Linux Audio

Check our new training course

Buildroot integration, development and maintenance

Need a Buildroot system for your embedded project?
Loading...
Note: File does not exist in v3.1.
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Thermal throttle event support code (such as syslog messaging and rate
  4 * limiting) that was factored out from x86_64 (mce_intel.c) and i386 (p4.c).
  5 *
  6 * This allows consistent reporting of CPU thermal throttle events.
  7 *
  8 * Maintains a counter in /sys that keeps track of the number of thermal
  9 * events, such that the user knows how bad the thermal problem might be
 10 * (since the logging to syslog is rate limited).
 11 *
 12 * Author: Dmitriy Zavin (dmitriyz@google.com)
 13 *
 14 * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
 15 *          Inspired by Ross Biro's and Al Borchers' counter code.
 16 */
 17#include <linux/interrupt.h>
 18#include <linux/notifier.h>
 19#include <linux/jiffies.h>
 20#include <linux/kernel.h>
 21#include <linux/percpu.h>
 22#include <linux/export.h>
 23#include <linux/types.h>
 24#include <linux/init.h>
 25#include <linux/smp.h>
 26#include <linux/cpu.h>
 27
 28#include <asm/processor.h>
 29#include <asm/thermal.h>
 30#include <asm/traps.h>
 31#include <asm/apic.h>
 32#include <asm/irq.h>
 33#include <asm/msr.h>
 34
 35#include "intel_hfi.h"
 36#include "thermal_interrupt.h"
 37
 38/* How long to wait between reporting thermal events */
 39#define CHECK_INTERVAL		(300 * HZ)
 40
 41#define THERMAL_THROTTLING_EVENT	0
 42#define POWER_LIMIT_EVENT		1
 43
 44/**
 45 * struct _thermal_state - Represent the current thermal event state
 46 * @next_check:			Stores the next timestamp, when it is allowed
 47 *				to log the next warning message.
 48 * @last_interrupt_time:	Stores the timestamp for the last threshold
 49 *				high event.
 50 * @therm_work:			Delayed workqueue structure
 51 * @count:			Stores the current running count for thermal
 52 *				or power threshold interrupts.
 53 * @last_count:			Stores the previous running count for thermal
 54 *				or power threshold interrupts.
 55 * @max_time_ms:		This shows the maximum amount of time CPU was
 56 *				in throttled state for a single thermal
 57 *				threshold high to low state.
 58 * @total_time_ms:		This is a cumulative time during which CPU was
 59 *				in the throttled state.
 60 * @rate_control_active:	Set when a throttling message is logged.
 61 *				This is used for the purpose of rate-control.
 62 * @new_event:			Stores the last high/low status of the
 63 *				THERM_STATUS_PROCHOT or
 64 *				THERM_STATUS_POWER_LIMIT.
 65 * @level:			Stores whether this _thermal_state instance is
 66 *				for a CORE level or for PACKAGE level.
 67 * @sample_index:		Index for storing the next sample in the buffer
 68 *				temp_samples[].
 69 * @sample_count:		Total number of samples collected in the buffer
 70 *				temp_samples[].
 71 * @average:			The last moving average of temperature samples
 72 * @baseline_temp:		Temperature at which thermal threshold high
 73 *				interrupt was generated.
 74 * @temp_samples:		Storage for temperature samples to calculate
 75 *				moving average.
 76 *
 77 * This structure is used to represent data related to thermal state for a CPU.
 78 * There is a separate storage for core and package level for each CPU.
