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  */
80 struct _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 
98 struct 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 */
110 int (*platform_thermal_notify)(__u64 msr_val);
111 EXPORT_SYMBOL(platform_thermal_notify);
112 
113 /* Callback to handle core package threshold_interrupts */
114 int (*platform_thermal_package_notify)(__u64 msr_val);
115 EXPORT_SYMBOL_GPL(platform_thermal_package_notify);
116 
117 /* Callback support of rate control, return true, if
118  * callback has rate control */
119 bool (*platform_thermal_package_rate_control)(void);
120 EXPORT_SYMBOL_GPL(platform_thermal_package_rate_control);
121 
122 
123 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
124 
125 static atomic_t therm_throt_en	= ATOMIC_INIT(0);
126 
127 static 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 									\
137 static 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 
156 define_therm_throt_device_show_func(core_throttle, count);
157 define_therm_throt_device_one_ro(core_throttle_count);
158 
159 define_therm_throt_device_show_func(core_power_limit, count);
160 define_therm_throt_device_one_ro(core_power_limit_count);
161 
162 define_therm_throt_device_show_func(package_throttle, count);
163 define_therm_throt_device_one_ro(package_throttle_count);
164 
165 define_therm_throt_device_show_func(package_power_limit, count);
166 define_therm_throt_device_one_ro(package_power_limit_count);
167 
168 define_therm_throt_device_show_func(core_throttle, max_time_ms);
169 define_therm_throt_device_one_ro(core_throttle_max_time_ms);
170 
171 define_therm_throt_device_show_func(package_throttle, max_time_ms);
172 define_therm_throt_device_one_ro(package_throttle_max_time_ms);
173 
174 define_therm_throt_device_show_func(core_throttle, total_time_ms);
175 define_therm_throt_device_one_ro(core_throttle_total_time_ms);
176 
177 define_therm_throt_device_show_func(package_throttle, total_time_ms);
178 define_therm_throt_device_one_ro(package_throttle_total_time_ms);
179 
180 static 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 
187 static const struct attribute_group thermal_attr_group = {
188 	.attrs	= thermal_throttle_attrs,
189 	.name	= "thermal_throttle"
190 };
191 #endif /* CONFIG_SYSFS */
192 
193 #define CORE_LEVEL	0
194 #define PACKAGE_LEVEL	1
195 
196 #define THERM_THROT_POLL_INTERVAL	HZ
197 #define THERM_STATUS_PROCHOT_LOG	BIT(1)
198 
199 #define THERM_STATUS_CLEAR_CORE_MASK (BIT(1) | BIT(3) | BIT(5) | BIT(7) | BIT(9) | BIT(11) | BIT(13) | BIT(15))
200 #define THERM_STATUS_CLEAR_PKG_MASK  (BIT(1) | BIT(3) | BIT(5) | BIT(7) | BIT(9) | BIT(11))
201 
202 static void clear_therm_status_log(int level)
203 {
204 	int msr;
205 	u64 mask, msr_val;
206 
207 	if (level == CORE_LEVEL) {
208 		msr  = MSR_IA32_THERM_STATUS;
209 		mask = THERM_STATUS_CLEAR_CORE_MASK;
210 	} else {
211 		msr  = MSR_IA32_PACKAGE_THERM_STATUS;
212 		mask = THERM_STATUS_CLEAR_PKG_MASK;
213 	}
214 
215 	rdmsrl(msr, msr_val);
216 	msr_val &= mask;
217 	wrmsrl(msr, msr_val & ~THERM_STATUS_PROCHOT_LOG);
218 }
219 
220 static void get_therm_status(int level, bool *proc_hot, u8 *temp)
221 {
222 	int msr;
223 	u64 msr_val;
224 
225 	if (level == CORE_LEVEL)
226 		msr = MSR_IA32_THERM_STATUS;
227 	else
228 		msr = MSR_IA32_PACKAGE_THERM_STATUS;
229 
230 	rdmsrl(msr, msr_val);
231 	if (msr_val & THERM_STATUS_PROCHOT_LOG)
232 		*proc_hot = true;
233 	else
234 		*proc_hot = false;
235 
236 	*temp = (msr_val >> 16) & 0x7F;
237 }
238 
239 static void __maybe_unused throttle_active_work(struct work_struct *work)
240 {
241 	struct _thermal_state *state = container_of(to_delayed_work(work),
242 						struct _thermal_state, therm_work);
