[PATCH v4] Move modern AMD cpufreq support to acpi-cpufreq

[PATCH v4] Move modern AMD cpufreq support to acpi-cpufreq

[Matthew Garrett]

The programming model for cpufreq on current AMD CPUs is almost identical
to the one used on Intel and VIA hardware. This patchset merges support
into acpi-cpufreq and removes it from powernow-k8. V4 addresses all the 
review feedback received so far.

[PATCH v4 1/7] x86: Add AMD HW_PSTATE cpu feature bit and MSRs

[Matthew Garrett]

AMD CPUs that support hardware P-state setting expose this via cpuid. Add
the bit to the generic code so the ACPI cpufreq driver can make use of it,
and also add the related MSRs.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 arch/x86/include/asm/cpufeature.h |    1 +
 arch/x86/include/asm/msr-index.h  |    2 ++
 arch/x86/kernel/cpu/scattered.c   |    1 +
 3 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h
index 71cc380..acf597f 100644
--- a/arch/x86/include/asm/cpufeature.h
+++ b/arch/x86/include/asm/cpufeature.h
@@ -174,6 +174,7 @@
 #define X86_FEATURE_PLN		(7*32+ 5) /* Intel Power Limit Notification */
 #define X86_FEATURE_PTS		(7*32+ 6) /* Intel Package Thermal Status */
 #define X86_FEATURE_DTS		(7*32+ 7) /* Digital Thermal Sensor */
+#define X86_FEATURE_HW_PSTATE	(7*32+ 8) /* AMD hardware P states */
 
 /* Virtualization flags: Linux defined, word 8 */
 #define X86_FEATURE_TPR_SHADOW  (8*32+ 0) /* Intel TPR Shadow */
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 485b4f1..1772978 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -234,6 +234,8 @@
 
 #define MSR_IA32_PERF_STATUS		0x00000198
 #define MSR_IA32_PERF_CTL		0x00000199
+#define MSR_AMD_PERF_STATUS		0xc0010063
+#define MSR_AMD_PERF_CTL		0xc0010062
 
 #define MSR_IA32_MPERF			0x000000e7
 #define MSR_IA32_APERF			0x000000e8
diff --git a/arch/x86/kernel/cpu/scattered.c b/arch/x86/kernel/cpu/scattered.c
index c7f64e6..4adffc4 100644
--- a/arch/x86/kernel/cpu/scattered.c
+++ b/arch/x86/kernel/cpu/scattered.c
@@ -39,6 +39,7 @@ void __cpuinit init_scattered_cpuid_features(struct cpuinfo_x86 *c)
 		{ X86_FEATURE_APERFMPERF,	CR_ECX, 0, 0x00000006, 0 },
 		{ X86_FEATURE_EPB,		CR_ECX, 3, 0x00000006, 0 },
 		{ X86_FEATURE_XSAVEOPT,		CR_EAX,	0, 0x0000000d, 1 },
+		{ X86_FEATURE_HW_PSTATE,	CR_EDX, 7, 0x80000007, 0 },
 		{ X86_FEATURE_CPB,		CR_EDX, 9, 0x80000007, 0 },
 		{ X86_FEATURE_NPT,		CR_EDX, 0, 0x8000000a, 0 },
 		{ X86_FEATURE_LBRV,		CR_EDX, 1, 0x8000000a, 0 },
-- 
1.7.6

[PATCH v4 2/7] acpi-cpufreq: Add support for modern AMD CPUs

[Matthew Garrett]

The programming model for P-states on modern AMD CPUs is very similar to
that of Intel and VIA. It makes sense to consolidate this support into one
driver rather than duplicating functionality between two of them. This
patch adds support for AMDs with hardware P-state control to acpi-cpufreq.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 drivers/cpufreq/acpi-cpufreq.c |   43 ++++++++++++++++++++++++++++++++++-----
 1 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
index 596d5dd..298d451 100644
--- a/drivers/cpufreq/acpi-cpufreq.c
+++ b/drivers/cpufreq/acpi-cpufreq.c
@@ -54,10 +54,12 @@ MODULE_LICENSE("GPL");
 enum {
 	UNDEFINED_CAPABLE = 0,
 	SYSTEM_INTEL_MSR_CAPABLE,
+	SYSTEM_AMD_MSR_CAPABLE,
 	SYSTEM_IO_CAPABLE,
 };
 
 #define INTEL_MSR_RANGE		(0xffff)
+#define AMD_MSR_RANGE		(0x7)
 
 struct acpi_cpufreq_data {
 	struct acpi_processor_performance *acpi_data;
@@ -82,6 +84,13 @@ static int check_est_cpu(unsigned int cpuid)
 	return cpu_has(cpu, X86_FEATURE_EST);
 }
 
+static int check_amd_hwpstate_cpu(unsigned int cpuid)
+{
+	struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
+
+	return cpu_has(cpu, X86_FEATURE_HW_PSTATE);
+}
+
 static unsigned extract_io(u32 value, struct acpi_cpufreq_data *data)
 {
 	struct acpi_processor_performance *perf;
@@ -101,7 +110,11 @@ static unsigned extract_msr(u32 msr, struct acpi_cpufreq_data *data)
 	int i;
 	struct acpi_processor_performance *perf;
 
-	msr &= INTEL_MSR_RANGE;
+	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
+		msr &= AMD_MSR_RANGE;
+	else
+		msr &= INTEL_MSR_RANGE;
+
 	perf = data->acpi_data;
 
 	for (i = 0; data->freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
@@ -115,6 +128,7 @@ static unsigned extract_freq(u32 val, struct acpi_cpufreq_data *data)
 {
 	switch (data->cpu_feature) {
 	case SYSTEM_INTEL_MSR_CAPABLE:
+	case SYSTEM_AMD_MSR_CAPABLE:
 		return extract_msr(val, data);
 	case SYSTEM_IO_CAPABLE:
 		return extract_io(val, data);
@@ -150,6 +164,7 @@ static void do_drv_read(void *_cmd)
 
 	switch (cmd->type) {
 	case SYSTEM_INTEL_MSR_CAPABLE:
+	case SYSTEM_AMD_MSR_CAPABLE:
 		rdmsr(cmd->addr.msr.reg, cmd->val, h);
 		break;
 	case SYSTEM_IO_CAPABLE:
@@ -174,6 +189,9 @@ static void do_drv_write(void *_cmd)
 		lo = (lo & ~INTEL_MSR_RANGE) | (cmd->val & INTEL_MSR_RANGE);
 		wrmsr(cmd->addr.msr.reg, lo, hi);
 		break;
+	case SYSTEM_AMD_MSR_CAPABLE:
+		wrmsr(cmd->addr.msr.reg, cmd->val, 0);
+		break;
 	case SYSTEM_IO_CAPABLE:
 		acpi_os_write_port((acpi_io_address)cmd->addr.io.port,
 				cmd->val,
@@ -217,6 +235,10 @@ static u32 get_cur_val(const struct cpumask *mask)
 		cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
 		cmd.addr.msr.reg = MSR_IA32_PERF_STATUS;
 		break;
+	case SYSTEM_AMD_MSR_CAPABLE:
+		cmd.type = SYSTEM_AMD_MSR_CAPABLE;
+		cmd.addr.msr.reg = MSR_AMD_PERF_STATUS;
+		break;
 	case SYSTEM_IO_CAPABLE:
 		cmd.type = SYSTEM_IO_CAPABLE;
 		perf = per_cpu(acfreq_data, cpumask_first(mask))->acpi_data;
@@ -326,6 +348,11 @@ static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 		cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
 		cmd.val = (u32) perf->states[next_perf_state].control;
 		break;
+	case SYSTEM_AMD_MSR_CAPABLE:
+		cmd.type = SYSTEM_AMD_MSR_CAPABLE;
+		cmd.addr.msr.reg = MSR_AMD_PERF_CTL;
+		cmd.val = (u32) perf->states[next_perf_state].control;
+		break;
 	case SYSTEM_IO_CAPABLE:
 		cmd.type = SYSTEM_IO_CAPABLE;
 		cmd.addr.io.port = perf->control_register.address;
@@ -580,12 +607,16 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
 		break;
 	case ACPI_ADR_SPACE_FIXED_HARDWARE:
 		pr_debug("HARDWARE addr space\n");
-		if (!check_est_cpu(cpu)) {
-			result = -ENODEV;
-			goto err_unreg;
+		if (check_est_cpu(cpu)) {
+			data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
+			break;
 		}
-		data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
-		break;
+		if (check_amd_hwpstate_cpu(cpu)) {
+			data->cpu_feature = SYSTEM_AMD_MSR_CAPABLE;
+			break;
+		}
+		result = -ENODEV;
+		goto err_unreg;
 	default:
 		pr_debug("Unknown addr space %d\n",
 			(u32) (perf->control_register.space_id));
-- 
1.7.6

[PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Matthew Garrett]

One feature present in powernow-k8 that isn't present in acpi-cpufreq is
support for enabling or disabling AMD's core performance boost technology.
This patch adds that support to acpi-cpufreq, but also extends it to allow
Intel's dynamic acceleration to be disabled via the same interface. The
sysfs entry retains the cpb name for compatibility purposes.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 drivers/cpufreq/acpi-cpufreq.c |  191 ++++++++++++++++++++++++++++++++++++++++
 1 files changed, 191 insertions(+), 0 deletions(-)

diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
index 298d451..0a92277 100644
--- a/drivers/cpufreq/acpi-cpufreq.c
+++ b/drivers/cpufreq/acpi-cpufreq.c
@@ -76,6 +76,107 @@ static struct acpi_processor_performance __percpu *acpi_perf_data;
 static struct cpufreq_driver acpi_cpufreq_driver;
 
 static unsigned int acpi_pstate_strict;
+static bool cpb_enabled, cpb_supported;
+static struct msr __percpu *msrs;
+
+static bool cpb_state(unsigned int cpu)
+{
+	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
+	u32 lo, hi;
+	u64 msr;
+
+	switch (data->cpu_feature) {
+	case SYSTEM_INTEL_MSR_CAPABLE:
+		rdmsr_on_cpu(cpu, MSR_IA32_MISC_ENABLE, &lo, &hi);
+		msr = lo | ((u64)hi << 32);
+		return !(msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE);
+	case SYSTEM_AMD_MSR_CAPABLE:
+		rdmsr_on_cpu(cpu, MSR_K7_HWCR, &lo, &hi);
+		msr = lo | ((u64)hi << 32);
+		return !(msr & BIT(25));
+	}
+	return false;
+}
+
+static void _cpb_toggle_msrs(bool enable)
+{
+	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, 0);
+	u32 cpu;
+
+	get_online_cpus();
+
+	switch (data->cpu_feature) {
+	case SYSTEM_INTEL_MSR_CAPABLE:
+		rdmsr_on_cpus(cpu_online_mask, MSR_IA32_MISC_ENABLE, msrs);
+
+		for_each_cpu(cpu, cpu_online_mask) {
+			struct msr *reg = per_cpu_ptr(msrs, cpu);
+			if (enable)
+				reg->q &= ~MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
+			else
+				reg->q |= MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
+		}
+
+		wrmsr_on_cpus(cpu_online_mask, MSR_IA32_MISC_ENABLE, msrs);
+		break;
+	case SYSTEM_AMD_MSR_CAPABLE:
+		rdmsr_on_cpus(cpu_online_mask, MSR_K7_HWCR, msrs);
+
+		for_each_cpu(cpu, cpu_online_mask) {
+			struct msr *reg = per_cpu_ptr(msrs, cpu);
+			if (enable)
+				reg->l &= ~BIT(25);
+			else
+				reg->l |= BIT(25);
+		}
+
+		wrmsr_on_cpus(cpu_online_mask, MSR_K7_HWCR, msrs);
+		break;
+	}
+
+	put_online_cpus();
+}
+
+static void cpb_toggle(bool enable)
+{
+	if (enable && !cpb_enabled) {
+		cpb_enabled = true;
+		_cpb_toggle_msrs(enable);
+		pr_debug("Core Boosting enabled.\n");
+	} else if (!enable && cpb_enabled) {
+		cpb_enabled = false;
+		_cpb_toggle_msrs(enable);
+		pr_debug("Core Boosting disabled.\n");
+	}
+}
+
+static ssize_t store_cpb(struct cpufreq_policy *policy, const char *buf,
+			 size_t count)
+{
+	int ret = -EINVAL;
+	unsigned long val = 0;
+
+	ret = strict_strtoul(buf, 10, &val);
+	if (!ret && (val == 0 || val == 1) && cpb_supported)
+		cpb_toggle(val);
+	else
+		return -EINVAL;
+
+	return count;
+}
+
+static ssize_t show_cpb(struct cpufreq_policy *policy, char *buf)
+{
+	unsigned int cpu = policy->cpu;
+
+	return sprintf(buf, "%u\n", cpb_state(cpu));
+}
+
+#define define_one_rw(_name) \
+static struct freq_attr _name = \
+	__ATTR(_name, 0644, show_##_name, store_##_name)
+
+define_one_rw(cpb);
 
 static int check_est_cpu(unsigned int cpuid)
 {
@@ -446,6 +547,70 @@ static void free_acpi_perf_data(void)
 	free_percpu(acpi_perf_data);
 }
 
+static int cpb_notify(struct notifier_block *nb, unsigned long action,
+		      void *hcpu)
+{
+	unsigned cpu = (long)hcpu;
+	u32 lo, hi;
+	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
+	int msr;
+	u64 bit;
+
+	switch (data->cpu_feature) {
+	case SYSTEM_INTEL_MSR_CAPABLE:
+		msr = MSR_IA32_MISC_ENABLE;
+		bit = MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
+		break;
+	case SYSTEM_AMD_MSR_CAPABLE:
+		msr = MSR_K7_HWCR;
+		bit = BIT(25);
+		break;
+	default:
+		return NOTIFY_OK;
+	}
+
+	/*
+	 * Clear the boost-disable flag on the CPU_DOWN path so that
+	 * this cpu cannot block the remaining ones from boosting. On
+	 * the CPU_UP path we simply keep the boost-disable flag in
+	 * sync with the current global state.
+	 */
+
+	switch (action) {
+	case CPU_UP_PREPARE:
+	case CPU_UP_PREPARE_FROZEN:
+		if (!cpb_enabled) {
+			rdmsr_on_cpu(cpu, msr, &lo, &hi);
+			if (bit < BIT(32))
+				lo |= bit;
+			else
+				hi |= (bit >> 32);
+			wrmsr_on_cpu(cpu, msr, lo, hi);
+		}
+		break;
+
+	case CPU_DOWN_PREPARE:
+	case CPU_DOWN_PREPARE_FROZEN:
+		rdmsr_on_cpu(cpu, msr, &lo, &hi);
+		if (bit < BIT(32))
+			lo &= ~bit;
+		else
+			hi &= ~(bit >> 32);
+		wrmsr_on_cpu(cpu, msr, lo, hi);
+		break;
+
+	default:
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+
+static struct notifier_block cpb_nb = {
+	.notifier_call          = cpb_notify,
+};
+
 /*
  * acpi_cpufreq_early_init - initialize ACPI P-States library
  *
@@ -666,6 +831,24 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
 	if (result)
 		goto err_freqfree;
 
+	if (boot_cpu_has(X86_FEATURE_CPB) || boot_cpu_has(X86_FEATURE_IDA)) {
+		msrs = msrs_alloc();
+
+		if (!msrs) {
+			result = -ENOMEM;
+			goto err_freqfree;
+		}
+
+		cpb_supported = true;
+
+		cpb_enabled = cpb_state(0);
+
+		register_cpu_notifier(&cpb_nb);
+
+		/* Make sure all CPUs have the same state */
+		_cpb_toggle_msrs(cpb_enabled);
+	}
+
 	if (perf->states[0].core_frequency * 1000 != policy->cpuinfo.max_freq)
 		printk(KERN_WARNING FW_WARN "P-state 0 is not max freq\n");
 
