xref: /linux/kernel/cpu.c (revision c80544dc0b87bb65038355e7aafdc30be16b26ab)
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 = %x) \n",
101 				 p->comm, p->pid, cpu, p->state, p->flags);
102 	}
103 	write_unlock_irq(&tasklist_lock);
104 }
105 
106 struct take_cpu_down_param {
107 	unsigned long mod;
108 	void *hcpu;
109 };
110 
111 /* Take this CPU down. */
112 static int take_cpu_down(void *_param)
113 {
114 	struct take_cpu_down_param *param = _param;
115 	int err;
116 
117 	raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
118 				param->hcpu);
119 	/* Ensure this CPU doesn't handle any more interrupts. */
120 	err = __cpu_disable();
121 	if (err < 0)
122 		return err;
123 
124 	/* Force idle task to run as soon as we yield: it should
125 	   immediately notice cpu is offline and die quickly. */
126 	sched_idle_next();
127 	return 0;
128 }
129 
130 /* Requires cpu_add_remove_lock to be held */
131 static int _cpu_down(unsigned int cpu, int tasks_frozen)
132 {
133 	int err, nr_calls = 0;
134 	struct task_struct *p;
135 	cpumask_t old_allowed, tmp;
136 	void *hcpu = (void *)(long)cpu;
137 	unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
138 	struct take_cpu_down_param tcd_param = {
139 		.mod = mod,
140 		.hcpu = hcpu,
141 	};
142 
143 	if (num_online_cpus() == 1)
144 		return -EBUSY;
145 
146 	if (!cpu_online(cpu))
147 		return -EINVAL;
148 
149 	raw_notifier_call_chain(&cpu_chain, CPU_LOCK_ACQUIRE, hcpu);
150 	err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
151 					hcpu, -1, &nr_calls);
152 	if (err == NOTIFY_BAD) {
153 		nr_calls--;
154 		__raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
155 					  hcpu, nr_calls, NULL);
156 		printk("%s: attempt to take down CPU %u failed\n",
157 				__FUNCTION__, cpu);
158 		err = -EINVAL;
159 		goto out_release;
160 	}
161 
162 	/* Ensure that we are not runnable on dying cpu */
163 	old_allowed = current->cpus_allowed;
164 	tmp = CPU_MASK_ALL;
165 	cpu_clear(cpu, tmp);
166 	set_cpus_allowed(current, tmp);
167 
168 	mutex_lock(&cpu_bitmask_lock);
169 	p = __stop_machine_run(take_cpu_down, &tcd_param, cpu);
170 	mutex_unlock(&cpu_bitmask_lock);
171 
172 	if (IS_ERR(p) || cpu_online(cpu)) {
173 		/* CPU didn't die: tell everyone.  Can't complain. */
174 		if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
175 					    hcpu) == NOTIFY_BAD)
176 			BUG();
177 
178 		if (IS_ERR(p)) {
179 			err = PTR_ERR(p);
180 			goto out_allowed;
181 		}
182 		goto out_thread;
183 	}
184 
185 	/* Wait for it to sleep (leaving idle task). */
186 	while (!idle_cpu(cpu))
187 		yield();
188 
189 	/* This actually kills the CPU. */
190 	__cpu_die(cpu);
191 
192 	/* CPU is completely dead: tell everyone.  Too late to complain. */
193 	if (raw_notifier_call_chain(&cpu_chain, CPU_DEAD | mod,
194 				    hcpu) == NOTIFY_BAD)
195 		BUG();
196 
197 	check_for_tasks(cpu);
198 
199 out_thread:
200 	err = kthread_stop(p);
201 out_allowed:
202 	set_cpus_allowed(current, old_allowed);
203 out_release:
204 	raw_notifier_call_chain(&cpu_chain, CPU_LOCK_RELEASE, hcpu);
205 	return err;
206 }
207 
208 int cpu_down(unsigned int cpu)
209 {
210 	int err = 0;
211 
212 	mutex_lock(&cpu_add_remove_lock);
213 	if (cpu_hotplug_disabled)
214 		err = -EBUSY;
215 	else
216 		err = _cpu_down(cpu, 0);
217 
218 	mutex_unlock(&cpu_add_remove_lock);
