Pileus Git - ~andy/linux/blob - 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/plpar_wrappers.h>
  21
  22 struct cpuidle_driver pseries_idle_driver = {
  23         .name             = "pseries_idle",
  24         .owner            = THIS_MODULE,
  25 };
  26
  27 #define MAX_IDLE_STATE_COUNT    2
  28
  29 static int max_idle_state = MAX_IDLE_STATE_COUNT - 1;
  30 static struct cpuidle_state *cpuidle_state_table;
  31
  32 static inline void idle_loop_prolog(unsigned long *in_purr)
  33 {
  34         *in_purr = mfspr(SPRN_PURR);
  35         /*
  36          * Indicate to the HV that we are idle. Now would be
  37          * a good time to find other work to dispatch.
  38          */
  39         get_lppaca()->idle = 1;
  40 }
  41
  42 static inline void idle_loop_epilog(unsigned long in_purr)
  43 {
  44         u64 wait_cycles;
  45
  46         wait_cycles = be64_to_cpu(get_lppaca()->wait_state_cycles);
  47         wait_cycles += mfspr(SPRN_PURR) - in_purr;
  48         get_lppaca()->wait_state_cycles = cpu_to_be64(wait_cycles);
  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                 HMT_low();
  65                 HMT_very_low();
  66         }
  67
  68         HMT_medium();
  69         clear_thread_flag(TIF_POLLING_NRFLAG);
  70         smp_mb();
  71
  72         idle_loop_epilog(in_purr);
  73
  74         return index;
  75 }
  76
  77 static void check_and_cede_processor(void)
  78 {
  79         /*
  80          * Ensure our interrupt state is properly tracked,
  81          * also checks if no interrupt has occurred while we
  82          * were soft-disabled
  83          */
  84         if (prep_irq_for_idle()) {
  85                 cede_processor();
  86 #ifdef CONFIG_TRACE_IRQFLAGS
  87                 /* Ensure that H_CEDE returns with IRQs on */
  88                 if (WARN_ON(!(mfmsr() & MSR_EE)))
  89                         __hard_irq_enable();
  90 #endif
  91         }
  92 }
  93
  94 static int dedicated_cede_loop(struct cpuidle_device *dev,
  95                                 struct cpuidle_driver *drv,
  96                                 int index)
  97 {
  98         unsigned long in_purr;
  99
 100         idle_loop_prolog(&in_purr);
 101         get_lppaca()->donate_dedicated_cpu = 1;
 102
 103         HMT_medium();
 104         check_and_cede_processor();
 105
 106         get_lppaca()->donate_dedicated_cpu = 0;
 107
 108         idle_loop_epilog(in_purr);
 109
 110         return index;
 111 }
 112
 113 static int shared_cede_loop(struct cpuidle_device *dev,
 114                         struct cpuidle_driver *drv,
 115                         int index)
 116 {
 117         unsigned long in_purr;
 118
 119         idle_loop_prolog(&in_purr);
 120
 121         /*
 122          * Yield the processor to the hypervisor.  We return if
 123          * an external interrupt occurs (which are driven prior
 124          * to returning here) or if a prod occurs from another
 125          * processor. When returning here, external interrupts
 126          * are enabled.
 127          */
 128         check_and_cede_processor();
 129
 130         idle_loop_epilog(in_purr);
 131
 132         return index;
 133 }
 134
 135 /*
 136  * States for dedicated partition case.
 137  */
 138 static struct cpuidle_state dedicated_states[MAX_IDLE_STATE_COUNT] = {
 139         { /* Snooze */
 140                 .name = "snooze",
 141                 .desc = "snooze",
 142                 .flags = CPUIDLE_FLAG_TIME_VALID,
 143                 .exit_latency = 0,
 144                 .target_residency = 0,
 145                 .enter = &snooze_loop },
 146         { /* CEDE */
 147                 .name = "CEDE",
 148                 .desc = "CEDE",
 149                 .flags = CPUIDLE_FLAG_TIME_VALID,
 150                 .exit_latency = 10,
 151                 .target_residency = 100,
 152                 .enter = &dedicated_cede_loop },
 153 };
 154
 155 /*
 156  * States for shared partition case.