243 	unsigned int i, avg, this_cpu = smp_processor_id();
244 	u64 now = get_jiffies_64();
245 	bool hot;
246 	u8 temp;
247 
248 	get_therm_status(state->level, &hot, &temp);
249 	/* temperature value is offset from the max so lesser means hotter */
250 	if (!hot && temp > state->baseline_temp) {
251 		if (state->rate_control_active)
252 			pr_info("CPU%d: %s temperature/speed normal (total events = %lu)\n",
253 				this_cpu,
254 				state->level == CORE_LEVEL ? "Core" : "Package",
255 				state->count);
256 
257 		state->rate_control_active = false;
258 		return;
259 	}
260 
261 	if (time_before64(now, state->next_check) &&
262 			  state->rate_control_active)
263 		goto re_arm;
264 
265 	state->next_check = now + CHECK_INTERVAL;
266 
267 	if (state->count != state->last_count) {
268 		/* There was one new thermal interrupt */
269 		state->last_count = state->count;
270 		state->average = 0;
271 		state->sample_count = 0;
272 		state->sample_index = 0;
273 	}
274 
275 	state->temp_samples[state->sample_index] = temp;
276 	state->sample_count++;
277 	state->sample_index = (state->sample_index + 1) % ARRAY_SIZE(state->temp_samples);
278 	if (state->sample_count < ARRAY_SIZE(state->temp_samples))
279 		goto re_arm;
280 
281 	avg = 0;
282 	for (i = 0; i < ARRAY_SIZE(state->temp_samples); ++i)
283 		avg += state->temp_samples[i];
284 
285 	avg /= ARRAY_SIZE(state->temp_samples);
286 
287 	if (state->average > avg) {
288 		pr_warn("CPU%d: %s temperature is above threshold, cpu clock is throttled (total events = %lu)\n",
289 			this_cpu,
290 			state->level == CORE_LEVEL ? "Core" : "Package",
291 			state->count);
292 		state->rate_control_active = true;
293 	}
294 
295 	state->average = avg;
296 
297 re_arm:
298 	clear_therm_status_log(state->level);
299 	schedule_delayed_work_on(this_cpu, &state->therm_work, THERM_THROT_POLL_INTERVAL);
300 }
301 
302 /***
303  * therm_throt_process - Process thermal throttling event from interrupt
304  * @curr: Whether the condition is current or not (boolean), since the
305  *        thermal interrupt normally gets called both when the thermal
306  *        event begins and once the event has ended.
307  *
308  * This function is called by the thermal interrupt after the
309  * IRQ has been acknowledged.
310  *
311  * It will take care of rate limiting and printing messages to the syslog.
312  */
313 static void therm_throt_process(bool new_event, int event, int level)
314 {
315 	struct _thermal_state *state;
316 	unsigned int this_cpu = smp_processor_id();
317 	bool old_event;
318 	u64 now;
319 	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
320 
321 	now = get_jiffies_64();
322 	if (level == CORE_LEVEL) {
323 		if (event == THERMAL_THROTTLING_EVENT)
324 			state = &pstate->core_throttle;
325 		else if (event == POWER_LIMIT_EVENT)
326 			state = &pstate->core_power_limit;
327 		else
328 			return;
329 	} else if (level == PACKAGE_LEVEL) {
330 		if (event == THERMAL_THROTTLING_EVENT)
331 			state = &pstate->package_throttle;
332 		else if (event == POWER_LIMIT_EVENT)
333 			state = &pstate->package_power_limit;
334 		else
335 			return;
336 	} else
337 		return;
338 
339 	old_event = state->new_event;
340 	state->new_event = new_event;
341 
342 	if (new_event)
343 		state->count++;
344 
345 	if (event != THERMAL_THROTTLING_EVENT)
346 		return;
347 
348 	if (new_event && !state->last_interrupt_time) {
349 		bool hot;
350 		u8 temp;
351 
352 		get_therm_status(state->level, &hot, &temp);
353 		/*
354 		 * Ignore short temperature spike as the system is not close
355 		 * to PROCHOT. 10C offset is large enough to ignore. It is
356 		 * already dropped from the high threshold temperature.
