162306a36Sopenharmony_ciWhat: /sys/kernel/mm/numa/ 262306a36Sopenharmony_ciDate: June 2021 362306a36Sopenharmony_ciContact: Linux memory management mailing list <linux-mm@kvack.org> 462306a36Sopenharmony_ciDescription: Interface for NUMA 562306a36Sopenharmony_ci 662306a36Sopenharmony_ciWhat: /sys/kernel/mm/numa/demotion_enabled 762306a36Sopenharmony_ciDate: June 2021 862306a36Sopenharmony_ciContact: Linux memory management mailing list <linux-mm@kvack.org> 962306a36Sopenharmony_ciDescription: Enable/disable demoting pages during reclaim 1062306a36Sopenharmony_ci 1162306a36Sopenharmony_ci Page migration during reclaim is intended for systems 1262306a36Sopenharmony_ci with tiered memory configurations. These systems have 1362306a36Sopenharmony_ci multiple types of memory with varied performance 1462306a36Sopenharmony_ci characteristics instead of plain NUMA systems where 1562306a36Sopenharmony_ci the same kind of memory is found at varied distances. 1662306a36Sopenharmony_ci Allowing page migration during reclaim enables these 1762306a36Sopenharmony_ci systems to migrate pages from fast tiers to slow tiers 1862306a36Sopenharmony_ci when the fast tier is under pressure. This migration 1962306a36Sopenharmony_ci is performed before swap. It may move data to a NUMA 2062306a36Sopenharmony_ci node that does not fall into the cpuset of the 2162306a36Sopenharmony_ci allocating process which might be construed to violate 2262306a36Sopenharmony_ci the guarantees of cpusets. This should not be enabled 2362306a36Sopenharmony_ci on systems which need strict cpuset location 2462306a36Sopenharmony_ci guarantees. 25