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/kernel.h>
  20 #include <linux/percpu.h>
  21 #include <linux/sysdev.h>
  22 #include <linux/types.h>
  23 #include <linux/init.h>
  24 #include <linux/smp.h>
  25 #include <linux/cpu.h>
  26
  27 #include <asm/processor.h>
  28 #include <asm/system.h>
  29 #include <asm/apic.h>
  30 #include <asm/idle.h>
  31 #include <asm/mce.h>
  32 #include <asm/msr.h>
  33
  34 /* How long to wait between reporting thermal events */
  35 #define CHECK_INTERVAL          (300 * HZ)
  36
  37 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
  38 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
  39
  40 static atomic_t therm_throt_en          = ATOMIC_INIT(0);
  41
  42 #ifdef CONFIG_SYSFS
  43 #define define_therm_throt_sysdev_one_ro(_name)                         \
  44         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
  45
  46 #define define_therm_throt_sysdev_show_func(name)                       \
  47 static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,   \
  48                                         struct sysdev_attribute *attr,  \
  49                                               char *buf)                \
  50 {                                                                       \
  51         unsigned int cpu = dev->id;                                     \
  52         ssize_t ret;                                                    \
  53                                                                         \
  54         preempt_disable();      /* CPU hotplug */                       \
  55         if (cpu_online(cpu))                                            \
  56                 ret = sprintf(buf, "%lu\n",                             \
  57                               per_cpu(thermal_throttle_##name, cpu));   \
  58         else                                                            \
  59                 ret = 0;                                                \
  60         preempt_enable();                                               \
  61                                                                         \
  62         return ret;                                                     \
  63 }
  64
  65 define_therm_throt_sysdev_show_func(count);
  66 define_therm_throt_sysdev_one_ro(count);
  67
  68 static struct attribute *thermal_throttle_attrs[] = {
  69         &attr_count.attr,
  70         NULL
  71 };
  72
  73 static struct attribute_group thermal_throttle_attr_group = {
  74         .attrs  = thermal_throttle_attrs,
  75         .name   = "thermal_throttle"
  76 };
  77 #endif /* CONFIG_SYSFS */
  78
  79 /***
  80  * therm_throt_process - Process thermal throttling event from interrupt
  81  * @curr: Whether the condition is current or not (boolean), since the
  82  *        thermal interrupt normally gets called both when the thermal
  83  *        event begins and once the event has ended.
  84  *
  85  * This function is called by the thermal interrupt after the
  86  * IRQ has been acknowledged.
  87  *
  88  * It will take care of rate limiting and printing messages to the syslog.
  89  *
  90  * Returns: 0 : Event should NOT be further logged, i.e. still in
  91  *              "timeout" from previous log message.
  92  *          1 : Event should be logged further, and a message has been
  93  *              printed to the syslog.
  94  */
  95 static int therm_throt_process(int curr)
  96 {
  97         unsigned int cpu = smp_processor_id();
  98         __u64 tmp_jiffs = get_jiffies_64();
  99
 100         if (curr)
 101                 __get_cpu_var(thermal_throttle_count)++;
 102
 103         if (time_before64(tmp_jiffs, __get_cpu_var(next_check)))
 104                 return 0;
 105
 106         __get_cpu_var(next_check) = tmp_jiffs + CHECK_INTERVAL;
 107
 108         /* if we just entered the thermal event */
 109         if (curr) {
 110                 printk(KERN_CRIT "CPU%d: Temperature above threshold, "
 111                        "cpu clock throttled (total events = %lu)\n", cpu,
 112                        __get_cpu_var(thermal_throttle_count));
 113
 114                 add_taint(TAINT_MACHINE_CHECK);
 115         } else {
 116                 printk(KERN_CRIT "CPU%d: Temperature/speed normal\n", cpu);
 117         }
 118
 119         return 1;
 120 }
 121
 122 #ifdef CONFIG_SYSFS
 123 /* Add/Remove thermal_throttle interface for CPU device: */
 124 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
 125 {
 126         return sysfs_create_group(&sys_dev->kobj,
 127                                   &thermal_throttle_attr_group);
 128 }
 129
 130 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
 131 {
 132         sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
 133 }
 134
 135 /* Mutex protecting device creation against CPU hotplug: */
 136 static DEFINE_MUTEX(therm_cpu_lock);
 137
 138 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
 139 static __cpuinit int
 140 thermal_throttle_cpu_callback(struct notifier_block *nfb,
 141                               unsigned long action,
 142                               void *hcpu)
 143 {
