watchdog.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780
  1. // SPDX-License-Identifier: GPL-2.0
  2. /*
  3. * Detect hard and soft lockups on a system
  4. *
  5. * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
  6. *
  7. * Note: Most of this code is borrowed heavily from the original softlockup
  8. * detector, so thanks to Ingo for the initial implementation.
  9. * Some chunks also taken from the old x86-specific nmi watchdog code, thanks
  10. * to those contributors as well.
  11. */
  12. #define pr_fmt(fmt) "watchdog: " fmt
  13. #include <linux/mm.h>
  14. #include <linux/cpu.h>
  15. #include <linux/nmi.h>
  16. #include <linux/init.h>
  17. #include <linux/module.h>
  18. #include <linux/sysctl.h>
  19. #include <linux/tick.h>
  20. #include <linux/sched/clock.h>
  21. #include <linux/sched/debug.h>
  22. #include <linux/sched/isolation.h>
  23. #include <linux/stop_machine.h>
  24. #include <asm/irq_regs.h>
  25. #include <linux/kvm_para.h>
  26. static DEFINE_MUTEX(watchdog_mutex);
  27. #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
  28. # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED)
  29. # define NMI_WATCHDOG_DEFAULT 1
  30. #else
  31. # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED)
  32. # define NMI_WATCHDOG_DEFAULT 0
  33. #endif
  34. unsigned long __read_mostly watchdog_enabled;
  35. int __read_mostly watchdog_user_enabled = 1;
  36. int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT;
  37. int __read_mostly soft_watchdog_user_enabled = 1;
  38. int __read_mostly watchdog_thresh = 10;
  39. int __read_mostly nmi_watchdog_available;
  40. struct cpumask watchdog_allowed_mask __read_mostly;
  41. struct cpumask watchdog_cpumask __read_mostly;
  42. unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
  43. #ifdef CONFIG_HARDLOCKUP_DETECTOR
  44. /*
  45. * Should we panic when a soft-lockup or hard-lockup occurs:
  46. */
  47. unsigned int __read_mostly hardlockup_panic =
  48. CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
  49. /*
  50. * We may not want to enable hard lockup detection by default in all cases,
  51. * for example when running the kernel as a guest on a hypervisor. In these
  52. * cases this function can be called to disable hard lockup detection. This
  53. * function should only be executed once by the boot processor before the
  54. * kernel command line parameters are parsed, because otherwise it is not
  55. * possible to override this in hardlockup_panic_setup().
  56. */
  57. void __init hardlockup_detector_disable(void)
  58. {
  59. nmi_watchdog_user_enabled = 0;
  60. }
  61. static int __init hardlockup_panic_setup(char *str)
  62. {
  63. if (!strncmp(str, "panic", 5))
  64. hardlockup_panic = 1;
  65. else if (!strncmp(str, "nopanic", 7))
  66. hardlockup_panic = 0;
  67. else if (!strncmp(str, "0", 1))
  68. nmi_watchdog_user_enabled = 0;
  69. else if (!strncmp(str, "1", 1))
  70. nmi_watchdog_user_enabled = 1;
  71. return 1;
  72. }
  73. __setup("nmi_watchdog=", hardlockup_panic_setup);
  74. # ifdef CONFIG_SMP
  75. int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
  76. static int __init hardlockup_all_cpu_backtrace_setup(char *str)
  77. {
  78. sysctl_hardlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
  79. return 1;
  80. }
  81. __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
  82. # endif /* CONFIG_SMP */
  83. #endif /* CONFIG_HARDLOCKUP_DETECTOR */
  84. /*
  85. * These functions can be overridden if an architecture implements its
  86. * own hardlockup detector.
  87. *
  88. * watchdog_nmi_enable/disable can be implemented to start and stop when
  89. * softlockup watchdog threads start and stop. The arch must select the
  90. * SOFTLOCKUP_DETECTOR Kconfig.
