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 | |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 12 | #define pr_fmt(fmt) "NMI watchdog: " fmt |
| 13 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 14 | #include <linux/mm.h> |
| 15 | #include <linux/cpu.h> |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 16 | #include <linux/device.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 17 | #include <linux/nmi.h> |
| 18 | #include <linux/init.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 19 | #include <linux/module.h> |
| 20 | #include <linux/sysctl.h> |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 21 | #include <linux/smpboot.h> |
Clark Williams | 8bd75c7 | 2013-02-07 09:47:07 -0600 | [diff] [blame] | 22 | #include <linux/sched/rt.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> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 25 | |
| 26 | #include <asm/irq_regs.h> |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 27 | #include <linux/kvm_para.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 28 | #include <linux/perf_event.h> |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 29 | #include <linux/kthread.h> |
Prasad Sodagudi | 309fb62 | 2017-03-01 15:03:05 -0800 | [diff] [blame^] | 30 | #include <soc/qcom/watchdog.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 31 | |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 32 | /* |
| 33 | * The run state of the lockup detectors is controlled by the content of the |
| 34 | * 'watchdog_enabled' variable. Each lockup detector has its dedicated bit - |
| 35 | * bit 0 for the hard lockup detector and bit 1 for the soft lockup detector. |
| 36 | * |
| 37 | * 'watchdog_user_enabled', 'nmi_watchdog_enabled' and 'soft_watchdog_enabled' |
| 38 | * are variables that are only used as an 'interface' between the parameters |
| 39 | * in /proc/sys/kernel and the internal state bits in 'watchdog_enabled'. The |
| 40 | * 'watchdog_thresh' variable is handled differently because its value is not |
| 41 | * boolean, and the lockup detectors are 'suspended' while 'watchdog_thresh' |
| 42 | * is equal zero. |
| 43 | */ |
| 44 | #define NMI_WATCHDOG_ENABLED_BIT 0 |
| 45 | #define SOFT_WATCHDOG_ENABLED_BIT 1 |
| 46 | #define NMI_WATCHDOG_ENABLED (1 << NMI_WATCHDOG_ENABLED_BIT) |
| 47 | #define SOFT_WATCHDOG_ENABLED (1 << SOFT_WATCHDOG_ENABLED_BIT) |
| 48 | |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 49 | static DEFINE_MUTEX(watchdog_proc_mutex); |
| 50 | |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 51 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
| 52 | static unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED|NMI_WATCHDOG_ENABLED; |
| 53 | #else |
| 54 | static unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED; |
| 55 | #endif |
| 56 | int __read_mostly nmi_watchdog_enabled; |
| 57 | int __read_mostly soft_watchdog_enabled; |
| 58 | int __read_mostly watchdog_user_enabled; |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 59 | int __read_mostly watchdog_thresh = 10; |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 60 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 61 | #ifdef CONFIG_SMP |
| 62 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; |
Jiri Kosina | 5553787 | 2015-11-05 18:44:41 -0800 | [diff] [blame] | 63 | int __read_mostly sysctl_hardlockup_all_cpu_backtrace; |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 64 | #else |
| 65 | #define sysctl_softlockup_all_cpu_backtrace 0 |
Jiri Kosina | 5553787 | 2015-11-05 18:44:41 -0800 | [diff] [blame] | 66 | #define sysctl_hardlockup_all_cpu_backtrace 0 |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 67 | #endif |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 68 | static struct cpumask watchdog_cpumask __read_mostly; |
| 69 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); |
| 70 | |
| 71 | /* Helper for online, unparked cpus. */ |
| 72 | #define for_each_watchdog_cpu(cpu) \ |
| 73 | for_each_cpu_and((cpu), cpu_online_mask, &watchdog_cpumask) |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 74 | |
Ulrich Obergfell | ec6a906 | 2015-09-04 15:45:28 -0700 | [diff] [blame] | 75 | /* |
| 76 | * The 'watchdog_running' variable is set to 1 when the watchdog threads |
| 77 | * are registered/started and is set to 0 when the watchdog threads are |
| 78 | * unregistered/stopped, so it is an indicator whether the threads exist. |
| 79 | */ |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 80 | static int __read_mostly watchdog_running; |
Ulrich Obergfell | ec6a906 | 2015-09-04 15:45:28 -0700 | [diff] [blame] | 81 | /* |
| 82 | * If a subsystem has a need to deactivate the watchdog temporarily, it |
| 83 | * can use the suspend/resume interface to achieve this. The content of |
| 84 | * the 'watchdog_suspended' variable reflects this state. Existing threads |
| 85 | * are parked/unparked by the lockup_detector_{suspend|resume} functions |
| 86 | * (see comment blocks pertaining to those functions for further details). |
| 87 | * |
| 88 | * 'watchdog_suspended' also prevents threads from being registered/started |
| 89 | * or unregistered/stopped via parameters in /proc/sys/kernel, so the state |
| 90 | * of 'watchdog_running' cannot change while the watchdog is deactivated |
| 91 | * temporarily (see related code in 'proc' handlers). |
| 92 | */ |
| 93 | static int __read_mostly watchdog_suspended; |
| 94 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 95 | static u64 __read_mostly sample_period; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 96 | |
| 97 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); |
| 98 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); |
| 99 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 100 | static DEFINE_PER_CPU(unsigned int, watchdog_en); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 101 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 102 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 103 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
| 104 | static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 105 | static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved); |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 106 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | cafcd80 | 2010-05-14 11:11:21 -0400 | [diff] [blame] | 107 | static DEFINE_PER_CPU(bool, hard_watchdog_warn); |
| 108 | static DEFINE_PER_CPU(bool, watchdog_nmi_touch); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 109 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 110 | #endif |
| 111 | #ifdef CONFIG_HARDLOCKUP_DETECTOR_OTHER_CPU |
| 112 | static cpumask_t __read_mostly watchdog_cpus; |
| 113 | #endif |
| 114 | #ifdef CONFIG_HARDLOCKUP_DETECTOR_NMI |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 115 | static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); |
| 116 | #endif |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 117 | static unsigned long soft_lockup_nmi_warn; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 118 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 119 | /* boot commands */ |
| 120 | /* |
| 121 | * Should we panic when a soft-lockup or hard-lockup occurs: |
| 122 | */ |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 123 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | ac1f591 | 2015-11-05 18:44:44 -0800 | [diff] [blame] | 124 | unsigned int __read_mostly hardlockup_panic = |
Don Zickus | fef2c9b | 2011-03-22 16:34:16 -0700 | [diff] [blame] | 125 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; |
Brian Norris | de361bf | 2016-02-29 17:42:29 -0800 | [diff] [blame] | 126 | static unsigned long __maybe_unused hardlockup_allcpu_dumped; |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 127 | /* |
| 128 | * We may not want to enable hard lockup detection by default in all cases, |
| 129 | * for example when running the kernel as a guest on a hypervisor. In these |
