Re: [PATCH v7 3/3] PM: Introduce Intel PowerClamp Driver

From: Zhang Rui
Date: Thu Jan 17 2013 - 09:35:30 EST


On Wed, 2013-01-16 at 05:11 -0800, Jacob Pan wrote:
> Intel PowerClamp driver performs synchronized idle injection across
> all online CPUs. The goal is to maintain a given package level C-state
> ratio.
>
> Compared to other throttling methods already exist in the kernel,
> such as ACPI PAD (taking CPUs offline) and clock modulation, this is often
> more efficient in terms of performance per watt.
>
> Please refer to Documentation/thermal/intel_powerclamp.txt for more details.
>
> Signed-off-by: Arjan van de Ven <arjan@xxxxxxxxxxxxxxx>
> Signed-off-by: Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>

applied to thermal -next.

thanks,
rui

> ---
> Documentation/thermal/intel_powerclamp.txt | 307 +++++++++++
> drivers/thermal/Kconfig | 10 +
> drivers/thermal/Makefile | 2 +
> drivers/thermal/intel_powerclamp.c | 788 ++++++++++++++++++++++++++++
> 4 files changed, 1107 insertions(+)
> create mode 100644 Documentation/thermal/intel_powerclamp.txt
> create mode 100644 drivers/thermal/intel_powerclamp.c
>
> diff --git a/Documentation/thermal/intel_powerclamp.txt b/Documentation/thermal/intel_powerclamp.txt
> new file mode 100644
> index 0000000..332de4a
> --- /dev/null
> +++ b/Documentation/thermal/intel_powerclamp.txt
> @@ -0,0 +1,307 @@
> + =======================
> + INTEL POWERCLAMP DRIVER
> + =======================
> +By: Arjan van de Ven <arjan@xxxxxxxxxxxxxxx>
> + Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>
> +
> +Contents:
> + (*) Introduction
> + - Goals and Objectives
> +
> + (*) Theory of Operation
> + - Idle Injection
> + - Calibration
> +
> + (*) Performance Analysis
> + - Effectiveness and Limitations
> + - Power vs Performance
> + - Scalability
> + - Calibration
> + - Comparison with Alternative Techniques
> +
> + (*) Usage and Interfaces
> + - Generic Thermal Layer (sysfs)
> + - Kernel APIs (TBD)
> +
> +============
> +INTRODUCTION
> +============
> +
> +Consider the situation where a systemâs power consumption must be
> +reduced at runtime, due to power budget, thermal constraint, or noise
> +level, and where active cooling is not preferred. Software managed
> +passive power reduction must be performed to prevent the hardware
> +actions that are designed for catastrophic scenarios.
> +
> +Currently, P-states, T-states (clock modulation), and CPU offlining
> +are used for CPU throttling.
> +
> +On Intel CPUs, C-states provide effective power reduction, but so far
> +theyâre only used opportunistically, based on workload. With the
> +development of intel_powerclamp driver, the method of synchronizing
> +idle injection across all online CPU threads was introduced. The goal
> +is to achieve forced and controllable C-state residency.
> +
> +Test/Analysis has been made in the areas of power, performance,
> +scalability, and user experience. In many cases, clear advantage is
> +shown over taking the CPU offline or modulating the CPU clock.
> +
> +
> +===================
> +THEORY OF OPERATION
> +===================
> +
> +Idle Injection
> +--------------
> +
> +On modern Intel processors (Nehalem or later), package level C-state
> +residency is available in MSRs, thus also available to the kernel.
> +
> +These MSRs are:
> + #define MSR_PKG_C2_RESIDENCY 0x60D
> + #define MSR_PKG_C3_RESIDENCY 0x3F8
> + #define MSR_PKG_C6_RESIDENCY 0x3F9
> + #define MSR_PKG_C7_RESIDENCY 0x3FA
> +
> +If the kernel can also inject idle time to the system, then a
> +closed-loop control system can be established that manages package
> +level C-state. The intel_powerclamp driver is conceived as such a
> +control system, where the target set point is a user-selected idle
> +ratio (based on power reduction), and the error is the difference
> +between the actual package level C-state residency ratio and the target idle
> +ratio.
> +
> +Injection is controlled by high priority kernel threads, spawned for
> +each online CPU.
> +
> +These kernel threads, with SCHED_FIFO class, are created to perform
> +clamping actions of controlled duty ratio and duration. Each per-CPU
> +thread synchronizes its idle time and duration, based on the rounding
> +of jiffies, so accumulated errors can be prevented to avoid a jittery
> +effect. Threads are also bound to the CPU such that they cannot be
> +migrated, unless the CPU is taken offline. In this case, threads
> +belong to the offlined CPUs will be terminated immediately.
> +
> +Running as SCHED_FIFO and relatively high priority, also allows such
> +scheme to work for both preemptable and non-preemptable kernels.
> +Alignment of idle time around jiffies ensures scalability for HZ
> +values. This effect can be better visualized using a Perf timechart.
> +The following diagram shows the behavior of kernel thread
> +kidle_inject/cpu. During idle injection, it runs monitor/mwait idle
> +for a given "duration", then relinquishes the CPU to other tasks,
> +until the next time interval.
> +
> +The NOHZ schedule tick is disabled during idle time, but interrupts
> +are not masked. Tests show that the extra wakeups from scheduler tick
> +have a dramatic impact on the effectiveness of the powerclamp driver
> +on large scale systems (Westmere system with 80 processors).