 79 */
 80struct _thermal_state {
 81	u64			next_check;
 82	u64			last_interrupt_time;
 83	struct delayed_work	therm_work;
 84	unsigned long		count;
 85	unsigned long		last_count;
 86	unsigned long		max_time_ms;
 87	unsigned long		total_time_ms;
 88	bool			rate_control_active;
 89	bool			new_event;
 90	u8			level;
 91	u8			sample_index;
 92	u8			sample_count;
 93	u8			average;
 94	u8			baseline_temp;
 95	u8			temp_samples[3];
 96};
 97
 98struct thermal_state {
 99	struct _thermal_state core_throttle;
100	struct _thermal_state core_power_limit;
101	struct _thermal_state package_throttle;
102	struct _thermal_state package_power_limit;
103	struct _thermal_state core_thresh0;
104	struct _thermal_state core_thresh1;
105	struct _thermal_state pkg_thresh0;
106	struct _thermal_state pkg_thresh1;
107};
108
109/* Callback to handle core threshold interrupts */
110int (*platform_thermal_notify)(__u64 msr_val);
111EXPORT_SYMBOL(platform_thermal_notify);
112
113/* Callback to handle core package threshold_interrupts */
114int (*platform_thermal_package_notify)(__u64 msr_val);
115EXPORT_SYMBOL_GPL(platform_thermal_package_notify);
116
117/* Callback support of rate control, return true, if
118 * callback has rate control */
119bool (*platform_thermal_package_rate_control)(void);
120EXPORT_SYMBOL_GPL(platform_thermal_package_rate_control);
121
122
123static DEFINE_PER_CPU(struct thermal_state, thermal_state);
124
125static atomic_t therm_throt_en	= ATOMIC_INIT(0);
126
127static u32 lvtthmr_init __read_mostly;
128
129#ifdef CONFIG_SYSFS
130#define define_therm_throt_device_one_ro(_name)				\
131	static DEVICE_ATTR(_name, 0444,					\
132			   therm_throt_device_show_##_name,		\
133				   NULL)				\
134
135#define define_therm_throt_device_show_func(event, name)		\
136									\
137static ssize_t therm_throt_device_show_##event##_##name(		\
138			struct device *dev,				\
139			struct device_attribute *attr,			\
140			char *buf)					\
141{									\
142	unsigned int cpu = dev->id;					\
143	ssize_t ret;							\
144									\
145	preempt_disable();	/* CPU hotplug */			\
146	if (cpu_online(cpu)) {						\
147		ret = sprintf(buf, "%lu\n",				\
148			      per_cpu(thermal_state, cpu).event.name);	\
149	} else								\
150		ret = 0;						\
151	preempt_enable();						\
152									\
153	return ret;							\
154}
155
156define_therm_throt_device_show_func(core_throttle, count);
157define_therm_throt_device_one_ro(core_throttle_count);
158
159define_therm_throt_device_show_func(core_power_limit, count);
160define_therm_throt_device_one_ro(core_power_limit_count);
161
162define_therm_throt_device_show_func(package_throttle, count);
163define_therm_throt_device_one_ro(package_throttle_count);
164
165define_therm_throt_device_show_func(package_power_limit, count);
166define_therm_throt_device_one_ro(package_power_limit_count);
167
168define_therm_throt_device_show_func(core_throttle, max_time_ms);
169define_therm_throt_device_one_ro(core_throttle_max_time_ms);
170
171define_therm_throt_device_show_func(package_throttle, max_time_ms);
172define_therm_throt_device_one_ro(package_throttle_max_time_ms);
173
174define_therm_throt_device_show_func(core_throttle, total_time_ms);
175define_therm_throt_device_one_ro(core_throttle_total_time_ms);
176
177define_therm_throt_device_show_func(package_throttle, total_time_ms);
178define_therm_throt_device_one_ro(package_throttle_total_time_ms);
179
180static struct attribute *thermal_throttle_attrs[] = {
181	&dev_attr_core_throttle_count.attr,
182	&dev_attr_core_throttle_max_time_ms.attr,
183	&dev_attr_core_throttle_total_time_ms.attr,
184	NULL
185};
186
187static const struct attribute_group thermal_attr_group = {
188	.attrs	= thermal_throttle_attrs,
