diff options
| -rw-r--r-- | kernel/nmi_watchdog.c | 259 |
1 files changed, 0 insertions, 259 deletions
diff --git a/kernel/nmi_watchdog.c b/kernel/nmi_watchdog.c deleted file mode 100644 index a79d211c30df..000000000000 --- a/kernel/nmi_watchdog.c +++ /dev/null | |||
| @@ -1,259 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * Detect Hard Lockups using the NMI | ||
| 3 | * | ||
| 4 | * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc. | ||
| 5 | * | ||
| 6 | * this code detects hard lockups: incidents in where on a CPU | ||
| 7 | * the kernel does not respond to anything except NMI. | ||
| 8 | * | ||
| 9 | * Note: Most of this code is borrowed heavily from softlockup.c, | ||
| 10 | * so thanks to Ingo for the initial implementation. | ||
| 11 | * Some chunks also taken from arch/x86/kernel/apic/nmi.c, thanks | ||
| 12 | * to those contributors as well. | ||
| 13 | */ | ||
| 14 | |||
| 15 | #include <linux/mm.h> | ||
| 16 | #include <linux/cpu.h> | ||
| 17 | #include <linux/nmi.h> | ||
| 18 | #include <linux/init.h> | ||
| 19 | #include <linux/delay.h> | ||
| 20 | #include <linux/freezer.h> | ||
| 21 | #include <linux/lockdep.h> | ||
| 22 | #include <linux/notifier.h> | ||
| 23 | #include <linux/module.h> | ||
| 24 | #include <linux/sysctl.h> | ||
| 25 | |||
| 26 | #include <asm/irq_regs.h> | ||
| 27 | #include <linux/perf_event.h> | ||
| 28 | |||
| 29 | static DEFINE_PER_CPU(struct perf_event *, nmi_watchdog_ev); | ||
| 30 | static DEFINE_PER_CPU(int, nmi_watchdog_touch); | ||
| 31 | static DEFINE_PER_CPU(long, alert_counter); | ||
| 32 | |||
| 33 | static int panic_on_timeout; | ||
| 34 | |||
| 35 | void touch_nmi_watchdog(void) | ||
| 36 | { | ||
| 37 | __raw_get_cpu_var(nmi_watchdog_touch) = 1; | ||
| 38 | touch_softlockup_watchdog(); | ||
| 39 | } | ||
| 40 | EXPORT_SYMBOL(touch_nmi_watchdog); | ||
| 41 | |||
| 42 | void touch_all_nmi_watchdog(void) | ||
| 43 | { | ||
| 44 | int cpu; | ||
| 45 | |||
| 46 | for_each_online_cpu(cpu) | ||
| 47 | per_cpu(nmi_watchdog_touch, cpu) = 1; | ||
| 48 | touch_softlockup_watchdog(); | ||
| 49 | } | ||
| 50 | |||
| 51 | static int __init setup_nmi_watchdog(char *str) | ||
| 52 | { | ||
| 53 | if (!strncmp(str, "panic", 5)) { | ||
| 54 | panic_on_timeout = 1; | ||
| 55 | str = strchr(str, ','); | ||
| 56 | if (!str) | ||
| 57 | return 1; | ||
| 58 | ++str; | ||
| 59 | } | ||
| 60 | return 1; | ||
| 61 | } | ||
| 62 | __setup("nmi_watchdog=", setup_nmi_watchdog); | ||
| 63 | |||
| 64 | struct perf_event_attr wd_hw_attr = { | ||
| 65 | .type = PERF_TYPE_HARDWARE, | ||
| 66 | .config = PERF_COUNT_HW_CPU_CYCLES, | ||
| 67 | .size = sizeof(struct perf_event_attr), | ||
| 68 | .pinned = 1, | ||
| 69 | .disabled = 1, | ||
| 70 | }; | ||
| 71 | |||
| 72 | struct perf_event_attr wd_sw_attr = { | ||
| 73 | .type = PERF_TYPE_SOFTWARE, | ||
| 74 | .config = PERF_COUNT_SW_CPU_CLOCK, | ||
| 75 | .size = sizeof(struct perf_event_attr), | ||
| 76 | .pinned = 1, | ||
| 77 | .disabled = 1, | ||
| 78 | }; | ||
| 79 | |||
| 80 | void wd_overflow(struct perf_event *event, int nmi, | ||
| 81 | struct perf_sample_data *data, | ||
