Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751348Ab0KYFxE (ORCPT ); Thu, 25 Nov 2010 00:53:04 -0500 Received: from mga11.intel.com ([192.55.52.93]:42469 "EHLO mga11.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750903Ab0KYFxC (ORCPT ); Thu, 25 Nov 2010 00:53:02 -0500 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.59,252,1288594800"; d="scan'208";a="630153584" Message-ID: <4CEDF9BB.9030002@linux.intel.com> Date: Thu, 25 Nov 2010 13:52:59 +0800 From: Chen Gong User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.1; zh-CN; rv:1.9.2.12) Gecko/20101027 Thunderbird/3.1.6 MIME-Version: 1.0 To: Len Brown CC: Greg Kroah-Hartman , linux-pm@lists.linux-foundation.org, linux-kernel@vger.kernel.org, linux-acpi@vger.kernel.org, x86@kernel.org Subject: Re: [PATCH v2] tools: create power/x86/x86_energy_perf_policy References: In-Reply-To: Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 14888 Lines: 521 于 11/24/2010 1:31 PM, Len Brown 写道: > From: Len Brown > > MSR_IA32_ENERGY_PERF_BIAS first became available on Westmere Xeon. > It is implemented in all Sandy Bridge processors -- mobile, desktop and server. > It is expected to become increasingly important in subsequent generations. > > x86_energy_perf_policy is a user-space utility to set this > hardware energy vs performance policy hint in the processor. > Most systems would benefit from "x86_energy_perf_policy normal" > at system startup, as the hardware default is maximum performance > at the expense of energy efficiency. > > Linux-2.6.36 added "epb" to /proc/cpuinfo to indicate > if an x86 processor supports MSR_IA32_ENERGY_PERF_BIAS, > though the kernel does not actually program the MSR. > > In March, Venkatesh Pallipadi proposed a small driver > that programmed MSR_IA32_ENERGY_PERF_BIAS, based on > the cpufreq governor in use. It also offered > a boot-time cmdline option to override. > http://lkml.org/lkml/2010/3/4/457 > But hiding the hardware policy behind the > governor choice was deemed "kinda icky". > > So in June, I proposed a generic user/kernel API to > consolidate the power/performance policy trade-off. > "RFC: /sys/power/policy_preference" > http://lkml.org/lkml/2010/6/16/399 > That is my preference for implementing this capability, > but I received no support on the list. > > So in September, I sent x86_energy_perf_policy.c to LKML, > a user-space utility that scribbles directly to the MSR. > http://lkml.org/lkml/2010/9/28/246 > > Here is the same utility re-sent, this time proposed > to reside in the kernel tools directory. > > Signed-off-by: Len Brown > --- > v2 > create man page > minor tweaks in response to review comments > > tools/power/x86/x86_energy_perf_policy/Makefile | 8 + > .../x86_energy_perf_policy.8 | 104 +++++++ > .../x86_energy_perf_policy.c | 325 ++++++++++++++++++++ > > diff --git a/tools/power/x86/x86_energy_perf_policy/Makefile b/tools/power/x86/x86_energy_perf_policy/Makefile > new file mode 100644 > index 0000000..f458237 > --- /dev/null > +++ b/tools/power/x86/x86_energy_perf_policy/Makefile > @@ -0,0 +1,8 @@ > +x86_energy_perf_policy : x86_energy_perf_policy.c > + > +clean : > + rm -f x86_energy_perf_policy > + > +install : > + install x86_energy_perf_policy /usr/bin/ > + install x86_energy_perf_policy.8 /usr/share/man/man8/ > diff --git a/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.8 b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.8 > new file mode 100644 > index 0000000..8eaaad6 > --- /dev/null > +++ b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.8 > @@ -0,0 +1,104 @@ > +.