Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 2 | /* |
| 3 | * Detect hard and soft lockups on a system |
| 4 | * |
| 5 | * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc. |
| 6 | * |
Fernando Luis Vázquez Cao | 86f5e6a | 2012-02-09 17:42:22 -0500 | [diff] [blame] | 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 |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 10 | * to those contributors as well. |
| 11 | */ |
| 12 | |
Kefeng Wang | 5f92a7b | 2017-07-14 14:49:46 -0700 | [diff] [blame] | 13 | #define pr_fmt(fmt) "watchdog: " fmt |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 14 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 15 | #include <linux/mm.h> |
| 16 | #include <linux/cpu.h> |
| 17 | #include <linux/nmi.h> |
| 18 | #include <linux/init.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 19 | #include <linux/module.h> |
| 20 | #include <linux/sysctl.h> |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 21 | #include <linux/tick.h> |
Ingo Molnar | e601757 | 2017-02-01 16:36:40 +0100 | [diff] [blame] | 22 | #include <linux/sched/clock.h> |
Ingo Molnar | b17b015 | 2017-02-08 18:51:35 +0100 | [diff] [blame] | 23 | #include <linux/sched/debug.h> |
Frederic Weisbecker | 7863406 | 2017-10-27 04:42:28 +0200 | [diff] [blame] | 24 | #include <linux/sched/isolation.h> |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 25 | #include <linux/stop_machine.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 26 | |
| 27 | #include <asm/irq_regs.h> |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 28 | #include <linux/kvm_para.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 29 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 30 | static DEFINE_MUTEX(watchdog_mutex); |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 31 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 32 | #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG) |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 33 | # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED) |
| 34 | # define NMI_WATCHDOG_DEFAULT 1 |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 35 | #else |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 36 | # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED) |
| 37 | # define NMI_WATCHDOG_DEFAULT 0 |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 38 | #endif |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 39 | |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 40 | unsigned long __read_mostly watchdog_enabled; |
| 41 | int __read_mostly watchdog_user_enabled = 1; |
| 42 | int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT; |
| 43 | int __read_mostly soft_watchdog_user_enabled = 1; |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 44 | int __read_mostly watchdog_thresh = 10; |
Valdis Kletnieks | 48084ab | 2019-03-12 05:33:48 -0400 | [diff] [blame] | 45 | static int __read_mostly nmi_watchdog_available; |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 46 | |
Valdis Kletnieks | 48084ab | 2019-03-12 05:33:48 -0400 | [diff] [blame] | 47 | static struct cpumask watchdog_allowed_mask __read_mostly; |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 48 | |
| 49 | struct cpumask watchdog_cpumask __read_mostly; |
| 50 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); |
| 51 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 52 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 53 | /* |
| 54 | * Should we panic when a soft-lockup or hard-lockup occurs: |
| 55 | */ |
| 56 | unsigned int __read_mostly hardlockup_panic = |
| 57 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; |
| 58 | /* |
| 59 | * We may not want to enable hard lockup detection by default in all cases, |
| 60 | * for example when running the kernel as a guest on a hypervisor. In these |
| 61 | * cases this function can be called to disable hard lockup detection. This |
| 62 | * function should only be executed once by the boot processor before the |
| 63 | * kernel command line parameters are parsed, because otherwise it is not |
| 64 | * possible to override this in hardlockup_panic_setup(). |
| 65 | */ |
Thomas Gleixner | 7a35582 | 2017-09-12 21:37:02 +0200 | [diff] [blame] | 66 | void __init hardlockup_detector_disable(void) |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 67 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 68 | nmi_watchdog_user_enabled = 0; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 69 | } |
| 70 | |
| 71 | static int __init hardlockup_panic_setup(char *str) |
| 72 | { |
| 73 | if (!strncmp(str, "panic", 5)) |
| 74 | hardlockup_panic = 1; |
| 75 | else if (!strncmp(str, "nopanic", 7)) |
| 76 | hardlockup_panic = 0; |
| 77 | else if (!strncmp(str, "0", 1)) |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 78 | nmi_watchdog_user_enabled = 0; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 79 | else if (!strncmp(str, "1", 1)) |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 80 | nmi_watchdog_user_enabled = 1; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 81 | return 1; |
| 82 | } |
| 83 | __setup("nmi_watchdog=", hardlockup_panic_setup); |
| 84 | |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 85 | # ifdef CONFIG_SMP |
| 86 | int __read_mostly sysctl_hardlockup_all_cpu_backtrace; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 87 | |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 88 | static int __init hardlockup_all_cpu_backtrace_setup(char *str) |
| 89 | { |
| 90 | sysctl_hardlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0); |
| 91 | return 1; |
