On Tue, Apr 05, 2016 at 07:10:04AM +0200, Juergen Gross wrote:

+int smp_call_on_cpu(unsigned int cpu, bool pin, int (*func)(void *), void *par)

Why .pin and not .phys? .pin does not (to me) reflect the
hypervisor/physical-cpu thing.

Also, as per smp_call_function_single() would it not be more consistent
to make this the last argument?

+{
+	struct smp_call_on_cpu_struct sscs = {
+		.work = __WORK_INITIALIZER(sscs.work, smp_call_on_cpu_callback),
+		.done = COMPLETION_INITIALIZER_ONSTACK(sscs.done),
+		.func = func,
+		.data = par,
+		.cpu  = pin ? cpu : -1,
+	};
+
+	if (cpu >= nr_cpu_ids)

You might want to also include cpu_online().

	if (cpu >= nr_cpu_ids || !cpu_online(cpu))

+		return -ENXIO;

Seeing how its fairly hard to schedule work on a cpu that's not actually
there.

+
+	queue_work_on(cpu, system_wq, &sscs.work);
+	wait_for_completion(&sscs.done);
+
+	return sscs.ret;
+}