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> |
| 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> |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 22 | #include <linux/tick.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 23 | |
| 24 | #include <asm/irq_regs.h> |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 25 | #include <linux/kvm_para.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 26 | #include <linux/perf_event.h> |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame^] | 27 | #include <linux/kthread.h> |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 28 | |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 29 | /* |
| 30 | * The run state of the lockup detectors is controlled by the content of the |
| 31 | * 'watchdog_enabled' variable. Each lockup detector has its dedicated bit - |
| 32 | * bit 0 for the hard lockup detector and bit 1 for the soft lockup detector. |
| 33 | * |
| 34 | * 'watchdog_user_enabled', 'nmi_watchdog_enabled' and 'soft_watchdog_enabled' |
| 35 | * are variables that are only used as an 'interface' between the parameters |
| 36 | * in /proc/sys/kernel and the internal state bits in 'watchdog_enabled'. The |
| 37 | * 'watchdog_thresh' variable is handled differently because its value is not |
| 38 | * boolean, and the lockup detectors are 'suspended' while 'watchdog_thresh' |
| 39 | * is equal zero. |
| 40 | */ |
| 41 | #define NMI_WATCHDOG_ENABLED_BIT 0 |
| 42 | #define SOFT_WATCHDOG_ENABLED_BIT 1 |
| 43 | #define NMI_WATCHDOG_ENABLED (1 << NMI_WATCHDOG_ENABLED_BIT) |
| 44 | #define SOFT_WATCHDOG_ENABLED (1 << SOFT_WATCHDOG_ENABLED_BIT) |
| 45 | |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 46 | static DEFINE_MUTEX(watchdog_proc_mutex); |
| 47 | |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 48 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
| 49 | static unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED|NMI_WATCHDOG_ENABLED; |
| 50 | #else |
| 51 | static unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED; |
| 52 | #endif |
| 53 | int __read_mostly nmi_watchdog_enabled; |
| 54 | int __read_mostly soft_watchdog_enabled; |
| 55 | int __read_mostly watchdog_user_enabled; |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 56 | int __read_mostly watchdog_thresh = 10; |
Ulrich Obergfell | 84d56e6 | 2015-04-14 15:43:55 -0700 | [diff] [blame] | 57 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 58 | #ifdef CONFIG_SMP |
| 59 | int __read_mostly sysctl_softlockup_all_cpu_backtrace; |
| 60 | #else |
| 61 | #define sysctl_softlockup_all_cpu_backtrace 0 |
| 62 | #endif |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 63 | static struct cpumask watchdog_cpumask __read_mostly; |
| 64 | unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask); |
| 65 | |
| 66 | /* Helper for online, unparked cpus. */ |
| 67 | #define for_each_watchdog_cpu(cpu) \ |
| 68 | for_each_cpu_and((cpu), cpu_online_mask, &watchdog_cpumask) |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 69 | |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 70 | static int __read_mostly watchdog_running; |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 71 | static u64 __read_mostly sample_period; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 72 | |
| 73 | static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); |
| 74 | static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); |
| 75 | static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); |
| 76 | static DEFINE_PER_CPU(bool, softlockup_touch_sync); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 77 | static DEFINE_PER_CPU(bool, soft_watchdog_warn); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 78 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); |
| 79 | static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 80 | static DEFINE_PER_CPU(struct task_struct *, softlockup_task_ptr_saved); |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 81 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | cafcd80 | 2010-05-14 11:11:21 -0400 | [diff] [blame] | 82 | static DEFINE_PER_CPU(bool, hard_watchdog_warn); |
| 83 | static DEFINE_PER_CPU(bool, watchdog_nmi_touch); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 84 | static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); |
| 85 | static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); |
| 86 | #endif |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 87 | static unsigned long soft_lockup_nmi_warn; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 88 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 89 | /* boot commands */ |
| 90 | /* |
| 91 | * Should we panic when a soft-lockup or hard-lockup occurs: |
| 92 | */ |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 93 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | fef2c9b | 2011-03-22 16:34:16 -0700 | [diff] [blame] | 94 | static int hardlockup_panic = |
| 95 | CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 96 | /* |
| 97 | * We may not want to enable hard lockup detection by default in all cases, |
| 98 | * for example when running the kernel as a guest on a hypervisor. In these |
| 99 | * cases this function can be called to disable hard lockup detection. This |
| 100 | * function should only be executed once by the boot processor before the |
| 101 | * kernel command line parameters are parsed, because otherwise it is not |
| 102 | * possible to override this in hardlockup_panic_setup(). |
| 103 | */ |
Ulrich Obergfell | 692297d | 2015-04-14 15:44:19 -0700 | [diff] [blame] | 104 | void hardlockup_detector_disable(void) |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 105 | { |
Ulrich Obergfell | 692297d | 2015-04-14 15:44:19 -0700 | [diff] [blame] | 106 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 107 | } |
| 108 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 109 | static int __init hardlockup_panic_setup(char *str) |
| 110 | { |
| 111 | if (!strncmp(str, "panic", 5)) |
| 112 | hardlockup_panic = 1; |
Don Zickus | fef2c9b | 2011-03-22 16:34:16 -0700 | [diff] [blame] | 113 | else if (!strncmp(str, "nopanic", 7)) |
| 114 | hardlockup_panic = 0; |