| 92 | } |
| 93 | __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup); |
| 94 | # endif /* CONFIG_SMP */ |
| 95 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 96 | |
Ulrich Obergfell | ec6a906 | 2015-09-04 15:45:28 -0700 | [diff] [blame] | 97 | /* |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 98 | * These functions can be overridden if an architecture implements its |
| 99 | * own hardlockup detector. |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 100 | * |
| 101 | * watchdog_nmi_enable/disable can be implemented to start and stop when |
| 102 | * softlockup watchdog threads start and stop. The arch must select the |
| 103 | * SOFTLOCKUP_DETECTOR Kconfig. |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 104 | */ |
| 105 | int __weak watchdog_nmi_enable(unsigned int cpu) |
| 106 | { |
Thomas Gleixner | 146c9d0 | 2017-09-12 21:37:21 +0200 | [diff] [blame] | 107 | hardlockup_detector_perf_enable(); |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 108 | return 0; |
| 109 | } |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 110 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 111 | void __weak watchdog_nmi_disable(unsigned int cpu) |
| 112 | { |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 113 | hardlockup_detector_perf_disable(); |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 114 | } |
| 115 | |
Thomas Gleixner | a994a31 | 2017-09-12 21:37:19 +0200 | [diff] [blame] | 116 | /* Return 0, if a NMI watchdog is available. Error code otherwise */ |
| 117 | int __weak __init watchdog_nmi_probe(void) |
| 118 | { |
| 119 | return hardlockup_detector_perf_init(); |
| 120 | } |
| 121 | |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 122 | /** |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 123 | * watchdog_nmi_stop - Stop the watchdog for reconfiguration |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 124 | * |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 125 | * The reconfiguration steps are: |
| 126 | * watchdog_nmi_stop(); |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 127 | * update_variables(); |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 128 | * watchdog_nmi_start(); |
| 129 | */ |
| 130 | void __weak watchdog_nmi_stop(void) { } |
| 131 | |
| 132 | /** |
| 133 | * watchdog_nmi_start - Start the watchdog after reconfiguration |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 134 | * |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 135 | * Counterpart to watchdog_nmi_stop(). |
| 136 | * |
| 137 | * The following variables have been updated in update_variables() and |
| 138 | * contain the currently valid configuration: |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 139 | * - watchdog_enabled |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 140 | * - watchdog_thresh |
| 141 | * - watchdog_cpumask |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 142 | */ |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 143 | void __weak watchdog_nmi_start(void) { } |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 144 | |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 145 | /** |
| 146 | * lockup_detector_update_enable - Update the sysctl enable bit |
| 147 | * |
| 148 | * Caller needs to make sure that the NMI/perf watchdogs are off, so this |
| 149 | * can't race with watchdog_nmi_disable(). |
| 150 | */ |
| 151 | static void lockup_detector_update_enable(void) |
| 152 | { |
| 153 | watchdog_enabled = 0; |
| 154 | if (!watchdog_user_enabled) |
| 155 | return; |
Thomas Gleixner | a994a31 | 2017-09-12 21:37:19 +0200 | [diff] [blame] | 156 | if (nmi_watchdog_available && nmi_watchdog_user_enabled) |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 157 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; |
| 158 | if (soft_watchdog_user_enabled) |
| 159 | watchdog_enabled |= SOFT_WATCHDOG_ENABLED; |
| 160 | } |
| 161 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 162 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR |
| 163 | |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 164 | /* Global variables, exported for sysctl */ |
| 165 | unsigned int __read_mostly softlockup_panic = |
| 166 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 167 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 168 | static bool softlockup_initialized __read_mostly; |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 169 | static u64 __read_mostly sample_period; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 170 | |
| 171 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 172 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
| 173 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 174 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 175 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
| 176 | static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 177 | static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 178 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 179 | static unsigned long soft_lockup_nmi_warn; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 180 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 181 | static int __init softlockup_panic_setup(char *str) |
| 182 | { |
| 183 | softlockup_panic = simple_strtoul(str, NULL, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 184 | return 1; |
| 185 | } |
| 186 | __setup("softlockup_panic=", softlockup_panic_setup); |
| 187 | |
| 188 | static int __init nowatchdog_setup(char *str) |
| 189 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 190 | watchdog_user_enabled = 0; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 191 | return 1; |