189	.name	= "thermal_throttle"
190};
191#endif /* CONFIG_SYSFS */
192
193#define THERM_THROT_POLL_INTERVAL	HZ
194#define THERM_STATUS_PROCHOT_LOG	BIT(1)
195
196#define THERM_STATUS_CLEAR_CORE_MASK (BIT(1) | BIT(3) | BIT(5) | BIT(7) | BIT(9) | BIT(11) | BIT(13) | BIT(15))
197#define THERM_STATUS_CLEAR_PKG_MASK  (BIT(1) | BIT(3) | BIT(5) | BIT(7) | BIT(9) | BIT(11))
198
199/*
200 * Clear the bits in package thermal status register for bit = 1
201 * in bitmask
202 */
203void thermal_clear_package_intr_status(int level, u64 bit_mask)
204{
205	u64 msr_val;
206	int msr;
207
208	if (level == CORE_LEVEL) {
209		msr  = MSR_IA32_THERM_STATUS;
210		msr_val = THERM_STATUS_CLEAR_CORE_MASK;
211	} else {
212		msr  = MSR_IA32_PACKAGE_THERM_STATUS;
213		msr_val = THERM_STATUS_CLEAR_PKG_MASK;
214		if (boot_cpu_has(X86_FEATURE_HFI))
215			msr_val |= BIT(26);
216
217	}
218
219	msr_val &= ~bit_mask;
220	wrmsrl(msr, msr_val);
221}
222EXPORT_SYMBOL_GPL(thermal_clear_package_intr_status);
223
224static void get_therm_status(int level, bool *proc_hot, u8 *temp)
225{
226	int msr;
227	u64 msr_val;
228
229	if (level == CORE_LEVEL)
230		msr = MSR_IA32_THERM_STATUS;
231	else
232		msr = MSR_IA32_PACKAGE_THERM_STATUS;
233
234	rdmsrl(msr, msr_val);
235	if (msr_val & THERM_STATUS_PROCHOT_LOG)
236		*proc_hot = true;
237	else
238		*proc_hot = false;
239
240	*temp = (msr_val >> 16) & 0x7F;
241}
242
243static void __maybe_unused throttle_active_work(struct work_struct *work)
244{
245	struct _thermal_state *state = container_of(to_delayed_work(work),
246						struct _thermal_state, therm_work);
247	unsigned int i, avg, this_cpu = smp_processor_id();
248	u64 now = get_jiffies_64();
249	bool hot;
250	u8 temp;
251
252	get_therm_status(state->level, &hot, &temp);
253	/* temperature value is offset from the max so lesser means hotter */
254	if (!hot && temp > state->baseline_temp) {
255		if (state->rate_control_active)
256			pr_info("CPU%d: %s temperature/speed normal (total events = %lu)\n",
257				this_cpu,
258				state->level == CORE_LEVEL ? "Core" : "Package",
259				state->count);
260
261		state->rate_control_active = false;
262		return;
263	}
264
265	if (time_before64(now, state->next_check) &&
266			  state->rate_control_active)
267		goto re_arm;
268
269	state->next_check = now + CHECK_INTERVAL;
270
271	if (state->count != state->last_count) {
272		/* There was one new thermal interrupt */
273		state->last_count = state->count;
274		state->average = 0;
275		state->sample_count = 0;
276		state->sample_index = 0;
277	}
278
279	state->temp_samples[state->sample_index] = temp;
280	state->sample_count++;
281	state->sample_index = (state->sample_index + 1) % ARRAY_SIZE(state->temp_samples);
282	if (state->sample_count < ARRAY_SIZE(state->temp_samples))
283		goto re_arm;
284
285	avg = 0;
286	for (i = 0; i < ARRAY_SIZE(state->temp_samples); ++i)
287		avg += state->temp_samples[i];
288
289	avg /= ARRAY_SIZE(state->temp_samples);
290
291	if (state->average > avg) {
292		pr_warn("CPU%d: %s temperature is above threshold, cpu clock is throttled (total events = %lu)\n",
293			this_cpu,
294			state->level == CORE_LEVEL ? "Core" : "Package",
295			state->count);
296		state->rate_control_active = true;
297	}
298
299	state->average = avg;
300
301re_arm:
302	thermal_clear_package_intr_status(state->level, THERM_STATUS_PROCHOT_LOG);
303	schedule_delayed_work_on(this_cpu, &state->therm_work, THERM_THROT_POLL_INTERVAL);
304}
305
306/***
307 * therm_throt_process - Process thermal throttling event from interrupt
308 * @curr: Whether the condition is current or not (boolean), since the
309 *        thermal interrupt normally gets called both when the thermal
310 *        event begins and once the event has ended.
