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) |
| 35 | unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED | |
| 36 | NMI_WATCHDOG_ENABLED; |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 37 | #else |
Babu Moger | 249e52e | 2016-12-14 15:06:21 -0800 | [diff] [blame] | 38 | unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED; |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 39 | #endif |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 40 | |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame^] | 41 | int __read_mostly nmi_watchdog_user_enabled; |
| 42 | int __read_mostly soft_watchdog_user_enabled; |
| 43 | int __read_mostly watchdog_user_enabled; |
| 44 | int __read_mostly watchdog_thresh = 10; |
| 45 | |
| 46 | struct cpumask watchdog_allowed_mask __read_mostly; |
| 47 | static bool softlockup_threads_initialized __read_mostly; |
| 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 | { |
| 68 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
| 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)) |
| 78 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
| 79 | else if (!strncmp(str, "1", 1)) |
| 80 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; |
| 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 | { |
| 107 | return 0; |
| 108 | } |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 109 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 110 | void __weak watchdog_nmi_disable(unsigned int cpu) |
| 111 | { |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 112 | hardlockup_detector_perf_disable(); |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 113 | } |
| 114 | |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 115 | /* |
| 116 | * watchdog_nmi_reconfigure can be implemented to be notified after any |
| 117 | * watchdog configuration change. The arch hardlockup watchdog should |
| 118 | * respond to the following variables: |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame^] | 119 | * - watchdog_enabled |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 120 | * - watchdog_thresh |
| 121 | * - watchdog_cpumask |
| 122 | * - sysctl_hardlockup_all_cpu_backtrace |
| 123 | * - hardlockup_panic |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 124 | */ |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 125 | void __weak watchdog_nmi_reconfigure(void) { } |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 126 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 127 | #ifdef CONFIG_SOFTLOCKUP_DETECTOR |
| 128 | |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 129 | /* Global variables, exported for sysctl */ |
| 130 | unsigned int __read_mostly softlockup_panic = |
| 131 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 132 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 133 | static u64 __read_mostly sample_period; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 134 | |
| 135 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); |
| 136 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); |
| 137 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
| 138 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 139 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 140 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
| 141 | static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 142 | static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 143 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 144 | static unsigned long soft_lockup_nmi_warn; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 145 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 146 | static int __init softlockup_panic_setup(char *str) |
| 147 | { |
| 148 | softlockup_panic = simple_strtoul(str, NULL, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 149 | return 1; |
| 150 | } |
| 151 | __setup("softlockup_panic=", softlockup_panic_setup); |
| 152 | |
| 153 | static int __init nowatchdog_setup(char *str) |
| 154 | { |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 155 | watchdog_enabled = 0; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 156 | return 1; |
| 157 | } |
| 158 | __setup("nowatchdog", nowatchdog_setup); |
| 159 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 160 | static int __init nosoftlockup_setup(char *str) |
| 161 | { |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 162 | watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 163 | return 1; |
| 164 | } |
| 165 | __setup("nosoftlockup", nosoftlockup_setup); |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 166 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 167 | #ifdef CONFIG_SMP |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 168 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; |
| 169 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 170 | static int __init softlockup_all_cpu_backtrace_setup(char *str) |
| 171 | { |