@@ -749,6 +932,7 @@ static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
 
 static struct freq_attr *acpi_cpufreq_attr[] = {
 	&cpufreq_freq_attr_scaling_available_freqs,
+	&cpb,
 	NULL,
 };
 
@@ -788,6 +972,13 @@ static void __exit acpi_cpufreq_exit(void)
 {
 	pr_debug("acpi_cpufreq_exit\n");
 
+	if (msrs) {
+		unregister_cpu_notifier(&cpb_nb);
+
+		msrs_free(msrs);
+		msrs = NULL;
+	}
+
 	cpufreq_unregister_driver(&acpi_cpufreq_driver);
 
 	free_acpi_perf_data();
-- 
1.7.6

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Borislav Petkov]

On Mon, Jul 18, 2011 at 12:37:38PM -0400, Matthew Garrett wrote:
> One feature present in powernow-k8 that isn't present in acpi-cpufreq is
> support for enabling or disabling AMD's core performance boost technology.
> This patch adds that support to acpi-cpufreq, but also extends it to allow
> Intel's dynamic acceleration to be disabled via the same interface. The
> sysfs entry retains the cpb name for compatibility purposes.
> 
> Signed-off-by: Matthew Garrett <mjg@redhat.com>
> ---
>  drivers/cpufreq/acpi-cpufreq.c |  191 ++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 191 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
> index 298d451..0a92277 100644
> --- a/drivers/cpufreq/acpi-cpufreq.c
> +++ b/drivers/cpufreq/acpi-cpufreq.c
> @@ -76,6 +76,107 @@ static struct acpi_processor_performance __percpu *acpi_perf_data;
>  static struct cpufreq_driver acpi_cpufreq_driver;
>  
>  static unsigned int acpi_pstate_strict;
> +static bool cpb_enabled, cpb_supported;
> +static struct msr __percpu *msrs;
> +
> +static bool cpb_state(unsigned int cpu)
> +{
> +	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
> +	u32 lo, hi;
> +	u64 msr;
> +
> +	switch (data->cpu_feature) {
> +	case SYSTEM_INTEL_MSR_CAPABLE:
> +		rdmsr_on_cpu(cpu, MSR_IA32_MISC_ENABLE, &lo, &hi);
> +		msr = lo | ((u64)hi << 32);
> +		return !(msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE);
> +	case SYSTEM_AMD_MSR_CAPABLE:
> +		rdmsr_on_cpu(cpu, MSR_K7_HWCR, &lo, &hi);
> +		msr = lo | ((u64)hi << 32);
> +		return !(msr & BIT(25));
> +	}
> +	return false;
> +}

You need to handle the case where boosting is not even supported by the
CPU. For example, a Barcelona CPU does not boost but I get

$ cat /sys/devices/system/cpu/cpu0/cpufreq/boost
1

This is because this bit in the register is simply reserved and it
countains some random value (0 in this case).

What you need to do is look at the cpu_enabled variable which is caching
the current boost state. Besides, you don't want to do a MSR read
everytime you 'cat' the sysfs entry. See how powernow-k8 does it. I'm
guessing with Intel it should be similar.

Thanks.

-- 
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Matthew Garrett]

On Fri, Jul 22, 2011 at 04:41:43PM +0200, Borislav Petkov wrote:

> You need to handle the case where boosting is not even supported by the
> CPU. For example, a Barcelona CPU does not boost but I get
> 
> $ cat /sys/devices/system/cpu/cpu0/cpufreq/boost
> 1

Ah, yeah. Fair enough.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Andy Lutomirski]

On 07/18/2011 12:37 PM, Matthew Garrett wrote:
> One feature present in powernow-k8 that isn't present in acpi-cpufreq is
> support for enabling or disabling AMD's core performance boost technology.
> This patch adds that support to acpi-cpufreq, but also extends it to allow
> Intel's dynamic acceleration to be disabled via the same interface. The
> sysfs entry retains the cpb name for compatibility purposes.
>
> Signed-off-by: Matthew Garrett<mjg@redhat.com>
> ---
>   drivers/cpufreq/acpi-cpufreq.c |  191 ++++++++++++++++++++++++++++++++++++++++
>   1 files changed, 191 insertions(+), 0 deletions(-)
>

How is this different from setting 3400000 vs 3401000 as the max freq? 
(I'm probably missing something about the way that Intel's turbo mode 
works.)

> +static ssize_t store_cpb(struct cpufreq_policy *policy, const char *buf,
> +			 size_t count)
> +{
> +	int ret = -EINVAL;
> +	unsigned long val = 0;
> +
> +	ret = strict_strtoul(buf, 10,&val);
> +	if (!ret&&  (val == 0 || val == 1)&&  cpb_supported)
> +		cpb_toggle(val);

Shouldn't this set the flag only on policy->cpu (as opposed to all 
online cpus?)

Also, you're missing a space before &&.

--Andy

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Matthew Garrett]

On Sat, Jul 23, 2011 at 11:06:29PM -0400, Andy Lutomirski wrote:
> On 07/18/2011 12:37 PM, Matthew Garrett wrote:
> >One feature present in powernow-k8 that isn't present in acpi-cpufreq is
> >support for enabling or disabling AMD's core performance boost technology.
> >This patch adds that support to acpi-cpufreq, but also extends it to allow
> >Intel's dynamic acceleration to be disabled via the same interface. The
> >sysfs entry retains the cpb name for compatibility purposes.
> >
> >Signed-off-by: Matthew Garrett<mjg@redhat.com>
> >---
> >  drivers/cpufreq/acpi-cpufreq.c |  191 ++++++++++++++++++++++++++++++++++++++++
> >  1 files changed, 191 insertions(+), 0 deletions(-)
> >
> 
> How is this different from setting 3400000 vs 3401000 as the max
> freq? (I'm probably missing something about the way that Intel's
> turbo mode works.)

It's not really, it just means that there's a shared interface to 
achieve the same goal.

> >+	unsigned long val = 0;
> >+
> >+	ret = strict_strtoul(buf, 10,&val);
> >+	if (!ret&&  (val == 0 || val == 1)&&  cpb_supported)
> >+		cpb_toggle(val);
> 
> Shouldn't this set the flag only on policy->cpu (as opposed to all
> online cpus?)

It's effectively a per-package MSR rather than a per-thread or per-core 
one. We could make this per-package rather than system-wide, but I don't 
really know of any use cases that would benefit from the extra code. If 
anyone cares enough it could be added, but I'm a great believer in 
letting people come up with arguments for adding something before doing 
it...

> Also, you're missing a space before &&.

Think that's a cut and paste error. I'll fix it up.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Andy Lutomirski]

On 07/18/2011 12:37 PM, Matthew Garrett wrote:
> One feature present in powernow-k8 that isn't present in acpi-cpufreq is
> support for enabling or disabling AMD's core performance boost technology.
> This patch adds that support to acpi-cpufreq, but also extends it to allow
> Intel's dynamic acceleration to be disabled via the same interface. The
> sysfs entry retains the cpb name for compatibility purposes.
>
> Signed-off-by: Matthew Garrett<mjg@redhat.com>
> ---
>   drivers/cpufreq/acpi-cpufreq.c |  191 ++++++++++++++++++++++++++++++++++++++++
>   1 files changed, 191 insertions(+), 0 deletions(-)
>

> +			if (enable)
> +				reg->q&= ~MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
> +			else
> +				reg->q |= MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
> +		}

I was curious and looked it up.  Intel SDM volume 3, 14.3.2.1 says:

Opportunistic processor performance operation can be disabled by setting 
bit 38 of
IA32_MISC_ENABLE. This mechanism is intended for BIOS only. If
IA32_MISC_ENABLE[38] is set, CPUID.06H:EAX[1] will return 0.

and the next section says:

System software can temporarily disengage opportunistic processor 
performance
operation by setting bit 32 of the IA32_PERF_CTL MSR (0199H), using a read-
modify-write sequence on the MSR.