311 *
312 * This function is called by the thermal interrupt after the
313 * IRQ has been acknowledged.
314 *
315 * It will take care of rate limiting and printing messages to the syslog.
316 */
317static void therm_throt_process(bool new_event, int event, int level)
318{
319	struct _thermal_state *state;
320	unsigned int this_cpu = smp_processor_id();
321	bool old_event;
322	u64 now;
323	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
324
325	now = get_jiffies_64();
326	if (level == CORE_LEVEL) {
327		if (event == THERMAL_THROTTLING_EVENT)
328			state = &pstate->core_throttle;
329		else if (event == POWER_LIMIT_EVENT)
330			state = &pstate->core_power_limit;
331		else
332			return;
333	} else if (level == PACKAGE_LEVEL) {
334		if (event == THERMAL_THROTTLING_EVENT)
335			state = &pstate->package_throttle;
336		else if (event == POWER_LIMIT_EVENT)
337			state = &pstate->package_power_limit;
338		else
339			return;
340	} else
341		return;
342
343	old_event = state->new_event;
344	state->new_event = new_event;
345
346	if (new_event)
347		state->count++;
348
349	if (event != THERMAL_THROTTLING_EVENT)
350		return;
351
352	if (new_event && !state->last_interrupt_time) {
353		bool hot;
354		u8 temp;
355
356		get_therm_status(state->level, &hot, &temp);
357		/*
358		 * Ignore short temperature spike as the system is not close
359		 * to PROCHOT. 10C offset is large enough to ignore. It is
360		 * already dropped from the high threshold temperature.
361		 */
362		if (temp > 10)
363			return;
364
365		state->baseline_temp = temp;
366		state->last_interrupt_time = now;
367		schedule_delayed_work_on(this_cpu, &state->therm_work, THERM_THROT_POLL_INTERVAL);
368	} else if (old_event && state->last_interrupt_time) {
369		unsigned long throttle_time;
370
371		throttle_time = jiffies_delta_to_msecs(now - state->last_interrupt_time);
372		if (throttle_time > state->max_time_ms)
373			state->max_time_ms = throttle_time;
374		state->total_time_ms += throttle_time;
375		state->last_interrupt_time = 0;
376	}
377}
378
379static int thresh_event_valid(int level, int event)
380{
381	struct _thermal_state *state;
382	unsigned int this_cpu = smp_processor_id();
383	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
384	u64 now = get_jiffies_64();
385
386	if (level == PACKAGE_LEVEL)
387		state = (event == 0) ? &pstate->pkg_thresh0 :
388						&pstate->pkg_thresh1;
389	else
390		state = (event == 0) ? &pstate->core_thresh0 :
391						&pstate->core_thresh1;
392
393	if (time_before64(now, state->next_check))
394		return 0;
395
396	state->next_check = now + CHECK_INTERVAL;
397
398	return 1;
399}
400
401static bool int_pln_enable;
402static int __init int_pln_enable_setup(char *s)
403{
404	int_pln_enable = true;
405
406	return 1;
407}
408__setup("int_pln_enable", int_pln_enable_setup);
409
410#ifdef CONFIG_SYSFS
411/* Add/Remove thermal_throttle interface for CPU device: */
412static int thermal_throttle_add_dev(struct device *dev, unsigned int cpu)
413{
414	int err;
415	struct cpuinfo_x86 *c = &cpu_data(cpu);
416
417	err = sysfs_create_group(&dev->kobj, &thermal_attr_group);
418	if (err)
419		return err;
420
421	if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable) {
422		err = sysfs_add_file_to_group(&dev->kobj,
423					      &dev_attr_core_power_limit_count.attr,
424					      thermal_attr_group.name);
425		if (err)
426			goto del_group;
427	}
428
429	if (cpu_has(c, X86_FEATURE_PTS)) {
430		err = sysfs_add_file_to_group(&dev->kobj,
431					      &dev_attr_package_throttle_count.attr,
432					      thermal_attr_group.name);
433		if (err)
434			goto del_group;
435
436		err = sysfs_add_file_to_group(&dev->kobj,
