x86: mce: Clean up thermal throttling state tracking code
[safe/jmp/linux-2.6] / arch / x86 / kernel / cpu / mcheck / therm_throt.c
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 /*
38  * Current thermal throttling state:
39  */
40 struct thermal_state {
41         bool                    is_throttled;
42
43         u64                     next_check;
44         unsigned long           throttle_count;
45 };
46
47 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
48
49 static atomic_t therm_throt_en  = ATOMIC_INIT(0);
50
51 #ifdef CONFIG_SYSFS
52 #define define_therm_throt_sysdev_one_ro(_name)                         \
53         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
54
55 #define define_therm_throt_sysdev_show_func(name)                       \
56                                                                         \
57 static ssize_t therm_throt_sysdev_show_##name(                          \
58                         struct sys_device *dev,                         \
59                         struct sysdev_attribute *attr,                  \
60                         char *buf)                                      \
61 {                                                                       \
62         unsigned int cpu = dev->id;                                     \
63         ssize_t ret;                                                    \
64                                                                         \
65         preempt_disable();      /* CPU hotplug */                       \
66         if (cpu_online(cpu))                                            \
67                 ret = sprintf(buf, "%lu\n",                             \
68                               per_cpu(thermal_state, cpu).name);        \
69         else                                                            \
70                 ret = 0;                                                \
71         preempt_enable();                                               \
72                                                                         \
73         return ret;                                                     \
74 }
75
76 define_therm_throt_sysdev_show_func(throttle_count);
77 define_therm_throt_sysdev_one_ro(throttle_count);
78
79 static struct attribute *thermal_throttle_attrs[] = {
80         &attr_throttle_count.attr,
81         NULL
82 };
83
84 static struct attribute_group thermal_throttle_attr_group = {
85         .attrs  = thermal_throttle_attrs,
86         .name   = "thermal_throttle"
87 };
88 #endif /* CONFIG_SYSFS */
89
90 /***
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.
95  *
96  * This function is called by the thermal interrupt after the
97  * IRQ has been acknowledged.
98  *
99  * It will take care of rate limiting and printing messages to the syslog.
100  *
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.
105  */
106 static int therm_throt_process(bool is_throttled)
107 {
108         struct thermal_state *state;
109         unsigned int this_cpu;
110         bool was_throttled;
111         u64 now;
112
113         this_cpu = smp_processor_id();
114         now = get_jiffies_64();
115         state = &per_cpu(thermal_state, this_cpu);
116
117         was_throttled = state->is_throttled;
118         state->is_throttled = is_throttled;
119
120         if (is_throttled)
121                 state->throttle_count++;
122
123         if (!(was_throttled ^ is_throttled) &&
124                         time_before64(now, state->next_check))
125                 return 0;
126
127         state->next_check = now + CHECK_INTERVAL;
128
129         /* if we just entered the thermal event */
130         if (is_throttled) {
131                 printk(KERN_CRIT "CPU%d: Temperature above threshold, cpu clock throttled (total events = %lu)\n", this_cpu, state->throttle_count);
132
133                 add_taint(TAINT_MACHINE_CHECK);
134                 return 1;
135         }
136         if (was_throttled) {
137                 printk(KERN_INFO "CPU%d: Temperature/speed normal\n", this_cpu);
138                 return 1;
139         }
140
141         return 0;
142 }
143
144 #ifdef CONFIG_SYSFS
145 /* Add/Remove thermal_throttle interface for CPU device: */
146 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
147 {
148         return sysfs_create_group(&sys_dev->kobj,
149                                   &thermal_throttle_attr_group);
150 }
151
152 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
153 {
154         sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
155 }
156
157 /* Mutex protecting device creation against CPU hotplug: */
158 static DEFINE_MUTEX(therm_cpu_lock);
159
160 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
161 static __cpuinit int
162 thermal_throttle_cpu_callback(struct notifier_block *nfb,
163                               unsigned long action,