\" This page Copyright (C) 2010 Len Brown > +.\" Distributed under the GPL, Copyleft 1994. > +.TH X86_ENERGY_PERF_POLICY 8 > +.SH NAME > +x86_energy_perf_policy \- read or write MSR_IA32_ENERGY_PERF_BIAS > +.SH SYNOPSIS > +.ft B > +.B x86_energy_perf_policy > +.RB [ "\-c cpu" ] > +.RB [ "\-v" ] > +.RB "\-r" > +.br > +.B x86_energy_perf_policy > +.RB [ "\-c cpu" ] > +.RB [ "\-v" ] > +.RB 'performance' > +.br > +.B x86_energy_perf_policy > +.RB [ "\-c cpu" ] > +.RB [ "\-v" ] > +.RB 'normal' > +.br > +.B x86_energy_perf_policy > +.RB [ "\-c cpu" ] > +.RB [ "\-v" ] > +.RB 'powersave' > +.br > +.B x86_energy_perf_policy > +.RB [ "\-c cpu" ] > +.RB [ "\-v" ] > +.RB n > +.br > +.SH DESCRIPTION > +\fBx86_energy_perf_policy\fP > +allows software to convey > +its policy for the relative importance of performance > +versus energy savings to the processor. > + > +The processor uses this information in model-specific ways > +when it must select trade-offs between performance and > +energy efficiency. > + > +This policy hint does not supersede Processor Performance states > +(P-states) or CPU Idle power states (C-states), but allows > +software to have influence where it would otherwise be unable > +to express a preference. > + > +For example, this setting may tell the hardware how > +aggressively or conservatively to control frequency > +in the "turbo range" above the explicitly OS-controlled > +P-state frequency range. It may also tell the hardware > +how aggressively is should enter the OS requested C-states. > + > +Support for this feature is indicated by CPUID.06H.ECX.bit3 > +per the Intel Architectures Software Developer's Manual. > + > +.SS Options > +\fB-c\fP limits operation to a single CPU. > +The default is to operate on all CPUs. > +Note that MSR_IA32_ENERGY_PERF_BIAS is defined per > +logical processor, but that the initial implementations > +of the MSR were shared among all processors in each package. > +.PP > +\fB-v\fP increases verbosity. By default > +x86_energy_perf_policy is silent. > +.PP > +\fB-r\fP is for "read-only" mode - the unchanged state > +is read and displayed. > +.PP > +.I performance > +Set a policy where performance is paramount. > +The processor will be unwilling to sacrifice any performance > +for the sake of energy saving. This is the hardware default. > +.PP > +.I normal > +Set a policy with a normal balance between performance and energy efficiency. > +The processor will tolerate minor performance compromise > +for potentially significant energy savings. > +This reasonable default for most desktops and servers. > +.PP > +.I powersave > +Set a policy where the processor can accept > +a measurable performance hit to maximize energy efficiency. > +.PP > +.I n > +Set MSR_IA32_ENERGY_PERF_BIAS to the specified number. > +The range of valid numbers is 0-15, where 0 is maximum > +performance and 15 is maximum energy efficiency. > + > +.SH NOTES > +.B "x86_energy_perf_policy " > +runs only as root. > +.SH FILES > +.ta > +.nf > +/dev/cpu/*/msr > +.fi > + > +.SH "SEE ALSO" > +msr(4) > +.PP > +.SH AUTHORS > +.nf > +Written by Len Brown > diff --git a/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c > new file mode 100644 > index 0000000..b539923 > --- /dev/null > +++ b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c > @@ -0,0 +1,325 @@ > +/* > + * x86_energy_perf_policy -- set the energy versus performance > + * policy preference bias on recent X86 processors. > + */ > +/* > + * Copyright (c) 2010, Intel Corporation. > + * Len Brown > + * > + * 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. > + */ > + > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > + > +unsigned int verbose; /* set with -v */ > +unsigned int read_only; /* set with -r */ > +char *progname; > +unsigned long long new_bias; > +int cpu = -1; > + > +/* > + * Usage: > + * > + * -c cpu: limit action to a single CPU (default is all CPUs) > + * -v: verbose output (can invoke more than once) > + * -r: read-only, don't change any settings > + * > + * performance > + * Performance is paramount. > + * Unwilling to sacrafice any performance > + * for the sake of energy saving. (hardware default) > + * > + * normal > + * Can tolerate minor performance compromise > + * for potentially significant energy savings. > + * (reasonable default for most desktops and servers) > + * > + * powersave > + * Can tolerate significant performance hit > + * to maximize energy savings. > + * > + * n > + * a numerical value to write to the underlying MSR. > + */ > +void usage(void) > +{ > + printf("%s: [-c cpu] [-v] " > + "(-r | 'performance' | 'normal' | 'powersave' | n)\n", > + progname); > + exit(1); > +} > + > +#define MSR_IA32_ENERGY_PERF_BIAS 0x000001b0 > + > +#define BIAS_PERFORMANCE 0 > +#define BIAS_BALANCE 6 > +#define BIAS_POWERSAVE 15 > + > +void cmdline(int argc, char **argv) > +{ > + int opt; > + > + progname = argv[0]; > + > + while ((opt = getopt(argc, argv, "+rvc:")) != -1) { > + switch (opt) { > + case 'c': > + cpu = atoi(optarg); > + break; > + case 'r': > + read_only = 1; > + break; > + case 'v': > + verbose++; > + break; > + default: > + usage(); > + } > + } > + /* if -r, then should be no additional optind */ > + if (read_only&& (argc> optind)) > + usage(); > + > + /* > + * if no -r , then must be one additional optind > + */ > + if (!read_only) { > + > + if (argc != optind + 1) { > + printf("must supply -r or policy param\n"); > + usage(); > + } > + > + if (!strcmp("performance", argv[optind])) { > + new_bias = BIAS_PERFORMANCE; > + } else if (!strcmp("normal", argv[optind])) { > + new_bias = BIAS_BALANCE; > + } else if (!strcmp("powersave", argv[optind])) { > + new_bias = BIAS_POWERSAVE; > + } else { > + char *endptr; > + > + new_bias = strtoull(argv[optind],&endptr, 0); > + if (endptr == argv[optind] || > + new_bias> BIAS_POWERSAVE) { > + fprintf(stderr, "invalid value: %s\n", > + argv[optind]); > + usage(); > + } > + } > + } > +} > + > +/* > + * validate_cpuid() > + * returns on success, quietly exits on failure (make verbose with -v) > + */ > +void validate_cpuid(void) > +{ > + unsigned int eax, ebx, ecx, edx, max_level; > + char brand[16]; > + unsigned int fms, family, model, stepping; > + > + eax = ebx = ecx = edx = 0; > + > + asm("cpuid" : "=a" (max_level), "=b" (ebx), "=c" (ecx), > + "=d" (edx) : "a" (0)); > + > + if (ebx != 0x756e6547 || edx != 0x49656e69 || ecx != 0x6c65746e) { > + if (verbose) > + fprintf(stderr, "%.4s%.4s%.4s != GenuineIntel", > + (char *)&ebx, (char *)&edx, (char *)&ecx); > + exit(1); > + } > + > + asm("cpuid" : "=a" (fms), "=c" (ecx), "=d" (edx) : "a" (1) : "ebx"); > + family = (fms>> 8)& 0xf; > + model = (fms>> 4)& 0xf; > + stepping = fms& 0xf; > + if (family == 6 || family == 0xf) > + model += ((fms>> 16)& 0xf)<< 4; > + > + if (verbose> 1) > + printf("CPUID %s %d levels family:model:stepping " > + "0x%x:%x:%x (%d:%d:%d)\n", brand, max_level, > + family, model, stepping, family, model, stepping); > + > + if (!(edx& (1<< 5))) { > + if (verbose) > + printf("CPUID: no MSR\n"); > + exit(1); > + } > + > + /* > + * Support for MSR_IA32_ENERGY_PERF_BIAS > + * is indicated by CPUID.06H.ECX.bit3 > + */ > + asm("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (6)); > + if (verbose) > + printf("CPUID.06H.ECX: 0x%x\n", ecx); > + if (!