Would that be a better way to control boost mode on Intel chips?

--Andy

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Matthew Garrett]

On Sat, Jul 23, 2011 at 11:15:02PM -0400, Andy Lutomirski wrote:

> I was curious and looked it up.  Intel SDM volume 3, 14.3.2.1 says:
> 
> Opportunistic processor performance operation can be disabled by
> setting bit 38 of
> IA32_MISC_ENABLE. This mechanism is intended for BIOS only. If
> IA32_MISC_ENABLE[38] is set, CPUID.06H:EAX[1] will return 0.

Hm. Interesting. The observed behaviour is that it works fine if we 
toggle this ourselves, and I don't remember the version of the docs I 
was looking at adding the BIOS proviso. Having said that...

> System software can temporarily disengage opportunistic processor
> performance
> operation by setting bit 32 of the IA32_PERF_CTL MSR (0199H), using a read-
> modify-write sequence on the MSR.

This complicates things a little, since right now we just write the 
firmware's P state value directly into PERF_CTL. We'd need to add code 
to acpi_cpufreq_target to make sure that it masked that bit off. It's a 
little more awkward, but if we're being told not to do it by just 
hitting the bit in MISC_ENABLE it's probably worth it. I'll try to 
handle that this week.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Andrew Lutomirski]

On Sat, Jul 23, 2011 at 11:33 PM, Matthew Garrett <mjg@redhat.com> wrote:
> On Sat, Jul 23, 2011 at 11:15:02PM -0400, Andy Lutomirski wrote:
>
>> I was curious and looked it up.  Intel SDM volume 3, 14.3.2.1 says:
>>
>> Opportunistic processor performance operation can be disabled by
>> setting bit 38 of
>> IA32_MISC_ENABLE. This mechanism is intended for BIOS only. If
>> IA32_MISC_ENABLE[38] is set, CPUID.06H:EAX[1] will return 0.
>
> Hm. Interesting. The observed behaviour is that it works fine if we
> toggle this ourselves, and I don't remember the version of the docs I
> was looking at adding the BIOS proviso. Having said that...
>
>> System software can temporarily disengage opportunistic processor
>> performance
>> operation by setting bit 32 of the IA32_PERF_CTL MSR (0199H), using a read-
>> modify-write sequence on the MSR.
>
> This complicates things a little, since right now we just write the
> firmware's P state value directly into PERF_CTL. We'd need to add code
> to acpi_cpufreq_target to make sure that it masked that bit off. It's a
> little more awkward, but if we're being told not to do it by just
> hitting the bit in MISC_ENABLE it's probably worth it. I'll try to
> handle that this week.

This way may give the benefit of making it work per core instead of
per package.  The manual is rather unclear on this point.

I actually have a use case for this.  I have a system that keeps a
bunch of cores under moderate load.  I have one thread in particular
that needs to be fast, and I'd like to disable boosting on the other
cores to keep more thermal and power headroom available for the one
thread that cares.

Of course, I can probably accomplish this with scaling_max_freq as well.

--Andy

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Matthew Garrett]

On Sat, Jul 23, 2011 at 11:42:50PM -0400, Andrew Lutomirski wrote:
> On Sat, Jul 23, 2011 at 11:33 PM, Matthew Garrett <mjg@redhat.com> wrote:
> > This complicates things a little, since right now we just write the
> > firmware's P state value directly into PERF_CTL. We'd need to add code
> > to acpi_cpufreq_target to make sure that it masked that bit off. It's a
> > little more awkward, but if we're being told not to do it by just
> > hitting the bit in MISC_ENABLE it's probably worth it. I'll try to
> > handle that this week.
> 
> This way may give the benefit of making it work per core instead of
> per package.  The manual is rather unclear on this point.

Being able to enter turbo mode typically requires coordination between 
the cores in order to ensure that the package remains within limits. The 
AMD implementation certainly disables their equivalent entirely if any 
core in the package has it disabled. I haven't verified that Intel 
behaviour is identical, but it wouldn't surprise me. I can try to check 
that.

> I actually have a use case for this.  I have a system that keeps a
> bunch of cores under moderate load.  I have one thread in particular
> that needs to be fast, and I'd like to disable boosting on the other
> cores to keep more thermal and power headroom available for the one
> thread that cares.

Are the other threads sufficiently opportunistic to use extra CPU power 
if it's available to them? You'll generally only get turbo if the other 
cores are in C6, so even if turbo is disabled on a specific core it'll 
probably prevent another core from entering turbo if anything's 
executing on it. You'd arguably want it to be able to get into turbo so 
it can hit C6 more quickly and let the other thread use the extra 
headroom.
-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: [PATCH v4 3/7] acpi-cpufreq: Add support for disabling dynamic overclocking

[Borislav Petkov]

On Sat, Jul 23, 2011 at 11:49:37PM -0400, Matthew Garrett wrote:
> > This way may give the benefit of making it work per core instead of
> > per package.  The manual is rather unclear on this point.
> 
> Being able to enter turbo mode typically requires coordination between 
> the cores in order to ensure that the package remains within limits. The 
> AMD implementation certainly disables their equivalent entirely if any 
> core in the package has it disabled. I haven't verified that Intel 
> behaviour is identical, but it wouldn't surprise me. I can try to check 
> that.

We actually can do both on family 12h - per core and per package
disable. Family 10h, revE does only per-package disable. We opted for
having a single interface for sw and always do per-package disable
simply because we have no usecases for per-core disable and I agree with
Matthew - we'll implement it only when it's needed.

> > I actually have a use case for this.  I have a system that keeps a
> > bunch of cores under moderate load.  I have one thread in particular
> > that needs to be fast, and I'd like to disable boosting on the other
> > cores to keep more thermal and power headroom available for the one
> > thread that cares.
> 
> Are the other threads sufficiently opportunistic to use extra CPU power 
> if it's available to them? You'll generally only get turbo if the other 
> cores are in C6, so even if turbo is disabled on a specific core it'll 
> probably prevent another core from entering turbo if anything's 
> executing on it. You'd arguably want it to be able to get into turbo so 
> it can hit C6 more quickly and let the other thread use the extra 
> headroom.

Right, so we were looking for per-core disable use cases too while
discussing this internally. Andy, limiting the other cores to a higher
P-state (lower freq) and letting the one core boost with a higher
headroom might actually give you more bang than turning off boost on the
n-1 cores. This definitely needs some experimenting and measurements
before you can say for sure. And it all depends on the specific workload
and boosting algorithm.

There's this cpufreq-aperf tool in cpufrequtils which shows
you the boosted freq and C-state residency of the cores, or
<tools/power/x86/turbostat/> in the kernel sources - you might wanna do
some measurements with those to actually have some hard data for your
workload ...

HTH.

-- 
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551

[PATCH v4 4/7] ACPI: Add fixups for AMD P-state figures

[Matthew Garrett]

Some AMD systems may round the frequencies in ACPI tables to 100MHz
boundaries. We can obtain the real frequencies from MSRs, so add a quirk
to fix these frequencies up on AMD systems.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 arch/x86/include/asm/msr-index.h |    1 +
 drivers/acpi/processor_perflib.c |   30 ++++++++++++++++++++++++++++++
 2 files changed, 31 insertions(+), 0 deletions(-)

diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 1772978..1449569 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -234,6 +234,7 @@
 
 #define MSR_IA32_PERF_STATUS		0x00000198
 #define MSR_IA32_PERF_CTL		0x00000199
+#define MSR_AMD_PSTATE_DEF_BASE		0xc0010064
 #define MSR_AMD_PERF_STATUS		0xc0010063
 #define MSR_AMD_PERF_CTL		0xc0010062
 
diff --git a/drivers/acpi/processor_perflib.c b/drivers/acpi/processor_perflib.c
index 85b3237..aed97d7 100644
--- a/drivers/acpi/processor_perflib.c
+++ b/drivers/acpi/processor_perflib.c
@@ -302,6 +302,34 @@ static int acpi_processor_get_performance_control(struct acpi_processor *pr)
 	return result;
 }
 