| 192 | } |
| 193 | __setup("nowatchdog", nowatchdog_setup); |
| 194 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 195 | static int __init nosoftlockup_setup(char *str) |
| 196 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 197 | soft_watchdog_user_enabled = 0; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 198 | return 1; |
| 199 | } |
| 200 | __setup("nosoftlockup", nosoftlockup_setup); |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 201 | |
Laurence Oberman | 1129505 | 2018-11-01 09:30:18 -0400 | [diff] [blame] | 202 | static int __init watchdog_thresh_setup(char *str) |
| 203 | { |
| 204 | get_option(&str, &watchdog_thresh); |
| 205 | return 1; |
| 206 | } |
| 207 | __setup("watchdog_thresh=", watchdog_thresh_setup); |
| 208 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 209 | #ifdef CONFIG_SMP |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 210 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; |
| 211 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 212 | static int __init softlockup_all_cpu_backtrace_setup(char *str) |
| 213 | { |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 214 | sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0); |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 215 | return 1; |
| 216 | } |
| 217 | __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 218 | #endif |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 219 | |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 220 | static void __lockup_detector_cleanup(void); |
| 221 | |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 222 | /* |
| 223 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- |
| 224 | * lockups can have false positives under extreme conditions. So we generally |
| 225 | * want a higher threshold for soft lockups than for hard lockups. So we couple |
| 226 | * the thresholds with a factor: we make the soft threshold twice the amount of |
| 227 | * time the hard threshold is. |
| 228 | */ |
Ingo Molnar | 6e9101a | 2011-05-24 05:43:18 +0200 | [diff] [blame] | 229 | static int get_softlockup_thresh(void) |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 230 | { |
| 231 | return watchdog_thresh * 2; |
| 232 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 233 | |
| 234 | /* |
| 235 | * Returns seconds, approximately. We don't need nanosecond |
| 236 | * resolution, and we don't need to waste time with a big divide when |
| 237 | * 2^30ns == 1.074s. |
| 238 | */ |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 239 | static unsigned long get_timestamp(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 240 | { |
Cyril Bur | 545a2bf | 2015-02-12 15:01:24 -0800 | [diff] [blame] | 241 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 242 | } |
| 243 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 244 | static void set_sample_period(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 245 | { |
| 246 | /* |
Mandeep Singh Baines | 586692a | 2011-05-22 22:10:22 -0700 | [diff] [blame] | 247 | * convert watchdog_thresh from seconds to ns |
Fernando Luis Vázquez Cao | 86f5e6a | 2012-02-09 17:42:22 -0500 | [diff] [blame] | 248 | * the divide by 5 is to give hrtimer several chances (two |
| 249 | * or three with the current relation between the soft |
| 250 | * and hard thresholds) to increment before the |
| 251 | * hardlockup detector generates a warning |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 252 | */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 253 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
Thomas Gleixner | 7edaeb6 | 2017-08-15 09:50:13 +0200 | [diff] [blame] | 254 | watchdog_update_hrtimer_threshold(sample_period); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 255 | } |
| 256 | |
| 257 | /* Commands for resetting the watchdog */ |
| 258 | static void __touch_watchdog(void) |
| 259 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 260 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 261 | } |
| 262 | |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 263 | /** |
| 264 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls |
| 265 | * |
| 266 | * Call when the scheduler may have stalled for legitimate reasons |
| 267 | * preventing the watchdog task from executing - e.g. the scheduler |
| 268 | * entering idle state. This should only be used for scheduler events. |
| 269 | * Use touch_softlockup_watchdog() for everything else. |
| 270 | */ |
Vincent Whitchurch | cb9d7fd | 2018-08-21 17:25:07 +0200 | [diff] [blame] | 271 | notrace void touch_softlockup_watchdog_sched(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 272 | { |
Andrew Morton | 7861144 | 2014-04-18 15:07:12 -0700 | [diff] [blame] | 273 | /* |
| 274 | * Preemption can be enabled. It doesn't matter which CPU's timestamp |
| 275 | * gets zeroed here, so use the raw_ operation. |
| 276 | */ |
| 277 | raw_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 278 | } |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 279 | |
Vincent Whitchurch | cb9d7fd | 2018-08-21 17:25:07 +0200 | [diff] [blame] | 280 | notrace void touch_softlockup_watchdog(void) |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 281 | { |
| 282 | touch_softlockup_watchdog_sched(); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 283 | wq_watchdog_touch(raw_smp_processor_id()); |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 284 | } |