 157  */
 158 static struct cpuidle_state shared_states[MAX_IDLE_STATE_COUNT] = {
 159         { /* Shared Cede */
 160                 .name = "Shared Cede",
 161                 .desc = "Shared Cede",
 162                 .flags = CPUIDLE_FLAG_TIME_VALID,
 163                 .exit_latency = 0,
 164                 .target_residency = 0,
 165                 .enter = &shared_cede_loop },
 166 };
 167
 168 void update_smt_snooze_delay(int cpu, int residency)
 169 {
 170         struct cpuidle_driver *drv = cpuidle_get_driver();
 171         struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
 172
 173         if (cpuidle_state_table != dedicated_states)
 174                 return;
 175
 176         if (residency < 0) {
 177                 /* Disable the Nap state on that cpu */
 178                 if (dev)
 179                         dev->states_usage[1].disable = 1;
 180         } else
 181                 if (drv)
 182                         drv->states[1].target_residency = residency;
 183 }
 184
 185 static int pseries_cpuidle_add_cpu_notifier(struct notifier_block *n,
 186                         unsigned long action, void *hcpu)
 187 {
 188         int hotcpu = (unsigned long)hcpu;
 189         struct cpuidle_device *dev =
 190                         per_cpu_ptr(cpuidle_devices, hotcpu);
 191
 192         if (dev && cpuidle_get_driver()) {
 193                 switch (action) {
 194                 case CPU_ONLINE:
 195                 case CPU_ONLINE_FROZEN:
 196                         cpuidle_pause_and_lock();
 197                         cpuidle_enable_device(dev);
 198                         cpuidle_resume_and_unlock();
 199                         break;
 200
 201                 case CPU_DEAD:
 202                 case CPU_DEAD_FROZEN:
 203                         cpuidle_pause_and_lock();
 204                         cpuidle_disable_device(dev);
 205                         cpuidle_resume_and_unlock();
 206                         break;
 207
 208                 default:
 209                         return NOTIFY_DONE;
 210                 }
 211         }
 212         return NOTIFY_OK;
 213 }
 214
 215 static struct notifier_block setup_hotplug_notifier = {
 216         .notifier_call = pseries_cpuidle_add_cpu_notifier,
 217 };
 218
 219 /*
 220  * pseries_cpuidle_driver_init()
 221  */
 222 static int pseries_cpuidle_driver_init(void)
 223 {
 224         int idle_state;
 225         struct cpuidle_driver *drv = &pseries_idle_driver;
 226
 227         drv->state_count = 0;
 228
 229         for (idle_state = 0; idle_state < MAX_IDLE_STATE_COUNT; ++idle_state) {
 230
 231                 if (idle_state > max_idle_state)
 232                         break;
 233
 234                 /* is the state not enabled? */
 235                 if (cpuidle_state_table[idle_state].enter == NULL)
 236                         continue;
 237
 238                 drv->states[drv->state_count] = /* structure copy */
 239                         cpuidle_state_table[idle_state];
 240
 241                 drv->state_count += 1;
 242         }
 243
 244         return 0;
 245 }
 246
 247 /*
 248  * pseries_idle_probe()
 249  * Choose state table for shared versus dedicated partition
 250  */
 251 static int pseries_idle_probe(void)
 252 {
 253
 254         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 255                 return -ENODEV;
 256
 257         if (cpuidle_disable != IDLE_NO_OVERRIDE)
 258                 return -ENODEV;
 259
 260         if (max_idle_state == 0) {
 261                 printk(KERN_DEBUG "pseries processor idle disabled.\n");
 262                 return -EPERM;
 263         }
 264
 265         if (lppaca_shared_proc(get_lppaca()))
 266                 cpuidle_state_table = shared_states;
 267         else
 268                 cpuidle_state_table = dedicated_states;
 269
 270         return 0;
 271 }
 272
 273 static int __init pseries_processor_idle_init(void)
 274 {
 275         int retval;
 276
 277         retval = pseries_idle_probe();
 278         if (retval)
 279                 return retval;
 280
 281         pseries_cpuidle_driver_init();
 282         retval = cpuidle_register(&pseries_idle_driver, NULL);
 283         if (retval) {
 284                 printk(KERN_DEBUG "Registration of pseries driver failed.\n");
 285                 return retval;
 286         }
 287
 288         register_cpu_notifier(&setup_hotplug_notifier);
 289         printk(KERN_DEBUG "pseries_idle_driver registered\n");
 290
 291         return 0;
 292 }
 293
 294 static void __exit pseries_processor_idle_exit(void)
 295 {
 296
 297         unregister_cpu_notifier(&setup_hotplug_notifier);
 298         cpuidle_unregister(&pseries_idle_driver);
 299
 300         return;
 301 }
 302
 303 module_init(pseries_processor_idle_init);
 304 module_exit(pseries_processor_idle_exit);
 305
 306 MODULE_AUTHOR("Deepthi Dharwar <deepthi@linux.vnet.ibm.com>");
 307 MODULE_DESCRIPTION("Cpuidle driver for POWER");
 308 MODULE_LICENSE("GPL");