| 130 | * cases this function can be called to disable hard lockup detection. This |
| 131 | * function should only be executed once by the boot processor before the |
| 132 | * kernel command line parameters are parsed, because otherwise it is not |
| 133 | * possible to override this in hardlockup_panic_setup(). |
| 134 | */ |
Ulrich Obergfell | 692297d | 2015-04-14 15:44:19 -0700 | [diff] [blame] | 135 | void hardlockup_detector_disable(void) |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 136 | { |
Ulrich Obergfell | 692297d | 2015-04-14 15:44:19 -0700 | [diff] [blame] | 137 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 138 | } |
| 139 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 140 | static int __init hardlockup_panic_setup(char *str) |
| 141 | { |
| 142 | if (!strncmp(str, "panic", 5)) |
| 143 | hardlockup_panic = 1; |
Don Zickus | fef2c9b | 2011-03-22 16:34:16 -0700 | [diff] [blame] | 144 | else if (!strncmp(str, "nopanic", 7)) |
| 145 | hardlockup_panic = 0; |
Don Zickus | 5dc3055 | 2010-11-29 17:07:17 -0500 | [diff] [blame] | 146 | else if (!strncmp(str, "0", 1)) |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 147 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
| 148 | else if (!strncmp(str, "1", 1)) |
| 149 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 150 | return 1; |
| 151 | } |
| 152 | __setup("nmi_watchdog=", hardlockup_panic_setup); |
| 153 | #endif |
| 154 | |
| 155 | unsigned int __read_mostly softlockup_panic = |
| 156 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; |
| 157 | |
| 158 | static int __init softlockup_panic_setup(char *str) |
| 159 | { |
| 160 | softlockup_panic = simple_strtoul(str, NULL, 0); |
| 161 | |
| 162 | return 1; |
| 163 | } |
| 164 | __setup("softlockup_panic=", softlockup_panic_setup); |
| 165 | |
| 166 | static int __init nowatchdog_setup(char *str) |
| 167 | { |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 168 | watchdog_enabled = 0; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 169 | return 1; |
| 170 | } |
| 171 | __setup("nowatchdog", nowatchdog_setup); |
| 172 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 173 | static int __init nosoftlockup_setup(char *str) |
| 174 | { |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 175 | watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 176 | return 1; |
| 177 | } |
| 178 | __setup("nosoftlockup", nosoftlockup_setup); |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 179 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 180 | #ifdef CONFIG_SMP |
| 181 | static int __init softlockup_all_cpu_backtrace_setup(char *str) |
| 182 | { |
| 183 | sysctl_softlockup_all_cpu_backtrace = |
| 184 | !!simple_strtol(str, NULL, 0); |
| 185 | return 1; |
| 186 | } |
| 187 | __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); |
Jiri Kosina | 5553787 | 2015-11-05 18:44:41 -0800 | [diff] [blame] | 188 | static int __init hardlockup_all_cpu_backtrace_setup(char *str) |
| 189 | { |
| 190 | sysctl_hardlockup_all_cpu_backtrace = |
| 191 | !!simple_strtol(str, NULL, 0); |
| 192 | return 1; |
| 193 | } |
| 194 | __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup); |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 195 | #endif |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 196 | |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 197 | /* |
| 198 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- |
| 199 | * lockups can have false positives under extreme conditions. So we generally |
| 200 | * want a higher threshold for soft lockups than for hard lockups. So we couple |
| 201 | * the thresholds with a factor: we make the soft threshold twice the amount of |
| 202 | * time the hard threshold is. |
| 203 | */ |
Ingo Molnar | 6e9101a | 2011-05-24 05:43:18 +0200 | [diff] [blame] | 204 | static int get_softlockup_thresh(void) |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 205 | { |
| 206 | return watchdog_thresh * 2; |
| 207 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 208 | |
| 209 | /* |
| 210 | * Returns seconds, approximately. We don't need nanosecond |
| 211 | * resolution, and we don't need to waste time with a big divide when |
| 212 | * 2^30ns == 1.074s. |
| 213 | */ |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 214 | static unsigned long get_timestamp(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 215 | { |
Cyril Bur | 545a2bf | 2015-02-12 15:01:24 -0800 | [diff] [blame] | 216 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 217 | } |
| 218 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 219 | static void set_sample_period(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 220 | { |
| 221 | /* |
Mandeep Singh Baines | 586692a | 2011-05-22 22:10:22 -0700 | [diff] [blame] | 222 | * convert watchdog_thresh from seconds to ns |
Fernando Luis Vázquez Cao | 86f5e6a | 2012-02-09 17:42:22 -0500 | [diff] [blame] | 223 | * the divide by 5 is to give hrtimer several chances (two |
| 224 | * or three with the current relation between the soft |
| 225 | * and hard thresholds) to increment before the |
| 226 | * hardlockup detector generates a warning |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 227 | */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 228 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 229 | } |
| 230 | |
| 231 | /* Commands for resetting the watchdog */ |
| 232 | static void __touch_watchdog(void) |
| 233 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 234 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 235 | } |
| 236 | |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 237 | /** |
| 238 | * touch_softlockup_watchdog_sched - touch watchdog on scheduler stalls |
| 239 | * |
| 240 | * Call when the scheduler may have stalled for legitimate reasons |
| 241 | * preventing the watchdog task from executing - e.g. the scheduler |
| 242 | * entering idle state. This should only be used for scheduler events. |
| 243 | * Use touch_softlockup_watchdog() for everything else. |
| 244 | */ |
| 245 | void touch_softlockup_watchdog_sched(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 246 | { |
Andrew Morton | 7861144 | 2014-04-18 15:07:12 -0700 | [diff] [blame] | 247 | /* |
| 248 | * Preemption can be enabled. It doesn't matter which CPU's timestamp |
| 249 | * gets zeroed here, so use the raw_ operation. |
| 250 | */ |
| 251 | raw_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 252 | } |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 253 | |
| 254 | void touch_softlockup_watchdog(void) |
| 255 | { |
| 256 | touch_softlockup_watchdog_sched(); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 257 | wq_watchdog_touch(raw_smp_processor_id()); |
Tejun Heo | 03e0d46 | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 258 | } |
Ingo Molnar | 0167c78 | 2010-05-13 08:53:33 +0200 | [diff] [blame] | 259 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 260 | |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 261 | void touch_all_softlockup_watchdogs(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 262 | { |
| 263 | int cpu; |
| 264 | |
| 265 | /* |
| 266 | * this is done lockless |
| 267 | * do we care if a 0 races with a timestamp? |
| 268 | * all it means is the softlock check starts one cycle later |
| 269 | */ |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 270 | for_each_watchdog_cpu(cpu) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 271 | per_cpu(watchdog_touch_ts, cpu) = 0; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 272 | wq_watchdog_touch(-1); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 273 | } |
| 274 | |
Don Zickus | cafcd80 | 2010-05-14 11:11:21 -0400 | [diff] [blame] | 275 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 276 | void touch_nmi_watchdog(void) |
| 277 | { |
Ben Zhang | 62572e2 | 2014-04-03 14:47:18 -0700 | [diff] [blame] | 278 | /* |
| 279 | * Using __raw here because some code paths have |
