[PATCH v3 0/2] cgroup/cpuset: fix DL attach accounting

[Guopeng Zhang]

Hi,

This v3 series contains two cpuset fixes for SCHED_DEADLINE attach
accounting.

Patch 1 fixes an internal cpuset_can_attach() failure path where
temporary DL migration state can be left behind if a later per-task
check fails before cpuset marks attach_in_progress.

Patch 2 keeps cpuset DL bandwidth reservation aligned with the condition
used by set_cpus_allowed_dl() for source-side bandwidth removal. It keeps
counting all migrating DL tasks for cpuset task accounting, but reserves
destination DL bandwidth only for tasks that actually need a root-domain
bandwidth move.

Guopeng Zhang (2):
  cgroup/cpuset: reset DL migration state on can_attach() failure
  cgroup/cpuset: reserve DL bandwidth only for root-domain moves

 include/linux/sched/deadline.h  |  9 ++++++++
 kernel/cgroup/cpuset-internal.h |  1 +
 kernel/cgroup/cpuset.c          | 39 ++++++++++++++++++---------------
 kernel/sched/deadline.c         | 13 ++++++++---
 4 files changed, 41 insertions(+), 21 deletions(-)

---
Changes in v3:
- Patch 1: use common ret != 0 cleanup in cpuset_can_attach(), as
  suggested by Waiman Long and Chen Ridong.
- Patch 2: drop task_cpu_possible_mask() / attach-target-mask handling
  as suggested by Waiman Long.
- Patch 2: keep the change limited to reserving DL bandwidth only for
  tasks that need a root-domain bandwidth move.
- Leave the broader can_attach()/attach() transaction model unchanged.

Changes in v2:
- Split the original change into two patches.
- Add a separate fix for resetting pending DL migration state on
  cpuset_can_attach() failure.
- Clarify that nr_migrate_dl_tasks counts all migrating DL tasks for
  cpuset task accounting, while sum_migrate_dl_bw only tracks bandwidth
  needing destination root-domain reservation.

v2:
  https://lore.kernel.org/all/20260507103310.35849-1-zhangguopeng@xxxxxxxxxx/

v1:
  https://lore.kernel.org/all/20260421083449.95750-1-zhangguopeng@xxxxxxxxxx

-- 
2.43.0