. Commit 449846913eb6('mm: huge_memory: add thp mapping align control') has implemented the sysfs interface control of 64 KB aligned mapping in arm64, based on this for thp_align_mapping to add a new bit to control pmd size page alignment, where the 64K alignment mapping enable for arm64 will depend on the new bits being turned on. it is disabled by default, and could be enabled by /sys/kernel/mm/transparent_hugepage/thp_mapping_align. Signed-off-by: Kefeng Wang Signed-off-by: Ze Zuo --- Documentation/admin-guide/mm/transhuge.rst | 19 +++++++++++-------- include/linux/huge_mm.h | 5 +++++ mm/huge_memory.c | 14 ++++++++++++-- mm/mmap.c | 2 +- 4 files changed, 29 insertions(+), 11 deletions(-) diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst index 54046f487f15..aeff60881fd0 100644 --- a/Documentation/admin-guide/mm/transhuge.rst +++ b/Documentation/admin-guide/mm/transhuge.rst @@ -211,14 +211,17 @@ possible to enable/disable it by configurate the corresponding bit:: echo 0x2 >/sys/kernel/mm/transparent_hugepage/thp_exec_enabled echo 0x3 >/sys/kernel/mm/transparent_hugepage/thp_exec_enabled -The kernel could try to enable other larger size mappings align other -than THP size, eg, 64K on arm64, BIT0 for file mapping, BIT1 for anon -mapping, it is disabled by default, and could enable this feature by -writing the corresponding bit to 1:: - - echo 0x1 >/sys/kernel/mm/transparent_hugepage/thp_mapping_align - echo 0x2 >/sys/kernel/mm/transparent_hugepage/thp_mapping_align - echo 0x3 >/sys/kernel/mm/transparent_hugepage/thp_mapping_align +The kernel could try to enable mappings for different sizes, eg, 64K for +arm64, BIT0 for file mapping, BIT1 for anonymous mapping, and THP size +page, BIT3 for anonymous mapping, where 64K for arm64 is dependent on +BIT3 being turned on, the above feature are disabled by default, and +could enable the above feature by writing the corresponding bit to 1:: + + echo 0x4 >/sys/kernel/mm/transparent_hugepage/thp_mapping_align + echo 0x5 >/sys/kernel/mm/transparent_hugepage/thp_mapping_align + echo 0x6 >/sys/kernel/mm/transparent_hugepage/thp_mapping_align + echo 0x7 >/sys/kernel/mm/transparent_hugepage/thp_mapping_align + The kernel could enable high-orders(greated than PAGE_ALLOC_COSTLY_ORDER, only support order 4 for now) be stored on PCP lists(except PMD order), which could diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index 3e5f7064e2de..6dcdcc47dd5e 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -54,6 +54,7 @@ enum transparent_hugepage_flag { TRANSPARENT_HUGEPAGE_FILE_EXEC_MTHP_FLAG, TRANSPARENT_HUGEPAGE_FILE_MAPPING_ALIGN_FLAG, TRANSPARENT_HUGEPAGE_ANON_MAPPING_ALIGN_FLAG, + TRANSPARENT_HUGEPAGE_ANON_MAPPING_PMD_ALIGN_FLAG, }; struct kobject; @@ -288,6 +289,10 @@ static inline void count_mthp_stat(int order, enum mthp_stat_item item) (transparent_hugepage_flags & \ (1<