drivers/watchdog/watchdog_dev.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  *      watchdog_dev.c
   4  *
   5  *      (c) Copyright 2008-2011 Alan Cox <alan@lxorguk.ukuu.org.uk>,
   6  *                                              All Rights Reserved.
   7  *
   8  *      (c) Copyright 2008-2011 Wim Van Sebroeck <wim@iguana.be>.
   9  *
  10  *      (c) Copyright 2021 Hewlett Packard Enterprise Development LP.
  11  *
  12  *      This source code is part of the generic code that can be used
  13  *      by all the watchdog timer drivers.
  14  *
  15  *      This part of the generic code takes care of the following
  16  *      misc device: /dev/watchdog.
  17  *
  18  *      Based on source code of the following authors:
  19  *        Matt Domsch <Matt_Domsch@dell.com>,
  20  *        Rob Radez <rob@osinvestor.com>,
  21  *        Rusty Lynch <rusty@linux.co.intel.com>
  22  *        Satyam Sharma <satyam@infradead.org>
  23  *        Randy Dunlap <randy.dunlap@oracle.com>
  24  *
  25  *      Neither Alan Cox, CymruNet Ltd., Wim Van Sebroeck nor Iguana vzw.
  26  *      admit liability nor provide warranty for any of this software.
  27  *      This material is provided "AS-IS" and at no charge.
  28  */
  29
  30 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  31
  32 #include <linux/cdev.h>         /* For character device */
  33 #include <linux/errno.h>        /* For the -ENODEV/... values */
  34 #include <linux/fs.h>           /* For file operations */
  35 #include <linux/init.h>         /* For __init/__exit/... */
  36 #include <linux/hrtimer.h>      /* For hrtimers */
  37 #include <linux/kernel.h>       /* For printk/panic/... */
  38 #include <linux/kstrtox.h>      /* For kstrto* */
  39 #include <linux/kthread.h>      /* For kthread_work */
  40 #include <linux/miscdevice.h>   /* For handling misc devices */
  41 #include <linux/module.h>       /* For module stuff/... */
  42 #include <linux/mutex.h>        /* For mutexes */
  43 #include <linux/slab.h>         /* For memory functions */
  44 #include <linux/types.h>        /* For standard types (like size_t) */
  45 #include <linux/watchdog.h>     /* For watchdog specific items */
  46 #include <linux/uaccess.h>      /* For copy_to_user/put_user/... */
  47
  48 #include "watchdog_core.h"
  49 #include "watchdog_pretimeout.h"
  50
  51 #include <trace/events/watchdog.h>
  52
  53 /* the dev_t structure to store the dynamically allocated watchdog devices */
  54 static dev_t watchdog_devt;
  55 /* Reference to watchdog device behind /dev/watchdog */
  56 static struct watchdog_core_data *old_wd_data;
  57
  58 static struct kthread_worker *watchdog_kworker;
  59
  60 static bool handle_boot_enabled =
  61         IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED);
  62
  63 static unsigned open_timeout = CONFIG_WATCHDOG_OPEN_TIMEOUT;
  64
  65 static bool watchdog_past_open_deadline(struct watchdog_core_data *data)
  66 {
  67         return ktime_after(ktime_get(), data->open_deadline);
  68 }
  69
  70 static void watchdog_set_open_deadline(struct watchdog_core_data *data)
  71 {
  72         data->open_deadline = open_timeout ?
  73                 ktime_get() + ktime_set(open_timeout, 0) : KTIME_MAX;
  74 }
  75
  76 static inline bool watchdog_need_worker(struct watchdog_device *wdd)
  77 {
  78         /* All variables in milli-seconds */
  79         unsigned int hm = wdd->max_hw_heartbeat_ms;
  80         unsigned int t = wdd->timeout * 1000;
  81
  82         /*
  83          * A worker to generate heartbeat requests is needed if all of the
  84          * following conditions are true.
  85          * - Userspace activated the watchdog.
  86          * - The driver provided a value for the maximum hardware timeout, and
  87          *   thus is aware that the framework supports generating heartbeat
  88          *   requests.
  89          * - Userspace requests a longer timeout than the hardware can handle.
  90          *
  91          * Alternatively, if userspace has not opened the watchdog
  92          * device, we take care of feeding the watchdog if it is
  93          * running.
  94          */
  95         return (hm && watchdog_active(wdd) && t > hm) ||
  96                 (t && !watchdog_active(wdd) && watchdog_hw_running(wdd));
  97 }
  98
  99 static ktime_t watchdog_next_keepalive(struct watchdog_device *wdd)
 100 {
 101         struct watchdog_core_data *wd_data = wdd->wd_data;
 102         unsigned int timeout_ms = wdd->timeout * 1000;
 103         ktime_t keepalive_interval;
 104         ktime_t last_heartbeat, latest_heartbeat;
 105         ktime_t virt_timeout;
 106         unsigned int hw_heartbeat_ms;
 107
 108         if (watchdog_active(wdd))
 109                 virt_timeout = ktime_add(wd_data->last_keepalive,
 110                                          ms_to_ktime(timeout_ms));
 111         else
 112                 virt_timeout = wd_data->open_deadline;
 113
 114         hw_heartbeat_ms = min_not_zero(timeout_ms, wdd->max_hw_heartbeat_ms);
 115         keepalive_interval = ms_to_ktime(hw_heartbeat_ms / 2);
 116
 117         /*
 118          * To ensure that the watchdog times out wdd->timeout seconds
 119          * after the most recent ping from userspace, the last
 120          * worker ping has to come in hw_heartbeat_ms before this timeout.
