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         error = device_power_down(PMSG_SUSPEND);
 295         if (error) {
 296                 printk(KERN_ERR "PM: Some devices failed to power down\n");
 297                 goto Done;
 298         }
 299
 300         if (suspend_ops->prepare) {
 301                 error = suspend_ops->prepare();
 302                 if (error)
 303                         goto Power_up_devices;
 304         }
 305
 306         if (suspend_test(TEST_PLATFORM))
 307                 goto Platfrom_finish;
 308
 309         error = disable_nonboot_cpus();
 310         if (error || suspend_test(TEST_CPUS))
 311                 goto Enable_cpus;
 312
 313         arch_suspend_disable_irqs();
 314         BUG_ON(!irqs_disabled());
 315
 316         error = sysdev_suspend(PMSG_SUSPEND);
 317         if (!error) {
 318                 if (!suspend_test(TEST_CORE))
 319                         error = suspend_ops->enter(state);
 320                 sysdev_resume();
 321         }
 322
 323         arch_suspend_enable_irqs();
 324         BUG_ON(irqs_disabled());
 325
 326  Enable_cpus:
 327         enable_nonboot_cpus();
 328
 329  Platfrom_finish:
 330         if (suspend_ops->finish)
 331                 suspend_ops->finish();
 332
 333  Power_up_devices:
 334         device_power_up(PMSG_RESUME);
 335
 336  Done:
 337         device_pm_unlock();
 338
 339         return error;
 340 }
 341
 342 /**
 343  *      suspend_devices_and_enter - suspend devices and enter the desired system
 344  *                                  sleep state.
 345  *      @state:           state to enter
 346  */
 347 int suspend_devices_and_enter(suspend_state_t state)
 348 {
 349         int error;
 350
 351         if (!suspend_ops)
 352                 return -ENOSYS;
 353
 354         if (suspend_ops->begin) {
 355                 error = suspend_ops->begin(state);
 356                 if (error)
 357                         goto Close;
 358         }
 359         suspend_console();
 360         suspend_test_start();
 361         error = device_suspend(PMSG_SUSPEND);
 362         if (error) {
 363                 printk(KERN_ERR "PM: Some devices failed to suspend\n");
 364                 goto Recover_platform;
 365         }
 366         suspend_test_finish("suspend devices");
 367         if (suspend_test(TEST_DEVICES))
 368                 goto Recover_platform;
 369
 370         suspend_enter(state);
 371
 372  Resume_devices:
 373         suspend_test_start();
 374         device_resume(PMSG_RESUME);
 375         suspend_test_finish("resume devices");
 376         resume_console();
 377  Close:
 378         if (suspend_ops->end)
 379                 suspend_ops->end();
 380         return error;
 381
 382  Recover_platform:
 383         if (suspend_ops->recover)
 384                 suspend_ops->recover();
 385         goto Resume_devices;
 386 }
 387
 388 /**
 389  *      suspend_finish - Do final work before exiting suspend sequence.
 390  *
 391  *      Call platform code to clean up, restart processes, and free the
 392  *      console that we've allocated. This is not called for suspend-to-disk.
 393  */
 394 static void suspend_finish(void)
 395 {
 396         suspend_thaw_processes();
 397         usermodehelper_enable();
 398         pm_notifier_call_chain(PM_POST_SUSPEND);
 399         pm_restore_console();
 400 }
 401
 402
 403
 404
 405 static const char * const pm_states[PM_SUSPEND_MAX] = {
 406         [PM_SUSPEND_STANDBY]    = "standby",
 407         [PM_SUSPEND_MEM]        = "mem",
 408 };
 409
 410 static inline int valid_state(suspend_state_t state)
 411 {
 412         /* All states need lowlevel support and need to be valid
 413          * to the lowlevel implementation, no valid callback
 414          * implies that none are valid. */
 415         if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
 416                 return 0;
 417         return 1;
 418 }
 419
 420
 421 /**
 422  *      enter_state - Do common work of entering low-power state.
 423  *      @state:         pm_state structure for state we're entering.
 424  *
 425  *      Make sure we're the only ones trying to enter a sleep state. Fail
 426  *      if someone has beat us to it, since we don't want anything weird to
 427  *      happen when we wake up.
 428  *      Then, do the setup for suspend, enter the state, and cleaup (after
 429  *      we've woken up).
