tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1What: /sys/kernel/mm/numa/
2Date: June 2021
3Contact: Linux memory management mailing list <linux-mm@kvack.org>
4Description: Interface for NUMA
5
6What: /sys/kernel/mm/numa/demotion_enabled
7Date: June 2021
8Contact: Linux memory management mailing list <linux-mm@kvack.org>
9Description: Enable/disable demoting pages during reclaim
10
11 Page migration during reclaim is intended for systems
12 with tiered memory configurations. These systems have
13 multiple types of memory with varied performance
14 characteristics instead of plain NUMA systems where
15 the same kind of memory is found at varied distances.
16 Allowing page migration during reclaim enables these
17 systems to migrate pages from fast tiers to slow tiers
18 when the fast tier is under pressure. This migration
19 is performed before swap if an eligible numa node is
20 present in cpuset.mems for the cgroup (or if cpuset v1
21 is being used). If cpusets.mems changes at runtime, it
22 may move data to a NUMA node that does not fall into the
23 cpuset of the new cpusets.mems, which might be construed
24 to violate the guarantees of cpusets. Shared memory,
25 such as libraries, owned by another cgroup may still be
26 demoted and result in memory use on a node not present
27 in cpusets.mem. This should not be enabled on systems
28 which need strict cpuset location guarantees.