(ecx& (1<< 3))) { > + if (verbose) > + printf("CPUID: No MSR_IA32_ENERGY_PERF_BIAS\n"); > + exit(1); > + } > + return; /* success */ > +} > + > +unsigned long long get_msr(int cpu, int offset) > +{ > + unsigned long long msr; > + char msr_path[32]; > + int retval; > + int fd; > + > + sprintf(msr_path, "/dev/cpu/%d/msr", cpu); > + fd = open(msr_path, O_RDONLY); > + if (fd< 0) { > + printf("Try \"# modprobe msr\"\n"); > + perror(msr_path); > + exit(1); > + } > + > + retval = pread(fd,&msr, sizeof msr, offset); > + > + if (retval != sizeof msr) { > + printf("pread cpu%d 0x%x = %d\n", cpu, offset, retval); > + exit(-2); > + } > + close(fd); > + return msr; > +} > + > +unsigned long long put_msr(int cpu, unsigned long long new_msr, int offset) > +{ > + unsigned long long old_msr; > + char msr_path[32]; > + int retval; > + int fd; > + > + sprintf(msr_path, "/dev/cpu/%d/msr", cpu); > + fd = open(msr_path, O_RDWR); > + if (fd< 0) { > + perror(msr_path); > + exit(1); > + } > + > + retval = pread(fd,&old_msr, sizeof old_msr, offset); > + if (retval != sizeof old_msr) { > + perror("pwrite"); > + printf("pread cpu%d 0x%x = %d\n", cpu, offset, retval); > + exit(-2); > + } > + > + retval = pwrite(fd,&new_msr, sizeof new_msr, offset); > + if (retval != sizeof new_msr) { > + perror("pwrite"); > + printf("pwrite cpu%d 0x%x = %d\n", cpu, offset, retval); > + exit(-2); > + } > + > + close(fd); > + > + return old_msr; > +} > + > +void print_msr(int cpu) > +{ > + printf("cpu%d: 0x%016llx\n", > + cpu, get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS)); > +} > + > +void update_msr(int cpu) > +{ > + unsigned long long previous_msr; > + > + previous_msr = put_msr(cpu, new_bias, MSR_IA32_ENERGY_PERF_BIAS); > + > + if (verbose) > + printf("cpu%d msr0x%x 0x%016llx -> 0x%016llx\n", > + cpu, MSR_IA32_ENERGY_PERF_BIAS, previous_msr, new_bias); > + > + return; > +} > + > +char *proc_stat = "/proc/stat"; > +/* > + * run func() on every cpu in /dev/cpu > + */ > +void for_every_cpu(void (func)(int)) > +{ > + FILE *fp; > + int retval; > + > + fp = fopen(proc_stat, "r"); > + if (fp == NULL) { > + perror(proc_stat); > + exit(1); > + } > + > + retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n"); > + if (retval != 0) { > + perror("/proc/stat format"); > + exit(1); > + } > + > + while (1) { > + int cpu; > + > + retval = fscanf(fp, > + "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", > + &cpu); > + if (retval != 1) > + return; > + > + func(cpu); > + } > + fclose(fp); > +} > + > +int main(int argc, char **argv) > +{ > + cmdline(argc, argv); > + > + if (verbose> 1) > + printf("x86_energy_perf_policy Nov 24, 2010" > + " - Len Brown\n"); > + if (verbose> 1&& !read_only) > + printf("new_bias %lld\n", new_bias); > + > + validate_cpuid(); > + > + if (cpu != -1) { > + if (read_only) > + print_msr(cpu); > + else > + update_msr(cpu); > + } else { > + if (read_only) > + for_every_cpu(print_msr); > + else > + for_every_cpu(update_msr); > + } > + > + return 0; > +} > I have 2 questions. 1. the usage looks too simple. If I haven't read the comments in the source codes, I even can't know the exact meaning of these parameters. Such as -v, -vv etc. How about adding the comments as the part of the usage ? 2. the paramter "noraml | performance | powersave | n" looks weird. why it can't look like other paramter (-r, -v etc.). For example, I can't use it such as "./x86_energy_perf_policy -c 0 normal -v" -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/