[PATCH 0/5] Intel P states enhancements

[PATCH 0/5] Intel P states enhancements

[Srinivas Pandruvada]

This series enhances Intel P state drivers with the following features:
- When max_perf_pct is reduced in turbo range, we can change the turbo ratios
when platform allows. This is particularly useful in limiting performance with
HWP where whole range is turbo.
- Use Turbo Activation Ratio, when calculating max non turbo P state. This will
show now correct percentage in turbo range
- To calculate busy percent, the estimate is not correct when the max non turbo
is limited by tar, hence using physical max non turbo as before.
- Use ACPI _PSS and _PPC in intel_ptate driver.
- Avoid calculation for P state control value when cpufreq policy requests
frequency limits when matched in _PSS. Sometime calculations causes reduced
control value in boundary conditions.

Although they are independent patches, sending as series to help applying and
testing.

I appreciate review and testing on multiple platforms.

Srinivas Pandruvada (5):
  cpufreq: intel_p_state: Fix limiting turbo sub states
  cpufreq: intel_pstate: get P1 from TAR when available
  cpufreq: intel-pstate: Use separate max pstate for scaling
  cpufreq: intel_pstate: Use ACPI perf configuration
  cpufreq: intel_pstate: Avoid calculation for max/min

 arch/x86/include/asm/msr-index.h |   7 +
 drivers/cpufreq/Kconfig.x86      |   1 +
 drivers/cpufreq/intel_pstate.c   | 326 +++++++++++++++++++++++++++++++++++++--
 3 files changed, 324 insertions(+), 10 deletions(-)

-- 
1.9.3

[PATCH 1/5] cpufreq: intel_p_state: Fix limiting turbo sub states

[Srinivas Pandruvada]

Although the max_perf_pct reflects sub states in turbo range, we can't
really restrict to those states. This gives wrong impression that the
performance is reduced.
This can be achieved by restricting turbo ratio limits (MSR 0x1AD),
when bit 28 of platform info MSR allows (MSR 0xCE) is 1.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/cpufreq/intel_pstate.c | 93 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 92 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
index c204e5a..ff623a6 100644
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -80,6 +80,7 @@ struct pstate_data {
 	int	max_pstate;
 	int	scaling;
 	int	turbo_pstate;
+	u64	turbo_ratio_limit;
 };
 
 struct vid_data {
@@ -132,6 +133,8 @@ struct pstate_funcs {
 	int (*get_scaling)(void);
 	void (*set)(struct cpudata*, int pstate);
 	void (*get_vid)(struct cpudata *);
+	u64 (*get_turbo_ratio_limit)(struct cpudata *);
+	int (*set_turbo_ratio_limit)(struct cpudata *, u64, u64);
 };
 
 struct cpu_defaults {
@@ -425,6 +428,23 @@ static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 
+	if (pstate_funcs.set_turbo_ratio_limit) {
+		int max_perf_adj;
+		struct cpudata *cpu = all_cpu_data[0];
+
+		if (limits.max_sysfs_pct == 100)
+			max_perf_adj = cpu->pstate.turbo_ratio_limit;
+		else
+			max_perf_adj = fp_toint(mul_fp(int_tofp(
+					cpu->pstate.turbo_ratio_limit & 0xff),
+					limits.max_perf));
+
+		if (max_perf_adj > cpu->pstate.max_pstate)
+			pstate_funcs.set_turbo_ratio_limit(cpu,
+						cpu->pstate.turbo_ratio_limit,
+						max_perf_adj);
+	}
+
 	if (hwp_active)
 		intel_pstate_hwp_set();
 	return count;
@@ -617,6 +637,55 @@ static void core_set_pstate(struct cpudata *cpudata, int pstate)
 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
 }
 
+static u64 core_get_turbo_ratio_limit(struct cpudata *cpudata)
+{
+	u64 value;
+
+	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
+
+	return value;
+}
+
+static int core_set_turbo_ratio_limit(struct cpudata *cpudata, u64 def_ratio,
+			       u64 new_ratio)
+{
+	u64 value;
+
+	rdmsrl(MSR_PLATFORM_INFO, value);
+	if (value & BIT(28)) {
+		u64 ratio = 0;
+		u64 out_ratio = 0;
+		u8 max_ratio = new_ratio & 0xff;
+		int i;
+		/*
+		 * If caller provided reduced max ratio (one core active)
+		 * then use this for all other ratios, which are more
+		 * than the default ratio for those many cores active
+		 * for example if default ratio is 0x1a1b1c1d and new ratio
+		 * is 0x1b, then resultant ratio will be 0x1a1b1b1b
+		 */
+		for (i = 0; i < sizeof(def_ratio); ++i) {
+			if (def_ratio & 0xff) {
+				if (new_ratio & 0xff)
+					ratio = new_ratio & 0xff;
+				else {
+					if ((def_ratio & 0xff) > max_ratio)
+						ratio = max_ratio;
+					else
+						ratio = def_ratio & 0xff;
+				}
+				out_ratio |= (ratio << (i * 8));
+			}
+			def_ratio >>= 8;
+			new_ratio >>= 8;
+		}
+		wrmsrl(MSR_NHM_TURBO_RATIO_LIMIT, out_ratio);
+		return 0;
+	}
+
+	return -EPERM;
+}
+
 static int knl_get_turbo_pstate(void)
 {
 	u64 value;
@@ -645,6 +714,8 @@ static struct cpu_defaults core_params = {
 		.get_turbo = core_get_turbo_pstate,
 		.get_scaling = core_get_scaling,
 		.set = core_set_pstate,
+		.get_turbo_ratio_limit = core_get_turbo_ratio_limit,
+		.set_turbo_ratio_limit = core_set_turbo_ratio_limit,
 	},
 };
 
@@ -733,7 +804,10 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 	cpu->pstate.max_pstate = pstate_funcs.get_max();
 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 	cpu->pstate.scaling = pstate_funcs.get_scaling();
-
+	if (pstate_funcs.get_turbo_ratio_limit &&
+	    !cpu->pstate.turbo_ratio_limit)
+		cpu->pstate.turbo_ratio_limit =
+			pstate_funcs.get_turbo_ratio_limit(cpu);
 	if (pstate_funcs.get_vid)
 		pstate_funcs.get_vid(cpu);
 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate, false);
@@ -939,6 +1013,21 @@ static int intel_pstate_init_cpu(unsigned int cpunum)
 
 	intel_pstate_get_cpu_pstates(cpu);
 
+	/* readjust turbo limit ratio after resume or hotplug */
+	if (limits.max_sysfs_pct != 100 &&
+	    pstate_funcs.set_turbo_ratio_limit) {
+		int max_perf_adj;
+
+		max_perf_adj = fp_toint(mul_fp(int_tofp(
+					cpu->pstate.turbo_ratio_limit & 0xff),
+					limits.max_perf));
+
+		if (max_perf_adj > cpu->pstate.max_pstate)
+			pstate_funcs.set_turbo_ratio_limit(cpu,
+					cpu->pstate.turbo_ratio_limit,
+					max_perf_adj);
+	}
+
 	init_timer_deferrable(&cpu->timer);
 	cpu->timer.data = (unsigned long)cpu;
 	cpu->timer.expires = jiffies + HZ/100;
@@ -1100,6 +1189,8 @@ static void copy_cpu_funcs(struct pstate_funcs *funcs)
 	pstate_funcs.get_scaling = funcs->get_scaling;
 	pstate_funcs.set       = funcs->set;
 	pstate_funcs.get_vid   = funcs->get_vid;
+	pstate_funcs.set_turbo_ratio_limit = funcs->set_turbo_ratio_limit;
+	pstate_funcs.get_turbo_ratio_limit = funcs->get_turbo_ratio_limit;
 }
 
