Pull bugzilla-5737 into release branch

Conflicts:

	arch/x86_64/kernel/acpi/processor.c

arch/i386/kernel/acpi/processor.c

@@ -47,7 +47,7 @@ static void init_intel_pdc(struct acpi_processor *pr, struct cpuinfo_x86 *c)
 	buf[2] = ACPI_PDC_C_CAPABILITY_SMP;
 
 	if (cpu_has(c, X86_FEATURE_EST))
-		buf[2] |= ACPI_PDC_EST_CAPABILITY_SMP;
+		buf[2] |= ACPI_PDC_EST_CAPABILITY_SWSMP;
 
 	obj->type = ACPI_TYPE_BUFFER;
 	obj->buffer.length = 12;

arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.c

@@ -48,12 +48,13 @@ MODULE_LICENSE("GPL");
 
 
 struct cpufreq_acpi_io {
-	struct acpi_processor_performance	acpi_data;
+	struct acpi_processor_performance	*acpi_data;
 	struct cpufreq_frequency_table		*freq_table;
 	unsigned int				resume;
 };
 
 static struct cpufreq_acpi_io	*acpi_io_data[NR_CPUS];
+static struct acpi_processor_performance	*acpi_perf_data[NR_CPUS];
 
 static struct cpufreq_driver acpi_cpufreq_driver;
 
@@ -104,64 +105,43 @@ acpi_processor_set_performance (
 {
 	u16			port = 0;
 	u8			bit_width = 0;
+	int			i = 0;
 	int			ret = 0;
 	u32			value = 0;
-	int			i = 0;
-	struct cpufreq_freqs    cpufreq_freqs;
-	cpumask_t		saved_mask;
 	int			retval;
+	struct acpi_processor_performance	*perf;
 
 	dprintk("acpi_processor_set_performance\n");
 
-	/*
-	 * TBD: Use something other than set_cpus_allowed.
-	 * As set_cpus_allowed is a bit racy, 
-	 * with any other set_cpus_allowed for this process.
-	 */
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed(current, cpumask_of_cpu(cpu));
-	if (smp_processor_id() != cpu) {
-		return (-EAGAIN);
-	}
-	
-	if (state == data->acpi_data.state) {
+	retval = 0;
+	perf = data->acpi_data;	
+	if (state == perf->state) {
 		if (unlikely(data->resume)) {
 			dprintk("Called after resume, resetting to P%d\n", state);
 			data->resume = 0;
 		} else {
 			dprintk("Already at target state (P%d)\n", state);
-			retval = 0;
-			goto migrate_end;
+			return (retval);
 		}
 	}
 
-	dprintk("Transitioning from P%d to P%d\n",
-		data->acpi_data.state, state);
-
-	/* cpufreq frequency struct */
-	cpufreq_freqs.cpu = cpu;
-	cpufreq_freqs.old = data->freq_table[data->acpi_data.state].frequency;
-	cpufreq_freqs.new = data->freq_table[state].frequency;
-
-	/* notify cpufreq */
-	cpufreq_notify_transition(&cpufreq_freqs, CPUFREQ_PRECHANGE);
+	dprintk("Transitioning from P%d to P%d\n", perf->state, state);
 
 	/*
 	 * First we write the target state's 'control' value to the
 	 * control_register.
 	 */
 
-	port = data->acpi_data.control_register.address;
-	bit_width = data->acpi_data.control_register.bit_width;
-	value = (u32) data->acpi_data.states[state].control;
+	port = perf->control_register.address;
+	bit_width = perf->control_register.bit_width;
+	value = (u32) perf->states[state].control;
 
 	dprintk("Writing 0x%08x to port 0x%04x\n", value, port);
 
 	ret = acpi_processor_write_port(port, bit_width, value);
 	if (ret) {
 		dprintk("Invalid port width 0x%04x\n", bit_width);
-		retval = ret;
-		goto migrate_end;
+		return (ret);
 	}
 
 	/*
@@ -177,49 +157,36 @@ acpi_processor_set_performance (
 		 * before giving up.
 		 */
 
-		port = data->acpi_data.status_register.address;
-		bit_width = data->acpi_data.status_register.bit_width;
+		port = perf->status_register.address;
+		bit_width = perf->status_register.bit_width;
 
 		dprintk("Looking for 0x%08x from port 0x%04x\n",
-			(u32) data->acpi_data.states[state].status, port);
+			(u32) perf->states[state].status, port);
 
-		for (i=0; i<100; i++) {
+		for (i = 0; i < 100; i++) {
 			ret = acpi_processor_read_port(port, bit_width, &value);
 			if (ret) {	
 				dprintk("Invalid port width 0x%04x\n", bit_width);
-				retval = ret;
-				goto migrate_end;
+				return (ret);
 			}
-			if (value == (u32) data->acpi_data.states[state].status)
+			if (value == (u32) perf->states[state].status)
 				break;
 			udelay(10);
 		}
 	} else {
 		i = 0;
-		value = (u32) data->acpi_data.states[state].status;
+		value = (u32) perf->states[state].status;
 	}
 
