2010-11-15 16:01:26

by Len Brown

Subject: [PATCH RESEND] tools: add power/x86/turbostat - v2

From: Len Brown <[email protected]>

turbostat displays actual processor frequency on modern
Intel processors. Note that this capability depends on free-running
APERF and MPERF MSRs, which were cleared by the acpi_cpufreq
driver up through Linux-2.6.29.

On Intel Core i3/i5/i7 (Nehalem) and newer processors,
turbostat displays residency in idle power saving states.

See the comments in the source code for example usage.

Signed-off-by: Len Brown <[email protected]>
Reviewed-by: Otavio Salvador

tools/power/x86/turbostat/Makefile | 7 +
tools/power/x86/turbostat/turbostat.c | 1107 +++++++++++++++++++++++++++++++++
2 files changed, 1114 insertions(+), 0 deletions(-)
create mode 100644 tools/power/x86/turbostat/Makefile
create mode 100644 tools/power/x86/turbostat/turbostat.c

diff --git a/tools/power/x86/turbostat/Makefile b/tools/power/x86/turbostat/Makefile
new file mode 100644
index 0000000..5a18f55
--- /dev/null
+++ b/tools/power/x86/turbostat/Makefile
@@ -0,0 +1,7 @@
+turbostat : turbostat.c
+
+clean :
+ rm -f turbostat
+
+install :
+ install turbostat /usr/bin/turbostat
diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
new file mode 100644
index 0000000..e4526a6
--- /dev/null
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -0,0 +1,1107 @@
+/*
+ * turbostat -- show CPU frequency and C-state residency
+ * on modern Intel turbo-capable processors.
+ *
+ * Copyright (c) 2010, Intel Corporation.
+ * Len Brown <[email protected]>
+ *
+ * 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.
+ */
+
+/*
+ * Works properly on Nehalem and newer processors.
+ * Nehalem features an always-running TSC, plus hardware
+ * C-state residency MSRs.
+ *
+ * Works poorly on systems before Nehalem with
+ * a TSC that stops in deep C-states.
+ *
+ * Works properly on Linux-2.6.30 and later.
+ * Works poorly Linux-2.6.29 and earlier, as acpi-cpufreq
+ * used to clear APERF/MPERF counters on access.
+ *
+ * APERF, MPERF count non-halted cycles.
+ * Although it is not guaranteed by the architecture, we assume
+ * here that they count at TSC rate, which is true today.
+ *
+ * References:
+ * "Intel® Turbo Boost Technology
+ * in Intel® Core™ Microarchitecture (Nehalem) Based Processors"
+ * http://download.intel.com/design/processor/applnots/320354.pdf
+ *
+ * "Intel® 64 and IA-32 Architectures Software Developer's Manual
+ * Volume 3B: System Programming Guide"
+ * http://www.intel.com/products/processor/manuals/
+ */
+
+/*
+ * usage:
+ * # turbostat [-v] [-i interval_sec] [command [arg]...]
+ *
+ * examples:
+ *
+ * [root@x980]# ./turbostat
+ *core CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
+ * 0.04 1.62 3.38 0.11 0.00 99.85 0.00 95.07
+ * 0 0 0.04 1.62 3.38 0.06 0.00 99.90 0.00 95.07
+ * 0 6 0.02 1.62 3.38 0.08 0.00 99.90 0.00 95.07
+ * 1 2 0.10 1.62 3.38 0.29 0.00 99.61 0.00 95.07
+ * 1 8 0.11 1.62 3.38 0.28 0.00 99.61 0.00 95.07
+ * 2 4 0.01 1.62 3.38 0.01 0.00 99.98 0.00 95.07
+ * 2 10 0.01 1.61 3.38 0.02 0.00 99.98 0.00 95.07
+ * 8 1 0.07 1.62 3.38 0.15 0.00 99.78 0.00 95.07
+ * 8 7 0.03 1.62 3.38 0.19 0.00 99.78 0.00 95.07
+ * 9 3 0.01 1.62 3.38 0.02 0.00 99.98 0.00 95.07
+ * 9 9 0.01 1.62 3.38 0.02 0.00 99.98 0.00 95.07
+ * 10 5 0.01 1.62 3.38 0.13 0.00 99.86 0.00 95.07
+ * 10 11 0.08 1.62 3.38 0.05 0.00 99.86 0.00 95.07
+ *
+ * Without any parameters, turbostat prints out counters ever 5 seconds.
+ * (override interval with "-i sec" option).
+ *
+ * core is the processor core number
+ *
+ * CPU is the cpu number according to Linux, ala /dev/cpu
+ *
+ * The first row of data is a system wide summary, and thus has no core or CPU#.
+ *
+ * %c0 is the percent of the interval that the core retired instructions.
+ *
+ * GHz is the average clock rate while the core was in c0 state.
+ *
+ * TSC is the average GHz that the TSC ran during the entire interval.
+ *
+ * %c1, %c3, %c6 show the residency in hardware CPU core idle states.
+ * Note that these may not equal the software states requested by Linux.
+ *
+ * pc3%, pc6%, show package idle C-states.
+ *
+ * The "-v" option adds verbosity to the output: eg.
+ *
+ * CPUID GenuineIntel 11 levels family:model:stepping 0x6:2c:2 (6:44:2)
+ * 12 * 133 = 1600 MHz max efficiency
+ * 25 * 133 = 3333 MHz TSC frequency
+ * 26 * 133 = 3467 MHz max turbo 4 active cores
+ * 26 * 133 = 3467 MHz max turbo 3 active cores
+ * 27 * 133 = 3600 MHz max turbo 2 active cores
+ * 27 * 133 = 3600 MHz max turbo 1 active cores
+ *
+ * This tells you what you should be able to get, given nominal
+ * cooling and electrical capacity. Note that NHM-EX/WSM-EX do
+ * not tell us these max turbo frequencies like NHM/NHM-EP/WSM/WSM-EP do.
+ *
+ * If turbostat is invoked with a command, it will fork that command
+ * and output the statistics gathered while that command was running.
+ * eg. Here a cycle soaker is run on 1 CPU (see %c0) for a few seconds
+ * until ^C while the others are mostly idle:
+ *
+ * [root@x980 lenb]# ./turbostat cat /dev/zero > /dev/null
+ *
+ *^Ccore CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
+ * 8.49 3.63 3.38 16.23 0.66 74.63 0.00 0.00
+ * 0 0 1.22 3.62 3.38 32.18 0.00 66.60 0.00 0.00
+ * 0 6 0.40 3.61 3.38 33.00 0.00 66.60 0.00 0.00
+ * 1 2 0.11 3.14 3.38 0.19 3.95 95.75 0.00 0.00
+ * 1 8 0.05 2.88 3.38 0.25 3.95 95.75 0.00 0.00
+ * 2 4 0.00 3.13 3.38 0.02 0.00 99.98 0.00 0.00
+ * 2 10 0.00 3.09 3.38 0.02 0.00 99.98 0.00 0.00
+ * 8 1 0.04 3.50 3.38 14.43 0.00 85.54 0.00 0.00
+ * 8 7 0.03 2.98 3.38 14.43 0.00 85.54 0.00 0.00
+ * 9 3 0.00 3.16 3.38 100.00 0.00 0.00 0.00 0.00
+ * 9 9 99.93 3.63 3.38 0.06 0.00 0.00 0.00 0.00
+ * 10 5 0.01 2.82 3.38 0.08 0.00 99.91 0.00 0.00
+ * 10 11 0.02 3.36 3.38 0.06 0.00 99.91 0.00 0.00
+ * 6.950866 sec
+ *
+ * Above the cycles soaker drives cpu9 up 3.6 Ghz turbo limit
+ * while the other processors are generally in various states of idle.
+ *
+ * Note that cpu3 is an HT sibling sharing the same core (9)
+ * with cpu9, and thus it is unble to get more idle than c1
+ * when cpu9 is busy.
+ *
+ * Note that the arithmetic average of the GHz column above is 3.24,
+ * yet turbostat shows avg 3.61. This is a weighted average --
+ * where the weight is %c0. ie. it is the total number of
+ * unhalted cycles elapsed per time divided by the number of CPUs.
+ *
+ * Note that turbostat reads hardware counters, but doesn't write them.
+ * So it will not interfere with the OS or other programs, including
+ * multiple invocations of itself.
+ *
+ * turbostat depends on the Linux msr driver for /dev/cpu/.../msr
+ */
+
+#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 <dirent.h>
+
+#define MSR_TSC 0x10
+#define MSR_NEHALEM_PLATFORM_INFO 0xCE
+#define MSR_NEHALEM_TURBO_RATIO_LIMIT 0x1AD
+#define MSR_APERF 0xE8
+#define MSR_MPERF 0xE7
+#define MSR_PKG_C3_RESIDENCY 0x3F8
+#define MSR_PKG_C6_RESIDENCY 0x3F9
+#define MSR_CORE_C3_RESIDENCY 0x3FC
+#define MSR_CORE_C6_RESIDENCY 0x3FD
+
+char *proc_stat = "/proc/stat";
+unsigned int interval_sec = 5; /* set with -i interval_sec */
+unsigned int verbose; /* set with -v */
+unsigned int skip_c0;
+unsigned int skip_c1;
+unsigned int do_nhm_cstates;
+unsigned int do_snb_cstates;
+unsigned int do_aperf = 1; /* TBD set with CPUID */
+unsigned int units = 1000000000; /* Ghz etc */
+unsigned int do_non_stop_tsc;
+unsigned int do_nehalem_platform_info;
+unsigned int do_nehalem_turbo_ratio_limit;
+unsigned int extra_msr_offset;
+double bclk;
+unsigned int show_pkg;
+unsigned int show_core;
+unsigned int show_cpu;
+
+int aperf_mperf_unstable;
+int backwards_count;
+char *progname;
+int need_reinitialize;
+
+int num_cpus;
+
+typedef struct per_cpu_counters {
+ unsigned long long tsc; /* per thread */
+ unsigned long long aperf; /* per thread */
+ unsigned long long mperf; /* per thread */
+ unsigned long long c1; /* per thread (calculated) */
+ unsigned long long c3; /* per core */
+ unsigned long long c6; /* per core */
+ unsigned long long c7; /* per core */
+ unsigned long long pc2; /* per package */
+ unsigned long long pc3; /* per package */
+ unsigned long long pc6; /* per package */
+ unsigned long long pc7; /* per package */
+ unsigned long long extra_msr; /* per thread */
+ int pkg;
+ int core;
+ int cpu;
+ struct per_cpu_counters *next;
+} PCC;
+
+PCC *pcc_even;
+PCC *pcc_odd;
+PCC *pcc_delta;
+PCC *pcc_average;
+struct timeval tv_even;
+struct timeval tv_odd;
+struct timeval tv_delta;
+
+unsigned long long get_msr(int cpu, off_t offset)
+{
+ ssize_t retval;
+ unsigned long long msr;
+ char pathname[32];
+ int fd;
+
+ sprintf(pathname, "/dev/cpu/%d/msr", cpu);
+ fd = open(pathname, O_RDONLY);
+ if (fd < 0) {
+ perror(pathname);
+ need_reinitialize = 1;
+ return 0;
+ }
+
+ retval = pread(fd, &msr, sizeof msr, offset);
+ if (retval != sizeof msr) {
+ fprintf(stderr, "cpu%d pread(..., 0x%x) = %d\n",
+ cpu, offset, retval);
+ exit(-2);
+ }
+
+ close(fd);
+ return msr;
+}
+
+void print_header()
+{
+ if (show_pkg)
+ fprintf(stderr, "pkg ");
+ if (show_core)
+ fprintf(stderr, "core");
+ if (show_cpu)
+ fprintf(stderr, " CPU");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c0 ");
+ fprintf(stderr, " GHz");
+ fprintf(stderr, " TSC");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c1 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c3 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c6 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%pc3 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%pc6 ");
+ if (extra_msr_offset)
+ fprintf(stderr, " MSR 0x%x ", extra_msr_offset);
+
+ putc('\n', stderr);
+}
+
+void dump_pcc(PCC *pcc)
+{
+ fprintf(stderr, "package: %d ", pcc->pkg);
+ fprintf(stderr, "core:: %d ", pcc->core);
+ fprintf(stderr, "CPU: %d ", pcc->cpu);
+ fprintf(stderr, "TSC: %016llX\n", pcc->tsc);
+ fprintf(stderr, "c3: %016llX\n", pcc->c3);
+ fprintf(stderr, "c6: %016llX\n", pcc->c6);
+ fprintf(stderr, "c7: %016llX\n", pcc->c7);
+ fprintf(stderr, "aperf: %016llX\n", pcc->aperf);
+ fprintf(stderr, "pc2: %016llX\n", pcc->pc2);
+ fprintf(stderr, "pc3: %016llX\n", pcc->pc3);
+ fprintf(stderr, "pc6: %016llX\n", pcc->pc6);
+ fprintf(stderr, "pc7: %016llX\n", pcc->pc7);
+ fprintf(stderr, "msr0x%x: %016llX\n", extra_msr_offset, pcc->extra_msr);
+}
+
+void dump_list(PCC *pcc)
+{
+ printf("dump_list 0x%p\n", pcc);
+
+ for (; pcc; pcc = pcc->next)
+ dump_pcc(pcc);
+}
+
+void print_pcc(PCC *p)
+{
+ double interval_float;
+
+ interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
+
+ if (p == pcc_average) {
+ if (show_pkg)
+ fprintf(stderr, " ");
+ if (show_core)
+ fprintf(stderr, " ");
+ if (show_cpu)
+ fprintf(stderr, " ");
+ } else {
+ if (show_pkg)
+ fprintf(stderr, "%4d", p->pkg);
+ if (show_core)
+ fprintf(stderr, "%4d", p->core);
+ if (show_cpu)
+ fprintf(stderr, "%4d", p->cpu);
+ }
+
+ /* C0 */
+ if (do_nhm_cstates) {
+ if (!skip_c0)
+ fprintf(stderr, "%7.2f", 100.0 * p->mperf/p->tsc);
+ else
+ fprintf(stderr, " ****");
+ }
+
+ if (do_aperf) {
+ if (!aperf_mperf_unstable) {
+ fprintf(stderr, "%5.2f",
+ 1.0 * p->tsc / units * p->aperf /
+ p->mperf / interval_float);
+ } else {
+ if (p->aperf > p->tsc || p->mperf > p->tsc) {
+ fprintf(stderr, " ****");
+ } else {
+ fprintf(stderr, "%4.1f*",
+ 1.0 * p->tsc /
+ units * p->aperf /
+ p->mperf / interval_float);
+ }
+ }
+ }
+
+ fprintf(stderr, "%5.2f", 1.0 * p->tsc/units/interval_float);
+
+ if (do_nhm_cstates) {
+ if (!skip_c1)
+ fprintf(stderr, "%7.2f", 100.0 * p->c1/p->tsc);
+ else
+ fprintf(stderr, " ****");
+ }
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c3/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c6/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc3/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc6/p->tsc);
+ if (extra_msr_offset)
+ fprintf(stderr, " 0x%016llx", p->extra_msr);
+ putc('\n', stderr);
+}
+
+void print_counters(PCC *cnt)
+{
+ int i;
+ PCC *pcc;
+
+ print_header();
+
+ if (num_cpus > 1)
+ print_pcc(pcc_average);
+
+ for (pcc = cnt; pcc != NULL; pcc = pcc->next)
+ print_pcc(pcc);
+
+}
+
+#define SUBTRACT_COUNTER(after, before, delta) (delta = (after - before), (before > after))
+
+
+int compute_delta(PCC *after, PCC *before, PCC *delta)
+{
+ int i;
+ int errors = 0;
+ int perf_err = 0;
+
+ skip_c0 = skip_c1 = 0;
+
+ for ( ; after && before && delta;
+ after = after->next, before = before->next, delta = delta->next) {
+ if (before->cpu != after->cpu) {
+ printf("cpu configuration changed: %d != %d\n",
+ before->cpu, after->cpu);
+ return -1;
+ }
+
+ if (SUBTRACT_COUNTER(after->tsc, before->tsc, delta->tsc)) {
+ fprintf(stderr, "cpu%d TSC went backwards %llX to %llX\n",
+ i, before->tsc, after->tsc);
+ errors++;
+ }
+ /* check for TSC < 1 Mcycles over interval */
+ if (delta->tsc < (1000 * 1000)) {
+ fprintf(stderr, "Insanely slow TSC rate,"
+ " TSC stops in idle?\n");
+ fprintf(stderr, "You can disable all c-states"
+ " by booting with \"idle=poll\"\n");
+ fprintf(stderr, "or just the deep ones with"
+ " \"processor.max_cstate=1\"\n");
+ exit(-3);
+ }
+ if (SUBTRACT_COUNTER(after->c3, before->c3, delta->c3)) {
+ fprintf(stderr, "cpu%d c3 counter went backwards %llX to %llX\n",
+ i, before->c3, after->c3);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->c6, before->c6, delta->c6)) {
+ fprintf(stderr, "cpu%d c6 counter went backwards %llX to %llX\n",
+ i, before->c6, after->c6);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->c7, before->c7, delta->c7)) {
+ fprintf(stderr, "cpu%d c7 counter went backwards %llX to %llX\n",
+ i, before->c7, after->c7);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc2, before->pc2, delta->pc2)) {
+ fprintf(stderr, "cpu%d pc2 counter went backwards %llX to %llX\n",
+ i, before->pc2, after->pc2);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc3, before->pc3, delta->pc3)) {
+ fprintf(stderr, "cpu%d pc3 counter went backwards %llX to %llX\n",
+ i, before->pc3, after->pc3);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc6, before->pc6, delta->pc6)) {
+ fprintf(stderr, "cpu%d pc6 counter went backwards %llX to %llX\n",
+ i, before->pc6, after->pc6);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc7, before->pc7, delta->pc7)) {
+ fprintf(stderr, "cpu%d pc7 counter went backwards %llX to %llX\n",
+ i, before->pc7, after->pc7);
+ errors++;
+ }
+
+ perf_err = SUBTRACT_COUNTER(after->aperf, before->aperf, delta->aperf);
+ if (perf_err) {
+ fprintf(stderr, "cpu%d aperf counter went backwards %llX to %llX\n",
+ i, before->aperf, after->aperf);
+ }
+ perf_err |= SUBTRACT_COUNTER(after->mperf, before->mperf, delta->mperf);
+ if (perf_err) {
+ fprintf(stderr, "cpu%d mperf counter went backwards %llX to %llX\n",
+ i, before->mperf, after->mperf);
+ }
+ if (perf_err) {
+ if (!aperf_mperf_unstable) {
+ fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
+ fprintf(stderr, "* Frequency results do not cover entire interval *\n");
+ fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
+
+ aperf_mperf_unstable = 1;
+ }
+ /*
+ * mperf delta is likely a huge "positive" number
+ * can not use it for calculating c0 time
+ */
+ skip_c0 = 1;
+ skip_c1 = 1;
+ }
+
+ /*
+ * As mperf and tsc collection are not atomic,
+ * it is possible for mperf's non-halted cycles
+ * to exceed TSC's all cycles: show c1l = 0% in that case.
+ */
+ if (delta->mperf > delta->tsc)
+ delta->c1 = 0;
+ else /* normal case, derive c1 */
+ delta->c1 = delta->tsc - delta->mperf
+ - delta->c3 - delta->c6 - delta->c7;
+
+ if (delta->mperf == 0)
+ delta->mperf = 1; /* divide by 0 protection */
+
+ /*
+ * for "extra msr", just copy the latest w/o subtracting
+ */
+ delta->extra_msr = after->extra_msr;
+ if (errors) {
+ fprintf(stderr, "ERROR cpu%d before:\n", i);
+ dump_pcc(before);
+ fprintf(stderr, "ERROR cpu%d after:\n", i);
+ dump_pcc(after);
+ errors = 0;
+ }
+ }
+ return 0;
+}
+
+void compute_average(PCC *delta, PCC *avg)
+{
+ int i;
+ PCC *sum;
+
+ sum = calloc(1, sizeof(PCC));
+ if (sum == NULL) {
+ perror("calloc sum");
+ exit(-1);
+ }
+
+ for (; delta; delta = delta->next) {
+ sum->tsc += delta->tsc;
+ sum->c1 += delta->c1;
+ sum->c3 += delta->c3;
+ sum->c6 += delta->c6;
+ sum->c7 += delta->c7;
+ sum->aperf += delta->aperf;
+ sum->mperf += delta->mperf;
+ sum->pc2 += delta->pc2;
+ sum->pc3 += delta->pc3;
+ sum->pc6 += delta->pc6;
+ sum->pc7 += delta->pc7;
+ }
+ avg->tsc = sum->tsc/num_cpus;
+ avg->c1 = sum->c1/num_cpus;
+ avg->c3 = sum->c3/num_cpus;
+ avg->c6 = sum->c6/num_cpus;
+ avg->c7 = sum->c7/num_cpus;
+ avg->aperf = sum->aperf/num_cpus;
+ avg->mperf = sum->mperf/num_cpus;
+ avg->pc2 = sum->pc2/num_cpus;
+ avg->pc3 = sum->pc3/num_cpus;
+ avg->pc6 = sum->pc6/num_cpus;
+ avg->pc7 = sum->pc7/num_cpus;
+
+ free(sum);
+}
+
+void get_counters(PCC *pcc)
+{
+ for ( ; pcc; pcc = pcc->next) {
+ pcc->tsc = get_msr(pcc->cpu, MSR_TSC);
+ if (do_nhm_cstates)
+ pcc->c3 = get_msr(pcc->cpu, MSR_CORE_C3_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->c6 = get_msr(pcc->cpu, MSR_CORE_C6_RESIDENCY);
+ if (do_aperf)
+ pcc->aperf = get_msr(pcc->cpu, MSR_APERF);
+ if (do_aperf)
+ pcc->mperf = get_msr(pcc->cpu, MSR_MPERF);
+ if (do_nhm_cstates)
+ pcc->pc3 = get_msr(pcc->cpu, MSR_PKG_C3_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->pc6 = get_msr(pcc->cpu, MSR_PKG_C6_RESIDENCY);
+ if (extra_msr_offset)
+ pcc->extra_msr = get_msr(pcc->cpu, extra_msr_offset);
+ }
+}
+
+
+void print_nehalem_info()
+{
+ unsigned long long msr;
+ unsigned int ratio;
+
+ if (!do_nehalem_platform_info)
+ return;
+
+ msr = get_msr(0, MSR_NEHALEM_PLATFORM_INFO);
+
+ ratio = (msr >> 40) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 8) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz TSC frequency\n",
+ ratio, bclk, ratio * bclk);
+
+ if (verbose > 1)
+ fprintf(stderr, "MSR_NEHALEM_PLATFORM_INFO: 0x%llx\n", msr);
+
+ if (!do_nehalem_turbo_ratio_limit)
+ return;
+
+ msr = get_msr(0, MSR_NEHALEM_TURBO_RATIO_LIMIT);
+
+ ratio = (msr >> 24) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 16) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 8) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 0) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+}
+
+void free_counter_list(PCC *list)
+{
+ PCC *p;
+
+ for (p = list; p; ) {
+ PCC *free_me;
+
+ free_me = p;
+ p = p->next;
+ free(free_me);
+ }
+ return;
+}
+
+void free_all_counters(void)
+{
+ free_counter_list(pcc_even);
+ pcc_even = NULL;
+
+ free_counter_list(pcc_odd);
+ pcc_odd = NULL;
+
+ free_counter_list(pcc_delta);
+ pcc_delta = NULL;
+
+ free_counter_list(pcc_average);
+ pcc_average = NULL;
+}
+
+void insert_cpu_counters(PCC **list, PCC *new)
+{
+ PCC *prev, *next;
+
+ /*
+ * list was empty
+ */
+ if (*list == NULL) {
+ new->next = *list;
+ *list = new;
+ return;
+ }
+
+ /*
+ * insert on front of list.
+ * It is sorted by ascending package#, core#, cpu#
+ */
+ if (((*list)->pkg > new->pkg) ||
+ (((*list)->pkg == new->pkg) && ((*list)->core > new->core)) ||
+ (((*list)->pkg == new->pkg) && ((*list)->core == new->core) && ((*list)->cpu > new->cpu))) {
+ new->next = *list;
+ *list = new;
+ return;
+ }
+
+ prev = *list;
+
+ while (prev->next && (prev->next->pkg < new->pkg)) {
+ prev = prev->next;
+ show_pkg = 1; /* there is more than 1 package */
+ }
+
+ while (prev->next && (prev->next->pkg == new->pkg)
+ && (prev->next->core < new->core)) {
+ prev = prev->next;
+ show_core = 1; /* there is more than 1 core */
+ show_cpu = 1; /* there is more than 1 CPU (HW thread) */
+ }
+
+ while (prev->next && (prev->next->pkg == new->pkg)
+ && (prev->next->core == new->core)
+ && (prev->next->cpu < new->cpu)) {
+ prev = prev->next;
+ show_cpu = 1;
+ }
+
+ /*
+ * insert after "prev"
+ */
+ new->next = prev->next;
+ prev->next = new;
+
+ return;
+}
+
+void alloc_new_cpu_counters(int pkg, int core, int cpu)
+{
+ PCC *new;
+
+ if (verbose > 1)
+ printf("pkg%d core%d, cpu%d\n", pkg, core, cpu);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(-1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_odd, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(-1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_even, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(-1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_delta, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(-1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ pcc_average = new;
+}
+
+void re_initialize(void)
+{
+ printf("turbostat: topology changed, re-initializing.\n");
+ free_all_counters();
+ num_cpus = for_all_cpus(alloc_new_cpu_counters);
+ need_reinitialize = 0;
+ printf("num_cpus is now %d\n", num_cpus);
+}
+
+void dummy(int pkg, int core, int cpu) { return; }
+/*
+ * check to see if a cpu came on-line
+ */
+void verify_num_cpus()
+{
+ int new_num_cpus;
+
+ new_num_cpus = for_all_cpus(dummy);
+
+ if (new_num_cpus != num_cpus) {
+ if (verbose)
+ printf("num_cpus was %d, is now %d\n",
+ num_cpus, new_num_cpus);
+ need_reinitialize = 1;
+ }
+
+ return;
+}
+
+void turbostat_loop()
+{
+restart:
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+
+ while (1) {
+ int retval;
+
+ verify_num_cpus();
+ if (need_reinitialize) {
+ re_initialize();
+ goto restart;
+ }
+ sleep(interval_sec);
+ get_counters(pcc_odd);
+ gettimeofday(&tv_odd, (struct timezone *)NULL);
+
+ compute_delta(pcc_odd, pcc_even, pcc_delta);
+ timersub(&tv_odd, &tv_even, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ print_counters(pcc_delta);
+ if (need_reinitialize) {
+ re_initialize();
+ goto restart;
+ }
+ sleep(interval_sec);
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+ compute_delta(pcc_even, pcc_odd, pcc_delta);
+ timersub(&tv_even, &tv_odd, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ print_counters(pcc_delta);
+ }
+}
+
+void check_dev_msr()
+{
+ struct stat sb;
+
+ if (stat("/dev/cpu/0/msr", &sb)) {
+ fprintf(stderr, "no /dev/cpu/0/msr\n");
+ fprintf(stderr, "Try \"# modprobe msr\"\n");
+ exit(-5);
+ }
+}
+
+int has_nehalem_turbo_ratio_limit(unsigned int family, unsigned int model)
+{
+ if (family != 6)
+ return 0;
+
+ switch (model) {
+ case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
+ case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
+ case 0x1F: /* Core i7 and i5 Processor - Nehalem */
+ case 0x25: /* Westmere Client - Clarkdale, Arrandale */
+ case 0x2C: /* Westmere EP - Gulftown */
+ case 0x2A: /* SNB */
+ case 0x2D: /* SNB Xeon */
+ return 1;
+ case 0x2E: /* Nehalem-EX Xeon - Beckton */
+ case 0x2F: /* Westmere-EX Xeon - Eagleton */
+ default:
+ return 0;
+ }
+}
+
+int is_snb(unsigned int family, unsigned int model)
+{
+ switch (model) {
+ case 0x2A:
+ case 0x2D:
+ return 1;
+ }
+ return 0;
+}
+
+double discover_bclk(unsigned int family, unsigned int model)
+{
+ if (is_snb(family, model))
+ return 100.00;
+ else
+ return 133.33;
+}
+
+void check_cpuid()
+{
+ 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));
+
+ sprintf(brand, "%.4s%.4s%.4s", &ebx, &edx, &ecx);
+
+ if (strncmp(brand, "GenuineIntel", 12)) {
+ fprintf(stderr, "CPUID: %s != GenuineIntel\n", brand);
+ 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)
+ fprintf(stderr, "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))) {
+ fprintf(stderr, "CPUID: no MSR\n");
+ exit(-1);
+ }
+
+ ebx = ecx = edx = 0;
+ asm("cpuid" : "=a" (max_level), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x80000000));
+
+ if (max_level >= 0x80000007) {
+ /*
+ * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
+ */
+ asm("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x80000007));
+ do_non_stop_tsc = edx && (1 << 8);
+ }
+ do_nehalem_platform_info = do_non_stop_tsc;
+ do_nhm_cstates = do_non_stop_tsc;
+ do_snb_cstates = is_snb(family, model);
+ if (do_snb_cstates)
+ fprintf(stderr, "This version does not support SNB\n");
+ bclk = discover_bclk(family, model);
+
+ do_nehalem_turbo_ratio_limit = has_nehalem_turbo_ratio_limit(family, model);
+}
+
+
+usage() {
+ fprintf(stderr, "%s: [-v] [-i interval_sec] [-M MSR#] [command [arg]...]\n",
+ progname);
+}
+
+int get_physical_package_id(int cpu)
+{
+ char path[64];
+ FILE *filep;
+ int pkg;
+
+ sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
+ filep = fopen(path, "r");
+ if (filep == NULL) {
+ perror(path);
+ exit(-1);
+ }
+ fscanf(filep, "%d", &pkg);
+ fclose(filep);
+ return pkg;
+}
+
+int get_core_id(int cpu)
+{
+ char path[64];
+ FILE *filep;
+ int core;
+
+ sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
+ filep = fopen(path, "r");
+ if (filep == NULL) {
+ perror(path);
+ exit(-1);
+ }
+ fscanf(filep, "%d", &core);
+ fclose(filep);
+ return core;
+}
+
+/*
+ * in /dev/cpu/ return success for names that are numbers
+ * ie. filter out ".", "..", "microcode".
+ */
+int dir_filter(const struct dirent *dirp)
+{
+ if (isdigit(dirp->d_name[0]))
+ return 1;
+ else
+ return 0;
+}
+
+int open_dev_cpu_msr(int dummy1)
+{
+ return 0;
+}
+
+/*
+ * run func(index, cpu) on every cpu in /proc/stat
+ */
+
+int for_all_cpus(void (func)(int, int, int))
+{
+ FILE *fp;
+ int cpu_count;
+ 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);
+ }
+
+ for (cpu_count = 0; ; cpu_count++) {
+ int cpu;
+
+ retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu);
+ if (retval != 1)
+ return cpu_count;
+
+ func(get_physical_package_id(cpu), get_core_id(cpu), cpu);
+ }
+ fclose(fp);
+}
+
+void turbostat_init()
+{
+ int i;
+
+ check_cpuid();
+
+ check_dev_msr();
+
+ num_cpus = for_all_cpus(alloc_new_cpu_counters);
+
+ if (verbose)
+ print_nehalem_info();
+}
+
+int fork_it(char **argv)
+{
+ int retval;
+ pid_t child_pid;
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+
+ child_pid = fork();
+ if (!child_pid) {
+ /* child */
+ execvp(argv[0], argv);
+ } else {
+ int status;
+
+ /* parent */
+ if (child_pid == -1) {
+ perror("fork");
+ exit(-1);
+ }
+
+ signal(SIGINT, SIG_IGN);
+ signal(SIGQUIT, SIG_IGN);
+ if (waitpid(child_pid, &status, 0) == -1) {
+ perror("wait");
+ exit(-1);
+ }
+ }
+ get_counters(pcc_odd);
+ gettimeofday(&tv_odd, (struct timezone *)NULL);
+ retval = compute_delta(pcc_odd, pcc_even, pcc_delta);
+
+ timersub(&tv_odd, &tv_even, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ if (!retval)
+ print_counters(pcc_delta);
+
+ fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);;
+
+ return 0;
+}
+
+void cmdline(int argc, char **argv)
+{
+ int opt;
+
+ progname = argv[0];
+
+ while ((opt = getopt(argc, argv, "+vi:M:")) != -1) {
+ switch (opt) {
+ case 'v':
+ verbose++;
+ show_pkg = 1;
+ show_core = 1;
+ show_cpu = 1;
+ break;
+ case 'i':
+ interval_sec = atoi(optarg);
+ break;
+ case 'M':
+ sscanf(optarg, "%x", &extra_msr_offset);
+ if (verbose > 1)
+ fprintf(stderr, "MSR 0x%X\n", extra_msr_offset);
+ break;
+ default:
+ usage();
+ exit(-1);
+ }
+ }
+}
+
+int main(int argc, char **argv)
+{
+ cmdline(argc, argv);
+
+ if (verbose > 1)
+ fprintf(stderr, "turbostat Aug 25, 2010"
+ " - Len Brown <[email protected]>\n");
+ if (verbose > 1)
+ fprintf(stderr, "http://userweb.kernel.org/~lenb/acpi/utils/pmtools/turbostat/\n");
+
+ turbostat_init();
+
+ /*
+ * if any params left, it must be a command to fork
+ */
+ if (argc - optind)
+ return fork_it(argv + optind);
+ else
+ turbostat_loop();
+
+ return 0;
+}
--
1.7.3.2.90.gd4c43