> +
> +CPU0
> + ____________ ____________
> +kidle_inject/0 | sleep | mwait | sleep |
> + _________| |________| |_______
> + duration
> +CPU1
> + ____________ ____________
> +kidle_inject/1 | sleep | mwait | sleep |
> + _________| |________| |_______
> + ^
> + |
> + |
> + roundup(jiffies, interval)
> +
> +Only one CPU is allowed to collect statistics and update global
> +control parameters. This CPU is referred to as the controlling CPU in
> +this document. The controlling CPU is elected at runtime, with a
> +policy that favors BSP, taking into account the possibility of a CPU
> +hot-plug.
> +
> +In terms of dynamics of the idle control system, package level idle
> +time is considered largely as a non-causal system where its behavior
> +cannot be based on the past or current input. Therefore, the
> +intel_powerclamp driver attempts to enforce the desired idle time
> +instantly as given input (target idle ratio). After injection,
> +powerclamp moniors the actual idle for a given time window and adjust
> +the next injection accordingly to avoid over/under correction.
> +
> +When used in a causal control system, such as a temperature control,
> +it is up to the user of this driver to implement algorithms where
> +past samples and outputs are included in the feedback. For example, a
> +PID-based thermal controller can use the powerclamp driver to
> +maintain a desired target temperature, based on integral and
> +derivative gains of the past samples.
> +
> +
> +
> +Calibration
> +-----------
> +During scalability testing, it is observed that synchronized actions
> +among CPUs become challenging as the number of cores grows. This is
> +also true for the ability of a system to enter package level C-states.
> +
> +To make sure the intel_powerclamp driver scales well, online
> +calibration is implemented. The goals for doing such a calibration
> +are:
> +
> +a) determine the effective range of idle injection ratio
> +b) determine the amount of compensation needed at each target ratio
> +
> +Compensation to each target ratio consists of two parts:
> +
> + a) steady state error compensation
> + This is to offset the error occurring when the system can
> + enter idle without extra wakeups (such as external interrupts).
> +
> + b) dynamic error compensation
> + When an excessive amount of wakeups occurs during idle, an
> + additional idle ratio can be added to quiet interrupts, by
> + slowing down CPU activities.
> +
> +A debugfs file is provided for the user to examine compensation
> +progress and results, such as on a Westmere system.
> +[jacob@nex01 ~]$ cat
> +/sys/kernel/debug/intel_powerclamp/powerclamp_calib
> +controlling cpu: 0
> +pct confidence steady dynamic (compensation)
> +0 0 0 0
> +1 1 0 0
> +2 1 1 0
> +3 3 1 0
> +4 3 1 0
> +5 3 1 0
> +6 3 1 0
> +7 3 1 0
> +8 3 1 0
> +...
> +30 3 2 0
> +31 3 2 0
> +32 3 1 0
> +33 3 2 0
> +34 3 1 0
> +35 3 2 0
> +36 3 1 0
> +37 3 2 0
> +38 3 1 0
> +39 3 2 0
> +40 3 3 0
> +41 3 1 0
> +42 3 2 0
> +43 3 1 0
> +44 3 1 0
> +45 3 2 0
> +46 3 3 0
> +47 3 0 0
> +48 3 2 0
> +49 3 3 0
> +
> +Calibration occurs during runtime. No offline method is available.
> +Steady state compensation is used only when confidence levels of all
> +adjacent ratios have reached satisfactory level. A confidence level
> +is accumulated based on clean data collected at runtime. Data
> +collected during a period without extra interrupts is considered
> +clean.
> +
> +To compensate for excessive amounts of wakeup during idle, additional
> +idle time is injected when such a condition is detected. Currently,
> +we have a simple algorithm to double the injection ratio. A possible
> +enhancement might be to throttle the offending IRQ, such as delaying
> +EOI for level triggered interrupts. But it is a challenge to be
> +non-intrusive to the scheduler or the IRQ core code.
> +
> +
> +CPU Online/Offline
> +------------------
> +Per-CPU kernel threads are started/stopped upon receiving
> +notifications of CPU hotplug activities. The intel_powerclamp driver
> +keeps track of clamping kernel threads, even after they are migrated
> +to other CPUs, after a CPU offline event.
> +
> +
> +=====================
> +Performance Analysis
> +=====================
> +This section describes the general performance data collected on
> +multiple systems, including Westmere (80P) and Ivy Bridge (4P, 8P).
> +
> +Effectiveness and Limitations
> +-----------------------------
> +The maximum range that idle injection is allowed is capped at 50
> +percent. As mentioned earlier, since interrupts are allowed during
> +forced idle time, excessive interrupts could result in less
> +effectiveness. The extreme case would be doing a ping -f to generated
> +flooded network interrupts without much CPU acknowledgement. In this
> +case, little can be done from the idle injection threads. In most
> +normal cases, such as scp a large file, applications can be throttled
> +by the powerclamp driver, since slowing down the CPU also slows down
> +network protocol processing, which in turn reduces interrupts.
> +
> +When control parameters change at runtime by the controlling CPU, it
> +may take an additional period for the rest of the CPUs to catch up
> +with the changes. During this time, idle injection is out of sync,
> +thus not able to enter package C- states at the expected ratio. But
> +this effect is minor, in that in most cases change to the target
> +ratio is updated much less frequently than the idle injection
> +frequency.
