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 #include <asm/plpar_wrappers.h>
  22
  23 struct cpuidle_driver pseries_idle_driver = {
  24         .name             = "pseries_idle",
  25         .owner            = THIS_MODULE,
  26 };
  27
  28 #define MAX_IDLE_STATE_COUNT    2
  29
  30 static int max_idle_state = MAX_IDLE_STATE_COUNT - 1;
  31 static struct cpuidle_state *cpuidle_state_table;
  32
  33 static inline void idle_loop_prolog(unsigned long *in_purr)
  34 {
  35         *in_purr = mfspr(SPRN_PURR);
  36         /*
  37          * Indicate to the HV that we are idle. Now would be
  38          * a good time to find other work to dispatch.
  39          */
  40         get_lppaca()->idle = 1;
  41 }
  42
  43 static inline void idle_loop_epilog(unsigned long in_purr)
  44 {
  45         u64 wait_cycles;
  46
  47         wait_cycles = be64_to_cpu(get_lppaca()->wait_state_cycles);
  48         wait_cycles += mfspr(SPRN_PURR) - in_purr;
  49         get_lppaca()->wait_state_cycles = cpu_to_be64(wait_cycles);
  50         get_lppaca()->idle = 0;
  51 }
  52
  53 static int snooze_loop(struct cpuidle_device *dev,
  54                         struct cpuidle_driver *drv,
  55                         int index)
  56 {
  57         unsigned long in_purr;
  58         int cpu = dev->cpu;
  59
  60         idle_loop_prolog(&in_purr);
  61         local_irq_enable();
  62         set_thread_flag(TIF_POLLING_NRFLAG);
  63
  64         while ((!need_resched()) && cpu_online(cpu)) {
  65                 ppc64_runlatch_off();
  66                 HMT_low();
  67                 HMT_very_low();
  68         }
  69
  70         HMT_medium();
  71         clear_thread_flag(TIF_POLLING_NRFLAG);
  72         smp_mb();
  73
  74         idle_loop_epilog(in_purr);
  75
  76         return index;
  77 }
  78
  79 static void check_and_cede_processor(void)
  80 {
  81         /*
  82          * Ensure our interrupt state is properly tracked,
  83          * also checks if no interrupt has occurred while we
  84          * were soft-disabled
  85          */
  86         if (prep_irq_for_idle()) {
  87                 cede_processor();
  88 #ifdef CONFIG_TRACE_IRQFLAGS
  89                 /* Ensure that H_CEDE returns with IRQs on */
  90                 if (WARN_ON(!(mfmsr() & MSR_EE)))
  91                         __hard_irq_enable();
  92 #endif
  93         }
  94 }
  95
  96 static int dedicated_cede_loop(struct cpuidle_device *dev,
  97                                 struct cpuidle_driver *drv,
  98                                 int index)
  99 {
 100         unsigned long in_purr;
 101
 102         idle_loop_prolog(&in_purr);
 103         get_lppaca()->donate_dedicated_cpu = 1;
 104
 105         ppc64_runlatch_off();
 106         HMT_medium();
 107         check_and_cede_processor();
 108
 109         get_lppaca()->donate_dedicated_cpu = 0;
 110
 111         idle_loop_epilog(in_purr);
 112
 113         return index;
 114 }
 115
 116 static int shared_cede_loop(struct cpuidle_device *dev,
 117                         struct cpuidle_driver *drv,
 118                         int index)
 119 {
 120         unsigned long in_purr;
 121
 122         idle_loop_prolog(&in_purr);
 123
 124         /*
 125          * Yield the processor to the hypervisor.  We return if
 126          * an external interrupt occurs (which are driven prior
 127          * to returning here) or if a prod occurs from another
 128          * processor. When returning here, external interrupts
 129          * are enabled.
 130          */
 131         check_and_cede_processor();
 132
 133         idle_loop_epilog(in_purr);
 134
 135         return index;
 136 }
 137
 138 /*
 139  * States for dedicated partition case.
 140  */
 141 static struct cpuidle_state dedicated_states[MAX_IDLE_STATE_COUNT] = {
 142         { /* Snooze */
 143                 .name = "snooze",
 144                 .desc = "snooze",
 145                 .flags = CPUIDLE_FLAG_TIME_VALID,
 146                 .exit_latency = 0,
 147                 .target_residency = 0,
 148                 .enter = &snooze_loop },
 149         { /* CEDE */
 150                 .name = "CEDE",
 151                 .desc = "CEDE",
 152                 .flags = CPUIDLE_FLAG_TIME_VALID,
 153                 .exit_latency = 10,
 154                 .target_residency = 100,
 155                 .enter = &dedicated_cede_loop },
 156 };
 157
 158 /*
 159  * States for shared partition case.
