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
33/* How long to wait between reporting thermal events */
34#define CHECK_INTERVAL		(300 * HZ)
35
36#define THERMAL_THROTTLING_EVENT	0
37#define POWER_LIMIT_EVENT		1
38
39/*
40 * Current thermal event state:
41 */
42struct _thermal_state {
43	bool			new_event;
44	int			event;
45	u64			next_check;
46	unsigned long		count;
47	unsigned long		last_count;
48};
49
50struct thermal_state {
51	struct _thermal_state core_throttle;
52	struct _thermal_state core_power_limit;
53	struct _thermal_state package_throttle;
54	struct _thermal_state package_power_limit;
55	struct _thermal_state core_thresh0;
56	struct _thermal_state core_thresh1;
57};
58
59/* Callback to handle core threshold interrupts */
60int (*platform_thermal_notify)(__u64 msr_val);
61EXPORT_SYMBOL(platform_thermal_notify);
62
63static DEFINE_PER_CPU(struct thermal_state, thermal_state);
64
65static atomic_t therm_throt_en	= ATOMIC_INIT(0);
66
67static u32 lvtthmr_init __read_mostly;
68
69#ifdef CONFIG_SYSFS
70#define define_therm_throt_device_one_ro(_name)				\
71	static DEVICE_ATTR(_name, 0444,					\
72			   therm_throt_device_show_##_name,		\
73				   NULL)				\
74
75#define define_therm_throt_device_show_func(event, name)		\
76									\
77static ssize_t therm_throt_device_show_##event##_##name(		\
78			struct device *dev,				\
79			struct device_attribute *attr,			\
80			char *buf)					\
81{									\
82	unsigned int cpu = dev->id;					\
83	ssize_t ret;							\
84									\
85	preempt_disable();	/* CPU hotplug */			\
86	if (cpu_online(cpu)) {						\
87		ret = sprintf(buf, "%lu\n",				\
88			      per_cpu(thermal_state, cpu).event.name);	\
89	} else								\
90		ret = 0;						\
91	preempt_enable();						\
92									\
93	return ret;							\
94}
95
96define_therm_throt_device_show_func(core_throttle, count);
97define_therm_throt_device_one_ro(core_throttle_count);
98
99define_therm_throt_device_show_func(core_power_limit, count);
100define_therm_throt_device_one_ro(core_power_limit_count);
101
102define_therm_throt_device_show_func(package_throttle, count);
103define_therm_throt_device_one_ro(package_throttle_count);
104
105define_therm_throt_device_show_func(package_power_limit, count);
106define_therm_throt_device_one_ro(package_power_limit_count);
107
108static struct attribute *thermal_throttle_attrs[] = {
109	&dev_attr_core_throttle_count.attr,
110	NULL
111};
112
113static struct attribute_group thermal_attr_group = {
114	.attrs	= thermal_throttle_attrs,
115	.name	= "thermal_throttle"
116};
117#endif /* CONFIG_SYSFS */
118
119#define CORE_LEVEL	0
120#define PACKAGE_LEVEL	1
121
122/***
123 * therm_throt_process - Process thermal throttling event from interrupt
124 * @curr: Whether the condition is current or not (boolean), since the
125 *        thermal interrupt normally gets called both when the thermal
126 *        event begins and once the event has ended.
127 *
128 * This function is called by the thermal interrupt after the
129 * IRQ has been acknowledged.
130 *
131 * It will take care of rate limiting and printing messages to the syslog.
132 *
133 * Returns: 0 : Event should NOT be further logged, i.e. still in
134 *              "timeout" from previous log message.
135 *          1 : Event should be logged further, and a message has been
136 *              printed to the syslog.
