cpu hotplug: thermal_throttle: fix cpu hotplug error handling
[linux-2.6.git] / arch / x86 / kernel / cpu / mcheck / therm_throt.c
1 /*
2  * linux/arch/i386/kernel/cpu/mcheck/therm_throt.c
3  *
4  * Thermal throttle event support code (such as syslog messaging and rate
5  * limiting) that was factored out from x86_64 (mce_intel.c) and i386 (p4.c).
6  * This allows consistent reporting of CPU thermal throttle events.
7  *
8  * Maintains a counter in /sys that keeps track of the number of thermal
9  * events, such that the user knows how bad the thermal problem might be
10  * (since the logging to syslog and mcelog is rate limited).
11  *
12  * Author: Dmitriy Zavin (dmitriyz@google.com)
13  *
14  * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
15  *          Inspired by Ross Biro's and Al Borchers' counter code.
16  */
17
18 #include <linux/percpu.h>
19 #include <linux/sysdev.h>
20 #include <linux/cpu.h>
21 #include <asm/cpu.h>
22 #include <linux/notifier.h>
23 #include <linux/jiffies.h>
24 #include <asm/therm_throt.h>
25
26 /* How long to wait between reporting thermal events */
27 #define CHECK_INTERVAL              (300 * HZ)
28
29 static DEFINE_PER_CPU(__u64, next_check) = INITIAL_JIFFIES;
30 static DEFINE_PER_CPU(unsigned long, thermal_throttle_count);
31 atomic_t therm_throt_en = ATOMIC_INIT(0);
32
33 #ifdef CONFIG_SYSFS
34 #define define_therm_throt_sysdev_one_ro(_name)                              \
35         static SYSDEV_ATTR(_name, 0444, therm_throt_sysdev_show_##_name, NULL)
36
37 #define define_therm_throt_sysdev_show_func(name)                            \
38 static ssize_t therm_throt_sysdev_show_##name(struct sys_device *dev,        \
39                                               char *buf)                     \
40 {                                                                            \
41         unsigned int cpu = dev->id;                                          \
42         ssize_t ret;                                                         \
43                                                                              \
44         preempt_disable();              /* CPU hotplug */                    \
45         if (cpu_online(cpu))                                                 \
46                 ret = sprintf(buf, "%lu\n",                                  \
47                               per_cpu(thermal_throttle_##name, cpu));        \
48         else                                                                 \
49                 ret = 0;                                                     \
50         preempt_enable();                                                    \
51                                                                              \
52         return ret;                                                          \
53 }
54
55 define_therm_throt_sysdev_show_func(count);
56 define_therm_throt_sysdev_one_ro(count);
57
58 static struct attribute *thermal_throttle_attrs[] = {
59         &attr_count.attr,
60         NULL
61 };
62
63 static struct attribute_group thermal_throttle_attr_group = {
64         .attrs = thermal_throttle_attrs,
65         .name = "thermal_throttle"
66 };
67 #endif /* CONFIG_SYSFS */
68
69 /***
70  * therm_throt_process - Process thermal throttling event from interrupt
71  * @curr: Whether the condition is current or not (boolean), since the
72  *        thermal interrupt normally gets called both when the thermal
73  *        event begins and once the event has ended.
74  *
75  * This function is called by the thermal interrupt after the
76  * IRQ has been acknowledged.
77  *
78  * It will take care of rate limiting and printing messages to the syslog.
79  *
80  * Returns: 0 : Event should NOT be further logged, i.e. still in
81  *              "timeout" from previous log message.
82  *          1 : Event should be logged further, and a message has been
83  *              printed to the syslog.
84  */
85 int therm_throt_process(int curr)
86 {
87         unsigned int cpu = smp_processor_id();
88         __u64 tmp_jiffs = get_jiffies_64();
89
90         if (curr)
91                 __get_cpu_var(thermal_throttle_count)++;
92
93         if (time_before64(tmp_jiffs, __get_cpu_var(next_check)))
94                 return 0;
95
96         __get_cpu_var(next_check) = tmp_jiffs + CHECK_INTERVAL;
97
98         /* if we just entered the thermal event */
99         if (curr) {
100                 printk(KERN_CRIT "CPU%d: Temperature above threshold, "
101                        "cpu clock throttled (total events = %lu)\n", cpu,
102                        __get_cpu_var(thermal_throttle_count));
103
104                 add_taint(TAINT_MACHINE_CHECK);
105         } else {
106                 printk(KERN_CRIT "CPU%d: Temperature/speed normal\n", cpu);
107         }
108
109         return 1;
110 }
111
112 #ifdef CONFIG_SYSFS
113 /* Add/Remove thermal_throttle interface for CPU device */
114 static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev)
115 {
116         return sysfs_create_group(&sys_dev->kobj, &thermal_throttle_attr_group);
117 }
118
119 static __cpuinit void thermal_throttle_remove_dev(struct sys_device *sys_dev)
120 {
121         return sysfs_remove_group(&sys_dev->kobj, &thermal_throttle_attr_group);
122 }
123
124 /* Mutex protecting device creation against CPU hotplug */
125 static DEFINE_MUTEX(therm_cpu_lock);
126
127 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
128 static __cpuinit int thermal_throttle_cpu_callback(struct notifier_block *nfb,
129                                                    unsigned long action,
130                                                    void *hcpu)
131 {
132         unsigned int cpu = (unsigned long)hcpu;
133         struct sys_device *sys_dev;
134         int err = 0;
135
136         sys_dev = get_cpu_sysdev(cpu);
137         switch (action) {
138         case CPU_UP_PREPARE:
139         case CPU_UP_PREPARE_FROZEN:
140                 mutex_lock(&therm_cpu_lock);
141                 err = thermal_throttle_add_dev(sys_dev);
142                 mutex_unlock(&therm_cpu_lock);
143                 WARN_ON(err);
144                 break;
145         case CPU_UP_CANCELED:
146         case CPU_UP_CANCELED_FROZEN:
147         case CPU_DEAD:
148         case CPU_DEAD_FROZEN:
149                 mutex_lock(&therm_cpu_lock);
150                 thermal_throttle_remove_dev(sys_dev);
151                 mutex_unlock(&therm_cpu_lock);
152                 break;
153         }
154         return err ? NOTIFY_BAD : NOTIFY_OK;
155 }
156
157 static struct notifier_block thermal_throttle_cpu_notifier __cpuinitdata =
158 {
159         .notifier_call = thermal_throttle_cpu_callback,
160 };
161
162 static __init int thermal_throttle_init_device(void)
163 {
164         unsigned int cpu = 0;
165         int err;
166
167         if (!atomic_read(&therm_throt_en))
168                 return 0;
169
170         register_hotcpu_notifier(&thermal_throttle_cpu_notifier);
171
172 #ifdef CONFIG_HOTPLUG_CPU
173         mutex_lock(&therm_cpu_lock);
174 #endif
175         /* connect live CPUs to sysfs */
176         for_each_online_cpu(cpu) {
177                 err = thermal_throttle_add_dev(get_cpu_sysdev(cpu));
178                 WARN_ON(err);
179         }
180 #ifdef CONFIG_HOTPLUG_CPU
181         mutex_unlock(&therm_cpu_lock);
182 #endif
183
184         return 0;
185 }
186
187 device_initcall(thermal_throttle_init_device);
188 #endif /* CONFIG_SYSFS */