 121          */
 122         last_heartbeat = ktime_sub(virt_timeout, ms_to_ktime(hw_heartbeat_ms));
 123         latest_heartbeat = ktime_sub(last_heartbeat, ktime_get());
 124         if (ktime_before(latest_heartbeat, keepalive_interval))
 125                 return latest_heartbeat;
 126         return keepalive_interval;
 127 }
 128
 129 static inline void watchdog_update_worker(struct watchdog_device *wdd)
 130 {
 131         struct watchdog_core_data *wd_data = wdd->wd_data;
 132
 133         if (watchdog_need_worker(wdd)) {
 134                 ktime_t t = watchdog_next_keepalive(wdd);
 135
 136                 if (t > 0)
 137                         hrtimer_start(&wd_data->timer, t,
 138                                       HRTIMER_MODE_REL_HARD);
 139         } else {
 140                 hrtimer_cancel(&wd_data->timer);
 141         }
 142 }
 143
 144 static int __watchdog_ping(struct watchdog_device *wdd)
 145 {
 146         struct watchdog_core_data *wd_data = wdd->wd_data;
 147         ktime_t earliest_keepalive, now;
 148         int err;
 149
 150         earliest_keepalive = ktime_add(wd_data->last_hw_keepalive,
 151                                        ms_to_ktime(wdd->min_hw_heartbeat_ms));
 152         now = ktime_get();
 153
 154         if (ktime_after(earliest_keepalive, now)) {
 155                 hrtimer_start(&wd_data->timer,
 156                               ktime_sub(earliest_keepalive, now),
 157                               HRTIMER_MODE_REL_HARD);
 158                 return 0;
 159         }
 160
 161         wd_data->last_hw_keepalive = now;
 162
 163         if (wdd->ops->ping) {
 164                 err = wdd->ops->ping(wdd);  /* ping the watchdog */
 165                 trace_watchdog_ping(wdd, err);
 166         } else {
 167                 err = wdd->ops->start(wdd); /* restart watchdog */
 168                 trace_watchdog_start(wdd, err);
 169         }
 170
 171         if (err == 0)
 172                 watchdog_hrtimer_pretimeout_start(wdd);
 173
 174         watchdog_update_worker(wdd);
 175
 176         return err;
 177 }
 178
 179 /*
 180  * watchdog_ping - ping the watchdog
 181  * @wdd: The watchdog device to ping
 182  *
 183  * If the watchdog has no own ping operation then it needs to be
 184  * restarted via the start operation. This wrapper function does
 185  * exactly that.
 186  * We only ping when the watchdog device is running.
 187  * The caller must hold wd_data->lock.
 188  *
 189  * Return: 0 on success, error otherwise.
 190  */
 191 static int watchdog_ping(struct watchdog_device *wdd)
 192 {
 193         struct watchdog_core_data *wd_data = wdd->wd_data;
 194
 195         if (!watchdog_active(wdd) && !watchdog_hw_running(wdd))
 196                 return 0;
 197
 198         set_bit(_WDOG_KEEPALIVE, &wd_data->status);
 199
 200         wd_data->last_keepalive = ktime_get();
 201         return __watchdog_ping(wdd);
 202 }
 203
 204 static bool watchdog_worker_should_ping(struct watchdog_core_data *wd_data)
 205 {
 206         struct watchdog_device *wdd = wd_data->wdd;
 207
 208         if (!wdd)
 209                 return false;
 210
 211         if (watchdog_active(wdd))
 212                 return true;
 213
 214         return watchdog_hw_running(wdd) && !watchdog_past_open_deadline(wd_data);
 215 }
 216
 217 static void watchdog_ping_work(struct kthread_work *work)
 218 {
 219         struct watchdog_core_data *wd_data;
 220
 221         wd_data = container_of(work, struct watchdog_core_data, work);
 222
 223         mutex_lock(&wd_data->lock);
 224         if (watchdog_worker_should_ping(wd_data))
 225                 __watchdog_ping(wd_data->wdd);
 226         mutex_unlock(&wd_data->lock);
 227 }
 228
 229 static enum hrtimer_restart watchdog_timer_expired(struct hrtimer *timer)
 230 {
 231         struct watchdog_core_data *wd_data;
 232
 233         wd_data = container_of(timer, struct watchdog_core_data, timer);
 234
 235         kthread_queue_work(watchdog_kworker, &wd_data->work);
 236         return HRTIMER_NORESTART;
 237 }
 238
 239 /*
 240  * watchdog_start - wrapper to start the watchdog
 241  * @wdd: The watchdog device to start
 242  *
 243  * Start the watchdog if it is not active and mark it active.
 244  * The caller must hold wd_data->lock.
 245  *
 246  * Return: 0 on success or a negative errno code for failure.
 247  */
 248 static int watchdog_start(struct watchdog_device *wdd)
 249 {
 250         struct watchdog_core_data *wd_data = wdd->wd_data;
 251         ktime_t started_at;
 252         int err;
 253
 254         if (watchdog_active(wdd))
 255                 return 0;
 256
 257         set_bit(_WDOG_KEEPALIVE, &wd_data->status);
 258
 259         started_at = ktime_get();
 260         if (watchdog_hw_running(wdd) && wdd->ops->ping) {
 261                 err = __watchdog_ping(wdd);
 262                 if (err == 0) {
 263                         set_bit(WDOG_ACTIVE, &wdd->status);
 264                         watchdog_hrtimer_pretimeout_start(wdd);
 265                 }
 266         } else {
 267                 err = wdd->ops->start(wdd);
 268                 trace_watchdog_start(wdd, err);
 269                 if (err == 0) {
 270                         set_bit(WDOG_ACTIVE, &wdd->status);
 271                         wd_data->last_keepalive = started_at;
 272                         wd_data->last_hw_keepalive = started_at;
 273                         watchdog_update_worker(wdd);
 274                         watchdog_hrtimer_pretimeout_start(wdd);
 275                 }
 276         }
 277
 278         return err;
 279 }
 280
 281 /*
 282  * watchdog_stop - wrapper to stop the watchdog
 283  * @wdd: The watchdog device to stop
 284  *
 285  * Stop the watchdog if it is still active and unmark it active.
 286  * If the 'nowayout' feature was set, the watchdog cannot be stopped.
 287  * The caller must hold wd_data->lock.
 288  *
 289  * Return: 0 on success or a negative errno code for failure.
