arch/x86/kernel/cpu/mcheck/therm_throt.c

   1 /*
   2  * Thermal throttle event support code (such as syslog messaging and rate
   3  * limiting) that was factored out from x86_64 (mce_intel.c) and i386 (p4.c).
   4  *
   5  * This allows consistent reporting of CPU thermal throttle events.
   6  *
   7  * Maintains a counter in /sys that keeps track of the number of thermal
   8  * events, such that the user knows how bad the thermal problem might be
   9  * (since the logging to syslog and mcelog is rate limited).
  10  *
  11  * Author: Dmitriy Zavin (dmitriyz@google.com)
  12  *
  13  * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
  14  *          Inspired by Ross Biro's and Al Borchers' counter code.
  15  */
  16 #include <linux/interrupt.h>
  17 #include <linux/notifier.h>
  18 #include <linux/jiffies.h>
  19 #include <linux/percpu.h>
  20 #include <linux/sysdev.h>
  21 #include <linux/cpu.h>
  22
  23 #include <asm/therm_throt.h>
  24 #include <asm/idle.h>
  25 #include <asm/mce.h>
  26
  27 /* How long to wait between reporting thermal events */
  28 #define CHECK_INTERVAL          (300 * HZ)
  29
  30 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
  31 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
  32
  33 atomic_t therm_throt_en         = ATOMIC_INIT(0);
  34
  35 #ifdef CONFIG_SYSFS
  36 #define define_therm_throt_sysdev_one_ro(_name)                         \
  37         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
  38
  39 #define define_therm_throt_sysdev_show_func(name)                       \
  40 static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,   \
  41                                         struct sysdev_attribute *attr,  \
  42                                               char *buf)                \
  43 {                                                                       \
  44         unsigned int cpu = dev->id;                                     \
  45         ssize_t ret;                                                    \
  46                                                                         \
  47         preempt_disable();      /* CPU hotplug */                       \
  48         if (cpu_online(cpu))                                            \
  49                 ret = sprintf(buf, "%lu\n",                             \
  50                               per_cpu(thermal_throttle_##name, cpu));   \
  51         else                                                            \
  52                 ret = 0;                                                \
  53         preempt_enable();                                               \
  54                                                                         \
  55         return ret;                                                     \
  56 }
  57
  58 define_therm_throt_sysdev_show_func(count);
  59 define_therm_throt_sysdev_one_ro(count);
  60
  61 static struct attribute *thermal_throttle_attrs[] = {
  62         &attr_count.attr,
  63         NULL
  64 };
  65
  66 static struct attribute_group thermal_throttle_attr_group = {
  67         .attrs  = thermal_throttle_attrs,
  68         .name   = "thermal_throttle"
  69 };
  70 #endif /* CONFIG_SYSFS */
  71
  72 /***
  73  * therm_throt_process - Process thermal throttling event from interrupt
  74  * @curr: Whether the condition is current or not (boolean), since the
  75  *        thermal interrupt normally gets called both when the thermal
  76  *        event begins and once the event has ended.
  77  *
  78  * This function is called by the thermal interrupt after the
  79  * IRQ has been acknowledged.
  80  *
  81  * It will take care of rate limiting and printing messages to the syslog.
  82  *
  83  * Returns: 0 : Event should NOT be further logged, i.e. still in
  84  *              "timeout" from previous log message.
  85  *          1 : Event should be logged further, and a message has been
  86  *              printed to the syslog.
