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/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>
27 #include <asm/processor.h>
28 #include <asm/system.h>
34 /* How long to wait between reporting thermal events */
35 #define CHECK_INTERVAL (300 * HZ)
38 * Current thermal throttling state:
40 struct thermal_state
{
44 unsigned long throttle_count
;
47 static DEFINE_PER_CPU(struct thermal_state
, thermal_state
);
49 static atomic_t therm_throt_en
= ATOMIC_INIT(0);
52 #define define_therm_throt_sysdev_one_ro(_name) \
53 static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
55 #define define_therm_throt_sysdev_show_func(name) \
57 static ssize_t therm_throt_sysdev_show_##name( \
58 struct sys_device *dev, \
59 struct sysdev_attribute *attr, \
62 unsigned int cpu = dev->id; \
65 preempt_disable(); /* CPU hotplug */ \
66 if (cpu_online(cpu)) \
67 ret = sprintf(buf, "%lu\n", \
68 per_cpu(thermal_state, cpu).name); \
76 define_therm_throt_sysdev_show_func(throttle_count
);
77 define_therm_throt_sysdev_one_ro(throttle_count
);
79 static struct attribute
*thermal_throttle_attrs
[] = {
80 &attr_throttle_count
.attr
,
84 static struct attribute_group thermal_throttle_attr_group
= {
85 .attrs
= thermal_throttle_attrs
,
86 .name
= "thermal_throttle"
88 #endif /* CONFIG_SYSFS */
91 * therm_throt_process - Process thermal throttling event from interrupt
92 * @curr: Whether the condition is current or not (boolean), since the
93 * thermal interrupt normally gets called both when the thermal
94 * event begins and once the event has ended.
96 * This function is called by the thermal interrupt after the
97 * IRQ has been acknowledged.
99 * It will take care of rate limiting and printing messages to the syslog.
101 * Returns: 0 : Event should NOT be further logged, i.e. still in
102 * "timeout" from previous log message.
103 * 1 : Event should be logged further, and a message has been
104 * printed to the syslog.
106 static int therm_throt_process(bool is_throttled
)
108 struct thermal_state
*state
;
109 unsigned int this_cpu
;
113 this_cpu
= smp_processor_id();
114 now
= get_jiffies_64();
115 state
= &per_cpu(thermal_state
, this_cpu
);
117 was_throttled
= state
->is_throttled
;
118 state
->is_throttled
= is_throttled
;
121 state
->throttle_count
++;
123 if (!(was_throttled
^ is_throttled
) &&
124 time_before64(now
, state
->next_check
))
127 state
->next_check
= now
+ CHECK_INTERVAL
;
129 /* if we just entered the thermal event */
131 printk(KERN_CRIT
"CPU%d: Temperature above threshold, cpu clock throttled (total events = %lu)\n", this_cpu
, state
->throttle_count
);
133 add_taint(TAINT_MACHINE_CHECK
);
137 printk(KERN_INFO
"CPU%d: Temperature/speed normal\n", this_cpu
);
145 /* Add/Remove thermal_throttle interface for CPU device: */
146 static __cpuinit
int thermal_throttle_add_dev(struct sys_device
*sys_dev
)
148 return sysfs_create_group(&sys_dev
->kobj
,
149 &thermal_throttle_attr_group
);
152 static __cpuinit
void thermal_throttle_remove_dev(struct sys_device
*sys_dev
)
154 sysfs_remove_group(&sys_dev
->kobj
, &thermal_throttle_attr_group
);
157 /* Mutex protecting device creation against CPU hotplug: */
158 static DEFINE_MUTEX(therm_cpu_lock
);
160 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
162 thermal_throttle_cpu_callback(struct notifier_block
*nfb
,
163 unsigned long action
,
166 unsigned int cpu
= (unsigned long)hcpu
;