--8323328-1715002308-1288149298=:3209--

2010-11-15 17:16:52

by Otavio Salvador

Subject: Re: [PATCH RESEND] tools: add power/x86/turbostat - v2

On Mon, Nov 15, 2010 at 2:01 PM, Len Brown <[email protected]> wrote:
> From: Len Brown <[email protected]>
>
> turbostat displays actual processor frequency on modern
> Intel processors. ?Note that this capability depends on free-running
> APERF and MPERF MSRs, which were cleared by the acpi_cpufreq
> driver up through Linux-2.6.29.
>
> On Intel Core i3/i5/i7 (Nehalem) and newer processors,
> turbostat displays residency in idle power saving states.
>
> See the comments in the source code for example usage.
>
> Signed-off-by: Len Brown <[email protected]>
> Reviewed-by: Otavio Salvador

Reviewed-by: Otavio Salvador <[email protected]>

--
Otavio Salvador? ? ? ? ? ? ? ? ? O.S. Systems
E-mail: [email protected]? http://www.ossystems.com.br
Mobile: +55 53 9981-7854? ? ? ?? http://projetos.ossystems.com.br

2010-11-16 10:29:23

by Andreas Herrmann

Subject: Re: [PATCH RESEND] tools: add power/x86/turbostat - v2

On Mon, Nov 15, 2010 at 11:01:05AM -0500, Len Brown wrote:
> From: Len Brown <[email protected]>
>
> turbostat displays actual processor frequency on modern
> Intel processors. Note that this capability depends on free-running
> APERF and MPERF MSRs, which were cleared by the acpi_cpufreq
> driver up through Linux-2.6.29.

Why isn't cpufreq-aperf sufficient for showing the actual frequency?
The tool can handle both AMD and Intel CPUs.

There is a CPUID flag which can be checked for APERF/MPERF existence,
no need to enforce a general vendor check like this:

$ ./turbostat
CPUID: AuthenticAMD != GenuineIntel

> On Intel Core i3/i5/i7 (Nehalem) and newer processors,
> turbostat displays residency in idle power saving states.

Can't this be added to cpuidle-info (see branch cpupowerutils in
git://git.kernel.org/pub/scm/utils/kernel/cpufreq/cpufrequtils.git)?