357 		 */
358 		if (temp > 10)
359 			return;
360 
361 		state->baseline_temp = temp;
362 		state->last_interrupt_time = now;
363 		schedule_delayed_work_on(this_cpu, &state->therm_work, THERM_THROT_POLL_INTERVAL);
364 	} else if (old_event && state->last_interrupt_time) {
365 		unsigned long throttle_time;
366 
367 		throttle_time = jiffies_delta_to_msecs(now - state->last_interrupt_time);
368 		if (throttle_time > state->max_time_ms)
369 			state->max_time_ms = throttle_time;
370 		state->total_time_ms += throttle_time;
371 		state->last_interrupt_time = 0;
372 	}
373 }
374 
375 static int thresh_event_valid(int level, int event)
376 {
377 	struct _thermal_state *state;
378 	unsigned int this_cpu = smp_processor_id();
379 	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
380 	u64 now = get_jiffies_64();
381 
382 	if (level == PACKAGE_LEVEL)
383 		state = (event == 0) ? &pstate->pkg_thresh0 :
384 						&pstate->pkg_thresh1;
385 	else
386 		state = (event == 0) ? &pstate->core_thresh0 :
387 						&pstate->core_thresh1;
388 
389 	if (time_before64(now, state->next_check))
390 		return 0;
391 
392 	state->next_check = now + CHECK_INTERVAL;
393 
394 	return 1;
395 }
396 
397 static bool int_pln_enable;
398 static int __init int_pln_enable_setup(char *s)
399 {
400 	int_pln_enable = true;
401 
402 	return 1;
403 }
404 __setup("int_pln_enable", int_pln_enable_setup);
405 
406 #ifdef CONFIG_SYSFS
407 /* Add/Remove thermal_throttle interface for CPU device: */
408 static int thermal_throttle_add_dev(struct device *dev, unsigned int cpu)
409 {
410 	int err;
411 	struct cpuinfo_x86 *c = &cpu_data(cpu);
412 
413 	err = sysfs_create_group(&dev->kobj, &thermal_attr_group);
414 	if (err)
415 		return err;
416 
417 	if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable) {
418 		err = sysfs_add_file_to_group(&dev->kobj,
419 					      &dev_attr_core_power_limit_count.attr,
420 					      thermal_attr_group.name);
421 		if (err)
422 			goto del_group;
423 	}
424 
425 	if (cpu_has(c, X86_FEATURE_PTS)) {
426 		err = sysfs_add_file_to_group(&dev->kobj,
427 					      &dev_attr_package_throttle_count.attr,
428 					      thermal_attr_group.name);
429 		if (err)
430 			goto del_group;
431 
432 		err = sysfs_add_file_to_group(&dev->kobj,
433 					      &dev_attr_package_throttle_max_time_ms.attr,
434 					      thermal_attr_group.name);
435 		if (err)
436 			goto del_group;
437 
438 		err = sysfs_add_file_to_group(&dev->kobj,
439 					      &dev_attr_package_throttle_total_time_ms.attr,
440 					      thermal_attr_group.name);
441 		if (err)
442 			goto del_group;
443 
444 		if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable) {
445 			err = sysfs_add_file_to_group(&dev->kobj,
446 					&dev_attr_package_power_limit_count.attr,
447 					thermal_attr_group.name);
448 			if (err)
449 				goto del_group;
450 		}
451 	}
452 
453 	return 0;
454 
455 del_group:
456 	sysfs_remove_group(&dev->kobj, &thermal_attr_group);
457 
458 	return err;
459 }
460 
461 static void thermal_throttle_remove_dev(struct device *dev)
462 {
463 	sysfs_remove_group(&dev->kobj, &thermal_attr_group);
464 }
465 
466 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
467 static int thermal_throttle_online(unsigned int cpu)
468 {
469 	struct thermal_state *state = &per_cpu(thermal_state, cpu);
470 	struct device *dev = get_cpu_device(cpu);
471 	u32 l;
472 
473 	state->package_throttle.level = PACKAGE_LEVEL;
474 	state->core_throttle.level = CORE_LEVEL;
475 
476 	INIT_DELAYED_WORK(&state->package_throttle.therm_work, throttle_active_work);
477 	INIT_DELAYED_WORK(&state->core_throttle.therm_work, throttle_active_work);
478 
479 	/*
480 	 * The first CPU coming online will enable the HFI. Usually this causes
481 	 * hardware to issue an HFI thermal interrupt. Such interrupt will reach
482 	 * the CPU once we enable the thermal vector in the local APIC.