Don Zickus | 5dc3055 | 2010-11-29 17:07:17 -0500 | [diff] [blame] | 115 | else if (!strncmp(str, "0", 1)) |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 116 | watchdog_enabled &= ~NMI_WATCHDOG_ENABLED; |
| 117 | else if (!strncmp(str, "1", 1)) |
| 118 | watchdog_enabled |= NMI_WATCHDOG_ENABLED; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 119 | return 1; |
| 120 | } |
| 121 | __setup("nmi_watchdog=", hardlockup_panic_setup); |
| 122 | #endif |
| 123 | |
| 124 | unsigned int __read_mostly softlockup_panic = |
| 125 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; |
| 126 | |
| 127 | static int __init softlockup_panic_setup(char *str) |
| 128 | { |
| 129 | softlockup_panic = simple_strtoul(str, NULL, 0); |
| 130 | |
| 131 | return 1; |
| 132 | } |
| 133 | __setup("softlockup_panic=", softlockup_panic_setup); |
| 134 | |
| 135 | static int __init nowatchdog_setup(char *str) |
| 136 | { |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 137 | watchdog_enabled = 0; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 138 | return 1; |
| 139 | } |
| 140 | __setup("nowatchdog", nowatchdog_setup); |
| 141 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 142 | static int __init nosoftlockup_setup(char *str) |
| 143 | { |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 144 | watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 145 | return 1; |
| 146 | } |
| 147 | __setup("nosoftlockup", nosoftlockup_setup); |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 148 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 149 | #ifdef CONFIG_SMP |
| 150 | static int __init softlockup_all_cpu_backtrace_setup(char *str) |
| 151 | { |
| 152 | sysctl_softlockup_all_cpu_backtrace = |
| 153 | !!simple_strtol(str, NULL, 0); |
| 154 | return 1; |
| 155 | } |
| 156 | __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup); |
| 157 | #endif |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 158 | |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 159 | /* |
| 160 | * Hard-lockup warnings should be triggered after just a few seconds. Soft- |
| 161 | * lockups can have false positives under extreme conditions. So we generally |
| 162 | * want a higher threshold for soft lockups than for hard lockups. So we couple |
| 163 | * the thresholds with a factor: we make the soft threshold twice the amount of |
| 164 | * time the hard threshold is. |
| 165 | */ |
Ingo Molnar | 6e9101a | 2011-05-24 05:43:18 +0200 | [diff] [blame] | 166 | static int get_softlockup_thresh(void) |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 167 | { |
| 168 | return watchdog_thresh * 2; |
| 169 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 170 | |
| 171 | /* |
| 172 | * Returns seconds, approximately. We don't need nanosecond |
| 173 | * resolution, and we don't need to waste time with a big divide when |
| 174 | * 2^30ns == 1.074s. |
| 175 | */ |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 176 | static unsigned long get_timestamp(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 177 | { |
Cyril Bur | 545a2bf | 2015-02-12 15:01:24 -0800 | [diff] [blame] | 178 | return running_clock() >> 30LL; /* 2^30 ~= 10^9 */ |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 179 | } |
| 180 | |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 181 | static void set_sample_period(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 182 | { |
| 183 | /* |
Mandeep Singh Baines | 586692a | 2011-05-22 22:10:22 -0700 | [diff] [blame] | 184 | * convert watchdog_thresh from seconds to ns |
Fernando Luis Vázquez Cao | 86f5e6a | 2012-02-09 17:42:22 -0500 | [diff] [blame] | 185 | * the divide by 5 is to give hrtimer several chances (two |
| 186 | * or three with the current relation between the soft |
| 187 | * and hard thresholds) to increment before the |
| 188 | * hardlockup detector generates a warning |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 189 | */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 190 | sample_period = get_softlockup_thresh() * ((u64)NSEC_PER_SEC / 5); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 191 | } |
| 192 | |
| 193 | /* Commands for resetting the watchdog */ |
| 194 | static void __touch_watchdog(void) |
| 195 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 196 | __this_cpu_write(watchdog_touch_ts, get_timestamp()); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 197 | } |
| 198 | |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 199 | void touch_softlockup_watchdog(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 200 | { |
Andrew Morton | 7861144 | 2014-04-18 15:07:12 -0700 | [diff] [blame] | 201 | /* |
| 202 | * Preemption can be enabled. It doesn't matter which CPU's timestamp |
| 203 | * gets zeroed here, so use the raw_ operation. |
| 204 | */ |
| 205 | raw_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 206 | } |
Ingo Molnar | 0167c78 | 2010-05-13 08:53:33 +0200 | [diff] [blame] | 207 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 208 | |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 209 | void touch_all_softlockup_watchdogs(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 210 | { |
| 211 | int cpu; |
| 212 | |
| 213 | /* |
| 214 | * this is done lockless |
| 215 | * do we care if a 0 races with a timestamp? |
| 216 | * all it means is the softlock check starts one cycle later |
| 217 | */ |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 218 | for_each_watchdog_cpu(cpu) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 219 | per_cpu(watchdog_touch_ts, cpu) = 0; |
| 220 | } |
| 221 | |
Don Zickus | cafcd80 | 2010-05-14 11:11:21 -0400 | [diff] [blame] | 222 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 223 | void touch_nmi_watchdog(void) |
| 224 | { |
Ben Zhang | 62572e2 | 2014-04-03 14:47:18 -0700 | [diff] [blame] | 225 | /* |
| 226 | * Using __raw here because some code paths have |
| 227 | * preemption enabled. If preemption is enabled |
| 228 | * then interrupts should be enabled too, in which |
| 229 | * case we shouldn't have to worry about the watchdog |
| 230 | * going off. |
| 231 | */ |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 232 | raw_cpu_write(watchdog_nmi_touch, true); |
