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
  26 #include "power.h"
  27
  28 DEFINE_MUTEX(pm_mutex);
  29
  30 unsigned int pm_flags;
  31 EXPORT_SYMBOL(pm_flags);
  32
  33 #ifdef CONFIG_PM_SLEEP
  34
  35 /* Routines for PM-transition notifications */
  36
  37 static BLOCKING_NOTIFIER_HEAD(pm_chain_head);
  38
  39 int register_pm_notifier(struct notifier_block *nb)
  40 {
  41         return blocking_notifier_chain_register(&pm_chain_head, nb);
  42 }
  43 EXPORT_SYMBOL_GPL(register_pm_notifier);
  44
  45 int unregister_pm_notifier(struct notifier_block *nb)
  46 {
  47         return blocking_notifier_chain_unregister(&pm_chain_head, nb);
  48 }
  49 EXPORT_SYMBOL_GPL(unregister_pm_notifier);
  50
  51 int pm_notifier_call_chain(unsigned long val)
  52 {
  53         return (blocking_notifier_call_chain(&pm_chain_head, val, NULL)
  54                         == NOTIFY_BAD) ? -EINVAL : 0;
  55 }
  56
  57 #ifdef CONFIG_PM_DEBUG
  58 int pm_test_level = TEST_NONE;
  59
  60 static const char * const pm_tests[__TEST_AFTER_LAST] = {
  61         [TEST_NONE] = "none",
  62         [TEST_CORE] = "core",
  63         [TEST_CPUS] = "processors",
  64         [TEST_PLATFORM] = "platform",
  65         [TEST_DEVICES] = "devices",
  66         [TEST_FREEZER] = "freezer",
  67 };
  68
  69 static ssize_t pm_test_show(struct kobject *kobj, struct kobj_attribute *attr,
  70                                 char *buf)
  71 {
  72         char *s = buf;
  73         int level;
  74
  75         for (level = TEST_FIRST; level <= TEST_MAX; level++)
  76                 if (pm_tests[level]) {
  77                         if (level == pm_test_level)
  78                                 s += sprintf(s, "[%s] ", pm_tests[level]);
  79                         else
  80                                 s += sprintf(s, "%s ", pm_tests[level]);
  81                 }
  82
  83         if (s != buf)
  84                 /* convert the last space to a newline */
  85                 *(s-1) = '\n';
  86
  87         return (s - buf);
  88 }
  89
  90 static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
  91                                 const char *buf, size_t n)
  92 {
  93         const char * const *s;
  94         int level;
  95         char *p;
  96         int len;
  97         int error = -EINVAL;
  98
  99         p = memchr(buf, '\n', n);
 100         len = p ? p - buf : n;
 101
 102         mutex_lock(&pm_mutex);
 103
 104         level = TEST_FIRST;
 105         for (s = &pm_tests[level]; level <= TEST_MAX; s++, level++)
 106                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len)) {
 107                         pm_test_level = level;
 108                         error = 0;
 109                         break;
 110                 }
 111
 112         mutex_unlock(&pm_mutex);
 113
 114         return error ? error : n;
 115 }
 116
 117 power_attr(pm_test);
 118 #endif /* CONFIG_PM_DEBUG */
 119
 120 #endif /* CONFIG_PM_SLEEP */
 121
 122 #ifdef CONFIG_SUSPEND
 123
 124 static int suspend_test(int level)
 125 {
 126 #ifdef CONFIG_PM_DEBUG
 127         if (pm_test_level == level) {
 128                 printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n");
 129                 mdelay(5000);
 130                 return 1;
 131         }
 132 #endif /* !CONFIG_PM_DEBUG */
 133         return 0;
 134 }
 135
 136 #ifdef CONFIG_PM_TEST_SUSPEND
 137
 138 /*
 139  * We test the system suspend code by setting an RTC wakealarm a short
 140  * time in the future, then suspending.  Suspending the devices won't
 141  * normally take long ... some systems only need a few milliseconds.
 142  *
 143  * The time it takes is system-specific though, so when we test this
 144  * during system bootup we allow a LOT of time.
 145  */
 146 #define TEST_SUSPEND_SECONDS    5
 147
 148 static unsigned long suspend_test_start_time;
 149
 150 static void suspend_test_start(void)
 151 {
 152         /* FIXME Use better timebase than "jiffies", ideally a clocksource.
 153          * What we want is a hardware counter that will work correctly even
 154          * during the irqs-are-off stages of the suspend/resume cycle...