| 82 | struct pt_regs *regs) | ||
| 83 | { | ||
| 84 | int cpu = smp_processor_id(); | ||
| 85 | int touched = 0; | ||
| 86 | |||
| 87 | if (__get_cpu_var(nmi_watchdog_touch)) { | ||
| 88 | per_cpu(nmi_watchdog_touch, cpu) = 0; | ||
| 89 | touched = 1; | ||
| 90 | } | ||
| 91 | |||
| 92 | /* check to see if the cpu is doing anything */ | ||
| 93 | if (!touched && hw_nmi_is_cpu_stuck(regs)) { | ||
| 94 | /* | ||
| 95 | * Ayiee, looks like this CPU is stuck ... | ||
| 96 | * wait a few IRQs (5 seconds) before doing the oops ... | ||
| 97 | */ | ||
| 98 | per_cpu(alert_counter, cpu) += 1; | ||
| 99 | if (per_cpu(alert_counter, cpu) == 5) { | ||
| 100 | if (panic_on_timeout) | ||
| 101 | panic("NMI Watchdog detected LOCKUP on cpu %d", cpu); | ||
| 102 | else | ||
| 103 | WARN(1, "NMI Watchdog detected LOCKUP on cpu %d", cpu); | ||
| 104 | } | ||
| 105 | } else { | ||
| 106 | per_cpu(alert_counter, cpu) = 0; | ||
| 107 | } | ||
| 108 | |||
| 109 | return; | ||
| 110 | } | ||
| 111 | |||
| 112 | static int enable_nmi_watchdog(int cpu) | ||
| 113 | { | ||
| 114 | struct perf_event *event; | ||
| 115 | struct perf_event_attr *wd_attr; | ||
| 116 | |||
| 117 | event = per_cpu(nmi_watchdog_ev, cpu); | ||
| 118 | if (event && event->state > PERF_EVENT_STATE_OFF) | ||
| 119 | return 0; | ||
| 120 | |||
| 121 | if (event == NULL) { | ||
| 122 | /* Try to register using hardware perf events first */ | ||
| 123 | wd_attr = &wd_hw_attr; | ||
| 124 | wd_attr->sample_period = hw_nmi_get_sample_period(); | ||
| 125 | event = perf_event_create_kernel_counter(wd_attr, cpu, -1, wd_overflow); | ||
| 126 | if (IS_ERR(event)) { | ||
| 127 | /* hardware doesn't exist or not supported, fallback to software events */ | ||
| 128 | printk(KERN_INFO "nmi_watchdog: hardware not available, trying software events\n"); | ||
| 129 | wd_attr = &wd_sw_attr; | ||
| 130 | wd_attr->sample_period = NSEC_PER_SEC; | ||
| 131 | event = perf_event_create_kernel_counter(wd_attr, cpu, -1, wd_overflow); | ||
| 132 | if (IS_ERR(event)) { | ||
| 133 | printk(KERN_ERR "nmi watchdog failed to create perf event on %i: %p\n", cpu, event); | ||
| 134 | return -1; | ||
| 135 | } | ||
| 136 | } | ||
| 137 | per_cpu(nmi_watchdog_ev, cpu) = event; | ||
| 138 | } | ||
| 139 | perf_event_enable(per_cpu(nmi_watchdog_ev, cpu)); | ||
| 140 | return 0; | ||
| 141 | } | ||
| 142 | |||
| 143 | static void disable_nmi_watchdog(int cpu) | ||
| 144 | { | ||
| 145 | struct perf_event *event; | ||
| 146 | |||
| 147 | event = per_cpu(nmi_watchdog_ev, cpu); | ||
| 148 | if (event) { | ||
| 149 | perf_event_disable(per_cpu(nmi_watchdog_ev, cpu)); | ||
| 150 | per_cpu(nmi_watchdog_ev, cpu) = NULL; | ||
| 151 | perf_event_release_kernel(event); | ||
| 152 | } | ||
| 153 | } | ||
| 154 | |||
| 155 | #ifdef CONFIG_SYSCTL | ||
| 156 | /* | ||
| 157 | * proc handler for /proc/sys/kernel/nmi_watchdog | ||
| 158 | */ | ||
| 159 | int nmi_watchdog_enabled; | ||
| 160 | |||
| 161 | int proc_nmi_enabled(struct ctl_table *table, int write, | ||
| 162 | void __user *buffer, size_t *length, loff_t *ppos) | ||
| 163 | { | ||
| 164 | int cpu; | ||
