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
196static u64 therm_intr_core_clear_mask;
197static u64 therm_intr_pkg_clear_mask;
198
199static void thermal_intr_init_core_clear_mask(void)
200{
201 if (therm_intr_core_clear_mask)
202 return;
203
204 /*
205 * Reference: Intel SDM Volume 4
206 * "Table 2-2. IA-32 Architectural MSRs", MSR 0x19C
207 * IA32_THERM_STATUS.
208 */
209
210 /*
211 * Bit 1, 3, 5: CPUID.01H:EDX[22] = 1. This driver will not
212 * enable interrupts, when 0 as it checks for X86_FEATURE_ACPI.
213 */
214 therm_intr_core_clear_mask = (BIT(1) | BIT(3) | BIT(5));
215
216 /*
217 * Bit 7 and 9: Thermal Threshold #1 and #2 log
218 * If CPUID.01H:ECX[8] = 1
219 */
220 if (boot_cpu_has(X86_FEATURE_TM2))
221 therm_intr_core_clear_mask |= (BIT(7) | BIT(9));
222
223 /* Bit 11: Power Limitation log (R/WC0) If CPUID.06H:EAX[4] = 1 */
224 if (boot_cpu_has(X86_FEATURE_PLN))
225 therm_intr_core_clear_mask |= BIT(11);
226
227 /*
228 * Bit 13: Current Limit log (R/WC0) If CPUID.06H:EAX[7] = 1
229 * Bit 15: Cross Domain Limit log (R/WC0) If CPUID.06H:EAX[7] = 1
230 */
231 if (boot_cpu_has(X86_FEATURE_HWP))
232 therm_intr_core_clear_mask |= (BIT(13) | BIT(15));
233}
234
235static void thermal_intr_init_pkg_clear_mask(void)
236{
237 if (therm_intr_pkg_clear_mask)
238 return;
239
240 /*
241 * Reference: Intel SDM Volume 4
242 * "Table 2-2. IA-32 Architectural MSRs", MSR 0x1B1
243 * IA32_PACKAGE_THERM_STATUS.
244 */
245
246 /* All bits except BIT 26 depend on CPUID.06H: EAX[6] = 1 */
247 if (boot_cpu_has(X86_FEATURE_PTS))
248 therm_intr_pkg_clear_mask = (BIT(1) | BIT(3) | BIT(5) | BIT(7) | BIT(9) | BIT(11));
249
250 /*
251 * Intel SDM Volume 2A: Thermal and Power Management Leaf
252 * Bit 26: CPUID.06H: EAX[19] = 1
253 */
254 if (boot_cpu_has(X86_FEATURE_HFI))
255 therm_intr_pkg_clear_mask |= BIT(26);
256}
257
258/*
259 * Clear the bits in package thermal status register for bit = 1
260 * in bitmask
261 */
262void thermal_clear_package_intr_status(int level, u64 bit_mask)
263{
264 u64 msr_val;
265 int msr;
266
267 if (level == CORE_LEVEL) {
268 msr = MSR_IA32_THERM_STATUS;
269 msr_val = therm_intr_core_clear_mask;
270 } else {
271 msr = MSR_IA32_PACKAGE_THERM_STATUS;
272 msr_val = therm_intr_pkg_clear_mask;
273 }
274
275 msr_val &= ~bit_mask;
276 wrmsrq(msr, val: msr_val);
277}
278EXPORT_SYMBOL_GPL(thermal_clear_package_intr_status);
279
280static void get_therm_status(int level, bool *proc_hot, u8 *temp)
281{
282 int msr;
283 u64 msr_val;
284
285 if (level == CORE_LEVEL)
286 msr = MSR_IA32_THERM_STATUS;
287 else
288 msr = MSR_IA32_PACKAGE_THERM_STATUS;
289
290 rdmsrq(msr, msr_val);
291 if (msr_val & THERM_STATUS_PROCHOT_LOG)
292 *proc_hot = true;
293 else
294 *proc_hot = false;
295
296 *temp = (msr_val >> 16) & 0x7F;
297}
298
299static void __maybe_unused throttle_active_work(struct work_struct *work)
300{
301 struct _thermal_state *state = container_of(to_delayed_work(work),
302 struct _thermal_state, therm_work);
303 unsigned int i, avg, this_cpu = smp_processor_id();
304 u64 now = get_jiffies_64();
305 bool hot;
306 u8 temp;
307
308 get_therm_status(level: state->level, proc_hot: &hot, temp: &temp);
309 /* temperature value is offset from the max so lesser means hotter */
310 if (!hot && temp > state->baseline_temp) {
