Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 1 | /* |
| 2 | * Detect hard and soft lockups on a system |
| 3 | * |
| 4 | * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc. |
| 5 | * |
Fernando Luis Vázquez Cao | 86f5e6a | 2012-02-09 17:42:22 -0500 | [diff] [blame] | 6 | * Note: Most of this code is borrowed heavily from the original softlockup |
| 7 | * detector, so thanks to Ingo for the initial implementation. |
| 8 | * 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] | 9 | * to those contributors as well. |
| 10 | */ |
| 11 | |
Kefeng Wang | 5f92a7b | 2017-07-14 14:49:46 -0700 | [diff] [blame] | 12 | #define pr_fmt(fmt) "watchdog: " fmt |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 13 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 14 | #include <linux/mm.h> |
| 15 | #include <linux/cpu.h> |
| 16 | #include <linux/nmi.h> |
| 17 | #include <linux/init.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 18 | #include <linux/module.h> |
| 19 | #include <linux/sysctl.h> |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 20 | #include <linux/smpboot.h> |
Clark Williams | 8bd75c7 | 2013-02-07 09:47:07 -0600 | [diff] [blame] | 21 | #include <linux/sched/rt.h> |
Ingo Molnar | ae7e81c | 2017-02-01 18:07:51 +0100 | [diff] [blame] | 22 | #include <uapi/linux/sched/types.h> |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 23 | #include <linux/tick.h> |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 24 | #include <linux/workqueue.h> |
Ingo Molnar | e601757 | 2017-02-01 16:36:40 +0100 | [diff] [blame] | 25 | #include <linux/sched/clock.h> |
Ingo Molnar | b17b015 | 2017-02-08 18:51:35 +0100 | [diff] [blame] | 26 | #include <linux/sched/debug.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 27 | |
| 28 | #include <asm/irq_regs.h> |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 29 | #include <linux/kvm_para.h> |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 30 | #include <linux/kthread.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 31 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 32 | static DEFINE_MUTEX(watchdog_mutex); |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 33 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 34 | #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG) |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 35 | # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED) |
| 36 | # define NMI_WATCHDOG_DEFAULT 1 |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 37 | #else |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 38 | # define WATCHDOG_DEFAULT (SOFT_WATCHDOG_ENABLED) |
| 39 | # define NMI_WATCHDOG_DEFAULT 0 |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 40 | #endif |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 41 | |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 42 | unsigned long __read_mostly watchdog_enabled; |
| 43 | int __read_mostly watchdog_user_enabled = 1; |
| 44 | int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT; |
| 45 | int __read_mostly soft_watchdog_user_enabled = 1; |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 46 | int __read_mostly watchdog_thresh = 10; |
| 47 | |
| 48 | struct cpumask watchdog_allowed_mask __read_mostly; |
| 49 | static bool softlockup_threads_initialized __read_mostly; |
| 50 | |
| 51 | struct cpumask watchdog_cpumask __read_mostly; |
| 52 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); |
| 53 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 54 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 55 | /* |
| 56 | * Should we panic when a soft-lockup or hard-lockup occurs: |
| 57 | */ |
| 58 | unsigned int __read_mostly hardlockup_panic = |
| 59 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; |
| 60 | /* |
| 61 | * We may not want to enable hard lockup detection by default in all cases, |
| 62 | * for example when running the kernel as a guest on a hypervisor. In these |
| 63 | * cases this function can be called to disable hard lockup detection. This |
| 64 | * function should only be executed once by the boot processor before the |
| 65 | * kernel command line parameters are parsed, because otherwise it is not |
| 66 | * possible to override this in hardlockup_panic_setup(). |
| 67 | */ |
Thomas Gleixner | 7a35582 | 2017-09-12 21:37:02 +0200 | [diff] [blame] | 68 | void __init hardlockup_detector_disable(void) |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 69 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 70 | nmi_watchdog_user_enabled = 0; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 71 | } |
| 72 | |
| 73 | static int __init hardlockup_panic_setup(char *str) |
| 74 | { |
| 75 | if (!strncmp(str, "panic", 5)) |
| 76 | hardlockup_panic = 1; |
| 77 | else if (!strncmp(str, "nopanic", 7)) |
| 78 | hardlockup_panic = 0; |
| 79 | else if (!strncmp(str, "0", 1)) |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 80 | nmi_watchdog_user_enabled = 0; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 81 | else if (!strncmp(str, "1", 1)) |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 82 | nmi_watchdog_user_enabled = 1; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 83 | return 1; |