  91. */
  92. int __weak watchdog_nmi_enable(unsigned int cpu)
  93. {
  94. hardlockup_detector_perf_enable();
  95. return 0;
  96. }
  97. void __weak watchdog_nmi_disable(unsigned int cpu)
  98. {
  99. hardlockup_detector_perf_disable();
  100. }
  101. /* Return 0, if a NMI watchdog is available. Error code otherwise */
  102. int __weak __init watchdog_nmi_probe(void)
  103. {
  104. return hardlockup_detector_perf_init();
  105. }
  106. /**
  107. * watchdog_nmi_stop - Stop the watchdog for reconfiguration
  108. *
  109. * The reconfiguration steps are:
  110. * watchdog_nmi_stop();
  111. * update_variables();
  112. * watchdog_nmi_start();
  113. */
  114. void __weak watchdog_nmi_stop(void) { }
  115. /**
  116. * watchdog_nmi_start - Start the watchdog after reconfiguration
  117. *
  118. * Counterpart to watchdog_nmi_stop().
  119. *
  120. * The following variables have been updated in update_variables() and
  121. * contain the currently valid configuration:
  122. * - watchdog_enabled
  123. * - watchdog_thresh
  124. * - watchdog_cpumask
  125. */
  126. void __weak watchdog_nmi_start(void) { }
  127. /**
  128. * lockup_detector_update_enable - Update the sysctl enable bit
  129. *
  130. * Caller needs to make sure that the NMI/perf watchdogs are off, so this
  131. * can't race with watchdog_nmi_disable().
  132. */
  133. static void lockup_detector_update_enable(void)
  134. {
  135. watchdog_enabled = 0;
  136. if (!watchdog_user_enabled)
  137. return;
  138. if (nmi_watchdog_available && nmi_watchdog_user_enabled)
  139. watchdog_enabled |= NMI_WATCHDOG_ENABLED;
  140. if (soft_watchdog_user_enabled)
  141. watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
  142. }
  143. #ifdef CONFIG_SOFTLOCKUP_DETECTOR
  144. /* Global variables, exported for sysctl */
  145. unsigned int __read_mostly softlockup_panic =
  146. CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
  147. static bool softlockup_initialized __read_mostly;
  148. static u64 __read_mostly sample_period;
  149. static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
  150. static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
  151. static DEFINE_PER_CPU(bool, softlockup_touch_sync);
  152. static DEFINE_PER_CPU(bool, soft_watchdog_warn);
  153. static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
  154. static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
  155. static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved);
  156. static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
  157. static unsigned long soft_lockup_nmi_warn;
  158. static int __init softlockup_panic_setup(char *str)
  159. {
  160. softlockup_panic = simple_strtoul(str, NULL, 0);
  161. return 1;
  162. }
  163. __setup("softlockup_panic=", softlockup_panic_setup);
  164. static int __init nowatchdog_setup(char *str)
  165. {
  166. watchdog_user_enabled = 0;
  167. return 1;
  168. }
  169. __setup("nowatchdog", nowatchdog_setup);
  170. static int __init nosoftlockup_setup(char *str)
  171. {
  172. soft_watchdog_user_enabled = 0;
  173. return 1;
  174. }
  175. __setup("nosoftlockup", nosoftlockup_setup);
  176. #ifdef CONFIG_SMP
  177. int __read_mostly sysctl_softlockup_all_cpu_backtrace;
  178. static int __init softlockup_all_cpu_backtrace_setup(char *str)
  179. {
  180. sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
  181. return 1;
  182. }
  183. __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
  184. #endif
  185. static void __lockup_detector_cleanup(void);
  186. /*
  187. * Hard-lockup warnings should be triggered after just a few seconds. Soft-
  188. * lockups can have false positives under extreme conditions. So we generally
  189. * want a higher threshold for soft lockups than for hard lockups. So we couple
  190. * the thresholds with a factor: we make the soft threshold twice the amount of
  191. * time the hard threshold is.
  192. */
  193. static int get_softlockup_thresh(void)
  194. {
  195. return watchdog_thresh * 2;
  196. }
  197. /*
  198. * Returns seconds, approximately. We don't need nanosecond
  199. * resolution, and we don't need to waste time with a big divide when
  200. * 2^30ns == 1.074s.
