From mboxrd@z Thu Jan 1 00:00:00 1970 From: Zhang Rui Subject: Re: [PATCH v7 3/3] PM: Introduce Intel PowerClamp Driver Date: Thu, 17 Jan 2013 15:03:05 +0800 Message-ID: <1358406185.20319.3.camel@rzhang1-mobl4> References: <1358341895-4369-1-git-send-email-jacob.jun.pan@linux.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from mga11.intel.com ([192.55.52.93]:28336 "EHLO mga11.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758608Ab3AQHDM (ORCPT ); Thu, 17 Jan 2013 02:03:12 -0500 In-Reply-To: <1358341895-4369-1-git-send-email-jacob.jun.pan@linux.intel.com> Sender: linux-pm-owner@vger.kernel.org List-Id: linux-pm@vger.kernel.org To: Jacob Pan Cc: Linux PM , LKML , Rafael Wysocki , Len Brown , Arjan van de Ven 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-stat= e > ratio. >=20 > 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. >=20 > Please refer to Documentation/thermal/intel_powerclamp.txt for more d= etails. >=20 > Signed-off-by: Arjan van de Ven > Signed-off-by: Jacob Pan Reviewed-by: Zhang Rui Rafael, can you take this please? I'm not going to push anything except bug fixes until next merge window. 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 >=20 > diff --git a/Documentation/thermal/intel_powerclamp.txt b/Documentati= on/thermal/intel_powerclamp.txt > new file mode 100644 > index 0000000..332de4a > --- /dev/null > +++ b/Documentation/thermal/intel_powerclamp.txt > @@ -0,0 +1,307 @@ > + =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D > + INTEL POWERCLAMP DRIVER > + =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D > +By: Arjan van de Ven > + Jacob Pan > + > +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) > + > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > +INTRODUCTION > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > + > +Consider the situation where a system=E2=80=99s power consumption mu= st be > +reduced at runtime, due to power budget, thermal constraint, or nois= e > +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 fa= r > +they=E2=80=99re 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 goa= l > +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. > + > + > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > +THEORY OF OPERATION > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > + > +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 tar= get 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 roundin= g > +of jiffies, so accumulated errors can be prevented to avoid a jitter= y > +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 tic= k > +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 i= n > +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 adjus= t > +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-state= s. > + > +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, additiona= l > +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. > + > + > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > +Performance Analysis > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > +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 generate= d > +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 throttle= d > +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/8= P > +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). > + > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > +Usage and Interfaces > +=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > +The powerclamp driver is registered to the generic thermal layer as = a > +cooling device. Currently, it=E2=80=99s not bound to any thermal zon= es. > + > +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_stat= e > +" > + > +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% natura= l > +idle time, powerclamp kernel threads will do idle injection, which > +appear running to the scheduler. But the overall system idle is stil= l > +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%s= i, 0.0%st > +Mem: 3943228k total, 1689632k used, 2253596k free, 74960k buf= fers > +Swap: 4087804k total, 0k used, 4087804k free, 945336k cac= hed > + > + 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_i= nject/0 > + 3344 root -51 0 0 0 0 D 25 0.0 0:01.60 kidle_i= nject/3 > + 3342 root -51 0 0 0 0 D 25 0.0 0:01.61 kidle_i= nject/1 > + 3343 root -51 0 0 0 0 D 25 0.0 0:01.60 kidle_i= nject/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 lo= w to > cool down the CPU. > =20 > +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. Thi= s > + enforce idle time which results in more package C-state residency= =2E 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-$(CONFIG_RCAR_THERMAL) +=3D rcar_thermal.o > obj-$(CONFIG_EXYNOS_THERMAL) +=3D exynos_thermal.o > obj-$(CONFIG_DB8500_THERMAL) +=3D db8500_thermal.o > obj-$(CONFIG_DB8500_CPUFREQ_COOLING) +=3D db8500_cpufreq_cooling.o > +obj-$(CONFIG_INTEL_POWERCLAMP) +=3D intel_powerclamp.o > + > diff --git a/drivers/thermal/intel_powerclamp.c b/drivers/thermal/int= el_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 > + * Jacob Pan > + * > + * This program is free software; you can redistribute it and/or mod= ify 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 WI= THOUT > + * ANY WARRANTY; without even the implied warranty of MERCHANTABILIT= Y or > + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public Lic= ense 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, curre= ntly a fixed > + * compensation is added to clamping duration when exce= ssive amount > + * of wakeups are observed during idle time. the reason= is that in > + * case of external interrupts without need for ack, cl= amping down > + * cpu in non-irq context does not reduce irq. for majo= rity of the > + * cases, clamping down cpu does help reduce irq as wel= l, we should > + * be able to differenciate the two cases and give a qu= antitative > + * solution for the irqs that we can control. perhaps b= ased on > + * get_cpu_iowait_time_us() > + * > + * 2. synchronization with other hw blocks > + * > + * > + */ > + > +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt > + > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > + > +#include > +#include > +#include > +#include > +#include > +#include > + > +#define MAX_TARGET_RATIO (50U) > +/* For each undisturbed clamping period (no extra wake ups during id= le 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 pe= r 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 =3D 0; > + unsigned long new_duration; > + > + ret =3D 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 =3D -EINVAL; > + } > + > + duration =3D clamp(new_duration, 6ul, 25ul); > + smp_mb(); > + > +exit: > + > + return ret; > +} > + > +static struct kernel_param_ops duration_ops =3D { > + .set =3D duration_set, > + .get =3D param_get_int, > +}; > + > + > +module_param_cb(duration, &duration_ops, &duration, 0644); > +MODULE_PARM_DESC(duration, "forced idle time for each attempt in mse= c."); > + > +struct powerclamp_calibration_data { > + unsigned long confidence; /* used for calibration, basically a cou= nter > + * 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 idl= e > + * 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_para= m *kp) > +{ > + int ret =3D 0; > + unsigned long new_window_size; > + > + ret =3D 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 =3D -EINVAL; > + } > + > + window_size =3D clamp(new_window_size, 2ul, 10ul); > + smp_mb(); > + > +exit_win: > + > + return ret; > +} > + > +static struct kernel_param_ops window_size_ops =3D { > + .set =3D window_size_set, > + .get =3D 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 =3D 0; > + unsigned int highest_subcstate =3D 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 >>=3D MWAIT_SUBSTATE_SIZE; > + for (i =3D 0; i < 7 && edx; i++, edx >>=3D MWAIT_SUBSTATE_SIZE) { > + if (edx & MWAIT_SUBSTATE_MASK) { > + highest_cstate =3D i; > + highest_subcstate =3D edx & MWAIT_SUBSTATE_MASK; > + } > + } > + target_mwait =3D (highest_cstate << MWAIT_SUBSTATE_SIZE) | > + (highest_subcstate - 1); > + > +} > + > +static u64 pkg_state_counter(void) > +{ > + u64 val; > + u64 count =3D 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 +=3D val; > + else > + skip_c2 =3D true; > + } > + > + if (!skip_c3) { > + if (!rdmsrl_safe(MSR_PKG_C3_RESIDENCY, &val)) > + count +=3D val; > + else > + skip_c3 =3D true; > + } > + > + if (!skip_c6) { > + if (!rdmsrl_safe(MSR_PKG_C6_RESIDENCY, &val)) > + count +=3D val; > + else > + skip_c6 =3D true; > + } > + > + if (!skip_c7) { > + if (!rdmsrl_safe(MSR_PKG_C7_RESIDENCY, &val)) > + count +=3D val; > + else > + skip_c7 =3D 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 =3D 0; > + > + /* we only use compensation if all adjacent ones are good */ > + if (ratio =3D=3D 1 && > + cal_data[ratio].confidence >=3D CONFIDENCE_OK && > + cal_data[ratio + 1].confidence >=3D CONFIDENCE_OK && > + cal_data[ratio + 2].confidence >=3D CONFIDENCE_OK) { > + comp =3D (cal_data[ratio].steady_comp + > + cal_data[ratio + 