sysctl extern cleanup: module
[linux-2.6.git] / kernel / cpu.c
index f8f9468..f8cced2 100644 (file)
@@ -15,7 +15,8 @@
 #include <linux/stop_machine.h>
 #include <linux/mutex.h>
 
-/* Serializes the updates to cpu_online_map, cpu_present_map */
+#ifdef CONFIG_SMP
+/* Serializes the updates to cpu_online_mask, cpu_present_mask */
 static DEFINE_MUTEX(cpu_add_remove_lock);
 
 static __cpuinitdata RAW_NOTIFIER_HEAD(cpu_chain);
@@ -33,18 +34,11 @@ static struct {
         * an ongoing cpu hotplug operation.
         */
        int refcount;
-       wait_queue_head_t writer_queue;
-} cpu_hotplug;
-
-#define writer_exists() (cpu_hotplug.active_writer != NULL)
-
-void __init cpu_hotplug_init(void)
-{
-       cpu_hotplug.active_writer = NULL;
-       mutex_init(&cpu_hotplug.lock);
-       cpu_hotplug.refcount = 0;
-       init_waitqueue_head(&cpu_hotplug.writer_queue);
-}
+} cpu_hotplug = {
+       .active_writer = NULL,
+       .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
+       .refcount = 0,
+};
 
 #ifdef CONFIG_HOTPLUG_CPU
 
@@ -65,11 +59,8 @@ void put_online_cpus(void)
        if (cpu_hotplug.active_writer == current)
                return;
        mutex_lock(&cpu_hotplug.lock);
-       cpu_hotplug.refcount--;
-
-       if (unlikely(writer_exists()) && !cpu_hotplug.refcount)
-               wake_up(&cpu_hotplug.writer_queue);
-
+       if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
+               wake_up_process(cpu_hotplug.active_writer);
        mutex_unlock(&cpu_hotplug.lock);
 
 }
@@ -79,7 +70,7 @@ EXPORT_SYMBOL_GPL(put_online_cpus);
 
 /*
  * The following two API's must be used when attempting
- * to serialize the updates to cpu_online_map, cpu_present_map.
+ * to serialize the updates to cpu_online_mask, cpu_present_mask.
  */
 void cpu_maps_update_begin(void)
 {
@@ -98,8 +89,8 @@ void cpu_maps_update_done(void)
  * Note that during a cpu-hotplug operation, the new readers, if any,
  * will be blocked by the cpu_hotplug.lock
  *
- * Since cpu_maps_update_begin is always called after invoking
- * cpu_maps_update_begin, we can be sure that only one writer is active.
+ * 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
@@ -115,19 +106,16 @@ void cpu_maps_update_done(void)
  */
 static void cpu_hotplug_begin(void)
 {
-       DECLARE_WAITQUEUE(wait, current);
-
-       mutex_lock(&cpu_hotplug.lock);
-
        cpu_hotplug.active_writer = current;
-       add_wait_queue_exclusive(&cpu_hotplug.writer_queue, &wait);
-       while (cpu_hotplug.refcount) {
-               set_current_state(TASK_UNINTERRUPTIBLE);
+
+       for (;;) {
+               mutex_lock(&cpu_hotplug.lock);
+               if (likely(!cpu_hotplug.refcount))
+                       break;
+               __set_current_state(TASK_UNINTERRUPTIBLE);
                mutex_unlock(&cpu_hotplug.lock);
                schedule();
-               mutex_lock(&cpu_hotplug.lock);
        }
-       remove_wait_queue_locked(&cpu_hotplug.writer_queue, &wait);
 }
 
 static void cpu_hotplug_done(void)
@@ -163,13 +151,13 @@ static inline void check_for_tasks(int cpu)
 
        write_lock_irq(&tasklist_lock);
        for_each_process(p) {
-               if (task_cpu(p) == cpu &&
+               if (task_cpu(p) == cpu && p->state == TASK_RUNNING &&
                    (!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);
+                       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);
 }
@@ -185,13 +173,14 @@ static int __ref take_cpu_down(void *_param)
        struct take_cpu_down_param *param = _param;
        int err;
 
