1/*
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).
4 *
5 * This allows consistent reporting of CPU thermal throttle events.
6 *
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).
10 *
11 * Author: Dmitriy Zavin (dmitriyz@google.com)
12 *
13 * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
14 *          Inspired by Ross Biro's and Al Borchers' counter code.
15 */
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>
26
27#include <asm/processor.h>
28#include <asm/apic.h>
29#include <asm/idle.h>
30#include <asm/mce.h>
31#include <asm/msr.h>
32#include <asm/trace/irq_vectors.h>
33
34/* How long to wait between reporting thermal events */
35#define CHECK_INTERVAL		(300 * HZ)
36
37#define THERMAL_THROTTLING_EVENT	0
38#define POWER_LIMIT_EVENT		1
39
40/*
41 * Current thermal event state:
42 */
43struct _thermal_state {
44	bool			new_event;
45	int			event;
46	u64			next_check;
47	unsigned long		count;
48	unsigned long		last_count;
49};
50
51struct 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;
60};
61
62/* Callback to handle core threshold interrupts */
63int (*platform_thermal_notify)(__u64 msr_val);
64EXPORT_SYMBOL(platform_thermal_notify);
65
66/* Callback to handle core package threshold_interrupts */
67int (*platform_thermal_package_notify)(__u64 msr_val);
68EXPORT_SYMBOL_GPL(platform_thermal_package_notify);
69
70/* Callback support of rate control, return true, if
71 * callback has rate control */
72bool (*platform_thermal_package_rate_control)(void);
73EXPORT_SYMBOL_GPL(platform_thermal_package_rate_control);
74
75
76static DEFINE_PER_CPU(struct thermal_state, thermal_state);
77
78static atomic_t therm_throt_en	= ATOMIC_INIT(0);
79
80static u32 lvtthmr_init __read_mostly;
81
82#ifdef CONFIG_SYSFS
83#define define_therm_throt_device_one_ro(_name)				\
84	static DEVICE_ATTR(_name, 0444,					\
85			   therm_throt_device_show_##_name,		\
86				   NULL)				\
87
88#define define_therm_throt_device_show_func(event, name)		\
89									\
90static ssize_t therm_throt_device_show_##event##_##name(		\
91			struct device *dev,				\
92			struct device_attribute *attr,			\
93			char *buf)					\
94{									\
95	unsigned int cpu = dev->id;					\
96	ssize_t ret;							\
97									\
98	preempt_disable();	/* CPU hotplug */			\
99	if (cpu_online(cpu)) {						\
100		ret = sprintf(buf, "%lu\n",				\
101			      per_cpu(thermal_state, cpu).event.name);	\
102	} else								\
103		ret = 0;						\
104	preempt_enable();						\
105									\
106	return ret;							\
107}
108
109define_therm_throt_device_show_func(core_throttle, count);
110define_therm_throt_device_one_ro(core_throttle_count);
111
112define_therm_throt_device_show_func(core_power_limit, count);
113define_therm_throt_device_one_ro(core_power_limit_count);
114
115define_therm_throt_device_show_func(package_throttle, count);
116define_therm_throt_device_one_ro(package_throttle_count);
117
118define_therm_throt_device_show_func(package_power_limit, count);
119define_therm_throt_device_one_ro(package_power_limit_count);
120
121static struct attribute *thermal_throttle_attrs[] = {
122	&dev_attr_core_throttle_count.attr,
123	NULL
124};
125
126static struct attribute_group thermal_attr_group = {
127	.attrs	= thermal_throttle_attrs,
128	.name	= "thermal_throttle"
129};
130#endif /* CONFIG_SYSFS */
131
132#define CORE_LEVEL	0
133#define PACKAGE_LEVEL	1
134
135/***
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.
140 *
141 * This function is called by the thermal interrupt after the
142 * IRQ has been acknowledged.
143 *
144 * It will take care of rate limiting and printing messages to the syslog.
145 *
146 * Returns: 0 : Event should NOT be further logged, i.e. still in
147 *              "timeout" from previous log message.
148 *          1 : Event should be logged further, and a message has been
149 *              printed to the syslog.
