[PATCH v4 07/11] x86/intel_pstate: the main boby of the intel_pstate driver

[PATCH v4 07/11] x86/intel_pstate: the main boby of the intel_pstate driver

[Wei Wang]

The intel_pstate driver is ported following its kernel code logic
(commit: 93f0822d).In order to port the Linux source file with
minimal modifications, some of the variable types are kept intact
(e.g. "int current_pstae", would otherwise be changed to
"unsigned int").

In the kernel, a user can adjust the limits via sysfs
(limits.min_sysfs_pct/max_sysfs_pct). In Xen, the
policy->limits.min_perf_pct/max_perf_pct acts as the transit station.
A user interacts with it via xenpm.

The new xen/include/asm-x86/cpufreq.h header file is added.

v4 changes:
1) changed the identation to be a "Tab" (same as Linux intel_pstate),
   instead of 4 "+$";
2) added a new header file, xen/include/asm-x86/cpufreq.h.

Signed-off-by: Wei Wang <wei.w.wang@intel.com>
---
 xen/arch/x86/acpi/cpufreq/Makefile       |   1 +
 xen/arch/x86/acpi/cpufreq/intel_pstate.c | 870 +++++++++++++++++++++++++++++++
 xen/include/asm-x86/cpufreq.h            |  34 ++
 xen/include/asm-x86/msr-index.h          |   3 +
 4 files changed, 908 insertions(+)
 create mode 100644 xen/arch/x86/acpi/cpufreq/intel_pstate.c
 create mode 100644 xen/include/asm-x86/cpufreq.h