| 165 | |||
| 166 | if (!write) { | ||
| 167 | struct perf_event *event; | ||
| 168 | for_each_online_cpu(cpu) { | ||
| 169 | event = per_cpu(nmi_watchdog_ev, cpu); | ||
| 170 | if (event && event->state > PERF_EVENT_STATE_OFF) { | ||
| 171 | nmi_watchdog_enabled = 1; | ||
| 172 | break; | ||
| 173 | } | ||
| 174 | } | ||
| 175 | proc_dointvec(table, write, buffer, length, ppos); | ||
| 176 | return 0; | ||
| 177 | } | ||
| 178 | |||
| 179 | touch_all_nmi_watchdog(); | ||
| 180 | proc_dointvec(table, write, buffer, length, ppos); | ||
| 181 | if (nmi_watchdog_enabled) { | ||
| 182 | for_each_online_cpu(cpu) | ||
| 183 | if (enable_nmi_watchdog(cpu)) { | ||
| 184 | printk(KERN_ERR "NMI watchdog failed configuration, " | ||
| 185 | " can not be enabled\n"); | ||
| 186 | } | ||
| 187 | } else { | ||
| 188 | for_each_online_cpu(cpu) | ||
| 189 | disable_nmi_watchdog(cpu); | ||
| 190 | } | ||
| 191 | return 0; | ||
| 192 | } | ||
| 193 | |||
| 194 | #endif /* CONFIG_SYSCTL */ | ||
| 195 | |||
| 196 | /* | ||
| 197 | * Create/destroy watchdog threads as CPUs come and go: | ||
| 198 | */ | ||
| 199 | static int __cpuinit | ||
| 200 | cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) | ||
| 201 | { | ||
| 202 | int hotcpu = (unsigned long)hcpu; | ||
| 203 | |||
| 204 | switch (action) { | ||
| 205 | case CPU_UP_PREPARE: | ||
| 206 | case CPU_UP_PREPARE_FROZEN: | ||
| 207 | per_cpu(nmi_watchdog_touch, hotcpu) = 0; | ||
| 208 | break; | ||
| 209 | case CPU_ONLINE: | ||
| 210 | case CPU_ONLINE_FROZEN: | ||
| 211 | if (enable_nmi_watchdog(hotcpu)) | ||
| 212 | return NOTIFY_BAD; | ||
| 213 | break; | ||
| 214 | #ifdef CONFIG_HOTPLUG_CPU | ||
| 215 | case CPU_UP_CANCELED: | ||
| 216 | case CPU_UP_CANCELED_FROZEN: | ||
| 217 | disable_nmi_watchdog(hotcpu); | ||
| 218 | case CPU_DEAD: | ||
| 219 | case CPU_DEAD_FROZEN: | ||
| 220 | break; | ||
| 221 | #endif /* CONFIG_HOTPLUG_CPU */ | ||
| 222 | } | ||
| 223 | return NOTIFY_OK; | ||
| 224 | } | ||
| 225 | |||
| 226 | static struct notifier_block __cpuinitdata cpu_nfb = { | ||
| 227 | .notifier_call = cpu_callback | ||
| 228 | }; | ||
| 229 | |||
| 230 | static int __initdata nonmi_watchdog; | ||
| 231 | |||
| 232 | static int __init nonmi_watchdog_setup(char *str) | ||
| 233 | { | ||
| 234 | nonmi_watchdog = 1; | ||
| 235 | return 1; | ||
| 236 | } | ||
| 237 | __setup("nonmi_watchdog", nonmi_watchdog_setup); | ||
| 238 | |||
| 239 | static int __init spawn_nmi_watchdog_task(void) | ||
| 240 | { | ||
| 241 | void *cpu = (void *)(long)smp_processor_id(); | ||
| 242 | int err; | ||
| 243 | |||
| 244 | if (nonmi_watchdog) | ||
| 245 | return 0; | ||
| 246 | |||
| 247 | printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n"); | ||
| 248 | |||
| 249 | err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); | ||
| 250 | if (err == NOTIFY_BAD) { | ||
| 251 | BUG(); | ||
| 252 | return 1; | ||
| 253 | } | ||
| 254 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); | ||
| 255 | register_cpu_notifier(&cpu_nfb); | ||
| 256 | |||
| 257 | return 0; | ||
| 258 | } | ||
| 259 | early_initcall(spawn_nmi_watchdog_task); | ||
