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
;
45 unsigned long last_throttle_count
;
48 static DEFINE_PER_CPU(struct thermal_state
, thermal_state
);
50 static atomic_t therm_throt_en
= ATOMIC_INIT(0);
52 static u32 lvtthmr_init __read_mostly
;
55 #define define_therm_throt_sysdev_one_ro(_name) \
56 static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
58 #define define_therm_throt_sysdev_show_func(name) \
60 static ssize_t therm_throt_sysdev_show_##name( \
61 struct sys_device *dev, \
62 struct sysdev_attribute *attr, \
65 unsigned int cpu = dev->id; \
68 preempt_disable(); /* CPU hotplug */ \
69 if (cpu_online(cpu)) \
70 ret = sprintf(buf, "%lu\n", \
71 per_cpu(thermal_state, cpu).name); \
79 define_therm_throt_sysdev_show_func(throttle_count
);
80 define_therm_throt_sysdev_one_ro(throttle_count
);
82 static struct attribute
*thermal_throttle_attrs
[] = {
83 &attr_throttle_count
.attr
,
87 static struct attribute_group thermal_throttle_attr_group
= {
88 .attrs
= thermal_throttle_attrs
,
89 .name
= "thermal_throttle"
91 #endif /* CONFIG_SYSFS */
94 * therm_throt_process - Process thermal throttling event from interrupt
95 * @curr: Whether the condition is current or not (boolean), since the
96 * thermal interrupt normally gets called both when the thermal
97 * event begins and once the event has ended.
99 * This function is called by the thermal interrupt after the
100 * IRQ has been acknowledged.
102 * It will take care of rate limiting and printing messages to the syslog.
104 * Returns: 0 : Event should NOT be further logged, i.e. still in
105 * "timeout" from previous log message.
106 * 1 : Event should be logged further, and a message has been
107 * printed to the syslog.
109 static int therm_throt_process(bool is_throttled
)
111 struct thermal_state
*state
;
112 unsigned int this_cpu
;
116 this_cpu
= smp_processor_id();
117 now
= get_jiffies_64();
118 state
= &per_cpu(thermal_state
, this_cpu
);
120 was_throttled
= state
->is_throttled
;
121 state
->is_throttled
= is_throttled
;
124 state
->throttle_count
++;
126 if (time_before64(now
, state
->next_check
) &&
127 state
->throttle_count
!= state
->last_throttle_count
)
130 state
->next_check
= now
+ CHECK_INTERVAL
;
131 state
->last_throttle_count
= state
->throttle_count
;
133 /* if we just entered the thermal event */
135 printk(KERN_CRIT
"CPU%d: Temperature above threshold, cpu clock throttled (total events = %lu)\n", this_cpu
, state
->throttle_count
);
137 add_taint(TAINT_MACHINE_CHECK
);
141 printk(KERN_INFO
"CPU%d: Temperature/speed normal\n", this_cpu
);
149 /* Add/Remove thermal_throttle interface for CPU device: */
150 static __cpuinit
int thermal_throttle_add_dev(struct sys_device
*sys_dev
)
152 return sysfs_create_group(&sys_dev
->kobj
,
153 &thermal_throttle_attr_group
);
156 static __cpuinit
void thermal_throttle_remove_dev(struct sys_device
*sys_dev
)
158 sysfs_remove_group(&sys_dev
->kobj
, &thermal_throttle_attr_group
);
161 /* Mutex protecting device creation against CPU hotplug: */
162 static DEFINE_MUTEX(therm_cpu_lock
);
164 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
166 thermal_throttle_cpu_callback(struct notifier_block
*nfb
,
167 unsigned long action
,
170 unsigned int cpu
= (unsigned long)hcpu
;
171 struct sys_device
*sys_dev
;
174 sys_dev
= get_cpu_sysdev(cpu
);
178 case CPU_UP_PREPARE_FROZEN
:
179 mutex_lock(&therm_cpu_lock
);
180 err
= thermal_throttle_add_dev(sys_dev
);
181 mutex_unlock(&therm_cpu_lock
);
184 case CPU_UP_CANCELED
:
185 case CPU_UP_CANCELED_FROZEN
:
187 case CPU_DEAD_FROZEN
:
188 mutex_lock(&therm_cpu_lock
);
189 thermal_throttle_remove_dev(sys_dev
);
190 mutex_unlock(&therm_cpu_lock
);
193 return err
? NOTIFY_BAD
: NOTIFY_OK
;
196 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata
=
198 .notifier_call
= thermal_throttle_cpu_callback
,
201 static __init