 #if IS_ENABLED(CONFIG_ACPI)
-- 
1.9.3

[PATCH 2/5] cpufreq: intel_pstate: get P1 from TAR when available

[Srinivas Pandruvada]

After Ivybridge, the max non turbo ratio obtained from platform info msr
is not always guaranteed P1 on client platforms. The max non turbo
activation ratio (TAR), determines the max for the current level of TDP.
The ratio in platform info is physical max. The TAR MSR can be locked,
so updating this value is not possible on all platforms.
This change gets this ratio from MSR TURBO_ACTIVATION_RATIO if
available,
but also do some sanity checking to make sure that this value is
correct.
The sanity check involves reading the TDP ratio for the current tdp
control value when platform has configurable TDP present and matching
TAC
with this.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 arch/x86/include/asm/msr-index.h |  7 +++++++
 drivers/cpufreq/intel_pstate.c   | 39 +++++++++++++++++++++++++++++++++++----
 2 files changed, 42 insertions(+), 4 deletions(-)

diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index f2f2cd6..e8e6fac 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -184,6 +184,13 @@
 #define MSR_GFX_PERF_LIMIT_REASONS	0x000006B0
 #define MSR_RING_PERF_LIMIT_REASONS	0x000006B1
 
+/* Config TDP MSRs */
+#define MSR_CONFIG_TDP_NOMINAL		0x00000648
+#define MSR_CONFIG_TDP_LEVEL1		0x00000649
+#define MSR_CONFIG_TDP_LEVEL2		0x0000064A
+#define MSR_CONFIG_TDP_CONTROL		0x0000064B
+#define MSR_TURBO_ACTIVATION_RATIO	0x0000064C
+
 /* Hardware P state interface */
 #define MSR_PPERF			0x0000064e
 #define MSR_PERF_LIMIT_REASONS		0x0000064f
diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
index ff623a6..2641102 100644
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -43,7 +43,6 @@
 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
 #define fp_toint(X) ((X) >> FRAC_BITS)
 
-
 static inline int32_t mul_fp(int32_t x, int32_t y)
 {
 	return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
@@ -602,10 +601,42 @@ static int core_get_min_pstate(void)
 
 static int core_get_max_pstate(void)
 {
-	u64 value;
+	u64 tar;
+	u64 plat_info;
+	int max_pstate;
+	int err;
+
+	rdmsrl(MSR_PLATFORM_INFO, plat_info);
+	max_pstate = (plat_info >> 8) & 0xFF;
+
+	err = rdmsrl_safe(MSR_TURBO_ACTIVATION_RATIO, &tar);
+	if (!err) {
+		/* Do some sanity checking for safety */
+		if (plat_info & 0x600000000) {
+			u64 tdp_ctrl;
+			u64 tdp_ratio;
+			int tdp_msr;
+
+			err = rdmsrl_safe(MSR_CONFIG_TDP_CONTROL, &tdp_ctrl);
+			if (err)
+				goto skip_tar;
+
+			tdp_msr = MSR_CONFIG_TDP_NOMINAL + tdp_ctrl;
+			err = rdmsrl_safe(tdp_msr, &tdp_ratio);
+			if (err)
+				goto skip_tar;
+
+			if (tdp_ratio - 1 == tar) {
+				max_pstate = tar;
+				pr_debug("max_pstate=TAC %x\n", max_pstate);
+			} else {
+				goto skip_tar;
+			}
+		}
+	}
 
-	rdmsrl(MSR_PLATFORM_INFO, value);
-	return (value >> 8) & 0xFF;
+skip_tar:
+	return max_pstate;
 }
 
 static int core_get_turbo_pstate(void)
-- 
1.9.3

[PATCH 3/5] cpufreq: intel-pstate: Use separate max pstate for scaling

[Srinivas Pandruvada]

Systems with configurable TDP have multiple max non turbo p state. Intel
P state uses max non turbo P state for scaling. But using the real max
non turbo p state causes underestimation of next P state. So using
the physical max non turbo P state as before for scaling.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/cpufreq/intel_pstate.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
index 2641102..a349462 100644
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -77,6 +77,7 @@ struct pstate_data {
 	int	current_pstate;
 	int	min_pstate;
 	int	max_pstate;
+	int	max_pstate_physical;
 	int	scaling;
 	int	turbo_pstate;
 	u64	turbo_ratio_limit;
@@ -127,6 +128,7 @@ struct pstate_adjust_policy {
 
 struct pstate_funcs {
 	int (*get_max)(void);
+	int (*get_max_physical)(void);
 	int (*get_min)(void);
 	int (*get_turbo)(void);
 	int (*get_scaling)(void);
@@ -599,6 +601,14 @@ static int core_get_min_pstate(void)
 	return (value >> 40) & 0xFF;
 }
 
+static int core_get_max_pstate_physical(void)
+{
+	u64 value;
+
+	rdmsrl(MSR_PLATFORM_INFO, value);
+	return (value >> 8) & 0xFF;
+}
+
 static int core_get_max_pstate(void)
 {
 	u64 tar;
@@ -741,6 +751,7 @@ static struct cpu_defaults core_params = {
 	},
 	.funcs = {
 		.get_max = core_get_max_pstate,
+		.get_max_physical = core_get_max_pstate_physical,
 		.get_min = core_get_min_pstate,
 		.get_turbo = core_get_turbo_pstate,
 		.get_scaling = core_get_scaling,
@@ -761,6 +772,7 @@ static struct cpu_defaults byt_params = {
 	},
 	.funcs = {
 		.get_max = byt_get_max_pstate,
+		.get_max_physical = byt_get_max_pstate,
 		.get_min = byt_get_min_pstate,
 		.get_turbo = byt_get_turbo_pstate,
 		.set = byt_set_pstate,
@@ -780,6 +792,7 @@ static struct cpu_defaults knl_params = {
 	},
 	.funcs = {
 		.get_max = core_get_max_pstate,
+		.get_max_physical = core_get_max_pstate_physical,
 		.get_min = core_get_min_pstate,
 		.get_turbo = knl_get_turbo_pstate,
 		.set = core_set_pstate,
@@ -833,6 +846,7 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 {
 	cpu->pstate.min_pstate = pstate_funcs.get_min();
 	cpu->pstate.max_pstate = pstate_funcs.get_max();
+	cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical();
 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 	cpu->pstate.scaling = pstate_funcs.get_scaling();
 	if (pstate_funcs.get_turbo_ratio_limit &&
@@ -854,7 +868,8 @@ static inline void intel_pstate_calc_busy(struct cpudata *cpu)
 
 	sample->freq = fp_toint(
 		mul_fp(int_tofp(
-			cpu->pstate.max_pstate * cpu->pstate.scaling / 100),
+			cpu->pstate.max_pstate_physical *
+			cpu->pstate.scaling / 100),
 			core_pct));
 
 	sample->core_pct_busy = (int32_t)core_pct;
@@ -922,7 +937,7 @@ static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 	 * specified pstate.
 	 */
 	core_busy = cpu->sample.core_pct_busy;
-	max_pstate = int_tofp(cpu->pstate.max_pstate);
+	max_pstate = int_tofp(cpu->pstate.max_pstate_physical);
 	current_pstate = int_tofp(cpu->pstate.current_pstate);
 	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 
@@ -1215,6 +1230,7 @@ static void copy_pid_params(struct pstate_adjust_policy *policy)
 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 {
 	pstate_funcs.get_max   = funcs->get_max;
+	pstate_funcs.get_max_physical = funcs->get_max_physical;
 	pstate_funcs.get_min   = funcs->get_min;
 	pstate_funcs.get_turbo = funcs->get_turbo;
 	pstate_funcs.get_scaling = funcs->get_scaling;
-- 
1.9.3