311 if (state->rate_control_active)
312 pr_info("CPU%d: %s temperature/speed normal (total events = %lu)\n",
313 this_cpu,
314 state->level == CORE_LEVEL ? "Core" : "Package",
315 state->count);
316
317 state->rate_control_active = false;
318 return;
319 }
320
321 if (time_before64(now, state->next_check) &&
322 state->rate_control_active)
323 goto re_arm;
324
325 state->next_check = now + CHECK_INTERVAL;
326
327 if (state->count != state->last_count) {
328 /* There was one new thermal interrupt */
329 state->last_count = state->count;
330 state->average = 0;
331 state->sample_count = 0;
332 state->sample_index = 0;
333 }
334
335 state->temp_samples[state->sample_index] = temp;
336 state->sample_count++;
337 state->sample_index = (state->sample_index + 1) % ARRAY_SIZE(state->temp_samples);
338 if (state->sample_count < ARRAY_SIZE(state->temp_samples))
339 goto re_arm;
340
341 avg = 0;
342 for (i = 0; i < ARRAY_SIZE(state->temp_samples); ++i)
343 avg += state->temp_samples[i];
344
345 avg /= ARRAY_SIZE(state->temp_samples);
346
347 if (state->average > avg) {
348 pr_warn("CPU%d: %s temperature is above threshold, cpu clock is throttled (total events = %lu)\n",
349 this_cpu,
350 state->level == CORE_LEVEL ? "Core" : "Package",
351 state->count);
352 state->rate_control_active = true;
353 }
354
355 state->average = avg;
356
357re_arm:
358 thermal_clear_package_intr_status(state->level, THERM_STATUS_PROCHOT_LOG);
359 schedule_delayed_work_on(cpu: this_cpu, dwork: &state->therm_work, THERM_THROT_POLL_INTERVAL);
360}
361
362/***
363 * therm_throt_process - Process thermal throttling event from interrupt
364 * @curr: Whether the condition is current or not (boolean), since the
365 * thermal interrupt normally gets called both when the thermal
366 * event begins and once the event has ended.
367 *
368 * This function is called by the thermal interrupt after the
369 * IRQ has been acknowledged.
370 *
371 * It will take care of rate limiting and printing messages to the syslog.
372 */
373static void therm_throt_process(bool new_event, int event, int level)
374{
375 struct _thermal_state *state;
376 unsigned int this_cpu = smp_processor_id();
377 bool old_event;
378 u64 now;
379 struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
380
381 now = get_jiffies_64();
382 if (level == CORE_LEVEL) {
383 if (event == THERMAL_THROTTLING_EVENT)
384 state = &pstate->core_throttle;
385 else if (event == POWER_LIMIT_EVENT)
386 state = &pstate->core_power_limit;
387 else
388 return;
389 } else if (level == PACKAGE_LEVEL) {
390 if (event == THERMAL_THROTTLING_EVENT)
391 state = &pstate->package_throttle;
392 else if (event == POWER_LIMIT_EVENT)
393 state = &pstate->package_power_limit;
394 else
395 return;
396 } else
397 return;
398
399 old_event = state->new_event;
400 state->new_event = new_event;
401
402 if (new_event)
403 state->count++;
404
405 if (event != THERMAL_THROTTLING_EVENT)
406 return;
407
408 if (new_event && !state->last_interrupt_time) {
409 bool hot;
410 u8 temp;
411
412 get_therm_status(level: state->level, proc_hot: &hot, temp: &temp);
413 /*
414 * Ignore short temperature spike as the system is not close
415 * to PROCHOT. 10C offset is large enough to ignore. It is
416 * already dropped from the high threshold temperature.
