arch/powerpc/platforms/pseries/processor_idle.c

   1 /*
   2  *  processor_idle - idle state cpuidle driver.
   3  *  Adapted from drivers/idle/intel_idle.c and
   4  *  drivers/acpi/processor_idle.c
   5  *
   6  */
   7
   8 #include <linux/kernel.h>
   9 #include <linux/module.h>
  10 #include <linux/init.h>
  11 #include <linux/moduleparam.h>
  12 #include <linux/cpuidle.h>
  13 #include <linux/cpu.h>
  14 #include <linux/notifier.h>
  15
  16 #include <asm/paca.h>
  17 #include <asm/reg.h>
  18 #include <asm/machdep.h>
  19 #include <asm/firmware.h>
  20 #include <asm/runlatch.h>
  21
  22 #include "plpar_wrappers.h"
  23 #include "pseries.h"
  24
  25 struct cpuidle_driver pseries_idle_driver = {
  26         .name             = "pseries_idle",
  27         .owner            = THIS_MODULE,
  28 };
  29
  30 #define MAX_IDLE_STATE_COUNT    2
  31
  32 static int max_idle_state = MAX_IDLE_STATE_COUNT - 1;
  33 static struct cpuidle_device __percpu *pseries_cpuidle_devices;
  34 static struct cpuidle_state *cpuidle_state_table;
  35
  36 static inline void idle_loop_prolog(unsigned long *in_purr)
  37 {
  38         *in_purr = mfspr(SPRN_PURR);
  39         /*
  40          * Indicate to the HV that we are idle. Now would be
  41          * a good time to find other work to dispatch.
  42          */
  43         get_lppaca()->idle = 1;
  44 }
  45
  46 static inline void idle_loop_epilog(unsigned long in_purr)
  47 {
  48         get_lppaca()->wait_state_cycles += mfspr(SPRN_PURR) - in_purr;
  49         get_lppaca()->idle = 0;
  50 }
  51
  52 static int snooze_loop(struct cpuidle_device *dev,
  53                         struct cpuidle_driver *drv,
  54                         int index)
  55 {
  56         unsigned long in_purr;
  57         int cpu = dev->cpu;
  58
  59         idle_loop_prolog(&in_purr);
  60         local_irq_enable();
  61         set_thread_flag(TIF_POLLING_NRFLAG);
  62
  63         while ((!need_resched()) && cpu_online(cpu)) {
  64                 ppc64_runlatch_off();
  65                 HMT_low();
  66                 HMT_very_low();
  67         }
  68
  69         HMT_medium();
  70         clear_thread_flag(TIF_POLLING_NRFLAG);
  71         smp_mb();
  72
  73         idle_loop_epilog(in_purr);
  74
  75         return index;
  76 }
  77
  78 static void check_and_cede_processor(void)
  79 {
  80         /*
  81          * Ensure our interrupt state is properly tracked,
  82          * also checks if no interrupt has occurred while we
  83          * were soft-disabled
  84          */
  85         if (prep_irq_for_idle()) {
  86                 cede_processor();
  87 #ifdef CONFIG_TRACE_IRQFLAGS
  88                 /* Ensure that H_CEDE returns with IRQs on */
  89                 if (WARN_ON(!(mfmsr() & MSR_EE)))
  90                         __hard_irq_enable();
  91 #endif
  92         }
  93 }
  94
  95 static int dedicated_cede_loop(struct cpuidle_device *dev,
  96                                 struct cpuidle_driver *drv,
  97                                 int index)
  98 {
  99         unsigned long in_purr;
 100
 101         idle_loop_prolog(&in_purr);
 102         get_lppaca()->donate_dedicated_cpu = 1;
 103
 104         ppc64_runlatch_off();
 105         HMT_medium();
 106         check_and_cede_processor();
 107
 108         get_lppaca()->donate_dedicated_cpu = 0;
 109
 110         idle_loop_epilog(in_purr);
 111
 112         return index;
 113 }
 114
 115 static int shared_cede_loop(struct cpuidle_device *dev,
 116                         struct cpuidle_driver *drv,
 117                         int index)
 118 {
 119         unsigned long in_purr;
 120
 121         idle_loop_prolog(&in_purr);
 122
 123         /*
 124          * Yield the processor to the hypervisor.  We return if
 125          * an external interrupt occurs (which are driven prior
 126          * to returning here) or if a prod occurs from another
 127          * processor. When returning here, external interrupts
 128          * are enabled.