[PATCH 4/5] cpufreq: intel_pstate: Use ACPI perf configuration

[Srinivas Pandruvada]

Use ACPI _PSS to limit the Intel P State turbo, max and min ratios.
This driver uses acpi processor perf lib calls to register performance.
The following logic is used to adjust Intel P state driver limits:
- If there is no turbo entry in _PSS, then disable Intel P state turbo
and limit to non turbo max
- If the non turbo max ratio is more than _PSS max non turbo value, then
set the max non turbo ratio to _PSS non turbo max
- If the min ratio is less than _PSS min then change the min ratio
matching _PSS min
- Scale the _PSS turbo frequency to max turbo frequency based on control
value.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/cpufreq/Kconfig.x86    |   1 +
 drivers/cpufreq/intel_pstate.c | 130 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 131 insertions(+)

diff --git a/drivers/cpufreq/Kconfig.x86 b/drivers/cpufreq/Kconfig.x86
index c59bdcb..1c6affd 100644
--- a/drivers/cpufreq/Kconfig.x86
+++ b/drivers/cpufreq/Kconfig.x86
@@ -5,6 +5,7 @@
 config X86_INTEL_PSTATE
        bool "Intel P state control"
        depends on X86
+       select ACPI_PROCESSOR
        help
           This driver provides a P state for Intel core processors.
 	  The driver implements an internal governor and will become
diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
index a349462..6eafb55 100644
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -34,6 +34,8 @@
 #include <asm/cpu_device_id.h>
 #include <asm/cpufeature.h>
 
+#include <acpi/processor.h>
+
 #define BYT_RATIOS		0x66a
 #define BYT_VIDS		0x66b
 #define BYT_TURBO_RATIOS	0x66c
@@ -114,6 +116,9 @@ struct cpudata {
 	u64	prev_mperf;
 	u64	prev_tsc;
 	struct sample sample;
+#if IS_ENABLED(CONFIG_ACPI)
+	struct acpi_processor_performance acpi_perf_data;
+#endif
 };
 
 static struct cpudata **all_cpu_data;
@@ -146,6 +151,7 @@ struct cpu_defaults {
 static struct pstate_adjust_policy pid_params;
 static struct pstate_funcs pstate_funcs;
 static int hwp_active;
+static int no_acpi_perf;
 
 struct perf_limits {
 	int no_turbo;
@@ -173,6 +179,116 @@ static struct perf_limits limits = {
 	.min_sysfs_pct = 0,
 };
 
+#if IS_ENABLED(CONFIG_ACPI)
+static int intel_pstate_init_perf_limits(struct cpufreq_policy *policy)
+{
+	struct cpudata *cpu;
+	int ret;
+	int min_perf;
+	bool turbo_absent = false;
+	int i;
+
+	cpu = all_cpu_data[policy->cpu];
+
+	if (!cpu->acpi_perf_data.shared_cpu_map &&
+	    zalloc_cpumask_var_node(&cpu->acpi_perf_data.shared_cpu_map,
+				    GFP_KERNEL, cpu_to_node(policy->cpu))) {
+		return -ENOMEM;
+	}
+
+	ret = acpi_processor_register_performance(&cpu->acpi_perf_data,
+						  policy->cpu);
+	if (ret)
+		return ret;
+
+	if (cpu->acpi_perf_data.control_register.space_id !=
+						ACPI_ADR_SPACE_FIXED_HARDWARE)
+		return -EIO;
+
+	pr_debug("CPU%u - ACPI _PSS perf data\n", policy->cpu);
+	for (i = 0; i < cpu->acpi_perf_data.state_count; i++)
+		pr_debug("     %cP%d: %u MHz, %u mW, 0x%x\n",
+			 (i == cpu->acpi_perf_data.state ? '*' : ' '), i,
+			 (u32) cpu->acpi_perf_data.states[i].core_frequency,
+			 (u32) cpu->acpi_perf_data.states[i].power,
+			 (u32) cpu->acpi_perf_data.states[i].control);
+
+	if (cpu->acpi_perf_data.state_count < 2)
+		return 0;
+
+	/* Check if there is a turbo freq in _PSS */
+	if ((cpu->acpi_perf_data.states[0].core_frequency % 2) == 0) {
+		pr_debug("no turbo range exists in _PSS\n");
+		limits.no_turbo = limits.turbo_disabled = 1;
+		cpu->pstate.turbo_pstate = cpu->pstate.max_pstate;
+		turbo_absent = true;
+	}
+
+	/* Check if the max non turbo p state < Intel P state max */
+	for (i = 0; i < cpu->acpi_perf_data.state_count; i++) {
+		if ((cpu->acpi_perf_data.states[i].core_frequency % 2) == 0) {
+			int perf_max_ctl =
+				cpu->acpi_perf_data.states[i].control >> 8;
+			if (cpu->pstate.max_pstate > perf_max_ctl)
+				cpu->pstate.max_pstate = perf_max_ctl;
+			break;
+		}
+	}
+
+	/* check If min perf > Intel P State min */
+	min_perf = cpu->acpi_perf_data.states[
+			cpu->acpi_perf_data.state_count - 1].control >> 8;
+	if (min_perf > cpu->pstate.min_pstate) {
+		cpu->pstate.min_pstate = min_perf;
+		policy->cpuinfo.min_freq = cpu->pstate.min_pstate *
+						cpu->pstate.scaling;
+	}
+
+	if (turbo_absent)
+		policy->cpuinfo.max_freq =
+			cpu->pstate.max_pstate * cpu->pstate.scaling;
+	else {
+		policy->cpuinfo.max_freq =
+			cpu->pstate.turbo_pstate * cpu->pstate.scaling;
+		/* scale freq to intel pstate turbo scale */
+		cpu->acpi_perf_data.states[0].core_frequency =
+				cpu->pstate.scaling *
+				(cpu->acpi_perf_data.states[0].control >> 8);
+	}
+
+	pr_debug("Updated limits 0x%x 0x%x 0x %x\n", cpu->pstate.min_pstate,
+		 cpu->pstate.max_pstate, cpu->pstate.turbo_pstate);
+	pr_debug("policy max_freq=%d Khz min_freq = %d KHz\n",
+		 policy->cpuinfo.max_freq, policy->cpuinfo.min_freq);
+
+	return 0;
+}
+
+static int intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
+{
+	struct cpudata *cpu;
+
+	if (!no_acpi_perf)
+		return 0;
+
+	cpu = all_cpu_data[policy->cpu];
+	acpi_processor_unregister_performance(&cpu->acpi_perf_data,
+					      policy->cpu);
+	return 0;
+}
+
+#else
+static int intel_pstate_init_perf_limits(struct cpufreq_policy *policy)
+{
+	return 0;
+}
+
+static int intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
+{
+	return 0;
+}
+#endif
+
 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 			     int deadband, int integral) {
 	pid->setpoint = setpoint;
@@ -1186,18 +1302,29 @@ static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
 	policy->cpuinfo.max_freq =
 		cpu->pstate.turbo_pstate * cpu->pstate.scaling;
+	if (!no_acpi_perf)
+		intel_pstate_init_perf_limits(policy);
+	/* If there is no acpi perf data or error, we ignore and use Intel P
+	 * state calculated limits
+	 */
 	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 	cpumask_set_cpu(policy->cpu, policy->cpus);
 
 	return 0;
 }
 
+static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
+{
+	return intel_pstate_exit_perf_limits(policy);
+}
+
 static struct cpufreq_driver intel_pstate_driver = {
 	.flags		= CPUFREQ_CONST_LOOPS,
 	.verify		= intel_pstate_verify_policy,
 	.setpolicy	= intel_pstate_set_policy,
 	.get		= intel_pstate_get,
 	.init		= intel_pstate_cpu_init,
+	.exit		= intel_pstate_cpu_exit,
 	.stop_cpu	= intel_pstate_stop_cpu,
 	.name		= "intel_pstate",
 };
@@ -1429,6 +1556,9 @@ static int __init intel_pstate_setup(char *str)
 		force_load = 1;
 	if (!strcmp(str, "hwp_only"))
 		hwp_only = 1;
+	if (!strcmp(str, "no_acpi_perf"))
+		no_acpi_perf = 1;
+
 	return 0;
 }
 early_param("intel_pstate", intel_pstate_setup);
-- 
1.9.3

