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/sysdev.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/system.h>
29 #include <asm/apic.h>
30 #include <asm/idle.h>
31 #include <asm/mce.h>
32 #include <asm/msr.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  */
43 struct _thermal_state {
44 	bool			new_event;
45 	int			event;
46 	u64			next_check;
47 	unsigned long		count;
48 	unsigned long		last_count;
49 };
50 
51 struct 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 };
59 
60 /* Callback to handle core threshold interrupts */
61 int (*platform_thermal_notify)(__u64 msr_val);
62 EXPORT_SYMBOL(platform_thermal_notify);
63 
64 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
65 
66 static atomic_t therm_throt_en	= ATOMIC_INIT(0);
67 
68 static u32 lvtthmr_init __read_mostly;
69 
70 #ifdef CONFIG_SYSFS
71 #define define_therm_throt_sysdev_one_ro(_name)				\
72 	static SYSDEV_ATTR(_name, 0444,					\
73 			   therm_throt_sysdev_show_##_name,		\
74 				   NULL)				\
75 
76 #define define_therm_throt_sysdev_show_func(event, name)		\
77 									\
78 static ssize_t therm_throt_sysdev_show_##event##_##name(		\
79 			struct sys_device *dev,				\
80 			struct sysdev_attribute *attr,			\
81 			char *buf)					\
82 {									\
83 	unsigned int cpu = dev->id;					\
84 	ssize_t ret;							\
85 									\
86 	preempt_disable();	/* CPU hotplug */			\
87 	if (cpu_online(cpu)) {						\
88 		ret = sprintf(buf, "%lu\n",				\
89 			      per_cpu(thermal_state, cpu).event.name);	\
90 	} else								\
91 		ret = 0;						\
92 	preempt_enable();						\
93 									\
94 	return ret;							\
95 }
96 
97 define_therm_throt_sysdev_show_func(core_throttle, count);
98 define_therm_throt_sysdev_one_ro(core_throttle_count);
99 
100 define_therm_throt_sysdev_show_func(core_power_limit, count);
101 define_therm_throt_sysdev_one_ro(core_power_limit_count);
102 
103 define_therm_throt_sysdev_show_func(package_throttle, count);
104 define_therm_throt_sysdev_one_ro(package_throttle_count);
105 
106 define_therm_throt_sysdev_show_func(package_power_limit, count);
107 define_therm_throt_sysdev_one_ro(package_power_limit_count);
108 
109 static struct attribute *thermal_throttle_attrs[] = {
110 	&attr_core_throttle_count.attr,
111 	NULL
112 };
113 
114 static struct attribute_group thermal_attr_group = {
115 	.attrs	= thermal_throttle_attrs,
116 	.name	= "thermal_throttle"
117 };
118 #endif /* CONFIG_SYSFS */
119 
120 #define CORE_LEVEL	0
121 #define PACKAGE_LEVEL	1
122 
123 /***
124  * therm_throt_process - Process thermal throttling event from interrupt
125  * @curr: Whether the condition is current or not (boolean), since the
126  *        thermal interrupt normally gets called both when the thermal
127  *        event begins and once the event has ended.
128  *
129  * This function is called by the thermal interrupt after the
130  * IRQ has been acknowledged.
131  *
132  * It will take care of rate limiting and printing messages to the syslog.
133  *
134  * Returns: 0 : Event should NOT be further logged, i.e. still in
135  *              "timeout" from previous log message.
136  *          1 : Event should be logged further, and a message has been
137  *              printed to the syslog.
