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).
5 * This allows consistent reporting of CPU thermal throttle events.
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).
11 * Author: Dmitriy Zavin (dmitriyz@google.com)
13 * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
14 * Inspired by Ross Biro's and Al Borchers' counter code.
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/export.h>
22 #include <linux/types.h>
23 #include <linux/init.h>
24 #include <linux/smp.h>
25 #include <linux/cpu.h>
27 #include <asm/processor.h>
32 #include <asm/trace/irq_vectors.h>
34 /* How long to wait between reporting thermal events */
35 #define CHECK_INTERVAL (300 * HZ)
37 #define THERMAL_THROTTLING_EVENT 0
38 #define POWER_LIMIT_EVENT 1
41 * Current thermal event state:
43 struct _thermal_state {
48 unsigned long last_count;
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 struct _thermal_state core_thresh0;
57 struct _thermal_state core_thresh1;
60 /* Callback to handle core threshold interrupts */
61 int (*platform_thermal_notify)(__u64 msr_val);
62 EXPORT_SYMBOL(platform_thermal_notify);
64 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
66 static atomic_t therm_throt_en = ATOMIC_INIT(0);
68 static u32 lvtthmr_init __read_mostly;
71 #define define_therm_throt_device_one_ro(_name) \
72 static DEVICE_ATTR(_name, 0444, \
73 therm_throt_device_show_##_name, \
76 #define define_therm_throt_device_show_func(event, name) \
78 static ssize_t therm_throt_device_show_##event##_##name( \
80 struct device_attribute *attr, \
83 unsigned int cpu = dev->id; \
86 preempt_disable(); /* CPU hotplug */ \
87 if (cpu_online(cpu)) { \
88 ret = sprintf(buf, "%lu\n", \
89 per_cpu(thermal_state, cpu).event.name); \
97 define_therm_throt_device_show_func(core_throttle, count);
98 define_therm_throt_device_one_ro(core_throttle_count);
100 define_therm_throt_device_show_func(core_power_limit, count);
101 define_therm_throt_device_one_ro(core_power_limit_count);
103 define_therm_throt_device_show_func(package_throttle, count);
104 define_therm_throt_device_one_ro(package_throttle_count);
106 define_therm_throt_device_show_func(package_power_limit, count);
107 define_therm_throt_device_one_ro(package_power_limit_count);
109 static struct attribute *thermal_throttle_attrs[] = {
110 &dev_attr_core_throttle_count.attr,
114 static struct attribute_group thermal_attr_group = {
115 .attrs = thermal_throttle_attrs,
116 .name = "thermal_throttle"
118 #endif /* CONFIG_SYSFS */
121 #define PACKAGE_LEVEL 1
124 * therm_throt_process - Process thermal throttling event from interrupt
125 * @curr: Whether the condition is current or not (boolean), since the
126 * thermal interrupt normally gets called both when the thermal
127 * event begins and once the event has ended.
129 * This function is called by the thermal interrupt after the
130 * IRQ has been acknowledged.
132 * It will take care of rate limiting and printing messages to the syslog.
134 * Returns: 0 : Event should NOT be further logged, i.e. still in
135 * "timeout" from previous log message.
136 * 1 : Event should be logged further, and a message has been
137 * printed to the syslog.
