/* * turbostat -- show CPU frequency and C-state residency * on modern Intel turbo-capable processors. * * Copyright (c) 2012 Intel Corporation. * Len Brown * * This program is free software; you can redistribute it and/or modify it * under the terms and conditions of the GNU General Public License, * version 2, as published by the Free Software Foundation. * * This program is distributed in the hope it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for * more details. * * You should have received a copy of the GNU General Public License along with * this program; if not, write to the Free Software Foundation, Inc., * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. */ #define _GNU_SOURCE #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define MSR_NEHALEM_PLATFORM_INFO 0xCE #define MSR_NEHALEM_TURBO_RATIO_LIMIT 0x1AD #define MSR_IVT_TURBO_RATIO_LIMIT 0x1AE #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 summary_only; /* set with -s */ 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 do_ivt_turbo_ratio_limit; unsigned int extra_msr_offset32; unsigned int extra_msr_offset64; unsigned int extra_delta_offset32; unsigned int extra_delta_offset64; double bclk; unsigned int show_pkg; unsigned int show_core; unsigned int show_cpu; unsigned int show_pkg_only; unsigned int show_core_only; char *output_buffer, *outp; int aperf_mperf_unstable; int backwards_count; char *progname; cpu_set_t *cpu_present_set, *cpu_affinity_set; size_t cpu_present_setsize, cpu_affinity_setsize; struct thread_data { unsigned long long tsc; unsigned long long aperf; unsigned long long mperf; unsigned long long c1; /* derived */ unsigned long long extra_msr64; unsigned long long extra_delta64; unsigned long long extra_msr32; unsigned long long extra_delta32; unsigned int cpu_id; unsigned int flags; #define CPU_IS_FIRST_THREAD_IN_CORE 0x2 #define CPU_IS_FIRST_CORE_IN_PACKAGE 0x4 } *thread_even, *thread_odd; struct core_data { unsigned long long c3; unsigned long long c6; unsigned long long c7; unsigned int core_id; } *core_even, *core_odd; struct pkg_data { unsigned long long pc2; unsigned long long pc3; unsigned long long pc6; unsigned long long pc7; unsigned int package_id; } *package_even, *package_odd; #define ODD_COUNTERS thread_odd, core_odd, package_odd #define EVEN_COUNTERS thread_even, core_even, package_even #define GET_THREAD(thread_base, thread_no, core_no, pkg_no) \ (thread_base + (pkg_no) * topo.num_cores_per_pkg * \ topo.num_threads_per_core + \ (core_no) * topo.num_threads_per_core + (thread_no)) #define GET_CORE(core_base, core_no, pkg_no) \ (core_base + (pkg_no) * topo.num_cores_per_pkg + (core_no)) #define GET_PKG(pkg_base, pkg_no) (pkg_base + pkg_no) struct system_summary { struct thread_data threads; struct core_data cores; struct pkg_data packages; } sum, average; struct topo_params { int num_packages; int num_cpus; int num_cores; int max_cpu_num; int num_cores_per_pkg; int num_threads_per_core; } topo; struct timeval tv_even, tv_odd, tv_delta; void setup_all_buffers(void); int cpu_is_not_present(int cpu) { return !CPU_ISSET_S(cpu, cpu_present_setsize, cpu_present_set); } /* * run func(thread, core, package) in topology order * skip non-present cpus */ int for_all_cpus(int (func)(struct thread_data *, struct core_data *, struct pkg_data *), struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base) { int retval, pkg_no, core_no, thread_no; for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { for (thread_no = 0; thread_no < topo.num_threads_per_core; ++thread_no) { struct thread_data *t; struct core_data *c; struct pkg_data *p; t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); if (cpu_is_not_present(t->cpu_id)) continue; c = GET_CORE(core_base, core_no, pkg_no); p = GET_PKG(pkg_base, pkg_no); retval = func(t, c, p); if (retval) return retval; } } } return 0; } int cpu_migrate(int cpu) { CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); CPU_SET_S(cpu, cpu_affinity_setsize, cpu_affinity_set); if (sched_setaffinity(0, cpu_affinity_setsize, cpu_affinity_set) == -1) return -1; else return 0; } int get_msr(int cpu, off_t offset, unsigned long long *msr) { ssize_t retval; char pathname[32]; int fd; sprintf(pathname, "/dev/cpu/%d/msr", cpu); fd = open(pathname, O_RDONLY); if (fd < 0) return -1; retval = pread(fd, msr, sizeof *msr, offset); close(fd); if (retval != sizeof *msr) return -1; return 0; } void print_header(void) { if (show_pkg) outp += sprintf(outp, "pk"); if (show_pkg) outp += sprintf(outp, " "); if (show_core) outp += sprintf(outp, "cor"); if (show_cpu) outp += sprintf(outp, " CPU"); if (show_pkg || show_core || show_cpu) outp += sprintf(outp, " "); if (do_nhm_cstates) outp += sprintf(outp, " %%c0"); if (has_aperf) outp += sprintf(outp, " GHz"); outp += sprintf(outp, " TSC"); if (extra_delta_offset32) outp += sprintf(outp, " delta 0x%03X", extra_delta_offset32); if (extra_delta_offset64) outp += sprintf(outp, " DELTA 0x%03X", extra_delta_offset64); if (extra_msr_offset32) outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset32); if (extra_msr_offset64) outp += sprintf(outp, " MSR 0x%03X", extra_msr_offset64); if (do_nhm_cstates) outp += sprintf(outp, " %%c1"); if (do_nhm_cstates) outp += sprintf(outp, " %%c3"); if (do_nhm_cstates) outp += sprintf(outp, " %%c6"); if (do_snb_cstates) outp += sprintf(outp, " %%c7"); if (do_snb_cstates) outp += sprintf(outp, " %%pc2"); if (do_nhm_cstates) outp += sprintf(outp, " %%pc3"); if (do_nhm_cstates) outp += sprintf(outp, " %%pc6"); if (do_snb_cstates) outp += sprintf(outp, " %%pc7"); outp += sprintf(outp, "\n"); } int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { fprintf(stderr, "t %p, c %p, p %p\n", t, c, p); if (t) { fprintf(stderr, "CPU: %d flags 0x%x\n", t->cpu_id, t->flags); fprintf(stderr, "TSC: %016llX\n", t->tsc); fprintf(stderr, "aperf: %016llX\n", t->aperf); fprintf(stderr, "mperf: %016llX\n", t->mperf); fprintf(stderr, "c1: %016llX\n", t->c1); fprintf(stderr, "msr0x%x: %08llX\n", extra_delta_offset32, t->extra_delta32); fprintf(stderr, "msr0x%x: %016llX\n", extra_delta_offset64, t->extra_delta64); fprintf(stderr, "msr0x%x: %08llX\n", extra_msr_offset32, t->extra_msr32); fprintf(stderr, "msr0x%x: %016llX\n", extra_msr_offset64, t->extra_msr64); } if (c) { fprintf(stderr, "core: %d\n", c->core_id); fprintf(stderr, "c3: %016llX\n", c->c3); fprintf(stderr, "c6: %016llX\n", c->c6); fprintf(stderr, "c7: %016llX\n", c->c7); } if (p) { fprintf(stderr, "package: %d\n", p->package_id); fprintf(stderr, "pc2: %016llX\n", p->pc2); fprintf(stderr, "pc3: %016llX\n", p->pc3); fprintf(stderr, "pc6: %016llX\n", p->pc6); fprintf(stderr, "pc7: %016llX\n", p->pc7); } return 0; } /* * column formatting convention & formats * package: "pk" 2 columns %2d * core: "cor" 3 columns %3d * CPU: "CPU" 3 columns %3d * GHz: "GHz" 3 columns %3.2 * TSC: "TSC" 3 columns %3.2 * percentage " %pc3" %6.2 */ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { double interval_float; /* if showing only 1st thread in core and this isn't one, bail out */ if (show_core_only && !(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) return 0; /* if showing only 1st thread in pkg and this isn't one, bail out */ if (show_pkg_only && !