> +
> +Scalability
> +-----------
> +Tests also show a minor, but measurable, difference between the 4P/8P
> +Ivy Bridge system and the 80P Westmere server under 50% idle ratio.
> +More compensation is needed on Westmere for the same amount of
> +target idle ratio. The compensation also increases as the idle ratio
> +gets larger. The above reason constitutes the need for the
> +calibration code.
> +
> +On the IVB 8P system, compared to an offline CPU, powerclamp can
> +achieve up to 40% better performance per watt. (measured by a spin
> +counter summed over per CPU counting threads spawned for all running
> +CPUs).
> +
> +====================
> +Usage and Interfaces
> +====================
> +The powerclamp driver is registered to the generic thermal layer as a
> +cooling device. Currently, itâs not bound to any thermal zones.
> +
> +jacob@chromoly:/sys/class/thermal/cooling_device14$ grep . *
> +cur_state:0
> +max_state:50
> +type:intel_powerclamp
> +
> +Example usage:
> +- To inject 25% idle time
> +$ sudo sh -c "echo 25 > /sys/class/thermal/cooling_device80/cur_state
> +"
> +
> +If the system is not busy and has more than 25% idle time already,
> +then the powerclamp driver will not start idle injection. Using Top
> +will not show idle injection kernel threads.
> +
> +If the system is busy (spin test below) and has less than 25% natural
> +idle time, powerclamp kernel threads will do idle injection, which
> +appear running to the scheduler. But the overall system idle is still
> +reflected. In this example, 24.1% idle is shown. This helps the
> +system admin or user determine the cause of slowdown, when a
> +powerclamp driver is in action.
> +
> +
> +Tasks: 197 total, 1 running, 196 sleeping, 0 stopped, 0 zombie
> +Cpu(s): 71.2%us, 4.7%sy, 0.0%ni, 24.1%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
> +Mem: 3943228k total, 1689632k used, 2253596k free, 74960k buffers
> +Swap: 4087804k total, 0k used, 4087804k free, 945336k cached
> +
> + PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> + 3352 jacob 20 0 262m 644 428 S 286 0.0 0:17.16 spin
> + 3341 root -51 0 0 0 0 D 25 0.0 0:01.62 kidle_inject/0
> + 3344 root -51 0 0 0 0 D 25 0.0 0:01.60 kidle_inject/3
> + 3342 root -51 0 0 0 0 D 25 0.0 0:01.61 kidle_inject/1
> + 3343 root -51 0 0 0 0 D 25 0.0 0:01.60 kidle_inject/2
> + 2935 jacob 20 0 696m 125m 35m S 5 3.3 0:31.11 firefox
> + 1546 root 20 0 158m 20m 6640 S 3 0.5 0:26.97 Xorg
> + 2100 jacob 20 0 1223m 88m 30m S 3 2.3 0:23.68 compiz
> +
> +Tests have shown that by using the powerclamp driver as a cooling
> +device, a PID based userspace thermal controller can manage to
> +control CPU temperature effectively, when no other thermal influence
> +is added. For example, a UltraBook user can compile the kernel under
> +certain temperature (below most active trip points).
> diff --git a/drivers/thermal/Kconfig b/drivers/thermal/Kconfig
> index c2c77d1..7d90ab8 100644
> --- a/drivers/thermal/Kconfig
> +++ b/drivers/thermal/Kconfig
> @@ -122,4 +122,14 @@ config DB8500_CPUFREQ_COOLING
> bound cpufreq cooling device turns active to set CPU frequency low to
> cool down the CPU.
>
> +config INTEL_POWERCLAMP
> + tristate "Intel PowerClamp idle injection driver"
> + depends on THERMAL
> + depends on X86
> + depends on CPU_SUP_INTEL
> + help
> + Enable this to enable Intel PowerClamp idle injection driver. This
> + enforce idle time which results in more package C-state residency. The
> + user interface is exposed via generic thermal framework.
> +
> endif
> diff --git a/drivers/thermal/Makefile b/drivers/thermal/Makefile
> index d8da683..574f5f5 100644
> --- a/drivers/thermal/Makefile
> +++ b/drivers/thermal/Makefile
> @@ -18,3 +18,5 @@ obj-$(CONFIG_RCAR_THERMAL) += rcar_thermal.o
> obj-$(CONFIG_EXYNOS_THERMAL) += exynos_thermal.o
> obj-$(CONFIG_DB8500_THERMAL) += db8500_thermal.o
> obj-$(CONFIG_DB8500_CPUFREQ_COOLING) += db8500_cpufreq_cooling.o
> +obj-$(CONFIG_INTEL_POWERCLAMP) += intel_powerclamp.o
> +
> diff --git a/drivers/thermal/intel_powerclamp.c b/drivers/thermal/intel_powerclamp.c
> new file mode 100644
> index 0000000..81ebf87
> --- /dev/null
> +++ b/drivers/thermal/intel_powerclamp.c
> @@ -0,0 +1,788 @@
> +/*
> + * intel_powerclamp.c - package c-state idle injection
> + *
> + * Copyright (c) 2012, Intel Corporation.
