1 // SPDX-License-Identifier: GPL-2.0-only
3  * Thermal throttle event support code (such as syslog messaging and rate
6  * This allows consistent reporting of CPU thermal throttle events.
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
26 #include <linux/cpu.h>
29 #include <asm/thermal.h>
38 /* How long to wait between reporting thermal events */
45  * struct _thermal_state - Represent the current thermal event state
51  * @count:			Stores the current running count for thermal
53  * @last_count:			Stores the previous running count for thermal
55  * @max_time_ms:		This shows the maximum amount of time CPU was
56  *				in throttled state for a single thermal
58  * @total_time_ms:		This is a cumulative time during which CPU was
61  *				This is used for the purpose of rate-control.
72  * @baseline_temp:		Temperature at which thermal threshold high
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.
142 	unsigned int cpu = dev->id;					\
145 	preempt_disable();	/* CPU hotplug */			\
146 	if (cpu_online(cpu)) {						\
148 			      per_cpu(thermal_state, cpu).event.name);	\
206 	 * "Table 2-2. IA-32 Architectural MSRs", MSR 0x19C  in thermal_intr_init_core_clear_mask()
217 	 * Bit 7 and 9: Thermal Threshold #1 and #2 log  in thermal_intr_init_core_clear_mask()
242 	 * "Table 2-2. IA-32 Architectural MSRs", MSR 0x1B1  in thermal_intr_init_pkg_clear_mask()
251 	 * Intel SDM Volume 2A: Thermal and Power Management Leaf  in thermal_intr_init_pkg_clear_mask()
259  * Clear the bits in package thermal status register for bit = 1
308 	get_therm_status(state->level, &hot, &temp);  in throttle_active_work()
310 	if (!hot && temp > state->baseline_temp) {  in throttle_active_work()
311 		if (state->rate_control_active)  in throttle_active_work()
312 			pr_info("CPU%d: %s temperature/speed normal (total events = %lu)\n",  in throttle_active_work()
314 				state->level == CORE_LEVEL ? "Core" : "Package",  in throttle_active_work()
315 				state->count);  in throttle_active_work()
317 		state->rate_control_active = false;  in throttle_active_work()
321 	if (time_before64(now, state->next_check) &&  in throttle_active_work()
322 			  state->rate_control_active)  in throttle_active_work()
325 	state->next_check = now + CHECK_INTERVAL;  in throttle_active_work()
327 	if (state->count != state->last_count) {  in throttle_active_work()
328 		/* There was one new thermal interrupt */  in throttle_active_work()
329 		state->last_count = state->count;  in throttle_active_work()
330 		state->average = 0;  in throttle_active_work()
331 		state->sample_count = 0;  in throttle_active_work()
332 		state->sample_index = 0;  in throttle_active_work()
335 	state->temp_samples[state->sample_index] = temp;  in throttle_active_work()
336 	state->sample_count++;  in throttle_active_work()
337 	state->sample_index = (state->sample_index + 1) % ARRAY_SIZE(state->temp_samples);  in throttle_active_work()
338 	if (state->sample_count < ARRAY_SIZE(state->temp_samples))  in throttle_active_work()
342 	for (i = 0; i < ARRAY_SIZE(state->temp_samples); ++i)  in throttle_active_work()
343 		avg += state->temp_samples[i];  in throttle_active_work()
345 	avg /= ARRAY_SIZE(state->temp_samples);  in throttle_active_work()
347 	if (state->average > avg) {  in throttle_active_work()
348 		pr_warn("CPU%d: %s temperature is above threshold, cpu clock is throttled (total events = %lu)\n",  in throttle_active_work()
350 			state->level == CORE_LEVEL ? "Core" : "Package",  in throttle_active_work()
351 			state->count);  in throttle_active_work()
352 		state->rate_control_active = true;  in throttle_active_work()
355 	state->average = avg;  in throttle_active_work()
358 	thermal_clear_package_intr_status(state->level, THERM_STATUS_PROCHOT_LOG);  in throttle_active_work()
359 	schedule_delayed_work_on(this_cpu, &state->therm_work, THERM_THROT_POLL_INTERVAL);  in throttle_active_work()
363  * therm_throt_process - Process thermal throttling event from interrupt
365  *        thermal interrupt normally gets called both when the thermal
368  * This function is called by the thermal interrupt after the
384 			state = &pstate->core_throttle;  in therm_throt_process()
386 			state = &pstate->core_power_limit;  in therm_throt_process()
391 			state = &pstate->package_throttle;  in therm_throt_process()
393 			state = &pstate->package_power_limit;  in therm_throt_process()
399 	old_event = state->new_event;  in therm_throt_process()
400 	state->new_event = new_event;  in therm_throt_process()
403 		state->count++;  in therm_throt_process()
408 	if (new_event && !state->last_interrupt_time) {  in therm_throt_process()
412 		get_therm_status(state->level, &hot, &temp);  in therm_throt_process()
421 		state->baseline_temp = temp;  in therm_throt_process()
422 		state->last_interrupt_time = now;  in therm_throt_process()
423 		schedule_delayed_work_on(this_cpu, &state->therm_work, THERM_THROT_POLL_INTERVAL);  in therm_throt_process()
424 	} else if (old_event && state->last_interrupt_time) {  in therm_throt_process()