diff --git a/xen/arch/x86/acpi/cpufreq/Makefile b/xen/arch/x86/acpi/cpufreq/Makefile
index f75da9b..99fa9f4 100644
--- a/xen/arch/x86/acpi/cpufreq/Makefile
+++ b/xen/arch/x86/acpi/cpufreq/Makefile
@@ -1,2 +1,3 @@
 obj-y += cpufreq.o
+obj-y += intel_pstate.o
 obj-y += powernow.o
diff --git a/xen/arch/x86/acpi/cpufreq/intel_pstate.c b/xen/arch/x86/acpi/cpufreq/intel_pstate.c
new file mode 100644
index 0000000..19c74cc
--- /dev/null
+++ b/xen/arch/x86/acpi/cpufreq/intel_pstate.c
@@ -0,0 +1,870 @@
+#include <xen/kernel.h>
+#include <xen/types.h>
+#include <xen/init.h>
+#include <xen/bitmap.h>
+#include <xen/cpumask.h>
+#include <xen/timer.h>
+#include <asm/msr.h>
+#include <asm/msr-index.h>
+#include <asm/processor.h>
+#include <asm/div64.h>
+#include <asm/cpufreq.h>
+#include <acpi/cpufreq/cpufreq.h>
+
+#define BYT_RATIOS	  0x66a
+#define BYT_VIDS	  0x66b
+#define BYT_TURBO_RATIOS  0x66c
+#define BYT_TURBO_VIDS	  0x66d
+
+#define FRAC_BITS 8
+#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;
+}
+
+static inline int32_t div_fp(int32_t x, int32_t y)
+{
+	return div_s64((int64_t)x << FRAC_BITS, y);
+}
+
+static inline int ceiling_fp(int32_t x)
+{
+	int mask, ret;
+
+	ret = fp_toint(x);
+	mask = (1 << FRAC_BITS) - 1;
+	if (x & mask)
+		ret += 1;
+	return ret;
+}
+
+struct sample {
+	int32_t core_pct_busy;
+	u64 aperf;
+	u64 mperf;
+	int freq;
+	s_time_t time;
+};
+
+struct pstate_data {
+	int	current_pstate;
+	int	min_pstate;
+	int	max_pstate;
+	int	scaling;
+	int	turbo_pstate;
+};
+
+struct vid_data {
+	int min;
+	int max;
+	int turbo;
+	int32_t ratio;
+};
+
+struct _pid {
+	int setpoint;
+	int32_t integral;
+	int32_t p_gain;
+	int32_t i_gain;
+	int32_t d_gain;
+	int deadband;
+	int32_t last_err;
+};
+
+struct cpudata {
+	int cpu;
+
+	struct timer timer;
+
+	struct pstate_data pstate;
+	struct vid_data vid;
+	struct _pid pid;
+
+	s_time_t last_sample_time;
+	u64	prev_aperf;
+	u64	prev_mperf;
+	struct sample sample;
+};
+
+static struct cpudata **all_cpu_data;
+
+struct pstate_adjust_policy {
+	int sample_rate_ms;
+	int deadband;
+	int setpoint;
+	int p_gain_pct;
+	int d_gain_pct;
+	int i_gain_pct;
+};
+
+struct pstate_funcs {
+	int (*get_max)(void);
+	int (*get_min)(void);
+	int (*get_turbo)(void);
+	int (*get_scaling)(void);
+	void (*set)(struct perf_limits *, struct cpudata *, int pstate);
+	void (*get_vid)(struct cpudata *);
+};
+
+struct cpu_defaults {
+	struct pstate_adjust_policy pid_policy;
+	struct pstate_funcs funcs;
+};
+
+static struct pstate_adjust_policy pid_params;
+static struct pstate_funcs pstate_funcs;
+
+static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
+			     int deadband, int integral) {
+	pid->setpoint = setpoint;
+	pid->deadband  = deadband;
+	pid->integral  = int_tofp(integral);
+	pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
+}
+
+static inline void pid_p_gain_set(struct _pid *pid, int percent)
+{
+	pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
+}
+
+static inline void pid_i_gain_set(struct _pid *pid, int percent)
+{
+	pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
+}
+
+static inline void pid_d_gain_set(struct _pid *pid, int percent)
+{
+	pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
+}
+
+static signed int pid_calc(struct _pid *pid, int32_t busy)
+{
+	signed int result;
+	int32_t pterm, dterm, fp_error;
+	int32_t integral_limit;
+
+	fp_error = int_tofp(pid->setpoint) - busy;
+
+	if (ABS(fp_error) <= int_tofp(pid->deadband))
+		return 0;
+
+	pterm = mul_fp(pid->p_gain, fp_error);
+
+	pid->integral += fp_error;
+
+	/*
+	 * We limit the integral here so that it will never
+	 * get higher than 30.  This prevents it from becoming
+	 * too large an input over long periods of time and allows
+	 * it to get factored out sooner.
+	 * The value of 30 was chosen through experimentation.
+	 */
+	integral_limit = int_tofp(30);
+	if (pid->integral > integral_limit)
+		pid->integral = integral_limit;
+	if (pid->integral < -integral_limit)
+		pid->integral = -integral_limit;
+
+	dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
+	pid->last_err = fp_error;
+
+	result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
+	result = result + (1 << (FRAC_BITS-1));
+	return (signed int)fp_toint(result);
+}
+
+static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
+{
+	pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
+	pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
+	pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
+
+	pid_reset(&cpu->pid, pid_params.setpoint, 100, pid_params.deadband, 0);
