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