Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / cpuidle / cpuidle-powernv.c
diff --git a/kernel/drivers/cpuidle/cpuidle-powernv.c b/kernel/drivers/cpuidle/cpuidle-powernv.c
new file mode 100644 (file)
index 0000000..3442764
--- /dev/null
@@ -0,0 +1,290 @@
+/*
+ *  cpuidle-powernv - idle state cpuidle driver.
+ *  Adapted from drivers/cpuidle/cpuidle-pseries
+ *
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/moduleparam.h>
+#include <linux/cpuidle.h>
+#include <linux/cpu.h>
+#include <linux/notifier.h>
+#include <linux/clockchips.h>
+#include <linux/of.h>
+#include <linux/slab.h>
+
+#include <asm/machdep.h>
+#include <asm/firmware.h>
+#include <asm/opal.h>
+#include <asm/runlatch.h>
+
+#define MAX_POWERNV_IDLE_STATES        8
+
+struct cpuidle_driver powernv_idle_driver = {
+       .name             = "powernv_idle",
+       .owner            = THIS_MODULE,
+};
+
+static int max_idle_state;
+static struct cpuidle_state *cpuidle_state_table;
+
+static int snooze_loop(struct cpuidle_device *dev,
+                       struct cpuidle_driver *drv,
+                       int index)
+{
+       local_irq_enable();
+       set_thread_flag(TIF_POLLING_NRFLAG);
+
+       ppc64_runlatch_off();
+       while (!need_resched()) {
+               HMT_low();
+               HMT_very_low();
+       }
+
+       HMT_medium();
+       ppc64_runlatch_on();
+       clear_thread_flag(TIF_POLLING_NRFLAG);
+       smp_mb();
+       return index;
+}
+
+static int nap_loop(struct cpuidle_device *dev,
+                       struct cpuidle_driver *drv,
+                       int index)
+{
+       ppc64_runlatch_off();
+       power7_idle();
+       ppc64_runlatch_on();
+       return index;
+}
+
+/* Register for fastsleep only in oneshot mode of broadcast */
+#ifdef CONFIG_TICK_ONESHOT
+static int fastsleep_loop(struct cpuidle_device *dev,
+                               struct cpuidle_driver *drv,
+                               int index)
+{
+       unsigned long old_lpcr = mfspr(SPRN_LPCR);
+       unsigned long new_lpcr;
+
+       if (unlikely(system_state < SYSTEM_RUNNING))
+               return index;
+
+       new_lpcr = old_lpcr;
+       /* Do not exit powersave upon decrementer as we've setup the timer
+        * offload.
+        */
+       new_lpcr &= ~LPCR_PECE1;
+
+       mtspr(SPRN_LPCR, new_lpcr);
+       power7_sleep();
+
+       mtspr(SPRN_LPCR, old_lpcr);
+
+       return index;
+}
+#endif
+/*
+ * States for dedicated partition case.
+ */
+static struct cpuidle_state powernv_states[MAX_POWERNV_IDLE_STATES] = {
+       { /* Snooze */
+               .name = "snooze",
+               .desc = "snooze",
+               .exit_latency = 0,
+               .target_residency = 0,
+               .enter = &snooze_loop },
+};
+
+static int powernv_cpuidle_add_cpu_notifier(struct notifier_block *n,
+                       unsigned long action, void *hcpu)
+{
+       int hotcpu = (unsigned long)hcpu;
+       struct cpuidle_device *dev =
+                               per_cpu(cpuidle_devices, hotcpu);
+
+       if (dev && cpuidle_get_driver()) {
+               switch (action) {
+               case CPU_ONLINE:
+               case CPU_ONLINE_FROZEN:
+                       cpuidle_pause_and_lock();
+                       cpuidle_enable_device(dev);
+                       cpuidle_resume_and_unlock();
+                       break;
+
+               case CPU_DEAD:
+               case CPU_DEAD_FROZEN:
+                       cpuidle_pause_and_lock();
+                       cpuidle_disable_device(dev);
+                       cpuidle_resume_and_unlock();
+                       break;
+
+               default:
+                       return NOTIFY_DONE;
+               }
+       }
+       return NOTIFY_OK;
+}
+
+static struct notifier_block setup_hotplug_notifier = {
+       .notifier_call = powernv_cpuidle_add_cpu_notifier,
+};
+
+/*
+ * powernv_cpuidle_driver_init()
+ */
+static int powernv_cpuidle_driver_init(void)
+{
+       int idle_state;
+       struct cpuidle_driver *drv = &powernv_idle_driver;
+
+       drv->state_count = 0;
+
+       for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
+               /* Is the state not enabled? */
+               if (cpuidle_state_table[idle_state].enter == NULL)
+                       continue;
+
+               drv->states[drv->state_count] = /* structure copy */
+                       cpuidle_state_table[idle_state];
+
+               drv->state_count += 1;
+       }
+
+       return 0;
+}
+
+static int powernv_add_idle_states(void)
+{
+       struct device_node *power_mgt;
+       int nr_idle_states = 1; /* Snooze */
+       int dt_idle_states;