+#ifdef CONFIG_X86
+/*
+ * Some AMDs have 50MHz frequency multiples, but only provide 100MHz rounding
+ * in their ACPI data. Calculate the real values and fix up the _PSS data.
+ */
+static void amd_fixup_frequency(struct acpi_processor_px *px, int i)
+{
+	u32 hi, lo, fid, did;
+	int index = px->control & 0x00000007;
+
+	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
+		return;
+
+	if ((boot_cpu_data.x86 == 0x10 && boot_cpu_data.x86_model < 10)
+	    || boot_cpu_data.x86 == 0x11) {
+		rdmsr(MSR_AMD_PSTATE_DEF_BASE + index, lo, hi);
+		fid = lo & 0x3f;
+		did = (lo >> 6) & 7;
+		if (boot_cpu_data.x86 == 0x10)
+			px->core_frequency = (100 * (fid + 0x10)) >> did;
+		else
+			px->core_frequency = (100 * (fid + 8)) >> did;
+	}
+}
+#else
+static void amd_fixup_frequency(struct acpi_processor_px *px, int i) {};
+#endif
+
 static int acpi_processor_get_performance_states(struct acpi_processor *pr)
 {
 	int result = 0;
@@ -356,6 +384,8 @@ static int acpi_processor_get_performance_states(struct acpi_processor *pr)
 			goto end;
 		}
 
+		amd_fixup_frequency(px, i);
+
 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 				  "State [%d]: core_frequency[%d] power[%d] transition_latency[%d] bus_master_latency[%d] control[0x%x] status[0x%x]\n",
 				  i,
-- 
1.7.6

[PATCH v4 5/7] cpufreq: Add compatibility hack to powernow-k8

[Matthew Garrett]

cpufreq modules are often loaded from init scripts that assume that all
recent AMD systems will use powernow-k8, so we should ensure that loading
it triggers a load of acpi-cpufreq if the latter is built as a module.
This avoids the problem of users ending up without any cpufreq support
after the transition.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 drivers/cpufreq/powernow-k8.c |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/drivers/cpufreq/powernow-k8.c b/drivers/cpufreq/powernow-k8.c
index bce576d..8de4446 100644
--- a/drivers/cpufreq/powernow-k8.c
+++ b/drivers/cpufreq/powernow-k8.c
@@ -1549,8 +1549,12 @@ static int __cpuinit powernowk8_init(void)
 			supported_cpus++;
 	}
 
-	if (supported_cpus != num_online_cpus())
+	if (supported_cpus != num_online_cpus()) {
+		if (static_cpu_has(X86_FEATURE_HW_PSTATE))
+			request_module("acpi_cpufreq");
+
 		return -ENODEV;
+	}
 
 	printk(KERN_INFO PFX "Found %d %s (%d cpu cores) (" VERSION ")\n",
 		num_online_nodes(), boot_cpu_data.x86_model_id, supported_cpus);
-- 
1.7.6

[PATCH v4 6/7] cpufreq: Remove support for hardware P-state chips from powernow-k8

[Matthew Garrett]

These chips are now supported by acpi-cpufreq, so we can delete all the
code handling them.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 drivers/cpufreq/Makefile      |    2 +-
 drivers/cpufreq/powernow-k8.c |  375 +++--------------------------------------
 drivers/cpufreq/powernow-k8.h |   32 ----
 3 files changed, 26 insertions(+), 383 deletions(-)

diff --git a/drivers/cpufreq/Makefile b/drivers/cpufreq/Makefile
index e2fc2d2..f8813d4 100644
--- a/drivers/cpufreq/Makefile
+++ b/drivers/cpufreq/Makefile
@@ -19,7 +19,7 @@ obj-$(CONFIG_CPU_FREQ_TABLE)		+= freq_table.o
 # K8 systems. ACPI is preferred to all other hardware-specific drivers.
 # speedstep-* is preferred over p4-clockmod.
 
-obj-$(CONFIG_X86_POWERNOW_K8)		+= powernow-k8.o mperf.o
+obj-$(CONFIG_X86_POWERNOW_K8)		+= powernow-k8.o
 obj-$(CONFIG_X86_ACPI_CPUFREQ)		+= acpi-cpufreq.o mperf.o
 obj-$(CONFIG_X86_PCC_CPUFREQ)		+= pcc-cpufreq.o
 obj-$(CONFIG_X86_POWERNOW_K6)		+= powernow-k6.o
diff --git a/drivers/cpufreq/powernow-k8.c b/drivers/cpufreq/powernow-k8.c
index 8de4446..783ba82 100644
--- a/drivers/cpufreq/powernow-k8.c
+++ b/drivers/cpufreq/powernow-k8.c
@@ -45,19 +45,12 @@
 #define PFX "powernow-k8: "
 #define VERSION "version 2.20.00"
 #include "powernow-k8.h"
-#include "mperf.h"
 
 /* serialize freq changes  */
 static DEFINE_MUTEX(fidvid_mutex);
 
 static DEFINE_PER_CPU(struct powernow_k8_data *, powernow_data);
 
-static int cpu_family = CPU_OPTERON;
-
-/* core performance boost */
-static bool cpb_capable, cpb_enabled;
-static struct msr __percpu *msrs;
-
 static struct cpufreq_driver cpufreq_amd64_driver;
 
 #ifndef CONFIG_SMP
@@ -79,12 +72,6 @@ static u32 find_khz_freq_from_fid(u32 fid)
 	return 1000 * find_freq_from_fid(fid);
 }
 
-static u32 find_khz_freq_from_pstate(struct cpufreq_frequency_table *data,
-		u32 pstate)
-{
-	return data[pstate].frequency;
-}
-
 /* Return the vco fid for an input fid
  *
  * Each "low" fid has corresponding "high" fid, and you can get to "low" fids
@@ -107,9 +94,6 @@ static int pending_bit_stuck(void)
 {
 	u32 lo, hi;
 
-	if (cpu_family == CPU_HW_PSTATE)
-		return 0;
-
 	rdmsr(MSR_FIDVID_STATUS, lo, hi);
 	return lo & MSR_S_LO_CHANGE_PENDING ? 1 : 0;
 }
@@ -123,20 +107,6 @@ static int query_current_values_with_pending_wait(struct powernow_k8_data *data)
 	u32 lo, hi;
 	u32 i = 0;
 
-	if (cpu_family == CPU_HW_PSTATE) {
-		rdmsr(MSR_PSTATE_STATUS, lo, hi);
-		i = lo & HW_PSTATE_MASK;
-		data->currpstate = i;
-
-		/*
-		 * a workaround for family 11h erratum 311 might cause
-		 * an "out-of-range Pstate if the core is in Pstate-0
-		 */
-		if ((boot_cpu_data.x86 == 0x11) && (i >= data->numps))
-			data->currpstate = HW_PSTATE_0;
-
-		return 0;
-	}
 	do {
 		if (i++ > 10000) {
 			pr_debug("detected change pending stuck\n");
@@ -293,14 +263,6 @@ static int decrease_vid_code_by_step(struct powernow_k8_data *data,
 	return 0;
 }
 