137 */
138static int therm_throt_process(bool new_event, int event, int level)
139{
140	struct _thermal_state *state;
141	unsigned int this_cpu = smp_processor_id();
142	bool old_event;
143	u64 now;
144	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
145
146	now = get_jiffies_64();
147	if (level == CORE_LEVEL) {
148		if (event == THERMAL_THROTTLING_EVENT)
149			state = &pstate->core_throttle;
150		else if (event == POWER_LIMIT_EVENT)
151			state = &pstate->core_power_limit;
152		else
153			 return 0;
154	} else if (level == PACKAGE_LEVEL) {
155		if (event == THERMAL_THROTTLING_EVENT)
156			state = &pstate->package_throttle;
157		else if (event == POWER_LIMIT_EVENT)
158			state = &pstate->package_power_limit;
159		else
160			return 0;
161	} else
162		return 0;
163
164	old_event = state->new_event;
165	state->new_event = new_event;
166
167	if (new_event)
168		state->count++;
169
170	if (time_before64(now, state->next_check) &&
171			state->count != state->last_count)
172		return 0;
173
174	state->next_check = now + CHECK_INTERVAL;
175	state->last_count = state->count;
176
177	/* if we just entered the thermal event */
178	if (new_event) {
179		if (event == THERMAL_THROTTLING_EVENT)
180			printk(KERN_CRIT "CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",
181				this_cpu,
182				level == CORE_LEVEL ? "Core" : "Package",
183				state->count);
184		else
185			printk(KERN_CRIT "CPU%d: %s power limit notification (total events = %lu)\n",
186				this_cpu,
187				level == CORE_LEVEL ? "Core" : "Package",
188				state->count);
189		return 1;
190	}
191	if (old_event) {
192		if (event == THERMAL_THROTTLING_EVENT)
193			printk(KERN_INFO "CPU%d: %s temperature/speed normal\n",
194				this_cpu,
195				level == CORE_LEVEL ? "Core" : "Package");
196		else
197			printk(KERN_INFO "CPU%d: %s power limit normal\n",
198				this_cpu,
199				level == CORE_LEVEL ? "Core" : "Package");
200		return 1;
201	}
202
203	return 0;
204}
205
206static int thresh_event_valid(int event)
207{
208	struct _thermal_state *state;
209	unsigned int this_cpu = smp_processor_id();
210	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
211	u64 now = get_jiffies_64();
212
213	state = (event == 0) ? &pstate->core_thresh0 : &pstate->core_thresh1;
214
215	if (time_before64(now, state->next_check))
216		return 0;
217
218	state->next_check = now + CHECK_INTERVAL;
219	return 1;
220}
221
222#ifdef CONFIG_SYSFS
223/* Add/Remove thermal_throttle interface for CPU device: */
224static __cpuinit int thermal_throttle_add_dev(struct device *dev,
225				unsigned int cpu)
226{
227	int err;
228	struct cpuinfo_x86 *c = &cpu_data(cpu);
229
230	err = sysfs_create_group(&dev->kobj, &thermal_attr_group);
231	if (err)
232		return err;
233
234	if (cpu_has(c, X86_FEATURE_PLN))
235		err = sysfs_add_file_to_group(&dev->kobj,
236					      &dev_attr_core_power_limit_count.attr,
237					      thermal_attr_group.name);
238	if (cpu_has(c, X86_FEATURE_PTS)) {
239		err = sysfs_add_file_to_group(&dev->kobj,
240					      &dev_attr_package_throttle_count.attr,
241					      thermal_attr_group.name);
242		if (cpu_has(c, X86_FEATURE_PLN))
243			err = sysfs_add_file_to_group(&dev->kobj,
244					&dev_attr_package_power_limit_count.attr,
245					thermal_attr_group.name);
246	}
247
248	return err;
249}
250
251static __cpuinit void thermal_throttle_remove_dev(struct device *dev)
252{
253	sysfs_remove_group(&dev->kobj, &thermal_attr_group);
254}
255
256/* Mutex protecting device creation against CPU hotplug: */
257static DEFINE_MUTEX(therm_cpu_lock);
258
259/* Get notified when a cpu comes on/off. Be hotplug friendly. */
260static __cpuinit int
261thermal_throttle_cpu_callback(struct notifier_block *nfb,
262			      unsigned long action,
