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 #define THERMAL_THROTTLING_EVENT        0
  38 #define POWER_LIMIT_EVENT               1
  39
  40 /*
  41  * Current thermal event state:
  42  */
  43 struct _thermal_state {
  44         bool                    new_event;
  45         int                     event;
  46         u64                     next_check;
  47         unsigned long           count;
  48         unsigned long           last_count;
  49 };
  50
  51 struct thermal_state {
  52         struct _thermal_state core_throttle;
  53         struct _thermal_state core_power_limit;
  54         struct _thermal_state package_throttle;
  55         struct _thermal_state package_power_limit;
  56 };
  57
  58 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
  59
  60 static atomic_t therm_throt_en  = ATOMIC_INIT(0);
  61
  62 static u32 lvtthmr_init __read_mostly;
  63
  64 #ifdef CONFIG_SYSFS
  65 #define define_therm_throt_sysdev_one_ro(_name)                         \
  66         static SYSDEV_ATTR(_name, 0444,                                 \
  67                            therm_throt_sysdev_show_##_name,             \
  68                                    NULL)                                \
  69
  70 #define define_therm_throt_sysdev_show_func(event, name)                \
  71                                                                         \
  72 static ssize_t therm_throt_sysdev_show_##event##_##name(                \
  73                         struct sys_device *dev,                         \
  74                         struct sysdev_attribute *attr,                  \
  75                         char *buf)                                      \
  76 {                                                                       \
  77         unsigned int cpu = dev->id;                                     \
  78         ssize_t ret;                                                    \
  79                                                                         \
  80         preempt_disable();      /* CPU hotplug */                       \
  81         if (cpu_online(cpu)) {                                          \
  82                 ret = sprintf(buf, "%lu\n",                             \
  83                               per_cpu(thermal_state, cpu).event.name);  \
  84         } else                                                          \
  85                 ret = 0;                                                \
  86         preempt_enable();                                               \
  87                                                                         \
  88         return ret;                                                     \
  89 }
  90
  91 define_therm_throt_sysdev_show_func(core_throttle, count);
  92 define_therm_throt_sysdev_one_ro(core_throttle_count);
  93
  94 define_therm_throt_sysdev_show_func(core_power_limit, count);
  95 define_therm_throt_sysdev_one_ro(core_power_limit_count);
  96
  97 define_therm_throt_sysdev_show_func(package_throttle, count);
  98 define_therm_throt_sysdev_one_ro(package_throttle_count);
  99
 100 define_therm_throt_sysdev_show_func(package_power_limit, count);
 101 define_therm_throt_sysdev_one_ro(package_power_limit_count);
 102
 103 static struct attribute *thermal_throttle_attrs[] = {
 104         &attr_core_throttle_count.attr,
 105         NULL
 106 };
 107
 108 static struct attribute_group thermal_attr_group = {
 109         .attrs  = thermal_throttle_attrs,
 110         .name   = "thermal_throttle"
 111 };
 112 #endif /* CONFIG_SYSFS */
 113
 114 #define CORE_LEVEL      0
 115 #define PACKAGE_LEVEL   1
 116
 117 /***
 118  * therm_throt_process - Process thermal throttling event from interrupt
 119  * @curr: Whether the condition is current or not (boolean), since the
 120  *        thermal interrupt normally gets called both when the thermal
 121  *        event begins and once the event has ended.
 122  *
 123  * This function is called by the thermal interrupt after the
 124  * IRQ has been acknowledged.
 125  *
 126  * It will take care of rate limiting and printing messages to the syslog.
 127  *
 128  * Returns: 0 : Event should NOT be further logged, i.e. still in
 129  *              "timeout" from previous log message.
 130  *          1 : Event should be logged further, and a message has been
 131  *              printed to the syslog.