Don Zickus | 332fbdb | 2010-05-07 17:11:45 -0400 | [diff] [blame] | 233 | touch_softlockup_watchdog(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 234 | } |
| 235 | EXPORT_SYMBOL(touch_nmi_watchdog); |
| 236 | |
Don Zickus | cafcd80 | 2010-05-14 11:11:21 -0400 | [diff] [blame] | 237 | #endif |
| 238 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 239 | void touch_softlockup_watchdog_sync(void) |
| 240 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 241 | __this_cpu_write(softlockup_touch_sync, true); |
| 242 | __this_cpu_write(watchdog_touch_ts, 0); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 243 | } |
| 244 | |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 245 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 246 | /* watchdog detector functions */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 247 | static int is_hardlockup(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 248 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 249 | unsigned long hrint = __this_cpu_read(hrtimer_interrupts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 250 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 251 | if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 252 | return 1; |
| 253 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 254 | __this_cpu_write(hrtimer_interrupts_saved, hrint); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 255 | return 0; |
| 256 | } |
| 257 | #endif |
| 258 | |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 259 | static int is_softlockup(unsigned long touch_ts) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 260 | { |
Namhyung Kim | c06b4f1 | 2012-12-27 11:49:44 +0900 | [diff] [blame] | 261 | unsigned long now = get_timestamp(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 262 | |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 263 | if (watchdog_enabled & SOFT_WATCHDOG_ENABLED) { |
| 264 | /* Warn about unreasonable delays. */ |
| 265 | if (time_after(now, touch_ts + get_softlockup_thresh())) |
| 266 | return now - touch_ts; |
| 267 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 268 | return 0; |
| 269 | } |
| 270 | |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 271 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Cyrill Gorcunov | 1880c4a | 2011-06-23 16:49:18 +0400 | [diff] [blame] | 272 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 273 | static struct perf_event_attr wd_hw_attr = { |
| 274 | .type = PERF_TYPE_HARDWARE, |
| 275 | .config = PERF_COUNT_HW_CPU_CYCLES, |
| 276 | .size = sizeof(struct perf_event_attr), |
| 277 | .pinned = 1, |
| 278 | .disabled = 1, |
| 279 | }; |
| 280 | |
| 281 | /* Callback function for perf event subsystem */ |
Peter Zijlstra | a8b0ca1 | 2011-06-27 14:41:57 +0200 | [diff] [blame] | 282 | static void watchdog_overflow_callback(struct perf_event *event, |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 283 | struct perf_sample_data *data, |
| 284 | struct pt_regs *regs) |
| 285 | { |
Peter Zijlstra | c6db67c | 2010-08-20 11:49:15 +0200 | [diff] [blame] | 286 | /* Ensure the watchdog never gets throttled */ |
| 287 | event->hw.interrupts = 0; |
| 288 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 289 | if (__this_cpu_read(watchdog_nmi_touch) == true) { |
| 290 | __this_cpu_write(watchdog_nmi_touch, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 291 | return; |
| 292 | } |
| 293 | |
| 294 | /* check for a hardlockup |
| 295 | * This is done by making sure our timer interrupt |
| 296 | * is incrementing. The timer interrupt should have |
| 297 | * fired multiple times before we overflow'd. If it hasn't |
| 298 | * then this is a good indication the cpu is stuck |
| 299 | */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 300 | if (is_hardlockup()) { |
| 301 | int this_cpu = smp_processor_id(); |
| 302 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 303 | /* only print hardlockups once */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 304 | if (__this_cpu_read(hard_watchdog_warn) == true) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 305 | return; |
| 306 | |
| 307 | if (hardlockup_panic) |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 308 | panic("Watchdog detected hard LOCKUP on cpu %d", |
| 309 | this_cpu); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 310 | else |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 311 | WARN(1, "Watchdog detected hard LOCKUP on cpu %d", |
| 312 | this_cpu); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 313 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 314 | __this_cpu_write(hard_watchdog_warn, true); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 315 | return; |
| 316 | } |
| 317 | |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 318 | __this_cpu_write(hard_watchdog_warn, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 319 | return; |
| 320 | } |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 321 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
| 322 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 323 | static void watchdog_interrupt_count(void) |
| 324 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 325 | __this_cpu_inc(hrtimer_interrupts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 326 | } |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 327 | |
| 328 | static int watchdog_nmi_enable(unsigned int cpu); |
| 329 | static void watchdog_nmi_disable(unsigned int cpu); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 330 | |
| 331 | /* watchdog kicker functions */ |
| 332 | static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) |
| 333 | { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 334 | unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 335 | struct pt_regs *regs = get_irq_regs(); |
| 336 | int duration; |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 337 | int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 338 | |
| 339 | /* kick the hardlockup detector */ |
| 340 | watchdog_interrupt_count(); |
| 341 | |
| 342 | /* kick the softlockup detector */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 343 | wake_up_process(__this_cpu_read(softlockup_watchdog)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 344 | |