1].steady_comp + > + cal_data[ratio + 2].steady_comp) / 3; > + } else if (ratio =3D=3D MAX_TARGET_RATIO - 1 && > + cal_data[ratio].confidence >=3D CONFIDENCE_OK && > + cal_data[ratio - 1].confidence >=3D CONFIDENCE_OK && > + cal_data[ratio - 2].confidence >=3D CONFIDENCE_OK) { > + comp =3D (cal_data[ratio].steady_comp + > + cal_data[ratio - 1].steady_comp + > + cal_data[ratio - 2].steady_comp) / 3; > + } else if (cal_data[ratio].confidence >=3D CONFIDENCE_OK && > + cal_data[ratio - 1].confidence >=3D CONFIDENCE_OK && > + cal_data[ratio + 1].confidence >=3D CONFIDENCE_OK) { > + comp =3D (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 =3D ratio; > + /* do not exceed limit */ > + if (comp + ratio >=3D MAX_TARGET_RATIO) > + comp =3D MAX_TARGET_RATIO - ratio - 1; > + > + return comp; > +} > + > +static void adjust_compensation(int target_ratio, unsigned int win) > +{ > + int delta; > + struct powerclamp_calibration_data *d =3D &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 >=3D CONFIDENCE_OK || > + atomic_read(&idle_wakeup_counter) > > + win * num_online_cpus()) > + return; > + > + delta =3D set_target_ratio - current_ratio; > + /* filter out bad data */ > + if (delta >=3D 0 && delta <=3D (1+target_ratio/10)) { > + if (d->steady_comp) > + d->steady_comp =3D > + roundup(delta+d->steady_comp, 2)/2; > + else > + d->steady_comp =3D 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 =3D pkg_state_counter(); > + rdtscll(tsc_now); > + > + /* calculate pkg cstate vs tsc ratio */ > + if (!msr_last || !tsc_last) > + current_ratio =3D 1; > + else if (tsc_now-tsc_last) > + current_ratio =3D 100*(msr_now-msr_last)/ > + (tsc_now-tsc_last); > + > + /* update record */ > + msr_last =3D msr_now; > + tsc_last =3D tsc_now; > + > + adjust_compensation(target_ratio, win); > + /* > + * too many external interrupts, set flag such > + * that we can take measure later. > + */ > + reduce_irq =3D atomic_read(&idle_wakeup_counter) >=3D > + 2 * win * num_online_cpus(); > + > + atomic_set(&idle_wakeup_counter, 0); > + /* if we are above target+guard, skip */ > + return set_target_ratio + guard <=3D current_ratio; > +} > + > +static int clamp_thread(void *arg) > +{ > + int cpunr =3D (unsigned long)arg; > + DEFINE_TIMER(wakeup_timer, noop_timer, 0, 0); > + static const struct sched_param param =3D { > + .sched_priority =3D MAX_USER_RT_PRIO/2, > + }; > + unsigned int count =3D 0; > + unsigned int target_ratio; > + > + set_bit(cpunr, cpu_clamping_mask); > + set_freezable(); > + init_timer_on_stack(&wakeup_timer); > + sched_setscheduler(current, SCHED_FIFO, ¶m); > + > + while (true =3D=3D clamping && !kthread_should_stop() && > + cpu_online(cpunr)) { > + int sleeptime; > + unsigned long target_jiffies; > + unsigned int guard; > + unsigned int compensation =3D 0; > + int interval; /* jiffies to sleep for each attempt */ > + unsigned int duration_jiffies =3D 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 =3D set_target_ratio; > + guard =3D 1 + target_ratio/20; > + window_size_now =3D 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 =3D get_compensation(target_ratio); > + interval =3D duration_jiffies*100/(target_ratio+compensation); > + > + /* align idle time */ > + target_jiffies =3D roundup(jiffies, interval); > + sleeptime =3D target_jiffies - jiffies; > + if (sleeptime <=3D 0) > + sleeptime =3D 1; > + schedule_timeout_interruptible(sleeptime); > + /* > + * only elected controlling cpu can collect stats and update > + * control parameters. > + */ > + if (cpunr =3D=3D control_cpu && !(count%window_size_now)) { > + should_skip =3D > + powerclamp_adjust_controls(target_ratio, > + guard, window_size_now); > + smp_mb(); > + } > + > + if (should_skip) > + continue; > + > + target_jiffies =3D 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 =3D 1; > + unsigned long eax =3D 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 *)¤t_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 =3D pkg_state_counter(); > + rdtscll(tsc_now); > + jiffies_now =3D jiffies; > + > + /* calculate pkg cstate vs tsc ratio */ > + if (!msr_last || !tsc_last) > + pkg_cstate_ratio_cur =3D 1; > + else { > + if (tsc_now - tsc_last) > + pkg_cstate_ratio_cur =3D 100 * (msr_now - msr_last)/ > + (tsc_now - tsc_last); > + } > + > + /* update record */ > + msr_last =3D msr_now; > + jiffies_last =3D jiffies_now; > + tsc_last =3D tsc_now; > + > + if (true =3D=3D 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 =3D clamp(set_target_ratio, 0U, MAX_TARGET_RATIO); > + /* prevent