BTW, gcc v4.4 throws several warnings when compiling your
code:

linux-2.6/tools/power/x86/turbostat $ make
cc -Wall turbostat.c -o turbostat
turbostat.c: In function ‘get_msr’:
turbostat.c:234: warning: implicit declaration of function ‘pread’
turbostat.c:237: warning: format ‘%x’ expects type ‘unsigned int’, but argument 4 has type ‘off_t’
turbostat.c:237: warning: format ‘%d’ expects type ‘int’, but argument 5 has type ‘ssize_t’
turbostat.c: In function ‘print_counters’:
turbostat.c:368: warning: unused variable ‘i’
turbostat.c: In function ‘compute_average’:
turbostat.c:508: warning: unused variable ‘i’
turbostat.c: In function ‘insert_cpu_counters’:
turbostat.c:642: warning: unused variable ‘next’
turbostat.c: In function ‘re_initialize’:
turbostat.c:747: warning: implicit declaration of function ‘for_all_cpus’
turbostat.c: In function ‘turbostat_loop’:
turbostat.c:779: warning: unused variable ‘retval’
turbostat.c: In function ‘check_cpuid’:
turbostat.c:868: warning: format ‘%.4s’ expects type ‘char *’, but argument 3 has type ‘unsigned int *’
turbostat.c:868: warning: format ‘%.4s’ expects type ‘char *’, but argument 4 has type ‘unsigned int *’
turbostat.c:868: warning: format ‘%.4s’ expects type ‘char *’, but argument 5 has type ‘unsigned int *’
turbostat.c:870: warning: implicit declaration of function ‘strncmp’
turbostat.c: At top level:
turbostat.c:912: warning: return type defaults to ‘int’
turbostat.c: In function ‘dir_filter’:
turbostat.c:957: warning: implicit declaration of function ‘isdigit’
turbostat.c: In function ‘turbostat_init’:
turbostat.c:1004: warning: unused variable ‘i’
turbostat.c: In function ‘fork_it’:
turbostat.c:1038: warning: implicit declaration of function ‘waitpid’
turbostat.c: In function ‘usage’:
turbostat.c:915: warning: control reaches end of non-void function
turbostat.c: In function ‘compute_delta’:
turbostat.c:401: warning: ‘i’ may be used uninitialized in this function


Regards,
Andreas

2010-11-17 06:45:12

by Len Brown

Subject: Re: [PATCH RESEND] tools: add power/x86/turbostat - v2

On Tue, 16 Nov 2010, Andreas Herrmann wrote:

> On Mon, Nov 15, 2010 at 11:01:05AM -0500, Len Brown wrote:
> > From: Len Brown <[email protected]>
> >
> > turbostat displays actual processor frequency on modern
> > Intel processors. Note that this capability depends on free-running
> > APERF and MPERF MSRs, which were cleared by the acpi_cpufreq
> > driver up through Linux-2.6.29.
>
> Why isn't cpufreq-aperf sufficient for showing the actual frequency?
> The tool can handle both AMD and Intel CPUs.

Looking over its source code, I don't see any reason that cpufreq-aperf
wouldn't be sufficient for showing actual frequency, were that your goal.
I hadn't heard of this tool until your note. I complied it, but I guess
I'm too lazy to figure out why it didnt' run.
It seems somewhat elaborate as compared to the task at hand...

> There is a CPUID flag which can be checked for APERF/MPERF existence,
> no need to enforce a general vendor check like this:
>
> $ ./turbostat
> CPUID: AuthenticAMD != GenuineIntel

turbostat isn't very interesting on systems that don't
have a reliable TSC or C-state residency MSRs. AFAIK,
AMD has neither. If you have information to the contrary
and can test it on AMD hardware, I'd be happy to enable AMD,
as I have neither.

> > On Intel Core i3/i5/i7 (Nehalem) and newer processors,
> > turbostat displays residency in idle power saving states.
>
> Can't this be added to cpuidle-info (see branch cpupowerutils in
> git://git.kernel.org/pub/scm/utils/kernel/cpufreq/cpufrequtils.git)?

I suppose it could if somebody were in need of a hobby.
However, if you want a big tool in this space,
you'd probably be better off running powertop2,
which will incorporate what turbostat does, and more.

> BTW, gcc v4.4 throws several warnings when compiling your
> code:

thanks, I'll fix these.

cheers,
Len Brown, Intel Open Source Technology Center

2010-11-24 04:58:33

by Len Brown

Subject: [PATCH v3] tools: create power/x86/turbostat

turbostat is a tool to help verify proper operation of CPU idle power saving
states, processor performance states, and turbo-mode on modern x86 processors.

Signed-off-by: Len Brown <[email protected]>
---
v3:
Move source comments to a man page
Allow operation on AMD
Add support for new counters in Sandy Bridge

tools/power/x86/turbostat/Makefile | 8 +
tools/power/x86/turbostat/turbostat.8 | 172 ++++++
tools/power/x86/turbostat/turbostat.c | 1047 +++++++++++++++++++++++++++++++++
3 files changed, 1227 insertions(+), 0 deletions(-)

