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/init.h>
  18 #include <linux/console.h>
  19 #include <linux/cpu.h>
  20 #include <linux/resume-trace.h>
  21 #include <linux/freezer.h>
  22 #include <linux/vmstat.h>
  23 #include <linux/syscalls.h>
  24 #include <linux/ftrace.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 int suspend_test(int level)
  61 {
  62         if (pm_test_level == level) {
  63                 printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n");
  64                 mdelay(5000);
  65                 return 1;
  66         }
  67         return 0;
  68 }
  69
  70 static const char * const pm_tests[__TEST_AFTER_LAST] = {
  71         [TEST_NONE] = "none",
  72         [TEST_CORE] = "core",
  73         [TEST_CPUS] = "processors",
  74         [TEST_PLATFORM] = "platform",
  75         [TEST_DEVICES] = "devices",
  76         [TEST_FREEZER] = "freezer",
  77 };
  78
  79 static ssize_t pm_test_show(struct kobject *kobj, struct kobj_attribute *attr,
  80                                 char *buf)
  81 {
  82         char *s = buf;
  83         int level;
  84
  85         for (level = TEST_FIRST; level <= TEST_MAX; level++)
  86                 if (pm_tests[level]) {
  87                         if (level == pm_test_level)
  88                                 s += sprintf(s, "[%s] ", pm_tests[level]);
  89                         else
  90                                 s += sprintf(s, "%s ", pm_tests[level]);
  91                 }
  92
  93         if (s != buf)
  94                 /* convert the last space to a newline */
  95                 *(s-1) = '\n';
  96
  97         return (s - buf);
  98 }
  99
 100 static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
 101                                 const char *buf, size_t n)
 102 {
 103         const char * const *s;
 104         int level;
 105         char *p;
 106         int len;
 107         int error = -EINVAL;
 108
 109         p = memchr(buf, '\n', n);
 110         len = p ? p - buf : n;
 111
 112         mutex_lock(&pm_mutex);
 113
 114         level = TEST_FIRST;
 115         for (s = &pm_tests[level]; level <= TEST_MAX; s++, level++)
 116                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len)) {
 117                         pm_test_level = level;
 118                         error = 0;
 119                         break;
 120                 }
 121
 122         mutex_unlock(&pm_mutex);
 123
 124         return error ? error : n;
 125 }
 126
 127 power_attr(pm_test);
 128 #else /* !CONFIG_PM_DEBUG */
 129 static inline int suspend_test(int level) { return 0; }
 130 #endif /* !CONFIG_PM_DEBUG */
 131
 132 #endif /* CONFIG_PM_SLEEP */
 133
 134 #ifdef CONFIG_SUSPEND
 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_ON(msec > (TEST_SUSPEND_SECONDS * 1000));
 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         if (suspend_freeze_processes()) {
 241                 error = -EAGAIN;
 242                 goto Thaw;
 243         }
 244
 245         free_pages = global_page_state(NR_FREE_PAGES);
 246         if (free_pages < FREE_PAGE_NUMBER) {
 247                 pr_debug("PM: free some memory\n");
 248                 shrink_all_memory(FREE_PAGE_NUMBER - free_pages);
 249                 if (nr_free_pages() < FREE_PAGE_NUMBER) {
 250                         error = -ENOMEM;
 251                         printk(KERN_ERR "PM: No enough memory\n");
 252                 }
 253         }
 254         if (!error)
 255                 return 0;
 256
 257  Thaw:
 258         suspend_thaw_processes();
 259  Finish:
 260         pm_notifier_call_chain(PM_POST_SUSPEND);
 261         pm_restore_console();
 262         return error;
 263 }
 264
 265 /* default implementation */
 266 void __attribute__ ((weak)) arch_suspend_disable_irqs(void)
 267 {
 268         local_irq_disable();
 269 }
 270
 271 /* default implementation */
 272 void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
 273 {
 274         local_irq_enable();
 275 }
 276
 277 /**
 278  *      suspend_enter - enter the desired system sleep state.
 279  *      @state:         state to enter
 280  *
 281  *      This function should be called after devices have been suspended.
