linux 2.6.16.38 w/ vs2.0.3-rc1
[linux-2.6.git] / drivers / cpufreq / cpufreq.c
index 29b2fa5..9582de1 100644 (file)
@@ -5,9 +5,7 @@
  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
  *
  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
- *     Added handling for CPU hotplug
- *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
- *     Fix handling for CPU hotplug -- affected CPUs
+ *                     Added handling for CPU hotplug
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 as
@@ -46,14 +44,15 @@ static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
 static void handle_update(void *data);
 
 /**
- * Two notifier lists: the "policy" list is involved in the
- * validation process for a new CPU frequency policy; the
+ * Two notifier lists: the "policy" list is involved in the 
+ * validation process for a new CPU frequency policy; the 
  * "transition" list for kernel code that needs to handle
  * changes to devices when the CPU clock speed changes.
  * The mutex locks both lists.
  */
-static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
-static BLOCKING_NOTIFIER_HEAD(cpufreq_transition_notifier_list);
+static struct notifier_block *cpufreq_policy_notifier_list;
+static struct notifier_block *cpufreq_transition_notifier_list;
+static DECLARE_RWSEM (cpufreq_notifier_rwsem);
 
 
 static LIST_HEAD(cpufreq_governor_list);
@@ -152,7 +151,7 @@ void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt
        va_list args;
        unsigned int len;
        unsigned long flags;
-
+       
        WARN_ON(!prefix);
        if (type & debug) {
                spin_lock_irqsave(&disable_ratelimit_lock, flags);
@@ -199,7 +198,7 @@ static inline void cpufreq_debug_disable_ratelimit(void) { return; }
  *
  * This function alters the system "loops_per_jiffy" for the clock
  * speed change. Note that loops_per_jiffy cannot be updated on SMP
- * systems as each CPU might be scaled differently. So, use the arch
+ * systems as each CPU might be scaled differently. So, use the arch 
  * per-CPU loops_per_jiffy value wherever possible.
  */
 #ifndef CONFIG_SMP
@@ -234,7 +233,7 @@ static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) {
  *
  * This function calls the transition notifiers and the "adjust_jiffies"
  * function. It is called twice on all CPU frequency changes that have
- * external effects.
+ * external effects. 
  */
 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 {
@@ -246,11 +245,13 @@ void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
        dprintk("notification %u of frequency transition to %u kHz\n",
                state, freqs->new);
 
+       down_read(&cpufreq_notifier_rwsem);
+
        policy = cpufreq_cpu_data[freqs->cpu];
        switch (state) {
 
        case CPUFREQ_PRECHANGE:
-               /* detect if the driver reported a value as "old frequency"
+               /* detect if the driver reported a value as "old frequency" 
                 * which is not equal to what the cpufreq core thinks is
                 * "old frequency".
                 */
@@ -263,19 +264,20 @@ void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
                                freqs->old = policy->cur;
                        }
                }
-               blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
-                               CPUFREQ_PRECHANGE, freqs);
+               notifier_call_chain(&cpufreq_transition_notifier_list,
+                                       CPUFREQ_PRECHANGE, freqs);
                adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
                break;
 
        case CPUFREQ_POSTCHANGE:
                adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
-               blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
-                               CPUFREQ_POSTCHANGE, freqs);
+               notifier_call_chain(&cpufreq_transition_notifier_list,
+                                       CPUFREQ_POSTCHANGE, freqs);
                if (likely(policy) && likely(policy->cpu == freqs->cpu))
                        policy->cur = freqs->new;
                break;
        }
+       up_read(&cpufreq_notifier_rwsem);
 }
 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 
@@ -319,6 +321,7 @@ out:
        }
        return -EINVAL;
 }
+EXPORT_SYMBOL_GPL(cpufreq_parse_governor);
 
 
 /* drivers/base/cpu.c */
@@ -332,11 +335,11 @@ extern struct sysdev_class cpu_sysdev_class;
  * "unsigned int".
  */
 
-#define show_one(file_name, object)                    \
-static ssize_t show_##file_name                                \
-(struct cpufreq_policy * policy, char *buf)            \
-{                                                      \
-       return sprintf (buf, "%u\n", policy->object);   \
+#define show_one(file_name, object)                                    \
+static ssize_t show_##file_name                                        \
+(struct cpufreq_policy * policy, char *buf)                            \
+{                                                                      \
+       return sprintf (buf, "%u\n", policy->object);                   \
 }
 
