drivers/cpuidle/cpuidle-powernv.c

   1 /*
   2  *  cpuidle-powernv - idle state cpuidle driver.
   3  *  Adapted from drivers/cpuidle/cpuidle-pseries
   4  *
   5  */
   6
   7 #include <linux/kernel.h>
   8 #include <linux/module.h>
   9 #include <linux/init.h>
  10 #include <linux/moduleparam.h>
  11 #include <linux/cpuidle.h>
  12 #include <linux/cpu.h>
  13 #include <linux/notifier.h>
  14 #include <linux/clockchips.h>
  15 #include <linux/of.h>
  16 #include <linux/slab.h>
  17
  18 #include <asm/machdep.h>
  19 #include <asm/firmware.h>
  20 #include <asm/opal.h>
  21 #include <asm/runlatch.h>
  22
  23 #define MAX_POWERNV_IDLE_STATES 8
  24
  25 struct cpuidle_driver powernv_idle_driver = {
  26         .name             = "powernv_idle",
  27         .owner            = THIS_MODULE,
  28 };
  29
  30 static int max_idle_state;
  31 static struct cpuidle_state *cpuidle_state_table;
  32 static u64 snooze_timeout;
  33 static bool snooze_timeout_en;
  34
  35 static int snooze_loop(struct cpuidle_device *dev,
  36                         struct cpuidle_driver *drv,
  37                         int index)
  38 {
  39         u64 snooze_exit_time;
  40
  41         local_irq_enable();
  42         set_thread_flag(TIF_POLLING_NRFLAG);
  43
  44         snooze_exit_time = get_tb() + snooze_timeout;
  45         ppc64_runlatch_off();
  46         while (!need_resched()) {
  47                 HMT_low();
  48                 HMT_very_low();
  49                 if (snooze_timeout_en && get_tb() > snooze_exit_time)
  50                         break;
  51         }
  52
  53         HMT_medium();
  54         ppc64_runlatch_on();
  55         clear_thread_flag(TIF_POLLING_NRFLAG);
  56         smp_mb();
  57         return index;
  58 }
  59
  60 static int nap_loop(struct cpuidle_device *dev,
  61                         struct cpuidle_driver *drv,
  62                         int index)
  63 {
  64         ppc64_runlatch_off();
  65         power7_idle();
  66         ppc64_runlatch_on();
  67         return index;
  68 }
  69
  70 /* Register for fastsleep only in oneshot mode of broadcast */
  71 #ifdef CONFIG_TICK_ONESHOT
  72 static int fastsleep_loop(struct cpuidle_device *dev,
  73                                 struct cpuidle_driver *drv,
  74                                 int index)
  75 {
  76         unsigned long old_lpcr = mfspr(SPRN_LPCR);
  77         unsigned long new_lpcr;
  78
  79         if (unlikely(system_state < SYSTEM_RUNNING))
  80                 return index;
  81
  82         new_lpcr = old_lpcr;
  83         /* Do not exit powersave upon decrementer as we've setup the timer
  84          * offload.
  85          */
  86         new_lpcr &= ~LPCR_PECE1;
  87
  88         mtspr(SPRN_LPCR, new_lpcr);
  89         power7_sleep();
  90
  91         mtspr(SPRN_LPCR, old_lpcr);
  92
  93         return index;
  94 }
  95 #endif
  96 /*
  97  * States for dedicated partition case.