| 280 | * preemption enabled. If preemption is enabled |
| 281 | * then interrupts should be enabled too, in which |
| 282 | * case we shouldn't have to worry about the watchdog |
| 283 | * going off. |
| 284 | */ |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 285 | raw_cpu_write(watchdog_nmi_touch, true); |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 286 | touch_softlockup_watchdog(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 287 | } |
| 288 | EXPORT_SYMBOL(touch_nmi_watchdog); |
| 289 | |
Don Zickus | cafcd80 | 2010-05-14 11:11:21 -0400 | [diff] [blame] | 290 | #endif |
| 291 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 292 | void touch_softlockup_watchdog_sync(void) |
| 293 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 294 | __this_cpu_write(softlockup_touch_sync, true); |
| 295 | __this_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 296 | } |
| 297 | |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 298 | #ifdef CONFIG_HARDLOCKUP_DETECTOR_NMI |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 299 | /* watchdog detector functions */ |
Yaowei Bai | 451637e | 2015-11-05 18:44:24 -0800 | [diff] [blame] | 300 | static bool is_hardlockup(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 301 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 302 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 303 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 304 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
Yaowei Bai | 451637e | 2015-11-05 18:44:24 -0800 | [diff] [blame] | 305 | return true; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 306 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 307 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
Yaowei Bai | 451637e | 2015-11-05 18:44:24 -0800 | [diff] [blame] | 308 | return false; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 309 | } |
| 310 | #endif |
| 311 | |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 312 | #ifdef CONFIG_HARDLOCKUP_DETECTOR_OTHER_CPU |
| 313 | static unsigned int watchdog_next_cpu(unsigned int cpu) |
| 314 | { |
| 315 | cpumask_t cpus = watchdog_cpus; |
| 316 | unsigned int next_cpu; |
| 317 | |
| 318 | next_cpu = cpumask_next(cpu, &cpus); |
| 319 | if (next_cpu >= nr_cpu_ids) |
| 320 | next_cpu = cpumask_first(&cpus); |
| 321 | |
| 322 | if (next_cpu == cpu) |
| 323 | return nr_cpu_ids; |
| 324 | |
| 325 | return next_cpu; |
| 326 | } |
| 327 | |
| 328 | static int is_hardlockup_other_cpu(unsigned int cpu) |
| 329 | { |
| 330 | unsigned long hrint = per_cpu(hrtimer_interrupts, cpu); |
| 331 | |
| 332 | if (per_cpu(hrtimer_interrupts_saved, cpu) == hrint) |
| 333 | return 1; |
| 334 | |
| 335 | per_cpu(hrtimer_interrupts_saved, cpu) = hrint; |
| 336 | return 0; |
| 337 | } |
| 338 | |
| 339 | static void watchdog_check_hardlockup_other_cpu(void) |
| 340 | { |
| 341 | unsigned int next_cpu; |
| 342 | |
| 343 | /* |
| 344 | * Test for hardlockups every 3 samples. The sample period is |
| 345 | * watchdog_thresh * 2 / 5, so 3 samples gets us back to slightly over |
| 346 | * watchdog_thresh (over by 20%). |
| 347 | */ |
| 348 | if (__this_cpu_read(hrtimer_interrupts) % 3 != 0) |
| 349 | return; |
| 350 | |
| 351 | /* check for a hardlockup on the next cpu */ |
| 352 | next_cpu = watchdog_next_cpu(smp_processor_id()); |
| 353 | if (next_cpu >= nr_cpu_ids) |
| 354 | return; |
| 355 | |
| 356 | smp_rmb(); |
| 357 | |
| 358 | if (per_cpu(watchdog_nmi_touch, next_cpu) == true) { |
| 359 | per_cpu(watchdog_nmi_touch, next_cpu) = false; |
| 360 | return; |
| 361 | } |
| 362 | |
| 363 | if (is_hardlockup_other_cpu(next_cpu)) { |
| 364 | /* only warn once */ |
| 365 | if (per_cpu(hard_watchdog_warn, next_cpu) == true) |
| 366 | return; |
| 367 | |
Prasad Sodagudi | 309fb62 | 2017-03-01 15:03:05 -0800 | [diff] [blame^] | 368 | if (hardlockup_panic) { |
| 369 | pr_err("Watchdog detected hard LOCKUP on cpu %u", |
| 370 | next_cpu); |
| 371 | msm_trigger_wdog_bite(); |
| 372 | } |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 373 | else |
| 374 | WARN(1, "Watchdog detected hard LOCKUP on cpu %u", next_cpu); |
| 375 | |
| 376 | per_cpu(hard_watchdog_warn, next_cpu) = true; |
| 377 | } else { |
| 378 | per_cpu(hard_watchdog_warn, next_cpu) = false; |
| 379 | } |
| 380 | } |
| 381 | #else |
| 382 | static inline void watchdog_check_hardlockup_other_cpu(void) { return; } |
| 383 | #endif |
| 384 | |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 385 | static int is_softlockup(unsigned long touch_ts) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 386 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 387 | unsigned long now = get_timestamp(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 388 | |
Ulrich Obergfell | 39d2da2 | 2015-11-05 18:44:56 -0800 | [diff] [blame] | 389 | if ((watchdog_enabled & SOFT_WATCHDOG_ENABLED) && watchdog_thresh){ |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 390 | /* Warn about unreasonable delays. */ |
| 391 | if (time_after(now, touch_ts + get_softlockup_thresh())) |
| 392 | return now - touch_ts; |
| 393 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 394 | return 0; |
| 395 | } |
| 396 | |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 397 | #ifdef CONFIG_HARDLOCKUP_DETECTOR_NMI |
Cyrill Gorcunov | 1880c4a | 2011-06-23 16:49:18 +0400 | [diff] [blame] | 398 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 399 | static struct perf_event_attr wd_hw_attr = { |
| 400 | .type = PERF_TYPE_HARDWARE, |
| 401 | .config = PERF_COUNT_HW_CPU_CYCLES, |
| 402 | .size = sizeof(struct perf_event_attr), |
| 403 | .pinned = 1, |
| 404 | .disabled = 1, |
| 405 | }; |
| 406 | |
| 407 | /* Callback function for perf event subsystem */ |
Peter Zijlstra | a8b0ca1 | 2011-06-27 14:41:57 +0200 | [diff] [blame] | 408 | static void watchdog_overflow_callback(struct perf_event *event, |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 409 | struct perf_sample_data *data, |
| 410 | struct pt_regs *regs) |
| 411 | { |
Peter Zijlstra | c6db67c | 2010-08-20 11:49:15 +0200 | [diff] [blame] | 412 | /* Ensure the watchdog never gets throttled */ |
| 413 | event->hw.interrupts = 0; |
| 414 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 415 | if (__this_cpu_read(watchdog_nmi_touch) == true) { |
| 416 | __this_cpu_write(watchdog_nmi_touch, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 417 | return; |
| 418 | } |
| 419 | |
| 420 | /* check for a hardlockup |
| 421 | * This is done by making sure our timer interrupt |
| 422 | * is incrementing. The timer interrupt should have |
| 423 | * fired multiple times before we overflow'd. If it hasn't |
| 424 | * then this is a good indication the cpu is stuck |
| 425 | */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 426 | if (is_hardlockup()) { |
| 427 | int this_cpu = smp_processor_id(); |
| 428 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 429 | /* only print hardlockups once */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 430 | if (__this_cpu_read(hard_watchdog_warn) == true) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 431 | return; |
| 432 | |
Jiri Kosina | 5553787 | 2015-11-05 18:44:41 -0800 | [diff] [blame] | 433 | pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu); |
Prasad Sodagudi | 309fb62 | 2017-03-01 15:03:05 -0800 | [diff] [blame^] | 434 | if (hardlockup_panic) |
| 435 | msm_trigger_wdog_bite(); |
| 436 | |
Jiri Kosina | 5553787 | 2015-11-05 18:44:41 -0800 | [diff] [blame] | 437 | print_modules(); |
| 438 | print_irqtrace_events(current); |
| 439 | if (regs) |
| 440 | show_regs(regs); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 441 | else |
Jiri Kosina | 5553787 | 2015-11-05 18:44:41 -0800 | [diff] [blame] | 442 | dump_stack(); |
| 443 | |
| 444 | /* |
| 445 | * Perform all-CPU dump only once to avoid multiple hardlockups |
| 446 | * generating interleaving traces |
| 447 | */ |
| 448 | if (sysctl_hardlockup_all_cpu_backtrace && |