+}
+
+static inline void intel_pstate_reset_all_pid(void)
+{
+	unsigned int cpu;
+
+	for_each_online_cpu(cpu) {
+		if (all_cpu_data[cpu])
+			intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
+	}
+}
+
+static inline void update_turbo_state(struct cpufreq_policy *policy)
+{
+	u64 misc_en;
+	struct cpudata *cpu;
+
+	cpu = all_cpu_data[policy->cpu];
+	rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
+	policy->limits.turbo_disabled =
+		(misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
+			cpu->pstate.max_pstate == cpu->pstate.turbo_pstate);
+}
+
+#define BYT_TURBO_CONTROL_BIT 32
+#define BYT_MIN_PSTATE(val) (((value) >> 8) & 0x7f)
+#define BYT_MAX_PSTATE(val) (((value) >> 16) & 0x7f)
+#define BYT_TURBO_PSTATE(value) ((value) & 0x7f)
+static int byt_get_min_pstate(void)
+{
+	u64 value;
+
+	rdmsrl(BYT_RATIOS, value);
+	return BYT_MIN_PSTATE(val);
+}
+
+static int byt_get_max_pstate(void)
+{
+	u64 value;
+
+	rdmsrl(BYT_RATIOS, value);
+	return BYT_MAX_PSTATE(val);
+}
+
+static int byt_get_turbo_pstate(void)
+{
+	u64 value;
+
+	rdmsrl(BYT_TURBO_RATIOS, value);
+	return BYT_TURBO_PSTATE(value);
+}
+
+static void byt_set_pstate(struct perf_limits *limits,
+				struct cpudata *cpudata, int pstate)
+{
+	u64 val;
+	int32_t vid_fp;
+	u32 vid;
+
+	val = pstate << 8;
+	if (limits->no_turbo && !limits->turbo_disabled)
+		val |= (u64)1 << BYT_TURBO_CONTROL_BIT;
+
+	vid_fp = cpudata->vid.min + mul_fp(
+		int_tofp(pstate - cpudata->pstate.min_pstate),
+		cpudata->vid.ratio);
+
+	vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
+	vid = ceiling_fp(vid_fp);
+
+	if (pstate > cpudata->pstate.max_pstate)
+		vid = cpudata->vid.turbo;
+
+	val |= vid;
+
+	wrmsrl(MSR_IA32_PERF_CTL, val);
+}
+
+#define BYT_BCLK_FREQS 5
+#define TO_FREQ_TABLE_IDX_MASK 0x7
+static const int byt_freq_table[BYT_BCLK_FREQS] = { 833, 1000, 1333, 1167, 800};
+
+static int byt_get_scaling(void)
+{
+	u64 value;
+	int i;
+
+	rdmsrl(MSR_FSB_FREQ, value);
+	i = value & TO_FREQ_TABLE_IDX_MASK;
+
+	BUG_ON(i > BYT_BCLK_FREQS);
+
+	return byt_freq_table[i] * 100;
+}
+
+static void byt_get_vid(struct cpudata *cpudata)
+{
+	u64 value;
+
+	rdmsrl(BYT_VIDS, value);
+	cpudata->vid.min = int_tofp(BYT_MIN_PSTATE(val));
+	cpudata->vid.max = int_tofp(BYT_MAX_PSTATE(val));
+	cpudata->vid.ratio = div_fp(
+		cpudata->vid.max - cpudata->vid.min,
+		int_tofp(cpudata->pstate.max_pstate -
+			cpudata->pstate.min_pstate));
+
+	rdmsrl(BYT_TURBO_VIDS, value);
+	cpudata->vid.turbo = BYT_TURBO_PSTATE(value);
+}
+
+#define SCALING_FACTOR 100000
+#define CORE_TURBO_CONTROL_BIT 32
+#define CORE_MIN_PSTATE(val) (((value) >> 40) & 0xff)
+#define CORE_MAX_PSTATE(val) (((value) >> 8) & 0xff)
+#define CORE_TURBO_PSTATE(value) ((value) & 0xff)
+static int core_get_min_pstate(void)
+{
+	u64 value;
+
+	rdmsrl(MSR_INTEL_PLATFORM_INFO, value);
+	return CORE_MIN_PSTATE(val);
+}
+
+static int core_get_max_pstate(void)
+{
+	u64 value;
+
+	rdmsrl(MSR_INTEL_PLATFORM_INFO, value);
+	return CORE_MAX_PSTATE(val);
+}
+
+static int core_get_turbo_pstate(void)
+{
+	u64 value;
+	int nont, ret;
+
+	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
+	nont = core_get_max_pstate();
+	ret = CORE_TURBO_PSTATE(value);
+	if (ret <= nont)
+		ret = nont;
+	return ret;
+}
+
+static inline int core_get_scaling(void)
+{
+	return SCALING_FACTOR;
+}
+
+static void core_set_pstate(struct perf_limits *limits,
+				struct cpudata *cpudata, int pstate)
+{
+	u64 val;
+
+	val = pstate << 8;
+	if (limits->no_turbo && !limits->turbo_disabled)
+		val |= (u64)1 << CORE_TURBO_CONTROL_BIT;
+
+	wrmsrl(MSR_IA32_PERF_CTL, val);
+}
+
+static const struct cpu_defaults core_params = {
+	.pid_policy = {
+		.sample_rate_ms = 10,
+		.deadband = 0,
+		.setpoint = 97,
+		.p_gain_pct = 20,
+		.d_gain_pct = 0,
+		.i_gain_pct = 0,
+	},
+	.funcs = {
+		.get_max = core_get_max_pstate,
+		.get_min = core_get_min_pstate,
+		.get_turbo = core_get_turbo_pstate,
+		.get_scaling = core_get_scaling,
+		.set = core_set_pstate,
+	},
+};
+
+static const struct cpu_defaults byt_params = {
+	.pid_policy = {
+		.sample_rate_ms = 10,
+		.deadband = 0,
+		.setpoint = 97,
+		.p_gain_pct = 14,
+		.d_gain_pct = 0,
+		.i_gain_pct = 4,
+	},
+	.funcs = {
+		.get_max = byt_get_max_pstate,
+		.get_min = byt_get_min_pstate,
+		.get_turbo = byt_get_turbo_pstate,
+		.set = byt_set_pstate,
+		.get_scaling = byt_get_scaling,
+		.get_vid = byt_get_vid,
+	},
+};
+
+static void intel_pstate_get_min_max(struct perf_limits *limits,
+				struct cpudata *cpu, int *min, int *max)
