Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git…

…/davej/cpufreq

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/davej/cpufreq:
  [CPUFREQ] Powernow-k8: support family 0xf with 2 low p-states
  [CPUFREQ] fix (utter) cpufreq_add_dev mess
  [CPUFREQ] Cleanup locking in conservative governor
  [CPUFREQ] Cleanup locking in ondemand governor
  [CPUFREQ] Mark policy_rwsem as going static in cpufreq.c wont be exported
  [CPUFREQ] Eliminate the recent lockdep warnings in cpufreq

Documentation/feature-removal-schedule.txt

@@ -458,3 +458,13 @@ Why:	Remove the old legacy 32bit machine check code. This has been
 	but the old version has been kept around for easier testing. Note this
 	doesn't impact the old P5 and WinChip machine check handlers.
 Who:	Andi Kleen <andi@firstfloor.org>
+
+----------------------------
+
+What:	lock_policy_rwsem_* and unlock_policy_rwsem_* will not be
+	exported interface anymore.
+When:	2.6.33
+Why:	cpu_policy_rwsem has a new cleaner definition making it local to
+	cpufreq core and contained inside cpufreq.c. Other dependent
+	drivers should not use it in order to safely avoid lockdep issues.
+Who:	Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>

arch/x86/kernel/cpu/cpufreq/powernow-k8.c

@@ -299,7 +299,7 @@ static int transition_pstate(struct powernow_k8_data *data, u32 pstate)
 static int transition_fid_vid(struct powernow_k8_data *data,
 		u32 reqfid, u32 reqvid)
 {
-	if (core_voltage_pre_transition(data, reqvid))
+	if (core_voltage_pre_transition(data, reqvid, reqfid))
 		return 1;
 
 	if (core_frequency_transition(data, reqfid))
@@ -327,17 +327,20 @@ static int transition_fid_vid(struct powernow_k8_data *data,
 
 /* Phase 1 - core voltage transition ... setup voltage */
 static int core_voltage_pre_transition(struct powernow_k8_data *data,
-		u32 reqvid)
+		u32 reqvid, u32 reqfid)
 {
 	u32 rvosteps = data->rvo;
 	u32 savefid = data->currfid;
-	u32 maxvid, lo;
+	u32 maxvid, lo, rvomult = 1;
 
 	dprintk("ph1 (cpu%d): start, currfid 0x%x, currvid 0x%x, "
 		"reqvid 0x%x, rvo 0x%x\n",
 		smp_processor_id(),
 		data->currfid, data->currvid, reqvid, data->rvo);
 
+	if ((savefid < LO_FID_TABLE_TOP) && (reqfid < LO_FID_TABLE_TOP))
+		rvomult = 2;
+	rvosteps *= rvomult;
 	rdmsr(MSR_FIDVID_STATUS, lo, maxvid);
 	maxvid = 0x1f & (maxvid >> 16);
 	dprintk("ph1 maxvid=0x%x\n", maxvid);
@@ -351,7 +354,8 @@ static int core_voltage_pre_transition(struct powernow_k8_data *data,
 			return 1;
 	}
 
-	while ((rvosteps > 0) && ((data->rvo + data->currvid) > reqvid)) {
+	while ((rvosteps > 0) &&
+			((rvomult * data->rvo + data->currvid) > reqvid)) {
 		if (data->currvid == maxvid) {
 			rvosteps = 0;
 		} else {
@@ -384,13 +388,6 @@ static int core_frequency_transition(struct powernow_k8_data *data, u32 reqfid)
 	u32 vcoreqfid, vcocurrfid, vcofiddiff;
 	u32 fid_interval, savevid = data->currvid;
 
-	if ((reqfid < HI_FID_TABLE_BOTTOM) &&
-	    (data->currfid < HI_FID_TABLE_BOTTOM)) {
-		printk(KERN_ERR PFX "ph2: illegal lo-lo transition "
-				"0x%x 0x%x\n", reqfid, data->currfid);
-		return 1;
-	}
-
 	if (data->currfid == reqfid) {
 		printk(KERN_ERR PFX "ph2 null fid transition 0x%x\n",
 				data->currfid);
@@ -407,6 +404,9 @@ static int core_frequency_transition(struct powernow_k8_data *data, u32 reqfid)
 	vcofiddiff = vcocurrfid > vcoreqfid ? vcocurrfid - vcoreqfid
 	    : vcoreqfid - vcocurrfid;
 