| 84 | } |
| 85 | __setup("nmi_watchdog=", hardlockup_panic_setup); |
| 86 | |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 87 | # ifdef CONFIG_SMP |
| 88 | int __read_mostly sysctl_hardlockup_all_cpu_backtrace; |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 89 | |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 90 | static int __init hardlockup_all_cpu_backtrace_setup(char *str) |
| 91 | { |
| 92 | sysctl_hardlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0); |
| 93 | return 1; |
| 94 | } |
| 95 | __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup); |
| 96 | # endif /* CONFIG_SMP */ |
| 97 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 98 | |
Ulrich Obergfell | ec6a906 | 2015-09-04 15:45:28 -0700 | [diff] [blame] | 99 | /* |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 100 | * These functions can be overridden if an architecture implements its |
| 101 | * own hardlockup detector. |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 102 | * |
| 103 | * watchdog_nmi_enable/disable can be implemented to start and stop when |
| 104 | * softlockup watchdog threads start and stop. The arch must select the |
| 105 | * SOFTLOCKUP_DETECTOR Kconfig. |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 106 | */ |
| 107 | int __weak watchdog_nmi_enable(unsigned int cpu) |
| 108 | { |
| 109 | return 0; |
| 110 | } |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 111 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 112 | void __weak watchdog_nmi_disable(unsigned int cpu) |
| 113 | { |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 114 | hardlockup_detector_perf_disable(); |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 115 | } |
| 116 | |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 117 | /** |
| 118 | * watchdog_nmi_reconfigure - Optional function to reconfigure NMI watchdogs |
| 119 | * @run: If false stop the watchdogs on all enabled CPUs |
| 120 | * If true start the watchdogs on all enabled CPUs |
| 121 | * |
| 122 | * The core call order is: |
| 123 | * watchdog_nmi_reconfigure(false); |
| 124 | * update_variables(); |
| 125 | * watchdog_nmi_reconfigure(true); |
| 126 | * |
| 127 | * The second call which starts the watchdogs again guarantees that the |
| 128 | * following variables are stable across the call. |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 129 | * - watchdog_enabled |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 130 | * - watchdog_thresh |
| 131 | * - watchdog_cpumask |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 132 | * |
| 133 | * After the call the variables can be changed again. |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 134 | */ |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 135 | void __weak watchdog_nmi_reconfigure(bool run) { } |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 136 | |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 137 | /** |
| 138 | * lockup_detector_update_enable - Update the sysctl enable bit |
| 139 | * |
| 140 | * Caller needs to make sure that the NMI/perf watchdogs are off, so this |
| 141 | * can't race with watchdog_nmi_disable(). |
| 142 | */ |
| 143 | static void lockup_detector_update_enable(void) |
| 144 | { |
| 145 | watchdog_enabled = 0; |
| 146 | if (!watchdog_user_enabled) |
| 147 | return; |
| 148 | if (nmi_watchdog_user_enabled) |
| 149 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; |
| 150 | if (soft_watchdog_user_enabled) |
| 151 | watchdog_enabled |= SOFT_WATCHDOG_ENABLED; |
| 152 | } |
| 153 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 154 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR |
| 155 | |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 156 | /* Global variables, exported for sysctl */ |
| 157 | unsigned int __read_mostly softlockup_panic = |
| 158 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 159 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 160 | static u64 __read_mostly sample_period; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 161 | |
| 162 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); |
| 163 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); |
| 164 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
| 165 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 166 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 167 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
| 168 | static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 169 | static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 170 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 171 | static unsigned long soft_lockup_nmi_warn; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 172 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 173 | static int __init softlockup_panic_setup(char *str) |
| 174 | { |
| 175 | softlockup_panic = simple_strtoul(str, NULL, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 176 | return 1; |
| 177 | } |
| 178 | __setup("softlockup_panic=", softlockup_panic_setup); |
| 179 | |