  201. */
  202. static unsigned long get_timestamp(void)
  203. {
  204. return running_clock() >> 30LL; /* 2^30 ~= 10^9 */
  205. }
  206. static void set_sample_period(void)
  207. {
  208. /*
  209. * convert watchdog_thresh from seconds to ns
  210. * the divide by 5 is to give hrtimer several chances (two
  211. * or three with the current relation between the soft
  212. * and hard thresholds) to increment before the
  213. * hardlockup detector generates a warning
  214. */
  215. sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5);
  216. watchdog_update_hrtimer_threshold(sample_period);
  217. }
  218. /* Commands for resetting the watchdog */
  219. static void __touch_watchdog(void)
  220. {
  221. __this_cpu_write(watchdog_touch_ts, get_timestamp());
  222. }
  223. /**
  224. * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls
  225. *
  226. * Call when the scheduler may have stalled for legitimate reasons
  227. * preventing the watchdog task from executing - e.g. the scheduler
  228. * entering idle state. This should only be used for scheduler events.
  229. * Use touch_softlockup_watchdog() for everything else.
  230. */
  231. void touch_softlockup_watchdog_sched(void)
  232. {
  233. /*
  234. * Preemption can be enabled. It doesn't matter which CPU's timestamp
  235. * gets zeroed here, so use the raw_ operation.
  236. */
  237. raw_cpu_write(watchdog_touch_ts, 0);
  238. }
  239. void touch_softlockup_watchdog(void)
  240. {
  241. touch_softlockup_watchdog_sched();
  242. wq_watchdog_touch(raw_smp_processor_id());
  243. }
  244. EXPORT_SYMBOL(touch_softlockup_watchdog);
  245. void touch_all_softlockup_watchdogs(void)
  246. {
  247. int cpu;
  248. /*
  249. * watchdog_mutex cannpt be taken here, as this might be called
  250. * from (soft)interrupt context, so the access to
  251. * watchdog_allowed_cpumask might race with a concurrent update.
  252. *
  253. * The watchdog time stamp can race against a concurrent real
  254. * update as well, the only side effect might be a cycle delay for
  255. * the softlockup check.
  256. */
  257. for_each_cpu(cpu, &watchdog_allowed_mask)
  258. per_cpu(watchdog_touch_ts, cpu) = 0;
  259. wq_watchdog_touch(-1);
  260. }
  261. void touch_softlockup_watchdog_sync(void)
  262. {
  263. __this_cpu_write(softlockup_touch_sync, true);
  264. __this_cpu_write(watchdog_touch_ts, 0);
  265. }
  266. static int is_softlockup(unsigned long touch_ts)
  267. {
  268. unsigned long now = get_timestamp();
  269. if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){
  270. /* Warn about unreasonable delays. */
  271. if (time_after(now, touch_ts + get_softlockup_thresh()))
  272. return now - touch_ts;
  273. }
  274. return 0;
  275. }
  276. /* watchdog detector functions */
  277. bool is_hardlockup(void)
  278. {
  279. unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
  280. if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
  281. return true;
  282. __this_cpu_write(hrtimer_interrupts_saved, hrint);
  283. return false;
  284. }
  285. static void watchdog_interrupt_count(void)
  286. {
  287. __this_cpu_inc(hrtimer_interrupts);
  288. }
  289. static DEFINE_PER_CPU(struct completion, softlockup_completion);
  290. static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
  291. /*
  292. * The watchdog thread function - touches the timestamp.
  293. *
  294. * It only runs once every sample_period seconds (4 seconds by
  295. * default) to reset the softlockup timestamp. If this gets delayed
  296. * for more than 2*watchdog_thresh seconds then the debug-printout
  297. * triggers in watchdog_timer_fn().
  298. */
  299. static int softlockup_fn(void *data)
  300. {
  301. __this_cpu_write(soft_lockup_hrtimer_cnt,
  302. __this_cpu_read(hrtimer_interrupts));
  303. __touch_watchdog();
  304. complete(this_cpu_ptr(&softlockup_completion));
  305. return 0;
  306. }
  307. /* watchdog kicker functions */
  308. static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
  309. {
  310. unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts);
  311. struct pt_regs *regs = get_irq_regs();
  312. int duration;
  313. int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
  314. if (!watchdog_enabled)
  315. return HRTIMER_NORESTART;
  316. /* kick the hardlockup detector */
  317. watchdog_interrupt_count();
  318. /* kick the softlockup detector */
  319. if (completion_done(this_cpu_ptr(&softlockup_completion))) {
  320. reinit_completion(this_cpu_ptr(&softlockup_completion));
  321. stop_one_cpu_nowait(smp_processor_id(),
  322. softlockup_fn, NULL,
  323. this_cpu_ptr(&softlockup_stop_work));
  324. }
  325. /* .. and repeat */
  326. hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
  327. if (touch_ts == 0) {
  328. if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
  329. /*
  330. * If the time stamp was touched atomically
  331. * make sure the scheduler tick is up to date.