diff --git a/tools/power/x86/turbostat/Makefile b/tools/power/x86/turbostat/Makefile
new file mode 100644
index 0000000..fd8e1f1
--- /dev/null
+++ b/tools/power/x86/turbostat/Makefile
@@ -0,0 +1,8 @@
+turbostat : turbostat.c
+
+clean :
+ rm -f turbostat
+
+install :
+ install turbostat /usr/bin/turbostat
+ install turbostat.8 /usr/share/man/man8
diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8
new file mode 100644
index 0000000..ff75125
--- /dev/null
+++ b/tools/power/x86/turbostat/turbostat.8
@@ -0,0 +1,172 @@
+.TH TURBOSTAT 8
+.SH NAME
+turbostat \- Report processor frequency and idle statistics
+.SH SYNOPSIS
+.ft B
+.B turbostat
+.RB [ "\-v" ]
+.RB [ "\-M MSR#" ]
+.RB command
+.br
+.B turbostat
+.RB [ "\-v" ]
+.RB [ "\-M MSR#" ]
+.RB [ "\-i interval_sec" ]
+.SH DESCRIPTION
+\fBturbostat \fP reports processor topology, frequency
+and idle power state statistics on modern X86 processors.
+Either \fBcommand\fP is forked and statistics are printed
+upon its completion, or statistics are printed periodically.
+
+\fBturbostat \fP
+requires that the processor
+supports an "invariant" TSC, plus the APERF and MPERF MSRs.
+\fBturbostat \fP will report idle cpu power state residency
+on processors that additionally support C-state residency counters.
+
+.SS Options
+The \fB-v\fP option increases verbosity.
+.PP
+The \fB-M MSR#\fP option dumps the specified MSR,
+in addition to the usual frequency and idle statistics.
+.PP
+The \fB-i interval_sec\fP option prints statistics every \fiinterval_sec\fP seconds.
+The default is 5 seconds.
+.PP
+The \fBcommand\fP parameter forks \fBcommand\fP and upon its exit,
+displays the statistics gathered since it was forked.
+.PP
+.SH FIELD DESCRIPTIONS
+.nf
+\fBpkg\fP processor package number.
+\fBcore\fP processor core number.
+\fBCPU\fP Linux CPU (logical processor) number.
+\fB%c0\fP percent of the interval that the CPU retired instructions.
+\fBGHz\fP average clock rate while the CPU was in c0 state.
+\fBTSC\fP average GHz that the TSC ran during the entire interval.
+\fB%c1, %c3, %c6\fP show the percentage residency in hardware core idle states.
+\fB%pc3, %pc6\fP percentage residency in hardware package idle states.
+.fi
+.PP
+.SH EXAMPLE
+Without any parameters, turbostat prints out counters ever 5 seconds.
+(override interval with "-i sec" option, or specify a command
+for turbostat to fork).
+
+The first row of statistics reflect the average for the entire system.
+Subsequent rows show per-CPU statistics.
+
+.nf
+[root@x980]# ./turbostat
+core CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
+ 0.04 1.62 3.38 0.11 0.00 99.85 0.00 95.07
+ 0 0 0.04 1.62 3.38 0.06 0.00 99.90 0.00 95.07
+ 0 6 0.02 1.62 3.38 0.08 0.00 99.90 0.00 95.07
+ 1 2 0.10 1.62 3.38 0.29 0.00 99.61 0.00 95.07
+ 1 8 0.11 1.62 3.38 0.28 0.00 99.61 0.00 95.07
+ 2 4 0.01 1.62 3.38 0.01 0.00 99.98 0.00 95.07
+ 2 10 0.01 1.61 3.38 0.02 0.00 99.98 0.00 95.07
+ 8 1 0.07 1.62 3.38 0.15 0.00 99.78 0.00 95.07
+ 8 7 0.03 1.62 3.38 0.19 0.00 99.78 0.00 95.07
+ 9 3 0.01 1.62 3.38 0.02 0.00 99.98 0.00 95.07
+ 9 9 0.01 1.62 3.38 0.02 0.00 99.98 0.00 95.07
+ 10 5 0.01 1.62 3.38 0.13 0.00 99.86 0.00 95.07
+ 10 11 0.08 1.62 3.38 0.05 0.00 99.86 0.00 95.07
+.fi
+.SH VERBOSE EXAMPLE
+The "-v" option adds verbosity to the output:
+
+.nf
+GenuineIntel 11 CPUID levels; family:model:stepping 0x6:2c:2 (6:44:2)
+12 * 133 = 1600 MHz max efficiency
+25 * 133 = 3333 MHz TSC frequency
+26 * 133 = 3467 MHz max turbo 4 active cores
+26 * 133 = 3467 MHz max turbo 3 active cores
+27 * 133 = 3600 MHz max turbo 2 active cores
+27 * 133 = 3600 MHz max turbo 1 active cores
+
+.fi
+The \fBmax efficiency\fP frequency, a.k.a. Low Frequency Mode, is the frequency
+available at the minimum package voltage. The \fBTSC frequency\fP is the nominal
+maximum frequency of the processor if turbo-mode were not available. This frequency
+should be sustainable on all CPUs indefinitely, given nominal power and cooling.
+The remaining rows show what maximum turbo frequency is possible
+depending on the number of idle cores. Note that this information is
+not available on all processors.
+.SH FORK EXAMPLE
+If turbostat is invoked with a command, it will fork that command
+and output the statistics gathered when the command exits.
+eg. Here a cycle soaker is run on 1 CPU (see %c0) for a few seconds
+until ^C while the other CPUs are mostly idle:
+
+.nf
+[root@x980 lenb]# ./turbostat cat /dev/zero > /dev/null
+
+^Ccore CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
+ 8.49 3.63 3.38 16.23 0.66 74.63 0.00 0.00
+ 0 0 1.22 3.62 3.38 32.18 0.00 66.60 0.00 0.00
+ 0 6 0.40 3.61 3.38 33.00 0.00 66.60 0.00 0.00
+ 1 2 0.11 3.14 3.38 0.19 3.95 95.75 0.00 0.00
+ 1 8 0.05 2.88 3.38 0.25 3.95 95.75 0.00 0.00
+ 2 4 0.00 3.13 3.38 0.02 0.00 99.98 0.00 0.00
+ 2 10 0.00 3.09 3.38 0.02 0.00 99.98 0.00 0.00
+ 8 1 0.04 3.50 3.38 14.43 0.00 85.54 0.00 0.00
+ 8 7 0.03 2.98 3.38 14.43 0.00 85.54 0.00 0.00
+ 9 3 0.00 3.16 3.38 100.00 0.00 0.00 0.00 0.00
+ 9 9 99.93 3.63 3.38 0.06 0.00 0.00 0.00 0.00
+ 10 5 0.01 2.82 3.38 0.08 0.00 99.91 0.00 0.00
+ 10 11 0.02 3.36 3.38 0.06 0.00 99.91 0.00 0.00
+6.950866 sec
+
+.fi
+Above the cycle soaker drives cpu9 up 3.6 Ghz turbo limit
+while the other processors are generally in various states of idle.
+
+Note that cpu3 is an HT sibling sharing core9
+with cpu9, and thus it is unable to get to an idle state
+deeper than c1 while cpu9 is busy.
+
+Note that turbostat reports average GHz of 3.61, while
+the arithmetic average of the GHz column above is 3.24.
+This is a weighted average, where the weight is %c0. ie. it is the total number of
+un-halted cycles elapsed per time divided by the number of CPUs.
+.SH NOTES
+
+.B "turbostat "
+must be run as root.
+
+.B "turbostat "
+reads hardware counters, but doesn't write them.
+So it will not interfere with the OS or other programs, including
+multiple invocations of itself.
+
+\fBturbostat \fP
+may work poorly on Linux-2.6.20 through 2.6.29,
+as \fBacpi-cpufreq \fPperiodically cleared the APERF and MPERF
+in those kernels.
+
+The APERF, MPERF MSRs are defined to count non-halted cycles.
+Although it is not guaranteed by the architecture, turbostat assumes
+that they count at TSC rate, which is true on all processors tested to date.
+
+.SH REFERENCES
+"Intel® Turbo Boost Technology
+in Intel® Core™ Microarchitecture (Nehalem) Based Processors"
+http://download.intel.com/design/processor/applnots/320354.pdf
+
+"Intel® 64 and IA-32 Architectures Software Developer's Manual
+Volume 3B: System Programming Guide"
+http://www.intel.com/products/processor/manuals/
+
+.SH FILES
+.ta
+.nf
+/dev/cpu/*/msr
+.fi
+
+.SH "SEE ALSO"
+msr(4), vmstat(8)
+.PP
+.SH AUTHORS
+.nf
+Written by Len Brown <[email protected]>
diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
new file mode 100644
index 0000000..b8001b1
--- /dev/null
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -0,0 +1,1047 @@
+/*
+ * turbostat -- show CPU frequency and C-state residency
+ * on modern Intel turbo-capable processors.
+ *
+ * Copyright (c) 2010, Intel Corporation.
+ * Len Brown <[email protected]>
+ *
+ * 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/wait.h>
+#include <sys/stat.h>
+#include <sys/resource.h>
+#include <fcntl.h>
+#include <signal.h>
+#include <sys/time.h>
+#include <stdlib.h>
+#include <dirent.h>
+#include <string.h>
+#include <ctype.h>
+
+#define MSR_TSC 0x10
+#define MSR_NEHALEM_PLATFORM_INFO 0xCE
+#define MSR_NEHALEM_TURBO_RATIO_LIMIT 0x1AD
+#define MSR_APERF 0xE8
+#define MSR_MPERF 0xE7
+#define MSR_PKG_C2_RESIDENCY 0x60D /* SNB only */
+#define MSR_PKG_C3_RESIDENCY 0x3F8
+#define MSR_PKG_C6_RESIDENCY 0x3F9
+#define MSR_PKG_C7_RESIDENCY 0x3FA /* SNB only */
+#define MSR_CORE_C3_RESIDENCY 0x3FC
+#define MSR_CORE_C6_RESIDENCY 0x3FD
+#define MSR_CORE_C7_RESIDENCY 0x3FE /* SNB only */
+
+char *proc_stat = "/proc/stat";
+unsigned int interval_sec = 5; /* set with -i interval_sec */
+unsigned int verbose; /* set with -v */
+unsigned int skip_c0;
+unsigned int skip_c1;
+unsigned int do_nhm_cstates;
+unsigned int do_snb_cstates;
+unsigned int has_aperf;
+unsigned int units = 1000000000; /* Ghz etc */
+unsigned int genuine_intel;
+unsigned int has_invariant_tsc;
+unsigned int do_nehalem_platform_info;
+unsigned int do_nehalem_turbo_ratio_limit;
+unsigned int extra_msr_offset;
+double bclk;
+unsigned int show_pkg;
+unsigned int show_core;
+unsigned int show_cpu;
+
+int aperf_mperf_unstable;
+int backwards_count;
+char *progname;
+int need_reinitialize;
+
+int num_cpus;
+
+typedef struct per_cpu_counters {
+ unsigned long long tsc; /* per thread */
+ unsigned long long aperf; /* per thread */
+ unsigned long long mperf; /* per thread */
+ unsigned long long c1; /* per thread (calculated) */
+ unsigned long long c3; /* per core */
+ unsigned long long c6; /* per core */
+ unsigned long long c7; /* per core */
+ unsigned long long pc2; /* per package */
+ unsigned long long pc3; /* per package */
+ unsigned long long pc6; /* per package */
+ unsigned long long pc7; /* per package */
+ unsigned long long extra_msr; /* per thread */
+ int pkg;
+ int core;
+ int cpu;
+ struct per_cpu_counters *next;
+} PCC;
+
+PCC *pcc_even;
+PCC *pcc_odd;
+PCC *pcc_delta;
+PCC *pcc_average;
+struct timeval tv_even;
+struct timeval tv_odd;
+struct timeval tv_delta;
+
+unsigned long long get_msr(int cpu, off_t offset)
+{
+ ssize_t retval;
+ unsigned long long msr;
+ char pathname[32];
+ int fd;
+
+ sprintf(pathname, "/dev/cpu/%d/msr", cpu);
+ fd = open(pathname, O_RDONLY);
+ if (fd < 0) {
+ perror(pathname);
+ need_reinitialize = 1;
+ return 0;
+ }
+
+ retval = pread(fd, &msr, sizeof msr, offset);
+ if (retval != sizeof msr) {
+ fprintf(stderr, "cpu%d pread(..., 0x%zx) = %jd\n",
+ cpu, offset, retval);
+ exit(-2);
+ }
+
+ close(fd);
+ return msr;
+}
+
+void print_header()
+{
+ if (show_pkg)
+ fprintf(stderr, "pkg ");
+ if (show_core)
+ fprintf(stderr, "core");
+ if (show_cpu)
+ fprintf(stderr, " CPU");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c0 ");
+ if (has_aperf)
+ fprintf(stderr, " GHz");
+ fprintf(stderr, " TSC");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c1 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c3 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c6 ");
+ if (do_snb_cstates)
+ fprintf(stderr, " %%c7 ");
+ if (do_snb_cstates)
+ fprintf(stderr, " %%pc2 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%pc3 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%pc6 ");
+ if (do_snb_cstates)
+ fprintf(stderr, " %%pc7 ");
+ if (extra_msr_offset)
+ fprintf(stderr, " MSR 0x%x ", extra_msr_offset);
+
+ putc('\n', stderr);
+}
+
+void dump_pcc(PCC *pcc)
+{
+ fprintf(stderr, "package: %d ", pcc->pkg);
+ fprintf(stderr, "core:: %d ", pcc->core);
+ fprintf(stderr, "CPU: %d ", pcc->cpu);
+ fprintf(stderr, "TSC: %016llX\n", pcc->tsc);
+ fprintf(stderr, "c3: %016llX\n", pcc->c3);
+ fprintf(stderr, "c6: %016llX\n", pcc->c6);
+ fprintf(stderr, "c7: %016llX\n", pcc->c7);
+ fprintf(stderr, "aperf: %016llX\n", pcc->aperf);
+ fprintf(stderr, "pc2: %016llX\n", pcc->pc2);
+ fprintf(stderr, "pc3: %016llX\n", pcc->pc3);
+ fprintf(stderr, "pc6: %016llX\n", pcc->pc6);
+ fprintf(stderr, "pc7: %016llX\n", pcc->pc7);
+ fprintf(stderr, "msr0x%x: %016llX\n", extra_msr_offset, pcc->extra_msr);
+}
+
+void dump_list(PCC *pcc)
+{
+ printf("dump_list 0x%p\n", pcc);
+
+ for (; pcc; pcc = pcc->next)
+ dump_pcc(pcc);
+}
+
+void print_pcc(PCC *p)
+{
+ double interval_float;
+
+ interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
+
+ /* topology columns, print blanks on 1st (average) line */
+ if (p == pcc_average) {
+ if (show_pkg)
+ fprintf(stderr, " ");
+ if (show_core)
+ fprintf(stderr, " ");
+ if (show_cpu)
+ fprintf(stderr, " ");
+ } else {
+ if (show_pkg)
+ fprintf(stderr, "%4d", p->pkg);
+ if (show_core)
+ fprintf(stderr, "%4d", p->core);
+ if (show_cpu)
+ fprintf(stderr, "%4d", p->cpu);
+ }
+
+ /* %c0 */
+ if (do_nhm_cstates) {
+ if (!skip_c0)
+ fprintf(stderr, "%7.2f", 100.0 * p->mperf/p->tsc);
+ else
+ fprintf(stderr, " ****");
+ }
+
+ /* GHz */
+ if (has_aperf) {
+ if (!aperf_mperf_unstable) {
+ fprintf(stderr, "%5.2f",
+ 1.0 * p->tsc / units * p->aperf /
+ p->mperf / interval_float);
+ } else {
+ if (p->aperf > p->tsc || p->mperf > p->tsc) {
+ fprintf(stderr, " ****");
+ } else {
+ fprintf(stderr, "%4.1f*",
+ 1.0 * p->tsc /
+ units * p->aperf /
+ p->mperf / interval_float);
+ }
+ }
+ }
+
+ /* TSC */
+ fprintf(stderr, "%5.2f", 1.0 * p->tsc/units/interval_float);
+
+ if (do_nhm_cstates) {
+ if (!skip_c1)
+ fprintf(stderr, "%7.2f", 100.0 * p->c1/p->tsc);
+ else
+ fprintf(stderr, " ****");
+ }
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c3/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c6/p->tsc);
+ if (do_snb_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c7/p->tsc);
+ if (do_snb_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc2/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc3/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc6/p->tsc);
+ if (do_snb_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc7/p->tsc);
+ if (extra_msr_offset)
+ fprintf(stderr, " 0x%016llx", p->extra_msr);
+ putc('\n', stderr);
+}
+
+void print_counters(PCC *cnt)
+{
+ PCC *pcc;
+
+ print_header();
+
+ if (num_cpus > 1)
+ print_pcc(pcc_average);
+
+ for (pcc = cnt; pcc != NULL; pcc = pcc->next)
+ print_pcc(pcc);
+
+}
+
+#define SUBTRACT_COUNTER(after, before, delta) (delta = (after - before), (before > after))
+
+
+int compute_delta(PCC *after, PCC *before, PCC *delta)
+{
+ int errors = 0;
+ int perf_err = 0;
+
+ skip_c0 = skip_c1 = 0;
+
+ for ( ; after && before && delta;
+ after = after->next, before = before->next, delta = delta->next) {
+ if (before->cpu != after->cpu) {
+ printf("cpu configuration changed: %d != %d\n",
+ before->cpu, after->cpu);
+ return -1;
+ }
+
+ if (SUBTRACT_COUNTER(after->tsc, before->tsc, delta->tsc)) {
+ fprintf(stderr, "cpu%d TSC went backwards %llX to %llX\n",
+ before->cpu, before->tsc, after->tsc);
+ errors++;
+ }
+ /* check for TSC < 1 Mcycles over interval */
+ if (delta->tsc < (1000 * 1000)) {
+ fprintf(stderr, "Insanely slow TSC rate,"
+ " TSC stops in idle?\n");
+ fprintf(stderr, "You can disable all c-states"
+ " by booting with \"idle=poll\"\n");
+ fprintf(stderr, "or just the deep ones with"
+ " \"processor.max_cstate=1\"\n");
+ exit(-3);
+ }
+ if (SUBTRACT_COUNTER(after->c3, before->c3, delta->c3)) {
+ fprintf(stderr, "cpu%d c3 counter went backwards %llX to %llX\n",
+ before->cpu, before->c3, after->c3);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->c6, before->c6, delta->c6)) {
+ fprintf(stderr, "cpu%d c6 counter went backwards %llX to %llX\n",
+ before->cpu, before->c6, after->c6);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->c7, before->c7, delta->c7)) {
+ fprintf(stderr, "cpu%d c7 counter went backwards %llX to %llX\n",
+ before->cpu, before->c7, after->c7);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc2, before->pc2, delta->pc2)) {
+ fprintf(stderr, "cpu%d pc2 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc2, after->pc2);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc3, before->pc3, delta->pc3)) {
+ fprintf(stderr, "cpu%d pc3 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc3, after->pc3);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc6, before->pc6, delta->pc6)) {
+ fprintf(stderr, "cpu%d pc6 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc6, after->pc6);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc7, before->pc7, delta->pc7)) {
+ fprintf(stderr, "cpu%d pc7 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc7, after->pc7);
+ errors++;
+ }
+
+ perf_err = SUBTRACT_COUNTER(after->aperf, before->aperf, delta->aperf);
+ if (perf_err) {
+ fprintf(stderr, "cpu%d aperf counter went backwards %llX to %llX\n",
+ before->cpu, before->aperf, after->aperf);
+ }
+ perf_err |= SUBTRACT_COUNTER(after->mperf, before->mperf, delta->mperf);
+ if (perf_err) {
+ fprintf(stderr, "cpu%d mperf counter went backwards %llX to %llX\n",
+ before->cpu, before->mperf, after->mperf);
+ }
+ if (perf_err) {
+ if (!aperf_mperf_unstable) {
+ fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
+ fprintf(stderr, "* Frequency results do not cover entire interval *\n");
+ fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
+
+ aperf_mperf_unstable = 1;
+ }
+ /*
+ * mperf delta is likely a huge "positive" number
+ * can not use it for calculating c0 time
+ */
+ skip_c0 = 1;
+ skip_c1 = 1;
+ }
+
+ /*
+ * As mperf and tsc collection are not atomic,
+ * it is possible for mperf's non-halted cycles
+ * to exceed TSC's all cycles: show c1 = 0% in that case.
+ */
+ if (delta->mperf > delta->tsc)
+ delta->c1 = 0;
+ else /* normal case, derive c1 */
+ delta->c1 = delta->tsc - delta->mperf
+ - delta->c3 - delta->c6 - delta->c7;
+
+ if (delta->mperf == 0)
+ delta->mperf = 1; /* divide by 0 protection */
+
+ /*
+ * for "extra msr", just copy the latest w/o subtracting
+ */
+ delta->extra_msr = after->extra_msr;
+ if (errors) {
+ fprintf(stderr, "ERROR cpu%d before:\n", before->cpu);
+ dump_pcc(before);
+ fprintf(stderr, "ERROR cpu%d after:\n", before->cpu);
+ dump_pcc(after);
+ errors = 0;
+ }
+ }
+ return 0;
+}
+
+void compute_average(PCC *delta, PCC *avg)
+{
+ PCC *sum;
+
+ sum = calloc(1, sizeof(PCC));
+ if (sum == NULL) {
+ perror("calloc sum");
+ exit(1);
+ }
+
+ for (; delta; delta = delta->next) {
+ sum->tsc += delta->tsc;
+ sum->c1 += delta->c1;
+ sum->c3 += delta->c3;
+ sum->c6 += delta->c6;
+ sum->c7 += delta->c7;
+ sum->aperf += delta->aperf;
+ sum->mperf += delta->mperf;
+ sum->pc2 += delta->pc2;
+ sum->pc3 += delta->pc3;
+ sum->pc6 += delta->pc6;
+ sum->pc7 += delta->pc7;
+ }
+ avg->tsc = sum->tsc/num_cpus;
+ avg->c1 = sum->c1/num_cpus;
+ avg->c3 = sum->c3/num_cpus;
+ avg->c6 = sum->c6/num_cpus;
+ avg->c7 = sum->c7/num_cpus;
+ avg->aperf = sum->aperf/num_cpus;
+ avg->mperf = sum->mperf/num_cpus;
+ avg->pc2 = sum->pc2/num_cpus;
+ avg->pc3 = sum->pc3/num_cpus;
+ avg->pc6 = sum->pc6/num_cpus;
+ avg->pc7 = sum->pc7/num_cpus;
+
+ free(sum);
+}
+
+void get_counters(PCC *pcc)
+{
+ for ( ; pcc; pcc = pcc->next) {
+ pcc->tsc = get_msr(pcc->cpu, MSR_TSC);
+ if (do_nhm_cstates)
+ pcc->c3 = get_msr(pcc->cpu, MSR_CORE_C3_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->c6 = get_msr(pcc->cpu, MSR_CORE_C6_RESIDENCY);
+ if (do_snb_cstates)
+ pcc->c7 = get_msr(pcc->cpu, MSR_CORE_C7_RESIDENCY);
+ if (has_aperf)
+ pcc->aperf = get_msr(pcc->cpu, MSR_APERF);
+ if (has_aperf)
+ pcc->mperf = get_msr(pcc->cpu, MSR_MPERF);
+ if (do_snb_cstates)
+ pcc->pc2 = get_msr(pcc->cpu, MSR_PKG_C2_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->pc3 = get_msr(pcc->cpu, MSR_PKG_C3_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->pc6 = get_msr(pcc->cpu, MSR_PKG_C6_RESIDENCY);
+ if (do_snb_cstates)
+ pcc->pc7 = get_msr(pcc->cpu, MSR_PKG_C7_RESIDENCY);
+ if (extra_msr_offset)
+ pcc->extra_msr = get_msr(pcc->cpu, extra_msr_offset);
+ }
+}
+
+
+void print_nehalem_info()
+{
+ unsigned long long msr;
+ unsigned int ratio;
+
+ if (!do_nehalem_platform_info)
+ return;
+
+ msr = get_msr(0, MSR_NEHALEM_PLATFORM_INFO);
+
+ ratio = (msr >> 40) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 8) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz TSC frequency\n",
+ ratio, bclk, ratio * bclk);
+
+ if (verbose > 1)
+ fprintf(stderr, "MSR_NEHALEM_PLATFORM_INFO: 0x%llx\n", msr);
+
+ if (!do_nehalem_turbo_ratio_limit)
+ return;
+
+ msr = get_msr(0, MSR_NEHALEM_TURBO_RATIO_LIMIT);
+
+ ratio = (msr >> 24) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 16) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 8) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 0) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+}
+
+void free_counter_list(PCC *list)
+{
+ PCC *p;
+
+ for (p = list; p; ) {
+ PCC *free_me;
+
+ free_me = p;
+ p = p->next;
+ free(free_me);
+ }
+ return;
+}
+
+void free_all_counters(void)
+{
+ free_counter_list(pcc_even);
+ pcc_even = NULL;
+
+ free_counter_list(pcc_odd);
+ pcc_odd = NULL;
+
+ free_counter_list(pcc_delta);
+ pcc_delta = NULL;
+
+ free_counter_list(pcc_average);
+ pcc_average = NULL;
+}
+
+void insert_cpu_counters(PCC **list, PCC *new)
+{
+ PCC *prev;
+
+ /*
+ * list was empty
+ */
+ if (*list == NULL) {
+ new->next = *list;
+ *list = new;
+ return;
+ }
+
+ show_cpu = 1; /* there is more than one CPU */
+
+ /*
+ * insert on front of list.
+ * It is sorted by ascending package#, core#, cpu#
+ */
+ if (((*list)->pkg > new->pkg) ||
+ (((*list)->pkg == new->pkg) && ((*list)->core > new->core)) ||
+ (((*list)->pkg == new->pkg) && ((*list)->core == new->core) && ((*list)->cpu > new->cpu))) {
+ new->next = *list;
+ *list = new;
+ return;
+ }
+
+ prev = *list;
+
+ while (prev->next && (prev->next->pkg < new->pkg)) {
+ prev = prev->next;
+ show_pkg = 1; /* there is more than 1 package */
+ }
+
+ while (prev->next && (prev->next->pkg == new->pkg)
+ && (prev->next->core < new->core)) {
+ prev = prev->next;
+ show_core = 1; /* there is more than 1 core */
+ }
+
+ while (prev->next && (prev->next->pkg == new->pkg)
+ && (prev->next->core == new->core)
+ && (prev->next->cpu < new->cpu)) {
+ prev = prev->next;
+ }
+
+ /*
+ * insert after "prev"
+ */
+ new->next = prev->next;
+ prev->next = new;
+
+ return;
+}
+
+void alloc_new_cpu_counters(int pkg, int core, int cpu)
+{
+ PCC *new;
+
+ if (verbose > 1)
+ printf("pkg%d core%d, cpu%d\n", pkg, core, cpu);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_odd, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_even, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_delta, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ pcc_average = new;
+}
+
+int get_physical_package_id(int cpu)
+{
+ char path[64];
+ FILE *filep;
+ int pkg;
+
+ sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
+ filep = fopen(path, "r");
+ if (filep == NULL) {
+ perror(path);
+ exit(1);
+ }
+ fscanf(filep, "%d", &pkg);
+ fclose(filep);
+ return pkg;
+}
+
+int get_core_id(int cpu)
+{
+ char path[64];
+ FILE *filep;
+ int core;
+
+ sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
+ filep = fopen(path, "r");
+ if (filep == NULL) {
+ perror(path);
+ exit(1);
+ }
+ fscanf(filep, "%d", &core);
+ fclose(filep);
+ return core;
+}
+
+/*
+ * run func(index, cpu) on every cpu in /proc/stat
+ */
+
+int for_all_cpus(void (func)(int, int, int))
+{
+ FILE *fp;
+ int cpu_count;
+ 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);
+ }
+
+ for (cpu_count = 0; ; cpu_count++) {
+ int cpu;
+
+ retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu);
+ if (retval != 1)
+ return cpu_count;
+
+ func(get_physical_package_id(cpu), get_core_id(cpu), cpu);
+ }
+ fclose(fp);
+}
+
+void re_initialize(void)
+{
+ printf("turbostat: topology changed, re-initializing.\n");
+ free_all_counters();
+ num_cpus = for_all_cpus(alloc_new_cpu_counters);
+ need_reinitialize = 0;
+ printf("num_cpus is now %d\n", num_cpus);
+}
+
+void dummy(int pkg, int core, int cpu) { return; }
+/*
+ * check to see if a cpu came on-line
+ */
+void verify_num_cpus()
+{
+ int new_num_cpus;
+
+ new_num_cpus = for_all_cpus(dummy);
+
+ if (new_num_cpus != num_cpus) {
+ if (verbose)
+ printf("num_cpus was %d, is now %d\n",
+ num_cpus, new_num_cpus);
+ need_reinitialize = 1;
+ }
+
+ return;
+}
+
+void turbostat_loop()
+{
+restart:
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+
+ while (1) {
+ verify_num_cpus();
+ if (need_reinitialize) {
+ re_initialize();
+ goto restart;
+ }
+ sleep(interval_sec);
+ get_counters(pcc_odd);
+ gettimeofday(&tv_odd, (struct timezone *)NULL);
+
+ compute_delta(pcc_odd, pcc_even, pcc_delta);
+ timersub(&tv_odd, &tv_even, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ print_counters(pcc_delta);
+ if (need_reinitialize) {
+ re_initialize();
+ goto restart;
+ }
+ sleep(interval_sec);
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+ compute_delta(pcc_even, pcc_odd, pcc_delta);
+ timersub(&tv_even, &tv_odd, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ print_counters(pcc_delta);
+ }
+}
+
+void check_dev_msr()
+{
+ struct stat sb;
+
+ if (stat("/dev/cpu/0/msr", &sb)) {
+ fprintf(stderr, "no /dev/cpu/0/msr\n");
+ fprintf(stderr, "Try \"# modprobe msr\"\n");
+ exit(-5);
+ }
+}
+
+void check_super_user()
+{
+ if (getuid() != 0) {
+ fprintf(stderr, "must be root\n");
+ exit(-6);
+ }
+}
+
+int has_nehalem_turbo_ratio_limit(unsigned int family, unsigned int model)
+{
+ if (!genuine_intel)
+ return 0;
+
+ if (family != 6)
+ return 0;
+
+ switch (model) {
+ case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
+ case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
+ case 0x1F: /* Core i7 and i5 Processor - Nehalem */
+ case 0x25: /* Westmere Client - Clarkdale, Arrandale */
+ case 0x2C: /* Westmere EP - Gulftown */
+ case 0x2A: /* SNB */
+ case 0x2D: /* SNB Xeon */
+ return 1;
+ case 0x2E: /* Nehalem-EX Xeon - Beckton */
+ case 0x2F: /* Westmere-EX Xeon - Eagleton */
+ default:
+ return 0;
+ }
+}
+
+int is_snb(unsigned int family, unsigned int model)
+{
+ if (!genuine_intel)
+ return 0;
+
+ switch (model) {
+ case 0x2A:
+ case 0x2D:
+ return 1;
+ }
+ return 0;
+}
+
+double discover_bclk(unsigned int family, unsigned int model)
+{
+ if (is_snb(family, model))
+ return 100.00;
+ else
+ return 133.33;
+}
+
+void check_cpuid()
+{
+ unsigned int eax, ebx, ecx, edx, max_level;
+ 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)
+ genuine_intel = 1;
+
+ if (verbose)
+ fprintf(stderr, "%.4s%.4s%.4s ",
+ (char *)&ebx, (char *)&edx, (char *)&ecx);
+
+ 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)
+ fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
+ max_level, family, model, stepping, family, model, stepping);
+
+ if (!(edx & (1 << 5))) {
+ fprintf(stderr, "CPUID: no MSR\n");
+ exit(1);
+ }
+
+ /*
+ * check max extended function levels of CPUID.
+ * This is needed to check for invariant TSC.
+ * This check is valid for both Intel and AMD.
+ */
+ ebx = ecx = edx = 0;
+ asm("cpuid" : "=a" (max_level), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x80000000));
+
+ if (max_level < 0x80000007) {
+ fprintf(stderr, "CPUID: no invariant TSC (max_level 0x%x)\n", max_level);
+ exit(1);
+ }
+
+ /*
+ * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
+ * this check is valid for both Intel and AMD
+ */
+ asm("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x80000007));
+ has_invariant_tsc = edx && (1 << 8);
+
+ if (!has_invariant_tsc) {
+ fprintf(stderr, "No invariant TSC\n");
+ exit(1);
+ }
+
+ /*
+ * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
+ * this check is valid for both Intel and AMD
+ */
+
+ asm("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x6));
+ has_aperf = ecx && (1 << 0);
+ if (!has_aperf) {
+ fprintf(stderr, "No APERF MSR\n");
+ exit(1);
+ }
+
+ do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
+ do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
+ do_snb_cstates = is_snb(family, model);
+ bclk = discover_bclk(family, model);
+
+ do_nehalem_turbo_ratio_limit = has_nehalem_turbo_ratio_limit(family, model);
+}
+
+
+void usage()
+{
+ fprintf(stderr, "%s: [-v] [-M MSR#] [-i interval_sec | command ...]\n",
+ progname);
+ exit(1);
+}
+
+
+/*
+ * in /dev/cpu/ return success for names that are numbers
+ * ie. filter out ".", "..", "microcode".
+ */
+int dir_filter(const struct dirent *dirp)
+{
+ if (isdigit(dirp->d_name[0]))
+ return 1;
+ else
+ return 0;
+}
+
+int open_dev_cpu_msr(int dummy1)
+{
+ return 0;
+}
+
+void turbostat_init()
+{
+ check_cpuid();
+
+ check_dev_msr();
+ check_super_user();
+
+ num_cpus = for_all_cpus(alloc_new_cpu_counters);
+
+ if (verbose)
+ print_nehalem_info();
+}
+
+int fork_it(char **argv)
+{
+ int retval;
+ pid_t child_pid;
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+
+ child_pid = fork();
+ if (!child_pid) {
+ /* child */
+ execvp(argv[0], argv);
+ } else {
+ int status;
+
+ /* parent */
+ if (child_pid == -1) {
+ perror("fork");
+ exit(1);
+ }
+
+ signal(SIGINT, SIG_IGN);
+ signal(SIGQUIT, SIG_IGN);
+ if (waitpid(child_pid, &status, 0) == -1) {
+ perror("wait");
+ exit(1);
+ }
+ }
+ get_counters(pcc_odd);
+ gettimeofday(&tv_odd, (struct timezone *)NULL);
+ retval = compute_delta(pcc_odd, pcc_even, pcc_delta);
+
+ timersub(&tv_odd, &tv_even, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ if (!retval)
+ print_counters(pcc_delta);
+
+ fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);;
+
+ return 0;
+}
+
+void cmdline(int argc, char **argv)
+{
+ int opt;
+
+ progname = argv[0];
+
+ while ((opt = getopt(argc, argv, "+vi:M:")) != -1) {
+ switch (opt) {
+ case 'v':
+ verbose++;
+ break;
+ case 'i':
+ interval_sec = atoi(optarg);
+ break;
+ case 'M':
+ sscanf(optarg, "%x", &extra_msr_offset);
+ if (verbose > 1)
+ fprintf(stderr, "MSR 0x%X\n", extra_msr_offset);
+ break;
+ default:
+ usage();
+ }
+ }
+}
+
+int main(int argc, char **argv)
+{
+ cmdline(argc, argv);
+
+ if (verbose > 1)
+ fprintf(stderr, "turbostat Nov 16, 2010"
+ " - Len Brown <[email protected]>\n");
+ if (verbose > 1)
+ fprintf(stderr, "http://userweb.kernel.org/~lenb/acpi/utils/pmtools/turbostat/\n");
+
+ turbostat_init();
+
+ /*
+ * if any params left, it must be a command to fork
+ */
+ if (argc - optind)
+ return fork_it(argv + optind);
+ else
+ turbostat_loop();
+
+ return 0;
+}

