[PATCH] fix "kvm: add vm exit profiling"
[linux-2.6] / kernel / cpu.c
1 /* CPU control.
2  * (C) 2001, 2002, 2003, 2004 Rusty Russell
3  *
4  * This code is licenced under the GPL.
5  */
6 #include <linux/proc_fs.h>
7 #include <linux/smp.h>
8 #include <linux/init.h>
9 #include <linux/notifier.h>
10 #include <linux/sched.h>
11 #include <linux/unistd.h>
12 #include <linux/cpu.h>
13 #include <linux/module.h>
14 #include <linux/kthread.h>
15 #include <linux/stop_machine.h>
16 #include <linux/mutex.h>
17
18 /* This protects CPUs going up and down... */
19 static DEFINE_MUTEX(cpu_add_remove_lock);
20 static DEFINE_MUTEX(cpu_bitmask_lock);
21
22 static __cpuinitdata RAW_NOTIFIER_HEAD(cpu_chain);
23
24 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
25  * Should always be manipulated under cpu_add_remove_lock
26  */
27 static int cpu_hotplug_disabled;
28
29 #ifdef CONFIG_HOTPLUG_CPU
30
31 /* Crappy recursive lock-takers in cpufreq! Complain loudly about idiots */
32 static struct task_struct *recursive;
33 static int recursive_depth;
34
35 void lock_cpu_hotplug(void)
36 {
37         struct task_struct *tsk = current;
38
39         if (tsk == recursive) {
40                 static int warnings = 10;
41                 if (warnings) {
42                         printk(KERN_ERR "Lukewarm IQ detected in hotplug locking\n");
43                         WARN_ON(1);
44                         warnings--;
45                 }
46                 recursive_depth++;
47                 return;
48         }
49         mutex_lock(&cpu_bitmask_lock);
50         recursive = tsk;
51 }
52 EXPORT_SYMBOL_GPL(lock_cpu_hotplug);
53
54 void unlock_cpu_hotplug(void)
55 {
56         WARN_ON(recursive != current);
57         if (recursive_depth) {
58                 recursive_depth--;
59                 return;
60         }
61         recursive = NULL;
62         mutex_unlock(&cpu_bitmask_lock);
63 }
64 EXPORT_SYMBOL_GPL(unlock_cpu_hotplug);
65
66 #endif  /* CONFIG_HOTPLUG_CPU */
67
68 /* Need to know about CPUs going up/down? */
69 int __cpuinit register_cpu_notifier(struct notifier_block *nb)
70 {
71         int ret;
72         mutex_lock(&cpu_add_remove_lock);
73         ret = raw_notifier_chain_register(&cpu_chain, nb);
74         mutex_unlock(&cpu_add_remove_lock);
75         return ret;
76 }
77
78 #ifdef CONFIG_HOTPLUG_CPU
79
80 EXPORT_SYMBOL(register_cpu_notifier);
81
82 void unregister_cpu_notifier(struct notifier_block *nb)
83 {
84         mutex_lock(&cpu_add_remove_lock);
85         raw_notifier_chain_unregister(&cpu_chain, nb);
86         mutex_unlock(&cpu_add_remove_lock);
87 }
88 EXPORT_SYMBOL(unregister_cpu_notifier);
89
90 static inline void check_for_tasks(int cpu)
91 {
92         struct task_struct *p;
93
94         write_lock_irq(&tasklist_lock);
95         for_each_process(p) {
96                 if (task_cpu(p) == cpu &&
97                     (!cputime_eq(p->utime, cputime_zero) ||
98                      !cputime_eq(p->stime, cputime_zero)))
99                         printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d\
100                                 (state = %ld, flags = %lx) \n",
101                                  p->comm, p->pid, cpu, p->state, p->flags);
102         }
103         write_unlock_irq(&tasklist_lock);
104 }
105
106 /* Take this CPU down. */
107 static int take_cpu_down(void *unused)
108 {
109         int err;
110
111         /* Ensure this CPU doesn't handle any more interrupts. */
112         err = __cpu_disable();
113         if (err < 0)
114                 return err;
115
116         /* Force idle task to run as soon as we yield: it should
117            immediately notice cpu is offline and die quickly. */
118         sched_idle_next();
119         return 0;
120 }
121
122 /* Requires cpu_add_remove_lock to be held */
123 static int _cpu_down(unsigned int cpu)
124 {
125         int err;
126         struct task_struct *p;
127         cpumask_t old_allowed, tmp;
128
129         if (num_online_cpus() == 1)
130                 return -EBUSY;
131
132         if (!cpu_online(cpu))
133                 return -EINVAL;
134
135         err = raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE,
136                                                 (void *)(long)cpu);
137         if (err == NOTIFY_BAD) {
138                 printk("%s: attempt to take down CPU %u failed\n",
139                                 __FUNCTION__, cpu);
140                 return -EINVAL;
141         }
142
143         /* Ensure that we are not runnable on dying cpu */
144         old_allowed = current->cpus_allowed;
145         tmp = CPU_MASK_ALL;
146         cpu_clear(cpu, tmp);
147         set_cpus_allowed(current, tmp);
148
149         mutex_lock(&cpu_bitmask_lock);
150         p = __stop_machine_run(take_cpu_down, NULL, cpu);
151         mutex_unlock(&cpu_bitmask_lock);
152
153         if (IS_ERR(p) || cpu_online(cpu)) {
154                 /* CPU didn't die: tell everyone.  Can't complain. */
155                 if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED,
156                                 (void *)(long)cpu) == NOTIFY_BAD)
157                         BUG();
158
159                 if (IS_ERR(p)) {
160                         err = PTR_ERR(p);
161                         goto out_allowed;