> + *
> + * Authors:
> + * Arjan van de Ven <arjan@xxxxxxxxxxxxxxx>
> + * Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms and conditions of the GNU General Public License,
> + * version 2, as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope 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.,
> + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
> + *
> + *
> + * TODO:
> + * 1. better handle wakeup from external interrupts, currently a fixed
> + * compensation is added to clamping duration when excessive amount
> + * of wakeups are observed during idle time. the reason is that in
> + * case of external interrupts without need for ack, clamping down
> + * cpu in non-irq context does not reduce irq. for majority of the
> + * cases, clamping down cpu does help reduce irq as well, we should
> + * be able to differenciate the two cases and give a quantitative
> + * solution for the irqs that we can control. perhaps based on
> + * get_cpu_iowait_time_us()
> + *
> + * 2. synchronization with other hw blocks
> + *
> + *
> + */
> +
> +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> +
> +#include <linux/module.h>
> +#include <linux/kernel.h>
> +#include <linux/delay.h>
> +#include <linux/kthread.h>
> +#include <linux/freezer.h>
> +#include <linux/cpu.h>
> +#include <linux/thermal.h>
> +#include <linux/slab.h>
> +#include <linux/tick.h>
> +#include <linux/debugfs.h>
> +#include <linux/seq_file.h>
> +
> +#include <asm/nmi.h>
> +#include <asm/msr.h>
> +#include <asm/mwait.h>
> +#include <asm/cpu_device_id.h>
> +#include <asm/idle.h>
> +#include <asm/hardirq.h>
> +
> +#define MAX_TARGET_RATIO (50U)
> +/* For each undisturbed clamping period (no extra wake ups during idle time),
> + * we increment the confidence counter for the given target ratio.
> + * CONFIDENCE_OK defines the level where runtime calibration results are
> + * valid.
> + */
> +#define CONFIDENCE_OK (3)
> +/* Default idle injection duration, driver adjust sleep time to meet target
> + * idle ratio. Similar to frequency modulation.
> + */
> +#define DEFAULT_DURATION_JIFFIES (6)
> +
> +static unsigned int target_mwait;
> +static struct dentry *debug_dir;
> +
> +/* user selected target */
> +static unsigned int set_target_ratio;
> +static unsigned int current_ratio;
> +static bool should_skip;
> +static bool reduce_irq;
> +static atomic_t idle_wakeup_counter;
> +static unsigned int control_cpu; /* The cpu assigned to collect stat and update
> + * control parameters. default to BSP but BSP
> + * can be offlined.
> + */
> +static bool clamping;
> +
> +
> +static struct task_struct __percpu **powerclamp_thread;
> +static struct thermal_cooling_device *cooling_dev;
> +static unsigned long *cpu_clamping_mask; /* bit map for tracking per cpu
> + * clamping thread
> + */
> +
> +static unsigned int duration;
> +static unsigned int pkg_cstate_ratio_cur;
> +static unsigned int window_size;
> +
> +static int duration_set(const char *arg, const struct kernel_param *kp)
> +{
> + int ret = 0;
> + unsigned long new_duration;
> +
> + ret = kstrtoul(arg, 10, &new_duration);
> + if (ret)
> + goto exit;
> + if (new_duration > 25 || new_duration < 6) {
> + pr_err("Out of recommended range %lu, between 6-25ms\n",
> + new_duration);
> + ret = -EINVAL;
> + }
> +
> + duration = clamp(new_duration, 6ul, 25ul);
> + smp_mb();
> +
> +exit:
> +
> + return ret;
> +}
> +
> +static struct kernel_param_ops duration_ops = {
> + .set = duration_set,
> + .get = param_get_int,
> +};
> +
> +
> +module_param_cb(duration, &duration_ops, &duration, 0644);
> +MODULE_PARM_DESC(duration, "forced idle time for each attempt in msec.");
> +
> +struct powerclamp_calibration_data {
> + unsigned long confidence; /* used for calibration, basically a counter
> + * gets incremented each time a clamping
> + * period is completed without extra wakeups
> + * once that counter is reached given level,
> + * compensation is deemed usable.
> + */
> + unsigned long steady_comp; /* steady state compensation used when
> + * no extra wakeups occurred.
> + */
> + unsigned long dynamic_comp; /* compensate excessive wakeup from idle
> + * mostly from external interrupts.