+{
+	int max_perf = cpu->pstate.turbo_pstate;
+	int max_perf_adj;
+	int min_perf;
+
+	if (limits->no_turbo || limits->turbo_disabled)
+		max_perf = cpu->pstate.max_pstate;
+
+	/* performance can be limited by user through xenpm */
+	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);
+}
+
+static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
+{
+	int max_perf, min_perf;
+	struct cpufreq_policy *policy;
+	struct perf_limits *limits;
+
+	policy = per_cpu(cpufreq_cpu_policy, cpu->cpu);
+	limits = &policy->limits;
+
+	update_turbo_state(policy);
+
+	if (limits->turbo_disabled)
+		policy->turbo = CPUFREQ_TURBO_UNSUPPORTED;
+	else if (limits->no_turbo)
+		policy->turbo = CPUFREQ_TURBO_DISABLED;
+	else
+		policy->turbo = CPUFREQ_TURBO_ENABLED;
+
+	intel_pstate_get_min_max(limits, cpu, &min_perf, &max_perf);
+
+	pstate = clamp_t(int, pstate, min_perf, max_perf);
+
+	if (pstate == cpu->pstate.current_pstate)
+		return;
+
+	cpu->pstate.current_pstate = pstate;
+	policy->cur = pstate * SCALING_FACTOR;
+
+	pstate_funcs.set(limits, cpu, pstate);
+}
+
+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.turbo_pstate = pstate_funcs.get_turbo();
+	cpu->pstate.scaling = pstate_funcs.get_scaling();
+
+	if (pstate_funcs.get_vid)
+		pstate_funcs.get_vid(cpu);
+	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
+}
+
+static inline void intel_pstate_calc_busy(struct cpudata *cpu)
+{
+	struct sample *sample = &cpu->sample;
+	int64_t core_pct;
+
+	core_pct = int_tofp(sample->aperf) * int_tofp(100);
+	core_pct = div64_u64(core_pct, int_tofp(sample->mperf));
+
+	sample->freq = fp_toint(
+		mul_fp(int_tofp(
+			cpu->pstate.max_pstate * cpu->pstate.scaling / 100),
+			core_pct));
+
+	sample->core_pct_busy = (int32_t)core_pct;
+}
+
+static inline void intel_pstate_sample(struct cpudata *cpu)
+{
+	u64 aperf, mperf;
+	unsigned long flags;
+
+	local_irq_save(flags);
+	rdmsrl(MSR_IA32_APERF, aperf);
+	rdmsrl(MSR_IA32_MPERF, mperf);
+	local_irq_restore(flags);
+
+	cpu->last_sample_time = cpu->sample.time;
+	cpu->sample.time = get_s_time();
+	cpu->sample.aperf = aperf;
+	cpu->sample.mperf = mperf;
+	cpu->sample.aperf -= cpu->prev_aperf;
+	cpu->sample.mperf -= cpu->prev_mperf;
+
+	intel_pstate_calc_busy(cpu);
+
+	cpu->prev_aperf = aperf;
+	cpu->prev_mperf = mperf;
+}
+
+static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
+{
+	set_timer(&cpu->timer, NOW() + MILLISECS(pid_params.sample_rate_ms));
+}
+
+static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
+{
+	int32_t core_busy, max_pstate, current_pstate, sample_ratio;
+	u32 duration_us;
+	u32 sample_time_us;
+
+	/*
+	 * core_busy is the ratio of actual performance to max
+	 * max_pstate is the max non turbo pstate available
+	 * current_pstate was the pstate that was requested during
+	 * the last sample period.
+	 *
+	 * We normalize core_busy, which was our actual percent
+	 * performance to what we requested during the last sample
+	 * period. The result will be a percentage of busy at a
+	 * specified pstate.
+	 */
+	core_busy = cpu->sample.core_pct_busy;
+	max_pstate = int_tofp(cpu->pstate.max_pstate);
+	current_pstate = int_tofp(cpu->pstate.current_pstate);
+	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
+
+	/*
+	 * Since we have a deferred timer, it will not fire unless
+	 * we are in C0.  So, determine if the actual elapsed time
+	 * is significantly greater (3x) than our sample interval. If it
+	 * is, then we were idle for a long enough period of time
+	 * to adjust our busyness.
+	 */
+	sample_time_us = pid_params.sample_rate_ms  * 1000ULL;
+	duration_us = (u32)((s_time_t)(cpu->sample.time - cpu->last_sample_time)
+			/ 1000);
+	if (duration_us > sample_time_us * 3) {
+		sample_ratio = div_fp(int_tofp(sample_time_us),
+				      int_tofp(duration_us));
+		core_busy = mul_fp(core_busy, sample_ratio);
+	}
+
+	return core_busy;
+}
+
+static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
+{
+	int32_t busy_scaled;
+	struct _pid *pid;
+	signed int ctl;
+
+	pid = &cpu->pid;
+	busy_scaled = intel_pstate_get_scaled_busy(cpu);
+
+	ctl = pid_calc(pid, busy_scaled);
+
+	/* Negative values of ctl increase the pstate and vice versa */
+	intel_pstate_set_pstate(cpu, cpu->pstate.current_pstate - ctl);
+}
+
+static void intel_pstate_timer_func(void *data)
+{
+	struct cpudata *cpu = (struct cpudata *) data;
+
+	intel_pstate_sample(cpu);
+
+	intel_pstate_adjust_busy_pstate(cpu);
+
+	intel_pstate_set_sample_time(cpu);
+}
+
+#define ICPU(model, policy) \
+	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