| 345 | /* .. and repeat */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 346 | hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 347 | |
| 348 | if (touch_ts == 0) { |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 349 | if (unlikely(__this_cpu_read(softlockup_touch_sync))) { |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 350 | /* |
| 351 | * If the time stamp was touched atomically |
| 352 | * make sure the scheduler tick is up to date. |
| 353 | */ |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 354 | __this_cpu_write(softlockup_touch_sync, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 355 | sched_clock_tick(); |
| 356 | } |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 357 | |
| 358 | /* Clear the guest paused flag on watchdog reset */ |
| 359 | kvm_check_and_clear_guest_paused(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 360 | __touch_watchdog(); |
| 361 | return HRTIMER_RESTART; |
| 362 | } |
| 363 | |
| 364 | /* check for a softlockup |
| 365 | * This is done by making sure a high priority task is |
| 366 | * being scheduled. The task touches the watchdog to |
| 367 | * indicate it is getting cpu time. If it hasn't then |
| 368 | * this is a good indication some task is hogging the cpu |
| 369 | */ |
Don Zickus | 26e09c6 | 2010-05-17 18:06:04 -0400 | [diff] [blame] | 370 | duration = is_softlockup(touch_ts); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 371 | if (unlikely(duration)) { |
Eric B Munson | 5d1c0f4 | 2012-03-10 14:37:28 -0500 | [diff] [blame] | 372 | /* |
| 373 | * If a virtual machine is stopped by the host it can look to |
| 374 | * the watchdog like a soft lockup, check to see if the host |
| 375 | * stopped the vm before we issue the warning |
| 376 | */ |
| 377 | if (kvm_check_and_clear_guest_paused()) |
| 378 | return HRTIMER_RESTART; |
| 379 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 380 | /* only warn once */ |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 381 | if (__this_cpu_read(soft_watchdog_warn) == true) { |
| 382 | /* |
| 383 | * When multiple processes are causing softlockups the |
| 384 | * softlockup detector only warns on the first one |
| 385 | * because the code relies on a full quiet cycle to |
| 386 | * re-arm. The second process prevents the quiet cycle |
| 387 | * and never gets reported. Use task pointers to detect |
| 388 | * this. |
| 389 | */ |
| 390 | if (__this_cpu_read(softlockup_task_ptr_saved) != |
| 391 | current) { |
| 392 | __this_cpu_write(soft_watchdog_warn, false); |
| 393 | __touch_watchdog(); |
| 394 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 395 | return HRTIMER_RESTART; |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 396 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 397 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 398 | if (softlockup_all_cpu_backtrace) { |
| 399 | /* Prevent multiple soft-lockup reports if one cpu is already |
| 400 | * engaged in dumping cpu back traces |
| 401 | */ |
| 402 | if (test_and_set_bit(0, &soft_lockup_nmi_warn)) { |
| 403 | /* Someone else will report us. Let's give up */ |
| 404 | __this_cpu_write(soft_watchdog_warn, true); |
| 405 | return HRTIMER_RESTART; |
| 406 | } |
| 407 | } |
| 408 | |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 409 | 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] | 410 | smp_processor_id(), duration, |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 411 | current->comm, task_pid_nr(current)); |
chai wen | b1a8de1 | 2014-10-09 15:25:17 -0700 | [diff] [blame] | 412 | __this_cpu_write(softlockup_task_ptr_saved, current); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 413 | print_modules(); |
| 414 | print_irqtrace_events(current); |
| 415 | if (regs) |
| 416 | show_regs(regs); |
| 417 | else |
| 418 | dump_stack(); |
| 419 | |
Aaron Tomlin | ed23587 | 2014-06-23 13:22:05 -0700 | [diff] [blame] | 420 | if (softlockup_all_cpu_backtrace) { |
| 421 | /* Avoid generating two back traces for current |
| 422 | * given that one is already made above |
| 423 | */ |
| 424 | trigger_allbutself_cpu_backtrace(); |
| 425 | |
| 426 | clear_bit(0, &soft_lockup_nmi_warn); |
| 427 | /* Barrier to sync with other cpus */ |
| 428 | smp_mb__after_atomic(); |
| 429 | } |
| 430 | |
Josh Hunt | 69361ee | 2014-08-08 14:22:31 -0700 | [diff] [blame] | 431 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 432 | if (softlockup_panic) |
| 433 | panic("softlockup: hung tasks"); |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 434 | __this_cpu_write(soft_watchdog_warn, true); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 435 | } else |
Christoph Lameter | 909ea96 | 2010-12-08 16:22:55 +0100 | [diff] [blame] | 436 | __this_cpu_write(soft_watchdog_warn, false); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 437 | |
| 438 | return HRTIMER_RESTART; |
| 439 | } |
| 440 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 441 | static void watchdog_set_prio(unsigned int policy, unsigned int prio) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 442 | { |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 443 | struct sched_param param = { .sched_priority = prio }; |
| 444 | |
| 445 | sched_setscheduler(current, policy, ¶m); |
| 446 | } |
| 447 | |
| 448 | static void watchdog_enable(unsigned int cpu) |
| 449 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 450 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 451 | |
Bjørn Mork | 3935e895 | 2012-12-19 20:51:31 +0100 | [diff] [blame] | 452 | /* kick off the timer for the hardlockup detector */ |
| 453 | hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
| 454 | hrtimer->function = watchdog_timer_fn; |
| 455 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 456 | /* Enable the perf event */ |
| 457 | watchdog_nmi_enable(cpu); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 458 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 459 | /* done here because hrtimer_start can only pin to smp_processor_id() */ |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 460 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 461 | HRTIMER_MODE_REL_PINNED); |
| 462 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 463 | /* initialize timestamp */ |