> + */
> +};
> +
> +static struct powerclamp_calibration_data cal_data[MAX_TARGET_RATIO];
> +
> +static int window_size_set(const char *arg, const struct kernel_param *kp)
> +{
> + int ret = 0;
> + unsigned long new_window_size;
> +
> + ret = kstrtoul(arg, 10, &new_window_size);
> + if (ret)
> + goto exit_win;
> + if (new_window_size > 10 || new_window_size < 2) {
> + pr_err("Out of recommended window size %lu, between 2-10\n",
> + new_window_size);
> + ret = -EINVAL;
> + }
> +
> + window_size = clamp(new_window_size, 2ul, 10ul);
> + smp_mb();
> +
> +exit_win:
> +
> + return ret;
> +}
> +
> +static struct kernel_param_ops window_size_ops = {
> + .set = window_size_set,
> + .get = param_get_int,
> +};
> +
> +module_param_cb(window_size, &window_size_ops, &window_size, 0644);
> +MODULE_PARM_DESC(window_size, "sliding window in number of clamping cycles\n"
> + "\tpowerclamp controls idle ratio within this window. larger\n"
> + "\twindow size results in slower response time but more smooth\n"
> + "\tclamping results. default to 2.");
> +
> +static void find_target_mwait(void)
> +{
> + unsigned int eax, ebx, ecx, edx;
> + unsigned int highest_cstate = 0;
> + unsigned int highest_subcstate = 0;
> + int i;
> +
> + if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
> + return;
> +
> + cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
> +
> + if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
> + !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
> + return;
> +
> + edx >>= MWAIT_SUBSTATE_SIZE;
> + for (i = 0; i < 7 && edx; i++, edx >>= MWAIT_SUBSTATE_SIZE) {
> + if (edx & MWAIT_SUBSTATE_MASK) {
> + highest_cstate = i;
> + highest_subcstate = edx & MWAIT_SUBSTATE_MASK;
> + }
> + }
> + target_mwait = (highest_cstate << MWAIT_SUBSTATE_SIZE) |
> + (highest_subcstate - 1);
> +
> +}
> +
> +static u64 pkg_state_counter(void)
> +{
> + u64 val;
> + u64 count = 0;
> +
> + static bool skip_c2;
> + static bool skip_c3;
> + static bool skip_c6;
> + static bool skip_c7;
> +
> + if (!skip_c2) {
> + if (!rdmsrl_safe(MSR_PKG_C2_RESIDENCY, &val))
> + count += val;
> + else
> + skip_c2 = true;
> + }
> +
> + if (!skip_c3) {
> + if (!rdmsrl_safe(MSR_PKG_C3_RESIDENCY, &val))
> + count += val;
> + else
> + skip_c3 = true;
> + }
> +
> + if (!skip_c6) {
> + if (!rdmsrl_safe(MSR_PKG_C6_RESIDENCY, &val))
> + count += val;
> + else
> + skip_c6 = true;
> + }
> +
> + if (!skip_c7) {
> + if (!rdmsrl_safe(MSR_PKG_C7_RESIDENCY, &val))
> + count += val;
> + else
> + skip_c7 = true;
> + }
> +
> + return count;
> +}
> +
> +static void noop_timer(unsigned long foo)
> +{
> + /* empty... just the fact that we get the interrupt wakes us up */
> +}
> +
> +static unsigned int get_compensation(int ratio)
> +{
> + unsigned int comp = 0;
> +
> + /* we only use compensation if all adjacent ones are good */
> + if (ratio == 1 &&
> + cal_data[ratio].confidence >= CONFIDENCE_OK &&
> + cal_data[ratio + 1].confidence >= CONFIDENCE_OK &&
> + cal_data[ratio + 2].confidence >= CONFIDENCE_OK) {
> + comp = (cal_data[ratio].steady_comp +
> + cal_data[ratio + 1].steady_comp +
> + cal_data[ratio + 2].steady_comp) / 3;
> + } else if (ratio == MAX_TARGET_RATIO - 1 &&
> + cal_data[ratio].confidence >= CONFIDENCE_OK &&
> + cal_data[ratio - 1].confidence >= CONFIDENCE_OK &&
> + cal_data[ratio - 2].confidence >= CONFIDENCE_OK) {
> + comp = (cal_data[ratio].steady_comp +
> + cal_data[ratio - 1].steady_comp +
> + cal_data[ratio - 2].steady_comp) / 3;
> + } else if (cal_data[ratio].confidence >= CONFIDENCE_OK &&
> + cal_data[ratio - 1].confidence >= CONFIDENCE_OK &&
> + cal_data[ratio + 1].confidence >= CONFIDENCE_OK) {
> + comp = (cal_data[ratio].steady_comp +
> + cal_data[ratio - 1].steady_comp +
> + cal_data[ratio + 1].steady_comp) / 3;
> + }
> +
> + /* REVISIT: simple penalty of double idle injection */
> + if (reduce_irq)
> + comp = ratio;
> + /* do not exceed limit */
> + if (comp + ratio >= MAX_TARGET_RATIO)
> + comp = MAX_TARGET_RATIO - ratio - 1;
> +
> + return comp;
> +}
> +
> +static void adjust_compensation(int target_ratio, unsigned int win)
> +{
> + int delta;
> + struct powerclamp_calibration_data *d = &cal_data[target_ratio];
> +
> + /*
> + * adjust compensations if confidence level has not been reached or
> + * there are too many wakeups during the last idle injection period, we
> + * cannot trust the data for compensation.
> + */
> + if (d->confidence >= CONFIDENCE_OK ||
> + atomic_read(&idle_wakeup_counter) >
> + win * num_online_cpus())
> + return;
> +
> + delta = set_target_ratio - current_ratio;
> + /* filter out bad data */
> + if (delta >= 0 && delta <= (1+target_ratio/10)) {
> + if (d->steady_comp)
> + d->steady_comp =
> + roundup(delta+d->steady_comp, 2)/2;
> + else
> + d->steady_comp = delta;
> + d->confidence++;
> + }
> +}
> +
> +static bool powerclamp_adjust_controls(unsigned int target_ratio,
> + unsigned int guard, unsigned int win)
> +{
> + static u64 msr_last, tsc_last;
> + u64 msr_now, tsc_now;
> +
> + /* check result for the last window */
> + msr_now = pkg_state_counter();
> + rdtscll(tsc_now);
> +
> + /* calculate pkg cstate vs tsc ratio */
> + if (!msr_last || !tsc_last)
> + current_ratio = 1;
> + else if (tsc_now-tsc_last)
> + current_ratio = 100*(msr_now-msr_last)/
> + (tsc_now-tsc_last);
> +
> + /* update record */
> + msr_last = msr_now;
> + tsc_last = tsc_now;
> +
> + adjust_compensation(target_ratio, win);
> + /*
> + * too many external interrupts, set flag such
> + * that we can take measure later.