-	/* notify cpufreq */
-	cpufreq_notify_transition(&cpufreq_freqs, CPUFREQ_POSTCHANGE);
-
-	if (unlikely(value != (u32) data->acpi_data.states[state].status)) {
-		unsigned int tmp = cpufreq_freqs.new;
-		cpufreq_freqs.new = cpufreq_freqs.old;
-		cpufreq_freqs.old = tmp;
-		cpufreq_notify_transition(&cpufreq_freqs, CPUFREQ_PRECHANGE);
-		cpufreq_notify_transition(&cpufreq_freqs, CPUFREQ_POSTCHANGE);
+	if (unlikely(value != (u32) perf->states[state].status)) {
 		printk(KERN_WARNING "acpi-cpufreq: Transition failed\n");
 		retval = -ENODEV;
-		goto migrate_end;
+		return (retval);
 	}
 
 	dprintk("Transition successful after %d microseconds\n", i * 10);
 
-	data->acpi_data.state = state;
-
-	retval = 0;
-migrate_end:
-	set_cpus_allowed(current, saved_mask);
+	perf->state = state;
 	return (retval);
 }
 
@@ -231,8 +198,17 @@ acpi_cpufreq_target (
 	unsigned int relation)
 {
 	struct cpufreq_acpi_io *data = acpi_io_data[policy->cpu];
+	struct acpi_processor_performance *perf;
+	struct cpufreq_freqs freqs;
+	cpumask_t online_policy_cpus;
+	cpumask_t saved_mask;
+	cpumask_t set_mask;
+	cpumask_t covered_cpus;
+	unsigned int cur_state = 0;
 	unsigned int next_state = 0;
 	unsigned int result = 0;
+	unsigned int j;
+	unsigned int tmp;
 
 	dprintk("acpi_cpufreq_setpolicy\n");
 
@@ -241,11 +217,95 @@ acpi_cpufreq_target (
 			target_freq,
 			relation,
 			&next_state);
-	if (result)
+	if (unlikely(result))
 		return (result);
 
-	result = acpi_processor_set_performance (data, policy->cpu, next_state);
+	perf = data->acpi_data;
+	cur_state = perf->state;
+	freqs.old = data->freq_table[cur_state].frequency;
+	freqs.new = data->freq_table[next_state].frequency;
+
+#ifdef CONFIG_HOTPLUG_CPU
+	/* cpufreq holds the hotplug lock, so we are safe from here on */
+	cpus_and(online_policy_cpus, cpu_online_map, policy->cpus);
+#else
+	online_policy_cpus = policy->cpus;
+#endif
+
+	for_each_cpu_mask(j, online_policy_cpus) {
+		freqs.cpu = j;
+		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+	}
+
+	/*
+	 * We need to call driver->target() on all or any CPU in
+	 * policy->cpus, depending on policy->shared_type.
+	 */
+	saved_mask = current->cpus_allowed;
+	cpus_clear(covered_cpus);
+	for_each_cpu_mask(j, online_policy_cpus) {
+		/*
+		 * Support for SMP systems.
+		 * Make sure we are running on CPU that wants to change freq
+		 */
+		cpus_clear(set_mask);
+		if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
+			cpus_or(set_mask, set_mask, online_policy_cpus);
+		else
+			cpu_set(j, set_mask);
+
+		set_cpus_allowed(current, set_mask);
+		if (unlikely(!cpu_isset(smp_processor_id(), set_mask))) {
+			dprintk("couldn't limit to CPUs in this domain\n");
+			result = -EAGAIN;
+			break;
+		}
+
+		result = acpi_processor_set_performance (data, j, next_state);
+		if (result) {
+			result = -EAGAIN;
+			break;
+		}
+
+		if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
+			break;
  
+		cpu_set(j, covered_cpus);
+	}
+
+	for_each_cpu_mask(j, online_policy_cpus) {
+		freqs.cpu = j;
+		cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+	}
+
+	if (unlikely(result)) {
+		/*
+		 * We have failed halfway through the frequency change.
+		 * We have sent callbacks to online_policy_cpus and
+		 * acpi_processor_set_performance() has been called on 
+		 * coverd_cpus. Best effort undo..
+		 */
+
+		if (!cpus_empty(covered_cpus)) {
+			for_each_cpu_mask(j, covered_cpus) {
+				policy->cpu = j;
+				acpi_processor_set_performance (data, 
+						j, 
+						cur_state);
+			}
+		}
+
+		tmp = freqs.new;
+		freqs.new = freqs.old;
+		freqs.old = tmp;
+		for_each_cpu_mask(j, online_policy_cpus) {
+			freqs.cpu = j;
+			cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+			cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+		}
+	}
+
+	set_cpus_allowed(current, saved_mask);
 	return (result);
 }
 
@@ -271,30 +331,65 @@ acpi_cpufreq_guess_freq (
 	struct cpufreq_acpi_io	*data,
 	unsigned int		cpu)
 {
+	struct acpi_processor_performance	*perf = data->acpi_data;
+
 	if (cpu_khz) {
 		/* search the closest match to cpu_khz */
 		unsigned int i;
 		unsigned long freq;
-		unsigned long freqn = data->acpi_data.states[0].core_frequency * 1000;
+		unsigned long freqn = perf->states[0].core_frequency * 1000;
 
-		for (i=0; i < (data->acpi_data.state_count - 1); i++) {
+		for (i = 0; i < (perf->state_count - 1); i++) {
 			freq = freqn;
-			freqn = data->acpi_data.states[i+1].core_frequency * 1000;
+			freqn = perf->states[i+1].core_frequency * 1000;
 			if ((2 * cpu_khz) > (freqn + freq)) {
-				data->acpi_data.state = i;
+				perf->state = i;
 				return (freq);
 			}
 		}
-		data->acpi_data.state = data->acpi_data.state_count - 1;
+		perf->state = perf->state_count - 1;
 		return (freqn);
-	} else
+	} else {
 		/* assume CPU is at P0... */
-		data->acpi_data.state = 0;
-		return data->acpi_data.states[0].core_frequency * 1000;
-	
+		perf->state = 0;
+		return perf->states[0].core_frequency * 1000;
+	}
 }
 