| 449 | !test_and_set_bit(0, &hardlockup_allcpu_dumped)) |
| 450 | trigger_allbutself_cpu_backtrace(); |
| 451 | |
| 452 | if (hardlockup_panic) |
Hidehiro Kawai | 58c5661 | 2015-12-14 11:19:10 +0100 | [diff] [blame] | 453 | nmi_panic(regs, "Hard LOCKUP"); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 454 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 455 | __this_cpu_write(hard_watchdog_warn, true); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 456 | return; |
| 457 | } |
| 458 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 459 | __this_cpu_write(hard_watchdog_warn, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 460 | return; |
| 461 | } |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 462 | #endif /* CONFIG_HARDLOCKUP_DETECTOR_NMI */ |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 463 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 464 | static void watchdog_interrupt_count(void) |
| 465 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 466 | __this_cpu_inc(hrtimer_interrupts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 467 | } |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 468 | |
| 469 | static int watchdog_nmi_enable(unsigned int cpu); |
| 470 | static void watchdog_nmi_disable(unsigned int cpu); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 471 | |
Ulrich Obergfell | 58cf690 | 2015-11-05 18:44:30 -0800 | [diff] [blame] | 472 | static int watchdog_enable_all_cpus(void); |
| 473 | static void watchdog_disable_all_cpus(void); |
| 474 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 475 | /* watchdog kicker functions */ |
| 476 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) |
| 477 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 478 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 479 | struct pt_regs *regs = get_irq_regs(); |
| 480 | int duration; |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 481 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 482 | |
| 483 | /* kick the hardlockup detector */ |
| 484 | watchdog_interrupt_count(); |
| 485 | |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 486 | /* test for hardlockups on the next cpu */ |
| 487 | watchdog_check_hardlockup_other_cpu(); |
| 488 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 489 | /* kick the softlockup detector */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 490 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 491 | |
| 492 | /* .. and repeat */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 493 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 494 | |
| 495 | if (touch_ts == 0) { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 496 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 497 | /* |
| 498 | * If the time stamp was touched atomically |
| 499 | * make sure the scheduler tick is up to date. |
| 500 | */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 501 | __this_cpu_write(softlockup_touch_sync, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 502 | sched_clock_tick(); |
| 503 | } |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 504 | |
| 505 | /* Clear the guest paused flag on watchdog reset */ |
| 506 | kvm_check_and_clear_guest_paused(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 507 | __touch_watchdog(); |
| 508 | return HRTIMER_RESTART; |
| 509 | } |
| 510 | |
| 511 | /* check for a softlockup |
| 512 | * This is done by making sure a high priority task is |
| 513 | * being scheduled. The task touches the watchdog to |
| 514 | * indicate it is getting cpu time. If it hasn't then |
| 515 | * this is a good indication some task is hogging the cpu |
| 516 | */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 517 | duration = is_softlockup(touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 518 | if (unlikely(duration)) { |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 519 | /* |
| 520 | * If a virtual machine is stopped by the host it can look to |
| 521 | * the watchdog like a soft lockup, check to see if the host |
| 522 | * stopped the vm before we issue the warning |
| 523 | */ |
| 524 | if (kvm_check_and_clear_guest_paused()) |
| 525 | return HRTIMER_RESTART; |
| 526 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 527 | /* only warn once */ |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 528 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
| 529 | /* |
| 530 | * When multiple processes are causing softlockups the |
| 531 | * softlockup detector only warns on the first one |
| 532 | * because the code relies on a full quiet cycle to |
| 533 | * re-arm. The second process prevents the quiet cycle |
| 534 | * and never gets reported. Use task pointers to detect |
| 535 | * this. |
| 536 | */ |
| 537 | if (__this_cpu_read(softlockup_task_ptr_saved) != |
| 538 | current) { |
| 539 | __this_cpu_write(soft_watchdog_warn, false); |
| 540 | __touch_watchdog(); |
| 541 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 542 | return HRTIMER_RESTART; |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 543 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 544 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 545 | if (softlockup_all_cpu_backtrace) { |
| 546 | /* Prevent multiple soft-lockup reports if one cpu is already |
| 547 | * engaged in dumping cpu back traces |
| 548 | */ |
| 549 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { |
| 550 | /* Someone else will report us. Let's give up */ |
| 551 | __this_cpu_write(soft_watchdog_warn, true); |
| 552 | return HRTIMER_RESTART; |
| 553 | } |
| 554 | } |
| 555 | |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 556 | 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] | 557 | smp_processor_id(), duration, |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 558 | current->comm, task_pid_nr(current)); |
Prasad Sodagudi | 309fb62 | 2017-03-01 15:03:05 -0800 | [diff] [blame^] | 559 | |
| 560 | if (softlockup_panic) |
| 561 | msm_trigger_wdog_bite(); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 562 | __this_cpu_write(softlockup_task_ptr_saved, current); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 563 | print_modules(); |
| 564 | print_irqtrace_events(current); |
| 565 | if (regs) |
| 566 | show_regs(regs); |
| 567 | else |
| 568 | dump_stack(); |
| 569 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 570 | if (softlockup_all_cpu_backtrace) { |
| 571 | /* Avoid generating two back traces for current |
| 572 | * given that one is already made above |
| 573 | */ |
| 574 | trigger_allbutself_cpu_backtrace(); |
| 575 | |
| 576 | clear_bit(0, &soft_lockup_nmi_warn); |
| 577 | /* Barrier to sync with other cpus */ |
| 578 | smp_mb__after_atomic(); |
| 579 | } |
| 580 | |
Josh Hunt | 69361ee | 2014-08-08 14:22:31 -0700 | [diff] [blame] | 581 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 582 | if (softlockup_panic) |
| 583 | panic("softlockup: hung tasks"); |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 584 | __this_cpu_write(soft_watchdog_warn, true); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 585 | } else |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 586 | __this_cpu_write(soft_watchdog_warn, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 587 | |
| 588 | return HRTIMER_RESTART; |
| 589 | } |
| 590 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 591 | static void watchdog_set_prio(unsigned int policy, unsigned int prio) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 592 | { |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 593 | struct sched_param param = { .sched_priority = prio }; |
| 594 | |
| 595 | sched_setscheduler(current, policy, ¶m); |
| 596 | } |
| 597 | |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 598 | void watchdog_enable(unsigned int cpu) |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 599 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 600 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 601 | unsigned int *enabled = raw_cpu_ptr(&watchdog_en); |