138  */
therm_throt_process(bool new_event,int event,int level)139 static int therm_throt_process(bool new_event, int event, int level)
140 {
141 	struct _thermal_state *state;
142 	unsigned int this_cpu = smp_processor_id();
143 	bool old_event;
144 	u64 now;
145 	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
146 
147 	now = get_jiffies_64();
148 	if (level == CORE_LEVEL) {
149 		if (event == THERMAL_THROTTLING_EVENT)
150 			state = &pstate->core_throttle;
151 		else if (event == POWER_LIMIT_EVENT)
152 			state = &pstate->core_power_limit;
153 		else
154 			 return 0;
155 	} else if (level == PACKAGE_LEVEL) {
156 		if (event == THERMAL_THROTTLING_EVENT)
157 			state = &pstate->package_throttle;
158 		else if (event == POWER_LIMIT_EVENT)
159 			state = &pstate->package_power_limit;
160 		else
161 			return 0;
162 	} else
163 		return 0;
164 
165 	old_event = state->new_event;
166 	state->new_event = new_event;
167 
168 	if (new_event)
169 		state->count++;
170 
171 	if (time_before64(now, state->next_check) &&
172 			state->count != state->last_count)
173 		return 0;
174 
175 	state->next_check = now + CHECK_INTERVAL;
176 	state->last_count = state->count;
177 
178 	/* if we just entered the thermal event */
179 	if (new_event) {
180 		if (event == THERMAL_THROTTLING_EVENT)
181 			printk(KERN_CRIT "CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",
182 				this_cpu,
183 				level == CORE_LEVEL ? "Core" : "Package",
184 				state->count);
185 		else
186 			printk(KERN_CRIT "CPU%d: %s power limit notification (total events = %lu)\n",
187 				this_cpu,
188 				level == CORE_LEVEL ? "Core" : "Package",
189 				state->count);
190 
191 		add_taint(TAINT_MACHINE_CHECK);
192 		return 1;
193 	}
194 	if (old_event) {
195 		if (event == THERMAL_THROTTLING_EVENT)
196 			printk(KERN_INFO "CPU%d: %s temperature/speed normal\n",
197 				this_cpu,
198 				level == CORE_LEVEL ? "Core" : "Package");
199 		else
200 			printk(KERN_INFO "CPU%d: %s power limit normal\n",
201 				this_cpu,
202 				level == CORE_LEVEL ? "Core" : "Package");
203 		return 1;
204 	}
205 
206 	return 0;
207 }
208 
thresh_event_valid(int event)209 static int thresh_event_valid(int event)
210 {
211 	struct _thermal_state *state;
212 	unsigned int this_cpu = smp_processor_id();
213 	struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
214 	u64 now = get_jiffies_64();
215 
216 	state = (event == 0) ? &pstate->core_thresh0 : &pstate->core_thresh1;
217 
218 	if (time_before64(now, state->next_check))
219 		return 0;
220 
221 	state->next_check = now + CHECK_INTERVAL;
222 	return 1;
223 }
224 
225 #ifdef CONFIG_SYSFS
226 /* Add/Remove thermal_throttle interface for CPU device: */
thermal_throttle_add_dev(struct sys_device * sys_dev,unsigned int cpu)227 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev,
228 				unsigned int cpu)
229 {
230 	int err;
231 	struct cpuinfo_x86 *c = &cpu_data(cpu);
232 
233 	err = sysfs_create_group(&sys_dev->kobj, &thermal_attr_group);
234 	if (err)
235 		return err;
236 
237 	if (cpu_has(c, X86_FEATURE_PLN))
238 		err = sysfs_add_file_to_group(&sys_dev->kobj,
239 					      &attr_core_power_limit_count.attr,
240 					      thermal_attr_group.name);
241 	if (cpu_has(c, X86_FEATURE_PTS)) {
242 		err = sysfs_add_file_to_group(&sys_dev->kobj,
243 					      &attr_package_throttle_count.attr,
244 					      thermal_attr_group.name);
245 		if (cpu_has(c, X86_FEATURE_PLN))
246 			err = sysfs_add_file_to_group(&sys_dev->kobj,
247 					&attr_package_power_limit_count.attr,
248 					thermal_attr_group.name);
249 	}
250 
251 	return err;
252 }
253 
thermal_throttle_remove_dev(struct sys_device * sys_dev)254 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
255 {
256 	sysfs_remove_group(&sys_dev->kobj, &thermal_attr_group);
257 }
258 
259 /* Mutex protecting device creation against CPU hotplug: */
260 static DEFINE_MUTEX(therm_cpu_lock);
261 
262 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
263 static __cpuinit int
thermal_throttle_cpu_callback(struct notifier_block * nfb,unsigned long action,void * hcpu)264 thermal_throttle_cpu_callback(struct notifier_block *nfb,
265 			      unsigned long action,
266 			      void *hcpu)
267 {
268 	unsigned int cpu = (unsigned long)hcpu;
269 	struct sys_device *sys_dev;
270 	int err = 0;
271 
272 	sys_dev = get_cpu_sysdev(cpu);
273 
274 	switch (action) {
275 	case CPU_UP_PREPARE:
276 	case CPU_UP_PREPARE_FROZEN:
277 		mutex_lock(&therm_cpu_lock);
278 		err = thermal_throttle_add_dev(sys_dev, cpu);
279 		mutex_unlock(&therm_cpu_lock);
280 		WARN_ON(err);
281 		break;
282 	case CPU_UP_CANCELED:
283 	case CPU_UP_CANCELED_FROZEN:
284 	case CPU_DEAD:
285 	case CPU_DEAD_FROZEN:
286 		mutex_lock(&therm_cpu_lock);
287 		thermal_throttle_remove_dev(sys_dev);
288 		mutex_unlock(&therm_cpu_lock);
289 		break;
290 	}
291 	return notifier_from_errno(err);
292 }
293 
294 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
295 {
296 	.notifier_call = thermal_throttle_cpu_callback,
297 };
298 
thermal_throttle_init_device(void)299 static __init int thermal_throttle_init_device(void)
300 {
301 	unsigned int cpu = 0;
302 	int err;
303 
304 	if (!atomic_read(&therm_throt_en))
305 		return 0;
306 
307 	register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
308 
309 #ifdef CONFIG_HOTPLUG_CPU
310 	mutex_lock(&therm_cpu_lock);
311 #endif
312 	/* connect live CPUs to sysfs */
313 	for_each_online_cpu(cpu) {
314 		err = thermal_throttle_add_dev(get_cpu_sysdev(cpu), cpu);
315 		WARN_ON(err);
316 	}
317 #ifdef CONFIG_HOTPLUG_CPU
318 	mutex_unlock(&therm_cpu_lock);
319 #endif
320 
321 	return 0;
322 }
323 device_initcall(thermal_throttle_init_device);
324 
325 #endif /* CONFIG_SYSFS */
326 
327 /*
328  * Set up the most two significant bit to notify mce log that this thermal
329  * event type.