+			&policy##_params }
+
+static const struct x86_cpu_id intel_pstate_cpu_ids[] __initconst = {
+	ICPU(0x2a, core),
+	ICPU(0x2d, core),
+	ICPU(0x37, byt),
+	ICPU(0x3a, core),
+	ICPU(0x3c, core),
+	ICPU(0x3d, core),
+	ICPU(0x3e, core),
+	ICPU(0x3f, core),
+	ICPU(0x45, core),
+	ICPU(0x46, core),
+	ICPU(0x47, core),
+	ICPU(0x4c, byt),
+	ICPU(0x4e, core),
+	ICPU(0x4f, core),
+	ICPU(0x56, core),
+	{}
+};
+
+static int intel_pstate_init_cpu(unsigned int cpunum)
+{
+	struct cpudata *cpu;
+	s_time_t expires;
+
+	if (!all_cpu_data[cpunum])
+		all_cpu_data[cpunum] = xzalloc(struct cpudata);
+	if (!all_cpu_data[cpunum])
+		return -ENOMEM;
+
+	cpu = all_cpu_data[cpunum];
+
+	cpu->cpu = cpunum;
+	intel_pstate_get_cpu_pstates(cpu);
+
+	init_timer(&cpu->timer, intel_pstate_timer_func, cpu, cpunum);
+	expires = NOW() + MILLISECS(10);
+
+	intel_pstate_busy_pid_reset(cpu);
+	intel_pstate_sample(cpu);
+
+	set_timer(&cpu->timer, expires);
+
+	return 0;
+}
+
+static int intel_pstate_set_policy(struct cpufreq_policy *policy)
+{
+	struct perf_limits *limits = &policy->limits;
+	uint32_t cur_gov = policy->internal_gov->cur_gov;
+
+	if (!policy->cpuinfo.max_freq)
+		return -ENODEV;
+
+	switch (cur_gov) {
+	case INTERNAL_GOV_PERFORMANCE:
+		limits->no_turbo = 0;
+		limits->max_perf_pct = 100;
+		limits->max_perf = int_tofp(1);
+		limits->min_perf_pct = 100;
+		limits->min_perf = int_tofp(1);
+		break;
+	case INTERNAL_GOV_POWERSAVE:
+		limits->min_perf =
+			div_fp(int_tofp(limits->min_policy_pct),
+						int_tofp(100));
+		limits->max_perf = limits->min_perf;
+		limits->min_perf_pct = limits->min_policy_pct;
+		limits->max_perf_pct = limits->min_perf_pct;
+		break;
+	case INTERNAL_GOV_USERSPACE:
+		limits->max_perf =
+			div_fp(int_tofp(limits->max_perf_pct),
+						int_tofp(100));
+		limits->min_perf = limits->max_perf;
+		limits->min_perf_pct = limits->max_perf_pct;
+		break;
+	case INTERNAL_GOV_ONDEMAND:
+	default:
+		limits->min_perf =
+			div_fp(int_tofp(limits->min_perf_pct), int_tofp(100));
+		limits->max_perf =
+			div_fp(int_tofp(limits->max_perf_pct), int_tofp(100));
+		break;
+	}
+
+	return 0;
+}
+
+static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
+{
+	uint32_t cur_gov = policy->internal_gov->cur_gov;
+
+	cpufreq_verify_within_limits(policy, policy->cpuinfo.min_freq,
+						 policy->cpuinfo.max_freq);
+
+	switch(cur_gov) {
+	case INTERNAL_GOV_PERFORMANCE:
+	case INTERNAL_GOV_POWERSAVE:
+	case INTERNAL_GOV_USERSPACE:
+	case INTERNAL_GOV_ONDEMAND:
+					return 0;
+	default:
+					return -EINVAL;
+	}
+}
+
+static void intel_pstate_internal_gov_release(struct internal_governor *gov)
+{
+	xfree(gov->avail_gov);
+	xfree(gov);
+}
+
+static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
+{
+	int cpu_num = policy->cpu;
+	struct cpudata *cpu = all_cpu_data[cpu_num];
+
+	kill_timer(&all_cpu_data[cpu_num]->timer);
+
+	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
+
+	intel_pstate_internal_gov_release(policy->internal_gov);
+
+	return 0;
+}
+
+static int intel_pstate_turbo_update(int cpuid, struct cpufreq_policy *policy)
+{
+	struct cpudata *cpu = all_cpu_data[policy->cpu];
+	struct perf_limits *limits = &policy->limits;
+
+	update_turbo_state(policy);
+	if (limits->turbo_disabled) {
+		printk("Turbo disabled by BIOS or not supported on CPU\n");
+		return -EINVAL;
+	}
+	limits->no_turbo = policy->turbo == CPUFREQ_TURBO_ENABLED ? 0 : 1;
+
+	if (limits->no_turbo)
+		policy->cpuinfo.max_freq =
+			cpu->pstate.max_pstate * cpu->pstate.scaling;
+	else
+		policy->cpuinfo.max_freq =
+			cpu->pstate.turbo_pstate * cpu->pstate.scaling;
+
+	policy->max = clamp_t(unsigned int, policy->max,
+			policy->cpuinfo.min_freq, policy->cpuinfo.max_freq);
+
+	return 0;
+}
+
+static int get_turbo_pct(struct cpudata *cpu)
+{
+	int total, no_turbo, turbo_pct;
+	uint32_t turbo_fp;
+
+	total = cpu->pstate.turbo_pstate - cpu->pstate.min_pstate + 1;
+	no_turbo = cpu->pstate.max_pstate - cpu->pstate.min_pstate + 1;
+	turbo_fp = div_fp(int_tofp(no_turbo), int_tofp(total));
+	turbo_pct = 100 - fp_toint(mul_fp(turbo_fp, int_tofp(100)));
+	return turbo_pct;
+}
+
+#define INTEL_PSTATE_GOV_NUM 4
+static struct internal_governor* intel_pstate_internal_gov_init(void)
+{
+	unsigned int i = 0;
+	struct internal_governor *gov;
+	char *avail_gov;
+
+	gov = xzalloc(struct internal_governor);
+	if (!gov)
+		return NULL;
+	avail_gov = xzalloc_array(char,
+			INTEL_PSTATE_GOV_NUM * CPUFREQ_NAME_LEN);
+	if (!avail_gov)
+		return NULL;
+
+	gov->avail_gov = avail_gov;
+