| 464 | watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1); |
| 465 | __touch_watchdog(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 466 | } |
| 467 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 468 | static void watchdog_disable(unsigned int cpu) |
| 469 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 470 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 471 | |
| 472 | watchdog_set_prio(SCHED_NORMAL, 0); |
| 473 | hrtimer_cancel(hrtimer); |
| 474 | /* disable the perf event */ |
| 475 | watchdog_nmi_disable(cpu); |
| 476 | } |
| 477 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 478 | static void watchdog_cleanup(unsigned int cpu, bool online) |
| 479 | { |
| 480 | watchdog_disable(cpu); |
| 481 | } |
| 482 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 483 | static int watchdog_should_run(unsigned int cpu) |
| 484 | { |
| 485 | return __this_cpu_read(hrtimer_interrupts) != |
| 486 | __this_cpu_read(soft_lockup_hrtimer_cnt); |
| 487 | } |
| 488 | |
| 489 | /* |
| 490 | * The watchdog thread function - touches the timestamp. |
| 491 | * |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 492 | * It only runs once every sample_period seconds (4 seconds by |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 493 | * default) to reset the softlockup timestamp. If this gets delayed |
| 494 | * for more than 2*watchdog_thresh seconds then the debug-printout |
| 495 | * triggers in watchdog_timer_fn(). |
| 496 | */ |
| 497 | static void watchdog(unsigned int cpu) |
| 498 | { |
| 499 | __this_cpu_write(soft_lockup_hrtimer_cnt, |
| 500 | __this_cpu_read(hrtimer_interrupts)); |
| 501 | __touch_watchdog(); |
Ulrich Obergfell | bcfba4f | 2015-04-14 15:44:10 -0700 | [diff] [blame] | 502 | |
| 503 | /* |
| 504 | * watchdog_nmi_enable() clears the NMI_WATCHDOG_ENABLED bit in the |
| 505 | * failure path. Check for failures that can occur asynchronously - |
| 506 | * for example, when CPUs are on-lined - and shut down the hardware |
| 507 | * perf event on each CPU accordingly. |
| 508 | * |
| 509 | * The only non-obvious place this bit can be cleared is through |
| 510 | * watchdog_nmi_enable(), so a pr_info() is placed there. Placing a |
| 511 | * pr_info here would be too noisy as it would result in a message |
| 512 | * every few seconds if the hardlockup was disabled but the softlockup |
| 513 | * enabled. |
| 514 | */ |
| 515 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) |
| 516 | watchdog_nmi_disable(cpu); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 517 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 518 | |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 519 | #ifdef CONFIG_HARDLOCKUP_DETECTOR |
Don Zickus | a702704 | 2012-06-13 09:35:48 -0400 | [diff] [blame] | 520 | /* |
| 521 | * People like the simple clean cpu node info on boot. |
| 522 | * Reduce the watchdog noise by only printing messages |
| 523 | * that are different from what cpu0 displayed. |
| 524 | */ |
| 525 | static unsigned long cpu0_err; |
| 526 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 527 | static int watchdog_nmi_enable(unsigned int cpu) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 528 | { |
| 529 | struct perf_event_attr *wd_attr; |
| 530 | struct perf_event *event = per_cpu(watchdog_ev, cpu); |
| 531 | |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 532 | /* nothing to do if the hard lockup detector is disabled */ |
| 533 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) |
| 534 | goto out; |
Ulrich Obergfell | 6e7458a | 2014-10-13 15:55:35 -0700 | [diff] [blame] | 535 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 536 | /* is it already setup and enabled? */ |
| 537 | if (event && event->state > PERF_EVENT_STATE_OFF) |
| 538 | goto out; |
| 539 | |
| 540 | /* it is setup but not enabled */ |
| 541 | if (event != NULL) |
| 542 | goto out_enable; |
| 543 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 544 | wd_attr = &wd_hw_attr; |
Mandeep Singh Baines | 4eec42f | 2011-05-22 22:10:23 -0700 | [diff] [blame] | 545 | wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh); |
Cyrill Gorcunov | 1880c4a | 2011-06-23 16:49:18 +0400 | [diff] [blame] | 546 | |
| 547 | /* Try to register using hardware perf events */ |
Avi Kivity | 4dc0da8 | 2011-06-29 18:42:35 +0300 | [diff] [blame] | 548 | 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] | 549 | |
| 550 | /* save cpu0 error for future comparision */ |
| 551 | if (cpu == 0 && IS_ERR(event)) |
| 552 | cpu0_err = PTR_ERR(event); |
| 553 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 554 | if (!IS_ERR(event)) { |
Don Zickus | a702704 | 2012-06-13 09:35:48 -0400 | [diff] [blame] | 555 | /* only print for cpu0 or different than cpu0 */ |
| 556 | if (cpu == 0 || cpu0_err) |
| 557 | 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] | 558 | goto out_save; |
| 559 | } |
| 560 | |
Ulrich Obergfell | bcfba4f | 2015-04-14 15:44:10 -0700 | [diff] [blame] | 561 | /* |
| 562 | * Disable the hard lockup detector if _any_ CPU fails to set up |
| 563 | * set up the hardware perf event. The watchdog() function checks |
| 564 | * the NMI_WATCHDOG_ENABLED bit periodically. |
| 565 | * |
| 566 | * The barriers are for syncing up watchdog_enabled across all the |
| 567 | * cpus, as clear_bit() does not use barriers. |
| 568 | */ |
| 569 | smp_mb__before_atomic(); |
| 570 | clear_bit(NMI_WATCHDOG_ENABLED_BIT, &watchdog_enabled); |
| 571 | smp_mb__after_atomic(); |
| 572 | |
Don Zickus | a702704 | 2012-06-13 09:35:48 -0400 | [diff] [blame] | 573 | /* skip displaying the same error again */ |
| 574 | if (cpu > 0 && (PTR_ERR(event) == cpu0_err)) |
| 575 | return PTR_ERR(event); |
Don Zickus | 5651f7f | 2011-02-09 14:02:33 -0500 | [diff] [blame] | 576 | |
| 577 | /* vary the KERN level based on the returned errno */ |
| 578 | if (PTR_ERR(event) == -EOPNOTSUPP) |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 579 | pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu); |
Don Zickus | 5651f7f | 2011-02-09 14:02:33 -0500 | [diff] [blame] | 580 | else if (PTR_ERR(event) == -ENOENT) |
Fabian Frederick | 656c3b7 | 2014-08-06 16:04:03 -0700 | [diff] [blame] | 581 | pr_warn("disabled (cpu%i): hardware events not enabled\n", |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 582 | cpu); |