 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
@@ -345,8 +348,6 @@ show_one(scaling_min_freq, min);
 show_one(scaling_max_freq, max);
 show_one(scaling_cur_freq, cur);
 
-static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
-
 /**
  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
  */
@@ -365,10 +366,7 @@ static ssize_t store_##file_name                                   \
        if (ret != 1)                                                   \
                return -EINVAL;                                         \
                                                                        \
-       mutex_lock(&policy->lock);                                      \
-       ret = __cpufreq_set_policy(policy, &new_policy);                \
-       policy->user_policy.object = policy->object;                    \
-       mutex_unlock(&policy->lock);                                    \
+       ret = cpufreq_set_policy(&new_policy);                          \
                                                                        \
        return ret ? ret : count;                                       \
 }
@@ -406,8 +404,8 @@ static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 /**
  * store_scaling_governor - store policy for the specified CPU
  */
-static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
-                                      const char *buf, size_t count)
+static ssize_t store_scaling_governor (struct cpufreq_policy * policy, 
+                                      const char *buf, size_t count) 
 {
        unsigned int ret = -EINVAL;
        char    str_governor[16];
@@ -424,15 +422,7 @@ static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
        if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
                return -EINVAL;
 
-       /* Do not use cpufreq_set_policy here or the user_policy.max
-          will be wrongly overridden */
-       mutex_lock(&policy->lock);
-       ret = __cpufreq_set_policy(policy, &new_policy);
-
-       policy->user_policy.policy = policy->policy;
-       policy->user_policy.governor = policy->governor;
-       mutex_unlock(&policy->lock);
-
+       ret = cpufreq_set_policy(&new_policy);
        return ret ? ret : count;
 }
 
@@ -538,7 +528,7 @@ static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
        return ret;
 }
 
-static ssize_t store(struct kobject * kobj, struct attribute * attr,
+static ssize_t store(struct kobject * kobj, struct attribute * attr, 
                     const char * buf, size_t count)
 {
        struct cpufreq_policy * policy = to_policy(kobj);
@@ -574,7 +564,7 @@ static struct kobj_type ktype_cpufreq = {
 /**
  * cpufreq_add_dev - add a CPU device
  *
- * Adds the cpufreq interface for a CPU device.
+ * Adds the cpufreq interface for a CPU device. 
  */
 static int cpufreq_add_dev (struct sys_device * sys_dev)
 {
@@ -583,12 +573,8 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        struct cpufreq_policy new_policy;
        struct cpufreq_policy *policy;
        struct freq_attr **drv_attr;
-       struct sys_device *cpu_sys_dev;
        unsigned long flags;
        unsigned int j;
-#ifdef CONFIG_SMP
-       struct cpufreq_policy *managed_policy;
-#endif
 
        if (cpu_is_offline(cpu))
                return 0;
@@ -601,7 +587,8 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
         * CPU because it is in the same boat. */
        policy = cpufreq_cpu_get(cpu);
        if (unlikely(policy)) {
-               cpufreq_cpu_put(policy);
+               dprintk("CPU already managed, adding link\n");
+               sysfs_create_link(&sys_dev->kobj, &policy->kobj, "cpufreq");
                cpufreq_debug_enable_ratelimit();
                return 0;
        }
@@ -636,32 +623,6 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
                goto err_out;
        }
 
-#ifdef CONFIG_SMP
-       for_each_cpu_mask(j, policy->cpus) {
-               if (cpu == j)
-                       continue;
-
-               /* check for existing affected CPUs.  They may not be aware
-                * of it due to CPU Hotplug.
-                */
-               managed_policy = cpufreq_cpu_get(j);
-               if (unlikely(managed_policy)) {
-                       spin_lock_irqsave(&cpufreq_driver_lock, flags);
-                       managed_policy->cpus = policy->cpus;
-                       cpufreq_cpu_data[cpu] = managed_policy;
-                       spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
-
-                       dprintk("CPU already managed, adding link\n");
-                       sysfs_create_link(&sys_dev->kobj,
-                                         &managed_policy->kobj, "cpufreq");
-
-                       cpufreq_debug_enable_ratelimit();
-                       mutex_unlock(&policy->lock);
-                       ret = 0;
-                       goto err_out_driver_exit; /* call driver->exit() */
-               }
-       }
-#endif
        memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 
        /* prepare interface data */
@@ -689,26 +650,12 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        for_each_cpu_mask(j, policy->cpus)
                cpufreq_cpu_data[j] = policy;
        spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
-
-       /* symlink affected CPUs */
-       for_each_cpu_mask(j, policy->cpus) {
-               if (j == cpu)
-                       continue;
-               if (!cpu_online(j))
-                       continue;
-
-               dprintk("CPU %u already managed, adding link\n", j);
-               cpufreq_cpu_get(cpu);
-               cpu_sys_dev = get_cpu_sysdev(j);
-               sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
-                                 "cpufreq");
-       }
-
        policy->governor = NULL; /* to assure that the starting sequence is
                                  * run in cpufreq_set_policy */
        mutex_unlock(&policy->lock);
-
+       
        /* set default policy */
+       
        ret = cpufreq_set_policy(&new_policy);
        if (ret) {
                dprintk("setting policy failed\n");
@@ -718,7 +665,7 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        module_put(cpufreq_driver->owner);
        dprintk("initialization complete\n");
        cpufreq_debug_enable_ratelimit();
-
+       
        return 0;
 
