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

   1 /*
   2  *
   3  * Thermal throttle event support code (such as syslog messaging and rate
   4  * limiting) that was factored out from x86_64 (mce_intel.c) and i386 (p4.c).
   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
  17 #include <linux/percpu.h>
  18 #include <linux/sysdev.h>
  19 #include <linux/cpu.h>
  20 #include <asm/cpu.h>
  21 #include <linux/notifier.h>
  22 #include <linux/jiffies.h>
  23 #include <asm/therm_throt.h>
  24
  25 /* How long to wait between reporting thermal events */
  26 #define CHECK_INTERVAL              (300 * HZ)
  27
  28 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
  29 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
  30 atomic_t therm_throt_en = ATOMIC_INIT(0);
  31
  32 #ifdef CONFIG_SYSFS
  33 #define define_therm_throt_sysdev_one_ro(_name)                              \
  34         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
  35
  36 #define define_therm_throt_sysdev_show_func(name)                            \
  37 static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,        \
  38                                               char *buf)                     \
  39 {                                                                            \
  40         unsigned int cpu = dev->id;                                          \
  41         ssize_t ret;                                                         \
  42                                                                              \
  43         preempt_disable();              /* CPU hotplug */                    \
  44         if (cpu_online(cpu))                                                 \
  45                 ret = sprintf(buf, "%lu\n",                                  \
  46                               per_cpu(thermal_throttle_##name, cpu));        \
  47         else                                                                 \
  48                 ret = 0;                                                     \
  49         preempt_enable();                                                    \
  50                                                                              \
  51         return ret;                                                          \
  52 }
  53
  54 define_therm_throt_sysdev_show_func(count);
  55 define_therm_throt_sysdev_one_ro(count);
  56
  57 static struct attribute *thermal_throttle_attrs[] = {
  58         &attr_count.attr,
  59         NULL
  60 };
  61
  62 static struct attribute_group thermal_throttle_attr_group = {
  63         .attrs = thermal_throttle_attrs,
  64         .name = "thermal_throttle"
  65 };
  66 #endif /* CONFIG_SYSFS */
  67
  68 /***
  69  * therm_throt_process - Process thermal throttling event from interrupt
  70  * @curr: Whether the condition is current or not (boolean), since the
  71  *        thermal interrupt normally gets called both when the thermal
  72  *        event begins and once the event has ended.
  73  *
  74  * This function is called by the thermal interrupt after the
  75  * IRQ has been acknowledged.
  76  *
  77  * It will take care of rate limiting and printing messages to the syslog.
  78  *
  79  * Returns: 0 : Event should NOT be further logged, i.e. still in
  80  *              "timeout" from previous log message.
  81  *          1 : Event should be logged further, and a message has been
  82  *              printed to the syslog.
  83  */
  84 int therm_throt_process(int curr)
  85 {
  86         unsigned int cpu = smp_processor_id();
  87         __u64 tmp_jiffs = get_jiffies_64();
  88
  89         if (curr)
  90                 __get_cpu_var(thermal_throttle_count)++;
  91
  92         if (time_before64(tmp_jiffs, __get_cpu_var(next_check)))
  93                 return 0;
  94
  95         __get_cpu_var(next_check) = tmp_jiffs + CHECK_INTERVAL;
  96
  97         /* if we just entered the thermal event */
  98         if (curr) {
  99                 printk(KERN_CRIT "CPU%d: Temperature above threshold, "
 100                        "cpu clock throttled (total events = %lu)\n", cpu,
 101                        __get_cpu_var(thermal_throttle_count));
 102
 103                 add_taint(TAINT_MACHINE_CHECK);
 104         } else {
 105                 printk(KERN_CRIT "CPU%d: Temperature/speed normal\n", cpu);
 106         }
 107
 108         return 1;
 109 }
 110
 111 #ifdef CONFIG_SYSFS
 112 /* Add/Remove thermal_throttle interface for CPU device */
 113 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
 114 {
 115         return sysfs_create_group(&sys_dev->kobj, &thermal_throttle_attr_group);
 116 }
 117
 118 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
 119 {
 120         sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
 121 }
 122
 123 /* Mutex protecting device creation against CPU hotplug */
 124 static DEFINE_MUTEX(therm_cpu_lock);
 125
 126 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
 127 static __cpuinit int thermal_throttle_cpu_callback(struct notifier_block *nfb,
 128                                                    unsigned long action,
 129                                                    void *hcpu)
 130 {
 131         unsigned int cpu = (unsigned long)hcpu;
 132         struct sys_device *sys_dev;
 133         int err = 0;
 134
 135         sys_dev = get_cpu_sysdev(cpu);
 136         switch (action) {
 137         case CPU_UP_PREPARE:
 138         case CPU_UP_PREPARE_FROZEN:
 139                 mutex_lock(&therm_cpu_lock);
 140                 err = thermal_throttle_add_dev(sys_dev);
 141                 mutex_unlock(&therm_cpu_lock);
 142                 WARN_ON(err);
 143                 break;
 144         case CPU_UP_CANCELED:
 145         case CPU_UP_CANCELED_FROZEN:
 146         case CPU_DEAD:
 147         case CPU_DEAD_FROZEN:
 148                 mutex_lock(&therm_cpu_lock);
 149                 thermal_throttle_remove_dev(sys_dev);
 150                 mutex_unlock(&therm_cpu_lock);
 151                 break;
 152         }
 153         return err ? NOTIFY_BAD : NOTIFY_OK;
 154 }
 155
 156 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
 157 {
 158         .notifier_call = thermal_throttle_cpu_callback,
 159 };
 160
 161 static __init int thermal_throttle_init_device(void)
 162 {
 163         unsigned int cpu = 0;
 164         int err;
 165
 166         if (!atomic_read(&therm_throt_en))
 167                 return 0;
 168
 169         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
 170
 171 #ifdef CONFIG_HOTPLUG_CPU
 172         mutex_lock(&therm_cpu_lock);
 173 #endif
 174         /* connect live CPUs to sysfs */
 175         for_each_online_cpu(cpu) {
 176                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
 177                 WARN_ON(err);
 178         }
 179 #ifdef CONFIG_HOTPLUG_CPU
 180         mutex_unlock(&therm_cpu_lock);
 181 #endif
 182
 183         return 0;
 184 }
 185
 186 device_initcall(thermal_throttle_init_device);
 187 #endif /* CONFIG_SYSFS */