+       u32 *latency_ns, *residency_ns, *flags;
+       int i, rc;
+
+       /* Currently we have snooze statically defined */
+
+       power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
+       if (!power_mgt) {
+               pr_warn("opal: PowerMgmt Node not found\n");
+               goto out;
+       }
+
+       /* Read values of any property to determine the num of idle states */
+       dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags");
+       if (dt_idle_states < 0) {
+               pr_warn("cpuidle-powernv: no idle states found in the DT\n");
+               goto out;
+       }
+
+       flags = kzalloc(sizeof(*flags) * dt_idle_states, GFP_KERNEL);
+       if (of_property_read_u32_array(power_mgt,
+                       "ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
+               pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n");
+               goto out_free_flags;
+       }
+
+       latency_ns = kzalloc(sizeof(*latency_ns) * dt_idle_states, GFP_KERNEL);
+       rc = of_property_read_u32_array(power_mgt,
+               "ibm,cpu-idle-state-latencies-ns", latency_ns, dt_idle_states);
+       if (rc) {
+               pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
+               goto out_free_latency;
+       }
+
+       residency_ns = kzalloc(sizeof(*residency_ns) * dt_idle_states, GFP_KERNEL);
+       rc = of_property_read_u32_array(power_mgt,
+               "ibm,cpu-idle-state-residency-ns", residency_ns, dt_idle_states);
+
+       for (i = 0; i < dt_idle_states; i++) {
+
+               /*
+                * Cpuidle accepts exit_latency and target_residency in us.
+                * Use default target_residency values if f/w does not expose it.
+                */
+               if (flags[i] & OPAL_PM_NAP_ENABLED) {
+                       /* Add NAP state */
+                       strcpy(powernv_states[nr_idle_states].name, "Nap");
+                       strcpy(powernv_states[nr_idle_states].desc, "Nap");
+                       powernv_states[nr_idle_states].flags = 0;
+                       powernv_states[nr_idle_states].target_residency = 100;
+                       powernv_states[nr_idle_states].enter = &nap_loop;
+               }
+
+               /*
+                * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
+                * within this config dependency check.
+                */
+#ifdef CONFIG_TICK_ONESHOT
+               if (flags[i] & OPAL_PM_SLEEP_ENABLED ||
+                       flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) {
+                       /* Add FASTSLEEP state */
+                       strcpy(powernv_states[nr_idle_states].name, "FastSleep");
+                       strcpy(powernv_states[nr_idle_states].desc, "FastSleep");
+                       powernv_states[nr_idle_states].flags = CPUIDLE_FLAG_TIMER_STOP;
+                       powernv_states[nr_idle_states].target_residency = 300000;
+                       powernv_states[nr_idle_states].enter = &fastsleep_loop;
+               }
+#endif
+               powernv_states[nr_idle_states].exit_latency =
+                               ((unsigned int)latency_ns[i]) / 1000;
+
+               if (!rc) {
+                       powernv_states[nr_idle_states].target_residency =
+                               ((unsigned int)residency_ns[i]) / 1000;
+               }
+
+               nr_idle_states++;
+       }
+
+       kfree(residency_ns);
+out_free_latency:
+       kfree(latency_ns);
+out_free_flags:
+       kfree(flags);
+out:
+       return nr_idle_states;
+}
+
+/*
+ * powernv_idle_probe()
+ * Choose state table for shared versus dedicated partition
+ */
+static int powernv_idle_probe(void)
+{
+       if (cpuidle_disable != IDLE_NO_OVERRIDE)
+               return -ENODEV;
+
+       if (firmware_has_feature(FW_FEATURE_OPALv3)) {
+               cpuidle_state_table = powernv_states;
+               /* Device tree can indicate more idle states */
+               max_idle_state = powernv_add_idle_states();
+       } else
+               return -ENODEV;
+
+       return 0;
+}
+
+static int __init powernv_processor_idle_init(void)
+{
+       int retval;
+
+       retval = powernv_idle_probe();
+       if (retval)
+               return retval;
+
+       powernv_cpuidle_driver_init();
+       retval = cpuidle_register(&powernv_idle_driver, NULL);
+       if (retval) {
+               printk(KERN_DEBUG "Registration of powernv driver failed.\n");
+               return retval;
+       }
+
+       register_cpu_notifier(&setup_hotplug_notifier);
+       printk(KERN_DEBUG "powernv_idle_driver registered\n");
+       return 0;
+}
+
+device_initcall(powernv_processor_idle_init);