 
+/*
+ * acpi_cpufreq_early_init - initialize ACPI P-States library
+ *
+ * Initialize the ACPI P-States library (drivers/acpi/processor_perflib.c)
+ * in order to determine correct frequency and voltage pairings. We can
+ * do _PDC and _PSD and find out the processor dependency for the
+ * actual init that will happen later...
+ */
+static int acpi_cpufreq_early_init_acpi(void)
+{
+	struct acpi_processor_performance	*data;
+	unsigned int				i, j;
+
+	dprintk("acpi_cpufreq_early_init\n");
+
+	for_each_cpu(i) {
+		data = kzalloc(sizeof(struct acpi_processor_performance), 
+			GFP_KERNEL);
+		if (!data) {
+			for_each_cpu(j) {
+				kfree(acpi_perf_data[j]);
+				acpi_perf_data[j] = NULL;
+			}
+			return (-ENOMEM);
+		}
+		acpi_perf_data[i] = data;
+	}
+
+	/* Do initialization in ACPI core */
+	acpi_processor_preregister_performance(acpi_perf_data);
+	return 0;
+}
+
 static int
 acpi_cpufreq_cpu_init (
 	struct cpufreq_policy   *policy)
@@ -304,41 +399,51 @@ acpi_cpufreq_cpu_init (
 	struct cpufreq_acpi_io	*data;
 	unsigned int		result = 0;
 	struct cpuinfo_x86 *c = &cpu_data[policy->cpu];
+	struct acpi_processor_performance	*perf;
 
 	dprintk("acpi_cpufreq_cpu_init\n");
 
+	if (!acpi_perf_data[cpu])
+		return (-ENODEV);
+
 	data = kzalloc(sizeof(struct cpufreq_acpi_io), GFP_KERNEL);
 	if (!data)
 		return (-ENOMEM);
 
+	data->acpi_data = acpi_perf_data[cpu];
 	acpi_io_data[cpu] = data;
 
-	result = acpi_processor_register_performance(&data->acpi_data, cpu);
+	result = acpi_processor_register_performance(data->acpi_data, cpu);
 
 	if (result)
 		goto err_free;
 
+	perf = data->acpi_data;
+	policy->cpus = perf->shared_cpu_map;
+	policy->shared_type = perf->shared_type;
+
 	if (cpu_has(c, X86_FEATURE_CONSTANT_TSC)) {
 		acpi_cpufreq_driver.flags |= CPUFREQ_CONST_LOOPS;
 	}
 
 	/* capability check */
-	if (data->acpi_data.state_count <= 1) {
+	if (perf->state_count <= 1) {
 		dprintk("No P-States\n");
 		result = -ENODEV;
 		goto err_unreg;
 	}
-	if ((data->acpi_data.control_register.space_id != ACPI_ADR_SPACE_SYSTEM_IO) ||
-	    (data->acpi_data.status_register.space_id != ACPI_ADR_SPACE_SYSTEM_IO)) {
+
+	if ((perf->control_register.space_id != ACPI_ADR_SPACE_SYSTEM_IO) ||
+	    (perf->status_register.space_id != ACPI_ADR_SPACE_SYSTEM_IO)) {
 		dprintk("Unsupported address space [%d, %d]\n",
-			(u32) (data->acpi_data.control_register.space_id),
-			(u32) (data->acpi_data.status_register.space_id));
+			(u32) (perf->control_register.space_id),
+			(u32) (perf->status_register.space_id));
 		result = -ENODEV;
 		goto err_unreg;
 	}
 
 	/* alloc freq_table */
-	data->freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) * (data->acpi_data.state_count + 1), GFP_KERNEL);
+	data->freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) * (perf->state_count + 1), GFP_KERNEL);
 	if (!data->freq_table) {
 		result = -ENOMEM;
 		goto err_unreg;
@@ -346,9 +451,9 @@ acpi_cpufreq_cpu_init (
 
 	/* detect transition latency */
 	policy->cpuinfo.transition_latency = 0;
-	for (i=0; i<data->acpi_data.state_count; i++) {
-		if ((data->acpi_data.states[i].transition_latency * 1000) > policy->cpuinfo.transition_latency)
-			policy->cpuinfo.transition_latency = data->acpi_data.states[i].transition_latency * 1000;
+	for (i=0; i<perf->state_count; i++) {
+		if ((perf->states[i].transition_latency * 1000) > policy->cpuinfo.transition_latency)
+			policy->cpuinfo.transition_latency = perf->states[i].transition_latency * 1000;
 	}
 	policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
 
@@ -356,11 +461,11 @@ acpi_cpufreq_cpu_init (
 	policy->cur = acpi_cpufreq_guess_freq(data, policy->cpu);
 
