Hello, Tianyi. On Sun, Sep 06, 2026 at 10:40:29PM +0800, Tianyi Chen wrote: > + if (cpu >= 0 && > + (!bpf_cpumask_test_cpu(cpu, cast_mask(mask)) || > + !bpf_cpumask_test_cpu(cpu, p->cpus_ptr))) > + cpu = -ERANGE;
The syscall runs with migration disabled. With CONFIG_PREEMPT_RCU, preemption between selection and this check can make migrate_disable_switch() narrow p->cpus_ptr to the running CPU, rejecting a valid selection on another CPU. Please check the requested affinity instead. > + SCX_FAIL_IF(sched_getaffinity(0, sizeof(original), &original), > + "Failed to get affinity (%d)", errno); Please use dynamically allocated masks. The fixed cpu_set_t makes sched_getaffinity() fail with EINVAL when nr_cpu_ids exceeds 1024, even if the task only uses low-numbered CPUs. Thanks. -- tejun
