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 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>
28 #include <asm/traps.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;
58 struct _thermal_state pkg_thresh0;
59 struct _thermal_state pkg_thresh1;
62 /* Callback to handle core threshold interrupts */
63 int (*platform_thermal_notify)(__u64 msr_val);
64 EXPORT_SYMBOL(platform_thermal_notify);
66 /* Callback to handle core package threshold_interrupts */
67 int (*platform_thermal_package_notify)(__u64 msr_val);
68 EXPORT_SYMBOL_GPL(platform_thermal_package_notify);
70 /* Callback support of rate control, return true, if
71 * callback has rate control */
72 bool (*platform_thermal_package_rate_control)(void);
73 EXPORT_SYMBOL_GPL(platform_thermal_package_rate_control);
76 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
78 static atomic_t therm_throt_en = ATOMIC_INIT(0);
80 static u32 lvtthmr_init __read_mostly;
83 #define define_therm_throt_device_one_ro(_name) \
84 static DEVICE_ATTR(_name, 0444, \
85 therm_throt_device_show_##_name, \
88 #define define_therm_throt_device_show_func(event, name) \
90 static ssize_t therm_throt_device_show_##event##_##name( \
92 struct device_attribute *attr, \
95 unsigned int cpu = dev->id; \
98 preempt_disable(); /* CPU hotplug */ \
99 if (cpu_online(cpu)) { \
100 ret = sprintf(buf, "%lu\n", \
101 per_cpu(thermal_state, cpu).event.name); \
109 define_therm_throt_device_show_func(core_throttle, count);
110 define_therm_throt_device_one_ro(core_throttle_count);
112 define_therm_throt_device_show_func(core_power_limit, count);
113 define_therm_throt_device_one_ro(core_power_limit_count);
115 define_therm_throt_device_show_func(package_throttle, count);
116 define_therm_throt_device_one_ro(package_throttle_count);
118 define_therm_throt_device_show_func(package_power_limit, count);
119 define_therm_throt_device_one_ro(package_power_limit_count);
121 static struct attribute *thermal_throttle_attrs[] = {
122 &dev_attr_core_throttle_count.attr,
126 static const struct attribute_group thermal_attr_group = {
127 .attrs = thermal_throttle_attrs,
128 .name = "thermal_throttle"
130 #endif /* CONFIG_SYSFS */
133 #define PACKAGE_LEVEL 1
136 * therm_throt_process - Process thermal throttling event from interrupt
137 * @curr: Whether the condition is current or not (boolean), since the
138 * thermal interrupt normally gets called both when the thermal
139 * event begins and once the event has ended.
141 * This function is called by the thermal interrupt after the
142 * IRQ has been acknowledged.
144 * It will take care of rate limiting and printing messages to the syslog.
146 static void therm_throt_process(bool new_event, int event, int level)
148 struct _thermal_state *state;
149 unsigned int this_cpu = smp_processor_id();
152 struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
154 now = get_jiffies_64();
155 if (level == CORE_LEVEL) {
156 if (event == THERMAL_THROTTLING_EVENT)
157 state = &pstate->core_throttle;
158 else if (event == POWER_LIMIT_EVENT)
159 state = &pstate->core_power_limit;
162 } else if (level == PACKAGE_LEVEL) {
163 if (event == THERMAL_THROTTLING_EVENT)
164 state = &pstate->package_throttle;
165 else if (event == POWER_LIMIT_EVENT)
166 state = &pstate->package_power_limit;
172 old_event = state->new_event;
173 state->new_event = new_event;
178 if (time_before64(now, state->next_check) &&
179 state->count != state->last_count)
182 state->next_check = now + CHECK_INTERVAL;
183 state->last_count = state->count;
185 /* if we just entered the thermal event */
187 if (event == THERMAL_THROTTLING_EVENT)
188 pr_warn("CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",
190 level == CORE_LEVEL ? "Core" : "Package",