 	/* table init */
-	for (i=0; i<=data->acpi_data.state_count; i++)
+	for (i=0; i<=perf->state_count; i++)
 	{
 		data->freq_table[i].index = i;
-		if (i<data->acpi_data.state_count)
-			data->freq_table[i].frequency = data->acpi_data.states[i].core_frequency * 1000;
+		if (i<perf->state_count)
+			data->freq_table[i].frequency = perf->states[i].core_frequency * 1000;
 		else
 			data->freq_table[i].frequency = CPUFREQ_TABLE_END;
 	}
@@ -375,12 +480,12 @@ acpi_cpufreq_cpu_init (
 
 	printk(KERN_INFO "acpi-cpufreq: CPU%u - ACPI performance management activated.\n",
 	       cpu);
-	for (i = 0; i < data->acpi_data.state_count; i++)
+	for (i = 0; i < perf->state_count; i++)
 		dprintk("     %cP%d: %d MHz, %d mW, %d uS\n",
-			(i == data->acpi_data.state?'*':' '), i,
-			(u32) data->acpi_data.states[i].core_frequency,
-			(u32) data->acpi_data.states[i].power,
-			(u32) data->acpi_data.states[i].transition_latency);
+			(i == perf->state?'*':' '), i,
+			(u32) perf->states[i].core_frequency,
+			(u32) perf->states[i].power,
+			(u32) perf->states[i].transition_latency);
 
 	cpufreq_frequency_table_get_attr(data->freq_table, policy->cpu);
 	
@@ -395,7 +500,7 @@ acpi_cpufreq_cpu_init (
  err_freqfree:
 	kfree(data->freq_table);
  err_unreg:
-	acpi_processor_unregister_performance(&data->acpi_data, cpu);
+	acpi_processor_unregister_performance(perf, cpu);
  err_free:
 	kfree(data);
 	acpi_io_data[cpu] = NULL;
@@ -416,7 +521,7 @@ acpi_cpufreq_cpu_exit (
 	if (data) {
 		cpufreq_frequency_table_put_attr(policy->cpu);
 		acpi_io_data[policy->cpu] = NULL;
-		acpi_processor_unregister_performance(&data->acpi_data, policy->cpu);
+		acpi_processor_unregister_performance(data->acpi_data, policy->cpu);
 		kfree(data);
 	}
 
@@ -462,6 +567,9 @@ acpi_cpufreq_init (void)
 
 	dprintk("acpi_cpufreq_init\n");
 
+	result = acpi_cpufreq_early_init_acpi();
+
+	if (!result)
  		result = cpufreq_register_driver(&acpi_cpufreq_driver);
 	
 	return (result);
@@ -471,10 +579,15 @@ acpi_cpufreq_init (void)
 static void __exit
 acpi_cpufreq_exit (void)
 {
+	unsigned int	i;
 	dprintk("acpi_cpufreq_exit\n");
 
 	cpufreq_unregister_driver(&acpi_cpufreq_driver);
 
+	for_each_cpu(i) {
+		kfree(acpi_perf_data[i]);
+		acpi_perf_data[i] = NULL;
+	}
 	return;
 }
 

arch/i386/kernel/cpu/cpufreq/speedstep-centrino.c

@@ -351,7 +351,36 @@ static unsigned int get_cur_freq(unsigned int cpu)
 
 #ifdef CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI
 
-static struct acpi_processor_performance p;
+static struct acpi_processor_performance *acpi_perf_data[NR_CPUS];
+
+/*
+ * centrino_cpu_early_init_acpi - Do the preregistering with ACPI P-States
+ * library
+ *
+ * Before doing the actual init, we need to do _PSD related setup whenever
+ * supported by the BIOS. These are handled by this early_init routine.
+ */
+static int centrino_cpu_early_init_acpi(void)
+{
+	unsigned int	i, j;
+	struct acpi_processor_performance	*data;
+
+	for_each_cpu(i) {
+		data = kzalloc(sizeof(struct acpi_processor_performance), 
+				GFP_KERNEL);
+		if (!data) {
+			for_each_cpu(j) {
+				kfree(acpi_perf_data[j]);
+				acpi_perf_data[j] = NULL;
+			}
+			return (-ENOMEM);
+		}
+		acpi_perf_data[i] = data;
+	}
+
+	acpi_processor_preregister_performance(acpi_perf_data);
+	return 0;
+}
 
 /*
  * centrino_cpu_init_acpi - register with ACPI P-States library
@@ -365,46 +394,51 @@ static int centrino_cpu_init_acpi(struct cpufreq_policy *policy)
 	unsigned long			cur_freq;
 	int				result = 0, i;
 	unsigned int			cpu = policy->cpu;
+	struct acpi_processor_performance	*p;
+
+	p = acpi_perf_data[cpu];
 
 	/* register with ACPI core */
-	if (acpi_processor_register_performance(&p, cpu)) {
+	if (acpi_processor_register_performance(p, cpu)) {
 		dprintk(KERN_INFO PFX "obtaining ACPI data failed\n");
 		return -EIO;
 	}
+	policy->cpus = p->shared_cpu_map;
+	policy->shared_type = p->shared_type;
 
 	/* verify the acpi_data */
-	if (p.state_count <= 1) {
+	if (p->state_count <= 1) {
 		dprintk("No P-States\n");
 		result = -ENODEV;
 		goto err_unreg;
 	}
 
-	if ((p.control_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
-	    (p.status_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
+	if ((p->control_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
+	    (p->status_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
 		dprintk("Invalid control/status registers (%x - %x)\n",
-			p.control_register.space_id, p.status_register.space_id);
+			p->control_register.space_id, p->status_register.space_id);
 		result = -EIO;
 		goto err_unreg;
 	}
 
