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/therm_throt.h>
  28 #include <asm/processor.h>
  29 #include <asm/system.h>
  30 #include <asm/apic.h>
  31 #include <asm/idle.h>
  32 #include <asm/mce.h>
  33 #include <asm/msr.h>
  34
  35 /* How long to wait between reporting thermal events */
  36 #define CHECK_INTERVAL          (300 * HZ)
  37
  38 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
  39 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
  40
  41 atomic_t therm_throt_en         = ATOMIC_INIT(0);
  42
  43 #ifdef CONFIG_SYSFS
  44 #define define_therm_throt_sysdev_one_ro(_name)                         \
  45         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
  46
  47 #define define_therm_throt_sysdev_show_func(name)                       \
  48 static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,   \
  49                                         struct sysdev_attribute *attr,  \
  50                                               char *buf)                \
  51 {                                                                       \
  52         unsigned int cpu = dev->id;                                     \
  53         ssize_t ret;                                                    \
  54                                                                         \
  55         preempt_disable();      /* CPU hotplug */                       \
  56         if (cpu_online(cpu))                                            \
  57                 ret = sprintf(buf, "%lu\n",                             \
  58                               per_cpu(thermal_throttle_##name, cpu));   \
  59         else                                                            \
  60                 ret = 0;                                                \
  61         preempt_enable();                                               \
  62                                                                         \
  63         return ret;                                                     \
  64 }
  65
  66 define_therm_throt_sysdev_show_func(count);
  67 define_therm_throt_sysdev_one_ro(count);
  68
  69 static struct attribute *thermal_throttle_attrs[] = {
  70         &attr_count.attr,
  71         NULL
  72 };
  73
  74 static struct attribute_group thermal_throttle_attr_group = {
  75         .attrs  = thermal_throttle_attrs,
  76         .name   = "thermal_throttle"
  77 };
  78 #endif /* CONFIG_SYSFS */
  79
  80 /***
  81  * therm_throt_process - Process thermal throttling event from interrupt
  82  * @curr: Whether the condition is current or not (boolean), since the
  83  *        thermal interrupt normally gets called both when the thermal
  84  *        event begins and once the event has ended.
  85  *
  86  * This function is called by the thermal interrupt after the
  87  * IRQ has been acknowledged.
  88  *
  89  * It will take care of rate limiting and printing messages to the syslog.
  90  *
  91  * Returns: 0 : Event should NOT be further logged, i.e. still in
  92  *              "timeout" from previous log message.
  93  *          1 : Event should be logged further, and a message has been
  94  *              printed to the syslog.
  95  */
  96 int therm_throt_process(int curr)
  97 {
  98         unsigned int cpu = smp_processor_id();
  99         __u64 tmp_jiffs = get_jiffies_64();
 100
 101         if (curr)
 102                 __get_cpu_var(thermal_throttle_count)++;
 103
 104         if (time_before64(tmp_jiffs, __get_cpu_var(next_check)))
 105                 return 0;
 106
 107         __get_cpu_var(next_check) = tmp_jiffs + CHECK_INTERVAL;
 108
 109         /* if we just entered the thermal event */
 110         if (curr) {
 111                 printk(KERN_CRIT "CPU%d: Temperature above threshold, "
 112                        "cpu clock throttled (total events = %lu)\n", cpu,
 113                        __get_cpu_var(thermal_throttle_count));
 114
 115                 add_taint(TAINT_MACHINE_CHECK);
 116         } else {
 117                 printk(KERN_CRIT "CPU%d: Temperature/speed normal\n", cpu);
 118         }
 119
 120         return 1;
 121 }
 122
 123 #ifdef CONFIG_SYSFS
 124 /* Add/Remove thermal_throttle interface for CPU device: */
 125 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
 126 {
 127         return sysfs_create_group(&sys_dev->kobj,
 128                                   &thermal_throttle_attr_group);
 129 }
 130
 131 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
 132 {
 133         sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
 134 }
 135
 136 /* Mutex protecting device creation against CPU hotplug: */
 137 static DEFINE_MUTEX(therm_cpu_lock);
 138
 139 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
 140 static __cpuinit int
 141 thermal_throttle_cpu_callback(struct notifier_block *nfb,
 142                               unsigned long action,
 143                               void *hcpu)
 144 {
 145         unsigned int cpu = (unsigned long)hcpu;
 146         struct sys_device *sys_dev;