Thomas Gleixner | 368a7e2 | 2017-09-12 21:37:07 +0200 | [diff] [blame] | 172 | sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0); |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 173 | return 1; |
| 174 | } |
| 175 | __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 176 | #endif |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 177 | |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 178 | static void __lockup_detector_cleanup(void); |
| 179 | |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 180 | /* |
| 181 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- |
| 182 | * lockups can have false positives under extreme conditions. So we generally |
| 183 | * want a higher threshold for soft lockups than for hard lockups. So we couple |
| 184 | * the thresholds with a factor: we make the soft threshold twice the amount of |
| 185 | * time the hard threshold is. |
| 186 | */ |
Ingo Molnar | 6e9101a | 2011-05-24 05:43:18 +0200 | [diff] [blame] | 187 | static int get_softlockup_thresh(void) |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 188 | { |
| 189 | return watchdog_thresh * 2; |
| 190 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 191 | |
| 192 | /* |
| 193 | * Returns seconds, approximately. We don't need nanosecond |
| 194 | * resolution, and we don't need to waste time with a big divide when |
| 195 | * 2^30ns == 1.074s. |
| 196 | */ |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 197 | static unsigned long get_timestamp(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 198 | { |
Cyril Bur | 545a2bf | 2015-02-12 15:01:24 -0800 | [diff] [blame] | 199 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 200 | } |
| 201 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 202 | static void set_sample_period(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 203 | { |
| 204 | /* |
Mandeep Singh Baines | 586692a | 2011-05-22 22:10:22 -0700 | [diff] [blame] | 205 | * convert watchdog_thresh from seconds to ns |
Fernando Luis Vázquez Cao | 86f5e6a | 2012-02-09 17:42:22 -0500 | [diff] [blame] | 206 | * the divide by 5 is to give hrtimer several chances (two |
| 207 | * or three with the current relation between the soft |
| 208 | * and hard thresholds) to increment before the |
| 209 | * hardlockup detector generates a warning |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 210 | */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 211 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
Thomas Gleixner | 7edaeb6 | 2017-08-15 09:50:13 +0200 | [diff] [blame] | 212 | watchdog_update_hrtimer_threshold(sample_period); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 213 | } |
| 214 | |
| 215 | /* Commands for resetting the watchdog */ |
| 216 | static void __touch_watchdog(void) |
| 217 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 218 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 219 | } |
| 220 | |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 221 | /** |
| 222 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls |
| 223 | * |
| 224 | * Call when the scheduler may have stalled for legitimate reasons |
| 225 | * preventing the watchdog task from executing - e.g. the scheduler |
| 226 | * entering idle state. This should only be used for scheduler events. |
| 227 | * Use touch_softlockup_watchdog() for everything else. |
| 228 | */ |
| 229 | void touch_softlockup_watchdog_sched(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 230 | { |
Andrew Morton | 7861144 | 2014-04-18 15:07:12 -0700 | [diff] [blame] | 231 | /* |
| 232 | * Preemption can be enabled. It doesn't matter which CPU's timestamp |
| 233 | * gets zeroed here, so use the raw_ operation. |
| 234 | */ |
| 235 | raw_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 236 | } |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 237 | |
| 238 | void touch_softlockup_watchdog(void) |
| 239 | { |
| 240 | touch_softlockup_watchdog_sched(); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 241 | wq_watchdog_touch(raw_smp_processor_id()); |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 242 | } |
Ingo Molnar | 0167c78 | 2010-05-13 08:53:33 +0200 | [diff] [blame] | 243 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 244 | |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 245 | void touch_all_softlockup_watchdogs(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 246 | { |
| 247 | int cpu; |
| 248 | |
| 249 | /* |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 250 | * watchdog_mutex cannpt be taken here, as this might be called |
| 251 | * from (soft)interrupt context, so the access to |
| 252 | * watchdog_allowed_cpumask might race with a concurrent update. |
| 253 | * |
| 254 | * The watchdog time stamp can race against a concurrent real |
| 255 | * update as well, the only side effect might be a cycle delay for |
