lkml news

LKML News v5.16-rc2

[PATCH -V10 0/6] NUMA balancing: optimize memory placement for memory tiering system (Huang Ying) https://lkml.kernel.org/r/20211116013522.140575-1-ying.huang@intel.com This patchset optimizes the pages placement on tiered memory systems using DRAM and PMEM. [RFC 0/3] page table check (Pasha Tatashin) https://lkml.kernel.org/r/20211116220038.116484-1-pasha.tatashin@soleen.com Pasha found broken page ref issue in 4.14 kernel. The page was leaked into another process, so they were able to find the issue by examining the page tables. This patchset tries to prevent this kind of memory corruption issues by checking the page tables in time of inserting entries into user page tables.

LKML News v5.16-rc1

[PATCH v4 0/3] mm: add new syscall set_mempolicy_home_node (Aneesh Kumar K.V) https://lkml.kernel.org/r/20211101050206.549050-1-aneesh.kumar@linux.ibm.com This patchset introduces a new syscall that can be used to set a home node for the MPOL_BIND and MPOL_PREFERRED_MANY. [RFC PATCH 0/3] Improve the migration stats (Baolin Wang) https://lkml.kernel.org/r/cover.1635936218.git.baolin.wang@linux.alibaba.com This RFC patchset fixes the migration stats that broken here and there. incoming (Andrew Morton) https://lkml.kernel.org/r/20211105133408.cccbb98b71a77d5e8430aba1@linux-foundation.org The first pull request for v5.16 from Andrew Morton. Patches for DAMON subsystem including DAMOS, physical address space support, and DAMON-based proactive reclamation are also included here.

LKML News v5.15

Dynamically allocated memory descriptors (Matthew Wilcox) https://lkml.kernel.org/r/YXcLqcFhDq3uUwIj@casper.infradead.org In this mail, Matthew explains his idea on making struct page only two words and how it could work with buddy allocator freelists. [RFC] mm: Disable NUMA_BALANCING_DEFAULT_ENABLED and TRANSPARENT_HUGEPAGE on PREEMPT_RT (Sebastian Andrzej Siewior) https://lkml.kernel.org/r/20211026165100.ahz5bkx44lrrw5pt@linutronix.de Mel told RT patchset could incur spikes on NUMA machines when NUMA_BALANCING and THP are turned on. Sebastian request comments on disabling those on RT enabled kernel by default, but still trying to find a good test for that.

LKML News v5.15-rc7

incoming (Andrew Morton) https://lkml.kernel.org/r/20211018151438.f2246e2656c041b6753a8bdd@linux-foundation.org 19 mm subsystem fix patches have sent to Linus Torvalds. [PATCH v11 1/3] mm: rearrange madvise code to allow for reuse (Suren Baghdasaryan) https://lkml.kernel.org/r/20211019215511.3771969-1-surenb@google.com This commit refactor some madvise code for vma updates so that those can be reused by prctl. mmotm 2021-10-20-20-40 uploaded (Andrew Morton) https://lkml.kernel.org/r/20211021034104.F6VxKBQcG%akpm@linux-foundation.org Andrew updated mmotm. It’s based on 5.15-rc6, adds 74 new patches modifies 29 existing patches, and drops 22 patches. In total, 366 patches in the queue.

LKML News v5.15-rc6

mmotm 2021-10-10-18-19 uploaded (Andrew Morton) https://lkml.kernel.org/r/20211011012001.mSbkDke0y%akpm@linux-foundation.org Update of mmotm. 33 new patches added and 7 old patches dropped. 285 patches in total. [PATCH v2 0/2] Introduce simple pidfd to task helper (Christian Brauner) https://lkml.kernel.org/r/20211011133245.1703103-1-brauner@kernel.org This patchset adds a simple helper to make code clean without duplications. [PATCH v0.7 0/5] sched,mm,x86/uaccess: implement User Managed Concurrency Groups (Peter Oskolkov) https://lkml.kernel.org/r/20211012232522.714898-1-posk@google.com This patchset implements the kernel parts for user-space m:n threading. If this is finalized and merged, user-space applications will be able to do the scheduling of their threads on their own.

