v14: - Fix a bug about cpumask handling in patch 4 by using CONFIG_CPUMASK_OFFSTACK #ifdef block. - Add patch 12 to show descriptive name when reading cpuset.sched.partition (suggested by Tejun). - Change the function prototype of free_cpumasks() to match that of alloc_cpumasks() (suggested by Tom Hromatka). v13: - A major rewrite of the partition code so that there will be no auto-turning off anymore. Instead, the partition root can enter into an error state that can be restored back to a partition root later on. - Patches 1 and 9 are the same as previous version, the rests are either new or substantially revised. v12: - Take out the debugging patch to print partitions. - Add a patch to force turning off partition flag if newly modified CPU list doesn't meet the requirement of being a partition root. - Remove some unneeded checking code in update_reserved_cpumask(). v11: - Change the "domain_root" name to "partition" as suggested by Peter and update the documentation and code accordingly. - Remove the dying cgroup check in update_reserved_cpus() as the check may not be needed after all. - Document the effect of losing CPU affinity after offling all the cpus in a partition. - There is no other major code changes in this version. v11 patch: https://lkml.org/lkml/2018/6/24/30 v12 patch: https://lkml.org/lkml/2018/8/27/423 v13 patch: https://lkml.org/lkml/2018/10/12/861 The purpose of this patchset is to provide a basic set of cpuset control files for cgroup v2. This basic set includes the non-root "cpus", "mems" and "sched.partition". The "cpus.effective" and "mems.effective" will appear in all cpuset-enabled cgroups. The new control file that is unique to v2 is "sched.partition". It is a tristate flag file that designates if a cgroup is the root of a new scheduling domain or partition with its own set of unique list of CPUs from scheduling perspective disjointed from other partitions. An user can write only "1" or "0" into this file to turn on and off partition root. Depending on circumstances, a partition root may become erroneous and has a flag value of -1. However, if condition becomes favorable again, it can be changed back to a partition root automatically. The root cgroup is always a partition root. Multiple levels of partitions are supported with some limitations. So a container partition root can behave like a real root. When a partition root cgroup is removed, its list of exclusive CPUs will be returned back to the parent's cpus.effective automatically. A container root can be a partition root with sub-partitions created underneath it. One difference from the real root is that the "cpuset.sched.partition" flag isn't present in the real root, but is present in a container root. This is also true for other cpuset control files as well as those from the other controllers. This is a general issue that is not going to be addressed here in this patchset. This patchset does not exclude the possibility of adding more features in the future after careful consideration. Patch 1 enables cpuset in cgroup v2 with cpus, mems and their effective counterparts. Patch 2 defines new data structures to support partitioning. Patch 3 simplifies the allocation and freeing of cpumasks in the cpuset code and prepares for use by subsequent patches. Patch 4 adds a new "sched.partition" control file for setting up multiple scheduling domains or partitions. A partition root implies cpu_exclusive. Patch 5 makes new "sched.partition" file to have a new error value of -1 which indicates that the partition root enters into an erroneous state where some of the constraints of a partition root (like cpu_exclusive) will still hold but it is not a real partition root anymore. This allows the cpuset to change back to a partition root later on automatically if the conditions become favorable again. Patch 6 adds tracking of the number of cpusets that use the parent's effective_cpus in order to make sure that those cpusets will be properly updated if their parents effective cpus changes because of changes in sibling partitions. Patch 7 makes the hotplug code deal with partition root properly. Patch 8 updates the scheduling domain genaration code to work with the new partition feature. Patch 9 exposes cpus.effective and mems.effective to the root cgroup as enabling child partitions will take CPUs away from the root cgroup. So it will be nice to monitor what CPUs are left there. Patch 10 updates the cgroup v2 documentation file with information about the new "sched.partition" file. Patch 11 adds a new read-only "cpus.subpartitions" file that list the CPUs in the subparts_cpus mask in the cpuset data structure when the command line option "cgroup_debug" is specified. This is mostly used for debugging and verification purposes. Patch 12 changes the output of reading cpuset.sched.partition from integer to a descriptive text similar to what cgroup.type is doing. A test script with various cpuset configurations was run on both regular and debug kernels with this patchset applied to verify that the cpusets behaved appropriate without unexpected error. Waiman Long (12): cpuset: Enable cpuset controller in default hierarchy cpuset: Define data structures to support scheduling partition cpuset: Simply allocation and freeing of cpumasks cpuset: Add new v2 cpuset.sched.partition flag cpuset: Add an error state to cpuset.sched.partition cpuset: Track cpusets that use parent's effective_cpus cpuset: Make CPU hotplug work with partition cpuset: Make generate_sched_domains() work with partition cpuset: Expose cpus.effective and mems.effective on cgroup v2 root cpuset: Add documentation about the new "cpuset.sched.partition" flag cpuset: Expose cpuset.cpus.subpartitions with cgroup_debug cpuset: Show descriptive text when reading cpuset.sched.partition Documentation/admin-guide/cgroup-v2.rst | 175 ++++- include/linux/cgroup-defs.h | 1 + kernel/cgroup/cgroup-internal.h | 2 + kernel/cgroup/cgroup.c | 14 +- kernel/cgroup/cpuset.c | 909 ++++++++++++++++++++++-- kernel/cgroup/debug.c | 4 +- 6 files changed, 1030 insertions(+), 75 deletions(-) -- 2.18.0