417 */
418 if (temp > 10)
419 return;
420
421 state->baseline_temp = temp;
422 state->last_interrupt_time = now;
423 schedule_delayed_work_on(cpu: this_cpu, dwork: &state->therm_work, THERM_THROT_POLL_INTERVAL);
424 } else if (old_event && state->last_interrupt_time) {
425 unsigned long throttle_time;
426
427 throttle_time = jiffies_delta_to_msecs(delta: now - state->last_interrupt_time);
428 if (throttle_time > state->max_time_ms)
429 state->max_time_ms = throttle_time;
430 state->total_time_ms += throttle_time;
431 state->last_interrupt_time = 0;
432 }
433}
434
435static int thresh_event_valid(int level, int event)
436{
437 struct _thermal_state *state;
438 unsigned int this_cpu = smp_processor_id();
439 struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
440 u64 now = get_jiffies_64();
441
442 if (level == PACKAGE_LEVEL)
443 state = (event == 0) ? &pstate->pkg_thresh0 :
444 &pstate->pkg_thresh1;
445 else
446 state = (event == 0) ? &pstate->core_thresh0 :
447 &pstate->core_thresh1;
448
449 if (time_before64(now, state->next_check))
450 return 0;
451
452 state->next_check = now + CHECK_INTERVAL;
453
454 return 1;
455}
456
457static bool int_pln_enable;
458static int __init int_pln_enable_setup(char *s)
459{
460 int_pln_enable = true;
461
462 return 1;
463}
464__setup("int_pln_enable", int_pln_enable_setup);
465
466#ifdef CONFIG_SYSFS
467/* Add/Remove thermal_throttle interface for CPU device: */
468static int thermal_throttle_add_dev(struct device *dev, unsigned int cpu)
469{
470 int err;
471 struct cpuinfo_x86 *c = &cpu_data(cpu);
472
473 err = sysfs_create_group(kobj: &dev->kobj, grp: &thermal_attr_group);
474 if (err)
475 return err;
476
477 if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable) {
478 err = sysfs_add_file_to_group(kobj: &dev->kobj,
479 attr: &dev_attr_core_power_limit_count.attr,
480 group: thermal_attr_group.name);
481 if (err)
482 goto del_group;
483 }
484
485 if (cpu_has(c, X86_FEATURE_PTS)) {
486 err = sysfs_add_file_to_group(kobj: &dev->kobj,
487 attr: &dev_attr_package_throttle_count.attr,
488 group: thermal_attr_group.name);
489 if (err)
490 goto del_group;
491
492 err = sysfs_add_file_to_group(kobj: &dev->kobj,
493 attr: &dev_attr_package_throttle_max_time_ms.attr,
494 group: thermal_attr_group.name);
495 if (err)
496 goto del_group;
497
498 err = sysfs_add_file_to_group(kobj: &dev->kobj,
499 attr: &dev_attr_package_throttle_total_time_ms.attr,
500 group: thermal_attr_group.name);
501 if (err)
502 goto del_group;
503
504 if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable) {
505 err = sysfs_add_file_to_group(kobj: &dev->kobj,
506 attr: &dev_attr_package_power_limit_count.attr,
507 group: thermal_attr_group.name);
508 if (err)
509 goto del_group;
510 }
511 }
512
513 return 0;
514
515del_group:
516 sysfs_remove_group(kobj: &dev->kobj, grp: &thermal_attr_group);
517
518 return err;
519}
520
521static void thermal_throttle_remove_dev(struct device *dev)
522{
523 sysfs_remove_group(kobj: &dev->kobj, grp: &thermal_attr_group);
524}
525
526/* Get notified when a cpu comes on/off. Be hotplug friendly. */
527static int thermal_throttle_online(unsigned int cpu)
528{
529 struct thermal_state *state = &per_cpu(thermal_state, cpu);
530 struct device *dev = get_cpu_device(cpu);
531 u32 l;
532
533 state->package_throttle.level = PACKAGE_LEVEL;
534 state->core_throttle.level = CORE_LEVEL;
535
536 INIT_DELAYED_WORK(&state->package_throttle.therm_work, throttle_active_work);
537 INIT_DELAYED_WORK(&state->core_throttle.therm_work, throttle_active_work);
538
539 /*
540 * The first CPU coming online will enable the HFI. Usually this causes
541 * hardware to issue an HFI thermal interrupt. Such interrupt will reach
542 * the CPU once we enable the thermal vector in the local APIC.