> + */
> + reduce_irq = atomic_read(&idle_wakeup_counter) >=
> + 2 * win * num_online_cpus();
> +
> + atomic_set(&idle_wakeup_counter, 0);
> + /* if we are above target+guard, skip */
> + return set_target_ratio + guard <= current_ratio;
> +}
> +
> +static int clamp_thread(void *arg)
> +{
> + int cpunr = (unsigned long)arg;
> + DEFINE_TIMER(wakeup_timer, noop_timer, 0, 0);
> + static const struct sched_param param = {
> + .sched_priority = MAX_USER_RT_PRIO/2,
> + };
> + unsigned int count = 0;
> + unsigned int target_ratio;
> +
> + set_bit(cpunr, cpu_clamping_mask);
> + set_freezable();
> + init_timer_on_stack(&wakeup_timer);
> + sched_setscheduler(current, SCHED_FIFO, &param);
> +
> + while (true == clamping && !kthread_should_stop() &&
> + cpu_online(cpunr)) {
> + int sleeptime;
> + unsigned long target_jiffies;
> + unsigned int guard;
> + unsigned int compensation = 0;
> + int interval; /* jiffies to sleep for each attempt */
> + unsigned int duration_jiffies = msecs_to_jiffies(duration);
> + unsigned int window_size_now;
> +
> + try_to_freeze();
> + /*
> + * make sure user selected ratio does not take effect until
> + * the next round. adjust target_ratio if user has changed
> + * target such that we can converge quickly.
> + */
> + target_ratio = set_target_ratio;
> + guard = 1 + target_ratio/20;
> + window_size_now = window_size;
> + count++;
> +
> + /*
> + * systems may have different ability to enter package level
> + * c-states, thus we need to compensate the injected idle ratio
> + * to achieve the actual target reported by the HW.
> + */
> + compensation = get_compensation(target_ratio);
> + interval = duration_jiffies*100/(target_ratio+compensation);
> +
> + /* align idle time */
> + target_jiffies = roundup(jiffies, interval);
> + sleeptime = target_jiffies - jiffies;
> + if (sleeptime <= 0)
> + sleeptime = 1;
> + schedule_timeout_interruptible(sleeptime);
> + /*
> + * only elected controlling cpu can collect stats and update
> + * control parameters.
> + */
> + if (cpunr == control_cpu && !(count%window_size_now)) {
> + should_skip =
> + powerclamp_adjust_controls(target_ratio,
> + guard, window_size_now);
> + smp_mb();
> + }
> +
> + if (should_skip)
> + continue;
> +
> + target_jiffies = jiffies + duration_jiffies;
> + mod_timer(&wakeup_timer, target_jiffies);
> + if (unlikely(local_softirq_pending()))
> + continue;
> + /*
> + * stop tick sched during idle time, interrupts are still
> + * allowed. thus jiffies are updated properly.
> + */
> + preempt_disable();
> + tick_nohz_idle_enter();
> + /* mwait until target jiffies is reached */
> + while (time_before(jiffies, target_jiffies)) {
> + unsigned long ecx = 1;
> + unsigned long eax = target_mwait;
> +
> + /*
> + * REVISIT: may call enter_idle() to notify drivers who
> + * can save power during cpu idle. same for exit_idle()
> + */
> + local_touch_nmi();
> + stop_critical_timings();
> + __monitor((void *)&current_thread_info()->flags, 0, 0);
> + cpu_relax(); /* allow HT sibling to run */
> + __mwait(eax, ecx);
> + start_critical_timings();
> + atomic_inc(&idle_wakeup_counter);
> + }
> + tick_nohz_idle_exit();
> + preempt_enable_no_resched();
> + }
> + del_timer_sync(&wakeup_timer);
> + clear_bit(cpunr, cpu_clamping_mask);
> +
> + return 0;
> +}
> +
> +/*
> + * 1 HZ polling while clamping is active, useful for userspace
> + * to monitor actual idle ratio.