 160  */
 161 static struct cpuidle_state shared_states[MAX_IDLE_STATE_COUNT] = {
 162         { /* Shared Cede */
 163                 .name = "Shared Cede",
 164                 .desc = "Shared Cede",
 165                 .flags = CPUIDLE_FLAG_TIME_VALID,
 166                 .exit_latency = 0,
 167                 .target_residency = 0,
 168                 .enter = &shared_cede_loop },
 169 };
 170
 171 void update_smt_snooze_delay(int cpu, int residency)
 172 {
 173         struct cpuidle_driver *drv = cpuidle_get_driver();
 174         struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
 175
 176         if (cpuidle_state_table != dedicated_states)
 177                 return;
 178
 179         if (residency < 0) {
 180                 /* Disable the Nap state on that cpu */
 181                 if (dev)
 182                         dev->states_usage[1].disable = 1;
 183         } else
 184                 if (drv)
 185                         drv->states[1].target_residency = residency;
 186 }
 187
 188 static int pseries_cpuidle_add_cpu_notifier(struct notifier_block *n,
 189                         unsigned long action, void *hcpu)
 190 {
 191         int hotcpu = (unsigned long)hcpu;
 192         struct cpuidle_device *dev =
 193                         per_cpu_ptr(cpuidle_devices, hotcpu);
 194
 195         if (dev && cpuidle_get_driver()) {
 196                 switch (action) {
 197                 case CPU_ONLINE:
 198                 case CPU_ONLINE_FROZEN:
 199                         cpuidle_pause_and_lock();
 200                         cpuidle_enable_device(dev);
 201                         cpuidle_resume_and_unlock();
 202                         break;
 203
 204                 case CPU_DEAD:
 205                 case CPU_DEAD_FROZEN:
 206                         cpuidle_pause_and_lock();
 207                         cpuidle_disable_device(dev);
 208                         cpuidle_resume_and_unlock();
 209                         break;
 210
 211                 default:
 212                         return NOTIFY_DONE;
 213                 }
 214         }
 215         return NOTIFY_OK;
 216 }
 217
 218 static struct notifier_block setup_hotplug_notifier = {
 219         .notifier_call = pseries_cpuidle_add_cpu_notifier,
 220 };
 221
 222 /*
 223  * pseries_cpuidle_driver_init()
 224  */
 225 static int pseries_cpuidle_driver_init(void)
 226 {
 227         int idle_state;
 228         struct cpuidle_driver *drv = &pseries_idle_driver;
 229
 230         drv->state_count = 0;
 231
 232         for (idle_state = 0; idle_state < MAX_IDLE_STATE_COUNT; ++idle_state) {
 233
 234                 if (idle_state > max_idle_state)
 235                         break;
 236
 237                 /* is the state not enabled? */
 238                 if (cpuidle_state_table[idle_state].enter == NULL)
 239                         continue;
 240
 241                 drv->states[drv->state_count] = /* structure copy */
 242                         cpuidle_state_table[idle_state];
 243
 244                 drv->state_count += 1;
 245         }
 246
 247         return 0;
 248 }
 249
 250 /*
 251  * pseries_idle_probe()
 252  * Choose state table for shared versus dedicated partition
 253  */
 254 static int pseries_idle_probe(void)
 255 {
 256
 257         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
 258                 return -ENODEV;
 259
 260         if (cpuidle_disable != IDLE_NO_OVERRIDE)
 261                 return -ENODEV;
 262
 263         if (max_idle_state == 0) {
 264                 printk(KERN_DEBUG "pseries processor idle disabled.\n");
 265                 return -EPERM;
 266         }
 267
 268         if (lppaca_shared_proc(get_lppaca()))
 269                 cpuidle_state_table = shared_states;
 270         else
 271                 cpuidle_state_table = dedicated_states;
 272
 273         return 0;
 274 }
 275
 276 static int __init pseries_processor_idle_init(void)
 277 {
 278         int retval;
 279
 280         retval = pseries_idle_probe();
 281         if (retval)
 282                 return retval;
 283
 284         pseries_cpuidle_driver_init();
 285         retval = cpuidle_register(&pseries_idle_driver, NULL);
 286         if (retval) {
 287                 printk(KERN_DEBUG "Registration of pseries driver failed.\n");
 288                 return retval;
 289         }
 290
 291         register_cpu_notifier(&setup_hotplug_notifier);
 292         printk(KERN_DEBUG "pseries_idle_driver registered\n");
 293
 294         return 0;
 295 }
 296
 297 static void __exit pseries_processor_idle_exit(void)
 298 {
 299
 300         unregister_cpu_notifier(&setup_hotplug_notifier);
 301         cpuidle_unregister(&pseries_idle_driver);
 302
 303         return;
 304 }
 305
 306 module_init(pseries_processor_idle_init);
 307 module_exit(pseries_processor_idle_exit);
 308
 309 MODULE_AUTHOR("Deepthi Dharwar <deepthi@linux.vnet.ibm.com>");
 310 MODULE_DESCRIPTION("Cpuidle driver for POWER");
 311 MODULE_LICENSE("GPL");