kernel/power/main.c

   1 /*
   2  * kernel/power/main.c - PM subsystem core functionality.
   3  *
   4  * Copyright (c) 2003 Patrick Mochel
   5  * Copyright (c) 2003 Open Source Development Lab
   6  *
   7  * This file is released under the GPLv2
   8  *
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/suspend.h>
  13 #include <linux/kobject.h>
  14 #include <linux/string.h>
  15 #include <linux/delay.h>
  16 #include <linux/errno.h>
  17 #include <linux/kmod.h>
  18 #include <linux/init.h>
  19 #include <linux/console.h>
  20 #include <linux/cpu.h>
  21 #include <linux/resume-trace.h>
  22 #include <linux/freezer.h>
  23 #include <linux/vmstat.h>
  24 #include <linux/syscalls.h>
  25 #include <linux/ftrace.h>
  26
  27 #include "power.h"
  28
  29 DEFINE_MUTEX(pm_mutex);
  30
  31 unsigned int pm_flags;
  32 EXPORT_SYMBOL(pm_flags);
  33
  34 #ifdef CONFIG_PM_SLEEP
  35
  36 /* Routines for PM-transition notifications */
  37
  38 static BLOCKING_NOTIFIER_HEAD(pm_chain_head);
  39
  40 int register_pm_notifier(struct notifier_block *nb)
  41 {
  42         return blocking_notifier_chain_register(&pm_chain_head, nb);
  43 }
  44 EXPORT_SYMBOL_GPL(register_pm_notifier);
  45
  46 int unregister_pm_notifier(struct notifier_block *nb)
  47 {
  48         return blocking_notifier_chain_unregister(&pm_chain_head, nb);
  49 }
  50 EXPORT_SYMBOL_GPL(unregister_pm_notifier);
  51
  52 int pm_notifier_call_chain(unsigned long val)
  53 {
  54         return (blocking_notifier_call_chain(&pm_chain_head, val, NULL)
  55                         == NOTIFY_BAD) ? -EINVAL : 0;
  56 }
  57
  58 #ifdef CONFIG_PM_DEBUG
  59 int pm_test_level = TEST_NONE;
  60
  61 static int suspend_test(int level)
  62 {
  63         if (pm_test_level == level) {
  64                 printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n");
  65                 mdelay(5000);
  66                 return 1;
  67         }
  68         return 0;
  69 }
  70
  71 static const char * const pm_tests[__TEST_AFTER_LAST] = {
  72         [TEST_NONE] = "none",
  73         [TEST_CORE] = "core",
  74         [TEST_CPUS] = "processors",
  75         [TEST_PLATFORM] = "platform",
  76         [TEST_DEVICES] = "devices",
  77         [TEST_FREEZER] = "freezer",
  78 };
  79
  80 static ssize_t pm_test_show(struct kobject *kobj, struct kobj_attribute *attr,
  81                                 char *buf)
  82 {
  83         char *s = buf;
  84         int level;
  85
  86         for (level = TEST_FIRST; level <= TEST_MAX; level++)
  87                 if (pm_tests[level]) {
  88                         if (level == pm_test_level)
  89                                 s += sprintf(s, "[%s] ", pm_tests[level]);
  90                         else
  91                                 s += sprintf(s, "%s ", pm_tests[level]);
  92                 }
  93
  94         if (s != buf)
  95                 /* convert the last space to a newline */
  96                 *(s-1) = '\n';
  97
  98         return (s - buf);
  99 }
 100
 101 static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
 102                                 const char *buf, size_t n)
 103 {
 104         const char * const *s;
 105         int level;
 106         char *p;
 107         int len;
 108         int error = -EINVAL;
 109
 110         p = memchr(buf, '\n', n);
 111         len = p ? p - buf : n;
 112
 113         mutex_lock(&pm_mutex);
 114
 115         level = TEST_FIRST;
 116         for (s = &pm_tests[level]; level <= TEST_MAX; s++, level++)
 117                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len)) {
 118                         pm_test_level = level;
 119                         error = 0;
 120                         break;
 121                 }
 122
 123         mutex_unlock(&pm_mutex);
 124
 125         return error ? error : n;
 126 }
 127
 128 power_attr(pm_test);
 129 #else /* !CONFIG_PM_DEBUG */
 130 static inline int suspend_test(int level) { return 0; }
 131 #endif /* !CONFIG_PM_DEBUG */
 132
 133 #endif /* CONFIG_PM_SLEEP */
 134
 135 #ifdef CONFIG_SUSPEND
 136
 137 #ifdef CONFIG_PM_TEST_SUSPEND
 138
 139 /*
 140  * We test the system suspend code by setting an RTC wakealarm a short
 141  * time in the future, then suspending.  Suspending the devices won't
 142  * normally take long ... some systems only need a few milliseconds.