> + */
> +static void poll_pkg_cstate(struct work_struct *dummy);
> +static DECLARE_DELAYED_WORK(poll_pkg_cstate_work, poll_pkg_cstate);
> +static void poll_pkg_cstate(struct work_struct *dummy)
> +{
> + static u64 msr_last;
> + static u64 tsc_last;
> + static unsigned long jiffies_last;
> +
> + u64 msr_now;
> + unsigned long jiffies_now;
> + u64 tsc_now;
> +
> + msr_now = pkg_state_counter();
> + rdtscll(tsc_now);
> + jiffies_now = jiffies;
> +
> + /* calculate pkg cstate vs tsc ratio */
> + if (!msr_last || !tsc_last)
> + pkg_cstate_ratio_cur = 1;
> + else {
> + if (tsc_now - tsc_last)
> + pkg_cstate_ratio_cur = 100 * (msr_now - msr_last)/
> + (tsc_now - tsc_last);
> + }
> +
> + /* update record */
> + msr_last = msr_now;
> + jiffies_last = jiffies_now;
> + tsc_last = tsc_now;
> +
> + if (true == clamping)
> + schedule_delayed_work(&poll_pkg_cstate_work, HZ);
> +}
> +
> +static int start_power_clamp(void)
> +{
> + unsigned long cpu;
> + struct task_struct *thread;
> +
> + /* check if pkg cstate counter is completely 0, abort in this case */
> + if (!pkg_state_counter()) {
> + pr_err("pkg cstate counter not functional, abort\n");
> + return -EINVAL;
> + }
> +
> + set_target_ratio = clamp(set_target_ratio, 0U, MAX_TARGET_RATIO);
> + /* prevent cpu hotplug */
> + get_online_cpus();
> +
> + /* prefer BSP */
> + control_cpu = 0;
> + if (!cpu_online(control_cpu))
> + control_cpu = smp_processor_id();
> +
> + clamping = true;
> + schedule_delayed_work(&poll_pkg_cstate_work, 0);
> +
> + /* start one thread per online cpu */
> + for_each_online_cpu(cpu) {
> + struct task_struct **p =
> + per_cpu_ptr(powerclamp_thread, cpu);
> +
> + thread = kthread_create_on_node(clamp_thread,
> + (void *) cpu,
> + cpu_to_node(cpu),
> + "kidle_inject/%ld", cpu);
> + /* bind to cpu here */
> + if (likely(!IS_ERR(thread))) {
> + kthread_bind(thread, cpu);
> + wake_up_process(thread);
> + *p = thread;
> + }
> +
> + }
> + put_online_cpus();
> +
> + return 0;
> +}
> +
> +static void end_power_clamp(void)
> +{
> + int i;
> + struct task_struct *thread;
> +
> + clamping = false;
> + /*
> + * make clamping visible to other cpus and give per cpu clamping threads
> + * sometime to exit, or gets killed later.
> + */
> + smp_mb();
> + msleep(20);
> + if (bitmap_weight(cpu_clamping_mask, num_possible_cpus())) {
> + for_each_set_bit(i, cpu_clamping_mask, num_possible_cpus()) {
> + pr_debug("clamping thread for cpu %d alive, kill\n", i);
> + thread = *per_cpu_ptr(powerclamp_thread, i);
> + kthread_stop(thread);
> + }
> + }
> +}
> +
> +static int powerclamp_cpu_callback(struct notifier_block *nfb,
> + unsigned long action, void *hcpu)
> +{
> + unsigned long cpu = (unsigned long)hcpu;
> + struct task_struct *thread;
> + struct task_struct **percpu_thread =
> + per_cpu_ptr(powerclamp_thread, cpu);
> +
> + if (false == clamping)
> + goto exit_ok;
> +
> + switch (action) {
> + case CPU_ONLINE:
> + thread = kthread_create_on_node(clamp_thread,
> + (void *) cpu,
> + cpu_to_node(cpu),
> + "kidle_inject/%lu", cpu);
> + if (likely(!IS_ERR(thread))) {
> + kthread_bind(thread, cpu);
> + wake_up_process(thread);
> + *percpu_thread = thread;
> + }
> + /* prefer BSP as controlling CPU */
> + if (cpu == 0) {
> + control_cpu = 0;
> + smp_mb();
> + }
> + break;
> + case CPU_DEAD:
> + if (test_bit(cpu, cpu_clamping_mask)) {
> + pr_err("cpu %lu dead but powerclamping thread is not\n",
> + cpu);
> + kthread_stop(*percpu_thread);
> + }
> + if (cpu == control_cpu) {
> + control_cpu = smp_processor_id();
> + smp_mb();
> + }
> + }
> +
> +exit_ok:
> + return NOTIFY_OK;
> +}
> +
> +static struct notifier_block powerclamp_cpu_notifier = {
> + .notifier_call = powerclamp_cpu_callback,
> +};
> +
> +static int powerclamp_get_max_state(struct thermal_cooling_device *cdev,
> + unsigned long *state)
> +{
> + *state = MAX_TARGET_RATIO;
> +
> + return 0;
> +}
> +
> +static int powerclamp_get_cur_state(struct thermal_cooling_device *cdev,
> + unsigned long *state)
> +{
> + if (true == clamping)
> + *state = pkg_cstate_ratio_cur;
> + else
> + /* to save power, do not poll idle ratio while not clamping */
> + *state = -1; /* indicates invalid state */
> +
> + return 0;
> +}
> +
> +static int powerclamp_set_cur_state(struct thermal_cooling_device *cdev,
> + unsigned long new_target_ratio)
> +{
> + int ret = 0;
> +
> + new_target_ratio = clamp(new_target_ratio, 0UL,
> + (unsigned long) (MAX_TARGET_RATIO-1));
> + if (set_target_ratio == 0 && new_target_ratio > 0) {
> + pr_info("Start idle injection to reduce power\n");
> + set_target_ratio = new_target_ratio;