150 */
151static int therm_throt_process(bool new_event, int event, int level)
152{
153	struct _thermal_state *state;
154	unsigned int this_cpu = smp_processor_id();
155	bool old_event;
156	u64 now;
157	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
158
159	now = get_jiffies_64();
160	if (level == CORE_LEVEL) {
161		if (event == THERMAL_THROTTLING_EVENT)
162			state = &pstate->core_throttle;
163		else if (event == POWER_LIMIT_EVENT)
164			state = &pstate->core_power_limit;
165		else
166			 return 0;
167	} else if (level == PACKAGE_LEVEL) {
168		if (event == THERMAL_THROTTLING_EVENT)
169			state = &pstate->package_throttle;
170		else if (event == POWER_LIMIT_EVENT)
171			state = &pstate->package_power_limit;
172		else
173			return 0;
174	} else
175		return 0;
176
177	old_event = state->new_event;
178	state->new_event = new_event;
179
180	if (new_event)
181		state->count++;
182
183	if (time_before64(now, state->next_check) &&
184			state->count != state->last_count)
185		return 0;
186
187	state->next_check = now + CHECK_INTERVAL;
188	state->last_count = state->count;
189
190	/* if we just entered the thermal event */
191	if (new_event) {
192		if (event == THERMAL_THROTTLING_EVENT)
193			printk(KERN_CRIT "CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",
194				this_cpu,
195				level == CORE_LEVEL ? "Core" : "Package",
196				state->count);
197		return 1;
198	}
199	if (old_event) {
200		if (event == THERMAL_THROTTLING_EVENT)
201			printk(KERN_INFO "CPU%d: %s temperature/speed normal\n",
202				this_cpu,
203				level == CORE_LEVEL ? "Core" : "Package");
204		return 1;
205	}
206
207	return 0;
208}
209
210static int thresh_event_valid(int level, int event)
211{
212	struct _thermal_state *state;
213	unsigned int this_cpu = smp_processor_id();
214	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
215	u64 now = get_jiffies_64();
216
217	if (level == PACKAGE_LEVEL)
218		state = (event == 0) ? &pstate->pkg_thresh0 :
219						&pstate->pkg_thresh1;
220	else
221		state = (event == 0) ? &pstate->core_thresh0 :
222						&pstate->core_thresh1;
223
224	if (time_before64(now, state->next_check))
225		return 0;
226
227	state->next_check = now + CHECK_INTERVAL;
228
229	return 1;
230}
231
232static bool int_pln_enable;
233static int __init int_pln_enable_setup(char *s)
234{
235	int_pln_enable = true;
236
237	return 1;
238}
239__setup("int_pln_enable", int_pln_enable_setup);
240
241#ifdef CONFIG_SYSFS
242/* Add/Remove thermal_throttle interface for CPU device: */
243static int thermal_throttle_add_dev(struct device *dev, unsigned int cpu)
244{
245	int err;
246	struct cpuinfo_x86 *c = &cpu_data(cpu);
247
248	err = sysfs_create_group(&dev->kobj, &thermal_attr_group);
249	if (err)
250		return err;
251
252	if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
253		err = sysfs_add_file_to_group(&dev->kobj,
254					      &dev_attr_core_power_limit_count.attr,
255					      thermal_attr_group.name);
256	if (cpu_has(c, X86_FEATURE_PTS)) {
257		err = sysfs_add_file_to_group(&dev->kobj,
258					      &dev_attr_package_throttle_count.attr,
259					      thermal_attr_group.name);
260		if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
261			err = sysfs_add_file_to_group(&dev->kobj,
262					&dev_attr_package_power_limit_count.attr,
263					thermal_attr_group.name);
264	}
265
266	return err;
267}
268
269static void thermal_throttle_remove_dev(struct device *dev)
270{
271	sysfs_remove_group(&dev->kobj, &thermal_attr_group);
272}
273
274/* Get notified when a cpu comes on/off. Be hotplug friendly. */
275static int
276thermal_throttle_cpu_callback(struct notifier_block *nfb,
277			      unsigned long action,
278			      void *hcpu)
279{
280	unsigned int cpu = (unsigned long)hcpu;
281	struct device *dev;
282	int err = 0;
283
284	dev = get_cpu_device(cpu);
285
286	switch (action) {
287	case CPU_UP_PREPARE:
288	case CPU_UP_PREPARE_FROZEN:
289		err = thermal_throttle_add_dev(dev, cpu);
290		WARN_ON(err);
291		break;
292	case CPU_UP_CANCELED:
293	case CPU_UP_CANCELED_FROZEN:
294	case CPU_DEAD:
295	case CPU_DEAD_FROZEN:
296		thermal_throttle_remove_dev(dev);
297		break;
298	}
299	return notifier_from_errno(err);
300}
301
302static struct notifier_block thermal_throttle_cpu_notifier =
303{
304	.notifier_call = thermal_throttle_cpu_callback,
305};
306
307static __init int thermal_throttle_init_device(void)
308{
309	unsigned int cpu = 0;
310	int err;
311
312	if (!atomic_read(&therm_throt_en))
313		return 0;
314
315	cpu_notifier_register_begin();
316
317	/* connect live CPUs to sysfs */
318	for_each_online_cpu(cpu) {
319		err = thermal_throttle_add_dev(get_cpu_device(cpu), cpu);
320		WARN_ON(err);
321	}
322
323	__register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
324	cpu_notifier_register_done();
325
326	return 0;
327}
328device_initcall(thermal_throttle_init_device);
