If a task it attached to a cgroup then its CPU-mask will be set to the mask of the cgroup. Upon enabling the cpuset controller it will be set to all online CPUs. All running user tasks, that changed their CPU-mask, will have their mask altered without knowing it. Tasks with unchanged CPU-mask are set to all online CPUs and this change is a nop. If the task is already using a subset of the allowed (new) CPUs, skip changing the mask. This will preserve the CPU-mask as long as it contains only allowed CPUs from the new CPU-mask. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> --- kernel/cgroup/cpuset.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index b474289c15b82..8d5126684f9e6 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -2527,7 +2527,8 @@ static void cpuset_attach(struct cgroup_taskset *tset) * can_attach beforehand should guarantee that this doesn't * fail. TODO: have a better way to handle failure here */ - WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); + if (!cpumask_subset(&task->cpus_mask, cpus_attach)) + WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); cpuset_update_task_spread_flag(cs, task); -- 2.37.2