 282  */
 283 static int suspend_enter(suspend_state_t state)
 284 {
 285         int error = 0;
 286
 287         device_pm_lock();
 288         arch_suspend_disable_irqs();
 289         BUG_ON(!irqs_disabled());
 290
 291         if ((error = device_power_down(PMSG_SUSPEND))) {
 292                 printk(KERN_ERR "PM: Some devices failed to power down\n");
 293                 goto Done;
 294         }
 295
 296         if (!suspend_test(TEST_CORE))
 297                 error = suspend_ops->enter(state);
 298
 299         device_power_up(PMSG_RESUME);
 300  Done:
 301         arch_suspend_enable_irqs();
 302         BUG_ON(irqs_disabled());
 303         device_pm_unlock();
 304         return error;
 305 }
 306
 307 /**
 308  *      suspend_devices_and_enter - suspend devices and enter the desired system
 309  *                                  sleep state.
 310  *      @state:           state to enter
 311  */
 312 int suspend_devices_and_enter(suspend_state_t state)
 313 {
 314         int error, ftrace_save;
 315
 316         if (!suspend_ops)
 317                 return -ENOSYS;
 318
 319         if (suspend_ops->begin) {
 320                 error = suspend_ops->begin(state);
 321                 if (error)
 322                         goto Close;
 323         }
 324         suspend_console();
 325         ftrace_save = __ftrace_enabled_save();
 326         suspend_test_start();
 327         error = device_suspend(PMSG_SUSPEND);
 328         if (error) {
 329                 printk(KERN_ERR "PM: Some devices failed to suspend\n");
 330                 goto Recover_platform;
 331         }
 332         suspend_test_finish("suspend devices");
 333         if (suspend_test(TEST_DEVICES))
 334                 goto Recover_platform;
 335
 336         if (suspend_ops->prepare) {
 337                 error = suspend_ops->prepare();
 338                 if (error)
 339                         goto Resume_devices;
 340         }
 341
 342         if (suspend_test(TEST_PLATFORM))
 343                 goto Finish;
 344
 345         error = disable_nonboot_cpus();
 346         if (!error && !suspend_test(TEST_CPUS))
 347                 suspend_enter(state);
 348
 349         enable_nonboot_cpus();
 350  Finish:
 351         if (suspend_ops->finish)
 352                 suspend_ops->finish();
 353  Resume_devices:
 354         suspend_test_start();
 355         device_resume(PMSG_RESUME);
 356         suspend_test_finish("resume devices");
 357         __ftrace_enabled_restore(ftrace_save);
 358         resume_console();
 359  Close:
 360         if (suspend_ops->end)
 361                 suspend_ops->end();
 362         return error;
 363
 364  Recover_platform:
 365         if (suspend_ops->recover)
 366                 suspend_ops->recover();
 367         goto Resume_devices;
 368 }
 369
 370 /**
 371  *      suspend_finish - Do final work before exiting suspend sequence.
 372  *
 373  *      Call platform code to clean up, restart processes, and free the
 374  *      console that we've allocated. This is not called for suspend-to-disk.
 375  */
 376 static void suspend_finish(void)
 377 {
 378         suspend_thaw_processes();
 379         pm_notifier_call_chain(PM_POST_SUSPEND);
 380         pm_restore_console();
 381 }
 382
 383
 384
 385
 386 static const char * const pm_states[PM_SUSPEND_MAX] = {
 387         [PM_SUSPEND_STANDBY]    = "standby",
 388         [PM_SUSPEND_MEM]        = "mem",
 389 };
 390
 391 static inline int valid_state(suspend_state_t state)
 392 {
 393         /* All states need lowlevel support and need to be valid
 394          * to the lowlevel implementation, no valid callback
 395          * implies that none are valid. */
 396         if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
 397                 return 0;
 398         return 1;
 399 }
 400
 401
 402 /**
 403  *      enter_state - Do common work of entering low-power state.
 404  *      @state:         pm_state structure for state we're entering.
 405  *
 406  *      Make sure we're the only ones trying to enter a sleep state. Fail
 407  *      if someone has beat us to it, since we don't want anything weird to
 408  *      happen when we wake up.
 409  *      Then, do the setup for suspend, enter the state, and cleaup (after
 410  *      we've woken up).
 411  */
 412 static int enter_state(suspend_state_t state)
 413 {
 414         int error;
 415
 416         if (!valid_state(state))
 417                 return -ENODEV;
 418
 419         if (!mutex_trylock(&pm_mutex))
 420                 return -EBUSY;
 421
 422         printk(KERN_INFO "PM: Syncing filesystems ... ");
 423         sys_sync();
 424         printk("done.\n");
 425
 426         pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
 427         error = suspend_prepare();
 428         if (error)
 429                 goto Unlock;
 430
 431         if (suspend_test(TEST_FREEZER))
 432                 goto Finish;
 433
 434         pr_debug("PM: Entering %s sleep\n", pm_states[state]);
 435         error = suspend_devices_and_enter(state);
 436
 437  Finish:
 438         pr_debug("PM: Finishing wakeup.\n");
 439         suspend_finish();
 440  Unlock:
 441         mutex_unlock(&pm_mutex);
 442         return error;
 443 }
 444
 445
 446 /**
 447  *      pm_suspend - Externally visible function for suspending system.