330  * This is a temp solution. May be changed in the future with mce log
331  * infrasture.
332  */
333 #define CORE_THROTTLED		(0)
334 #define CORE_POWER_LIMIT	((__u64)1 << 62)
335 #define PACKAGE_THROTTLED	((__u64)2 << 62)
336 #define PACKAGE_POWER_LIMIT	((__u64)3 << 62)
337 
notify_thresholds(__u64 msr_val)338 static void notify_thresholds(__u64 msr_val)
339 {
340 	/* check whether the interrupt handler is defined;
341 	 * otherwise simply return
342 	 */
343 	if (!platform_thermal_notify)
344 		return;
345 
346 	/* lower threshold reached */
347 	if ((msr_val & THERM_LOG_THRESHOLD0) &&	thresh_event_valid(0))
348 		platform_thermal_notify(msr_val);
349 	/* higher threshold reached */
350 	if ((msr_val & THERM_LOG_THRESHOLD1) && thresh_event_valid(1))
351 		platform_thermal_notify(msr_val);
352 }
353 
354 /* Thermal transition interrupt handler */
intel_thermal_interrupt(void)355 static void intel_thermal_interrupt(void)
356 {
357 	__u64 msr_val;
358 	struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
359 
360 	rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
361 
362 	/* Check for violation of core thermal thresholds*/
363 	notify_thresholds(msr_val);
364 
365 	if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
366 				THERMAL_THROTTLING_EVENT,
367 				CORE_LEVEL) != 0)
368 		mce_log_therm_throt_event(CORE_THROTTLED | msr_val);
369 
370 	if (cpu_has(c, X86_FEATURE_PLN))
371 		if (therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
372 					POWER_LIMIT_EVENT,
373 					CORE_LEVEL) != 0)
374 			mce_log_therm_throt_event(CORE_POWER_LIMIT | msr_val);
375 
376 	if (cpu_has(c, X86_FEATURE_PTS)) {
377 		rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
378 		if (therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
379 					THERMAL_THROTTLING_EVENT,
380 					PACKAGE_LEVEL) != 0)
381 			mce_log_therm_throt_event(PACKAGE_THROTTLED | msr_val);
382 		if (cpu_has(c, X86_FEATURE_PLN))
383 			if (therm_throt_process(msr_val &
384 					PACKAGE_THERM_STATUS_POWER_LIMIT,
385 					POWER_LIMIT_EVENT,
386 					PACKAGE_LEVEL) != 0)
387 				mce_log_therm_throt_event(PACKAGE_POWER_LIMIT
388 							  | msr_val);
389 	}
390 }
391 
unexpected_thermal_interrupt(void)392 static void unexpected_thermal_interrupt(void)
393 {
394 	printk(KERN_ERR "CPU%d: Unexpected LVT thermal interrupt!\n",
395 			smp_processor_id());
396 	add_taint(TAINT_MACHINE_CHECK);
397 }
398 
399 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
400 
smp_thermal_interrupt(struct pt_regs * regs)401 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
402 {
403 	exit_idle();
404 	irq_enter();
405 	inc_irq_stat(irq_thermal_count);
406 	smp_thermal_vector();
407 	irq_exit();
408 	/* Ack only at the end to avoid potential reentry */
409 	ack_APIC_irq();
410 }
411 
412 /* Thermal monitoring depends on APIC, ACPI and clock modulation */
intel_thermal_supported(struct cpuinfo_x86 * c)413 static int intel_thermal_supported(struct cpuinfo_x86 *c)
414 {
415 	if (!cpu_has_apic)
416 		return 0;
417 	if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
418 		return 0;
419 	return 1;
420 }
421 
mcheck_intel_therm_init(void)422 void __init mcheck_intel_therm_init(void)
423 {
424 	/*
425 	 * This function is only called on boot CPU. Save the init thermal
426 	 * LVT value on BSP and use that value to restore APs' thermal LVT
427 	 * entry BIOS programmed later
428 	 */
429 	if (intel_thermal_supported(&boot_cpu_data))
430 		lvtthmr_init = apic_read(APIC_LVTTHMR);
431 }
432 