  332. */
  333. __this_cpu_write(softlockup_touch_sync, false);
  334. sched_clock_tick();
  335. }
  336. /* Clear the guest paused flag on watchdog reset */
  337. kvm_check_and_clear_guest_paused();
  338. __touch_watchdog();
  339. return HRTIMER_RESTART;
  340. }
  341. /* check for a softlockup
  342. * This is done by making sure a high priority task is
  343. * being scheduled. The task touches the watchdog to
  344. * indicate it is getting cpu time. If it hasn't then
  345. * this is a good indication some task is hogging the cpu
  346. */
  347. duration = is_softlockup(touch_ts);
  348. if (unlikely(duration)) {
  349. /*
  350. * If a virtual machine is stopped by the host it can look to
  351. * the watchdog like a soft lockup, check to see if the host
  352. * stopped the vm before we issue the warning
  353. */
  354. if (kvm_check_and_clear_guest_paused())
  355. return HRTIMER_RESTART;
  356. /* only warn once */
  357. if (__this_cpu_read(soft_watchdog_warn) == true) {
  358. /*
  359. * When multiple processes are causing softlockups the
  360. * softlockup detector only warns on the first one
  361. * because the code relies on a full quiet cycle to
  362. * re-arm. The second process prevents the quiet cycle
  363. * and never gets reported. Use task pointers to detect
  364. * this.
  365. */
  366. if (__this_cpu_read(softlockup_task_ptr_saved) !=
  367. current) {
  368. __this_cpu_write(soft_watchdog_warn, false);
  369. __touch_watchdog();
  370. }
  371. return HRTIMER_RESTART;
  372. }
  373. if (softlockup_all_cpu_backtrace) {
  374. /* Prevent multiple soft-lockup reports if one cpu is already
  375. * engaged in dumping cpu back traces
  376. */
  377. if (test_and_set_bit(0, &soft_lockup_nmi_warn)) {
  378. /* Someone else will report us. Let's give up */
  379. __this_cpu_write(soft_watchdog_warn, true);
  380. return HRTIMER_RESTART;
  381. }
  382. }
  383. pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
  384. smp_processor_id(), duration,
  385. current->comm, task_pid_nr(current));
  386. __this_cpu_write(softlockup_task_ptr_saved, current);
  387. print_modules();
  388. print_irqtrace_events(current);
  389. if (regs)
  390. show_regs(regs);
  391. else
  392. dump_stack();
  393. if (softlockup_all_cpu_backtrace) {
  394. /* Avoid generating two back traces for current
  395. * given that one is already made above
  396. */
  397. trigger_allbutself_cpu_backtrace();
  398. clear_bit(0, &soft_lockup_nmi_warn);
  399. /* Barrier to sync with other cpus */
  400. smp_mb__after_atomic();
  401. }
  402. add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK);
  403. if (softlockup_panic)
  404. panic("softlockup: hung tasks");
  405. __this_cpu_write(soft_watchdog_warn, true);
  406. } else
  407. __this_cpu_write(soft_watchdog_warn, false);
  408. return HRTIMER_RESTART;
  409. }
  410. static void watchdog_enable(unsigned int cpu)
  411. {
  412. struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
  413. struct completion *done = this_cpu_ptr(&softlockup_completion);
  414. WARN_ON_ONCE(cpu != smp_processor_id());
  415. init_completion(done);
  416. complete(done);
  417. /*
  418. * Start the timer first to prevent the NMI watchdog triggering
  419. * before the timer has a chance to fire.
  420. */
  421. hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
  422. hrtimer->function = watchdog_timer_fn;
  423. hrtimer_start(hrtimer, ns_to_ktime(sample_period),
  424. HRTIMER_MODE_REL_PINNED);
  425. /* Initialize timestamp */
  426. __touch_watchdog();
  427. /* Enable the perf event */
  428. if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
  429. watchdog_nmi_enable(cpu);
  430. }
  431. static void watchdog_disable(unsigned int cpu)
  432. {
  433. struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
  434. WARN_ON_ONCE(cpu != smp_processor_id());
  435. /*
  436. * Disable the perf event first. That prevents that a large delay
  437. * between disabling the timer and disabling the perf event causes
  438. * the perf NMI to detect a false positive.