| 256 | * the softlockup check. |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 257 | */ |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 258 | for_each_cpu(cpu, &watchdog_allowed_mask) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 259 | per_cpu(watchdog_touch_ts, cpu) = 0; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 260 | wq_watchdog_touch(-1); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 261 | } |
| 262 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 263 | void touch_softlockup_watchdog_sync(void) |
| 264 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 265 | __this_cpu_write(softlockup_touch_sync, true); |
| 266 | __this_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 267 | } |
| 268 | |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 269 | static int is_softlockup(unsigned long touch_ts) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 270 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 271 | unsigned long now = get_timestamp(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 272 | |
Ulrich Obergfell | 39d2da2 | 2015-11-05 18:44:56 -0800 | [diff] [blame] | 273 | if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 274 | /* Warn about unreasonable delays. */ |
| 275 | if (time_after(now, touch_ts + get_softlockup_thresh())) |
| 276 | return now - touch_ts; |
| 277 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 278 | return 0; |
| 279 | } |
| 280 | |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 281 | /* watchdog detector functions */ |
| 282 | bool is_hardlockup(void) |
| 283 | { |
| 284 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
| 285 | |
| 286 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
| 287 | return true; |
| 288 | |
| 289 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
| 290 | return false; |
| 291 | } |
| 292 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 293 | static void watchdog_interrupt_count(void) |
| 294 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 295 | __this_cpu_inc(hrtimer_interrupts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 296 | } |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 297 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 298 | /* watchdog kicker functions */ |
| 299 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) |
| 300 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 301 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 302 | struct pt_regs *regs = get_irq_regs(); |
| 303 | int duration; |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 304 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 305 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 306 | if (!watchdog_enabled) |
Don Zickus | b94f511 | 2017-01-24 15:17:53 -0800 | [diff] [blame] | 307 | return HRTIMER_NORESTART; |
| 308 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 309 | /* kick the hardlockup detector */ |
| 310 | watchdog_interrupt_count(); |
| 311 | |
| 312 | /* kick the softlockup detector */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 313 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 314 | |
| 315 | /* .. and repeat */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 316 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 317 | |
| 318 | if (touch_ts == 0) { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 319 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 320 | /* |
| 321 | * If the time stamp was touched atomically |
| 322 | * make sure the scheduler tick is up to date. |
| 323 | */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 324 | __this_cpu_write(softlockup_touch_sync, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 325 | sched_clock_tick(); |
| 326 | } |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 327 | |
| 328 | /* Clear the guest paused flag on watchdog reset */ |
| 329 | kvm_check_and_clear_guest_paused(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 330 | __touch_watchdog(); |
| 331 | return HRTIMER_RESTART; |
| 332 | } |
| 333 | |
| 334 | /* check for a softlockup |
| 335 | * This is done by making sure a high priority task is |
| 336 | * being scheduled. The task touches the watchdog to |
| 337 | * indicate it is getting cpu time. If it hasn't then |
| 338 | * this is a good indication some task is hogging the cpu |
| 339 | */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 340 | duration = is_softlockup(touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 341 | if (unlikely(duration)) { |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 342 | /* |
| 343 | * If a virtual machine is stopped by the host it can look to |
| 344 | * the watchdog like a soft lockup, check to see if the host |
| 345 | * stopped the vm before we issue the warning |
| 346 | */ |
| 347 | if (kvm_check_and_clear_guest_paused()) |
| 348 | return HRTIMER_RESTART; |
| 349 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 350 | /* only warn once */ |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 351 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