139 static int therm_throt_process(bool new_event, int event, int level)
141 struct _thermal_state *state;
142 unsigned int this_cpu = smp_processor_id();
145 struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
147 now = get_jiffies_64();
148 if (level == CORE_LEVEL) {
149 if (event == THERMAL_THROTTLING_EVENT)
150 state = &pstate->core_throttle;
151 else if (event == POWER_LIMIT_EVENT)
152 state = &pstate->core_power_limit;
155 } else if (level == PACKAGE_LEVEL) {
156 if (event == THERMAL_THROTTLING_EVENT)
157 state = &pstate->package_throttle;
158 else if (event == POWER_LIMIT_EVENT)
159 state = &pstate->package_power_limit;
165 old_event = state->new_event;
166 state->new_event = new_event;
171 if (time_before64(now, state->next_check) &&
172 state->count != state->last_count)
175 state->next_check = now + CHECK_INTERVAL;
176 state->last_count = state->count;
178 /* if we just entered the thermal event */
180 if (event == THERMAL_THROTTLING_EVENT)
181 printk(KERN_CRIT "CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",
183 level == CORE_LEVEL ? "Core" : "Package",
188 if (event == THERMAL_THROTTLING_EVENT)
189 printk(KERN_INFO "CPU%d: %s temperature/speed normal\n",
191 level == CORE_LEVEL ? "Core" : "Package");
198 static int thresh_event_valid(int event)
200 struct _thermal_state *state;
201 unsigned int this_cpu = smp_processor_id();
202 struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
203 u64 now = get_jiffies_64();
205 state = (event == 0) ? &pstate->core_thresh0 : &pstate->core_thresh1;
207 if (time_before64(now, state->next_check))
210 state->next_check = now + CHECK_INTERVAL;
214 static bool int_pln_enable;
215 static int __init int_pln_enable_setup(char *s)
217 int_pln_enable = true;
221 __setup("int_pln_enable", int_pln_enable_setup);
224 /* Add/Remove thermal_throttle interface for CPU device: */
225 static __cpuinit int thermal_throttle_add_dev(struct device *dev,
229 struct cpuinfo_x86 *c = &cpu_data(cpu);
231 err = sysfs_create_group(&dev->kobj, &thermal_attr_group);
235 if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
236 err = sysfs_add_file_to_group(&dev->kobj,
237 &dev_attr_core_power_limit_count.attr,
238 thermal_attr_group.name);
239 if (cpu_has(c, X86_FEATURE_PTS)) {
240 err = sysfs_add_file_to_group(&dev->kobj,
241 &dev_attr_package_throttle_count.attr,
242 thermal_attr_group.name);
243 if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
244 err = sysfs_add_file_to_group(&dev->kobj,
245 &dev_attr_package_power_limit_count.attr,
246 thermal_attr_group.name);
252 static __cpuinit void thermal_throttle_remove_dev(struct device *dev)
254 sysfs_remove_group(&dev->kobj, &thermal_attr_group);
257 /* Mutex protecting device creation against CPU hotplug: */
258 static DEFINE_MUTEX(therm_cpu_lock);
260 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
262 thermal_throttle_cpu_callback(struct notifier_block *nfb,
263 unsigned long action,
266 unsigned int cpu = (unsigned long)hcpu;
270 dev = get_cpu_device(cpu);
274 case CPU_UP_PREPARE_FROZEN:
275 mutex_lock(&therm_cpu_lock);
276 err = thermal_throttle_add_dev(dev, cpu);
277 mutex_unlock(&therm_cpu_lock);
280 case CPU_UP_CANCELED:
281 case CPU_UP_CANCELED_FROZEN:
283 case CPU_DEAD_FROZEN:
284 mutex_lock(&therm_cpu_lock);
285 thermal_throttle_remove_dev(dev);
286 mutex_unlock(&therm_cpu_lock);
289 return notifier_from_errno(err);
292 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
294 .notifier_call = thermal_throttle_cpu_callback,
297 static __init int thermal_throttle_init_device(void)
299 unsigned int cpu = 0;
302 if (!atomic_read(&therm_throt_en))
305 register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
307 #ifdef CONFIG_HOTPLUG_CPU
308 mutex_lock(&therm_cpu_lock);
310 /* connect live CPUs to sysfs */
311 for_each_online_cpu(cpu) {
312 err = thermal_throttle_add_dev(get_cpu_device(cpu), cpu);
315 #ifdef CONFIG_HOTPLUG_CPU
316 mutex_unlock(&therm_cpu_lock);
321 device_initcall(thermal_throttle_init_device);
323 #endif /* CONFIG_SYSFS */
325 static void notify_thresholds(__u64 msr_val)
327 /* check whether the interrupt handler is defined;
328 * otherwise simply return
330 if (!platform_thermal_notify)
333 /* lower threshold reached */
334 if ((msr_val & THERM_LOG_THRESHOLD0) && thresh_event_valid(0))
335 platform_thermal_notify(msr_val);
336 /* higher threshold reached */
337 if ((msr_val & THERM_LOG_THRESHOLD1) && thresh_event_valid(1))
338 platform_thermal_notify(msr_val);
341 /* Thermal transition interrupt handler */
342 static void intel_thermal_interrupt(void)
346 rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
348 /* Check for violation of core thermal thresholds*/
349 notify_thresholds(msr_val);
351 if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
352 THERMAL_THROTTLING_EVENT,
354 mce_log_therm_throt_event(msr_val);
356 if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