[PATCH 5/5] cpufreq: intel_pstate: Avoid calculation for max/min

[Srinivas Pandruvada]

When requested from cpufreq to set policy, look into _pss and get
control values, instead of using max/min perf calculations. These
calculation misses next control state in boundary conditions.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
 drivers/cpufreq/intel_pstate.c | 48 +++++++++++++++++++++++++++++++++++++-----
 1 file changed, 43 insertions(+), 5 deletions(-)

diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
index 6eafb55..ead751e 100644
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -164,6 +164,8 @@ struct perf_limits {
 	int max_sysfs_pct;
 	int min_policy_pct;
 	int min_sysfs_pct;
+	int max_perf_ctl;
+	int min_perf_ctl;
 };
 
 static struct perf_limits limits = {
@@ -177,6 +179,8 @@ static struct perf_limits limits = {
 	.max_sysfs_pct = 100,
 	.min_policy_pct = 0,
 	.min_sysfs_pct = 0,
+	.max_perf_ctl = 0,
+	.min_perf_ctl = 0,
 };
 
 #if IS_ENABLED(CONFIG_ACPI)
@@ -929,12 +933,23 @@ static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 	 * policy, or by cpu specific default values determined through
 	 * experimentation.
 	 */
-	max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
-	*max = clamp_t(int, max_perf_adj,
-			cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
+	if (limits.max_perf_ctl && limits.max_sysfs_pct >=
+						limits.max_policy_pct) {
+		*max = limits.max_perf_ctl;
+	} else {
+		max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf),
+					limits.max_perf));
+		*max = clamp_t(int, max_perf_adj, cpu->pstate.min_pstate,
+			       cpu->pstate.turbo_pstate);
+	}
 
-	min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
-	*min = clamp_t(int, min_perf, cpu->pstate.min_pstate, max_perf);
+	if (limits.min_perf_ctl) {
+		*min = limits.min_perf_ctl;
+	} else {
+		min_perf = fp_toint(mul_fp(int_tofp(max_perf),
+				    limits.min_perf));
+		*min = clamp_t(int, min_perf, cpu->pstate.min_pstate, max_perf);
+	}
 }
 
 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate, bool force)
@@ -1223,6 +1238,11 @@ static unsigned int intel_pstate_get(unsigned int cpu_num)
 
 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 {
+	struct cpudata *cpu;
+	int i;
+
+	pr_debug("%s max %u policy->max %u\n", __func__,
+		 policy->cpuinfo.max_freq, policy->max);
 	if (!policy->cpuinfo.max_freq)
 		return -ENODEV;
 
@@ -1235,6 +1255,8 @@ static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 		limits.max_perf_pct = 100;
 		limits.max_perf = int_tofp(1);
 		limits.no_turbo = 0;
+		limits.max_perf_ctl = 0;
+		limits.min_perf_ctl = 0;
 		return 0;
 	}
 
@@ -1248,6 +1270,22 @@ static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 
+	cpu = all_cpu_data[policy->cpu];
+	for (i = 0; i < cpu->acpi_perf_data.state_count; i++) {
+		if ((cpu->acpi_perf_data.states[i].control >> 8) *
+		    cpu->pstate.scaling == policy->max)
+			limits.max_perf_ctl =
+				cpu->acpi_perf_data.states[i].control >> 8;
+		if ((cpu->acpi_perf_data.states[i].control >> 8) *
+		    cpu->pstate.scaling == policy->min)
+			limits.min_perf_ctl =
+				cpu->acpi_perf_data.states[i].control >> 8;
+	}
+
+	pr_debug("max %u policy->max %u max_perf_ctl 0x%x min_perf_ctl 0x%x\n",
+		 policy->cpuinfo.max_freq, policy->max, limits.max_perf_ctl,
+		 limits.min_perf_ctl);
+
 	if (hwp_active)
 		intel_pstate_hwp_set();
 
-- 
1.9.3

Re: [PATCH 3/5] cpufreq: intel-pstate: Use separate max pstate for scaling

[Rafael J. Wysocki]

On Wednesday, August 26, 2015 01:32:39 PM Srinivas Pandruvada wrote:
> Systems with configurable TDP have multiple max non turbo p state. Intel
> P state uses max non turbo P state for scaling. But using the real max
> non turbo p state causes underestimation of next P state. So using
> the physical max non turbo P state as before for scaling.

What does "physical" mean here?

> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> ---
>  drivers/cpufreq/intel_pstate.c | 20 ++++++++++++++++++--
>  1 file changed, 18 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
> index 2641102..a349462 100644
> --- a/drivers/cpufreq/intel_pstate.c
> +++ b/drivers/cpufreq/intel_pstate.c
> @@ -77,6 +77,7 @@ struct pstate_data {
>  	int	current_pstate;
>  	int	min_pstate;
>  	int	max_pstate;
> +	int	max_pstate_physical;
>  	int	scaling;
>  	int	turbo_pstate;
>  	u64	turbo_ratio_limit;
> @@ -127,6 +128,7 @@ struct pstate_adjust_policy {
>  
>  struct pstate_funcs {
>  	int (*get_max)(void);
> +	int (*get_max_physical)(void);
>  	int (*get_min)(void);
>  	int (*get_turbo)(void);
>  	int (*get_scaling)(void);
> @@ -599,6 +601,14 @@ static int core_get_min_pstate(void)
>  	return (value >> 40) & 0xFF;
>  }
>  
> +static int core_get_max_pstate_physical(void)
> +{
> +	u64 value;
> +
> +	rdmsrl(MSR_PLATFORM_INFO, value);
> +	return (value >> 8) & 0xFF;
> +}
> +
>  static int core_get_max_pstate(void)
>  {
>  	u64 tar;
> @@ -741,6 +751,7 @@ static struct cpu_defaults core_params = {
>  	},
>  	.funcs = {
>  		.get_max = core_get_max_pstate,
> +		.get_max_physical = core_get_max_pstate_physical,
>  		.get_min = core_get_min_pstate,
>  		.get_turbo = core_get_turbo_pstate,
>  		.get_scaling = core_get_scaling,
> @@ -761,6 +772,7 @@ static struct cpu_defaults byt_params = {
>  	},
>  	.funcs = {
>  		.get_max = byt_get_max_pstate,
> +		.get_max_physical = byt_get_max_pstate,
>  		.get_min = byt_get_min_pstate,
>  		.get_turbo = byt_get_turbo_pstate,
>  		.set = byt_set_pstate,
> @@ -780,6 +792,7 @@ static struct cpu_defaults knl_params = {
>  	},
>  	.funcs = {
>  		.get_max = core_get_max_pstate,
> +		.get_max_physical = core_get_max_pstate_physical,
>  		.get_min = core_get_min_pstate,
>  		.get_turbo = knl_get_turbo_pstate,
>  		.set = core_set_pstate,
> @@ -833,6 +846,7 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
>  {
>  	cpu->pstate.min_pstate = pstate_funcs.get_min();
>  	cpu->pstate.max_pstate = pstate_funcs.get_max();
> +	cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical();
>  	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
>  	cpu->pstate.scaling = pstate_funcs.get_scaling();
>  	if (pstate_funcs.get_turbo_ratio_limit &&
> @@ -854,7 +868,8 @@ static inline void intel_pstate_calc_busy(struct cpudata *cpu)
>  
>  	sample->freq = fp_toint(
>  		mul_fp(int_tofp(
> -			cpu->pstate.max_pstate * cpu->pstate.scaling / 100),
> +			cpu->pstate.max_pstate_physical *
> +			cpu->pstate.scaling / 100),
>  			core_pct));
>  
>  	sample->core_pct_busy = (int32_t)core_pct;
> @@ -922,7 +937,7 @@ static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
>  	 * specified pstate.
>  	 */
>  	core_busy = cpu->sample.core_pct_busy;
> -	max_pstate = int_tofp(cpu->pstate.max_pstate);
> +	max_pstate = int_tofp(cpu->pstate.max_pstate_physical);
>  	current_pstate = int_tofp(cpu->pstate.current_pstate);
>  	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
>  
> @@ -1215,6 +1230,7 @@ static void copy_pid_params(struct pstate_adjust_policy *policy)
>  static void copy_cpu_funcs(struct pstate_funcs *funcs)
>  {
>  	pstate_funcs.get_max   = funcs->get_max;
> +	pstate_funcs.get_max_physical = funcs->get_max_physical;
>  	pstate_funcs.get_min   = funcs->get_min;
>  	pstate_funcs.get_turbo = funcs->get_turbo;
>  	pstate_funcs.get_scaling = funcs->get_scaling;
> 