| 352 | /* |
| 353 | * When multiple processes are causing softlockups the |
| 354 | * softlockup detector only warns on the first one |
| 355 | * because the code relies on a full quiet cycle to |
| 356 | * re-arm. The second process prevents the quiet cycle |
| 357 | * and never gets reported. Use task pointers to detect |
| 358 | * this. |
| 359 | */ |
| 360 | if (__this_cpu_read(softlockup_task_ptr_saved) != |
| 361 | current) { |
| 362 | __this_cpu_write(soft_watchdog_warn, false); |
| 363 | __touch_watchdog(); |
| 364 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 365 | return HRTIMER_RESTART; |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 366 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 367 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 368 | if (softlockup_all_cpu_backtrace) { |
| 369 | /* Prevent multiple soft-lockup reports if one cpu is already |
| 370 | * engaged in dumping cpu back traces |
| 371 | */ |
| 372 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { |
| 373 | /* Someone else will report us. Let's give up */ |
| 374 | __this_cpu_write(soft_watchdog_warn, true); |
| 375 | return HRTIMER_RESTART; |
| 376 | } |
| 377 | } |
| 378 | |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 379 | 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] | 380 | smp_processor_id(), duration, |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 381 | current->comm, task_pid_nr(current)); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 382 | __this_cpu_write(softlockup_task_ptr_saved, current); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 383 | print_modules(); |
| 384 | print_irqtrace_events(current); |
| 385 | if (regs) |
| 386 | show_regs(regs); |
| 387 | else |
| 388 | dump_stack(); |
| 389 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 390 | if (softlockup_all_cpu_backtrace) { |
| 391 | /* Avoid generating two back traces for current |
| 392 | * given that one is already made above |
| 393 | */ |
| 394 | trigger_allbutself_cpu_backtrace(); |
| 395 | |
| 396 | clear_bit(0, &soft_lockup_nmi_warn); |
| 397 | /* Barrier to sync with other cpus */ |
| 398 | smp_mb__after_atomic(); |
| 399 | } |
| 400 | |
Josh Hunt | 69361ee | 2014-08-08 14:22:31 -0700 | [diff] [blame] | 401 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 402 | if (softlockup_panic) |
| 403 | panic("softlockup: hung tasks"); |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 404 | __this_cpu_write(soft_watchdog_warn, true); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 405 | } else |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 406 | __this_cpu_write(soft_watchdog_warn, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 407 | |
| 408 | return HRTIMER_RESTART; |
| 409 | } |
| 410 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 411 | static void watchdog_set_prio(unsigned int policy, unsigned int prio) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 412 | { |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 413 | struct sched_param param = { .sched_priority = prio }; |
| 414 | |
| 415 | sched_setscheduler(current, policy, ¶m); |
| 416 | } |
| 417 | |
| 418 | static void watchdog_enable(unsigned int cpu) |
| 419 | { |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 420 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 421 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 422 | /* |
| 423 | * Start the timer first to prevent the NMI watchdog triggering |
| 424 | * before the timer has a chance to fire. |
| 425 | */ |
Bjørn Mork | 3935e895 | 2012-12-19 20:51:31 +0100 | [diff] [blame] | 426 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
| 427 | hrtimer->function = watchdog_timer_fn; |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 428 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 429 | HRTIMER_MODE_REL_PINNED); |
| 430 | |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 431 | /* Initialize timestamp */ |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 432 | __touch_watchdog(); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 433 | /* Enable the perf event */ |
| 434 | watchdog_nmi_enable(cpu); |
| 435 | |
| 436 | watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 437 | } |
| 438 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 439 | static void watchdog_disable(unsigned int cpu) |
| 440 | { |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 441 | struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 442 | |
| 443 | watchdog_set_prio(SCHED_NORMAL, 0); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 444 | /* |
| 445 | * Disable the perf event first. That prevents that a large delay |
| 446 | * between disabling the timer and disabling the perf event causes |
| 447 | * the perf NMI to detect a false positive. |
| 448 | */ |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 449 | watchdog_nmi_disable(cpu); |
