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 /*
  38  * Current thermal throttling state:
  39  */
  40 struct thermal_state {
  41         bool                    is_throttled;
  42
  43         u64                     next_check;
  44         unsigned long           throttle_count;
  45         unsigned long           last_throttle_count;
  46 };
  47
  48 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
  49
  50 static atomic_t therm_throt_en  = ATOMIC_INIT(0);
  51
  52 #ifdef CONFIG_SYSFS
  53 #define define_therm_throt_sysdev_one_ro(_name)                         \
  54         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
  55
  56 #define define_therm_throt_sysdev_show_func(name)                       \
  57                                                                         \
  58 static ssize_t therm_throt_sysdev_show_##name(                          \
  59                         struct sys_device *dev,                         \
  60                         struct sysdev_attribute *attr,                  \
  61                         char *buf)                                      \
  62 {                                                                       \
  63         unsigned int cpu = dev->id;                                     \
  64         ssize_t ret;                                                    \
  65                                                                         \
  66         preempt_disable();      /* CPU hotplug */                       \
  67         if (cpu_online(cpu))                                            \
  68                 ret = sprintf(buf, "%lu\n",                             \
  69                               per_cpu(thermal_state, cpu).name);        \
  70         else                                                            \
  71                 ret = 0;                                                \
  72         preempt_enable();                                               \
  73                                                                         \
  74         return ret;                                                     \
  75 }
  76
  77 define_therm_throt_sysdev_show_func(throttle_count);
  78 define_therm_throt_sysdev_one_ro(throttle_count);
  79
  80 static struct attribute *thermal_throttle_attrs[] = {
  81         &attr_throttle_count.attr,
  82         NULL
  83 };
  84
  85 static struct attribute_group thermal_throttle_attr_group = {
  86         .attrs  = thermal_throttle_attrs,
  87         .name   = "thermal_throttle"
  88 };
  89 #endif /* CONFIG_SYSFS */
  90
  91 /***
  92  * therm_throt_process - Process thermal throttling event from interrupt
  93  * @curr: Whether the condition is current or not (boolean), since the
  94  *        thermal interrupt normally gets called both when the thermal
  95  *        event begins and once the event has ended.
  96  *
  97  * This function is called by the thermal interrupt after the
  98  * IRQ has been acknowledged.
  99  *
 100  * It will take care of rate limiting and printing messages to the syslog.
 101  *
 102  * Returns: 0 : Event should NOT be further logged, i.e. still in
 103  *              "timeout" from previous log message.
 104  *          1 : Event should be logged further, and a message has been
 105  *              printed to the syslog.
 106  */
 107 static int therm_throt_process(bool is_throttled)
 108 {
 109         struct thermal_state *state;
 110         unsigned int this_cpu;
 111         bool was_throttled;
 112         u64 now;
 113
 114         this_cpu = smp_processor_id();
 115         now = get_jiffies_64();
 116         state = &per_cpu(thermal_state, this_cpu);
 117
 118         was_throttled = state->is_throttled;
 119         state->is_throttled = is_throttled;
 120
 121         if (is_throttled)
 122                 state->throttle_count++;
 123
 124         if (time_before64(now, state->next_check) &&
 125                         state->throttle_count != state->last_throttle_count)
 126                 return 0;
 127
 128         state->next_check = now + CHECK_INTERVAL;
 129         state->last_throttle_count = state->throttle_count;
 130
 131         /* if we just entered the thermal event */
 132         if (is_throttled) {
 133                 printk(KERN_CRIT "CPU%d: Temperature above threshold, cpu clock throttled (total events = %lu)\n", this_cpu, state->throttle_count);
 134
 135                 add_taint(TAINT_MACHINE_CHECK);
 136                 return 1;
 137         }
 138         if (was_throttled) {
 139                 printk(KERN_INFO "CPU%d: Temperature/speed normal\n", this_cpu);
 140                 return 1;
 141         }
 142
 143         return 0;
 144 }
 145
 146 #ifdef CONFIG_SYSFS
 147 /* Add/Remove thermal_throttle interface for CPU device: */
 148 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
 149 {
 150         return sysfs_create_group(&sys_dev->kobj,
 151                                   &thermal_throttle_attr_group);
 152 }
 153
 154 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
 155 {
 156         sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
 157 }
 158
 159 /* Mutex protecting device creation against CPU hotplug: */
 160 static DEFINE_MUTEX(therm_cpu_lock);
 161
 162 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
 163 static __cpuinit int
 164 thermal_throttle_cpu_callback(struct notifier_block *nfb,
 165                               unsigned long action,
 166                               void *hcpu)
 167 {