+	i += scnprintf(&avail_gov[0], CPUFREQ_NAME_LEN, "%s ", "performance");
+	i += scnprintf(&avail_gov[i], CPUFREQ_NAME_LEN, "%s ", "powersave");
+	i += scnprintf(&avail_gov[i], CPUFREQ_NAME_LEN, "%s ", "userspace");
+	i += scnprintf(&avail_gov[i], CPUFREQ_NAME_LEN, "%s ", "ondemand");
+	avail_gov[i-1] = '\0';
+	gov->gov_num = INTEL_PSTATE_GOV_NUM;
+	gov->cur_gov = INTERNAL_GOV_ONDEMAND;
+	return gov;
+}
+
+static int intel_pstate_cpu_setup(struct cpufreq_policy *policy)
+{
+	struct cpudata *cpu;
+	struct perf_limits *limits = &policy->limits;
+	int rc;
+
+	rc = intel_pstate_init_cpu(policy->cpu);
+	if (rc)
+		return rc;
+
+	policy->internal_gov = intel_pstate_internal_gov_init();
+	if (!policy->internal_gov)
+		return -ENOMEM;
+
+	cpu = all_cpu_data[policy->cpu];
+	policy->min = cpu->pstate.min_pstate * cpu->pstate.scaling;
+	policy->max = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
+
+	/* cpuinfo and default policy values */
+	policy->cpuinfo.min_freq =
+		cpu->pstate.min_pstate * cpu->pstate.scaling;
+	policy->cpuinfo.max_freq =
+		cpu->pstate.turbo_pstate * cpu->pstate.scaling;
+	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
+	cpumask_set_cpu(policy->cpu, policy->cpus);
+
+	limits->no_turbo = 0;
+	limits->turbo_disabled = 0;
+	limits->turbo_pct = get_turbo_pct(cpu);
+	limits->min_policy_pct =
+		(policy->min * 100) / policy->cpuinfo.max_freq;
+	limits->min_policy_pct =
+		clamp_t(uint32_t, limits->min_policy_pct, 0, 100);
+	limits->max_policy_pct =
+		(policy->max * 100) / policy->cpuinfo.max_freq;
+	limits->max_policy_pct =
+		clamp_t(uint32_t, limits->max_policy_pct, 0, 100);
+	limits->max_perf_pct   = limits->max_policy_pct;
+	limits->min_perf_pct   = limits->min_policy_pct;
+
+	return 0;
+}
+
+static struct cpufreq_driver intel_pstate_driver = {
+	.verify	   = intel_pstate_verify_policy,
+	.setpolicy	= intel_pstate_set_policy,
+	.init		 = intel_pstate_cpu_setup,
+	.exit		 = intel_pstate_cpu_exit,
+	.update	   = intel_pstate_turbo_update,
+	.name		 = "intel_pstate",
+};
+
+static int intel_pstate_msrs_not_valid(void)
+{
+	if (!pstate_funcs.get_max() ||
+		!pstate_funcs.get_min() ||
+		!pstate_funcs.get_turbo())
+		return -ENODEV;
+
+	return 0;
+}
+
+static void __init copy_pid_params(struct pstate_adjust_policy *policy)
+{
+	pid_params.sample_rate_ms = policy->sample_rate_ms;
+	pid_params.p_gain_pct = policy->p_gain_pct;
+	pid_params.i_gain_pct = policy->i_gain_pct;
+	pid_params.d_gain_pct = policy->d_gain_pct;
+	pid_params.deadband = policy->deadband;
+	pid_params.setpoint = policy->setpoint;
+}
+
+static void __init copy_cpu_funcs(struct pstate_funcs *funcs)
+{
+	pstate_funcs.get_max   = funcs->get_max;
+	pstate_funcs.get_min   = funcs->get_min;
+	pstate_funcs.get_turbo = funcs->get_turbo;
+	pstate_funcs.get_scaling = funcs->get_scaling;
+	pstate_funcs.set	   = funcs->set;
+	pstate_funcs.get_vid   = funcs->get_vid;
+}
+
+int __init intel_pstate_init(void)
+{
+	int cpu, rc = 0;
+	const struct x86_cpu_id *id;
+	struct cpu_defaults *cpu_info;
+
+	id = x86_match_cpu(intel_pstate_cpu_ids);
+	if (!id)
+		return -ENODEV;
+
+	cpu_info = (struct cpu_defaults *)id->driver_data;
+
+	copy_pid_params(&cpu_info->pid_policy);
+	copy_cpu_funcs(&cpu_info->funcs);
+
+	if (intel_pstate_msrs_not_valid())
+		return -ENODEV;
+
+	all_cpu_data = xzalloc_array(struct cpudata *, NR_CPUS);
+	if (!all_cpu_data)
+		return -ENOMEM;
+
+	rc = cpufreq_register_driver(&intel_pstate_driver);
+	if (rc)
+		goto out;
+
+	return rc;
+out:
+	for_each_online_cpu(cpu) {
+		if (all_cpu_data[cpu]) {
+			kill_timer(&all_cpu_data[cpu]->timer);
+			xfree(all_cpu_data[cpu]);
+		}
+	}
+	xfree(all_cpu_data);
+	return -ENODEV;
+}
diff --git a/xen/include/asm-x86/cpufreq.h b/xen/include/asm-x86/cpufreq.h
new file mode 100644
index 0000000..94410f8
--- /dev/null
+++ b/xen/include/asm-x86/cpufreq.h
@@ -0,0 +1,34 @@
+#ifndef _ASM_X86_CPUFREQ_H
+#define _ASM_X86_CPUFREQ_H
+
+/*
+ *  Copyright (C) 2015 Wei Wang <wei.w.wang@intel.com>
+ *
+ * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ */
+
+extern int intel_pstate_init(void);
+
+/*
+ * Maximum transition latency is in nanoseconds - if it's unknown,
+ * CPUFREQ_ETERNAL shall be used.
+ */
+#define CPUFREQ_ETERNAL        (-1)
+
+#endif /* _ASM_X86_CPUFREQ_H */
diff --git a/xen/include/asm-x86/msr-index.h b/xen/include/asm-x86/msr-index.h
index 83f2f70..57945d9 100644
--- a/xen/include/asm-x86/msr-index.h
+++ b/xen/include/asm-x86/msr-index.h
@@ -52,6 +52,8 @@
 #define MSR_IA32_MCG_STATUS		0x0000017a
 #define MSR_IA32_MCG_CTL		0x0000017b
 