Thomas Gleixner | 01f0a02 | 2017-09-12 21:37:05 +0200 | [diff] [blame] | 450 | hrtimer_cancel(hrtimer); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 451 | } |
| 452 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 453 | static void watchdog_cleanup(unsigned int cpu, bool online) |
| 454 | { |
| 455 | watchdog_disable(cpu); |
| 456 | } |
| 457 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 458 | static int watchdog_should_run(unsigned int cpu) |
| 459 | { |
| 460 | return __this_cpu_read(hrtimer_interrupts) != |
| 461 | __this_cpu_read(soft_lockup_hrtimer_cnt); |
| 462 | } |
| 463 | |
| 464 | /* |
| 465 | * The watchdog thread function - touches the timestamp. |
| 466 | * |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 467 | * It only runs once every sample_period seconds (4 seconds by |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 468 | * default) to reset the softlockup timestamp. If this gets delayed |
| 469 | * for more than 2*watchdog_thresh seconds then the debug-printout |
| 470 | * triggers in watchdog_timer_fn(). |
| 471 | */ |
| 472 | static void watchdog(unsigned int cpu) |
| 473 | { |
| 474 | __this_cpu_write(soft_lockup_hrtimer_cnt, |
| 475 | __this_cpu_read(hrtimer_interrupts)); |
| 476 | __touch_watchdog(); |
| 477 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 478 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 479 | static struct smp_hotplug_thread watchdog_threads = { |
| 480 | .store = &softlockup_watchdog, |
| 481 | .thread_should_run = watchdog_should_run, |
| 482 | .thread_fn = watchdog, |
| 483 | .thread_comm = "watchdog/%u", |
| 484 | .setup = watchdog_enable, |
| 485 | .cleanup = watchdog_cleanup, |
| 486 | .park = watchdog_disable, |
| 487 | .unpark = watchdog_enable, |
| 488 | }; |
| 489 | |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 490 | static void softlockup_update_smpboot_threads(void) |
| 491 | { |
| 492 | lockdep_assert_held(&watchdog_mutex); |
| 493 | |
| 494 | if (!softlockup_threads_initialized) |
| 495 | return; |
| 496 | |
| 497 | smpboot_update_cpumask_percpu_thread(&watchdog_threads, |
| 498 | &watchdog_allowed_mask); |
| 499 | __lockup_detector_cleanup(); |
| 500 | } |
| 501 | |
| 502 | /* Temporarily park all watchdog threads */ |
| 503 | static void softlockup_park_all_threads(void) |
| 504 | { |
| 505 | cpumask_clear(&watchdog_allowed_mask); |
| 506 | softlockup_update_smpboot_threads(); |
| 507 | } |
| 508 | |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 509 | /* Unpark enabled threads */ |
| 510 | static void softlockup_unpark_threads(void) |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 511 | { |
| 512 | cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask); |
| 513 | softlockup_update_smpboot_threads(); |
| 514 | } |
| 515 | |
| 516 | static void softlockup_reconfigure_threads(bool enabled) |
| 517 | { |
| 518 | softlockup_park_all_threads(); |
| 519 | set_sample_period(); |
| 520 | if (enabled) |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 521 | softlockup_unpark_threads(); |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 522 | } |
| 523 | |
| 524 | /* |
| 525 | * Create the watchdog thread infrastructure. |
| 526 | * |
| 527 | * The threads are not unparked as watchdog_allowed_mask is empty. When |
| 528 | * the threads are sucessfully initialized, take the proper locks and |
| 529 | * unpark the threads in the watchdog_cpumask if the watchdog is enabled. |
| 530 | */ |
| 531 | static __init void softlockup_init_threads(void) |
| 532 | { |
| 533 | int ret; |
| 534 | |
| 535 | /* |
| 536 | * If sysctl is off and watchdog got disabled on the command line, |
| 537 | * nothing to do here. |
| 538 | */ |
| 539 | if (!IS_ENABLED(CONFIG_SYSCTL) && |
| 540 | !(watchdog_enabled && watchdog_thresh)) |
| 541 | return; |
| 542 | |
| 543 | ret = smpboot_register_percpu_thread_cpumask(&watchdog_threads, |
| 544 | &watchdog_allowed_mask); |
| 545 | if (ret) { |
| 546 | pr_err("Failed to initialize soft lockup detector threads\n"); |
| 547 | return; |
| 548 | } |
| 549 | |
| 550 | mutex_lock(&watchdog_mutex); |
| 551 | softlockup_threads_initialized = true; |
| 552 | softlockup_reconfigure_threads(watchdog_enabled && watchdog_thresh); |
| 553 | mutex_unlock(&watchdog_mutex); |
| 554 | } |
| 555 | |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 556 | #else /* CONFIG_SOFTLOCKUP_DETECTOR */ |
| 557 | static inline int watchdog_park_threads(void) { return 0; } |
| 558 | static inline void watchdog_unpark_threads(void) { } |
| 559 | static inline int watchdog_enable_all_cpus(void) { return 0; } |
| 560 | static inline void watchdog_disable_all_cpus(void) { } |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 561 | static inline void softlockup_init_threads(void) { } |
Thomas Gleixner | 2eb2527 | 2017-09-12 21:37:10 +0200 | [diff] [blame] | 562 | static inline void softlockup_reconfigure_threads(bool enabled) { } |
Thomas Gleixner | 2b9d7f2 | 2017-09-12 21:37:06 +0200 | [diff] [blame] | 563 | #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */ |
Nicholas Piggin | 05a4a95 | 2017-07-12 14:35:46 -0700 | [diff] [blame] | 564 | |
Thomas Gleixner | 941154b | 2017-09-12 21:37:04 +0200 | [diff] [blame] | 565 | static void __lockup_detector_cleanup(void) |
| 566 | { |
| 567 | lockdep_assert_held(&watchdog_mutex); |
| 568 | hardlockup_detector_perf_cleanup(); |
| 569 | } |
| 570 | |
| 571 | /** |
| 572 | * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes |
| 573 | * |
| 574 | * Caller must not hold the cpu hotplug rwsem. |
| 575 | */ |
| 576 | void lockup_detector_cleanup(void) |
| 577 | { |
| 578 | mutex_lock(&watchdog_mutex); |
| 579 | __lockup_detector_cleanup(); |
| 580 | mutex_unlock(&watchdog_mutex); |
| 581 | } |
| 582 | |
Thomas Gleixner | 6554fd8 | 2017-09-12 21:36:57 +0200 | [diff] [blame] | 583 | /** |
| 584 | * lockup_detector_soft_poweroff - Interface to stop lockup detector(s) |
| 585 | * |
| 586 | * Special interface for parisc. It prevents lockup detector warnings from |
| 587 | * the default pm_poweroff() function which busy loops forever. |
| 588 | */ |
| 589 | void lockup_detector_soft_poweroff(void) |
| 590 | { |
| 591 | watchdog_enabled = 0; |
| 592 | } |
| 593 | |
Ulrich Obergfell | 58cf690 | 2015-11-05 18:44:30 -0800 | [diff] [blame] | 594 | #ifdef CONFIG_SYSCTL |
| 595 | |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 596 | /* Propagate any changes to the watchdog threads */ |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 597 | static void proc_watchdog_update(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 598 | { |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 599 | /* Remove impossible cpus to keep sysctl output clean. */ |
| 600 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask); |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 601 | softlockup_reconfigure_threads(watchdog_enabled && watchdog_thresh); |
Nicholas Piggin | a10a842 | 2017-07-12 14:35:49 -0700 | [diff] [blame] | 602 | watchdog_nmi_reconfigure(); |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 603 | } |
| 604 | |
| 605 | /* |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 606 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter |
| 607 | * |
Thomas Gleixner | 7feeb9c | 2017-09-12 21:37:15 +0200 | [diff] [blame^] | 608 | * caller | table->data points to | 'which' |
| 609 | * -------------------|----------------------------|-------------------------- |
| 610 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED | |
| 611 | * | | SOFT_WATCHDOG_ENABLED |
| 612 | * -------------------|----------------------------|-------------------------- |
| 613 | * proc_nmi_watchdog | nmi_watchdog_user_enabled | NMI_WATCHDOG_ENABLED |
| 614 | * -------------------|----------------------------|-------------------------- |
| 615 | * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 616 | */ |
| 617 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, |
| 618 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 619 | { |
| 620 | int err, old, new; |
| 621 | int *watchdog_param = (int *)table->data; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 622 | |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 623 | cpu_hotplug_disable(); |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 624 | mutex_lock(&watchdog_mutex); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 625 | |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 626 | /* |
| 627 | * If the parameter is being read return the state of the corresponding |
| 628 | * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the |
| 629 | * run state of the lockup detectors. |
| 630 | */ |
| 631 | if (!write) { |
| 632 | *watchdog_param = (watchdog_enabled & which) != 0; |
| 633 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 634 | } else { |
| 635 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 636 | if (err) |
| 637 | goto out; |
| 638 | |
| 639 | /* |
| 640 | * There is a race window between fetching the current value |
| 641 | * from 'watchdog_enabled' and storing the new value. During |
| 642 | * this race window, watchdog_nmi_enable() can sneak in and |
| 643 | * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'. |
| 644 | * The 'cmpxchg' detects this race and the loop retries. |
| 645 | */ |
| 646 | do { |
| 647 | old = watchdog_enabled; |
| 648 | /* |
| 649 | * If the parameter value is not zero set the |
| 650 | * corresponding bit(s), else clear it(them). |
| 651 | */ |
| 652 | if (*watchdog_param) |
| 653 | new = old | which; |
| 654 | else |
| 655 | new = old & ~which; |
| 656 | } while (cmpxchg(&watchdog_enabled, old, new) != old); |
| 657 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 658 | if (old != new) |
| 659 | proc_watchdog_update(); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 660 | } |
| 661 | out: |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 662 | mutex_unlock(&watchdog_mutex); |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 663 | cpu_hotplug_enable(); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 664 | return err; |