-       raw_notifier_call_chain(&cpu_chain, CPU_DYING | param->mod,
-                               param->hcpu);
        /* 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();
@@ -202,8 +191,7 @@ static int __ref take_cpu_down(void *_param)
 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 {
        int err, nr_calls = 0;
-       struct task_struct *p;
-       cpumask_t old_allowed, tmp;
+       cpumask_var_t old_allowed;
        void *hcpu = (void *)(long)cpu;
        unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
        struct take_cpu_down_param tcd_param = {
@@ -217,39 +205,40 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
        if (!cpu_online(cpu))
                return -EINVAL;
 
+       if (!alloc_cpumask_var(&old_allowed, GFP_KERNEL))
+               return -ENOMEM;
+
        cpu_hotplug_begin();
+       set_cpu_active(cpu, false);
        err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
                                        hcpu, -1, &nr_calls);
        if (err == NOTIFY_BAD) {
+               set_cpu_active(cpu, true);
+
                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",
-                               __FUNCTION__, cpu);
+                               __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);
+       cpumask_copy(old_allowed, &current->cpus_allowed);
+       set_cpus_allowed_ptr(current, cpu_active_mask);
 
-       p = __stop_machine_run(take_cpu_down, &tcd_param, cpu);
-
-       if (IS_ERR(p) || cpu_online(cpu)) {
+       err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
+       if (err) {
+               set_cpu_active(cpu, true);
                /* CPU didn't die: tell everyone.  Can't complain. */
                if (raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
                                            hcpu) == NOTIFY_BAD)
                        BUG();
 
-               if (IS_ERR(p)) {
-                       err = PTR_ERR(p);
-                       goto out_allowed;
-               }
-               goto out_thread;
+               goto out_allowed;
        }
+       BUG_ON(cpu_online(cpu));
 
        /* Wait for it to sleep (leaving idle task). */
        while (!idle_cpu(cpu))
@@ -265,28 +254,41 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 
        check_for_tasks(cpu);
 
-out_thread:
-       err = kthread_stop(p);
 out_allowed:
-       set_cpus_allowed_ptr(current, &old_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();
+       }
+       free_cpumask_var(old_allowed);
        return err;
 }
 
 int __ref cpu_down(unsigned int cpu)
 {
-       int err = 0;
+       int err;
 
+       err = stop_machine_create();
+       if (err)
+               return err;
        cpu_maps_update_begin();
-       if (cpu_hotplug_disabled)
+
+       if (cpu_hotplug_disabled) {
                err = -EBUSY;
-       else
-               err = _cpu_down(cpu, 0);
+               goto out;
+       }
 
+       err = _cpu_down(cpu, 0);
+
+out:
        cpu_maps_update_done();
+       stop_machine_destroy();
        return err;
 }
+EXPORT_SYMBOL(cpu_down);
 #endif /*CONFIG_HOTPLUG_CPU*/
 
 /* Requires cpu_add_remove_lock to be held */
@@ -305,7 +307,7 @@ static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
        if (ret == NOTIFY_BAD) {
                nr_calls--;
                printk("%s: attempt to bring up CPU %u failed\n",
-                               __FUNCTION__, cpu);
+                               __func__, cpu);
                ret = -EINVAL;
                goto out_notify;
        }
@@ -316,6 +318,8 @@ static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
                goto out_notify;
        BUG_ON(!cpu_online(cpu));
 
+       set_cpu_active(cpu, true);
+
        /* Now call notifier in preparation. */
        raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);
 
@@ -331,10 +335,10 @@ out_notify:
 int __cpuinit cpu_up(unsigned int cpu)
 {
        int err = 0;
-       if (!cpu_isset(cpu, cpu_possible_map)) {
+       if (!cpu_possible(cpu)) {
                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) || defined(CONFIG_S390)
+#if defined(CONFIG_IA64)
                printk(KERN_ERR "please check additional_cpus= boot "
                                "parameter\n");
 #endif
@@ -342,42 +346,51 @@ int __cpuinit cpu_up(unsigned int cpu)
        }
 
        cpu_maps_update_begin();
-       if (cpu_hotplug_disabled)
+
+       if (cpu_hotplug_disabled) {
                err = -EBUSY;
-       else
-               err = _cpu_up(cpu, 0);
+               goto out;
+       }
 