  98  */
  99 static struct cpuidle_state powernv_states[MAX_POWERNV_IDLE_STATES] = {
 100         { /* Snooze */
 101                 .name = "snooze",
 102                 .desc = "snooze",
 103                 .exit_latency = 0,
 104                 .target_residency = 0,
 105                 .enter = &snooze_loop },
 106 };
 107
 108 static int powernv_cpuidle_add_cpu_notifier(struct notifier_block *n,
 109                         unsigned long action, void *hcpu)
 110 {
 111         int hotcpu = (unsigned long)hcpu;
 112         struct cpuidle_device *dev =
 113                                 per_cpu(cpuidle_devices, hotcpu);
 114
 115         if (dev && cpuidle_get_driver()) {
 116                 switch (action) {
 117                 case CPU_ONLINE:
 118                 case CPU_ONLINE_FROZEN:
 119                         cpuidle_pause_and_lock();
 120                         cpuidle_enable_device(dev);
 121                         cpuidle_resume_and_unlock();
 122                         break;
 123
 124                 case CPU_DEAD:
 125                 case CPU_DEAD_FROZEN:
 126                         cpuidle_pause_and_lock();
 127                         cpuidle_disable_device(dev);
 128                         cpuidle_resume_and_unlock();
 129                         break;
 130
 131                 default:
 132                         return NOTIFY_DONE;
 133                 }
 134         }
 135         return NOTIFY_OK;
 136 }
 137
 138 static struct notifier_block setup_hotplug_notifier = {
 139         .notifier_call = powernv_cpuidle_add_cpu_notifier,
 140 };
 141
 142 /*
 143  * powernv_cpuidle_driver_init()
 144  */
 145 static int powernv_cpuidle_driver_init(void)
 146 {
 147         int idle_state;
 148         struct cpuidle_driver *drv = &powernv_idle_driver;
 149
 150         drv->state_count = 0;
 151
 152         for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
 153                 /* Is the state not enabled? */
 154                 if (cpuidle_state_table[idle_state].enter == NULL)
 155                         continue;
 156
 157                 drv->states[drv->state_count] = /* structure copy */
 158                         cpuidle_state_table[idle_state];
 159
 160                 drv->state_count += 1;
 161         }
 162
 163         /*
 164          * On the PowerNV platform cpu_present may be less than cpu_possible in
 165          * cases when firmware detects the CPU, but it is not available to the
 166          * OS.  If CONFIG_HOTPLUG_CPU=n, then such CPUs are not hotplugable at
 167          * run time and hence cpu_devices are not created for those CPUs by the
 168          * generic topology_init().
 169          *
 170          * drv->cpumask defaults to cpu_possible_mask in
 171          * __cpuidle_driver_init().  This breaks cpuidle on PowerNV where
 172          * cpu_devices are not created for CPUs in cpu_possible_mask that
 173          * cannot be hot-added later at run time.
 174          *
 175          * Trying cpuidle_register_device() on a CPU without a cpu_device is
 176          * incorrect, so pass a correct CPU mask to the generic cpuidle driver.
 177          */
 178
 179         drv->cpumask = (struct cpumask *)cpu_present_mask;
 180
 181         return 0;
 182 }
 183
 184 static int powernv_add_idle_states(void)
 185 {
 186         struct device_node *power_mgt;
 187         int nr_idle_states = 1; /* Snooze */
 188         int dt_idle_states;
 189         u32 *latency_ns, *residency_ns, *flags;
 190         int i, rc;
 191
 192         /* Currently we have snooze statically defined */
 193
 194         power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
 195         if (!power_mgt) {
 196                 pr_warn("opal: PowerMgmt Node not found\n");
 197                 goto out;
 198         }
 199
 200         /* Read values of any property to determine the num of idle states */
 201         dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags");
 202         if (dt_idle_states < 0) {
 203                 pr_warn("cpuidle-powernv: no idle states found in the DT\n");
 204                 goto out;
 205         }
 206
 207         flags = kzalloc(sizeof(*flags) * dt_idle_states, GFP_KERNEL);
 208         if (of_property_read_u32_array(power_mgt,
 209                         "ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
 210                 pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n");
 211                 goto out_free_flags;
 212         }
 213
 214         latency_ns = kzalloc(sizeof(*latency_ns) * dt_idle_states, GFP_KERNEL);
 215         rc = of_property_read_u32_array(power_mgt,
 216                 "ibm,cpu-idle-state-latencies-ns", latency_ns, dt_idle_states);
 217         if (rc) {
 218                 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
 219                 goto out_free_latency;
 220         }
 221
 222         residency_ns = kzalloc(sizeof(*residency_ns) * dt_idle_states, GFP_KERNEL);
 223         rc = of_property_read_u32_array(power_mgt,
 224                 "ibm,cpu-idle-state-residency-ns", residency_ns, dt_idle_states);
 225
 226         for (i = 0; i < dt_idle_states; i++) {
 227
 228                 /*
 229                  * Cpuidle accepts exit_latency and target_residency in us.