| 180 | static int __init nowatchdog_setup(char *str) |
| 181 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 182 | watchdog_user_enabled = 0; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 183 | return 1; |
| 184 | } |
| 185 | __setup("nowatchdog", nowatchdog_setup); |
| 186 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 187 | static int __init nosoftlockup_setup(char *str) |
| 188 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 189 | soft_watchdog_user_enabled = 0; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 190 | return 1; |
| 191 | } |
| 192 | __setup("nosoftlockup", nosoftlockup_setup); |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 193 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 194 | #ifdef CONFIG_SMP |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 195 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; |
| 196 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 197 | static int __init softlockup_all_cpu_backtrace_setup(char *str) |
| 198 | { |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 199 | sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0); |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 200 | return 1; |
| 201 | } |
| 202 | __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 203 | #endif |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 204 | |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 205 | static void __lockup_detector_cleanup(void); |
| 206 | |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 207 | /* |
| 208 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- |
| 209 | * lockups can have false positives under extreme conditions. So we generally |
| 210 | * want a higher threshold for soft lockups than for hard lockups. So we couple |
| 211 | * the thresholds with a factor: we make the soft threshold twice the amount of |
| 212 | * time the hard threshold is. |
| 213 | */ |
Ingo Molnar | 6e9101a | 2011-05-24 05:43:18 +0200 | [diff] [blame] | 214 | static int get_softlockup_thresh(void) |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 215 | { |
| 216 | return watchdog_thresh * 2; |
| 217 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 218 | |
| 219 | /* |
| 220 | * Returns seconds, approximately. We don't need nanosecond |
| 221 | * resolution, and we don't need to waste time with a big divide when |
| 222 | * 2^30ns == 1.074s. |
| 223 | */ |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 224 | static unsigned long get_timestamp(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 225 | { |
Cyril Bur | 545a2bf | 2015-02-12 15:01:24 -0800 | [diff] [blame] | 226 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 227 | } |
| 228 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 229 | static void set_sample_period(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 230 | { |
| 231 | /* |
Mandeep Singh Baines | 586692a | 2011-05-22 22:10:22 -0700 | [diff] [blame] | 232 | * convert watchdog_thresh from seconds to ns |
Fernando Luis Vázquez Cao | 86f5e6a | 2012-02-09 17:42:22 -0500 | [diff] [blame] | 233 | * the divide by 5 is to give hrtimer several chances (two |
| 234 | * or three with the current relation between the soft |
| 235 | * and hard thresholds) to increment before the |
| 236 | * hardlockup detector generates a warning |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 237 | */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 238 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
Thomas Gleixner | 7edaeb6 | 2017-08-15 09:50:13 +0200 | [diff] [blame] | 239 | watchdog_update_hrtimer_threshold(sample_period); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 240 | } |
| 241 | |
| 242 | /* Commands for resetting the watchdog */ |
| 243 | static void __touch_watchdog(void) |
| 244 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 245 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 246 | } |
| 247 | |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 248 | /** |
| 249 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls |
| 250 | * |
| 251 | * Call when the scheduler may have stalled for legitimate reasons |
| 252 | * preventing the watchdog task from executing - e.g. the scheduler |
| 253 | * entering idle state. This should only be used for scheduler events. |
| 254 | * Use touch_softlockup_watchdog() for everything else. |
| 255 | */ |
| 256 | void touch_softlockup_watchdog_sched(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 257 | { |
Andrew Morton | 7861144 | 2014-04-18 15:07:12 -0700 | [diff] [blame] | 258 | /* |
| 259 | * Preemption can be enabled. It doesn't matter which CPU's timestamp |
| 260 | * gets zeroed here, so use the raw_ operation. |
| 261 | */ |
| 262 | raw_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 263 | } |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 264 | |
| 265 | void touch_softlockup_watchdog(void) |
| 266 | { |
| 267 | touch_softlockup_watchdog_sched(); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 268 | wq_watchdog_touch(raw_smp_processor_id()); |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 269 | } |