-	for (i=0; i<p.state_count; i++) {
-		if (p.states[i].control != p.states[i].status) {
+	for (i=0; i<p->state_count; i++) {
+		if (p->states[i].control != p->states[i].status) {
 			dprintk("Different control (%llu) and status values (%llu)\n",
-				p.states[i].control, p.states[i].status);
+				p->states[i].control, p->states[i].status);
 			result = -EINVAL;
 			goto err_unreg;
 		}
 
-		if (!p.states[i].core_frequency) {
+		if (!p->states[i].core_frequency) {
 			dprintk("Zero core frequency for state %u\n", i);
 			result = -EINVAL;
 			goto err_unreg;
 		}
 
-		if (p.states[i].core_frequency > p.states[0].core_frequency) {
+		if (p->states[i].core_frequency > p->states[0].core_frequency) {
 			dprintk("P%u has larger frequency (%llu) than P0 (%llu), skipping\n", i,
-				p.states[i].core_frequency, p.states[0].core_frequency);
-			p.states[i].core_frequency = 0;
+				p->states[i].core_frequency, p->states[0].core_frequency);
+			p->states[i].core_frequency = 0;
 			continue;
 		}
 	}
@@ -416,26 +450,26 @@ static int centrino_cpu_init_acpi(struct cpufreq_policy *policy)
 	}
 
 	centrino_model[cpu]->model_name=NULL;
-	centrino_model[cpu]->max_freq = p.states[0].core_frequency * 1000;
+	centrino_model[cpu]->max_freq = p->states[0].core_frequency * 1000;
 	centrino_model[cpu]->op_points =  kmalloc(sizeof(struct cpufreq_frequency_table) *
-					     (p.state_count + 1), GFP_KERNEL);
+					     (p->state_count + 1), GFP_KERNEL);
         if (!centrino_model[cpu]->op_points) {
                 result = -ENOMEM;
                 goto err_kfree;
         }
 
-        for (i=0; i<p.state_count; i++) {
-		centrino_model[cpu]->op_points[i].index = p.states[i].control;
-		centrino_model[cpu]->op_points[i].frequency = p.states[i].core_frequency * 1000;
+        for (i=0; i<p->state_count; i++) {
+		centrino_model[cpu]->op_points[i].index = p->states[i].control;
+		centrino_model[cpu]->op_points[i].frequency = p->states[i].core_frequency * 1000;
 		dprintk("adding state %i with frequency %u and control value %04x\n", 
 			i, centrino_model[cpu]->op_points[i].frequency, centrino_model[cpu]->op_points[i].index);
 	}
-	centrino_model[cpu]->op_points[p.state_count].frequency = CPUFREQ_TABLE_END;
+	centrino_model[cpu]->op_points[p->state_count].frequency = CPUFREQ_TABLE_END;
 
 	cur_freq = get_cur_freq(cpu);
 
-	for (i=0; i<p.state_count; i++) {
-		if (!p.states[i].core_frequency) {
+	for (i=0; i<p->state_count; i++) {
+		if (!p->states[i].core_frequency) {
 			dprintk("skipping state %u\n", i);
 			centrino_model[cpu]->op_points[i].frequency = CPUFREQ_ENTRY_INVALID;
 			continue;
@@ -451,7 +485,7 @@ static int centrino_cpu_init_acpi(struct cpufreq_policy *policy)
 		}
 
 		if (cur_freq == centrino_model[cpu]->op_points[i].frequency)
-			p.state = i;
+			p->state = i;
 	}
 
 	/* notify BIOS that we exist */
@@ -464,12 +498,13 @@ static int centrino_cpu_init_acpi(struct cpufreq_policy *policy)
  err_kfree:
 	kfree(centrino_model[cpu]);
  err_unreg:
-	acpi_processor_unregister_performance(&p, cpu);
+	acpi_processor_unregister_performance(p, cpu);
 	dprintk(KERN_INFO PFX "invalid ACPI data\n");
 	return (result);
 }
 #else
 static inline int centrino_cpu_init_acpi(struct cpufreq_policy *policy) { return -ENODEV; }
+static inline int centrino_cpu_early_init_acpi(void) { return 0; }
 #endif
 
 static int centrino_cpu_init(struct cpufreq_policy *policy)
@@ -555,11 +590,16 @@ static int centrino_cpu_exit(struct cpufreq_policy *policy)
 
 #ifdef CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI
 	if (!centrino_model[cpu]->model_name) {
+		static struct acpi_processor_performance *p;
+
+		if (acpi_perf_data[cpu]) {
+			p = acpi_perf_data[cpu];
 			dprintk("unregistering and freeing ACPI data\n");
-		acpi_processor_unregister_performance(&p, cpu);
+			acpi_processor_unregister_performance(p, cpu);
 			kfree(centrino_model[cpu]->op_points);
 			kfree(centrino_model[cpu]);
 		}
+	}
 #endif
 
 	centrino_model[cpu] = NULL;
@@ -592,63 +632,128 @@ static int centrino_target (struct cpufreq_policy *policy,
 			    unsigned int relation)
 {
 	unsigned int    newstate = 0;
-	unsigned int	msr, oldmsr, h, cpu = policy->cpu;
+	unsigned int	msr, oldmsr = 0, h = 0, cpu = policy->cpu;
 	struct cpufreq_freqs	freqs;
+	cpumask_t		online_policy_cpus;
 	cpumask_t		saved_mask;
-	int			retval;
+	cpumask_t		set_mask;
+	cpumask_t		covered_cpus;
+	int			retval = 0;
+	unsigned int		j, k, first_cpu, tmp;
 