164                               void *hcpu)
165 {
166         unsigned int cpu = (unsigned long)hcpu;
167         struct sys_device *sys_dev;
168         int err = 0;
169
170         sys_dev = get_cpu_sysdev(cpu);
171
172         switch (action) {
173         case CPU_UP_PREPARE:
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);
178                 WARN_ON(err);
179                 break;
180         case CPU_UP_CANCELED:
181         case CPU_UP_CANCELED_FROZEN:
182         case CPU_DEAD:
183         case CPU_DEAD_FROZEN:
184                 mutex_lock(&therm_cpu_lock);
185                 thermal_throttle_remove_dev(sys_dev);
186                 mutex_unlock(&therm_cpu_lock);
187                 break;
188         }
189         return err ? NOTIFY_BAD : NOTIFY_OK;
190 }
191
192 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
193 {
194         .notifier_call = thermal_throttle_cpu_callback,
195 };
196
197 static __init int thermal_throttle_init_device(void)
198 {
199         unsigned int cpu = 0;
200         int err;
201
202         if (!atomic_read(&therm_throt_en))
203                 return 0;
204
205         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
206
207 #ifdef CONFIG_HOTPLUG_CPU
208         mutex_lock(&therm_cpu_lock);
209 #endif
210         /* connect live CPUs to sysfs */
211         for_each_online_cpu(cpu) {
212                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
213                 WARN_ON(err);
214         }
215 #ifdef CONFIG_HOTPLUG_CPU
216         mutex_unlock(&therm_cpu_lock);
217 #endif
218
219         return 0;
220 }
221 device_initcall(thermal_throttle_init_device);
222
223 #endif /* CONFIG_SYSFS */
224
225 /* Thermal transition interrupt handler */
226 static void intel_thermal_interrupt(void)
227 {
228         __u64 msr_val;
229
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);
233 }
234
235 static void unexpected_thermal_interrupt(void)
236 {
237         printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
238                         smp_processor_id());
239         add_taint(TAINT_MACHINE_CHECK);
240 }
241
242 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
243
244 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
245 {
246         exit_idle();
247         irq_enter();
248         inc_irq_stat(irq_thermal_count);
249         smp_thermal_vector();
250         irq_exit();
251         /* Ack only at the end to avoid potential reentry */
252         ack_APIC_irq();
253 }
254
255 void intel_init_thermal(struct cpuinfo_x86 *c)
256 {
257         unsigned int cpu = smp_processor_id();
258         int tm2 = 0;
259         u32 l, h;
260
261         /* Thermal monitoring depends on ACPI and clock modulation*/
262         if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
263                 return;
264
265         /*
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:
269          */
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)) {
273                 printk(KERN_DEBUG
274                        "CPU%d: Thermal monitoring handled by SMI\n", cpu);
275                 return;
276         }
277
278         /* Check whether a vector already exists */
279         if (h & APIC_VECTOR_MASK) {
280                 printk(KERN_DEBUG
281                        "CPU%d: Thermal LVT vector (%#x) already installed\n",
282                        cpu, (h & APIC_VECTOR_MASK));
283                 return;
284         }
285
286         /* early Pentium M models use different method for enabling TM2 */
287         if (cpu_has(c, X86_FEATURE_TM2)) {
288                 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
289                         rdmsr(MSR_THERM2_CTL, l, h);
290                         if (l & MSR_THERM2_CTL_TM_SELECT)
291                                 tm2 = 1;
292                 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
293                         tm2 = 1;
294         }
295
296         /* We'll mask the thermal vector in the lapic till we're ready: */
297         h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
298         apic_write(APIC_LVTTHMR, h);
299
300         rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
301         wrmsr(MSR_IA32_THERM_INTERRUPT,
302                 l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
303
304         smp_thermal_vector = intel_thermal_interrupt;
305
306         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
307         wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
308
309         /* Unmask the thermal vector: */
310         l = apic_read(APIC_LVTTHMR);
311         apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
312
313         printk(KERN_INFO "CPU%d: Thermal monitoring enabled (%s)\n",
314                cpu, tm2 ? "TM2" : "TM1");
315
316         /* enable thermal throttle processing */
317         atomic_set(&therm_throt_en, 1);
318 }