Ingo Molnar | 0167c78 | 2010-05-13 08:53:33 +0200 | [diff] [blame] | 285 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 286 | |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 287 | void touch_all_softlockup_watchdogs(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 288 | { |
| 289 | int cpu; |
| 290 | |
| 291 | /* |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 292 | * watchdog_mutex cannpt be taken here, as this might be called |
| 293 | * from (soft)interrupt context, so the access to |
| 294 | * watchdog_allowed_cpumask might race with a concurrent update. |
| 295 | * |
| 296 | * The watchdog time stamp can race against a concurrent real |
| 297 | * update as well, the only side effect might be a cycle delay for |
| 298 | * the softlockup check. |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 299 | */ |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 300 | for_each_cpu(cpu, &watchdog_allowed_mask) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 301 | per_cpu(watchdog_touch_ts, cpu) = 0; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 302 | wq_watchdog_touch(-1); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 303 | } |
| 304 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 305 | void touch_softlockup_watchdog_sync(void) |
| 306 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 307 | __this_cpu_write(softlockup_touch_sync, true); |
| 308 | __this_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 309 | } |
| 310 | |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 311 | static int is_softlockup(unsigned long touch_ts) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 312 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 313 | unsigned long now = get_timestamp(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 314 | |
Ulrich Obergfell | 39d2da2 | 2015-11-05 18:44:56 -0800 | [diff] [blame] | 315 | if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 316 | /* Warn about unreasonable delays. */ |
| 317 | if (time_after(now, touch_ts + get_softlockup_thresh())) |
| 318 | return now - touch_ts; |
| 319 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 320 | return 0; |
| 321 | } |
| 322 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 323 | /* watchdog detector functions */ |
| 324 | bool is_hardlockup(void) |
| 325 | { |
| 326 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
| 327 | |
| 328 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
| 329 | return true; |
| 330 | |
| 331 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
| 332 | return false; |
| 333 | } |
| 334 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 335 | static void watchdog_interrupt_count(void) |
| 336 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 337 | __this_cpu_inc(hrtimer_interrupts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 338 | } |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 339 | |
Peter Zijlstra | be45bf5 | 2018-07-13 12:42:08 +0200 | [diff] [blame] | 340 | static DEFINE_PER_CPU(struct completion, softlockup_completion); |
| 341 | static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work); |
| 342 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 343 | /* |
| 344 | * The watchdog thread function - touches the timestamp. |
| 345 | * |
| 346 | * It only runs once every sample_period seconds (4 seconds by |
| 347 | * default) to reset the softlockup timestamp. If this gets delayed |
| 348 | * for more than 2*watchdog_thresh seconds then the debug-printout |
| 349 | * triggers in watchdog_timer_fn(). |
| 350 | */ |
| 351 | static int softlockup_fn(void *data) |
| 352 | { |
| 353 | __this_cpu_write(soft_lockup_hrtimer_cnt, |
| 354 | __this_cpu_read(hrtimer_interrupts)); |
| 355 | __touch_watchdog(); |
Peter Zijlstra | be45bf5 | 2018-07-13 12:42:08 +0200 | [diff] [blame] | 356 | complete(this_cpu_ptr(&softlockup_completion)); |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 357 | |
| 358 | return 0; |
| 359 | } |
| 360 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 361 | /* watchdog kicker functions */ |
| 362 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) |
| 363 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 364 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 365 | struct pt_regs *regs = get_irq_regs(); |
| 366 | int duration; |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 367 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 368 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 369 | if (!watchdog_enabled) |
Don Zickus | b94f511 | 2017-01-24 15:17:53 -0800 | [diff] [blame] | 370 | return HRTIMER_NORESTART; |
| 371 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 372 | /* kick the hardlockup detector */ |
| 373 | watchdog_interrupt_count(); |
| 374 | |
| 375 | /* kick the softlockup detector */ |
Peter Zijlstra | be45bf5 | 2018-07-13 12:42:08 +0200 | [diff] [blame] | 376 | if (completion_done(this_cpu_ptr(&softlockup_completion))) { |
| 377 | reinit_completion(this_cpu_ptr(&softlockup_completion)); |
| 378 | stop_one_cpu_nowait(smp_processor_id(), |
| 379 | softlockup_fn, NULL, |
| 380 | this_cpu_ptr(&softlockup_stop_work)); |
| 381 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 382 | |
| 383 | /* .. and repeat */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 384 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 385 | |
| 386 | if (touch_ts == 0) { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 387 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 388 | /* |
| 389 | * If the time stamp was touched atomically |
| 390 | * make sure the scheduler tick is up to date. |
