linux kernel-2.6.28.6 / CPU control

/* CPU control.
 * (C) 2001, 2002, 2003, 2004 Rusty Russell
 *
 * This code is licenced under the GPL.
 */
#include <linux/proc_fs.h>
#include <linux/smp.h>
#include <linux/init.h>
#include <linux/notifier.h>
#include <linux/sched.h>
#include <linux/unistd.h>
#include <linux/cpu.h>
#include <linux/module.h>
#include <linux/kthread.h>
#include <linux/stop_machine.h>
#include <linux/mutex.h>

/*
 * Represents all cpu's present in the system
 * In systems capable of hotplug, this map could dynamically grow
 * as new cpu's are detected in the system via any platform specific
 * method, such as ACPI for e.g.
 */
cpumask_t cpu_present_map __read_mostly;
EXPORT_SYMBOL(cpu_present_map);

#ifndef CONFIG_SMP

/*
 * Represents all cpu's that are currently online.
 */
cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL;
EXPORT_SYMBOL(cpu_online_map);

cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL;
EXPORT_SYMBOL(cpu_possible_map);

#else /* CONFIG_SMP */

/* Serializes the updates to cpu_online_map, cpu_present_map */
static DEFINE_MUTEX(cpu_add_remove_lock);

static __cpuinitdata RAW_NOTIFIER_HEAD(cpu_chain);

/* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
 * Should always be manipulated under cpu_add_remove_lock
 */
static int cpu_hotplug_disabled;

static struct {
 struct task_struct *active_writer;
 struct mutex lock; /* Synchronizes accesses to refcount, */
 /*
  * Also blocks the new readers during
  * an ongoing cpu hotplug operation.
  */
 int refcount;
} cpu_hotplug;

void __init cpu_hotplug_init(void)
{
 cpu_hotplug.active_writer = NULL;
 mutex_init(&cpu_hotplug.lock);
 cpu_hotplug.refcount = 0;
}

cpumask_t cpu_active_map;

#ifdef CONFIG_HOTPLUG_CPU

void get_online_cpus(void)
{
 might_sleep();
 if (cpu_hotplug.active_writer == current)
  return;
 mutex_lock(&cpu_hotplug.lock);
 cpu_hotplug.refcount++;
 mutex_unlock(&cpu_hotplug.lock);

}
EXPORT_SYMBOL_GPL(get_online_cpus);

void put_online_cpus(void)
{
 if (cpu_hotplug.active_writer == current)
  return;
 mutex_lock(&cpu_hotplug.lock);
 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
  wake_up_process(cpu_hotplug.active_writer);
 mutex_unlock(&cpu_hotplug.lock);

}
EXPORT_SYMBOL_GPL(put_online_cpus);

#endif /* CONFIG_HOTPLUG_CPU */

/*
 * The following two API's must be used when attempting
 * to serialize the updates to cpu_online_map, cpu_present_map.
 */
void cpu_maps_update_begin(void)
{
 mutex_lock(&cpu_add_remove_lock);
}

void cpu_maps_update_done(void)
{
 mutex_unlock(&cpu_add_remove_lock);
}

/*
 * This ensures that the hotplug operation can begin only when the
 * refcount goes to zero.
 *
 * Note that during a cpu-hotplug operation, the new readers, if any,
 * will be blocked by the cpu_hotplug.lock
 *
 * Since cpu_hotplug_begin() is always called after invoking
 * cpu_maps_update_begin(), we can be sure that only one writer is active.
 *
 * Note that theoretically, there is a possibility of a livelock:
 * - Refcount goes to zero, last reader wakes up the sleeping
 *   writer.
 * - Last reader unlocks the cpu_hotplug.lock.
 * - A new reader arrives at this moment, bumps up the refcount.
 * - The writer acquires the cpu_hotplug.lock finds the refcount
 *   non zero and goes to sleep again.
 *
 * However, this is very difficult to achieve in practice since
 * get_online_cpus() not an api which is called all that often.
 *
 */
static void cpu_hotplug_begin(void)
{
 cpu_hotplug.active_writer = current;

 for (;;) {
  mutex_lock(&cpu_hotplug.lock);
  if (likely(!cpu_hotplug.refcount))
   break;
  __set_current_state(TASK_UNINTERRUPTIBLE);
  mutex_unlock(&cpu_hotplug.lock);
  schedule();
 }
}

static void cpu_hotplug_done(void)
{
 cpu_hotplug.active_writer = NULL;
 mutex_unlock(&cpu_hotplug.lock);
}
/* Need to know about CPUs going up/down? */
int __ref register_cpu_notifier(struct notifier_block *nb)
{
 int ret;
 cpu_maps_update_begin();
 ret = raw_notifier_chain_register(&cpu_chain, nb);
 cpu_maps_update_done();
 return ret;
}

