unrelated; please remove

If you do want to distinguish between the two cases for EPERM, maybe it makes sense to check if we have cpus set first.

Also, maybe move this code into a separate function (setupScheduler), as we have the same 8 lines of code in both setns_init_linux.go and standard_init_linux.go

maybe it makes sense to check if we have cpus set first.

I also considered it when writing the code, but there is another corner case is that if the cpus value includes all cpus in the machine, it will be no error. So I don't want to check cpus value.

nit: errors.New.

I think it's better to use s.Nice directly (drop the intermediate variable) here.

I think the original author used an intermediate variable because repeating config.Scheduler.Nice three times was too much.

But, just in case an intermediate variable is preferred, it is better to limit its scope:

if n := s.Nice; n < -20 || n > 19 {
	return fmt.Errorf("invalid scheduler.nice: %d", n)
}

Maybe don't make this a special case.

I.e. something like

if err := unix.SchedSetAttr(0, attr, 0); err != nil {
    if errors.Is(err, unix.EPERM) && config.Cgroups.CpusetCpus != "" {
        return errors.New("process scheduler can not be used together with AllowedCPUs")
    }
    return fmt.Errorf("error setting scheduler: %w", err)
}