329
330#endif /* CONFIG_SYSFS */
331
332static void notify_package_thresholds(__u64 msr_val)
333{
334	bool notify_thres_0 = false;
335	bool notify_thres_1 = false;
336
337	if (!platform_thermal_package_notify)
338		return;
339
340	/* lower threshold check */
341	if (msr_val & THERM_LOG_THRESHOLD0)
342		notify_thres_0 = true;
343	/* higher threshold check */
344	if (msr_val & THERM_LOG_THRESHOLD1)
345		notify_thres_1 = true;
346
347	if (!notify_thres_0 && !notify_thres_1)
348		return;
349
350	if (platform_thermal_package_rate_control &&
351		platform_thermal_package_rate_control()) {
352		/* Rate control is implemented in callback */
353		platform_thermal_package_notify(msr_val);
354		return;
355	}
356
357	/* lower threshold reached */
358	if (notify_thres_0 && thresh_event_valid(PACKAGE_LEVEL, 0))
359		platform_thermal_package_notify(msr_val);
360	/* higher threshold reached */
361	if (notify_thres_1 && thresh_event_valid(PACKAGE_LEVEL, 1))
362		platform_thermal_package_notify(msr_val);
363}
364
365static void notify_thresholds(__u64 msr_val)
366{
367	/* check whether the interrupt handler is defined;
368	 * otherwise simply return
369	 */
370	if (!platform_thermal_notify)
371		return;
372
373	/* lower threshold reached */
374	if ((msr_val & THERM_LOG_THRESHOLD0) &&
375			thresh_event_valid(CORE_LEVEL, 0))
376		platform_thermal_notify(msr_val);
377	/* higher threshold reached */
378	if ((msr_val & THERM_LOG_THRESHOLD1) &&
379			thresh_event_valid(CORE_LEVEL, 1))
380		platform_thermal_notify(msr_val);
381}
382
383/* Thermal transition interrupt handler */
384static void intel_thermal_interrupt(void)
385{
386	__u64 msr_val;
387
388	rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
389
390	/* Check for violation of core thermal thresholds*/
391	notify_thresholds(msr_val);
392
393	if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
394				THERMAL_THROTTLING_EVENT,
395				CORE_LEVEL) != 0)
396		mce_log_therm_throt_event(msr_val);
397
398	if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
399		therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
400					POWER_LIMIT_EVENT,
401					CORE_LEVEL);
402
403	if (this_cpu_has(X86_FEATURE_PTS)) {
404		rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
405		/* check violations of package thermal thresholds */
406		notify_package_thresholds(msr_val);
407		therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
408					THERMAL_THROTTLING_EVENT,
409					PACKAGE_LEVEL);
410		if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
411			therm_throt_process(msr_val &
412					PACKAGE_THERM_STATUS_POWER_LIMIT,
413					POWER_LIMIT_EVENT,
414					PACKAGE_LEVEL);
415	}
416}
417
418static void unexpected_thermal_interrupt(void)
419{
420	printk(KERN_ERR "CPU%d: Unexpected LVT thermal interrupt!\n",
421			smp_processor_id());
422}
423
424static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
425
426static inline void __smp_thermal_interrupt(void)
427{
428	inc_irq_stat(irq_thermal_count);
429	smp_thermal_vector();
430}
431
432asmlinkage __visible void smp_thermal_interrupt(struct pt_regs *regs)
433{
434	entering_irq();
435	__smp_thermal_interrupt();
436	exiting_ack_irq();
437}
438
439asmlinkage __visible void smp_trace_thermal_interrupt(struct pt_regs *regs)
440{
441	entering_irq();
442	trace_thermal_apic_entry(THERMAL_APIC_VECTOR);
443	__smp_thermal_interrupt();
444	trace_thermal_apic_exit(THERMAL_APIC_VECTOR);
445	exiting_ack_irq();
446}
447
448/* Thermal monitoring depends on APIC, ACPI and clock modulation */
449static int intel_thermal_supported(struct cpuinfo_x86 *c)
450{
451	if (!cpu_has_apic)
452		return 0;
453	if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
454		return 0;
455	return 1;
456}
457
458void __init mcheck_intel_therm_init(void)
459{
460	/*
461	 * This function is only called on boot CPU. Save the init thermal
462	 * LVT value on BSP and use that value to restore APs' thermal LVT
463	 * entry BIOS programmed later
464	 */
465	if (intel_thermal_supported(&boot_cpu_data))
466		lvtthmr_init = apic_read(APIC_LVTTHMR);
467}
468
469void intel_init_thermal(struct cpuinfo_x86 *c)
470{
471	unsigned int cpu = smp_processor_id();
472	int tm2 = 0;
473	u32 l, h;
474
475	if (!intel_thermal_supported(c))
476		return;
477
478	/*
479	 * First check if its enabled already, in which case there might
480	 * be some SMM goo which handles it, so we can't even put a handler
481	 * since it might be delivered via SMI already:
482	 */
483	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
484
485	h = lvtthmr_init;
486	/*
487	 * The initial value of thermal LVT entries on all APs always reads
488	 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
489	 * sequence to them and LVT registers are reset to 0s except for
490	 * the mask bits which are set to 1s when APs receive INIT IPI.