 143  *
 144  * The time it takes is system-specific though, so when we test this
 145  * during system bootup we allow a LOT of time.
 146  */
 147 #define TEST_SUSPEND_SECONDS    5
 148
 149 static unsigned long suspend_test_start_time;
 150
 151 static void suspend_test_start(void)
 152 {
 153         /* FIXME Use better timebase than "jiffies", ideally a clocksource.
 154          * What we want is a hardware counter that will work correctly even
 155          * during the irqs-are-off stages of the suspend/resume cycle...
 156          */
 157         suspend_test_start_time = jiffies;
 158 }
 159
 160 static void suspend_test_finish(const char *label)
 161 {
 162         long nj = jiffies - suspend_test_start_time;
 163         unsigned msec;
 164
 165         msec = jiffies_to_msecs(abs(nj));
 166         pr_info("PM: %s took %d.%03d seconds\n", label,
 167                         msec / 1000, msec % 1000);
 168
 169         /* Warning on suspend means the RTC alarm period needs to be
 170          * larger -- the system was sooo slooowwww to suspend that the
 171          * alarm (should have) fired before the system went to sleep!
 172          *
 173          * Warning on either suspend or resume also means the system
 174          * has some performance issues.  The stack dump of a WARN_ON
 175          * is more likely to get the right attention than a printk...
 176          */
 177         WARN_ON(msec > (TEST_SUSPEND_SECONDS * 1000));
 178 }
 179
 180 #else
 181
 182 static void suspend_test_start(void)
 183 {
 184 }
 185
 186 static void suspend_test_finish(const char *label)
 187 {
 188 }
 189
 190 #endif
 191
 192 /* This is just an arbitrary number */
 193 #define FREE_PAGE_NUMBER (100)
 194
 195 static struct platform_suspend_ops *suspend_ops;
 196
 197 /**
 198  *      suspend_set_ops - Set the global suspend method table.
 199  *      @ops:   Pointer to ops structure.
 200  */
 201
 202 void suspend_set_ops(struct platform_suspend_ops *ops)
 203 {
 204         mutex_lock(&pm_mutex);
 205         suspend_ops = ops;
 206         mutex_unlock(&pm_mutex);
 207 }
 208
 209 /**
 210  * suspend_valid_only_mem - generic memory-only valid callback
 211  *
 212  * Platform drivers that implement mem suspend only and only need
 213  * to check for that in their .valid callback can use this instead
 214  * of rolling their own .valid callback.
 215  */
 216 int suspend_valid_only_mem(suspend_state_t state)
 217 {
 218         return state == PM_SUSPEND_MEM;
 219 }
 220
 221 /**
 222  *      suspend_prepare - Do prep work before entering low-power state.
 223  *
 224  *      This is common code that is called for each state that we're entering.
 225  *      Run suspend notifiers, allocate a console and stop all processes.