Don Zickus | 5651f7f | 2011-02-09 14:02:33 -0500 | [diff] [blame] | 583 | else |
Andrew Morton | 4501980 | 2012-03-23 15:01:55 -0700 | [diff] [blame] | 584 | pr_err("disabled (cpu%i): unable to create perf event: %ld\n", |
| 585 | cpu, PTR_ERR(event)); |
Ulrich Obergfell | bcfba4f | 2015-04-14 15:44:10 -0700 | [diff] [blame] | 586 | |
| 587 | pr_info("Shutting down hard lockup detector on all cpus\n"); |
| 588 | |
Akinobu Mita | eac2433 | 2010-08-31 23:00:08 -0400 | [diff] [blame] | 589 | return PTR_ERR(event); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 590 | |
| 591 | /* success path */ |
| 592 | out_save: |
| 593 | per_cpu(watchdog_ev, cpu) = event; |
| 594 | out_enable: |
| 595 | perf_event_enable(per_cpu(watchdog_ev, cpu)); |
| 596 | out: |
| 597 | return 0; |
| 598 | } |
| 599 | |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 600 | static void watchdog_nmi_disable(unsigned int cpu) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 601 | { |
| 602 | struct perf_event *event = per_cpu(watchdog_ev, cpu); |
| 603 | |
| 604 | if (event) { |
| 605 | perf_event_disable(event); |
| 606 | per_cpu(watchdog_ev, cpu) = NULL; |
| 607 | |
| 608 | /* should be in cleanup, but blocks oprofile */ |
| 609 | perf_event_release_kernel(event); |
| 610 | } |
Ulrich Obergfell | df57714 | 2014-08-11 10:49:25 -0400 | [diff] [blame] | 611 | if (cpu == 0) { |
| 612 | /* watchdog_nmi_enable() expects this to be zero initially. */ |
| 613 | cpu0_err = 0; |
| 614 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 615 | } |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 616 | |
| 617 | void watchdog_nmi_enable_all(void) |
| 618 | { |
| 619 | int cpu; |
| 620 | |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 621 | mutex_lock(&watchdog_proc_mutex); |
| 622 | |
| 623 | if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED)) |
| 624 | goto unlock; |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 625 | |
| 626 | get_online_cpus(); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 627 | for_each_watchdog_cpu(cpu) |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 628 | watchdog_nmi_enable(cpu); |
| 629 | put_online_cpus(); |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 630 | |
| 631 | unlock: |
Michal Hocko | 1173ff0 | 2015-05-19 09:07:27 +0200 | [diff] [blame] | 632 | mutex_unlock(&watchdog_proc_mutex); |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 633 | } |
| 634 | |
| 635 | void watchdog_nmi_disable_all(void) |
| 636 | { |
| 637 | int cpu; |
| 638 | |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 639 | mutex_lock(&watchdog_proc_mutex); |
| 640 | |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 641 | if (!watchdog_running) |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 642 | goto unlock; |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 643 | |
| 644 | get_online_cpus(); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 645 | for_each_watchdog_cpu(cpu) |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 646 | watchdog_nmi_disable(cpu); |
| 647 | put_online_cpus(); |
Peter Zijlstra | ab992dc | 2015-05-18 11:31:50 +0200 | [diff] [blame] | 648 | |
| 649 | unlock: |
| 650 | mutex_unlock(&watchdog_proc_mutex); |
Stephane Eranian | b3738d2 | 2014-11-17 20:07:03 +0100 | [diff] [blame] | 651 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 652 | #else |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 653 | static int watchdog_nmi_enable(unsigned int cpu) { return 0; } |
| 654 | static void watchdog_nmi_disable(unsigned int cpu) { return; } |
Frederic Weisbecker | 23637d4 | 2010-05-15 23:15:20 +0200 | [diff] [blame] | 655 | #endif /* CONFIG_HARDLOCKUP_DETECTOR */ |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 656 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 657 | static struct smp_hotplug_thread watchdog_threads = { |
| 658 | .store = &softlockup_watchdog, |
| 659 | .thread_should_run = watchdog_should_run, |
| 660 | .thread_fn = watchdog, |
| 661 | .thread_comm = "watchdog/%u", |
| 662 | .setup = watchdog_enable, |
| 663 | .cleanup = watchdog_cleanup, |
| 664 | .park = watchdog_disable, |
| 665 | .unpark = watchdog_enable, |
| 666 | }; |
| 667 | |
Ulrich Obergfell | 81a4bee | 2015-09-04 15:45:15 -0700 | [diff] [blame^] | 668 | /* |
| 669 | * park all watchdog threads that are specified in 'watchdog_cpumask' |
| 670 | */ |
| 671 | static int watchdog_park_threads(void) |
| 672 | { |
| 673 | int cpu, ret = 0; |
| 674 | |
| 675 | get_online_cpus(); |
| 676 | for_each_watchdog_cpu(cpu) { |
| 677 | ret = kthread_park(per_cpu(softlockup_watchdog, cpu)); |
| 678 | if (ret) |
| 679 | break; |
| 680 | } |
| 681 | if (ret) { |
| 682 | for_each_watchdog_cpu(cpu) |
| 683 | kthread_unpark(per_cpu(softlockup_watchdog, cpu)); |
| 684 | } |
| 685 | put_online_cpus(); |
| 686 | |
| 687 | return ret; |
| 688 | } |
| 689 | |
| 690 | /* |
| 691 | * unpark all watchdog threads that are specified in 'watchdog_cpumask' |
| 692 | */ |
| 693 | static void watchdog_unpark_threads(void) |
| 694 | { |
| 695 | int cpu; |
| 696 | |
| 697 | get_online_cpus(); |
| 698 | for_each_watchdog_cpu(cpu) |
| 699 | kthread_unpark(per_cpu(softlockup_watchdog, cpu)); |
| 700 | put_online_cpus(); |
| 701 | } |
| 702 | |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 703 | static void restart_watchdog_hrtimer(void *info) |
| 704 | { |
Christoph Lameter | f7f66b0 | 2014-08-17 12:30:34 -0500 | [diff] [blame] | 705 | struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer); |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 706 | int ret; |
| 707 | |
| 708 | /* |
| 709 | * No need to cancel and restart hrtimer if it is currently executing |
| 710 | * because it will reprogram itself with the new period now. |
| 711 | * We should never see it unqueued here because we are running per-cpu |
| 712 | * with interrupts disabled. |
| 713 | */ |
| 714 | ret = hrtimer_try_to_cancel(hrtimer); |
| 715 | if (ret == 1) |
| 716 | hrtimer_start(hrtimer, ns_to_ktime(sample_period), |
| 717 | HRTIMER_MODE_REL_PINNED); |
| 718 | } |
| 719 | |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 720 | static void update_watchdog(int cpu) |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 721 | { |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 722 | /* |