219 	return err;
220 }
221 #endif /*CONFIG_HOTPLUG_CPU*/
222 
223 /* Requires cpu_add_remove_lock to be held */
224 static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
225 {
226 	int ret, nr_calls = 0;
227 	void *hcpu = (void *)(long)cpu;
228 	unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
229 
230 	if (cpu_online(cpu) || !cpu_present(cpu))
231 		return -EINVAL;
232 
233 	raw_notifier_call_chain(&cpu_chain, CPU_LOCK_ACQUIRE, hcpu);
234 	ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
235 							-1, &nr_calls);
236 	if (ret == NOTIFY_BAD) {
237 		nr_calls--;
238 		printk("%s: attempt to bring up CPU %u failed\n",
239 				__FUNCTION__, cpu);
240 		ret = -EINVAL;
241 		goto out_notify;
242 	}
243 
244 	/* Arch-specific enabling code. */
245 	mutex_lock(&cpu_bitmask_lock);
246 	ret = __cpu_up(cpu);
247 	mutex_unlock(&cpu_bitmask_lock);
248 	if (ret != 0)
249 		goto out_notify;
250 	BUG_ON(!cpu_online(cpu));
251 
252 	/* Now call notifier in preparation. */
253 	raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);
254 
255 out_notify:
256 	if (ret != 0)
257 		__raw_notifier_call_chain(&cpu_chain,
258 				CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
259 	raw_notifier_call_chain(&cpu_chain, CPU_LOCK_RELEASE, hcpu);
260 
261 	return ret;
262 }
263 
264 int __cpuinit cpu_up(unsigned int cpu)
265 {
266 	int err = 0;
267 
268 	mutex_lock(&cpu_add_remove_lock);
269 	if (cpu_hotplug_disabled)
270 		err = -EBUSY;
271 	else
272 		err = _cpu_up(cpu, 0);
273 
274 	mutex_unlock(&cpu_add_remove_lock);
275 	return err;
276 }
277 
278 #ifdef CONFIG_PM_SLEEP_SMP
279 static cpumask_t frozen_cpus;
280 
281 int disable_nonboot_cpus(void)
282 {
283 	int cpu, first_cpu, error = 0;
284 
285 	mutex_lock(&cpu_add_remove_lock);
286 	first_cpu = first_cpu(cpu_online_map);
287 	/* We take down all of the non-boot CPUs in one shot to avoid races
288 	 * with the userspace trying to use the CPU hotplug at the same time
289 	 */
290 	cpus_clear(frozen_cpus);
291 	printk("Disabling non-boot CPUs ...\n");
292 	for_each_online_cpu(cpu) {
293 		if (cpu == first_cpu)
294 			continue;
295 		error = _cpu_down(cpu, 1);
296 		if (!error) {
297 			cpu_set(cpu, frozen_cpus);
298 			printk("CPU%d is down\n", cpu);
299 		} else {
300 			printk(KERN_ERR "Error taking CPU%d down: %d\n",
301 				cpu, error);
302 			break;
303 		}
304 	}
305 	if (!error) {
306 		BUG_ON(num_online_cpus() > 1);
307 		/* Make sure the CPUs won't be enabled by someone else */
308 		cpu_hotplug_disabled = 1;
309 	} else {
310 		printk(KERN_ERR "Non-boot CPUs are not disabled\n");
311 	}
312 	mutex_unlock(&cpu_add_remove_lock);
313 	return error;
314 }
315 
316 void enable_nonboot_cpus(void)
317 {
318 	int cpu, error;
319 
320 	/* Allow everyone to use the CPU hotplug again */
321 	mutex_lock(&cpu_add_remove_lock);
322 	cpu_hotplug_disabled = 0;
323 	if (cpus_empty(frozen_cpus))
324 		goto out;
325 
326 	printk("Enabling non-boot CPUs ...\n");
327 	for_each_cpu_mask(cpu, frozen_cpus) {
328 		error = _cpu_up(cpu, 1);
329 		if (!error) {
330 			printk("CPU%d is up\n", cpu);
331 			continue;
332 		}
333 		printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
334 	}
335 	cpus_clear(frozen_cpus);
336 out:
337 	mutex_unlock(&cpu_add_remove_lock);
338 }
339 #endif /* CONFIG_PM_SLEEP_SMP */
340