 144         unsigned int cpu = (unsigned long)hcpu;
 145         struct sys_device *sys_dev;
 146         int err = 0;
 147
 148         sys_dev = get_cpu_sysdev(cpu);
 149
 150         switch (action) {
 151         case CPU_UP_PREPARE:
 152         case CPU_UP_PREPARE_FROZEN:
 153                 mutex_lock(&therm_cpu_lock);
 154                 err = thermal_throttle_add_dev(sys_dev);
 155                 mutex_unlock(&therm_cpu_lock);
 156                 WARN_ON(err);
 157                 break;
 158         case CPU_UP_CANCELED:
 159         case CPU_UP_CANCELED_FROZEN:
 160         case CPU_DEAD:
 161         case CPU_DEAD_FROZEN:
 162                 mutex_lock(&therm_cpu_lock);
 163                 thermal_throttle_remove_dev(sys_dev);
 164                 mutex_unlock(&therm_cpu_lock);
 165                 break;
 166         }
 167         return err ? NOTIFY_BAD : NOTIFY_OK;
 168 }
 169
 170 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
 171 {
 172         .notifier_call = thermal_throttle_cpu_callback,
 173 };
 174
 175 static __init int thermal_throttle_init_device(void)
 176 {
 177         unsigned int cpu = 0;
 178         int err;
 179
 180         if (!atomic_read(&therm_throt_en))
 181                 return 0;
 182
 183         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
 184
 185 #ifdef CONFIG_HOTPLUG_CPU
 186         mutex_lock(&therm_cpu_lock);
 187 #endif
 188         /* connect live CPUs to sysfs */
 189         for_each_online_cpu(cpu) {
 190                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
 191                 WARN_ON(err);
 192         }
 193 #ifdef CONFIG_HOTPLUG_CPU
 194         mutex_unlock(&therm_cpu_lock);
 195 #endif
 196
 197         return 0;
 198 }
 199 device_initcall(thermal_throttle_init_device);
 200
 201 #endif /* CONFIG_SYSFS */
 202
 203 /* Thermal transition interrupt handler */
 204 static void intel_thermal_interrupt(void)
 205 {
 206         __u64 msr_val;
 207
 208         rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
 209         if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
 210                 mce_log_therm_throt_event(msr_val);
 211 }
 212
 213 static void unexpected_thermal_interrupt(void)
 214 {
 215         printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
 216                         smp_processor_id());
 217         add_taint(TAINT_MACHINE_CHECK);
 218 }
 219
 220 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
 221
 222 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
 223 {
 224         exit_idle();
 225         irq_enter();
 226         inc_irq_stat(irq_thermal_count);
 227         smp_thermal_vector();
 228         irq_exit();
 229         /* Ack only at the end to avoid potential reentry */
 230         ack_APIC_irq();
 231 }
 232
 233 void intel_init_thermal(struct cpuinfo_x86 *c)
 234 {
 235         unsigned int cpu = smp_processor_id();
 236         int tm2 = 0;
 237         u32 l, h;
 238
 239         /* Thermal monitoring depends on ACPI and clock modulation*/
 240         if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
 241                 return;
 242
 243         /*
 244          * First check if its enabled already, in which case there might
 245          * be some SMM goo which handles it, so we can't even put a handler
 246          * since it might be delivered via SMI already:
 247          */
 248         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 249         h = apic_read(APIC_LVTTHMR);
 250         if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
 251                 printk(KERN_DEBUG
 252                        "CPU%d: Thermal monitoring handled by SMI\n", cpu);
 253                 return;
 254         }
 255
 256         if (cpu_has(c, X86_FEATURE_TM2) && (l & MSR_IA32_MISC_ENABLE_TM2))
 257                 tm2 = 1;
 258
 259         /* Check whether a vector already exists */
 260         if (h & APIC_VECTOR_MASK) {
 261                 printk(KERN_DEBUG
 262                        "CPU%d: Thermal LVT vector (%#x) already installed\n",
 263                        cpu, (h & APIC_VECTOR_MASK));
 264                 return;
 265         }
 266
 267         /* We'll mask the thermal vector in the lapic till we're ready: */
 268         h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
 269         apic_write(APIC_LVTTHMR, h);
 270
 271         rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
 272         wrmsr(MSR_IA32_THERM_INTERRUPT,
 273                 l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
 274
 275         smp_thermal_vector = intel_thermal_interrupt;
 276
 277         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 278         wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
 279
 280         /* Unmask the thermal vector: */
 281         l = apic_read(APIC_LVTTHMR);
 282         apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
 283
 284         printk(KERN_INFO "CPU%d: Thermal monitoring enabled (%s)\n",
 285                cpu, tm2 ? "TM2" : "TM1");
 286
 287         /* enable thermal throttle processing */
 288         atomic_set(&therm_throt_en, 1);
 289 }