195 if (event == THERMAL_THROTTLING_EVENT)
196 pr_info("CPU%d: %s temperature/speed normal\n", this_cpu,
197 level == CORE_LEVEL ? "Core" : "Package");
202 static int thresh_event_valid(int level, int event)
204 struct _thermal_state *state;
205 unsigned int this_cpu = smp_processor_id();
206 struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
207 u64 now = get_jiffies_64();
209 if (level == PACKAGE_LEVEL)
210 state = (event == 0) ? &pstate->pkg_thresh0 :
211 &pstate->pkg_thresh1;
213 state = (event == 0) ? &pstate->core_thresh0 :
214 &pstate->core_thresh1;
216 if (time_before64(now, state->next_check))
219 state->next_check = now + CHECK_INTERVAL;
224 static bool int_pln_enable;
225 static int __init int_pln_enable_setup(char *s)
227 int_pln_enable = true;
231 __setup("int_pln_enable", int_pln_enable_setup);
234 /* Add/Remove thermal_throttle interface for CPU device: */
235 static int thermal_throttle_add_dev(struct device *dev, unsigned int cpu)
238 struct cpuinfo_x86 *c = &cpu_data(cpu);
240 err = sysfs_create_group(&dev->kobj, &thermal_attr_group);
244 if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
245 err = sysfs_add_file_to_group(&dev->kobj,
246 &dev_attr_core_power_limit_count.attr,
247 thermal_attr_group.name);
248 if (cpu_has(c, X86_FEATURE_PTS)) {
249 err = sysfs_add_file_to_group(&dev->kobj,
250 &dev_attr_package_throttle_count.attr,
251 thermal_attr_group.name);
252 if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
253 err = sysfs_add_file_to_group(&dev->kobj,
254 &dev_attr_package_power_limit_count.attr,
255 thermal_attr_group.name);
261 static void thermal_throttle_remove_dev(struct device *dev)
263 sysfs_remove_group(&dev->kobj, &thermal_attr_group);
266 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
267 static int thermal_throttle_online(unsigned int cpu)
269 struct device *dev = get_cpu_device(cpu);
271 return thermal_throttle_add_dev(dev, cpu);
274 static int thermal_throttle_offline(unsigned int cpu)
276 struct device *dev = get_cpu_device(cpu);
278 thermal_throttle_remove_dev(dev);
282 static __init int thermal_throttle_init_device(void)
286 if (!atomic_read(&therm_throt_en))
289 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/therm:online",
290 thermal_throttle_online,
291 thermal_throttle_offline);
292 return ret < 0 ? ret : 0;
294 device_initcall(thermal_throttle_init_device);
296 #endif /* CONFIG_SYSFS */
298 static void notify_package_thresholds(__u64 msr_val)
300 bool notify_thres_0 = false;
301 bool notify_thres_1 = false;
303 if (!platform_thermal_package_notify)
306 /* lower threshold check */
307 if (msr_val & THERM_LOG_THRESHOLD0)
308 notify_thres_0 = true;
309 /* higher threshold check */
310 if (msr_val & THERM_LOG_THRESHOLD1)
311 notify_thres_1 = true;
313 if (!notify_thres_0 && !notify_thres_1)
316 if (platform_thermal_package_rate_control &&
317 platform_thermal_package_rate_control()) {
318 /* Rate control is implemented in callback */
319 platform_thermal_package_notify(msr_val);
323 /* lower threshold reached */
324 if (notify_thres_0 && thresh_event_valid(PACKAGE_LEVEL, 0))
325 platform_thermal_package_notify(msr_val);
326 /* higher threshold reached */
327 if (notify_thres_1 && thresh_event_valid(PACKAGE_LEVEL, 1))
328 platform_thermal_package_notify(msr_val);
331 static void notify_thresholds(__u64 msr_val)
333 /* check whether the interrupt handler is defined;
334 * otherwise simply return
336 if (!platform_thermal_notify)
339 /* lower threshold reached */
340 if ((msr_val & THERM_LOG_THRESHOLD0) &&
341 thresh_event_valid(CORE_LEVEL, 0))
342 platform_thermal_notify(msr_val);
343 /* higher threshold reached */
344 if ((msr_val & THERM_LOG_THRESHOLD1) &&
345 thresh_event_valid(CORE_LEVEL, 1))
346 platform_thermal_notify(msr_val);