| 391 | */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 392 | __this_cpu_write(softlockup_touch_sync, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 393 | sched_clock_tick(); |
| 394 | } |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 395 | |
| 396 | /* Clear the guest paused flag on watchdog reset */ |
| 397 | kvm_check_and_clear_guest_paused(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 398 | __touch_watchdog(); |
| 399 | return HRTIMER_RESTART; |
| 400 | } |
| 401 | |
| 402 | /* check for a softlockup |
| 403 | * This is done by making sure a high priority task is |
| 404 | * being scheduled. The task touches the watchdog to |
| 405 | * indicate it is getting cpu time. If it hasn't then |
| 406 | * this is a good indication some task is hogging the cpu |
| 407 | */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 408 | duration = is_softlockup(touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 409 | if (unlikely(duration)) { |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 410 | /* |
| 411 | * If a virtual machine is stopped by the host it can look to |
| 412 | * the watchdog like a soft lockup, check to see if the host |
| 413 | * stopped the vm before we issue the warning |
| 414 | */ |
| 415 | if (kvm_check_and_clear_guest_paused()) |
| 416 | return HRTIMER_RESTART; |
| 417 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 418 | /* only warn once */ |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 419 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
| 420 | /* |
| 421 | * When multiple processes are causing softlockups the |
| 422 | * softlockup detector only warns on the first one |
| 423 | * because the code relies on a full quiet cycle to |
| 424 | * re-arm. The second process prevents the quiet cycle |
| 425 | * and never gets reported. Use task pointers to detect |
| 426 | * this. |
| 427 | */ |
| 428 | if (__this_cpu_read(softlockup_task_ptr_saved) != |
| 429 | current) { |
| 430 | __this_cpu_write(soft_watchdog_warn, false); |
| 431 | __touch_watchdog(); |
| 432 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 433 | return HRTIMER_RESTART; |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 434 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 435 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 436 | if (softlockup_all_cpu_backtrace) { |
| 437 | /* Prevent multiple soft-lockup reports if one cpu is already |
| 438 | * engaged in dumping cpu back traces |
| 439 | */ |
| 440 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { |
| 441 | /* Someone else will report us. Let's give up */ |
| 442 | __this_cpu_write(soft_watchdog_warn, true); |
| 443 | return HRTIMER_RESTART; |
| 444 | } |
| 445 | } |
| 446 | |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 447 | pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 448 | smp_processor_id(), duration, |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 449 | current->comm, task_pid_nr(current)); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 450 | __this_cpu_write(softlockup_task_ptr_saved, current); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 451 | print_modules(); |
| 452 | print_irqtrace_events(current); |
| 453 | if (regs) |
| 454 | show_regs(regs); |
| 455 | else |
| 456 | dump_stack(); |
| 457 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 458 | if (softlockup_all_cpu_backtrace) { |
| 459 | /* Avoid generating two back traces for current |
| 460 | * given that one is already made above |
| 461 | */ |
| 462 | trigger_allbutself_cpu_backtrace(); |
| 463 | |
| 464 | clear_bit(0, &soft_lockup_nmi_warn); |
| 465 | /* Barrier to sync with other cpus */ |
| 466 | smp_mb__after_atomic(); |
| 467 | } |
| 468 | |
Josh Hunt | 69361ee | 2014-08-08 14:22:31 -0700 | [diff] [blame] | 469 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 470 | if (softlockup_panic) |
| 471 | panic("softlockup: hung tasks"); |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 472 | __this_cpu_write(soft_watchdog_warn, true); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 473 | } else |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 474 | __this_cpu_write(soft_watchdog_warn, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 475 | |
| 476 | return HRTIMER_RESTART; |
| 477 | } |
| 478 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 479 | static void watchdog_enable(unsigned int cpu) |
| 480 | { |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 481 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
Peter Zijlstra | be45bf5 | 2018-07-13 12:42:08 +0200 | [diff] [blame] | 482 | struct completion *done = this_cpu_ptr(&softlockup_completion); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 483 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 484 | WARN_ON_ONCE(cpu != smp_processor_id()); |
| 485 | |
Peter Zijlstra | be45bf5 | 2018-07-13 12:42:08 +0200 | [diff] [blame] | 486 | init_completion(done); |
| 487 | complete(done); |
| 488 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 489 | /* |
| 490 | * Start the timer first to prevent the NMI watchdog triggering |
| 491 | * before the timer has a chance to fire. |
| 492 | */ |
Bjørn Mork | 3935e895 | 2012-12-19 20:51:31 +0100 | [diff] [blame] | 493 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
| 494 | hrtimer->function = watchdog_timer_fn; |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 495 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 496 | HRTIMER_MODE_REL_PINNED); |
| 497 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 498 | /* Initialize timestamp */ |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 499 | __touch_watchdog(); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 500 | /* Enable the perf event */ |