-/* Change hardware pstate by single MSR write */
-static int transition_pstate(struct powernow_k8_data *data, u32 pstate)
-{
-	wrmsr(MSR_PSTATE_CTRL, pstate, 0);
-	data->currpstate = pstate;
-	return 0;
-}
-
 /* Change Opteron/Athlon64 fid and vid, by the 3 phases. */
 static int transition_fid_vid(struct powernow_k8_data *data,
 		u32 reqfid, u32 reqvid)
@@ -551,15 +513,8 @@ static void check_supported_cpu(void *_rc)
 				"Power state transitions not supported\n");
 			return;
 		}
-	} else { /* must be a HW Pstate capable processor */
-		cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
-		if ((edx & USE_HW_PSTATE) == USE_HW_PSTATE)
-			cpu_family = CPU_HW_PSTATE;
-		else
-			return;
+		*rc = 0;
 	}
-
-	*rc = 0;
 }
 
 static int check_pst_table(struct powernow_k8_data *data, struct pst_s *pst,
@@ -623,18 +578,11 @@ static void print_basics(struct powernow_k8_data *data)
 	for (j = 0; j < data->numps; j++) {
 		if (data->powernow_table[j].frequency !=
 				CPUFREQ_ENTRY_INVALID) {
-			if (cpu_family == CPU_HW_PSTATE) {
-				printk(KERN_INFO PFX
-					"   %d : pstate %d (%d MHz)\n", j,
-					data->powernow_table[j].index,
-					data->powernow_table[j].frequency/1000);
-			} else {
 				printk(KERN_INFO PFX
 					"fid 0x%x (%d MHz), vid 0x%x\n",
 					data->powernow_table[j].index & 0xff,
 					data->powernow_table[j].frequency/1000,
 					data->powernow_table[j].index >> 8);
-			}
 		}
 	}
 	if (data->batps)
@@ -642,20 +590,6 @@ static void print_basics(struct powernow_k8_data *data)
 				data->batps);
 }
 
-static u32 freq_from_fid_did(u32 fid, u32 did)
-{
-	u32 mhz = 0;
-
-	if (boot_cpu_data.x86 == 0x10)
-		mhz = (100 * (fid + 0x10)) >> did;
-	else if (boot_cpu_data.x86 == 0x11)
-		mhz = (100 * (fid + 8)) >> did;
-	else
-		BUG();
-
-	return mhz * 1000;
-}
-
 static int fill_powernow_table(struct powernow_k8_data *data,
 		struct pst_s *pst, u8 maxvid)
 {
@@ -815,7 +749,7 @@ static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data,
 {
 	u64 control;
 
-	if (!data->acpi_data.state_count || (cpu_family == CPU_HW_PSTATE))
+	if (!data->acpi_data.state_count)
 		return;
 
 	control = data->acpi_data.states[index].control;
@@ -866,10 +800,7 @@ static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
 	data->numps = data->acpi_data.state_count;
 	powernow_k8_acpi_pst_values(data, 0);
 
-	if (cpu_family == CPU_HW_PSTATE)
-		ret_val = fill_powernow_table_pstate(data, powernow_table);
-	else
-		ret_val = fill_powernow_table_fidvid(data, powernow_table);
+	ret_val = fill_powernow_table_fidvid(data, powernow_table);
 	if (ret_val)
 		goto err_out_mem;
 
@@ -906,47 +837,6 @@ err_out:
 	return ret_val;
 }
 
-static int fill_powernow_table_pstate(struct powernow_k8_data *data,
-		struct cpufreq_frequency_table *powernow_table)
-{
-	int i;
-	u32 hi = 0, lo = 0;
-	rdmsr(MSR_PSTATE_CUR_LIMIT, lo, hi);
-	data->max_hw_pstate = (lo & HW_PSTATE_MAX_MASK) >> HW_PSTATE_MAX_SHIFT;
-
-	for (i = 0; i < data->acpi_data.state_count; i++) {
-		u32 index;
-
-		index = data->acpi_data.states[i].control & HW_PSTATE_MASK;
-		if (index > data->max_hw_pstate) {
-			printk(KERN_ERR PFX "invalid pstate %d - "
-					"bad value %d.\n", i, index);
-			printk(KERN_ERR PFX "Please report to BIOS "
-					"manufacturer\n");
-			invalidate_entry(powernow_table, i);
-			continue;
-		}
-		rdmsr(MSR_PSTATE_DEF_BASE + index, lo, hi);
-		if (!(hi & HW_PSTATE_VALID_MASK)) {
-			pr_debug("invalid pstate %d, ignoring\n", index);
-			invalidate_entry(powernow_table, i);
-			continue;
-		}
-
-		powernow_table[i].index = index;
-
-		/* Frequency may be rounded for these */
-		if ((boot_cpu_data.x86 == 0x10 && boot_cpu_data.x86_model < 10)
-				 || boot_cpu_data.x86 == 0x11) {
-			powernow_table[i].frequency =
-				freq_from_fid_did(lo & 0x3f, (lo >> 6) & 7);
-		} else
-			powernow_table[i].frequency =
-				data->acpi_data.states[i].core_frequency * 1000;
-	}
-	return 0;
-}
-
 static int fill_powernow_table_fidvid(struct powernow_k8_data *data,
 		struct cpufreq_frequency_table *powernow_table)
 {
@@ -1023,15 +913,8 @@ static int get_transition_latency(struct powernow_k8_data *data)
 			max_latency = cur_latency;
 	}
 	if (max_latency == 0) {
-		/*
-		 * Fam 11h and later may return 0 as transition latency. This
-		 * is intended and means "very fast". While cpufreq core and
-		 * governors currently can handle that gracefully, better set it
-		 * to 1 to avoid problems in the future.
-		 */
-		if (boot_cpu_data.x86 < 0x11)
-			printk(KERN_ERR FW_WARN PFX "Invalid zero transition "
-				"latency\n");
+		printk(KERN_ERR FW_WARN PFX "Invalid zero transition "
+		       "latency\n");
 		max_latency = 1;
 	}
 	/* value in usecs, needs to be in nanoseconds */
@@ -1091,40 +974,6 @@ static int transition_frequency_fidvid(struct powernow_k8_data *data,
 	return res;
 }
 
-/* Take a frequency, and issue the hardware pstate transition command */
-static int transition_frequency_pstate(struct powernow_k8_data *data,
-		unsigned int index)
-{
-	u32 pstate = 0;
-	int res, i;
-	struct cpufreq_freqs freqs;
-
-	pr_debug("cpu %d transition to index %u\n", smp_processor_id(), index);
-
-	/* get MSR index for hardware pstate transition */
-	pstate = index & HW_PSTATE_MASK;
-	if (pstate > data->max_hw_pstate)
-		return -EINVAL;
-
-	freqs.old = find_khz_freq_from_pstate(data->powernow_table,
-			data->currpstate);
-	freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
-
-	for_each_cpu(i, data->available_cores) {
-		freqs.cpu = i;
-		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
-	}
-
-	res = transition_pstate(data, pstate);
-	freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
-
-	for_each_cpu(i, data->available_cores) {
-		freqs.cpu = i;
-		cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
-	}
-	return res;
-}
-
 /* Driver entry point to switch to the target frequency */
 static int powernowk8_target(struct cpufreq_policy *pol,
 		unsigned targfreq, unsigned relation)
@@ -1166,18 +1015,16 @@ static int powernowk8_target(struct cpufreq_policy *pol,
 	if (query_current_values_with_pending_wait(data))
 		goto err_out;
 
-	if (cpu_family != CPU_HW_PSTATE) {
-		pr_debug("targ: curr fid 0x%x, vid 0x%x\n",
-		data->currfid, data->currvid);
+	pr_debug("targ: curr fid 0x%x, vid 0x%x\n",
+		 data->currfid, data->currvid);
 
-		if ((checkvid != data->currvid) ||
-		    (checkfid != data->currfid)) {
-			printk(KERN_INFO PFX
-				"error - out of sync, fix 0x%x 0x%x, "
-				"vid 0x%x 0x%x\n",
-				checkfid, data->currfid,
-				checkvid, data->currvid);
-		}
+	if ((checkvid != data->currvid) ||
+	    (checkfid != data->currfid)) {
+		printk(KERN_INFO PFX
+		       "error - out of sync, fix 0x%x 0x%x, "
+		       "vid 0x%x 0x%x\n",
+		       checkfid, data->currfid,
+		       checkvid, data->currvid);
 	}
 
 	if (cpufreq_frequency_table_target(pol, data->powernow_table,
@@ -1188,10 +1035,8 @@ static int powernowk8_target(struct cpufreq_policy *pol,
 
 	powernow_k8_acpi_pst_values(data, newstate);
 
