Maximum Frequency Invariance has to be part of Cpu Invariance because
Frequency Invariance deals only with differences in load-tracking
introduces by Dynamic Frequency Scaling and not with limiting the
possible range of cpu frequency.
By placing Maximum Frequency Invariance into Cpu Invariance,
load-tracking is scaled via arch_scale_cpu_capacity()
in __update_load_avg() and cpu capacity is scaled via
arch_scale_cpu_capacity() in update_cpu_capacity().
To be able to save the extra multiplication in the scheduler hotpath
(__update_load_avg()) we could:
1 Inform cpufreq about base cpu capacity at boot and let it handle
scale_cpu_capacity() as well.
2 Use the cpufreq policy callback which would update a per-cpu current
cpu_scale and this value would be return in scale_cpu_capacity().
3 Use per-cpu current max_freq_scale and current cpu_scale with the
current patch.
Including <linux/cpufreq.h> in topology.h like for the arm arch doesn't
work because of CONFIG_COMPAT=y (Kernel support for 32-bit EL0).
That's why cpufreq_scale_max_freq_capacity() has to be declared extern
in topology.h.
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
#ifdef CONFIG_CPU_FREQ
#define arch_scale_freq_capacity cpufreq_scale_freq_capacity
extern unsigned long cpufreq_scale_freq_capacity(struct sched_domain *sd, int cpu);
+extern unsigned long cpufreq_scale_max_freq_capacity(int cpu);
#endif
#define arch_scale_cpu_capacity scale_cpu_capacity
extern unsigned long scale_cpu_capacity(struct sched_domain *sd, int cpu);
unsigned long scale_cpu_capacity(struct sched_domain *sd, int cpu)
{
+#ifdef CONFIG_CPU_FREQ
+ unsigned long max_freq_scale = cpufreq_scale_max_freq_capacity(cpu);
+
+ return per_cpu(cpu_scale, cpu) * max_freq_scale >> SCHED_CAPACITY_SHIFT;
+#else
return per_cpu(cpu_scale, cpu);
+#endif
}
static void set_capacity_scale(unsigned int cpu, unsigned long capacity)