 155          */
 156         suspend_test_start_time = jiffies;
 157 }
 158
 159 static void suspend_test_finish(const char *label)
 160 {
 161         long nj = jiffies - suspend_test_start_time;
 162         unsigned msec;
 163
 164         msec = jiffies_to_msecs(abs(nj));
 165         pr_info("PM: %s took %d.%03d seconds\n", label,
 166                         msec / 1000, msec % 1000);
 167
 168         /* Warning on suspend means the RTC alarm period needs to be
 169          * larger -- the system was sooo slooowwww to suspend that the
 170          * alarm (should have) fired before the system went to sleep!
 171          *
 172          * Warning on either suspend or resume also means the system
 173          * has some performance issues.  The stack dump of a WARN_ON
 174          * is more likely to get the right attention than a printk...
 175          */
 176         WARN(msec > (TEST_SUSPEND_SECONDS * 1000), "Component: %s\n", label);
 177 }
 178
 179 #else
 180
 181 static void suspend_test_start(void)
 182 {
 183 }
 184
 185 static void suspend_test_finish(const char *label)
 186 {
 187 }
 188
 189 #endif
 190
 191 /* This is just an arbitrary number */
 192 #define FREE_PAGE_NUMBER (100)
 193
 194 static struct platform_suspend_ops *suspend_ops;
 195
 196 /**
 197  *      suspend_set_ops - Set the global suspend method table.
 198  *      @ops:   Pointer to ops structure.
 199  */
 200
 201 void suspend_set_ops(struct platform_suspend_ops *ops)
 202 {
 203         mutex_lock(&pm_mutex);
 204         suspend_ops = ops;
 205         mutex_unlock(&pm_mutex);
 206 }
 207
 208 /**
 209  * suspend_valid_only_mem - generic memory-only valid callback
 210  *
 211  * Platform drivers that implement mem suspend only and only need
 212  * to check for that in their .valid callback can use this instead
 213  * of rolling their own .valid callback.
 214  */
 215 int suspend_valid_only_mem(suspend_state_t state)
 216 {
 217         return state == PM_SUSPEND_MEM;
 218 }
 219
 220 /**
 221  *      suspend_prepare - Do prep work before entering low-power state.
 222  *
 223  *      This is common code that is called for each state that we're entering.
 224  *      Run suspend notifiers, allocate a console and stop all processes.
 225  */
 226 static int suspend_prepare(void)
 227 {
 228         int error;
 229         unsigned int free_pages;
 230
 231         if (!suspend_ops || !suspend_ops->enter)
 232                 return -EPERM;
 233
 234         pm_prepare_console();
 235
 236         error = pm_notifier_call_chain(PM_SUSPEND_PREPARE);
 237         if (error)
 238                 goto Finish;
 239
 240         error = usermodehelper_disable();
 241         if (error)
 242                 goto Finish;
 243
 244         if (suspend_freeze_processes()) {
 245                 error = -EAGAIN;
 246                 goto Thaw;
 247         }
 248
 249         free_pages = global_page_state(NR_FREE_PAGES);
 250         if (free_pages < FREE_PAGE_NUMBER) {
 251                 pr_debug("PM: free some memory\n");
 252                 shrink_all_memory(FREE_PAGE_NUMBER - free_pages);
 253                 if (nr_free_pages() < FREE_PAGE_NUMBER) {
 254                         error = -ENOMEM;
 255                         printk(KERN_ERR "PM: No enough memory\n");
 256                 }
 257         }
 258         if (!error)
 259                 return 0;
 260
 261  Thaw:
 262         suspend_thaw_processes();
 263         usermodehelper_enable();
 264  Finish:
 265         pm_notifier_call_chain(PM_POST_SUSPEND);
 266         pm_restore_console();
 267         return error;
 268 }
 269
 270 /* default implementation */
 271 void __attribute__ ((weak)) arch_suspend_disable_irqs(void)
 272 {
 273         local_irq_disable();
 274 }
 275
 276 /* default implementation */
 277 void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
 278 {
 279         local_irq_enable();
 280 }
 281
 282 /**
 283  *      suspend_enter - enter the desired system sleep state.
 284  *      @state:         state to enter
 285  *
 286  *      This function should be called after devices have been suspended.