162                 }
163                 goto out_thread;
164         }
165
166         /* Wait for it to sleep (leaving idle task). */
167         while (!idle_cpu(cpu))
168                 yield();
169
170         /* This actually kills the CPU. */
171         __cpu_die(cpu);
172
173         /* Move it here so it can run. */
174         kthread_bind(p, get_cpu());
175         put_cpu();
176
177         /* CPU is completely dead: tell everyone.  Too late to complain. */
178         if (raw_notifier_call_chain(&cpu_chain, CPU_DEAD,
179                         (void *)(long)cpu) == NOTIFY_BAD)
180                 BUG();
181
182         check_for_tasks(cpu);
183
184 out_thread:
185         err = kthread_stop(p);
186 out_allowed:
187         set_cpus_allowed(current, old_allowed);
188         return err;
189 }
190
191 int cpu_down(unsigned int cpu)
192 {
193         int err = 0;
194
195         mutex_lock(&cpu_add_remove_lock);
196         if (cpu_hotplug_disabled)
197                 err = -EBUSY;
198         else
199                 err = _cpu_down(cpu);
200
201         mutex_unlock(&cpu_add_remove_lock);
202         return err;
203 }
204 #endif /*CONFIG_HOTPLUG_CPU*/
205
206 /* Requires cpu_add_remove_lock to be held */
207 static int __cpuinit _cpu_up(unsigned int cpu)
208 {
209         int ret;
210         void *hcpu = (void *)(long)cpu;
211
212         if (cpu_online(cpu) || !cpu_present(cpu))
213                 return -EINVAL;
214
215         ret = raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE, hcpu);
216         if (ret == NOTIFY_BAD) {
217                 printk("%s: attempt to bring up CPU %u failed\n",
218                                 __FUNCTION__, cpu);
219                 ret = -EINVAL;
220                 goto out_notify;
221         }
222
223         /* Arch-specific enabling code. */
224         mutex_lock(&cpu_bitmask_lock);
225         ret = __cpu_up(cpu);
226         mutex_unlock(&cpu_bitmask_lock);
227         if (ret != 0)
228                 goto out_notify;
229         BUG_ON(!cpu_online(cpu));
230
231         /* Now call notifier in preparation. */
232         raw_notifier_call_chain(&cpu_chain, CPU_ONLINE, hcpu);
233
234 out_notify:
235         if (ret != 0)
236                 raw_notifier_call_chain(&cpu_chain,
237                                 CPU_UP_CANCELED, hcpu);
238
239         return ret;
240 }
241
242 int __cpuinit cpu_up(unsigned int cpu)
243 {
244         int err = 0;
245
246         mutex_lock(&cpu_add_remove_lock);
247         if (cpu_hotplug_disabled)
248                 err = -EBUSY;
249         else
250                 err = _cpu_up(cpu);
251
252         mutex_unlock(&cpu_add_remove_lock);
253         return err;
254 }
255
256 #ifdef CONFIG_SUSPEND_SMP
257 static cpumask_t frozen_cpus;
258
259 int disable_nonboot_cpus(void)
260 {
261         int cpu, first_cpu, error = 0;
262
263         mutex_lock(&cpu_add_remove_lock);
264         first_cpu = first_cpu(cpu_present_map);
265         if (!cpu_online(first_cpu)) {
266                 error = _cpu_up(first_cpu);
267                 if (error) {
268                         printk(KERN_ERR "Could not bring CPU%d up.\n",
269                                 first_cpu);
270                         goto out;
271                 }
272         }
273
274         /* We take down all of the non-boot CPUs in one shot to avoid races
275          * with the userspace trying to use the CPU hotplug at the same time
276          */
277         cpus_clear(frozen_cpus);
278         printk("Disabling non-boot CPUs ...\n");
279         for_each_online_cpu(cpu) {
280                 if (cpu == first_cpu)
281                         continue;
282                 error = _cpu_down(cpu);
283                 if (!error) {
284                         cpu_set(cpu, frozen_cpus);
285                         printk("CPU%d is down\n", cpu);
286                 } else {
287                         printk(KERN_ERR "Error taking CPU%d down: %d\n",
288                                 cpu, error);
289                         break;
290                 }
291         }
292         if (!error) {
293                 BUG_ON(num_online_cpus() > 1);
294                 /* Make sure the CPUs won't be enabled by someone else */
295                 cpu_hotplug_disabled = 1;
296         } else {
297                 printk(KERN_ERR "Non-boot CPUs are not disabled\n");
298         }
299 out:
300         mutex_unlock(&cpu_add_remove_lock);
301         return error;
302 }
303
304 void enable_nonboot_cpus(void)
305 {
306         int cpu, error;
307
308         /* Allow everyone to use the CPU hotplug again */
309         mutex_lock(&cpu_add_remove_lock);
310         cpu_hotplug_disabled = 0;
311         mutex_unlock(&cpu_add_remove_lock);
312
313         printk("Enabling non-boot CPUs ...\n");
314         for_each_cpu_mask(cpu, frozen_cpus) {
315                 error = cpu_up(cpu);
316                 if (!error) {
317                         printk("CPU%d is up\n", cpu);
318                         continue;
319                 }
320                 printk(KERN_WARNING "Error taking CPU%d up: %d\n",
321                         cpu, error);
322         }
323         cpus_clear(frozen_cpus);
324 }
325 #endif