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

Re: [PATCH 4/5] cpufreq: intel_pstate: Use ACPI perf configuration

[Rafael J. Wysocki]

On Wednesday, August 26, 2015 01:32:40 PM Srinivas Pandruvada wrote:
> Use ACPI _PSS to limit the Intel P State turbo, max and min ratios.
> This driver uses acpi processor perf lib calls to register performance.
> The following logic is used to adjust Intel P state driver limits:
> - If there is no turbo entry in _PSS, then disable Intel P state turbo
> and limit to non turbo max
> - If the non turbo max ratio is more than _PSS max non turbo value, then
> set the max non turbo ratio to _PSS non turbo max
> - If the min ratio is less than _PSS min then change the min ratio
> matching _PSS min
> - Scale the _PSS turbo frequency to max turbo frequency based on control
> value.

The new command line switch should be mentioned in the changelog too.

> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> ---
>  drivers/cpufreq/Kconfig.x86    |   1 +
>  drivers/cpufreq/intel_pstate.c | 130 +++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 131 insertions(+)
> 
> diff --git a/drivers/cpufreq/Kconfig.x86 b/drivers/cpufreq/Kconfig.x86
> index c59bdcb..1c6affd 100644
> --- a/drivers/cpufreq/Kconfig.x86
> +++ b/drivers/cpufreq/Kconfig.x86
> @@ -5,6 +5,7 @@
>  config X86_INTEL_PSTATE
>         bool "Intel P state control"
>         depends on X86
> +       select ACPI_PROCESSOR

	select ACPI_PROCESSOR if ACPI

or there will be a broken dependency if ACPI is not selected.

>         help
>            This driver provides a P state for Intel core processors.
>  	  The driver implements an internal governor and will become
> diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
> index a349462..6eafb55 100644
> --- a/drivers/cpufreq/intel_pstate.c
> +++ b/drivers/cpufreq/intel_pstate.c
> @@ -34,6 +34,8 @@
>  #include <asm/cpu_device_id.h>
>  #include <asm/cpufeature.h>
>  
> +#include <acpi/processor.h>
> +
>  #define BYT_RATIOS		0x66a
>  #define BYT_VIDS		0x66b
>  #define BYT_TURBO_RATIOS	0x66c
> @@ -114,6 +116,9 @@ struct cpudata {
>  	u64	prev_mperf;
>  	u64	prev_tsc;
>  	struct sample sample;
> +#if IS_ENABLED(CONFIG_ACPI)
> +	struct acpi_processor_performance acpi_perf_data;
> +#endif
>  };
>  
>  static struct cpudata **all_cpu_data;
> @@ -146,6 +151,7 @@ struct cpu_defaults {
>  static struct pstate_adjust_policy pid_params;
>  static struct pstate_funcs pstate_funcs;
>  static int hwp_active;
> +static int no_acpi_perf;

Why don't you call it "no_acpi"?  The "perf" part seems to be redundant.

>  
>  struct perf_limits {
>  	int no_turbo;
> @@ -173,6 +179,116 @@ static struct perf_limits limits = {
>  	.min_sysfs_pct = 0,
>  };
>  
> +#if IS_ENABLED(CONFIG_ACPI)
> +static int intel_pstate_init_perf_limits(struct cpufreq_policy *policy)
> +{
> +	struct cpudata *cpu;
> +	int ret;
> +	int min_perf;
> +	bool turbo_absent = false;
> +	int i;
> +
> +	cpu = all_cpu_data[policy->cpu];
> +
> +	if (!cpu->acpi_perf_data.shared_cpu_map &&
> +	    zalloc_cpumask_var_node(&cpu->acpi_perf_data.shared_cpu_map,
> +				    GFP_KERNEL, cpu_to_node(policy->cpu))) {
> +		return -ENOMEM;
> +	}
> +
> +	ret = acpi_processor_register_performance(&cpu->acpi_perf_data,
> +						  policy->cpu);
> +	if (ret)
> +		return ret;
> +
> +	if (cpu->acpi_perf_data.control_register.space_id !=
> +						ACPI_ADR_SPACE_FIXED_HARDWARE)
> +		return -EIO;
> +
> +	pr_debug("CPU%u - ACPI _PSS perf data\n", policy->cpu);
> +	for (i = 0; i < cpu->acpi_perf_data.state_count; i++)
> +		pr_debug("     %cP%d: %u MHz, %u mW, 0x%x\n",
> +			 (i == cpu->acpi_perf_data.state ? '*' : ' '), i,
> +			 (u32) cpu->acpi_perf_data.states[i].core_frequency,
> +			 (u32) cpu->acpi_perf_data.states[i].power,
> +			 (u32) cpu->acpi_perf_data.states[i].control);
> +
> +	if (cpu->acpi_perf_data.state_count < 2)
> +		return 0;

So what happens if we do that?  We'll not use ACPI limits, right?

> +
> +	/* Check if there is a turbo freq in _PSS */
> +	if ((cpu->acpi_perf_data.states[0].core_frequency % 2) == 0) {

Where does this check come from?

> +		pr_debug("no turbo range exists in _PSS\n");
> +		limits.no_turbo = limits.turbo_disabled = 1;
> +		cpu->pstate.turbo_pstate = cpu->pstate.max_pstate;
> +		turbo_absent = true;
> +	}
> +
> +	/* Check if the max non turbo p state < Intel P state max */
> +	for (i = 0; i < cpu->acpi_perf_data.state_count; i++) {
> +		if ((cpu->acpi_perf_data.states[i].core_frequency % 2) == 0) {

And this?

Generally, this reads "if core_frequency of state i is even", but this is
a number from the firmware, so I'm not sure why it being even or odd has
to be meaningful in any way?

> +			int perf_max_ctl =
> +				cpu->acpi_perf_data.states[i].control >> 8;
> +			if (cpu->pstate.max_pstate > perf_max_ctl)
> +				cpu->pstate.max_pstate = perf_max_ctl;
> +			break;
> +		}
> +	}
> +
> +	/* check If min perf > Intel P State min */
> +	min_perf = cpu->acpi_perf_data.states[
> +			cpu->acpi_perf_data.state_count - 1].control >> 8;
> +	if (min_perf > cpu->pstate.min_pstate) {
> +		cpu->pstate.min_pstate = min_perf;

What's the guarantee that min_perf can be used as min_pstate?