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) return 0; interval_float = tv_delta.tv_sec + tv_delta.tv_usec/1000000.0; /* topo columns, print blanks on 1st (average) line */ if (t == &average.threads) { if (show_pkg) outp += sprintf(outp, " "); if (show_pkg && show_core) outp += sprintf(outp, " "); if (show_core) outp += sprintf(outp, " "); if (show_cpu) outp += sprintf(outp, " " " "); } else { if (show_pkg) { if (p) outp += sprintf(outp, "%2d", p->package_id); else outp += sprintf(outp, " "); } if (show_pkg && show_core) outp += sprintf(outp, " "); if (show_core) { if (c) outp += sprintf(outp, "%3d", c->core_id); else outp += sprintf(outp, " "); } if (show_cpu) outp += sprintf(outp, " %3d", t->cpu_id); } /* %c0 */ if (do_nhm_cstates) { if (show_pkg || show_core || show_cpu) outp += sprintf(outp, " "); if (!skip_c0) outp += sprintf(outp, "%6.2f", 100.0 * t->mperf/t->tsc); else outp += sprintf(outp, " ****"); } /* GHz */ if (has_aperf) { if (!aperf_mperf_unstable) { outp += sprintf(outp, " %3.2f", 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float); } else { if (t->aperf > t->tsc || t->mperf > t->tsc) { outp += sprintf(outp, " ***"); } else { outp += sprintf(outp, "%3.1f*", 1.0 * t->tsc / units * t->aperf / t->mperf / interval_float); } } } /* TSC */ outp += sprintf(outp, "%5.2f", 1.0 * t->tsc/units/interval_float); /* delta */ if (extra_delta_offset32) outp += sprintf(outp, " %11llu", t->extra_delta32); /* DELTA */ if (extra_delta_offset64) outp += sprintf(outp, " %11llu", t->extra_delta64); /* msr */ if (extra_msr_offset32) outp += sprintf(outp, " 0x%08llx", t->extra_msr32); /* MSR */ if (extra_msr_offset64) outp += sprintf(outp, " 0x%016llx", t->extra_msr64); if (do_nhm_cstates) { if (!skip_c1) outp += sprintf(outp, " %6.2f", 100.0 * t->c1/t->tsc); else outp += sprintf(outp, " ****"); } /* print per-core data only for 1st thread in core */ if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) goto done; if (do_nhm_cstates) outp += sprintf(outp, " %6.2f", 100.0 * c->c3/t->tsc); if (do_nhm_cstates) outp += sprintf(outp, " %6.2f", 100.0 * c->c6/t->tsc); if (do_snb_cstates) outp += sprintf(outp, " %6.2f", 100.0 * c->c7/t->tsc); /* print per-package data only for 1st core in package */ if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) goto done; if (do_snb_cstates) outp += sprintf(outp, " %6.2f", 100.0 * p->pc2/t->tsc); if (do_nhm_cstates) outp += sprintf(outp, " %6.2f", 100.0 * p->pc3/t->tsc); if (do_nhm_cstates) outp += sprintf(outp, " %6.2f", 100.0 * p->pc6/t->tsc); if (do_snb_cstates) outp += sprintf(outp, " %6.2f", 100.0 * p->pc7/t->tsc); done: outp += sprintf(outp, "\n"); return 0; } void flush_stdout() { fputs(output_buffer, stdout); outp = output_buffer; } void flush_stderr() { fputs(output_buffer, stderr); outp = output_buffer; } void format_all_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { static int printed; if (!printed || !summary_only) print_header(); if (topo.num_cpus > 1) format_counters(&average.threads, &average.cores, &average.packages); printed = 1; if (summary_only) return; for_all_cpus(format_counters, t, c, p); } void delta_package(struct pkg_data *new, struct pkg_data *old) { old->pc2 = new->pc2 - old->pc2; old->pc3 = new->pc3 - old->pc3; old->pc6 = new->pc6 - old->pc6; old->pc7 = new->pc7 - old->pc7; } void delta_core(struct core_data *new, struct core_data *old) { old->c3 = new->c3 - old->c3; old->c6 = new->c6 - old->c6; old->c7 = new->c7 - old->c7; } /* * old = new - old */ void delta_thread(struct thread_data *new, struct thread_data *old, struct core_data *core_delta) { old->tsc = new->tsc - old->tsc; /* check for TSC < 1 Mcycles over interval */ if (old->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); } old->c1 = new->c1 - old->c1; if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) { old->aperf = new->aperf - old->aperf; old->mperf = new->mperf - old->mperf; } else { 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 counter collection is not atomic, * it is possible for mperf's non-halted cycles + idle states * to exceed TSC's all cycles: show c1 = 0% in that case. */ if ((old->mperf + core_delta->c3 + core_delta->c6 + core_delta->c7) > old->tsc) old->c1 = 0; else { /* normal case, derive c1 */ old->c1 = old->tsc - old->mperf - core_delta->c3 - core_delta->c6 - core_delta->c7; } if (old->mperf == 0) { if (verbose > 1) fprintf(stderr, "cpu%d MPERF 0!