2010-12-07 06:01:09

by Len Brown

Subject: [PATCH v4] tools: create power/x86/turbostat

From: Len Brown <[email protected]>

turbostat is a tool to help verify proper operation
of processor power management states on modern x86.

Signed-off-by: Len Brown <[email protected]>
---
v4: fix open file descriptor leak, reported by David C. Wong

tools/power/x86/turbostat/Makefile | 8 +
tools/power/x86/turbostat/turbostat.8 | 172 ++++++
tools/power/x86/turbostat/turbostat.c | 1048 +++++++++++++++++++++++++++++++++
3 files changed, 1228 insertions(+), 0 deletions(-)

diff --git a/tools/power/x86/turbostat/Makefile b/tools/power/x86/turbostat/Makefile
new file mode 100644
index 0000000..fd8e1f1
--- /dev/null
+++ b/tools/power/x86/turbostat/Makefile
@@ -0,0 +1,8 @@
+turbostat : turbostat.c
+
+clean :
+ rm -f turbostat
+
+install :
+ install turbostat /usr/bin/turbostat
+ install turbostat.8 /usr/share/man/man8
diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8
new file mode 100644
index 0000000..ff75125
--- /dev/null
+++ b/tools/power/x86/turbostat/turbostat.8
@@ -0,0 +1,172 @@
+.TH TURBOSTAT 8
+.SH NAME
+turbostat \- Report processor frequency and idle statistics
+.SH SYNOPSIS
+.ft B
+.B turbostat
+.RB [ "\-v" ]
+.RB [ "\-M MSR#" ]
+.RB command
+.br
+.B turbostat
+.RB [ "\-v" ]
+.RB [ "\-M MSR#" ]
+.RB [ "\-i interval_sec" ]
+.SH DESCRIPTION
+\fBturbostat \fP reports processor topology, frequency
+and idle power state statistics on modern X86 processors.
+Either \fBcommand\fP is forked and statistics are printed
+upon its completion, or statistics are printed periodically.
+
+\fBturbostat \fP
+requires that the processor
+supports an "invariant" TSC, plus the APERF and MPERF MSRs.
+\fBturbostat \fP will report idle cpu power state residency
+on processors that additionally support C-state residency counters.
+
+.SS Options
+The \fB-v\fP option increases verbosity.
+.PP
+The \fB-M MSR#\fP option dumps the specified MSR,
+in addition to the usual frequency and idle statistics.
+.PP
+The \fB-i interval_sec\fP option prints statistics every \fiinterval_sec\fP seconds.
+The default is 5 seconds.
+.PP
+The \fBcommand\fP parameter forks \fBcommand\fP and upon its exit,
+displays the statistics gathered since it was forked.
+.PP
+.SH FIELD DESCRIPTIONS
+.nf
+\fBpkg\fP processor package number.
+\fBcore\fP processor core number.
+\fBCPU\fP Linux CPU (logical processor) number.
+\fB%c0\fP percent of the interval that the CPU retired instructions.
+\fBGHz\fP average clock rate while the CPU was in c0 state.
+\fBTSC\fP average GHz that the TSC ran during the entire interval.
+\fB%c1, %c3, %c6\fP show the percentage residency in hardware core idle states.
+\fB%pc3, %pc6\fP percentage residency in hardware package idle states.
+.fi
+.PP
+.SH EXAMPLE
+Without any parameters, turbostat prints out counters ever 5 seconds.
+(override interval with "-i sec" option, or specify a command
+for turbostat to fork).
+
+The first row of statistics reflect the average for the entire system.
+Subsequent rows show per-CPU statistics.
+
+.nf
+[root@x980]# ./turbostat
+core CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
+ 0.04 1.62 3.38 0.11 0.00 99.85 0.00 95.07
+ 0 0 0.04 1.62 3.38 0.06 0.00 99.90 0.00 95.07
+ 0 6 0.02 1.62 3.38 0.08 0.00 99.90 0.00 95.07
+ 1 2 0.10 1.62 3.38 0.29 0.00 99.61 0.00 95.07
+ 1 8 0.11 1.62 3.38 0.28 0.00 99.61 0.00 95.07
+ 2 4 0.01 1.62 3.38 0.01 0.00 99.98 0.00 95.07
+ 2 10 0.01 1.61 3.38 0.02 0.00 99.98 0.00 95.07
+ 8 1 0.07 1.62 3.38 0.15 0.00 99.78 0.00 95.07
+ 8 7 0.03 1.62 3.38 0.19 0.00 99.78 0.00 95.07
+ 9 3 0.01 1.62 3.38 0.02 0.00 99.98 0.00 95.07
+ 9 9 0.01 1.62 3.38 0.02 0.00 99.98 0.00 95.07
+ 10 5 0.01 1.62 3.38 0.13 0.00 99.86 0.00 95.07
+ 10 11 0.08 1.62 3.38 0.05 0.00 99.86 0.00 95.07
+.fi
+.SH VERBOSE EXAMPLE
+The "-v" option adds verbosity to the output:
+
+.nf
+GenuineIntel 11 CPUID levels; family:model:stepping 0x6:2c:2 (6:44:2)
+12 * 133 = 1600 MHz max efficiency
+25 * 133 = 3333 MHz TSC frequency
+26 * 133 = 3467 MHz max turbo 4 active cores
+26 * 133 = 3467 MHz max turbo 3 active cores
+27 * 133 = 3600 MHz max turbo 2 active cores
+27 * 133 = 3600 MHz max turbo 1 active cores
+
+.fi
+The \fBmax efficiency\fP frequency, a.k.a. Low Frequency Mode, is the frequency
+available at the minimum package voltage. The \fBTSC frequency\fP is the nominal
+maximum frequency of the processor if turbo-mode were not available. This frequency
+should be sustainable on all CPUs indefinitely, given nominal power and cooling.
+The remaining rows show what maximum turbo frequency is possible
+depending on the number of idle cores. Note that this information is
+not available on all processors.
+.SH FORK EXAMPLE
+If turbostat is invoked with a command, it will fork that command
+and output the statistics gathered when the command exits.
+eg. Here a cycle soaker is run on 1 CPU (see %c0) for a few seconds
+until ^C while the other CPUs are mostly idle:
+
+.nf
+[root@x980 lenb]# ./turbostat cat /dev/zero > /dev/null
+
+^Ccore CPU %c0 GHz TSC %c1 %c3 %c6 %pc3 %pc6
+ 8.49 3.63 3.38 16.23 0.66 74.63 0.00 0.00
+ 0 0 1.22 3.62 3.38 32.18 0.00 66.60 0.00 0.00
+ 0 6 0.40 3.61 3.38 33.00 0.00 66.60 0.00 0.00
+ 1 2 0.11 3.14 3.38 0.19 3.95 95.75 0.00 0.00
+ 1 8 0.05 2.88 3.38 0.25 3.95 95.75 0.00 0.00
+ 2 4 0.00 3.13 3.38 0.02 0.00 99.98 0.00 0.00
+ 2 10 0.00 3.09 3.38 0.02 0.00 99.98 0.00 0.00
+ 8 1 0.04 3.50 3.38 14.43 0.00 85.54 0.00 0.00
+ 8 7 0.03 2.98 3.38 14.43 0.00 85.54 0.00 0.00
+ 9 3 0.00 3.16 3.38 100.00 0.00 0.00 0.00 0.00
+ 9 9 99.93 3.63 3.38 0.06 0.00 0.00 0.00 0.00
+ 10 5 0.01 2.82 3.38 0.08 0.00 99.91 0.00 0.00
+ 10 11 0.02 3.36 3.38 0.06 0.00 99.91 0.00 0.00
+6.950866 sec
+
+.fi
+Above the cycle soaker drives cpu9 up 3.6 Ghz turbo limit
+while the other processors are generally in various states of idle.
+
+Note that cpu3 is an HT sibling sharing core9
+with cpu9, and thus it is unable to get to an idle state
+deeper than c1 while cpu9 is busy.
+
+Note that turbostat reports average GHz of 3.61, while
+the arithmetic average of the GHz column above is 3.24.
+This is a weighted average, where the weight is %c0. ie. it is the total number of
+un-halted cycles elapsed per time divided by the number of CPUs.
+.SH NOTES
+
+.B "turbostat "
+must be run as root.
+
+.B "turbostat "
+reads hardware counters, but doesn't write them.
+So it will not interfere with the OS or other programs, including
+multiple invocations of itself.
+
+\fBturbostat \fP
+may work poorly on Linux-2.6.20 through 2.6.29,
+as \fBacpi-cpufreq \fPperiodically cleared the APERF and MPERF
+in those kernels.
+
+The APERF, MPERF MSRs are defined to count non-halted cycles.
+Although it is not guaranteed by the architecture, turbostat assumes
+that they count at TSC rate, which is true on all processors tested to date.
+
+.SH REFERENCES
+"Intel® Turbo Boost Technology
+in Intel® Core™ Microarchitecture (Nehalem) Based Processors"
+http://download.intel.com/design/processor/applnots/320354.pdf
+
+"Intel® 64 and IA-32 Architectures Software Developer's Manual
+Volume 3B: System Programming Guide"
+http://www.intel.com/products/processor/manuals/
+
+.SH FILES
+.ta
+.nf
+/dev/cpu/*/msr
+.fi
+
+.SH "SEE ALSO"
+msr(4), vmstat(8)
+.PP
+.SH AUTHORS
+.nf
+Written by Len Brown <[email protected]>
diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
new file mode 100644
index 0000000..4c6983d
--- /dev/null
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -0,0 +1,1048 @@
+/*
+ * turbostat -- show CPU frequency and C-state residency
+ * on modern Intel turbo-capable processors.
+ *
+ * Copyright (c) 2010, Intel Corporation.
+ * Len Brown <[email protected]>
+ *
+ * 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/wait.h>
+#include <sys/stat.h>
+#include <sys/resource.h>
+#include <fcntl.h>
+#include <signal.h>
+#include <sys/time.h>
+#include <stdlib.h>
+#include <dirent.h>
+#include <string.h>
+#include <ctype.h>
+
+#define MSR_TSC 0x10
+#define MSR_NEHALEM_PLATFORM_INFO 0xCE
+#define MSR_NEHALEM_TURBO_RATIO_LIMIT 0x1AD
+#define MSR_APERF 0xE8
+#define MSR_MPERF 0xE7
+#define MSR_PKG_C2_RESIDENCY 0x60D /* SNB only */
+#define MSR_PKG_C3_RESIDENCY 0x3F8
+#define MSR_PKG_C6_RESIDENCY 0x3F9
+#define MSR_PKG_C7_RESIDENCY 0x3FA /* SNB only */
+#define MSR_CORE_C3_RESIDENCY 0x3FC
+#define MSR_CORE_C6_RESIDENCY 0x3FD
+#define MSR_CORE_C7_RESIDENCY 0x3FE /* SNB only */
+
+char *proc_stat = "/proc/stat";
+unsigned int interval_sec = 5; /* set with -i interval_sec */
+unsigned int verbose; /* set with -v */
+unsigned int skip_c0;
+unsigned int skip_c1;
+unsigned int do_nhm_cstates;
+unsigned int do_snb_cstates;
+unsigned int has_aperf;
+unsigned int units = 1000000000; /* Ghz etc */
+unsigned int genuine_intel;
+unsigned int has_invariant_tsc;
+unsigned int do_nehalem_platform_info;
+unsigned int do_nehalem_turbo_ratio_limit;
+unsigned int extra_msr_offset;
+double bclk;
+unsigned int show_pkg;
+unsigned int show_core;
+unsigned int show_cpu;
+
+int aperf_mperf_unstable;
+int backwards_count;
+char *progname;
+int need_reinitialize;
+
+int num_cpus;
+
+typedef struct per_cpu_counters {
+ unsigned long long tsc; /* per thread */
+ unsigned long long aperf; /* per thread */
+ unsigned long long mperf; /* per thread */
+ unsigned long long c1; /* per thread (calculated) */
+ unsigned long long c3; /* per core */
+ unsigned long long c6; /* per core */
+ unsigned long long c7; /* per core */
+ unsigned long long pc2; /* per package */
+ unsigned long long pc3; /* per package */
+ unsigned long long pc6; /* per package */
+ unsigned long long pc7; /* per package */
+ unsigned long long extra_msr; /* per thread */
+ int pkg;
+ int core;
+ int cpu;
+ struct per_cpu_counters *next;
+} PCC;
+
+PCC *pcc_even;
+PCC *pcc_odd;
+PCC *pcc_delta;
+PCC *pcc_average;
+struct timeval tv_even;
+struct timeval tv_odd;
+struct timeval tv_delta;
+
+unsigned long long get_msr(int cpu, off_t offset)
+{
+ ssize_t retval;
+ unsigned long long msr;
+ char pathname[32];
+ int fd;
+
+ sprintf(pathname, "/dev/cpu/%d/msr", cpu);
+ fd = open(pathname, O_RDONLY);
+ if (fd < 0) {
+ perror(pathname);
+ need_reinitialize = 1;
+ return 0;
+ }
+
+ retval = pread(fd, &msr, sizeof msr, offset);
+ if (retval != sizeof msr) {
+ fprintf(stderr, "cpu%d pread(..., 0x%zx) = %jd\n",
+ cpu, offset, retval);
+ exit(-2);
+ }
+
+ close(fd);
+ return msr;
+}
+
+void print_header()
+{
+ if (show_pkg)
+ fprintf(stderr, "pkg ");
+ if (show_core)
+ fprintf(stderr, "core");
+ if (show_cpu)
+ fprintf(stderr, " CPU");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c0 ");
+ if (has_aperf)
+ fprintf(stderr, " GHz");
+ fprintf(stderr, " TSC");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c1 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c3 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%c6 ");
+ if (do_snb_cstates)
+ fprintf(stderr, " %%c7 ");
+ if (do_snb_cstates)
+ fprintf(stderr, " %%pc2 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%pc3 ");
+ if (do_nhm_cstates)
+ fprintf(stderr, " %%pc6 ");
+ if (do_snb_cstates)
+ fprintf(stderr, " %%pc7 ");
+ if (extra_msr_offset)
+ fprintf(stderr, " MSR 0x%x ", extra_msr_offset);
+
+ putc('\n', stderr);
+}
+
+void dump_pcc(PCC *pcc)
+{
+ fprintf(stderr, "package: %d ", pcc->pkg);
+ fprintf(stderr, "core:: %d ", pcc->core);
+ fprintf(stderr, "CPU: %d ", pcc->cpu);
+ fprintf(stderr, "TSC: %016llX\n", pcc->tsc);
+ fprintf(stderr, "c3: %016llX\n", pcc->c3);
+ fprintf(stderr, "c6: %016llX\n", pcc->c6);
+ fprintf(stderr, "c7: %016llX\n", pcc->c7);
+ fprintf(stderr, "aperf: %016llX\n", pcc->aperf);
+ fprintf(stderr, "pc2: %016llX\n", pcc->pc2);
+ fprintf(stderr, "pc3: %016llX\n", pcc->pc3);
+ fprintf(stderr, "pc6: %016llX\n", pcc->pc6);
+ fprintf(stderr, "pc7: %016llX\n", pcc->pc7);
+ fprintf(stderr, "msr0x%x: %016llX\n", extra_msr_offset, pcc->extra_msr);
+}
+
+void dump_list(PCC *pcc)
+{
+ printf("dump_list 0x%p\n", pcc);
+
+ for (; pcc; pcc = pcc->next)
+ dump_pcc(pcc);
+}
+
+void print_pcc(PCC *p)
+{
+ double interval_float;
+
+ interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;
+
+ /* topology columns, print blanks on 1st (average) line */
+ if (p == pcc_average) {
+ if (show_pkg)
+ fprintf(stderr, " ");
+ if (show_core)
+ fprintf(stderr, " ");
+ if (show_cpu)
+ fprintf(stderr, " ");
+ } else {
+ if (show_pkg)
+ fprintf(stderr, "%4d", p->pkg);
+ if (show_core)
+ fprintf(stderr, "%4d", p->core);
+ if (show_cpu)
+ fprintf(stderr, "%4d", p->cpu);
+ }
+
+ /* %c0 */
+ if (do_nhm_cstates) {
+ if (!skip_c0)
+ fprintf(stderr, "%7.2f", 100.0 * p->mperf/p->tsc);
+ else
+ fprintf(stderr, " ****");
+ }
+
+ /* GHz */
+ if (has_aperf) {
+ if (!aperf_mperf_unstable) {
+ fprintf(stderr, "%5.2f",
+ 1.0 * p->tsc / units * p->aperf /
+ p->mperf / interval_float);
+ } else {
+ if (p->aperf > p->tsc || p->mperf > p->tsc) {
+ fprintf(stderr, " ****");
+ } else {
+ fprintf(stderr, "%4.1f*",
+ 1.0 * p->tsc /
+ units * p->aperf /
+ p->mperf / interval_float);
+ }
+ }
+ }
+
+ /* TSC */
+ fprintf(stderr, "%5.2f", 1.0 * p->tsc/units/interval_float);
+
+ if (do_nhm_cstates) {
+ if (!skip_c1)
+ fprintf(stderr, "%7.2f", 100.0 * p->c1/p->tsc);
+ else
+ fprintf(stderr, " ****");
+ }
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c3/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c6/p->tsc);
+ if (do_snb_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->c7/p->tsc);
+ if (do_snb_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc2/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc3/p->tsc);
+ if (do_nhm_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc6/p->tsc);
+ if (do_snb_cstates)
+ fprintf(stderr, "%7.2f", 100.0 * p->pc7/p->tsc);
+ if (extra_msr_offset)
+ fprintf(stderr, " 0x%016llx", p->extra_msr);
+ putc('\n', stderr);
+}
+
+void print_counters(PCC *cnt)
+{
+ PCC *pcc;
+
+ print_header();
+
+ if (num_cpus > 1)
+ print_pcc(pcc_average);
+
+ for (pcc = cnt; pcc != NULL; pcc = pcc->next)
+ print_pcc(pcc);
+
+}
+
+#define SUBTRACT_COUNTER(after, before, delta) (delta = (after - before), (before > after))
+
+
+int compute_delta(PCC *after, PCC *before, PCC *delta)
+{
+ int errors = 0;
+ int perf_err = 0;
+
+ skip_c0 = skip_c1 = 0;
+
+ for ( ; after && before && delta;
+ after = after->next, before = before->next, delta = delta->next) {
+ if (before->cpu != after->cpu) {
+ printf("cpu configuration changed: %d != %d\n",
+ before->cpu, after->cpu);
+ return -1;
+ }
+
+ if (SUBTRACT_COUNTER(after->tsc, before->tsc, delta->tsc)) {
+ fprintf(stderr, "cpu%d TSC went backwards %llX to %llX\n",
+ before->cpu, before->tsc, after->tsc);
+ errors++;
+ }
+ /* check for TSC < 1 Mcycles over interval */
+ if (delta->tsc < (1000 * 1000)) {
+ fprintf(stderr, "Insanely slow TSC rate,"
+ " TSC stops in idle?\n");
+ fprintf(stderr, "You can disable all c-states"
+ " by booting with \"idle=poll\"\n");
+ fprintf(stderr, "or just the deep ones with"
+ " \"processor.max_cstate=1\"\n");
+ exit(-3);
+ }
+ if (SUBTRACT_COUNTER(after->c3, before->c3, delta->c3)) {
+ fprintf(stderr, "cpu%d c3 counter went backwards %llX to %llX\n",
+ before->cpu, before->c3, after->c3);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->c6, before->c6, delta->c6)) {
