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 = 0;
 291
 292         device_pm_lock();
 293         arch_suspend_disable_irqs();
 294         BUG_ON(!irqs_disabled());
 295
 296         if ((error = device_power_down(PMSG_SUSPEND))) {
 297                 printk(KERN_ERR "PM: Some devices failed to power down\n");
 298                 goto Done;
 299         }
 300
 301         error = sysdev_suspend(PMSG_SUSPEND);
 302         if (!error) {
 303                 if (!suspend_test(TEST_CORE))
 304                         error = suspend_ops->enter(state);
 305                 sysdev_resume();
 306         }
 307
 308         device_power_up(PMSG_RESUME);
 309  Done:
 310         arch_suspend_enable_irqs();
 311         BUG_ON(irqs_disabled());
 312         device_pm_unlock();
 313         return error;
 314 }
 315
 316 /**
 317  *      suspend_devices_and_enter - suspend devices and enter the desired system
 318  *                                  sleep state.
 319  *      @state:           state to enter
 320  */
 321 int suspend_devices_and_enter(suspend_state_t state)
 322 {
 323         int error;
 324
 325         if (!suspend_ops)
 326                 return -ENOSYS;
 327
 328         if (suspend_ops->begin) {
 329                 error = suspend_ops->begin(state);
 330                 if (error)
 331                         goto Close;
 332         }
 333         suspend_console();
 334         suspend_test_start();
 335         error = device_suspend(PMSG_SUSPEND);
 336         if (error) {
 337                 printk(KERN_ERR "PM: Some devices failed to suspend\n");
 338                 goto Recover_platform;
 339         }
 340         suspend_test_finish("suspend devices");
 341         if (suspend_test(TEST_DEVICES))
 342                 goto Recover_platform;
 343
 344         if (suspend_ops->prepare) {
 345                 error = suspend_ops->prepare();
 346                 if (error)
 347                         goto Resume_devices;
 348         }
 349
 350         if (suspend_test(TEST_PLATFORM))
 351                 goto Finish;
 352
 353         error = disable_nonboot_cpus();
 354         if (!error && !suspend_test(TEST_CPUS))
 355                 suspend_enter(state);
 356
 357         enable_nonboot_cpus();
 358  Finish:
 359         if (suspend_ops->finish)
 360                 suspend_ops->finish();
 361  Resume_devices:
 362         suspend_test_start();
 363         device_resume(PMSG_RESUME);
 364         suspend_test_finish("resume devices");
 365         resume_console();
 366  Close:
 367         if (suspend_ops->end)
 368                 suspend_ops->end();
 369         return error;
 370
 371  Recover_platform:
 372         if (suspend_ops->recover)
 373                 suspend_ops->recover();
 374         goto Resume_devices;
 375 }
 376
 377 /**
 378  *      suspend_finish - Do final work before exiting suspend sequence.
 379  *
 380  *      Call platform code to clean up, restart processes, and free the
 381  *      console that we've allocated. This is not called for suspend-to-disk.
 382  */
 383 static void suspend_finish(void)
 384 {
 385         suspend_thaw_processes();
 386         usermodehelper_enable();
 387         pm_notifier_call_chain(PM_POST_SUSPEND);
 388         pm_restore_console();
 389 }
 390
 391
 392
 393
 394 static const char * const pm_states[PM_SUSPEND_MAX] = {
 395         [PM_SUSPEND_STANDBY]    = "standby",
 396         [PM_SUSPEND_MEM]        = "mem",
 397 };
 398
 399 static inline int valid_state(suspend_state_t state)
 400 {
 401         /* All states need lowlevel support and need to be valid
 402          * to the lowlevel implementation, no valid callback
 403          * implies that none are valid. */
 404         if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
 405                 return 0;
 406         return 1;
 407 }
 408
 409
 410 /**
 411  *      enter_state - Do common work of entering low-power state.
 412  *      @state:         pm_state structure for state we're entering.
 413  *
 414  *      Make sure we're the only ones trying to enter a sleep state. Fail
 415  *      if someone has beat us to it, since we don't want anything weird to
 416  *      happen when we wake up.
 417  *      Then, do the setup for suspend, enter the state, and cleaup (after
 418  *      we've woken up).