 287  */
 288 static int suspend_enter(suspend_state_t state)
 289 {
 290         int error;
 291
 292         device_pm_lock();
 293
 294         if (suspend_ops->prepare) {
 295                 error = suspend_ops->prepare();
 296                 if (error)
 297                         goto Done;
 298         }
 299
 300         error = device_power_down(PMSG_SUSPEND);
 301         if (error) {
 302                 printk(KERN_ERR "PM: Some devices failed to power down\n");
 303                 goto Platfrom_finish;
 304         }
 305
 306         if (suspend_ops->prepare_late) {
 307                 error = suspend_ops->prepare_late();
 308                 if (error)
 309                         goto Power_up_devices;
 310         }
 311
 312         if (suspend_test(TEST_PLATFORM))
 313                 goto Platform_wake;
 314
 315         error = disable_nonboot_cpus();
 316         if (error || suspend_test(TEST_CPUS))
 317                 goto Enable_cpus;
 318
 319         arch_suspend_disable_irqs();
 320         BUG_ON(!irqs_disabled());
 321
 322         error = sysdev_suspend(PMSG_SUSPEND);
 323         if (!error) {
 324                 if (!suspend_test(TEST_CORE))
 325                         error = suspend_ops->enter(state);
 326                 sysdev_resume();
 327         }
 328
 329         arch_suspend_enable_irqs();
 330         BUG_ON(irqs_disabled());
 331
 332  Enable_cpus:
 333         enable_nonboot_cpus();
 334
 335  Platform_wake:
 336         if (suspend_ops->wake)
 337                 suspend_ops->wake();
 338
 339  Power_up_devices:
 340         device_power_up(PMSG_RESUME);
 341
 342  Platfrom_finish:
 343         if (suspend_ops->finish)
 344                 suspend_ops->finish();
 345
 346  Done:
 347         device_pm_unlock();
 348
 349         return error;
 350 }
 351
 352 /**
 353  *      suspend_devices_and_enter - suspend devices and enter the desired system
 354  *                                  sleep state.
 355  *      @state:           state to enter
 356  */
 357 int suspend_devices_and_enter(suspend_state_t state)
 358 {
 359         int error;
 360
 361         if (!suspend_ops)
 362                 return -ENOSYS;
 363
 364         if (suspend_ops->begin) {
 365                 error = suspend_ops->begin(state);
 366                 if (error)
 367                         goto Close;
 368         }
 369         suspend_console();
 370         suspend_test_start();
 371         error = device_suspend(PMSG_SUSPEND);
 372         if (error) {
 373                 printk(KERN_ERR "PM: Some devices failed to suspend\n");
 374                 goto Recover_platform;
 375         }
 376         suspend_test_finish("suspend devices");
 377         if (suspend_test(TEST_DEVICES))
 378                 goto Recover_platform;
 379
 380         suspend_enter(state);
 381
 382  Resume_devices:
 383         suspend_test_start();
 384         device_resume(PMSG_RESUME);
 385         suspend_test_finish("resume devices");
 386         resume_console();
 387  Close:
 388         if (suspend_ops->end)
 389                 suspend_ops->end();
 390         return error;
 391
 392  Recover_platform:
 393         if (suspend_ops->recover)
 394                 suspend_ops->recover();
 395         goto Resume_devices;
 396 }
 397
 398 /**
 399  *      suspend_finish - Do final work before exiting suspend sequence.
 400  *
 401  *      Call platform code to clean up, restart processes, and free the
 402  *      console that we've allocated. This is not called for suspend-to-disk.
 403  */
 404 static void suspend_finish(void)
 405 {
 406         suspend_thaw_processes();
 407         usermodehelper_enable();
 408         pm_notifier_call_chain(PM_POST_SUSPEND);
 409         pm_restore_console();
 410 }
 411
 412
 413
 414
 415 static const char * const pm_states[PM_SUSPEND_MAX] = {
 416         [PM_SUSPEND_STANDBY]    = "standby",
 417         [PM_SUSPEND_MEM]        = "mem",
 418 };
 419
 420 static inline int valid_state(suspend_state_t state)
 421 {
 422         /* All states need lowlevel support and need to be valid
 423          * to the lowlevel implementation, no valid callback
 424          * implies that none are valid. */
 425         if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
 426                 return 0;
 427         return 1;
 428 }
 429
 430
 431 /**
 432  *      enter_state - Do common work of entering low-power state.
 433  *      @state:         pm_state structure for state we're entering.
 434  *
 435  *      Make sure we're the only ones trying to enter a sleep state. Fail
 436  *      if someone has beat us to it, since we don't want anything weird to
 437  *      happen when we wake up.
 438  *      Then, do the setup for suspend, enter the state, and cleaup (after
 439  *      we've woken up).