 168         unsigned int cpu = (unsigned long)hcpu;
 169         struct sys_device *sys_dev;
 170         int err = 0;
 171
 172         sys_dev = get_cpu_sysdev(cpu);
 173
 174         switch (action) {
 175         case CPU_UP_PREPARE:
 176         case CPU_UP_PREPARE_FROZEN:
 177                 mutex_lock(&therm_cpu_lock);
 178                 err = thermal_throttle_add_dev(sys_dev);
 179                 mutex_unlock(&therm_cpu_lock);
 180                 WARN_ON(err);
 181                 break;
 182         case CPU_UP_CANCELED:
 183         case CPU_UP_CANCELED_FROZEN:
 184         case CPU_DEAD:
 185         case CPU_DEAD_FROZEN:
 186                 mutex_lock(&therm_cpu_lock);
 187                 thermal_throttle_remove_dev(sys_dev);
 188                 mutex_unlock(&therm_cpu_lock);
 189                 break;
 190         }
 191         return err ? NOTIFY_BAD : NOTIFY_OK;
 192 }
 193
 194 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
 195 {
 196         .notifier_call = thermal_throttle_cpu_callback,
 197 };
 198
 199 static __init int thermal_throttle_init_device(void)
 200 {
 201         unsigned int cpu = 0;
 202         int err;
 203
 204         if (!atomic_read(&therm_throt_en))
 205                 return 0;
 206
 207         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
 208
 209 #ifdef CONFIG_HOTPLUG_CPU
 210         mutex_lock(&therm_cpu_lock);
 211 #endif
 212         /* connect live CPUs to sysfs */
 213         for_each_online_cpu(cpu) {
 214                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
 215                 WARN_ON(err);
 216         }
 217 #ifdef CONFIG_HOTPLUG_CPU
 218         mutex_unlock(&therm_cpu_lock);
 219 #endif
 220
 221         return 0;
 222 }
 223 device_initcall(thermal_throttle_init_device);
 224
 225 #endif /* CONFIG_SYSFS */
 226
 227 /* Thermal transition interrupt handler */
 228 static void intel_thermal_interrupt(void)
 229 {
 230         __u64 msr_val;
 231
 232         rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
 233         if (therm_throt_process((msr_val & THERM_STATUS_PROCHOT) != 0))
 234                 mce_log_therm_throt_event(msr_val);
 235 }
 236
 237 static void unexpected_thermal_interrupt(void)
 238 {
 239         printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
 240                         smp_processor_id());
 241         add_taint(TAINT_MACHINE_CHECK);
 242 }
 243
 244 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
 245
 246 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
 247 {
 248         exit_idle();
 249         irq_enter();
 250         inc_irq_stat(irq_thermal_count);
 251         smp_thermal_vector();
 252         irq_exit();
 253         /* Ack only at the end to avoid potential reentry */
 254         ack_APIC_irq();
 255 }
 256
 257 void intel_init_thermal(struct cpuinfo_x86 *c)
 258 {
 259         unsigned int cpu = smp_processor_id();
 260         int tm2 = 0;
 261         u32 l, h;
 262
 263         /* Thermal monitoring depends on ACPI and clock modulation*/
 264         if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
 265                 return;
 266
 267         /*
 268          * First check if its enabled already, in which case there might
 269          * be some SMM goo which handles it, so we can't even put a handler
 270          * since it might be delivered via SMI already:
 271          */
 272         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 273         h = apic_read(APIC_LVTTHMR);
 274         if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
 275                 printk(KERN_DEBUG
 276                        "CPU%d: Thermal monitoring handled by SMI\n", cpu);
 277                 return;
 278         }
 279
 280         /* Check whether a vector already exists */
 281         if (h & APIC_VECTOR_MASK) {
 282                 printk(KERN_DEBUG
 283                        "CPU%d: Thermal LVT vector (%#x) already installed\n",
 284                        cpu, (h & APIC_VECTOR_MASK));
 285                 return;
 286         }
 287
 288         /* early Pentium M models use different method for enabling TM2 */
 289         if (cpu_has(c, X86_FEATURE_TM2)) {
 290                 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
 291                         rdmsr(MSR_THERM2_CTL, l, h);
 292                         if (l & MSR_THERM2_CTL_TM_SELECT)
 293                                 tm2 = 1;
 294                 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
 295                         tm2 = 1;
 296         }
 297
 298         /* We'll mask the thermal vector in the lapic till we're ready: */
 299         h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
 300         apic_write(APIC_LVTTHMR, h);
 301
 302         rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
 303         wrmsr(MSR_IA32_THERM_INTERRUPT,
 304                 l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
 305
 306         smp_thermal_vector = intel_thermal_interrupt;
 307
 308         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 309         wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
 310
 311         /* Unmask the thermal vector: */
 312         l = apic_read(APIC_LVTTHMR);
 313         apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
 314
 315         printk(KERN_INFO "CPU%d: Thermal monitoring enabled (%s)\n",
 316                cpu, tm2 ? "TM2" : "TM1");
 317
 318         /* enable thermal throttle processing */
 319         atomic_set(&therm_throt_en, 1);
 320 }