[PATCH 2/2] cpufreq: scpi: remove arm_big_little dependency

On 10/01/18 04:19, Viresh Kumar wrote:

On 09-01-18, 16:49, Sudeep Holla wrote:

quoted

+static int
+scpi_get_sharing_cpus(struct device *cpu_dev, struct cpumask *cpumask)
 {
-	return scpi_ops->get_transition_latency(cpu_dev);
+	int cpu, domain, tdomain;
+	struct device *tcpu_dev;
+
+	domain = scpi_ops->device_domain_id(cpu_dev);
+	if (domain < 0)
+		return domain;
+
+	for_each_possible_cpu(cpu) {
+		if (cpu == cpu_dev->id)
+			continue;
+
+		tcpu_dev = get_cpu_device(cpu);
+		if (!tcpu_dev)
+			continue;
+
+		tdomain = scpi_ops->device_domain_id(tcpu_dev);
+		if (tdomain == domain)
+			cpumask_set_cpu(cpu, cpumask);
+	}
+
+	return 0;
 }

So this is the main thing you want to achieve ? i.e. get the
policy->cpus from scpi_ops, right ?

From the looks of it yes, but...

Why can't we update .init_opp_table() to include policy as a parameter
and let individual stub drivers update policy->cpus then ?

Possible, again but ...

There are few reasons why I would like remove scpi dependency on bL driver:

1. It has a notion of big and little which may not be true but that not
   much of a problem.

2. MAX_CLUSTER = 2 and scpi is getting used on multi-cluster systems
   though it was first tested on Juno which was bL system. There are
   quite a few internal FVP platforms that can manage to run with just
   proper DT with this change.

3. raw_cpu_to_cluster still calls topology_physical_package_id which
   breaks on these platforms and also with introduction of ACPI PPTT it
   will break on all ARM64 platforms.

4. We can still leave usage of topology_physical_package_id as is in
   arm_big_little.c worst case if it's going to be used only on ARM32
   For ARM64, topology_physical_package_id will be changed to give
   actual physical socket number which will be 0 for all multi-cluster
   (including bL) systems as along as they are single chip.

-- 
Regards,
Sudeep