 129          */
 130         check_and_cede_processor();
 131
 132         idle_loop_epilog(in_purr);
 133
 134         return index;
 135 }
 136
 137 /*
 138  * States for dedicated partition case.
 139  */
 140 static struct cpuidle_state dedicated_states[MAX_IDLE_STATE_COUNT] = {
 141         { /* Snooze */
 142                 .name = "snooze",
 143                 .desc = "snooze",
 144                 .flags = CPUIDLE_FLAG_TIME_VALID,
 145                 .exit_latency = 0,
 146                 .target_residency = 0,
 147                 .enter = &snooze_loop },
 148         { /* CEDE */
 149                 .name = "CEDE",
 150                 .desc = "CEDE",
 151                 .flags = CPUIDLE_FLAG_TIME_VALID,
 152                 .exit_latency = 10,
 153                 .target_residency = 100,
 154                 .enter = &dedicated_cede_loop },
 155 };
 156
 157 /*
 158  * States for shared partition case.
 159  */
 160 static struct cpuidle_state shared_states[MAX_IDLE_STATE_COUNT] = {
 161         { /* Shared Cede */
 162                 .name = "Shared Cede",
 163                 .desc = "Shared Cede",
 164                 .flags = CPUIDLE_FLAG_TIME_VALID,
 165                 .exit_latency = 0,
 166                 .target_residency = 0,
 167                 .enter = &shared_cede_loop },
 168 };
 169
 170 void update_smt_snooze_delay(int cpu, int residency)
 171 {
 172         struct cpuidle_driver *drv = cpuidle_get_driver();
 173         struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
 174
 175         if (cpuidle_state_table != dedicated_states)
 176                 return;
 177
 178         if (residency < 0) {
 179                 /* Disable the Nap state on that cpu */
 180                 if (dev)
 181                         dev->states_usage[1].disable = 1;
 182         } else
 183                 if (drv)
 184                         drv->states[1].target_residency = residency;
 185 }
 186
 187 static int pseries_cpuidle_add_cpu_notifier(struct notifier_block *n,
 188                         unsigned long action, void *hcpu)
 189 {
 190         int hotcpu = (unsigned long)hcpu;
 191         struct cpuidle_device *dev =
 192                         per_cpu_ptr(pseries_cpuidle_devices, hotcpu);
 193
 194         if (dev && cpuidle_get_driver()) {
 195                 switch (action) {
 196                 case CPU_ONLINE:
 197                 case CPU_ONLINE_FROZEN:
 198                         cpuidle_pause_and_lock();
 199                         cpuidle_enable_device(dev);
 200                         cpuidle_resume_and_unlock();
 201                         break;
 202
 203                 case CPU_DEAD:
 204                 case CPU_DEAD_FROZEN:
 205                         cpuidle_pause_and_lock();
 206                         cpuidle_disable_device(dev);
 207                         cpuidle_resume_and_unlock();
 208                         break;
 209
 210                 default:
 211                         return NOTIFY_DONE;
 212                 }
 213         }
 214         return NOTIFY_OK;
 215 }
 216
 217 static struct notifier_block setup_hotplug_notifier = {
 218         .notifier_call = pseries_cpuidle_add_cpu_notifier,
 219 };
 220
 221 /*
 222  * pseries_cpuidle_driver_init()
 223  */
 224 static int pseries_cpuidle_driver_init(void)
 225 {
 226         int idle_state;
 227         struct cpuidle_driver *drv = &pseries_idle_driver;
 228
 229         drv->state_count = 0;
 230
 231         for (idle_state = 0; idle_state < MAX_IDLE_STATE_COUNT; ++idle_state) {
 232
 233                 if (idle_state > max_idle_state)
 234                         break;
 235
 236                 /* is the state not enabled? */
 237                 if (cpuidle_state_table[idle_state].enter == NULL)
 238                         continue;
 239
