[PATCH] Disable CPU hotplug during suspend
[linux-2.6.git] / 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 BLOCKING_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         mutex_unlock(&cpu_bitmask_lock);
62         recursive = NULL;
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         return blocking_notifier_chain_register(&cpu_chain, nb);
72 }
73
74 #ifdef CONFIG_HOTPLUG_CPU
75
76 EXPORT_SYMBOL(register_cpu_notifier);
77
78 void unregister_cpu_notifier(struct notifier_block *nb)
79 {
80         blocking_notifier_chain_unregister(&cpu_chain, nb);
81 }
82 EXPORT_SYMBOL(unregister_cpu_notifier);
83
84 static inline void check_for_tasks(int cpu)
85 {
86         struct task_struct *p;
87
88         write_lock_irq(&tasklist_lock);
89         for_each_process(p) {
90                 if (task_cpu(p) == cpu &&
91                     (!cputime_eq(p->utime, cputime_zero) ||
92                      !cputime_eq(p->stime, cputime_zero)))
93                         printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d\
94                                 (state = %ld, flags = %lx) \n",
95                                  p->comm, p->pid, cpu, p->state, p->flags);
96         }
97         write_unlock_irq(&tasklist_lock);
98 }
99
100 /* Take this CPU down. */
101 static int take_cpu_down(void *unused)
102 {
103         int err;
104
105         /* Ensure this CPU doesn't handle any more interrupts. */
106         err = __cpu_disable();
107         if (err < 0)
108                 return err;
109
110         /* Force idle task to run as soon as we yield: it should
111            immediately notice cpu is offline and die quickly. */
112         sched_idle_next();
113         return 0;
114 }
115
116 /* Requires cpu_add_remove_lock to be held */
117 static int _cpu_down(unsigned int cpu)
118 {
119         int err;
120         struct task_struct *p;
121         cpumask_t old_allowed, tmp;
122
123         if (num_online_cpus() == 1)
124                 return -EBUSY;
125
126         if (!cpu_online(cpu))
127                 return -EINVAL;
128
129         err = blocking_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE,
130                                                 (void *)(long)cpu);
131         if (err == NOTIFY_BAD) {
132                 printk("%s: attempt to take down CPU %u failed\n",
133                                 __FUNCTION__, cpu);
134                 return -EINVAL;
135         }
136
137         /* Ensure that we are not runnable on dying cpu */
138         old_allowed = current->cpus_allowed;
139         tmp = CPU_MASK_ALL;
140         cpu_clear(cpu, tmp);
141         set_cpus_allowed(current, tmp);
142
143         mutex_lock(&cpu_bitmask_lock);
144         p = __stop_machine_run(take_cpu_down, NULL, cpu);
145         mutex_unlock(&cpu_bitmask_lock);
146
147         if (IS_ERR(p)) {
148                 /* CPU didn't die: tell everyone.  Can't complain. */
149                 if (blocking_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED,
150                                 (void *)(long)cpu) == NOTIFY_BAD)
151                         BUG();
152
153                 err = PTR_ERR(p);
154                 goto out_allowed;
155         }
156
157         if (cpu_online(cpu))
158                 goto out_thread;
159
160         /* Wait for it to sleep (leaving idle task). */
161         while (!idle_cpu(cpu))
162                 yield();
163
164         /* This actually kills the CPU. */
165         __cpu_die(cpu);
166
167         /* Move it here so it can run. */
168         kthread_bind(p, get_cpu());
169         put_cpu();
170
171         /* CPU is completely dead: tell everyone.  Too late to complain. */
172         if (blocking_notifier_call_chain(&cpu_chain, CPU_DEAD,
173                         (void *)(long)cpu) == NOTIFY_BAD)
174                 BUG();
175
176         check_for_tasks(cpu);
177
178 out_thread:
179         err = kthread_stop(p);
180 out_allowed:
181         set_cpus_allowed(current, old_allowed);
182         return err;
183 }
184
185 int cpu_down(unsigned int cpu)
186 {
187         int err = 0;
188
189         mutex_lock(&cpu_add_remove_lock);
190         if (cpu_hotplug_disabled)
191                 err = -EBUSY;
192         else
193                 err = _cpu_down(cpu);
194
195         mutex_unlock(&cpu_add_remove_lock);