+ fprintf(stderr, "cpu%d c6 counter went backwards %llX to %llX\n",
+ before->cpu, before->c6, after->c6);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->c7, before->c7, delta->c7)) {
+ fprintf(stderr, "cpu%d c7 counter went backwards %llX to %llX\n",
+ before->cpu, before->c7, after->c7);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc2, before->pc2, delta->pc2)) {
+ fprintf(stderr, "cpu%d pc2 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc2, after->pc2);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc3, before->pc3, delta->pc3)) {
+ fprintf(stderr, "cpu%d pc3 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc3, after->pc3);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc6, before->pc6, delta->pc6)) {
+ fprintf(stderr, "cpu%d pc6 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc6, after->pc6);
+ errors++;
+ }
+ if (SUBTRACT_COUNTER(after->pc7, before->pc7, delta->pc7)) {
+ fprintf(stderr, "cpu%d pc7 counter went backwards %llX to %llX\n",
+ before->cpu, before->pc7, after->pc7);
+ errors++;
+ }
+
+ perf_err = SUBTRACT_COUNTER(after->aperf, before->aperf, delta->aperf);
+ if (perf_err) {
+ fprintf(stderr, "cpu%d aperf counter went backwards %llX to %llX\n",
+ before->cpu, before->aperf, after->aperf);
+ }
+ perf_err |= SUBTRACT_COUNTER(after->mperf, before->mperf, delta->mperf);
+ if (perf_err) {
+ fprintf(stderr, "cpu%d mperf counter went backwards %llX to %llX\n",
+ before->cpu, before->mperf, after->mperf);
+ }
+ if (perf_err) {
+ if (!aperf_mperf_unstable) {
+ fprintf(stderr, "%s: APERF or MPERF went backwards *\n", progname);
+ fprintf(stderr, "* Frequency results do not cover entire interval *\n");
+ fprintf(stderr, "* fix this by running Linux-2.6.30 or later *\n");
+
+ aperf_mperf_unstable = 1;
+ }
+ /*
+ * mperf delta is likely a huge "positive" number
+ * can not use it for calculating c0 time
+ */
+ skip_c0 = 1;
+ skip_c1 = 1;
+ }
+
+ /*
+ * As mperf and tsc collection are not atomic,
+ * it is possible for mperf's non-halted cycles
+ * to exceed TSC's all cycles: show c1 = 0% in that case.
+ */
+ if (delta->mperf > delta->tsc)
+ delta->c1 = 0;
+ else /* normal case, derive c1 */
+ delta->c1 = delta->tsc - delta->mperf
+ - delta->c3 - delta->c6 - delta->c7;
+
+ if (delta->mperf == 0)
+ delta->mperf = 1; /* divide by 0 protection */
+
+ /*
+ * for "extra msr", just copy the latest w/o subtracting
+ */
+ delta->extra_msr = after->extra_msr;
+ if (errors) {
+ fprintf(stderr, "ERROR cpu%d before:\n", before->cpu);
+ dump_pcc(before);
+ fprintf(stderr, "ERROR cpu%d after:\n", before->cpu);
+ dump_pcc(after);
+ errors = 0;
+ }
+ }
+ return 0;
+}
+
+void compute_average(PCC *delta, PCC *avg)
+{
+ PCC *sum;
+
+ sum = calloc(1, sizeof(PCC));
+ if (sum == NULL) {
+ perror("calloc sum");
+ exit(1);
+ }
+
+ for (; delta; delta = delta->next) {
+ sum->tsc += delta->tsc;
+ sum->c1 += delta->c1;
+ sum->c3 += delta->c3;
+ sum->c6 += delta->c6;
+ sum->c7 += delta->c7;
+ sum->aperf += delta->aperf;
+ sum->mperf += delta->mperf;
+ sum->pc2 += delta->pc2;
+ sum->pc3 += delta->pc3;
+ sum->pc6 += delta->pc6;
+ sum->pc7 += delta->pc7;
+ }
+ avg->tsc = sum->tsc/num_cpus;
+ avg->c1 = sum->c1/num_cpus;
+ avg->c3 = sum->c3/num_cpus;
+ avg->c6 = sum->c6/num_cpus;
+ avg->c7 = sum->c7/num_cpus;
+ avg->aperf = sum->aperf/num_cpus;
+ avg->mperf = sum->mperf/num_cpus;
+ avg->pc2 = sum->pc2/num_cpus;
+ avg->pc3 = sum->pc3/num_cpus;
+ avg->pc6 = sum->pc6/num_cpus;
+ avg->pc7 = sum->pc7/num_cpus;
+
+ free(sum);
+}
+
+void get_counters(PCC *pcc)
+{
+ for ( ; pcc; pcc = pcc->next) {
+ pcc->tsc = get_msr(pcc->cpu, MSR_TSC);
+ if (do_nhm_cstates)
+ pcc->c3 = get_msr(pcc->cpu, MSR_CORE_C3_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->c6 = get_msr(pcc->cpu, MSR_CORE_C6_RESIDENCY);
+ if (do_snb_cstates)
+ pcc->c7 = get_msr(pcc->cpu, MSR_CORE_C7_RESIDENCY);
+ if (has_aperf)
+ pcc->aperf = get_msr(pcc->cpu, MSR_APERF);
+ if (has_aperf)
+ pcc->mperf = get_msr(pcc->cpu, MSR_MPERF);
+ if (do_snb_cstates)
+ pcc->pc2 = get_msr(pcc->cpu, MSR_PKG_C2_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->pc3 = get_msr(pcc->cpu, MSR_PKG_C3_RESIDENCY);
+ if (do_nhm_cstates)
+ pcc->pc6 = get_msr(pcc->cpu, MSR_PKG_C6_RESIDENCY);
+ if (do_snb_cstates)
+ pcc->pc7 = get_msr(pcc->cpu, MSR_PKG_C7_RESIDENCY);
+ if (extra_msr_offset)
+ pcc->extra_msr = get_msr(pcc->cpu, extra_msr_offset);
+ }
+}
+
+
+void print_nehalem_info()
+{
+ unsigned long long msr;
+ unsigned int ratio;
+
+ if (!do_nehalem_platform_info)
+ return;
+
+ msr = get_msr(0, MSR_NEHALEM_PLATFORM_INFO);
+
+ ratio = (msr >> 40) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 8) & 0xFF;
+ fprintf(stderr, "%d * %.0f = %.0f MHz TSC frequency\n",
+ ratio, bclk, ratio * bclk);
+
+ if (verbose > 1)
+ fprintf(stderr, "MSR_NEHALEM_PLATFORM_INFO: 0x%llx\n", msr);
+
+ if (!do_nehalem_turbo_ratio_limit)
+ return;
+
+ msr = get_msr(0, MSR_NEHALEM_TURBO_RATIO_LIMIT);
+
+ ratio = (msr >> 24) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 16) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 8) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+ ratio = (msr >> 0) & 0xFF;
+ if (ratio)
+ fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
+ ratio, bclk, ratio * bclk);
+
+}
+
+void free_counter_list(PCC *list)
+{
+ PCC *p;
+
+ for (p = list; p; ) {
+ PCC *free_me;
+
+ free_me = p;
+ p = p->next;
+ free(free_me);
+ }
+ return;
+}
+
+void free_all_counters(void)
+{
+ free_counter_list(pcc_even);
+ pcc_even = NULL;
+
+ free_counter_list(pcc_odd);
+ pcc_odd = NULL;
+
+ free_counter_list(pcc_delta);
+ pcc_delta = NULL;
+
+ free_counter_list(pcc_average);
+ pcc_average = NULL;
+}
+
+void insert_cpu_counters(PCC **list, PCC *new)
+{
+ PCC *prev;
+
+ /*
+ * list was empty
+ */
+ if (*list == NULL) {
+ new->next = *list;
+ *list = new;
+ return;
+ }
+
+ show_cpu = 1; /* there is more than one CPU */
+
+ /*
+ * insert on front of list.
+ * It is sorted by ascending package#, core#, cpu#
+ */
+ if (((*list)->pkg > new->pkg) ||
+ (((*list)->pkg == new->pkg) && ((*list)->core > new->core)) ||
+ (((*list)->pkg == new->pkg) && ((*list)->core == new->core) && ((*list)->cpu > new->cpu))) {
+ new->next = *list;
+ *list = new;
+ return;
+ }
+
+ prev = *list;
+
+ while (prev->next && (prev->next->pkg < new->pkg)) {
+ prev = prev->next;
+ show_pkg = 1; /* there is more than 1 package */
+ }
+
+ while (prev->next && (prev->next->pkg == new->pkg)
+ && (prev->next->core < new->core)) {
+ prev = prev->next;
+ show_core = 1; /* there is more than 1 core */
+ }
+
+ while (prev->next && (prev->next->pkg == new->pkg)
+ && (prev->next->core == new->core)
+ && (prev->next->cpu < new->cpu)) {
+ prev = prev->next;
+ }
+
+ /*
+ * insert after "prev"
+ */
+ new->next = prev->next;
+ prev->next = new;
+
+ return;
+}
+
+void alloc_new_cpu_counters(int pkg, int core, int cpu)
+{
+ PCC *new;
+
+ if (verbose > 1)
+ printf("pkg%d core%d, cpu%d\n", pkg, core, cpu);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_odd, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_even, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ insert_cpu_counters(&pcc_delta, new);
+
+ new = (PCC *)calloc(1, sizeof(PCC));
+ if (new == NULL) {
+ perror("calloc");
+ exit(1);
+ }
+ new->pkg = pkg;
+ new->core = core;
+ new->cpu = cpu;
+ pcc_average = new;
+}
+
+int get_physical_package_id(int cpu)
+{
+ char path[64];
+ FILE *filep;
+ int pkg;
+
+ sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
+ filep = fopen(path, "r");
+ if (filep == NULL) {
+ perror(path);
+ exit(1);
+ }
+ fscanf(filep, "%d", &pkg);
+ fclose(filep);
+ return pkg;
+}
+
+int get_core_id(int cpu)
+{
+ char path[64];
+ FILE *filep;
+ int core;
+
+ sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
+ filep = fopen(path, "r");
+ if (filep == NULL) {
+ perror(path);
+ exit(1);
+ }
+ fscanf(filep, "%d", &core);
+ fclose(filep);
+ return core;
+}
+
+/*
+ * run func(index, cpu) on every cpu in /proc/stat
+ */
+
+int for_all_cpus(void (func)(int, int, int))
+{
+ FILE *fp;
+ int cpu_count;
+ 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);
+ }
+
+ for (cpu_count = 0; ; cpu_count++) {
+ int cpu;
+
+ retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu);
+ if (retval != 1)
+ break;
+
+ func(get_physical_package_id(cpu), get_core_id(cpu), cpu);
+ }
+ fclose(fp);
+ return cpu_count;
+}
+
+void re_initialize(void)
+{
+ printf("turbostat: topology changed, re-initializing.\n");
+ free_all_counters();
+ num_cpus = for_all_cpus(alloc_new_cpu_counters);
+ need_reinitialize = 0;
+ printf("num_cpus is now %d\n", num_cpus);
+}
+
+void dummy(int pkg, int core, int cpu) { return; }
+/*
+ * check to see if a cpu came on-line
+ */
+void verify_num_cpus()
+{
+ int new_num_cpus;
+
+ new_num_cpus = for_all_cpus(dummy);
+
+ if (new_num_cpus != num_cpus) {
+ if (verbose)
+ printf("num_cpus was %d, is now %d\n",
+ num_cpus, new_num_cpus);
+ need_reinitialize = 1;
+ }
+
+ return;
+}
+
+void turbostat_loop()
+{
+restart:
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+
+ while (1) {
+ verify_num_cpus();
+ if (need_reinitialize) {
+ re_initialize();
+ goto restart;
+ }
+ sleep(interval_sec);
+ get_counters(pcc_odd);
+ gettimeofday(&tv_odd, (struct timezone *)NULL);
+
+ compute_delta(pcc_odd, pcc_even, pcc_delta);
+ timersub(&tv_odd, &tv_even, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ print_counters(pcc_delta);
+ if (need_reinitialize) {
+ re_initialize();
+ goto restart;
+ }
+ sleep(interval_sec);
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+ compute_delta(pcc_even, pcc_odd, pcc_delta);
+ timersub(&tv_even, &tv_odd, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ print_counters(pcc_delta);
+ }
+}
+
+void check_dev_msr()
+{
+ struct stat sb;
+
+ if (stat("/dev/cpu/0/msr", &sb)) {
+ fprintf(stderr, "no /dev/cpu/0/msr\n");
+ fprintf(stderr, "Try \"# modprobe msr\"\n");
+ exit(-5);
+ }
+}
+
+void check_super_user()
+{
+ if (getuid() != 0) {
+ fprintf(stderr, "must be root\n");
+ exit(-6);
+ }
+}
+
+int has_nehalem_turbo_ratio_limit(unsigned int family, unsigned int model)
+{
+ if (!genuine_intel)
+ return 0;
+
+ if (family != 6)
+ return 0;
+
+ switch (model) {
+ case 0x1A: /* Core i7, Xeon 5500 series - Bloomfield, Gainstown NHM-EP */
+ case 0x1E: /* Core i7 and i5 Processor - Clarksfield, Lynnfield, Jasper Forest */
+ case 0x1F: /* Core i7 and i5 Processor - Nehalem */
+ case 0x25: /* Westmere Client - Clarkdale, Arrandale */
+ case 0x2C: /* Westmere EP - Gulftown */
+ case 0x2A: /* SNB */
+ case 0x2D: /* SNB Xeon */
+ return 1;
+ case 0x2E: /* Nehalem-EX Xeon - Beckton */
+ case 0x2F: /* Westmere-EX Xeon - Eagleton */
+ default:
+ return 0;
+ }
+}
+
+int is_snb(unsigned int family, unsigned int model)
+{
+ if (!genuine_intel)
+ return 0;
+
+ switch (model) {
+ case 0x2A:
+ case 0x2D:
+ return 1;
+ }
+ return 0;
+}
+
+double discover_bclk(unsigned int family, unsigned int model)
+{
+ if (is_snb(family, model))
+ return 100.00;
+ else
+ return 133.33;
+}
+
+void check_cpuid()
+{
+ unsigned int eax, ebx, ecx, edx, max_level;
+ 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)
+ genuine_intel = 1;
+
+ if (verbose)
+ fprintf(stderr, "%.4s%.4s%.4s ",
+ (char *)&ebx, (char *)&edx, (char *)&ecx);
+
+ 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)
+ fprintf(stderr, "%d CPUID levels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
+ max_level, family, model, stepping, family, model, stepping);
+
+ if (!(edx & (1 << 5))) {
+ fprintf(stderr, "CPUID: no MSR\n");
+ exit(1);
+ }
+
+ /*
+ * check max extended function levels of CPUID.
+ * This is needed to check for invariant TSC.
+ * This check is valid for both Intel and AMD.
+ */
+ ebx = ecx = edx = 0;
+ asm("cpuid" : "=a" (max_level), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x80000000));
+
+ if (max_level < 0x80000007) {
+ fprintf(stderr, "CPUID: no invariant TSC (max_level 0x%x)\n", max_level);
+ exit(1);
+ }
+
+ /*
+ * Non-Stop TSC is advertised by CPUID.EAX=0x80000007: EDX.bit8
+ * this check is valid for both Intel and AMD
+ */
+ asm("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x80000007));
+ has_invariant_tsc = edx && (1 << 8);
+
+ if (!has_invariant_tsc) {
+ fprintf(stderr, "No invariant TSC\n");
+ exit(1);
+ }
+
+ /*
+ * APERF/MPERF is advertised by CPUID.EAX=0x6: ECX.bit0
+ * this check is valid for both Intel and AMD
+ */
+
+ asm("cpuid" : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (0x6));
+ has_aperf = ecx && (1 << 0);
+ if (!has_aperf) {
+ fprintf(stderr, "No APERF MSR\n");
+ exit(1);
+ }
+
+ do_nehalem_platform_info = genuine_intel && has_invariant_tsc;
+ do_nhm_cstates = genuine_intel; /* all Intel w/ non-stop TSC have NHM counters */
+ do_snb_cstates = is_snb(family, model);
+ bclk = discover_bclk(family, model);
+
+ do_nehalem_turbo_ratio_limit = has_nehalem_turbo_ratio_limit(family, model);
+}
+
+
+void usage()
+{
+ fprintf(stderr, "%s: [-v] [-M MSR#] [-i interval_sec | command ...]\n",
+ progname);
+ exit(1);
+}
+
+
+/*
+ * in /dev/cpu/ return success for names that are numbers
+ * ie. filter out ".", "..", "microcode".
+ */
+int dir_filter(const struct dirent *dirp)
+{
+ if (isdigit(dirp->d_name[0]))
+ return 1;
+ else
+ return 0;
+}
+
+int open_dev_cpu_msr(int dummy1)
+{
+ return 0;
+}
+
+void turbostat_init()
+{
+ check_cpuid();
+
+ check_dev_msr();
+ check_super_user();
+
+ num_cpus = for_all_cpus(alloc_new_cpu_counters);
+
+ if (verbose)
+ print_nehalem_info();
+}
+
+int fork_it(char **argv)
+{
+ int retval;
+ pid_t child_pid;
+ get_counters(pcc_even);
+ gettimeofday(&tv_even, (struct timezone *)NULL);
+
+ child_pid = fork();
+ if (!child_pid) {
+ /* child */
+ execvp(argv[0], argv);
+ } else {
+ int status;
+
+ /* parent */
+ if (child_pid == -1) {
+ perror("fork");
+ exit(1);
+ }
+
+ signal(SIGINT, SIG_IGN);
+ signal(SIGQUIT, SIG_IGN);
+ if (waitpid(child_pid, &status, 0) == -1) {
+ perror("wait");
+ exit(1);
+ }
+ }
+ get_counters(pcc_odd);
+ gettimeofday(&tv_odd, (struct timezone *)NULL);
+ retval = compute_delta(pcc_odd, pcc_even, pcc_delta);
+
+ timersub(&tv_odd, &tv_even, &tv_delta);
+ compute_average(pcc_delta, pcc_average);
+ if (!retval)
+ print_counters(pcc_delta);
+
+ fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);;
+
+ return 0;
+}
+
+void cmdline(int argc, char **argv)
+{
+ int opt;
+
+ progname = argv[0];
+
+ while ((opt = getopt(argc, argv, "+vi:M:")) != -1) {
+ switch (opt) {
+ case 'v':
+ verbose++;
+ break;
+ case 'i':
+ interval_sec = atoi(optarg);
+ break;
+ case 'M':
+ sscanf(optarg, "%x", &extra_msr_offset);
+ if (verbose > 1)
+ fprintf(stderr, "MSR 0x%X\n", extra_msr_offset);
+ break;
+ default:
+ usage();
+ }
+ }
+}
+
+int main(int argc, char **argv)
+{
+ cmdline(argc, argv);
+
+ if (verbose > 1)
+ fprintf(stderr, "turbostat Dec 6, 2010"
+ " - Len Brown <[email protected]>\n");
+ if (verbose > 1)
+ fprintf(stderr, "http://userweb.kernel.org/~lenb/acpi/utils/pmtools/turbostat/\n");
+
+ turbostat_init();
+
+ /*
+ * if any params left, it must be a command to fork
+ */
+ if (argc - optind)
+ return fork_it(argv + optind);
+ else
+ turbostat_loop();
+
+ return 0;
+}