 290  */
 291 static int watchdog_stop(struct watchdog_device *wdd)
 292 {
 293         int err = 0;
 294
 295         if (!watchdog_active(wdd))
 296                 return 0;
 297
 298         if (test_bit(WDOG_NO_WAY_OUT, &wdd->status)) {
 299                 pr_info("watchdog%d: nowayout prevents watchdog being stopped!\n",
 300                         wdd->id);
 301                 return -EBUSY;
 302         }
 303
 304         if (wdd->ops->stop) {
 305                 clear_bit(WDOG_HW_RUNNING, &wdd->status);
 306                 err = wdd->ops->stop(wdd);
 307                 trace_watchdog_stop(wdd, err);
 308         } else {
 309                 set_bit(WDOG_HW_RUNNING, &wdd->status);
 310         }
 311
 312         if (err == 0) {
 313                 clear_bit(WDOG_ACTIVE, &wdd->status);
 314                 watchdog_update_worker(wdd);
 315                 watchdog_hrtimer_pretimeout_stop(wdd);
 316         }
 317
 318         return err;
 319 }
 320
 321 /*
 322  * watchdog_get_status - wrapper to get the watchdog status
 323  * @wdd: The watchdog device to get the status from
 324  *
 325  * Get the watchdog's status flags.
 326  * The caller must hold wd_data->lock.
 327  *
 328  * Return: watchdog's status flags.
 329  */
 330 static unsigned int watchdog_get_status(struct watchdog_device *wdd)
 331 {
 332         struct watchdog_core_data *wd_data = wdd->wd_data;
 333         unsigned int status;
 334
 335         if (wdd->ops->status)
 336                 status = wdd->ops->status(wdd);
 337         else
 338                 status = wdd->bootstatus & (WDIOF_CARDRESET |
 339                                             WDIOF_OVERHEAT |
 340                                             WDIOF_FANFAULT |
 341                                             WDIOF_EXTERN1 |
 342                                             WDIOF_EXTERN2 |
 343                                             WDIOF_POWERUNDER |
 344                                             WDIOF_POWEROVER);
 345
 346         if (test_bit(_WDOG_ALLOW_RELEASE, &wd_data->status))
 347                 status |= WDIOF_MAGICCLOSE;
 348
 349         if (test_and_clear_bit(_WDOG_KEEPALIVE, &wd_data->status))
 350                 status |= WDIOF_KEEPALIVEPING;
 351
 352         if (IS_ENABLED(CONFIG_WATCHDOG_HRTIMER_PRETIMEOUT))
 353                 status |= WDIOF_PRETIMEOUT;
 354
 355         return status;
 356 }
 357
 358 /*
 359  * watchdog_set_timeout - set the watchdog timer timeout
 360  * @wdd:        The watchdog device to set the timeout for
 361  * @timeout:    Timeout to set in seconds
 362  *
 363  * The caller must hold wd_data->lock.
 364  *
 365  * Return: 0 if successful, error otherwise.
 366  */
 367 static int watchdog_set_timeout(struct watchdog_device *wdd,
 368                                                         unsigned int timeout)
 369 {
 370         int err = 0;
 371
 372         if (!(wdd->info->options & WDIOF_SETTIMEOUT))
 373                 return -EOPNOTSUPP;
 374
 375         if (watchdog_timeout_invalid(wdd, timeout))
 376                 return -EINVAL;
 377
 378         if (wdd->ops->set_timeout) {
 379                 err = wdd->ops->set_timeout(wdd, timeout);
 380                 trace_watchdog_set_timeout(wdd, timeout, err);
 381         } else {
 382                 wdd->timeout = timeout;
 383                 /* Disable pretimeout if it doesn't fit the new timeout */
 384                 if (wdd->pretimeout >= wdd->timeout)
 385                         wdd->pretimeout = 0;
 386         }
 387
 388         watchdog_update_worker(wdd);
 389
 390         return err;
 391 }
 392
 393 /*
 394  * watchdog_set_pretimeout - set the watchdog timer pretimeout
 395  * @wdd:        The watchdog device to set the timeout for
 396  * @timeout:    pretimeout to set in seconds
 397  *
 398  * Return: 0 if successful, error otherwise.
 399  */
 400 static int watchdog_set_pretimeout(struct watchdog_device *wdd,
 401                                    unsigned int timeout)
 402 {
 403         int err = 0;
 404
 405         if (!watchdog_have_pretimeout(wdd))
 406                 return -EOPNOTSUPP;
 407
 408         if (watchdog_pretimeout_invalid(wdd, timeout))
 409                 return -EINVAL;
 410
 411         if (wdd->ops->set_pretimeout && (wdd->info->options & WDIOF_PRETIMEOUT))
 412                 err = wdd->ops->set_pretimeout(wdd, timeout);
 413         else
 414                 wdd->pretimeout = timeout;
 415
 416         return err;
 417 }
 418
 419 /*
 420  * watchdog_get_timeleft - wrapper to get the time left before a reboot
 421  * @wdd:        The watchdog device to get the remaining time from
 422  * @timeleft:   The time that's left
 423  *
 424  * Get the time before a watchdog will reboot (if not pinged).
 425  * The caller must hold wd_data->lock.
 426  *
 427  * Return: 0 if successful, error otherwise.