 430  */
 431 static int enter_state(suspend_state_t state)
 432 {
 433         int error;
 434
 435         if (!valid_state(state))
 436                 return -ENODEV;
 437
 438         if (!mutex_trylock(&pm_mutex))
 439                 return -EBUSY;
 440
 441         printk(KERN_INFO "PM: Syncing filesystems ... ");
 442         sys_sync();
 443         printk("done.\n");
 444
 445         pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
 446         error = suspend_prepare();
 447         if (error)
 448                 goto Unlock;
 449
 450         if (suspend_test(TEST_FREEZER))
 451                 goto Finish;
 452
 453         pr_debug("PM: Entering %s sleep\n", pm_states[state]);
 454         error = suspend_devices_and_enter(state);
 455
 456  Finish:
 457         pr_debug("PM: Finishing wakeup.\n");
 458         suspend_finish();
 459  Unlock:
 460         mutex_unlock(&pm_mutex);
 461         return error;
 462 }
 463
 464
 465 /**
 466  *      pm_suspend - Externally visible function for suspending system.
 467  *      @state:         Enumerated value of state to enter.
 468  *
 469  *      Determine whether or not value is within range, get state
 470  *      structure, and enter (above).
 471  */
 472
 473 int pm_suspend(suspend_state_t state)
 474 {
 475         if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
 476                 return enter_state(state);
 477         return -EINVAL;
 478 }
 479
 480 EXPORT_SYMBOL(pm_suspend);
 481
 482 #endif /* CONFIG_SUSPEND */
 483
 484 struct kobject *power_kobj;
 485
 486 /**
 487  *      state - control system power state.
 488  *
 489  *      show() returns what states are supported, which is hard-coded to
 490  *      'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
 491  *      'disk' (Suspend-to-Disk).
 492  *
 493  *      store() accepts one of those strings, translates it into the
 494  *      proper enumerated value, and initiates a suspend transition.
 495  */
 496
 497 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
 498                           char *buf)
 499 {
 500         char *s = buf;
 501 #ifdef CONFIG_SUSPEND
 502         int i;
 503
 504         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 505                 if (pm_states[i] && valid_state(i))
 506                         s += sprintf(s,"%s ", pm_states[i]);
 507         }
 508 #endif
 509 #ifdef CONFIG_HIBERNATION
 510         s += sprintf(s, "%s\n", "disk");
 511 #else
 512         if (s != buf)
 513                 /* convert the last space to a newline */
 514                 *(s-1) = '\n';
 515 #endif
 516         return (s - buf);
 517 }
 518
 519 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
 520                            const char *buf, size_t n)
 521 {
 522 #ifdef CONFIG_SUSPEND
 523         suspend_state_t state = PM_SUSPEND_STANDBY;
 524         const char * const *s;
 525 #endif
 526         char *p;
 527         int len;
 528         int error = -EINVAL;
 529
 530         p = memchr(buf, '\n', n);
 531         len = p ? p - buf : n;
 532
 533         /* First, check if we are requested to hibernate */
 534         if (len == 4 && !strncmp(buf, "disk", len)) {
 535                 error = hibernate();
 536   goto Exit;
 537         }
 538
 539 #ifdef CONFIG_SUSPEND
 540         for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
 541                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
 542                         break;
 543         }
 544         if (state < PM_SUSPEND_MAX && *s)
 545                 error = enter_state(state);
 546 #endif
 547
 548  Exit:
 549         return error ? error : n;
 550 }
 551
 552 power_attr(state);
 553
 554 #ifdef CONFIG_PM_TRACE
 555 int pm_trace_enabled;
 556
 557 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
 558                              char *buf)
 559 {
 560         return sprintf(buf, "%d\n", pm_trace_enabled);
 561 }
 562
 563 static ssize_t
 564 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
 565                const char *buf, size_t n)
 566 {
 567         int val;
 568
 569         if (sscanf(buf, "%d", &val) == 1) {
 570                 pm_trace_enabled = !!val;
 571                 return n;
 572         }
 573         return -EINVAL;
 574 }
 575
 576 power_attr(pm_trace);
 577 #endif /* CONFIG_PM_TRACE */
 578
 579 static struct attribute * g[] = {
 580         &state_attr.attr,
 581 #ifdef CONFIG_PM_TRACE
 582         &pm_trace_attr.attr,
 583 #endif
 584 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
 585         &pm_test_attr.attr,
 586 #endif
 587         NULL,
 588 };
 589
 590 static struct attribute_group attr_group = {
 591         .attrs = g,
 592 };
 593
 594
 595 static int __init pm_init(void)
 596 {
 597         power_kobj = kobject_create_and_add("power", NULL);
 598         if (!power_kobj)
 599                 return -ENOMEM;
 600         return sysfs_create_group(power_kobj, &attr_group);
 601 }
 602
 603 core_initcall(pm_init);
 604
 605
 606 #ifdef CONFIG_PM_TEST_SUSPEND
 607
 608 #include <linux/rtc.h>
 609
 610 /*
 611  * To test system suspend, we need a hands-off mechanism to resume the
 612  * system.  RTCs wake alarms are a common self-contained mechanism.