  439. */
  440. watchdog_nmi_disable(cpu);
  441. hrtimer_cancel(hrtimer);
  442. wait_for_completion(this_cpu_ptr(&softlockup_completion));
  443. }
  444. static int softlockup_stop_fn(void *data)
  445. {
  446. watchdog_disable(smp_processor_id());
  447. return 0;
  448. }
  449. static void softlockup_stop_all(void)
  450. {
  451. int cpu;
  452. if (!softlockup_initialized)
  453. return;
  454. for_each_cpu(cpu, &watchdog_allowed_mask)
  455. smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false);
  456. cpumask_clear(&watchdog_allowed_mask);
  457. }
  458. static int softlockup_start_fn(void *data)
  459. {
  460. watchdog_enable(smp_processor_id());
  461. return 0;
  462. }
  463. static void softlockup_start_all(void)
  464. {
  465. int cpu;
  466. cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
  467. for_each_cpu(cpu, &watchdog_allowed_mask)
  468. smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false);
  469. }
  470. int lockup_detector_online_cpu(unsigned int cpu)
  471. {
  472. watchdog_enable(cpu);
  473. return 0;
  474. }
  475. int lockup_detector_offline_cpu(unsigned int cpu)
  476. {
  477. watchdog_disable(cpu);
  478. return 0;
  479. }
  480. static void lockup_detector_reconfigure(void)
  481. {
  482. cpus_read_lock();
  483. watchdog_nmi_stop();
  484. softlockup_stop_all();
  485. set_sample_period();
  486. lockup_detector_update_enable();
  487. if (watchdog_enabled && watchdog_thresh)
  488. softlockup_start_all();
  489. watchdog_nmi_start();
  490. cpus_read_unlock();
  491. /*
  492. * Must be called outside the cpus locked section to prevent
  493. * recursive locking in the perf code.
  494. */
  495. __lockup_detector_cleanup();
  496. }
  497. /*
  498. * Create the watchdog thread infrastructure and configure the detector(s).
  499. *
  500. * The threads are not unparked as watchdog_allowed_mask is empty. When
  501. * the threads are sucessfully initialized, take the proper locks and
  502. * unpark the threads in the watchdog_cpumask if the watchdog is enabled.
  503. */
  504. static __init void lockup_detector_setup(void)
  505. {
  506. /*
  507. * If sysctl is off and watchdog got disabled on the command line,
  508. * nothing to do here.
  509. */
  510. lockup_detector_update_enable();
  511. if (!IS_ENABLED(CONFIG_SYSCTL) &&
  512. !(watchdog_enabled && watchdog_thresh))
  513. return;
  514. mutex_lock(&watchdog_mutex);
  515. lockup_detector_reconfigure();
  516. softlockup_initialized = true;
  517. mutex_unlock(&watchdog_mutex);
  518. }
  519. #else /* CONFIG_SOFTLOCKUP_DETECTOR */
  520. static void lockup_detector_reconfigure(void)
  521. {
  522. cpus_read_lock();
  523. watchdog_nmi_stop();
  524. lockup_detector_update_enable();
  525. watchdog_nmi_start();
  526. cpus_read_unlock();
  527. }
  528. static inline void lockup_detector_setup(void)
  529. {
  530. lockup_detector_reconfigure();
  531. }
  532. #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
  533. static void __lockup_detector_cleanup(void)
  534. {
  535. lockdep_assert_held(&watchdog_mutex);
  536. hardlockup_detector_perf_cleanup();
  537. }
  538. /**
  539. * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
  540. *
  541. * Caller must not hold the cpu hotplug rwsem.
  542. */
  543. void lockup_detector_cleanup(void)
  544. {
  545. mutex_lock(&watchdog_mutex);
  546. __lockup_detector_cleanup();
  547. mutex_unlock(&watchdog_mutex);
  548. }
  549. /**
  550. * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
  551. *
  552. * Special interface for parisc. It prevents lockup detector warnings from
  553. * the default pm_poweroff() function which busy loops forever.