 440  */
 441 static int enter_state(suspend_state_t state)
 442 {
 443         int error;
 444
 445         if (!valid_state(state))
 446                 return -ENODEV;
 447
 448         if (!mutex_trylock(&pm_mutex))
 449                 return -EBUSY;
 450
 451         printk(KERN_INFO "PM: Syncing filesystems ... ");
 452         sys_sync();
 453         printk("done.\n");
 454
 455         pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
 456         error = suspend_prepare();
 457         if (error)
 458                 goto Unlock;
 459
 460         if (suspend_test(TEST_FREEZER))
 461                 goto Finish;
 462
 463         pr_debug("PM: Entering %s sleep\n", pm_states[state]);
 464         error = suspend_devices_and_enter(state);
 465
 466  Finish:
 467         pr_debug("PM: Finishing wakeup.\n");
 468         suspend_finish();
 469  Unlock:
 470         mutex_unlock(&pm_mutex);
 471         return error;
 472 }
 473
 474
 475 /**
 476  *      pm_suspend - Externally visible function for suspending system.
 477  *      @state:         Enumerated value of state to enter.
 478  *
 479  *      Determine whether or not value is within range, get state
 480  *      structure, and enter (above).
 481  */
 482
 483 int pm_suspend(suspend_state_t state)
 484 {
 485         if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
 486                 return enter_state(state);
 487         return -EINVAL;
 488 }
 489
 490 EXPORT_SYMBOL(pm_suspend);
 491
 492 #endif /* CONFIG_SUSPEND */
 493
 494 struct kobject *power_kobj;
 495
 496 /**
 497  *      state - control system power state.
 498  *
 499  *      show() returns what states are supported, which is hard-coded to
 500  *      'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
 501  *      'disk' (Suspend-to-Disk).
 502  *
 503  *      store() accepts one of those strings, translates it into the
 504  *      proper enumerated value, and initiates a suspend transition.
 505  */
 506
 507 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
 508                           char *buf)
 509 {
 510         char *s = buf;
 511 #ifdef CONFIG_SUSPEND
 512         int i;
 513
 514         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 515                 if (pm_states[i] && valid_state(i))
 516                         s += sprintf(s,"%s ", pm_states[i]);
 517         }
 518 #endif
 519 #ifdef CONFIG_HIBERNATION
 520         s += sprintf(s, "%s\n", "disk");
 521 #else
 522         if (s != buf)
 523                 /* convert the last space to a newline */
 524                 *(s-1) = '\n';
 525 #endif
 526         return (s - buf);
 527 }
 528
 529 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
 530                            const char *buf, size_t n)
 531 {
 532 #ifdef CONFIG_SUSPEND
 533         suspend_state_t state = PM_SUSPEND_STANDBY;
 534         const char * const *s;
 535 #endif
 536         char *p;
 537         int len;
 538         int error = -EINVAL;
 539
 540         p = memchr(buf, '\n', n);
 541         len = p ? p - buf : n;
 542
 543         /* First, check if we are requested to hibernate */
 544         if (len == 4 && !strncmp(buf, "disk", len)) {
 545                 error = hibernate();
 546   goto Exit;
 547         }
 548
 549 #ifdef CONFIG_SUSPEND
 550         for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
 551                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
 552                         break;
 553         }
 554         if (state < PM_SUSPEND_MAX && *s)
 555                 error = enter_state(state);
 556 #endif
 557
 558  Exit:
 559         return error ? error : n;
 560 }
 561
 562 power_attr(state);
 563
 564 #ifdef CONFIG_PM_TRACE
 565 int pm_trace_enabled;
 566
 567 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
 568                              char *buf)
 569 {
 570         return sprintf(buf, "%d\n", pm_trace_enabled);
 571 }
 572
 573 static ssize_t
 574 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
 575                const char *buf, size_t n)
 576 {
 577         int val;
 578
 579         if (sscanf(buf, "%d", &val) == 1) {
 580                 pm_trace_enabled = !!val;
 581                 return n;
 582         }
 583         return -EINVAL;
 584 }
 585
 586 power_attr(pm_trace);
 587 #endif /* CONFIG_PM_TRACE */
 588
 589 static struct attribute * g[] = {
 590         &state_attr.attr,
 591 #ifdef CONFIG_PM_TRACE
 592         &pm_trace_attr.attr,
 593 #endif
 594 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
 595         &pm_test_attr.attr,
 596 #endif
 597         NULL,
 598 };
 599
 600 static struct attribute_group attr_group = {
 601         .attrs = g,
 602 };
 603
 604
 605 static int __init pm_init(void)
 606 {
 607         power_kobj = kobject_create_and_add("power", NULL);
 608         if (!power_kobj)
 609                 return -ENOMEM;
 610         return sysfs_create_group(power_kobj, &attr_group);
 611 }
 612
 613 core_initcall(pm_init);
 614
 615
 616 #ifdef CONFIG_PM_TEST_SUSPEND
 617
 618 #include <linux/rtc.h>
 619
 620 /*
 621  * To test system suspend, we need a hands-off mechanism to resume the
 622  * system.  RTCs wake alarms are a common self-contained mechanism.