-	if (centrino_model[cpu] == NULL)
+	if (unlikely(centrino_model[cpu] == NULL))
 		return -ENODEV;
 
+	if (unlikely(cpufreq_frequency_table_target(policy,
+			centrino_model[cpu]->op_points,
+			target_freq,
+			relation,
+			&newstate))) {
+		return -EINVAL;
+	}
+
+#ifdef CONFIG_HOTPLUG_CPU
+	/* cpufreq holds the hotplug lock, so we are safe from here on */
+	cpus_and(online_policy_cpus, cpu_online_map, policy->cpus);
+#else
+	online_policy_cpus = policy->cpus;
+#endif
+
+	saved_mask = current->cpus_allowed;
+	first_cpu = 1;
+	cpus_clear(covered_cpus);
+	for_each_cpu_mask(j, online_policy_cpus) {
 		/*
 		 * Support for SMP systems.
-	 * Make sure we are running on the CPU that wants to change frequency
+		 * Make sure we are running on CPU that wants to change freq
 		 */
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed(current, policy->cpus);
-	if (!cpu_isset(smp_processor_id(), policy->cpus)) {
-		dprintk("couldn't limit to CPUs in this domain\n");
-		return(-EAGAIN);
-	}
+		cpus_clear(set_mask);
+		if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
+			cpus_or(set_mask, set_mask, online_policy_cpus);
+		else
+			cpu_set(j, set_mask);
 
-	if (cpufreq_frequency_table_target(policy, centrino_model[cpu]->op_points, target_freq,
-					   relation, &newstate)) {
-		retval = -EINVAL;
+		set_cpus_allowed(current, set_mask);
+		if (unlikely(!cpu_isset(smp_processor_id(), set_mask))) {
+			dprintk("couldn't limit to CPUs in this domain\n");
+			retval = -EAGAIN;
+			if (first_cpu) {
+				/* We haven't started the transition yet. */
 				goto migrate_end;
 			}
+			break;
+		}
 
 		msr = centrino_model[cpu]->op_points[newstate].index;
-	rdmsr(MSR_IA32_PERF_CTL, oldmsr, h);
 
+		if (first_cpu) {
+			rdmsr(MSR_IA32_PERF_CTL, oldmsr, h);
 			if (msr == (oldmsr & 0xffff)) {
+				dprintk("no change needed - msr was and needs "
+					"to be %x\n", oldmsr);
 				retval = 0;
-		dprintk("no change needed - msr was and needs to be %x\n", oldmsr);
 				goto migrate_end;
 			}
 
-	freqs.cpu = cpu;
 			freqs.old = extract_clock(oldmsr, cpu, 0);
 			freqs.new = extract_clock(msr, cpu, 0);
 
 			dprintk("target=%dkHz old=%d new=%d msr=%04x\n",
 				target_freq, freqs.old, freqs.new, msr);
 
-	cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+			for_each_cpu_mask(k, online_policy_cpus) {
+				freqs.cpu = k;
+				cpufreq_notify_transition(&freqs,
+					CPUFREQ_PRECHANGE);
+			}
 
-	/* all but 16 LSB are "reserved", so treat them with
-	   care */
+			first_cpu = 0;
+			/* all but 16 LSB are reserved, treat them with care */
 			oldmsr &= ~0xffff;
 			msr &= 0xffff;
 			oldmsr |= msr;
+		}
 
 		wrmsr(MSR_IA32_PERF_CTL, oldmsr, h);
+		if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY)
+			break;
 
+		cpu_set(j, covered_cpus);
+	}
+
+	for_each_cpu_mask(k, online_policy_cpus) {
+		freqs.cpu = k;
 		cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+	}
+
+	if (unlikely(retval)) {
+		/*
+		 * We have failed halfway through the frequency change.
+		 * We have sent callbacks to policy->cpus and
+		 * MSRs have already been written on coverd_cpus.
+		 * Best effort undo..
+		 */
+
+		if (!cpus_empty(covered_cpus)) {
+			for_each_cpu_mask(j, covered_cpus) {
+				set_cpus_allowed(current, cpumask_of_cpu(j));
+				wrmsr(MSR_IA32_PERF_CTL, oldmsr, h);
+			}
+		}
+
+		tmp = freqs.new;
+		freqs.new = freqs.old;
+		freqs.old = tmp;
+		for_each_cpu_mask(j, online_policy_cpus) {
+			freqs.cpu = j;
+			cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+			cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+		}
+	}
 
-	retval = 0;
 migrate_end:
 	set_cpus_allowed(current, saved_mask);
-	return (retval);
+	return 0;
 }
 
 static struct freq_attr* centrino_attr[] = {
@@ -690,12 +795,25 @@ static int __init centrino_init(void)
 	if (!cpu_has(cpu, X86_FEATURE_EST))
 		return -ENODEV;
 
+	centrino_cpu_early_init_acpi();
+
 	return cpufreq_register_driver(&centrino_driver);
 }
 
 static void __exit centrino_exit(void)
 {
+#ifdef CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI
+	unsigned int j;
+#endif
+	
 	cpufreq_unregister_driver(&centrino_driver);
+
+#ifdef CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI
+	for_each_cpu(j) {
+		kfree(acpi_perf_data[j]);
+		acpi_perf_data[j] = NULL;
+	}
+#endif
 }
 
 MODULE_AUTHOR ("Jeremy Fitzhardinge <jeremy@goop.org>");

drivers/acpi/processor_perflib.c

@@ -554,6 +554,230 @@ static void acpi_cpufreq_remove_file(struct acpi_processor *pr)
 }
 #endif				/* CONFIG_X86_ACPI_CPUFREQ_PROC_INTF */
 