 226  */
 227 static int suspend_prepare(void)
 228 {
 229         int error;
 230         unsigned int free_pages;
 231
 232         if (!suspend_ops || !suspend_ops->enter)
 233                 return -EPERM;
 234
 235         pm_prepare_console();
 236
 237         error = pm_notifier_call_chain(PM_SUSPEND_PREPARE);
 238         if (error)
 239                 goto Finish;
 240
 241         error = usermodehelper_disable();
 242         if (error)
 243                 goto Finish;
 244
 245         if (suspend_freeze_processes()) {
 246                 error = -EAGAIN;
 247                 goto Thaw;
 248         }
 249
 250         free_pages = global_page_state(NR_FREE_PAGES);
 251         if (free_pages < FREE_PAGE_NUMBER) {
 252                 pr_debug("PM: free some memory\n");
 253                 shrink_all_memory(FREE_PAGE_NUMBER - free_pages);
 254                 if (nr_free_pages() < FREE_PAGE_NUMBER) {
 255                         error = -ENOMEM;
 256                         printk(KERN_ERR "PM: No enough memory\n");
 257                 }
 258         }
 259         if (!error)
 260                 return 0;
 261
 262  Thaw:
 263         suspend_thaw_processes();
 264         usermodehelper_enable();
 265  Finish:
 266         pm_notifier_call_chain(PM_POST_SUSPEND);
 267         pm_restore_console();
 268         return error;
 269 }
 270
 271 /* default implementation */
 272 void __attribute__ ((weak)) arch_suspend_disable_irqs(void)
 273 {
 274         local_irq_disable();
 275 }
 276
 277 /* default implementation */
 278 void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
 279 {
 280         local_irq_enable();
 281 }
 282
 283 /**
 284  *      suspend_enter - enter the desired system sleep state.
 285  *      @state:         state to enter
 286  *
 287  *      This function should be called after devices have been suspended.
 288  */
 289 static int suspend_enter(suspend_state_t state)
 290 {
 291         int error = 0;
 292
 293         device_pm_lock();
 294         arch_suspend_disable_irqs();
 295         BUG_ON(!irqs_disabled());
 296
 297         if ((error = device_power_down(PMSG_SUSPEND))) {
 298                 printk(KERN_ERR "PM: Some devices failed to power down\n");
 299                 goto Done;
 300         }
 301
 302         if (!suspend_test(TEST_CORE))
 303                 error = suspend_ops->enter(state);
 304
 305         device_power_up(PMSG_RESUME);
 306  Done:
 307         arch_suspend_enable_irqs();
 308         BUG_ON(irqs_disabled());
 309         device_pm_unlock();
 310         return error;
 311 }
 312
 313 /**
 314  *      suspend_devices_and_enter - suspend devices and enter the desired system
 315  *                                  sleep state.
 316  *      @state:           state to enter
 317  */
 318 int suspend_devices_and_enter(suspend_state_t state)
 319 {
 320         int error, ftrace_save;
 321
 322         if (!suspend_ops)
 323                 return -ENOSYS;
 324
 325         if (suspend_ops->begin) {
 326                 error = suspend_ops->begin(state);
 327                 if (error)
 328                         goto Close;
 329         }
 330         suspend_console();
 331         ftrace_save = __ftrace_enabled_save();
 332         suspend_test_start();
 333         error = device_suspend(PMSG_SUSPEND);
 334         if (error) {
 335                 printk(KERN_ERR "PM: Some devices failed to suspend\n");
 336                 goto Recover_platform;
 337         }
 338         suspend_test_finish("suspend devices");
 339         if (suspend_test(TEST_DEVICES))
 340                 goto Recover_platform;
 341
 342         if (suspend_ops->prepare) {
 343                 error = suspend_ops->prepare();
 344                 if (error)
 345                         goto Resume_devices;
 346         }
 347
 348         if (suspend_test(TEST_PLATFORM))
 349                 goto Finish;
 350
 351         error = disable_nonboot_cpus();
 352         if (!error && !suspend_test(TEST_CPUS))
 353                 suspend_enter(state);
 354
 355         enable_nonboot_cpus();
 356  Finish:
 357         if (suspend_ops->finish)
 358                 suspend_ops->finish();
 359  Resume_devices:
 360         suspend_test_start();
 361         device_resume(PMSG_RESUME);
 362         suspend_test_finish("resume devices");
 363         __ftrace_enabled_restore(ftrace_save);
 364         resume_console();
 365  Close:
 366         if (suspend_ops->end)
 367                 suspend_ops->end();
 368         return error;
 369
 370  Recover_platform:
 371         if (suspend_ops->recover)
 372                 suspend_ops->recover();
 373         goto Resume_devices;
 374 }
 375
 376 /**
 377  *      suspend_finish - Do final work before exiting suspend sequence.