 428  */
 429 static int watchdog_get_timeleft(struct watchdog_device *wdd,
 430                                                         unsigned int *timeleft)
 431 {
 432         *timeleft = 0;
 433
 434         if (!wdd->ops->get_timeleft)
 435                 return -EOPNOTSUPP;
 436
 437         *timeleft = wdd->ops->get_timeleft(wdd);
 438
 439         return 0;
 440 }
 441
 442 #ifdef CONFIG_WATCHDOG_SYSFS
 443 static ssize_t nowayout_show(struct device *dev, struct device_attribute *attr,
 444                                 char *buf)
 445 {
 446         struct watchdog_device *wdd = dev_get_drvdata(dev);
 447
 448         return sysfs_emit(buf, "%d\n", !!test_bit(WDOG_NO_WAY_OUT,
 449                                                   &wdd->status));
 450 }
 451
 452 static ssize_t nowayout_store(struct device *dev, struct device_attribute *attr,
 453                                 const char *buf, size_t len)
 454 {
 455         struct watchdog_device *wdd = dev_get_drvdata(dev);
 456         unsigned int value;
 457         int ret;
 458
 459         ret = kstrtouint(buf, 0, &value);
 460         if (ret)
 461                 return ret;
 462         if (value > 1)
 463                 return -EINVAL;
 464         /* nowayout cannot be disabled once set */
 465         if (test_bit(WDOG_NO_WAY_OUT, &wdd->status) && !value)
 466                 return -EPERM;
 467         watchdog_set_nowayout(wdd, value);
 468         return len;
 469 }
 470 static DEVICE_ATTR_RW(nowayout);
 471
 472 static ssize_t status_show(struct device *dev, struct device_attribute *attr,
 473                                 char *buf)
 474 {
 475         struct watchdog_device *wdd = dev_get_drvdata(dev);
 476         struct watchdog_core_data *wd_data = wdd->wd_data;
 477         unsigned int status;
 478
 479         mutex_lock(&wd_data->lock);
 480         status = watchdog_get_status(wdd);
 481         mutex_unlock(&wd_data->lock);
 482
 483         return sysfs_emit(buf, "0x%x\n", status);
 484 }
 485 static DEVICE_ATTR_RO(status);
 486
 487 static ssize_t bootstatus_show(struct device *dev,
 488                                 struct device_attribute *attr, char *buf)
 489 {
 490         struct watchdog_device *wdd = dev_get_drvdata(dev);
 491
 492         return sysfs_emit(buf, "%u\n", wdd->bootstatus);
 493 }
 494 static DEVICE_ATTR_RO(bootstatus);
 495
 496 static ssize_t timeleft_show(struct device *dev, struct device_attribute *attr,
 497                                 char *buf)
 498 {
 499         struct watchdog_device *wdd = dev_get_drvdata(dev);
 500         struct watchdog_core_data *wd_data = wdd->wd_data;
 501         ssize_t status;
 502         unsigned int val;
 503
 504         mutex_lock(&wd_data->lock);
 505         status = watchdog_get_timeleft(wdd, &val);
 506         mutex_unlock(&wd_data->lock);
 507         if (!status)
 508                 status = sysfs_emit(buf, "%u\n", val);
 509
 510         return status;
 511 }
 512 static DEVICE_ATTR_RO(timeleft);
 513
 514 static ssize_t timeout_show(struct device *dev, struct device_attribute *attr,
 515                                 char *buf)
 516 {
 517         struct watchdog_device *wdd = dev_get_drvdata(dev);
 518
 519         return sysfs_emit(buf, "%u\n", wdd->timeout);
 520 }
 521 static DEVICE_ATTR_RO(timeout);
 522
 523 static ssize_t min_timeout_show(struct device *dev,
 524                                 struct device_attribute *attr, char *buf)
 525 {
 526         struct watchdog_device *wdd = dev_get_drvdata(dev);
 527
 528         return sysfs_emit(buf, "%u\n", wdd->min_timeout);
 529 }
 530 static DEVICE_ATTR_RO(min_timeout);
 531
 532 static ssize_t max_timeout_show(struct device *dev,
 533                                 struct device_attribute *attr, char *buf)
 534 {
 535         struct watchdog_device *wdd = dev_get_drvdata(dev);
 536
 537         return sysfs_emit(buf, "%u\n", wdd->max_timeout);
 538 }
 539 static DEVICE_ATTR_RO(max_timeout);
 540
 541 static ssize_t pretimeout_show(struct device *dev,
 542                                struct device_attribute *attr, char *buf)
 543 {
 544         struct watchdog_device *wdd = dev_get_drvdata(dev);
 545
 546         return sysfs_emit(buf, "%u\n", wdd->pretimeout);
 547 }
 548 static DEVICE_ATTR_RO(pretimeout);
 549
 550 static ssize_t fw_version_show(struct device *dev, struct device_attribute *attr,
 551                                char *buf)
 552 {
 553         struct watchdog_device *wdd = dev_get_drvdata(dev);
 554
 555         return sysfs_emit(buf, "%d\n", wdd->info->firmware_version);
 556 }
 557 static DEVICE_ATTR_RO(fw_version);
 558
 559 static ssize_t identity_show(struct device *dev, struct device_attribute *attr,
 560                                 char *buf)
 561 {
 562         struct watchdog_device *wdd = dev_get_drvdata(dev);
 563
 564         return sysfs_emit(buf, "%s\n", wdd->info->identity);
 565 }
 566 static DEVICE_ATTR_RO(identity);
 567
 568 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
 569                                 char *buf)
 570 {
 571         struct watchdog_device *wdd = dev_get_drvdata(dev);
 572
 573         if (watchdog_active(wdd))
 574                 return sysfs_emit(buf, "active\n");
 575
 576         return sysfs_emit(buf, "inactive\n");
 577 }
 578 static DEVICE_ATTR_RO(state);
 579
 580 static ssize_t pretimeout_available_governors_show(struct device *dev,
 581                                    struct device_attribute *attr, char *buf)
 582 {
 583         return watchdog_pretimeout_available_governors_get(buf);
 584 }
 585 static DEVICE_ATTR_RO(pretimeout_available_governors);
 586
 587 static ssize_t pretimeout_governor_show(struct device *dev,
 588                                         struct device_attribute *attr,
 589                                         char *buf)
 590 {
 591         struct watchdog_device *wdd = dev_get_drvdata(dev);
 592
 593         return watchdog_pretimeout_governor_get(wdd, buf);
 594 }
 595
 596 static ssize_t pretimeout_governor_store(struct device *dev,
 597                                          struct device_attribute *attr,
 598                                          const char *buf, size_t count)
 599 {
 600         struct watchdog_device *wdd = dev_get_drvdata(dev);
 601         int ret = watchdog_pretimeout_governor_set(wdd, buf);
 602
 603         if (!ret)
 604                 ret = count;
 605
 606         return ret;
 607 }
 608 static DEVICE_ATTR_RW(pretimeout_governor);
 609
 610 static umode_t wdt_is_visible(struct kobject *kobj, struct attribute *attr,
 611                                 int n)
 612 {
 613         struct device *dev = kobj_to_dev(kobj);
 614         struct watchdog_device *wdd = dev_get_drvdata(dev);
 615         umode_t mode = attr->mode;
 616
 617         if (attr == &dev_attr_timeleft.attr && !wdd->ops->get_timeleft)
 618                 mode = 0;
 619         else if (attr == &dev_attr_pretimeout.attr && !watchdog_have_pretimeout(wdd))
 620                 mode = 0;
 621         else if ((attr == &dev_attr_pretimeout_governor.attr ||
 622                   attr == &dev_attr_pretimeout_available_governors.attr) &&
 623                  (!watchdog_have_pretimeout(wdd) || !IS_ENABLED(CONFIG_WATCHDOG_PRETIMEOUT_GOV)))
 624                 mode = 0;
 625
 626         return mode;
 627 }
 628 static struct attribute *wdt_attrs[] = {
 629         &dev_attr_state.attr,
 630         &dev_attr_fw_version.attr,
 631         &dev_attr_identity.attr,
 632         &dev_attr_timeout.attr,
 633         &dev_attr_min_timeout.attr,
 634         &dev_attr_max_timeout.attr,
 635         &dev_attr_pretimeout.attr,
 636         &dev_attr_timeleft.attr,
 637         &dev_attr_bootstatus.attr,
 638         &dev_attr_status.attr,
 639         &dev_attr_nowayout.attr,
 640         &dev_attr_pretimeout_governor.attr,
 641         &dev_attr_pretimeout_available_governors.attr,
 642         NULL,
 643 };
 644
 645 static const struct attribute_group wdt_group = {
 646         .attrs = wdt_attrs,
 647         .is_visible = wdt_is_visible,
 648 };
 649 __ATTRIBUTE_GROUPS(wdt);
 650 #else
 651 #define wdt_groups      NULL
 652 #endif
 653
 654 /*
 655  * watchdog_ioctl_op - call the watchdog drivers ioctl op if defined
 656  * @wdd: The watchdog device to do the ioctl on
 657  * @cmd: Watchdog command
 658  * @arg: Argument pointer
 659  *
 660  * The caller must hold wd_data->lock.