357 therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
361 if (this_cpu_has(X86_FEATURE_PTS)) {
362 rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
363 therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
364 THERMAL_THROTTLING_EVENT,
366 if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
367 therm_throt_process(msr_val &
368 PACKAGE_THERM_STATUS_POWER_LIMIT,
374 static void unexpected_thermal_interrupt(void)
376 printk(KERN_ERR "CPU%d: Unexpected LVT thermal interrupt!\n",
380 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
382 static inline void __smp_thermal_interrupt(void)
384 inc_irq_stat(irq_thermal_count);
385 smp_thermal_vector();
388 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
391 __smp_thermal_interrupt();
395 asmlinkage void smp_trace_thermal_interrupt(struct pt_regs *regs)
398 trace_thermal_apic_entry(THERMAL_APIC_VECTOR);
399 __smp_thermal_interrupt();
400 trace_thermal_apic_exit(THERMAL_APIC_VECTOR);
404 /* Thermal monitoring depends on APIC, ACPI and clock modulation */
405 static int intel_thermal_supported(struct cpuinfo_x86 *c)
409 if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
414 void __init mcheck_intel_therm_init(void)
417 * This function is only called on boot CPU. Save the init thermal
418 * LVT value on BSP and use that value to restore APs' thermal LVT
419 * entry BIOS programmed later
421 if (intel_thermal_supported(&boot_cpu_data))
422 lvtthmr_init = apic_read(APIC_LVTTHMR);
425 void intel_init_thermal(struct cpuinfo_x86 *c)
427 unsigned int cpu = smp_processor_id();
431 if (!intel_thermal_supported(c))
435 * First check if its enabled already, in which case there might
436 * be some SMM goo which handles it, so we can't even put a handler
437 * since it might be delivered via SMI already:
439 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
443 * The initial value of thermal LVT entries on all APs always reads
444 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
445 * sequence to them and LVT registers are reset to 0s except for
446 * the mask bits which are set to 1s when APs receive INIT IPI.
447 * If BIOS takes over the thermal interrupt and sets its interrupt
448 * delivery mode to SMI (not fixed), it restores the value that the
449 * BIOS has programmed on AP based on BSP's info we saved since BIOS
450 * is always setting the same value for all threads/cores.
452 if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
453 apic_write(APIC_LVTTHMR, lvtthmr_init);
456 if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
458 "CPU%d: Thermal monitoring handled by SMI\n", cpu);
462 /* Check whether a vector already exists */
463 if (h & APIC_VECTOR_MASK) {
465 "CPU%d: Thermal LVT vector (%#x) already installed\n",
466 cpu, (h & APIC_VECTOR_MASK));
470 /* early Pentium M models use different method for enabling TM2 */
471 if (cpu_has(c, X86_FEATURE_TM2)) {
472 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
473 rdmsr(MSR_THERM2_CTL, l, h);
474 if (l & MSR_THERM2_CTL_TM_SELECT)
476 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
480 /* We'll mask the thermal vector in the lapic till we're ready: */
481 h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
482 apic_write(APIC_LVTTHMR, h);
484 rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
485 if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
486 wrmsr(MSR_IA32_THERM_INTERRUPT,
487 (l | (THERM_INT_LOW_ENABLE
488 | THERM_INT_HIGH_ENABLE)) & ~THERM_INT_PLN_ENABLE, h);
489 else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
490 wrmsr(MSR_IA32_THERM_INTERRUPT,
491 l | (THERM_INT_LOW_ENABLE
492 | THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
494 wrmsr(MSR_IA32_THERM_INTERRUPT,
495 l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
497 if (cpu_has(c, X86_FEATURE_PTS)) {
498 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
499 if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
500 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
501 (l | (PACKAGE_THERM_INT_LOW_ENABLE
502 | PACKAGE_THERM_INT_HIGH_ENABLE))
503 & ~PACKAGE_THERM_INT_PLN_ENABLE, h);
504 else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
505 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
506 l | (PACKAGE_THERM_INT_LOW_ENABLE
507 | PACKAGE_THERM_INT_HIGH_ENABLE
508 | PACKAGE_THERM_INT_PLN_ENABLE), h);
510 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
511 l | (PACKAGE_THERM_INT_LOW_ENABLE
512 | PACKAGE_THERM_INT_HIGH_ENABLE), h);
515 smp_thermal_vector = intel_thermal_interrupt;
517 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
518 wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
520 /* Unmask the thermal vector: */
521 l = apic_read(APIC_LVTTHMR);
522 apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
524 printk_once(KERN_INFO "CPU0: Thermal monitoring enabled (%s)\n",
525 tm2 ? "TM2" : "TM1");
527 /* enable thermal throttle processing */
528 atomic_set(&therm_throt_en, 1);