This patch series is RESEND of the previous patches on psi subsystem. A few weeks passed since the last review, so I put them together and resend for more convenient review and merge.
Patch 1 add PSI_CPU_FULL state means all non-idle tasks in a cgroup are delayed on the CPU resource which used by others outside of the cgroup or throttled by the cgroup cpu.max configuration. Patch 2 use ONCPU state and the current in_memstall flag to detect reclaim, remove the hook in timer tick to make code more concise and maintainable. And patch 3 adds unlikely() annotations to move the pressure state branches out of line to eliminate undesirable jumps during wakeup and sleeps. Patch 4 optimize the voluntary sleep switch by remove one call of psi_group_change() for every common cgroup ancestor of the two tasks. Chengming Zhou (3): psi: Add PSI_CPU_FULL state psi: Use ONCPU state tracking machinery to detect reclaim psi: Optimize task switch inside shared cgroups Johannes Weiner (1): psi: pressure states are unlikely include/linux/psi.h | 1 - include/linux/psi_types.h | 3 +- kernel/sched/core.c | 1 - kernel/sched/psi.c | 122 ++++++++++++++++++++++++---------------------- kernel/sched/stats.h | 37 +++++--------- 5 files changed, 78 insertions(+), 86 deletions(-) -- 2.11.0