Ingo Molnar | 0167c78 | 2010-05-13 08:53:33 +0200 | [diff] [blame] | 270 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 271 | |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 272 | void touch_all_softlockup_watchdogs(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 273 | { |
| 274 | int cpu; |
| 275 | |
| 276 | /* |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 277 | * watchdog_mutex cannpt be taken here, as this might be called |
| 278 | * from (soft)interrupt context, so the access to |
| 279 | * watchdog_allowed_cpumask might race with a concurrent update. |
| 280 | * |
| 281 | * The watchdog time stamp can race against a concurrent real |
| 282 | * update as well, the only side effect might be a cycle delay for |
| 283 | * the softlockup check. |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 284 | */ |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 285 | for_each_cpu(cpu, &watchdog_allowed_mask) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 286 | per_cpu(watchdog_touch_ts, cpu) = 0; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 287 | wq_watchdog_touch(-1); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 288 | } |
| 289 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 290 | void touch_softlockup_watchdog_sync(void) |
| 291 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 292 | __this_cpu_write(softlockup_touch_sync, true); |
| 293 | __this_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 294 | } |
| 295 | |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 296 | static int is_softlockup(unsigned long touch_ts) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 297 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 298 | unsigned long now = get_timestamp(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 299 | |
Ulrich Obergfell | 39d2da2 | 2015-11-05 18:44:56 -0800 | [diff] [blame] | 300 | if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 301 | /* Warn about unreasonable delays. */ |
| 302 | if (time_after(now, touch_ts + get_softlockup_thresh())) |
| 303 | return now - touch_ts; |
| 304 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 305 | return 0; |
| 306 | } |
| 307 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 308 | /* watchdog detector functions */ |
| 309 | bool is_hardlockup(void) |
| 310 | { |
| 311 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
| 312 | |
| 313 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
| 314 | return true; |
| 315 | |
| 316 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
| 317 | return false; |
| 318 | } |
| 319 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 320 | static void watchdog_interrupt_count(void) |
| 321 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 322 | __this_cpu_inc(hrtimer_interrupts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 323 | } |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 324 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 325 | /* watchdog kicker functions */ |
| 326 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) |
| 327 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 328 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 329 | struct pt_regs *regs = get_irq_regs(); |
| 330 | int duration; |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 331 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 332 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 333 | if (!watchdog_enabled) |
Don Zickus | b94f511 | 2017-01-24 15:17:53 -0800 | [diff] [blame] | 334 | return HRTIMER_NORESTART; |
| 335 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 336 | /* kick the hardlockup detector */ |
| 337 | watchdog_interrupt_count(); |
| 338 | |
| 339 | /* kick the softlockup detector */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 340 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 341 | |
| 342 | /* .. and repeat */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 343 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 344 | |
| 345 | if (touch_ts == 0) { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 346 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 347 | /* |
| 348 | * If the time stamp was touched atomically |
| 349 | * make sure the scheduler tick is up to date. |
| 350 | */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 351 | __this_cpu_write(softlockup_touch_sync, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 352 | sched_clock_tick(); |
| 353 | } |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 354 | |
| 355 | /* Clear the guest paused flag on watchdog reset */ |
| 356 | kvm_check_and_clear_guest_paused(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 357 | __touch_watchdog(); |
| 358 | return HRTIMER_RESTART; |
| 359 | } |
| 360 | |
| 361 | /* check for a softlockup |
| 362 | * This is done by making sure a high priority task is |
| 363 | * being scheduled. The task touches the watchdog to |
| 364 | * indicate it is getting cpu time. If it hasn't then |
| 365 | * this is a good indication some task is hogging the cpu |