 132  */
 133 static int therm_throt_process(bool new_event, int event, int level)
 134 {
 135         struct _thermal_state *state;
 136         unsigned int this_cpu = smp_processor_id();
 137         bool old_event;
 138         u64 now;
 139         struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
 140
 141         now = get_jiffies_64();
 142         if (level == CORE_LEVEL) {
 143                 if (event == THERMAL_THROTTLING_EVENT)
 144                         state = &pstate->core_throttle;
 145                 else if (event == POWER_LIMIT_EVENT)
 146                         state = &pstate->core_power_limit;
 147                 else
 148                          return 0;
 149         } else if (level == PACKAGE_LEVEL) {
 150                 if (event == THERMAL_THROTTLING_EVENT)
 151                         state = &pstate->package_throttle;
 152                 else if (event == POWER_LIMIT_EVENT)
 153                         state = &pstate->package_power_limit;
 154                 else
 155                         return 0;
 156         } else
 157                 return 0;
 158
 159         old_event = state->new_event;
 160         state->new_event = new_event;
 161
 162         if (new_event)
 163                 state->count++;
 164
 165         if (time_before64(now, state->next_check) &&
 166                         state->count != state->last_count)
 167                 return 0;
 168
 169         state->next_check = now + CHECK_INTERVAL;
 170         state->last_count = state->count;
 171
 172         /* if we just entered the thermal event */
 173         if (new_event) {
 174                 if (event == THERMAL_THROTTLING_EVENT)
 175                         printk(KERN_CRIT "CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",
 176                                 this_cpu,
 177                                 level == CORE_LEVEL ? "Core" : "Package",
 178                                 state->count);
 179                 else
 180                         printk(KERN_CRIT "CPU%d: %s power limit notification (total events = %lu)\n",
 181                                 this_cpu,
 182                                 level == CORE_LEVEL ? "Core" : "Package",
 183                                 state->count);
 184
 185                 add_taint(TAINT_MACHINE_CHECK);
 186                 return 1;
 187         }
 188         if (old_event) {
 189                 if (event == THERMAL_THROTTLING_EVENT)
 190                         printk(KERN_INFO "CPU%d: %s temperature/speed normal\n",
 191                                 this_cpu,
 192                                 level == CORE_LEVEL ? "Core" : "Package");
 193                 else
 194                         printk(KERN_INFO "CPU%d: %s power limit normal\n",
 195                                 this_cpu,
 196                                 level == CORE_LEVEL ? "Core" : "Package");
 197                 return 1;
 198         }
 199
 200         return 0;
 201 }
 202
 203 #ifdef CONFIG_SYSFS
 204 /* Add/Remove thermal_throttle interface for CPU device: */
 205 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
 206 {
 207         int err;
 208         struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
 209
 210         err = sysfs_create_group(&sys_dev->kobj, &thermal_attr_group);
 211         if (err)
 212                 return err;
 213
 214         if (cpu_has(c, X86_FEATURE_PLN))
 215                 err = sysfs_add_file_to_group(&sys_dev->kobj,
 216                                               &attr_core_power_limit_count.attr,
 217                                               thermal_attr_group.name);
 218         if (cpu_has(c, X86_FEATURE_PTS))
 219                 err = sysfs_add_file_to_group(&sys_dev->kobj,
 220                                               &attr_package_throttle_count.attr,
 221                                               thermal_attr_group.name);
 222                 if (cpu_has(c, X86_FEATURE_PLN))
 223                         err = sysfs_add_file_to_group(&sys_dev->kobj,
 224                                         &attr_package_power_limit_count.attr,
 225                                         thermal_attr_group.name);
 226
 227         return err;
 228 }
 229
 230 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
 231 {
 232         sysfs_remove_group(&sys_dev->kobj, &thermal_attr_group);
 233 }
 234
 235 /* Mutex protecting device creation against CPU hotplug: */
 236 static DEFINE_MUTEX(therm_cpu_lock);
 237
 238 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
 239 static __cpuinit int
 240 thermal_throttle_cpu_callback(struct notifier_block *nfb,
 241                               unsigned long action,
 242                               void *hcpu)
 243 {
 244         unsigned int cpu = (unsigned long)hcpu;