-	if (cpu_family == CPU_HW_PSTATE)
-		ret = transition_frequency_pstate(data, newstate);
-	else
-		ret = transition_frequency_fidvid(data, newstate);
+	ret = transition_frequency_fidvid(data, newstate);
+
 	if (ret) {
 		printk(KERN_ERR PFX "transition frequency failed\n");
 		ret = 1;
@@ -1200,11 +1045,7 @@ static int powernowk8_target(struct cpufreq_policy *pol,
 	}
 	mutex_unlock(&fidvid_mutex);
 
-	if (cpu_family == CPU_HW_PSTATE)
-		pol->cur = find_khz_freq_from_pstate(data->powernow_table,
-				newstate);
-	else
-		pol->cur = find_khz_freq_from_fid(data->currfid);
+	pol->cur = find_khz_freq_from_fid(data->currfid);
 	ret = 0;
 
 err_out:
@@ -1244,8 +1085,7 @@ static void __cpuinit powernowk8_cpu_init_on_cpu(void *_init_on_cpu)
 		return;
 	}
 
-	if (cpu_family == CPU_OPTERON)
-		fidvid_msr_init();
+	fidvid_msr_init();
 
 	init_on_cpu->rc = 0;
 }
@@ -1259,7 +1099,6 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 	struct powernow_k8_data *data;
 	struct init_on_cpu init_on_cpu;
 	int rc;
-	struct cpuinfo_x86 *c = &cpu_data(pol->cpu);
 
 	if (!cpu_online(pol->cpu))
 		return -ENODEV;
@@ -1275,7 +1114,6 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 	}
 
 	data->cpu = pol->cpu;
-	data->currpstate = HW_PSTATE_INVALID;
 
 	if (powernow_k8_cpu_init_acpi(data)) {
 		/*
@@ -1312,17 +1150,10 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 	if (rc != 0)
 		goto err_out_exit_acpi;
 
-	if (cpu_family == CPU_HW_PSTATE)
-		cpumask_copy(pol->cpus, cpumask_of(pol->cpu));
-	else
-		cpumask_copy(pol->cpus, cpu_core_mask(pol->cpu));
+	cpumask_copy(pol->cpus, cpu_core_mask(pol->cpu));
 	data->available_cores = pol->cpus;
 
-	if (cpu_family == CPU_HW_PSTATE)
-		pol->cur = find_khz_freq_from_pstate(data->powernow_table,
-				data->currpstate);
-	else
-		pol->cur = find_khz_freq_from_fid(data->currfid);
+	pol->cur = find_khz_freq_from_fid(data->currfid);
 	pr_debug("policy current frequency %d kHz\n", pol->cur);
 
 	/* min/max the cpu is capable of */
@@ -1334,18 +1165,10 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 		return -EINVAL;
 	}
 
-	/* Check for APERF/MPERF support in hardware */
-	if (cpu_has(c, X86_FEATURE_APERFMPERF))
-		cpufreq_amd64_driver.getavg = cpufreq_get_measured_perf;
-
 	cpufreq_frequency_table_get_attr(data->powernow_table, pol->cpu);
 
-	if (cpu_family == CPU_HW_PSTATE)
-		pr_debug("cpu_init done, current pstate 0x%x\n",
-				data->currpstate);
-	else
-		pr_debug("cpu_init done, current fid 0x%x, vid 0x%x\n",
-			data->currfid, data->currvid);
+	pr_debug("cpu_init done, current fid 0x%x, vid 0x%x\n",
+		 data->currfid, data->currvid);
 
 	per_cpu(powernow_data, pol->cpu) = data;
 
@@ -1398,88 +1221,15 @@ static unsigned int powernowk8_get(unsigned int cpu)
 	if (err)
 		goto out;
 
-	if (cpu_family == CPU_HW_PSTATE)
-		khz = find_khz_freq_from_pstate(data->powernow_table,
-						data->currpstate);
-	else
-		khz = find_khz_freq_from_fid(data->currfid);
+	khz = find_khz_freq_from_fid(data->currfid);
 
 
 out:
 	return khz;
 }
 
-static void _cpb_toggle_msrs(bool t)
-{
-	int cpu;
-
-	get_online_cpus();
-
-	rdmsr_on_cpus(cpu_online_mask, MSR_K7_HWCR, msrs);
-
-	for_each_cpu(cpu, cpu_online_mask) {
-		struct msr *reg = per_cpu_ptr(msrs, cpu);
-		if (t)
-			reg->l &= ~BIT(25);
-		else
-			reg->l |= BIT(25);
-	}
-	wrmsr_on_cpus(cpu_online_mask, MSR_K7_HWCR, msrs);
-
-	put_online_cpus();
-}
-
-/*
- * Switch on/off core performance boosting.
- *
- * 0=disable
- * 1=enable.
- */
-static void cpb_toggle(bool t)
-{
-	if (!cpb_capable)
-		return;
-
-	if (t && !cpb_enabled) {
-		cpb_enabled = true;
-		_cpb_toggle_msrs(t);
-		printk(KERN_INFO PFX "Core Boosting enabled.\n");
-	} else if (!t && cpb_enabled) {
-		cpb_enabled = false;
-		_cpb_toggle_msrs(t);
-		printk(KERN_INFO PFX "Core Boosting disabled.\n");
-	}
-}
-
-static ssize_t store_cpb(struct cpufreq_policy *policy, const char *buf,
-				 size_t count)
-{
-	int ret = -EINVAL;
-	unsigned long val = 0;
-
-	ret = strict_strtoul(buf, 10, &val);
-	if (!ret && (val == 0 || val == 1) && cpb_capable)
-		cpb_toggle(val);
-	else
-		return -EINVAL;
-
-	return count;
-}
-
-static ssize_t show_cpb(struct cpufreq_policy *policy, char *buf)
-{
-	return sprintf(buf, "%u\n", cpb_enabled);
-}
-
-#define define_one_rw(_name) \
-static struct freq_attr _name = \
-__ATTR(_name, 0644, show_##_name, store_##_name)
-
-define_one_rw(cpb);
-
 static struct freq_attr *powernow_k8_attr[] = {
 	&cpufreq_freq_attr_scaling_available_freqs,
-	&cpb,
 	NULL,
 };
 
@@ -1495,51 +1245,10 @@ static struct cpufreq_driver cpufreq_amd64_driver = {
 	.attr		= powernow_k8_attr,
 };
 
-/*
- * Clear the boost-disable flag on the CPU_DOWN path so that this cpu
- * cannot block the remaining ones from boosting. On the CPU_UP path we
- * simply keep the boost-disable flag in sync with the current global
- * state.
- */
-static int cpb_notify(struct notifier_block *nb, unsigned long action,
-		      void *hcpu)
-{
-	unsigned cpu = (long)hcpu;
-	u32 lo, hi;
-
-	switch (action) {
-	case CPU_UP_PREPARE:
-	case CPU_UP_PREPARE_FROZEN:
-
-		if (!cpb_enabled) {
-			rdmsr_on_cpu(cpu, MSR_K7_HWCR, &lo, &hi);
-			lo |= BIT(25);
-			wrmsr_on_cpu(cpu, MSR_K7_HWCR, lo, hi);
-		}
-		break;
-
-	case CPU_DOWN_PREPARE:
-	case CPU_DOWN_PREPARE_FROZEN:
-		rdmsr_on_cpu(cpu, MSR_K7_HWCR, &lo, &hi);
-		lo &= ~BIT(25);
-		wrmsr_on_cpu(cpu, MSR_K7_HWCR, lo, hi);
-		break;
-
-	default:
-		break;
-	}
-
-	return NOTIFY_OK;
-}
-
-static struct notifier_block cpb_nb = {
-	.notifier_call		= cpb_notify,
-};
-
 /* driver entry point for init */
 static int __cpuinit powernowk8_init(void)
 {
-	unsigned int i, supported_cpus = 0, cpu;
+	unsigned int i, supported_cpus = 0;
 	int rv;
 
 	for_each_online_cpu(i) {
@@ -1559,35 +1268,8 @@ static int __cpuinit powernowk8_init(void)
 	printk(KERN_INFO PFX "Found %d %s (%d cpu cores) (" VERSION ")\n",
 		num_online_nodes(), boot_cpu_data.x86_model_id, supported_cpus);
 