| 366 | */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 367 | duration = is_softlockup(touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 368 | if (unlikely(duration)) { |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 369 | /* |
| 370 | * If a virtual machine is stopped by the host it can look to |
| 371 | * the watchdog like a soft lockup, check to see if the host |
| 372 | * stopped the vm before we issue the warning |
| 373 | */ |
| 374 | if (kvm_check_and_clear_guest_paused()) |
| 375 | return HRTIMER_RESTART; |
| 376 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 377 | /* only warn once */ |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 378 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
| 379 | /* |
| 380 | * When multiple processes are causing softlockups the |
| 381 | * softlockup detector only warns on the first one |
| 382 | * because the code relies on a full quiet cycle to |
| 383 | * re-arm. The second process prevents the quiet cycle |
| 384 | * and never gets reported. Use task pointers to detect |
| 385 | * this. |
| 386 | */ |
| 387 | if (__this_cpu_read(softlockup_task_ptr_saved) != |
| 388 | current) { |
| 389 | __this_cpu_write(soft_watchdog_warn, false); |
| 390 | __touch_watchdog(); |
| 391 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 392 | return HRTIMER_RESTART; |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 393 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 394 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 395 | if (softlockup_all_cpu_backtrace) { |
| 396 | /* Prevent multiple soft-lockup reports if one cpu is already |
| 397 | * engaged in dumping cpu back traces |
| 398 | */ |
| 399 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { |
| 400 | /* Someone else will report us. Let's give up */ |
| 401 | __this_cpu_write(soft_watchdog_warn, true); |
| 402 | return HRTIMER_RESTART; |
| 403 | } |
| 404 | } |
| 405 | |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 406 | 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] | 407 | smp_processor_id(), duration, |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 408 | current->comm, task_pid_nr(current)); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 409 | __this_cpu_write(softlockup_task_ptr_saved, current); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 410 | print_modules(); |
| 411 | print_irqtrace_events(current); |
| 412 | if (regs) |
| 413 | show_regs(regs); |
| 414 | else |
| 415 | dump_stack(); |
| 416 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 417 | if (softlockup_all_cpu_backtrace) { |
| 418 | /* Avoid generating two back traces for current |
| 419 | * given that one is already made above |
| 420 | */ |
| 421 | trigger_allbutself_cpu_backtrace(); |
| 422 | |
| 423 | clear_bit(0, &soft_lockup_nmi_warn); |
| 424 | /* Barrier to sync with other cpus */ |
| 425 | smp_mb__after_atomic(); |
| 426 | } |
| 427 | |
Josh Hunt | 69361ee | 2014-08-08 14:22:31 -0700 | [diff] [blame] | 428 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 429 | if (softlockup_panic) |
| 430 | panic("softlockup: hung tasks"); |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 431 | __this_cpu_write(soft_watchdog_warn, true); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 432 | } else |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 433 | __this_cpu_write(soft_watchdog_warn, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 434 | |
| 435 | return HRTIMER_RESTART; |
| 436 | } |
| 437 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 438 | static void watchdog_set_prio(unsigned int policy, unsigned int prio) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 439 | { |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 440 | struct sched_param param = { .sched_priority = prio }; |
| 441 | |
| 442 | sched_setscheduler(current, policy, ¶m); |
| 443 | } |
| 444 | |
| 445 | static void watchdog_enable(unsigned int cpu) |
| 446 | { |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 447 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 448 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 449 | /* |
| 450 | * Start the timer first to prevent the NMI watchdog triggering |
| 451 | * before the timer has a chance to fire. |
| 452 | */ |
Bjørn Mork | 3935e895 | 2012-12-19 20:51:31 +0100 | [diff] [blame] | 453 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
| 454 | hrtimer->function = watchdog_timer_fn; |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 455 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 456 | HRTIMER_MODE_REL_PINNED); |
| 457 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 458 | /* Initialize timestamp */ |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 459 | __touch_watchdog(); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 460 | /* Enable the perf event */ |
| 461 | watchdog_nmi_enable(cpu); |
| 462 | |
| 463 | watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 464 | } |
| 465 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 466 | static void watchdog_disable(unsigned int cpu) |
| 467 | { |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 468 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 469 | |