+static int acpi_processor_get_psd(struct acpi_processor	*pr)
+{
+	int result = 0;
+	acpi_status status = AE_OK;
+	struct acpi_buffer buffer = {ACPI_ALLOCATE_BUFFER, NULL};
+	struct acpi_buffer format = {sizeof("NNNNN"), "NNNNN"};
+	struct acpi_buffer state = {0, NULL};
+	union acpi_object  *psd = NULL;
+	struct acpi_psd_package *pdomain;
+
+	status = acpi_evaluate_object(pr->handle, "_PSD", NULL, &buffer);
+	if (ACPI_FAILURE(status)) {
+		return -ENODEV;
+	}
+
+	psd = (union acpi_object *) buffer.pointer;
+	if (!psd || (psd->type != ACPI_TYPE_PACKAGE)) {
+		ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Invalid _PSD data\n"));
+		result = -EFAULT;
+		goto end;
+	}
+
+	if (psd->package.count != 1) {
+		ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Invalid _PSD data\n"));
+		result = -EFAULT;
+		goto end;
+	}
+
+	pdomain = &(pr->performance->domain_info);
+
+	state.length = sizeof(struct acpi_psd_package);
+	state.pointer = pdomain;
+
+	status = acpi_extract_package(&(psd->package.elements[0]),
+		&format, &state);
+	if (ACPI_FAILURE(status)) {
+		ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Invalid _PSD data\n"));
+		result = -EFAULT;
+		goto end;
+	}
+
+	if (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES) {
+		ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Unknown _PSD:num_entries\n"));
+		result = -EFAULT;
+		goto end;
+	}
+
+	if (pdomain->revision != ACPI_PSD_REV0_REVISION) {
+		ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Unknown _PSD:revision\n"));
+		result = -EFAULT;
+		goto end;
+	}
+
+end:
+	acpi_os_free(buffer.pointer);
+	return result;
+}
+
+int acpi_processor_preregister_performance(
+		struct acpi_processor_performance **performance)
+{
+	int count, count_target;
+	int retval = 0;
+	unsigned int i, j;
+	cpumask_t covered_cpus;
+	struct acpi_processor *pr;
+	struct acpi_psd_package *pdomain;
+	struct acpi_processor *match_pr;
+	struct acpi_psd_package *match_pdomain;
+
+	down(&performance_sem);
+
+	retval = 0;
+
+	/* Call _PSD for all CPUs */
+	for_each_possible_cpu(i) {
+		pr = processors[i];
+		if (!pr) {
+			/* Look only at processors in ACPI namespace */
+			continue;
+		}
+
+		if (pr->performance) {
+			retval = -EBUSY;
+			continue;
+		}
+
+		if (!performance || !performance[i]) {
+			retval = -EINVAL;
+			continue;
+		}
+
+		pr->performance = performance[i];
+		cpu_set(i, pr->performance->shared_cpu_map);
+		if (acpi_processor_get_psd(pr)) {
+			retval = -EINVAL;
+			continue;
+		}
+	}
+	if (retval)
+		goto err_ret;
+
+	/*
+	 * Now that we have _PSD data from all CPUs, lets setup P-state 
+	 * domain info.
+	 */
+	for_each_possible_cpu(i) {
+		pr = processors[i];
+		if (!pr)
+			continue;
+
+		/* Basic validity check for domain info */
+		pdomain = &(pr->performance->domain_info);
+		if ((pdomain->revision != ACPI_PSD_REV0_REVISION) ||
+		    (pdomain->num_entries != ACPI_PSD_REV0_ENTRIES)) {
+			retval = -EINVAL;
+			goto err_ret;
+		}
+		if (pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ALL &&
+		    pdomain->coord_type != DOMAIN_COORD_TYPE_SW_ANY &&
+		    pdomain->coord_type != DOMAIN_COORD_TYPE_HW_ALL) {
+			retval = -EINVAL;
+			goto err_ret;
+		}
+	}
+
+	cpus_clear(covered_cpus);
+	for_each_possible_cpu(i) {
+		pr = processors[i];
+		if (!pr)
+			continue;
+
+		if (cpu_isset(i, covered_cpus))
+			continue;
+
+		pdomain = &(pr->performance->domain_info);
+		cpu_set(i, pr->performance->shared_cpu_map);
+		cpu_set(i, covered_cpus);
+		if (pdomain->num_processors <= 1)
+			continue;
+
+		/* Validate the Domain info */
+		count_target = pdomain->num_processors;
+		count = 1;
+		if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ALL ||
+		    pdomain->coord_type == DOMAIN_COORD_TYPE_HW_ALL) {
+			pr->performance->shared_type = CPUFREQ_SHARED_TYPE_ALL;
+		} else if (pdomain->coord_type == DOMAIN_COORD_TYPE_SW_ANY) {
+			pr->performance->shared_type = CPUFREQ_SHARED_TYPE_ANY;
+		}
+
+		for_each_possible_cpu(j) {
+			if (i == j)
+				continue;
+
+			match_pr = processors[j];
+			if (!match_pr)
+				continue;
+
+			match_pdomain = &(match_pr->performance->domain_info);
+			if (match_pdomain->domain != pdomain->domain)
+				continue;
+
+			/* Here i and j are in the same domain */
+
+			if (match_pdomain->num_processors != count_target) {
+				retval = -EINVAL;
+				goto err_ret;
+			}
+
+			if (pdomain->coord_type != match_pdomain->coord_type) {
+				retval = -EINVAL;
+				goto err_ret;
+			}
+
+			cpu_set(j, covered_cpus);
+			cpu_set(j, pr->performance->shared_cpu_map);
+			count++;
+		}
+
+		for_each_possible_cpu(j) {
+			if (i == j)
+				continue;
+
+			match_pr = processors[j];
+			if (!match_pr)
+				continue;
+
+			match_pdomain = &(match_pr->performance->domain_info);
+			if (match_pdomain->domain != pdomain->domain)
+				continue;
+
+			match_pr->performance->shared_type = 
+					pr->performance->shared_type;
+			match_pr->performance->shared_cpu_map =
+				pr->performance->shared_cpu_map;
+		}
+	}
+
+err_ret:
+	if (retval) {
+		ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "Error while parsing _PSD domain information. Assuming no coordination\n"));
+	}
+
+	for_each_possible_cpu(i) {
+		pr = processors[i];
+		if (!pr || !pr->performance)
+			continue;
+
+		/* Assume no coordination on any error parsing domain info */
+		if (retval) {
+			cpus_clear(pr->performance->shared_cpu_map);
+			cpu_set(i, pr->performance->shared_cpu_map);
+			pr->performance->shared_type = CPUFREQ_SHARED_TYPE_ALL;
+		}
+		pr->performance = NULL; /* Will be set for real in register */
+	}
+
+	up(&performance_sem);
+	return retval;
+}
+EXPORT_SYMBOL(acpi_processor_preregister_performance);
+
+
 int
 acpi_processor_register_performance(struct acpi_processor_performance
 				    *performance, unsigned int cpu)