 419  */
 420 static int enter_state(suspend_state_t state)
 421 {
 422         int error;
 423
 424         if (!valid_state(state))
 425                 return -ENODEV;
 426
 427         if (!mutex_trylock(&pm_mutex))
 428                 return -EBUSY;
 429
 430         printk(KERN_INFO "PM: Syncing filesystems ... ");
 431         sys_sync();
 432         printk("done.\n");
 433
 434         pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
 435         error = suspend_prepare();
 436         if (error)
 437                 goto Unlock;
 438
 439         if (suspend_test(TEST_FREEZER))
 440                 goto Finish;
 441
 442         pr_debug("PM: Entering %s sleep\n", pm_states[state]);
 443         error = suspend_devices_and_enter(state);
 444
 445  Finish:
 446         pr_debug("PM: Finishing wakeup.\n");
 447         suspend_finish();
 448  Unlock:
 449         mutex_unlock(&pm_mutex);
 450         return error;
 451 }
 452
 453
 454 /**
 455  *      pm_suspend - Externally visible function for suspending system.
 456  *      @state:         Enumerated value of state to enter.
 457  *
 458  *      Determine whether or not value is within range, get state
 459  *      structure, and enter (above).
 460  */
 461
 462 int pm_suspend(suspend_state_t state)
 463 {
 464         if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
 465                 return enter_state(state);
 466         return -EINVAL;
 467 }
 468
 469 EXPORT_SYMBOL(pm_suspend);
 470
 471 #endif /* CONFIG_SUSPEND */
 472
 473 struct kobject *power_kobj;
 474
 475 /**
 476  *      state - control system power state.
 477  *
 478  *      show() returns what states are supported, which is hard-coded to
 479  *      'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
 480  *      'disk' (Suspend-to-Disk).
 481  *
 482  *      store() accepts one of those strings, translates it into the
 483  *      proper enumerated value, and initiates a suspend transition.
 484  */
 485
 486 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
 487                           char *buf)
 488 {
 489         char *s = buf;
 490 #ifdef CONFIG_SUSPEND
 491         int i;
 492
 493         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 494                 if (pm_states[i] && valid_state(i))
 495                         s += sprintf(s,"%s ", pm_states[i]);
 496         }
 497 #endif
 498 #ifdef CONFIG_HIBERNATION
 499         s += sprintf(s, "%s\n", "disk");
 500 #else
 501         if (s != buf)
 502                 /* convert the last space to a newline */
 503                 *(s-1) = '\n';
 504 #endif
 505         return (s - buf);
 506 }
 507
 508 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
 509                            const char *buf, size_t n)
 510 {
 511 #ifdef CONFIG_SUSPEND
 512         suspend_state_t state = PM_SUSPEND_STANDBY;
 513         const char * const *s;
 514 #endif
 515         char *p;
 516         int len;
 517         int error = -EINVAL;
 518
 519         p = memchr(buf, '\n', n);
 520         len = p ? p - buf : n;
 521
 522         /* First, check if we are requested to hibernate */
 523         if (len == 4 && !strncmp(buf, "disk", len)) {
 524                 error = hibernate();
 525   goto Exit;
 526         }
 527
 528 #ifdef CONFIG_SUSPEND
 529         for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
 530                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
 531                         break;
 532         }
 533         if (state < PM_SUSPEND_MAX && *s)
 534                 error = enter_state(state);
 535 #endif
 536
 537  Exit:
 538         return error ? error : n;
 539 }
 540
 541 power_attr(state);
 542
 543 #ifdef CONFIG_PM_TRACE
 544 int pm_trace_enabled;
 545
 546 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
 547                              char *buf)
 548 {
 549         return sprintf(buf, "%d\n", pm_trace_enabled);
 550 }
 551
 552 static ssize_t
 553 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
 554                const char *buf, size_t n)
 555 {
 556         int val;
 557
 558         if (sscanf(buf, "%d", &val) == 1) {
 559                 pm_trace_enabled = !!val;
 560                 return n;
 561         }
 562         return -EINVAL;
 563 }
 564
 565 power_attr(pm_trace);
 566 #endif /* CONFIG_PM_TRACE */
 567
 568 static struct attribute * g[] = {
 569         &state_attr.attr,
 570 #ifdef CONFIG_PM_TRACE
 571         &pm_trace_attr.attr,
 572 #endif
 573 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
 574         &pm_test_attr.attr,
 575 #endif
 576         NULL,
 577 };
 578
 579 static struct attribute_group attr_group = {
 580         .attrs = g,
 581 };
 582
 583
 584 static int __init pm_init(void)
 585 {
 586         power_kobj = kobject_create_and_add("power", NULL);
 587         if (!power_kobj)
 588                 return -ENOMEM;
 589         return sysfs_create_group(power_kobj, &attr_group);
 590 }
 591
 592 core_initcall(pm_init);
 593
 594
 595 #ifdef CONFIG_PM_TEST_SUSPEND
 596
 597 #include <linux/rtc.h>
 598
 599 /*
 600  * To test system suspend, we need a hands-off mechanism to resume the
 601  * system.  RTCs wake alarms are a common self-contained mechanism.