 661  *
 662  * Return: 0 if successful, error otherwise.
 663  */
 664 static int watchdog_ioctl_op(struct watchdog_device *wdd, unsigned int cmd,
 665                                                         unsigned long arg)
 666 {
 667         if (!wdd->ops->ioctl)
 668                 return -ENOIOCTLCMD;
 669
 670         return wdd->ops->ioctl(wdd, cmd, arg);
 671 }
 672
 673 /*
 674  * watchdog_write - writes to the watchdog
 675  * @file:       File from VFS
 676  * @data:       User address of data
 677  * @len:        Length of data
 678  * @ppos:       Pointer to the file offset
 679  *
 680  * A write to a watchdog device is defined as a keepalive ping.
 681  * Writing the magic 'V' sequence allows the next close to turn
 682  * off the watchdog (if 'nowayout' is not set).
 683  *
 684  * Return: @len if successful, error otherwise.
 685  */
 686 static ssize_t watchdog_write(struct file *file, const char __user *data,
 687                                                 size_t len, loff_t *ppos)
 688 {
 689         struct watchdog_core_data *wd_data = file->private_data;
 690         struct watchdog_device *wdd;
 691         int err;
 692         size_t i;
 693         char c;
 694
 695         if (len == 0)
 696                 return 0;
 697
 698         /*
 699          * Note: just in case someone wrote the magic character
 700          * five months ago...
 701          */
 702         clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
 703
 704         /* scan to see whether or not we got the magic character */
 705         for (i = 0; i != len; i++) {
 706                 if (get_user(c, data + i))
 707                         return -EFAULT;
 708                 if (c == 'V')
 709                         set_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
 710         }
 711
 712         /* someone wrote to us, so we send the watchdog a keepalive ping */
 713
 714         err = -ENODEV;
 715         mutex_lock(&wd_data->lock);
 716         wdd = wd_data->wdd;
 717         if (wdd)
 718                 err = watchdog_ping(wdd);
 719         mutex_unlock(&wd_data->lock);
 720
 721         if (err < 0)
 722                 return err;
 723
 724         return len;
 725 }
 726
 727 /*
 728  * watchdog_ioctl - handle the different ioctl's for the watchdog device
 729  * @file:       File handle to the device
 730  * @cmd:        Watchdog command
 731  * @arg:        Argument pointer
 732  *
 733  * The watchdog API defines a common set of functions for all watchdogs
 734  * according to their available features.
 735  *
 736  * Return: 0 if successful, error otherwise.
 737  */
 738
 739 static long watchdog_ioctl(struct file *file, unsigned int cmd,
 740                                                         unsigned long arg)
 741 {
 742         struct watchdog_core_data *wd_data = file->private_data;
 743         void __user *argp = (void __user *)arg;
 744         struct watchdog_device *wdd;
 745         int __user *p = argp;
 746         unsigned int val;
 747         int err;
 748
 749         mutex_lock(&wd_data->lock);
 750
 751         wdd = wd_data->wdd;
 752         if (!wdd) {
 753                 err = -ENODEV;
 754                 goto out_ioctl;
 755         }
 756
 757         err = watchdog_ioctl_op(wdd, cmd, arg);
 758         if (err != -ENOIOCTLCMD)
 759                 goto out_ioctl;
 760
 761         switch (cmd) {
 762         case WDIOC_GETSUPPORT:
 763                 err = copy_to_user(argp, wdd->info,
 764                         sizeof(struct watchdog_info)) ? -EFAULT : 0;
 765                 break;
 766         case WDIOC_GETSTATUS:
 767                 val = watchdog_get_status(wdd);
 768                 err = put_user(val, p);
 769                 break;
 770         case WDIOC_GETBOOTSTATUS:
 771                 err = put_user(wdd->bootstatus, p);
 772                 break;
 773         case WDIOC_SETOPTIONS:
 774                 if (get_user(val, p)) {
 775                         err = -EFAULT;
 776                         break;
 777                 }
 778                 if (val & WDIOS_DISABLECARD) {
 779                         err = watchdog_stop(wdd);
 780                         if (err < 0)
 781                                 break;
 782                 }
 783                 if (val & WDIOS_ENABLECARD)
 784                         err = watchdog_start(wdd);
 785                 break;
 786         case WDIOC_KEEPALIVE:
 787                 if (!(wdd->info->options & WDIOF_KEEPALIVEPING)) {
 788                         err = -EOPNOTSUPP;
 789                         break;
 790                 }
 791                 err = watchdog_ping(wdd);
 792                 break;
 793         case WDIOC_SETTIMEOUT:
 794                 if (get_user(val, p)) {
 795                         err = -EFAULT;
 796                         break;
 797                 }
 798                 err = watchdog_set_timeout(wdd, val);
 799                 if (err < 0)
 800                         break;
 801                 /* If the watchdog is active then we send a keepalive ping
 802                  * to make sure that the watchdog keep's running (and if
 803                  * possible that it takes the new timeout) */
 804                 err = watchdog_ping(wdd);
 805                 if (err < 0)
 806                         break;
 807                 fallthrough;
 808         case WDIOC_GETTIMEOUT:
 809                 /* timeout == 0 means that we don't know the timeout */
 810                 if (wdd->timeout == 0) {
 811                         err = -EOPNOTSUPP;
 812                         break;
 813                 }
 814                 err = put_user(wdd->timeout, p);
 815                 break;
 816         case WDIOC_GETTIMELEFT:
 817                 err = watchdog_get_timeleft(wdd, &val);
 818                 if (err < 0)
 819                         break;
 820                 err = put_user(val, p);
 821                 break;
 822         case WDIOC_SETPRETIMEOUT:
 823                 if (get_user(val, p)) {
 824                         err = -EFAULT;
 825                         break;
 826                 }
 827                 err = watchdog_set_pretimeout(wdd, val);
 828                 break;
 829         case WDIOC_GETPRETIMEOUT:
 830                 err = put_user(wdd->pretimeout, p);
 831                 break;
 832         default:
 833                 err = -ENOTTY;
 834                 break;
 835         }
 836
 837 out_ioctl:
 838         mutex_unlock(&wd_data->lock);
 839         return err;
 840 }
 841
 842 /*
 843  * watchdog_open - open the /dev/watchdog* devices
 844  * @inode:      Inode of device
 845  * @file:       File handle to device
 846  *
 847  * When the /dev/watchdog* device gets opened, we start the watchdog.