437					      &dev_attr_package_throttle_max_time_ms.attr,
438					      thermal_attr_group.name);
439		if (err)
440			goto del_group;
441
442		err = sysfs_add_file_to_group(&dev->kobj,
443					      &dev_attr_package_throttle_total_time_ms.attr,
444					      thermal_attr_group.name);
445		if (err)
446			goto del_group;
447
448		if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable) {
449			err = sysfs_add_file_to_group(&dev->kobj,
450					&dev_attr_package_power_limit_count.attr,
451					thermal_attr_group.name);
452			if (err)
453				goto del_group;
454		}
455	}
456
457	return 0;
458
459del_group:
460	sysfs_remove_group(&dev->kobj, &thermal_attr_group);
461
462	return err;
463}
464
465static void thermal_throttle_remove_dev(struct device *dev)
466{
467	sysfs_remove_group(&dev->kobj, &thermal_attr_group);
468}
469
470/* Get notified when a cpu comes on/off. Be hotplug friendly. */
471static int thermal_throttle_online(unsigned int cpu)
472{
473	struct thermal_state *state = &per_cpu(thermal_state, cpu);
474	struct device *dev = get_cpu_device(cpu);
475	u32 l;
476
477	state->package_throttle.level = PACKAGE_LEVEL;
478	state->core_throttle.level = CORE_LEVEL;
479
480	INIT_DELAYED_WORK(&state->package_throttle.therm_work, throttle_active_work);
481	INIT_DELAYED_WORK(&state->core_throttle.therm_work, throttle_active_work);
482
483	/*
484	 * The first CPU coming online will enable the HFI. Usually this causes
485	 * hardware to issue an HFI thermal interrupt. Such interrupt will reach
486	 * the CPU once we enable the thermal vector in the local APIC.
487	 */
488	intel_hfi_online(cpu);
489
490	/* Unmask the thermal vector after the above workqueues are initialized. */
491	l = apic_read(APIC_LVTTHMR);
492	apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
493
494	return thermal_throttle_add_dev(dev, cpu);
495}
496
497static int thermal_throttle_offline(unsigned int cpu)
498{
499	struct thermal_state *state = &per_cpu(thermal_state, cpu);
500	struct device *dev = get_cpu_device(cpu);
501	u32 l;
502
503	/* Mask the thermal vector before draining evtl. pending work */
504	l = apic_read(APIC_LVTTHMR);
505	apic_write(APIC_LVTTHMR, l | APIC_LVT_MASKED);
506
507	intel_hfi_offline(cpu);
508
509	cancel_delayed_work_sync(&state->package_throttle.therm_work);
510	cancel_delayed_work_sync(&state->core_throttle.therm_work);
511
512	state->package_throttle.rate_control_active = false;
513	state->core_throttle.rate_control_active = false;
514
515	thermal_throttle_remove_dev(dev);
516	return 0;
517}
518
519static __init int thermal_throttle_init_device(void)
520{
521	int ret;
522
523	if (!atomic_read(&therm_throt_en))
524		return 0;
525
526	intel_hfi_init();
527
528	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/therm:online",
529				thermal_throttle_online,
530				thermal_throttle_offline);
531	return ret < 0 ? ret : 0;
532}
533device_initcall(thermal_throttle_init_device);
534
535#endif /* CONFIG_SYSFS */
536
537static void notify_package_thresholds(__u64 msr_val)
538{
539	bool notify_thres_0 = false;
540	bool notify_thres_1 = false;
541
542	if (!platform_thermal_package_notify)
543		return;
544
545	/* lower threshold check */
546	if (msr_val & THERM_LOG_THRESHOLD0)
547		notify_thres_0 = true;
548	/* higher threshold check */
549	if (msr_val & THERM_LOG_THRESHOLD1)
550		notify_thres_1 = true;
551
552	if (!notify_thres_0 && !notify_thres_1)
553		return;
554
555	if (platform_thermal_package_rate_control &&
556		platform_thermal_package_rate_control()) {
557		/* Rate control is implemented in callback */
558		platform_thermal_package_notify(msr_val);
559		return;
560	}
561
562	/* lower threshold reached */