+	if ((reqfid <= LO_FID_TABLE_TOP) && (data->currfid <= LO_FID_TABLE_TOP))
+		vcofiddiff = 0;
+
 	while (vcofiddiff > 2) {
 		(data->currfid & 1) ? (fid_interval = 1) : (fid_interval = 2);
 
@@ -1081,14 +1081,6 @@ static int transition_frequency_fidvid(struct powernow_k8_data *data,
 		return 0;
 	}
 
-	if ((fid < HI_FID_TABLE_BOTTOM) &&
-	    (data->currfid < HI_FID_TABLE_BOTTOM)) {
-		printk(KERN_ERR PFX
-		       "ignoring illegal change in lo freq table-%x to 0x%x\n",
-		       data->currfid, fid);
-		return 1;
-	}
-
 	dprintk("cpu %d, changing to fid 0x%x, vid 0x%x\n",
 		smp_processor_id(), fid, vid);
 	freqs.old = find_khz_freq_from_fid(data->currfid);

arch/x86/kernel/cpu/cpufreq/powernow-k8.h

@@ -215,7 +215,8 @@ struct pst_s {
 
 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "powernow-k8", msg)
 
-static int core_voltage_pre_transition(struct powernow_k8_data *data, u32 reqvid);
+static int core_voltage_pre_transition(struct powernow_k8_data *data,
+	u32 reqvid, u32 regfid);
 static int core_voltage_post_transition(struct powernow_k8_data *data, u32 reqvid);
 static int core_frequency_transition(struct powernow_k8_data *data, u32 reqfid);
 

drivers/cpufreq/cpufreq.c

@@ -761,6 +761,10 @@ static struct kobj_type ktype_cpufreq = {
  * cpufreq_add_dev - add a CPU device
  *
  * Adds the cpufreq interface for a CPU device.
+ *
+ * The Oracle says: try running cpufreq registration/unregistration concurrently
+ * with with cpu hotplugging and all hell will break loose. Tried to clean this
+ * mess up, but more thorough testing is needed. - Mathieu
  */
 static int cpufreq_add_dev(struct sys_device *sys_dev)
 {
@@ -804,23 +808,21 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
 		goto nomem_out;
 	}
 	if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL)) {
-		kfree(policy);
 		ret = -ENOMEM;
-		goto nomem_out;
+		goto err_free_policy;
 	}
 	if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL)) {
-		free_cpumask_var(policy->cpus);
-		kfree(policy);
 		ret = -ENOMEM;
-		goto nomem_out;
+		goto err_free_cpumask;
 	}
 
 	policy->cpu = cpu;
 	cpumask_copy(policy->cpus, cpumask_of(cpu));
 
 	/* Initially set CPU itself as the policy_cpu */
 	per_cpu(policy_cpu, cpu) = cpu;
-	lock_policy_rwsem_write(cpu);
+	ret = (lock_policy_rwsem_write(cpu) < 0);
+	WARN_ON(ret);
 
 	init_completion(&policy->kobj_unregister);
 	INIT_WORK(&policy->update, handle_update);
@@ -833,7 +835,7 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
 	ret = cpufreq_driver->init(policy);
 	if (ret) {
 		dprintk("initialization failed\n");
-		goto err_out;
+		goto err_unlock_policy;
 	}
 	policy->user_policy.min = policy->min;
 	policy->user_policy.max = policy->max;
@@ -858,15 +860,21 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
 		/* Check for existing affected CPUs.
 		 * They may not be aware of it due to CPU Hotplug.
 		 */
-		managed_policy = cpufreq_cpu_get(j);		/* FIXME: Where is this released?  What about error paths? */
+		managed_policy = cpufreq_cpu_get(j);
 		if (unlikely(managed_policy)) {
 
 			/* Set proper policy_cpu */
 			unlock_policy_rwsem_write(cpu);
 			per_cpu(policy_cpu, cpu) = managed_policy->cpu;
 