 
@@ -777,11 +724,10 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
 
 #ifdef CONFIG_SMP
        /* if this isn't the CPU which is the parent of the kobj, we
-        * only need to unlink, put and exit
+        * only need to unlink, put and exit 
         */
        if (unlikely(cpu != data->cpu)) {
                dprintk("removing link\n");
-               cpu_clear(cpu, data->cpus);
                spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
                sysfs_remove_link(&sys_dev->kobj, "cpufreq");
                cpufreq_cpu_put(data);
@@ -794,7 +740,7 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
        if (!kobject_get(&data->kobj)) {
                spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
                cpufreq_debug_enable_ratelimit();
-               return -EFAULT;
+               return -EFAULT;
        }
 
 #ifdef CONFIG_SMP
@@ -837,7 +783,7 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
        kobject_put(&data->kobj);
 
        /* we need to make sure that the underlying kobj is actually
-        * not referenced anymore by anybody before we proceed with
+        * not referenced anymore by anybody before we proceed with 
         * unloading.
         */
        dprintk("waiting for dropping of refcount\n");
@@ -885,7 +831,7 @@ static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigne
 }
 
 
-/**
+/** 
  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
  * @cpu: CPU number
  *
@@ -909,7 +855,7 @@ unsigned int cpufreq_quick_get(unsigned int cpu)
 EXPORT_SYMBOL(cpufreq_quick_get);
 
 
-/**
+/** 
  * cpufreq_get - get the current CPU frequency (in kHz)
  * @cpu: CPU number
  *
@@ -1014,7 +960,7 @@ static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
                freqs.old = cpu_policy->cur;
                freqs.new = cur_freq;
 
-               blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
+               notifier_call_chain(&cpufreq_transition_notifier_list,
                                    CPUFREQ_SUSPENDCHANGE, &freqs);
                adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
 
@@ -1095,8 +1041,7 @@ static int cpufreq_resume(struct sys_device * sysdev)
                        freqs.old = cpu_policy->cur;
                        freqs.new = cur_freq;
 
-                       blocking_notifier_call_chain(
-                                       &cpufreq_transition_notifier_list,
+                       notifier_call_chain(&cpufreq_transition_notifier_list,
                                        CPUFREQ_RESUMECHANGE, &freqs);
                        adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
 
@@ -1127,30 +1072,30 @@ static struct sysdev_driver cpufreq_sysdev_driver = {
  *     @nb: notifier function to register
  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
  *
- *     Add a driver to one of two lists: either a list of drivers that
+ *     Add a driver to one of two lists: either a list of drivers that 
  *      are notified about clock rate changes (once before and once after
  *      the transition), or a list of drivers that are notified about
  *      changes in cpufreq policy.
  *
  *     This function may sleep, and has the same return conditions as
- *     blocking_notifier_chain_register.
+ *     notifier_chain_register.
  */
 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
 {
        int ret;
 