> +		policy->cpuinfo.min_freq = cpu->pstate.min_pstate *
> +						cpu->pstate.scaling;
> +	}
> +
> +	if (turbo_absent)
> +		policy->cpuinfo.max_freq =
> +			cpu->pstate.max_pstate * cpu->pstate.scaling;
> +	else {
> +		policy->cpuinfo.max_freq =
> +			cpu->pstate.turbo_pstate * cpu->pstate.scaling;
> +		/* scale freq to intel pstate turbo scale */
> +		cpu->acpi_perf_data.states[0].core_frequency =
> +				cpu->pstate.scaling *
> +				(cpu->acpi_perf_data.states[0].control >> 8);
> +	}
> +
> +	pr_debug("Updated limits 0x%x 0x%x 0x %x\n", cpu->pstate.min_pstate,
> +		 cpu->pstate.max_pstate, cpu->pstate.turbo_pstate);
> +	pr_debug("policy max_freq=%d Khz min_freq = %d KHz\n",
> +		 policy->cpuinfo.max_freq, policy->cpuinfo.min_freq);
> +
> +	return 0;
> +}
> +
> +static int intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
> +{
> +	struct cpudata *cpu;
> +
> +	if (!no_acpi_perf)
> +		return 0;
> +
> +	cpu = all_cpu_data[policy->cpu];
> +	acpi_processor_unregister_performance(&cpu->acpi_perf_data,
> +					      policy->cpu);
> +	return 0;
> +}
> +
> +#else
> +static int intel_pstate_init_perf_limits(struct cpufreq_policy *policy)
> +{
> +	return 0;
> +}
> +
> +static int intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
> +{
> +	return 0;
> +}
> +#endif
> +
>  static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
>  			     int deadband, int integral) {
>  	pid->setpoint = setpoint;
> @@ -1186,18 +1302,29 @@ static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
>  	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
>  	policy->cpuinfo.max_freq =
>  		cpu->pstate.turbo_pstate * cpu->pstate.scaling;
> +	if (!no_acpi_perf)
> +		intel_pstate_init_perf_limits(policy);
> +	/* If there is no acpi perf data or error, we ignore and use Intel P
> +	 * state calculated limits
> +	 */
>  	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
>  	cpumask_set_cpu(policy->cpu, policy->cpus);
>  
>  	return 0;
>  }
>  
> +static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
> +{
> +	return intel_pstate_exit_perf_limits(policy);
> +}
> +
>  static struct cpufreq_driver intel_pstate_driver = {
>  	.flags		= CPUFREQ_CONST_LOOPS,
>  	.verify		= intel_pstate_verify_policy,
>  	.setpolicy	= intel_pstate_set_policy,
>  	.get		= intel_pstate_get,
>  	.init		= intel_pstate_cpu_init,
> +	.exit		= intel_pstate_cpu_exit,
>  	.stop_cpu	= intel_pstate_stop_cpu,
>  	.name		= "intel_pstate",
>  };
> @@ -1429,6 +1556,9 @@ static int __init intel_pstate_setup(char *str)
>  		force_load = 1;
>  	if (!strcmp(str, "hwp_only"))
>  		hwp_only = 1;
> +	if (!strcmp(str, "no_acpi_perf"))
> +		no_acpi_perf = 1;
> +
>  	return 0;
>  }
>  early_param("intel_pstate", intel_pstate_setup);
> 

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

Re: [PATCH 3/5] cpufreq: intel-pstate: Use separate max pstate for scaling

[Srinivas Pandruvada]

On Thu, 2015-08-27 at 01:14 +0200, Rafael J. Wysocki wrote:
> On Wednesday, August 26, 2015 01:32:39 PM Srinivas Pandruvada wrote:
> > Systems with configurable TDP have multiple max non turbo p state. 
> > Intel
> > P state uses max non turbo P state for scaling. But using the real 
> > max
> > non turbo p state causes underestimation of next P state. So using
> > the physical max non turbo P state as before for scaling.
> 
> What does "physical" mean here?
That is the max non turbo p state a particular processor family can
reach. For example a Broadwell client processor, the max non turbo P
state can be up to 0x17. This I am calling physical. May be not a good
choice of word?
OEMs may not want the non turbo P state to be this high and can reduce
by changing turbo activation ratio.

Thanks,
Srinivas
> > Signed-off-by: Srinivas Pandruvada <
> > srinivas.pandruvada@linux.intel.com>
> > ---
> >  drivers/cpufreq/intel_pstate.c | 20 ++++++++++++++++++--
> >  1 file changed, 18 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/cpufreq/intel_pstate.c 
> > b/drivers/cpufreq/intel_pstate.c
> > index 2641102..a349462 100644
> > --- a/drivers/cpufreq/intel_pstate.c
> > +++ b/drivers/cpufreq/intel_pstate.c
> > @@ -77,6 +77,7 @@ struct pstate_data {
> >  	int	current_pstate;
> >  	int	min_pstate;
> >  	int	max_pstate;
> > +	int	max_pstate_physical;
> >  	int	scaling;
> >  	int	turbo_pstate;
> >  	u64	turbo_ratio_limit;
> > @@ -127,6 +128,7 @@ struct pstate_adjust_policy {
> >  
> >  struct pstate_funcs {
> >  	int (*get_max)(void);
> > +	int (*get_max_physical)(void);
> >  	int (*get_min)(void);
> >  	int (*get_turbo)(void);
> >  	int (*get_scaling)(void);
> > @@ -599,6 +601,14 @@ static int core_get_min_pstate(void)
> >  	return (value >> 40) & 0xFF;
> >  }
> >  
> > +static int core_get_max_pstate_physical(void)
> > +{
> > +	u64 value;
> > +
> > +	rdmsrl(MSR_PLATFORM_INFO, value);
> > +	return (value >> 8) & 0xFF;
> > +}
> > +
> >  static int core_get_max_pstate(void)
> >  {
> >  	u64 tar;
> > @@ -741,6 +751,7 @@ static struct cpu_defaults core_params = {
> >  	},
> >  	.funcs = {
> >  		.get_max = core_get_max_pstate,
> > +		.get_max_physical = core_get_max_pstate_physical,
> >  		.get_min = core_get_min_pstate,
> >  		.get_turbo = core_get_turbo_pstate,
> >  		.get_scaling = core_get_scaling,
> > @@ -761,6 +772,7 @@ static struct cpu_defaults byt_params = {
> >  	},
> >  	.funcs = {
> >  		.get_max = byt_get_max_pstate,
> > +		.get_max_physical = byt_get_max_pstate,
> >  		.get_min = byt_get_min_pstate,
> >  		.get_turbo = byt_get_turbo_pstate,
> >  		.set = byt_set_pstate,
> > @@ -780,6 +792,7 @@ static struct cpu_defaults knl_params = {
> >  	},
> >  	.funcs = {
> >  		.get_max = core_get_max_pstate,
> > +		.get_max_physical = core_get_max_pstate_physical,
> >  		.get_min = core_get_min_pstate,
> >  		.get_turbo = knl_get_turbo_pstate,
> >  		.set = core_set_pstate,
> > @@ -833,6 +846,7 @@ static void intel_pstate_get_cpu_pstates(struct 
> > cpudata *cpu)
> >  {
> >  	cpu->pstate.min_pstate = pstate_funcs.get_min();
> >  	cpu->pstate.max_pstate = pstate_funcs.get_max();
> > +	cpu->pstate.max_pstate_physical = 
> > pstate_funcs.get_max_physical();
> >  	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
> >  	cpu->pstate.scaling = pstate_funcs.get_scaling();
> >  	if (pstate_funcs.get_turbo_ratio_limit &&
> > @@ -854,7 +868,8 @@ static inline void 
> > intel_pstate_calc_busy(struct cpudata *cpu)
> >  
> >  	sample->freq = fp_toint(
> >  		mul_fp(int_tofp(
> > -			cpu->pstate.max_pstate * cpu
> > ->pstate.scaling / 100),
> > +			cpu->pstate.max_pstate_physical *
> > +			cpu->pstate.scaling / 100),
> >  			core_pct));
> >  
> >  	sample->core_pct_busy = (int32_t)core_pct;
> > @@ -922,7 +937,7 @@ static inline int32_t 
> > intel_pstate_get_scaled_busy(struct cpudata *cpu)
> >  	 * specified pstate.
> >  	 */
> >  	core_busy = cpu->sample.core_pct_busy;
> > -	max_pstate = int_tofp(cpu->pstate.max_pstate);
> > +	max_pstate = int_tofp(cpu->pstate.max_pstate_physical);
> >  	current_pstate = int_tofp(cpu->pstate.current_pstate);
> >  	core_busy = mul_fp(core_busy, div_fp(max_pstate, 
> > current_pstate));
> >  
> > @@ -1215,6 +1230,7 @@ static void copy_pid_params(struct 
> > pstate_adjust_policy *policy)
> >  static void copy_cpu_funcs(struct pstate_funcs *funcs)
> >  {
> >  	pstate_funcs.get_max   = funcs->get_max;
> > +	pstate_funcs.get_max_physical = funcs->get_max_physical;
> >  	pstate_funcs.get_min   = funcs->get_min;
> >  	pstate_funcs.get_turbo = funcs->get_turbo;
> >  	pstate_funcs.get_scaling = funcs->get_scaling;
> > 
> 