| 602 | |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 603 | if (*enabled) |
| 604 | return; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 605 | |
Bjørn Mork | 3935e895 | 2012-12-19 20:51:31 +0100 | [diff] [blame] | 606 | /* kick off the timer for the hardlockup detector */ |
| 607 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
| 608 | hrtimer->function = watchdog_timer_fn; |
| 609 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 610 | /* Enable the perf event */ |
| 611 | watchdog_nmi_enable(cpu); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 612 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 613 | /* done here because hrtimer_start can only pin to smp_processor_id() */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 614 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 615 | HRTIMER_MODE_REL_PINNED); |
| 616 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 617 | /* initialize timestamp */ |
| 618 | watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1); |
| 619 | __touch_watchdog(); |
Olav Haugan | 1111347 | 2016-11-03 15:10:57 -0700 | [diff] [blame] | 620 | |
| 621 | /* |
| 622 | * Need to ensure above operations are observed by other CPUs before |
| 623 | * indicating that timer is enabled. This is to synchronize core |
| 624 | * isolation and hotplug. Core isolation will wait for this flag to be |
| 625 | * set. |
| 626 | */ |
| 627 | mb(); |
| 628 | *enabled = 1; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 629 | } |
| 630 | |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 631 | void watchdog_disable(unsigned int cpu) |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 632 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 633 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 634 | unsigned int *enabled = raw_cpu_ptr(&watchdog_en); |
| 635 | |
Olav Haugan | d8354e6 | 2016-08-18 16:49:44 -0700 | [diff] [blame] | 636 | if (!*enabled) |
| 637 | return; |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 638 | |
| 639 | watchdog_set_prio(SCHED_NORMAL, 0); |
| 640 | hrtimer_cancel(hrtimer); |
| 641 | /* disable the perf event */ |
| 642 | watchdog_nmi_disable(cpu); |
Olav Haugan | 1111347 | 2016-11-03 15:10:57 -0700 | [diff] [blame] | 643 | |
| 644 | /* |
| 645 | * No need for barrier here since disabling the watchdog is |
| 646 | * synchronized with hotplug lock |
| 647 | */ |
| 648 | *enabled = 0; |
| 649 | } |
| 650 | |
| 651 | bool watchdog_configured(unsigned int cpu) |
| 652 | { |
| 653 | return *per_cpu_ptr(&watchdog_en, cpu); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 654 | } |
| 655 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 656 | static void watchdog_cleanup(unsigned int cpu, bool online) |
| 657 | { |
| 658 | watchdog_disable(cpu); |
| 659 | } |
| 660 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 661 | static int watchdog_should_run(unsigned int cpu) |
| 662 | { |
| 663 | return __this_cpu_read(hrtimer_interrupts) != |
| 664 | __this_cpu_read(soft_lockup_hrtimer_cnt); |
| 665 | } |
| 666 | |
| 667 | /* |
| 668 | * The watchdog thread function - touches the timestamp. |
| 669 | * |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 670 | * It only runs once every sample_period seconds (4 seconds by |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 671 | * default) to reset the softlockup timestamp. If this gets delayed |
| 672 | * for more than 2*watchdog_thresh seconds then the debug-printout |
| 673 | * triggers in watchdog_timer_fn(). |
| 674 | */ |
| 675 | static void watchdog(unsigned int cpu) |
| 676 | { |
| 677 | __this_cpu_write(soft_lockup_hrtimer_cnt, |
| 678 | __this_cpu_read(hrtimer_interrupts)); |
| 679 | __touch_watchdog(); |
Ulrich Obergfell | bcfba4f | 2015-04-14 15:44:10 -0700 | [diff] [blame] | 680 | |
| 681 | /* |
| 682 | * watchdog_nmi_enable() clears the NMI_WATCHDOG_ENABLED bit in the |
| 683 | * failure path. Check for failures that can occur asynchronously - |
| 684 | * for example, when CPUs are on-lined - and shut down the hardware |
| 685 | * perf event on each CPU accordingly. |
| 686 | * |
| 687 | * The only non-obvious place this bit can be cleared is through |
| 688 | * watchdog_nmi_enable(), so a pr_info() is placed there. Placing a |
| 689 | * pr_info here would be too noisy as it would result in a message |
| 690 | * every few seconds if the hardlockup was disabled but the softlockup |
| 691 | * enabled. |
| 692 | */ |
| 693 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) |
| 694 | watchdog_nmi_disable(cpu); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 695 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 696 | |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 697 | #ifdef CONFIG_HARDLOCKUP_DETECTOR_NMI |
Don Zickus | a702704 | 2012-06-13 09:35:48 -0400 | [diff] [blame] | 698 | /* |
| 699 | * People like the simple clean cpu node info on boot. |
| 700 | * Reduce the watchdog noise by only printing messages |
| 701 | * that are different from what cpu0 displayed. |
| 702 | */ |
| 703 | static unsigned long cpu0_err; |
| 704 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 705 | static int watchdog_nmi_enable(unsigned int cpu) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 706 | { |
| 707 | struct perf_event_attr *wd_attr; |
| 708 | struct perf_event *event = per_cpu(watchdog_ev, cpu); |
| 709 | |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 710 | /* nothing to do if the hard lockup detector is disabled */ |
| 711 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) |
| 712 | goto out; |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 713 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 714 | /* is it already setup and enabled? */ |
| 715 | if (event && event->state > PERF_EVENT_STATE_OFF) |
| 716 | goto out; |
| 717 | |
| 718 | /* it is setup but not enabled */ |
| 719 | if (event != NULL) |
| 720 | goto out_enable; |
| 721 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 722 | wd_attr = &wd_hw_attr; |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 723 | wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh); |
Cyrill Gorcunov | 1880c4a | 2011-06-23 16:49:18 +0400 | [diff] [blame] | 724 | |
| 725 | /* Try to register using hardware perf events */ |
Avi Kivity | 4dc0da8 | 2011-06-29 18:42:35 +0300 | [diff] [blame] | 726 | event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL); |
Don Zickus | a702704 | 2012-06-13 09:35:48 -0400 | [diff] [blame] | 727 | |
| 728 | /* save cpu0 error for future comparision */ |
| 729 | if (cpu == 0 && IS_ERR(event)) |
| 730 | cpu0_err = PTR_ERR(event); |
| 731 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 732 | if (!IS_ERR(event)) { |
Don Zickus | a702704 | 2012-06-13 09:35:48 -0400 | [diff] [blame] | 733 | /* only print for cpu0 or different than cpu0 */ |
| 734 | if (cpu == 0 || cpu0_err) |
| 735 | pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n"); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 736 | goto out_save; |
| 737 | } |
| 738 | |
Ulrich Obergfell | bcfba4f | 2015-04-14 15:44:10 -0700 | [diff] [blame] | 739 | /* |
| 740 | * Disable the hard lockup detector if _any_ CPU fails to set up |
| 741 | * set up the hardware perf event. The watchdog() function checks |
| 742 | * the NMI_WATCHDOG_ENABLED bit periodically. |
| 743 | * |
| 744 | * The barriers are for syncing up watchdog_enabled across all the |
| 745 | * cpus, as clear_bit() does not use barriers. |
| 746 | */ |
| 747 | smp_mb__before_atomic(); |
| 748 | clear_bit(NMI_WATCHDOG_ENABLED_BIT, &watchdog_enabled); |
| 749 | smp_mb__after_atomic(); |
| 750 | |
Don Zickus | a702704 | 2012-06-13 09:35:48 -0400 | [diff] [blame] | 751 | /* skip displaying the same error again */ |
| 752 | if (cpu > 0 && (PTR_ERR(event) == cpu0_err)) |
| 753 | return PTR_ERR(event); |