483 	 */
484 	intel_hfi_online(cpu);
485 
486 	/* Unmask the thermal vector after the above workqueues are initialized. */
487 	l = apic_read(APIC_LVTTHMR);
488 	apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
489 
490 	return thermal_throttle_add_dev(dev, cpu);
491 }
492 
493 static int thermal_throttle_offline(unsigned int cpu)
494 {
495 	struct thermal_state *state = &per_cpu(thermal_state, cpu);
496 	struct device *dev = get_cpu_device(cpu);
497 	u32 l;
498 
499 	/* Mask the thermal vector before draining evtl. pending work */
500 	l = apic_read(APIC_LVTTHMR);
501 	apic_write(APIC_LVTTHMR, l | APIC_LVT_MASKED);
502 
503 	intel_hfi_offline(cpu);
504 
505 	cancel_delayed_work_sync(&state->package_throttle.therm_work);
506 	cancel_delayed_work_sync(&state->core_throttle.therm_work);
507 
508 	state->package_throttle.rate_control_active = false;
509 	state->core_throttle.rate_control_active = false;
510 
511 	thermal_throttle_remove_dev(dev);
512 	return 0;
513 }
514 
515 static __init int thermal_throttle_init_device(void)
516 {
517 	int ret;
518 
519 	if (!atomic_read(&therm_throt_en))
520 		return 0;
521 
522 	intel_hfi_init();
523 
524 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/therm:online",
525 				thermal_throttle_online,
526 				thermal_throttle_offline);
527 	return ret < 0 ? ret : 0;
528 }
529 device_initcall(thermal_throttle_init_device);
530 
531 #endif /* CONFIG_SYSFS */
532 
533 static void notify_package_thresholds(__u64 msr_val)
534 {
535 	bool notify_thres_0 = false;
536 	bool notify_thres_1 = false;
537 
538 	if (!platform_thermal_package_notify)
539 		return;
540 
541 	/* lower threshold check */
542 	if (msr_val & THERM_LOG_THRESHOLD0)
543 		notify_thres_0 = true;
544 	/* higher threshold check */
545 	if (msr_val & THERM_LOG_THRESHOLD1)
546 		notify_thres_1 = true;
547 
548 	if (!notify_thres_0 && !notify_thres_1)
549 		return;
550 
551 	if (platform_thermal_package_rate_control &&
552 		platform_thermal_package_rate_control()) {
553 		/* Rate control is implemented in callback */
554 		platform_thermal_package_notify(msr_val);
555 		return;
556 	}
557 
558 	/* lower threshold reached */
559 	if (notify_thres_0 && thresh_event_valid(PACKAGE_LEVEL, 0))
560 		platform_thermal_package_notify(msr_val);
561 	/* higher threshold reached */
562 	if (notify_thres_1 && thresh_event_valid(PACKAGE_LEVEL, 1))
563 		platform_thermal_package_notify(msr_val);
564 }
565 
566 static void notify_thresholds(__u64 msr_val)
567 {
568 	/* check whether the interrupt handler is defined;
569 	 * otherwise simply return
570 	 */
571 	if (!platform_thermal_notify)
572 		return;
573 
574 	/* lower threshold reached */
575 	if ((msr_val & THERM_LOG_THRESHOLD0) &&
576 			thresh_event_valid(CORE_LEVEL, 0))
577 		platform_thermal_notify(msr_val);
578 	/* higher threshold reached */
579 	if ((msr_val & THERM_LOG_THRESHOLD1) &&
580 			thresh_event_valid(CORE_LEVEL, 1))
581 		platform_thermal_notify(msr_val);
582 }
583 
584 void __weak notify_hwp_interrupt(void)
585 {
586 	wrmsrl_safe(MSR_HWP_STATUS, 0);
587 }
588 
589 /* Thermal transition interrupt handler */
590 void intel_thermal_interrupt(void)
591 {
592 	__u64 msr_val;
593 
594 	if (static_cpu_has(X86_FEATURE_HWP))
595 		notify_hwp_interrupt();
596 
597 	rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
598 
599 	/* Check for violation of core thermal thresholds*/
600 	notify_thresholds(msr_val);
601 
602 	therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
603 			    THERMAL_THROTTLING_EVENT,
604 			    CORE_LEVEL);
605 
606 	if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
607 		therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
608 					POWER_LIMIT_EVENT,
609 					CORE_LEVEL);
610 
611 	if (this_cpu_has(X86_FEATURE_PTS)) {
612 		rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
613 		/* check violations of package thermal thresholds */
614 		notify_package_thresholds(msr_val);
615 		therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
616 					THERMAL_THROTTLING_EVENT,
617 					PACKAGE_LEVEL);
618 		if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
619 			therm_throt_process(msr_val &
620 					PACKAGE_THERM_STATUS_POWER_LIMIT,
621 					POWER_LIMIT_EVENT,
622 					PACKAGE_LEVEL);
623 
624 		if (this_cpu_has(X86_FEATURE_HFI))
625 			intel_hfi_process_event(msr_val &
626 						PACKAGE_THERM_STATUS_HFI_UPDATED);
627 	}