> + ret = start_power_clamp();
> + goto exit_set;
> + } else if (set_target_ratio > 0 && new_target_ratio == 0) {
> + pr_info("Stop forced idle injection\n");
> + set_target_ratio = 0;
> + end_power_clamp();
> + } else /* adjust currently running */ {
> + set_target_ratio = new_target_ratio;
> + /* make new set_target_ratio visible to other cpus */
> + smp_mb();
> + }
> +
> +exit_set:
> + return ret;
> +}
> +
> +/* bind to generic thermal layer as cooling device*/
> +static struct thermal_cooling_device_ops powerclamp_cooling_ops = {
> + .get_max_state = powerclamp_get_max_state,
> + .get_cur_state = powerclamp_get_cur_state,
> + .set_cur_state = powerclamp_set_cur_state,
> +};
> +
> +/* runs on Nehalem and later */
> +static const struct x86_cpu_id intel_powerclamp_ids[] = {
> + { X86_VENDOR_INTEL, 6, 0x1a},
> + { X86_VENDOR_INTEL, 6, 0x1c},
> + { X86_VENDOR_INTEL, 6, 0x1e},
> + { X86_VENDOR_INTEL, 6, 0x1f},
> + { X86_VENDOR_INTEL, 6, 0x25},
> + { X86_VENDOR_INTEL, 6, 0x26},
> + { X86_VENDOR_INTEL, 6, 0x2a},
> + { X86_VENDOR_INTEL, 6, 0x2c},
> + { X86_VENDOR_INTEL, 6, 0x2d},
> + { X86_VENDOR_INTEL, 6, 0x2e},
> + { X86_VENDOR_INTEL, 6, 0x2f},
> + { X86_VENDOR_INTEL, 6, 0x3a},
> + {}
> +};
> +MODULE_DEVICE_TABLE(x86cpu, intel_powerclamp_ids);
> +
> +static int powerclamp_probe(void)
> +{
> + if (!x86_match_cpu(intel_powerclamp_ids)) {
> + pr_err("Intel powerclamp does not run on family %d model %d\n",
> + boot_cpu_data.x86, boot_cpu_data.x86_model);
> + return -ENODEV;
> + }
> + if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC) ||
> + !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) ||
> + !boot_cpu_has(X86_FEATURE_MWAIT) ||
> + !boot_cpu_has(X86_FEATURE_ARAT))
> + return -ENODEV;
> +
> + /* find the deepest mwait value */
> + find_target_mwait();
> +
> + return 0;
> +}
> +
> +static int powerclamp_debug_show(struct seq_file *m, void *unused)
> +{
> + int i = 0;
> +
> + seq_printf(m, "controlling cpu: %d\n", control_cpu);
> + seq_printf(m, "pct confidence steady dynamic (compensation)\n");
> + for (i = 0; i < MAX_TARGET_RATIO; i++) {
> + seq_printf(m, "%d\t%lu\t%lu\t%lu\n",
> + i,
> + cal_data[i].confidence,
> + cal_data[i].steady_comp,
> + cal_data[i].dynamic_comp);
> + }
> +
> + return 0;
> +}
> +
> +static int powerclamp_debug_open(struct inode *inode,
> + struct file *file)
> +{
> + return single_open(file, powerclamp_debug_show, inode->i_private);
> +}
> +
> +static const struct file_operations powerclamp_debug_fops = {
> + .open = powerclamp_debug_open,
> + .read = seq_read,
> + .llseek = seq_lseek,
> + .release = single_release,
> + .owner = THIS_MODULE,
> +};
> +
> +static inline void powerclamp_create_debug_files(void)
> +{
> + debug_dir = debugfs_create_dir("intel_powerclamp", NULL);
> + if (!debug_dir)
> + return;
> +
> + if (!debugfs_create_file("powerclamp_calib", S_IRUGO, debug_dir,
> + cal_data, &powerclamp_debug_fops))
> + goto file_error;
> +
> + return;
> +
> +file_error:
> + debugfs_remove_recursive(debug_dir);
> +}
> +
> +static int powerclamp_init(void)
> +{
> + int retval;
> + int bitmap_size;
> +
> + bitmap_size = BITS_TO_LONGS(num_possible_cpus()) * sizeof(long);
> + cpu_clamping_mask = kzalloc(bitmap_size, GFP_KERNEL);
> + if (!cpu_clamping_mask)
> + return -ENOMEM;
> +
> + /* probe cpu features and ids here */
> + retval = powerclamp_probe();
> + if (retval)
> + return retval;
> + /* set default limit, maybe adjusted during runtime based on feedback */
> + window_size = 2;
> + register_hotcpu_notifier(&powerclamp_cpu_notifier);
> + powerclamp_thread = alloc_percpu(struct task_struct *);
> + cooling_dev = thermal_cooling_device_register("intel_powerclamp", NULL,
> + &powerclamp_cooling_ops);
> + if (IS_ERR(cooling_dev))
> + return -ENODEV;
> +
> + if (!duration)
> + duration = jiffies_to_msecs(DEFAULT_DURATION_JIFFIES);
> + powerclamp_create_debug_files();
> +
> + return 0;
> +}
> +module_init(powerclamp_init);
> +
> +static void powerclamp_exit(void)
> +{
> + unregister_hotcpu_notifier(&powerclamp_cpu_notifier);
> + end_power_clamp();
> + free_percpu(powerclamp_thread);
> + thermal_cooling_device_unregister(cooling_dev);
> + kfree(cpu_clamping_mask);
> +
> + cancel_delayed_work_sync(&poll_pkg_cstate_work);
> + debugfs_remove_recursive(debug_dir);
> +}
> +module_exit(powerclamp_exit);
> +
> +MODULE_LICENSE("GPL");
> +MODULE_AUTHOR("Arjan van de Ven <arjan@xxxxxxxxxxxxxxx>");
> +MODULE_AUTHOR("Jacob Pan <jacob.jun.pan@xxxxxxxxxxxxxxx>");
> +MODULE_DESCRIPTION("Package Level C-state Idle Injection for Intel CPUs");


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/