-	if (boot_cpu_has(X86_FEATURE_CPB)) {
-
-		cpb_capable = true;
-
-		msrs = msrs_alloc();
-		if (!msrs) {
-			printk(KERN_ERR "%s: Error allocating msrs!\n", __func__);
-			return -ENOMEM;
-		}
-
-		register_cpu_notifier(&cpb_nb);
-
-		rdmsr_on_cpus(cpu_online_mask, MSR_K7_HWCR, msrs);
-
-		for_each_cpu(cpu, cpu_online_mask) {
-			struct msr *reg = per_cpu_ptr(msrs, cpu);
-			cpb_enabled |= !(!!(reg->l & BIT(25)));
-		}
-
-		printk(KERN_INFO PFX "Core Performance Boosting: %s.\n",
-			(cpb_enabled ? "on" : "off"));
-	}
-
 	rv = cpufreq_register_driver(&cpufreq_amd64_driver);
-	if (rv < 0 && boot_cpu_has(X86_FEATURE_CPB)) {
-		unregister_cpu_notifier(&cpb_nb);
-		msrs_free(msrs);
-		msrs = NULL;
-	}
+
 	return rv;
 }
 
@@ -1596,13 +1278,6 @@ static void __exit powernowk8_exit(void)
 {
 	pr_debug("exit\n");
 
-	if (boot_cpu_has(X86_FEATURE_CPB)) {
-		msrs_free(msrs);
-		msrs = NULL;
-
-		unregister_cpu_notifier(&cpb_nb);
-	}
-
 	cpufreq_unregister_driver(&cpufreq_amd64_driver);
 }
 
diff --git a/drivers/cpufreq/powernow-k8.h b/drivers/cpufreq/powernow-k8.h
index 3744d26..79329d4 100644
--- a/drivers/cpufreq/powernow-k8.h
+++ b/drivers/cpufreq/powernow-k8.h
@@ -5,24 +5,11 @@
  *  http://www.gnu.org/licenses/gpl.html
  */
 
-enum pstate {
-	HW_PSTATE_INVALID = 0xff,
-	HW_PSTATE_0 = 0,
-	HW_PSTATE_1 = 1,
-	HW_PSTATE_2 = 2,
-	HW_PSTATE_3 = 3,
-	HW_PSTATE_4 = 4,
-	HW_PSTATE_5 = 5,
-	HW_PSTATE_6 = 6,
-	HW_PSTATE_7 = 7,
-};
-
 struct powernow_k8_data {
 	unsigned int cpu;
 
 	u32 numps;  /* number of p-states */
 	u32 batps;  /* number of p-states supported on battery */
-	u32 max_hw_pstate; /* maximum legal hardware pstate */
 
 	/* these values are constant when the PSB is used to determine
 	 * vid/fid pairings, but are modified during the ->target() call
@@ -37,7 +24,6 @@ struct powernow_k8_data {
 	/* keep track of the current fid / vid or pstate */
 	u32 currvid;
 	u32 currfid;
-	enum pstate currpstate;
 
 	/* the powernow_table includes all frequency and vid/fid pairings:
 	 * fid are the lower 8 bits of the index, vid are the upper 8 bits.
@@ -97,23 +83,6 @@ struct powernow_k8_data {
 #define MSR_S_HI_CURRENT_VID      0x0000003f
 #define MSR_C_HI_STP_GNT_BENIGN	  0x00000001
 
-
-/* Hardware Pstate _PSS and MSR definitions */
-#define USE_HW_PSTATE		0x00000080
-#define HW_PSTATE_MASK 		0x00000007
-#define HW_PSTATE_VALID_MASK 	0x80000000
-#define HW_PSTATE_MAX_MASK	0x000000f0
-#define HW_PSTATE_MAX_SHIFT	4
-#define MSR_PSTATE_DEF_BASE 	0xc0010064 /* base of Pstate MSRs */
-#define MSR_PSTATE_STATUS 	0xc0010063 /* Pstate Status MSR */
-#define MSR_PSTATE_CTRL 	0xc0010062 /* Pstate control MSR */
-#define MSR_PSTATE_CUR_LIMIT	0xc0010061 /* pstate current limit MSR */
-
-/* define the two driver architectures */
-#define CPU_OPTERON 0
-#define CPU_HW_PSTATE 1
-
-
 /*
  * There are restrictions frequencies have to follow:
  * - only 1 entry in the low fid table ( <=1.4GHz )
@@ -218,5 +187,4 @@ static int core_frequency_transition(struct powernow_k8_data *data, u32 reqfid);
 
 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, unsigned int index);
 
-static int fill_powernow_table_pstate(struct powernow_k8_data *data, struct cpufreq_frequency_table *powernow_table);
 static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpufreq_frequency_table *powernow_table);
-- 
1.7.6

Re: [PATCH v4 6/7] cpufreq: Remove support for hardware P-state chips from powernow-k8

[Borislav Petkov]

On Mon, Jul 18, 2011 at 12:37:41PM -0400, Matthew Garrett wrote:

...

>         /* min/max the cpu is capable of */
> @@ -1334,18 +1165,10 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
>                 return -EINVAL;
>         }
> 
> -       /* Check for APERF/MPERF support in hardware */
> -       if (cpu_has(c, X86_FEATURE_APERFMPERF))
> -               cpufreq_amd64_driver.getavg = cpufreq_get_measured_perf;
> -

Btw, this makes cpufreq_get_measured_perf used only in acpi-cpufreq now
and <drivers/cpufreq/mperf.c> can be merged into acpi-cpufreq, thus
removing linking it into both drivers in <drivers/cpufreq/Makefile>.

Also, can you put the next set of patches in a git repo somewhere so
that I could pull them for testing instead of applying them one by one?

Thanks.

-- 
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551

[PATCH v4 7/7] cpufreq: Add boost alias to cpb

[Matthew Garrett]

cpb is AMD's name for the dynamic overclocking present in their chips.
Intel provide similar functionality, so add "boost" as an alias. In the
long run we can then transition to using that rather than cpb.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 Documentation/cpu-freq/user-guide.txt |    5 +++++
 drivers/cpufreq/acpi-cpufreq.c        |    8 +++-----
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/Documentation/cpu-freq/user-guide.txt b/Documentation/cpu-freq/user-guide.txt
index 04f6b32..a692d3f 100644
--- a/Documentation/cpu-freq/user-guide.txt
+++ b/Documentation/cpu-freq/user-guide.txt
@@ -214,6 +214,11 @@ bios_limit :			If the BIOS tells the OS to limit a CPU to
 				which can be detected through the generic
 				thermal driver.
 
+boost:				Some CPUs support clocking themselves to above
+				their rated frequency depending on temperature
+				and power constraints. This file allows this
+				to be enabled or disabled.
+
 If you have selected the "userspace" governor which allows you to
 set the CPU operating frequency to a specific value, you can read out
 the current frequency in
diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
index 0a92277..30c66cc 100644
--- a/drivers/cpufreq/acpi-cpufreq.c
+++ b/drivers/cpufreq/acpi-cpufreq.c
@@ -172,11 +172,8 @@ static ssize_t show_cpb(struct cpufreq_policy *policy, char *buf)
 	return sprintf(buf, "%u\n", cpb_state(cpu));
 }
 
-#define define_one_rw(_name) \
-static struct freq_attr _name = \
-	__ATTR(_name, 0644, show_##_name, store_##_name)
-
-define_one_rw(cpb);
+static struct freq_attr cpb = __ATTR(cpb, 0644, show_cpb, store_cpb);
+static struct freq_attr boost = __ATTR(boost, 0644, show_cpb, store_cpb);
 
 static int check_est_cpu(unsigned int cpuid)
 {
@@ -933,6 +930,7 @@ static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
 static struct freq_attr *acpi_cpufreq_attr[] = {
 	&cpufreq_freq_attr_scaling_available_freqs,
 	&cpb,
+	&boost,
 	NULL,
 };
 
-- 
1.7.6