 613  */
 614
 615 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
 616 {
 617         static char err_readtime[] __initdata =
 618                 KERN_ERR "PM: can't read %s time, err %d\n";
 619         static char err_wakealarm [] __initdata =
 620                 KERN_ERR "PM: can't set %s wakealarm, err %d\n";
 621         static char err_suspend[] __initdata =
 622                 KERN_ERR "PM: suspend test failed, error %d\n";
 623         static char info_test[] __initdata =
 624                 KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
 625
 626         unsigned long           now;
 627         struct rtc_wkalrm       alm;
 628         int                     status;
 629
 630         /* this may fail if the RTC hasn't been initialized */
 631         status = rtc_read_time(rtc, &alm.time);
 632         if (status < 0) {
 633                 printk(err_readtime, dev_name(&rtc->dev), status);
 634                 return;
 635         }
 636         rtc_tm_to_time(&alm.time, &now);
 637
 638         memset(&alm, 0, sizeof alm);
 639         rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
 640         alm.enabled = true;
 641
 642         status = rtc_set_alarm(rtc, &alm);
 643         if (status < 0) {
 644                 printk(err_wakealarm, dev_name(&rtc->dev), status);
 645                 return;
 646         }
 647
 648         if (state == PM_SUSPEND_MEM) {
 649                 printk(info_test, pm_states[state]);
 650                 status = pm_suspend(state);
 651                 if (status == -ENODEV)
 652                         state = PM_SUSPEND_STANDBY;
 653         }
 654         if (state == PM_SUSPEND_STANDBY) {
 655                 printk(info_test, pm_states[state]);
 656                 status = pm_suspend(state);
 657         }
 658         if (status < 0)
 659                 printk(err_suspend, status);
 660
 661         /* Some platforms can't detect that the alarm triggered the
 662          * wakeup, or (accordingly) disable it after it afterwards.
 663          * It's supposed to give oneshot behavior; cope.
 664          */
 665         alm.enabled = false;
 666         rtc_set_alarm(rtc, &alm);
 667 }
 668
 669 static int __init has_wakealarm(struct device *dev, void *name_ptr)
 670 {
 671         struct rtc_device *candidate = to_rtc_device(dev);
 672
 673         if (!candidate->ops->set_alarm)
 674                 return 0;
 675         if (!device_may_wakeup(candidate->dev.parent))
 676                 return 0;
 677
 678         *(const char **)name_ptr = dev_name(dev);
 679         return 1;
 680 }
 681
 682 /*
 683  * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
 684  * at startup time.  They're normally disabled, for faster boot and because
 685  * we can't know which states really work on this particular system.
 686  */
 687 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
 688
 689 static char warn_bad_state[] __initdata =
 690         KERN_WARNING "PM: can't test '%s' suspend state\n";
 691
 692 static int __init setup_test_suspend(char *value)
 693 {
 694         unsigned i;
 695
 696         /* "=mem" ==> "mem" */
 697         value++;
 698         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 699                 if (!pm_states[i])
 700                         continue;
 701                 if (strcmp(pm_states[i], value) != 0)
 702                         continue;
 703                 test_state = (__force suspend_state_t) i;
 704                 return 0;
 705         }
 706         printk(warn_bad_state, value);
 707         return 0;
 708 }
 709 __setup("test_suspend", setup_test_suspend);
 710
 711 static int __init test_suspend(void)
 712 {
 713         static char             warn_no_rtc[] __initdata =
 714                 KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
 715
 716         char                    *pony = NULL;
 717         struct rtc_device       *rtc = NULL;
 718
 719         /* PM is initialized by now; is that state testable? */
 720         if (test_state == PM_SUSPEND_ON)
 721                 goto done;
 722         if (!valid_state(test_state)) {
 723                 printk(warn_bad_state, pm_states[test_state]);
 724                 goto done;
 725         }
 726
 727         /* RTCs have initialized by now too ... can we use one? */
 728         class_find_device(rtc_class, NULL, &pony, has_wakealarm);
 729         if (pony)
 730                 rtc = rtc_class_open(pony);
 731         if (!rtc) {
 732                 printk(warn_no_rtc);
 733                 goto done;
 734         }
 735
 736         /* go for it */
 737         test_wakealarm(rtc, test_state);
 738         rtc_class_close(rtc);
 739 done:
 740         return 0;
 741 }
 742 late_initcall(test_suspend);
 743
 744 #endif /* CONFIG_PM_TEST_SUSPEND */