The CPU bandwidth quota and burst controls currently make the order of
userspace writes significant. A burst configured while quota is unlimited can
prevent a later finite quota from being installed, and increasing quota and
burst in the burst-first order can fail with EINVAL.
This series keeps the configured burst value independent of the current quota
and applies the quota-relative clamp when CFS refills runtime. It also adds a
selftest for both write orders and documents the resulting behavior for cgroup
v1 and cgroup v2.
Changes in v2:
- apply the clamp at refill time instead of changing the configured burst;
- retain the value reported by cpu.max.burst and cpu.cfs_burst_us;
- add write-order selftests and update the CPU bandwidth documentation.
Testing:
make -C tools/testing/selftests TARGETS=cgroup
Zhe Liu (3):
sched/fair: Remove the write-order dependency between cpu.max and
cpu.max.burst
selftests: cgroup: Test CPU quota and burst write order
Documentation: describe CPU quota and burst ordering
Documentation/admin-guide/cgroup-v2.rst | 5 ++-
Documentation/scheduler/sched-bwc.rst | 21 ++++++-----
kernel/sched/core.c | 3 +-
kernel/sched/fair.c | 3 +-
tools/testing/selftests/cgroup/test_cpu.c | 40 +++++++++++++++++++++++
5 files changed, 59 insertions(+), 13 deletions(-)
--
2.25.1