263			      void *hcpu)
264{
265	unsigned int cpu = (unsigned long)hcpu;
266	struct device *dev;
267	int err = 0;
268
269	dev = get_cpu_device(cpu);
270
271	switch (action) {
272	case CPU_UP_PREPARE:
273	case CPU_UP_PREPARE_FROZEN:
274		mutex_lock(&therm_cpu_lock);
275		err = thermal_throttle_add_dev(dev, cpu);
276		mutex_unlock(&therm_cpu_lock);
277		WARN_ON(err);
278		break;
279	case CPU_UP_CANCELED:
280	case CPU_UP_CANCELED_FROZEN:
281	case CPU_DEAD:
282	case CPU_DEAD_FROZEN:
283		mutex_lock(&therm_cpu_lock);
284		thermal_throttle_remove_dev(dev);
285		mutex_unlock(&therm_cpu_lock);
286		break;
287	}
288	return notifier_from_errno(err);
289}
290
291static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
292{
293	.notifier_call = thermal_throttle_cpu_callback,
294};
295
296static __init int thermal_throttle_init_device(void)
297{
298	unsigned int cpu = 0;
299	int err;
300
301	if (!atomic_read(&therm_throt_en))
302		return 0;
303
304	register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
305
306#ifdef CONFIG_HOTPLUG_CPU
307	mutex_lock(&therm_cpu_lock);
308#endif
309	/* connect live CPUs to sysfs */
310	for_each_online_cpu(cpu) {
311		err = thermal_throttle_add_dev(get_cpu_device(cpu), cpu);
312		WARN_ON(err);
313	}
314#ifdef CONFIG_HOTPLUG_CPU
315	mutex_unlock(&therm_cpu_lock);
316#endif
317
318	return 0;
319}
320device_initcall(thermal_throttle_init_device);
321
322#endif /* CONFIG_SYSFS */
323
324static void notify_thresholds(__u64 msr_val)
325{
326	/* check whether the interrupt handler is defined;
327	 * otherwise simply return
328	 */
329	if (!platform_thermal_notify)
330		return;
331
332	/* lower threshold reached */
333	if ((msr_val & THERM_LOG_THRESHOLD0) &&	thresh_event_valid(0))
334		platform_thermal_notify(msr_val);
335	/* higher threshold reached */
336	if ((msr_val & THERM_LOG_THRESHOLD1) && thresh_event_valid(1))
337		platform_thermal_notify(msr_val);
338}
339
340/* Thermal transition interrupt handler */
341static void intel_thermal_interrupt(void)
342{
343	__u64 msr_val;
344
345	rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
346
347	/* Check for violation of core thermal thresholds*/
348	notify_thresholds(msr_val);
349
350	if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
351				THERMAL_THROTTLING_EVENT,
352				CORE_LEVEL) != 0)
353		mce_log_therm_throt_event(msr_val);
354
355	if (this_cpu_has(X86_FEATURE_PLN))
356		therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
357					POWER_LIMIT_EVENT,
358					CORE_LEVEL);
359
360	if (this_cpu_has(X86_FEATURE_PTS)) {
361		rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
362		therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
363					THERMAL_THROTTLING_EVENT,
364					PACKAGE_LEVEL);
365		if (this_cpu_has(X86_FEATURE_PLN))
366			therm_throt_process(msr_val &
367					PACKAGE_THERM_STATUS_POWER_LIMIT,
368					POWER_LIMIT_EVENT,
369					PACKAGE_LEVEL);
370	}
371}
372
373static void unexpected_thermal_interrupt(void)
374{
375	printk(KERN_ERR "CPU%d: Unexpected LVT thermal interrupt!\n",
376			smp_processor_id());
377}
378
379static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
380
381asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
382{
383	irq_enter();
384	exit_idle();
385	inc_irq_stat(irq_thermal_count);
386	smp_thermal_vector();
387	irq_exit();
388	/* Ack only at the end to avoid potential reentry */
389	ack_APIC_irq();
390}
391
392/* Thermal monitoring depends on APIC, ACPI and clock modulation */
393static int intel_thermal_supported(struct cpuinfo_x86 *c)
394{
395	if (!cpu_has_apic)
396		return 0;
397	if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
398		return 0;
399	return 1;
400}
401
402void __init mcheck_intel_therm_init(void)