 848  * Watch out: the /dev/watchdog device is single open, so we make sure
 849  * it can only be opened once.
 850  *
 851  * Return: 0 if successful, error otherwise.
 852  */
 853 static int watchdog_open(struct inode *inode, struct file *file)
 854 {
 855         struct watchdog_core_data *wd_data;
 856         struct watchdog_device *wdd;
 857         bool hw_running;
 858         int err;
 859
 860         /* Get the corresponding watchdog device */
 861         if (imajor(inode) == MISC_MAJOR)
 862                 wd_data = old_wd_data;
 863         else
 864                 wd_data = container_of(inode->i_cdev, struct watchdog_core_data,
 865                                        cdev);
 866
 867         /* the watchdog is single open! */
 868         if (test_and_set_bit(_WDOG_DEV_OPEN, &wd_data->status))
 869                 return -EBUSY;
 870
 871         wdd = wd_data->wdd;
 872
 873         /*
 874          * If the /dev/watchdog device is open, we don't want the module
 875          * to be unloaded.
 876          */
 877         hw_running = watchdog_hw_running(wdd);
 878         if (!hw_running && !try_module_get(wdd->ops->owner)) {
 879                 err = -EBUSY;
 880                 goto out_clear;
 881         }
 882
 883         err = watchdog_start(wdd);
 884         if (err < 0)
 885                 goto out_mod;
 886
 887         file->private_data = wd_data;
 888
 889         if (!hw_running)
 890                 get_device(&wd_data->dev);
 891
 892         /*
 893          * open_timeout only applies for the first open from
 894          * userspace. Set open_deadline to infinity so that the kernel
 895          * will take care of an always-running hardware watchdog in
 896          * case the device gets magic-closed or WDIOS_DISABLECARD is
 897          * applied.
 898          */
 899         wd_data->open_deadline = KTIME_MAX;
 900
 901         /* dev/watchdog is a virtual (and thus non-seekable) filesystem */
 902         return stream_open(inode, file);
 903
 904 out_mod:
 905         module_put(wd_data->wdd->ops->owner);
 906 out_clear:
 907         clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
 908         return err;
 909 }
 910
 911 static void watchdog_core_data_release(struct device *dev)
 912 {
 913         struct watchdog_core_data *wd_data;
 914
 915         wd_data = container_of(dev, struct watchdog_core_data, dev);
 916
 917         kfree(wd_data);
 918 }
 919
 920 /*
 921  * watchdog_release - release the watchdog device
 922  * @inode:      Inode of device
 923  * @file:       File handle to device
 924  *
 925  * This is the code for when /dev/watchdog gets closed. We will only
 926  * stop the watchdog when we have received the magic char (and nowayout
 927  * was not set), else the watchdog will keep running.
 928  *
 929  * Always returns 0.
 930  */
 931 static int watchdog_release(struct inode *inode, struct file *file)
 932 {
 933         struct watchdog_core_data *wd_data = file->private_data;
 934         struct watchdog_device *wdd;
 935         int err = -EBUSY;
 936         bool running;
 937
 938         mutex_lock(&wd_data->lock);
 939
 940         wdd = wd_data->wdd;
 941         if (!wdd)
 942                 goto done;
 943
 944         /*
 945          * We only stop the watchdog if we received the magic character
 946          * or if WDIOF_MAGICCLOSE is not set. If nowayout was set then
 947          * watchdog_stop will fail.
 948          */
 949         if (!watchdog_active(wdd))
 950                 err = 0;
 951         else if (test_and_clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status) ||
 952                  !(wdd->info->options & WDIOF_MAGICCLOSE))
 953                 err = watchdog_stop(wdd);
 954
 955         /* If the watchdog was not stopped, send a keepalive ping */
 956         if (err < 0) {
 957                 pr_crit("watchdog%d: watchdog did not stop!\n", wdd->id);
 958                 watchdog_ping(wdd);
 959         }
 960
 961         watchdog_update_worker(wdd);
 962
 963         /* make sure that /dev/watchdog can be re-opened */
 964         clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
 965
 966 done:
 967         running = wdd && watchdog_hw_running(wdd);
 968         mutex_unlock(&wd_data->lock);
 969         /*
 970          * Allow the owner module to be unloaded again unless the watchdog
 971          * is still running. If the watchdog is still running, it can not
 972          * be stopped, and its driver must not be unloaded.