| 470 | watchdog_set_prio(SCHED_NORMAL, 0); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 471 | /* |
| 472 | * Disable the perf event first. That prevents that a large delay |
| 473 | * between disabling the timer and disabling the perf event causes |
| 474 | * the perf NMI to detect a false positive. |
| 475 | */ |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 476 | watchdog_nmi_disable(cpu); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 477 | hrtimer_cancel(hrtimer); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 478 | } |
| 479 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 480 | static void watchdog_cleanup(unsigned int cpu, bool online) |
| 481 | { |
| 482 | watchdog_disable(cpu); |
| 483 | } |
| 484 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 485 | static int watchdog_should_run(unsigned int cpu) |
| 486 | { |
| 487 | return __this_cpu_read(hrtimer_interrupts) != |
| 488 | __this_cpu_read(soft_lockup_hrtimer_cnt); |
| 489 | } |
| 490 | |
| 491 | /* |
| 492 | * The watchdog thread function - touches the timestamp. |
| 493 | * |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 494 | * It only runs once every sample_period seconds (4 seconds by |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 495 | * default) to reset the softlockup timestamp. If this gets delayed |
| 496 | * for more than 2*watchdog_thresh seconds then the debug-printout |
| 497 | * triggers in watchdog_timer_fn(). |
| 498 | */ |
| 499 | static void watchdog(unsigned int cpu) |
| 500 | { |
| 501 | __this_cpu_write(soft_lockup_hrtimer_cnt, |
| 502 | __this_cpu_read(hrtimer_interrupts)); |
| 503 | __touch_watchdog(); |
| 504 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 505 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 506 | static struct smp_hotplug_thread watchdog_threads = { |
| 507 | .store = &softlockup_watchdog, |
| 508 | .thread_should_run = watchdog_should_run, |
| 509 | .thread_fn = watchdog, |
| 510 | .thread_comm = "watchdog/%u", |
| 511 | .setup = watchdog_enable, |
| 512 | .cleanup = watchdog_cleanup, |
| 513 | .park = watchdog_disable, |
| 514 | .unpark = watchdog_enable, |
| 515 | }; |
| 516 | |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 517 | static void softlockup_update_smpboot_threads(void) |
| 518 | { |
| 519 | lockdep_assert_held(&watchdog_mutex); |
| 520 | |
| 521 | if (!softlockup_threads_initialized) |
| 522 | return; |
| 523 | |
| 524 | smpboot_update_cpumask_percpu_thread(&watchdog_threads, |
| 525 | &watchdog_allowed_mask); |
| 526 | __lockup_detector_cleanup(); |
| 527 | } |
| 528 | |
| 529 | /* Temporarily park all watchdog threads */ |
| 530 | static void softlockup_park_all_threads(void) |
| 531 | { |
| 532 | cpumask_clear(&watchdog_allowed_mask); |
| 533 | softlockup_update_smpboot_threads(); |
| 534 | } |
| 535 | |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 536 | /* Unpark enabled threads */ |
| 537 | static void softlockup_unpark_threads(void) |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 538 | { |
| 539 | cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask); |
| 540 | softlockup_update_smpboot_threads(); |
| 541 | } |
| 542 | |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 543 | static void softlockup_reconfigure_threads(void) |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 544 | { |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 545 | watchdog_nmi_reconfigure(false); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 546 | softlockup_park_all_threads(); |
| 547 | set_sample_period(); |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 548 | lockup_detector_update_enable(); |
| 549 | if (watchdog_enabled && watchdog_thresh) |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 550 | softlockup_unpark_threads(); |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 551 | watchdog_nmi_reconfigure(true); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 552 | } |
| 553 | |
| 554 | /* |
| 555 | * Create the watchdog thread infrastructure. |
| 556 | * |
| 557 | * The threads are not unparked as watchdog_allowed_mask is empty. When |
| 558 | * the threads are sucessfully initialized, take the proper locks and |
| 559 | * unpark the threads in the watchdog_cpumask if the watchdog is enabled. |
| 560 | */ |
| 561 | static __init void softlockup_init_threads(void) |
| 562 | { |
| 563 | int ret; |
| 564 | |
| 565 | /* |
| 566 | * If sysctl is off and watchdog got disabled on the command line, |
| 567 | * nothing to do here. |
| 568 | */ |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 569 | lockup_detector_update_enable(); |
| 570 | |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 571 | if (!IS_ENABLED(CONFIG_SYSCTL) && |
| 572 | !(watchdog_enabled && watchdog_thresh)) |
| 573 | return; |
| 574 | |
| 575 | ret = smpboot_register_percpu_thread_cpumask(&watchdog_threads, |
| 576 | &watchdog_allowed_mask); |
| 577 | if (ret) { |
| 578 | pr_err("Failed to initialize soft lockup detector threads\n"); |
| 579 | return; |
| 580 | } |
| 581 | |
| 582 | mutex_lock(&watchdog_mutex); |