cpu hotplug */ > + get_online_cpus(); > + > + /* prefer BSP */ > + control_cpu =3D 0; > + if (!cpu_online(control_cpu)) > + control_cpu =3D smp_processor_id(); > + > + clamping =3D true; > + schedule_delayed_work(&poll_pkg_cstate_work, 0); > + > + /* start one thread per online cpu */ > + for_each_online_cpu(cpu) { > + struct task_struct **p =3D > + per_cpu_ptr(powerclamp_thread, cpu); > + > + thread =3D 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 =3D thread; > + } > + > + } > + put_online_cpus(); > + > + return 0; > +} > + > +static void end_power_clamp(void) > +{ > + int i; > + struct task_struct *thread; > + > + clamping =3D false; > + /* > + * make clamping visible to other cpus and give per cpu clamping th= reads > + * 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 =3D *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 =3D (unsigned long)hcpu; > + struct task_struct *thread; > + struct task_struct **percpu_thread =3D > + per_cpu_ptr(powerclamp_thread, cpu); > + > + if (false =3D=3D clamping) > + goto exit_ok; > + > + switch (action) { > + case CPU_ONLINE: > + thread =3D 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 =3D thread; > + } > + /* prefer BSP as controlling CPU */ > + if (cpu =3D=3D 0) { > + control_cpu =3D 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 =3D=3D control_cpu) { > + control_cpu =3D smp_processor_id(); > + smp_mb(); > + } > + } > + > +exit_ok: > + return NOTIFY_OK; > +} > + > +static struct notifier_block powerclamp_cpu_notifier =3D { > + .notifier_call =3D powerclamp_cpu_callback, > +}; > + > +static int powerclamp_get_max_state(struct thermal_cooling_device *c= dev, > + unsigned long *state) > +{ > + *state =3D MAX_TARGET_RATIO; > + > + return 0; > +} > + > +static int powerclamp_get_cur_state(struct thermal_cooling_device *c= dev, > + unsigned long *state) > +{ > + if (true =3D=3D clamping) > + *state =3D pkg_cstate_ratio_cur; > + else > + /* to save power, do not poll idle ratio while not clamping */ > + *state =3D -1; /* indicates invalid state */ > + > + return 0; > +} > + > +static int powerclamp_set_cur_state(struct thermal_cooling_device *c= dev, > + unsigned long new_target_ratio) > +{ > + int ret =3D 0; > + > + new_target_ratio =3D clamp(new_target_ratio, 0UL, > + (unsigned long) (MAX_TARGET_RATIO-1)); > + if (set_target_ratio =3D=3D 0 && new_target_ratio > 0) { > + pr_info("Start idle injection to reduce power\n"); > + set_target_ratio =3D new_target_ratio; > + ret =3D start_power_clamp(); > + goto exit_set; > + } else if (set_target_ratio > 0 && new_target_ratio =3D=3D 0) { > + pr_info("Stop forced idle injection\n"); > + set_target_ratio =3D 0; > + end_power_clamp(); > + } else /* adjust currently running */ { > + set_target_ratio =3D 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 =3D = { > + .get_max_state =3D powerclamp_get_max_state, > + .get_cur_state =3D powerclamp_get_cur_state, > + .set_cur_state =3D powerclamp_set_cur_state, > +}; > + > +/* runs on Nehalem and later */ > +static const struct x86_cpu_id intel_powerclamp_ids[] =3D { > + { 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 =3D 0; > + > + seq_printf(m, "controlling cpu: %d\n", control_cpu); > + seq_printf(m, "pct confidence steady dynamic (compensation)\n"); > + for (i =3D 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 =3D { > + .open =3D powerclamp_debug_open, > + .read =3D seq_read, > + .llseek =3D seq_lseek, > + .release =3D single_release, > + .owner =3D THIS_MODULE, > +}; > + > +static inline void powerclamp_create_debug_files(void) > +{ > + debug_dir =3D 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 =3D BITS_TO_LONGS(num_possible_cpus()) * sizeof(long); > + cpu_clamping_mask =3D kzalloc(bitmap_size, GFP_KERNEL); > + if (!cpu_clamping_mask) > + return -ENOMEM; > + > + /* probe cpu features and ids here */ > + retval =3D powerclamp_probe(); > + if (retval) > + return retval; > + /* set default limit, maybe adjusted during runtime based on feedba= ck */ > + window_size =3D 2; > + register_hotcpu_notifier(&powerclamp_cpu_notifier); > + powerclamp_thread =3D alloc_percpu(struct task_struct *); > + cooling_dev =3D thermal_cooling_device_register("intel_powerclamp",= NULL, > + &powerclamp_cooling_ops); > + if (IS_ERR(cooling_dev)) > + return -ENODEV; > + > + if (!duration) > + duration =3D 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 "); > +MODULE_AUTHOR("Jacob Pan "); > +MODULE_DESCRIPTION("Package Level C-state Idle Injection for Intel C= PUs");