Zswap backing memory is currently not tracked (and limited) on a per-cgroup basis. As a result, workloads can escape their memory containment and cause resource priority inversions on a shared host. E.g. a lo-pri group fills the global zswap pool and forces a hi-pri group out to disk. Also, zswap doesn't benefit all workloads equally. Some even suffer when memory contents compress poorly, and are better off going to disk swap directly. On a host with mixed workloads, it's currently not possible to enable zswap for one workload but not for the other. This series implements missing cgroup awareness and control for zswap to address both issues. More details on interface and implementation in patch 5. Patches 1-3 clean up related and adjacent options in Kconfig. Not dependencies, just things I noticed during development. Based on v5.18-rc4-mmots-2022-04-26-19-34-5-g5e1fdb02de7a. Documentation/admin-guide/cgroup-v2.rst | 21 ++ drivers/block/zram/Kconfig | 3 +- fs/proc/meminfo.c | 7 + include/linux/memcontrol.h | 54 +++ include/linux/swap.h | 5 + include/linux/vm_event_item.h | 4 + init/Kconfig | 123 ------- mm/Kconfig | 523 +++++++++++++++++++----------- mm/memcontrol.c | 196 ++++++++++- mm/vmstat.c | 4 + mm/zswap.c | 50 ++- 11 files changed, 648 insertions(+), 342 deletions(-)