543 */
544 intel_hfi_online(cpu);
545
546 /* Unmask the thermal vector after the above workqueues are initialized. */
547 l = apic_read(APIC_LVTTHMR);
548 apic_write(APIC_LVTTHMR, val: l & ~APIC_LVT_MASKED);
549
550 return thermal_throttle_add_dev(dev, cpu);
551}
552
553static int thermal_throttle_offline(unsigned int cpu)
554{
555 struct thermal_state *state = &per_cpu(thermal_state, cpu);
556 struct device *dev = get_cpu_device(cpu);
557 u32 l;
558
559 /* Mask the thermal vector before draining evtl. pending work */
560 l = apic_read(APIC_LVTTHMR);
561 apic_write(APIC_LVTTHMR, val: l | APIC_LVT_MASKED);
562
563 intel_hfi_offline(cpu);
564
565 cancel_delayed_work_sync(dwork: &state->package_throttle.therm_work);
566 cancel_delayed_work_sync(dwork: &state->core_throttle.therm_work);
567
568 state->package_throttle.rate_control_active = false;
569 state->core_throttle.rate_control_active = false;
570
571 thermal_throttle_remove_dev(dev);
572 return 0;
573}
574
575static __init int thermal_throttle_init_device(void)
576{
577 int ret;
578
579 if (!atomic_read(v: &therm_throt_en))
580 return 0;
581
582 intel_hfi_init();
583
584 ret = cpuhp_setup_state(state: CPUHP_AP_ONLINE_DYN, name: "x86/therm:online",
585 startup: thermal_throttle_online,
586 teardown: thermal_throttle_offline);
587 return ret < 0 ? ret : 0;
588}
589device_initcall(thermal_throttle_init_device);
590
591#endif /* CONFIG_SYSFS */
592
593static void notify_package_thresholds(__u64 msr_val)
594{
595 bool notify_thres_0 = false;
596 bool notify_thres_1 = false;
597
598 if (!platform_thermal_package_notify)
599 return;
600
601 /* lower threshold check */
602 if (msr_val & THERM_LOG_THRESHOLD0)
603 notify_thres_0 = true;
604 /* higher threshold check */
605 if (msr_val & THERM_LOG_THRESHOLD1)
606 notify_thres_1 = true;
607
608 if (!notify_thres_0 && !notify_thres_1)
609 return;
610
611 if (platform_thermal_package_rate_control &&
612 platform_thermal_package_rate_control()) {
613 /* Rate control is implemented in callback */
614 platform_thermal_package_notify(msr_val);
615 return;
616 }
617
618 /* lower threshold reached */
619 if (notify_thres_0 && thresh_event_valid(PACKAGE_LEVEL, event: 0))
620 platform_thermal_package_notify(msr_val);
621 /* higher threshold reached */
622 if (notify_thres_1 && thresh_event_valid(PACKAGE_LEVEL, event: 1))
623 platform_thermal_package_notify(msr_val);
624}
625
626static void notify_thresholds(__u64 msr_val)
627{
628 /* check whether the interrupt handler is defined;
629 * otherwise simply return
630 */
631 if (!platform_thermal_notify)
632 return;
633
634 /* lower threshold reached */
635 if ((msr_val & THERM_LOG_THRESHOLD0) &&
636 thresh_event_valid(CORE_LEVEL, event: 0))
637 platform_thermal_notify(msr_val);
638 /* higher threshold reached */
639 if ((msr_val & THERM_LOG_THRESHOLD1) &&
640 thresh_event_valid(CORE_LEVEL, event: 1))
641 platform_thermal_notify(msr_val);
642}
643
644void __weak notify_hwp_interrupt(void)
645{
646 wrmsrq_safe(MSR_HWP_STATUS, val: 0);
647}
648
649/* Thermal transition interrupt handler */
650void intel_thermal_interrupt(void)
651{
652 __u64 msr_val;
653
654 if (static_cpu_has(X86_FEATURE_HWP))
655 notify_hwp_interrupt();
656
657 rdmsrq(MSR_IA32_THERM_STATUS, msr_val);
658
659 /* Check for violation of core thermal thresholds*/
660 notify_thresholds(msr_val);
661
662 therm_throt_process(new_event: msr_val & THERM_STATUS_PROCHOT,
663 THERMAL_THROTTLING_EVENT,
664 CORE_LEVEL);
665
666 if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
667 therm_throt_process(new_event: msr_val & THERM_STATUS_POWER_LIMIT,
668 POWER_LIMIT_EVENT,
669 CORE_LEVEL);
670
671 if (this_cpu_has(X86_FEATURE_PTS)) {
672 rdmsrq(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
673 /* check violations of package thermal thresholds */
674 notify_package_thresholds(msr_val);
675 therm_throt_process(new_event: msr_val & PACKAGE_THERM_STATUS_PROCHOT,
676 THERMAL_THROTTLING_EVENT,
677 PACKAGE_LEVEL);
678 if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
679 therm_throt_process(new_event: msr_val &
680 PACKAGE_THERM_STATUS_POWER_LIMIT,
681 POWER_LIMIT_EVENT,
682 PACKAGE_LEVEL);
683
684 if (this_cpu_has(X86_FEATURE_HFI))
685 intel_hfi_process_event(pkg_therm_status_msr_val: msr_val &
686 PACKAGE_THERM_STATUS_HFI_UPDATED);
687 }
688}
689
690/* Thermal monitoring depends on APIC, ACPI and clock modulation */
691static int intel_thermal_supported(struct cpuinfo_x86 *c)
692{
693 if (!boot_cpu_has(X86_FEATURE_APIC))
694 return 0;
695 if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
696 return 0;
697 return 1;
698}
699
700bool x86_thermal_enabled(void)
701{
702 return atomic_read(v: &therm_throt_en);
703}
704
705void __init therm_lvt_init(void)
706{
707 /*
708 * This function is only called on boot CPU. Save the init thermal
709 * LVT value on BSP and use that value to restore APs' thermal LVT
710 * entry BIOS programmed later
711 */
712 if (intel_thermal_supported(c: &boot_cpu_data))
713 lvtthmr_init = apic_read(APIC_LVTTHMR);
714}
715
716void intel_init_thermal(struct cpuinfo_x86 *c)
717{
718 unsigned int cpu = smp_processor_id();
719 int tm2 = 0;
720 u32 l, h;
721
722 if (!intel_thermal_supported(c))
723 return;
724
725 /*
726 * First check if its enabled already, in which case there might
727 * be some SMM goo which handles it, so we can't even put a handler
728 * since it might be delivered via SMI already:
729 */
730 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
731
732 h = lvtthmr_init;
733 /*
734 * The initial value of thermal LVT entries on all APs always reads
735 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
736 * sequence to them and LVT registers are reset to 0s except for
737 * the mask bits which are set to 1s when APs receive INIT IPI.