intel_init_thermal(struct cpuinfo_x86 * c)433 void intel_init_thermal(struct cpuinfo_x86 *c)
434 {
435 	unsigned int cpu = smp_processor_id();
436 	int tm2 = 0;
437 	u32 l, h;
438 
439 	if (!intel_thermal_supported(c))
440 		return;
441 
442 	/*
443 	 * First check if its enabled already, in which case there might
444 	 * be some SMM goo which handles it, so we can't even put a handler
445 	 * since it might be delivered via SMI already:
446 	 */
447 	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
448 
449 	h = lvtthmr_init;
450 	/*
451 	 * The initial value of thermal LVT entries on all APs always reads
452 	 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
453 	 * sequence to them and LVT registers are reset to 0s except for
454 	 * the mask bits which are set to 1s when APs receive INIT IPI.
455 	 * If BIOS takes over the thermal interrupt and sets its interrupt
456 	 * delivery mode to SMI (not fixed), it restores the value that the
457 	 * BIOS has programmed on AP based on BSP's info we saved since BIOS
458 	 * is always setting the same value for all threads/cores.
459 	 */
460 	if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
461 		apic_write(APIC_LVTTHMR, lvtthmr_init);
462 
463 
464 	if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
465 		printk(KERN_DEBUG
466 		       "CPU%d: Thermal monitoring handled by SMI\n", cpu);
467 		return;
468 	}
469 
470 	/* Check whether a vector already exists */
471 	if (h & APIC_VECTOR_MASK) {
472 		printk(KERN_DEBUG
473 		       "CPU%d: Thermal LVT vector (%#x) already installed\n",
474 		       cpu, (h & APIC_VECTOR_MASK));
475 		return;
476 	}
477 
478 	/* early Pentium M models use different method for enabling TM2 */
479 	if (cpu_has(c, X86_FEATURE_TM2)) {
480 		if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
481 			rdmsr(MSR_THERM2_CTL, l, h);
482 			if (l & MSR_THERM2_CTL_TM_SELECT)
483 				tm2 = 1;
484 		} else if (l & MSR_IA32_MISC_ENABLE_TM2)
485 			tm2 = 1;
486 	}
487 
488 	/* We'll mask the thermal vector in the lapic till we're ready: */
489 	h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
490 	apic_write(APIC_LVTTHMR, h);
491 
492 	rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
493 	if (cpu_has(c, X86_FEATURE_PLN))
494 		wrmsr(MSR_IA32_THERM_INTERRUPT,
495 		      l | (THERM_INT_LOW_ENABLE
496 			| THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
497 	else
498 		wrmsr(MSR_IA32_THERM_INTERRUPT,
499 		      l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
500 
501 	if (cpu_has(c, X86_FEATURE_PTS)) {
502 		rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
503 		if (cpu_has(c, X86_FEATURE_PLN))
504 			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
505 			      l | (PACKAGE_THERM_INT_LOW_ENABLE
506 				| PACKAGE_THERM_INT_HIGH_ENABLE
507 				| PACKAGE_THERM_INT_PLN_ENABLE), h);
508 		else
509 			wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
510 			      l | (PACKAGE_THERM_INT_LOW_ENABLE
511 				| PACKAGE_THERM_INT_HIGH_ENABLE), h);
512 	}
513 
514 	smp_thermal_vector = intel_thermal_interrupt;
515 
516 	rdmsr(MSR_IA32_MISC_ENABLE, l, h);
517 	wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
518 
519 	/* Unmask the thermal vector: */
520 	l = apic_read(APIC_LVTTHMR);
521 	apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
522 
523 	printk_once(KERN_INFO "CPU0: Thermal monitoring enabled (%s)\n",
524 		       tm2 ? "TM2" : "TM1");
525 
526 	/* enable thermal throttle processing */
527 	atomic_set(&therm_throt_en, 1);
528 }
529