563	if (notify_thres_0 && thresh_event_valid(PACKAGE_LEVEL, 0))
564		platform_thermal_package_notify(msr_val);
565	/* higher threshold reached */
566	if (notify_thres_1 && thresh_event_valid(PACKAGE_LEVEL, 1))
567		platform_thermal_package_notify(msr_val);
568}
569
570static void notify_thresholds(__u64 msr_val)
571{
572	/* check whether the interrupt handler is defined;
573	 * otherwise simply return
574	 */
575	if (!platform_thermal_notify)
576		return;
577
578	/* lower threshold reached */
579	if ((msr_val & THERM_LOG_THRESHOLD0) &&
580			thresh_event_valid(CORE_LEVEL, 0))
581		platform_thermal_notify(msr_val);
582	/* higher threshold reached */
583	if ((msr_val & THERM_LOG_THRESHOLD1) &&
584			thresh_event_valid(CORE_LEVEL, 1))
585		platform_thermal_notify(msr_val);
586}
587
588void __weak notify_hwp_interrupt(void)
589{
590	wrmsrl_safe(MSR_HWP_STATUS, 0);
591}
592
593/* Thermal transition interrupt handler */
594void intel_thermal_interrupt(void)
595{
596	__u64 msr_val;
597
598	if (static_cpu_has(X86_FEATURE_HWP))
599		notify_hwp_interrupt();
600
601	rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
602
603	/* Check for violation of core thermal thresholds*/
604	notify_thresholds(msr_val);
605
606	therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
607			    THERMAL_THROTTLING_EVENT,
608			    CORE_LEVEL);
609
610	if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
611		therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
612					POWER_LIMIT_EVENT,
613					CORE_LEVEL);
614
615	if (this_cpu_has(X86_FEATURE_PTS)) {
616		rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
617		/* check violations of package thermal thresholds */
618		notify_package_thresholds(msr_val);
619		therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
620					THERMAL_THROTTLING_EVENT,
621					PACKAGE_LEVEL);
622		if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
623			therm_throt_process(msr_val &
624					PACKAGE_THERM_STATUS_POWER_LIMIT,
625					POWER_LIMIT_EVENT,
626					PACKAGE_LEVEL);
627
628		if (this_cpu_has(X86_FEATURE_HFI))
629			intel_hfi_process_event(msr_val &
630						PACKAGE_THERM_STATUS_HFI_UPDATED);
631	}
632}
633
634/* Thermal monitoring depends on APIC, ACPI and clock modulation */
635static int intel_thermal_supported(struct cpuinfo_x86 *c)
636{
637	if (!boot_cpu_has(X86_FEATURE_APIC))
638		return 0;
639	if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
640		return 0;
641	return 1;
642}
643
644bool x86_thermal_enabled(void)
645{
646	return atomic_read(&therm_throt_en);
647}
648
649void __init therm_lvt_init(void)
650{
651	/*
652	 * This function is only called on boot CPU. Save the init thermal
653	 * LVT value on BSP and use that value to restore APs' thermal LVT
654	 * entry BIOS programmed later
655	 */
656	if (intel_thermal_supported(&boot_cpu_data))
657		lvtthmr_init = apic_read(APIC_LVTTHMR);
658}
659
660void intel_init_thermal(struct cpuinfo_x86 *c)
661{
662	unsigned int cpu = smp_processor_id();
663	int tm2 = 0;
664	u32 l, h;
665
666	if (!intel_thermal_supported(c))
667		return;
668
669	/*
670	 * First check if its enabled already, in which case there might
671	 * be some SMM goo which handles it, so we can't even put a handler
672	 * since it might be delivered via SMI already:
673	 */
674	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
675
676	h = lvtthmr_init;
677	/*
678	 * The initial value of thermal LVT entries on all APs always reads
679	 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
680	 * sequence to them and LVT registers are reset to 0s except for
681	 * the mask bits which are set to 1s when APs receive INIT IPI.