349 /* Thermal transition interrupt handler */
350 static void intel_thermal_interrupt(void)
354 if (static_cpu_has(X86_FEATURE_HWP))
355 wrmsrl_safe(MSR_HWP_STATUS, 0);
357 rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
359 /* Check for violation of core thermal thresholds*/
360 notify_thresholds(msr_val);
362 therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
363 THERMAL_THROTTLING_EVENT,
366 if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
367 therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
371 if (this_cpu_has(X86_FEATURE_PTS)) {
372 rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
373 /* check violations of package thermal thresholds */
374 notify_package_thresholds(msr_val);
375 therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
376 THERMAL_THROTTLING_EVENT,
378 if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
379 therm_throt_process(msr_val &
380 PACKAGE_THERM_STATUS_POWER_LIMIT,
386 static void unexpected_thermal_interrupt(void)
388 pr_err("CPU%d: Unexpected LVT thermal interrupt!\n",
392 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
394 asmlinkage __visible void __irq_entry smp_thermal_interrupt(struct pt_regs *regs)
397 trace_thermal_apic_entry(THERMAL_APIC_VECTOR);
398 inc_irq_stat(irq_thermal_count);
399 smp_thermal_vector();
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)
407 if (!boot_cpu_has(X86_FEATURE_APIC))
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)) {
457 if (system_state == SYSTEM_BOOTING)
458 pr_debug("CPU%d: Thermal monitoring handled by SMI\n", cpu);
462 /* early Pentium M models use different method for enabling TM2 */
463 if (cpu_has(c, X86_FEATURE_TM2)) {
464 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
465 rdmsr(MSR_THERM2_CTL, l, h);
466 if (l & MSR_THERM2_CTL_TM_SELECT)
468 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
472 /* We'll mask the thermal vector in the lapic till we're ready: */
473 h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
474 apic_write(APIC_LVTTHMR, h);
476 rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
477 if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
478 wrmsr(MSR_IA32_THERM_INTERRUPT,
479 (l | (THERM_INT_LOW_ENABLE
480 | THERM_INT_HIGH_ENABLE)) & ~THERM_INT_PLN_ENABLE, h);
481 else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
482 wrmsr(MSR_IA32_THERM_INTERRUPT,
483 l | (THERM_INT_LOW_ENABLE
484 | THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
486 wrmsr(MSR_IA32_THERM_INTERRUPT,
487 l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
489 if (cpu_has(c, X86_FEATURE_PTS)) {
490 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
491 if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
492 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
493 (l | (PACKAGE_THERM_INT_LOW_ENABLE
494 | PACKAGE_THERM_INT_HIGH_ENABLE))
495 & ~PACKAGE_THERM_INT_PLN_ENABLE, h);
496 else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
497 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
498 l | (PACKAGE_THERM_INT_LOW_ENABLE
499 | PACKAGE_THERM_INT_HIGH_ENABLE
500 | PACKAGE_THERM_INT_PLN_ENABLE), h);
502 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
503 l | (PACKAGE_THERM_INT_LOW_ENABLE
504 | PACKAGE_THERM_INT_HIGH_ENABLE), h);
507 smp_thermal_vector = intel_thermal_interrupt;
509 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
510 wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
512 /* Unmask the thermal vector: */
513 l = apic_read(APIC_LVTTHMR);
514 apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
516 pr_info_once("CPU0: Thermal monitoring enabled (%s)\n",
517 tm2 ? "TM2" : "TM1");
519 /* enable thermal throttle processing */
520 atomic_set(&therm_throt_en, 1);