+#define MSR_NHM_TURBO_RATIO_LIMIT	0x000001ad
+
 #define MSR_IA32_PEBS_ENABLE		0x000003f1
 #define MSR_IA32_DS_AREA		0x00000600
 #define MSR_IA32_PERF_CAPABILITIES	0x00000345
@@ -319,6 +321,7 @@
 #define MSR_IA32_MISC_ENABLE_MONITOR_ENABLE (1<<18)
 #define MSR_IA32_MISC_ENABLE_LIMIT_CPUID  (1<<22)
 #define MSR_IA32_MISC_ENABLE_XTPR_DISABLE (1<<23)
+#define MSR_IA32_MISC_ENABLE_TURBO_DISABLE (1ULL<<38)
 
 #define MSR_IA32_TSC_DEADLINE		0x000006E0
 #define MSR_IA32_ENERGY_PERF_BIAS	0x000001b0
-- 
1.9.1

Re: [PATCH v4 07/11] x86/intel_pstate: the main boby of the intel_pstate driver

[Jan Beulich]

>>> On 25.06.15 at 13:16, <wei.w.wang@intel.com> wrote:
> +int __init intel_pstate_init(void)
> +{
> +	int cpu, rc = 0;
> +	const struct x86_cpu_id *id;
> +	struct cpu_defaults *cpu_info;
> +
> +	id = x86_match_cpu(intel_pstate_cpu_ids);
> +	if (!id)
> +		return -ENODEV;
> +
> +	cpu_info = (struct cpu_defaults *)id->driver_data;
> +
> +	copy_pid_params(&cpu_info->pid_policy);
> +	copy_cpu_funcs(&cpu_info->funcs);
> +
> +	if (intel_pstate_msrs_not_valid())
> +		return -ENODEV;
> +
> +	all_cpu_data = xzalloc_array(struct cpudata *, NR_CPUS);
> +	if (!all_cpu_data)
> +		return -ENOMEM;
> +
> +	rc = cpufreq_register_driver(&intel_pstate_driver);
> +	if (rc)
> +		goto out;
> +
> +	return rc;
> +out:
> +	for_each_online_cpu(cpu) {
> +		if (all_cpu_data[cpu]) {
> +			kill_timer(&all_cpu_data[cpu]->timer);
> +			xfree(all_cpu_data[cpu]);
> +		}
> +	}

I have a hard time seeing where in this function the setup
happens that is being undone here (keeping in mind that the
notifier registration inside cpufreq_register_driver() doesn't
actually call the notifier function).

And then, looking at the diff between this and what Linux 4.2-rc3
has (which admittedly looks a little newer than what you sent, so
I already subtract some of the delta), it is significantly larger than
the source file itself. That surely doesn't suggest a clone-with-
minimal-delta. Yet as said before - either you do that, or you
accept us picking at things you inherited from Linux.