 378  *
 379  *      Call platform code to clean up, restart processes, and free the
 380  *      console that we've allocated. This is not called for suspend-to-disk.
 381  */
 382 static void suspend_finish(void)
 383 {
 384         suspend_thaw_processes();
 385         usermodehelper_enable();
 386         pm_notifier_call_chain(PM_POST_SUSPEND);
 387         pm_restore_console();
 388 }
 389
 390
 391
 392
 393 static const char * const pm_states[PM_SUSPEND_MAX] = {
 394         [PM_SUSPEND_STANDBY]    = "standby",
 395         [PM_SUSPEND_MEM]        = "mem",
 396 };
 397
 398 static inline int valid_state(suspend_state_t state)
 399 {
 400         /* All states need lowlevel support and need to be valid
 401          * to the lowlevel implementation, no valid callback
 402          * implies that none are valid. */
 403         if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
 404                 return 0;
 405         return 1;
 406 }
 407
 408
 409 /**
 410  *      enter_state - Do common work of entering low-power state.
 411  *      @state:         pm_state structure for state we're entering.
 412  *
 413  *      Make sure we're the only ones trying to enter a sleep state. Fail
 414  *      if someone has beat us to it, since we don't want anything weird to
 415  *      happen when we wake up.
 416  *      Then, do the setup for suspend, enter the state, and cleaup (after
 417  *      we've woken up).
 418  */
 419 static int enter_state(suspend_state_t state)
 420 {
 421         int error;
 422
 423         if (!valid_state(state))
 424                 return -ENODEV;
 425
 426         if (!mutex_trylock(&pm_mutex))
 427                 return -EBUSY;
 428
 429         printk(KERN_INFO "PM: Syncing filesystems ... ");
 430         sys_sync();
 431         printk("done.\n");
 432
 433         pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
 434         error = suspend_prepare();
 435         if (error)
 436                 goto Unlock;
 437
 438         if (suspend_test(TEST_FREEZER))
 439                 goto Finish;
 440
 441         pr_debug("PM: Entering %s sleep\n", pm_states[state]);
 442         error = suspend_devices_and_enter(state);
 443
 444  Finish:
 445         pr_debug("PM: Finishing wakeup.\n");
 446         suspend_finish();
 447  Unlock:
 448         mutex_unlock(&pm_mutex);
 449         return error;
 450 }
 451
 452
 453 /**
 454  *      pm_suspend - Externally visible function for suspending system.
 455  *      @state:         Enumerated value of state to enter.
 456  *
 457  *      Determine whether or not value is within range, get state
 458  *      structure, and enter (above).
 459  */
 460
 461 int pm_suspend(suspend_state_t state)
 462 {
 463         if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
 464                 return enter_state(state);
 465         return -EINVAL;
 466 }
 467
 468 EXPORT_SYMBOL(pm_suspend);
 469
 470 #endif /* CONFIG_SUSPEND */
 471
 472 struct kobject *power_kobj;
 473
 474 /**
 475  *      state - control system power state.
 476  *
 477  *      show() returns what states are supported, which is hard-coded to
 478  *      'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
 479  *      'disk' (Suspend-to-Disk).