Thomas Gleixner | 146c9d0 | 2017-09-12 21:37:21 +0200 | [diff] [blame] | 501 | if (watchdog_enabled & NMI_WATCHDOG_ENABLED) |
| 502 | watchdog_nmi_enable(cpu); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 503 | } |
| 504 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 505 | static void watchdog_disable(unsigned int cpu) |
| 506 | { |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 507 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 508 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 509 | WARN_ON_ONCE(cpu != smp_processor_id()); |
| 510 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 511 | /* |
| 512 | * Disable the perf event first. That prevents that a large delay |
| 513 | * between disabling the timer and disabling the perf event causes |
| 514 | * the perf NMI to detect a false positive. |
| 515 | */ |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 516 | watchdog_nmi_disable(cpu); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 517 | hrtimer_cancel(hrtimer); |
Peter Zijlstra | be45bf5 | 2018-07-13 12:42:08 +0200 | [diff] [blame] | 518 | wait_for_completion(this_cpu_ptr(&softlockup_completion)); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 519 | } |
| 520 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 521 | static int softlockup_stop_fn(void *data) |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 522 | { |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 523 | watchdog_disable(smp_processor_id()); |
| 524 | return 0; |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 525 | } |
| 526 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 527 | static void softlockup_stop_all(void) |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 528 | { |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 529 | int cpu; |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 530 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 531 | if (!softlockup_initialized) |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 532 | return; |
| 533 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 534 | for_each_cpu(cpu, &watchdog_allowed_mask) |
| 535 | smp_call_on_cpu(cpu, softlockup_stop_fn, NULL, false); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 536 | |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 537 | cpumask_clear(&watchdog_allowed_mask); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 538 | } |
| 539 | |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 540 | static int softlockup_start_fn(void *data) |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 541 | { |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 542 | watchdog_enable(smp_processor_id()); |
| 543 | return 0; |
| 544 | } |
| 545 | |
| 546 | static void softlockup_start_all(void) |
| 547 | { |
| 548 | int cpu; |
| 549 | |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 550 | cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask); |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 551 | for_each_cpu(cpu, &watchdog_allowed_mask) |
| 552 | smp_call_on_cpu(cpu, softlockup_start_fn, NULL, false); |
| 553 | } |
| 554 | |
| 555 | int lockup_detector_online_cpu(unsigned int cpu) |
| 556 | { |
Thomas Gleixner | 7dd4761 | 2019-03-26 22:51:02 +0100 | [diff] [blame] | 557 | if (cpumask_test_cpu(cpu, &watchdog_allowed_mask)) |
| 558 | watchdog_enable(cpu); |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 559 | return 0; |
| 560 | } |
| 561 | |
| 562 | int lockup_detector_offline_cpu(unsigned int cpu) |
| 563 | { |
Thomas Gleixner | 7dd4761 | 2019-03-26 22:51:02 +0100 | [diff] [blame] | 564 | if (cpumask_test_cpu(cpu, &watchdog_allowed_mask)) |
| 565 | watchdog_disable(cpu); |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 566 | return 0; |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 567 | } |
| 568 | |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 569 | static void lockup_detector_reconfigure(void) |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 570 | { |
Thomas Gleixner | e31d688 | 2017-10-03 16:37:53 +0200 | [diff] [blame] | 571 | cpus_read_lock(); |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 572 | watchdog_nmi_stop(); |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 573 | |
| 574 | softlockup_stop_all(); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 575 | set_sample_period(); |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 576 | lockup_detector_update_enable(); |
| 577 | if (watchdog_enabled && watchdog_thresh) |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 578 | softlockup_start_all(); |
| 579 | |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 580 | watchdog_nmi_start(); |
Thomas Gleixner | e31d688 | 2017-10-03 16:37:53 +0200 | [diff] [blame] | 581 | cpus_read_unlock(); |
| 582 | /* |
| 583 | * Must be called outside the cpus locked section to prevent |
| 584 | * recursive locking in the perf code. |
| 585 | */ |
| 586 | __lockup_detector_cleanup(); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 587 | } |
| 588 | |
| 589 | /* |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 590 | * Create the watchdog thread infrastructure and configure the detector(s). |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 591 | * |
| 592 | * The threads are not unparked as watchdog_allowed_mask is empty. When |
Arash Fotouhi | 76e1552 | 2019-03-22 19:28:32 -0700 | [diff] [blame] | 593 | * the threads are successfully initialized, take the proper locks and |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 594 | * unpark the threads in the watchdog_cpumask if the watchdog is enabled. |