| 583 | softlockup_threads_initialized = true; |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 584 | softlockup_reconfigure_threads(); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 585 | mutex_unlock(&watchdog_mutex); |
| 586 | } |
| 587 | |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 588 | #else /* CONFIG_SOFTLOCKUP_DETECTOR */ |
| 589 | static inline int watchdog_park_threads(void) { return 0; } |
| 590 | static inline void watchdog_unpark_threads(void) { } |
| 591 | static inline int watchdog_enable_all_cpus(void) { return 0; } |
| 592 | static inline void watchdog_disable_all_cpus(void) { } |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 593 | static inline void softlockup_init_threads(void) { } |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 594 | static void softlockup_reconfigure_threads(void) |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 595 | { |
| 596 | watchdog_nmi_reconfigure(false); |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 597 | lockup_detector_update_enable(); |
Thomas Gleixner | 6592ad2 | 2017-09-12 21:37:16 +0200 | [diff] [blame] | 598 | watchdog_nmi_reconfigure(true); |
| 599 | } |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 600 | #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */ |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 601 | |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 602 | static void __lockup_detector_cleanup(void) |
| 603 | { |
| 604 | lockdep_assert_held(&watchdog_mutex); |
| 605 | hardlockup_detector_perf_cleanup(); |
| 606 | } |
| 607 | |
| 608 | /** |
| 609 | * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes |
| 610 | * |
| 611 | * Caller must not hold the cpu hotplug rwsem. |
| 612 | */ |
| 613 | void lockup_detector_cleanup(void) |
| 614 | { |
| 615 | mutex_lock(&watchdog_mutex); |
| 616 | __lockup_detector_cleanup(); |
| 617 | mutex_unlock(&watchdog_mutex); |
| 618 | } |
| 619 | |
Thomas Gleixner | 6554fd8 | 2017-09-12 21:36:57 +0200 | [diff] [blame] | 620 | /** |
| 621 | * lockup_detector_soft_poweroff - Interface to stop lockup detector(s) |
| 622 | * |
| 623 | * Special interface for parisc. It prevents lockup detector warnings from |
| 624 | * the default pm_poweroff() function which busy loops forever. |
| 625 | */ |
| 626 | void lockup_detector_soft_poweroff(void) |
| 627 | { |
| 628 | watchdog_enabled = 0; |
| 629 | } |
| 630 | |
Ulrich Obergfell | 58cf690 | 2015-11-05 18:44:30 -0800 | [diff] [blame] | 631 | #ifdef CONFIG_SYSCTL |
| 632 | |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 633 | /* Propagate any changes to the watchdog threads */ |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 634 | static void proc_watchdog_update(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 635 | { |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 636 | /* Remove impossible cpus to keep sysctl output clean. */ |
| 637 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask); |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 638 | softlockup_reconfigure_threads(); |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 639 | } |
| 640 | |
| 641 | /* |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 642 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter |
| 643 | * |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame] | 644 | * caller | table->data points to | 'which' |
| 645 | * -------------------|----------------------------|-------------------------- |
| 646 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED | |
| 647 | * | | SOFT_WATCHDOG_ENABLED |
| 648 | * -------------------|----------------------------|-------------------------- |
| 649 | * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
| 650 | * -------------------|----------------------------|-------------------------- |
| 651 | * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 652 | */ |
| 653 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, |
| 654 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 655 | { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 656 | int err, old, *param = table->data; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 657 | |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 658 | cpu_hotplug_disable(); |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 659 | mutex_lock(&watchdog_mutex); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 660 | |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 661 | if (!write) { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 662 | /* |
| 663 | * On read synchronize the userspace interface. This is a |
| 664 | * racy snapshot. |
| 665 | */ |
| 666 | *param = (watchdog_enabled & which) != 0; |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 667 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 668 | } else { |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 669 | old = READ_ONCE(*param); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 670 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Thomas Gleixner | 0915498 | 2017-09-12 21:37:17 +0200 | [diff] [blame^] | 671 | if (!err && old != READ_ONCE(*param)) |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 672 | proc_watchdog_update(); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 673 | } |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 674 | mutex_unlock(&watchdog_mutex); |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 675 | cpu_hotplug_enable(); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 676 | return err; |