 480  *
 481  *      store() accepts one of those strings, translates it into the
 482  *      proper enumerated value, and initiates a suspend transition.
 483  */
 484
 485 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
 486                           char *buf)
 487 {
 488         char *s = buf;
 489 #ifdef CONFIG_SUSPEND
 490         int i;
 491
 492         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 493                 if (pm_states[i] && valid_state(i))
 494                         s += sprintf(s,"%s ", pm_states[i]);
 495         }
 496 #endif
 497 #ifdef CONFIG_HIBERNATION
 498         s += sprintf(s, "%s\n", "disk");
 499 #else
 500         if (s != buf)
 501                 /* convert the last space to a newline */
 502                 *(s-1) = '\n';
 503 #endif
 504         return (s - buf);
 505 }
 506
 507 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
 508                            const char *buf, size_t n)
 509 {
 510 #ifdef CONFIG_SUSPEND
 511         suspend_state_t state = PM_SUSPEND_STANDBY;
 512         const char * const *s;
 513 #endif
 514         char *p;
 515         int len;
 516         int error = -EINVAL;
 517
 518         p = memchr(buf, '\n', n);
 519         len = p ? p - buf : n;
 520
 521         /* First, check if we are requested to hibernate */
 522         if (len == 4 && !strncmp(buf, "disk", len)) {
 523                 error = hibernate();
 524   goto Exit;
 525         }
 526
 527 #ifdef CONFIG_SUSPEND
 528         for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
 529                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
 530                         break;
 531         }
 532         if (state < PM_SUSPEND_MAX && *s)
 533                 error = enter_state(state);
 534 #endif
 535
 536  Exit:
 537         return error ? error : n;
 538 }
 539
 540 power_attr(state);
 541
 542 #ifdef CONFIG_PM_TRACE
 543 int pm_trace_enabled;
 544
 545 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
 546                              char *buf)
 547 {
 548         return sprintf(buf, "%d\n", pm_trace_enabled);
 549 }
 550
 551 static ssize_t
 552 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
 553                const char *buf, size_t n)
 554 {
 555         int val;
 556
 557         if (sscanf(buf, "%d", &val) == 1) {
 558                 pm_trace_enabled = !!val;
 559                 return n;
 560         }
 561         return -EINVAL;
 562 }
 563
 564 power_attr(pm_trace);
 565 #endif /* CONFIG_PM_TRACE */
 566
 567 static struct attribute * g[] = {
 568         &state_attr.attr,
 569 #ifdef CONFIG_PM_TRACE
 570         &pm_trace_attr.attr,
 571 #endif
 572 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
 573         &pm_test_attr.attr,
 574 #endif
 575         NULL,
 576 };
 577
 578 static struct attribute_group attr_group = {
 579         .attrs = g,
 580 };
 581
 582
 583 static int __init pm_init(void)
 584 {
 585         power_kobj = kobject_create_and_add("power", NULL);
 586         if (!power_kobj)
 587                 return -ENOMEM;
 588         return sysfs_create_group(power_kobj, &attr_group);
 589 }
 590
 591 core_initcall(pm_init);
 592
 593
 594 #ifdef CONFIG_PM_TEST_SUSPEND
 595
 596 #include <linux/rtc.h>
 597
 598 /*
 599  * To test system suspend, we need a hands-off mechanism to resume the
 600  * system.  RTCs wake alarms are a common self-contained mechanism.
