PM: Add user-space wake lock api.
[linux-2.6.git] / kernel / cpu.c
index ff071e0..12b7458 100644 (file)
 #include <linux/kthread.h>
 #include <linux/stop_machine.h>
 #include <linux/mutex.h>
-#include <asm/tboot.h>
+#include <linux/gfp.h>
 
 #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);
+/*
+ * The following two API's must be used when attempting
+ * to serialize the updates to cpu_online_mask, cpu_present_mask.
+ */
+void cpu_maps_update_begin(void)
+{
+       mutex_lock(&cpu_add_remove_lock);
+}
+
+void cpu_maps_update_done(void)
+{
+       mutex_unlock(&cpu_add_remove_lock);
+}
+
+static 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;
 
+#ifdef CONFIG_HOTPLUG_CPU
+
 static struct {
        struct task_struct *active_writer;
        struct mutex lock; /* Synchronizes accesses to refcount, */
@@ -41,8 +57,6 @@ static struct {
        .refcount = 0,
 };
 
-#ifdef CONFIG_HOTPLUG_CPU
-
 void get_online_cpus(void)
 {
        might_sleep();
@@ -67,22 +81,6 @@ void put_online_cpus(void)
 }
 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_mask, cpu_present_mask.
- */
-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.
@@ -124,6 +122,12 @@ static void cpu_hotplug_done(void)
        cpu_hotplug.active_writer = NULL;
        mutex_unlock(&cpu_hotplug.lock);
 }
+
+#else /* #if CONFIG_HOTPLUG_CPU */
+static void cpu_hotplug_begin(void) {}
+static void cpu_hotplug_done(void) {}
+#endif /* #else #if CONFIG_HOTPLUG_CPU */
+
 /* Need to know about CPUs going up/down? */
 int __ref register_cpu_notifier(struct notifier_block *nb)
 {
@@ -134,8 +138,28 @@ int __ref register_cpu_notifier(struct notifier_block *nb)
        return ret;
 }
 
+static int __cpu_notify(unsigned long val, void *v, int nr_to_call,
+                       int *nr_calls)
+{
+       int ret;
+
+       ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call,
+                                       nr_calls);
+
+       return notifier_to_errno(ret);
+}
+
+static int cpu_notify(unsigned long val, void *v)
+{
+       return __cpu_notify(val, v, -1, NULL);
+}
+
 #ifdef CONFIG_HOTPLUG_CPU
 
+static void cpu_notify_nofail(unsigned long val, void *v)
+{
+       BUG_ON(cpu_notify(val, v));
+}
 EXPORT_SYMBOL(register_cpu_notifier);
 
 void __ref unregister_cpu_notifier(struct notifier_block *nb)
@@ -152,13 +176,13 @@ 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);
 }
@@ -179,12 +203,7 @@ static int __ref take_cpu_down(void *_param)
        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();
+       cpu_notify(CPU_DYING | param->mod, param->hcpu);
        return 0;
 }
 
@@ -192,7 +211,6 @@ static int __ref take_cpu_down(void *_param)
 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 {
        int err, nr_calls = 0;
-       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 = {
@@ -206,62 +224,48 @@ 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();
-       err = __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_PREPARE | mod,
-                                       hcpu, -1, &nr_calls);
-       if (err == NOTIFY_BAD) {
+
+       err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
+       if (err) {
                nr_calls--;
-               __raw_notifier_call_chain(&cpu_chain, CPU_DOWN_FAILED | mod,
-                                         hcpu, nr_calls, NULL);
+               __cpu_notify(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 */
-       cpumask_copy(old_allowed, &current->cpus_allowed);
-       set_cpus_allowed_ptr(current,
-                            cpumask_of(cpumask_any_but(cpu_online_mask, cpu)));
-
        err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
        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();
+               cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
 
-               goto out_allowed;
+               goto out_release;
        }
        BUG_ON(cpu_online(cpu));
 
-       /* Wait for it to sleep (leaving idle task). */
+       /*
+        * The migration_call() CPU_DYING callback will have removed all
+        * runnable tasks from the cpu, there's only the idle task left now
+        * that the migration thread is done doing the stop_machine thing.
+        *
+        * Wait for the stop thread to go away.
+        */
        while (!idle_cpu(cpu))
-               yield();
+               cpu_relax();
 
        /* 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();
+       cpu_notify_nofail(CPU_DEAD | mod, hcpu);
 
        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();
-       }
-       free_cpumask_var(old_allowed);
+       if (!err)
+               cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
        return err;
 }
 
@@ -269,9 +273,6 @@ int __ref cpu_down(unsigned int cpu)
 {
        int err;
 
-       err = stop_machine_create();
-       if (err)
-               return err;
        cpu_maps_update_begin();
 
        if (cpu_hotplug_disabled) {
@@ -279,26 +280,10 @@ int __ref cpu_down(unsigned int cpu)
                goto out;
        }
 