| 677 | } |
| 678 | |
| 679 | /* |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 680 | * /proc/sys/kernel/watchdog |
| 681 | */ |
| 682 | int proc_watchdog(struct ctl_table *table, int write, |
| 683 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 684 | { |
| 685 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, |
| 686 | table, write, buffer, lenp, ppos); |
| 687 | } |
| 688 | |
| 689 | /* |
| 690 | * /proc/sys/kernel/nmi_watchdog |
| 691 | */ |
| 692 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
| 693 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 694 | { |
| 695 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, |
| 696 | table, write, buffer, lenp, ppos); |
| 697 | } |
| 698 | |
| 699 | /* |
| 700 | * /proc/sys/kernel/soft_watchdog |
| 701 | */ |
| 702 | int proc_soft_watchdog(struct ctl_table *table, int write, |
| 703 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 704 | { |
| 705 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, |
| 706 | table, write, buffer, lenp, ppos); |
| 707 | } |
| 708 | |
| 709 | /* |
| 710 | * /proc/sys/kernel/watchdog_thresh |
| 711 | */ |
| 712 | int proc_watchdog_thresh(struct ctl_table *table, int write, |
| 713 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 714 | { |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 715 | int err, old; |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 716 | |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 717 | cpu_hotplug_disable(); |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 718 | mutex_lock(&watchdog_mutex); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 719 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 720 | old = READ_ONCE(watchdog_thresh); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 721 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 722 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 723 | if (!err && write && old != READ_ONCE(watchdog_thresh)) |
| 724 | proc_watchdog_update(); |
Mandeep Singh Baines | e04ab2b | 2011-05-22 22:10:21 -0700 | [diff] [blame] | 725 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 726 | mutex_unlock(&watchdog_mutex); |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 727 | cpu_hotplug_enable(); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 728 | return err; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 729 | } |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 730 | |
| 731 | /* |
| 732 | * The cpumask is the mask of possible cpus that the watchdog can run |
| 733 | * on, not the mask of cpus it is actually running on. This allows the |
| 734 | * user to specify a mask that will include cpus that have not yet |
| 735 | * been brought online, if desired. |
| 736 | */ |
| 737 | int proc_watchdog_cpumask(struct ctl_table *table, int write, |
| 738 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 739 | { |
| 740 | int err; |
| 741 | |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 742 | cpu_hotplug_disable(); |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 743 | mutex_lock(&watchdog_mutex); |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 744 | |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 745 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
Thomas Gleixner | 05ba3de | 2017-09-12 21:37:08 +0200 | [diff] [blame] | 746 | if (!err && write) |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 747 | proc_watchdog_update(); |
Thomas Gleixner | 5490125 | 2017-09-12 21:36:59 +0200 | [diff] [blame] | 748 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 749 | mutex_unlock(&watchdog_mutex); |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 750 | cpu_hotplug_enable(); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 751 | return err; |
| 752 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 753 | #endif /* CONFIG_SYSCTL */ |
| 754 | |
Peter Zijlstra | 004417a | 2010-11-25 18:38:29 +0100 | [diff] [blame] | 755 | void __init lockup_detector_init(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 756 | { |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 757 | #ifdef CONFIG_NO_HZ_FULL |
| 758 | if (tick_nohz_full_enabled()) { |
Frederic Weisbecker | 314b08ff | 2015-09-04 15:45:09 -0700 | [diff] [blame] | 759 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
| 760 | cpumask_copy(&watchdog_cpumask, housekeeping_mask); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 761 | } else |
| 762 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 763 | #else |
| 764 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 765 | #endif |
| 766 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 767 | softlockup_init_threads(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 768 | } |