  554. */
  555. void lockup_detector_soft_poweroff(void)
  556. {
  557. watchdog_enabled = 0;
  558. }
  559. #ifdef CONFIG_SYSCTL
  560. /* Propagate any changes to the watchdog threads */
  561. static void proc_watchdog_update(void)
  562. {
  563. /* Remove impossible cpus to keep sysctl output clean. */
  564. cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
  565. lockup_detector_reconfigure();
  566. }
  567. /*
  568. * common function for watchdog, nmi_watchdog and soft_watchdog parameter
  569. *
  570. * caller | table->data points to | 'which'
  571. * -------------------|----------------------------|--------------------------
  572. * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
  573. * | | SOFT_WATCHDOG_ENABLED
  574. * -------------------|----------------------------|--------------------------
  575. * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED
  576. * -------------------|----------------------------|--------------------------
  577. * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
  578. */
  579. static int proc_watchdog_common(int which, struct ctl_table *table, int write,
  580. void __user *buffer, size_t *lenp, loff_t *ppos)
  581. {
  582. int err, old, *param = table->data;
  583. mutex_lock(&watchdog_mutex);
  584. if (!write) {
  585. /*
  586. * On read synchronize the userspace interface. This is a
  587. * racy snapshot.
  588. */
  589. *param = (watchdog_enabled & which) != 0;
  590. err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
  591. } else {
  592. old = READ_ONCE(*param);
  593. err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
  594. if (!err && old != READ_ONCE(*param))
  595. proc_watchdog_update();
  596. }
  597. mutex_unlock(&watchdog_mutex);
  598. return err;
  599. }
  600. /*
  601. * /proc/sys/kernel/watchdog
  602. */
  603. int proc_watchdog(struct ctl_table *table, int write,
  604. void __user *buffer, size_t *lenp, loff_t *ppos)
  605. {
  606. return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED,
  607. table, write, buffer, lenp, ppos);
  608. }
  609. /*
  610. * /proc/sys/kernel/nmi_watchdog
  611. */
  612. int proc_nmi_watchdog(struct ctl_table *table, int write,
  613. void __user *buffer, size_t *lenp, loff_t *ppos)
  614. {
  615. if (!nmi_watchdog_available && write)
  616. return -ENOTSUPP;
  617. return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
  618. table, write, buffer, lenp, ppos);
  619. }
  620. /*
  621. * /proc/sys/kernel/soft_watchdog
  622. */
  623. int proc_soft_watchdog(struct ctl_table *table, int write,
  624. void __user *buffer, size_t *lenp, loff_t *ppos)
  625. {
  626. return proc_watchdog_common(SOFT_WATCHDOG_ENABLED,
  627. table, write, buffer, lenp, ppos);
  628. }
  629. /*
  630. * /proc/sys/kernel/watchdog_thresh
  631. */
  632. int proc_watchdog_thresh(struct ctl_table *table, int write,
  633. void __user *buffer, size_t *lenp, loff_t *ppos)
  634. {
  635. int err, old;
  636. mutex_lock(&watchdog_mutex);
  637. old = READ_ONCE(watchdog_thresh);
  638. err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
  639. if (!err && write && old != READ_ONCE(watchdog_thresh))
  640. proc_watchdog_update();
  641. mutex_unlock(&watchdog_mutex);
  642. return err;
  643. }
  644. /*
  645. * The cpumask is the mask of possible cpus that the watchdog can run
  646. * on, not the mask of cpus it is actually running on. This allows the
  647. * user to specify a mask that will include cpus that have not yet
  648. * been brought online, if desired.
  649. */
  650. int proc_watchdog_cpumask(struct ctl_table *table, int write,
  651. void __user *buffer, size_t *lenp, loff_t *ppos)
  652. {
  653. int err;
  654. mutex_lock(&watchdog_mutex);
  655. err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
  656. if (!err && write)
  657. proc_watchdog_update();
  658. mutex_unlock(&watchdog_mutex);
  659. return err;
  660. }
  661. #endif /* CONFIG_SYSCTL */
  662. void __init lockup_detector_init(void)
  663. {
  664. if (tick_nohz_full_enabled())
  665. pr_info("Disabling watchdog on nohz_full cores by default\n");
  666. cpumask_copy(&watchdog_cpumask,
  667. housekeeping_cpumask(HK_FLAG_TIMER));
  668. if (!watchdog_nmi_probe())
  669. nmi_watchdog_available = true;
  670. lockup_detector_setup();
  671. }