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).
5 * This allows consistent reporting of CPU thermal throttle events.
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).
11 * Author: Dmitriy Zavin (dmitriyz@google.com)
13 * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
14 * Inspired by Ross Biro's and Al Borchers' counter code.
16 #include <linux/interrupt.h>
17 #include <linux/notifier.h>
18 #include <linux/jiffies.h>
19 #include <linux/percpu.h>
20 #include <linux/sysdev.h>
21 #include <linux/cpu.h>
23 #include <asm/therm_throt.h>
27 /* How long to wait between reporting thermal events */
28 #define CHECK_INTERVAL (300 * HZ)
30 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
31 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
33 atomic_t therm_throt_en = ATOMIC_INIT(0);
36 #define define_therm_throt_sysdev_one_ro(_name) \
37 static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
39 #define define_therm_throt_sysdev_show_func(name) \
40 static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev, \
41 struct sysdev_attribute *attr, \
44 unsigned int cpu = dev->id; \
47 preempt_disable(); /* CPU hotplug */ \
48 if (cpu_online(cpu)) \
49 ret = sprintf(buf, "%lu\n", \
50 per_cpu(thermal_throttle_##name, cpu)); \
58 define_therm_throt_sysdev_show_func(count);
59 define_therm_throt_sysdev_one_ro(count);
61 static struct attribute *thermal_throttle_attrs[] = {
66 static struct attribute_group thermal_throttle_attr_group = {
67 .attrs = thermal_throttle_attrs,
68 .name = "thermal_throttle"
70 #endif /* CONFIG_SYSFS */
73 * therm_throt_process - Process thermal throttling event from interrupt
74 * @curr: Whether the condition is current or not (boolean), since the
75 * thermal interrupt normally gets called both when the thermal
76 * event begins and once the event has ended.
78 * This function is called by the thermal interrupt after the
79 * IRQ has been acknowledged.
81 * It will take care of rate limiting and printing messages to the syslog.
83 * Returns: 0 : Event should NOT be further logged, i.e. still in
84 * "timeout" from previous log message.
85 * 1 : Event should be logged further, and a message has been
86 * printed to the syslog.
88 int therm_throt_process(int curr)
90 unsigned int cpu = smp_processor_id();
91 __u64 tmp_jiffs = get_jiffies_64();
94 __get_cpu_var(thermal_throttle_count)++;
96 if (time_before64(tmp_jiffs, __get_cpu_var(next_check)))
99 __get_cpu_var(next_check) = tmp_jiffs + CHECK_INTERVAL;
101 /* if we just entered the thermal event */
103 printk(KERN_CRIT "CPU%d: Temperature above threshold, "
104 "cpu clock throttled (total events = %lu)\n", cpu,
105 __get_cpu_var(thermal_throttle_count));
107 add_taint(TAINT_MACHINE_CHECK);
109 printk(KERN_CRIT "CPU%d: Temperature/speed normal\n", cpu);
116 /* Add/Remove thermal_throttle interface for CPU device: */
117 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
119 return sysfs_create_group(&sys_dev->kobj,
120 &thermal_throttle_attr_group);
123 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
125 sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
128 /* Mutex protecting device creation against CPU hotplug: */
129 static DEFINE_MUTEX(therm_cpu_lock);
131 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
133 thermal_throttle_cpu_callback(struct notifier_block *nfb,
134 unsigned long action,
137 unsigned int cpu = (unsigned long)hcpu;
138 struct sys_device *sys_dev;
141 sys_dev = get_cpu_sysdev(cpu);
145 case CPU_UP_PREPARE_FROZEN:
146 mutex_lock(&therm_cpu_lock);
147 err = thermal_throttle_add_dev(sys_dev);
148 mutex_unlock(&therm_cpu_lock);
151 case CPU_UP_CANCELED:
152 case CPU_UP_CANCELED_FROZEN:
154 case CPU_DEAD_FROZEN:
155 mutex_lock(&therm_cpu_lock);
156 thermal_throttle_remove_dev(sys_dev);
157 mutex_unlock(&therm_cpu_lock);
160 return err ? NOTIFY_BAD : NOTIFY_OK;
163 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
165 .notifier_call = thermal_throttle_cpu_callback,
168 static __init int thermal_throttle_init_device(void)
170 unsigned int cpu = 0;
173 if (!atomic_read(&therm_throt_en))
176 register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
178 #ifdef CONFIG_HOTPLUG_CPU
179 mutex_lock(&therm_cpu_lock);
181 /* connect live CPUs to sysfs */
182 for_each_online_cpu(cpu) {
183 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
186 #ifdef CONFIG_HOTPLUG_CPU
187 mutex_unlock(&therm_cpu_lock);
192 device_initcall(thermal_throttle_init_device);
194 #endif /* CONFIG_SYSFS */
196 /* Thermal transition interrupt handler */
197 void intel_thermal_interrupt(void)
201 rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
202 if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
203 mce_log_therm_throt_event(msr_val);
206 static void unexpected_thermal_interrupt(void)
208 printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
210 add_taint(TAINT_MACHINE_CHECK);
213 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
215 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
219 inc_irq_stat(irq_thermal_count);
220 smp_thermal_vector();
222 /* Ack only at the end to avoid potential reentry */
226 void intel_set_thermal_handler(void)
228 smp_thermal_vector = intel_thermal_interrupt;