Re: [PATCH 4/5] cpufreq: intel_pstate: Use ACPI perf configuration

[Srinivas Pandruvada]

On Thu, 2015-08-27 at 03:11 +0200, Rafael J. Wysocki wrote:
> On Wednesday, August 26, 2015 01:32:40 PM Srinivas Pandruvada wrote:
> > Use ACPI _PSS to limit the Intel P State turbo, max and min ratios.
> > This driver uses acpi processor perf lib calls to register 
> > performance.
> > The following logic is used to adjust Intel P state driver limits:
> > - If there is no turbo entry in _PSS, then disable Intel P state 
> > turbo
> > and limit to non turbo max
> > - If the non turbo max ratio is more than _PSS max non turbo value, 
> > then
> > set the max non turbo ratio to _PSS non turbo max
> > - If the min ratio is less than _PSS min then change the min ratio
> > matching _PSS min
> > - Scale the _PSS turbo frequency to max turbo frequency based on 
> > control
> > value.
> 
> The new command line switch should be mentioned in the changelog too.
Missed it. Also need to update the kernel command line document.
> 
> > Signed-off-by: Srinivas Pandruvada <
> > srinivas.pandruvada@linux.intel.com>
> > ---
> >  drivers/cpufreq/Kconfig.x86    |   1 +
> >  drivers/cpufreq/intel_pstate.c | 130 
> > +++++++++++++++++++++++++++++++++++++++++
> >  2 files changed, 131 insertions(+)
> > 
> > diff --git a/drivers/cpufreq/Kconfig.x86 
> > b/drivers/cpufreq/Kconfig.x86
> > index c59bdcb..1c6affd 100644
> > --- a/drivers/cpufreq/Kconfig.x86
> > +++ b/drivers/cpufreq/Kconfig.x86
> > @@ -5,6 +5,7 @@
> >  config X86_INTEL_PSTATE
> >         bool "Intel P state control"
> >         depends on X86
> > +       select ACPI_PROCESSOR
> 
> 	select ACPI_PROCESSOR if ACPI
> 
> or there will be a broken dependency if ACPI is not selected.
I will fix this in next patchset.
> 
> >         help
> >            This driver provides a P state for Intel core 
> > processors.
> >  	  The driver implements an internal governor and will 
> > become
> > diff --git a/drivers/cpufreq/intel_pstate.c 
> > b/drivers/cpufreq/intel_pstate.c
> > index a349462..6eafb55 100644
> > --- a/drivers/cpufreq/intel_pstate.c
> > +++ b/drivers/cpufreq/intel_pstate.c
> > @@ -34,6 +34,8 @@
> >  #include <asm/cpu_device_id.h>
> >  #include <asm/cpufeature.h>
> >  
> > +#include <acpi/processor.h>
> > +
> >  #define BYT_RATIOS		0x66a
> >  #define BYT_VIDS		0x66b
> >  #define BYT_TURBO_RATIOS	0x66c
> > @@ -114,6 +116,9 @@ struct cpudata {
> >  	u64	prev_mperf;
> >  	u64	prev_tsc;
> >  	struct sample sample;
> > +#if IS_ENABLED(CONFIG_ACPI)
> > +	struct acpi_processor_performance acpi_perf_data;
> > +#endif
> >  };
> >  
> >  static struct cpudata **all_cpu_data;
> > @@ -146,6 +151,7 @@ struct cpu_defaults {
> >  static struct pstate_adjust_policy pid_params;
> >  static struct pstate_funcs pstate_funcs;
> >  static int hwp_active;
> > +static int no_acpi_perf;
> 
> Why don't you call it "no_acpi"?  The "perf" part seems to be 
> redundant.
> 
OK. I will change this.
> >  
> >  struct perf_limits {
> >  	int no_turbo;
> > @@ -173,6 +179,116 @@ static struct perf_limits limits = {
> >  	.min_sysfs_pct = 0,
> >  };
> >  
> > +#if IS_ENABLED(CONFIG_ACPI)
> > +static int intel_pstate_init_perf_limits(struct cpufreq_policy 
> > *policy)
> > +{
> > +	struct cpudata *cpu;
> > +	int ret;
> > +	int min_perf;
> > +	bool turbo_absent = false;
> > +	int i;
> > +
> > +	cpu = all_cpu_data[policy->cpu];
> > +
> > +	if (!cpu->acpi_perf_data.shared_cpu_map &&
> > +	    zalloc_cpumask_var_node(&cpu
> > ->acpi_perf_data.shared_cpu_map,
> > +				    GFP_KERNEL, cpu_to_node(policy
> > ->cpu))) {
> > +		return -ENOMEM;
> > +	}
> > +
> > +	ret = acpi_processor_register_performance(&cpu
> > ->acpi_perf_data,
> > +						  policy->cpu);
> > +	if (ret)
> > +		return ret;
> > +
> > +	if (cpu->acpi_perf_data.control_register.space_id !=
> > +						ACPI_ADR_SPACE_FIX
> > ED_HARDWARE)
> > +		return -EIO;
> > +
> > +	pr_debug("CPU%u - ACPI _PSS perf data\n", policy->cpu);
> > +	for (i = 0; i < cpu->acpi_perf_data.state_count; i++)
> > +		pr_debug("     %cP%d: %u MHz, %u mW, 0x%x\n",
> > +			 (i == cpu->acpi_perf_data.state ? '*' : ' 
> > '), i,
> > +			 (u32) cpu
> > ->acpi_perf_data.states[i].core_frequency,
> > +			 (u32) cpu
> > ->acpi_perf_data.states[i].power,
> > +			 (u32) cpu
> > ->acpi_perf_data.states[i].control);
> > +
> > +	if (cpu->acpi_perf_data.state_count < 2)
> > +		return 0;
> 
> So what happens if we do that?  We'll not use ACPI limits, right?
> 
Yes, we will continue as before with Intel P state calculated limits
based on MSRs.
> > +
> > +	/* Check if there is a turbo freq in _PSS */
> > +	if ((cpu->acpi_perf_data.states[0].core_frequency % 2) == 
> > 0) {
> 
> Where does this check come from?
Isn't general convention to specify turbo frequency (max non turbo freq
+ 1 MHz)? Usually these are specified by BIOS writer guides.
> 
> > +		pr_debug("no turbo range exists in _PSS\n");
> > +		limits.no_turbo = limits.turbo_disabled = 1;
> > +		cpu->pstate.turbo_pstate = cpu->pstate.max_pstate;
> > +		turbo_absent = true;
> > +	}
> > +
> > +	/* Check if the max non turbo p state < Intel P state max 
> > */
> > +	for (i = 0; i < cpu->acpi_perf_data.state_count; i++) {
> > +		if ((cpu->acpi_perf_data.states[i].core_frequency 
> > % 2) == 0) {
> 
> And this?
> 
> Generally, this reads "if core_frequency of state i is even", but 
> this is
> a number from the firmware, so I'm not sure why it being even or odd 
> has
> to be meaningful in any way?
> 
The first even frequency will be max non turbo. There can't be a
frequncy 2301 Mhz in Intel Processors (which is indicating a turbo
frequency). For example:
[    1.233597]      *P0: 2301 MHz, 15000 mW, 0x1d00
[    1.233598]       P1: 2300 MHz, 15000 mW, 0x1700
[    1.233599]       P2: 2200 MHz, 14088 mW, 0x1600
[    1.233600]       P3: 2000 MHz, 12462 mW, 0x1400
[    1.233601]       P4: 1900 MHz, 11745 mW, 0x1300
[    1.233602]       P5: 1800 MHz, 11042 mW, 0x1200
[    1.233603]       P6: 1700 MHz, 10215 mW, 0x1100
[    1.233604]       P7: 1500 MHz, 8882 mW, 0xf00
[    1.233605]       P8: 1400 MHz, 8101 mW, 0xe00
[    1.233606]       P9: 1300 MHz, 7471 mW, 0xd00
[    1.233607]       P10: 1100 MHz, 6121 mW, 0xb00
[    1.233608]       P11: 1000 MHz, 5531 mW, 0xa00
[    1.233610]       P12: 900 MHz, 4955 mW, 0x900
[    1.233611]       P13: 800 MHz, 4267 mW, 0x800
[    1.233612]       P14: 600 MHz, 3182 mW, 0x600
[    1.233613]       P15: 500 MHz, 2538 mW, 0x500