628 }
629 
630 /* Thermal monitoring depends on APIC, ACPI and clock modulation */
631 static int intel_thermal_supported(struct cpuinfo_x86 *c)
632 {
633 	if (!boot_cpu_has(X86_FEATURE_APIC))
634 		return 0;
635 	if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
636 		return 0;
637 	return 1;
638 }
639 
640 bool x86_thermal_enabled(void)
641 {
642 	return atomic_read(&therm_throt_en);
643 }
644 
645 void __init therm_lvt_init(void)
646 {
647 	/*
648 	 * This function is only called on boot CPU. Save the init thermal
649 	 * LVT value on BSP and use that value to restore APs' thermal LVT
650 	 * entry BIOS programmed later
651 	 */
652 	if (intel_thermal_supported(&boot_cpu_data))
653 		lvtthmr_init = apic_read(APIC_LVTTHMR);
654 }
655 
656 void intel_init_thermal(struct cpuinfo_x86 *c)
657 {
658 	unsigned int cpu = smp_processor_id();
659 	int tm2 = 0;
660 	u32 l, h;
661 
662 	if (!intel_thermal_supported(c))
663 		return;
664 
665 	/*
666 	 * First check if its enabled already, in which case there might
667 	 * be some SMM goo which handles it, so we can't even put a handler
668 	 * since it might be delivered via SMI already:
669 	 */
670 	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
671 
672 	h = lvtthmr_init;
673 	/*
674 	 * The initial value of thermal LVT entries on all APs always reads
675 	 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
676 	 * sequence to them and LVT registers are reset to 0s except for
677 	 * the mask bits which are set to 1s when APs receive INIT IPI.
678 	 * If BIOS takes over the thermal interrupt and sets its interrupt
679 	 * delivery mode to SMI (not fixed), it restores the value that the
680 	 * BIOS has programmed on AP based on BSP's info we saved since BIOS
681 	 * is always setting the same value for all threads/cores.
682 	 */
683 	if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
684 		apic_write(APIC_LVTTHMR, lvtthmr_init);
685 
686 
687 	if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
688 		if (system_state == SYSTEM_BOOTING)
689 			pr_debug("CPU%d: Thermal monitoring handled by SMI\n", cpu);
690 		return;
691 	}
692 
693 	/* early Pentium M models use different method for enabling TM2 */
694 	if (cpu_has(c, X86_FEATURE_TM2)) {
695 		if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
696 			rdmsr(MSR_THERM2_CTL, l, h);
697 			if (l & MSR_THERM2_CTL_TM_SELECT)
698 				tm2 = 1;
699 		} else if (l & MSR_IA32_MISC_ENABLE_TM2)
700 			tm2 = 1;
701 	}
702 
703 	/* We'll mask the thermal vector in the lapic till we're ready: */
704 	h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
705 	apic_write(APIC_LVTTHMR, h);
706 
707 	rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
708 	if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
709 		wrmsr(MSR_IA32_THERM_INTERRUPT,
710 			(l | (THERM_INT_LOW_ENABLE
711 			| THERM_INT_HIGH_ENABLE)) & ~THERM_INT_PLN_ENABLE, h);
712 	else 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
717 		wrmsr(MSR_IA32_THERM_INTERRUPT,
718 		      l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
719 
720 	if (cpu_has(c, X86_FEATURE_PTS)) {
721 		rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
722 		if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
723 			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
724 				(l | (PACKAGE_THERM_INT_LOW_ENABLE
725 				| PACKAGE_THERM_INT_HIGH_ENABLE))
726 				& ~PACKAGE_THERM_INT_PLN_ENABLE, h);
727 		else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
728 			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
729 				l | (PACKAGE_THERM_INT_LOW_ENABLE
730 				| PACKAGE_THERM_INT_HIGH_ENABLE
731 				| PACKAGE_THERM_INT_PLN_ENABLE), h);
732 		else
733 			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
734 			      l | (PACKAGE_THERM_INT_LOW_ENABLE
735 				| PACKAGE_THERM_INT_HIGH_ENABLE), h);
736 
737 		if (cpu_has(c, X86_FEATURE_HFI)) {
738 			rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
739 			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
740 			      l | PACKAGE_THERM_INT_HFI_ENABLE, h);
741 		}
742 	}
743 
744 	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
745 	wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
746 
747 	pr_info_once("CPU0: Thermal monitoring enabled (%s)\n",
748 		      tm2 ? "TM2" : "TM1");
749 
750 	/* enable thermal throttle processing */
751 	atomic_set(&therm_throt_en, 1);
752 }
753