 973          */
 974         if (!running) {
 975                 module_put(wd_data->cdev.owner);
 976                 put_device(&wd_data->dev);
 977         }
 978         return 0;
 979 }
 980
 981 static const struct file_operations watchdog_fops = {
 982         .owner          = THIS_MODULE,
 983         .write          = watchdog_write,
 984         .unlocked_ioctl = watchdog_ioctl,
 985         .compat_ioctl   = compat_ptr_ioctl,
 986         .open           = watchdog_open,
 987         .release        = watchdog_release,
 988 };
 989
 990 static struct miscdevice watchdog_miscdev = {
 991         .minor          = WATCHDOG_MINOR,
 992         .name           = "watchdog",
 993         .fops           = &watchdog_fops,
 994 };
 995
 996 static struct class watchdog_class = {
 997         .name =         "watchdog",
 998         .owner =        THIS_MODULE,
 999         .dev_groups =   wdt_groups,
1000 };
1001
1002 /*
1003  * watchdog_cdev_register - register watchdog character device
1004  * @wdd: Watchdog device
1005  *
1006  * Register a watchdog character device including handling the legacy
1007  * /dev/watchdog node. /dev/watchdog is actually a miscdevice and
1008  * thus we set it up like that.
1009  *
1010  * Return: 0 if successful, error otherwise.
1011  */
1012 static int watchdog_cdev_register(struct watchdog_device *wdd)
1013 {
1014         struct watchdog_core_data *wd_data;
1015         int err;
1016
1017         wd_data = kzalloc(sizeof(struct watchdog_core_data), GFP_KERNEL);
1018         if (!wd_data)
1019                 return -ENOMEM;
1020         mutex_init(&wd_data->lock);
1021
1022         wd_data->wdd = wdd;
1023         wdd->wd_data = wd_data;
1024
1025         if (IS_ERR_OR_NULL(watchdog_kworker)) {
1026                 kfree(wd_data);
1027                 return -ENODEV;
1028         }
1029
1030         device_initialize(&wd_data->dev);
1031         wd_data->dev.devt = MKDEV(MAJOR(watchdog_devt), wdd->id);
1032         wd_data->dev.class = &watchdog_class;
1033         wd_data->dev.parent = wdd->parent;
1034         wd_data->dev.groups = wdd->groups;
1035         wd_data->dev.release = watchdog_core_data_release;
1036         dev_set_drvdata(&wd_data->dev, wdd);
1037         err = dev_set_name(&wd_data->dev, "watchdog%d", wdd->id);
1038         if (err) {
1039                 put_device(&wd_data->dev);
1040                 return err;
1041         }
1042
1043         kthread_init_work(&wd_data->work, watchdog_ping_work);
1044         hrtimer_init(&wd_data->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
1045         wd_data->timer.function = watchdog_timer_expired;
1046         watchdog_hrtimer_pretimeout_init(wdd);
1047
1048         if (wdd->id == 0) {
1049                 old_wd_data = wd_data;
1050                 watchdog_miscdev.parent = wdd->parent;
1051                 err = misc_register(&watchdog_miscdev);
1052                 if (err != 0) {
1053                         pr_err("%s: cannot register miscdev on minor=%d (err=%d).\n",
1054                                 wdd->info->identity, WATCHDOG_MINOR, err);
1055                         if (err == -EBUSY)
1056                                 pr_err("%s: a legacy watchdog module is probably present.\n",
1057                                         wdd->info->identity);
1058                         old_wd_data = NULL;
1059                         put_device(&wd_data->dev);
1060                         return err;
1061                 }
1062         }
1063
1064         /* Fill in the data structures */
1065         cdev_init(&wd_data->cdev, &watchdog_fops);
1066
1067         /* Add the device */
1068         err = cdev_device_add(&wd_data->cdev, &wd_data->dev);
1069         if (err) {
1070                 pr_err("watchdog%d unable to add device %d:%d\n",
1071                         wdd->id,  MAJOR(watchdog_devt), wdd->id);
1072                 if (wdd->id == 0) {
1073                         misc_deregister(&watchdog_miscdev);
1074                         old_wd_data = NULL;
1075                 }
1076                 put_device(&wd_data->dev);
1077                 return err;
1078         }
1079
1080         wd_data->cdev.owner = wdd->ops->owner;
1081
1082         /* Record time of most recent heartbeat as 'just before now'. */
1083         wd_data->last_hw_keepalive = ktime_sub(ktime_get(), 1);
1084         watchdog_set_open_deadline(wd_data);
1085
1086         /*
1087          * If the watchdog is running, prevent its driver from being unloaded,
1088          * and schedule an immediate ping.
1089          */
1090         if (watchdog_hw_running(wdd)) {
1091                 __module_get(wdd->ops->owner);
1092                 get_device(&wd_data->dev);
1093                 if (handle_boot_enabled)
1094                         hrtimer_start(&wd_data->timer, 0,
1095                                       HRTIMER_MODE_REL_HARD);
1096                 else
1097                         pr_info("watchdog%d running and kernel based pre-userspace handler disabled\n",
1098                                 wdd->id);
1099         }
1100
1101         return 0;
1102 }
1103
1104 /*
1105  * watchdog_cdev_unregister - unregister watchdog character device
1106  * @wdd: Watchdog device
1107  *
1108  * Unregister watchdog character device and if needed the legacy
1109  * /dev/watchdog device.
1110  */
1111 static void watchdog_cdev_unregister(struct watchdog_device *wdd)
1112 {
1113         struct watchdog_core_data *wd_data = wdd->wd_data;
1114
1115         cdev_device_del(&wd_data->cdev, &wd_data->dev);
1116         if (wdd->id == 0) {
1117                 misc_deregister(&watchdog_miscdev);
1118                 old_wd_data = NULL;
1119         }
1120
1121         if (watchdog_active(wdd) &&
1122             test_bit(WDOG_STOP_ON_UNREGISTER, &wdd->status)) {
1123                 watchdog_stop(wdd);
1124         }
1125
1126         watchdog_hrtimer_pretimeout_stop(wdd);
1127
1128         mutex_lock(&wd_data->lock);
1129         wd_data->wdd = NULL;
1130         wdd->wd_data = NULL;
1131         mutex_unlock(&wd_data->lock);
1132
1133         hrtimer_cancel(&wd_data->timer);
1134         kthread_cancel_work_sync(&wd_data->work);
1135
1136         put_device(&wd_data->dev);
1137 }
1138
1139 /**
1140  * watchdog_dev_register - register a watchdog device
1141  * @wdd: Watchdog device
1142  *
1143  * Register a watchdog device including handling the legacy
1144  * /dev/watchdog node. /dev/watchdog is actually a miscdevice and
1145  * thus we set it up like that.