 448  *      @state:         Enumerated value of state to enter.
 449  *
 450  *      Determine whether or not value is within range, get state
 451  *      structure, and enter (above).
 452  */
 453
 454 int pm_suspend(suspend_state_t state)
 455 {
 456         if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
 457                 return enter_state(state);
 458         return -EINVAL;
 459 }
 460
 461 EXPORT_SYMBOL(pm_suspend);
 462
 463 #endif /* CONFIG_SUSPEND */
 464
 465 struct kobject *power_kobj;
 466
 467 /**
 468  *      state - control system power state.
 469  *
 470  *      show() returns what states are supported, which is hard-coded to
 471  *      'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
 472  *      'disk' (Suspend-to-Disk).
 473  *
 474  *      store() accepts one of those strings, translates it into the
 475  *      proper enumerated value, and initiates a suspend transition.
 476  */
 477
 478 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
 479                           char *buf)
 480 {
 481         char *s = buf;
 482 #ifdef CONFIG_SUSPEND
 483         int i;
 484
 485         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 486                 if (pm_states[i] && valid_state(i))
 487                         s += sprintf(s,"%s ", pm_states[i]);
 488         }
 489 #endif
 490 #ifdef CONFIG_HIBERNATION
 491         s += sprintf(s, "%s\n", "disk");
 492 #else
 493         if (s != buf)
 494                 /* convert the last space to a newline */
 495                 *(s-1) = '\n';
 496 #endif
 497         return (s - buf);
 498 }
 499
 500 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
 501                            const char *buf, size_t n)
 502 {
 503 #ifdef CONFIG_SUSPEND
 504         suspend_state_t state = PM_SUSPEND_STANDBY;
 505         const char * const *s;
 506 #endif
 507         char *p;
 508         int len;
 509         int error = -EINVAL;
 510
 511         p = memchr(buf, '\n', n);
 512         len = p ? p - buf : n;
 513
 514         /* First, check if we are requested to hibernate */
 515         if (len == 4 && !strncmp(buf, "disk", len)) {
 516                 error = hibernate();
 517   goto Exit;
 518         }
 519
 520 #ifdef CONFIG_SUSPEND
 521         for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
 522                 if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
 523                         break;
 524         }
 525         if (state < PM_SUSPEND_MAX && *s)
 526                 error = enter_state(state);
 527 #endif
 528
 529  Exit:
 530         return error ? error : n;
 531 }
 532
 533 power_attr(state);
 534
 535 #ifdef CONFIG_PM_TRACE
 536 int pm_trace_enabled;
 537
 538 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
 539                              char *buf)
 540 {
 541         return sprintf(buf, "%d\n", pm_trace_enabled);
 542 }
 543
 544 static ssize_t
 545 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
 546                const char *buf, size_t n)
 547 {
 548         int val;
 549
 550         if (sscanf(buf, "%d", &val) == 1) {
 551                 pm_trace_enabled = !!val;
 552                 return n;
 553         }
 554         return -EINVAL;
 555 }
 556
 557 power_attr(pm_trace);
 558 #endif /* CONFIG_PM_TRACE */
 559
 560 static struct attribute * g[] = {
 561         &state_attr.attr,
 562 #ifdef CONFIG_PM_TRACE
 563         &pm_trace_attr.attr,
 564 #endif
 565 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
 566         &pm_test_attr.attr,
 567 #endif
 568         NULL,
 569 };
 570
 571 static struct attribute_group attr_group = {
 572         .attrs = g,
 573 };
 574
 575
 576 static int __init pm_init(void)
 577 {
 578         power_kobj = kobject_create_and_add("power", NULL);
 579         if (!power_kobj)
 580                 return -ENOMEM;
 581         return sysfs_create_group(power_kobj, &attr_group);
 582 }
 583
 584 core_initcall(pm_init);
 585
 586
 587 #ifdef CONFIG_PM_TEST_SUSPEND
 588
 589 #include <linux/rtc.h>
 590
 591 /*
 592  * To test system suspend, we need a hands-off mechanism to resume the
 593  * system.  RTCs wake alarms are a common self-contained mechanism.