  87  */
  88 int therm_throt_process(int curr)
  89 {
  90         unsigned int cpu = smp_processor_id();
  91         __u64 tmp_jiffs = get_jiffies_64();
  92
  93         if (curr)
  94                 __get_cpu_var(thermal_throttle_count)++;
  95
  96         if (time_before64(tmp_jiffs, __get_cpu_var(next_check)))
  97                 return 0;
  98
  99         __get_cpu_var(next_check) = tmp_jiffs + CHECK_INTERVAL;
 100
 101         /* if we just entered the thermal event */
 102         if (curr) {
 103                 printk(KERN_CRIT "CPU%d: Temperature above threshold, "
 104                        "cpu clock throttled (total events = %lu)\n", cpu,
 105                        __get_cpu_var(thermal_throttle_count));
 106
 107                 add_taint(TAINT_MACHINE_CHECK);
 108         } else {
 109                 printk(KERN_CRIT "CPU%d: Temperature/speed normal\n", cpu);
 110         }
 111
 112         return 1;
 113 }
 114
 115 #ifdef CONFIG_SYSFS
 116 /* Add/Remove thermal_throttle interface for CPU device: */
 117 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
 118 {
 119         return sysfs_create_group(&sys_dev->kobj,
 120                                   &thermal_throttle_attr_group);
 121 }
 122
 123 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
 124 {
 125         sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
 126 }
 127
 128 /* Mutex protecting device creation against CPU hotplug: */
 129 static DEFINE_MUTEX(therm_cpu_lock);
 130
 131 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
 132 static __cpuinit int
 133 thermal_throttle_cpu_callback(struct notifier_block *nfb,
 134                               unsigned long action,
 135                               void *hcpu)
 136 {
 137         unsigned int cpu = (unsigned long)hcpu;
 138         struct sys_device *sys_dev;
 139         int err = 0;
 140
 141         sys_dev = get_cpu_sysdev(cpu);
 142
 143         switch (action) {
 144         case CPU_UP_PREPARE:
 145         case CPU_UP_PREPARE_FROZEN:
 146                 mutex_lock(&therm_cpu_lock);
 147                 err = thermal_throttle_add_dev(sys_dev);
 148                 mutex_unlock(&therm_cpu_lock);
 149                 WARN_ON(err);
 150                 break;
 151         case CPU_UP_CANCELED:
 152         case CPU_UP_CANCELED_FROZEN:
 153         case CPU_DEAD:
 154         case CPU_DEAD_FROZEN:
 155                 mutex_lock(&therm_cpu_lock);
 156                 thermal_throttle_remove_dev(sys_dev);
 157                 mutex_unlock(&therm_cpu_lock);
 158                 break;
 159         }
 160         return err ? NOTIFY_BAD : NOTIFY_OK;
 161 }
 162
 163 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
 164 {
 165         .notifier_call = thermal_throttle_cpu_callback,
 166 };
 167
 168 static __init int thermal_throttle_init_device(void)
 169 {
 170         unsigned int cpu = 0;
 171         int err;
 172
 173         if (!atomic_read(&therm_throt_en))
 174                 return 0;
 175
 176         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
 177
 178 #ifdef CONFIG_HOTPLUG_CPU
 179         mutex_lock(&therm_cpu_lock);
 180 #endif
 181         /* connect live CPUs to sysfs */
 182         for_each_online_cpu(cpu) {
 183                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
 184                 WARN_ON(err);
 185         }
 186 #ifdef CONFIG_HOTPLUG_CPU
 187         mutex_unlock(&therm_cpu_lock);
 188 #endif
 189
 190         return 0;
 191 }
 192 device_initcall(thermal_throttle_init_device);
 193
 194 #endif /* CONFIG_SYSFS */
 195
 196 /* Thermal transition interrupt handler */
 197 void intel_thermal_interrupt(void)
 198 {
 199         __u64 msr_val;
 200
 201         rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
 202         if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
 203                 mce_log_therm_throt_event(msr_val);
 204 }
 205
 206 static void unexpected_thermal_interrupt(void)
 207 {
 208         printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
 209                         smp_processor_id());
 210         add_taint(TAINT_MACHINE_CHECK);
 211 }
 212
 213 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
 214
 215 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
 216 {
 217         exit_idle();
 218         irq_enter();
 219         inc_irq_stat(irq_thermal_count);
 220         smp_thermal_vector();
 221         irq_exit();
 222         /* Ack only at the end to avoid potential reentry */
 223         ack_APIC_irq();
 224 }
 225
 226 void intel_set_thermal_handler(void)
 227 {
 228         smp_thermal_vector = intel_thermal_interrupt;
 229 }