LKML News v5.15-rc5

[PATCH 0/2] Introduce simple pidfd to task helper (Christian Brauner) https://lkml.kernel.org/r/20211004125050.1153693-1-christian.brauner@ubuntu.com This patchset adds a pidfd handling wrapper for removing duplications. [PATCH 00/62] Separate struct slab from struct page (Matthew Wilcox) https://lkml.kernel.org/r/20211004134650.4031813-1-willy@infradead.org This huge patchset separates struct slab from struct page to make struct page simple and therefore easier to handle. [PATCH v3 00/66] Introducing the Maple Tree (Liam Howlett) https://lkml.kernel.org/r/20211005012959.1110504-1-Liam.Howlett@oracle.com This patchset introduces Maple tree, which is intended to replace the rb tree for vmas, to improve the scalability.

LKML News v5.15-rc4

[RFC PATCH 0/8] mm/madvise: support process_madvise(MADV_DONTNEED) (Nadav Amit) https://lkml.kernel.org/r/20210926161259.238054-1-namit@vmware.com This patchset makes process_madvise() to support MADV_DONTNEED hint. mmotm 2021-09-28-19-05 uploaded (Andrew Morton) https://lkml.kernel.org/r/20210929020604.DyNimbxao%akpm@linux-foundation.org The mm-of-the-moment snapshot has been updated. 13 drop, 45 add, 37 modified. 184 patches in the broken/ in total. [PATCH 1/2] memcg: flush stats only if updated (Shakeel Butt) https://lkml.kernel.org/r/20210930044711.2892660-1-shakeelb@google.com This patchset makes memcg stats updated only when needed, as it is currently flushed on every refault and reclaim iteration.

LKML News v5.15-rc3

[RFC PATCH] Introducing lockless cache built on top of slab allocator (Hyeonggon Yoo) https://lkml.kernel.org/r/20210919164239.49905-1-42.hyeyoo@gmail.com The block layer now has a lockless cache, which increased its IOPS. This RFC generalizes it so that it can be used in some other area, such as NAPI. [PATCH v5 0/3] virtio-mem: disallow mapping virtio-mem memory via /dev/mem (David Hildenbrand) https://lkml.kernel.org/r/20210920142856.17758-1-david@redhat.com This patchset makes a basic infrastructure to exclude some physical memory regions that marked as IORESOURCE_SYSTEM_RAM from /dev/mem access, even though they are not marked as IORESOURCE_BUSY.

LKML News v5.15-rc2

[PATCH v1] vmpressure: wake up work only when there is registration event (wangyong) https://lkml.kernel.org/r/1631548441-2784-1-git-send-email-wang.yong12@zte.com.cn Currently, vmpressure works are awaken even though there are no vmpressure callbacks. This commit avoids such unnecessary wakeup by counting the number of registered events and skip waking up works if the number is zero. [PATCH v3 00/76] Optimize list lru memory consumption (Muchun Song) https://lkml.kernel.org/r/20210914072938.6440-1-songmuchun@bytedance.com Muchun found kmalloc-32 for list_lru_one allocation consuming 6 GiB of memory on their server.

LKML News v5.15-rc1

[RFC PATCH] mm/oom: detect and kill task which has allocation forbidden by cpuset limit (Feng Tang) https://lkml.kernel.org/r/1630399085-70431-1-git-send-email-feng.tang@intel.com Docker needs to allocate memory in non-movable zones for creation of an OS (e.g., GFP_HIGHUSER). However, if it is binded to a cpuset which has only movable zone, the allocation fails and OOM storm attacks other innocent processes. This patch detects the case and kill only the allocation failing process. Folios: Can we resolve this please?