2010-12-07 14:29:18

by Thiago Farina

Subject: Re: [PATCH v4] tools: create power/x86/turbostat

On Tue, Dec 7, 2010 at 4:00 AM, Len Brown <[email protected]> wrote:
> From: Len Brown <[email protected]>
>
> turbostat is a tool to help verify proper operation
> of processor power management states on modern x86.
>
> Signed-off-by: Len Brown <[email protected]>
> ---
> v4: fix open file descriptor leak, reported by David C. Wong
>
>  tools/power/x86/turbostat/Makefile    |    8 +
>  tools/power/x86/turbostat/turbostat.8 |  172 ++++++
>  tools/power/x86/turbostat/turbostat.c | 1048 +++++++++++++++++++++++++++++++++
>  3 files changed, 1228 insertions(+), 0 deletions(-)
>
> diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
> new file mode 100644
> index 0000000..4c6983d
> --- /dev/null
> +++ b/tools/power/x86/turbostat/turbostat.c
> @@ -0,0 +1,1048 @@
> +/*
> + * turbostat -- show CPU frequency and C-state residency
> + * on modern Intel turbo-capable processors.
> + *
> + * Copyright (c) 2010, Intel Corporation.
> + * Len Brown <[email protected]>
> + *
> + * 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/wait.h>
> +#include <sys/stat.h>
> +#include <sys/resource.h>
> +#include <fcntl.h>
> +#include <signal.h>
> +#include <sys/time.h>
> +#include <stdlib.h>
> +#include <dirent.h>
> +#include <string.h>
> +#include <ctype.h>
> +
> +#define MSR_TSC        0x10
> +#define MSR_NEHALEM_PLATFORM_INFO      0xCE
> +#define MSR_NEHALEM_TURBO_RATIO_LIMIT  0x1AD
> +#define MSR_APERF      0xE8
> +#define MSR_MPERF      0xE7
> +#define MSR_PKG_C2_RESIDENCY   0x60D   /* SNB only */
> +#define MSR_PKG_C3_RESIDENCY   0x3F8
> +#define MSR_PKG_C6_RESIDENCY   0x3F9
> +#define MSR_PKG_C7_RESIDENCY   0x3FA   /* SNB only */
> +#define MSR_CORE_C3_RESIDENCY  0x3FC
> +#define MSR_CORE_C6_RESIDENCY  0x3FD
> +#define MSR_CORE_C7_RESIDENCY  0x3FE   /* SNB only */
> +
> +char *proc_stat = "/proc/stat";
> +unsigned int interval_sec = 5; /* set with -i interval_sec */
> +unsigned int verbose;          /* set with -v */
> +unsigned int skip_c0;
> +unsigned int skip_c1;
> +unsigned int do_nhm_cstates;
> +unsigned int do_snb_cstates;
> +unsigned int has_aperf;
> +unsigned int units = 1000000000;       /* Ghz etc */
> +unsigned int genuine_intel;
> +unsigned int has_invariant_tsc;
> +unsigned int do_nehalem_platform_info;
> +unsigned int do_nehalem_turbo_ratio_limit;
> +unsigned int extra_msr_offset;
> +double bclk;
> +unsigned int show_pkg;
> +unsigned int show_core;
> +unsigned int show_cpu;
> +
> +int aperf_mperf_unstable;
> +int backwards_count;
> +char *progname;
> +int need_reinitialize;
> +
> +int num_cpus;
> +
> +typedef struct per_cpu_counters {
> +       unsigned long long tsc;         /* per thread */
> +       unsigned long long aperf;       /* per thread */
> +       unsigned long long mperf;       /* per thread */
> +       unsigned long long c1;  /* per thread (calculated) */
> +       unsigned long long c3;  /* per core */
> +       unsigned long long c6;  /* per core */
> +       unsigned long long c7;  /* per core */
> +       unsigned long long pc2; /* per package */
> +       unsigned long long pc3; /* per package */
> +       unsigned long long pc6; /* per package */
> +       unsigned long long pc7; /* per package */
> +       unsigned long long extra_msr;   /* per thread */
> +       int pkg;
> +       int core;
> +       int cpu;
> +       struct per_cpu_counters *next;
> +} PCC;
> +

Why the typedef? Isn't preferable to use struct per_cpu_counters all
around (yeah I know it requires more typing, but is less obfuscated).

> +PCC *pcc_even;
> +PCC *pcc_odd;
> +PCC *pcc_delta;
> +PCC *pcc_average;
> +struct timeval tv_even;
> +struct timeval tv_odd;
> +struct timeval tv_delta;
> +
> +unsigned long long get_msr(int cpu, off_t offset)
> +{
> +       ssize_t retval;
> +       unsigned long long msr;
> +       char pathname[32];
> +       int fd;
> +
> +       sprintf(pathname, "/dev/cpu/%d/msr", cpu);
> +       fd = open(pathname, O_RDONLY);
> +       if (fd < 0) {
> +               perror(pathname);
> +               need_reinitialize = 1;
> +               return 0;
> +       }
> +
> +       retval = pread(fd, &msr, sizeof msr, offset);
> +       if (retval != sizeof msr) {
> +               fprintf(stderr, "cpu%d pread(..., 0x%zx) = %jd\n",
> +                       cpu, offset, retval);
> +               exit(-2);
> +       }
> +
> +       close(fd);
> +       return msr;
> +}
> +
> +void print_header()
use void in the paramenter? Also make this static?

> +
> +#define SUBTRACT_COUNTER(after, before, delta) (delta = (after - before), (before > after))
> +
> +
Remove this extra blank line?

> +
> +void get_counters(PCC *pcc)
> +{
> +       for ( ; pcc; pcc = pcc->next) {
> +               pcc->tsc = get_msr(pcc->cpu, MSR_TSC);
> +               if (do_nhm_cstates)
> +                       pcc->c3 = get_msr(pcc->cpu, MSR_CORE_C3_RESIDENCY);
> +               if (do_nhm_cstates)
> +                       pcc->c6 = get_msr(pcc->cpu, MSR_CORE_C6_RESIDENCY);
> +               if (do_snb_cstates)
> +                       pcc->c7 = get_msr(pcc->cpu, MSR_CORE_C7_RESIDENCY);
> +               if (has_aperf)
> +                       pcc->aperf = get_msr(pcc->cpu, MSR_APERF);
> +               if (has_aperf)
> +                       pcc->mperf = get_msr(pcc->cpu, MSR_MPERF);
> +               if (do_snb_cstates)
> +                       pcc->pc2 = get_msr(pcc->cpu, MSR_PKG_C2_RESIDENCY);
> +               if (do_nhm_cstates)
> +                       pcc->pc3 = get_msr(pcc->cpu, MSR_PKG_C3_RESIDENCY);
> +               if (do_nhm_cstates)
> +                       pcc->pc6 = get_msr(pcc->cpu, MSR_PKG_C6_RESIDENCY);
> +               if (do_snb_cstates)
> +                       pcc->pc7 = get_msr(pcc->cpu, MSR_PKG_C7_RESIDENCY);
> +               if (extra_msr_offset)
> +                       pcc->extra_msr = get_msr(pcc->cpu, extra_msr_offset);
> +       }
> +}
> +
> +
Remove this blank line too?

> +void print_nehalem_info()
> +{
Add void into the parameter?

> +       unsigned long long msr;
> +       unsigned int ratio;
> +
> +       if (!do_nehalem_platform_info)
> +               return;
> +
> +       msr = get_msr(0, MSR_NEHALEM_PLATFORM_INFO);
> +
> +       ratio = (msr >> 40) & 0xFF;
> +       fprintf(stderr, "%d * %.0f = %.0f MHz max efficiency\n",
> +               ratio, bclk, ratio * bclk);
> +
> +       ratio = (msr >> 8) & 0xFF;
> +       fprintf(stderr, "%d * %.0f = %.0f MHz TSC frequency\n",
> +               ratio, bclk, ratio * bclk);
> +
> +       if (verbose > 1)
> +               fprintf(stderr, "MSR_NEHALEM_PLATFORM_INFO: 0x%llx\n", msr);
> +
> +       if (!do_nehalem_turbo_ratio_limit)
> +               return;
> +
> +       msr = get_msr(0, MSR_NEHALEM_TURBO_RATIO_LIMIT);
> +
> +       ratio = (msr >> 24) & 0xFF;
> +       if (ratio)
> +               fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 4 active cores\n",
> +                       ratio, bclk, ratio * bclk);
> +
> +       ratio = (msr >> 16) & 0xFF;
> +       if (ratio)
> +               fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 3 active cores\n",
> +                       ratio, bclk, ratio * bclk);
> +
> +       ratio = (msr >> 8) & 0xFF;
> +       if (ratio)
> +               fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 2 active cores\n",
> +                       ratio, bclk, ratio * bclk);
> +
> +       ratio = (msr >> 0) & 0xFF;
> +       if (ratio)
> +               fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 1 active cores\n",
> +                       ratio, bclk, ratio * bclk);
> +
> +}
> +
> +void free_counter_list(PCC *list)
> +{
> +       PCC *p;
> +
> +       for (p = list; p; ) {
> +               PCC *free_me;
> +
> +               free_me = p;
> +               p = p->next;
> +               free(free_me);
> +       }
> +       return;
Why the return here? Isn't redundant?

> +
> +void insert_cpu_counters(PCC **list, PCC *new)
> +{
> +       PCC *prev;
> +
> +       /*
> +        * list was empty
> +        */
> +       if (*list == NULL) {
> +               new->next = *list;
> +               *list = new;
> +               return;
> +       }
> +
> +       show_cpu = 1;   /* there is more than one CPU */
> +
> +       /*
> +        * insert on front of list.
> +        * It is sorted by ascending package#, core#, cpu#
> +        */
> +       if (((*list)->pkg > new->pkg) ||
> +           (((*list)->pkg == new->pkg) && ((*list)->core > new->core)) ||
> +           (((*list)->pkg == new->pkg) && ((*list)->core == new->core) && ((*list)->cpu > new->cpu))) {
> +               new->next = *list;
> +               *list = new;
> +               return;
> +       }
> +
> +       prev = *list;
> +
> +       while (prev->next && (prev->next->pkg < new->pkg)) {
> +               prev = prev->next;
> +               show_pkg = 1;   /* there is more than 1 package */
> +       }
> +
> +       while (prev->next && (prev->next->pkg == new->pkg)
> +               && (prev->next->core < new->core)) {
> +               prev = prev->next;
> +               show_core = 1;  /* there is more than 1 core */
> +       }
> +
> +       while (prev->next && (prev->next->pkg == new->pkg)
> +               && (prev->next->core == new->core)
> +               && (prev->next->cpu < new->cpu)) {
> +               prev = prev->next;
> +       }
> +
> +       /*
> +        * insert after "prev"
> +        */
> +       new->next = prev->next;
> +       prev->next = new;
> +
> +       return;
Same thing here about the return.

