[PATCH 00/22] mm: introduce N_MEMORY_COMMON
From: Gregory Price (Meta)
Date: Sat Nov 14 2026 - 02:10:19 EST
This is phase 1 of the Private NUMA node series.
I will send v6 "Private Memory NUMA nodes" (phases 2 and 3) separately
next week for further discussion - but I pulled this out ahead to make
the core filtering mechanism easier to review.
The series is now broken up like so:
Phase 1: Introduce N_MEMORY_COMMON to define "general purpose" (common)
memory as a concept separate from a node having memory as a
topological feature (N_MEMORY).
This phase is a no-op for existing systems, but creates the
baseline for ensuring private node memory is default opted-out
of being touched by MM (much like ZONE_DEVICE).
Phase 2: Introduce N_MEMORY_PRIVATE to define memory not part of the
general purpose pool. With this we can introduce a minimum set
of KUnit tests for the page allocator to demonstrate the
isolation mechanism.
At this point it will be possible for a driver to re-use the
buddy allocator to make uniform order-N allocations.
Phase 3: Introduce feature bits (COMPACTION, RECLAIM, MEMPOLICY) which
is the minimum viable set to allow userland to manage a private
memory node via mbind() and migrate_pages().
This enables management of HBM and Coherent GPU memory via
userland tools (numactl, cpuset) while protecting them from
incidental allocation (no fallbacks, only explicit placement).
The memory could even be hotplugged in ZONE_NORMAL and pinned.
Actual cover letter for this portion
====================================
N_MEMORY has two meanings today:
1) this node has memory
2) ordinary allocations and MM services can use this node's memory
The two are the same today. They stop being the same once a NUMA node
with memory isn't part of the general allocation pool.
This series adds N_MEMORY_COMMON for the second meaning and converts
the relevant N_MEMORY users accordingly. N_MEMORY continues reporting
memory topology - and some users (cpusets) still utilize it.
At the end of this series, N_MEMORY == N_MEMORY_COMMON, so the series
changes no behavior for existing systems. A later series will add memory
outside N_MEMORY_COMMON without having to audit every N_MEMORY user again.
Rather than add new hooks for page-table and pfn walkers, we consolidate
existing ZONE_DEVICE checks (page/folio_is_zone_device()) into a general
`page/folio_is_common_memory()` check.
The series is laid out as:
1-3 helper functions to check node states and a folio's
membership to a common memory node. N_MEMORY_COMMON itself, and
the `has_common_memory` sysfs ABI.
4 zonelists: the fallback lists hold common nodes, and a node's
__GFP_THISNODE list is built only for a common node
5-21 N_MEMORY->N_MEMORY_COMMON and zone_device -> common_memory
conversions. One component per patch.
22 removal of first_memory_node/next_memory_node, which no longer
have users after this series.
Each conversion restricts or adjusts:
- an MM service (reclaim, compaction, KSM, collapse, etc)
- an allocation target to common memory
- accounting meant for ordinary allocations from common memory only
has_common_memory lists the same nodes as has_memory until memory
outside N_MEMORY_COMMON exists. It is added now so that userspace, such
as libnuma, can start choosing allocation targets from it, and so that
the meaning is fixed before anything makes the two lists differ.
first/next_memory_node() is replaced by first/next_node_state().
Gregory Price (Meta) (22):
mm: add node-state-filtered zone and node iterators
mm: publish and clear node memory states via helpers
mm: introduce the common-memory node state
mm/page_alloc: build zonelists from common memory
mm/mempolicy: keep interleave and tmpfs mount policies on common memory
mm: base hash distribution on common memory nodes
mm/memory_hotplug: migrate offlined pages to common memory
mm/memory-tiers: restrict demotion topology to common memory
mm/ksm: restrict merging to common memory
mm/khugepaged: restrict collapse to common memory
mm/mempolicy: don't NUMA-hint folios outside common memory
mm/damon: restrict DAMON-native folio operations to common memory
mm/hugetlb: restrict pool growth to common memory
mm/page_alloc: count common memory nodes for kswapd recovery
mm/oom_kill: compare OOM constraints against common memory
tty/sysrq: use common memory for the manual OOM zonelist
mm/vmscan: restrict reclaim to common memory
mm/compaction: restrict compaction to common memory
mm/migrate: restrict move_pages() destinations to common memory
PM: hibernate: size the image from common memory
mm/khugepaged: base the recommended min_free_kbytes on common memory
nodemask: remove the now-unused first_memory_node/next_memory_node
Documentation/ABI/stable/sysfs-devices-node | 16 +++++++
Documentation/mm/physical_memory.rst | 8 +++-
drivers/base/node.c | 2 +
drivers/tty/sysrq.c | 3 +-
include/linux/mmzone.h | 46 +++++++++++++++++++++
include/linux/nodemask.h | 38 +++++++++++++----
kernel/power/snapshot.c | 7 +++-
mm/compaction.c | 14 +++----
mm/damon/ops-common.c | 7 ++--
mm/damon/vaddr.c | 2 +-
mm/hugetlb.c | 44 +++++++++-----------
mm/internal.h | 18 ++++++++
mm/khugepaged.c | 12 ++++--
mm/ksm.c | 8 ++--
mm/memory-tiers.c | 23 ++++++-----
mm/memory_hotplug.c | 25 ++++++-----
mm/mempolicy.c | 30 +++++++-------
mm/migrate.c | 2 +-
mm/mm_init.c | 16 ++++---
mm/oom_kill.c | 7 ++--
mm/page_alloc.c | 43 +++++++++++++------
mm/vmscan.c | 26 ++++++------
22 files changed, 266 insertions(+), 131 deletions(-)
--
2.53.1-Meta