#ifdef CONFIG_HOTPLUG_CPU

EXPORT_SYMBOL(register_cpu_notifier);

void __ref unregister_cpu_notifier(struct notifier_block *nb)
{
 cpu_maps_update_begin();
 raw_notifier_chain_unregister(&cpu_chain, nb);
 cpu_maps_update_done();
}
EXPORT_SYMBOL(unregister_cpu_notifier);

static inline void check_for_tasks(int cpu)
{
 struct task_struct *p;

 write_lock_irq(&tasklist_lock);
 for_each_process(p) {
  if (task_cpu(p) == cpu &&
      (!cputime_eq(p->utime, cputime_zero) ||
       !cputime_eq(p->stime, cputime_zero)))
   printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d/
    (state = %ld, flags = %x) /n",
     p->comm, task_pid_nr(p), cpu,
     p->state, p->flags);
 }
 write_unlock_irq(&tasklist_lock);
}

struct take_cpu_down_param {
 unsigned long mod;
 void *hcpu;
};

/* Take this CPU down. */
static int __ref take_cpu_down(void *_param)
{
 struct take_cpu_down_param *param = _param;
 int err;

 /* Ensure this CPU doesn't handle any more interrupts. */
 err = __cpu_disable();
 if (err < 0)
  return err;

 raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
    param->hcpu);

 /* Force idle task to run as soon as we yield: it should
    immediately notice cpu is offline and die quickly. */
 sched_idle_next();
 return 0;
}

/* Requires cpu_add_remove_lock to be held */
static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
{
 int err, nr_calls = 0;
 cpumask_t old_allowed, tmp;
 void *hcpu = (void *)(long)cpu;
 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
 struct take_cpu_down_param tcd_param = {
  .mod = mod,
  .hcpu = hcpu,
 };

 if (num_online_cpus() == 1)
  return -EBUSY;

 if (!cpu_online(cpu))
  return -EINVAL;

 cpu_hotplug_begin();
 err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
     hcpu, -1, &nr_calls);
 if (err == NOTIFY_BAD) {
  nr_calls--;
  __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
       hcpu, nr_calls, NULL);
  printk("%s: attempt to take down CPU %u failed/n",
    __func__, cpu);
  err = -EINVAL;
  goto out_release;
 }

 /* Ensure that we are not runnable on dying cpu */
 old_allowed = current->cpus_allowed;
 cpus_setall(tmp);
 cpu_clear(cpu, tmp);
 set_cpus_allowed_ptr(current, &tmp);
 tmp = cpumask_of_cpu(cpu);

 err = __stop_machine(take_cpu_down, &tcd_param, &tmp);
 if (err) {
  /* CPU didn't die: tell everyone.  Can't complain. */
  if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
         hcpu) == NOTIFY_BAD)
   BUG();

  goto out_allowed;
 }
 BUG_ON(cpu_online(cpu));

 /* Wait for it to sleep (leaving idle task). */
 while (!idle_cpu(cpu))
  yield();

 /* This actually kills the CPU. */
 __cpu_die(cpu);

 /* CPU is completely dead: tell everyone.  Too late to complain. */
 if (raw_notifier_call_chain(&cpu_chain, CPU_DEAD | mod,
        hcpu) == NOTIFY_BAD)
  BUG();

 check_for_tasks(cpu);

out_allowed:
 set_cpus_allowed_ptr(current, &old_allowed);
out_release:
 cpu_hotplug_done();
 if (!err) {
  if (raw_notifier_call_chain(&cpu_chain, CPU_POST_DEAD | mod,
         hcpu) == NOTIFY_BAD)
   BUG();
 }
 return err;
}

int __ref cpu_down(unsigned int cpu)
{
 int err = 0;

 cpu_maps_update_begin();

 if (cpu_hotplug_disabled) {
  err = -EBUSY;
  goto out;
 }

 cpu_clear(cpu, cpu_active_map);

 /*
  * Make sure the all cpus did the reschedule and are not
  * using stale version of the cpu_active_map.
  * This is not strictly necessary becuase stop_machine()
  * that we run down the line already provides the required
  * synchronization. But it's really a side effect and we do not
  * want to depend on the innards of the stop_machine here.
  */
 synchronize_sched();

 err = _cpu_down(cpu, 0);

 if (cpu_online(cpu))
  cpu_set(cpu, cpu_active_map);

out:
 cpu_maps_update_done();
 return err;
}
EXPORT_SYMBOL(cpu_down);
#endif /*CONFIG_HOTPLUG_CPU*/

/* Requires cpu_add_remove_lock to be held */
static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
{
 int ret, nr_calls = 0;
 void *hcpu = (void *)(long)cpu;
 unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;

 if (cpu_online(cpu) || !cpu_present(cpu))
  return -EINVAL;

 cpu_hotplug_begin();
 ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
       -1, &nr_calls);
 if (ret == NOTIFY_BAD) {
  nr_calls--;
  printk("%s: attempt to bring up CPU %u failed/n",
    __func__, cpu);
  ret = -EINVAL;
  goto out_notify;
 }