1146  *
1147  * Return: 0 if successful, error otherwise.
1148  */
1149 int watchdog_dev_register(struct watchdog_device *wdd)
1150 {
1151         int ret;
1152
1153         ret = watchdog_cdev_register(wdd);
1154         if (ret)
1155                 return ret;
1156
1157         ret = watchdog_register_pretimeout(wdd);
1158         if (ret)
1159                 watchdog_cdev_unregister(wdd);
1160
1161         return ret;
1162 }
1163
1164 /**
1165  * watchdog_dev_unregister - unregister a watchdog device
1166  * @wdd: watchdog device
1167  *
1168  * Unregister watchdog device and if needed the legacy
1169  * /dev/watchdog device.
1170  */
1171 void watchdog_dev_unregister(struct watchdog_device *wdd)
1172 {
1173         watchdog_unregister_pretimeout(wdd);
1174         watchdog_cdev_unregister(wdd);
1175 }
1176
1177 /**
1178  * watchdog_set_last_hw_keepalive - set last HW keepalive time for watchdog
1179  * @wdd:                Watchdog device
1180  * @last_ping_ms:       Time since last HW heartbeat
1181  *
1182  * Adjusts the last known HW keepalive time for a watchdog timer.
1183  * This is needed if the watchdog is already running when the probe
1184  * function is called, and it can't be pinged immediately. This
1185  * function must be called immediately after watchdog registration,
1186  * and min_hw_heartbeat_ms must be set for this to be useful.
1187  *
1188  * Return: 0 if successful, error otherwise.
1189  */
1190 int watchdog_set_last_hw_keepalive(struct watchdog_device *wdd,
1191                                    unsigned int last_ping_ms)
1192 {
1193         struct watchdog_core_data *wd_data;
1194         ktime_t now;
1195
1196         if (!wdd)
1197                 return -EINVAL;
1198
1199         wd_data = wdd->wd_data;
1200
1201         now = ktime_get();
1202
1203         wd_data->last_hw_keepalive = ktime_sub(now, ms_to_ktime(last_ping_ms));
1204
1205         if (watchdog_hw_running(wdd) && handle_boot_enabled)
1206                 return __watchdog_ping(wdd);
1207
1208         return 0;
1209 }
1210 EXPORT_SYMBOL_GPL(watchdog_set_last_hw_keepalive);
1211
1212 /**
1213  * watchdog_dev_init - init dev part of watchdog core
1214  *
1215  * Allocate a range of chardev nodes to use for watchdog devices.
1216  *
1217  * Return: 0 if successful, error otherwise.
1218  */
1219 int __init watchdog_dev_init(void)
1220 {
1221         int err;
1222
1223         watchdog_kworker = kthread_create_worker(0, "watchdogd");
1224         if (IS_ERR(watchdog_kworker)) {
1225                 pr_err("Failed to create watchdog kworker\n");
1226                 return PTR_ERR(watchdog_kworker);
1227         }
1228         sched_set_fifo(watchdog_kworker->task);
1229
1230         err = class_register(&watchdog_class);
1231         if (err < 0) {
1232                 pr_err("couldn't register class\n");
1233                 goto err_register;
1234         }
1235
1236         err = alloc_chrdev_region(&watchdog_devt, 0, MAX_DOGS, "watchdog");
1237         if (err < 0) {
1238                 pr_err("watchdog: unable to allocate char dev region\n");
1239                 goto err_alloc;
1240         }
1241
1242         return 0;
1243
1244 err_alloc:
1245         class_unregister(&watchdog_class);
1246 err_register:
1247         kthread_destroy_worker(watchdog_kworker);
1248         return err;
1249 }
1250
1251 /**
1252  * watchdog_dev_exit - exit dev part of watchdog core
1253  *
1254  * Release the range of chardev nodes used for watchdog devices.
1255  */
1256 void __exit watchdog_dev_exit(void)
1257 {
1258         unregister_chrdev_region(watchdog_devt, MAX_DOGS);
1259         class_unregister(&watchdog_class);
1260         kthread_destroy_worker(watchdog_kworker);
1261 }
1262
1263 int watchdog_dev_suspend(struct watchdog_device *wdd)
1264 {
1265         struct watchdog_core_data *wd_data = wdd->wd_data;
1266         int ret = 0;
1267
1268         if (!wdd->wd_data)
1269                 return -ENODEV;
1270
1271         /* ping for the last time before suspend */
1272         mutex_lock(&wd_data->lock);
1273         if (watchdog_worker_should_ping(wd_data))
1274                 ret = __watchdog_ping(wd_data->wdd);
1275         mutex_unlock(&wd_data->lock);
1276
1277         if (ret)
1278                 return ret;
1279
1280         /*
1281          * make sure that watchdog worker will not kick in when the wdog is
1282          * suspended
1283          */
1284         hrtimer_cancel(&wd_data->timer);
1285         kthread_cancel_work_sync(&wd_data->work);
1286
1287         return 0;
1288 }
1289
1290 int watchdog_dev_resume(struct watchdog_device *wdd)
1291 {
1292         struct watchdog_core_data *wd_data = wdd->wd_data;
1293         int ret = 0;
1294
1295         if (!wdd->wd_data)
1296                 return -ENODEV;
1297
1298         /*
1299          * __watchdog_ping will also retrigger hrtimer and therefore restore the
1300          * ping worker if needed.
1301          */
1302         mutex_lock(&wd_data->lock);
1303         if (watchdog_worker_should_ping(wd_data))
1304                 ret = __watchdog_ping(wd_data->wdd);
1305         mutex_unlock(&wd_data->lock);
1306
1307         return ret;
1308 }
1309
1310 module_param(handle_boot_enabled, bool, 0444);
1311 MODULE_PARM_DESC(handle_boot_enabled,
1312         "Watchdog core auto-updates boot enabled watchdogs before userspace takes over (default="
1313         __MODULE_STRING(IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED)) ")");
1314
1315 module_param(open_timeout, uint, 0644);
1316 MODULE_PARM_DESC(open_timeout,
1317         "Maximum time (in seconds, 0 means infinity) for userspace to take over a running watchdog (default="
1318         __MODULE_STRING(CONFIG_WATCHDOG_OPEN_TIMEOUT) ")");