Added support for starting CPUs in parallel on EyeQ to speed up boot time.
On EyeQ5, booting 8 CPUs is now ~90ms faster.
On EyeQ6, booting 32 CPUs is now ~650ms faster.
Signed-off-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
select USB_UHCI_BIG_ENDIAN_DESC if CPU_BIG_ENDIAN
select USB_UHCI_BIG_ENDIAN_MMIO if CPU_BIG_ENDIAN
select USE_OF
+ select HOTPLUG_PARALLEL if SMP
help
Select this to build a kernel supporting EyeQ SoC from Mobileye.
select MIPS_CM
select MIPS_CPS_PM if HOTPLUG_CPU
select SMP
+ select HOTPLUG_SMT if HOTPLUG_PARALLEL
select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
select SYNC_R4K if (CEVT_R4K || CSRC_R4K)
select SYS_SUPPORTS_HOTPLUG_CPU
#define topology_core_id(cpu) (cpu_core(&cpu_data[cpu]))
#define topology_core_cpumask(cpu) (&cpu_core_map[cpu])
#define topology_sibling_cpumask(cpu) (&cpu_sibling_map[cpu])
+
+extern struct cpumask __cpu_primary_thread_mask;
+#define cpu_primary_thread_mask ((const struct cpumask *)&__cpu_primary_thread_mask)
#endif
#endif /* __ASM_TOPOLOGY_H */
/* Use the number of VPEs in cluster 0 core 0 for smp_num_siblings */
if (!cl && !c)
smp_num_siblings = core_vpes;
+ cpumask_set_cpu(nvpes, &__cpu_primary_thread_mask);
for (v = 0; v < min_t(int, core_vpes, NR_CPUS - nvpes); v++) {
cpu_set_cluster(&cpu_data[nvpes + v], cl);
cl = cpu_cluster(¤t_cpu_data);
c = cpu_core(¤t_cpu_data);
cluster_bootcfg = &mips_cps_cluster_bootcfg[cl];
+ cpu_smt_set_num_threads(core_vpes, core_vpes);
core_bootcfg = &cluster_bootcfg->core_config[c];
bitmap_set(cluster_bootcfg->core_power, cpu_core(¤t_cpu_data), 1);
atomic_set(&core_bootcfg->vpe_mask, 1 << cpu_vpe_id(¤t_cpu_data));
cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
EXPORT_SYMBOL(cpu_core_map);
+#ifndef CONFIG_HOTPLUG_PARALLEL
static DECLARE_COMPLETION(cpu_starting);
static DECLARE_COMPLETION(cpu_running);
+#endif
/*
* A logical cpu mask containing only one VPE per core to
cpumask_t cpu_coherent_mask;
+struct cpumask __cpu_primary_thread_mask __read_mostly;
+
unsigned int smp_max_threads __initdata = UINT_MAX;
static int __init early_nosmt(char *s)
set_cpu_core_map(cpu);
cpumask_set_cpu(cpu, &cpu_coherent_mask);
+#ifdef CONFIG_HOTPLUG_PARALLEL
+ cpuhp_ap_sync_alive();
+#endif
notify_cpu_starting(cpu);
+#ifndef CONFIG_HOTPLUG_PARALLEL
/* Notify boot CPU that we're starting & ready to sync counters */
complete(&cpu_starting);
+#endif
synchronise_count_slave(cpu);
calculate_cpu_foreign_map();
+#ifndef CONFIG_HOTPLUG_PARALLEL
/*
* Notify boot CPU that we're up & online and it can safely return
* from __cpu_up
*/
complete(&cpu_running);
+#endif
/*
* irq will be enabled in ->smp_finish(), enabling it too early
set_cpu_online(0, true);
}
+#ifdef CONFIG_HOTPLUG_PARALLEL
+int arch_cpuhp_kick_ap_alive(unsigned int cpu, struct task_struct *tidle)
+{
+ return mp_ops->boot_secondary(cpu, tidle);
+}
+#else
int __cpu_up(unsigned int cpu, struct task_struct *tidle)
{
int err;
wait_for_completion(&cpu_running);
return 0;
}
+#endif
#ifdef CONFIG_PROFILING
/* Not really SMP stuff ... */