 240                 drv->states[drv->state_count] = /* structure copy */
 241                         cpuidle_state_table[idle_state];
 242
 243                 drv->state_count += 1;
 244         }
 245
 246         return 0;
 247 }
 248
 249 /* pseries_idle_devices_uninit(void)
 250  * unregister cpuidle devices and de-allocate memory
 251  */
 252 static void pseries_idle_devices_uninit(void)
 253 {
 254         int i;
 255         struct cpuidle_device *dev;
 256
 257         for_each_possible_cpu(i) {
 258                 dev = per_cpu_ptr(pseries_cpuidle_devices, i);
 259                 cpuidle_unregister_device(dev);
 260         }
 261
 262         free_percpu(pseries_cpuidle_devices);
 263         return;
 264 }
 265
 266 /* pseries_idle_devices_init()
 267  * allocate, initialize and register cpuidle device
 268  */
 269 static int pseries_idle_devices_init(void)
 270 {
 271         int i;
 272         struct cpuidle_driver *drv = &pseries_idle_driver;
 273         struct cpuidle_device *dev;
 274
 275         pseries_cpuidle_devices = alloc_percpu(struct cpuidle_device);
 276         if (pseries_cpuidle_devices == NULL)
 277                 return -ENOMEM;
 278
 279         for_each_possible_cpu(i) {
 280                 dev = per_cpu_ptr(pseries_cpuidle_devices, i);
 281                 dev->state_count = drv->state_count;
 282                 dev->cpu = i;
 283                 if (cpuidle_register_device(dev)) {
 284                         printk(KERN_DEBUG \
 285                                 "cpuidle_register_device %d failed!\n", i);
 286                         return -EIO;
 287                 }
 288         }
 289
 290         return 0;
 291 }
 292
 293 /*
 294  * pseries_idle_probe()
 295  * Choose state table for shared versus dedicated partition
 296  */
 297 static int pseries_idle_probe(void)
 298 {
 299
 300         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 301                 return -ENODEV;
 302
 303         if (cpuidle_disable != IDLE_NO_OVERRIDE)
 304                 return -ENODEV;
 305
 306         if (max_idle_state == 0) {
 307                 printk(KERN_DEBUG "pseries processor idle disabled.\n");
 308                 return -EPERM;
 309         }
 310
 311         if (get_lppaca()->shared_proc)
 312                 cpuidle_state_table = shared_states;
 313         else
 314                 cpuidle_state_table = dedicated_states;
 315
 316         return 0;
 317 }
 318
 319 static int __init pseries_processor_idle_init(void)
 320 {
 321         int retval;
 322
 323         retval = pseries_idle_probe();
 324         if (retval)
 325                 return retval;
 326
 327         pseries_cpuidle_driver_init();
 328         retval = cpuidle_register_driver(&pseries_idle_driver);
 329         if (retval) {
 330                 printk(KERN_DEBUG "Registration of pseries driver failed.\n");
 331                 return retval;
 332         }
 333
 334         retval = pseries_idle_devices_init();
 335         if (retval) {
 336                 pseries_idle_devices_uninit();
 337                 cpuidle_unregister_driver(&pseries_idle_driver);
 338                 return retval;
 339         }
 340
 341         register_cpu_notifier(&setup_hotplug_notifier);
 342         printk(KERN_DEBUG "pseries_idle_driver registered\n");
 343
 344         return 0;
 345 }
 346
 347 static void __exit pseries_processor_idle_exit(void)
 348 {
 349
 350         unregister_cpu_notifier(&setup_hotplug_notifier);
 351         pseries_idle_devices_uninit();
 352         cpuidle_unregister_driver(&pseries_idle_driver);
 353
 354         return;
 355 }
 356
 357 module_init(pseries_processor_idle_init);
 358 module_exit(pseries_processor_idle_exit);
 359
 360 MODULE_AUTHOR("Deepthi Dharwar <deepthi@linux.vnet.ibm.com>");
 361 MODULE_DESCRIPTION("Cpuidle driver for POWER");
 362 MODULE_LICENSE("GPL");