| 665 | } |
| 666 | |
| 667 | /* |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 668 | * /proc/sys/kernel/watchdog |
| 669 | */ |
| 670 | int proc_watchdog(struct ctl_table *table, int write, |
| 671 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 672 | { |
| 673 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, |
| 674 | table, write, buffer, lenp, ppos); |
| 675 | } |
| 676 | |
| 677 | /* |
| 678 | * /proc/sys/kernel/nmi_watchdog |
| 679 | */ |
| 680 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
| 681 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 682 | { |
| 683 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, |
| 684 | table, write, buffer, lenp, ppos); |
| 685 | } |
| 686 | |
| 687 | /* |
| 688 | * /proc/sys/kernel/soft_watchdog |
| 689 | */ |
| 690 | int proc_soft_watchdog(struct ctl_table *table, int write, |
| 691 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 692 | { |
| 693 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, |
| 694 | table, write, buffer, lenp, ppos); |
| 695 | } |
| 696 | |
| 697 | /* |
| 698 | * /proc/sys/kernel/watchdog_thresh |
| 699 | */ |
| 700 | int proc_watchdog_thresh(struct ctl_table *table, int write, |
| 701 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 702 | { |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 703 | int err, old; |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 704 | |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 705 | cpu_hotplug_disable(); |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 706 | mutex_lock(&watchdog_mutex); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 707 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 708 | old = READ_ONCE(watchdog_thresh); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 709 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 710 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 711 | if (!err && write && old != READ_ONCE(watchdog_thresh)) |
| 712 | proc_watchdog_update(); |
Mandeep Singh Baines | e04ab2b | 2011-05-22 22:10:21 -0700 | [diff] [blame] | 713 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 714 | mutex_unlock(&watchdog_mutex); |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 715 | cpu_hotplug_enable(); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 716 | return err; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 717 | } |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 718 | |
| 719 | /* |
| 720 | * The cpumask is the mask of possible cpus that the watchdog can run |
| 721 | * on, not the mask of cpus it is actually running on. This allows the |
| 722 | * user to specify a mask that will include cpus that have not yet |
| 723 | * been brought online, if desired. |
| 724 | */ |
| 725 | int proc_watchdog_cpumask(struct ctl_table *table, int write, |
| 726 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 727 | { |
| 728 | int err; |
| 729 | |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 730 | cpu_hotplug_disable(); |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 731 | mutex_lock(&watchdog_mutex); |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 732 | |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 733 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
Thomas Gleixner | 05ba3de | 2017-09-12 21:37:08 +0200 | [diff] [blame] | 734 | if (!err && write) |
Thomas Gleixner | e8b62b2 | 2017-09-12 21:37:12 +0200 | [diff] [blame] | 735 | proc_watchdog_update(); |
Thomas Gleixner | 5490125 | 2017-09-12 21:36:59 +0200 | [diff] [blame] | 736 | |
Thomas Gleixner | 946d197 | 2017-09-12 21:37:01 +0200 | [diff] [blame] | 737 | mutex_unlock(&watchdog_mutex); |
Thomas Gleixner | b7a3498 | 2017-09-12 21:37:00 +0200 | [diff] [blame] | 738 | cpu_hotplug_enable(); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 739 | return err; |
| 740 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 741 | #endif /* CONFIG_SYSCTL */ |
| 742 | |
Peter Zijlstra | 004417a | 2010-11-25 18:38:29 +0100 | [diff] [blame] | 743 | void __init lockup_detector_init(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 744 | { |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 745 | #ifdef CONFIG_NO_HZ_FULL |
| 746 | if (tick_nohz_full_enabled()) { |
Frederic Weisbecker | 314b08ff | 2015-09-04 15:45:09 -0700 | [diff] [blame] | 747 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
| 748 | cpumask_copy(&watchdog_cpumask, housekeeping_mask); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 749 | } else |
| 750 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 751 | #else |
| 752 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 753 | #endif |
| 754 | |
Thomas Gleixner | d57108d | 2017-09-12 21:37:11 +0200 | [diff] [blame] | 755 | softlockup_init_threads(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 756 | } |