-       set_cpu_active(cpu, false);
-
-       /*
-        * Make sure the all cpus did the reschedule and are not
-        * using stale version of the cpu_active_mask.
-        * 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))
-               set_cpu_active(cpu, true);
-
 out:
        cpu_maps_update_done();
-       stop_machine_destroy();
        return err;
 }
 EXPORT_SYMBOL(cpu_down);
@@ -315,13 +300,11 @@ static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
                return -EINVAL;
 
        cpu_hotplug_begin();
-       ret = __raw_notifier_call_chain(&cpu_chain, CPU_UP_PREPARE | mod, hcpu,
-                                                       -1, &nr_calls);
-       if (ret == NOTIFY_BAD) {
+       ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
+       if (ret) {
                nr_calls--;
-               printk("%s: attempt to bring up CPU %u failed\n",
+               printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
                                __func__, cpu);
-               ret = -EINVAL;
                goto out_notify;
        }
 
@@ -331,15 +314,12 @@ 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);
+       cpu_notify(CPU_ONLINE | mod, hcpu);
 
 out_notify:
        if (ret != 0)
-               __raw_notifier_call_chain(&cpu_chain,
-                               CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
+               __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
        cpu_hotplug_done();
 
        return ret;
@@ -348,16 +328,44 @@ out_notify:
 int __cpuinit cpu_up(unsigned int cpu)
 {
        int err = 0;
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+       int nid;
+       pg_data_t       *pgdat;
+#endif
+
        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)
+#if defined(CONFIG_IA64)
                printk(KERN_ERR "please check additional_cpus= boot "
                                "parameter\n");
 #endif
                return -EINVAL;
        }
 
+#ifdef CONFIG_MEMORY_HOTPLUG
+       nid = cpu_to_node(cpu);
+       if (!node_online(nid)) {
+               err = mem_online_node(nid);
+               if (err)
+                       return err;
+       }
+
+       pgdat = NODE_DATA(nid);
+       if (!pgdat) {
+               printk(KERN_ERR
+                       "Can't online cpu %d due to NULL pgdat\n", cpu);
+               return -ENOMEM;
+       }
+
+       if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
+               mutex_lock(&zonelists_mutex);
+               build_all_zonelists(NULL);
+               mutex_unlock(&zonelists_mutex);
+       }
+#endif
+
        cpu_maps_update_begin();
 
        if (cpu_hotplug_disabled) {
@@ -375,36 +383,42 @@ out:
 #ifdef CONFIG_PM_SLEEP_SMP
 static cpumask_var_t frozen_cpus;
 
+void __weak arch_disable_nonboot_cpus_begin(void)
+{
+}
+
+void __weak arch_disable_nonboot_cpus_end(void)
+{
+}
+
 int disable_nonboot_cpus(void)
 {
-       int cpu, first_cpu, error, num_cpus = 0;
+       int cpu, first_cpu, error = 0;
 
-       error = stop_machine_create();
-       if (error)
-               return error;
        cpu_maps_update_begin();
        first_cpu = cpumask_first(cpu_online_mask);
-       /* We take down all of the non-boot CPUs in one shot to avoid races
+       /*
+        * 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
         */
        cpumask_clear(frozen_cpus);
+       arch_disable_nonboot_cpus_begin();
+
        printk("Disabling non-boot CPUs ...\n");
        for_each_online_cpu(cpu) {
                if (cpu == first_cpu)
                        continue;
-               num_cpus++;
                error = _cpu_down(cpu, 1);
-               if (!error) {
+               if (!error)
                        cpumask_set_cpu(cpu, frozen_cpus);
-                       printk("CPU%d is down\n", cpu);
-               } else {
+               else {
                        printk(KERN_ERR "Error taking CPU%d down: %d\n",
                                cpu, error);
                        break;
                }
        }
-       /* ensure all CPUs have gone into wait-for-SIPI */
-       error |= tboot_wait_for_aps(num_cpus);
+
+       arch_disable_nonboot_cpus_end();
 
        if (!error) {
                BUG_ON(num_online_cpus() > 1);
@@ -414,10 +428,17 @@ 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;
@@ -428,15 +449,21 @@ void __ref enable_nonboot_cpus(void)
        if (cpumask_empty(frozen_cpus))
                goto out;
 
-       printk("Enabling non-boot CPUs ...\n");
+       printk(KERN_INFO "Enabling non-boot CPUs ...\n");
+
+       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);
+                       printk(KERN_INFO "CPU%d is up\n", cpu);
                        continue;
                }
                printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
        }
+
+       arch_enable_nonboot_cpus_end();
+
        cpumask_clear(frozen_cpus);
 out:
        cpu_maps_update_done();
@@ -467,7 +494,7 @@ void __cpuinit notify_cpu_starting(unsigned int cpu)
        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);
+       cpu_notify(val, (void *)(long)cpu);
 }
 
 #endif /* CONFIG_SMP */
@@ -481,7 +508,7 @@ void __cpuinit notify_cpu_starting(unsigned int cpu)
  */
 
 /* 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_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)