 245         struct sys_device *sys_dev;
 246         int err = 0;
 247
 248         sys_dev = get_cpu_sysdev(cpu);
 249
 250         switch (action) {
 251         case CPU_UP_PREPARE:
 252         case CPU_UP_PREPARE_FROZEN:
 253                 mutex_lock(&therm_cpu_lock);
 254                 err = thermal_throttle_add_dev(sys_dev);
 255                 mutex_unlock(&therm_cpu_lock);
 256                 WARN_ON(err);
 257                 break;
 258         case CPU_UP_CANCELED:
 259         case CPU_UP_CANCELED_FROZEN:
 260         case CPU_DEAD:
 261         case CPU_DEAD_FROZEN:
 262                 mutex_lock(&therm_cpu_lock);
 263                 thermal_throttle_remove_dev(sys_dev);
 264                 mutex_unlock(&therm_cpu_lock);
 265                 break;
 266         }
 267         return notifier_from_errno(err);
 268 }
 269
 270 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
 271 {
 272         .notifier_call = thermal_throttle_cpu_callback,
 273 };
 274
 275 static __init int thermal_throttle_init_device(void)
 276 {
 277         unsigned int cpu = 0;
 278         int err;
 279
 280         if (!atomic_read(&therm_throt_en))
 281                 return 0;
 282
 283         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
 284
 285 #ifdef CONFIG_HOTPLUG_CPU
 286         mutex_lock(&therm_cpu_lock);
 287 #endif
 288         /* connect live CPUs to sysfs */
 289         for_each_online_cpu(cpu) {
 290                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
 291                 WARN_ON(err);
 292         }
 293 #ifdef CONFIG_HOTPLUG_CPU
 294         mutex_unlock(&therm_cpu_lock);
 295 #endif
 296
 297         return 0;
 298 }
 299 device_initcall(thermal_throttle_init_device);
 300
 301 #endif /* CONFIG_SYSFS */
 302
 303 /*
 304  * Set up the most two significant bit to notify mce log that this thermal
 305  * event type.
 306  * This is a temp solution. May be changed in the future with mce log
 307  * infrasture.
 308  */
 309 #define CORE_THROTTLED          (0)
 310 #define CORE_POWER_LIMIT        ((__u64)1 << 62)
 311 #define PACKAGE_THROTTLED       ((__u64)2 << 62)
 312 #define PACKAGE_POWER_LIMIT     ((__u64)3 << 62)
 313
 314 /* Thermal transition interrupt handler */
 315 static void intel_thermal_interrupt(void)
 316 {
 317         __u64 msr_val;
 318         struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
 319
 320         rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
 321
 322         if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
 323                                 THERMAL_THROTTLING_EVENT,
 324                                 CORE_LEVEL) != 0)
 325                 mce_log_therm_throt_event(CORE_THROTTLED | msr_val);
 326
 327         if (cpu_has(c, X86_FEATURE_PLN))
 328                 if (therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
 329                                         POWER_LIMIT_EVENT,
 330                                         CORE_LEVEL) != 0)
 331                         mce_log_therm_throt_event(CORE_POWER_LIMIT | msr_val);
 332
 333         if (cpu_has(c, X86_FEATURE_PTS)) {
 334                 rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
 335                 if (therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
 336                                         THERMAL_THROTTLING_EVENT,
 337                                         PACKAGE_LEVEL) != 0)
 338                         mce_log_therm_throt_event(PACKAGE_THROTTLED | msr_val);
 339                 if (cpu_has(c, X86_FEATURE_PLN))
 340                         if (therm_throt_process(msr_val &
 341                                         PACKAGE_THERM_STATUS_POWER_LIMIT,
 342                                         POWER_LIMIT_EVENT,
 343                                         PACKAGE_LEVEL) != 0)
 344                                 mce_log_therm_throt_event(PACKAGE_POWER_LIMIT
 345                                                           | msr_val);
 346         }
 347 }
 348
 349 static void unexpected_thermal_interrupt(void)
 350 {
 351         printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
 352                         smp_processor_id());
 353         add_taint(TAINT_MACHINE_CHECK);
 354 }
 355
 356 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
 357
 358 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
 359 {
 360         exit_idle();
 361         irq_enter();
 362         inc_irq_stat(irq_thermal_count);
 363         smp_thermal_vector();
 364         irq_exit();
 365         /* Ack only at the end to avoid potential reentry */
 366         ack_APIC_irq();
 367 }
 368
 369 /* Thermal monitoring depends on APIC, ACPI and clock modulation */
 370 static int intel_thermal_supported(struct cpuinfo_x86 *c)
 371 {
 372         if (!cpu_has_apic)