 147         int err = 0;
 148
 149         sys_dev = get_cpu_sysdev(cpu);
 150
 151         switch (action) {
 152         case CPU_UP_PREPARE:
 153         case CPU_UP_PREPARE_FROZEN:
 154                 mutex_lock(&therm_cpu_lock);
 155                 err = thermal_throttle_add_dev(sys_dev);
 156                 mutex_unlock(&therm_cpu_lock);
 157                 WARN_ON(err);
 158                 break;
 159         case CPU_UP_CANCELED:
 160         case CPU_UP_CANCELED_FROZEN:
 161         case CPU_DEAD:
 162         case CPU_DEAD_FROZEN:
 163                 mutex_lock(&therm_cpu_lock);
 164                 thermal_throttle_remove_dev(sys_dev);
 165                 mutex_unlock(&therm_cpu_lock);
 166                 break;
 167         }
 168         return err ? NOTIFY_BAD : NOTIFY_OK;
 169 }
 170
 171 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
 172 {
 173         .notifier_call = thermal_throttle_cpu_callback,
 174 };
 175
 176 static __init int thermal_throttle_init_device(void)
 177 {
 178         unsigned int cpu = 0;
 179         int err;
 180
 181         if (!atomic_read(&therm_throt_en))
 182                 return 0;
 183
 184         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
 185
 186 #ifdef CONFIG_HOTPLUG_CPU
 187         mutex_lock(&therm_cpu_lock);
 188 #endif
 189         /* connect live CPUs to sysfs */
 190         for_each_online_cpu(cpu) {
 191                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
 192                 WARN_ON(err);
 193         }
 194 #ifdef CONFIG_HOTPLUG_CPU
 195         mutex_unlock(&therm_cpu_lock);
 196 #endif
 197
 198         return 0;
 199 }
 200 device_initcall(thermal_throttle_init_device);
 201
 202 #endif /* CONFIG_SYSFS */
 203
 204 /* Thermal transition interrupt handler */
 205 static void intel_thermal_interrupt(void)
 206 {
 207         __u64 msr_val;
 208
 209         rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
 210         if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
 211                 mce_log_therm_throt_event(msr_val);
 212 }
 213
 214 static void unexpected_thermal_interrupt(void)
 215 {
 216         printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
 217                         smp_processor_id());
 218         add_taint(TAINT_MACHINE_CHECK);
 219 }
 220
 221 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
 222
 223 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
 224 {
 225         exit_idle();
 226         irq_enter();
 227         inc_irq_stat(irq_thermal_count);
 228         smp_thermal_vector();
 229         irq_exit();
 230         /* Ack only at the end to avoid potential reentry */
 231         ack_APIC_irq();
 232 }
 233
 234 void intel_init_thermal(struct cpuinfo_x86 *c)
 235 {
 236         unsigned int cpu = smp_processor_id();
 237         int tm2 = 0;
 238         u32 l, h;
 239
 240         /* Thermal monitoring depends on ACPI and clock modulation*/
 241         if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
 242                 return;
 243
 244         /*
 245          * First check if its enabled already, in which case there might
 246          * be some SMM goo which handles it, so we can't even put a handler
 247          * since it might be delivered via SMI already:
 248          */
 249         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 250         h = apic_read(APIC_LVTTHMR);
 251         if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
 252                 printk(KERN_DEBUG
 253                        "CPU%d: Thermal monitoring handled by SMI\n", cpu);
 254                 return;
 255         }
 256
 257         if (cpu_has(c, X86_FEATURE_TM2) && (l & MSR_IA32_MISC_ENABLE_TM2))
 258                 tm2 = 1;
 259
 260         /* Check whether a vector already exists */
 261         if (h & APIC_VECTOR_MASK) {
 262                 printk(KERN_DEBUG
 263                        "CPU%d: Thermal LVT vector (%#x) already installed\n",
 264                        cpu, (h & APIC_VECTOR_MASK));
 265                 return;
 266         }
 267
 268         /* We'll mask the thermal vector in the lapic till we're ready: */
 269         h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
 270         apic_write(APIC_LVTTHMR, h);
 271
 272         rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
 273         wrmsr(MSR_IA32_THERM_INTERRUPT,
 274                 l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
 275
 276         smp_thermal_vector = intel_thermal_interrupt;
 277
 278         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 279         wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
 280
 281         /* Unmask the thermal vector: */
 282         l = apic_read(APIC_LVTTHMR);
 283         apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
 284
 285         printk(KERN_INFO "CPU%d: Thermal monitoring enabled (%s)\n",
 286                cpu, tm2 ? "TM2" : "TM1");
 287
 288         /* enable thermal throttle processing */
 289         atomic_set(&therm_throt_en, 1);
 290 }