 602  */
 603
 604 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
 605 {
 606         static char err_readtime[] __initdata =
 607                 KERN_ERR "PM: can't read %s time, err %d\n";
 608         static char err_wakealarm [] __initdata =
 609                 KERN_ERR "PM: can't set %s wakealarm, err %d\n";
 610         static char err_suspend[] __initdata =
 611                 KERN_ERR "PM: suspend test failed, error %d\n";
 612         static char info_test[] __initdata =
 613                 KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
 614
 615         unsigned long           now;
 616         struct rtc_wkalrm       alm;
 617         int                     status;
 618
 619         /* this may fail if the RTC hasn't been initialized */
 620         status = rtc_read_time(rtc, &alm.time);
 621         if (status < 0) {
 622                 printk(err_readtime, dev_name(&rtc->dev), status);
 623                 return;
 624         }
 625         rtc_tm_to_time(&alm.time, &now);
 626
 627         memset(&alm, 0, sizeof alm);
 628         rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
 629         alm.enabled = true;
 630
 631         status = rtc_set_alarm(rtc, &alm);
 632         if (status < 0) {
 633                 printk(err_wakealarm, dev_name(&rtc->dev), status);
 634                 return;
 635         }
 636
 637         if (state == PM_SUSPEND_MEM) {
 638                 printk(info_test, pm_states[state]);
 639                 status = pm_suspend(state);
 640                 if (status == -ENODEV)
 641                         state = PM_SUSPEND_STANDBY;
 642         }
 643         if (state == PM_SUSPEND_STANDBY) {
 644                 printk(info_test, pm_states[state]);
 645                 status = pm_suspend(state);
 646         }
 647         if (status < 0)
 648                 printk(err_suspend, status);
 649
 650         /* Some platforms can't detect that the alarm triggered the
 651          * wakeup, or (accordingly) disable it after it afterwards.
 652          * It's supposed to give oneshot behavior; cope.
 653          */
 654         alm.enabled = false;
 655         rtc_set_alarm(rtc, &alm);
 656 }
 657
 658 static int __init has_wakealarm(struct device *dev, void *name_ptr)
 659 {
 660         struct rtc_device *candidate = to_rtc_device(dev);
 661
 662         if (!candidate->ops->set_alarm)
 663                 return 0;
 664         if (!device_may_wakeup(candidate->dev.parent))
 665                 return 0;
 666
 667         *(const char **)name_ptr = dev_name(dev);
 668         return 1;
 669 }
 670
 671 /*
 672  * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
 673  * at startup time.  They're normally disabled, for faster boot and because
 674  * we can't know which states really work on this particular system.
 675  */
 676 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
 677
 678 static char warn_bad_state[] __initdata =
 679         KERN_WARNING "PM: can't test '%s' suspend state\n";
 680
 681 static int __init setup_test_suspend(char *value)
 682 {
 683         unsigned i;
 684
 685         /* "=mem" ==> "mem" */
 686         value++;
 687         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 688                 if (!pm_states[i])
 689                         continue;
 690                 if (strcmp(pm_states[i], value) != 0)
 691                         continue;
 692                 test_state = (__force suspend_state_t) i;
 693                 return 0;
 694         }
 695         printk(warn_bad_state, value);
 696         return 0;
 697 }
 698 __setup("test_suspend", setup_test_suspend);
 699
 700 static int __init test_suspend(void)
 701 {
 702         static char             warn_no_rtc[] __initdata =
 703                 KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
 704
 705         char                    *pony = NULL;
 706         struct rtc_device       *rtc = NULL;
 707
 708         /* PM is initialized by now; is that state testable? */
 709         if (test_state == PM_SUSPEND_ON)
 710                 goto done;
 711         if (!valid_state(test_state)) {
 712                 printk(warn_bad_state, pm_states[test_state]);
 713                 goto done;
 714         }
 715
 716         /* RTCs have initialized by now too ... can we use one? */
 717         class_find_device(rtc_class, NULL, &pony, has_wakealarm);
 718         if (pony)
 719                 rtc = rtc_class_open(pony);
 720         if (!rtc) {
 721                 printk(warn_no_rtc);
 722                 goto done;
 723         }
 724
 725         /* go for it */
 726         test_wakealarm(rtc, test_state);
 727         rtc_class_close(rtc);
 728 done:
 729         return 0;
 730 }
 731 late_initcall(test_suspend);
 732
 733 #endif /* CONFIG_PM_TEST_SUSPEND */