 623  */
 624
 625 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
 626 {
 627         static char err_readtime[] __initdata =
 628                 KERN_ERR "PM: can't read %s time, err %d\n";
 629         static char err_wakealarm [] __initdata =
 630                 KERN_ERR "PM: can't set %s wakealarm, err %d\n";
 631         static char err_suspend[] __initdata =
 632                 KERN_ERR "PM: suspend test failed, error %d\n";
 633         static char info_test[] __initdata =
 634                 KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
 635
 636         unsigned long           now;
 637         struct rtc_wkalrm       alm;
 638         int                     status;
 639
 640         /* this may fail if the RTC hasn't been initialized */
 641         status = rtc_read_time(rtc, &alm.time);
 642         if (status < 0) {
 643                 printk(err_readtime, dev_name(&rtc->dev), status);
 644                 return;
 645         }
 646         rtc_tm_to_time(&alm.time, &now);
 647
 648         memset(&alm, 0, sizeof alm);
 649         rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
 650         alm.enabled = true;
 651
 652         status = rtc_set_alarm(rtc, &alm);
 653         if (status < 0) {
 654                 printk(err_wakealarm, dev_name(&rtc->dev), status);
 655                 return;
 656         }
 657
 658         if (state == PM_SUSPEND_MEM) {
 659                 printk(info_test, pm_states[state]);
 660                 status = pm_suspend(state);
 661                 if (status == -ENODEV)
 662                         state = PM_SUSPEND_STANDBY;
 663         }
 664         if (state == PM_SUSPEND_STANDBY) {
 665                 printk(info_test, pm_states[state]);
 666                 status = pm_suspend(state);
 667         }
 668         if (status < 0)
 669                 printk(err_suspend, status);
 670
 671         /* Some platforms can't detect that the alarm triggered the
 672          * wakeup, or (accordingly) disable it after it afterwards.
 673          * It's supposed to give oneshot behavior; cope.
 674          */
 675         alm.enabled = false;
 676         rtc_set_alarm(rtc, &alm);
 677 }
 678
 679 static int __init has_wakealarm(struct device *dev, void *name_ptr)
 680 {
 681         struct rtc_device *candidate = to_rtc_device(dev);
 682
 683         if (!candidate->ops->set_alarm)
 684                 return 0;
 685         if (!device_may_wakeup(candidate->dev.parent))
 686                 return 0;
 687
 688         *(const char **)name_ptr = dev_name(dev);
 689         return 1;
 690 }
 691
 692 /*
 693  * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
 694  * at startup time.  They're normally disabled, for faster boot and because
 695  * we can't know which states really work on this particular system.
 696  */
 697 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
 698
 699 static char warn_bad_state[] __initdata =
 700         KERN_WARNING "PM: can't test '%s' suspend state\n";
 701
 702 static int __init setup_test_suspend(char *value)
 703 {
 704         unsigned i;
 705
 706         /* "=mem" ==> "mem" */
 707         value++;
 708         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 709                 if (!pm_states[i])
 710                         continue;
 711                 if (strcmp(pm_states[i], value) != 0)
 712                         continue;
 713                 test_state = (__force suspend_state_t) i;
 714                 return 0;
 715         }
 716         printk(warn_bad_state, value);
 717         return 0;
 718 }
 719 __setup("test_suspend", setup_test_suspend);
 720
 721 static int __init test_suspend(void)
 722 {
 723         static char             warn_no_rtc[] __initdata =
 724                 KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
 725
 726         char                    *pony = NULL;
 727         struct rtc_device       *rtc = NULL;
 728
 729         /* PM is initialized by now; is that state testable? */
 730         if (test_state == PM_SUSPEND_ON)
 731                 goto done;
 732         if (!valid_state(test_state)) {
 733                 printk(warn_bad_state, pm_states[test_state]);
 734                 goto done;
 735         }
 736
 737         /* RTCs have initialized by now too ... can we use one? */
 738         class_find_device(rtc_class, NULL, &pony, has_wakealarm);
 739         if (pony)
 740                 rtc = rtc_class_open(pony);
 741         if (!rtc) {
 742                 printk(warn_no_rtc);
 743                 goto done;
 744         }
 745
 746         /* go for it */
 747         test_wakealarm(rtc, test_state);
 748         rtc_class_close(rtc);
 749 done:
 750         return 0;
 751 }
 752 late_initcall(test_suspend);
 753
 754 #endif /* CONFIG_PM_TEST_SUSPEND */