Re: [PATCH v2] tools: create power/x86/x86_energy_perf_policy

[Chen Gong <gong.chen@linux.intel.com>]

于 11/24/2010 1:31 PM, Len Brown 写道:
> From: Len Brown<len.brown@intel.com>
>
> 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<len.brown@intel.com>
> ---
> 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<len.brown@intel.com>
> +.\"  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<len.brown@intel.com>
> 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<len.brown@intel.com>
> + *
> + * 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<stdio.h>
> +#include<unistd.h>
> +#include<sys/types.h>
> +#include<sys/stat.h>
> +#include<sys/resource.h>
> +#include<fcntl.h>
> +#include<signal.h>
> +#include<sys/time.h>
> +#include<stdlib.h>
> +#include<string.h>
> +
> +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<lenb@kernel.org>\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"
_______________________________________________
linux-pm mailing list
linux-pm@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/linux-pm