int thermal_throttle_init_device(void)
203 unsigned int cpu
= 0;
206 if (!atomic_read(&therm_throt_en
))
209 register_hotcpu_notifier(&thermal_throttle_cpu_notifier
);
211 #ifdef CONFIG_HOTPLUG_CPU
212 mutex_lock(&therm_cpu_lock
);
214 /* connect live CPUs to sysfs */
215 for_each_online_cpu(cpu
) {
216 err
= thermal_throttle_add_dev(get_cpu_sysdev(cpu
));
219 #ifdef CONFIG_HOTPLUG_CPU
220 mutex_unlock(&therm_cpu_lock
);
225 device_initcall(thermal_throttle_init_device
);
227 #endif /* CONFIG_SYSFS */
229 /* Thermal transition interrupt handler */
230 static void intel_thermal_interrupt(void)
234 rdmsrl(MSR_IA32_THERM_STATUS
, msr_val
);
235 if (therm_throt_process((msr_val
& THERM_STATUS_PROCHOT
) != 0))
236 mce_log_therm_throt_event(msr_val
);
239 static void unexpected_thermal_interrupt(void)
241 printk(KERN_ERR
"CPU%d: Unexpected LVT TMR interrupt!\n",
243 add_taint(TAINT_MACHINE_CHECK
);
246 static void (*smp_thermal_vector
)(void) = unexpected_thermal_interrupt
;
248 asmlinkage
void smp_thermal_interrupt(struct pt_regs
*regs
)
252 inc_irq_stat(irq_thermal_count
);
253 smp_thermal_vector();
255 /* Ack only at the end to avoid potential reentry */
259 void __init
mcheck_intel_therm_init(void)
262 * This function is only called on boot CPU. Save the init thermal
263 * LVT value on BSP and use that value to restore APs' thermal LVT
264 * entry BIOS programmed later
266 if (cpu_has(&boot_cpu_data
, X86_FEATURE_ACPI
) &&
267 cpu_has(&boot_cpu_data
, X86_FEATURE_ACC
))
268 lvtthmr_init
= apic_read(APIC_LVTTHMR
);
271 void __init
intel_init_thermal(struct cpuinfo_x86
*c
)
273 unsigned int cpu
= smp_processor_id();
277 /* Thermal monitoring depends on ACPI and clock modulation*/
278 if (!cpu_has(c
, X86_FEATURE_ACPI
) || !cpu_has(c
, X86_FEATURE_ACC
))
282 * First check if its enabled already, in which case there might
283 * be some SMM goo which handles it, so we can't even put a handler
284 * since it might be delivered via SMI already:
286 rdmsr(MSR_IA32_MISC_ENABLE
, l
, h
);
289 * The initial value of thermal LVT entries on all APs always reads
290 * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
291 * sequence to them and LVT registers are reset to 0s except for
292 * the mask bits which are set to 1s when APs receive INIT IPI.
293 * Always restore the value that BIOS has programmed on AP based on
294 * BSP's info we saved since BIOS is always setting the same value
295 * for all threads/cores
297 apic_write(APIC_LVTTHMR
, lvtthmr_init
);
301 if ((l
& MSR_IA32_MISC_ENABLE_TM1
) && (h
& APIC_DM_SMI
)) {
303 "CPU%d: Thermal monitoring handled by SMI\n", cpu
);
307 /* Check whether a vector already exists */
308 if (h
& APIC_VECTOR_MASK
) {
310 "CPU%d: Thermal LVT vector (%#x) already installed\n",
311 cpu
, (h
& APIC_VECTOR_MASK
));
315 /* early Pentium M models use different method for enabling TM2 */
316 if (cpu_has(c
, X86_FEATURE_TM2
)) {
317 if (c
->x86
== 6 && (c
->x86_model
== 9 || c
->x86_model
== 13)) {
318 rdmsr(MSR_THERM2_CTL
, l
, h
);
319 if (l
& MSR_THERM2_CTL_TM_SELECT
)
321 } else if (l
& MSR_IA32_MISC_ENABLE_TM2
)
325 /* We'll mask the thermal vector in the lapic till we're ready: */
326 h
= THERMAL_APIC_VECTOR
| APIC_DM_FIXED
| APIC_LVT_MASKED
;
327 apic_write(APIC_LVTTHMR
, h
);
329 rdmsr(MSR_IA32_THERM_INTERRUPT
, l
, h
);
330 wrmsr(MSR_IA32_THERM_INTERRUPT
,
331 l
| (THERM_INT_LOW_ENABLE
| THERM_INT_HIGH_ENABLE
), h
);
333 smp_thermal_vector
= intel_thermal_interrupt
;
335 rdmsr(MSR_IA32_MISC_ENABLE
, l
, h
);
336 wrmsr(MSR_IA32_MISC_ENABLE
, l
| MSR_IA32_MISC_ENABLE_TM1
, h
);
338 /* Unmask the thermal vector: */
339 l
= apic_read(APIC_LVTTHMR
);
340 apic_write(APIC_LVTTHMR
, l
& ~APIC_LVT_MASKED
);
342 printk(KERN_INFO
"CPU%d: Thermal monitoring enabled (%s)\n",
343 cpu
, tm2
? "TM2" : "TM1");
345 /* enable thermal throttle processing */
346 atomic_set(&therm_throt_en
, 1);