Jan

Re: [PATCH v4 07/11] x86/intel_pstate: the main boby of the intel_pstate driver

[Wang, Wei W]

On 24/07/2015 21:54,  Jan Beulich wrote:
> >>> On 25.06.15 at 13:16, <wei.w.wang@intel.com> wrote:
> > +int __init intel_pstate_init(void)
> > +{
> > +	int cpu, rc = 0;
> > +	const struct x86_cpu_id *id;
> > +	struct cpu_defaults *cpu_info;
> > +
> > +	id = x86_match_cpu(intel_pstate_cpu_ids);
> > +	if (!id)
> > +		return -ENODEV;
> > +
> > +	cpu_info = (struct cpu_defaults *)id->driver_data;
> > +
> > +	copy_pid_params(&cpu_info->pid_policy);
> > +	copy_cpu_funcs(&cpu_info->funcs);
> > +
> > +	if (intel_pstate_msrs_not_valid())
> > +		return -ENODEV;
> > +
> > +	all_cpu_data = xzalloc_array(struct cpudata *, NR_CPUS);
> > +	if (!all_cpu_data)
> > +		return -ENOMEM;
> > +
> > +	rc = cpufreq_register_driver(&intel_pstate_driver);
> > +	if (rc)
> > +		goto out;
> > +
> > +	return rc;
> > +out:
> > +	for_each_online_cpu(cpu) {
> > +		if (all_cpu_data[cpu]) {
> > +			kill_timer(&all_cpu_data[cpu]->timer);
> > +			xfree(all_cpu_data[cpu]);
> > +		}
> > +	}
> 
> I have a hard time seeing where in this function the setup happens that is
> being undone here (keeping in mind that the notifier registration inside
> cpufreq_register_driver() doesn't actually call the notifier function).
> 
> And then, looking at the diff between this and what Linux 4.2-rc3 has (which
> admittedly looks a little newer than what you sent, so I already subtract
> some of the delta), it is significantly larger than the source file itself. That
> surely doesn't suggest a clone-with- minimal-delta. Yet as said before - either
> you do that, or you accept us picking at things you inherited from Linux.

I think it's better to choose the latter - picking out things that are useful for us from Linux.
Can you please take a look this patch and summarize the comments? Thanks.

Best,
Wei

Re: [PATCH v4 07/11] x86/intel_pstate: the main boby of the intel_pstate driver

[Jan Beulich]

>>> On 27.07.15 at 11:30, <wei.w.wang@intel.com> wrote:
> On 24/07/2015 21:54,  Jan Beulich wrote:
>> >>> On 25.06.15 at 13:16, <wei.w.wang@intel.com> wrote:
>> > +int __init intel_pstate_init(void)
>> > +{
>> > +	int cpu, rc = 0;
>> > +	const struct x86_cpu_id *id;
>> > +	struct cpu_defaults *cpu_info;
>> > +
>> > +	id = x86_match_cpu(intel_pstate_cpu_ids);
>> > +	if (!id)
>> > +		return -ENODEV;
>> > +
>> > +	cpu_info = (struct cpu_defaults *)id->driver_data;
>> > +
>> > +	copy_pid_params(&cpu_info->pid_policy);
>> > +	copy_cpu_funcs(&cpu_info->funcs);
>> > +
>> > +	if (intel_pstate_msrs_not_valid())
>> > +		return -ENODEV;
>> > +
>> > +	all_cpu_data = xzalloc_array(struct cpudata *, NR_CPUS);
>> > +	if (!all_cpu_data)
>> > +		return -ENOMEM;
>> > +
>> > +	rc = cpufreq_register_driver(&intel_pstate_driver);
>> > +	if (rc)
>> > +		goto out;
>> > +
>> > +	return rc;
>> > +out:
>> > +	for_each_online_cpu(cpu) {
>> > +		if (all_cpu_data[cpu]) {
>> > +			kill_timer(&all_cpu_data[cpu]->timer);
>> > +			xfree(all_cpu_data[cpu]);
>> > +		}
>> > +	}
>> 
>> I have a hard time seeing where in this function the setup happens that is
>> being undone here (keeping in mind that the notifier registration inside
>> cpufreq_register_driver() doesn't actually call the notifier function).
>> 
>> And then, looking at the diff between this and what Linux 4.2-rc3 has (which
>> admittedly looks a little newer than what you sent, so I already subtract
>> some of the delta), it is significantly larger than the source file itself. 
> That
>> surely doesn't suggest a clone-with- minimal-delta. Yet as said before - either
>> you do that, or you accept us picking at things you inherited from Linux.
> 
> I think it's better to choose the latter - picking out things that are useful 
> for us from Linux.
> Can you please take a look this patch and summarize the comments? Thanks.

I'm sorry, but for a first round I'd rather expect _you_ to go through
the code you intend to add and spot possible problems. Only then, on
a submission where you state that you did so, would I want to invest
time in sanity checking things.

And then I hope you realize that the clone-with-minimal-delta would
have benefits on the maintenance side going forward (fewer manual
adjustments needed due to non-applying Linux side changes).

Jan