-			if (lock_policy_rwsem_write(cpu) < 0)
-				goto err_out_driver_exit;
+			if (lock_policy_rwsem_write(cpu) < 0) {
+				/* Should not go through policy unlock path */
+				if (cpufreq_driver->exit)
+					cpufreq_driver->exit(policy);
+				ret = -EBUSY;
+				cpufreq_cpu_put(managed_policy);
+				goto err_free_cpumask;
+			}
 
 			spin_lock_irqsave(&cpufreq_driver_lock, flags);
 			cpumask_copy(managed_policy->cpus, policy->cpus);
@@ -877,12 +885,14 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
 			ret = sysfs_create_link(&sys_dev->kobj,
 						&managed_policy->kobj,
 						"cpufreq");
-			if (ret)
-				goto err_out_driver_exit;
-
-			cpufreq_debug_enable_ratelimit();
-			ret = 0;
-			goto err_out_driver_exit; /* call driver->exit() */
+			if (!ret)
+				cpufreq_cpu_put(managed_policy);
+			/*
+			 * Success. We only needed to be added to the mask.
+			 * Call driver->exit() because only the cpu parent of
+			 * the kobj needed to call init().
+			 */
+			goto out_driver_exit; /* call driver->exit() */
 		}
 	}
 #endif
@@ -892,25 +902,25 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
 	ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq, &sys_dev->kobj,
 				   "cpufreq");
 	if (ret)
-		goto err_out_driver_exit;
+		goto out_driver_exit;
 
 	/* set up files for this cpu device */
 	drv_attr = cpufreq_driver->attr;
 	while ((drv_attr) && (*drv_attr)) {
 		ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 		if (ret)
-			goto err_out_driver_exit;
+			goto err_out_kobj_put;
 		drv_attr++;
 	}
 	if (cpufreq_driver->get) {
 		ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 		if (ret)
-			goto err_out_driver_exit;
+			goto err_out_kobj_put;
 	}
 	if (cpufreq_driver->target) {
 		ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 		if (ret)
-			goto err_out_driver_exit;
+			goto err_out_kobj_put;
 	}
 
 	spin_lock_irqsave(&cpufreq_driver_lock, flags);
@@ -928,12 +938,14 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
 			continue;
 
 		dprintk("CPU %u already managed, adding link\n", j);
-		cpufreq_cpu_get(cpu);
+		managed_policy = cpufreq_cpu_get(cpu);
 		cpu_sys_dev = get_cpu_sysdev(j);
 		ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 					"cpufreq");
-		if (ret)
+		if (ret) {
+			cpufreq_cpu_put(managed_policy);
 			goto err_out_unregister;
+		}
 	}
 
 	policy->governor = NULL; /* to assure that the starting sequence is
@@ -965,17 +977,20 @@ static int cpufreq_add_dev(struct sys_device *sys_dev)
 		per_cpu(cpufreq_cpu_data, j) = NULL;
 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 
+err_out_kobj_put:
 	kobject_put(&policy->kobj);
 	wait_for_completion(&policy->kobj_unregister);
 
-err_out_driver_exit:
+out_driver_exit:
 	if (cpufreq_driver->exit)
 		cpufreq_driver->exit(policy);
 
-err_out:
+err_unlock_policy:
 	unlock_policy_rwsem_write(cpu);
+err_free_cpumask:
+	free_cpumask_var(policy->cpus);
+err_free_policy:
 	kfree(policy);
-
 nomem_out:
 	module_put(cpufreq_driver->owner);
 module_out:
@@ -1070,8 +1085,6 @@ static int __cpufreq_remove_dev(struct sys_device *sys_dev)
 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 #endif
 
-	unlock_policy_rwsem_write(cpu);
-
 	if (cpufreq_driver->target)
 		__cpufreq_governor(data, CPUFREQ_GOV_STOP);
 
@@ -1088,6 +1101,8 @@ static int __cpufreq_remove_dev(struct sys_device *sys_dev)
 	if (cpufreq_driver->exit)
 		cpufreq_driver->exit(data);
 
+	unlock_policy_rwsem_write(cpu);
+
 	free_cpumask_var(data->related_cpus);
 	free_cpumask_var(data->cpus);
 	kfree(data);