491	 * If BIOS takes over the thermal interrupt and sets its interrupt
492	 * delivery mode to SMI (not fixed), it restores the value that the
493	 * BIOS has programmed on AP based on BSP's info we saved since BIOS
494	 * is always setting the same value for all threads/cores.
495	 */
496	if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
497		apic_write(APIC_LVTTHMR, lvtthmr_init);
498
499
500	if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
501		if (system_state == SYSTEM_BOOTING)
502			printk(KERN_DEBUG "CPU%d: Thermal monitoring handled by SMI\n", cpu);
503		return;
504	}
505
506	/* Check whether a vector already exists */
507	if (h & APIC_VECTOR_MASK) {
508		printk(KERN_DEBUG
509		       "CPU%d: Thermal LVT vector (%#x) already installed\n",
510		       cpu, (h & APIC_VECTOR_MASK));
511		return;
512	}
513
514	/* early Pentium M models use different method for enabling TM2 */
515	if (cpu_has(c, X86_FEATURE_TM2)) {
516		if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
517			rdmsr(MSR_THERM2_CTL, l, h);
518			if (l & MSR_THERM2_CTL_TM_SELECT)
519				tm2 = 1;
520		} else if (l & MSR_IA32_MISC_ENABLE_TM2)
521			tm2 = 1;
522	}
523
524	/* We'll mask the thermal vector in the lapic till we're ready: */
525	h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
526	apic_write(APIC_LVTTHMR, h);
527
528	rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
529	if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
530		wrmsr(MSR_IA32_THERM_INTERRUPT,
531			(l | (THERM_INT_LOW_ENABLE
532			| THERM_INT_HIGH_ENABLE)) & ~THERM_INT_PLN_ENABLE, h);
533	else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
534		wrmsr(MSR_IA32_THERM_INTERRUPT,
535			l | (THERM_INT_LOW_ENABLE
536			| THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
537	else
538		wrmsr(MSR_IA32_THERM_INTERRUPT,
539		      l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
540
541	if (cpu_has(c, X86_FEATURE_PTS)) {
542		rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
543		if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
544			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
545				(l | (PACKAGE_THERM_INT_LOW_ENABLE
546				| PACKAGE_THERM_INT_HIGH_ENABLE))
547				& ~PACKAGE_THERM_INT_PLN_ENABLE, h);
548		else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
549			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
550				l | (PACKAGE_THERM_INT_LOW_ENABLE
551				| PACKAGE_THERM_INT_HIGH_ENABLE
552				| PACKAGE_THERM_INT_PLN_ENABLE), h);
553		else
554			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
555			      l | (PACKAGE_THERM_INT_LOW_ENABLE
556				| PACKAGE_THERM_INT_HIGH_ENABLE), h);
557	}
558
559	smp_thermal_vector = intel_thermal_interrupt;
560
561	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
562	wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
563
564	/* Unmask the thermal vector: */
565	l = apic_read(APIC_LVTTHMR);
566	apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
567
568	printk_once(KERN_INFO "CPU0: Thermal monitoring enabled (%s)\n",
569		       tm2 ? "TM2" : "TM1");
570
571	/* enable thermal throttle processing */
572	atomic_set(&therm_throt_en, 1);
573}
574