\n", old->cpu_id); old->mperf = 1; /* divide by 0 protection */ } old->extra_delta32 = new->extra_delta32 - old->extra_delta32; old->extra_delta32 &= 0xFFFFFFFF; old->extra_delta64 = new->extra_delta64 - old->extra_delta64; /* * Extra MSR is just a snapshot, simply copy latest w/o subtracting */ old->extra_msr32 = new->extra_msr32; old->extra_msr64 = new->extra_msr64; } int delta_cpu(struct thread_data *t, struct core_data *c, struct pkg_data *p, struct thread_data *t2, struct core_data *c2, struct pkg_data *p2) { /* calculate core delta only for 1st thread in core */ if (t->flags & CPU_IS_FIRST_THREAD_IN_CORE) delta_core(c, c2); /* always calculate thread delta */ delta_thread(t, t2, c2); /* c2 is core delta */ /* calculate package delta only for 1st core in package */ if (t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE) delta_package(p, p2); return 0; } void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { t->tsc = 0; t->aperf = 0; t->mperf = 0; t->c1 = 0; t->extra_delta32 = 0; t->extra_delta64 = 0; /* tells format_counters to dump all fields from this set */ t->flags = CPU_IS_FIRST_THREAD_IN_CORE | CPU_IS_FIRST_CORE_IN_PACKAGE; c->c3 = 0; c->c6 = 0; c->c7 = 0; p->pc2 = 0; p->pc3 = 0; p->pc6 = 0; p->pc7 = 0; } int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { average.threads.tsc += t->tsc; average.threads.aperf += t->aperf; average.threads.mperf += t->mperf; average.threads.c1 += t->c1; average.threads.extra_delta32 += t->extra_delta32; average.threads.extra_delta64 += t->extra_delta64; /* sum per-core values only for 1st thread in core */ if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) return 0; average.cores.c3 += c->c3; average.cores.c6 += c->c6; average.cores.c7 += c->c7; /* sum per-pkg values only for 1st core in pkg */ if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) return 0; average.packages.pc2 += p->pc2; average.packages.pc3 += p->pc3; average.packages.pc6 += p->pc6; average.packages.pc7 += p->pc7; return 0; } /* * sum the counters for all cpus in the system * compute the weighted average */ void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data *p) { clear_counters(&average.threads, &average.cores, &average.packages); for_all_cpus(sum_counters, t, c, p); average.threads.tsc /= topo.num_cpus; average.threads.aperf /= topo.num_cpus; average.threads.mperf /= topo.num_cpus; average.threads.c1 /= topo.num_cpus; average.threads.extra_delta32 /= topo.num_cpus; average.threads.extra_delta32 &= 0xFFFFFFFF; average.threads.extra_delta64 /= topo.num_cpus; average.cores.c3 /= topo.num_cores; average.cores.c6 /= topo.num_cores; average.cores.c7 /= topo.num_cores; average.packages.pc2 /= topo.num_packages; average.packages.pc3 /= topo.num_packages; average.packages.pc6 /= topo.num_packages; average.packages.pc7 /= topo.num_packages; } static unsigned long long rdtsc(void) { unsigned int low, high; asm volatile("rdtsc" : "=a" (low), "=d" (high)); return low | ((unsigned long long)high) << 32; } /* * get_counters(...) * migrate to cpu * acquire and record local counters for that cpu */ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { int cpu = t->cpu_id; if (cpu_migrate(cpu)) return -1; t->tsc = rdtsc(); /* we are running on local CPU of interest */ if (has_aperf) { if (get_msr(cpu, MSR_APERF, &t->aperf)) return -3; if (get_msr(cpu, MSR_MPERF, &t->mperf)) return -4; } if (extra_delta_offset32) { if (get_msr(cpu, extra_delta_offset32, &t->extra_delta32)) return -5; t->extra_delta32 &= 0xFFFFFFFF; } if (extra_delta_offset64) if (get_msr(cpu, extra_delta_offset64, &t->extra_delta64)) return -5; if (extra_msr_offset32) { if (get_msr(cpu, extra_msr_offset32, &t->extra_msr32)) return -5; t->extra_msr32 &= 0xFFFFFFFF; } if (extra_msr_offset64) if (get_msr(cpu, extra_msr_offset64, &t->extra_msr64)) return -5; /* collect core counters only for 1st thread in core */ if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE)) return 0; if (do_nhm_cstates) { if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3)) return -6; if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6)) return -7; } if (do_snb_cstates) if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7)) return -8; /* collect package counters only for 1st core in package */ if (!(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE)) return 0; if (do_nhm_cstates) { if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3)) return -9; if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6)) return -10; } if (do_snb_cstates) { if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2)) return -11; if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7)) return -12; } return 0; } void print_verbose_header(void) { unsigned long long msr; unsigned int ratio; if (!do_nehalem_platform_info) return; get_msr(0, MSR_NEHALEM_PLATFORM_INFO, &msr); if (verbose > 1) fprintf(stderr, "MSR_NEHALEM_PLATFORM_INFO: 0x%llx\n", msr); 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 (!do_ivt_turbo_ratio_limit) goto print_nhm_turbo_ratio_limits; get_msr(0, MSR_IVT_TURBO_RATIO_LIMIT, &msr); if (verbose > 1) fprintf(stderr, "MSR_IVT_TURBO_RATIO_LIMIT: 0x%llx\n", msr); ratio = (msr >> 56) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 16 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 48) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 15 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 40) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 14 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 32) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 13 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 24) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 12 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 16) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 11 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 8) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 10 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 0) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 9 active cores\n", ratio, bclk, ratio * bclk); print_nhm_turbo_ratio_limits: if (!do_nehalem_turbo_ratio_limit) return; get_msr(0, MSR_NEHALEM_TURBO_RATIO_LIMIT, &msr); if (verbose > 1) fprintf(stderr, "MSR_NEHALEM_TURBO_RATIO_LIMIT: 0x%llx\n", msr); ratio = (msr >> 56) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 8 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 48) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 7 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 40) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 6 active cores\n", ratio, bclk, ratio * bclk); ratio = (msr >> 32) & 0xFF; if (ratio) fprintf(stderr, "%d * %.0f = %.0f MHz max turbo 5 active cores\n", ratio, bclk, ratio * bclk); 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_all_buffers(void) { CPU_FREE(cpu_present_set); cpu_present_set = NULL; cpu_present_set = 0; CPU_FREE(cpu_affinity_set); cpu_affinity_set = NULL; cpu_affinity_setsize = 0; free(thread_even); free(core_even); free(package_even); thread_even = NULL; core_even = NULL; package_even = NULL; free(thread_odd); free(core_odd); free(package_odd); thread_odd = NULL; core_odd = NULL; package_odd = NULL; free(output_buffer); output_buffer = NULL; outp = NULL; } /* * cpu_is_first_sibling_in_core(cpu) * return 1 if given CPU is 1st HT sibling in the core */ int cpu_is_first_sibling_in_core(int cpu) { char path[64]; FILE *filep; int first_cpu; sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu); filep = fopen(path, "r"); if (filep == NULL) { perror(path); exit(1); } fscanf(filep, "%d", &first_cpu); fclose(filep); return (cpu == first_cpu); } /* * cpu_is_first_core_in_package(cpu) * return 1 if given CPU is 1st core in package */ int cpu_is_first_core_in_package(int cpu) { char path[64]; FILE *filep; int first_cpu; sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/core_siblings_list", cpu); filep = fopen(path, "r"); if (filep == NULL) { perror(path); exit(1); } fscanf(filep, "%d", &first_cpu); fclose(filep); return (cpu == first_cpu); } int get_physical_package_id(int cpu) { char path[80]; 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[80]; 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; } int get_num_ht_siblings(int cpu) { char path[80]; FILE *filep; int sib1, sib2; int matches; char character; sprintf(path, "/sys/devices/system/cpu/cpu%d/topology/thread_siblings_list", cpu); filep = fopen(path, "r"); if (filep == NULL) { perror(path); exit(1); } /* * file format: * if a pair of number with a character between: 2 siblings (eg. 1-2, or 1,4) * otherwinse 1 sibling (self). */ matches = fscanf(filep, "%d%c%d\n", &sib1, &character, &sib2); fclose(filep); if (matches == 3) return 2; else return 1; } /* * run func(thread, core, package) in topology order * skip non-present cpus */ int for_all_cpus_2(int (func)(struct thread_data *, struct core_data *, struct pkg_data *, struct thread_data *, struct core_data *, struct pkg_data *), struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base, struct thread_data *thread_base2, struct core_data *core_base2, struct pkg_data *pkg_base2) { int retval, pkg_no, core_no, thread_no; for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { for (core_no = 0; core_no < topo.num_cores_per_pkg; ++core_no) { for (thread_no = 0; thread_no < topo.num_threads_per_core; ++thread_no) { struct thread_data *t, *t2; struct core_data *c, *c2; struct pkg_data *p, *p2; t = GET_THREAD(thread_base, thread_no, core_no, pkg_no); if (cpu_is_not_present(t->cpu_id)) continue; t2 = GET_THREAD(thread_base2, thread_no, core_no, pkg_no); c = GET_CORE(core_base, core_no, pkg_no); c2 = GET_CORE(core_base2, core_no, pkg_no); p = GET_PKG(pkg_base, pkg_no); p2 = GET_PKG(pkg_base2, pkg_no); retval = func(t, c, p, t2, c2, p2); if (retval) return retval; } } } return 0; } /* * run func(cpu) on every cpu in /proc/stat * return max_cpu number */ int for_all_proc_cpus(int (func)(int)) { FILE *fp; int cpu_num; int retval; fp = fopen(proc_stat, "r"); if (fp == NULL) { perror(proc_stat); exit(1); } retval = fscanf(fp, "cpu %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n"); if (retval != 0) { perror("/proc/stat format"); exit(1); } while (1) { retval = fscanf(fp, "cpu%u %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d\n", &cpu_num); if (retval != 1) break; retval = func(cpu_num); if (retval) { fclose(fp); return(retval); } } fclose(fp); return 0; } void re_initialize(void) { free_all_buffers(); setup_all_buffers(); printf("turbostat: re-initialized with num_cpus %d\n", topo.num_cpus); } /* * count_cpus() * remember the last one seen, it will be the max */ int count_cpus(int cpu) { if (topo.max_cpu_num < cpu) topo.max_cpu_num = cpu; topo.num_cpus += 1; return 0; } int mark_cpu_present(int cpu) { CPU_SET_S(cpu, cpu_present_setsize, cpu_present_set); return 0; } void turbostat_loop() { int retval; restart: retval = for_all_cpus(get_counters, EVEN_COUNTERS); if (retval) { re_initialize(); goto restart; } gettimeofday(&tv_even, (struct timezone *)NULL); while (1) { if (for_all_proc_cpus(cpu_is_not_present)) { re_initialize(); goto restart; } sleep(interval_sec); retval = for_all_cpus(get_counters, ODD_COUNTERS); if (retval) { re_initialize(); goto restart; } gettimeofday(&tv_odd, (struct timezone *)NULL); timersub(&tv_odd, &tv_even, &tv_delta); for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); compute_average(EVEN_COUNTERS); format_all_counters(EVEN_COUNTERS); flush_stdout(); sleep(interval_sec); retval = for_all_cpus(get_counters, EVEN_COUNTERS); if (retval) { re_initialize(); goto restart; } gettimeofday(&tv_even, (struct timezone *)NULL); timersub(&tv_even, &tv_odd, &tv_delta); for_all_cpus_2(delta_cpu, EVEN_COUNTERS, ODD_COUNTERS); compute_average(ODD_COUNTERS); format_all_counters(ODD_COUNTERS); flush_stdout(); } } 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 */ case 0x3A: /* IVB */ case 0x3E: /* IVB Xeon */ return 1; case 0x2E: /* Nehalem-EX Xeon - Beckton */ case 0x2F: /* Westmere-EX Xeon - Eagleton */ default: return 0; } } int has_ivt_turbo_ratio_limit(unsigned int family, unsigned int model) { if (!genuine_intel) return 0; if (family != 6) return 0; switch (model) { case 0x3E: /* IVB Xeon */ return 1; default: return 0; } } int is_snb(unsigned int family, unsigned int model) { if (!genuine_intel) return 0; switch (model) { case 0x2A: case 0x2D: case 0x3A: /* IVB */ case 0x3E: /* IVB Xeon */ 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); do_ivt_turbo_ratio_limit = has_ivt_turbo_ratio_limit(family, model); } void usage() { fprintf(stderr, "%s: [-v][-d MSR#][-D MSR#][-m MSR#][-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 topology_probe() { int i; int max_core_id = 0; int max_package_id = 0; int max_siblings = 0; struct cpu_topology { int core_id; int physical_package_id; } *cpus; /* Initialize num_cpus, max_cpu_num */ topo.num_cpus = 0; topo.max_cpu_num = 0; for_all_proc_cpus(count_cpus); if (!summary_only && topo.num_cpus > 1) show_cpu = 1; if (verbose > 1) fprintf(stderr, "num_cpus %d max_cpu_num %d\n", topo.num_cpus, topo.max_cpu_num); cpus = calloc(1, (topo.max_cpu_num + 1) * sizeof(struct cpu_topology)); if (cpus == NULL) { perror("calloc cpus"); exit(1); } /* * Allocate and initialize cpu_present_set */ cpu_present_set = CPU_ALLOC((topo.max_cpu_num + 1)); if (cpu_present_set == NULL) { perror("CPU_ALLOC"); exit(3); } cpu_present_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); CPU_ZERO_S(cpu_present_setsize, cpu_present_set); for_all_proc_cpus(mark_cpu_present); /* * Allocate and initialize cpu_affinity_set */ cpu_affinity_set = CPU_ALLOC((topo.max_cpu_num + 1)); if (cpu_affinity_set == NULL) { perror("CPU_ALLOC"); exit(3); } cpu_affinity_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); CPU_ZERO_S(cpu_affinity_setsize, cpu_affinity_set); /* * For online cpus * find max_core_id, max_package_id */ for (i = 0; i <= topo.max_cpu_num; ++i) { int siblings; if (cpu_is_not_present(i)) { if (verbose > 1) fprintf(stderr, "cpu%d NOT PRESENT\n", i); continue; } cpus[i].core_id = get_core_id(i); if (cpus[i].core_id > max_core_id) max_core_id = cpus[i].core_id; cpus[i].physical_package_id = get_physical_package_id(i); if (cpus[i].physical_package_id > max_package_id) max_package_id = cpus[i].physical_package_id; siblings = get_num_ht_siblings(i); if (siblings > max_siblings) max_siblings = siblings; if (verbose > 1) fprintf(stderr, "cpu %d pkg %d core %d\n", i, cpus[i].physical_package_id, cpus[i].core_id); } topo.num_cores_per_pkg = max_core_id + 1; if (verbose > 1) fprintf(stderr, "max_core_id %d, sizing for %d cores per package\n", max_core_id, topo.num_cores_per_pkg); if (!summary_only && topo.num_cores_per_pkg > 1) show_core = 1; topo.num_packages = max_package_id + 1; if (verbose > 1) fprintf(stderr, "max_package_id %d, sizing for %d packages\n", max_package_id, topo.num_packages); if (!summary_only && topo.num_packages > 1) show_pkg = 1; topo.num_threads_per_core = max_siblings; if (verbose > 1) fprintf(stderr, "max_siblings %d\n", max_siblings); free(cpus); } void allocate_counters(struct