| 595 | */ |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 596 | static __init void lockup_detector_setup(void) |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 597 | { |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 598 | /* |
| 599 | * If sysctl is off and watchdog got disabled on the command line, |
| 600 | * nothing to do here. |
| 601 | */ |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 602 | lockup_detector_update_enable(); |
| 603 | |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 604 | if (!IS_ENABLED(CONFIG_SYSCTL) && |
| 605 | !(watchdog_enabled && watchdog_thresh)) |
| 606 | return; |
| 607 | |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 608 | mutex_lock(&watchdog_mutex); |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 609 | lockup_detector_reconfigure(); |
Peter Zijlstra | 9cf5773 | 2018-06-07 10:52:03 +0200 | [diff] [blame] | 610 | softlockup_initialized = true; |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 611 | mutex_unlock(&watchdog_mutex); |
| 612 | } |
| 613 | |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 614 | #else /* CONFIG_SOFTLOCKUP_DETECTOR */ |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 615 | static void lockup_detector_reconfigure(void) |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 616 | { |
Thomas Gleixner | e31d688 | 2017-10-03 16:37:53 +0200 | [diff] [blame] | 617 | cpus_read_lock(); |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 618 | watchdog_nmi_stop(); |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 619 | lockup_detector_update_enable(); |
Thomas Gleixner | 6b9dc48 | 2017-10-02 12:34:50 +0200 | [diff] [blame] | 620 | watchdog_nmi_start(); |
Thomas Gleixner | e31d688 | 2017-10-03 16:37:53 +0200 | [diff] [blame] | 621 | cpus_read_unlock(); |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 622 | } |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 623 | static inline void lockup_detector_setup(void) |
Thomas Gleixner | 34ddaa3 | 2017-10-03 16:39:02 +0200 | [diff] [blame] | 624 | { |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 625 | lockup_detector_reconfigure(); |
Thomas Gleixner | 34ddaa3 | 2017-10-03 16:39:02 +0200 | [diff] [blame] | 626 | } |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 627 | #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */ |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 628 | |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 629 | static void __lockup_detector_cleanup(void) |
| 630 | { |
| 631 | lockdep_assert_held(&watchdog_mutex); |
| 632 | hardlockup_detector_perf_cleanup(); |
| 633 | } |
| 634 | |
| 635 | /** |
| 636 | * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes |
| 637 | * |
| 638 | * Caller must not hold the cpu hotplug rwsem. |
| 639 | */ |
| 640 | void lockup_detector_cleanup(void) |
| 641 | { |
| 642 | mutex_lock(&watchdog_mutex); |
| 643 | __lockup_detector_cleanup(); |
| 644 | mutex_unlock(&watchdog_mutex); |
| 645 | } |
| 646 | |
Thomas Gleixner | 6554fd8 | 2017-09-12 21:36:57 +0200 | [diff] [blame] | 647 | /** |
| 648 | * lockup_detector_soft_poweroff - Interface to stop lockup detector(s) |
| 649 | * |
| 650 | * Special interface for parisc. It prevents lockup detector warnings from |
| 651 | * the default pm_poweroff() function which busy loops forever. |
| 652 | */ |
| 653 | void lockup_detector_soft_poweroff(void) |
| 654 | { |
| 655 | watchdog_enabled = 0; |
| 656 | } |
| 657 | |
Ulrich Obergfell | 58cf690 | 2015-11-05 18:44:30 -0800 | [diff] [blame] | 658 | #ifdef CONFIG_SYSCTL |
| 659 | |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 660 | /* Propagate any changes to the watchdog threads */ |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 661 | static void proc_watchdog_update(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 662 | { |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 663 | /* Remove impossible cpus to keep sysctl output clean. */ |
| 664 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask); |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 665 | lockup_detector_reconfigure(); |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 666 | } |
| 667 | |
| 668 | /* |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 669 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter |
| 670 | * |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 671 | * caller | table->data points to | 'which' |
| 672 | * -------------------|----------------------------|-------------------------- |
| 673 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED | |
| 674 | * | | SOFT_WATCHDOG_ENABLED |
| 675 | * -------------------|----------------------------|-------------------------- |
| 676 | * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
| 677 | * -------------------|----------------------------|-------------------------- |
| 678 | * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 679 | */ |
| 680 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, |
| 681 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 682 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 683 | int err, old, *param = table->data; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 684 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 685 | mutex_lock(&watchdog_mutex); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 686 | |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 687 | if (!write) { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 688 | /* |