 373                 return 0;
 374         if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
 375                 return 0;
 376         return 1;
 377 }
 378
 379 void __init mcheck_intel_therm_init(void)
 380 {
 381         /*
 382          * This function is only called on boot CPU. Save the init thermal
 383          * LVT value on BSP and use that value to restore APs' thermal LVT
 384          * entry BIOS programmed later
 385          */
 386         if (intel_thermal_supported(&boot_cpu_data))
 387                 lvtthmr_init = apic_read(APIC_LVTTHMR);
 388 }
 389
 390 void intel_init_thermal(struct cpuinfo_x86 *c)
 391 {
 392         unsigned int cpu = smp_processor_id();
 393         int tm2 = 0;
 394         u32 l, h;
 395
 396         if (!intel_thermal_supported(c))
 397                 return;
 398
 399         /*
 400          * First check if its enabled already, in which case there might
 401          * be some SMM goo which handles it, so we can't even put a handler
 402          * since it might be delivered via SMI already:
 403          */
 404         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 405
 406         /*
 407          * The initial value of thermal LVT entries on all APs always reads
 408          * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
 409          * sequence to them and LVT registers are reset to 0s except for
 410          * the mask bits which are set to 1s when APs receive INIT IPI.
 411          * Always restore the value that BIOS has programmed on AP based on
 412          * BSP's info we saved since BIOS is always setting the same value
 413          * for all threads/cores
 414          */
 415         apic_write(APIC_LVTTHMR, lvtthmr_init);
 416
 417         h = lvtthmr_init;
 418
 419         if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
 420                 printk(KERN_DEBUG
 421                        "CPU%d: Thermal monitoring handled by SMI\n", cpu);
 422                 return;
 423         }
 424
 425         /* Check whether a vector already exists */
 426         if (h & APIC_VECTOR_MASK) {
 427                 printk(KERN_DEBUG
 428                        "CPU%d: Thermal LVT vector (%#x) already installed\n",
 429                        cpu, (h & APIC_VECTOR_MASK));
 430                 return;
 431         }
 432
 433         /* early Pentium M models use different method for enabling TM2 */
 434         if (cpu_has(c, X86_FEATURE_TM2)) {
 435                 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
 436                         rdmsr(MSR_THERM2_CTL, l, h);
 437                         if (l & MSR_THERM2_CTL_TM_SELECT)
 438                                 tm2 = 1;
 439                 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
 440                         tm2 = 1;
 441         }
 442
 443         /* We'll mask the thermal vector in the lapic till we're ready: */
 444         h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
 445         apic_write(APIC_LVTTHMR, h);
 446
 447         rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
 448         if (cpu_has(c, X86_FEATURE_PLN))
 449                 wrmsr(MSR_IA32_THERM_INTERRUPT,
 450                       l | (THERM_INT_LOW_ENABLE
 451                         | THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
 452         else
 453                 wrmsr(MSR_IA32_THERM_INTERRUPT,
 454                       l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
 455
 456         if (cpu_has(c, X86_FEATURE_PTS)) {
 457                 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
 458                 if (cpu_has(c, X86_FEATURE_PLN))
 459                         wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
 460                               l | (PACKAGE_THERM_INT_LOW_ENABLE
 461                                 | PACKAGE_THERM_INT_HIGH_ENABLE
 462                                 | PACKAGE_THERM_INT_PLN_ENABLE), h);
 463                 else
 464                         wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
 465                               l | (PACKAGE_THERM_INT_LOW_ENABLE
 466                                 | PACKAGE_THERM_INT_HIGH_ENABLE), h);
 467         }
 468
 469         smp_thermal_vector = intel_thermal_interrupt;
 470
 471         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
 472         wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
 473
 474         /* Unmask the thermal vector: */
 475         l = apic_read(APIC_LVTTHMR);
 476         apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
 477
 478         printk_once(KERN_INFO "CPU0: Thermal monitoring enabled (%s)\n",
 479                        tm2 ? "TM2" : "TM1");
 480
 481         /* enable thermal throttle processing */
 482         atomic_set(&therm_throt_en, 1);
 483 }