 230                  * Use default target_residency values if f/w does not expose it.
 231                  */
 232                 if (flags[i] & OPAL_PM_NAP_ENABLED) {
 233                         /* Add NAP state */
 234                         strcpy(powernv_states[nr_idle_states].name, "Nap");
 235                         strcpy(powernv_states[nr_idle_states].desc, "Nap");
 236                         powernv_states[nr_idle_states].flags = 0;
 237                         powernv_states[nr_idle_states].target_residency = 100;
 238                         powernv_states[nr_idle_states].enter = &nap_loop;
 239                 }
 240
 241                 /*
 242                  * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
 243                  * within this config dependency check.
 244                  */
 245 #ifdef CONFIG_TICK_ONESHOT
 246                 if (flags[i] & OPAL_PM_SLEEP_ENABLED ||
 247                         flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) {
 248                         /* Add FASTSLEEP state */
 249                         strcpy(powernv_states[nr_idle_states].name, "FastSleep");
 250                         strcpy(powernv_states[nr_idle_states].desc, "FastSleep");
 251                         powernv_states[nr_idle_states].flags = CPUIDLE_FLAG_TIMER_STOP;
 252                         powernv_states[nr_idle_states].target_residency = 300000;
 253                         powernv_states[nr_idle_states].enter = &fastsleep_loop;
 254                 }
 255 #endif
 256                 powernv_states[nr_idle_states].exit_latency =
 257                                 ((unsigned int)latency_ns[i]) / 1000;
 258
 259                 if (!rc) {
 260                         powernv_states[nr_idle_states].target_residency =
 261                                 ((unsigned int)residency_ns[i]) / 1000;
 262                 }
 263
 264                 nr_idle_states++;
 265         }
 266
 267         kfree(residency_ns);
 268 out_free_latency:
 269         kfree(latency_ns);
 270 out_free_flags:
 271         kfree(flags);
 272 out:
 273         return nr_idle_states;
 274 }
 275
 276 /*
 277  * powernv_idle_probe()
 278  * Choose state table for shared versus dedicated partition
 279  */
 280 static int powernv_idle_probe(void)
 281 {
 282         if (cpuidle_disable != IDLE_NO_OVERRIDE)
 283                 return -ENODEV;
 284
 285         if (firmware_has_feature(FW_FEATURE_OPALv3)) {
 286                 cpuidle_state_table = powernv_states;
 287                 /* Device tree can indicate more idle states */
 288                 max_idle_state = powernv_add_idle_states();
 289                 if (max_idle_state > 1) {
 290                         snooze_timeout_en = true;
 291                         snooze_timeout = powernv_states[1].target_residency *
 292                                          tb_ticks_per_usec;
 293                 }
 294         } else
 295                 return -ENODEV;
 296
 297         return 0;
 298 }
 299
 300 static int __init powernv_processor_idle_init(void)
 301 {
 302         int retval;
 303
 304         retval = powernv_idle_probe();
 305         if (retval)
 306                 return retval;
 307
 308         powernv_cpuidle_driver_init();
 309         retval = cpuidle_register(&powernv_idle_driver, NULL);
 310         if (retval) {
 311                 printk(KERN_DEBUG "Registration of powernv driver failed.\n");
 312                 return retval;
 313         }
 314
 315         register_cpu_notifier(&setup_hotplug_notifier);
 316         printk(KERN_DEBUG "powernv_idle_driver registered\n");
 317         return 0;
 318 }
 319
 320 device_initcall(powernv_processor_idle_init);