| 723 | * Make sure that perf event counter will adopt to a new |
| 724 | * sampling period. Updating the sampling period directly would |
| 725 | * be much nicer but we do not have an API for that now so |
| 726 | * let's use a big hammer. |
| 727 | * Hrtimer will adopt the new period on the next tick but this |
| 728 | * might be late already so we have to restart the timer as well. |
| 729 | */ |
| 730 | watchdog_nmi_disable(cpu); |
Frederic Weisbecker | e0a23b06 | 2014-02-24 16:40:00 +0100 | [diff] [blame] | 731 | smp_call_function_single(cpu, restart_watchdog_hrtimer, NULL, 1); |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 732 | watchdog_nmi_enable(cpu); |
| 733 | } |
| 734 | |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 735 | static void update_watchdog_all_cpus(void) |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 736 | { |
| 737 | int cpu; |
| 738 | |
| 739 | get_online_cpus(); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 740 | for_each_watchdog_cpu(cpu) |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 741 | update_watchdog(cpu); |
Michal Hocko | 9809b18 | 2013-09-24 15:27:30 -0700 | [diff] [blame] | 742 | put_online_cpus(); |
| 743 | } |
| 744 | |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 745 | static int watchdog_enable_all_cpus(void) |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 746 | { |
| 747 | int err = 0; |
| 748 | |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 749 | if (!watchdog_running) { |
Frederic Weisbecker | 230ec93 | 2015-09-04 15:45:06 -0700 | [diff] [blame] | 750 | err = smpboot_register_percpu_thread_cpumask(&watchdog_threads, |
| 751 | &watchdog_cpumask); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 752 | if (err) |
| 753 | pr_err("Failed to create watchdog threads, disabled\n"); |
Frederic Weisbecker | 230ec93 | 2015-09-04 15:45:06 -0700 | [diff] [blame] | 754 | else |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 755 | watchdog_running = 1; |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 756 | } else { |
| 757 | /* |
| 758 | * Enable/disable the lockup detectors or |
| 759 | * change the sample period 'on the fly'. |
| 760 | */ |
| 761 | update_watchdog_all_cpus(); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 762 | } |
| 763 | |
| 764 | return err; |
| 765 | } |
| 766 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 767 | /* prepare/enable/disable routines */ |
Vasily Averin | 4ff8195 | 2011-10-31 17:11:18 -0700 | [diff] [blame] | 768 | /* sysctl functions */ |
| 769 | #ifdef CONFIG_SYSCTL |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 770 | static void watchdog_disable_all_cpus(void) |
| 771 | { |
Frederic Weisbecker | 3c00ea8 | 2013-05-19 20:45:15 +0200 | [diff] [blame] | 772 | if (watchdog_running) { |
| 773 | watchdog_running = 0; |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 774 | smpboot_unregister_percpu_thread(&watchdog_threads); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 775 | } |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 776 | } |
| 777 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 778 | /* |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 779 | * Update the run state of the lockup detectors. |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 780 | */ |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 781 | static int proc_watchdog_update(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 782 | { |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 783 | int err = 0; |
| 784 | |
| 785 | /* |
| 786 | * Watchdog threads won't be started if they are already active. |
| 787 | * The 'watchdog_running' variable in watchdog_*_all_cpus() takes |
| 788 | * care of this. If those threads are already active, the sample |
| 789 | * period will be updated and the lockup detectors will be enabled |
| 790 | * or disabled 'on the fly'. |
| 791 | */ |
| 792 | if (watchdog_enabled && watchdog_thresh) |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 793 | err = watchdog_enable_all_cpus(); |
Ulrich Obergfell | a0c9cbb | 2015-04-14 15:43:58 -0700 | [diff] [blame] | 794 | else |
| 795 | watchdog_disable_all_cpus(); |
| 796 | |
| 797 | return err; |
| 798 | |
| 799 | } |
| 800 | |
| 801 | /* |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 802 | * common function for watchdog, nmi_watchdog and soft_watchdog parameter |
| 803 | * |
| 804 | * caller | table->data points to | 'which' contains the flag(s) |
| 805 | * -------------------|-----------------------|----------------------------- |
| 806 | * proc_watchdog | watchdog_user_enabled | NMI_WATCHDOG_ENABLED or'ed |
| 807 | * | | with SOFT_WATCHDOG_ENABLED |
| 808 | * -------------------|-----------------------|----------------------------- |
| 809 | * proc_nmi_watchdog | nmi_watchdog_enabled | NMI_WATCHDOG_ENABLED |
| 810 | * -------------------|-----------------------|----------------------------- |
| 811 | * proc_soft_watchdog | soft_watchdog_enabled | SOFT_WATCHDOG_ENABLED |
| 812 | */ |
| 813 | static int proc_watchdog_common(int which, struct ctl_table *table, int write, |
| 814 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 815 | { |
| 816 | int err, old, new; |
| 817 | int *watchdog_param = (int *)table->data; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 818 | |
Michal Hocko | 359e6fa | 2013-09-24 15:27:29 -0700 | [diff] [blame] | 819 | mutex_lock(&watchdog_proc_mutex); |
Thomas Gleixner | bcd951c | 2012-07-16 10:42:38 +0000 | [diff] [blame] | 820 | |
Ulrich Obergfell | ef246a2 | 2015-04-14 15:44:05 -0700 | [diff] [blame] | 821 | /* |
| 822 | * If the parameter is being read return the state of the corresponding |
| 823 | * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the |
| 824 | * run state of the lockup detectors. |
| 825 | */ |
| 826 | if (!write) { |
| 827 | *watchdog_param = (watchdog_enabled & which) != 0; |
| 828 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 829 | } else { |
| 830 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
| 831 | if (err) |
| 832 | goto out; |
| 833 | |
| 834 | /* |
| 835 | * There is a race window between fetching the current value |