 594  */
 595
 596 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
 597 {
 598         static char err_readtime[] __initdata =
 599                 KERN_ERR "PM: can't read %s time, err %d\n";
 600         static char err_wakealarm [] __initdata =
 601                 KERN_ERR "PM: can't set %s wakealarm, err %d\n";
 602         static char err_suspend[] __initdata =
 603                 KERN_ERR "PM: suspend test failed, error %d\n";
 604         static char info_test[] __initdata =
 605                 KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
 606
 607         unsigned long           now;
 608         struct rtc_wkalrm       alm;
 609         int                     status;
 610
 611         /* this may fail if the RTC hasn't been initialized */
 612         status = rtc_read_time(rtc, &alm.time);
 613         if (status < 0) {
 614                 printk(err_readtime, rtc->dev.bus_id, status);
 615                 return;
 616         }
 617         rtc_tm_to_time(&alm.time, &now);
 618
 619         memset(&alm, 0, sizeof alm);
 620         rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
 621         alm.enabled = true;
 622
 623         status = rtc_set_alarm(rtc, &alm);
 624         if (status < 0) {
 625                 printk(err_wakealarm, rtc->dev.bus_id, status);
 626                 return;
 627         }
 628
 629         if (state == PM_SUSPEND_MEM) {
 630                 printk(info_test, pm_states[state]);
 631                 status = pm_suspend(state);
 632                 if (status == -ENODEV)
 633                         state = PM_SUSPEND_STANDBY;
 634         }
 635         if (state == PM_SUSPEND_STANDBY) {
 636                 printk(info_test, pm_states[state]);
 637                 status = pm_suspend(state);
 638         }
 639         if (status < 0)
 640                 printk(err_suspend, status);
 641
 642         /* Some platforms can't detect that the alarm triggered the
 643          * wakeup, or (accordingly) disable it after it afterwards.
 644          * It's supposed to give oneshot behavior; cope.
 645          */
 646         alm.enabled = false;
 647         rtc_set_alarm(rtc, &alm);
 648 }
 649
 650 static int __init has_wakealarm(struct device *dev, void *name_ptr)
 651 {
 652         struct rtc_device *candidate = to_rtc_device(dev);
 653
 654         if (!candidate->ops->set_alarm)
 655                 return 0;
 656         if (!device_may_wakeup(candidate->dev.parent))
 657                 return 0;
 658
 659         *(char **)name_ptr = dev->bus_id;
 660         return 1;
 661 }
 662
 663 /*
 664  * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
 665  * at startup time.  They're normally disabled, for faster boot and because
 666  * we can't know which states really work on this particular system.
 667  */
 668 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
 669
 670 static char warn_bad_state[] __initdata =
 671         KERN_WARNING "PM: can't test '%s' suspend state\n";
 672
 673 static int __init setup_test_suspend(char *value)
 674 {
 675         unsigned i;
 676
 677         /* "=mem" ==> "mem" */
 678         value++;
 679         for (i = 0; i < PM_SUSPEND_MAX; i++) {
 680                 if (!pm_states[i])
 681                         continue;
 682                 if (strcmp(pm_states[i], value) != 0)
 683                         continue;
 684                 test_state = (__force suspend_state_t) i;
 685                 return 0;
 686         }
 687         printk(warn_bad_state, value);
 688         return 0;
 689 }
 690 __setup("test_suspend", setup_test_suspend);
 691
 692 static int __init test_suspend(void)
 693 {
 694         static char             warn_no_rtc[] __initdata =
 695                 KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
 696
 697         char                    *pony = NULL;
 698         struct rtc_device       *rtc = NULL;
 699
 700         /* PM is initialized by now; is that state testable? */
 701         if (test_state == PM_SUSPEND_ON)
 702                 goto done;
 703         if (!valid_state(test_state)) {
 704                 printk(warn_bad_state, pm_states[test_state]);
 705                 goto done;
 706         }
 707
 708         /* RTCs have initialized by now too ... can we use one? */
 709         class_find_device(rtc_class, NULL, &pony, has_wakealarm);
 710         if (pony)
 711                 rtc = rtc_class_open(pony);
 712         if (!rtc) {
 713                 printk(warn_no_rtc);
 714                 goto done;
 715         }
 716
 717         /* go for it */
 718         test_wakealarm(rtc, test_state);
 719         rtc_class_close(rtc);
 720 done:
 721         return 0;
 722 }
 723 late_initcall(test_suspend);
 724
 725 #endif /* CONFIG_PM_TEST_SUSPEND */