thread_data **t, struct core_data **c, struct pkg_data **p) { int i; *t = calloc(topo.num_threads_per_core * topo.num_cores_per_pkg * topo.num_packages, sizeof(struct thread_data)); if (*t == NULL) goto error; for (i = 0; i < topo.num_threads_per_core * topo.num_cores_per_pkg * topo.num_packages; i++) (*t)[i].cpu_id = -1; *c = calloc(topo.num_cores_per_pkg * topo.num_packages, sizeof(struct core_data)); if (*c == NULL) goto error; for (i = 0; i < topo.num_cores_per_pkg * topo.num_packages; i++) (*c)[i].core_id = -1; *p = calloc(topo.num_packages, sizeof(struct pkg_data)); if (*p == NULL) goto error; for (i = 0; i < topo.num_packages; i++) (*p)[i].package_id = i; return; error: perror("calloc counters"); exit(1); } /* * init_counter() * * set cpu_id, core_num, pkg_num * set FIRST_THREAD_IN_CORE and FIRST_CORE_IN_PACKAGE * * increment topo.num_cores when 1st core in pkg seen */ void init_counter(struct thread_data *thread_base, struct core_data *core_base, struct pkg_data *pkg_base, int thread_num, int core_num, int pkg_num, int cpu_id) { struct thread_data *t; struct core_data *c; struct pkg_data *p; t = GET_THREAD(thread_base, thread_num, core_num, pkg_num); c = GET_CORE(core_base, core_num, pkg_num); p = GET_PKG(pkg_base, pkg_num); t->cpu_id = cpu_id; if (thread_num == 0) { t->flags |= CPU_IS_FIRST_THREAD_IN_CORE; if (cpu_is_first_core_in_package(cpu_id)) t->flags |= CPU_IS_FIRST_CORE_IN_PACKAGE; } c->core_id = core_num; p->package_id = pkg_num; } int initialize_counters(int cpu_id) { int my_thread_id, my_core_id, my_package_id; my_package_id = get_physical_package_id(cpu_id); my_core_id = get_core_id(cpu_id); if (cpu_is_first_sibling_in_core(cpu_id)) { my_thread_id = 0; topo.num_cores++; } else { my_thread_id = 1; } init_counter(EVEN_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); init_counter(ODD_COUNTERS, my_thread_id, my_core_id, my_package_id, cpu_id); return 0; } void allocate_output_buffer() { output_buffer = calloc(1, (1 + topo.num_cpus) * 128); outp = output_buffer; if (outp == NULL) { perror("calloc"); exit(-1); } } void setup_all_buffers(void) { topology_probe(); allocate_counters(&thread_even, &core_even, &package_even); allocate_counters(&thread_odd, &core_odd, &package_odd); allocate_output_buffer(); for_all_proc_cpus(initialize_counters); } void turbostat_init() { check_cpuid(); check_dev_msr(); check_super_user(); setup_all_buffers(); if (verbose) print_verbose_header(); } int fork_it(char **argv) { pid_t child_pid; for_all_cpus(get_counters, EVEN_COUNTERS); /* clear affinity side-effect of get_counters() */ sched_setaffinity(0, cpu_present_setsize, cpu_present_set); 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); } } /* * n.b. fork_it() does not check for errors from for_all_cpus() * because re-starting is problematic when forking */ for_all_cpus(get_counters, ODD_COUNTERS); gettimeofday(&tv_odd, (struct timezone *)NULL); timersub(&tv_odd, &tv_even, &tv_delta); for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS); compute_average(EVEN_COUNTERS); format_all_counters(EVEN_COUNTERS); flush_stderr(); 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, "+cpsvid:D:m:M:")) != -1) { switch (opt) { case 'c': show_core_only++; break; case 'p': show_pkg_only++; break; case 's': summary_only++; break; case 'v': verbose++; break; case 'i': interval_sec = atoi(optarg); break; case 'd': sscanf(optarg, "%x", &extra_delta_offset32); break; case 'D': sscanf(optarg, "%x", &extra_delta_offset64); break; case 'm': sscanf(optarg, "%x", &extra_msr_offset32); break; case 'M': sscanf(optarg, "%x", &extra_msr_offset64); break; default: usage(); } } } int main(int argc, char **argv) { cmdline(argc, argv); if (verbose > 1) fprintf(stderr, "turbostat v2.0 May 16, 2012" " - Len Brown \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; }