Don Zickus | 5651f7f | 2011-02-09 14:02:33 -0500 | [diff] [blame] | 754 | |
| 755 | /* vary the KERN level based on the returned errno */ |
| 756 | if (PTR_ERR(event) == -EOPNOTSUPP) |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 757 | pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu); |
Don Zickus | 5651f7f | 2011-02-09 14:02:33 -0500 | [diff] [blame] | 758 | else if (PTR_ERR(event) == -ENOENT) |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 759 | pr_warn("disabled (cpu%i): hardware events not enabled\n", |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 760 | cpu); |
Don Zickus | 5651f7f | 2011-02-09 14:02:33 -0500 | [diff] [blame] | 761 | else |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 762 | pr_err("disabled (cpu%i): unable to create perf event: %ld\n", |
| 763 | cpu, PTR_ERR(event)); |
Ulrich Obergfell | bcfba4f | 2015-04-14 15:44:10 -0700 | [diff] [blame] | 764 | |
| 765 | pr_info("Shutting down hard lockup detector on all cpus\n"); |
| 766 | |
Akinobu Mita | eac2433 | 2010-08-31 23:00:08 -0400 | [diff] [blame] | 767 | return PTR_ERR(event); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 768 | |
| 769 | /* success path */ |
| 770 | out_save: |
| 771 | per_cpu(watchdog_ev, cpu) = event; |
| 772 | out_enable: |
| 773 | perf_event_enable(per_cpu(watchdog_ev, cpu)); |
| 774 | out: |
| 775 | return 0; |
| 776 | } |
| 777 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 778 | static void watchdog_nmi_disable(unsigned int cpu) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 779 | { |
| 780 | struct perf_event *event = per_cpu(watchdog_ev, cpu); |
| 781 | |
| 782 | if (event) { |
| 783 | perf_event_disable(event); |
| 784 | per_cpu(watchdog_ev, cpu) = NULL; |
| 785 | |
| 786 | /* should be in cleanup, but blocks oprofile */ |
| 787 | perf_event_release_kernel(event); |
| 788 | } |
Ulrich Obergfell | df57714 | 2014-08-11 10:49:25 -0400 | [diff] [blame] | 789 | if (cpu == 0) { |
| 790 | /* watchdog_nmi_enable() expects this to be zero initially. */ |
| 791 | cpu0_err = 0; |
| 792 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 793 | } |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 794 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 795 | #else |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 796 | #ifdef CONFIG_HARDLOCKUP_DETECTOR_OTHER_CPU |
| 797 | static int watchdog_nmi_enable(unsigned int cpu) |
| 798 | { |
| 799 | /* |
| 800 | * The new cpu will be marked online before the first hrtimer interrupt |
| 801 | * runs on it. If another cpu tests for a hardlockup on the new cpu |
| 802 | * before it has run its first hrtimer, it will get a false positive. |
| 803 | * Touch the watchdog on the new cpu to delay the first check for at |
| 804 | * least 3 sampling periods to guarantee one hrtimer has run on the new |
| 805 | * cpu. |
| 806 | */ |
| 807 | per_cpu(watchdog_nmi_touch, cpu) = true; |
| 808 | smp_wmb(); |
| 809 | cpumask_set_cpu(cpu, &watchdog_cpus); |
| 810 | return 0; |
| 811 | } |
| 812 | |
| 813 | static void watchdog_nmi_disable(unsigned int cpu) |
| 814 | { |
| 815 | unsigned int next_cpu = watchdog_next_cpu(cpu); |
| 816 | |
| 817 | /* |
| 818 | * Offlining this cpu will cause the cpu before this one to start |
| 819 | * checking the one after this one. If this cpu just finished checking |
| 820 | * the next cpu and updating hrtimer_interrupts_saved, and then the |
| 821 | * previous cpu checks it within one sample period, it will trigger a |
| 822 | * false positive. Touch the watchdog on the next cpu to prevent it. |
| 823 | */ |
| 824 | if (next_cpu < nr_cpu_ids) |
| 825 | per_cpu(watchdog_nmi_touch, next_cpu) = true; |
| 826 | smp_wmb(); |
| 827 | cpumask_clear_cpu(cpu, &watchdog_cpus); |
| 828 | } |
| 829 | #else |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 830 | static int watchdog_nmi_enable(unsigned int cpu) { return 0; } |
| 831 | static void watchdog_nmi_disable(unsigned int cpu) { return; } |
Colin Cross | b9180a4 | 2013-01-11 13:51:48 -0800 | [diff] [blame] | 832 | #endif /* CONFIG_HARDLOCKUP_DETECTOR_OTHER_CPU */ |
| 833 | #endif /* CONFIG_HARDLOCKUP_DETECTOR_NMI */ |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 834 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 835 | static struct smp_hotplug_thread watchdog_threads = { |
| 836 | .store = &softlockup_watchdog, |
| 837 | .thread_should_run = watchdog_should_run, |
| 838 | .thread_fn = watchdog, |
| 839 | .thread_comm = "watchdog/%u", |
| 840 | .setup = watchdog_enable, |
| 841 | .cleanup = watchdog_cleanup, |
| 842 | .park = watchdog_disable, |
| 843 | .unpark = watchdog_enable, |
| 844 | }; |
| 845 | |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 846 | /* |
| 847 | * park all watchdog threads that are specified in 'watchdog_cpumask' |
Ulrich Obergfell | ee7fed5 | 2015-11-05 18:44:39 -0800 | [diff] [blame] | 848 | * |
| 849 | * This function returns an error if kthread_park() of a watchdog thread |
| 850 | * fails. In this situation, the watchdog threads of some CPUs can already |
| 851 | * be parked and the watchdog threads of other CPUs can still be runnable. |
| 852 | * Callers are expected to handle this special condition as appropriate in |
| 853 | * their context. |
Ulrich Obergfell | a2a45b8 | 2015-11-05 18:44:53 -0800 | [diff] [blame] | 854 | * |
| 855 | * This function may only be called in a context that is protected against |
| 856 | * races with CPU hotplug - for example, via get_online_cpus(). |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 857 | */ |
| 858 | static int watchdog_park_threads(void) |
| 859 | { |
| 860 | int cpu, ret = 0; |
| 861 | |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 862 | for_each_watchdog_cpu(cpu) { |
| 863 | ret = kthread_park(per_cpu(softlockup_watchdog, cpu)); |
| 864 | if (ret) |
| 865 | break; |
| 866 | } |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 867 | |
| 868 | return ret; |
| 869 | } |
| 870 | |
| 871 | /* |
| 872 | * unpark all watchdog threads that are specified in 'watchdog_cpumask' |
Ulrich Obergfell | a2a45b8 | 2015-11-05 18:44:53 -0800 | [diff] [blame] | 873 | * |
| 874 | * This function may only be called in a context that is protected against |
| 875 | * races with CPU hotplug - for example, via get_online_cpus(). |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 876 | */ |
| 877 | static void watchdog_unpark_threads(void) |
| 878 | { |
| 879 | int cpu; |
| 880 | |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 881 | for_each_watchdog_cpu(cpu) |
| 882 | kthread_unpark(per_cpu(softlockup_watchdog, cpu)); |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame] | 883 | } |
| 884 | |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 885 | /* |
| 886 | * Suspend the hard and soft lockup detector by parking the watchdog threads. |
| 887 | */ |
Ulrich Obergfell | ec6a906 | 2015-09-04 15:45:28 -0700 | [diff] [blame] | 888 | int lockup_detector_suspend(void) |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 889 | { |
| 890 | int ret = 0; |
| 891 | |
Ulrich Obergfell | ee89e71 | 2015-11-05 18:44:47 -0800 | [diff] [blame] | 892 | get_online_cpus(); |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 893 | mutex_lock(&watchdog_proc_mutex); |
| 894 | /* |
| 895 | * Multiple suspend requests can be active in parallel (counted by |
| 896 | * the 'watchdog_suspended' variable). If the watchdog threads are |
| 897 | * running, the first caller takes care that they will be parked. |
| 898 | * The state of 'watchdog_running' cannot change while a suspend |
Ulrich Obergfell | ec6a906 | 2015-09-04 15:45:28 -0700 | [diff] [blame] | 899 | * request is active (see related code in 'proc' handlers). |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 900 | */ |
| 901 | if (watchdog_running && !watchdog_suspended) |
| 902 | ret = watchdog_park_threads(); |
| 903 | |
| 904 | if (ret == 0) |
| 905 | watchdog_suspended++; |
Ulrich Obergfell | c993590 | 2015-11-05 18:44:36 -0800 | [diff] [blame] | 906 | else { |
