[PATCH v1 3/3] arch_topology: Scale CPU capacity if without CPU raw capacity

[Leo Yan <leo.yan@linaro.org>]

Unlike a typical Arm big.LITTLE architecture, some Arm systems (like
Qualcomm SoC msm8996 and msm8939) have two clusters, all CPUs in two
clusters have the same micro architecture, but some CPUs are "fast" and
other are "slow".  On this kind platform, all CPUs have the same raw CPU
capacity but "fast" CPUs have higher maximum frequency than "slow" ones.

Let's see an example, there have two clusters and every cluster have 4
CPUs, every CPU has the same raw CPU capacity.  The cluster 0 has the
maximum frequency 1497.6MHz and the cluster 1 has the maximum frequency
1113.6MHz, if don't specify "capacity-dmips-mhz" in DT, the we will
get below result:

  # cat /sys/devices/system/cpu/cpu*/cpu_capacity
  1024
  1024
  1024
  1024
  1024
  1024
  1024
  1024

If "capacity-dmips-mhz" property is not specified for CPU nodes, the
kernel will fallback to default capacity value SCHED_CAPACITY_SCALE
(1024).  Though CPUs in different clusters have different maximum
frequencies, kernel skips to scale CPU capacity so that every CPU
capacity is always SCHED_CAPACITY_SCALE (1024).

This patch is to scale CPU capacity even though "capacity-dmips-mhz"
property is not specified in DT.  If "capacity-dmips-mhz" property is
absent in DT binding, the array "raw_capacity" is not allocated so we
rollback to use SCHED_CAPACITY_SCALE as raw CPU capacity and proceed
to scale CPU capacity based on maximum frequency.

After apply this patch, we can get below result for up elaborated
platform:

  # cat /sys/devices/system/cpu/cpu*/cpu_capacity
  1024
  1024
  1024
  1024
  761
  761
  761
  761

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 drivers/base/arch_topology.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
index 0687576e880b..ef1fa2e417ea 100644
--- a/drivers/base/arch_topology.c
+++ b/drivers/base/arch_topology.c
@@ -267,20 +267,25 @@ void topology_normalize_cpu_scale(void)
 {
 	u64 capacity;
 	u64 capacity_scale;
+	u32 raw_cpu_capacity;
 	int cpu;
 
-	if (!raw_capacity)
+	if (cap_parsing_failed)
 		return;
 
 	capacity_scale = 1;
 	for_each_possible_cpu(cpu) {
-		capacity = raw_capacity[cpu] * per_cpu(freq_factor, cpu);
+		raw_cpu_capacity =
+			raw_capacity ? raw_capacity[cpu] : SCHED_CAPACITY_SCALE;
+		capacity = raw_cpu_capacity * per_cpu(freq_factor, cpu);
 		capacity_scale = max(capacity, capacity_scale);
 	}
 
 	pr_debug("cpu_capacity: capacity_scale=%llu\n", capacity_scale);
 	for_each_possible_cpu(cpu) {
-		capacity = raw_capacity[cpu] * per_cpu(freq_factor, cpu);
+		raw_cpu_capacity =
+			raw_capacity ? raw_capacity[cpu] : SCHED_CAPACITY_SCALE;
+		capacity = raw_cpu_capacity * per_cpu(freq_factor, cpu);
 		capacity = div64_u64(capacity << SCHED_CAPACITY_SHIFT,
 			capacity_scale);
 		topology_set_cpu_scale(cpu, capacity);
@@ -373,7 +378,7 @@ init_cpu_capacity_callback(struct notifier_block *nb,
 	struct cpufreq_policy *policy = data;
 	int cpu;
 
-	if (!raw_capacity)
+	if (cap_parsing_failed)
 		return 0;
 
 	if (val != CPUFREQ_CREATE_POLICY)
@@ -412,7 +417,7 @@ static int __init register_cpufreq_notifier(void)
 	 * until we have the necessary code to parse the cpu capacity, so
 	 * skip registering cpufreq notifier.
 	 */
-	if (!acpi_disabled || !raw_capacity)
+	if (!acpi_disabled || cap_parsing_failed)
 		return -EINVAL;
 
 	if (!alloc_cpumask_var(&cpus_to_visit, GFP_KERNEL))
-- 
2.25.1