738 * If BIOS takes over the thermal interrupt and sets its interrupt
739 * delivery mode to SMI (not fixed), it restores the value that the
740 * BIOS has programmed on AP based on BSP's info we saved since BIOS
741 * is always setting the same value for all threads/cores.
742 */
743 if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
744 apic_write(APIC_LVTTHMR, val: lvtthmr_init);
745
746
747 if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
748 if (system_state == SYSTEM_BOOTING)
749 pr_debug("CPU%d: Thermal monitoring handled by SMI\n", cpu);
750 return;
751 }
752
753 /* early Pentium M models use different method for enabling TM2 */
754 if (cpu_has(c, X86_FEATURE_TM2)) {
755 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
756 rdmsr(MSR_THERM2_CTL, l, h);
757 if (l & MSR_THERM2_CTL_TM_SELECT)
758 tm2 = 1;
759 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
760 tm2 = 1;
761 }
762
763 /* We'll mask the thermal vector in the lapic till we're ready: */
764 h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
765 apic_write(APIC_LVTTHMR, val: h);
766
767 thermal_intr_init_core_clear_mask();
768 thermal_intr_init_pkg_clear_mask();
769
770 rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
771 if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
772 wrmsr(MSR_IA32_THERM_INTERRUPT,
773 low: (l | (THERM_INT_LOW_ENABLE
774 | THERM_INT_HIGH_ENABLE)) & ~THERM_INT_PLN_ENABLE, high: h);
775 else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
776 wrmsr(MSR_IA32_THERM_INTERRUPT,
777 low: l | (THERM_INT_LOW_ENABLE
778 | THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), high: h);
779 else
780 wrmsr(MSR_IA32_THERM_INTERRUPT,
781 low: l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), high: h);
782
783 if (cpu_has(c, X86_FEATURE_PTS)) {
784 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
785 if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
786 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
787 low: (l | (PACKAGE_THERM_INT_LOW_ENABLE
788 | PACKAGE_THERM_INT_HIGH_ENABLE))
789 & ~PACKAGE_THERM_INT_PLN_ENABLE, high: h);
790 else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
791 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
792 low: l | (PACKAGE_THERM_INT_LOW_ENABLE
793 | PACKAGE_THERM_INT_HIGH_ENABLE
794 | PACKAGE_THERM_INT_PLN_ENABLE), high: h);
795 else
796 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
797 low: l | (PACKAGE_THERM_INT_LOW_ENABLE
798 | PACKAGE_THERM_INT_HIGH_ENABLE), high: h);
799
800 if (cpu_has(c, X86_FEATURE_HFI)) {
801 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
802 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
803 low: l | PACKAGE_THERM_INT_HFI_ENABLE, high: h);
804 }
805 }
806
807 rdmsr(MSR_IA32_MISC_ENABLE, l, h);
808 wrmsr(MSR_IA32_MISC_ENABLE, low: l | MSR_IA32_MISC_ENABLE_TM1, high: h);
809
810 pr_info_once("CPU0: Thermal monitoring enabled (%s)\n",
811 tm2 ? "TM2" : "TM1");
812
813 /* enable thermal throttle processing */
814 atomic_set(v: &therm_throt_en, i: 1);
815}
816