+       down_write(&cpufreq_notifier_rwsem);
        switch (list) {
        case CPUFREQ_TRANSITION_NOTIFIER:
-               ret = blocking_notifier_chain_register(
-                               &cpufreq_transition_notifier_list, nb);
+               ret = notifier_chain_register(&cpufreq_transition_notifier_list, nb);
                break;
        case CPUFREQ_POLICY_NOTIFIER:
-               ret = blocking_notifier_chain_register(
-                               &cpufreq_policy_notifier_list, nb);
+               ret = notifier_chain_register(&cpufreq_policy_notifier_list, nb);
                break;
        default:
                ret = -EINVAL;
        }
+       up_write(&cpufreq_notifier_rwsem);
 
        return ret;
 }
@@ -1165,24 +1110,24 @@ EXPORT_SYMBOL(cpufreq_register_notifier);
  *     Remove a driver from the CPU frequency notifier list.
  *
  *     This function may sleep, and has the same return conditions as
- *     blocking_notifier_chain_unregister.
+ *     notifier_chain_unregister.
  */
 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
 {
        int ret;
 
+       down_write(&cpufreq_notifier_rwsem);
        switch (list) {
        case CPUFREQ_TRANSITION_NOTIFIER:
-               ret = blocking_notifier_chain_unregister(
-                               &cpufreq_transition_notifier_list, nb);
+               ret = notifier_chain_unregister(&cpufreq_transition_notifier_list, nb);
                break;
        case CPUFREQ_POLICY_NOTIFIER:
-               ret = blocking_notifier_chain_unregister(
-                               &cpufreq_policy_notifier_list, nb);
+               ret = notifier_chain_unregister(&cpufreq_policy_notifier_list, nb);
                break;
        default:
                ret = -EINVAL;
        }
+       up_write(&cpufreq_notifier_rwsem);
 
        return ret;
 }
@@ -1280,7 +1225,7 @@ int cpufreq_register_governor(struct cpufreq_governor *governor)
                return -EINVAL;
 
        mutex_lock(&cpufreq_governor_mutex);
-
+       
        list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
                if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
                        mutex_unlock(&cpufreq_governor_mutex);
@@ -1289,7 +1234,7 @@ int cpufreq_register_governor(struct cpufreq_governor *governor)
        }
        list_add(&governor->governor_list, &cpufreq_governor_list);
 
-       mutex_unlock(&cpufreq_governor_mutex);
+       mutex_unlock(&cpufreq_governor_mutex);
        return 0;
 }
 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
@@ -1354,23 +1299,29 @@ static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_poli
        if (ret)
                goto error_out;
 
+       down_read(&cpufreq_notifier_rwsem);
+
        /* adjust if necessary - all reasons */
-       blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
-                       CPUFREQ_ADJUST, policy);
+       notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_ADJUST,
+                           policy);
 
        /* adjust if necessary - hardware incompatibility*/
-       blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
-                       CPUFREQ_INCOMPATIBLE, policy);
+       notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_INCOMPATIBLE,
+                           policy);
 
        /* verify the cpu speed can be set within this limit,
           which might be different to the first one */
        ret = cpufreq_driver->verify(policy);
-       if (ret)
+       if (ret) {
+               up_read(&cpufreq_notifier_rwsem);
                goto error_out;
+       }
 
        /* notification of the new policy */
-       blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
-                       CPUFREQ_NOTIFY, policy);
+       notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_NOTIFY,
+                           policy);
+
+       up_read(&cpufreq_notifier_rwsem);
 
        data->min = policy->min;
        data->max = policy->max;
@@ -1497,7 +1448,7 @@ int cpufreq_update_policy(unsigned int cpu)
 }
 EXPORT_SYMBOL(cpufreq_update_policy);
 
-static int cpufreq_cpu_callback(struct notifier_block *nfb,
+static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
                                        unsigned long action, void *hcpu)
 {
        unsigned int cpu = (unsigned long)hcpu;
@@ -1546,9 +1497,9 @@ static struct notifier_block cpufreq_cpu_notifier =
  * @driver_data: A struct cpufreq_driver containing the values#
  * submitted by the CPU Frequency driver.
  *
- *   Registers a CPU Frequency driver to this core code. This code
+ *   Registers a CPU Frequency driver to this core code. This code 
  * returns zero on success, -EBUSY when another driver got here first
- * (and isn't unregistered in the meantime).
+ * (and isn't unregistered in the meantime). 
  *
  */
 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
@@ -1609,7 +1560,7 @@ EXPORT_SYMBOL_GPL(cpufreq_register_driver);
 /**
  * cpufreq_unregister_driver - unregister the current CPUFreq driver
  *
- *    Unregister the current CPUFreq driver. Only call this if you have
+ *    Unregister the current CPUFreq driver. Only call this if you have 
  * the right to do so, i.e. if you have succeeded in initialising before!
  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
  * currently not initialised.