Let me think if I can use control values (the last column in the above
log) instead of using odd/even.

> > +			int perf_max_ctl =
> > +				cpu
> > ->acpi_perf_data.states[i].control >> 8;
> > +			if (cpu->pstate.max_pstate > perf_max_ctl)
> > +				cpu->pstate.max_pstate = 
> > perf_max_ctl;
> > +			break;
> > +		}
> > +	}
> > +
> > +	/* check If min perf > Intel P State min */
> > +	min_perf = cpu->acpi_perf_data.states[
> > +			cpu->acpi_perf_data.state_count - 
> > 1].control >> 8;
> > +	if (min_perf > cpu->pstate.min_pstate) {
> > +		cpu->pstate.min_pstate = min_perf;
> 
> What's the guarantee that min_perf can be used as min_pstate?
Intel P state min is the lowest possible P state from the processor
platform info MSR. So if the ACPI specified min_perf is more, then it
should work. It is possible that ACPI is setting floor for P states.
> > +		policy->cpuinfo.min_freq = cpu->pstate.min_pstate 
> > *
> > +						cpu
> > ->pstate.scaling;
> > +	}
> > +
> > +	if (turbo_absent)
> > +		policy->cpuinfo.max_freq =
> > +			cpu->pstate.max_pstate * cpu
> > ->pstate.scaling;
> > +	else {
> > +		policy->cpuinfo.max_freq =
> > +			cpu->pstate.turbo_pstate * cpu
> > ->pstate.scaling;
> > +		/* scale freq to intel pstate turbo scale */
> > +		cpu->acpi_perf_data.states[0].core_frequency =
> > +				cpu->pstate.scaling *
> > +				(cpu
> > ->acpi_perf_data.states[0].control >> 8);
> > +	}
> > +
> > +	pr_debug("Updated limits 0x%x 0x%x 0x %x\n", cpu
> > ->pstate.min_pstate,
> > +		 cpu->pstate.max_pstate, cpu
> > ->pstate.turbo_pstate);
> > +	pr_debug("policy max_freq=%d Khz min_freq = %d KHz\n",
> > +		 policy->cpuinfo.max_freq, policy
> > ->cpuinfo.min_freq);
> > +
> > +	return 0;
> > +}
> > +
> > +static int intel_pstate_exit_perf_limits(struct cpufreq_policy 
> > *policy)
> > +{
> > +	struct cpudata *cpu;
> > +
> > +	if (!no_acpi_perf)
> > +		return 0;
> > +
> > +	cpu = all_cpu_data[policy->cpu];
> > +	acpi_processor_unregister_performance(&cpu
> > ->acpi_perf_data,
> > +					      policy->cpu);
> > +	return 0;
> > +}
> > +
> > +#else
> > +static int intel_pstate_init_perf_limits(struct cpufreq_policy 
> > *policy)
> > +{
> > +	return 0;
> > +}
> > +
> > +static int intel_pstate_exit_perf_limits(struct cpufreq_policy 
> > *policy)
> > +{
> > +	return 0;
> > +}
> > +#endif
> > +
> >  static inline void pid_reset(struct _pid *pid, int setpoint, int 
> > busy,
> >  			     int deadband, int integral) {
> >  	pid->setpoint = setpoint;
> > @@ -1186,18 +1302,29 @@ static int intel_pstate_cpu_init(struct 
> > cpufreq_policy *policy)
> >  	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu
> > ->pstate.scaling;
> >  	policy->cpuinfo.max_freq =
> >  		cpu->pstate.turbo_pstate * cpu->pstate.scaling;
> > +	if (!no_acpi_perf)
> > +		intel_pstate_init_perf_limits(policy);
> > +	/* If there is no acpi perf data or error, we ignore and 
> > use Intel P
> > +	 * state calculated limits
> > +	 */
> >  	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
> >  	cpumask_set_cpu(policy->cpu, policy->cpus);
> >  
> >  	return 0;
> >  }
> >  
> > +static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
> > +{
> > +	return intel_pstate_exit_perf_limits(policy);
> > +}
> > +
> >  static struct cpufreq_driver intel_pstate_driver = {
> >  	.flags		= CPUFREQ_CONST_LOOPS,
> >  	.verify		= intel_pstate_verify_policy,
> >  	.setpolicy	= intel_pstate_set_policy,
> >  	.get		= intel_pstate_get,
> >  	.init		= intel_pstate_cpu_init,
> > +	.exit		= intel_pstate_cpu_exit,
> >  	.stop_cpu	= intel_pstate_stop_cpu,
> >  	.name		= "intel_pstate",
> >  };
> > @@ -1429,6 +1556,9 @@ static int __init intel_pstate_setup(char 
> > *str)
> >  		force_load = 1;
> >  	if (!strcmp(str, "hwp_only"))
> >  		hwp_only = 1;
> > +	if (!strcmp(str, "no_acpi_perf"))
> > +		no_acpi_perf = 1;
> > +
> >  	return 0;
> >  }
> >  early_param("intel_pstate", intel_pstate_setup);
> > 
>