+       err = _cpu_up(cpu, 0);
+
+out:
        cpu_maps_update_done();
        return err;
 }
 
 #ifdef CONFIG_PM_SLEEP_SMP
-static cpumask_t frozen_cpus;
+static cpumask_var_t frozen_cpus;
 
 int disable_nonboot_cpus(void)
 {
-       int cpu, first_cpu, error = 0;
+       int cpu, first_cpu, error;
 
+       error = stop_machine_create();
+       if (error)
+               return error;
        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
+       first_cpu = cpumask_first(cpu_online_mask);
+       /*
+        * 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);
+       cpumask_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 {
+               if (!error)
+                       cpumask_set_cpu(cpu, frozen_cpus);
+               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 */
@@ -386,9 +399,18 @@ int disable_nonboot_cpus(void)
                printk(KERN_ERR "Non-boot CPUs are not disabled\n");
        }
        cpu_maps_update_done();
+       stop_machine_destroy();
        return error;
 }
 
+void __weak arch_enable_nonboot_cpus_begin(void)
+{
+}
+
+void __weak arch_enable_nonboot_cpus_end(void)
+{
+}
+
 void __ref enable_nonboot_cpus(void)
 {
        int cpu, error;
@@ -396,11 +418,14 @@ void __ref enable_nonboot_cpus(void)
        /* Allow everyone to use the CPU hotplug again */
        cpu_maps_update_begin();
        cpu_hotplug_disabled = 0;
-       if (cpus_empty(frozen_cpus))
+       if (cpumask_empty(frozen_cpus))
                goto out;
 
        printk("Enabling non-boot CPUs ...\n");
-       for_each_cpu_mask(cpu, frozen_cpus) {
+
+       arch_enable_nonboot_cpus_begin();
+
+       for_each_cpu(cpu, frozen_cpus) {
                error = _cpu_up(cpu, 1);
                if (!error) {
                        printk("CPU%d is up\n", cpu);
@@ -408,8 +433,136 @@ void __ref enable_nonboot_cpus(void)
                }
                printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
        }
-       cpus_clear(frozen_cpus);
+
+       arch_enable_nonboot_cpus_end();
+
+       cpumask_clear(frozen_cpus);
 out:
        cpu_maps_update_done();
 }
+
+static int alloc_frozen_cpus(void)
+{
+       if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
+               return -ENOMEM;
+       return 0;
+}
+core_initcall(alloc_frozen_cpus);
+#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 (frozen_cpus != NULL && cpumask_test_cpu(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() 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);
+
+#ifdef CONFIG_INIT_ALL_POSSIBLE
+static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
+       = CPU_BITS_ALL;
+#else
+static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
+#endif
+const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
+EXPORT_SYMBOL(cpu_possible_mask);
+
+static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
+const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
+EXPORT_SYMBOL(cpu_online_mask);
+
+static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
+const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
+EXPORT_SYMBOL(cpu_present_mask);
+
+static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
+const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
+EXPORT_SYMBOL(cpu_active_mask);
+
+void set_cpu_possible(unsigned int cpu, bool possible)
+{
+       if (possible)
+               cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
+       else
+               cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
+}
+
+void set_cpu_present(unsigned int cpu, bool present)
+{
+       if (present)
+               cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
+       else
+               cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
+}
+
+void set_cpu_online(unsigned int cpu, bool online)
+{
+       if (online)
+               cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
+       else
+               cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
+}
+
+void set_cpu_active(unsigned int cpu, bool active)
+{
+       if (active)
+               cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
+       else
+               cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
+}
+
+void init_cpu_present(const struct cpumask *src)
+{
+       cpumask_copy(to_cpumask(cpu_present_bits), src);
+}
+
+void init_cpu_possible(const struct cpumask *src)
+{
+       cpumask_copy(to_cpumask(cpu_possible_bits), src);
+}
+
+void init_cpu_online(const struct cpumask *src)
+{
+       cpumask_copy(to_cpumask(cpu_online_bits), src);
+}