403{
404	/*
405	 * This function is only called on boot CPU. Save the init thermal
406	 * LVT value on BSP and use that value to restore APs' thermal LVT
407	 * entry BIOS programmed later
408	 */
409	if (intel_thermal_supported(&boot_cpu_data))
410		lvtthmr_init = apic_read(APIC_LVTTHMR);
411}
412
413void intel_init_thermal(struct cpuinfo_x86 *c)
414{
415	unsigned int cpu = smp_processor_id();
416	int tm2 = 0;
417	u32 l, h;
418
419	if (!intel_thermal_supported(c))
420		return;
421
422	/*
423	 * First check if its enabled already, in which case there might
424	 * be some SMM goo which handles it, so we can't even put a handler
425	 * since it might be delivered via SMI already:
426	 */
427	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
428
429	h = lvtthmr_init;
430	/*
431	 * The initial value of thermal LVT entries on all APs always reads
432	 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
433	 * sequence to them and LVT registers are reset to 0s except for
434	 * the mask bits which are set to 1s when APs receive INIT IPI.
435	 * If BIOS takes over the thermal interrupt and sets its interrupt
436	 * delivery mode to SMI (not fixed), it restores the value that the
437	 * BIOS has programmed on AP based on BSP's info we saved since BIOS
438	 * is always setting the same value for all threads/cores.
439	 */
440	if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
441		apic_write(APIC_LVTTHMR, lvtthmr_init);
442
443
444	if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
445		printk(KERN_DEBUG
446		       "CPU%d: Thermal monitoring handled by SMI\n", cpu);
447		return;
448	}
449
450	/* Check whether a vector already exists */
451	if (h & APIC_VECTOR_MASK) {
452		printk(KERN_DEBUG
453		       "CPU%d: Thermal LVT vector (%#x) already installed\n",
454		       cpu, (h & APIC_VECTOR_MASK));
455		return;
456	}
457
458	/* early Pentium M models use different method for enabling TM2 */
459	if (cpu_has(c, X86_FEATURE_TM2)) {
460		if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
461			rdmsr(MSR_THERM2_CTL, l, h);
462			if (l & MSR_THERM2_CTL_TM_SELECT)
463				tm2 = 1;
464		} else if (l & MSR_IA32_MISC_ENABLE_TM2)
465			tm2 = 1;
466	}
467
468	/* We'll mask the thermal vector in the lapic till we're ready: */
469	h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
470	apic_write(APIC_LVTTHMR, h);
471
472	rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
473	if (cpu_has(c, X86_FEATURE_PLN))
474		wrmsr(MSR_IA32_THERM_INTERRUPT,
475		      l | (THERM_INT_LOW_ENABLE
476			| THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
477	else
478		wrmsr(MSR_IA32_THERM_INTERRUPT,
479		      l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
480
481	if (cpu_has(c, X86_FEATURE_PTS)) {
482		rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
483		if (cpu_has(c, X86_FEATURE_PLN))
484			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
485			      l | (PACKAGE_THERM_INT_LOW_ENABLE
486				| PACKAGE_THERM_INT_HIGH_ENABLE
487				| PACKAGE_THERM_INT_PLN_ENABLE), h);
488		else
489			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
490			      l | (PACKAGE_THERM_INT_LOW_ENABLE
491				| PACKAGE_THERM_INT_HIGH_ENABLE), h);
492	}
493
494	smp_thermal_vector = intel_thermal_interrupt;
495
496	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
497	wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
498
499	/* Unmask the thermal vector: */
500	l = apic_read(APIC_LVTTHMR);
501	apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
502
503	printk_once(KERN_INFO "CPU0: Thermal monitoring enabled (%s)\n",
504		       tm2 ? "TM2" : "TM1");
505
506	/* enable thermal throttle processing */
507	atomic_set(&therm_throt_en, 1);
508}
509