167 struct sys_device
*sys_dev
;
170 sys_dev
= get_cpu_sysdev(cpu
);
174 case CPU_UP_PREPARE_FROZEN
:
175 mutex_lock(&therm_cpu_lock
);
176 err
= thermal_throttle_add_dev(sys_dev
);
177 mutex_unlock(&therm_cpu_lock
);
180 case CPU_UP_CANCELED
:
181 case CPU_UP_CANCELED_FROZEN
:
183 case CPU_DEAD_FROZEN
:
184 mutex_lock(&therm_cpu_lock
);
185 thermal_throttle_remove_dev(sys_dev
);
186 mutex_unlock(&therm_cpu_lock
);
189 return err
? NOTIFY_BAD
: NOTIFY_OK
;
192 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata
=
194 .notifier_call
= thermal_throttle_cpu_callback
,
197 static __init
int thermal_throttle_init_device(void)
199 unsigned int cpu
= 0;
202 if (!atomic_read(&therm_throt_en
))
205 register_hotcpu_notifier(&thermal_throttle_cpu_notifier
);
207 #ifdef CONFIG_HOTPLUG_CPU
208 mutex_lock(&therm_cpu_lock
);
210 /* connect live CPUs to sysfs */
211 for_each_online_cpu(cpu
) {
212 err
= thermal_throttle_add_dev(get_cpu_sysdev(cpu
));
215 #ifdef CONFIG_HOTPLUG_CPU
216 mutex_unlock(&therm_cpu_lock
);
221 device_initcall(thermal_throttle_init_device
);
223 #endif /* CONFIG_SYSFS */
225 /* Thermal transition interrupt handler */
226 static void intel_thermal_interrupt(void)
230 rdmsrl(MSR_IA32_THERM_STATUS
, msr_val
);
231 if (therm_throt_process((msr_val
& THERM_STATUS_PROCHOT
) != 0))
232 mce_log_therm_throt_event(msr_val
);
235 static void unexpected_thermal_interrupt(void)
237 printk(KERN_ERR
"CPU%d: Unexpected LVT TMR interrupt!\n",
239 add_taint(TAINT_MACHINE_CHECK
);
242 static void (*smp_thermal_vector
)(void) = unexpected_thermal_interrupt
;
244 asmlinkage
void smp_thermal_interrupt(struct pt_regs
*regs
)
248 inc_irq_stat(irq_thermal_count
);
249 smp_thermal_vector();
251 /* Ack only at the end to avoid potential reentry */
255 void intel_init_thermal(struct cpuinfo_x86
*c
)
257 unsigned int cpu
= smp_processor_id();
261 /* Thermal monitoring depends on ACPI and clock modulation*/
262 if (!cpu_has(c
, X86_FEATURE_ACPI
) || !cpu_has(c
, X86_FEATURE_ACC
))
266 * First check if its enabled already, in which case there might
267 * be some SMM goo which handles it, so we can't even put a handler
268 * since it might be delivered via SMI already:
270 rdmsr(MSR_IA32_MISC_ENABLE
, l
, h
);
271 h
= apic_read(APIC_LVTTHMR
);
272 if ((l
& MSR_IA32_MISC_ENABLE_TM1
) && (h
& APIC_DM_SMI
)) {
274 "CPU%d: Thermal monitoring handled by SMI\n", cpu
);
278 if (cpu_has(c
, X86_FEATURE_TM2
) && (l
& MSR_IA32_MISC_ENABLE_TM2
))
281 /* Check whether a vector already exists */
282 if (h
& APIC_VECTOR_MASK
) {
284 "CPU%d: Thermal LVT vector (%#x) already installed\n",
285 cpu
, (h
& APIC_VECTOR_MASK
));
289 /* We'll mask the thermal vector in the lapic till we're ready: */
290 h
= THERMAL_APIC_VECTOR
| APIC_DM_FIXED
| APIC_LVT_MASKED
;
291 apic_write(APIC_LVTTHMR
, h
);
293 rdmsr(MSR_IA32_THERM_INTERRUPT
, l
, h
);
294 wrmsr(MSR_IA32_THERM_INTERRUPT
,
295 l
| (THERM_INT_LOW_ENABLE
| THERM_INT_HIGH_ENABLE
), h
);
297 smp_thermal_vector
= intel_thermal_interrupt
;
299 rdmsr(MSR_IA32_MISC_ENABLE
, l
, h
);
300 wrmsr(MSR_IA32_MISC_ENABLE
, l
| MSR_IA32_MISC_ENABLE_TM1
, h
);
302 /* Unmask the thermal vector: */
303 l
= apic_read(APIC_LVTTHMR
);
304 apic_write(APIC_LVTTHMR
, l
& ~APIC_LVT_MASKED
);
306 printk(KERN_INFO
"CPU%d: Thermal monitoring enabled (%s)\n",
307 cpu
, tm2
? "TM2" : "TM1");
309 /* enable thermal throttle processing */
310 atomic_set(&therm_throt_en
, 1);