Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / cpuidle / cpuidle-arm.c
diff --git a/kernel/drivers/cpuidle/cpuidle-arm.c b/kernel/drivers/cpuidle/cpuidle-arm.c
new file mode 100644 (file)
index 0000000..545069d
--- /dev/null
@@ -0,0 +1,163 @@
+/*
+ * ARM/ARM64 generic CPU idle driver.
+ *
+ * Copyright (C) 2014 ARM Ltd.
+ * Author: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#define pr_fmt(fmt) "CPUidle arm: " fmt
+
+#include <linux/cpuidle.h>
+#include <linux/cpumask.h>
+#include <linux/cpu_pm.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/slab.h>
+
+#include <asm/cpuidle.h>
+
+#include "dt_idle_states.h"
+
+/*
+ * arm_enter_idle_state - Programs CPU to enter the specified state
+ *
+ * dev: cpuidle device
+ * drv: cpuidle driver
+ * idx: state index
+ *
+ * Called from the CPUidle framework to program the device to the
+ * specified target state selected by the governor.
+ */
+static int arm_enter_idle_state(struct cpuidle_device *dev,
+                               struct cpuidle_driver *drv, int idx)
+{
+       int ret;
+
+       if (!idx) {
+               cpu_do_idle();
+               return idx;
+       }
+
+       ret = cpu_pm_enter();
+       if (!ret) {
+               /*
+                * Pass idle state index to cpu_suspend which in turn will
+                * call the CPU ops suspend protocol with idle index as a
+                * parameter.
+                */
+               arm_cpuidle_suspend(idx);
+
+               cpu_pm_exit();
+       }
+
+       return ret ? -1 : idx;
+}
+
+static struct cpuidle_driver arm_idle_driver = {
+       .name = "arm_idle",
+       .owner = THIS_MODULE,
+       /*
+        * State at index 0 is standby wfi and considered standard
+        * on all ARM platforms. If in some platforms simple wfi
+        * can't be used as "state 0", DT bindings must be implemented
+        * to work around this issue and allow installing a special
+        * handler for idle state index 0.
+        */
+       .states[0] = {
+               .enter                  = arm_enter_idle_state,
+               .exit_latency           = 1,
+               .target_residency       = 1,
+               .power_usage            = UINT_MAX,
+               .name                   = "WFI",
+               .desc                   = "ARM WFI",
+       }
+};
+
+static const struct of_device_id arm_idle_state_match[] __initconst = {
+       { .compatible = "arm,idle-state",
+         .data = arm_enter_idle_state },
+       { },
+};
+
+/*
+ * arm_idle_init
+ *
+ * Registers the arm specific cpuidle driver with the cpuidle
+ * framework. It relies on core code to parse the idle states
+ * and initialize them using driver data structures accordingly.
+ */
+static int __init arm_idle_init(void)
+{
+       int cpu, ret;
+       struct cpuidle_driver *drv = &arm_idle_driver;
+       struct cpuidle_device *dev;
+
+       /*
+        * Initialize idle states data, starting at index 1.
+        * This driver is DT only, if no DT idle states are detected (ret == 0)
+        * let the driver initialization fail accordingly since there is no
+        * reason to initialize the idle driver if only wfi is supported.
+        */
+       ret = dt_init_idle_driver(drv, arm_idle_state_match, 1);
+       if (ret <= 0)
+               return ret ? : -ENODEV;
+
+       ret = cpuidle_register_driver(drv);
+       if (ret) {
+               pr_err("Failed to register cpuidle driver\n");
+               return ret;
+       }
+
+       /*
+        * Call arch CPU operations in order to initialize
+        * idle states suspend back-end specific data
+        */
+       for_each_possible_cpu(cpu) {
+               ret = arm_cpuidle_init(cpu);
+
+               /*
+                * Skip the cpuidle device initialization if the reported
+                * failure is a HW misconfiguration/breakage (-ENXIO).
+                */
+               if (ret == -ENXIO)
+                       continue;
+
+               if (ret) {
+                       pr_err("CPU %d failed to init idle CPU ops\n", cpu);
+                       goto out_fail;
+               }
+
+               dev = kzalloc(sizeof(*dev), GFP_KERNEL);
+               if (!dev) {
+                       pr_err("Failed to allocate cpuidle device\n");
+                       goto out_fail;
+               }
+               dev->cpu = cpu;
+
+               ret = cpuidle_register_device(dev);
+               if (ret) {
+                       pr_err("Failed to register cpuidle device for CPU %d\n",
+                              cpu);
+                       kfree(dev);
+                       goto out_fail;
+               }
+       }
+
+       return 0;
+out_fail:
+       while (--cpu >= 0) {
+               dev = per_cpu(cpuidle_devices, cpu);
+               cpuidle_unregister_device(dev);
+               kfree(dev);
+       }
+
+       cpuidle_unregister_driver(drv);
+
+       return ret;
+}
+device_initcall(arm_idle_init);