Merge branch 'x86-mce-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6.git] / 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 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
38 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
39 static DEFINE_PER_CPU(bool, thermal_throttle_active);
40
41 static atomic_t therm_throt_en          = ATOMIC_INIT(0);
42
43 #ifdef CONFIG_SYSFS
44 #define define_therm_throt_sysdev_one_ro(_name)                         \
45         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
46
47 #define define_therm_throt_sysdev_show_func(name)                       \
48 static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,   \
49                                         struct sysdev_attribute *attr,  \
50                                               char *buf)                \
51 {                                                                       \
52         unsigned int cpu = dev->id;                                     \
53         ssize_t ret;                                                    \
54                                                                         \
55         preempt_disable();      /* CPU hotplug */                       \
56         if (cpu_online(cpu))                                            \
57                 ret = sprintf(buf, "%lu\n",                             \
58                               per_cpu(thermal_throttle_##name, cpu));   \
59         else                                                            \
60                 ret = 0;                                                \
61         preempt_enable();                                               \
62                                                                         \
63         return ret;                                                     \
64 }
65
66 define_therm_throt_sysdev_show_func(count);
67 define_therm_throt_sysdev_one_ro(count);
68
69 static struct attribute *thermal_throttle_attrs[] = {
70         &attr_count.attr,
71         NULL
72 };
73
74 static struct attribute_group thermal_throttle_attr_group = {
75         .attrs  = thermal_throttle_attrs,
76         .name   = "thermal_throttle"
77 };
78 #endif /* CONFIG_SYSFS */
79
80 /***
81  * therm_throt_process - Process thermal throttling event from interrupt
82  * @curr: Whether the condition is current or not (boolean), since the
83  *        thermal interrupt normally gets called both when the thermal
84  *        event begins and once the event has ended.
85  *
86  * This function is called by the thermal interrupt after the
87  * IRQ has been acknowledged.
88  *
89  * It will take care of rate limiting and printing messages to the syslog.
90  *
91  * Returns: 0 : Event should NOT be further logged, i.e. still in
92  *              "timeout" from previous log message.
93  *          1 : Event should be logged further, and a message has been
94  *              printed to the syslog.
95  */
96 static int therm_throt_process(int curr)
97 {
98         unsigned int cpu = smp_processor_id();
99         __u64 tmp_jiffs = get_jiffies_64();
100         bool was_throttled = __get_cpu_var(thermal_throttle_active);
101         bool is_throttled = __get_cpu_var(thermal_throttle_active) = curr;
102
103         if (is_throttled)
104                 __get_cpu_var(thermal_throttle_count)++;
105
106         if (!(was_throttled ^ is_throttled) &&
107             time_before64(tmp_jiffs, __get_cpu_var(next_check)))
108                 return 0;
109
110         __get_cpu_var(next_check) = tmp_jiffs + CHECK_INTERVAL;
111
112         /* if we just entered the thermal event */
113         if (is_throttled) {
114                 printk(KERN_CRIT "CPU%d: Temperature above threshold, "
115                        "cpu clock throttled (total events = %lu)\n",
116                        cpu, __get_cpu_var(thermal_throttle_count));
117
118                 add_taint(TAINT_MACHINE_CHECK);
119                 return 1;
120         }
121         if (was_throttled) {
122                 printk(KERN_INFO "CPU%d: Temperature/speed normal\n", cpu);
123                 return 1;
124         }
125
126         return 0;
127 }
128
129 #ifdef CONFIG_SYSFS
130 /* Add/Remove thermal_throttle interface for CPU device: */
131 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
132 {
133         return sysfs_create_group(&sys_dev->kobj,
134                                   &thermal_throttle_attr_group);
135 }
136
137 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
138 {
139         sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
140 }
141
142 /* Mutex protecting device creation against CPU hotplug: */
143 static DEFINE_MUTEX(therm_cpu_lock);
144
145 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
146 static __cpuinit int
147 thermal_throttle_cpu_callback(struct notifier_block *nfb,
148                               unsigned long action,
149                               void *hcpu)
150 {
151         unsigned int cpu = (unsigned long)hcpu;
152         struct sys_device *sys_dev;
153         int err = 0;
154
155         sys_dev = get_cpu_sysdev(cpu);
156
157         switch (action) {
158         case CPU_UP_PREPARE:
159         case CPU_UP_PREPARE_FROZEN:
160                 mutex_lock(&therm_cpu_lock);
161                 err = thermal_throttle_add_dev(sys_dev);
162                 mutex_unlock(&therm_cpu_lock);
163                 WARN_ON(err);
164                 break;
165         case CPU_UP_CANCELED:
166         case CPU_UP_CANCELED_FROZEN:
167         case CPU_DEAD:
168         case CPU_DEAD_FROZEN:
169                 mutex_lock(&therm_cpu_lock);
170                 thermal_throttle_remove_dev(sys_dev);
171                 mutex_unlock(&therm_cpu_lock);
172                 break;
173         }
174         return err ? NOTIFY_BAD : NOTIFY_OK;
175 }
176
177 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
178 {
179         .notifier_call = thermal_throttle_cpu_callback,
180 };
181
182 static __init int thermal_throttle_init_device(void)
183 {
184         unsigned int cpu = 0;
185         int err;
186
187         if (!atomic_read(&therm_throt_en))
188                 return 0;
189
190         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
191
192 #ifdef CONFIG_HOTPLUG_CPU
193         mutex_lock(&therm_cpu_lock);
194 #endif
195         /* connect live CPUs to sysfs */
196         for_each_online_cpu(cpu) {
197                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
198                 WARN_ON(err);
199         }
200 #ifdef CONFIG_HOTPLUG_CPU
201         mutex_unlock(&therm_cpu_lock);
202 #endif
203
204         return 0;
205 }
206 device_initcall(thermal_throttle_init_device);
207
208 #endif /* CONFIG_SYSFS */
209
210 /* Thermal transition interrupt handler */
211 static void intel_thermal_interrupt(void)
212 {
213         __u64 msr_val;
214
215         rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
216         if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT))
217                 mce_log_therm_throt_event(msr_val);
218 }
219
220 static void unexpected_thermal_interrupt(void)
221 {
222         printk(KERN_ERR "CPU%d: Unexpected LVT TMR interrupt!\n",
223                         smp_processor_id());
224         add_taint(TAINT_MACHINE_CHECK);
225 }
226
227 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
228
229 asmlinkage void smp_thermal_interrupt(struct pt_regs *regs)
230 {
231         exit_idle();
232         irq_enter();
233         inc_irq_stat(irq_thermal_count);
234         smp_thermal_vector();
235         irq_exit();
236         /* Ack only at the end to avoid potential reentry */
237         ack_APIC_irq();
238 }
239
240 void intel_init_thermal(struct cpuinfo_x86 *c)
241 {
242         unsigned int cpu = smp_processor_id();
243         int tm2 = 0;
244         u32 l, h;
245
246         /* Thermal monitoring depends on ACPI and clock modulation*/
247         if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
248                 return;
249
250         /*
251          * First check if its enabled already, in which case there might
252          * be some SMM goo which handles it, so we can't even put a handler
253          * since it might be delivered via SMI already:
254          */
255         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
256         h = apic_read(APIC_LVTTHMR);
257         if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
258                 printk(KERN_DEBUG
259                        "CPU%d: Thermal monitoring handled by SMI\n", cpu);
260                 return;
261         }
262
263         /* Check whether a vector already exists */
264         if (h & APIC_VECTOR_MASK) {
265                 printk(KERN_DEBUG
266                        "CPU%d: Thermal LVT vector (%#x) already installed\n",
267                        cpu, (h & APIC_VECTOR_MASK));
268                 return;
269         }
270
271         /* early Pentium M models use different method for enabling TM2 */
272         if (cpu_has(c, X86_FEATURE_TM2)) {
273                 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
274                         rdmsr(MSR_THERM2_CTL, l, h);
275                         if (l & MSR_THERM2_CTL_TM_SELECT)
276                                 tm2 = 1;
277                 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
278                         tm2 = 1;
279         }
280
281         /* We'll mask the thermal vector in the lapic till we're ready: */
282         h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
283         apic_write(APIC_LVTTHMR, h);
284
285         rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
286         wrmsr(MSR_IA32_THERM_INTERRUPT,
287                 l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
288
289         smp_thermal_vector = intel_thermal_interrupt;
290
291         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
292         wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
293
294         /* Unmask the thermal vector: */
295         l = apic_read(APIC_LVTTHMR);
296         apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
297
298         printk(KERN_INFO "CPU%d: Thermal monitoring enabled (%s)\n",
299                cpu, tm2 ? "TM2" : "TM1");
300
301         /* enable thermal throttle processing */
302         atomic_set(&therm_throt_en, 1);
303 }