| 689 | * On read synchronize the userspace interface. This is a |
| 690 | * racy snapshot. |
| 691 | */ |
| 692 | *param = (watchdog_enabled & which) != 0; |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 693 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 694 | } else { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 695 | old = READ_ONCE(*param); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 696 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame] | 697 | if (!err && old != READ_ONCE(*param)) |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 698 | proc_watchdog_update(); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 699 | } |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 700 | mutex_unlock(&watchdog_mutex); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 701 | return err; |
| 702 | } |
| 703 | |
| 704 | /* |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 705 | * /proc/sys/kernel/watchdog |
| 706 | */ |
| 707 | int proc_watchdog(struct ctl_table *table, int write, |
| 708 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 709 | { |
| 710 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, |
| 711 | table, write, buffer, lenp, ppos); |
| 712 | } |
| 713 | |
| 714 | /* |
| 715 | * /proc/sys/kernel/nmi_watchdog |
| 716 | */ |
| 717 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
| 718 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 719 | { |
Thomas Gleixner | a994a31 | 2017-09-12 21:37:19 +0200 | [diff] [blame] | 720 | if (!nmi_watchdog_available && write) |
| 721 | return -ENOTSUPP; |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 722 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, |
| 723 | table, write, buffer, lenp, ppos); |
| 724 | } |
| 725 | |
| 726 | /* |
| 727 | * /proc/sys/kernel/soft_watchdog |
| 728 | */ |
| 729 | int proc_soft_watchdog(struct ctl_table *table, int write, |
| 730 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 731 | { |
| 732 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, |
| 733 | table, write, buffer, lenp, ppos); |
| 734 | } |
| 735 | |
| 736 | /* |
| 737 | * /proc/sys/kernel/watchdog_thresh |
| 738 | */ |
| 739 | int proc_watchdog_thresh(struct ctl_table *table, int write, |
| 740 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 741 | { |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 742 | int err, old; |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 743 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 744 | mutex_lock(&watchdog_mutex); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 745 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 746 | old = READ_ONCE(watchdog_thresh); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 747 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 748 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 749 | if (!err && write && old != READ_ONCE(watchdog_thresh)) |
| 750 | proc_watchdog_update(); |
Mandeep Singh Baines | e04ab2b | 2011-05-22 22:10:21 -0700 | [diff] [blame] | 751 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 752 | mutex_unlock(&watchdog_mutex); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 753 | return err; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 754 | } |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 755 | |
| 756 | /* |
| 757 | * The cpumask is the mask of possible cpus that the watchdog can run |
| 758 | * on, not the mask of cpus it is actually running on. This allows the |
| 759 | * user to specify a mask that will include cpus that have not yet |
| 760 | * been brought online, if desired. |
| 761 | */ |
| 762 | int proc_watchdog_cpumask(struct ctl_table *table, int write, |
| 763 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 764 | { |
| 765 | int err; |
| 766 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 767 | mutex_lock(&watchdog_mutex); |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 768 | |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 769 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
Thomas Gleixner | 05ba3de | 2017-09-12 21:37:08 +0200 | [diff] [blame] | 770 | if (!err && write) |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 771 | proc_watchdog_update(); |
Thomas Gleixner | 5490125 | 2017-09-12 21:36:59 +0200 | [diff] [blame] | 772 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 773 | mutex_unlock(&watchdog_mutex); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 774 | return err; |
| 775 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 776 | #endif /* CONFIG_SYSCTL */ |
| 777 | |
Peter Zijlstra | 004417a | 2010-11-25 18:38:29 +0100 | [diff] [blame] | 778 | void __init lockup_detector_init(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 779 | { |
Frederic Weisbecker | 13316b3 | 2017-10-27 04:42:29 +0200 | [diff] [blame] | 780 | if (tick_nohz_full_enabled()) |
Frederic Weisbecker | 314b08ff | 2015-09-04 15:45:09 -0700 | [diff] [blame] | 781 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
Frederic Weisbecker | 13316b3 | 2017-10-27 04:42:29 +0200 | [diff] [blame] | 782 | |
Frederic Weisbecker | de20155 | 2017-10-27 04:42:35 +0200 | [diff] [blame] | 783 | cpumask_copy(&watchdog_cpumask, |
| 784 | housekeeping_cpumask(HK_FLAG_TIMER)); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 785 | |
Thomas Gleixner | a994a31 | 2017-09-12 21:37:19 +0200 | [diff] [blame] | 786 | if (!watchdog_nmi_probe()) |
| 787 | nmi_watchdog_available = true; |
Thomas Gleixner | 5587185 | 2017-10-04 10:03:04 +0200 | [diff] [blame] | 788 | lockup_detector_setup(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 789 | } |