196         return err;
197 }
198 #endif /*CONFIG_HOTPLUG_CPU*/
199
200 /* Requires cpu_add_remove_lock to be held */
201 static int __devinit _cpu_up(unsigned int cpu)
202 {
203         int ret;
204         void *hcpu = (void *)(long)cpu;
205
206         if (cpu_online(cpu) || !cpu_present(cpu))
207                 return -EINVAL;
208
209         ret = blocking_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE, hcpu);
210         if (ret == NOTIFY_BAD) {
211                 printk("%s: attempt to bring up CPU %u failed\n",
212                                 __FUNCTION__, cpu);
213                 ret = -EINVAL;
214                 goto out_notify;
215         }
216
217         /* Arch-specific enabling code. */
218         mutex_lock(&cpu_bitmask_lock);
219         ret = __cpu_up(cpu);
220         mutex_unlock(&cpu_bitmask_lock);
221         if (ret != 0)
222                 goto out_notify;
223         BUG_ON(!cpu_online(cpu));
224
225         /* Now call notifier in preparation. */
226         blocking_notifier_call_chain(&cpu_chain, CPU_ONLINE, hcpu);
227
228 out_notify:
229         if (ret != 0)
230                 blocking_notifier_call_chain(&cpu_chain,
231                                 CPU_UP_CANCELED, hcpu);
232
233         return ret;
234 }
235
236 int __devinit cpu_up(unsigned int cpu)
237 {
238         int err = 0;
239
240         mutex_lock(&cpu_add_remove_lock);
241         if (cpu_hotplug_disabled)
242                 err = -EBUSY;
243         else
244                 err = _cpu_up(cpu);
245
246         mutex_unlock(&cpu_add_remove_lock);
247         return err;
248 }
249
250 #ifdef CONFIG_SUSPEND_SMP
251 static cpumask_t frozen_cpus;
252
253 int disable_nonboot_cpus(void)
254 {
255         int cpu, first_cpu, error;
256
257         mutex_lock(&cpu_add_remove_lock);
258         first_cpu = first_cpu(cpu_present_map);
259         if (!cpu_online(first_cpu)) {
260                 error = _cpu_up(first_cpu);
261                 if (error) {
262                         printk(KERN_ERR "Could not bring CPU%d up.\n",
263                                 first_cpu);
264                         goto out;
265                 }
266         }
267         error = set_cpus_allowed(current, cpumask_of_cpu(first_cpu));
268         if (error) {
269                 printk(KERN_ERR "Could not run on CPU%d\n", first_cpu);
270                 goto out;
271         }
272         /* We take down all of the non-boot CPUs in one shot to avoid races
273          * with the userspace trying to use the CPU hotplug at the same time
274          */
275         cpus_clear(frozen_cpus);
276         printk("Disabling non-boot CPUs ...\n");
277         for_each_online_cpu(cpu) {
278                 if (cpu == first_cpu)
279                         continue;
280                 error = _cpu_down(cpu);
281                 if (!error) {
282                         cpu_set(cpu, frozen_cpus);
283                         printk("CPU%d is down\n", cpu);
284                 } else {
285                         printk(KERN_ERR "Error taking CPU%d down: %d\n",
286                                 cpu, error);
287                         break;
288                 }
289         }
290         if (!error) {
291                 BUG_ON(num_online_cpus() > 1);
292                 /* Make sure the CPUs won't be enabled by someone else */
293                 cpu_hotplug_disabled = 1;
294         } else {
295                 printk(KERN_ERR "Non-boot CPUs are not disabled");
296         }
297 out:
298         mutex_unlock(&cpu_add_remove_lock);
299         return error;
300 }
301
302 void enable_nonboot_cpus(void)
303 {
304         int cpu, error;
305
306         /* Allow everyone to use the CPU hotplug again */
307         mutex_lock(&cpu_add_remove_lock);
308         cpu_hotplug_disabled = 0;
309         mutex_unlock(&cpu_add_remove_lock);
310
311         printk("Enabling non-boot CPUs ...\n");
312         for_each_cpu_mask(cpu, frozen_cpus) {
313                 error = cpu_up(cpu);
314                 if (!error) {
315                         printk("CPU%d is up\n", cpu);
316                         continue;
317                 }
318                 printk(KERN_WARNING "Error taking CPU%d up: %d\n",
319                         cpu, error);
320         }
321         cpus_clear(frozen_cpus);
322 }
323 #endif