682	 * If BIOS takes over the thermal interrupt and sets its interrupt
683	 * delivery mode to SMI (not fixed), it restores the value that the
684	 * BIOS has programmed on AP based on BSP's info we saved since BIOS
685	 * is always setting the same value for all threads/cores.
686	 */
687	if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
688		apic_write(APIC_LVTTHMR, lvtthmr_init);
689
690
691	if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
692		if (system_state == SYSTEM_BOOTING)
693			pr_debug("CPU%d: Thermal monitoring handled by SMI\n", cpu);
694		return;
695	}
696
697	/* early Pentium M models use different method for enabling TM2 */
698	if (cpu_has(c, X86_FEATURE_TM2)) {
699		if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
700			rdmsr(MSR_THERM2_CTL, l, h);
701			if (l & MSR_THERM2_CTL_TM_SELECT)
702				tm2 = 1;
703		} else if (l & MSR_IA32_MISC_ENABLE_TM2)
704			tm2 = 1;
705	}
706
707	/* We'll mask the thermal vector in the lapic till we're ready: */
708	h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
709	apic_write(APIC_LVTTHMR, h);
710
711	rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
712	if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
713		wrmsr(MSR_IA32_THERM_INTERRUPT,
714			(l | (THERM_INT_LOW_ENABLE
715			| THERM_INT_HIGH_ENABLE)) & ~THERM_INT_PLN_ENABLE, h);
716	else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
717		wrmsr(MSR_IA32_THERM_INTERRUPT,
718			l | (THERM_INT_LOW_ENABLE
719			| THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
720	else
721		wrmsr(MSR_IA32_THERM_INTERRUPT,
722		      l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
723
724	if (cpu_has(c, X86_FEATURE_PTS)) {
725		rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
726		if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
727			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
728				(l | (PACKAGE_THERM_INT_LOW_ENABLE
729				| PACKAGE_THERM_INT_HIGH_ENABLE))
730				& ~PACKAGE_THERM_INT_PLN_ENABLE, h);
731		else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
732			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
733				l | (PACKAGE_THERM_INT_LOW_ENABLE
734				| PACKAGE_THERM_INT_HIGH_ENABLE
735				| PACKAGE_THERM_INT_PLN_ENABLE), h);
736		else
737			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
738			      l | (PACKAGE_THERM_INT_LOW_ENABLE
739				| PACKAGE_THERM_INT_HIGH_ENABLE), h);
740
741		if (cpu_has(c, X86_FEATURE_HFI)) {
742			rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
743			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
744			      l | PACKAGE_THERM_INT_HFI_ENABLE, h);
745		}
746	}
747
748	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
749	wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
750
751	pr_info_once("CPU0: Thermal monitoring enabled (%s)\n",
752		      tm2 ? "TM2" : "TM1");
753
754	/* enable thermal throttle processing */
755	atomic_set(&therm_throt_en, 1);
756}