include/acpi/pdc_intel.h

@@ -18,6 +18,11 @@
 					 ACPI_PDC_C_C1_HALT | \
 					 ACPI_PDC_P_FFH)
 
+#define ACPI_PDC_EST_CAPABILITY_SWSMP	(ACPI_PDC_SMP_C1PT | \
+					 ACPI_PDC_C_C1_HALT | \
+					 ACPI_PDC_SMP_P_SWCOORD | \
+					 ACPI_PDC_P_FFH)
+
 #define ACPI_PDC_C_CAPABILITY_SMP	(ACPI_PDC_SMP_C2C3 | \
 					 ACPI_PDC_SMP_C1PT | \
 					 ACPI_PDC_C_C1_HALT)

include/acpi/processor.h

@@ -3,6 +3,7 @@
 
 #include <linux/kernel.h>
 #include <linux/config.h>
+#include <linux/cpu.h>
 
 #include <asm/acpi.h>
 
@@ -18,6 +19,17 @@
 
 #define ACPI_PDC_REVISION_ID		0x1
 
+#define ACPI_PSD_REV0_REVISION		0 /* Support for _PSD as in ACPI 3.0 */
+#define ACPI_PSD_REV0_ENTRIES		5
+
+/*
+ * Types of coordination defined in ACPI 3.0. Same macros can be used across
+ * P, C and T states
+ */
+#define DOMAIN_COORD_TYPE_SW_ALL	0xfc
+#define DOMAIN_COORD_TYPE_SW_ANY	0xfd
+#define DOMAIN_COORD_TYPE_HW_ALL	0xfe
+
 /* Power Management */
 
 struct acpi_processor_cx;
@@ -66,6 +78,14 @@ struct acpi_processor_power {
 
 /* Performance Management */
 
+struct acpi_psd_package {
+	acpi_integer num_entries;
+	acpi_integer revision;
+	acpi_integer domain;
+	acpi_integer coord_type;
+	acpi_integer num_processors;
+} __attribute__ ((packed));
+
 struct acpi_pct_register {
 	u8 descriptor;
 	u16 length;
@@ -92,7 +112,9 @@ struct acpi_processor_performance {
 	struct acpi_pct_register status_register;
 	unsigned int state_count;
 	struct acpi_processor_px *states;
-
+	struct acpi_psd_package domain_info;
+	cpumask_t shared_cpu_map;
+	unsigned int shared_type;
 };
 
 /* Throttling Control */
@@ -161,6 +183,9 @@ struct acpi_processor_errata {
 	} piix4;
 };
 
+extern int acpi_processor_preregister_performance(
+		struct acpi_processor_performance **performance);
+
 extern int acpi_processor_register_performance(struct acpi_processor_performance
 					       *performance, unsigned int cpu);
 extern void acpi_processor_unregister_performance(struct

include/linux/cpufreq.h

@@ -73,6 +73,8 @@ struct cpufreq_real_policy {
 
 struct cpufreq_policy {
 	cpumask_t		cpus;	/* affected CPUs */
+	unsigned int		shared_type; /* ANY or ALL affected CPUs
+						should set cpufreq */
 	unsigned int		cpu;    /* cpu nr of registered CPU */
 	struct cpufreq_cpuinfo	cpuinfo;/* see above */
 
@@ -99,6 +101,8 @@ struct cpufreq_policy {
 #define CPUFREQ_INCOMPATIBLE	(1)
 #define CPUFREQ_NOTIFY		(2)
 
+#define CPUFREQ_SHARED_TYPE_ALL	(0) /* All dependent CPUs should set freq */
+#define CPUFREQ_SHARED_TYPE_ANY	(1) /* Freq can be set from any dependent CPU */
 
 /******************** cpufreq transition notifiers *******************/