Hi, the ongoing versioning of the cgroup user interface gives us a chance to clean up the memcg control interface and fix a lot of inconsistencies and ugliness that crept in over time. This series adds a minimal set of control files to the new memcg interface to get basic memcg functionality going in unified hierarchy: - memory.current: a read-only file that shows current memory usage. - memory.high: a file that allows setting a high limit on the memory usage. This is an elastic limit, which is enforced via direct reclaim, so allocators are throttled once it's reached, but it can be exceeded and does not trigger OOM kills. This should be a much more suitable default upper boundary for the majority of use cases that are better off with some elasticity than with sudden OOM kills. - memory.max: a file that allows setting a maximum limit on memory usage which is ultimately enforced by OOM killing tasks in the group. This is for setups that want strict isolation at the cost of task death above a certain point. However, even those can still combine the max limit with the high limit to approach OOM situations gracefully and with time to intervene. - memory.vmstat: vmstat-style per-memcg statistics. Very minimal for now (lru stats, allocations and frees, faults), but fixing fundamental issues of the old memory.stat file, including gross misnomers like pgpgin/pgpgout for pages charged/uncharged etc. Documentation/cgroups/unified-hierarchy.txt | 18 +++ include/linux/res_counter.h | 29 +++++ include/linux/swap.h | 3 +- kernel/res_counter.c | 3 + mm/memcontrol.c | 177 +++++++++++++++++++++++++--- mm/vmscan.c | 3 +- 6 files changed, 216 insertions(+), 17 deletions(-) -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>