> +
> +int get_physical_package_id(int cpu)
> +{
> +       char path[64];
> +       FILE *filep;
> +       int pkg;
> +
> +       sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/physical_package_id", cpu);
> +       filep = fopen(path, "r");
> +       if (filep == NULL) {
> +               perror(path);
> +               exit(1);
> +       }
> +       fscanf(filep, "%d", &pkg);
> +       fclose(filep);
> +       return pkg;
> +}
> +
> +int get_core_id(int cpu)
> +{
> +       char path[64];
> +       FILE *filep;
> +       int core;
> +
> +       sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_id", cpu);
> +       filep = fopen(path, "r");
> +       if (filep == NULL) {
> +               perror(path);
> +               exit(1);
> +       }
> +       fscanf(filep, "%d", &core);
> +       fclose(filep);
> +       return core;
> +}
> +
> +/*
> + * run func(index, cpu) on every cpu in /proc/stat
> + */
> +
> +int for_all_cpus(void (func)(int, int, int))
> +{
> +       FILE *fp;
> +       int cpu_count;
> +       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);
> +       }
> +
> +       for (cpu_count = 0; ; cpu_count++) {
> +               int cpu;
> +
> +               retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu);
> +               if (retval != 1)
> +                       break;
> +
> +               func(get_physical_package_id(cpu), get_core_id(cpu), cpu);
> +       }
> +       fclose(fp);
> +       return cpu_count;
> +}
> +
> +void re_initialize(void)
> +{
> +       printf("turbostat: topology changed, re-initializing.\n");
> +       free_all_counters();
> +       num_cpus = for_all_cpus(alloc_new_cpu_counters);
> +       need_reinitialize = 0;
> +       printf("num_cpus is now %d\n", num_cpus);
> +}
> +
> +void dummy(int pkg, int core, int cpu) { return; }
> +/*
> + * check to see if a cpu came on-line
> + */
> +void verify_num_cpus()
Use void on parameter?

> +{
> +       int new_num_cpus;
> +
> +       new_num_cpus = for_all_cpus(dummy);
> +
> +       if (new_num_cpus != num_cpus) {
> +               if (verbose)
> +                       printf("num_cpus was %d, is now  %d\n",
> +                               num_cpus, new_num_cpus);
> +               need_reinitialize = 1;
> +       }
> +
> +       return;
> +}
No need of this return, right?

2011-02-11 04:43:39

by Len Brown

Subject: Re: [PATCH v4] tools: create power/x86/turbostat

From: Len Brown <[email protected]>

Follow kernel style more closely.
Nuke typdef, re-name "per cpu counters" simply to be counters etc.

This patch changes no functionality.

Suggested-by: Thiago Farina <[email protected]>
Signed-off-by: Len Brown <[email protected]>
---
Thanks for the feedback, Thiago.

-Len

tools/power/x86/turbostat/turbostat.c | 196 ++++++++++++++++-----------------
1 files changed, 96 insertions(+), 100 deletions(-)

diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
index 4c6983d..87d309e 100644
--- a/tools/power/x86/turbostat/turbostat.c
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -72,7 +72,7 @@ int need_reinitialize;

int num_cpus;

-typedef struct per_cpu_counters {
+struct counters {
unsigned long long tsc; /* per thread */
unsigned long long aperf; /* per thread */
unsigned long long mperf; /* per thread */
@@ -88,13 +88,13 @@ typedef struct per_cpu_counters {
int pkg;
int core;
int cpu;
- struct per_cpu_counters *next;
-} PCC;
+ struct counters *next;
+};

-PCC *pcc_even;
-PCC *pcc_odd;
-PCC *pcc_delta;
-PCC *pcc_average;
+struct counters *cnt_even;
+struct counters *cnt_odd;
+struct counters *cnt_delta;
+struct counters *cnt_average;
struct timeval tv_even;
struct timeval tv_odd;
struct timeval tv_delta;
@@ -125,7 +125,7 @@ unsigned long long get_msr(int cpu, off_t offset)
return msr;
}

-void print_header()
+void print_header(void)
{
if (show_pkg)
fprintf(stderr, "pkg ");
@@ -160,39 +160,39 @@ void print_header()
putc('\n', stderr);
}

-void dump_pcc(PCC *pcc)
+void dump_cnt(struct counters *cnt)
{
- fprintf(stderr, "package: %d ", pcc->pkg);
- fprintf(stderr, "core:: %d ", pcc->core);
- fprintf(stderr, "CPU: %d ", pcc->cpu);
- fprintf(stderr, "TSC: %016llX\n", pcc->tsc);
- fprintf(stderr, "c3: %016llX\n", pcc->c3);
- fprintf(stderr, "c6: %016llX\n", pcc->c6);
- fprintf(stderr, "c7: %016llX\n", pcc->c7);
- fprintf(stderr, "aperf: %016llX\n", pcc->aperf);
- fprintf(stderr, "pc2: %016llX\n", pcc->pc2);
- fprintf(stderr, "pc3: %016llX\n", pcc->pc3);
- fprintf(stderr, "pc6: %016llX\n", pcc->pc6);
- fprintf(stderr, "pc7: %016llX\n", pcc->pc7);
- fprintf(stderr, "msr0x%x: %016llX\n", extra_msr_offset, pcc->extra_msr);
+ fprintf(stderr, "package: %d ", cnt->pkg);
+ fprintf(stderr, "core:: %d ", cnt->core);
+ fprintf(stderr, "CPU: %d ", cnt->cpu);
+ fprintf(stderr, "TSC: %016llX\n", cnt->tsc);
+ fprintf(stderr, "c3: %016llX\n", cnt->c3);
+ fprintf(stderr, "c6: %016llX\n", cnt->c6);
+ fprintf(stderr, "c7: %016llX\n", cnt->c7);
+ fprintf(stderr, "aperf: %016llX\n", cnt->aperf);
+ fprintf(stderr, "pc2: %016llX\n", cnt->pc2);
+ fprintf(stderr, "pc3: %016llX\n", cnt->pc3);
+ fprintf(stderr, "pc6: %016llX\n", cnt->pc6);
+ fprintf(stderr, "pc7: %016llX\n", cnt->pc7);
+ fprintf(stderr, "msr0x%x: %016llX\n", extra_msr_offset, cnt->extra_msr);
}

-void dump_list(PCC *pcc)
+void dump_list(struct counters *cnt)
{
- printf("dump_list 0x%p\n", pcc);
+ printf("dump_list 0x%p\n", cnt);

- for (; pcc; pcc = pcc->next)
- dump_pcc(pcc);
+ for (; cnt; cnt = cnt->next)
+ dump_cnt(cnt);
}

-void print_pcc(PCC *p)
+void print_cnt(struct counters *p)
{
double interval_float;

interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0;

/* topology columns, print blanks on 1st (average) line */
- if (p == pcc_average) {
+ if (p == cnt_average) {
if (show_pkg)
fprintf(stderr, " ");
if (show_core)
@@ -262,24 +262,24 @@ void print_pcc(PCC *p)
putc('\n', stderr);
}

-void print_counters(PCC *cnt)
+void print_counters(struct counters *counters)
{
- PCC *pcc;
+ struct counters *cnt;

print_header();

if (num_cpus > 1)
- print_pcc(pcc_average);
+ print_cnt(cnt_average);

- for (pcc = cnt; pcc != NULL; pcc = pcc->next)
- print_pcc(pcc);
+ for (cnt = counters; cnt != NULL; cnt = cnt->next)
+ print_cnt(cnt);

}

#define SUBTRACT_COUNTER(after, before, delta) (delta = (after - before), (before > after))

-
-int compute_delta(PCC *after, PCC *before, PCC *delta)
+int compute_delta(struct counters *after,
+ struct counters *before, struct counters *delta)
{
int errors = 0;
int perf_err = 0;
@@ -391,20 +391,20 @@ int compute_delta(PCC *after, PCC *before, PCC *delta)
delta->extra_msr = after->extra_msr;
if (errors) {
fprintf(stderr, "ERROR cpu%d before:\n", before->cpu);
- dump_pcc(before);
+ dump_cnt(before);
fprintf(stderr, "ERROR cpu%d after:\n", before->cpu);
- dump_pcc(after);
+ dump_cnt(after);
errors = 0;
}
}
return 0;
}

-void compute_average(PCC *delta, PCC *avg)
+void compute_average(struct counters *delta, struct counters *avg)
{
- PCC *sum;
+ struct counters *sum;

- sum = calloc(1, sizeof(PCC));
+ sum = calloc(1, sizeof(struct counters));
if (sum == NULL) {
perror("calloc sum");
exit(1);
@@ -438,35 +438,34 @@ void compute_average(PCC *delta, PCC *avg)
free(sum);
}

-void get_counters(PCC *pcc)
+void get_counters(struct counters *cnt)
{
- for ( ; pcc; pcc = pcc->next) {
- pcc->tsc = get_msr(pcc->cpu, MSR_TSC);
+ for ( ; cnt; cnt = cnt->next) {
+ cnt->tsc = get_msr(cnt->cpu, MSR_TSC);
if (do_nhm_cstates)
- pcc->c3 = get_msr(pcc->cpu, MSR_CORE_C3_RESIDENCY);
+ cnt->c3 = get_msr(cnt->cpu, MSR_CORE_C3_RESIDENCY);
if (do_nhm_cstates)
- pcc->c6 = get_msr(pcc->cpu, MSR_CORE_C6_RESIDENCY);
+ cnt->c6 = get_msr(cnt->cpu, MSR_CORE_C6_RESIDENCY);
if (do_snb_cstates)
- pcc->c7 = get_msr(pcc->cpu, MSR_CORE_C7_RESIDENCY);
+ cnt->c7 = get_msr(cnt->cpu, MSR_CORE_C7_RESIDENCY);
if (has_aperf)
- pcc->aperf = get_msr(pcc->cpu, MSR_APERF);
+ cnt->aperf = get_msr(cnt->cpu, MSR_APERF);
if (has_aperf)
- pcc->mperf = get_msr(pcc->cpu, MSR_MPERF);
+ cnt->mperf = get_msr(cnt->cpu, MSR_MPERF);
if (do_snb_cstates)
- pcc->pc2 = get_msr(pcc->cpu, MSR_PKG_C2_RESIDENCY);
+ cnt->pc2 = get_msr(cnt->cpu, MSR_PKG_C2_RESIDENCY);
if (do_nhm_cstates)
- pcc->pc3 = get_msr(pcc->cpu, MSR_PKG_C3_RESIDENCY);
+ cnt->pc3 = get_msr(cnt->cpu, MSR_PKG_C3_RESIDENCY);
if (do_nhm_cstates)
- pcc->pc6 = get_msr(pcc->cpu, MSR_PKG_C6_RESIDENCY);
+ cnt->pc6 = get_msr(cnt->cpu, MSR_PKG_C6_RESIDENCY);
if (do_snb_cstates)
- pcc->pc7 = get_msr(pcc->cpu, MSR_PKG_C7_RESIDENCY);
+ cnt->pc7 = get_msr(cnt->cpu, MSR_PKG_C7_RESIDENCY);
if (extra_msr_offset)
- pcc->extra_msr = get_msr(pcc->cpu, extra_msr_offset);
+ cnt->extra_msr = get_msr(cnt->cpu, extra_msr_offset);
}
}

-
-void print_nehalem_info()
+void print_nehalem_info(void)
{
unsigned long long msr;
unsigned int ratio;
@@ -514,38 +513,38 @@ void print_nehalem_info()

}

-void free_counter_list(PCC *list)
+void free_counter_list(struct counters *list)
{
- PCC *p;
+ struct counters *p;

for (p = list; p; ) {
- PCC *free_me;
+ struct counters *free_me;

free_me = p;
p = p->next;
free(free_me);
}
- return;
}

void free_all_counters(void)
{
- free_counter_list(pcc_even);
- pcc_even = NULL;
+ free_counter_list(cnt_even);
+ cnt_even = NULL;

- free_counter_list(pcc_odd);
- pcc_odd = NULL;
+ free_counter_list(cnt_odd);
+ cnt_odd = NULL;

- free_counter_list(pcc_delta);
- pcc_delta = NULL;
+ free_counter_list(cnt_delta);
+ cnt_delta = NULL;

- free_counter_list(pcc_average);
- pcc_average = NULL;
+ free_counter_list(cnt_average);
+ cnt_average = NULL;
}

-void insert_cpu_counters(PCC **list, PCC *new)
+void insert_counters(struct counters **list,
+ struct counters *new)
{
- PCC *prev;
+ struct counters *prev;

/*
* list was empty
@@ -594,18 +593,16 @@ void insert_cpu_counters(PCC **list, PCC *new)
*/
new->next = prev->next;
prev->next = new;
-
- return;
}

-void alloc_new_cpu_counters(int pkg, int core, int cpu)
+void alloc_new_counters(int pkg, int core, int cpu)
{
- PCC *new;
+ struct counters *new;

if (verbose > 1)
printf("pkg%d core%d, cpu%d\n", pkg, core, cpu);

- new = (PCC *)calloc(1, sizeof(PCC));
+ new = (struct counters *)calloc(1, sizeof(struct counters));
if (new == NULL) {
perror("calloc");
exit(1);
@@ -613,9 +610,10 @@ void alloc_new_cpu_counters(int pkg, int core, int cpu)
new->pkg = pkg;
new->core = core;
new->cpu = cpu;
- insert_cpu_counters(&pcc_odd, new);
+ insert_counters(&cnt_odd, new);

- new = (PCC *)calloc(1, sizeof(PCC));
+ new = (struct counters *)calloc(1,
+ sizeof(struct counters));
if (new == NULL) {
perror("calloc");
exit(1);
@@ -623,9 +621,9 @@ void alloc_new_cpu_counters(int pkg, int core, int cpu)
new->pkg = pkg;
new->core = core;
new->cpu = cpu;
- insert_cpu_counters(&pcc_even, new);
+ insert_counters(&cnt_even, new);

- new = (PCC *)calloc(1, sizeof(PCC));
+ new = (struct counters *)calloc(1, sizeof(struct counters));
if (new == NULL) {
perror("calloc");
exit(1);
@@ -633,9 +631,9 @@ void alloc_new_cpu_counters(int pkg, int core, int cpu)
new->pkg = pkg;
new->core = core;
new->cpu = cpu;
- insert_cpu_counters(&pcc_delta, new);
+ insert_counters(&cnt_delta, new);

- new = (PCC *)calloc(1, sizeof(PCC));
+ new = (struct counters *)calloc(1, sizeof(struct counters));
if (new == NULL) {
perror("calloc");
exit(1);
@@ -643,7 +641,7 @@ void alloc_new_cpu_counters(int pkg, int core, int cpu)
new->pkg = pkg;
new->core = core;
new->cpu = cpu;
- pcc_average = new;
+ cnt_average = new;
}

int get_physical_package_id(int cpu)
@@ -719,7 +717,7 @@ void re_initialize(void)
{
printf("turbostat: topology changed, re-initializing.\n");
free_all_counters();
- num_cpus = for_all_cpus(alloc_new_cpu_counters);
+ num_cpus = for_all_cpus(alloc_new_counters);
need_reinitialize = 0;
printf("num_cpus is now %d\n", num_cpus);
}
@@ -728,7 +726,7 @@ void dummy(int pkg, int core, int cpu) { return; }
/*
* check to see if a cpu came on-line
*/
-void verify_num_cpus()
+void verify_num_cpus(void)
{
int new_num_cpus;

@@ -740,14 +738,12 @@ void verify_num_cpus()
num_cpus, new_num_cpus);
need_reinitialize = 1;
}
-
- return;
}

void turbostat_loop()
{
restart:
- get_counters(pcc_even);
+ get_counters(cnt_even);
gettimeofday(&tv_even, (struct timezone *)NULL);

while (1) {
@@ -757,24 +753,24 @@ restart:
goto restart;
}
sleep(interval_sec);
- get_counters(pcc_odd);
+ get_counters(cnt_odd);
gettimeofday(&tv_odd, (struct timezone *)NULL);

- compute_delta(pcc_odd, pcc_even, pcc_delta);
+ compute_delta(cnt_odd, cnt_even, cnt_delta);
timersub(&tv_odd, &tv_even, &tv_delta);
- compute_average(pcc_delta, pcc_average);
- print_counters(pcc_delta);
+ compute_average(cnt_delta, cnt_average);
+ print_counters(cnt_delta);
if (need_reinitialize) {
re_initialize();
goto restart;
}
sleep(interval_sec);
- get_counters(pcc_even);
+ get_counters(cnt_even);
gettimeofday(&tv_even, (struct timezone *)NULL);
- compute_delta(pcc_even, pcc_odd, pcc_delta);
+ compute_delta(cnt_even, cnt_odd, cnt_delta);
timersub(&tv_even, &tv_odd, &tv_delta);
- compute_average(pcc_delta, pcc_average);
- print_counters(pcc_delta);
+ compute_average(cnt_delta, cnt_average);
+ print_counters(cnt_delta);
}
}

@@ -952,7 +948,7 @@ void turbostat_init()
check_dev_msr();
check_super_user();

- num_cpus = for_all_cpus(alloc_new_cpu_counters);
+ num_cpus = for_all_cpus(alloc_new_counters);

if (verbose)
print_nehalem_info();
@@ -962,7 +958,7 @@ int fork_it(char **argv)
{
int retval;
pid_t child_pid;
- get_counters(pcc_even);
+ get_counters(cnt_even);
gettimeofday(&tv_even, (struct timezone *)NULL);

child_pid = fork();
@@ -985,14 +981,14 @@ int fork_it(char **argv)
exit(1);
}
}
- get_counters(pcc_odd);
+ get_counters(cnt_odd);
gettimeofday(&tv_odd, (struct timezone *)NULL);
- retval = compute_delta(pcc_odd, pcc_even, pcc_delta);
+ retval = compute_delta(cnt_odd, cnt_even, cnt_delta);

timersub(&tv_odd, &tv_even, &tv_delta);
- compute_average(pcc_delta, pcc_average);
+ compute_average(cnt_delta, cnt_average);
if (!retval)
- print_counters(pcc_delta);
+ print_counters(cnt_delta);

fprintf(stderr, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec/1000000.0);;

--
1.7.4.45.g1a9f