 601  */
 602
 603 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
 604 {
 605         static char err_readtime[] __initdata =
 606                 KERN_ERR "PM: can't read %s time, err %d\n";
 607         static char err_wakealarm [] __initdata =
 608                 KERN_ERR "PM: can't set %s wakealarm, err %d\n";
 609         static char err_suspend[] __initdata =
 610                 KERN_ERR "PM: suspend test failed, error %d\n";
 611         static char info_test[] __initdata =
 612                 KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
 613
 614         unsigned long           now;
 615         struct rtc_wkalrm       alm;
 616         int                     status;
 617
 618         /* this may fail if the RTC hasn't been initialized */
 619         status = rtc_read_time(rtc, &alm.time);
 620         if (status < 0) {
 621                 printk(err_readtime, rtc->dev.bus_id, status);
 622                 return;
 623         }
 624         rtc_tm_to_time(&alm.time, &now);
 625
 626         memset(&alm, 0, sizeof alm);
 627         rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
 628         alm.enabled = true;
 629
 630         status = rtc_set_alarm(rtc, &alm);
 631         if (status < 0) {
 632                 printk(err_wakealarm, rtc->dev.bus_id, status);
 633                 return;
 634         }
 635
 636         if (state == PM_SUSPEND_MEM) {
 637                 printk(info_test, pm_states[state]);
 638                 status = pm_suspend(state);
 639                 if (status == -ENODEV)
 640                         state = PM_SUSPEND_STANDBY;
 641         }
 642         if (state == PM_SUSPEND_STANDBY) {
 643                 printk(info_test, pm_states[state]);
 644                 status = pm_suspend(state);
 645         }
 646         if (status < 0)
 647                 printk(err_suspend, status);
 648
 649         /* Some platforms can't detect that the alarm triggered the
 650          * wakeup, or (accordingly) disable it after it afterwards.
 651          * It's supposed to give oneshot behavior; cope.
 652          */
 653         alm.enabled = false;
 654         rtc_set_alarm(rtc, &alm);
 655 }
 656
 657 static int __init has_wakealarm(struct device *dev, void *name_ptr)
 658 {
 659         struct rtc_device *candidate = to_rtc_device(dev);
 660
 661         if (!candidate->ops->set_alarm)
 662                 return 0;
 663         if (!device_may_wakeup(candidate->dev.parent))
 664                 return 0;
 665
 666         *(char **)name_ptr = dev->bus_id;
 667         return 1;
 668 }
 669
 670 /*
 671  * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
 672  * at startup time.  They're normally disabled, for faster boot and because
 673  * we can't know which states really work on this particular system.
 674  */
 675 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
 676
 677 static char warn_bad_state[] __initdata =
 678         KERN_WARNING "PM: can't test '%s' suspend state\n";
 679
 680 static int __init setup_test_suspend(char *value)
 681 {
 682         unsigned i;
 683
 684         /* "=mem" ==> "mem" */
 685         value++;
 686         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 687                 if (!pm_states[i])
 688                         continue;
 689                 if (strcmp(pm_states[i], value) != 0)
 690                         continue;
 691                 test_state = (__force suspend_state_t) i;
 692                 return 0;
 693         }
 694         printk(warn_bad_state, value);
 695         return 0;
 696 }
 697 __setup("test_suspend", setup_test_suspend);
 698
 699 static int __init test_suspend(void)
 700 {
 701         static char             warn_no_rtc[] __initdata =
 702                 KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
 703
 704         char                    *pony = NULL;
 705         struct rtc_device       *rtc = NULL;
 706
 707         /* PM is initialized by now; is that state testable? */
 708         if (test_state == PM_SUSPEND_ON)
 709                 goto done;
 710         if (!valid_state(test_state)) {
 711                 printk(warn_bad_state, pm_states[test_state]);
 712                 goto done;
 713         }
 714
 715         /* RTCs have initialized by now too ... can we use one? */
 716         class_find_device(rtc_class, NULL, &pony, has_wakealarm);
 717         if (pony)
 718                 rtc = rtc_class_open(pony);
 719         if (!rtc) {
 720                 printk(warn_no_rtc);
 721                 goto done;
 722         }
 723
 724         /* go for it */
 725         test_wakealarm(rtc, test_state);
 726         rtc_class_close(rtc);
 727 done:
 728         return 0;
 729 }
 730 late_initcall(test_suspend);
 731
 732 #endif /* CONFIG_PM_TEST_SUSPEND */