| 907 | watchdog_disable_all_cpus(); |
| 908 | pr_err("Failed to suspend lockup detectors, disabled\n"); |
| 909 | watchdog_enabled = 0; |
| 910 | } |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 911 | |
| 912 | mutex_unlock(&watchdog_proc_mutex); |
| 913 | |
| 914 | return ret; |
| 915 | } |
| 916 | |
| 917 | /* |
| 918 | * Resume the hard and soft lockup detector by unparking the watchdog threads. |
| 919 | */ |
Ulrich Obergfell | ec6a906 | 2015-09-04 15:45:28 -0700 | [diff] [blame] | 920 | void lockup_detector_resume(void) |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 921 | { |
| 922 | mutex_lock(&watchdog_proc_mutex); |
| 923 | |
| 924 | watchdog_suspended--; |
| 925 | /* |
| 926 | * The watchdog threads are unparked if they were previously running |
| 927 | * and if there is no more active suspend request. |
| 928 | */ |
| 929 | if (watchdog_running && !watchdog_suspended) |
| 930 | watchdog_unpark_threads(); |
| 931 | |
| 932 | mutex_unlock(&watchdog_proc_mutex); |
Ulrich Obergfell | ee89e71 | 2015-11-05 18:44:47 -0800 | [diff] [blame] | 933 | put_online_cpus(); |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 934 | } |
| 935 | |
Ulrich Obergfell | b43cb43 | 2015-11-05 18:44:33 -0800 | [diff] [blame] | 936 | static int update_watchdog_all_cpus(void) |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 937 | { |
Ulrich Obergfell | b43cb43 | 2015-11-05 18:44:33 -0800 | [diff] [blame] | 938 | int ret; |
| 939 | |
| 940 | ret = watchdog_park_threads(); |
| 941 | if (ret) |
| 942 | return ret; |
| 943 | |
Ulrich Obergfell | d4bdd0b | 2015-09-04 15:45:21 -0700 | [diff] [blame] | 944 | watchdog_unpark_threads(); |
Ulrich Obergfell | b43cb43 | 2015-11-05 18:44:33 -0800 | [diff] [blame] | 945 | |
| 946 | return 0; |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 947 | } |
| 948 | |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 949 | static int watchdog_enable_all_cpus(void) |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 950 | { |
| 951 | int err = 0; |
| 952 | |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 953 | if (!watchdog_running) { |
Frederic Weisbecker | 230ec93 | 2015-09-04 15:45:06 -0700 | [diff] [blame] | 954 | err = smpboot_register_percpu_thread_cpumask(&watchdog_threads, |
| 955 | &watchdog_cpumask); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 956 | if (err) |
| 957 | pr_err("Failed to create watchdog threads, disabled\n"); |
Frederic Weisbecker | 230ec93 | 2015-09-04 15:45:06 -0700 | [diff] [blame] | 958 | else |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 959 | watchdog_running = 1; |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 960 | } else { |
| 961 | /* |
| 962 | * Enable/disable the lockup detectors or |
| 963 | * change the sample period 'on the fly'. |
| 964 | */ |
Ulrich Obergfell | b43cb43 | 2015-11-05 18:44:33 -0800 | [diff] [blame] | 965 | err = update_watchdog_all_cpus(); |
| 966 | |
| 967 | if (err) { |
| 968 | watchdog_disable_all_cpus(); |
| 969 | pr_err("Failed to update lockup detectors, disabled\n"); |
| 970 | } |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 971 | } |
| 972 | |
Ulrich Obergfell | b43cb43 | 2015-11-05 18:44:33 -0800 | [diff] [blame] | 973 | if (err) |
| 974 | watchdog_enabled = 0; |
| 975 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 976 | return err; |
| 977 | } |
| 978 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 979 | static void watchdog_disable_all_cpus(void) |
| 980 | { |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 981 | if (watchdog_running) { |
| 982 | watchdog_running = 0; |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 983 | smpboot_unregister_percpu_thread(&watchdog_threads); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 984 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 985 | } |
| 986 | |
Ulrich Obergfell | 58cf690 | 2015-11-05 18:44:30 -0800 | [diff] [blame] | 987 | #ifdef CONFIG_SYSCTL |
| 988 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 989 | /* |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 990 | * Update the run state of the lockup detectors. |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 991 | */ |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 992 | static int proc_watchdog_update(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 993 | { |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 994 | int err = 0; |
| 995 | |
| 996 | /* |
| 997 | * Watchdog threads won't be started if they are already active. |
| 998 | * The 'watchdog_running' variable in watchdog_*_all_cpus() takes |
| 999 | * care of this. If those threads are already active, the sample |
| 1000 | * period will be updated and the lockup detectors will be enabled |
| 1001 | * or disabled 'on the fly'. |
| 1002 | */ |
| 1003 | if (watchdog_enabled && watchdog_thresh) |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 1004 | err = watchdog_enable_all_cpus(); |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 1005 | else |
| 1006 | watchdog_disable_all_cpus(); |
| 1007 | |
| 1008 | return err; |
| 1009 | |
| 1010 | } |
| 1011 | |
| 1012 | /* |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 1013 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter |
| 1014 | * |
| 1015 | * caller | table->data points to | 'which' contains the flag(s) |
| 1016 | * -------------------|-----------------------|----------------------------- |
| 1017 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED or'ed |
| 1018 | * | | with SOFT_WATCHDOG_ENABLED |
| 1019 | * -------------------|-----------------------|----------------------------- |
| 1020 | * proc_nmi_watchdog | nmi_watchdog_enabled | NMI_WATCHDOG_ENABLED |
| 1021 | * -------------------|-----------------------|----------------------------- |
| 1022 | * proc_soft_watchdog | soft_watchdog_enabled | SOFT_WATCHDOG_ENABLED |
| 1023 | */ |
| 1024 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, |
| 1025 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 1026 | { |
| 1027 | int err, old, new; |
| 1028 | int *watchdog_param = (int *)table->data; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 1029 | |
Ulrich Obergfell | 8614dde | 2015-11-05 18:44:50 -0800 | [diff] [blame] | 1030 | get_online_cpus(); |
Michal Hocko | 359e6fa | 2013-09-24 15:27:29 -0700 | [diff] [blame] | 1031 | mutex_lock(&watchdog_proc_mutex); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 1032 | |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 1033 | if (watchdog_suspended) { |
| 1034 | /* no parameter changes allowed while watchdog is suspended */ |
| 1035 | err = -EAGAIN; |
| 1036 | goto out; |
| 1037 | } |
| 1038 | |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 1039 | /* |
| 1040 | * If the parameter is being read return the state of the corresponding |
| 1041 | * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the |
| 1042 | * run state of the lockup detectors. |
| 1043 | */ |
| 1044 | if (!write) { |
| 1045 | *watchdog_param = (watchdog_enabled & which) != 0; |
| 1046 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 1047 | } else { |
| 1048 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 1049 | if (err) |
| 1050 | goto out; |
| 1051 | |
| 1052 | /* |
| 1053 | * There is a race window between fetching the current value |
| 1054 | * from 'watchdog_enabled' and storing the new value. During |
| 1055 | * this race window, watchdog_nmi_enable() can sneak in and |
| 1056 | * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'. |
| 1057 | * The 'cmpxchg' detects this race and the loop retries. |
| 1058 | */ |
| 1059 | do { |
| 1060 | old = watchdog_enabled; |
| 1061 | /* |
| 1062 | * If the parameter value is not zero set the |
| 1063 | * corresponding bit(s), else clear it(them). |
| 1064 | */ |
| 1065 | if (*watchdog_param) |
| 1066 | new = old | which; |
| 1067 | else |
| 1068 | new = old & ~which; |
| 1069 | } while (cmpxchg(&watchdog_enabled, old, new) != old); |