 /* Arch-specific enabling code. */
 ret = __cpu_up(cpu);
 if (ret != 0)
  goto out_notify;
 BUG_ON(!cpu_online(cpu));

 cpu_set(cpu, cpu_active_map);

 /* Now call notifier in preparation. */
 raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);

out_notify:
 if (ret != 0)
  __raw_notifier_call_chain(&cpu_chain,
    CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
 cpu_hotplug_done();

 return ret;
}

int __cpuinit cpu_up(unsigned int cpu)
{
 int err = 0;
 if (!cpu_isset(cpu, cpu_possible_map)) {
  printk(KERN_ERR "can't online cpu %d because it is not "
   "configured as may-hotadd at boot time/n", cpu);
#if defined(CONFIG_IA64) || defined(CONFIG_X86_64)
  printk(KERN_ERR "please check additional_cpus= boot "
    "parameter/n");
#endif
  return -EINVAL;
 }

 cpu_maps_update_begin();

 if (cpu_hotplug_disabled) {
  err = -EBUSY;
  goto out;
 }

 err = _cpu_up(cpu, 0);

out:
 cpu_maps_update_done();
 return err;
}

#ifdef CONFIG_PM_SLEEP_SMP
static cpumask_t frozen_cpus;

int disable_nonboot_cpus(void)
{
 int cpu, first_cpu, error = 0;

 cpu_maps_update_begin();
 first_cpu = first_cpu(cpu_online_map);
 /* We take down all of the non-boot CPUs in one shot to avoid races
  * with the userspace trying to use the CPU hotplug at the same time
  */
 cpus_clear(frozen_cpus);
 printk("Disabling non-boot CPUs .../n");
 for_each_online_cpu(cpu) {
  if (cpu == first_cpu)
   continue;
  error = _cpu_down(cpu, 1);
  if (!error) {
   cpu_set(cpu, frozen_cpus);
   printk("CPU%d is down/n", cpu);
  } else {
   printk(KERN_ERR "Error taking CPU%d down: %d/n",
    cpu, error);
   break;
  }
 }
 if (!error) {
  BUG_ON(num_online_cpus() > 1);
  /* Make sure the CPUs won't be enabled by someone else */
  cpu_hotplug_disabled = 1;
 } else {
  printk(KERN_ERR "Non-boot CPUs are not disabled/n");
 }
 cpu_maps_update_done();
 return error;
}

void __ref enable_nonboot_cpus(void)
{
 int cpu, error;

 /* Allow everyone to use the CPU hotplug again */
 cpu_maps_update_begin();
 cpu_hotplug_disabled = 0;
 if (cpus_empty(frozen_cpus))
  goto out;

 printk("Enabling non-boot CPUs .../n");
 for_each_cpu_mask_nr(cpu, frozen_cpus) {
  error = _cpu_up(cpu, 1);
  if (!error) {
   printk("CPU%d is up/n", cpu);
   continue;
  }
  printk(KERN_WARNING "Error taking CPU%d up: %d/n", cpu, error);
 }
 cpus_clear(frozen_cpus);
out:
 cpu_maps_update_done();
}
#endif /* CONFIG_PM_SLEEP_SMP */

/**
 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
 * @cpu: cpu that just started
 *
 * This function calls the cpu_chain notifiers with CPU_STARTING.
 * It must be called by the arch code on the new cpu, before the new cpu
 * enables interrupts and before the "boot" cpu returns from __cpu_up().
 */
void __cpuinit notify_cpu_starting(unsigned int cpu)
{
 unsigned long val = CPU_STARTING;

#ifdef CONFIG_PM_SLEEP_SMP
 if (cpu_isset(cpu, frozen_cpus))
  val = CPU_STARTING_FROZEN;
#endif /* CONFIG_PM_SLEEP_SMP */
 raw_notifier_call_chain(&cpu_chain, val, (void *)(long)cpu);
}

#endif /* CONFIG_SMP */

/*
 * cpu_bit_bitmap[] is a special, "compressed" data structure that
 * represents all NR_CPUS bits binary values of 1<<nr.
 *
 * It is used by cpumask_of_cpu() to get a constant address to a CPU
 * mask value that has a single bit set only.
 */

/* cpu_bit_bitmap[0] is empty - so we can back into it */
#define MASK_DECLARE_1(x) [x+1][0] = 1UL << (x)
#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)

const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {

 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
#if BITS_PER_LONG > 32
 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
#endif
};
EXPORT_SYMBOL_GPL(cpu_bit_bitmap);

const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
EXPORT_SYMBOL(cpu_all_bits);