427 		throttle_time = jiffies_delta_to_msecs(now - state->last_interrupt_time);  in therm_throt_process()
428 		if (throttle_time > state->max_time_ms)  in therm_throt_process()
429 			state->max_time_ms = throttle_time;  in therm_throt_process()
430 		state->total_time_ms += throttle_time;  in therm_throt_process()
431 		state->last_interrupt_time = 0;  in therm_throt_process()
443 		state = (event == 0) ? &pstate->pkg_thresh0 :  in thresh_event_valid()
444 						&pstate->pkg_thresh1;  in thresh_event_valid()
446 		state = (event == 0) ? &pstate->core_thresh0 :  in thresh_event_valid()
447 						&pstate->core_thresh1;  in thresh_event_valid()
449 	if (time_before64(now, state->next_check))  in thresh_event_valid()
452 	state->next_check = now + CHECK_INTERVAL;  in thresh_event_valid()
467 /* Add/Remove thermal_throttle interface for CPU device: */
468 static int thermal_throttle_add_dev(struct device *dev, unsigned int cpu)  in thermal_throttle_add_dev()  argument
471 	struct cpuinfo_x86 *c = &cpu_data(cpu);  in thermal_throttle_add_dev()
473 	err = sysfs_create_group(&dev->kobj, &thermal_attr_group);  in thermal_throttle_add_dev()
478 		err = sysfs_add_file_to_group(&dev->kobj,  in thermal_throttle_add_dev()
486 		err = sysfs_add_file_to_group(&dev->kobj,  in thermal_throttle_add_dev()
492 		err = sysfs_add_file_to_group(&dev->kobj,  in thermal_throttle_add_dev()
498 		err = sysfs_add_file_to_group(&dev->kobj,  in thermal_throttle_add_dev()
505 			err = sysfs_add_file_to_group(&dev->kobj,  in thermal_throttle_add_dev()
516 	sysfs_remove_group(&dev->kobj, &thermal_attr_group);  in thermal_throttle_add_dev()
523 	sysfs_remove_group(&dev->kobj, &thermal_attr_group);  in thermal_throttle_remove_dev()
526 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
527 static int thermal_throttle_online(unsigned int cpu)  in thermal_throttle_online()  argument
529 	struct thermal_state *state = &per_cpu(thermal_state, cpu);  in thermal_throttle_online()
530 	struct device *dev = get_cpu_device(cpu);  in thermal_throttle_online()
533 	state->package_throttle.level = PACKAGE_LEVEL;  in thermal_throttle_online()
534 	state->core_throttle.level = CORE_LEVEL;  in thermal_throttle_online()
536 	INIT_DELAYED_WORK(&state->package_throttle.therm_work, throttle_active_work);  in thermal_throttle_online()
537 	INIT_DELAYED_WORK(&state->core_throttle.therm_work, throttle_active_work);  in thermal_throttle_online()
540 	 * The first CPU coming online will enable the HFI. Usually this causes  in thermal_throttle_online()
541 	 * hardware to issue an HFI thermal interrupt. Such interrupt will reach  in thermal_throttle_online()
542 	 * the CPU once we enable the thermal vector in the local APIC.  in thermal_throttle_online()
544 	intel_hfi_online(cpu);  in thermal_throttle_online()
546 	/* Unmask the thermal vector after the above workqueues are initialized. */  in thermal_throttle_online()
550 	return thermal_throttle_add_dev(dev, cpu);  in thermal_throttle_online()
553 static int thermal_throttle_offline(unsigned int cpu)  in thermal_throttle_offline()  argument
555 	struct thermal_state *state = &per_cpu(thermal_state, cpu);  in thermal_throttle_offline()
556 	struct device *dev = get_cpu_device(cpu);  in thermal_throttle_offline()
559 	/* Mask the thermal vector before draining evtl. pending work */  in thermal_throttle_offline()
563 	intel_hfi_offline(cpu);  in thermal_throttle_offline()
565 	cancel_delayed_work_sync(&state->package_throttle.therm_work);  in thermal_throttle_offline()
566 	cancel_delayed_work_sync(&state->core_throttle.therm_work);  in thermal_throttle_offline()
568 	state->package_throttle.rate_control_active = false;  in thermal_throttle_offline()
569 	state->core_throttle.rate_control_active = false;  in thermal_throttle_offline()
649 /* Thermal transition interrupt handler */
659 	/* Check for violation of core thermal thresholds*/  in intel_thermal_interrupt()
673 		/* check violations of package thermal thresholds */  in intel_thermal_interrupt()
690 /* Thermal monitoring depends on APIC, ACPI and clock modulation */
708 	 * This function is only called on boot CPU. Save the init thermal  in therm_lvt_init()
709 	 * LVT value on BSP and use that value to restore APs' thermal LVT  in therm_lvt_init()
718 	unsigned int cpu = smp_processor_id();  in intel_init_thermal()  local
734 	 * The initial value of thermal LVT entries on all APs always reads  in intel_init_thermal()
735 	 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI  in intel_init_thermal()
738 	 * If BIOS takes over the thermal interrupt and sets its interrupt  in intel_init_thermal()
749 			pr_debug("CPU%d: Thermal monitoring handled by SMI\n", cpu);  in intel_init_thermal()
755 		if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {  in intel_init_thermal()
763 	/* We'll mask the thermal vector in the lapic till we're ready: */  in intel_init_thermal()
810 	pr_info_once("CPU0: Thermal monitoring enabled (%s)\n",  in intel_init_thermal()
813 	/* enable thermal throttle processing */  in intel_init_thermal()