| 1070 | |
| 1071 | /* |
Ulrich Obergfell | b43cb43 | 2015-11-05 18:44:33 -0800 | [diff] [blame] | 1072 | * Update the run state of the lockup detectors. There is _no_ |
| 1073 | * need to check the value returned by proc_watchdog_update() |
| 1074 | * and to restore the previous value of 'watchdog_enabled' as |
| 1075 | * both lockup detectors are disabled if proc_watchdog_update() |
| 1076 | * returns an error. |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 1077 | */ |
Joshua Hunt | a1ee193 | 2016-03-17 14:17:23 -0700 | [diff] [blame] | 1078 | if (old == new) |
| 1079 | goto out; |
| 1080 | |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 1081 | err = proc_watchdog_update(); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 1082 | } |
| 1083 | out: |
| 1084 | mutex_unlock(&watchdog_proc_mutex); |
Ulrich Obergfell | 8614dde | 2015-11-05 18:44:50 -0800 | [diff] [blame] | 1085 | put_online_cpus(); |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 1086 | return err; |
| 1087 | } |
| 1088 | |
| 1089 | /* |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 1090 | * /proc/sys/kernel/watchdog |
| 1091 | */ |
| 1092 | int proc_watchdog(struct ctl_table *table, int write, |
| 1093 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 1094 | { |
| 1095 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, |
| 1096 | table, write, buffer, lenp, ppos); |
| 1097 | } |
| 1098 | |
| 1099 | /* |
| 1100 | * /proc/sys/kernel/nmi_watchdog |
| 1101 | */ |
| 1102 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
| 1103 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 1104 | { |
| 1105 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, |
| 1106 | table, write, buffer, lenp, ppos); |
| 1107 | } |
| 1108 | |
| 1109 | /* |
| 1110 | * /proc/sys/kernel/soft_watchdog |
| 1111 | */ |
| 1112 | int proc_soft_watchdog(struct ctl_table *table, int write, |
| 1113 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 1114 | { |
| 1115 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, |
| 1116 | table, write, buffer, lenp, ppos); |
| 1117 | } |
| 1118 | |
| 1119 | /* |
| 1120 | * /proc/sys/kernel/watchdog_thresh |
| 1121 | */ |
| 1122 | int proc_watchdog_thresh(struct ctl_table *table, int write, |
| 1123 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 1124 | { |
Joshua Hunt | a1ee193 | 2016-03-17 14:17:23 -0700 | [diff] [blame] | 1125 | int err, old, new; |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 1126 | |
Ulrich Obergfell | 8614dde | 2015-11-05 18:44:50 -0800 | [diff] [blame] | 1127 | get_online_cpus(); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 1128 | mutex_lock(&watchdog_proc_mutex); |
| 1129 | |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 1130 | if (watchdog_suspended) { |
| 1131 | /* no parameter changes allowed while watchdog is suspended */ |
| 1132 | err = -EAGAIN; |
| 1133 | goto out; |
| 1134 | } |
| 1135 | |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 1136 | old = ACCESS_ONCE(watchdog_thresh); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 1137 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 1138 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 1139 | if (err || !write) |
Michal Hocko | 359e6fa | 2013-09-24 15:27:29 -0700 | [diff] [blame] | 1140 | goto out; |
Mandeep Singh Baines | e04ab2b | 2011-05-22 22:10:21 -0700 | [diff] [blame] | 1141 | |
anish kumar | b66a235 | 2013-03-12 14:44:08 -0400 | [diff] [blame] | 1142 | /* |
Ulrich Obergfell | d283c64 | 2015-11-05 18:44:27 -0800 | [diff] [blame] | 1143 | * Update the sample period. Restore on failure. |
anish kumar | b66a235 | 2013-03-12 14:44:08 -0400 | [diff] [blame] | 1144 | */ |
Joshua Hunt | a1ee193 | 2016-03-17 14:17:23 -0700 | [diff] [blame] | 1145 | new = ACCESS_ONCE(watchdog_thresh); |
| 1146 | if (old == new) |
| 1147 | goto out; |
| 1148 | |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 1149 | set_sample_period(); |
| 1150 | err = proc_watchdog_update(); |
Ulrich Obergfell | d283c64 | 2015-11-05 18:44:27 -0800 | [diff] [blame] | 1151 | if (err) { |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 1152 | watchdog_thresh = old; |
Ulrich Obergfell | d283c64 | 2015-11-05 18:44:27 -0800 | [diff] [blame] | 1153 | set_sample_period(); |
| 1154 | } |
Michal Hocko | 359e6fa | 2013-09-24 15:27:29 -0700 | [diff] [blame] | 1155 | out: |
| 1156 | mutex_unlock(&watchdog_proc_mutex); |
Ulrich Obergfell | 8614dde | 2015-11-05 18:44:50 -0800 | [diff] [blame] | 1157 | put_online_cpus(); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 1158 | return err; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 1159 | } |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 1160 | |
| 1161 | /* |
| 1162 | * The cpumask is the mask of possible cpus that the watchdog can run |
| 1163 | * on, not the mask of cpus it is actually running on. This allows the |
| 1164 | * user to specify a mask that will include cpus that have not yet |
| 1165 | * been brought online, if desired. |
| 1166 | */ |
| 1167 | int proc_watchdog_cpumask(struct ctl_table *table, int write, |
| 1168 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 1169 | { |
| 1170 | int err; |
| 1171 | |
Ulrich Obergfell | 8614dde | 2015-11-05 18:44:50 -0800 | [diff] [blame] | 1172 | get_online_cpus(); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 1173 | mutex_lock(&watchdog_proc_mutex); |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 1174 | |
| 1175 | if (watchdog_suspended) { |
| 1176 | /* no parameter changes allowed while watchdog is suspended */ |
| 1177 | err = -EAGAIN; |
| 1178 | goto out; |
| 1179 | } |
| 1180 | |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 1181 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
| 1182 | if (!err && write) { |
| 1183 | /* Remove impossible cpus to keep sysctl output cleaner. */ |
| 1184 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, |
| 1185 | cpu_possible_mask); |
| 1186 | |
| 1187 | if (watchdog_running) { |
| 1188 | /* |
| 1189 | * Failure would be due to being unable to allocate |
| 1190 | * a temporary cpumask, so we are likely not in a |
| 1191 | * position to do much else to make things better. |
| 1192 | */ |
| 1193 | if (smpboot_update_cpumask_percpu_thread( |
| 1194 | &watchdog_threads, &watchdog_cpumask) != 0) |
| 1195 | pr_err("cpumask update failed\n"); |
| 1196 | } |
| 1197 | } |
Ulrich Obergfell | 8c073d2 | 2015-09-04 15:45:18 -0700 | [diff] [blame] | 1198 | out: |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 1199 | mutex_unlock(&watchdog_proc_mutex); |
Ulrich Obergfell | 8614dde | 2015-11-05 18:44:50 -0800 | [diff] [blame] | 1200 | put_online_cpus(); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 1201 | return err; |
| 1202 | } |
| 1203 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 1204 | #endif /* CONFIG_SYSCTL */ |
| 1205 | |
Peter Zijlstra | 004417a | 2010-11-25 18:38:29 +0100 | [diff] [blame] | 1206 | void __init lockup_detector_init(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 1207 | { |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 1208 | set_sample_period(); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 1209 | |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 1210 | #ifdef CONFIG_NO_HZ_FULL |
| 1211 | if (tick_nohz_full_enabled()) { |
Frederic Weisbecker | 314b08ff | 2015-09-04 15:45:09 -0700 | [diff] [blame] | 1212 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
| 1213 | cpumask_copy(&watchdog_cpumask, housekeeping_mask); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 1214 | } else |
| 1215 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 1216 | #else |
| 1217 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 1218 | #endif |
| 1219 | |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 1220 | if (watchdog_enabled) |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 1221 | watchdog_enable_all_cpus(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 1222 | } |