| 836 | * from 'watchdog_enabled' and storing the new value. During |
| 837 | * this race window, watchdog_nmi_enable() can sneak in and |
| 838 | * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'. |
| 839 | * The 'cmpxchg' detects this race and the loop retries. |
| 840 | */ |
| 841 | do { |
| 842 | old = watchdog_enabled; |
| 843 | /* |
| 844 | * If the parameter value is not zero set the |
| 845 | * corresponding bit(s), else clear it(them). |
| 846 | */ |
| 847 | if (*watchdog_param) |
| 848 | new = old | which; |
| 849 | else |
| 850 | new = old & ~which; |
| 851 | } while (cmpxchg(&watchdog_enabled, old, new) != old); |
| 852 | |
| 853 | /* |
| 854 | * Update the run state of the lockup detectors. |
| 855 | * Restore 'watchdog_enabled' on failure. |
| 856 | */ |
| 857 | err = proc_watchdog_update(); |
| 858 | if (err) |
| 859 | watchdog_enabled = old; |
| 860 | } |
| 861 | out: |
| 862 | mutex_unlock(&watchdog_proc_mutex); |
| 863 | return err; |
| 864 | } |
| 865 | |
| 866 | /* |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 867 | * /proc/sys/kernel/watchdog |
| 868 | */ |
| 869 | int proc_watchdog(struct ctl_table *table, int write, |
| 870 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 871 | { |
| 872 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED|SOFT_WATCHDOG_ENABLED, |
| 873 | table, write, buffer, lenp, ppos); |
| 874 | } |
| 875 | |
| 876 | /* |
| 877 | * /proc/sys/kernel/nmi_watchdog |
| 878 | */ |
| 879 | int proc_nmi_watchdog(struct ctl_table *table, int write, |
| 880 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 881 | { |
| 882 | return proc_watchdog_common(NMI_WATCHDOG_ENABLED, |
| 883 | table, write, buffer, lenp, ppos); |
| 884 | } |
| 885 | |
| 886 | /* |
| 887 | * /proc/sys/kernel/soft_watchdog |
| 888 | */ |
| 889 | int proc_soft_watchdog(struct ctl_table *table, int write, |
| 890 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 891 | { |
| 892 | return proc_watchdog_common(SOFT_WATCHDOG_ENABLED, |
| 893 | table, write, buffer, lenp, ppos); |
| 894 | } |
| 895 | |
| 896 | /* |
| 897 | * /proc/sys/kernel/watchdog_thresh |
| 898 | */ |
| 899 | int proc_watchdog_thresh(struct ctl_table *table, int write, |
| 900 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 901 | { |
| 902 | int err, old; |
| 903 | |
| 904 | mutex_lock(&watchdog_proc_mutex); |
| 905 | |
| 906 | old = ACCESS_ONCE(watchdog_thresh); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 907 | err = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 908 | |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 909 | if (err || !write) |
Michal Hocko | 359e6fa | 2013-09-24 15:27:29 -0700 | [diff] [blame] | 910 | goto out; |
Mandeep Singh Baines | e04ab2b | 2011-05-22 22:10:21 -0700 | [diff] [blame] | 911 | |
anish kumar | b66a235 | 2013-03-12 14:44:08 -0400 | [diff] [blame] | 912 | /* |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 913 | * Update the sample period. |
| 914 | * Restore 'watchdog_thresh' on failure. |
anish kumar | b66a235 | 2013-03-12 14:44:08 -0400 | [diff] [blame] | 915 | */ |
Ulrich Obergfell | 83a80a3 | 2015-04-14 15:44:08 -0700 | [diff] [blame] | 916 | set_sample_period(); |
| 917 | err = proc_watchdog_update(); |
| 918 | if (err) |
| 919 | watchdog_thresh = old; |
Michal Hocko | 359e6fa | 2013-09-24 15:27:29 -0700 | [diff] [blame] | 920 | out: |
| 921 | mutex_unlock(&watchdog_proc_mutex); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 922 | return err; |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 923 | } |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 924 | |
| 925 | /* |
| 926 | * The cpumask is the mask of possible cpus that the watchdog can run |
| 927 | * on, not the mask of cpus it is actually running on. This allows the |
| 928 | * user to specify a mask that will include cpus that have not yet |
| 929 | * been brought online, if desired. |
| 930 | */ |
| 931 | int proc_watchdog_cpumask(struct ctl_table *table, int write, |
| 932 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 933 | { |
| 934 | int err; |
| 935 | |
| 936 | mutex_lock(&watchdog_proc_mutex); |
| 937 | err = proc_do_large_bitmap(table, write, buffer, lenp, ppos); |
| 938 | if (!err && write) { |
| 939 | /* Remove impossible cpus to keep sysctl output cleaner. */ |
| 940 | cpumask_and(&watchdog_cpumask, &watchdog_cpumask, |
| 941 | cpu_possible_mask); |
| 942 | |
| 943 | if (watchdog_running) { |
| 944 | /* |
| 945 | * Failure would be due to being unable to allocate |
| 946 | * a temporary cpumask, so we are likely not in a |
| 947 | * position to do much else to make things better. |
| 948 | */ |
| 949 | if (smpboot_update_cpumask_percpu_thread( |
| 950 | &watchdog_threads, &watchdog_cpumask) != 0) |
| 951 | pr_err("cpumask update failed\n"); |
| 952 | } |
| 953 | } |
| 954 | mutex_unlock(&watchdog_proc_mutex); |
| 955 | return err; |
| 956 | } |
| 957 | |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 958 | #endif /* CONFIG_SYSCTL */ |
| 959 | |
Peter Zijlstra | 004417a | 2010-11-25 18:38:29 +0100 | [diff] [blame] | 960 | void __init lockup_detector_init(void) |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 961 | { |
Chuansheng Liu | 0f34c40 | 2012-12-17 15:59:50 -0800 | [diff] [blame] | 962 | set_sample_period(); |
Frederic Weisbecker | b8900bc | 2013-06-06 15:42:53 +0200 | [diff] [blame] | 963 | |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 964 | #ifdef CONFIG_NO_HZ_FULL |
| 965 | if (tick_nohz_full_enabled()) { |
Frederic Weisbecker | 314b08ff | 2015-09-04 15:45:09 -0700 | [diff] [blame] | 966 | pr_info("Disabling watchdog on nohz_full cores by default\n"); |
| 967 | cpumask_copy(&watchdog_cpumask, housekeeping_mask); |
Chris Metcalf | fe4ba3c | 2015-06-24 16:55:45 -0700 | [diff] [blame] | 968 | } else |
| 969 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 970 | #else |
| 971 | cpumask_copy(&watchdog_cpumask, cpu_possible_mask); |
| 972 | #endif |
| 973 | |
Ulrich Obergfell | 195daf6 | 2015-04-14 15:44:13 -0700 | [diff] [blame] | 974 | if (watchdog_enabled) |
Ulrich Obergfell | b2f57c3 | 2015-04-14 15:44:16 -0700 | [diff] [blame] | 975 | watchdog_enable_all_cpus(); |
Don Zickus | 58687ac | 2010-05-07 17:11:44 -0400 | [diff] [blame] | 976 | } |